nfs_clvnops.c revision 317579
1/*-
2 * Copyright (c) 1989, 1993
3 *	The Regents of the University of California.  All rights reserved.
4 *
5 * This code is derived from software contributed to Berkeley by
6 * Rick Macklem at The University of Guelph.
7 *
8 * Redistribution and use in source and binary forms, with or without
9 * modification, are permitted provided that the following conditions
10 * are met:
11 * 1. Redistributions of source code must retain the above copyright
12 *    notice, this list of conditions and the following disclaimer.
13 * 2. Redistributions in binary form must reproduce the above copyright
14 *    notice, this list of conditions and the following disclaimer in the
15 *    documentation and/or other materials provided with the distribution.
16 * 4. Neither the name of the University nor the names of its contributors
17 *    may be used to endorse or promote products derived from this software
18 *    without specific prior written permission.
19 *
20 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
21 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
22 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
23 * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
24 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
25 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
26 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
27 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
28 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
29 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
30 * SUCH DAMAGE.
31 *
32 *	from nfs_vnops.c	8.16 (Berkeley) 5/27/95
33 */
34
35#include <sys/cdefs.h>
36__FBSDID("$FreeBSD: stable/10/sys/fs/nfsclient/nfs_clvnops.c 317579 2017-04-29 01:04:39Z rmacklem $");
37
38/*
39 * vnode op calls for Sun NFS version 2, 3 and 4
40 */
41
42#include "opt_kdtrace.h"
43#include "opt_inet.h"
44
45#include <sys/param.h>
46#include <sys/kernel.h>
47#include <sys/systm.h>
48#include <sys/resourcevar.h>
49#include <sys/proc.h>
50#include <sys/mount.h>
51#include <sys/bio.h>
52#include <sys/buf.h>
53#include <sys/jail.h>
54#include <sys/malloc.h>
55#include <sys/mbuf.h>
56#include <sys/namei.h>
57#include <sys/socket.h>
58#include <sys/vnode.h>
59#include <sys/dirent.h>
60#include <sys/fcntl.h>
61#include <sys/lockf.h>
62#include <sys/stat.h>
63#include <sys/sysctl.h>
64#include <sys/signalvar.h>
65
66#include <vm/vm.h>
67#include <vm/vm_extern.h>
68#include <vm/vm_object.h>
69
70#include <fs/nfs/nfsport.h>
71#include <fs/nfsclient/nfsnode.h>
72#include <fs/nfsclient/nfsmount.h>
73#include <fs/nfsclient/nfs.h>
74#include <fs/nfsclient/nfs_kdtrace.h>
75
76#include <net/if.h>
77#include <netinet/in.h>
78#include <netinet/in_var.h>
79
80#include <nfs/nfs_lock.h>
81
82#ifdef KDTRACE_HOOKS
83#include <sys/dtrace_bsd.h>
84
85dtrace_nfsclient_accesscache_flush_probe_func_t
86		dtrace_nfscl_accesscache_flush_done_probe;
87uint32_t	nfscl_accesscache_flush_done_id;
88
89dtrace_nfsclient_accesscache_get_probe_func_t
90		dtrace_nfscl_accesscache_get_hit_probe,
91		dtrace_nfscl_accesscache_get_miss_probe;
92uint32_t	nfscl_accesscache_get_hit_id;
93uint32_t	nfscl_accesscache_get_miss_id;
94
95dtrace_nfsclient_accesscache_load_probe_func_t
96		dtrace_nfscl_accesscache_load_done_probe;
97uint32_t	nfscl_accesscache_load_done_id;
98#endif /* !KDTRACE_HOOKS */
99
100/* Defs */
101#define	TRUE	1
102#define	FALSE	0
103
104extern struct nfsstats newnfsstats;
105extern int nfsrv_useacl;
106extern int nfscl_debuglevel;
107MALLOC_DECLARE(M_NEWNFSREQ);
108
109/*
110 * Ifdef for FreeBSD-current merged buffer cache. It is unfortunate that these
111 * calls are not in getblk() and brelse() so that they would not be necessary
112 * here.
113 */
114#ifndef B_VMIO
115#define	vfs_busy_pages(bp, f)
116#endif
117
118static vop_read_t	nfsfifo_read;
119static vop_write_t	nfsfifo_write;
120static vop_close_t	nfsfifo_close;
121static int	nfs_setattrrpc(struct vnode *, struct vattr *, struct ucred *,
122		    struct thread *);
123static vop_lookup_t	nfs_lookup;
124static vop_create_t	nfs_create;
125static vop_mknod_t	nfs_mknod;
126static vop_open_t	nfs_open;
127static vop_pathconf_t	nfs_pathconf;
128static vop_close_t	nfs_close;
129static vop_access_t	nfs_access;
130static vop_getattr_t	nfs_getattr;
131static vop_setattr_t	nfs_setattr;
132static vop_read_t	nfs_read;
133static vop_fsync_t	nfs_fsync;
134static vop_remove_t	nfs_remove;
135static vop_link_t	nfs_link;
136static vop_rename_t	nfs_rename;
137static vop_mkdir_t	nfs_mkdir;
138static vop_rmdir_t	nfs_rmdir;
139static vop_symlink_t	nfs_symlink;
140static vop_readdir_t	nfs_readdir;
141static vop_strategy_t	nfs_strategy;
142static	int	nfs_lookitup(struct vnode *, char *, int,
143		    struct ucred *, struct thread *, struct nfsnode **);
144static	int	nfs_sillyrename(struct vnode *, struct vnode *,
145		    struct componentname *);
146static vop_access_t	nfsspec_access;
147static vop_readlink_t	nfs_readlink;
148static vop_print_t	nfs_print;
149static vop_advlock_t	nfs_advlock;
150static vop_advlockasync_t nfs_advlockasync;
151static vop_getacl_t nfs_getacl;
152static vop_setacl_t nfs_setacl;
153static vop_set_text_t nfs_set_text;
154
155/*
156 * Global vfs data structures for nfs
157 */
158struct vop_vector newnfs_vnodeops = {
159	.vop_default =		&default_vnodeops,
160	.vop_access =		nfs_access,
161	.vop_advlock =		nfs_advlock,
162	.vop_advlockasync =	nfs_advlockasync,
163	.vop_close =		nfs_close,
164	.vop_create =		nfs_create,
165	.vop_fsync =		nfs_fsync,
166	.vop_getattr =		nfs_getattr,
167	.vop_getpages =		ncl_getpages,
168	.vop_putpages =		ncl_putpages,
169	.vop_inactive =		ncl_inactive,
170	.vop_link =		nfs_link,
171	.vop_lookup =		nfs_lookup,
172	.vop_mkdir =		nfs_mkdir,
173	.vop_mknod =		nfs_mknod,
174	.vop_open =		nfs_open,
175	.vop_pathconf =		nfs_pathconf,
176	.vop_print =		nfs_print,
177	.vop_read =		nfs_read,
178	.vop_readdir =		nfs_readdir,
179	.vop_readlink =		nfs_readlink,
180	.vop_reclaim =		ncl_reclaim,
181	.vop_remove =		nfs_remove,
182	.vop_rename =		nfs_rename,
183	.vop_rmdir =		nfs_rmdir,
184	.vop_setattr =		nfs_setattr,
185	.vop_strategy =		nfs_strategy,
186	.vop_symlink =		nfs_symlink,
187	.vop_write =		ncl_write,
188	.vop_getacl =		nfs_getacl,
189	.vop_setacl =		nfs_setacl,
190	.vop_set_text =		nfs_set_text,
191};
192
193struct vop_vector newnfs_fifoops = {
194	.vop_default =		&fifo_specops,
195	.vop_access =		nfsspec_access,
196	.vop_close =		nfsfifo_close,
197	.vop_fsync =		nfs_fsync,
198	.vop_getattr =		nfs_getattr,
199	.vop_inactive =		ncl_inactive,
200	.vop_print =		nfs_print,
201	.vop_read =		nfsfifo_read,
202	.vop_reclaim =		ncl_reclaim,
203	.vop_setattr =		nfs_setattr,
204	.vop_write =		nfsfifo_write,
205};
206
207static int nfs_mknodrpc(struct vnode *dvp, struct vnode **vpp,
208    struct componentname *cnp, struct vattr *vap);
209static int nfs_removerpc(struct vnode *dvp, struct vnode *vp, char *name,
210    int namelen, struct ucred *cred, struct thread *td);
211static int nfs_renamerpc(struct vnode *fdvp, struct vnode *fvp,
212    char *fnameptr, int fnamelen, struct vnode *tdvp, struct vnode *tvp,
213    char *tnameptr, int tnamelen, struct ucred *cred, struct thread *td);
214static int nfs_renameit(struct vnode *sdvp, struct vnode *svp,
215    struct componentname *scnp, struct sillyrename *sp);
216
217/*
218 * Global variables
219 */
220#define	DIRHDSIZ	(sizeof (struct dirent) - (MAXNAMLEN + 1))
221
222SYSCTL_DECL(_vfs_nfs);
223
224static int	nfsaccess_cache_timeout = NFS_MAXATTRTIMO;
225SYSCTL_INT(_vfs_nfs, OID_AUTO, access_cache_timeout, CTLFLAG_RW,
226	   &nfsaccess_cache_timeout, 0, "NFS ACCESS cache timeout");
227
228static int	nfs_prime_access_cache = 0;
229SYSCTL_INT(_vfs_nfs, OID_AUTO, prime_access_cache, CTLFLAG_RW,
230	   &nfs_prime_access_cache, 0,
231	   "Prime NFS ACCESS cache when fetching attributes");
232
233static int	newnfs_commit_on_close = 0;
234SYSCTL_INT(_vfs_nfs, OID_AUTO, commit_on_close, CTLFLAG_RW,
235    &newnfs_commit_on_close, 0, "write+commit on close, else only write");
236
237static int	nfs_clean_pages_on_close = 1;
238SYSCTL_INT(_vfs_nfs, OID_AUTO, clean_pages_on_close, CTLFLAG_RW,
239	   &nfs_clean_pages_on_close, 0, "NFS clean dirty pages on close");
240
241int newnfs_directio_enable = 0;
242SYSCTL_INT(_vfs_nfs, OID_AUTO, nfs_directio_enable, CTLFLAG_RW,
243	   &newnfs_directio_enable, 0, "Enable NFS directio");
244
245int nfs_keep_dirty_on_error;
246SYSCTL_INT(_vfs_nfs, OID_AUTO, nfs_keep_dirty_on_error, CTLFLAG_RW,
247    &nfs_keep_dirty_on_error, 0, "Retry pageout if error returned");
248
249/*
250 * This sysctl allows other processes to mmap a file that has been opened
251 * O_DIRECT by a process.  In general, having processes mmap the file while
252 * Direct IO is in progress can lead to Data Inconsistencies.  But, we allow
253 * this by default to prevent DoS attacks - to prevent a malicious user from
254 * opening up files O_DIRECT preventing other users from mmap'ing these
255 * files.  "Protected" environments where stricter consistency guarantees are
256 * required can disable this knob.  The process that opened the file O_DIRECT
257 * cannot mmap() the file, because mmap'ed IO on an O_DIRECT open() is not
258 * meaningful.
259 */
260int newnfs_directio_allow_mmap = 1;
261SYSCTL_INT(_vfs_nfs, OID_AUTO, nfs_directio_allow_mmap, CTLFLAG_RW,
262	   &newnfs_directio_allow_mmap, 0, "Enable mmaped IO on file with O_DIRECT opens");
263
264#if 0
265SYSCTL_INT(_vfs_nfs, OID_AUTO, access_cache_hits, CTLFLAG_RD,
266	   &newnfsstats.accesscache_hits, 0, "NFS ACCESS cache hit count");
267
268SYSCTL_INT(_vfs_nfs, OID_AUTO, access_cache_misses, CTLFLAG_RD,
269	   &newnfsstats.accesscache_misses, 0, "NFS ACCESS cache miss count");
270#endif
271
272#define	NFSACCESS_ALL (NFSACCESS_READ | NFSACCESS_MODIFY		\
273			 | NFSACCESS_EXTEND | NFSACCESS_EXECUTE	\
274			 | NFSACCESS_DELETE | NFSACCESS_LOOKUP)
275
276/*
277 * SMP Locking Note :
278 * The list of locks after the description of the lock is the ordering
279 * of other locks acquired with the lock held.
280 * np->n_mtx : Protects the fields in the nfsnode.
281       VM Object Lock
282       VI_MTX (acquired indirectly)
283 * nmp->nm_mtx : Protects the fields in the nfsmount.
284       rep->r_mtx
285 * ncl_iod_mutex : Global lock, protects shared nfsiod state.
286 * nfs_reqq_mtx : Global lock, protects the nfs_reqq list.
287       nmp->nm_mtx
288       rep->r_mtx
289 * rep->r_mtx : Protects the fields in an nfsreq.
290 */
291
292static int
293nfs34_access_otw(struct vnode *vp, int wmode, struct thread *td,
294    struct ucred *cred, u_int32_t *retmode)
295{
296	int error = 0, attrflag, i, lrupos;
297	u_int32_t rmode;
298	struct nfsnode *np = VTONFS(vp);
299	struct nfsvattr nfsva;
300
301	error = nfsrpc_accessrpc(vp, wmode, cred, td, &nfsva, &attrflag,
302	    &rmode, NULL);
303	if (attrflag)
304		(void) nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0, 1);
305	if (!error) {
306		lrupos = 0;
307		mtx_lock(&np->n_mtx);
308		for (i = 0; i < NFS_ACCESSCACHESIZE; i++) {
309			if (np->n_accesscache[i].uid == cred->cr_uid) {
310				np->n_accesscache[i].mode = rmode;
311				np->n_accesscache[i].stamp = time_second;
312				break;
313			}
314			if (i > 0 && np->n_accesscache[i].stamp <
315			    np->n_accesscache[lrupos].stamp)
316				lrupos = i;
317		}
318		if (i == NFS_ACCESSCACHESIZE) {
319			np->n_accesscache[lrupos].uid = cred->cr_uid;
320			np->n_accesscache[lrupos].mode = rmode;
321			np->n_accesscache[lrupos].stamp = time_second;
322		}
323		mtx_unlock(&np->n_mtx);
324		if (retmode != NULL)
325			*retmode = rmode;
326		KDTRACE_NFS_ACCESSCACHE_LOAD_DONE(vp, cred->cr_uid, rmode, 0);
327	} else if (NFS_ISV4(vp)) {
328		error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0);
329	}
330#ifdef KDTRACE_HOOKS
331	if (error != 0)
332		KDTRACE_NFS_ACCESSCACHE_LOAD_DONE(vp, cred->cr_uid, 0,
333		    error);
334#endif
335	return (error);
336}
337
338/*
339 * nfs access vnode op.
340 * For nfs version 2, just return ok. File accesses may fail later.
341 * For nfs version 3, use the access rpc to check accessibility. If file modes
342 * are changed on the server, accesses might still fail later.
343 */
344static int
345nfs_access(struct vop_access_args *ap)
346{
347	struct vnode *vp = ap->a_vp;
348	int error = 0, i, gotahit;
349	u_int32_t mode, wmode, rmode;
350	int v34 = NFS_ISV34(vp);
351	struct nfsnode *np = VTONFS(vp);
352
353	/*
354	 * Disallow write attempts on filesystems mounted read-only;
355	 * unless the file is a socket, fifo, or a block or character
356	 * device resident on the filesystem.
357	 */
358	if ((ap->a_accmode & (VWRITE | VAPPEND | VWRITE_NAMED_ATTRS |
359	    VDELETE_CHILD | VWRITE_ATTRIBUTES | VDELETE | VWRITE_ACL |
360	    VWRITE_OWNER)) != 0 && (vp->v_mount->mnt_flag & MNT_RDONLY) != 0) {
361		switch (vp->v_type) {
362		case VREG:
363		case VDIR:
364		case VLNK:
365			return (EROFS);
366		default:
367			break;
368		}
369	}
370	/*
371	 * For nfs v3 or v4, check to see if we have done this recently, and if
372	 * so return our cached result instead of making an ACCESS call.
373	 * If not, do an access rpc, otherwise you are stuck emulating
374	 * ufs_access() locally using the vattr. This may not be correct,
375	 * since the server may apply other access criteria such as
376	 * client uid-->server uid mapping that we do not know about.
377	 */
378	if (v34) {
379		if (ap->a_accmode & VREAD)
380			mode = NFSACCESS_READ;
381		else
382			mode = 0;
383		if (vp->v_type != VDIR) {
384			if (ap->a_accmode & VWRITE)
385				mode |= (NFSACCESS_MODIFY | NFSACCESS_EXTEND);
386			if (ap->a_accmode & VAPPEND)
387				mode |= NFSACCESS_EXTEND;
388			if (ap->a_accmode & VEXEC)
389				mode |= NFSACCESS_EXECUTE;
390			if (ap->a_accmode & VDELETE)
391				mode |= NFSACCESS_DELETE;
392		} else {
393			if (ap->a_accmode & VWRITE)
394				mode |= (NFSACCESS_MODIFY | NFSACCESS_EXTEND);
395			if (ap->a_accmode & VAPPEND)
396				mode |= NFSACCESS_EXTEND;
397			if (ap->a_accmode & VEXEC)
398				mode |= NFSACCESS_LOOKUP;
399			if (ap->a_accmode & VDELETE)
400				mode |= NFSACCESS_DELETE;
401			if (ap->a_accmode & VDELETE_CHILD)
402				mode |= NFSACCESS_MODIFY;
403		}
404		/* XXX safety belt, only make blanket request if caching */
405		if (nfsaccess_cache_timeout > 0) {
406			wmode = NFSACCESS_READ | NFSACCESS_MODIFY |
407				NFSACCESS_EXTEND | NFSACCESS_EXECUTE |
408				NFSACCESS_DELETE | NFSACCESS_LOOKUP;
409		} else {
410			wmode = mode;
411		}
412
413		/*
414		 * Does our cached result allow us to give a definite yes to
415		 * this request?
416		 */
417		gotahit = 0;
418		mtx_lock(&np->n_mtx);
419		for (i = 0; i < NFS_ACCESSCACHESIZE; i++) {
420			if (ap->a_cred->cr_uid == np->n_accesscache[i].uid) {
421			    if (time_second < (np->n_accesscache[i].stamp
422				+ nfsaccess_cache_timeout) &&
423				(np->n_accesscache[i].mode & mode) == mode) {
424				NFSINCRGLOBAL(newnfsstats.accesscache_hits);
425				gotahit = 1;
426			    }
427			    break;
428			}
429		}
430		mtx_unlock(&np->n_mtx);
431#ifdef KDTRACE_HOOKS
432		if (gotahit != 0)
433			KDTRACE_NFS_ACCESSCACHE_GET_HIT(vp,
434			    ap->a_cred->cr_uid, mode);
435		else
436			KDTRACE_NFS_ACCESSCACHE_GET_MISS(vp,
437			    ap->a_cred->cr_uid, mode);
438#endif
439		if (gotahit == 0) {
440			/*
441			 * Either a no, or a don't know.  Go to the wire.
442			 */
443			NFSINCRGLOBAL(newnfsstats.accesscache_misses);
444		        error = nfs34_access_otw(vp, wmode, ap->a_td,
445			    ap->a_cred, &rmode);
446			if (!error &&
447			    (rmode & mode) != mode)
448				error = EACCES;
449		}
450		return (error);
451	} else {
452		if ((error = nfsspec_access(ap)) != 0) {
453			return (error);
454		}
455		/*
456		 * Attempt to prevent a mapped root from accessing a file
457		 * which it shouldn't.  We try to read a byte from the file
458		 * if the user is root and the file is not zero length.
459		 * After calling nfsspec_access, we should have the correct
460		 * file size cached.
461		 */
462		mtx_lock(&np->n_mtx);
463		if (ap->a_cred->cr_uid == 0 && (ap->a_accmode & VREAD)
464		    && VTONFS(vp)->n_size > 0) {
465			struct iovec aiov;
466			struct uio auio;
467			char buf[1];
468
469			mtx_unlock(&np->n_mtx);
470			aiov.iov_base = buf;
471			aiov.iov_len = 1;
472			auio.uio_iov = &aiov;
473			auio.uio_iovcnt = 1;
474			auio.uio_offset = 0;
475			auio.uio_resid = 1;
476			auio.uio_segflg = UIO_SYSSPACE;
477			auio.uio_rw = UIO_READ;
478			auio.uio_td = ap->a_td;
479
480			if (vp->v_type == VREG)
481				error = ncl_readrpc(vp, &auio, ap->a_cred);
482			else if (vp->v_type == VDIR) {
483				char* bp;
484				bp = malloc(NFS_DIRBLKSIZ, M_TEMP, M_WAITOK);
485				aiov.iov_base = bp;
486				aiov.iov_len = auio.uio_resid = NFS_DIRBLKSIZ;
487				error = ncl_readdirrpc(vp, &auio, ap->a_cred,
488				    ap->a_td);
489				free(bp, M_TEMP);
490			} else if (vp->v_type == VLNK)
491				error = ncl_readlinkrpc(vp, &auio, ap->a_cred);
492			else
493				error = EACCES;
494		} else
495			mtx_unlock(&np->n_mtx);
496		return (error);
497	}
498}
499
500
501/*
502 * nfs open vnode op
503 * Check to see if the type is ok
504 * and that deletion is not in progress.
505 * For paged in text files, you will need to flush the page cache
506 * if consistency is lost.
507 */
508/* ARGSUSED */
509static int
510nfs_open(struct vop_open_args *ap)
511{
512	struct vnode *vp = ap->a_vp;
513	struct nfsnode *np = VTONFS(vp);
514	struct vattr vattr;
515	int error;
516	int fmode = ap->a_mode;
517	struct ucred *cred;
518
519	if (vp->v_type != VREG && vp->v_type != VDIR && vp->v_type != VLNK)
520		return (EOPNOTSUPP);
521
522	/*
523	 * For NFSv4, we need to do the Open Op before cache validation,
524	 * so that we conform to RFC3530 Sec. 9.3.1.
525	 */
526	if (NFS_ISV4(vp)) {
527		error = nfsrpc_open(vp, fmode, ap->a_cred, ap->a_td);
528		if (error) {
529			error = nfscl_maperr(ap->a_td, error, (uid_t)0,
530			    (gid_t)0);
531			return (error);
532		}
533	}
534
535	/*
536	 * Now, if this Open will be doing reading, re-validate/flush the
537	 * cache, so that Close/Open coherency is maintained.
538	 */
539	mtx_lock(&np->n_mtx);
540	if (np->n_flag & NMODIFIED) {
541		mtx_unlock(&np->n_mtx);
542		error = ncl_vinvalbuf(vp, V_SAVE, ap->a_td, 1);
543		if (error == EINTR || error == EIO) {
544			if (NFS_ISV4(vp))
545				(void) nfsrpc_close(vp, 0, ap->a_td);
546			return (error);
547		}
548		mtx_lock(&np->n_mtx);
549		np->n_attrstamp = 0;
550		KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(vp);
551		if (vp->v_type == VDIR)
552			np->n_direofoffset = 0;
553		mtx_unlock(&np->n_mtx);
554		error = VOP_GETATTR(vp, &vattr, ap->a_cred);
555		if (error) {
556			if (NFS_ISV4(vp))
557				(void) nfsrpc_close(vp, 0, ap->a_td);
558			return (error);
559		}
560		mtx_lock(&np->n_mtx);
561		np->n_mtime = vattr.va_mtime;
562		if (NFS_ISV4(vp))
563			np->n_change = vattr.va_filerev;
564	} else {
565		mtx_unlock(&np->n_mtx);
566		error = VOP_GETATTR(vp, &vattr, ap->a_cred);
567		if (error) {
568			if (NFS_ISV4(vp))
569				(void) nfsrpc_close(vp, 0, ap->a_td);
570			return (error);
571		}
572		mtx_lock(&np->n_mtx);
573		if ((NFS_ISV4(vp) && np->n_change != vattr.va_filerev) ||
574		    NFS_TIMESPEC_COMPARE(&np->n_mtime, &vattr.va_mtime)) {
575			if (vp->v_type == VDIR)
576				np->n_direofoffset = 0;
577			mtx_unlock(&np->n_mtx);
578			error = ncl_vinvalbuf(vp, V_SAVE, ap->a_td, 1);
579			if (error == EINTR || error == EIO) {
580				if (NFS_ISV4(vp))
581					(void) nfsrpc_close(vp, 0, ap->a_td);
582				return (error);
583			}
584			mtx_lock(&np->n_mtx);
585			np->n_mtime = vattr.va_mtime;
586			if (NFS_ISV4(vp))
587				np->n_change = vattr.va_filerev;
588		}
589	}
590
591	/*
592	 * If the object has >= 1 O_DIRECT active opens, we disable caching.
593	 */
594	if (newnfs_directio_enable && (fmode & O_DIRECT) &&
595	    (vp->v_type == VREG)) {
596		if (np->n_directio_opens == 0) {
597			mtx_unlock(&np->n_mtx);
598			error = ncl_vinvalbuf(vp, V_SAVE, ap->a_td, 1);
599			if (error) {
600				if (NFS_ISV4(vp))
601					(void) nfsrpc_close(vp, 0, ap->a_td);
602				return (error);
603			}
604			mtx_lock(&np->n_mtx);
605			np->n_flag |= NNONCACHE;
606		}
607		np->n_directio_opens++;
608	}
609
610	/* If opened for writing via NFSv4.1 or later, mark that for pNFS. */
611	if (NFSHASPNFS(VFSTONFS(vp->v_mount)) && (fmode & FWRITE) != 0)
612		np->n_flag |= NWRITEOPENED;
613
614	/*
615	 * If this is an open for writing, capture a reference to the
616	 * credentials, so they can be used by ncl_putpages(). Using
617	 * these write credentials is preferable to the credentials of
618	 * whatever thread happens to be doing the VOP_PUTPAGES() since
619	 * the write RPCs are less likely to fail with EACCES.
620	 */
621	if ((fmode & FWRITE) != 0) {
622		cred = np->n_writecred;
623		np->n_writecred = crhold(ap->a_cred);
624	} else
625		cred = NULL;
626	mtx_unlock(&np->n_mtx);
627
628	if (cred != NULL)
629		crfree(cred);
630	vnode_create_vobject(vp, vattr.va_size, ap->a_td);
631	return (0);
632}
633
634/*
635 * nfs close vnode op
636 * What an NFS client should do upon close after writing is a debatable issue.
637 * Most NFS clients push delayed writes to the server upon close, basically for
638 * two reasons:
639 * 1 - So that any write errors may be reported back to the client process
640 *     doing the close system call. By far the two most likely errors are
641 *     NFSERR_NOSPC and NFSERR_DQUOT to indicate space allocation failure.
642 * 2 - To put a worst case upper bound on cache inconsistency between
643 *     multiple clients for the file.
644 * There is also a consistency problem for Version 2 of the protocol w.r.t.
645 * not being able to tell if other clients are writing a file concurrently,
646 * since there is no way of knowing if the changed modify time in the reply
647 * is only due to the write for this client.
648 * (NFS Version 3 provides weak cache consistency data in the reply that
649 *  should be sufficient to detect and handle this case.)
650 *
651 * The current code does the following:
652 * for NFS Version 2 - play it safe and flush/invalidate all dirty buffers
653 * for NFS Version 3 - flush dirty buffers to the server but don't invalidate
654 *                     or commit them (this satisfies 1 and 2 except for the
655 *                     case where the server crashes after this close but
656 *                     before the commit RPC, which is felt to be "good
657 *                     enough". Changing the last argument to ncl_flush() to
658 *                     a 1 would force a commit operation, if it is felt a
659 *                     commit is necessary now.
660 * for NFS Version 4 - flush the dirty buffers and commit them, if
661 *		       nfscl_mustflush() says this is necessary.
662 *                     It is necessary if there is no write delegation held,
663 *                     in order to satisfy open/close coherency.
664 *                     If the file isn't cached on local stable storage,
665 *                     it may be necessary in order to detect "out of space"
666 *                     errors from the server, if the write delegation
667 *                     issued by the server doesn't allow the file to grow.
668 */
669/* ARGSUSED */
670static int
671nfs_close(struct vop_close_args *ap)
672{
673	struct vnode *vp = ap->a_vp;
674	struct nfsnode *np = VTONFS(vp);
675	struct nfsvattr nfsva;
676	struct ucred *cred;
677	int error = 0, ret, localcred = 0;
678	int fmode = ap->a_fflag;
679
680	if ((vp->v_mount->mnt_kern_flag & MNTK_UNMOUNTF))
681		return (0);
682	/*
683	 * During shutdown, a_cred isn't valid, so just use root.
684	 */
685	if (ap->a_cred == NOCRED) {
686		cred = newnfs_getcred();
687		localcred = 1;
688	} else {
689		cred = ap->a_cred;
690	}
691	if (vp->v_type == VREG) {
692	    /*
693	     * Examine and clean dirty pages, regardless of NMODIFIED.
694	     * This closes a major hole in close-to-open consistency.
695	     * We want to push out all dirty pages (and buffers) on
696	     * close, regardless of whether they were dirtied by
697	     * mmap'ed writes or via write().
698	     */
699	    if (nfs_clean_pages_on_close && vp->v_object) {
700		VM_OBJECT_WLOCK(vp->v_object);
701		vm_object_page_clean(vp->v_object, 0, 0, 0);
702		VM_OBJECT_WUNLOCK(vp->v_object);
703	    }
704	    mtx_lock(&np->n_mtx);
705	    if (np->n_flag & NMODIFIED) {
706		mtx_unlock(&np->n_mtx);
707		if (NFS_ISV3(vp)) {
708		    /*
709		     * Under NFSv3 we have dirty buffers to dispose of.  We
710		     * must flush them to the NFS server.  We have the option
711		     * of waiting all the way through the commit rpc or just
712		     * waiting for the initial write.  The default is to only
713		     * wait through the initial write so the data is in the
714		     * server's cache, which is roughly similar to the state
715		     * a standard disk subsystem leaves the file in on close().
716		     *
717		     * We cannot clear the NMODIFIED bit in np->n_flag due to
718		     * potential races with other processes, and certainly
719		     * cannot clear it if we don't commit.
720		     * These races occur when there is no longer the old
721		     * traditional vnode locking implemented for Vnode Ops.
722		     */
723		    int cm = newnfs_commit_on_close ? 1 : 0;
724		    error = ncl_flush(vp, MNT_WAIT, ap->a_td, cm, 0);
725		    /* np->n_flag &= ~NMODIFIED; */
726		} else if (NFS_ISV4(vp)) {
727			if (nfscl_mustflush(vp) != 0) {
728				int cm = newnfs_commit_on_close ? 1 : 0;
729				error = ncl_flush(vp, MNT_WAIT, ap->a_td,
730				    cm, 0);
731				/*
732				 * as above w.r.t races when clearing
733				 * NMODIFIED.
734				 * np->n_flag &= ~NMODIFIED;
735				 */
736			}
737		} else
738		    error = ncl_vinvalbuf(vp, V_SAVE, ap->a_td, 1);
739		mtx_lock(&np->n_mtx);
740	    }
741 	    /*
742 	     * Invalidate the attribute cache in all cases.
743 	     * An open is going to fetch fresh attrs any way, other procs
744 	     * on this node that have file open will be forced to do an
745 	     * otw attr fetch, but this is safe.
746	     * --> A user found that their RPC count dropped by 20% when
747	     *     this was commented out and I can't see any requirement
748	     *     for it, so I've disabled it when negative lookups are
749	     *     enabled. (What does this have to do with negative lookup
750	     *     caching? Well nothing, except it was reported by the
751	     *     same user that needed negative lookup caching and I wanted
752	     *     there to be a way to disable it to see if it
753	     *     is the cause of some caching/coherency issue that might
754	     *     crop up.)
755 	     */
756	    if (VFSTONFS(vp->v_mount)->nm_negnametimeo == 0) {
757		    np->n_attrstamp = 0;
758		    KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(vp);
759	    }
760	    if (np->n_flag & NWRITEERR) {
761		np->n_flag &= ~NWRITEERR;
762		error = np->n_error;
763	    }
764	    mtx_unlock(&np->n_mtx);
765	}
766
767	if (NFS_ISV4(vp)) {
768		/*
769		 * Get attributes so "change" is up to date.
770		 */
771		if (error == 0 && nfscl_mustflush(vp) != 0 &&
772		    vp->v_type == VREG &&
773		    (VFSTONFS(vp->v_mount)->nm_flag & NFSMNT_NOCTO) == 0) {
774			ret = nfsrpc_getattr(vp, cred, ap->a_td, &nfsva,
775			    NULL);
776			if (!ret) {
777				np->n_change = nfsva.na_filerev;
778				(void) nfscl_loadattrcache(&vp, &nfsva, NULL,
779				    NULL, 0, 0);
780			}
781		}
782
783		/*
784		 * and do the close.
785		 */
786		ret = nfsrpc_close(vp, 0, ap->a_td);
787		if (!error && ret)
788			error = ret;
789		if (error)
790			error = nfscl_maperr(ap->a_td, error, (uid_t)0,
791			    (gid_t)0);
792	}
793	if (newnfs_directio_enable)
794		KASSERT((np->n_directio_asyncwr == 0),
795			("nfs_close: dirty unflushed (%d) directio buffers\n",
796			 np->n_directio_asyncwr));
797	if (newnfs_directio_enable && (fmode & O_DIRECT) && (vp->v_type == VREG)) {
798		mtx_lock(&np->n_mtx);
799		KASSERT((np->n_directio_opens > 0),
800			("nfs_close: unexpectedly value (0) of n_directio_opens\n"));
801		np->n_directio_opens--;
802		if (np->n_directio_opens == 0)
803			np->n_flag &= ~NNONCACHE;
804		mtx_unlock(&np->n_mtx);
805	}
806	if (localcred)
807		NFSFREECRED(cred);
808	return (error);
809}
810
811/*
812 * nfs getattr call from vfs.
813 */
814static int
815nfs_getattr(struct vop_getattr_args *ap)
816{
817	struct vnode *vp = ap->a_vp;
818	struct thread *td = curthread;	/* XXX */
819	struct nfsnode *np = VTONFS(vp);
820	int error = 0;
821	struct nfsvattr nfsva;
822	struct vattr *vap = ap->a_vap;
823	struct vattr vattr;
824
825	/*
826	 * Update local times for special files.
827	 */
828	mtx_lock(&np->n_mtx);
829	if (np->n_flag & (NACC | NUPD))
830		np->n_flag |= NCHG;
831	mtx_unlock(&np->n_mtx);
832	/*
833	 * First look in the cache.
834	 */
835	if (ncl_getattrcache(vp, &vattr) == 0) {
836		vap->va_type = vattr.va_type;
837		vap->va_mode = vattr.va_mode;
838		vap->va_nlink = vattr.va_nlink;
839		vap->va_uid = vattr.va_uid;
840		vap->va_gid = vattr.va_gid;
841		vap->va_fsid = vattr.va_fsid;
842		vap->va_fileid = vattr.va_fileid;
843		vap->va_size = vattr.va_size;
844		vap->va_blocksize = vattr.va_blocksize;
845		vap->va_atime = vattr.va_atime;
846		vap->va_mtime = vattr.va_mtime;
847		vap->va_ctime = vattr.va_ctime;
848		vap->va_gen = vattr.va_gen;
849		vap->va_flags = vattr.va_flags;
850		vap->va_rdev = vattr.va_rdev;
851		vap->va_bytes = vattr.va_bytes;
852		vap->va_filerev = vattr.va_filerev;
853		/*
854		 * Get the local modify time for the case of a write
855		 * delegation.
856		 */
857		nfscl_deleggetmodtime(vp, &vap->va_mtime);
858		return (0);
859	}
860
861	if (NFS_ISV34(vp) && nfs_prime_access_cache &&
862	    nfsaccess_cache_timeout > 0) {
863		NFSINCRGLOBAL(newnfsstats.accesscache_misses);
864		nfs34_access_otw(vp, NFSACCESS_ALL, td, ap->a_cred, NULL);
865		if (ncl_getattrcache(vp, ap->a_vap) == 0) {
866			nfscl_deleggetmodtime(vp, &ap->a_vap->va_mtime);
867			return (0);
868		}
869	}
870	error = nfsrpc_getattr(vp, ap->a_cred, td, &nfsva, NULL);
871	if (!error)
872		error = nfscl_loadattrcache(&vp, &nfsva, vap, NULL, 0, 0);
873	if (!error) {
874		/*
875		 * Get the local modify time for the case of a write
876		 * delegation.
877		 */
878		nfscl_deleggetmodtime(vp, &vap->va_mtime);
879	} else if (NFS_ISV4(vp)) {
880		error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0);
881	}
882	return (error);
883}
884
885/*
886 * nfs setattr call.
887 */
888static int
889nfs_setattr(struct vop_setattr_args *ap)
890{
891	struct vnode *vp = ap->a_vp;
892	struct nfsnode *np = VTONFS(vp);
893	struct thread *td = curthread;	/* XXX */
894	struct vattr *vap = ap->a_vap;
895	int error = 0;
896	u_quad_t tsize;
897
898#ifndef nolint
899	tsize = (u_quad_t)0;
900#endif
901
902	/*
903	 * Setting of flags and marking of atimes are not supported.
904	 */
905	if (vap->va_flags != VNOVAL)
906		return (EOPNOTSUPP);
907
908	/*
909	 * Disallow write attempts if the filesystem is mounted read-only.
910	 */
911  	if ((vap->va_flags != VNOVAL || vap->va_uid != (uid_t)VNOVAL ||
912	    vap->va_gid != (gid_t)VNOVAL || vap->va_atime.tv_sec != VNOVAL ||
913	    vap->va_mtime.tv_sec != VNOVAL || vap->va_mode != (mode_t)VNOVAL) &&
914	    (vp->v_mount->mnt_flag & MNT_RDONLY))
915		return (EROFS);
916	if (vap->va_size != VNOVAL) {
917 		switch (vp->v_type) {
918 		case VDIR:
919 			return (EISDIR);
920 		case VCHR:
921 		case VBLK:
922 		case VSOCK:
923 		case VFIFO:
924			if (vap->va_mtime.tv_sec == VNOVAL &&
925			    vap->va_atime.tv_sec == VNOVAL &&
926			    vap->va_mode == (mode_t)VNOVAL &&
927			    vap->va_uid == (uid_t)VNOVAL &&
928			    vap->va_gid == (gid_t)VNOVAL)
929				return (0);
930 			vap->va_size = VNOVAL;
931 			break;
932 		default:
933			/*
934			 * Disallow write attempts if the filesystem is
935			 * mounted read-only.
936			 */
937			if (vp->v_mount->mnt_flag & MNT_RDONLY)
938				return (EROFS);
939			/*
940			 *  We run vnode_pager_setsize() early (why?),
941			 * we must set np->n_size now to avoid vinvalbuf
942			 * V_SAVE races that might setsize a lower
943			 * value.
944			 */
945			mtx_lock(&np->n_mtx);
946			tsize = np->n_size;
947			mtx_unlock(&np->n_mtx);
948			error = ncl_meta_setsize(vp, ap->a_cred, td,
949			    vap->va_size);
950			mtx_lock(&np->n_mtx);
951 			if (np->n_flag & NMODIFIED) {
952			    tsize = np->n_size;
953			    mtx_unlock(&np->n_mtx);
954 			    if (vap->va_size == 0)
955 				error = ncl_vinvalbuf(vp, 0, td, 1);
956 			    else
957 				error = ncl_vinvalbuf(vp, V_SAVE, td, 1);
958 			    if (error) {
959				vnode_pager_setsize(vp, tsize);
960				return (error);
961			    }
962			    /*
963			     * Call nfscl_delegmodtime() to set the modify time
964			     * locally, as required.
965			     */
966			    nfscl_delegmodtime(vp);
967 			} else
968			    mtx_unlock(&np->n_mtx);
969			/*
970			 * np->n_size has already been set to vap->va_size
971			 * in ncl_meta_setsize(). We must set it again since
972			 * nfs_loadattrcache() could be called through
973			 * ncl_meta_setsize() and could modify np->n_size.
974			 */
975			mtx_lock(&np->n_mtx);
976 			np->n_vattr.na_size = np->n_size = vap->va_size;
977			mtx_unlock(&np->n_mtx);
978  		};
979  	} else {
980		mtx_lock(&np->n_mtx);
981		if ((vap->va_mtime.tv_sec != VNOVAL || vap->va_atime.tv_sec != VNOVAL) &&
982		    (np->n_flag & NMODIFIED) && vp->v_type == VREG) {
983			mtx_unlock(&np->n_mtx);
984			if ((error = ncl_vinvalbuf(vp, V_SAVE, td, 1)) != 0 &&
985			    (error == EINTR || error == EIO))
986				return (error);
987		} else
988			mtx_unlock(&np->n_mtx);
989	}
990	error = nfs_setattrrpc(vp, vap, ap->a_cred, td);
991	if (error && vap->va_size != VNOVAL) {
992		mtx_lock(&np->n_mtx);
993		np->n_size = np->n_vattr.na_size = tsize;
994		vnode_pager_setsize(vp, tsize);
995		mtx_unlock(&np->n_mtx);
996	}
997	return (error);
998}
999
1000/*
1001 * Do an nfs setattr rpc.
1002 */
1003static int
1004nfs_setattrrpc(struct vnode *vp, struct vattr *vap, struct ucred *cred,
1005    struct thread *td)
1006{
1007	struct nfsnode *np = VTONFS(vp);
1008	int error, ret, attrflag, i;
1009	struct nfsvattr nfsva;
1010
1011	if (NFS_ISV34(vp)) {
1012		mtx_lock(&np->n_mtx);
1013		for (i = 0; i < NFS_ACCESSCACHESIZE; i++)
1014			np->n_accesscache[i].stamp = 0;
1015		np->n_flag |= NDELEGMOD;
1016		mtx_unlock(&np->n_mtx);
1017		KDTRACE_NFS_ACCESSCACHE_FLUSH_DONE(vp);
1018	}
1019	error = nfsrpc_setattr(vp, vap, NULL, cred, td, &nfsva, &attrflag,
1020	    NULL);
1021	if (attrflag) {
1022		ret = nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0, 1);
1023		if (ret && !error)
1024			error = ret;
1025	}
1026	if (error && NFS_ISV4(vp))
1027		error = nfscl_maperr(td, error, vap->va_uid, vap->va_gid);
1028	return (error);
1029}
1030
1031/*
1032 * nfs lookup call, one step at a time...
1033 * First look in cache
1034 * If not found, unlock the directory nfsnode and do the rpc
1035 */
1036static int
1037nfs_lookup(struct vop_lookup_args *ap)
1038{
1039	struct componentname *cnp = ap->a_cnp;
1040	struct vnode *dvp = ap->a_dvp;
1041	struct vnode **vpp = ap->a_vpp;
1042	struct mount *mp = dvp->v_mount;
1043	int flags = cnp->cn_flags;
1044	struct vnode *newvp;
1045	struct nfsmount *nmp;
1046	struct nfsnode *np, *newnp;
1047	int error = 0, attrflag, dattrflag, ltype, ncticks;
1048	struct thread *td = cnp->cn_thread;
1049	struct nfsfh *nfhp;
1050	struct nfsvattr dnfsva, nfsva;
1051	struct vattr vattr;
1052	struct timespec nctime;
1053
1054	*vpp = NULLVP;
1055	if ((flags & ISLASTCN) && (mp->mnt_flag & MNT_RDONLY) &&
1056	    (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME))
1057		return (EROFS);
1058	if (dvp->v_type != VDIR)
1059		return (ENOTDIR);
1060	nmp = VFSTONFS(mp);
1061	np = VTONFS(dvp);
1062
1063	/* For NFSv4, wait until any remove is done. */
1064	mtx_lock(&np->n_mtx);
1065	while (NFSHASNFSV4(nmp) && (np->n_flag & NREMOVEINPROG)) {
1066		np->n_flag |= NREMOVEWANT;
1067		(void) msleep((caddr_t)np, &np->n_mtx, PZERO, "nfslkup", 0);
1068	}
1069	mtx_unlock(&np->n_mtx);
1070
1071	if ((error = VOP_ACCESS(dvp, VEXEC, cnp->cn_cred, td)) != 0)
1072		return (error);
1073	error = cache_lookup(dvp, vpp, cnp, &nctime, &ncticks);
1074	if (error > 0 && error != ENOENT)
1075		return (error);
1076	if (error == -1) {
1077		/*
1078		 * Lookups of "." are special and always return the
1079		 * current directory.  cache_lookup() already handles
1080		 * associated locking bookkeeping, etc.
1081		 */
1082		if (cnp->cn_namelen == 1 && cnp->cn_nameptr[0] == '.') {
1083			/* XXX: Is this really correct? */
1084			if (cnp->cn_nameiop != LOOKUP &&
1085			    (flags & ISLASTCN))
1086				cnp->cn_flags |= SAVENAME;
1087			return (0);
1088		}
1089
1090		/*
1091		 * We only accept a positive hit in the cache if the
1092		 * change time of the file matches our cached copy.
1093		 * Otherwise, we discard the cache entry and fallback
1094		 * to doing a lookup RPC.  We also only trust cache
1095		 * entries for less than nm_nametimeo seconds.
1096		 *
1097		 * To better handle stale file handles and attributes,
1098		 * clear the attribute cache of this node if it is a
1099		 * leaf component, part of an open() call, and not
1100		 * locally modified before fetching the attributes.
1101		 * This should allow stale file handles to be detected
1102		 * here where we can fall back to a LOOKUP RPC to
1103		 * recover rather than having nfs_open() detect the
1104		 * stale file handle and failing open(2) with ESTALE.
1105		 */
1106		newvp = *vpp;
1107		newnp = VTONFS(newvp);
1108		if (!(nmp->nm_flag & NFSMNT_NOCTO) &&
1109		    (flags & (ISLASTCN | ISOPEN)) == (ISLASTCN | ISOPEN) &&
1110		    !(newnp->n_flag & NMODIFIED)) {
1111			mtx_lock(&newnp->n_mtx);
1112			newnp->n_attrstamp = 0;
1113			KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(newvp);
1114			mtx_unlock(&newnp->n_mtx);
1115		}
1116		if (nfscl_nodeleg(newvp, 0) == 0 ||
1117		    ((u_int)(ticks - ncticks) < (nmp->nm_nametimeo * hz) &&
1118		    VOP_GETATTR(newvp, &vattr, cnp->cn_cred) == 0 &&
1119		    timespeccmp(&vattr.va_ctime, &nctime, ==))) {
1120			NFSINCRGLOBAL(newnfsstats.lookupcache_hits);
1121			if (cnp->cn_nameiop != LOOKUP &&
1122			    (flags & ISLASTCN))
1123				cnp->cn_flags |= SAVENAME;
1124			return (0);
1125		}
1126		cache_purge(newvp);
1127		if (dvp != newvp)
1128			vput(newvp);
1129		else
1130			vrele(newvp);
1131		*vpp = NULLVP;
1132	} else if (error == ENOENT) {
1133		if (dvp->v_iflag & VI_DOOMED)
1134			return (ENOENT);
1135		/*
1136		 * We only accept a negative hit in the cache if the
1137		 * modification time of the parent directory matches
1138		 * the cached copy in the name cache entry.
1139		 * Otherwise, we discard all of the negative cache
1140		 * entries for this directory.  We also only trust
1141		 * negative cache entries for up to nm_negnametimeo
1142		 * seconds.
1143		 */
1144		if ((u_int)(ticks - ncticks) < (nmp->nm_negnametimeo * hz) &&
1145		    VOP_GETATTR(dvp, &vattr, cnp->cn_cred) == 0 &&
1146		    timespeccmp(&vattr.va_mtime, &nctime, ==)) {
1147			NFSINCRGLOBAL(newnfsstats.lookupcache_hits);
1148			return (ENOENT);
1149		}
1150		cache_purge_negative(dvp);
1151	}
1152
1153	error = 0;
1154	newvp = NULLVP;
1155	NFSINCRGLOBAL(newnfsstats.lookupcache_misses);
1156	error = nfsrpc_lookup(dvp, cnp->cn_nameptr, cnp->cn_namelen,
1157	    cnp->cn_cred, td, &dnfsva, &nfsva, &nfhp, &attrflag, &dattrflag,
1158	    NULL);
1159	if (dattrflag)
1160		(void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, NULL, 0, 1);
1161	if (error) {
1162		if (newvp != NULLVP) {
1163			vput(newvp);
1164			*vpp = NULLVP;
1165		}
1166
1167		if (error != ENOENT) {
1168			if (NFS_ISV4(dvp))
1169				error = nfscl_maperr(td, error, (uid_t)0,
1170				    (gid_t)0);
1171			return (error);
1172		}
1173
1174		/* The requested file was not found. */
1175		if ((cnp->cn_nameiop == CREATE || cnp->cn_nameiop == RENAME) &&
1176		    (flags & ISLASTCN)) {
1177			/*
1178			 * XXX: UFS does a full VOP_ACCESS(dvp,
1179			 * VWRITE) here instead of just checking
1180			 * MNT_RDONLY.
1181			 */
1182			if (mp->mnt_flag & MNT_RDONLY)
1183				return (EROFS);
1184			cnp->cn_flags |= SAVENAME;
1185			return (EJUSTRETURN);
1186		}
1187
1188		if ((cnp->cn_flags & MAKEENTRY) != 0 && dattrflag) {
1189			/*
1190			 * Cache the modification time of the parent
1191			 * directory from the post-op attributes in
1192			 * the name cache entry.  The negative cache
1193			 * entry will be ignored once the directory
1194			 * has changed.  Don't bother adding the entry
1195			 * if the directory has already changed.
1196			 */
1197			mtx_lock(&np->n_mtx);
1198			if (timespeccmp(&np->n_vattr.na_mtime,
1199			    &dnfsva.na_mtime, ==)) {
1200				mtx_unlock(&np->n_mtx);
1201				cache_enter_time(dvp, NULL, cnp,
1202				    &dnfsva.na_mtime, NULL);
1203			} else
1204				mtx_unlock(&np->n_mtx);
1205		}
1206		return (ENOENT);
1207	}
1208
1209	/*
1210	 * Handle RENAME case...
1211	 */
1212	if (cnp->cn_nameiop == RENAME && (flags & ISLASTCN)) {
1213		if (NFS_CMPFH(np, nfhp->nfh_fh, nfhp->nfh_len)) {
1214			FREE((caddr_t)nfhp, M_NFSFH);
1215			return (EISDIR);
1216		}
1217		error = nfscl_nget(mp, dvp, nfhp, cnp, td, &np, NULL,
1218		    LK_EXCLUSIVE);
1219		if (error)
1220			return (error);
1221		newvp = NFSTOV(np);
1222		if (attrflag)
1223			(void) nfscl_loadattrcache(&newvp, &nfsva, NULL, NULL,
1224			    0, 1);
1225		*vpp = newvp;
1226		cnp->cn_flags |= SAVENAME;
1227		return (0);
1228	}
1229
1230	if (flags & ISDOTDOT) {
1231		ltype = NFSVOPISLOCKED(dvp);
1232		error = vfs_busy(mp, MBF_NOWAIT);
1233		if (error != 0) {
1234			vfs_ref(mp);
1235			NFSVOPUNLOCK(dvp, 0);
1236			error = vfs_busy(mp, 0);
1237			NFSVOPLOCK(dvp, ltype | LK_RETRY);
1238			vfs_rel(mp);
1239			if (error == 0 && (dvp->v_iflag & VI_DOOMED)) {
1240				vfs_unbusy(mp);
1241				error = ENOENT;
1242			}
1243			if (error != 0)
1244				return (error);
1245		}
1246		NFSVOPUNLOCK(dvp, 0);
1247		error = nfscl_nget(mp, dvp, nfhp, cnp, td, &np, NULL,
1248		    cnp->cn_lkflags);
1249		if (error == 0)
1250			newvp = NFSTOV(np);
1251		vfs_unbusy(mp);
1252		if (newvp != dvp)
1253			NFSVOPLOCK(dvp, ltype | LK_RETRY);
1254		if (dvp->v_iflag & VI_DOOMED) {
1255			if (error == 0) {
1256				if (newvp == dvp)
1257					vrele(newvp);
1258				else
1259					vput(newvp);
1260			}
1261			error = ENOENT;
1262		}
1263		if (error != 0)
1264			return (error);
1265		if (attrflag)
1266			(void) nfscl_loadattrcache(&newvp, &nfsva, NULL, NULL,
1267			    0, 1);
1268	} else if (NFS_CMPFH(np, nfhp->nfh_fh, nfhp->nfh_len)) {
1269		FREE((caddr_t)nfhp, M_NFSFH);
1270		VREF(dvp);
1271		newvp = dvp;
1272		if (attrflag)
1273			(void) nfscl_loadattrcache(&newvp, &nfsva, NULL, NULL,
1274			    0, 1);
1275	} else {
1276		error = nfscl_nget(mp, dvp, nfhp, cnp, td, &np, NULL,
1277		    cnp->cn_lkflags);
1278		if (error)
1279			return (error);
1280		newvp = NFSTOV(np);
1281		if (attrflag)
1282			(void) nfscl_loadattrcache(&newvp, &nfsva, NULL, NULL,
1283			    0, 1);
1284		else if ((flags & (ISLASTCN | ISOPEN)) == (ISLASTCN | ISOPEN) &&
1285		    !(np->n_flag & NMODIFIED)) {
1286			/*
1287			 * Flush the attribute cache when opening a
1288			 * leaf node to ensure that fresh attributes
1289			 * are fetched in nfs_open() since we did not
1290			 * fetch attributes from the LOOKUP reply.
1291			 */
1292			mtx_lock(&np->n_mtx);
1293			np->n_attrstamp = 0;
1294			KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(newvp);
1295			mtx_unlock(&np->n_mtx);
1296		}
1297	}
1298	if (cnp->cn_nameiop != LOOKUP && (flags & ISLASTCN))
1299		cnp->cn_flags |= SAVENAME;
1300	if ((cnp->cn_flags & MAKEENTRY) &&
1301	    (cnp->cn_nameiop != DELETE || !(flags & ISLASTCN)) &&
1302	    attrflag != 0 && (newvp->v_type != VDIR || dattrflag != 0))
1303		cache_enter_time(dvp, newvp, cnp, &nfsva.na_ctime,
1304		    newvp->v_type != VDIR ? NULL : &dnfsva.na_ctime);
1305	*vpp = newvp;
1306	return (0);
1307}
1308
1309/*
1310 * nfs read call.
1311 * Just call ncl_bioread() to do the work.
1312 */
1313static int
1314nfs_read(struct vop_read_args *ap)
1315{
1316	struct vnode *vp = ap->a_vp;
1317
1318	switch (vp->v_type) {
1319	case VREG:
1320		return (ncl_bioread(vp, ap->a_uio, ap->a_ioflag, ap->a_cred));
1321	case VDIR:
1322		return (EISDIR);
1323	default:
1324		return (EOPNOTSUPP);
1325	}
1326}
1327
1328/*
1329 * nfs readlink call
1330 */
1331static int
1332nfs_readlink(struct vop_readlink_args *ap)
1333{
1334	struct vnode *vp = ap->a_vp;
1335
1336	if (vp->v_type != VLNK)
1337		return (EINVAL);
1338	return (ncl_bioread(vp, ap->a_uio, 0, ap->a_cred));
1339}
1340
1341/*
1342 * Do a readlink rpc.
1343 * Called by ncl_doio() from below the buffer cache.
1344 */
1345int
1346ncl_readlinkrpc(struct vnode *vp, struct uio *uiop, struct ucred *cred)
1347{
1348	int error, ret, attrflag;
1349	struct nfsvattr nfsva;
1350
1351	error = nfsrpc_readlink(vp, uiop, cred, uiop->uio_td, &nfsva,
1352	    &attrflag, NULL);
1353	if (attrflag) {
1354		ret = nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0, 1);
1355		if (ret && !error)
1356			error = ret;
1357	}
1358	if (error && NFS_ISV4(vp))
1359		error = nfscl_maperr(uiop->uio_td, error, (uid_t)0, (gid_t)0);
1360	return (error);
1361}
1362
1363/*
1364 * nfs read rpc call
1365 * Ditto above
1366 */
1367int
1368ncl_readrpc(struct vnode *vp, struct uio *uiop, struct ucred *cred)
1369{
1370	int error, ret, attrflag;
1371	struct nfsvattr nfsva;
1372	struct nfsmount *nmp;
1373
1374	nmp = VFSTONFS(vnode_mount(vp));
1375	error = EIO;
1376	attrflag = 0;
1377	if (NFSHASPNFS(nmp))
1378		error = nfscl_doiods(vp, uiop, NULL, NULL,
1379		    NFSV4OPEN_ACCESSREAD, cred, uiop->uio_td);
1380	NFSCL_DEBUG(4, "readrpc: aft doiods=%d\n", error);
1381	if (error != 0)
1382		error = nfsrpc_read(vp, uiop, cred, uiop->uio_td, &nfsva,
1383		    &attrflag, NULL);
1384	if (attrflag) {
1385		ret = nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0, 1);
1386		if (ret && !error)
1387			error = ret;
1388	}
1389	if (error && NFS_ISV4(vp))
1390		error = nfscl_maperr(uiop->uio_td, error, (uid_t)0, (gid_t)0);
1391	return (error);
1392}
1393
1394/*
1395 * nfs write call
1396 */
1397int
1398ncl_writerpc(struct vnode *vp, struct uio *uiop, struct ucred *cred,
1399    int *iomode, int *must_commit, int called_from_strategy)
1400{
1401	struct nfsvattr nfsva;
1402	int error, attrflag, ret;
1403	struct nfsmount *nmp;
1404
1405	nmp = VFSTONFS(vnode_mount(vp));
1406	error = EIO;
1407	attrflag = 0;
1408	if (NFSHASPNFS(nmp))
1409		error = nfscl_doiods(vp, uiop, iomode, must_commit,
1410		    NFSV4OPEN_ACCESSWRITE, cred, uiop->uio_td);
1411	NFSCL_DEBUG(4, "writerpc: aft doiods=%d\n", error);
1412	if (error != 0)
1413		error = nfsrpc_write(vp, uiop, iomode, must_commit, cred,
1414		    uiop->uio_td, &nfsva, &attrflag, NULL,
1415		    called_from_strategy);
1416	if (attrflag) {
1417		if (VTONFS(vp)->n_flag & ND_NFSV4)
1418			ret = nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 1,
1419			    1);
1420		else
1421			ret = nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0,
1422			    1);
1423		if (ret && !error)
1424			error = ret;
1425	}
1426	if (DOINGASYNC(vp))
1427		*iomode = NFSWRITE_FILESYNC;
1428	if (error && NFS_ISV4(vp))
1429		error = nfscl_maperr(uiop->uio_td, error, (uid_t)0, (gid_t)0);
1430	return (error);
1431}
1432
1433/*
1434 * nfs mknod rpc
1435 * For NFS v2 this is a kludge. Use a create rpc but with the IFMT bits of the
1436 * mode set to specify the file type and the size field for rdev.
1437 */
1438static int
1439nfs_mknodrpc(struct vnode *dvp, struct vnode **vpp, struct componentname *cnp,
1440    struct vattr *vap)
1441{
1442	struct nfsvattr nfsva, dnfsva;
1443	struct vnode *newvp = NULL;
1444	struct nfsnode *np = NULL, *dnp;
1445	struct nfsfh *nfhp;
1446	struct vattr vattr;
1447	int error = 0, attrflag, dattrflag;
1448	u_int32_t rdev;
1449
1450	if (vap->va_type == VCHR || vap->va_type == VBLK)
1451		rdev = vap->va_rdev;
1452	else if (vap->va_type == VFIFO || vap->va_type == VSOCK)
1453		rdev = 0xffffffff;
1454	else
1455		return (EOPNOTSUPP);
1456	if ((error = VOP_GETATTR(dvp, &vattr, cnp->cn_cred)))
1457		return (error);
1458	error = nfsrpc_mknod(dvp, cnp->cn_nameptr, cnp->cn_namelen, vap,
1459	    rdev, vap->va_type, cnp->cn_cred, cnp->cn_thread, &dnfsva,
1460	    &nfsva, &nfhp, &attrflag, &dattrflag, NULL);
1461	if (!error) {
1462		if (!nfhp)
1463			(void) nfsrpc_lookup(dvp, cnp->cn_nameptr,
1464			    cnp->cn_namelen, cnp->cn_cred, cnp->cn_thread,
1465			    &dnfsva, &nfsva, &nfhp, &attrflag, &dattrflag,
1466			    NULL);
1467		if (nfhp)
1468			error = nfscl_nget(dvp->v_mount, dvp, nfhp, cnp,
1469			    cnp->cn_thread, &np, NULL, LK_EXCLUSIVE);
1470	}
1471	if (dattrflag)
1472		(void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, NULL, 0, 1);
1473	if (!error) {
1474		newvp = NFSTOV(np);
1475		if (attrflag != 0) {
1476			error = nfscl_loadattrcache(&newvp, &nfsva, NULL, NULL,
1477			    0, 1);
1478			if (error != 0)
1479				vput(newvp);
1480		}
1481	}
1482	if (!error) {
1483		*vpp = newvp;
1484	} else if (NFS_ISV4(dvp)) {
1485		error = nfscl_maperr(cnp->cn_thread, error, vap->va_uid,
1486		    vap->va_gid);
1487	}
1488	dnp = VTONFS(dvp);
1489	mtx_lock(&dnp->n_mtx);
1490	dnp->n_flag |= NMODIFIED;
1491	if (!dattrflag) {
1492		dnp->n_attrstamp = 0;
1493		KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(dvp);
1494	}
1495	mtx_unlock(&dnp->n_mtx);
1496	return (error);
1497}
1498
1499/*
1500 * nfs mknod vop
1501 * just call nfs_mknodrpc() to do the work.
1502 */
1503/* ARGSUSED */
1504static int
1505nfs_mknod(struct vop_mknod_args *ap)
1506{
1507	return (nfs_mknodrpc(ap->a_dvp, ap->a_vpp, ap->a_cnp, ap->a_vap));
1508}
1509
1510static struct mtx nfs_cverf_mtx;
1511MTX_SYSINIT(nfs_cverf_mtx, &nfs_cverf_mtx, "NFS create verifier mutex",
1512    MTX_DEF);
1513
1514static nfsquad_t
1515nfs_get_cverf(void)
1516{
1517	static nfsquad_t cverf;
1518	nfsquad_t ret;
1519	static int cverf_initialized = 0;
1520
1521	mtx_lock(&nfs_cverf_mtx);
1522	if (cverf_initialized == 0) {
1523		cverf.lval[0] = arc4random();
1524		cverf.lval[1] = arc4random();
1525		cverf_initialized = 1;
1526	} else
1527		cverf.qval++;
1528	ret = cverf;
1529	mtx_unlock(&nfs_cverf_mtx);
1530
1531	return (ret);
1532}
1533
1534/*
1535 * nfs file create call
1536 */
1537static int
1538nfs_create(struct vop_create_args *ap)
1539{
1540	struct vnode *dvp = ap->a_dvp;
1541	struct vattr *vap = ap->a_vap;
1542	struct componentname *cnp = ap->a_cnp;
1543	struct nfsnode *np = NULL, *dnp;
1544	struct vnode *newvp = NULL;
1545	struct nfsmount *nmp;
1546	struct nfsvattr dnfsva, nfsva;
1547	struct nfsfh *nfhp;
1548	nfsquad_t cverf;
1549	int error = 0, attrflag, dattrflag, fmode = 0;
1550	struct vattr vattr;
1551
1552	/*
1553	 * Oops, not for me..
1554	 */
1555	if (vap->va_type == VSOCK)
1556		return (nfs_mknodrpc(dvp, ap->a_vpp, cnp, vap));
1557
1558	if ((error = VOP_GETATTR(dvp, &vattr, cnp->cn_cred)))
1559		return (error);
1560	if (vap->va_vaflags & VA_EXCLUSIVE)
1561		fmode |= O_EXCL;
1562	dnp = VTONFS(dvp);
1563	nmp = VFSTONFS(vnode_mount(dvp));
1564again:
1565	/* For NFSv4, wait until any remove is done. */
1566	mtx_lock(&dnp->n_mtx);
1567	while (NFSHASNFSV4(nmp) && (dnp->n_flag & NREMOVEINPROG)) {
1568		dnp->n_flag |= NREMOVEWANT;
1569		(void) msleep((caddr_t)dnp, &dnp->n_mtx, PZERO, "nfscrt", 0);
1570	}
1571	mtx_unlock(&dnp->n_mtx);
1572
1573	cverf = nfs_get_cverf();
1574	error = nfsrpc_create(dvp, cnp->cn_nameptr, cnp->cn_namelen,
1575	    vap, cverf, fmode, cnp->cn_cred, cnp->cn_thread, &dnfsva, &nfsva,
1576	    &nfhp, &attrflag, &dattrflag, NULL);
1577	if (!error) {
1578		if (nfhp == NULL)
1579			(void) nfsrpc_lookup(dvp, cnp->cn_nameptr,
1580			    cnp->cn_namelen, cnp->cn_cred, cnp->cn_thread,
1581			    &dnfsva, &nfsva, &nfhp, &attrflag, &dattrflag,
1582			    NULL);
1583		if (nfhp != NULL)
1584			error = nfscl_nget(dvp->v_mount, dvp, nfhp, cnp,
1585			    cnp->cn_thread, &np, NULL, LK_EXCLUSIVE);
1586	}
1587	if (dattrflag)
1588		(void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, NULL, 0, 1);
1589	if (!error) {
1590		newvp = NFSTOV(np);
1591		if (attrflag == 0)
1592			error = nfsrpc_getattr(newvp, cnp->cn_cred,
1593			    cnp->cn_thread, &nfsva, NULL);
1594		if (error == 0)
1595			error = nfscl_loadattrcache(&newvp, &nfsva, NULL, NULL,
1596			    0, 1);
1597	}
1598	if (error) {
1599		if (newvp != NULL) {
1600			vput(newvp);
1601			newvp = NULL;
1602		}
1603		if (NFS_ISV34(dvp) && (fmode & O_EXCL) &&
1604		    error == NFSERR_NOTSUPP) {
1605			fmode &= ~O_EXCL;
1606			goto again;
1607		}
1608	} else if (NFS_ISV34(dvp) && (fmode & O_EXCL)) {
1609		if (nfscl_checksattr(vap, &nfsva)) {
1610			error = nfsrpc_setattr(newvp, vap, NULL, cnp->cn_cred,
1611			    cnp->cn_thread, &nfsva, &attrflag, NULL);
1612			if (error && (vap->va_uid != (uid_t)VNOVAL ||
1613			    vap->va_gid != (gid_t)VNOVAL)) {
1614				/* try again without setting uid/gid */
1615				vap->va_uid = (uid_t)VNOVAL;
1616				vap->va_gid = (uid_t)VNOVAL;
1617				error = nfsrpc_setattr(newvp, vap, NULL,
1618				    cnp->cn_cred, cnp->cn_thread, &nfsva,
1619				    &attrflag, NULL);
1620			}
1621			if (attrflag)
1622				(void) nfscl_loadattrcache(&newvp, &nfsva, NULL,
1623				    NULL, 0, 1);
1624			if (error != 0)
1625				vput(newvp);
1626		}
1627	}
1628	if (!error) {
1629		if ((cnp->cn_flags & MAKEENTRY) && attrflag)
1630			cache_enter_time(dvp, newvp, cnp, &nfsva.na_ctime,
1631			    NULL);
1632		*ap->a_vpp = newvp;
1633	} else if (NFS_ISV4(dvp)) {
1634		error = nfscl_maperr(cnp->cn_thread, error, vap->va_uid,
1635		    vap->va_gid);
1636	}
1637	mtx_lock(&dnp->n_mtx);
1638	dnp->n_flag |= NMODIFIED;
1639	if (!dattrflag) {
1640		dnp->n_attrstamp = 0;
1641		KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(dvp);
1642	}
1643	mtx_unlock(&dnp->n_mtx);
1644	return (error);
1645}
1646
1647/*
1648 * nfs file remove call
1649 * To try and make nfs semantics closer to ufs semantics, a file that has
1650 * other processes using the vnode is renamed instead of removed and then
1651 * removed later on the last close.
1652 * - If v_usecount > 1
1653 *	  If a rename is not already in the works
1654 *	     call nfs_sillyrename() to set it up
1655 *     else
1656 *	  do the remove rpc
1657 */
1658static int
1659nfs_remove(struct vop_remove_args *ap)
1660{
1661	struct vnode *vp = ap->a_vp;
1662	struct vnode *dvp = ap->a_dvp;
1663	struct componentname *cnp = ap->a_cnp;
1664	struct nfsnode *np = VTONFS(vp);
1665	int error = 0;
1666	struct vattr vattr;
1667
1668	KASSERT((cnp->cn_flags & HASBUF) != 0, ("nfs_remove: no name"));
1669	KASSERT(vrefcnt(vp) > 0, ("nfs_remove: bad v_usecount"));
1670	if (vp->v_type == VDIR)
1671		error = EPERM;
1672	else if (vrefcnt(vp) == 1 || (np->n_sillyrename &&
1673	    VOP_GETATTR(vp, &vattr, cnp->cn_cred) == 0 &&
1674	    vattr.va_nlink > 1)) {
1675		/*
1676		 * Purge the name cache so that the chance of a lookup for
1677		 * the name succeeding while the remove is in progress is
1678		 * minimized. Without node locking it can still happen, such
1679		 * that an I/O op returns ESTALE, but since you get this if
1680		 * another host removes the file..
1681		 */
1682		cache_purge(vp);
1683		/*
1684		 * throw away biocache buffers, mainly to avoid
1685		 * unnecessary delayed writes later.
1686		 */
1687		error = ncl_vinvalbuf(vp, 0, cnp->cn_thread, 1);
1688		/* Do the rpc */
1689		if (error != EINTR && error != EIO)
1690			error = nfs_removerpc(dvp, vp, cnp->cn_nameptr,
1691			    cnp->cn_namelen, cnp->cn_cred, cnp->cn_thread);
1692		/*
1693		 * Kludge City: If the first reply to the remove rpc is lost..
1694		 *   the reply to the retransmitted request will be ENOENT
1695		 *   since the file was in fact removed
1696		 *   Therefore, we cheat and return success.
1697		 */
1698		if (error == ENOENT)
1699			error = 0;
1700	} else if (!np->n_sillyrename)
1701		error = nfs_sillyrename(dvp, vp, cnp);
1702	mtx_lock(&np->n_mtx);
1703	np->n_attrstamp = 0;
1704	mtx_unlock(&np->n_mtx);
1705	KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(vp);
1706	return (error);
1707}
1708
1709/*
1710 * nfs file remove rpc called from nfs_inactive
1711 */
1712int
1713ncl_removeit(struct sillyrename *sp, struct vnode *vp)
1714{
1715	/*
1716	 * Make sure that the directory vnode is still valid.
1717	 * XXX we should lock sp->s_dvp here.
1718	 */
1719	if (sp->s_dvp->v_type == VBAD)
1720		return (0);
1721	return (nfs_removerpc(sp->s_dvp, vp, sp->s_name, sp->s_namlen,
1722	    sp->s_cred, NULL));
1723}
1724
1725/*
1726 * Nfs remove rpc, called from nfs_remove() and ncl_removeit().
1727 */
1728static int
1729nfs_removerpc(struct vnode *dvp, struct vnode *vp, char *name,
1730    int namelen, struct ucred *cred, struct thread *td)
1731{
1732	struct nfsvattr dnfsva;
1733	struct nfsnode *dnp = VTONFS(dvp);
1734	int error = 0, dattrflag;
1735
1736	mtx_lock(&dnp->n_mtx);
1737	dnp->n_flag |= NREMOVEINPROG;
1738	mtx_unlock(&dnp->n_mtx);
1739	error = nfsrpc_remove(dvp, name, namelen, vp, cred, td, &dnfsva,
1740	    &dattrflag, NULL);
1741	mtx_lock(&dnp->n_mtx);
1742	if ((dnp->n_flag & NREMOVEWANT)) {
1743		dnp->n_flag &= ~(NREMOVEWANT | NREMOVEINPROG);
1744		mtx_unlock(&dnp->n_mtx);
1745		wakeup((caddr_t)dnp);
1746	} else {
1747		dnp->n_flag &= ~NREMOVEINPROG;
1748		mtx_unlock(&dnp->n_mtx);
1749	}
1750	if (dattrflag)
1751		(void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, NULL, 0, 1);
1752	mtx_lock(&dnp->n_mtx);
1753	dnp->n_flag |= NMODIFIED;
1754	if (!dattrflag) {
1755		dnp->n_attrstamp = 0;
1756		KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(dvp);
1757	}
1758	mtx_unlock(&dnp->n_mtx);
1759	if (error && NFS_ISV4(dvp))
1760		error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0);
1761	return (error);
1762}
1763
1764/*
1765 * nfs file rename call
1766 */
1767static int
1768nfs_rename(struct vop_rename_args *ap)
1769{
1770	struct vnode *fvp = ap->a_fvp;
1771	struct vnode *tvp = ap->a_tvp;
1772	struct vnode *fdvp = ap->a_fdvp;
1773	struct vnode *tdvp = ap->a_tdvp;
1774	struct componentname *tcnp = ap->a_tcnp;
1775	struct componentname *fcnp = ap->a_fcnp;
1776	struct nfsnode *fnp = VTONFS(ap->a_fvp);
1777	struct nfsnode *tdnp = VTONFS(ap->a_tdvp);
1778	struct nfsv4node *newv4 = NULL;
1779	int error;
1780
1781	KASSERT((tcnp->cn_flags & HASBUF) != 0 &&
1782	    (fcnp->cn_flags & HASBUF) != 0, ("nfs_rename: no name"));
1783	/* Check for cross-device rename */
1784	if ((fvp->v_mount != tdvp->v_mount) ||
1785	    (tvp && (fvp->v_mount != tvp->v_mount))) {
1786		error = EXDEV;
1787		goto out;
1788	}
1789
1790	if (fvp == tvp) {
1791		printf("nfs_rename: fvp == tvp (can't happen)\n");
1792		error = 0;
1793		goto out;
1794	}
1795	if ((error = NFSVOPLOCK(fvp, LK_EXCLUSIVE)) != 0)
1796		goto out;
1797
1798	/*
1799	 * We have to flush B_DELWRI data prior to renaming
1800	 * the file.  If we don't, the delayed-write buffers
1801	 * can be flushed out later after the file has gone stale
1802	 * under NFSV3.  NFSV2 does not have this problem because
1803	 * ( as far as I can tell ) it flushes dirty buffers more
1804	 * often.
1805	 *
1806	 * Skip the rename operation if the fsync fails, this can happen
1807	 * due to the server's volume being full, when we pushed out data
1808	 * that was written back to our cache earlier. Not checking for
1809	 * this condition can result in potential (silent) data loss.
1810	 */
1811	error = VOP_FSYNC(fvp, MNT_WAIT, fcnp->cn_thread);
1812	NFSVOPUNLOCK(fvp, 0);
1813	if (!error && tvp)
1814		error = VOP_FSYNC(tvp, MNT_WAIT, tcnp->cn_thread);
1815	if (error)
1816		goto out;
1817
1818	/*
1819	 * If the tvp exists and is in use, sillyrename it before doing the
1820	 * rename of the new file over it.
1821	 * XXX Can't sillyrename a directory.
1822	 */
1823	if (tvp && vrefcnt(tvp) > 1 && !VTONFS(tvp)->n_sillyrename &&
1824		tvp->v_type != VDIR && !nfs_sillyrename(tdvp, tvp, tcnp)) {
1825		vput(tvp);
1826		tvp = NULL;
1827	}
1828
1829	error = nfs_renamerpc(fdvp, fvp, fcnp->cn_nameptr, fcnp->cn_namelen,
1830	    tdvp, tvp, tcnp->cn_nameptr, tcnp->cn_namelen, tcnp->cn_cred,
1831	    tcnp->cn_thread);
1832
1833	if (error == 0 && NFS_ISV4(tdvp)) {
1834		/*
1835		 * For NFSv4, check to see if it is the same name and
1836		 * replace the name, if it is different.
1837		 */
1838		MALLOC(newv4, struct nfsv4node *,
1839		    sizeof (struct nfsv4node) +
1840		    tdnp->n_fhp->nfh_len + tcnp->cn_namelen - 1,
1841		    M_NFSV4NODE, M_WAITOK);
1842		mtx_lock(&tdnp->n_mtx);
1843		mtx_lock(&fnp->n_mtx);
1844		if (fnp->n_v4 != NULL && fvp->v_type == VREG &&
1845		    (fnp->n_v4->n4_namelen != tcnp->cn_namelen ||
1846		      NFSBCMP(tcnp->cn_nameptr, NFS4NODENAME(fnp->n_v4),
1847		      tcnp->cn_namelen) ||
1848		      tdnp->n_fhp->nfh_len != fnp->n_v4->n4_fhlen ||
1849		      NFSBCMP(tdnp->n_fhp->nfh_fh, fnp->n_v4->n4_data,
1850			tdnp->n_fhp->nfh_len))) {
1851#ifdef notdef
1852{ char nnn[100]; int nnnl;
1853nnnl = (tcnp->cn_namelen < 100) ? tcnp->cn_namelen : 99;
1854bcopy(tcnp->cn_nameptr, nnn, nnnl);
1855nnn[nnnl] = '\0';
1856printf("ren replace=%s\n",nnn);
1857}
1858#endif
1859			FREE((caddr_t)fnp->n_v4, M_NFSV4NODE);
1860			fnp->n_v4 = newv4;
1861			newv4 = NULL;
1862			fnp->n_v4->n4_fhlen = tdnp->n_fhp->nfh_len;
1863			fnp->n_v4->n4_namelen = tcnp->cn_namelen;
1864			NFSBCOPY(tdnp->n_fhp->nfh_fh, fnp->n_v4->n4_data,
1865			    tdnp->n_fhp->nfh_len);
1866			NFSBCOPY(tcnp->cn_nameptr,
1867			    NFS4NODENAME(fnp->n_v4), tcnp->cn_namelen);
1868		}
1869		mtx_unlock(&tdnp->n_mtx);
1870		mtx_unlock(&fnp->n_mtx);
1871		if (newv4 != NULL)
1872			FREE((caddr_t)newv4, M_NFSV4NODE);
1873	}
1874
1875	if (fvp->v_type == VDIR) {
1876		if (tvp != NULL && tvp->v_type == VDIR)
1877			cache_purge(tdvp);
1878		cache_purge(fdvp);
1879	}
1880
1881out:
1882	if (tdvp == tvp)
1883		vrele(tdvp);
1884	else
1885		vput(tdvp);
1886	if (tvp)
1887		vput(tvp);
1888	vrele(fdvp);
1889	vrele(fvp);
1890	/*
1891	 * Kludge: Map ENOENT => 0 assuming that it is a reply to a retry.
1892	 */
1893	if (error == ENOENT)
1894		error = 0;
1895	return (error);
1896}
1897
1898/*
1899 * nfs file rename rpc called from nfs_remove() above
1900 */
1901static int
1902nfs_renameit(struct vnode *sdvp, struct vnode *svp, struct componentname *scnp,
1903    struct sillyrename *sp)
1904{
1905
1906	return (nfs_renamerpc(sdvp, svp, scnp->cn_nameptr, scnp->cn_namelen,
1907	    sdvp, NULL, sp->s_name, sp->s_namlen, scnp->cn_cred,
1908	    scnp->cn_thread));
1909}
1910
1911/*
1912 * Do an nfs rename rpc. Called from nfs_rename() and nfs_renameit().
1913 */
1914static int
1915nfs_renamerpc(struct vnode *fdvp, struct vnode *fvp, char *fnameptr,
1916    int fnamelen, struct vnode *tdvp, struct vnode *tvp, char *tnameptr,
1917    int tnamelen, struct ucred *cred, struct thread *td)
1918{
1919	struct nfsvattr fnfsva, tnfsva;
1920	struct nfsnode *fdnp = VTONFS(fdvp);
1921	struct nfsnode *tdnp = VTONFS(tdvp);
1922	int error = 0, fattrflag, tattrflag;
1923
1924	error = nfsrpc_rename(fdvp, fvp, fnameptr, fnamelen, tdvp, tvp,
1925	    tnameptr, tnamelen, cred, td, &fnfsva, &tnfsva, &fattrflag,
1926	    &tattrflag, NULL, NULL);
1927	mtx_lock(&fdnp->n_mtx);
1928	fdnp->n_flag |= NMODIFIED;
1929	if (fattrflag != 0) {
1930		mtx_unlock(&fdnp->n_mtx);
1931		(void) nfscl_loadattrcache(&fdvp, &fnfsva, NULL, NULL, 0, 1);
1932	} else {
1933		fdnp->n_attrstamp = 0;
1934		mtx_unlock(&fdnp->n_mtx);
1935		KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(fdvp);
1936	}
1937	mtx_lock(&tdnp->n_mtx);
1938	tdnp->n_flag |= NMODIFIED;
1939	if (tattrflag != 0) {
1940		mtx_unlock(&tdnp->n_mtx);
1941		(void) nfscl_loadattrcache(&tdvp, &tnfsva, NULL, NULL, 0, 1);
1942	} else {
1943		tdnp->n_attrstamp = 0;
1944		mtx_unlock(&tdnp->n_mtx);
1945		KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(tdvp);
1946	}
1947	if (error && NFS_ISV4(fdvp))
1948		error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0);
1949	return (error);
1950}
1951
1952/*
1953 * nfs hard link create call
1954 */
1955static int
1956nfs_link(struct vop_link_args *ap)
1957{
1958	struct vnode *vp = ap->a_vp;
1959	struct vnode *tdvp = ap->a_tdvp;
1960	struct componentname *cnp = ap->a_cnp;
1961	struct nfsnode *np, *tdnp;
1962	struct nfsvattr nfsva, dnfsva;
1963	int error = 0, attrflag, dattrflag;
1964
1965	/*
1966	 * Push all writes to the server, so that the attribute cache
1967	 * doesn't get "out of sync" with the server.
1968	 * XXX There should be a better way!
1969	 */
1970	VOP_FSYNC(vp, MNT_WAIT, cnp->cn_thread);
1971
1972	error = nfsrpc_link(tdvp, vp, cnp->cn_nameptr, cnp->cn_namelen,
1973	    cnp->cn_cred, cnp->cn_thread, &dnfsva, &nfsva, &attrflag,
1974	    &dattrflag, NULL);
1975	tdnp = VTONFS(tdvp);
1976	mtx_lock(&tdnp->n_mtx);
1977	tdnp->n_flag |= NMODIFIED;
1978	if (dattrflag != 0) {
1979		mtx_unlock(&tdnp->n_mtx);
1980		(void) nfscl_loadattrcache(&tdvp, &dnfsva, NULL, NULL, 0, 1);
1981	} else {
1982		tdnp->n_attrstamp = 0;
1983		mtx_unlock(&tdnp->n_mtx);
1984		KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(tdvp);
1985	}
1986	if (attrflag)
1987		(void) nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0, 1);
1988	else {
1989		np = VTONFS(vp);
1990		mtx_lock(&np->n_mtx);
1991		np->n_attrstamp = 0;
1992		mtx_unlock(&np->n_mtx);
1993		KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(vp);
1994	}
1995	/*
1996	 * If negative lookup caching is enabled, I might as well
1997	 * add an entry for this node. Not necessary for correctness,
1998	 * but if negative caching is enabled, then the system
1999	 * must care about lookup caching hit rate, so...
2000	 */
2001	if (VFSTONFS(vp->v_mount)->nm_negnametimeo != 0 &&
2002	    (cnp->cn_flags & MAKEENTRY) && attrflag != 0 && error == 0) {
2003		cache_enter_time(tdvp, vp, cnp, &nfsva.na_ctime, NULL);
2004	}
2005	if (error && NFS_ISV4(vp))
2006		error = nfscl_maperr(cnp->cn_thread, error, (uid_t)0,
2007		    (gid_t)0);
2008	return (error);
2009}
2010
2011/*
2012 * nfs symbolic link create call
2013 */
2014static int
2015nfs_symlink(struct vop_symlink_args *ap)
2016{
2017	struct vnode *dvp = ap->a_dvp;
2018	struct vattr *vap = ap->a_vap;
2019	struct componentname *cnp = ap->a_cnp;
2020	struct nfsvattr nfsva, dnfsva;
2021	struct nfsfh *nfhp;
2022	struct nfsnode *np = NULL, *dnp;
2023	struct vnode *newvp = NULL;
2024	int error = 0, attrflag, dattrflag, ret;
2025
2026	vap->va_type = VLNK;
2027	error = nfsrpc_symlink(dvp, cnp->cn_nameptr, cnp->cn_namelen,
2028	    ap->a_target, vap, cnp->cn_cred, cnp->cn_thread, &dnfsva,
2029	    &nfsva, &nfhp, &attrflag, &dattrflag, NULL);
2030	if (nfhp) {
2031		ret = nfscl_nget(dvp->v_mount, dvp, nfhp, cnp, cnp->cn_thread,
2032		    &np, NULL, LK_EXCLUSIVE);
2033		if (!ret)
2034			newvp = NFSTOV(np);
2035		else if (!error)
2036			error = ret;
2037	}
2038	if (newvp != NULL) {
2039		if (attrflag)
2040			(void) nfscl_loadattrcache(&newvp, &nfsva, NULL, NULL,
2041			    0, 1);
2042	} else if (!error) {
2043		/*
2044		 * If we do not have an error and we could not extract the
2045		 * newvp from the response due to the request being NFSv2, we
2046		 * have to do a lookup in order to obtain a newvp to return.
2047		 */
2048		error = nfs_lookitup(dvp, cnp->cn_nameptr, cnp->cn_namelen,
2049		    cnp->cn_cred, cnp->cn_thread, &np);
2050		if (!error)
2051			newvp = NFSTOV(np);
2052	}
2053	if (error) {
2054		if (newvp)
2055			vput(newvp);
2056		if (NFS_ISV4(dvp))
2057			error = nfscl_maperr(cnp->cn_thread, error,
2058			    vap->va_uid, vap->va_gid);
2059	} else {
2060		*ap->a_vpp = newvp;
2061	}
2062
2063	dnp = VTONFS(dvp);
2064	mtx_lock(&dnp->n_mtx);
2065	dnp->n_flag |= NMODIFIED;
2066	if (dattrflag != 0) {
2067		mtx_unlock(&dnp->n_mtx);
2068		(void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, NULL, 0, 1);
2069	} else {
2070		dnp->n_attrstamp = 0;
2071		mtx_unlock(&dnp->n_mtx);
2072		KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(dvp);
2073	}
2074	/*
2075	 * If negative lookup caching is enabled, I might as well
2076	 * add an entry for this node. Not necessary for correctness,
2077	 * but if negative caching is enabled, then the system
2078	 * must care about lookup caching hit rate, so...
2079	 */
2080	if (VFSTONFS(dvp->v_mount)->nm_negnametimeo != 0 &&
2081	    (cnp->cn_flags & MAKEENTRY) && attrflag != 0 && error == 0) {
2082		cache_enter_time(dvp, newvp, cnp, &nfsva.na_ctime, NULL);
2083	}
2084	return (error);
2085}
2086
2087/*
2088 * nfs make dir call
2089 */
2090static int
2091nfs_mkdir(struct vop_mkdir_args *ap)
2092{
2093	struct vnode *dvp = ap->a_dvp;
2094	struct vattr *vap = ap->a_vap;
2095	struct componentname *cnp = ap->a_cnp;
2096	struct nfsnode *np = NULL, *dnp;
2097	struct vnode *newvp = NULL;
2098	struct vattr vattr;
2099	struct nfsfh *nfhp;
2100	struct nfsvattr nfsva, dnfsva;
2101	int error = 0, attrflag, dattrflag, ret;
2102
2103	if ((error = VOP_GETATTR(dvp, &vattr, cnp->cn_cred)) != 0)
2104		return (error);
2105	vap->va_type = VDIR;
2106	error = nfsrpc_mkdir(dvp, cnp->cn_nameptr, cnp->cn_namelen,
2107	    vap, cnp->cn_cred, cnp->cn_thread, &dnfsva, &nfsva, &nfhp,
2108	    &attrflag, &dattrflag, NULL);
2109	dnp = VTONFS(dvp);
2110	mtx_lock(&dnp->n_mtx);
2111	dnp->n_flag |= NMODIFIED;
2112	if (dattrflag != 0) {
2113		mtx_unlock(&dnp->n_mtx);
2114		(void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, NULL, 0, 1);
2115	} else {
2116		dnp->n_attrstamp = 0;
2117		mtx_unlock(&dnp->n_mtx);
2118		KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(dvp);
2119	}
2120	if (nfhp) {
2121		ret = nfscl_nget(dvp->v_mount, dvp, nfhp, cnp, cnp->cn_thread,
2122		    &np, NULL, LK_EXCLUSIVE);
2123		if (!ret) {
2124			newvp = NFSTOV(np);
2125			if (attrflag)
2126			   (void) nfscl_loadattrcache(&newvp, &nfsva, NULL,
2127				NULL, 0, 1);
2128		} else if (!error)
2129			error = ret;
2130	}
2131	if (!error && newvp == NULL) {
2132		error = nfs_lookitup(dvp, cnp->cn_nameptr, cnp->cn_namelen,
2133		    cnp->cn_cred, cnp->cn_thread, &np);
2134		if (!error) {
2135			newvp = NFSTOV(np);
2136			if (newvp->v_type != VDIR)
2137				error = EEXIST;
2138		}
2139	}
2140	if (error) {
2141		if (newvp)
2142			vput(newvp);
2143		if (NFS_ISV4(dvp))
2144			error = nfscl_maperr(cnp->cn_thread, error,
2145			    vap->va_uid, vap->va_gid);
2146	} else {
2147		/*
2148		 * If negative lookup caching is enabled, I might as well
2149		 * add an entry for this node. Not necessary for correctness,
2150		 * but if negative caching is enabled, then the system
2151		 * must care about lookup caching hit rate, so...
2152		 */
2153		if (VFSTONFS(dvp->v_mount)->nm_negnametimeo != 0 &&
2154		    (cnp->cn_flags & MAKEENTRY) &&
2155		    attrflag != 0 && dattrflag != 0)
2156			cache_enter_time(dvp, newvp, cnp, &nfsva.na_ctime,
2157			    &dnfsva.na_ctime);
2158		*ap->a_vpp = newvp;
2159	}
2160	return (error);
2161}
2162
2163/*
2164 * nfs remove directory call
2165 */
2166static int
2167nfs_rmdir(struct vop_rmdir_args *ap)
2168{
2169	struct vnode *vp = ap->a_vp;
2170	struct vnode *dvp = ap->a_dvp;
2171	struct componentname *cnp = ap->a_cnp;
2172	struct nfsnode *dnp;
2173	struct nfsvattr dnfsva;
2174	int error, dattrflag;
2175
2176	if (dvp == vp)
2177		return (EINVAL);
2178	error = nfsrpc_rmdir(dvp, cnp->cn_nameptr, cnp->cn_namelen,
2179	    cnp->cn_cred, cnp->cn_thread, &dnfsva, &dattrflag, NULL);
2180	dnp = VTONFS(dvp);
2181	mtx_lock(&dnp->n_mtx);
2182	dnp->n_flag |= NMODIFIED;
2183	if (dattrflag != 0) {
2184		mtx_unlock(&dnp->n_mtx);
2185		(void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, NULL, 0, 1);
2186	} else {
2187		dnp->n_attrstamp = 0;
2188		mtx_unlock(&dnp->n_mtx);
2189		KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(dvp);
2190	}
2191
2192	cache_purge(dvp);
2193	cache_purge(vp);
2194	if (error && NFS_ISV4(dvp))
2195		error = nfscl_maperr(cnp->cn_thread, error, (uid_t)0,
2196		    (gid_t)0);
2197	/*
2198	 * Kludge: Map ENOENT => 0 assuming that you have a reply to a retry.
2199	 */
2200	if (error == ENOENT)
2201		error = 0;
2202	return (error);
2203}
2204
2205/*
2206 * nfs readdir call
2207 */
2208static int
2209nfs_readdir(struct vop_readdir_args *ap)
2210{
2211	struct vnode *vp = ap->a_vp;
2212	struct nfsnode *np = VTONFS(vp);
2213	struct uio *uio = ap->a_uio;
2214	ssize_t tresid, left;
2215	int error = 0;
2216	struct vattr vattr;
2217
2218	if (ap->a_eofflag != NULL)
2219		*ap->a_eofflag = 0;
2220	if (vp->v_type != VDIR)
2221		return(EPERM);
2222
2223	/*
2224	 * First, check for hit on the EOF offset cache
2225	 */
2226	if (np->n_direofoffset > 0 && uio->uio_offset >= np->n_direofoffset &&
2227	    (np->n_flag & NMODIFIED) == 0) {
2228		if (VOP_GETATTR(vp, &vattr, ap->a_cred) == 0) {
2229			mtx_lock(&np->n_mtx);
2230			if ((NFS_ISV4(vp) && np->n_change == vattr.va_filerev) ||
2231			    !NFS_TIMESPEC_COMPARE(&np->n_mtime, &vattr.va_mtime)) {
2232				mtx_unlock(&np->n_mtx);
2233				NFSINCRGLOBAL(newnfsstats.direofcache_hits);
2234				if (ap->a_eofflag != NULL)
2235					*ap->a_eofflag = 1;
2236				return (0);
2237			} else
2238				mtx_unlock(&np->n_mtx);
2239		}
2240	}
2241
2242	/*
2243	 * NFS always guarantees that directory entries don't straddle
2244	 * DIRBLKSIZ boundaries.  As such, we need to limit the size
2245	 * to an exact multiple of DIRBLKSIZ, to avoid copying a partial
2246	 * directory entry.
2247	 */
2248	left = uio->uio_resid % DIRBLKSIZ;
2249	if (left == uio->uio_resid)
2250		return (EINVAL);
2251	uio->uio_resid -= left;
2252
2253	/*
2254	 * Call ncl_bioread() to do the real work.
2255	 */
2256	tresid = uio->uio_resid;
2257	error = ncl_bioread(vp, uio, 0, ap->a_cred);
2258
2259	if (!error && uio->uio_resid == tresid) {
2260		NFSINCRGLOBAL(newnfsstats.direofcache_misses);
2261		if (ap->a_eofflag != NULL)
2262			*ap->a_eofflag = 1;
2263	}
2264
2265	/* Add the partial DIRBLKSIZ (left) back in. */
2266	uio->uio_resid += left;
2267	return (error);
2268}
2269
2270/*
2271 * Readdir rpc call.
2272 * Called from below the buffer cache by ncl_doio().
2273 */
2274int
2275ncl_readdirrpc(struct vnode *vp, struct uio *uiop, struct ucred *cred,
2276    struct thread *td)
2277{
2278	struct nfsvattr nfsva;
2279	nfsuint64 *cookiep, cookie;
2280	struct nfsnode *dnp = VTONFS(vp);
2281	struct nfsmount *nmp = VFSTONFS(vp->v_mount);
2282	int error = 0, eof, attrflag;
2283
2284	KASSERT(uiop->uio_iovcnt == 1 &&
2285	    (uiop->uio_offset & (DIRBLKSIZ - 1)) == 0 &&
2286	    (uiop->uio_resid & (DIRBLKSIZ - 1)) == 0,
2287	    ("nfs readdirrpc bad uio"));
2288
2289	/*
2290	 * If there is no cookie, assume directory was stale.
2291	 */
2292	ncl_dircookie_lock(dnp);
2293	cookiep = ncl_getcookie(dnp, uiop->uio_offset, 0);
2294	if (cookiep) {
2295		cookie = *cookiep;
2296		ncl_dircookie_unlock(dnp);
2297	} else {
2298		ncl_dircookie_unlock(dnp);
2299		return (NFSERR_BAD_COOKIE);
2300	}
2301
2302	if (NFSHASNFSV3(nmp) && !NFSHASGOTFSINFO(nmp))
2303		(void)ncl_fsinfo(nmp, vp, cred, td);
2304
2305	error = nfsrpc_readdir(vp, uiop, &cookie, cred, td, &nfsva,
2306	    &attrflag, &eof, NULL);
2307	if (attrflag)
2308		(void) nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0, 1);
2309
2310	if (!error) {
2311		/*
2312		 * We are now either at the end of the directory or have filled
2313		 * the block.
2314		 */
2315		if (eof)
2316			dnp->n_direofoffset = uiop->uio_offset;
2317		else {
2318			if (uiop->uio_resid > 0)
2319				printf("EEK! readdirrpc resid > 0\n");
2320			ncl_dircookie_lock(dnp);
2321			cookiep = ncl_getcookie(dnp, uiop->uio_offset, 1);
2322			*cookiep = cookie;
2323			ncl_dircookie_unlock(dnp);
2324		}
2325	} else if (NFS_ISV4(vp)) {
2326		error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0);
2327	}
2328	return (error);
2329}
2330
2331/*
2332 * NFS V3 readdir plus RPC. Used in place of ncl_readdirrpc().
2333 */
2334int
2335ncl_readdirplusrpc(struct vnode *vp, struct uio *uiop, struct ucred *cred,
2336    struct thread *td)
2337{
2338	struct nfsvattr nfsva;
2339	nfsuint64 *cookiep, cookie;
2340	struct nfsnode *dnp = VTONFS(vp);
2341	struct nfsmount *nmp = VFSTONFS(vp->v_mount);
2342	int error = 0, attrflag, eof;
2343
2344	KASSERT(uiop->uio_iovcnt == 1 &&
2345	    (uiop->uio_offset & (DIRBLKSIZ - 1)) == 0 &&
2346	    (uiop->uio_resid & (DIRBLKSIZ - 1)) == 0,
2347	    ("nfs readdirplusrpc bad uio"));
2348
2349	/*
2350	 * If there is no cookie, assume directory was stale.
2351	 */
2352	ncl_dircookie_lock(dnp);
2353	cookiep = ncl_getcookie(dnp, uiop->uio_offset, 0);
2354	if (cookiep) {
2355		cookie = *cookiep;
2356		ncl_dircookie_unlock(dnp);
2357	} else {
2358		ncl_dircookie_unlock(dnp);
2359		return (NFSERR_BAD_COOKIE);
2360	}
2361
2362	if (NFSHASNFSV3(nmp) && !NFSHASGOTFSINFO(nmp))
2363		(void)ncl_fsinfo(nmp, vp, cred, td);
2364	error = nfsrpc_readdirplus(vp, uiop, &cookie, cred, td, &nfsva,
2365	    &attrflag, &eof, NULL);
2366	if (attrflag)
2367		(void) nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0, 1);
2368
2369	if (!error) {
2370		/*
2371		 * We are now either at end of the directory or have filled the
2372		 * the block.
2373		 */
2374		if (eof)
2375			dnp->n_direofoffset = uiop->uio_offset;
2376		else {
2377			if (uiop->uio_resid > 0)
2378				printf("EEK! readdirplusrpc resid > 0\n");
2379			ncl_dircookie_lock(dnp);
2380			cookiep = ncl_getcookie(dnp, uiop->uio_offset, 1);
2381			*cookiep = cookie;
2382			ncl_dircookie_unlock(dnp);
2383		}
2384	} else if (NFS_ISV4(vp)) {
2385		error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0);
2386	}
2387	return (error);
2388}
2389
2390/*
2391 * Silly rename. To make the NFS filesystem that is stateless look a little
2392 * more like the "ufs" a remove of an active vnode is translated to a rename
2393 * to a funny looking filename that is removed by nfs_inactive on the
2394 * nfsnode. There is the potential for another process on a different client
2395 * to create the same funny name between the nfs_lookitup() fails and the
2396 * nfs_rename() completes, but...
2397 */
2398static int
2399nfs_sillyrename(struct vnode *dvp, struct vnode *vp, struct componentname *cnp)
2400{
2401	struct sillyrename *sp;
2402	struct nfsnode *np;
2403	int error;
2404	short pid;
2405	unsigned int lticks;
2406
2407	cache_purge(dvp);
2408	np = VTONFS(vp);
2409	KASSERT(vp->v_type != VDIR, ("nfs: sillyrename dir"));
2410	MALLOC(sp, struct sillyrename *, sizeof (struct sillyrename),
2411	    M_NEWNFSREQ, M_WAITOK);
2412	sp->s_cred = crhold(cnp->cn_cred);
2413	sp->s_dvp = dvp;
2414	VREF(dvp);
2415
2416	/*
2417	 * Fudge together a funny name.
2418	 * Changing the format of the funny name to accomodate more
2419	 * sillynames per directory.
2420	 * The name is now changed to .nfs.<ticks>.<pid>.4, where ticks is
2421	 * CPU ticks since boot.
2422	 */
2423	pid = cnp->cn_thread->td_proc->p_pid;
2424	lticks = (unsigned int)ticks;
2425	for ( ; ; ) {
2426		sp->s_namlen = sprintf(sp->s_name,
2427				       ".nfs.%08x.%04x4.4", lticks,
2428				       pid);
2429		if (nfs_lookitup(dvp, sp->s_name, sp->s_namlen, sp->s_cred,
2430				 cnp->cn_thread, NULL))
2431			break;
2432		lticks++;
2433	}
2434	error = nfs_renameit(dvp, vp, cnp, sp);
2435	if (error)
2436		goto bad;
2437	error = nfs_lookitup(dvp, sp->s_name, sp->s_namlen, sp->s_cred,
2438		cnp->cn_thread, &np);
2439	np->n_sillyrename = sp;
2440	return (0);
2441bad:
2442	vrele(sp->s_dvp);
2443	crfree(sp->s_cred);
2444	free((caddr_t)sp, M_NEWNFSREQ);
2445	return (error);
2446}
2447
2448/*
2449 * Look up a file name and optionally either update the file handle or
2450 * allocate an nfsnode, depending on the value of npp.
2451 * npp == NULL	--> just do the lookup
2452 * *npp == NULL --> allocate a new nfsnode and make sure attributes are
2453 *			handled too
2454 * *npp != NULL --> update the file handle in the vnode
2455 */
2456static int
2457nfs_lookitup(struct vnode *dvp, char *name, int len, struct ucred *cred,
2458    struct thread *td, struct nfsnode **npp)
2459{
2460	struct vnode *newvp = NULL, *vp;
2461	struct nfsnode *np, *dnp = VTONFS(dvp);
2462	struct nfsfh *nfhp, *onfhp;
2463	struct nfsvattr nfsva, dnfsva;
2464	struct componentname cn;
2465	int error = 0, attrflag, dattrflag;
2466	u_int hash;
2467
2468	error = nfsrpc_lookup(dvp, name, len, cred, td, &dnfsva, &nfsva,
2469	    &nfhp, &attrflag, &dattrflag, NULL);
2470	if (dattrflag)
2471		(void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, NULL, 0, 1);
2472	if (npp && !error) {
2473		if (*npp != NULL) {
2474		    np = *npp;
2475		    vp = NFSTOV(np);
2476		    /*
2477		     * For NFSv4, check to see if it is the same name and
2478		     * replace the name, if it is different.
2479		     */
2480		    if (np->n_v4 != NULL && nfsva.na_type == VREG &&
2481			(np->n_v4->n4_namelen != len ||
2482			 NFSBCMP(name, NFS4NODENAME(np->n_v4), len) ||
2483			 dnp->n_fhp->nfh_len != np->n_v4->n4_fhlen ||
2484			 NFSBCMP(dnp->n_fhp->nfh_fh, np->n_v4->n4_data,
2485			 dnp->n_fhp->nfh_len))) {
2486#ifdef notdef
2487{ char nnn[100]; int nnnl;
2488nnnl = (len < 100) ? len : 99;
2489bcopy(name, nnn, nnnl);
2490nnn[nnnl] = '\0';
2491printf("replace=%s\n",nnn);
2492}
2493#endif
2494			    FREE((caddr_t)np->n_v4, M_NFSV4NODE);
2495			    MALLOC(np->n_v4, struct nfsv4node *,
2496				sizeof (struct nfsv4node) +
2497				dnp->n_fhp->nfh_len + len - 1,
2498				M_NFSV4NODE, M_WAITOK);
2499			    np->n_v4->n4_fhlen = dnp->n_fhp->nfh_len;
2500			    np->n_v4->n4_namelen = len;
2501			    NFSBCOPY(dnp->n_fhp->nfh_fh, np->n_v4->n4_data,
2502				dnp->n_fhp->nfh_len);
2503			    NFSBCOPY(name, NFS4NODENAME(np->n_v4), len);
2504		    }
2505		    hash = fnv_32_buf(nfhp->nfh_fh, nfhp->nfh_len,
2506			FNV1_32_INIT);
2507		    onfhp = np->n_fhp;
2508		    /*
2509		     * Rehash node for new file handle.
2510		     */
2511		    vfs_hash_rehash(vp, hash);
2512		    np->n_fhp = nfhp;
2513		    if (onfhp != NULL)
2514			FREE((caddr_t)onfhp, M_NFSFH);
2515		    newvp = NFSTOV(np);
2516		} else if (NFS_CMPFH(dnp, nfhp->nfh_fh, nfhp->nfh_len)) {
2517		    FREE((caddr_t)nfhp, M_NFSFH);
2518		    VREF(dvp);
2519		    newvp = dvp;
2520		} else {
2521		    cn.cn_nameptr = name;
2522		    cn.cn_namelen = len;
2523		    error = nfscl_nget(dvp->v_mount, dvp, nfhp, &cn, td,
2524			&np, NULL, LK_EXCLUSIVE);
2525		    if (error)
2526			return (error);
2527		    newvp = NFSTOV(np);
2528		}
2529		if (!attrflag && *npp == NULL) {
2530			if (newvp == dvp)
2531				vrele(newvp);
2532			else
2533				vput(newvp);
2534			return (ENOENT);
2535		}
2536		if (attrflag)
2537			(void) nfscl_loadattrcache(&newvp, &nfsva, NULL, NULL,
2538			    0, 1);
2539	}
2540	if (npp && *npp == NULL) {
2541		if (error) {
2542			if (newvp) {
2543				if (newvp == dvp)
2544					vrele(newvp);
2545				else
2546					vput(newvp);
2547			}
2548		} else
2549			*npp = np;
2550	}
2551	if (error && NFS_ISV4(dvp))
2552		error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0);
2553	return (error);
2554}
2555
2556/*
2557 * Nfs Version 3 and 4 commit rpc
2558 */
2559int
2560ncl_commit(struct vnode *vp, u_quad_t offset, int cnt, struct ucred *cred,
2561   struct thread *td)
2562{
2563	struct nfsvattr nfsva;
2564	struct nfsmount *nmp = VFSTONFS(vp->v_mount);
2565	int error, attrflag;
2566
2567	mtx_lock(&nmp->nm_mtx);
2568	if ((nmp->nm_state & NFSSTA_HASWRITEVERF) == 0) {
2569		mtx_unlock(&nmp->nm_mtx);
2570		return (0);
2571	}
2572	mtx_unlock(&nmp->nm_mtx);
2573	error = nfsrpc_commit(vp, offset, cnt, cred, td, &nfsva,
2574	    &attrflag, NULL);
2575	if (attrflag != 0)
2576		(void) nfscl_loadattrcache(&vp, &nfsva, NULL, NULL,
2577		    0, 1);
2578	if (error != 0 && NFS_ISV4(vp))
2579		error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0);
2580	return (error);
2581}
2582
2583/*
2584 * Strategy routine.
2585 * For async requests when nfsiod(s) are running, queue the request by
2586 * calling ncl_asyncio(), otherwise just all ncl_doio() to do the
2587 * request.
2588 */
2589static int
2590nfs_strategy(struct vop_strategy_args *ap)
2591{
2592	struct buf *bp = ap->a_bp;
2593	struct ucred *cr;
2594
2595	KASSERT(!(bp->b_flags & B_DONE),
2596	    ("nfs_strategy: buffer %p unexpectedly marked B_DONE", bp));
2597	BUF_ASSERT_HELD(bp);
2598
2599	if (bp->b_iocmd == BIO_READ)
2600		cr = bp->b_rcred;
2601	else
2602		cr = bp->b_wcred;
2603
2604	/*
2605	 * If the op is asynchronous and an i/o daemon is waiting
2606	 * queue the request, wake it up and wait for completion
2607	 * otherwise just do it ourselves.
2608	 */
2609	if ((bp->b_flags & B_ASYNC) == 0 ||
2610	    ncl_asyncio(VFSTONFS(ap->a_vp->v_mount), bp, NOCRED, curthread))
2611		(void) ncl_doio(ap->a_vp, bp, cr, curthread, 1);
2612	return (0);
2613}
2614
2615/*
2616 * fsync vnode op. Just call ncl_flush() with commit == 1.
2617 */
2618/* ARGSUSED */
2619static int
2620nfs_fsync(struct vop_fsync_args *ap)
2621{
2622
2623	if (ap->a_vp->v_type != VREG) {
2624		/*
2625		 * For NFS, metadata is changed synchronously on the server,
2626		 * so there is nothing to flush. Also, ncl_flush() clears
2627		 * the NMODIFIED flag and that shouldn't be done here for
2628		 * directories.
2629		 */
2630		return (0);
2631	}
2632	return (ncl_flush(ap->a_vp, ap->a_waitfor, ap->a_td, 1, 0));
2633}
2634
2635/*
2636 * Flush all the blocks associated with a vnode.
2637 * 	Walk through the buffer pool and push any dirty pages
2638 *	associated with the vnode.
2639 * If the called_from_renewthread argument is TRUE, it has been called
2640 * from the NFSv4 renew thread and, as such, cannot block indefinitely
2641 * waiting for a buffer write to complete.
2642 */
2643int
2644ncl_flush(struct vnode *vp, int waitfor, struct thread *td,
2645    int commit, int called_from_renewthread)
2646{
2647	struct nfsnode *np = VTONFS(vp);
2648	struct buf *bp;
2649	int i;
2650	struct buf *nbp;
2651	struct nfsmount *nmp = VFSTONFS(vp->v_mount);
2652	int error = 0, slptimeo = 0, slpflag = 0, retv, bvecpos;
2653	int passone = 1, trycnt = 0;
2654	u_quad_t off, endoff, toff;
2655	struct ucred* wcred = NULL;
2656	struct buf **bvec = NULL;
2657	struct bufobj *bo;
2658#ifndef NFS_COMMITBVECSIZ
2659#define	NFS_COMMITBVECSIZ	20
2660#endif
2661	struct buf *bvec_on_stack[NFS_COMMITBVECSIZ];
2662	int bvecsize = 0, bveccount;
2663
2664	if (called_from_renewthread != 0)
2665		slptimeo = hz;
2666	if (nmp->nm_flag & NFSMNT_INT)
2667		slpflag = PCATCH;
2668	if (!commit)
2669		passone = 0;
2670	bo = &vp->v_bufobj;
2671	/*
2672	 * A b_flags == (B_DELWRI | B_NEEDCOMMIT) block has been written to the
2673	 * server, but has not been committed to stable storage on the server
2674	 * yet. On the first pass, the byte range is worked out and the commit
2675	 * rpc is done. On the second pass, ncl_writebp() is called to do the
2676	 * job.
2677	 */
2678again:
2679	off = (u_quad_t)-1;
2680	endoff = 0;
2681	bvecpos = 0;
2682	if (NFS_ISV34(vp) && commit) {
2683		if (bvec != NULL && bvec != bvec_on_stack)
2684			free(bvec, M_TEMP);
2685		/*
2686		 * Count up how many buffers waiting for a commit.
2687		 */
2688		bveccount = 0;
2689		BO_LOCK(bo);
2690		TAILQ_FOREACH_SAFE(bp, &bo->bo_dirty.bv_hd, b_bobufs, nbp) {
2691			if (!BUF_ISLOCKED(bp) &&
2692			    (bp->b_flags & (B_DELWRI | B_NEEDCOMMIT))
2693				== (B_DELWRI | B_NEEDCOMMIT))
2694				bveccount++;
2695		}
2696		/*
2697		 * Allocate space to remember the list of bufs to commit.  It is
2698		 * important to use M_NOWAIT here to avoid a race with nfs_write.
2699		 * If we can't get memory (for whatever reason), we will end up
2700		 * committing the buffers one-by-one in the loop below.
2701		 */
2702		if (bveccount > NFS_COMMITBVECSIZ) {
2703			/*
2704			 * Release the vnode interlock to avoid a lock
2705			 * order reversal.
2706			 */
2707			BO_UNLOCK(bo);
2708			bvec = (struct buf **)
2709				malloc(bveccount * sizeof(struct buf *),
2710				       M_TEMP, M_NOWAIT);
2711			BO_LOCK(bo);
2712			if (bvec == NULL) {
2713				bvec = bvec_on_stack;
2714				bvecsize = NFS_COMMITBVECSIZ;
2715			} else
2716				bvecsize = bveccount;
2717		} else {
2718			bvec = bvec_on_stack;
2719			bvecsize = NFS_COMMITBVECSIZ;
2720		}
2721		TAILQ_FOREACH_SAFE(bp, &bo->bo_dirty.bv_hd, b_bobufs, nbp) {
2722			if (bvecpos >= bvecsize)
2723				break;
2724			if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT, NULL)) {
2725				nbp = TAILQ_NEXT(bp, b_bobufs);
2726				continue;
2727			}
2728			if ((bp->b_flags & (B_DELWRI | B_NEEDCOMMIT)) !=
2729			    (B_DELWRI | B_NEEDCOMMIT)) {
2730				BUF_UNLOCK(bp);
2731				nbp = TAILQ_NEXT(bp, b_bobufs);
2732				continue;
2733			}
2734			BO_UNLOCK(bo);
2735			bremfree(bp);
2736			/*
2737			 * Work out if all buffers are using the same cred
2738			 * so we can deal with them all with one commit.
2739			 *
2740			 * NOTE: we are not clearing B_DONE here, so we have
2741			 * to do it later on in this routine if we intend to
2742			 * initiate I/O on the bp.
2743			 *
2744			 * Note: to avoid loopback deadlocks, we do not
2745			 * assign b_runningbufspace.
2746			 */
2747			if (wcred == NULL)
2748				wcred = bp->b_wcred;
2749			else if (wcred != bp->b_wcred)
2750				wcred = NOCRED;
2751			vfs_busy_pages(bp, 1);
2752
2753			BO_LOCK(bo);
2754			/*
2755			 * bp is protected by being locked, but nbp is not
2756			 * and vfs_busy_pages() may sleep.  We have to
2757			 * recalculate nbp.
2758			 */
2759			nbp = TAILQ_NEXT(bp, b_bobufs);
2760
2761			/*
2762			 * A list of these buffers is kept so that the
2763			 * second loop knows which buffers have actually
2764			 * been committed. This is necessary, since there
2765			 * may be a race between the commit rpc and new
2766			 * uncommitted writes on the file.
2767			 */
2768			bvec[bvecpos++] = bp;
2769			toff = ((u_quad_t)bp->b_blkno) * DEV_BSIZE +
2770				bp->b_dirtyoff;
2771			if (toff < off)
2772				off = toff;
2773			toff += (u_quad_t)(bp->b_dirtyend - bp->b_dirtyoff);
2774			if (toff > endoff)
2775				endoff = toff;
2776		}
2777		BO_UNLOCK(bo);
2778	}
2779	if (bvecpos > 0) {
2780		/*
2781		 * Commit data on the server, as required.
2782		 * If all bufs are using the same wcred, then use that with
2783		 * one call for all of them, otherwise commit each one
2784		 * separately.
2785		 */
2786		if (wcred != NOCRED)
2787			retv = ncl_commit(vp, off, (int)(endoff - off),
2788					  wcred, td);
2789		else {
2790			retv = 0;
2791			for (i = 0; i < bvecpos; i++) {
2792				off_t off, size;
2793				bp = bvec[i];
2794				off = ((u_quad_t)bp->b_blkno) * DEV_BSIZE +
2795					bp->b_dirtyoff;
2796				size = (u_quad_t)(bp->b_dirtyend
2797						  - bp->b_dirtyoff);
2798				retv = ncl_commit(vp, off, (int)size,
2799						  bp->b_wcred, td);
2800				if (retv) break;
2801			}
2802		}
2803
2804		if (retv == NFSERR_STALEWRITEVERF)
2805			ncl_clearcommit(vp->v_mount);
2806
2807		/*
2808		 * Now, either mark the blocks I/O done or mark the
2809		 * blocks dirty, depending on whether the commit
2810		 * succeeded.
2811		 */
2812		for (i = 0; i < bvecpos; i++) {
2813			bp = bvec[i];
2814			bp->b_flags &= ~(B_NEEDCOMMIT | B_CLUSTEROK);
2815			if (retv) {
2816				/*
2817				 * Error, leave B_DELWRI intact
2818				 */
2819				vfs_unbusy_pages(bp);
2820				brelse(bp);
2821			} else {
2822				/*
2823				 * Success, remove B_DELWRI ( bundirty() ).
2824				 *
2825				 * b_dirtyoff/b_dirtyend seem to be NFS
2826				 * specific.  We should probably move that
2827				 * into bundirty(). XXX
2828				 */
2829				bufobj_wref(bo);
2830				bp->b_flags |= B_ASYNC;
2831				bundirty(bp);
2832				bp->b_flags &= ~B_DONE;
2833				bp->b_ioflags &= ~BIO_ERROR;
2834				bp->b_dirtyoff = bp->b_dirtyend = 0;
2835				bufdone(bp);
2836			}
2837		}
2838	}
2839
2840	/*
2841	 * Start/do any write(s) that are required.
2842	 */
2843loop:
2844	BO_LOCK(bo);
2845	TAILQ_FOREACH_SAFE(bp, &bo->bo_dirty.bv_hd, b_bobufs, nbp) {
2846		if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT, NULL)) {
2847			if (waitfor != MNT_WAIT || passone)
2848				continue;
2849
2850			error = BUF_TIMELOCK(bp,
2851			    LK_EXCLUSIVE | LK_SLEEPFAIL | LK_INTERLOCK,
2852			    BO_LOCKPTR(bo), "nfsfsync", slpflag, slptimeo);
2853			if (error == 0) {
2854				BUF_UNLOCK(bp);
2855				goto loop;
2856			}
2857			if (error == ENOLCK) {
2858				error = 0;
2859				goto loop;
2860			}
2861			if (called_from_renewthread != 0) {
2862				/*
2863				 * Return EIO so the flush will be retried
2864				 * later.
2865				 */
2866				error = EIO;
2867				goto done;
2868			}
2869			if (newnfs_sigintr(nmp, td)) {
2870				error = EINTR;
2871				goto done;
2872			}
2873			if (slpflag == PCATCH) {
2874				slpflag = 0;
2875				slptimeo = 2 * hz;
2876			}
2877			goto loop;
2878		}
2879		if ((bp->b_flags & B_DELWRI) == 0)
2880			panic("nfs_fsync: not dirty");
2881		if ((passone || !commit) && (bp->b_flags & B_NEEDCOMMIT)) {
2882			BUF_UNLOCK(bp);
2883			continue;
2884		}
2885		BO_UNLOCK(bo);
2886		bremfree(bp);
2887		if (passone || !commit)
2888		    bp->b_flags |= B_ASYNC;
2889		else
2890		    bp->b_flags |= B_ASYNC;
2891		bwrite(bp);
2892		if (newnfs_sigintr(nmp, td)) {
2893			error = EINTR;
2894			goto done;
2895		}
2896		goto loop;
2897	}
2898	if (passone) {
2899		passone = 0;
2900		BO_UNLOCK(bo);
2901		goto again;
2902	}
2903	if (waitfor == MNT_WAIT) {
2904		while (bo->bo_numoutput) {
2905			error = bufobj_wwait(bo, slpflag, slptimeo);
2906			if (error) {
2907			    BO_UNLOCK(bo);
2908			    if (called_from_renewthread != 0) {
2909				/*
2910				 * Return EIO so that the flush will be
2911				 * retried later.
2912				 */
2913				error = EIO;
2914				goto done;
2915			    }
2916			    error = newnfs_sigintr(nmp, td);
2917			    if (error)
2918				goto done;
2919			    if (slpflag == PCATCH) {
2920				slpflag = 0;
2921				slptimeo = 2 * hz;
2922			    }
2923			    BO_LOCK(bo);
2924			}
2925		}
2926		if (bo->bo_dirty.bv_cnt != 0 && commit) {
2927			BO_UNLOCK(bo);
2928			goto loop;
2929		}
2930		/*
2931		 * Wait for all the async IO requests to drain
2932		 */
2933		BO_UNLOCK(bo);
2934		mtx_lock(&np->n_mtx);
2935		while (np->n_directio_asyncwr > 0) {
2936			np->n_flag |= NFSYNCWAIT;
2937			error = newnfs_msleep(td, &np->n_directio_asyncwr,
2938			    &np->n_mtx, slpflag | (PRIBIO + 1),
2939			    "nfsfsync", 0);
2940			if (error) {
2941				if (newnfs_sigintr(nmp, td)) {
2942					mtx_unlock(&np->n_mtx);
2943					error = EINTR;
2944					goto done;
2945				}
2946			}
2947		}
2948		mtx_unlock(&np->n_mtx);
2949	} else
2950		BO_UNLOCK(bo);
2951	if (NFSHASPNFS(nmp)) {
2952		nfscl_layoutcommit(vp, td);
2953		/*
2954		 * Invalidate the attribute cache, since writes to a DS
2955		 * won't update the size attribute.
2956		 */
2957		mtx_lock(&np->n_mtx);
2958		np->n_attrstamp = 0;
2959	} else
2960		mtx_lock(&np->n_mtx);
2961	if (np->n_flag & NWRITEERR) {
2962		error = np->n_error;
2963		np->n_flag &= ~NWRITEERR;
2964	}
2965  	if (commit && bo->bo_dirty.bv_cnt == 0 &&
2966	    bo->bo_numoutput == 0 && np->n_directio_asyncwr == 0)
2967  		np->n_flag &= ~NMODIFIED;
2968	mtx_unlock(&np->n_mtx);
2969done:
2970	if (bvec != NULL && bvec != bvec_on_stack)
2971		free(bvec, M_TEMP);
2972	if (error == 0 && commit != 0 && waitfor == MNT_WAIT &&
2973	    (bo->bo_dirty.bv_cnt != 0 || bo->bo_numoutput != 0 ||
2974	    np->n_directio_asyncwr != 0)) {
2975		if (trycnt++ < 5) {
2976			/* try, try again... */
2977			passone = 1;
2978			wcred = NULL;
2979			bvec = NULL;
2980			bvecsize = 0;
2981			goto again;
2982		}
2983		vn_printf(vp, "ncl_flush failed");
2984		error = called_from_renewthread != 0 ? EIO : EBUSY;
2985	}
2986	return (error);
2987}
2988
2989/*
2990 * NFS advisory byte-level locks.
2991 */
2992static int
2993nfs_advlock(struct vop_advlock_args *ap)
2994{
2995	struct vnode *vp = ap->a_vp;
2996	struct ucred *cred;
2997	struct nfsnode *np = VTONFS(ap->a_vp);
2998	struct proc *p = (struct proc *)ap->a_id;
2999	struct thread *td = curthread;	/* XXX */
3000	struct vattr va;
3001	int ret, error = EOPNOTSUPP;
3002	u_quad_t size;
3003
3004	if (NFS_ISV4(vp) && (ap->a_flags & (F_POSIX | F_FLOCK)) != 0) {
3005		if (vp->v_type != VREG)
3006			return (EINVAL);
3007		if ((ap->a_flags & F_POSIX) != 0)
3008			cred = p->p_ucred;
3009		else
3010			cred = td->td_ucred;
3011		NFSVOPLOCK(vp, LK_EXCLUSIVE | LK_RETRY);
3012		if (vp->v_iflag & VI_DOOMED) {
3013			NFSVOPUNLOCK(vp, 0);
3014			return (EBADF);
3015		}
3016
3017		/*
3018		 * If this is unlocking a write locked region, flush and
3019		 * commit them before unlocking. This is required by
3020		 * RFC3530 Sec. 9.3.2.
3021		 */
3022		if (ap->a_op == F_UNLCK &&
3023		    nfscl_checkwritelocked(vp, ap->a_fl, cred, td, ap->a_id,
3024		    ap->a_flags))
3025			(void) ncl_flush(vp, MNT_WAIT, td, 1, 0);
3026
3027		/*
3028		 * Loop around doing the lock op, while a blocking lock
3029		 * must wait for the lock op to succeed.
3030		 */
3031		do {
3032			ret = nfsrpc_advlock(vp, np->n_size, ap->a_op,
3033			    ap->a_fl, 0, cred, td, ap->a_id, ap->a_flags);
3034			if (ret == NFSERR_DENIED && (ap->a_flags & F_WAIT) &&
3035			    ap->a_op == F_SETLK) {
3036				NFSVOPUNLOCK(vp, 0);
3037				error = nfs_catnap(PZERO | PCATCH, ret,
3038				    "ncladvl");
3039				if (error)
3040					return (EINTR);
3041				NFSVOPLOCK(vp, LK_EXCLUSIVE | LK_RETRY);
3042				if (vp->v_iflag & VI_DOOMED) {
3043					NFSVOPUNLOCK(vp, 0);
3044					return (EBADF);
3045				}
3046			}
3047		} while (ret == NFSERR_DENIED && (ap->a_flags & F_WAIT) &&
3048		     ap->a_op == F_SETLK);
3049		if (ret == NFSERR_DENIED) {
3050			NFSVOPUNLOCK(vp, 0);
3051			return (EAGAIN);
3052		} else if (ret == EINVAL || ret == EBADF || ret == EINTR) {
3053			NFSVOPUNLOCK(vp, 0);
3054			return (ret);
3055		} else if (ret != 0) {
3056			NFSVOPUNLOCK(vp, 0);
3057			return (EACCES);
3058		}
3059
3060		/*
3061		 * Now, if we just got a lock, invalidate data in the buffer
3062		 * cache, as required, so that the coherency conforms with
3063		 * RFC3530 Sec. 9.3.2.
3064		 */
3065		if (ap->a_op == F_SETLK) {
3066			if ((np->n_flag & NMODIFIED) == 0) {
3067				np->n_attrstamp = 0;
3068				KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(vp);
3069				ret = VOP_GETATTR(vp, &va, cred);
3070			}
3071			if ((np->n_flag & NMODIFIED) || ret ||
3072			    np->n_change != va.va_filerev) {
3073				(void) ncl_vinvalbuf(vp, V_SAVE, td, 1);
3074				np->n_attrstamp = 0;
3075				KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(vp);
3076				ret = VOP_GETATTR(vp, &va, cred);
3077				if (!ret) {
3078					np->n_mtime = va.va_mtime;
3079					np->n_change = va.va_filerev;
3080				}
3081			}
3082			/* Mark that a file lock has been acquired. */
3083			mtx_lock(&np->n_mtx);
3084			np->n_flag |= NHASBEENLOCKED;
3085			mtx_unlock(&np->n_mtx);
3086		}
3087		NFSVOPUNLOCK(vp, 0);
3088		return (0);
3089	} else if (!NFS_ISV4(vp)) {
3090		error = NFSVOPLOCK(vp, LK_SHARED);
3091		if (error)
3092			return (error);
3093		if ((VFSTONFS(vp->v_mount)->nm_flag & NFSMNT_NOLOCKD) != 0) {
3094			size = VTONFS(vp)->n_size;
3095			NFSVOPUNLOCK(vp, 0);
3096			error = lf_advlock(ap, &(vp->v_lockf), size);
3097		} else {
3098			if (nfs_advlock_p != NULL)
3099				error = nfs_advlock_p(ap);
3100			else {
3101				NFSVOPUNLOCK(vp, 0);
3102				error = ENOLCK;
3103			}
3104		}
3105		if (error == 0 && ap->a_op == F_SETLK) {
3106			error = NFSVOPLOCK(vp, LK_SHARED);
3107			if (error == 0) {
3108				/* Mark that a file lock has been acquired. */
3109				mtx_lock(&np->n_mtx);
3110				np->n_flag |= NHASBEENLOCKED;
3111				mtx_unlock(&np->n_mtx);
3112				NFSVOPUNLOCK(vp, 0);
3113			}
3114		}
3115	}
3116	return (error);
3117}
3118
3119/*
3120 * NFS advisory byte-level locks.
3121 */
3122static int
3123nfs_advlockasync(struct vop_advlockasync_args *ap)
3124{
3125	struct vnode *vp = ap->a_vp;
3126	u_quad_t size;
3127	int error;
3128
3129	if (NFS_ISV4(vp))
3130		return (EOPNOTSUPP);
3131	error = NFSVOPLOCK(vp, LK_SHARED);
3132	if (error)
3133		return (error);
3134	if ((VFSTONFS(vp->v_mount)->nm_flag & NFSMNT_NOLOCKD) != 0) {
3135		size = VTONFS(vp)->n_size;
3136		NFSVOPUNLOCK(vp, 0);
3137		error = lf_advlockasync(ap, &(vp->v_lockf), size);
3138	} else {
3139		NFSVOPUNLOCK(vp, 0);
3140		error = EOPNOTSUPP;
3141	}
3142	return (error);
3143}
3144
3145/*
3146 * Print out the contents of an nfsnode.
3147 */
3148static int
3149nfs_print(struct vop_print_args *ap)
3150{
3151	struct vnode *vp = ap->a_vp;
3152	struct nfsnode *np = VTONFS(vp);
3153
3154	printf("\tfileid %ld fsid 0x%x", np->n_vattr.na_fileid,
3155	    np->n_vattr.na_fsid);
3156	if (vp->v_type == VFIFO)
3157		fifo_printinfo(vp);
3158	printf("\n");
3159	return (0);
3160}
3161
3162/*
3163 * This is the "real" nfs::bwrite(struct buf*).
3164 * We set B_CACHE if this is a VMIO buffer.
3165 */
3166int
3167ncl_writebp(struct buf *bp, int force __unused, struct thread *td)
3168{
3169	int s;
3170	int oldflags = bp->b_flags;
3171#if 0
3172	int retv = 1;
3173	off_t off;
3174#endif
3175
3176	BUF_ASSERT_HELD(bp);
3177
3178	if (bp->b_flags & B_INVAL) {
3179		brelse(bp);
3180		return(0);
3181	}
3182
3183	bp->b_flags |= B_CACHE;
3184
3185	/*
3186	 * Undirty the bp.  We will redirty it later if the I/O fails.
3187	 */
3188
3189	s = splbio();
3190	bundirty(bp);
3191	bp->b_flags &= ~B_DONE;
3192	bp->b_ioflags &= ~BIO_ERROR;
3193	bp->b_iocmd = BIO_WRITE;
3194
3195	bufobj_wref(bp->b_bufobj);
3196	curthread->td_ru.ru_oublock++;
3197	splx(s);
3198
3199	/*
3200	 * Note: to avoid loopback deadlocks, we do not
3201	 * assign b_runningbufspace.
3202	 */
3203	vfs_busy_pages(bp, 1);
3204
3205	BUF_KERNPROC(bp);
3206	bp->b_iooffset = dbtob(bp->b_blkno);
3207	bstrategy(bp);
3208
3209	if( (oldflags & B_ASYNC) == 0) {
3210		int rtval = bufwait(bp);
3211
3212		if (oldflags & B_DELWRI) {
3213			s = splbio();
3214			reassignbuf(bp);
3215			splx(s);
3216		}
3217		brelse(bp);
3218		return (rtval);
3219	}
3220
3221	return (0);
3222}
3223
3224/*
3225 * nfs special file access vnode op.
3226 * Essentially just get vattr and then imitate iaccess() since the device is
3227 * local to the client.
3228 */
3229static int
3230nfsspec_access(struct vop_access_args *ap)
3231{
3232	struct vattr *vap;
3233	struct ucred *cred = ap->a_cred;
3234	struct vnode *vp = ap->a_vp;
3235	accmode_t accmode = ap->a_accmode;
3236	struct vattr vattr;
3237	int error;
3238
3239	/*
3240	 * Disallow write attempts on filesystems mounted read-only;
3241	 * unless the file is a socket, fifo, or a block or character
3242	 * device resident on the filesystem.
3243	 */
3244	if ((accmode & VWRITE) && (vp->v_mount->mnt_flag & MNT_RDONLY)) {
3245		switch (vp->v_type) {
3246		case VREG:
3247		case VDIR:
3248		case VLNK:
3249			return (EROFS);
3250		default:
3251			break;
3252		}
3253	}
3254	vap = &vattr;
3255	error = VOP_GETATTR(vp, vap, cred);
3256	if (error)
3257		goto out;
3258	error  = vaccess(vp->v_type, vap->va_mode, vap->va_uid, vap->va_gid,
3259	    accmode, cred, NULL);
3260out:
3261	return error;
3262}
3263
3264/*
3265 * Read wrapper for fifos.
3266 */
3267static int
3268nfsfifo_read(struct vop_read_args *ap)
3269{
3270	struct nfsnode *np = VTONFS(ap->a_vp);
3271	int error;
3272
3273	/*
3274	 * Set access flag.
3275	 */
3276	mtx_lock(&np->n_mtx);
3277	np->n_flag |= NACC;
3278	vfs_timestamp(&np->n_atim);
3279	mtx_unlock(&np->n_mtx);
3280	error = fifo_specops.vop_read(ap);
3281	return error;
3282}
3283
3284/*
3285 * Write wrapper for fifos.
3286 */
3287static int
3288nfsfifo_write(struct vop_write_args *ap)
3289{
3290	struct nfsnode *np = VTONFS(ap->a_vp);
3291
3292	/*
3293	 * Set update flag.
3294	 */
3295	mtx_lock(&np->n_mtx);
3296	np->n_flag |= NUPD;
3297	vfs_timestamp(&np->n_mtim);
3298	mtx_unlock(&np->n_mtx);
3299	return(fifo_specops.vop_write(ap));
3300}
3301
3302/*
3303 * Close wrapper for fifos.
3304 *
3305 * Update the times on the nfsnode then do fifo close.
3306 */
3307static int
3308nfsfifo_close(struct vop_close_args *ap)
3309{
3310	struct vnode *vp = ap->a_vp;
3311	struct nfsnode *np = VTONFS(vp);
3312	struct vattr vattr;
3313	struct timespec ts;
3314
3315	mtx_lock(&np->n_mtx);
3316	if (np->n_flag & (NACC | NUPD)) {
3317		vfs_timestamp(&ts);
3318		if (np->n_flag & NACC)
3319			np->n_atim = ts;
3320		if (np->n_flag & NUPD)
3321			np->n_mtim = ts;
3322		np->n_flag |= NCHG;
3323		if (vrefcnt(vp) == 1 &&
3324		    (vp->v_mount->mnt_flag & MNT_RDONLY) == 0) {
3325			VATTR_NULL(&vattr);
3326			if (np->n_flag & NACC)
3327				vattr.va_atime = np->n_atim;
3328			if (np->n_flag & NUPD)
3329				vattr.va_mtime = np->n_mtim;
3330			mtx_unlock(&np->n_mtx);
3331			(void)VOP_SETATTR(vp, &vattr, ap->a_cred);
3332			goto out;
3333		}
3334	}
3335	mtx_unlock(&np->n_mtx);
3336out:
3337	return (fifo_specops.vop_close(ap));
3338}
3339
3340/*
3341 * Just call ncl_writebp() with the force argument set to 1.
3342 *
3343 * NOTE: B_DONE may or may not be set in a_bp on call.
3344 */
3345static int
3346nfs_bwrite(struct buf *bp)
3347{
3348
3349	return (ncl_writebp(bp, 1, curthread));
3350}
3351
3352struct buf_ops buf_ops_newnfs = {
3353	.bop_name	=	"buf_ops_nfs",
3354	.bop_write	=	nfs_bwrite,
3355	.bop_strategy	=	bufstrategy,
3356	.bop_sync	=	bufsync,
3357	.bop_bdflush	=	bufbdflush,
3358};
3359
3360static int
3361nfs_getacl(struct vop_getacl_args *ap)
3362{
3363	int error;
3364
3365	if (ap->a_type != ACL_TYPE_NFS4)
3366		return (EOPNOTSUPP);
3367	error = nfsrpc_getacl(ap->a_vp, ap->a_cred, ap->a_td, ap->a_aclp,
3368	    NULL);
3369	if (error > NFSERR_STALE) {
3370		(void) nfscl_maperr(ap->a_td, error, (uid_t)0, (gid_t)0);
3371		error = EPERM;
3372	}
3373	return (error);
3374}
3375
3376static int
3377nfs_setacl(struct vop_setacl_args *ap)
3378{
3379	int error;
3380
3381	if (ap->a_type != ACL_TYPE_NFS4)
3382		return (EOPNOTSUPP);
3383	error = nfsrpc_setacl(ap->a_vp, ap->a_cred, ap->a_td, ap->a_aclp,
3384	    NULL);
3385	if (error > NFSERR_STALE) {
3386		(void) nfscl_maperr(ap->a_td, error, (uid_t)0, (gid_t)0);
3387		error = EPERM;
3388	}
3389	return (error);
3390}
3391
3392static int
3393nfs_set_text(struct vop_set_text_args *ap)
3394{
3395	struct vnode *vp = ap->a_vp;
3396	struct nfsnode *np;
3397
3398	/*
3399	 * If the text file has been mmap'd, flush any dirty pages to the
3400	 * buffer cache and then...
3401	 * Make sure all writes are pushed to the NFS server.  If this is not
3402	 * done, the modify time of the file can change while the text
3403	 * file is being executed.  This will cause the process that is
3404	 * executing the text file to be terminated.
3405	 */
3406	if (vp->v_object != NULL) {
3407		VM_OBJECT_WLOCK(vp->v_object);
3408		vm_object_page_clean(vp->v_object, 0, 0, OBJPC_SYNC);
3409		VM_OBJECT_WUNLOCK(vp->v_object);
3410	}
3411
3412	/* And, finally, make sure that n_mtime is up to date. */
3413	np = VTONFS(vp);
3414	mtx_lock(&np->n_mtx);
3415	np->n_mtime = np->n_vattr.na_mtime;
3416	mtx_unlock(&np->n_mtx);
3417
3418	vp->v_vflag |= VV_TEXT;
3419	return (0);
3420}
3421
3422/*
3423 * Return POSIX pathconf information applicable to nfs filesystems.
3424 */
3425static int
3426nfs_pathconf(struct vop_pathconf_args *ap)
3427{
3428	struct nfsv3_pathconf pc;
3429	struct nfsvattr nfsva;
3430	struct vnode *vp = ap->a_vp;
3431	struct thread *td = curthread;
3432	int attrflag, error;
3433
3434	if ((NFS_ISV34(vp) && (ap->a_name == _PC_LINK_MAX ||
3435	    ap->a_name == _PC_NAME_MAX || ap->a_name == _PC_CHOWN_RESTRICTED ||
3436	    ap->a_name == _PC_NO_TRUNC)) ||
3437	    (NFS_ISV4(vp) && ap->a_name == _PC_ACL_NFS4)) {
3438		/*
3439		 * Since only the above 4 a_names are returned by the NFSv3
3440		 * Pathconf RPC, there is no point in doing it for others.
3441		 * For NFSv4, the Pathconf RPC (actually a Getattr Op.) can
3442		 * be used for _PC_NFS4_ACL as well.
3443		 */
3444		error = nfsrpc_pathconf(vp, &pc, td->td_ucred, td, &nfsva,
3445		    &attrflag, NULL);
3446		if (attrflag != 0)
3447			(void) nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0,
3448			    1);
3449		if (error != 0)
3450			return (error);
3451	} else {
3452		/*
3453		 * For NFSv2 (or NFSv3 when not one of the above 4 a_names),
3454		 * just fake them.
3455		 */
3456		pc.pc_linkmax = LINK_MAX;
3457		pc.pc_namemax = NFS_MAXNAMLEN;
3458		pc.pc_notrunc = 1;
3459		pc.pc_chownrestricted = 1;
3460		pc.pc_caseinsensitive = 0;
3461		pc.pc_casepreserving = 1;
3462		error = 0;
3463	}
3464	switch (ap->a_name) {
3465	case _PC_LINK_MAX:
3466		*ap->a_retval = pc.pc_linkmax;
3467		break;
3468	case _PC_NAME_MAX:
3469		*ap->a_retval = pc.pc_namemax;
3470		break;
3471	case _PC_PATH_MAX:
3472		*ap->a_retval = PATH_MAX;
3473		break;
3474	case _PC_PIPE_BUF:
3475		*ap->a_retval = PIPE_BUF;
3476		break;
3477	case _PC_CHOWN_RESTRICTED:
3478		*ap->a_retval = pc.pc_chownrestricted;
3479		break;
3480	case _PC_NO_TRUNC:
3481		*ap->a_retval = pc.pc_notrunc;
3482		break;
3483	case _PC_ACL_EXTENDED:
3484		*ap->a_retval = 0;
3485		break;
3486	case _PC_ACL_NFS4:
3487		if (NFS_ISV4(vp) && nfsrv_useacl != 0 && attrflag != 0 &&
3488		    NFSISSET_ATTRBIT(&nfsva.na_suppattr, NFSATTRBIT_ACL))
3489			*ap->a_retval = 1;
3490		else
3491			*ap->a_retval = 0;
3492		break;
3493	case _PC_ACL_PATH_MAX:
3494		if (NFS_ISV4(vp))
3495			*ap->a_retval = ACL_MAX_ENTRIES;
3496		else
3497			*ap->a_retval = 3;
3498		break;
3499	case _PC_MAC_PRESENT:
3500		*ap->a_retval = 0;
3501		break;
3502	case _PC_ASYNC_IO:
3503		/* _PC_ASYNC_IO should have been handled by upper layers. */
3504		KASSERT(0, ("_PC_ASYNC_IO should not get here"));
3505		error = EINVAL;
3506		break;
3507	case _PC_PRIO_IO:
3508		*ap->a_retval = 0;
3509		break;
3510	case _PC_SYNC_IO:
3511		*ap->a_retval = 0;
3512		break;
3513	case _PC_ALLOC_SIZE_MIN:
3514		*ap->a_retval = vp->v_mount->mnt_stat.f_bsize;
3515		break;
3516	case _PC_FILESIZEBITS:
3517		if (NFS_ISV34(vp))
3518			*ap->a_retval = 64;
3519		else
3520			*ap->a_retval = 32;
3521		break;
3522	case _PC_REC_INCR_XFER_SIZE:
3523		*ap->a_retval = vp->v_mount->mnt_stat.f_iosize;
3524		break;
3525	case _PC_REC_MAX_XFER_SIZE:
3526		*ap->a_retval = -1; /* means ``unlimited'' */
3527		break;
3528	case _PC_REC_MIN_XFER_SIZE:
3529		*ap->a_retval = vp->v_mount->mnt_stat.f_iosize;
3530		break;
3531	case _PC_REC_XFER_ALIGN:
3532		*ap->a_retval = PAGE_SIZE;
3533		break;
3534	case _PC_SYMLINK_MAX:
3535		*ap->a_retval = NFS_MAXPATHLEN;
3536		break;
3537
3538	default:
3539		error = EINVAL;
3540		break;
3541	}
3542	return (error);
3543}
3544
3545