xref: /freebsd/sys/fs/nfsclient/nfs_clvnops.c (revision d396c67f26b079c2808002c07212d9df9818a11b)
1 /*-
2  * SPDX-License-Identifier: BSD-3-Clause
3  *
4  * Copyright (c) 1989, 1993
5  *	The Regents of the University of California.  All rights reserved.
6  *
7  * This code is derived from software contributed to Berkeley by
8  * Rick Macklem at The University of Guelph.
9  *
10  * Redistribution and use in source and binary forms, with or without
11  * modification, are permitted provided that the following conditions
12  * are met:
13  * 1. Redistributions of source code must retain the above copyright
14  *    notice, this list of conditions and the following disclaimer.
15  * 2. Redistributions in binary form must reproduce the above copyright
16  *    notice, this list of conditions and the following disclaimer in the
17  *    documentation and/or other materials provided with the distribution.
18  * 3. Neither the name of the University nor the names of its contributors
19  *    may be used to endorse or promote products derived from this software
20  *    without specific prior written permission.
21  *
22  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
23  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
24  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
25  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
26  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
27  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
28  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
29  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
30  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
31  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
32  * SUCH DAMAGE.
33  *
34  *	from nfs_vnops.c	8.16 (Berkeley) 5/27/95
35  */
36 
37 #include <sys/cdefs.h>
38 __FBSDID("$FreeBSD$");
39 
40 /*
41  * vnode op calls for Sun NFS version 2, 3 and 4
42  */
43 
44 #include "opt_inet.h"
45 
46 #include <sys/param.h>
47 #include <sys/kernel.h>
48 #include <sys/systm.h>
49 #include <sys/resourcevar.h>
50 #include <sys/proc.h>
51 #include <sys/mount.h>
52 #include <sys/bio.h>
53 #include <sys/buf.h>
54 #include <sys/extattr.h>
55 #include <sys/filio.h>
56 #include <sys/jail.h>
57 #include <sys/malloc.h>
58 #include <sys/mbuf.h>
59 #include <sys/namei.h>
60 #include <sys/socket.h>
61 #include <sys/vnode.h>
62 #include <sys/dirent.h>
63 #include <sys/fcntl.h>
64 #include <sys/lockf.h>
65 #include <sys/stat.h>
66 #include <sys/sysctl.h>
67 #include <sys/signalvar.h>
68 
69 #include <vm/vm.h>
70 #include <vm/vm_extern.h>
71 #include <vm/vm_object.h>
72 
73 #include <fs/nfs/nfsport.h>
74 #include <fs/nfsclient/nfsnode.h>
75 #include <fs/nfsclient/nfsmount.h>
76 #include <fs/nfsclient/nfs.h>
77 #include <fs/nfsclient/nfs_kdtrace.h>
78 
79 #include <net/if.h>
80 #include <netinet/in.h>
81 #include <netinet/in_var.h>
82 
83 #include <nfs/nfs_lock.h>
84 
85 #ifdef KDTRACE_HOOKS
86 #include <sys/dtrace_bsd.h>
87 
88 dtrace_nfsclient_accesscache_flush_probe_func_t
89 		dtrace_nfscl_accesscache_flush_done_probe;
90 uint32_t	nfscl_accesscache_flush_done_id;
91 
92 dtrace_nfsclient_accesscache_get_probe_func_t
93 		dtrace_nfscl_accesscache_get_hit_probe,
94 		dtrace_nfscl_accesscache_get_miss_probe;
95 uint32_t	nfscl_accesscache_get_hit_id;
96 uint32_t	nfscl_accesscache_get_miss_id;
97 
98 dtrace_nfsclient_accesscache_load_probe_func_t
99 		dtrace_nfscl_accesscache_load_done_probe;
100 uint32_t	nfscl_accesscache_load_done_id;
101 #endif /* !KDTRACE_HOOKS */
102 
103 /* Defs */
104 #define	TRUE	1
105 #define	FALSE	0
106 
107 extern struct nfsstatsv1 nfsstatsv1;
108 extern int nfsrv_useacl;
109 extern int nfscl_debuglevel;
110 MALLOC_DECLARE(M_NEWNFSREQ);
111 
112 static vop_read_t	nfsfifo_read;
113 static vop_write_t	nfsfifo_write;
114 static vop_close_t	nfsfifo_close;
115 static int	nfs_setattrrpc(struct vnode *, struct vattr *, struct ucred *,
116 		    struct thread *);
117 static vop_lookup_t	nfs_lookup;
118 static vop_create_t	nfs_create;
119 static vop_mknod_t	nfs_mknod;
120 static vop_open_t	nfs_open;
121 static vop_pathconf_t	nfs_pathconf;
122 static vop_close_t	nfs_close;
123 static vop_access_t	nfs_access;
124 static vop_getattr_t	nfs_getattr;
125 static vop_setattr_t	nfs_setattr;
126 static vop_read_t	nfs_read;
127 static vop_fsync_t	nfs_fsync;
128 static vop_remove_t	nfs_remove;
129 static vop_link_t	nfs_link;
130 static vop_rename_t	nfs_rename;
131 static vop_mkdir_t	nfs_mkdir;
132 static vop_rmdir_t	nfs_rmdir;
133 static vop_symlink_t	nfs_symlink;
134 static vop_readdir_t	nfs_readdir;
135 static vop_strategy_t	nfs_strategy;
136 static	int	nfs_lookitup(struct vnode *, char *, int,
137 		    struct ucred *, struct thread *, struct nfsnode **);
138 static	int	nfs_sillyrename(struct vnode *, struct vnode *,
139 		    struct componentname *);
140 static vop_access_t	nfsspec_access;
141 static vop_readlink_t	nfs_readlink;
142 static vop_print_t	nfs_print;
143 static vop_advlock_t	nfs_advlock;
144 static vop_advlockasync_t nfs_advlockasync;
145 static vop_getacl_t nfs_getacl;
146 static vop_setacl_t nfs_setacl;
147 static vop_advise_t nfs_advise;
148 static vop_allocate_t nfs_allocate;
149 static vop_copy_file_range_t nfs_copy_file_range;
150 static vop_ioctl_t nfs_ioctl;
151 static vop_getextattr_t nfs_getextattr;
152 static vop_setextattr_t nfs_setextattr;
153 static vop_listextattr_t nfs_listextattr;
154 static vop_deleteextattr_t nfs_deleteextattr;
155 static vop_lock1_t	nfs_lock;
156 
157 /*
158  * Global vfs data structures for nfs
159  */
160 
161 static struct vop_vector newnfs_vnodeops_nosig = {
162 	.vop_default =		&default_vnodeops,
163 	.vop_access =		nfs_access,
164 	.vop_advlock =		nfs_advlock,
165 	.vop_advlockasync =	nfs_advlockasync,
166 	.vop_close =		nfs_close,
167 	.vop_create =		nfs_create,
168 	.vop_fsync =		nfs_fsync,
169 	.vop_getattr =		nfs_getattr,
170 	.vop_getpages =		ncl_getpages,
171 	.vop_putpages =		ncl_putpages,
172 	.vop_inactive =		ncl_inactive,
173 	.vop_link =		nfs_link,
174 	.vop_lock1 =		nfs_lock,
175 	.vop_lookup =		nfs_lookup,
176 	.vop_mkdir =		nfs_mkdir,
177 	.vop_mknod =		nfs_mknod,
178 	.vop_open =		nfs_open,
179 	.vop_pathconf =		nfs_pathconf,
180 	.vop_print =		nfs_print,
181 	.vop_read =		nfs_read,
182 	.vop_readdir =		nfs_readdir,
183 	.vop_readlink =		nfs_readlink,
184 	.vop_reclaim =		ncl_reclaim,
185 	.vop_remove =		nfs_remove,
186 	.vop_rename =		nfs_rename,
187 	.vop_rmdir =		nfs_rmdir,
188 	.vop_setattr =		nfs_setattr,
189 	.vop_strategy =		nfs_strategy,
190 	.vop_symlink =		nfs_symlink,
191 	.vop_write =		ncl_write,
192 	.vop_getacl =		nfs_getacl,
193 	.vop_setacl =		nfs_setacl,
194 	.vop_advise =		nfs_advise,
195 	.vop_allocate =		nfs_allocate,
196 	.vop_copy_file_range =	nfs_copy_file_range,
197 	.vop_ioctl =		nfs_ioctl,
198 	.vop_getextattr =	nfs_getextattr,
199 	.vop_setextattr =	nfs_setextattr,
200 	.vop_listextattr =	nfs_listextattr,
201 	.vop_deleteextattr =	nfs_deleteextattr,
202 };
203 VFS_VOP_VECTOR_REGISTER(newnfs_vnodeops_nosig);
204 
205 static int
206 nfs_vnodeops_bypass(struct vop_generic_args *a)
207 {
208 
209 	return (vop_sigdefer(&newnfs_vnodeops_nosig, a));
210 }
211 
212 struct vop_vector newnfs_vnodeops = {
213 	.vop_default =		&default_vnodeops,
214 	.vop_bypass =		nfs_vnodeops_bypass,
215 };
216 VFS_VOP_VECTOR_REGISTER(newnfs_vnodeops);
217 
218 static struct vop_vector newnfs_fifoops_nosig = {
219 	.vop_default =		&fifo_specops,
220 	.vop_access =		nfsspec_access,
221 	.vop_close =		nfsfifo_close,
222 	.vop_fsync =		nfs_fsync,
223 	.vop_getattr =		nfs_getattr,
224 	.vop_inactive =		ncl_inactive,
225 	.vop_pathconf =		nfs_pathconf,
226 	.vop_print =		nfs_print,
227 	.vop_read =		nfsfifo_read,
228 	.vop_reclaim =		ncl_reclaim,
229 	.vop_setattr =		nfs_setattr,
230 	.vop_write =		nfsfifo_write,
231 };
232 VFS_VOP_VECTOR_REGISTER(newnfs_fifoops_nosig);
233 
234 static int
235 nfs_fifoops_bypass(struct vop_generic_args *a)
236 {
237 
238 	return (vop_sigdefer(&newnfs_fifoops_nosig, a));
239 }
240 
241 struct vop_vector newnfs_fifoops = {
242 	.vop_default =		&default_vnodeops,
243 	.vop_bypass =		nfs_fifoops_bypass,
244 };
245 VFS_VOP_VECTOR_REGISTER(newnfs_fifoops);
246 
247 static int nfs_mknodrpc(struct vnode *dvp, struct vnode **vpp,
248     struct componentname *cnp, struct vattr *vap);
249 static int nfs_removerpc(struct vnode *dvp, struct vnode *vp, char *name,
250     int namelen, struct ucred *cred, struct thread *td);
251 static int nfs_renamerpc(struct vnode *fdvp, struct vnode *fvp,
252     char *fnameptr, int fnamelen, struct vnode *tdvp, struct vnode *tvp,
253     char *tnameptr, int tnamelen, struct ucred *cred, struct thread *td);
254 static int nfs_renameit(struct vnode *sdvp, struct vnode *svp,
255     struct componentname *scnp, struct sillyrename *sp);
256 
257 /*
258  * Global variables
259  */
260 SYSCTL_DECL(_vfs_nfs);
261 
262 static int	nfsaccess_cache_timeout = NFS_MAXATTRTIMO;
263 SYSCTL_INT(_vfs_nfs, OID_AUTO, access_cache_timeout, CTLFLAG_RW,
264 	   &nfsaccess_cache_timeout, 0, "NFS ACCESS cache timeout");
265 
266 static int	nfs_prime_access_cache = 0;
267 SYSCTL_INT(_vfs_nfs, OID_AUTO, prime_access_cache, CTLFLAG_RW,
268 	   &nfs_prime_access_cache, 0,
269 	   "Prime NFS ACCESS cache when fetching attributes");
270 
271 static int	newnfs_commit_on_close = 0;
272 SYSCTL_INT(_vfs_nfs, OID_AUTO, commit_on_close, CTLFLAG_RW,
273     &newnfs_commit_on_close, 0, "write+commit on close, else only write");
274 
275 static int	nfs_clean_pages_on_close = 1;
276 SYSCTL_INT(_vfs_nfs, OID_AUTO, clean_pages_on_close, CTLFLAG_RW,
277 	   &nfs_clean_pages_on_close, 0, "NFS clean dirty pages on close");
278 
279 int newnfs_directio_enable = 0;
280 SYSCTL_INT(_vfs_nfs, OID_AUTO, nfs_directio_enable, CTLFLAG_RW,
281 	   &newnfs_directio_enable, 0, "Enable NFS directio");
282 
283 int nfs_keep_dirty_on_error;
284 SYSCTL_INT(_vfs_nfs, OID_AUTO, nfs_keep_dirty_on_error, CTLFLAG_RW,
285     &nfs_keep_dirty_on_error, 0, "Retry pageout if error returned");
286 
287 /*
288  * This sysctl allows other processes to mmap a file that has been opened
289  * O_DIRECT by a process.  In general, having processes mmap the file while
290  * Direct IO is in progress can lead to Data Inconsistencies.  But, we allow
291  * this by default to prevent DoS attacks - to prevent a malicious user from
292  * opening up files O_DIRECT preventing other users from mmap'ing these
293  * files.  "Protected" environments where stricter consistency guarantees are
294  * required can disable this knob.  The process that opened the file O_DIRECT
295  * cannot mmap() the file, because mmap'ed IO on an O_DIRECT open() is not
296  * meaningful.
297  */
298 int newnfs_directio_allow_mmap = 1;
299 SYSCTL_INT(_vfs_nfs, OID_AUTO, nfs_directio_allow_mmap, CTLFLAG_RW,
300 	   &newnfs_directio_allow_mmap, 0, "Enable mmaped IO on file with O_DIRECT opens");
301 
302 #define	NFSACCESS_ALL (NFSACCESS_READ | NFSACCESS_MODIFY		\
303 			 | NFSACCESS_EXTEND | NFSACCESS_EXECUTE	\
304 			 | NFSACCESS_DELETE | NFSACCESS_LOOKUP)
305 
306 /*
307  * SMP Locking Note :
308  * The list of locks after the description of the lock is the ordering
309  * of other locks acquired with the lock held.
310  * np->n_mtx : Protects the fields in the nfsnode.
311        VM Object Lock
312        VI_MTX (acquired indirectly)
313  * nmp->nm_mtx : Protects the fields in the nfsmount.
314        rep->r_mtx
315  * ncl_iod_mutex : Global lock, protects shared nfsiod state.
316  * nfs_reqq_mtx : Global lock, protects the nfs_reqq list.
317        nmp->nm_mtx
318        rep->r_mtx
319  * rep->r_mtx : Protects the fields in an nfsreq.
320  */
321 
322 static int
323 nfs_lock(struct vop_lock1_args *ap)
324 {
325 	struct vnode *vp;
326 	struct nfsnode *np;
327 	u_quad_t nsize;
328 	int error, lktype;
329 	bool onfault;
330 
331 	vp = ap->a_vp;
332 	lktype = ap->a_flags & LK_TYPE_MASK;
333 	error = VOP_LOCK1_APV(&default_vnodeops, ap);
334 	if (error != 0 || vp->v_op != &newnfs_vnodeops)
335 		return (error);
336 	np = VTONFS(vp);
337 	if (np == NULL)
338 		return (0);
339 	NFSLOCKNODE(np);
340 	if ((np->n_flag & NVNSETSZSKIP) == 0 || (lktype != LK_SHARED &&
341 	    lktype != LK_EXCLUSIVE && lktype != LK_UPGRADE &&
342 	    lktype != LK_TRYUPGRADE)) {
343 		NFSUNLOCKNODE(np);
344 		return (0);
345 	}
346 	onfault = (ap->a_flags & LK_EATTR_MASK) == LK_NOWAIT &&
347 	    (ap->a_flags & LK_INIT_MASK) == LK_CANRECURSE &&
348 	    (lktype == LK_SHARED || lktype == LK_EXCLUSIVE);
349 	if (onfault && vp->v_vnlock->lk_recurse == 0) {
350 		/*
351 		 * Force retry in vm_fault(), to make the lock request
352 		 * sleepable, which allows us to piggy-back the
353 		 * sleepable call to vnode_pager_setsize().
354 		 */
355 		NFSUNLOCKNODE(np);
356 		VOP_UNLOCK(vp);
357 		return (EBUSY);
358 	}
359 	if ((ap->a_flags & LK_NOWAIT) != 0 ||
360 	    (lktype == LK_SHARED && vp->v_vnlock->lk_recurse > 0)) {
361 		NFSUNLOCKNODE(np);
362 		return (0);
363 	}
364 	if (lktype == LK_SHARED) {
365 		NFSUNLOCKNODE(np);
366 		VOP_UNLOCK(vp);
367 		ap->a_flags &= ~(LK_TYPE_MASK | LK_INTERLOCK);
368 		ap->a_flags |= LK_EXCLUSIVE;
369 		error = VOP_LOCK1_APV(&default_vnodeops, ap);
370 		if (error != 0 || vp->v_op != &newnfs_vnodeops)
371 			return (error);
372 		if (vp->v_data == NULL)
373 			goto downgrade;
374 		MPASS(vp->v_data == np);
375 		NFSLOCKNODE(np);
376 		if ((np->n_flag & NVNSETSZSKIP) == 0) {
377 			NFSUNLOCKNODE(np);
378 			goto downgrade;
379 		}
380 	}
381 	np->n_flag &= ~NVNSETSZSKIP;
382 	nsize = np->n_size;
383 	NFSUNLOCKNODE(np);
384 	vnode_pager_setsize(vp, nsize);
385 downgrade:
386 	if (lktype == LK_SHARED) {
387 		ap->a_flags &= ~(LK_TYPE_MASK | LK_INTERLOCK);
388 		ap->a_flags |= LK_DOWNGRADE;
389 		(void)VOP_LOCK1_APV(&default_vnodeops, ap);
390 	}
391 	return (0);
392 }
393 
394 static int
395 nfs34_access_otw(struct vnode *vp, int wmode, struct thread *td,
396     struct ucred *cred, u_int32_t *retmode)
397 {
398 	int error = 0, attrflag, i, lrupos;
399 	u_int32_t rmode;
400 	struct nfsnode *np = VTONFS(vp);
401 	struct nfsvattr nfsva;
402 
403 	error = nfsrpc_accessrpc(vp, wmode, cred, td, &nfsva, &attrflag,
404 	    &rmode, NULL);
405 	if (attrflag)
406 		(void) nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0, 1);
407 	if (!error) {
408 		lrupos = 0;
409 		NFSLOCKNODE(np);
410 		for (i = 0; i < NFS_ACCESSCACHESIZE; i++) {
411 			if (np->n_accesscache[i].uid == cred->cr_uid) {
412 				np->n_accesscache[i].mode = rmode;
413 				np->n_accesscache[i].stamp = time_second;
414 				break;
415 			}
416 			if (i > 0 && np->n_accesscache[i].stamp <
417 			    np->n_accesscache[lrupos].stamp)
418 				lrupos = i;
419 		}
420 		if (i == NFS_ACCESSCACHESIZE) {
421 			np->n_accesscache[lrupos].uid = cred->cr_uid;
422 			np->n_accesscache[lrupos].mode = rmode;
423 			np->n_accesscache[lrupos].stamp = time_second;
424 		}
425 		NFSUNLOCKNODE(np);
426 		if (retmode != NULL)
427 			*retmode = rmode;
428 		KDTRACE_NFS_ACCESSCACHE_LOAD_DONE(vp, cred->cr_uid, rmode, 0);
429 	} else if (NFS_ISV4(vp)) {
430 		error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0);
431 	}
432 #ifdef KDTRACE_HOOKS
433 	if (error != 0)
434 		KDTRACE_NFS_ACCESSCACHE_LOAD_DONE(vp, cred->cr_uid, 0,
435 		    error);
436 #endif
437 	return (error);
438 }
439 
440 /*
441  * nfs access vnode op.
442  * For nfs version 2, just return ok. File accesses may fail later.
443  * For nfs version 3, use the access rpc to check accessibility. If file modes
444  * are changed on the server, accesses might still fail later.
445  */
446 static int
447 nfs_access(struct vop_access_args *ap)
448 {
449 	struct vnode *vp = ap->a_vp;
450 	int error = 0, i, gotahit;
451 	u_int32_t mode, wmode, rmode;
452 	int v34 = NFS_ISV34(vp);
453 	struct nfsnode *np = VTONFS(vp);
454 
455 	/*
456 	 * Disallow write attempts on filesystems mounted read-only;
457 	 * unless the file is a socket, fifo, or a block or character
458 	 * device resident on the filesystem.
459 	 */
460 	if ((ap->a_accmode & (VWRITE | VAPPEND | VWRITE_NAMED_ATTRS |
461 	    VDELETE_CHILD | VWRITE_ATTRIBUTES | VDELETE | VWRITE_ACL |
462 	    VWRITE_OWNER)) != 0 && (vp->v_mount->mnt_flag & MNT_RDONLY) != 0) {
463 		switch (vp->v_type) {
464 		case VREG:
465 		case VDIR:
466 		case VLNK:
467 			return (EROFS);
468 		default:
469 			break;
470 		}
471 	}
472 	/*
473 	 * For nfs v3 or v4, check to see if we have done this recently, and if
474 	 * so return our cached result instead of making an ACCESS call.
475 	 * If not, do an access rpc, otherwise you are stuck emulating
476 	 * ufs_access() locally using the vattr. This may not be correct,
477 	 * since the server may apply other access criteria such as
478 	 * client uid-->server uid mapping that we do not know about.
479 	 */
480 	if (v34) {
481 		if (ap->a_accmode & VREAD)
482 			mode = NFSACCESS_READ;
483 		else
484 			mode = 0;
485 		if (vp->v_type != VDIR) {
486 			if (ap->a_accmode & VWRITE)
487 				mode |= (NFSACCESS_MODIFY | NFSACCESS_EXTEND);
488 			if (ap->a_accmode & VAPPEND)
489 				mode |= NFSACCESS_EXTEND;
490 			if (ap->a_accmode & VEXEC)
491 				mode |= NFSACCESS_EXECUTE;
492 			if (ap->a_accmode & VDELETE)
493 				mode |= NFSACCESS_DELETE;
494 		} else {
495 			if (ap->a_accmode & VWRITE)
496 				mode |= (NFSACCESS_MODIFY | NFSACCESS_EXTEND);
497 			if (ap->a_accmode & VAPPEND)
498 				mode |= NFSACCESS_EXTEND;
499 			if (ap->a_accmode & VEXEC)
500 				mode |= NFSACCESS_LOOKUP;
501 			if (ap->a_accmode & VDELETE)
502 				mode |= NFSACCESS_DELETE;
503 			if (ap->a_accmode & VDELETE_CHILD)
504 				mode |= NFSACCESS_MODIFY;
505 		}
506 		/* XXX safety belt, only make blanket request if caching */
507 		if (nfsaccess_cache_timeout > 0) {
508 			wmode = NFSACCESS_READ | NFSACCESS_MODIFY |
509 				NFSACCESS_EXTEND | NFSACCESS_EXECUTE |
510 				NFSACCESS_DELETE | NFSACCESS_LOOKUP;
511 		} else {
512 			wmode = mode;
513 		}
514 
515 		/*
516 		 * Does our cached result allow us to give a definite yes to
517 		 * this request?
518 		 */
519 		gotahit = 0;
520 		NFSLOCKNODE(np);
521 		for (i = 0; i < NFS_ACCESSCACHESIZE; i++) {
522 			if (ap->a_cred->cr_uid == np->n_accesscache[i].uid) {
523 			    if (time_second < (np->n_accesscache[i].stamp
524 				+ nfsaccess_cache_timeout) &&
525 				(np->n_accesscache[i].mode & mode) == mode) {
526 				NFSINCRGLOBAL(nfsstatsv1.accesscache_hits);
527 				gotahit = 1;
528 			    }
529 			    break;
530 			}
531 		}
532 		NFSUNLOCKNODE(np);
533 #ifdef KDTRACE_HOOKS
534 		if (gotahit != 0)
535 			KDTRACE_NFS_ACCESSCACHE_GET_HIT(vp,
536 			    ap->a_cred->cr_uid, mode);
537 		else
538 			KDTRACE_NFS_ACCESSCACHE_GET_MISS(vp,
539 			    ap->a_cred->cr_uid, mode);
540 #endif
541 		if (gotahit == 0) {
542 			/*
543 			 * Either a no, or a don't know.  Go to the wire.
544 			 */
545 			NFSINCRGLOBAL(nfsstatsv1.accesscache_misses);
546 		        error = nfs34_access_otw(vp, wmode, ap->a_td,
547 			    ap->a_cred, &rmode);
548 			if (!error &&
549 			    (rmode & mode) != mode)
550 				error = EACCES;
551 		}
552 		return (error);
553 	} else {
554 		if ((error = nfsspec_access(ap)) != 0) {
555 			return (error);
556 		}
557 		/*
558 		 * Attempt to prevent a mapped root from accessing a file
559 		 * which it shouldn't.  We try to read a byte from the file
560 		 * if the user is root and the file is not zero length.
561 		 * After calling nfsspec_access, we should have the correct
562 		 * file size cached.
563 		 */
564 		NFSLOCKNODE(np);
565 		if (ap->a_cred->cr_uid == 0 && (ap->a_accmode & VREAD)
566 		    && VTONFS(vp)->n_size > 0) {
567 			struct iovec aiov;
568 			struct uio auio;
569 			char buf[1];
570 
571 			NFSUNLOCKNODE(np);
572 			aiov.iov_base = buf;
573 			aiov.iov_len = 1;
574 			auio.uio_iov = &aiov;
575 			auio.uio_iovcnt = 1;
576 			auio.uio_offset = 0;
577 			auio.uio_resid = 1;
578 			auio.uio_segflg = UIO_SYSSPACE;
579 			auio.uio_rw = UIO_READ;
580 			auio.uio_td = ap->a_td;
581 
582 			if (vp->v_type == VREG)
583 				error = ncl_readrpc(vp, &auio, ap->a_cred);
584 			else if (vp->v_type == VDIR) {
585 				char* bp;
586 				bp = malloc(NFS_DIRBLKSIZ, M_TEMP, M_WAITOK);
587 				aiov.iov_base = bp;
588 				aiov.iov_len = auio.uio_resid = NFS_DIRBLKSIZ;
589 				error = ncl_readdirrpc(vp, &auio, ap->a_cred,
590 				    ap->a_td);
591 				free(bp, M_TEMP);
592 			} else if (vp->v_type == VLNK)
593 				error = ncl_readlinkrpc(vp, &auio, ap->a_cred);
594 			else
595 				error = EACCES;
596 		} else
597 			NFSUNLOCKNODE(np);
598 		return (error);
599 	}
600 }
601 
602 /*
603  * nfs open vnode op
604  * Check to see if the type is ok
605  * and that deletion is not in progress.
606  * For paged in text files, you will need to flush the page cache
607  * if consistency is lost.
608  */
609 /* ARGSUSED */
610 static int
611 nfs_open(struct vop_open_args *ap)
612 {
613 	struct vnode *vp = ap->a_vp;
614 	struct nfsnode *np = VTONFS(vp);
615 	struct vattr vattr;
616 	int error;
617 	int fmode = ap->a_mode;
618 	struct ucred *cred;
619 	vm_object_t obj;
620 
621 	if (vp->v_type != VREG && vp->v_type != VDIR && vp->v_type != VLNK)
622 		return (EOPNOTSUPP);
623 
624 	/*
625 	 * For NFSv4, we need to do the Open Op before cache validation,
626 	 * so that we conform to RFC3530 Sec. 9.3.1.
627 	 */
628 	if (NFS_ISV4(vp)) {
629 		error = nfsrpc_open(vp, fmode, ap->a_cred, ap->a_td);
630 		if (error) {
631 			error = nfscl_maperr(ap->a_td, error, (uid_t)0,
632 			    (gid_t)0);
633 			return (error);
634 		}
635 	}
636 
637 	/*
638 	 * Now, if this Open will be doing reading, re-validate/flush the
639 	 * cache, so that Close/Open coherency is maintained.
640 	 */
641 	NFSLOCKNODE(np);
642 	if (np->n_flag & NMODIFIED) {
643 		NFSUNLOCKNODE(np);
644 		error = ncl_vinvalbuf(vp, V_SAVE, ap->a_td, 1);
645 		if (error == EINTR || error == EIO) {
646 			if (NFS_ISV4(vp))
647 				(void) nfsrpc_close(vp, 0, ap->a_td);
648 			return (error);
649 		}
650 		NFSLOCKNODE(np);
651 		np->n_attrstamp = 0;
652 		KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(vp);
653 		if (vp->v_type == VDIR)
654 			np->n_direofoffset = 0;
655 		NFSUNLOCKNODE(np);
656 		error = VOP_GETATTR(vp, &vattr, ap->a_cred);
657 		if (error) {
658 			if (NFS_ISV4(vp))
659 				(void) nfsrpc_close(vp, 0, ap->a_td);
660 			return (error);
661 		}
662 		NFSLOCKNODE(np);
663 		np->n_mtime = vattr.va_mtime;
664 		if (NFS_ISV4(vp))
665 			np->n_change = vattr.va_filerev;
666 	} else {
667 		NFSUNLOCKNODE(np);
668 		error = VOP_GETATTR(vp, &vattr, ap->a_cred);
669 		if (error) {
670 			if (NFS_ISV4(vp))
671 				(void) nfsrpc_close(vp, 0, ap->a_td);
672 			return (error);
673 		}
674 		NFSLOCKNODE(np);
675 		if ((NFS_ISV4(vp) && np->n_change != vattr.va_filerev) ||
676 		    NFS_TIMESPEC_COMPARE(&np->n_mtime, &vattr.va_mtime)) {
677 			if (vp->v_type == VDIR)
678 				np->n_direofoffset = 0;
679 			NFSUNLOCKNODE(np);
680 			error = ncl_vinvalbuf(vp, V_SAVE, ap->a_td, 1);
681 			if (error == EINTR || error == EIO) {
682 				if (NFS_ISV4(vp))
683 					(void) nfsrpc_close(vp, 0, ap->a_td);
684 				return (error);
685 			}
686 			NFSLOCKNODE(np);
687 			np->n_mtime = vattr.va_mtime;
688 			if (NFS_ISV4(vp))
689 				np->n_change = vattr.va_filerev;
690 		}
691 	}
692 
693 	/*
694 	 * If the object has >= 1 O_DIRECT active opens, we disable caching.
695 	 */
696 	if (newnfs_directio_enable && (fmode & O_DIRECT) &&
697 	    (vp->v_type == VREG)) {
698 		if (np->n_directio_opens == 0) {
699 			NFSUNLOCKNODE(np);
700 			error = ncl_vinvalbuf(vp, V_SAVE, ap->a_td, 1);
701 			if (error) {
702 				if (NFS_ISV4(vp))
703 					(void) nfsrpc_close(vp, 0, ap->a_td);
704 				return (error);
705 			}
706 			NFSLOCKNODE(np);
707 			np->n_flag |= NNONCACHE;
708 		}
709 		np->n_directio_opens++;
710 	}
711 
712 	/* If opened for writing via NFSv4.1 or later, mark that for pNFS. */
713 	if (NFSHASPNFS(VFSTONFS(vp->v_mount)) && (fmode & FWRITE) != 0)
714 		np->n_flag |= NWRITEOPENED;
715 
716 	/*
717 	 * If this is an open for writing, capture a reference to the
718 	 * credentials, so they can be used by ncl_putpages(). Using
719 	 * these write credentials is preferable to the credentials of
720 	 * whatever thread happens to be doing the VOP_PUTPAGES() since
721 	 * the write RPCs are less likely to fail with EACCES.
722 	 */
723 	if ((fmode & FWRITE) != 0) {
724 		cred = np->n_writecred;
725 		np->n_writecred = crhold(ap->a_cred);
726 	} else
727 		cred = NULL;
728 	NFSUNLOCKNODE(np);
729 
730 	if (cred != NULL)
731 		crfree(cred);
732 	vnode_create_vobject(vp, vattr.va_size, ap->a_td);
733 
734 	/*
735 	 * If the text file has been mmap'd, flush any dirty pages to the
736 	 * buffer cache and then...
737 	 * Make sure all writes are pushed to the NFS server.  If this is not
738 	 * done, the modify time of the file can change while the text
739 	 * file is being executed.  This will cause the process that is
740 	 * executing the text file to be terminated.
741 	 */
742 	if (vp->v_writecount <= -1) {
743 		if ((obj = vp->v_object) != NULL &&
744 		    vm_object_mightbedirty(obj)) {
745 			VM_OBJECT_WLOCK(obj);
746 			vm_object_page_clean(obj, 0, 0, OBJPC_SYNC);
747 			VM_OBJECT_WUNLOCK(obj);
748 		}
749 
750 		/* Now, flush the buffer cache. */
751 		ncl_flush(vp, MNT_WAIT, curthread, 0, 0);
752 
753 		/* And, finally, make sure that n_mtime is up to date. */
754 		np = VTONFS(vp);
755 		NFSLOCKNODE(np);
756 		np->n_mtime = np->n_vattr.na_mtime;
757 		NFSUNLOCKNODE(np);
758 	}
759 	return (0);
760 }
761 
762 /*
763  * nfs close vnode op
764  * What an NFS client should do upon close after writing is a debatable issue.
765  * Most NFS clients push delayed writes to the server upon close, basically for
766  * two reasons:
767  * 1 - So that any write errors may be reported back to the client process
768  *     doing the close system call. By far the two most likely errors are
769  *     NFSERR_NOSPC and NFSERR_DQUOT to indicate space allocation failure.
770  * 2 - To put a worst case upper bound on cache inconsistency between
771  *     multiple clients for the file.
772  * There is also a consistency problem for Version 2 of the protocol w.r.t.
773  * not being able to tell if other clients are writing a file concurrently,
774  * since there is no way of knowing if the changed modify time in the reply
775  * is only due to the write for this client.
776  * (NFS Version 3 provides weak cache consistency data in the reply that
777  *  should be sufficient to detect and handle this case.)
778  *
779  * The current code does the following:
780  * for NFS Version 2 - play it safe and flush/invalidate all dirty buffers
781  * for NFS Version 3 - flush dirty buffers to the server but don't invalidate
782  *                     or commit them (this satisfies 1 and 2 except for the
783  *                     case where the server crashes after this close but
784  *                     before the commit RPC, which is felt to be "good
785  *                     enough". Changing the last argument to ncl_flush() to
786  *                     a 1 would force a commit operation, if it is felt a
787  *                     commit is necessary now.
788  * for NFS Version 4 - flush the dirty buffers and commit them, if
789  *		       nfscl_mustflush() says this is necessary.
790  *                     It is necessary if there is no write delegation held,
791  *                     in order to satisfy open/close coherency.
792  *                     If the file isn't cached on local stable storage,
793  *                     it may be necessary in order to detect "out of space"
794  *                     errors from the server, if the write delegation
795  *                     issued by the server doesn't allow the file to grow.
796  */
797 /* ARGSUSED */
798 static int
799 nfs_close(struct vop_close_args *ap)
800 {
801 	struct vnode *vp = ap->a_vp;
802 	struct nfsnode *np = VTONFS(vp);
803 	struct nfsvattr nfsva;
804 	struct ucred *cred;
805 	int error = 0, ret, localcred = 0;
806 	int fmode = ap->a_fflag;
807 
808 	if (NFSCL_FORCEDISM(vp->v_mount))
809 		return (0);
810 	/*
811 	 * During shutdown, a_cred isn't valid, so just use root.
812 	 */
813 	if (ap->a_cred == NOCRED) {
814 		cred = newnfs_getcred();
815 		localcred = 1;
816 	} else {
817 		cred = ap->a_cred;
818 	}
819 	if (vp->v_type == VREG) {
820 	    /*
821 	     * Examine and clean dirty pages, regardless of NMODIFIED.
822 	     * This closes a major hole in close-to-open consistency.
823 	     * We want to push out all dirty pages (and buffers) on
824 	     * close, regardless of whether they were dirtied by
825 	     * mmap'ed writes or via write().
826 	     */
827 	    if (nfs_clean_pages_on_close && vp->v_object) {
828 		VM_OBJECT_WLOCK(vp->v_object);
829 		vm_object_page_clean(vp->v_object, 0, 0, 0);
830 		VM_OBJECT_WUNLOCK(vp->v_object);
831 	    }
832 	    NFSLOCKNODE(np);
833 	    if (np->n_flag & NMODIFIED) {
834 		NFSUNLOCKNODE(np);
835 		if (NFS_ISV3(vp)) {
836 		    /*
837 		     * Under NFSv3 we have dirty buffers to dispose of.  We
838 		     * must flush them to the NFS server.  We have the option
839 		     * of waiting all the way through the commit rpc or just
840 		     * waiting for the initial write.  The default is to only
841 		     * wait through the initial write so the data is in the
842 		     * server's cache, which is roughly similar to the state
843 		     * a standard disk subsystem leaves the file in on close().
844 		     *
845 		     * We cannot clear the NMODIFIED bit in np->n_flag due to
846 		     * potential races with other processes, and certainly
847 		     * cannot clear it if we don't commit.
848 		     * These races occur when there is no longer the old
849 		     * traditional vnode locking implemented for Vnode Ops.
850 		     */
851 		    int cm = newnfs_commit_on_close ? 1 : 0;
852 		    error = ncl_flush(vp, MNT_WAIT, ap->a_td, cm, 0);
853 		    /* np->n_flag &= ~NMODIFIED; */
854 		} else if (NFS_ISV4(vp)) {
855 			if (nfscl_mustflush(vp) != 0) {
856 				int cm = newnfs_commit_on_close ? 1 : 0;
857 				error = ncl_flush(vp, MNT_WAIT, ap->a_td,
858 				    cm, 0);
859 				/*
860 				 * as above w.r.t races when clearing
861 				 * NMODIFIED.
862 				 * np->n_flag &= ~NMODIFIED;
863 				 */
864 			}
865 		} else {
866 			error = ncl_vinvalbuf(vp, V_SAVE, ap->a_td, 1);
867 		}
868 		NFSLOCKNODE(np);
869 	    }
870  	    /*
871  	     * Invalidate the attribute cache in all cases.
872  	     * An open is going to fetch fresh attrs any way, other procs
873  	     * on this node that have file open will be forced to do an
874  	     * otw attr fetch, but this is safe.
875 	     * --> A user found that their RPC count dropped by 20% when
876 	     *     this was commented out and I can't see any requirement
877 	     *     for it, so I've disabled it when negative lookups are
878 	     *     enabled. (What does this have to do with negative lookup
879 	     *     caching? Well nothing, except it was reported by the
880 	     *     same user that needed negative lookup caching and I wanted
881 	     *     there to be a way to disable it to see if it
882 	     *     is the cause of some caching/coherency issue that might
883 	     *     crop up.)
884  	     */
885 	    if (VFSTONFS(vp->v_mount)->nm_negnametimeo == 0) {
886 		    np->n_attrstamp = 0;
887 		    KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(vp);
888 	    }
889 	    if (np->n_flag & NWRITEERR) {
890 		np->n_flag &= ~NWRITEERR;
891 		error = np->n_error;
892 	    }
893 	    NFSUNLOCKNODE(np);
894 	}
895 
896 	if (NFS_ISV4(vp)) {
897 		/*
898 		 * Get attributes so "change" is up to date.
899 		 */
900 		if (error == 0 && nfscl_mustflush(vp) != 0 &&
901 		    vp->v_type == VREG &&
902 		    (VFSTONFS(vp->v_mount)->nm_flag & NFSMNT_NOCTO) == 0) {
903 			ret = nfsrpc_getattr(vp, cred, ap->a_td, &nfsva,
904 			    NULL);
905 			if (!ret) {
906 				np->n_change = nfsva.na_filerev;
907 				(void) nfscl_loadattrcache(&vp, &nfsva, NULL,
908 				    NULL, 0, 0);
909 			}
910 		}
911 
912 		/*
913 		 * and do the close.
914 		 */
915 		ret = nfsrpc_close(vp, 0, ap->a_td);
916 		if (!error && ret)
917 			error = ret;
918 		if (error)
919 			error = nfscl_maperr(ap->a_td, error, (uid_t)0,
920 			    (gid_t)0);
921 	}
922 	if (newnfs_directio_enable)
923 		KASSERT((np->n_directio_asyncwr == 0),
924 			("nfs_close: dirty unflushed (%d) directio buffers\n",
925 			 np->n_directio_asyncwr));
926 	if (newnfs_directio_enable && (fmode & O_DIRECT) && (vp->v_type == VREG)) {
927 		NFSLOCKNODE(np);
928 		KASSERT((np->n_directio_opens > 0),
929 			("nfs_close: unexpectedly value (0) of n_directio_opens\n"));
930 		np->n_directio_opens--;
931 		if (np->n_directio_opens == 0)
932 			np->n_flag &= ~NNONCACHE;
933 		NFSUNLOCKNODE(np);
934 	}
935 	if (localcred)
936 		NFSFREECRED(cred);
937 	return (error);
938 }
939 
940 /*
941  * nfs getattr call from vfs.
942  */
943 static int
944 nfs_getattr(struct vop_getattr_args *ap)
945 {
946 	struct vnode *vp = ap->a_vp;
947 	struct thread *td = curthread;	/* XXX */
948 	struct nfsnode *np = VTONFS(vp);
949 	int error = 0;
950 	struct nfsvattr nfsva;
951 	struct vattr *vap = ap->a_vap;
952 	struct vattr vattr;
953 
954 	/*
955 	 * Update local times for special files.
956 	 */
957 	NFSLOCKNODE(np);
958 	if (np->n_flag & (NACC | NUPD))
959 		np->n_flag |= NCHG;
960 	NFSUNLOCKNODE(np);
961 	/*
962 	 * First look in the cache.
963 	 */
964 	if (ncl_getattrcache(vp, &vattr) == 0) {
965 		ncl_copy_vattr(vap, &vattr);
966 
967 		/*
968 		 * Get the local modify time for the case of a write
969 		 * delegation.
970 		 */
971 		nfscl_deleggetmodtime(vp, &vap->va_mtime);
972 		return (0);
973 	}
974 
975 	if (NFS_ISV34(vp) && nfs_prime_access_cache &&
976 	    nfsaccess_cache_timeout > 0) {
977 		NFSINCRGLOBAL(nfsstatsv1.accesscache_misses);
978 		nfs34_access_otw(vp, NFSACCESS_ALL, td, ap->a_cred, NULL);
979 		if (ncl_getattrcache(vp, ap->a_vap) == 0) {
980 			nfscl_deleggetmodtime(vp, &ap->a_vap->va_mtime);
981 			return (0);
982 		}
983 	}
984 	error = nfsrpc_getattr(vp, ap->a_cred, td, &nfsva, NULL);
985 	if (!error)
986 		error = nfscl_loadattrcache(&vp, &nfsva, vap, NULL, 0, 0);
987 	if (!error) {
988 		/*
989 		 * Get the local modify time for the case of a write
990 		 * delegation.
991 		 */
992 		nfscl_deleggetmodtime(vp, &vap->va_mtime);
993 	} else if (NFS_ISV4(vp)) {
994 		error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0);
995 	}
996 	return (error);
997 }
998 
999 /*
1000  * nfs setattr call.
1001  */
1002 static int
1003 nfs_setattr(struct vop_setattr_args *ap)
1004 {
1005 	struct vnode *vp = ap->a_vp;
1006 	struct nfsnode *np = VTONFS(vp);
1007 	struct thread *td = curthread;	/* XXX */
1008 	struct vattr *vap = ap->a_vap;
1009 	int error = 0;
1010 	u_quad_t tsize;
1011 
1012 #ifndef nolint
1013 	tsize = (u_quad_t)0;
1014 #endif
1015 
1016 	/*
1017 	 * Setting of flags and marking of atimes are not supported.
1018 	 */
1019 	if (vap->va_flags != VNOVAL)
1020 		return (EOPNOTSUPP);
1021 
1022 	/*
1023 	 * Disallow write attempts if the filesystem is mounted read-only.
1024 	 */
1025   	if ((vap->va_flags != VNOVAL || vap->va_uid != (uid_t)VNOVAL ||
1026 	    vap->va_gid != (gid_t)VNOVAL || vap->va_atime.tv_sec != VNOVAL ||
1027 	    vap->va_mtime.tv_sec != VNOVAL ||
1028 	    vap->va_birthtime.tv_sec != VNOVAL ||
1029 	    vap->va_mode != (mode_t)VNOVAL) &&
1030 	    (vp->v_mount->mnt_flag & MNT_RDONLY))
1031 		return (EROFS);
1032 	if (vap->va_size != VNOVAL) {
1033  		switch (vp->v_type) {
1034  		case VDIR:
1035  			return (EISDIR);
1036  		case VCHR:
1037  		case VBLK:
1038  		case VSOCK:
1039  		case VFIFO:
1040 			if (vap->va_mtime.tv_sec == VNOVAL &&
1041 			    vap->va_atime.tv_sec == VNOVAL &&
1042 			    vap->va_birthtime.tv_sec == VNOVAL &&
1043 			    vap->va_mode == (mode_t)VNOVAL &&
1044 			    vap->va_uid == (uid_t)VNOVAL &&
1045 			    vap->va_gid == (gid_t)VNOVAL)
1046 				return (0);
1047  			vap->va_size = VNOVAL;
1048  			break;
1049  		default:
1050 			/*
1051 			 * Disallow write attempts if the filesystem is
1052 			 * mounted read-only.
1053 			 */
1054 			if (vp->v_mount->mnt_flag & MNT_RDONLY)
1055 				return (EROFS);
1056 			/*
1057 			 *  We run vnode_pager_setsize() early (why?),
1058 			 * we must set np->n_size now to avoid vinvalbuf
1059 			 * V_SAVE races that might setsize a lower
1060 			 * value.
1061 			 */
1062 			NFSLOCKNODE(np);
1063 			tsize = np->n_size;
1064 			NFSUNLOCKNODE(np);
1065 			error = ncl_meta_setsize(vp, td, vap->va_size);
1066 			NFSLOCKNODE(np);
1067  			if (np->n_flag & NMODIFIED) {
1068 			    tsize = np->n_size;
1069 			    NFSUNLOCKNODE(np);
1070 			    error = ncl_vinvalbuf(vp, vap->va_size == 0 ?
1071 			        0 : V_SAVE, td, 1);
1072 			    if (error != 0) {
1073 				    vnode_pager_setsize(vp, tsize);
1074 				    return (error);
1075 			    }
1076 			    /*
1077 			     * Call nfscl_delegmodtime() to set the modify time
1078 			     * locally, as required.
1079 			     */
1080 			    nfscl_delegmodtime(vp);
1081  			} else
1082 			    NFSUNLOCKNODE(np);
1083 			/*
1084 			 * np->n_size has already been set to vap->va_size
1085 			 * in ncl_meta_setsize(). We must set it again since
1086 			 * nfs_loadattrcache() could be called through
1087 			 * ncl_meta_setsize() and could modify np->n_size.
1088 			 */
1089 			NFSLOCKNODE(np);
1090  			np->n_vattr.na_size = np->n_size = vap->va_size;
1091 			NFSUNLOCKNODE(np);
1092   		}
1093   	} else {
1094 		NFSLOCKNODE(np);
1095 		if ((vap->va_mtime.tv_sec != VNOVAL || vap->va_atime.tv_sec != VNOVAL) &&
1096 		    (np->n_flag & NMODIFIED) && vp->v_type == VREG) {
1097 			NFSUNLOCKNODE(np);
1098 			error = ncl_vinvalbuf(vp, V_SAVE, td, 1);
1099 			if (error == EINTR || error == EIO)
1100 				return (error);
1101 		} else
1102 			NFSUNLOCKNODE(np);
1103 	}
1104 	error = nfs_setattrrpc(vp, vap, ap->a_cred, td);
1105 	if (error && vap->va_size != VNOVAL) {
1106 		NFSLOCKNODE(np);
1107 		np->n_size = np->n_vattr.na_size = tsize;
1108 		vnode_pager_setsize(vp, tsize);
1109 		NFSUNLOCKNODE(np);
1110 	}
1111 	return (error);
1112 }
1113 
1114 /*
1115  * Do an nfs setattr rpc.
1116  */
1117 static int
1118 nfs_setattrrpc(struct vnode *vp, struct vattr *vap, struct ucred *cred,
1119     struct thread *td)
1120 {
1121 	struct nfsnode *np = VTONFS(vp);
1122 	int error, ret, attrflag, i;
1123 	struct nfsvattr nfsva;
1124 
1125 	if (NFS_ISV34(vp)) {
1126 		NFSLOCKNODE(np);
1127 		for (i = 0; i < NFS_ACCESSCACHESIZE; i++)
1128 			np->n_accesscache[i].stamp = 0;
1129 		np->n_flag |= NDELEGMOD;
1130 		NFSUNLOCKNODE(np);
1131 		KDTRACE_NFS_ACCESSCACHE_FLUSH_DONE(vp);
1132 	}
1133 	error = nfsrpc_setattr(vp, vap, NULL, cred, td, &nfsva, &attrflag,
1134 	    NULL);
1135 	if (attrflag) {
1136 		ret = nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0, 1);
1137 		if (ret && !error)
1138 			error = ret;
1139 	}
1140 	if (error && NFS_ISV4(vp))
1141 		error = nfscl_maperr(td, error, vap->va_uid, vap->va_gid);
1142 	return (error);
1143 }
1144 
1145 /*
1146  * nfs lookup call, one step at a time...
1147  * First look in cache
1148  * If not found, unlock the directory nfsnode and do the rpc
1149  */
1150 static int
1151 nfs_lookup(struct vop_lookup_args *ap)
1152 {
1153 	struct componentname *cnp = ap->a_cnp;
1154 	struct vnode *dvp = ap->a_dvp;
1155 	struct vnode **vpp = ap->a_vpp;
1156 	struct mount *mp = dvp->v_mount;
1157 	int flags = cnp->cn_flags;
1158 	struct vnode *newvp;
1159 	struct nfsmount *nmp;
1160 	struct nfsnode *np, *newnp;
1161 	int error = 0, attrflag, dattrflag, ltype, ncticks;
1162 	struct thread *td = cnp->cn_thread;
1163 	struct nfsfh *nfhp;
1164 	struct nfsvattr dnfsva, nfsva;
1165 	struct vattr vattr;
1166 	struct timespec nctime;
1167 	uint32_t openmode;
1168 
1169 	*vpp = NULLVP;
1170 	if ((flags & ISLASTCN) && (mp->mnt_flag & MNT_RDONLY) &&
1171 	    (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME))
1172 		return (EROFS);
1173 	if (dvp->v_type != VDIR)
1174 		return (ENOTDIR);
1175 	nmp = VFSTONFS(mp);
1176 	np = VTONFS(dvp);
1177 
1178 	/* For NFSv4, wait until any remove is done. */
1179 	NFSLOCKNODE(np);
1180 	while (NFSHASNFSV4(nmp) && (np->n_flag & NREMOVEINPROG)) {
1181 		np->n_flag |= NREMOVEWANT;
1182 		(void) msleep((caddr_t)np, &np->n_mtx, PZERO, "nfslkup", 0);
1183 	}
1184 	NFSUNLOCKNODE(np);
1185 
1186 	error = vn_dir_check_exec(dvp, cnp);
1187 	if (error != 0)
1188 		return (error);
1189 	error = cache_lookup(dvp, vpp, cnp, &nctime, &ncticks);
1190 	if (error > 0 && error != ENOENT)
1191 		return (error);
1192 	if (error == -1) {
1193 		/*
1194 		 * Lookups of "." are special and always return the
1195 		 * current directory.  cache_lookup() already handles
1196 		 * associated locking bookkeeping, etc.
1197 		 */
1198 		if (cnp->cn_namelen == 1 && cnp->cn_nameptr[0] == '.') {
1199 			/* XXX: Is this really correct? */
1200 			if (cnp->cn_nameiop != LOOKUP &&
1201 			    (flags & ISLASTCN))
1202 				cnp->cn_flags |= SAVENAME;
1203 			return (0);
1204 		}
1205 
1206 		/*
1207 		 * We only accept a positive hit in the cache if the
1208 		 * change time of the file matches our cached copy.
1209 		 * Otherwise, we discard the cache entry and fallback
1210 		 * to doing a lookup RPC.  We also only trust cache
1211 		 * entries for less than nm_nametimeo seconds.
1212 		 *
1213 		 * To better handle stale file handles and attributes,
1214 		 * clear the attribute cache of this node if it is a
1215 		 * leaf component, part of an open() call, and not
1216 		 * locally modified before fetching the attributes.
1217 		 * This should allow stale file handles to be detected
1218 		 * here where we can fall back to a LOOKUP RPC to
1219 		 * recover rather than having nfs_open() detect the
1220 		 * stale file handle and failing open(2) with ESTALE.
1221 		 */
1222 		newvp = *vpp;
1223 		newnp = VTONFS(newvp);
1224 		if (!(nmp->nm_flag & NFSMNT_NOCTO) &&
1225 		    (flags & (ISLASTCN | ISOPEN)) == (ISLASTCN | ISOPEN) &&
1226 		    !(newnp->n_flag & NMODIFIED)) {
1227 			NFSLOCKNODE(newnp);
1228 			newnp->n_attrstamp = 0;
1229 			KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(newvp);
1230 			NFSUNLOCKNODE(newnp);
1231 		}
1232 		if (nfscl_nodeleg(newvp, 0) == 0 ||
1233 		    ((u_int)(ticks - ncticks) < (nmp->nm_nametimeo * hz) &&
1234 		    VOP_GETATTR(newvp, &vattr, cnp->cn_cred) == 0 &&
1235 		    timespeccmp(&vattr.va_ctime, &nctime, ==))) {
1236 			NFSINCRGLOBAL(nfsstatsv1.lookupcache_hits);
1237 			if (cnp->cn_nameiop != LOOKUP &&
1238 			    (flags & ISLASTCN))
1239 				cnp->cn_flags |= SAVENAME;
1240 			return (0);
1241 		}
1242 		cache_purge(newvp);
1243 		if (dvp != newvp)
1244 			vput(newvp);
1245 		else
1246 			vrele(newvp);
1247 		*vpp = NULLVP;
1248 	} else if (error == ENOENT) {
1249 		if (VN_IS_DOOMED(dvp))
1250 			return (ENOENT);
1251 		/*
1252 		 * We only accept a negative hit in the cache if the
1253 		 * modification time of the parent directory matches
1254 		 * the cached copy in the name cache entry.
1255 		 * Otherwise, we discard all of the negative cache
1256 		 * entries for this directory.  We also only trust
1257 		 * negative cache entries for up to nm_negnametimeo
1258 		 * seconds.
1259 		 */
1260 		if ((u_int)(ticks - ncticks) < (nmp->nm_negnametimeo * hz) &&
1261 		    VOP_GETATTR(dvp, &vattr, cnp->cn_cred) == 0 &&
1262 		    timespeccmp(&vattr.va_mtime, &nctime, ==)) {
1263 			NFSINCRGLOBAL(nfsstatsv1.lookupcache_hits);
1264 			return (ENOENT);
1265 		}
1266 		cache_purge_negative(dvp);
1267 	}
1268 
1269 	/*
1270 	 * If this an NFSv4.1/4.2 mount using the "oneopenown" mount
1271 	 * option, it is possible to do the Open operation in the same
1272 	 * compound as Lookup, so long as delegations are not being
1273 	 * issued.  This saves doing a separate RPC for Open.
1274 	 */
1275 	openmode = 0;
1276 	NFSLOCKMNT(nmp);
1277 	if (NFSHASNFSV4N(nmp) && NFSHASONEOPENOWN(nmp) &&
1278 	    (nmp->nm_privflag & NFSMNTP_DELEGISSUED) == 0 &&
1279 	    (!NFSMNT_RDONLY(mp) || (flags & OPENWRITE) == 0) &&
1280 	    (flags & (ISLASTCN | ISOPEN)) == (ISLASTCN | ISOPEN)) {
1281 		if ((flags & OPENREAD) != 0)
1282 			openmode |= NFSV4OPEN_ACCESSREAD;
1283 		if ((flags & OPENWRITE) != 0)
1284 			openmode |= NFSV4OPEN_ACCESSWRITE;
1285 	}
1286 	NFSUNLOCKMNT(nmp);
1287 
1288 	newvp = NULLVP;
1289 	NFSINCRGLOBAL(nfsstatsv1.lookupcache_misses);
1290 	error = nfsrpc_lookup(dvp, cnp->cn_nameptr, cnp->cn_namelen,
1291 	    cnp->cn_cred, td, &dnfsva, &nfsva, &nfhp, &attrflag, &dattrflag,
1292 	    NULL, openmode);
1293 	if (dattrflag)
1294 		(void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, NULL, 0, 1);
1295 	if (error) {
1296 		if (newvp != NULLVP) {
1297 			vput(newvp);
1298 			*vpp = NULLVP;
1299 		}
1300 
1301 		if (error != ENOENT) {
1302 			if (NFS_ISV4(dvp))
1303 				error = nfscl_maperr(td, error, (uid_t)0,
1304 				    (gid_t)0);
1305 			return (error);
1306 		}
1307 
1308 		/* The requested file was not found. */
1309 		if ((cnp->cn_nameiop == CREATE || cnp->cn_nameiop == RENAME) &&
1310 		    (flags & ISLASTCN)) {
1311 			/*
1312 			 * XXX: UFS does a full VOP_ACCESS(dvp,
1313 			 * VWRITE) here instead of just checking
1314 			 * MNT_RDONLY.
1315 			 */
1316 			if (mp->mnt_flag & MNT_RDONLY)
1317 				return (EROFS);
1318 			cnp->cn_flags |= SAVENAME;
1319 			return (EJUSTRETURN);
1320 		}
1321 
1322 		if ((cnp->cn_flags & MAKEENTRY) != 0 && dattrflag) {
1323 			/*
1324 			 * Cache the modification time of the parent
1325 			 * directory from the post-op attributes in
1326 			 * the name cache entry.  The negative cache
1327 			 * entry will be ignored once the directory
1328 			 * has changed.  Don't bother adding the entry
1329 			 * if the directory has already changed.
1330 			 */
1331 			NFSLOCKNODE(np);
1332 			if (timespeccmp(&np->n_vattr.na_mtime,
1333 			    &dnfsva.na_mtime, ==)) {
1334 				NFSUNLOCKNODE(np);
1335 				cache_enter_time(dvp, NULL, cnp,
1336 				    &dnfsva.na_mtime, NULL);
1337 			} else
1338 				NFSUNLOCKNODE(np);
1339 		}
1340 		return (ENOENT);
1341 	}
1342 
1343 	/*
1344 	 * Handle RENAME case...
1345 	 */
1346 	if (cnp->cn_nameiop == RENAME && (flags & ISLASTCN)) {
1347 		if (NFS_CMPFH(np, nfhp->nfh_fh, nfhp->nfh_len)) {
1348 			free(nfhp, M_NFSFH);
1349 			return (EISDIR);
1350 		}
1351 		error = nfscl_nget(mp, dvp, nfhp, cnp, td, &np, NULL,
1352 		    LK_EXCLUSIVE);
1353 		if (error)
1354 			return (error);
1355 		newvp = NFSTOV(np);
1356 		if (attrflag)
1357 			(void) nfscl_loadattrcache(&newvp, &nfsva, NULL, NULL,
1358 			    0, 1);
1359 		*vpp = newvp;
1360 		cnp->cn_flags |= SAVENAME;
1361 		return (0);
1362 	}
1363 
1364 	if (flags & ISDOTDOT) {
1365 		ltype = NFSVOPISLOCKED(dvp);
1366 		error = vfs_busy(mp, MBF_NOWAIT);
1367 		if (error != 0) {
1368 			vfs_ref(mp);
1369 			NFSVOPUNLOCK(dvp);
1370 			error = vfs_busy(mp, 0);
1371 			NFSVOPLOCK(dvp, ltype | LK_RETRY);
1372 			vfs_rel(mp);
1373 			if (error == 0 && VN_IS_DOOMED(dvp)) {
1374 				vfs_unbusy(mp);
1375 				error = ENOENT;
1376 			}
1377 			if (error != 0)
1378 				return (error);
1379 		}
1380 		NFSVOPUNLOCK(dvp);
1381 		error = nfscl_nget(mp, dvp, nfhp, cnp, td, &np, NULL,
1382 		    cnp->cn_lkflags);
1383 		if (error == 0)
1384 			newvp = NFSTOV(np);
1385 		vfs_unbusy(mp);
1386 		if (newvp != dvp)
1387 			NFSVOPLOCK(dvp, ltype | LK_RETRY);
1388 		if (VN_IS_DOOMED(dvp)) {
1389 			if (error == 0) {
1390 				if (newvp == dvp)
1391 					vrele(newvp);
1392 				else
1393 					vput(newvp);
1394 			}
1395 			error = ENOENT;
1396 		}
1397 		if (error != 0)
1398 			return (error);
1399 		if (attrflag)
1400 			(void) nfscl_loadattrcache(&newvp, &nfsva, NULL, NULL,
1401 			    0, 1);
1402 	} else if (NFS_CMPFH(np, nfhp->nfh_fh, nfhp->nfh_len)) {
1403 		free(nfhp, M_NFSFH);
1404 		VREF(dvp);
1405 		newvp = dvp;
1406 		if (attrflag)
1407 			(void) nfscl_loadattrcache(&newvp, &nfsva, NULL, NULL,
1408 			    0, 1);
1409 	} else {
1410 		error = nfscl_nget(mp, dvp, nfhp, cnp, td, &np, NULL,
1411 		    cnp->cn_lkflags);
1412 		if (error)
1413 			return (error);
1414 		newvp = NFSTOV(np);
1415 		if (attrflag)
1416 			(void) nfscl_loadattrcache(&newvp, &nfsva, NULL, NULL,
1417 			    0, 1);
1418 		else if ((flags & (ISLASTCN | ISOPEN)) == (ISLASTCN | ISOPEN) &&
1419 		    !(np->n_flag & NMODIFIED)) {
1420 			/*
1421 			 * Flush the attribute cache when opening a
1422 			 * leaf node to ensure that fresh attributes
1423 			 * are fetched in nfs_open() since we did not
1424 			 * fetch attributes from the LOOKUP reply.
1425 			 */
1426 			NFSLOCKNODE(np);
1427 			np->n_attrstamp = 0;
1428 			KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(newvp);
1429 			NFSUNLOCKNODE(np);
1430 		}
1431 	}
1432 	if (cnp->cn_nameiop != LOOKUP && (flags & ISLASTCN))
1433 		cnp->cn_flags |= SAVENAME;
1434 	if ((cnp->cn_flags & MAKEENTRY) && dvp != newvp &&
1435 	    (cnp->cn_nameiop != DELETE || !(flags & ISLASTCN)) &&
1436 	    attrflag != 0 && (newvp->v_type != VDIR || dattrflag != 0))
1437 		cache_enter_time(dvp, newvp, cnp, &nfsva.na_ctime,
1438 		    newvp->v_type != VDIR ? NULL : &dnfsva.na_ctime);
1439 	*vpp = newvp;
1440 	return (0);
1441 }
1442 
1443 /*
1444  * nfs read call.
1445  * Just call ncl_bioread() to do the work.
1446  */
1447 static int
1448 nfs_read(struct vop_read_args *ap)
1449 {
1450 	struct vnode *vp = ap->a_vp;
1451 
1452 	switch (vp->v_type) {
1453 	case VREG:
1454 		return (ncl_bioread(vp, ap->a_uio, ap->a_ioflag, ap->a_cred));
1455 	case VDIR:
1456 		return (EISDIR);
1457 	default:
1458 		return (EOPNOTSUPP);
1459 	}
1460 }
1461 
1462 /*
1463  * nfs readlink call
1464  */
1465 static int
1466 nfs_readlink(struct vop_readlink_args *ap)
1467 {
1468 	struct vnode *vp = ap->a_vp;
1469 
1470 	if (vp->v_type != VLNK)
1471 		return (EINVAL);
1472 	return (ncl_bioread(vp, ap->a_uio, 0, ap->a_cred));
1473 }
1474 
1475 /*
1476  * Do a readlink rpc.
1477  * Called by ncl_doio() from below the buffer cache.
1478  */
1479 int
1480 ncl_readlinkrpc(struct vnode *vp, struct uio *uiop, struct ucred *cred)
1481 {
1482 	int error, ret, attrflag;
1483 	struct nfsvattr nfsva;
1484 
1485 	error = nfsrpc_readlink(vp, uiop, cred, uiop->uio_td, &nfsva,
1486 	    &attrflag, NULL);
1487 	if (attrflag) {
1488 		ret = nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0, 1);
1489 		if (ret && !error)
1490 			error = ret;
1491 	}
1492 	if (error && NFS_ISV4(vp))
1493 		error = nfscl_maperr(uiop->uio_td, error, (uid_t)0, (gid_t)0);
1494 	return (error);
1495 }
1496 
1497 /*
1498  * nfs read rpc call
1499  * Ditto above
1500  */
1501 int
1502 ncl_readrpc(struct vnode *vp, struct uio *uiop, struct ucred *cred)
1503 {
1504 	int error, ret, attrflag;
1505 	struct nfsvattr nfsva;
1506 	struct nfsmount *nmp;
1507 
1508 	nmp = VFSTONFS(vp->v_mount);
1509 	error = EIO;
1510 	attrflag = 0;
1511 	if (NFSHASPNFS(nmp))
1512 		error = nfscl_doiods(vp, uiop, NULL, NULL,
1513 		    NFSV4OPEN_ACCESSREAD, 0, cred, uiop->uio_td);
1514 	NFSCL_DEBUG(4, "readrpc: aft doiods=%d\n", error);
1515 	if (error != 0)
1516 		error = nfsrpc_read(vp, uiop, cred, uiop->uio_td, &nfsva,
1517 		    &attrflag, NULL);
1518 	if (attrflag) {
1519 		ret = nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0, 1);
1520 		if (ret && !error)
1521 			error = ret;
1522 	}
1523 	if (error && NFS_ISV4(vp))
1524 		error = nfscl_maperr(uiop->uio_td, error, (uid_t)0, (gid_t)0);
1525 	return (error);
1526 }
1527 
1528 /*
1529  * nfs write call
1530  */
1531 int
1532 ncl_writerpc(struct vnode *vp, struct uio *uiop, struct ucred *cred,
1533     int *iomode, int *must_commit, int called_from_strategy)
1534 {
1535 	struct nfsvattr nfsva;
1536 	int error, attrflag, ret;
1537 	struct nfsmount *nmp;
1538 
1539 	nmp = VFSTONFS(vp->v_mount);
1540 	error = EIO;
1541 	attrflag = 0;
1542 	if (NFSHASPNFS(nmp))
1543 		error = nfscl_doiods(vp, uiop, iomode, must_commit,
1544 		    NFSV4OPEN_ACCESSWRITE, 0, cred, uiop->uio_td);
1545 	NFSCL_DEBUG(4, "writerpc: aft doiods=%d\n", error);
1546 	if (error != 0)
1547 		error = nfsrpc_write(vp, uiop, iomode, must_commit, cred,
1548 		    uiop->uio_td, &nfsva, &attrflag, NULL,
1549 		    called_from_strategy);
1550 	if (attrflag) {
1551 		if (VTONFS(vp)->n_flag & ND_NFSV4)
1552 			ret = nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 1,
1553 			    1);
1554 		else
1555 			ret = nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0,
1556 			    1);
1557 		if (ret && !error)
1558 			error = ret;
1559 	}
1560 	if (DOINGASYNC(vp))
1561 		*iomode = NFSWRITE_FILESYNC;
1562 	if (error && NFS_ISV4(vp))
1563 		error = nfscl_maperr(uiop->uio_td, error, (uid_t)0, (gid_t)0);
1564 	return (error);
1565 }
1566 
1567 /*
1568  * nfs mknod rpc
1569  * For NFS v2 this is a kludge. Use a create rpc but with the IFMT bits of the
1570  * mode set to specify the file type and the size field for rdev.
1571  */
1572 static int
1573 nfs_mknodrpc(struct vnode *dvp, struct vnode **vpp, struct componentname *cnp,
1574     struct vattr *vap)
1575 {
1576 	struct nfsvattr nfsva, dnfsva;
1577 	struct vnode *newvp = NULL;
1578 	struct nfsnode *np = NULL, *dnp;
1579 	struct nfsfh *nfhp;
1580 	struct vattr vattr;
1581 	int error = 0, attrflag, dattrflag;
1582 	u_int32_t rdev;
1583 
1584 	if (vap->va_type == VCHR || vap->va_type == VBLK)
1585 		rdev = vap->va_rdev;
1586 	else if (vap->va_type == VFIFO || vap->va_type == VSOCK)
1587 		rdev = 0xffffffff;
1588 	else
1589 		return (EOPNOTSUPP);
1590 	if ((error = VOP_GETATTR(dvp, &vattr, cnp->cn_cred)))
1591 		return (error);
1592 	error = nfsrpc_mknod(dvp, cnp->cn_nameptr, cnp->cn_namelen, vap,
1593 	    rdev, vap->va_type, cnp->cn_cred, cnp->cn_thread, &dnfsva,
1594 	    &nfsva, &nfhp, &attrflag, &dattrflag, NULL);
1595 	if (!error) {
1596 		if (!nfhp)
1597 			(void) nfsrpc_lookup(dvp, cnp->cn_nameptr,
1598 			    cnp->cn_namelen, cnp->cn_cred, cnp->cn_thread,
1599 			    &dnfsva, &nfsva, &nfhp, &attrflag, &dattrflag,
1600 			    NULL, 0);
1601 		if (nfhp)
1602 			error = nfscl_nget(dvp->v_mount, dvp, nfhp, cnp,
1603 			    cnp->cn_thread, &np, NULL, LK_EXCLUSIVE);
1604 	}
1605 	if (dattrflag)
1606 		(void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, NULL, 0, 1);
1607 	if (!error) {
1608 		newvp = NFSTOV(np);
1609 		if (attrflag != 0) {
1610 			error = nfscl_loadattrcache(&newvp, &nfsva, NULL, NULL,
1611 			    0, 1);
1612 			if (error != 0)
1613 				vput(newvp);
1614 		}
1615 	}
1616 	if (!error) {
1617 		*vpp = newvp;
1618 	} else if (NFS_ISV4(dvp)) {
1619 		error = nfscl_maperr(cnp->cn_thread, error, vap->va_uid,
1620 		    vap->va_gid);
1621 	}
1622 	dnp = VTONFS(dvp);
1623 	NFSLOCKNODE(dnp);
1624 	dnp->n_flag |= NMODIFIED;
1625 	if (!dattrflag) {
1626 		dnp->n_attrstamp = 0;
1627 		KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(dvp);
1628 	}
1629 	NFSUNLOCKNODE(dnp);
1630 	return (error);
1631 }
1632 
1633 /*
1634  * nfs mknod vop
1635  * just call nfs_mknodrpc() to do the work.
1636  */
1637 /* ARGSUSED */
1638 static int
1639 nfs_mknod(struct vop_mknod_args *ap)
1640 {
1641 	return (nfs_mknodrpc(ap->a_dvp, ap->a_vpp, ap->a_cnp, ap->a_vap));
1642 }
1643 
1644 static struct mtx nfs_cverf_mtx;
1645 MTX_SYSINIT(nfs_cverf_mtx, &nfs_cverf_mtx, "NFS create verifier mutex",
1646     MTX_DEF);
1647 
1648 static nfsquad_t
1649 nfs_get_cverf(void)
1650 {
1651 	static nfsquad_t cverf;
1652 	nfsquad_t ret;
1653 	static int cverf_initialized = 0;
1654 
1655 	mtx_lock(&nfs_cverf_mtx);
1656 	if (cverf_initialized == 0) {
1657 		cverf.lval[0] = arc4random();
1658 		cverf.lval[1] = arc4random();
1659 		cverf_initialized = 1;
1660 	} else
1661 		cverf.qval++;
1662 	ret = cverf;
1663 	mtx_unlock(&nfs_cverf_mtx);
1664 
1665 	return (ret);
1666 }
1667 
1668 /*
1669  * nfs file create call
1670  */
1671 static int
1672 nfs_create(struct vop_create_args *ap)
1673 {
1674 	struct vnode *dvp = ap->a_dvp;
1675 	struct vattr *vap = ap->a_vap;
1676 	struct componentname *cnp = ap->a_cnp;
1677 	struct nfsnode *np = NULL, *dnp;
1678 	struct vnode *newvp = NULL;
1679 	struct nfsmount *nmp;
1680 	struct nfsvattr dnfsva, nfsva;
1681 	struct nfsfh *nfhp;
1682 	nfsquad_t cverf;
1683 	int error = 0, attrflag, dattrflag, fmode = 0;
1684 	struct vattr vattr;
1685 
1686 	/*
1687 	 * Oops, not for me..
1688 	 */
1689 	if (vap->va_type == VSOCK)
1690 		return (nfs_mknodrpc(dvp, ap->a_vpp, cnp, vap));
1691 
1692 	if ((error = VOP_GETATTR(dvp, &vattr, cnp->cn_cred)))
1693 		return (error);
1694 	if (vap->va_vaflags & VA_EXCLUSIVE)
1695 		fmode |= O_EXCL;
1696 	dnp = VTONFS(dvp);
1697 	nmp = VFSTONFS(dvp->v_mount);
1698 again:
1699 	/* For NFSv4, wait until any remove is done. */
1700 	NFSLOCKNODE(dnp);
1701 	while (NFSHASNFSV4(nmp) && (dnp->n_flag & NREMOVEINPROG)) {
1702 		dnp->n_flag |= NREMOVEWANT;
1703 		(void) msleep((caddr_t)dnp, &dnp->n_mtx, PZERO, "nfscrt", 0);
1704 	}
1705 	NFSUNLOCKNODE(dnp);
1706 
1707 	cverf = nfs_get_cverf();
1708 	error = nfsrpc_create(dvp, cnp->cn_nameptr, cnp->cn_namelen,
1709 	    vap, cverf, fmode, cnp->cn_cred, cnp->cn_thread, &dnfsva, &nfsva,
1710 	    &nfhp, &attrflag, &dattrflag, NULL);
1711 	if (!error) {
1712 		if (nfhp == NULL)
1713 			(void) nfsrpc_lookup(dvp, cnp->cn_nameptr,
1714 			    cnp->cn_namelen, cnp->cn_cred, cnp->cn_thread,
1715 			    &dnfsva, &nfsva, &nfhp, &attrflag, &dattrflag,
1716 			    NULL, 0);
1717 		if (nfhp != NULL)
1718 			error = nfscl_nget(dvp->v_mount, dvp, nfhp, cnp,
1719 			    cnp->cn_thread, &np, NULL, LK_EXCLUSIVE);
1720 	}
1721 	if (dattrflag)
1722 		(void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, NULL, 0, 1);
1723 	if (!error) {
1724 		newvp = NFSTOV(np);
1725 		if (attrflag == 0)
1726 			error = nfsrpc_getattr(newvp, cnp->cn_cred,
1727 			    cnp->cn_thread, &nfsva, NULL);
1728 		if (error == 0)
1729 			error = nfscl_loadattrcache(&newvp, &nfsva, NULL, NULL,
1730 			    0, 1);
1731 	}
1732 	if (error) {
1733 		if (newvp != NULL) {
1734 			vput(newvp);
1735 			newvp = NULL;
1736 		}
1737 		if (NFS_ISV34(dvp) && (fmode & O_EXCL) &&
1738 		    error == NFSERR_NOTSUPP) {
1739 			fmode &= ~O_EXCL;
1740 			goto again;
1741 		}
1742 	} else if (NFS_ISV34(dvp) && (fmode & O_EXCL)) {
1743 		if (nfscl_checksattr(vap, &nfsva)) {
1744 			error = nfsrpc_setattr(newvp, vap, NULL, cnp->cn_cred,
1745 			    cnp->cn_thread, &nfsva, &attrflag, NULL);
1746 			if (error && (vap->va_uid != (uid_t)VNOVAL ||
1747 			    vap->va_gid != (gid_t)VNOVAL)) {
1748 				/* try again without setting uid/gid */
1749 				vap->va_uid = (uid_t)VNOVAL;
1750 				vap->va_gid = (uid_t)VNOVAL;
1751 				error = nfsrpc_setattr(newvp, vap, NULL,
1752 				    cnp->cn_cred, cnp->cn_thread, &nfsva,
1753 				    &attrflag, NULL);
1754 			}
1755 			if (attrflag)
1756 				(void) nfscl_loadattrcache(&newvp, &nfsva, NULL,
1757 				    NULL, 0, 1);
1758 			if (error != 0)
1759 				vput(newvp);
1760 		}
1761 	}
1762 	if (!error) {
1763 		if ((cnp->cn_flags & MAKEENTRY) && attrflag) {
1764 			if (dvp != newvp)
1765 				cache_enter_time(dvp, newvp, cnp,
1766 				    &nfsva.na_ctime, NULL);
1767 			else
1768 				printf("nfs_create: bogus NFS server returned "
1769 				    "the directory as the new file object\n");
1770 		}
1771 		*ap->a_vpp = newvp;
1772 	} else if (NFS_ISV4(dvp)) {
1773 		error = nfscl_maperr(cnp->cn_thread, error, vap->va_uid,
1774 		    vap->va_gid);
1775 	}
1776 	NFSLOCKNODE(dnp);
1777 	dnp->n_flag |= NMODIFIED;
1778 	if (!dattrflag) {
1779 		dnp->n_attrstamp = 0;
1780 		KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(dvp);
1781 	}
1782 	NFSUNLOCKNODE(dnp);
1783 	return (error);
1784 }
1785 
1786 /*
1787  * nfs file remove call
1788  * To try and make nfs semantics closer to ufs semantics, a file that has
1789  * other processes using the vnode is renamed instead of removed and then
1790  * removed later on the last close.
1791  * - If v_usecount > 1
1792  *	  If a rename is not already in the works
1793  *	     call nfs_sillyrename() to set it up
1794  *     else
1795  *	  do the remove rpc
1796  */
1797 static int
1798 nfs_remove(struct vop_remove_args *ap)
1799 {
1800 	struct vnode *vp = ap->a_vp;
1801 	struct vnode *dvp = ap->a_dvp;
1802 	struct componentname *cnp = ap->a_cnp;
1803 	struct nfsnode *np = VTONFS(vp);
1804 	int error = 0;
1805 	struct vattr vattr;
1806 
1807 	KASSERT((cnp->cn_flags & HASBUF) != 0, ("nfs_remove: no name"));
1808 	KASSERT(vrefcnt(vp) > 0, ("nfs_remove: bad v_usecount"));
1809 	if (vp->v_type == VDIR)
1810 		error = EPERM;
1811 	else if (vrefcnt(vp) == 1 || (np->n_sillyrename &&
1812 	    VOP_GETATTR(vp, &vattr, cnp->cn_cred) == 0 &&
1813 	    vattr.va_nlink > 1)) {
1814 		/*
1815 		 * Purge the name cache so that the chance of a lookup for
1816 		 * the name succeeding while the remove is in progress is
1817 		 * minimized. Without node locking it can still happen, such
1818 		 * that an I/O op returns ESTALE, but since you get this if
1819 		 * another host removes the file..
1820 		 */
1821 		cache_purge(vp);
1822 		/*
1823 		 * throw away biocache buffers, mainly to avoid
1824 		 * unnecessary delayed writes later.
1825 		 */
1826 		error = ncl_vinvalbuf(vp, 0, cnp->cn_thread, 1);
1827 		if (error != EINTR && error != EIO)
1828 			/* Do the rpc */
1829 			error = nfs_removerpc(dvp, vp, cnp->cn_nameptr,
1830 			    cnp->cn_namelen, cnp->cn_cred, cnp->cn_thread);
1831 		/*
1832 		 * Kludge City: If the first reply to the remove rpc is lost..
1833 		 *   the reply to the retransmitted request will be ENOENT
1834 		 *   since the file was in fact removed
1835 		 *   Therefore, we cheat and return success.
1836 		 */
1837 		if (error == ENOENT)
1838 			error = 0;
1839 	} else if (!np->n_sillyrename)
1840 		error = nfs_sillyrename(dvp, vp, cnp);
1841 	NFSLOCKNODE(np);
1842 	np->n_attrstamp = 0;
1843 	NFSUNLOCKNODE(np);
1844 	KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(vp);
1845 	return (error);
1846 }
1847 
1848 /*
1849  * nfs file remove rpc called from nfs_inactive
1850  */
1851 int
1852 ncl_removeit(struct sillyrename *sp, struct vnode *vp)
1853 {
1854 	/*
1855 	 * Make sure that the directory vnode is still valid.
1856 	 * XXX we should lock sp->s_dvp here.
1857 	 */
1858 	if (sp->s_dvp->v_type == VBAD)
1859 		return (0);
1860 	return (nfs_removerpc(sp->s_dvp, vp, sp->s_name, sp->s_namlen,
1861 	    sp->s_cred, NULL));
1862 }
1863 
1864 /*
1865  * Nfs remove rpc, called from nfs_remove() and ncl_removeit().
1866  */
1867 static int
1868 nfs_removerpc(struct vnode *dvp, struct vnode *vp, char *name,
1869     int namelen, struct ucred *cred, struct thread *td)
1870 {
1871 	struct nfsvattr dnfsva;
1872 	struct nfsnode *dnp = VTONFS(dvp);
1873 	int error = 0, dattrflag;
1874 
1875 	NFSLOCKNODE(dnp);
1876 	dnp->n_flag |= NREMOVEINPROG;
1877 	NFSUNLOCKNODE(dnp);
1878 	error = nfsrpc_remove(dvp, name, namelen, vp, cred, td, &dnfsva,
1879 	    &dattrflag, NULL);
1880 	NFSLOCKNODE(dnp);
1881 	if ((dnp->n_flag & NREMOVEWANT)) {
1882 		dnp->n_flag &= ~(NREMOVEWANT | NREMOVEINPROG);
1883 		NFSUNLOCKNODE(dnp);
1884 		wakeup((caddr_t)dnp);
1885 	} else {
1886 		dnp->n_flag &= ~NREMOVEINPROG;
1887 		NFSUNLOCKNODE(dnp);
1888 	}
1889 	if (dattrflag)
1890 		(void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, NULL, 0, 1);
1891 	NFSLOCKNODE(dnp);
1892 	dnp->n_flag |= NMODIFIED;
1893 	if (!dattrflag) {
1894 		dnp->n_attrstamp = 0;
1895 		KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(dvp);
1896 	}
1897 	NFSUNLOCKNODE(dnp);
1898 	if (error && NFS_ISV4(dvp))
1899 		error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0);
1900 	return (error);
1901 }
1902 
1903 /*
1904  * nfs file rename call
1905  */
1906 static int
1907 nfs_rename(struct vop_rename_args *ap)
1908 {
1909 	struct vnode *fvp = ap->a_fvp;
1910 	struct vnode *tvp = ap->a_tvp;
1911 	struct vnode *fdvp = ap->a_fdvp;
1912 	struct vnode *tdvp = ap->a_tdvp;
1913 	struct componentname *tcnp = ap->a_tcnp;
1914 	struct componentname *fcnp = ap->a_fcnp;
1915 	struct nfsnode *fnp = VTONFS(ap->a_fvp);
1916 	struct nfsnode *tdnp = VTONFS(ap->a_tdvp);
1917 	struct nfsv4node *newv4 = NULL;
1918 	int error;
1919 
1920 	KASSERT((tcnp->cn_flags & HASBUF) != 0 &&
1921 	    (fcnp->cn_flags & HASBUF) != 0, ("nfs_rename: no name"));
1922 	/* Check for cross-device rename */
1923 	if ((fvp->v_mount != tdvp->v_mount) ||
1924 	    (tvp && (fvp->v_mount != tvp->v_mount))) {
1925 		error = EXDEV;
1926 		goto out;
1927 	}
1928 
1929 	if (fvp == tvp) {
1930 		printf("nfs_rename: fvp == tvp (can't happen)\n");
1931 		error = 0;
1932 		goto out;
1933 	}
1934 	if ((error = NFSVOPLOCK(fvp, LK_EXCLUSIVE)) != 0)
1935 		goto out;
1936 
1937 	/*
1938 	 * We have to flush B_DELWRI data prior to renaming
1939 	 * the file.  If we don't, the delayed-write buffers
1940 	 * can be flushed out later after the file has gone stale
1941 	 * under NFSV3.  NFSV2 does not have this problem because
1942 	 * ( as far as I can tell ) it flushes dirty buffers more
1943 	 * often.
1944 	 *
1945 	 * Skip the rename operation if the fsync fails, this can happen
1946 	 * due to the server's volume being full, when we pushed out data
1947 	 * that was written back to our cache earlier. Not checking for
1948 	 * this condition can result in potential (silent) data loss.
1949 	 */
1950 	error = VOP_FSYNC(fvp, MNT_WAIT, fcnp->cn_thread);
1951 	NFSVOPUNLOCK(fvp);
1952 	if (!error && tvp)
1953 		error = VOP_FSYNC(tvp, MNT_WAIT, tcnp->cn_thread);
1954 	if (error)
1955 		goto out;
1956 
1957 	/*
1958 	 * If the tvp exists and is in use, sillyrename it before doing the
1959 	 * rename of the new file over it.
1960 	 * XXX Can't sillyrename a directory.
1961 	 */
1962 	if (tvp && vrefcnt(tvp) > 1 && !VTONFS(tvp)->n_sillyrename &&
1963 		tvp->v_type != VDIR && !nfs_sillyrename(tdvp, tvp, tcnp)) {
1964 		vput(tvp);
1965 		tvp = NULL;
1966 	}
1967 
1968 	error = nfs_renamerpc(fdvp, fvp, fcnp->cn_nameptr, fcnp->cn_namelen,
1969 	    tdvp, tvp, tcnp->cn_nameptr, tcnp->cn_namelen, tcnp->cn_cred,
1970 	    tcnp->cn_thread);
1971 
1972 	if (error == 0 && NFS_ISV4(tdvp)) {
1973 		/*
1974 		 * For NFSv4, check to see if it is the same name and
1975 		 * replace the name, if it is different.
1976 		 */
1977 		newv4 = malloc(
1978 		    sizeof (struct nfsv4node) +
1979 		    tdnp->n_fhp->nfh_len + tcnp->cn_namelen - 1,
1980 		    M_NFSV4NODE, M_WAITOK);
1981 		NFSLOCKNODE(tdnp);
1982 		NFSLOCKNODE(fnp);
1983 		if (fnp->n_v4 != NULL && fvp->v_type == VREG &&
1984 		    (fnp->n_v4->n4_namelen != tcnp->cn_namelen ||
1985 		      NFSBCMP(tcnp->cn_nameptr, NFS4NODENAME(fnp->n_v4),
1986 		      tcnp->cn_namelen) ||
1987 		      tdnp->n_fhp->nfh_len != fnp->n_v4->n4_fhlen ||
1988 		      NFSBCMP(tdnp->n_fhp->nfh_fh, fnp->n_v4->n4_data,
1989 			tdnp->n_fhp->nfh_len))) {
1990 #ifdef notdef
1991 { char nnn[100]; int nnnl;
1992 nnnl = (tcnp->cn_namelen < 100) ? tcnp->cn_namelen : 99;
1993 bcopy(tcnp->cn_nameptr, nnn, nnnl);
1994 nnn[nnnl] = '\0';
1995 printf("ren replace=%s\n",nnn);
1996 }
1997 #endif
1998 			free(fnp->n_v4, M_NFSV4NODE);
1999 			fnp->n_v4 = newv4;
2000 			newv4 = NULL;
2001 			fnp->n_v4->n4_fhlen = tdnp->n_fhp->nfh_len;
2002 			fnp->n_v4->n4_namelen = tcnp->cn_namelen;
2003 			NFSBCOPY(tdnp->n_fhp->nfh_fh, fnp->n_v4->n4_data,
2004 			    tdnp->n_fhp->nfh_len);
2005 			NFSBCOPY(tcnp->cn_nameptr,
2006 			    NFS4NODENAME(fnp->n_v4), tcnp->cn_namelen);
2007 		}
2008 		NFSUNLOCKNODE(tdnp);
2009 		NFSUNLOCKNODE(fnp);
2010 		if (newv4 != NULL)
2011 			free(newv4, M_NFSV4NODE);
2012 	}
2013 
2014 	if (fvp->v_type == VDIR) {
2015 		if (tvp != NULL && tvp->v_type == VDIR)
2016 			cache_purge(tdvp);
2017 		cache_purge(fdvp);
2018 	}
2019 
2020 out:
2021 	if (tdvp == tvp)
2022 		vrele(tdvp);
2023 	else
2024 		vput(tdvp);
2025 	if (tvp)
2026 		vput(tvp);
2027 	vrele(fdvp);
2028 	vrele(fvp);
2029 	/*
2030 	 * Kludge: Map ENOENT => 0 assuming that it is a reply to a retry.
2031 	 */
2032 	if (error == ENOENT)
2033 		error = 0;
2034 	return (error);
2035 }
2036 
2037 /*
2038  * nfs file rename rpc called from nfs_remove() above
2039  */
2040 static int
2041 nfs_renameit(struct vnode *sdvp, struct vnode *svp, struct componentname *scnp,
2042     struct sillyrename *sp)
2043 {
2044 
2045 	return (nfs_renamerpc(sdvp, svp, scnp->cn_nameptr, scnp->cn_namelen,
2046 	    sdvp, NULL, sp->s_name, sp->s_namlen, scnp->cn_cred,
2047 	    scnp->cn_thread));
2048 }
2049 
2050 /*
2051  * Do an nfs rename rpc. Called from nfs_rename() and nfs_renameit().
2052  */
2053 static int
2054 nfs_renamerpc(struct vnode *fdvp, struct vnode *fvp, char *fnameptr,
2055     int fnamelen, struct vnode *tdvp, struct vnode *tvp, char *tnameptr,
2056     int tnamelen, struct ucred *cred, struct thread *td)
2057 {
2058 	struct nfsvattr fnfsva, tnfsva;
2059 	struct nfsnode *fdnp = VTONFS(fdvp);
2060 	struct nfsnode *tdnp = VTONFS(tdvp);
2061 	int error = 0, fattrflag, tattrflag;
2062 
2063 	error = nfsrpc_rename(fdvp, fvp, fnameptr, fnamelen, tdvp, tvp,
2064 	    tnameptr, tnamelen, cred, td, &fnfsva, &tnfsva, &fattrflag,
2065 	    &tattrflag, NULL, NULL);
2066 	NFSLOCKNODE(fdnp);
2067 	fdnp->n_flag |= NMODIFIED;
2068 	if (fattrflag != 0) {
2069 		NFSUNLOCKNODE(fdnp);
2070 		(void) nfscl_loadattrcache(&fdvp, &fnfsva, NULL, NULL, 0, 1);
2071 	} else {
2072 		fdnp->n_attrstamp = 0;
2073 		NFSUNLOCKNODE(fdnp);
2074 		KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(fdvp);
2075 	}
2076 	NFSLOCKNODE(tdnp);
2077 	tdnp->n_flag |= NMODIFIED;
2078 	if (tattrflag != 0) {
2079 		NFSUNLOCKNODE(tdnp);
2080 		(void) nfscl_loadattrcache(&tdvp, &tnfsva, NULL, NULL, 0, 1);
2081 	} else {
2082 		tdnp->n_attrstamp = 0;
2083 		NFSUNLOCKNODE(tdnp);
2084 		KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(tdvp);
2085 	}
2086 	if (error && NFS_ISV4(fdvp))
2087 		error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0);
2088 	return (error);
2089 }
2090 
2091 /*
2092  * nfs hard link create call
2093  */
2094 static int
2095 nfs_link(struct vop_link_args *ap)
2096 {
2097 	struct vnode *vp = ap->a_vp;
2098 	struct vnode *tdvp = ap->a_tdvp;
2099 	struct componentname *cnp = ap->a_cnp;
2100 	struct nfsnode *np, *tdnp;
2101 	struct nfsvattr nfsva, dnfsva;
2102 	int error = 0, attrflag, dattrflag;
2103 
2104 	/*
2105 	 * Push all writes to the server, so that the attribute cache
2106 	 * doesn't get "out of sync" with the server.
2107 	 * XXX There should be a better way!
2108 	 */
2109 	VOP_FSYNC(vp, MNT_WAIT, cnp->cn_thread);
2110 
2111 	error = nfsrpc_link(tdvp, vp, cnp->cn_nameptr, cnp->cn_namelen,
2112 	    cnp->cn_cred, cnp->cn_thread, &dnfsva, &nfsva, &attrflag,
2113 	    &dattrflag, NULL);
2114 	tdnp = VTONFS(tdvp);
2115 	NFSLOCKNODE(tdnp);
2116 	tdnp->n_flag |= NMODIFIED;
2117 	if (dattrflag != 0) {
2118 		NFSUNLOCKNODE(tdnp);
2119 		(void) nfscl_loadattrcache(&tdvp, &dnfsva, NULL, NULL, 0, 1);
2120 	} else {
2121 		tdnp->n_attrstamp = 0;
2122 		NFSUNLOCKNODE(tdnp);
2123 		KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(tdvp);
2124 	}
2125 	if (attrflag)
2126 		(void) nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0, 1);
2127 	else {
2128 		np = VTONFS(vp);
2129 		NFSLOCKNODE(np);
2130 		np->n_attrstamp = 0;
2131 		NFSUNLOCKNODE(np);
2132 		KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(vp);
2133 	}
2134 	/*
2135 	 * If negative lookup caching is enabled, I might as well
2136 	 * add an entry for this node. Not necessary for correctness,
2137 	 * but if negative caching is enabled, then the system
2138 	 * must care about lookup caching hit rate, so...
2139 	 */
2140 	if (VFSTONFS(vp->v_mount)->nm_negnametimeo != 0 &&
2141 	    (cnp->cn_flags & MAKEENTRY) && attrflag != 0 && error == 0) {
2142 		if (tdvp != vp)
2143 			cache_enter_time(tdvp, vp, cnp, &nfsva.na_ctime, NULL);
2144 		else
2145 			printf("nfs_link: bogus NFS server returned "
2146 			    "the directory as the new link\n");
2147 	}
2148 	if (error && NFS_ISV4(vp))
2149 		error = nfscl_maperr(cnp->cn_thread, error, (uid_t)0,
2150 		    (gid_t)0);
2151 	return (error);
2152 }
2153 
2154 /*
2155  * nfs symbolic link create call
2156  */
2157 static int
2158 nfs_symlink(struct vop_symlink_args *ap)
2159 {
2160 	struct vnode *dvp = ap->a_dvp;
2161 	struct vattr *vap = ap->a_vap;
2162 	struct componentname *cnp = ap->a_cnp;
2163 	struct nfsvattr nfsva, dnfsva;
2164 	struct nfsfh *nfhp;
2165 	struct nfsnode *np = NULL, *dnp;
2166 	struct vnode *newvp = NULL;
2167 	int error = 0, attrflag, dattrflag, ret;
2168 
2169 	vap->va_type = VLNK;
2170 	error = nfsrpc_symlink(dvp, cnp->cn_nameptr, cnp->cn_namelen,
2171 	    ap->a_target, vap, cnp->cn_cred, cnp->cn_thread, &dnfsva,
2172 	    &nfsva, &nfhp, &attrflag, &dattrflag, NULL);
2173 	if (nfhp) {
2174 		ret = nfscl_nget(dvp->v_mount, dvp, nfhp, cnp, cnp->cn_thread,
2175 		    &np, NULL, LK_EXCLUSIVE);
2176 		if (!ret)
2177 			newvp = NFSTOV(np);
2178 		else if (!error)
2179 			error = ret;
2180 	}
2181 	if (newvp != NULL) {
2182 		if (attrflag)
2183 			(void) nfscl_loadattrcache(&newvp, &nfsva, NULL, NULL,
2184 			    0, 1);
2185 	} else if (!error) {
2186 		/*
2187 		 * If we do not have an error and we could not extract the
2188 		 * newvp from the response due to the request being NFSv2, we
2189 		 * have to do a lookup in order to obtain a newvp to return.
2190 		 */
2191 		error = nfs_lookitup(dvp, cnp->cn_nameptr, cnp->cn_namelen,
2192 		    cnp->cn_cred, cnp->cn_thread, &np);
2193 		if (!error)
2194 			newvp = NFSTOV(np);
2195 	}
2196 	if (error) {
2197 		if (newvp)
2198 			vput(newvp);
2199 		if (NFS_ISV4(dvp))
2200 			error = nfscl_maperr(cnp->cn_thread, error,
2201 			    vap->va_uid, vap->va_gid);
2202 	} else {
2203 		*ap->a_vpp = newvp;
2204 	}
2205 
2206 	dnp = VTONFS(dvp);
2207 	NFSLOCKNODE(dnp);
2208 	dnp->n_flag |= NMODIFIED;
2209 	if (dattrflag != 0) {
2210 		NFSUNLOCKNODE(dnp);
2211 		(void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, NULL, 0, 1);
2212 	} else {
2213 		dnp->n_attrstamp = 0;
2214 		NFSUNLOCKNODE(dnp);
2215 		KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(dvp);
2216 	}
2217 	/*
2218 	 * If negative lookup caching is enabled, I might as well
2219 	 * add an entry for this node. Not necessary for correctness,
2220 	 * but if negative caching is enabled, then the system
2221 	 * must care about lookup caching hit rate, so...
2222 	 */
2223 	if (VFSTONFS(dvp->v_mount)->nm_negnametimeo != 0 &&
2224 	    (cnp->cn_flags & MAKEENTRY) && attrflag != 0 && error == 0) {
2225 		if (dvp != newvp)
2226 			cache_enter_time(dvp, newvp, cnp, &nfsva.na_ctime,
2227 			    NULL);
2228 		else
2229 			printf("nfs_symlink: bogus NFS server returned "
2230 			    "the directory as the new file object\n");
2231 	}
2232 	return (error);
2233 }
2234 
2235 /*
2236  * nfs make dir call
2237  */
2238 static int
2239 nfs_mkdir(struct vop_mkdir_args *ap)
2240 {
2241 	struct vnode *dvp = ap->a_dvp;
2242 	struct vattr *vap = ap->a_vap;
2243 	struct componentname *cnp = ap->a_cnp;
2244 	struct nfsnode *np = NULL, *dnp;
2245 	struct vnode *newvp = NULL;
2246 	struct vattr vattr;
2247 	struct nfsfh *nfhp;
2248 	struct nfsvattr nfsva, dnfsva;
2249 	int error = 0, attrflag, dattrflag, ret;
2250 
2251 	if ((error = VOP_GETATTR(dvp, &vattr, cnp->cn_cred)) != 0)
2252 		return (error);
2253 	vap->va_type = VDIR;
2254 	error = nfsrpc_mkdir(dvp, cnp->cn_nameptr, cnp->cn_namelen,
2255 	    vap, cnp->cn_cred, cnp->cn_thread, &dnfsva, &nfsva, &nfhp,
2256 	    &attrflag, &dattrflag, NULL);
2257 	dnp = VTONFS(dvp);
2258 	NFSLOCKNODE(dnp);
2259 	dnp->n_flag |= NMODIFIED;
2260 	if (dattrflag != 0) {
2261 		NFSUNLOCKNODE(dnp);
2262 		(void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, NULL, 0, 1);
2263 	} else {
2264 		dnp->n_attrstamp = 0;
2265 		NFSUNLOCKNODE(dnp);
2266 		KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(dvp);
2267 	}
2268 	if (nfhp) {
2269 		ret = nfscl_nget(dvp->v_mount, dvp, nfhp, cnp, cnp->cn_thread,
2270 		    &np, NULL, LK_EXCLUSIVE);
2271 		if (!ret) {
2272 			newvp = NFSTOV(np);
2273 			if (attrflag)
2274 			   (void) nfscl_loadattrcache(&newvp, &nfsva, NULL,
2275 				NULL, 0, 1);
2276 		} else if (!error)
2277 			error = ret;
2278 	}
2279 	if (!error && newvp == NULL) {
2280 		error = nfs_lookitup(dvp, cnp->cn_nameptr, cnp->cn_namelen,
2281 		    cnp->cn_cred, cnp->cn_thread, &np);
2282 		if (!error) {
2283 			newvp = NFSTOV(np);
2284 			if (newvp->v_type != VDIR)
2285 				error = EEXIST;
2286 		}
2287 	}
2288 	if (error) {
2289 		if (newvp)
2290 			vput(newvp);
2291 		if (NFS_ISV4(dvp))
2292 			error = nfscl_maperr(cnp->cn_thread, error,
2293 			    vap->va_uid, vap->va_gid);
2294 	} else {
2295 		/*
2296 		 * If negative lookup caching is enabled, I might as well
2297 		 * add an entry for this node. Not necessary for correctness,
2298 		 * but if negative caching is enabled, then the system
2299 		 * must care about lookup caching hit rate, so...
2300 		 */
2301 		if (VFSTONFS(dvp->v_mount)->nm_negnametimeo != 0 &&
2302 		    (cnp->cn_flags & MAKEENTRY) &&
2303 		    attrflag != 0 && dattrflag != 0) {
2304 			if (dvp != newvp)
2305 				cache_enter_time(dvp, newvp, cnp,
2306 				    &nfsva.na_ctime, &dnfsva.na_ctime);
2307 			else
2308 				printf("nfs_mkdir: bogus NFS server returned "
2309 				    "the directory that the directory was "
2310 				    "created in as the new file object\n");
2311 		}
2312 		*ap->a_vpp = newvp;
2313 	}
2314 	return (error);
2315 }
2316 
2317 /*
2318  * nfs remove directory call
2319  */
2320 static int
2321 nfs_rmdir(struct vop_rmdir_args *ap)
2322 {
2323 	struct vnode *vp = ap->a_vp;
2324 	struct vnode *dvp = ap->a_dvp;
2325 	struct componentname *cnp = ap->a_cnp;
2326 	struct nfsnode *dnp;
2327 	struct nfsvattr dnfsva;
2328 	int error, dattrflag;
2329 
2330 	if (dvp == vp)
2331 		return (EINVAL);
2332 	error = nfsrpc_rmdir(dvp, cnp->cn_nameptr, cnp->cn_namelen,
2333 	    cnp->cn_cred, cnp->cn_thread, &dnfsva, &dattrflag, NULL);
2334 	dnp = VTONFS(dvp);
2335 	NFSLOCKNODE(dnp);
2336 	dnp->n_flag |= NMODIFIED;
2337 	if (dattrflag != 0) {
2338 		NFSUNLOCKNODE(dnp);
2339 		(void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, NULL, 0, 1);
2340 	} else {
2341 		dnp->n_attrstamp = 0;
2342 		NFSUNLOCKNODE(dnp);
2343 		KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(dvp);
2344 	}
2345 
2346 	cache_purge(dvp);
2347 	cache_purge(vp);
2348 	if (error && NFS_ISV4(dvp))
2349 		error = nfscl_maperr(cnp->cn_thread, error, (uid_t)0,
2350 		    (gid_t)0);
2351 	/*
2352 	 * Kludge: Map ENOENT => 0 assuming that you have a reply to a retry.
2353 	 */
2354 	if (error == ENOENT)
2355 		error = 0;
2356 	return (error);
2357 }
2358 
2359 /*
2360  * nfs readdir call
2361  */
2362 static int
2363 nfs_readdir(struct vop_readdir_args *ap)
2364 {
2365 	struct vnode *vp = ap->a_vp;
2366 	struct nfsnode *np = VTONFS(vp);
2367 	struct uio *uio = ap->a_uio;
2368 	ssize_t tresid, left;
2369 	int error = 0;
2370 	struct vattr vattr;
2371 
2372 	if (ap->a_eofflag != NULL)
2373 		*ap->a_eofflag = 0;
2374 	if (vp->v_type != VDIR)
2375 		return(EPERM);
2376 
2377 	/*
2378 	 * First, check for hit on the EOF offset cache
2379 	 */
2380 	NFSLOCKNODE(np);
2381 	if (np->n_direofoffset > 0 && uio->uio_offset >= np->n_direofoffset &&
2382 	    (np->n_flag & NMODIFIED) == 0) {
2383 		NFSUNLOCKNODE(np);
2384 		if (VOP_GETATTR(vp, &vattr, ap->a_cred) == 0) {
2385 			NFSLOCKNODE(np);
2386 			if ((NFS_ISV4(vp) && np->n_change == vattr.va_filerev) ||
2387 			    !NFS_TIMESPEC_COMPARE(&np->n_mtime, &vattr.va_mtime)) {
2388 				NFSUNLOCKNODE(np);
2389 				NFSINCRGLOBAL(nfsstatsv1.direofcache_hits);
2390 				if (ap->a_eofflag != NULL)
2391 					*ap->a_eofflag = 1;
2392 				return (0);
2393 			} else
2394 				NFSUNLOCKNODE(np);
2395 		}
2396 	} else
2397 		NFSUNLOCKNODE(np);
2398 
2399 	/*
2400 	 * NFS always guarantees that directory entries don't straddle
2401 	 * DIRBLKSIZ boundaries.  As such, we need to limit the size
2402 	 * to an exact multiple of DIRBLKSIZ, to avoid copying a partial
2403 	 * directory entry.
2404 	 */
2405 	left = uio->uio_resid % DIRBLKSIZ;
2406 	if (left == uio->uio_resid)
2407 		return (EINVAL);
2408 	uio->uio_resid -= left;
2409 
2410 	/*
2411 	 * Call ncl_bioread() to do the real work.
2412 	 */
2413 	tresid = uio->uio_resid;
2414 	error = ncl_bioread(vp, uio, 0, ap->a_cred);
2415 
2416 	if (!error && uio->uio_resid == tresid) {
2417 		NFSINCRGLOBAL(nfsstatsv1.direofcache_misses);
2418 		if (ap->a_eofflag != NULL)
2419 			*ap->a_eofflag = 1;
2420 	}
2421 
2422 	/* Add the partial DIRBLKSIZ (left) back in. */
2423 	uio->uio_resid += left;
2424 	return (error);
2425 }
2426 
2427 /*
2428  * Readdir rpc call.
2429  * Called from below the buffer cache by ncl_doio().
2430  */
2431 int
2432 ncl_readdirrpc(struct vnode *vp, struct uio *uiop, struct ucred *cred,
2433     struct thread *td)
2434 {
2435 	struct nfsvattr nfsva;
2436 	nfsuint64 *cookiep, cookie;
2437 	struct nfsnode *dnp = VTONFS(vp);
2438 	struct nfsmount *nmp = VFSTONFS(vp->v_mount);
2439 	int error = 0, eof, attrflag;
2440 
2441 	KASSERT(uiop->uio_iovcnt == 1 &&
2442 	    (uiop->uio_offset & (DIRBLKSIZ - 1)) == 0 &&
2443 	    (uiop->uio_resid & (DIRBLKSIZ - 1)) == 0,
2444 	    ("nfs readdirrpc bad uio"));
2445 
2446 	/*
2447 	 * If there is no cookie, assume directory was stale.
2448 	 */
2449 	ncl_dircookie_lock(dnp);
2450 	NFSUNLOCKNODE(dnp);
2451 	cookiep = ncl_getcookie(dnp, uiop->uio_offset, 0);
2452 	if (cookiep) {
2453 		cookie = *cookiep;
2454 		ncl_dircookie_unlock(dnp);
2455 	} else {
2456 		ncl_dircookie_unlock(dnp);
2457 		return (NFSERR_BAD_COOKIE);
2458 	}
2459 
2460 	if (NFSHASNFSV3(nmp) && !NFSHASGOTFSINFO(nmp))
2461 		(void)ncl_fsinfo(nmp, vp, cred, td);
2462 
2463 	error = nfsrpc_readdir(vp, uiop, &cookie, cred, td, &nfsva,
2464 	    &attrflag, &eof, NULL);
2465 	if (attrflag)
2466 		(void) nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0, 1);
2467 
2468 	if (!error) {
2469 		/*
2470 		 * We are now either at the end of the directory or have filled
2471 		 * the block.
2472 		 */
2473 		if (eof) {
2474 			NFSLOCKNODE(dnp);
2475 			dnp->n_direofoffset = uiop->uio_offset;
2476 			NFSUNLOCKNODE(dnp);
2477 		} else {
2478 			if (uiop->uio_resid > 0)
2479 				printf("EEK! readdirrpc resid > 0\n");
2480 			ncl_dircookie_lock(dnp);
2481 			NFSUNLOCKNODE(dnp);
2482 			cookiep = ncl_getcookie(dnp, uiop->uio_offset, 1);
2483 			*cookiep = cookie;
2484 			ncl_dircookie_unlock(dnp);
2485 		}
2486 	} else if (NFS_ISV4(vp)) {
2487 		error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0);
2488 	}
2489 	return (error);
2490 }
2491 
2492 /*
2493  * NFS V3 readdir plus RPC. Used in place of ncl_readdirrpc().
2494  */
2495 int
2496 ncl_readdirplusrpc(struct vnode *vp, struct uio *uiop, struct ucred *cred,
2497     struct thread *td)
2498 {
2499 	struct nfsvattr nfsva;
2500 	nfsuint64 *cookiep, cookie;
2501 	struct nfsnode *dnp = VTONFS(vp);
2502 	struct nfsmount *nmp = VFSTONFS(vp->v_mount);
2503 	int error = 0, attrflag, eof;
2504 
2505 	KASSERT(uiop->uio_iovcnt == 1 &&
2506 	    (uiop->uio_offset & (DIRBLKSIZ - 1)) == 0 &&
2507 	    (uiop->uio_resid & (DIRBLKSIZ - 1)) == 0,
2508 	    ("nfs readdirplusrpc bad uio"));
2509 
2510 	/*
2511 	 * If there is no cookie, assume directory was stale.
2512 	 */
2513 	ncl_dircookie_lock(dnp);
2514 	NFSUNLOCKNODE(dnp);
2515 	cookiep = ncl_getcookie(dnp, uiop->uio_offset, 0);
2516 	if (cookiep) {
2517 		cookie = *cookiep;
2518 		ncl_dircookie_unlock(dnp);
2519 	} else {
2520 		ncl_dircookie_unlock(dnp);
2521 		return (NFSERR_BAD_COOKIE);
2522 	}
2523 
2524 	if (NFSHASNFSV3(nmp) && !NFSHASGOTFSINFO(nmp))
2525 		(void)ncl_fsinfo(nmp, vp, cred, td);
2526 	error = nfsrpc_readdirplus(vp, uiop, &cookie, cred, td, &nfsva,
2527 	    &attrflag, &eof, NULL);
2528 	if (attrflag)
2529 		(void) nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0, 1);
2530 
2531 	if (!error) {
2532 		/*
2533 		 * We are now either at end of the directory or have filled the
2534 		 * the block.
2535 		 */
2536 		if (eof) {
2537 			NFSLOCKNODE(dnp);
2538 			dnp->n_direofoffset = uiop->uio_offset;
2539 			NFSUNLOCKNODE(dnp);
2540 		} else {
2541 			if (uiop->uio_resid > 0)
2542 				printf("EEK! readdirplusrpc resid > 0\n");
2543 			ncl_dircookie_lock(dnp);
2544 			NFSUNLOCKNODE(dnp);
2545 			cookiep = ncl_getcookie(dnp, uiop->uio_offset, 1);
2546 			*cookiep = cookie;
2547 			ncl_dircookie_unlock(dnp);
2548 		}
2549 	} else if (NFS_ISV4(vp)) {
2550 		error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0);
2551 	}
2552 	return (error);
2553 }
2554 
2555 /*
2556  * Silly rename. To make the NFS filesystem that is stateless look a little
2557  * more like the "ufs" a remove of an active vnode is translated to a rename
2558  * to a funny looking filename that is removed by nfs_inactive on the
2559  * nfsnode. There is the potential for another process on a different client
2560  * to create the same funny name between the nfs_lookitup() fails and the
2561  * nfs_rename() completes, but...
2562  */
2563 static int
2564 nfs_sillyrename(struct vnode *dvp, struct vnode *vp, struct componentname *cnp)
2565 {
2566 	struct sillyrename *sp;
2567 	struct nfsnode *np;
2568 	int error;
2569 	short pid;
2570 	unsigned int lticks;
2571 
2572 	cache_purge(dvp);
2573 	np = VTONFS(vp);
2574 	KASSERT(vp->v_type != VDIR, ("nfs: sillyrename dir"));
2575 	sp = malloc(sizeof (struct sillyrename),
2576 	    M_NEWNFSREQ, M_WAITOK);
2577 	sp->s_cred = crhold(cnp->cn_cred);
2578 	sp->s_dvp = dvp;
2579 	VREF(dvp);
2580 
2581 	/*
2582 	 * Fudge together a funny name.
2583 	 * Changing the format of the funny name to accommodate more
2584 	 * sillynames per directory.
2585 	 * The name is now changed to .nfs.<ticks>.<pid>.4, where ticks is
2586 	 * CPU ticks since boot.
2587 	 */
2588 	pid = cnp->cn_thread->td_proc->p_pid;
2589 	lticks = (unsigned int)ticks;
2590 	for ( ; ; ) {
2591 		sp->s_namlen = sprintf(sp->s_name,
2592 				       ".nfs.%08x.%04x4.4", lticks,
2593 				       pid);
2594 		if (nfs_lookitup(dvp, sp->s_name, sp->s_namlen, sp->s_cred,
2595 				 cnp->cn_thread, NULL))
2596 			break;
2597 		lticks++;
2598 	}
2599 	error = nfs_renameit(dvp, vp, cnp, sp);
2600 	if (error)
2601 		goto bad;
2602 	error = nfs_lookitup(dvp, sp->s_name, sp->s_namlen, sp->s_cred,
2603 		cnp->cn_thread, &np);
2604 	np->n_sillyrename = sp;
2605 	return (0);
2606 bad:
2607 	vrele(sp->s_dvp);
2608 	crfree(sp->s_cred);
2609 	free(sp, M_NEWNFSREQ);
2610 	return (error);
2611 }
2612 
2613 /*
2614  * Look up a file name and optionally either update the file handle or
2615  * allocate an nfsnode, depending on the value of npp.
2616  * npp == NULL	--> just do the lookup
2617  * *npp == NULL --> allocate a new nfsnode and make sure attributes are
2618  *			handled too
2619  * *npp != NULL --> update the file handle in the vnode
2620  */
2621 static int
2622 nfs_lookitup(struct vnode *dvp, char *name, int len, struct ucred *cred,
2623     struct thread *td, struct nfsnode **npp)
2624 {
2625 	struct vnode *newvp = NULL, *vp;
2626 	struct nfsnode *np, *dnp = VTONFS(dvp);
2627 	struct nfsfh *nfhp, *onfhp;
2628 	struct nfsvattr nfsva, dnfsva;
2629 	struct componentname cn;
2630 	int error = 0, attrflag, dattrflag;
2631 	u_int hash;
2632 
2633 	error = nfsrpc_lookup(dvp, name, len, cred, td, &dnfsva, &nfsva,
2634 	    &nfhp, &attrflag, &dattrflag, NULL, 0);
2635 	if (dattrflag)
2636 		(void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, NULL, 0, 1);
2637 	if (npp && !error) {
2638 		if (*npp != NULL) {
2639 		    np = *npp;
2640 		    vp = NFSTOV(np);
2641 		    /*
2642 		     * For NFSv4, check to see if it is the same name and
2643 		     * replace the name, if it is different.
2644 		     */
2645 		    if (np->n_v4 != NULL && nfsva.na_type == VREG &&
2646 			(np->n_v4->n4_namelen != len ||
2647 			 NFSBCMP(name, NFS4NODENAME(np->n_v4), len) ||
2648 			 dnp->n_fhp->nfh_len != np->n_v4->n4_fhlen ||
2649 			 NFSBCMP(dnp->n_fhp->nfh_fh, np->n_v4->n4_data,
2650 			 dnp->n_fhp->nfh_len))) {
2651 #ifdef notdef
2652 { char nnn[100]; int nnnl;
2653 nnnl = (len < 100) ? len : 99;
2654 bcopy(name, nnn, nnnl);
2655 nnn[nnnl] = '\0';
2656 printf("replace=%s\n",nnn);
2657 }
2658 #endif
2659 			    free(np->n_v4, M_NFSV4NODE);
2660 			    np->n_v4 = malloc(
2661 				sizeof (struct nfsv4node) +
2662 				dnp->n_fhp->nfh_len + len - 1,
2663 				M_NFSV4NODE, M_WAITOK);
2664 			    np->n_v4->n4_fhlen = dnp->n_fhp->nfh_len;
2665 			    np->n_v4->n4_namelen = len;
2666 			    NFSBCOPY(dnp->n_fhp->nfh_fh, np->n_v4->n4_data,
2667 				dnp->n_fhp->nfh_len);
2668 			    NFSBCOPY(name, NFS4NODENAME(np->n_v4), len);
2669 		    }
2670 		    hash = fnv_32_buf(nfhp->nfh_fh, nfhp->nfh_len,
2671 			FNV1_32_INIT);
2672 		    onfhp = np->n_fhp;
2673 		    /*
2674 		     * Rehash node for new file handle.
2675 		     */
2676 		    vfs_hash_rehash(vp, hash);
2677 		    np->n_fhp = nfhp;
2678 		    if (onfhp != NULL)
2679 			free(onfhp, M_NFSFH);
2680 		    newvp = NFSTOV(np);
2681 		} else if (NFS_CMPFH(dnp, nfhp->nfh_fh, nfhp->nfh_len)) {
2682 		    free(nfhp, M_NFSFH);
2683 		    VREF(dvp);
2684 		    newvp = dvp;
2685 		} else {
2686 		    cn.cn_nameptr = name;
2687 		    cn.cn_namelen = len;
2688 		    error = nfscl_nget(dvp->v_mount, dvp, nfhp, &cn, td,
2689 			&np, NULL, LK_EXCLUSIVE);
2690 		    if (error)
2691 			return (error);
2692 		    newvp = NFSTOV(np);
2693 		}
2694 		if (!attrflag && *npp == NULL) {
2695 			if (newvp == dvp)
2696 				vrele(newvp);
2697 			else
2698 				vput(newvp);
2699 			return (ENOENT);
2700 		}
2701 		if (attrflag)
2702 			(void) nfscl_loadattrcache(&newvp, &nfsva, NULL, NULL,
2703 			    0, 1);
2704 	}
2705 	if (npp && *npp == NULL) {
2706 		if (error) {
2707 			if (newvp) {
2708 				if (newvp == dvp)
2709 					vrele(newvp);
2710 				else
2711 					vput(newvp);
2712 			}
2713 		} else
2714 			*npp = np;
2715 	}
2716 	if (error && NFS_ISV4(dvp))
2717 		error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0);
2718 	return (error);
2719 }
2720 
2721 /*
2722  * Nfs Version 3 and 4 commit rpc
2723  */
2724 int
2725 ncl_commit(struct vnode *vp, u_quad_t offset, int cnt, struct ucred *cred,
2726    struct thread *td)
2727 {
2728 	struct nfsvattr nfsva;
2729 	struct nfsmount *nmp = VFSTONFS(vp->v_mount);
2730 	struct nfsnode *np;
2731 	struct uio uio;
2732 	int error, attrflag;
2733 
2734 	np = VTONFS(vp);
2735 	error = EIO;
2736 	attrflag = 0;
2737 	if (NFSHASPNFS(nmp) && (np->n_flag & NDSCOMMIT) != 0) {
2738 		uio.uio_offset = offset;
2739 		uio.uio_resid = cnt;
2740 		error = nfscl_doiods(vp, &uio, NULL, NULL,
2741 		    NFSV4OPEN_ACCESSWRITE, 1, cred, td);
2742 		if (error != 0) {
2743 			NFSLOCKNODE(np);
2744 			np->n_flag &= ~NDSCOMMIT;
2745 			NFSUNLOCKNODE(np);
2746 		}
2747 	}
2748 	if (error != 0) {
2749 		mtx_lock(&nmp->nm_mtx);
2750 		if ((nmp->nm_state & NFSSTA_HASWRITEVERF) == 0) {
2751 			mtx_unlock(&nmp->nm_mtx);
2752 			return (0);
2753 		}
2754 		mtx_unlock(&nmp->nm_mtx);
2755 		error = nfsrpc_commit(vp, offset, cnt, cred, td, &nfsva,
2756 		    &attrflag, NULL);
2757 	}
2758 	if (attrflag != 0)
2759 		(void) nfscl_loadattrcache(&vp, &nfsva, NULL, NULL,
2760 		    0, 1);
2761 	if (error != 0 && NFS_ISV4(vp))
2762 		error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0);
2763 	return (error);
2764 }
2765 
2766 /*
2767  * Strategy routine.
2768  * For async requests when nfsiod(s) are running, queue the request by
2769  * calling ncl_asyncio(), otherwise just all ncl_doio() to do the
2770  * request.
2771  */
2772 static int
2773 nfs_strategy(struct vop_strategy_args *ap)
2774 {
2775 	struct buf *bp;
2776 	struct vnode *vp;
2777 	struct ucred *cr;
2778 
2779 	bp = ap->a_bp;
2780 	vp = ap->a_vp;
2781 	KASSERT(bp->b_vp == vp, ("missing b_getvp"));
2782 	KASSERT(!(bp->b_flags & B_DONE),
2783 	    ("nfs_strategy: buffer %p unexpectedly marked B_DONE", bp));
2784 
2785 	if (vp->v_type == VREG && bp->b_blkno == bp->b_lblkno)
2786 		bp->b_blkno = bp->b_lblkno * (vp->v_bufobj.bo_bsize /
2787 		    DEV_BSIZE);
2788 	if (bp->b_iocmd == BIO_READ)
2789 		cr = bp->b_rcred;
2790 	else
2791 		cr = bp->b_wcred;
2792 
2793 	/*
2794 	 * If the op is asynchronous and an i/o daemon is waiting
2795 	 * queue the request, wake it up and wait for completion
2796 	 * otherwise just do it ourselves.
2797 	 */
2798 	if ((bp->b_flags & B_ASYNC) == 0 ||
2799 	    ncl_asyncio(VFSTONFS(vp->v_mount), bp, NOCRED, curthread))
2800 		(void) ncl_doio(vp, bp, cr, curthread, 1);
2801 	return (0);
2802 }
2803 
2804 /*
2805  * fsync vnode op. Just call ncl_flush() with commit == 1.
2806  */
2807 /* ARGSUSED */
2808 static int
2809 nfs_fsync(struct vop_fsync_args *ap)
2810 {
2811 
2812 	if (ap->a_vp->v_type != VREG) {
2813 		/*
2814 		 * For NFS, metadata is changed synchronously on the server,
2815 		 * so there is nothing to flush. Also, ncl_flush() clears
2816 		 * the NMODIFIED flag and that shouldn't be done here for
2817 		 * directories.
2818 		 */
2819 		return (0);
2820 	}
2821 	return (ncl_flush(ap->a_vp, ap->a_waitfor, ap->a_td, 1, 0));
2822 }
2823 
2824 /*
2825  * Flush all the blocks associated with a vnode.
2826  * 	Walk through the buffer pool and push any dirty pages
2827  *	associated with the vnode.
2828  * If the called_from_renewthread argument is TRUE, it has been called
2829  * from the NFSv4 renew thread and, as such, cannot block indefinitely
2830  * waiting for a buffer write to complete.
2831  */
2832 int
2833 ncl_flush(struct vnode *vp, int waitfor, struct thread *td,
2834     int commit, int called_from_renewthread)
2835 {
2836 	struct nfsnode *np = VTONFS(vp);
2837 	struct buf *bp;
2838 	int i;
2839 	struct buf *nbp;
2840 	struct nfsmount *nmp = VFSTONFS(vp->v_mount);
2841 	int error = 0, slptimeo = 0, slpflag = 0, retv, bvecpos;
2842 	int passone = 1, trycnt = 0;
2843 	u_quad_t off, endoff, toff;
2844 	struct ucred* wcred = NULL;
2845 	struct buf **bvec = NULL;
2846 	struct bufobj *bo;
2847 #ifndef NFS_COMMITBVECSIZ
2848 #define	NFS_COMMITBVECSIZ	20
2849 #endif
2850 	struct buf *bvec_on_stack[NFS_COMMITBVECSIZ];
2851 	u_int bvecsize = 0, bveccount;
2852 
2853 	if (called_from_renewthread != 0)
2854 		slptimeo = hz;
2855 	if (nmp->nm_flag & NFSMNT_INT)
2856 		slpflag = PCATCH;
2857 	if (!commit)
2858 		passone = 0;
2859 	bo = &vp->v_bufobj;
2860 	/*
2861 	 * A b_flags == (B_DELWRI | B_NEEDCOMMIT) block has been written to the
2862 	 * server, but has not been committed to stable storage on the server
2863 	 * yet. On the first pass, the byte range is worked out and the commit
2864 	 * rpc is done. On the second pass, ncl_writebp() is called to do the
2865 	 * job.
2866 	 */
2867 again:
2868 	off = (u_quad_t)-1;
2869 	endoff = 0;
2870 	bvecpos = 0;
2871 	if (NFS_ISV34(vp) && commit) {
2872 		if (bvec != NULL && bvec != bvec_on_stack)
2873 			free(bvec, M_TEMP);
2874 		/*
2875 		 * Count up how many buffers waiting for a commit.
2876 		 */
2877 		bveccount = 0;
2878 		BO_LOCK(bo);
2879 		TAILQ_FOREACH_SAFE(bp, &bo->bo_dirty.bv_hd, b_bobufs, nbp) {
2880 			if (!BUF_ISLOCKED(bp) &&
2881 			    (bp->b_flags & (B_DELWRI | B_NEEDCOMMIT))
2882 				== (B_DELWRI | B_NEEDCOMMIT))
2883 				bveccount++;
2884 		}
2885 		/*
2886 		 * Allocate space to remember the list of bufs to commit.  It is
2887 		 * important to use M_NOWAIT here to avoid a race with nfs_write.
2888 		 * If we can't get memory (for whatever reason), we will end up
2889 		 * committing the buffers one-by-one in the loop below.
2890 		 */
2891 		if (bveccount > NFS_COMMITBVECSIZ) {
2892 			/*
2893 			 * Release the vnode interlock to avoid a lock
2894 			 * order reversal.
2895 			 */
2896 			BO_UNLOCK(bo);
2897 			bvec = (struct buf **)
2898 				malloc(bveccount * sizeof(struct buf *),
2899 				       M_TEMP, M_NOWAIT);
2900 			BO_LOCK(bo);
2901 			if (bvec == NULL) {
2902 				bvec = bvec_on_stack;
2903 				bvecsize = NFS_COMMITBVECSIZ;
2904 			} else
2905 				bvecsize = bveccount;
2906 		} else {
2907 			bvec = bvec_on_stack;
2908 			bvecsize = NFS_COMMITBVECSIZ;
2909 		}
2910 		TAILQ_FOREACH_SAFE(bp, &bo->bo_dirty.bv_hd, b_bobufs, nbp) {
2911 			if (bvecpos >= bvecsize)
2912 				break;
2913 			if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT, NULL)) {
2914 				nbp = TAILQ_NEXT(bp, b_bobufs);
2915 				continue;
2916 			}
2917 			if ((bp->b_flags & (B_DELWRI | B_NEEDCOMMIT)) !=
2918 			    (B_DELWRI | B_NEEDCOMMIT)) {
2919 				BUF_UNLOCK(bp);
2920 				nbp = TAILQ_NEXT(bp, b_bobufs);
2921 				continue;
2922 			}
2923 			BO_UNLOCK(bo);
2924 			bremfree(bp);
2925 			/*
2926 			 * Work out if all buffers are using the same cred
2927 			 * so we can deal with them all with one commit.
2928 			 *
2929 			 * NOTE: we are not clearing B_DONE here, so we have
2930 			 * to do it later on in this routine if we intend to
2931 			 * initiate I/O on the bp.
2932 			 *
2933 			 * Note: to avoid loopback deadlocks, we do not
2934 			 * assign b_runningbufspace.
2935 			 */
2936 			if (wcred == NULL)
2937 				wcred = bp->b_wcred;
2938 			else if (wcred != bp->b_wcred)
2939 				wcred = NOCRED;
2940 			vfs_busy_pages(bp, 1);
2941 
2942 			BO_LOCK(bo);
2943 			/*
2944 			 * bp is protected by being locked, but nbp is not
2945 			 * and vfs_busy_pages() may sleep.  We have to
2946 			 * recalculate nbp.
2947 			 */
2948 			nbp = TAILQ_NEXT(bp, b_bobufs);
2949 
2950 			/*
2951 			 * A list of these buffers is kept so that the
2952 			 * second loop knows which buffers have actually
2953 			 * been committed. This is necessary, since there
2954 			 * may be a race between the commit rpc and new
2955 			 * uncommitted writes on the file.
2956 			 */
2957 			bvec[bvecpos++] = bp;
2958 			toff = ((u_quad_t)bp->b_blkno) * DEV_BSIZE +
2959 				bp->b_dirtyoff;
2960 			if (toff < off)
2961 				off = toff;
2962 			toff += (u_quad_t)(bp->b_dirtyend - bp->b_dirtyoff);
2963 			if (toff > endoff)
2964 				endoff = toff;
2965 		}
2966 		BO_UNLOCK(bo);
2967 	}
2968 	if (bvecpos > 0) {
2969 		/*
2970 		 * Commit data on the server, as required.
2971 		 * If all bufs are using the same wcred, then use that with
2972 		 * one call for all of them, otherwise commit each one
2973 		 * separately.
2974 		 */
2975 		if (wcred != NOCRED)
2976 			retv = ncl_commit(vp, off, (int)(endoff - off),
2977 					  wcred, td);
2978 		else {
2979 			retv = 0;
2980 			for (i = 0; i < bvecpos; i++) {
2981 				off_t off, size;
2982 				bp = bvec[i];
2983 				off = ((u_quad_t)bp->b_blkno) * DEV_BSIZE +
2984 					bp->b_dirtyoff;
2985 				size = (u_quad_t)(bp->b_dirtyend
2986 						  - bp->b_dirtyoff);
2987 				retv = ncl_commit(vp, off, (int)size,
2988 						  bp->b_wcred, td);
2989 				if (retv) break;
2990 			}
2991 		}
2992 
2993 		if (retv == NFSERR_STALEWRITEVERF)
2994 			ncl_clearcommit(vp->v_mount);
2995 
2996 		/*
2997 		 * Now, either mark the blocks I/O done or mark the
2998 		 * blocks dirty, depending on whether the commit
2999 		 * succeeded.
3000 		 */
3001 		for (i = 0; i < bvecpos; i++) {
3002 			bp = bvec[i];
3003 			bp->b_flags &= ~(B_NEEDCOMMIT | B_CLUSTEROK);
3004 			if (retv) {
3005 				/*
3006 				 * Error, leave B_DELWRI intact
3007 				 */
3008 				vfs_unbusy_pages(bp);
3009 				brelse(bp);
3010 			} else {
3011 				/*
3012 				 * Success, remove B_DELWRI ( bundirty() ).
3013 				 *
3014 				 * b_dirtyoff/b_dirtyend seem to be NFS
3015 				 * specific.  We should probably move that
3016 				 * into bundirty(). XXX
3017 				 */
3018 				bufobj_wref(bo);
3019 				bp->b_flags |= B_ASYNC;
3020 				bundirty(bp);
3021 				bp->b_flags &= ~B_DONE;
3022 				bp->b_ioflags &= ~BIO_ERROR;
3023 				bp->b_dirtyoff = bp->b_dirtyend = 0;
3024 				bufdone(bp);
3025 			}
3026 		}
3027 	}
3028 
3029 	/*
3030 	 * Start/do any write(s) that are required.
3031 	 */
3032 loop:
3033 	BO_LOCK(bo);
3034 	TAILQ_FOREACH_SAFE(bp, &bo->bo_dirty.bv_hd, b_bobufs, nbp) {
3035 		if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT, NULL)) {
3036 			if (waitfor != MNT_WAIT || passone)
3037 				continue;
3038 
3039 			error = BUF_TIMELOCK(bp,
3040 			    LK_EXCLUSIVE | LK_SLEEPFAIL | LK_INTERLOCK,
3041 			    BO_LOCKPTR(bo), "nfsfsync", slpflag, slptimeo);
3042 			if (error == 0) {
3043 				BUF_UNLOCK(bp);
3044 				goto loop;
3045 			}
3046 			if (error == ENOLCK) {
3047 				error = 0;
3048 				goto loop;
3049 			}
3050 			if (called_from_renewthread != 0) {
3051 				/*
3052 				 * Return EIO so the flush will be retried
3053 				 * later.
3054 				 */
3055 				error = EIO;
3056 				goto done;
3057 			}
3058 			if (newnfs_sigintr(nmp, td)) {
3059 				error = EINTR;
3060 				goto done;
3061 			}
3062 			if (slpflag == PCATCH) {
3063 				slpflag = 0;
3064 				slptimeo = 2 * hz;
3065 			}
3066 			goto loop;
3067 		}
3068 		if ((bp->b_flags & B_DELWRI) == 0)
3069 			panic("nfs_fsync: not dirty");
3070 		if ((passone || !commit) && (bp->b_flags & B_NEEDCOMMIT)) {
3071 			BUF_UNLOCK(bp);
3072 			continue;
3073 		}
3074 		BO_UNLOCK(bo);
3075 		bremfree(bp);
3076 		bp->b_flags |= B_ASYNC;
3077 		bwrite(bp);
3078 		if (newnfs_sigintr(nmp, td)) {
3079 			error = EINTR;
3080 			goto done;
3081 		}
3082 		goto loop;
3083 	}
3084 	if (passone) {
3085 		passone = 0;
3086 		BO_UNLOCK(bo);
3087 		goto again;
3088 	}
3089 	if (waitfor == MNT_WAIT) {
3090 		while (bo->bo_numoutput) {
3091 			error = bufobj_wwait(bo, slpflag, slptimeo);
3092 			if (error) {
3093 			    BO_UNLOCK(bo);
3094 			    if (called_from_renewthread != 0) {
3095 				/*
3096 				 * Return EIO so that the flush will be
3097 				 * retried later.
3098 				 */
3099 				error = EIO;
3100 				goto done;
3101 			    }
3102 			    error = newnfs_sigintr(nmp, td);
3103 			    if (error)
3104 				goto done;
3105 			    if (slpflag == PCATCH) {
3106 				slpflag = 0;
3107 				slptimeo = 2 * hz;
3108 			    }
3109 			    BO_LOCK(bo);
3110 			}
3111 		}
3112 		if (bo->bo_dirty.bv_cnt != 0 && commit) {
3113 			BO_UNLOCK(bo);
3114 			goto loop;
3115 		}
3116 		/*
3117 		 * Wait for all the async IO requests to drain
3118 		 */
3119 		BO_UNLOCK(bo);
3120 		NFSLOCKNODE(np);
3121 		while (np->n_directio_asyncwr > 0) {
3122 			np->n_flag |= NFSYNCWAIT;
3123 			error = newnfs_msleep(td, &np->n_directio_asyncwr,
3124 			    &np->n_mtx, slpflag | (PRIBIO + 1),
3125 			    "nfsfsync", 0);
3126 			if (error) {
3127 				if (newnfs_sigintr(nmp, td)) {
3128 					NFSUNLOCKNODE(np);
3129 					error = EINTR;
3130 					goto done;
3131 				}
3132 			}
3133 		}
3134 		NFSUNLOCKNODE(np);
3135 	} else
3136 		BO_UNLOCK(bo);
3137 	if (NFSHASPNFS(nmp)) {
3138 		nfscl_layoutcommit(vp, td);
3139 		/*
3140 		 * Invalidate the attribute cache, since writes to a DS
3141 		 * won't update the size attribute.
3142 		 */
3143 		NFSLOCKNODE(np);
3144 		np->n_attrstamp = 0;
3145 	} else
3146 		NFSLOCKNODE(np);
3147 	if (np->n_flag & NWRITEERR) {
3148 		error = np->n_error;
3149 		np->n_flag &= ~NWRITEERR;
3150 	}
3151   	if (commit && bo->bo_dirty.bv_cnt == 0 &&
3152 	    bo->bo_numoutput == 0 && np->n_directio_asyncwr == 0)
3153   		np->n_flag &= ~NMODIFIED;
3154 	NFSUNLOCKNODE(np);
3155 done:
3156 	if (bvec != NULL && bvec != bvec_on_stack)
3157 		free(bvec, M_TEMP);
3158 	if (error == 0 && commit != 0 && waitfor == MNT_WAIT &&
3159 	    (bo->bo_dirty.bv_cnt != 0 || bo->bo_numoutput != 0 ||
3160 	    np->n_directio_asyncwr != 0)) {
3161 		if (trycnt++ < 5) {
3162 			/* try, try again... */
3163 			passone = 1;
3164 			wcred = NULL;
3165 			bvec = NULL;
3166 			bvecsize = 0;
3167 			goto again;
3168 		}
3169 		vn_printf(vp, "ncl_flush failed");
3170 		error = called_from_renewthread != 0 ? EIO : EBUSY;
3171 	}
3172 	return (error);
3173 }
3174 
3175 /*
3176  * NFS advisory byte-level locks.
3177  */
3178 static int
3179 nfs_advlock(struct vop_advlock_args *ap)
3180 {
3181 	struct vnode *vp = ap->a_vp;
3182 	struct ucred *cred;
3183 	struct nfsnode *np = VTONFS(ap->a_vp);
3184 	struct proc *p = (struct proc *)ap->a_id;
3185 	struct thread *td = curthread;	/* XXX */
3186 	struct vattr va;
3187 	int ret, error;
3188 	u_quad_t size;
3189 	struct nfsmount *nmp;
3190 
3191 	error = NFSVOPLOCK(vp, LK_SHARED);
3192 	if (error != 0)
3193 		return (EBADF);
3194 	if (NFS_ISV4(vp) && (ap->a_flags & (F_POSIX | F_FLOCK)) != 0) {
3195 		if (vp->v_type != VREG) {
3196 			error = EINVAL;
3197 			goto out;
3198 		}
3199 		if ((ap->a_flags & F_POSIX) != 0)
3200 			cred = p->p_ucred;
3201 		else
3202 			cred = td->td_ucred;
3203 		NFSVOPLOCK(vp, LK_UPGRADE | LK_RETRY);
3204 		if (VN_IS_DOOMED(vp)) {
3205 			error = EBADF;
3206 			goto out;
3207 		}
3208 
3209 		/*
3210 		 * If this is unlocking a write locked region, flush and
3211 		 * commit them before unlocking. This is required by
3212 		 * RFC3530 Sec. 9.3.2.
3213 		 */
3214 		if (ap->a_op == F_UNLCK &&
3215 		    nfscl_checkwritelocked(vp, ap->a_fl, cred, td, ap->a_id,
3216 		    ap->a_flags))
3217 			(void) ncl_flush(vp, MNT_WAIT, td, 1, 0);
3218 
3219 		/*
3220 		 * Mark NFS node as might have acquired a lock.
3221 		 * This is separate from NHASBEENLOCKED, because it must
3222 		 * be done before the nfsrpc_advlock() call, which might
3223 		 * add a nfscllock structure to the client state.
3224 		 * It is used to check for the case where a nfscllock
3225 		 * state structure cannot exist for the file.
3226 		 * Only done for "oneopenown" NFSv4.1/4.2 mounts.
3227 		 */
3228 		nmp = VFSTONFS(vp->v_mount);
3229 		if (NFSHASNFSV4N(nmp) && NFSHASONEOPENOWN(nmp)) {
3230 			NFSLOCKNODE(np);
3231 			np->n_flag |= NMIGHTBELOCKED;
3232 			NFSUNLOCKNODE(np);
3233 		}
3234 
3235 		/*
3236 		 * Loop around doing the lock op, while a blocking lock
3237 		 * must wait for the lock op to succeed.
3238 		 */
3239 		do {
3240 			ret = nfsrpc_advlock(vp, np->n_size, ap->a_op,
3241 			    ap->a_fl, 0, cred, td, ap->a_id, ap->a_flags);
3242 			if (ret == NFSERR_DENIED && (ap->a_flags & F_WAIT) &&
3243 			    ap->a_op == F_SETLK) {
3244 				NFSVOPUNLOCK(vp);
3245 				error = nfs_catnap(PZERO | PCATCH, ret,
3246 				    "ncladvl");
3247 				if (error)
3248 					return (EINTR);
3249 				NFSVOPLOCK(vp, LK_EXCLUSIVE | LK_RETRY);
3250 				if (VN_IS_DOOMED(vp)) {
3251 					error = EBADF;
3252 					goto out;
3253 				}
3254 			}
3255 		} while (ret == NFSERR_DENIED && (ap->a_flags & F_WAIT) &&
3256 		     ap->a_op == F_SETLK);
3257 		if (ret == NFSERR_DENIED) {
3258 			error = EAGAIN;
3259 			goto out;
3260 		} else if (ret == EINVAL || ret == EBADF || ret == EINTR) {
3261 			error = ret;
3262 			goto out;
3263 		} else if (ret != 0) {
3264 			error = EACCES;
3265 			goto out;
3266 		}
3267 
3268 		/*
3269 		 * Now, if we just got a lock, invalidate data in the buffer
3270 		 * cache, as required, so that the coherency conforms with
3271 		 * RFC3530 Sec. 9.3.2.
3272 		 */
3273 		if (ap->a_op == F_SETLK) {
3274 			if ((np->n_flag & NMODIFIED) == 0) {
3275 				np->n_attrstamp = 0;
3276 				KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(vp);
3277 				ret = VOP_GETATTR(vp, &va, cred);
3278 			}
3279 			if ((np->n_flag & NMODIFIED) || ret ||
3280 			    np->n_change != va.va_filerev) {
3281 				(void) ncl_vinvalbuf(vp, V_SAVE, td, 1);
3282 				np->n_attrstamp = 0;
3283 				KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(vp);
3284 				ret = VOP_GETATTR(vp, &va, cred);
3285 				if (!ret) {
3286 					np->n_mtime = va.va_mtime;
3287 					np->n_change = va.va_filerev;
3288 				}
3289 			}
3290 			/* Mark that a file lock has been acquired. */
3291 			NFSLOCKNODE(np);
3292 			np->n_flag |= NHASBEENLOCKED;
3293 			NFSUNLOCKNODE(np);
3294 		}
3295 	} else if (!NFS_ISV4(vp)) {
3296 		if ((VFSTONFS(vp->v_mount)->nm_flag & NFSMNT_NOLOCKD) != 0) {
3297 			size = VTONFS(vp)->n_size;
3298 			NFSVOPUNLOCK(vp);
3299 			error = lf_advlock(ap, &(vp->v_lockf), size);
3300 		} else {
3301 			if (nfs_advlock_p != NULL)
3302 				error = nfs_advlock_p(ap);
3303 			else {
3304 				NFSVOPUNLOCK(vp);
3305 				error = ENOLCK;
3306 			}
3307 		}
3308 		if (error == 0 && ap->a_op == F_SETLK) {
3309 			error = NFSVOPLOCK(vp, LK_SHARED);
3310 			if (error == 0) {
3311 				/* Mark that a file lock has been acquired. */
3312 				NFSLOCKNODE(np);
3313 				np->n_flag |= NHASBEENLOCKED;
3314 				NFSUNLOCKNODE(np);
3315 				NFSVOPUNLOCK(vp);
3316 			}
3317 		}
3318 		return (error);
3319 	} else
3320 		error = EOPNOTSUPP;
3321 out:
3322 	NFSVOPUNLOCK(vp);
3323 	return (error);
3324 }
3325 
3326 /*
3327  * NFS advisory byte-level locks.
3328  */
3329 static int
3330 nfs_advlockasync(struct vop_advlockasync_args *ap)
3331 {
3332 	struct vnode *vp = ap->a_vp;
3333 	u_quad_t size;
3334 	int error;
3335 
3336 	if (NFS_ISV4(vp))
3337 		return (EOPNOTSUPP);
3338 	error = NFSVOPLOCK(vp, LK_SHARED);
3339 	if (error)
3340 		return (error);
3341 	if ((VFSTONFS(vp->v_mount)->nm_flag & NFSMNT_NOLOCKD) != 0) {
3342 		size = VTONFS(vp)->n_size;
3343 		NFSVOPUNLOCK(vp);
3344 		error = lf_advlockasync(ap, &(vp->v_lockf), size);
3345 	} else {
3346 		NFSVOPUNLOCK(vp);
3347 		error = EOPNOTSUPP;
3348 	}
3349 	return (error);
3350 }
3351 
3352 /*
3353  * Print out the contents of an nfsnode.
3354  */
3355 static int
3356 nfs_print(struct vop_print_args *ap)
3357 {
3358 	struct vnode *vp = ap->a_vp;
3359 	struct nfsnode *np = VTONFS(vp);
3360 
3361 	printf("\tfileid %jd fsid 0x%jx", (uintmax_t)np->n_vattr.na_fileid,
3362 	    (uintmax_t)np->n_vattr.na_fsid);
3363 	if (vp->v_type == VFIFO)
3364 		fifo_printinfo(vp);
3365 	printf("\n");
3366 	return (0);
3367 }
3368 
3369 /*
3370  * This is the "real" nfs::bwrite(struct buf*).
3371  * We set B_CACHE if this is a VMIO buffer.
3372  */
3373 int
3374 ncl_writebp(struct buf *bp, int force __unused, struct thread *td)
3375 {
3376 	int oldflags, rtval;
3377 
3378 	if (bp->b_flags & B_INVAL) {
3379 		brelse(bp);
3380 		return (0);
3381 	}
3382 
3383 	oldflags = bp->b_flags;
3384 	bp->b_flags |= B_CACHE;
3385 
3386 	/*
3387 	 * Undirty the bp.  We will redirty it later if the I/O fails.
3388 	 */
3389 	bundirty(bp);
3390 	bp->b_flags &= ~B_DONE;
3391 	bp->b_ioflags &= ~BIO_ERROR;
3392 	bp->b_iocmd = BIO_WRITE;
3393 
3394 	bufobj_wref(bp->b_bufobj);
3395 	curthread->td_ru.ru_oublock++;
3396 
3397 	/*
3398 	 * Note: to avoid loopback deadlocks, we do not
3399 	 * assign b_runningbufspace.
3400 	 */
3401 	vfs_busy_pages(bp, 1);
3402 
3403 	BUF_KERNPROC(bp);
3404 	bp->b_iooffset = dbtob(bp->b_blkno);
3405 	bstrategy(bp);
3406 
3407 	if ((oldflags & B_ASYNC) != 0)
3408 		return (0);
3409 
3410 	rtval = bufwait(bp);
3411 	if (oldflags & B_DELWRI)
3412 		reassignbuf(bp);
3413 	brelse(bp);
3414 	return (rtval);
3415 }
3416 
3417 /*
3418  * nfs special file access vnode op.
3419  * Essentially just get vattr and then imitate iaccess() since the device is
3420  * local to the client.
3421  */
3422 static int
3423 nfsspec_access(struct vop_access_args *ap)
3424 {
3425 	struct vattr *vap;
3426 	struct ucred *cred = ap->a_cred;
3427 	struct vnode *vp = ap->a_vp;
3428 	accmode_t accmode = ap->a_accmode;
3429 	struct vattr vattr;
3430 	int error;
3431 
3432 	/*
3433 	 * Disallow write attempts on filesystems mounted read-only;
3434 	 * unless the file is a socket, fifo, or a block or character
3435 	 * device resident on the filesystem.
3436 	 */
3437 	if ((accmode & VWRITE) && (vp->v_mount->mnt_flag & MNT_RDONLY)) {
3438 		switch (vp->v_type) {
3439 		case VREG:
3440 		case VDIR:
3441 		case VLNK:
3442 			return (EROFS);
3443 		default:
3444 			break;
3445 		}
3446 	}
3447 	vap = &vattr;
3448 	error = VOP_GETATTR(vp, vap, cred);
3449 	if (error)
3450 		goto out;
3451 	error = vaccess(vp->v_type, vap->va_mode, vap->va_uid, vap->va_gid,
3452 	    accmode, cred);
3453 out:
3454 	return error;
3455 }
3456 
3457 /*
3458  * Read wrapper for fifos.
3459  */
3460 static int
3461 nfsfifo_read(struct vop_read_args *ap)
3462 {
3463 	struct nfsnode *np = VTONFS(ap->a_vp);
3464 	int error;
3465 
3466 	/*
3467 	 * Set access flag.
3468 	 */
3469 	NFSLOCKNODE(np);
3470 	np->n_flag |= NACC;
3471 	vfs_timestamp(&np->n_atim);
3472 	NFSUNLOCKNODE(np);
3473 	error = fifo_specops.vop_read(ap);
3474 	return error;
3475 }
3476 
3477 /*
3478  * Write wrapper for fifos.
3479  */
3480 static int
3481 nfsfifo_write(struct vop_write_args *ap)
3482 {
3483 	struct nfsnode *np = VTONFS(ap->a_vp);
3484 
3485 	/*
3486 	 * Set update flag.
3487 	 */
3488 	NFSLOCKNODE(np);
3489 	np->n_flag |= NUPD;
3490 	vfs_timestamp(&np->n_mtim);
3491 	NFSUNLOCKNODE(np);
3492 	return(fifo_specops.vop_write(ap));
3493 }
3494 
3495 /*
3496  * Close wrapper for fifos.
3497  *
3498  * Update the times on the nfsnode then do fifo close.
3499  */
3500 static int
3501 nfsfifo_close(struct vop_close_args *ap)
3502 {
3503 	struct vnode *vp = ap->a_vp;
3504 	struct nfsnode *np = VTONFS(vp);
3505 	struct vattr vattr;
3506 	struct timespec ts;
3507 
3508 	NFSLOCKNODE(np);
3509 	if (np->n_flag & (NACC | NUPD)) {
3510 		vfs_timestamp(&ts);
3511 		if (np->n_flag & NACC)
3512 			np->n_atim = ts;
3513 		if (np->n_flag & NUPD)
3514 			np->n_mtim = ts;
3515 		np->n_flag |= NCHG;
3516 		if (vrefcnt(vp) == 1 &&
3517 		    (vp->v_mount->mnt_flag & MNT_RDONLY) == 0) {
3518 			VATTR_NULL(&vattr);
3519 			if (np->n_flag & NACC)
3520 				vattr.va_atime = np->n_atim;
3521 			if (np->n_flag & NUPD)
3522 				vattr.va_mtime = np->n_mtim;
3523 			NFSUNLOCKNODE(np);
3524 			(void)VOP_SETATTR(vp, &vattr, ap->a_cred);
3525 			goto out;
3526 		}
3527 	}
3528 	NFSUNLOCKNODE(np);
3529 out:
3530 	return (fifo_specops.vop_close(ap));
3531 }
3532 
3533 /*
3534  * Just call ncl_writebp() with the force argument set to 1.
3535  *
3536  * NOTE: B_DONE may or may not be set in a_bp on call.
3537  */
3538 static int
3539 nfs_bwrite(struct buf *bp)
3540 {
3541 
3542 	return (ncl_writebp(bp, 1, curthread));
3543 }
3544 
3545 struct buf_ops buf_ops_newnfs = {
3546 	.bop_name	=	"buf_ops_nfs",
3547 	.bop_write	=	nfs_bwrite,
3548 	.bop_strategy	=	bufstrategy,
3549 	.bop_sync	=	bufsync,
3550 	.bop_bdflush	=	bufbdflush,
3551 };
3552 
3553 static int
3554 nfs_getacl(struct vop_getacl_args *ap)
3555 {
3556 	int error;
3557 
3558 	if (ap->a_type != ACL_TYPE_NFS4)
3559 		return (EOPNOTSUPP);
3560 	error = nfsrpc_getacl(ap->a_vp, ap->a_cred, ap->a_td, ap->a_aclp,
3561 	    NULL);
3562 	if (error > NFSERR_STALE) {
3563 		(void) nfscl_maperr(ap->a_td, error, (uid_t)0, (gid_t)0);
3564 		error = EPERM;
3565 	}
3566 	return (error);
3567 }
3568 
3569 static int
3570 nfs_setacl(struct vop_setacl_args *ap)
3571 {
3572 	int error;
3573 
3574 	if (ap->a_type != ACL_TYPE_NFS4)
3575 		return (EOPNOTSUPP);
3576 	error = nfsrpc_setacl(ap->a_vp, ap->a_cred, ap->a_td, ap->a_aclp,
3577 	    NULL);
3578 	if (error > NFSERR_STALE) {
3579 		(void) nfscl_maperr(ap->a_td, error, (uid_t)0, (gid_t)0);
3580 		error = EPERM;
3581 	}
3582 	return (error);
3583 }
3584 
3585 /*
3586  * VOP_ADVISE for NFS.
3587  * Just return 0 for any errors, since it is just a hint.
3588  */
3589 static int
3590 nfs_advise(struct vop_advise_args *ap)
3591 {
3592 	struct thread *td = curthread;
3593 	struct nfsmount *nmp;
3594 	uint64_t len;
3595 	int error;
3596 
3597 	/*
3598 	 * First do vop_stdadvise() to handle the buffer cache.
3599 	 */
3600 	error = vop_stdadvise(ap);
3601 	if (error != 0)
3602 		return (error);
3603 	if (ap->a_start < 0 || ap->a_end < 0)
3604 		return (0);
3605 	if (ap->a_end == OFF_MAX)
3606 		len = 0;
3607 	else if (ap->a_end < ap->a_start)
3608 		return (0);
3609 	else
3610 		len = ap->a_end - ap->a_start + 1;
3611 	nmp = VFSTONFS(ap->a_vp->v_mount);
3612 	mtx_lock(&nmp->nm_mtx);
3613 	if (!NFSHASNFSV4(nmp) || nmp->nm_minorvers < NFSV42_MINORVERSION ||
3614 	    (NFSHASPNFS(nmp) && (nmp->nm_privflag & NFSMNTP_IOADVISETHRUMDS) ==
3615 	    0) || (nmp->nm_privflag & NFSMNTP_NOADVISE) != 0) {
3616 		mtx_unlock(&nmp->nm_mtx);
3617 		return (0);
3618 	}
3619 	mtx_unlock(&nmp->nm_mtx);
3620 	error = nfsrpc_advise(ap->a_vp, ap->a_start, len, ap->a_advice,
3621 	    td->td_ucred, td);
3622 	if (error == NFSERR_NOTSUPP) {
3623 		mtx_lock(&nmp->nm_mtx);
3624 		nmp->nm_privflag |= NFSMNTP_NOADVISE;
3625 		mtx_unlock(&nmp->nm_mtx);
3626 	}
3627 	return (0);
3628 }
3629 
3630 /*
3631  * nfs allocate call
3632  */
3633 static int
3634 nfs_allocate(struct vop_allocate_args *ap)
3635 {
3636 	struct vnode *vp = ap->a_vp;
3637 	struct thread *td = curthread;
3638 	struct nfsvattr nfsva;
3639 	struct nfsmount *nmp;
3640 	int attrflag, error, ret;
3641 
3642 	attrflag = 0;
3643 	nmp = VFSTONFS(vp->v_mount);
3644 	mtx_lock(&nmp->nm_mtx);
3645 	if (NFSHASNFSV4(nmp) && nmp->nm_minorvers >= NFSV42_MINORVERSION &&
3646 	    (nmp->nm_privflag & NFSMNTP_NOALLOCATE) == 0) {
3647 		mtx_unlock(&nmp->nm_mtx);
3648 		/*
3649 		 * Flush first to ensure that the allocate adds to the
3650 		 * file's allocation on the server.
3651 		 */
3652 		error = ncl_flush(vp, MNT_WAIT, td, 1, 0);
3653 		if (error == 0)
3654 			error = nfsrpc_allocate(vp, *ap->a_offset, *ap->a_len,
3655 			    &nfsva, &attrflag, td->td_ucred, td, NULL);
3656 		if (error == 0) {
3657 			*ap->a_offset += *ap->a_len;
3658 			*ap->a_len = 0;
3659 		} else if (error == NFSERR_NOTSUPP) {
3660 			mtx_lock(&nmp->nm_mtx);
3661 			nmp->nm_privflag |= NFSMNTP_NOALLOCATE;
3662 			mtx_unlock(&nmp->nm_mtx);
3663 		}
3664 	} else {
3665 		mtx_unlock(&nmp->nm_mtx);
3666 		error = EIO;
3667 	}
3668 	/*
3669 	 * If the NFS server cannot perform the Allocate operation, just call
3670 	 * vop_stdallocate() to perform it.
3671 	 */
3672 	if (error != 0)
3673 		error = vop_stdallocate(ap);
3674 	if (attrflag != 0) {
3675 		ret = nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0, 1);
3676 		if (error == 0 && ret != 0)
3677 			error = ret;
3678 	}
3679 	if (error != 0)
3680 		error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0);
3681 	return (error);
3682 }
3683 
3684 /*
3685  * nfs copy_file_range call
3686  */
3687 static int
3688 nfs_copy_file_range(struct vop_copy_file_range_args *ap)
3689 {
3690 	struct vnode *invp = ap->a_invp;
3691 	struct vnode *outvp = ap->a_outvp;
3692 	struct mount *mp;
3693 	struct nfsvattr innfsva, outnfsva;
3694 	struct vattr *vap;
3695 	struct uio io;
3696 	struct nfsmount *nmp;
3697 	size_t len, len2;
3698 	int error, inattrflag, outattrflag, ret, ret2;
3699 	off_t inoff, outoff;
3700 	bool consecutive, must_commit, tryoutcred;
3701 
3702 	ret = ret2 = 0;
3703 	nmp = VFSTONFS(invp->v_mount);
3704 	mtx_lock(&nmp->nm_mtx);
3705 	/* NFSv4.2 Copy is not permitted for infile == outfile. */
3706 	if (!NFSHASNFSV4(nmp) || nmp->nm_minorvers < NFSV42_MINORVERSION ||
3707 	    (nmp->nm_privflag & NFSMNTP_NOCOPY) != 0 || invp == outvp) {
3708 		mtx_unlock(&nmp->nm_mtx);
3709 		error = vn_generic_copy_file_range(ap->a_invp, ap->a_inoffp,
3710 		    ap->a_outvp, ap->a_outoffp, ap->a_lenp, ap->a_flags,
3711 		    ap->a_incred, ap->a_outcred, ap->a_fsizetd);
3712 		return (error);
3713 	}
3714 	mtx_unlock(&nmp->nm_mtx);
3715 
3716 	/* Lock both vnodes, avoiding risk of deadlock. */
3717 	do {
3718 		mp = NULL;
3719 		error = vn_start_write(outvp, &mp, V_WAIT);
3720 		if (error == 0) {
3721 			error = vn_lock(outvp, LK_EXCLUSIVE);
3722 			if (error == 0) {
3723 				error = vn_lock(invp, LK_SHARED | LK_NOWAIT);
3724 				if (error == 0)
3725 					break;
3726 				VOP_UNLOCK(outvp);
3727 				if (mp != NULL)
3728 					vn_finished_write(mp);
3729 				mp = NULL;
3730 				error = vn_lock(invp, LK_SHARED);
3731 				if (error == 0)
3732 					VOP_UNLOCK(invp);
3733 			}
3734 		}
3735 		if (mp != NULL)
3736 			vn_finished_write(mp);
3737 	} while (error == 0);
3738 	if (error != 0)
3739 		return (error);
3740 
3741 	/*
3742 	 * Do the vn_rlimit_fsize() check.  Should this be above the VOP layer?
3743 	 */
3744 	io.uio_offset = *ap->a_outoffp;
3745 	io.uio_resid = *ap->a_lenp;
3746 	error = vn_rlimit_fsize(outvp, &io, ap->a_fsizetd);
3747 
3748 	/*
3749 	 * Flush the input file so that the data is up to date before
3750 	 * the copy.  Flush writes for the output file so that they
3751 	 * do not overwrite the data copied to the output file by the Copy.
3752 	 * Set the commit argument for both flushes so that the data is on
3753 	 * stable storage before the Copy RPC.  This is done in case the
3754 	 * server reboots during the Copy and needs to be redone.
3755 	 */
3756 	if (error == 0)
3757 		error = ncl_flush(invp, MNT_WAIT, curthread, 1, 0);
3758 	if (error == 0)
3759 		error = ncl_flush(outvp, MNT_WAIT, curthread, 1, 0);
3760 
3761 	/* Do the actual NFSv4.2 RPC. */
3762 	len = *ap->a_lenp;
3763 	mtx_lock(&nmp->nm_mtx);
3764 	if ((nmp->nm_privflag & NFSMNTP_NOCONSECUTIVE) == 0)
3765 		consecutive = true;
3766 	else
3767 		consecutive = false;
3768 	mtx_unlock(&nmp->nm_mtx);
3769 	inoff = *ap->a_inoffp;
3770 	outoff = *ap->a_outoffp;
3771 	tryoutcred = true;
3772 	must_commit = false;
3773 	if (error == 0) {
3774 		vap = &VTONFS(invp)->n_vattr.na_vattr;
3775 		error = VOP_GETATTR(invp, vap, ap->a_incred);
3776 		if (error == 0) {
3777 			/*
3778 			 * Clip "len" at va_size so that RFC compliant servers
3779 			 * will not reply NFSERR_INVAL.
3780 			 * Setting "len == 0" for the RPC would be preferred,
3781 			 * but some Linux servers do not support that.
3782 			 */
3783 			if (inoff >= vap->va_size)
3784 				*ap->a_lenp = len = 0;
3785 			else if (inoff + len > vap->va_size)
3786 				*ap->a_lenp = len = vap->va_size - inoff;
3787 		} else
3788 			error = 0;
3789 	}
3790 
3791 	/*
3792 	 * len will be set to 0 upon a successful Copy RPC.
3793 	 * As such, this only loops when the Copy RPC needs to be retried.
3794 	 */
3795 	while (len > 0 && error == 0) {
3796 		inattrflag = outattrflag = 0;
3797 		len2 = len;
3798 		if (tryoutcred)
3799 			error = nfsrpc_copy_file_range(invp, ap->a_inoffp,
3800 			    outvp, ap->a_outoffp, &len2, ap->a_flags,
3801 			    &inattrflag, &innfsva, &outattrflag, &outnfsva,
3802 			    ap->a_outcred, consecutive, &must_commit);
3803 		else
3804 			error = nfsrpc_copy_file_range(invp, ap->a_inoffp,
3805 			    outvp, ap->a_outoffp, &len2, ap->a_flags,
3806 			    &inattrflag, &innfsva, &outattrflag, &outnfsva,
3807 			    ap->a_incred, consecutive, &must_commit);
3808 		if (inattrflag != 0)
3809 			ret = nfscl_loadattrcache(&invp, &innfsva, NULL, NULL,
3810 			    0, 1);
3811 		if (outattrflag != 0)
3812 			ret2 = nfscl_loadattrcache(&outvp, &outnfsva, NULL,
3813 			    NULL, 1, 1);
3814 		if (error == 0) {
3815 			if (consecutive == false) {
3816 				if (len2 == len) {
3817 					mtx_lock(&nmp->nm_mtx);
3818 					nmp->nm_privflag |=
3819 					    NFSMNTP_NOCONSECUTIVE;
3820 					mtx_unlock(&nmp->nm_mtx);
3821 				} else
3822 					error = NFSERR_OFFLOADNOREQS;
3823 			}
3824 			*ap->a_lenp = len2;
3825 			len = 0;
3826 			if (len2 > 0 && must_commit && error == 0)
3827 				error = ncl_commit(outvp, outoff, *ap->a_lenp,
3828 				    ap->a_outcred, curthread);
3829 			if (error == 0 && ret != 0)
3830 				error = ret;
3831 			if (error == 0 && ret2 != 0)
3832 				error = ret2;
3833 		} else if (error == NFSERR_OFFLOADNOREQS && consecutive) {
3834 			/*
3835 			 * Try consecutive == false, which is ok only if all
3836 			 * bytes are copied.
3837 			 * If only some bytes were copied when consecutive
3838 			 * is false, there is no way to know which bytes
3839 			 * still need to be written.
3840 			 */
3841 			consecutive = false;
3842 			error = 0;
3843 		} else if (error == NFSERR_ACCES && tryoutcred) {
3844 			/* Try again with incred. */
3845 			tryoutcred = false;
3846 			error = 0;
3847 		}
3848 		if (error == NFSERR_STALEWRITEVERF) {
3849 			/*
3850 			 * Server rebooted, so do it all again.
3851 			 */
3852 			*ap->a_inoffp = inoff;
3853 			*ap->a_outoffp = outoff;
3854 			len = *ap->a_lenp;
3855 			must_commit = false;
3856 			error = 0;
3857 		}
3858 	}
3859 	VOP_UNLOCK(invp);
3860 	VOP_UNLOCK(outvp);
3861 	if (mp != NULL)
3862 		vn_finished_write(mp);
3863 	if (error == NFSERR_NOTSUPP || error == NFSERR_OFFLOADNOREQS ||
3864 	    error == NFSERR_ACCES) {
3865 		/*
3866 		 * Unlike the NFSv4.2 Copy, vn_generic_copy_file_range() can
3867 		 * use a_incred for the read and a_outcred for the write, so
3868 		 * try this for NFSERR_ACCES failures for the Copy.
3869 		 * For NFSERR_NOTSUPP and NFSERR_OFFLOADNOREQS, the Copy can
3870 		 * never succeed, so disable it.
3871 		 */
3872 		if (error != NFSERR_ACCES) {
3873 			/* Can never do Copy on this mount. */
3874 			mtx_lock(&nmp->nm_mtx);
3875 			nmp->nm_privflag |= NFSMNTP_NOCOPY;
3876 			mtx_unlock(&nmp->nm_mtx);
3877 		}
3878 		*ap->a_inoffp = inoff;
3879 		*ap->a_outoffp = outoff;
3880 		error = vn_generic_copy_file_range(ap->a_invp, ap->a_inoffp,
3881 		    ap->a_outvp, ap->a_outoffp, ap->a_lenp, ap->a_flags,
3882 		    ap->a_incred, ap->a_outcred, ap->a_fsizetd);
3883 	} else if (error != 0)
3884 		*ap->a_lenp = 0;
3885 
3886 	if (error != 0)
3887 		error = nfscl_maperr(curthread, error, (uid_t)0, (gid_t)0);
3888 	return (error);
3889 }
3890 
3891 /*
3892  * nfs ioctl call
3893  */
3894 static int
3895 nfs_ioctl(struct vop_ioctl_args *ap)
3896 {
3897 	struct vnode *vp = ap->a_vp;
3898 	struct nfsvattr nfsva;
3899 	struct nfsmount *nmp;
3900 	int attrflag, content, error, ret;
3901 	bool eof = false;			/* shut up compiler. */
3902 
3903 	if (vp->v_type != VREG)
3904 		return (ENOTTY);
3905 	nmp = VFSTONFS(vp->v_mount);
3906 	if (!NFSHASNFSV4(nmp) || nmp->nm_minorvers < NFSV42_MINORVERSION) {
3907 		error = vop_stdioctl(ap);
3908 		return (error);
3909 	}
3910 
3911 	/* Do the actual NFSv4.2 RPC. */
3912 	switch (ap->a_command) {
3913 	case FIOSEEKDATA:
3914 		content = NFSV4CONTENT_DATA;
3915 		break;
3916 	case FIOSEEKHOLE:
3917 		content = NFSV4CONTENT_HOLE;
3918 		break;
3919 	default:
3920 		return (ENOTTY);
3921 	}
3922 
3923 	error = vn_lock(vp, LK_SHARED);
3924 	if (error != 0)
3925 		return (EBADF);
3926 	attrflag = 0;
3927 	if (*((off_t *)ap->a_data) >= VTONFS(vp)->n_size)
3928 		error = ENXIO;
3929 	else {
3930 		/*
3931 		 * Flush all writes, so that the server is up to date.
3932 		 * Although a Commit is not required, the commit argument
3933 		 * is set so that, for a pNFS File/Flexible File Layout
3934 		 * server, the LayoutCommit will be done to ensure the file
3935 		 * size is up to date on the Metadata Server.
3936 		 */
3937 		error = ncl_flush(vp, MNT_WAIT, ap->a_td, 1, 0);
3938 		if (error == 0)
3939 			error = nfsrpc_seek(vp, (off_t *)ap->a_data, &eof,
3940 			    content, ap->a_cred, &nfsva, &attrflag);
3941 		/* If at eof for FIOSEEKDATA, return ENXIO. */
3942 		if (eof && error == 0 && content == NFSV4CONTENT_DATA)
3943 			error = ENXIO;
3944 	}
3945 	if (attrflag != 0) {
3946 		ret = nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0, 1);
3947 		if (error == 0 && ret != 0)
3948 			error = ret;
3949 	}
3950 	NFSVOPUNLOCK(vp);
3951 
3952 	if (error != 0)
3953 		error = ENXIO;
3954 	return (error);
3955 }
3956 
3957 /*
3958  * nfs getextattr call
3959  */
3960 static int
3961 nfs_getextattr(struct vop_getextattr_args *ap)
3962 {
3963 	struct vnode *vp = ap->a_vp;
3964 	struct nfsmount *nmp;
3965 	struct ucred *cred;
3966 	struct thread *td = ap->a_td;
3967 	struct nfsvattr nfsva;
3968 	ssize_t len;
3969 	int attrflag, error, ret;
3970 
3971 	nmp = VFSTONFS(vp->v_mount);
3972 	mtx_lock(&nmp->nm_mtx);
3973 	if (!NFSHASNFSV4(nmp) || nmp->nm_minorvers < NFSV42_MINORVERSION ||
3974 	    (nmp->nm_privflag & NFSMNTP_NOXATTR) != 0 ||
3975 	    ap->a_attrnamespace != EXTATTR_NAMESPACE_USER) {
3976 		mtx_unlock(&nmp->nm_mtx);
3977 		return (EOPNOTSUPP);
3978 	}
3979 	mtx_unlock(&nmp->nm_mtx);
3980 
3981 	cred = ap->a_cred;
3982 	if (cred == NULL)
3983 		cred = td->td_ucred;
3984 	/* Do the actual NFSv4.2 Optional Extended Attribute (RFC-8276) RPC. */
3985 	attrflag = 0;
3986 	error = nfsrpc_getextattr(vp, ap->a_name, ap->a_uio, &len, &nfsva,
3987 	    &attrflag, cred, td);
3988 	if (attrflag != 0) {
3989 		ret = nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0, 1);
3990 		if (error == 0 && ret != 0)
3991 			error = ret;
3992 	}
3993 	if (error == 0 && ap->a_size != NULL)
3994 		*ap->a_size = len;
3995 
3996 	switch (error) {
3997 	case NFSERR_NOTSUPP:
3998 	case NFSERR_OPILLEGAL:
3999 		mtx_lock(&nmp->nm_mtx);
4000 		nmp->nm_privflag |= NFSMNTP_NOXATTR;
4001 		mtx_unlock(&nmp->nm_mtx);
4002 		error = EOPNOTSUPP;
4003 		break;
4004 	case NFSERR_NOXATTR:
4005 	case NFSERR_XATTR2BIG:
4006 		error = ENOATTR;
4007 		break;
4008 	default:
4009 		error = nfscl_maperr(td, error, 0, 0);
4010 		break;
4011 	}
4012 	return (error);
4013 }
4014 
4015 /*
4016  * nfs setextattr call
4017  */
4018 static int
4019 nfs_setextattr(struct vop_setextattr_args *ap)
4020 {
4021 	struct vnode *vp = ap->a_vp;
4022 	struct nfsmount *nmp;
4023 	struct ucred *cred;
4024 	struct thread *td = ap->a_td;
4025 	struct nfsvattr nfsva;
4026 	int attrflag, error, ret;
4027 
4028 	nmp = VFSTONFS(vp->v_mount);
4029 	mtx_lock(&nmp->nm_mtx);
4030 	if (!NFSHASNFSV4(nmp) || nmp->nm_minorvers < NFSV42_MINORVERSION ||
4031 	    (nmp->nm_privflag & NFSMNTP_NOXATTR) != 0 ||
4032 	    ap->a_attrnamespace != EXTATTR_NAMESPACE_USER) {
4033 		mtx_unlock(&nmp->nm_mtx);
4034 		return (EOPNOTSUPP);
4035 	}
4036 	mtx_unlock(&nmp->nm_mtx);
4037 
4038 	if (ap->a_uio->uio_resid < 0)
4039 		return (EINVAL);
4040 	cred = ap->a_cred;
4041 	if (cred == NULL)
4042 		cred = td->td_ucred;
4043 	/* Do the actual NFSv4.2 Optional Extended Attribute (RFC-8276) RPC. */
4044 	attrflag = 0;
4045 	error = nfsrpc_setextattr(vp, ap->a_name, ap->a_uio, &nfsva,
4046 	    &attrflag, cred, td);
4047 	if (attrflag != 0) {
4048 		ret = nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0, 1);
4049 		if (error == 0 && ret != 0)
4050 			error = ret;
4051 	}
4052 
4053 	switch (error) {
4054 	case NFSERR_NOTSUPP:
4055 	case NFSERR_OPILLEGAL:
4056 		mtx_lock(&nmp->nm_mtx);
4057 		nmp->nm_privflag |= NFSMNTP_NOXATTR;
4058 		mtx_unlock(&nmp->nm_mtx);
4059 		error = EOPNOTSUPP;
4060 		break;
4061 	case NFSERR_NOXATTR:
4062 	case NFSERR_XATTR2BIG:
4063 		error = ENOATTR;
4064 		break;
4065 	default:
4066 		error = nfscl_maperr(td, error, 0, 0);
4067 		break;
4068 	}
4069 	return (error);
4070 }
4071 
4072 /*
4073  * nfs listextattr call
4074  */
4075 static int
4076 nfs_listextattr(struct vop_listextattr_args *ap)
4077 {
4078 	struct vnode *vp = ap->a_vp;
4079 	struct nfsmount *nmp;
4080 	struct ucred *cred;
4081 	struct thread *td = ap->a_td;
4082 	struct nfsvattr nfsva;
4083 	size_t len, len2;
4084 	uint64_t cookie;
4085 	int attrflag, error, ret;
4086 	bool eof;
4087 
4088 	nmp = VFSTONFS(vp->v_mount);
4089 	mtx_lock(&nmp->nm_mtx);
4090 	if (!NFSHASNFSV4(nmp) || nmp->nm_minorvers < NFSV42_MINORVERSION ||
4091 	    (nmp->nm_privflag & NFSMNTP_NOXATTR) != 0 ||
4092 	    ap->a_attrnamespace != EXTATTR_NAMESPACE_USER) {
4093 		mtx_unlock(&nmp->nm_mtx);
4094 		return (EOPNOTSUPP);
4095 	}
4096 	mtx_unlock(&nmp->nm_mtx);
4097 
4098 	cred = ap->a_cred;
4099 	if (cred == NULL)
4100 		cred = td->td_ucred;
4101 
4102 	/* Loop around doing List Extended Attribute RPCs. */
4103 	eof = false;
4104 	cookie = 0;
4105 	len2 = 0;
4106 	error = 0;
4107 	while (!eof && error == 0) {
4108 		len = nmp->nm_rsize;
4109 		attrflag = 0;
4110 		error = nfsrpc_listextattr(vp, &cookie, ap->a_uio, &len, &eof,
4111 		    &nfsva, &attrflag, cred, td);
4112 		if (attrflag != 0) {
4113 			ret = nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0,
4114 			    1);
4115 			if (error == 0 && ret != 0)
4116 				error = ret;
4117 		}
4118 		if (error == 0) {
4119 			len2 += len;
4120 			if (len2 > SSIZE_MAX)
4121 				error = ENOATTR;
4122 		}
4123 	}
4124 	if (error == 0 && ap->a_size != NULL)
4125 		*ap->a_size = len2;
4126 
4127 	switch (error) {
4128 	case NFSERR_NOTSUPP:
4129 	case NFSERR_OPILLEGAL:
4130 		mtx_lock(&nmp->nm_mtx);
4131 		nmp->nm_privflag |= NFSMNTP_NOXATTR;
4132 		mtx_unlock(&nmp->nm_mtx);
4133 		error = EOPNOTSUPP;
4134 		break;
4135 	case NFSERR_NOXATTR:
4136 	case NFSERR_XATTR2BIG:
4137 		error = ENOATTR;
4138 		break;
4139 	default:
4140 		error = nfscl_maperr(td, error, 0, 0);
4141 		break;
4142 	}
4143 	return (error);
4144 }
4145 
4146 /*
4147  * nfs setextattr call
4148  */
4149 static int
4150 nfs_deleteextattr(struct vop_deleteextattr_args *ap)
4151 {
4152 	struct vnode *vp = ap->a_vp;
4153 	struct nfsmount *nmp;
4154 	struct nfsvattr nfsva;
4155 	int attrflag, error, ret;
4156 
4157 	nmp = VFSTONFS(vp->v_mount);
4158 	mtx_lock(&nmp->nm_mtx);
4159 	if (!NFSHASNFSV4(nmp) || nmp->nm_minorvers < NFSV42_MINORVERSION ||
4160 	    (nmp->nm_privflag & NFSMNTP_NOXATTR) != 0 ||
4161 	    ap->a_attrnamespace != EXTATTR_NAMESPACE_USER) {
4162 		mtx_unlock(&nmp->nm_mtx);
4163 		return (EOPNOTSUPP);
4164 	}
4165 	mtx_unlock(&nmp->nm_mtx);
4166 
4167 	/* Do the actual NFSv4.2 Optional Extended Attribute (RFC-8276) RPC. */
4168 	attrflag = 0;
4169 	error = nfsrpc_rmextattr(vp, ap->a_name, &nfsva, &attrflag, ap->a_cred,
4170 	    ap->a_td);
4171 	if (attrflag != 0) {
4172 		ret = nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0, 1);
4173 		if (error == 0 && ret != 0)
4174 			error = ret;
4175 	}
4176 
4177 	switch (error) {
4178 	case NFSERR_NOTSUPP:
4179 	case NFSERR_OPILLEGAL:
4180 		mtx_lock(&nmp->nm_mtx);
4181 		nmp->nm_privflag |= NFSMNTP_NOXATTR;
4182 		mtx_unlock(&nmp->nm_mtx);
4183 		error = EOPNOTSUPP;
4184 		break;
4185 	case NFSERR_NOXATTR:
4186 	case NFSERR_XATTR2BIG:
4187 		error = ENOATTR;
4188 		break;
4189 	default:
4190 		error = nfscl_maperr(ap->a_td, error, 0, 0);
4191 		break;
4192 	}
4193 	return (error);
4194 }
4195 
4196 /*
4197  * Return POSIX pathconf information applicable to nfs filesystems.
4198  */
4199 static int
4200 nfs_pathconf(struct vop_pathconf_args *ap)
4201 {
4202 	struct nfsv3_pathconf pc;
4203 	struct nfsvattr nfsva;
4204 	struct vnode *vp = ap->a_vp;
4205 	struct nfsmount *nmp;
4206 	struct thread *td = curthread;
4207 	off_t off;
4208 	bool eof;
4209 	int attrflag, error;
4210 
4211 	if ((NFS_ISV34(vp) && (ap->a_name == _PC_LINK_MAX ||
4212 	    ap->a_name == _PC_NAME_MAX || ap->a_name == _PC_CHOWN_RESTRICTED ||
4213 	    ap->a_name == _PC_NO_TRUNC)) ||
4214 	    (NFS_ISV4(vp) && ap->a_name == _PC_ACL_NFS4)) {
4215 		/*
4216 		 * Since only the above 4 a_names are returned by the NFSv3
4217 		 * Pathconf RPC, there is no point in doing it for others.
4218 		 * For NFSv4, the Pathconf RPC (actually a Getattr Op.) can
4219 		 * be used for _PC_NFS4_ACL as well.
4220 		 */
4221 		error = nfsrpc_pathconf(vp, &pc, td->td_ucred, td, &nfsva,
4222 		    &attrflag, NULL);
4223 		if (attrflag != 0)
4224 			(void) nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0,
4225 			    1);
4226 		if (error != 0)
4227 			return (error);
4228 	} else {
4229 		/*
4230 		 * For NFSv2 (or NFSv3 when not one of the above 4 a_names),
4231 		 * just fake them.
4232 		 */
4233 		pc.pc_linkmax = NFS_LINK_MAX;
4234 		pc.pc_namemax = NFS_MAXNAMLEN;
4235 		pc.pc_notrunc = 1;
4236 		pc.pc_chownrestricted = 1;
4237 		pc.pc_caseinsensitive = 0;
4238 		pc.pc_casepreserving = 1;
4239 		error = 0;
4240 	}
4241 	switch (ap->a_name) {
4242 	case _PC_LINK_MAX:
4243 #ifdef _LP64
4244 		*ap->a_retval = pc.pc_linkmax;
4245 #else
4246 		*ap->a_retval = MIN(LONG_MAX, pc.pc_linkmax);
4247 #endif
4248 		break;
4249 	case _PC_NAME_MAX:
4250 		*ap->a_retval = pc.pc_namemax;
4251 		break;
4252 	case _PC_PIPE_BUF:
4253 		if (ap->a_vp->v_type == VDIR || ap->a_vp->v_type == VFIFO)
4254 			*ap->a_retval = PIPE_BUF;
4255 		else
4256 			error = EINVAL;
4257 		break;
4258 	case _PC_CHOWN_RESTRICTED:
4259 		*ap->a_retval = pc.pc_chownrestricted;
4260 		break;
4261 	case _PC_NO_TRUNC:
4262 		*ap->a_retval = pc.pc_notrunc;
4263 		break;
4264 	case _PC_ACL_NFS4:
4265 		if (NFS_ISV4(vp) && nfsrv_useacl != 0 && attrflag != 0 &&
4266 		    NFSISSET_ATTRBIT(&nfsva.na_suppattr, NFSATTRBIT_ACL))
4267 			*ap->a_retval = 1;
4268 		else
4269 			*ap->a_retval = 0;
4270 		break;
4271 	case _PC_ACL_PATH_MAX:
4272 		if (NFS_ISV4(vp))
4273 			*ap->a_retval = ACL_MAX_ENTRIES;
4274 		else
4275 			*ap->a_retval = 3;
4276 		break;
4277 	case _PC_PRIO_IO:
4278 		*ap->a_retval = 0;
4279 		break;
4280 	case _PC_SYNC_IO:
4281 		*ap->a_retval = 0;
4282 		break;
4283 	case _PC_ALLOC_SIZE_MIN:
4284 		*ap->a_retval = vp->v_mount->mnt_stat.f_bsize;
4285 		break;
4286 	case _PC_FILESIZEBITS:
4287 		if (NFS_ISV34(vp))
4288 			*ap->a_retval = 64;
4289 		else
4290 			*ap->a_retval = 32;
4291 		break;
4292 	case _PC_REC_INCR_XFER_SIZE:
4293 		*ap->a_retval = vp->v_mount->mnt_stat.f_iosize;
4294 		break;
4295 	case _PC_REC_MAX_XFER_SIZE:
4296 		*ap->a_retval = -1; /* means ``unlimited'' */
4297 		break;
4298 	case _PC_REC_MIN_XFER_SIZE:
4299 		*ap->a_retval = vp->v_mount->mnt_stat.f_iosize;
4300 		break;
4301 	case _PC_REC_XFER_ALIGN:
4302 		*ap->a_retval = PAGE_SIZE;
4303 		break;
4304 	case _PC_SYMLINK_MAX:
4305 		*ap->a_retval = NFS_MAXPATHLEN;
4306 		break;
4307 	case _PC_MIN_HOLE_SIZE:
4308 		/* Only some NFSv4.2 servers support Seek for Holes. */
4309 		*ap->a_retval = 0;
4310 		nmp = VFSTONFS(vp->v_mount);
4311 		if (NFS_ISV4(vp) && nmp->nm_minorvers == NFSV42_MINORVERSION) {
4312 			/*
4313 			 * NFSv4.2 doesn't have an attribute for hole size,
4314 			 * so all we can do is see if the Seek operation is
4315 			 * supported and then use f_iosize as a "best guess".
4316 			 */
4317 			mtx_lock(&nmp->nm_mtx);
4318 			if ((nmp->nm_privflag & NFSMNTP_SEEKTESTED) == 0) {
4319 				mtx_unlock(&nmp->nm_mtx);
4320 				off = 0;
4321 				attrflag = 0;
4322 				error = nfsrpc_seek(vp, &off, &eof,
4323 				    NFSV4CONTENT_HOLE, td->td_ucred, &nfsva,
4324 				    &attrflag);
4325 				if (attrflag != 0)
4326 					nfscl_loadattrcache(&vp, &nfsva,
4327 					    NULL, NULL, 0, 1);
4328 				mtx_lock(&nmp->nm_mtx);
4329 				if (error == NFSERR_NOTSUPP)
4330 					nmp->nm_privflag |= NFSMNTP_SEEKTESTED;
4331 				else
4332 					nmp->nm_privflag |= NFSMNTP_SEEKTESTED |
4333 					    NFSMNTP_SEEK;
4334 				error = 0;
4335 			}
4336 			if ((nmp->nm_privflag & NFSMNTP_SEEK) != 0)
4337 				*ap->a_retval = vp->v_mount->mnt_stat.f_iosize;
4338 			mtx_unlock(&nmp->nm_mtx);
4339 		}
4340 		break;
4341 
4342 	default:
4343 		error = vop_stdpathconf(ap);
4344 		break;
4345 	}
4346 	return (error);
4347 }
4348