xref: /freebsd/sys/fs/nfsclient/nfs_clvnops.c (revision 1f88aa09417f1cfb3929fd37531b1ab51213c2d6)
1 /*-
2  * SPDX-License-Identifier: BSD-3-Clause
3  *
4  * Copyright (c) 1989, 1993
5  *	The Regents of the University of California.  All rights reserved.
6  *
7  * This code is derived from software contributed to Berkeley by
8  * Rick Macklem at The University of Guelph.
9  *
10  * Redistribution and use in source and binary forms, with or without
11  * modification, are permitted provided that the following conditions
12  * are met:
13  * 1. Redistributions of source code must retain the above copyright
14  *    notice, this list of conditions and the following disclaimer.
15  * 2. Redistributions in binary form must reproduce the above copyright
16  *    notice, this list of conditions and the following disclaimer in the
17  *    documentation and/or other materials provided with the distribution.
18  * 3. Neither the name of the University nor the names of its contributors
19  *    may be used to endorse or promote products derived from this software
20  *    without specific prior written permission.
21  *
22  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
23  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
24  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
25  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
26  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
27  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
28  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
29  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
30  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
31  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
32  * SUCH DAMAGE.
33  *
34  *	from nfs_vnops.c	8.16 (Berkeley) 5/27/95
35  */
36 
37 #include <sys/cdefs.h>
38 __FBSDID("$FreeBSD$");
39 
40 /*
41  * vnode op calls for Sun NFS version 2, 3 and 4
42  */
43 
44 #include "opt_inet.h"
45 
46 #include <sys/param.h>
47 #include <sys/kernel.h>
48 #include <sys/systm.h>
49 #include <sys/resourcevar.h>
50 #include <sys/proc.h>
51 #include <sys/mount.h>
52 #include <sys/bio.h>
53 #include <sys/buf.h>
54 #include <sys/extattr.h>
55 #include <sys/filio.h>
56 #include <sys/jail.h>
57 #include <sys/malloc.h>
58 #include <sys/mbuf.h>
59 #include <sys/namei.h>
60 #include <sys/socket.h>
61 #include <sys/vnode.h>
62 #include <sys/dirent.h>
63 #include <sys/fcntl.h>
64 #include <sys/lockf.h>
65 #include <sys/stat.h>
66 #include <sys/sysctl.h>
67 #include <sys/signalvar.h>
68 
69 #include <vm/vm.h>
70 #include <vm/vm_extern.h>
71 #include <vm/vm_object.h>
72 
73 #include <fs/nfs/nfsport.h>
74 #include <fs/nfsclient/nfsnode.h>
75 #include <fs/nfsclient/nfsmount.h>
76 #include <fs/nfsclient/nfs.h>
77 #include <fs/nfsclient/nfs_kdtrace.h>
78 
79 #include <net/if.h>
80 #include <netinet/in.h>
81 #include <netinet/in_var.h>
82 
83 #include <nfs/nfs_lock.h>
84 
85 #ifdef KDTRACE_HOOKS
86 #include <sys/dtrace_bsd.h>
87 
88 dtrace_nfsclient_accesscache_flush_probe_func_t
89 		dtrace_nfscl_accesscache_flush_done_probe;
90 uint32_t	nfscl_accesscache_flush_done_id;
91 
92 dtrace_nfsclient_accesscache_get_probe_func_t
93 		dtrace_nfscl_accesscache_get_hit_probe,
94 		dtrace_nfscl_accesscache_get_miss_probe;
95 uint32_t	nfscl_accesscache_get_hit_id;
96 uint32_t	nfscl_accesscache_get_miss_id;
97 
98 dtrace_nfsclient_accesscache_load_probe_func_t
99 		dtrace_nfscl_accesscache_load_done_probe;
100 uint32_t	nfscl_accesscache_load_done_id;
101 #endif /* !KDTRACE_HOOKS */
102 
103 /* Defs */
104 #define	TRUE	1
105 #define	FALSE	0
106 
107 extern struct nfsstatsv1 nfsstatsv1;
108 extern int nfsrv_useacl;
109 extern int nfscl_debuglevel;
110 MALLOC_DECLARE(M_NEWNFSREQ);
111 
112 static vop_read_t	nfsfifo_read;
113 static vop_write_t	nfsfifo_write;
114 static vop_close_t	nfsfifo_close;
115 static int	nfs_setattrrpc(struct vnode *, struct vattr *, struct ucred *,
116 		    struct thread *);
117 static vop_lookup_t	nfs_lookup;
118 static vop_create_t	nfs_create;
119 static vop_mknod_t	nfs_mknod;
120 static vop_open_t	nfs_open;
121 static vop_pathconf_t	nfs_pathconf;
122 static vop_close_t	nfs_close;
123 static vop_access_t	nfs_access;
124 static vop_getattr_t	nfs_getattr;
125 static vop_setattr_t	nfs_setattr;
126 static vop_read_t	nfs_read;
127 static vop_fsync_t	nfs_fsync;
128 static vop_remove_t	nfs_remove;
129 static vop_link_t	nfs_link;
130 static vop_rename_t	nfs_rename;
131 static vop_mkdir_t	nfs_mkdir;
132 static vop_rmdir_t	nfs_rmdir;
133 static vop_symlink_t	nfs_symlink;
134 static vop_readdir_t	nfs_readdir;
135 static vop_strategy_t	nfs_strategy;
136 static	int	nfs_lookitup(struct vnode *, char *, int,
137 		    struct ucred *, struct thread *, struct nfsnode **);
138 static	int	nfs_sillyrename(struct vnode *, struct vnode *,
139 		    struct componentname *);
140 static vop_access_t	nfsspec_access;
141 static vop_readlink_t	nfs_readlink;
142 static vop_print_t	nfs_print;
143 static vop_advlock_t	nfs_advlock;
144 static vop_advlockasync_t nfs_advlockasync;
145 static vop_getacl_t nfs_getacl;
146 static vop_setacl_t nfs_setacl;
147 static vop_advise_t nfs_advise;
148 static vop_allocate_t nfs_allocate;
149 static vop_copy_file_range_t nfs_copy_file_range;
150 static vop_ioctl_t nfs_ioctl;
151 static vop_getextattr_t nfs_getextattr;
152 static vop_setextattr_t nfs_setextattr;
153 static vop_listextattr_t nfs_listextattr;
154 static vop_deleteextattr_t nfs_deleteextattr;
155 static vop_lock1_t	nfs_lock;
156 
157 /*
158  * Global vfs data structures for nfs
159  */
160 
161 static struct vop_vector newnfs_vnodeops_nosig = {
162 	.vop_default =		&default_vnodeops,
163 	.vop_access =		nfs_access,
164 	.vop_advlock =		nfs_advlock,
165 	.vop_advlockasync =	nfs_advlockasync,
166 	.vop_close =		nfs_close,
167 	.vop_create =		nfs_create,
168 	.vop_fsync =		nfs_fsync,
169 	.vop_getattr =		nfs_getattr,
170 	.vop_getpages =		ncl_getpages,
171 	.vop_putpages =		ncl_putpages,
172 	.vop_inactive =		ncl_inactive,
173 	.vop_link =		nfs_link,
174 	.vop_lock1 =		nfs_lock,
175 	.vop_lookup =		nfs_lookup,
176 	.vop_mkdir =		nfs_mkdir,
177 	.vop_mknod =		nfs_mknod,
178 	.vop_open =		nfs_open,
179 	.vop_pathconf =		nfs_pathconf,
180 	.vop_print =		nfs_print,
181 	.vop_read =		nfs_read,
182 	.vop_readdir =		nfs_readdir,
183 	.vop_readlink =		nfs_readlink,
184 	.vop_reclaim =		ncl_reclaim,
185 	.vop_remove =		nfs_remove,
186 	.vop_rename =		nfs_rename,
187 	.vop_rmdir =		nfs_rmdir,
188 	.vop_setattr =		nfs_setattr,
189 	.vop_strategy =		nfs_strategy,
190 	.vop_symlink =		nfs_symlink,
191 	.vop_write =		ncl_write,
192 	.vop_getacl =		nfs_getacl,
193 	.vop_setacl =		nfs_setacl,
194 	.vop_advise =		nfs_advise,
195 	.vop_allocate =		nfs_allocate,
196 	.vop_copy_file_range =	nfs_copy_file_range,
197 	.vop_ioctl =		nfs_ioctl,
198 	.vop_getextattr =	nfs_getextattr,
199 	.vop_setextattr =	nfs_setextattr,
200 	.vop_listextattr =	nfs_listextattr,
201 	.vop_deleteextattr =	nfs_deleteextattr,
202 };
203 VFS_VOP_VECTOR_REGISTER(newnfs_vnodeops_nosig);
204 
205 static int
206 nfs_vnodeops_bypass(struct vop_generic_args *a)
207 {
208 
209 	return (vop_sigdefer(&newnfs_vnodeops_nosig, a));
210 }
211 
212 struct vop_vector newnfs_vnodeops = {
213 	.vop_default =		&default_vnodeops,
214 	.vop_bypass =		nfs_vnodeops_bypass,
215 };
216 VFS_VOP_VECTOR_REGISTER(newnfs_vnodeops);
217 
218 static struct vop_vector newnfs_fifoops_nosig = {
219 	.vop_default =		&fifo_specops,
220 	.vop_access =		nfsspec_access,
221 	.vop_close =		nfsfifo_close,
222 	.vop_fsync =		nfs_fsync,
223 	.vop_getattr =		nfs_getattr,
224 	.vop_inactive =		ncl_inactive,
225 	.vop_pathconf =		nfs_pathconf,
226 	.vop_print =		nfs_print,
227 	.vop_read =		nfsfifo_read,
228 	.vop_reclaim =		ncl_reclaim,
229 	.vop_setattr =		nfs_setattr,
230 	.vop_write =		nfsfifo_write,
231 };
232 VFS_VOP_VECTOR_REGISTER(newnfs_fifoops_nosig);
233 
234 static int
235 nfs_fifoops_bypass(struct vop_generic_args *a)
236 {
237 
238 	return (vop_sigdefer(&newnfs_fifoops_nosig, a));
239 }
240 
241 struct vop_vector newnfs_fifoops = {
242 	.vop_default =		&default_vnodeops,
243 	.vop_bypass =		nfs_fifoops_bypass,
244 };
245 VFS_VOP_VECTOR_REGISTER(newnfs_fifoops);
246 
247 static int nfs_mknodrpc(struct vnode *dvp, struct vnode **vpp,
248     struct componentname *cnp, struct vattr *vap);
249 static int nfs_removerpc(struct vnode *dvp, struct vnode *vp, char *name,
250     int namelen, struct ucred *cred, struct thread *td);
251 static int nfs_renamerpc(struct vnode *fdvp, struct vnode *fvp,
252     char *fnameptr, int fnamelen, struct vnode *tdvp, struct vnode *tvp,
253     char *tnameptr, int tnamelen, struct ucred *cred, struct thread *td);
254 static int nfs_renameit(struct vnode *sdvp, struct vnode *svp,
255     struct componentname *scnp, struct sillyrename *sp);
256 
257 /*
258  * Global variables
259  */
260 SYSCTL_DECL(_vfs_nfs);
261 
262 static int	nfsaccess_cache_timeout = NFS_MAXATTRTIMO;
263 SYSCTL_INT(_vfs_nfs, OID_AUTO, access_cache_timeout, CTLFLAG_RW,
264 	   &nfsaccess_cache_timeout, 0, "NFS ACCESS cache timeout");
265 
266 static int	nfs_prime_access_cache = 0;
267 SYSCTL_INT(_vfs_nfs, OID_AUTO, prime_access_cache, CTLFLAG_RW,
268 	   &nfs_prime_access_cache, 0,
269 	   "Prime NFS ACCESS cache when fetching attributes");
270 
271 static int	newnfs_commit_on_close = 0;
272 SYSCTL_INT(_vfs_nfs, OID_AUTO, commit_on_close, CTLFLAG_RW,
273     &newnfs_commit_on_close, 0, "write+commit on close, else only write");
274 
275 static int	nfs_clean_pages_on_close = 1;
276 SYSCTL_INT(_vfs_nfs, OID_AUTO, clean_pages_on_close, CTLFLAG_RW,
277 	   &nfs_clean_pages_on_close, 0, "NFS clean dirty pages on close");
278 
279 int newnfs_directio_enable = 0;
280 SYSCTL_INT(_vfs_nfs, OID_AUTO, nfs_directio_enable, CTLFLAG_RW,
281 	   &newnfs_directio_enable, 0, "Enable NFS directio");
282 
283 int nfs_keep_dirty_on_error;
284 SYSCTL_INT(_vfs_nfs, OID_AUTO, nfs_keep_dirty_on_error, CTLFLAG_RW,
285     &nfs_keep_dirty_on_error, 0, "Retry pageout if error returned");
286 
287 /*
288  * This sysctl allows other processes to mmap a file that has been opened
289  * O_DIRECT by a process.  In general, having processes mmap the file while
290  * Direct IO is in progress can lead to Data Inconsistencies.  But, we allow
291  * this by default to prevent DoS attacks - to prevent a malicious user from
292  * opening up files O_DIRECT preventing other users from mmap'ing these
293  * files.  "Protected" environments where stricter consistency guarantees are
294  * required can disable this knob.  The process that opened the file O_DIRECT
295  * cannot mmap() the file, because mmap'ed IO on an O_DIRECT open() is not
296  * meaningful.
297  */
298 int newnfs_directio_allow_mmap = 1;
299 SYSCTL_INT(_vfs_nfs, OID_AUTO, nfs_directio_allow_mmap, CTLFLAG_RW,
300 	   &newnfs_directio_allow_mmap, 0, "Enable mmaped IO on file with O_DIRECT opens");
301 
302 #define	NFSACCESS_ALL (NFSACCESS_READ | NFSACCESS_MODIFY		\
303 			 | NFSACCESS_EXTEND | NFSACCESS_EXECUTE	\
304 			 | NFSACCESS_DELETE | NFSACCESS_LOOKUP)
305 
306 /*
307  * SMP Locking Note :
308  * The list of locks after the description of the lock is the ordering
309  * of other locks acquired with the lock held.
310  * np->n_mtx : Protects the fields in the nfsnode.
311        VM Object Lock
312        VI_MTX (acquired indirectly)
313  * nmp->nm_mtx : Protects the fields in the nfsmount.
314        rep->r_mtx
315  * ncl_iod_mutex : Global lock, protects shared nfsiod state.
316  * nfs_reqq_mtx : Global lock, protects the nfs_reqq list.
317        nmp->nm_mtx
318        rep->r_mtx
319  * rep->r_mtx : Protects the fields in an nfsreq.
320  */
321 
322 static int
323 nfs_lock(struct vop_lock1_args *ap)
324 {
325 	struct vnode *vp;
326 	struct nfsnode *np;
327 	u_quad_t nsize;
328 	int error, lktype;
329 	bool onfault;
330 
331 	vp = ap->a_vp;
332 	lktype = ap->a_flags & LK_TYPE_MASK;
333 	error = VOP_LOCK1_APV(&default_vnodeops, ap);
334 	if (error != 0 || vp->v_op != &newnfs_vnodeops)
335 		return (error);
336 	np = VTONFS(vp);
337 	if (np == NULL)
338 		return (0);
339 	NFSLOCKNODE(np);
340 	if ((np->n_flag & NVNSETSZSKIP) == 0 || (lktype != LK_SHARED &&
341 	    lktype != LK_EXCLUSIVE && lktype != LK_UPGRADE &&
342 	    lktype != LK_TRYUPGRADE)) {
343 		NFSUNLOCKNODE(np);
344 		return (0);
345 	}
346 	onfault = (ap->a_flags & LK_EATTR_MASK) == LK_NOWAIT &&
347 	    (ap->a_flags & LK_INIT_MASK) == LK_CANRECURSE &&
348 	    (lktype == LK_SHARED || lktype == LK_EXCLUSIVE);
349 	if (onfault && vp->v_vnlock->lk_recurse == 0) {
350 		/*
351 		 * Force retry in vm_fault(), to make the lock request
352 		 * sleepable, which allows us to piggy-back the
353 		 * sleepable call to vnode_pager_setsize().
354 		 */
355 		NFSUNLOCKNODE(np);
356 		VOP_UNLOCK(vp);
357 		return (EBUSY);
358 	}
359 	if ((ap->a_flags & LK_NOWAIT) != 0 ||
360 	    (lktype == LK_SHARED && vp->v_vnlock->lk_recurse > 0)) {
361 		NFSUNLOCKNODE(np);
362 		return (0);
363 	}
364 	if (lktype == LK_SHARED) {
365 		NFSUNLOCKNODE(np);
366 		VOP_UNLOCK(vp);
367 		ap->a_flags &= ~(LK_TYPE_MASK | LK_INTERLOCK);
368 		ap->a_flags |= LK_EXCLUSIVE;
369 		error = VOP_LOCK1_APV(&default_vnodeops, ap);
370 		if (error != 0 || vp->v_op != &newnfs_vnodeops)
371 			return (error);
372 		if (vp->v_data == NULL)
373 			goto downgrade;
374 		MPASS(vp->v_data == np);
375 		NFSLOCKNODE(np);
376 		if ((np->n_flag & NVNSETSZSKIP) == 0) {
377 			NFSUNLOCKNODE(np);
378 			goto downgrade;
379 		}
380 	}
381 	np->n_flag &= ~NVNSETSZSKIP;
382 	nsize = np->n_size;
383 	NFSUNLOCKNODE(np);
384 	vnode_pager_setsize(vp, nsize);
385 downgrade:
386 	if (lktype == LK_SHARED) {
387 		ap->a_flags &= ~(LK_TYPE_MASK | LK_INTERLOCK);
388 		ap->a_flags |= LK_DOWNGRADE;
389 		(void)VOP_LOCK1_APV(&default_vnodeops, ap);
390 	}
391 	return (0);
392 }
393 
394 static int
395 nfs34_access_otw(struct vnode *vp, int wmode, struct thread *td,
396     struct ucred *cred, u_int32_t *retmode)
397 {
398 	int error = 0, attrflag, i, lrupos;
399 	u_int32_t rmode;
400 	struct nfsnode *np = VTONFS(vp);
401 	struct nfsvattr nfsva;
402 
403 	error = nfsrpc_accessrpc(vp, wmode, cred, td, &nfsva, &attrflag,
404 	    &rmode, NULL);
405 	if (attrflag)
406 		(void) nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0, 1);
407 	if (!error) {
408 		lrupos = 0;
409 		NFSLOCKNODE(np);
410 		for (i = 0; i < NFS_ACCESSCACHESIZE; i++) {
411 			if (np->n_accesscache[i].uid == cred->cr_uid) {
412 				np->n_accesscache[i].mode = rmode;
413 				np->n_accesscache[i].stamp = time_second;
414 				break;
415 			}
416 			if (i > 0 && np->n_accesscache[i].stamp <
417 			    np->n_accesscache[lrupos].stamp)
418 				lrupos = i;
419 		}
420 		if (i == NFS_ACCESSCACHESIZE) {
421 			np->n_accesscache[lrupos].uid = cred->cr_uid;
422 			np->n_accesscache[lrupos].mode = rmode;
423 			np->n_accesscache[lrupos].stamp = time_second;
424 		}
425 		NFSUNLOCKNODE(np);
426 		if (retmode != NULL)
427 			*retmode = rmode;
428 		KDTRACE_NFS_ACCESSCACHE_LOAD_DONE(vp, cred->cr_uid, rmode, 0);
429 	} else if (NFS_ISV4(vp)) {
430 		error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0);
431 	}
432 #ifdef KDTRACE_HOOKS
433 	if (error != 0)
434 		KDTRACE_NFS_ACCESSCACHE_LOAD_DONE(vp, cred->cr_uid, 0,
435 		    error);
436 #endif
437 	return (error);
438 }
439 
440 /*
441  * nfs access vnode op.
442  * For nfs version 2, just return ok. File accesses may fail later.
443  * For nfs version 3, use the access rpc to check accessibility. If file modes
444  * are changed on the server, accesses might still fail later.
445  */
446 static int
447 nfs_access(struct vop_access_args *ap)
448 {
449 	struct vnode *vp = ap->a_vp;
450 	int error = 0, i, gotahit;
451 	u_int32_t mode, wmode, rmode;
452 	int v34 = NFS_ISV34(vp);
453 	struct nfsnode *np = VTONFS(vp);
454 
455 	/*
456 	 * Disallow write attempts on filesystems mounted read-only;
457 	 * unless the file is a socket, fifo, or a block or character
458 	 * device resident on the filesystem.
459 	 */
460 	if ((ap->a_accmode & (VWRITE | VAPPEND | VWRITE_NAMED_ATTRS |
461 	    VDELETE_CHILD | VWRITE_ATTRIBUTES | VDELETE | VWRITE_ACL |
462 	    VWRITE_OWNER)) != 0 && (vp->v_mount->mnt_flag & MNT_RDONLY) != 0) {
463 		switch (vp->v_type) {
464 		case VREG:
465 		case VDIR:
466 		case VLNK:
467 			return (EROFS);
468 		default:
469 			break;
470 		}
471 	}
472 	/*
473 	 * For nfs v3 or v4, check to see if we have done this recently, and if
474 	 * so return our cached result instead of making an ACCESS call.
475 	 * If not, do an access rpc, otherwise you are stuck emulating
476 	 * ufs_access() locally using the vattr. This may not be correct,
477 	 * since the server may apply other access criteria such as
478 	 * client uid-->server uid mapping that we do not know about.
479 	 */
480 	if (v34) {
481 		if (ap->a_accmode & VREAD)
482 			mode = NFSACCESS_READ;
483 		else
484 			mode = 0;
485 		if (vp->v_type != VDIR) {
486 			if (ap->a_accmode & VWRITE)
487 				mode |= (NFSACCESS_MODIFY | NFSACCESS_EXTEND);
488 			if (ap->a_accmode & VAPPEND)
489 				mode |= NFSACCESS_EXTEND;
490 			if (ap->a_accmode & VEXEC)
491 				mode |= NFSACCESS_EXECUTE;
492 			if (ap->a_accmode & VDELETE)
493 				mode |= NFSACCESS_DELETE;
494 		} else {
495 			if (ap->a_accmode & VWRITE)
496 				mode |= (NFSACCESS_MODIFY | NFSACCESS_EXTEND);
497 			if (ap->a_accmode & VAPPEND)
498 				mode |= NFSACCESS_EXTEND;
499 			if (ap->a_accmode & VEXEC)
500 				mode |= NFSACCESS_LOOKUP;
501 			if (ap->a_accmode & VDELETE)
502 				mode |= NFSACCESS_DELETE;
503 			if (ap->a_accmode & VDELETE_CHILD)
504 				mode |= NFSACCESS_MODIFY;
505 		}
506 		/* XXX safety belt, only make blanket request if caching */
507 		if (nfsaccess_cache_timeout > 0) {
508 			wmode = NFSACCESS_READ | NFSACCESS_MODIFY |
509 				NFSACCESS_EXTEND | NFSACCESS_EXECUTE |
510 				NFSACCESS_DELETE | NFSACCESS_LOOKUP;
511 		} else {
512 			wmode = mode;
513 		}
514 
515 		/*
516 		 * Does our cached result allow us to give a definite yes to
517 		 * this request?
518 		 */
519 		gotahit = 0;
520 		NFSLOCKNODE(np);
521 		for (i = 0; i < NFS_ACCESSCACHESIZE; i++) {
522 			if (ap->a_cred->cr_uid == np->n_accesscache[i].uid) {
523 			    if (time_second < (np->n_accesscache[i].stamp
524 				+ nfsaccess_cache_timeout) &&
525 				(np->n_accesscache[i].mode & mode) == mode) {
526 				NFSINCRGLOBAL(nfsstatsv1.accesscache_hits);
527 				gotahit = 1;
528 			    }
529 			    break;
530 			}
531 		}
532 		NFSUNLOCKNODE(np);
533 #ifdef KDTRACE_HOOKS
534 		if (gotahit != 0)
535 			KDTRACE_NFS_ACCESSCACHE_GET_HIT(vp,
536 			    ap->a_cred->cr_uid, mode);
537 		else
538 			KDTRACE_NFS_ACCESSCACHE_GET_MISS(vp,
539 			    ap->a_cred->cr_uid, mode);
540 #endif
541 		if (gotahit == 0) {
542 			/*
543 			 * Either a no, or a don't know.  Go to the wire.
544 			 */
545 			NFSINCRGLOBAL(nfsstatsv1.accesscache_misses);
546 		        error = nfs34_access_otw(vp, wmode, ap->a_td,
547 			    ap->a_cred, &rmode);
548 			if (!error &&
549 			    (rmode & mode) != mode)
550 				error = EACCES;
551 		}
552 		return (error);
553 	} else {
554 		if ((error = nfsspec_access(ap)) != 0) {
555 			return (error);
556 		}
557 		/*
558 		 * Attempt to prevent a mapped root from accessing a file
559 		 * which it shouldn't.  We try to read a byte from the file
560 		 * if the user is root and the file is not zero length.
561 		 * After calling nfsspec_access, we should have the correct
562 		 * file size cached.
563 		 */
564 		NFSLOCKNODE(np);
565 		if (ap->a_cred->cr_uid == 0 && (ap->a_accmode & VREAD)
566 		    && VTONFS(vp)->n_size > 0) {
567 			struct iovec aiov;
568 			struct uio auio;
569 			char buf[1];
570 
571 			NFSUNLOCKNODE(np);
572 			aiov.iov_base = buf;
573 			aiov.iov_len = 1;
574 			auio.uio_iov = &aiov;
575 			auio.uio_iovcnt = 1;
576 			auio.uio_offset = 0;
577 			auio.uio_resid = 1;
578 			auio.uio_segflg = UIO_SYSSPACE;
579 			auio.uio_rw = UIO_READ;
580 			auio.uio_td = ap->a_td;
581 
582 			if (vp->v_type == VREG)
583 				error = ncl_readrpc(vp, &auio, ap->a_cred);
584 			else if (vp->v_type == VDIR) {
585 				char* bp;
586 				bp = malloc(NFS_DIRBLKSIZ, M_TEMP, M_WAITOK);
587 				aiov.iov_base = bp;
588 				aiov.iov_len = auio.uio_resid = NFS_DIRBLKSIZ;
589 				error = ncl_readdirrpc(vp, &auio, ap->a_cred,
590 				    ap->a_td);
591 				free(bp, M_TEMP);
592 			} else if (vp->v_type == VLNK)
593 				error = ncl_readlinkrpc(vp, &auio, ap->a_cred);
594 			else
595 				error = EACCES;
596 		} else
597 			NFSUNLOCKNODE(np);
598 		return (error);
599 	}
600 }
601 
602 /*
603  * nfs open vnode op
604  * Check to see if the type is ok
605  * and that deletion is not in progress.
606  * For paged in text files, you will need to flush the page cache
607  * if consistency is lost.
608  */
609 /* ARGSUSED */
610 static int
611 nfs_open(struct vop_open_args *ap)
612 {
613 	struct vnode *vp = ap->a_vp;
614 	struct nfsnode *np = VTONFS(vp);
615 	struct vattr vattr;
616 	int error;
617 	int fmode = ap->a_mode;
618 	struct ucred *cred;
619 	vm_object_t obj;
620 
621 	if (vp->v_type != VREG && vp->v_type != VDIR && vp->v_type != VLNK)
622 		return (EOPNOTSUPP);
623 
624 	/*
625 	 * For NFSv4, we need to do the Open Op before cache validation,
626 	 * so that we conform to RFC3530 Sec. 9.3.1.
627 	 */
628 	if (NFS_ISV4(vp)) {
629 		error = nfsrpc_open(vp, fmode, ap->a_cred, ap->a_td);
630 		if (error) {
631 			error = nfscl_maperr(ap->a_td, error, (uid_t)0,
632 			    (gid_t)0);
633 			return (error);
634 		}
635 	}
636 
637 	/*
638 	 * Now, if this Open will be doing reading, re-validate/flush the
639 	 * cache, so that Close/Open coherency is maintained.
640 	 */
641 	NFSLOCKNODE(np);
642 	if (np->n_flag & NMODIFIED) {
643 		NFSUNLOCKNODE(np);
644 		error = ncl_vinvalbuf(vp, V_SAVE, ap->a_td, 1);
645 		if (error == EINTR || error == EIO) {
646 			if (NFS_ISV4(vp))
647 				(void) nfsrpc_close(vp, 0, ap->a_td);
648 			return (error);
649 		}
650 		NFSLOCKNODE(np);
651 		np->n_attrstamp = 0;
652 		KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(vp);
653 		if (vp->v_type == VDIR)
654 			np->n_direofoffset = 0;
655 		NFSUNLOCKNODE(np);
656 		error = VOP_GETATTR(vp, &vattr, ap->a_cred);
657 		if (error) {
658 			if (NFS_ISV4(vp))
659 				(void) nfsrpc_close(vp, 0, ap->a_td);
660 			return (error);
661 		}
662 		NFSLOCKNODE(np);
663 		np->n_mtime = vattr.va_mtime;
664 		if (NFS_ISV4(vp))
665 			np->n_change = vattr.va_filerev;
666 	} else {
667 		NFSUNLOCKNODE(np);
668 		error = VOP_GETATTR(vp, &vattr, ap->a_cred);
669 		if (error) {
670 			if (NFS_ISV4(vp))
671 				(void) nfsrpc_close(vp, 0, ap->a_td);
672 			return (error);
673 		}
674 		NFSLOCKNODE(np);
675 		if ((NFS_ISV4(vp) && np->n_change != vattr.va_filerev) ||
676 		    NFS_TIMESPEC_COMPARE(&np->n_mtime, &vattr.va_mtime)) {
677 			if (vp->v_type == VDIR)
678 				np->n_direofoffset = 0;
679 			NFSUNLOCKNODE(np);
680 			error = ncl_vinvalbuf(vp, V_SAVE, ap->a_td, 1);
681 			if (error == EINTR || error == EIO) {
682 				if (NFS_ISV4(vp))
683 					(void) nfsrpc_close(vp, 0, ap->a_td);
684 				return (error);
685 			}
686 			NFSLOCKNODE(np);
687 			np->n_mtime = vattr.va_mtime;
688 			if (NFS_ISV4(vp))
689 				np->n_change = vattr.va_filerev;
690 		}
691 	}
692 
693 	/*
694 	 * If the object has >= 1 O_DIRECT active opens, we disable caching.
695 	 */
696 	if (newnfs_directio_enable && (fmode & O_DIRECT) &&
697 	    (vp->v_type == VREG)) {
698 		if (np->n_directio_opens == 0) {
699 			NFSUNLOCKNODE(np);
700 			error = ncl_vinvalbuf(vp, V_SAVE, ap->a_td, 1);
701 			if (error) {
702 				if (NFS_ISV4(vp))
703 					(void) nfsrpc_close(vp, 0, ap->a_td);
704 				return (error);
705 			}
706 			NFSLOCKNODE(np);
707 			np->n_flag |= NNONCACHE;
708 		}
709 		np->n_directio_opens++;
710 	}
711 
712 	/* If opened for writing via NFSv4.1 or later, mark that for pNFS. */
713 	if (NFSHASPNFS(VFSTONFS(vp->v_mount)) && (fmode & FWRITE) != 0)
714 		np->n_flag |= NWRITEOPENED;
715 
716 	/*
717 	 * If this is an open for writing, capture a reference to the
718 	 * credentials, so they can be used by ncl_putpages(). Using
719 	 * these write credentials is preferable to the credentials of
720 	 * whatever thread happens to be doing the VOP_PUTPAGES() since
721 	 * the write RPCs are less likely to fail with EACCES.
722 	 */
723 	if ((fmode & FWRITE) != 0) {
724 		cred = np->n_writecred;
725 		np->n_writecred = crhold(ap->a_cred);
726 	} else
727 		cred = NULL;
728 	NFSUNLOCKNODE(np);
729 
730 	if (cred != NULL)
731 		crfree(cred);
732 	vnode_create_vobject(vp, vattr.va_size, ap->a_td);
733 
734 	/*
735 	 * If the text file has been mmap'd, flush any dirty pages to the
736 	 * buffer cache and then...
737 	 * Make sure all writes are pushed to the NFS server.  If this is not
738 	 * done, the modify time of the file can change while the text
739 	 * file is being executed.  This will cause the process that is
740 	 * executing the text file to be terminated.
741 	 */
742 	if (vp->v_writecount <= -1) {
743 		if ((obj = vp->v_object) != NULL &&
744 		    vm_object_mightbedirty(obj)) {
745 			VM_OBJECT_WLOCK(obj);
746 			vm_object_page_clean(obj, 0, 0, OBJPC_SYNC);
747 			VM_OBJECT_WUNLOCK(obj);
748 		}
749 
750 		/* Now, flush the buffer cache. */
751 		ncl_flush(vp, MNT_WAIT, curthread, 0, 0);
752 
753 		/* And, finally, make sure that n_mtime is up to date. */
754 		np = VTONFS(vp);
755 		NFSLOCKNODE(np);
756 		np->n_mtime = np->n_vattr.na_mtime;
757 		NFSUNLOCKNODE(np);
758 	}
759 	return (0);
760 }
761 
762 /*
763  * nfs close vnode op
764  * What an NFS client should do upon close after writing is a debatable issue.
765  * Most NFS clients push delayed writes to the server upon close, basically for
766  * two reasons:
767  * 1 - So that any write errors may be reported back to the client process
768  *     doing the close system call. By far the two most likely errors are
769  *     NFSERR_NOSPC and NFSERR_DQUOT to indicate space allocation failure.
770  * 2 - To put a worst case upper bound on cache inconsistency between
771  *     multiple clients for the file.
772  * There is also a consistency problem for Version 2 of the protocol w.r.t.
773  * not being able to tell if other clients are writing a file concurrently,
774  * since there is no way of knowing if the changed modify time in the reply
775  * is only due to the write for this client.
776  * (NFS Version 3 provides weak cache consistency data in the reply that
777  *  should be sufficient to detect and handle this case.)
778  *
779  * The current code does the following:
780  * for NFS Version 2 - play it safe and flush/invalidate all dirty buffers
781  * for NFS Version 3 - flush dirty buffers to the server but don't invalidate
782  *                     or commit them (this satisfies 1 and 2 except for the
783  *                     case where the server crashes after this close but
784  *                     before the commit RPC, which is felt to be "good
785  *                     enough". Changing the last argument to ncl_flush() to
786  *                     a 1 would force a commit operation, if it is felt a
787  *                     commit is necessary now.
788  * for NFS Version 4 - flush the dirty buffers and commit them, if
789  *		       nfscl_mustflush() says this is necessary.
790  *                     It is necessary if there is no write delegation held,
791  *                     in order to satisfy open/close coherency.
792  *                     If the file isn't cached on local stable storage,
793  *                     it may be necessary in order to detect "out of space"
794  *                     errors from the server, if the write delegation
795  *                     issued by the server doesn't allow the file to grow.
796  */
797 /* ARGSUSED */
798 static int
799 nfs_close(struct vop_close_args *ap)
800 {
801 	struct vnode *vp = ap->a_vp;
802 	struct nfsnode *np = VTONFS(vp);
803 	struct nfsvattr nfsva;
804 	struct ucred *cred;
805 	int error = 0, ret, localcred = 0;
806 	int fmode = ap->a_fflag;
807 
808 	if (NFSCL_FORCEDISM(vp->v_mount))
809 		return (0);
810 	/*
811 	 * During shutdown, a_cred isn't valid, so just use root.
812 	 */
813 	if (ap->a_cred == NOCRED) {
814 		cred = newnfs_getcred();
815 		localcred = 1;
816 	} else {
817 		cred = ap->a_cred;
818 	}
819 	if (vp->v_type == VREG) {
820 	    /*
821 	     * Examine and clean dirty pages, regardless of NMODIFIED.
822 	     * This closes a major hole in close-to-open consistency.
823 	     * We want to push out all dirty pages (and buffers) on
824 	     * close, regardless of whether they were dirtied by
825 	     * mmap'ed writes or via write().
826 	     */
827 	    if (nfs_clean_pages_on_close && vp->v_object) {
828 		VM_OBJECT_WLOCK(vp->v_object);
829 		vm_object_page_clean(vp->v_object, 0, 0, 0);
830 		VM_OBJECT_WUNLOCK(vp->v_object);
831 	    }
832 	    NFSLOCKNODE(np);
833 	    if (np->n_flag & NMODIFIED) {
834 		NFSUNLOCKNODE(np);
835 		if (NFS_ISV3(vp)) {
836 		    /*
837 		     * Under NFSv3 we have dirty buffers to dispose of.  We
838 		     * must flush them to the NFS server.  We have the option
839 		     * of waiting all the way through the commit rpc or just
840 		     * waiting for the initial write.  The default is to only
841 		     * wait through the initial write so the data is in the
842 		     * server's cache, which is roughly similar to the state
843 		     * a standard disk subsystem leaves the file in on close().
844 		     *
845 		     * We cannot clear the NMODIFIED bit in np->n_flag due to
846 		     * potential races with other processes, and certainly
847 		     * cannot clear it if we don't commit.
848 		     * These races occur when there is no longer the old
849 		     * traditional vnode locking implemented for Vnode Ops.
850 		     */
851 		    int cm = newnfs_commit_on_close ? 1 : 0;
852 		    error = ncl_flush(vp, MNT_WAIT, ap->a_td, cm, 0);
853 		    /* np->n_flag &= ~NMODIFIED; */
854 		} else if (NFS_ISV4(vp)) {
855 			if (nfscl_mustflush(vp) != 0) {
856 				int cm = newnfs_commit_on_close ? 1 : 0;
857 				error = ncl_flush(vp, MNT_WAIT, ap->a_td,
858 				    cm, 0);
859 				/*
860 				 * as above w.r.t races when clearing
861 				 * NMODIFIED.
862 				 * np->n_flag &= ~NMODIFIED;
863 				 */
864 			}
865 		} else {
866 			error = ncl_vinvalbuf(vp, V_SAVE, ap->a_td, 1);
867 		}
868 		NFSLOCKNODE(np);
869 	    }
870  	    /*
871  	     * Invalidate the attribute cache in all cases.
872  	     * An open is going to fetch fresh attrs any way, other procs
873  	     * on this node that have file open will be forced to do an
874  	     * otw attr fetch, but this is safe.
875 	     * --> A user found that their RPC count dropped by 20% when
876 	     *     this was commented out and I can't see any requirement
877 	     *     for it, so I've disabled it when negative lookups are
878 	     *     enabled. (What does this have to do with negative lookup
879 	     *     caching? Well nothing, except it was reported by the
880 	     *     same user that needed negative lookup caching and I wanted
881 	     *     there to be a way to disable it to see if it
882 	     *     is the cause of some caching/coherency issue that might
883 	     *     crop up.)
884  	     */
885 	    if (VFSTONFS(vp->v_mount)->nm_negnametimeo == 0) {
886 		    np->n_attrstamp = 0;
887 		    KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(vp);
888 	    }
889 	    if (np->n_flag & NWRITEERR) {
890 		np->n_flag &= ~NWRITEERR;
891 		error = np->n_error;
892 	    }
893 	    NFSUNLOCKNODE(np);
894 	}
895 
896 	if (NFS_ISV4(vp)) {
897 		/*
898 		 * Get attributes so "change" is up to date.
899 		 */
900 		if (error == 0 && nfscl_mustflush(vp) != 0 &&
901 		    vp->v_type == VREG &&
902 		    (VFSTONFS(vp->v_mount)->nm_flag & NFSMNT_NOCTO) == 0) {
903 			ret = nfsrpc_getattr(vp, cred, ap->a_td, &nfsva,
904 			    NULL);
905 			if (!ret) {
906 				np->n_change = nfsva.na_filerev;
907 				(void) nfscl_loadattrcache(&vp, &nfsva, NULL,
908 				    NULL, 0, 0);
909 			}
910 		}
911 
912 		/*
913 		 * and do the close.
914 		 */
915 		ret = nfsrpc_close(vp, 0, ap->a_td);
916 		if (!error && ret)
917 			error = ret;
918 		if (error)
919 			error = nfscl_maperr(ap->a_td, error, (uid_t)0,
920 			    (gid_t)0);
921 	}
922 	if (newnfs_directio_enable)
923 		KASSERT((np->n_directio_asyncwr == 0),
924 			("nfs_close: dirty unflushed (%d) directio buffers\n",
925 			 np->n_directio_asyncwr));
926 	if (newnfs_directio_enable && (fmode & O_DIRECT) && (vp->v_type == VREG)) {
927 		NFSLOCKNODE(np);
928 		KASSERT((np->n_directio_opens > 0),
929 			("nfs_close: unexpectedly value (0) of n_directio_opens\n"));
930 		np->n_directio_opens--;
931 		if (np->n_directio_opens == 0)
932 			np->n_flag &= ~NNONCACHE;
933 		NFSUNLOCKNODE(np);
934 	}
935 	if (localcred)
936 		NFSFREECRED(cred);
937 	return (error);
938 }
939 
940 /*
941  * nfs getattr call from vfs.
942  */
943 static int
944 nfs_getattr(struct vop_getattr_args *ap)
945 {
946 	struct vnode *vp = ap->a_vp;
947 	struct thread *td = curthread;	/* XXX */
948 	struct nfsnode *np = VTONFS(vp);
949 	int error = 0;
950 	struct nfsvattr nfsva;
951 	struct vattr *vap = ap->a_vap;
952 	struct vattr vattr;
953 
954 	/*
955 	 * Update local times for special files.
956 	 */
957 	NFSLOCKNODE(np);
958 	if (np->n_flag & (NACC | NUPD))
959 		np->n_flag |= NCHG;
960 	NFSUNLOCKNODE(np);
961 	/*
962 	 * First look in the cache.
963 	 */
964 	if (ncl_getattrcache(vp, &vattr) == 0) {
965 		ncl_copy_vattr(vap, &vattr);
966 
967 		/*
968 		 * Get the local modify time for the case of a write
969 		 * delegation.
970 		 */
971 		nfscl_deleggetmodtime(vp, &vap->va_mtime);
972 		return (0);
973 	}
974 
975 	if (NFS_ISV34(vp) && nfs_prime_access_cache &&
976 	    nfsaccess_cache_timeout > 0) {
977 		NFSINCRGLOBAL(nfsstatsv1.accesscache_misses);
978 		nfs34_access_otw(vp, NFSACCESS_ALL, td, ap->a_cred, NULL);
979 		if (ncl_getattrcache(vp, ap->a_vap) == 0) {
980 			nfscl_deleggetmodtime(vp, &ap->a_vap->va_mtime);
981 			return (0);
982 		}
983 	}
984 	error = nfsrpc_getattr(vp, ap->a_cred, td, &nfsva, NULL);
985 	if (!error)
986 		error = nfscl_loadattrcache(&vp, &nfsva, vap, NULL, 0, 0);
987 	if (!error) {
988 		/*
989 		 * Get the local modify time for the case of a write
990 		 * delegation.
991 		 */
992 		nfscl_deleggetmodtime(vp, &vap->va_mtime);
993 	} else if (NFS_ISV4(vp)) {
994 		error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0);
995 	}
996 	return (error);
997 }
998 
999 /*
1000  * nfs setattr call.
1001  */
1002 static int
1003 nfs_setattr(struct vop_setattr_args *ap)
1004 {
1005 	struct vnode *vp = ap->a_vp;
1006 	struct nfsnode *np = VTONFS(vp);
1007 	struct thread *td = curthread;	/* XXX */
1008 	struct vattr *vap = ap->a_vap;
1009 	int error = 0;
1010 	u_quad_t tsize;
1011 
1012 #ifndef nolint
1013 	tsize = (u_quad_t)0;
1014 #endif
1015 
1016 	/*
1017 	 * Setting of flags and marking of atimes are not supported.
1018 	 */
1019 	if (vap->va_flags != VNOVAL)
1020 		return (EOPNOTSUPP);
1021 
1022 	/*
1023 	 * Disallow write attempts if the filesystem is mounted read-only.
1024 	 */
1025   	if ((vap->va_flags != VNOVAL || vap->va_uid != (uid_t)VNOVAL ||
1026 	    vap->va_gid != (gid_t)VNOVAL || vap->va_atime.tv_sec != VNOVAL ||
1027 	    vap->va_mtime.tv_sec != VNOVAL ||
1028 	    vap->va_birthtime.tv_sec != VNOVAL ||
1029 	    vap->va_mode != (mode_t)VNOVAL) &&
1030 	    (vp->v_mount->mnt_flag & MNT_RDONLY))
1031 		return (EROFS);
1032 	if (vap->va_size != VNOVAL) {
1033  		switch (vp->v_type) {
1034  		case VDIR:
1035  			return (EISDIR);
1036  		case VCHR:
1037  		case VBLK:
1038  		case VSOCK:
1039  		case VFIFO:
1040 			if (vap->va_mtime.tv_sec == VNOVAL &&
1041 			    vap->va_atime.tv_sec == VNOVAL &&
1042 			    vap->va_birthtime.tv_sec == VNOVAL &&
1043 			    vap->va_mode == (mode_t)VNOVAL &&
1044 			    vap->va_uid == (uid_t)VNOVAL &&
1045 			    vap->va_gid == (gid_t)VNOVAL)
1046 				return (0);
1047  			vap->va_size = VNOVAL;
1048  			break;
1049  		default:
1050 			/*
1051 			 * Disallow write attempts if the filesystem is
1052 			 * mounted read-only.
1053 			 */
1054 			if (vp->v_mount->mnt_flag & MNT_RDONLY)
1055 				return (EROFS);
1056 			/*
1057 			 *  We run vnode_pager_setsize() early (why?),
1058 			 * we must set np->n_size now to avoid vinvalbuf
1059 			 * V_SAVE races that might setsize a lower
1060 			 * value.
1061 			 */
1062 			NFSLOCKNODE(np);
1063 			tsize = np->n_size;
1064 			NFSUNLOCKNODE(np);
1065 			error = ncl_meta_setsize(vp, td, vap->va_size);
1066 			NFSLOCKNODE(np);
1067  			if (np->n_flag & NMODIFIED) {
1068 			    tsize = np->n_size;
1069 			    NFSUNLOCKNODE(np);
1070 			    error = ncl_vinvalbuf(vp, vap->va_size == 0 ?
1071 			        0 : V_SAVE, td, 1);
1072 			    if (error != 0) {
1073 				    vnode_pager_setsize(vp, tsize);
1074 				    return (error);
1075 			    }
1076 			    /*
1077 			     * Call nfscl_delegmodtime() to set the modify time
1078 			     * locally, as required.
1079 			     */
1080 			    nfscl_delegmodtime(vp);
1081  			} else
1082 			    NFSUNLOCKNODE(np);
1083 			/*
1084 			 * np->n_size has already been set to vap->va_size
1085 			 * in ncl_meta_setsize(). We must set it again since
1086 			 * nfs_loadattrcache() could be called through
1087 			 * ncl_meta_setsize() and could modify np->n_size.
1088 			 */
1089 			NFSLOCKNODE(np);
1090  			np->n_vattr.na_size = np->n_size = vap->va_size;
1091 			NFSUNLOCKNODE(np);
1092   		}
1093   	} else {
1094 		NFSLOCKNODE(np);
1095 		if ((vap->va_mtime.tv_sec != VNOVAL || vap->va_atime.tv_sec != VNOVAL) &&
1096 		    (np->n_flag & NMODIFIED) && vp->v_type == VREG) {
1097 			NFSUNLOCKNODE(np);
1098 			error = ncl_vinvalbuf(vp, V_SAVE, td, 1);
1099 			if (error == EINTR || error == EIO)
1100 				return (error);
1101 		} else
1102 			NFSUNLOCKNODE(np);
1103 	}
1104 	error = nfs_setattrrpc(vp, vap, ap->a_cred, td);
1105 	if (error && vap->va_size != VNOVAL) {
1106 		NFSLOCKNODE(np);
1107 		np->n_size = np->n_vattr.na_size = tsize;
1108 		vnode_pager_setsize(vp, tsize);
1109 		NFSUNLOCKNODE(np);
1110 	}
1111 	return (error);
1112 }
1113 
1114 /*
1115  * Do an nfs setattr rpc.
1116  */
1117 static int
1118 nfs_setattrrpc(struct vnode *vp, struct vattr *vap, struct ucred *cred,
1119     struct thread *td)
1120 {
1121 	struct nfsnode *np = VTONFS(vp);
1122 	int error, ret, attrflag, i;
1123 	struct nfsvattr nfsva;
1124 
1125 	if (NFS_ISV34(vp)) {
1126 		NFSLOCKNODE(np);
1127 		for (i = 0; i < NFS_ACCESSCACHESIZE; i++)
1128 			np->n_accesscache[i].stamp = 0;
1129 		np->n_flag |= NDELEGMOD;
1130 		NFSUNLOCKNODE(np);
1131 		KDTRACE_NFS_ACCESSCACHE_FLUSH_DONE(vp);
1132 	}
1133 	error = nfsrpc_setattr(vp, vap, NULL, cred, td, &nfsva, &attrflag,
1134 	    NULL);
1135 	if (attrflag) {
1136 		ret = nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0, 1);
1137 		if (ret && !error)
1138 			error = ret;
1139 	}
1140 	if (error && NFS_ISV4(vp))
1141 		error = nfscl_maperr(td, error, vap->va_uid, vap->va_gid);
1142 	return (error);
1143 }
1144 
1145 /*
1146  * nfs lookup call, one step at a time...
1147  * First look in cache
1148  * If not found, unlock the directory nfsnode and do the rpc
1149  */
1150 static int
1151 nfs_lookup(struct vop_lookup_args *ap)
1152 {
1153 	struct componentname *cnp = ap->a_cnp;
1154 	struct vnode *dvp = ap->a_dvp;
1155 	struct vnode **vpp = ap->a_vpp;
1156 	struct mount *mp = dvp->v_mount;
1157 	int flags = cnp->cn_flags;
1158 	struct vnode *newvp;
1159 	struct nfsmount *nmp;
1160 	struct nfsnode *np, *newnp;
1161 	int error = 0, attrflag, dattrflag, ltype, ncticks;
1162 	struct thread *td = cnp->cn_thread;
1163 	struct nfsfh *nfhp;
1164 	struct nfsvattr dnfsva, nfsva;
1165 	struct vattr vattr;
1166 	struct timespec nctime;
1167 
1168 	*vpp = NULLVP;
1169 	if ((flags & ISLASTCN) && (mp->mnt_flag & MNT_RDONLY) &&
1170 	    (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME))
1171 		return (EROFS);
1172 	if (dvp->v_type != VDIR)
1173 		return (ENOTDIR);
1174 	nmp = VFSTONFS(mp);
1175 	np = VTONFS(dvp);
1176 
1177 	/* For NFSv4, wait until any remove is done. */
1178 	NFSLOCKNODE(np);
1179 	while (NFSHASNFSV4(nmp) && (np->n_flag & NREMOVEINPROG)) {
1180 		np->n_flag |= NREMOVEWANT;
1181 		(void) msleep((caddr_t)np, &np->n_mtx, PZERO, "nfslkup", 0);
1182 	}
1183 	NFSUNLOCKNODE(np);
1184 
1185 	error = vn_dir_check_exec(dvp, cnp);
1186 	if (error != 0)
1187 		return (error);
1188 	error = cache_lookup(dvp, vpp, cnp, &nctime, &ncticks);
1189 	if (error > 0 && error != ENOENT)
1190 		return (error);
1191 	if (error == -1) {
1192 		/*
1193 		 * Lookups of "." are special and always return the
1194 		 * current directory.  cache_lookup() already handles
1195 		 * associated locking bookkeeping, etc.
1196 		 */
1197 		if (cnp->cn_namelen == 1 && cnp->cn_nameptr[0] == '.') {
1198 			/* XXX: Is this really correct? */
1199 			if (cnp->cn_nameiop != LOOKUP &&
1200 			    (flags & ISLASTCN))
1201 				cnp->cn_flags |= SAVENAME;
1202 			return (0);
1203 		}
1204 
1205 		/*
1206 		 * We only accept a positive hit in the cache if the
1207 		 * change time of the file matches our cached copy.
1208 		 * Otherwise, we discard the cache entry and fallback
1209 		 * to doing a lookup RPC.  We also only trust cache
1210 		 * entries for less than nm_nametimeo seconds.
1211 		 *
1212 		 * To better handle stale file handles and attributes,
1213 		 * clear the attribute cache of this node if it is a
1214 		 * leaf component, part of an open() call, and not
1215 		 * locally modified before fetching the attributes.
1216 		 * This should allow stale file handles to be detected
1217 		 * here where we can fall back to a LOOKUP RPC to
1218 		 * recover rather than having nfs_open() detect the
1219 		 * stale file handle and failing open(2) with ESTALE.
1220 		 */
1221 		newvp = *vpp;
1222 		newnp = VTONFS(newvp);
1223 		if (!(nmp->nm_flag & NFSMNT_NOCTO) &&
1224 		    (flags & (ISLASTCN | ISOPEN)) == (ISLASTCN | ISOPEN) &&
1225 		    !(newnp->n_flag & NMODIFIED)) {
1226 			NFSLOCKNODE(newnp);
1227 			newnp->n_attrstamp = 0;
1228 			KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(newvp);
1229 			NFSUNLOCKNODE(newnp);
1230 		}
1231 		if (nfscl_nodeleg(newvp, 0) == 0 ||
1232 		    ((u_int)(ticks - ncticks) < (nmp->nm_nametimeo * hz) &&
1233 		    VOP_GETATTR(newvp, &vattr, cnp->cn_cred) == 0 &&
1234 		    timespeccmp(&vattr.va_ctime, &nctime, ==))) {
1235 			NFSINCRGLOBAL(nfsstatsv1.lookupcache_hits);
1236 			if (cnp->cn_nameiop != LOOKUP &&
1237 			    (flags & ISLASTCN))
1238 				cnp->cn_flags |= SAVENAME;
1239 			return (0);
1240 		}
1241 		cache_purge(newvp);
1242 		if (dvp != newvp)
1243 			vput(newvp);
1244 		else
1245 			vrele(newvp);
1246 		*vpp = NULLVP;
1247 	} else if (error == ENOENT) {
1248 		if (VN_IS_DOOMED(dvp))
1249 			return (ENOENT);
1250 		/*
1251 		 * We only accept a negative hit in the cache if the
1252 		 * modification time of the parent directory matches
1253 		 * the cached copy in the name cache entry.
1254 		 * Otherwise, we discard all of the negative cache
1255 		 * entries for this directory.  We also only trust
1256 		 * negative cache entries for up to nm_negnametimeo
1257 		 * seconds.
1258 		 */
1259 		if ((u_int)(ticks - ncticks) < (nmp->nm_negnametimeo * hz) &&
1260 		    VOP_GETATTR(dvp, &vattr, cnp->cn_cred) == 0 &&
1261 		    timespeccmp(&vattr.va_mtime, &nctime, ==)) {
1262 			NFSINCRGLOBAL(nfsstatsv1.lookupcache_hits);
1263 			return (ENOENT);
1264 		}
1265 		cache_purge_negative(dvp);
1266 	}
1267 
1268 	newvp = NULLVP;
1269 	NFSINCRGLOBAL(nfsstatsv1.lookupcache_misses);
1270 	error = nfsrpc_lookup(dvp, cnp->cn_nameptr, cnp->cn_namelen,
1271 	    cnp->cn_cred, td, &dnfsva, &nfsva, &nfhp, &attrflag, &dattrflag,
1272 	    NULL);
1273 	if (dattrflag)
1274 		(void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, NULL, 0, 1);
1275 	if (error) {
1276 		if (newvp != NULLVP) {
1277 			vput(newvp);
1278 			*vpp = NULLVP;
1279 		}
1280 
1281 		if (error != ENOENT) {
1282 			if (NFS_ISV4(dvp))
1283 				error = nfscl_maperr(td, error, (uid_t)0,
1284 				    (gid_t)0);
1285 			return (error);
1286 		}
1287 
1288 		/* The requested file was not found. */
1289 		if ((cnp->cn_nameiop == CREATE || cnp->cn_nameiop == RENAME) &&
1290 		    (flags & ISLASTCN)) {
1291 			/*
1292 			 * XXX: UFS does a full VOP_ACCESS(dvp,
1293 			 * VWRITE) here instead of just checking
1294 			 * MNT_RDONLY.
1295 			 */
1296 			if (mp->mnt_flag & MNT_RDONLY)
1297 				return (EROFS);
1298 			cnp->cn_flags |= SAVENAME;
1299 			return (EJUSTRETURN);
1300 		}
1301 
1302 		if ((cnp->cn_flags & MAKEENTRY) != 0 && dattrflag) {
1303 			/*
1304 			 * Cache the modification time of the parent
1305 			 * directory from the post-op attributes in
1306 			 * the name cache entry.  The negative cache
1307 			 * entry will be ignored once the directory
1308 			 * has changed.  Don't bother adding the entry
1309 			 * if the directory has already changed.
1310 			 */
1311 			NFSLOCKNODE(np);
1312 			if (timespeccmp(&np->n_vattr.na_mtime,
1313 			    &dnfsva.na_mtime, ==)) {
1314 				NFSUNLOCKNODE(np);
1315 				cache_enter_time(dvp, NULL, cnp,
1316 				    &dnfsva.na_mtime, NULL);
1317 			} else
1318 				NFSUNLOCKNODE(np);
1319 		}
1320 		return (ENOENT);
1321 	}
1322 
1323 	/*
1324 	 * Handle RENAME case...
1325 	 */
1326 	if (cnp->cn_nameiop == RENAME && (flags & ISLASTCN)) {
1327 		if (NFS_CMPFH(np, nfhp->nfh_fh, nfhp->nfh_len)) {
1328 			free(nfhp, M_NFSFH);
1329 			return (EISDIR);
1330 		}
1331 		error = nfscl_nget(mp, dvp, nfhp, cnp, td, &np, NULL,
1332 		    LK_EXCLUSIVE);
1333 		if (error)
1334 			return (error);
1335 		newvp = NFSTOV(np);
1336 		if (attrflag)
1337 			(void) nfscl_loadattrcache(&newvp, &nfsva, NULL, NULL,
1338 			    0, 1);
1339 		*vpp = newvp;
1340 		cnp->cn_flags |= SAVENAME;
1341 		return (0);
1342 	}
1343 
1344 	if (flags & ISDOTDOT) {
1345 		ltype = NFSVOPISLOCKED(dvp);
1346 		error = vfs_busy(mp, MBF_NOWAIT);
1347 		if (error != 0) {
1348 			vfs_ref(mp);
1349 			NFSVOPUNLOCK(dvp);
1350 			error = vfs_busy(mp, 0);
1351 			NFSVOPLOCK(dvp, ltype | LK_RETRY);
1352 			vfs_rel(mp);
1353 			if (error == 0 && VN_IS_DOOMED(dvp)) {
1354 				vfs_unbusy(mp);
1355 				error = ENOENT;
1356 			}
1357 			if (error != 0)
1358 				return (error);
1359 		}
1360 		NFSVOPUNLOCK(dvp);
1361 		error = nfscl_nget(mp, dvp, nfhp, cnp, td, &np, NULL,
1362 		    cnp->cn_lkflags);
1363 		if (error == 0)
1364 			newvp = NFSTOV(np);
1365 		vfs_unbusy(mp);
1366 		if (newvp != dvp)
1367 			NFSVOPLOCK(dvp, ltype | LK_RETRY);
1368 		if (VN_IS_DOOMED(dvp)) {
1369 			if (error == 0) {
1370 				if (newvp == dvp)
1371 					vrele(newvp);
1372 				else
1373 					vput(newvp);
1374 			}
1375 			error = ENOENT;
1376 		}
1377 		if (error != 0)
1378 			return (error);
1379 		if (attrflag)
1380 			(void) nfscl_loadattrcache(&newvp, &nfsva, NULL, NULL,
1381 			    0, 1);
1382 	} else if (NFS_CMPFH(np, nfhp->nfh_fh, nfhp->nfh_len)) {
1383 		free(nfhp, M_NFSFH);
1384 		VREF(dvp);
1385 		newvp = dvp;
1386 		if (attrflag)
1387 			(void) nfscl_loadattrcache(&newvp, &nfsva, NULL, NULL,
1388 			    0, 1);
1389 	} else {
1390 		error = nfscl_nget(mp, dvp, nfhp, cnp, td, &np, NULL,
1391 		    cnp->cn_lkflags);
1392 		if (error)
1393 			return (error);
1394 		newvp = NFSTOV(np);
1395 		if (attrflag)
1396 			(void) nfscl_loadattrcache(&newvp, &nfsva, NULL, NULL,
1397 			    0, 1);
1398 		else if ((flags & (ISLASTCN | ISOPEN)) == (ISLASTCN | ISOPEN) &&
1399 		    !(np->n_flag & NMODIFIED)) {
1400 			/*
1401 			 * Flush the attribute cache when opening a
1402 			 * leaf node to ensure that fresh attributes
1403 			 * are fetched in nfs_open() since we did not
1404 			 * fetch attributes from the LOOKUP reply.
1405 			 */
1406 			NFSLOCKNODE(np);
1407 			np->n_attrstamp = 0;
1408 			KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(newvp);
1409 			NFSUNLOCKNODE(np);
1410 		}
1411 	}
1412 	if (cnp->cn_nameiop != LOOKUP && (flags & ISLASTCN))
1413 		cnp->cn_flags |= SAVENAME;
1414 	if ((cnp->cn_flags & MAKEENTRY) && dvp != newvp &&
1415 	    (cnp->cn_nameiop != DELETE || !(flags & ISLASTCN)) &&
1416 	    attrflag != 0 && (newvp->v_type != VDIR || dattrflag != 0))
1417 		cache_enter_time(dvp, newvp, cnp, &nfsva.na_ctime,
1418 		    newvp->v_type != VDIR ? NULL : &dnfsva.na_ctime);
1419 	*vpp = newvp;
1420 	return (0);
1421 }
1422 
1423 /*
1424  * nfs read call.
1425  * Just call ncl_bioread() to do the work.
1426  */
1427 static int
1428 nfs_read(struct vop_read_args *ap)
1429 {
1430 	struct vnode *vp = ap->a_vp;
1431 
1432 	switch (vp->v_type) {
1433 	case VREG:
1434 		return (ncl_bioread(vp, ap->a_uio, ap->a_ioflag, ap->a_cred));
1435 	case VDIR:
1436 		return (EISDIR);
1437 	default:
1438 		return (EOPNOTSUPP);
1439 	}
1440 }
1441 
1442 /*
1443  * nfs readlink call
1444  */
1445 static int
1446 nfs_readlink(struct vop_readlink_args *ap)
1447 {
1448 	struct vnode *vp = ap->a_vp;
1449 
1450 	if (vp->v_type != VLNK)
1451 		return (EINVAL);
1452 	return (ncl_bioread(vp, ap->a_uio, 0, ap->a_cred));
1453 }
1454 
1455 /*
1456  * Do a readlink rpc.
1457  * Called by ncl_doio() from below the buffer cache.
1458  */
1459 int
1460 ncl_readlinkrpc(struct vnode *vp, struct uio *uiop, struct ucred *cred)
1461 {
1462 	int error, ret, attrflag;
1463 	struct nfsvattr nfsva;
1464 
1465 	error = nfsrpc_readlink(vp, uiop, cred, uiop->uio_td, &nfsva,
1466 	    &attrflag, NULL);
1467 	if (attrflag) {
1468 		ret = nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0, 1);
1469 		if (ret && !error)
1470 			error = ret;
1471 	}
1472 	if (error && NFS_ISV4(vp))
1473 		error = nfscl_maperr(uiop->uio_td, error, (uid_t)0, (gid_t)0);
1474 	return (error);
1475 }
1476 
1477 /*
1478  * nfs read rpc call
1479  * Ditto above
1480  */
1481 int
1482 ncl_readrpc(struct vnode *vp, struct uio *uiop, struct ucred *cred)
1483 {
1484 	int error, ret, attrflag;
1485 	struct nfsvattr nfsva;
1486 	struct nfsmount *nmp;
1487 
1488 	nmp = VFSTONFS(vp->v_mount);
1489 	error = EIO;
1490 	attrflag = 0;
1491 	if (NFSHASPNFS(nmp))
1492 		error = nfscl_doiods(vp, uiop, NULL, NULL,
1493 		    NFSV4OPEN_ACCESSREAD, 0, cred, uiop->uio_td);
1494 	NFSCL_DEBUG(4, "readrpc: aft doiods=%d\n", error);
1495 	if (error != 0)
1496 		error = nfsrpc_read(vp, uiop, cred, uiop->uio_td, &nfsva,
1497 		    &attrflag, NULL);
1498 	if (attrflag) {
1499 		ret = nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0, 1);
1500 		if (ret && !error)
1501 			error = ret;
1502 	}
1503 	if (error && NFS_ISV4(vp))
1504 		error = nfscl_maperr(uiop->uio_td, error, (uid_t)0, (gid_t)0);
1505 	return (error);
1506 }
1507 
1508 /*
1509  * nfs write call
1510  */
1511 int
1512 ncl_writerpc(struct vnode *vp, struct uio *uiop, struct ucred *cred,
1513     int *iomode, int *must_commit, int called_from_strategy)
1514 {
1515 	struct nfsvattr nfsva;
1516 	int error, attrflag, ret;
1517 	struct nfsmount *nmp;
1518 
1519 	nmp = VFSTONFS(vp->v_mount);
1520 	error = EIO;
1521 	attrflag = 0;
1522 	if (NFSHASPNFS(nmp))
1523 		error = nfscl_doiods(vp, uiop, iomode, must_commit,
1524 		    NFSV4OPEN_ACCESSWRITE, 0, cred, uiop->uio_td);
1525 	NFSCL_DEBUG(4, "writerpc: aft doiods=%d\n", error);
1526 	if (error != 0)
1527 		error = nfsrpc_write(vp, uiop, iomode, must_commit, cred,
1528 		    uiop->uio_td, &nfsva, &attrflag, NULL,
1529 		    called_from_strategy);
1530 	if (attrflag) {
1531 		if (VTONFS(vp)->n_flag & ND_NFSV4)
1532 			ret = nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 1,
1533 			    1);
1534 		else
1535 			ret = nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0,
1536 			    1);
1537 		if (ret && !error)
1538 			error = ret;
1539 	}
1540 	if (DOINGASYNC(vp))
1541 		*iomode = NFSWRITE_FILESYNC;
1542 	if (error && NFS_ISV4(vp))
1543 		error = nfscl_maperr(uiop->uio_td, error, (uid_t)0, (gid_t)0);
1544 	return (error);
1545 }
1546 
1547 /*
1548  * nfs mknod rpc
1549  * For NFS v2 this is a kludge. Use a create rpc but with the IFMT bits of the
1550  * mode set to specify the file type and the size field for rdev.
1551  */
1552 static int
1553 nfs_mknodrpc(struct vnode *dvp, struct vnode **vpp, struct componentname *cnp,
1554     struct vattr *vap)
1555 {
1556 	struct nfsvattr nfsva, dnfsva;
1557 	struct vnode *newvp = NULL;
1558 	struct nfsnode *np = NULL, *dnp;
1559 	struct nfsfh *nfhp;
1560 	struct vattr vattr;
1561 	int error = 0, attrflag, dattrflag;
1562 	u_int32_t rdev;
1563 
1564 	if (vap->va_type == VCHR || vap->va_type == VBLK)
1565 		rdev = vap->va_rdev;
1566 	else if (vap->va_type == VFIFO || vap->va_type == VSOCK)
1567 		rdev = 0xffffffff;
1568 	else
1569 		return (EOPNOTSUPP);
1570 	if ((error = VOP_GETATTR(dvp, &vattr, cnp->cn_cred)))
1571 		return (error);
1572 	error = nfsrpc_mknod(dvp, cnp->cn_nameptr, cnp->cn_namelen, vap,
1573 	    rdev, vap->va_type, cnp->cn_cred, cnp->cn_thread, &dnfsva,
1574 	    &nfsva, &nfhp, &attrflag, &dattrflag, NULL);
1575 	if (!error) {
1576 		if (!nfhp)
1577 			(void) nfsrpc_lookup(dvp, cnp->cn_nameptr,
1578 			    cnp->cn_namelen, cnp->cn_cred, cnp->cn_thread,
1579 			    &dnfsva, &nfsva, &nfhp, &attrflag, &dattrflag,
1580 			    NULL);
1581 		if (nfhp)
1582 			error = nfscl_nget(dvp->v_mount, dvp, nfhp, cnp,
1583 			    cnp->cn_thread, &np, NULL, LK_EXCLUSIVE);
1584 	}
1585 	if (dattrflag)
1586 		(void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, NULL, 0, 1);
1587 	if (!error) {
1588 		newvp = NFSTOV(np);
1589 		if (attrflag != 0) {
1590 			error = nfscl_loadattrcache(&newvp, &nfsva, NULL, NULL,
1591 			    0, 1);
1592 			if (error != 0)
1593 				vput(newvp);
1594 		}
1595 	}
1596 	if (!error) {
1597 		*vpp = newvp;
1598 	} else if (NFS_ISV4(dvp)) {
1599 		error = nfscl_maperr(cnp->cn_thread, error, vap->va_uid,
1600 		    vap->va_gid);
1601 	}
1602 	dnp = VTONFS(dvp);
1603 	NFSLOCKNODE(dnp);
1604 	dnp->n_flag |= NMODIFIED;
1605 	if (!dattrflag) {
1606 		dnp->n_attrstamp = 0;
1607 		KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(dvp);
1608 	}
1609 	NFSUNLOCKNODE(dnp);
1610 	return (error);
1611 }
1612 
1613 /*
1614  * nfs mknod vop
1615  * just call nfs_mknodrpc() to do the work.
1616  */
1617 /* ARGSUSED */
1618 static int
1619 nfs_mknod(struct vop_mknod_args *ap)
1620 {
1621 	return (nfs_mknodrpc(ap->a_dvp, ap->a_vpp, ap->a_cnp, ap->a_vap));
1622 }
1623 
1624 static struct mtx nfs_cverf_mtx;
1625 MTX_SYSINIT(nfs_cverf_mtx, &nfs_cverf_mtx, "NFS create verifier mutex",
1626     MTX_DEF);
1627 
1628 static nfsquad_t
1629 nfs_get_cverf(void)
1630 {
1631 	static nfsquad_t cverf;
1632 	nfsquad_t ret;
1633 	static int cverf_initialized = 0;
1634 
1635 	mtx_lock(&nfs_cverf_mtx);
1636 	if (cverf_initialized == 0) {
1637 		cverf.lval[0] = arc4random();
1638 		cverf.lval[1] = arc4random();
1639 		cverf_initialized = 1;
1640 	} else
1641 		cverf.qval++;
1642 	ret = cverf;
1643 	mtx_unlock(&nfs_cverf_mtx);
1644 
1645 	return (ret);
1646 }
1647 
1648 /*
1649  * nfs file create call
1650  */
1651 static int
1652 nfs_create(struct vop_create_args *ap)
1653 {
1654 	struct vnode *dvp = ap->a_dvp;
1655 	struct vattr *vap = ap->a_vap;
1656 	struct componentname *cnp = ap->a_cnp;
1657 	struct nfsnode *np = NULL, *dnp;
1658 	struct vnode *newvp = NULL;
1659 	struct nfsmount *nmp;
1660 	struct nfsvattr dnfsva, nfsva;
1661 	struct nfsfh *nfhp;
1662 	nfsquad_t cverf;
1663 	int error = 0, attrflag, dattrflag, fmode = 0;
1664 	struct vattr vattr;
1665 
1666 	/*
1667 	 * Oops, not for me..
1668 	 */
1669 	if (vap->va_type == VSOCK)
1670 		return (nfs_mknodrpc(dvp, ap->a_vpp, cnp, vap));
1671 
1672 	if ((error = VOP_GETATTR(dvp, &vattr, cnp->cn_cred)))
1673 		return (error);
1674 	if (vap->va_vaflags & VA_EXCLUSIVE)
1675 		fmode |= O_EXCL;
1676 	dnp = VTONFS(dvp);
1677 	nmp = VFSTONFS(dvp->v_mount);
1678 again:
1679 	/* For NFSv4, wait until any remove is done. */
1680 	NFSLOCKNODE(dnp);
1681 	while (NFSHASNFSV4(nmp) && (dnp->n_flag & NREMOVEINPROG)) {
1682 		dnp->n_flag |= NREMOVEWANT;
1683 		(void) msleep((caddr_t)dnp, &dnp->n_mtx, PZERO, "nfscrt", 0);
1684 	}
1685 	NFSUNLOCKNODE(dnp);
1686 
1687 	cverf = nfs_get_cverf();
1688 	error = nfsrpc_create(dvp, cnp->cn_nameptr, cnp->cn_namelen,
1689 	    vap, cverf, fmode, cnp->cn_cred, cnp->cn_thread, &dnfsva, &nfsva,
1690 	    &nfhp, &attrflag, &dattrflag, NULL);
1691 	if (!error) {
1692 		if (nfhp == NULL)
1693 			(void) nfsrpc_lookup(dvp, cnp->cn_nameptr,
1694 			    cnp->cn_namelen, cnp->cn_cred, cnp->cn_thread,
1695 			    &dnfsva, &nfsva, &nfhp, &attrflag, &dattrflag,
1696 			    NULL);
1697 		if (nfhp != NULL)
1698 			error = nfscl_nget(dvp->v_mount, dvp, nfhp, cnp,
1699 			    cnp->cn_thread, &np, NULL, LK_EXCLUSIVE);
1700 	}
1701 	if (dattrflag)
1702 		(void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, NULL, 0, 1);
1703 	if (!error) {
1704 		newvp = NFSTOV(np);
1705 		if (attrflag == 0)
1706 			error = nfsrpc_getattr(newvp, cnp->cn_cred,
1707 			    cnp->cn_thread, &nfsva, NULL);
1708 		if (error == 0)
1709 			error = nfscl_loadattrcache(&newvp, &nfsva, NULL, NULL,
1710 			    0, 1);
1711 	}
1712 	if (error) {
1713 		if (newvp != NULL) {
1714 			vput(newvp);
1715 			newvp = NULL;
1716 		}
1717 		if (NFS_ISV34(dvp) && (fmode & O_EXCL) &&
1718 		    error == NFSERR_NOTSUPP) {
1719 			fmode &= ~O_EXCL;
1720 			goto again;
1721 		}
1722 	} else if (NFS_ISV34(dvp) && (fmode & O_EXCL)) {
1723 		if (nfscl_checksattr(vap, &nfsva)) {
1724 			error = nfsrpc_setattr(newvp, vap, NULL, cnp->cn_cred,
1725 			    cnp->cn_thread, &nfsva, &attrflag, NULL);
1726 			if (error && (vap->va_uid != (uid_t)VNOVAL ||
1727 			    vap->va_gid != (gid_t)VNOVAL)) {
1728 				/* try again without setting uid/gid */
1729 				vap->va_uid = (uid_t)VNOVAL;
1730 				vap->va_gid = (uid_t)VNOVAL;
1731 				error = nfsrpc_setattr(newvp, vap, NULL,
1732 				    cnp->cn_cred, cnp->cn_thread, &nfsva,
1733 				    &attrflag, NULL);
1734 			}
1735 			if (attrflag)
1736 				(void) nfscl_loadattrcache(&newvp, &nfsva, NULL,
1737 				    NULL, 0, 1);
1738 			if (error != 0)
1739 				vput(newvp);
1740 		}
1741 	}
1742 	if (!error) {
1743 		if ((cnp->cn_flags & MAKEENTRY) && attrflag) {
1744 			if (dvp != newvp)
1745 				cache_enter_time(dvp, newvp, cnp,
1746 				    &nfsva.na_ctime, NULL);
1747 			else
1748 				printf("nfs_create: bogus NFS server returned "
1749 				    "the directory as the new file object\n");
1750 		}
1751 		*ap->a_vpp = newvp;
1752 	} else if (NFS_ISV4(dvp)) {
1753 		error = nfscl_maperr(cnp->cn_thread, error, vap->va_uid,
1754 		    vap->va_gid);
1755 	}
1756 	NFSLOCKNODE(dnp);
1757 	dnp->n_flag |= NMODIFIED;
1758 	if (!dattrflag) {
1759 		dnp->n_attrstamp = 0;
1760 		KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(dvp);
1761 	}
1762 	NFSUNLOCKNODE(dnp);
1763 	return (error);
1764 }
1765 
1766 /*
1767  * nfs file remove call
1768  * To try and make nfs semantics closer to ufs semantics, a file that has
1769  * other processes using the vnode is renamed instead of removed and then
1770  * removed later on the last close.
1771  * - If v_usecount > 1
1772  *	  If a rename is not already in the works
1773  *	     call nfs_sillyrename() to set it up
1774  *     else
1775  *	  do the remove rpc
1776  */
1777 static int
1778 nfs_remove(struct vop_remove_args *ap)
1779 {
1780 	struct vnode *vp = ap->a_vp;
1781 	struct vnode *dvp = ap->a_dvp;
1782 	struct componentname *cnp = ap->a_cnp;
1783 	struct nfsnode *np = VTONFS(vp);
1784 	int error = 0;
1785 	struct vattr vattr;
1786 
1787 	KASSERT((cnp->cn_flags & HASBUF) != 0, ("nfs_remove: no name"));
1788 	KASSERT(vrefcnt(vp) > 0, ("nfs_remove: bad v_usecount"));
1789 	if (vp->v_type == VDIR)
1790 		error = EPERM;
1791 	else if (vrefcnt(vp) == 1 || (np->n_sillyrename &&
1792 	    VOP_GETATTR(vp, &vattr, cnp->cn_cred) == 0 &&
1793 	    vattr.va_nlink > 1)) {
1794 		/*
1795 		 * Purge the name cache so that the chance of a lookup for
1796 		 * the name succeeding while the remove is in progress is
1797 		 * minimized. Without node locking it can still happen, such
1798 		 * that an I/O op returns ESTALE, but since you get this if
1799 		 * another host removes the file..
1800 		 */
1801 		cache_purge(vp);
1802 		/*
1803 		 * throw away biocache buffers, mainly to avoid
1804 		 * unnecessary delayed writes later.
1805 		 */
1806 		error = ncl_vinvalbuf(vp, 0, cnp->cn_thread, 1);
1807 		if (error != EINTR && error != EIO)
1808 			/* Do the rpc */
1809 			error = nfs_removerpc(dvp, vp, cnp->cn_nameptr,
1810 			    cnp->cn_namelen, cnp->cn_cred, cnp->cn_thread);
1811 		/*
1812 		 * Kludge City: If the first reply to the remove rpc is lost..
1813 		 *   the reply to the retransmitted request will be ENOENT
1814 		 *   since the file was in fact removed
1815 		 *   Therefore, we cheat and return success.
1816 		 */
1817 		if (error == ENOENT)
1818 			error = 0;
1819 	} else if (!np->n_sillyrename)
1820 		error = nfs_sillyrename(dvp, vp, cnp);
1821 	NFSLOCKNODE(np);
1822 	np->n_attrstamp = 0;
1823 	NFSUNLOCKNODE(np);
1824 	KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(vp);
1825 	return (error);
1826 }
1827 
1828 /*
1829  * nfs file remove rpc called from nfs_inactive
1830  */
1831 int
1832 ncl_removeit(struct sillyrename *sp, struct vnode *vp)
1833 {
1834 	/*
1835 	 * Make sure that the directory vnode is still valid.
1836 	 * XXX we should lock sp->s_dvp here.
1837 	 */
1838 	if (sp->s_dvp->v_type == VBAD)
1839 		return (0);
1840 	return (nfs_removerpc(sp->s_dvp, vp, sp->s_name, sp->s_namlen,
1841 	    sp->s_cred, NULL));
1842 }
1843 
1844 /*
1845  * Nfs remove rpc, called from nfs_remove() and ncl_removeit().
1846  */
1847 static int
1848 nfs_removerpc(struct vnode *dvp, struct vnode *vp, char *name,
1849     int namelen, struct ucred *cred, struct thread *td)
1850 {
1851 	struct nfsvattr dnfsva;
1852 	struct nfsnode *dnp = VTONFS(dvp);
1853 	int error = 0, dattrflag;
1854 
1855 	NFSLOCKNODE(dnp);
1856 	dnp->n_flag |= NREMOVEINPROG;
1857 	NFSUNLOCKNODE(dnp);
1858 	error = nfsrpc_remove(dvp, name, namelen, vp, cred, td, &dnfsva,
1859 	    &dattrflag, NULL);
1860 	NFSLOCKNODE(dnp);
1861 	if ((dnp->n_flag & NREMOVEWANT)) {
1862 		dnp->n_flag &= ~(NREMOVEWANT | NREMOVEINPROG);
1863 		NFSUNLOCKNODE(dnp);
1864 		wakeup((caddr_t)dnp);
1865 	} else {
1866 		dnp->n_flag &= ~NREMOVEINPROG;
1867 		NFSUNLOCKNODE(dnp);
1868 	}
1869 	if (dattrflag)
1870 		(void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, NULL, 0, 1);
1871 	NFSLOCKNODE(dnp);
1872 	dnp->n_flag |= NMODIFIED;
1873 	if (!dattrflag) {
1874 		dnp->n_attrstamp = 0;
1875 		KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(dvp);
1876 	}
1877 	NFSUNLOCKNODE(dnp);
1878 	if (error && NFS_ISV4(dvp))
1879 		error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0);
1880 	return (error);
1881 }
1882 
1883 /*
1884  * nfs file rename call
1885  */
1886 static int
1887 nfs_rename(struct vop_rename_args *ap)
1888 {
1889 	struct vnode *fvp = ap->a_fvp;
1890 	struct vnode *tvp = ap->a_tvp;
1891 	struct vnode *fdvp = ap->a_fdvp;
1892 	struct vnode *tdvp = ap->a_tdvp;
1893 	struct componentname *tcnp = ap->a_tcnp;
1894 	struct componentname *fcnp = ap->a_fcnp;
1895 	struct nfsnode *fnp = VTONFS(ap->a_fvp);
1896 	struct nfsnode *tdnp = VTONFS(ap->a_tdvp);
1897 	struct nfsv4node *newv4 = NULL;
1898 	int error;
1899 
1900 	KASSERT((tcnp->cn_flags & HASBUF) != 0 &&
1901 	    (fcnp->cn_flags & HASBUF) != 0, ("nfs_rename: no name"));
1902 	/* Check for cross-device rename */
1903 	if ((fvp->v_mount != tdvp->v_mount) ||
1904 	    (tvp && (fvp->v_mount != tvp->v_mount))) {
1905 		error = EXDEV;
1906 		goto out;
1907 	}
1908 
1909 	if (fvp == tvp) {
1910 		printf("nfs_rename: fvp == tvp (can't happen)\n");
1911 		error = 0;
1912 		goto out;
1913 	}
1914 	if ((error = NFSVOPLOCK(fvp, LK_EXCLUSIVE)) != 0)
1915 		goto out;
1916 
1917 	/*
1918 	 * We have to flush B_DELWRI data prior to renaming
1919 	 * the file.  If we don't, the delayed-write buffers
1920 	 * can be flushed out later after the file has gone stale
1921 	 * under NFSV3.  NFSV2 does not have this problem because
1922 	 * ( as far as I can tell ) it flushes dirty buffers more
1923 	 * often.
1924 	 *
1925 	 * Skip the rename operation if the fsync fails, this can happen
1926 	 * due to the server's volume being full, when we pushed out data
1927 	 * that was written back to our cache earlier. Not checking for
1928 	 * this condition can result in potential (silent) data loss.
1929 	 */
1930 	error = VOP_FSYNC(fvp, MNT_WAIT, fcnp->cn_thread);
1931 	NFSVOPUNLOCK(fvp);
1932 	if (!error && tvp)
1933 		error = VOP_FSYNC(tvp, MNT_WAIT, tcnp->cn_thread);
1934 	if (error)
1935 		goto out;
1936 
1937 	/*
1938 	 * If the tvp exists and is in use, sillyrename it before doing the
1939 	 * rename of the new file over it.
1940 	 * XXX Can't sillyrename a directory.
1941 	 */
1942 	if (tvp && vrefcnt(tvp) > 1 && !VTONFS(tvp)->n_sillyrename &&
1943 		tvp->v_type != VDIR && !nfs_sillyrename(tdvp, tvp, tcnp)) {
1944 		vput(tvp);
1945 		tvp = NULL;
1946 	}
1947 
1948 	error = nfs_renamerpc(fdvp, fvp, fcnp->cn_nameptr, fcnp->cn_namelen,
1949 	    tdvp, tvp, tcnp->cn_nameptr, tcnp->cn_namelen, tcnp->cn_cred,
1950 	    tcnp->cn_thread);
1951 
1952 	if (error == 0 && NFS_ISV4(tdvp)) {
1953 		/*
1954 		 * For NFSv4, check to see if it is the same name and
1955 		 * replace the name, if it is different.
1956 		 */
1957 		newv4 = malloc(
1958 		    sizeof (struct nfsv4node) +
1959 		    tdnp->n_fhp->nfh_len + tcnp->cn_namelen - 1,
1960 		    M_NFSV4NODE, M_WAITOK);
1961 		NFSLOCKNODE(tdnp);
1962 		NFSLOCKNODE(fnp);
1963 		if (fnp->n_v4 != NULL && fvp->v_type == VREG &&
1964 		    (fnp->n_v4->n4_namelen != tcnp->cn_namelen ||
1965 		      NFSBCMP(tcnp->cn_nameptr, NFS4NODENAME(fnp->n_v4),
1966 		      tcnp->cn_namelen) ||
1967 		      tdnp->n_fhp->nfh_len != fnp->n_v4->n4_fhlen ||
1968 		      NFSBCMP(tdnp->n_fhp->nfh_fh, fnp->n_v4->n4_data,
1969 			tdnp->n_fhp->nfh_len))) {
1970 #ifdef notdef
1971 { char nnn[100]; int nnnl;
1972 nnnl = (tcnp->cn_namelen < 100) ? tcnp->cn_namelen : 99;
1973 bcopy(tcnp->cn_nameptr, nnn, nnnl);
1974 nnn[nnnl] = '\0';
1975 printf("ren replace=%s\n",nnn);
1976 }
1977 #endif
1978 			free(fnp->n_v4, M_NFSV4NODE);
1979 			fnp->n_v4 = newv4;
1980 			newv4 = NULL;
1981 			fnp->n_v4->n4_fhlen = tdnp->n_fhp->nfh_len;
1982 			fnp->n_v4->n4_namelen = tcnp->cn_namelen;
1983 			NFSBCOPY(tdnp->n_fhp->nfh_fh, fnp->n_v4->n4_data,
1984 			    tdnp->n_fhp->nfh_len);
1985 			NFSBCOPY(tcnp->cn_nameptr,
1986 			    NFS4NODENAME(fnp->n_v4), tcnp->cn_namelen);
1987 		}
1988 		NFSUNLOCKNODE(tdnp);
1989 		NFSUNLOCKNODE(fnp);
1990 		if (newv4 != NULL)
1991 			free(newv4, M_NFSV4NODE);
1992 	}
1993 
1994 	if (fvp->v_type == VDIR) {
1995 		if (tvp != NULL && tvp->v_type == VDIR)
1996 			cache_purge(tdvp);
1997 		cache_purge(fdvp);
1998 	}
1999 
2000 out:
2001 	if (tdvp == tvp)
2002 		vrele(tdvp);
2003 	else
2004 		vput(tdvp);
2005 	if (tvp)
2006 		vput(tvp);
2007 	vrele(fdvp);
2008 	vrele(fvp);
2009 	/*
2010 	 * Kludge: Map ENOENT => 0 assuming that it is a reply to a retry.
2011 	 */
2012 	if (error == ENOENT)
2013 		error = 0;
2014 	return (error);
2015 }
2016 
2017 /*
2018  * nfs file rename rpc called from nfs_remove() above
2019  */
2020 static int
2021 nfs_renameit(struct vnode *sdvp, struct vnode *svp, struct componentname *scnp,
2022     struct sillyrename *sp)
2023 {
2024 
2025 	return (nfs_renamerpc(sdvp, svp, scnp->cn_nameptr, scnp->cn_namelen,
2026 	    sdvp, NULL, sp->s_name, sp->s_namlen, scnp->cn_cred,
2027 	    scnp->cn_thread));
2028 }
2029 
2030 /*
2031  * Do an nfs rename rpc. Called from nfs_rename() and nfs_renameit().
2032  */
2033 static int
2034 nfs_renamerpc(struct vnode *fdvp, struct vnode *fvp, char *fnameptr,
2035     int fnamelen, struct vnode *tdvp, struct vnode *tvp, char *tnameptr,
2036     int tnamelen, struct ucred *cred, struct thread *td)
2037 {
2038 	struct nfsvattr fnfsva, tnfsva;
2039 	struct nfsnode *fdnp = VTONFS(fdvp);
2040 	struct nfsnode *tdnp = VTONFS(tdvp);
2041 	int error = 0, fattrflag, tattrflag;
2042 
2043 	error = nfsrpc_rename(fdvp, fvp, fnameptr, fnamelen, tdvp, tvp,
2044 	    tnameptr, tnamelen, cred, td, &fnfsva, &tnfsva, &fattrflag,
2045 	    &tattrflag, NULL, NULL);
2046 	NFSLOCKNODE(fdnp);
2047 	fdnp->n_flag |= NMODIFIED;
2048 	if (fattrflag != 0) {
2049 		NFSUNLOCKNODE(fdnp);
2050 		(void) nfscl_loadattrcache(&fdvp, &fnfsva, NULL, NULL, 0, 1);
2051 	} else {
2052 		fdnp->n_attrstamp = 0;
2053 		NFSUNLOCKNODE(fdnp);
2054 		KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(fdvp);
2055 	}
2056 	NFSLOCKNODE(tdnp);
2057 	tdnp->n_flag |= NMODIFIED;
2058 	if (tattrflag != 0) {
2059 		NFSUNLOCKNODE(tdnp);
2060 		(void) nfscl_loadattrcache(&tdvp, &tnfsva, NULL, NULL, 0, 1);
2061 	} else {
2062 		tdnp->n_attrstamp = 0;
2063 		NFSUNLOCKNODE(tdnp);
2064 		KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(tdvp);
2065 	}
2066 	if (error && NFS_ISV4(fdvp))
2067 		error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0);
2068 	return (error);
2069 }
2070 
2071 /*
2072  * nfs hard link create call
2073  */
2074 static int
2075 nfs_link(struct vop_link_args *ap)
2076 {
2077 	struct vnode *vp = ap->a_vp;
2078 	struct vnode *tdvp = ap->a_tdvp;
2079 	struct componentname *cnp = ap->a_cnp;
2080 	struct nfsnode *np, *tdnp;
2081 	struct nfsvattr nfsva, dnfsva;
2082 	int error = 0, attrflag, dattrflag;
2083 
2084 	/*
2085 	 * Push all writes to the server, so that the attribute cache
2086 	 * doesn't get "out of sync" with the server.
2087 	 * XXX There should be a better way!
2088 	 */
2089 	VOP_FSYNC(vp, MNT_WAIT, cnp->cn_thread);
2090 
2091 	error = nfsrpc_link(tdvp, vp, cnp->cn_nameptr, cnp->cn_namelen,
2092 	    cnp->cn_cred, cnp->cn_thread, &dnfsva, &nfsva, &attrflag,
2093 	    &dattrflag, NULL);
2094 	tdnp = VTONFS(tdvp);
2095 	NFSLOCKNODE(tdnp);
2096 	tdnp->n_flag |= NMODIFIED;
2097 	if (dattrflag != 0) {
2098 		NFSUNLOCKNODE(tdnp);
2099 		(void) nfscl_loadattrcache(&tdvp, &dnfsva, NULL, NULL, 0, 1);
2100 	} else {
2101 		tdnp->n_attrstamp = 0;
2102 		NFSUNLOCKNODE(tdnp);
2103 		KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(tdvp);
2104 	}
2105 	if (attrflag)
2106 		(void) nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0, 1);
2107 	else {
2108 		np = VTONFS(vp);
2109 		NFSLOCKNODE(np);
2110 		np->n_attrstamp = 0;
2111 		NFSUNLOCKNODE(np);
2112 		KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(vp);
2113 	}
2114 	/*
2115 	 * If negative lookup caching is enabled, I might as well
2116 	 * add an entry for this node. Not necessary for correctness,
2117 	 * but if negative caching is enabled, then the system
2118 	 * must care about lookup caching hit rate, so...
2119 	 */
2120 	if (VFSTONFS(vp->v_mount)->nm_negnametimeo != 0 &&
2121 	    (cnp->cn_flags & MAKEENTRY) && attrflag != 0 && error == 0) {
2122 		if (tdvp != vp)
2123 			cache_enter_time(tdvp, vp, cnp, &nfsva.na_ctime, NULL);
2124 		else
2125 			printf("nfs_link: bogus NFS server returned "
2126 			    "the directory as the new link\n");
2127 	}
2128 	if (error && NFS_ISV4(vp))
2129 		error = nfscl_maperr(cnp->cn_thread, error, (uid_t)0,
2130 		    (gid_t)0);
2131 	return (error);
2132 }
2133 
2134 /*
2135  * nfs symbolic link create call
2136  */
2137 static int
2138 nfs_symlink(struct vop_symlink_args *ap)
2139 {
2140 	struct vnode *dvp = ap->a_dvp;
2141 	struct vattr *vap = ap->a_vap;
2142 	struct componentname *cnp = ap->a_cnp;
2143 	struct nfsvattr nfsva, dnfsva;
2144 	struct nfsfh *nfhp;
2145 	struct nfsnode *np = NULL, *dnp;
2146 	struct vnode *newvp = NULL;
2147 	int error = 0, attrflag, dattrflag, ret;
2148 
2149 	vap->va_type = VLNK;
2150 	error = nfsrpc_symlink(dvp, cnp->cn_nameptr, cnp->cn_namelen,
2151 	    ap->a_target, vap, cnp->cn_cred, cnp->cn_thread, &dnfsva,
2152 	    &nfsva, &nfhp, &attrflag, &dattrflag, NULL);
2153 	if (nfhp) {
2154 		ret = nfscl_nget(dvp->v_mount, dvp, nfhp, cnp, cnp->cn_thread,
2155 		    &np, NULL, LK_EXCLUSIVE);
2156 		if (!ret)
2157 			newvp = NFSTOV(np);
2158 		else if (!error)
2159 			error = ret;
2160 	}
2161 	if (newvp != NULL) {
2162 		if (attrflag)
2163 			(void) nfscl_loadattrcache(&newvp, &nfsva, NULL, NULL,
2164 			    0, 1);
2165 	} else if (!error) {
2166 		/*
2167 		 * If we do not have an error and we could not extract the
2168 		 * newvp from the response due to the request being NFSv2, we
2169 		 * have to do a lookup in order to obtain a newvp to return.
2170 		 */
2171 		error = nfs_lookitup(dvp, cnp->cn_nameptr, cnp->cn_namelen,
2172 		    cnp->cn_cred, cnp->cn_thread, &np);
2173 		if (!error)
2174 			newvp = NFSTOV(np);
2175 	}
2176 	if (error) {
2177 		if (newvp)
2178 			vput(newvp);
2179 		if (NFS_ISV4(dvp))
2180 			error = nfscl_maperr(cnp->cn_thread, error,
2181 			    vap->va_uid, vap->va_gid);
2182 	} else {
2183 		*ap->a_vpp = newvp;
2184 	}
2185 
2186 	dnp = VTONFS(dvp);
2187 	NFSLOCKNODE(dnp);
2188 	dnp->n_flag |= NMODIFIED;
2189 	if (dattrflag != 0) {
2190 		NFSUNLOCKNODE(dnp);
2191 		(void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, NULL, 0, 1);
2192 	} else {
2193 		dnp->n_attrstamp = 0;
2194 		NFSUNLOCKNODE(dnp);
2195 		KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(dvp);
2196 	}
2197 	/*
2198 	 * If negative lookup caching is enabled, I might as well
2199 	 * add an entry for this node. Not necessary for correctness,
2200 	 * but if negative caching is enabled, then the system
2201 	 * must care about lookup caching hit rate, so...
2202 	 */
2203 	if (VFSTONFS(dvp->v_mount)->nm_negnametimeo != 0 &&
2204 	    (cnp->cn_flags & MAKEENTRY) && attrflag != 0 && error == 0) {
2205 		if (dvp != newvp)
2206 			cache_enter_time(dvp, newvp, cnp, &nfsva.na_ctime,
2207 			    NULL);
2208 		else
2209 			printf("nfs_symlink: bogus NFS server returned "
2210 			    "the directory as the new file object\n");
2211 	}
2212 	return (error);
2213 }
2214 
2215 /*
2216  * nfs make dir call
2217  */
2218 static int
2219 nfs_mkdir(struct vop_mkdir_args *ap)
2220 {
2221 	struct vnode *dvp = ap->a_dvp;
2222 	struct vattr *vap = ap->a_vap;
2223 	struct componentname *cnp = ap->a_cnp;
2224 	struct nfsnode *np = NULL, *dnp;
2225 	struct vnode *newvp = NULL;
2226 	struct vattr vattr;
2227 	struct nfsfh *nfhp;
2228 	struct nfsvattr nfsva, dnfsva;
2229 	int error = 0, attrflag, dattrflag, ret;
2230 
2231 	if ((error = VOP_GETATTR(dvp, &vattr, cnp->cn_cred)) != 0)
2232 		return (error);
2233 	vap->va_type = VDIR;
2234 	error = nfsrpc_mkdir(dvp, cnp->cn_nameptr, cnp->cn_namelen,
2235 	    vap, cnp->cn_cred, cnp->cn_thread, &dnfsva, &nfsva, &nfhp,
2236 	    &attrflag, &dattrflag, NULL);
2237 	dnp = VTONFS(dvp);
2238 	NFSLOCKNODE(dnp);
2239 	dnp->n_flag |= NMODIFIED;
2240 	if (dattrflag != 0) {
2241 		NFSUNLOCKNODE(dnp);
2242 		(void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, NULL, 0, 1);
2243 	} else {
2244 		dnp->n_attrstamp = 0;
2245 		NFSUNLOCKNODE(dnp);
2246 		KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(dvp);
2247 	}
2248 	if (nfhp) {
2249 		ret = nfscl_nget(dvp->v_mount, dvp, nfhp, cnp, cnp->cn_thread,
2250 		    &np, NULL, LK_EXCLUSIVE);
2251 		if (!ret) {
2252 			newvp = NFSTOV(np);
2253 			if (attrflag)
2254 			   (void) nfscl_loadattrcache(&newvp, &nfsva, NULL,
2255 				NULL, 0, 1);
2256 		} else if (!error)
2257 			error = ret;
2258 	}
2259 	if (!error && newvp == NULL) {
2260 		error = nfs_lookitup(dvp, cnp->cn_nameptr, cnp->cn_namelen,
2261 		    cnp->cn_cred, cnp->cn_thread, &np);
2262 		if (!error) {
2263 			newvp = NFSTOV(np);
2264 			if (newvp->v_type != VDIR)
2265 				error = EEXIST;
2266 		}
2267 	}
2268 	if (error) {
2269 		if (newvp)
2270 			vput(newvp);
2271 		if (NFS_ISV4(dvp))
2272 			error = nfscl_maperr(cnp->cn_thread, error,
2273 			    vap->va_uid, vap->va_gid);
2274 	} else {
2275 		/*
2276 		 * If negative lookup caching is enabled, I might as well
2277 		 * add an entry for this node. Not necessary for correctness,
2278 		 * but if negative caching is enabled, then the system
2279 		 * must care about lookup caching hit rate, so...
2280 		 */
2281 		if (VFSTONFS(dvp->v_mount)->nm_negnametimeo != 0 &&
2282 		    (cnp->cn_flags & MAKEENTRY) &&
2283 		    attrflag != 0 && dattrflag != 0) {
2284 			if (dvp != newvp)
2285 				cache_enter_time(dvp, newvp, cnp,
2286 				    &nfsva.na_ctime, &dnfsva.na_ctime);
2287 			else
2288 				printf("nfs_mkdir: bogus NFS server returned "
2289 				    "the directory that the directory was "
2290 				    "created in as the new file object\n");
2291 		}
2292 		*ap->a_vpp = newvp;
2293 	}
2294 	return (error);
2295 }
2296 
2297 /*
2298  * nfs remove directory call
2299  */
2300 static int
2301 nfs_rmdir(struct vop_rmdir_args *ap)
2302 {
2303 	struct vnode *vp = ap->a_vp;
2304 	struct vnode *dvp = ap->a_dvp;
2305 	struct componentname *cnp = ap->a_cnp;
2306 	struct nfsnode *dnp;
2307 	struct nfsvattr dnfsva;
2308 	int error, dattrflag;
2309 
2310 	if (dvp == vp)
2311 		return (EINVAL);
2312 	error = nfsrpc_rmdir(dvp, cnp->cn_nameptr, cnp->cn_namelen,
2313 	    cnp->cn_cred, cnp->cn_thread, &dnfsva, &dattrflag, NULL);
2314 	dnp = VTONFS(dvp);
2315 	NFSLOCKNODE(dnp);
2316 	dnp->n_flag |= NMODIFIED;
2317 	if (dattrflag != 0) {
2318 		NFSUNLOCKNODE(dnp);
2319 		(void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, NULL, 0, 1);
2320 	} else {
2321 		dnp->n_attrstamp = 0;
2322 		NFSUNLOCKNODE(dnp);
2323 		KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(dvp);
2324 	}
2325 
2326 	cache_purge(dvp);
2327 	cache_purge(vp);
2328 	if (error && NFS_ISV4(dvp))
2329 		error = nfscl_maperr(cnp->cn_thread, error, (uid_t)0,
2330 		    (gid_t)0);
2331 	/*
2332 	 * Kludge: Map ENOENT => 0 assuming that you have a reply to a retry.
2333 	 */
2334 	if (error == ENOENT)
2335 		error = 0;
2336 	return (error);
2337 }
2338 
2339 /*
2340  * nfs readdir call
2341  */
2342 static int
2343 nfs_readdir(struct vop_readdir_args *ap)
2344 {
2345 	struct vnode *vp = ap->a_vp;
2346 	struct nfsnode *np = VTONFS(vp);
2347 	struct uio *uio = ap->a_uio;
2348 	ssize_t tresid, left;
2349 	int error = 0;
2350 	struct vattr vattr;
2351 
2352 	if (ap->a_eofflag != NULL)
2353 		*ap->a_eofflag = 0;
2354 	if (vp->v_type != VDIR)
2355 		return(EPERM);
2356 
2357 	/*
2358 	 * First, check for hit on the EOF offset cache
2359 	 */
2360 	NFSLOCKNODE(np);
2361 	if (np->n_direofoffset > 0 && uio->uio_offset >= np->n_direofoffset &&
2362 	    (np->n_flag & NMODIFIED) == 0) {
2363 		NFSUNLOCKNODE(np);
2364 		if (VOP_GETATTR(vp, &vattr, ap->a_cred) == 0) {
2365 			NFSLOCKNODE(np);
2366 			if ((NFS_ISV4(vp) && np->n_change == vattr.va_filerev) ||
2367 			    !NFS_TIMESPEC_COMPARE(&np->n_mtime, &vattr.va_mtime)) {
2368 				NFSUNLOCKNODE(np);
2369 				NFSINCRGLOBAL(nfsstatsv1.direofcache_hits);
2370 				if (ap->a_eofflag != NULL)
2371 					*ap->a_eofflag = 1;
2372 				return (0);
2373 			} else
2374 				NFSUNLOCKNODE(np);
2375 		}
2376 	} else
2377 		NFSUNLOCKNODE(np);
2378 
2379 	/*
2380 	 * NFS always guarantees that directory entries don't straddle
2381 	 * DIRBLKSIZ boundaries.  As such, we need to limit the size
2382 	 * to an exact multiple of DIRBLKSIZ, to avoid copying a partial
2383 	 * directory entry.
2384 	 */
2385 	left = uio->uio_resid % DIRBLKSIZ;
2386 	if (left == uio->uio_resid)
2387 		return (EINVAL);
2388 	uio->uio_resid -= left;
2389 
2390 	/*
2391 	 * Call ncl_bioread() to do the real work.
2392 	 */
2393 	tresid = uio->uio_resid;
2394 	error = ncl_bioread(vp, uio, 0, ap->a_cred);
2395 
2396 	if (!error && uio->uio_resid == tresid) {
2397 		NFSINCRGLOBAL(nfsstatsv1.direofcache_misses);
2398 		if (ap->a_eofflag != NULL)
2399 			*ap->a_eofflag = 1;
2400 	}
2401 
2402 	/* Add the partial DIRBLKSIZ (left) back in. */
2403 	uio->uio_resid += left;
2404 	return (error);
2405 }
2406 
2407 /*
2408  * Readdir rpc call.
2409  * Called from below the buffer cache by ncl_doio().
2410  */
2411 int
2412 ncl_readdirrpc(struct vnode *vp, struct uio *uiop, struct ucred *cred,
2413     struct thread *td)
2414 {
2415 	struct nfsvattr nfsva;
2416 	nfsuint64 *cookiep, cookie;
2417 	struct nfsnode *dnp = VTONFS(vp);
2418 	struct nfsmount *nmp = VFSTONFS(vp->v_mount);
2419 	int error = 0, eof, attrflag;
2420 
2421 	KASSERT(uiop->uio_iovcnt == 1 &&
2422 	    (uiop->uio_offset & (DIRBLKSIZ - 1)) == 0 &&
2423 	    (uiop->uio_resid & (DIRBLKSIZ - 1)) == 0,
2424 	    ("nfs readdirrpc bad uio"));
2425 
2426 	/*
2427 	 * If there is no cookie, assume directory was stale.
2428 	 */
2429 	ncl_dircookie_lock(dnp);
2430 	NFSUNLOCKNODE(dnp);
2431 	cookiep = ncl_getcookie(dnp, uiop->uio_offset, 0);
2432 	if (cookiep) {
2433 		cookie = *cookiep;
2434 		ncl_dircookie_unlock(dnp);
2435 	} else {
2436 		ncl_dircookie_unlock(dnp);
2437 		return (NFSERR_BAD_COOKIE);
2438 	}
2439 
2440 	if (NFSHASNFSV3(nmp) && !NFSHASGOTFSINFO(nmp))
2441 		(void)ncl_fsinfo(nmp, vp, cred, td);
2442 
2443 	error = nfsrpc_readdir(vp, uiop, &cookie, cred, td, &nfsva,
2444 	    &attrflag, &eof, NULL);
2445 	if (attrflag)
2446 		(void) nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0, 1);
2447 
2448 	if (!error) {
2449 		/*
2450 		 * We are now either at the end of the directory or have filled
2451 		 * the block.
2452 		 */
2453 		if (eof) {
2454 			NFSLOCKNODE(dnp);
2455 			dnp->n_direofoffset = uiop->uio_offset;
2456 			NFSUNLOCKNODE(dnp);
2457 		} else {
2458 			if (uiop->uio_resid > 0)
2459 				printf("EEK! readdirrpc resid > 0\n");
2460 			ncl_dircookie_lock(dnp);
2461 			NFSUNLOCKNODE(dnp);
2462 			cookiep = ncl_getcookie(dnp, uiop->uio_offset, 1);
2463 			*cookiep = cookie;
2464 			ncl_dircookie_unlock(dnp);
2465 		}
2466 	} else if (NFS_ISV4(vp)) {
2467 		error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0);
2468 	}
2469 	return (error);
2470 }
2471 
2472 /*
2473  * NFS V3 readdir plus RPC. Used in place of ncl_readdirrpc().
2474  */
2475 int
2476 ncl_readdirplusrpc(struct vnode *vp, struct uio *uiop, struct ucred *cred,
2477     struct thread *td)
2478 {
2479 	struct nfsvattr nfsva;
2480 	nfsuint64 *cookiep, cookie;
2481 	struct nfsnode *dnp = VTONFS(vp);
2482 	struct nfsmount *nmp = VFSTONFS(vp->v_mount);
2483 	int error = 0, attrflag, eof;
2484 
2485 	KASSERT(uiop->uio_iovcnt == 1 &&
2486 	    (uiop->uio_offset & (DIRBLKSIZ - 1)) == 0 &&
2487 	    (uiop->uio_resid & (DIRBLKSIZ - 1)) == 0,
2488 	    ("nfs readdirplusrpc bad uio"));
2489 
2490 	/*
2491 	 * If there is no cookie, assume directory was stale.
2492 	 */
2493 	ncl_dircookie_lock(dnp);
2494 	NFSUNLOCKNODE(dnp);
2495 	cookiep = ncl_getcookie(dnp, uiop->uio_offset, 0);
2496 	if (cookiep) {
2497 		cookie = *cookiep;
2498 		ncl_dircookie_unlock(dnp);
2499 	} else {
2500 		ncl_dircookie_unlock(dnp);
2501 		return (NFSERR_BAD_COOKIE);
2502 	}
2503 
2504 	if (NFSHASNFSV3(nmp) && !NFSHASGOTFSINFO(nmp))
2505 		(void)ncl_fsinfo(nmp, vp, cred, td);
2506 	error = nfsrpc_readdirplus(vp, uiop, &cookie, cred, td, &nfsva,
2507 	    &attrflag, &eof, NULL);
2508 	if (attrflag)
2509 		(void) nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0, 1);
2510 
2511 	if (!error) {
2512 		/*
2513 		 * We are now either at end of the directory or have filled the
2514 		 * the block.
2515 		 */
2516 		if (eof) {
2517 			NFSLOCKNODE(dnp);
2518 			dnp->n_direofoffset = uiop->uio_offset;
2519 			NFSUNLOCKNODE(dnp);
2520 		} else {
2521 			if (uiop->uio_resid > 0)
2522 				printf("EEK! readdirplusrpc resid > 0\n");
2523 			ncl_dircookie_lock(dnp);
2524 			NFSUNLOCKNODE(dnp);
2525 			cookiep = ncl_getcookie(dnp, uiop->uio_offset, 1);
2526 			*cookiep = cookie;
2527 			ncl_dircookie_unlock(dnp);
2528 		}
2529 	} else if (NFS_ISV4(vp)) {
2530 		error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0);
2531 	}
2532 	return (error);
2533 }
2534 
2535 /*
2536  * Silly rename. To make the NFS filesystem that is stateless look a little
2537  * more like the "ufs" a remove of an active vnode is translated to a rename
2538  * to a funny looking filename that is removed by nfs_inactive on the
2539  * nfsnode. There is the potential for another process on a different client
2540  * to create the same funny name between the nfs_lookitup() fails and the
2541  * nfs_rename() completes, but...
2542  */
2543 static int
2544 nfs_sillyrename(struct vnode *dvp, struct vnode *vp, struct componentname *cnp)
2545 {
2546 	struct sillyrename *sp;
2547 	struct nfsnode *np;
2548 	int error;
2549 	short pid;
2550 	unsigned int lticks;
2551 
2552 	cache_purge(dvp);
2553 	np = VTONFS(vp);
2554 	KASSERT(vp->v_type != VDIR, ("nfs: sillyrename dir"));
2555 	sp = malloc(sizeof (struct sillyrename),
2556 	    M_NEWNFSREQ, M_WAITOK);
2557 	sp->s_cred = crhold(cnp->cn_cred);
2558 	sp->s_dvp = dvp;
2559 	VREF(dvp);
2560 
2561 	/*
2562 	 * Fudge together a funny name.
2563 	 * Changing the format of the funny name to accommodate more
2564 	 * sillynames per directory.
2565 	 * The name is now changed to .nfs.<ticks>.<pid>.4, where ticks is
2566 	 * CPU ticks since boot.
2567 	 */
2568 	pid = cnp->cn_thread->td_proc->p_pid;
2569 	lticks = (unsigned int)ticks;
2570 	for ( ; ; ) {
2571 		sp->s_namlen = sprintf(sp->s_name,
2572 				       ".nfs.%08x.%04x4.4", lticks,
2573 				       pid);
2574 		if (nfs_lookitup(dvp, sp->s_name, sp->s_namlen, sp->s_cred,
2575 				 cnp->cn_thread, NULL))
2576 			break;
2577 		lticks++;
2578 	}
2579 	error = nfs_renameit(dvp, vp, cnp, sp);
2580 	if (error)
2581 		goto bad;
2582 	error = nfs_lookitup(dvp, sp->s_name, sp->s_namlen, sp->s_cred,
2583 		cnp->cn_thread, &np);
2584 	np->n_sillyrename = sp;
2585 	return (0);
2586 bad:
2587 	vrele(sp->s_dvp);
2588 	crfree(sp->s_cred);
2589 	free(sp, M_NEWNFSREQ);
2590 	return (error);
2591 }
2592 
2593 /*
2594  * Look up a file name and optionally either update the file handle or
2595  * allocate an nfsnode, depending on the value of npp.
2596  * npp == NULL	--> just do the lookup
2597  * *npp == NULL --> allocate a new nfsnode and make sure attributes are
2598  *			handled too
2599  * *npp != NULL --> update the file handle in the vnode
2600  */
2601 static int
2602 nfs_lookitup(struct vnode *dvp, char *name, int len, struct ucred *cred,
2603     struct thread *td, struct nfsnode **npp)
2604 {
2605 	struct vnode *newvp = NULL, *vp;
2606 	struct nfsnode *np, *dnp = VTONFS(dvp);
2607 	struct nfsfh *nfhp, *onfhp;
2608 	struct nfsvattr nfsva, dnfsva;
2609 	struct componentname cn;
2610 	int error = 0, attrflag, dattrflag;
2611 	u_int hash;
2612 
2613 	error = nfsrpc_lookup(dvp, name, len, cred, td, &dnfsva, &nfsva,
2614 	    &nfhp, &attrflag, &dattrflag, NULL);
2615 	if (dattrflag)
2616 		(void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, NULL, 0, 1);
2617 	if (npp && !error) {
2618 		if (*npp != NULL) {
2619 		    np = *npp;
2620 		    vp = NFSTOV(np);
2621 		    /*
2622 		     * For NFSv4, check to see if it is the same name and
2623 		     * replace the name, if it is different.
2624 		     */
2625 		    if (np->n_v4 != NULL && nfsva.na_type == VREG &&
2626 			(np->n_v4->n4_namelen != len ||
2627 			 NFSBCMP(name, NFS4NODENAME(np->n_v4), len) ||
2628 			 dnp->n_fhp->nfh_len != np->n_v4->n4_fhlen ||
2629 			 NFSBCMP(dnp->n_fhp->nfh_fh, np->n_v4->n4_data,
2630 			 dnp->n_fhp->nfh_len))) {
2631 #ifdef notdef
2632 { char nnn[100]; int nnnl;
2633 nnnl = (len < 100) ? len : 99;
2634 bcopy(name, nnn, nnnl);
2635 nnn[nnnl] = '\0';
2636 printf("replace=%s\n",nnn);
2637 }
2638 #endif
2639 			    free(np->n_v4, M_NFSV4NODE);
2640 			    np->n_v4 = malloc(
2641 				sizeof (struct nfsv4node) +
2642 				dnp->n_fhp->nfh_len + len - 1,
2643 				M_NFSV4NODE, M_WAITOK);
2644 			    np->n_v4->n4_fhlen = dnp->n_fhp->nfh_len;
2645 			    np->n_v4->n4_namelen = len;
2646 			    NFSBCOPY(dnp->n_fhp->nfh_fh, np->n_v4->n4_data,
2647 				dnp->n_fhp->nfh_len);
2648 			    NFSBCOPY(name, NFS4NODENAME(np->n_v4), len);
2649 		    }
2650 		    hash = fnv_32_buf(nfhp->nfh_fh, nfhp->nfh_len,
2651 			FNV1_32_INIT);
2652 		    onfhp = np->n_fhp;
2653 		    /*
2654 		     * Rehash node for new file handle.
2655 		     */
2656 		    vfs_hash_rehash(vp, hash);
2657 		    np->n_fhp = nfhp;
2658 		    if (onfhp != NULL)
2659 			free(onfhp, M_NFSFH);
2660 		    newvp = NFSTOV(np);
2661 		} else if (NFS_CMPFH(dnp, nfhp->nfh_fh, nfhp->nfh_len)) {
2662 		    free(nfhp, M_NFSFH);
2663 		    VREF(dvp);
2664 		    newvp = dvp;
2665 		} else {
2666 		    cn.cn_nameptr = name;
2667 		    cn.cn_namelen = len;
2668 		    error = nfscl_nget(dvp->v_mount, dvp, nfhp, &cn, td,
2669 			&np, NULL, LK_EXCLUSIVE);
2670 		    if (error)
2671 			return (error);
2672 		    newvp = NFSTOV(np);
2673 		}
2674 		if (!attrflag && *npp == NULL) {
2675 			if (newvp == dvp)
2676 				vrele(newvp);
2677 			else
2678 				vput(newvp);
2679 			return (ENOENT);
2680 		}
2681 		if (attrflag)
2682 			(void) nfscl_loadattrcache(&newvp, &nfsva, NULL, NULL,
2683 			    0, 1);
2684 	}
2685 	if (npp && *npp == NULL) {
2686 		if (error) {
2687 			if (newvp) {
2688 				if (newvp == dvp)
2689 					vrele(newvp);
2690 				else
2691 					vput(newvp);
2692 			}
2693 		} else
2694 			*npp = np;
2695 	}
2696 	if (error && NFS_ISV4(dvp))
2697 		error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0);
2698 	return (error);
2699 }
2700 
2701 /*
2702  * Nfs Version 3 and 4 commit rpc
2703  */
2704 int
2705 ncl_commit(struct vnode *vp, u_quad_t offset, int cnt, struct ucred *cred,
2706    struct thread *td)
2707 {
2708 	struct nfsvattr nfsva;
2709 	struct nfsmount *nmp = VFSTONFS(vp->v_mount);
2710 	struct nfsnode *np;
2711 	struct uio uio;
2712 	int error, attrflag;
2713 
2714 	np = VTONFS(vp);
2715 	error = EIO;
2716 	attrflag = 0;
2717 	if (NFSHASPNFS(nmp) && (np->n_flag & NDSCOMMIT) != 0) {
2718 		uio.uio_offset = offset;
2719 		uio.uio_resid = cnt;
2720 		error = nfscl_doiods(vp, &uio, NULL, NULL,
2721 		    NFSV4OPEN_ACCESSWRITE, 1, cred, td);
2722 		if (error != 0) {
2723 			NFSLOCKNODE(np);
2724 			np->n_flag &= ~NDSCOMMIT;
2725 			NFSUNLOCKNODE(np);
2726 		}
2727 	}
2728 	if (error != 0) {
2729 		mtx_lock(&nmp->nm_mtx);
2730 		if ((nmp->nm_state & NFSSTA_HASWRITEVERF) == 0) {
2731 			mtx_unlock(&nmp->nm_mtx);
2732 			return (0);
2733 		}
2734 		mtx_unlock(&nmp->nm_mtx);
2735 		error = nfsrpc_commit(vp, offset, cnt, cred, td, &nfsva,
2736 		    &attrflag, NULL);
2737 	}
2738 	if (attrflag != 0)
2739 		(void) nfscl_loadattrcache(&vp, &nfsva, NULL, NULL,
2740 		    0, 1);
2741 	if (error != 0 && NFS_ISV4(vp))
2742 		error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0);
2743 	return (error);
2744 }
2745 
2746 /*
2747  * Strategy routine.
2748  * For async requests when nfsiod(s) are running, queue the request by
2749  * calling ncl_asyncio(), otherwise just all ncl_doio() to do the
2750  * request.
2751  */
2752 static int
2753 nfs_strategy(struct vop_strategy_args *ap)
2754 {
2755 	struct buf *bp;
2756 	struct vnode *vp;
2757 	struct ucred *cr;
2758 
2759 	bp = ap->a_bp;
2760 	vp = ap->a_vp;
2761 	KASSERT(bp->b_vp == vp, ("missing b_getvp"));
2762 	KASSERT(!(bp->b_flags & B_DONE),
2763 	    ("nfs_strategy: buffer %p unexpectedly marked B_DONE", bp));
2764 
2765 	if (vp->v_type == VREG && bp->b_blkno == bp->b_lblkno)
2766 		bp->b_blkno = bp->b_lblkno * (vp->v_bufobj.bo_bsize /
2767 		    DEV_BSIZE);
2768 	if (bp->b_iocmd == BIO_READ)
2769 		cr = bp->b_rcred;
2770 	else
2771 		cr = bp->b_wcred;
2772 
2773 	/*
2774 	 * If the op is asynchronous and an i/o daemon is waiting
2775 	 * queue the request, wake it up and wait for completion
2776 	 * otherwise just do it ourselves.
2777 	 */
2778 	if ((bp->b_flags & B_ASYNC) == 0 ||
2779 	    ncl_asyncio(VFSTONFS(vp->v_mount), bp, NOCRED, curthread))
2780 		(void) ncl_doio(vp, bp, cr, curthread, 1);
2781 	return (0);
2782 }
2783 
2784 /*
2785  * fsync vnode op. Just call ncl_flush() with commit == 1.
2786  */
2787 /* ARGSUSED */
2788 static int
2789 nfs_fsync(struct vop_fsync_args *ap)
2790 {
2791 
2792 	if (ap->a_vp->v_type != VREG) {
2793 		/*
2794 		 * For NFS, metadata is changed synchronously on the server,
2795 		 * so there is nothing to flush. Also, ncl_flush() clears
2796 		 * the NMODIFIED flag and that shouldn't be done here for
2797 		 * directories.
2798 		 */
2799 		return (0);
2800 	}
2801 	return (ncl_flush(ap->a_vp, ap->a_waitfor, ap->a_td, 1, 0));
2802 }
2803 
2804 /*
2805  * Flush all the blocks associated with a vnode.
2806  * 	Walk through the buffer pool and push any dirty pages
2807  *	associated with the vnode.
2808  * If the called_from_renewthread argument is TRUE, it has been called
2809  * from the NFSv4 renew thread and, as such, cannot block indefinitely
2810  * waiting for a buffer write to complete.
2811  */
2812 int
2813 ncl_flush(struct vnode *vp, int waitfor, struct thread *td,
2814     int commit, int called_from_renewthread)
2815 {
2816 	struct nfsnode *np = VTONFS(vp);
2817 	struct buf *bp;
2818 	int i;
2819 	struct buf *nbp;
2820 	struct nfsmount *nmp = VFSTONFS(vp->v_mount);
2821 	int error = 0, slptimeo = 0, slpflag = 0, retv, bvecpos;
2822 	int passone = 1, trycnt = 0;
2823 	u_quad_t off, endoff, toff;
2824 	struct ucred* wcred = NULL;
2825 	struct buf **bvec = NULL;
2826 	struct bufobj *bo;
2827 #ifndef NFS_COMMITBVECSIZ
2828 #define	NFS_COMMITBVECSIZ	20
2829 #endif
2830 	struct buf *bvec_on_stack[NFS_COMMITBVECSIZ];
2831 	u_int bvecsize = 0, bveccount;
2832 
2833 	if (called_from_renewthread != 0)
2834 		slptimeo = hz;
2835 	if (nmp->nm_flag & NFSMNT_INT)
2836 		slpflag = PCATCH;
2837 	if (!commit)
2838 		passone = 0;
2839 	bo = &vp->v_bufobj;
2840 	/*
2841 	 * A b_flags == (B_DELWRI | B_NEEDCOMMIT) block has been written to the
2842 	 * server, but has not been committed to stable storage on the server
2843 	 * yet. On the first pass, the byte range is worked out and the commit
2844 	 * rpc is done. On the second pass, ncl_writebp() is called to do the
2845 	 * job.
2846 	 */
2847 again:
2848 	off = (u_quad_t)-1;
2849 	endoff = 0;
2850 	bvecpos = 0;
2851 	if (NFS_ISV34(vp) && commit) {
2852 		if (bvec != NULL && bvec != bvec_on_stack)
2853 			free(bvec, M_TEMP);
2854 		/*
2855 		 * Count up how many buffers waiting for a commit.
2856 		 */
2857 		bveccount = 0;
2858 		BO_LOCK(bo);
2859 		TAILQ_FOREACH_SAFE(bp, &bo->bo_dirty.bv_hd, b_bobufs, nbp) {
2860 			if (!BUF_ISLOCKED(bp) &&
2861 			    (bp->b_flags & (B_DELWRI | B_NEEDCOMMIT))
2862 				== (B_DELWRI | B_NEEDCOMMIT))
2863 				bveccount++;
2864 		}
2865 		/*
2866 		 * Allocate space to remember the list of bufs to commit.  It is
2867 		 * important to use M_NOWAIT here to avoid a race with nfs_write.
2868 		 * If we can't get memory (for whatever reason), we will end up
2869 		 * committing the buffers one-by-one in the loop below.
2870 		 */
2871 		if (bveccount > NFS_COMMITBVECSIZ) {
2872 			/*
2873 			 * Release the vnode interlock to avoid a lock
2874 			 * order reversal.
2875 			 */
2876 			BO_UNLOCK(bo);
2877 			bvec = (struct buf **)
2878 				malloc(bveccount * sizeof(struct buf *),
2879 				       M_TEMP, M_NOWAIT);
2880 			BO_LOCK(bo);
2881 			if (bvec == NULL) {
2882 				bvec = bvec_on_stack;
2883 				bvecsize = NFS_COMMITBVECSIZ;
2884 			} else
2885 				bvecsize = bveccount;
2886 		} else {
2887 			bvec = bvec_on_stack;
2888 			bvecsize = NFS_COMMITBVECSIZ;
2889 		}
2890 		TAILQ_FOREACH_SAFE(bp, &bo->bo_dirty.bv_hd, b_bobufs, nbp) {
2891 			if (bvecpos >= bvecsize)
2892 				break;
2893 			if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT, NULL)) {
2894 				nbp = TAILQ_NEXT(bp, b_bobufs);
2895 				continue;
2896 			}
2897 			if ((bp->b_flags & (B_DELWRI | B_NEEDCOMMIT)) !=
2898 			    (B_DELWRI | B_NEEDCOMMIT)) {
2899 				BUF_UNLOCK(bp);
2900 				nbp = TAILQ_NEXT(bp, b_bobufs);
2901 				continue;
2902 			}
2903 			BO_UNLOCK(bo);
2904 			bremfree(bp);
2905 			/*
2906 			 * Work out if all buffers are using the same cred
2907 			 * so we can deal with them all with one commit.
2908 			 *
2909 			 * NOTE: we are not clearing B_DONE here, so we have
2910 			 * to do it later on in this routine if we intend to
2911 			 * initiate I/O on the bp.
2912 			 *
2913 			 * Note: to avoid loopback deadlocks, we do not
2914 			 * assign b_runningbufspace.
2915 			 */
2916 			if (wcred == NULL)
2917 				wcred = bp->b_wcred;
2918 			else if (wcred != bp->b_wcred)
2919 				wcred = NOCRED;
2920 			vfs_busy_pages(bp, 1);
2921 
2922 			BO_LOCK(bo);
2923 			/*
2924 			 * bp is protected by being locked, but nbp is not
2925 			 * and vfs_busy_pages() may sleep.  We have to
2926 			 * recalculate nbp.
2927 			 */
2928 			nbp = TAILQ_NEXT(bp, b_bobufs);
2929 
2930 			/*
2931 			 * A list of these buffers is kept so that the
2932 			 * second loop knows which buffers have actually
2933 			 * been committed. This is necessary, since there
2934 			 * may be a race between the commit rpc and new
2935 			 * uncommitted writes on the file.
2936 			 */
2937 			bvec[bvecpos++] = bp;
2938 			toff = ((u_quad_t)bp->b_blkno) * DEV_BSIZE +
2939 				bp->b_dirtyoff;
2940 			if (toff < off)
2941 				off = toff;
2942 			toff += (u_quad_t)(bp->b_dirtyend - bp->b_dirtyoff);
2943 			if (toff > endoff)
2944 				endoff = toff;
2945 		}
2946 		BO_UNLOCK(bo);
2947 	}
2948 	if (bvecpos > 0) {
2949 		/*
2950 		 * Commit data on the server, as required.
2951 		 * If all bufs are using the same wcred, then use that with
2952 		 * one call for all of them, otherwise commit each one
2953 		 * separately.
2954 		 */
2955 		if (wcred != NOCRED)
2956 			retv = ncl_commit(vp, off, (int)(endoff - off),
2957 					  wcred, td);
2958 		else {
2959 			retv = 0;
2960 			for (i = 0; i < bvecpos; i++) {
2961 				off_t off, size;
2962 				bp = bvec[i];
2963 				off = ((u_quad_t)bp->b_blkno) * DEV_BSIZE +
2964 					bp->b_dirtyoff;
2965 				size = (u_quad_t)(bp->b_dirtyend
2966 						  - bp->b_dirtyoff);
2967 				retv = ncl_commit(vp, off, (int)size,
2968 						  bp->b_wcred, td);
2969 				if (retv) break;
2970 			}
2971 		}
2972 
2973 		if (retv == NFSERR_STALEWRITEVERF)
2974 			ncl_clearcommit(vp->v_mount);
2975 
2976 		/*
2977 		 * Now, either mark the blocks I/O done or mark the
2978 		 * blocks dirty, depending on whether the commit
2979 		 * succeeded.
2980 		 */
2981 		for (i = 0; i < bvecpos; i++) {
2982 			bp = bvec[i];
2983 			bp->b_flags &= ~(B_NEEDCOMMIT | B_CLUSTEROK);
2984 			if (retv) {
2985 				/*
2986 				 * Error, leave B_DELWRI intact
2987 				 */
2988 				vfs_unbusy_pages(bp);
2989 				brelse(bp);
2990 			} else {
2991 				/*
2992 				 * Success, remove B_DELWRI ( bundirty() ).
2993 				 *
2994 				 * b_dirtyoff/b_dirtyend seem to be NFS
2995 				 * specific.  We should probably move that
2996 				 * into bundirty(). XXX
2997 				 */
2998 				bufobj_wref(bo);
2999 				bp->b_flags |= B_ASYNC;
3000 				bundirty(bp);
3001 				bp->b_flags &= ~B_DONE;
3002 				bp->b_ioflags &= ~BIO_ERROR;
3003 				bp->b_dirtyoff = bp->b_dirtyend = 0;
3004 				bufdone(bp);
3005 			}
3006 		}
3007 	}
3008 
3009 	/*
3010 	 * Start/do any write(s) that are required.
3011 	 */
3012 loop:
3013 	BO_LOCK(bo);
3014 	TAILQ_FOREACH_SAFE(bp, &bo->bo_dirty.bv_hd, b_bobufs, nbp) {
3015 		if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT, NULL)) {
3016 			if (waitfor != MNT_WAIT || passone)
3017 				continue;
3018 
3019 			error = BUF_TIMELOCK(bp,
3020 			    LK_EXCLUSIVE | LK_SLEEPFAIL | LK_INTERLOCK,
3021 			    BO_LOCKPTR(bo), "nfsfsync", slpflag, slptimeo);
3022 			if (error == 0) {
3023 				BUF_UNLOCK(bp);
3024 				goto loop;
3025 			}
3026 			if (error == ENOLCK) {
3027 				error = 0;
3028 				goto loop;
3029 			}
3030 			if (called_from_renewthread != 0) {
3031 				/*
3032 				 * Return EIO so the flush will be retried
3033 				 * later.
3034 				 */
3035 				error = EIO;
3036 				goto done;
3037 			}
3038 			if (newnfs_sigintr(nmp, td)) {
3039 				error = EINTR;
3040 				goto done;
3041 			}
3042 			if (slpflag == PCATCH) {
3043 				slpflag = 0;
3044 				slptimeo = 2 * hz;
3045 			}
3046 			goto loop;
3047 		}
3048 		if ((bp->b_flags & B_DELWRI) == 0)
3049 			panic("nfs_fsync: not dirty");
3050 		if ((passone || !commit) && (bp->b_flags & B_NEEDCOMMIT)) {
3051 			BUF_UNLOCK(bp);
3052 			continue;
3053 		}
3054 		BO_UNLOCK(bo);
3055 		bremfree(bp);
3056 		bp->b_flags |= B_ASYNC;
3057 		bwrite(bp);
3058 		if (newnfs_sigintr(nmp, td)) {
3059 			error = EINTR;
3060 			goto done;
3061 		}
3062 		goto loop;
3063 	}
3064 	if (passone) {
3065 		passone = 0;
3066 		BO_UNLOCK(bo);
3067 		goto again;
3068 	}
3069 	if (waitfor == MNT_WAIT) {
3070 		while (bo->bo_numoutput) {
3071 			error = bufobj_wwait(bo, slpflag, slptimeo);
3072 			if (error) {
3073 			    BO_UNLOCK(bo);
3074 			    if (called_from_renewthread != 0) {
3075 				/*
3076 				 * Return EIO so that the flush will be
3077 				 * retried later.
3078 				 */
3079 				error = EIO;
3080 				goto done;
3081 			    }
3082 			    error = newnfs_sigintr(nmp, td);
3083 			    if (error)
3084 				goto done;
3085 			    if (slpflag == PCATCH) {
3086 				slpflag = 0;
3087 				slptimeo = 2 * hz;
3088 			    }
3089 			    BO_LOCK(bo);
3090 			}
3091 		}
3092 		if (bo->bo_dirty.bv_cnt != 0 && commit) {
3093 			BO_UNLOCK(bo);
3094 			goto loop;
3095 		}
3096 		/*
3097 		 * Wait for all the async IO requests to drain
3098 		 */
3099 		BO_UNLOCK(bo);
3100 		NFSLOCKNODE(np);
3101 		while (np->n_directio_asyncwr > 0) {
3102 			np->n_flag |= NFSYNCWAIT;
3103 			error = newnfs_msleep(td, &np->n_directio_asyncwr,
3104 			    &np->n_mtx, slpflag | (PRIBIO + 1),
3105 			    "nfsfsync", 0);
3106 			if (error) {
3107 				if (newnfs_sigintr(nmp, td)) {
3108 					NFSUNLOCKNODE(np);
3109 					error = EINTR;
3110 					goto done;
3111 				}
3112 			}
3113 		}
3114 		NFSUNLOCKNODE(np);
3115 	} else
3116 		BO_UNLOCK(bo);
3117 	if (NFSHASPNFS(nmp)) {
3118 		nfscl_layoutcommit(vp, td);
3119 		/*
3120 		 * Invalidate the attribute cache, since writes to a DS
3121 		 * won't update the size attribute.
3122 		 */
3123 		NFSLOCKNODE(np);
3124 		np->n_attrstamp = 0;
3125 	} else
3126 		NFSLOCKNODE(np);
3127 	if (np->n_flag & NWRITEERR) {
3128 		error = np->n_error;
3129 		np->n_flag &= ~NWRITEERR;
3130 	}
3131   	if (commit && bo->bo_dirty.bv_cnt == 0 &&
3132 	    bo->bo_numoutput == 0 && np->n_directio_asyncwr == 0)
3133   		np->n_flag &= ~NMODIFIED;
3134 	NFSUNLOCKNODE(np);
3135 done:
3136 	if (bvec != NULL && bvec != bvec_on_stack)
3137 		free(bvec, M_TEMP);
3138 	if (error == 0 && commit != 0 && waitfor == MNT_WAIT &&
3139 	    (bo->bo_dirty.bv_cnt != 0 || bo->bo_numoutput != 0 ||
3140 	    np->n_directio_asyncwr != 0)) {
3141 		if (trycnt++ < 5) {
3142 			/* try, try again... */
3143 			passone = 1;
3144 			wcred = NULL;
3145 			bvec = NULL;
3146 			bvecsize = 0;
3147 			goto again;
3148 		}
3149 		vn_printf(vp, "ncl_flush failed");
3150 		error = called_from_renewthread != 0 ? EIO : EBUSY;
3151 	}
3152 	return (error);
3153 }
3154 
3155 /*
3156  * NFS advisory byte-level locks.
3157  */
3158 static int
3159 nfs_advlock(struct vop_advlock_args *ap)
3160 {
3161 	struct vnode *vp = ap->a_vp;
3162 	struct ucred *cred;
3163 	struct nfsnode *np = VTONFS(ap->a_vp);
3164 	struct proc *p = (struct proc *)ap->a_id;
3165 	struct thread *td = curthread;	/* XXX */
3166 	struct vattr va;
3167 	int ret, error;
3168 	u_quad_t size;
3169 	struct nfsmount *nmp;
3170 
3171 	error = NFSVOPLOCK(vp, LK_SHARED);
3172 	if (error != 0)
3173 		return (EBADF);
3174 	if (NFS_ISV4(vp) && (ap->a_flags & (F_POSIX | F_FLOCK)) != 0) {
3175 		if (vp->v_type != VREG) {
3176 			error = EINVAL;
3177 			goto out;
3178 		}
3179 		if ((ap->a_flags & F_POSIX) != 0)
3180 			cred = p->p_ucred;
3181 		else
3182 			cred = td->td_ucred;
3183 		NFSVOPLOCK(vp, LK_UPGRADE | LK_RETRY);
3184 		if (VN_IS_DOOMED(vp)) {
3185 			error = EBADF;
3186 			goto out;
3187 		}
3188 
3189 		/*
3190 		 * If this is unlocking a write locked region, flush and
3191 		 * commit them before unlocking. This is required by
3192 		 * RFC3530 Sec. 9.3.2.
3193 		 */
3194 		if (ap->a_op == F_UNLCK &&
3195 		    nfscl_checkwritelocked(vp, ap->a_fl, cred, td, ap->a_id,
3196 		    ap->a_flags))
3197 			(void) ncl_flush(vp, MNT_WAIT, td, 1, 0);
3198 
3199 		/*
3200 		 * Mark NFS node as might have acquired a lock.
3201 		 * This is separate from NHASBEENLOCKED, because it must
3202 		 * be done before the nfsrpc_advlock() call, which might
3203 		 * add a nfscllock structure to the client state.
3204 		 * It is used to check for the case where a nfscllock
3205 		 * state structure cannot exist for the file.
3206 		 * Only done for "oneopenown" NFSv4.1/4.2 mounts.
3207 		 */
3208 		nmp = VFSTONFS(vp->v_mount);
3209 		if (NFSHASNFSV4N(nmp) && NFSHASONEOPENOWN(nmp)) {
3210 			NFSLOCKNODE(np);
3211 			np->n_flag |= NMIGHTBELOCKED;
3212 			NFSUNLOCKNODE(np);
3213 		}
3214 
3215 		/*
3216 		 * Loop around doing the lock op, while a blocking lock
3217 		 * must wait for the lock op to succeed.
3218 		 */
3219 		do {
3220 			ret = nfsrpc_advlock(vp, np->n_size, ap->a_op,
3221 			    ap->a_fl, 0, cred, td, ap->a_id, ap->a_flags);
3222 			if (ret == NFSERR_DENIED && (ap->a_flags & F_WAIT) &&
3223 			    ap->a_op == F_SETLK) {
3224 				NFSVOPUNLOCK(vp);
3225 				error = nfs_catnap(PZERO | PCATCH, ret,
3226 				    "ncladvl");
3227 				if (error)
3228 					return (EINTR);
3229 				NFSVOPLOCK(vp, LK_EXCLUSIVE | LK_RETRY);
3230 				if (VN_IS_DOOMED(vp)) {
3231 					error = EBADF;
3232 					goto out;
3233 				}
3234 			}
3235 		} while (ret == NFSERR_DENIED && (ap->a_flags & F_WAIT) &&
3236 		     ap->a_op == F_SETLK);
3237 		if (ret == NFSERR_DENIED) {
3238 			error = EAGAIN;
3239 			goto out;
3240 		} else if (ret == EINVAL || ret == EBADF || ret == EINTR) {
3241 			error = ret;
3242 			goto out;
3243 		} else if (ret != 0) {
3244 			error = EACCES;
3245 			goto out;
3246 		}
3247 
3248 		/*
3249 		 * Now, if we just got a lock, invalidate data in the buffer
3250 		 * cache, as required, so that the coherency conforms with
3251 		 * RFC3530 Sec. 9.3.2.
3252 		 */
3253 		if (ap->a_op == F_SETLK) {
3254 			if ((np->n_flag & NMODIFIED) == 0) {
3255 				np->n_attrstamp = 0;
3256 				KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(vp);
3257 				ret = VOP_GETATTR(vp, &va, cred);
3258 			}
3259 			if ((np->n_flag & NMODIFIED) || ret ||
3260 			    np->n_change != va.va_filerev) {
3261 				(void) ncl_vinvalbuf(vp, V_SAVE, td, 1);
3262 				np->n_attrstamp = 0;
3263 				KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(vp);
3264 				ret = VOP_GETATTR(vp, &va, cred);
3265 				if (!ret) {
3266 					np->n_mtime = va.va_mtime;
3267 					np->n_change = va.va_filerev;
3268 				}
3269 			}
3270 			/* Mark that a file lock has been acquired. */
3271 			NFSLOCKNODE(np);
3272 			np->n_flag |= NHASBEENLOCKED;
3273 			NFSUNLOCKNODE(np);
3274 		}
3275 	} else if (!NFS_ISV4(vp)) {
3276 		if ((VFSTONFS(vp->v_mount)->nm_flag & NFSMNT_NOLOCKD) != 0) {
3277 			size = VTONFS(vp)->n_size;
3278 			NFSVOPUNLOCK(vp);
3279 			error = lf_advlock(ap, &(vp->v_lockf), size);
3280 		} else {
3281 			if (nfs_advlock_p != NULL)
3282 				error = nfs_advlock_p(ap);
3283 			else {
3284 				NFSVOPUNLOCK(vp);
3285 				error = ENOLCK;
3286 			}
3287 		}
3288 		if (error == 0 && ap->a_op == F_SETLK) {
3289 			error = NFSVOPLOCK(vp, LK_SHARED);
3290 			if (error == 0) {
3291 				/* Mark that a file lock has been acquired. */
3292 				NFSLOCKNODE(np);
3293 				np->n_flag |= NHASBEENLOCKED;
3294 				NFSUNLOCKNODE(np);
3295 				NFSVOPUNLOCK(vp);
3296 			}
3297 		}
3298 		return (error);
3299 	} else
3300 		error = EOPNOTSUPP;
3301 out:
3302 	NFSVOPUNLOCK(vp);
3303 	return (error);
3304 }
3305 
3306 /*
3307  * NFS advisory byte-level locks.
3308  */
3309 static int
3310 nfs_advlockasync(struct vop_advlockasync_args *ap)
3311 {
3312 	struct vnode *vp = ap->a_vp;
3313 	u_quad_t size;
3314 	int error;
3315 
3316 	if (NFS_ISV4(vp))
3317 		return (EOPNOTSUPP);
3318 	error = NFSVOPLOCK(vp, LK_SHARED);
3319 	if (error)
3320 		return (error);
3321 	if ((VFSTONFS(vp->v_mount)->nm_flag & NFSMNT_NOLOCKD) != 0) {
3322 		size = VTONFS(vp)->n_size;
3323 		NFSVOPUNLOCK(vp);
3324 		error = lf_advlockasync(ap, &(vp->v_lockf), size);
3325 	} else {
3326 		NFSVOPUNLOCK(vp);
3327 		error = EOPNOTSUPP;
3328 	}
3329 	return (error);
3330 }
3331 
3332 /*
3333  * Print out the contents of an nfsnode.
3334  */
3335 static int
3336 nfs_print(struct vop_print_args *ap)
3337 {
3338 	struct vnode *vp = ap->a_vp;
3339 	struct nfsnode *np = VTONFS(vp);
3340 
3341 	printf("\tfileid %jd fsid 0x%jx", (uintmax_t)np->n_vattr.na_fileid,
3342 	    (uintmax_t)np->n_vattr.na_fsid);
3343 	if (vp->v_type == VFIFO)
3344 		fifo_printinfo(vp);
3345 	printf("\n");
3346 	return (0);
3347 }
3348 
3349 /*
3350  * This is the "real" nfs::bwrite(struct buf*).
3351  * We set B_CACHE if this is a VMIO buffer.
3352  */
3353 int
3354 ncl_writebp(struct buf *bp, int force __unused, struct thread *td)
3355 {
3356 	int oldflags, rtval;
3357 
3358 	if (bp->b_flags & B_INVAL) {
3359 		brelse(bp);
3360 		return (0);
3361 	}
3362 
3363 	oldflags = bp->b_flags;
3364 	bp->b_flags |= B_CACHE;
3365 
3366 	/*
3367 	 * Undirty the bp.  We will redirty it later if the I/O fails.
3368 	 */
3369 	bundirty(bp);
3370 	bp->b_flags &= ~B_DONE;
3371 	bp->b_ioflags &= ~BIO_ERROR;
3372 	bp->b_iocmd = BIO_WRITE;
3373 
3374 	bufobj_wref(bp->b_bufobj);
3375 	curthread->td_ru.ru_oublock++;
3376 
3377 	/*
3378 	 * Note: to avoid loopback deadlocks, we do not
3379 	 * assign b_runningbufspace.
3380 	 */
3381 	vfs_busy_pages(bp, 1);
3382 
3383 	BUF_KERNPROC(bp);
3384 	bp->b_iooffset = dbtob(bp->b_blkno);
3385 	bstrategy(bp);
3386 
3387 	if ((oldflags & B_ASYNC) != 0)
3388 		return (0);
3389 
3390 	rtval = bufwait(bp);
3391 	if (oldflags & B_DELWRI)
3392 		reassignbuf(bp);
3393 	brelse(bp);
3394 	return (rtval);
3395 }
3396 
3397 /*
3398  * nfs special file access vnode op.
3399  * Essentially just get vattr and then imitate iaccess() since the device is
3400  * local to the client.
3401  */
3402 static int
3403 nfsspec_access(struct vop_access_args *ap)
3404 {
3405 	struct vattr *vap;
3406 	struct ucred *cred = ap->a_cred;
3407 	struct vnode *vp = ap->a_vp;
3408 	accmode_t accmode = ap->a_accmode;
3409 	struct vattr vattr;
3410 	int error;
3411 
3412 	/*
3413 	 * Disallow write attempts on filesystems mounted read-only;
3414 	 * unless the file is a socket, fifo, or a block or character
3415 	 * device resident on the filesystem.
3416 	 */
3417 	if ((accmode & VWRITE) && (vp->v_mount->mnt_flag & MNT_RDONLY)) {
3418 		switch (vp->v_type) {
3419 		case VREG:
3420 		case VDIR:
3421 		case VLNK:
3422 			return (EROFS);
3423 		default:
3424 			break;
3425 		}
3426 	}
3427 	vap = &vattr;
3428 	error = VOP_GETATTR(vp, vap, cred);
3429 	if (error)
3430 		goto out;
3431 	error = vaccess(vp->v_type, vap->va_mode, vap->va_uid, vap->va_gid,
3432 	    accmode, cred);
3433 out:
3434 	return error;
3435 }
3436 
3437 /*
3438  * Read wrapper for fifos.
3439  */
3440 static int
3441 nfsfifo_read(struct vop_read_args *ap)
3442 {
3443 	struct nfsnode *np = VTONFS(ap->a_vp);
3444 	int error;
3445 
3446 	/*
3447 	 * Set access flag.
3448 	 */
3449 	NFSLOCKNODE(np);
3450 	np->n_flag |= NACC;
3451 	vfs_timestamp(&np->n_atim);
3452 	NFSUNLOCKNODE(np);
3453 	error = fifo_specops.vop_read(ap);
3454 	return error;
3455 }
3456 
3457 /*
3458  * Write wrapper for fifos.
3459  */
3460 static int
3461 nfsfifo_write(struct vop_write_args *ap)
3462 {
3463 	struct nfsnode *np = VTONFS(ap->a_vp);
3464 
3465 	/*
3466 	 * Set update flag.
3467 	 */
3468 	NFSLOCKNODE(np);
3469 	np->n_flag |= NUPD;
3470 	vfs_timestamp(&np->n_mtim);
3471 	NFSUNLOCKNODE(np);
3472 	return(fifo_specops.vop_write(ap));
3473 }
3474 
3475 /*
3476  * Close wrapper for fifos.
3477  *
3478  * Update the times on the nfsnode then do fifo close.
3479  */
3480 static int
3481 nfsfifo_close(struct vop_close_args *ap)
3482 {
3483 	struct vnode *vp = ap->a_vp;
3484 	struct nfsnode *np = VTONFS(vp);
3485 	struct vattr vattr;
3486 	struct timespec ts;
3487 
3488 	NFSLOCKNODE(np);
3489 	if (np->n_flag & (NACC | NUPD)) {
3490 		vfs_timestamp(&ts);
3491 		if (np->n_flag & NACC)
3492 			np->n_atim = ts;
3493 		if (np->n_flag & NUPD)
3494 			np->n_mtim = ts;
3495 		np->n_flag |= NCHG;
3496 		if (vrefcnt(vp) == 1 &&
3497 		    (vp->v_mount->mnt_flag & MNT_RDONLY) == 0) {
3498 			VATTR_NULL(&vattr);
3499 			if (np->n_flag & NACC)
3500 				vattr.va_atime = np->n_atim;
3501 			if (np->n_flag & NUPD)
3502 				vattr.va_mtime = np->n_mtim;
3503 			NFSUNLOCKNODE(np);
3504 			(void)VOP_SETATTR(vp, &vattr, ap->a_cred);
3505 			goto out;
3506 		}
3507 	}
3508 	NFSUNLOCKNODE(np);
3509 out:
3510 	return (fifo_specops.vop_close(ap));
3511 }
3512 
3513 /*
3514  * Just call ncl_writebp() with the force argument set to 1.
3515  *
3516  * NOTE: B_DONE may or may not be set in a_bp on call.
3517  */
3518 static int
3519 nfs_bwrite(struct buf *bp)
3520 {
3521 
3522 	return (ncl_writebp(bp, 1, curthread));
3523 }
3524 
3525 struct buf_ops buf_ops_newnfs = {
3526 	.bop_name	=	"buf_ops_nfs",
3527 	.bop_write	=	nfs_bwrite,
3528 	.bop_strategy	=	bufstrategy,
3529 	.bop_sync	=	bufsync,
3530 	.bop_bdflush	=	bufbdflush,
3531 };
3532 
3533 static int
3534 nfs_getacl(struct vop_getacl_args *ap)
3535 {
3536 	int error;
3537 
3538 	if (ap->a_type != ACL_TYPE_NFS4)
3539 		return (EOPNOTSUPP);
3540 	error = nfsrpc_getacl(ap->a_vp, ap->a_cred, ap->a_td, ap->a_aclp,
3541 	    NULL);
3542 	if (error > NFSERR_STALE) {
3543 		(void) nfscl_maperr(ap->a_td, error, (uid_t)0, (gid_t)0);
3544 		error = EPERM;
3545 	}
3546 	return (error);
3547 }
3548 
3549 static int
3550 nfs_setacl(struct vop_setacl_args *ap)
3551 {
3552 	int error;
3553 
3554 	if (ap->a_type != ACL_TYPE_NFS4)
3555 		return (EOPNOTSUPP);
3556 	error = nfsrpc_setacl(ap->a_vp, ap->a_cred, ap->a_td, ap->a_aclp,
3557 	    NULL);
3558 	if (error > NFSERR_STALE) {
3559 		(void) nfscl_maperr(ap->a_td, error, (uid_t)0, (gid_t)0);
3560 		error = EPERM;
3561 	}
3562 	return (error);
3563 }
3564 
3565 /*
3566  * VOP_ADVISE for NFS.
3567  * Just return 0 for any errors, since it is just a hint.
3568  */
3569 static int
3570 nfs_advise(struct vop_advise_args *ap)
3571 {
3572 	struct thread *td = curthread;
3573 	struct nfsmount *nmp;
3574 	uint64_t len;
3575 	int error;
3576 
3577 	/*
3578 	 * First do vop_stdadvise() to handle the buffer cache.
3579 	 */
3580 	error = vop_stdadvise(ap);
3581 	if (error != 0)
3582 		return (error);
3583 	if (ap->a_start < 0 || ap->a_end < 0)
3584 		return (0);
3585 	if (ap->a_end == OFF_MAX)
3586 		len = 0;
3587 	else if (ap->a_end < ap->a_start)
3588 		return (0);
3589 	else
3590 		len = ap->a_end - ap->a_start + 1;
3591 	nmp = VFSTONFS(ap->a_vp->v_mount);
3592 	mtx_lock(&nmp->nm_mtx);
3593 	if (!NFSHASNFSV4(nmp) || nmp->nm_minorvers < NFSV42_MINORVERSION ||
3594 	    (NFSHASPNFS(nmp) && (nmp->nm_privflag & NFSMNTP_IOADVISETHRUMDS) ==
3595 	    0) || (nmp->nm_privflag & NFSMNTP_NOADVISE) != 0) {
3596 		mtx_unlock(&nmp->nm_mtx);
3597 		return (0);
3598 	}
3599 	mtx_unlock(&nmp->nm_mtx);
3600 	error = nfsrpc_advise(ap->a_vp, ap->a_start, len, ap->a_advice,
3601 	    td->td_ucred, td);
3602 	if (error == NFSERR_NOTSUPP) {
3603 		mtx_lock(&nmp->nm_mtx);
3604 		nmp->nm_privflag |= NFSMNTP_NOADVISE;
3605 		mtx_unlock(&nmp->nm_mtx);
3606 	}
3607 	return (0);
3608 }
3609 
3610 /*
3611  * nfs allocate call
3612  */
3613 static int
3614 nfs_allocate(struct vop_allocate_args *ap)
3615 {
3616 	struct vnode *vp = ap->a_vp;
3617 	struct thread *td = curthread;
3618 	struct nfsvattr nfsva;
3619 	struct nfsmount *nmp;
3620 	int attrflag, error, ret;
3621 
3622 	attrflag = 0;
3623 	nmp = VFSTONFS(vp->v_mount);
3624 	mtx_lock(&nmp->nm_mtx);
3625 	if (NFSHASNFSV4(nmp) && nmp->nm_minorvers >= NFSV42_MINORVERSION &&
3626 	    (nmp->nm_privflag & NFSMNTP_NOALLOCATE) == 0) {
3627 		mtx_unlock(&nmp->nm_mtx);
3628 		/*
3629 		 * Flush first to ensure that the allocate adds to the
3630 		 * file's allocation on the server.
3631 		 */
3632 		error = ncl_flush(vp, MNT_WAIT, td, 1, 0);
3633 		if (error == 0)
3634 			error = nfsrpc_allocate(vp, *ap->a_offset, *ap->a_len,
3635 			    &nfsva, &attrflag, td->td_ucred, td, NULL);
3636 		if (error == 0) {
3637 			*ap->a_offset += *ap->a_len;
3638 			*ap->a_len = 0;
3639 		} else if (error == NFSERR_NOTSUPP) {
3640 			mtx_lock(&nmp->nm_mtx);
3641 			nmp->nm_privflag |= NFSMNTP_NOALLOCATE;
3642 			mtx_unlock(&nmp->nm_mtx);
3643 		}
3644 	} else {
3645 		mtx_unlock(&nmp->nm_mtx);
3646 		error = EIO;
3647 	}
3648 	/*
3649 	 * If the NFS server cannot perform the Allocate operation, just call
3650 	 * vop_stdallocate() to perform it.
3651 	 */
3652 	if (error != 0)
3653 		error = vop_stdallocate(ap);
3654 	if (attrflag != 0) {
3655 		ret = nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0, 1);
3656 		if (error == 0 && ret != 0)
3657 			error = ret;
3658 	}
3659 	if (error != 0)
3660 		error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0);
3661 	return (error);
3662 }
3663 
3664 /*
3665  * nfs copy_file_range call
3666  */
3667 static int
3668 nfs_copy_file_range(struct vop_copy_file_range_args *ap)
3669 {
3670 	struct vnode *invp = ap->a_invp;
3671 	struct vnode *outvp = ap->a_outvp;
3672 	struct mount *mp;
3673 	struct nfsvattr innfsva, outnfsva;
3674 	struct vattr *vap;
3675 	struct uio io;
3676 	struct nfsmount *nmp;
3677 	size_t len, len2;
3678 	int error, inattrflag, outattrflag, ret, ret2;
3679 	off_t inoff, outoff;
3680 	bool consecutive, must_commit, tryoutcred;
3681 
3682 	ret = ret2 = 0;
3683 	nmp = VFSTONFS(invp->v_mount);
3684 	mtx_lock(&nmp->nm_mtx);
3685 	/* NFSv4.2 Copy is not permitted for infile == outfile. */
3686 	if (!NFSHASNFSV4(nmp) || nmp->nm_minorvers < NFSV42_MINORVERSION ||
3687 	    (nmp->nm_privflag & NFSMNTP_NOCOPY) != 0 || invp == outvp) {
3688 		mtx_unlock(&nmp->nm_mtx);
3689 		error = vn_generic_copy_file_range(ap->a_invp, ap->a_inoffp,
3690 		    ap->a_outvp, ap->a_outoffp, ap->a_lenp, ap->a_flags,
3691 		    ap->a_incred, ap->a_outcred, ap->a_fsizetd);
3692 		return (error);
3693 	}
3694 	mtx_unlock(&nmp->nm_mtx);
3695 
3696 	/* Lock both vnodes, avoiding risk of deadlock. */
3697 	do {
3698 		mp = NULL;
3699 		error = vn_start_write(outvp, &mp, V_WAIT);
3700 		if (error == 0) {
3701 			error = vn_lock(outvp, LK_EXCLUSIVE);
3702 			if (error == 0) {
3703 				error = vn_lock(invp, LK_SHARED | LK_NOWAIT);
3704 				if (error == 0)
3705 					break;
3706 				VOP_UNLOCK(outvp);
3707 				if (mp != NULL)
3708 					vn_finished_write(mp);
3709 				mp = NULL;
3710 				error = vn_lock(invp, LK_SHARED);
3711 				if (error == 0)
3712 					VOP_UNLOCK(invp);
3713 			}
3714 		}
3715 		if (mp != NULL)
3716 			vn_finished_write(mp);
3717 	} while (error == 0);
3718 	if (error != 0)
3719 		return (error);
3720 
3721 	/*
3722 	 * Do the vn_rlimit_fsize() check.  Should this be above the VOP layer?
3723 	 */
3724 	io.uio_offset = *ap->a_outoffp;
3725 	io.uio_resid = *ap->a_lenp;
3726 	error = vn_rlimit_fsize(outvp, &io, ap->a_fsizetd);
3727 
3728 	/*
3729 	 * Flush the input file so that the data is up to date before
3730 	 * the copy.  Flush writes for the output file so that they
3731 	 * do not overwrite the data copied to the output file by the Copy.
3732 	 * Set the commit argument for both flushes so that the data is on
3733 	 * stable storage before the Copy RPC.  This is done in case the
3734 	 * server reboots during the Copy and needs to be redone.
3735 	 */
3736 	if (error == 0)
3737 		error = ncl_flush(invp, MNT_WAIT, curthread, 1, 0);
3738 	if (error == 0)
3739 		error = ncl_flush(outvp, MNT_WAIT, curthread, 1, 0);
3740 
3741 	/* Do the actual NFSv4.2 RPC. */
3742 	len = *ap->a_lenp;
3743 	mtx_lock(&nmp->nm_mtx);
3744 	if ((nmp->nm_privflag & NFSMNTP_NOCONSECUTIVE) == 0)
3745 		consecutive = true;
3746 	else
3747 		consecutive = false;
3748 	mtx_unlock(&nmp->nm_mtx);
3749 	inoff = *ap->a_inoffp;
3750 	outoff = *ap->a_outoffp;
3751 	tryoutcred = true;
3752 	must_commit = false;
3753 	if (error == 0) {
3754 		vap = &VTONFS(invp)->n_vattr.na_vattr;
3755 		error = VOP_GETATTR(invp, vap, ap->a_incred);
3756 		if (error == 0) {
3757 			/*
3758 			 * Clip "len" at va_size so that RFC compliant servers
3759 			 * will not reply NFSERR_INVAL.
3760 			 * Setting "len == 0" for the RPC would be preferred,
3761 			 * but some Linux servers do not support that.
3762 			 */
3763 			if (inoff >= vap->va_size)
3764 				*ap->a_lenp = len = 0;
3765 			else if (inoff + len > vap->va_size)
3766 				*ap->a_lenp = len = vap->va_size - inoff;
3767 		} else
3768 			error = 0;
3769 	}
3770 
3771 	/*
3772 	 * len will be set to 0 upon a successful Copy RPC.
3773 	 * As such, this only loops when the Copy RPC needs to be retried.
3774 	 */
3775 	while (len > 0 && error == 0) {
3776 		inattrflag = outattrflag = 0;
3777 		len2 = len;
3778 		if (tryoutcred)
3779 			error = nfsrpc_copy_file_range(invp, ap->a_inoffp,
3780 			    outvp, ap->a_outoffp, &len2, ap->a_flags,
3781 			    &inattrflag, &innfsva, &outattrflag, &outnfsva,
3782 			    ap->a_outcred, consecutive, &must_commit);
3783 		else
3784 			error = nfsrpc_copy_file_range(invp, ap->a_inoffp,
3785 			    outvp, ap->a_outoffp, &len2, ap->a_flags,
3786 			    &inattrflag, &innfsva, &outattrflag, &outnfsva,
3787 			    ap->a_incred, consecutive, &must_commit);
3788 		if (inattrflag != 0)
3789 			ret = nfscl_loadattrcache(&invp, &innfsva, NULL, NULL,
3790 			    0, 1);
3791 		if (outattrflag != 0)
3792 			ret2 = nfscl_loadattrcache(&outvp, &outnfsva, NULL,
3793 			    NULL, 1, 1);
3794 		if (error == 0) {
3795 			if (consecutive == false) {
3796 				if (len2 == len) {
3797 					mtx_lock(&nmp->nm_mtx);
3798 					nmp->nm_privflag |=
3799 					    NFSMNTP_NOCONSECUTIVE;
3800 					mtx_unlock(&nmp->nm_mtx);
3801 				} else
3802 					error = NFSERR_OFFLOADNOREQS;
3803 			}
3804 			*ap->a_lenp = len2;
3805 			len = 0;
3806 			if (len2 > 0 && must_commit && error == 0)
3807 				error = ncl_commit(outvp, outoff, *ap->a_lenp,
3808 				    ap->a_outcred, curthread);
3809 			if (error == 0 && ret != 0)
3810 				error = ret;
3811 			if (error == 0 && ret2 != 0)
3812 				error = ret2;
3813 		} else if (error == NFSERR_OFFLOADNOREQS && consecutive) {
3814 			/*
3815 			 * Try consecutive == false, which is ok only if all
3816 			 * bytes are copied.
3817 			 * If only some bytes were copied when consecutive
3818 			 * is false, there is no way to know which bytes
3819 			 * still need to be written.
3820 			 */
3821 			consecutive = false;
3822 			error = 0;
3823 		} else if (error == NFSERR_ACCES && tryoutcred) {
3824 			/* Try again with incred. */
3825 			tryoutcred = false;
3826 			error = 0;
3827 		}
3828 		if (error == NFSERR_STALEWRITEVERF) {
3829 			/*
3830 			 * Server rebooted, so do it all again.
3831 			 */
3832 			*ap->a_inoffp = inoff;
3833 			*ap->a_outoffp = outoff;
3834 			len = *ap->a_lenp;
3835 			must_commit = false;
3836 			error = 0;
3837 		}
3838 	}
3839 	VOP_UNLOCK(invp);
3840 	VOP_UNLOCK(outvp);
3841 	if (mp != NULL)
3842 		vn_finished_write(mp);
3843 	if (error == NFSERR_NOTSUPP || error == NFSERR_OFFLOADNOREQS ||
3844 	    error == NFSERR_ACCES) {
3845 		/*
3846 		 * Unlike the NFSv4.2 Copy, vn_generic_copy_file_range() can
3847 		 * use a_incred for the read and a_outcred for the write, so
3848 		 * try this for NFSERR_ACCES failures for the Copy.
3849 		 * For NFSERR_NOTSUPP and NFSERR_OFFLOADNOREQS, the Copy can
3850 		 * never succeed, so disable it.
3851 		 */
3852 		if (error != NFSERR_ACCES) {
3853 			/* Can never do Copy on this mount. */
3854 			mtx_lock(&nmp->nm_mtx);
3855 			nmp->nm_privflag |= NFSMNTP_NOCOPY;
3856 			mtx_unlock(&nmp->nm_mtx);
3857 		}
3858 		*ap->a_inoffp = inoff;
3859 		*ap->a_outoffp = outoff;
3860 		error = vn_generic_copy_file_range(ap->a_invp, ap->a_inoffp,
3861 		    ap->a_outvp, ap->a_outoffp, ap->a_lenp, ap->a_flags,
3862 		    ap->a_incred, ap->a_outcred, ap->a_fsizetd);
3863 	} else if (error != 0)
3864 		*ap->a_lenp = 0;
3865 
3866 	if (error != 0)
3867 		error = nfscl_maperr(curthread, error, (uid_t)0, (gid_t)0);
3868 	return (error);
3869 }
3870 
3871 /*
3872  * nfs ioctl call
3873  */
3874 static int
3875 nfs_ioctl(struct vop_ioctl_args *ap)
3876 {
3877 	struct vnode *vp = ap->a_vp;
3878 	struct nfsvattr nfsva;
3879 	struct nfsmount *nmp;
3880 	int attrflag, content, error, ret;
3881 	bool eof = false;			/* shut up compiler. */
3882 
3883 	if (vp->v_type != VREG)
3884 		return (ENOTTY);
3885 	nmp = VFSTONFS(vp->v_mount);
3886 	if (!NFSHASNFSV4(nmp) || nmp->nm_minorvers < NFSV42_MINORVERSION) {
3887 		error = vop_stdioctl(ap);
3888 		return (error);
3889 	}
3890 
3891 	/* Do the actual NFSv4.2 RPC. */
3892 	switch (ap->a_command) {
3893 	case FIOSEEKDATA:
3894 		content = NFSV4CONTENT_DATA;
3895 		break;
3896 	case FIOSEEKHOLE:
3897 		content = NFSV4CONTENT_HOLE;
3898 		break;
3899 	default:
3900 		return (ENOTTY);
3901 	}
3902 
3903 	error = vn_lock(vp, LK_SHARED);
3904 	if (error != 0)
3905 		return (EBADF);
3906 	attrflag = 0;
3907 	if (*((off_t *)ap->a_data) >= VTONFS(vp)->n_size)
3908 		error = ENXIO;
3909 	else {
3910 		/*
3911 		 * Flush all writes, so that the server is up to date.
3912 		 * Although a Commit is not required, the commit argument
3913 		 * is set so that, for a pNFS File/Flexible File Layout
3914 		 * server, the LayoutCommit will be done to ensure the file
3915 		 * size is up to date on the Metadata Server.
3916 		 */
3917 		error = ncl_flush(vp, MNT_WAIT, ap->a_td, 1, 0);
3918 		if (error == 0)
3919 			error = nfsrpc_seek(vp, (off_t *)ap->a_data, &eof,
3920 			    content, ap->a_cred, &nfsva, &attrflag);
3921 		/* If at eof for FIOSEEKDATA, return ENXIO. */
3922 		if (eof && error == 0 && content == NFSV4CONTENT_DATA)
3923 			error = ENXIO;
3924 	}
3925 	if (attrflag != 0) {
3926 		ret = nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0, 1);
3927 		if (error == 0 && ret != 0)
3928 			error = ret;
3929 	}
3930 	NFSVOPUNLOCK(vp);
3931 
3932 	if (error != 0)
3933 		error = ENXIO;
3934 	return (error);
3935 }
3936 
3937 /*
3938  * nfs getextattr call
3939  */
3940 static int
3941 nfs_getextattr(struct vop_getextattr_args *ap)
3942 {
3943 	struct vnode *vp = ap->a_vp;
3944 	struct nfsmount *nmp;
3945 	struct ucred *cred;
3946 	struct thread *td = ap->a_td;
3947 	struct nfsvattr nfsva;
3948 	ssize_t len;
3949 	int attrflag, error, ret;
3950 
3951 	nmp = VFSTONFS(vp->v_mount);
3952 	mtx_lock(&nmp->nm_mtx);
3953 	if (!NFSHASNFSV4(nmp) || nmp->nm_minorvers < NFSV42_MINORVERSION ||
3954 	    (nmp->nm_privflag & NFSMNTP_NOXATTR) != 0 ||
3955 	    ap->a_attrnamespace != EXTATTR_NAMESPACE_USER) {
3956 		mtx_unlock(&nmp->nm_mtx);
3957 		return (EOPNOTSUPP);
3958 	}
3959 	mtx_unlock(&nmp->nm_mtx);
3960 
3961 	cred = ap->a_cred;
3962 	if (cred == NULL)
3963 		cred = td->td_ucred;
3964 	/* Do the actual NFSv4.2 Optional Extended Attribute (RFC-8276) RPC. */
3965 	attrflag = 0;
3966 	error = nfsrpc_getextattr(vp, ap->a_name, ap->a_uio, &len, &nfsva,
3967 	    &attrflag, cred, td);
3968 	if (attrflag != 0) {
3969 		ret = nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0, 1);
3970 		if (error == 0 && ret != 0)
3971 			error = ret;
3972 	}
3973 	if (error == 0 && ap->a_size != NULL)
3974 		*ap->a_size = len;
3975 
3976 	switch (error) {
3977 	case NFSERR_NOTSUPP:
3978 	case NFSERR_OPILLEGAL:
3979 		mtx_lock(&nmp->nm_mtx);
3980 		nmp->nm_privflag |= NFSMNTP_NOXATTR;
3981 		mtx_unlock(&nmp->nm_mtx);
3982 		error = EOPNOTSUPP;
3983 		break;
3984 	case NFSERR_NOXATTR:
3985 	case NFSERR_XATTR2BIG:
3986 		error = ENOATTR;
3987 		break;
3988 	default:
3989 		error = nfscl_maperr(td, error, 0, 0);
3990 		break;
3991 	}
3992 	return (error);
3993 }
3994 
3995 /*
3996  * nfs setextattr call
3997  */
3998 static int
3999 nfs_setextattr(struct vop_setextattr_args *ap)
4000 {
4001 	struct vnode *vp = ap->a_vp;
4002 	struct nfsmount *nmp;
4003 	struct ucred *cred;
4004 	struct thread *td = ap->a_td;
4005 	struct nfsvattr nfsva;
4006 	int attrflag, error, ret;
4007 
4008 	nmp = VFSTONFS(vp->v_mount);
4009 	mtx_lock(&nmp->nm_mtx);
4010 	if (!NFSHASNFSV4(nmp) || nmp->nm_minorvers < NFSV42_MINORVERSION ||
4011 	    (nmp->nm_privflag & NFSMNTP_NOXATTR) != 0 ||
4012 	    ap->a_attrnamespace != EXTATTR_NAMESPACE_USER) {
4013 		mtx_unlock(&nmp->nm_mtx);
4014 		return (EOPNOTSUPP);
4015 	}
4016 	mtx_unlock(&nmp->nm_mtx);
4017 
4018 	if (ap->a_uio->uio_resid < 0)
4019 		return (EINVAL);
4020 	cred = ap->a_cred;
4021 	if (cred == NULL)
4022 		cred = td->td_ucred;
4023 	/* Do the actual NFSv4.2 Optional Extended Attribute (RFC-8276) RPC. */
4024 	attrflag = 0;
4025 	error = nfsrpc_setextattr(vp, ap->a_name, ap->a_uio, &nfsva,
4026 	    &attrflag, cred, td);
4027 	if (attrflag != 0) {
4028 		ret = nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0, 1);
4029 		if (error == 0 && ret != 0)
4030 			error = ret;
4031 	}
4032 
4033 	switch (error) {
4034 	case NFSERR_NOTSUPP:
4035 	case NFSERR_OPILLEGAL:
4036 		mtx_lock(&nmp->nm_mtx);
4037 		nmp->nm_privflag |= NFSMNTP_NOXATTR;
4038 		mtx_unlock(&nmp->nm_mtx);
4039 		error = EOPNOTSUPP;
4040 		break;
4041 	case NFSERR_NOXATTR:
4042 	case NFSERR_XATTR2BIG:
4043 		error = ENOATTR;
4044 		break;
4045 	default:
4046 		error = nfscl_maperr(td, error, 0, 0);
4047 		break;
4048 	}
4049 	return (error);
4050 }
4051 
4052 /*
4053  * nfs listextattr call
4054  */
4055 static int
4056 nfs_listextattr(struct vop_listextattr_args *ap)
4057 {
4058 	struct vnode *vp = ap->a_vp;
4059 	struct nfsmount *nmp;
4060 	struct ucred *cred;
4061 	struct thread *td = ap->a_td;
4062 	struct nfsvattr nfsva;
4063 	size_t len, len2;
4064 	uint64_t cookie;
4065 	int attrflag, error, ret;
4066 	bool eof;
4067 
4068 	nmp = VFSTONFS(vp->v_mount);
4069 	mtx_lock(&nmp->nm_mtx);
4070 	if (!NFSHASNFSV4(nmp) || nmp->nm_minorvers < NFSV42_MINORVERSION ||
4071 	    (nmp->nm_privflag & NFSMNTP_NOXATTR) != 0 ||
4072 	    ap->a_attrnamespace != EXTATTR_NAMESPACE_USER) {
4073 		mtx_unlock(&nmp->nm_mtx);
4074 		return (EOPNOTSUPP);
4075 	}
4076 	mtx_unlock(&nmp->nm_mtx);
4077 
4078 	cred = ap->a_cred;
4079 	if (cred == NULL)
4080 		cred = td->td_ucred;
4081 
4082 	/* Loop around doing List Extended Attribute RPCs. */
4083 	eof = false;
4084 	cookie = 0;
4085 	len2 = 0;
4086 	error = 0;
4087 	while (!eof && error == 0) {
4088 		len = nmp->nm_rsize;
4089 		attrflag = 0;
4090 		error = nfsrpc_listextattr(vp, &cookie, ap->a_uio, &len, &eof,
4091 		    &nfsva, &attrflag, cred, td);
4092 		if (attrflag != 0) {
4093 			ret = nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0,
4094 			    1);
4095 			if (error == 0 && ret != 0)
4096 				error = ret;
4097 		}
4098 		if (error == 0) {
4099 			len2 += len;
4100 			if (len2 > SSIZE_MAX)
4101 				error = ENOATTR;
4102 		}
4103 	}
4104 	if (error == 0 && ap->a_size != NULL)
4105 		*ap->a_size = len2;
4106 
4107 	switch (error) {
4108 	case NFSERR_NOTSUPP:
4109 	case NFSERR_OPILLEGAL:
4110 		mtx_lock(&nmp->nm_mtx);
4111 		nmp->nm_privflag |= NFSMNTP_NOXATTR;
4112 		mtx_unlock(&nmp->nm_mtx);
4113 		error = EOPNOTSUPP;
4114 		break;
4115 	case NFSERR_NOXATTR:
4116 	case NFSERR_XATTR2BIG:
4117 		error = ENOATTR;
4118 		break;
4119 	default:
4120 		error = nfscl_maperr(td, error, 0, 0);
4121 		break;
4122 	}
4123 	return (error);
4124 }
4125 
4126 /*
4127  * nfs setextattr call
4128  */
4129 static int
4130 nfs_deleteextattr(struct vop_deleteextattr_args *ap)
4131 {
4132 	struct vnode *vp = ap->a_vp;
4133 	struct nfsmount *nmp;
4134 	struct nfsvattr nfsva;
4135 	int attrflag, error, ret;
4136 
4137 	nmp = VFSTONFS(vp->v_mount);
4138 	mtx_lock(&nmp->nm_mtx);
4139 	if (!NFSHASNFSV4(nmp) || nmp->nm_minorvers < NFSV42_MINORVERSION ||
4140 	    (nmp->nm_privflag & NFSMNTP_NOXATTR) != 0 ||
4141 	    ap->a_attrnamespace != EXTATTR_NAMESPACE_USER) {
4142 		mtx_unlock(&nmp->nm_mtx);
4143 		return (EOPNOTSUPP);
4144 	}
4145 	mtx_unlock(&nmp->nm_mtx);
4146 
4147 	/* Do the actual NFSv4.2 Optional Extended Attribute (RFC-8276) RPC. */
4148 	attrflag = 0;
4149 	error = nfsrpc_rmextattr(vp, ap->a_name, &nfsva, &attrflag, ap->a_cred,
4150 	    ap->a_td);
4151 	if (attrflag != 0) {
4152 		ret = nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0, 1);
4153 		if (error == 0 && ret != 0)
4154 			error = ret;
4155 	}
4156 
4157 	switch (error) {
4158 	case NFSERR_NOTSUPP:
4159 	case NFSERR_OPILLEGAL:
4160 		mtx_lock(&nmp->nm_mtx);
4161 		nmp->nm_privflag |= NFSMNTP_NOXATTR;
4162 		mtx_unlock(&nmp->nm_mtx);
4163 		error = EOPNOTSUPP;
4164 		break;
4165 	case NFSERR_NOXATTR:
4166 	case NFSERR_XATTR2BIG:
4167 		error = ENOATTR;
4168 		break;
4169 	default:
4170 		error = nfscl_maperr(ap->a_td, error, 0, 0);
4171 		break;
4172 	}
4173 	return (error);
4174 }
4175 
4176 /*
4177  * Return POSIX pathconf information applicable to nfs filesystems.
4178  */
4179 static int
4180 nfs_pathconf(struct vop_pathconf_args *ap)
4181 {
4182 	struct nfsv3_pathconf pc;
4183 	struct nfsvattr nfsva;
4184 	struct vnode *vp = ap->a_vp;
4185 	struct nfsmount *nmp;
4186 	struct thread *td = curthread;
4187 	off_t off;
4188 	bool eof;
4189 	int attrflag, error;
4190 
4191 	if ((NFS_ISV34(vp) && (ap->a_name == _PC_LINK_MAX ||
4192 	    ap->a_name == _PC_NAME_MAX || ap->a_name == _PC_CHOWN_RESTRICTED ||
4193 	    ap->a_name == _PC_NO_TRUNC)) ||
4194 	    (NFS_ISV4(vp) && ap->a_name == _PC_ACL_NFS4)) {
4195 		/*
4196 		 * Since only the above 4 a_names are returned by the NFSv3
4197 		 * Pathconf RPC, there is no point in doing it for others.
4198 		 * For NFSv4, the Pathconf RPC (actually a Getattr Op.) can
4199 		 * be used for _PC_NFS4_ACL as well.
4200 		 */
4201 		error = nfsrpc_pathconf(vp, &pc, td->td_ucred, td, &nfsva,
4202 		    &attrflag, NULL);
4203 		if (attrflag != 0)
4204 			(void) nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0,
4205 			    1);
4206 		if (error != 0)
4207 			return (error);
4208 	} else {
4209 		/*
4210 		 * For NFSv2 (or NFSv3 when not one of the above 4 a_names),
4211 		 * just fake them.
4212 		 */
4213 		pc.pc_linkmax = NFS_LINK_MAX;
4214 		pc.pc_namemax = NFS_MAXNAMLEN;
4215 		pc.pc_notrunc = 1;
4216 		pc.pc_chownrestricted = 1;
4217 		pc.pc_caseinsensitive = 0;
4218 		pc.pc_casepreserving = 1;
4219 		error = 0;
4220 	}
4221 	switch (ap->a_name) {
4222 	case _PC_LINK_MAX:
4223 #ifdef _LP64
4224 		*ap->a_retval = pc.pc_linkmax;
4225 #else
4226 		*ap->a_retval = MIN(LONG_MAX, pc.pc_linkmax);
4227 #endif
4228 		break;
4229 	case _PC_NAME_MAX:
4230 		*ap->a_retval = pc.pc_namemax;
4231 		break;
4232 	case _PC_PIPE_BUF:
4233 		if (ap->a_vp->v_type == VDIR || ap->a_vp->v_type == VFIFO)
4234 			*ap->a_retval = PIPE_BUF;
4235 		else
4236 			error = EINVAL;
4237 		break;
4238 	case _PC_CHOWN_RESTRICTED:
4239 		*ap->a_retval = pc.pc_chownrestricted;
4240 		break;
4241 	case _PC_NO_TRUNC:
4242 		*ap->a_retval = pc.pc_notrunc;
4243 		break;
4244 	case _PC_ACL_NFS4:
4245 		if (NFS_ISV4(vp) && nfsrv_useacl != 0 && attrflag != 0 &&
4246 		    NFSISSET_ATTRBIT(&nfsva.na_suppattr, NFSATTRBIT_ACL))
4247 			*ap->a_retval = 1;
4248 		else
4249 			*ap->a_retval = 0;
4250 		break;
4251 	case _PC_ACL_PATH_MAX:
4252 		if (NFS_ISV4(vp))
4253 			*ap->a_retval = ACL_MAX_ENTRIES;
4254 		else
4255 			*ap->a_retval = 3;
4256 		break;
4257 	case _PC_PRIO_IO:
4258 		*ap->a_retval = 0;
4259 		break;
4260 	case _PC_SYNC_IO:
4261 		*ap->a_retval = 0;
4262 		break;
4263 	case _PC_ALLOC_SIZE_MIN:
4264 		*ap->a_retval = vp->v_mount->mnt_stat.f_bsize;
4265 		break;
4266 	case _PC_FILESIZEBITS:
4267 		if (NFS_ISV34(vp))
4268 			*ap->a_retval = 64;
4269 		else
4270 			*ap->a_retval = 32;
4271 		break;
4272 	case _PC_REC_INCR_XFER_SIZE:
4273 		*ap->a_retval = vp->v_mount->mnt_stat.f_iosize;
4274 		break;
4275 	case _PC_REC_MAX_XFER_SIZE:
4276 		*ap->a_retval = -1; /* means ``unlimited'' */
4277 		break;
4278 	case _PC_REC_MIN_XFER_SIZE:
4279 		*ap->a_retval = vp->v_mount->mnt_stat.f_iosize;
4280 		break;
4281 	case _PC_REC_XFER_ALIGN:
4282 		*ap->a_retval = PAGE_SIZE;
4283 		break;
4284 	case _PC_SYMLINK_MAX:
4285 		*ap->a_retval = NFS_MAXPATHLEN;
4286 		break;
4287 	case _PC_MIN_HOLE_SIZE:
4288 		/* Only some NFSv4.2 servers support Seek for Holes. */
4289 		*ap->a_retval = 0;
4290 		nmp = VFSTONFS(vp->v_mount);
4291 		if (NFS_ISV4(vp) && nmp->nm_minorvers == NFSV42_MINORVERSION) {
4292 			/*
4293 			 * NFSv4.2 doesn't have an attribute for hole size,
4294 			 * so all we can do is see if the Seek operation is
4295 			 * supported and then use f_iosize as a "best guess".
4296 			 */
4297 			mtx_lock(&nmp->nm_mtx);
4298 			if ((nmp->nm_privflag & NFSMNTP_SEEKTESTED) == 0) {
4299 				mtx_unlock(&nmp->nm_mtx);
4300 				off = 0;
4301 				attrflag = 0;
4302 				error = nfsrpc_seek(vp, &off, &eof,
4303 				    NFSV4CONTENT_HOLE, td->td_ucred, &nfsva,
4304 				    &attrflag);
4305 				if (attrflag != 0)
4306 					nfscl_loadattrcache(&vp, &nfsva,
4307 					    NULL, NULL, 0, 1);
4308 				mtx_lock(&nmp->nm_mtx);
4309 				if (error == NFSERR_NOTSUPP)
4310 					nmp->nm_privflag |= NFSMNTP_SEEKTESTED;
4311 				else
4312 					nmp->nm_privflag |= NFSMNTP_SEEKTESTED |
4313 					    NFSMNTP_SEEK;
4314 				error = 0;
4315 			}
4316 			if ((nmp->nm_privflag & NFSMNTP_SEEK) != 0)
4317 				*ap->a_retval = vp->v_mount->mnt_stat.f_iosize;
4318 			mtx_unlock(&nmp->nm_mtx);
4319 		}
4320 		break;
4321 
4322 	default:
4323 		error = vop_stdpathconf(ap);
4324 		break;
4325 	}
4326 	return (error);
4327 }
4328