xref: /freebsd/sys/fs/nfsserver/nfs_nfsdport.c (revision 9d54812421274e490dc5f0fe4722ab8d35d9b258)
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/callout.h>
51 #include <sys/filio.h>
52 #include <sys/hash.h>
53 #include <sys/sysctl.h>
54 #include <nlm/nlm_prot.h>
55 #include <nlm/nlm.h>
56 
57 FEATURE(nfsd, "NFSv4 server");
58 
59 extern u_int32_t newnfs_true, newnfs_false, newnfs_xdrneg1;
60 extern int nfsrv_useacl;
61 extern int newnfs_numnfsd;
62 extern struct mount nfsv4root_mnt;
63 extern struct nfsrv_stablefirst nfsrv_stablefirst;
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 struct callout nfsd_callout;
101 
102 static int nfssvc_srvcall(struct thread *, struct nfssvc_args *,
103     struct ucred *);
104 
105 int nfsrv_enable_crossmntpt = 1;
106 static int nfs_commit_blks;
107 static int nfs_commit_miss;
108 extern int nfsrv_issuedelegs;
109 extern int nfsrv_dolocallocks;
110 extern int nfsd_enable_stringtouid;
111 extern struct nfsdevicehead nfsrv_devidhead;
112 
113 static int nfsrv_createiovec(int, struct mbuf **, struct mbuf **,
114     struct iovec **);
115 static int nfsrv_createiovec_extpgs(int, int, struct mbuf **,
116     struct mbuf **, struct iovec **);
117 static int nfsrv_createiovecw(int, struct mbuf *, char *, struct iovec **,
118     int *);
119 static void nfsrv_pnfscreate(struct vnode *, struct vattr *, struct ucred *,
120     NFSPROC_T *);
121 static void nfsrv_pnfsremovesetup(struct vnode *, NFSPROC_T *, struct vnode **,
122     int *, char *, fhandle_t *);
123 static void nfsrv_pnfsremove(struct vnode **, int, char *, fhandle_t *,
124     NFSPROC_T *);
125 static int nfsrv_proxyds(struct vnode *, off_t, int, struct ucred *,
126     struct thread *, int, struct mbuf **, char *, struct mbuf **,
127     struct nfsvattr *, struct acl *, off_t *, int, bool *);
128 static int nfsrv_setextattr(struct vnode *, struct nfsvattr *, NFSPROC_T *);
129 static int nfsrv_readdsrpc(fhandle_t *, off_t, int, struct ucred *,
130     NFSPROC_T *, struct nfsmount *, struct mbuf **, struct mbuf **);
131 static int nfsrv_writedsrpc(fhandle_t *, off_t, int, struct ucred *,
132     NFSPROC_T *, struct vnode *, struct nfsmount **, int, struct mbuf **,
133     char *, int *);
134 static int nfsrv_allocatedsrpc(fhandle_t *, off_t, off_t, struct ucred *,
135     NFSPROC_T *, struct vnode *, struct nfsmount **, int, int *);
136 static int nfsrv_deallocatedsrpc(fhandle_t *, off_t, off_t, struct ucred *,
137     NFSPROC_T *, struct vnode *, struct nfsmount **, int, int *);
138 static int nfsrv_setacldsrpc(fhandle_t *, struct ucred *, NFSPROC_T *,
139     struct vnode *, struct nfsmount **, int, struct acl *, int *);
140 static int nfsrv_setattrdsrpc(fhandle_t *, struct ucred *, NFSPROC_T *,
141     struct vnode *, struct nfsmount **, int, struct nfsvattr *, int *);
142 static int nfsrv_getattrdsrpc(fhandle_t *, struct ucred *, NFSPROC_T *,
143     struct vnode *, struct nfsmount *, struct nfsvattr *);
144 static int nfsrv_seekdsrpc(fhandle_t *, off_t *, int, bool *, struct ucred *,
145     NFSPROC_T *, struct nfsmount *);
146 static int nfsrv_putfhname(fhandle_t *, char *);
147 static int nfsrv_pnfslookupds(struct vnode *, struct vnode *,
148     struct pnfsdsfile *, struct vnode **, NFSPROC_T *);
149 static void nfsrv_pnfssetfh(struct vnode *, struct pnfsdsfile *, char *, char *,
150     struct vnode *, NFSPROC_T *);
151 static int nfsrv_dsremove(struct vnode *, char *, struct ucred *, NFSPROC_T *);
152 static int nfsrv_dssetacl(struct vnode *, struct acl *, struct ucred *,
153     NFSPROC_T *);
154 static int nfsrv_pnfsstatfs(struct statfs *, struct mount *);
155 
156 int nfs_pnfsio(task_fn_t *, void *);
157 
158 SYSCTL_NODE(_vfs, OID_AUTO, nfsd, CTLFLAG_RW | CTLFLAG_MPSAFE, 0,
159     "NFS server");
160 SYSCTL_INT(_vfs_nfsd, OID_AUTO, mirrormnt, CTLFLAG_RW,
161     &nfsrv_enable_crossmntpt, 0, "Enable nfsd to cross mount points");
162 SYSCTL_INT(_vfs_nfsd, OID_AUTO, commit_blks, CTLFLAG_RW, &nfs_commit_blks,
163     0, "");
164 SYSCTL_INT(_vfs_nfsd, OID_AUTO, commit_miss, CTLFLAG_RW, &nfs_commit_miss,
165     0, "");
166 SYSCTL_INT(_vfs_nfsd, OID_AUTO, issue_delegations, CTLFLAG_RW,
167     &nfsrv_issuedelegs, 0, "Enable nfsd to issue delegations");
168 SYSCTL_INT(_vfs_nfsd, OID_AUTO, enable_locallocks, CTLFLAG_RW,
169     &nfsrv_dolocallocks, 0, "Enable nfsd to acquire local locks on files");
170 SYSCTL_INT(_vfs_nfsd, OID_AUTO, debuglevel, CTLFLAG_RW, &nfsd_debuglevel,
171     0, "Debug level for NFS server");
172 SYSCTL_INT(_vfs_nfsd, OID_AUTO, enable_stringtouid, CTLFLAG_RW,
173     &nfsd_enable_stringtouid, 0, "Enable nfsd to accept numeric owner_names");
174 static int nfsrv_pnfsgetdsattr = 1;
175 SYSCTL_INT(_vfs_nfsd, OID_AUTO, pnfsgetdsattr, CTLFLAG_RW,
176     &nfsrv_pnfsgetdsattr, 0, "When set getattr gets DS attributes via RPC");
177 
178 /*
179  * nfsrv_dsdirsize can only be increased and only when the nfsd threads are
180  * not running.
181  * The dsN subdirectories for the increased values must have been created
182  * on all DS servers before this increase is done.
183  */
184 u_int	nfsrv_dsdirsize = 20;
185 static int
186 sysctl_dsdirsize(SYSCTL_HANDLER_ARGS)
187 {
188 	int error, newdsdirsize;
189 
190 	newdsdirsize = nfsrv_dsdirsize;
191 	error = sysctl_handle_int(oidp, &newdsdirsize, 0, req);
192 	if (error != 0 || req->newptr == NULL)
193 		return (error);
194 	if (newdsdirsize <= nfsrv_dsdirsize || newdsdirsize > 10000 ||
195 	    newnfs_numnfsd != 0)
196 		return (EINVAL);
197 	nfsrv_dsdirsize = newdsdirsize;
198 	return (0);
199 }
200 SYSCTL_PROC(_vfs_nfsd, OID_AUTO, dsdirsize,
201     CTLTYPE_UINT | CTLFLAG_MPSAFE | CTLFLAG_RW, 0, sizeof(nfsrv_dsdirsize),
202     sysctl_dsdirsize, "IU", "Number of dsN subdirs on the DS servers");
203 
204 /*
205  * nfs_srvmaxio can only be increased and only when the nfsd threads are
206  * not running.  The setting must be a power of 2, with the current limit of
207  * 1Mbyte.
208  */
209 static int
210 sysctl_srvmaxio(SYSCTL_HANDLER_ARGS)
211 {
212 	int error;
213 	u_int newsrvmaxio;
214 	uint64_t tval;
215 
216 	newsrvmaxio = nfs_srvmaxio;
217 	error = sysctl_handle_int(oidp, &newsrvmaxio, 0, req);
218 	if (error != 0 || req->newptr == NULL)
219 		return (error);
220 	if (newsrvmaxio == nfs_srvmaxio)
221 		return (0);
222 	if (newsrvmaxio < nfs_srvmaxio) {
223 		printf("nfsd: vfs.nfsd.srvmaxio can only be increased\n");
224 		return (EINVAL);
225 	}
226 	if (newsrvmaxio > 1048576) {
227 		printf("nfsd: vfs.nfsd.srvmaxio cannot be > 1Mbyte\n");
228 		return (EINVAL);
229 	}
230 	if ((newsrvmaxio & (newsrvmaxio - 1)) != 0) {
231 		printf("nfsd: vfs.nfsd.srvmaxio must be a power of 2\n");
232 		return (EINVAL);
233 	}
234 
235 	/*
236 	 * Check that kern.ipc.maxsockbuf is large enough for
237 	 * newsrviomax, given the setting of vfs.nfs.bufpackets.
238 	 */
239 	if ((newsrvmaxio + NFS_MAXXDR) * nfs_bufpackets >
240 	    sb_max_adj) {
241 		/*
242 		 * Suggest vfs.nfs.bufpackets * maximum RPC message for
243 		 * sb_max_adj.
244 		 */
245 		tval = (newsrvmaxio + NFS_MAXXDR) * nfs_bufpackets;
246 
247 		/*
248 		 * Convert suggested sb_max_adj value to a suggested
249 		 * sb_max value, which is what is set via kern.ipc.maxsockbuf.
250 		 * Perform the inverse calculation of (from uipc_sockbuf.c):
251 		 * sb_max_adj = (u_quad_t)sb_max * MCLBYTES /
252 		 *     (MSIZE + MCLBYTES);
253 		 * XXX If the calculation of sb_max_adj from sb_max changes,
254 		 *     this calculation must be changed as well.
255 		 */
256 		tval *= (MSIZE + MCLBYTES);  /* Brackets for readability. */
257 		tval += MCLBYTES - 1;        /* Round up divide. */
258 		tval /= MCLBYTES;
259 		printf("nfsd: set kern.ipc.maxsockbuf to a minimum of "
260 		    "%ju to support %ubyte NFS I/O\n", (uintmax_t)tval,
261 		    newsrvmaxio);
262 		return (EINVAL);
263 	}
264 
265 	NFSD_LOCK();
266 	if (newnfs_numnfsd != 0) {
267 		NFSD_UNLOCK();
268 		printf("nfsd: cannot set vfs.nfsd.srvmaxio when nfsd "
269 		    "threads are running\n");
270 		return (EINVAL);
271 	}
272 
273 
274 	nfs_srvmaxio = newsrvmaxio;
275 	NFSD_UNLOCK();
276 	return (0);
277 }
278 SYSCTL_PROC(_vfs_nfsd, OID_AUTO, srvmaxio,
279     CTLTYPE_UINT | CTLFLAG_MPSAFE | CTLFLAG_RW, NULL, 0,
280     sysctl_srvmaxio, "IU", "Maximum I/O size in bytes");
281 
282 #define	MAX_REORDERED_RPC	16
283 #define	NUM_HEURISTIC		1031
284 #define	NHUSE_INIT		64
285 #define	NHUSE_INC		16
286 #define	NHUSE_MAX		2048
287 
288 static struct nfsheur {
289 	struct vnode *nh_vp;	/* vp to match (unreferenced pointer) */
290 	off_t nh_nextoff;	/* next offset for sequential detection */
291 	int nh_use;		/* use count for selection */
292 	int nh_seqcount;	/* heuristic */
293 } nfsheur[NUM_HEURISTIC];
294 
295 /*
296  * Heuristic to detect sequential operation.
297  */
298 static struct nfsheur *
299 nfsrv_sequential_heuristic(struct uio *uio, struct vnode *vp)
300 {
301 	struct nfsheur *nh;
302 	int hi, try;
303 
304 	/* Locate best candidate. */
305 	try = 32;
306 	hi = ((int)(vm_offset_t)vp / sizeof(struct vnode)) % NUM_HEURISTIC;
307 	nh = &nfsheur[hi];
308 	while (try--) {
309 		if (nfsheur[hi].nh_vp == vp) {
310 			nh = &nfsheur[hi];
311 			break;
312 		}
313 		if (nfsheur[hi].nh_use > 0)
314 			--nfsheur[hi].nh_use;
315 		hi = (hi + 1) % NUM_HEURISTIC;
316 		if (nfsheur[hi].nh_use < nh->nh_use)
317 			nh = &nfsheur[hi];
318 	}
319 
320 	/* Initialize hint if this is a new file. */
321 	if (nh->nh_vp != vp) {
322 		nh->nh_vp = vp;
323 		nh->nh_nextoff = uio->uio_offset;
324 		nh->nh_use = NHUSE_INIT;
325 		if (uio->uio_offset == 0)
326 			nh->nh_seqcount = 4;
327 		else
328 			nh->nh_seqcount = 1;
329 	}
330 
331 	/* Calculate heuristic. */
332 	if ((uio->uio_offset == 0 && nh->nh_seqcount > 0) ||
333 	    uio->uio_offset == nh->nh_nextoff) {
334 		/* See comments in vfs_vnops.c:sequential_heuristic(). */
335 		nh->nh_seqcount += howmany(uio->uio_resid, 16384);
336 		if (nh->nh_seqcount > IO_SEQMAX)
337 			nh->nh_seqcount = IO_SEQMAX;
338 	} else if (qabs(uio->uio_offset - nh->nh_nextoff) <= MAX_REORDERED_RPC *
339 	    imax(vp->v_mount->mnt_stat.f_iosize, uio->uio_resid)) {
340 		/* Probably a reordered RPC, leave seqcount alone. */
341 	} else if (nh->nh_seqcount > 1) {
342 		nh->nh_seqcount /= 2;
343 	} else {
344 		nh->nh_seqcount = 0;
345 	}
346 	nh->nh_use += NHUSE_INC;
347 	if (nh->nh_use > NHUSE_MAX)
348 		nh->nh_use = NHUSE_MAX;
349 	return (nh);
350 }
351 
352 /*
353  * Get attributes into nfsvattr structure.
354  */
355 int
356 nfsvno_getattr(struct vnode *vp, struct nfsvattr *nvap,
357     struct nfsrv_descript *nd, struct thread *p, int vpislocked,
358     nfsattrbit_t *attrbitp)
359 {
360 	int error, gotattr, lockedit = 0;
361 	struct nfsvattr na;
362 
363 	if (vpislocked == 0) {
364 		/*
365 		 * When vpislocked == 0, the vnode is either exclusively
366 		 * locked by this thread or not locked by this thread.
367 		 * As such, shared lock it, if not exclusively locked.
368 		 */
369 		if (NFSVOPISLOCKED(vp) != LK_EXCLUSIVE) {
370 			lockedit = 1;
371 			NFSVOPLOCK(vp, LK_SHARED | LK_RETRY);
372 		}
373 	}
374 
375 	/*
376 	 * Acquire the Change, Size, TimeAccess, TimeModify and SpaceUsed
377 	 * attributes, as required.
378 	 * This needs to be done for regular files if:
379 	 * - non-NFSv4 RPCs or
380 	 * - when attrbitp == NULL or
381 	 * - an NFSv4 RPC with any of the above attributes in attrbitp.
382 	 * A return of 0 for nfsrv_proxyds() indicates that it has acquired
383 	 * these attributes.  nfsrv_proxyds() will return an error if the
384 	 * server is not a pNFS one.
385 	 */
386 	gotattr = 0;
387 	if (vp->v_type == VREG && nfsrv_devidcnt > 0 && (attrbitp == NULL ||
388 	    (nd->nd_flag & ND_NFSV4) == 0 ||
389 	    NFSISSET_ATTRBIT(attrbitp, NFSATTRBIT_CHANGE) ||
390 	    NFSISSET_ATTRBIT(attrbitp, NFSATTRBIT_SIZE) ||
391 	    NFSISSET_ATTRBIT(attrbitp, NFSATTRBIT_TIMEACCESS) ||
392 	    NFSISSET_ATTRBIT(attrbitp, NFSATTRBIT_TIMEMODIFY) ||
393 	    NFSISSET_ATTRBIT(attrbitp, NFSATTRBIT_SPACEUSED))) {
394 		error = nfsrv_proxyds(vp, 0, 0, nd->nd_cred, p,
395 		    NFSPROC_GETATTR, NULL, NULL, NULL, &na, NULL, NULL, 0,
396 		    NULL);
397 		if (error == 0)
398 			gotattr = 1;
399 	}
400 
401 	error = VOP_GETATTR(vp, &nvap->na_vattr, nd->nd_cred);
402 	if (lockedit != 0)
403 		NFSVOPUNLOCK(vp);
404 
405 	/*
406 	 * If we got the Change, Size and Modify Time from the DS,
407 	 * replace them.
408 	 */
409 	if (gotattr != 0) {
410 		nvap->na_atime = na.na_atime;
411 		nvap->na_mtime = na.na_mtime;
412 		nvap->na_filerev = na.na_filerev;
413 		nvap->na_size = na.na_size;
414 		nvap->na_bytes = na.na_bytes;
415 	}
416 	NFSD_DEBUG(4, "nfsvno_getattr: gotattr=%d err=%d chg=%ju\n", gotattr,
417 	    error, (uintmax_t)na.na_filerev);
418 
419 	NFSEXITCODE(error);
420 	return (error);
421 }
422 
423 /*
424  * Get a file handle for a vnode.
425  */
426 int
427 nfsvno_getfh(struct vnode *vp, fhandle_t *fhp, struct thread *p)
428 {
429 	int error;
430 
431 	NFSBZERO((caddr_t)fhp, sizeof(fhandle_t));
432 	fhp->fh_fsid = vp->v_mount->mnt_stat.f_fsid;
433 	error = VOP_VPTOFH(vp, &fhp->fh_fid);
434 
435 	NFSEXITCODE(error);
436 	return (error);
437 }
438 
439 /*
440  * Perform access checking for vnodes obtained from file handles that would
441  * refer to files already opened by a Unix client. You cannot just use
442  * vn_writechk() and VOP_ACCESSX() for two reasons.
443  * 1 - You must check for exported rdonly as well as MNT_RDONLY for the write
444  *     case.
445  * 2 - The owner is to be given access irrespective of mode bits for some
446  *     operations, so that processes that chmod after opening a file don't
447  *     break.
448  */
449 int
450 nfsvno_accchk(struct vnode *vp, accmode_t accmode, struct ucred *cred,
451     struct nfsexstuff *exp, struct thread *p, int override, int vpislocked,
452     u_int32_t *supportedtypep)
453 {
454 	struct vattr vattr;
455 	int error = 0, getret = 0;
456 
457 	if (vpislocked == 0) {
458 		if (NFSVOPLOCK(vp, LK_SHARED) != 0) {
459 			error = EPERM;
460 			goto out;
461 		}
462 	}
463 	if (accmode & VWRITE) {
464 		/* Just vn_writechk() changed to check rdonly */
465 		/*
466 		 * Disallow write attempts on read-only file systems;
467 		 * unless the file is a socket or a block or character
468 		 * device resident on the file system.
469 		 */
470 		if (NFSVNO_EXRDONLY(exp) ||
471 		    (vp->v_mount->mnt_flag & MNT_RDONLY)) {
472 			switch (vp->v_type) {
473 			case VREG:
474 			case VDIR:
475 			case VLNK:
476 				error = EROFS;
477 			default:
478 				break;
479 			}
480 		}
481 		/*
482 		 * If there's shared text associated with
483 		 * the inode, try to free it up once.  If
484 		 * we fail, we can't allow writing.
485 		 */
486 		if (VOP_IS_TEXT(vp) && error == 0)
487 			error = ETXTBSY;
488 	}
489 	if (error != 0) {
490 		if (vpislocked == 0)
491 			NFSVOPUNLOCK(vp);
492 		goto out;
493 	}
494 
495 	/*
496 	 * Should the override still be applied when ACLs are enabled?
497 	 */
498 	error = VOP_ACCESSX(vp, accmode, cred, p);
499 	if (error != 0 && (accmode & (VDELETE | VDELETE_CHILD))) {
500 		/*
501 		 * Try again with VEXPLICIT_DENY, to see if the test for
502 		 * deletion is supported.
503 		 */
504 		error = VOP_ACCESSX(vp, accmode | VEXPLICIT_DENY, cred, p);
505 		if (error == 0) {
506 			if (vp->v_type == VDIR) {
507 				accmode &= ~(VDELETE | VDELETE_CHILD);
508 				accmode |= VWRITE;
509 				error = VOP_ACCESSX(vp, accmode, cred, p);
510 			} else if (supportedtypep != NULL) {
511 				*supportedtypep &= ~NFSACCESS_DELETE;
512 			}
513 		}
514 	}
515 
516 	/*
517 	 * Allow certain operations for the owner (reads and writes
518 	 * on files that are already open).
519 	 */
520 	if (override != NFSACCCHK_NOOVERRIDE &&
521 	    (error == EPERM || error == EACCES)) {
522 		if (cred->cr_uid == 0 && (override & NFSACCCHK_ALLOWROOT))
523 			error = 0;
524 		else if (override & NFSACCCHK_ALLOWOWNER) {
525 			getret = VOP_GETATTR(vp, &vattr, cred);
526 			if (getret == 0 && cred->cr_uid == vattr.va_uid)
527 				error = 0;
528 		}
529 	}
530 	if (vpislocked == 0)
531 		NFSVOPUNLOCK(vp);
532 
533 out:
534 	NFSEXITCODE(error);
535 	return (error);
536 }
537 
538 /*
539  * Set attribute(s) vnop.
540  */
541 int
542 nfsvno_setattr(struct vnode *vp, struct nfsvattr *nvap, struct ucred *cred,
543     struct thread *p, struct nfsexstuff *exp)
544 {
545 	u_quad_t savsize = 0;
546 	int error, savedit;
547 	time_t savbtime;
548 
549 	/*
550 	 * If this is an exported file system and a pNFS service is running,
551 	 * don't VOP_SETATTR() of size for the MDS file system.
552 	 */
553 	savedit = 0;
554 	error = 0;
555 	if (vp->v_type == VREG && (vp->v_mount->mnt_flag & MNT_EXPORTED) != 0 &&
556 	    nfsrv_devidcnt != 0 && nvap->na_vattr.va_size != VNOVAL &&
557 	    nvap->na_vattr.va_size > 0) {
558 		savsize = nvap->na_vattr.va_size;
559 		nvap->na_vattr.va_size = VNOVAL;
560 		if (nvap->na_vattr.va_uid != (uid_t)VNOVAL ||
561 		    nvap->na_vattr.va_gid != (gid_t)VNOVAL ||
562 		    nvap->na_vattr.va_mode != (mode_t)VNOVAL ||
563 		    nvap->na_vattr.va_atime.tv_sec != VNOVAL ||
564 		    nvap->na_vattr.va_mtime.tv_sec != VNOVAL)
565 			savedit = 1;
566 		else
567 			savedit = 2;
568 	}
569 	if (savedit != 2)
570 		error = VOP_SETATTR(vp, &nvap->na_vattr, cred);
571 	if (savedit != 0)
572 		nvap->na_vattr.va_size = savsize;
573 	if (error == 0 && (nvap->na_vattr.va_uid != (uid_t)VNOVAL ||
574 	    nvap->na_vattr.va_gid != (gid_t)VNOVAL ||
575 	    nvap->na_vattr.va_size != VNOVAL ||
576 	    nvap->na_vattr.va_mode != (mode_t)VNOVAL ||
577 	    nvap->na_vattr.va_atime.tv_sec != VNOVAL ||
578 	    nvap->na_vattr.va_mtime.tv_sec != VNOVAL)) {
579 		/* Never modify birthtime on a DS file. */
580 		savbtime = nvap->na_vattr.va_birthtime.tv_sec;
581 		nvap->na_vattr.va_birthtime.tv_sec = VNOVAL;
582 		/* For a pNFS server, set the attributes on the DS file. */
583 		error = nfsrv_proxyds(vp, 0, 0, cred, p, NFSPROC_SETATTR,
584 		    NULL, NULL, NULL, nvap, NULL, NULL, 0, NULL);
585 		nvap->na_vattr.va_birthtime.tv_sec = savbtime;
586 		if (error == ENOENT)
587 			error = 0;
588 	}
589 	NFSEXITCODE(error);
590 	return (error);
591 }
592 
593 /*
594  * Set up nameidata for a lookup() call and do it.
595  */
596 int
597 nfsvno_namei(struct nfsrv_descript *nd, struct nameidata *ndp,
598     struct vnode *dp, int islocked, struct nfsexstuff *exp,
599     struct vnode **retdirp)
600 {
601 	struct componentname *cnp = &ndp->ni_cnd;
602 	int i;
603 	struct iovec aiov;
604 	struct uio auio;
605 	int lockleaf = (cnp->cn_flags & LOCKLEAF) != 0, linklen;
606 	int error = 0;
607 	char *cp;
608 
609 	*retdirp = NULL;
610 	cnp->cn_nameptr = cnp->cn_pnbuf;
611 	ndp->ni_lcf = 0;
612 	/*
613 	 * Extract and set starting directory.
614 	 */
615 	if (dp->v_type != VDIR) {
616 		if (islocked)
617 			vput(dp);
618 		else
619 			vrele(dp);
620 		nfsvno_relpathbuf(ndp);
621 		error = ENOTDIR;
622 		goto out1;
623 	}
624 	if (islocked)
625 		NFSVOPUNLOCK(dp);
626 	VREF(dp);
627 	*retdirp = dp;
628 	if (NFSVNO_EXRDONLY(exp))
629 		cnp->cn_flags |= RDONLY;
630 	ndp->ni_segflg = UIO_SYSSPACE;
631 
632 	if (nd->nd_flag & ND_PUBLOOKUP) {
633 		ndp->ni_loopcnt = 0;
634 		if (cnp->cn_pnbuf[0] == '/') {
635 			vrele(dp);
636 			/*
637 			 * Check for degenerate pathnames here, since lookup()
638 			 * panics on them.
639 			 */
640 			for (i = 1; i < ndp->ni_pathlen; i++)
641 				if (cnp->cn_pnbuf[i] != '/')
642 					break;
643 			if (i == ndp->ni_pathlen) {
644 				error = NFSERR_ACCES;
645 				goto out;
646 			}
647 			dp = rootvnode;
648 			VREF(dp);
649 		}
650 	} else if ((nfsrv_enable_crossmntpt == 0 && NFSVNO_EXPORTED(exp)) ||
651 	    (nd->nd_flag & ND_NFSV4) == 0) {
652 		/*
653 		 * Only cross mount points for NFSv4 when doing a
654 		 * mount while traversing the file system above
655 		 * the mount point, unless nfsrv_enable_crossmntpt is set.
656 		 */
657 		cnp->cn_flags |= NOCROSSMOUNT;
658 	}
659 
660 	/*
661 	 * Initialize for scan, set ni_startdir and bump ref on dp again
662 	 * because lookup() will dereference ni_startdir.
663 	 */
664 
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 = vfs_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 = nfsvno_setattr(*vpp, nvap,
1264 				    nd->nd_cred, p, exp);
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 = nfsvno_setattr(vp,
1934 					    nvap, cred, p, exp);
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 	uint64_t *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 				txdr_hyper(*cookiep, tl);
2263 			} else {
2264 				NFSM_BUILD(tl, u_int32_t *, NFSX_UNSIGNED);
2265 				*tl = txdr_unsigned(*cookiep);
2266 			}
2267 		}
2268 		cpos += dp->d_reclen;
2269 		dp = (struct dirent *)cpos;
2270 		cookiep++;
2271 		ncookies--;
2272 	}
2273 	if (cpos < cend)
2274 		eofflag = 0;
2275 	NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
2276 	*tl++ = newnfs_false;
2277 	if (eofflag)
2278 		*tl = newnfs_true;
2279 	else
2280 		*tl = newnfs_false;
2281 	free(rbuf, M_TEMP);
2282 	free(cookies, M_TEMP);
2283 
2284 out:
2285 	NFSEXITCODE2(0, nd);
2286 	return (0);
2287 nfsmout:
2288 	vput(vp);
2289 	NFSEXITCODE2(error, nd);
2290 	return (error);
2291 }
2292 
2293 /*
2294  * Readdirplus for V3 and Readdir for V4.
2295  */
2296 int
2297 nfsrvd_readdirplus(struct nfsrv_descript *nd, int isdgram,
2298     struct vnode *vp, struct nfsexstuff *exp)
2299 {
2300 	struct dirent *dp;
2301 	u_int32_t *tl;
2302 	int dirlen;
2303 	char *cpos, *cend, *rbuf;
2304 	struct vnode *nvp;
2305 	fhandle_t nfh;
2306 	struct nfsvattr nva, at, *nvap = &nva;
2307 	struct mbuf *mb0, *mb1;
2308 	struct nfsreferral *refp;
2309 	int nlen, r, error = 0, getret = 1, usevget = 1;
2310 	int siz, cnt, fullsiz, eofflag, ncookies, entrycnt;
2311 	caddr_t bpos0, bpos1;
2312 	u_int64_t off, toff, verf __unused;
2313 	uint64_t *cookies = NULL, *cookiep;
2314 	nfsattrbit_t attrbits, rderrbits, savbits;
2315 	struct uio io;
2316 	struct iovec iv;
2317 	struct componentname cn;
2318 	int at_root, is_ufs, is_zfs, needs_unbusy, supports_nfsv4acls;
2319 	struct mount *mp, *new_mp;
2320 	uint64_t mounted_on_fileno;
2321 	struct thread *p = curthread;
2322 	int bextpg0, bextpg1, bextpgsiz0, bextpgsiz1;
2323 
2324 	if (nd->nd_repstat) {
2325 		nfsrv_postopattr(nd, getret, &at);
2326 		goto out;
2327 	}
2328 	NFSM_DISSECT(tl, u_int32_t *, 6 * NFSX_UNSIGNED);
2329 	off = fxdr_hyper(tl);
2330 	toff = off;
2331 	tl += 2;
2332 	verf = fxdr_hyper(tl);
2333 	tl += 2;
2334 	siz = fxdr_unsigned(int, *tl++);
2335 	cnt = fxdr_unsigned(int, *tl);
2336 
2337 	/*
2338 	 * Use the server's maximum data transfer size as the upper bound
2339 	 * on reply datalen.
2340 	 */
2341 	if (cnt > NFS_SRVMAXDATA(nd) || cnt < 0)
2342 		cnt = NFS_SRVMAXDATA(nd);
2343 
2344 	/*
2345 	 * siz is a "hint" of how much directory information (name, fileid,
2346 	 * cookie) should be in the reply. At least one client "hints" 0,
2347 	 * so I set it to cnt for that case. I also round it up to the
2348 	 * next multiple of DIRBLKSIZ.
2349 	 * Since the size of a Readdirplus directory entry reply will always
2350 	 * be greater than a directory entry returned by VOP_READDIR(), it
2351 	 * does not make sense to read more than NFS_SRVMAXDATA() via
2352 	 * VOP_READDIR().
2353 	 */
2354 	if (siz <= 0)
2355 		siz = cnt;
2356 	else if (siz > NFS_SRVMAXDATA(nd))
2357 		siz = NFS_SRVMAXDATA(nd);
2358 	siz = ((siz + DIRBLKSIZ - 1) & ~(DIRBLKSIZ - 1));
2359 
2360 	if (nd->nd_flag & ND_NFSV4) {
2361 		error = nfsrv_getattrbits(nd, &attrbits, NULL, NULL);
2362 		if (error)
2363 			goto nfsmout;
2364 		NFSSET_ATTRBIT(&savbits, &attrbits);
2365 		NFSCLRNOTFILLABLE_ATTRBIT(&attrbits, nd);
2366 		NFSZERO_ATTRBIT(&rderrbits);
2367 		NFSSETBIT_ATTRBIT(&rderrbits, NFSATTRBIT_RDATTRERROR);
2368 	} else {
2369 		NFSZERO_ATTRBIT(&attrbits);
2370 	}
2371 	fullsiz = siz;
2372 	nd->nd_repstat = getret = nfsvno_getattr(vp, &at, nd, p, 1, NULL);
2373 #if 0
2374 	if (!nd->nd_repstat) {
2375 	    if (off && verf != at.na_filerev) {
2376 		/*
2377 		 * va_filerev is not sufficient as a cookie verifier,
2378 		 * since it is not supposed to change when entries are
2379 		 * removed/added unless that offset cookies returned to
2380 		 * the client are no longer valid.
2381 		 */
2382 		if (nd->nd_flag & ND_NFSV4) {
2383 			nd->nd_repstat = NFSERR_NOTSAME;
2384 		} else {
2385 			nd->nd_repstat = NFSERR_BAD_COOKIE;
2386 		}
2387 	    }
2388 	}
2389 #endif
2390 	if (!nd->nd_repstat && vp->v_type != VDIR)
2391 		nd->nd_repstat = NFSERR_NOTDIR;
2392 	if (!nd->nd_repstat && cnt == 0)
2393 		nd->nd_repstat = NFSERR_TOOSMALL;
2394 	if (!nd->nd_repstat)
2395 		nd->nd_repstat = nfsvno_accchk(vp, VEXEC,
2396 		    nd->nd_cred, exp, p, NFSACCCHK_NOOVERRIDE,
2397 		    NFSACCCHK_VPISLOCKED, NULL);
2398 	if (nd->nd_repstat) {
2399 		vput(vp);
2400 		if (nd->nd_flag & ND_NFSV3)
2401 			nfsrv_postopattr(nd, getret, &at);
2402 		goto out;
2403 	}
2404 	is_ufs = strcmp(vp->v_mount->mnt_vfc->vfc_name, "ufs") == 0;
2405 	is_zfs = strcmp(vp->v_mount->mnt_vfc->vfc_name, "zfs") == 0;
2406 
2407 	rbuf = malloc(siz, M_TEMP, M_WAITOK);
2408 again:
2409 	eofflag = 0;
2410 	if (cookies) {
2411 		free(cookies, M_TEMP);
2412 		cookies = NULL;
2413 	}
2414 
2415 	iv.iov_base = rbuf;
2416 	iv.iov_len = siz;
2417 	io.uio_iov = &iv;
2418 	io.uio_iovcnt = 1;
2419 	io.uio_offset = (off_t)off;
2420 	io.uio_resid = siz;
2421 	io.uio_segflg = UIO_SYSSPACE;
2422 	io.uio_rw = UIO_READ;
2423 	io.uio_td = NULL;
2424 	nd->nd_repstat = VOP_READDIR(vp, &io, nd->nd_cred, &eofflag, &ncookies,
2425 	    &cookies);
2426 	off = (u_int64_t)io.uio_offset;
2427 	if (io.uio_resid)
2428 		siz -= io.uio_resid;
2429 
2430 	getret = nfsvno_getattr(vp, &at, nd, p, 1, NULL);
2431 
2432 	if (!cookies && !nd->nd_repstat)
2433 		nd->nd_repstat = NFSERR_PERM;
2434 	if (!nd->nd_repstat)
2435 		nd->nd_repstat = getret;
2436 	if (nd->nd_repstat) {
2437 		vput(vp);
2438 		if (cookies)
2439 			free(cookies, M_TEMP);
2440 		free(rbuf, M_TEMP);
2441 		if (nd->nd_flag & ND_NFSV3)
2442 			nfsrv_postopattr(nd, getret, &at);
2443 		goto out;
2444 	}
2445 	/*
2446 	 * If nothing read, return eof
2447 	 * rpc reply
2448 	 */
2449 	if (siz == 0) {
2450 		vput(vp);
2451 		if (nd->nd_flag & ND_NFSV3)
2452 			nfsrv_postopattr(nd, getret, &at);
2453 		NFSM_BUILD(tl, u_int32_t *, 4 * NFSX_UNSIGNED);
2454 		txdr_hyper(at.na_filerev, tl);
2455 		tl += 2;
2456 		*tl++ = newnfs_false;
2457 		*tl = newnfs_true;
2458 		free(cookies, M_TEMP);
2459 		free(rbuf, M_TEMP);
2460 		goto out;
2461 	}
2462 
2463 	/*
2464 	 * Check for degenerate cases of nothing useful read.
2465 	 * If so go try again
2466 	 */
2467 	cpos = rbuf;
2468 	cend = rbuf + siz;
2469 	dp = (struct dirent *)cpos;
2470 	cookiep = cookies;
2471 
2472 	/*
2473 	 * For some reason FreeBSD's ufs_readdir() chooses to back the
2474 	 * directory offset up to a block boundary, so it is necessary to
2475 	 * skip over the records that precede the requested offset. This
2476 	 * requires the assumption that file offset cookies monotonically
2477 	 * increase.
2478 	 */
2479 	while (cpos < cend && ncookies > 0 &&
2480 	  (dp->d_fileno == 0 || dp->d_type == DT_WHT ||
2481 	   (is_ufs == 1 && ((u_quad_t)(*cookiep)) <= toff) ||
2482 	   ((nd->nd_flag & ND_NFSV4) &&
2483 	    ((dp->d_namlen == 1 && dp->d_name[0] == '.') ||
2484 	     (dp->d_namlen==2 && dp->d_name[0]=='.' && dp->d_name[1]=='.'))))) {
2485 		cpos += dp->d_reclen;
2486 		dp = (struct dirent *)cpos;
2487 		cookiep++;
2488 		ncookies--;
2489 	}
2490 	if (cpos >= cend || ncookies == 0) {
2491 		siz = fullsiz;
2492 		toff = off;
2493 		goto again;
2494 	}
2495 
2496 	/*
2497 	 * Busy the file system so that the mount point won't go away
2498 	 * and, as such, VFS_VGET() can be used safely.
2499 	 */
2500 	mp = vp->v_mount;
2501 	vfs_ref(mp);
2502 	NFSVOPUNLOCK(vp);
2503 	nd->nd_repstat = vfs_busy(mp, 0);
2504 	vfs_rel(mp);
2505 	if (nd->nd_repstat != 0) {
2506 		vrele(vp);
2507 		free(cookies, M_TEMP);
2508 		free(rbuf, M_TEMP);
2509 		if (nd->nd_flag & ND_NFSV3)
2510 			nfsrv_postopattr(nd, getret, &at);
2511 		goto out;
2512 	}
2513 
2514 	/*
2515 	 * Check to see if entries in this directory can be safely acquired
2516 	 * via VFS_VGET() or if a switch to VOP_LOOKUP() is required.
2517 	 * ZFS snapshot directories need VOP_LOOKUP(), so that any
2518 	 * automount of the snapshot directory that is required will
2519 	 * be done.
2520 	 * This needs to be done here for NFSv4, since NFSv4 never does
2521 	 * a VFS_VGET() for "." or "..".
2522 	 */
2523 	if (is_zfs == 1) {
2524 		r = VFS_VGET(mp, at.na_fileid, LK_SHARED, &nvp);
2525 		if (r == EOPNOTSUPP) {
2526 			usevget = 0;
2527 			cn.cn_nameiop = LOOKUP;
2528 			cn.cn_lkflags = LK_SHARED | LK_RETRY;
2529 			cn.cn_cred = nd->nd_cred;
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 						}
2630 						cn.cn_nameptr = dp->d_name;
2631 						cn.cn_namelen = nlen;
2632 						cn.cn_flags = ISLASTCN |
2633 						    NOFOLLOW | LOCKLEAF;
2634 						if (nlen == 2 &&
2635 						    dp->d_name[0] == '.' &&
2636 						    dp->d_name[1] == '.')
2637 							cn.cn_flags |=
2638 							    ISDOTDOT;
2639 						if (NFSVOPLOCK(vp, LK_SHARED)
2640 						    != 0) {
2641 							nd->nd_repstat = EPERM;
2642 							break;
2643 						}
2644 						if ((vp->v_vflag & VV_ROOT) != 0
2645 						    && (cn.cn_flags & ISDOTDOT)
2646 						    != 0) {
2647 							vref(vp);
2648 							nvp = vp;
2649 							r = 0;
2650 						} else {
2651 							r = VOP_LOOKUP(vp, &nvp,
2652 							    &cn);
2653 							if (vp != nvp)
2654 								NFSVOPUNLOCK(vp);
2655 						}
2656 					}
2657 
2658 					/*
2659 					 * For NFSv4, check to see if nvp is
2660 					 * a mount point and get the mount
2661 					 * point vnode, as required.
2662 					 */
2663 					if (r == 0 &&
2664 					    nfsrv_enable_crossmntpt != 0 &&
2665 					    (nd->nd_flag & ND_NFSV4) != 0 &&
2666 					    nvp->v_type == VDIR &&
2667 					    nvp->v_mountedhere != NULL) {
2668 						new_mp = nvp->v_mountedhere;
2669 						r = vfs_busy(new_mp, 0);
2670 						vput(nvp);
2671 						nvp = NULL;
2672 						if (r == 0) {
2673 							r = VFS_ROOT(new_mp,
2674 							    LK_SHARED, &nvp);
2675 							needs_unbusy = 1;
2676 							if (r == 0)
2677 								at_root = 1;
2678 						}
2679 					}
2680 				}
2681 
2682 				/*
2683 				 * If we failed to look up the entry, then it
2684 				 * has become invalid, most likely removed.
2685 				 */
2686 				if (r != 0) {
2687 					if (needs_unbusy)
2688 						vfs_unbusy(new_mp);
2689 					goto invalid;
2690 				}
2691 				KASSERT(refp != NULL || nvp != NULL,
2692 				    ("%s: undetected lookup error", __func__));
2693 
2694 				if (refp == NULL &&
2695 				    ((nd->nd_flag & ND_NFSV3) ||
2696 				     NFSNONZERO_ATTRBIT(&attrbits))) {
2697 					r = nfsvno_getfh(nvp, &nfh, p);
2698 					if (!r)
2699 					    r = nfsvno_getattr(nvp, nvap, nd, p,
2700 						1, &attrbits);
2701 					if (r == 0 && is_zfs == 1 &&
2702 					    nfsrv_enable_crossmntpt != 0 &&
2703 					    (nd->nd_flag & ND_NFSV4) != 0 &&
2704 					    nvp->v_type == VDIR &&
2705 					    vp->v_mount != nvp->v_mount) {
2706 					    /*
2707 					     * For a ZFS snapshot, there is a
2708 					     * pseudo mount that does not set
2709 					     * v_mountedhere, so it needs to
2710 					     * be detected via a different
2711 					     * mount structure.
2712 					     */
2713 					    at_root = 1;
2714 					    if (new_mp == mp)
2715 						new_mp = nvp->v_mount;
2716 					}
2717 				}
2718 
2719 				/*
2720 				 * If we failed to get attributes of the entry,
2721 				 * then just skip it for NFSv3 (the traditional
2722 				 * behavior in the old NFS server).
2723 				 * For NFSv4 the behavior is controlled by
2724 				 * RDATTRERROR: we either ignore the error or
2725 				 * fail the request.
2726 				 * Note that RDATTRERROR is never set for NFSv3.
2727 				 */
2728 				if (r != 0) {
2729 					if (!NFSISSET_ATTRBIT(&attrbits,
2730 					    NFSATTRBIT_RDATTRERROR)) {
2731 						vput(nvp);
2732 						if (needs_unbusy != 0)
2733 							vfs_unbusy(new_mp);
2734 						if ((nd->nd_flag & ND_NFSV3))
2735 							goto invalid;
2736 						nd->nd_repstat = r;
2737 						break;
2738 					}
2739 				}
2740 			}
2741 
2742 			/*
2743 			 * Build the directory record xdr
2744 			 */
2745 			if (nd->nd_flag & ND_NFSV3) {
2746 				NFSM_BUILD(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
2747 				*tl++ = newnfs_true;
2748 				*tl++ = 0;
2749 				*tl = txdr_unsigned(dp->d_fileno);
2750 				dirlen += nfsm_strtom(nd, dp->d_name, nlen);
2751 				NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
2752 				txdr_hyper(*cookiep, tl);
2753 				nfsrv_postopattr(nd, 0, nvap);
2754 				dirlen += nfsm_fhtom(nd,(u_int8_t *)&nfh,0,1);
2755 				dirlen += (5*NFSX_UNSIGNED+NFSX_V3POSTOPATTR);
2756 				if (nvp != NULL)
2757 					vput(nvp);
2758 			} else {
2759 				NFSM_BUILD(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
2760 				*tl++ = newnfs_true;
2761 				txdr_hyper(*cookiep, tl);
2762 				dirlen += nfsm_strtom(nd, dp->d_name, nlen);
2763 				if (nvp != NULL) {
2764 					supports_nfsv4acls =
2765 					    nfs_supportsnfsv4acls(nvp);
2766 					NFSVOPUNLOCK(nvp);
2767 				} else
2768 					supports_nfsv4acls = 0;
2769 				if (refp != NULL) {
2770 					dirlen += nfsrv_putreferralattr(nd,
2771 					    &savbits, refp, 0,
2772 					    &nd->nd_repstat);
2773 					if (nd->nd_repstat) {
2774 						if (nvp != NULL)
2775 							vrele(nvp);
2776 						if (needs_unbusy != 0)
2777 							vfs_unbusy(new_mp);
2778 						break;
2779 					}
2780 				} else if (r) {
2781 					dirlen += nfsvno_fillattr(nd, new_mp,
2782 					    nvp, nvap, &nfh, r, &rderrbits,
2783 					    nd->nd_cred, p, isdgram, 0,
2784 					    supports_nfsv4acls, at_root,
2785 					    mounted_on_fileno);
2786 				} else {
2787 					dirlen += nfsvno_fillattr(nd, new_mp,
2788 					    nvp, nvap, &nfh, r, &attrbits,
2789 					    nd->nd_cred, p, isdgram, 0,
2790 					    supports_nfsv4acls, at_root,
2791 					    mounted_on_fileno);
2792 				}
2793 				if (nvp != NULL)
2794 					vrele(nvp);
2795 				dirlen += (3 * NFSX_UNSIGNED);
2796 			}
2797 			if (needs_unbusy != 0)
2798 				vfs_unbusy(new_mp);
2799 			if (dirlen <= cnt)
2800 				entrycnt++;
2801 		}
2802 invalid:
2803 		cpos += dp->d_reclen;
2804 		dp = (struct dirent *)cpos;
2805 		cookiep++;
2806 		ncookies--;
2807 	}
2808 	vrele(vp);
2809 	vfs_unbusy(mp);
2810 
2811 	/*
2812 	 * If dirlen > cnt, we must strip off the last entry. If that
2813 	 * results in an empty reply, report NFSERR_TOOSMALL.
2814 	 */
2815 	if (dirlen > cnt || nd->nd_repstat) {
2816 		if (!nd->nd_repstat && entrycnt == 0)
2817 			nd->nd_repstat = NFSERR_TOOSMALL;
2818 		if (nd->nd_repstat) {
2819 			nfsm_trimtrailing(nd, mb0, bpos0, bextpg0, bextpgsiz0);
2820 			if (nd->nd_flag & ND_NFSV3)
2821 				nfsrv_postopattr(nd, getret, &at);
2822 		} else
2823 			nfsm_trimtrailing(nd, mb1, bpos1, bextpg1, bextpgsiz1);
2824 		eofflag = 0;
2825 	} else if (cpos < cend)
2826 		eofflag = 0;
2827 	if (!nd->nd_repstat) {
2828 		NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
2829 		*tl++ = newnfs_false;
2830 		if (eofflag)
2831 			*tl = newnfs_true;
2832 		else
2833 			*tl = newnfs_false;
2834 	}
2835 	free(cookies, M_TEMP);
2836 	free(rbuf, M_TEMP);
2837 
2838 out:
2839 	NFSEXITCODE2(0, nd);
2840 	return (0);
2841 nfsmout:
2842 	vput(vp);
2843 	NFSEXITCODE2(error, nd);
2844 	return (error);
2845 }
2846 
2847 /*
2848  * Get the settable attributes out of the mbuf list.
2849  * (Return 0 or EBADRPC)
2850  */
2851 int
2852 nfsrv_sattr(struct nfsrv_descript *nd, vnode_t vp, struct nfsvattr *nvap,
2853     nfsattrbit_t *attrbitp, NFSACL_T *aclp, struct thread *p)
2854 {
2855 	u_int32_t *tl;
2856 	struct nfsv2_sattr *sp;
2857 	int error = 0, toclient = 0;
2858 
2859 	switch (nd->nd_flag & (ND_NFSV2 | ND_NFSV3 | ND_NFSV4)) {
2860 	case ND_NFSV2:
2861 		NFSM_DISSECT(sp, struct nfsv2_sattr *, NFSX_V2SATTR);
2862 		/*
2863 		 * Some old clients didn't fill in the high order 16bits.
2864 		 * --> check the low order 2 bytes for 0xffff
2865 		 */
2866 		if ((fxdr_unsigned(int, sp->sa_mode) & 0xffff) != 0xffff)
2867 			nvap->na_mode = nfstov_mode(sp->sa_mode);
2868 		if (sp->sa_uid != newnfs_xdrneg1)
2869 			nvap->na_uid = fxdr_unsigned(uid_t, sp->sa_uid);
2870 		if (sp->sa_gid != newnfs_xdrneg1)
2871 			nvap->na_gid = fxdr_unsigned(gid_t, sp->sa_gid);
2872 		if (sp->sa_size != newnfs_xdrneg1)
2873 			nvap->na_size = fxdr_unsigned(u_quad_t, sp->sa_size);
2874 		if (sp->sa_atime.nfsv2_sec != newnfs_xdrneg1) {
2875 #ifdef notyet
2876 			fxdr_nfsv2time(&sp->sa_atime, &nvap->na_atime);
2877 #else
2878 			nvap->na_atime.tv_sec =
2879 				fxdr_unsigned(u_int32_t,sp->sa_atime.nfsv2_sec);
2880 			nvap->na_atime.tv_nsec = 0;
2881 #endif
2882 		}
2883 		if (sp->sa_mtime.nfsv2_sec != newnfs_xdrneg1)
2884 			fxdr_nfsv2time(&sp->sa_mtime, &nvap->na_mtime);
2885 		break;
2886 	case ND_NFSV3:
2887 		NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2888 		if (*tl == newnfs_true) {
2889 			NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2890 			nvap->na_mode = nfstov_mode(*tl);
2891 		}
2892 		NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2893 		if (*tl == newnfs_true) {
2894 			NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2895 			nvap->na_uid = fxdr_unsigned(uid_t, *tl);
2896 		}
2897 		NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2898 		if (*tl == newnfs_true) {
2899 			NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2900 			nvap->na_gid = fxdr_unsigned(gid_t, *tl);
2901 		}
2902 		NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2903 		if (*tl == newnfs_true) {
2904 			NFSM_DISSECT(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
2905 			nvap->na_size = fxdr_hyper(tl);
2906 		}
2907 		NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2908 		switch (fxdr_unsigned(int, *tl)) {
2909 		case NFSV3SATTRTIME_TOCLIENT:
2910 			NFSM_DISSECT(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
2911 			fxdr_nfsv3time(tl, &nvap->na_atime);
2912 			toclient = 1;
2913 			break;
2914 		case NFSV3SATTRTIME_TOSERVER:
2915 			vfs_timestamp(&nvap->na_atime);
2916 			nvap->na_vaflags |= VA_UTIMES_NULL;
2917 			break;
2918 		}
2919 		NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2920 		switch (fxdr_unsigned(int, *tl)) {
2921 		case NFSV3SATTRTIME_TOCLIENT:
2922 			NFSM_DISSECT(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
2923 			fxdr_nfsv3time(tl, &nvap->na_mtime);
2924 			nvap->na_vaflags &= ~VA_UTIMES_NULL;
2925 			break;
2926 		case NFSV3SATTRTIME_TOSERVER:
2927 			vfs_timestamp(&nvap->na_mtime);
2928 			if (!toclient)
2929 				nvap->na_vaflags |= VA_UTIMES_NULL;
2930 			break;
2931 		}
2932 		break;
2933 	case ND_NFSV4:
2934 		error = nfsv4_sattr(nd, vp, nvap, attrbitp, aclp, p);
2935 	}
2936 nfsmout:
2937 	NFSEXITCODE2(error, nd);
2938 	return (error);
2939 }
2940 
2941 /*
2942  * Handle the setable attributes for V4.
2943  * Returns NFSERR_BADXDR if it can't be parsed, 0 otherwise.
2944  */
2945 int
2946 nfsv4_sattr(struct nfsrv_descript *nd, vnode_t vp, struct nfsvattr *nvap,
2947     nfsattrbit_t *attrbitp, NFSACL_T *aclp, struct thread *p)
2948 {
2949 	u_int32_t *tl;
2950 	int attrsum = 0;
2951 	int i, j;
2952 	int error, attrsize, bitpos, aclsize, aceerr, retnotsup = 0;
2953 	int moderet, toclient = 0;
2954 	u_char *cp, namestr[NFSV4_SMALLSTR + 1];
2955 	uid_t uid;
2956 	gid_t gid;
2957 	u_short mode, mask;		/* Same type as va_mode. */
2958 	struct vattr va;
2959 
2960 	error = nfsrv_getattrbits(nd, attrbitp, NULL, &retnotsup);
2961 	if (error)
2962 		goto nfsmout;
2963 	NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2964 	attrsize = fxdr_unsigned(int, *tl);
2965 
2966 	/*
2967 	 * Loop around getting the setable attributes. If an unsupported
2968 	 * one is found, set nd_repstat == NFSERR_ATTRNOTSUPP and return.
2969 	 */
2970 	if (retnotsup) {
2971 		nd->nd_repstat = NFSERR_ATTRNOTSUPP;
2972 		bitpos = NFSATTRBIT_MAX;
2973 	} else {
2974 		bitpos = 0;
2975 	}
2976 	moderet = 0;
2977 	for (; bitpos < NFSATTRBIT_MAX; bitpos++) {
2978 	    if (attrsum > attrsize) {
2979 		error = NFSERR_BADXDR;
2980 		goto nfsmout;
2981 	    }
2982 	    if (NFSISSET_ATTRBIT(attrbitp, bitpos))
2983 		switch (bitpos) {
2984 		case NFSATTRBIT_SIZE:
2985 			NFSM_DISSECT(tl, u_int32_t *, NFSX_HYPER);
2986                      if (vp != NULL && vp->v_type != VREG) {
2987                             error = (vp->v_type == VDIR) ? NFSERR_ISDIR :
2988                                 NFSERR_INVAL;
2989                             goto nfsmout;
2990 			}
2991 			nvap->na_size = fxdr_hyper(tl);
2992 			attrsum += NFSX_HYPER;
2993 			break;
2994 		case NFSATTRBIT_ACL:
2995 			error = nfsrv_dissectacl(nd, aclp, true, &aceerr,
2996 			    &aclsize, p);
2997 			if (error)
2998 				goto nfsmout;
2999 			if (aceerr && !nd->nd_repstat)
3000 				nd->nd_repstat = aceerr;
3001 			attrsum += aclsize;
3002 			break;
3003 		case NFSATTRBIT_ARCHIVE:
3004 			NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
3005 			if (!nd->nd_repstat)
3006 				nd->nd_repstat = NFSERR_ATTRNOTSUPP;
3007 			attrsum += NFSX_UNSIGNED;
3008 			break;
3009 		case NFSATTRBIT_HIDDEN:
3010 			NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
3011 			if (!nd->nd_repstat)
3012 				nd->nd_repstat = NFSERR_ATTRNOTSUPP;
3013 			attrsum += NFSX_UNSIGNED;
3014 			break;
3015 		case NFSATTRBIT_MIMETYPE:
3016 			NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
3017 			i = fxdr_unsigned(int, *tl);
3018 			error = nfsm_advance(nd, NFSM_RNDUP(i), -1);
3019 			if (error)
3020 				goto nfsmout;
3021 			if (!nd->nd_repstat)
3022 				nd->nd_repstat = NFSERR_ATTRNOTSUPP;
3023 			attrsum += (NFSX_UNSIGNED + NFSM_RNDUP(i));
3024 			break;
3025 		case NFSATTRBIT_MODE:
3026 			moderet = NFSERR_INVAL;	/* Can't do MODESETMASKED. */
3027 			NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
3028 			nvap->na_mode = nfstov_mode(*tl);
3029 			attrsum += NFSX_UNSIGNED;
3030 			break;
3031 		case NFSATTRBIT_OWNER:
3032 			NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
3033 			j = fxdr_unsigned(int, *tl);
3034 			if (j < 0) {
3035 				error = NFSERR_BADXDR;
3036 				goto nfsmout;
3037 			}
3038 			if (j > NFSV4_SMALLSTR)
3039 				cp = malloc(j + 1, M_NFSSTRING, M_WAITOK);
3040 			else
3041 				cp = namestr;
3042 			error = nfsrv_mtostr(nd, cp, j);
3043 			if (error) {
3044 				if (j > NFSV4_SMALLSTR)
3045 					free(cp, M_NFSSTRING);
3046 				goto nfsmout;
3047 			}
3048 			if (!nd->nd_repstat) {
3049 				nd->nd_repstat = nfsv4_strtouid(nd, cp, j,
3050 				    &uid);
3051 				if (!nd->nd_repstat)
3052 					nvap->na_uid = uid;
3053 			}
3054 			if (j > NFSV4_SMALLSTR)
3055 				free(cp, M_NFSSTRING);
3056 			attrsum += (NFSX_UNSIGNED + NFSM_RNDUP(j));
3057 			break;
3058 		case NFSATTRBIT_OWNERGROUP:
3059 			NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
3060 			j = fxdr_unsigned(int, *tl);
3061 			if (j < 0) {
3062 				error = NFSERR_BADXDR;
3063 				goto nfsmout;
3064 			}
3065 			if (j > NFSV4_SMALLSTR)
3066 				cp = malloc(j + 1, M_NFSSTRING, M_WAITOK);
3067 			else
3068 				cp = namestr;
3069 			error = nfsrv_mtostr(nd, cp, j);
3070 			if (error) {
3071 				if (j > NFSV4_SMALLSTR)
3072 					free(cp, M_NFSSTRING);
3073 				goto nfsmout;
3074 			}
3075 			if (!nd->nd_repstat) {
3076 				nd->nd_repstat = nfsv4_strtogid(nd, cp, j,
3077 				    &gid);
3078 				if (!nd->nd_repstat)
3079 					nvap->na_gid = gid;
3080 			}
3081 			if (j > NFSV4_SMALLSTR)
3082 				free(cp, M_NFSSTRING);
3083 			attrsum += (NFSX_UNSIGNED + NFSM_RNDUP(j));
3084 			break;
3085 		case NFSATTRBIT_SYSTEM:
3086 			NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
3087 			if (!nd->nd_repstat)
3088 				nd->nd_repstat = NFSERR_ATTRNOTSUPP;
3089 			attrsum += NFSX_UNSIGNED;
3090 			break;
3091 		case NFSATTRBIT_TIMEACCESSSET:
3092 			NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
3093 			attrsum += NFSX_UNSIGNED;
3094 			if (fxdr_unsigned(int, *tl)==NFSV4SATTRTIME_TOCLIENT) {
3095 			    NFSM_DISSECT(tl, u_int32_t *, NFSX_V4TIME);
3096 			    fxdr_nfsv4time(tl, &nvap->na_atime);
3097 			    toclient = 1;
3098 			    attrsum += NFSX_V4TIME;
3099 			} else {
3100 			    vfs_timestamp(&nvap->na_atime);
3101 			    nvap->na_vaflags |= VA_UTIMES_NULL;
3102 			}
3103 			break;
3104 		case NFSATTRBIT_TIMEBACKUP:
3105 			NFSM_DISSECT(tl, u_int32_t *, NFSX_V4TIME);
3106 			if (!nd->nd_repstat)
3107 				nd->nd_repstat = NFSERR_ATTRNOTSUPP;
3108 			attrsum += NFSX_V4TIME;
3109 			break;
3110 		case NFSATTRBIT_TIMECREATE:
3111 			NFSM_DISSECT(tl, u_int32_t *, NFSX_V4TIME);
3112 			fxdr_nfsv4time(tl, &nvap->na_btime);
3113 			attrsum += NFSX_V4TIME;
3114 			break;
3115 		case NFSATTRBIT_TIMEMODIFYSET:
3116 			NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
3117 			attrsum += NFSX_UNSIGNED;
3118 			if (fxdr_unsigned(int, *tl)==NFSV4SATTRTIME_TOCLIENT) {
3119 			    NFSM_DISSECT(tl, u_int32_t *, NFSX_V4TIME);
3120 			    fxdr_nfsv4time(tl, &nvap->na_mtime);
3121 			    nvap->na_vaflags &= ~VA_UTIMES_NULL;
3122 			    attrsum += NFSX_V4TIME;
3123 			} else {
3124 			    vfs_timestamp(&nvap->na_mtime);
3125 			    if (!toclient)
3126 				nvap->na_vaflags |= VA_UTIMES_NULL;
3127 			}
3128 			break;
3129 		case NFSATTRBIT_MODESETMASKED:
3130 			NFSM_DISSECT(tl, uint32_t *, 2 * NFSX_UNSIGNED);
3131 			mode = fxdr_unsigned(u_short, *tl++);
3132 			mask = fxdr_unsigned(u_short, *tl);
3133 			/*
3134 			 * vp == NULL implies an Open/Create operation.
3135 			 * This attribute can only be used for Setattr and
3136 			 * only for NFSv4.1 or higher.
3137 			 * If moderet != 0, a mode attribute has also been
3138 			 * specified and this attribute cannot be done in the
3139 			 * same Setattr operation.
3140 			 */
3141 			if ((nd->nd_flag & ND_NFSV41) == 0)
3142 				nd->nd_repstat = NFSERR_ATTRNOTSUPP;
3143 			else if ((mode & ~07777) != 0 || (mask & ~07777) != 0 ||
3144 			    vp == NULL)
3145 				nd->nd_repstat = NFSERR_INVAL;
3146 			else if (moderet == 0)
3147 				moderet = VOP_GETATTR(vp, &va, nd->nd_cred);
3148 			if (moderet == 0)
3149 				nvap->na_mode = (mode & mask) |
3150 				    (va.va_mode & ~mask);
3151 			else
3152 				nd->nd_repstat = moderet;
3153 			attrsum += 2 * NFSX_UNSIGNED;
3154 			break;
3155 		default:
3156 			nd->nd_repstat = NFSERR_ATTRNOTSUPP;
3157 			/*
3158 			 * set bitpos so we drop out of the loop.
3159 			 */
3160 			bitpos = NFSATTRBIT_MAX;
3161 			break;
3162 		}
3163 	}
3164 
3165 	/*
3166 	 * some clients pad the attrlist, so we need to skip over the
3167 	 * padding.
3168 	 */
3169 	if (attrsum > attrsize) {
3170 		error = NFSERR_BADXDR;
3171 	} else {
3172 		attrsize = NFSM_RNDUP(attrsize);
3173 		if (attrsum < attrsize)
3174 			error = nfsm_advance(nd, attrsize - attrsum, -1);
3175 	}
3176 nfsmout:
3177 	NFSEXITCODE2(error, nd);
3178 	return (error);
3179 }
3180 
3181 /*
3182  * Check/setup export credentials.
3183  */
3184 int
3185 nfsd_excred(struct nfsrv_descript *nd, struct nfsexstuff *exp,
3186     struct ucred *credanon, bool testsec)
3187 {
3188 	int error;
3189 
3190 	/*
3191 	 * Check/setup credentials.
3192 	 */
3193 	if (nd->nd_flag & ND_GSS)
3194 		exp->nes_exflag &= ~MNT_EXPORTANON;
3195 
3196 	/*
3197 	 * Check to see if the operation is allowed for this security flavor.
3198 	 */
3199 	error = 0;
3200 	if (testsec) {
3201 		error = nfsvno_testexp(nd, exp);
3202 		if (error != 0)
3203 			goto out;
3204 	}
3205 
3206 	/*
3207 	 * Check to see if the file system is exported V4 only.
3208 	 */
3209 	if (NFSVNO_EXV4ONLY(exp) && !(nd->nd_flag & ND_NFSV4)) {
3210 		error = NFSERR_PROGNOTV4;
3211 		goto out;
3212 	}
3213 
3214 	/*
3215 	 * Now, map the user credentials.
3216 	 * (Note that ND_AUTHNONE will only be set for an NFSv3
3217 	 *  Fsinfo RPC. If set for anything else, this code might need
3218 	 *  to change.)
3219 	 */
3220 	if (NFSVNO_EXPORTED(exp)) {
3221 		if (((nd->nd_flag & ND_GSS) == 0 && nd->nd_cred->cr_uid == 0) ||
3222 		     NFSVNO_EXPORTANON(exp) ||
3223 		     (nd->nd_flag & ND_AUTHNONE) != 0) {
3224 			nd->nd_cred->cr_uid = credanon->cr_uid;
3225 			nd->nd_cred->cr_gid = credanon->cr_gid;
3226 			crsetgroups(nd->nd_cred, credanon->cr_ngroups,
3227 			    credanon->cr_groups);
3228 		} else if ((nd->nd_flag & ND_GSS) == 0) {
3229 			/*
3230 			 * If using AUTH_SYS, call nfsrv_getgrpscred() to see
3231 			 * if there is a replacement credential with a group
3232 			 * list set up by "nfsuserd -manage-gids".
3233 			 * If there is no replacement, nfsrv_getgrpscred()
3234 			 * simply returns its argument.
3235 			 */
3236 			nd->nd_cred = nfsrv_getgrpscred(nd->nd_cred);
3237 		}
3238 	}
3239 
3240 out:
3241 	NFSEXITCODE2(error, nd);
3242 	return (error);
3243 }
3244 
3245 /*
3246  * Check exports.
3247  */
3248 int
3249 nfsvno_checkexp(struct mount *mp, struct sockaddr *nam, struct nfsexstuff *exp,
3250     struct ucred **credp)
3251 {
3252 	int error;
3253 
3254 	error = VFS_CHECKEXP(mp, nam, &exp->nes_exflag, credp,
3255 	    &exp->nes_numsecflavor, exp->nes_secflavors);
3256 	if (error) {
3257 		if (nfs_rootfhset) {
3258 			exp->nes_exflag = 0;
3259 			exp->nes_numsecflavor = 0;
3260 			error = 0;
3261 		}
3262 	} else if (exp->nes_numsecflavor < 1 || exp->nes_numsecflavor >
3263 	    MAXSECFLAVORS) {
3264 		printf("nfsvno_checkexp: numsecflavors out of range\n");
3265 		exp->nes_numsecflavor = 0;
3266 		error = EACCES;
3267 	}
3268 	NFSEXITCODE(error);
3269 	return (error);
3270 }
3271 
3272 /*
3273  * Get a vnode for a file handle and export stuff.
3274  */
3275 int
3276 nfsvno_fhtovp(struct mount *mp, fhandle_t *fhp, struct sockaddr *nam,
3277     int lktype, struct vnode **vpp, struct nfsexstuff *exp,
3278     struct ucred **credp)
3279 {
3280 	int error;
3281 
3282 	*credp = NULL;
3283 	exp->nes_numsecflavor = 0;
3284 	error = VFS_FHTOVP(mp, &fhp->fh_fid, lktype, vpp);
3285 	if (error != 0)
3286 		/* Make sure the server replies ESTALE to the client. */
3287 		error = ESTALE;
3288 	if (nam && !error) {
3289 		error = VFS_CHECKEXP(mp, nam, &exp->nes_exflag, credp,
3290 		    &exp->nes_numsecflavor, exp->nes_secflavors);
3291 		if (error) {
3292 			if (nfs_rootfhset) {
3293 				exp->nes_exflag = 0;
3294 				exp->nes_numsecflavor = 0;
3295 				error = 0;
3296 			} else {
3297 				vput(*vpp);
3298 			}
3299 		} else if (exp->nes_numsecflavor < 1 || exp->nes_numsecflavor >
3300 		    MAXSECFLAVORS) {
3301 			printf("nfsvno_fhtovp: numsecflavors out of range\n");
3302 			exp->nes_numsecflavor = 0;
3303 			error = EACCES;
3304 			vput(*vpp);
3305 		}
3306 	}
3307 	NFSEXITCODE(error);
3308 	return (error);
3309 }
3310 
3311 /*
3312  * nfsd_fhtovp() - convert a fh to a vnode ptr
3313  * 	- look up fsid in mount list (if not found ret error)
3314  *	- get vp and export rights by calling nfsvno_fhtovp()
3315  *	- if cred->cr_uid == 0 or MNT_EXPORTANON set it to credanon
3316  *	  for AUTH_SYS
3317  *	- if mpp != NULL, return the mount point so that it can
3318  *	  be used for vn_finished_write() by the caller
3319  */
3320 void
3321 nfsd_fhtovp(struct nfsrv_descript *nd, struct nfsrvfh *nfp, int lktype,
3322     struct vnode **vpp, struct nfsexstuff *exp,
3323     struct mount **mpp, int startwrite, int nextop)
3324 {
3325 	struct mount *mp, *mpw;
3326 	struct ucred *credanon;
3327 	fhandle_t *fhp;
3328 	int error;
3329 
3330 	if (mpp != NULL)
3331 		*mpp = NULL;
3332 	*vpp = NULL;
3333 	fhp = (fhandle_t *)nfp->nfsrvfh_data;
3334 	mp = vfs_busyfs(&fhp->fh_fsid);
3335 	if (mp == NULL) {
3336 		nd->nd_repstat = ESTALE;
3337 		goto out;
3338 	}
3339 
3340 	if (startwrite) {
3341 		mpw = mp;
3342 		error = vn_start_write(NULL, &mpw, V_WAIT);
3343 		if (error != 0) {
3344 			mpw = NULL;
3345 			vfs_unbusy(mp);
3346 			nd->nd_repstat = ESTALE;
3347 			goto out;
3348 		}
3349 		if (lktype == LK_SHARED && !(MNT_SHARED_WRITES(mp)))
3350 			lktype = LK_EXCLUSIVE;
3351 	} else
3352 		mpw = NULL;
3353 
3354 	nd->nd_repstat = nfsvno_fhtovp(mp, fhp, nd->nd_nam, lktype, vpp, exp,
3355 	    &credanon);
3356 	vfs_unbusy(mp);
3357 
3358 	/*
3359 	 * For NFSv4 without a pseudo root fs, unexported file handles
3360 	 * can be returned, so that Lookup works everywhere.
3361 	 */
3362 	if (!nd->nd_repstat && exp->nes_exflag == 0 &&
3363 	    !(nd->nd_flag & ND_NFSV4)) {
3364 		vput(*vpp);
3365 		*vpp = NULL;
3366 		nd->nd_repstat = EACCES;
3367 	}
3368 
3369 	/*
3370 	 * Personally, I've never seen any point in requiring a
3371 	 * reserved port#, since only in the rare case where the
3372 	 * clients are all boxes with secure system privileges,
3373 	 * does it provide any enhanced security, but... some people
3374 	 * believe it to be useful and keep putting this code back in.
3375 	 * (There is also some "security checker" out there that
3376 	 *  complains if the nfs server doesn't enforce this.)
3377 	 * However, note the following:
3378 	 * RFC3530 (NFSv4) specifies that a reserved port# not be
3379 	 *	required.
3380 	 * RFC2623 recommends that, if a reserved port# is checked for,
3381 	 *	that there be a way to turn that off--> ifdef'd.
3382 	 */
3383 #ifdef NFS_REQRSVPORT
3384 	if (!nd->nd_repstat) {
3385 		struct sockaddr_in *saddr;
3386 		struct sockaddr_in6 *saddr6;
3387 
3388 		saddr = NFSSOCKADDR(nd->nd_nam, struct sockaddr_in *);
3389 		saddr6 = NFSSOCKADDR(nd->nd_nam, struct sockaddr_in6 *);
3390 		if (!(nd->nd_flag & ND_NFSV4) &&
3391 		    ((saddr->sin_family == AF_INET &&
3392 		      ntohs(saddr->sin_port) >= IPPORT_RESERVED) ||
3393 		     (saddr6->sin6_family == AF_INET6 &&
3394 		      ntohs(saddr6->sin6_port) >= IPPORT_RESERVED))) {
3395 			vput(*vpp);
3396 			nd->nd_repstat = (NFSERR_AUTHERR | AUTH_TOOWEAK);
3397 		}
3398 	}
3399 #endif	/* NFS_REQRSVPORT */
3400 
3401 	/*
3402 	 * Check/setup credentials.
3403 	 */
3404 	if (!nd->nd_repstat) {
3405 		nd->nd_saveduid = nd->nd_cred->cr_uid;
3406 		nd->nd_repstat = nfsd_excred(nd, exp, credanon,
3407 		    nfsrv_checkwrongsec(nd, nextop, (*vpp)->v_type));
3408 		if (nd->nd_repstat)
3409 			vput(*vpp);
3410 	}
3411 	if (credanon != NULL)
3412 		crfree(credanon);
3413 	if (nd->nd_repstat) {
3414 		vn_finished_write(mpw);
3415 		*vpp = NULL;
3416 	} else if (mpp != NULL) {
3417 		*mpp = mpw;
3418 	}
3419 
3420 out:
3421 	NFSEXITCODE2(0, nd);
3422 }
3423 
3424 /*
3425  * glue for fp.
3426  */
3427 static int
3428 fp_getfvp(struct thread *p, int fd, struct file **fpp, struct vnode **vpp)
3429 {
3430 	struct filedesc *fdp;
3431 	struct file *fp;
3432 	int error = 0;
3433 
3434 	fdp = p->td_proc->p_fd;
3435 	if (fd < 0 || fd >= fdp->fd_nfiles ||
3436 	    (fp = fdp->fd_ofiles[fd].fde_file) == NULL) {
3437 		error = EBADF;
3438 		goto out;
3439 	}
3440 	*fpp = fp;
3441 
3442 out:
3443 	NFSEXITCODE(error);
3444 	return (error);
3445 }
3446 
3447 /*
3448  * Called from nfssvc() to update the exports list. Just call
3449  * vfs_export(). This has to be done, since the v4 root fake fs isn't
3450  * in the mount list.
3451  */
3452 int
3453 nfsrv_v4rootexport(void *argp, struct ucred *cred, struct thread *p)
3454 {
3455 	struct nfsex_args *nfsexargp = (struct nfsex_args *)argp;
3456 	int error = 0;
3457 	struct nameidata nd;
3458 	fhandle_t fh;
3459 
3460 	error = vfs_export(&nfsv4root_mnt, &nfsexargp->export);
3461 	if ((nfsexargp->export.ex_flags & MNT_DELEXPORT) != 0)
3462 		nfs_rootfhset = 0;
3463 	else if (error == 0) {
3464 		if (nfsexargp->fspec == NULL) {
3465 			error = EPERM;
3466 			goto out;
3467 		}
3468 		/*
3469 		 * If fspec != NULL, this is the v4root path.
3470 		 */
3471 		NDINIT(&nd, LOOKUP, FOLLOW, UIO_USERSPACE, nfsexargp->fspec);
3472 		if ((error = namei(&nd)) != 0)
3473 			goto out;
3474 		error = nfsvno_getfh(nd.ni_vp, &fh, p);
3475 		vrele(nd.ni_vp);
3476 		if (!error) {
3477 			nfs_rootfh.nfsrvfh_len = NFSX_MYFH;
3478 			NFSBCOPY((caddr_t)&fh,
3479 			    nfs_rootfh.nfsrvfh_data,
3480 			    sizeof (fhandle_t));
3481 			nfs_rootfhset = 1;
3482 		}
3483 	}
3484 
3485 out:
3486 	NFSEXITCODE(error);
3487 	return (error);
3488 }
3489 
3490 /*
3491  * This function needs to test to see if the system is near its limit
3492  * for memory allocation via malloc() or mget() and return True iff
3493  * either of these resources are near their limit.
3494  * XXX (For now, this is just a stub.)
3495  */
3496 int nfsrv_testmalloclimit = 0;
3497 int
3498 nfsrv_mallocmget_limit(void)
3499 {
3500 	static int printmesg = 0;
3501 	static int testval = 1;
3502 
3503 	if (nfsrv_testmalloclimit && (testval++ % 1000) == 0) {
3504 		if ((printmesg++ % 100) == 0)
3505 			printf("nfsd: malloc/mget near limit\n");
3506 		return (1);
3507 	}
3508 	return (0);
3509 }
3510 
3511 /*
3512  * BSD specific initialization of a mount point.
3513  */
3514 void
3515 nfsd_mntinit(void)
3516 {
3517 	static int inited = 0;
3518 
3519 	if (inited)
3520 		return;
3521 	inited = 1;
3522 	nfsv4root_mnt.mnt_flag = (MNT_RDONLY | MNT_EXPORTED);
3523 	TAILQ_INIT(&nfsv4root_mnt.mnt_nvnodelist);
3524 	TAILQ_INIT(&nfsv4root_mnt.mnt_lazyvnodelist);
3525 	nfsv4root_mnt.mnt_export = NULL;
3526 	TAILQ_INIT(&nfsv4root_opt);
3527 	TAILQ_INIT(&nfsv4root_newopt);
3528 	nfsv4root_mnt.mnt_opt = &nfsv4root_opt;
3529 	nfsv4root_mnt.mnt_optnew = &nfsv4root_newopt;
3530 	nfsv4root_mnt.mnt_nvnodelistsize = 0;
3531 	nfsv4root_mnt.mnt_lazyvnodelistsize = 0;
3532 }
3533 
3534 static void
3535 nfsd_timer(void *arg)
3536 {
3537 
3538 	nfsrv_servertimer();
3539 	callout_reset_sbt(&nfsd_callout, SBT_1S, SBT_1S, nfsd_timer, NULL, 0);
3540 }
3541 
3542 /*
3543  * Get a vnode for a file handle, without checking exports, etc.
3544  */
3545 struct vnode *
3546 nfsvno_getvp(fhandle_t *fhp)
3547 {
3548 	struct mount *mp;
3549 	struct vnode *vp;
3550 	int error;
3551 
3552 	mp = vfs_busyfs(&fhp->fh_fsid);
3553 	if (mp == NULL)
3554 		return (NULL);
3555 	error = VFS_FHTOVP(mp, &fhp->fh_fid, LK_EXCLUSIVE, &vp);
3556 	vfs_unbusy(mp);
3557 	if (error)
3558 		return (NULL);
3559 	return (vp);
3560 }
3561 
3562 /*
3563  * Do a local VOP_ADVLOCK().
3564  */
3565 int
3566 nfsvno_advlock(struct vnode *vp, int ftype, u_int64_t first,
3567     u_int64_t end, struct thread *td)
3568 {
3569 	int error = 0;
3570 	struct flock fl;
3571 	u_int64_t tlen;
3572 
3573 	if (nfsrv_dolocallocks == 0)
3574 		goto out;
3575 	ASSERT_VOP_UNLOCKED(vp, "nfsvno_advlock: vp locked");
3576 
3577 	fl.l_whence = SEEK_SET;
3578 	fl.l_type = ftype;
3579 	fl.l_start = (off_t)first;
3580 	if (end == NFS64BITSSET) {
3581 		fl.l_len = 0;
3582 	} else {
3583 		tlen = end - first;
3584 		fl.l_len = (off_t)tlen;
3585 	}
3586 	/*
3587 	 * For FreeBSD8, the l_pid and l_sysid must be set to the same
3588 	 * values for all calls, so that all locks will be held by the
3589 	 * nfsd server. (The nfsd server handles conflicts between the
3590 	 * various clients.)
3591 	 * Since an NFSv4 lockowner is a ClientID plus an array of up to 1024
3592 	 * bytes, so it can't be put in l_sysid.
3593 	 */
3594 	if (nfsv4_sysid == 0)
3595 		nfsv4_sysid = nlm_acquire_next_sysid();
3596 	fl.l_pid = (pid_t)0;
3597 	fl.l_sysid = (int)nfsv4_sysid;
3598 
3599 	if (ftype == F_UNLCK)
3600 		error = VOP_ADVLOCK(vp, (caddr_t)td->td_proc, F_UNLCK, &fl,
3601 		    (F_POSIX | F_REMOTE));
3602 	else
3603 		error = VOP_ADVLOCK(vp, (caddr_t)td->td_proc, F_SETLK, &fl,
3604 		    (F_POSIX | F_REMOTE));
3605 
3606 out:
3607 	NFSEXITCODE(error);
3608 	return (error);
3609 }
3610 
3611 /*
3612  * Check the nfsv4 root exports.
3613  */
3614 int
3615 nfsvno_v4rootexport(struct nfsrv_descript *nd)
3616 {
3617 	struct ucred *credanon;
3618 	int error = 0, numsecflavor, secflavors[MAXSECFLAVORS], i;
3619 	uint64_t exflags;
3620 
3621 	error = vfs_stdcheckexp(&nfsv4root_mnt, nd->nd_nam, &exflags,
3622 	    &credanon, &numsecflavor, secflavors);
3623 	if (error) {
3624 		error = NFSERR_PROGUNAVAIL;
3625 		goto out;
3626 	}
3627 	if (credanon != NULL)
3628 		crfree(credanon);
3629 	for (i = 0; i < numsecflavor; i++) {
3630 		if (secflavors[i] == AUTH_SYS)
3631 			nd->nd_flag |= ND_EXAUTHSYS;
3632 		else if (secflavors[i] == RPCSEC_GSS_KRB5)
3633 			nd->nd_flag |= ND_EXGSS;
3634 		else if (secflavors[i] == RPCSEC_GSS_KRB5I)
3635 			nd->nd_flag |= ND_EXGSSINTEGRITY;
3636 		else if (secflavors[i] == RPCSEC_GSS_KRB5P)
3637 			nd->nd_flag |= ND_EXGSSPRIVACY;
3638 	}
3639 
3640 	/* And set ND_EXxx flags for TLS. */
3641 	if ((exflags & MNT_EXTLS) != 0) {
3642 		nd->nd_flag |= ND_EXTLS;
3643 		if ((exflags & MNT_EXTLSCERT) != 0)
3644 			nd->nd_flag |= ND_EXTLSCERT;
3645 		if ((exflags & MNT_EXTLSCERTUSER) != 0)
3646 			nd->nd_flag |= ND_EXTLSCERTUSER;
3647 	}
3648 
3649 out:
3650 	NFSEXITCODE(error);
3651 	return (error);
3652 }
3653 
3654 /*
3655  * Nfs server pseudo system call for the nfsd's
3656  */
3657 /*
3658  * MPSAFE
3659  */
3660 static int
3661 nfssvc_nfsd(struct thread *td, struct nfssvc_args *uap)
3662 {
3663 	struct file *fp;
3664 	struct nfsd_addsock_args sockarg;
3665 	struct nfsd_nfsd_args nfsdarg;
3666 	struct nfsd_nfsd_oargs onfsdarg;
3667 	struct nfsd_pnfsd_args pnfsdarg;
3668 	struct vnode *vp, *nvp, *curdvp;
3669 	struct pnfsdsfile *pf;
3670 	struct nfsdevice *ds, *fds;
3671 	cap_rights_t rights;
3672 	int buflen, error, ret;
3673 	char *buf, *cp, *cp2, *cp3;
3674 	char fname[PNFS_FILENAME_LEN + 1];
3675 
3676 	if (uap->flag & NFSSVC_NFSDADDSOCK) {
3677 		error = copyin(uap->argp, (caddr_t)&sockarg, sizeof (sockarg));
3678 		if (error)
3679 			goto out;
3680 		/*
3681 		 * Since we don't know what rights might be required,
3682 		 * pretend that we need them all. It is better to be too
3683 		 * careful than too reckless.
3684 		 */
3685 		error = fget(td, sockarg.sock,
3686 		    cap_rights_init_one(&rights, CAP_SOCK_SERVER), &fp);
3687 		if (error != 0)
3688 			goto out;
3689 		if (fp->f_type != DTYPE_SOCKET) {
3690 			fdrop(fp, td);
3691 			error = EPERM;
3692 			goto out;
3693 		}
3694 		error = nfsrvd_addsock(fp);
3695 		fdrop(fp, td);
3696 	} else if (uap->flag & NFSSVC_NFSDNFSD) {
3697 		if (uap->argp == NULL) {
3698 			error = EINVAL;
3699 			goto out;
3700 		}
3701 		if ((uap->flag & NFSSVC_NEWSTRUCT) == 0) {
3702 			error = copyin(uap->argp, &onfsdarg, sizeof(onfsdarg));
3703 			if (error == 0) {
3704 				nfsdarg.principal = onfsdarg.principal;
3705 				nfsdarg.minthreads = onfsdarg.minthreads;
3706 				nfsdarg.maxthreads = onfsdarg.maxthreads;
3707 				nfsdarg.version = 1;
3708 				nfsdarg.addr = NULL;
3709 				nfsdarg.addrlen = 0;
3710 				nfsdarg.dnshost = NULL;
3711 				nfsdarg.dnshostlen = 0;
3712 				nfsdarg.dspath = NULL;
3713 				nfsdarg.dspathlen = 0;
3714 				nfsdarg.mdspath = NULL;
3715 				nfsdarg.mdspathlen = 0;
3716 				nfsdarg.mirrorcnt = 1;
3717 			}
3718 		} else
3719 			error = copyin(uap->argp, &nfsdarg, sizeof(nfsdarg));
3720 		if (error)
3721 			goto out;
3722 		if (nfsdarg.addrlen > 0 && nfsdarg.addrlen < 10000 &&
3723 		    nfsdarg.dnshostlen > 0 && nfsdarg.dnshostlen < 10000 &&
3724 		    nfsdarg.dspathlen > 0 && nfsdarg.dspathlen < 10000 &&
3725 		    nfsdarg.mdspathlen > 0 && nfsdarg.mdspathlen < 10000 &&
3726 		    nfsdarg.mirrorcnt >= 1 &&
3727 		    nfsdarg.mirrorcnt <= NFSDEV_MAXMIRRORS &&
3728 		    nfsdarg.addr != NULL && nfsdarg.dnshost != NULL &&
3729 		    nfsdarg.dspath != NULL && nfsdarg.mdspath != NULL) {
3730 			NFSD_DEBUG(1, "addrlen=%d dspathlen=%d dnslen=%d"
3731 			    " mdspathlen=%d mirrorcnt=%d\n", nfsdarg.addrlen,
3732 			    nfsdarg.dspathlen, nfsdarg.dnshostlen,
3733 			    nfsdarg.mdspathlen, nfsdarg.mirrorcnt);
3734 			cp = malloc(nfsdarg.addrlen + 1, M_TEMP, M_WAITOK);
3735 			error = copyin(nfsdarg.addr, cp, nfsdarg.addrlen);
3736 			if (error != 0) {
3737 				free(cp, M_TEMP);
3738 				goto out;
3739 			}
3740 			cp[nfsdarg.addrlen] = '\0';	/* Ensure nul term. */
3741 			nfsdarg.addr = cp;
3742 			cp = malloc(nfsdarg.dnshostlen + 1, M_TEMP, M_WAITOK);
3743 			error = copyin(nfsdarg.dnshost, cp, nfsdarg.dnshostlen);
3744 			if (error != 0) {
3745 				free(nfsdarg.addr, M_TEMP);
3746 				free(cp, M_TEMP);
3747 				goto out;
3748 			}
3749 			cp[nfsdarg.dnshostlen] = '\0';	/* Ensure nul term. */
3750 			nfsdarg.dnshost = cp;
3751 			cp = malloc(nfsdarg.dspathlen + 1, M_TEMP, M_WAITOK);
3752 			error = copyin(nfsdarg.dspath, cp, nfsdarg.dspathlen);
3753 			if (error != 0) {
3754 				free(nfsdarg.addr, M_TEMP);
3755 				free(nfsdarg.dnshost, M_TEMP);
3756 				free(cp, M_TEMP);
3757 				goto out;
3758 			}
3759 			cp[nfsdarg.dspathlen] = '\0';	/* Ensure nul term. */
3760 			nfsdarg.dspath = cp;
3761 			cp = malloc(nfsdarg.mdspathlen + 1, M_TEMP, M_WAITOK);
3762 			error = copyin(nfsdarg.mdspath, cp, nfsdarg.mdspathlen);
3763 			if (error != 0) {
3764 				free(nfsdarg.addr, M_TEMP);
3765 				free(nfsdarg.dnshost, M_TEMP);
3766 				free(nfsdarg.dspath, M_TEMP);
3767 				free(cp, M_TEMP);
3768 				goto out;
3769 			}
3770 			cp[nfsdarg.mdspathlen] = '\0';	/* Ensure nul term. */
3771 			nfsdarg.mdspath = cp;
3772 		} else {
3773 			nfsdarg.addr = NULL;
3774 			nfsdarg.addrlen = 0;
3775 			nfsdarg.dnshost = NULL;
3776 			nfsdarg.dnshostlen = 0;
3777 			nfsdarg.dspath = NULL;
3778 			nfsdarg.dspathlen = 0;
3779 			nfsdarg.mdspath = NULL;
3780 			nfsdarg.mdspathlen = 0;
3781 			nfsdarg.mirrorcnt = 1;
3782 		}
3783 		nfsd_timer(NULL);
3784 		error = nfsrvd_nfsd(td, &nfsdarg);
3785 		free(nfsdarg.addr, M_TEMP);
3786 		free(nfsdarg.dnshost, M_TEMP);
3787 		free(nfsdarg.dspath, M_TEMP);
3788 		free(nfsdarg.mdspath, M_TEMP);
3789 	} else if (uap->flag & NFSSVC_PNFSDS) {
3790 		error = copyin(uap->argp, &pnfsdarg, sizeof(pnfsdarg));
3791 		if (error == 0 && (pnfsdarg.op == PNFSDOP_DELDSSERVER ||
3792 		    pnfsdarg.op == PNFSDOP_FORCEDELDS)) {
3793 			cp = malloc(PATH_MAX + 1, M_TEMP, M_WAITOK);
3794 			error = copyinstr(pnfsdarg.dspath, cp, PATH_MAX + 1,
3795 			    NULL);
3796 			if (error == 0)
3797 				error = nfsrv_deldsserver(pnfsdarg.op, cp, td);
3798 			free(cp, M_TEMP);
3799 		} else if (error == 0 && pnfsdarg.op == PNFSDOP_COPYMR) {
3800 			cp = malloc(PATH_MAX + 1, M_TEMP, M_WAITOK);
3801 			buflen = sizeof(*pf) * NFSDEV_MAXMIRRORS;
3802 			buf = malloc(buflen, M_TEMP, M_WAITOK);
3803 			error = copyinstr(pnfsdarg.mdspath, cp, PATH_MAX + 1,
3804 			    NULL);
3805 			NFSD_DEBUG(4, "pnfsdcopymr cp mdspath=%d\n", error);
3806 			if (error == 0 && pnfsdarg.dspath != NULL) {
3807 				cp2 = malloc(PATH_MAX + 1, M_TEMP, M_WAITOK);
3808 				error = copyinstr(pnfsdarg.dspath, cp2,
3809 				    PATH_MAX + 1, NULL);
3810 				NFSD_DEBUG(4, "pnfsdcopymr cp dspath=%d\n",
3811 				    error);
3812 			} else
3813 				cp2 = NULL;
3814 			if (error == 0 && pnfsdarg.curdspath != NULL) {
3815 				cp3 = malloc(PATH_MAX + 1, M_TEMP, M_WAITOK);
3816 				error = copyinstr(pnfsdarg.curdspath, cp3,
3817 				    PATH_MAX + 1, NULL);
3818 				NFSD_DEBUG(4, "pnfsdcopymr cp curdspath=%d\n",
3819 				    error);
3820 			} else
3821 				cp3 = NULL;
3822 			curdvp = NULL;
3823 			fds = NULL;
3824 			if (error == 0)
3825 				error = nfsrv_mdscopymr(cp, cp2, cp3, buf,
3826 				    &buflen, fname, td, &vp, &nvp, &pf, &ds,
3827 				    &fds);
3828 			NFSD_DEBUG(4, "nfsrv_mdscopymr=%d\n", error);
3829 			if (error == 0) {
3830 				if (pf->dsf_dir >= nfsrv_dsdirsize) {
3831 					printf("copymr: dsdir out of range\n");
3832 					pf->dsf_dir = 0;
3833 				}
3834 				NFSD_DEBUG(4, "copymr: buflen=%d\n", buflen);
3835 				error = nfsrv_copymr(vp, nvp,
3836 				    ds->nfsdev_dsdir[pf->dsf_dir], ds, pf,
3837 				    (struct pnfsdsfile *)buf,
3838 				    buflen / sizeof(*pf), td->td_ucred, td);
3839 				vput(vp);
3840 				vput(nvp);
3841 				if (fds != NULL && error == 0) {
3842 					curdvp = fds->nfsdev_dsdir[pf->dsf_dir];
3843 					ret = vn_lock(curdvp, LK_EXCLUSIVE);
3844 					if (ret == 0) {
3845 						nfsrv_dsremove(curdvp, fname,
3846 						    td->td_ucred, td);
3847 						NFSVOPUNLOCK(curdvp);
3848 					}
3849 				}
3850 				NFSD_DEBUG(4, "nfsrv_copymr=%d\n", error);
3851 			}
3852 			free(cp, M_TEMP);
3853 			free(cp2, M_TEMP);
3854 			free(cp3, M_TEMP);
3855 			free(buf, M_TEMP);
3856 		}
3857 	} else {
3858 		error = nfssvc_srvcall(td, uap, td->td_ucred);
3859 	}
3860 
3861 out:
3862 	NFSEXITCODE(error);
3863 	return (error);
3864 }
3865 
3866 static int
3867 nfssvc_srvcall(struct thread *p, struct nfssvc_args *uap, struct ucred *cred)
3868 {
3869 	struct nfsex_args export;
3870 	struct nfsex_oldargs oexp;
3871 	struct file *fp = NULL;
3872 	int stablefd, i, len;
3873 	struct nfsd_clid adminrevoke;
3874 	struct nfsd_dumplist dumplist;
3875 	struct nfsd_dumpclients *dumpclients;
3876 	struct nfsd_dumplocklist dumplocklist;
3877 	struct nfsd_dumplocks *dumplocks;
3878 	struct nameidata nd;
3879 	vnode_t vp;
3880 	int error = EINVAL, igotlock;
3881 	struct proc *procp;
3882 	gid_t *grps;
3883 	static int suspend_nfsd = 0;
3884 
3885 	if (uap->flag & NFSSVC_PUBLICFH) {
3886 		NFSBZERO((caddr_t)&nfs_pubfh.nfsrvfh_data,
3887 		    sizeof (fhandle_t));
3888 		error = copyin(uap->argp,
3889 		    &nfs_pubfh.nfsrvfh_data, sizeof (fhandle_t));
3890 		if (!error)
3891 			nfs_pubfhset = 1;
3892 	} else if ((uap->flag & (NFSSVC_V4ROOTEXPORT | NFSSVC_NEWSTRUCT)) ==
3893 	    (NFSSVC_V4ROOTEXPORT | NFSSVC_NEWSTRUCT)) {
3894 		error = copyin(uap->argp,(caddr_t)&export,
3895 		    sizeof (struct nfsex_args));
3896 		if (!error) {
3897 			grps = NULL;
3898 			if (export.export.ex_ngroups > NGROUPS_MAX ||
3899 			    export.export.ex_ngroups < 0)
3900 				error = EINVAL;
3901 			else if (export.export.ex_ngroups > 0) {
3902 				grps = malloc(export.export.ex_ngroups *
3903 				    sizeof(gid_t), M_TEMP, M_WAITOK);
3904 				error = copyin(export.export.ex_groups, grps,
3905 				    export.export.ex_ngroups * sizeof(gid_t));
3906 				export.export.ex_groups = grps;
3907 			} else
3908 				export.export.ex_groups = NULL;
3909 			if (!error)
3910 				error = nfsrv_v4rootexport(&export, cred, p);
3911 			free(grps, M_TEMP);
3912 		}
3913 	} else if ((uap->flag & (NFSSVC_V4ROOTEXPORT | NFSSVC_NEWSTRUCT)) ==
3914 	    NFSSVC_V4ROOTEXPORT) {
3915 		error = copyin(uap->argp,(caddr_t)&oexp,
3916 		    sizeof (struct nfsex_oldargs));
3917 		if (!error) {
3918 			memset(&export.export, 0, sizeof(export.export));
3919 			export.export.ex_flags = (uint64_t)oexp.export.ex_flags;
3920 			export.export.ex_root = oexp.export.ex_root;
3921 			export.export.ex_uid = oexp.export.ex_anon.cr_uid;
3922 			export.export.ex_ngroups =
3923 			    oexp.export.ex_anon.cr_ngroups;
3924 			export.export.ex_groups = NULL;
3925 			if (export.export.ex_ngroups > XU_NGROUPS ||
3926 			    export.export.ex_ngroups < 0)
3927 				error = EINVAL;
3928 			else if (export.export.ex_ngroups > 0) {
3929 				export.export.ex_groups = malloc(
3930 				    export.export.ex_ngroups * sizeof(gid_t),
3931 				    M_TEMP, M_WAITOK);
3932 				for (i = 0; i < export.export.ex_ngroups; i++)
3933 					export.export.ex_groups[i] =
3934 					    oexp.export.ex_anon.cr_groups[i];
3935 			}
3936 			export.export.ex_addr = oexp.export.ex_addr;
3937 			export.export.ex_addrlen = oexp.export.ex_addrlen;
3938 			export.export.ex_mask = oexp.export.ex_mask;
3939 			export.export.ex_masklen = oexp.export.ex_masklen;
3940 			export.export.ex_indexfile = oexp.export.ex_indexfile;
3941 			export.export.ex_numsecflavors =
3942 			    oexp.export.ex_numsecflavors;
3943 			if (export.export.ex_numsecflavors >= MAXSECFLAVORS ||
3944 			    export.export.ex_numsecflavors < 0)
3945 				error = EINVAL;
3946 			else {
3947 				for (i = 0; i < export.export.ex_numsecflavors;
3948 				    i++)
3949 					export.export.ex_secflavors[i] =
3950 					    oexp.export.ex_secflavors[i];
3951 			}
3952 			export.fspec = oexp.fspec;
3953 			if (error == 0)
3954 				error = nfsrv_v4rootexport(&export, cred, p);
3955 			free(export.export.ex_groups, M_TEMP);
3956 		}
3957 	} else if (uap->flag & NFSSVC_NOPUBLICFH) {
3958 		nfs_pubfhset = 0;
3959 		error = 0;
3960 	} else if (uap->flag & NFSSVC_STABLERESTART) {
3961 		error = copyin(uap->argp, (caddr_t)&stablefd,
3962 		    sizeof (int));
3963 		if (!error)
3964 			error = fp_getfvp(p, stablefd, &fp, &vp);
3965 		if (!error && (NFSFPFLAG(fp) & (FREAD | FWRITE)) != (FREAD | FWRITE))
3966 			error = EBADF;
3967 		if (!error && newnfs_numnfsd != 0)
3968 			error = EPERM;
3969 		if (!error) {
3970 			nfsrv_stablefirst.nsf_fp = fp;
3971 			nfsrv_setupstable(p);
3972 		}
3973 	} else if (uap->flag & NFSSVC_ADMINREVOKE) {
3974 		error = copyin(uap->argp, (caddr_t)&adminrevoke,
3975 		    sizeof (struct nfsd_clid));
3976 		if (!error)
3977 			error = nfsrv_adminrevoke(&adminrevoke, p);
3978 	} else if (uap->flag & NFSSVC_DUMPCLIENTS) {
3979 		error = copyin(uap->argp, (caddr_t)&dumplist,
3980 		    sizeof (struct nfsd_dumplist));
3981 		if (!error && (dumplist.ndl_size < 1 ||
3982 			dumplist.ndl_size > NFSRV_MAXDUMPLIST))
3983 			error = EPERM;
3984 		if (!error) {
3985 		    len = sizeof (struct nfsd_dumpclients) * dumplist.ndl_size;
3986 		    dumpclients = malloc(len, M_TEMP, M_WAITOK | M_ZERO);
3987 		    nfsrv_dumpclients(dumpclients, dumplist.ndl_size);
3988 		    error = copyout(dumpclients, dumplist.ndl_list, len);
3989 		    free(dumpclients, M_TEMP);
3990 		}
3991 	} else if (uap->flag & NFSSVC_DUMPLOCKS) {
3992 		error = copyin(uap->argp, (caddr_t)&dumplocklist,
3993 		    sizeof (struct nfsd_dumplocklist));
3994 		if (!error && (dumplocklist.ndllck_size < 1 ||
3995 			dumplocklist.ndllck_size > NFSRV_MAXDUMPLIST))
3996 			error = EPERM;
3997 		if (!error)
3998 			error = nfsrv_lookupfilename(&nd,
3999 				dumplocklist.ndllck_fname, p);
4000 		if (!error) {
4001 			len = sizeof (struct nfsd_dumplocks) *
4002 				dumplocklist.ndllck_size;
4003 			dumplocks = malloc(len, M_TEMP, M_WAITOK | M_ZERO);
4004 			nfsrv_dumplocks(nd.ni_vp, dumplocks,
4005 			    dumplocklist.ndllck_size, p);
4006 			vput(nd.ni_vp);
4007 			error = copyout(dumplocks, dumplocklist.ndllck_list,
4008 			    len);
4009 			free(dumplocks, M_TEMP);
4010 		}
4011 	} else if (uap->flag & NFSSVC_BACKUPSTABLE) {
4012 		procp = p->td_proc;
4013 		PROC_LOCK(procp);
4014 		nfsd_master_pid = procp->p_pid;
4015 		bcopy(procp->p_comm, nfsd_master_comm, MAXCOMLEN + 1);
4016 		nfsd_master_start = procp->p_stats->p_start;
4017 		nfsd_master_proc = procp;
4018 		PROC_UNLOCK(procp);
4019 	} else if ((uap->flag & NFSSVC_SUSPENDNFSD) != 0) {
4020 		NFSLOCKV4ROOTMUTEX();
4021 		if (suspend_nfsd == 0) {
4022 			/* Lock out all nfsd threads */
4023 			do {
4024 				igotlock = nfsv4_lock(&nfsd_suspend_lock, 1,
4025 				    NULL, NFSV4ROOTLOCKMUTEXPTR, NULL);
4026 			} while (igotlock == 0 && suspend_nfsd == 0);
4027 			suspend_nfsd = 1;
4028 		}
4029 		NFSUNLOCKV4ROOTMUTEX();
4030 		error = 0;
4031 	} else if ((uap->flag & NFSSVC_RESUMENFSD) != 0) {
4032 		NFSLOCKV4ROOTMUTEX();
4033 		if (suspend_nfsd != 0) {
4034 			nfsv4_unlock(&nfsd_suspend_lock, 0);
4035 			suspend_nfsd = 0;
4036 		}
4037 		NFSUNLOCKV4ROOTMUTEX();
4038 		error = 0;
4039 	}
4040 
4041 	NFSEXITCODE(error);
4042 	return (error);
4043 }
4044 
4045 /*
4046  * Check exports.
4047  * Returns 0 if ok, 1 otherwise.
4048  */
4049 int
4050 nfsvno_testexp(struct nfsrv_descript *nd, struct nfsexstuff *exp)
4051 {
4052 	int i;
4053 
4054 	if ((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 #ifdef notnow
4062 		/* There is currently no auth_stat for this. */
4063 		else if ((nd->nd_flag & ND_TLS) == 0)
4064 			return (NFSERR_AUTHERR | AUTH_NEEDS_TLS);
4065 		else
4066 			return (NFSERR_AUTHERR | AUTH_NEEDS_TLS_MUTUAL_HOST);
4067 #endif
4068 		else
4069 			return (NFSERR_AUTHERR | AUTH_TOOWEAK);
4070 	}
4071 
4072 	/*
4073 	 * RFC2623 suggests that the NFSv3 Fsinfo RPC be allowed to use
4074 	 * AUTH_NONE or AUTH_SYS for file systems requiring RPCSEC_GSS.
4075 	 */
4076 	if ((nd->nd_flag & ND_NFSV3) != 0 && nd->nd_procnum == NFSPROC_FSINFO)
4077 		return (0);
4078 
4079 	/*
4080 	 * This seems odd, but allow the case where the security flavor
4081 	 * list is empty. This happens when NFSv4 is traversing non-exported
4082 	 * file systems. Exported file systems should always have a non-empty
4083 	 * security flavor list.
4084 	 */
4085 	if (exp->nes_numsecflavor == 0)
4086 		return (0);
4087 
4088 	for (i = 0; i < exp->nes_numsecflavor; i++) {
4089 		/*
4090 		 * The tests for privacy and integrity must be first,
4091 		 * since ND_GSS is set for everything but AUTH_SYS.
4092 		 */
4093 		if (exp->nes_secflavors[i] == RPCSEC_GSS_KRB5P &&
4094 		    (nd->nd_flag & ND_GSSPRIVACY))
4095 			return (0);
4096 		if (exp->nes_secflavors[i] == RPCSEC_GSS_KRB5I &&
4097 		    (nd->nd_flag & ND_GSSINTEGRITY))
4098 			return (0);
4099 		if (exp->nes_secflavors[i] == RPCSEC_GSS_KRB5 &&
4100 		    (nd->nd_flag & ND_GSS))
4101 			return (0);
4102 		if (exp->nes_secflavors[i] == AUTH_SYS &&
4103 		    (nd->nd_flag & ND_GSS) == 0)
4104 			return (0);
4105 	}
4106 	if ((nd->nd_flag & ND_NFSV4) != 0)
4107 		return (NFSERR_WRONGSEC);
4108 	return (NFSERR_AUTHERR | AUTH_TOOWEAK);
4109 }
4110 
4111 /*
4112  * Calculate a hash value for the fid in a file handle.
4113  */
4114 uint32_t
4115 nfsrv_hashfh(fhandle_t *fhp)
4116 {
4117 	uint32_t hashval;
4118 
4119 	hashval = hash32_buf(&fhp->fh_fid, sizeof(struct fid), 0);
4120 	return (hashval);
4121 }
4122 
4123 /*
4124  * Calculate a hash value for the sessionid.
4125  */
4126 uint32_t
4127 nfsrv_hashsessionid(uint8_t *sessionid)
4128 {
4129 	uint32_t hashval;
4130 
4131 	hashval = hash32_buf(sessionid, NFSX_V4SESSIONID, 0);
4132 	return (hashval);
4133 }
4134 
4135 /*
4136  * Signal the userland master nfsd to backup the stable restart file.
4137  */
4138 void
4139 nfsrv_backupstable(void)
4140 {
4141 	struct proc *procp;
4142 
4143 	if (nfsd_master_proc != NULL) {
4144 		procp = pfind(nfsd_master_pid);
4145 		/* Try to make sure it is the correct process. */
4146 		if (procp == nfsd_master_proc &&
4147 		    procp->p_stats->p_start.tv_sec ==
4148 		    nfsd_master_start.tv_sec &&
4149 		    procp->p_stats->p_start.tv_usec ==
4150 		    nfsd_master_start.tv_usec &&
4151 		    strcmp(procp->p_comm, nfsd_master_comm) == 0)
4152 			kern_psignal(procp, SIGUSR2);
4153 		else
4154 			nfsd_master_proc = NULL;
4155 
4156 		if (procp != NULL)
4157 			PROC_UNLOCK(procp);
4158 	}
4159 }
4160 
4161 /*
4162  * Create a DS data file for nfsrv_pnfscreate(). Called for each mirror.
4163  * The arguments are in a structure, so that they can be passed through
4164  * taskqueue for a kernel process to execute this function.
4165  */
4166 struct nfsrvdscreate {
4167 	int			done;
4168 	int			inprog;
4169 	struct task		tsk;
4170 	struct ucred		*tcred;
4171 	struct vnode		*dvp;
4172 	NFSPROC_T		*p;
4173 	struct pnfsdsfile	*pf;
4174 	int			err;
4175 	fhandle_t		fh;
4176 	struct vattr		va;
4177 	struct vattr		createva;
4178 };
4179 
4180 int
4181 nfsrv_dscreate(struct vnode *dvp, struct vattr *vap, struct vattr *nvap,
4182     fhandle_t *fhp, struct pnfsdsfile *pf, struct pnfsdsattr *dsa,
4183     char *fnamep, struct ucred *tcred, NFSPROC_T *p, struct vnode **nvpp)
4184 {
4185 	struct vnode *nvp;
4186 	struct nameidata named;
4187 	struct vattr va;
4188 	char *bufp;
4189 	u_long *hashp;
4190 	struct nfsnode *np;
4191 	struct nfsmount *nmp;
4192 	int error;
4193 
4194 	NFSNAMEICNDSET(&named.ni_cnd, tcred, CREATE,
4195 	    LOCKPARENT | LOCKLEAF | SAVESTART | NOCACHE);
4196 	nfsvno_setpathbuf(&named, &bufp, &hashp);
4197 	named.ni_cnd.cn_lkflags = LK_EXCLUSIVE;
4198 	named.ni_cnd.cn_nameptr = bufp;
4199 	if (fnamep != NULL) {
4200 		strlcpy(bufp, fnamep, PNFS_FILENAME_LEN + 1);
4201 		named.ni_cnd.cn_namelen = strlen(bufp);
4202 	} else
4203 		named.ni_cnd.cn_namelen = nfsrv_putfhname(fhp, bufp);
4204 	NFSD_DEBUG(4, "nfsrv_dscreate: dvp=%p fname=%s\n", dvp, bufp);
4205 
4206 	/* Create the date file in the DS mount. */
4207 	error = NFSVOPLOCK(dvp, LK_EXCLUSIVE);
4208 	if (error == 0) {
4209 		error = VOP_CREATE(dvp, &nvp, &named.ni_cnd, vap);
4210 		vref(dvp);
4211 		VOP_VPUT_PAIR(dvp, error == 0 ? &nvp : NULL, false);
4212 		if (error == 0) {
4213 			/* Set the ownership of the file. */
4214 			error = VOP_SETATTR(nvp, nvap, tcred);
4215 			NFSD_DEBUG(4, "nfsrv_dscreate:"
4216 			    " setattr-uid=%d\n", error);
4217 			if (error != 0)
4218 				vput(nvp);
4219 		}
4220 		if (error != 0)
4221 			printf("pNFS: pnfscreate failed=%d\n", error);
4222 	} else
4223 		printf("pNFS: pnfscreate vnlock=%d\n", error);
4224 	if (error == 0) {
4225 		np = VTONFS(nvp);
4226 		nmp = VFSTONFS(nvp->v_mount);
4227 		if (strcmp(nvp->v_mount->mnt_vfc->vfc_name, "nfs")
4228 		    != 0 || nmp->nm_nam->sa_len > sizeof(
4229 		    struct sockaddr_in6) ||
4230 		    np->n_fhp->nfh_len != NFSX_MYFH) {
4231 			printf("Bad DS file: fstype=%s salen=%d"
4232 			    " fhlen=%d\n",
4233 			    nvp->v_mount->mnt_vfc->vfc_name,
4234 			    nmp->nm_nam->sa_len, np->n_fhp->nfh_len);
4235 			error = ENOENT;
4236 		}
4237 
4238 		/* Set extattrs for the DS on the MDS file. */
4239 		if (error == 0) {
4240 			if (dsa != NULL) {
4241 				error = VOP_GETATTR(nvp, &va, tcred);
4242 				if (error == 0) {
4243 					dsa->dsa_filerev = va.va_filerev;
4244 					dsa->dsa_size = va.va_size;
4245 					dsa->dsa_atime = va.va_atime;
4246 					dsa->dsa_mtime = va.va_mtime;
4247 					dsa->dsa_bytes = va.va_bytes;
4248 				}
4249 			}
4250 			if (error == 0) {
4251 				NFSBCOPY(np->n_fhp->nfh_fh, &pf->dsf_fh,
4252 				    NFSX_MYFH);
4253 				NFSBCOPY(nmp->nm_nam, &pf->dsf_sin,
4254 				    nmp->nm_nam->sa_len);
4255 				NFSBCOPY(named.ni_cnd.cn_nameptr,
4256 				    pf->dsf_filename,
4257 				    sizeof(pf->dsf_filename));
4258 			}
4259 		} else
4260 			printf("pNFS: pnfscreate can't get DS"
4261 			    " attr=%d\n", error);
4262 		if (nvpp != NULL && error == 0)
4263 			*nvpp = nvp;
4264 		else
4265 			vput(nvp);
4266 	}
4267 	nfsvno_relpathbuf(&named);
4268 	return (error);
4269 }
4270 
4271 /*
4272  * Start up the thread that will execute nfsrv_dscreate().
4273  */
4274 static void
4275 start_dscreate(void *arg, int pending)
4276 {
4277 	struct nfsrvdscreate *dsc;
4278 
4279 	dsc = (struct nfsrvdscreate *)arg;
4280 	dsc->err = nfsrv_dscreate(dsc->dvp, &dsc->createva, &dsc->va, &dsc->fh,
4281 	    dsc->pf, NULL, NULL, dsc->tcred, dsc->p, NULL);
4282 	dsc->done = 1;
4283 	NFSD_DEBUG(4, "start_dscreate: err=%d\n", dsc->err);
4284 }
4285 
4286 /*
4287  * Create a pNFS data file on the Data Server(s).
4288  */
4289 static void
4290 nfsrv_pnfscreate(struct vnode *vp, struct vattr *vap, struct ucred *cred,
4291     NFSPROC_T *p)
4292 {
4293 	struct nfsrvdscreate *dsc, *tdsc = NULL;
4294 	struct nfsdevice *ds, *tds, *fds;
4295 	struct mount *mp;
4296 	struct pnfsdsfile *pf, *tpf;
4297 	struct pnfsdsattr dsattr;
4298 	struct vattr va;
4299 	struct vnode *dvp[NFSDEV_MAXMIRRORS];
4300 	struct nfsmount *nmp;
4301 	fhandle_t fh;
4302 	uid_t vauid;
4303 	gid_t vagid;
4304 	u_short vamode;
4305 	struct ucred *tcred;
4306 	int dsdir[NFSDEV_MAXMIRRORS], error, i, mirrorcnt, ret;
4307 	int failpos, timo;
4308 
4309 	/* Get a DS server directory in a round-robin order. */
4310 	mirrorcnt = 1;
4311 	mp = vp->v_mount;
4312 	ds = fds = NULL;
4313 	NFSDDSLOCK();
4314 	/*
4315 	 * Search for the first entry that handles this MDS fs, but use the
4316 	 * first entry for all MDS fs's otherwise.
4317 	 */
4318 	TAILQ_FOREACH(tds, &nfsrv_devidhead, nfsdev_list) {
4319 		if (tds->nfsdev_nmp != NULL) {
4320 			if (tds->nfsdev_mdsisset == 0 && ds == NULL)
4321 				ds = tds;
4322 			else if (tds->nfsdev_mdsisset != 0 && fsidcmp(
4323 			    &mp->mnt_stat.f_fsid, &tds->nfsdev_mdsfsid) == 0) {
4324 				ds = fds = tds;
4325 				break;
4326 			}
4327 		}
4328 	}
4329 	if (ds == NULL) {
4330 		NFSDDSUNLOCK();
4331 		NFSD_DEBUG(4, "nfsrv_pnfscreate: no srv\n");
4332 		return;
4333 	}
4334 	i = dsdir[0] = ds->nfsdev_nextdir;
4335 	ds->nfsdev_nextdir = (ds->nfsdev_nextdir + 1) % nfsrv_dsdirsize;
4336 	dvp[0] = ds->nfsdev_dsdir[i];
4337 	tds = TAILQ_NEXT(ds, nfsdev_list);
4338 	if (nfsrv_maxpnfsmirror > 1 && tds != NULL) {
4339 		TAILQ_FOREACH_FROM(tds, &nfsrv_devidhead, nfsdev_list) {
4340 			if (tds->nfsdev_nmp != NULL &&
4341 			    ((tds->nfsdev_mdsisset == 0 && fds == NULL) ||
4342 			     (tds->nfsdev_mdsisset != 0 && fds != NULL &&
4343 			      fsidcmp(&mp->mnt_stat.f_fsid,
4344 			      &tds->nfsdev_mdsfsid) == 0))) {
4345 				dsdir[mirrorcnt] = i;
4346 				dvp[mirrorcnt] = tds->nfsdev_dsdir[i];
4347 				mirrorcnt++;
4348 				if (mirrorcnt >= nfsrv_maxpnfsmirror)
4349 					break;
4350 			}
4351 		}
4352 	}
4353 	/* Put at end of list to implement round-robin usage. */
4354 	TAILQ_REMOVE(&nfsrv_devidhead, ds, nfsdev_list);
4355 	TAILQ_INSERT_TAIL(&nfsrv_devidhead, ds, nfsdev_list);
4356 	NFSDDSUNLOCK();
4357 	dsc = NULL;
4358 	if (mirrorcnt > 1)
4359 		tdsc = dsc = malloc(sizeof(*dsc) * (mirrorcnt - 1), M_TEMP,
4360 		    M_WAITOK | M_ZERO);
4361 	tpf = pf = malloc(sizeof(*pf) * nfsrv_maxpnfsmirror, M_TEMP, M_WAITOK |
4362 	    M_ZERO);
4363 
4364 	error = nfsvno_getfh(vp, &fh, p);
4365 	if (error == 0)
4366 		error = VOP_GETATTR(vp, &va, cred);
4367 	if (error == 0) {
4368 		/* Set the attributes for "vp" to Setattr the DS vp. */
4369 		vauid = va.va_uid;
4370 		vagid = va.va_gid;
4371 		vamode = va.va_mode;
4372 		VATTR_NULL(&va);
4373 		va.va_uid = vauid;
4374 		va.va_gid = vagid;
4375 		va.va_mode = vamode;
4376 		va.va_size = 0;
4377 	} else
4378 		printf("pNFS: pnfscreate getfh+attr=%d\n", error);
4379 
4380 	NFSD_DEBUG(4, "nfsrv_pnfscreate: cruid=%d crgid=%d\n", cred->cr_uid,
4381 	    cred->cr_gid);
4382 	/* Make data file name based on FH. */
4383 	tcred = newnfs_getcred();
4384 
4385 	/*
4386 	 * Create the file on each DS mirror, using kernel process(es) for the
4387 	 * additional mirrors.
4388 	 */
4389 	failpos = -1;
4390 	for (i = 0; i < mirrorcnt - 1 && error == 0; i++, tpf++, tdsc++) {
4391 		tpf->dsf_dir = dsdir[i];
4392 		tdsc->tcred = tcred;
4393 		tdsc->p = p;
4394 		tdsc->pf = tpf;
4395 		tdsc->createva = *vap;
4396 		NFSBCOPY(&fh, &tdsc->fh, sizeof(fh));
4397 		tdsc->va = va;
4398 		tdsc->dvp = dvp[i];
4399 		tdsc->done = 0;
4400 		tdsc->inprog = 0;
4401 		tdsc->err = 0;
4402 		ret = EIO;
4403 		if (nfs_pnfsiothreads != 0) {
4404 			ret = nfs_pnfsio(start_dscreate, tdsc);
4405 			NFSD_DEBUG(4, "nfsrv_pnfscreate: nfs_pnfsio=%d\n", ret);
4406 		}
4407 		if (ret != 0) {
4408 			ret = nfsrv_dscreate(dvp[i], vap, &va, &fh, tpf, NULL,
4409 			    NULL, tcred, p, NULL);
4410 			if (ret != 0) {
4411 				KASSERT(error == 0, ("nfsrv_dscreate err=%d",
4412 				    error));
4413 				if (failpos == -1 && nfsds_failerr(ret))
4414 					failpos = i;
4415 				else
4416 					error = ret;
4417 			}
4418 		}
4419 	}
4420 	if (error == 0) {
4421 		tpf->dsf_dir = dsdir[mirrorcnt - 1];
4422 		error = nfsrv_dscreate(dvp[mirrorcnt - 1], vap, &va, &fh, tpf,
4423 		    &dsattr, NULL, tcred, p, NULL);
4424 		if (failpos == -1 && mirrorcnt > 1 && nfsds_failerr(error)) {
4425 			failpos = mirrorcnt - 1;
4426 			error = 0;
4427 		}
4428 	}
4429 	timo = hz / 50;		/* Wait for 20msec. */
4430 	if (timo < 1)
4431 		timo = 1;
4432 	/* Wait for kernel task(s) to complete. */
4433 	for (tdsc = dsc, i = 0; i < mirrorcnt - 1; i++, tdsc++) {
4434 		while (tdsc->inprog != 0 && tdsc->done == 0)
4435 			tsleep(&tdsc->tsk, PVFS, "srvdcr", timo);
4436 		if (tdsc->err != 0) {
4437 			if (failpos == -1 && nfsds_failerr(tdsc->err))
4438 				failpos = i;
4439 			else if (error == 0)
4440 				error = tdsc->err;
4441 		}
4442 	}
4443 
4444 	/*
4445 	 * If failpos has been set, that mirror has failed, so it needs
4446 	 * to be disabled.
4447 	 */
4448 	if (failpos >= 0) {
4449 		nmp = VFSTONFS(dvp[failpos]->v_mount);
4450 		NFSLOCKMNT(nmp);
4451 		if ((nmp->nm_privflag & (NFSMNTP_FORCEDISM |
4452 		     NFSMNTP_CANCELRPCS)) == 0) {
4453 			nmp->nm_privflag |= NFSMNTP_CANCELRPCS;
4454 			NFSUNLOCKMNT(nmp);
4455 			ds = nfsrv_deldsnmp(PNFSDOP_DELDSSERVER, nmp, p);
4456 			NFSD_DEBUG(4, "dscreatfail fail=%d ds=%p\n", failpos,
4457 			    ds);
4458 			if (ds != NULL)
4459 				nfsrv_killrpcs(nmp);
4460 			NFSLOCKMNT(nmp);
4461 			nmp->nm_privflag &= ~NFSMNTP_CANCELRPCS;
4462 			wakeup(nmp);
4463 		}
4464 		NFSUNLOCKMNT(nmp);
4465 	}
4466 
4467 	NFSFREECRED(tcred);
4468 	if (error == 0) {
4469 		ASSERT_VOP_ELOCKED(vp, "nfsrv_pnfscreate vp");
4470 
4471 		NFSD_DEBUG(4, "nfsrv_pnfscreate: mirrorcnt=%d maxmirror=%d\n",
4472 		    mirrorcnt, nfsrv_maxpnfsmirror);
4473 		/*
4474 		 * For all mirrors that couldn't be created, fill in the
4475 		 * *pf structure, but with an IP address == 0.0.0.0.
4476 		 */
4477 		tpf = pf + mirrorcnt;
4478 		for (i = mirrorcnt; i < nfsrv_maxpnfsmirror; i++, tpf++) {
4479 			*tpf = *pf;
4480 			tpf->dsf_sin.sin_family = AF_INET;
4481 			tpf->dsf_sin.sin_len = sizeof(struct sockaddr_in);
4482 			tpf->dsf_sin.sin_addr.s_addr = 0;
4483 			tpf->dsf_sin.sin_port = 0;
4484 		}
4485 
4486 		error = vn_extattr_set(vp, IO_NODELOCKED,
4487 		    EXTATTR_NAMESPACE_SYSTEM, "pnfsd.dsfile",
4488 		    sizeof(*pf) * nfsrv_maxpnfsmirror, (char *)pf, p);
4489 		if (error == 0)
4490 			error = vn_extattr_set(vp, IO_NODELOCKED,
4491 			    EXTATTR_NAMESPACE_SYSTEM, "pnfsd.dsattr",
4492 			    sizeof(dsattr), (char *)&dsattr, p);
4493 		if (error != 0)
4494 			printf("pNFS: pnfscreate setextattr=%d\n",
4495 			    error);
4496 	} else
4497 		printf("pNFS: pnfscreate=%d\n", error);
4498 	free(pf, M_TEMP);
4499 	free(dsc, M_TEMP);
4500 }
4501 
4502 /*
4503  * Get the information needed to remove the pNFS Data Server file from the
4504  * Metadata file.  Upon success, ddvp is set non-NULL to the locked
4505  * DS directory vnode.  The caller must unlock *ddvp when done with it.
4506  */
4507 static void
4508 nfsrv_pnfsremovesetup(struct vnode *vp, NFSPROC_T *p, struct vnode **dvpp,
4509     int *mirrorcntp, char *fname, fhandle_t *fhp)
4510 {
4511 	struct vattr va;
4512 	struct ucred *tcred;
4513 	char *buf;
4514 	int buflen, error;
4515 
4516 	dvpp[0] = NULL;
4517 	/* If not an exported regular file or not a pNFS server, just return. */
4518 	if (vp->v_type != VREG || (vp->v_mount->mnt_flag & MNT_EXPORTED) == 0 ||
4519 	    nfsrv_devidcnt == 0)
4520 		return;
4521 
4522 	/* Check to see if this is the last hard link. */
4523 	tcred = newnfs_getcred();
4524 	error = VOP_GETATTR(vp, &va, tcred);
4525 	NFSFREECRED(tcred);
4526 	if (error != 0) {
4527 		printf("pNFS: nfsrv_pnfsremovesetup getattr=%d\n", error);
4528 		return;
4529 	}
4530 	if (va.va_nlink > 1)
4531 		return;
4532 
4533 	error = nfsvno_getfh(vp, fhp, p);
4534 	if (error != 0) {
4535 		printf("pNFS: nfsrv_pnfsremovesetup getfh=%d\n", error);
4536 		return;
4537 	}
4538 
4539 	buflen = 1024;
4540 	buf = malloc(buflen, M_TEMP, M_WAITOK);
4541 	/* Get the directory vnode for the DS mount and the file handle. */
4542 	error = nfsrv_dsgetsockmnt(vp, 0, buf, &buflen, mirrorcntp, p, dvpp,
4543 	    NULL, NULL, fname, NULL, NULL, NULL, NULL, NULL);
4544 	free(buf, M_TEMP);
4545 	if (error != 0)
4546 		printf("pNFS: nfsrv_pnfsremovesetup getsockmnt=%d\n", error);
4547 }
4548 
4549 /*
4550  * Remove a DS data file for nfsrv_pnfsremove(). Called for each mirror.
4551  * The arguments are in a structure, so that they can be passed through
4552  * taskqueue for a kernel process to execute this function.
4553  */
4554 struct nfsrvdsremove {
4555 	int			done;
4556 	int			inprog;
4557 	struct task		tsk;
4558 	struct ucred		*tcred;
4559 	struct vnode		*dvp;
4560 	NFSPROC_T		*p;
4561 	int			err;
4562 	char			fname[PNFS_FILENAME_LEN + 1];
4563 };
4564 
4565 static int
4566 nfsrv_dsremove(struct vnode *dvp, char *fname, struct ucred *tcred,
4567     NFSPROC_T *p)
4568 {
4569 	struct nameidata named;
4570 	struct vnode *nvp;
4571 	char *bufp;
4572 	u_long *hashp;
4573 	int error;
4574 
4575 	error = NFSVOPLOCK(dvp, LK_EXCLUSIVE);
4576 	if (error != 0)
4577 		return (error);
4578 	named.ni_cnd.cn_nameiop = DELETE;
4579 	named.ni_cnd.cn_lkflags = LK_EXCLUSIVE | LK_RETRY;
4580 	named.ni_cnd.cn_cred = tcred;
4581 	named.ni_cnd.cn_flags = ISLASTCN | LOCKPARENT | LOCKLEAF | SAVENAME;
4582 	nfsvno_setpathbuf(&named, &bufp, &hashp);
4583 	named.ni_cnd.cn_nameptr = bufp;
4584 	named.ni_cnd.cn_namelen = strlen(fname);
4585 	strlcpy(bufp, fname, NAME_MAX);
4586 	NFSD_DEBUG(4, "nfsrv_pnfsremove: filename=%s\n", bufp);
4587 	error = VOP_LOOKUP(dvp, &nvp, &named.ni_cnd);
4588 	NFSD_DEBUG(4, "nfsrv_pnfsremove: aft LOOKUP=%d\n", error);
4589 	if (error == 0) {
4590 		error = VOP_REMOVE(dvp, nvp, &named.ni_cnd);
4591 		vput(nvp);
4592 	}
4593 	NFSVOPUNLOCK(dvp);
4594 	nfsvno_relpathbuf(&named);
4595 	if (error != 0)
4596 		printf("pNFS: nfsrv_pnfsremove failed=%d\n", error);
4597 	return (error);
4598 }
4599 
4600 /*
4601  * Start up the thread that will execute nfsrv_dsremove().
4602  */
4603 static void
4604 start_dsremove(void *arg, int pending)
4605 {
4606 	struct nfsrvdsremove *dsrm;
4607 
4608 	dsrm = (struct nfsrvdsremove *)arg;
4609 	dsrm->err = nfsrv_dsremove(dsrm->dvp, dsrm->fname, dsrm->tcred,
4610 	    dsrm->p);
4611 	dsrm->done = 1;
4612 	NFSD_DEBUG(4, "start_dsremove: err=%d\n", dsrm->err);
4613 }
4614 
4615 /*
4616  * Remove a pNFS data file from a Data Server.
4617  * nfsrv_pnfsremovesetup() must have been called before the MDS file was
4618  * removed to set up the dvp and fill in the FH.
4619  */
4620 static void
4621 nfsrv_pnfsremove(struct vnode **dvp, int mirrorcnt, char *fname, fhandle_t *fhp,
4622     NFSPROC_T *p)
4623 {
4624 	struct ucred *tcred;
4625 	struct nfsrvdsremove *dsrm, *tdsrm;
4626 	struct nfsdevice *ds;
4627 	struct nfsmount *nmp;
4628 	int failpos, i, ret, timo;
4629 
4630 	tcred = newnfs_getcred();
4631 	dsrm = NULL;
4632 	if (mirrorcnt > 1)
4633 		dsrm = malloc(sizeof(*dsrm) * mirrorcnt - 1, M_TEMP, M_WAITOK);
4634 	/*
4635 	 * Remove the file on each DS mirror, using kernel process(es) for the
4636 	 * additional mirrors.
4637 	 */
4638 	failpos = -1;
4639 	for (tdsrm = dsrm, i = 0; i < mirrorcnt - 1; i++, tdsrm++) {
4640 		tdsrm->tcred = tcred;
4641 		tdsrm->p = p;
4642 		tdsrm->dvp = dvp[i];
4643 		strlcpy(tdsrm->fname, fname, PNFS_FILENAME_LEN + 1);
4644 		tdsrm->inprog = 0;
4645 		tdsrm->done = 0;
4646 		tdsrm->err = 0;
4647 		ret = EIO;
4648 		if (nfs_pnfsiothreads != 0) {
4649 			ret = nfs_pnfsio(start_dsremove, tdsrm);
4650 			NFSD_DEBUG(4, "nfsrv_pnfsremove: nfs_pnfsio=%d\n", ret);
4651 		}
4652 		if (ret != 0) {
4653 			ret = nfsrv_dsremove(dvp[i], fname, tcred, p);
4654 			if (failpos == -1 && nfsds_failerr(ret))
4655 				failpos = i;
4656 		}
4657 	}
4658 	ret = nfsrv_dsremove(dvp[mirrorcnt - 1], fname, tcred, p);
4659 	if (failpos == -1 && mirrorcnt > 1 && nfsds_failerr(ret))
4660 		failpos = mirrorcnt - 1;
4661 	timo = hz / 50;		/* Wait for 20msec. */
4662 	if (timo < 1)
4663 		timo = 1;
4664 	/* Wait for kernel task(s) to complete. */
4665 	for (tdsrm = dsrm, i = 0; i < mirrorcnt - 1; i++, tdsrm++) {
4666 		while (tdsrm->inprog != 0 && tdsrm->done == 0)
4667 			tsleep(&tdsrm->tsk, PVFS, "srvdsrm", timo);
4668 		if (failpos == -1 && nfsds_failerr(tdsrm->err))
4669 			failpos = i;
4670 	}
4671 
4672 	/*
4673 	 * If failpos has been set, that mirror has failed, so it needs
4674 	 * to be disabled.
4675 	 */
4676 	if (failpos >= 0) {
4677 		nmp = VFSTONFS(dvp[failpos]->v_mount);
4678 		NFSLOCKMNT(nmp);
4679 		if ((nmp->nm_privflag & (NFSMNTP_FORCEDISM |
4680 		     NFSMNTP_CANCELRPCS)) == 0) {
4681 			nmp->nm_privflag |= NFSMNTP_CANCELRPCS;
4682 			NFSUNLOCKMNT(nmp);
4683 			ds = nfsrv_deldsnmp(PNFSDOP_DELDSSERVER, nmp, p);
4684 			NFSD_DEBUG(4, "dsremovefail fail=%d ds=%p\n", failpos,
4685 			    ds);
4686 			if (ds != NULL)
4687 				nfsrv_killrpcs(nmp);
4688 			NFSLOCKMNT(nmp);
4689 			nmp->nm_privflag &= ~NFSMNTP_CANCELRPCS;
4690 			wakeup(nmp);
4691 		}
4692 		NFSUNLOCKMNT(nmp);
4693 	}
4694 
4695 	/* Get rid all layouts for the file. */
4696 	nfsrv_freefilelayouts(fhp);
4697 
4698 	NFSFREECRED(tcred);
4699 	free(dsrm, M_TEMP);
4700 }
4701 
4702 /*
4703  * Generate a file name based on the file handle and put it in *bufp.
4704  * Return the number of bytes generated.
4705  */
4706 static int
4707 nfsrv_putfhname(fhandle_t *fhp, char *bufp)
4708 {
4709 	int i;
4710 	uint8_t *cp;
4711 	const uint8_t *hexdigits = "0123456789abcdef";
4712 
4713 	cp = (uint8_t *)fhp;
4714 	for (i = 0; i < sizeof(*fhp); i++) {
4715 		bufp[2 * i] = hexdigits[(*cp >> 4) & 0xf];
4716 		bufp[2 * i + 1] = hexdigits[*cp++ & 0xf];
4717 	}
4718 	bufp[2 * i] = '\0';
4719 	return (2 * i);
4720 }
4721 
4722 /*
4723  * Update the Metadata file's attributes from the DS file when a Read/Write
4724  * layout is returned.
4725  * Basically just call nfsrv_proxyds() with procedure == NFSPROC_LAYOUTRETURN
4726  * so that it does a nfsrv_getattrdsrpc() and nfsrv_setextattr() on the DS file.
4727  */
4728 int
4729 nfsrv_updatemdsattr(struct vnode *vp, struct nfsvattr *nap, NFSPROC_T *p)
4730 {
4731 	struct ucred *tcred;
4732 	int error;
4733 
4734 	/* Do this as root so that it won't fail with EACCES. */
4735 	tcred = newnfs_getcred();
4736 	error = nfsrv_proxyds(vp, 0, 0, tcred, p, NFSPROC_LAYOUTRETURN,
4737 	    NULL, NULL, NULL, nap, NULL, NULL, 0, NULL);
4738 	NFSFREECRED(tcred);
4739 	return (error);
4740 }
4741 
4742 /*
4743  * Set the NFSv4 ACL on the DS file to the same ACL as the MDS file.
4744  */
4745 static int
4746 nfsrv_dssetacl(struct vnode *vp, struct acl *aclp, struct ucred *cred,
4747     NFSPROC_T *p)
4748 {
4749 	int error;
4750 
4751 	error = nfsrv_proxyds(vp, 0, 0, cred, p, NFSPROC_SETACL,
4752 	    NULL, NULL, NULL, NULL, aclp, NULL, 0, NULL);
4753 	return (error);
4754 }
4755 
4756 static int
4757 nfsrv_proxyds(struct vnode *vp, off_t off, int cnt, struct ucred *cred,
4758     struct thread *p, int ioproc, struct mbuf **mpp, char *cp,
4759     struct mbuf **mpp2, struct nfsvattr *nap, struct acl *aclp,
4760     off_t *offp, int content, bool *eofp)
4761 {
4762 	struct nfsmount *nmp[NFSDEV_MAXMIRRORS], *failnmp;
4763 	fhandle_t fh[NFSDEV_MAXMIRRORS];
4764 	struct vnode *dvp[NFSDEV_MAXMIRRORS];
4765 	struct nfsdevice *ds;
4766 	struct pnfsdsattr dsattr;
4767 	struct opnfsdsattr odsattr;
4768 	char *buf;
4769 	int buflen, error, failpos, i, mirrorcnt, origmircnt, trycnt;
4770 
4771 	NFSD_DEBUG(4, "in nfsrv_proxyds\n");
4772 	/*
4773 	 * If not a regular file, not exported or not a pNFS server,
4774 	 * just return ENOENT.
4775 	 */
4776 	if (vp->v_type != VREG || (vp->v_mount->mnt_flag & MNT_EXPORTED) == 0 ||
4777 	    nfsrv_devidcnt == 0)
4778 		return (ENOENT);
4779 
4780 	buflen = 1024;
4781 	buf = malloc(buflen, M_TEMP, M_WAITOK);
4782 	error = 0;
4783 
4784 	/*
4785 	 * For Getattr, get the Change attribute (va_filerev) and size (va_size)
4786 	 * from the MetaData file's extended attribute.
4787 	 */
4788 	if (ioproc == NFSPROC_GETATTR) {
4789 		error = vn_extattr_get(vp, IO_NODELOCKED,
4790 		    EXTATTR_NAMESPACE_SYSTEM, "pnfsd.dsattr", &buflen, buf,
4791 		    p);
4792 		if (error == 0) {
4793 			if (buflen == sizeof(odsattr)) {
4794 				NFSBCOPY(buf, &odsattr, buflen);
4795 				nap->na_filerev = odsattr.dsa_filerev;
4796 				nap->na_size = odsattr.dsa_size;
4797 				nap->na_atime = odsattr.dsa_atime;
4798 				nap->na_mtime = odsattr.dsa_mtime;
4799 				/*
4800 				 * Fake na_bytes by rounding up na_size.
4801 				 * Since we don't know the block size, just
4802 				 * use BLKDEV_IOSIZE.
4803 				 */
4804 				nap->na_bytes = (odsattr.dsa_size +
4805 				    BLKDEV_IOSIZE - 1) & ~(BLKDEV_IOSIZE - 1);
4806 			} else if (buflen == sizeof(dsattr)) {
4807 				NFSBCOPY(buf, &dsattr, buflen);
4808 				nap->na_filerev = dsattr.dsa_filerev;
4809 				nap->na_size = dsattr.dsa_size;
4810 				nap->na_atime = dsattr.dsa_atime;
4811 				nap->na_mtime = dsattr.dsa_mtime;
4812 				nap->na_bytes = dsattr.dsa_bytes;
4813 			} else
4814 				error = ENXIO;
4815 		}
4816 		if (error == 0) {
4817 			/*
4818 			 * If nfsrv_pnfsgetdsattr is 0 or nfsrv_checkdsattr()
4819 			 * returns 0, just return now.  nfsrv_checkdsattr()
4820 			 * returns 0 if there is no Read/Write layout
4821 			 * plus either an Open/Write_access or Write
4822 			 * delegation issued to a client for the file.
4823 			 */
4824 			if (nfsrv_pnfsgetdsattr == 0 ||
4825 			    nfsrv_checkdsattr(vp, p) == 0) {
4826 				free(buf, M_TEMP);
4827 				return (error);
4828 			}
4829 		}
4830 
4831 		/*
4832 		 * Clear ENOATTR so the code below will attempt to do a
4833 		 * nfsrv_getattrdsrpc() to get the attributes and (re)create
4834 		 * the extended attribute.
4835 		 */
4836 		if (error == ENOATTR)
4837 			error = 0;
4838 	}
4839 
4840 	origmircnt = -1;
4841 	trycnt = 0;
4842 tryagain:
4843 	if (error == 0) {
4844 		buflen = 1024;
4845 		if (ioproc == NFSPROC_READDS && NFSVOPISLOCKED(vp) ==
4846 		    LK_EXCLUSIVE)
4847 			printf("nfsrv_proxyds: Readds vp exclusively locked\n");
4848 		error = nfsrv_dsgetsockmnt(vp, LK_SHARED, buf, &buflen,
4849 		    &mirrorcnt, p, dvp, fh, NULL, NULL, NULL, NULL, NULL,
4850 		    NULL, NULL);
4851 		if (error == 0) {
4852 			for (i = 0; i < mirrorcnt; i++)
4853 				nmp[i] = VFSTONFS(dvp[i]->v_mount);
4854 		} else
4855 			printf("pNFS: proxy getextattr sockaddr=%d\n", error);
4856 	} else
4857 		printf("pNFS: nfsrv_dsgetsockmnt=%d\n", error);
4858 	if (error == 0) {
4859 		failpos = -1;
4860 		if (origmircnt == -1)
4861 			origmircnt = mirrorcnt;
4862 		/*
4863 		 * If failpos is set to a mirror#, then that mirror has
4864 		 * failed and will be disabled. For Read, Getattr and Seek, the
4865 		 * function only tries one mirror, so if that mirror has
4866 		 * failed, it will need to be retried. As such, increment
4867 		 * tryitagain for these cases.
4868 		 * For Write, Setattr and Setacl, the function tries all
4869 		 * mirrors and will not return an error for the case where
4870 		 * one mirror has failed. For these cases, the functioning
4871 		 * mirror(s) will have been modified, so a retry isn't
4872 		 * necessary. These functions will set failpos for the
4873 		 * failed mirror#.
4874 		 */
4875 		if (ioproc == NFSPROC_READDS) {
4876 			error = nfsrv_readdsrpc(fh, off, cnt, cred, p, nmp[0],
4877 			    mpp, mpp2);
4878 			if (nfsds_failerr(error) && mirrorcnt > 1) {
4879 				/*
4880 				 * Setting failpos will cause the mirror
4881 				 * to be disabled and then a retry of this
4882 				 * read is required.
4883 				 */
4884 				failpos = 0;
4885 				error = 0;
4886 				trycnt++;
4887 			}
4888 		} else if (ioproc == NFSPROC_WRITEDS)
4889 			error = nfsrv_writedsrpc(fh, off, cnt, cred, p, vp,
4890 			    &nmp[0], mirrorcnt, mpp, cp, &failpos);
4891 		else if (ioproc == NFSPROC_SETATTR)
4892 			error = nfsrv_setattrdsrpc(fh, cred, p, vp, &nmp[0],
4893 			    mirrorcnt, nap, &failpos);
4894 		else if (ioproc == NFSPROC_SETACL)
4895 			error = nfsrv_setacldsrpc(fh, cred, p, vp, &nmp[0],
4896 			    mirrorcnt, aclp, &failpos);
4897 		else if (ioproc == NFSPROC_SEEKDS) {
4898 			error = nfsrv_seekdsrpc(fh, offp, content, eofp, cred,
4899 			    p, nmp[0]);
4900 			if (nfsds_failerr(error) && mirrorcnt > 1) {
4901 				/*
4902 				 * Setting failpos will cause the mirror
4903 				 * to be disabled and then a retry of this
4904 				 * read is required.
4905 				 */
4906 				failpos = 0;
4907 				error = 0;
4908 				trycnt++;
4909 			}
4910 		} else if (ioproc == NFSPROC_ALLOCATE)
4911 			error = nfsrv_allocatedsrpc(fh, off, *offp, cred, p, vp,
4912 			    &nmp[0], mirrorcnt, &failpos);
4913 		else if (ioproc == NFSPROC_DEALLOCATE)
4914 			error = nfsrv_deallocatedsrpc(fh, off, *offp, cred, p,
4915 			    vp, &nmp[0], mirrorcnt, &failpos);
4916 		else {
4917 			error = nfsrv_getattrdsrpc(&fh[mirrorcnt - 1], cred, p,
4918 			    vp, nmp[mirrorcnt - 1], nap);
4919 			if (nfsds_failerr(error) && mirrorcnt > 1) {
4920 				/*
4921 				 * Setting failpos will cause the mirror
4922 				 * to be disabled and then a retry of this
4923 				 * getattr is required.
4924 				 */
4925 				failpos = mirrorcnt - 1;
4926 				error = 0;
4927 				trycnt++;
4928 			}
4929 		}
4930 		ds = NULL;
4931 		if (failpos >= 0) {
4932 			failnmp = nmp[failpos];
4933 			NFSLOCKMNT(failnmp);
4934 			if ((failnmp->nm_privflag & (NFSMNTP_FORCEDISM |
4935 			     NFSMNTP_CANCELRPCS)) == 0) {
4936 				failnmp->nm_privflag |= NFSMNTP_CANCELRPCS;
4937 				NFSUNLOCKMNT(failnmp);
4938 				ds = nfsrv_deldsnmp(PNFSDOP_DELDSSERVER,
4939 				    failnmp, p);
4940 				NFSD_DEBUG(4, "dsldsnmp fail=%d ds=%p\n",
4941 				    failpos, ds);
4942 				if (ds != NULL)
4943 					nfsrv_killrpcs(failnmp);
4944 				NFSLOCKMNT(failnmp);
4945 				failnmp->nm_privflag &= ~NFSMNTP_CANCELRPCS;
4946 				wakeup(failnmp);
4947 			}
4948 			NFSUNLOCKMNT(failnmp);
4949 		}
4950 		for (i = 0; i < mirrorcnt; i++)
4951 			NFSVOPUNLOCK(dvp[i]);
4952 		NFSD_DEBUG(4, "nfsrv_proxyds: aft RPC=%d trya=%d\n", error,
4953 		    trycnt);
4954 		/* Try the Read/Getattr again if a mirror was deleted. */
4955 		if (ds != NULL && trycnt > 0 && trycnt < origmircnt)
4956 			goto tryagain;
4957 	} else {
4958 		/* Return ENOENT for any Extended Attribute error. */
4959 		error = ENOENT;
4960 	}
4961 	free(buf, M_TEMP);
4962 	NFSD_DEBUG(4, "nfsrv_proxyds: error=%d\n", error);
4963 	return (error);
4964 }
4965 
4966 /*
4967  * Get the DS mount point, fh and directory from the "pnfsd.dsfile" extended
4968  * attribute.
4969  * newnmpp - If it points to a non-NULL nmp, that is the destination and needs
4970  *           to be checked.  If it points to a NULL nmp, then it returns
4971  *           a suitable destination.
4972  * curnmp - If non-NULL, it is the source mount for the copy.
4973  */
4974 int
4975 nfsrv_dsgetsockmnt(struct vnode *vp, int lktype, char *buf, int *buflenp,
4976     int *mirrorcntp, NFSPROC_T *p, struct vnode **dvpp, fhandle_t *fhp,
4977     char *devid, char *fnamep, struct vnode **nvpp, struct nfsmount **newnmpp,
4978     struct nfsmount *curnmp, int *ippos, int *dsdirp)
4979 {
4980 	struct vnode *dvp, *nvp = NULL, **tdvpp;
4981 	struct mount *mp;
4982 	struct nfsmount *nmp, *newnmp;
4983 	struct sockaddr *sad;
4984 	struct sockaddr_in *sin;
4985 	struct nfsdevice *ds, *tds, *fndds;
4986 	struct pnfsdsfile *pf;
4987 	uint32_t dsdir;
4988 	int error, fhiszero, fnd, gotone, i, mirrorcnt;
4989 
4990 	ASSERT_VOP_LOCKED(vp, "nfsrv_dsgetsockmnt vp");
4991 	*mirrorcntp = 1;
4992 	tdvpp = dvpp;
4993 	if (nvpp != NULL)
4994 		*nvpp = NULL;
4995 	if (dvpp != NULL)
4996 		*dvpp = NULL;
4997 	if (ippos != NULL)
4998 		*ippos = -1;
4999 	if (newnmpp != NULL)
5000 		newnmp = *newnmpp;
5001 	else
5002 		newnmp = NULL;
5003 	mp = vp->v_mount;
5004 	error = vn_extattr_get(vp, IO_NODELOCKED, EXTATTR_NAMESPACE_SYSTEM,
5005 	    "pnfsd.dsfile", buflenp, buf, p);
5006 	mirrorcnt = *buflenp / sizeof(*pf);
5007 	if (error == 0 && (mirrorcnt < 1 || mirrorcnt > NFSDEV_MAXMIRRORS ||
5008 	    *buflenp != sizeof(*pf) * mirrorcnt))
5009 		error = ENOATTR;
5010 
5011 	pf = (struct pnfsdsfile *)buf;
5012 	/* If curnmp != NULL, check for a match in the mirror list. */
5013 	if (curnmp != NULL && error == 0) {
5014 		fnd = 0;
5015 		for (i = 0; i < mirrorcnt; i++, pf++) {
5016 			sad = (struct sockaddr *)&pf->dsf_sin;
5017 			if (nfsaddr2_match(sad, curnmp->nm_nam)) {
5018 				if (ippos != NULL)
5019 					*ippos = i;
5020 				fnd = 1;
5021 				break;
5022 			}
5023 		}
5024 		if (fnd == 0)
5025 			error = ENXIO;
5026 	}
5027 
5028 	gotone = 0;
5029 	pf = (struct pnfsdsfile *)buf;
5030 	NFSD_DEBUG(4, "nfsrv_dsgetsockmnt: mirrorcnt=%d err=%d\n", mirrorcnt,
5031 	    error);
5032 	for (i = 0; i < mirrorcnt && error == 0; i++, pf++) {
5033 		fhiszero = 0;
5034 		sad = (struct sockaddr *)&pf->dsf_sin;
5035 		sin = &pf->dsf_sin;
5036 		dsdir = pf->dsf_dir;
5037 		if (dsdir >= nfsrv_dsdirsize) {
5038 			printf("nfsrv_dsgetsockmnt: dsdir=%d\n", dsdir);
5039 			error = ENOATTR;
5040 		} else if (nvpp != NULL && newnmp != NULL &&
5041 		    nfsaddr2_match(sad, newnmp->nm_nam))
5042 			error = EEXIST;
5043 		if (error == 0) {
5044 			if (ippos != NULL && curnmp == NULL &&
5045 			    sad->sa_family == AF_INET &&
5046 			    sin->sin_addr.s_addr == 0)
5047 				*ippos = i;
5048 			if (NFSBCMP(&zerofh, &pf->dsf_fh, sizeof(zerofh)) == 0)
5049 				fhiszero = 1;
5050 			/* Use the socket address to find the mount point. */
5051 			fndds = NULL;
5052 			NFSDDSLOCK();
5053 			/* Find a match for the IP address. */
5054 			TAILQ_FOREACH(ds, &nfsrv_devidhead, nfsdev_list) {
5055 				if (ds->nfsdev_nmp != NULL) {
5056 					dvp = ds->nfsdev_dvp;
5057 					nmp = VFSTONFS(dvp->v_mount);
5058 					if (nmp != ds->nfsdev_nmp)
5059 						printf("different2 nmp %p %p\n",
5060 						    nmp, ds->nfsdev_nmp);
5061 					if (nfsaddr2_match(sad, nmp->nm_nam)) {
5062 						fndds = ds;
5063 						break;
5064 					}
5065 				}
5066 			}
5067 			if (fndds != NULL && newnmpp != NULL &&
5068 			    newnmp == NULL) {
5069 				/* Search for a place to make a mirror copy. */
5070 				TAILQ_FOREACH(tds, &nfsrv_devidhead,
5071 				    nfsdev_list) {
5072 					if (tds->nfsdev_nmp != NULL &&
5073 					    fndds != tds &&
5074 					    ((tds->nfsdev_mdsisset == 0 &&
5075 					      fndds->nfsdev_mdsisset == 0) ||
5076 					     (tds->nfsdev_mdsisset != 0 &&
5077 					      fndds->nfsdev_mdsisset != 0 &&
5078 					      fsidcmp(&tds->nfsdev_mdsfsid,
5079 					      &mp->mnt_stat.f_fsid) == 0))) {
5080 						*newnmpp = tds->nfsdev_nmp;
5081 						break;
5082 					}
5083 				}
5084 				if (tds != NULL) {
5085 					/*
5086 					 * Move this entry to the end of the
5087 					 * list, so it won't be selected as
5088 					 * easily the next time.
5089 					 */
5090 					TAILQ_REMOVE(&nfsrv_devidhead, tds,
5091 					    nfsdev_list);
5092 					TAILQ_INSERT_TAIL(&nfsrv_devidhead, tds,
5093 					    nfsdev_list);
5094 				}
5095 			}
5096 			NFSDDSUNLOCK();
5097 			if (fndds != NULL) {
5098 				dvp = fndds->nfsdev_dsdir[dsdir];
5099 				if (lktype != 0 || fhiszero != 0 ||
5100 				    (nvpp != NULL && *nvpp == NULL)) {
5101 					if (fhiszero != 0)
5102 						error = vn_lock(dvp,
5103 						    LK_EXCLUSIVE);
5104 					else if (lktype != 0)
5105 						error = vn_lock(dvp, lktype);
5106 					else
5107 						error = vn_lock(dvp, LK_SHARED);
5108 					/*
5109 					 * If the file handle is all 0's, try to
5110 					 * do a Lookup against the DS to acquire
5111 					 * it.
5112 					 * If dvpp == NULL or the Lookup fails,
5113 					 * unlock dvp after the call.
5114 					 */
5115 					if (error == 0 && (fhiszero != 0 ||
5116 					    (nvpp != NULL && *nvpp == NULL))) {
5117 						error = nfsrv_pnfslookupds(vp,
5118 						    dvp, pf, &nvp, p);
5119 						if (error == 0) {
5120 							if (fhiszero != 0)
5121 								nfsrv_pnfssetfh(
5122 								    vp, pf,
5123 								    devid,
5124 								    fnamep,
5125 								    nvp, p);
5126 							if (nvpp != NULL &&
5127 							    *nvpp == NULL) {
5128 								*nvpp = nvp;
5129 								*dsdirp = dsdir;
5130 							} else
5131 								vput(nvp);
5132 						}
5133 						if (error != 0 || lktype == 0)
5134 							NFSVOPUNLOCK(dvp);
5135 					}
5136 				}
5137 				if (error == 0) {
5138 					gotone++;
5139 					NFSD_DEBUG(4, "gotone=%d\n", gotone);
5140 					if (devid != NULL) {
5141 						NFSBCOPY(fndds->nfsdev_deviceid,
5142 						    devid, NFSX_V4DEVICEID);
5143 						devid += NFSX_V4DEVICEID;
5144 					}
5145 					if (dvpp != NULL)
5146 						*tdvpp++ = dvp;
5147 					if (fhp != NULL)
5148 						NFSBCOPY(&pf->dsf_fh, fhp++,
5149 						    NFSX_MYFH);
5150 					if (fnamep != NULL && gotone == 1)
5151 						strlcpy(fnamep,
5152 						    pf->dsf_filename,
5153 						    sizeof(pf->dsf_filename));
5154 				} else
5155 					NFSD_DEBUG(4, "nfsrv_dsgetsockmnt "
5156 					    "err=%d\n", error);
5157 			}
5158 		}
5159 	}
5160 	if (error == 0 && gotone == 0)
5161 		error = ENOENT;
5162 
5163 	NFSD_DEBUG(4, "eo nfsrv_dsgetsockmnt: gotone=%d err=%d\n", gotone,
5164 	    error);
5165 	if (error == 0)
5166 		*mirrorcntp = gotone;
5167 	else {
5168 		if (gotone > 0 && dvpp != NULL) {
5169 			/*
5170 			 * If the error didn't occur on the first one and
5171 			 * dvpp != NULL, the one(s) prior to the failure will
5172 			 * have locked dvp's that need to be unlocked.
5173 			 */
5174 			for (i = 0; i < gotone; i++) {
5175 				NFSVOPUNLOCK(*dvpp);
5176 				*dvpp++ = NULL;
5177 			}
5178 		}
5179 		/*
5180 		 * If it found the vnode to be copied from before a failure,
5181 		 * it needs to be vput()'d.
5182 		 */
5183 		if (nvpp != NULL && *nvpp != NULL) {
5184 			vput(*nvpp);
5185 			*nvpp = NULL;
5186 		}
5187 	}
5188 	return (error);
5189 }
5190 
5191 /*
5192  * Set the extended attribute for the Change attribute.
5193  */
5194 static int
5195 nfsrv_setextattr(struct vnode *vp, struct nfsvattr *nap, NFSPROC_T *p)
5196 {
5197 	struct pnfsdsattr dsattr;
5198 	int error;
5199 
5200 	ASSERT_VOP_ELOCKED(vp, "nfsrv_setextattr vp");
5201 	dsattr.dsa_filerev = nap->na_filerev;
5202 	dsattr.dsa_size = nap->na_size;
5203 	dsattr.dsa_atime = nap->na_atime;
5204 	dsattr.dsa_mtime = nap->na_mtime;
5205 	dsattr.dsa_bytes = nap->na_bytes;
5206 	error = vn_extattr_set(vp, IO_NODELOCKED, EXTATTR_NAMESPACE_SYSTEM,
5207 	    "pnfsd.dsattr", sizeof(dsattr), (char *)&dsattr, p);
5208 	if (error != 0)
5209 		printf("pNFS: setextattr=%d\n", error);
5210 	return (error);
5211 }
5212 
5213 static int
5214 nfsrv_readdsrpc(fhandle_t *fhp, off_t off, int len, struct ucred *cred,
5215     NFSPROC_T *p, struct nfsmount *nmp, struct mbuf **mpp, struct mbuf **mpendp)
5216 {
5217 	uint32_t *tl;
5218 	struct nfsrv_descript *nd;
5219 	nfsv4stateid_t st;
5220 	struct mbuf *m, *m2;
5221 	int error = 0, retlen, tlen, trimlen;
5222 
5223 	NFSD_DEBUG(4, "in nfsrv_readdsrpc\n");
5224 	nd = malloc(sizeof(*nd), M_TEMP, M_WAITOK | M_ZERO);
5225 	*mpp = NULL;
5226 	/*
5227 	 * Use a stateid where other is an alternating 01010 pattern and
5228 	 * seqid is 0xffffffff.  This value is not defined as special by
5229 	 * the RFC and is used by the FreeBSD NFS server to indicate an
5230 	 * MDS->DS proxy operation.
5231 	 */
5232 	st.other[0] = 0x55555555;
5233 	st.other[1] = 0x55555555;
5234 	st.other[2] = 0x55555555;
5235 	st.seqid = 0xffffffff;
5236 	nfscl_reqstart(nd, NFSPROC_READDS, nmp, (u_int8_t *)fhp, sizeof(*fhp),
5237 	    NULL, NULL, 0, 0, cred);
5238 	nfsm_stateidtom(nd, &st, NFSSTATEID_PUTSTATEID);
5239 	NFSM_BUILD(tl, uint32_t *, NFSX_UNSIGNED * 3);
5240 	txdr_hyper(off, tl);
5241 	*(tl + 2) = txdr_unsigned(len);
5242 	error = newnfs_request(nd, nmp, NULL, &nmp->nm_sockreq, NULL, p, cred,
5243 	    NFS_PROG, NFS_VER4, NULL, 1, NULL, NULL);
5244 	if (error != 0) {
5245 		free(nd, M_TEMP);
5246 		return (error);
5247 	}
5248 	if (nd->nd_repstat == 0) {
5249 		NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
5250 		NFSM_STRSIZ(retlen, len);
5251 		if (retlen > 0) {
5252 			/* Trim off the pre-data XDR from the mbuf chain. */
5253 			m = nd->nd_mrep;
5254 			while (m != NULL && m != nd->nd_md) {
5255 				if (m->m_next == nd->nd_md) {
5256 					m->m_next = NULL;
5257 					m_freem(nd->nd_mrep);
5258 					nd->nd_mrep = m = nd->nd_md;
5259 				} else
5260 					m = m->m_next;
5261 			}
5262 			if (m == NULL) {
5263 				printf("nfsrv_readdsrpc: busted mbuf list\n");
5264 				error = ENOENT;
5265 				goto nfsmout;
5266 			}
5267 
5268 			/*
5269 			 * Now, adjust first mbuf so that any XDR before the
5270 			 * read data is skipped over.
5271 			 */
5272 			trimlen = nd->nd_dpos - mtod(m, char *);
5273 			if (trimlen > 0) {
5274 				m->m_len -= trimlen;
5275 				NFSM_DATAP(m, trimlen);
5276 			}
5277 
5278 			/*
5279 			 * Truncate the mbuf chain at retlen bytes of data,
5280 			 * plus XDR padding that brings the length up to a
5281 			 * multiple of 4.
5282 			 */
5283 			tlen = NFSM_RNDUP(retlen);
5284 			do {
5285 				if (m->m_len >= tlen) {
5286 					m->m_len = tlen;
5287 					tlen = 0;
5288 					m2 = m->m_next;
5289 					m->m_next = NULL;
5290 					m_freem(m2);
5291 					break;
5292 				}
5293 				tlen -= m->m_len;
5294 				m = m->m_next;
5295 			} while (m != NULL);
5296 			if (tlen > 0) {
5297 				printf("nfsrv_readdsrpc: busted mbuf list\n");
5298 				error = ENOENT;
5299 				goto nfsmout;
5300 			}
5301 			*mpp = nd->nd_mrep;
5302 			*mpendp = m;
5303 			nd->nd_mrep = NULL;
5304 		}
5305 	} else
5306 		error = nd->nd_repstat;
5307 nfsmout:
5308 	/* If nd->nd_mrep is already NULL, this is a no-op. */
5309 	m_freem(nd->nd_mrep);
5310 	free(nd, M_TEMP);
5311 	NFSD_DEBUG(4, "nfsrv_readdsrpc error=%d\n", error);
5312 	return (error);
5313 }
5314 
5315 /*
5316  * Do a write RPC on a DS data file, using this structure for the arguments,
5317  * so that this function can be executed by a separate kernel process.
5318  */
5319 struct nfsrvwritedsdorpc {
5320 	int			done;
5321 	int			inprog;
5322 	struct task		tsk;
5323 	fhandle_t		fh;
5324 	off_t			off;
5325 	int			len;
5326 	struct nfsmount		*nmp;
5327 	struct ucred		*cred;
5328 	NFSPROC_T		*p;
5329 	struct mbuf		*m;
5330 	int			err;
5331 };
5332 
5333 static int
5334 nfsrv_writedsdorpc(struct nfsmount *nmp, fhandle_t *fhp, off_t off, int len,
5335     struct nfsvattr *nap, struct mbuf *m, struct ucred *cred, NFSPROC_T *p)
5336 {
5337 	uint32_t *tl;
5338 	struct nfsrv_descript *nd;
5339 	nfsattrbit_t attrbits;
5340 	nfsv4stateid_t st;
5341 	int commit, error, retlen;
5342 
5343 	nd = malloc(sizeof(*nd), M_TEMP, M_WAITOK | M_ZERO);
5344 	nfscl_reqstart(nd, NFSPROC_WRITE, nmp, (u_int8_t *)fhp,
5345 	    sizeof(fhandle_t), NULL, NULL, 0, 0, cred);
5346 
5347 	/*
5348 	 * Use a stateid where other is an alternating 01010 pattern and
5349 	 * seqid is 0xffffffff.  This value is not defined as special by
5350 	 * the RFC and is used by the FreeBSD NFS server to indicate an
5351 	 * MDS->DS proxy operation.
5352 	 */
5353 	st.other[0] = 0x55555555;
5354 	st.other[1] = 0x55555555;
5355 	st.other[2] = 0x55555555;
5356 	st.seqid = 0xffffffff;
5357 	nfsm_stateidtom(nd, &st, NFSSTATEID_PUTSTATEID);
5358 	NFSM_BUILD(tl, u_int32_t *, NFSX_HYPER + 2 * NFSX_UNSIGNED);
5359 	txdr_hyper(off, tl);
5360 	tl += 2;
5361 	/*
5362 	 * Do all writes FileSync, since the server doesn't hold onto dirty
5363 	 * buffers.  Since clients should be accessing the DS servers directly
5364 	 * using the pNFS layouts, this just needs to work correctly as a
5365 	 * fallback.
5366 	 */
5367 	*tl++ = txdr_unsigned(NFSWRITE_FILESYNC);
5368 	*tl = txdr_unsigned(len);
5369 	NFSD_DEBUG(4, "nfsrv_writedsdorpc: len=%d\n", len);
5370 
5371 	/* Put data in mbuf chain. */
5372 	nd->nd_mb->m_next = m;
5373 
5374 	/* Set nd_mb and nd_bpos to end of data. */
5375 	while (m->m_next != NULL)
5376 		m = m->m_next;
5377 	nd->nd_mb = m;
5378 	nfsm_set(nd, m->m_len);
5379 	NFSD_DEBUG(4, "nfsrv_writedsdorpc: lastmb len=%d\n", m->m_len);
5380 
5381 	/* Do a Getattr for the attributes that change upon writing. */
5382 	NFSZERO_ATTRBIT(&attrbits);
5383 	NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_SIZE);
5384 	NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_CHANGE);
5385 	NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_TIMEACCESS);
5386 	NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_TIMEMODIFY);
5387 	NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_SPACEUSED);
5388 	NFSM_BUILD(tl, u_int32_t *, NFSX_UNSIGNED);
5389 	*tl = txdr_unsigned(NFSV4OP_GETATTR);
5390 	(void) nfsrv_putattrbit(nd, &attrbits);
5391 	error = newnfs_request(nd, nmp, NULL, &nmp->nm_sockreq, NULL, p,
5392 	    cred, NFS_PROG, NFS_VER4, NULL, 1, NULL, NULL);
5393 	if (error != 0) {
5394 		free(nd, M_TEMP);
5395 		return (error);
5396 	}
5397 	NFSD_DEBUG(4, "nfsrv_writedsdorpc: aft writerpc=%d\n", nd->nd_repstat);
5398 	/* Get rid of weak cache consistency data for now. */
5399 	if ((nd->nd_flag & (ND_NOMOREDATA | ND_NFSV4 | ND_V4WCCATTR)) ==
5400 	    (ND_NFSV4 | ND_V4WCCATTR)) {
5401 		error = nfsv4_loadattr(nd, NULL, nap, NULL, NULL, 0, NULL, NULL,
5402 		    NULL, NULL, NULL, 0, NULL, NULL, NULL, NULL, NULL);
5403 		NFSD_DEBUG(4, "nfsrv_writedsdorpc: wcc attr=%d\n", error);
5404 		if (error != 0)
5405 			goto nfsmout;
5406 		/*
5407 		 * Get rid of Op# and status for next op.
5408 		 */
5409 		NFSM_DISSECT(tl, uint32_t *, 2 * NFSX_UNSIGNED);
5410 		if (*++tl != 0)
5411 			nd->nd_flag |= ND_NOMOREDATA;
5412 	}
5413 	if (nd->nd_repstat == 0) {
5414 		NFSM_DISSECT(tl, uint32_t *, 2 * NFSX_UNSIGNED + NFSX_VERF);
5415 		retlen = fxdr_unsigned(int, *tl++);
5416 		commit = fxdr_unsigned(int, *tl);
5417 		if (commit != NFSWRITE_FILESYNC)
5418 			error = NFSERR_IO;
5419 		NFSD_DEBUG(4, "nfsrv_writedsdorpc:retlen=%d commit=%d err=%d\n",
5420 		    retlen, commit, error);
5421 	} else
5422 		error = nd->nd_repstat;
5423 	/* We have no use for the Write Verifier since we use FileSync. */
5424 
5425 	/*
5426 	 * Get the Change, Size, Access Time and Modify Time attributes and set
5427 	 * on the Metadata file, so its attributes will be what the file's
5428 	 * would be if it had been written.
5429 	 */
5430 	if (error == 0) {
5431 		NFSM_DISSECT(tl, uint32_t *, 2 * NFSX_UNSIGNED);
5432 		error = nfsv4_loadattr(nd, NULL, nap, NULL, NULL, 0, NULL, NULL,
5433 		    NULL, NULL, NULL, 0, NULL, NULL, NULL, NULL, NULL);
5434 	}
5435 	NFSD_DEBUG(4, "nfsrv_writedsdorpc: aft loadattr=%d\n", error);
5436 nfsmout:
5437 	m_freem(nd->nd_mrep);
5438 	free(nd, M_TEMP);
5439 	NFSD_DEBUG(4, "nfsrv_writedsdorpc error=%d\n", error);
5440 	return (error);
5441 }
5442 
5443 /*
5444  * Start up the thread that will execute nfsrv_writedsdorpc().
5445  */
5446 static void
5447 start_writedsdorpc(void *arg, int pending)
5448 {
5449 	struct nfsrvwritedsdorpc *drpc;
5450 
5451 	drpc = (struct nfsrvwritedsdorpc *)arg;
5452 	drpc->err = nfsrv_writedsdorpc(drpc->nmp, &drpc->fh, drpc->off,
5453 	    drpc->len, NULL, drpc->m, drpc->cred, drpc->p);
5454 	drpc->done = 1;
5455 	NFSD_DEBUG(4, "start_writedsdorpc: err=%d\n", drpc->err);
5456 }
5457 
5458 static int
5459 nfsrv_writedsrpc(fhandle_t *fhp, off_t off, int len, struct ucred *cred,
5460     NFSPROC_T *p, struct vnode *vp, struct nfsmount **nmpp, int mirrorcnt,
5461     struct mbuf **mpp, char *cp, int *failposp)
5462 {
5463 	struct nfsrvwritedsdorpc *drpc, *tdrpc = NULL;
5464 	struct nfsvattr na;
5465 	struct mbuf *m;
5466 	int error, i, offs, ret, timo;
5467 
5468 	NFSD_DEBUG(4, "in nfsrv_writedsrpc\n");
5469 	KASSERT(*mpp != NULL, ("nfsrv_writedsrpc: NULL mbuf chain"));
5470 	drpc = NULL;
5471 	if (mirrorcnt > 1)
5472 		tdrpc = drpc = malloc(sizeof(*drpc) * (mirrorcnt - 1), M_TEMP,
5473 		    M_WAITOK);
5474 
5475 	/* Calculate offset in mbuf chain that data starts. */
5476 	offs = cp - mtod(*mpp, char *);
5477 	NFSD_DEBUG(4, "nfsrv_writedsrpc: mcopy offs=%d len=%d\n", offs, len);
5478 
5479 	/*
5480 	 * Do the write RPC for every DS, using a separate kernel process
5481 	 * for every DS except the last one.
5482 	 */
5483 	error = 0;
5484 	for (i = 0; i < mirrorcnt - 1; i++, tdrpc++) {
5485 		tdrpc->done = 0;
5486 		NFSBCOPY(fhp, &tdrpc->fh, sizeof(*fhp));
5487 		tdrpc->off = off;
5488 		tdrpc->len = len;
5489 		tdrpc->nmp = *nmpp;
5490 		tdrpc->cred = cred;
5491 		tdrpc->p = p;
5492 		tdrpc->inprog = 0;
5493 		tdrpc->err = 0;
5494 		tdrpc->m = m_copym(*mpp, offs, NFSM_RNDUP(len), M_WAITOK);
5495 		ret = EIO;
5496 		if (nfs_pnfsiothreads != 0) {
5497 			ret = nfs_pnfsio(start_writedsdorpc, tdrpc);
5498 			NFSD_DEBUG(4, "nfsrv_writedsrpc: nfs_pnfsio=%d\n",
5499 			    ret);
5500 		}
5501 		if (ret != 0) {
5502 			ret = nfsrv_writedsdorpc(*nmpp, fhp, off, len, NULL,
5503 			    tdrpc->m, cred, p);
5504 			if (nfsds_failerr(ret) && *failposp == -1)
5505 				*failposp = i;
5506 			else if (error == 0 && ret != 0)
5507 				error = ret;
5508 		}
5509 		nmpp++;
5510 		fhp++;
5511 	}
5512 	m = m_copym(*mpp, offs, NFSM_RNDUP(len), M_WAITOK);
5513 	ret = nfsrv_writedsdorpc(*nmpp, fhp, off, len, &na, m, cred, p);
5514 	if (nfsds_failerr(ret) && *failposp == -1 && mirrorcnt > 1)
5515 		*failposp = mirrorcnt - 1;
5516 	else if (error == 0 && ret != 0)
5517 		error = ret;
5518 	if (error == 0)
5519 		error = nfsrv_setextattr(vp, &na, p);
5520 	NFSD_DEBUG(4, "nfsrv_writedsrpc: aft setextat=%d\n", error);
5521 	tdrpc = drpc;
5522 	timo = hz / 50;		/* Wait for 20msec. */
5523 	if (timo < 1)
5524 		timo = 1;
5525 	for (i = 0; i < mirrorcnt - 1; i++, tdrpc++) {
5526 		/* Wait for RPCs on separate threads to complete. */
5527 		while (tdrpc->inprog != 0 && tdrpc->done == 0)
5528 			tsleep(&tdrpc->tsk, PVFS, "srvwrds", timo);
5529 		if (nfsds_failerr(tdrpc->err) && *failposp == -1)
5530 			*failposp = i;
5531 		else if (error == 0 && tdrpc->err != 0)
5532 			error = tdrpc->err;
5533 	}
5534 	free(drpc, M_TEMP);
5535 	return (error);
5536 }
5537 
5538 /*
5539  * Do a allocate RPC on a DS data file, using this structure for the arguments,
5540  * so that this function can be executed by a separate kernel process.
5541  */
5542 struct nfsrvallocatedsdorpc {
5543 	int			done;
5544 	int			inprog;
5545 	struct task		tsk;
5546 	fhandle_t		fh;
5547 	off_t			off;
5548 	off_t			len;
5549 	struct nfsmount		*nmp;
5550 	struct ucred		*cred;
5551 	NFSPROC_T		*p;
5552 	int			err;
5553 };
5554 
5555 static int
5556 nfsrv_allocatedsdorpc(struct nfsmount *nmp, fhandle_t *fhp, off_t off,
5557     off_t len, struct nfsvattr *nap, struct ucred *cred, NFSPROC_T *p)
5558 {
5559 	uint32_t *tl;
5560 	struct nfsrv_descript *nd;
5561 	nfsattrbit_t attrbits;
5562 	nfsv4stateid_t st;
5563 	int error;
5564 
5565 	nd = malloc(sizeof(*nd), M_TEMP, M_WAITOK | M_ZERO);
5566 	nfscl_reqstart(nd, NFSPROC_ALLOCATE, nmp, (u_int8_t *)fhp,
5567 	    sizeof(fhandle_t), NULL, NULL, 0, 0, cred);
5568 
5569 	/*
5570 	 * Use a stateid where other is an alternating 01010 pattern and
5571 	 * seqid is 0xffffffff.  This value is not defined as special by
5572 	 * the RFC and is used by the FreeBSD NFS server to indicate an
5573 	 * MDS->DS proxy operation.
5574 	 */
5575 	st.other[0] = 0x55555555;
5576 	st.other[1] = 0x55555555;
5577 	st.other[2] = 0x55555555;
5578 	st.seqid = 0xffffffff;
5579 	nfsm_stateidtom(nd, &st, NFSSTATEID_PUTSTATEID);
5580 	NFSM_BUILD(tl, uint32_t *, 2 * NFSX_HYPER + NFSX_UNSIGNED);
5581 	txdr_hyper(off, tl); tl += 2;
5582 	txdr_hyper(len, tl); tl += 2;
5583 	NFSD_DEBUG(4, "nfsrv_allocatedsdorpc: len=%jd\n", (intmax_t)len);
5584 
5585 	*tl = txdr_unsigned(NFSV4OP_GETATTR);
5586 	NFSGETATTR_ATTRBIT(&attrbits);
5587 	nfsrv_putattrbit(nd, &attrbits);
5588 	error = newnfs_request(nd, nmp, NULL, &nmp->nm_sockreq, NULL, p,
5589 	    cred, NFS_PROG, NFS_VER4, NULL, 1, NULL, NULL);
5590 	if (error != 0) {
5591 		free(nd, M_TEMP);
5592 		return (error);
5593 	}
5594 	NFSD_DEBUG(4, "nfsrv_allocatedsdorpc: aft allocaterpc=%d\n",
5595 	    nd->nd_repstat);
5596 	if (nd->nd_repstat == 0) {
5597 		NFSM_DISSECT(tl, uint32_t *, 2 * NFSX_UNSIGNED);
5598 		error = nfsv4_loadattr(nd, NULL, nap, NULL, NULL, 0, NULL, NULL,
5599 		    NULL, NULL, NULL, 0, NULL, NULL, NULL, NULL, NULL);
5600 	} else
5601 		error = nd->nd_repstat;
5602 	NFSD_DEBUG(4, "nfsrv_allocatedsdorpc: aft loadattr=%d\n", error);
5603 nfsmout:
5604 	m_freem(nd->nd_mrep);
5605 	free(nd, M_TEMP);
5606 	NFSD_DEBUG(4, "nfsrv_allocatedsdorpc error=%d\n", error);
5607 	return (error);
5608 }
5609 
5610 /*
5611  * Start up the thread that will execute nfsrv_allocatedsdorpc().
5612  */
5613 static void
5614 start_allocatedsdorpc(void *arg, int pending)
5615 {
5616 	struct nfsrvallocatedsdorpc *drpc;
5617 
5618 	drpc = (struct nfsrvallocatedsdorpc *)arg;
5619 	drpc->err = nfsrv_allocatedsdorpc(drpc->nmp, &drpc->fh, drpc->off,
5620 	    drpc->len, NULL, drpc->cred, drpc->p);
5621 	drpc->done = 1;
5622 	NFSD_DEBUG(4, "start_allocatedsdorpc: err=%d\n", drpc->err);
5623 }
5624 
5625 static int
5626 nfsrv_allocatedsrpc(fhandle_t *fhp, off_t off, off_t len, struct ucred *cred,
5627     NFSPROC_T *p, struct vnode *vp, struct nfsmount **nmpp, int mirrorcnt,
5628     int *failposp)
5629 {
5630 	struct nfsrvallocatedsdorpc *drpc, *tdrpc = NULL;
5631 	struct nfsvattr na;
5632 	int error, i, ret, timo;
5633 
5634 	NFSD_DEBUG(4, "in nfsrv_allocatedsrpc\n");
5635 	drpc = NULL;
5636 	if (mirrorcnt > 1)
5637 		tdrpc = drpc = malloc(sizeof(*drpc) * (mirrorcnt - 1), M_TEMP,
5638 		    M_WAITOK);
5639 
5640 	/*
5641 	 * Do the allocate RPC for every DS, using a separate kernel process
5642 	 * for every DS except the last one.
5643 	 */
5644 	error = 0;
5645 	for (i = 0; i < mirrorcnt - 1; i++, tdrpc++) {
5646 		tdrpc->done = 0;
5647 		NFSBCOPY(fhp, &tdrpc->fh, sizeof(*fhp));
5648 		tdrpc->off = off;
5649 		tdrpc->len = len;
5650 		tdrpc->nmp = *nmpp;
5651 		tdrpc->cred = cred;
5652 		tdrpc->p = p;
5653 		tdrpc->inprog = 0;
5654 		tdrpc->err = 0;
5655 		ret = EIO;
5656 		if (nfs_pnfsiothreads != 0) {
5657 			ret = nfs_pnfsio(start_allocatedsdorpc, tdrpc);
5658 			NFSD_DEBUG(4, "nfsrv_allocatedsrpc: nfs_pnfsio=%d\n",
5659 			    ret);
5660 		}
5661 		if (ret != 0) {
5662 			ret = nfsrv_allocatedsdorpc(*nmpp, fhp, off, len, NULL,
5663 			    cred, p);
5664 			if (nfsds_failerr(ret) && *failposp == -1)
5665 				*failposp = i;
5666 			else if (error == 0 && ret != 0)
5667 				error = ret;
5668 		}
5669 		nmpp++;
5670 		fhp++;
5671 	}
5672 	ret = nfsrv_allocatedsdorpc(*nmpp, fhp, off, len, &na, cred, p);
5673 	if (nfsds_failerr(ret) && *failposp == -1 && mirrorcnt > 1)
5674 		*failposp = mirrorcnt - 1;
5675 	else if (error == 0 && ret != 0)
5676 		error = ret;
5677 	if (error == 0)
5678 		error = nfsrv_setextattr(vp, &na, p);
5679 	NFSD_DEBUG(4, "nfsrv_allocatedsrpc: aft setextat=%d\n", error);
5680 	tdrpc = drpc;
5681 	timo = hz / 50;		/* Wait for 20msec. */
5682 	if (timo < 1)
5683 		timo = 1;
5684 	for (i = 0; i < mirrorcnt - 1; i++, tdrpc++) {
5685 		/* Wait for RPCs on separate threads to complete. */
5686 		while (tdrpc->inprog != 0 && tdrpc->done == 0)
5687 			tsleep(&tdrpc->tsk, PVFS, "srvalds", timo);
5688 		if (nfsds_failerr(tdrpc->err) && *failposp == -1)
5689 			*failposp = i;
5690 		else if (error == 0 && tdrpc->err != 0)
5691 			error = tdrpc->err;
5692 	}
5693 	free(drpc, M_TEMP);
5694 	return (error);
5695 }
5696 
5697 /*
5698  * Do a deallocate RPC on a DS data file, using this structure for the
5699  * arguments, so that this function can be executed by a separate kernel
5700  * process.
5701  */
5702 struct nfsrvdeallocatedsdorpc {
5703 	int			done;
5704 	int			inprog;
5705 	struct task		tsk;
5706 	fhandle_t		fh;
5707 	off_t			off;
5708 	off_t			len;
5709 	struct nfsmount		*nmp;
5710 	struct ucred		*cred;
5711 	NFSPROC_T		*p;
5712 	int			err;
5713 };
5714 
5715 static int
5716 nfsrv_deallocatedsdorpc(struct nfsmount *nmp, fhandle_t *fhp, off_t off,
5717     off_t len, struct nfsvattr *nap, struct ucred *cred, NFSPROC_T *p)
5718 {
5719 	uint32_t *tl;
5720 	struct nfsrv_descript *nd;
5721 	nfsattrbit_t attrbits;
5722 	nfsv4stateid_t st;
5723 	int error;
5724 
5725 	nd = malloc(sizeof(*nd), M_TEMP, M_WAITOK | M_ZERO);
5726 	nfscl_reqstart(nd, NFSPROC_DEALLOCATE, nmp, (u_int8_t *)fhp,
5727 	    sizeof(fhandle_t), NULL, NULL, 0, 0, cred);
5728 
5729 	/*
5730 	 * Use a stateid where other is an alternating 01010 pattern and
5731 	 * seqid is 0xffffffff.  This value is not defined as special by
5732 	 * the RFC and is used by the FreeBSD NFS server to indicate an
5733 	 * MDS->DS proxy operation.
5734 	 */
5735 	st.other[0] = 0x55555555;
5736 	st.other[1] = 0x55555555;
5737 	st.other[2] = 0x55555555;
5738 	st.seqid = 0xffffffff;
5739 	nfsm_stateidtom(nd, &st, NFSSTATEID_PUTSTATEID);
5740 	NFSM_BUILD(tl, uint32_t *, 2 * NFSX_HYPER + NFSX_UNSIGNED);
5741 	txdr_hyper(off, tl); tl += 2;
5742 	txdr_hyper(len, tl); tl += 2;
5743 	NFSD_DEBUG(4, "nfsrv_deallocatedsdorpc: len=%jd\n", (intmax_t)len);
5744 
5745 	/* Do a Getattr for the attributes that change upon writing. */
5746 	NFSZERO_ATTRBIT(&attrbits);
5747 	NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_SIZE);
5748 	NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_CHANGE);
5749 	NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_TIMEACCESS);
5750 	NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_TIMEMODIFY);
5751 	NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_SPACEUSED);
5752 	*tl = txdr_unsigned(NFSV4OP_GETATTR);
5753 	nfsrv_putattrbit(nd, &attrbits);
5754 	error = newnfs_request(nd, nmp, NULL, &nmp->nm_sockreq, NULL, p,
5755 	    cred, NFS_PROG, NFS_VER4, NULL, 1, NULL, NULL);
5756 	if (error != 0) {
5757 		free(nd, M_TEMP);
5758 		return (error);
5759 	}
5760 	NFSD_DEBUG(4, "nfsrv_deallocatedsdorpc: aft deallocaterpc=%d\n",
5761 	    nd->nd_repstat);
5762 	/* Get rid of weak cache consistency data for now. */
5763 	if ((nd->nd_flag & (ND_NOMOREDATA | ND_NFSV4 | ND_V4WCCATTR)) ==
5764 	    (ND_NFSV4 | ND_V4WCCATTR)) {
5765 		error = nfsv4_loadattr(nd, NULL, nap, NULL, NULL, 0, NULL, NULL,
5766 		    NULL, NULL, NULL, 0, NULL, NULL, NULL, NULL, NULL);
5767 		NFSD_DEBUG(4, "nfsrv_deallocatedsdorpc: wcc attr=%d\n", error);
5768 		if (error != 0)
5769 			goto nfsmout;
5770 		/*
5771 		 * Get rid of Op# and status for next op.
5772 		 */
5773 		NFSM_DISSECT(tl, uint32_t *, 2 * NFSX_UNSIGNED);
5774 		if (*++tl != 0)
5775 			nd->nd_flag |= ND_NOMOREDATA;
5776 	}
5777 	if (nd->nd_repstat == 0) {
5778 		NFSM_DISSECT(tl, uint32_t *, 2 * NFSX_UNSIGNED);
5779 		error = nfsv4_loadattr(nd, NULL, nap, NULL, NULL, 0, NULL, NULL,
5780 		    NULL, NULL, NULL, 0, NULL, NULL, NULL, NULL, NULL);
5781 	} else
5782 		error = nd->nd_repstat;
5783 	NFSD_DEBUG(4, "nfsrv_deallocatedsdorpc: aft loadattr=%d\n", error);
5784 nfsmout:
5785 	m_freem(nd->nd_mrep);
5786 	free(nd, M_TEMP);
5787 	NFSD_DEBUG(4, "nfsrv_deallocatedsdorpc error=%d\n", error);
5788 	return (error);
5789 }
5790 
5791 /*
5792  * Start up the thread that will execute nfsrv_deallocatedsdorpc().
5793  */
5794 static void
5795 start_deallocatedsdorpc(void *arg, int pending)
5796 {
5797 	struct nfsrvdeallocatedsdorpc *drpc;
5798 
5799 	drpc = (struct nfsrvdeallocatedsdorpc *)arg;
5800 	drpc->err = nfsrv_deallocatedsdorpc(drpc->nmp, &drpc->fh, drpc->off,
5801 	    drpc->len, NULL, drpc->cred, drpc->p);
5802 	drpc->done = 1;
5803 	NFSD_DEBUG(4, "start_deallocatedsdorpc: err=%d\n", drpc->err);
5804 }
5805 
5806 static int
5807 nfsrv_deallocatedsrpc(fhandle_t *fhp, off_t off, off_t len, struct ucred *cred,
5808     NFSPROC_T *p, struct vnode *vp, struct nfsmount **nmpp, int mirrorcnt,
5809     int *failposp)
5810 {
5811 	struct nfsrvdeallocatedsdorpc *drpc, *tdrpc = NULL;
5812 	struct nfsvattr na;
5813 	int error, i, ret, timo;
5814 
5815 	NFSD_DEBUG(4, "in nfsrv_deallocatedsrpc\n");
5816 	drpc = NULL;
5817 	if (mirrorcnt > 1)
5818 		tdrpc = drpc = malloc(sizeof(*drpc) * (mirrorcnt - 1), M_TEMP,
5819 		    M_WAITOK);
5820 
5821 	/*
5822 	 * Do the deallocate RPC for every DS, using a separate kernel process
5823 	 * for every DS except the last one.
5824 	 */
5825 	error = 0;
5826 	for (i = 0; i < mirrorcnt - 1; i++, tdrpc++) {
5827 		tdrpc->done = 0;
5828 		NFSBCOPY(fhp, &tdrpc->fh, sizeof(*fhp));
5829 		tdrpc->off = off;
5830 		tdrpc->len = len;
5831 		tdrpc->nmp = *nmpp;
5832 		tdrpc->cred = cred;
5833 		tdrpc->p = p;
5834 		tdrpc->inprog = 0;
5835 		tdrpc->err = 0;
5836 		ret = EIO;
5837 		if (nfs_pnfsiothreads != 0) {
5838 			ret = nfs_pnfsio(start_deallocatedsdorpc, tdrpc);
5839 			NFSD_DEBUG(4, "nfsrv_deallocatedsrpc: nfs_pnfsio=%d\n",
5840 			    ret);
5841 		}
5842 		if (ret != 0) {
5843 			ret = nfsrv_deallocatedsdorpc(*nmpp, fhp, off, len,
5844 			    NULL, cred, p);
5845 			if (nfsds_failerr(ret) && *failposp == -1)
5846 				*failposp = i;
5847 			else if (error == 0 && ret != 0)
5848 				error = ret;
5849 		}
5850 		nmpp++;
5851 		fhp++;
5852 	}
5853 	ret = nfsrv_deallocatedsdorpc(*nmpp, fhp, off, len, &na, cred, p);
5854 	if (nfsds_failerr(ret) && *failposp == -1 && mirrorcnt > 1)
5855 		*failposp = mirrorcnt - 1;
5856 	else if (error == 0 && ret != 0)
5857 		error = ret;
5858 	if (error == 0)
5859 		error = nfsrv_setextattr(vp, &na, p);
5860 	NFSD_DEBUG(4, "nfsrv_deallocatedsrpc: aft setextat=%d\n", error);
5861 	tdrpc = drpc;
5862 	timo = hz / 50;		/* Wait for 20msec. */
5863 	if (timo < 1)
5864 		timo = 1;
5865 	for (i = 0; i < mirrorcnt - 1; i++, tdrpc++) {
5866 		/* Wait for RPCs on separate threads to complete. */
5867 		while (tdrpc->inprog != 0 && tdrpc->done == 0)
5868 			tsleep(&tdrpc->tsk, PVFS, "srvalds", timo);
5869 		if (nfsds_failerr(tdrpc->err) && *failposp == -1)
5870 			*failposp = i;
5871 		else if (error == 0 && tdrpc->err != 0)
5872 			error = tdrpc->err;
5873 	}
5874 	free(drpc, M_TEMP);
5875 	return (error);
5876 }
5877 
5878 static int
5879 nfsrv_setattrdsdorpc(fhandle_t *fhp, struct ucred *cred, NFSPROC_T *p,
5880     struct vnode *vp, struct nfsmount *nmp, struct nfsvattr *nap,
5881     struct nfsvattr *dsnap)
5882 {
5883 	uint32_t *tl;
5884 	struct nfsrv_descript *nd;
5885 	nfsv4stateid_t st;
5886 	nfsattrbit_t attrbits;
5887 	int error;
5888 
5889 	NFSD_DEBUG(4, "in nfsrv_setattrdsdorpc\n");
5890 	nd = malloc(sizeof(*nd), M_TEMP, M_WAITOK | M_ZERO);
5891 	/*
5892 	 * Use a stateid where other is an alternating 01010 pattern and
5893 	 * seqid is 0xffffffff.  This value is not defined as special by
5894 	 * the RFC and is used by the FreeBSD NFS server to indicate an
5895 	 * MDS->DS proxy operation.
5896 	 */
5897 	st.other[0] = 0x55555555;
5898 	st.other[1] = 0x55555555;
5899 	st.other[2] = 0x55555555;
5900 	st.seqid = 0xffffffff;
5901 	nfscl_reqstart(nd, NFSPROC_SETATTR, nmp, (u_int8_t *)fhp, sizeof(*fhp),
5902 	    NULL, NULL, 0, 0, cred);
5903 	nfsm_stateidtom(nd, &st, NFSSTATEID_PUTSTATEID);
5904 	nfscl_fillsattr(nd, &nap->na_vattr, vp, NFSSATTR_FULL, 0);
5905 
5906 	/* Do a Getattr for the attributes that change due to writing. */
5907 	NFSZERO_ATTRBIT(&attrbits);
5908 	NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_SIZE);
5909 	NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_CHANGE);
5910 	NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_TIMEACCESS);
5911 	NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_TIMEMODIFY);
5912 	NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_SPACEUSED);
5913 	NFSM_BUILD(tl, u_int32_t *, NFSX_UNSIGNED);
5914 	*tl = txdr_unsigned(NFSV4OP_GETATTR);
5915 	(void) nfsrv_putattrbit(nd, &attrbits);
5916 	error = newnfs_request(nd, nmp, NULL, &nmp->nm_sockreq, NULL, p, cred,
5917 	    NFS_PROG, NFS_VER4, NULL, 1, NULL, NULL);
5918 	if (error != 0) {
5919 		free(nd, M_TEMP);
5920 		return (error);
5921 	}
5922 	NFSD_DEBUG(4, "nfsrv_setattrdsdorpc: aft setattrrpc=%d\n",
5923 	    nd->nd_repstat);
5924 	/* Get rid of weak cache consistency data for now. */
5925 	if ((nd->nd_flag & (ND_NOMOREDATA | ND_NFSV4 | ND_V4WCCATTR)) ==
5926 	    (ND_NFSV4 | ND_V4WCCATTR)) {
5927 		error = nfsv4_loadattr(nd, NULL, dsnap, NULL, NULL, 0, NULL,
5928 		    NULL, NULL, NULL, NULL, 0, NULL, NULL, NULL, NULL, NULL);
5929 		NFSD_DEBUG(4, "nfsrv_setattrdsdorpc: wcc attr=%d\n", error);
5930 		if (error != 0)
5931 			goto nfsmout;
5932 		/*
5933 		 * Get rid of Op# and status for next op.
5934 		 */
5935 		NFSM_DISSECT(tl, uint32_t *, 2 * NFSX_UNSIGNED);
5936 		if (*++tl != 0)
5937 			nd->nd_flag |= ND_NOMOREDATA;
5938 	}
5939 	error = nfsrv_getattrbits(nd, &attrbits, NULL, NULL);
5940 	if (error != 0)
5941 		goto nfsmout;
5942 	if (nd->nd_repstat != 0)
5943 		error = nd->nd_repstat;
5944 	/*
5945 	 * Get the Change, Size, Access Time and Modify Time attributes and set
5946 	 * on the Metadata file, so its attributes will be what the file's
5947 	 * would be if it had been written.
5948 	 */
5949 	if (error == 0) {
5950 		NFSM_DISSECT(tl, uint32_t *, 2 * NFSX_UNSIGNED);
5951 		error = nfsv4_loadattr(nd, NULL, dsnap, NULL, NULL, 0, NULL,
5952 		    NULL, NULL, NULL, NULL, 0, NULL, NULL, NULL, NULL, NULL);
5953 	}
5954 	NFSD_DEBUG(4, "nfsrv_setattrdsdorpc: aft setattr loadattr=%d\n", error);
5955 nfsmout:
5956 	m_freem(nd->nd_mrep);
5957 	free(nd, M_TEMP);
5958 	NFSD_DEBUG(4, "nfsrv_setattrdsdorpc error=%d\n", error);
5959 	return (error);
5960 }
5961 
5962 struct nfsrvsetattrdsdorpc {
5963 	int			done;
5964 	int			inprog;
5965 	struct task		tsk;
5966 	fhandle_t		fh;
5967 	struct nfsmount		*nmp;
5968 	struct vnode		*vp;
5969 	struct ucred		*cred;
5970 	NFSPROC_T		*p;
5971 	struct nfsvattr		na;
5972 	struct nfsvattr		dsna;
5973 	int			err;
5974 };
5975 
5976 /*
5977  * Start up the thread that will execute nfsrv_setattrdsdorpc().
5978  */
5979 static void
5980 start_setattrdsdorpc(void *arg, int pending)
5981 {
5982 	struct nfsrvsetattrdsdorpc *drpc;
5983 
5984 	drpc = (struct nfsrvsetattrdsdorpc *)arg;
5985 	drpc->err = nfsrv_setattrdsdorpc(&drpc->fh, drpc->cred, drpc->p,
5986 	    drpc->vp, drpc->nmp, &drpc->na, &drpc->dsna);
5987 	drpc->done = 1;
5988 }
5989 
5990 static int
5991 nfsrv_setattrdsrpc(fhandle_t *fhp, struct ucred *cred, NFSPROC_T *p,
5992     struct vnode *vp, struct nfsmount **nmpp, int mirrorcnt,
5993     struct nfsvattr *nap, int *failposp)
5994 {
5995 	struct nfsrvsetattrdsdorpc *drpc, *tdrpc = NULL;
5996 	struct nfsvattr na;
5997 	int error, i, ret, timo;
5998 
5999 	NFSD_DEBUG(4, "in nfsrv_setattrdsrpc\n");
6000 	drpc = NULL;
6001 	if (mirrorcnt > 1)
6002 		tdrpc = drpc = malloc(sizeof(*drpc) * (mirrorcnt - 1), M_TEMP,
6003 		    M_WAITOK);
6004 
6005 	/*
6006 	 * Do the setattr RPC for every DS, using a separate kernel process
6007 	 * for every DS except the last one.
6008 	 */
6009 	error = 0;
6010 	for (i = 0; i < mirrorcnt - 1; i++, tdrpc++) {
6011 		tdrpc->done = 0;
6012 		tdrpc->inprog = 0;
6013 		NFSBCOPY(fhp, &tdrpc->fh, sizeof(*fhp));
6014 		tdrpc->nmp = *nmpp;
6015 		tdrpc->vp = vp;
6016 		tdrpc->cred = cred;
6017 		tdrpc->p = p;
6018 		tdrpc->na = *nap;
6019 		tdrpc->err = 0;
6020 		ret = EIO;
6021 		if (nfs_pnfsiothreads != 0) {
6022 			ret = nfs_pnfsio(start_setattrdsdorpc, tdrpc);
6023 			NFSD_DEBUG(4, "nfsrv_setattrdsrpc: nfs_pnfsio=%d\n",
6024 			    ret);
6025 		}
6026 		if (ret != 0) {
6027 			ret = nfsrv_setattrdsdorpc(fhp, cred, p, vp, *nmpp, nap,
6028 			    &na);
6029 			if (nfsds_failerr(ret) && *failposp == -1)
6030 				*failposp = i;
6031 			else if (error == 0 && ret != 0)
6032 				error = ret;
6033 		}
6034 		nmpp++;
6035 		fhp++;
6036 	}
6037 	ret = nfsrv_setattrdsdorpc(fhp, cred, p, vp, *nmpp, nap, &na);
6038 	if (nfsds_failerr(ret) && *failposp == -1 && mirrorcnt > 1)
6039 		*failposp = mirrorcnt - 1;
6040 	else if (error == 0 && ret != 0)
6041 		error = ret;
6042 	if (error == 0)
6043 		error = nfsrv_setextattr(vp, &na, p);
6044 	NFSD_DEBUG(4, "nfsrv_setattrdsrpc: aft setextat=%d\n", error);
6045 	tdrpc = drpc;
6046 	timo = hz / 50;		/* Wait for 20msec. */
6047 	if (timo < 1)
6048 		timo = 1;
6049 	for (i = 0; i < mirrorcnt - 1; i++, tdrpc++) {
6050 		/* Wait for RPCs on separate threads to complete. */
6051 		while (tdrpc->inprog != 0 && tdrpc->done == 0)
6052 			tsleep(&tdrpc->tsk, PVFS, "srvsads", timo);
6053 		if (nfsds_failerr(tdrpc->err) && *failposp == -1)
6054 			*failposp = i;
6055 		else if (error == 0 && tdrpc->err != 0)
6056 			error = tdrpc->err;
6057 	}
6058 	free(drpc, M_TEMP);
6059 	return (error);
6060 }
6061 
6062 /*
6063  * Do a Setattr of an NFSv4 ACL on the DS file.
6064  */
6065 static int
6066 nfsrv_setacldsdorpc(fhandle_t *fhp, struct ucred *cred, NFSPROC_T *p,
6067     struct vnode *vp, struct nfsmount *nmp, struct acl *aclp)
6068 {
6069 	struct nfsrv_descript *nd;
6070 	nfsv4stateid_t st;
6071 	nfsattrbit_t attrbits;
6072 	int error;
6073 
6074 	NFSD_DEBUG(4, "in nfsrv_setacldsdorpc\n");
6075 	nd = malloc(sizeof(*nd), M_TEMP, M_WAITOK | M_ZERO);
6076 	/*
6077 	 * Use a stateid where other is an alternating 01010 pattern and
6078 	 * seqid is 0xffffffff.  This value is not defined as special by
6079 	 * the RFC and is used by the FreeBSD NFS server to indicate an
6080 	 * MDS->DS proxy operation.
6081 	 */
6082 	st.other[0] = 0x55555555;
6083 	st.other[1] = 0x55555555;
6084 	st.other[2] = 0x55555555;
6085 	st.seqid = 0xffffffff;
6086 	nfscl_reqstart(nd, NFSPROC_SETACL, nmp, (u_int8_t *)fhp, sizeof(*fhp),
6087 	    NULL, NULL, 0, 0, cred);
6088 	nfsm_stateidtom(nd, &st, NFSSTATEID_PUTSTATEID);
6089 	NFSZERO_ATTRBIT(&attrbits);
6090 	NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_ACL);
6091 	/*
6092 	 * The "vp" argument to nfsv4_fillattr() is only used for vnode_type(),
6093 	 * so passing in the metadata "vp" will be ok, since it is of
6094 	 * the same type (VREG).
6095 	 */
6096 	nfsv4_fillattr(nd, NULL, vp, aclp, NULL, NULL, 0, &attrbits, NULL,
6097 	    NULL, 0, 0, 0, 0, 0, NULL);
6098 	error = newnfs_request(nd, nmp, NULL, &nmp->nm_sockreq, NULL, p, cred,
6099 	    NFS_PROG, NFS_VER4, NULL, 1, NULL, NULL);
6100 	if (error != 0) {
6101 		free(nd, M_TEMP);
6102 		return (error);
6103 	}
6104 	NFSD_DEBUG(4, "nfsrv_setacldsdorpc: aft setaclrpc=%d\n",
6105 	    nd->nd_repstat);
6106 	error = nd->nd_repstat;
6107 	m_freem(nd->nd_mrep);
6108 	free(nd, M_TEMP);
6109 	return (error);
6110 }
6111 
6112 struct nfsrvsetacldsdorpc {
6113 	int			done;
6114 	int			inprog;
6115 	struct task		tsk;
6116 	fhandle_t		fh;
6117 	struct nfsmount		*nmp;
6118 	struct vnode		*vp;
6119 	struct ucred		*cred;
6120 	NFSPROC_T		*p;
6121 	struct acl		*aclp;
6122 	int			err;
6123 };
6124 
6125 /*
6126  * Start up the thread that will execute nfsrv_setacldsdorpc().
6127  */
6128 static void
6129 start_setacldsdorpc(void *arg, int pending)
6130 {
6131 	struct nfsrvsetacldsdorpc *drpc;
6132 
6133 	drpc = (struct nfsrvsetacldsdorpc *)arg;
6134 	drpc->err = nfsrv_setacldsdorpc(&drpc->fh, drpc->cred, drpc->p,
6135 	    drpc->vp, drpc->nmp, drpc->aclp);
6136 	drpc->done = 1;
6137 }
6138 
6139 static int
6140 nfsrv_setacldsrpc(fhandle_t *fhp, struct ucred *cred, NFSPROC_T *p,
6141     struct vnode *vp, struct nfsmount **nmpp, int mirrorcnt, struct acl *aclp,
6142     int *failposp)
6143 {
6144 	struct nfsrvsetacldsdorpc *drpc, *tdrpc = NULL;
6145 	int error, i, ret, timo;
6146 
6147 	NFSD_DEBUG(4, "in nfsrv_setacldsrpc\n");
6148 	drpc = NULL;
6149 	if (mirrorcnt > 1)
6150 		tdrpc = drpc = malloc(sizeof(*drpc) * (mirrorcnt - 1), M_TEMP,
6151 		    M_WAITOK);
6152 
6153 	/*
6154 	 * Do the setattr RPC for every DS, using a separate kernel process
6155 	 * for every DS except the last one.
6156 	 */
6157 	error = 0;
6158 	for (i = 0; i < mirrorcnt - 1; i++, tdrpc++) {
6159 		tdrpc->done = 0;
6160 		tdrpc->inprog = 0;
6161 		NFSBCOPY(fhp, &tdrpc->fh, sizeof(*fhp));
6162 		tdrpc->nmp = *nmpp;
6163 		tdrpc->vp = vp;
6164 		tdrpc->cred = cred;
6165 		tdrpc->p = p;
6166 		tdrpc->aclp = aclp;
6167 		tdrpc->err = 0;
6168 		ret = EIO;
6169 		if (nfs_pnfsiothreads != 0) {
6170 			ret = nfs_pnfsio(start_setacldsdorpc, tdrpc);
6171 			NFSD_DEBUG(4, "nfsrv_setacldsrpc: nfs_pnfsio=%d\n",
6172 			    ret);
6173 		}
6174 		if (ret != 0) {
6175 			ret = nfsrv_setacldsdorpc(fhp, cred, p, vp, *nmpp,
6176 			    aclp);
6177 			if (nfsds_failerr(ret) && *failposp == -1)
6178 				*failposp = i;
6179 			else if (error == 0 && ret != 0)
6180 				error = ret;
6181 		}
6182 		nmpp++;
6183 		fhp++;
6184 	}
6185 	ret = nfsrv_setacldsdorpc(fhp, cred, p, vp, *nmpp, aclp);
6186 	if (nfsds_failerr(ret) && *failposp == -1 && mirrorcnt > 1)
6187 		*failposp = mirrorcnt - 1;
6188 	else if (error == 0 && ret != 0)
6189 		error = ret;
6190 	NFSD_DEBUG(4, "nfsrv_setacldsrpc: aft setextat=%d\n", error);
6191 	tdrpc = drpc;
6192 	timo = hz / 50;		/* Wait for 20msec. */
6193 	if (timo < 1)
6194 		timo = 1;
6195 	for (i = 0; i < mirrorcnt - 1; i++, tdrpc++) {
6196 		/* Wait for RPCs on separate threads to complete. */
6197 		while (tdrpc->inprog != 0 && tdrpc->done == 0)
6198 			tsleep(&tdrpc->tsk, PVFS, "srvacds", timo);
6199 		if (nfsds_failerr(tdrpc->err) && *failposp == -1)
6200 			*failposp = i;
6201 		else if (error == 0 && tdrpc->err != 0)
6202 			error = tdrpc->err;
6203 	}
6204 	free(drpc, M_TEMP);
6205 	return (error);
6206 }
6207 
6208 /*
6209  * Getattr call to the DS for the attributes that change due to writing.
6210  */
6211 static int
6212 nfsrv_getattrdsrpc(fhandle_t *fhp, struct ucred *cred, NFSPROC_T *p,
6213     struct vnode *vp, struct nfsmount *nmp, struct nfsvattr *nap)
6214 {
6215 	struct nfsrv_descript *nd;
6216 	int error;
6217 	nfsattrbit_t attrbits;
6218 
6219 	NFSD_DEBUG(4, "in nfsrv_getattrdsrpc\n");
6220 	nd = malloc(sizeof(*nd), M_TEMP, M_WAITOK | M_ZERO);
6221 	nfscl_reqstart(nd, NFSPROC_GETATTR, nmp, (u_int8_t *)fhp,
6222 	    sizeof(fhandle_t), NULL, NULL, 0, 0, cred);
6223 	NFSZERO_ATTRBIT(&attrbits);
6224 	NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_SIZE);
6225 	NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_CHANGE);
6226 	NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_TIMEACCESS);
6227 	NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_TIMEMODIFY);
6228 	NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_SPACEUSED);
6229 	(void) nfsrv_putattrbit(nd, &attrbits);
6230 	error = newnfs_request(nd, nmp, NULL, &nmp->nm_sockreq, NULL, p, cred,
6231 	    NFS_PROG, NFS_VER4, NULL, 1, NULL, NULL);
6232 	if (error != 0) {
6233 		free(nd, M_TEMP);
6234 		return (error);
6235 	}
6236 	NFSD_DEBUG(4, "nfsrv_getattrdsrpc: aft getattrrpc=%d\n",
6237 	    nd->nd_repstat);
6238 	if (nd->nd_repstat == 0) {
6239 		error = nfsv4_loadattr(nd, NULL, nap, NULL, NULL, 0,
6240 		    NULL, NULL, NULL, NULL, NULL, 0, NULL, NULL, NULL,
6241 		    NULL, NULL);
6242 		/*
6243 		 * We can only save the updated values in the extended
6244 		 * attribute if the vp is exclusively locked.
6245 		 * This should happen when any of the following operations
6246 		 * occur on the vnode:
6247 		 *    Close, Delegreturn, LayoutCommit, LayoutReturn
6248 		 * As such, the updated extended attribute should get saved
6249 		 * before nfsrv_checkdsattr() returns 0 and allows the cached
6250 		 * attributes to be returned without calling this function.
6251 		 */
6252 		if (error == 0 && VOP_ISLOCKED(vp) == LK_EXCLUSIVE) {
6253 			error = nfsrv_setextattr(vp, nap, p);
6254 			NFSD_DEBUG(4, "nfsrv_getattrdsrpc: aft setextat=%d\n",
6255 			    error);
6256 		}
6257 	} else
6258 		error = nd->nd_repstat;
6259 	m_freem(nd->nd_mrep);
6260 	free(nd, M_TEMP);
6261 	NFSD_DEBUG(4, "nfsrv_getattrdsrpc error=%d\n", error);
6262 	return (error);
6263 }
6264 
6265 /*
6266  * Seek call to a DS.
6267  */
6268 static int
6269 nfsrv_seekdsrpc(fhandle_t *fhp, off_t *offp, int content, bool *eofp,
6270     struct ucred *cred, NFSPROC_T *p, struct nfsmount *nmp)
6271 {
6272 	uint32_t *tl;
6273 	struct nfsrv_descript *nd;
6274 	nfsv4stateid_t st;
6275 	int error;
6276 
6277 	NFSD_DEBUG(4, "in nfsrv_seekdsrpc\n");
6278 	/*
6279 	 * Use a stateid where other is an alternating 01010 pattern and
6280 	 * seqid is 0xffffffff.  This value is not defined as special by
6281 	 * the RFC and is used by the FreeBSD NFS server to indicate an
6282 	 * MDS->DS proxy operation.
6283 	 */
6284 	st.other[0] = 0x55555555;
6285 	st.other[1] = 0x55555555;
6286 	st.other[2] = 0x55555555;
6287 	st.seqid = 0xffffffff;
6288 	nd = malloc(sizeof(*nd), M_TEMP, M_WAITOK | M_ZERO);
6289 	nfscl_reqstart(nd, NFSPROC_SEEKDS, nmp, (u_int8_t *)fhp,
6290 	    sizeof(fhandle_t), NULL, NULL, 0, 0, cred);
6291 	nfsm_stateidtom(nd, &st, NFSSTATEID_PUTSTATEID);
6292 	NFSM_BUILD(tl, uint32_t *, NFSX_HYPER + NFSX_UNSIGNED);
6293 	txdr_hyper(*offp, tl); tl += 2;
6294 	*tl = txdr_unsigned(content);
6295 	error = newnfs_request(nd, nmp, NULL, &nmp->nm_sockreq, NULL, p, cred,
6296 	    NFS_PROG, NFS_VER4, NULL, 1, NULL, NULL);
6297 	if (error != 0) {
6298 		free(nd, M_TEMP);
6299 		return (error);
6300 	}
6301 	NFSD_DEBUG(4, "nfsrv_seekdsrpc: aft seekrpc=%d\n", nd->nd_repstat);
6302 	if (nd->nd_repstat == 0) {
6303 		NFSM_DISSECT(tl, uint32_t *, NFSX_UNSIGNED + NFSX_HYPER);
6304 		if (*tl++ == newnfs_true)
6305 			*eofp = true;
6306 		else
6307 			*eofp = false;
6308 		*offp = fxdr_hyper(tl);
6309 	} else
6310 		error = nd->nd_repstat;
6311 nfsmout:
6312 	m_freem(nd->nd_mrep);
6313 	free(nd, M_TEMP);
6314 	NFSD_DEBUG(4, "nfsrv_seekdsrpc error=%d\n", error);
6315 	return (error);
6316 }
6317 
6318 /*
6319  * Get the device id and file handle for a DS file.
6320  */
6321 int
6322 nfsrv_dsgetdevandfh(struct vnode *vp, NFSPROC_T *p, int *mirrorcntp,
6323     fhandle_t *fhp, char *devid)
6324 {
6325 	int buflen, error;
6326 	char *buf;
6327 
6328 	buflen = 1024;
6329 	buf = malloc(buflen, M_TEMP, M_WAITOK);
6330 	error = nfsrv_dsgetsockmnt(vp, 0, buf, &buflen, mirrorcntp, p, NULL,
6331 	    fhp, devid, NULL, NULL, NULL, NULL, NULL, NULL);
6332 	free(buf, M_TEMP);
6333 	return (error);
6334 }
6335 
6336 /*
6337  * Do a Lookup against the DS for the filename.
6338  */
6339 static int
6340 nfsrv_pnfslookupds(struct vnode *vp, struct vnode *dvp, struct pnfsdsfile *pf,
6341     struct vnode **nvpp, NFSPROC_T *p)
6342 {
6343 	struct nameidata named;
6344 	struct ucred *tcred;
6345 	char *bufp;
6346 	u_long *hashp;
6347 	struct vnode *nvp;
6348 	int error;
6349 
6350 	tcred = newnfs_getcred();
6351 	named.ni_cnd.cn_nameiop = LOOKUP;
6352 	named.ni_cnd.cn_lkflags = LK_SHARED | LK_RETRY;
6353 	named.ni_cnd.cn_cred = tcred;
6354 	named.ni_cnd.cn_flags = ISLASTCN | LOCKPARENT | LOCKLEAF | SAVENAME;
6355 	nfsvno_setpathbuf(&named, &bufp, &hashp);
6356 	named.ni_cnd.cn_nameptr = bufp;
6357 	named.ni_cnd.cn_namelen = strlen(pf->dsf_filename);
6358 	strlcpy(bufp, pf->dsf_filename, NAME_MAX);
6359 	NFSD_DEBUG(4, "nfsrv_pnfslookupds: filename=%s\n", bufp);
6360 	error = VOP_LOOKUP(dvp, &nvp, &named.ni_cnd);
6361 	NFSD_DEBUG(4, "nfsrv_pnfslookupds: aft LOOKUP=%d\n", error);
6362 	NFSFREECRED(tcred);
6363 	nfsvno_relpathbuf(&named);
6364 	if (error == 0)
6365 		*nvpp = nvp;
6366 	NFSD_DEBUG(4, "eo nfsrv_pnfslookupds=%d\n", error);
6367 	return (error);
6368 }
6369 
6370 /*
6371  * Set the file handle to the correct one.
6372  */
6373 static void
6374 nfsrv_pnfssetfh(struct vnode *vp, struct pnfsdsfile *pf, char *devid,
6375     char *fnamep, struct vnode *nvp, NFSPROC_T *p)
6376 {
6377 	struct nfsnode *np;
6378 	int ret = 0;
6379 
6380 	np = VTONFS(nvp);
6381 	NFSBCOPY(np->n_fhp->nfh_fh, &pf->dsf_fh, NFSX_MYFH);
6382 	/*
6383 	 * We can only do a vn_set_extattr() if the vnode is exclusively
6384 	 * locked and vn_start_write() has been done.  If devid != NULL or
6385 	 * fnamep != NULL or the vnode is shared locked, vn_start_write()
6386 	 * may not have been done.
6387 	 * If not done now, it will be done on a future call.
6388 	 */
6389 	if (devid == NULL && fnamep == NULL && NFSVOPISLOCKED(vp) ==
6390 	    LK_EXCLUSIVE)
6391 		ret = vn_extattr_set(vp, IO_NODELOCKED,
6392 		    EXTATTR_NAMESPACE_SYSTEM, "pnfsd.dsfile", sizeof(*pf),
6393 		    (char *)pf, p);
6394 	NFSD_DEBUG(4, "eo nfsrv_pnfssetfh=%d\n", ret);
6395 }
6396 
6397 /*
6398  * Cause RPCs waiting on "nmp" to fail.  This is called for a DS mount point
6399  * when the DS has failed.
6400  */
6401 void
6402 nfsrv_killrpcs(struct nfsmount *nmp)
6403 {
6404 
6405 	/*
6406 	 * Call newnfs_nmcancelreqs() to cause
6407 	 * any RPCs in progress on the mount point to
6408 	 * fail.
6409 	 * This will cause any process waiting for an
6410 	 * RPC to complete while holding a vnode lock
6411 	 * on the mounted-on vnode (such as "df" or
6412 	 * a non-forced "umount") to fail.
6413 	 * This will unlock the mounted-on vnode so
6414 	 * a forced dismount can succeed.
6415 	 * The NFSMNTP_CANCELRPCS flag should be set when this function is
6416 	 * called.
6417 	 */
6418 	newnfs_nmcancelreqs(nmp);
6419 }
6420 
6421 /*
6422  * Sum up the statfs info for each of the DSs, so that the client will
6423  * receive the total for all DSs.
6424  */
6425 static int
6426 nfsrv_pnfsstatfs(struct statfs *sf, struct mount *mp)
6427 {
6428 	struct statfs *tsf;
6429 	struct nfsdevice *ds;
6430 	struct vnode **dvpp, **tdvpp, *dvp;
6431 	uint64_t tot;
6432 	int cnt, error = 0, i;
6433 
6434 	if (nfsrv_devidcnt <= 0)
6435 		return (ENXIO);
6436 	dvpp = mallocarray(nfsrv_devidcnt, sizeof(*dvpp), M_TEMP, M_WAITOK);
6437 	tsf = malloc(sizeof(*tsf), M_TEMP, M_WAITOK);
6438 
6439 	/* Get an array of the dvps for the DSs. */
6440 	tdvpp = dvpp;
6441 	i = 0;
6442 	NFSDDSLOCK();
6443 	/* First, search for matches for same file system. */
6444 	TAILQ_FOREACH(ds, &nfsrv_devidhead, nfsdev_list) {
6445 		if (ds->nfsdev_nmp != NULL && ds->nfsdev_mdsisset != 0 &&
6446 		    fsidcmp(&ds->nfsdev_mdsfsid, &mp->mnt_stat.f_fsid) == 0) {
6447 			if (++i > nfsrv_devidcnt)
6448 				break;
6449 			*tdvpp++ = ds->nfsdev_dvp;
6450 		}
6451 	}
6452 	/*
6453 	 * If no matches for same file system, total all servers not assigned
6454 	 * to a file system.
6455 	 */
6456 	if (i == 0) {
6457 		TAILQ_FOREACH(ds, &nfsrv_devidhead, nfsdev_list) {
6458 			if (ds->nfsdev_nmp != NULL &&
6459 			    ds->nfsdev_mdsisset == 0) {
6460 				if (++i > nfsrv_devidcnt)
6461 					break;
6462 				*tdvpp++ = ds->nfsdev_dvp;
6463 			}
6464 		}
6465 	}
6466 	NFSDDSUNLOCK();
6467 	cnt = i;
6468 
6469 	/* Do a VFS_STATFS() for each of the DSs and sum them up. */
6470 	tdvpp = dvpp;
6471 	for (i = 0; i < cnt && error == 0; i++) {
6472 		dvp = *tdvpp++;
6473 		error = VFS_STATFS(dvp->v_mount, tsf);
6474 		if (error == 0) {
6475 			if (sf->f_bsize == 0) {
6476 				if (tsf->f_bsize > 0)
6477 					sf->f_bsize = tsf->f_bsize;
6478 				else
6479 					sf->f_bsize = 8192;
6480 			}
6481 			if (tsf->f_blocks > 0) {
6482 				if (sf->f_bsize != tsf->f_bsize) {
6483 					tot = tsf->f_blocks * tsf->f_bsize;
6484 					sf->f_blocks += (tot / sf->f_bsize);
6485 				} else
6486 					sf->f_blocks += tsf->f_blocks;
6487 			}
6488 			if (tsf->f_bfree > 0) {
6489 				if (sf->f_bsize != tsf->f_bsize) {
6490 					tot = tsf->f_bfree * tsf->f_bsize;
6491 					sf->f_bfree += (tot / sf->f_bsize);
6492 				} else
6493 					sf->f_bfree += tsf->f_bfree;
6494 			}
6495 			if (tsf->f_bavail > 0) {
6496 				if (sf->f_bsize != tsf->f_bsize) {
6497 					tot = tsf->f_bavail * tsf->f_bsize;
6498 					sf->f_bavail += (tot / sf->f_bsize);
6499 				} else
6500 					sf->f_bavail += tsf->f_bavail;
6501 			}
6502 		}
6503 	}
6504 	free(tsf, M_TEMP);
6505 	free(dvpp, M_TEMP);
6506 	return (error);
6507 }
6508 
6509 /*
6510  * Set an NFSv4 acl.
6511  */
6512 int
6513 nfsrv_setacl(struct vnode *vp, NFSACL_T *aclp, struct ucred *cred, NFSPROC_T *p)
6514 {
6515 	int error;
6516 
6517 	if (nfsrv_useacl == 0 || nfs_supportsnfsv4acls(vp) == 0) {
6518 		error = NFSERR_ATTRNOTSUPP;
6519 		goto out;
6520 	}
6521 	/*
6522 	 * With NFSv4 ACLs, chmod(2) may need to add additional entries.
6523 	 * Make sure it has enough room for that - splitting every entry
6524 	 * into two and appending "canonical six" entries at the end.
6525 	 * Cribbed out of kern/vfs_acl.c - Rick M.
6526 	 */
6527 	if (aclp->acl_cnt > (ACL_MAX_ENTRIES - 6) / 2) {
6528 		error = NFSERR_ATTRNOTSUPP;
6529 		goto out;
6530 	}
6531 	error = VOP_SETACL(vp, ACL_TYPE_NFS4, aclp, cred, p);
6532 	if (error == 0) {
6533 		error = nfsrv_dssetacl(vp, aclp, cred, p);
6534 		if (error == ENOENT)
6535 			error = 0;
6536 	}
6537 
6538 out:
6539 	NFSEXITCODE(error);
6540 	return (error);
6541 }
6542 
6543 /*
6544  * Seek vnode op call (actually it is a VOP_IOCTL()).
6545  * This function is called with the vnode locked, but unlocks and vrele()s
6546  * the vp before returning.
6547  */
6548 int
6549 nfsvno_seek(struct nfsrv_descript *nd, struct vnode *vp, u_long cmd,
6550     off_t *offp, int content, bool *eofp, struct ucred *cred, NFSPROC_T *p)
6551 {
6552 	struct nfsvattr at;
6553 	int error, ret;
6554 
6555 	ASSERT_VOP_LOCKED(vp, "nfsvno_seek vp");
6556 	/*
6557 	 * Attempt to seek on a DS file. A return of ENOENT implies
6558 	 * there is no DS file to seek on.
6559 	 */
6560 	error = nfsrv_proxyds(vp, 0, 0, cred, p, NFSPROC_SEEKDS, NULL,
6561 	    NULL, NULL, NULL, NULL, offp, content, eofp);
6562 	if (error != ENOENT) {
6563 		vput(vp);
6564 		return (error);
6565 	}
6566 
6567 	/*
6568 	 * Do the VOP_IOCTL() call.  For the case where *offp == file_size,
6569 	 * VOP_IOCTL() will return ENXIO.  However, the correct reply for
6570 	 * NFSv4.2 is *eofp == true and error == 0 for this case.
6571 	 */
6572 	NFSVOPUNLOCK(vp);
6573 	error = VOP_IOCTL(vp, cmd, offp, 0, cred, p);
6574 	*eofp = false;
6575 	if (error == ENXIO || (error == 0 && cmd == FIOSEEKHOLE)) {
6576 		/* Handle the cases where we might be at EOF. */
6577 		ret = nfsvno_getattr(vp, &at, nd, p, 0, NULL);
6578 		if (ret == 0 && *offp == at.na_size) {
6579 			*eofp = true;
6580 			error = 0;
6581 		}
6582 		if (ret != 0 && error == 0)
6583 			error = ret;
6584 	}
6585 	vrele(vp);
6586 	NFSEXITCODE(error);
6587 	return (error);
6588 }
6589 
6590 /*
6591  * Allocate vnode op call.
6592  */
6593 int
6594 nfsvno_allocate(struct vnode *vp, off_t off, off_t len, struct ucred *cred,
6595     NFSPROC_T *p)
6596 {
6597 	int error;
6598 	off_t olen;
6599 
6600 	ASSERT_VOP_ELOCKED(vp, "nfsvno_allocate vp");
6601 	/*
6602 	 * Attempt to allocate on a DS file. A return of ENOENT implies
6603 	 * there is no DS file to allocate on.
6604 	 */
6605 	error = nfsrv_proxyds(vp, off, 0, cred, p, NFSPROC_ALLOCATE, NULL,
6606 	    NULL, NULL, NULL, NULL, &len, 0, NULL);
6607 	if (error != ENOENT)
6608 		return (error);
6609 
6610 	/*
6611 	 * Do the actual VOP_ALLOCATE(), looping so long as
6612 	 * progress is being made, to achieve completion.
6613 	 */
6614 	do {
6615 		olen = len;
6616 		error = VOP_ALLOCATE(vp, &off, &len, IO_SYNC, cred);
6617 		if (error == 0 && len > 0 && olen > len)
6618 			maybe_yield();
6619 	} while (error == 0 && len > 0 && olen > len);
6620 	if (error == 0 && len > 0)
6621 		error = NFSERR_IO;
6622 	NFSEXITCODE(error);
6623 	return (error);
6624 }
6625 
6626 /*
6627  * Deallocate vnode op call.
6628  */
6629 int
6630 nfsvno_deallocate(struct vnode *vp, off_t off, off_t len, struct ucred *cred,
6631     NFSPROC_T *p)
6632 {
6633 	int error;
6634 	off_t olen;
6635 
6636 	ASSERT_VOP_ELOCKED(vp, "nfsvno_deallocate vp");
6637 	/*
6638 	 * Attempt to deallocate on a DS file. A return of ENOENT implies
6639 	 * there is no DS file to deallocate on.
6640 	 */
6641 	error = nfsrv_proxyds(vp, off, 0, cred, p, NFSPROC_DEALLOCATE, NULL,
6642 	    NULL, NULL, NULL, NULL, &len, 0, NULL);
6643 	if (error != ENOENT)
6644 		return (error);
6645 
6646 	/*
6647 	 * Do the actual VOP_DEALLOCATE(), looping so long as
6648 	 * progress is being made, to achieve completion.
6649 	 */
6650 	do {
6651 		olen = len;
6652 		error = VOP_DEALLOCATE(vp, &off, &len, 0, IO_SYNC, cred);
6653 		if (error == 0 && len > 0 && olen > len)
6654 			maybe_yield();
6655 	} while (error == 0 && len > 0 && olen > len);
6656 	if (error == 0 && len > 0)
6657 		error = NFSERR_IO;
6658 	NFSEXITCODE(error);
6659 	return (error);
6660 }
6661 
6662 /*
6663  * Get Extended Atribute vnode op into an mbuf list.
6664  */
6665 int
6666 nfsvno_getxattr(struct vnode *vp, char *name, uint32_t maxresp,
6667     struct ucred *cred, uint64_t flag, int maxextsiz, struct thread *p,
6668     struct mbuf **mpp, struct mbuf **mpendp, int *lenp)
6669 {
6670 	struct iovec *iv;
6671 	struct uio io, *uiop = &io;
6672 	struct mbuf *m, *m2;
6673 	int alen, error, len, tlen;
6674 	size_t siz;
6675 
6676 	/* First, find out the size of the extended attribute. */
6677 	error = VOP_GETEXTATTR(vp, EXTATTR_NAMESPACE_USER, name, NULL,
6678 	    &siz, cred, p);
6679 	if (error != 0)
6680 		return (NFSERR_NOXATTR);
6681 	if (siz > maxresp - NFS_MAXXDR)
6682 		return (NFSERR_XATTR2BIG);
6683 	len = siz;
6684 	tlen = NFSM_RNDUP(len);
6685 	if (tlen > 0) {
6686 		/*
6687 		 * If cnt > MCLBYTES and the reply will not be saved, use
6688 		 * ext_pgs mbufs for TLS.
6689 		 * For NFSv4.0, we do not know for sure if the reply will
6690 		 * be saved, so do not use ext_pgs mbufs for NFSv4.0.
6691 		 * Always use ext_pgs mbufs if ND_EXTPG is set.
6692 		 */
6693 		if ((flag & ND_EXTPG) != 0 || (tlen > MCLBYTES &&
6694 		    (flag & (ND_TLS | ND_SAVEREPLY)) == ND_TLS &&
6695 		    (flag & (ND_NFSV4 | ND_NFSV41)) != ND_NFSV4))
6696 			uiop->uio_iovcnt = nfsrv_createiovec_extpgs(tlen,
6697 			    maxextsiz, &m, &m2, &iv);
6698 		else
6699 			uiop->uio_iovcnt = nfsrv_createiovec(tlen, &m, &m2,
6700 			    &iv);
6701 		uiop->uio_iov = iv;
6702 	} else {
6703 		uiop->uio_iovcnt = 0;
6704 		uiop->uio_iov = iv = NULL;
6705 		m = m2 = NULL;
6706 	}
6707 	uiop->uio_offset = 0;
6708 	uiop->uio_resid = tlen;
6709 	uiop->uio_rw = UIO_READ;
6710 	uiop->uio_segflg = UIO_SYSSPACE;
6711 	uiop->uio_td = p;
6712 #ifdef MAC
6713 	error = mac_vnode_check_getextattr(cred, vp, EXTATTR_NAMESPACE_USER,
6714 	    name);
6715 	if (error != 0)
6716 		goto out;
6717 #endif
6718 
6719 	if (tlen > 0)
6720 		error = VOP_GETEXTATTR(vp, EXTATTR_NAMESPACE_USER, name, uiop,
6721 		    NULL, cred, p);
6722 	if (error != 0)
6723 		goto out;
6724 	if (uiop->uio_resid > 0) {
6725 		alen = tlen;
6726 		len = tlen - uiop->uio_resid;
6727 		tlen = NFSM_RNDUP(len);
6728 		if (alen != tlen)
6729 			printf("nfsvno_getxattr: weird size read\n");
6730 		if (tlen == 0) {
6731 			m_freem(m);
6732 			m = m2 = NULL;
6733 		} else if (alen != tlen || tlen != len)
6734 			m2 = nfsrv_adj(m, alen - tlen, tlen - len);
6735 	}
6736 	*lenp = len;
6737 	*mpp = m;
6738 	*mpendp = m2;
6739 
6740 out:
6741 	if (error != 0) {
6742 		if (m != NULL)
6743 			m_freem(m);
6744 		*lenp = 0;
6745 	}
6746 	free(iv, M_TEMP);
6747 	NFSEXITCODE(error);
6748 	return (error);
6749 }
6750 
6751 /*
6752  * Set Extended attribute vnode op from an mbuf list.
6753  */
6754 int
6755 nfsvno_setxattr(struct vnode *vp, char *name, int len, struct mbuf *m,
6756     char *cp, struct ucred *cred, struct thread *p)
6757 {
6758 	struct iovec *iv;
6759 	struct uio uio, *uiop = &uio;
6760 	int cnt, error;
6761 
6762 	error = 0;
6763 #ifdef MAC
6764 	error = mac_vnode_check_setextattr(cred, vp, EXTATTR_NAMESPACE_USER,
6765 	    name);
6766 #endif
6767 	if (error != 0)
6768 		goto out;
6769 
6770 	uiop->uio_rw = UIO_WRITE;
6771 	uiop->uio_segflg = UIO_SYSSPACE;
6772 	uiop->uio_td = p;
6773 	uiop->uio_offset = 0;
6774 	uiop->uio_resid = len;
6775 	if (len > 0) {
6776 		error = nfsrv_createiovecw(len, m, cp, &iv, &cnt);
6777 		uiop->uio_iov = iv;
6778 		uiop->uio_iovcnt = cnt;
6779 	} else {
6780 		uiop->uio_iov = iv = NULL;
6781 		uiop->uio_iovcnt = 0;
6782 	}
6783 	if (error == 0) {
6784 		error = VOP_SETEXTATTR(vp, EXTATTR_NAMESPACE_USER, name, uiop,
6785 		    cred, p);
6786 		free(iv, M_TEMP);
6787 	}
6788 
6789 out:
6790 	NFSEXITCODE(error);
6791 	return (error);
6792 }
6793 
6794 /*
6795  * Remove Extended attribute vnode op.
6796  */
6797 int
6798 nfsvno_rmxattr(struct nfsrv_descript *nd, struct vnode *vp, char *name,
6799     struct ucred *cred, struct thread *p)
6800 {
6801 	int error;
6802 
6803 	/*
6804 	 * Get rid of any delegations.  I am not sure why this is required,
6805 	 * but RFC-8276 says so.
6806 	 */
6807 	error = nfsrv_checkremove(vp, 0, nd, nd->nd_clientid, p);
6808 	if (error != 0)
6809 		goto out;
6810 #ifdef MAC
6811 	error = mac_vnode_check_deleteextattr(cred, vp, EXTATTR_NAMESPACE_USER,
6812 	    name);
6813 	if (error != 0)
6814 		goto out;
6815 #endif
6816 
6817 	error = VOP_DELETEEXTATTR(vp, EXTATTR_NAMESPACE_USER, name, cred, p);
6818 	if (error == EOPNOTSUPP)
6819 		error = VOP_SETEXTATTR(vp, EXTATTR_NAMESPACE_USER, name, NULL,
6820 		    cred, p);
6821 out:
6822 	NFSEXITCODE(error);
6823 	return (error);
6824 }
6825 
6826 /*
6827  * List Extended Atribute vnode op into an mbuf list.
6828  */
6829 int
6830 nfsvno_listxattr(struct vnode *vp, uint64_t cookie, struct ucred *cred,
6831     struct thread *p, u_char **bufp, uint32_t *lenp, bool *eofp)
6832 {
6833 	struct iovec iv;
6834 	struct uio io;
6835 	int error;
6836 	size_t siz;
6837 
6838 	*bufp = NULL;
6839 	/* First, find out the size of the extended attribute. */
6840 	error = VOP_LISTEXTATTR(vp, EXTATTR_NAMESPACE_USER, NULL, &siz, cred,
6841 	    p);
6842 	if (error != 0)
6843 		return (NFSERR_NOXATTR);
6844 	if (siz <= cookie) {
6845 		*lenp = 0;
6846 		*eofp = true;
6847 		goto out;
6848 	}
6849 	if (siz > cookie + *lenp) {
6850 		siz = cookie + *lenp;
6851 		*eofp = false;
6852 	} else
6853 		*eofp = true;
6854 	/* Just choose a sanity limit of 10Mbytes for malloc(M_TEMP). */
6855 	if (siz > 10 * 1024 * 1024) {
6856 		error = NFSERR_XATTR2BIG;
6857 		goto out;
6858 	}
6859 	*bufp = malloc(siz, M_TEMP, M_WAITOK);
6860 	iv.iov_base = *bufp;
6861 	iv.iov_len = siz;
6862 	io.uio_iovcnt = 1;
6863 	io.uio_iov = &iv;
6864 	io.uio_offset = 0;
6865 	io.uio_resid = siz;
6866 	io.uio_rw = UIO_READ;
6867 	io.uio_segflg = UIO_SYSSPACE;
6868 	io.uio_td = p;
6869 #ifdef MAC
6870 	error = mac_vnode_check_listextattr(cred, vp, EXTATTR_NAMESPACE_USER);
6871 	if (error != 0)
6872 		goto out;
6873 #endif
6874 
6875 	error = VOP_LISTEXTATTR(vp, EXTATTR_NAMESPACE_USER, &io, NULL, cred,
6876 	    p);
6877 	if (error != 0)
6878 		goto out;
6879 	if (io.uio_resid > 0)
6880 		siz -= io.uio_resid;
6881 	*lenp = siz;
6882 
6883 out:
6884 	if (error != 0) {
6885 		free(*bufp, M_TEMP);
6886 		*bufp = NULL;
6887 	}
6888 	NFSEXITCODE(error);
6889 	return (error);
6890 }
6891 
6892 /*
6893  * Trim trailing data off the mbuf list being built.
6894  */
6895 void
6896 nfsm_trimtrailing(struct nfsrv_descript *nd, struct mbuf *mb, char *bpos,
6897     int bextpg, int bextpgsiz)
6898 {
6899 	vm_page_t pg;
6900 	int fullpgsiz, i;
6901 
6902 	if (mb->m_next != NULL) {
6903 		m_freem(mb->m_next);
6904 		mb->m_next = NULL;
6905 	}
6906 	if ((mb->m_flags & M_EXTPG) != 0) {
6907 		KASSERT(bextpg >= 0 && bextpg < mb->m_epg_npgs,
6908 		    ("nfsm_trimtrailing: bextpg out of range"));
6909 		KASSERT(bpos == (char *)(void *)
6910 		    PHYS_TO_DMAP(mb->m_epg_pa[bextpg]) + PAGE_SIZE - bextpgsiz,
6911 		    ("nfsm_trimtrailing: bextpgsiz bad!"));
6912 
6913 		/* First, get rid of any pages after this position. */
6914 		for (i = mb->m_epg_npgs - 1; i > bextpg; i--) {
6915 			pg = PHYS_TO_VM_PAGE(mb->m_epg_pa[i]);
6916 			vm_page_unwire_noq(pg);
6917 			vm_page_free(pg);
6918 		}
6919 		mb->m_epg_npgs = bextpg + 1;
6920 		if (bextpg == 0)
6921 			fullpgsiz = PAGE_SIZE - mb->m_epg_1st_off;
6922 		else
6923 			fullpgsiz = PAGE_SIZE;
6924 		mb->m_epg_last_len = fullpgsiz - bextpgsiz;
6925 		mb->m_len = m_epg_pagelen(mb, 0, mb->m_epg_1st_off);
6926 		for (i = 1; i < mb->m_epg_npgs; i++)
6927 			mb->m_len += m_epg_pagelen(mb, i, 0);
6928 		nd->nd_bextpgsiz = bextpgsiz;
6929 		nd->nd_bextpg = bextpg;
6930 	} else
6931 		mb->m_len = bpos - mtod(mb, char *);
6932 	nd->nd_mb = mb;
6933 	nd->nd_bpos = bpos;
6934 }
6935 
6936 
6937 /*
6938  * Check to see if a put file handle operation should test for
6939  * NFSERR_WRONGSEC, although NFSv3 actually returns NFSERR_AUTHERR.
6940  * When Open is the next operation, NFSERR_WRONGSEC cannot be
6941  * replied for the Open cases that use a component.  This can
6942  * be identified by the fact that the file handle's type is VDIR.
6943  */
6944 bool
6945 nfsrv_checkwrongsec(struct nfsrv_descript *nd, int nextop, enum vtype vtyp)
6946 {
6947 
6948 	if ((nd->nd_flag & ND_NFSV4) == 0)
6949 		return (true);
6950 
6951 	if ((nd->nd_flag & ND_LASTOP) != 0)
6952 		return (false);
6953 
6954 	if (nextop == NFSV4OP_PUTROOTFH || nextop == NFSV4OP_PUTFH ||
6955 	    nextop == NFSV4OP_PUTPUBFH || nextop == NFSV4OP_RESTOREFH ||
6956 	    nextop == NFSV4OP_LOOKUP || nextop == NFSV4OP_LOOKUPP ||
6957 	    nextop == NFSV4OP_SECINFO || nextop == NFSV4OP_SECINFONONAME)
6958 		return (false);
6959 	if (nextop == NFSV4OP_OPEN && vtyp == VDIR)
6960 		return (false);
6961 	return (true);
6962 }
6963 
6964 /*
6965  * Check DSs marked no space.
6966  */
6967 void
6968 nfsrv_checknospc(void)
6969 {
6970 	struct statfs *tsf;
6971 	struct nfsdevice *ds;
6972 	struct vnode **dvpp, **tdvpp, *dvp;
6973 	char *devid, *tdevid;
6974 	int cnt, error = 0, i;
6975 
6976 	if (nfsrv_devidcnt <= 0)
6977 		return;
6978 	dvpp = mallocarray(nfsrv_devidcnt, sizeof(*dvpp), M_TEMP, M_WAITOK);
6979 	devid = malloc(nfsrv_devidcnt * NFSX_V4DEVICEID, M_TEMP, M_WAITOK);
6980 	tsf = malloc(sizeof(*tsf), M_TEMP, M_WAITOK);
6981 
6982 	/* Get an array of the dvps for the DSs. */
6983 	tdvpp = dvpp;
6984 	tdevid = devid;
6985 	i = 0;
6986 	NFSDDSLOCK();
6987 	/* First, search for matches for same file system. */
6988 	TAILQ_FOREACH(ds, &nfsrv_devidhead, nfsdev_list) {
6989 		if (ds->nfsdev_nmp != NULL && ds->nfsdev_nospc) {
6990 			if (++i > nfsrv_devidcnt)
6991 				break;
6992 			*tdvpp++ = ds->nfsdev_dvp;
6993 			NFSBCOPY(ds->nfsdev_deviceid, tdevid, NFSX_V4DEVICEID);
6994 			tdevid += NFSX_V4DEVICEID;
6995 		}
6996 	}
6997 	NFSDDSUNLOCK();
6998 
6999 	/* Do a VFS_STATFS() for each of the DSs and clear no space. */
7000 	cnt = i;
7001 	tdvpp = dvpp;
7002 	tdevid = devid;
7003 	for (i = 0; i < cnt && error == 0; i++) {
7004 		dvp = *tdvpp++;
7005 		error = VFS_STATFS(dvp->v_mount, tsf);
7006 		if (error == 0 && tsf->f_bavail > 0) {
7007 			NFSD_DEBUG(1, "nfsrv_checknospc: reset nospc\n");
7008 			nfsrv_marknospc(tdevid, false);
7009 		}
7010 		tdevid += NFSX_V4DEVICEID;
7011 	}
7012 	free(tsf, M_TEMP);
7013 	free(dvpp, M_TEMP);
7014 	free(devid, M_TEMP);
7015 }
7016 
7017 extern int (*nfsd_call_nfsd)(struct thread *, struct nfssvc_args *);
7018 
7019 /*
7020  * Called once to initialize data structures...
7021  */
7022 static int
7023 nfsd_modevent(module_t mod, int type, void *data)
7024 {
7025 	int error = 0, i;
7026 	static int loaded = 0;
7027 
7028 	switch (type) {
7029 	case MOD_LOAD:
7030 		if (loaded)
7031 			goto out;
7032 		newnfs_portinit();
7033 		for (i = 0; i < NFSRVCACHE_HASHSIZE; i++) {
7034 			mtx_init(&nfsrchash_table[i].mtx, "nfsrtc", NULL,
7035 			    MTX_DEF);
7036 			mtx_init(&nfsrcahash_table[i].mtx, "nfsrtca", NULL,
7037 			    MTX_DEF);
7038 		}
7039 		mtx_init(&nfsrc_udpmtx, "nfsuc", NULL, MTX_DEF);
7040 		mtx_init(&nfs_v4root_mutex, "nfs4rt", NULL, MTX_DEF);
7041 		mtx_init(&nfsv4root_mnt.mnt_mtx, "nfs4mnt", NULL, MTX_DEF);
7042 		mtx_init(&nfsrv_dontlistlock_mtx, "nfs4dnl", NULL, MTX_DEF);
7043 		mtx_init(&nfsrv_recalllock_mtx, "nfs4rec", NULL, MTX_DEF);
7044 		lockinit(&nfsv4root_mnt.mnt_explock, PVFS, "explock", 0, 0);
7045 		callout_init(&nfsd_callout, 1);
7046 		nfsrvd_initcache();
7047 		nfsd_init();
7048 		NFSD_LOCK();
7049 		nfsrvd_init(0);
7050 		NFSD_UNLOCK();
7051 		nfsd_mntinit();
7052 #ifdef VV_DISABLEDELEG
7053 		vn_deleg_ops.vndeleg_recall = nfsd_recalldelegation;
7054 		vn_deleg_ops.vndeleg_disable = nfsd_disabledelegation;
7055 #endif
7056 		nfsd_call_nfsd = nfssvc_nfsd;
7057 		loaded = 1;
7058 		break;
7059 
7060 	case MOD_UNLOAD:
7061 		if (newnfs_numnfsd != 0) {
7062 			error = EBUSY;
7063 			break;
7064 		}
7065 
7066 #ifdef VV_DISABLEDELEG
7067 		vn_deleg_ops.vndeleg_recall = NULL;
7068 		vn_deleg_ops.vndeleg_disable = NULL;
7069 #endif
7070 		nfsd_call_nfsd = NULL;
7071 		callout_drain(&nfsd_callout);
7072 
7073 		/* Clean out all NFSv4 state. */
7074 		nfsrv_throwawayallstate(curthread);
7075 
7076 		/* Clean the NFS server reply cache */
7077 		nfsrvd_cleancache();
7078 
7079 		/* Free up the krpc server pool. */
7080 		if (nfsrvd_pool != NULL)
7081 			svcpool_destroy(nfsrvd_pool);
7082 
7083 		/* and get rid of the locks */
7084 		for (i = 0; i < NFSRVCACHE_HASHSIZE; i++) {
7085 			mtx_destroy(&nfsrchash_table[i].mtx);
7086 			mtx_destroy(&nfsrcahash_table[i].mtx);
7087 		}
7088 		mtx_destroy(&nfsrc_udpmtx);
7089 		mtx_destroy(&nfs_v4root_mutex);
7090 		mtx_destroy(&nfsv4root_mnt.mnt_mtx);
7091 		mtx_destroy(&nfsrv_dontlistlock_mtx);
7092 		mtx_destroy(&nfsrv_recalllock_mtx);
7093 		for (i = 0; i < nfsrv_sessionhashsize; i++)
7094 			mtx_destroy(&nfssessionhash[i].mtx);
7095 		if (nfslayouthash != NULL) {
7096 			for (i = 0; i < nfsrv_layouthashsize; i++)
7097 				mtx_destroy(&nfslayouthash[i].mtx);
7098 			free(nfslayouthash, M_NFSDSESSION);
7099 		}
7100 		lockdestroy(&nfsv4root_mnt.mnt_explock);
7101 		free(nfsclienthash, M_NFSDCLIENT);
7102 		free(nfslockhash, M_NFSDLOCKFILE);
7103 		free(nfssessionhash, M_NFSDSESSION);
7104 		loaded = 0;
7105 		break;
7106 	default:
7107 		error = EOPNOTSUPP;
7108 		break;
7109 	}
7110 
7111 out:
7112 	NFSEXITCODE(error);
7113 	return (error);
7114 }
7115 static moduledata_t nfsd_mod = {
7116 	"nfsd",
7117 	nfsd_modevent,
7118 	NULL,
7119 };
7120 DECLARE_MODULE(nfsd, nfsd_mod, SI_SUB_VFS, SI_ORDER_ANY);
7121 
7122 /* So that loader and kldload(2) can find us, wherever we are.. */
7123 MODULE_VERSION(nfsd, 1);
7124 MODULE_DEPEND(nfsd, nfscommon, 1, 1, 1);
7125 MODULE_DEPEND(nfsd, nfslockd, 1, 1, 1);
7126 MODULE_DEPEND(nfsd, krpc, 1, 1, 1);
7127 MODULE_DEPEND(nfsd, nfssvc, 1, 1, 1);
7128