xref: /freebsd/sys/fs/nfsserver/nfs_nfsdport.c (revision 21b492ed51aa6ff8008a8aa83333b1de30288a15)
1 /*-
2  * SPDX-License-Identifier: BSD-3-Clause
3  *
4  * Copyright (c) 1989, 1993
5  *	The Regents of the University of California.  All rights reserved.
6  *
7  * This code is derived from software contributed to Berkeley by
8  * Rick Macklem at The University of Guelph.
9  *
10  * Redistribution and use in source and binary forms, with or without
11  * modification, are permitted provided that the following conditions
12  * are met:
13  * 1. Redistributions of source code must retain the above copyright
14  *    notice, this list of conditions and the following disclaimer.
15  * 2. Redistributions in binary form must reproduce the above copyright
16  *    notice, this list of conditions and the following disclaimer in the
17  *    documentation and/or other materials provided with the distribution.
18  * 3. Neither the name of the University nor the names of its contributors
19  *    may be used to endorse or promote products derived from this software
20  *    without specific prior written permission.
21  *
22  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
23  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
24  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
25  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
26  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
27  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
28  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
29  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
30  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
31  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
32  * SUCH DAMAGE.
33  *
34  */
35 
36 #include <sys/cdefs.h>
37 __FBSDID("$FreeBSD$");
38 
39 #include <sys/capsicum.h>
40 #include <sys/extattr.h>
41 
42 /*
43  * Functions that perform the vfs operations required by the routines in
44  * nfsd_serv.c. It is hoped that this change will make the server more
45  * portable.
46  */
47 
48 #include <fs/nfs/nfsport.h>
49 #include <security/mac/mac_framework.h>
50 #include <sys/filio.h>
51 #include <sys/hash.h>
52 #include <sys/sysctl.h>
53 #include <nlm/nlm_prot.h>
54 #include <nlm/nlm.h>
55 
56 FEATURE(nfsd, "NFSv4 server");
57 
58 extern u_int32_t newnfs_true, newnfs_false, newnfs_xdrneg1;
59 extern int nfsrv_useacl;
60 extern int newnfs_numnfsd;
61 extern struct mount nfsv4root_mnt;
62 extern struct nfsrv_stablefirst nfsrv_stablefirst;
63 extern void (*nfsd_call_servertimer)(void);
64 extern SVCPOOL	*nfsrvd_pool;
65 extern struct nfsv4lock nfsd_suspend_lock;
66 extern struct nfsclienthashhead *nfsclienthash;
67 extern struct nfslockhashhead *nfslockhash;
68 extern struct nfssessionhash *nfssessionhash;
69 extern int nfsrv_sessionhashsize;
70 extern struct nfsstatsv1 nfsstatsv1;
71 extern struct nfslayouthash *nfslayouthash;
72 extern int nfsrv_layouthashsize;
73 extern struct mtx nfsrv_dslock_mtx;
74 extern int nfs_pnfsiothreads;
75 extern struct nfsdontlisthead nfsrv_dontlisthead;
76 extern volatile int nfsrv_dontlistlen;
77 extern volatile int nfsrv_devidcnt;
78 extern int nfsrv_maxpnfsmirror;
79 extern uint32_t nfs_srvmaxio;
80 extern int nfs_bufpackets;
81 extern u_long sb_max_adj;
82 struct vfsoptlist nfsv4root_opt, nfsv4root_newopt;
83 NFSDLOCKMUTEX;
84 NFSSTATESPINLOCK;
85 struct nfsrchash_bucket nfsrchash_table[NFSRVCACHE_HASHSIZE];
86 struct nfsrchash_bucket nfsrcahash_table[NFSRVCACHE_HASHSIZE];
87 struct mtx nfsrc_udpmtx;
88 struct mtx nfs_v4root_mutex;
89 struct mtx nfsrv_dontlistlock_mtx;
90 struct mtx nfsrv_recalllock_mtx;
91 struct nfsrvfh nfs_rootfh, nfs_pubfh;
92 int nfs_pubfhset = 0, nfs_rootfhset = 0;
93 struct proc *nfsd_master_proc = NULL;
94 int nfsd_debuglevel = 0;
95 static pid_t nfsd_master_pid = (pid_t)-1;
96 static char nfsd_master_comm[MAXCOMLEN + 1];
97 static struct timeval nfsd_master_start;
98 static uint32_t nfsv4_sysid = 0;
99 static fhandle_t zerofh;
100 
101 static int nfssvc_srvcall(struct thread *, struct nfssvc_args *,
102     struct ucred *);
103 
104 int nfsrv_enable_crossmntpt = 1;
105 static int nfs_commit_blks;
106 static int nfs_commit_miss;
107 extern int nfsrv_issuedelegs;
108 extern int nfsrv_dolocallocks;
109 extern int nfsd_enable_stringtouid;
110 extern struct nfsdevicehead nfsrv_devidhead;
111 
112 static int nfsrv_createiovec(int, struct mbuf **, struct mbuf **,
113     struct iovec **);
114 static int nfsrv_createiovec_extpgs(int, int, struct mbuf **,
115     struct mbuf **, struct iovec **);
116 static int nfsrv_createiovecw(int, struct mbuf *, char *, struct iovec **,
117     int *);
118 static void nfsrv_pnfscreate(struct vnode *, struct vattr *, struct ucred *,
119     NFSPROC_T *);
120 static void nfsrv_pnfsremovesetup(struct vnode *, NFSPROC_T *, struct vnode **,
121     int *, char *, fhandle_t *);
122 static void nfsrv_pnfsremove(struct vnode **, int, char *, fhandle_t *,
123     NFSPROC_T *);
124 static int nfsrv_proxyds(struct vnode *, off_t, int, struct ucred *,
125     struct thread *, int, struct mbuf **, char *, struct mbuf **,
126     struct nfsvattr *, struct acl *, off_t *, int, bool *);
127 static int nfsrv_setextattr(struct vnode *, struct nfsvattr *, NFSPROC_T *);
128 static int nfsrv_readdsrpc(fhandle_t *, off_t, int, struct ucred *,
129     NFSPROC_T *, struct nfsmount *, struct mbuf **, struct mbuf **);
130 static int nfsrv_writedsrpc(fhandle_t *, off_t, int, struct ucred *,
131     NFSPROC_T *, struct vnode *, struct nfsmount **, int, struct mbuf **,
132     char *, int *);
133 static int nfsrv_allocatedsrpc(fhandle_t *, off_t, off_t, struct ucred *,
134     NFSPROC_T *, struct vnode *, struct nfsmount **, int, int *);
135 static int nfsrv_deallocatedsrpc(fhandle_t *, off_t, off_t, struct ucred *,
136     NFSPROC_T *, struct vnode *, struct nfsmount **, int, int *);
137 static int nfsrv_setacldsrpc(fhandle_t *, struct ucred *, NFSPROC_T *,
138     struct vnode *, struct nfsmount **, int, struct acl *, int *);
139 static int nfsrv_setattrdsrpc(fhandle_t *, struct ucred *, NFSPROC_T *,
140     struct vnode *, struct nfsmount **, int, struct nfsvattr *, int *);
141 static int nfsrv_getattrdsrpc(fhandle_t *, struct ucred *, NFSPROC_T *,
142     struct vnode *, struct nfsmount *, struct nfsvattr *);
143 static int nfsrv_seekdsrpc(fhandle_t *, off_t *, int, bool *, struct ucred *,
144     NFSPROC_T *, struct nfsmount *);
145 static int nfsrv_putfhname(fhandle_t *, char *);
146 static int nfsrv_pnfslookupds(struct vnode *, struct vnode *,
147     struct pnfsdsfile *, struct vnode **, NFSPROC_T *);
148 static void nfsrv_pnfssetfh(struct vnode *, struct pnfsdsfile *, char *, char *,
149     struct vnode *, NFSPROC_T *);
150 static int nfsrv_dsremove(struct vnode *, char *, struct ucred *, NFSPROC_T *);
151 static int nfsrv_dssetacl(struct vnode *, struct acl *, struct ucred *,
152     NFSPROC_T *);
153 static int nfsrv_pnfsstatfs(struct statfs *, struct mount *);
154 
155 int nfs_pnfsio(task_fn_t *, void *);
156 
157 SYSCTL_NODE(_vfs, OID_AUTO, nfsd, CTLFLAG_RW | CTLFLAG_MPSAFE, 0,
158     "NFS server");
159 SYSCTL_INT(_vfs_nfsd, OID_AUTO, mirrormnt, CTLFLAG_RW,
160     &nfsrv_enable_crossmntpt, 0, "Enable nfsd to cross mount points");
161 SYSCTL_INT(_vfs_nfsd, OID_AUTO, commit_blks, CTLFLAG_RW, &nfs_commit_blks,
162     0, "");
163 SYSCTL_INT(_vfs_nfsd, OID_AUTO, commit_miss, CTLFLAG_RW, &nfs_commit_miss,
164     0, "");
165 SYSCTL_INT(_vfs_nfsd, OID_AUTO, issue_delegations, CTLFLAG_RW,
166     &nfsrv_issuedelegs, 0, "Enable nfsd to issue delegations");
167 SYSCTL_INT(_vfs_nfsd, OID_AUTO, enable_locallocks, CTLFLAG_RW,
168     &nfsrv_dolocallocks, 0, "Enable nfsd to acquire local locks on files");
169 SYSCTL_INT(_vfs_nfsd, OID_AUTO, debuglevel, CTLFLAG_RW, &nfsd_debuglevel,
170     0, "Debug level for NFS server");
171 SYSCTL_INT(_vfs_nfsd, OID_AUTO, enable_stringtouid, CTLFLAG_RW,
172     &nfsd_enable_stringtouid, 0, "Enable nfsd to accept numeric owner_names");
173 static int nfsrv_pnfsgetdsattr = 1;
174 SYSCTL_INT(_vfs_nfsd, OID_AUTO, pnfsgetdsattr, CTLFLAG_RW,
175     &nfsrv_pnfsgetdsattr, 0, "When set getattr gets DS attributes via RPC");
176 
177 /*
178  * nfsrv_dsdirsize can only be increased and only when the nfsd threads are
179  * not running.
180  * The dsN subdirectories for the increased values must have been created
181  * on all DS servers before this increase is done.
182  */
183 u_int	nfsrv_dsdirsize = 20;
184 static int
185 sysctl_dsdirsize(SYSCTL_HANDLER_ARGS)
186 {
187 	int error, newdsdirsize;
188 
189 	newdsdirsize = nfsrv_dsdirsize;
190 	error = sysctl_handle_int(oidp, &newdsdirsize, 0, req);
191 	if (error != 0 || req->newptr == NULL)
192 		return (error);
193 	if (newdsdirsize <= nfsrv_dsdirsize || newdsdirsize > 10000 ||
194 	    newnfs_numnfsd != 0)
195 		return (EINVAL);
196 	nfsrv_dsdirsize = newdsdirsize;
197 	return (0);
198 }
199 SYSCTL_PROC(_vfs_nfsd, OID_AUTO, dsdirsize,
200     CTLTYPE_UINT | CTLFLAG_MPSAFE | CTLFLAG_RW, 0, sizeof(nfsrv_dsdirsize),
201     sysctl_dsdirsize, "IU", "Number of dsN subdirs on the DS servers");
202 
203 /*
204  * nfs_srvmaxio can only be increased and only when the nfsd threads are
205  * not running.  The setting must be a power of 2, with the current limit of
206  * 1Mbyte.
207  */
208 static int
209 sysctl_srvmaxio(SYSCTL_HANDLER_ARGS)
210 {
211 	int error;
212 	u_int newsrvmaxio;
213 	uint64_t tval;
214 
215 	newsrvmaxio = nfs_srvmaxio;
216 	error = sysctl_handle_int(oidp, &newsrvmaxio, 0, req);
217 	if (error != 0 || req->newptr == NULL)
218 		return (error);
219 	if (newsrvmaxio == nfs_srvmaxio)
220 		return (0);
221 	if (newsrvmaxio < nfs_srvmaxio) {
222 		printf("nfsd: vfs.nfsd.srvmaxio can only be increased\n");
223 		return (EINVAL);
224 	}
225 	if (newsrvmaxio > 1048576) {
226 		printf("nfsd: vfs.nfsd.srvmaxio cannot be > 1Mbyte\n");
227 		return (EINVAL);
228 	}
229 	if ((newsrvmaxio & (newsrvmaxio - 1)) != 0) {
230 		printf("nfsd: vfs.nfsd.srvmaxio must be a power of 2\n");
231 		return (EINVAL);
232 	}
233 
234 	/*
235 	 * Check that kern.ipc.maxsockbuf is large enough for
236 	 * newsrviomax, given the setting of vfs.nfs.bufpackets.
237 	 */
238 	if ((newsrvmaxio + NFS_MAXXDR) * nfs_bufpackets >
239 	    sb_max_adj) {
240 		/*
241 		 * Suggest vfs.nfs.bufpackets * maximum RPC message for
242 		 * sb_max_adj.
243 		 */
244 		tval = (newsrvmaxio + NFS_MAXXDR) * nfs_bufpackets;
245 
246 		/*
247 		 * Convert suggested sb_max_adj value to a suggested
248 		 * sb_max value, which is what is set via kern.ipc.maxsockbuf.
249 		 * Perform the inverse calculation of (from uipc_sockbuf.c):
250 		 * sb_max_adj = (u_quad_t)sb_max * MCLBYTES /
251 		 *     (MSIZE + MCLBYTES);
252 		 * XXX If the calculation of sb_max_adj from sb_max changes,
253 		 *     this calculation must be changed as well.
254 		 */
255 		tval *= (MSIZE + MCLBYTES);  /* Brackets for readability. */
256 		tval += MCLBYTES - 1;        /* Round up divide. */
257 		tval /= MCLBYTES;
258 		printf("nfsd: set kern.ipc.maxsockbuf to a minimum of "
259 		    "%ju to support %ubyte NFS I/O\n", (uintmax_t)tval,
260 		    newsrvmaxio);
261 		return (EINVAL);
262 	}
263 
264 	NFSD_LOCK();
265 	if (newnfs_numnfsd != 0) {
266 		NFSD_UNLOCK();
267 		printf("nfsd: cannot set vfs.nfsd.srvmaxio when nfsd "
268 		    "threads are running\n");
269 		return (EINVAL);
270 	}
271 
272 
273 	nfs_srvmaxio = newsrvmaxio;
274 	NFSD_UNLOCK();
275 	return (0);
276 }
277 SYSCTL_PROC(_vfs_nfsd, OID_AUTO, srvmaxio,
278     CTLTYPE_UINT | CTLFLAG_MPSAFE | CTLFLAG_RW, NULL, 0,
279     sysctl_srvmaxio, "IU", "Maximum I/O size in bytes");
280 
281 #define	MAX_REORDERED_RPC	16
282 #define	NUM_HEURISTIC		1031
283 #define	NHUSE_INIT		64
284 #define	NHUSE_INC		16
285 #define	NHUSE_MAX		2048
286 
287 static struct nfsheur {
288 	struct vnode *nh_vp;	/* vp to match (unreferenced pointer) */
289 	off_t nh_nextoff;	/* next offset for sequential detection */
290 	int nh_use;		/* use count for selection */
291 	int nh_seqcount;	/* heuristic */
292 } nfsheur[NUM_HEURISTIC];
293 
294 /*
295  * Heuristic to detect sequential operation.
296  */
297 static struct nfsheur *
298 nfsrv_sequential_heuristic(struct uio *uio, struct vnode *vp)
299 {
300 	struct nfsheur *nh;
301 	int hi, try;
302 
303 	/* Locate best candidate. */
304 	try = 32;
305 	hi = ((int)(vm_offset_t)vp / sizeof(struct vnode)) % NUM_HEURISTIC;
306 	nh = &nfsheur[hi];
307 	while (try--) {
308 		if (nfsheur[hi].nh_vp == vp) {
309 			nh = &nfsheur[hi];
310 			break;
311 		}
312 		if (nfsheur[hi].nh_use > 0)
313 			--nfsheur[hi].nh_use;
314 		hi = (hi + 1) % NUM_HEURISTIC;
315 		if (nfsheur[hi].nh_use < nh->nh_use)
316 			nh = &nfsheur[hi];
317 	}
318 
319 	/* Initialize hint if this is a new file. */
320 	if (nh->nh_vp != vp) {
321 		nh->nh_vp = vp;
322 		nh->nh_nextoff = uio->uio_offset;
323 		nh->nh_use = NHUSE_INIT;
324 		if (uio->uio_offset == 0)
325 			nh->nh_seqcount = 4;
326 		else
327 			nh->nh_seqcount = 1;
328 	}
329 
330 	/* Calculate heuristic. */
331 	if ((uio->uio_offset == 0 && nh->nh_seqcount > 0) ||
332 	    uio->uio_offset == nh->nh_nextoff) {
333 		/* See comments in vfs_vnops.c:sequential_heuristic(). */
334 		nh->nh_seqcount += howmany(uio->uio_resid, 16384);
335 		if (nh->nh_seqcount > IO_SEQMAX)
336 			nh->nh_seqcount = IO_SEQMAX;
337 	} else if (qabs(uio->uio_offset - nh->nh_nextoff) <= MAX_REORDERED_RPC *
338 	    imax(vp->v_mount->mnt_stat.f_iosize, uio->uio_resid)) {
339 		/* Probably a reordered RPC, leave seqcount alone. */
340 	} else if (nh->nh_seqcount > 1) {
341 		nh->nh_seqcount /= 2;
342 	} else {
343 		nh->nh_seqcount = 0;
344 	}
345 	nh->nh_use += NHUSE_INC;
346 	if (nh->nh_use > NHUSE_MAX)
347 		nh->nh_use = NHUSE_MAX;
348 	return (nh);
349 }
350 
351 /*
352  * Get attributes into nfsvattr structure.
353  */
354 int
355 nfsvno_getattr(struct vnode *vp, struct nfsvattr *nvap,
356     struct nfsrv_descript *nd, struct thread *p, int vpislocked,
357     nfsattrbit_t *attrbitp)
358 {
359 	int error, gotattr, lockedit = 0;
360 	struct nfsvattr na;
361 
362 	if (vpislocked == 0) {
363 		/*
364 		 * When vpislocked == 0, the vnode is either exclusively
365 		 * locked by this thread or not locked by this thread.
366 		 * As such, shared lock it, if not exclusively locked.
367 		 */
368 		if (NFSVOPISLOCKED(vp) != LK_EXCLUSIVE) {
369 			lockedit = 1;
370 			NFSVOPLOCK(vp, LK_SHARED | LK_RETRY);
371 		}
372 	}
373 
374 	/*
375 	 * Acquire the Change, Size, TimeAccess, TimeModify and SpaceUsed
376 	 * attributes, as required.
377 	 * This needs to be done for regular files if:
378 	 * - non-NFSv4 RPCs or
379 	 * - when attrbitp == NULL or
380 	 * - an NFSv4 RPC with any of the above attributes in attrbitp.
381 	 * A return of 0 for nfsrv_proxyds() indicates that it has acquired
382 	 * these attributes.  nfsrv_proxyds() will return an error if the
383 	 * server is not a pNFS one.
384 	 */
385 	gotattr = 0;
386 	if (vp->v_type == VREG && nfsrv_devidcnt > 0 && (attrbitp == NULL ||
387 	    (nd->nd_flag & ND_NFSV4) == 0 ||
388 	    NFSISSET_ATTRBIT(attrbitp, NFSATTRBIT_CHANGE) ||
389 	    NFSISSET_ATTRBIT(attrbitp, NFSATTRBIT_SIZE) ||
390 	    NFSISSET_ATTRBIT(attrbitp, NFSATTRBIT_TIMEACCESS) ||
391 	    NFSISSET_ATTRBIT(attrbitp, NFSATTRBIT_TIMEMODIFY) ||
392 	    NFSISSET_ATTRBIT(attrbitp, NFSATTRBIT_SPACEUSED))) {
393 		error = nfsrv_proxyds(vp, 0, 0, nd->nd_cred, p,
394 		    NFSPROC_GETATTR, NULL, NULL, NULL, &na, NULL, NULL, 0,
395 		    NULL);
396 		if (error == 0)
397 			gotattr = 1;
398 	}
399 
400 	error = VOP_GETATTR(vp, &nvap->na_vattr, nd->nd_cred);
401 	if (lockedit != 0)
402 		NFSVOPUNLOCK(vp);
403 
404 	/*
405 	 * If we got the Change, Size and Modify Time from the DS,
406 	 * replace them.
407 	 */
408 	if (gotattr != 0) {
409 		nvap->na_atime = na.na_atime;
410 		nvap->na_mtime = na.na_mtime;
411 		nvap->na_filerev = na.na_filerev;
412 		nvap->na_size = na.na_size;
413 		nvap->na_bytes = na.na_bytes;
414 	}
415 	NFSD_DEBUG(4, "nfsvno_getattr: gotattr=%d err=%d chg=%ju\n", gotattr,
416 	    error, (uintmax_t)na.na_filerev);
417 
418 	NFSEXITCODE(error);
419 	return (error);
420 }
421 
422 /*
423  * Get a file handle for a vnode.
424  */
425 int
426 nfsvno_getfh(struct vnode *vp, fhandle_t *fhp, struct thread *p)
427 {
428 	int error;
429 
430 	NFSBZERO((caddr_t)fhp, sizeof(fhandle_t));
431 	fhp->fh_fsid = vp->v_mount->mnt_stat.f_fsid;
432 	error = VOP_VPTOFH(vp, &fhp->fh_fid);
433 
434 	NFSEXITCODE(error);
435 	return (error);
436 }
437 
438 /*
439  * Perform access checking for vnodes obtained from file handles that would
440  * refer to files already opened by a Unix client. You cannot just use
441  * vn_writechk() and VOP_ACCESSX() for two reasons.
442  * 1 - You must check for exported rdonly as well as MNT_RDONLY for the write
443  *     case.
444  * 2 - The owner is to be given access irrespective of mode bits for some
445  *     operations, so that processes that chmod after opening a file don't
446  *     break.
447  */
448 int
449 nfsvno_accchk(struct vnode *vp, accmode_t accmode, struct ucred *cred,
450     struct nfsexstuff *exp, struct thread *p, int override, int vpislocked,
451     u_int32_t *supportedtypep)
452 {
453 	struct vattr vattr;
454 	int error = 0, getret = 0;
455 
456 	if (vpislocked == 0) {
457 		if (NFSVOPLOCK(vp, LK_SHARED) != 0) {
458 			error = EPERM;
459 			goto out;
460 		}
461 	}
462 	if (accmode & VWRITE) {
463 		/* Just vn_writechk() changed to check rdonly */
464 		/*
465 		 * Disallow write attempts on read-only file systems;
466 		 * unless the file is a socket or a block or character
467 		 * device resident on the file system.
468 		 */
469 		if (NFSVNO_EXRDONLY(exp) ||
470 		    (vp->v_mount->mnt_flag & MNT_RDONLY)) {
471 			switch (vp->v_type) {
472 			case VREG:
473 			case VDIR:
474 			case VLNK:
475 				error = EROFS;
476 			default:
477 				break;
478 			}
479 		}
480 		/*
481 		 * If there's shared text associated with
482 		 * the inode, try to free it up once.  If
483 		 * we fail, we can't allow writing.
484 		 */
485 		if (VOP_IS_TEXT(vp) && error == 0)
486 			error = ETXTBSY;
487 	}
488 	if (error != 0) {
489 		if (vpislocked == 0)
490 			NFSVOPUNLOCK(vp);
491 		goto out;
492 	}
493 
494 	/*
495 	 * Should the override still be applied when ACLs are enabled?
496 	 */
497 	error = VOP_ACCESSX(vp, accmode, cred, p);
498 	if (error != 0 && (accmode & (VDELETE | VDELETE_CHILD))) {
499 		/*
500 		 * Try again with VEXPLICIT_DENY, to see if the test for
501 		 * deletion is supported.
502 		 */
503 		error = VOP_ACCESSX(vp, accmode | VEXPLICIT_DENY, cred, p);
504 		if (error == 0) {
505 			if (vp->v_type == VDIR) {
506 				accmode &= ~(VDELETE | VDELETE_CHILD);
507 				accmode |= VWRITE;
508 				error = VOP_ACCESSX(vp, accmode, cred, p);
509 			} else if (supportedtypep != NULL) {
510 				*supportedtypep &= ~NFSACCESS_DELETE;
511 			}
512 		}
513 	}
514 
515 	/*
516 	 * Allow certain operations for the owner (reads and writes
517 	 * on files that are already open).
518 	 */
519 	if (override != NFSACCCHK_NOOVERRIDE &&
520 	    (error == EPERM || error == EACCES)) {
521 		if (cred->cr_uid == 0 && (override & NFSACCCHK_ALLOWROOT))
522 			error = 0;
523 		else if (override & NFSACCCHK_ALLOWOWNER) {
524 			getret = VOP_GETATTR(vp, &vattr, cred);
525 			if (getret == 0 && cred->cr_uid == vattr.va_uid)
526 				error = 0;
527 		}
528 	}
529 	if (vpislocked == 0)
530 		NFSVOPUNLOCK(vp);
531 
532 out:
533 	NFSEXITCODE(error);
534 	return (error);
535 }
536 
537 /*
538  * Set attribute(s) vnop.
539  */
540 int
541 nfsvno_setattr(struct vnode *vp, struct nfsvattr *nvap, struct ucred *cred,
542     struct thread *p, struct nfsexstuff *exp)
543 {
544 	u_quad_t savsize = 0;
545 	int error, savedit;
546 	time_t savbtime;
547 
548 	/*
549 	 * If this is an exported file system and a pNFS service is running,
550 	 * don't VOP_SETATTR() of size for the MDS file system.
551 	 */
552 	savedit = 0;
553 	error = 0;
554 	if (vp->v_type == VREG && (vp->v_mount->mnt_flag & MNT_EXPORTED) != 0 &&
555 	    nfsrv_devidcnt != 0 && nvap->na_vattr.va_size != VNOVAL &&
556 	    nvap->na_vattr.va_size > 0) {
557 		savsize = nvap->na_vattr.va_size;
558 		nvap->na_vattr.va_size = VNOVAL;
559 		if (nvap->na_vattr.va_uid != (uid_t)VNOVAL ||
560 		    nvap->na_vattr.va_gid != (gid_t)VNOVAL ||
561 		    nvap->na_vattr.va_mode != (mode_t)VNOVAL ||
562 		    nvap->na_vattr.va_atime.tv_sec != VNOVAL ||
563 		    nvap->na_vattr.va_mtime.tv_sec != VNOVAL)
564 			savedit = 1;
565 		else
566 			savedit = 2;
567 	}
568 	if (savedit != 2)
569 		error = VOP_SETATTR(vp, &nvap->na_vattr, cred);
570 	if (savedit != 0)
571 		nvap->na_vattr.va_size = savsize;
572 	if (error == 0 && (nvap->na_vattr.va_uid != (uid_t)VNOVAL ||
573 	    nvap->na_vattr.va_gid != (gid_t)VNOVAL ||
574 	    nvap->na_vattr.va_size != VNOVAL ||
575 	    nvap->na_vattr.va_mode != (mode_t)VNOVAL ||
576 	    nvap->na_vattr.va_atime.tv_sec != VNOVAL ||
577 	    nvap->na_vattr.va_mtime.tv_sec != VNOVAL)) {
578 		/* Never modify birthtime on a DS file. */
579 		savbtime = nvap->na_vattr.va_birthtime.tv_sec;
580 		nvap->na_vattr.va_birthtime.tv_sec = VNOVAL;
581 		/* For a pNFS server, set the attributes on the DS file. */
582 		error = nfsrv_proxyds(vp, 0, 0, cred, p, NFSPROC_SETATTR,
583 		    NULL, NULL, NULL, nvap, NULL, NULL, 0, NULL);
584 		nvap->na_vattr.va_birthtime.tv_sec = savbtime;
585 		if (error == ENOENT)
586 			error = 0;
587 	}
588 	NFSEXITCODE(error);
589 	return (error);
590 }
591 
592 /*
593  * Set up nameidata for a lookup() call and do it.
594  */
595 int
596 nfsvno_namei(struct nfsrv_descript *nd, struct nameidata *ndp,
597     struct vnode *dp, int islocked, struct nfsexstuff *exp, struct thread *p,
598     struct vnode **retdirp)
599 {
600 	struct componentname *cnp = &ndp->ni_cnd;
601 	int i;
602 	struct iovec aiov;
603 	struct uio auio;
604 	int lockleaf = (cnp->cn_flags & LOCKLEAF) != 0, linklen;
605 	int error = 0;
606 	char *cp;
607 
608 	*retdirp = NULL;
609 	cnp->cn_nameptr = cnp->cn_pnbuf;
610 	ndp->ni_lcf = 0;
611 	/*
612 	 * Extract and set starting directory.
613 	 */
614 	if (dp->v_type != VDIR) {
615 		if (islocked)
616 			vput(dp);
617 		else
618 			vrele(dp);
619 		nfsvno_relpathbuf(ndp);
620 		error = ENOTDIR;
621 		goto out1;
622 	}
623 	if (islocked)
624 		NFSVOPUNLOCK(dp);
625 	VREF(dp);
626 	*retdirp = dp;
627 	if (NFSVNO_EXRDONLY(exp))
628 		cnp->cn_flags |= RDONLY;
629 	ndp->ni_segflg = UIO_SYSSPACE;
630 
631 	if (nd->nd_flag & ND_PUBLOOKUP) {
632 		ndp->ni_loopcnt = 0;
633 		if (cnp->cn_pnbuf[0] == '/') {
634 			vrele(dp);
635 			/*
636 			 * Check for degenerate pathnames here, since lookup()
637 			 * panics on them.
638 			 */
639 			for (i = 1; i < ndp->ni_pathlen; i++)
640 				if (cnp->cn_pnbuf[i] != '/')
641 					break;
642 			if (i == ndp->ni_pathlen) {
643 				error = NFSERR_ACCES;
644 				goto out;
645 			}
646 			dp = rootvnode;
647 			VREF(dp);
648 		}
649 	} else if ((nfsrv_enable_crossmntpt == 0 && NFSVNO_EXPORTED(exp)) ||
650 	    (nd->nd_flag & ND_NFSV4) == 0) {
651 		/*
652 		 * Only cross mount points for NFSv4 when doing a
653 		 * mount while traversing the file system above
654 		 * the mount point, unless nfsrv_enable_crossmntpt is set.
655 		 */
656 		cnp->cn_flags |= NOCROSSMOUNT;
657 	}
658 
659 	/*
660 	 * Initialize for scan, set ni_startdir and bump ref on dp again
661 	 * because lookup() will dereference ni_startdir.
662 	 */
663 
664 	cnp->cn_thread = p;
665 	ndp->ni_startdir = dp;
666 	ndp->ni_rootdir = rootvnode;
667 	ndp->ni_topdir = NULL;
668 
669 	if (!lockleaf)
670 		cnp->cn_flags |= LOCKLEAF;
671 	for (;;) {
672 		cnp->cn_nameptr = cnp->cn_pnbuf;
673 		/*
674 		 * Call lookup() to do the real work.  If an error occurs,
675 		 * ndp->ni_vp and ni_dvp are left uninitialized or NULL and
676 		 * we do not have to dereference anything before returning.
677 		 * In either case ni_startdir will be dereferenced and NULLed
678 		 * out.
679 		 */
680 		error = lookup(ndp);
681 		if (error)
682 			break;
683 
684 		/*
685 		 * Check for encountering a symbolic link.  Trivial
686 		 * termination occurs if no symlink encountered.
687 		 */
688 		if ((cnp->cn_flags & ISSYMLINK) == 0) {
689 			if ((cnp->cn_flags & (SAVENAME | SAVESTART)) == 0)
690 				nfsvno_relpathbuf(ndp);
691 			if (ndp->ni_vp && !lockleaf)
692 				NFSVOPUNLOCK(ndp->ni_vp);
693 			break;
694 		}
695 
696 		/*
697 		 * Validate symlink
698 		 */
699 		if ((cnp->cn_flags & LOCKPARENT) && ndp->ni_pathlen == 1)
700 			NFSVOPUNLOCK(ndp->ni_dvp);
701 		if (!(nd->nd_flag & ND_PUBLOOKUP)) {
702 			error = EINVAL;
703 			goto badlink2;
704 		}
705 
706 		if (ndp->ni_loopcnt++ >= MAXSYMLINKS) {
707 			error = ELOOP;
708 			goto badlink2;
709 		}
710 		if (ndp->ni_pathlen > 1)
711 			cp = uma_zalloc(namei_zone, M_WAITOK);
712 		else
713 			cp = cnp->cn_pnbuf;
714 		aiov.iov_base = cp;
715 		aiov.iov_len = MAXPATHLEN;
716 		auio.uio_iov = &aiov;
717 		auio.uio_iovcnt = 1;
718 		auio.uio_offset = 0;
719 		auio.uio_rw = UIO_READ;
720 		auio.uio_segflg = UIO_SYSSPACE;
721 		auio.uio_td = NULL;
722 		auio.uio_resid = MAXPATHLEN;
723 		error = VOP_READLINK(ndp->ni_vp, &auio, cnp->cn_cred);
724 		if (error) {
725 		badlink1:
726 			if (ndp->ni_pathlen > 1)
727 				uma_zfree(namei_zone, cp);
728 		badlink2:
729 			vrele(ndp->ni_dvp);
730 			vput(ndp->ni_vp);
731 			break;
732 		}
733 		linklen = MAXPATHLEN - auio.uio_resid;
734 		if (linklen == 0) {
735 			error = ENOENT;
736 			goto badlink1;
737 		}
738 		if (linklen + ndp->ni_pathlen >= MAXPATHLEN) {
739 			error = ENAMETOOLONG;
740 			goto badlink1;
741 		}
742 
743 		/*
744 		 * Adjust or replace path
745 		 */
746 		if (ndp->ni_pathlen > 1) {
747 			NFSBCOPY(ndp->ni_next, cp + linklen, ndp->ni_pathlen);
748 			uma_zfree(namei_zone, cnp->cn_pnbuf);
749 			cnp->cn_pnbuf = cp;
750 		} else
751 			cnp->cn_pnbuf[linklen] = '\0';
752 		ndp->ni_pathlen += linklen;
753 
754 		/*
755 		 * Cleanup refs for next loop and check if root directory
756 		 * should replace current directory.  Normally ni_dvp
757 		 * becomes the new base directory and is cleaned up when
758 		 * we loop.  Explicitly null pointers after invalidation
759 		 * to clarify operation.
760 		 */
761 		vput(ndp->ni_vp);
762 		ndp->ni_vp = NULL;
763 
764 		if (cnp->cn_pnbuf[0] == '/') {
765 			vrele(ndp->ni_dvp);
766 			ndp->ni_dvp = ndp->ni_rootdir;
767 			VREF(ndp->ni_dvp);
768 		}
769 		ndp->ni_startdir = ndp->ni_dvp;
770 		ndp->ni_dvp = NULL;
771 	}
772 	if (!lockleaf)
773 		cnp->cn_flags &= ~LOCKLEAF;
774 
775 out:
776 	if (error) {
777 		nfsvno_relpathbuf(ndp);
778 		ndp->ni_vp = NULL;
779 		ndp->ni_dvp = NULL;
780 		ndp->ni_startdir = NULL;
781 	} else if ((ndp->ni_cnd.cn_flags & (WANTPARENT|LOCKPARENT)) == 0) {
782 		ndp->ni_dvp = NULL;
783 	}
784 
785 out1:
786 	NFSEXITCODE2(error, nd);
787 	return (error);
788 }
789 
790 /*
791  * Set up a pathname buffer and return a pointer to it and, optionally
792  * set a hash pointer.
793  */
794 void
795 nfsvno_setpathbuf(struct nameidata *ndp, char **bufpp, u_long **hashpp)
796 {
797 	struct componentname *cnp = &ndp->ni_cnd;
798 
799 	cnp->cn_flags |= (NOMACCHECK | HASBUF);
800 	cnp->cn_pnbuf = uma_zalloc(namei_zone, M_WAITOK);
801 	if (hashpp != NULL)
802 		*hashpp = NULL;
803 	*bufpp = cnp->cn_pnbuf;
804 }
805 
806 /*
807  * Release the above path buffer, if not released by nfsvno_namei().
808  */
809 void
810 nfsvno_relpathbuf(struct nameidata *ndp)
811 {
812 
813 	if ((ndp->ni_cnd.cn_flags & HASBUF) == 0)
814 		panic("nfsrelpath");
815 	uma_zfree(namei_zone, ndp->ni_cnd.cn_pnbuf);
816 	ndp->ni_cnd.cn_flags &= ~HASBUF;
817 }
818 
819 /*
820  * Readlink vnode op into an mbuf list.
821  */
822 int
823 nfsvno_readlink(struct vnode *vp, struct ucred *cred, int maxextsiz,
824     struct thread *p, struct mbuf **mpp, struct mbuf **mpendp, int *lenp)
825 {
826 	struct iovec *iv;
827 	struct uio io, *uiop = &io;
828 	struct mbuf *mp, *mp3;
829 	int len, tlen, error = 0;
830 
831 	len = NFS_MAXPATHLEN;
832 	if (maxextsiz > 0)
833 		uiop->uio_iovcnt = nfsrv_createiovec_extpgs(len, maxextsiz,
834 		    &mp3, &mp, &iv);
835 	else
836 		uiop->uio_iovcnt = nfsrv_createiovec(len, &mp3, &mp, &iv);
837 	uiop->uio_iov = iv;
838 	uiop->uio_offset = 0;
839 	uiop->uio_resid = len;
840 	uiop->uio_rw = UIO_READ;
841 	uiop->uio_segflg = UIO_SYSSPACE;
842 	uiop->uio_td = NULL;
843 	error = VOP_READLINK(vp, uiop, cred);
844 	free(iv, M_TEMP);
845 	if (error) {
846 		m_freem(mp3);
847 		*lenp = 0;
848 		goto out;
849 	}
850 	if (uiop->uio_resid > 0) {
851 		len -= uiop->uio_resid;
852 		tlen = NFSM_RNDUP(len);
853 		if (tlen == 0) {
854 			m_freem(mp3);
855 			mp3 = mp = NULL;
856 		} else if (tlen != NFS_MAXPATHLEN || tlen != len)
857 			mp = nfsrv_adj(mp3, NFS_MAXPATHLEN - tlen,
858 			    tlen - len);
859 	}
860 	*lenp = len;
861 	*mpp = mp3;
862 	*mpendp = mp;
863 
864 out:
865 	NFSEXITCODE(error);
866 	return (error);
867 }
868 
869 /*
870  * Create an mbuf chain and an associated iovec that can be used to Read
871  * or Getextattr of data.
872  * Upon success, return pointers to the first and last mbufs in the chain
873  * plus the malloc'd iovec and its iovlen.
874  */
875 static int
876 nfsrv_createiovec(int len, struct mbuf **mpp, struct mbuf **mpendp,
877     struct iovec **ivp)
878 {
879 	struct mbuf *m, *m2 = NULL, *m3;
880 	struct iovec *iv;
881 	int i, left, siz;
882 
883 	left = len;
884 	m3 = NULL;
885 	/*
886 	 * Generate the mbuf list with the uio_iov ref. to it.
887 	 */
888 	i = 0;
889 	while (left > 0) {
890 		NFSMGET(m);
891 		MCLGET(m, M_WAITOK);
892 		m->m_len = 0;
893 		siz = min(M_TRAILINGSPACE(m), left);
894 		left -= siz;
895 		i++;
896 		if (m3)
897 			m2->m_next = m;
898 		else
899 			m3 = m;
900 		m2 = m;
901 	}
902 	*ivp = iv = malloc(i * sizeof (struct iovec), M_TEMP, M_WAITOK);
903 	m = m3;
904 	left = len;
905 	i = 0;
906 	while (left > 0) {
907 		if (m == NULL)
908 			panic("nfsrv_createiovec iov");
909 		siz = min(M_TRAILINGSPACE(m), left);
910 		if (siz > 0) {
911 			iv->iov_base = mtod(m, caddr_t) + m->m_len;
912 			iv->iov_len = siz;
913 			m->m_len += siz;
914 			left -= siz;
915 			iv++;
916 			i++;
917 		}
918 		m = m->m_next;
919 	}
920 	*mpp = m3;
921 	*mpendp = m2;
922 	return (i);
923 }
924 
925 /*
926  * Create an mbuf chain and an associated iovec that can be used to Read
927  * or Getextattr of data.
928  * Upon success, return pointers to the first and last mbufs in the chain
929  * plus the malloc'd iovec and its iovlen.
930  * Same as above, but creates ext_pgs mbuf(s).
931  */
932 static int
933 nfsrv_createiovec_extpgs(int len, int maxextsiz, struct mbuf **mpp,
934     struct mbuf **mpendp, struct iovec **ivp)
935 {
936 	struct mbuf *m, *m2 = NULL, *m3;
937 	struct iovec *iv;
938 	int i, left, pgno, siz;
939 
940 	left = len;
941 	m3 = NULL;
942 	/*
943 	 * Generate the mbuf list with the uio_iov ref. to it.
944 	 */
945 	i = 0;
946 	while (left > 0) {
947 		siz = min(left, maxextsiz);
948 		m = mb_alloc_ext_plus_pages(siz, M_WAITOK);
949 		left -= siz;
950 		i += m->m_epg_npgs;
951 		if (m3 != NULL)
952 			m2->m_next = m;
953 		else
954 			m3 = m;
955 		m2 = m;
956 	}
957 	*ivp = iv = malloc(i * sizeof (struct iovec), M_TEMP, M_WAITOK);
958 	m = m3;
959 	left = len;
960 	i = 0;
961 	pgno = 0;
962 	while (left > 0) {
963 		if (m == NULL)
964 			panic("nfsvno_createiovec_extpgs iov");
965 		siz = min(PAGE_SIZE, left);
966 		if (siz > 0) {
967 			iv->iov_base = (void *)PHYS_TO_DMAP(m->m_epg_pa[pgno]);
968 			iv->iov_len = siz;
969 			m->m_len += siz;
970 			if (pgno == m->m_epg_npgs - 1)
971 				m->m_epg_last_len = siz;
972 			left -= siz;
973 			iv++;
974 			i++;
975 			pgno++;
976 		}
977 		if (pgno == m->m_epg_npgs && left > 0) {
978 			m = m->m_next;
979 			if (m == NULL)
980 				panic("nfsvno_createiovec_extpgs iov");
981 			pgno = 0;
982 		}
983 	}
984 	*mpp = m3;
985 	*mpendp = m2;
986 	return (i);
987 }
988 
989 /*
990  * Read vnode op call into mbuf list.
991  */
992 int
993 nfsvno_read(struct vnode *vp, off_t off, int cnt, struct ucred *cred,
994     int maxextsiz, struct thread *p, struct mbuf **mpp,
995     struct mbuf **mpendp)
996 {
997 	struct mbuf *m;
998 	struct iovec *iv;
999 	int error = 0, len, tlen, ioflag = 0;
1000 	struct mbuf *m3;
1001 	struct uio io, *uiop = &io;
1002 	struct nfsheur *nh;
1003 
1004 	/*
1005 	 * Attempt to read from a DS file. A return of ENOENT implies
1006 	 * there is no DS file to read.
1007 	 */
1008 	error = nfsrv_proxyds(vp, off, cnt, cred, p, NFSPROC_READDS, mpp,
1009 	    NULL, mpendp, NULL, NULL, NULL, 0, NULL);
1010 	if (error != ENOENT)
1011 		return (error);
1012 
1013 	len = NFSM_RNDUP(cnt);
1014 	if (maxextsiz > 0)
1015 		uiop->uio_iovcnt = nfsrv_createiovec_extpgs(len, maxextsiz,
1016 		    &m3, &m, &iv);
1017 	else
1018 		uiop->uio_iovcnt = nfsrv_createiovec(len, &m3, &m, &iv);
1019 	uiop->uio_iov = iv;
1020 	uiop->uio_offset = off;
1021 	uiop->uio_resid = len;
1022 	uiop->uio_rw = UIO_READ;
1023 	uiop->uio_segflg = UIO_SYSSPACE;
1024 	uiop->uio_td = NULL;
1025 	nh = nfsrv_sequential_heuristic(uiop, vp);
1026 	ioflag |= nh->nh_seqcount << IO_SEQSHIFT;
1027 	/* XXX KDM make this more systematic? */
1028 	nfsstatsv1.srvbytes[NFSV4OP_READ] += uiop->uio_resid;
1029 	error = VOP_READ(vp, uiop, IO_NODELOCKED | ioflag, cred);
1030 	free(iv, M_TEMP);
1031 	if (error) {
1032 		m_freem(m3);
1033 		*mpp = NULL;
1034 		goto out;
1035 	}
1036 	nh->nh_nextoff = uiop->uio_offset;
1037 	tlen = len - uiop->uio_resid;
1038 	cnt = cnt < tlen ? cnt : tlen;
1039 	tlen = NFSM_RNDUP(cnt);
1040 	if (tlen == 0) {
1041 		m_freem(m3);
1042 		m3 = m = NULL;
1043 	} else if (len != tlen || tlen != cnt)
1044 		m = nfsrv_adj(m3, len - tlen, tlen - cnt);
1045 	*mpp = m3;
1046 	*mpendp = m;
1047 
1048 out:
1049 	NFSEXITCODE(error);
1050 	return (error);
1051 }
1052 
1053 /*
1054  * Create the iovec for the mbuf chain passed in as an argument.
1055  * The "cp" argument is where the data starts within the first mbuf in
1056  * the chain. It returns the iovec and the iovcnt.
1057  */
1058 static int
1059 nfsrv_createiovecw(int retlen, struct mbuf *m, char *cp, struct iovec **ivpp,
1060     int *iovcntp)
1061 {
1062 	struct mbuf *mp;
1063 	struct iovec *ivp;
1064 	int cnt, i, len;
1065 
1066 	/*
1067 	 * Loop through the mbuf chain, counting how many mbufs are a
1068 	 * part of this write operation, so the iovec size is known.
1069 	 */
1070 	cnt = 0;
1071 	len = retlen;
1072 	mp = m;
1073 	i = mtod(mp, caddr_t) + mp->m_len - cp;
1074 	while (len > 0) {
1075 		if (i > 0) {
1076 			len -= i;
1077 			cnt++;
1078 		}
1079 		mp = mp->m_next;
1080 		if (!mp) {
1081 			if (len > 0)
1082 				return (EBADRPC);
1083 		} else
1084 			i = mp->m_len;
1085 	}
1086 
1087 	/* Now, create the iovec. */
1088 	mp = m;
1089 	*ivpp = ivp = malloc(cnt * sizeof (struct iovec), M_TEMP,
1090 	    M_WAITOK);
1091 	*iovcntp = cnt;
1092 	i = mtod(mp, caddr_t) + mp->m_len - cp;
1093 	len = retlen;
1094 	while (len > 0) {
1095 		if (mp == NULL)
1096 			panic("nfsrv_createiovecw");
1097 		if (i > 0) {
1098 			i = min(i, len);
1099 			ivp->iov_base = cp;
1100 			ivp->iov_len = i;
1101 			ivp++;
1102 			len -= i;
1103 		}
1104 		mp = mp->m_next;
1105 		if (mp) {
1106 			i = mp->m_len;
1107 			cp = mtod(mp, caddr_t);
1108 		}
1109 	}
1110 	return (0);
1111 }
1112 
1113 /*
1114  * Write vnode op from an mbuf list.
1115  */
1116 int
1117 nfsvno_write(struct vnode *vp, off_t off, int retlen, int *stable,
1118     struct mbuf *mp, char *cp, struct ucred *cred, struct thread *p)
1119 {
1120 	struct iovec *iv;
1121 	int cnt, ioflags, error;
1122 	struct uio io, *uiop = &io;
1123 	struct nfsheur *nh;
1124 
1125 	/*
1126 	 * Attempt to write to a DS file. A return of ENOENT implies
1127 	 * there is no DS file to write.
1128 	 */
1129 	error = nfsrv_proxyds(vp, off, retlen, cred, p, NFSPROC_WRITEDS,
1130 	    &mp, cp, NULL, NULL, NULL, NULL, 0, NULL);
1131 	if (error != ENOENT) {
1132 		*stable = NFSWRITE_FILESYNC;
1133 		return (error);
1134 	}
1135 
1136 	if (*stable == NFSWRITE_UNSTABLE)
1137 		ioflags = IO_NODELOCKED;
1138 	else
1139 		ioflags = (IO_SYNC | IO_NODELOCKED);
1140 	error = nfsrv_createiovecw(retlen, mp, cp, &iv, &cnt);
1141 	if (error != 0)
1142 		return (error);
1143 	uiop->uio_iov = iv;
1144 	uiop->uio_iovcnt = cnt;
1145 	uiop->uio_resid = retlen;
1146 	uiop->uio_rw = UIO_WRITE;
1147 	uiop->uio_segflg = UIO_SYSSPACE;
1148 	NFSUIOPROC(uiop, p);
1149 	uiop->uio_offset = off;
1150 	nh = nfsrv_sequential_heuristic(uiop, vp);
1151 	ioflags |= nh->nh_seqcount << IO_SEQSHIFT;
1152 	/* XXX KDM make this more systematic? */
1153 	nfsstatsv1.srvbytes[NFSV4OP_WRITE] += uiop->uio_resid;
1154 	error = VOP_WRITE(vp, uiop, ioflags, cred);
1155 	if (error == 0)
1156 		nh->nh_nextoff = uiop->uio_offset;
1157 	free(iv, M_TEMP);
1158 
1159 	NFSEXITCODE(error);
1160 	return (error);
1161 }
1162 
1163 /*
1164  * Common code for creating a regular file (plus special files for V2).
1165  */
1166 int
1167 nfsvno_createsub(struct nfsrv_descript *nd, struct nameidata *ndp,
1168     struct vnode **vpp, struct nfsvattr *nvap, int *exclusive_flagp,
1169     int32_t *cverf, NFSDEV_T rdev, struct nfsexstuff *exp)
1170 {
1171 	u_quad_t tempsize;
1172 	int error;
1173 	struct thread *p = curthread;
1174 
1175 	error = nd->nd_repstat;
1176 	if (!error && ndp->ni_vp == NULL) {
1177 		if (nvap->na_type == VREG || nvap->na_type == VSOCK) {
1178 			vrele(ndp->ni_startdir);
1179 			error = VOP_CREATE(ndp->ni_dvp,
1180 			    &ndp->ni_vp, &ndp->ni_cnd, &nvap->na_vattr);
1181 			/* For a pNFS server, create the data file on a DS. */
1182 			if (error == 0 && nvap->na_type == VREG) {
1183 				/*
1184 				 * Create a data file on a DS for a pNFS server.
1185 				 * This function just returns if not
1186 				 * running a pNFS DS or the creation fails.
1187 				 */
1188 				nfsrv_pnfscreate(ndp->ni_vp, &nvap->na_vattr,
1189 				    nd->nd_cred, p);
1190 			}
1191 			VOP_VPUT_PAIR(ndp->ni_dvp, error == 0 ? &ndp->ni_vp :
1192 			    NULL, false);
1193 			nfsvno_relpathbuf(ndp);
1194 			if (!error) {
1195 				if (*exclusive_flagp) {
1196 					*exclusive_flagp = 0;
1197 					NFSVNO_ATTRINIT(nvap);
1198 					nvap->na_atime.tv_sec = cverf[0];
1199 					nvap->na_atime.tv_nsec = cverf[1];
1200 					error = VOP_SETATTR(ndp->ni_vp,
1201 					    &nvap->na_vattr, nd->nd_cred);
1202 					if (error != 0) {
1203 						vput(ndp->ni_vp);
1204 						ndp->ni_vp = NULL;
1205 						error = NFSERR_NOTSUPP;
1206 					}
1207 				}
1208 			}
1209 		/*
1210 		 * NFS V2 Only. nfsrvd_mknod() does this for V3.
1211 		 * (This implies, just get out on an error.)
1212 		 */
1213 		} else if (nvap->na_type == VCHR || nvap->na_type == VBLK ||
1214 			nvap->na_type == VFIFO) {
1215 			if (nvap->na_type == VCHR && rdev == 0xffffffff)
1216 				nvap->na_type = VFIFO;
1217                         if (nvap->na_type != VFIFO &&
1218 			    (error = priv_check_cred(nd->nd_cred, PRIV_VFS_MKNOD_DEV))) {
1219 				vrele(ndp->ni_startdir);
1220 				nfsvno_relpathbuf(ndp);
1221 				vput(ndp->ni_dvp);
1222 				goto out;
1223 			}
1224 			nvap->na_rdev = rdev;
1225 			error = VOP_MKNOD(ndp->ni_dvp, &ndp->ni_vp,
1226 			    &ndp->ni_cnd, &nvap->na_vattr);
1227 			VOP_VPUT_PAIR(ndp->ni_dvp, error == 0 ? &ndp->ni_vp :
1228 			    NULL, false);
1229 			nfsvno_relpathbuf(ndp);
1230 			vrele(ndp->ni_startdir);
1231 			if (error)
1232 				goto out;
1233 		} else {
1234 			vrele(ndp->ni_startdir);
1235 			nfsvno_relpathbuf(ndp);
1236 			vput(ndp->ni_dvp);
1237 			error = ENXIO;
1238 			goto out;
1239 		}
1240 		*vpp = ndp->ni_vp;
1241 	} else {
1242 		/*
1243 		 * Handle cases where error is already set and/or
1244 		 * the file exists.
1245 		 * 1 - clean up the lookup
1246 		 * 2 - iff !error and na_size set, truncate it
1247 		 */
1248 		vrele(ndp->ni_startdir);
1249 		nfsvno_relpathbuf(ndp);
1250 		*vpp = ndp->ni_vp;
1251 		if (ndp->ni_dvp == *vpp)
1252 			vrele(ndp->ni_dvp);
1253 		else
1254 			vput(ndp->ni_dvp);
1255 		if (!error && nvap->na_size != VNOVAL) {
1256 			error = nfsvno_accchk(*vpp, VWRITE,
1257 			    nd->nd_cred, exp, p, NFSACCCHK_NOOVERRIDE,
1258 			    NFSACCCHK_VPISLOCKED, NULL);
1259 			if (!error) {
1260 				tempsize = nvap->na_size;
1261 				NFSVNO_ATTRINIT(nvap);
1262 				nvap->na_size = tempsize;
1263 				error = VOP_SETATTR(*vpp,
1264 				    &nvap->na_vattr, nd->nd_cred);
1265 			}
1266 		}
1267 		if (error)
1268 			vput(*vpp);
1269 	}
1270 
1271 out:
1272 	NFSEXITCODE(error);
1273 	return (error);
1274 }
1275 
1276 /*
1277  * Do a mknod vnode op.
1278  */
1279 int
1280 nfsvno_mknod(struct nameidata *ndp, struct nfsvattr *nvap, struct ucred *cred,
1281     struct thread *p)
1282 {
1283 	int error = 0;
1284 	enum vtype vtyp;
1285 
1286 	vtyp = nvap->na_type;
1287 	/*
1288 	 * Iff doesn't exist, create it.
1289 	 */
1290 	if (ndp->ni_vp) {
1291 		vrele(ndp->ni_startdir);
1292 		nfsvno_relpathbuf(ndp);
1293 		vput(ndp->ni_dvp);
1294 		vrele(ndp->ni_vp);
1295 		error = EEXIST;
1296 		goto out;
1297 	}
1298 	if (vtyp != VCHR && vtyp != VBLK && vtyp != VSOCK && vtyp != VFIFO) {
1299 		vrele(ndp->ni_startdir);
1300 		nfsvno_relpathbuf(ndp);
1301 		vput(ndp->ni_dvp);
1302 		error = NFSERR_BADTYPE;
1303 		goto out;
1304 	}
1305 	if (vtyp == VSOCK) {
1306 		vrele(ndp->ni_startdir);
1307 		error = VOP_CREATE(ndp->ni_dvp, &ndp->ni_vp,
1308 		    &ndp->ni_cnd, &nvap->na_vattr);
1309 		VOP_VPUT_PAIR(ndp->ni_dvp, error == 0 ? &ndp->ni_vp : NULL,
1310 		    false);
1311 		nfsvno_relpathbuf(ndp);
1312 	} else {
1313 		if (nvap->na_type != VFIFO &&
1314 		    (error = priv_check_cred(cred, PRIV_VFS_MKNOD_DEV))) {
1315 			vrele(ndp->ni_startdir);
1316 			nfsvno_relpathbuf(ndp);
1317 			vput(ndp->ni_dvp);
1318 			goto out;
1319 		}
1320 		error = VOP_MKNOD(ndp->ni_dvp, &ndp->ni_vp,
1321 		    &ndp->ni_cnd, &nvap->na_vattr);
1322 		VOP_VPUT_PAIR(ndp->ni_dvp, error == 0 ? &ndp->ni_vp : NULL,
1323 		    false);
1324 		nfsvno_relpathbuf(ndp);
1325 		vrele(ndp->ni_startdir);
1326 		/*
1327 		 * Since VOP_MKNOD returns the ni_vp, I can't
1328 		 * see any reason to do the lookup.
1329 		 */
1330 	}
1331 
1332 out:
1333 	NFSEXITCODE(error);
1334 	return (error);
1335 }
1336 
1337 /*
1338  * Mkdir vnode op.
1339  */
1340 int
1341 nfsvno_mkdir(struct nameidata *ndp, struct nfsvattr *nvap, uid_t saved_uid,
1342     struct ucred *cred, struct thread *p, struct nfsexstuff *exp)
1343 {
1344 	int error = 0;
1345 
1346 	if (ndp->ni_vp != NULL) {
1347 		if (ndp->ni_dvp == ndp->ni_vp)
1348 			vrele(ndp->ni_dvp);
1349 		else
1350 			vput(ndp->ni_dvp);
1351 		vrele(ndp->ni_vp);
1352 		nfsvno_relpathbuf(ndp);
1353 		error = EEXIST;
1354 		goto out;
1355 	}
1356 	error = VOP_MKDIR(ndp->ni_dvp, &ndp->ni_vp, &ndp->ni_cnd,
1357 	    &nvap->na_vattr);
1358 	VOP_VPUT_PAIR(ndp->ni_dvp, error == 0 ? &ndp->ni_vp : NULL, false);
1359 	nfsvno_relpathbuf(ndp);
1360 
1361 out:
1362 	NFSEXITCODE(error);
1363 	return (error);
1364 }
1365 
1366 /*
1367  * symlink vnode op.
1368  */
1369 int
1370 nfsvno_symlink(struct nameidata *ndp, struct nfsvattr *nvap, char *pathcp,
1371     int pathlen, int not_v2, uid_t saved_uid, struct ucred *cred, struct thread *p,
1372     struct nfsexstuff *exp)
1373 {
1374 	int error = 0;
1375 
1376 	if (ndp->ni_vp) {
1377 		vrele(ndp->ni_startdir);
1378 		nfsvno_relpathbuf(ndp);
1379 		if (ndp->ni_dvp == ndp->ni_vp)
1380 			vrele(ndp->ni_dvp);
1381 		else
1382 			vput(ndp->ni_dvp);
1383 		vrele(ndp->ni_vp);
1384 		error = EEXIST;
1385 		goto out;
1386 	}
1387 
1388 	error = VOP_SYMLINK(ndp->ni_dvp, &ndp->ni_vp, &ndp->ni_cnd,
1389 	    &nvap->na_vattr, pathcp);
1390 	/*
1391 	 * Although FreeBSD still had the lookup code in
1392 	 * it for 7/current, there doesn't seem to be any
1393 	 * point, since VOP_SYMLINK() returns the ni_vp.
1394 	 * Just vput it for v2.
1395 	 */
1396 	VOP_VPUT_PAIR(ndp->ni_dvp, &ndp->ni_vp, !not_v2 && error == 0);
1397 	vrele(ndp->ni_startdir);
1398 	nfsvno_relpathbuf(ndp);
1399 
1400 out:
1401 	NFSEXITCODE(error);
1402 	return (error);
1403 }
1404 
1405 /*
1406  * Parse symbolic link arguments.
1407  * This function has an ugly side effect. It will malloc() an area for
1408  * the symlink and set iov_base to point to it, only if it succeeds.
1409  * So, if it returns with uiop->uio_iov->iov_base != NULL, that must
1410  * be FREE'd later.
1411  */
1412 int
1413 nfsvno_getsymlink(struct nfsrv_descript *nd, struct nfsvattr *nvap,
1414     struct thread *p, char **pathcpp, int *lenp)
1415 {
1416 	u_int32_t *tl;
1417 	char *pathcp = NULL;
1418 	int error = 0, len;
1419 	struct nfsv2_sattr *sp;
1420 
1421 	*pathcpp = NULL;
1422 	*lenp = 0;
1423 	if ((nd->nd_flag & ND_NFSV3) &&
1424 	    (error = nfsrv_sattr(nd, NULL, nvap, NULL, NULL, p)))
1425 		goto nfsmout;
1426 	NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
1427 	len = fxdr_unsigned(int, *tl);
1428 	if (len > NFS_MAXPATHLEN || len <= 0) {
1429 		error = EBADRPC;
1430 		goto nfsmout;
1431 	}
1432 	pathcp = malloc(len + 1, M_TEMP, M_WAITOK);
1433 	error = nfsrv_mtostr(nd, pathcp, len);
1434 	if (error)
1435 		goto nfsmout;
1436 	if (nd->nd_flag & ND_NFSV2) {
1437 		NFSM_DISSECT(sp, struct nfsv2_sattr *, NFSX_V2SATTR);
1438 		nvap->na_mode = fxdr_unsigned(u_int16_t, sp->sa_mode);
1439 	}
1440 	*pathcpp = pathcp;
1441 	*lenp = len;
1442 	NFSEXITCODE2(0, nd);
1443 	return (0);
1444 nfsmout:
1445 	if (pathcp)
1446 		free(pathcp, M_TEMP);
1447 	NFSEXITCODE2(error, nd);
1448 	return (error);
1449 }
1450 
1451 /*
1452  * Remove a non-directory object.
1453  */
1454 int
1455 nfsvno_removesub(struct nameidata *ndp, int is_v4, struct ucred *cred,
1456     struct thread *p, struct nfsexstuff *exp)
1457 {
1458 	struct vnode *vp, *dsdvp[NFSDEV_MAXMIRRORS];
1459 	int error = 0, mirrorcnt;
1460 	char fname[PNFS_FILENAME_LEN + 1];
1461 	fhandle_t fh;
1462 
1463 	vp = ndp->ni_vp;
1464 	dsdvp[0] = NULL;
1465 	if (vp->v_type == VDIR)
1466 		error = NFSERR_ISDIR;
1467 	else if (is_v4)
1468 		error = nfsrv_checkremove(vp, 1, NULL, (nfsquad_t)((u_quad_t)0),
1469 		    p);
1470 	if (error == 0)
1471 		nfsrv_pnfsremovesetup(vp, p, dsdvp, &mirrorcnt, fname, &fh);
1472 	if (!error)
1473 		error = VOP_REMOVE(ndp->ni_dvp, vp, &ndp->ni_cnd);
1474 	if (error == 0 && dsdvp[0] != NULL)
1475 		nfsrv_pnfsremove(dsdvp, mirrorcnt, fname, &fh, p);
1476 	if (ndp->ni_dvp == vp)
1477 		vrele(ndp->ni_dvp);
1478 	else
1479 		vput(ndp->ni_dvp);
1480 	vput(vp);
1481 	if ((ndp->ni_cnd.cn_flags & SAVENAME) != 0)
1482 		nfsvno_relpathbuf(ndp);
1483 	NFSEXITCODE(error);
1484 	return (error);
1485 }
1486 
1487 /*
1488  * Remove a directory.
1489  */
1490 int
1491 nfsvno_rmdirsub(struct nameidata *ndp, int is_v4, struct ucred *cred,
1492     struct thread *p, struct nfsexstuff *exp)
1493 {
1494 	struct vnode *vp;
1495 	int error = 0;
1496 
1497 	vp = ndp->ni_vp;
1498 	if (vp->v_type != VDIR) {
1499 		error = ENOTDIR;
1500 		goto out;
1501 	}
1502 	/*
1503 	 * No rmdir "." please.
1504 	 */
1505 	if (ndp->ni_dvp == vp) {
1506 		error = EINVAL;
1507 		goto out;
1508 	}
1509 	/*
1510 	 * The root of a mounted filesystem cannot be deleted.
1511 	 */
1512 	if (vp->v_vflag & VV_ROOT)
1513 		error = EBUSY;
1514 out:
1515 	if (!error)
1516 		error = VOP_RMDIR(ndp->ni_dvp, vp, &ndp->ni_cnd);
1517 	if (ndp->ni_dvp == vp)
1518 		vrele(ndp->ni_dvp);
1519 	else
1520 		vput(ndp->ni_dvp);
1521 	vput(vp);
1522 	if ((ndp->ni_cnd.cn_flags & SAVENAME) != 0)
1523 		nfsvno_relpathbuf(ndp);
1524 	NFSEXITCODE(error);
1525 	return (error);
1526 }
1527 
1528 /*
1529  * Rename vnode op.
1530  */
1531 int
1532 nfsvno_rename(struct nameidata *fromndp, struct nameidata *tondp,
1533     u_int32_t ndstat, u_int32_t ndflag, struct ucred *cred, struct thread *p)
1534 {
1535 	struct vnode *fvp, *tvp, *tdvp, *dsdvp[NFSDEV_MAXMIRRORS];
1536 	int error = 0, mirrorcnt;
1537 	char fname[PNFS_FILENAME_LEN + 1];
1538 	fhandle_t fh;
1539 
1540 	dsdvp[0] = NULL;
1541 	fvp = fromndp->ni_vp;
1542 	if (ndstat) {
1543 		vrele(fromndp->ni_dvp);
1544 		vrele(fvp);
1545 		error = ndstat;
1546 		goto out1;
1547 	}
1548 	tdvp = tondp->ni_dvp;
1549 	tvp = tondp->ni_vp;
1550 	if (tvp != NULL) {
1551 		if (fvp->v_type == VDIR && tvp->v_type != VDIR) {
1552 			error = (ndflag & ND_NFSV2) ? EISDIR : EEXIST;
1553 			goto out;
1554 		} else if (fvp->v_type != VDIR && tvp->v_type == VDIR) {
1555 			error = (ndflag & ND_NFSV2) ? ENOTDIR : EEXIST;
1556 			goto out;
1557 		}
1558 		if (tvp->v_type == VDIR && tvp->v_mountedhere) {
1559 			error = (ndflag & ND_NFSV2) ? ENOTEMPTY : EXDEV;
1560 			goto out;
1561 		}
1562 
1563 		/*
1564 		 * A rename to '.' or '..' results in a prematurely
1565 		 * unlocked vnode on FreeBSD5, so I'm just going to fail that
1566 		 * here.
1567 		 */
1568 		if ((tondp->ni_cnd.cn_namelen == 1 &&
1569 		     tondp->ni_cnd.cn_nameptr[0] == '.') ||
1570 		    (tondp->ni_cnd.cn_namelen == 2 &&
1571 		     tondp->ni_cnd.cn_nameptr[0] == '.' &&
1572 		     tondp->ni_cnd.cn_nameptr[1] == '.')) {
1573 			error = EINVAL;
1574 			goto out;
1575 		}
1576 	}
1577 	if (fvp->v_type == VDIR && fvp->v_mountedhere) {
1578 		error = (ndflag & ND_NFSV2) ? ENOTEMPTY : EXDEV;
1579 		goto out;
1580 	}
1581 	if (fvp->v_mount != tdvp->v_mount) {
1582 		error = (ndflag & ND_NFSV2) ? ENOTEMPTY : EXDEV;
1583 		goto out;
1584 	}
1585 	if (fvp == tdvp) {
1586 		error = (ndflag & ND_NFSV2) ? ENOTEMPTY : EINVAL;
1587 		goto out;
1588 	}
1589 	if (fvp == tvp) {
1590 		/*
1591 		 * If source and destination are the same, there is nothing to
1592 		 * do. Set error to -1 to indicate this.
1593 		 */
1594 		error = -1;
1595 		goto out;
1596 	}
1597 	if (ndflag & ND_NFSV4) {
1598 		if (NFSVOPLOCK(fvp, LK_EXCLUSIVE) == 0) {
1599 			error = nfsrv_checkremove(fvp, 0, NULL,
1600 			    (nfsquad_t)((u_quad_t)0), p);
1601 			NFSVOPUNLOCK(fvp);
1602 		} else
1603 			error = EPERM;
1604 		if (tvp && !error)
1605 			error = nfsrv_checkremove(tvp, 1, NULL,
1606 			    (nfsquad_t)((u_quad_t)0), p);
1607 	} else {
1608 		/*
1609 		 * For NFSv2 and NFSv3, try to get rid of the delegation, so
1610 		 * that the NFSv4 client won't be confused by the rename.
1611 		 * Since nfsd_recalldelegation() can only be called on an
1612 		 * unlocked vnode at this point and fvp is the file that will
1613 		 * still exist after the rename, just do fvp.
1614 		 */
1615 		nfsd_recalldelegation(fvp, p);
1616 	}
1617 	if (error == 0 && tvp != NULL) {
1618 		nfsrv_pnfsremovesetup(tvp, p, dsdvp, &mirrorcnt, fname, &fh);
1619 		NFSD_DEBUG(4, "nfsvno_rename: pnfsremovesetup"
1620 		    " dsdvp=%p\n", dsdvp[0]);
1621 	}
1622 out:
1623 	if (!error) {
1624 		error = VOP_RENAME(fromndp->ni_dvp, fromndp->ni_vp,
1625 		    &fromndp->ni_cnd, tondp->ni_dvp, tondp->ni_vp,
1626 		    &tondp->ni_cnd);
1627 	} else {
1628 		if (tdvp == tvp)
1629 			vrele(tdvp);
1630 		else
1631 			vput(tdvp);
1632 		if (tvp)
1633 			vput(tvp);
1634 		vrele(fromndp->ni_dvp);
1635 		vrele(fvp);
1636 		if (error == -1)
1637 			error = 0;
1638 	}
1639 
1640 	/*
1641 	 * If dsdvp[0] != NULL, it was set up by nfsrv_pnfsremovesetup() and
1642 	 * if the rename succeeded, the DS file for the tvp needs to be
1643 	 * removed.
1644 	 */
1645 	if (error == 0 && dsdvp[0] != NULL) {
1646 		nfsrv_pnfsremove(dsdvp, mirrorcnt, fname, &fh, p);
1647 		NFSD_DEBUG(4, "nfsvno_rename: pnfsremove\n");
1648 	}
1649 
1650 	vrele(tondp->ni_startdir);
1651 	nfsvno_relpathbuf(tondp);
1652 out1:
1653 	vrele(fromndp->ni_startdir);
1654 	nfsvno_relpathbuf(fromndp);
1655 	NFSEXITCODE(error);
1656 	return (error);
1657 }
1658 
1659 /*
1660  * Link vnode op.
1661  */
1662 int
1663 nfsvno_link(struct nameidata *ndp, struct vnode *vp, struct ucred *cred,
1664     struct thread *p, struct nfsexstuff *exp)
1665 {
1666 	struct vnode *xp;
1667 	int error = 0;
1668 
1669 	xp = ndp->ni_vp;
1670 	if (xp != NULL) {
1671 		error = EEXIST;
1672 	} else {
1673 		xp = ndp->ni_dvp;
1674 		if (vp->v_mount != xp->v_mount)
1675 			error = EXDEV;
1676 	}
1677 	if (!error) {
1678 		NFSVOPLOCK(vp, LK_EXCLUSIVE | LK_RETRY);
1679 		if (!VN_IS_DOOMED(vp))
1680 			error = VOP_LINK(ndp->ni_dvp, vp, &ndp->ni_cnd);
1681 		else
1682 			error = EPERM;
1683 		if (ndp->ni_dvp == vp) {
1684 			vrele(ndp->ni_dvp);
1685 			NFSVOPUNLOCK(vp);
1686 		} else {
1687 			vref(vp);
1688 			VOP_VPUT_PAIR(ndp->ni_dvp, &vp, true);
1689 		}
1690 	} else {
1691 		if (ndp->ni_dvp == ndp->ni_vp)
1692 			vrele(ndp->ni_dvp);
1693 		else
1694 			vput(ndp->ni_dvp);
1695 		if (ndp->ni_vp)
1696 			vrele(ndp->ni_vp);
1697 	}
1698 	nfsvno_relpathbuf(ndp);
1699 	NFSEXITCODE(error);
1700 	return (error);
1701 }
1702 
1703 /*
1704  * Do the fsync() appropriate for the commit.
1705  */
1706 int
1707 nfsvno_fsync(struct vnode *vp, u_int64_t off, int cnt, struct ucred *cred,
1708     struct thread *td)
1709 {
1710 	int error = 0;
1711 
1712 	/*
1713 	 * RFC 1813 3.3.21: if count is 0, a flush from offset to the end of
1714 	 * file is done.  At this time VOP_FSYNC does not accept offset and
1715 	 * byte count parameters so call VOP_FSYNC the whole file for now.
1716 	 * The same is true for NFSv4: RFC 3530 Sec. 14.2.3.
1717 	 * File systems that do not use the buffer cache (as indicated
1718 	 * by MNTK_USES_BCACHE not being set) must use VOP_FSYNC().
1719 	 */
1720 	if (cnt == 0 || cnt > MAX_COMMIT_COUNT ||
1721 	    (vp->v_mount->mnt_kern_flag & MNTK_USES_BCACHE) == 0) {
1722 		/*
1723 		 * Give up and do the whole thing
1724 		 */
1725 		if (vp->v_object && vm_object_mightbedirty(vp->v_object)) {
1726 			VM_OBJECT_WLOCK(vp->v_object);
1727 			vm_object_page_clean(vp->v_object, 0, 0, OBJPC_SYNC);
1728 			VM_OBJECT_WUNLOCK(vp->v_object);
1729 		}
1730 		error = VOP_FSYNC(vp, MNT_WAIT, td);
1731 	} else {
1732 		/*
1733 		 * Locate and synchronously write any buffers that fall
1734 		 * into the requested range.  Note:  we are assuming that
1735 		 * f_iosize is a power of 2.
1736 		 */
1737 		int iosize = vp->v_mount->mnt_stat.f_iosize;
1738 		int iomask = iosize - 1;
1739 		struct bufobj *bo;
1740 		daddr_t lblkno;
1741 
1742 		/*
1743 		 * Align to iosize boundary, super-align to page boundary.
1744 		 */
1745 		if (off & iomask) {
1746 			cnt += off & iomask;
1747 			off &= ~(u_quad_t)iomask;
1748 		}
1749 		if (off & PAGE_MASK) {
1750 			cnt += off & PAGE_MASK;
1751 			off &= ~(u_quad_t)PAGE_MASK;
1752 		}
1753 		lblkno = off / iosize;
1754 
1755 		if (vp->v_object && vm_object_mightbedirty(vp->v_object)) {
1756 			VM_OBJECT_WLOCK(vp->v_object);
1757 			vm_object_page_clean(vp->v_object, off, off + cnt,
1758 			    OBJPC_SYNC);
1759 			VM_OBJECT_WUNLOCK(vp->v_object);
1760 		}
1761 
1762 		bo = &vp->v_bufobj;
1763 		BO_LOCK(bo);
1764 		while (cnt > 0) {
1765 			struct buf *bp;
1766 
1767 			/*
1768 			 * If we have a buffer and it is marked B_DELWRI we
1769 			 * have to lock and write it.  Otherwise the prior
1770 			 * write is assumed to have already been committed.
1771 			 *
1772 			 * gbincore() can return invalid buffers now so we
1773 			 * have to check that bit as well (though B_DELWRI
1774 			 * should not be set if B_INVAL is set there could be
1775 			 * a race here since we haven't locked the buffer).
1776 			 */
1777 			if ((bp = gbincore(&vp->v_bufobj, lblkno)) != NULL) {
1778 				if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_SLEEPFAIL |
1779 				    LK_INTERLOCK, BO_LOCKPTR(bo)) == ENOLCK) {
1780 					BO_LOCK(bo);
1781 					continue; /* retry */
1782 				}
1783 			    	if ((bp->b_flags & (B_DELWRI|B_INVAL)) ==
1784 				    B_DELWRI) {
1785 					bremfree(bp);
1786 					bp->b_flags &= ~B_ASYNC;
1787 					bwrite(bp);
1788 					++nfs_commit_miss;
1789 				} else
1790 					BUF_UNLOCK(bp);
1791 				BO_LOCK(bo);
1792 			}
1793 			++nfs_commit_blks;
1794 			if (cnt < iosize)
1795 				break;
1796 			cnt -= iosize;
1797 			++lblkno;
1798 		}
1799 		BO_UNLOCK(bo);
1800 	}
1801 	NFSEXITCODE(error);
1802 	return (error);
1803 }
1804 
1805 /*
1806  * Statfs vnode op.
1807  */
1808 int
1809 nfsvno_statfs(struct vnode *vp, struct statfs *sf)
1810 {
1811 	struct statfs *tsf;
1812 	int error;
1813 
1814 	tsf = NULL;
1815 	if (nfsrv_devidcnt > 0) {
1816 		/* For a pNFS service, get the DS numbers. */
1817 		tsf = malloc(sizeof(*tsf), M_TEMP, M_WAITOK | M_ZERO);
1818 		error = nfsrv_pnfsstatfs(tsf, vp->v_mount);
1819 		if (error != 0) {
1820 			free(tsf, M_TEMP);
1821 			tsf = NULL;
1822 		}
1823 	}
1824 	error = VFS_STATFS(vp->v_mount, sf);
1825 	if (error == 0) {
1826 		if (tsf != NULL) {
1827 			sf->f_blocks = tsf->f_blocks;
1828 			sf->f_bavail = tsf->f_bavail;
1829 			sf->f_bfree = tsf->f_bfree;
1830 			sf->f_bsize = tsf->f_bsize;
1831 		}
1832 		/*
1833 		 * Since NFS handles these values as unsigned on the
1834 		 * wire, there is no way to represent negative values,
1835 		 * so set them to 0. Without this, they will appear
1836 		 * to be very large positive values for clients like
1837 		 * Solaris10.
1838 		 */
1839 		if (sf->f_bavail < 0)
1840 			sf->f_bavail = 0;
1841 		if (sf->f_ffree < 0)
1842 			sf->f_ffree = 0;
1843 	}
1844 	free(tsf, M_TEMP);
1845 	NFSEXITCODE(error);
1846 	return (error);
1847 }
1848 
1849 /*
1850  * Do the vnode op stuff for Open. Similar to nfsvno_createsub(), but
1851  * must handle nfsrv_opencheck() calls after any other access checks.
1852  */
1853 void
1854 nfsvno_open(struct nfsrv_descript *nd, struct nameidata *ndp,
1855     nfsquad_t clientid, nfsv4stateid_t *stateidp, struct nfsstate *stp,
1856     int *exclusive_flagp, struct nfsvattr *nvap, int32_t *cverf, int create,
1857     NFSACL_T *aclp, nfsattrbit_t *attrbitp, struct ucred *cred,
1858     struct nfsexstuff *exp, struct vnode **vpp)
1859 {
1860 	struct vnode *vp = NULL;
1861 	u_quad_t tempsize;
1862 	struct nfsexstuff nes;
1863 	struct thread *p = curthread;
1864 
1865 	if (ndp->ni_vp == NULL)
1866 		nd->nd_repstat = nfsrv_opencheck(clientid,
1867 		    stateidp, stp, NULL, nd, p, nd->nd_repstat);
1868 	if (!nd->nd_repstat) {
1869 		if (ndp->ni_vp == NULL) {
1870 			vrele(ndp->ni_startdir);
1871 			nd->nd_repstat = VOP_CREATE(ndp->ni_dvp,
1872 			    &ndp->ni_vp, &ndp->ni_cnd, &nvap->na_vattr);
1873 			/* For a pNFS server, create the data file on a DS. */
1874 			if (nd->nd_repstat == 0) {
1875 				/*
1876 				 * Create a data file on a DS for a pNFS server.
1877 				 * This function just returns if not
1878 				 * running a pNFS DS or the creation fails.
1879 				 */
1880 				nfsrv_pnfscreate(ndp->ni_vp, &nvap->na_vattr,
1881 				    cred, p);
1882 			}
1883 			VOP_VPUT_PAIR(ndp->ni_dvp, nd->nd_repstat == 0 ?
1884 			    &ndp->ni_vp : NULL, false);
1885 			nfsvno_relpathbuf(ndp);
1886 			if (!nd->nd_repstat) {
1887 				if (*exclusive_flagp) {
1888 					*exclusive_flagp = 0;
1889 					NFSVNO_ATTRINIT(nvap);
1890 					nvap->na_atime.tv_sec = cverf[0];
1891 					nvap->na_atime.tv_nsec = cverf[1];
1892 					nd->nd_repstat = VOP_SETATTR(ndp->ni_vp,
1893 					    &nvap->na_vattr, cred);
1894 					if (nd->nd_repstat != 0) {
1895 						vput(ndp->ni_vp);
1896 						ndp->ni_vp = NULL;
1897 						nd->nd_repstat = NFSERR_NOTSUPP;
1898 					} else
1899 						NFSSETBIT_ATTRBIT(attrbitp,
1900 						    NFSATTRBIT_TIMEACCESS);
1901 				} else {
1902 					nfsrv_fixattr(nd, ndp->ni_vp, nvap,
1903 					    aclp, p, attrbitp, exp);
1904 				}
1905 			}
1906 			vp = ndp->ni_vp;
1907 		} else {
1908 			if (ndp->ni_startdir)
1909 				vrele(ndp->ni_startdir);
1910 			nfsvno_relpathbuf(ndp);
1911 			vp = ndp->ni_vp;
1912 			if (create == NFSV4OPEN_CREATE) {
1913 				if (ndp->ni_dvp == vp)
1914 					vrele(ndp->ni_dvp);
1915 				else
1916 					vput(ndp->ni_dvp);
1917 			}
1918 			if (NFSVNO_ISSETSIZE(nvap) && vp->v_type == VREG) {
1919 				if (ndp->ni_cnd.cn_flags & RDONLY)
1920 					NFSVNO_SETEXRDONLY(&nes);
1921 				else
1922 					NFSVNO_EXINIT(&nes);
1923 				nd->nd_repstat = nfsvno_accchk(vp,
1924 				    VWRITE, cred, &nes, p,
1925 				    NFSACCCHK_NOOVERRIDE,
1926 				    NFSACCCHK_VPISLOCKED, NULL);
1927 				nd->nd_repstat = nfsrv_opencheck(clientid,
1928 				    stateidp, stp, vp, nd, p, nd->nd_repstat);
1929 				if (!nd->nd_repstat) {
1930 					tempsize = nvap->na_size;
1931 					NFSVNO_ATTRINIT(nvap);
1932 					nvap->na_size = tempsize;
1933 					nd->nd_repstat = VOP_SETATTR(vp,
1934 					    &nvap->na_vattr, cred);
1935 				}
1936 			} else if (vp->v_type == VREG) {
1937 				nd->nd_repstat = nfsrv_opencheck(clientid,
1938 				    stateidp, stp, vp, nd, p, nd->nd_repstat);
1939 			}
1940 		}
1941 	} else {
1942 		if (ndp->ni_cnd.cn_flags & HASBUF)
1943 			nfsvno_relpathbuf(ndp);
1944 		if (ndp->ni_startdir && create == NFSV4OPEN_CREATE) {
1945 			vrele(ndp->ni_startdir);
1946 			if (ndp->ni_dvp == ndp->ni_vp)
1947 				vrele(ndp->ni_dvp);
1948 			else
1949 				vput(ndp->ni_dvp);
1950 			if (ndp->ni_vp)
1951 				vput(ndp->ni_vp);
1952 		}
1953 	}
1954 	*vpp = vp;
1955 
1956 	NFSEXITCODE2(0, nd);
1957 }
1958 
1959 /*
1960  * Updates the file rev and sets the mtime and ctime
1961  * to the current clock time, returning the va_filerev and va_Xtime
1962  * values.
1963  * Return ESTALE to indicate the vnode is VIRF_DOOMED.
1964  */
1965 int
1966 nfsvno_updfilerev(struct vnode *vp, struct nfsvattr *nvap,
1967     struct nfsrv_descript *nd, struct thread *p)
1968 {
1969 	struct vattr va;
1970 
1971 	VATTR_NULL(&va);
1972 	vfs_timestamp(&va.va_mtime);
1973 	if (NFSVOPISLOCKED(vp) != LK_EXCLUSIVE) {
1974 		NFSVOPLOCK(vp, LK_UPGRADE | LK_RETRY);
1975 		if (VN_IS_DOOMED(vp))
1976 			return (ESTALE);
1977 	}
1978 	(void) VOP_SETATTR(vp, &va, nd->nd_cred);
1979 	(void) nfsvno_getattr(vp, nvap, nd, p, 1, NULL);
1980 	return (0);
1981 }
1982 
1983 /*
1984  * Glue routine to nfsv4_fillattr().
1985  */
1986 int
1987 nfsvno_fillattr(struct nfsrv_descript *nd, struct mount *mp, struct vnode *vp,
1988     struct nfsvattr *nvap, fhandle_t *fhp, int rderror, nfsattrbit_t *attrbitp,
1989     struct ucred *cred, struct thread *p, int isdgram, int reterr,
1990     int supports_nfsv4acls, int at_root, uint64_t mounted_on_fileno)
1991 {
1992 	struct statfs *sf;
1993 	int error;
1994 
1995 	sf = NULL;
1996 	if (nfsrv_devidcnt > 0 &&
1997 	    (NFSISSET_ATTRBIT(attrbitp, NFSATTRBIT_SPACEAVAIL) ||
1998 	     NFSISSET_ATTRBIT(attrbitp, NFSATTRBIT_SPACEFREE) ||
1999 	     NFSISSET_ATTRBIT(attrbitp, NFSATTRBIT_SPACETOTAL))) {
2000 		sf = malloc(sizeof(*sf), M_TEMP, M_WAITOK | M_ZERO);
2001 		error = nfsrv_pnfsstatfs(sf, mp);
2002 		if (error != 0) {
2003 			free(sf, M_TEMP);
2004 			sf = NULL;
2005 		}
2006 	}
2007 	error = nfsv4_fillattr(nd, mp, vp, NULL, &nvap->na_vattr, fhp, rderror,
2008 	    attrbitp, cred, p, isdgram, reterr, supports_nfsv4acls, at_root,
2009 	    mounted_on_fileno, sf);
2010 	free(sf, M_TEMP);
2011 	NFSEXITCODE2(0, nd);
2012 	return (error);
2013 }
2014 
2015 /* Since the Readdir vnode ops vary, put the entire functions in here. */
2016 /*
2017  * nfs readdir service
2018  * - mallocs what it thinks is enough to read
2019  *	count rounded up to a multiple of DIRBLKSIZ <= NFS_MAXREADDIR
2020  * - calls VOP_READDIR()
2021  * - loops around building the reply
2022  *	if the output generated exceeds count break out of loop
2023  *	The NFSM_CLGET macro is used here so that the reply will be packed
2024  *	tightly in mbuf clusters.
2025  * - it trims out records with d_fileno == 0
2026  *	this doesn't matter for Unix clients, but they might confuse clients
2027  *	for other os'.
2028  * - it trims out records with d_type == DT_WHT
2029  *	these cannot be seen through NFS (unless we extend the protocol)
2030  *     The alternate call nfsrvd_readdirplus() does lookups as well.
2031  * PS: The NFS protocol spec. does not clarify what the "count" byte
2032  *	argument is a count of.. just name strings and file id's or the
2033  *	entire reply rpc or ...
2034  *	I tried just file name and id sizes and it confused the Sun client,
2035  *	so I am using the full rpc size now. The "paranoia.." comment refers
2036  *	to including the status longwords that are not a part of the dir.
2037  *	"entry" structures, but are in the rpc.
2038  */
2039 int
2040 nfsrvd_readdir(struct nfsrv_descript *nd, int isdgram,
2041     struct vnode *vp, struct nfsexstuff *exp)
2042 {
2043 	struct dirent *dp;
2044 	u_int32_t *tl;
2045 	int dirlen;
2046 	char *cpos, *cend, *rbuf;
2047 	struct nfsvattr at;
2048 	int nlen, error = 0, getret = 1;
2049 	int siz, cnt, fullsiz, eofflag, ncookies;
2050 	u_int64_t off, toff, verf __unused;
2051 	u_long *cookies = NULL, *cookiep;
2052 	struct uio io;
2053 	struct iovec iv;
2054 	int is_ufs;
2055 	struct thread *p = curthread;
2056 
2057 	if (nd->nd_repstat) {
2058 		nfsrv_postopattr(nd, getret, &at);
2059 		goto out;
2060 	}
2061 	if (nd->nd_flag & ND_NFSV2) {
2062 		NFSM_DISSECT(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
2063 		off = fxdr_unsigned(u_quad_t, *tl++);
2064 	} else {
2065 		NFSM_DISSECT(tl, u_int32_t *, 5 * NFSX_UNSIGNED);
2066 		off = fxdr_hyper(tl);
2067 		tl += 2;
2068 		verf = fxdr_hyper(tl);
2069 		tl += 2;
2070 	}
2071 	toff = off;
2072 	cnt = fxdr_unsigned(int, *tl);
2073 	if (cnt > NFS_SRVMAXDATA(nd) || cnt < 0)
2074 		cnt = NFS_SRVMAXDATA(nd);
2075 	siz = ((cnt + DIRBLKSIZ - 1) & ~(DIRBLKSIZ - 1));
2076 	fullsiz = siz;
2077 	if (nd->nd_flag & ND_NFSV3) {
2078 		nd->nd_repstat = getret = nfsvno_getattr(vp, &at, nd, p, 1,
2079 		    NULL);
2080 #if 0
2081 		/*
2082 		 * va_filerev is not sufficient as a cookie verifier,
2083 		 * since it is not supposed to change when entries are
2084 		 * removed/added unless that offset cookies returned to
2085 		 * the client are no longer valid.
2086 		 */
2087 		if (!nd->nd_repstat && toff && verf != at.na_filerev)
2088 			nd->nd_repstat = NFSERR_BAD_COOKIE;
2089 #endif
2090 	}
2091 	if (!nd->nd_repstat && vp->v_type != VDIR)
2092 		nd->nd_repstat = NFSERR_NOTDIR;
2093 	if (nd->nd_repstat == 0 && cnt == 0) {
2094 		if (nd->nd_flag & ND_NFSV2)
2095 			/* NFSv2 does not have NFSERR_TOOSMALL */
2096 			nd->nd_repstat = EPERM;
2097 		else
2098 			nd->nd_repstat = NFSERR_TOOSMALL;
2099 	}
2100 	if (!nd->nd_repstat)
2101 		nd->nd_repstat = nfsvno_accchk(vp, VEXEC,
2102 		    nd->nd_cred, exp, p, NFSACCCHK_NOOVERRIDE,
2103 		    NFSACCCHK_VPISLOCKED, NULL);
2104 	if (nd->nd_repstat) {
2105 		vput(vp);
2106 		if (nd->nd_flag & ND_NFSV3)
2107 			nfsrv_postopattr(nd, getret, &at);
2108 		goto out;
2109 	}
2110 	is_ufs = strcmp(vp->v_mount->mnt_vfc->vfc_name, "ufs") == 0;
2111 	rbuf = malloc(siz, M_TEMP, M_WAITOK);
2112 again:
2113 	eofflag = 0;
2114 	if (cookies) {
2115 		free(cookies, M_TEMP);
2116 		cookies = NULL;
2117 	}
2118 
2119 	iv.iov_base = rbuf;
2120 	iv.iov_len = siz;
2121 	io.uio_iov = &iv;
2122 	io.uio_iovcnt = 1;
2123 	io.uio_offset = (off_t)off;
2124 	io.uio_resid = siz;
2125 	io.uio_segflg = UIO_SYSSPACE;
2126 	io.uio_rw = UIO_READ;
2127 	io.uio_td = NULL;
2128 	nd->nd_repstat = VOP_READDIR(vp, &io, nd->nd_cred, &eofflag, &ncookies,
2129 	    &cookies);
2130 	off = (u_int64_t)io.uio_offset;
2131 	if (io.uio_resid)
2132 		siz -= io.uio_resid;
2133 
2134 	if (!cookies && !nd->nd_repstat)
2135 		nd->nd_repstat = NFSERR_PERM;
2136 	if (nd->nd_flag & ND_NFSV3) {
2137 		getret = nfsvno_getattr(vp, &at, nd, p, 1, NULL);
2138 		if (!nd->nd_repstat)
2139 			nd->nd_repstat = getret;
2140 	}
2141 
2142 	/*
2143 	 * Handles the failed cases. nd->nd_repstat == 0 past here.
2144 	 */
2145 	if (nd->nd_repstat) {
2146 		vput(vp);
2147 		free(rbuf, M_TEMP);
2148 		if (cookies)
2149 			free(cookies, M_TEMP);
2150 		if (nd->nd_flag & ND_NFSV3)
2151 			nfsrv_postopattr(nd, getret, &at);
2152 		goto out;
2153 	}
2154 	/*
2155 	 * If nothing read, return eof
2156 	 * rpc reply
2157 	 */
2158 	if (siz == 0) {
2159 		vput(vp);
2160 		if (nd->nd_flag & ND_NFSV2) {
2161 			NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
2162 		} else {
2163 			nfsrv_postopattr(nd, getret, &at);
2164 			NFSM_BUILD(tl, u_int32_t *, 4 * NFSX_UNSIGNED);
2165 			txdr_hyper(at.na_filerev, tl);
2166 			tl += 2;
2167 		}
2168 		*tl++ = newnfs_false;
2169 		*tl = newnfs_true;
2170 		free(rbuf, M_TEMP);
2171 		free(cookies, M_TEMP);
2172 		goto out;
2173 	}
2174 
2175 	/*
2176 	 * Check for degenerate cases of nothing useful read.
2177 	 * If so go try again
2178 	 */
2179 	cpos = rbuf;
2180 	cend = rbuf + siz;
2181 	dp = (struct dirent *)cpos;
2182 	cookiep = cookies;
2183 
2184 	/*
2185 	 * For some reason FreeBSD's ufs_readdir() chooses to back the
2186 	 * directory offset up to a block boundary, so it is necessary to
2187 	 * skip over the records that precede the requested offset. This
2188 	 * requires the assumption that file offset cookies monotonically
2189 	 * increase.
2190 	 */
2191 	while (cpos < cend && ncookies > 0 &&
2192 	    (dp->d_fileno == 0 || dp->d_type == DT_WHT ||
2193 	     (is_ufs == 1 && ((u_quad_t)(*cookiep)) <= toff))) {
2194 		cpos += dp->d_reclen;
2195 		dp = (struct dirent *)cpos;
2196 		cookiep++;
2197 		ncookies--;
2198 	}
2199 	if (cpos >= cend || ncookies == 0) {
2200 		siz = fullsiz;
2201 		toff = off;
2202 		goto again;
2203 	}
2204 	vput(vp);
2205 
2206 	/*
2207 	 * If cnt > MCLBYTES and the reply will not be saved, use
2208 	 * ext_pgs mbufs for TLS.
2209 	 * For NFSv4.0, we do not know for sure if the reply will
2210 	 * be saved, so do not use ext_pgs mbufs for NFSv4.0.
2211 	 */
2212 	if (cnt > MCLBYTES && siz > MCLBYTES &&
2213 	    (nd->nd_flag & (ND_TLS | ND_EXTPG | ND_SAVEREPLY)) == ND_TLS &&
2214 	    (nd->nd_flag & (ND_NFSV4 | ND_NFSV41)) != ND_NFSV4)
2215 		nd->nd_flag |= ND_EXTPG;
2216 
2217 	/*
2218 	 * dirlen is the size of the reply, including all XDR and must
2219 	 * not exceed cnt. For NFSv2, RFC1094 didn't clearly indicate
2220 	 * if the XDR should be included in "count", but to be safe, we do.
2221 	 * (Include the two booleans at the end of the reply in dirlen now.)
2222 	 */
2223 	if (nd->nd_flag & ND_NFSV3) {
2224 		nfsrv_postopattr(nd, getret, &at);
2225 		NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
2226 		txdr_hyper(at.na_filerev, tl);
2227 		dirlen = NFSX_V3POSTOPATTR + NFSX_VERF + 2 * NFSX_UNSIGNED;
2228 	} else {
2229 		dirlen = 2 * NFSX_UNSIGNED;
2230 	}
2231 
2232 	/* Loop through the records and build reply */
2233 	while (cpos < cend && ncookies > 0) {
2234 		nlen = dp->d_namlen;
2235 		if (dp->d_fileno != 0 && dp->d_type != DT_WHT &&
2236 			nlen <= NFS_MAXNAMLEN) {
2237 			if (nd->nd_flag & ND_NFSV3)
2238 				dirlen += (6*NFSX_UNSIGNED + NFSM_RNDUP(nlen));
2239 			else
2240 				dirlen += (4*NFSX_UNSIGNED + NFSM_RNDUP(nlen));
2241 			if (dirlen > cnt) {
2242 				eofflag = 0;
2243 				break;
2244 			}
2245 
2246 			/*
2247 			 * Build the directory record xdr from
2248 			 * the dirent entry.
2249 			 */
2250 			if (nd->nd_flag & ND_NFSV3) {
2251 				NFSM_BUILD(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
2252 				*tl++ = newnfs_true;
2253 				*tl++ = 0;
2254 			} else {
2255 				NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
2256 				*tl++ = newnfs_true;
2257 			}
2258 			*tl = txdr_unsigned(dp->d_fileno);
2259 			(void) nfsm_strtom(nd, dp->d_name, nlen);
2260 			if (nd->nd_flag & ND_NFSV3) {
2261 				NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
2262 				*tl++ = 0;
2263 			} else
2264 				NFSM_BUILD(tl, u_int32_t *, NFSX_UNSIGNED);
2265 			*tl = txdr_unsigned(*cookiep);
2266 		}
2267 		cpos += dp->d_reclen;
2268 		dp = (struct dirent *)cpos;
2269 		cookiep++;
2270 		ncookies--;
2271 	}
2272 	if (cpos < cend)
2273 		eofflag = 0;
2274 	NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
2275 	*tl++ = newnfs_false;
2276 	if (eofflag)
2277 		*tl = newnfs_true;
2278 	else
2279 		*tl = newnfs_false;
2280 	free(rbuf, M_TEMP);
2281 	free(cookies, M_TEMP);
2282 
2283 out:
2284 	NFSEXITCODE2(0, nd);
2285 	return (0);
2286 nfsmout:
2287 	vput(vp);
2288 	NFSEXITCODE2(error, nd);
2289 	return (error);
2290 }
2291 
2292 /*
2293  * Readdirplus for V3 and Readdir for V4.
2294  */
2295 int
2296 nfsrvd_readdirplus(struct nfsrv_descript *nd, int isdgram,
2297     struct vnode *vp, struct nfsexstuff *exp)
2298 {
2299 	struct dirent *dp;
2300 	u_int32_t *tl;
2301 	int dirlen;
2302 	char *cpos, *cend, *rbuf;
2303 	struct vnode *nvp;
2304 	fhandle_t nfh;
2305 	struct nfsvattr nva, at, *nvap = &nva;
2306 	struct mbuf *mb0, *mb1;
2307 	struct nfsreferral *refp;
2308 	int nlen, r, error = 0, getret = 1, usevget = 1;
2309 	int siz, cnt, fullsiz, eofflag, ncookies, entrycnt;
2310 	caddr_t bpos0, bpos1;
2311 	u_int64_t off, toff, verf;
2312 	u_long *cookies = NULL, *cookiep;
2313 	nfsattrbit_t attrbits, rderrbits, savbits;
2314 	struct uio io;
2315 	struct iovec iv;
2316 	struct componentname cn;
2317 	int at_root, is_ufs, is_zfs, needs_unbusy, supports_nfsv4acls;
2318 	struct mount *mp, *new_mp;
2319 	uint64_t mounted_on_fileno;
2320 	struct thread *p = curthread;
2321 	int bextpg0, bextpg1, bextpgsiz0, bextpgsiz1;
2322 
2323 	if (nd->nd_repstat) {
2324 		nfsrv_postopattr(nd, getret, &at);
2325 		goto out;
2326 	}
2327 	NFSM_DISSECT(tl, u_int32_t *, 6 * NFSX_UNSIGNED);
2328 	off = fxdr_hyper(tl);
2329 	toff = off;
2330 	tl += 2;
2331 	verf = fxdr_hyper(tl);
2332 	tl += 2;
2333 	siz = fxdr_unsigned(int, *tl++);
2334 	cnt = fxdr_unsigned(int, *tl);
2335 
2336 	/*
2337 	 * Use the server's maximum data transfer size as the upper bound
2338 	 * on reply datalen.
2339 	 */
2340 	if (cnt > NFS_SRVMAXDATA(nd) || cnt < 0)
2341 		cnt = NFS_SRVMAXDATA(nd);
2342 
2343 	/*
2344 	 * siz is a "hint" of how much directory information (name, fileid,
2345 	 * cookie) should be in the reply. At least one client "hints" 0,
2346 	 * so I set it to cnt for that case. I also round it up to the
2347 	 * next multiple of DIRBLKSIZ.
2348 	 * Since the size of a Readdirplus directory entry reply will always
2349 	 * be greater than a directory entry returned by VOP_READDIR(), it
2350 	 * does not make sense to read more than NFS_SRVMAXDATA() via
2351 	 * VOP_READDIR().
2352 	 */
2353 	if (siz <= 0)
2354 		siz = cnt;
2355 	else if (siz > NFS_SRVMAXDATA(nd))
2356 		siz = NFS_SRVMAXDATA(nd);
2357 	siz = ((siz + DIRBLKSIZ - 1) & ~(DIRBLKSIZ - 1));
2358 
2359 	if (nd->nd_flag & ND_NFSV4) {
2360 		error = nfsrv_getattrbits(nd, &attrbits, NULL, NULL);
2361 		if (error)
2362 			goto nfsmout;
2363 		NFSSET_ATTRBIT(&savbits, &attrbits);
2364 		NFSCLRNOTFILLABLE_ATTRBIT(&attrbits, nd);
2365 		NFSZERO_ATTRBIT(&rderrbits);
2366 		NFSSETBIT_ATTRBIT(&rderrbits, NFSATTRBIT_RDATTRERROR);
2367 	} else {
2368 		NFSZERO_ATTRBIT(&attrbits);
2369 	}
2370 	fullsiz = siz;
2371 	nd->nd_repstat = getret = nfsvno_getattr(vp, &at, nd, p, 1, NULL);
2372 #if 0
2373 	if (!nd->nd_repstat) {
2374 	    if (off && verf != at.na_filerev) {
2375 		/*
2376 		 * va_filerev is not sufficient as a cookie verifier,
2377 		 * since it is not supposed to change when entries are
2378 		 * removed/added unless that offset cookies returned to
2379 		 * the client are no longer valid.
2380 		 */
2381 		if (nd->nd_flag & ND_NFSV4) {
2382 			nd->nd_repstat = NFSERR_NOTSAME;
2383 		} else {
2384 			nd->nd_repstat = NFSERR_BAD_COOKIE;
2385 		}
2386 	    }
2387 	}
2388 #endif
2389 	if (!nd->nd_repstat && vp->v_type != VDIR)
2390 		nd->nd_repstat = NFSERR_NOTDIR;
2391 	if (!nd->nd_repstat && cnt == 0)
2392 		nd->nd_repstat = NFSERR_TOOSMALL;
2393 	if (!nd->nd_repstat)
2394 		nd->nd_repstat = nfsvno_accchk(vp, VEXEC,
2395 		    nd->nd_cred, exp, p, NFSACCCHK_NOOVERRIDE,
2396 		    NFSACCCHK_VPISLOCKED, NULL);
2397 	if (nd->nd_repstat) {
2398 		vput(vp);
2399 		if (nd->nd_flag & ND_NFSV3)
2400 			nfsrv_postopattr(nd, getret, &at);
2401 		goto out;
2402 	}
2403 	is_ufs = strcmp(vp->v_mount->mnt_vfc->vfc_name, "ufs") == 0;
2404 	is_zfs = strcmp(vp->v_mount->mnt_vfc->vfc_name, "zfs") == 0;
2405 
2406 	rbuf = malloc(siz, M_TEMP, M_WAITOK);
2407 again:
2408 	eofflag = 0;
2409 	if (cookies) {
2410 		free(cookies, M_TEMP);
2411 		cookies = NULL;
2412 	}
2413 
2414 	iv.iov_base = rbuf;
2415 	iv.iov_len = siz;
2416 	io.uio_iov = &iv;
2417 	io.uio_iovcnt = 1;
2418 	io.uio_offset = (off_t)off;
2419 	io.uio_resid = siz;
2420 	io.uio_segflg = UIO_SYSSPACE;
2421 	io.uio_rw = UIO_READ;
2422 	io.uio_td = NULL;
2423 	nd->nd_repstat = VOP_READDIR(vp, &io, nd->nd_cred, &eofflag, &ncookies,
2424 	    &cookies);
2425 	off = (u_int64_t)io.uio_offset;
2426 	if (io.uio_resid)
2427 		siz -= io.uio_resid;
2428 
2429 	getret = nfsvno_getattr(vp, &at, nd, p, 1, NULL);
2430 
2431 	if (!cookies && !nd->nd_repstat)
2432 		nd->nd_repstat = NFSERR_PERM;
2433 	if (!nd->nd_repstat)
2434 		nd->nd_repstat = getret;
2435 	if (nd->nd_repstat) {
2436 		vput(vp);
2437 		if (cookies)
2438 			free(cookies, M_TEMP);
2439 		free(rbuf, M_TEMP);
2440 		if (nd->nd_flag & ND_NFSV3)
2441 			nfsrv_postopattr(nd, getret, &at);
2442 		goto out;
2443 	}
2444 	/*
2445 	 * If nothing read, return eof
2446 	 * rpc reply
2447 	 */
2448 	if (siz == 0) {
2449 		vput(vp);
2450 		if (nd->nd_flag & ND_NFSV3)
2451 			nfsrv_postopattr(nd, getret, &at);
2452 		NFSM_BUILD(tl, u_int32_t *, 4 * NFSX_UNSIGNED);
2453 		txdr_hyper(at.na_filerev, tl);
2454 		tl += 2;
2455 		*tl++ = newnfs_false;
2456 		*tl = newnfs_true;
2457 		free(cookies, M_TEMP);
2458 		free(rbuf, M_TEMP);
2459 		goto out;
2460 	}
2461 
2462 	/*
2463 	 * Check for degenerate cases of nothing useful read.
2464 	 * If so go try again
2465 	 */
2466 	cpos = rbuf;
2467 	cend = rbuf + siz;
2468 	dp = (struct dirent *)cpos;
2469 	cookiep = cookies;
2470 
2471 	/*
2472 	 * For some reason FreeBSD's ufs_readdir() chooses to back the
2473 	 * directory offset up to a block boundary, so it is necessary to
2474 	 * skip over the records that precede the requested offset. This
2475 	 * requires the assumption that file offset cookies monotonically
2476 	 * increase.
2477 	 */
2478 	while (cpos < cend && ncookies > 0 &&
2479 	  (dp->d_fileno == 0 || dp->d_type == DT_WHT ||
2480 	   (is_ufs == 1 && ((u_quad_t)(*cookiep)) <= toff) ||
2481 	   ((nd->nd_flag & ND_NFSV4) &&
2482 	    ((dp->d_namlen == 1 && dp->d_name[0] == '.') ||
2483 	     (dp->d_namlen==2 && dp->d_name[0]=='.' && dp->d_name[1]=='.'))))) {
2484 		cpos += dp->d_reclen;
2485 		dp = (struct dirent *)cpos;
2486 		cookiep++;
2487 		ncookies--;
2488 	}
2489 	if (cpos >= cend || ncookies == 0) {
2490 		siz = fullsiz;
2491 		toff = off;
2492 		goto again;
2493 	}
2494 
2495 	/*
2496 	 * Busy the file system so that the mount point won't go away
2497 	 * and, as such, VFS_VGET() can be used safely.
2498 	 */
2499 	mp = vp->v_mount;
2500 	vfs_ref(mp);
2501 	NFSVOPUNLOCK(vp);
2502 	nd->nd_repstat = vfs_busy(mp, 0);
2503 	vfs_rel(mp);
2504 	if (nd->nd_repstat != 0) {
2505 		vrele(vp);
2506 		free(cookies, M_TEMP);
2507 		free(rbuf, M_TEMP);
2508 		if (nd->nd_flag & ND_NFSV3)
2509 			nfsrv_postopattr(nd, getret, &at);
2510 		goto out;
2511 	}
2512 
2513 	/*
2514 	 * Check to see if entries in this directory can be safely acquired
2515 	 * via VFS_VGET() or if a switch to VOP_LOOKUP() is required.
2516 	 * ZFS snapshot directories need VOP_LOOKUP(), so that any
2517 	 * automount of the snapshot directory that is required will
2518 	 * be done.
2519 	 * This needs to be done here for NFSv4, since NFSv4 never does
2520 	 * a VFS_VGET() for "." or "..".
2521 	 */
2522 	if (is_zfs == 1) {
2523 		r = VFS_VGET(mp, at.na_fileid, LK_SHARED, &nvp);
2524 		if (r == EOPNOTSUPP) {
2525 			usevget = 0;
2526 			cn.cn_nameiop = LOOKUP;
2527 			cn.cn_lkflags = LK_SHARED | LK_RETRY;
2528 			cn.cn_cred = nd->nd_cred;
2529 			cn.cn_thread = p;
2530 		} else if (r == 0)
2531 			vput(nvp);
2532 	}
2533 
2534 	/*
2535 	 * If the reply is likely to exceed MCLBYTES and the reply will
2536 	 * not be saved, use ext_pgs mbufs for TLS.
2537 	 * It is difficult to predict how large each entry will be and
2538 	 * how many entries have been read, so just assume the directory
2539 	 * entries grow by a factor of 4 when attributes are included.
2540 	 * For NFSv4.0, we do not know for sure if the reply will
2541 	 * be saved, so do not use ext_pgs mbufs for NFSv4.0.
2542 	 */
2543 	if (cnt > MCLBYTES && siz > MCLBYTES / 4 &&
2544 	    (nd->nd_flag & (ND_TLS | ND_EXTPG | ND_SAVEREPLY)) == ND_TLS &&
2545 	    (nd->nd_flag & (ND_NFSV4 | ND_NFSV41)) != ND_NFSV4)
2546 		nd->nd_flag |= ND_EXTPG;
2547 
2548 	/*
2549 	 * Save this position, in case there is an error before one entry
2550 	 * is created.
2551 	 */
2552 	mb0 = nd->nd_mb;
2553 	bpos0 = nd->nd_bpos;
2554 	bextpg0 = nd->nd_bextpg;
2555 	bextpgsiz0 = nd->nd_bextpgsiz;
2556 
2557 	/*
2558 	 * Fill in the first part of the reply.
2559 	 * dirlen is the reply length in bytes and cannot exceed cnt.
2560 	 * (Include the two booleans at the end of the reply in dirlen now,
2561 	 *  so we recognize when we have exceeded cnt.)
2562 	 */
2563 	if (nd->nd_flag & ND_NFSV3) {
2564 		dirlen = NFSX_V3POSTOPATTR + NFSX_VERF + 2 * NFSX_UNSIGNED;
2565 		nfsrv_postopattr(nd, getret, &at);
2566 	} else {
2567 		dirlen = NFSX_VERF + 2 * NFSX_UNSIGNED;
2568 	}
2569 	NFSM_BUILD(tl, u_int32_t *, NFSX_VERF);
2570 	txdr_hyper(at.na_filerev, tl);
2571 
2572 	/*
2573 	 * Save this position, in case there is an empty reply needed.
2574 	 */
2575 	mb1 = nd->nd_mb;
2576 	bpos1 = nd->nd_bpos;
2577 	bextpg1 = nd->nd_bextpg;
2578 	bextpgsiz1 = nd->nd_bextpgsiz;
2579 
2580 	/* Loop through the records and build reply */
2581 	entrycnt = 0;
2582 	while (cpos < cend && ncookies > 0 && dirlen < cnt) {
2583 		nlen = dp->d_namlen;
2584 		if (dp->d_fileno != 0 && dp->d_type != DT_WHT &&
2585 		    nlen <= NFS_MAXNAMLEN &&
2586 		    ((nd->nd_flag & ND_NFSV3) || nlen > 2 ||
2587 		     (nlen==2 && (dp->d_name[0]!='.' || dp->d_name[1]!='.'))
2588 		      || (nlen == 1 && dp->d_name[0] != '.'))) {
2589 			/*
2590 			 * Save the current position in the reply, in case
2591 			 * this entry exceeds cnt.
2592 			 */
2593 			mb1 = nd->nd_mb;
2594 			bpos1 = nd->nd_bpos;
2595 			bextpg1 = nd->nd_bextpg;
2596 			bextpgsiz1 = nd->nd_bextpgsiz;
2597 
2598 			/*
2599 			 * For readdir_and_lookup get the vnode using
2600 			 * the file number.
2601 			 */
2602 			nvp = NULL;
2603 			refp = NULL;
2604 			r = 0;
2605 			at_root = 0;
2606 			needs_unbusy = 0;
2607 			new_mp = mp;
2608 			mounted_on_fileno = (uint64_t)dp->d_fileno;
2609 			if ((nd->nd_flag & ND_NFSV3) ||
2610 			    NFSNONZERO_ATTRBIT(&savbits)) {
2611 				if (nd->nd_flag & ND_NFSV4)
2612 					refp = nfsv4root_getreferral(NULL,
2613 					    vp, dp->d_fileno);
2614 				if (refp == NULL) {
2615 					if (usevget)
2616 						r = VFS_VGET(mp, dp->d_fileno,
2617 						    LK_SHARED, &nvp);
2618 					else
2619 						r = EOPNOTSUPP;
2620 					if (r == EOPNOTSUPP) {
2621 						if (usevget) {
2622 							usevget = 0;
2623 							cn.cn_nameiop = LOOKUP;
2624 							cn.cn_lkflags =
2625 							    LK_SHARED |
2626 							    LK_RETRY;
2627 							cn.cn_cred =
2628 							    nd->nd_cred;
2629 							cn.cn_thread = p;
2630 						}
2631 						cn.cn_nameptr = dp->d_name;
2632 						cn.cn_namelen = nlen;
2633 						cn.cn_flags = ISLASTCN |
2634 						    NOFOLLOW | LOCKLEAF;
2635 						if (nlen == 2 &&
2636 						    dp->d_name[0] == '.' &&
2637 						    dp->d_name[1] == '.')
2638 							cn.cn_flags |=
2639 							    ISDOTDOT;
2640 						if (NFSVOPLOCK(vp, LK_SHARED)
2641 						    != 0) {
2642 							nd->nd_repstat = EPERM;
2643 							break;
2644 						}
2645 						if ((vp->v_vflag & VV_ROOT) != 0
2646 						    && (cn.cn_flags & ISDOTDOT)
2647 						    != 0) {
2648 							vref(vp);
2649 							nvp = vp;
2650 							r = 0;
2651 						} else {
2652 							r = VOP_LOOKUP(vp, &nvp,
2653 							    &cn);
2654 							if (vp != nvp)
2655 								NFSVOPUNLOCK(vp);
2656 						}
2657 					}
2658 
2659 					/*
2660 					 * For NFSv4, check to see if nvp is
2661 					 * a mount point and get the mount
2662 					 * point vnode, as required.
2663 					 */
2664 					if (r == 0 &&
2665 					    nfsrv_enable_crossmntpt != 0 &&
2666 					    (nd->nd_flag & ND_NFSV4) != 0 &&
2667 					    nvp->v_type == VDIR &&
2668 					    nvp->v_mountedhere != NULL) {
2669 						new_mp = nvp->v_mountedhere;
2670 						r = vfs_busy(new_mp, 0);
2671 						vput(nvp);
2672 						nvp = NULL;
2673 						if (r == 0) {
2674 							r = VFS_ROOT(new_mp,
2675 							    LK_SHARED, &nvp);
2676 							needs_unbusy = 1;
2677 							if (r == 0)
2678 								at_root = 1;
2679 						}
2680 					}
2681 				}
2682 
2683 				/*
2684 				 * If we failed to look up the entry, then it
2685 				 * has become invalid, most likely removed.
2686 				 */
2687 				if (r != 0) {
2688 					if (needs_unbusy)
2689 						vfs_unbusy(new_mp);
2690 					goto invalid;
2691 				}
2692 				KASSERT(refp != NULL || nvp != NULL,
2693 				    ("%s: undetected lookup error", __func__));
2694 
2695 				if (refp == NULL &&
2696 				    ((nd->nd_flag & ND_NFSV3) ||
2697 				     NFSNONZERO_ATTRBIT(&attrbits))) {
2698 					r = nfsvno_getfh(nvp, &nfh, p);
2699 					if (!r)
2700 					    r = nfsvno_getattr(nvp, nvap, nd, p,
2701 						1, &attrbits);
2702 					if (r == 0 && is_zfs == 1 &&
2703 					    nfsrv_enable_crossmntpt != 0 &&
2704 					    (nd->nd_flag & ND_NFSV4) != 0 &&
2705 					    nvp->v_type == VDIR &&
2706 					    vp->v_mount != nvp->v_mount) {
2707 					    /*
2708 					     * For a ZFS snapshot, there is a
2709 					     * pseudo mount that does not set
2710 					     * v_mountedhere, so it needs to
2711 					     * be detected via a different
2712 					     * mount structure.
2713 					     */
2714 					    at_root = 1;
2715 					    if (new_mp == mp)
2716 						new_mp = nvp->v_mount;
2717 					}
2718 				}
2719 
2720 				/*
2721 				 * If we failed to get attributes of the entry,
2722 				 * then just skip it for NFSv3 (the traditional
2723 				 * behavior in the old NFS server).
2724 				 * For NFSv4 the behavior is controlled by
2725 				 * RDATTRERROR: we either ignore the error or
2726 				 * fail the request.
2727 				 * Note that RDATTRERROR is never set for NFSv3.
2728 				 */
2729 				if (r != 0) {
2730 					if (!NFSISSET_ATTRBIT(&attrbits,
2731 					    NFSATTRBIT_RDATTRERROR)) {
2732 						vput(nvp);
2733 						if (needs_unbusy != 0)
2734 							vfs_unbusy(new_mp);
2735 						if ((nd->nd_flag & ND_NFSV3))
2736 							goto invalid;
2737 						nd->nd_repstat = r;
2738 						break;
2739 					}
2740 				}
2741 			}
2742 
2743 			/*
2744 			 * Build the directory record xdr
2745 			 */
2746 			if (nd->nd_flag & ND_NFSV3) {
2747 				NFSM_BUILD(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
2748 				*tl++ = newnfs_true;
2749 				*tl++ = 0;
2750 				*tl = txdr_unsigned(dp->d_fileno);
2751 				dirlen += nfsm_strtom(nd, dp->d_name, nlen);
2752 				NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
2753 				*tl++ = 0;
2754 				*tl = txdr_unsigned(*cookiep);
2755 				nfsrv_postopattr(nd, 0, nvap);
2756 				dirlen += nfsm_fhtom(nd,(u_int8_t *)&nfh,0,1);
2757 				dirlen += (5*NFSX_UNSIGNED+NFSX_V3POSTOPATTR);
2758 				if (nvp != NULL)
2759 					vput(nvp);
2760 			} else {
2761 				NFSM_BUILD(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
2762 				*tl++ = newnfs_true;
2763 				*tl++ = 0;
2764 				*tl = txdr_unsigned(*cookiep);
2765 				dirlen += nfsm_strtom(nd, dp->d_name, nlen);
2766 				if (nvp != NULL) {
2767 					supports_nfsv4acls =
2768 					    nfs_supportsnfsv4acls(nvp);
2769 					NFSVOPUNLOCK(nvp);
2770 				} else
2771 					supports_nfsv4acls = 0;
2772 				if (refp != NULL) {
2773 					dirlen += nfsrv_putreferralattr(nd,
2774 					    &savbits, refp, 0,
2775 					    &nd->nd_repstat);
2776 					if (nd->nd_repstat) {
2777 						if (nvp != NULL)
2778 							vrele(nvp);
2779 						if (needs_unbusy != 0)
2780 							vfs_unbusy(new_mp);
2781 						break;
2782 					}
2783 				} else if (r) {
2784 					dirlen += nfsvno_fillattr(nd, new_mp,
2785 					    nvp, nvap, &nfh, r, &rderrbits,
2786 					    nd->nd_cred, p, isdgram, 0,
2787 					    supports_nfsv4acls, at_root,
2788 					    mounted_on_fileno);
2789 				} else {
2790 					dirlen += nfsvno_fillattr(nd, new_mp,
2791 					    nvp, nvap, &nfh, r, &attrbits,
2792 					    nd->nd_cred, p, isdgram, 0,
2793 					    supports_nfsv4acls, at_root,
2794 					    mounted_on_fileno);
2795 				}
2796 				if (nvp != NULL)
2797 					vrele(nvp);
2798 				dirlen += (3 * NFSX_UNSIGNED);
2799 			}
2800 			if (needs_unbusy != 0)
2801 				vfs_unbusy(new_mp);
2802 			if (dirlen <= cnt)
2803 				entrycnt++;
2804 		}
2805 invalid:
2806 		cpos += dp->d_reclen;
2807 		dp = (struct dirent *)cpos;
2808 		cookiep++;
2809 		ncookies--;
2810 	}
2811 	vrele(vp);
2812 	vfs_unbusy(mp);
2813 
2814 	/*
2815 	 * If dirlen > cnt, we must strip off the last entry. If that
2816 	 * results in an empty reply, report NFSERR_TOOSMALL.
2817 	 */
2818 	if (dirlen > cnt || nd->nd_repstat) {
2819 		if (!nd->nd_repstat && entrycnt == 0)
2820 			nd->nd_repstat = NFSERR_TOOSMALL;
2821 		if (nd->nd_repstat) {
2822 			nfsm_trimtrailing(nd, mb0, bpos0, bextpg0, bextpgsiz0);
2823 			if (nd->nd_flag & ND_NFSV3)
2824 				nfsrv_postopattr(nd, getret, &at);
2825 		} else
2826 			nfsm_trimtrailing(nd, mb1, bpos1, bextpg1, bextpgsiz1);
2827 		eofflag = 0;
2828 	} else if (cpos < cend)
2829 		eofflag = 0;
2830 	if (!nd->nd_repstat) {
2831 		NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
2832 		*tl++ = newnfs_false;
2833 		if (eofflag)
2834 			*tl = newnfs_true;
2835 		else
2836 			*tl = newnfs_false;
2837 	}
2838 	free(cookies, M_TEMP);
2839 	free(rbuf, M_TEMP);
2840 
2841 out:
2842 	NFSEXITCODE2(0, nd);
2843 	return (0);
2844 nfsmout:
2845 	vput(vp);
2846 	NFSEXITCODE2(error, nd);
2847 	return (error);
2848 }
2849 
2850 /*
2851  * Get the settable attributes out of the mbuf list.
2852  * (Return 0 or EBADRPC)
2853  */
2854 int
2855 nfsrv_sattr(struct nfsrv_descript *nd, vnode_t vp, struct nfsvattr *nvap,
2856     nfsattrbit_t *attrbitp, NFSACL_T *aclp, struct thread *p)
2857 {
2858 	u_int32_t *tl;
2859 	struct nfsv2_sattr *sp;
2860 	int error = 0, toclient = 0;
2861 
2862 	switch (nd->nd_flag & (ND_NFSV2 | ND_NFSV3 | ND_NFSV4)) {
2863 	case ND_NFSV2:
2864 		NFSM_DISSECT(sp, struct nfsv2_sattr *, NFSX_V2SATTR);
2865 		/*
2866 		 * Some old clients didn't fill in the high order 16bits.
2867 		 * --> check the low order 2 bytes for 0xffff
2868 		 */
2869 		if ((fxdr_unsigned(int, sp->sa_mode) & 0xffff) != 0xffff)
2870 			nvap->na_mode = nfstov_mode(sp->sa_mode);
2871 		if (sp->sa_uid != newnfs_xdrneg1)
2872 			nvap->na_uid = fxdr_unsigned(uid_t, sp->sa_uid);
2873 		if (sp->sa_gid != newnfs_xdrneg1)
2874 			nvap->na_gid = fxdr_unsigned(gid_t, sp->sa_gid);
2875 		if (sp->sa_size != newnfs_xdrneg1)
2876 			nvap->na_size = fxdr_unsigned(u_quad_t, sp->sa_size);
2877 		if (sp->sa_atime.nfsv2_sec != newnfs_xdrneg1) {
2878 #ifdef notyet
2879 			fxdr_nfsv2time(&sp->sa_atime, &nvap->na_atime);
2880 #else
2881 			nvap->na_atime.tv_sec =
2882 				fxdr_unsigned(u_int32_t,sp->sa_atime.nfsv2_sec);
2883 			nvap->na_atime.tv_nsec = 0;
2884 #endif
2885 		}
2886 		if (sp->sa_mtime.nfsv2_sec != newnfs_xdrneg1)
2887 			fxdr_nfsv2time(&sp->sa_mtime, &nvap->na_mtime);
2888 		break;
2889 	case ND_NFSV3:
2890 		NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2891 		if (*tl == newnfs_true) {
2892 			NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2893 			nvap->na_mode = nfstov_mode(*tl);
2894 		}
2895 		NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2896 		if (*tl == newnfs_true) {
2897 			NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2898 			nvap->na_uid = fxdr_unsigned(uid_t, *tl);
2899 		}
2900 		NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2901 		if (*tl == newnfs_true) {
2902 			NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2903 			nvap->na_gid = fxdr_unsigned(gid_t, *tl);
2904 		}
2905 		NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2906 		if (*tl == newnfs_true) {
2907 			NFSM_DISSECT(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
2908 			nvap->na_size = fxdr_hyper(tl);
2909 		}
2910 		NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2911 		switch (fxdr_unsigned(int, *tl)) {
2912 		case NFSV3SATTRTIME_TOCLIENT:
2913 			NFSM_DISSECT(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
2914 			fxdr_nfsv3time(tl, &nvap->na_atime);
2915 			toclient = 1;
2916 			break;
2917 		case NFSV3SATTRTIME_TOSERVER:
2918 			vfs_timestamp(&nvap->na_atime);
2919 			nvap->na_vaflags |= VA_UTIMES_NULL;
2920 			break;
2921 		}
2922 		NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2923 		switch (fxdr_unsigned(int, *tl)) {
2924 		case NFSV3SATTRTIME_TOCLIENT:
2925 			NFSM_DISSECT(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
2926 			fxdr_nfsv3time(tl, &nvap->na_mtime);
2927 			nvap->na_vaflags &= ~VA_UTIMES_NULL;
2928 			break;
2929 		case NFSV3SATTRTIME_TOSERVER:
2930 			vfs_timestamp(&nvap->na_mtime);
2931 			if (!toclient)
2932 				nvap->na_vaflags |= VA_UTIMES_NULL;
2933 			break;
2934 		}
2935 		break;
2936 	case ND_NFSV4:
2937 		error = nfsv4_sattr(nd, vp, nvap, attrbitp, aclp, p);
2938 	}
2939 nfsmout:
2940 	NFSEXITCODE2(error, nd);
2941 	return (error);
2942 }
2943 
2944 /*
2945  * Handle the setable attributes for V4.
2946  * Returns NFSERR_BADXDR if it can't be parsed, 0 otherwise.
2947  */
2948 int
2949 nfsv4_sattr(struct nfsrv_descript *nd, vnode_t vp, struct nfsvattr *nvap,
2950     nfsattrbit_t *attrbitp, NFSACL_T *aclp, struct thread *p)
2951 {
2952 	u_int32_t *tl;
2953 	int attrsum = 0;
2954 	int i, j;
2955 	int error, attrsize, bitpos, aclsize, aceerr, retnotsup = 0;
2956 	int moderet, toclient = 0;
2957 	u_char *cp, namestr[NFSV4_SMALLSTR + 1];
2958 	uid_t uid;
2959 	gid_t gid;
2960 	u_short mode, mask;		/* Same type as va_mode. */
2961 	struct vattr va;
2962 
2963 	error = nfsrv_getattrbits(nd, attrbitp, NULL, &retnotsup);
2964 	if (error)
2965 		goto nfsmout;
2966 	NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2967 	attrsize = fxdr_unsigned(int, *tl);
2968 
2969 	/*
2970 	 * Loop around getting the setable attributes. If an unsupported
2971 	 * one is found, set nd_repstat == NFSERR_ATTRNOTSUPP and return.
2972 	 */
2973 	if (retnotsup) {
2974 		nd->nd_repstat = NFSERR_ATTRNOTSUPP;
2975 		bitpos = NFSATTRBIT_MAX;
2976 	} else {
2977 		bitpos = 0;
2978 	}
2979 	moderet = 0;
2980 	for (; bitpos < NFSATTRBIT_MAX; bitpos++) {
2981 	    if (attrsum > attrsize) {
2982 		error = NFSERR_BADXDR;
2983 		goto nfsmout;
2984 	    }
2985 	    if (NFSISSET_ATTRBIT(attrbitp, bitpos))
2986 		switch (bitpos) {
2987 		case NFSATTRBIT_SIZE:
2988 			NFSM_DISSECT(tl, u_int32_t *, NFSX_HYPER);
2989                      if (vp != NULL && vp->v_type != VREG) {
2990                             error = (vp->v_type == VDIR) ? NFSERR_ISDIR :
2991                                 NFSERR_INVAL;
2992                             goto nfsmout;
2993 			}
2994 			nvap->na_size = fxdr_hyper(tl);
2995 			attrsum += NFSX_HYPER;
2996 			break;
2997 		case NFSATTRBIT_ACL:
2998 			error = nfsrv_dissectacl(nd, aclp, &aceerr, &aclsize,
2999 			    p);
3000 			if (error)
3001 				goto nfsmout;
3002 			if (aceerr && !nd->nd_repstat)
3003 				nd->nd_repstat = aceerr;
3004 			attrsum += aclsize;
3005 			break;
3006 		case NFSATTRBIT_ARCHIVE:
3007 			NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
3008 			if (!nd->nd_repstat)
3009 				nd->nd_repstat = NFSERR_ATTRNOTSUPP;
3010 			attrsum += NFSX_UNSIGNED;
3011 			break;
3012 		case NFSATTRBIT_HIDDEN:
3013 			NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
3014 			if (!nd->nd_repstat)
3015 				nd->nd_repstat = NFSERR_ATTRNOTSUPP;
3016 			attrsum += NFSX_UNSIGNED;
3017 			break;
3018 		case NFSATTRBIT_MIMETYPE:
3019 			NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
3020 			i = fxdr_unsigned(int, *tl);
3021 			error = nfsm_advance(nd, NFSM_RNDUP(i), -1);
3022 			if (error)
3023 				goto nfsmout;
3024 			if (!nd->nd_repstat)
3025 				nd->nd_repstat = NFSERR_ATTRNOTSUPP;
3026 			attrsum += (NFSX_UNSIGNED + NFSM_RNDUP(i));
3027 			break;
3028 		case NFSATTRBIT_MODE:
3029 			moderet = NFSERR_INVAL;	/* Can't do MODESETMASKED. */
3030 			NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
3031 			nvap->na_mode = nfstov_mode(*tl);
3032 			attrsum += NFSX_UNSIGNED;
3033 			break;
3034 		case NFSATTRBIT_OWNER:
3035 			NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
3036 			j = fxdr_unsigned(int, *tl);
3037 			if (j < 0) {
3038 				error = NFSERR_BADXDR;
3039 				goto nfsmout;
3040 			}
3041 			if (j > NFSV4_SMALLSTR)
3042 				cp = malloc(j + 1, M_NFSSTRING, M_WAITOK);
3043 			else
3044 				cp = namestr;
3045 			error = nfsrv_mtostr(nd, cp, j);
3046 			if (error) {
3047 				if (j > NFSV4_SMALLSTR)
3048 					free(cp, M_NFSSTRING);
3049 				goto nfsmout;
3050 			}
3051 			if (!nd->nd_repstat) {
3052 				nd->nd_repstat = nfsv4_strtouid(nd, cp, j,
3053 				    &uid);
3054 				if (!nd->nd_repstat)
3055 					nvap->na_uid = uid;
3056 			}
3057 			if (j > NFSV4_SMALLSTR)
3058 				free(cp, M_NFSSTRING);
3059 			attrsum += (NFSX_UNSIGNED + NFSM_RNDUP(j));
3060 			break;
3061 		case NFSATTRBIT_OWNERGROUP:
3062 			NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
3063 			j = fxdr_unsigned(int, *tl);
3064 			if (j < 0) {
3065 				error = NFSERR_BADXDR;
3066 				goto nfsmout;
3067 			}
3068 			if (j > NFSV4_SMALLSTR)
3069 				cp = malloc(j + 1, M_NFSSTRING, M_WAITOK);
3070 			else
3071 				cp = namestr;
3072 			error = nfsrv_mtostr(nd, cp, j);
3073 			if (error) {
3074 				if (j > NFSV4_SMALLSTR)
3075 					free(cp, M_NFSSTRING);
3076 				goto nfsmout;
3077 			}
3078 			if (!nd->nd_repstat) {
3079 				nd->nd_repstat = nfsv4_strtogid(nd, cp, j,
3080 				    &gid);
3081 				if (!nd->nd_repstat)
3082 					nvap->na_gid = gid;
3083 			}
3084 			if (j > NFSV4_SMALLSTR)
3085 				free(cp, M_NFSSTRING);
3086 			attrsum += (NFSX_UNSIGNED + NFSM_RNDUP(j));
3087 			break;
3088 		case NFSATTRBIT_SYSTEM:
3089 			NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
3090 			if (!nd->nd_repstat)
3091 				nd->nd_repstat = NFSERR_ATTRNOTSUPP;
3092 			attrsum += NFSX_UNSIGNED;
3093 			break;
3094 		case NFSATTRBIT_TIMEACCESSSET:
3095 			NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
3096 			attrsum += NFSX_UNSIGNED;
3097 			if (fxdr_unsigned(int, *tl)==NFSV4SATTRTIME_TOCLIENT) {
3098 			    NFSM_DISSECT(tl, u_int32_t *, NFSX_V4TIME);
3099 			    fxdr_nfsv4time(tl, &nvap->na_atime);
3100 			    toclient = 1;
3101 			    attrsum += NFSX_V4TIME;
3102 			} else {
3103 			    vfs_timestamp(&nvap->na_atime);
3104 			    nvap->na_vaflags |= VA_UTIMES_NULL;
3105 			}
3106 			break;
3107 		case NFSATTRBIT_TIMEBACKUP:
3108 			NFSM_DISSECT(tl, u_int32_t *, NFSX_V4TIME);
3109 			if (!nd->nd_repstat)
3110 				nd->nd_repstat = NFSERR_ATTRNOTSUPP;
3111 			attrsum += NFSX_V4TIME;
3112 			break;
3113 		case NFSATTRBIT_TIMECREATE:
3114 			NFSM_DISSECT(tl, u_int32_t *, NFSX_V4TIME);
3115 			fxdr_nfsv4time(tl, &nvap->na_btime);
3116 			attrsum += NFSX_V4TIME;
3117 			break;
3118 		case NFSATTRBIT_TIMEMODIFYSET:
3119 			NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
3120 			attrsum += NFSX_UNSIGNED;
3121 			if (fxdr_unsigned(int, *tl)==NFSV4SATTRTIME_TOCLIENT) {
3122 			    NFSM_DISSECT(tl, u_int32_t *, NFSX_V4TIME);
3123 			    fxdr_nfsv4time(tl, &nvap->na_mtime);
3124 			    nvap->na_vaflags &= ~VA_UTIMES_NULL;
3125 			    attrsum += NFSX_V4TIME;
3126 			} else {
3127 			    vfs_timestamp(&nvap->na_mtime);
3128 			    if (!toclient)
3129 				nvap->na_vaflags |= VA_UTIMES_NULL;
3130 			}
3131 			break;
3132 		case NFSATTRBIT_MODESETMASKED:
3133 			NFSM_DISSECT(tl, uint32_t *, 2 * NFSX_UNSIGNED);
3134 			mode = fxdr_unsigned(u_short, *tl++);
3135 			mask = fxdr_unsigned(u_short, *tl);
3136 			/*
3137 			 * vp == NULL implies an Open/Create operation.
3138 			 * This attribute can only be used for Setattr and
3139 			 * only for NFSv4.1 or higher.
3140 			 * If moderet != 0, a mode attribute has also been
3141 			 * specified and this attribute cannot be done in the
3142 			 * same Setattr operation.
3143 			 */
3144 			if ((nd->nd_flag & ND_NFSV41) == 0)
3145 				nd->nd_repstat = NFSERR_ATTRNOTSUPP;
3146 			else if ((mode & ~07777) != 0 || (mask & ~07777) != 0 ||
3147 			    vp == NULL)
3148 				nd->nd_repstat = NFSERR_INVAL;
3149 			else if (moderet == 0)
3150 				moderet = VOP_GETATTR(vp, &va, nd->nd_cred);
3151 			if (moderet == 0)
3152 				nvap->na_mode = (mode & mask) |
3153 				    (va.va_mode & ~mask);
3154 			else
3155 				nd->nd_repstat = moderet;
3156 			attrsum += 2 * NFSX_UNSIGNED;
3157 			break;
3158 		default:
3159 			nd->nd_repstat = NFSERR_ATTRNOTSUPP;
3160 			/*
3161 			 * set bitpos so we drop out of the loop.
3162 			 */
3163 			bitpos = NFSATTRBIT_MAX;
3164 			break;
3165 		}
3166 	}
3167 
3168 	/*
3169 	 * some clients pad the attrlist, so we need to skip over the
3170 	 * padding.
3171 	 */
3172 	if (attrsum > attrsize) {
3173 		error = NFSERR_BADXDR;
3174 	} else {
3175 		attrsize = NFSM_RNDUP(attrsize);
3176 		if (attrsum < attrsize)
3177 			error = nfsm_advance(nd, attrsize - attrsum, -1);
3178 	}
3179 nfsmout:
3180 	NFSEXITCODE2(error, nd);
3181 	return (error);
3182 }
3183 
3184 /*
3185  * Check/setup export credentials.
3186  */
3187 int
3188 nfsd_excred(struct nfsrv_descript *nd, struct nfsexstuff *exp,
3189     struct ucred *credanon, bool testsec)
3190 {
3191 	int error;
3192 
3193 	/*
3194 	 * Check/setup credentials.
3195 	 */
3196 	if (nd->nd_flag & ND_GSS)
3197 		exp->nes_exflag &= ~MNT_EXPORTANON;
3198 
3199 	/*
3200 	 * Check to see if the operation is allowed for this security flavor.
3201 	 */
3202 	error = 0;
3203 	if (testsec) {
3204 		error = nfsvno_testexp(nd, exp);
3205 		if (error != 0)
3206 			goto out;
3207 	}
3208 
3209 	/*
3210 	 * Check to see if the file system is exported V4 only.
3211 	 */
3212 	if (NFSVNO_EXV4ONLY(exp) && !(nd->nd_flag & ND_NFSV4)) {
3213 		error = NFSERR_PROGNOTV4;
3214 		goto out;
3215 	}
3216 
3217 	/*
3218 	 * Now, map the user credentials.
3219 	 * (Note that ND_AUTHNONE will only be set for an NFSv3
3220 	 *  Fsinfo RPC. If set for anything else, this code might need
3221 	 *  to change.)
3222 	 */
3223 	if (NFSVNO_EXPORTED(exp)) {
3224 		if (((nd->nd_flag & ND_GSS) == 0 && nd->nd_cred->cr_uid == 0) ||
3225 		     NFSVNO_EXPORTANON(exp) ||
3226 		     (nd->nd_flag & ND_AUTHNONE) != 0) {
3227 			nd->nd_cred->cr_uid = credanon->cr_uid;
3228 			nd->nd_cred->cr_gid = credanon->cr_gid;
3229 			crsetgroups(nd->nd_cred, credanon->cr_ngroups,
3230 			    credanon->cr_groups);
3231 		} else if ((nd->nd_flag & ND_GSS) == 0) {
3232 			/*
3233 			 * If using AUTH_SYS, call nfsrv_getgrpscred() to see
3234 			 * if there is a replacement credential with a group
3235 			 * list set up by "nfsuserd -manage-gids".
3236 			 * If there is no replacement, nfsrv_getgrpscred()
3237 			 * simply returns its argument.
3238 			 */
3239 			nd->nd_cred = nfsrv_getgrpscred(nd->nd_cred);
3240 		}
3241 	}
3242 
3243 out:
3244 	NFSEXITCODE2(error, nd);
3245 	return (error);
3246 }
3247 
3248 /*
3249  * Check exports.
3250  */
3251 int
3252 nfsvno_checkexp(struct mount *mp, struct sockaddr *nam, struct nfsexstuff *exp,
3253     struct ucred **credp)
3254 {
3255 	int error;
3256 
3257 	error = VFS_CHECKEXP(mp, nam, &exp->nes_exflag, credp,
3258 	    &exp->nes_numsecflavor, exp->nes_secflavors);
3259 	if (error) {
3260 		if (nfs_rootfhset) {
3261 			exp->nes_exflag = 0;
3262 			exp->nes_numsecflavor = 0;
3263 			error = 0;
3264 		}
3265 	} else if (exp->nes_numsecflavor < 1 || exp->nes_numsecflavor >
3266 	    MAXSECFLAVORS) {
3267 		printf("nfsvno_checkexp: numsecflavors out of range\n");
3268 		exp->nes_numsecflavor = 0;
3269 		error = EACCES;
3270 	}
3271 	NFSEXITCODE(error);
3272 	return (error);
3273 }
3274 
3275 /*
3276  * Get a vnode for a file handle and export stuff.
3277  */
3278 int
3279 nfsvno_fhtovp(struct mount *mp, fhandle_t *fhp, struct sockaddr *nam,
3280     int lktype, struct vnode **vpp, struct nfsexstuff *exp,
3281     struct ucred **credp)
3282 {
3283 	int error;
3284 
3285 	*credp = NULL;
3286 	exp->nes_numsecflavor = 0;
3287 	error = VFS_FHTOVP(mp, &fhp->fh_fid, lktype, vpp);
3288 	if (error != 0)
3289 		/* Make sure the server replies ESTALE to the client. */
3290 		error = ESTALE;
3291 	if (nam && !error) {
3292 		error = VFS_CHECKEXP(mp, nam, &exp->nes_exflag, credp,
3293 		    &exp->nes_numsecflavor, exp->nes_secflavors);
3294 		if (error) {
3295 			if (nfs_rootfhset) {
3296 				exp->nes_exflag = 0;
3297 				exp->nes_numsecflavor = 0;
3298 				error = 0;
3299 			} else {
3300 				vput(*vpp);
3301 			}
3302 		} else if (exp->nes_numsecflavor < 1 || exp->nes_numsecflavor >
3303 		    MAXSECFLAVORS) {
3304 			printf("nfsvno_fhtovp: numsecflavors out of range\n");
3305 			exp->nes_numsecflavor = 0;
3306 			error = EACCES;
3307 			vput(*vpp);
3308 		}
3309 	}
3310 	NFSEXITCODE(error);
3311 	return (error);
3312 }
3313 
3314 /*
3315  * nfsd_fhtovp() - convert a fh to a vnode ptr
3316  * 	- look up fsid in mount list (if not found ret error)
3317  *	- get vp and export rights by calling nfsvno_fhtovp()
3318  *	- if cred->cr_uid == 0 or MNT_EXPORTANON set it to credanon
3319  *	  for AUTH_SYS
3320  *	- if mpp != NULL, return the mount point so that it can
3321  *	  be used for vn_finished_write() by the caller
3322  */
3323 void
3324 nfsd_fhtovp(struct nfsrv_descript *nd, struct nfsrvfh *nfp, int lktype,
3325     struct vnode **vpp, struct nfsexstuff *exp,
3326     struct mount **mpp, int startwrite, int nextop)
3327 {
3328 	struct mount *mp, *mpw;
3329 	struct ucred *credanon;
3330 	fhandle_t *fhp;
3331 	int error;
3332 
3333 	if (mpp != NULL)
3334 		*mpp = NULL;
3335 	*vpp = NULL;
3336 	fhp = (fhandle_t *)nfp->nfsrvfh_data;
3337 	mp = vfs_busyfs(&fhp->fh_fsid);
3338 	if (mp == NULL) {
3339 		nd->nd_repstat = ESTALE;
3340 		goto out;
3341 	}
3342 
3343 	if (startwrite) {
3344 		mpw = mp;
3345 		error = vn_start_write(NULL, &mpw, V_WAIT);
3346 		if (error != 0) {
3347 			mpw = NULL;
3348 			vfs_unbusy(mp);
3349 			nd->nd_repstat = ESTALE;
3350 			goto out;
3351 		}
3352 		if (lktype == LK_SHARED && !(MNT_SHARED_WRITES(mp)))
3353 			lktype = LK_EXCLUSIVE;
3354 	} else
3355 		mpw = NULL;
3356 
3357 	nd->nd_repstat = nfsvno_fhtovp(mp, fhp, nd->nd_nam, lktype, vpp, exp,
3358 	    &credanon);
3359 	vfs_unbusy(mp);
3360 
3361 	/*
3362 	 * For NFSv4 without a pseudo root fs, unexported file handles
3363 	 * can be returned, so that Lookup works everywhere.
3364 	 */
3365 	if (!nd->nd_repstat && exp->nes_exflag == 0 &&
3366 	    !(nd->nd_flag & ND_NFSV4)) {
3367 		vput(*vpp);
3368 		*vpp = NULL;
3369 		nd->nd_repstat = EACCES;
3370 	}
3371 
3372 	/*
3373 	 * Personally, I've never seen any point in requiring a
3374 	 * reserved port#, since only in the rare case where the
3375 	 * clients are all boxes with secure system privileges,
3376 	 * does it provide any enhanced security, but... some people
3377 	 * believe it to be useful and keep putting this code back in.
3378 	 * (There is also some "security checker" out there that
3379 	 *  complains if the nfs server doesn't enforce this.)
3380 	 * However, note the following:
3381 	 * RFC3530 (NFSv4) specifies that a reserved port# not be
3382 	 *	required.
3383 	 * RFC2623 recommends that, if a reserved port# is checked for,
3384 	 *	that there be a way to turn that off--> ifdef'd.
3385 	 */
3386 #ifdef NFS_REQRSVPORT
3387 	if (!nd->nd_repstat) {
3388 		struct sockaddr_in *saddr;
3389 		struct sockaddr_in6 *saddr6;
3390 
3391 		saddr = NFSSOCKADDR(nd->nd_nam, struct sockaddr_in *);
3392 		saddr6 = NFSSOCKADDR(nd->nd_nam, struct sockaddr_in6 *);
3393 		if (!(nd->nd_flag & ND_NFSV4) &&
3394 		    ((saddr->sin_family == AF_INET &&
3395 		      ntohs(saddr->sin_port) >= IPPORT_RESERVED) ||
3396 		     (saddr6->sin6_family == AF_INET6 &&
3397 		      ntohs(saddr6->sin6_port) >= IPPORT_RESERVED))) {
3398 			vput(*vpp);
3399 			nd->nd_repstat = (NFSERR_AUTHERR | AUTH_TOOWEAK);
3400 		}
3401 	}
3402 #endif	/* NFS_REQRSVPORT */
3403 
3404 	/*
3405 	 * Check/setup credentials.
3406 	 */
3407 	if (!nd->nd_repstat) {
3408 		nd->nd_saveduid = nd->nd_cred->cr_uid;
3409 		nd->nd_repstat = nfsd_excred(nd, exp, credanon,
3410 		    nfsrv_checkwrongsec(nd, nextop, (*vpp)->v_type));
3411 		if (nd->nd_repstat)
3412 			vput(*vpp);
3413 	}
3414 	if (credanon != NULL)
3415 		crfree(credanon);
3416 	if (nd->nd_repstat) {
3417 		vn_finished_write(mpw);
3418 		*vpp = NULL;
3419 	} else if (mpp != NULL) {
3420 		*mpp = mpw;
3421 	}
3422 
3423 out:
3424 	NFSEXITCODE2(0, nd);
3425 }
3426 
3427 /*
3428  * glue for fp.
3429  */
3430 static int
3431 fp_getfvp(struct thread *p, int fd, struct file **fpp, struct vnode **vpp)
3432 {
3433 	struct filedesc *fdp;
3434 	struct file *fp;
3435 	int error = 0;
3436 
3437 	fdp = p->td_proc->p_fd;
3438 	if (fd < 0 || fd >= fdp->fd_nfiles ||
3439 	    (fp = fdp->fd_ofiles[fd].fde_file) == NULL) {
3440 		error = EBADF;
3441 		goto out;
3442 	}
3443 	*fpp = fp;
3444 
3445 out:
3446 	NFSEXITCODE(error);
3447 	return (error);
3448 }
3449 
3450 /*
3451  * Called from nfssvc() to update the exports list. Just call
3452  * vfs_export(). This has to be done, since the v4 root fake fs isn't
3453  * in the mount list.
3454  */
3455 int
3456 nfsrv_v4rootexport(void *argp, struct ucred *cred, struct thread *p)
3457 {
3458 	struct nfsex_args *nfsexargp = (struct nfsex_args *)argp;
3459 	int error = 0;
3460 	struct nameidata nd;
3461 	fhandle_t fh;
3462 
3463 	error = vfs_export(&nfsv4root_mnt, &nfsexargp->export);
3464 	if ((nfsexargp->export.ex_flags & MNT_DELEXPORT) != 0)
3465 		nfs_rootfhset = 0;
3466 	else if (error == 0) {
3467 		if (nfsexargp->fspec == NULL) {
3468 			error = EPERM;
3469 			goto out;
3470 		}
3471 		/*
3472 		 * If fspec != NULL, this is the v4root path.
3473 		 */
3474 		NDINIT(&nd, LOOKUP, FOLLOW, UIO_USERSPACE,
3475 		    nfsexargp->fspec, p);
3476 		if ((error = namei(&nd)) != 0)
3477 			goto out;
3478 		error = nfsvno_getfh(nd.ni_vp, &fh, p);
3479 		vrele(nd.ni_vp);
3480 		if (!error) {
3481 			nfs_rootfh.nfsrvfh_len = NFSX_MYFH;
3482 			NFSBCOPY((caddr_t)&fh,
3483 			    nfs_rootfh.nfsrvfh_data,
3484 			    sizeof (fhandle_t));
3485 			nfs_rootfhset = 1;
3486 		}
3487 	}
3488 
3489 out:
3490 	NFSEXITCODE(error);
3491 	return (error);
3492 }
3493 
3494 /*
3495  * This function needs to test to see if the system is near its limit
3496  * for memory allocation via malloc() or mget() and return True iff
3497  * either of these resources are near their limit.
3498  * XXX (For now, this is just a stub.)
3499  */
3500 int nfsrv_testmalloclimit = 0;
3501 int
3502 nfsrv_mallocmget_limit(void)
3503 {
3504 	static int printmesg = 0;
3505 	static int testval = 1;
3506 
3507 	if (nfsrv_testmalloclimit && (testval++ % 1000) == 0) {
3508 		if ((printmesg++ % 100) == 0)
3509 			printf("nfsd: malloc/mget near limit\n");
3510 		return (1);
3511 	}
3512 	return (0);
3513 }
3514 
3515 /*
3516  * BSD specific initialization of a mount point.
3517  */
3518 void
3519 nfsd_mntinit(void)
3520 {
3521 	static int inited = 0;
3522 
3523 	if (inited)
3524 		return;
3525 	inited = 1;
3526 	nfsv4root_mnt.mnt_flag = (MNT_RDONLY | MNT_EXPORTED);
3527 	TAILQ_INIT(&nfsv4root_mnt.mnt_nvnodelist);
3528 	TAILQ_INIT(&nfsv4root_mnt.mnt_lazyvnodelist);
3529 	nfsv4root_mnt.mnt_export = NULL;
3530 	TAILQ_INIT(&nfsv4root_opt);
3531 	TAILQ_INIT(&nfsv4root_newopt);
3532 	nfsv4root_mnt.mnt_opt = &nfsv4root_opt;
3533 	nfsv4root_mnt.mnt_optnew = &nfsv4root_newopt;
3534 	nfsv4root_mnt.mnt_nvnodelistsize = 0;
3535 	nfsv4root_mnt.mnt_lazyvnodelistsize = 0;
3536 }
3537 
3538 /*
3539  * Get a vnode for a file handle, without checking exports, etc.
3540  */
3541 struct vnode *
3542 nfsvno_getvp(fhandle_t *fhp)
3543 {
3544 	struct mount *mp;
3545 	struct vnode *vp;
3546 	int error;
3547 
3548 	mp = vfs_busyfs(&fhp->fh_fsid);
3549 	if (mp == NULL)
3550 		return (NULL);
3551 	error = VFS_FHTOVP(mp, &fhp->fh_fid, LK_EXCLUSIVE, &vp);
3552 	vfs_unbusy(mp);
3553 	if (error)
3554 		return (NULL);
3555 	return (vp);
3556 }
3557 
3558 /*
3559  * Do a local VOP_ADVLOCK().
3560  */
3561 int
3562 nfsvno_advlock(struct vnode *vp, int ftype, u_int64_t first,
3563     u_int64_t end, struct thread *td)
3564 {
3565 	int error = 0;
3566 	struct flock fl;
3567 	u_int64_t tlen;
3568 
3569 	if (nfsrv_dolocallocks == 0)
3570 		goto out;
3571 	ASSERT_VOP_UNLOCKED(vp, "nfsvno_advlock: vp locked");
3572 
3573 	fl.l_whence = SEEK_SET;
3574 	fl.l_type = ftype;
3575 	fl.l_start = (off_t)first;
3576 	if (end == NFS64BITSSET) {
3577 		fl.l_len = 0;
3578 	} else {
3579 		tlen = end - first;
3580 		fl.l_len = (off_t)tlen;
3581 	}
3582 	/*
3583 	 * For FreeBSD8, the l_pid and l_sysid must be set to the same
3584 	 * values for all calls, so that all locks will be held by the
3585 	 * nfsd server. (The nfsd server handles conflicts between the
3586 	 * various clients.)
3587 	 * Since an NFSv4 lockowner is a ClientID plus an array of up to 1024
3588 	 * bytes, so it can't be put in l_sysid.
3589 	 */
3590 	if (nfsv4_sysid == 0)
3591 		nfsv4_sysid = nlm_acquire_next_sysid();
3592 	fl.l_pid = (pid_t)0;
3593 	fl.l_sysid = (int)nfsv4_sysid;
3594 
3595 	if (ftype == F_UNLCK)
3596 		error = VOP_ADVLOCK(vp, (caddr_t)td->td_proc, F_UNLCK, &fl,
3597 		    (F_POSIX | F_REMOTE));
3598 	else
3599 		error = VOP_ADVLOCK(vp, (caddr_t)td->td_proc, F_SETLK, &fl,
3600 		    (F_POSIX | F_REMOTE));
3601 
3602 out:
3603 	NFSEXITCODE(error);
3604 	return (error);
3605 }
3606 
3607 /*
3608  * Check the nfsv4 root exports.
3609  */
3610 int
3611 nfsvno_v4rootexport(struct nfsrv_descript *nd)
3612 {
3613 	struct ucred *credanon;
3614 	int error = 0, numsecflavor, secflavors[MAXSECFLAVORS], i;
3615 	uint64_t exflags;
3616 
3617 	error = vfs_stdcheckexp(&nfsv4root_mnt, nd->nd_nam, &exflags,
3618 	    &credanon, &numsecflavor, secflavors);
3619 	if (error) {
3620 		error = NFSERR_PROGUNAVAIL;
3621 		goto out;
3622 	}
3623 	if (credanon != NULL)
3624 		crfree(credanon);
3625 	for (i = 0; i < numsecflavor; i++) {
3626 		if (secflavors[i] == AUTH_SYS)
3627 			nd->nd_flag |= ND_EXAUTHSYS;
3628 		else if (secflavors[i] == RPCSEC_GSS_KRB5)
3629 			nd->nd_flag |= ND_EXGSS;
3630 		else if (secflavors[i] == RPCSEC_GSS_KRB5I)
3631 			nd->nd_flag |= ND_EXGSSINTEGRITY;
3632 		else if (secflavors[i] == RPCSEC_GSS_KRB5P)
3633 			nd->nd_flag |= ND_EXGSSPRIVACY;
3634 	}
3635 
3636 	/* And set ND_EXxx flags for TLS. */
3637 	if ((exflags & MNT_EXTLS) != 0) {
3638 		nd->nd_flag |= ND_EXTLS;
3639 		if ((exflags & MNT_EXTLSCERT) != 0)
3640 			nd->nd_flag |= ND_EXTLSCERT;
3641 		if ((exflags & MNT_EXTLSCERTUSER) != 0)
3642 			nd->nd_flag |= ND_EXTLSCERTUSER;
3643 	}
3644 
3645 out:
3646 	NFSEXITCODE(error);
3647 	return (error);
3648 }
3649 
3650 /*
3651  * Nfs server pseudo system call for the nfsd's
3652  */
3653 /*
3654  * MPSAFE
3655  */
3656 static int
3657 nfssvc_nfsd(struct thread *td, struct nfssvc_args *uap)
3658 {
3659 	struct file *fp;
3660 	struct nfsd_addsock_args sockarg;
3661 	struct nfsd_nfsd_args nfsdarg;
3662 	struct nfsd_nfsd_oargs onfsdarg;
3663 	struct nfsd_pnfsd_args pnfsdarg;
3664 	struct vnode *vp, *nvp, *curdvp;
3665 	struct pnfsdsfile *pf;
3666 	struct nfsdevice *ds, *fds;
3667 	cap_rights_t rights;
3668 	int buflen, error, ret;
3669 	char *buf, *cp, *cp2, *cp3;
3670 	char fname[PNFS_FILENAME_LEN + 1];
3671 
3672 	if (uap->flag & NFSSVC_NFSDADDSOCK) {
3673 		error = copyin(uap->argp, (caddr_t)&sockarg, sizeof (sockarg));
3674 		if (error)
3675 			goto out;
3676 		/*
3677 		 * Since we don't know what rights might be required,
3678 		 * pretend that we need them all. It is better to be too
3679 		 * careful than too reckless.
3680 		 */
3681 		error = fget(td, sockarg.sock,
3682 		    cap_rights_init_one(&rights, CAP_SOCK_SERVER), &fp);
3683 		if (error != 0)
3684 			goto out;
3685 		if (fp->f_type != DTYPE_SOCKET) {
3686 			fdrop(fp, td);
3687 			error = EPERM;
3688 			goto out;
3689 		}
3690 		error = nfsrvd_addsock(fp);
3691 		fdrop(fp, td);
3692 	} else if (uap->flag & NFSSVC_NFSDNFSD) {
3693 		if (uap->argp == NULL) {
3694 			error = EINVAL;
3695 			goto out;
3696 		}
3697 		if ((uap->flag & NFSSVC_NEWSTRUCT) == 0) {
3698 			error = copyin(uap->argp, &onfsdarg, sizeof(onfsdarg));
3699 			if (error == 0) {
3700 				nfsdarg.principal = onfsdarg.principal;
3701 				nfsdarg.minthreads = onfsdarg.minthreads;
3702 				nfsdarg.maxthreads = onfsdarg.maxthreads;
3703 				nfsdarg.version = 1;
3704 				nfsdarg.addr = NULL;
3705 				nfsdarg.addrlen = 0;
3706 				nfsdarg.dnshost = NULL;
3707 				nfsdarg.dnshostlen = 0;
3708 				nfsdarg.dspath = NULL;
3709 				nfsdarg.dspathlen = 0;
3710 				nfsdarg.mdspath = NULL;
3711 				nfsdarg.mdspathlen = 0;
3712 				nfsdarg.mirrorcnt = 1;
3713 			}
3714 		} else
3715 			error = copyin(uap->argp, &nfsdarg, sizeof(nfsdarg));
3716 		if (error)
3717 			goto out;
3718 		if (nfsdarg.addrlen > 0 && nfsdarg.addrlen < 10000 &&
3719 		    nfsdarg.dnshostlen > 0 && nfsdarg.dnshostlen < 10000 &&
3720 		    nfsdarg.dspathlen > 0 && nfsdarg.dspathlen < 10000 &&
3721 		    nfsdarg.mdspathlen > 0 && nfsdarg.mdspathlen < 10000 &&
3722 		    nfsdarg.mirrorcnt >= 1 &&
3723 		    nfsdarg.mirrorcnt <= NFSDEV_MAXMIRRORS &&
3724 		    nfsdarg.addr != NULL && nfsdarg.dnshost != NULL &&
3725 		    nfsdarg.dspath != NULL && nfsdarg.mdspath != NULL) {
3726 			NFSD_DEBUG(1, "addrlen=%d dspathlen=%d dnslen=%d"
3727 			    " mdspathlen=%d mirrorcnt=%d\n", nfsdarg.addrlen,
3728 			    nfsdarg.dspathlen, nfsdarg.dnshostlen,
3729 			    nfsdarg.mdspathlen, nfsdarg.mirrorcnt);
3730 			cp = malloc(nfsdarg.addrlen + 1, M_TEMP, M_WAITOK);
3731 			error = copyin(nfsdarg.addr, cp, nfsdarg.addrlen);
3732 			if (error != 0) {
3733 				free(cp, M_TEMP);
3734 				goto out;
3735 			}
3736 			cp[nfsdarg.addrlen] = '\0';	/* Ensure nul term. */
3737 			nfsdarg.addr = cp;
3738 			cp = malloc(nfsdarg.dnshostlen + 1, M_TEMP, M_WAITOK);
3739 			error = copyin(nfsdarg.dnshost, cp, nfsdarg.dnshostlen);
3740 			if (error != 0) {
3741 				free(nfsdarg.addr, M_TEMP);
3742 				free(cp, M_TEMP);
3743 				goto out;
3744 			}
3745 			cp[nfsdarg.dnshostlen] = '\0';	/* Ensure nul term. */
3746 			nfsdarg.dnshost = cp;
3747 			cp = malloc(nfsdarg.dspathlen + 1, M_TEMP, M_WAITOK);
3748 			error = copyin(nfsdarg.dspath, cp, nfsdarg.dspathlen);
3749 			if (error != 0) {
3750 				free(nfsdarg.addr, M_TEMP);
3751 				free(nfsdarg.dnshost, M_TEMP);
3752 				free(cp, M_TEMP);
3753 				goto out;
3754 			}
3755 			cp[nfsdarg.dspathlen] = '\0';	/* Ensure nul term. */
3756 			nfsdarg.dspath = cp;
3757 			cp = malloc(nfsdarg.mdspathlen + 1, M_TEMP, M_WAITOK);
3758 			error = copyin(nfsdarg.mdspath, cp, nfsdarg.mdspathlen);
3759 			if (error != 0) {
3760 				free(nfsdarg.addr, M_TEMP);
3761 				free(nfsdarg.dnshost, M_TEMP);
3762 				free(nfsdarg.dspath, M_TEMP);
3763 				free(cp, M_TEMP);
3764 				goto out;
3765 			}
3766 			cp[nfsdarg.mdspathlen] = '\0';	/* Ensure nul term. */
3767 			nfsdarg.mdspath = cp;
3768 		} else {
3769 			nfsdarg.addr = NULL;
3770 			nfsdarg.addrlen = 0;
3771 			nfsdarg.dnshost = NULL;
3772 			nfsdarg.dnshostlen = 0;
3773 			nfsdarg.dspath = NULL;
3774 			nfsdarg.dspathlen = 0;
3775 			nfsdarg.mdspath = NULL;
3776 			nfsdarg.mdspathlen = 0;
3777 			nfsdarg.mirrorcnt = 1;
3778 		}
3779 		error = nfsrvd_nfsd(td, &nfsdarg);
3780 		free(nfsdarg.addr, M_TEMP);
3781 		free(nfsdarg.dnshost, M_TEMP);
3782 		free(nfsdarg.dspath, M_TEMP);
3783 		free(nfsdarg.mdspath, M_TEMP);
3784 	} else if (uap->flag & NFSSVC_PNFSDS) {
3785 		error = copyin(uap->argp, &pnfsdarg, sizeof(pnfsdarg));
3786 		if (error == 0 && (pnfsdarg.op == PNFSDOP_DELDSSERVER ||
3787 		    pnfsdarg.op == PNFSDOP_FORCEDELDS)) {
3788 			cp = malloc(PATH_MAX + 1, M_TEMP, M_WAITOK);
3789 			error = copyinstr(pnfsdarg.dspath, cp, PATH_MAX + 1,
3790 			    NULL);
3791 			if (error == 0)
3792 				error = nfsrv_deldsserver(pnfsdarg.op, cp, td);
3793 			free(cp, M_TEMP);
3794 		} else if (error == 0 && pnfsdarg.op == PNFSDOP_COPYMR) {
3795 			cp = malloc(PATH_MAX + 1, M_TEMP, M_WAITOK);
3796 			buflen = sizeof(*pf) * NFSDEV_MAXMIRRORS;
3797 			buf = malloc(buflen, M_TEMP, M_WAITOK);
3798 			error = copyinstr(pnfsdarg.mdspath, cp, PATH_MAX + 1,
3799 			    NULL);
3800 			NFSD_DEBUG(4, "pnfsdcopymr cp mdspath=%d\n", error);
3801 			if (error == 0 && pnfsdarg.dspath != NULL) {
3802 				cp2 = malloc(PATH_MAX + 1, M_TEMP, M_WAITOK);
3803 				error = copyinstr(pnfsdarg.dspath, cp2,
3804 				    PATH_MAX + 1, NULL);
3805 				NFSD_DEBUG(4, "pnfsdcopymr cp dspath=%d\n",
3806 				    error);
3807 			} else
3808 				cp2 = NULL;
3809 			if (error == 0 && pnfsdarg.curdspath != NULL) {
3810 				cp3 = malloc(PATH_MAX + 1, M_TEMP, M_WAITOK);
3811 				error = copyinstr(pnfsdarg.curdspath, cp3,
3812 				    PATH_MAX + 1, NULL);
3813 				NFSD_DEBUG(4, "pnfsdcopymr cp curdspath=%d\n",
3814 				    error);
3815 			} else
3816 				cp3 = NULL;
3817 			curdvp = NULL;
3818 			fds = NULL;
3819 			if (error == 0)
3820 				error = nfsrv_mdscopymr(cp, cp2, cp3, buf,
3821 				    &buflen, fname, td, &vp, &nvp, &pf, &ds,
3822 				    &fds);
3823 			NFSD_DEBUG(4, "nfsrv_mdscopymr=%d\n", error);
3824 			if (error == 0) {
3825 				if (pf->dsf_dir >= nfsrv_dsdirsize) {
3826 					printf("copymr: dsdir out of range\n");
3827 					pf->dsf_dir = 0;
3828 				}
3829 				NFSD_DEBUG(4, "copymr: buflen=%d\n", buflen);
3830 				error = nfsrv_copymr(vp, nvp,
3831 				    ds->nfsdev_dsdir[pf->dsf_dir], ds, pf,
3832 				    (struct pnfsdsfile *)buf,
3833 				    buflen / sizeof(*pf), td->td_ucred, td);
3834 				vput(vp);
3835 				vput(nvp);
3836 				if (fds != NULL && error == 0) {
3837 					curdvp = fds->nfsdev_dsdir[pf->dsf_dir];
3838 					ret = vn_lock(curdvp, LK_EXCLUSIVE);
3839 					if (ret == 0) {
3840 						nfsrv_dsremove(curdvp, fname,
3841 						    td->td_ucred, td);
3842 						NFSVOPUNLOCK(curdvp);
3843 					}
3844 				}
3845 				NFSD_DEBUG(4, "nfsrv_copymr=%d\n", error);
3846 			}
3847 			free(cp, M_TEMP);
3848 			free(cp2, M_TEMP);
3849 			free(cp3, M_TEMP);
3850 			free(buf, M_TEMP);
3851 		}
3852 	} else {
3853 		error = nfssvc_srvcall(td, uap, td->td_ucred);
3854 	}
3855 
3856 out:
3857 	NFSEXITCODE(error);
3858 	return (error);
3859 }
3860 
3861 static int
3862 nfssvc_srvcall(struct thread *p, struct nfssvc_args *uap, struct ucred *cred)
3863 {
3864 	struct nfsex_args export;
3865 	struct nfsex_oldargs oexp;
3866 	struct file *fp = NULL;
3867 	int stablefd, i, len;
3868 	struct nfsd_clid adminrevoke;
3869 	struct nfsd_dumplist dumplist;
3870 	struct nfsd_dumpclients *dumpclients;
3871 	struct nfsd_dumplocklist dumplocklist;
3872 	struct nfsd_dumplocks *dumplocks;
3873 	struct nameidata nd;
3874 	vnode_t vp;
3875 	int error = EINVAL, igotlock;
3876 	struct proc *procp;
3877 	gid_t *grps;
3878 	static int suspend_nfsd = 0;
3879 
3880 	if (uap->flag & NFSSVC_PUBLICFH) {
3881 		NFSBZERO((caddr_t)&nfs_pubfh.nfsrvfh_data,
3882 		    sizeof (fhandle_t));
3883 		error = copyin(uap->argp,
3884 		    &nfs_pubfh.nfsrvfh_data, sizeof (fhandle_t));
3885 		if (!error)
3886 			nfs_pubfhset = 1;
3887 	} else if ((uap->flag & (NFSSVC_V4ROOTEXPORT | NFSSVC_NEWSTRUCT)) ==
3888 	    (NFSSVC_V4ROOTEXPORT | NFSSVC_NEWSTRUCT)) {
3889 		error = copyin(uap->argp,(caddr_t)&export,
3890 		    sizeof (struct nfsex_args));
3891 		if (!error) {
3892 			grps = NULL;
3893 			if (export.export.ex_ngroups > NGROUPS_MAX ||
3894 			    export.export.ex_ngroups < 0)
3895 				error = EINVAL;
3896 			else if (export.export.ex_ngroups > 0) {
3897 				grps = malloc(export.export.ex_ngroups *
3898 				    sizeof(gid_t), M_TEMP, M_WAITOK);
3899 				error = copyin(export.export.ex_groups, grps,
3900 				    export.export.ex_ngroups * sizeof(gid_t));
3901 				export.export.ex_groups = grps;
3902 			} else
3903 				export.export.ex_groups = NULL;
3904 			if (!error)
3905 				error = nfsrv_v4rootexport(&export, cred, p);
3906 			free(grps, M_TEMP);
3907 		}
3908 	} else if ((uap->flag & (NFSSVC_V4ROOTEXPORT | NFSSVC_NEWSTRUCT)) ==
3909 	    NFSSVC_V4ROOTEXPORT) {
3910 		error = copyin(uap->argp,(caddr_t)&oexp,
3911 		    sizeof (struct nfsex_oldargs));
3912 		if (!error) {
3913 			memset(&export.export, 0, sizeof(export.export));
3914 			export.export.ex_flags = (uint64_t)oexp.export.ex_flags;
3915 			export.export.ex_root = oexp.export.ex_root;
3916 			export.export.ex_uid = oexp.export.ex_anon.cr_uid;
3917 			export.export.ex_ngroups =
3918 			    oexp.export.ex_anon.cr_ngroups;
3919 			export.export.ex_groups = NULL;
3920 			if (export.export.ex_ngroups > XU_NGROUPS ||
3921 			    export.export.ex_ngroups < 0)
3922 				error = EINVAL;
3923 			else if (export.export.ex_ngroups > 0) {
3924 				export.export.ex_groups = malloc(
3925 				    export.export.ex_ngroups * sizeof(gid_t),
3926 				    M_TEMP, M_WAITOK);
3927 				for (i = 0; i < export.export.ex_ngroups; i++)
3928 					export.export.ex_groups[i] =
3929 					    oexp.export.ex_anon.cr_groups[i];
3930 			}
3931 			export.export.ex_addr = oexp.export.ex_addr;
3932 			export.export.ex_addrlen = oexp.export.ex_addrlen;
3933 			export.export.ex_mask = oexp.export.ex_mask;
3934 			export.export.ex_masklen = oexp.export.ex_masklen;
3935 			export.export.ex_indexfile = oexp.export.ex_indexfile;
3936 			export.export.ex_numsecflavors =
3937 			    oexp.export.ex_numsecflavors;
3938 			if (export.export.ex_numsecflavors >= MAXSECFLAVORS ||
3939 			    export.export.ex_numsecflavors < 0)
3940 				error = EINVAL;
3941 			else {
3942 				for (i = 0; i < export.export.ex_numsecflavors;
3943 				    i++)
3944 					export.export.ex_secflavors[i] =
3945 					    oexp.export.ex_secflavors[i];
3946 			}
3947 			export.fspec = oexp.fspec;
3948 			if (error == 0)
3949 				error = nfsrv_v4rootexport(&export, cred, p);
3950 			free(export.export.ex_groups, M_TEMP);
3951 		}
3952 	} else if (uap->flag & NFSSVC_NOPUBLICFH) {
3953 		nfs_pubfhset = 0;
3954 		error = 0;
3955 	} else if (uap->flag & NFSSVC_STABLERESTART) {
3956 		error = copyin(uap->argp, (caddr_t)&stablefd,
3957 		    sizeof (int));
3958 		if (!error)
3959 			error = fp_getfvp(p, stablefd, &fp, &vp);
3960 		if (!error && (NFSFPFLAG(fp) & (FREAD | FWRITE)) != (FREAD | FWRITE))
3961 			error = EBADF;
3962 		if (!error && newnfs_numnfsd != 0)
3963 			error = EPERM;
3964 		if (!error) {
3965 			nfsrv_stablefirst.nsf_fp = fp;
3966 			nfsrv_setupstable(p);
3967 		}
3968 	} else if (uap->flag & NFSSVC_ADMINREVOKE) {
3969 		error = copyin(uap->argp, (caddr_t)&adminrevoke,
3970 		    sizeof (struct nfsd_clid));
3971 		if (!error)
3972 			error = nfsrv_adminrevoke(&adminrevoke, p);
3973 	} else if (uap->flag & NFSSVC_DUMPCLIENTS) {
3974 		error = copyin(uap->argp, (caddr_t)&dumplist,
3975 		    sizeof (struct nfsd_dumplist));
3976 		if (!error && (dumplist.ndl_size < 1 ||
3977 			dumplist.ndl_size > NFSRV_MAXDUMPLIST))
3978 			error = EPERM;
3979 		if (!error) {
3980 		    len = sizeof (struct nfsd_dumpclients) * dumplist.ndl_size;
3981 		    dumpclients = malloc(len, M_TEMP, M_WAITOK | M_ZERO);
3982 		    nfsrv_dumpclients(dumpclients, dumplist.ndl_size);
3983 		    error = copyout(dumpclients, dumplist.ndl_list, len);
3984 		    free(dumpclients, M_TEMP);
3985 		}
3986 	} else if (uap->flag & NFSSVC_DUMPLOCKS) {
3987 		error = copyin(uap->argp, (caddr_t)&dumplocklist,
3988 		    sizeof (struct nfsd_dumplocklist));
3989 		if (!error && (dumplocklist.ndllck_size < 1 ||
3990 			dumplocklist.ndllck_size > NFSRV_MAXDUMPLIST))
3991 			error = EPERM;
3992 		if (!error)
3993 			error = nfsrv_lookupfilename(&nd,
3994 				dumplocklist.ndllck_fname, p);
3995 		if (!error) {
3996 			len = sizeof (struct nfsd_dumplocks) *
3997 				dumplocklist.ndllck_size;
3998 			dumplocks = malloc(len, M_TEMP, M_WAITOK | M_ZERO);
3999 			nfsrv_dumplocks(nd.ni_vp, dumplocks,
4000 			    dumplocklist.ndllck_size, p);
4001 			vput(nd.ni_vp);
4002 			error = copyout(dumplocks, dumplocklist.ndllck_list,
4003 			    len);
4004 			free(dumplocks, M_TEMP);
4005 		}
4006 	} else if (uap->flag & NFSSVC_BACKUPSTABLE) {
4007 		procp = p->td_proc;
4008 		PROC_LOCK(procp);
4009 		nfsd_master_pid = procp->p_pid;
4010 		bcopy(procp->p_comm, nfsd_master_comm, MAXCOMLEN + 1);
4011 		nfsd_master_start = procp->p_stats->p_start;
4012 		nfsd_master_proc = procp;
4013 		PROC_UNLOCK(procp);
4014 	} else if ((uap->flag & NFSSVC_SUSPENDNFSD) != 0) {
4015 		NFSLOCKV4ROOTMUTEX();
4016 		if (suspend_nfsd == 0) {
4017 			/* Lock out all nfsd threads */
4018 			do {
4019 				igotlock = nfsv4_lock(&nfsd_suspend_lock, 1,
4020 				    NULL, NFSV4ROOTLOCKMUTEXPTR, NULL);
4021 			} while (igotlock == 0 && suspend_nfsd == 0);
4022 			suspend_nfsd = 1;
4023 		}
4024 		NFSUNLOCKV4ROOTMUTEX();
4025 		error = 0;
4026 	} else if ((uap->flag & NFSSVC_RESUMENFSD) != 0) {
4027 		NFSLOCKV4ROOTMUTEX();
4028 		if (suspend_nfsd != 0) {
4029 			nfsv4_unlock(&nfsd_suspend_lock, 0);
4030 			suspend_nfsd = 0;
4031 		}
4032 		NFSUNLOCKV4ROOTMUTEX();
4033 		error = 0;
4034 	}
4035 
4036 	NFSEXITCODE(error);
4037 	return (error);
4038 }
4039 
4040 /*
4041  * Check exports.
4042  * Returns 0 if ok, 1 otherwise.
4043  */
4044 int
4045 nfsvno_testexp(struct nfsrv_descript *nd, struct nfsexstuff *exp)
4046 {
4047 	int i;
4048 
4049 	/*
4050 	 * Allow NFSv3 Fsinfo per RFC2623.
4051 	 */
4052 	if (((nd->nd_flag & ND_NFSV4) != 0 ||
4053 	     nd->nd_procnum != NFSPROC_FSINFO) &&
4054 	    ((NFSVNO_EXTLS(exp) && (nd->nd_flag & ND_TLS) == 0) ||
4055 	     (NFSVNO_EXTLSCERT(exp) &&
4056 	      (nd->nd_flag & ND_TLSCERT) == 0) ||
4057 	     (NFSVNO_EXTLSCERTUSER(exp) &&
4058 	      (nd->nd_flag & ND_TLSCERTUSER) == 0))) {
4059 		if ((nd->nd_flag & ND_NFSV4) != 0)
4060 			return (NFSERR_WRONGSEC);
4061 		else if ((nd->nd_flag & ND_TLS) == 0)
4062 			return (NFSERR_AUTHERR | AUTH_NEEDS_TLS);
4063 		else
4064 			return (NFSERR_AUTHERR | AUTH_NEEDS_TLS_MUTUAL_HOST);
4065 	}
4066 
4067 	/*
4068 	 * This seems odd, but allow the case where the security flavor
4069 	 * list is empty. This happens when NFSv4 is traversing non-exported
4070 	 * file systems. Exported file systems should always have a non-empty
4071 	 * security flavor list.
4072 	 */
4073 	if (exp->nes_numsecflavor == 0)
4074 		return (0);
4075 
4076 	for (i = 0; i < exp->nes_numsecflavor; i++) {
4077 		/*
4078 		 * The tests for privacy and integrity must be first,
4079 		 * since ND_GSS is set for everything but AUTH_SYS.
4080 		 */
4081 		if (exp->nes_secflavors[i] == RPCSEC_GSS_KRB5P &&
4082 		    (nd->nd_flag & ND_GSSPRIVACY))
4083 			return (0);
4084 		if (exp->nes_secflavors[i] == RPCSEC_GSS_KRB5I &&
4085 		    (nd->nd_flag & ND_GSSINTEGRITY))
4086 			return (0);
4087 		if (exp->nes_secflavors[i] == RPCSEC_GSS_KRB5 &&
4088 		    (nd->nd_flag & ND_GSS))
4089 			return (0);
4090 		if (exp->nes_secflavors[i] == AUTH_SYS &&
4091 		    (nd->nd_flag & ND_GSS) == 0)
4092 			return (0);
4093 	}
4094 	if ((nd->nd_flag & ND_NFSV4) != 0)
4095 		return (NFSERR_WRONGSEC);
4096 	return (NFSERR_AUTHERR | AUTH_TOOWEAK);
4097 }
4098 
4099 /*
4100  * Calculate a hash value for the fid in a file handle.
4101  */
4102 uint32_t
4103 nfsrv_hashfh(fhandle_t *fhp)
4104 {
4105 	uint32_t hashval;
4106 
4107 	hashval = hash32_buf(&fhp->fh_fid, sizeof(struct fid), 0);
4108 	return (hashval);
4109 }
4110 
4111 /*
4112  * Calculate a hash value for the sessionid.
4113  */
4114 uint32_t
4115 nfsrv_hashsessionid(uint8_t *sessionid)
4116 {
4117 	uint32_t hashval;
4118 
4119 	hashval = hash32_buf(sessionid, NFSX_V4SESSIONID, 0);
4120 	return (hashval);
4121 }
4122 
4123 /*
4124  * Signal the userland master nfsd to backup the stable restart file.
4125  */
4126 void
4127 nfsrv_backupstable(void)
4128 {
4129 	struct proc *procp;
4130 
4131 	if (nfsd_master_proc != NULL) {
4132 		procp = pfind(nfsd_master_pid);
4133 		/* Try to make sure it is the correct process. */
4134 		if (procp == nfsd_master_proc &&
4135 		    procp->p_stats->p_start.tv_sec ==
4136 		    nfsd_master_start.tv_sec &&
4137 		    procp->p_stats->p_start.tv_usec ==
4138 		    nfsd_master_start.tv_usec &&
4139 		    strcmp(procp->p_comm, nfsd_master_comm) == 0)
4140 			kern_psignal(procp, SIGUSR2);
4141 		else
4142 			nfsd_master_proc = NULL;
4143 
4144 		if (procp != NULL)
4145 			PROC_UNLOCK(procp);
4146 	}
4147 }
4148 
4149 /*
4150  * Create a DS data file for nfsrv_pnfscreate(). Called for each mirror.
4151  * The arguments are in a structure, so that they can be passed through
4152  * taskqueue for a kernel process to execute this function.
4153  */
4154 struct nfsrvdscreate {
4155 	int			done;
4156 	int			inprog;
4157 	struct task		tsk;
4158 	struct ucred		*tcred;
4159 	struct vnode		*dvp;
4160 	NFSPROC_T		*p;
4161 	struct pnfsdsfile	*pf;
4162 	int			err;
4163 	fhandle_t		fh;
4164 	struct vattr		va;
4165 	struct vattr		createva;
4166 };
4167 
4168 int
4169 nfsrv_dscreate(struct vnode *dvp, struct vattr *vap, struct vattr *nvap,
4170     fhandle_t *fhp, struct pnfsdsfile *pf, struct pnfsdsattr *dsa,
4171     char *fnamep, struct ucred *tcred, NFSPROC_T *p, struct vnode **nvpp)
4172 {
4173 	struct vnode *nvp;
4174 	struct nameidata named;
4175 	struct vattr va;
4176 	char *bufp;
4177 	u_long *hashp;
4178 	struct nfsnode *np;
4179 	struct nfsmount *nmp;
4180 	int error;
4181 
4182 	NFSNAMEICNDSET(&named.ni_cnd, tcred, CREATE,
4183 	    LOCKPARENT | LOCKLEAF | SAVESTART | NOCACHE);
4184 	nfsvno_setpathbuf(&named, &bufp, &hashp);
4185 	named.ni_cnd.cn_lkflags = LK_EXCLUSIVE;
4186 	named.ni_cnd.cn_thread = p;
4187 	named.ni_cnd.cn_nameptr = bufp;
4188 	if (fnamep != NULL) {
4189 		strlcpy(bufp, fnamep, PNFS_FILENAME_LEN + 1);
4190 		named.ni_cnd.cn_namelen = strlen(bufp);
4191 	} else
4192 		named.ni_cnd.cn_namelen = nfsrv_putfhname(fhp, bufp);
4193 	NFSD_DEBUG(4, "nfsrv_dscreate: dvp=%p fname=%s\n", dvp, bufp);
4194 
4195 	/* Create the date file in the DS mount. */
4196 	error = NFSVOPLOCK(dvp, LK_EXCLUSIVE);
4197 	if (error == 0) {
4198 		error = VOP_CREATE(dvp, &nvp, &named.ni_cnd, vap);
4199 		vref(dvp);
4200 		VOP_VPUT_PAIR(dvp, error == 0 ? &nvp : NULL, false);
4201 		if (error == 0) {
4202 			/* Set the ownership of the file. */
4203 			error = VOP_SETATTR(nvp, nvap, tcred);
4204 			NFSD_DEBUG(4, "nfsrv_dscreate:"
4205 			    " setattr-uid=%d\n", error);
4206 			if (error != 0)
4207 				vput(nvp);
4208 		}
4209 		if (error != 0)
4210 			printf("pNFS: pnfscreate failed=%d\n", error);
4211 	} else
4212 		printf("pNFS: pnfscreate vnlock=%d\n", error);
4213 	if (error == 0) {
4214 		np = VTONFS(nvp);
4215 		nmp = VFSTONFS(nvp->v_mount);
4216 		if (strcmp(nvp->v_mount->mnt_vfc->vfc_name, "nfs")
4217 		    != 0 || nmp->nm_nam->sa_len > sizeof(
4218 		    struct sockaddr_in6) ||
4219 		    np->n_fhp->nfh_len != NFSX_MYFH) {
4220 			printf("Bad DS file: fstype=%s salen=%d"
4221 			    " fhlen=%d\n",
4222 			    nvp->v_mount->mnt_vfc->vfc_name,
4223 			    nmp->nm_nam->sa_len, np->n_fhp->nfh_len);
4224 			error = ENOENT;
4225 		}
4226 
4227 		/* Set extattrs for the DS on the MDS file. */
4228 		if (error == 0) {
4229 			if (dsa != NULL) {
4230 				error = VOP_GETATTR(nvp, &va, tcred);
4231 				if (error == 0) {
4232 					dsa->dsa_filerev = va.va_filerev;
4233 					dsa->dsa_size = va.va_size;
4234 					dsa->dsa_atime = va.va_atime;
4235 					dsa->dsa_mtime = va.va_mtime;
4236 					dsa->dsa_bytes = va.va_bytes;
4237 				}
4238 			}
4239 			if (error == 0) {
4240 				NFSBCOPY(np->n_fhp->nfh_fh, &pf->dsf_fh,
4241 				    NFSX_MYFH);
4242 				NFSBCOPY(nmp->nm_nam, &pf->dsf_sin,
4243 				    nmp->nm_nam->sa_len);
4244 				NFSBCOPY(named.ni_cnd.cn_nameptr,
4245 				    pf->dsf_filename,
4246 				    sizeof(pf->dsf_filename));
4247 			}
4248 		} else
4249 			printf("pNFS: pnfscreate can't get DS"
4250 			    " attr=%d\n", error);
4251 		if (nvpp != NULL && error == 0)
4252 			*nvpp = nvp;
4253 		else
4254 			vput(nvp);
4255 	}
4256 	nfsvno_relpathbuf(&named);
4257 	return (error);
4258 }
4259 
4260 /*
4261  * Start up the thread that will execute nfsrv_dscreate().
4262  */
4263 static void
4264 start_dscreate(void *arg, int pending)
4265 {
4266 	struct nfsrvdscreate *dsc;
4267 
4268 	dsc = (struct nfsrvdscreate *)arg;
4269 	dsc->err = nfsrv_dscreate(dsc->dvp, &dsc->createva, &dsc->va, &dsc->fh,
4270 	    dsc->pf, NULL, NULL, dsc->tcred, dsc->p, NULL);
4271 	dsc->done = 1;
4272 	NFSD_DEBUG(4, "start_dscreate: err=%d\n", dsc->err);
4273 }
4274 
4275 /*
4276  * Create a pNFS data file on the Data Server(s).
4277  */
4278 static void
4279 nfsrv_pnfscreate(struct vnode *vp, struct vattr *vap, struct ucred *cred,
4280     NFSPROC_T *p)
4281 {
4282 	struct nfsrvdscreate *dsc, *tdsc = NULL;
4283 	struct nfsdevice *ds, *tds, *fds;
4284 	struct mount *mp;
4285 	struct pnfsdsfile *pf, *tpf;
4286 	struct pnfsdsattr dsattr;
4287 	struct vattr va;
4288 	struct vnode *dvp[NFSDEV_MAXMIRRORS];
4289 	struct nfsmount *nmp;
4290 	fhandle_t fh;
4291 	uid_t vauid;
4292 	gid_t vagid;
4293 	u_short vamode;
4294 	struct ucred *tcred;
4295 	int dsdir[NFSDEV_MAXMIRRORS], error, i, mirrorcnt, ret;
4296 	int failpos, timo;
4297 
4298 	/* Get a DS server directory in a round-robin order. */
4299 	mirrorcnt = 1;
4300 	mp = vp->v_mount;
4301 	ds = fds = NULL;
4302 	NFSDDSLOCK();
4303 	/*
4304 	 * Search for the first entry that handles this MDS fs, but use the
4305 	 * first entry for all MDS fs's otherwise.
4306 	 */
4307 	TAILQ_FOREACH(tds, &nfsrv_devidhead, nfsdev_list) {
4308 		if (tds->nfsdev_nmp != NULL) {
4309 			if (tds->nfsdev_mdsisset == 0 && ds == NULL)
4310 				ds = tds;
4311 			else if (tds->nfsdev_mdsisset != 0 && fsidcmp(
4312 			    &mp->mnt_stat.f_fsid, &tds->nfsdev_mdsfsid) == 0) {
4313 				ds = fds = tds;
4314 				break;
4315 			}
4316 		}
4317 	}
4318 	if (ds == NULL) {
4319 		NFSDDSUNLOCK();
4320 		NFSD_DEBUG(4, "nfsrv_pnfscreate: no srv\n");
4321 		return;
4322 	}
4323 	i = dsdir[0] = ds->nfsdev_nextdir;
4324 	ds->nfsdev_nextdir = (ds->nfsdev_nextdir + 1) % nfsrv_dsdirsize;
4325 	dvp[0] = ds->nfsdev_dsdir[i];
4326 	tds = TAILQ_NEXT(ds, nfsdev_list);
4327 	if (nfsrv_maxpnfsmirror > 1 && tds != NULL) {
4328 		TAILQ_FOREACH_FROM(tds, &nfsrv_devidhead, nfsdev_list) {
4329 			if (tds->nfsdev_nmp != NULL &&
4330 			    ((tds->nfsdev_mdsisset == 0 && fds == NULL) ||
4331 			     (tds->nfsdev_mdsisset != 0 && fds != NULL &&
4332 			      fsidcmp(&mp->mnt_stat.f_fsid,
4333 			      &tds->nfsdev_mdsfsid) == 0))) {
4334 				dsdir[mirrorcnt] = i;
4335 				dvp[mirrorcnt] = tds->nfsdev_dsdir[i];
4336 				mirrorcnt++;
4337 				if (mirrorcnt >= nfsrv_maxpnfsmirror)
4338 					break;
4339 			}
4340 		}
4341 	}
4342 	/* Put at end of list to implement round-robin usage. */
4343 	TAILQ_REMOVE(&nfsrv_devidhead, ds, nfsdev_list);
4344 	TAILQ_INSERT_TAIL(&nfsrv_devidhead, ds, nfsdev_list);
4345 	NFSDDSUNLOCK();
4346 	dsc = NULL;
4347 	if (mirrorcnt > 1)
4348 		tdsc = dsc = malloc(sizeof(*dsc) * (mirrorcnt - 1), M_TEMP,
4349 		    M_WAITOK | M_ZERO);
4350 	tpf = pf = malloc(sizeof(*pf) * nfsrv_maxpnfsmirror, M_TEMP, M_WAITOK |
4351 	    M_ZERO);
4352 
4353 	error = nfsvno_getfh(vp, &fh, p);
4354 	if (error == 0)
4355 		error = VOP_GETATTR(vp, &va, cred);
4356 	if (error == 0) {
4357 		/* Set the attributes for "vp" to Setattr the DS vp. */
4358 		vauid = va.va_uid;
4359 		vagid = va.va_gid;
4360 		vamode = va.va_mode;
4361 		VATTR_NULL(&va);
4362 		va.va_uid = vauid;
4363 		va.va_gid = vagid;
4364 		va.va_mode = vamode;
4365 		va.va_size = 0;
4366 	} else
4367 		printf("pNFS: pnfscreate getfh+attr=%d\n", error);
4368 
4369 	NFSD_DEBUG(4, "nfsrv_pnfscreate: cruid=%d crgid=%d\n", cred->cr_uid,
4370 	    cred->cr_gid);
4371 	/* Make data file name based on FH. */
4372 	tcred = newnfs_getcred();
4373 
4374 	/*
4375 	 * Create the file on each DS mirror, using kernel process(es) for the
4376 	 * additional mirrors.
4377 	 */
4378 	failpos = -1;
4379 	for (i = 0; i < mirrorcnt - 1 && error == 0; i++, tpf++, tdsc++) {
4380 		tpf->dsf_dir = dsdir[i];
4381 		tdsc->tcred = tcred;
4382 		tdsc->p = p;
4383 		tdsc->pf = tpf;
4384 		tdsc->createva = *vap;
4385 		NFSBCOPY(&fh, &tdsc->fh, sizeof(fh));
4386 		tdsc->va = va;
4387 		tdsc->dvp = dvp[i];
4388 		tdsc->done = 0;
4389 		tdsc->inprog = 0;
4390 		tdsc->err = 0;
4391 		ret = EIO;
4392 		if (nfs_pnfsiothreads != 0) {
4393 			ret = nfs_pnfsio(start_dscreate, tdsc);
4394 			NFSD_DEBUG(4, "nfsrv_pnfscreate: nfs_pnfsio=%d\n", ret);
4395 		}
4396 		if (ret != 0) {
4397 			ret = nfsrv_dscreate(dvp[i], vap, &va, &fh, tpf, NULL,
4398 			    NULL, tcred, p, NULL);
4399 			if (ret != 0) {
4400 				KASSERT(error == 0, ("nfsrv_dscreate err=%d",
4401 				    error));
4402 				if (failpos == -1 && nfsds_failerr(ret))
4403 					failpos = i;
4404 				else
4405 					error = ret;
4406 			}
4407 		}
4408 	}
4409 	if (error == 0) {
4410 		tpf->dsf_dir = dsdir[mirrorcnt - 1];
4411 		error = nfsrv_dscreate(dvp[mirrorcnt - 1], vap, &va, &fh, tpf,
4412 		    &dsattr, NULL, tcred, p, NULL);
4413 		if (failpos == -1 && mirrorcnt > 1 && nfsds_failerr(error)) {
4414 			failpos = mirrorcnt - 1;
4415 			error = 0;
4416 		}
4417 	}
4418 	timo = hz / 50;		/* Wait for 20msec. */
4419 	if (timo < 1)
4420 		timo = 1;
4421 	/* Wait for kernel task(s) to complete. */
4422 	for (tdsc = dsc, i = 0; i < mirrorcnt - 1; i++, tdsc++) {
4423 		while (tdsc->inprog != 0 && tdsc->done == 0)
4424 			tsleep(&tdsc->tsk, PVFS, "srvdcr", timo);
4425 		if (tdsc->err != 0) {
4426 			if (failpos == -1 && nfsds_failerr(tdsc->err))
4427 				failpos = i;
4428 			else if (error == 0)
4429 				error = tdsc->err;
4430 		}
4431 	}
4432 
4433 	/*
4434 	 * If failpos has been set, that mirror has failed, so it needs
4435 	 * to be disabled.
4436 	 */
4437 	if (failpos >= 0) {
4438 		nmp = VFSTONFS(dvp[failpos]->v_mount);
4439 		NFSLOCKMNT(nmp);
4440 		if ((nmp->nm_privflag & (NFSMNTP_FORCEDISM |
4441 		     NFSMNTP_CANCELRPCS)) == 0) {
4442 			nmp->nm_privflag |= NFSMNTP_CANCELRPCS;
4443 			NFSUNLOCKMNT(nmp);
4444 			ds = nfsrv_deldsnmp(PNFSDOP_DELDSSERVER, nmp, p);
4445 			NFSD_DEBUG(4, "dscreatfail fail=%d ds=%p\n", failpos,
4446 			    ds);
4447 			if (ds != NULL)
4448 				nfsrv_killrpcs(nmp);
4449 			NFSLOCKMNT(nmp);
4450 			nmp->nm_privflag &= ~NFSMNTP_CANCELRPCS;
4451 			wakeup(nmp);
4452 		}
4453 		NFSUNLOCKMNT(nmp);
4454 	}
4455 
4456 	NFSFREECRED(tcred);
4457 	if (error == 0) {
4458 		ASSERT_VOP_ELOCKED(vp, "nfsrv_pnfscreate vp");
4459 
4460 		NFSD_DEBUG(4, "nfsrv_pnfscreate: mirrorcnt=%d maxmirror=%d\n",
4461 		    mirrorcnt, nfsrv_maxpnfsmirror);
4462 		/*
4463 		 * For all mirrors that couldn't be created, fill in the
4464 		 * *pf structure, but with an IP address == 0.0.0.0.
4465 		 */
4466 		tpf = pf + mirrorcnt;
4467 		for (i = mirrorcnt; i < nfsrv_maxpnfsmirror; i++, tpf++) {
4468 			*tpf = *pf;
4469 			tpf->dsf_sin.sin_family = AF_INET;
4470 			tpf->dsf_sin.sin_len = sizeof(struct sockaddr_in);
4471 			tpf->dsf_sin.sin_addr.s_addr = 0;
4472 			tpf->dsf_sin.sin_port = 0;
4473 		}
4474 
4475 		error = vn_extattr_set(vp, IO_NODELOCKED,
4476 		    EXTATTR_NAMESPACE_SYSTEM, "pnfsd.dsfile",
4477 		    sizeof(*pf) * nfsrv_maxpnfsmirror, (char *)pf, p);
4478 		if (error == 0)
4479 			error = vn_extattr_set(vp, IO_NODELOCKED,
4480 			    EXTATTR_NAMESPACE_SYSTEM, "pnfsd.dsattr",
4481 			    sizeof(dsattr), (char *)&dsattr, p);
4482 		if (error != 0)
4483 			printf("pNFS: pnfscreate setextattr=%d\n",
4484 			    error);
4485 	} else
4486 		printf("pNFS: pnfscreate=%d\n", error);
4487 	free(pf, M_TEMP);
4488 	free(dsc, M_TEMP);
4489 }
4490 
4491 /*
4492  * Get the information needed to remove the pNFS Data Server file from the
4493  * Metadata file.  Upon success, ddvp is set non-NULL to the locked
4494  * DS directory vnode.  The caller must unlock *ddvp when done with it.
4495  */
4496 static void
4497 nfsrv_pnfsremovesetup(struct vnode *vp, NFSPROC_T *p, struct vnode **dvpp,
4498     int *mirrorcntp, char *fname, fhandle_t *fhp)
4499 {
4500 	struct vattr va;
4501 	struct ucred *tcred;
4502 	char *buf;
4503 	int buflen, error;
4504 
4505 	dvpp[0] = NULL;
4506 	/* If not an exported regular file or not a pNFS server, just return. */
4507 	if (vp->v_type != VREG || (vp->v_mount->mnt_flag & MNT_EXPORTED) == 0 ||
4508 	    nfsrv_devidcnt == 0)
4509 		return;
4510 
4511 	/* Check to see if this is the last hard link. */
4512 	tcred = newnfs_getcred();
4513 	error = VOP_GETATTR(vp, &va, tcred);
4514 	NFSFREECRED(tcred);
4515 	if (error != 0) {
4516 		printf("pNFS: nfsrv_pnfsremovesetup getattr=%d\n", error);
4517 		return;
4518 	}
4519 	if (va.va_nlink > 1)
4520 		return;
4521 
4522 	error = nfsvno_getfh(vp, fhp, p);
4523 	if (error != 0) {
4524 		printf("pNFS: nfsrv_pnfsremovesetup getfh=%d\n", error);
4525 		return;
4526 	}
4527 
4528 	buflen = 1024;
4529 	buf = malloc(buflen, M_TEMP, M_WAITOK);
4530 	/* Get the directory vnode for the DS mount and the file handle. */
4531 	error = nfsrv_dsgetsockmnt(vp, 0, buf, &buflen, mirrorcntp, p, dvpp,
4532 	    NULL, NULL, fname, NULL, NULL, NULL, NULL, NULL);
4533 	free(buf, M_TEMP);
4534 	if (error != 0)
4535 		printf("pNFS: nfsrv_pnfsremovesetup getsockmnt=%d\n", error);
4536 }
4537 
4538 /*
4539  * Remove a DS data file for nfsrv_pnfsremove(). Called for each mirror.
4540  * The arguments are in a structure, so that they can be passed through
4541  * taskqueue for a kernel process to execute this function.
4542  */
4543 struct nfsrvdsremove {
4544 	int			done;
4545 	int			inprog;
4546 	struct task		tsk;
4547 	struct ucred		*tcred;
4548 	struct vnode		*dvp;
4549 	NFSPROC_T		*p;
4550 	int			err;
4551 	char			fname[PNFS_FILENAME_LEN + 1];
4552 };
4553 
4554 static int
4555 nfsrv_dsremove(struct vnode *dvp, char *fname, struct ucred *tcred,
4556     NFSPROC_T *p)
4557 {
4558 	struct nameidata named;
4559 	struct vnode *nvp;
4560 	char *bufp;
4561 	u_long *hashp;
4562 	int error;
4563 
4564 	error = NFSVOPLOCK(dvp, LK_EXCLUSIVE);
4565 	if (error != 0)
4566 		return (error);
4567 	named.ni_cnd.cn_nameiop = DELETE;
4568 	named.ni_cnd.cn_lkflags = LK_EXCLUSIVE | LK_RETRY;
4569 	named.ni_cnd.cn_cred = tcred;
4570 	named.ni_cnd.cn_thread = p;
4571 	named.ni_cnd.cn_flags = ISLASTCN | LOCKPARENT | LOCKLEAF | SAVENAME;
4572 	nfsvno_setpathbuf(&named, &bufp, &hashp);
4573 	named.ni_cnd.cn_nameptr = bufp;
4574 	named.ni_cnd.cn_namelen = strlen(fname);
4575 	strlcpy(bufp, fname, NAME_MAX);
4576 	NFSD_DEBUG(4, "nfsrv_pnfsremove: filename=%s\n", bufp);
4577 	error = VOP_LOOKUP(dvp, &nvp, &named.ni_cnd);
4578 	NFSD_DEBUG(4, "nfsrv_pnfsremove: aft LOOKUP=%d\n", error);
4579 	if (error == 0) {
4580 		error = VOP_REMOVE(dvp, nvp, &named.ni_cnd);
4581 		vput(nvp);
4582 	}
4583 	NFSVOPUNLOCK(dvp);
4584 	nfsvno_relpathbuf(&named);
4585 	if (error != 0)
4586 		printf("pNFS: nfsrv_pnfsremove failed=%d\n", error);
4587 	return (error);
4588 }
4589 
4590 /*
4591  * Start up the thread that will execute nfsrv_dsremove().
4592  */
4593 static void
4594 start_dsremove(void *arg, int pending)
4595 {
4596 	struct nfsrvdsremove *dsrm;
4597 
4598 	dsrm = (struct nfsrvdsremove *)arg;
4599 	dsrm->err = nfsrv_dsremove(dsrm->dvp, dsrm->fname, dsrm->tcred,
4600 	    dsrm->p);
4601 	dsrm->done = 1;
4602 	NFSD_DEBUG(4, "start_dsremove: err=%d\n", dsrm->err);
4603 }
4604 
4605 /*
4606  * Remove a pNFS data file from a Data Server.
4607  * nfsrv_pnfsremovesetup() must have been called before the MDS file was
4608  * removed to set up the dvp and fill in the FH.
4609  */
4610 static void
4611 nfsrv_pnfsremove(struct vnode **dvp, int mirrorcnt, char *fname, fhandle_t *fhp,
4612     NFSPROC_T *p)
4613 {
4614 	struct ucred *tcred;
4615 	struct nfsrvdsremove *dsrm, *tdsrm;
4616 	struct nfsdevice *ds;
4617 	struct nfsmount *nmp;
4618 	int failpos, i, ret, timo;
4619 
4620 	tcred = newnfs_getcred();
4621 	dsrm = NULL;
4622 	if (mirrorcnt > 1)
4623 		dsrm = malloc(sizeof(*dsrm) * mirrorcnt - 1, M_TEMP, M_WAITOK);
4624 	/*
4625 	 * Remove the file on each DS mirror, using kernel process(es) for the
4626 	 * additional mirrors.
4627 	 */
4628 	failpos = -1;
4629 	for (tdsrm = dsrm, i = 0; i < mirrorcnt - 1; i++, tdsrm++) {
4630 		tdsrm->tcred = tcred;
4631 		tdsrm->p = p;
4632 		tdsrm->dvp = dvp[i];
4633 		strlcpy(tdsrm->fname, fname, PNFS_FILENAME_LEN + 1);
4634 		tdsrm->inprog = 0;
4635 		tdsrm->done = 0;
4636 		tdsrm->err = 0;
4637 		ret = EIO;
4638 		if (nfs_pnfsiothreads != 0) {
4639 			ret = nfs_pnfsio(start_dsremove, tdsrm);
4640 			NFSD_DEBUG(4, "nfsrv_pnfsremove: nfs_pnfsio=%d\n", ret);
4641 		}
4642 		if (ret != 0) {
4643 			ret = nfsrv_dsremove(dvp[i], fname, tcred, p);
4644 			if (failpos == -1 && nfsds_failerr(ret))
4645 				failpos = i;
4646 		}
4647 	}
4648 	ret = nfsrv_dsremove(dvp[mirrorcnt - 1], fname, tcred, p);
4649 	if (failpos == -1 && mirrorcnt > 1 && nfsds_failerr(ret))
4650 		failpos = mirrorcnt - 1;
4651 	timo = hz / 50;		/* Wait for 20msec. */
4652 	if (timo < 1)
4653 		timo = 1;
4654 	/* Wait for kernel task(s) to complete. */
4655 	for (tdsrm = dsrm, i = 0; i < mirrorcnt - 1; i++, tdsrm++) {
4656 		while (tdsrm->inprog != 0 && tdsrm->done == 0)
4657 			tsleep(&tdsrm->tsk, PVFS, "srvdsrm", timo);
4658 		if (failpos == -1 && nfsds_failerr(tdsrm->err))
4659 			failpos = i;
4660 	}
4661 
4662 	/*
4663 	 * If failpos has been set, that mirror has failed, so it needs
4664 	 * to be disabled.
4665 	 */
4666 	if (failpos >= 0) {
4667 		nmp = VFSTONFS(dvp[failpos]->v_mount);
4668 		NFSLOCKMNT(nmp);
4669 		if ((nmp->nm_privflag & (NFSMNTP_FORCEDISM |
4670 		     NFSMNTP_CANCELRPCS)) == 0) {
4671 			nmp->nm_privflag |= NFSMNTP_CANCELRPCS;
4672 			NFSUNLOCKMNT(nmp);
4673 			ds = nfsrv_deldsnmp(PNFSDOP_DELDSSERVER, nmp, p);
4674 			NFSD_DEBUG(4, "dsremovefail fail=%d ds=%p\n", failpos,
4675 			    ds);
4676 			if (ds != NULL)
4677 				nfsrv_killrpcs(nmp);
4678 			NFSLOCKMNT(nmp);
4679 			nmp->nm_privflag &= ~NFSMNTP_CANCELRPCS;
4680 			wakeup(nmp);
4681 		}
4682 		NFSUNLOCKMNT(nmp);
4683 	}
4684 
4685 	/* Get rid all layouts for the file. */
4686 	nfsrv_freefilelayouts(fhp);
4687 
4688 	NFSFREECRED(tcred);
4689 	free(dsrm, M_TEMP);
4690 }
4691 
4692 /*
4693  * Generate a file name based on the file handle and put it in *bufp.
4694  * Return the number of bytes generated.
4695  */
4696 static int
4697 nfsrv_putfhname(fhandle_t *fhp, char *bufp)
4698 {
4699 	int i;
4700 	uint8_t *cp;
4701 	const uint8_t *hexdigits = "0123456789abcdef";
4702 
4703 	cp = (uint8_t *)fhp;
4704 	for (i = 0; i < sizeof(*fhp); i++) {
4705 		bufp[2 * i] = hexdigits[(*cp >> 4) & 0xf];
4706 		bufp[2 * i + 1] = hexdigits[*cp++ & 0xf];
4707 	}
4708 	bufp[2 * i] = '\0';
4709 	return (2 * i);
4710 }
4711 
4712 /*
4713  * Update the Metadata file's attributes from the DS file when a Read/Write
4714  * layout is returned.
4715  * Basically just call nfsrv_proxyds() with procedure == NFSPROC_LAYOUTRETURN
4716  * so that it does a nfsrv_getattrdsrpc() and nfsrv_setextattr() on the DS file.
4717  */
4718 int
4719 nfsrv_updatemdsattr(struct vnode *vp, struct nfsvattr *nap, NFSPROC_T *p)
4720 {
4721 	struct ucred *tcred;
4722 	int error;
4723 
4724 	/* Do this as root so that it won't fail with EACCES. */
4725 	tcred = newnfs_getcred();
4726 	error = nfsrv_proxyds(vp, 0, 0, tcred, p, NFSPROC_LAYOUTRETURN,
4727 	    NULL, NULL, NULL, nap, NULL, NULL, 0, NULL);
4728 	NFSFREECRED(tcred);
4729 	return (error);
4730 }
4731 
4732 /*
4733  * Set the NFSv4 ACL on the DS file to the same ACL as the MDS file.
4734  */
4735 static int
4736 nfsrv_dssetacl(struct vnode *vp, struct acl *aclp, struct ucred *cred,
4737     NFSPROC_T *p)
4738 {
4739 	int error;
4740 
4741 	error = nfsrv_proxyds(vp, 0, 0, cred, p, NFSPROC_SETACL,
4742 	    NULL, NULL, NULL, NULL, aclp, NULL, 0, NULL);
4743 	return (error);
4744 }
4745 
4746 static int
4747 nfsrv_proxyds(struct vnode *vp, off_t off, int cnt, struct ucred *cred,
4748     struct thread *p, int ioproc, struct mbuf **mpp, char *cp,
4749     struct mbuf **mpp2, struct nfsvattr *nap, struct acl *aclp,
4750     off_t *offp, int content, bool *eofp)
4751 {
4752 	struct nfsmount *nmp[NFSDEV_MAXMIRRORS], *failnmp;
4753 	fhandle_t fh[NFSDEV_MAXMIRRORS];
4754 	struct vnode *dvp[NFSDEV_MAXMIRRORS];
4755 	struct nfsdevice *ds;
4756 	struct pnfsdsattr dsattr;
4757 	struct opnfsdsattr odsattr;
4758 	char *buf;
4759 	int buflen, error, failpos, i, mirrorcnt, origmircnt, trycnt;
4760 
4761 	NFSD_DEBUG(4, "in nfsrv_proxyds\n");
4762 	/*
4763 	 * If not a regular file, not exported or not a pNFS server,
4764 	 * just return ENOENT.
4765 	 */
4766 	if (vp->v_type != VREG || (vp->v_mount->mnt_flag & MNT_EXPORTED) == 0 ||
4767 	    nfsrv_devidcnt == 0)
4768 		return (ENOENT);
4769 
4770 	buflen = 1024;
4771 	buf = malloc(buflen, M_TEMP, M_WAITOK);
4772 	error = 0;
4773 
4774 	/*
4775 	 * For Getattr, get the Change attribute (va_filerev) and size (va_size)
4776 	 * from the MetaData file's extended attribute.
4777 	 */
4778 	if (ioproc == NFSPROC_GETATTR) {
4779 		error = vn_extattr_get(vp, IO_NODELOCKED,
4780 		    EXTATTR_NAMESPACE_SYSTEM, "pnfsd.dsattr", &buflen, buf,
4781 		    p);
4782 		if (error == 0) {
4783 			if (buflen == sizeof(odsattr)) {
4784 				NFSBCOPY(buf, &odsattr, buflen);
4785 				nap->na_filerev = odsattr.dsa_filerev;
4786 				nap->na_size = odsattr.dsa_size;
4787 				nap->na_atime = odsattr.dsa_atime;
4788 				nap->na_mtime = odsattr.dsa_mtime;
4789 				/*
4790 				 * Fake na_bytes by rounding up na_size.
4791 				 * Since we don't know the block size, just
4792 				 * use BLKDEV_IOSIZE.
4793 				 */
4794 				nap->na_bytes = (odsattr.dsa_size +
4795 				    BLKDEV_IOSIZE - 1) & ~(BLKDEV_IOSIZE - 1);
4796 			} else if (buflen == sizeof(dsattr)) {
4797 				NFSBCOPY(buf, &dsattr, buflen);
4798 				nap->na_filerev = dsattr.dsa_filerev;
4799 				nap->na_size = dsattr.dsa_size;
4800 				nap->na_atime = dsattr.dsa_atime;
4801 				nap->na_mtime = dsattr.dsa_mtime;
4802 				nap->na_bytes = dsattr.dsa_bytes;
4803 			} else
4804 				error = ENXIO;
4805 		}
4806 		if (error == 0) {
4807 			/*
4808 			 * If nfsrv_pnfsgetdsattr is 0 or nfsrv_checkdsattr()
4809 			 * returns 0, just return now.  nfsrv_checkdsattr()
4810 			 * returns 0 if there is no Read/Write layout
4811 			 * plus either an Open/Write_access or Write
4812 			 * delegation issued to a client for the file.
4813 			 */
4814 			if (nfsrv_pnfsgetdsattr == 0 ||
4815 			    nfsrv_checkdsattr(vp, p) == 0) {
4816 				free(buf, M_TEMP);
4817 				return (error);
4818 			}
4819 		}
4820 
4821 		/*
4822 		 * Clear ENOATTR so the code below will attempt to do a
4823 		 * nfsrv_getattrdsrpc() to get the attributes and (re)create
4824 		 * the extended attribute.
4825 		 */
4826 		if (error == ENOATTR)
4827 			error = 0;
4828 	}
4829 
4830 	origmircnt = -1;
4831 	trycnt = 0;
4832 tryagain:
4833 	if (error == 0) {
4834 		buflen = 1024;
4835 		if (ioproc == NFSPROC_READDS && NFSVOPISLOCKED(vp) ==
4836 		    LK_EXCLUSIVE)
4837 			printf("nfsrv_proxyds: Readds vp exclusively locked\n");
4838 		error = nfsrv_dsgetsockmnt(vp, LK_SHARED, buf, &buflen,
4839 		    &mirrorcnt, p, dvp, fh, NULL, NULL, NULL, NULL, NULL,
4840 		    NULL, NULL);
4841 		if (error == 0) {
4842 			for (i = 0; i < mirrorcnt; i++)
4843 				nmp[i] = VFSTONFS(dvp[i]->v_mount);
4844 		} else
4845 			printf("pNFS: proxy getextattr sockaddr=%d\n", error);
4846 	} else
4847 		printf("pNFS: nfsrv_dsgetsockmnt=%d\n", error);
4848 	if (error == 0) {
4849 		failpos = -1;
4850 		if (origmircnt == -1)
4851 			origmircnt = mirrorcnt;
4852 		/*
4853 		 * If failpos is set to a mirror#, then that mirror has
4854 		 * failed and will be disabled. For Read, Getattr and Seek, the
4855 		 * function only tries one mirror, so if that mirror has
4856 		 * failed, it will need to be retried. As such, increment
4857 		 * tryitagain for these cases.
4858 		 * For Write, Setattr and Setacl, the function tries all
4859 		 * mirrors and will not return an error for the case where
4860 		 * one mirror has failed. For these cases, the functioning
4861 		 * mirror(s) will have been modified, so a retry isn't
4862 		 * necessary. These functions will set failpos for the
4863 		 * failed mirror#.
4864 		 */
4865 		if (ioproc == NFSPROC_READDS) {
4866 			error = nfsrv_readdsrpc(fh, off, cnt, cred, p, nmp[0],
4867 			    mpp, mpp2);
4868 			if (nfsds_failerr(error) && mirrorcnt > 1) {
4869 				/*
4870 				 * Setting failpos will cause the mirror
4871 				 * to be disabled and then a retry of this
4872 				 * read is required.
4873 				 */
4874 				failpos = 0;
4875 				error = 0;
4876 				trycnt++;
4877 			}
4878 		} else if (ioproc == NFSPROC_WRITEDS)
4879 			error = nfsrv_writedsrpc(fh, off, cnt, cred, p, vp,
4880 			    &nmp[0], mirrorcnt, mpp, cp, &failpos);
4881 		else if (ioproc == NFSPROC_SETATTR)
4882 			error = nfsrv_setattrdsrpc(fh, cred, p, vp, &nmp[0],
4883 			    mirrorcnt, nap, &failpos);
4884 		else if (ioproc == NFSPROC_SETACL)
4885 			error = nfsrv_setacldsrpc(fh, cred, p, vp, &nmp[0],
4886 			    mirrorcnt, aclp, &failpos);
4887 		else if (ioproc == NFSPROC_SEEKDS) {
4888 			error = nfsrv_seekdsrpc(fh, offp, content, eofp, cred,
4889 			    p, nmp[0]);
4890 			if (nfsds_failerr(error) && mirrorcnt > 1) {
4891 				/*
4892 				 * Setting failpos will cause the mirror
4893 				 * to be disabled and then a retry of this
4894 				 * read is required.
4895 				 */
4896 				failpos = 0;
4897 				error = 0;
4898 				trycnt++;
4899 			}
4900 		} else if (ioproc == NFSPROC_ALLOCATE)
4901 			error = nfsrv_allocatedsrpc(fh, off, *offp, cred, p, vp,
4902 			    &nmp[0], mirrorcnt, &failpos);
4903 		else if (ioproc == NFSPROC_DEALLOCATE)
4904 			error = nfsrv_deallocatedsrpc(fh, off, *offp, cred, p,
4905 			    vp, &nmp[0], mirrorcnt, &failpos);
4906 		else {
4907 			error = nfsrv_getattrdsrpc(&fh[mirrorcnt - 1], cred, p,
4908 			    vp, nmp[mirrorcnt - 1], nap);
4909 			if (nfsds_failerr(error) && mirrorcnt > 1) {
4910 				/*
4911 				 * Setting failpos will cause the mirror
4912 				 * to be disabled and then a retry of this
4913 				 * getattr is required.
4914 				 */
4915 				failpos = mirrorcnt - 1;
4916 				error = 0;
4917 				trycnt++;
4918 			}
4919 		}
4920 		ds = NULL;
4921 		if (failpos >= 0) {
4922 			failnmp = nmp[failpos];
4923 			NFSLOCKMNT(failnmp);
4924 			if ((failnmp->nm_privflag & (NFSMNTP_FORCEDISM |
4925 			     NFSMNTP_CANCELRPCS)) == 0) {
4926 				failnmp->nm_privflag |= NFSMNTP_CANCELRPCS;
4927 				NFSUNLOCKMNT(failnmp);
4928 				ds = nfsrv_deldsnmp(PNFSDOP_DELDSSERVER,
4929 				    failnmp, p);
4930 				NFSD_DEBUG(4, "dsldsnmp fail=%d ds=%p\n",
4931 				    failpos, ds);
4932 				if (ds != NULL)
4933 					nfsrv_killrpcs(failnmp);
4934 				NFSLOCKMNT(failnmp);
4935 				failnmp->nm_privflag &= ~NFSMNTP_CANCELRPCS;
4936 				wakeup(failnmp);
4937 			}
4938 			NFSUNLOCKMNT(failnmp);
4939 		}
4940 		for (i = 0; i < mirrorcnt; i++)
4941 			NFSVOPUNLOCK(dvp[i]);
4942 		NFSD_DEBUG(4, "nfsrv_proxyds: aft RPC=%d trya=%d\n", error,
4943 		    trycnt);
4944 		/* Try the Read/Getattr again if a mirror was deleted. */
4945 		if (ds != NULL && trycnt > 0 && trycnt < origmircnt)
4946 			goto tryagain;
4947 	} else {
4948 		/* Return ENOENT for any Extended Attribute error. */
4949 		error = ENOENT;
4950 	}
4951 	free(buf, M_TEMP);
4952 	NFSD_DEBUG(4, "nfsrv_proxyds: error=%d\n", error);
4953 	return (error);
4954 }
4955 
4956 /*
4957  * Get the DS mount point, fh and directory from the "pnfsd.dsfile" extended
4958  * attribute.
4959  * newnmpp - If it points to a non-NULL nmp, that is the destination and needs
4960  *           to be checked.  If it points to a NULL nmp, then it returns
4961  *           a suitable destination.
4962  * curnmp - If non-NULL, it is the source mount for the copy.
4963  */
4964 int
4965 nfsrv_dsgetsockmnt(struct vnode *vp, int lktype, char *buf, int *buflenp,
4966     int *mirrorcntp, NFSPROC_T *p, struct vnode **dvpp, fhandle_t *fhp,
4967     char *devid, char *fnamep, struct vnode **nvpp, struct nfsmount **newnmpp,
4968     struct nfsmount *curnmp, int *ippos, int *dsdirp)
4969 {
4970 	struct vnode *dvp, *nvp = NULL, **tdvpp;
4971 	struct mount *mp;
4972 	struct nfsmount *nmp, *newnmp;
4973 	struct sockaddr *sad;
4974 	struct sockaddr_in *sin;
4975 	struct nfsdevice *ds, *tds, *fndds;
4976 	struct pnfsdsfile *pf;
4977 	uint32_t dsdir;
4978 	int error, fhiszero, fnd, gotone, i, mirrorcnt;
4979 
4980 	ASSERT_VOP_LOCKED(vp, "nfsrv_dsgetsockmnt vp");
4981 	*mirrorcntp = 1;
4982 	tdvpp = dvpp;
4983 	if (nvpp != NULL)
4984 		*nvpp = NULL;
4985 	if (dvpp != NULL)
4986 		*dvpp = NULL;
4987 	if (ippos != NULL)
4988 		*ippos = -1;
4989 	if (newnmpp != NULL)
4990 		newnmp = *newnmpp;
4991 	else
4992 		newnmp = NULL;
4993 	mp = vp->v_mount;
4994 	error = vn_extattr_get(vp, IO_NODELOCKED, EXTATTR_NAMESPACE_SYSTEM,
4995 	    "pnfsd.dsfile", buflenp, buf, p);
4996 	mirrorcnt = *buflenp / sizeof(*pf);
4997 	if (error == 0 && (mirrorcnt < 1 || mirrorcnt > NFSDEV_MAXMIRRORS ||
4998 	    *buflenp != sizeof(*pf) * mirrorcnt))
4999 		error = ENOATTR;
5000 
5001 	pf = (struct pnfsdsfile *)buf;
5002 	/* If curnmp != NULL, check for a match in the mirror list. */
5003 	if (curnmp != NULL && error == 0) {
5004 		fnd = 0;
5005 		for (i = 0; i < mirrorcnt; i++, pf++) {
5006 			sad = (struct sockaddr *)&pf->dsf_sin;
5007 			if (nfsaddr2_match(sad, curnmp->nm_nam)) {
5008 				if (ippos != NULL)
5009 					*ippos = i;
5010 				fnd = 1;
5011 				break;
5012 			}
5013 		}
5014 		if (fnd == 0)
5015 			error = ENXIO;
5016 	}
5017 
5018 	gotone = 0;
5019 	pf = (struct pnfsdsfile *)buf;
5020 	NFSD_DEBUG(4, "nfsrv_dsgetsockmnt: mirrorcnt=%d err=%d\n", mirrorcnt,
5021 	    error);
5022 	for (i = 0; i < mirrorcnt && error == 0; i++, pf++) {
5023 		fhiszero = 0;
5024 		sad = (struct sockaddr *)&pf->dsf_sin;
5025 		sin = &pf->dsf_sin;
5026 		dsdir = pf->dsf_dir;
5027 		if (dsdir >= nfsrv_dsdirsize) {
5028 			printf("nfsrv_dsgetsockmnt: dsdir=%d\n", dsdir);
5029 			error = ENOATTR;
5030 		} else if (nvpp != NULL && newnmp != NULL &&
5031 		    nfsaddr2_match(sad, newnmp->nm_nam))
5032 			error = EEXIST;
5033 		if (error == 0) {
5034 			if (ippos != NULL && curnmp == NULL &&
5035 			    sad->sa_family == AF_INET &&
5036 			    sin->sin_addr.s_addr == 0)
5037 				*ippos = i;
5038 			if (NFSBCMP(&zerofh, &pf->dsf_fh, sizeof(zerofh)) == 0)
5039 				fhiszero = 1;
5040 			/* Use the socket address to find the mount point. */
5041 			fndds = NULL;
5042 			NFSDDSLOCK();
5043 			/* Find a match for the IP address. */
5044 			TAILQ_FOREACH(ds, &nfsrv_devidhead, nfsdev_list) {
5045 				if (ds->nfsdev_nmp != NULL) {
5046 					dvp = ds->nfsdev_dvp;
5047 					nmp = VFSTONFS(dvp->v_mount);
5048 					if (nmp != ds->nfsdev_nmp)
5049 						printf("different2 nmp %p %p\n",
5050 						    nmp, ds->nfsdev_nmp);
5051 					if (nfsaddr2_match(sad, nmp->nm_nam)) {
5052 						fndds = ds;
5053 						break;
5054 					}
5055 				}
5056 			}
5057 			if (fndds != NULL && newnmpp != NULL &&
5058 			    newnmp == NULL) {
5059 				/* Search for a place to make a mirror copy. */
5060 				TAILQ_FOREACH(tds, &nfsrv_devidhead,
5061 				    nfsdev_list) {
5062 					if (tds->nfsdev_nmp != NULL &&
5063 					    fndds != tds &&
5064 					    ((tds->nfsdev_mdsisset == 0 &&
5065 					      fndds->nfsdev_mdsisset == 0) ||
5066 					     (tds->nfsdev_mdsisset != 0 &&
5067 					      fndds->nfsdev_mdsisset != 0 &&
5068 					      fsidcmp(&tds->nfsdev_mdsfsid,
5069 					      &mp->mnt_stat.f_fsid) == 0))) {
5070 						*newnmpp = tds->nfsdev_nmp;
5071 						break;
5072 					}
5073 				}
5074 				if (tds != NULL) {
5075 					/*
5076 					 * Move this entry to the end of the
5077 					 * list, so it won't be selected as
5078 					 * easily the next time.
5079 					 */
5080 					TAILQ_REMOVE(&nfsrv_devidhead, tds,
5081 					    nfsdev_list);
5082 					TAILQ_INSERT_TAIL(&nfsrv_devidhead, tds,
5083 					    nfsdev_list);
5084 				}
5085 			}
5086 			NFSDDSUNLOCK();
5087 			if (fndds != NULL) {
5088 				dvp = fndds->nfsdev_dsdir[dsdir];
5089 				if (lktype != 0 || fhiszero != 0 ||
5090 				    (nvpp != NULL && *nvpp == NULL)) {
5091 					if (fhiszero != 0)
5092 						error = vn_lock(dvp,
5093 						    LK_EXCLUSIVE);
5094 					else if (lktype != 0)
5095 						error = vn_lock(dvp, lktype);
5096 					else
5097 						error = vn_lock(dvp, LK_SHARED);
5098 					/*
5099 					 * If the file handle is all 0's, try to
5100 					 * do a Lookup against the DS to acquire
5101 					 * it.
5102 					 * If dvpp == NULL or the Lookup fails,
5103 					 * unlock dvp after the call.
5104 					 */
5105 					if (error == 0 && (fhiszero != 0 ||
5106 					    (nvpp != NULL && *nvpp == NULL))) {
5107 						error = nfsrv_pnfslookupds(vp,
5108 						    dvp, pf, &nvp, p);
5109 						if (error == 0) {
5110 							if (fhiszero != 0)
5111 								nfsrv_pnfssetfh(
5112 								    vp, pf,
5113 								    devid,
5114 								    fnamep,
5115 								    nvp, p);
5116 							if (nvpp != NULL &&
5117 							    *nvpp == NULL) {
5118 								*nvpp = nvp;
5119 								*dsdirp = dsdir;
5120 							} else
5121 								vput(nvp);
5122 						}
5123 						if (error != 0 || lktype == 0)
5124 							NFSVOPUNLOCK(dvp);
5125 					}
5126 				}
5127 				if (error == 0) {
5128 					gotone++;
5129 					NFSD_DEBUG(4, "gotone=%d\n", gotone);
5130 					if (devid != NULL) {
5131 						NFSBCOPY(fndds->nfsdev_deviceid,
5132 						    devid, NFSX_V4DEVICEID);
5133 						devid += NFSX_V4DEVICEID;
5134 					}
5135 					if (dvpp != NULL)
5136 						*tdvpp++ = dvp;
5137 					if (fhp != NULL)
5138 						NFSBCOPY(&pf->dsf_fh, fhp++,
5139 						    NFSX_MYFH);
5140 					if (fnamep != NULL && gotone == 1)
5141 						strlcpy(fnamep,
5142 						    pf->dsf_filename,
5143 						    sizeof(pf->dsf_filename));
5144 				} else
5145 					NFSD_DEBUG(4, "nfsrv_dsgetsockmnt "
5146 					    "err=%d\n", error);
5147 			}
5148 		}
5149 	}
5150 	if (error == 0 && gotone == 0)
5151 		error = ENOENT;
5152 
5153 	NFSD_DEBUG(4, "eo nfsrv_dsgetsockmnt: gotone=%d err=%d\n", gotone,
5154 	    error);
5155 	if (error == 0)
5156 		*mirrorcntp = gotone;
5157 	else {
5158 		if (gotone > 0 && dvpp != NULL) {
5159 			/*
5160 			 * If the error didn't occur on the first one and
5161 			 * dvpp != NULL, the one(s) prior to the failure will
5162 			 * have locked dvp's that need to be unlocked.
5163 			 */
5164 			for (i = 0; i < gotone; i++) {
5165 				NFSVOPUNLOCK(*dvpp);
5166 				*dvpp++ = NULL;
5167 			}
5168 		}
5169 		/*
5170 		 * If it found the vnode to be copied from before a failure,
5171 		 * it needs to be vput()'d.
5172 		 */
5173 		if (nvpp != NULL && *nvpp != NULL) {
5174 			vput(*nvpp);
5175 			*nvpp = NULL;
5176 		}
5177 	}
5178 	return (error);
5179 }
5180 
5181 /*
5182  * Set the extended attribute for the Change attribute.
5183  */
5184 static int
5185 nfsrv_setextattr(struct vnode *vp, struct nfsvattr *nap, NFSPROC_T *p)
5186 {
5187 	struct pnfsdsattr dsattr;
5188 	int error;
5189 
5190 	ASSERT_VOP_ELOCKED(vp, "nfsrv_setextattr vp");
5191 	dsattr.dsa_filerev = nap->na_filerev;
5192 	dsattr.dsa_size = nap->na_size;
5193 	dsattr.dsa_atime = nap->na_atime;
5194 	dsattr.dsa_mtime = nap->na_mtime;
5195 	dsattr.dsa_bytes = nap->na_bytes;
5196 	error = vn_extattr_set(vp, IO_NODELOCKED, EXTATTR_NAMESPACE_SYSTEM,
5197 	    "pnfsd.dsattr", sizeof(dsattr), (char *)&dsattr, p);
5198 	if (error != 0)
5199 		printf("pNFS: setextattr=%d\n", error);
5200 	return (error);
5201 }
5202 
5203 static int
5204 nfsrv_readdsrpc(fhandle_t *fhp, off_t off, int len, struct ucred *cred,
5205     NFSPROC_T *p, struct nfsmount *nmp, struct mbuf **mpp, struct mbuf **mpendp)
5206 {
5207 	uint32_t *tl;
5208 	struct nfsrv_descript *nd;
5209 	nfsv4stateid_t st;
5210 	struct mbuf *m, *m2;
5211 	int error = 0, retlen, tlen, trimlen;
5212 
5213 	NFSD_DEBUG(4, "in nfsrv_readdsrpc\n");
5214 	nd = malloc(sizeof(*nd), M_TEMP, M_WAITOK | M_ZERO);
5215 	*mpp = NULL;
5216 	/*
5217 	 * Use a stateid where other is an alternating 01010 pattern and
5218 	 * seqid is 0xffffffff.  This value is not defined as special by
5219 	 * the RFC and is used by the FreeBSD NFS server to indicate an
5220 	 * MDS->DS proxy operation.
5221 	 */
5222 	st.other[0] = 0x55555555;
5223 	st.other[1] = 0x55555555;
5224 	st.other[2] = 0x55555555;
5225 	st.seqid = 0xffffffff;
5226 	nfscl_reqstart(nd, NFSPROC_READDS, nmp, (u_int8_t *)fhp, sizeof(*fhp),
5227 	    NULL, NULL, 0, 0);
5228 	nfsm_stateidtom(nd, &st, NFSSTATEID_PUTSTATEID);
5229 	NFSM_BUILD(tl, uint32_t *, NFSX_UNSIGNED * 3);
5230 	txdr_hyper(off, tl);
5231 	*(tl + 2) = txdr_unsigned(len);
5232 	error = newnfs_request(nd, nmp, NULL, &nmp->nm_sockreq, NULL, p, cred,
5233 	    NFS_PROG, NFS_VER4, NULL, 1, NULL, NULL);
5234 	if (error != 0) {
5235 		free(nd, M_TEMP);
5236 		return (error);
5237 	}
5238 	if (nd->nd_repstat == 0) {
5239 		NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
5240 		NFSM_STRSIZ(retlen, len);
5241 		if (retlen > 0) {
5242 			/* Trim off the pre-data XDR from the mbuf chain. */
5243 			m = nd->nd_mrep;
5244 			while (m != NULL && m != nd->nd_md) {
5245 				if (m->m_next == nd->nd_md) {
5246 					m->m_next = NULL;
5247 					m_freem(nd->nd_mrep);
5248 					nd->nd_mrep = m = nd->nd_md;
5249 				} else
5250 					m = m->m_next;
5251 			}
5252 			if (m == NULL) {
5253 				printf("nfsrv_readdsrpc: busted mbuf list\n");
5254 				error = ENOENT;
5255 				goto nfsmout;
5256 			}
5257 
5258 			/*
5259 			 * Now, adjust first mbuf so that any XDR before the
5260 			 * read data is skipped over.
5261 			 */
5262 			trimlen = nd->nd_dpos - mtod(m, char *);
5263 			if (trimlen > 0) {
5264 				m->m_len -= trimlen;
5265 				NFSM_DATAP(m, trimlen);
5266 			}
5267 
5268 			/*
5269 			 * Truncate the mbuf chain at retlen bytes of data,
5270 			 * plus XDR padding that brings the length up to a
5271 			 * multiple of 4.
5272 			 */
5273 			tlen = NFSM_RNDUP(retlen);
5274 			do {
5275 				if (m->m_len >= tlen) {
5276 					m->m_len = tlen;
5277 					tlen = 0;
5278 					m2 = m->m_next;
5279 					m->m_next = NULL;
5280 					m_freem(m2);
5281 					break;
5282 				}
5283 				tlen -= m->m_len;
5284 				m = m->m_next;
5285 			} while (m != NULL);
5286 			if (tlen > 0) {
5287 				printf("nfsrv_readdsrpc: busted mbuf list\n");
5288 				error = ENOENT;
5289 				goto nfsmout;
5290 			}
5291 			*mpp = nd->nd_mrep;
5292 			*mpendp = m;
5293 			nd->nd_mrep = NULL;
5294 		}
5295 	} else
5296 		error = nd->nd_repstat;
5297 nfsmout:
5298 	/* If nd->nd_mrep is already NULL, this is a no-op. */
5299 	m_freem(nd->nd_mrep);
5300 	free(nd, M_TEMP);
5301 	NFSD_DEBUG(4, "nfsrv_readdsrpc error=%d\n", error);
5302 	return (error);
5303 }
5304 
5305 /*
5306  * Do a write RPC on a DS data file, using this structure for the arguments,
5307  * so that this function can be executed by a separate kernel process.
5308  */
5309 struct nfsrvwritedsdorpc {
5310 	int			done;
5311 	int			inprog;
5312 	struct task		tsk;
5313 	fhandle_t		fh;
5314 	off_t			off;
5315 	int			len;
5316 	struct nfsmount		*nmp;
5317 	struct ucred		*cred;
5318 	NFSPROC_T		*p;
5319 	struct mbuf		*m;
5320 	int			err;
5321 };
5322 
5323 static int
5324 nfsrv_writedsdorpc(struct nfsmount *nmp, fhandle_t *fhp, off_t off, int len,
5325     struct nfsvattr *nap, struct mbuf *m, struct ucred *cred, NFSPROC_T *p)
5326 {
5327 	uint32_t *tl;
5328 	struct nfsrv_descript *nd;
5329 	nfsattrbit_t attrbits;
5330 	nfsv4stateid_t st;
5331 	int commit, error, retlen;
5332 
5333 	nd = malloc(sizeof(*nd), M_TEMP, M_WAITOK | M_ZERO);
5334 	nfscl_reqstart(nd, NFSPROC_WRITE, nmp, (u_int8_t *)fhp,
5335 	    sizeof(fhandle_t), NULL, NULL, 0, 0);
5336 
5337 	/*
5338 	 * Use a stateid where other is an alternating 01010 pattern and
5339 	 * seqid is 0xffffffff.  This value is not defined as special by
5340 	 * the RFC and is used by the FreeBSD NFS server to indicate an
5341 	 * MDS->DS proxy operation.
5342 	 */
5343 	st.other[0] = 0x55555555;
5344 	st.other[1] = 0x55555555;
5345 	st.other[2] = 0x55555555;
5346 	st.seqid = 0xffffffff;
5347 	nfsm_stateidtom(nd, &st, NFSSTATEID_PUTSTATEID);
5348 	NFSM_BUILD(tl, u_int32_t *, NFSX_HYPER + 2 * NFSX_UNSIGNED);
5349 	txdr_hyper(off, tl);
5350 	tl += 2;
5351 	/*
5352 	 * Do all writes FileSync, since the server doesn't hold onto dirty
5353 	 * buffers.  Since clients should be accessing the DS servers directly
5354 	 * using the pNFS layouts, this just needs to work correctly as a
5355 	 * fallback.
5356 	 */
5357 	*tl++ = txdr_unsigned(NFSWRITE_FILESYNC);
5358 	*tl = txdr_unsigned(len);
5359 	NFSD_DEBUG(4, "nfsrv_writedsdorpc: len=%d\n", len);
5360 
5361 	/* Put data in mbuf chain. */
5362 	nd->nd_mb->m_next = m;
5363 
5364 	/* Set nd_mb and nd_bpos to end of data. */
5365 	while (m->m_next != NULL)
5366 		m = m->m_next;
5367 	nd->nd_mb = m;
5368 	nfsm_set(nd, m->m_len);
5369 	NFSD_DEBUG(4, "nfsrv_writedsdorpc: lastmb len=%d\n", m->m_len);
5370 
5371 	/* Do a Getattr for the attributes that change upon writing. */
5372 	NFSZERO_ATTRBIT(&attrbits);
5373 	NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_SIZE);
5374 	NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_CHANGE);
5375 	NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_TIMEACCESS);
5376 	NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_TIMEMODIFY);
5377 	NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_SPACEUSED);
5378 	NFSM_BUILD(tl, u_int32_t *, NFSX_UNSIGNED);
5379 	*tl = txdr_unsigned(NFSV4OP_GETATTR);
5380 	(void) nfsrv_putattrbit(nd, &attrbits);
5381 	error = newnfs_request(nd, nmp, NULL, &nmp->nm_sockreq, NULL, p,
5382 	    cred, NFS_PROG, NFS_VER4, NULL, 1, NULL, NULL);
5383 	if (error != 0) {
5384 		free(nd, M_TEMP);
5385 		return (error);
5386 	}
5387 	NFSD_DEBUG(4, "nfsrv_writedsdorpc: aft writerpc=%d\n", nd->nd_repstat);
5388 	/* Get rid of weak cache consistency data for now. */
5389 	if ((nd->nd_flag & (ND_NOMOREDATA | ND_NFSV4 | ND_V4WCCATTR)) ==
5390 	    (ND_NFSV4 | ND_V4WCCATTR)) {
5391 		error = nfsv4_loadattr(nd, NULL, nap, NULL, NULL, 0, NULL, NULL,
5392 		    NULL, NULL, NULL, 0, NULL, NULL, NULL, NULL, NULL);
5393 		NFSD_DEBUG(4, "nfsrv_writedsdorpc: wcc attr=%d\n", error);
5394 		if (error != 0)
5395 			goto nfsmout;
5396 		/*
5397 		 * Get rid of Op# and status for next op.
5398 		 */
5399 		NFSM_DISSECT(tl, uint32_t *, 2 * NFSX_UNSIGNED);
5400 		if (*++tl != 0)
5401 			nd->nd_flag |= ND_NOMOREDATA;
5402 	}
5403 	if (nd->nd_repstat == 0) {
5404 		NFSM_DISSECT(tl, uint32_t *, 2 * NFSX_UNSIGNED + NFSX_VERF);
5405 		retlen = fxdr_unsigned(int, *tl++);
5406 		commit = fxdr_unsigned(int, *tl);
5407 		if (commit != NFSWRITE_FILESYNC)
5408 			error = NFSERR_IO;
5409 		NFSD_DEBUG(4, "nfsrv_writedsdorpc:retlen=%d commit=%d err=%d\n",
5410 		    retlen, commit, error);
5411 	} else
5412 		error = nd->nd_repstat;
5413 	/* We have no use for the Write Verifier since we use FileSync. */
5414 
5415 	/*
5416 	 * Get the Change, Size, Access Time and Modify Time attributes and set
5417 	 * on the Metadata file, so its attributes will be what the file's
5418 	 * would be if it had been written.
5419 	 */
5420 	if (error == 0) {
5421 		NFSM_DISSECT(tl, uint32_t *, 2 * NFSX_UNSIGNED);
5422 		error = nfsv4_loadattr(nd, NULL, nap, NULL, NULL, 0, NULL, NULL,
5423 		    NULL, NULL, NULL, 0, NULL, NULL, NULL, NULL, NULL);
5424 	}
5425 	NFSD_DEBUG(4, "nfsrv_writedsdorpc: aft loadattr=%d\n", error);
5426 nfsmout:
5427 	m_freem(nd->nd_mrep);
5428 	free(nd, M_TEMP);
5429 	NFSD_DEBUG(4, "nfsrv_writedsdorpc error=%d\n", error);
5430 	return (error);
5431 }
5432 
5433 /*
5434  * Start up the thread that will execute nfsrv_writedsdorpc().
5435  */
5436 static void
5437 start_writedsdorpc(void *arg, int pending)
5438 {
5439 	struct nfsrvwritedsdorpc *drpc;
5440 
5441 	drpc = (struct nfsrvwritedsdorpc *)arg;
5442 	drpc->err = nfsrv_writedsdorpc(drpc->nmp, &drpc->fh, drpc->off,
5443 	    drpc->len, NULL, drpc->m, drpc->cred, drpc->p);
5444 	drpc->done = 1;
5445 	NFSD_DEBUG(4, "start_writedsdorpc: err=%d\n", drpc->err);
5446 }
5447 
5448 static int
5449 nfsrv_writedsrpc(fhandle_t *fhp, off_t off, int len, struct ucred *cred,
5450     NFSPROC_T *p, struct vnode *vp, struct nfsmount **nmpp, int mirrorcnt,
5451     struct mbuf **mpp, char *cp, int *failposp)
5452 {
5453 	struct nfsrvwritedsdorpc *drpc, *tdrpc = NULL;
5454 	struct nfsvattr na;
5455 	struct mbuf *m;
5456 	int error, i, offs, ret, timo;
5457 
5458 	NFSD_DEBUG(4, "in nfsrv_writedsrpc\n");
5459 	KASSERT(*mpp != NULL, ("nfsrv_writedsrpc: NULL mbuf chain"));
5460 	drpc = NULL;
5461 	if (mirrorcnt > 1)
5462 		tdrpc = drpc = malloc(sizeof(*drpc) * (mirrorcnt - 1), M_TEMP,
5463 		    M_WAITOK);
5464 
5465 	/* Calculate offset in mbuf chain that data starts. */
5466 	offs = cp - mtod(*mpp, char *);
5467 	NFSD_DEBUG(4, "nfsrv_writedsrpc: mcopy offs=%d len=%d\n", offs, len);
5468 
5469 	/*
5470 	 * Do the write RPC for every DS, using a separate kernel process
5471 	 * for every DS except the last one.
5472 	 */
5473 	error = 0;
5474 	for (i = 0; i < mirrorcnt - 1; i++, tdrpc++) {
5475 		tdrpc->done = 0;
5476 		NFSBCOPY(fhp, &tdrpc->fh, sizeof(*fhp));
5477 		tdrpc->off = off;
5478 		tdrpc->len = len;
5479 		tdrpc->nmp = *nmpp;
5480 		tdrpc->cred = cred;
5481 		tdrpc->p = p;
5482 		tdrpc->inprog = 0;
5483 		tdrpc->err = 0;
5484 		tdrpc->m = m_copym(*mpp, offs, NFSM_RNDUP(len), M_WAITOK);
5485 		ret = EIO;
5486 		if (nfs_pnfsiothreads != 0) {
5487 			ret = nfs_pnfsio(start_writedsdorpc, tdrpc);
5488 			NFSD_DEBUG(4, "nfsrv_writedsrpc: nfs_pnfsio=%d\n",
5489 			    ret);
5490 		}
5491 		if (ret != 0) {
5492 			ret = nfsrv_writedsdorpc(*nmpp, fhp, off, len, NULL,
5493 			    tdrpc->m, cred, p);
5494 			if (nfsds_failerr(ret) && *failposp == -1)
5495 				*failposp = i;
5496 			else if (error == 0 && ret != 0)
5497 				error = ret;
5498 		}
5499 		nmpp++;
5500 		fhp++;
5501 	}
5502 	m = m_copym(*mpp, offs, NFSM_RNDUP(len), M_WAITOK);
5503 	ret = nfsrv_writedsdorpc(*nmpp, fhp, off, len, &na, m, cred, p);
5504 	if (nfsds_failerr(ret) && *failposp == -1 && mirrorcnt > 1)
5505 		*failposp = mirrorcnt - 1;
5506 	else if (error == 0 && ret != 0)
5507 		error = ret;
5508 	if (error == 0)
5509 		error = nfsrv_setextattr(vp, &na, p);
5510 	NFSD_DEBUG(4, "nfsrv_writedsrpc: aft setextat=%d\n", error);
5511 	tdrpc = drpc;
5512 	timo = hz / 50;		/* Wait for 20msec. */
5513 	if (timo < 1)
5514 		timo = 1;
5515 	for (i = 0; i < mirrorcnt - 1; i++, tdrpc++) {
5516 		/* Wait for RPCs on separate threads to complete. */
5517 		while (tdrpc->inprog != 0 && tdrpc->done == 0)
5518 			tsleep(&tdrpc->tsk, PVFS, "srvwrds", timo);
5519 		if (nfsds_failerr(tdrpc->err) && *failposp == -1)
5520 			*failposp = i;
5521 		else if (error == 0 && tdrpc->err != 0)
5522 			error = tdrpc->err;
5523 	}
5524 	free(drpc, M_TEMP);
5525 	return (error);
5526 }
5527 
5528 /*
5529  * Do a allocate RPC on a DS data file, using this structure for the arguments,
5530  * so that this function can be executed by a separate kernel process.
5531  */
5532 struct nfsrvallocatedsdorpc {
5533 	int			done;
5534 	int			inprog;
5535 	struct task		tsk;
5536 	fhandle_t		fh;
5537 	off_t			off;
5538 	off_t			len;
5539 	struct nfsmount		*nmp;
5540 	struct ucred		*cred;
5541 	NFSPROC_T		*p;
5542 	int			err;
5543 };
5544 
5545 static int
5546 nfsrv_allocatedsdorpc(struct nfsmount *nmp, fhandle_t *fhp, off_t off,
5547     off_t len, struct nfsvattr *nap, struct ucred *cred, NFSPROC_T *p)
5548 {
5549 	uint32_t *tl;
5550 	struct nfsrv_descript *nd;
5551 	nfsattrbit_t attrbits;
5552 	nfsv4stateid_t st;
5553 	int error;
5554 
5555 	nd = malloc(sizeof(*nd), M_TEMP, M_WAITOK | M_ZERO);
5556 	nfscl_reqstart(nd, NFSPROC_ALLOCATE, nmp, (u_int8_t *)fhp,
5557 	    sizeof(fhandle_t), NULL, NULL, 0, 0);
5558 
5559 	/*
5560 	 * Use a stateid where other is an alternating 01010 pattern and
5561 	 * seqid is 0xffffffff.  This value is not defined as special by
5562 	 * the RFC and is used by the FreeBSD NFS server to indicate an
5563 	 * MDS->DS proxy operation.
5564 	 */
5565 	st.other[0] = 0x55555555;
5566 	st.other[1] = 0x55555555;
5567 	st.other[2] = 0x55555555;
5568 	st.seqid = 0xffffffff;
5569 	nfsm_stateidtom(nd, &st, NFSSTATEID_PUTSTATEID);
5570 	NFSM_BUILD(tl, uint32_t *, 2 * NFSX_HYPER + NFSX_UNSIGNED);
5571 	txdr_hyper(off, tl); tl += 2;
5572 	txdr_hyper(len, tl); tl += 2;
5573 	NFSD_DEBUG(4, "nfsrv_allocatedsdorpc: len=%jd\n", (intmax_t)len);
5574 
5575 	*tl = txdr_unsigned(NFSV4OP_GETATTR);
5576 	NFSGETATTR_ATTRBIT(&attrbits);
5577 	nfsrv_putattrbit(nd, &attrbits);
5578 	error = newnfs_request(nd, nmp, NULL, &nmp->nm_sockreq, NULL, p,
5579 	    cred, NFS_PROG, NFS_VER4, NULL, 1, NULL, NULL);
5580 	if (error != 0) {
5581 		free(nd, M_TEMP);
5582 		return (error);
5583 	}
5584 	NFSD_DEBUG(4, "nfsrv_allocatedsdorpc: aft allocaterpc=%d\n",
5585 	    nd->nd_repstat);
5586 	if (nd->nd_repstat == 0) {
5587 		NFSM_DISSECT(tl, uint32_t *, 2 * NFSX_UNSIGNED);
5588 		error = nfsv4_loadattr(nd, NULL, nap, NULL, NULL, 0, NULL, NULL,
5589 		    NULL, NULL, NULL, 0, NULL, NULL, NULL, NULL, NULL);
5590 	} else
5591 		error = nd->nd_repstat;
5592 	NFSD_DEBUG(4, "nfsrv_allocatedsdorpc: aft loadattr=%d\n", error);
5593 nfsmout:
5594 	m_freem(nd->nd_mrep);
5595 	free(nd, M_TEMP);
5596 	NFSD_DEBUG(4, "nfsrv_allocatedsdorpc error=%d\n", error);
5597 	return (error);
5598 }
5599 
5600 /*
5601  * Start up the thread that will execute nfsrv_allocatedsdorpc().
5602  */
5603 static void
5604 start_allocatedsdorpc(void *arg, int pending)
5605 {
5606 	struct nfsrvallocatedsdorpc *drpc;
5607 
5608 	drpc = (struct nfsrvallocatedsdorpc *)arg;
5609 	drpc->err = nfsrv_allocatedsdorpc(drpc->nmp, &drpc->fh, drpc->off,
5610 	    drpc->len, NULL, drpc->cred, drpc->p);
5611 	drpc->done = 1;
5612 	NFSD_DEBUG(4, "start_allocatedsdorpc: err=%d\n", drpc->err);
5613 }
5614 
5615 static int
5616 nfsrv_allocatedsrpc(fhandle_t *fhp, off_t off, off_t len, struct ucred *cred,
5617     NFSPROC_T *p, struct vnode *vp, struct nfsmount **nmpp, int mirrorcnt,
5618     int *failposp)
5619 {
5620 	struct nfsrvallocatedsdorpc *drpc, *tdrpc = NULL;
5621 	struct nfsvattr na;
5622 	int error, i, ret, timo;
5623 
5624 	NFSD_DEBUG(4, "in nfsrv_allocatedsrpc\n");
5625 	drpc = NULL;
5626 	if (mirrorcnt > 1)
5627 		tdrpc = drpc = malloc(sizeof(*drpc) * (mirrorcnt - 1), M_TEMP,
5628 		    M_WAITOK);
5629 
5630 	/*
5631 	 * Do the allocate RPC for every DS, using a separate kernel process
5632 	 * for every DS except the last one.
5633 	 */
5634 	error = 0;
5635 	for (i = 0; i < mirrorcnt - 1; i++, tdrpc++) {
5636 		tdrpc->done = 0;
5637 		NFSBCOPY(fhp, &tdrpc->fh, sizeof(*fhp));
5638 		tdrpc->off = off;
5639 		tdrpc->len = len;
5640 		tdrpc->nmp = *nmpp;
5641 		tdrpc->cred = cred;
5642 		tdrpc->p = p;
5643 		tdrpc->inprog = 0;
5644 		tdrpc->err = 0;
5645 		ret = EIO;
5646 		if (nfs_pnfsiothreads != 0) {
5647 			ret = nfs_pnfsio(start_allocatedsdorpc, tdrpc);
5648 			NFSD_DEBUG(4, "nfsrv_allocatedsrpc: nfs_pnfsio=%d\n",
5649 			    ret);
5650 		}
5651 		if (ret != 0) {
5652 			ret = nfsrv_allocatedsdorpc(*nmpp, fhp, off, len, NULL,
5653 			    cred, p);
5654 			if (nfsds_failerr(ret) && *failposp == -1)
5655 				*failposp = i;
5656 			else if (error == 0 && ret != 0)
5657 				error = ret;
5658 		}
5659 		nmpp++;
5660 		fhp++;
5661 	}
5662 	ret = nfsrv_allocatedsdorpc(*nmpp, fhp, off, len, &na, cred, p);
5663 	if (nfsds_failerr(ret) && *failposp == -1 && mirrorcnt > 1)
5664 		*failposp = mirrorcnt - 1;
5665 	else if (error == 0 && ret != 0)
5666 		error = ret;
5667 	if (error == 0)
5668 		error = nfsrv_setextattr(vp, &na, p);
5669 	NFSD_DEBUG(4, "nfsrv_allocatedsrpc: aft setextat=%d\n", error);
5670 	tdrpc = drpc;
5671 	timo = hz / 50;		/* Wait for 20msec. */
5672 	if (timo < 1)
5673 		timo = 1;
5674 	for (i = 0; i < mirrorcnt - 1; i++, tdrpc++) {
5675 		/* Wait for RPCs on separate threads to complete. */
5676 		while (tdrpc->inprog != 0 && tdrpc->done == 0)
5677 			tsleep(&tdrpc->tsk, PVFS, "srvalds", timo);
5678 		if (nfsds_failerr(tdrpc->err) && *failposp == -1)
5679 			*failposp = i;
5680 		else if (error == 0 && tdrpc->err != 0)
5681 			error = tdrpc->err;
5682 	}
5683 	free(drpc, M_TEMP);
5684 	return (error);
5685 }
5686 
5687 /*
5688  * Do a deallocate RPC on a DS data file, using this structure for the
5689  * arguments, so that this function can be executed by a separate kernel
5690  * process.
5691  */
5692 struct nfsrvdeallocatedsdorpc {
5693 	int			done;
5694 	int			inprog;
5695 	struct task		tsk;
5696 	fhandle_t		fh;
5697 	off_t			off;
5698 	off_t			len;
5699 	struct nfsmount		*nmp;
5700 	struct ucred		*cred;
5701 	NFSPROC_T		*p;
5702 	int			err;
5703 };
5704 
5705 static int
5706 nfsrv_deallocatedsdorpc(struct nfsmount *nmp, fhandle_t *fhp, off_t off,
5707     off_t len, struct nfsvattr *nap, struct ucred *cred, NFSPROC_T *p)
5708 {
5709 	uint32_t *tl;
5710 	struct nfsrv_descript *nd;
5711 	nfsattrbit_t attrbits;
5712 	nfsv4stateid_t st;
5713 	int error;
5714 
5715 	nd = malloc(sizeof(*nd), M_TEMP, M_WAITOK | M_ZERO);
5716 	nfscl_reqstart(nd, NFSPROC_DEALLOCATE, nmp, (u_int8_t *)fhp,
5717 	    sizeof(fhandle_t), NULL, NULL, 0, 0);
5718 
5719 	/*
5720 	 * Use a stateid where other is an alternating 01010 pattern and
5721 	 * seqid is 0xffffffff.  This value is not defined as special by
5722 	 * the RFC and is used by the FreeBSD NFS server to indicate an
5723 	 * MDS->DS proxy operation.
5724 	 */
5725 	st.other[0] = 0x55555555;
5726 	st.other[1] = 0x55555555;
5727 	st.other[2] = 0x55555555;
5728 	st.seqid = 0xffffffff;
5729 	nfsm_stateidtom(nd, &st, NFSSTATEID_PUTSTATEID);
5730 	NFSM_BUILD(tl, uint32_t *, 2 * NFSX_HYPER + NFSX_UNSIGNED);
5731 	txdr_hyper(off, tl); tl += 2;
5732 	txdr_hyper(len, tl); tl += 2;
5733 	NFSD_DEBUG(4, "nfsrv_deallocatedsdorpc: len=%jd\n", (intmax_t)len);
5734 
5735 	*tl = txdr_unsigned(NFSV4OP_GETATTR);
5736 	NFSGETATTR_ATTRBIT(&attrbits);
5737 	nfsrv_putattrbit(nd, &attrbits);
5738 	error = newnfs_request(nd, nmp, NULL, &nmp->nm_sockreq, NULL, p,
5739 	    cred, NFS_PROG, NFS_VER4, NULL, 1, NULL, NULL);
5740 	if (error != 0) {
5741 		free(nd, M_TEMP);
5742 		return (error);
5743 	}
5744 	NFSD_DEBUG(4, "nfsrv_deallocatedsdorpc: aft allocaterpc=%d\n",
5745 	    nd->nd_repstat);
5746 	if (nd->nd_repstat == 0) {
5747 		NFSM_DISSECT(tl, uint32_t *, 2 * NFSX_UNSIGNED);
5748 		error = nfsv4_loadattr(nd, NULL, nap, NULL, NULL, 0, NULL, NULL,
5749 		    NULL, NULL, NULL, 0, NULL, NULL, NULL, NULL, NULL);
5750 	} else
5751 		error = nd->nd_repstat;
5752 	NFSD_DEBUG(4, "nfsrv_deallocatedsdorpc: aft loadattr=%d\n", error);
5753 nfsmout:
5754 	m_freem(nd->nd_mrep);
5755 	free(nd, M_TEMP);
5756 	NFSD_DEBUG(4, "nfsrv_deallocatedsdorpc error=%d\n", error);
5757 	return (error);
5758 }
5759 
5760 /*
5761  * Start up the thread that will execute nfsrv_deallocatedsdorpc().
5762  */
5763 static void
5764 start_deallocatedsdorpc(void *arg, int pending)
5765 {
5766 	struct nfsrvdeallocatedsdorpc *drpc;
5767 
5768 	drpc = (struct nfsrvdeallocatedsdorpc *)arg;
5769 	drpc->err = nfsrv_deallocatedsdorpc(drpc->nmp, &drpc->fh, drpc->off,
5770 	    drpc->len, NULL, drpc->cred, drpc->p);
5771 	drpc->done = 1;
5772 	NFSD_DEBUG(4, "start_deallocatedsdorpc: err=%d\n", drpc->err);
5773 }
5774 
5775 static int
5776 nfsrv_deallocatedsrpc(fhandle_t *fhp, off_t off, off_t len, struct ucred *cred,
5777     NFSPROC_T *p, struct vnode *vp, struct nfsmount **nmpp, int mirrorcnt,
5778     int *failposp)
5779 {
5780 	struct nfsrvdeallocatedsdorpc *drpc, *tdrpc = NULL;
5781 	struct nfsvattr na;
5782 	int error, i, ret, timo;
5783 
5784 	NFSD_DEBUG(4, "in nfsrv_deallocatedsrpc\n");
5785 	drpc = NULL;
5786 	if (mirrorcnt > 1)
5787 		tdrpc = drpc = malloc(sizeof(*drpc) * (mirrorcnt - 1), M_TEMP,
5788 		    M_WAITOK);
5789 
5790 	/*
5791 	 * Do the deallocate RPC for every DS, using a separate kernel process
5792 	 * for every DS except the last one.
5793 	 */
5794 	error = 0;
5795 	for (i = 0; i < mirrorcnt - 1; i++, tdrpc++) {
5796 		tdrpc->done = 0;
5797 		NFSBCOPY(fhp, &tdrpc->fh, sizeof(*fhp));
5798 		tdrpc->off = off;
5799 		tdrpc->len = len;
5800 		tdrpc->nmp = *nmpp;
5801 		tdrpc->cred = cred;
5802 		tdrpc->p = p;
5803 		tdrpc->inprog = 0;
5804 		tdrpc->err = 0;
5805 		ret = EIO;
5806 		if (nfs_pnfsiothreads != 0) {
5807 			ret = nfs_pnfsio(start_deallocatedsdorpc, tdrpc);
5808 			NFSD_DEBUG(4, "nfsrv_deallocatedsrpc: nfs_pnfsio=%d\n",
5809 			    ret);
5810 		}
5811 		if (ret != 0) {
5812 			ret = nfsrv_deallocatedsdorpc(*nmpp, fhp, off, len,
5813 			    NULL, cred, p);
5814 			if (nfsds_failerr(ret) && *failposp == -1)
5815 				*failposp = i;
5816 			else if (error == 0 && ret != 0)
5817 				error = ret;
5818 		}
5819 		nmpp++;
5820 		fhp++;
5821 	}
5822 	ret = nfsrv_deallocatedsdorpc(*nmpp, fhp, off, len, &na, cred, p);
5823 	if (nfsds_failerr(ret) && *failposp == -1 && mirrorcnt > 1)
5824 		*failposp = mirrorcnt - 1;
5825 	else if (error == 0 && ret != 0)
5826 		error = ret;
5827 	if (error == 0)
5828 		error = nfsrv_setextattr(vp, &na, p);
5829 	NFSD_DEBUG(4, "nfsrv_deallocatedsrpc: aft setextat=%d\n", error);
5830 	tdrpc = drpc;
5831 	timo = hz / 50;		/* Wait for 20msec. */
5832 	if (timo < 1)
5833 		timo = 1;
5834 	for (i = 0; i < mirrorcnt - 1; i++, tdrpc++) {
5835 		/* Wait for RPCs on separate threads to complete. */
5836 		while (tdrpc->inprog != 0 && tdrpc->done == 0)
5837 			tsleep(&tdrpc->tsk, PVFS, "srvalds", timo);
5838 		if (nfsds_failerr(tdrpc->err) && *failposp == -1)
5839 			*failposp = i;
5840 		else if (error == 0 && tdrpc->err != 0)
5841 			error = tdrpc->err;
5842 	}
5843 	free(drpc, M_TEMP);
5844 	return (error);
5845 }
5846 
5847 static int
5848 nfsrv_setattrdsdorpc(fhandle_t *fhp, struct ucred *cred, NFSPROC_T *p,
5849     struct vnode *vp, struct nfsmount *nmp, struct nfsvattr *nap,
5850     struct nfsvattr *dsnap)
5851 {
5852 	uint32_t *tl;
5853 	struct nfsrv_descript *nd;
5854 	nfsv4stateid_t st;
5855 	nfsattrbit_t attrbits;
5856 	int error;
5857 
5858 	NFSD_DEBUG(4, "in nfsrv_setattrdsdorpc\n");
5859 	nd = malloc(sizeof(*nd), M_TEMP, M_WAITOK | M_ZERO);
5860 	/*
5861 	 * Use a stateid where other is an alternating 01010 pattern and
5862 	 * seqid is 0xffffffff.  This value is not defined as special by
5863 	 * the RFC and is used by the FreeBSD NFS server to indicate an
5864 	 * MDS->DS proxy operation.
5865 	 */
5866 	st.other[0] = 0x55555555;
5867 	st.other[1] = 0x55555555;
5868 	st.other[2] = 0x55555555;
5869 	st.seqid = 0xffffffff;
5870 	nfscl_reqstart(nd, NFSPROC_SETATTR, nmp, (u_int8_t *)fhp, sizeof(*fhp),
5871 	    NULL, NULL, 0, 0);
5872 	nfsm_stateidtom(nd, &st, NFSSTATEID_PUTSTATEID);
5873 	nfscl_fillsattr(nd, &nap->na_vattr, vp, NFSSATTR_FULL, 0);
5874 
5875 	/* Do a Getattr for the attributes that change due to writing. */
5876 	NFSZERO_ATTRBIT(&attrbits);
5877 	NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_SIZE);
5878 	NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_CHANGE);
5879 	NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_TIMEACCESS);
5880 	NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_TIMEMODIFY);
5881 	NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_SPACEUSED);
5882 	NFSM_BUILD(tl, u_int32_t *, NFSX_UNSIGNED);
5883 	*tl = txdr_unsigned(NFSV4OP_GETATTR);
5884 	(void) nfsrv_putattrbit(nd, &attrbits);
5885 	error = newnfs_request(nd, nmp, NULL, &nmp->nm_sockreq, NULL, p, cred,
5886 	    NFS_PROG, NFS_VER4, NULL, 1, NULL, NULL);
5887 	if (error != 0) {
5888 		free(nd, M_TEMP);
5889 		return (error);
5890 	}
5891 	NFSD_DEBUG(4, "nfsrv_setattrdsdorpc: aft setattrrpc=%d\n",
5892 	    nd->nd_repstat);
5893 	/* Get rid of weak cache consistency data for now. */
5894 	if ((nd->nd_flag & (ND_NOMOREDATA | ND_NFSV4 | ND_V4WCCATTR)) ==
5895 	    (ND_NFSV4 | ND_V4WCCATTR)) {
5896 		error = nfsv4_loadattr(nd, NULL, dsnap, NULL, NULL, 0, NULL,
5897 		    NULL, NULL, NULL, NULL, 0, NULL, NULL, NULL, NULL, NULL);
5898 		NFSD_DEBUG(4, "nfsrv_setattrdsdorpc: wcc attr=%d\n", error);
5899 		if (error != 0)
5900 			goto nfsmout;
5901 		/*
5902 		 * Get rid of Op# and status for next op.
5903 		 */
5904 		NFSM_DISSECT(tl, uint32_t *, 2 * NFSX_UNSIGNED);
5905 		if (*++tl != 0)
5906 			nd->nd_flag |= ND_NOMOREDATA;
5907 	}
5908 	error = nfsrv_getattrbits(nd, &attrbits, NULL, NULL);
5909 	if (error != 0)
5910 		goto nfsmout;
5911 	if (nd->nd_repstat != 0)
5912 		error = nd->nd_repstat;
5913 	/*
5914 	 * Get the Change, Size, Access Time and Modify Time attributes and set
5915 	 * on the Metadata file, so its attributes will be what the file's
5916 	 * would be if it had been written.
5917 	 */
5918 	if (error == 0) {
5919 		NFSM_DISSECT(tl, uint32_t *, 2 * NFSX_UNSIGNED);
5920 		error = nfsv4_loadattr(nd, NULL, dsnap, NULL, NULL, 0, NULL,
5921 		    NULL, NULL, NULL, NULL, 0, NULL, NULL, NULL, NULL, NULL);
5922 	}
5923 	NFSD_DEBUG(4, "nfsrv_setattrdsdorpc: aft setattr loadattr=%d\n", error);
5924 nfsmout:
5925 	m_freem(nd->nd_mrep);
5926 	free(nd, M_TEMP);
5927 	NFSD_DEBUG(4, "nfsrv_setattrdsdorpc error=%d\n", error);
5928 	return (error);
5929 }
5930 
5931 struct nfsrvsetattrdsdorpc {
5932 	int			done;
5933 	int			inprog;
5934 	struct task		tsk;
5935 	fhandle_t		fh;
5936 	struct nfsmount		*nmp;
5937 	struct vnode		*vp;
5938 	struct ucred		*cred;
5939 	NFSPROC_T		*p;
5940 	struct nfsvattr		na;
5941 	struct nfsvattr		dsna;
5942 	int			err;
5943 };
5944 
5945 /*
5946  * Start up the thread that will execute nfsrv_setattrdsdorpc().
5947  */
5948 static void
5949 start_setattrdsdorpc(void *arg, int pending)
5950 {
5951 	struct nfsrvsetattrdsdorpc *drpc;
5952 
5953 	drpc = (struct nfsrvsetattrdsdorpc *)arg;
5954 	drpc->err = nfsrv_setattrdsdorpc(&drpc->fh, drpc->cred, drpc->p,
5955 	    drpc->vp, drpc->nmp, &drpc->na, &drpc->dsna);
5956 	drpc->done = 1;
5957 }
5958 
5959 static int
5960 nfsrv_setattrdsrpc(fhandle_t *fhp, struct ucred *cred, NFSPROC_T *p,
5961     struct vnode *vp, struct nfsmount **nmpp, int mirrorcnt,
5962     struct nfsvattr *nap, int *failposp)
5963 {
5964 	struct nfsrvsetattrdsdorpc *drpc, *tdrpc = NULL;
5965 	struct nfsvattr na;
5966 	int error, i, ret, timo;
5967 
5968 	NFSD_DEBUG(4, "in nfsrv_setattrdsrpc\n");
5969 	drpc = NULL;
5970 	if (mirrorcnt > 1)
5971 		tdrpc = drpc = malloc(sizeof(*drpc) * (mirrorcnt - 1), M_TEMP,
5972 		    M_WAITOK);
5973 
5974 	/*
5975 	 * Do the setattr RPC for every DS, using a separate kernel process
5976 	 * for every DS except the last one.
5977 	 */
5978 	error = 0;
5979 	for (i = 0; i < mirrorcnt - 1; i++, tdrpc++) {
5980 		tdrpc->done = 0;
5981 		tdrpc->inprog = 0;
5982 		NFSBCOPY(fhp, &tdrpc->fh, sizeof(*fhp));
5983 		tdrpc->nmp = *nmpp;
5984 		tdrpc->vp = vp;
5985 		tdrpc->cred = cred;
5986 		tdrpc->p = p;
5987 		tdrpc->na = *nap;
5988 		tdrpc->err = 0;
5989 		ret = EIO;
5990 		if (nfs_pnfsiothreads != 0) {
5991 			ret = nfs_pnfsio(start_setattrdsdorpc, tdrpc);
5992 			NFSD_DEBUG(4, "nfsrv_setattrdsrpc: nfs_pnfsio=%d\n",
5993 			    ret);
5994 		}
5995 		if (ret != 0) {
5996 			ret = nfsrv_setattrdsdorpc(fhp, cred, p, vp, *nmpp, nap,
5997 			    &na);
5998 			if (nfsds_failerr(ret) && *failposp == -1)
5999 				*failposp = i;
6000 			else if (error == 0 && ret != 0)
6001 				error = ret;
6002 		}
6003 		nmpp++;
6004 		fhp++;
6005 	}
6006 	ret = nfsrv_setattrdsdorpc(fhp, cred, p, vp, *nmpp, nap, &na);
6007 	if (nfsds_failerr(ret) && *failposp == -1 && mirrorcnt > 1)
6008 		*failposp = mirrorcnt - 1;
6009 	else if (error == 0 && ret != 0)
6010 		error = ret;
6011 	if (error == 0)
6012 		error = nfsrv_setextattr(vp, &na, p);
6013 	NFSD_DEBUG(4, "nfsrv_setattrdsrpc: aft setextat=%d\n", error);
6014 	tdrpc = drpc;
6015 	timo = hz / 50;		/* Wait for 20msec. */
6016 	if (timo < 1)
6017 		timo = 1;
6018 	for (i = 0; i < mirrorcnt - 1; i++, tdrpc++) {
6019 		/* Wait for RPCs on separate threads to complete. */
6020 		while (tdrpc->inprog != 0 && tdrpc->done == 0)
6021 			tsleep(&tdrpc->tsk, PVFS, "srvsads", timo);
6022 		if (nfsds_failerr(tdrpc->err) && *failposp == -1)
6023 			*failposp = i;
6024 		else if (error == 0 && tdrpc->err != 0)
6025 			error = tdrpc->err;
6026 	}
6027 	free(drpc, M_TEMP);
6028 	return (error);
6029 }
6030 
6031 /*
6032  * Do a Setattr of an NFSv4 ACL on the DS file.
6033  */
6034 static int
6035 nfsrv_setacldsdorpc(fhandle_t *fhp, struct ucred *cred, NFSPROC_T *p,
6036     struct vnode *vp, struct nfsmount *nmp, struct acl *aclp)
6037 {
6038 	struct nfsrv_descript *nd;
6039 	nfsv4stateid_t st;
6040 	nfsattrbit_t attrbits;
6041 	int error;
6042 
6043 	NFSD_DEBUG(4, "in nfsrv_setacldsdorpc\n");
6044 	nd = malloc(sizeof(*nd), M_TEMP, M_WAITOK | M_ZERO);
6045 	/*
6046 	 * Use a stateid where other is an alternating 01010 pattern and
6047 	 * seqid is 0xffffffff.  This value is not defined as special by
6048 	 * the RFC and is used by the FreeBSD NFS server to indicate an
6049 	 * MDS->DS proxy operation.
6050 	 */
6051 	st.other[0] = 0x55555555;
6052 	st.other[1] = 0x55555555;
6053 	st.other[2] = 0x55555555;
6054 	st.seqid = 0xffffffff;
6055 	nfscl_reqstart(nd, NFSPROC_SETACL, nmp, (u_int8_t *)fhp, sizeof(*fhp),
6056 	    NULL, NULL, 0, 0);
6057 	nfsm_stateidtom(nd, &st, NFSSTATEID_PUTSTATEID);
6058 	NFSZERO_ATTRBIT(&attrbits);
6059 	NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_ACL);
6060 	/*
6061 	 * The "vp" argument to nfsv4_fillattr() is only used for vnode_type(),
6062 	 * so passing in the metadata "vp" will be ok, since it is of
6063 	 * the same type (VREG).
6064 	 */
6065 	nfsv4_fillattr(nd, NULL, vp, aclp, NULL, NULL, 0, &attrbits, NULL,
6066 	    NULL, 0, 0, 0, 0, 0, NULL);
6067 	error = newnfs_request(nd, nmp, NULL, &nmp->nm_sockreq, NULL, p, cred,
6068 	    NFS_PROG, NFS_VER4, NULL, 1, NULL, NULL);
6069 	if (error != 0) {
6070 		free(nd, M_TEMP);
6071 		return (error);
6072 	}
6073 	NFSD_DEBUG(4, "nfsrv_setacldsdorpc: aft setaclrpc=%d\n",
6074 	    nd->nd_repstat);
6075 	error = nd->nd_repstat;
6076 	m_freem(nd->nd_mrep);
6077 	free(nd, M_TEMP);
6078 	return (error);
6079 }
6080 
6081 struct nfsrvsetacldsdorpc {
6082 	int			done;
6083 	int			inprog;
6084 	struct task		tsk;
6085 	fhandle_t		fh;
6086 	struct nfsmount		*nmp;
6087 	struct vnode		*vp;
6088 	struct ucred		*cred;
6089 	NFSPROC_T		*p;
6090 	struct acl		*aclp;
6091 	int			err;
6092 };
6093 
6094 /*
6095  * Start up the thread that will execute nfsrv_setacldsdorpc().
6096  */
6097 static void
6098 start_setacldsdorpc(void *arg, int pending)
6099 {
6100 	struct nfsrvsetacldsdorpc *drpc;
6101 
6102 	drpc = (struct nfsrvsetacldsdorpc *)arg;
6103 	drpc->err = nfsrv_setacldsdorpc(&drpc->fh, drpc->cred, drpc->p,
6104 	    drpc->vp, drpc->nmp, drpc->aclp);
6105 	drpc->done = 1;
6106 }
6107 
6108 static int
6109 nfsrv_setacldsrpc(fhandle_t *fhp, struct ucred *cred, NFSPROC_T *p,
6110     struct vnode *vp, struct nfsmount **nmpp, int mirrorcnt, struct acl *aclp,
6111     int *failposp)
6112 {
6113 	struct nfsrvsetacldsdorpc *drpc, *tdrpc = NULL;
6114 	int error, i, ret, timo;
6115 
6116 	NFSD_DEBUG(4, "in nfsrv_setacldsrpc\n");
6117 	drpc = NULL;
6118 	if (mirrorcnt > 1)
6119 		tdrpc = drpc = malloc(sizeof(*drpc) * (mirrorcnt - 1), M_TEMP,
6120 		    M_WAITOK);
6121 
6122 	/*
6123 	 * Do the setattr RPC for every DS, using a separate kernel process
6124 	 * for every DS except the last one.
6125 	 */
6126 	error = 0;
6127 	for (i = 0; i < mirrorcnt - 1; i++, tdrpc++) {
6128 		tdrpc->done = 0;
6129 		tdrpc->inprog = 0;
6130 		NFSBCOPY(fhp, &tdrpc->fh, sizeof(*fhp));
6131 		tdrpc->nmp = *nmpp;
6132 		tdrpc->vp = vp;
6133 		tdrpc->cred = cred;
6134 		tdrpc->p = p;
6135 		tdrpc->aclp = aclp;
6136 		tdrpc->err = 0;
6137 		ret = EIO;
6138 		if (nfs_pnfsiothreads != 0) {
6139 			ret = nfs_pnfsio(start_setacldsdorpc, tdrpc);
6140 			NFSD_DEBUG(4, "nfsrv_setacldsrpc: nfs_pnfsio=%d\n",
6141 			    ret);
6142 		}
6143 		if (ret != 0) {
6144 			ret = nfsrv_setacldsdorpc(fhp, cred, p, vp, *nmpp,
6145 			    aclp);
6146 			if (nfsds_failerr(ret) && *failposp == -1)
6147 				*failposp = i;
6148 			else if (error == 0 && ret != 0)
6149 				error = ret;
6150 		}
6151 		nmpp++;
6152 		fhp++;
6153 	}
6154 	ret = nfsrv_setacldsdorpc(fhp, cred, p, vp, *nmpp, aclp);
6155 	if (nfsds_failerr(ret) && *failposp == -1 && mirrorcnt > 1)
6156 		*failposp = mirrorcnt - 1;
6157 	else if (error == 0 && ret != 0)
6158 		error = ret;
6159 	NFSD_DEBUG(4, "nfsrv_setacldsrpc: aft setextat=%d\n", error);
6160 	tdrpc = drpc;
6161 	timo = hz / 50;		/* Wait for 20msec. */
6162 	if (timo < 1)
6163 		timo = 1;
6164 	for (i = 0; i < mirrorcnt - 1; i++, tdrpc++) {
6165 		/* Wait for RPCs on separate threads to complete. */
6166 		while (tdrpc->inprog != 0 && tdrpc->done == 0)
6167 			tsleep(&tdrpc->tsk, PVFS, "srvacds", timo);
6168 		if (nfsds_failerr(tdrpc->err) && *failposp == -1)
6169 			*failposp = i;
6170 		else if (error == 0 && tdrpc->err != 0)
6171 			error = tdrpc->err;
6172 	}
6173 	free(drpc, M_TEMP);
6174 	return (error);
6175 }
6176 
6177 /*
6178  * Getattr call to the DS for the attributes that change due to writing.
6179  */
6180 static int
6181 nfsrv_getattrdsrpc(fhandle_t *fhp, struct ucred *cred, NFSPROC_T *p,
6182     struct vnode *vp, struct nfsmount *nmp, struct nfsvattr *nap)
6183 {
6184 	struct nfsrv_descript *nd;
6185 	int error;
6186 	nfsattrbit_t attrbits;
6187 
6188 	NFSD_DEBUG(4, "in nfsrv_getattrdsrpc\n");
6189 	nd = malloc(sizeof(*nd), M_TEMP, M_WAITOK | M_ZERO);
6190 	nfscl_reqstart(nd, NFSPROC_GETATTR, nmp, (u_int8_t *)fhp,
6191 	    sizeof(fhandle_t), NULL, NULL, 0, 0);
6192 	NFSZERO_ATTRBIT(&attrbits);
6193 	NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_SIZE);
6194 	NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_CHANGE);
6195 	NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_TIMEACCESS);
6196 	NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_TIMEMODIFY);
6197 	NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_SPACEUSED);
6198 	(void) nfsrv_putattrbit(nd, &attrbits);
6199 	error = newnfs_request(nd, nmp, NULL, &nmp->nm_sockreq, NULL, p, cred,
6200 	    NFS_PROG, NFS_VER4, NULL, 1, NULL, NULL);
6201 	if (error != 0) {
6202 		free(nd, M_TEMP);
6203 		return (error);
6204 	}
6205 	NFSD_DEBUG(4, "nfsrv_getattrdsrpc: aft getattrrpc=%d\n",
6206 	    nd->nd_repstat);
6207 	if (nd->nd_repstat == 0) {
6208 		error = nfsv4_loadattr(nd, NULL, nap, NULL, NULL, 0,
6209 		    NULL, NULL, NULL, NULL, NULL, 0, NULL, NULL, NULL,
6210 		    NULL, NULL);
6211 		/*
6212 		 * We can only save the updated values in the extended
6213 		 * attribute if the vp is exclusively locked.
6214 		 * This should happen when any of the following operations
6215 		 * occur on the vnode:
6216 		 *    Close, Delegreturn, LayoutCommit, LayoutReturn
6217 		 * As such, the updated extended attribute should get saved
6218 		 * before nfsrv_checkdsattr() returns 0 and allows the cached
6219 		 * attributes to be returned without calling this function.
6220 		 */
6221 		if (error == 0 && VOP_ISLOCKED(vp) == LK_EXCLUSIVE) {
6222 			error = nfsrv_setextattr(vp, nap, p);
6223 			NFSD_DEBUG(4, "nfsrv_getattrdsrpc: aft setextat=%d\n",
6224 			    error);
6225 		}
6226 	} else
6227 		error = nd->nd_repstat;
6228 	m_freem(nd->nd_mrep);
6229 	free(nd, M_TEMP);
6230 	NFSD_DEBUG(4, "nfsrv_getattrdsrpc error=%d\n", error);
6231 	return (error);
6232 }
6233 
6234 /*
6235  * Seek call to a DS.
6236  */
6237 static int
6238 nfsrv_seekdsrpc(fhandle_t *fhp, off_t *offp, int content, bool *eofp,
6239     struct ucred *cred, NFSPROC_T *p, struct nfsmount *nmp)
6240 {
6241 	uint32_t *tl;
6242 	struct nfsrv_descript *nd;
6243 	nfsv4stateid_t st;
6244 	int error;
6245 
6246 	NFSD_DEBUG(4, "in nfsrv_seekdsrpc\n");
6247 	/*
6248 	 * Use a stateid where other is an alternating 01010 pattern and
6249 	 * seqid is 0xffffffff.  This value is not defined as special by
6250 	 * the RFC and is used by the FreeBSD NFS server to indicate an
6251 	 * MDS->DS proxy operation.
6252 	 */
6253 	st.other[0] = 0x55555555;
6254 	st.other[1] = 0x55555555;
6255 	st.other[2] = 0x55555555;
6256 	st.seqid = 0xffffffff;
6257 	nd = malloc(sizeof(*nd), M_TEMP, M_WAITOK | M_ZERO);
6258 	nfscl_reqstart(nd, NFSPROC_SEEKDS, nmp, (u_int8_t *)fhp,
6259 	    sizeof(fhandle_t), NULL, NULL, 0, 0);
6260 	nfsm_stateidtom(nd, &st, NFSSTATEID_PUTSTATEID);
6261 	NFSM_BUILD(tl, uint32_t *, NFSX_HYPER + NFSX_UNSIGNED);
6262 	txdr_hyper(*offp, tl); tl += 2;
6263 	*tl = txdr_unsigned(content);
6264 	error = newnfs_request(nd, nmp, NULL, &nmp->nm_sockreq, NULL, p, cred,
6265 	    NFS_PROG, NFS_VER4, NULL, 1, NULL, NULL);
6266 	if (error != 0) {
6267 		free(nd, M_TEMP);
6268 		return (error);
6269 	}
6270 	NFSD_DEBUG(4, "nfsrv_seekdsrpc: aft seekrpc=%d\n", nd->nd_repstat);
6271 	if (nd->nd_repstat == 0) {
6272 		NFSM_DISSECT(tl, uint32_t *, NFSX_UNSIGNED + NFSX_HYPER);
6273 		if (*tl++ == newnfs_true)
6274 			*eofp = true;
6275 		else
6276 			*eofp = false;
6277 		*offp = fxdr_hyper(tl);
6278 	} else
6279 		error = nd->nd_repstat;
6280 nfsmout:
6281 	m_freem(nd->nd_mrep);
6282 	free(nd, M_TEMP);
6283 	NFSD_DEBUG(4, "nfsrv_seekdsrpc error=%d\n", error);
6284 	return (error);
6285 }
6286 
6287 /*
6288  * Get the device id and file handle for a DS file.
6289  */
6290 int
6291 nfsrv_dsgetdevandfh(struct vnode *vp, NFSPROC_T *p, int *mirrorcntp,
6292     fhandle_t *fhp, char *devid)
6293 {
6294 	int buflen, error;
6295 	char *buf;
6296 
6297 	buflen = 1024;
6298 	buf = malloc(buflen, M_TEMP, M_WAITOK);
6299 	error = nfsrv_dsgetsockmnt(vp, 0, buf, &buflen, mirrorcntp, p, NULL,
6300 	    fhp, devid, NULL, NULL, NULL, NULL, NULL, NULL);
6301 	free(buf, M_TEMP);
6302 	return (error);
6303 }
6304 
6305 /*
6306  * Do a Lookup against the DS for the filename.
6307  */
6308 static int
6309 nfsrv_pnfslookupds(struct vnode *vp, struct vnode *dvp, struct pnfsdsfile *pf,
6310     struct vnode **nvpp, NFSPROC_T *p)
6311 {
6312 	struct nameidata named;
6313 	struct ucred *tcred;
6314 	char *bufp;
6315 	u_long *hashp;
6316 	struct vnode *nvp;
6317 	int error;
6318 
6319 	tcred = newnfs_getcred();
6320 	named.ni_cnd.cn_nameiop = LOOKUP;
6321 	named.ni_cnd.cn_lkflags = LK_SHARED | LK_RETRY;
6322 	named.ni_cnd.cn_cred = tcred;
6323 	named.ni_cnd.cn_thread = p;
6324 	named.ni_cnd.cn_flags = ISLASTCN | LOCKPARENT | LOCKLEAF | SAVENAME;
6325 	nfsvno_setpathbuf(&named, &bufp, &hashp);
6326 	named.ni_cnd.cn_nameptr = bufp;
6327 	named.ni_cnd.cn_namelen = strlen(pf->dsf_filename);
6328 	strlcpy(bufp, pf->dsf_filename, NAME_MAX);
6329 	NFSD_DEBUG(4, "nfsrv_pnfslookupds: filename=%s\n", bufp);
6330 	error = VOP_LOOKUP(dvp, &nvp, &named.ni_cnd);
6331 	NFSD_DEBUG(4, "nfsrv_pnfslookupds: aft LOOKUP=%d\n", error);
6332 	NFSFREECRED(tcred);
6333 	nfsvno_relpathbuf(&named);
6334 	if (error == 0)
6335 		*nvpp = nvp;
6336 	NFSD_DEBUG(4, "eo nfsrv_pnfslookupds=%d\n", error);
6337 	return (error);
6338 }
6339 
6340 /*
6341  * Set the file handle to the correct one.
6342  */
6343 static void
6344 nfsrv_pnfssetfh(struct vnode *vp, struct pnfsdsfile *pf, char *devid,
6345     char *fnamep, struct vnode *nvp, NFSPROC_T *p)
6346 {
6347 	struct nfsnode *np;
6348 	int ret = 0;
6349 
6350 	np = VTONFS(nvp);
6351 	NFSBCOPY(np->n_fhp->nfh_fh, &pf->dsf_fh, NFSX_MYFH);
6352 	/*
6353 	 * We can only do a vn_set_extattr() if the vnode is exclusively
6354 	 * locked and vn_start_write() has been done.  If devid != NULL or
6355 	 * fnamep != NULL or the vnode is shared locked, vn_start_write()
6356 	 * may not have been done.
6357 	 * If not done now, it will be done on a future call.
6358 	 */
6359 	if (devid == NULL && fnamep == NULL && NFSVOPISLOCKED(vp) ==
6360 	    LK_EXCLUSIVE)
6361 		ret = vn_extattr_set(vp, IO_NODELOCKED,
6362 		    EXTATTR_NAMESPACE_SYSTEM, "pnfsd.dsfile", sizeof(*pf),
6363 		    (char *)pf, p);
6364 	NFSD_DEBUG(4, "eo nfsrv_pnfssetfh=%d\n", ret);
6365 }
6366 
6367 /*
6368  * Cause RPCs waiting on "nmp" to fail.  This is called for a DS mount point
6369  * when the DS has failed.
6370  */
6371 void
6372 nfsrv_killrpcs(struct nfsmount *nmp)
6373 {
6374 
6375 	/*
6376 	 * Call newnfs_nmcancelreqs() to cause
6377 	 * any RPCs in progress on the mount point to
6378 	 * fail.
6379 	 * This will cause any process waiting for an
6380 	 * RPC to complete while holding a vnode lock
6381 	 * on the mounted-on vnode (such as "df" or
6382 	 * a non-forced "umount") to fail.
6383 	 * This will unlock the mounted-on vnode so
6384 	 * a forced dismount can succeed.
6385 	 * The NFSMNTP_CANCELRPCS flag should be set when this function is
6386 	 * called.
6387 	 */
6388 	newnfs_nmcancelreqs(nmp);
6389 }
6390 
6391 /*
6392  * Sum up the statfs info for each of the DSs, so that the client will
6393  * receive the total for all DSs.
6394  */
6395 static int
6396 nfsrv_pnfsstatfs(struct statfs *sf, struct mount *mp)
6397 {
6398 	struct statfs *tsf;
6399 	struct nfsdevice *ds;
6400 	struct vnode **dvpp, **tdvpp, *dvp;
6401 	uint64_t tot;
6402 	int cnt, error = 0, i;
6403 
6404 	if (nfsrv_devidcnt <= 0)
6405 		return (ENXIO);
6406 	dvpp = mallocarray(nfsrv_devidcnt, sizeof(*dvpp), M_TEMP, M_WAITOK);
6407 	tsf = malloc(sizeof(*tsf), M_TEMP, M_WAITOK);
6408 
6409 	/* Get an array of the dvps for the DSs. */
6410 	tdvpp = dvpp;
6411 	i = 0;
6412 	NFSDDSLOCK();
6413 	/* First, search for matches for same file system. */
6414 	TAILQ_FOREACH(ds, &nfsrv_devidhead, nfsdev_list) {
6415 		if (ds->nfsdev_nmp != NULL && ds->nfsdev_mdsisset != 0 &&
6416 		    fsidcmp(&ds->nfsdev_mdsfsid, &mp->mnt_stat.f_fsid) == 0) {
6417 			if (++i > nfsrv_devidcnt)
6418 				break;
6419 			*tdvpp++ = ds->nfsdev_dvp;
6420 		}
6421 	}
6422 	/*
6423 	 * If no matches for same file system, total all servers not assigned
6424 	 * to a file system.
6425 	 */
6426 	if (i == 0) {
6427 		TAILQ_FOREACH(ds, &nfsrv_devidhead, nfsdev_list) {
6428 			if (ds->nfsdev_nmp != NULL &&
6429 			    ds->nfsdev_mdsisset == 0) {
6430 				if (++i > nfsrv_devidcnt)
6431 					break;
6432 				*tdvpp++ = ds->nfsdev_dvp;
6433 			}
6434 		}
6435 	}
6436 	NFSDDSUNLOCK();
6437 	cnt = i;
6438 
6439 	/* Do a VFS_STATFS() for each of the DSs and sum them up. */
6440 	tdvpp = dvpp;
6441 	for (i = 0; i < cnt && error == 0; i++) {
6442 		dvp = *tdvpp++;
6443 		error = VFS_STATFS(dvp->v_mount, tsf);
6444 		if (error == 0) {
6445 			if (sf->f_bsize == 0) {
6446 				if (tsf->f_bsize > 0)
6447 					sf->f_bsize = tsf->f_bsize;
6448 				else
6449 					sf->f_bsize = 8192;
6450 			}
6451 			if (tsf->f_blocks > 0) {
6452 				if (sf->f_bsize != tsf->f_bsize) {
6453 					tot = tsf->f_blocks * tsf->f_bsize;
6454 					sf->f_blocks += (tot / sf->f_bsize);
6455 				} else
6456 					sf->f_blocks += tsf->f_blocks;
6457 			}
6458 			if (tsf->f_bfree > 0) {
6459 				if (sf->f_bsize != tsf->f_bsize) {
6460 					tot = tsf->f_bfree * tsf->f_bsize;
6461 					sf->f_bfree += (tot / sf->f_bsize);
6462 				} else
6463 					sf->f_bfree += tsf->f_bfree;
6464 			}
6465 			if (tsf->f_bavail > 0) {
6466 				if (sf->f_bsize != tsf->f_bsize) {
6467 					tot = tsf->f_bavail * tsf->f_bsize;
6468 					sf->f_bavail += (tot / sf->f_bsize);
6469 				} else
6470 					sf->f_bavail += tsf->f_bavail;
6471 			}
6472 		}
6473 	}
6474 	free(tsf, M_TEMP);
6475 	free(dvpp, M_TEMP);
6476 	return (error);
6477 }
6478 
6479 /*
6480  * Set an NFSv4 acl.
6481  */
6482 int
6483 nfsrv_setacl(struct vnode *vp, NFSACL_T *aclp, struct ucred *cred, NFSPROC_T *p)
6484 {
6485 	int error;
6486 
6487 	if (nfsrv_useacl == 0 || nfs_supportsnfsv4acls(vp) == 0) {
6488 		error = NFSERR_ATTRNOTSUPP;
6489 		goto out;
6490 	}
6491 	/*
6492 	 * With NFSv4 ACLs, chmod(2) may need to add additional entries.
6493 	 * Make sure it has enough room for that - splitting every entry
6494 	 * into two and appending "canonical six" entries at the end.
6495 	 * Cribbed out of kern/vfs_acl.c - Rick M.
6496 	 */
6497 	if (aclp->acl_cnt > (ACL_MAX_ENTRIES - 6) / 2) {
6498 		error = NFSERR_ATTRNOTSUPP;
6499 		goto out;
6500 	}
6501 	error = VOP_SETACL(vp, ACL_TYPE_NFS4, aclp, cred, p);
6502 	if (error == 0) {
6503 		error = nfsrv_dssetacl(vp, aclp, cred, p);
6504 		if (error == ENOENT)
6505 			error = 0;
6506 	}
6507 
6508 out:
6509 	NFSEXITCODE(error);
6510 	return (error);
6511 }
6512 
6513 /*
6514  * Seek vnode op call (actually it is a VOP_IOCTL()).
6515  * This function is called with the vnode locked, but unlocks and vrele()s
6516  * the vp before returning.
6517  */
6518 int
6519 nfsvno_seek(struct nfsrv_descript *nd, struct vnode *vp, u_long cmd,
6520     off_t *offp, int content, bool *eofp, struct ucred *cred, NFSPROC_T *p)
6521 {
6522 	struct nfsvattr at;
6523 	int error, ret;
6524 
6525 	ASSERT_VOP_LOCKED(vp, "nfsvno_seek vp");
6526 	/*
6527 	 * Attempt to seek on a DS file. A return of ENOENT implies
6528 	 * there is no DS file to seek on.
6529 	 */
6530 	error = nfsrv_proxyds(vp, 0, 0, cred, p, NFSPROC_SEEKDS, NULL,
6531 	    NULL, NULL, NULL, NULL, offp, content, eofp);
6532 	if (error != ENOENT) {
6533 		vput(vp);
6534 		return (error);
6535 	}
6536 
6537 	/*
6538 	 * Do the VOP_IOCTL() call.  For the case where *offp == file_size,
6539 	 * VOP_IOCTL() will return ENXIO.  However, the correct reply for
6540 	 * NFSv4.2 is *eofp == true and error == 0 for this case.
6541 	 */
6542 	NFSVOPUNLOCK(vp);
6543 	error = VOP_IOCTL(vp, cmd, offp, 0, cred, p);
6544 	*eofp = false;
6545 	if (error == ENXIO || (error == 0 && cmd == FIOSEEKHOLE)) {
6546 		/* Handle the cases where we might be at EOF. */
6547 		ret = nfsvno_getattr(vp, &at, nd, p, 0, NULL);
6548 		if (ret == 0 && *offp == at.na_size) {
6549 			*eofp = true;
6550 			error = 0;
6551 		}
6552 		if (ret != 0 && error == 0)
6553 			error = ret;
6554 	}
6555 	vrele(vp);
6556 	NFSEXITCODE(error);
6557 	return (error);
6558 }
6559 
6560 /*
6561  * Allocate vnode op call.
6562  */
6563 int
6564 nfsvno_allocate(struct vnode *vp, off_t off, off_t len, struct ucred *cred,
6565     NFSPROC_T *p)
6566 {
6567 	int error;
6568 	off_t olen;
6569 
6570 	ASSERT_VOP_ELOCKED(vp, "nfsvno_allocate vp");
6571 	/*
6572 	 * Attempt to allocate on a DS file. A return of ENOENT implies
6573 	 * there is no DS file to allocate on.
6574 	 */
6575 	error = nfsrv_proxyds(vp, off, 0, cred, p, NFSPROC_ALLOCATE, NULL,
6576 	    NULL, NULL, NULL, NULL, &len, 0, NULL);
6577 	if (error != ENOENT)
6578 		return (error);
6579 
6580 	/*
6581 	 * Do the actual VOP_ALLOCATE(), looping so long as
6582 	 * progress is being made, to achieve completion.
6583 	 */
6584 	do {
6585 		olen = len;
6586 		error = VOP_ALLOCATE(vp, &off, &len);
6587 		if (error == 0 && len > 0 && olen > len)
6588 			maybe_yield();
6589 	} while (error == 0 && len > 0 && olen > len);
6590 	if (error == 0 && len > 0)
6591 		error = NFSERR_IO;
6592 	NFSEXITCODE(error);
6593 	return (error);
6594 }
6595 
6596 /*
6597  * Deallocate vnode op call.
6598  */
6599 int
6600 nfsvno_deallocate(struct vnode *vp, off_t off, off_t len, struct ucred *cred,
6601     NFSPROC_T *p)
6602 {
6603 	int error;
6604 	off_t olen;
6605 
6606 	ASSERT_VOP_ELOCKED(vp, "nfsvno_deallocate vp");
6607 	/*
6608 	 * Attempt to deallocate on a DS file. A return of ENOENT implies
6609 	 * there is no DS file to deallocate on.
6610 	 */
6611 	error = nfsrv_proxyds(vp, off, 0, cred, p, NFSPROC_DEALLOCATE, NULL,
6612 	    NULL, NULL, NULL, NULL, &len, 0, NULL);
6613 	if (error != ENOENT)
6614 		return (error);
6615 
6616 	/*
6617 	 * Do the actual VOP_DEALLOCATE(), looping so long as
6618 	 * progress is being made, to achieve completion.
6619 	 */
6620 	do {
6621 		olen = len;
6622 		error = VOP_DEALLOCATE(vp, &off, &len, 0, IO_SYNC, cred);
6623 		if (error == 0 && len > 0 && olen > len)
6624 			maybe_yield();
6625 	} while (error == 0 && len > 0 && olen > len);
6626 	if (error == 0 && len > 0)
6627 		error = NFSERR_IO;
6628 	NFSEXITCODE(error);
6629 	return (error);
6630 }
6631 
6632 /*
6633  * Get Extended Atribute vnode op into an mbuf list.
6634  */
6635 int
6636 nfsvno_getxattr(struct vnode *vp, char *name, uint32_t maxresp,
6637     struct ucred *cred, uint64_t flag, int maxextsiz, struct thread *p,
6638     struct mbuf **mpp, struct mbuf **mpendp, int *lenp)
6639 {
6640 	struct iovec *iv;
6641 	struct uio io, *uiop = &io;
6642 	struct mbuf *m, *m2;
6643 	int alen, error, len, tlen;
6644 	size_t siz;
6645 
6646 	/* First, find out the size of the extended attribute. */
6647 	error = VOP_GETEXTATTR(vp, EXTATTR_NAMESPACE_USER, name, NULL,
6648 	    &siz, cred, p);
6649 	if (error != 0)
6650 		return (NFSERR_NOXATTR);
6651 	if (siz > maxresp - NFS_MAXXDR)
6652 		return (NFSERR_XATTR2BIG);
6653 	len = siz;
6654 	tlen = NFSM_RNDUP(len);
6655 	if (tlen > 0) {
6656 		/*
6657 		 * If cnt > MCLBYTES and the reply will not be saved, use
6658 		 * ext_pgs mbufs for TLS.
6659 		 * For NFSv4.0, we do not know for sure if the reply will
6660 		 * be saved, so do not use ext_pgs mbufs for NFSv4.0.
6661 		 * Always use ext_pgs mbufs if ND_EXTPG is set.
6662 		 */
6663 		if ((flag & ND_EXTPG) != 0 || (tlen > MCLBYTES &&
6664 		    (flag & (ND_TLS | ND_SAVEREPLY)) == ND_TLS &&
6665 		    (flag & (ND_NFSV4 | ND_NFSV41)) != ND_NFSV4))
6666 			uiop->uio_iovcnt = nfsrv_createiovec_extpgs(tlen,
6667 			    maxextsiz, &m, &m2, &iv);
6668 		else
6669 			uiop->uio_iovcnt = nfsrv_createiovec(tlen, &m, &m2,
6670 			    &iv);
6671 		uiop->uio_iov = iv;
6672 	} else {
6673 		uiop->uio_iovcnt = 0;
6674 		uiop->uio_iov = iv = NULL;
6675 		m = m2 = NULL;
6676 	}
6677 	uiop->uio_offset = 0;
6678 	uiop->uio_resid = tlen;
6679 	uiop->uio_rw = UIO_READ;
6680 	uiop->uio_segflg = UIO_SYSSPACE;
6681 	uiop->uio_td = p;
6682 #ifdef MAC
6683 	error = mac_vnode_check_getextattr(cred, vp, EXTATTR_NAMESPACE_USER,
6684 	    name);
6685 	if (error != 0)
6686 		goto out;
6687 #endif
6688 
6689 	if (tlen > 0)
6690 		error = VOP_GETEXTATTR(vp, EXTATTR_NAMESPACE_USER, name, uiop,
6691 		    NULL, cred, p);
6692 	if (error != 0)
6693 		goto out;
6694 	if (uiop->uio_resid > 0) {
6695 		alen = tlen;
6696 		len = tlen - uiop->uio_resid;
6697 		tlen = NFSM_RNDUP(len);
6698 		if (alen != tlen)
6699 			printf("nfsvno_getxattr: weird size read\n");
6700 		if (tlen == 0) {
6701 			m_freem(m);
6702 			m = m2 = NULL;
6703 		} else if (alen != tlen || tlen != len)
6704 			m2 = nfsrv_adj(m, alen - tlen, tlen - len);
6705 	}
6706 	*lenp = len;
6707 	*mpp = m;
6708 	*mpendp = m2;
6709 
6710 out:
6711 	if (error != 0) {
6712 		if (m != NULL)
6713 			m_freem(m);
6714 		*lenp = 0;
6715 	}
6716 	free(iv, M_TEMP);
6717 	NFSEXITCODE(error);
6718 	return (error);
6719 }
6720 
6721 /*
6722  * Set Extended attribute vnode op from an mbuf list.
6723  */
6724 int
6725 nfsvno_setxattr(struct vnode *vp, char *name, int len, struct mbuf *m,
6726     char *cp, struct ucred *cred, struct thread *p)
6727 {
6728 	struct iovec *iv;
6729 	struct uio uio, *uiop = &uio;
6730 	int cnt, error;
6731 
6732 	error = 0;
6733 #ifdef MAC
6734 	error = mac_vnode_check_setextattr(cred, vp, EXTATTR_NAMESPACE_USER,
6735 	    name);
6736 #endif
6737 	if (error != 0)
6738 		goto out;
6739 
6740 	uiop->uio_rw = UIO_WRITE;
6741 	uiop->uio_segflg = UIO_SYSSPACE;
6742 	uiop->uio_td = p;
6743 	uiop->uio_offset = 0;
6744 	uiop->uio_resid = len;
6745 	if (len > 0) {
6746 		error = nfsrv_createiovecw(len, m, cp, &iv, &cnt);
6747 		uiop->uio_iov = iv;
6748 		uiop->uio_iovcnt = cnt;
6749 	} else {
6750 		uiop->uio_iov = iv = NULL;
6751 		uiop->uio_iovcnt = 0;
6752 	}
6753 	if (error == 0) {
6754 		error = VOP_SETEXTATTR(vp, EXTATTR_NAMESPACE_USER, name, uiop,
6755 		    cred, p);
6756 		free(iv, M_TEMP);
6757 	}
6758 
6759 out:
6760 	NFSEXITCODE(error);
6761 	return (error);
6762 }
6763 
6764 /*
6765  * Remove Extended attribute vnode op.
6766  */
6767 int
6768 nfsvno_rmxattr(struct nfsrv_descript *nd, struct vnode *vp, char *name,
6769     struct ucred *cred, struct thread *p)
6770 {
6771 	int error;
6772 
6773 	/*
6774 	 * Get rid of any delegations.  I am not sure why this is required,
6775 	 * but RFC-8276 says so.
6776 	 */
6777 	error = nfsrv_checkremove(vp, 0, nd, nd->nd_clientid, p);
6778 	if (error != 0)
6779 		goto out;
6780 #ifdef MAC
6781 	error = mac_vnode_check_deleteextattr(cred, vp, EXTATTR_NAMESPACE_USER,
6782 	    name);
6783 	if (error != 0)
6784 		goto out;
6785 #endif
6786 
6787 	error = VOP_DELETEEXTATTR(vp, EXTATTR_NAMESPACE_USER, name, cred, p);
6788 	if (error == EOPNOTSUPP)
6789 		error = VOP_SETEXTATTR(vp, EXTATTR_NAMESPACE_USER, name, NULL,
6790 		    cred, p);
6791 out:
6792 	NFSEXITCODE(error);
6793 	return (error);
6794 }
6795 
6796 /*
6797  * List Extended Atribute vnode op into an mbuf list.
6798  */
6799 int
6800 nfsvno_listxattr(struct vnode *vp, uint64_t cookie, struct ucred *cred,
6801     struct thread *p, u_char **bufp, uint32_t *lenp, bool *eofp)
6802 {
6803 	struct iovec iv;
6804 	struct uio io;
6805 	int error;
6806 	size_t siz;
6807 
6808 	*bufp = NULL;
6809 	/* First, find out the size of the extended attribute. */
6810 	error = VOP_LISTEXTATTR(vp, EXTATTR_NAMESPACE_USER, NULL, &siz, cred,
6811 	    p);
6812 	if (error != 0)
6813 		return (NFSERR_NOXATTR);
6814 	if (siz <= cookie) {
6815 		*lenp = 0;
6816 		*eofp = true;
6817 		goto out;
6818 	}
6819 	if (siz > cookie + *lenp) {
6820 		siz = cookie + *lenp;
6821 		*eofp = false;
6822 	} else
6823 		*eofp = true;
6824 	/* Just choose a sanity limit of 10Mbytes for malloc(M_TEMP). */
6825 	if (siz > 10 * 1024 * 1024) {
6826 		error = NFSERR_XATTR2BIG;
6827 		goto out;
6828 	}
6829 	*bufp = malloc(siz, M_TEMP, M_WAITOK);
6830 	iv.iov_base = *bufp;
6831 	iv.iov_len = siz;
6832 	io.uio_iovcnt = 1;
6833 	io.uio_iov = &iv;
6834 	io.uio_offset = 0;
6835 	io.uio_resid = siz;
6836 	io.uio_rw = UIO_READ;
6837 	io.uio_segflg = UIO_SYSSPACE;
6838 	io.uio_td = p;
6839 #ifdef MAC
6840 	error = mac_vnode_check_listextattr(cred, vp, EXTATTR_NAMESPACE_USER);
6841 	if (error != 0)
6842 		goto out;
6843 #endif
6844 
6845 	error = VOP_LISTEXTATTR(vp, EXTATTR_NAMESPACE_USER, &io, NULL, cred,
6846 	    p);
6847 	if (error != 0)
6848 		goto out;
6849 	if (io.uio_resid > 0)
6850 		siz -= io.uio_resid;
6851 	*lenp = siz;
6852 
6853 out:
6854 	if (error != 0) {
6855 		free(*bufp, M_TEMP);
6856 		*bufp = NULL;
6857 	}
6858 	NFSEXITCODE(error);
6859 	return (error);
6860 }
6861 
6862 /*
6863  * Trim trailing data off the mbuf list being built.
6864  */
6865 void
6866 nfsm_trimtrailing(struct nfsrv_descript *nd, struct mbuf *mb, char *bpos,
6867     int bextpg, int bextpgsiz)
6868 {
6869 	vm_page_t pg;
6870 	int fullpgsiz, i;
6871 
6872 	if (mb->m_next != NULL) {
6873 		m_freem(mb->m_next);
6874 		mb->m_next = NULL;
6875 	}
6876 	if ((mb->m_flags & M_EXTPG) != 0) {
6877 		KASSERT(bextpg >= 0 && bextpg < mb->m_epg_npgs,
6878 		    ("nfsm_trimtrailing: bextpg out of range"));
6879 		KASSERT(bpos == (char *)(void *)
6880 		    PHYS_TO_DMAP(mb->m_epg_pa[bextpg]) + PAGE_SIZE - bextpgsiz,
6881 		    ("nfsm_trimtrailing: bextpgsiz bad!"));
6882 
6883 		/* First, get rid of any pages after this position. */
6884 		for (i = mb->m_epg_npgs - 1; i > bextpg; i--) {
6885 			pg = PHYS_TO_VM_PAGE(mb->m_epg_pa[i]);
6886 			vm_page_unwire_noq(pg);
6887 			vm_page_free(pg);
6888 		}
6889 		mb->m_epg_npgs = bextpg + 1;
6890 		if (bextpg == 0)
6891 			fullpgsiz = PAGE_SIZE - mb->m_epg_1st_off;
6892 		else
6893 			fullpgsiz = PAGE_SIZE;
6894 		mb->m_epg_last_len = fullpgsiz - bextpgsiz;
6895 		mb->m_len = m_epg_pagelen(mb, 0, mb->m_epg_1st_off);
6896 		for (i = 1; i < mb->m_epg_npgs; i++)
6897 			mb->m_len += m_epg_pagelen(mb, i, 0);
6898 		nd->nd_bextpgsiz = bextpgsiz;
6899 		nd->nd_bextpg = bextpg;
6900 	} else
6901 		mb->m_len = bpos - mtod(mb, char *);
6902 	nd->nd_mb = mb;
6903 	nd->nd_bpos = bpos;
6904 }
6905 
6906 
6907 /*
6908  * Check to see if a put file handle operation should test for
6909  * NFSERR_WRONGSEC, although NFSv3 actually returns NFSERR_AUTHERR.
6910  * When Open is the next operation, NFSERR_WRONGSEC cannot be
6911  * replied for the Open cases that use a component.  Thia can
6912  * be identified by the fact that the file handle's type is VDIR.
6913  */
6914 bool
6915 nfsrv_checkwrongsec(struct nfsrv_descript *nd, int nextop, enum vtype vtyp)
6916 {
6917 
6918 	if ((nd->nd_flag & ND_NFSV4) == 0) {
6919 		if (nd->nd_procnum == NFSPROC_FSINFO)
6920 			return (false);
6921 		return (true);
6922 	}
6923 
6924 	if ((nd->nd_flag & ND_LASTOP) != 0)
6925 		return (false);
6926 
6927 	if (nextop == NFSV4OP_PUTROOTFH || nextop == NFSV4OP_PUTFH ||
6928 	    nextop == NFSV4OP_PUTPUBFH || nextop == NFSV4OP_RESTOREFH ||
6929 	    nextop == NFSV4OP_LOOKUP || nextop == NFSV4OP_LOOKUPP ||
6930 	    nextop == NFSV4OP_SECINFO || nextop == NFSV4OP_SECINFONONAME)
6931 		return (false);
6932 	if (nextop == NFSV4OP_OPEN && vtyp == VDIR)
6933 		return (false);
6934 	return (true);
6935 }
6936 
6937 extern int (*nfsd_call_nfsd)(struct thread *, struct nfssvc_args *);
6938 
6939 /*
6940  * Called once to initialize data structures...
6941  */
6942 static int
6943 nfsd_modevent(module_t mod, int type, void *data)
6944 {
6945 	int error = 0, i;
6946 	static int loaded = 0;
6947 
6948 	switch (type) {
6949 	case MOD_LOAD:
6950 		if (loaded)
6951 			goto out;
6952 		newnfs_portinit();
6953 		for (i = 0; i < NFSRVCACHE_HASHSIZE; i++) {
6954 			mtx_init(&nfsrchash_table[i].mtx, "nfsrtc", NULL,
6955 			    MTX_DEF);
6956 			mtx_init(&nfsrcahash_table[i].mtx, "nfsrtca", NULL,
6957 			    MTX_DEF);
6958 		}
6959 		mtx_init(&nfsrc_udpmtx, "nfsuc", NULL, MTX_DEF);
6960 		mtx_init(&nfs_v4root_mutex, "nfs4rt", NULL, MTX_DEF);
6961 		mtx_init(&nfsv4root_mnt.mnt_mtx, "nfs4mnt", NULL, MTX_DEF);
6962 		mtx_init(&nfsrv_dontlistlock_mtx, "nfs4dnl", NULL, MTX_DEF);
6963 		mtx_init(&nfsrv_recalllock_mtx, "nfs4rec", NULL, MTX_DEF);
6964 		lockinit(&nfsv4root_mnt.mnt_explock, PVFS, "explock", 0, 0);
6965 		nfsrvd_initcache();
6966 		nfsd_init();
6967 		NFSD_LOCK();
6968 		nfsrvd_init(0);
6969 		NFSD_UNLOCK();
6970 		nfsd_mntinit();
6971 #ifdef VV_DISABLEDELEG
6972 		vn_deleg_ops.vndeleg_recall = nfsd_recalldelegation;
6973 		vn_deleg_ops.vndeleg_disable = nfsd_disabledelegation;
6974 #endif
6975 		nfsd_call_servertimer = nfsrv_servertimer;
6976 		nfsd_call_nfsd = nfssvc_nfsd;
6977 		loaded = 1;
6978 		break;
6979 
6980 	case MOD_UNLOAD:
6981 		if (newnfs_numnfsd != 0) {
6982 			error = EBUSY;
6983 			break;
6984 		}
6985 
6986 #ifdef VV_DISABLEDELEG
6987 		vn_deleg_ops.vndeleg_recall = NULL;
6988 		vn_deleg_ops.vndeleg_disable = NULL;
6989 #endif
6990 		nfsd_call_servertimer = NULL;
6991 		nfsd_call_nfsd = NULL;
6992 
6993 		/* Clean out all NFSv4 state. */
6994 		nfsrv_throwawayallstate(curthread);
6995 
6996 		/* Clean the NFS server reply cache */
6997 		nfsrvd_cleancache();
6998 
6999 		/* Free up the krpc server pool. */
7000 		if (nfsrvd_pool != NULL)
7001 			svcpool_destroy(nfsrvd_pool);
7002 
7003 		/* and get rid of the locks */
7004 		for (i = 0; i < NFSRVCACHE_HASHSIZE; i++) {
7005 			mtx_destroy(&nfsrchash_table[i].mtx);
7006 			mtx_destroy(&nfsrcahash_table[i].mtx);
7007 		}
7008 		mtx_destroy(&nfsrc_udpmtx);
7009 		mtx_destroy(&nfs_v4root_mutex);
7010 		mtx_destroy(&nfsv4root_mnt.mnt_mtx);
7011 		mtx_destroy(&nfsrv_dontlistlock_mtx);
7012 		mtx_destroy(&nfsrv_recalllock_mtx);
7013 		for (i = 0; i < nfsrv_sessionhashsize; i++)
7014 			mtx_destroy(&nfssessionhash[i].mtx);
7015 		if (nfslayouthash != NULL) {
7016 			for (i = 0; i < nfsrv_layouthashsize; i++)
7017 				mtx_destroy(&nfslayouthash[i].mtx);
7018 			free(nfslayouthash, M_NFSDSESSION);
7019 		}
7020 		lockdestroy(&nfsv4root_mnt.mnt_explock);
7021 		free(nfsclienthash, M_NFSDCLIENT);
7022 		free(nfslockhash, M_NFSDLOCKFILE);
7023 		free(nfssessionhash, M_NFSDSESSION);
7024 		loaded = 0;
7025 		break;
7026 	default:
7027 		error = EOPNOTSUPP;
7028 		break;
7029 	}
7030 
7031 out:
7032 	NFSEXITCODE(error);
7033 	return (error);
7034 }
7035 static moduledata_t nfsd_mod = {
7036 	"nfsd",
7037 	nfsd_modevent,
7038 	NULL,
7039 };
7040 DECLARE_MODULE(nfsd, nfsd_mod, SI_SUB_VFS, SI_ORDER_ANY);
7041 
7042 /* So that loader and kldload(2) can find us, wherever we are.. */
7043 MODULE_VERSION(nfsd, 1);
7044 MODULE_DEPEND(nfsd, nfscommon, 1, 1, 1);
7045 MODULE_DEPEND(nfsd, nfslockd, 1, 1, 1);
7046 MODULE_DEPEND(nfsd, krpc, 1, 1, 1);
7047 MODULE_DEPEND(nfsd, nfssvc, 1, 1, 1);
7048