xref: /freebsd/sys/fs/nfsclient/nfs_clvnops.c (revision a7dea1671b87c07d2d266f836bfa8b58efc7c134)
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 /*
604  * nfs open vnode op
605  * Check to see if the type is ok
606  * and that deletion is not in progress.
607  * For paged in text files, you will need to flush the page cache
608  * if consistency is lost.
609  */
610 /* ARGSUSED */
611 static int
612 nfs_open(struct vop_open_args *ap)
613 {
614 	struct vnode *vp = ap->a_vp;
615 	struct nfsnode *np = VTONFS(vp);
616 	struct vattr vattr;
617 	int error;
618 	int fmode = ap->a_mode;
619 	struct ucred *cred;
620 	vm_object_t obj;
621 
622 	if (vp->v_type != VREG && vp->v_type != VDIR && vp->v_type != VLNK)
623 		return (EOPNOTSUPP);
624 
625 	/*
626 	 * For NFSv4, we need to do the Open Op before cache validation,
627 	 * so that we conform to RFC3530 Sec. 9.3.1.
628 	 */
629 	if (NFS_ISV4(vp)) {
630 		error = nfsrpc_open(vp, fmode, ap->a_cred, ap->a_td);
631 		if (error) {
632 			error = nfscl_maperr(ap->a_td, error, (uid_t)0,
633 			    (gid_t)0);
634 			return (error);
635 		}
636 	}
637 
638 	/*
639 	 * Now, if this Open will be doing reading, re-validate/flush the
640 	 * cache, so that Close/Open coherency is maintained.
641 	 */
642 	NFSLOCKNODE(np);
643 	if (np->n_flag & NMODIFIED) {
644 		NFSUNLOCKNODE(np);
645 		error = ncl_vinvalbuf(vp, V_SAVE, ap->a_td, 1);
646 		if (error == EINTR || error == EIO) {
647 			if (NFS_ISV4(vp))
648 				(void) nfsrpc_close(vp, 0, ap->a_td);
649 			return (error);
650 		}
651 		NFSLOCKNODE(np);
652 		np->n_attrstamp = 0;
653 		KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(vp);
654 		if (vp->v_type == VDIR)
655 			np->n_direofoffset = 0;
656 		NFSUNLOCKNODE(np);
657 		error = VOP_GETATTR(vp, &vattr, ap->a_cred);
658 		if (error) {
659 			if (NFS_ISV4(vp))
660 				(void) nfsrpc_close(vp, 0, ap->a_td);
661 			return (error);
662 		}
663 		NFSLOCKNODE(np);
664 		np->n_mtime = vattr.va_mtime;
665 		if (NFS_ISV4(vp))
666 			np->n_change = vattr.va_filerev;
667 	} else {
668 		NFSUNLOCKNODE(np);
669 		error = VOP_GETATTR(vp, &vattr, ap->a_cred);
670 		if (error) {
671 			if (NFS_ISV4(vp))
672 				(void) nfsrpc_close(vp, 0, ap->a_td);
673 			return (error);
674 		}
675 		NFSLOCKNODE(np);
676 		if ((NFS_ISV4(vp) && np->n_change != vattr.va_filerev) ||
677 		    NFS_TIMESPEC_COMPARE(&np->n_mtime, &vattr.va_mtime)) {
678 			if (vp->v_type == VDIR)
679 				np->n_direofoffset = 0;
680 			NFSUNLOCKNODE(np);
681 			error = ncl_vinvalbuf(vp, V_SAVE, ap->a_td, 1);
682 			if (error == EINTR || error == EIO) {
683 				if (NFS_ISV4(vp))
684 					(void) nfsrpc_close(vp, 0, ap->a_td);
685 				return (error);
686 			}
687 			NFSLOCKNODE(np);
688 			np->n_mtime = vattr.va_mtime;
689 			if (NFS_ISV4(vp))
690 				np->n_change = vattr.va_filerev;
691 		}
692 	}
693 
694 	/*
695 	 * If the object has >= 1 O_DIRECT active opens, we disable caching.
696 	 */
697 	if (newnfs_directio_enable && (fmode & O_DIRECT) &&
698 	    (vp->v_type == VREG)) {
699 		if (np->n_directio_opens == 0) {
700 			NFSUNLOCKNODE(np);
701 			error = ncl_vinvalbuf(vp, V_SAVE, ap->a_td, 1);
702 			if (error) {
703 				if (NFS_ISV4(vp))
704 					(void) nfsrpc_close(vp, 0, ap->a_td);
705 				return (error);
706 			}
707 			NFSLOCKNODE(np);
708 			np->n_flag |= NNONCACHE;
709 		}
710 		np->n_directio_opens++;
711 	}
712 
713 	/* If opened for writing via NFSv4.1 or later, mark that for pNFS. */
714 	if (NFSHASPNFS(VFSTONFS(vp->v_mount)) && (fmode & FWRITE) != 0)
715 		np->n_flag |= NWRITEOPENED;
716 
717 	/*
718 	 * If this is an open for writing, capture a reference to the
719 	 * credentials, so they can be used by ncl_putpages(). Using
720 	 * these write credentials is preferable to the credentials of
721 	 * whatever thread happens to be doing the VOP_PUTPAGES() since
722 	 * the write RPCs are less likely to fail with EACCES.
723 	 */
724 	if ((fmode & FWRITE) != 0) {
725 		cred = np->n_writecred;
726 		np->n_writecred = crhold(ap->a_cred);
727 	} else
728 		cred = NULL;
729 	NFSUNLOCKNODE(np);
730 
731 	if (cred != NULL)
732 		crfree(cred);
733 	vnode_create_vobject(vp, vattr.va_size, ap->a_td);
734 
735 	/*
736 	 * If the text file has been mmap'd, flush any dirty pages to the
737 	 * buffer cache and then...
738 	 * Make sure all writes are pushed to the NFS server.  If this is not
739 	 * done, the modify time of the file can change while the text
740 	 * file is being executed.  This will cause the process that is
741 	 * executing the text file to be terminated.
742 	 */
743 	if (vp->v_writecount <= -1) {
744 		if ((obj = vp->v_object) != NULL &&
745 		    vm_object_mightbedirty(obj)) {
746 			VM_OBJECT_WLOCK(obj);
747 			vm_object_page_clean(obj, 0, 0, OBJPC_SYNC);
748 			VM_OBJECT_WUNLOCK(obj);
749 		}
750 
751 		/* Now, flush the buffer cache. */
752 		ncl_flush(vp, MNT_WAIT, curthread, 0, 0);
753 
754 		/* And, finally, make sure that n_mtime is up to date. */
755 		np = VTONFS(vp);
756 		NFSLOCKNODE(np);
757 		np->n_mtime = np->n_vattr.na_mtime;
758 		NFSUNLOCKNODE(np);
759 	}
760 	return (0);
761 }
762 
763 /*
764  * nfs close vnode op
765  * What an NFS client should do upon close after writing is a debatable issue.
766  * Most NFS clients push delayed writes to the server upon close, basically for
767  * two reasons:
768  * 1 - So that any write errors may be reported back to the client process
769  *     doing the close system call. By far the two most likely errors are
770  *     NFSERR_NOSPC and NFSERR_DQUOT to indicate space allocation failure.
771  * 2 - To put a worst case upper bound on cache inconsistency between
772  *     multiple clients for the file.
773  * There is also a consistency problem for Version 2 of the protocol w.r.t.
774  * not being able to tell if other clients are writing a file concurrently,
775  * since there is no way of knowing if the changed modify time in the reply
776  * is only due to the write for this client.
777  * (NFS Version 3 provides weak cache consistency data in the reply that
778  *  should be sufficient to detect and handle this case.)
779  *
780  * The current code does the following:
781  * for NFS Version 2 - play it safe and flush/invalidate all dirty buffers
782  * for NFS Version 3 - flush dirty buffers to the server but don't invalidate
783  *                     or commit them (this satisfies 1 and 2 except for the
784  *                     case where the server crashes after this close but
785  *                     before the commit RPC, which is felt to be "good
786  *                     enough". Changing the last argument to ncl_flush() to
787  *                     a 1 would force a commit operation, if it is felt a
788  *                     commit is necessary now.
789  * for NFS Version 4 - flush the dirty buffers and commit them, if
790  *		       nfscl_mustflush() says this is necessary.
791  *                     It is necessary if there is no write delegation held,
792  *                     in order to satisfy open/close coherency.
793  *                     If the file isn't cached on local stable storage,
794  *                     it may be necessary in order to detect "out of space"
795  *                     errors from the server, if the write delegation
796  *                     issued by the server doesn't allow the file to grow.
797  */
798 /* ARGSUSED */
799 static int
800 nfs_close(struct vop_close_args *ap)
801 {
802 	struct vnode *vp = ap->a_vp;
803 	struct nfsnode *np = VTONFS(vp);
804 	struct nfsvattr nfsva;
805 	struct ucred *cred;
806 	int error = 0, ret, localcred = 0;
807 	int fmode = ap->a_fflag;
808 
809 	if (NFSCL_FORCEDISM(vp->v_mount))
810 		return (0);
811 	/*
812 	 * During shutdown, a_cred isn't valid, so just use root.
813 	 */
814 	if (ap->a_cred == NOCRED) {
815 		cred = newnfs_getcred();
816 		localcred = 1;
817 	} else {
818 		cred = ap->a_cred;
819 	}
820 	if (vp->v_type == VREG) {
821 	    /*
822 	     * Examine and clean dirty pages, regardless of NMODIFIED.
823 	     * This closes a major hole in close-to-open consistency.
824 	     * We want to push out all dirty pages (and buffers) on
825 	     * close, regardless of whether they were dirtied by
826 	     * mmap'ed writes or via write().
827 	     */
828 	    if (nfs_clean_pages_on_close && vp->v_object) {
829 		VM_OBJECT_WLOCK(vp->v_object);
830 		vm_object_page_clean(vp->v_object, 0, 0, 0);
831 		VM_OBJECT_WUNLOCK(vp->v_object);
832 	    }
833 	    NFSLOCKNODE(np);
834 	    if (np->n_flag & NMODIFIED) {
835 		NFSUNLOCKNODE(np);
836 		if (NFS_ISV3(vp)) {
837 		    /*
838 		     * Under NFSv3 we have dirty buffers to dispose of.  We
839 		     * must flush them to the NFS server.  We have the option
840 		     * of waiting all the way through the commit rpc or just
841 		     * waiting for the initial write.  The default is to only
842 		     * wait through the initial write so the data is in the
843 		     * server's cache, which is roughly similar to the state
844 		     * a standard disk subsystem leaves the file in on close().
845 		     *
846 		     * We cannot clear the NMODIFIED bit in np->n_flag due to
847 		     * potential races with other processes, and certainly
848 		     * cannot clear it if we don't commit.
849 		     * These races occur when there is no longer the old
850 		     * traditional vnode locking implemented for Vnode Ops.
851 		     */
852 		    int cm = newnfs_commit_on_close ? 1 : 0;
853 		    error = ncl_flush(vp, MNT_WAIT, ap->a_td, cm, 0);
854 		    /* np->n_flag &= ~NMODIFIED; */
855 		} else if (NFS_ISV4(vp)) {
856 			if (nfscl_mustflush(vp) != 0) {
857 				int cm = newnfs_commit_on_close ? 1 : 0;
858 				error = ncl_flush(vp, MNT_WAIT, ap->a_td,
859 				    cm, 0);
860 				/*
861 				 * as above w.r.t races when clearing
862 				 * NMODIFIED.
863 				 * np->n_flag &= ~NMODIFIED;
864 				 */
865 			}
866 		} else {
867 			error = ncl_vinvalbuf(vp, V_SAVE, ap->a_td, 1);
868 		}
869 		NFSLOCKNODE(np);
870 	    }
871  	    /*
872  	     * Invalidate the attribute cache in all cases.
873  	     * An open is going to fetch fresh attrs any way, other procs
874  	     * on this node that have file open will be forced to do an
875  	     * otw attr fetch, but this is safe.
876 	     * --> A user found that their RPC count dropped by 20% when
877 	     *     this was commented out and I can't see any requirement
878 	     *     for it, so I've disabled it when negative lookups are
879 	     *     enabled. (What does this have to do with negative lookup
880 	     *     caching? Well nothing, except it was reported by the
881 	     *     same user that needed negative lookup caching and I wanted
882 	     *     there to be a way to disable it to see if it
883 	     *     is the cause of some caching/coherency issue that might
884 	     *     crop up.)
885  	     */
886 	    if (VFSTONFS(vp->v_mount)->nm_negnametimeo == 0) {
887 		    np->n_attrstamp = 0;
888 		    KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(vp);
889 	    }
890 	    if (np->n_flag & NWRITEERR) {
891 		np->n_flag &= ~NWRITEERR;
892 		error = np->n_error;
893 	    }
894 	    NFSUNLOCKNODE(np);
895 	}
896 
897 	if (NFS_ISV4(vp)) {
898 		/*
899 		 * Get attributes so "change" is up to date.
900 		 */
901 		if (error == 0 && nfscl_mustflush(vp) != 0 &&
902 		    vp->v_type == VREG &&
903 		    (VFSTONFS(vp->v_mount)->nm_flag & NFSMNT_NOCTO) == 0) {
904 			ret = nfsrpc_getattr(vp, cred, ap->a_td, &nfsva,
905 			    NULL);
906 			if (!ret) {
907 				np->n_change = nfsva.na_filerev;
908 				(void) nfscl_loadattrcache(&vp, &nfsva, NULL,
909 				    NULL, 0, 0);
910 			}
911 		}
912 
913 		/*
914 		 * and do the close.
915 		 */
916 		ret = nfsrpc_close(vp, 0, ap->a_td);
917 		if (!error && ret)
918 			error = ret;
919 		if (error)
920 			error = nfscl_maperr(ap->a_td, error, (uid_t)0,
921 			    (gid_t)0);
922 	}
923 	if (newnfs_directio_enable)
924 		KASSERT((np->n_directio_asyncwr == 0),
925 			("nfs_close: dirty unflushed (%d) directio buffers\n",
926 			 np->n_directio_asyncwr));
927 	if (newnfs_directio_enable && (fmode & O_DIRECT) && (vp->v_type == VREG)) {
928 		NFSLOCKNODE(np);
929 		KASSERT((np->n_directio_opens > 0),
930 			("nfs_close: unexpectedly value (0) of n_directio_opens\n"));
931 		np->n_directio_opens--;
932 		if (np->n_directio_opens == 0)
933 			np->n_flag &= ~NNONCACHE;
934 		NFSUNLOCKNODE(np);
935 	}
936 	if (localcred)
937 		NFSFREECRED(cred);
938 	return (error);
939 }
940 
941 /*
942  * nfs getattr call from vfs.
943  */
944 static int
945 nfs_getattr(struct vop_getattr_args *ap)
946 {
947 	struct vnode *vp = ap->a_vp;
948 	struct thread *td = curthread;	/* XXX */
949 	struct nfsnode *np = VTONFS(vp);
950 	int error = 0;
951 	struct nfsvattr nfsva;
952 	struct vattr *vap = ap->a_vap;
953 	struct vattr vattr;
954 
955 	/*
956 	 * Update local times for special files.
957 	 */
958 	NFSLOCKNODE(np);
959 	if (np->n_flag & (NACC | NUPD))
960 		np->n_flag |= NCHG;
961 	NFSUNLOCKNODE(np);
962 	/*
963 	 * First look in the cache.
964 	 */
965 	if (ncl_getattrcache(vp, &vattr) == 0) {
966 		vap->va_type = vattr.va_type;
967 		vap->va_mode = vattr.va_mode;
968 		vap->va_nlink = vattr.va_nlink;
969 		vap->va_uid = vattr.va_uid;
970 		vap->va_gid = vattr.va_gid;
971 		vap->va_fsid = vattr.va_fsid;
972 		vap->va_fileid = vattr.va_fileid;
973 		vap->va_size = vattr.va_size;
974 		vap->va_blocksize = vattr.va_blocksize;
975 		vap->va_atime = vattr.va_atime;
976 		vap->va_mtime = vattr.va_mtime;
977 		vap->va_ctime = vattr.va_ctime;
978 		vap->va_gen = vattr.va_gen;
979 		vap->va_flags = vattr.va_flags;
980 		vap->va_rdev = vattr.va_rdev;
981 		vap->va_bytes = vattr.va_bytes;
982 		vap->va_filerev = vattr.va_filerev;
983 		/*
984 		 * Get the local modify time for the case of a write
985 		 * delegation.
986 		 */
987 		nfscl_deleggetmodtime(vp, &vap->va_mtime);
988 		return (0);
989 	}
990 
991 	if (NFS_ISV34(vp) && nfs_prime_access_cache &&
992 	    nfsaccess_cache_timeout > 0) {
993 		NFSINCRGLOBAL(nfsstatsv1.accesscache_misses);
994 		nfs34_access_otw(vp, NFSACCESS_ALL, td, ap->a_cred, NULL);
995 		if (ncl_getattrcache(vp, ap->a_vap) == 0) {
996 			nfscl_deleggetmodtime(vp, &ap->a_vap->va_mtime);
997 			return (0);
998 		}
999 	}
1000 	error = nfsrpc_getattr(vp, ap->a_cred, td, &nfsva, NULL);
1001 	if (!error)
1002 		error = nfscl_loadattrcache(&vp, &nfsva, vap, NULL, 0, 0);
1003 	if (!error) {
1004 		/*
1005 		 * Get the local modify time for the case of a write
1006 		 * delegation.
1007 		 */
1008 		nfscl_deleggetmodtime(vp, &vap->va_mtime);
1009 	} else if (NFS_ISV4(vp)) {
1010 		error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0);
1011 	}
1012 	return (error);
1013 }
1014 
1015 /*
1016  * nfs setattr call.
1017  */
1018 static int
1019 nfs_setattr(struct vop_setattr_args *ap)
1020 {
1021 	struct vnode *vp = ap->a_vp;
1022 	struct nfsnode *np = VTONFS(vp);
1023 	struct thread *td = curthread;	/* XXX */
1024 	struct vattr *vap = ap->a_vap;
1025 	int error = 0;
1026 	u_quad_t tsize;
1027 
1028 #ifndef nolint
1029 	tsize = (u_quad_t)0;
1030 #endif
1031 
1032 	/*
1033 	 * Setting of flags and marking of atimes are not supported.
1034 	 */
1035 	if (vap->va_flags != VNOVAL)
1036 		return (EOPNOTSUPP);
1037 
1038 	/*
1039 	 * Disallow write attempts if the filesystem is mounted read-only.
1040 	 */
1041   	if ((vap->va_flags != VNOVAL || vap->va_uid != (uid_t)VNOVAL ||
1042 	    vap->va_gid != (gid_t)VNOVAL || vap->va_atime.tv_sec != VNOVAL ||
1043 	    vap->va_mtime.tv_sec != VNOVAL || vap->va_mode != (mode_t)VNOVAL) &&
1044 	    (vp->v_mount->mnt_flag & MNT_RDONLY))
1045 		return (EROFS);
1046 	if (vap->va_size != VNOVAL) {
1047  		switch (vp->v_type) {
1048  		case VDIR:
1049  			return (EISDIR);
1050  		case VCHR:
1051  		case VBLK:
1052  		case VSOCK:
1053  		case VFIFO:
1054 			if (vap->va_mtime.tv_sec == VNOVAL &&
1055 			    vap->va_atime.tv_sec == VNOVAL &&
1056 			    vap->va_mode == (mode_t)VNOVAL &&
1057 			    vap->va_uid == (uid_t)VNOVAL &&
1058 			    vap->va_gid == (gid_t)VNOVAL)
1059 				return (0);
1060  			vap->va_size = VNOVAL;
1061  			break;
1062  		default:
1063 			/*
1064 			 * Disallow write attempts if the filesystem is
1065 			 * mounted read-only.
1066 			 */
1067 			if (vp->v_mount->mnt_flag & MNT_RDONLY)
1068 				return (EROFS);
1069 			/*
1070 			 *  We run vnode_pager_setsize() early (why?),
1071 			 * we must set np->n_size now to avoid vinvalbuf
1072 			 * V_SAVE races that might setsize a lower
1073 			 * value.
1074 			 */
1075 			NFSLOCKNODE(np);
1076 			tsize = np->n_size;
1077 			NFSUNLOCKNODE(np);
1078 			error = ncl_meta_setsize(vp, td, vap->va_size);
1079 			NFSLOCKNODE(np);
1080  			if (np->n_flag & NMODIFIED) {
1081 			    tsize = np->n_size;
1082 			    NFSUNLOCKNODE(np);
1083 			    error = ncl_vinvalbuf(vp, vap->va_size == 0 ?
1084 			        0 : V_SAVE, td, 1);
1085 			    if (error != 0) {
1086 				    vnode_pager_setsize(vp, tsize);
1087 				    return (error);
1088 			    }
1089 			    /*
1090 			     * Call nfscl_delegmodtime() to set the modify time
1091 			     * locally, as required.
1092 			     */
1093 			    nfscl_delegmodtime(vp);
1094  			} else
1095 			    NFSUNLOCKNODE(np);
1096 			/*
1097 			 * np->n_size has already been set to vap->va_size
1098 			 * in ncl_meta_setsize(). We must set it again since
1099 			 * nfs_loadattrcache() could be called through
1100 			 * ncl_meta_setsize() and could modify np->n_size.
1101 			 */
1102 			NFSLOCKNODE(np);
1103  			np->n_vattr.na_size = np->n_size = vap->va_size;
1104 			NFSUNLOCKNODE(np);
1105   		}
1106   	} else {
1107 		NFSLOCKNODE(np);
1108 		if ((vap->va_mtime.tv_sec != VNOVAL || vap->va_atime.tv_sec != VNOVAL) &&
1109 		    (np->n_flag & NMODIFIED) && vp->v_type == VREG) {
1110 			NFSUNLOCKNODE(np);
1111 			error = ncl_vinvalbuf(vp, V_SAVE, td, 1);
1112 			if (error == EINTR || error == EIO)
1113 				return (error);
1114 		} else
1115 			NFSUNLOCKNODE(np);
1116 	}
1117 	error = nfs_setattrrpc(vp, vap, ap->a_cred, td);
1118 	if (error && vap->va_size != VNOVAL) {
1119 		NFSLOCKNODE(np);
1120 		np->n_size = np->n_vattr.na_size = tsize;
1121 		vnode_pager_setsize(vp, tsize);
1122 		NFSUNLOCKNODE(np);
1123 	}
1124 	return (error);
1125 }
1126 
1127 /*
1128  * Do an nfs setattr rpc.
1129  */
1130 static int
1131 nfs_setattrrpc(struct vnode *vp, struct vattr *vap, struct ucred *cred,
1132     struct thread *td)
1133 {
1134 	struct nfsnode *np = VTONFS(vp);
1135 	int error, ret, attrflag, i;
1136 	struct nfsvattr nfsva;
1137 
1138 	if (NFS_ISV34(vp)) {
1139 		NFSLOCKNODE(np);
1140 		for (i = 0; i < NFS_ACCESSCACHESIZE; i++)
1141 			np->n_accesscache[i].stamp = 0;
1142 		np->n_flag |= NDELEGMOD;
1143 		NFSUNLOCKNODE(np);
1144 		KDTRACE_NFS_ACCESSCACHE_FLUSH_DONE(vp);
1145 	}
1146 	error = nfsrpc_setattr(vp, vap, NULL, cred, td, &nfsva, &attrflag,
1147 	    NULL);
1148 	if (attrflag) {
1149 		ret = nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0, 1);
1150 		if (ret && !error)
1151 			error = ret;
1152 	}
1153 	if (error && NFS_ISV4(vp))
1154 		error = nfscl_maperr(td, error, vap->va_uid, vap->va_gid);
1155 	return (error);
1156 }
1157 
1158 /*
1159  * nfs lookup call, one step at a time...
1160  * First look in cache
1161  * If not found, unlock the directory nfsnode and do the rpc
1162  */
1163 static int
1164 nfs_lookup(struct vop_lookup_args *ap)
1165 {
1166 	struct componentname *cnp = ap->a_cnp;
1167 	struct vnode *dvp = ap->a_dvp;
1168 	struct vnode **vpp = ap->a_vpp;
1169 	struct mount *mp = dvp->v_mount;
1170 	int flags = cnp->cn_flags;
1171 	struct vnode *newvp;
1172 	struct nfsmount *nmp;
1173 	struct nfsnode *np, *newnp;
1174 	int error = 0, attrflag, dattrflag, ltype, ncticks;
1175 	struct thread *td = cnp->cn_thread;
1176 	struct nfsfh *nfhp;
1177 	struct nfsvattr dnfsva, nfsva;
1178 	struct vattr vattr;
1179 	struct timespec nctime;
1180 
1181 	*vpp = NULLVP;
1182 	if ((flags & ISLASTCN) && (mp->mnt_flag & MNT_RDONLY) &&
1183 	    (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME))
1184 		return (EROFS);
1185 	if (dvp->v_type != VDIR)
1186 		return (ENOTDIR);
1187 	nmp = VFSTONFS(mp);
1188 	np = VTONFS(dvp);
1189 
1190 	/* For NFSv4, wait until any remove is done. */
1191 	NFSLOCKNODE(np);
1192 	while (NFSHASNFSV4(nmp) && (np->n_flag & NREMOVEINPROG)) {
1193 		np->n_flag |= NREMOVEWANT;
1194 		(void) msleep((caddr_t)np, &np->n_mtx, PZERO, "nfslkup", 0);
1195 	}
1196 	NFSUNLOCKNODE(np);
1197 
1198 	if ((error = VOP_ACCESS(dvp, VEXEC, cnp->cn_cred, td)) != 0)
1199 		return (error);
1200 	error = cache_lookup(dvp, vpp, cnp, &nctime, &ncticks);
1201 	if (error > 0 && error != ENOENT)
1202 		return (error);
1203 	if (error == -1) {
1204 		/*
1205 		 * Lookups of "." are special and always return the
1206 		 * current directory.  cache_lookup() already handles
1207 		 * associated locking bookkeeping, etc.
1208 		 */
1209 		if (cnp->cn_namelen == 1 && cnp->cn_nameptr[0] == '.') {
1210 			/* XXX: Is this really correct? */
1211 			if (cnp->cn_nameiop != LOOKUP &&
1212 			    (flags & ISLASTCN))
1213 				cnp->cn_flags |= SAVENAME;
1214 			return (0);
1215 		}
1216 
1217 		/*
1218 		 * We only accept a positive hit in the cache if the
1219 		 * change time of the file matches our cached copy.
1220 		 * Otherwise, we discard the cache entry and fallback
1221 		 * to doing a lookup RPC.  We also only trust cache
1222 		 * entries for less than nm_nametimeo seconds.
1223 		 *
1224 		 * To better handle stale file handles and attributes,
1225 		 * clear the attribute cache of this node if it is a
1226 		 * leaf component, part of an open() call, and not
1227 		 * locally modified before fetching the attributes.
1228 		 * This should allow stale file handles to be detected
1229 		 * here where we can fall back to a LOOKUP RPC to
1230 		 * recover rather than having nfs_open() detect the
1231 		 * stale file handle and failing open(2) with ESTALE.
1232 		 */
1233 		newvp = *vpp;
1234 		newnp = VTONFS(newvp);
1235 		if (!(nmp->nm_flag & NFSMNT_NOCTO) &&
1236 		    (flags & (ISLASTCN | ISOPEN)) == (ISLASTCN | ISOPEN) &&
1237 		    !(newnp->n_flag & NMODIFIED)) {
1238 			NFSLOCKNODE(newnp);
1239 			newnp->n_attrstamp = 0;
1240 			KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(newvp);
1241 			NFSUNLOCKNODE(newnp);
1242 		}
1243 		if (nfscl_nodeleg(newvp, 0) == 0 ||
1244 		    ((u_int)(ticks - ncticks) < (nmp->nm_nametimeo * hz) &&
1245 		    VOP_GETATTR(newvp, &vattr, cnp->cn_cred) == 0 &&
1246 		    timespeccmp(&vattr.va_ctime, &nctime, ==))) {
1247 			NFSINCRGLOBAL(nfsstatsv1.lookupcache_hits);
1248 			if (cnp->cn_nameiop != LOOKUP &&
1249 			    (flags & ISLASTCN))
1250 				cnp->cn_flags |= SAVENAME;
1251 			return (0);
1252 		}
1253 		cache_purge(newvp);
1254 		if (dvp != newvp)
1255 			vput(newvp);
1256 		else
1257 			vrele(newvp);
1258 		*vpp = NULLVP;
1259 	} else if (error == ENOENT) {
1260 		if (VN_IS_DOOMED(dvp))
1261 			return (ENOENT);
1262 		/*
1263 		 * We only accept a negative hit in the cache if the
1264 		 * modification time of the parent directory matches
1265 		 * the cached copy in the name cache entry.
1266 		 * Otherwise, we discard all of the negative cache
1267 		 * entries for this directory.  We also only trust
1268 		 * negative cache entries for up to nm_negnametimeo
1269 		 * seconds.
1270 		 */
1271 		if ((u_int)(ticks - ncticks) < (nmp->nm_negnametimeo * hz) &&
1272 		    VOP_GETATTR(dvp, &vattr, cnp->cn_cred) == 0 &&
1273 		    timespeccmp(&vattr.va_mtime, &nctime, ==)) {
1274 			NFSINCRGLOBAL(nfsstatsv1.lookupcache_hits);
1275 			return (ENOENT);
1276 		}
1277 		cache_purge_negative(dvp);
1278 	}
1279 
1280 	newvp = NULLVP;
1281 	NFSINCRGLOBAL(nfsstatsv1.lookupcache_misses);
1282 	error = nfsrpc_lookup(dvp, cnp->cn_nameptr, cnp->cn_namelen,
1283 	    cnp->cn_cred, td, &dnfsva, &nfsva, &nfhp, &attrflag, &dattrflag,
1284 	    NULL);
1285 	if (dattrflag)
1286 		(void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, NULL, 0, 1);
1287 	if (error) {
1288 		if (newvp != NULLVP) {
1289 			vput(newvp);
1290 			*vpp = NULLVP;
1291 		}
1292 
1293 		if (error != ENOENT) {
1294 			if (NFS_ISV4(dvp))
1295 				error = nfscl_maperr(td, error, (uid_t)0,
1296 				    (gid_t)0);
1297 			return (error);
1298 		}
1299 
1300 		/* The requested file was not found. */
1301 		if ((cnp->cn_nameiop == CREATE || cnp->cn_nameiop == RENAME) &&
1302 		    (flags & ISLASTCN)) {
1303 			/*
1304 			 * XXX: UFS does a full VOP_ACCESS(dvp,
1305 			 * VWRITE) here instead of just checking
1306 			 * MNT_RDONLY.
1307 			 */
1308 			if (mp->mnt_flag & MNT_RDONLY)
1309 				return (EROFS);
1310 			cnp->cn_flags |= SAVENAME;
1311 			return (EJUSTRETURN);
1312 		}
1313 
1314 		if ((cnp->cn_flags & MAKEENTRY) != 0 && dattrflag) {
1315 			/*
1316 			 * Cache the modification time of the parent
1317 			 * directory from the post-op attributes in
1318 			 * the name cache entry.  The negative cache
1319 			 * entry will be ignored once the directory
1320 			 * has changed.  Don't bother adding the entry
1321 			 * if the directory has already changed.
1322 			 */
1323 			NFSLOCKNODE(np);
1324 			if (timespeccmp(&np->n_vattr.na_mtime,
1325 			    &dnfsva.na_mtime, ==)) {
1326 				NFSUNLOCKNODE(np);
1327 				cache_enter_time(dvp, NULL, cnp,
1328 				    &dnfsva.na_mtime, NULL);
1329 			} else
1330 				NFSUNLOCKNODE(np);
1331 		}
1332 		return (ENOENT);
1333 	}
1334 
1335 	/*
1336 	 * Handle RENAME case...
1337 	 */
1338 	if (cnp->cn_nameiop == RENAME && (flags & ISLASTCN)) {
1339 		if (NFS_CMPFH(np, nfhp->nfh_fh, nfhp->nfh_len)) {
1340 			free(nfhp, M_NFSFH);
1341 			return (EISDIR);
1342 		}
1343 		error = nfscl_nget(mp, dvp, nfhp, cnp, td, &np, NULL,
1344 		    LK_EXCLUSIVE);
1345 		if (error)
1346 			return (error);
1347 		newvp = NFSTOV(np);
1348 		if (attrflag)
1349 			(void) nfscl_loadattrcache(&newvp, &nfsva, NULL, NULL,
1350 			    0, 1);
1351 		*vpp = newvp;
1352 		cnp->cn_flags |= SAVENAME;
1353 		return (0);
1354 	}
1355 
1356 	if (flags & ISDOTDOT) {
1357 		ltype = NFSVOPISLOCKED(dvp);
1358 		error = vfs_busy(mp, MBF_NOWAIT);
1359 		if (error != 0) {
1360 			vfs_ref(mp);
1361 			NFSVOPUNLOCK(dvp);
1362 			error = vfs_busy(mp, 0);
1363 			NFSVOPLOCK(dvp, ltype | LK_RETRY);
1364 			vfs_rel(mp);
1365 			if (error == 0 && VN_IS_DOOMED(dvp)) {
1366 				vfs_unbusy(mp);
1367 				error = ENOENT;
1368 			}
1369 			if (error != 0)
1370 				return (error);
1371 		}
1372 		NFSVOPUNLOCK(dvp);
1373 		error = nfscl_nget(mp, dvp, nfhp, cnp, td, &np, NULL,
1374 		    cnp->cn_lkflags);
1375 		if (error == 0)
1376 			newvp = NFSTOV(np);
1377 		vfs_unbusy(mp);
1378 		if (newvp != dvp)
1379 			NFSVOPLOCK(dvp, ltype | LK_RETRY);
1380 		if (VN_IS_DOOMED(dvp)) {
1381 			if (error == 0) {
1382 				if (newvp == dvp)
1383 					vrele(newvp);
1384 				else
1385 					vput(newvp);
1386 			}
1387 			error = ENOENT;
1388 		}
1389 		if (error != 0)
1390 			return (error);
1391 		if (attrflag)
1392 			(void) nfscl_loadattrcache(&newvp, &nfsva, NULL, NULL,
1393 			    0, 1);
1394 	} else if (NFS_CMPFH(np, nfhp->nfh_fh, nfhp->nfh_len)) {
1395 		free(nfhp, M_NFSFH);
1396 		VREF(dvp);
1397 		newvp = dvp;
1398 		if (attrflag)
1399 			(void) nfscl_loadattrcache(&newvp, &nfsva, NULL, NULL,
1400 			    0, 1);
1401 	} else {
1402 		error = nfscl_nget(mp, dvp, nfhp, cnp, td, &np, NULL,
1403 		    cnp->cn_lkflags);
1404 		if (error)
1405 			return (error);
1406 		newvp = NFSTOV(np);
1407 		if (attrflag)
1408 			(void) nfscl_loadattrcache(&newvp, &nfsva, NULL, NULL,
1409 			    0, 1);
1410 		else if ((flags & (ISLASTCN | ISOPEN)) == (ISLASTCN | ISOPEN) &&
1411 		    !(np->n_flag & NMODIFIED)) {
1412 			/*
1413 			 * Flush the attribute cache when opening a
1414 			 * leaf node to ensure that fresh attributes
1415 			 * are fetched in nfs_open() since we did not
1416 			 * fetch attributes from the LOOKUP reply.
1417 			 */
1418 			NFSLOCKNODE(np);
1419 			np->n_attrstamp = 0;
1420 			KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(newvp);
1421 			NFSUNLOCKNODE(np);
1422 		}
1423 	}
1424 	if (cnp->cn_nameiop != LOOKUP && (flags & ISLASTCN))
1425 		cnp->cn_flags |= SAVENAME;
1426 	if ((cnp->cn_flags & MAKEENTRY) &&
1427 	    (cnp->cn_nameiop != DELETE || !(flags & ISLASTCN)) &&
1428 	    attrflag != 0 && (newvp->v_type != VDIR || dattrflag != 0))
1429 		cache_enter_time(dvp, newvp, cnp, &nfsva.na_ctime,
1430 		    newvp->v_type != VDIR ? NULL : &dnfsva.na_ctime);
1431 	*vpp = newvp;
1432 	return (0);
1433 }
1434 
1435 /*
1436  * nfs read call.
1437  * Just call ncl_bioread() to do the work.
1438  */
1439 static int
1440 nfs_read(struct vop_read_args *ap)
1441 {
1442 	struct vnode *vp = ap->a_vp;
1443 
1444 	switch (vp->v_type) {
1445 	case VREG:
1446 		return (ncl_bioread(vp, ap->a_uio, ap->a_ioflag, ap->a_cred));
1447 	case VDIR:
1448 		return (EISDIR);
1449 	default:
1450 		return (EOPNOTSUPP);
1451 	}
1452 }
1453 
1454 /*
1455  * nfs readlink call
1456  */
1457 static int
1458 nfs_readlink(struct vop_readlink_args *ap)
1459 {
1460 	struct vnode *vp = ap->a_vp;
1461 
1462 	if (vp->v_type != VLNK)
1463 		return (EINVAL);
1464 	return (ncl_bioread(vp, ap->a_uio, 0, ap->a_cred));
1465 }
1466 
1467 /*
1468  * Do a readlink rpc.
1469  * Called by ncl_doio() from below the buffer cache.
1470  */
1471 int
1472 ncl_readlinkrpc(struct vnode *vp, struct uio *uiop, struct ucred *cred)
1473 {
1474 	int error, ret, attrflag;
1475 	struct nfsvattr nfsva;
1476 
1477 	error = nfsrpc_readlink(vp, uiop, cred, uiop->uio_td, &nfsva,
1478 	    &attrflag, NULL);
1479 	if (attrflag) {
1480 		ret = nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0, 1);
1481 		if (ret && !error)
1482 			error = ret;
1483 	}
1484 	if (error && NFS_ISV4(vp))
1485 		error = nfscl_maperr(uiop->uio_td, error, (uid_t)0, (gid_t)0);
1486 	return (error);
1487 }
1488 
1489 /*
1490  * nfs read rpc call
1491  * Ditto above
1492  */
1493 int
1494 ncl_readrpc(struct vnode *vp, struct uio *uiop, struct ucred *cred)
1495 {
1496 	int error, ret, attrflag;
1497 	struct nfsvattr nfsva;
1498 	struct nfsmount *nmp;
1499 
1500 	nmp = VFSTONFS(vnode_mount(vp));
1501 	error = EIO;
1502 	attrflag = 0;
1503 	if (NFSHASPNFS(nmp))
1504 		error = nfscl_doiods(vp, uiop, NULL, NULL,
1505 		    NFSV4OPEN_ACCESSREAD, 0, cred, uiop->uio_td);
1506 	NFSCL_DEBUG(4, "readrpc: aft doiods=%d\n", error);
1507 	if (error != 0)
1508 		error = nfsrpc_read(vp, uiop, cred, uiop->uio_td, &nfsva,
1509 		    &attrflag, NULL);
1510 	if (attrflag) {
1511 		ret = nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0, 1);
1512 		if (ret && !error)
1513 			error = ret;
1514 	}
1515 	if (error && NFS_ISV4(vp))
1516 		error = nfscl_maperr(uiop->uio_td, error, (uid_t)0, (gid_t)0);
1517 	return (error);
1518 }
1519 
1520 /*
1521  * nfs write call
1522  */
1523 int
1524 ncl_writerpc(struct vnode *vp, struct uio *uiop, struct ucred *cred,
1525     int *iomode, int *must_commit, int called_from_strategy)
1526 {
1527 	struct nfsvattr nfsva;
1528 	int error, attrflag, ret;
1529 	struct nfsmount *nmp;
1530 
1531 	nmp = VFSTONFS(vnode_mount(vp));
1532 	error = EIO;
1533 	attrflag = 0;
1534 	if (NFSHASPNFS(nmp))
1535 		error = nfscl_doiods(vp, uiop, iomode, must_commit,
1536 		    NFSV4OPEN_ACCESSWRITE, 0, cred, uiop->uio_td);
1537 	NFSCL_DEBUG(4, "writerpc: aft doiods=%d\n", error);
1538 	if (error != 0)
1539 		error = nfsrpc_write(vp, uiop, iomode, must_commit, cred,
1540 		    uiop->uio_td, &nfsva, &attrflag, NULL,
1541 		    called_from_strategy);
1542 	if (attrflag) {
1543 		if (VTONFS(vp)->n_flag & ND_NFSV4)
1544 			ret = nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 1,
1545 			    1);
1546 		else
1547 			ret = nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0,
1548 			    1);
1549 		if (ret && !error)
1550 			error = ret;
1551 	}
1552 	if (DOINGASYNC(vp))
1553 		*iomode = NFSWRITE_FILESYNC;
1554 	if (error && NFS_ISV4(vp))
1555 		error = nfscl_maperr(uiop->uio_td, error, (uid_t)0, (gid_t)0);
1556 	return (error);
1557 }
1558 
1559 /*
1560  * nfs mknod rpc
1561  * For NFS v2 this is a kludge. Use a create rpc but with the IFMT bits of the
1562  * mode set to specify the file type and the size field for rdev.
1563  */
1564 static int
1565 nfs_mknodrpc(struct vnode *dvp, struct vnode **vpp, struct componentname *cnp,
1566     struct vattr *vap)
1567 {
1568 	struct nfsvattr nfsva, dnfsva;
1569 	struct vnode *newvp = NULL;
1570 	struct nfsnode *np = NULL, *dnp;
1571 	struct nfsfh *nfhp;
1572 	struct vattr vattr;
1573 	int error = 0, attrflag, dattrflag;
1574 	u_int32_t rdev;
1575 
1576 	if (vap->va_type == VCHR || vap->va_type == VBLK)
1577 		rdev = vap->va_rdev;
1578 	else if (vap->va_type == VFIFO || vap->va_type == VSOCK)
1579 		rdev = 0xffffffff;
1580 	else
1581 		return (EOPNOTSUPP);
1582 	if ((error = VOP_GETATTR(dvp, &vattr, cnp->cn_cred)))
1583 		return (error);
1584 	error = nfsrpc_mknod(dvp, cnp->cn_nameptr, cnp->cn_namelen, vap,
1585 	    rdev, vap->va_type, cnp->cn_cred, cnp->cn_thread, &dnfsva,
1586 	    &nfsva, &nfhp, &attrflag, &dattrflag, NULL);
1587 	if (!error) {
1588 		if (!nfhp)
1589 			(void) nfsrpc_lookup(dvp, cnp->cn_nameptr,
1590 			    cnp->cn_namelen, cnp->cn_cred, cnp->cn_thread,
1591 			    &dnfsva, &nfsva, &nfhp, &attrflag, &dattrflag,
1592 			    NULL);
1593 		if (nfhp)
1594 			error = nfscl_nget(dvp->v_mount, dvp, nfhp, cnp,
1595 			    cnp->cn_thread, &np, NULL, LK_EXCLUSIVE);
1596 	}
1597 	if (dattrflag)
1598 		(void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, NULL, 0, 1);
1599 	if (!error) {
1600 		newvp = NFSTOV(np);
1601 		if (attrflag != 0) {
1602 			error = nfscl_loadattrcache(&newvp, &nfsva, NULL, NULL,
1603 			    0, 1);
1604 			if (error != 0)
1605 				vput(newvp);
1606 		}
1607 	}
1608 	if (!error) {
1609 		*vpp = newvp;
1610 	} else if (NFS_ISV4(dvp)) {
1611 		error = nfscl_maperr(cnp->cn_thread, error, vap->va_uid,
1612 		    vap->va_gid);
1613 	}
1614 	dnp = VTONFS(dvp);
1615 	NFSLOCKNODE(dnp);
1616 	dnp->n_flag |= NMODIFIED;
1617 	if (!dattrflag) {
1618 		dnp->n_attrstamp = 0;
1619 		KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(dvp);
1620 	}
1621 	NFSUNLOCKNODE(dnp);
1622 	return (error);
1623 }
1624 
1625 /*
1626  * nfs mknod vop
1627  * just call nfs_mknodrpc() to do the work.
1628  */
1629 /* ARGSUSED */
1630 static int
1631 nfs_mknod(struct vop_mknod_args *ap)
1632 {
1633 	return (nfs_mknodrpc(ap->a_dvp, ap->a_vpp, ap->a_cnp, ap->a_vap));
1634 }
1635 
1636 static struct mtx nfs_cverf_mtx;
1637 MTX_SYSINIT(nfs_cverf_mtx, &nfs_cverf_mtx, "NFS create verifier mutex",
1638     MTX_DEF);
1639 
1640 static nfsquad_t
1641 nfs_get_cverf(void)
1642 {
1643 	static nfsquad_t cverf;
1644 	nfsquad_t ret;
1645 	static int cverf_initialized = 0;
1646 
1647 	mtx_lock(&nfs_cverf_mtx);
1648 	if (cverf_initialized == 0) {
1649 		cverf.lval[0] = arc4random();
1650 		cverf.lval[1] = arc4random();
1651 		cverf_initialized = 1;
1652 	} else
1653 		cverf.qval++;
1654 	ret = cverf;
1655 	mtx_unlock(&nfs_cverf_mtx);
1656 
1657 	return (ret);
1658 }
1659 
1660 /*
1661  * nfs file create call
1662  */
1663 static int
1664 nfs_create(struct vop_create_args *ap)
1665 {
1666 	struct vnode *dvp = ap->a_dvp;
1667 	struct vattr *vap = ap->a_vap;
1668 	struct componentname *cnp = ap->a_cnp;
1669 	struct nfsnode *np = NULL, *dnp;
1670 	struct vnode *newvp = NULL;
1671 	struct nfsmount *nmp;
1672 	struct nfsvattr dnfsva, nfsva;
1673 	struct nfsfh *nfhp;
1674 	nfsquad_t cverf;
1675 	int error = 0, attrflag, dattrflag, fmode = 0;
1676 	struct vattr vattr;
1677 
1678 	/*
1679 	 * Oops, not for me..
1680 	 */
1681 	if (vap->va_type == VSOCK)
1682 		return (nfs_mknodrpc(dvp, ap->a_vpp, cnp, vap));
1683 
1684 	if ((error = VOP_GETATTR(dvp, &vattr, cnp->cn_cred)))
1685 		return (error);
1686 	if (vap->va_vaflags & VA_EXCLUSIVE)
1687 		fmode |= O_EXCL;
1688 	dnp = VTONFS(dvp);
1689 	nmp = VFSTONFS(vnode_mount(dvp));
1690 again:
1691 	/* For NFSv4, wait until any remove is done. */
1692 	NFSLOCKNODE(dnp);
1693 	while (NFSHASNFSV4(nmp) && (dnp->n_flag & NREMOVEINPROG)) {
1694 		dnp->n_flag |= NREMOVEWANT;
1695 		(void) msleep((caddr_t)dnp, &dnp->n_mtx, PZERO, "nfscrt", 0);
1696 	}
1697 	NFSUNLOCKNODE(dnp);
1698 
1699 	cverf = nfs_get_cverf();
1700 	error = nfsrpc_create(dvp, cnp->cn_nameptr, cnp->cn_namelen,
1701 	    vap, cverf, fmode, cnp->cn_cred, cnp->cn_thread, &dnfsva, &nfsva,
1702 	    &nfhp, &attrflag, &dattrflag, NULL);
1703 	if (!error) {
1704 		if (nfhp == NULL)
1705 			(void) nfsrpc_lookup(dvp, cnp->cn_nameptr,
1706 			    cnp->cn_namelen, cnp->cn_cred, cnp->cn_thread,
1707 			    &dnfsva, &nfsva, &nfhp, &attrflag, &dattrflag,
1708 			    NULL);
1709 		if (nfhp != NULL)
1710 			error = nfscl_nget(dvp->v_mount, dvp, nfhp, cnp,
1711 			    cnp->cn_thread, &np, NULL, LK_EXCLUSIVE);
1712 	}
1713 	if (dattrflag)
1714 		(void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, NULL, 0, 1);
1715 	if (!error) {
1716 		newvp = NFSTOV(np);
1717 		if (attrflag == 0)
1718 			error = nfsrpc_getattr(newvp, cnp->cn_cred,
1719 			    cnp->cn_thread, &nfsva, NULL);
1720 		if (error == 0)
1721 			error = nfscl_loadattrcache(&newvp, &nfsva, NULL, NULL,
1722 			    0, 1);
1723 	}
1724 	if (error) {
1725 		if (newvp != NULL) {
1726 			vput(newvp);
1727 			newvp = NULL;
1728 		}
1729 		if (NFS_ISV34(dvp) && (fmode & O_EXCL) &&
1730 		    error == NFSERR_NOTSUPP) {
1731 			fmode &= ~O_EXCL;
1732 			goto again;
1733 		}
1734 	} else if (NFS_ISV34(dvp) && (fmode & O_EXCL)) {
1735 		if (nfscl_checksattr(vap, &nfsva)) {
1736 			error = nfsrpc_setattr(newvp, vap, NULL, cnp->cn_cred,
1737 			    cnp->cn_thread, &nfsva, &attrflag, NULL);
1738 			if (error && (vap->va_uid != (uid_t)VNOVAL ||
1739 			    vap->va_gid != (gid_t)VNOVAL)) {
1740 				/* try again without setting uid/gid */
1741 				vap->va_uid = (uid_t)VNOVAL;
1742 				vap->va_gid = (uid_t)VNOVAL;
1743 				error = nfsrpc_setattr(newvp, vap, NULL,
1744 				    cnp->cn_cred, cnp->cn_thread, &nfsva,
1745 				    &attrflag, NULL);
1746 			}
1747 			if (attrflag)
1748 				(void) nfscl_loadattrcache(&newvp, &nfsva, NULL,
1749 				    NULL, 0, 1);
1750 			if (error != 0)
1751 				vput(newvp);
1752 		}
1753 	}
1754 	if (!error) {
1755 		if ((cnp->cn_flags & MAKEENTRY) && attrflag)
1756 			cache_enter_time(dvp, newvp, cnp, &nfsva.na_ctime,
1757 			    NULL);
1758 		*ap->a_vpp = newvp;
1759 	} else if (NFS_ISV4(dvp)) {
1760 		error = nfscl_maperr(cnp->cn_thread, error, vap->va_uid,
1761 		    vap->va_gid);
1762 	}
1763 	NFSLOCKNODE(dnp);
1764 	dnp->n_flag |= NMODIFIED;
1765 	if (!dattrflag) {
1766 		dnp->n_attrstamp = 0;
1767 		KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(dvp);
1768 	}
1769 	NFSUNLOCKNODE(dnp);
1770 	return (error);
1771 }
1772 
1773 /*
1774  * nfs file remove call
1775  * To try and make nfs semantics closer to ufs semantics, a file that has
1776  * other processes using the vnode is renamed instead of removed and then
1777  * removed later on the last close.
1778  * - If v_usecount > 1
1779  *	  If a rename is not already in the works
1780  *	     call nfs_sillyrename() to set it up
1781  *     else
1782  *	  do the remove rpc
1783  */
1784 static int
1785 nfs_remove(struct vop_remove_args *ap)
1786 {
1787 	struct vnode *vp = ap->a_vp;
1788 	struct vnode *dvp = ap->a_dvp;
1789 	struct componentname *cnp = ap->a_cnp;
1790 	struct nfsnode *np = VTONFS(vp);
1791 	int error = 0;
1792 	struct vattr vattr;
1793 
1794 	KASSERT((cnp->cn_flags & HASBUF) != 0, ("nfs_remove: no name"));
1795 	KASSERT(vrefcnt(vp) > 0, ("nfs_remove: bad v_usecount"));
1796 	if (vp->v_type == VDIR)
1797 		error = EPERM;
1798 	else if (vrefcnt(vp) == 1 || (np->n_sillyrename &&
1799 	    VOP_GETATTR(vp, &vattr, cnp->cn_cred) == 0 &&
1800 	    vattr.va_nlink > 1)) {
1801 		/*
1802 		 * Purge the name cache so that the chance of a lookup for
1803 		 * the name succeeding while the remove is in progress is
1804 		 * minimized. Without node locking it can still happen, such
1805 		 * that an I/O op returns ESTALE, but since you get this if
1806 		 * another host removes the file..
1807 		 */
1808 		cache_purge(vp);
1809 		/*
1810 		 * throw away biocache buffers, mainly to avoid
1811 		 * unnecessary delayed writes later.
1812 		 */
1813 		error = ncl_vinvalbuf(vp, 0, cnp->cn_thread, 1);
1814 		if (error != EINTR && error != EIO)
1815 			/* Do the rpc */
1816 			error = nfs_removerpc(dvp, vp, cnp->cn_nameptr,
1817 			    cnp->cn_namelen, cnp->cn_cred, cnp->cn_thread);
1818 		/*
1819 		 * Kludge City: If the first reply to the remove rpc is lost..
1820 		 *   the reply to the retransmitted request will be ENOENT
1821 		 *   since the file was in fact removed
1822 		 *   Therefore, we cheat and return success.
1823 		 */
1824 		if (error == ENOENT)
1825 			error = 0;
1826 	} else if (!np->n_sillyrename)
1827 		error = nfs_sillyrename(dvp, vp, cnp);
1828 	NFSLOCKNODE(np);
1829 	np->n_attrstamp = 0;
1830 	NFSUNLOCKNODE(np);
1831 	KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(vp);
1832 	return (error);
1833 }
1834 
1835 /*
1836  * nfs file remove rpc called from nfs_inactive
1837  */
1838 int
1839 ncl_removeit(struct sillyrename *sp, struct vnode *vp)
1840 {
1841 	/*
1842 	 * Make sure that the directory vnode is still valid.
1843 	 * XXX we should lock sp->s_dvp here.
1844 	 */
1845 	if (sp->s_dvp->v_type == VBAD)
1846 		return (0);
1847 	return (nfs_removerpc(sp->s_dvp, vp, sp->s_name, sp->s_namlen,
1848 	    sp->s_cred, NULL));
1849 }
1850 
1851 /*
1852  * Nfs remove rpc, called from nfs_remove() and ncl_removeit().
1853  */
1854 static int
1855 nfs_removerpc(struct vnode *dvp, struct vnode *vp, char *name,
1856     int namelen, struct ucred *cred, struct thread *td)
1857 {
1858 	struct nfsvattr dnfsva;
1859 	struct nfsnode *dnp = VTONFS(dvp);
1860 	int error = 0, dattrflag;
1861 
1862 	NFSLOCKNODE(dnp);
1863 	dnp->n_flag |= NREMOVEINPROG;
1864 	NFSUNLOCKNODE(dnp);
1865 	error = nfsrpc_remove(dvp, name, namelen, vp, cred, td, &dnfsva,
1866 	    &dattrflag, NULL);
1867 	NFSLOCKNODE(dnp);
1868 	if ((dnp->n_flag & NREMOVEWANT)) {
1869 		dnp->n_flag &= ~(NREMOVEWANT | NREMOVEINPROG);
1870 		NFSUNLOCKNODE(dnp);
1871 		wakeup((caddr_t)dnp);
1872 	} else {
1873 		dnp->n_flag &= ~NREMOVEINPROG;
1874 		NFSUNLOCKNODE(dnp);
1875 	}
1876 	if (dattrflag)
1877 		(void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, NULL, 0, 1);
1878 	NFSLOCKNODE(dnp);
1879 	dnp->n_flag |= NMODIFIED;
1880 	if (!dattrflag) {
1881 		dnp->n_attrstamp = 0;
1882 		KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(dvp);
1883 	}
1884 	NFSUNLOCKNODE(dnp);
1885 	if (error && NFS_ISV4(dvp))
1886 		error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0);
1887 	return (error);
1888 }
1889 
1890 /*
1891  * nfs file rename call
1892  */
1893 static int
1894 nfs_rename(struct vop_rename_args *ap)
1895 {
1896 	struct vnode *fvp = ap->a_fvp;
1897 	struct vnode *tvp = ap->a_tvp;
1898 	struct vnode *fdvp = ap->a_fdvp;
1899 	struct vnode *tdvp = ap->a_tdvp;
1900 	struct componentname *tcnp = ap->a_tcnp;
1901 	struct componentname *fcnp = ap->a_fcnp;
1902 	struct nfsnode *fnp = VTONFS(ap->a_fvp);
1903 	struct nfsnode *tdnp = VTONFS(ap->a_tdvp);
1904 	struct nfsv4node *newv4 = NULL;
1905 	int error;
1906 
1907 	KASSERT((tcnp->cn_flags & HASBUF) != 0 &&
1908 	    (fcnp->cn_flags & HASBUF) != 0, ("nfs_rename: no name"));
1909 	/* Check for cross-device rename */
1910 	if ((fvp->v_mount != tdvp->v_mount) ||
1911 	    (tvp && (fvp->v_mount != tvp->v_mount))) {
1912 		error = EXDEV;
1913 		goto out;
1914 	}
1915 
1916 	if (fvp == tvp) {
1917 		printf("nfs_rename: fvp == tvp (can't happen)\n");
1918 		error = 0;
1919 		goto out;
1920 	}
1921 	if ((error = NFSVOPLOCK(fvp, LK_EXCLUSIVE)) != 0)
1922 		goto out;
1923 
1924 	/*
1925 	 * We have to flush B_DELWRI data prior to renaming
1926 	 * the file.  If we don't, the delayed-write buffers
1927 	 * can be flushed out later after the file has gone stale
1928 	 * under NFSV3.  NFSV2 does not have this problem because
1929 	 * ( as far as I can tell ) it flushes dirty buffers more
1930 	 * often.
1931 	 *
1932 	 * Skip the rename operation if the fsync fails, this can happen
1933 	 * due to the server's volume being full, when we pushed out data
1934 	 * that was written back to our cache earlier. Not checking for
1935 	 * this condition can result in potential (silent) data loss.
1936 	 */
1937 	error = VOP_FSYNC(fvp, MNT_WAIT, fcnp->cn_thread);
1938 	NFSVOPUNLOCK(fvp);
1939 	if (!error && tvp)
1940 		error = VOP_FSYNC(tvp, MNT_WAIT, tcnp->cn_thread);
1941 	if (error)
1942 		goto out;
1943 
1944 	/*
1945 	 * If the tvp exists and is in use, sillyrename it before doing the
1946 	 * rename of the new file over it.
1947 	 * XXX Can't sillyrename a directory.
1948 	 */
1949 	if (tvp && vrefcnt(tvp) > 1 && !VTONFS(tvp)->n_sillyrename &&
1950 		tvp->v_type != VDIR && !nfs_sillyrename(tdvp, tvp, tcnp)) {
1951 		vput(tvp);
1952 		tvp = NULL;
1953 	}
1954 
1955 	error = nfs_renamerpc(fdvp, fvp, fcnp->cn_nameptr, fcnp->cn_namelen,
1956 	    tdvp, tvp, tcnp->cn_nameptr, tcnp->cn_namelen, tcnp->cn_cred,
1957 	    tcnp->cn_thread);
1958 
1959 	if (error == 0 && NFS_ISV4(tdvp)) {
1960 		/*
1961 		 * For NFSv4, check to see if it is the same name and
1962 		 * replace the name, if it is different.
1963 		 */
1964 		newv4 = malloc(
1965 		    sizeof (struct nfsv4node) +
1966 		    tdnp->n_fhp->nfh_len + tcnp->cn_namelen - 1,
1967 		    M_NFSV4NODE, M_WAITOK);
1968 		NFSLOCKNODE(tdnp);
1969 		NFSLOCKNODE(fnp);
1970 		if (fnp->n_v4 != NULL && fvp->v_type == VREG &&
1971 		    (fnp->n_v4->n4_namelen != tcnp->cn_namelen ||
1972 		      NFSBCMP(tcnp->cn_nameptr, NFS4NODENAME(fnp->n_v4),
1973 		      tcnp->cn_namelen) ||
1974 		      tdnp->n_fhp->nfh_len != fnp->n_v4->n4_fhlen ||
1975 		      NFSBCMP(tdnp->n_fhp->nfh_fh, fnp->n_v4->n4_data,
1976 			tdnp->n_fhp->nfh_len))) {
1977 #ifdef notdef
1978 { char nnn[100]; int nnnl;
1979 nnnl = (tcnp->cn_namelen < 100) ? tcnp->cn_namelen : 99;
1980 bcopy(tcnp->cn_nameptr, nnn, nnnl);
1981 nnn[nnnl] = '\0';
1982 printf("ren replace=%s\n",nnn);
1983 }
1984 #endif
1985 			free(fnp->n_v4, M_NFSV4NODE);
1986 			fnp->n_v4 = newv4;
1987 			newv4 = NULL;
1988 			fnp->n_v4->n4_fhlen = tdnp->n_fhp->nfh_len;
1989 			fnp->n_v4->n4_namelen = tcnp->cn_namelen;
1990 			NFSBCOPY(tdnp->n_fhp->nfh_fh, fnp->n_v4->n4_data,
1991 			    tdnp->n_fhp->nfh_len);
1992 			NFSBCOPY(tcnp->cn_nameptr,
1993 			    NFS4NODENAME(fnp->n_v4), tcnp->cn_namelen);
1994 		}
1995 		NFSUNLOCKNODE(tdnp);
1996 		NFSUNLOCKNODE(fnp);
1997 		if (newv4 != NULL)
1998 			free(newv4, M_NFSV4NODE);
1999 	}
2000 
2001 	if (fvp->v_type == VDIR) {
2002 		if (tvp != NULL && tvp->v_type == VDIR)
2003 			cache_purge(tdvp);
2004 		cache_purge(fdvp);
2005 	}
2006 
2007 out:
2008 	if (tdvp == tvp)
2009 		vrele(tdvp);
2010 	else
2011 		vput(tdvp);
2012 	if (tvp)
2013 		vput(tvp);
2014 	vrele(fdvp);
2015 	vrele(fvp);
2016 	/*
2017 	 * Kludge: Map ENOENT => 0 assuming that it is a reply to a retry.
2018 	 */
2019 	if (error == ENOENT)
2020 		error = 0;
2021 	return (error);
2022 }
2023 
2024 /*
2025  * nfs file rename rpc called from nfs_remove() above
2026  */
2027 static int
2028 nfs_renameit(struct vnode *sdvp, struct vnode *svp, struct componentname *scnp,
2029     struct sillyrename *sp)
2030 {
2031 
2032 	return (nfs_renamerpc(sdvp, svp, scnp->cn_nameptr, scnp->cn_namelen,
2033 	    sdvp, NULL, sp->s_name, sp->s_namlen, scnp->cn_cred,
2034 	    scnp->cn_thread));
2035 }
2036 
2037 /*
2038  * Do an nfs rename rpc. Called from nfs_rename() and nfs_renameit().
2039  */
2040 static int
2041 nfs_renamerpc(struct vnode *fdvp, struct vnode *fvp, char *fnameptr,
2042     int fnamelen, struct vnode *tdvp, struct vnode *tvp, char *tnameptr,
2043     int tnamelen, struct ucred *cred, struct thread *td)
2044 {
2045 	struct nfsvattr fnfsva, tnfsva;
2046 	struct nfsnode *fdnp = VTONFS(fdvp);
2047 	struct nfsnode *tdnp = VTONFS(tdvp);
2048 	int error = 0, fattrflag, tattrflag;
2049 
2050 	error = nfsrpc_rename(fdvp, fvp, fnameptr, fnamelen, tdvp, tvp,
2051 	    tnameptr, tnamelen, cred, td, &fnfsva, &tnfsva, &fattrflag,
2052 	    &tattrflag, NULL, NULL);
2053 	NFSLOCKNODE(fdnp);
2054 	fdnp->n_flag |= NMODIFIED;
2055 	if (fattrflag != 0) {
2056 		NFSUNLOCKNODE(fdnp);
2057 		(void) nfscl_loadattrcache(&fdvp, &fnfsva, NULL, NULL, 0, 1);
2058 	} else {
2059 		fdnp->n_attrstamp = 0;
2060 		NFSUNLOCKNODE(fdnp);
2061 		KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(fdvp);
2062 	}
2063 	NFSLOCKNODE(tdnp);
2064 	tdnp->n_flag |= NMODIFIED;
2065 	if (tattrflag != 0) {
2066 		NFSUNLOCKNODE(tdnp);
2067 		(void) nfscl_loadattrcache(&tdvp, &tnfsva, NULL, NULL, 0, 1);
2068 	} else {
2069 		tdnp->n_attrstamp = 0;
2070 		NFSUNLOCKNODE(tdnp);
2071 		KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(tdvp);
2072 	}
2073 	if (error && NFS_ISV4(fdvp))
2074 		error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0);
2075 	return (error);
2076 }
2077 
2078 /*
2079  * nfs hard link create call
2080  */
2081 static int
2082 nfs_link(struct vop_link_args *ap)
2083 {
2084 	struct vnode *vp = ap->a_vp;
2085 	struct vnode *tdvp = ap->a_tdvp;
2086 	struct componentname *cnp = ap->a_cnp;
2087 	struct nfsnode *np, *tdnp;
2088 	struct nfsvattr nfsva, dnfsva;
2089 	int error = 0, attrflag, dattrflag;
2090 
2091 	/*
2092 	 * Push all writes to the server, so that the attribute cache
2093 	 * doesn't get "out of sync" with the server.
2094 	 * XXX There should be a better way!
2095 	 */
2096 	VOP_FSYNC(vp, MNT_WAIT, cnp->cn_thread);
2097 
2098 	error = nfsrpc_link(tdvp, vp, cnp->cn_nameptr, cnp->cn_namelen,
2099 	    cnp->cn_cred, cnp->cn_thread, &dnfsva, &nfsva, &attrflag,
2100 	    &dattrflag, NULL);
2101 	tdnp = VTONFS(tdvp);
2102 	NFSLOCKNODE(tdnp);
2103 	tdnp->n_flag |= NMODIFIED;
2104 	if (dattrflag != 0) {
2105 		NFSUNLOCKNODE(tdnp);
2106 		(void) nfscl_loadattrcache(&tdvp, &dnfsva, NULL, NULL, 0, 1);
2107 	} else {
2108 		tdnp->n_attrstamp = 0;
2109 		NFSUNLOCKNODE(tdnp);
2110 		KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(tdvp);
2111 	}
2112 	if (attrflag)
2113 		(void) nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0, 1);
2114 	else {
2115 		np = VTONFS(vp);
2116 		NFSLOCKNODE(np);
2117 		np->n_attrstamp = 0;
2118 		NFSUNLOCKNODE(np);
2119 		KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(vp);
2120 	}
2121 	/*
2122 	 * If negative lookup caching is enabled, I might as well
2123 	 * add an entry for this node. Not necessary for correctness,
2124 	 * but if negative caching is enabled, then the system
2125 	 * must care about lookup caching hit rate, so...
2126 	 */
2127 	if (VFSTONFS(vp->v_mount)->nm_negnametimeo != 0 &&
2128 	    (cnp->cn_flags & MAKEENTRY) && attrflag != 0 && error == 0) {
2129 		cache_enter_time(tdvp, vp, cnp, &nfsva.na_ctime, NULL);
2130 	}
2131 	if (error && NFS_ISV4(vp))
2132 		error = nfscl_maperr(cnp->cn_thread, error, (uid_t)0,
2133 		    (gid_t)0);
2134 	return (error);
2135 }
2136 
2137 /*
2138  * nfs symbolic link create call
2139  */
2140 static int
2141 nfs_symlink(struct vop_symlink_args *ap)
2142 {
2143 	struct vnode *dvp = ap->a_dvp;
2144 	struct vattr *vap = ap->a_vap;
2145 	struct componentname *cnp = ap->a_cnp;
2146 	struct nfsvattr nfsva, dnfsva;
2147 	struct nfsfh *nfhp;
2148 	struct nfsnode *np = NULL, *dnp;
2149 	struct vnode *newvp = NULL;
2150 	int error = 0, attrflag, dattrflag, ret;
2151 
2152 	vap->va_type = VLNK;
2153 	error = nfsrpc_symlink(dvp, cnp->cn_nameptr, cnp->cn_namelen,
2154 	    ap->a_target, vap, cnp->cn_cred, cnp->cn_thread, &dnfsva,
2155 	    &nfsva, &nfhp, &attrflag, &dattrflag, NULL);
2156 	if (nfhp) {
2157 		ret = nfscl_nget(dvp->v_mount, dvp, nfhp, cnp, cnp->cn_thread,
2158 		    &np, NULL, LK_EXCLUSIVE);
2159 		if (!ret)
2160 			newvp = NFSTOV(np);
2161 		else if (!error)
2162 			error = ret;
2163 	}
2164 	if (newvp != NULL) {
2165 		if (attrflag)
2166 			(void) nfscl_loadattrcache(&newvp, &nfsva, NULL, NULL,
2167 			    0, 1);
2168 	} else if (!error) {
2169 		/*
2170 		 * If we do not have an error and we could not extract the
2171 		 * newvp from the response due to the request being NFSv2, we
2172 		 * have to do a lookup in order to obtain a newvp to return.
2173 		 */
2174 		error = nfs_lookitup(dvp, cnp->cn_nameptr, cnp->cn_namelen,
2175 		    cnp->cn_cred, cnp->cn_thread, &np);
2176 		if (!error)
2177 			newvp = NFSTOV(np);
2178 	}
2179 	if (error) {
2180 		if (newvp)
2181 			vput(newvp);
2182 		if (NFS_ISV4(dvp))
2183 			error = nfscl_maperr(cnp->cn_thread, error,
2184 			    vap->va_uid, vap->va_gid);
2185 	} else {
2186 		*ap->a_vpp = newvp;
2187 	}
2188 
2189 	dnp = VTONFS(dvp);
2190 	NFSLOCKNODE(dnp);
2191 	dnp->n_flag |= NMODIFIED;
2192 	if (dattrflag != 0) {
2193 		NFSUNLOCKNODE(dnp);
2194 		(void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, NULL, 0, 1);
2195 	} else {
2196 		dnp->n_attrstamp = 0;
2197 		NFSUNLOCKNODE(dnp);
2198 		KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(dvp);
2199 	}
2200 	/*
2201 	 * If negative lookup caching is enabled, I might as well
2202 	 * add an entry for this node. Not necessary for correctness,
2203 	 * but if negative caching is enabled, then the system
2204 	 * must care about lookup caching hit rate, so...
2205 	 */
2206 	if (VFSTONFS(dvp->v_mount)->nm_negnametimeo != 0 &&
2207 	    (cnp->cn_flags & MAKEENTRY) && attrflag != 0 && error == 0) {
2208 		cache_enter_time(dvp, newvp, cnp, &nfsva.na_ctime, NULL);
2209 	}
2210 	return (error);
2211 }
2212 
2213 /*
2214  * nfs make dir call
2215  */
2216 static int
2217 nfs_mkdir(struct vop_mkdir_args *ap)
2218 {
2219 	struct vnode *dvp = ap->a_dvp;
2220 	struct vattr *vap = ap->a_vap;
2221 	struct componentname *cnp = ap->a_cnp;
2222 	struct nfsnode *np = NULL, *dnp;
2223 	struct vnode *newvp = NULL;
2224 	struct vattr vattr;
2225 	struct nfsfh *nfhp;
2226 	struct nfsvattr nfsva, dnfsva;
2227 	int error = 0, attrflag, dattrflag, ret;
2228 
2229 	if ((error = VOP_GETATTR(dvp, &vattr, cnp->cn_cred)) != 0)
2230 		return (error);
2231 	vap->va_type = VDIR;
2232 	error = nfsrpc_mkdir(dvp, cnp->cn_nameptr, cnp->cn_namelen,
2233 	    vap, cnp->cn_cred, cnp->cn_thread, &dnfsva, &nfsva, &nfhp,
2234 	    &attrflag, &dattrflag, NULL);
2235 	dnp = VTONFS(dvp);
2236 	NFSLOCKNODE(dnp);
2237 	dnp->n_flag |= NMODIFIED;
2238 	if (dattrflag != 0) {
2239 		NFSUNLOCKNODE(dnp);
2240 		(void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, NULL, 0, 1);
2241 	} else {
2242 		dnp->n_attrstamp = 0;
2243 		NFSUNLOCKNODE(dnp);
2244 		KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(dvp);
2245 	}
2246 	if (nfhp) {
2247 		ret = nfscl_nget(dvp->v_mount, dvp, nfhp, cnp, cnp->cn_thread,
2248 		    &np, NULL, LK_EXCLUSIVE);
2249 		if (!ret) {
2250 			newvp = NFSTOV(np);
2251 			if (attrflag)
2252 			   (void) nfscl_loadattrcache(&newvp, &nfsva, NULL,
2253 				NULL, 0, 1);
2254 		} else if (!error)
2255 			error = ret;
2256 	}
2257 	if (!error && newvp == NULL) {
2258 		error = nfs_lookitup(dvp, cnp->cn_nameptr, cnp->cn_namelen,
2259 		    cnp->cn_cred, cnp->cn_thread, &np);
2260 		if (!error) {
2261 			newvp = NFSTOV(np);
2262 			if (newvp->v_type != VDIR)
2263 				error = EEXIST;
2264 		}
2265 	}
2266 	if (error) {
2267 		if (newvp)
2268 			vput(newvp);
2269 		if (NFS_ISV4(dvp))
2270 			error = nfscl_maperr(cnp->cn_thread, error,
2271 			    vap->va_uid, vap->va_gid);
2272 	} else {
2273 		/*
2274 		 * If negative lookup caching is enabled, I might as well
2275 		 * add an entry for this node. Not necessary for correctness,
2276 		 * but if negative caching is enabled, then the system
2277 		 * must care about lookup caching hit rate, so...
2278 		 */
2279 		if (VFSTONFS(dvp->v_mount)->nm_negnametimeo != 0 &&
2280 		    (cnp->cn_flags & MAKEENTRY) &&
2281 		    attrflag != 0 && dattrflag != 0)
2282 			cache_enter_time(dvp, newvp, cnp, &nfsva.na_ctime,
2283 			    &dnfsva.na_ctime);
2284 		*ap->a_vpp = newvp;
2285 	}
2286 	return (error);
2287 }
2288 
2289 /*
2290  * nfs remove directory call
2291  */
2292 static int
2293 nfs_rmdir(struct vop_rmdir_args *ap)
2294 {
2295 	struct vnode *vp = ap->a_vp;
2296 	struct vnode *dvp = ap->a_dvp;
2297 	struct componentname *cnp = ap->a_cnp;
2298 	struct nfsnode *dnp;
2299 	struct nfsvattr dnfsva;
2300 	int error, dattrflag;
2301 
2302 	if (dvp == vp)
2303 		return (EINVAL);
2304 	error = nfsrpc_rmdir(dvp, cnp->cn_nameptr, cnp->cn_namelen,
2305 	    cnp->cn_cred, cnp->cn_thread, &dnfsva, &dattrflag, NULL);
2306 	dnp = VTONFS(dvp);
2307 	NFSLOCKNODE(dnp);
2308 	dnp->n_flag |= NMODIFIED;
2309 	if (dattrflag != 0) {
2310 		NFSUNLOCKNODE(dnp);
2311 		(void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, NULL, 0, 1);
2312 	} else {
2313 		dnp->n_attrstamp = 0;
2314 		NFSUNLOCKNODE(dnp);
2315 		KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(dvp);
2316 	}
2317 
2318 	cache_purge(dvp);
2319 	cache_purge(vp);
2320 	if (error && NFS_ISV4(dvp))
2321 		error = nfscl_maperr(cnp->cn_thread, error, (uid_t)0,
2322 		    (gid_t)0);
2323 	/*
2324 	 * Kludge: Map ENOENT => 0 assuming that you have a reply to a retry.
2325 	 */
2326 	if (error == ENOENT)
2327 		error = 0;
2328 	return (error);
2329 }
2330 
2331 /*
2332  * nfs readdir call
2333  */
2334 static int
2335 nfs_readdir(struct vop_readdir_args *ap)
2336 {
2337 	struct vnode *vp = ap->a_vp;
2338 	struct nfsnode *np = VTONFS(vp);
2339 	struct uio *uio = ap->a_uio;
2340 	ssize_t tresid, left;
2341 	int error = 0;
2342 	struct vattr vattr;
2343 
2344 	if (ap->a_eofflag != NULL)
2345 		*ap->a_eofflag = 0;
2346 	if (vp->v_type != VDIR)
2347 		return(EPERM);
2348 
2349 	/*
2350 	 * First, check for hit on the EOF offset cache
2351 	 */
2352 	if (np->n_direofoffset > 0 && uio->uio_offset >= np->n_direofoffset &&
2353 	    (np->n_flag & NMODIFIED) == 0) {
2354 		if (VOP_GETATTR(vp, &vattr, ap->a_cred) == 0) {
2355 			NFSLOCKNODE(np);
2356 			if ((NFS_ISV4(vp) && np->n_change == vattr.va_filerev) ||
2357 			    !NFS_TIMESPEC_COMPARE(&np->n_mtime, &vattr.va_mtime)) {
2358 				NFSUNLOCKNODE(np);
2359 				NFSINCRGLOBAL(nfsstatsv1.direofcache_hits);
2360 				if (ap->a_eofflag != NULL)
2361 					*ap->a_eofflag = 1;
2362 				return (0);
2363 			} else
2364 				NFSUNLOCKNODE(np);
2365 		}
2366 	}
2367 
2368 	/*
2369 	 * NFS always guarantees that directory entries don't straddle
2370 	 * DIRBLKSIZ boundaries.  As such, we need to limit the size
2371 	 * to an exact multiple of DIRBLKSIZ, to avoid copying a partial
2372 	 * directory entry.
2373 	 */
2374 	left = uio->uio_resid % DIRBLKSIZ;
2375 	if (left == uio->uio_resid)
2376 		return (EINVAL);
2377 	uio->uio_resid -= left;
2378 
2379 	/*
2380 	 * Call ncl_bioread() to do the real work.
2381 	 */
2382 	tresid = uio->uio_resid;
2383 	error = ncl_bioread(vp, uio, 0, ap->a_cred);
2384 
2385 	if (!error && uio->uio_resid == tresid) {
2386 		NFSINCRGLOBAL(nfsstatsv1.direofcache_misses);
2387 		if (ap->a_eofflag != NULL)
2388 			*ap->a_eofflag = 1;
2389 	}
2390 
2391 	/* Add the partial DIRBLKSIZ (left) back in. */
2392 	uio->uio_resid += left;
2393 	return (error);
2394 }
2395 
2396 /*
2397  * Readdir rpc call.
2398  * Called from below the buffer cache by ncl_doio().
2399  */
2400 int
2401 ncl_readdirrpc(struct vnode *vp, struct uio *uiop, struct ucred *cred,
2402     struct thread *td)
2403 {
2404 	struct nfsvattr nfsva;
2405 	nfsuint64 *cookiep, cookie;
2406 	struct nfsnode *dnp = VTONFS(vp);
2407 	struct nfsmount *nmp = VFSTONFS(vp->v_mount);
2408 	int error = 0, eof, attrflag;
2409 
2410 	KASSERT(uiop->uio_iovcnt == 1 &&
2411 	    (uiop->uio_offset & (DIRBLKSIZ - 1)) == 0 &&
2412 	    (uiop->uio_resid & (DIRBLKSIZ - 1)) == 0,
2413 	    ("nfs readdirrpc bad uio"));
2414 
2415 	/*
2416 	 * If there is no cookie, assume directory was stale.
2417 	 */
2418 	ncl_dircookie_lock(dnp);
2419 	cookiep = ncl_getcookie(dnp, uiop->uio_offset, 0);
2420 	if (cookiep) {
2421 		cookie = *cookiep;
2422 		ncl_dircookie_unlock(dnp);
2423 	} else {
2424 		ncl_dircookie_unlock(dnp);
2425 		return (NFSERR_BAD_COOKIE);
2426 	}
2427 
2428 	if (NFSHASNFSV3(nmp) && !NFSHASGOTFSINFO(nmp))
2429 		(void)ncl_fsinfo(nmp, vp, cred, td);
2430 
2431 	error = nfsrpc_readdir(vp, uiop, &cookie, cred, td, &nfsva,
2432 	    &attrflag, &eof, NULL);
2433 	if (attrflag)
2434 		(void) nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0, 1);
2435 
2436 	if (!error) {
2437 		/*
2438 		 * We are now either at the end of the directory or have filled
2439 		 * the block.
2440 		 */
2441 		if (eof)
2442 			dnp->n_direofoffset = uiop->uio_offset;
2443 		else {
2444 			if (uiop->uio_resid > 0)
2445 				printf("EEK! readdirrpc resid > 0\n");
2446 			ncl_dircookie_lock(dnp);
2447 			cookiep = ncl_getcookie(dnp, uiop->uio_offset, 1);
2448 			*cookiep = cookie;
2449 			ncl_dircookie_unlock(dnp);
2450 		}
2451 	} else if (NFS_ISV4(vp)) {
2452 		error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0);
2453 	}
2454 	return (error);
2455 }
2456 
2457 /*
2458  * NFS V3 readdir plus RPC. Used in place of ncl_readdirrpc().
2459  */
2460 int
2461 ncl_readdirplusrpc(struct vnode *vp, struct uio *uiop, struct ucred *cred,
2462     struct thread *td)
2463 {
2464 	struct nfsvattr nfsva;
2465 	nfsuint64 *cookiep, cookie;
2466 	struct nfsnode *dnp = VTONFS(vp);
2467 	struct nfsmount *nmp = VFSTONFS(vp->v_mount);
2468 	int error = 0, attrflag, eof;
2469 
2470 	KASSERT(uiop->uio_iovcnt == 1 &&
2471 	    (uiop->uio_offset & (DIRBLKSIZ - 1)) == 0 &&
2472 	    (uiop->uio_resid & (DIRBLKSIZ - 1)) == 0,
2473 	    ("nfs readdirplusrpc bad uio"));
2474 
2475 	/*
2476 	 * If there is no cookie, assume directory was stale.
2477 	 */
2478 	ncl_dircookie_lock(dnp);
2479 	cookiep = ncl_getcookie(dnp, uiop->uio_offset, 0);
2480 	if (cookiep) {
2481 		cookie = *cookiep;
2482 		ncl_dircookie_unlock(dnp);
2483 	} else {
2484 		ncl_dircookie_unlock(dnp);
2485 		return (NFSERR_BAD_COOKIE);
2486 	}
2487 
2488 	if (NFSHASNFSV3(nmp) && !NFSHASGOTFSINFO(nmp))
2489 		(void)ncl_fsinfo(nmp, vp, cred, td);
2490 	error = nfsrpc_readdirplus(vp, uiop, &cookie, cred, td, &nfsva,
2491 	    &attrflag, &eof, NULL);
2492 	if (attrflag)
2493 		(void) nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0, 1);
2494 
2495 	if (!error) {
2496 		/*
2497 		 * We are now either at end of the directory or have filled the
2498 		 * the block.
2499 		 */
2500 		if (eof)
2501 			dnp->n_direofoffset = uiop->uio_offset;
2502 		else {
2503 			if (uiop->uio_resid > 0)
2504 				printf("EEK! readdirplusrpc resid > 0\n");
2505 			ncl_dircookie_lock(dnp);
2506 			cookiep = ncl_getcookie(dnp, uiop->uio_offset, 1);
2507 			*cookiep = cookie;
2508 			ncl_dircookie_unlock(dnp);
2509 		}
2510 	} else if (NFS_ISV4(vp)) {
2511 		error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0);
2512 	}
2513 	return (error);
2514 }
2515 
2516 /*
2517  * Silly rename. To make the NFS filesystem that is stateless look a little
2518  * more like the "ufs" a remove of an active vnode is translated to a rename
2519  * to a funny looking filename that is removed by nfs_inactive on the
2520  * nfsnode. There is the potential for another process on a different client
2521  * to create the same funny name between the nfs_lookitup() fails and the
2522  * nfs_rename() completes, but...
2523  */
2524 static int
2525 nfs_sillyrename(struct vnode *dvp, struct vnode *vp, struct componentname *cnp)
2526 {
2527 	struct sillyrename *sp;
2528 	struct nfsnode *np;
2529 	int error;
2530 	short pid;
2531 	unsigned int lticks;
2532 
2533 	cache_purge(dvp);
2534 	np = VTONFS(vp);
2535 	KASSERT(vp->v_type != VDIR, ("nfs: sillyrename dir"));
2536 	sp = malloc(sizeof (struct sillyrename),
2537 	    M_NEWNFSREQ, M_WAITOK);
2538 	sp->s_cred = crhold(cnp->cn_cred);
2539 	sp->s_dvp = dvp;
2540 	VREF(dvp);
2541 
2542 	/*
2543 	 * Fudge together a funny name.
2544 	 * Changing the format of the funny name to accommodate more
2545 	 * sillynames per directory.
2546 	 * The name is now changed to .nfs.<ticks>.<pid>.4, where ticks is
2547 	 * CPU ticks since boot.
2548 	 */
2549 	pid = cnp->cn_thread->td_proc->p_pid;
2550 	lticks = (unsigned int)ticks;
2551 	for ( ; ; ) {
2552 		sp->s_namlen = sprintf(sp->s_name,
2553 				       ".nfs.%08x.%04x4.4", lticks,
2554 				       pid);
2555 		if (nfs_lookitup(dvp, sp->s_name, sp->s_namlen, sp->s_cred,
2556 				 cnp->cn_thread, NULL))
2557 			break;
2558 		lticks++;
2559 	}
2560 	error = nfs_renameit(dvp, vp, cnp, sp);
2561 	if (error)
2562 		goto bad;
2563 	error = nfs_lookitup(dvp, sp->s_name, sp->s_namlen, sp->s_cred,
2564 		cnp->cn_thread, &np);
2565 	np->n_sillyrename = sp;
2566 	return (0);
2567 bad:
2568 	vrele(sp->s_dvp);
2569 	crfree(sp->s_cred);
2570 	free(sp, M_NEWNFSREQ);
2571 	return (error);
2572 }
2573 
2574 /*
2575  * Look up a file name and optionally either update the file handle or
2576  * allocate an nfsnode, depending on the value of npp.
2577  * npp == NULL	--> just do the lookup
2578  * *npp == NULL --> allocate a new nfsnode and make sure attributes are
2579  *			handled too
2580  * *npp != NULL --> update the file handle in the vnode
2581  */
2582 static int
2583 nfs_lookitup(struct vnode *dvp, char *name, int len, struct ucred *cred,
2584     struct thread *td, struct nfsnode **npp)
2585 {
2586 	struct vnode *newvp = NULL, *vp;
2587 	struct nfsnode *np, *dnp = VTONFS(dvp);
2588 	struct nfsfh *nfhp, *onfhp;
2589 	struct nfsvattr nfsva, dnfsva;
2590 	struct componentname cn;
2591 	int error = 0, attrflag, dattrflag;
2592 	u_int hash;
2593 
2594 	error = nfsrpc_lookup(dvp, name, len, cred, td, &dnfsva, &nfsva,
2595 	    &nfhp, &attrflag, &dattrflag, NULL);
2596 	if (dattrflag)
2597 		(void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, NULL, 0, 1);
2598 	if (npp && !error) {
2599 		if (*npp != NULL) {
2600 		    np = *npp;
2601 		    vp = NFSTOV(np);
2602 		    /*
2603 		     * For NFSv4, check to see if it is the same name and
2604 		     * replace the name, if it is different.
2605 		     */
2606 		    if (np->n_v4 != NULL && nfsva.na_type == VREG &&
2607 			(np->n_v4->n4_namelen != len ||
2608 			 NFSBCMP(name, NFS4NODENAME(np->n_v4), len) ||
2609 			 dnp->n_fhp->nfh_len != np->n_v4->n4_fhlen ||
2610 			 NFSBCMP(dnp->n_fhp->nfh_fh, np->n_v4->n4_data,
2611 			 dnp->n_fhp->nfh_len))) {
2612 #ifdef notdef
2613 { char nnn[100]; int nnnl;
2614 nnnl = (len < 100) ? len : 99;
2615 bcopy(name, nnn, nnnl);
2616 nnn[nnnl] = '\0';
2617 printf("replace=%s\n",nnn);
2618 }
2619 #endif
2620 			    free(np->n_v4, M_NFSV4NODE);
2621 			    np->n_v4 = malloc(
2622 				sizeof (struct nfsv4node) +
2623 				dnp->n_fhp->nfh_len + len - 1,
2624 				M_NFSV4NODE, M_WAITOK);
2625 			    np->n_v4->n4_fhlen = dnp->n_fhp->nfh_len;
2626 			    np->n_v4->n4_namelen = len;
2627 			    NFSBCOPY(dnp->n_fhp->nfh_fh, np->n_v4->n4_data,
2628 				dnp->n_fhp->nfh_len);
2629 			    NFSBCOPY(name, NFS4NODENAME(np->n_v4), len);
2630 		    }
2631 		    hash = fnv_32_buf(nfhp->nfh_fh, nfhp->nfh_len,
2632 			FNV1_32_INIT);
2633 		    onfhp = np->n_fhp;
2634 		    /*
2635 		     * Rehash node for new file handle.
2636 		     */
2637 		    vfs_hash_rehash(vp, hash);
2638 		    np->n_fhp = nfhp;
2639 		    if (onfhp != NULL)
2640 			free(onfhp, M_NFSFH);
2641 		    newvp = NFSTOV(np);
2642 		} else if (NFS_CMPFH(dnp, nfhp->nfh_fh, nfhp->nfh_len)) {
2643 		    free(nfhp, M_NFSFH);
2644 		    VREF(dvp);
2645 		    newvp = dvp;
2646 		} else {
2647 		    cn.cn_nameptr = name;
2648 		    cn.cn_namelen = len;
2649 		    error = nfscl_nget(dvp->v_mount, dvp, nfhp, &cn, td,
2650 			&np, NULL, LK_EXCLUSIVE);
2651 		    if (error)
2652 			return (error);
2653 		    newvp = NFSTOV(np);
2654 		}
2655 		if (!attrflag && *npp == NULL) {
2656 			if (newvp == dvp)
2657 				vrele(newvp);
2658 			else
2659 				vput(newvp);
2660 			return (ENOENT);
2661 		}
2662 		if (attrflag)
2663 			(void) nfscl_loadattrcache(&newvp, &nfsva, NULL, NULL,
2664 			    0, 1);
2665 	}
2666 	if (npp && *npp == NULL) {
2667 		if (error) {
2668 			if (newvp) {
2669 				if (newvp == dvp)
2670 					vrele(newvp);
2671 				else
2672 					vput(newvp);
2673 			}
2674 		} else
2675 			*npp = np;
2676 	}
2677 	if (error && NFS_ISV4(dvp))
2678 		error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0);
2679 	return (error);
2680 }
2681 
2682 /*
2683  * Nfs Version 3 and 4 commit rpc
2684  */
2685 int
2686 ncl_commit(struct vnode *vp, u_quad_t offset, int cnt, struct ucred *cred,
2687    struct thread *td)
2688 {
2689 	struct nfsvattr nfsva;
2690 	struct nfsmount *nmp = VFSTONFS(vp->v_mount);
2691 	struct nfsnode *np;
2692 	struct uio uio;
2693 	int error, attrflag;
2694 
2695 	np = VTONFS(vp);
2696 	error = EIO;
2697 	attrflag = 0;
2698 	if (NFSHASPNFS(nmp) && (np->n_flag & NDSCOMMIT) != 0) {
2699 		uio.uio_offset = offset;
2700 		uio.uio_resid = cnt;
2701 		error = nfscl_doiods(vp, &uio, NULL, NULL,
2702 		    NFSV4OPEN_ACCESSWRITE, 1, cred, td);
2703 		if (error != 0) {
2704 			NFSLOCKNODE(np);
2705 			np->n_flag &= ~NDSCOMMIT;
2706 			NFSUNLOCKNODE(np);
2707 		}
2708 	}
2709 	if (error != 0) {
2710 		mtx_lock(&nmp->nm_mtx);
2711 		if ((nmp->nm_state & NFSSTA_HASWRITEVERF) == 0) {
2712 			mtx_unlock(&nmp->nm_mtx);
2713 			return (0);
2714 		}
2715 		mtx_unlock(&nmp->nm_mtx);
2716 		error = nfsrpc_commit(vp, offset, cnt, cred, td, &nfsva,
2717 		    &attrflag, NULL);
2718 	}
2719 	if (attrflag != 0)
2720 		(void) nfscl_loadattrcache(&vp, &nfsva, NULL, NULL,
2721 		    0, 1);
2722 	if (error != 0 && NFS_ISV4(vp))
2723 		error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0);
2724 	return (error);
2725 }
2726 
2727 /*
2728  * Strategy routine.
2729  * For async requests when nfsiod(s) are running, queue the request by
2730  * calling ncl_asyncio(), otherwise just all ncl_doio() to do the
2731  * request.
2732  */
2733 static int
2734 nfs_strategy(struct vop_strategy_args *ap)
2735 {
2736 	struct buf *bp;
2737 	struct vnode *vp;
2738 	struct ucred *cr;
2739 
2740 	bp = ap->a_bp;
2741 	vp = ap->a_vp;
2742 	KASSERT(bp->b_vp == vp, ("missing b_getvp"));
2743 	KASSERT(!(bp->b_flags & B_DONE),
2744 	    ("nfs_strategy: buffer %p unexpectedly marked B_DONE", bp));
2745 
2746 	if (vp->v_type == VREG && bp->b_blkno == bp->b_lblkno)
2747 		bp->b_blkno = bp->b_lblkno * (vp->v_bufobj.bo_bsize /
2748 		    DEV_BSIZE);
2749 	if (bp->b_iocmd == BIO_READ)
2750 		cr = bp->b_rcred;
2751 	else
2752 		cr = bp->b_wcred;
2753 
2754 	/*
2755 	 * If the op is asynchronous and an i/o daemon is waiting
2756 	 * queue the request, wake it up and wait for completion
2757 	 * otherwise just do it ourselves.
2758 	 */
2759 	if ((bp->b_flags & B_ASYNC) == 0 ||
2760 	    ncl_asyncio(VFSTONFS(vp->v_mount), bp, NOCRED, curthread))
2761 		(void) ncl_doio(vp, bp, cr, curthread, 1);
2762 	return (0);
2763 }
2764 
2765 /*
2766  * fsync vnode op. Just call ncl_flush() with commit == 1.
2767  */
2768 /* ARGSUSED */
2769 static int
2770 nfs_fsync(struct vop_fsync_args *ap)
2771 {
2772 
2773 	if (ap->a_vp->v_type != VREG) {
2774 		/*
2775 		 * For NFS, metadata is changed synchronously on the server,
2776 		 * so there is nothing to flush. Also, ncl_flush() clears
2777 		 * the NMODIFIED flag and that shouldn't be done here for
2778 		 * directories.
2779 		 */
2780 		return (0);
2781 	}
2782 	return (ncl_flush(ap->a_vp, ap->a_waitfor, ap->a_td, 1, 0));
2783 }
2784 
2785 /*
2786  * Flush all the blocks associated with a vnode.
2787  * 	Walk through the buffer pool and push any dirty pages
2788  *	associated with the vnode.
2789  * If the called_from_renewthread argument is TRUE, it has been called
2790  * from the NFSv4 renew thread and, as such, cannot block indefinitely
2791  * waiting for a buffer write to complete.
2792  */
2793 int
2794 ncl_flush(struct vnode *vp, int waitfor, struct thread *td,
2795     int commit, int called_from_renewthread)
2796 {
2797 	struct nfsnode *np = VTONFS(vp);
2798 	struct buf *bp;
2799 	int i;
2800 	struct buf *nbp;
2801 	struct nfsmount *nmp = VFSTONFS(vp->v_mount);
2802 	int error = 0, slptimeo = 0, slpflag = 0, retv, bvecpos;
2803 	int passone = 1, trycnt = 0;
2804 	u_quad_t off, endoff, toff;
2805 	struct ucred* wcred = NULL;
2806 	struct buf **bvec = NULL;
2807 	struct bufobj *bo;
2808 #ifndef NFS_COMMITBVECSIZ
2809 #define	NFS_COMMITBVECSIZ	20
2810 #endif
2811 	struct buf *bvec_on_stack[NFS_COMMITBVECSIZ];
2812 	u_int bvecsize = 0, bveccount;
2813 
2814 	if (called_from_renewthread != 0)
2815 		slptimeo = hz;
2816 	if (nmp->nm_flag & NFSMNT_INT)
2817 		slpflag = PCATCH;
2818 	if (!commit)
2819 		passone = 0;
2820 	bo = &vp->v_bufobj;
2821 	/*
2822 	 * A b_flags == (B_DELWRI | B_NEEDCOMMIT) block has been written to the
2823 	 * server, but has not been committed to stable storage on the server
2824 	 * yet. On the first pass, the byte range is worked out and the commit
2825 	 * rpc is done. On the second pass, ncl_writebp() is called to do the
2826 	 * job.
2827 	 */
2828 again:
2829 	off = (u_quad_t)-1;
2830 	endoff = 0;
2831 	bvecpos = 0;
2832 	if (NFS_ISV34(vp) && commit) {
2833 		if (bvec != NULL && bvec != bvec_on_stack)
2834 			free(bvec, M_TEMP);
2835 		/*
2836 		 * Count up how many buffers waiting for a commit.
2837 		 */
2838 		bveccount = 0;
2839 		BO_LOCK(bo);
2840 		TAILQ_FOREACH_SAFE(bp, &bo->bo_dirty.bv_hd, b_bobufs, nbp) {
2841 			if (!BUF_ISLOCKED(bp) &&
2842 			    (bp->b_flags & (B_DELWRI | B_NEEDCOMMIT))
2843 				== (B_DELWRI | B_NEEDCOMMIT))
2844 				bveccount++;
2845 		}
2846 		/*
2847 		 * Allocate space to remember the list of bufs to commit.  It is
2848 		 * important to use M_NOWAIT here to avoid a race with nfs_write.
2849 		 * If we can't get memory (for whatever reason), we will end up
2850 		 * committing the buffers one-by-one in the loop below.
2851 		 */
2852 		if (bveccount > NFS_COMMITBVECSIZ) {
2853 			/*
2854 			 * Release the vnode interlock to avoid a lock
2855 			 * order reversal.
2856 			 */
2857 			BO_UNLOCK(bo);
2858 			bvec = (struct buf **)
2859 				malloc(bveccount * sizeof(struct buf *),
2860 				       M_TEMP, M_NOWAIT);
2861 			BO_LOCK(bo);
2862 			if (bvec == NULL) {
2863 				bvec = bvec_on_stack;
2864 				bvecsize = NFS_COMMITBVECSIZ;
2865 			} else
2866 				bvecsize = bveccount;
2867 		} else {
2868 			bvec = bvec_on_stack;
2869 			bvecsize = NFS_COMMITBVECSIZ;
2870 		}
2871 		TAILQ_FOREACH_SAFE(bp, &bo->bo_dirty.bv_hd, b_bobufs, nbp) {
2872 			if (bvecpos >= bvecsize)
2873 				break;
2874 			if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT, NULL)) {
2875 				nbp = TAILQ_NEXT(bp, b_bobufs);
2876 				continue;
2877 			}
2878 			if ((bp->b_flags & (B_DELWRI | B_NEEDCOMMIT)) !=
2879 			    (B_DELWRI | B_NEEDCOMMIT)) {
2880 				BUF_UNLOCK(bp);
2881 				nbp = TAILQ_NEXT(bp, b_bobufs);
2882 				continue;
2883 			}
2884 			BO_UNLOCK(bo);
2885 			bremfree(bp);
2886 			/*
2887 			 * Work out if all buffers are using the same cred
2888 			 * so we can deal with them all with one commit.
2889 			 *
2890 			 * NOTE: we are not clearing B_DONE here, so we have
2891 			 * to do it later on in this routine if we intend to
2892 			 * initiate I/O on the bp.
2893 			 *
2894 			 * Note: to avoid loopback deadlocks, we do not
2895 			 * assign b_runningbufspace.
2896 			 */
2897 			if (wcred == NULL)
2898 				wcred = bp->b_wcred;
2899 			else if (wcred != bp->b_wcred)
2900 				wcred = NOCRED;
2901 			vfs_busy_pages(bp, 1);
2902 
2903 			BO_LOCK(bo);
2904 			/*
2905 			 * bp is protected by being locked, but nbp is not
2906 			 * and vfs_busy_pages() may sleep.  We have to
2907 			 * recalculate nbp.
2908 			 */
2909 			nbp = TAILQ_NEXT(bp, b_bobufs);
2910 
2911 			/*
2912 			 * A list of these buffers is kept so that the
2913 			 * second loop knows which buffers have actually
2914 			 * been committed. This is necessary, since there
2915 			 * may be a race between the commit rpc and new
2916 			 * uncommitted writes on the file.
2917 			 */
2918 			bvec[bvecpos++] = bp;
2919 			toff = ((u_quad_t)bp->b_blkno) * DEV_BSIZE +
2920 				bp->b_dirtyoff;
2921 			if (toff < off)
2922 				off = toff;
2923 			toff += (u_quad_t)(bp->b_dirtyend - bp->b_dirtyoff);
2924 			if (toff > endoff)
2925 				endoff = toff;
2926 		}
2927 		BO_UNLOCK(bo);
2928 	}
2929 	if (bvecpos > 0) {
2930 		/*
2931 		 * Commit data on the server, as required.
2932 		 * If all bufs are using the same wcred, then use that with
2933 		 * one call for all of them, otherwise commit each one
2934 		 * separately.
2935 		 */
2936 		if (wcred != NOCRED)
2937 			retv = ncl_commit(vp, off, (int)(endoff - off),
2938 					  wcred, td);
2939 		else {
2940 			retv = 0;
2941 			for (i = 0; i < bvecpos; i++) {
2942 				off_t off, size;
2943 				bp = bvec[i];
2944 				off = ((u_quad_t)bp->b_blkno) * DEV_BSIZE +
2945 					bp->b_dirtyoff;
2946 				size = (u_quad_t)(bp->b_dirtyend
2947 						  - bp->b_dirtyoff);
2948 				retv = ncl_commit(vp, off, (int)size,
2949 						  bp->b_wcred, td);
2950 				if (retv) break;
2951 			}
2952 		}
2953 
2954 		if (retv == NFSERR_STALEWRITEVERF)
2955 			ncl_clearcommit(vp->v_mount);
2956 
2957 		/*
2958 		 * Now, either mark the blocks I/O done or mark the
2959 		 * blocks dirty, depending on whether the commit
2960 		 * succeeded.
2961 		 */
2962 		for (i = 0; i < bvecpos; i++) {
2963 			bp = bvec[i];
2964 			bp->b_flags &= ~(B_NEEDCOMMIT | B_CLUSTEROK);
2965 			if (retv) {
2966 				/*
2967 				 * Error, leave B_DELWRI intact
2968 				 */
2969 				vfs_unbusy_pages(bp);
2970 				brelse(bp);
2971 			} else {
2972 				/*
2973 				 * Success, remove B_DELWRI ( bundirty() ).
2974 				 *
2975 				 * b_dirtyoff/b_dirtyend seem to be NFS
2976 				 * specific.  We should probably move that
2977 				 * into bundirty(). XXX
2978 				 */
2979 				bufobj_wref(bo);
2980 				bp->b_flags |= B_ASYNC;
2981 				bundirty(bp);
2982 				bp->b_flags &= ~B_DONE;
2983 				bp->b_ioflags &= ~BIO_ERROR;
2984 				bp->b_dirtyoff = bp->b_dirtyend = 0;
2985 				bufdone(bp);
2986 			}
2987 		}
2988 	}
2989 
2990 	/*
2991 	 * Start/do any write(s) that are required.
2992 	 */
2993 loop:
2994 	BO_LOCK(bo);
2995 	TAILQ_FOREACH_SAFE(bp, &bo->bo_dirty.bv_hd, b_bobufs, nbp) {
2996 		if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT, NULL)) {
2997 			if (waitfor != MNT_WAIT || passone)
2998 				continue;
2999 
3000 			error = BUF_TIMELOCK(bp,
3001 			    LK_EXCLUSIVE | LK_SLEEPFAIL | LK_INTERLOCK,
3002 			    BO_LOCKPTR(bo), "nfsfsync", slpflag, slptimeo);
3003 			if (error == 0) {
3004 				BUF_UNLOCK(bp);
3005 				goto loop;
3006 			}
3007 			if (error == ENOLCK) {
3008 				error = 0;
3009 				goto loop;
3010 			}
3011 			if (called_from_renewthread != 0) {
3012 				/*
3013 				 * Return EIO so the flush will be retried
3014 				 * later.
3015 				 */
3016 				error = EIO;
3017 				goto done;
3018 			}
3019 			if (newnfs_sigintr(nmp, td)) {
3020 				error = EINTR;
3021 				goto done;
3022 			}
3023 			if (slpflag == PCATCH) {
3024 				slpflag = 0;
3025 				slptimeo = 2 * hz;
3026 			}
3027 			goto loop;
3028 		}
3029 		if ((bp->b_flags & B_DELWRI) == 0)
3030 			panic("nfs_fsync: not dirty");
3031 		if ((passone || !commit) && (bp->b_flags & B_NEEDCOMMIT)) {
3032 			BUF_UNLOCK(bp);
3033 			continue;
3034 		}
3035 		BO_UNLOCK(bo);
3036 		bremfree(bp);
3037 		bp->b_flags |= B_ASYNC;
3038 		bwrite(bp);
3039 		if (newnfs_sigintr(nmp, td)) {
3040 			error = EINTR;
3041 			goto done;
3042 		}
3043 		goto loop;
3044 	}
3045 	if (passone) {
3046 		passone = 0;
3047 		BO_UNLOCK(bo);
3048 		goto again;
3049 	}
3050 	if (waitfor == MNT_WAIT) {
3051 		while (bo->bo_numoutput) {
3052 			error = bufobj_wwait(bo, slpflag, slptimeo);
3053 			if (error) {
3054 			    BO_UNLOCK(bo);
3055 			    if (called_from_renewthread != 0) {
3056 				/*
3057 				 * Return EIO so that the flush will be
3058 				 * retried later.
3059 				 */
3060 				error = EIO;
3061 				goto done;
3062 			    }
3063 			    error = newnfs_sigintr(nmp, td);
3064 			    if (error)
3065 				goto done;
3066 			    if (slpflag == PCATCH) {
3067 				slpflag = 0;
3068 				slptimeo = 2 * hz;
3069 			    }
3070 			    BO_LOCK(bo);
3071 			}
3072 		}
3073 		if (bo->bo_dirty.bv_cnt != 0 && commit) {
3074 			BO_UNLOCK(bo);
3075 			goto loop;
3076 		}
3077 		/*
3078 		 * Wait for all the async IO requests to drain
3079 		 */
3080 		BO_UNLOCK(bo);
3081 		NFSLOCKNODE(np);
3082 		while (np->n_directio_asyncwr > 0) {
3083 			np->n_flag |= NFSYNCWAIT;
3084 			error = newnfs_msleep(td, &np->n_directio_asyncwr,
3085 			    &np->n_mtx, slpflag | (PRIBIO + 1),
3086 			    "nfsfsync", 0);
3087 			if (error) {
3088 				if (newnfs_sigintr(nmp, td)) {
3089 					NFSUNLOCKNODE(np);
3090 					error = EINTR;
3091 					goto done;
3092 				}
3093 			}
3094 		}
3095 		NFSUNLOCKNODE(np);
3096 	} else
3097 		BO_UNLOCK(bo);
3098 	if (NFSHASPNFS(nmp)) {
3099 		nfscl_layoutcommit(vp, td);
3100 		/*
3101 		 * Invalidate the attribute cache, since writes to a DS
3102 		 * won't update the size attribute.
3103 		 */
3104 		NFSLOCKNODE(np);
3105 		np->n_attrstamp = 0;
3106 	} else
3107 		NFSLOCKNODE(np);
3108 	if (np->n_flag & NWRITEERR) {
3109 		error = np->n_error;
3110 		np->n_flag &= ~NWRITEERR;
3111 	}
3112   	if (commit && bo->bo_dirty.bv_cnt == 0 &&
3113 	    bo->bo_numoutput == 0 && np->n_directio_asyncwr == 0)
3114   		np->n_flag &= ~NMODIFIED;
3115 	NFSUNLOCKNODE(np);
3116 done:
3117 	if (bvec != NULL && bvec != bvec_on_stack)
3118 		free(bvec, M_TEMP);
3119 	if (error == 0 && commit != 0 && waitfor == MNT_WAIT &&
3120 	    (bo->bo_dirty.bv_cnt != 0 || bo->bo_numoutput != 0 ||
3121 	    np->n_directio_asyncwr != 0)) {
3122 		if (trycnt++ < 5) {
3123 			/* try, try again... */
3124 			passone = 1;
3125 			wcred = NULL;
3126 			bvec = NULL;
3127 			bvecsize = 0;
3128 			goto again;
3129 		}
3130 		vn_printf(vp, "ncl_flush failed");
3131 		error = called_from_renewthread != 0 ? EIO : EBUSY;
3132 	}
3133 	return (error);
3134 }
3135 
3136 /*
3137  * NFS advisory byte-level locks.
3138  */
3139 static int
3140 nfs_advlock(struct vop_advlock_args *ap)
3141 {
3142 	struct vnode *vp = ap->a_vp;
3143 	struct ucred *cred;
3144 	struct nfsnode *np = VTONFS(ap->a_vp);
3145 	struct proc *p = (struct proc *)ap->a_id;
3146 	struct thread *td = curthread;	/* XXX */
3147 	struct vattr va;
3148 	int ret, error;
3149 	u_quad_t size;
3150 
3151 	error = NFSVOPLOCK(vp, LK_SHARED);
3152 	if (error != 0)
3153 		return (EBADF);
3154 	if (NFS_ISV4(vp) && (ap->a_flags & (F_POSIX | F_FLOCK)) != 0) {
3155 		if (vp->v_type != VREG) {
3156 			error = EINVAL;
3157 			goto out;
3158 		}
3159 		if ((ap->a_flags & F_POSIX) != 0)
3160 			cred = p->p_ucred;
3161 		else
3162 			cred = td->td_ucred;
3163 		NFSVOPLOCK(vp, LK_UPGRADE | LK_RETRY);
3164 		if (VN_IS_DOOMED(vp)) {
3165 			error = EBADF;
3166 			goto out;
3167 		}
3168 
3169 		/*
3170 		 * If this is unlocking a write locked region, flush and
3171 		 * commit them before unlocking. This is required by
3172 		 * RFC3530 Sec. 9.3.2.
3173 		 */
3174 		if (ap->a_op == F_UNLCK &&
3175 		    nfscl_checkwritelocked(vp, ap->a_fl, cred, td, ap->a_id,
3176 		    ap->a_flags))
3177 			(void) ncl_flush(vp, MNT_WAIT, td, 1, 0);
3178 
3179 		/*
3180 		 * Loop around doing the lock op, while a blocking lock
3181 		 * must wait for the lock op to succeed.
3182 		 */
3183 		do {
3184 			ret = nfsrpc_advlock(vp, np->n_size, ap->a_op,
3185 			    ap->a_fl, 0, cred, td, ap->a_id, ap->a_flags);
3186 			if (ret == NFSERR_DENIED && (ap->a_flags & F_WAIT) &&
3187 			    ap->a_op == F_SETLK) {
3188 				NFSVOPUNLOCK(vp);
3189 				error = nfs_catnap(PZERO | PCATCH, ret,
3190 				    "ncladvl");
3191 				if (error)
3192 					return (EINTR);
3193 				NFSVOPLOCK(vp, LK_EXCLUSIVE | LK_RETRY);
3194 				if (VN_IS_DOOMED(vp)) {
3195 					error = EBADF;
3196 					goto out;
3197 				}
3198 			}
3199 		} while (ret == NFSERR_DENIED && (ap->a_flags & F_WAIT) &&
3200 		     ap->a_op == F_SETLK);
3201 		if (ret == NFSERR_DENIED) {
3202 			error = EAGAIN;
3203 			goto out;
3204 		} else if (ret == EINVAL || ret == EBADF || ret == EINTR) {
3205 			error = ret;
3206 			goto out;
3207 		} else if (ret != 0) {
3208 			error = EACCES;
3209 			goto out;
3210 		}
3211 
3212 		/*
3213 		 * Now, if we just got a lock, invalidate data in the buffer
3214 		 * cache, as required, so that the coherency conforms with
3215 		 * RFC3530 Sec. 9.3.2.
3216 		 */
3217 		if (ap->a_op == F_SETLK) {
3218 			if ((np->n_flag & NMODIFIED) == 0) {
3219 				np->n_attrstamp = 0;
3220 				KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(vp);
3221 				ret = VOP_GETATTR(vp, &va, cred);
3222 			}
3223 			if ((np->n_flag & NMODIFIED) || ret ||
3224 			    np->n_change != va.va_filerev) {
3225 				(void) ncl_vinvalbuf(vp, V_SAVE, td, 1);
3226 				np->n_attrstamp = 0;
3227 				KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(vp);
3228 				ret = VOP_GETATTR(vp, &va, cred);
3229 				if (!ret) {
3230 					np->n_mtime = va.va_mtime;
3231 					np->n_change = va.va_filerev;
3232 				}
3233 			}
3234 			/* Mark that a file lock has been acquired. */
3235 			NFSLOCKNODE(np);
3236 			np->n_flag |= NHASBEENLOCKED;
3237 			NFSUNLOCKNODE(np);
3238 		}
3239 	} else if (!NFS_ISV4(vp)) {
3240 		if ((VFSTONFS(vp->v_mount)->nm_flag & NFSMNT_NOLOCKD) != 0) {
3241 			size = VTONFS(vp)->n_size;
3242 			NFSVOPUNLOCK(vp);
3243 			error = lf_advlock(ap, &(vp->v_lockf), size);
3244 		} else {
3245 			if (nfs_advlock_p != NULL)
3246 				error = nfs_advlock_p(ap);
3247 			else {
3248 				NFSVOPUNLOCK(vp);
3249 				error = ENOLCK;
3250 			}
3251 		}
3252 		if (error == 0 && ap->a_op == F_SETLK) {
3253 			error = NFSVOPLOCK(vp, LK_SHARED);
3254 			if (error == 0) {
3255 				/* Mark that a file lock has been acquired. */
3256 				NFSLOCKNODE(np);
3257 				np->n_flag |= NHASBEENLOCKED;
3258 				NFSUNLOCKNODE(np);
3259 				NFSVOPUNLOCK(vp);
3260 			}
3261 		}
3262 		return (error);
3263 	} else
3264 		error = EOPNOTSUPP;
3265 out:
3266 	NFSVOPUNLOCK(vp);
3267 	return (error);
3268 }
3269 
3270 /*
3271  * NFS advisory byte-level locks.
3272  */
3273 static int
3274 nfs_advlockasync(struct vop_advlockasync_args *ap)
3275 {
3276 	struct vnode *vp = ap->a_vp;
3277 	u_quad_t size;
3278 	int error;
3279 
3280 	if (NFS_ISV4(vp))
3281 		return (EOPNOTSUPP);
3282 	error = NFSVOPLOCK(vp, LK_SHARED);
3283 	if (error)
3284 		return (error);
3285 	if ((VFSTONFS(vp->v_mount)->nm_flag & NFSMNT_NOLOCKD) != 0) {
3286 		size = VTONFS(vp)->n_size;
3287 		NFSVOPUNLOCK(vp);
3288 		error = lf_advlockasync(ap, &(vp->v_lockf), size);
3289 	} else {
3290 		NFSVOPUNLOCK(vp);
3291 		error = EOPNOTSUPP;
3292 	}
3293 	return (error);
3294 }
3295 
3296 /*
3297  * Print out the contents of an nfsnode.
3298  */
3299 static int
3300 nfs_print(struct vop_print_args *ap)
3301 {
3302 	struct vnode *vp = ap->a_vp;
3303 	struct nfsnode *np = VTONFS(vp);
3304 
3305 	printf("\tfileid %jd fsid 0x%jx", (uintmax_t)np->n_vattr.na_fileid,
3306 	    (uintmax_t)np->n_vattr.na_fsid);
3307 	if (vp->v_type == VFIFO)
3308 		fifo_printinfo(vp);
3309 	printf("\n");
3310 	return (0);
3311 }
3312 
3313 /*
3314  * This is the "real" nfs::bwrite(struct buf*).
3315  * We set B_CACHE if this is a VMIO buffer.
3316  */
3317 int
3318 ncl_writebp(struct buf *bp, int force __unused, struct thread *td)
3319 {
3320 	int oldflags, rtval;
3321 
3322 	if (bp->b_flags & B_INVAL) {
3323 		brelse(bp);
3324 		return (0);
3325 	}
3326 
3327 	oldflags = bp->b_flags;
3328 	bp->b_flags |= B_CACHE;
3329 
3330 	/*
3331 	 * Undirty the bp.  We will redirty it later if the I/O fails.
3332 	 */
3333 	bundirty(bp);
3334 	bp->b_flags &= ~B_DONE;
3335 	bp->b_ioflags &= ~BIO_ERROR;
3336 	bp->b_iocmd = BIO_WRITE;
3337 
3338 	bufobj_wref(bp->b_bufobj);
3339 	curthread->td_ru.ru_oublock++;
3340 
3341 	/*
3342 	 * Note: to avoid loopback deadlocks, we do not
3343 	 * assign b_runningbufspace.
3344 	 */
3345 	vfs_busy_pages(bp, 1);
3346 
3347 	BUF_KERNPROC(bp);
3348 	bp->b_iooffset = dbtob(bp->b_blkno);
3349 	bstrategy(bp);
3350 
3351 	if ((oldflags & B_ASYNC) != 0)
3352 		return (0);
3353 
3354 	rtval = bufwait(bp);
3355 	if (oldflags & B_DELWRI)
3356 		reassignbuf(bp);
3357 	brelse(bp);
3358 	return (rtval);
3359 }
3360 
3361 /*
3362  * nfs special file access vnode op.
3363  * Essentially just get vattr and then imitate iaccess() since the device is
3364  * local to the client.
3365  */
3366 static int
3367 nfsspec_access(struct vop_access_args *ap)
3368 {
3369 	struct vattr *vap;
3370 	struct ucred *cred = ap->a_cred;
3371 	struct vnode *vp = ap->a_vp;
3372 	accmode_t accmode = ap->a_accmode;
3373 	struct vattr vattr;
3374 	int error;
3375 
3376 	/*
3377 	 * Disallow write attempts on filesystems mounted read-only;
3378 	 * unless the file is a socket, fifo, or a block or character
3379 	 * device resident on the filesystem.
3380 	 */
3381 	if ((accmode & VWRITE) && (vp->v_mount->mnt_flag & MNT_RDONLY)) {
3382 		switch (vp->v_type) {
3383 		case VREG:
3384 		case VDIR:
3385 		case VLNK:
3386 			return (EROFS);
3387 		default:
3388 			break;
3389 		}
3390 	}
3391 	vap = &vattr;
3392 	error = VOP_GETATTR(vp, vap, cred);
3393 	if (error)
3394 		goto out;
3395 	error  = vaccess(vp->v_type, vap->va_mode, vap->va_uid, vap->va_gid,
3396 	    accmode, cred, NULL);
3397 out:
3398 	return error;
3399 }
3400 
3401 /*
3402  * Read wrapper for fifos.
3403  */
3404 static int
3405 nfsfifo_read(struct vop_read_args *ap)
3406 {
3407 	struct nfsnode *np = VTONFS(ap->a_vp);
3408 	int error;
3409 
3410 	/*
3411 	 * Set access flag.
3412 	 */
3413 	NFSLOCKNODE(np);
3414 	np->n_flag |= NACC;
3415 	vfs_timestamp(&np->n_atim);
3416 	NFSUNLOCKNODE(np);
3417 	error = fifo_specops.vop_read(ap);
3418 	return error;
3419 }
3420 
3421 /*
3422  * Write wrapper for fifos.
3423  */
3424 static int
3425 nfsfifo_write(struct vop_write_args *ap)
3426 {
3427 	struct nfsnode *np = VTONFS(ap->a_vp);
3428 
3429 	/*
3430 	 * Set update flag.
3431 	 */
3432 	NFSLOCKNODE(np);
3433 	np->n_flag |= NUPD;
3434 	vfs_timestamp(&np->n_mtim);
3435 	NFSUNLOCKNODE(np);
3436 	return(fifo_specops.vop_write(ap));
3437 }
3438 
3439 /*
3440  * Close wrapper for fifos.
3441  *
3442  * Update the times on the nfsnode then do fifo close.
3443  */
3444 static int
3445 nfsfifo_close(struct vop_close_args *ap)
3446 {
3447 	struct vnode *vp = ap->a_vp;
3448 	struct nfsnode *np = VTONFS(vp);
3449 	struct vattr vattr;
3450 	struct timespec ts;
3451 
3452 	NFSLOCKNODE(np);
3453 	if (np->n_flag & (NACC | NUPD)) {
3454 		vfs_timestamp(&ts);
3455 		if (np->n_flag & NACC)
3456 			np->n_atim = ts;
3457 		if (np->n_flag & NUPD)
3458 			np->n_mtim = ts;
3459 		np->n_flag |= NCHG;
3460 		if (vrefcnt(vp) == 1 &&
3461 		    (vp->v_mount->mnt_flag & MNT_RDONLY) == 0) {
3462 			VATTR_NULL(&vattr);
3463 			if (np->n_flag & NACC)
3464 				vattr.va_atime = np->n_atim;
3465 			if (np->n_flag & NUPD)
3466 				vattr.va_mtime = np->n_mtim;
3467 			NFSUNLOCKNODE(np);
3468 			(void)VOP_SETATTR(vp, &vattr, ap->a_cred);
3469 			goto out;
3470 		}
3471 	}
3472 	NFSUNLOCKNODE(np);
3473 out:
3474 	return (fifo_specops.vop_close(ap));
3475 }
3476 
3477 /*
3478  * Just call ncl_writebp() with the force argument set to 1.
3479  *
3480  * NOTE: B_DONE may or may not be set in a_bp on call.
3481  */
3482 static int
3483 nfs_bwrite(struct buf *bp)
3484 {
3485 
3486 	return (ncl_writebp(bp, 1, curthread));
3487 }
3488 
3489 struct buf_ops buf_ops_newnfs = {
3490 	.bop_name	=	"buf_ops_nfs",
3491 	.bop_write	=	nfs_bwrite,
3492 	.bop_strategy	=	bufstrategy,
3493 	.bop_sync	=	bufsync,
3494 	.bop_bdflush	=	bufbdflush,
3495 };
3496 
3497 static int
3498 nfs_getacl(struct vop_getacl_args *ap)
3499 {
3500 	int error;
3501 
3502 	if (ap->a_type != ACL_TYPE_NFS4)
3503 		return (EOPNOTSUPP);
3504 	error = nfsrpc_getacl(ap->a_vp, ap->a_cred, ap->a_td, ap->a_aclp,
3505 	    NULL);
3506 	if (error > NFSERR_STALE) {
3507 		(void) nfscl_maperr(ap->a_td, error, (uid_t)0, (gid_t)0);
3508 		error = EPERM;
3509 	}
3510 	return (error);
3511 }
3512 
3513 static int
3514 nfs_setacl(struct vop_setacl_args *ap)
3515 {
3516 	int error;
3517 
3518 	if (ap->a_type != ACL_TYPE_NFS4)
3519 		return (EOPNOTSUPP);
3520 	error = nfsrpc_setacl(ap->a_vp, ap->a_cred, ap->a_td, ap->a_aclp,
3521 	    NULL);
3522 	if (error > NFSERR_STALE) {
3523 		(void) nfscl_maperr(ap->a_td, error, (uid_t)0, (gid_t)0);
3524 		error = EPERM;
3525 	}
3526 	return (error);
3527 }
3528 
3529 /*
3530  * VOP_ADVISE for NFS.
3531  * Just return 0 for any errors, since it is just a hint.
3532  */
3533 static int
3534 nfs_advise(struct vop_advise_args *ap)
3535 {
3536 	struct thread *td = curthread;
3537 	struct nfsmount *nmp;
3538 	uint64_t len;
3539 	int error;
3540 
3541 	/*
3542 	 * First do vop_stdadvise() to handle the buffer cache.
3543 	 */
3544 	error = vop_stdadvise(ap);
3545 	if (error != 0)
3546 		return (error);
3547 	if (ap->a_start < 0 || ap->a_end < 0)
3548 		return (0);
3549 	if (ap->a_end == OFF_MAX)
3550 		len = 0;
3551 	else if (ap->a_end < ap->a_start)
3552 		return (0);
3553 	else
3554 		len = ap->a_end - ap->a_start + 1;
3555 	nmp = VFSTONFS(ap->a_vp->v_mount);
3556 	mtx_lock(&nmp->nm_mtx);
3557 	if (!NFSHASNFSV4(nmp) || nmp->nm_minorvers < NFSV42_MINORVERSION ||
3558 	    (NFSHASPNFS(nmp) && (nmp->nm_privflag & NFSMNTP_IOADVISETHRUMDS) ==
3559 	    0) || (nmp->nm_privflag & NFSMNTP_NOADVISE) != 0) {
3560 		mtx_unlock(&nmp->nm_mtx);
3561 		return (0);
3562 	}
3563 	mtx_unlock(&nmp->nm_mtx);
3564 	error = nfsrpc_advise(ap->a_vp, ap->a_start, len, ap->a_advice,
3565 	    td->td_ucred, td);
3566 	if (error == NFSERR_NOTSUPP) {
3567 		mtx_lock(&nmp->nm_mtx);
3568 		nmp->nm_privflag |= NFSMNTP_NOADVISE;
3569 		mtx_unlock(&nmp->nm_mtx);
3570 	}
3571 	return (0);
3572 }
3573 
3574 /*
3575  * nfs allocate call
3576  */
3577 static int
3578 nfs_allocate(struct vop_allocate_args *ap)
3579 {
3580 	struct vnode *vp = ap->a_vp;
3581 	struct thread *td = curthread;
3582 	struct nfsvattr nfsva;
3583 	struct nfsmount *nmp;
3584 	int attrflag, error, ret;
3585 
3586 	attrflag = 0;
3587 	nmp = VFSTONFS(vp->v_mount);
3588 	mtx_lock(&nmp->nm_mtx);
3589 	if (NFSHASNFSV4(nmp) && nmp->nm_minorvers >= NFSV42_MINORVERSION &&
3590 	    (nmp->nm_privflag & NFSMNTP_NOALLOCATE) == 0) {
3591 		mtx_unlock(&nmp->nm_mtx);
3592 		/*
3593 		 * Flush first to ensure that the allocate adds to the
3594 		 * file's allocation on the server.
3595 		 */
3596 		error = ncl_flush(vp, MNT_WAIT, td, 1, 0);
3597 		if (error == 0)
3598 			error = nfsrpc_allocate(vp, *ap->a_offset, *ap->a_len,
3599 			    &nfsva, &attrflag, td->td_ucred, td, NULL);
3600 		if (error == 0) {
3601 			*ap->a_offset += *ap->a_len;
3602 			*ap->a_len = 0;
3603 		} else if (error == NFSERR_NOTSUPP) {
3604 			mtx_lock(&nmp->nm_mtx);
3605 			nmp->nm_privflag |= NFSMNTP_NOALLOCATE;
3606 			mtx_unlock(&nmp->nm_mtx);
3607 		}
3608 	} else {
3609 		mtx_unlock(&nmp->nm_mtx);
3610 		error = EIO;
3611 	}
3612 	/*
3613 	 * If the NFS server cannot perform the Allocate operation, just call
3614 	 * vop_stdallocate() to perform it.
3615 	 */
3616 	if (error != 0)
3617 		error = vop_stdallocate(ap);
3618 	if (attrflag != 0) {
3619 		ret = nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0, 1);
3620 		if (error == 0 && ret != 0)
3621 			error = ret;
3622 	}
3623 	if (error != 0)
3624 		error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0);
3625 	return (error);
3626 }
3627 
3628 /*
3629  * nfs copy_file_range call
3630  */
3631 static int
3632 nfs_copy_file_range(struct vop_copy_file_range_args *ap)
3633 {
3634 	struct vnode *invp = ap->a_invp;
3635 	struct vnode *outvp = ap->a_outvp;
3636 	struct mount *mp;
3637 	struct nfsvattr innfsva, outnfsva;
3638 	struct vattr *vap;
3639 	struct uio io;
3640 	struct nfsmount *nmp;
3641 	size_t len, len2, copiedlen;
3642 	int error, inattrflag, outattrflag, ret, ret2;
3643 	off_t inoff, outoff;
3644 	bool consecutive, must_commit, tryoutcred;
3645 
3646 	ret = ret2 = 0;
3647 	nmp = VFSTONFS(invp->v_mount);
3648 	mtx_lock(&nmp->nm_mtx);
3649 	/* NFSv4.2 Copy is not permitted for infile == outfile. */
3650 	if (!NFSHASNFSV4(nmp) || nmp->nm_minorvers < NFSV42_MINORVERSION ||
3651 	    (nmp->nm_privflag & NFSMNTP_NOCOPY) != 0 || invp == outvp) {
3652 		mtx_unlock(&nmp->nm_mtx);
3653 		error = vn_generic_copy_file_range(ap->a_invp, ap->a_inoffp,
3654 		    ap->a_outvp, ap->a_outoffp, ap->a_lenp, ap->a_flags,
3655 		    ap->a_incred, ap->a_outcred, ap->a_fsizetd);
3656 		return (error);
3657 	}
3658 	mtx_unlock(&nmp->nm_mtx);
3659 
3660 	/* Lock both vnodes, avoiding risk of deadlock. */
3661 	do {
3662 		mp = NULL;
3663 		error = vn_start_write(outvp, &mp, V_WAIT);
3664 		if (error == 0) {
3665 			error = vn_lock(outvp, LK_EXCLUSIVE);
3666 			if (error == 0) {
3667 				error = vn_lock(invp, LK_SHARED | LK_NOWAIT);
3668 				if (error == 0)
3669 					break;
3670 				VOP_UNLOCK(outvp);
3671 				if (mp != NULL)
3672 					vn_finished_write(mp);
3673 				mp = NULL;
3674 				error = vn_lock(invp, LK_SHARED);
3675 				if (error == 0)
3676 					VOP_UNLOCK(invp);
3677 			}
3678 		}
3679 		if (mp != NULL)
3680 			vn_finished_write(mp);
3681 	} while (error == 0);
3682 	if (error != 0)
3683 		return (error);
3684 
3685 	/*
3686 	 * Do the vn_rlimit_fsize() check.  Should this be above the VOP layer?
3687 	 */
3688 	io.uio_offset = *ap->a_outoffp;
3689 	io.uio_resid = *ap->a_lenp;
3690 	error = vn_rlimit_fsize(outvp, &io, ap->a_fsizetd);
3691 
3692 	/*
3693 	 * Flush the input file so that the data is up to date before
3694 	 * the copy.  Flush writes for the output file so that they
3695 	 * do not overwrite the data copied to the output file by the Copy.
3696 	 * Set the commit argument for both flushes so that the data is on
3697 	 * stable storage before the Copy RPC.  This is done in case the
3698 	 * server reboots during the Copy and needs to be redone.
3699 	 */
3700 	if (error == 0)
3701 		error = ncl_flush(invp, MNT_WAIT, curthread, 1, 0);
3702 	if (error == 0)
3703 		error = ncl_flush(outvp, MNT_WAIT, curthread, 1, 0);
3704 
3705 	/* Do the actual NFSv4.2 RPC. */
3706 	len = *ap->a_lenp;
3707 	mtx_lock(&nmp->nm_mtx);
3708 	if ((nmp->nm_privflag & NFSMNTP_NOCONSECUTIVE) == 0)
3709 		consecutive = true;
3710 	else
3711 		consecutive = false;
3712 	mtx_unlock(&nmp->nm_mtx);
3713 	inoff = *ap->a_inoffp;
3714 	outoff = *ap->a_outoffp;
3715 	tryoutcred = true;
3716 	must_commit = false;
3717 	if (error == 0) {
3718 		vap = &VTONFS(invp)->n_vattr.na_vattr;
3719 		error = VOP_GETATTR(invp, vap, ap->a_incred);
3720 		if (error == 0) {
3721 			/*
3722 			 * Clip "len" at va_size so that RFC compliant servers
3723 			 * will not reply NFSERR_INVAL.
3724 			 * Setting "len == 0" for the RPC would be preferred,
3725 			 * but some Linux servers do not support that.
3726 			 */
3727 			if (inoff >= vap->va_size)
3728 				*ap->a_lenp = len = 0;
3729 			else if (inoff + len > vap->va_size)
3730 				*ap->a_lenp = len = vap->va_size - inoff;
3731 		} else
3732 			error = 0;
3733 	}
3734 	copiedlen = 0;
3735 	while (len > 0 && error == 0) {
3736 		inattrflag = outattrflag = 0;
3737 		len2 = len;
3738 		if (tryoutcred)
3739 			error = nfsrpc_copy_file_range(invp, ap->a_inoffp,
3740 			    outvp, ap->a_outoffp, &len2, ap->a_flags,
3741 			    &inattrflag, &innfsva, &outattrflag, &outnfsva,
3742 			    ap->a_outcred, consecutive, &must_commit);
3743 		else
3744 			error = nfsrpc_copy_file_range(invp, ap->a_inoffp,
3745 			    outvp, ap->a_outoffp, &len2, ap->a_flags,
3746 			    &inattrflag, &innfsva, &outattrflag, &outnfsva,
3747 			    ap->a_incred, consecutive, &must_commit);
3748 		if (inattrflag != 0)
3749 			ret = nfscl_loadattrcache(&invp, &innfsva, NULL, NULL,
3750 			    0, 1);
3751 		if (outattrflag != 0)
3752 			ret2 = nfscl_loadattrcache(&outvp, &outnfsva, NULL,
3753 			    NULL, 1, 1);
3754 		if (error == 0) {
3755 			if (consecutive == false) {
3756 				if (len2 == len) {
3757 					mtx_lock(&nmp->nm_mtx);
3758 					nmp->nm_privflag |=
3759 					    NFSMNTP_NOCONSECUTIVE;
3760 					mtx_unlock(&nmp->nm_mtx);
3761 				} else
3762 					error = NFSERR_OFFLOADNOREQS;
3763 			}
3764 			/*
3765 			 * If the Copy returns a length == 0, it hit the
3766 			 * EOF on the input file.
3767 			 */
3768 			if (len2 == 0) {
3769 				*ap->a_lenp = copiedlen;
3770 				len = 0;
3771 			} else {
3772 				len -= len2;
3773 				copiedlen += len2;
3774 			}
3775 			if (len == 0 && must_commit && error == 0)
3776 				error = ncl_commit(outvp, outoff, *ap->a_lenp,
3777 				    ap->a_outcred, curthread);
3778 			if (error == 0 && ret != 0)
3779 				error = ret;
3780 			if (error == 0 && ret2 != 0)
3781 				error = ret2;
3782 		} else if (error == NFSERR_OFFLOADNOREQS && consecutive) {
3783 			/*
3784 			 * Try consecutive == false, which is ok only if all
3785 			 * bytes are copied.
3786 			 */
3787 			consecutive = false;
3788 			error = 0;
3789 		} else if (error == NFSERR_ACCES && tryoutcred) {
3790 			/* Try again with incred. */
3791 			tryoutcred = false;
3792 			error = 0;
3793 		}
3794 		if (error == NFSERR_STALEWRITEVERF) {
3795 			/*
3796 			 * Server rebooted, so do it all again.
3797 			 */
3798 			*ap->a_inoffp = inoff;
3799 			*ap->a_outoffp = outoff;
3800 			len = *ap->a_lenp;
3801 			must_commit = false;
3802 			error = 0;
3803 		}
3804 	}
3805 	VOP_UNLOCK(invp);
3806 	VOP_UNLOCK(outvp);
3807 	if (mp != NULL)
3808 		vn_finished_write(mp);
3809 	if (error == NFSERR_NOTSUPP || error == NFSERR_OFFLOADNOREQS ||
3810 	    error == NFSERR_ACCES) {
3811 		/*
3812 		 * Unlike the NFSv4.2 Copy, vn_generic_copy_file_range() can
3813 		 * use a_incred for the read and a_outcred for the write, so
3814 		 * try this for NFSERR_ACCES failures for the Copy.
3815 		 * For NFSERR_NOTSUPP and NFSERR_OFFLOADNOREQS, the Copy can
3816 		 * never succeed, so disable it.
3817 		 */
3818 		if (error != NFSERR_ACCES) {
3819 			/* Can never do Copy on this mount. */
3820 			mtx_lock(&nmp->nm_mtx);
3821 			nmp->nm_privflag |= NFSMNTP_NOCOPY;
3822 			mtx_unlock(&nmp->nm_mtx);
3823 		}
3824 		*ap->a_inoffp = inoff;
3825 		*ap->a_outoffp = outoff;
3826 		error = vn_generic_copy_file_range(ap->a_invp, ap->a_inoffp,
3827 		    ap->a_outvp, ap->a_outoffp, ap->a_lenp, ap->a_flags,
3828 		    ap->a_incred, ap->a_outcred, ap->a_fsizetd);
3829 	} else if (error != 0)
3830 		*ap->a_lenp = 0;
3831 
3832 	if (error != 0)
3833 		error = nfscl_maperr(curthread, error, (uid_t)0, (gid_t)0);
3834 	return (error);
3835 }
3836 
3837 /*
3838  * nfs ioctl call
3839  */
3840 static int
3841 nfs_ioctl(struct vop_ioctl_args *ap)
3842 {
3843 	struct vnode *vp = ap->a_vp;
3844 	struct nfsvattr nfsva;
3845 	struct nfsmount *nmp;
3846 	int attrflag, content, error, ret;
3847 	bool eof = false;			/* shut up compiler. */
3848 
3849 	if (vp->v_type != VREG)
3850 		return (ENOTTY);
3851 	nmp = VFSTONFS(vp->v_mount);
3852 	if (!NFSHASNFSV4(nmp) || nmp->nm_minorvers < NFSV42_MINORVERSION) {
3853 		error = vop_stdioctl(ap);
3854 		return (error);
3855 	}
3856 
3857 	/* Do the actual NFSv4.2 RPC. */
3858 	switch (ap->a_command) {
3859 	case FIOSEEKDATA:
3860 		content = NFSV4CONTENT_DATA;
3861 		break;
3862 	case FIOSEEKHOLE:
3863 		content = NFSV4CONTENT_HOLE;
3864 		break;
3865 	default:
3866 		return (ENOTTY);
3867 	}
3868 
3869 	error = vn_lock(vp, LK_SHARED);
3870 	if (error != 0)
3871 		return (EBADF);
3872 	attrflag = 0;
3873 	if (*((off_t *)ap->a_data) >= VTONFS(vp)->n_size)
3874 		error = ENXIO;
3875 	else {
3876 		/*
3877 		 * Flush all writes, so that the server is up to date.
3878 		 * Although a Commit is not required, the commit argument
3879 		 * is set so that, for a pNFS File/Flexible File Layout
3880 		 * server, the LayoutCommit will be done to ensure the file
3881 		 * size is up to date on the Metadata Server.
3882 		 */
3883 		error = ncl_flush(vp, MNT_WAIT, ap->a_td, 1, 0);
3884 		if (error == 0)
3885 			error = nfsrpc_seek(vp, (off_t *)ap->a_data, &eof,
3886 			    content, ap->a_cred, &nfsva, &attrflag);
3887 		/* If at eof for FIOSEEKDATA, return ENXIO. */
3888 		if (eof && error == 0 && content == NFSV4CONTENT_DATA)
3889 			error = ENXIO;
3890 	}
3891 	if (attrflag != 0) {
3892 		ret = nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0, 1);
3893 		if (error == 0 && ret != 0)
3894 			error = ret;
3895 	}
3896 	NFSVOPUNLOCK(vp);
3897 
3898 	if (error != 0)
3899 		error = ENXIO;
3900 	return (error);
3901 }
3902 
3903 /*
3904  * nfs getextattr call
3905  */
3906 static int
3907 nfs_getextattr(struct vop_getextattr_args *ap)
3908 {
3909 	struct vnode *vp = ap->a_vp;
3910 	struct nfsmount *nmp;
3911 	struct ucred *cred;
3912 	struct thread *td = ap->a_td;
3913 	struct nfsvattr nfsva;
3914 	ssize_t len;
3915 	int attrflag, error, ret;
3916 
3917 	nmp = VFSTONFS(vp->v_mount);
3918 	mtx_lock(&nmp->nm_mtx);
3919 	if (!NFSHASNFSV4(nmp) || nmp->nm_minorvers < NFSV42_MINORVERSION ||
3920 	    (nmp->nm_privflag & NFSMNTP_NOXATTR) != 0 ||
3921 	    ap->a_attrnamespace != EXTATTR_NAMESPACE_USER) {
3922 		mtx_unlock(&nmp->nm_mtx);
3923 		return (EOPNOTSUPP);
3924 	}
3925 	mtx_unlock(&nmp->nm_mtx);
3926 
3927 	cred = ap->a_cred;
3928 	if (cred == NULL)
3929 		cred = td->td_ucred;
3930 	/* Do the actual NFSv4.2 Optional Extended Attribute (RFC-8276) RPC. */
3931 	attrflag = 0;
3932 	error = nfsrpc_getextattr(vp, ap->a_name, ap->a_uio, &len, &nfsva,
3933 	    &attrflag, cred, td);
3934 	if (attrflag != 0) {
3935 		ret = nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0, 1);
3936 		if (error == 0 && ret != 0)
3937 			error = ret;
3938 	}
3939 	if (error == 0 && ap->a_size != NULL)
3940 		*ap->a_size = len;
3941 
3942 	switch (error) {
3943 	case NFSERR_NOTSUPP:
3944 	case NFSERR_OPILLEGAL:
3945 		mtx_lock(&nmp->nm_mtx);
3946 		nmp->nm_privflag |= NFSMNTP_NOXATTR;
3947 		mtx_unlock(&nmp->nm_mtx);
3948 		error = EOPNOTSUPP;
3949 		break;
3950 	case NFSERR_NOXATTR:
3951 	case NFSERR_XATTR2BIG:
3952 		error = ENOATTR;
3953 		break;
3954 	default:
3955 		error = nfscl_maperr(td, error, 0, 0);
3956 		break;
3957 	}
3958 	return (error);
3959 }
3960 
3961 /*
3962  * nfs setextattr call
3963  */
3964 static int
3965 nfs_setextattr(struct vop_setextattr_args *ap)
3966 {
3967 	struct vnode *vp = ap->a_vp;
3968 	struct nfsmount *nmp;
3969 	struct ucred *cred;
3970 	struct thread *td = ap->a_td;
3971 	struct nfsvattr nfsva;
3972 	int attrflag, error, ret;
3973 
3974 	nmp = VFSTONFS(vp->v_mount);
3975 	mtx_lock(&nmp->nm_mtx);
3976 	if (!NFSHASNFSV4(nmp) || nmp->nm_minorvers < NFSV42_MINORVERSION ||
3977 	    (nmp->nm_privflag & NFSMNTP_NOXATTR) != 0 ||
3978 	    ap->a_attrnamespace != EXTATTR_NAMESPACE_USER) {
3979 		mtx_unlock(&nmp->nm_mtx);
3980 		return (EOPNOTSUPP);
3981 	}
3982 	mtx_unlock(&nmp->nm_mtx);
3983 
3984 	if (ap->a_uio->uio_resid <= 0)
3985 		return (EINVAL);
3986 	cred = ap->a_cred;
3987 	if (cred == NULL)
3988 		cred = td->td_ucred;
3989 	/* Do the actual NFSv4.2 Optional Extended Attribute (RFC-8276) RPC. */
3990 	attrflag = 0;
3991 	error = nfsrpc_setextattr(vp, ap->a_name, ap->a_uio, &nfsva,
3992 	    &attrflag, cred, td);
3993 	if (attrflag != 0) {
3994 		ret = nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0, 1);
3995 		if (error == 0 && ret != 0)
3996 			error = ret;
3997 	}
3998 
3999 	switch (error) {
4000 	case NFSERR_NOTSUPP:
4001 	case NFSERR_OPILLEGAL:
4002 		mtx_lock(&nmp->nm_mtx);
4003 		nmp->nm_privflag |= NFSMNTP_NOXATTR;
4004 		mtx_unlock(&nmp->nm_mtx);
4005 		error = EOPNOTSUPP;
4006 		break;
4007 	case NFSERR_NOXATTR:
4008 	case NFSERR_XATTR2BIG:
4009 		error = ENOATTR;
4010 		break;
4011 	default:
4012 		error = nfscl_maperr(td, error, 0, 0);
4013 		break;
4014 	}
4015 	return (error);
4016 }
4017 
4018 /*
4019  * nfs listextattr call
4020  */
4021 static int
4022 nfs_listextattr(struct vop_listextattr_args *ap)
4023 {
4024 	struct vnode *vp = ap->a_vp;
4025 	struct nfsmount *nmp;
4026 	struct ucred *cred;
4027 	struct thread *td = ap->a_td;
4028 	struct nfsvattr nfsva;
4029 	size_t len, len2;
4030 	uint64_t cookie;
4031 	int attrflag, error, ret;
4032 	bool eof;
4033 
4034 	nmp = VFSTONFS(vp->v_mount);
4035 	mtx_lock(&nmp->nm_mtx);
4036 	if (!NFSHASNFSV4(nmp) || nmp->nm_minorvers < NFSV42_MINORVERSION ||
4037 	    (nmp->nm_privflag & NFSMNTP_NOXATTR) != 0 ||
4038 	    ap->a_attrnamespace != EXTATTR_NAMESPACE_USER) {
4039 		mtx_unlock(&nmp->nm_mtx);
4040 		return (EOPNOTSUPP);
4041 	}
4042 	mtx_unlock(&nmp->nm_mtx);
4043 
4044 	cred = ap->a_cred;
4045 	if (cred == NULL)
4046 		cred = td->td_ucred;
4047 
4048 	/* Loop around doing List Extended Attribute RPCs. */
4049 	eof = false;
4050 	cookie = 0;
4051 	len2 = 0;
4052 	error = 0;
4053 	while (!eof && error == 0) {
4054 		len = nmp->nm_rsize;
4055 		attrflag = 0;
4056 		error = nfsrpc_listextattr(vp, &cookie, ap->a_uio, &len, &eof,
4057 		    &nfsva, &attrflag, cred, td);
4058 		if (attrflag != 0) {
4059 			ret = nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0,
4060 			    1);
4061 			if (error == 0 && ret != 0)
4062 				error = ret;
4063 		}
4064 		if (error == 0) {
4065 			len2 += len;
4066 			if (len2 > SSIZE_MAX)
4067 				error = ENOATTR;
4068 		}
4069 	}
4070 	if (error == 0 && ap->a_size != NULL)
4071 		*ap->a_size = len2;
4072 
4073 	switch (error) {
4074 	case NFSERR_NOTSUPP:
4075 	case NFSERR_OPILLEGAL:
4076 		mtx_lock(&nmp->nm_mtx);
4077 		nmp->nm_privflag |= NFSMNTP_NOXATTR;
4078 		mtx_unlock(&nmp->nm_mtx);
4079 		error = EOPNOTSUPP;
4080 		break;
4081 	case NFSERR_NOXATTR:
4082 	case NFSERR_XATTR2BIG:
4083 		error = ENOATTR;
4084 		break;
4085 	default:
4086 		error = nfscl_maperr(td, error, 0, 0);
4087 		break;
4088 	}
4089 	return (error);
4090 }
4091 
4092 /*
4093  * nfs setextattr call
4094  */
4095 static int
4096 nfs_deleteextattr(struct vop_deleteextattr_args *ap)
4097 {
4098 	struct vnode *vp = ap->a_vp;
4099 	struct nfsmount *nmp;
4100 	struct nfsvattr nfsva;
4101 	int attrflag, error, ret;
4102 
4103 	nmp = VFSTONFS(vp->v_mount);
4104 	mtx_lock(&nmp->nm_mtx);
4105 	if (!NFSHASNFSV4(nmp) || nmp->nm_minorvers < NFSV42_MINORVERSION ||
4106 	    (nmp->nm_privflag & NFSMNTP_NOXATTR) != 0 ||
4107 	    ap->a_attrnamespace != EXTATTR_NAMESPACE_USER) {
4108 		mtx_unlock(&nmp->nm_mtx);
4109 		return (EOPNOTSUPP);
4110 	}
4111 	mtx_unlock(&nmp->nm_mtx);
4112 
4113 	/* Do the actual NFSv4.2 Optional Extended Attribute (RFC-8276) RPC. */
4114 	attrflag = 0;
4115 	error = nfsrpc_rmextattr(vp, ap->a_name, &nfsva, &attrflag, ap->a_cred,
4116 	    ap->a_td);
4117 	if (attrflag != 0) {
4118 		ret = nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0, 1);
4119 		if (error == 0 && ret != 0)
4120 			error = ret;
4121 	}
4122 
4123 	switch (error) {
4124 	case NFSERR_NOTSUPP:
4125 	case NFSERR_OPILLEGAL:
4126 		mtx_lock(&nmp->nm_mtx);
4127 		nmp->nm_privflag |= NFSMNTP_NOXATTR;
4128 		mtx_unlock(&nmp->nm_mtx);
4129 		error = EOPNOTSUPP;
4130 		break;
4131 	case NFSERR_NOXATTR:
4132 	case NFSERR_XATTR2BIG:
4133 		error = ENOATTR;
4134 		break;
4135 	default:
4136 		error = nfscl_maperr(ap->a_td, error, 0, 0);
4137 		break;
4138 	}
4139 	return (error);
4140 }
4141 
4142 /*
4143  * Return POSIX pathconf information applicable to nfs filesystems.
4144  */
4145 static int
4146 nfs_pathconf(struct vop_pathconf_args *ap)
4147 {
4148 	struct nfsv3_pathconf pc;
4149 	struct nfsvattr nfsva;
4150 	struct vnode *vp = ap->a_vp;
4151 	struct nfsmount *nmp;
4152 	struct thread *td = curthread;
4153 	off_t off;
4154 	bool eof;
4155 	int attrflag, error;
4156 
4157 	if ((NFS_ISV34(vp) && (ap->a_name == _PC_LINK_MAX ||
4158 	    ap->a_name == _PC_NAME_MAX || ap->a_name == _PC_CHOWN_RESTRICTED ||
4159 	    ap->a_name == _PC_NO_TRUNC)) ||
4160 	    (NFS_ISV4(vp) && ap->a_name == _PC_ACL_NFS4)) {
4161 		/*
4162 		 * Since only the above 4 a_names are returned by the NFSv3
4163 		 * Pathconf RPC, there is no point in doing it for others.
4164 		 * For NFSv4, the Pathconf RPC (actually a Getattr Op.) can
4165 		 * be used for _PC_NFS4_ACL as well.
4166 		 */
4167 		error = nfsrpc_pathconf(vp, &pc, td->td_ucred, td, &nfsva,
4168 		    &attrflag, NULL);
4169 		if (attrflag != 0)
4170 			(void) nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0,
4171 			    1);
4172 		if (error != 0)
4173 			return (error);
4174 	} else {
4175 		/*
4176 		 * For NFSv2 (or NFSv3 when not one of the above 4 a_names),
4177 		 * just fake them.
4178 		 */
4179 		pc.pc_linkmax = NFS_LINK_MAX;
4180 		pc.pc_namemax = NFS_MAXNAMLEN;
4181 		pc.pc_notrunc = 1;
4182 		pc.pc_chownrestricted = 1;
4183 		pc.pc_caseinsensitive = 0;
4184 		pc.pc_casepreserving = 1;
4185 		error = 0;
4186 	}
4187 	switch (ap->a_name) {
4188 	case _PC_LINK_MAX:
4189 #ifdef _LP64
4190 		*ap->a_retval = pc.pc_linkmax;
4191 #else
4192 		*ap->a_retval = MIN(LONG_MAX, pc.pc_linkmax);
4193 #endif
4194 		break;
4195 	case _PC_NAME_MAX:
4196 		*ap->a_retval = pc.pc_namemax;
4197 		break;
4198 	case _PC_PIPE_BUF:
4199 		if (ap->a_vp->v_type == VDIR || ap->a_vp->v_type == VFIFO)
4200 			*ap->a_retval = PIPE_BUF;
4201 		else
4202 			error = EINVAL;
4203 		break;
4204 	case _PC_CHOWN_RESTRICTED:
4205 		*ap->a_retval = pc.pc_chownrestricted;
4206 		break;
4207 	case _PC_NO_TRUNC:
4208 		*ap->a_retval = pc.pc_notrunc;
4209 		break;
4210 	case _PC_ACL_NFS4:
4211 		if (NFS_ISV4(vp) && nfsrv_useacl != 0 && attrflag != 0 &&
4212 		    NFSISSET_ATTRBIT(&nfsva.na_suppattr, NFSATTRBIT_ACL))
4213 			*ap->a_retval = 1;
4214 		else
4215 			*ap->a_retval = 0;
4216 		break;
4217 	case _PC_ACL_PATH_MAX:
4218 		if (NFS_ISV4(vp))
4219 			*ap->a_retval = ACL_MAX_ENTRIES;
4220 		else
4221 			*ap->a_retval = 3;
4222 		break;
4223 	case _PC_PRIO_IO:
4224 		*ap->a_retval = 0;
4225 		break;
4226 	case _PC_SYNC_IO:
4227 		*ap->a_retval = 0;
4228 		break;
4229 	case _PC_ALLOC_SIZE_MIN:
4230 		*ap->a_retval = vp->v_mount->mnt_stat.f_bsize;
4231 		break;
4232 	case _PC_FILESIZEBITS:
4233 		if (NFS_ISV34(vp))
4234 			*ap->a_retval = 64;
4235 		else
4236 			*ap->a_retval = 32;
4237 		break;
4238 	case _PC_REC_INCR_XFER_SIZE:
4239 		*ap->a_retval = vp->v_mount->mnt_stat.f_iosize;
4240 		break;
4241 	case _PC_REC_MAX_XFER_SIZE:
4242 		*ap->a_retval = -1; /* means ``unlimited'' */
4243 		break;
4244 	case _PC_REC_MIN_XFER_SIZE:
4245 		*ap->a_retval = vp->v_mount->mnt_stat.f_iosize;
4246 		break;
4247 	case _PC_REC_XFER_ALIGN:
4248 		*ap->a_retval = PAGE_SIZE;
4249 		break;
4250 	case _PC_SYMLINK_MAX:
4251 		*ap->a_retval = NFS_MAXPATHLEN;
4252 		break;
4253 	case _PC_MIN_HOLE_SIZE:
4254 		/* Only some NFSv4.2 servers support Seek for Holes. */
4255 		*ap->a_retval = 0;
4256 		nmp = VFSTONFS(vp->v_mount);
4257 		if (NFS_ISV4(vp) && nmp->nm_minorvers == NFSV42_MINORVERSION) {
4258 			/*
4259 			 * NFSv4.2 doesn't have an attribute for hole size,
4260 			 * so all we can do is see if the Seek operation is
4261 			 * supported and then use f_iosize as a "best guess".
4262 			 */
4263 			mtx_lock(&nmp->nm_mtx);
4264 			if ((nmp->nm_privflag & NFSMNTP_SEEKTESTED) == 0) {
4265 				mtx_unlock(&nmp->nm_mtx);
4266 				off = 0;
4267 				attrflag = 0;
4268 				error = nfsrpc_seek(vp, &off, &eof,
4269 				    NFSV4CONTENT_HOLE, td->td_ucred, &nfsva,
4270 				    &attrflag);
4271 				if (attrflag != 0)
4272 					nfscl_loadattrcache(&vp, &nfsva,
4273 					    NULL, NULL, 0, 1);
4274 				mtx_lock(&nmp->nm_mtx);
4275 				if (error == NFSERR_NOTSUPP)
4276 					nmp->nm_privflag |= NFSMNTP_SEEKTESTED;
4277 				else
4278 					nmp->nm_privflag |= NFSMNTP_SEEKTESTED |
4279 					    NFSMNTP_SEEK;
4280 				error = 0;
4281 			}
4282 			if ((nmp->nm_privflag & NFSMNTP_SEEK) != 0)
4283 				*ap->a_retval = vp->v_mount->mnt_stat.f_iosize;
4284 			mtx_unlock(&nmp->nm_mtx);
4285 		}
4286 		break;
4287 
4288 	default:
4289 		error = vop_stdpathconf(ap);
4290 		break;
4291 	}
4292 	return (error);
4293 }
4294 
4295