xref: /freebsd/sys/fs/nfsserver/nfs_nfsdport.c (revision b214fcceacad6b842545150664bd2695c1c2b34f)
1 /*-
2  * SPDX-License-Identifier: BSD-3-Clause
3  *
4  * Copyright (c) 1989, 1993
5  *	The Regents of the University of California.  All rights reserved.
6  *
7  * This code is derived from software contributed to Berkeley by
8  * Rick Macklem at The University of Guelph.
9  *
10  * Redistribution and use in source and binary forms, with or without
11  * modification, are permitted provided that the following conditions
12  * are met:
13  * 1. Redistributions of source code must retain the above copyright
14  *    notice, this list of conditions and the following disclaimer.
15  * 2. Redistributions in binary form must reproduce the above copyright
16  *    notice, this list of conditions and the following disclaimer in the
17  *    documentation and/or other materials provided with the distribution.
18  * 3. Neither the name of the University nor the names of its contributors
19  *    may be used to endorse or promote products derived from this software
20  *    without specific prior written permission.
21  *
22  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
23  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
24  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
25  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
26  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
27  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
28  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
29  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
30  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
31  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
32  * SUCH DAMAGE.
33  *
34  */
35 
36 #include <sys/cdefs.h>
37 __FBSDID("$FreeBSD$");
38 
39 #include <sys/capsicum.h>
40 #include <sys/extattr.h>
41 
42 /*
43  * Functions that perform the vfs operations required by the routines in
44  * nfsd_serv.c. It is hoped that this change will make the server more
45  * portable.
46  */
47 
48 #include <fs/nfs/nfsport.h>
49 #include <security/mac/mac_framework.h>
50 #include <sys/filio.h>
51 #include <sys/hash.h>
52 #include <sys/sysctl.h>
53 #include <nlm/nlm_prot.h>
54 #include <nlm/nlm.h>
55 
56 FEATURE(nfsd, "NFSv4 server");
57 
58 extern u_int32_t newnfs_true, newnfs_false, newnfs_xdrneg1;
59 extern int nfsrv_useacl;
60 extern int newnfs_numnfsd;
61 extern struct mount nfsv4root_mnt;
62 extern struct nfsrv_stablefirst nfsrv_stablefirst;
63 extern void (*nfsd_call_servertimer)(void);
64 extern SVCPOOL	*nfsrvd_pool;
65 extern struct nfsv4lock nfsd_suspend_lock;
66 extern struct nfsclienthashhead *nfsclienthash;
67 extern struct nfslockhashhead *nfslockhash;
68 extern struct nfssessionhash *nfssessionhash;
69 extern int nfsrv_sessionhashsize;
70 extern struct nfsstatsv1 nfsstatsv1;
71 extern struct nfslayouthash *nfslayouthash;
72 extern int nfsrv_layouthashsize;
73 extern struct mtx nfsrv_dslock_mtx;
74 extern int nfs_pnfsiothreads;
75 extern struct nfsdontlisthead nfsrv_dontlisthead;
76 extern volatile int nfsrv_dontlistlen;
77 extern volatile int nfsrv_devidcnt;
78 extern int nfsrv_maxpnfsmirror;
79 extern uint32_t nfs_srvmaxio;
80 extern int nfs_bufpackets;
81 extern u_long sb_max_adj;
82 struct vfsoptlist nfsv4root_opt, nfsv4root_newopt;
83 NFSDLOCKMUTEX;
84 NFSSTATESPINLOCK;
85 struct nfsrchash_bucket nfsrchash_table[NFSRVCACHE_HASHSIZE];
86 struct nfsrchash_bucket nfsrcahash_table[NFSRVCACHE_HASHSIZE];
87 struct mtx nfsrc_udpmtx;
88 struct mtx nfs_v4root_mutex;
89 struct mtx nfsrv_dontlistlock_mtx;
90 struct mtx nfsrv_recalllock_mtx;
91 struct nfsrvfh nfs_rootfh, nfs_pubfh;
92 int nfs_pubfhset = 0, nfs_rootfhset = 0;
93 struct proc *nfsd_master_proc = NULL;
94 int nfsd_debuglevel = 0;
95 static pid_t nfsd_master_pid = (pid_t)-1;
96 static char nfsd_master_comm[MAXCOMLEN + 1];
97 static struct timeval nfsd_master_start;
98 static uint32_t nfsv4_sysid = 0;
99 static fhandle_t zerofh;
100 
101 static int nfssvc_srvcall(struct thread *, struct nfssvc_args *,
102     struct ucred *);
103 
104 int nfsrv_enable_crossmntpt = 1;
105 static int nfs_commit_blks;
106 static int nfs_commit_miss;
107 extern int nfsrv_issuedelegs;
108 extern int nfsrv_dolocallocks;
109 extern int nfsd_enable_stringtouid;
110 extern struct nfsdevicehead nfsrv_devidhead;
111 
112 static int nfsrv_createiovec(int, struct mbuf **, struct mbuf **,
113     struct iovec **);
114 static int nfsrv_createiovec_extpgs(int, int, struct mbuf **,
115     struct mbuf **, struct iovec **);
116 static int nfsrv_createiovecw(int, struct mbuf *, char *, struct iovec **,
117     int *);
118 static void nfsrv_pnfscreate(struct vnode *, struct vattr *, struct ucred *,
119     NFSPROC_T *);
120 static void nfsrv_pnfsremovesetup(struct vnode *, NFSPROC_T *, struct vnode **,
121     int *, char *, fhandle_t *);
122 static void nfsrv_pnfsremove(struct vnode **, int, char *, fhandle_t *,
123     NFSPROC_T *);
124 static int nfsrv_proxyds(struct vnode *, off_t, int, struct ucred *,
125     struct thread *, int, struct mbuf **, char *, struct mbuf **,
126     struct nfsvattr *, struct acl *, off_t *, int, bool *);
127 static int nfsrv_setextattr(struct vnode *, struct nfsvattr *, NFSPROC_T *);
128 static int nfsrv_readdsrpc(fhandle_t *, off_t, int, struct ucred *,
129     NFSPROC_T *, struct nfsmount *, struct mbuf **, struct mbuf **);
130 static int nfsrv_writedsrpc(fhandle_t *, off_t, int, struct ucred *,
131     NFSPROC_T *, struct vnode *, struct nfsmount **, int, struct mbuf **,
132     char *, int *);
133 static int nfsrv_allocatedsrpc(fhandle_t *, off_t, off_t, struct ucred *,
134     NFSPROC_T *, struct vnode *, struct nfsmount **, int, int *);
135 static int nfsrv_deallocatedsrpc(fhandle_t *, off_t, off_t, struct ucred *,
136     NFSPROC_T *, struct vnode *, struct nfsmount **, int, int *);
137 static int nfsrv_setacldsrpc(fhandle_t *, struct ucred *, NFSPROC_T *,
138     struct vnode *, struct nfsmount **, int, struct acl *, int *);
139 static int nfsrv_setattrdsrpc(fhandle_t *, struct ucred *, NFSPROC_T *,
140     struct vnode *, struct nfsmount **, int, struct nfsvattr *, int *);
141 static int nfsrv_getattrdsrpc(fhandle_t *, struct ucred *, NFSPROC_T *,
142     struct vnode *, struct nfsmount *, struct nfsvattr *);
143 static int nfsrv_seekdsrpc(fhandle_t *, off_t *, int, bool *, struct ucred *,
144     NFSPROC_T *, struct nfsmount *);
145 static int nfsrv_putfhname(fhandle_t *, char *);
146 static int nfsrv_pnfslookupds(struct vnode *, struct vnode *,
147     struct pnfsdsfile *, struct vnode **, NFSPROC_T *);
148 static void nfsrv_pnfssetfh(struct vnode *, struct pnfsdsfile *, char *, char *,
149     struct vnode *, NFSPROC_T *);
150 static int nfsrv_dsremove(struct vnode *, char *, struct ucred *, NFSPROC_T *);
151 static int nfsrv_dssetacl(struct vnode *, struct acl *, struct ucred *,
152     NFSPROC_T *);
153 static int nfsrv_pnfsstatfs(struct statfs *, struct mount *);
154 
155 int nfs_pnfsio(task_fn_t *, void *);
156 
157 SYSCTL_NODE(_vfs, OID_AUTO, nfsd, CTLFLAG_RW | CTLFLAG_MPSAFE, 0,
158     "NFS server");
159 SYSCTL_INT(_vfs_nfsd, OID_AUTO, mirrormnt, CTLFLAG_RW,
160     &nfsrv_enable_crossmntpt, 0, "Enable nfsd to cross mount points");
161 SYSCTL_INT(_vfs_nfsd, OID_AUTO, commit_blks, CTLFLAG_RW, &nfs_commit_blks,
162     0, "");
163 SYSCTL_INT(_vfs_nfsd, OID_AUTO, commit_miss, CTLFLAG_RW, &nfs_commit_miss,
164     0, "");
165 SYSCTL_INT(_vfs_nfsd, OID_AUTO, issue_delegations, CTLFLAG_RW,
166     &nfsrv_issuedelegs, 0, "Enable nfsd to issue delegations");
167 SYSCTL_INT(_vfs_nfsd, OID_AUTO, enable_locallocks, CTLFLAG_RW,
168     &nfsrv_dolocallocks, 0, "Enable nfsd to acquire local locks on files");
169 SYSCTL_INT(_vfs_nfsd, OID_AUTO, debuglevel, CTLFLAG_RW, &nfsd_debuglevel,
170     0, "Debug level for NFS server");
171 SYSCTL_INT(_vfs_nfsd, OID_AUTO, enable_stringtouid, CTLFLAG_RW,
172     &nfsd_enable_stringtouid, 0, "Enable nfsd to accept numeric owner_names");
173 static int nfsrv_pnfsgetdsattr = 1;
174 SYSCTL_INT(_vfs_nfsd, OID_AUTO, pnfsgetdsattr, CTLFLAG_RW,
175     &nfsrv_pnfsgetdsattr, 0, "When set getattr gets DS attributes via RPC");
176 
177 /*
178  * nfsrv_dsdirsize can only be increased and only when the nfsd threads are
179  * not running.
180  * The dsN subdirectories for the increased values must have been created
181  * on all DS servers before this increase is done.
182  */
183 u_int	nfsrv_dsdirsize = 20;
184 static int
185 sysctl_dsdirsize(SYSCTL_HANDLER_ARGS)
186 {
187 	int error, newdsdirsize;
188 
189 	newdsdirsize = nfsrv_dsdirsize;
190 	error = sysctl_handle_int(oidp, &newdsdirsize, 0, req);
191 	if (error != 0 || req->newptr == NULL)
192 		return (error);
193 	if (newdsdirsize <= nfsrv_dsdirsize || newdsdirsize > 10000 ||
194 	    newnfs_numnfsd != 0)
195 		return (EINVAL);
196 	nfsrv_dsdirsize = newdsdirsize;
197 	return (0);
198 }
199 SYSCTL_PROC(_vfs_nfsd, OID_AUTO, dsdirsize,
200     CTLTYPE_UINT | CTLFLAG_MPSAFE | CTLFLAG_RW, 0, sizeof(nfsrv_dsdirsize),
201     sysctl_dsdirsize, "IU", "Number of dsN subdirs on the DS servers");
202 
203 /*
204  * nfs_srvmaxio can only be increased and only when the nfsd threads are
205  * not running.  The setting must be a power of 2, with the current limit of
206  * 1Mbyte.
207  */
208 static int
209 sysctl_srvmaxio(SYSCTL_HANDLER_ARGS)
210 {
211 	int error;
212 	u_int newsrvmaxio;
213 	uint64_t tval;
214 
215 	newsrvmaxio = nfs_srvmaxio;
216 	error = sysctl_handle_int(oidp, &newsrvmaxio, 0, req);
217 	if (error != 0 || req->newptr == NULL)
218 		return (error);
219 	if (newsrvmaxio == nfs_srvmaxio)
220 		return (0);
221 	if (newsrvmaxio < nfs_srvmaxio) {
222 		printf("nfsd: vfs.nfsd.srvmaxio can only be increased\n");
223 		return (EINVAL);
224 	}
225 	if (newsrvmaxio > 1048576) {
226 		printf("nfsd: vfs.nfsd.srvmaxio cannot be > 1Mbyte\n");
227 		return (EINVAL);
228 	}
229 	if ((newsrvmaxio & (newsrvmaxio - 1)) != 0) {
230 		printf("nfsd: vfs.nfsd.srvmaxio must be a power of 2\n");
231 		return (EINVAL);
232 	}
233 
234 	/*
235 	 * Check that kern.ipc.maxsockbuf is large enough for
236 	 * newsrviomax, given the setting of vfs.nfs.bufpackets.
237 	 */
238 	if ((newsrvmaxio + NFS_MAXXDR) * nfs_bufpackets >
239 	    sb_max_adj) {
240 		/*
241 		 * Suggest vfs.nfs.bufpackets * maximum RPC message for
242 		 * sb_max_adj.
243 		 */
244 		tval = (newsrvmaxio + NFS_MAXXDR) * nfs_bufpackets;
245 
246 		/*
247 		 * Convert suggested sb_max_adj value to a suggested
248 		 * sb_max value, which is what is set via kern.ipc.maxsockbuf.
249 		 * Perform the inverse calculation of (from uipc_sockbuf.c):
250 		 * sb_max_adj = (u_quad_t)sb_max * MCLBYTES /
251 		 *     (MSIZE + MCLBYTES);
252 		 * XXX If the calculation of sb_max_adj from sb_max changes,
253 		 *     this calculation must be changed as well.
254 		 */
255 		tval *= (MSIZE + MCLBYTES);  /* Brackets for readability. */
256 		tval += MCLBYTES - 1;        /* Round up divide. */
257 		tval /= MCLBYTES;
258 		printf("nfsd: set kern.ipc.maxsockbuf to a minimum of "
259 		    "%ju to support %ubyte NFS I/O\n", (uintmax_t)tval,
260 		    newsrvmaxio);
261 		return (EINVAL);
262 	}
263 
264 	NFSD_LOCK();
265 	if (newnfs_numnfsd != 0) {
266 		NFSD_UNLOCK();
267 		printf("nfsd: cannot set vfs.nfsd.srvmaxio when nfsd "
268 		    "threads are running\n");
269 		return (EINVAL);
270 	}
271 
272 
273 	nfs_srvmaxio = newsrvmaxio;
274 	NFSD_UNLOCK();
275 	return (0);
276 }
277 SYSCTL_PROC(_vfs_nfsd, OID_AUTO, srvmaxio,
278     CTLTYPE_UINT | CTLFLAG_MPSAFE | CTLFLAG_RW, NULL, 0,
279     sysctl_srvmaxio, "IU", "Maximum I/O size in bytes");
280 
281 #define	MAX_REORDERED_RPC	16
282 #define	NUM_HEURISTIC		1031
283 #define	NHUSE_INIT		64
284 #define	NHUSE_INC		16
285 #define	NHUSE_MAX		2048
286 
287 static struct nfsheur {
288 	struct vnode *nh_vp;	/* vp to match (unreferenced pointer) */
289 	off_t nh_nextoff;	/* next offset for sequential detection */
290 	int nh_use;		/* use count for selection */
291 	int nh_seqcount;	/* heuristic */
292 } nfsheur[NUM_HEURISTIC];
293 
294 /*
295  * Heuristic to detect sequential operation.
296  */
297 static struct nfsheur *
298 nfsrv_sequential_heuristic(struct uio *uio, struct vnode *vp)
299 {
300 	struct nfsheur *nh;
301 	int hi, try;
302 
303 	/* Locate best candidate. */
304 	try = 32;
305 	hi = ((int)(vm_offset_t)vp / sizeof(struct vnode)) % NUM_HEURISTIC;
306 	nh = &nfsheur[hi];
307 	while (try--) {
308 		if (nfsheur[hi].nh_vp == vp) {
309 			nh = &nfsheur[hi];
310 			break;
311 		}
312 		if (nfsheur[hi].nh_use > 0)
313 			--nfsheur[hi].nh_use;
314 		hi = (hi + 1) % NUM_HEURISTIC;
315 		if (nfsheur[hi].nh_use < nh->nh_use)
316 			nh = &nfsheur[hi];
317 	}
318 
319 	/* Initialize hint if this is a new file. */
320 	if (nh->nh_vp != vp) {
321 		nh->nh_vp = vp;
322 		nh->nh_nextoff = uio->uio_offset;
323 		nh->nh_use = NHUSE_INIT;
324 		if (uio->uio_offset == 0)
325 			nh->nh_seqcount = 4;
326 		else
327 			nh->nh_seqcount = 1;
328 	}
329 
330 	/* Calculate heuristic. */
331 	if ((uio->uio_offset == 0 && nh->nh_seqcount > 0) ||
332 	    uio->uio_offset == nh->nh_nextoff) {
333 		/* See comments in vfs_vnops.c:sequential_heuristic(). */
334 		nh->nh_seqcount += howmany(uio->uio_resid, 16384);
335 		if (nh->nh_seqcount > IO_SEQMAX)
336 			nh->nh_seqcount = IO_SEQMAX;
337 	} else if (qabs(uio->uio_offset - nh->nh_nextoff) <= MAX_REORDERED_RPC *
338 	    imax(vp->v_mount->mnt_stat.f_iosize, uio->uio_resid)) {
339 		/* Probably a reordered RPC, leave seqcount alone. */
340 	} else if (nh->nh_seqcount > 1) {
341 		nh->nh_seqcount /= 2;
342 	} else {
343 		nh->nh_seqcount = 0;
344 	}
345 	nh->nh_use += NHUSE_INC;
346 	if (nh->nh_use > NHUSE_MAX)
347 		nh->nh_use = NHUSE_MAX;
348 	return (nh);
349 }
350 
351 /*
352  * Get attributes into nfsvattr structure.
353  */
354 int
355 nfsvno_getattr(struct vnode *vp, struct nfsvattr *nvap,
356     struct nfsrv_descript *nd, struct thread *p, int vpislocked,
357     nfsattrbit_t *attrbitp)
358 {
359 	int error, gotattr, lockedit = 0;
360 	struct nfsvattr na;
361 
362 	if (vpislocked == 0) {
363 		/*
364 		 * When vpislocked == 0, the vnode is either exclusively
365 		 * locked by this thread or not locked by this thread.
366 		 * As such, shared lock it, if not exclusively locked.
367 		 */
368 		if (NFSVOPISLOCKED(vp) != LK_EXCLUSIVE) {
369 			lockedit = 1;
370 			NFSVOPLOCK(vp, LK_SHARED | LK_RETRY);
371 		}
372 	}
373 
374 	/*
375 	 * Acquire the Change, Size, TimeAccess, TimeModify and SpaceUsed
376 	 * attributes, as required.
377 	 * This needs to be done for regular files if:
378 	 * - non-NFSv4 RPCs or
379 	 * - when attrbitp == NULL or
380 	 * - an NFSv4 RPC with any of the above attributes in attrbitp.
381 	 * A return of 0 for nfsrv_proxyds() indicates that it has acquired
382 	 * these attributes.  nfsrv_proxyds() will return an error if the
383 	 * server is not a pNFS one.
384 	 */
385 	gotattr = 0;
386 	if (vp->v_type == VREG && nfsrv_devidcnt > 0 && (attrbitp == NULL ||
387 	    (nd->nd_flag & ND_NFSV4) == 0 ||
388 	    NFSISSET_ATTRBIT(attrbitp, NFSATTRBIT_CHANGE) ||
389 	    NFSISSET_ATTRBIT(attrbitp, NFSATTRBIT_SIZE) ||
390 	    NFSISSET_ATTRBIT(attrbitp, NFSATTRBIT_TIMEACCESS) ||
391 	    NFSISSET_ATTRBIT(attrbitp, NFSATTRBIT_TIMEMODIFY) ||
392 	    NFSISSET_ATTRBIT(attrbitp, NFSATTRBIT_SPACEUSED))) {
393 		error = nfsrv_proxyds(vp, 0, 0, nd->nd_cred, p,
394 		    NFSPROC_GETATTR, NULL, NULL, NULL, &na, NULL, NULL, 0,
395 		    NULL);
396 		if (error == 0)
397 			gotattr = 1;
398 	}
399 
400 	error = VOP_GETATTR(vp, &nvap->na_vattr, nd->nd_cred);
401 	if (lockedit != 0)
402 		NFSVOPUNLOCK(vp);
403 
404 	/*
405 	 * If we got the Change, Size and Modify Time from the DS,
406 	 * replace them.
407 	 */
408 	if (gotattr != 0) {
409 		nvap->na_atime = na.na_atime;
410 		nvap->na_mtime = na.na_mtime;
411 		nvap->na_filerev = na.na_filerev;
412 		nvap->na_size = na.na_size;
413 		nvap->na_bytes = na.na_bytes;
414 	}
415 	NFSD_DEBUG(4, "nfsvno_getattr: gotattr=%d err=%d chg=%ju\n", gotattr,
416 	    error, (uintmax_t)na.na_filerev);
417 
418 	NFSEXITCODE(error);
419 	return (error);
420 }
421 
422 /*
423  * Get a file handle for a vnode.
424  */
425 int
426 nfsvno_getfh(struct vnode *vp, fhandle_t *fhp, struct thread *p)
427 {
428 	int error;
429 
430 	NFSBZERO((caddr_t)fhp, sizeof(fhandle_t));
431 	fhp->fh_fsid = vp->v_mount->mnt_stat.f_fsid;
432 	error = VOP_VPTOFH(vp, &fhp->fh_fid);
433 
434 	NFSEXITCODE(error);
435 	return (error);
436 }
437 
438 /*
439  * Perform access checking for vnodes obtained from file handles that would
440  * refer to files already opened by a Unix client. You cannot just use
441  * vn_writechk() and VOP_ACCESSX() for two reasons.
442  * 1 - You must check for exported rdonly as well as MNT_RDONLY for the write
443  *     case.
444  * 2 - The owner is to be given access irrespective of mode bits for some
445  *     operations, so that processes that chmod after opening a file don't
446  *     break.
447  */
448 int
449 nfsvno_accchk(struct vnode *vp, accmode_t accmode, struct ucred *cred,
450     struct nfsexstuff *exp, struct thread *p, int override, int vpislocked,
451     u_int32_t *supportedtypep)
452 {
453 	struct vattr vattr;
454 	int error = 0, getret = 0;
455 
456 	if (vpislocked == 0) {
457 		if (NFSVOPLOCK(vp, LK_SHARED) != 0) {
458 			error = EPERM;
459 			goto out;
460 		}
461 	}
462 	if (accmode & VWRITE) {
463 		/* Just vn_writechk() changed to check rdonly */
464 		/*
465 		 * Disallow write attempts on read-only file systems;
466 		 * unless the file is a socket or a block or character
467 		 * device resident on the file system.
468 		 */
469 		if (NFSVNO_EXRDONLY(exp) ||
470 		    (vp->v_mount->mnt_flag & MNT_RDONLY)) {
471 			switch (vp->v_type) {
472 			case VREG:
473 			case VDIR:
474 			case VLNK:
475 				error = EROFS;
476 			default:
477 				break;
478 			}
479 		}
480 		/*
481 		 * If there's shared text associated with
482 		 * the inode, try to free it up once.  If
483 		 * we fail, we can't allow writing.
484 		 */
485 		if (VOP_IS_TEXT(vp) && error == 0)
486 			error = ETXTBSY;
487 	}
488 	if (error != 0) {
489 		if (vpislocked == 0)
490 			NFSVOPUNLOCK(vp);
491 		goto out;
492 	}
493 
494 	/*
495 	 * Should the override still be applied when ACLs are enabled?
496 	 */
497 	error = VOP_ACCESSX(vp, accmode, cred, p);
498 	if (error != 0 && (accmode & (VDELETE | VDELETE_CHILD))) {
499 		/*
500 		 * Try again with VEXPLICIT_DENY, to see if the test for
501 		 * deletion is supported.
502 		 */
503 		error = VOP_ACCESSX(vp, accmode | VEXPLICIT_DENY, cred, p);
504 		if (error == 0) {
505 			if (vp->v_type == VDIR) {
506 				accmode &= ~(VDELETE | VDELETE_CHILD);
507 				accmode |= VWRITE;
508 				error = VOP_ACCESSX(vp, accmode, cred, p);
509 			} else if (supportedtypep != NULL) {
510 				*supportedtypep &= ~NFSACCESS_DELETE;
511 			}
512 		}
513 	}
514 
515 	/*
516 	 * Allow certain operations for the owner (reads and writes
517 	 * on files that are already open).
518 	 */
519 	if (override != NFSACCCHK_NOOVERRIDE &&
520 	    (error == EPERM || error == EACCES)) {
521 		if (cred->cr_uid == 0 && (override & NFSACCCHK_ALLOWROOT))
522 			error = 0;
523 		else if (override & NFSACCCHK_ALLOWOWNER) {
524 			getret = VOP_GETATTR(vp, &vattr, cred);
525 			if (getret == 0 && cred->cr_uid == vattr.va_uid)
526 				error = 0;
527 		}
528 	}
529 	if (vpislocked == 0)
530 		NFSVOPUNLOCK(vp);
531 
532 out:
533 	NFSEXITCODE(error);
534 	return (error);
535 }
536 
537 /*
538  * Set attribute(s) vnop.
539  */
540 int
541 nfsvno_setattr(struct vnode *vp, struct nfsvattr *nvap, struct ucred *cred,
542     struct thread *p, struct nfsexstuff *exp)
543 {
544 	u_quad_t savsize = 0;
545 	int error, savedit;
546 	time_t savbtime;
547 
548 	/*
549 	 * If this is an exported file system and a pNFS service is running,
550 	 * don't VOP_SETATTR() of size for the MDS file system.
551 	 */
552 	savedit = 0;
553 	error = 0;
554 	if (vp->v_type == VREG && (vp->v_mount->mnt_flag & MNT_EXPORTED) != 0 &&
555 	    nfsrv_devidcnt != 0 && nvap->na_vattr.va_size != VNOVAL &&
556 	    nvap->na_vattr.va_size > 0) {
557 		savsize = nvap->na_vattr.va_size;
558 		nvap->na_vattr.va_size = VNOVAL;
559 		if (nvap->na_vattr.va_uid != (uid_t)VNOVAL ||
560 		    nvap->na_vattr.va_gid != (gid_t)VNOVAL ||
561 		    nvap->na_vattr.va_mode != (mode_t)VNOVAL ||
562 		    nvap->na_vattr.va_atime.tv_sec != VNOVAL ||
563 		    nvap->na_vattr.va_mtime.tv_sec != VNOVAL)
564 			savedit = 1;
565 		else
566 			savedit = 2;
567 	}
568 	if (savedit != 2)
569 		error = VOP_SETATTR(vp, &nvap->na_vattr, cred);
570 	if (savedit != 0)
571 		nvap->na_vattr.va_size = savsize;
572 	if (error == 0 && (nvap->na_vattr.va_uid != (uid_t)VNOVAL ||
573 	    nvap->na_vattr.va_gid != (gid_t)VNOVAL ||
574 	    nvap->na_vattr.va_size != VNOVAL ||
575 	    nvap->na_vattr.va_mode != (mode_t)VNOVAL ||
576 	    nvap->na_vattr.va_atime.tv_sec != VNOVAL ||
577 	    nvap->na_vattr.va_mtime.tv_sec != VNOVAL)) {
578 		/* Never modify birthtime on a DS file. */
579 		savbtime = nvap->na_vattr.va_birthtime.tv_sec;
580 		nvap->na_vattr.va_birthtime.tv_sec = VNOVAL;
581 		/* For a pNFS server, set the attributes on the DS file. */
582 		error = nfsrv_proxyds(vp, 0, 0, cred, p, NFSPROC_SETATTR,
583 		    NULL, NULL, NULL, nvap, NULL, NULL, 0, NULL);
584 		nvap->na_vattr.va_birthtime.tv_sec = savbtime;
585 		if (error == ENOENT)
586 			error = 0;
587 	}
588 	NFSEXITCODE(error);
589 	return (error);
590 }
591 
592 /*
593  * Set up nameidata for a lookup() call and do it.
594  */
595 int
596 nfsvno_namei(struct nfsrv_descript *nd, struct nameidata *ndp,
597     struct vnode *dp, int islocked, struct nfsexstuff *exp,
598     struct vnode **retdirp)
599 {
600 	struct componentname *cnp = &ndp->ni_cnd;
601 	int i;
602 	struct iovec aiov;
603 	struct uio auio;
604 	int lockleaf = (cnp->cn_flags & LOCKLEAF) != 0, linklen;
605 	int error = 0;
606 	char *cp;
607 
608 	*retdirp = NULL;
609 	cnp->cn_nameptr = cnp->cn_pnbuf;
610 	ndp->ni_lcf = 0;
611 	/*
612 	 * Extract and set starting directory.
613 	 */
614 	if (dp->v_type != VDIR) {
615 		if (islocked)
616 			vput(dp);
617 		else
618 			vrele(dp);
619 		nfsvno_relpathbuf(ndp);
620 		error = ENOTDIR;
621 		goto out1;
622 	}
623 	if (islocked)
624 		NFSVOPUNLOCK(dp);
625 	VREF(dp);
626 	*retdirp = dp;
627 	if (NFSVNO_EXRDONLY(exp))
628 		cnp->cn_flags |= RDONLY;
629 	ndp->ni_segflg = UIO_SYSSPACE;
630 
631 	if (nd->nd_flag & ND_PUBLOOKUP) {
632 		ndp->ni_loopcnt = 0;
633 		if (cnp->cn_pnbuf[0] == '/') {
634 			vrele(dp);
635 			/*
636 			 * Check for degenerate pathnames here, since lookup()
637 			 * panics on them.
638 			 */
639 			for (i = 1; i < ndp->ni_pathlen; i++)
640 				if (cnp->cn_pnbuf[i] != '/')
641 					break;
642 			if (i == ndp->ni_pathlen) {
643 				error = NFSERR_ACCES;
644 				goto out;
645 			}
646 			dp = rootvnode;
647 			VREF(dp);
648 		}
649 	} else if ((nfsrv_enable_crossmntpt == 0 && NFSVNO_EXPORTED(exp)) ||
650 	    (nd->nd_flag & ND_NFSV4) == 0) {
651 		/*
652 		 * Only cross mount points for NFSv4 when doing a
653 		 * mount while traversing the file system above
654 		 * the mount point, unless nfsrv_enable_crossmntpt is set.
655 		 */
656 		cnp->cn_flags |= NOCROSSMOUNT;
657 	}
658 
659 	/*
660 	 * Initialize for scan, set ni_startdir and bump ref on dp again
661 	 * because lookup() will dereference ni_startdir.
662 	 */
663 
664 	ndp->ni_startdir = dp;
665 	ndp->ni_rootdir = rootvnode;
666 	ndp->ni_topdir = NULL;
667 
668 	if (!lockleaf)
669 		cnp->cn_flags |= LOCKLEAF;
670 	for (;;) {
671 		cnp->cn_nameptr = cnp->cn_pnbuf;
672 		/*
673 		 * Call lookup() to do the real work.  If an error occurs,
674 		 * ndp->ni_vp and ni_dvp are left uninitialized or NULL and
675 		 * we do not have to dereference anything before returning.
676 		 * In either case ni_startdir will be dereferenced and NULLed
677 		 * out.
678 		 */
679 		error = lookup(ndp);
680 		if (error)
681 			break;
682 
683 		/*
684 		 * Check for encountering a symbolic link.  Trivial
685 		 * termination occurs if no symlink encountered.
686 		 */
687 		if ((cnp->cn_flags & ISSYMLINK) == 0) {
688 			if ((cnp->cn_flags & (SAVENAME | SAVESTART)) == 0)
689 				nfsvno_relpathbuf(ndp);
690 			if (ndp->ni_vp && !lockleaf)
691 				NFSVOPUNLOCK(ndp->ni_vp);
692 			break;
693 		}
694 
695 		/*
696 		 * Validate symlink
697 		 */
698 		if ((cnp->cn_flags & LOCKPARENT) && ndp->ni_pathlen == 1)
699 			NFSVOPUNLOCK(ndp->ni_dvp);
700 		if (!(nd->nd_flag & ND_PUBLOOKUP)) {
701 			error = EINVAL;
702 			goto badlink2;
703 		}
704 
705 		if (ndp->ni_loopcnt++ >= MAXSYMLINKS) {
706 			error = ELOOP;
707 			goto badlink2;
708 		}
709 		if (ndp->ni_pathlen > 1)
710 			cp = uma_zalloc(namei_zone, M_WAITOK);
711 		else
712 			cp = cnp->cn_pnbuf;
713 		aiov.iov_base = cp;
714 		aiov.iov_len = MAXPATHLEN;
715 		auio.uio_iov = &aiov;
716 		auio.uio_iovcnt = 1;
717 		auio.uio_offset = 0;
718 		auio.uio_rw = UIO_READ;
719 		auio.uio_segflg = UIO_SYSSPACE;
720 		auio.uio_td = NULL;
721 		auio.uio_resid = MAXPATHLEN;
722 		error = VOP_READLINK(ndp->ni_vp, &auio, cnp->cn_cred);
723 		if (error) {
724 		badlink1:
725 			if (ndp->ni_pathlen > 1)
726 				uma_zfree(namei_zone, cp);
727 		badlink2:
728 			vrele(ndp->ni_dvp);
729 			vput(ndp->ni_vp);
730 			break;
731 		}
732 		linklen = MAXPATHLEN - auio.uio_resid;
733 		if (linklen == 0) {
734 			error = ENOENT;
735 			goto badlink1;
736 		}
737 		if (linklen + ndp->ni_pathlen >= MAXPATHLEN) {
738 			error = ENAMETOOLONG;
739 			goto badlink1;
740 		}
741 
742 		/*
743 		 * Adjust or replace path
744 		 */
745 		if (ndp->ni_pathlen > 1) {
746 			NFSBCOPY(ndp->ni_next, cp + linklen, ndp->ni_pathlen);
747 			uma_zfree(namei_zone, cnp->cn_pnbuf);
748 			cnp->cn_pnbuf = cp;
749 		} else
750 			cnp->cn_pnbuf[linklen] = '\0';
751 		ndp->ni_pathlen += linklen;
752 
753 		/*
754 		 * Cleanup refs for next loop and check if root directory
755 		 * should replace current directory.  Normally ni_dvp
756 		 * becomes the new base directory and is cleaned up when
757 		 * we loop.  Explicitly null pointers after invalidation
758 		 * to clarify operation.
759 		 */
760 		vput(ndp->ni_vp);
761 		ndp->ni_vp = NULL;
762 
763 		if (cnp->cn_pnbuf[0] == '/') {
764 			vrele(ndp->ni_dvp);
765 			ndp->ni_dvp = ndp->ni_rootdir;
766 			VREF(ndp->ni_dvp);
767 		}
768 		ndp->ni_startdir = ndp->ni_dvp;
769 		ndp->ni_dvp = NULL;
770 	}
771 	if (!lockleaf)
772 		cnp->cn_flags &= ~LOCKLEAF;
773 
774 out:
775 	if (error) {
776 		nfsvno_relpathbuf(ndp);
777 		ndp->ni_vp = NULL;
778 		ndp->ni_dvp = NULL;
779 		ndp->ni_startdir = NULL;
780 	} else if ((ndp->ni_cnd.cn_flags & (WANTPARENT|LOCKPARENT)) == 0) {
781 		ndp->ni_dvp = NULL;
782 	}
783 
784 out1:
785 	NFSEXITCODE2(error, nd);
786 	return (error);
787 }
788 
789 /*
790  * Set up a pathname buffer and return a pointer to it and, optionally
791  * set a hash pointer.
792  */
793 void
794 nfsvno_setpathbuf(struct nameidata *ndp, char **bufpp, u_long **hashpp)
795 {
796 	struct componentname *cnp = &ndp->ni_cnd;
797 
798 	cnp->cn_flags |= (NOMACCHECK | HASBUF);
799 	cnp->cn_pnbuf = uma_zalloc(namei_zone, M_WAITOK);
800 	if (hashpp != NULL)
801 		*hashpp = NULL;
802 	*bufpp = cnp->cn_pnbuf;
803 }
804 
805 /*
806  * Release the above path buffer, if not released by nfsvno_namei().
807  */
808 void
809 nfsvno_relpathbuf(struct nameidata *ndp)
810 {
811 
812 	if ((ndp->ni_cnd.cn_flags & HASBUF) == 0)
813 		panic("nfsrelpath");
814 	uma_zfree(namei_zone, ndp->ni_cnd.cn_pnbuf);
815 	ndp->ni_cnd.cn_flags &= ~HASBUF;
816 }
817 
818 /*
819  * Readlink vnode op into an mbuf list.
820  */
821 int
822 nfsvno_readlink(struct vnode *vp, struct ucred *cred, int maxextsiz,
823     struct thread *p, struct mbuf **mpp, struct mbuf **mpendp, int *lenp)
824 {
825 	struct iovec *iv;
826 	struct uio io, *uiop = &io;
827 	struct mbuf *mp, *mp3;
828 	int len, tlen, error = 0;
829 
830 	len = NFS_MAXPATHLEN;
831 	if (maxextsiz > 0)
832 		uiop->uio_iovcnt = nfsrv_createiovec_extpgs(len, maxextsiz,
833 		    &mp3, &mp, &iv);
834 	else
835 		uiop->uio_iovcnt = nfsrv_createiovec(len, &mp3, &mp, &iv);
836 	uiop->uio_iov = iv;
837 	uiop->uio_offset = 0;
838 	uiop->uio_resid = len;
839 	uiop->uio_rw = UIO_READ;
840 	uiop->uio_segflg = UIO_SYSSPACE;
841 	uiop->uio_td = NULL;
842 	error = VOP_READLINK(vp, uiop, cred);
843 	free(iv, M_TEMP);
844 	if (error) {
845 		m_freem(mp3);
846 		*lenp = 0;
847 		goto out;
848 	}
849 	if (uiop->uio_resid > 0) {
850 		len -= uiop->uio_resid;
851 		tlen = NFSM_RNDUP(len);
852 		if (tlen == 0) {
853 			m_freem(mp3);
854 			mp3 = mp = NULL;
855 		} else if (tlen != NFS_MAXPATHLEN || tlen != len)
856 			mp = nfsrv_adj(mp3, NFS_MAXPATHLEN - tlen,
857 			    tlen - len);
858 	}
859 	*lenp = len;
860 	*mpp = mp3;
861 	*mpendp = mp;
862 
863 out:
864 	NFSEXITCODE(error);
865 	return (error);
866 }
867 
868 /*
869  * Create an mbuf chain and an associated iovec that can be used to Read
870  * or Getextattr of data.
871  * Upon success, return pointers to the first and last mbufs in the chain
872  * plus the malloc'd iovec and its iovlen.
873  */
874 static int
875 nfsrv_createiovec(int len, struct mbuf **mpp, struct mbuf **mpendp,
876     struct iovec **ivp)
877 {
878 	struct mbuf *m, *m2 = NULL, *m3;
879 	struct iovec *iv;
880 	int i, left, siz;
881 
882 	left = len;
883 	m3 = NULL;
884 	/*
885 	 * Generate the mbuf list with the uio_iov ref. to it.
886 	 */
887 	i = 0;
888 	while (left > 0) {
889 		NFSMGET(m);
890 		MCLGET(m, M_WAITOK);
891 		m->m_len = 0;
892 		siz = min(M_TRAILINGSPACE(m), left);
893 		left -= siz;
894 		i++;
895 		if (m3)
896 			m2->m_next = m;
897 		else
898 			m3 = m;
899 		m2 = m;
900 	}
901 	*ivp = iv = malloc(i * sizeof (struct iovec), M_TEMP, M_WAITOK);
902 	m = m3;
903 	left = len;
904 	i = 0;
905 	while (left > 0) {
906 		if (m == NULL)
907 			panic("nfsrv_createiovec iov");
908 		siz = min(M_TRAILINGSPACE(m), left);
909 		if (siz > 0) {
910 			iv->iov_base = mtod(m, caddr_t) + m->m_len;
911 			iv->iov_len = siz;
912 			m->m_len += siz;
913 			left -= siz;
914 			iv++;
915 			i++;
916 		}
917 		m = m->m_next;
918 	}
919 	*mpp = m3;
920 	*mpendp = m2;
921 	return (i);
922 }
923 
924 /*
925  * Create an mbuf chain and an associated iovec that can be used to Read
926  * or Getextattr of data.
927  * Upon success, return pointers to the first and last mbufs in the chain
928  * plus the malloc'd iovec and its iovlen.
929  * Same as above, but creates ext_pgs mbuf(s).
930  */
931 static int
932 nfsrv_createiovec_extpgs(int len, int maxextsiz, struct mbuf **mpp,
933     struct mbuf **mpendp, struct iovec **ivp)
934 {
935 	struct mbuf *m, *m2 = NULL, *m3;
936 	struct iovec *iv;
937 	int i, left, pgno, siz;
938 
939 	left = len;
940 	m3 = NULL;
941 	/*
942 	 * Generate the mbuf list with the uio_iov ref. to it.
943 	 */
944 	i = 0;
945 	while (left > 0) {
946 		siz = min(left, maxextsiz);
947 		m = mb_alloc_ext_plus_pages(siz, M_WAITOK);
948 		left -= siz;
949 		i += m->m_epg_npgs;
950 		if (m3 != NULL)
951 			m2->m_next = m;
952 		else
953 			m3 = m;
954 		m2 = m;
955 	}
956 	*ivp = iv = malloc(i * sizeof (struct iovec), M_TEMP, M_WAITOK);
957 	m = m3;
958 	left = len;
959 	i = 0;
960 	pgno = 0;
961 	while (left > 0) {
962 		if (m == NULL)
963 			panic("nfsvno_createiovec_extpgs iov");
964 		siz = min(PAGE_SIZE, left);
965 		if (siz > 0) {
966 			iv->iov_base = (void *)PHYS_TO_DMAP(m->m_epg_pa[pgno]);
967 			iv->iov_len = siz;
968 			m->m_len += siz;
969 			if (pgno == m->m_epg_npgs - 1)
970 				m->m_epg_last_len = siz;
971 			left -= siz;
972 			iv++;
973 			i++;
974 			pgno++;
975 		}
976 		if (pgno == m->m_epg_npgs && left > 0) {
977 			m = m->m_next;
978 			if (m == NULL)
979 				panic("nfsvno_createiovec_extpgs iov");
980 			pgno = 0;
981 		}
982 	}
983 	*mpp = m3;
984 	*mpendp = m2;
985 	return (i);
986 }
987 
988 /*
989  * Read vnode op call into mbuf list.
990  */
991 int
992 nfsvno_read(struct vnode *vp, off_t off, int cnt, struct ucred *cred,
993     int maxextsiz, struct thread *p, struct mbuf **mpp,
994     struct mbuf **mpendp)
995 {
996 	struct mbuf *m;
997 	struct iovec *iv;
998 	int error = 0, len, tlen, ioflag = 0;
999 	struct mbuf *m3;
1000 	struct uio io, *uiop = &io;
1001 	struct nfsheur *nh;
1002 
1003 	/*
1004 	 * Attempt to read from a DS file. A return of ENOENT implies
1005 	 * there is no DS file to read.
1006 	 */
1007 	error = nfsrv_proxyds(vp, off, cnt, cred, p, NFSPROC_READDS, mpp,
1008 	    NULL, mpendp, NULL, NULL, NULL, 0, NULL);
1009 	if (error != ENOENT)
1010 		return (error);
1011 
1012 	len = NFSM_RNDUP(cnt);
1013 	if (maxextsiz > 0)
1014 		uiop->uio_iovcnt = nfsrv_createiovec_extpgs(len, maxextsiz,
1015 		    &m3, &m, &iv);
1016 	else
1017 		uiop->uio_iovcnt = nfsrv_createiovec(len, &m3, &m, &iv);
1018 	uiop->uio_iov = iv;
1019 	uiop->uio_offset = off;
1020 	uiop->uio_resid = len;
1021 	uiop->uio_rw = UIO_READ;
1022 	uiop->uio_segflg = UIO_SYSSPACE;
1023 	uiop->uio_td = NULL;
1024 	nh = nfsrv_sequential_heuristic(uiop, vp);
1025 	ioflag |= nh->nh_seqcount << IO_SEQSHIFT;
1026 	/* XXX KDM make this more systematic? */
1027 	nfsstatsv1.srvbytes[NFSV4OP_READ] += uiop->uio_resid;
1028 	error = VOP_READ(vp, uiop, IO_NODELOCKED | ioflag, cred);
1029 	free(iv, M_TEMP);
1030 	if (error) {
1031 		m_freem(m3);
1032 		*mpp = NULL;
1033 		goto out;
1034 	}
1035 	nh->nh_nextoff = uiop->uio_offset;
1036 	tlen = len - uiop->uio_resid;
1037 	cnt = cnt < tlen ? cnt : tlen;
1038 	tlen = NFSM_RNDUP(cnt);
1039 	if (tlen == 0) {
1040 		m_freem(m3);
1041 		m3 = m = NULL;
1042 	} else if (len != tlen || tlen != cnt)
1043 		m = nfsrv_adj(m3, len - tlen, tlen - cnt);
1044 	*mpp = m3;
1045 	*mpendp = m;
1046 
1047 out:
1048 	NFSEXITCODE(error);
1049 	return (error);
1050 }
1051 
1052 /*
1053  * Create the iovec for the mbuf chain passed in as an argument.
1054  * The "cp" argument is where the data starts within the first mbuf in
1055  * the chain. It returns the iovec and the iovcnt.
1056  */
1057 static int
1058 nfsrv_createiovecw(int retlen, struct mbuf *m, char *cp, struct iovec **ivpp,
1059     int *iovcntp)
1060 {
1061 	struct mbuf *mp;
1062 	struct iovec *ivp;
1063 	int cnt, i, len;
1064 
1065 	/*
1066 	 * Loop through the mbuf chain, counting how many mbufs are a
1067 	 * part of this write operation, so the iovec size is known.
1068 	 */
1069 	cnt = 0;
1070 	len = retlen;
1071 	mp = m;
1072 	i = mtod(mp, caddr_t) + mp->m_len - cp;
1073 	while (len > 0) {
1074 		if (i > 0) {
1075 			len -= i;
1076 			cnt++;
1077 		}
1078 		mp = mp->m_next;
1079 		if (!mp) {
1080 			if (len > 0)
1081 				return (EBADRPC);
1082 		} else
1083 			i = mp->m_len;
1084 	}
1085 
1086 	/* Now, create the iovec. */
1087 	mp = m;
1088 	*ivpp = ivp = malloc(cnt * sizeof (struct iovec), M_TEMP,
1089 	    M_WAITOK);
1090 	*iovcntp = cnt;
1091 	i = mtod(mp, caddr_t) + mp->m_len - cp;
1092 	len = retlen;
1093 	while (len > 0) {
1094 		if (mp == NULL)
1095 			panic("nfsrv_createiovecw");
1096 		if (i > 0) {
1097 			i = min(i, len);
1098 			ivp->iov_base = cp;
1099 			ivp->iov_len = i;
1100 			ivp++;
1101 			len -= i;
1102 		}
1103 		mp = mp->m_next;
1104 		if (mp) {
1105 			i = mp->m_len;
1106 			cp = mtod(mp, caddr_t);
1107 		}
1108 	}
1109 	return (0);
1110 }
1111 
1112 /*
1113  * Write vnode op from an mbuf list.
1114  */
1115 int
1116 nfsvno_write(struct vnode *vp, off_t off, int retlen, int *stable,
1117     struct mbuf *mp, char *cp, struct ucred *cred, struct thread *p)
1118 {
1119 	struct iovec *iv;
1120 	int cnt, ioflags, error;
1121 	struct uio io, *uiop = &io;
1122 	struct nfsheur *nh;
1123 
1124 	/*
1125 	 * Attempt to write to a DS file. A return of ENOENT implies
1126 	 * there is no DS file to write.
1127 	 */
1128 	error = nfsrv_proxyds(vp, off, retlen, cred, p, NFSPROC_WRITEDS,
1129 	    &mp, cp, NULL, NULL, NULL, NULL, 0, NULL);
1130 	if (error != ENOENT) {
1131 		*stable = NFSWRITE_FILESYNC;
1132 		return (error);
1133 	}
1134 
1135 	if (*stable == NFSWRITE_UNSTABLE)
1136 		ioflags = IO_NODELOCKED;
1137 	else
1138 		ioflags = (IO_SYNC | IO_NODELOCKED);
1139 	error = nfsrv_createiovecw(retlen, mp, cp, &iv, &cnt);
1140 	if (error != 0)
1141 		return (error);
1142 	uiop->uio_iov = iv;
1143 	uiop->uio_iovcnt = cnt;
1144 	uiop->uio_resid = retlen;
1145 	uiop->uio_rw = UIO_WRITE;
1146 	uiop->uio_segflg = UIO_SYSSPACE;
1147 	NFSUIOPROC(uiop, p);
1148 	uiop->uio_offset = off;
1149 	nh = nfsrv_sequential_heuristic(uiop, vp);
1150 	ioflags |= nh->nh_seqcount << IO_SEQSHIFT;
1151 	/* XXX KDM make this more systematic? */
1152 	nfsstatsv1.srvbytes[NFSV4OP_WRITE] += uiop->uio_resid;
1153 	error = VOP_WRITE(vp, uiop, ioflags, cred);
1154 	if (error == 0)
1155 		nh->nh_nextoff = uiop->uio_offset;
1156 	free(iv, M_TEMP);
1157 
1158 	NFSEXITCODE(error);
1159 	return (error);
1160 }
1161 
1162 /*
1163  * Common code for creating a regular file (plus special files for V2).
1164  */
1165 int
1166 nfsvno_createsub(struct nfsrv_descript *nd, struct nameidata *ndp,
1167     struct vnode **vpp, struct nfsvattr *nvap, int *exclusive_flagp,
1168     int32_t *cverf, NFSDEV_T rdev, struct nfsexstuff *exp)
1169 {
1170 	u_quad_t tempsize;
1171 	int error;
1172 	struct thread *p = curthread;
1173 
1174 	error = nd->nd_repstat;
1175 	if (!error && ndp->ni_vp == NULL) {
1176 		if (nvap->na_type == VREG || nvap->na_type == VSOCK) {
1177 			vrele(ndp->ni_startdir);
1178 			error = VOP_CREATE(ndp->ni_dvp,
1179 			    &ndp->ni_vp, &ndp->ni_cnd, &nvap->na_vattr);
1180 			/* For a pNFS server, create the data file on a DS. */
1181 			if (error == 0 && nvap->na_type == VREG) {
1182 				/*
1183 				 * Create a data file on a DS for a pNFS server.
1184 				 * This function just returns if not
1185 				 * running a pNFS DS or the creation fails.
1186 				 */
1187 				nfsrv_pnfscreate(ndp->ni_vp, &nvap->na_vattr,
1188 				    nd->nd_cred, p);
1189 			}
1190 			VOP_VPUT_PAIR(ndp->ni_dvp, error == 0 ? &ndp->ni_vp :
1191 			    NULL, false);
1192 			nfsvno_relpathbuf(ndp);
1193 			if (!error) {
1194 				if (*exclusive_flagp) {
1195 					*exclusive_flagp = 0;
1196 					NFSVNO_ATTRINIT(nvap);
1197 					nvap->na_atime.tv_sec = cverf[0];
1198 					nvap->na_atime.tv_nsec = cverf[1];
1199 					error = VOP_SETATTR(ndp->ni_vp,
1200 					    &nvap->na_vattr, nd->nd_cred);
1201 					if (error != 0) {
1202 						vput(ndp->ni_vp);
1203 						ndp->ni_vp = NULL;
1204 						error = NFSERR_NOTSUPP;
1205 					}
1206 				}
1207 			}
1208 		/*
1209 		 * NFS V2 Only. nfsrvd_mknod() does this for V3.
1210 		 * (This implies, just get out on an error.)
1211 		 */
1212 		} else if (nvap->na_type == VCHR || nvap->na_type == VBLK ||
1213 			nvap->na_type == VFIFO) {
1214 			if (nvap->na_type == VCHR && rdev == 0xffffffff)
1215 				nvap->na_type = VFIFO;
1216                         if (nvap->na_type != VFIFO &&
1217 			    (error = priv_check_cred(nd->nd_cred, PRIV_VFS_MKNOD_DEV))) {
1218 				vrele(ndp->ni_startdir);
1219 				nfsvno_relpathbuf(ndp);
1220 				vput(ndp->ni_dvp);
1221 				goto out;
1222 			}
1223 			nvap->na_rdev = rdev;
1224 			error = VOP_MKNOD(ndp->ni_dvp, &ndp->ni_vp,
1225 			    &ndp->ni_cnd, &nvap->na_vattr);
1226 			VOP_VPUT_PAIR(ndp->ni_dvp, error == 0 ? &ndp->ni_vp :
1227 			    NULL, false);
1228 			nfsvno_relpathbuf(ndp);
1229 			vrele(ndp->ni_startdir);
1230 			if (error)
1231 				goto out;
1232 		} else {
1233 			vrele(ndp->ni_startdir);
1234 			nfsvno_relpathbuf(ndp);
1235 			vput(ndp->ni_dvp);
1236 			error = ENXIO;
1237 			goto out;
1238 		}
1239 		*vpp = ndp->ni_vp;
1240 	} else {
1241 		/*
1242 		 * Handle cases where error is already set and/or
1243 		 * the file exists.
1244 		 * 1 - clean up the lookup
1245 		 * 2 - iff !error and na_size set, truncate it
1246 		 */
1247 		vrele(ndp->ni_startdir);
1248 		nfsvno_relpathbuf(ndp);
1249 		*vpp = ndp->ni_vp;
1250 		if (ndp->ni_dvp == *vpp)
1251 			vrele(ndp->ni_dvp);
1252 		else
1253 			vput(ndp->ni_dvp);
1254 		if (!error && nvap->na_size != VNOVAL) {
1255 			error = nfsvno_accchk(*vpp, VWRITE,
1256 			    nd->nd_cred, exp, p, NFSACCCHK_NOOVERRIDE,
1257 			    NFSACCCHK_VPISLOCKED, NULL);
1258 			if (!error) {
1259 				tempsize = nvap->na_size;
1260 				NFSVNO_ATTRINIT(nvap);
1261 				nvap->na_size = tempsize;
1262 				error = VOP_SETATTR(*vpp,
1263 				    &nvap->na_vattr, nd->nd_cred);
1264 			}
1265 		}
1266 		if (error)
1267 			vput(*vpp);
1268 	}
1269 
1270 out:
1271 	NFSEXITCODE(error);
1272 	return (error);
1273 }
1274 
1275 /*
1276  * Do a mknod vnode op.
1277  */
1278 int
1279 nfsvno_mknod(struct nameidata *ndp, struct nfsvattr *nvap, struct ucred *cred,
1280     struct thread *p)
1281 {
1282 	int error = 0;
1283 	enum vtype vtyp;
1284 
1285 	vtyp = nvap->na_type;
1286 	/*
1287 	 * Iff doesn't exist, create it.
1288 	 */
1289 	if (ndp->ni_vp) {
1290 		vrele(ndp->ni_startdir);
1291 		nfsvno_relpathbuf(ndp);
1292 		vput(ndp->ni_dvp);
1293 		vrele(ndp->ni_vp);
1294 		error = EEXIST;
1295 		goto out;
1296 	}
1297 	if (vtyp != VCHR && vtyp != VBLK && vtyp != VSOCK && vtyp != VFIFO) {
1298 		vrele(ndp->ni_startdir);
1299 		nfsvno_relpathbuf(ndp);
1300 		vput(ndp->ni_dvp);
1301 		error = NFSERR_BADTYPE;
1302 		goto out;
1303 	}
1304 	if (vtyp == VSOCK) {
1305 		vrele(ndp->ni_startdir);
1306 		error = VOP_CREATE(ndp->ni_dvp, &ndp->ni_vp,
1307 		    &ndp->ni_cnd, &nvap->na_vattr);
1308 		VOP_VPUT_PAIR(ndp->ni_dvp, error == 0 ? &ndp->ni_vp : NULL,
1309 		    false);
1310 		nfsvno_relpathbuf(ndp);
1311 	} else {
1312 		if (nvap->na_type != VFIFO &&
1313 		    (error = priv_check_cred(cred, PRIV_VFS_MKNOD_DEV))) {
1314 			vrele(ndp->ni_startdir);
1315 			nfsvno_relpathbuf(ndp);
1316 			vput(ndp->ni_dvp);
1317 			goto out;
1318 		}
1319 		error = VOP_MKNOD(ndp->ni_dvp, &ndp->ni_vp,
1320 		    &ndp->ni_cnd, &nvap->na_vattr);
1321 		VOP_VPUT_PAIR(ndp->ni_dvp, error == 0 ? &ndp->ni_vp : NULL,
1322 		    false);
1323 		nfsvno_relpathbuf(ndp);
1324 		vrele(ndp->ni_startdir);
1325 		/*
1326 		 * Since VOP_MKNOD returns the ni_vp, I can't
1327 		 * see any reason to do the lookup.
1328 		 */
1329 	}
1330 
1331 out:
1332 	NFSEXITCODE(error);
1333 	return (error);
1334 }
1335 
1336 /*
1337  * Mkdir vnode op.
1338  */
1339 int
1340 nfsvno_mkdir(struct nameidata *ndp, struct nfsvattr *nvap, uid_t saved_uid,
1341     struct ucred *cred, struct thread *p, struct nfsexstuff *exp)
1342 {
1343 	int error = 0;
1344 
1345 	if (ndp->ni_vp != NULL) {
1346 		if (ndp->ni_dvp == ndp->ni_vp)
1347 			vrele(ndp->ni_dvp);
1348 		else
1349 			vput(ndp->ni_dvp);
1350 		vrele(ndp->ni_vp);
1351 		nfsvno_relpathbuf(ndp);
1352 		error = EEXIST;
1353 		goto out;
1354 	}
1355 	error = VOP_MKDIR(ndp->ni_dvp, &ndp->ni_vp, &ndp->ni_cnd,
1356 	    &nvap->na_vattr);
1357 	VOP_VPUT_PAIR(ndp->ni_dvp, error == 0 ? &ndp->ni_vp : NULL, false);
1358 	nfsvno_relpathbuf(ndp);
1359 
1360 out:
1361 	NFSEXITCODE(error);
1362 	return (error);
1363 }
1364 
1365 /*
1366  * symlink vnode op.
1367  */
1368 int
1369 nfsvno_symlink(struct nameidata *ndp, struct nfsvattr *nvap, char *pathcp,
1370     int pathlen, int not_v2, uid_t saved_uid, struct ucred *cred, struct thread *p,
1371     struct nfsexstuff *exp)
1372 {
1373 	int error = 0;
1374 
1375 	if (ndp->ni_vp) {
1376 		vrele(ndp->ni_startdir);
1377 		nfsvno_relpathbuf(ndp);
1378 		if (ndp->ni_dvp == ndp->ni_vp)
1379 			vrele(ndp->ni_dvp);
1380 		else
1381 			vput(ndp->ni_dvp);
1382 		vrele(ndp->ni_vp);
1383 		error = EEXIST;
1384 		goto out;
1385 	}
1386 
1387 	error = VOP_SYMLINK(ndp->ni_dvp, &ndp->ni_vp, &ndp->ni_cnd,
1388 	    &nvap->na_vattr, pathcp);
1389 	/*
1390 	 * Although FreeBSD still had the lookup code in
1391 	 * it for 7/current, there doesn't seem to be any
1392 	 * point, since VOP_SYMLINK() returns the ni_vp.
1393 	 * Just vput it for v2.
1394 	 */
1395 	VOP_VPUT_PAIR(ndp->ni_dvp, &ndp->ni_vp, !not_v2 && error == 0);
1396 	vrele(ndp->ni_startdir);
1397 	nfsvno_relpathbuf(ndp);
1398 
1399 out:
1400 	NFSEXITCODE(error);
1401 	return (error);
1402 }
1403 
1404 /*
1405  * Parse symbolic link arguments.
1406  * This function has an ugly side effect. It will malloc() an area for
1407  * the symlink and set iov_base to point to it, only if it succeeds.
1408  * So, if it returns with uiop->uio_iov->iov_base != NULL, that must
1409  * be FREE'd later.
1410  */
1411 int
1412 nfsvno_getsymlink(struct nfsrv_descript *nd, struct nfsvattr *nvap,
1413     struct thread *p, char **pathcpp, int *lenp)
1414 {
1415 	u_int32_t *tl;
1416 	char *pathcp = NULL;
1417 	int error = 0, len;
1418 	struct nfsv2_sattr *sp;
1419 
1420 	*pathcpp = NULL;
1421 	*lenp = 0;
1422 	if ((nd->nd_flag & ND_NFSV3) &&
1423 	    (error = nfsrv_sattr(nd, NULL, nvap, NULL, NULL, p)))
1424 		goto nfsmout;
1425 	NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
1426 	len = fxdr_unsigned(int, *tl);
1427 	if (len > NFS_MAXPATHLEN || len <= 0) {
1428 		error = EBADRPC;
1429 		goto nfsmout;
1430 	}
1431 	pathcp = malloc(len + 1, M_TEMP, M_WAITOK);
1432 	error = nfsrv_mtostr(nd, pathcp, len);
1433 	if (error)
1434 		goto nfsmout;
1435 	if (nd->nd_flag & ND_NFSV2) {
1436 		NFSM_DISSECT(sp, struct nfsv2_sattr *, NFSX_V2SATTR);
1437 		nvap->na_mode = fxdr_unsigned(u_int16_t, sp->sa_mode);
1438 	}
1439 	*pathcpp = pathcp;
1440 	*lenp = len;
1441 	NFSEXITCODE2(0, nd);
1442 	return (0);
1443 nfsmout:
1444 	if (pathcp)
1445 		free(pathcp, M_TEMP);
1446 	NFSEXITCODE2(error, nd);
1447 	return (error);
1448 }
1449 
1450 /*
1451  * Remove a non-directory object.
1452  */
1453 int
1454 nfsvno_removesub(struct nameidata *ndp, int is_v4, struct ucred *cred,
1455     struct thread *p, struct nfsexstuff *exp)
1456 {
1457 	struct vnode *vp, *dsdvp[NFSDEV_MAXMIRRORS];
1458 	int error = 0, mirrorcnt;
1459 	char fname[PNFS_FILENAME_LEN + 1];
1460 	fhandle_t fh;
1461 
1462 	vp = ndp->ni_vp;
1463 	dsdvp[0] = NULL;
1464 	if (vp->v_type == VDIR)
1465 		error = NFSERR_ISDIR;
1466 	else if (is_v4)
1467 		error = nfsrv_checkremove(vp, 1, NULL, (nfsquad_t)((u_quad_t)0),
1468 		    p);
1469 	if (error == 0)
1470 		nfsrv_pnfsremovesetup(vp, p, dsdvp, &mirrorcnt, fname, &fh);
1471 	if (!error)
1472 		error = VOP_REMOVE(ndp->ni_dvp, vp, &ndp->ni_cnd);
1473 	if (error == 0 && dsdvp[0] != NULL)
1474 		nfsrv_pnfsremove(dsdvp, mirrorcnt, fname, &fh, p);
1475 	if (ndp->ni_dvp == vp)
1476 		vrele(ndp->ni_dvp);
1477 	else
1478 		vput(ndp->ni_dvp);
1479 	vput(vp);
1480 	if ((ndp->ni_cnd.cn_flags & SAVENAME) != 0)
1481 		nfsvno_relpathbuf(ndp);
1482 	NFSEXITCODE(error);
1483 	return (error);
1484 }
1485 
1486 /*
1487  * Remove a directory.
1488  */
1489 int
1490 nfsvno_rmdirsub(struct nameidata *ndp, int is_v4, struct ucred *cred,
1491     struct thread *p, struct nfsexstuff *exp)
1492 {
1493 	struct vnode *vp;
1494 	int error = 0;
1495 
1496 	vp = ndp->ni_vp;
1497 	if (vp->v_type != VDIR) {
1498 		error = ENOTDIR;
1499 		goto out;
1500 	}
1501 	/*
1502 	 * No rmdir "." please.
1503 	 */
1504 	if (ndp->ni_dvp == vp) {
1505 		error = EINVAL;
1506 		goto out;
1507 	}
1508 	/*
1509 	 * The root of a mounted filesystem cannot be deleted.
1510 	 */
1511 	if (vp->v_vflag & VV_ROOT)
1512 		error = EBUSY;
1513 out:
1514 	if (!error)
1515 		error = VOP_RMDIR(ndp->ni_dvp, vp, &ndp->ni_cnd);
1516 	if (ndp->ni_dvp == vp)
1517 		vrele(ndp->ni_dvp);
1518 	else
1519 		vput(ndp->ni_dvp);
1520 	vput(vp);
1521 	if ((ndp->ni_cnd.cn_flags & SAVENAME) != 0)
1522 		nfsvno_relpathbuf(ndp);
1523 	NFSEXITCODE(error);
1524 	return (error);
1525 }
1526 
1527 /*
1528  * Rename vnode op.
1529  */
1530 int
1531 nfsvno_rename(struct nameidata *fromndp, struct nameidata *tondp,
1532     u_int32_t ndstat, u_int32_t ndflag, struct ucred *cred, struct thread *p)
1533 {
1534 	struct vnode *fvp, *tvp, *tdvp, *dsdvp[NFSDEV_MAXMIRRORS];
1535 	int error = 0, mirrorcnt;
1536 	char fname[PNFS_FILENAME_LEN + 1];
1537 	fhandle_t fh;
1538 
1539 	dsdvp[0] = NULL;
1540 	fvp = fromndp->ni_vp;
1541 	if (ndstat) {
1542 		vrele(fromndp->ni_dvp);
1543 		vrele(fvp);
1544 		error = ndstat;
1545 		goto out1;
1546 	}
1547 	tdvp = tondp->ni_dvp;
1548 	tvp = tondp->ni_vp;
1549 	if (tvp != NULL) {
1550 		if (fvp->v_type == VDIR && tvp->v_type != VDIR) {
1551 			error = (ndflag & ND_NFSV2) ? EISDIR : EEXIST;
1552 			goto out;
1553 		} else if (fvp->v_type != VDIR && tvp->v_type == VDIR) {
1554 			error = (ndflag & ND_NFSV2) ? ENOTDIR : EEXIST;
1555 			goto out;
1556 		}
1557 		if (tvp->v_type == VDIR && tvp->v_mountedhere) {
1558 			error = (ndflag & ND_NFSV2) ? ENOTEMPTY : EXDEV;
1559 			goto out;
1560 		}
1561 
1562 		/*
1563 		 * A rename to '.' or '..' results in a prematurely
1564 		 * unlocked vnode on FreeBSD5, so I'm just going to fail that
1565 		 * here.
1566 		 */
1567 		if ((tondp->ni_cnd.cn_namelen == 1 &&
1568 		     tondp->ni_cnd.cn_nameptr[0] == '.') ||
1569 		    (tondp->ni_cnd.cn_namelen == 2 &&
1570 		     tondp->ni_cnd.cn_nameptr[0] == '.' &&
1571 		     tondp->ni_cnd.cn_nameptr[1] == '.')) {
1572 			error = EINVAL;
1573 			goto out;
1574 		}
1575 	}
1576 	if (fvp->v_type == VDIR && fvp->v_mountedhere) {
1577 		error = (ndflag & ND_NFSV2) ? ENOTEMPTY : EXDEV;
1578 		goto out;
1579 	}
1580 	if (fvp->v_mount != tdvp->v_mount) {
1581 		error = (ndflag & ND_NFSV2) ? ENOTEMPTY : EXDEV;
1582 		goto out;
1583 	}
1584 	if (fvp == tdvp) {
1585 		error = (ndflag & ND_NFSV2) ? ENOTEMPTY : EINVAL;
1586 		goto out;
1587 	}
1588 	if (fvp == tvp) {
1589 		/*
1590 		 * If source and destination are the same, there is nothing to
1591 		 * do. Set error to -1 to indicate this.
1592 		 */
1593 		error = -1;
1594 		goto out;
1595 	}
1596 	if (ndflag & ND_NFSV4) {
1597 		if (NFSVOPLOCK(fvp, LK_EXCLUSIVE) == 0) {
1598 			error = nfsrv_checkremove(fvp, 0, NULL,
1599 			    (nfsquad_t)((u_quad_t)0), p);
1600 			NFSVOPUNLOCK(fvp);
1601 		} else
1602 			error = EPERM;
1603 		if (tvp && !error)
1604 			error = nfsrv_checkremove(tvp, 1, NULL,
1605 			    (nfsquad_t)((u_quad_t)0), p);
1606 	} else {
1607 		/*
1608 		 * For NFSv2 and NFSv3, try to get rid of the delegation, so
1609 		 * that the NFSv4 client won't be confused by the rename.
1610 		 * Since nfsd_recalldelegation() can only be called on an
1611 		 * unlocked vnode at this point and fvp is the file that will
1612 		 * still exist after the rename, just do fvp.
1613 		 */
1614 		nfsd_recalldelegation(fvp, p);
1615 	}
1616 	if (error == 0 && tvp != NULL) {
1617 		nfsrv_pnfsremovesetup(tvp, p, dsdvp, &mirrorcnt, fname, &fh);
1618 		NFSD_DEBUG(4, "nfsvno_rename: pnfsremovesetup"
1619 		    " dsdvp=%p\n", dsdvp[0]);
1620 	}
1621 out:
1622 	if (!error) {
1623 		error = VOP_RENAME(fromndp->ni_dvp, fromndp->ni_vp,
1624 		    &fromndp->ni_cnd, tondp->ni_dvp, tondp->ni_vp,
1625 		    &tondp->ni_cnd);
1626 	} else {
1627 		if (tdvp == tvp)
1628 			vrele(tdvp);
1629 		else
1630 			vput(tdvp);
1631 		if (tvp)
1632 			vput(tvp);
1633 		vrele(fromndp->ni_dvp);
1634 		vrele(fvp);
1635 		if (error == -1)
1636 			error = 0;
1637 	}
1638 
1639 	/*
1640 	 * If dsdvp[0] != NULL, it was set up by nfsrv_pnfsremovesetup() and
1641 	 * if the rename succeeded, the DS file for the tvp needs to be
1642 	 * removed.
1643 	 */
1644 	if (error == 0 && dsdvp[0] != NULL) {
1645 		nfsrv_pnfsremove(dsdvp, mirrorcnt, fname, &fh, p);
1646 		NFSD_DEBUG(4, "nfsvno_rename: pnfsremove\n");
1647 	}
1648 
1649 	vrele(tondp->ni_startdir);
1650 	nfsvno_relpathbuf(tondp);
1651 out1:
1652 	vrele(fromndp->ni_startdir);
1653 	nfsvno_relpathbuf(fromndp);
1654 	NFSEXITCODE(error);
1655 	return (error);
1656 }
1657 
1658 /*
1659  * Link vnode op.
1660  */
1661 int
1662 nfsvno_link(struct nameidata *ndp, struct vnode *vp, struct ucred *cred,
1663     struct thread *p, struct nfsexstuff *exp)
1664 {
1665 	struct vnode *xp;
1666 	int error = 0;
1667 
1668 	xp = ndp->ni_vp;
1669 	if (xp != NULL) {
1670 		error = EEXIST;
1671 	} else {
1672 		xp = ndp->ni_dvp;
1673 		if (vp->v_mount != xp->v_mount)
1674 			error = EXDEV;
1675 	}
1676 	if (!error) {
1677 		NFSVOPLOCK(vp, LK_EXCLUSIVE | LK_RETRY);
1678 		if (!VN_IS_DOOMED(vp))
1679 			error = VOP_LINK(ndp->ni_dvp, vp, &ndp->ni_cnd);
1680 		else
1681 			error = EPERM;
1682 		if (ndp->ni_dvp == vp) {
1683 			vrele(ndp->ni_dvp);
1684 			NFSVOPUNLOCK(vp);
1685 		} else {
1686 			vref(vp);
1687 			VOP_VPUT_PAIR(ndp->ni_dvp, &vp, true);
1688 		}
1689 	} else {
1690 		if (ndp->ni_dvp == ndp->ni_vp)
1691 			vrele(ndp->ni_dvp);
1692 		else
1693 			vput(ndp->ni_dvp);
1694 		if (ndp->ni_vp)
1695 			vrele(ndp->ni_vp);
1696 	}
1697 	nfsvno_relpathbuf(ndp);
1698 	NFSEXITCODE(error);
1699 	return (error);
1700 }
1701 
1702 /*
1703  * Do the fsync() appropriate for the commit.
1704  */
1705 int
1706 nfsvno_fsync(struct vnode *vp, u_int64_t off, int cnt, struct ucred *cred,
1707     struct thread *td)
1708 {
1709 	int error = 0;
1710 
1711 	/*
1712 	 * RFC 1813 3.3.21: if count is 0, a flush from offset to the end of
1713 	 * file is done.  At this time VOP_FSYNC does not accept offset and
1714 	 * byte count parameters so call VOP_FSYNC the whole file for now.
1715 	 * The same is true for NFSv4: RFC 3530 Sec. 14.2.3.
1716 	 * File systems that do not use the buffer cache (as indicated
1717 	 * by MNTK_USES_BCACHE not being set) must use VOP_FSYNC().
1718 	 */
1719 	if (cnt == 0 || cnt > MAX_COMMIT_COUNT ||
1720 	    (vp->v_mount->mnt_kern_flag & MNTK_USES_BCACHE) == 0) {
1721 		/*
1722 		 * Give up and do the whole thing
1723 		 */
1724 		if (vp->v_object && vm_object_mightbedirty(vp->v_object)) {
1725 			VM_OBJECT_WLOCK(vp->v_object);
1726 			vm_object_page_clean(vp->v_object, 0, 0, OBJPC_SYNC);
1727 			VM_OBJECT_WUNLOCK(vp->v_object);
1728 		}
1729 		error = VOP_FSYNC(vp, MNT_WAIT, td);
1730 	} else {
1731 		/*
1732 		 * Locate and synchronously write any buffers that fall
1733 		 * into the requested range.  Note:  we are assuming that
1734 		 * f_iosize is a power of 2.
1735 		 */
1736 		int iosize = vp->v_mount->mnt_stat.f_iosize;
1737 		int iomask = iosize - 1;
1738 		struct bufobj *bo;
1739 		daddr_t lblkno;
1740 
1741 		/*
1742 		 * Align to iosize boundary, super-align to page boundary.
1743 		 */
1744 		if (off & iomask) {
1745 			cnt += off & iomask;
1746 			off &= ~(u_quad_t)iomask;
1747 		}
1748 		if (off & PAGE_MASK) {
1749 			cnt += off & PAGE_MASK;
1750 			off &= ~(u_quad_t)PAGE_MASK;
1751 		}
1752 		lblkno = off / iosize;
1753 
1754 		if (vp->v_object && vm_object_mightbedirty(vp->v_object)) {
1755 			VM_OBJECT_WLOCK(vp->v_object);
1756 			vm_object_page_clean(vp->v_object, off, off + cnt,
1757 			    OBJPC_SYNC);
1758 			VM_OBJECT_WUNLOCK(vp->v_object);
1759 		}
1760 
1761 		bo = &vp->v_bufobj;
1762 		BO_LOCK(bo);
1763 		while (cnt > 0) {
1764 			struct buf *bp;
1765 
1766 			/*
1767 			 * If we have a buffer and it is marked B_DELWRI we
1768 			 * have to lock and write it.  Otherwise the prior
1769 			 * write is assumed to have already been committed.
1770 			 *
1771 			 * gbincore() can return invalid buffers now so we
1772 			 * have to check that bit as well (though B_DELWRI
1773 			 * should not be set if B_INVAL is set there could be
1774 			 * a race here since we haven't locked the buffer).
1775 			 */
1776 			if ((bp = gbincore(&vp->v_bufobj, lblkno)) != NULL) {
1777 				if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_SLEEPFAIL |
1778 				    LK_INTERLOCK, BO_LOCKPTR(bo)) == ENOLCK) {
1779 					BO_LOCK(bo);
1780 					continue; /* retry */
1781 				}
1782 			    	if ((bp->b_flags & (B_DELWRI|B_INVAL)) ==
1783 				    B_DELWRI) {
1784 					bremfree(bp);
1785 					bp->b_flags &= ~B_ASYNC;
1786 					bwrite(bp);
1787 					++nfs_commit_miss;
1788 				} else
1789 					BUF_UNLOCK(bp);
1790 				BO_LOCK(bo);
1791 			}
1792 			++nfs_commit_blks;
1793 			if (cnt < iosize)
1794 				break;
1795 			cnt -= iosize;
1796 			++lblkno;
1797 		}
1798 		BO_UNLOCK(bo);
1799 	}
1800 	NFSEXITCODE(error);
1801 	return (error);
1802 }
1803 
1804 /*
1805  * Statfs vnode op.
1806  */
1807 int
1808 nfsvno_statfs(struct vnode *vp, struct statfs *sf)
1809 {
1810 	struct statfs *tsf;
1811 	int error;
1812 
1813 	tsf = NULL;
1814 	if (nfsrv_devidcnt > 0) {
1815 		/* For a pNFS service, get the DS numbers. */
1816 		tsf = malloc(sizeof(*tsf), M_TEMP, M_WAITOK | M_ZERO);
1817 		error = nfsrv_pnfsstatfs(tsf, vp->v_mount);
1818 		if (error != 0) {
1819 			free(tsf, M_TEMP);
1820 			tsf = NULL;
1821 		}
1822 	}
1823 	error = VFS_STATFS(vp->v_mount, sf);
1824 	if (error == 0) {
1825 		if (tsf != NULL) {
1826 			sf->f_blocks = tsf->f_blocks;
1827 			sf->f_bavail = tsf->f_bavail;
1828 			sf->f_bfree = tsf->f_bfree;
1829 			sf->f_bsize = tsf->f_bsize;
1830 		}
1831 		/*
1832 		 * Since NFS handles these values as unsigned on the
1833 		 * wire, there is no way to represent negative values,
1834 		 * so set them to 0. Without this, they will appear
1835 		 * to be very large positive values for clients like
1836 		 * Solaris10.
1837 		 */
1838 		if (sf->f_bavail < 0)
1839 			sf->f_bavail = 0;
1840 		if (sf->f_ffree < 0)
1841 			sf->f_ffree = 0;
1842 	}
1843 	free(tsf, M_TEMP);
1844 	NFSEXITCODE(error);
1845 	return (error);
1846 }
1847 
1848 /*
1849  * Do the vnode op stuff for Open. Similar to nfsvno_createsub(), but
1850  * must handle nfsrv_opencheck() calls after any other access checks.
1851  */
1852 void
1853 nfsvno_open(struct nfsrv_descript *nd, struct nameidata *ndp,
1854     nfsquad_t clientid, nfsv4stateid_t *stateidp, struct nfsstate *stp,
1855     int *exclusive_flagp, struct nfsvattr *nvap, int32_t *cverf, int create,
1856     NFSACL_T *aclp, nfsattrbit_t *attrbitp, struct ucred *cred,
1857     struct nfsexstuff *exp, struct vnode **vpp)
1858 {
1859 	struct vnode *vp = NULL;
1860 	u_quad_t tempsize;
1861 	struct nfsexstuff nes;
1862 	struct thread *p = curthread;
1863 
1864 	if (ndp->ni_vp == NULL)
1865 		nd->nd_repstat = nfsrv_opencheck(clientid,
1866 		    stateidp, stp, NULL, nd, p, nd->nd_repstat);
1867 	if (!nd->nd_repstat) {
1868 		if (ndp->ni_vp == NULL) {
1869 			vrele(ndp->ni_startdir);
1870 			nd->nd_repstat = VOP_CREATE(ndp->ni_dvp,
1871 			    &ndp->ni_vp, &ndp->ni_cnd, &nvap->na_vattr);
1872 			/* For a pNFS server, create the data file on a DS. */
1873 			if (nd->nd_repstat == 0) {
1874 				/*
1875 				 * Create a data file on a DS for a pNFS server.
1876 				 * This function just returns if not
1877 				 * running a pNFS DS or the creation fails.
1878 				 */
1879 				nfsrv_pnfscreate(ndp->ni_vp, &nvap->na_vattr,
1880 				    cred, p);
1881 			}
1882 			VOP_VPUT_PAIR(ndp->ni_dvp, nd->nd_repstat == 0 ?
1883 			    &ndp->ni_vp : NULL, false);
1884 			nfsvno_relpathbuf(ndp);
1885 			if (!nd->nd_repstat) {
1886 				if (*exclusive_flagp) {
1887 					*exclusive_flagp = 0;
1888 					NFSVNO_ATTRINIT(nvap);
1889 					nvap->na_atime.tv_sec = cverf[0];
1890 					nvap->na_atime.tv_nsec = cverf[1];
1891 					nd->nd_repstat = VOP_SETATTR(ndp->ni_vp,
1892 					    &nvap->na_vattr, cred);
1893 					if (nd->nd_repstat != 0) {
1894 						vput(ndp->ni_vp);
1895 						ndp->ni_vp = NULL;
1896 						nd->nd_repstat = NFSERR_NOTSUPP;
1897 					} else
1898 						NFSSETBIT_ATTRBIT(attrbitp,
1899 						    NFSATTRBIT_TIMEACCESS);
1900 				} else {
1901 					nfsrv_fixattr(nd, ndp->ni_vp, nvap,
1902 					    aclp, p, attrbitp, exp);
1903 				}
1904 			}
1905 			vp = ndp->ni_vp;
1906 		} else {
1907 			if (ndp->ni_startdir)
1908 				vrele(ndp->ni_startdir);
1909 			nfsvno_relpathbuf(ndp);
1910 			vp = ndp->ni_vp;
1911 			if (create == NFSV4OPEN_CREATE) {
1912 				if (ndp->ni_dvp == vp)
1913 					vrele(ndp->ni_dvp);
1914 				else
1915 					vput(ndp->ni_dvp);
1916 			}
1917 			if (NFSVNO_ISSETSIZE(nvap) && vp->v_type == VREG) {
1918 				if (ndp->ni_cnd.cn_flags & RDONLY)
1919 					NFSVNO_SETEXRDONLY(&nes);
1920 				else
1921 					NFSVNO_EXINIT(&nes);
1922 				nd->nd_repstat = nfsvno_accchk(vp,
1923 				    VWRITE, cred, &nes, p,
1924 				    NFSACCCHK_NOOVERRIDE,
1925 				    NFSACCCHK_VPISLOCKED, NULL);
1926 				nd->nd_repstat = nfsrv_opencheck(clientid,
1927 				    stateidp, stp, vp, nd, p, nd->nd_repstat);
1928 				if (!nd->nd_repstat) {
1929 					tempsize = nvap->na_size;
1930 					NFSVNO_ATTRINIT(nvap);
1931 					nvap->na_size = tempsize;
1932 					nd->nd_repstat = VOP_SETATTR(vp,
1933 					    &nvap->na_vattr, cred);
1934 				}
1935 			} else if (vp->v_type == VREG) {
1936 				nd->nd_repstat = nfsrv_opencheck(clientid,
1937 				    stateidp, stp, vp, nd, p, nd->nd_repstat);
1938 			}
1939 		}
1940 	} else {
1941 		if (ndp->ni_cnd.cn_flags & HASBUF)
1942 			nfsvno_relpathbuf(ndp);
1943 		if (ndp->ni_startdir && create == NFSV4OPEN_CREATE) {
1944 			vrele(ndp->ni_startdir);
1945 			if (ndp->ni_dvp == ndp->ni_vp)
1946 				vrele(ndp->ni_dvp);
1947 			else
1948 				vput(ndp->ni_dvp);
1949 			if (ndp->ni_vp)
1950 				vput(ndp->ni_vp);
1951 		}
1952 	}
1953 	*vpp = vp;
1954 
1955 	NFSEXITCODE2(0, nd);
1956 }
1957 
1958 /*
1959  * Updates the file rev and sets the mtime and ctime
1960  * to the current clock time, returning the va_filerev and va_Xtime
1961  * values.
1962  * Return ESTALE to indicate the vnode is VIRF_DOOMED.
1963  */
1964 int
1965 nfsvno_updfilerev(struct vnode *vp, struct nfsvattr *nvap,
1966     struct nfsrv_descript *nd, struct thread *p)
1967 {
1968 	struct vattr va;
1969 
1970 	VATTR_NULL(&va);
1971 	vfs_timestamp(&va.va_mtime);
1972 	if (NFSVOPISLOCKED(vp) != LK_EXCLUSIVE) {
1973 		NFSVOPLOCK(vp, LK_UPGRADE | LK_RETRY);
1974 		if (VN_IS_DOOMED(vp))
1975 			return (ESTALE);
1976 	}
1977 	(void) VOP_SETATTR(vp, &va, nd->nd_cred);
1978 	(void) nfsvno_getattr(vp, nvap, nd, p, 1, NULL);
1979 	return (0);
1980 }
1981 
1982 /*
1983  * Glue routine to nfsv4_fillattr().
1984  */
1985 int
1986 nfsvno_fillattr(struct nfsrv_descript *nd, struct mount *mp, struct vnode *vp,
1987     struct nfsvattr *nvap, fhandle_t *fhp, int rderror, nfsattrbit_t *attrbitp,
1988     struct ucred *cred, struct thread *p, int isdgram, int reterr,
1989     int supports_nfsv4acls, int at_root, uint64_t mounted_on_fileno)
1990 {
1991 	struct statfs *sf;
1992 	int error;
1993 
1994 	sf = NULL;
1995 	if (nfsrv_devidcnt > 0 &&
1996 	    (NFSISSET_ATTRBIT(attrbitp, NFSATTRBIT_SPACEAVAIL) ||
1997 	     NFSISSET_ATTRBIT(attrbitp, NFSATTRBIT_SPACEFREE) ||
1998 	     NFSISSET_ATTRBIT(attrbitp, NFSATTRBIT_SPACETOTAL))) {
1999 		sf = malloc(sizeof(*sf), M_TEMP, M_WAITOK | M_ZERO);
2000 		error = nfsrv_pnfsstatfs(sf, mp);
2001 		if (error != 0) {
2002 			free(sf, M_TEMP);
2003 			sf = NULL;
2004 		}
2005 	}
2006 	error = nfsv4_fillattr(nd, mp, vp, NULL, &nvap->na_vattr, fhp, rderror,
2007 	    attrbitp, cred, p, isdgram, reterr, supports_nfsv4acls, at_root,
2008 	    mounted_on_fileno, sf);
2009 	free(sf, M_TEMP);
2010 	NFSEXITCODE2(0, nd);
2011 	return (error);
2012 }
2013 
2014 /* Since the Readdir vnode ops vary, put the entire functions in here. */
2015 /*
2016  * nfs readdir service
2017  * - mallocs what it thinks is enough to read
2018  *	count rounded up to a multiple of DIRBLKSIZ <= NFS_MAXREADDIR
2019  * - calls VOP_READDIR()
2020  * - loops around building the reply
2021  *	if the output generated exceeds count break out of loop
2022  *	The NFSM_CLGET macro is used here so that the reply will be packed
2023  *	tightly in mbuf clusters.
2024  * - it trims out records with d_fileno == 0
2025  *	this doesn't matter for Unix clients, but they might confuse clients
2026  *	for other os'.
2027  * - it trims out records with d_type == DT_WHT
2028  *	these cannot be seen through NFS (unless we extend the protocol)
2029  *     The alternate call nfsrvd_readdirplus() does lookups as well.
2030  * PS: The NFS protocol spec. does not clarify what the "count" byte
2031  *	argument is a count of.. just name strings and file id's or the
2032  *	entire reply rpc or ...
2033  *	I tried just file name and id sizes and it confused the Sun client,
2034  *	so I am using the full rpc size now. The "paranoia.." comment refers
2035  *	to including the status longwords that are not a part of the dir.
2036  *	"entry" structures, but are in the rpc.
2037  */
2038 int
2039 nfsrvd_readdir(struct nfsrv_descript *nd, int isdgram,
2040     struct vnode *vp, struct nfsexstuff *exp)
2041 {
2042 	struct dirent *dp;
2043 	u_int32_t *tl;
2044 	int dirlen;
2045 	char *cpos, *cend, *rbuf;
2046 	struct nfsvattr at;
2047 	int nlen, error = 0, getret = 1;
2048 	int siz, cnt, fullsiz, eofflag, ncookies;
2049 	u_int64_t off, toff, verf __unused;
2050 	uint64_t *cookies = NULL, *cookiep;
2051 	struct uio io;
2052 	struct iovec iv;
2053 	int is_ufs;
2054 	struct thread *p = curthread;
2055 
2056 	if (nd->nd_repstat) {
2057 		nfsrv_postopattr(nd, getret, &at);
2058 		goto out;
2059 	}
2060 	if (nd->nd_flag & ND_NFSV2) {
2061 		NFSM_DISSECT(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
2062 		off = fxdr_unsigned(u_quad_t, *tl++);
2063 	} else {
2064 		NFSM_DISSECT(tl, u_int32_t *, 5 * NFSX_UNSIGNED);
2065 		off = fxdr_hyper(tl);
2066 		tl += 2;
2067 		verf = fxdr_hyper(tl);
2068 		tl += 2;
2069 	}
2070 	toff = off;
2071 	cnt = fxdr_unsigned(int, *tl);
2072 	if (cnt > NFS_SRVMAXDATA(nd) || cnt < 0)
2073 		cnt = NFS_SRVMAXDATA(nd);
2074 	siz = ((cnt + DIRBLKSIZ - 1) & ~(DIRBLKSIZ - 1));
2075 	fullsiz = siz;
2076 	if (nd->nd_flag & ND_NFSV3) {
2077 		nd->nd_repstat = getret = nfsvno_getattr(vp, &at, nd, p, 1,
2078 		    NULL);
2079 #if 0
2080 		/*
2081 		 * va_filerev is not sufficient as a cookie verifier,
2082 		 * since it is not supposed to change when entries are
2083 		 * removed/added unless that offset cookies returned to
2084 		 * the client are no longer valid.
2085 		 */
2086 		if (!nd->nd_repstat && toff && verf != at.na_filerev)
2087 			nd->nd_repstat = NFSERR_BAD_COOKIE;
2088 #endif
2089 	}
2090 	if (!nd->nd_repstat && vp->v_type != VDIR)
2091 		nd->nd_repstat = NFSERR_NOTDIR;
2092 	if (nd->nd_repstat == 0 && cnt == 0) {
2093 		if (nd->nd_flag & ND_NFSV2)
2094 			/* NFSv2 does not have NFSERR_TOOSMALL */
2095 			nd->nd_repstat = EPERM;
2096 		else
2097 			nd->nd_repstat = NFSERR_TOOSMALL;
2098 	}
2099 	if (!nd->nd_repstat)
2100 		nd->nd_repstat = nfsvno_accchk(vp, VEXEC,
2101 		    nd->nd_cred, exp, p, NFSACCCHK_NOOVERRIDE,
2102 		    NFSACCCHK_VPISLOCKED, NULL);
2103 	if (nd->nd_repstat) {
2104 		vput(vp);
2105 		if (nd->nd_flag & ND_NFSV3)
2106 			nfsrv_postopattr(nd, getret, &at);
2107 		goto out;
2108 	}
2109 	is_ufs = strcmp(vp->v_mount->mnt_vfc->vfc_name, "ufs") == 0;
2110 	rbuf = malloc(siz, M_TEMP, M_WAITOK);
2111 again:
2112 	eofflag = 0;
2113 	if (cookies) {
2114 		free(cookies, M_TEMP);
2115 		cookies = NULL;
2116 	}
2117 
2118 	iv.iov_base = rbuf;
2119 	iv.iov_len = siz;
2120 	io.uio_iov = &iv;
2121 	io.uio_iovcnt = 1;
2122 	io.uio_offset = (off_t)off;
2123 	io.uio_resid = siz;
2124 	io.uio_segflg = UIO_SYSSPACE;
2125 	io.uio_rw = UIO_READ;
2126 	io.uio_td = NULL;
2127 	nd->nd_repstat = VOP_READDIR(vp, &io, nd->nd_cred, &eofflag, &ncookies,
2128 	    &cookies);
2129 	off = (u_int64_t)io.uio_offset;
2130 	if (io.uio_resid)
2131 		siz -= io.uio_resid;
2132 
2133 	if (!cookies && !nd->nd_repstat)
2134 		nd->nd_repstat = NFSERR_PERM;
2135 	if (nd->nd_flag & ND_NFSV3) {
2136 		getret = nfsvno_getattr(vp, &at, nd, p, 1, NULL);
2137 		if (!nd->nd_repstat)
2138 			nd->nd_repstat = getret;
2139 	}
2140 
2141 	/*
2142 	 * Handles the failed cases. nd->nd_repstat == 0 past here.
2143 	 */
2144 	if (nd->nd_repstat) {
2145 		vput(vp);
2146 		free(rbuf, M_TEMP);
2147 		if (cookies)
2148 			free(cookies, M_TEMP);
2149 		if (nd->nd_flag & ND_NFSV3)
2150 			nfsrv_postopattr(nd, getret, &at);
2151 		goto out;
2152 	}
2153 	/*
2154 	 * If nothing read, return eof
2155 	 * rpc reply
2156 	 */
2157 	if (siz == 0) {
2158 		vput(vp);
2159 		if (nd->nd_flag & ND_NFSV2) {
2160 			NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
2161 		} else {
2162 			nfsrv_postopattr(nd, getret, &at);
2163 			NFSM_BUILD(tl, u_int32_t *, 4 * NFSX_UNSIGNED);
2164 			txdr_hyper(at.na_filerev, tl);
2165 			tl += 2;
2166 		}
2167 		*tl++ = newnfs_false;
2168 		*tl = newnfs_true;
2169 		free(rbuf, M_TEMP);
2170 		free(cookies, M_TEMP);
2171 		goto out;
2172 	}
2173 
2174 	/*
2175 	 * Check for degenerate cases of nothing useful read.
2176 	 * If so go try again
2177 	 */
2178 	cpos = rbuf;
2179 	cend = rbuf + siz;
2180 	dp = (struct dirent *)cpos;
2181 	cookiep = cookies;
2182 
2183 	/*
2184 	 * For some reason FreeBSD's ufs_readdir() chooses to back the
2185 	 * directory offset up to a block boundary, so it is necessary to
2186 	 * skip over the records that precede the requested offset. This
2187 	 * requires the assumption that file offset cookies monotonically
2188 	 * increase.
2189 	 */
2190 	while (cpos < cend && ncookies > 0 &&
2191 	    (dp->d_fileno == 0 || dp->d_type == DT_WHT ||
2192 	     (is_ufs == 1 && ((u_quad_t)(*cookiep)) <= toff))) {
2193 		cpos += dp->d_reclen;
2194 		dp = (struct dirent *)cpos;
2195 		cookiep++;
2196 		ncookies--;
2197 	}
2198 	if (cpos >= cend || ncookies == 0) {
2199 		siz = fullsiz;
2200 		toff = off;
2201 		goto again;
2202 	}
2203 	vput(vp);
2204 
2205 	/*
2206 	 * If cnt > MCLBYTES and the reply will not be saved, use
2207 	 * ext_pgs mbufs for TLS.
2208 	 * For NFSv4.0, we do not know for sure if the reply will
2209 	 * be saved, so do not use ext_pgs mbufs for NFSv4.0.
2210 	 */
2211 	if (cnt > MCLBYTES && siz > MCLBYTES &&
2212 	    (nd->nd_flag & (ND_TLS | ND_EXTPG | ND_SAVEREPLY)) == ND_TLS &&
2213 	    (nd->nd_flag & (ND_NFSV4 | ND_NFSV41)) != ND_NFSV4)
2214 		nd->nd_flag |= ND_EXTPG;
2215 
2216 	/*
2217 	 * dirlen is the size of the reply, including all XDR and must
2218 	 * not exceed cnt. For NFSv2, RFC1094 didn't clearly indicate
2219 	 * if the XDR should be included in "count", but to be safe, we do.
2220 	 * (Include the two booleans at the end of the reply in dirlen now.)
2221 	 */
2222 	if (nd->nd_flag & ND_NFSV3) {
2223 		nfsrv_postopattr(nd, getret, &at);
2224 		NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
2225 		txdr_hyper(at.na_filerev, tl);
2226 		dirlen = NFSX_V3POSTOPATTR + NFSX_VERF + 2 * NFSX_UNSIGNED;
2227 	} else {
2228 		dirlen = 2 * NFSX_UNSIGNED;
2229 	}
2230 
2231 	/* Loop through the records and build reply */
2232 	while (cpos < cend && ncookies > 0) {
2233 		nlen = dp->d_namlen;
2234 		if (dp->d_fileno != 0 && dp->d_type != DT_WHT &&
2235 			nlen <= NFS_MAXNAMLEN) {
2236 			if (nd->nd_flag & ND_NFSV3)
2237 				dirlen += (6*NFSX_UNSIGNED + NFSM_RNDUP(nlen));
2238 			else
2239 				dirlen += (4*NFSX_UNSIGNED + NFSM_RNDUP(nlen));
2240 			if (dirlen > cnt) {
2241 				eofflag = 0;
2242 				break;
2243 			}
2244 
2245 			/*
2246 			 * Build the directory record xdr from
2247 			 * the dirent entry.
2248 			 */
2249 			if (nd->nd_flag & ND_NFSV3) {
2250 				NFSM_BUILD(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
2251 				*tl++ = newnfs_true;
2252 				*tl++ = 0;
2253 			} else {
2254 				NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
2255 				*tl++ = newnfs_true;
2256 			}
2257 			*tl = txdr_unsigned(dp->d_fileno);
2258 			(void) nfsm_strtom(nd, dp->d_name, nlen);
2259 			if (nd->nd_flag & ND_NFSV3) {
2260 				NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
2261 				txdr_hyper(*cookiep, tl);
2262 			} else {
2263 				NFSM_BUILD(tl, u_int32_t *, NFSX_UNSIGNED);
2264 				*tl = txdr_unsigned(*cookiep);
2265 			}
2266 		}
2267 		cpos += dp->d_reclen;
2268 		dp = (struct dirent *)cpos;
2269 		cookiep++;
2270 		ncookies--;
2271 	}
2272 	if (cpos < cend)
2273 		eofflag = 0;
2274 	NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
2275 	*tl++ = newnfs_false;
2276 	if (eofflag)
2277 		*tl = newnfs_true;
2278 	else
2279 		*tl = newnfs_false;
2280 	free(rbuf, M_TEMP);
2281 	free(cookies, M_TEMP);
2282 
2283 out:
2284 	NFSEXITCODE2(0, nd);
2285 	return (0);
2286 nfsmout:
2287 	vput(vp);
2288 	NFSEXITCODE2(error, nd);
2289 	return (error);
2290 }
2291 
2292 /*
2293  * Readdirplus for V3 and Readdir for V4.
2294  */
2295 int
2296 nfsrvd_readdirplus(struct nfsrv_descript *nd, int isdgram,
2297     struct vnode *vp, struct nfsexstuff *exp)
2298 {
2299 	struct dirent *dp;
2300 	u_int32_t *tl;
2301 	int dirlen;
2302 	char *cpos, *cend, *rbuf;
2303 	struct vnode *nvp;
2304 	fhandle_t nfh;
2305 	struct nfsvattr nva, at, *nvap = &nva;
2306 	struct mbuf *mb0, *mb1;
2307 	struct nfsreferral *refp;
2308 	int nlen, r, error = 0, getret = 1, usevget = 1;
2309 	int siz, cnt, fullsiz, eofflag, ncookies, entrycnt;
2310 	caddr_t bpos0, bpos1;
2311 	u_int64_t off, toff, verf __unused;
2312 	uint64_t *cookies = NULL, *cookiep;
2313 	nfsattrbit_t attrbits, rderrbits, savbits;
2314 	struct uio io;
2315 	struct iovec iv;
2316 	struct componentname cn;
2317 	int at_root, is_ufs, is_zfs, needs_unbusy, supports_nfsv4acls;
2318 	struct mount *mp, *new_mp;
2319 	uint64_t mounted_on_fileno;
2320 	struct thread *p = curthread;
2321 	int bextpg0, bextpg1, bextpgsiz0, bextpgsiz1;
2322 
2323 	if (nd->nd_repstat) {
2324 		nfsrv_postopattr(nd, getret, &at);
2325 		goto out;
2326 	}
2327 	NFSM_DISSECT(tl, u_int32_t *, 6 * NFSX_UNSIGNED);
2328 	off = fxdr_hyper(tl);
2329 	toff = off;
2330 	tl += 2;
2331 	verf = fxdr_hyper(tl);
2332 	tl += 2;
2333 	siz = fxdr_unsigned(int, *tl++);
2334 	cnt = fxdr_unsigned(int, *tl);
2335 
2336 	/*
2337 	 * Use the server's maximum data transfer size as the upper bound
2338 	 * on reply datalen.
2339 	 */
2340 	if (cnt > NFS_SRVMAXDATA(nd) || cnt < 0)
2341 		cnt = NFS_SRVMAXDATA(nd);
2342 
2343 	/*
2344 	 * siz is a "hint" of how much directory information (name, fileid,
2345 	 * cookie) should be in the reply. At least one client "hints" 0,
2346 	 * so I set it to cnt for that case. I also round it up to the
2347 	 * next multiple of DIRBLKSIZ.
2348 	 * Since the size of a Readdirplus directory entry reply will always
2349 	 * be greater than a directory entry returned by VOP_READDIR(), it
2350 	 * does not make sense to read more than NFS_SRVMAXDATA() via
2351 	 * VOP_READDIR().
2352 	 */
2353 	if (siz <= 0)
2354 		siz = cnt;
2355 	else if (siz > NFS_SRVMAXDATA(nd))
2356 		siz = NFS_SRVMAXDATA(nd);
2357 	siz = ((siz + DIRBLKSIZ - 1) & ~(DIRBLKSIZ - 1));
2358 
2359 	if (nd->nd_flag & ND_NFSV4) {
2360 		error = nfsrv_getattrbits(nd, &attrbits, NULL, NULL);
2361 		if (error)
2362 			goto nfsmout;
2363 		NFSSET_ATTRBIT(&savbits, &attrbits);
2364 		NFSCLRNOTFILLABLE_ATTRBIT(&attrbits, nd);
2365 		NFSZERO_ATTRBIT(&rderrbits);
2366 		NFSSETBIT_ATTRBIT(&rderrbits, NFSATTRBIT_RDATTRERROR);
2367 	} else {
2368 		NFSZERO_ATTRBIT(&attrbits);
2369 	}
2370 	fullsiz = siz;
2371 	nd->nd_repstat = getret = nfsvno_getattr(vp, &at, nd, p, 1, NULL);
2372 #if 0
2373 	if (!nd->nd_repstat) {
2374 	    if (off && verf != at.na_filerev) {
2375 		/*
2376 		 * va_filerev is not sufficient as a cookie verifier,
2377 		 * since it is not supposed to change when entries are
2378 		 * removed/added unless that offset cookies returned to
2379 		 * the client are no longer valid.
2380 		 */
2381 		if (nd->nd_flag & ND_NFSV4) {
2382 			nd->nd_repstat = NFSERR_NOTSAME;
2383 		} else {
2384 			nd->nd_repstat = NFSERR_BAD_COOKIE;
2385 		}
2386 	    }
2387 	}
2388 #endif
2389 	if (!nd->nd_repstat && vp->v_type != VDIR)
2390 		nd->nd_repstat = NFSERR_NOTDIR;
2391 	if (!nd->nd_repstat && cnt == 0)
2392 		nd->nd_repstat = NFSERR_TOOSMALL;
2393 	if (!nd->nd_repstat)
2394 		nd->nd_repstat = nfsvno_accchk(vp, VEXEC,
2395 		    nd->nd_cred, exp, p, NFSACCCHK_NOOVERRIDE,
2396 		    NFSACCCHK_VPISLOCKED, NULL);
2397 	if (nd->nd_repstat) {
2398 		vput(vp);
2399 		if (nd->nd_flag & ND_NFSV3)
2400 			nfsrv_postopattr(nd, getret, &at);
2401 		goto out;
2402 	}
2403 	is_ufs = strcmp(vp->v_mount->mnt_vfc->vfc_name, "ufs") == 0;
2404 	is_zfs = strcmp(vp->v_mount->mnt_vfc->vfc_name, "zfs") == 0;
2405 
2406 	rbuf = malloc(siz, M_TEMP, M_WAITOK);
2407 again:
2408 	eofflag = 0;
2409 	if (cookies) {
2410 		free(cookies, M_TEMP);
2411 		cookies = NULL;
2412 	}
2413 
2414 	iv.iov_base = rbuf;
2415 	iv.iov_len = siz;
2416 	io.uio_iov = &iv;
2417 	io.uio_iovcnt = 1;
2418 	io.uio_offset = (off_t)off;
2419 	io.uio_resid = siz;
2420 	io.uio_segflg = UIO_SYSSPACE;
2421 	io.uio_rw = UIO_READ;
2422 	io.uio_td = NULL;
2423 	nd->nd_repstat = VOP_READDIR(vp, &io, nd->nd_cred, &eofflag, &ncookies,
2424 	    &cookies);
2425 	off = (u_int64_t)io.uio_offset;
2426 	if (io.uio_resid)
2427 		siz -= io.uio_resid;
2428 
2429 	getret = nfsvno_getattr(vp, &at, nd, p, 1, NULL);
2430 
2431 	if (!cookies && !nd->nd_repstat)
2432 		nd->nd_repstat = NFSERR_PERM;
2433 	if (!nd->nd_repstat)
2434 		nd->nd_repstat = getret;
2435 	if (nd->nd_repstat) {
2436 		vput(vp);
2437 		if (cookies)
2438 			free(cookies, M_TEMP);
2439 		free(rbuf, M_TEMP);
2440 		if (nd->nd_flag & ND_NFSV3)
2441 			nfsrv_postopattr(nd, getret, &at);
2442 		goto out;
2443 	}
2444 	/*
2445 	 * If nothing read, return eof
2446 	 * rpc reply
2447 	 */
2448 	if (siz == 0) {
2449 		vput(vp);
2450 		if (nd->nd_flag & ND_NFSV3)
2451 			nfsrv_postopattr(nd, getret, &at);
2452 		NFSM_BUILD(tl, u_int32_t *, 4 * NFSX_UNSIGNED);
2453 		txdr_hyper(at.na_filerev, tl);
2454 		tl += 2;
2455 		*tl++ = newnfs_false;
2456 		*tl = newnfs_true;
2457 		free(cookies, M_TEMP);
2458 		free(rbuf, M_TEMP);
2459 		goto out;
2460 	}
2461 
2462 	/*
2463 	 * Check for degenerate cases of nothing useful read.
2464 	 * If so go try again
2465 	 */
2466 	cpos = rbuf;
2467 	cend = rbuf + siz;
2468 	dp = (struct dirent *)cpos;
2469 	cookiep = cookies;
2470 
2471 	/*
2472 	 * For some reason FreeBSD's ufs_readdir() chooses to back the
2473 	 * directory offset up to a block boundary, so it is necessary to
2474 	 * skip over the records that precede the requested offset. This
2475 	 * requires the assumption that file offset cookies monotonically
2476 	 * increase.
2477 	 */
2478 	while (cpos < cend && ncookies > 0 &&
2479 	  (dp->d_fileno == 0 || dp->d_type == DT_WHT ||
2480 	   (is_ufs == 1 && ((u_quad_t)(*cookiep)) <= toff) ||
2481 	   ((nd->nd_flag & ND_NFSV4) &&
2482 	    ((dp->d_namlen == 1 && dp->d_name[0] == '.') ||
2483 	     (dp->d_namlen==2 && dp->d_name[0]=='.' && dp->d_name[1]=='.'))))) {
2484 		cpos += dp->d_reclen;
2485 		dp = (struct dirent *)cpos;
2486 		cookiep++;
2487 		ncookies--;
2488 	}
2489 	if (cpos >= cend || ncookies == 0) {
2490 		siz = fullsiz;
2491 		toff = off;
2492 		goto again;
2493 	}
2494 
2495 	/*
2496 	 * Busy the file system so that the mount point won't go away
2497 	 * and, as such, VFS_VGET() can be used safely.
2498 	 */
2499 	mp = vp->v_mount;
2500 	vfs_ref(mp);
2501 	NFSVOPUNLOCK(vp);
2502 	nd->nd_repstat = vfs_busy(mp, 0);
2503 	vfs_rel(mp);
2504 	if (nd->nd_repstat != 0) {
2505 		vrele(vp);
2506 		free(cookies, M_TEMP);
2507 		free(rbuf, M_TEMP);
2508 		if (nd->nd_flag & ND_NFSV3)
2509 			nfsrv_postopattr(nd, getret, &at);
2510 		goto out;
2511 	}
2512 
2513 	/*
2514 	 * Check to see if entries in this directory can be safely acquired
2515 	 * via VFS_VGET() or if a switch to VOP_LOOKUP() is required.
2516 	 * ZFS snapshot directories need VOP_LOOKUP(), so that any
2517 	 * automount of the snapshot directory that is required will
2518 	 * be done.
2519 	 * This needs to be done here for NFSv4, since NFSv4 never does
2520 	 * a VFS_VGET() for "." or "..".
2521 	 */
2522 	if (is_zfs == 1) {
2523 		r = VFS_VGET(mp, at.na_fileid, LK_SHARED, &nvp);
2524 		if (r == EOPNOTSUPP) {
2525 			usevget = 0;
2526 			cn.cn_nameiop = LOOKUP;
2527 			cn.cn_lkflags = LK_SHARED | LK_RETRY;
2528 			cn.cn_cred = nd->nd_cred;
2529 		} else if (r == 0)
2530 			vput(nvp);
2531 	}
2532 
2533 	/*
2534 	 * If the reply is likely to exceed MCLBYTES and the reply will
2535 	 * not be saved, use ext_pgs mbufs for TLS.
2536 	 * It is difficult to predict how large each entry will be and
2537 	 * how many entries have been read, so just assume the directory
2538 	 * entries grow by a factor of 4 when attributes are included.
2539 	 * For NFSv4.0, we do not know for sure if the reply will
2540 	 * be saved, so do not use ext_pgs mbufs for NFSv4.0.
2541 	 */
2542 	if (cnt > MCLBYTES && siz > MCLBYTES / 4 &&
2543 	    (nd->nd_flag & (ND_TLS | ND_EXTPG | ND_SAVEREPLY)) == ND_TLS &&
2544 	    (nd->nd_flag & (ND_NFSV4 | ND_NFSV41)) != ND_NFSV4)
2545 		nd->nd_flag |= ND_EXTPG;
2546 
2547 	/*
2548 	 * Save this position, in case there is an error before one entry
2549 	 * is created.
2550 	 */
2551 	mb0 = nd->nd_mb;
2552 	bpos0 = nd->nd_bpos;
2553 	bextpg0 = nd->nd_bextpg;
2554 	bextpgsiz0 = nd->nd_bextpgsiz;
2555 
2556 	/*
2557 	 * Fill in the first part of the reply.
2558 	 * dirlen is the reply length in bytes and cannot exceed cnt.
2559 	 * (Include the two booleans at the end of the reply in dirlen now,
2560 	 *  so we recognize when we have exceeded cnt.)
2561 	 */
2562 	if (nd->nd_flag & ND_NFSV3) {
2563 		dirlen = NFSX_V3POSTOPATTR + NFSX_VERF + 2 * NFSX_UNSIGNED;
2564 		nfsrv_postopattr(nd, getret, &at);
2565 	} else {
2566 		dirlen = NFSX_VERF + 2 * NFSX_UNSIGNED;
2567 	}
2568 	NFSM_BUILD(tl, u_int32_t *, NFSX_VERF);
2569 	txdr_hyper(at.na_filerev, tl);
2570 
2571 	/*
2572 	 * Save this position, in case there is an empty reply needed.
2573 	 */
2574 	mb1 = nd->nd_mb;
2575 	bpos1 = nd->nd_bpos;
2576 	bextpg1 = nd->nd_bextpg;
2577 	bextpgsiz1 = nd->nd_bextpgsiz;
2578 
2579 	/* Loop through the records and build reply */
2580 	entrycnt = 0;
2581 	while (cpos < cend && ncookies > 0 && dirlen < cnt) {
2582 		nlen = dp->d_namlen;
2583 		if (dp->d_fileno != 0 && dp->d_type != DT_WHT &&
2584 		    nlen <= NFS_MAXNAMLEN &&
2585 		    ((nd->nd_flag & ND_NFSV3) || nlen > 2 ||
2586 		     (nlen==2 && (dp->d_name[0]!='.' || dp->d_name[1]!='.'))
2587 		      || (nlen == 1 && dp->d_name[0] != '.'))) {
2588 			/*
2589 			 * Save the current position in the reply, in case
2590 			 * this entry exceeds cnt.
2591 			 */
2592 			mb1 = nd->nd_mb;
2593 			bpos1 = nd->nd_bpos;
2594 			bextpg1 = nd->nd_bextpg;
2595 			bextpgsiz1 = nd->nd_bextpgsiz;
2596 
2597 			/*
2598 			 * For readdir_and_lookup get the vnode using
2599 			 * the file number.
2600 			 */
2601 			nvp = NULL;
2602 			refp = NULL;
2603 			r = 0;
2604 			at_root = 0;
2605 			needs_unbusy = 0;
2606 			new_mp = mp;
2607 			mounted_on_fileno = (uint64_t)dp->d_fileno;
2608 			if ((nd->nd_flag & ND_NFSV3) ||
2609 			    NFSNONZERO_ATTRBIT(&savbits)) {
2610 				if (nd->nd_flag & ND_NFSV4)
2611 					refp = nfsv4root_getreferral(NULL,
2612 					    vp, dp->d_fileno);
2613 				if (refp == NULL) {
2614 					if (usevget)
2615 						r = VFS_VGET(mp, dp->d_fileno,
2616 						    LK_SHARED, &nvp);
2617 					else
2618 						r = EOPNOTSUPP;
2619 					if (r == EOPNOTSUPP) {
2620 						if (usevget) {
2621 							usevget = 0;
2622 							cn.cn_nameiop = LOOKUP;
2623 							cn.cn_lkflags =
2624 							    LK_SHARED |
2625 							    LK_RETRY;
2626 							cn.cn_cred =
2627 							    nd->nd_cred;
2628 						}
2629 						cn.cn_nameptr = dp->d_name;
2630 						cn.cn_namelen = nlen;
2631 						cn.cn_flags = ISLASTCN |
2632 						    NOFOLLOW | LOCKLEAF;
2633 						if (nlen == 2 &&
2634 						    dp->d_name[0] == '.' &&
2635 						    dp->d_name[1] == '.')
2636 							cn.cn_flags |=
2637 							    ISDOTDOT;
2638 						if (NFSVOPLOCK(vp, LK_SHARED)
2639 						    != 0) {
2640 							nd->nd_repstat = EPERM;
2641 							break;
2642 						}
2643 						if ((vp->v_vflag & VV_ROOT) != 0
2644 						    && (cn.cn_flags & ISDOTDOT)
2645 						    != 0) {
2646 							vref(vp);
2647 							nvp = vp;
2648 							r = 0;
2649 						} else {
2650 							r = VOP_LOOKUP(vp, &nvp,
2651 							    &cn);
2652 							if (vp != nvp)
2653 								NFSVOPUNLOCK(vp);
2654 						}
2655 					}
2656 
2657 					/*
2658 					 * For NFSv4, check to see if nvp is
2659 					 * a mount point and get the mount
2660 					 * point vnode, as required.
2661 					 */
2662 					if (r == 0 &&
2663 					    nfsrv_enable_crossmntpt != 0 &&
2664 					    (nd->nd_flag & ND_NFSV4) != 0 &&
2665 					    nvp->v_type == VDIR &&
2666 					    nvp->v_mountedhere != NULL) {
2667 						new_mp = nvp->v_mountedhere;
2668 						r = vfs_busy(new_mp, 0);
2669 						vput(nvp);
2670 						nvp = NULL;
2671 						if (r == 0) {
2672 							r = VFS_ROOT(new_mp,
2673 							    LK_SHARED, &nvp);
2674 							needs_unbusy = 1;
2675 							if (r == 0)
2676 								at_root = 1;
2677 						}
2678 					}
2679 				}
2680 
2681 				/*
2682 				 * If we failed to look up the entry, then it
2683 				 * has become invalid, most likely removed.
2684 				 */
2685 				if (r != 0) {
2686 					if (needs_unbusy)
2687 						vfs_unbusy(new_mp);
2688 					goto invalid;
2689 				}
2690 				KASSERT(refp != NULL || nvp != NULL,
2691 				    ("%s: undetected lookup error", __func__));
2692 
2693 				if (refp == NULL &&
2694 				    ((nd->nd_flag & ND_NFSV3) ||
2695 				     NFSNONZERO_ATTRBIT(&attrbits))) {
2696 					r = nfsvno_getfh(nvp, &nfh, p);
2697 					if (!r)
2698 					    r = nfsvno_getattr(nvp, nvap, nd, p,
2699 						1, &attrbits);
2700 					if (r == 0 && is_zfs == 1 &&
2701 					    nfsrv_enable_crossmntpt != 0 &&
2702 					    (nd->nd_flag & ND_NFSV4) != 0 &&
2703 					    nvp->v_type == VDIR &&
2704 					    vp->v_mount != nvp->v_mount) {
2705 					    /*
2706 					     * For a ZFS snapshot, there is a
2707 					     * pseudo mount that does not set
2708 					     * v_mountedhere, so it needs to
2709 					     * be detected via a different
2710 					     * mount structure.
2711 					     */
2712 					    at_root = 1;
2713 					    if (new_mp == mp)
2714 						new_mp = nvp->v_mount;
2715 					}
2716 				}
2717 
2718 				/*
2719 				 * If we failed to get attributes of the entry,
2720 				 * then just skip it for NFSv3 (the traditional
2721 				 * behavior in the old NFS server).
2722 				 * For NFSv4 the behavior is controlled by
2723 				 * RDATTRERROR: we either ignore the error or
2724 				 * fail the request.
2725 				 * Note that RDATTRERROR is never set for NFSv3.
2726 				 */
2727 				if (r != 0) {
2728 					if (!NFSISSET_ATTRBIT(&attrbits,
2729 					    NFSATTRBIT_RDATTRERROR)) {
2730 						vput(nvp);
2731 						if (needs_unbusy != 0)
2732 							vfs_unbusy(new_mp);
2733 						if ((nd->nd_flag & ND_NFSV3))
2734 							goto invalid;
2735 						nd->nd_repstat = r;
2736 						break;
2737 					}
2738 				}
2739 			}
2740 
2741 			/*
2742 			 * Build the directory record xdr
2743 			 */
2744 			if (nd->nd_flag & ND_NFSV3) {
2745 				NFSM_BUILD(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
2746 				*tl++ = newnfs_true;
2747 				*tl++ = 0;
2748 				*tl = txdr_unsigned(dp->d_fileno);
2749 				dirlen += nfsm_strtom(nd, dp->d_name, nlen);
2750 				NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
2751 				txdr_hyper(*cookiep, tl);
2752 				nfsrv_postopattr(nd, 0, nvap);
2753 				dirlen += nfsm_fhtom(nd,(u_int8_t *)&nfh,0,1);
2754 				dirlen += (5*NFSX_UNSIGNED+NFSX_V3POSTOPATTR);
2755 				if (nvp != NULL)
2756 					vput(nvp);
2757 			} else {
2758 				NFSM_BUILD(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
2759 				*tl++ = newnfs_true;
2760 				txdr_hyper(*cookiep, tl);
2761 				dirlen += nfsm_strtom(nd, dp->d_name, nlen);
2762 				if (nvp != NULL) {
2763 					supports_nfsv4acls =
2764 					    nfs_supportsnfsv4acls(nvp);
2765 					NFSVOPUNLOCK(nvp);
2766 				} else
2767 					supports_nfsv4acls = 0;
2768 				if (refp != NULL) {
2769 					dirlen += nfsrv_putreferralattr(nd,
2770 					    &savbits, refp, 0,
2771 					    &nd->nd_repstat);
2772 					if (nd->nd_repstat) {
2773 						if (nvp != NULL)
2774 							vrele(nvp);
2775 						if (needs_unbusy != 0)
2776 							vfs_unbusy(new_mp);
2777 						break;
2778 					}
2779 				} else if (r) {
2780 					dirlen += nfsvno_fillattr(nd, new_mp,
2781 					    nvp, nvap, &nfh, r, &rderrbits,
2782 					    nd->nd_cred, p, isdgram, 0,
2783 					    supports_nfsv4acls, at_root,
2784 					    mounted_on_fileno);
2785 				} else {
2786 					dirlen += nfsvno_fillattr(nd, new_mp,
2787 					    nvp, nvap, &nfh, r, &attrbits,
2788 					    nd->nd_cred, p, isdgram, 0,
2789 					    supports_nfsv4acls, at_root,
2790 					    mounted_on_fileno);
2791 				}
2792 				if (nvp != NULL)
2793 					vrele(nvp);
2794 				dirlen += (3 * NFSX_UNSIGNED);
2795 			}
2796 			if (needs_unbusy != 0)
2797 				vfs_unbusy(new_mp);
2798 			if (dirlen <= cnt)
2799 				entrycnt++;
2800 		}
2801 invalid:
2802 		cpos += dp->d_reclen;
2803 		dp = (struct dirent *)cpos;
2804 		cookiep++;
2805 		ncookies--;
2806 	}
2807 	vrele(vp);
2808 	vfs_unbusy(mp);
2809 
2810 	/*
2811 	 * If dirlen > cnt, we must strip off the last entry. If that
2812 	 * results in an empty reply, report NFSERR_TOOSMALL.
2813 	 */
2814 	if (dirlen > cnt || nd->nd_repstat) {
2815 		if (!nd->nd_repstat && entrycnt == 0)
2816 			nd->nd_repstat = NFSERR_TOOSMALL;
2817 		if (nd->nd_repstat) {
2818 			nfsm_trimtrailing(nd, mb0, bpos0, bextpg0, bextpgsiz0);
2819 			if (nd->nd_flag & ND_NFSV3)
2820 				nfsrv_postopattr(nd, getret, &at);
2821 		} else
2822 			nfsm_trimtrailing(nd, mb1, bpos1, bextpg1, bextpgsiz1);
2823 		eofflag = 0;
2824 	} else if (cpos < cend)
2825 		eofflag = 0;
2826 	if (!nd->nd_repstat) {
2827 		NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
2828 		*tl++ = newnfs_false;
2829 		if (eofflag)
2830 			*tl = newnfs_true;
2831 		else
2832 			*tl = newnfs_false;
2833 	}
2834 	free(cookies, M_TEMP);
2835 	free(rbuf, M_TEMP);
2836 
2837 out:
2838 	NFSEXITCODE2(0, nd);
2839 	return (0);
2840 nfsmout:
2841 	vput(vp);
2842 	NFSEXITCODE2(error, nd);
2843 	return (error);
2844 }
2845 
2846 /*
2847  * Get the settable attributes out of the mbuf list.
2848  * (Return 0 or EBADRPC)
2849  */
2850 int
2851 nfsrv_sattr(struct nfsrv_descript *nd, vnode_t vp, struct nfsvattr *nvap,
2852     nfsattrbit_t *attrbitp, NFSACL_T *aclp, struct thread *p)
2853 {
2854 	u_int32_t *tl;
2855 	struct nfsv2_sattr *sp;
2856 	int error = 0, toclient = 0;
2857 
2858 	switch (nd->nd_flag & (ND_NFSV2 | ND_NFSV3 | ND_NFSV4)) {
2859 	case ND_NFSV2:
2860 		NFSM_DISSECT(sp, struct nfsv2_sattr *, NFSX_V2SATTR);
2861 		/*
2862 		 * Some old clients didn't fill in the high order 16bits.
2863 		 * --> check the low order 2 bytes for 0xffff
2864 		 */
2865 		if ((fxdr_unsigned(int, sp->sa_mode) & 0xffff) != 0xffff)
2866 			nvap->na_mode = nfstov_mode(sp->sa_mode);
2867 		if (sp->sa_uid != newnfs_xdrneg1)
2868 			nvap->na_uid = fxdr_unsigned(uid_t, sp->sa_uid);
2869 		if (sp->sa_gid != newnfs_xdrneg1)
2870 			nvap->na_gid = fxdr_unsigned(gid_t, sp->sa_gid);
2871 		if (sp->sa_size != newnfs_xdrneg1)
2872 			nvap->na_size = fxdr_unsigned(u_quad_t, sp->sa_size);
2873 		if (sp->sa_atime.nfsv2_sec != newnfs_xdrneg1) {
2874 #ifdef notyet
2875 			fxdr_nfsv2time(&sp->sa_atime, &nvap->na_atime);
2876 #else
2877 			nvap->na_atime.tv_sec =
2878 				fxdr_unsigned(u_int32_t,sp->sa_atime.nfsv2_sec);
2879 			nvap->na_atime.tv_nsec = 0;
2880 #endif
2881 		}
2882 		if (sp->sa_mtime.nfsv2_sec != newnfs_xdrneg1)
2883 			fxdr_nfsv2time(&sp->sa_mtime, &nvap->na_mtime);
2884 		break;
2885 	case ND_NFSV3:
2886 		NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2887 		if (*tl == newnfs_true) {
2888 			NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2889 			nvap->na_mode = nfstov_mode(*tl);
2890 		}
2891 		NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2892 		if (*tl == newnfs_true) {
2893 			NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2894 			nvap->na_uid = fxdr_unsigned(uid_t, *tl);
2895 		}
2896 		NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2897 		if (*tl == newnfs_true) {
2898 			NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2899 			nvap->na_gid = fxdr_unsigned(gid_t, *tl);
2900 		}
2901 		NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2902 		if (*tl == newnfs_true) {
2903 			NFSM_DISSECT(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
2904 			nvap->na_size = fxdr_hyper(tl);
2905 		}
2906 		NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2907 		switch (fxdr_unsigned(int, *tl)) {
2908 		case NFSV3SATTRTIME_TOCLIENT:
2909 			NFSM_DISSECT(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
2910 			fxdr_nfsv3time(tl, &nvap->na_atime);
2911 			toclient = 1;
2912 			break;
2913 		case NFSV3SATTRTIME_TOSERVER:
2914 			vfs_timestamp(&nvap->na_atime);
2915 			nvap->na_vaflags |= VA_UTIMES_NULL;
2916 			break;
2917 		}
2918 		NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2919 		switch (fxdr_unsigned(int, *tl)) {
2920 		case NFSV3SATTRTIME_TOCLIENT:
2921 			NFSM_DISSECT(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
2922 			fxdr_nfsv3time(tl, &nvap->na_mtime);
2923 			nvap->na_vaflags &= ~VA_UTIMES_NULL;
2924 			break;
2925 		case NFSV3SATTRTIME_TOSERVER:
2926 			vfs_timestamp(&nvap->na_mtime);
2927 			if (!toclient)
2928 				nvap->na_vaflags |= VA_UTIMES_NULL;
2929 			break;
2930 		}
2931 		break;
2932 	case ND_NFSV4:
2933 		error = nfsv4_sattr(nd, vp, nvap, attrbitp, aclp, p);
2934 	}
2935 nfsmout:
2936 	NFSEXITCODE2(error, nd);
2937 	return (error);
2938 }
2939 
2940 /*
2941  * Handle the setable attributes for V4.
2942  * Returns NFSERR_BADXDR if it can't be parsed, 0 otherwise.
2943  */
2944 int
2945 nfsv4_sattr(struct nfsrv_descript *nd, vnode_t vp, struct nfsvattr *nvap,
2946     nfsattrbit_t *attrbitp, NFSACL_T *aclp, struct thread *p)
2947 {
2948 	u_int32_t *tl;
2949 	int attrsum = 0;
2950 	int i, j;
2951 	int error, attrsize, bitpos, aclsize, aceerr, retnotsup = 0;
2952 	int moderet, toclient = 0;
2953 	u_char *cp, namestr[NFSV4_SMALLSTR + 1];
2954 	uid_t uid;
2955 	gid_t gid;
2956 	u_short mode, mask;		/* Same type as va_mode. */
2957 	struct vattr va;
2958 
2959 	error = nfsrv_getattrbits(nd, attrbitp, NULL, &retnotsup);
2960 	if (error)
2961 		goto nfsmout;
2962 	NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2963 	attrsize = fxdr_unsigned(int, *tl);
2964 
2965 	/*
2966 	 * Loop around getting the setable attributes. If an unsupported
2967 	 * one is found, set nd_repstat == NFSERR_ATTRNOTSUPP and return.
2968 	 */
2969 	if (retnotsup) {
2970 		nd->nd_repstat = NFSERR_ATTRNOTSUPP;
2971 		bitpos = NFSATTRBIT_MAX;
2972 	} else {
2973 		bitpos = 0;
2974 	}
2975 	moderet = 0;
2976 	for (; bitpos < NFSATTRBIT_MAX; bitpos++) {
2977 	    if (attrsum > attrsize) {
2978 		error = NFSERR_BADXDR;
2979 		goto nfsmout;
2980 	    }
2981 	    if (NFSISSET_ATTRBIT(attrbitp, bitpos))
2982 		switch (bitpos) {
2983 		case NFSATTRBIT_SIZE:
2984 			NFSM_DISSECT(tl, u_int32_t *, NFSX_HYPER);
2985                      if (vp != NULL && vp->v_type != VREG) {
2986                             error = (vp->v_type == VDIR) ? NFSERR_ISDIR :
2987                                 NFSERR_INVAL;
2988                             goto nfsmout;
2989 			}
2990 			nvap->na_size = fxdr_hyper(tl);
2991 			attrsum += NFSX_HYPER;
2992 			break;
2993 		case NFSATTRBIT_ACL:
2994 			error = nfsrv_dissectacl(nd, aclp, &aceerr, &aclsize,
2995 			    p);
2996 			if (error)
2997 				goto nfsmout;
2998 			if (aceerr && !nd->nd_repstat)
2999 				nd->nd_repstat = aceerr;
3000 			attrsum += aclsize;
3001 			break;
3002 		case NFSATTRBIT_ARCHIVE:
3003 			NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
3004 			if (!nd->nd_repstat)
3005 				nd->nd_repstat = NFSERR_ATTRNOTSUPP;
3006 			attrsum += NFSX_UNSIGNED;
3007 			break;
3008 		case NFSATTRBIT_HIDDEN:
3009 			NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
3010 			if (!nd->nd_repstat)
3011 				nd->nd_repstat = NFSERR_ATTRNOTSUPP;
3012 			attrsum += NFSX_UNSIGNED;
3013 			break;
3014 		case NFSATTRBIT_MIMETYPE:
3015 			NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
3016 			i = fxdr_unsigned(int, *tl);
3017 			error = nfsm_advance(nd, NFSM_RNDUP(i), -1);
3018 			if (error)
3019 				goto nfsmout;
3020 			if (!nd->nd_repstat)
3021 				nd->nd_repstat = NFSERR_ATTRNOTSUPP;
3022 			attrsum += (NFSX_UNSIGNED + NFSM_RNDUP(i));
3023 			break;
3024 		case NFSATTRBIT_MODE:
3025 			moderet = NFSERR_INVAL;	/* Can't do MODESETMASKED. */
3026 			NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
3027 			nvap->na_mode = nfstov_mode(*tl);
3028 			attrsum += NFSX_UNSIGNED;
3029 			break;
3030 		case NFSATTRBIT_OWNER:
3031 			NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
3032 			j = fxdr_unsigned(int, *tl);
3033 			if (j < 0) {
3034 				error = NFSERR_BADXDR;
3035 				goto nfsmout;
3036 			}
3037 			if (j > NFSV4_SMALLSTR)
3038 				cp = malloc(j + 1, M_NFSSTRING, M_WAITOK);
3039 			else
3040 				cp = namestr;
3041 			error = nfsrv_mtostr(nd, cp, j);
3042 			if (error) {
3043 				if (j > NFSV4_SMALLSTR)
3044 					free(cp, M_NFSSTRING);
3045 				goto nfsmout;
3046 			}
3047 			if (!nd->nd_repstat) {
3048 				nd->nd_repstat = nfsv4_strtouid(nd, cp, j,
3049 				    &uid);
3050 				if (!nd->nd_repstat)
3051 					nvap->na_uid = uid;
3052 			}
3053 			if (j > NFSV4_SMALLSTR)
3054 				free(cp, M_NFSSTRING);
3055 			attrsum += (NFSX_UNSIGNED + NFSM_RNDUP(j));
3056 			break;
3057 		case NFSATTRBIT_OWNERGROUP:
3058 			NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
3059 			j = fxdr_unsigned(int, *tl);
3060 			if (j < 0) {
3061 				error = NFSERR_BADXDR;
3062 				goto nfsmout;
3063 			}
3064 			if (j > NFSV4_SMALLSTR)
3065 				cp = malloc(j + 1, M_NFSSTRING, M_WAITOK);
3066 			else
3067 				cp = namestr;
3068 			error = nfsrv_mtostr(nd, cp, j);
3069 			if (error) {
3070 				if (j > NFSV4_SMALLSTR)
3071 					free(cp, M_NFSSTRING);
3072 				goto nfsmout;
3073 			}
3074 			if (!nd->nd_repstat) {
3075 				nd->nd_repstat = nfsv4_strtogid(nd, cp, j,
3076 				    &gid);
3077 				if (!nd->nd_repstat)
3078 					nvap->na_gid = gid;
3079 			}
3080 			if (j > NFSV4_SMALLSTR)
3081 				free(cp, M_NFSSTRING);
3082 			attrsum += (NFSX_UNSIGNED + NFSM_RNDUP(j));
3083 			break;
3084 		case NFSATTRBIT_SYSTEM:
3085 			NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
3086 			if (!nd->nd_repstat)
3087 				nd->nd_repstat = NFSERR_ATTRNOTSUPP;
3088 			attrsum += NFSX_UNSIGNED;
3089 			break;
3090 		case NFSATTRBIT_TIMEACCESSSET:
3091 			NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
3092 			attrsum += NFSX_UNSIGNED;
3093 			if (fxdr_unsigned(int, *tl)==NFSV4SATTRTIME_TOCLIENT) {
3094 			    NFSM_DISSECT(tl, u_int32_t *, NFSX_V4TIME);
3095 			    fxdr_nfsv4time(tl, &nvap->na_atime);
3096 			    toclient = 1;
3097 			    attrsum += NFSX_V4TIME;
3098 			} else {
3099 			    vfs_timestamp(&nvap->na_atime);
3100 			    nvap->na_vaflags |= VA_UTIMES_NULL;
3101 			}
3102 			break;
3103 		case NFSATTRBIT_TIMEBACKUP:
3104 			NFSM_DISSECT(tl, u_int32_t *, NFSX_V4TIME);
3105 			if (!nd->nd_repstat)
3106 				nd->nd_repstat = NFSERR_ATTRNOTSUPP;
3107 			attrsum += NFSX_V4TIME;
3108 			break;
3109 		case NFSATTRBIT_TIMECREATE:
3110 			NFSM_DISSECT(tl, u_int32_t *, NFSX_V4TIME);
3111 			fxdr_nfsv4time(tl, &nvap->na_btime);
3112 			attrsum += NFSX_V4TIME;
3113 			break;
3114 		case NFSATTRBIT_TIMEMODIFYSET:
3115 			NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
3116 			attrsum += NFSX_UNSIGNED;
3117 			if (fxdr_unsigned(int, *tl)==NFSV4SATTRTIME_TOCLIENT) {
3118 			    NFSM_DISSECT(tl, u_int32_t *, NFSX_V4TIME);
3119 			    fxdr_nfsv4time(tl, &nvap->na_mtime);
3120 			    nvap->na_vaflags &= ~VA_UTIMES_NULL;
3121 			    attrsum += NFSX_V4TIME;
3122 			} else {
3123 			    vfs_timestamp(&nvap->na_mtime);
3124 			    if (!toclient)
3125 				nvap->na_vaflags |= VA_UTIMES_NULL;
3126 			}
3127 			break;
3128 		case NFSATTRBIT_MODESETMASKED:
3129 			NFSM_DISSECT(tl, uint32_t *, 2 * NFSX_UNSIGNED);
3130 			mode = fxdr_unsigned(u_short, *tl++);
3131 			mask = fxdr_unsigned(u_short, *tl);
3132 			/*
3133 			 * vp == NULL implies an Open/Create operation.
3134 			 * This attribute can only be used for Setattr and
3135 			 * only for NFSv4.1 or higher.
3136 			 * If moderet != 0, a mode attribute has also been
3137 			 * specified and this attribute cannot be done in the
3138 			 * same Setattr operation.
3139 			 */
3140 			if ((nd->nd_flag & ND_NFSV41) == 0)
3141 				nd->nd_repstat = NFSERR_ATTRNOTSUPP;
3142 			else if ((mode & ~07777) != 0 || (mask & ~07777) != 0 ||
3143 			    vp == NULL)
3144 				nd->nd_repstat = NFSERR_INVAL;
3145 			else if (moderet == 0)
3146 				moderet = VOP_GETATTR(vp, &va, nd->nd_cred);
3147 			if (moderet == 0)
3148 				nvap->na_mode = (mode & mask) |
3149 				    (va.va_mode & ~mask);
3150 			else
3151 				nd->nd_repstat = moderet;
3152 			attrsum += 2 * NFSX_UNSIGNED;
3153 			break;
3154 		default:
3155 			nd->nd_repstat = NFSERR_ATTRNOTSUPP;
3156 			/*
3157 			 * set bitpos so we drop out of the loop.
3158 			 */
3159 			bitpos = NFSATTRBIT_MAX;
3160 			break;
3161 		}
3162 	}
3163 
3164 	/*
3165 	 * some clients pad the attrlist, so we need to skip over the
3166 	 * padding.
3167 	 */
3168 	if (attrsum > attrsize) {
3169 		error = NFSERR_BADXDR;
3170 	} else {
3171 		attrsize = NFSM_RNDUP(attrsize);
3172 		if (attrsum < attrsize)
3173 			error = nfsm_advance(nd, attrsize - attrsum, -1);
3174 	}
3175 nfsmout:
3176 	NFSEXITCODE2(error, nd);
3177 	return (error);
3178 }
3179 
3180 /*
3181  * Check/setup export credentials.
3182  */
3183 int
3184 nfsd_excred(struct nfsrv_descript *nd, struct nfsexstuff *exp,
3185     struct ucred *credanon, bool testsec)
3186 {
3187 	int error;
3188 
3189 	/*
3190 	 * Check/setup credentials.
3191 	 */
3192 	if (nd->nd_flag & ND_GSS)
3193 		exp->nes_exflag &= ~MNT_EXPORTANON;
3194 
3195 	/*
3196 	 * Check to see if the operation is allowed for this security flavor.
3197 	 */
3198 	error = 0;
3199 	if (testsec) {
3200 		error = nfsvno_testexp(nd, exp);
3201 		if (error != 0)
3202 			goto out;
3203 	}
3204 
3205 	/*
3206 	 * Check to see if the file system is exported V4 only.
3207 	 */
3208 	if (NFSVNO_EXV4ONLY(exp) && !(nd->nd_flag & ND_NFSV4)) {
3209 		error = NFSERR_PROGNOTV4;
3210 		goto out;
3211 	}
3212 
3213 	/*
3214 	 * Now, map the user credentials.
3215 	 * (Note that ND_AUTHNONE will only be set for an NFSv3
3216 	 *  Fsinfo RPC. If set for anything else, this code might need
3217 	 *  to change.)
3218 	 */
3219 	if (NFSVNO_EXPORTED(exp)) {
3220 		if (((nd->nd_flag & ND_GSS) == 0 && nd->nd_cred->cr_uid == 0) ||
3221 		     NFSVNO_EXPORTANON(exp) ||
3222 		     (nd->nd_flag & ND_AUTHNONE) != 0) {
3223 			nd->nd_cred->cr_uid = credanon->cr_uid;
3224 			nd->nd_cred->cr_gid = credanon->cr_gid;
3225 			crsetgroups(nd->nd_cred, credanon->cr_ngroups,
3226 			    credanon->cr_groups);
3227 		} else if ((nd->nd_flag & ND_GSS) == 0) {
3228 			/*
3229 			 * If using AUTH_SYS, call nfsrv_getgrpscred() to see
3230 			 * if there is a replacement credential with a group
3231 			 * list set up by "nfsuserd -manage-gids".
3232 			 * If there is no replacement, nfsrv_getgrpscred()
3233 			 * simply returns its argument.
3234 			 */
3235 			nd->nd_cred = nfsrv_getgrpscred(nd->nd_cred);
3236 		}
3237 	}
3238 
3239 out:
3240 	NFSEXITCODE2(error, nd);
3241 	return (error);
3242 }
3243 
3244 /*
3245  * Check exports.
3246  */
3247 int
3248 nfsvno_checkexp(struct mount *mp, struct sockaddr *nam, struct nfsexstuff *exp,
3249     struct ucred **credp)
3250 {
3251 	int error;
3252 
3253 	error = VFS_CHECKEXP(mp, nam, &exp->nes_exflag, credp,
3254 	    &exp->nes_numsecflavor, exp->nes_secflavors);
3255 	if (error) {
3256 		if (nfs_rootfhset) {
3257 			exp->nes_exflag = 0;
3258 			exp->nes_numsecflavor = 0;
3259 			error = 0;
3260 		}
3261 	} else if (exp->nes_numsecflavor < 1 || exp->nes_numsecflavor >
3262 	    MAXSECFLAVORS) {
3263 		printf("nfsvno_checkexp: numsecflavors out of range\n");
3264 		exp->nes_numsecflavor = 0;
3265 		error = EACCES;
3266 	}
3267 	NFSEXITCODE(error);
3268 	return (error);
3269 }
3270 
3271 /*
3272  * Get a vnode for a file handle and export stuff.
3273  */
3274 int
3275 nfsvno_fhtovp(struct mount *mp, fhandle_t *fhp, struct sockaddr *nam,
3276     int lktype, struct vnode **vpp, struct nfsexstuff *exp,
3277     struct ucred **credp)
3278 {
3279 	int error;
3280 
3281 	*credp = NULL;
3282 	exp->nes_numsecflavor = 0;
3283 	error = VFS_FHTOVP(mp, &fhp->fh_fid, lktype, vpp);
3284 	if (error != 0)
3285 		/* Make sure the server replies ESTALE to the client. */
3286 		error = ESTALE;
3287 	if (nam && !error) {
3288 		error = VFS_CHECKEXP(mp, nam, &exp->nes_exflag, credp,
3289 		    &exp->nes_numsecflavor, exp->nes_secflavors);
3290 		if (error) {
3291 			if (nfs_rootfhset) {
3292 				exp->nes_exflag = 0;
3293 				exp->nes_numsecflavor = 0;
3294 				error = 0;
3295 			} else {
3296 				vput(*vpp);
3297 			}
3298 		} else if (exp->nes_numsecflavor < 1 || exp->nes_numsecflavor >
3299 		    MAXSECFLAVORS) {
3300 			printf("nfsvno_fhtovp: numsecflavors out of range\n");
3301 			exp->nes_numsecflavor = 0;
3302 			error = EACCES;
3303 			vput(*vpp);
3304 		}
3305 	}
3306 	NFSEXITCODE(error);
3307 	return (error);
3308 }
3309 
3310 /*
3311  * nfsd_fhtovp() - convert a fh to a vnode ptr
3312  * 	- look up fsid in mount list (if not found ret error)
3313  *	- get vp and export rights by calling nfsvno_fhtovp()
3314  *	- if cred->cr_uid == 0 or MNT_EXPORTANON set it to credanon
3315  *	  for AUTH_SYS
3316  *	- if mpp != NULL, return the mount point so that it can
3317  *	  be used for vn_finished_write() by the caller
3318  */
3319 void
3320 nfsd_fhtovp(struct nfsrv_descript *nd, struct nfsrvfh *nfp, int lktype,
3321     struct vnode **vpp, struct nfsexstuff *exp,
3322     struct mount **mpp, int startwrite, int nextop)
3323 {
3324 	struct mount *mp, *mpw;
3325 	struct ucred *credanon;
3326 	fhandle_t *fhp;
3327 	int error;
3328 
3329 	if (mpp != NULL)
3330 		*mpp = NULL;
3331 	*vpp = NULL;
3332 	fhp = (fhandle_t *)nfp->nfsrvfh_data;
3333 	mp = vfs_busyfs(&fhp->fh_fsid);
3334 	if (mp == NULL) {
3335 		nd->nd_repstat = ESTALE;
3336 		goto out;
3337 	}
3338 
3339 	if (startwrite) {
3340 		mpw = mp;
3341 		error = vn_start_write(NULL, &mpw, V_WAIT);
3342 		if (error != 0) {
3343 			mpw = NULL;
3344 			vfs_unbusy(mp);
3345 			nd->nd_repstat = ESTALE;
3346 			goto out;
3347 		}
3348 		if (lktype == LK_SHARED && !(MNT_SHARED_WRITES(mp)))
3349 			lktype = LK_EXCLUSIVE;
3350 	} else
3351 		mpw = NULL;
3352 
3353 	nd->nd_repstat = nfsvno_fhtovp(mp, fhp, nd->nd_nam, lktype, vpp, exp,
3354 	    &credanon);
3355 	vfs_unbusy(mp);
3356 
3357 	/*
3358 	 * For NFSv4 without a pseudo root fs, unexported file handles
3359 	 * can be returned, so that Lookup works everywhere.
3360 	 */
3361 	if (!nd->nd_repstat && exp->nes_exflag == 0 &&
3362 	    !(nd->nd_flag & ND_NFSV4)) {
3363 		vput(*vpp);
3364 		*vpp = NULL;
3365 		nd->nd_repstat = EACCES;
3366 	}
3367 
3368 	/*
3369 	 * Personally, I've never seen any point in requiring a
3370 	 * reserved port#, since only in the rare case where the
3371 	 * clients are all boxes with secure system privileges,
3372 	 * does it provide any enhanced security, but... some people
3373 	 * believe it to be useful and keep putting this code back in.
3374 	 * (There is also some "security checker" out there that
3375 	 *  complains if the nfs server doesn't enforce this.)
3376 	 * However, note the following:
3377 	 * RFC3530 (NFSv4) specifies that a reserved port# not be
3378 	 *	required.
3379 	 * RFC2623 recommends that, if a reserved port# is checked for,
3380 	 *	that there be a way to turn that off--> ifdef'd.
3381 	 */
3382 #ifdef NFS_REQRSVPORT
3383 	if (!nd->nd_repstat) {
3384 		struct sockaddr_in *saddr;
3385 		struct sockaddr_in6 *saddr6;
3386 
3387 		saddr = NFSSOCKADDR(nd->nd_nam, struct sockaddr_in *);
3388 		saddr6 = NFSSOCKADDR(nd->nd_nam, struct sockaddr_in6 *);
3389 		if (!(nd->nd_flag & ND_NFSV4) &&
3390 		    ((saddr->sin_family == AF_INET &&
3391 		      ntohs(saddr->sin_port) >= IPPORT_RESERVED) ||
3392 		     (saddr6->sin6_family == AF_INET6 &&
3393 		      ntohs(saddr6->sin6_port) >= IPPORT_RESERVED))) {
3394 			vput(*vpp);
3395 			nd->nd_repstat = (NFSERR_AUTHERR | AUTH_TOOWEAK);
3396 		}
3397 	}
3398 #endif	/* NFS_REQRSVPORT */
3399 
3400 	/*
3401 	 * Check/setup credentials.
3402 	 */
3403 	if (!nd->nd_repstat) {
3404 		nd->nd_saveduid = nd->nd_cred->cr_uid;
3405 		nd->nd_repstat = nfsd_excred(nd, exp, credanon,
3406 		    nfsrv_checkwrongsec(nd, nextop, (*vpp)->v_type));
3407 		if (nd->nd_repstat)
3408 			vput(*vpp);
3409 	}
3410 	if (credanon != NULL)
3411 		crfree(credanon);
3412 	if (nd->nd_repstat) {
3413 		vn_finished_write(mpw);
3414 		*vpp = NULL;
3415 	} else if (mpp != NULL) {
3416 		*mpp = mpw;
3417 	}
3418 
3419 out:
3420 	NFSEXITCODE2(0, nd);
3421 }
3422 
3423 /*
3424  * glue for fp.
3425  */
3426 static int
3427 fp_getfvp(struct thread *p, int fd, struct file **fpp, struct vnode **vpp)
3428 {
3429 	struct filedesc *fdp;
3430 	struct file *fp;
3431 	int error = 0;
3432 
3433 	fdp = p->td_proc->p_fd;
3434 	if (fd < 0 || fd >= fdp->fd_nfiles ||
3435 	    (fp = fdp->fd_ofiles[fd].fde_file) == NULL) {
3436 		error = EBADF;
3437 		goto out;
3438 	}
3439 	*fpp = fp;
3440 
3441 out:
3442 	NFSEXITCODE(error);
3443 	return (error);
3444 }
3445 
3446 /*
3447  * Called from nfssvc() to update the exports list. Just call
3448  * vfs_export(). This has to be done, since the v4 root fake fs isn't
3449  * in the mount list.
3450  */
3451 int
3452 nfsrv_v4rootexport(void *argp, struct ucred *cred, struct thread *p)
3453 {
3454 	struct nfsex_args *nfsexargp = (struct nfsex_args *)argp;
3455 	int error = 0;
3456 	struct nameidata nd;
3457 	fhandle_t fh;
3458 
3459 	error = vfs_export(&nfsv4root_mnt, &nfsexargp->export);
3460 	if ((nfsexargp->export.ex_flags & MNT_DELEXPORT) != 0)
3461 		nfs_rootfhset = 0;
3462 	else if (error == 0) {
3463 		if (nfsexargp->fspec == NULL) {
3464 			error = EPERM;
3465 			goto out;
3466 		}
3467 		/*
3468 		 * If fspec != NULL, this is the v4root path.
3469 		 */
3470 		NDINIT(&nd, LOOKUP, FOLLOW, UIO_USERSPACE, nfsexargp->fspec);
3471 		if ((error = namei(&nd)) != 0)
3472 			goto out;
3473 		error = nfsvno_getfh(nd.ni_vp, &fh, p);
3474 		vrele(nd.ni_vp);
3475 		if (!error) {
3476 			nfs_rootfh.nfsrvfh_len = NFSX_MYFH;
3477 			NFSBCOPY((caddr_t)&fh,
3478 			    nfs_rootfh.nfsrvfh_data,
3479 			    sizeof (fhandle_t));
3480 			nfs_rootfhset = 1;
3481 		}
3482 	}
3483 
3484 out:
3485 	NFSEXITCODE(error);
3486 	return (error);
3487 }
3488 
3489 /*
3490  * This function needs to test to see if the system is near its limit
3491  * for memory allocation via malloc() or mget() and return True iff
3492  * either of these resources are near their limit.
3493  * XXX (For now, this is just a stub.)
3494  */
3495 int nfsrv_testmalloclimit = 0;
3496 int
3497 nfsrv_mallocmget_limit(void)
3498 {
3499 	static int printmesg = 0;
3500 	static int testval = 1;
3501 
3502 	if (nfsrv_testmalloclimit && (testval++ % 1000) == 0) {
3503 		if ((printmesg++ % 100) == 0)
3504 			printf("nfsd: malloc/mget near limit\n");
3505 		return (1);
3506 	}
3507 	return (0);
3508 }
3509 
3510 /*
3511  * BSD specific initialization of a mount point.
3512  */
3513 void
3514 nfsd_mntinit(void)
3515 {
3516 	static int inited = 0;
3517 
3518 	if (inited)
3519 		return;
3520 	inited = 1;
3521 	nfsv4root_mnt.mnt_flag = (MNT_RDONLY | MNT_EXPORTED);
3522 	TAILQ_INIT(&nfsv4root_mnt.mnt_nvnodelist);
3523 	TAILQ_INIT(&nfsv4root_mnt.mnt_lazyvnodelist);
3524 	nfsv4root_mnt.mnt_export = NULL;
3525 	TAILQ_INIT(&nfsv4root_opt);
3526 	TAILQ_INIT(&nfsv4root_newopt);
3527 	nfsv4root_mnt.mnt_opt = &nfsv4root_opt;
3528 	nfsv4root_mnt.mnt_optnew = &nfsv4root_newopt;
3529 	nfsv4root_mnt.mnt_nvnodelistsize = 0;
3530 	nfsv4root_mnt.mnt_lazyvnodelistsize = 0;
3531 }
3532 
3533 /*
3534  * Get a vnode for a file handle, without checking exports, etc.
3535  */
3536 struct vnode *
3537 nfsvno_getvp(fhandle_t *fhp)
3538 {
3539 	struct mount *mp;
3540 	struct vnode *vp;
3541 	int error;
3542 
3543 	mp = vfs_busyfs(&fhp->fh_fsid);
3544 	if (mp == NULL)
3545 		return (NULL);
3546 	error = VFS_FHTOVP(mp, &fhp->fh_fid, LK_EXCLUSIVE, &vp);
3547 	vfs_unbusy(mp);
3548 	if (error)
3549 		return (NULL);
3550 	return (vp);
3551 }
3552 
3553 /*
3554  * Do a local VOP_ADVLOCK().
3555  */
3556 int
3557 nfsvno_advlock(struct vnode *vp, int ftype, u_int64_t first,
3558     u_int64_t end, struct thread *td)
3559 {
3560 	int error = 0;
3561 	struct flock fl;
3562 	u_int64_t tlen;
3563 
3564 	if (nfsrv_dolocallocks == 0)
3565 		goto out;
3566 	ASSERT_VOP_UNLOCKED(vp, "nfsvno_advlock: vp locked");
3567 
3568 	fl.l_whence = SEEK_SET;
3569 	fl.l_type = ftype;
3570 	fl.l_start = (off_t)first;
3571 	if (end == NFS64BITSSET) {
3572 		fl.l_len = 0;
3573 	} else {
3574 		tlen = end - first;
3575 		fl.l_len = (off_t)tlen;
3576 	}
3577 	/*
3578 	 * For FreeBSD8, the l_pid and l_sysid must be set to the same
3579 	 * values for all calls, so that all locks will be held by the
3580 	 * nfsd server. (The nfsd server handles conflicts between the
3581 	 * various clients.)
3582 	 * Since an NFSv4 lockowner is a ClientID plus an array of up to 1024
3583 	 * bytes, so it can't be put in l_sysid.
3584 	 */
3585 	if (nfsv4_sysid == 0)
3586 		nfsv4_sysid = nlm_acquire_next_sysid();
3587 	fl.l_pid = (pid_t)0;
3588 	fl.l_sysid = (int)nfsv4_sysid;
3589 
3590 	if (ftype == F_UNLCK)
3591 		error = VOP_ADVLOCK(vp, (caddr_t)td->td_proc, F_UNLCK, &fl,
3592 		    (F_POSIX | F_REMOTE));
3593 	else
3594 		error = VOP_ADVLOCK(vp, (caddr_t)td->td_proc, F_SETLK, &fl,
3595 		    (F_POSIX | F_REMOTE));
3596 
3597 out:
3598 	NFSEXITCODE(error);
3599 	return (error);
3600 }
3601 
3602 /*
3603  * Check the nfsv4 root exports.
3604  */
3605 int
3606 nfsvno_v4rootexport(struct nfsrv_descript *nd)
3607 {
3608 	struct ucred *credanon;
3609 	int error = 0, numsecflavor, secflavors[MAXSECFLAVORS], i;
3610 	uint64_t exflags;
3611 
3612 	error = vfs_stdcheckexp(&nfsv4root_mnt, nd->nd_nam, &exflags,
3613 	    &credanon, &numsecflavor, secflavors);
3614 	if (error) {
3615 		error = NFSERR_PROGUNAVAIL;
3616 		goto out;
3617 	}
3618 	if (credanon != NULL)
3619 		crfree(credanon);
3620 	for (i = 0; i < numsecflavor; i++) {
3621 		if (secflavors[i] == AUTH_SYS)
3622 			nd->nd_flag |= ND_EXAUTHSYS;
3623 		else if (secflavors[i] == RPCSEC_GSS_KRB5)
3624 			nd->nd_flag |= ND_EXGSS;
3625 		else if (secflavors[i] == RPCSEC_GSS_KRB5I)
3626 			nd->nd_flag |= ND_EXGSSINTEGRITY;
3627 		else if (secflavors[i] == RPCSEC_GSS_KRB5P)
3628 			nd->nd_flag |= ND_EXGSSPRIVACY;
3629 	}
3630 
3631 	/* And set ND_EXxx flags for TLS. */
3632 	if ((exflags & MNT_EXTLS) != 0) {
3633 		nd->nd_flag |= ND_EXTLS;
3634 		if ((exflags & MNT_EXTLSCERT) != 0)
3635 			nd->nd_flag |= ND_EXTLSCERT;
3636 		if ((exflags & MNT_EXTLSCERTUSER) != 0)
3637 			nd->nd_flag |= ND_EXTLSCERTUSER;
3638 	}
3639 
3640 out:
3641 	NFSEXITCODE(error);
3642 	return (error);
3643 }
3644 
3645 /*
3646  * Nfs server pseudo system call for the nfsd's
3647  */
3648 /*
3649  * MPSAFE
3650  */
3651 static int
3652 nfssvc_nfsd(struct thread *td, struct nfssvc_args *uap)
3653 {
3654 	struct file *fp;
3655 	struct nfsd_addsock_args sockarg;
3656 	struct nfsd_nfsd_args nfsdarg;
3657 	struct nfsd_nfsd_oargs onfsdarg;
3658 	struct nfsd_pnfsd_args pnfsdarg;
3659 	struct vnode *vp, *nvp, *curdvp;
3660 	struct pnfsdsfile *pf;
3661 	struct nfsdevice *ds, *fds;
3662 	cap_rights_t rights;
3663 	int buflen, error, ret;
3664 	char *buf, *cp, *cp2, *cp3;
3665 	char fname[PNFS_FILENAME_LEN + 1];
3666 
3667 	if (uap->flag & NFSSVC_NFSDADDSOCK) {
3668 		error = copyin(uap->argp, (caddr_t)&sockarg, sizeof (sockarg));
3669 		if (error)
3670 			goto out;
3671 		/*
3672 		 * Since we don't know what rights might be required,
3673 		 * pretend that we need them all. It is better to be too
3674 		 * careful than too reckless.
3675 		 */
3676 		error = fget(td, sockarg.sock,
3677 		    cap_rights_init_one(&rights, CAP_SOCK_SERVER), &fp);
3678 		if (error != 0)
3679 			goto out;
3680 		if (fp->f_type != DTYPE_SOCKET) {
3681 			fdrop(fp, td);
3682 			error = EPERM;
3683 			goto out;
3684 		}
3685 		error = nfsrvd_addsock(fp);
3686 		fdrop(fp, td);
3687 	} else if (uap->flag & NFSSVC_NFSDNFSD) {
3688 		if (uap->argp == NULL) {
3689 			error = EINVAL;
3690 			goto out;
3691 		}
3692 		if ((uap->flag & NFSSVC_NEWSTRUCT) == 0) {
3693 			error = copyin(uap->argp, &onfsdarg, sizeof(onfsdarg));
3694 			if (error == 0) {
3695 				nfsdarg.principal = onfsdarg.principal;
3696 				nfsdarg.minthreads = onfsdarg.minthreads;
3697 				nfsdarg.maxthreads = onfsdarg.maxthreads;
3698 				nfsdarg.version = 1;
3699 				nfsdarg.addr = NULL;
3700 				nfsdarg.addrlen = 0;
3701 				nfsdarg.dnshost = NULL;
3702 				nfsdarg.dnshostlen = 0;
3703 				nfsdarg.dspath = NULL;
3704 				nfsdarg.dspathlen = 0;
3705 				nfsdarg.mdspath = NULL;
3706 				nfsdarg.mdspathlen = 0;
3707 				nfsdarg.mirrorcnt = 1;
3708 			}
3709 		} else
3710 			error = copyin(uap->argp, &nfsdarg, sizeof(nfsdarg));
3711 		if (error)
3712 			goto out;
3713 		if (nfsdarg.addrlen > 0 && nfsdarg.addrlen < 10000 &&
3714 		    nfsdarg.dnshostlen > 0 && nfsdarg.dnshostlen < 10000 &&
3715 		    nfsdarg.dspathlen > 0 && nfsdarg.dspathlen < 10000 &&
3716 		    nfsdarg.mdspathlen > 0 && nfsdarg.mdspathlen < 10000 &&
3717 		    nfsdarg.mirrorcnt >= 1 &&
3718 		    nfsdarg.mirrorcnt <= NFSDEV_MAXMIRRORS &&
3719 		    nfsdarg.addr != NULL && nfsdarg.dnshost != NULL &&
3720 		    nfsdarg.dspath != NULL && nfsdarg.mdspath != NULL) {
3721 			NFSD_DEBUG(1, "addrlen=%d dspathlen=%d dnslen=%d"
3722 			    " mdspathlen=%d mirrorcnt=%d\n", nfsdarg.addrlen,
3723 			    nfsdarg.dspathlen, nfsdarg.dnshostlen,
3724 			    nfsdarg.mdspathlen, nfsdarg.mirrorcnt);
3725 			cp = malloc(nfsdarg.addrlen + 1, M_TEMP, M_WAITOK);
3726 			error = copyin(nfsdarg.addr, cp, nfsdarg.addrlen);
3727 			if (error != 0) {
3728 				free(cp, M_TEMP);
3729 				goto out;
3730 			}
3731 			cp[nfsdarg.addrlen] = '\0';	/* Ensure nul term. */
3732 			nfsdarg.addr = cp;
3733 			cp = malloc(nfsdarg.dnshostlen + 1, M_TEMP, M_WAITOK);
3734 			error = copyin(nfsdarg.dnshost, cp, nfsdarg.dnshostlen);
3735 			if (error != 0) {
3736 				free(nfsdarg.addr, M_TEMP);
3737 				free(cp, M_TEMP);
3738 				goto out;
3739 			}
3740 			cp[nfsdarg.dnshostlen] = '\0';	/* Ensure nul term. */
3741 			nfsdarg.dnshost = cp;
3742 			cp = malloc(nfsdarg.dspathlen + 1, M_TEMP, M_WAITOK);
3743 			error = copyin(nfsdarg.dspath, cp, nfsdarg.dspathlen);
3744 			if (error != 0) {
3745 				free(nfsdarg.addr, M_TEMP);
3746 				free(nfsdarg.dnshost, M_TEMP);
3747 				free(cp, M_TEMP);
3748 				goto out;
3749 			}
3750 			cp[nfsdarg.dspathlen] = '\0';	/* Ensure nul term. */
3751 			nfsdarg.dspath = cp;
3752 			cp = malloc(nfsdarg.mdspathlen + 1, M_TEMP, M_WAITOK);
3753 			error = copyin(nfsdarg.mdspath, cp, nfsdarg.mdspathlen);
3754 			if (error != 0) {
3755 				free(nfsdarg.addr, M_TEMP);
3756 				free(nfsdarg.dnshost, M_TEMP);
3757 				free(nfsdarg.dspath, M_TEMP);
3758 				free(cp, M_TEMP);
3759 				goto out;
3760 			}
3761 			cp[nfsdarg.mdspathlen] = '\0';	/* Ensure nul term. */
3762 			nfsdarg.mdspath = cp;
3763 		} else {
3764 			nfsdarg.addr = NULL;
3765 			nfsdarg.addrlen = 0;
3766 			nfsdarg.dnshost = NULL;
3767 			nfsdarg.dnshostlen = 0;
3768 			nfsdarg.dspath = NULL;
3769 			nfsdarg.dspathlen = 0;
3770 			nfsdarg.mdspath = NULL;
3771 			nfsdarg.mdspathlen = 0;
3772 			nfsdarg.mirrorcnt = 1;
3773 		}
3774 		error = nfsrvd_nfsd(td, &nfsdarg);
3775 		free(nfsdarg.addr, M_TEMP);
3776 		free(nfsdarg.dnshost, M_TEMP);
3777 		free(nfsdarg.dspath, M_TEMP);
3778 		free(nfsdarg.mdspath, M_TEMP);
3779 	} else if (uap->flag & NFSSVC_PNFSDS) {
3780 		error = copyin(uap->argp, &pnfsdarg, sizeof(pnfsdarg));
3781 		if (error == 0 && (pnfsdarg.op == PNFSDOP_DELDSSERVER ||
3782 		    pnfsdarg.op == PNFSDOP_FORCEDELDS)) {
3783 			cp = malloc(PATH_MAX + 1, M_TEMP, M_WAITOK);
3784 			error = copyinstr(pnfsdarg.dspath, cp, PATH_MAX + 1,
3785 			    NULL);
3786 			if (error == 0)
3787 				error = nfsrv_deldsserver(pnfsdarg.op, cp, td);
3788 			free(cp, M_TEMP);
3789 		} else if (error == 0 && pnfsdarg.op == PNFSDOP_COPYMR) {
3790 			cp = malloc(PATH_MAX + 1, M_TEMP, M_WAITOK);
3791 			buflen = sizeof(*pf) * NFSDEV_MAXMIRRORS;
3792 			buf = malloc(buflen, M_TEMP, M_WAITOK);
3793 			error = copyinstr(pnfsdarg.mdspath, cp, PATH_MAX + 1,
3794 			    NULL);
3795 			NFSD_DEBUG(4, "pnfsdcopymr cp mdspath=%d\n", error);
3796 			if (error == 0 && pnfsdarg.dspath != NULL) {
3797 				cp2 = malloc(PATH_MAX + 1, M_TEMP, M_WAITOK);
3798 				error = copyinstr(pnfsdarg.dspath, cp2,
3799 				    PATH_MAX + 1, NULL);
3800 				NFSD_DEBUG(4, "pnfsdcopymr cp dspath=%d\n",
3801 				    error);
3802 			} else
3803 				cp2 = NULL;
3804 			if (error == 0 && pnfsdarg.curdspath != NULL) {
3805 				cp3 = malloc(PATH_MAX + 1, M_TEMP, M_WAITOK);
3806 				error = copyinstr(pnfsdarg.curdspath, cp3,
3807 				    PATH_MAX + 1, NULL);
3808 				NFSD_DEBUG(4, "pnfsdcopymr cp curdspath=%d\n",
3809 				    error);
3810 			} else
3811 				cp3 = NULL;
3812 			curdvp = NULL;
3813 			fds = NULL;
3814 			if (error == 0)
3815 				error = nfsrv_mdscopymr(cp, cp2, cp3, buf,
3816 				    &buflen, fname, td, &vp, &nvp, &pf, &ds,
3817 				    &fds);
3818 			NFSD_DEBUG(4, "nfsrv_mdscopymr=%d\n", error);
3819 			if (error == 0) {
3820 				if (pf->dsf_dir >= nfsrv_dsdirsize) {
3821 					printf("copymr: dsdir out of range\n");
3822 					pf->dsf_dir = 0;
3823 				}
3824 				NFSD_DEBUG(4, "copymr: buflen=%d\n", buflen);
3825 				error = nfsrv_copymr(vp, nvp,
3826 				    ds->nfsdev_dsdir[pf->dsf_dir], ds, pf,
3827 				    (struct pnfsdsfile *)buf,
3828 				    buflen / sizeof(*pf), td->td_ucred, td);
3829 				vput(vp);
3830 				vput(nvp);
3831 				if (fds != NULL && error == 0) {
3832 					curdvp = fds->nfsdev_dsdir[pf->dsf_dir];
3833 					ret = vn_lock(curdvp, LK_EXCLUSIVE);
3834 					if (ret == 0) {
3835 						nfsrv_dsremove(curdvp, fname,
3836 						    td->td_ucred, td);
3837 						NFSVOPUNLOCK(curdvp);
3838 					}
3839 				}
3840 				NFSD_DEBUG(4, "nfsrv_copymr=%d\n", error);
3841 			}
3842 			free(cp, M_TEMP);
3843 			free(cp2, M_TEMP);
3844 			free(cp3, M_TEMP);
3845 			free(buf, M_TEMP);
3846 		}
3847 	} else {
3848 		error = nfssvc_srvcall(td, uap, td->td_ucred);
3849 	}
3850 
3851 out:
3852 	NFSEXITCODE(error);
3853 	return (error);
3854 }
3855 
3856 static int
3857 nfssvc_srvcall(struct thread *p, struct nfssvc_args *uap, struct ucred *cred)
3858 {
3859 	struct nfsex_args export;
3860 	struct nfsex_oldargs oexp;
3861 	struct file *fp = NULL;
3862 	int stablefd, i, len;
3863 	struct nfsd_clid adminrevoke;
3864 	struct nfsd_dumplist dumplist;
3865 	struct nfsd_dumpclients *dumpclients;
3866 	struct nfsd_dumplocklist dumplocklist;
3867 	struct nfsd_dumplocks *dumplocks;
3868 	struct nameidata nd;
3869 	vnode_t vp;
3870 	int error = EINVAL, igotlock;
3871 	struct proc *procp;
3872 	gid_t *grps;
3873 	static int suspend_nfsd = 0;
3874 
3875 	if (uap->flag & NFSSVC_PUBLICFH) {
3876 		NFSBZERO((caddr_t)&nfs_pubfh.nfsrvfh_data,
3877 		    sizeof (fhandle_t));
3878 		error = copyin(uap->argp,
3879 		    &nfs_pubfh.nfsrvfh_data, sizeof (fhandle_t));
3880 		if (!error)
3881 			nfs_pubfhset = 1;
3882 	} else if ((uap->flag & (NFSSVC_V4ROOTEXPORT | NFSSVC_NEWSTRUCT)) ==
3883 	    (NFSSVC_V4ROOTEXPORT | NFSSVC_NEWSTRUCT)) {
3884 		error = copyin(uap->argp,(caddr_t)&export,
3885 		    sizeof (struct nfsex_args));
3886 		if (!error) {
3887 			grps = NULL;
3888 			if (export.export.ex_ngroups > NGROUPS_MAX ||
3889 			    export.export.ex_ngroups < 0)
3890 				error = EINVAL;
3891 			else if (export.export.ex_ngroups > 0) {
3892 				grps = malloc(export.export.ex_ngroups *
3893 				    sizeof(gid_t), M_TEMP, M_WAITOK);
3894 				error = copyin(export.export.ex_groups, grps,
3895 				    export.export.ex_ngroups * sizeof(gid_t));
3896 				export.export.ex_groups = grps;
3897 			} else
3898 				export.export.ex_groups = NULL;
3899 			if (!error)
3900 				error = nfsrv_v4rootexport(&export, cred, p);
3901 			free(grps, M_TEMP);
3902 		}
3903 	} else if ((uap->flag & (NFSSVC_V4ROOTEXPORT | NFSSVC_NEWSTRUCT)) ==
3904 	    NFSSVC_V4ROOTEXPORT) {
3905 		error = copyin(uap->argp,(caddr_t)&oexp,
3906 		    sizeof (struct nfsex_oldargs));
3907 		if (!error) {
3908 			memset(&export.export, 0, sizeof(export.export));
3909 			export.export.ex_flags = (uint64_t)oexp.export.ex_flags;
3910 			export.export.ex_root = oexp.export.ex_root;
3911 			export.export.ex_uid = oexp.export.ex_anon.cr_uid;
3912 			export.export.ex_ngroups =
3913 			    oexp.export.ex_anon.cr_ngroups;
3914 			export.export.ex_groups = NULL;
3915 			if (export.export.ex_ngroups > XU_NGROUPS ||
3916 			    export.export.ex_ngroups < 0)
3917 				error = EINVAL;
3918 			else if (export.export.ex_ngroups > 0) {
3919 				export.export.ex_groups = malloc(
3920 				    export.export.ex_ngroups * sizeof(gid_t),
3921 				    M_TEMP, M_WAITOK);
3922 				for (i = 0; i < export.export.ex_ngroups; i++)
3923 					export.export.ex_groups[i] =
3924 					    oexp.export.ex_anon.cr_groups[i];
3925 			}
3926 			export.export.ex_addr = oexp.export.ex_addr;
3927 			export.export.ex_addrlen = oexp.export.ex_addrlen;
3928 			export.export.ex_mask = oexp.export.ex_mask;
3929 			export.export.ex_masklen = oexp.export.ex_masklen;
3930 			export.export.ex_indexfile = oexp.export.ex_indexfile;
3931 			export.export.ex_numsecflavors =
3932 			    oexp.export.ex_numsecflavors;
3933 			if (export.export.ex_numsecflavors >= MAXSECFLAVORS ||
3934 			    export.export.ex_numsecflavors < 0)
3935 				error = EINVAL;
3936 			else {
3937 				for (i = 0; i < export.export.ex_numsecflavors;
3938 				    i++)
3939 					export.export.ex_secflavors[i] =
3940 					    oexp.export.ex_secflavors[i];
3941 			}
3942 			export.fspec = oexp.fspec;
3943 			if (error == 0)
3944 				error = nfsrv_v4rootexport(&export, cred, p);
3945 			free(export.export.ex_groups, M_TEMP);
3946 		}
3947 	} else if (uap->flag & NFSSVC_NOPUBLICFH) {
3948 		nfs_pubfhset = 0;
3949 		error = 0;
3950 	} else if (uap->flag & NFSSVC_STABLERESTART) {
3951 		error = copyin(uap->argp, (caddr_t)&stablefd,
3952 		    sizeof (int));
3953 		if (!error)
3954 			error = fp_getfvp(p, stablefd, &fp, &vp);
3955 		if (!error && (NFSFPFLAG(fp) & (FREAD | FWRITE)) != (FREAD | FWRITE))
3956 			error = EBADF;
3957 		if (!error && newnfs_numnfsd != 0)
3958 			error = EPERM;
3959 		if (!error) {
3960 			nfsrv_stablefirst.nsf_fp = fp;
3961 			nfsrv_setupstable(p);
3962 		}
3963 	} else if (uap->flag & NFSSVC_ADMINREVOKE) {
3964 		error = copyin(uap->argp, (caddr_t)&adminrevoke,
3965 		    sizeof (struct nfsd_clid));
3966 		if (!error)
3967 			error = nfsrv_adminrevoke(&adminrevoke, p);
3968 	} else if (uap->flag & NFSSVC_DUMPCLIENTS) {
3969 		error = copyin(uap->argp, (caddr_t)&dumplist,
3970 		    sizeof (struct nfsd_dumplist));
3971 		if (!error && (dumplist.ndl_size < 1 ||
3972 			dumplist.ndl_size > NFSRV_MAXDUMPLIST))
3973 			error = EPERM;
3974 		if (!error) {
3975 		    len = sizeof (struct nfsd_dumpclients) * dumplist.ndl_size;
3976 		    dumpclients = malloc(len, M_TEMP, M_WAITOK | M_ZERO);
3977 		    nfsrv_dumpclients(dumpclients, dumplist.ndl_size);
3978 		    error = copyout(dumpclients, dumplist.ndl_list, len);
3979 		    free(dumpclients, M_TEMP);
3980 		}
3981 	} else if (uap->flag & NFSSVC_DUMPLOCKS) {
3982 		error = copyin(uap->argp, (caddr_t)&dumplocklist,
3983 		    sizeof (struct nfsd_dumplocklist));
3984 		if (!error && (dumplocklist.ndllck_size < 1 ||
3985 			dumplocklist.ndllck_size > NFSRV_MAXDUMPLIST))
3986 			error = EPERM;
3987 		if (!error)
3988 			error = nfsrv_lookupfilename(&nd,
3989 				dumplocklist.ndllck_fname, p);
3990 		if (!error) {
3991 			len = sizeof (struct nfsd_dumplocks) *
3992 				dumplocklist.ndllck_size;
3993 			dumplocks = malloc(len, M_TEMP, M_WAITOK | M_ZERO);
3994 			nfsrv_dumplocks(nd.ni_vp, dumplocks,
3995 			    dumplocklist.ndllck_size, p);
3996 			vput(nd.ni_vp);
3997 			error = copyout(dumplocks, dumplocklist.ndllck_list,
3998 			    len);
3999 			free(dumplocks, M_TEMP);
4000 		}
4001 	} else if (uap->flag & NFSSVC_BACKUPSTABLE) {
4002 		procp = p->td_proc;
4003 		PROC_LOCK(procp);
4004 		nfsd_master_pid = procp->p_pid;
4005 		bcopy(procp->p_comm, nfsd_master_comm, MAXCOMLEN + 1);
4006 		nfsd_master_start = procp->p_stats->p_start;
4007 		nfsd_master_proc = procp;
4008 		PROC_UNLOCK(procp);
4009 	} else if ((uap->flag & NFSSVC_SUSPENDNFSD) != 0) {
4010 		NFSLOCKV4ROOTMUTEX();
4011 		if (suspend_nfsd == 0) {
4012 			/* Lock out all nfsd threads */
4013 			do {
4014 				igotlock = nfsv4_lock(&nfsd_suspend_lock, 1,
4015 				    NULL, NFSV4ROOTLOCKMUTEXPTR, NULL);
4016 			} while (igotlock == 0 && suspend_nfsd == 0);
4017 			suspend_nfsd = 1;
4018 		}
4019 		NFSUNLOCKV4ROOTMUTEX();
4020 		error = 0;
4021 	} else if ((uap->flag & NFSSVC_RESUMENFSD) != 0) {
4022 		NFSLOCKV4ROOTMUTEX();
4023 		if (suspend_nfsd != 0) {
4024 			nfsv4_unlock(&nfsd_suspend_lock, 0);
4025 			suspend_nfsd = 0;
4026 		}
4027 		NFSUNLOCKV4ROOTMUTEX();
4028 		error = 0;
4029 	}
4030 
4031 	NFSEXITCODE(error);
4032 	return (error);
4033 }
4034 
4035 /*
4036  * Check exports.
4037  * Returns 0 if ok, 1 otherwise.
4038  */
4039 int
4040 nfsvno_testexp(struct nfsrv_descript *nd, struct nfsexstuff *exp)
4041 {
4042 	int i;
4043 
4044 	/*
4045 	 * Allow NFSv3 Fsinfo per RFC2623.
4046 	 */
4047 	if (((nd->nd_flag & ND_NFSV4) != 0 ||
4048 	     nd->nd_procnum != NFSPROC_FSINFO) &&
4049 	    ((NFSVNO_EXTLS(exp) && (nd->nd_flag & ND_TLS) == 0) ||
4050 	     (NFSVNO_EXTLSCERT(exp) &&
4051 	      (nd->nd_flag & ND_TLSCERT) == 0) ||
4052 	     (NFSVNO_EXTLSCERTUSER(exp) &&
4053 	      (nd->nd_flag & ND_TLSCERTUSER) == 0))) {
4054 		if ((nd->nd_flag & ND_NFSV4) != 0)
4055 			return (NFSERR_WRONGSEC);
4056 		else if ((nd->nd_flag & ND_TLS) == 0)
4057 			return (NFSERR_AUTHERR | AUTH_NEEDS_TLS);
4058 		else
4059 			return (NFSERR_AUTHERR | AUTH_NEEDS_TLS_MUTUAL_HOST);
4060 	}
4061 
4062 	/*
4063 	 * This seems odd, but allow the case where the security flavor
4064 	 * list is empty. This happens when NFSv4 is traversing non-exported
4065 	 * file systems. Exported file systems should always have a non-empty
4066 	 * security flavor list.
4067 	 */
4068 	if (exp->nes_numsecflavor == 0)
4069 		return (0);
4070 
4071 	for (i = 0; i < exp->nes_numsecflavor; i++) {
4072 		/*
4073 		 * The tests for privacy and integrity must be first,
4074 		 * since ND_GSS is set for everything but AUTH_SYS.
4075 		 */
4076 		if (exp->nes_secflavors[i] == RPCSEC_GSS_KRB5P &&
4077 		    (nd->nd_flag & ND_GSSPRIVACY))
4078 			return (0);
4079 		if (exp->nes_secflavors[i] == RPCSEC_GSS_KRB5I &&
4080 		    (nd->nd_flag & ND_GSSINTEGRITY))
4081 			return (0);
4082 		if (exp->nes_secflavors[i] == RPCSEC_GSS_KRB5 &&
4083 		    (nd->nd_flag & ND_GSS))
4084 			return (0);
4085 		if (exp->nes_secflavors[i] == AUTH_SYS &&
4086 		    (nd->nd_flag & ND_GSS) == 0)
4087 			return (0);
4088 	}
4089 	if ((nd->nd_flag & ND_NFSV4) != 0)
4090 		return (NFSERR_WRONGSEC);
4091 	return (NFSERR_AUTHERR | AUTH_TOOWEAK);
4092 }
4093 
4094 /*
4095  * Calculate a hash value for the fid in a file handle.
4096  */
4097 uint32_t
4098 nfsrv_hashfh(fhandle_t *fhp)
4099 {
4100 	uint32_t hashval;
4101 
4102 	hashval = hash32_buf(&fhp->fh_fid, sizeof(struct fid), 0);
4103 	return (hashval);
4104 }
4105 
4106 /*
4107  * Calculate a hash value for the sessionid.
4108  */
4109 uint32_t
4110 nfsrv_hashsessionid(uint8_t *sessionid)
4111 {
4112 	uint32_t hashval;
4113 
4114 	hashval = hash32_buf(sessionid, NFSX_V4SESSIONID, 0);
4115 	return (hashval);
4116 }
4117 
4118 /*
4119  * Signal the userland master nfsd to backup the stable restart file.
4120  */
4121 void
4122 nfsrv_backupstable(void)
4123 {
4124 	struct proc *procp;
4125 
4126 	if (nfsd_master_proc != NULL) {
4127 		procp = pfind(nfsd_master_pid);
4128 		/* Try to make sure it is the correct process. */
4129 		if (procp == nfsd_master_proc &&
4130 		    procp->p_stats->p_start.tv_sec ==
4131 		    nfsd_master_start.tv_sec &&
4132 		    procp->p_stats->p_start.tv_usec ==
4133 		    nfsd_master_start.tv_usec &&
4134 		    strcmp(procp->p_comm, nfsd_master_comm) == 0)
4135 			kern_psignal(procp, SIGUSR2);
4136 		else
4137 			nfsd_master_proc = NULL;
4138 
4139 		if (procp != NULL)
4140 			PROC_UNLOCK(procp);
4141 	}
4142 }
4143 
4144 /*
4145  * Create a DS data file for nfsrv_pnfscreate(). Called for each mirror.
4146  * The arguments are in a structure, so that they can be passed through
4147  * taskqueue for a kernel process to execute this function.
4148  */
4149 struct nfsrvdscreate {
4150 	int			done;
4151 	int			inprog;
4152 	struct task		tsk;
4153 	struct ucred		*tcred;
4154 	struct vnode		*dvp;
4155 	NFSPROC_T		*p;
4156 	struct pnfsdsfile	*pf;
4157 	int			err;
4158 	fhandle_t		fh;
4159 	struct vattr		va;
4160 	struct vattr		createva;
4161 };
4162 
4163 int
4164 nfsrv_dscreate(struct vnode *dvp, struct vattr *vap, struct vattr *nvap,
4165     fhandle_t *fhp, struct pnfsdsfile *pf, struct pnfsdsattr *dsa,
4166     char *fnamep, struct ucred *tcred, NFSPROC_T *p, struct vnode **nvpp)
4167 {
4168 	struct vnode *nvp;
4169 	struct nameidata named;
4170 	struct vattr va;
4171 	char *bufp;
4172 	u_long *hashp;
4173 	struct nfsnode *np;
4174 	struct nfsmount *nmp;
4175 	int error;
4176 
4177 	NFSNAMEICNDSET(&named.ni_cnd, tcred, CREATE,
4178 	    LOCKPARENT | LOCKLEAF | SAVESTART | NOCACHE);
4179 	nfsvno_setpathbuf(&named, &bufp, &hashp);
4180 	named.ni_cnd.cn_lkflags = LK_EXCLUSIVE;
4181 	named.ni_cnd.cn_nameptr = bufp;
4182 	if (fnamep != NULL) {
4183 		strlcpy(bufp, fnamep, PNFS_FILENAME_LEN + 1);
4184 		named.ni_cnd.cn_namelen = strlen(bufp);
4185 	} else
4186 		named.ni_cnd.cn_namelen = nfsrv_putfhname(fhp, bufp);
4187 	NFSD_DEBUG(4, "nfsrv_dscreate: dvp=%p fname=%s\n", dvp, bufp);
4188 
4189 	/* Create the date file in the DS mount. */
4190 	error = NFSVOPLOCK(dvp, LK_EXCLUSIVE);
4191 	if (error == 0) {
4192 		error = VOP_CREATE(dvp, &nvp, &named.ni_cnd, vap);
4193 		vref(dvp);
4194 		VOP_VPUT_PAIR(dvp, error == 0 ? &nvp : NULL, false);
4195 		if (error == 0) {
4196 			/* Set the ownership of the file. */
4197 			error = VOP_SETATTR(nvp, nvap, tcred);
4198 			NFSD_DEBUG(4, "nfsrv_dscreate:"
4199 			    " setattr-uid=%d\n", error);
4200 			if (error != 0)
4201 				vput(nvp);
4202 		}
4203 		if (error != 0)
4204 			printf("pNFS: pnfscreate failed=%d\n", error);
4205 	} else
4206 		printf("pNFS: pnfscreate vnlock=%d\n", error);
4207 	if (error == 0) {
4208 		np = VTONFS(nvp);
4209 		nmp = VFSTONFS(nvp->v_mount);
4210 		if (strcmp(nvp->v_mount->mnt_vfc->vfc_name, "nfs")
4211 		    != 0 || nmp->nm_nam->sa_len > sizeof(
4212 		    struct sockaddr_in6) ||
4213 		    np->n_fhp->nfh_len != NFSX_MYFH) {
4214 			printf("Bad DS file: fstype=%s salen=%d"
4215 			    " fhlen=%d\n",
4216 			    nvp->v_mount->mnt_vfc->vfc_name,
4217 			    nmp->nm_nam->sa_len, np->n_fhp->nfh_len);
4218 			error = ENOENT;
4219 		}
4220 
4221 		/* Set extattrs for the DS on the MDS file. */
4222 		if (error == 0) {
4223 			if (dsa != NULL) {
4224 				error = VOP_GETATTR(nvp, &va, tcred);
4225 				if (error == 0) {
4226 					dsa->dsa_filerev = va.va_filerev;
4227 					dsa->dsa_size = va.va_size;
4228 					dsa->dsa_atime = va.va_atime;
4229 					dsa->dsa_mtime = va.va_mtime;
4230 					dsa->dsa_bytes = va.va_bytes;
4231 				}
4232 			}
4233 			if (error == 0) {
4234 				NFSBCOPY(np->n_fhp->nfh_fh, &pf->dsf_fh,
4235 				    NFSX_MYFH);
4236 				NFSBCOPY(nmp->nm_nam, &pf->dsf_sin,
4237 				    nmp->nm_nam->sa_len);
4238 				NFSBCOPY(named.ni_cnd.cn_nameptr,
4239 				    pf->dsf_filename,
4240 				    sizeof(pf->dsf_filename));
4241 			}
4242 		} else
4243 			printf("pNFS: pnfscreate can't get DS"
4244 			    " attr=%d\n", error);
4245 		if (nvpp != NULL && error == 0)
4246 			*nvpp = nvp;
4247 		else
4248 			vput(nvp);
4249 	}
4250 	nfsvno_relpathbuf(&named);
4251 	return (error);
4252 }
4253 
4254 /*
4255  * Start up the thread that will execute nfsrv_dscreate().
4256  */
4257 static void
4258 start_dscreate(void *arg, int pending)
4259 {
4260 	struct nfsrvdscreate *dsc;
4261 
4262 	dsc = (struct nfsrvdscreate *)arg;
4263 	dsc->err = nfsrv_dscreate(dsc->dvp, &dsc->createva, &dsc->va, &dsc->fh,
4264 	    dsc->pf, NULL, NULL, dsc->tcred, dsc->p, NULL);
4265 	dsc->done = 1;
4266 	NFSD_DEBUG(4, "start_dscreate: err=%d\n", dsc->err);
4267 }
4268 
4269 /*
4270  * Create a pNFS data file on the Data Server(s).
4271  */
4272 static void
4273 nfsrv_pnfscreate(struct vnode *vp, struct vattr *vap, struct ucred *cred,
4274     NFSPROC_T *p)
4275 {
4276 	struct nfsrvdscreate *dsc, *tdsc = NULL;
4277 	struct nfsdevice *ds, *tds, *fds;
4278 	struct mount *mp;
4279 	struct pnfsdsfile *pf, *tpf;
4280 	struct pnfsdsattr dsattr;
4281 	struct vattr va;
4282 	struct vnode *dvp[NFSDEV_MAXMIRRORS];
4283 	struct nfsmount *nmp;
4284 	fhandle_t fh;
4285 	uid_t vauid;
4286 	gid_t vagid;
4287 	u_short vamode;
4288 	struct ucred *tcred;
4289 	int dsdir[NFSDEV_MAXMIRRORS], error, i, mirrorcnt, ret;
4290 	int failpos, timo;
4291 
4292 	/* Get a DS server directory in a round-robin order. */
4293 	mirrorcnt = 1;
4294 	mp = vp->v_mount;
4295 	ds = fds = NULL;
4296 	NFSDDSLOCK();
4297 	/*
4298 	 * Search for the first entry that handles this MDS fs, but use the
4299 	 * first entry for all MDS fs's otherwise.
4300 	 */
4301 	TAILQ_FOREACH(tds, &nfsrv_devidhead, nfsdev_list) {
4302 		if (tds->nfsdev_nmp != NULL) {
4303 			if (tds->nfsdev_mdsisset == 0 && ds == NULL)
4304 				ds = tds;
4305 			else if (tds->nfsdev_mdsisset != 0 && fsidcmp(
4306 			    &mp->mnt_stat.f_fsid, &tds->nfsdev_mdsfsid) == 0) {
4307 				ds = fds = tds;
4308 				break;
4309 			}
4310 		}
4311 	}
4312 	if (ds == NULL) {
4313 		NFSDDSUNLOCK();
4314 		NFSD_DEBUG(4, "nfsrv_pnfscreate: no srv\n");
4315 		return;
4316 	}
4317 	i = dsdir[0] = ds->nfsdev_nextdir;
4318 	ds->nfsdev_nextdir = (ds->nfsdev_nextdir + 1) % nfsrv_dsdirsize;
4319 	dvp[0] = ds->nfsdev_dsdir[i];
4320 	tds = TAILQ_NEXT(ds, nfsdev_list);
4321 	if (nfsrv_maxpnfsmirror > 1 && tds != NULL) {
4322 		TAILQ_FOREACH_FROM(tds, &nfsrv_devidhead, nfsdev_list) {
4323 			if (tds->nfsdev_nmp != NULL &&
4324 			    ((tds->nfsdev_mdsisset == 0 && fds == NULL) ||
4325 			     (tds->nfsdev_mdsisset != 0 && fds != NULL &&
4326 			      fsidcmp(&mp->mnt_stat.f_fsid,
4327 			      &tds->nfsdev_mdsfsid) == 0))) {
4328 				dsdir[mirrorcnt] = i;
4329 				dvp[mirrorcnt] = tds->nfsdev_dsdir[i];
4330 				mirrorcnt++;
4331 				if (mirrorcnt >= nfsrv_maxpnfsmirror)
4332 					break;
4333 			}
4334 		}
4335 	}
4336 	/* Put at end of list to implement round-robin usage. */
4337 	TAILQ_REMOVE(&nfsrv_devidhead, ds, nfsdev_list);
4338 	TAILQ_INSERT_TAIL(&nfsrv_devidhead, ds, nfsdev_list);
4339 	NFSDDSUNLOCK();
4340 	dsc = NULL;
4341 	if (mirrorcnt > 1)
4342 		tdsc = dsc = malloc(sizeof(*dsc) * (mirrorcnt - 1), M_TEMP,
4343 		    M_WAITOK | M_ZERO);
4344 	tpf = pf = malloc(sizeof(*pf) * nfsrv_maxpnfsmirror, M_TEMP, M_WAITOK |
4345 	    M_ZERO);
4346 
4347 	error = nfsvno_getfh(vp, &fh, p);
4348 	if (error == 0)
4349 		error = VOP_GETATTR(vp, &va, cred);
4350 	if (error == 0) {
4351 		/* Set the attributes for "vp" to Setattr the DS vp. */
4352 		vauid = va.va_uid;
4353 		vagid = va.va_gid;
4354 		vamode = va.va_mode;
4355 		VATTR_NULL(&va);
4356 		va.va_uid = vauid;
4357 		va.va_gid = vagid;
4358 		va.va_mode = vamode;
4359 		va.va_size = 0;
4360 	} else
4361 		printf("pNFS: pnfscreate getfh+attr=%d\n", error);
4362 
4363 	NFSD_DEBUG(4, "nfsrv_pnfscreate: cruid=%d crgid=%d\n", cred->cr_uid,
4364 	    cred->cr_gid);
4365 	/* Make data file name based on FH. */
4366 	tcred = newnfs_getcred();
4367 
4368 	/*
4369 	 * Create the file on each DS mirror, using kernel process(es) for the
4370 	 * additional mirrors.
4371 	 */
4372 	failpos = -1;
4373 	for (i = 0; i < mirrorcnt - 1 && error == 0; i++, tpf++, tdsc++) {
4374 		tpf->dsf_dir = dsdir[i];
4375 		tdsc->tcred = tcred;
4376 		tdsc->p = p;
4377 		tdsc->pf = tpf;
4378 		tdsc->createva = *vap;
4379 		NFSBCOPY(&fh, &tdsc->fh, sizeof(fh));
4380 		tdsc->va = va;
4381 		tdsc->dvp = dvp[i];
4382 		tdsc->done = 0;
4383 		tdsc->inprog = 0;
4384 		tdsc->err = 0;
4385 		ret = EIO;
4386 		if (nfs_pnfsiothreads != 0) {
4387 			ret = nfs_pnfsio(start_dscreate, tdsc);
4388 			NFSD_DEBUG(4, "nfsrv_pnfscreate: nfs_pnfsio=%d\n", ret);
4389 		}
4390 		if (ret != 0) {
4391 			ret = nfsrv_dscreate(dvp[i], vap, &va, &fh, tpf, NULL,
4392 			    NULL, tcred, p, NULL);
4393 			if (ret != 0) {
4394 				KASSERT(error == 0, ("nfsrv_dscreate err=%d",
4395 				    error));
4396 				if (failpos == -1 && nfsds_failerr(ret))
4397 					failpos = i;
4398 				else
4399 					error = ret;
4400 			}
4401 		}
4402 	}
4403 	if (error == 0) {
4404 		tpf->dsf_dir = dsdir[mirrorcnt - 1];
4405 		error = nfsrv_dscreate(dvp[mirrorcnt - 1], vap, &va, &fh, tpf,
4406 		    &dsattr, NULL, tcred, p, NULL);
4407 		if (failpos == -1 && mirrorcnt > 1 && nfsds_failerr(error)) {
4408 			failpos = mirrorcnt - 1;
4409 			error = 0;
4410 		}
4411 	}
4412 	timo = hz / 50;		/* Wait for 20msec. */
4413 	if (timo < 1)
4414 		timo = 1;
4415 	/* Wait for kernel task(s) to complete. */
4416 	for (tdsc = dsc, i = 0; i < mirrorcnt - 1; i++, tdsc++) {
4417 		while (tdsc->inprog != 0 && tdsc->done == 0)
4418 			tsleep(&tdsc->tsk, PVFS, "srvdcr", timo);
4419 		if (tdsc->err != 0) {
4420 			if (failpos == -1 && nfsds_failerr(tdsc->err))
4421 				failpos = i;
4422 			else if (error == 0)
4423 				error = tdsc->err;
4424 		}
4425 	}
4426 
4427 	/*
4428 	 * If failpos has been set, that mirror has failed, so it needs
4429 	 * to be disabled.
4430 	 */
4431 	if (failpos >= 0) {
4432 		nmp = VFSTONFS(dvp[failpos]->v_mount);
4433 		NFSLOCKMNT(nmp);
4434 		if ((nmp->nm_privflag & (NFSMNTP_FORCEDISM |
4435 		     NFSMNTP_CANCELRPCS)) == 0) {
4436 			nmp->nm_privflag |= NFSMNTP_CANCELRPCS;
4437 			NFSUNLOCKMNT(nmp);
4438 			ds = nfsrv_deldsnmp(PNFSDOP_DELDSSERVER, nmp, p);
4439 			NFSD_DEBUG(4, "dscreatfail fail=%d ds=%p\n", failpos,
4440 			    ds);
4441 			if (ds != NULL)
4442 				nfsrv_killrpcs(nmp);
4443 			NFSLOCKMNT(nmp);
4444 			nmp->nm_privflag &= ~NFSMNTP_CANCELRPCS;
4445 			wakeup(nmp);
4446 		}
4447 		NFSUNLOCKMNT(nmp);
4448 	}
4449 
4450 	NFSFREECRED(tcred);
4451 	if (error == 0) {
4452 		ASSERT_VOP_ELOCKED(vp, "nfsrv_pnfscreate vp");
4453 
4454 		NFSD_DEBUG(4, "nfsrv_pnfscreate: mirrorcnt=%d maxmirror=%d\n",
4455 		    mirrorcnt, nfsrv_maxpnfsmirror);
4456 		/*
4457 		 * For all mirrors that couldn't be created, fill in the
4458 		 * *pf structure, but with an IP address == 0.0.0.0.
4459 		 */
4460 		tpf = pf + mirrorcnt;
4461 		for (i = mirrorcnt; i < nfsrv_maxpnfsmirror; i++, tpf++) {
4462 			*tpf = *pf;
4463 			tpf->dsf_sin.sin_family = AF_INET;
4464 			tpf->dsf_sin.sin_len = sizeof(struct sockaddr_in);
4465 			tpf->dsf_sin.sin_addr.s_addr = 0;
4466 			tpf->dsf_sin.sin_port = 0;
4467 		}
4468 
4469 		error = vn_extattr_set(vp, IO_NODELOCKED,
4470 		    EXTATTR_NAMESPACE_SYSTEM, "pnfsd.dsfile",
4471 		    sizeof(*pf) * nfsrv_maxpnfsmirror, (char *)pf, p);
4472 		if (error == 0)
4473 			error = vn_extattr_set(vp, IO_NODELOCKED,
4474 			    EXTATTR_NAMESPACE_SYSTEM, "pnfsd.dsattr",
4475 			    sizeof(dsattr), (char *)&dsattr, p);
4476 		if (error != 0)
4477 			printf("pNFS: pnfscreate setextattr=%d\n",
4478 			    error);
4479 	} else
4480 		printf("pNFS: pnfscreate=%d\n", error);
4481 	free(pf, M_TEMP);
4482 	free(dsc, M_TEMP);
4483 }
4484 
4485 /*
4486  * Get the information needed to remove the pNFS Data Server file from the
4487  * Metadata file.  Upon success, ddvp is set non-NULL to the locked
4488  * DS directory vnode.  The caller must unlock *ddvp when done with it.
4489  */
4490 static void
4491 nfsrv_pnfsremovesetup(struct vnode *vp, NFSPROC_T *p, struct vnode **dvpp,
4492     int *mirrorcntp, char *fname, fhandle_t *fhp)
4493 {
4494 	struct vattr va;
4495 	struct ucred *tcred;
4496 	char *buf;
4497 	int buflen, error;
4498 
4499 	dvpp[0] = NULL;
4500 	/* If not an exported regular file or not a pNFS server, just return. */
4501 	if (vp->v_type != VREG || (vp->v_mount->mnt_flag & MNT_EXPORTED) == 0 ||
4502 	    nfsrv_devidcnt == 0)
4503 		return;
4504 
4505 	/* Check to see if this is the last hard link. */
4506 	tcred = newnfs_getcred();
4507 	error = VOP_GETATTR(vp, &va, tcred);
4508 	NFSFREECRED(tcred);
4509 	if (error != 0) {
4510 		printf("pNFS: nfsrv_pnfsremovesetup getattr=%d\n", error);
4511 		return;
4512 	}
4513 	if (va.va_nlink > 1)
4514 		return;
4515 
4516 	error = nfsvno_getfh(vp, fhp, p);
4517 	if (error != 0) {
4518 		printf("pNFS: nfsrv_pnfsremovesetup getfh=%d\n", error);
4519 		return;
4520 	}
4521 
4522 	buflen = 1024;
4523 	buf = malloc(buflen, M_TEMP, M_WAITOK);
4524 	/* Get the directory vnode for the DS mount and the file handle. */
4525 	error = nfsrv_dsgetsockmnt(vp, 0, buf, &buflen, mirrorcntp, p, dvpp,
4526 	    NULL, NULL, fname, NULL, NULL, NULL, NULL, NULL);
4527 	free(buf, M_TEMP);
4528 	if (error != 0)
4529 		printf("pNFS: nfsrv_pnfsremovesetup getsockmnt=%d\n", error);
4530 }
4531 
4532 /*
4533  * Remove a DS data file for nfsrv_pnfsremove(). Called for each mirror.
4534  * The arguments are in a structure, so that they can be passed through
4535  * taskqueue for a kernel process to execute this function.
4536  */
4537 struct nfsrvdsremove {
4538 	int			done;
4539 	int			inprog;
4540 	struct task		tsk;
4541 	struct ucred		*tcred;
4542 	struct vnode		*dvp;
4543 	NFSPROC_T		*p;
4544 	int			err;
4545 	char			fname[PNFS_FILENAME_LEN + 1];
4546 };
4547 
4548 static int
4549 nfsrv_dsremove(struct vnode *dvp, char *fname, struct ucred *tcred,
4550     NFSPROC_T *p)
4551 {
4552 	struct nameidata named;
4553 	struct vnode *nvp;
4554 	char *bufp;
4555 	u_long *hashp;
4556 	int error;
4557 
4558 	error = NFSVOPLOCK(dvp, LK_EXCLUSIVE);
4559 	if (error != 0)
4560 		return (error);
4561 	named.ni_cnd.cn_nameiop = DELETE;
4562 	named.ni_cnd.cn_lkflags = LK_EXCLUSIVE | LK_RETRY;
4563 	named.ni_cnd.cn_cred = tcred;
4564 	named.ni_cnd.cn_flags = ISLASTCN | LOCKPARENT | LOCKLEAF | SAVENAME;
4565 	nfsvno_setpathbuf(&named, &bufp, &hashp);
4566 	named.ni_cnd.cn_nameptr = bufp;
4567 	named.ni_cnd.cn_namelen = strlen(fname);
4568 	strlcpy(bufp, fname, NAME_MAX);
4569 	NFSD_DEBUG(4, "nfsrv_pnfsremove: filename=%s\n", bufp);
4570 	error = VOP_LOOKUP(dvp, &nvp, &named.ni_cnd);
4571 	NFSD_DEBUG(4, "nfsrv_pnfsremove: aft LOOKUP=%d\n", error);
4572 	if (error == 0) {
4573 		error = VOP_REMOVE(dvp, nvp, &named.ni_cnd);
4574 		vput(nvp);
4575 	}
4576 	NFSVOPUNLOCK(dvp);
4577 	nfsvno_relpathbuf(&named);
4578 	if (error != 0)
4579 		printf("pNFS: nfsrv_pnfsremove failed=%d\n", error);
4580 	return (error);
4581 }
4582 
4583 /*
4584  * Start up the thread that will execute nfsrv_dsremove().
4585  */
4586 static void
4587 start_dsremove(void *arg, int pending)
4588 {
4589 	struct nfsrvdsremove *dsrm;
4590 
4591 	dsrm = (struct nfsrvdsremove *)arg;
4592 	dsrm->err = nfsrv_dsremove(dsrm->dvp, dsrm->fname, dsrm->tcred,
4593 	    dsrm->p);
4594 	dsrm->done = 1;
4595 	NFSD_DEBUG(4, "start_dsremove: err=%d\n", dsrm->err);
4596 }
4597 
4598 /*
4599  * Remove a pNFS data file from a Data Server.
4600  * nfsrv_pnfsremovesetup() must have been called before the MDS file was
4601  * removed to set up the dvp and fill in the FH.
4602  */
4603 static void
4604 nfsrv_pnfsremove(struct vnode **dvp, int mirrorcnt, char *fname, fhandle_t *fhp,
4605     NFSPROC_T *p)
4606 {
4607 	struct ucred *tcred;
4608 	struct nfsrvdsremove *dsrm, *tdsrm;
4609 	struct nfsdevice *ds;
4610 	struct nfsmount *nmp;
4611 	int failpos, i, ret, timo;
4612 
4613 	tcred = newnfs_getcred();
4614 	dsrm = NULL;
4615 	if (mirrorcnt > 1)
4616 		dsrm = malloc(sizeof(*dsrm) * mirrorcnt - 1, M_TEMP, M_WAITOK);
4617 	/*
4618 	 * Remove the file on each DS mirror, using kernel process(es) for the
4619 	 * additional mirrors.
4620 	 */
4621 	failpos = -1;
4622 	for (tdsrm = dsrm, i = 0; i < mirrorcnt - 1; i++, tdsrm++) {
4623 		tdsrm->tcred = tcred;
4624 		tdsrm->p = p;
4625 		tdsrm->dvp = dvp[i];
4626 		strlcpy(tdsrm->fname, fname, PNFS_FILENAME_LEN + 1);
4627 		tdsrm->inprog = 0;
4628 		tdsrm->done = 0;
4629 		tdsrm->err = 0;
4630 		ret = EIO;
4631 		if (nfs_pnfsiothreads != 0) {
4632 			ret = nfs_pnfsio(start_dsremove, tdsrm);
4633 			NFSD_DEBUG(4, "nfsrv_pnfsremove: nfs_pnfsio=%d\n", ret);
4634 		}
4635 		if (ret != 0) {
4636 			ret = nfsrv_dsremove(dvp[i], fname, tcred, p);
4637 			if (failpos == -1 && nfsds_failerr(ret))
4638 				failpos = i;
4639 		}
4640 	}
4641 	ret = nfsrv_dsremove(dvp[mirrorcnt - 1], fname, tcred, p);
4642 	if (failpos == -1 && mirrorcnt > 1 && nfsds_failerr(ret))
4643 		failpos = mirrorcnt - 1;
4644 	timo = hz / 50;		/* Wait for 20msec. */
4645 	if (timo < 1)
4646 		timo = 1;
4647 	/* Wait for kernel task(s) to complete. */
4648 	for (tdsrm = dsrm, i = 0; i < mirrorcnt - 1; i++, tdsrm++) {
4649 		while (tdsrm->inprog != 0 && tdsrm->done == 0)
4650 			tsleep(&tdsrm->tsk, PVFS, "srvdsrm", timo);
4651 		if (failpos == -1 && nfsds_failerr(tdsrm->err))
4652 			failpos = i;
4653 	}
4654 
4655 	/*
4656 	 * If failpos has been set, that mirror has failed, so it needs
4657 	 * to be disabled.
4658 	 */
4659 	if (failpos >= 0) {
4660 		nmp = VFSTONFS(dvp[failpos]->v_mount);
4661 		NFSLOCKMNT(nmp);
4662 		if ((nmp->nm_privflag & (NFSMNTP_FORCEDISM |
4663 		     NFSMNTP_CANCELRPCS)) == 0) {
4664 			nmp->nm_privflag |= NFSMNTP_CANCELRPCS;
4665 			NFSUNLOCKMNT(nmp);
4666 			ds = nfsrv_deldsnmp(PNFSDOP_DELDSSERVER, nmp, p);
4667 			NFSD_DEBUG(4, "dsremovefail fail=%d ds=%p\n", failpos,
4668 			    ds);
4669 			if (ds != NULL)
4670 				nfsrv_killrpcs(nmp);
4671 			NFSLOCKMNT(nmp);
4672 			nmp->nm_privflag &= ~NFSMNTP_CANCELRPCS;
4673 			wakeup(nmp);
4674 		}
4675 		NFSUNLOCKMNT(nmp);
4676 	}
4677 
4678 	/* Get rid all layouts for the file. */
4679 	nfsrv_freefilelayouts(fhp);
4680 
4681 	NFSFREECRED(tcred);
4682 	free(dsrm, M_TEMP);
4683 }
4684 
4685 /*
4686  * Generate a file name based on the file handle and put it in *bufp.
4687  * Return the number of bytes generated.
4688  */
4689 static int
4690 nfsrv_putfhname(fhandle_t *fhp, char *bufp)
4691 {
4692 	int i;
4693 	uint8_t *cp;
4694 	const uint8_t *hexdigits = "0123456789abcdef";
4695 
4696 	cp = (uint8_t *)fhp;
4697 	for (i = 0; i < sizeof(*fhp); i++) {
4698 		bufp[2 * i] = hexdigits[(*cp >> 4) & 0xf];
4699 		bufp[2 * i + 1] = hexdigits[*cp++ & 0xf];
4700 	}
4701 	bufp[2 * i] = '\0';
4702 	return (2 * i);
4703 }
4704 
4705 /*
4706  * Update the Metadata file's attributes from the DS file when a Read/Write
4707  * layout is returned.
4708  * Basically just call nfsrv_proxyds() with procedure == NFSPROC_LAYOUTRETURN
4709  * so that it does a nfsrv_getattrdsrpc() and nfsrv_setextattr() on the DS file.
4710  */
4711 int
4712 nfsrv_updatemdsattr(struct vnode *vp, struct nfsvattr *nap, NFSPROC_T *p)
4713 {
4714 	struct ucred *tcred;
4715 	int error;
4716 
4717 	/* Do this as root so that it won't fail with EACCES. */
4718 	tcred = newnfs_getcred();
4719 	error = nfsrv_proxyds(vp, 0, 0, tcred, p, NFSPROC_LAYOUTRETURN,
4720 	    NULL, NULL, NULL, nap, NULL, NULL, 0, NULL);
4721 	NFSFREECRED(tcred);
4722 	return (error);
4723 }
4724 
4725 /*
4726  * Set the NFSv4 ACL on the DS file to the same ACL as the MDS file.
4727  */
4728 static int
4729 nfsrv_dssetacl(struct vnode *vp, struct acl *aclp, struct ucred *cred,
4730     NFSPROC_T *p)
4731 {
4732 	int error;
4733 
4734 	error = nfsrv_proxyds(vp, 0, 0, cred, p, NFSPROC_SETACL,
4735 	    NULL, NULL, NULL, NULL, aclp, NULL, 0, NULL);
4736 	return (error);
4737 }
4738 
4739 static int
4740 nfsrv_proxyds(struct vnode *vp, off_t off, int cnt, struct ucred *cred,
4741     struct thread *p, int ioproc, struct mbuf **mpp, char *cp,
4742     struct mbuf **mpp2, struct nfsvattr *nap, struct acl *aclp,
4743     off_t *offp, int content, bool *eofp)
4744 {
4745 	struct nfsmount *nmp[NFSDEV_MAXMIRRORS], *failnmp;
4746 	fhandle_t fh[NFSDEV_MAXMIRRORS];
4747 	struct vnode *dvp[NFSDEV_MAXMIRRORS];
4748 	struct nfsdevice *ds;
4749 	struct pnfsdsattr dsattr;
4750 	struct opnfsdsattr odsattr;
4751 	char *buf;
4752 	int buflen, error, failpos, i, mirrorcnt, origmircnt, trycnt;
4753 
4754 	NFSD_DEBUG(4, "in nfsrv_proxyds\n");
4755 	/*
4756 	 * If not a regular file, not exported or not a pNFS server,
4757 	 * just return ENOENT.
4758 	 */
4759 	if (vp->v_type != VREG || (vp->v_mount->mnt_flag & MNT_EXPORTED) == 0 ||
4760 	    nfsrv_devidcnt == 0)
4761 		return (ENOENT);
4762 
4763 	buflen = 1024;
4764 	buf = malloc(buflen, M_TEMP, M_WAITOK);
4765 	error = 0;
4766 
4767 	/*
4768 	 * For Getattr, get the Change attribute (va_filerev) and size (va_size)
4769 	 * from the MetaData file's extended attribute.
4770 	 */
4771 	if (ioproc == NFSPROC_GETATTR) {
4772 		error = vn_extattr_get(vp, IO_NODELOCKED,
4773 		    EXTATTR_NAMESPACE_SYSTEM, "pnfsd.dsattr", &buflen, buf,
4774 		    p);
4775 		if (error == 0) {
4776 			if (buflen == sizeof(odsattr)) {
4777 				NFSBCOPY(buf, &odsattr, buflen);
4778 				nap->na_filerev = odsattr.dsa_filerev;
4779 				nap->na_size = odsattr.dsa_size;
4780 				nap->na_atime = odsattr.dsa_atime;
4781 				nap->na_mtime = odsattr.dsa_mtime;
4782 				/*
4783 				 * Fake na_bytes by rounding up na_size.
4784 				 * Since we don't know the block size, just
4785 				 * use BLKDEV_IOSIZE.
4786 				 */
4787 				nap->na_bytes = (odsattr.dsa_size +
4788 				    BLKDEV_IOSIZE - 1) & ~(BLKDEV_IOSIZE - 1);
4789 			} else if (buflen == sizeof(dsattr)) {
4790 				NFSBCOPY(buf, &dsattr, buflen);
4791 				nap->na_filerev = dsattr.dsa_filerev;
4792 				nap->na_size = dsattr.dsa_size;
4793 				nap->na_atime = dsattr.dsa_atime;
4794 				nap->na_mtime = dsattr.dsa_mtime;
4795 				nap->na_bytes = dsattr.dsa_bytes;
4796 			} else
4797 				error = ENXIO;
4798 		}
4799 		if (error == 0) {
4800 			/*
4801 			 * If nfsrv_pnfsgetdsattr is 0 or nfsrv_checkdsattr()
4802 			 * returns 0, just return now.  nfsrv_checkdsattr()
4803 			 * returns 0 if there is no Read/Write layout
4804 			 * plus either an Open/Write_access or Write
4805 			 * delegation issued to a client for the file.
4806 			 */
4807 			if (nfsrv_pnfsgetdsattr == 0 ||
4808 			    nfsrv_checkdsattr(vp, p) == 0) {
4809 				free(buf, M_TEMP);
4810 				return (error);
4811 			}
4812 		}
4813 
4814 		/*
4815 		 * Clear ENOATTR so the code below will attempt to do a
4816 		 * nfsrv_getattrdsrpc() to get the attributes and (re)create
4817 		 * the extended attribute.
4818 		 */
4819 		if (error == ENOATTR)
4820 			error = 0;
4821 	}
4822 
4823 	origmircnt = -1;
4824 	trycnt = 0;
4825 tryagain:
4826 	if (error == 0) {
4827 		buflen = 1024;
4828 		if (ioproc == NFSPROC_READDS && NFSVOPISLOCKED(vp) ==
4829 		    LK_EXCLUSIVE)
4830 			printf("nfsrv_proxyds: Readds vp exclusively locked\n");
4831 		error = nfsrv_dsgetsockmnt(vp, LK_SHARED, buf, &buflen,
4832 		    &mirrorcnt, p, dvp, fh, NULL, NULL, NULL, NULL, NULL,
4833 		    NULL, NULL);
4834 		if (error == 0) {
4835 			for (i = 0; i < mirrorcnt; i++)
4836 				nmp[i] = VFSTONFS(dvp[i]->v_mount);
4837 		} else
4838 			printf("pNFS: proxy getextattr sockaddr=%d\n", error);
4839 	} else
4840 		printf("pNFS: nfsrv_dsgetsockmnt=%d\n", error);
4841 	if (error == 0) {
4842 		failpos = -1;
4843 		if (origmircnt == -1)
4844 			origmircnt = mirrorcnt;
4845 		/*
4846 		 * If failpos is set to a mirror#, then that mirror has
4847 		 * failed and will be disabled. For Read, Getattr and Seek, the
4848 		 * function only tries one mirror, so if that mirror has
4849 		 * failed, it will need to be retried. As such, increment
4850 		 * tryitagain for these cases.
4851 		 * For Write, Setattr and Setacl, the function tries all
4852 		 * mirrors and will not return an error for the case where
4853 		 * one mirror has failed. For these cases, the functioning
4854 		 * mirror(s) will have been modified, so a retry isn't
4855 		 * necessary. These functions will set failpos for the
4856 		 * failed mirror#.
4857 		 */
4858 		if (ioproc == NFSPROC_READDS) {
4859 			error = nfsrv_readdsrpc(fh, off, cnt, cred, p, nmp[0],
4860 			    mpp, mpp2);
4861 			if (nfsds_failerr(error) && mirrorcnt > 1) {
4862 				/*
4863 				 * Setting failpos will cause the mirror
4864 				 * to be disabled and then a retry of this
4865 				 * read is required.
4866 				 */
4867 				failpos = 0;
4868 				error = 0;
4869 				trycnt++;
4870 			}
4871 		} else if (ioproc == NFSPROC_WRITEDS)
4872 			error = nfsrv_writedsrpc(fh, off, cnt, cred, p, vp,
4873 			    &nmp[0], mirrorcnt, mpp, cp, &failpos);
4874 		else if (ioproc == NFSPROC_SETATTR)
4875 			error = nfsrv_setattrdsrpc(fh, cred, p, vp, &nmp[0],
4876 			    mirrorcnt, nap, &failpos);
4877 		else if (ioproc == NFSPROC_SETACL)
4878 			error = nfsrv_setacldsrpc(fh, cred, p, vp, &nmp[0],
4879 			    mirrorcnt, aclp, &failpos);
4880 		else if (ioproc == NFSPROC_SEEKDS) {
4881 			error = nfsrv_seekdsrpc(fh, offp, content, eofp, cred,
4882 			    p, nmp[0]);
4883 			if (nfsds_failerr(error) && mirrorcnt > 1) {
4884 				/*
4885 				 * Setting failpos will cause the mirror
4886 				 * to be disabled and then a retry of this
4887 				 * read is required.
4888 				 */
4889 				failpos = 0;
4890 				error = 0;
4891 				trycnt++;
4892 			}
4893 		} else if (ioproc == NFSPROC_ALLOCATE)
4894 			error = nfsrv_allocatedsrpc(fh, off, *offp, cred, p, vp,
4895 			    &nmp[0], mirrorcnt, &failpos);
4896 		else if (ioproc == NFSPROC_DEALLOCATE)
4897 			error = nfsrv_deallocatedsrpc(fh, off, *offp, cred, p,
4898 			    vp, &nmp[0], mirrorcnt, &failpos);
4899 		else {
4900 			error = nfsrv_getattrdsrpc(&fh[mirrorcnt - 1], cred, p,
4901 			    vp, nmp[mirrorcnt - 1], nap);
4902 			if (nfsds_failerr(error) && mirrorcnt > 1) {
4903 				/*
4904 				 * Setting failpos will cause the mirror
4905 				 * to be disabled and then a retry of this
4906 				 * getattr is required.
4907 				 */
4908 				failpos = mirrorcnt - 1;
4909 				error = 0;
4910 				trycnt++;
4911 			}
4912 		}
4913 		ds = NULL;
4914 		if (failpos >= 0) {
4915 			failnmp = nmp[failpos];
4916 			NFSLOCKMNT(failnmp);
4917 			if ((failnmp->nm_privflag & (NFSMNTP_FORCEDISM |
4918 			     NFSMNTP_CANCELRPCS)) == 0) {
4919 				failnmp->nm_privflag |= NFSMNTP_CANCELRPCS;
4920 				NFSUNLOCKMNT(failnmp);
4921 				ds = nfsrv_deldsnmp(PNFSDOP_DELDSSERVER,
4922 				    failnmp, p);
4923 				NFSD_DEBUG(4, "dsldsnmp fail=%d ds=%p\n",
4924 				    failpos, ds);
4925 				if (ds != NULL)
4926 					nfsrv_killrpcs(failnmp);
4927 				NFSLOCKMNT(failnmp);
4928 				failnmp->nm_privflag &= ~NFSMNTP_CANCELRPCS;
4929 				wakeup(failnmp);
4930 			}
4931 			NFSUNLOCKMNT(failnmp);
4932 		}
4933 		for (i = 0; i < mirrorcnt; i++)
4934 			NFSVOPUNLOCK(dvp[i]);
4935 		NFSD_DEBUG(4, "nfsrv_proxyds: aft RPC=%d trya=%d\n", error,
4936 		    trycnt);
4937 		/* Try the Read/Getattr again if a mirror was deleted. */
4938 		if (ds != NULL && trycnt > 0 && trycnt < origmircnt)
4939 			goto tryagain;
4940 	} else {
4941 		/* Return ENOENT for any Extended Attribute error. */
4942 		error = ENOENT;
4943 	}
4944 	free(buf, M_TEMP);
4945 	NFSD_DEBUG(4, "nfsrv_proxyds: error=%d\n", error);
4946 	return (error);
4947 }
4948 
4949 /*
4950  * Get the DS mount point, fh and directory from the "pnfsd.dsfile" extended
4951  * attribute.
4952  * newnmpp - If it points to a non-NULL nmp, that is the destination and needs
4953  *           to be checked.  If it points to a NULL nmp, then it returns
4954  *           a suitable destination.
4955  * curnmp - If non-NULL, it is the source mount for the copy.
4956  */
4957 int
4958 nfsrv_dsgetsockmnt(struct vnode *vp, int lktype, char *buf, int *buflenp,
4959     int *mirrorcntp, NFSPROC_T *p, struct vnode **dvpp, fhandle_t *fhp,
4960     char *devid, char *fnamep, struct vnode **nvpp, struct nfsmount **newnmpp,
4961     struct nfsmount *curnmp, int *ippos, int *dsdirp)
4962 {
4963 	struct vnode *dvp, *nvp = NULL, **tdvpp;
4964 	struct mount *mp;
4965 	struct nfsmount *nmp, *newnmp;
4966 	struct sockaddr *sad;
4967 	struct sockaddr_in *sin;
4968 	struct nfsdevice *ds, *tds, *fndds;
4969 	struct pnfsdsfile *pf;
4970 	uint32_t dsdir;
4971 	int error, fhiszero, fnd, gotone, i, mirrorcnt;
4972 
4973 	ASSERT_VOP_LOCKED(vp, "nfsrv_dsgetsockmnt vp");
4974 	*mirrorcntp = 1;
4975 	tdvpp = dvpp;
4976 	if (nvpp != NULL)
4977 		*nvpp = NULL;
4978 	if (dvpp != NULL)
4979 		*dvpp = NULL;
4980 	if (ippos != NULL)
4981 		*ippos = -1;
4982 	if (newnmpp != NULL)
4983 		newnmp = *newnmpp;
4984 	else
4985 		newnmp = NULL;
4986 	mp = vp->v_mount;
4987 	error = vn_extattr_get(vp, IO_NODELOCKED, EXTATTR_NAMESPACE_SYSTEM,
4988 	    "pnfsd.dsfile", buflenp, buf, p);
4989 	mirrorcnt = *buflenp / sizeof(*pf);
4990 	if (error == 0 && (mirrorcnt < 1 || mirrorcnt > NFSDEV_MAXMIRRORS ||
4991 	    *buflenp != sizeof(*pf) * mirrorcnt))
4992 		error = ENOATTR;
4993 
4994 	pf = (struct pnfsdsfile *)buf;
4995 	/* If curnmp != NULL, check for a match in the mirror list. */
4996 	if (curnmp != NULL && error == 0) {
4997 		fnd = 0;
4998 		for (i = 0; i < mirrorcnt; i++, pf++) {
4999 			sad = (struct sockaddr *)&pf->dsf_sin;
5000 			if (nfsaddr2_match(sad, curnmp->nm_nam)) {
5001 				if (ippos != NULL)
5002 					*ippos = i;
5003 				fnd = 1;
5004 				break;
5005 			}
5006 		}
5007 		if (fnd == 0)
5008 			error = ENXIO;
5009 	}
5010 
5011 	gotone = 0;
5012 	pf = (struct pnfsdsfile *)buf;
5013 	NFSD_DEBUG(4, "nfsrv_dsgetsockmnt: mirrorcnt=%d err=%d\n", mirrorcnt,
5014 	    error);
5015 	for (i = 0; i < mirrorcnt && error == 0; i++, pf++) {
5016 		fhiszero = 0;
5017 		sad = (struct sockaddr *)&pf->dsf_sin;
5018 		sin = &pf->dsf_sin;
5019 		dsdir = pf->dsf_dir;
5020 		if (dsdir >= nfsrv_dsdirsize) {
5021 			printf("nfsrv_dsgetsockmnt: dsdir=%d\n", dsdir);
5022 			error = ENOATTR;
5023 		} else if (nvpp != NULL && newnmp != NULL &&
5024 		    nfsaddr2_match(sad, newnmp->nm_nam))
5025 			error = EEXIST;
5026 		if (error == 0) {
5027 			if (ippos != NULL && curnmp == NULL &&
5028 			    sad->sa_family == AF_INET &&
5029 			    sin->sin_addr.s_addr == 0)
5030 				*ippos = i;
5031 			if (NFSBCMP(&zerofh, &pf->dsf_fh, sizeof(zerofh)) == 0)
5032 				fhiszero = 1;
5033 			/* Use the socket address to find the mount point. */
5034 			fndds = NULL;
5035 			NFSDDSLOCK();
5036 			/* Find a match for the IP address. */
5037 			TAILQ_FOREACH(ds, &nfsrv_devidhead, nfsdev_list) {
5038 				if (ds->nfsdev_nmp != NULL) {
5039 					dvp = ds->nfsdev_dvp;
5040 					nmp = VFSTONFS(dvp->v_mount);
5041 					if (nmp != ds->nfsdev_nmp)
5042 						printf("different2 nmp %p %p\n",
5043 						    nmp, ds->nfsdev_nmp);
5044 					if (nfsaddr2_match(sad, nmp->nm_nam)) {
5045 						fndds = ds;
5046 						break;
5047 					}
5048 				}
5049 			}
5050 			if (fndds != NULL && newnmpp != NULL &&
5051 			    newnmp == NULL) {
5052 				/* Search for a place to make a mirror copy. */
5053 				TAILQ_FOREACH(tds, &nfsrv_devidhead,
5054 				    nfsdev_list) {
5055 					if (tds->nfsdev_nmp != NULL &&
5056 					    fndds != tds &&
5057 					    ((tds->nfsdev_mdsisset == 0 &&
5058 					      fndds->nfsdev_mdsisset == 0) ||
5059 					     (tds->nfsdev_mdsisset != 0 &&
5060 					      fndds->nfsdev_mdsisset != 0 &&
5061 					      fsidcmp(&tds->nfsdev_mdsfsid,
5062 					      &mp->mnt_stat.f_fsid) == 0))) {
5063 						*newnmpp = tds->nfsdev_nmp;
5064 						break;
5065 					}
5066 				}
5067 				if (tds != NULL) {
5068 					/*
5069 					 * Move this entry to the end of the
5070 					 * list, so it won't be selected as
5071 					 * easily the next time.
5072 					 */
5073 					TAILQ_REMOVE(&nfsrv_devidhead, tds,
5074 					    nfsdev_list);
5075 					TAILQ_INSERT_TAIL(&nfsrv_devidhead, tds,
5076 					    nfsdev_list);
5077 				}
5078 			}
5079 			NFSDDSUNLOCK();
5080 			if (fndds != NULL) {
5081 				dvp = fndds->nfsdev_dsdir[dsdir];
5082 				if (lktype != 0 || fhiszero != 0 ||
5083 				    (nvpp != NULL && *nvpp == NULL)) {
5084 					if (fhiszero != 0)
5085 						error = vn_lock(dvp,
5086 						    LK_EXCLUSIVE);
5087 					else if (lktype != 0)
5088 						error = vn_lock(dvp, lktype);
5089 					else
5090 						error = vn_lock(dvp, LK_SHARED);
5091 					/*
5092 					 * If the file handle is all 0's, try to
5093 					 * do a Lookup against the DS to acquire
5094 					 * it.
5095 					 * If dvpp == NULL or the Lookup fails,
5096 					 * unlock dvp after the call.
5097 					 */
5098 					if (error == 0 && (fhiszero != 0 ||
5099 					    (nvpp != NULL && *nvpp == NULL))) {
5100 						error = nfsrv_pnfslookupds(vp,
5101 						    dvp, pf, &nvp, p);
5102 						if (error == 0) {
5103 							if (fhiszero != 0)
5104 								nfsrv_pnfssetfh(
5105 								    vp, pf,
5106 								    devid,
5107 								    fnamep,
5108 								    nvp, p);
5109 							if (nvpp != NULL &&
5110 							    *nvpp == NULL) {
5111 								*nvpp = nvp;
5112 								*dsdirp = dsdir;
5113 							} else
5114 								vput(nvp);
5115 						}
5116 						if (error != 0 || lktype == 0)
5117 							NFSVOPUNLOCK(dvp);
5118 					}
5119 				}
5120 				if (error == 0) {
5121 					gotone++;
5122 					NFSD_DEBUG(4, "gotone=%d\n", gotone);
5123 					if (devid != NULL) {
5124 						NFSBCOPY(fndds->nfsdev_deviceid,
5125 						    devid, NFSX_V4DEVICEID);
5126 						devid += NFSX_V4DEVICEID;
5127 					}
5128 					if (dvpp != NULL)
5129 						*tdvpp++ = dvp;
5130 					if (fhp != NULL)
5131 						NFSBCOPY(&pf->dsf_fh, fhp++,
5132 						    NFSX_MYFH);
5133 					if (fnamep != NULL && gotone == 1)
5134 						strlcpy(fnamep,
5135 						    pf->dsf_filename,
5136 						    sizeof(pf->dsf_filename));
5137 				} else
5138 					NFSD_DEBUG(4, "nfsrv_dsgetsockmnt "
5139 					    "err=%d\n", error);
5140 			}
5141 		}
5142 	}
5143 	if (error == 0 && gotone == 0)
5144 		error = ENOENT;
5145 
5146 	NFSD_DEBUG(4, "eo nfsrv_dsgetsockmnt: gotone=%d err=%d\n", gotone,
5147 	    error);
5148 	if (error == 0)
5149 		*mirrorcntp = gotone;
5150 	else {
5151 		if (gotone > 0 && dvpp != NULL) {
5152 			/*
5153 			 * If the error didn't occur on the first one and
5154 			 * dvpp != NULL, the one(s) prior to the failure will
5155 			 * have locked dvp's that need to be unlocked.
5156 			 */
5157 			for (i = 0; i < gotone; i++) {
5158 				NFSVOPUNLOCK(*dvpp);
5159 				*dvpp++ = NULL;
5160 			}
5161 		}
5162 		/*
5163 		 * If it found the vnode to be copied from before a failure,
5164 		 * it needs to be vput()'d.
5165 		 */
5166 		if (nvpp != NULL && *nvpp != NULL) {
5167 			vput(*nvpp);
5168 			*nvpp = NULL;
5169 		}
5170 	}
5171 	return (error);
5172 }
5173 
5174 /*
5175  * Set the extended attribute for the Change attribute.
5176  */
5177 static int
5178 nfsrv_setextattr(struct vnode *vp, struct nfsvattr *nap, NFSPROC_T *p)
5179 {
5180 	struct pnfsdsattr dsattr;
5181 	int error;
5182 
5183 	ASSERT_VOP_ELOCKED(vp, "nfsrv_setextattr vp");
5184 	dsattr.dsa_filerev = nap->na_filerev;
5185 	dsattr.dsa_size = nap->na_size;
5186 	dsattr.dsa_atime = nap->na_atime;
5187 	dsattr.dsa_mtime = nap->na_mtime;
5188 	dsattr.dsa_bytes = nap->na_bytes;
5189 	error = vn_extattr_set(vp, IO_NODELOCKED, EXTATTR_NAMESPACE_SYSTEM,
5190 	    "pnfsd.dsattr", sizeof(dsattr), (char *)&dsattr, p);
5191 	if (error != 0)
5192 		printf("pNFS: setextattr=%d\n", error);
5193 	return (error);
5194 }
5195 
5196 static int
5197 nfsrv_readdsrpc(fhandle_t *fhp, off_t off, int len, struct ucred *cred,
5198     NFSPROC_T *p, struct nfsmount *nmp, struct mbuf **mpp, struct mbuf **mpendp)
5199 {
5200 	uint32_t *tl;
5201 	struct nfsrv_descript *nd;
5202 	nfsv4stateid_t st;
5203 	struct mbuf *m, *m2;
5204 	int error = 0, retlen, tlen, trimlen;
5205 
5206 	NFSD_DEBUG(4, "in nfsrv_readdsrpc\n");
5207 	nd = malloc(sizeof(*nd), M_TEMP, M_WAITOK | M_ZERO);
5208 	*mpp = NULL;
5209 	/*
5210 	 * Use a stateid where other is an alternating 01010 pattern and
5211 	 * seqid is 0xffffffff.  This value is not defined as special by
5212 	 * the RFC and is used by the FreeBSD NFS server to indicate an
5213 	 * MDS->DS proxy operation.
5214 	 */
5215 	st.other[0] = 0x55555555;
5216 	st.other[1] = 0x55555555;
5217 	st.other[2] = 0x55555555;
5218 	st.seqid = 0xffffffff;
5219 	nfscl_reqstart(nd, NFSPROC_READDS, nmp, (u_int8_t *)fhp, sizeof(*fhp),
5220 	    NULL, NULL, 0, 0);
5221 	nfsm_stateidtom(nd, &st, NFSSTATEID_PUTSTATEID);
5222 	NFSM_BUILD(tl, uint32_t *, NFSX_UNSIGNED * 3);
5223 	txdr_hyper(off, tl);
5224 	*(tl + 2) = txdr_unsigned(len);
5225 	error = newnfs_request(nd, nmp, NULL, &nmp->nm_sockreq, NULL, p, cred,
5226 	    NFS_PROG, NFS_VER4, NULL, 1, NULL, NULL);
5227 	if (error != 0) {
5228 		free(nd, M_TEMP);
5229 		return (error);
5230 	}
5231 	if (nd->nd_repstat == 0) {
5232 		NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
5233 		NFSM_STRSIZ(retlen, len);
5234 		if (retlen > 0) {
5235 			/* Trim off the pre-data XDR from the mbuf chain. */
5236 			m = nd->nd_mrep;
5237 			while (m != NULL && m != nd->nd_md) {
5238 				if (m->m_next == nd->nd_md) {
5239 					m->m_next = NULL;
5240 					m_freem(nd->nd_mrep);
5241 					nd->nd_mrep = m = nd->nd_md;
5242 				} else
5243 					m = m->m_next;
5244 			}
5245 			if (m == NULL) {
5246 				printf("nfsrv_readdsrpc: busted mbuf list\n");
5247 				error = ENOENT;
5248 				goto nfsmout;
5249 			}
5250 
5251 			/*
5252 			 * Now, adjust first mbuf so that any XDR before the
5253 			 * read data is skipped over.
5254 			 */
5255 			trimlen = nd->nd_dpos - mtod(m, char *);
5256 			if (trimlen > 0) {
5257 				m->m_len -= trimlen;
5258 				NFSM_DATAP(m, trimlen);
5259 			}
5260 
5261 			/*
5262 			 * Truncate the mbuf chain at retlen bytes of data,
5263 			 * plus XDR padding that brings the length up to a
5264 			 * multiple of 4.
5265 			 */
5266 			tlen = NFSM_RNDUP(retlen);
5267 			do {
5268 				if (m->m_len >= tlen) {
5269 					m->m_len = tlen;
5270 					tlen = 0;
5271 					m2 = m->m_next;
5272 					m->m_next = NULL;
5273 					m_freem(m2);
5274 					break;
5275 				}
5276 				tlen -= m->m_len;
5277 				m = m->m_next;
5278 			} while (m != NULL);
5279 			if (tlen > 0) {
5280 				printf("nfsrv_readdsrpc: busted mbuf list\n");
5281 				error = ENOENT;
5282 				goto nfsmout;
5283 			}
5284 			*mpp = nd->nd_mrep;
5285 			*mpendp = m;
5286 			nd->nd_mrep = NULL;
5287 		}
5288 	} else
5289 		error = nd->nd_repstat;
5290 nfsmout:
5291 	/* If nd->nd_mrep is already NULL, this is a no-op. */
5292 	m_freem(nd->nd_mrep);
5293 	free(nd, M_TEMP);
5294 	NFSD_DEBUG(4, "nfsrv_readdsrpc error=%d\n", error);
5295 	return (error);
5296 }
5297 
5298 /*
5299  * Do a write RPC on a DS data file, using this structure for the arguments,
5300  * so that this function can be executed by a separate kernel process.
5301  */
5302 struct nfsrvwritedsdorpc {
5303 	int			done;
5304 	int			inprog;
5305 	struct task		tsk;
5306 	fhandle_t		fh;
5307 	off_t			off;
5308 	int			len;
5309 	struct nfsmount		*nmp;
5310 	struct ucred		*cred;
5311 	NFSPROC_T		*p;
5312 	struct mbuf		*m;
5313 	int			err;
5314 };
5315 
5316 static int
5317 nfsrv_writedsdorpc(struct nfsmount *nmp, fhandle_t *fhp, off_t off, int len,
5318     struct nfsvattr *nap, struct mbuf *m, struct ucred *cred, NFSPROC_T *p)
5319 {
5320 	uint32_t *tl;
5321 	struct nfsrv_descript *nd;
5322 	nfsattrbit_t attrbits;
5323 	nfsv4stateid_t st;
5324 	int commit, error, retlen;
5325 
5326 	nd = malloc(sizeof(*nd), M_TEMP, M_WAITOK | M_ZERO);
5327 	nfscl_reqstart(nd, NFSPROC_WRITE, nmp, (u_int8_t *)fhp,
5328 	    sizeof(fhandle_t), NULL, NULL, 0, 0);
5329 
5330 	/*
5331 	 * Use a stateid where other is an alternating 01010 pattern and
5332 	 * seqid is 0xffffffff.  This value is not defined as special by
5333 	 * the RFC and is used by the FreeBSD NFS server to indicate an
5334 	 * MDS->DS proxy operation.
5335 	 */
5336 	st.other[0] = 0x55555555;
5337 	st.other[1] = 0x55555555;
5338 	st.other[2] = 0x55555555;
5339 	st.seqid = 0xffffffff;
5340 	nfsm_stateidtom(nd, &st, NFSSTATEID_PUTSTATEID);
5341 	NFSM_BUILD(tl, u_int32_t *, NFSX_HYPER + 2 * NFSX_UNSIGNED);
5342 	txdr_hyper(off, tl);
5343 	tl += 2;
5344 	/*
5345 	 * Do all writes FileSync, since the server doesn't hold onto dirty
5346 	 * buffers.  Since clients should be accessing the DS servers directly
5347 	 * using the pNFS layouts, this just needs to work correctly as a
5348 	 * fallback.
5349 	 */
5350 	*tl++ = txdr_unsigned(NFSWRITE_FILESYNC);
5351 	*tl = txdr_unsigned(len);
5352 	NFSD_DEBUG(4, "nfsrv_writedsdorpc: len=%d\n", len);
5353 
5354 	/* Put data in mbuf chain. */
5355 	nd->nd_mb->m_next = m;
5356 
5357 	/* Set nd_mb and nd_bpos to end of data. */
5358 	while (m->m_next != NULL)
5359 		m = m->m_next;
5360 	nd->nd_mb = m;
5361 	nfsm_set(nd, m->m_len);
5362 	NFSD_DEBUG(4, "nfsrv_writedsdorpc: lastmb len=%d\n", m->m_len);
5363 
5364 	/* Do a Getattr for the attributes that change upon writing. */
5365 	NFSZERO_ATTRBIT(&attrbits);
5366 	NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_SIZE);
5367 	NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_CHANGE);
5368 	NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_TIMEACCESS);
5369 	NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_TIMEMODIFY);
5370 	NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_SPACEUSED);
5371 	NFSM_BUILD(tl, u_int32_t *, NFSX_UNSIGNED);
5372 	*tl = txdr_unsigned(NFSV4OP_GETATTR);
5373 	(void) nfsrv_putattrbit(nd, &attrbits);
5374 	error = newnfs_request(nd, nmp, NULL, &nmp->nm_sockreq, NULL, p,
5375 	    cred, NFS_PROG, NFS_VER4, NULL, 1, NULL, NULL);
5376 	if (error != 0) {
5377 		free(nd, M_TEMP);
5378 		return (error);
5379 	}
5380 	NFSD_DEBUG(4, "nfsrv_writedsdorpc: aft writerpc=%d\n", nd->nd_repstat);
5381 	/* Get rid of weak cache consistency data for now. */
5382 	if ((nd->nd_flag & (ND_NOMOREDATA | ND_NFSV4 | ND_V4WCCATTR)) ==
5383 	    (ND_NFSV4 | ND_V4WCCATTR)) {
5384 		error = nfsv4_loadattr(nd, NULL, nap, NULL, NULL, 0, NULL, NULL,
5385 		    NULL, NULL, NULL, 0, NULL, NULL, NULL, NULL, NULL);
5386 		NFSD_DEBUG(4, "nfsrv_writedsdorpc: wcc attr=%d\n", error);
5387 		if (error != 0)
5388 			goto nfsmout;
5389 		/*
5390 		 * Get rid of Op# and status for next op.
5391 		 */
5392 		NFSM_DISSECT(tl, uint32_t *, 2 * NFSX_UNSIGNED);
5393 		if (*++tl != 0)
5394 			nd->nd_flag |= ND_NOMOREDATA;
5395 	}
5396 	if (nd->nd_repstat == 0) {
5397 		NFSM_DISSECT(tl, uint32_t *, 2 * NFSX_UNSIGNED + NFSX_VERF);
5398 		retlen = fxdr_unsigned(int, *tl++);
5399 		commit = fxdr_unsigned(int, *tl);
5400 		if (commit != NFSWRITE_FILESYNC)
5401 			error = NFSERR_IO;
5402 		NFSD_DEBUG(4, "nfsrv_writedsdorpc:retlen=%d commit=%d err=%d\n",
5403 		    retlen, commit, error);
5404 	} else
5405 		error = nd->nd_repstat;
5406 	/* We have no use for the Write Verifier since we use FileSync. */
5407 
5408 	/*
5409 	 * Get the Change, Size, Access Time and Modify Time attributes and set
5410 	 * on the Metadata file, so its attributes will be what the file's
5411 	 * would be if it had been written.
5412 	 */
5413 	if (error == 0) {
5414 		NFSM_DISSECT(tl, uint32_t *, 2 * NFSX_UNSIGNED);
5415 		error = nfsv4_loadattr(nd, NULL, nap, NULL, NULL, 0, NULL, NULL,
5416 		    NULL, NULL, NULL, 0, NULL, NULL, NULL, NULL, NULL);
5417 	}
5418 	NFSD_DEBUG(4, "nfsrv_writedsdorpc: aft loadattr=%d\n", error);
5419 nfsmout:
5420 	m_freem(nd->nd_mrep);
5421 	free(nd, M_TEMP);
5422 	NFSD_DEBUG(4, "nfsrv_writedsdorpc error=%d\n", error);
5423 	return (error);
5424 }
5425 
5426 /*
5427  * Start up the thread that will execute nfsrv_writedsdorpc().
5428  */
5429 static void
5430 start_writedsdorpc(void *arg, int pending)
5431 {
5432 	struct nfsrvwritedsdorpc *drpc;
5433 
5434 	drpc = (struct nfsrvwritedsdorpc *)arg;
5435 	drpc->err = nfsrv_writedsdorpc(drpc->nmp, &drpc->fh, drpc->off,
5436 	    drpc->len, NULL, drpc->m, drpc->cred, drpc->p);
5437 	drpc->done = 1;
5438 	NFSD_DEBUG(4, "start_writedsdorpc: err=%d\n", drpc->err);
5439 }
5440 
5441 static int
5442 nfsrv_writedsrpc(fhandle_t *fhp, off_t off, int len, struct ucred *cred,
5443     NFSPROC_T *p, struct vnode *vp, struct nfsmount **nmpp, int mirrorcnt,
5444     struct mbuf **mpp, char *cp, int *failposp)
5445 {
5446 	struct nfsrvwritedsdorpc *drpc, *tdrpc = NULL;
5447 	struct nfsvattr na;
5448 	struct mbuf *m;
5449 	int error, i, offs, ret, timo;
5450 
5451 	NFSD_DEBUG(4, "in nfsrv_writedsrpc\n");
5452 	KASSERT(*mpp != NULL, ("nfsrv_writedsrpc: NULL mbuf chain"));
5453 	drpc = NULL;
5454 	if (mirrorcnt > 1)
5455 		tdrpc = drpc = malloc(sizeof(*drpc) * (mirrorcnt - 1), M_TEMP,
5456 		    M_WAITOK);
5457 
5458 	/* Calculate offset in mbuf chain that data starts. */
5459 	offs = cp - mtod(*mpp, char *);
5460 	NFSD_DEBUG(4, "nfsrv_writedsrpc: mcopy offs=%d len=%d\n", offs, len);
5461 
5462 	/*
5463 	 * Do the write RPC for every DS, using a separate kernel process
5464 	 * for every DS except the last one.
5465 	 */
5466 	error = 0;
5467 	for (i = 0; i < mirrorcnt - 1; i++, tdrpc++) {
5468 		tdrpc->done = 0;
5469 		NFSBCOPY(fhp, &tdrpc->fh, sizeof(*fhp));
5470 		tdrpc->off = off;
5471 		tdrpc->len = len;
5472 		tdrpc->nmp = *nmpp;
5473 		tdrpc->cred = cred;
5474 		tdrpc->p = p;
5475 		tdrpc->inprog = 0;
5476 		tdrpc->err = 0;
5477 		tdrpc->m = m_copym(*mpp, offs, NFSM_RNDUP(len), M_WAITOK);
5478 		ret = EIO;
5479 		if (nfs_pnfsiothreads != 0) {
5480 			ret = nfs_pnfsio(start_writedsdorpc, tdrpc);
5481 			NFSD_DEBUG(4, "nfsrv_writedsrpc: nfs_pnfsio=%d\n",
5482 			    ret);
5483 		}
5484 		if (ret != 0) {
5485 			ret = nfsrv_writedsdorpc(*nmpp, fhp, off, len, NULL,
5486 			    tdrpc->m, cred, p);
5487 			if (nfsds_failerr(ret) && *failposp == -1)
5488 				*failposp = i;
5489 			else if (error == 0 && ret != 0)
5490 				error = ret;
5491 		}
5492 		nmpp++;
5493 		fhp++;
5494 	}
5495 	m = m_copym(*mpp, offs, NFSM_RNDUP(len), M_WAITOK);
5496 	ret = nfsrv_writedsdorpc(*nmpp, fhp, off, len, &na, m, cred, p);
5497 	if (nfsds_failerr(ret) && *failposp == -1 && mirrorcnt > 1)
5498 		*failposp = mirrorcnt - 1;
5499 	else if (error == 0 && ret != 0)
5500 		error = ret;
5501 	if (error == 0)
5502 		error = nfsrv_setextattr(vp, &na, p);
5503 	NFSD_DEBUG(4, "nfsrv_writedsrpc: aft setextat=%d\n", error);
5504 	tdrpc = drpc;
5505 	timo = hz / 50;		/* Wait for 20msec. */
5506 	if (timo < 1)
5507 		timo = 1;
5508 	for (i = 0; i < mirrorcnt - 1; i++, tdrpc++) {
5509 		/* Wait for RPCs on separate threads to complete. */
5510 		while (tdrpc->inprog != 0 && tdrpc->done == 0)
5511 			tsleep(&tdrpc->tsk, PVFS, "srvwrds", timo);
5512 		if (nfsds_failerr(tdrpc->err) && *failposp == -1)
5513 			*failposp = i;
5514 		else if (error == 0 && tdrpc->err != 0)
5515 			error = tdrpc->err;
5516 	}
5517 	free(drpc, M_TEMP);
5518 	return (error);
5519 }
5520 
5521 /*
5522  * Do a allocate RPC on a DS data file, using this structure for the arguments,
5523  * so that this function can be executed by a separate kernel process.
5524  */
5525 struct nfsrvallocatedsdorpc {
5526 	int			done;
5527 	int			inprog;
5528 	struct task		tsk;
5529 	fhandle_t		fh;
5530 	off_t			off;
5531 	off_t			len;
5532 	struct nfsmount		*nmp;
5533 	struct ucred		*cred;
5534 	NFSPROC_T		*p;
5535 	int			err;
5536 };
5537 
5538 static int
5539 nfsrv_allocatedsdorpc(struct nfsmount *nmp, fhandle_t *fhp, off_t off,
5540     off_t len, struct nfsvattr *nap, struct ucred *cred, NFSPROC_T *p)
5541 {
5542 	uint32_t *tl;
5543 	struct nfsrv_descript *nd;
5544 	nfsattrbit_t attrbits;
5545 	nfsv4stateid_t st;
5546 	int error;
5547 
5548 	nd = malloc(sizeof(*nd), M_TEMP, M_WAITOK | M_ZERO);
5549 	nfscl_reqstart(nd, NFSPROC_ALLOCATE, nmp, (u_int8_t *)fhp,
5550 	    sizeof(fhandle_t), NULL, NULL, 0, 0);
5551 
5552 	/*
5553 	 * Use a stateid where other is an alternating 01010 pattern and
5554 	 * seqid is 0xffffffff.  This value is not defined as special by
5555 	 * the RFC and is used by the FreeBSD NFS server to indicate an
5556 	 * MDS->DS proxy operation.
5557 	 */
5558 	st.other[0] = 0x55555555;
5559 	st.other[1] = 0x55555555;
5560 	st.other[2] = 0x55555555;
5561 	st.seqid = 0xffffffff;
5562 	nfsm_stateidtom(nd, &st, NFSSTATEID_PUTSTATEID);
5563 	NFSM_BUILD(tl, uint32_t *, 2 * NFSX_HYPER + NFSX_UNSIGNED);
5564 	txdr_hyper(off, tl); tl += 2;
5565 	txdr_hyper(len, tl); tl += 2;
5566 	NFSD_DEBUG(4, "nfsrv_allocatedsdorpc: len=%jd\n", (intmax_t)len);
5567 
5568 	*tl = txdr_unsigned(NFSV4OP_GETATTR);
5569 	NFSGETATTR_ATTRBIT(&attrbits);
5570 	nfsrv_putattrbit(nd, &attrbits);
5571 	error = newnfs_request(nd, nmp, NULL, &nmp->nm_sockreq, NULL, p,
5572 	    cred, NFS_PROG, NFS_VER4, NULL, 1, NULL, NULL);
5573 	if (error != 0) {
5574 		free(nd, M_TEMP);
5575 		return (error);
5576 	}
5577 	NFSD_DEBUG(4, "nfsrv_allocatedsdorpc: aft allocaterpc=%d\n",
5578 	    nd->nd_repstat);
5579 	if (nd->nd_repstat == 0) {
5580 		NFSM_DISSECT(tl, uint32_t *, 2 * NFSX_UNSIGNED);
5581 		error = nfsv4_loadattr(nd, NULL, nap, NULL, NULL, 0, NULL, NULL,
5582 		    NULL, NULL, NULL, 0, NULL, NULL, NULL, NULL, NULL);
5583 	} else
5584 		error = nd->nd_repstat;
5585 	NFSD_DEBUG(4, "nfsrv_allocatedsdorpc: aft loadattr=%d\n", error);
5586 nfsmout:
5587 	m_freem(nd->nd_mrep);
5588 	free(nd, M_TEMP);
5589 	NFSD_DEBUG(4, "nfsrv_allocatedsdorpc error=%d\n", error);
5590 	return (error);
5591 }
5592 
5593 /*
5594  * Start up the thread that will execute nfsrv_allocatedsdorpc().
5595  */
5596 static void
5597 start_allocatedsdorpc(void *arg, int pending)
5598 {
5599 	struct nfsrvallocatedsdorpc *drpc;
5600 
5601 	drpc = (struct nfsrvallocatedsdorpc *)arg;
5602 	drpc->err = nfsrv_allocatedsdorpc(drpc->nmp, &drpc->fh, drpc->off,
5603 	    drpc->len, NULL, drpc->cred, drpc->p);
5604 	drpc->done = 1;
5605 	NFSD_DEBUG(4, "start_allocatedsdorpc: err=%d\n", drpc->err);
5606 }
5607 
5608 static int
5609 nfsrv_allocatedsrpc(fhandle_t *fhp, off_t off, off_t len, struct ucred *cred,
5610     NFSPROC_T *p, struct vnode *vp, struct nfsmount **nmpp, int mirrorcnt,
5611     int *failposp)
5612 {
5613 	struct nfsrvallocatedsdorpc *drpc, *tdrpc = NULL;
5614 	struct nfsvattr na;
5615 	int error, i, ret, timo;
5616 
5617 	NFSD_DEBUG(4, "in nfsrv_allocatedsrpc\n");
5618 	drpc = NULL;
5619 	if (mirrorcnt > 1)
5620 		tdrpc = drpc = malloc(sizeof(*drpc) * (mirrorcnt - 1), M_TEMP,
5621 		    M_WAITOK);
5622 
5623 	/*
5624 	 * Do the allocate RPC for every DS, using a separate kernel process
5625 	 * for every DS except the last one.
5626 	 */
5627 	error = 0;
5628 	for (i = 0; i < mirrorcnt - 1; i++, tdrpc++) {
5629 		tdrpc->done = 0;
5630 		NFSBCOPY(fhp, &tdrpc->fh, sizeof(*fhp));
5631 		tdrpc->off = off;
5632 		tdrpc->len = len;
5633 		tdrpc->nmp = *nmpp;
5634 		tdrpc->cred = cred;
5635 		tdrpc->p = p;
5636 		tdrpc->inprog = 0;
5637 		tdrpc->err = 0;
5638 		ret = EIO;
5639 		if (nfs_pnfsiothreads != 0) {
5640 			ret = nfs_pnfsio(start_allocatedsdorpc, tdrpc);
5641 			NFSD_DEBUG(4, "nfsrv_allocatedsrpc: nfs_pnfsio=%d\n",
5642 			    ret);
5643 		}
5644 		if (ret != 0) {
5645 			ret = nfsrv_allocatedsdorpc(*nmpp, fhp, off, len, NULL,
5646 			    cred, p);
5647 			if (nfsds_failerr(ret) && *failposp == -1)
5648 				*failposp = i;
5649 			else if (error == 0 && ret != 0)
5650 				error = ret;
5651 		}
5652 		nmpp++;
5653 		fhp++;
5654 	}
5655 	ret = nfsrv_allocatedsdorpc(*nmpp, fhp, off, len, &na, cred, p);
5656 	if (nfsds_failerr(ret) && *failposp == -1 && mirrorcnt > 1)
5657 		*failposp = mirrorcnt - 1;
5658 	else if (error == 0 && ret != 0)
5659 		error = ret;
5660 	if (error == 0)
5661 		error = nfsrv_setextattr(vp, &na, p);
5662 	NFSD_DEBUG(4, "nfsrv_allocatedsrpc: aft setextat=%d\n", error);
5663 	tdrpc = drpc;
5664 	timo = hz / 50;		/* Wait for 20msec. */
5665 	if (timo < 1)
5666 		timo = 1;
5667 	for (i = 0; i < mirrorcnt - 1; i++, tdrpc++) {
5668 		/* Wait for RPCs on separate threads to complete. */
5669 		while (tdrpc->inprog != 0 && tdrpc->done == 0)
5670 			tsleep(&tdrpc->tsk, PVFS, "srvalds", timo);
5671 		if (nfsds_failerr(tdrpc->err) && *failposp == -1)
5672 			*failposp = i;
5673 		else if (error == 0 && tdrpc->err != 0)
5674 			error = tdrpc->err;
5675 	}
5676 	free(drpc, M_TEMP);
5677 	return (error);
5678 }
5679 
5680 /*
5681  * Do a deallocate RPC on a DS data file, using this structure for the
5682  * arguments, so that this function can be executed by a separate kernel
5683  * process.
5684  */
5685 struct nfsrvdeallocatedsdorpc {
5686 	int			done;
5687 	int			inprog;
5688 	struct task		tsk;
5689 	fhandle_t		fh;
5690 	off_t			off;
5691 	off_t			len;
5692 	struct nfsmount		*nmp;
5693 	struct ucred		*cred;
5694 	NFSPROC_T		*p;
5695 	int			err;
5696 };
5697 
5698 static int
5699 nfsrv_deallocatedsdorpc(struct nfsmount *nmp, fhandle_t *fhp, off_t off,
5700     off_t len, struct nfsvattr *nap, struct ucred *cred, NFSPROC_T *p)
5701 {
5702 	uint32_t *tl;
5703 	struct nfsrv_descript *nd;
5704 	nfsattrbit_t attrbits;
5705 	nfsv4stateid_t st;
5706 	int error;
5707 
5708 	nd = malloc(sizeof(*nd), M_TEMP, M_WAITOK | M_ZERO);
5709 	nfscl_reqstart(nd, NFSPROC_DEALLOCATE, nmp, (u_int8_t *)fhp,
5710 	    sizeof(fhandle_t), NULL, NULL, 0, 0);
5711 
5712 	/*
5713 	 * Use a stateid where other is an alternating 01010 pattern and
5714 	 * seqid is 0xffffffff.  This value is not defined as special by
5715 	 * the RFC and is used by the FreeBSD NFS server to indicate an
5716 	 * MDS->DS proxy operation.
5717 	 */
5718 	st.other[0] = 0x55555555;
5719 	st.other[1] = 0x55555555;
5720 	st.other[2] = 0x55555555;
5721 	st.seqid = 0xffffffff;
5722 	nfsm_stateidtom(nd, &st, NFSSTATEID_PUTSTATEID);
5723 	NFSM_BUILD(tl, uint32_t *, 2 * NFSX_HYPER + NFSX_UNSIGNED);
5724 	txdr_hyper(off, tl); tl += 2;
5725 	txdr_hyper(len, tl); tl += 2;
5726 	NFSD_DEBUG(4, "nfsrv_deallocatedsdorpc: len=%jd\n", (intmax_t)len);
5727 
5728 	/* Do a Getattr for the attributes that change upon writing. */
5729 	NFSZERO_ATTRBIT(&attrbits);
5730 	NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_SIZE);
5731 	NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_CHANGE);
5732 	NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_TIMEACCESS);
5733 	NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_TIMEMODIFY);
5734 	NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_SPACEUSED);
5735 	*tl = txdr_unsigned(NFSV4OP_GETATTR);
5736 	nfsrv_putattrbit(nd, &attrbits);
5737 	error = newnfs_request(nd, nmp, NULL, &nmp->nm_sockreq, NULL, p,
5738 	    cred, NFS_PROG, NFS_VER4, NULL, 1, NULL, NULL);
5739 	if (error != 0) {
5740 		free(nd, M_TEMP);
5741 		return (error);
5742 	}
5743 	NFSD_DEBUG(4, "nfsrv_deallocatedsdorpc: aft deallocaterpc=%d\n",
5744 	    nd->nd_repstat);
5745 	/* Get rid of weak cache consistency data for now. */
5746 	if ((nd->nd_flag & (ND_NOMOREDATA | ND_NFSV4 | ND_V4WCCATTR)) ==
5747 	    (ND_NFSV4 | ND_V4WCCATTR)) {
5748 		error = nfsv4_loadattr(nd, NULL, nap, NULL, NULL, 0, NULL, NULL,
5749 		    NULL, NULL, NULL, 0, NULL, NULL, NULL, NULL, NULL);
5750 		NFSD_DEBUG(4, "nfsrv_deallocatedsdorpc: wcc attr=%d\n", error);
5751 		if (error != 0)
5752 			goto nfsmout;
5753 		/*
5754 		 * Get rid of Op# and status for next op.
5755 		 */
5756 		NFSM_DISSECT(tl, uint32_t *, 2 * NFSX_UNSIGNED);
5757 		if (*++tl != 0)
5758 			nd->nd_flag |= ND_NOMOREDATA;
5759 	}
5760 	if (nd->nd_repstat == 0) {
5761 		NFSM_DISSECT(tl, uint32_t *, 2 * NFSX_UNSIGNED);
5762 		error = nfsv4_loadattr(nd, NULL, nap, NULL, NULL, 0, NULL, NULL,
5763 		    NULL, NULL, NULL, 0, NULL, NULL, NULL, NULL, NULL);
5764 	} else
5765 		error = nd->nd_repstat;
5766 	NFSD_DEBUG(4, "nfsrv_deallocatedsdorpc: aft loadattr=%d\n", error);
5767 nfsmout:
5768 	m_freem(nd->nd_mrep);
5769 	free(nd, M_TEMP);
5770 	NFSD_DEBUG(4, "nfsrv_deallocatedsdorpc error=%d\n", error);
5771 	return (error);
5772 }
5773 
5774 /*
5775  * Start up the thread that will execute nfsrv_deallocatedsdorpc().
5776  */
5777 static void
5778 start_deallocatedsdorpc(void *arg, int pending)
5779 {
5780 	struct nfsrvdeallocatedsdorpc *drpc;
5781 
5782 	drpc = (struct nfsrvdeallocatedsdorpc *)arg;
5783 	drpc->err = nfsrv_deallocatedsdorpc(drpc->nmp, &drpc->fh, drpc->off,
5784 	    drpc->len, NULL, drpc->cred, drpc->p);
5785 	drpc->done = 1;
5786 	NFSD_DEBUG(4, "start_deallocatedsdorpc: err=%d\n", drpc->err);
5787 }
5788 
5789 static int
5790 nfsrv_deallocatedsrpc(fhandle_t *fhp, off_t off, off_t len, struct ucred *cred,
5791     NFSPROC_T *p, struct vnode *vp, struct nfsmount **nmpp, int mirrorcnt,
5792     int *failposp)
5793 {
5794 	struct nfsrvdeallocatedsdorpc *drpc, *tdrpc = NULL;
5795 	struct nfsvattr na;
5796 	int error, i, ret, timo;
5797 
5798 	NFSD_DEBUG(4, "in nfsrv_deallocatedsrpc\n");
5799 	drpc = NULL;
5800 	if (mirrorcnt > 1)
5801 		tdrpc = drpc = malloc(sizeof(*drpc) * (mirrorcnt - 1), M_TEMP,
5802 		    M_WAITOK);
5803 
5804 	/*
5805 	 * Do the deallocate RPC for every DS, using a separate kernel process
5806 	 * for every DS except the last one.
5807 	 */
5808 	error = 0;
5809 	for (i = 0; i < mirrorcnt - 1; i++, tdrpc++) {
5810 		tdrpc->done = 0;
5811 		NFSBCOPY(fhp, &tdrpc->fh, sizeof(*fhp));
5812 		tdrpc->off = off;
5813 		tdrpc->len = len;
5814 		tdrpc->nmp = *nmpp;
5815 		tdrpc->cred = cred;
5816 		tdrpc->p = p;
5817 		tdrpc->inprog = 0;
5818 		tdrpc->err = 0;
5819 		ret = EIO;
5820 		if (nfs_pnfsiothreads != 0) {
5821 			ret = nfs_pnfsio(start_deallocatedsdorpc, tdrpc);
5822 			NFSD_DEBUG(4, "nfsrv_deallocatedsrpc: nfs_pnfsio=%d\n",
5823 			    ret);
5824 		}
5825 		if (ret != 0) {
5826 			ret = nfsrv_deallocatedsdorpc(*nmpp, fhp, off, len,
5827 			    NULL, cred, p);
5828 			if (nfsds_failerr(ret) && *failposp == -1)
5829 				*failposp = i;
5830 			else if (error == 0 && ret != 0)
5831 				error = ret;
5832 		}
5833 		nmpp++;
5834 		fhp++;
5835 	}
5836 	ret = nfsrv_deallocatedsdorpc(*nmpp, fhp, off, len, &na, cred, p);
5837 	if (nfsds_failerr(ret) && *failposp == -1 && mirrorcnt > 1)
5838 		*failposp = mirrorcnt - 1;
5839 	else if (error == 0 && ret != 0)
5840 		error = ret;
5841 	if (error == 0)
5842 		error = nfsrv_setextattr(vp, &na, p);
5843 	NFSD_DEBUG(4, "nfsrv_deallocatedsrpc: aft setextat=%d\n", error);
5844 	tdrpc = drpc;
5845 	timo = hz / 50;		/* Wait for 20msec. */
5846 	if (timo < 1)
5847 		timo = 1;
5848 	for (i = 0; i < mirrorcnt - 1; i++, tdrpc++) {
5849 		/* Wait for RPCs on separate threads to complete. */
5850 		while (tdrpc->inprog != 0 && tdrpc->done == 0)
5851 			tsleep(&tdrpc->tsk, PVFS, "srvalds", timo);
5852 		if (nfsds_failerr(tdrpc->err) && *failposp == -1)
5853 			*failposp = i;
5854 		else if (error == 0 && tdrpc->err != 0)
5855 			error = tdrpc->err;
5856 	}
5857 	free(drpc, M_TEMP);
5858 	return (error);
5859 }
5860 
5861 static int
5862 nfsrv_setattrdsdorpc(fhandle_t *fhp, struct ucred *cred, NFSPROC_T *p,
5863     struct vnode *vp, struct nfsmount *nmp, struct nfsvattr *nap,
5864     struct nfsvattr *dsnap)
5865 {
5866 	uint32_t *tl;
5867 	struct nfsrv_descript *nd;
5868 	nfsv4stateid_t st;
5869 	nfsattrbit_t attrbits;
5870 	int error;
5871 
5872 	NFSD_DEBUG(4, "in nfsrv_setattrdsdorpc\n");
5873 	nd = malloc(sizeof(*nd), M_TEMP, M_WAITOK | M_ZERO);
5874 	/*
5875 	 * Use a stateid where other is an alternating 01010 pattern and
5876 	 * seqid is 0xffffffff.  This value is not defined as special by
5877 	 * the RFC and is used by the FreeBSD NFS server to indicate an
5878 	 * MDS->DS proxy operation.
5879 	 */
5880 	st.other[0] = 0x55555555;
5881 	st.other[1] = 0x55555555;
5882 	st.other[2] = 0x55555555;
5883 	st.seqid = 0xffffffff;
5884 	nfscl_reqstart(nd, NFSPROC_SETATTR, nmp, (u_int8_t *)fhp, sizeof(*fhp),
5885 	    NULL, NULL, 0, 0);
5886 	nfsm_stateidtom(nd, &st, NFSSTATEID_PUTSTATEID);
5887 	nfscl_fillsattr(nd, &nap->na_vattr, vp, NFSSATTR_FULL, 0);
5888 
5889 	/* Do a Getattr for the attributes that change due to writing. */
5890 	NFSZERO_ATTRBIT(&attrbits);
5891 	NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_SIZE);
5892 	NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_CHANGE);
5893 	NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_TIMEACCESS);
5894 	NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_TIMEMODIFY);
5895 	NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_SPACEUSED);
5896 	NFSM_BUILD(tl, u_int32_t *, NFSX_UNSIGNED);
5897 	*tl = txdr_unsigned(NFSV4OP_GETATTR);
5898 	(void) nfsrv_putattrbit(nd, &attrbits);
5899 	error = newnfs_request(nd, nmp, NULL, &nmp->nm_sockreq, NULL, p, cred,
5900 	    NFS_PROG, NFS_VER4, NULL, 1, NULL, NULL);
5901 	if (error != 0) {
5902 		free(nd, M_TEMP);
5903 		return (error);
5904 	}
5905 	NFSD_DEBUG(4, "nfsrv_setattrdsdorpc: aft setattrrpc=%d\n",
5906 	    nd->nd_repstat);
5907 	/* Get rid of weak cache consistency data for now. */
5908 	if ((nd->nd_flag & (ND_NOMOREDATA | ND_NFSV4 | ND_V4WCCATTR)) ==
5909 	    (ND_NFSV4 | ND_V4WCCATTR)) {
5910 		error = nfsv4_loadattr(nd, NULL, dsnap, NULL, NULL, 0, NULL,
5911 		    NULL, NULL, NULL, NULL, 0, NULL, NULL, NULL, NULL, NULL);
5912 		NFSD_DEBUG(4, "nfsrv_setattrdsdorpc: wcc attr=%d\n", error);
5913 		if (error != 0)
5914 			goto nfsmout;
5915 		/*
5916 		 * Get rid of Op# and status for next op.
5917 		 */
5918 		NFSM_DISSECT(tl, uint32_t *, 2 * NFSX_UNSIGNED);
5919 		if (*++tl != 0)
5920 			nd->nd_flag |= ND_NOMOREDATA;
5921 	}
5922 	error = nfsrv_getattrbits(nd, &attrbits, NULL, NULL);
5923 	if (error != 0)
5924 		goto nfsmout;
5925 	if (nd->nd_repstat != 0)
5926 		error = nd->nd_repstat;
5927 	/*
5928 	 * Get the Change, Size, Access Time and Modify Time attributes and set
5929 	 * on the Metadata file, so its attributes will be what the file's
5930 	 * would be if it had been written.
5931 	 */
5932 	if (error == 0) {
5933 		NFSM_DISSECT(tl, uint32_t *, 2 * NFSX_UNSIGNED);
5934 		error = nfsv4_loadattr(nd, NULL, dsnap, NULL, NULL, 0, NULL,
5935 		    NULL, NULL, NULL, NULL, 0, NULL, NULL, NULL, NULL, NULL);
5936 	}
5937 	NFSD_DEBUG(4, "nfsrv_setattrdsdorpc: aft setattr loadattr=%d\n", error);
5938 nfsmout:
5939 	m_freem(nd->nd_mrep);
5940 	free(nd, M_TEMP);
5941 	NFSD_DEBUG(4, "nfsrv_setattrdsdorpc error=%d\n", error);
5942 	return (error);
5943 }
5944 
5945 struct nfsrvsetattrdsdorpc {
5946 	int			done;
5947 	int			inprog;
5948 	struct task		tsk;
5949 	fhandle_t		fh;
5950 	struct nfsmount		*nmp;
5951 	struct vnode		*vp;
5952 	struct ucred		*cred;
5953 	NFSPROC_T		*p;
5954 	struct nfsvattr		na;
5955 	struct nfsvattr		dsna;
5956 	int			err;
5957 };
5958 
5959 /*
5960  * Start up the thread that will execute nfsrv_setattrdsdorpc().
5961  */
5962 static void
5963 start_setattrdsdorpc(void *arg, int pending)
5964 {
5965 	struct nfsrvsetattrdsdorpc *drpc;
5966 
5967 	drpc = (struct nfsrvsetattrdsdorpc *)arg;
5968 	drpc->err = nfsrv_setattrdsdorpc(&drpc->fh, drpc->cred, drpc->p,
5969 	    drpc->vp, drpc->nmp, &drpc->na, &drpc->dsna);
5970 	drpc->done = 1;
5971 }
5972 
5973 static int
5974 nfsrv_setattrdsrpc(fhandle_t *fhp, struct ucred *cred, NFSPROC_T *p,
5975     struct vnode *vp, struct nfsmount **nmpp, int mirrorcnt,
5976     struct nfsvattr *nap, int *failposp)
5977 {
5978 	struct nfsrvsetattrdsdorpc *drpc, *tdrpc = NULL;
5979 	struct nfsvattr na;
5980 	int error, i, ret, timo;
5981 
5982 	NFSD_DEBUG(4, "in nfsrv_setattrdsrpc\n");
5983 	drpc = NULL;
5984 	if (mirrorcnt > 1)
5985 		tdrpc = drpc = malloc(sizeof(*drpc) * (mirrorcnt - 1), M_TEMP,
5986 		    M_WAITOK);
5987 
5988 	/*
5989 	 * Do the setattr RPC for every DS, using a separate kernel process
5990 	 * for every DS except the last one.
5991 	 */
5992 	error = 0;
5993 	for (i = 0; i < mirrorcnt - 1; i++, tdrpc++) {
5994 		tdrpc->done = 0;
5995 		tdrpc->inprog = 0;
5996 		NFSBCOPY(fhp, &tdrpc->fh, sizeof(*fhp));
5997 		tdrpc->nmp = *nmpp;
5998 		tdrpc->vp = vp;
5999 		tdrpc->cred = cred;
6000 		tdrpc->p = p;
6001 		tdrpc->na = *nap;
6002 		tdrpc->err = 0;
6003 		ret = EIO;
6004 		if (nfs_pnfsiothreads != 0) {
6005 			ret = nfs_pnfsio(start_setattrdsdorpc, tdrpc);
6006 			NFSD_DEBUG(4, "nfsrv_setattrdsrpc: nfs_pnfsio=%d\n",
6007 			    ret);
6008 		}
6009 		if (ret != 0) {
6010 			ret = nfsrv_setattrdsdorpc(fhp, cred, p, vp, *nmpp, nap,
6011 			    &na);
6012 			if (nfsds_failerr(ret) && *failposp == -1)
6013 				*failposp = i;
6014 			else if (error == 0 && ret != 0)
6015 				error = ret;
6016 		}
6017 		nmpp++;
6018 		fhp++;
6019 	}
6020 	ret = nfsrv_setattrdsdorpc(fhp, cred, p, vp, *nmpp, nap, &na);
6021 	if (nfsds_failerr(ret) && *failposp == -1 && mirrorcnt > 1)
6022 		*failposp = mirrorcnt - 1;
6023 	else if (error == 0 && ret != 0)
6024 		error = ret;
6025 	if (error == 0)
6026 		error = nfsrv_setextattr(vp, &na, p);
6027 	NFSD_DEBUG(4, "nfsrv_setattrdsrpc: aft setextat=%d\n", error);
6028 	tdrpc = drpc;
6029 	timo = hz / 50;		/* Wait for 20msec. */
6030 	if (timo < 1)
6031 		timo = 1;
6032 	for (i = 0; i < mirrorcnt - 1; i++, tdrpc++) {
6033 		/* Wait for RPCs on separate threads to complete. */
6034 		while (tdrpc->inprog != 0 && tdrpc->done == 0)
6035 			tsleep(&tdrpc->tsk, PVFS, "srvsads", timo);
6036 		if (nfsds_failerr(tdrpc->err) && *failposp == -1)
6037 			*failposp = i;
6038 		else if (error == 0 && tdrpc->err != 0)
6039 			error = tdrpc->err;
6040 	}
6041 	free(drpc, M_TEMP);
6042 	return (error);
6043 }
6044 
6045 /*
6046  * Do a Setattr of an NFSv4 ACL on the DS file.
6047  */
6048 static int
6049 nfsrv_setacldsdorpc(fhandle_t *fhp, struct ucred *cred, NFSPROC_T *p,
6050     struct vnode *vp, struct nfsmount *nmp, struct acl *aclp)
6051 {
6052 	struct nfsrv_descript *nd;
6053 	nfsv4stateid_t st;
6054 	nfsattrbit_t attrbits;
6055 	int error;
6056 
6057 	NFSD_DEBUG(4, "in nfsrv_setacldsdorpc\n");
6058 	nd = malloc(sizeof(*nd), M_TEMP, M_WAITOK | M_ZERO);
6059 	/*
6060 	 * Use a stateid where other is an alternating 01010 pattern and
6061 	 * seqid is 0xffffffff.  This value is not defined as special by
6062 	 * the RFC and is used by the FreeBSD NFS server to indicate an
6063 	 * MDS->DS proxy operation.
6064 	 */
6065 	st.other[0] = 0x55555555;
6066 	st.other[1] = 0x55555555;
6067 	st.other[2] = 0x55555555;
6068 	st.seqid = 0xffffffff;
6069 	nfscl_reqstart(nd, NFSPROC_SETACL, nmp, (u_int8_t *)fhp, sizeof(*fhp),
6070 	    NULL, NULL, 0, 0);
6071 	nfsm_stateidtom(nd, &st, NFSSTATEID_PUTSTATEID);
6072 	NFSZERO_ATTRBIT(&attrbits);
6073 	NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_ACL);
6074 	/*
6075 	 * The "vp" argument to nfsv4_fillattr() is only used for vnode_type(),
6076 	 * so passing in the metadata "vp" will be ok, since it is of
6077 	 * the same type (VREG).
6078 	 */
6079 	nfsv4_fillattr(nd, NULL, vp, aclp, NULL, NULL, 0, &attrbits, NULL,
6080 	    NULL, 0, 0, 0, 0, 0, NULL);
6081 	error = newnfs_request(nd, nmp, NULL, &nmp->nm_sockreq, NULL, p, cred,
6082 	    NFS_PROG, NFS_VER4, NULL, 1, NULL, NULL);
6083 	if (error != 0) {
6084 		free(nd, M_TEMP);
6085 		return (error);
6086 	}
6087 	NFSD_DEBUG(4, "nfsrv_setacldsdorpc: aft setaclrpc=%d\n",
6088 	    nd->nd_repstat);
6089 	error = nd->nd_repstat;
6090 	m_freem(nd->nd_mrep);
6091 	free(nd, M_TEMP);
6092 	return (error);
6093 }
6094 
6095 struct nfsrvsetacldsdorpc {
6096 	int			done;
6097 	int			inprog;
6098 	struct task		tsk;
6099 	fhandle_t		fh;
6100 	struct nfsmount		*nmp;
6101 	struct vnode		*vp;
6102 	struct ucred		*cred;
6103 	NFSPROC_T		*p;
6104 	struct acl		*aclp;
6105 	int			err;
6106 };
6107 
6108 /*
6109  * Start up the thread that will execute nfsrv_setacldsdorpc().
6110  */
6111 static void
6112 start_setacldsdorpc(void *arg, int pending)
6113 {
6114 	struct nfsrvsetacldsdorpc *drpc;
6115 
6116 	drpc = (struct nfsrvsetacldsdorpc *)arg;
6117 	drpc->err = nfsrv_setacldsdorpc(&drpc->fh, drpc->cred, drpc->p,
6118 	    drpc->vp, drpc->nmp, drpc->aclp);
6119 	drpc->done = 1;
6120 }
6121 
6122 static int
6123 nfsrv_setacldsrpc(fhandle_t *fhp, struct ucred *cred, NFSPROC_T *p,
6124     struct vnode *vp, struct nfsmount **nmpp, int mirrorcnt, struct acl *aclp,
6125     int *failposp)
6126 {
6127 	struct nfsrvsetacldsdorpc *drpc, *tdrpc = NULL;
6128 	int error, i, ret, timo;
6129 
6130 	NFSD_DEBUG(4, "in nfsrv_setacldsrpc\n");
6131 	drpc = NULL;
6132 	if (mirrorcnt > 1)
6133 		tdrpc = drpc = malloc(sizeof(*drpc) * (mirrorcnt - 1), M_TEMP,
6134 		    M_WAITOK);
6135 
6136 	/*
6137 	 * Do the setattr RPC for every DS, using a separate kernel process
6138 	 * for every DS except the last one.
6139 	 */
6140 	error = 0;
6141 	for (i = 0; i < mirrorcnt - 1; i++, tdrpc++) {
6142 		tdrpc->done = 0;
6143 		tdrpc->inprog = 0;
6144 		NFSBCOPY(fhp, &tdrpc->fh, sizeof(*fhp));
6145 		tdrpc->nmp = *nmpp;
6146 		tdrpc->vp = vp;
6147 		tdrpc->cred = cred;
6148 		tdrpc->p = p;
6149 		tdrpc->aclp = aclp;
6150 		tdrpc->err = 0;
6151 		ret = EIO;
6152 		if (nfs_pnfsiothreads != 0) {
6153 			ret = nfs_pnfsio(start_setacldsdorpc, tdrpc);
6154 			NFSD_DEBUG(4, "nfsrv_setacldsrpc: nfs_pnfsio=%d\n",
6155 			    ret);
6156 		}
6157 		if (ret != 0) {
6158 			ret = nfsrv_setacldsdorpc(fhp, cred, p, vp, *nmpp,
6159 			    aclp);
6160 			if (nfsds_failerr(ret) && *failposp == -1)
6161 				*failposp = i;
6162 			else if (error == 0 && ret != 0)
6163 				error = ret;
6164 		}
6165 		nmpp++;
6166 		fhp++;
6167 	}
6168 	ret = nfsrv_setacldsdorpc(fhp, cred, p, vp, *nmpp, aclp);
6169 	if (nfsds_failerr(ret) && *failposp == -1 && mirrorcnt > 1)
6170 		*failposp = mirrorcnt - 1;
6171 	else if (error == 0 && ret != 0)
6172 		error = ret;
6173 	NFSD_DEBUG(4, "nfsrv_setacldsrpc: aft setextat=%d\n", error);
6174 	tdrpc = drpc;
6175 	timo = hz / 50;		/* Wait for 20msec. */
6176 	if (timo < 1)
6177 		timo = 1;
6178 	for (i = 0; i < mirrorcnt - 1; i++, tdrpc++) {
6179 		/* Wait for RPCs on separate threads to complete. */
6180 		while (tdrpc->inprog != 0 && tdrpc->done == 0)
6181 			tsleep(&tdrpc->tsk, PVFS, "srvacds", timo);
6182 		if (nfsds_failerr(tdrpc->err) && *failposp == -1)
6183 			*failposp = i;
6184 		else if (error == 0 && tdrpc->err != 0)
6185 			error = tdrpc->err;
6186 	}
6187 	free(drpc, M_TEMP);
6188 	return (error);
6189 }
6190 
6191 /*
6192  * Getattr call to the DS for the attributes that change due to writing.
6193  */
6194 static int
6195 nfsrv_getattrdsrpc(fhandle_t *fhp, struct ucred *cred, NFSPROC_T *p,
6196     struct vnode *vp, struct nfsmount *nmp, struct nfsvattr *nap)
6197 {
6198 	struct nfsrv_descript *nd;
6199 	int error;
6200 	nfsattrbit_t attrbits;
6201 
6202 	NFSD_DEBUG(4, "in nfsrv_getattrdsrpc\n");
6203 	nd = malloc(sizeof(*nd), M_TEMP, M_WAITOK | M_ZERO);
6204 	nfscl_reqstart(nd, NFSPROC_GETATTR, nmp, (u_int8_t *)fhp,
6205 	    sizeof(fhandle_t), NULL, NULL, 0, 0);
6206 	NFSZERO_ATTRBIT(&attrbits);
6207 	NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_SIZE);
6208 	NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_CHANGE);
6209 	NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_TIMEACCESS);
6210 	NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_TIMEMODIFY);
6211 	NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_SPACEUSED);
6212 	(void) nfsrv_putattrbit(nd, &attrbits);
6213 	error = newnfs_request(nd, nmp, NULL, &nmp->nm_sockreq, NULL, p, cred,
6214 	    NFS_PROG, NFS_VER4, NULL, 1, NULL, NULL);
6215 	if (error != 0) {
6216 		free(nd, M_TEMP);
6217 		return (error);
6218 	}
6219 	NFSD_DEBUG(4, "nfsrv_getattrdsrpc: aft getattrrpc=%d\n",
6220 	    nd->nd_repstat);
6221 	if (nd->nd_repstat == 0) {
6222 		error = nfsv4_loadattr(nd, NULL, nap, NULL, NULL, 0,
6223 		    NULL, NULL, NULL, NULL, NULL, 0, NULL, NULL, NULL,
6224 		    NULL, NULL);
6225 		/*
6226 		 * We can only save the updated values in the extended
6227 		 * attribute if the vp is exclusively locked.
6228 		 * This should happen when any of the following operations
6229 		 * occur on the vnode:
6230 		 *    Close, Delegreturn, LayoutCommit, LayoutReturn
6231 		 * As such, the updated extended attribute should get saved
6232 		 * before nfsrv_checkdsattr() returns 0 and allows the cached
6233 		 * attributes to be returned without calling this function.
6234 		 */
6235 		if (error == 0 && VOP_ISLOCKED(vp) == LK_EXCLUSIVE) {
6236 			error = nfsrv_setextattr(vp, nap, p);
6237 			NFSD_DEBUG(4, "nfsrv_getattrdsrpc: aft setextat=%d\n",
6238 			    error);
6239 		}
6240 	} else
6241 		error = nd->nd_repstat;
6242 	m_freem(nd->nd_mrep);
6243 	free(nd, M_TEMP);
6244 	NFSD_DEBUG(4, "nfsrv_getattrdsrpc error=%d\n", error);
6245 	return (error);
6246 }
6247 
6248 /*
6249  * Seek call to a DS.
6250  */
6251 static int
6252 nfsrv_seekdsrpc(fhandle_t *fhp, off_t *offp, int content, bool *eofp,
6253     struct ucred *cred, NFSPROC_T *p, struct nfsmount *nmp)
6254 {
6255 	uint32_t *tl;
6256 	struct nfsrv_descript *nd;
6257 	nfsv4stateid_t st;
6258 	int error;
6259 
6260 	NFSD_DEBUG(4, "in nfsrv_seekdsrpc\n");
6261 	/*
6262 	 * Use a stateid where other is an alternating 01010 pattern and
6263 	 * seqid is 0xffffffff.  This value is not defined as special by
6264 	 * the RFC and is used by the FreeBSD NFS server to indicate an
6265 	 * MDS->DS proxy operation.
6266 	 */
6267 	st.other[0] = 0x55555555;
6268 	st.other[1] = 0x55555555;
6269 	st.other[2] = 0x55555555;
6270 	st.seqid = 0xffffffff;
6271 	nd = malloc(sizeof(*nd), M_TEMP, M_WAITOK | M_ZERO);
6272 	nfscl_reqstart(nd, NFSPROC_SEEKDS, nmp, (u_int8_t *)fhp,
6273 	    sizeof(fhandle_t), NULL, NULL, 0, 0);
6274 	nfsm_stateidtom(nd, &st, NFSSTATEID_PUTSTATEID);
6275 	NFSM_BUILD(tl, uint32_t *, NFSX_HYPER + NFSX_UNSIGNED);
6276 	txdr_hyper(*offp, tl); tl += 2;
6277 	*tl = txdr_unsigned(content);
6278 	error = newnfs_request(nd, nmp, NULL, &nmp->nm_sockreq, NULL, p, cred,
6279 	    NFS_PROG, NFS_VER4, NULL, 1, NULL, NULL);
6280 	if (error != 0) {
6281 		free(nd, M_TEMP);
6282 		return (error);
6283 	}
6284 	NFSD_DEBUG(4, "nfsrv_seekdsrpc: aft seekrpc=%d\n", nd->nd_repstat);
6285 	if (nd->nd_repstat == 0) {
6286 		NFSM_DISSECT(tl, uint32_t *, NFSX_UNSIGNED + NFSX_HYPER);
6287 		if (*tl++ == newnfs_true)
6288 			*eofp = true;
6289 		else
6290 			*eofp = false;
6291 		*offp = fxdr_hyper(tl);
6292 	} else
6293 		error = nd->nd_repstat;
6294 nfsmout:
6295 	m_freem(nd->nd_mrep);
6296 	free(nd, M_TEMP);
6297 	NFSD_DEBUG(4, "nfsrv_seekdsrpc error=%d\n", error);
6298 	return (error);
6299 }
6300 
6301 /*
6302  * Get the device id and file handle for a DS file.
6303  */
6304 int
6305 nfsrv_dsgetdevandfh(struct vnode *vp, NFSPROC_T *p, int *mirrorcntp,
6306     fhandle_t *fhp, char *devid)
6307 {
6308 	int buflen, error;
6309 	char *buf;
6310 
6311 	buflen = 1024;
6312 	buf = malloc(buflen, M_TEMP, M_WAITOK);
6313 	error = nfsrv_dsgetsockmnt(vp, 0, buf, &buflen, mirrorcntp, p, NULL,
6314 	    fhp, devid, NULL, NULL, NULL, NULL, NULL, NULL);
6315 	free(buf, M_TEMP);
6316 	return (error);
6317 }
6318 
6319 /*
6320  * Do a Lookup against the DS for the filename.
6321  */
6322 static int
6323 nfsrv_pnfslookupds(struct vnode *vp, struct vnode *dvp, struct pnfsdsfile *pf,
6324     struct vnode **nvpp, NFSPROC_T *p)
6325 {
6326 	struct nameidata named;
6327 	struct ucred *tcred;
6328 	char *bufp;
6329 	u_long *hashp;
6330 	struct vnode *nvp;
6331 	int error;
6332 
6333 	tcred = newnfs_getcred();
6334 	named.ni_cnd.cn_nameiop = LOOKUP;
6335 	named.ni_cnd.cn_lkflags = LK_SHARED | LK_RETRY;
6336 	named.ni_cnd.cn_cred = tcred;
6337 	named.ni_cnd.cn_flags = ISLASTCN | LOCKPARENT | LOCKLEAF | SAVENAME;
6338 	nfsvno_setpathbuf(&named, &bufp, &hashp);
6339 	named.ni_cnd.cn_nameptr = bufp;
6340 	named.ni_cnd.cn_namelen = strlen(pf->dsf_filename);
6341 	strlcpy(bufp, pf->dsf_filename, NAME_MAX);
6342 	NFSD_DEBUG(4, "nfsrv_pnfslookupds: filename=%s\n", bufp);
6343 	error = VOP_LOOKUP(dvp, &nvp, &named.ni_cnd);
6344 	NFSD_DEBUG(4, "nfsrv_pnfslookupds: aft LOOKUP=%d\n", error);
6345 	NFSFREECRED(tcred);
6346 	nfsvno_relpathbuf(&named);
6347 	if (error == 0)
6348 		*nvpp = nvp;
6349 	NFSD_DEBUG(4, "eo nfsrv_pnfslookupds=%d\n", error);
6350 	return (error);
6351 }
6352 
6353 /*
6354  * Set the file handle to the correct one.
6355  */
6356 static void
6357 nfsrv_pnfssetfh(struct vnode *vp, struct pnfsdsfile *pf, char *devid,
6358     char *fnamep, struct vnode *nvp, NFSPROC_T *p)
6359 {
6360 	struct nfsnode *np;
6361 	int ret = 0;
6362 
6363 	np = VTONFS(nvp);
6364 	NFSBCOPY(np->n_fhp->nfh_fh, &pf->dsf_fh, NFSX_MYFH);
6365 	/*
6366 	 * We can only do a vn_set_extattr() if the vnode is exclusively
6367 	 * locked and vn_start_write() has been done.  If devid != NULL or
6368 	 * fnamep != NULL or the vnode is shared locked, vn_start_write()
6369 	 * may not have been done.
6370 	 * If not done now, it will be done on a future call.
6371 	 */
6372 	if (devid == NULL && fnamep == NULL && NFSVOPISLOCKED(vp) ==
6373 	    LK_EXCLUSIVE)
6374 		ret = vn_extattr_set(vp, IO_NODELOCKED,
6375 		    EXTATTR_NAMESPACE_SYSTEM, "pnfsd.dsfile", sizeof(*pf),
6376 		    (char *)pf, p);
6377 	NFSD_DEBUG(4, "eo nfsrv_pnfssetfh=%d\n", ret);
6378 }
6379 
6380 /*
6381  * Cause RPCs waiting on "nmp" to fail.  This is called for a DS mount point
6382  * when the DS has failed.
6383  */
6384 void
6385 nfsrv_killrpcs(struct nfsmount *nmp)
6386 {
6387 
6388 	/*
6389 	 * Call newnfs_nmcancelreqs() to cause
6390 	 * any RPCs in progress on the mount point to
6391 	 * fail.
6392 	 * This will cause any process waiting for an
6393 	 * RPC to complete while holding a vnode lock
6394 	 * on the mounted-on vnode (such as "df" or
6395 	 * a non-forced "umount") to fail.
6396 	 * This will unlock the mounted-on vnode so
6397 	 * a forced dismount can succeed.
6398 	 * The NFSMNTP_CANCELRPCS flag should be set when this function is
6399 	 * called.
6400 	 */
6401 	newnfs_nmcancelreqs(nmp);
6402 }
6403 
6404 /*
6405  * Sum up the statfs info for each of the DSs, so that the client will
6406  * receive the total for all DSs.
6407  */
6408 static int
6409 nfsrv_pnfsstatfs(struct statfs *sf, struct mount *mp)
6410 {
6411 	struct statfs *tsf;
6412 	struct nfsdevice *ds;
6413 	struct vnode **dvpp, **tdvpp, *dvp;
6414 	uint64_t tot;
6415 	int cnt, error = 0, i;
6416 
6417 	if (nfsrv_devidcnt <= 0)
6418 		return (ENXIO);
6419 	dvpp = mallocarray(nfsrv_devidcnt, sizeof(*dvpp), M_TEMP, M_WAITOK);
6420 	tsf = malloc(sizeof(*tsf), M_TEMP, M_WAITOK);
6421 
6422 	/* Get an array of the dvps for the DSs. */
6423 	tdvpp = dvpp;
6424 	i = 0;
6425 	NFSDDSLOCK();
6426 	/* First, search for matches for same file system. */
6427 	TAILQ_FOREACH(ds, &nfsrv_devidhead, nfsdev_list) {
6428 		if (ds->nfsdev_nmp != NULL && ds->nfsdev_mdsisset != 0 &&
6429 		    fsidcmp(&ds->nfsdev_mdsfsid, &mp->mnt_stat.f_fsid) == 0) {
6430 			if (++i > nfsrv_devidcnt)
6431 				break;
6432 			*tdvpp++ = ds->nfsdev_dvp;
6433 		}
6434 	}
6435 	/*
6436 	 * If no matches for same file system, total all servers not assigned
6437 	 * to a file system.
6438 	 */
6439 	if (i == 0) {
6440 		TAILQ_FOREACH(ds, &nfsrv_devidhead, nfsdev_list) {
6441 			if (ds->nfsdev_nmp != NULL &&
6442 			    ds->nfsdev_mdsisset == 0) {
6443 				if (++i > nfsrv_devidcnt)
6444 					break;
6445 				*tdvpp++ = ds->nfsdev_dvp;
6446 			}
6447 		}
6448 	}
6449 	NFSDDSUNLOCK();
6450 	cnt = i;
6451 
6452 	/* Do a VFS_STATFS() for each of the DSs and sum them up. */
6453 	tdvpp = dvpp;
6454 	for (i = 0; i < cnt && error == 0; i++) {
6455 		dvp = *tdvpp++;
6456 		error = VFS_STATFS(dvp->v_mount, tsf);
6457 		if (error == 0) {
6458 			if (sf->f_bsize == 0) {
6459 				if (tsf->f_bsize > 0)
6460 					sf->f_bsize = tsf->f_bsize;
6461 				else
6462 					sf->f_bsize = 8192;
6463 			}
6464 			if (tsf->f_blocks > 0) {
6465 				if (sf->f_bsize != tsf->f_bsize) {
6466 					tot = tsf->f_blocks * tsf->f_bsize;
6467 					sf->f_blocks += (tot / sf->f_bsize);
6468 				} else
6469 					sf->f_blocks += tsf->f_blocks;
6470 			}
6471 			if (tsf->f_bfree > 0) {
6472 				if (sf->f_bsize != tsf->f_bsize) {
6473 					tot = tsf->f_bfree * tsf->f_bsize;
6474 					sf->f_bfree += (tot / sf->f_bsize);
6475 				} else
6476 					sf->f_bfree += tsf->f_bfree;
6477 			}
6478 			if (tsf->f_bavail > 0) {
6479 				if (sf->f_bsize != tsf->f_bsize) {
6480 					tot = tsf->f_bavail * tsf->f_bsize;
6481 					sf->f_bavail += (tot / sf->f_bsize);
6482 				} else
6483 					sf->f_bavail += tsf->f_bavail;
6484 			}
6485 		}
6486 	}
6487 	free(tsf, M_TEMP);
6488 	free(dvpp, M_TEMP);
6489 	return (error);
6490 }
6491 
6492 /*
6493  * Set an NFSv4 acl.
6494  */
6495 int
6496 nfsrv_setacl(struct vnode *vp, NFSACL_T *aclp, struct ucred *cred, NFSPROC_T *p)
6497 {
6498 	int error;
6499 
6500 	if (nfsrv_useacl == 0 || nfs_supportsnfsv4acls(vp) == 0) {
6501 		error = NFSERR_ATTRNOTSUPP;
6502 		goto out;
6503 	}
6504 	/*
6505 	 * With NFSv4 ACLs, chmod(2) may need to add additional entries.
6506 	 * Make sure it has enough room for that - splitting every entry
6507 	 * into two and appending "canonical six" entries at the end.
6508 	 * Cribbed out of kern/vfs_acl.c - Rick M.
6509 	 */
6510 	if (aclp->acl_cnt > (ACL_MAX_ENTRIES - 6) / 2) {
6511 		error = NFSERR_ATTRNOTSUPP;
6512 		goto out;
6513 	}
6514 	error = VOP_SETACL(vp, ACL_TYPE_NFS4, aclp, cred, p);
6515 	if (error == 0) {
6516 		error = nfsrv_dssetacl(vp, aclp, cred, p);
6517 		if (error == ENOENT)
6518 			error = 0;
6519 	}
6520 
6521 out:
6522 	NFSEXITCODE(error);
6523 	return (error);
6524 }
6525 
6526 /*
6527  * Seek vnode op call (actually it is a VOP_IOCTL()).
6528  * This function is called with the vnode locked, but unlocks and vrele()s
6529  * the vp before returning.
6530  */
6531 int
6532 nfsvno_seek(struct nfsrv_descript *nd, struct vnode *vp, u_long cmd,
6533     off_t *offp, int content, bool *eofp, struct ucred *cred, NFSPROC_T *p)
6534 {
6535 	struct nfsvattr at;
6536 	int error, ret;
6537 
6538 	ASSERT_VOP_LOCKED(vp, "nfsvno_seek vp");
6539 	/*
6540 	 * Attempt to seek on a DS file. A return of ENOENT implies
6541 	 * there is no DS file to seek on.
6542 	 */
6543 	error = nfsrv_proxyds(vp, 0, 0, cred, p, NFSPROC_SEEKDS, NULL,
6544 	    NULL, NULL, NULL, NULL, offp, content, eofp);
6545 	if (error != ENOENT) {
6546 		vput(vp);
6547 		return (error);
6548 	}
6549 
6550 	/*
6551 	 * Do the VOP_IOCTL() call.  For the case where *offp == file_size,
6552 	 * VOP_IOCTL() will return ENXIO.  However, the correct reply for
6553 	 * NFSv4.2 is *eofp == true and error == 0 for this case.
6554 	 */
6555 	NFSVOPUNLOCK(vp);
6556 	error = VOP_IOCTL(vp, cmd, offp, 0, cred, p);
6557 	*eofp = false;
6558 	if (error == ENXIO || (error == 0 && cmd == FIOSEEKHOLE)) {
6559 		/* Handle the cases where we might be at EOF. */
6560 		ret = nfsvno_getattr(vp, &at, nd, p, 0, NULL);
6561 		if (ret == 0 && *offp == at.na_size) {
6562 			*eofp = true;
6563 			error = 0;
6564 		}
6565 		if (ret != 0 && error == 0)
6566 			error = ret;
6567 	}
6568 	vrele(vp);
6569 	NFSEXITCODE(error);
6570 	return (error);
6571 }
6572 
6573 /*
6574  * Allocate vnode op call.
6575  */
6576 int
6577 nfsvno_allocate(struct vnode *vp, off_t off, off_t len, struct ucred *cred,
6578     NFSPROC_T *p)
6579 {
6580 	int error;
6581 	off_t olen;
6582 
6583 	ASSERT_VOP_ELOCKED(vp, "nfsvno_allocate vp");
6584 	/*
6585 	 * Attempt to allocate on a DS file. A return of ENOENT implies
6586 	 * there is no DS file to allocate on.
6587 	 */
6588 	error = nfsrv_proxyds(vp, off, 0, cred, p, NFSPROC_ALLOCATE, NULL,
6589 	    NULL, NULL, NULL, NULL, &len, 0, NULL);
6590 	if (error != ENOENT)
6591 		return (error);
6592 
6593 	/*
6594 	 * Do the actual VOP_ALLOCATE(), looping so long as
6595 	 * progress is being made, to achieve completion.
6596 	 */
6597 	do {
6598 		olen = len;
6599 		error = VOP_ALLOCATE(vp, &off, &len, IO_SYNC, cred);
6600 		if (error == 0 && len > 0 && olen > len)
6601 			maybe_yield();
6602 	} while (error == 0 && len > 0 && olen > len);
6603 	if (error == 0 && len > 0)
6604 		error = NFSERR_IO;
6605 	NFSEXITCODE(error);
6606 	return (error);
6607 }
6608 
6609 /*
6610  * Deallocate vnode op call.
6611  */
6612 int
6613 nfsvno_deallocate(struct vnode *vp, off_t off, off_t len, struct ucred *cred,
6614     NFSPROC_T *p)
6615 {
6616 	int error;
6617 	off_t olen;
6618 
6619 	ASSERT_VOP_ELOCKED(vp, "nfsvno_deallocate vp");
6620 	/*
6621 	 * Attempt to deallocate on a DS file. A return of ENOENT implies
6622 	 * there is no DS file to deallocate on.
6623 	 */
6624 	error = nfsrv_proxyds(vp, off, 0, cred, p, NFSPROC_DEALLOCATE, NULL,
6625 	    NULL, NULL, NULL, NULL, &len, 0, NULL);
6626 	if (error != ENOENT)
6627 		return (error);
6628 
6629 	/*
6630 	 * Do the actual VOP_DEALLOCATE(), looping so long as
6631 	 * progress is being made, to achieve completion.
6632 	 */
6633 	do {
6634 		olen = len;
6635 		error = VOP_DEALLOCATE(vp, &off, &len, 0, IO_SYNC, cred);
6636 		if (error == 0 && len > 0 && olen > len)
6637 			maybe_yield();
6638 	} while (error == 0 && len > 0 && olen > len);
6639 	if (error == 0 && len > 0)
6640 		error = NFSERR_IO;
6641 	NFSEXITCODE(error);
6642 	return (error);
6643 }
6644 
6645 /*
6646  * Get Extended Atribute vnode op into an mbuf list.
6647  */
6648 int
6649 nfsvno_getxattr(struct vnode *vp, char *name, uint32_t maxresp,
6650     struct ucred *cred, uint64_t flag, int maxextsiz, struct thread *p,
6651     struct mbuf **mpp, struct mbuf **mpendp, int *lenp)
6652 {
6653 	struct iovec *iv;
6654 	struct uio io, *uiop = &io;
6655 	struct mbuf *m, *m2;
6656 	int alen, error, len, tlen;
6657 	size_t siz;
6658 
6659 	/* First, find out the size of the extended attribute. */
6660 	error = VOP_GETEXTATTR(vp, EXTATTR_NAMESPACE_USER, name, NULL,
6661 	    &siz, cred, p);
6662 	if (error != 0)
6663 		return (NFSERR_NOXATTR);
6664 	if (siz > maxresp - NFS_MAXXDR)
6665 		return (NFSERR_XATTR2BIG);
6666 	len = siz;
6667 	tlen = NFSM_RNDUP(len);
6668 	if (tlen > 0) {
6669 		/*
6670 		 * If cnt > MCLBYTES and the reply will not be saved, use
6671 		 * ext_pgs mbufs for TLS.
6672 		 * For NFSv4.0, we do not know for sure if the reply will
6673 		 * be saved, so do not use ext_pgs mbufs for NFSv4.0.
6674 		 * Always use ext_pgs mbufs if ND_EXTPG is set.
6675 		 */
6676 		if ((flag & ND_EXTPG) != 0 || (tlen > MCLBYTES &&
6677 		    (flag & (ND_TLS | ND_SAVEREPLY)) == ND_TLS &&
6678 		    (flag & (ND_NFSV4 | ND_NFSV41)) != ND_NFSV4))
6679 			uiop->uio_iovcnt = nfsrv_createiovec_extpgs(tlen,
6680 			    maxextsiz, &m, &m2, &iv);
6681 		else
6682 			uiop->uio_iovcnt = nfsrv_createiovec(tlen, &m, &m2,
6683 			    &iv);
6684 		uiop->uio_iov = iv;
6685 	} else {
6686 		uiop->uio_iovcnt = 0;
6687 		uiop->uio_iov = iv = NULL;
6688 		m = m2 = NULL;
6689 	}
6690 	uiop->uio_offset = 0;
6691 	uiop->uio_resid = tlen;
6692 	uiop->uio_rw = UIO_READ;
6693 	uiop->uio_segflg = UIO_SYSSPACE;
6694 	uiop->uio_td = p;
6695 #ifdef MAC
6696 	error = mac_vnode_check_getextattr(cred, vp, EXTATTR_NAMESPACE_USER,
6697 	    name);
6698 	if (error != 0)
6699 		goto out;
6700 #endif
6701 
6702 	if (tlen > 0)
6703 		error = VOP_GETEXTATTR(vp, EXTATTR_NAMESPACE_USER, name, uiop,
6704 		    NULL, cred, p);
6705 	if (error != 0)
6706 		goto out;
6707 	if (uiop->uio_resid > 0) {
6708 		alen = tlen;
6709 		len = tlen - uiop->uio_resid;
6710 		tlen = NFSM_RNDUP(len);
6711 		if (alen != tlen)
6712 			printf("nfsvno_getxattr: weird size read\n");
6713 		if (tlen == 0) {
6714 			m_freem(m);
6715 			m = m2 = NULL;
6716 		} else if (alen != tlen || tlen != len)
6717 			m2 = nfsrv_adj(m, alen - tlen, tlen - len);
6718 	}
6719 	*lenp = len;
6720 	*mpp = m;
6721 	*mpendp = m2;
6722 
6723 out:
6724 	if (error != 0) {
6725 		if (m != NULL)
6726 			m_freem(m);
6727 		*lenp = 0;
6728 	}
6729 	free(iv, M_TEMP);
6730 	NFSEXITCODE(error);
6731 	return (error);
6732 }
6733 
6734 /*
6735  * Set Extended attribute vnode op from an mbuf list.
6736  */
6737 int
6738 nfsvno_setxattr(struct vnode *vp, char *name, int len, struct mbuf *m,
6739     char *cp, struct ucred *cred, struct thread *p)
6740 {
6741 	struct iovec *iv;
6742 	struct uio uio, *uiop = &uio;
6743 	int cnt, error;
6744 
6745 	error = 0;
6746 #ifdef MAC
6747 	error = mac_vnode_check_setextattr(cred, vp, EXTATTR_NAMESPACE_USER,
6748 	    name);
6749 #endif
6750 	if (error != 0)
6751 		goto out;
6752 
6753 	uiop->uio_rw = UIO_WRITE;
6754 	uiop->uio_segflg = UIO_SYSSPACE;
6755 	uiop->uio_td = p;
6756 	uiop->uio_offset = 0;
6757 	uiop->uio_resid = len;
6758 	if (len > 0) {
6759 		error = nfsrv_createiovecw(len, m, cp, &iv, &cnt);
6760 		uiop->uio_iov = iv;
6761 		uiop->uio_iovcnt = cnt;
6762 	} else {
6763 		uiop->uio_iov = iv = NULL;
6764 		uiop->uio_iovcnt = 0;
6765 	}
6766 	if (error == 0) {
6767 		error = VOP_SETEXTATTR(vp, EXTATTR_NAMESPACE_USER, name, uiop,
6768 		    cred, p);
6769 		free(iv, M_TEMP);
6770 	}
6771 
6772 out:
6773 	NFSEXITCODE(error);
6774 	return (error);
6775 }
6776 
6777 /*
6778  * Remove Extended attribute vnode op.
6779  */
6780 int
6781 nfsvno_rmxattr(struct nfsrv_descript *nd, struct vnode *vp, char *name,
6782     struct ucred *cred, struct thread *p)
6783 {
6784 	int error;
6785 
6786 	/*
6787 	 * Get rid of any delegations.  I am not sure why this is required,
6788 	 * but RFC-8276 says so.
6789 	 */
6790 	error = nfsrv_checkremove(vp, 0, nd, nd->nd_clientid, p);
6791 	if (error != 0)
6792 		goto out;
6793 #ifdef MAC
6794 	error = mac_vnode_check_deleteextattr(cred, vp, EXTATTR_NAMESPACE_USER,
6795 	    name);
6796 	if (error != 0)
6797 		goto out;
6798 #endif
6799 
6800 	error = VOP_DELETEEXTATTR(vp, EXTATTR_NAMESPACE_USER, name, cred, p);
6801 	if (error == EOPNOTSUPP)
6802 		error = VOP_SETEXTATTR(vp, EXTATTR_NAMESPACE_USER, name, NULL,
6803 		    cred, p);
6804 out:
6805 	NFSEXITCODE(error);
6806 	return (error);
6807 }
6808 
6809 /*
6810  * List Extended Atribute vnode op into an mbuf list.
6811  */
6812 int
6813 nfsvno_listxattr(struct vnode *vp, uint64_t cookie, struct ucred *cred,
6814     struct thread *p, u_char **bufp, uint32_t *lenp, bool *eofp)
6815 {
6816 	struct iovec iv;
6817 	struct uio io;
6818 	int error;
6819 	size_t siz;
6820 
6821 	*bufp = NULL;
6822 	/* First, find out the size of the extended attribute. */
6823 	error = VOP_LISTEXTATTR(vp, EXTATTR_NAMESPACE_USER, NULL, &siz, cred,
6824 	    p);
6825 	if (error != 0)
6826 		return (NFSERR_NOXATTR);
6827 	if (siz <= cookie) {
6828 		*lenp = 0;
6829 		*eofp = true;
6830 		goto out;
6831 	}
6832 	if (siz > cookie + *lenp) {
6833 		siz = cookie + *lenp;
6834 		*eofp = false;
6835 	} else
6836 		*eofp = true;
6837 	/* Just choose a sanity limit of 10Mbytes for malloc(M_TEMP). */
6838 	if (siz > 10 * 1024 * 1024) {
6839 		error = NFSERR_XATTR2BIG;
6840 		goto out;
6841 	}
6842 	*bufp = malloc(siz, M_TEMP, M_WAITOK);
6843 	iv.iov_base = *bufp;
6844 	iv.iov_len = siz;
6845 	io.uio_iovcnt = 1;
6846 	io.uio_iov = &iv;
6847 	io.uio_offset = 0;
6848 	io.uio_resid = siz;
6849 	io.uio_rw = UIO_READ;
6850 	io.uio_segflg = UIO_SYSSPACE;
6851 	io.uio_td = p;
6852 #ifdef MAC
6853 	error = mac_vnode_check_listextattr(cred, vp, EXTATTR_NAMESPACE_USER);
6854 	if (error != 0)
6855 		goto out;
6856 #endif
6857 
6858 	error = VOP_LISTEXTATTR(vp, EXTATTR_NAMESPACE_USER, &io, NULL, cred,
6859 	    p);
6860 	if (error != 0)
6861 		goto out;
6862 	if (io.uio_resid > 0)
6863 		siz -= io.uio_resid;
6864 	*lenp = siz;
6865 
6866 out:
6867 	if (error != 0) {
6868 		free(*bufp, M_TEMP);
6869 		*bufp = NULL;
6870 	}
6871 	NFSEXITCODE(error);
6872 	return (error);
6873 }
6874 
6875 /*
6876  * Trim trailing data off the mbuf list being built.
6877  */
6878 void
6879 nfsm_trimtrailing(struct nfsrv_descript *nd, struct mbuf *mb, char *bpos,
6880     int bextpg, int bextpgsiz)
6881 {
6882 	vm_page_t pg;
6883 	int fullpgsiz, i;
6884 
6885 	if (mb->m_next != NULL) {
6886 		m_freem(mb->m_next);
6887 		mb->m_next = NULL;
6888 	}
6889 	if ((mb->m_flags & M_EXTPG) != 0) {
6890 		KASSERT(bextpg >= 0 && bextpg < mb->m_epg_npgs,
6891 		    ("nfsm_trimtrailing: bextpg out of range"));
6892 		KASSERT(bpos == (char *)(void *)
6893 		    PHYS_TO_DMAP(mb->m_epg_pa[bextpg]) + PAGE_SIZE - bextpgsiz,
6894 		    ("nfsm_trimtrailing: bextpgsiz bad!"));
6895 
6896 		/* First, get rid of any pages after this position. */
6897 		for (i = mb->m_epg_npgs - 1; i > bextpg; i--) {
6898 			pg = PHYS_TO_VM_PAGE(mb->m_epg_pa[i]);
6899 			vm_page_unwire_noq(pg);
6900 			vm_page_free(pg);
6901 		}
6902 		mb->m_epg_npgs = bextpg + 1;
6903 		if (bextpg == 0)
6904 			fullpgsiz = PAGE_SIZE - mb->m_epg_1st_off;
6905 		else
6906 			fullpgsiz = PAGE_SIZE;
6907 		mb->m_epg_last_len = fullpgsiz - bextpgsiz;
6908 		mb->m_len = m_epg_pagelen(mb, 0, mb->m_epg_1st_off);
6909 		for (i = 1; i < mb->m_epg_npgs; i++)
6910 			mb->m_len += m_epg_pagelen(mb, i, 0);
6911 		nd->nd_bextpgsiz = bextpgsiz;
6912 		nd->nd_bextpg = bextpg;
6913 	} else
6914 		mb->m_len = bpos - mtod(mb, char *);
6915 	nd->nd_mb = mb;
6916 	nd->nd_bpos = bpos;
6917 }
6918 
6919 
6920 /*
6921  * Check to see if a put file handle operation should test for
6922  * NFSERR_WRONGSEC, although NFSv3 actually returns NFSERR_AUTHERR.
6923  * When Open is the next operation, NFSERR_WRONGSEC cannot be
6924  * replied for the Open cases that use a component.  Thia can
6925  * be identified by the fact that the file handle's type is VDIR.
6926  */
6927 bool
6928 nfsrv_checkwrongsec(struct nfsrv_descript *nd, int nextop, enum vtype vtyp)
6929 {
6930 
6931 	if ((nd->nd_flag & ND_NFSV4) == 0) {
6932 		if (nd->nd_procnum == NFSPROC_FSINFO)
6933 			return (false);
6934 		return (true);
6935 	}
6936 
6937 	if ((nd->nd_flag & ND_LASTOP) != 0)
6938 		return (false);
6939 
6940 	if (nextop == NFSV4OP_PUTROOTFH || nextop == NFSV4OP_PUTFH ||
6941 	    nextop == NFSV4OP_PUTPUBFH || nextop == NFSV4OP_RESTOREFH ||
6942 	    nextop == NFSV4OP_LOOKUP || nextop == NFSV4OP_LOOKUPP ||
6943 	    nextop == NFSV4OP_SECINFO || nextop == NFSV4OP_SECINFONONAME)
6944 		return (false);
6945 	if (nextop == NFSV4OP_OPEN && vtyp == VDIR)
6946 		return (false);
6947 	return (true);
6948 }
6949 
6950 /*
6951  * Check DSs marked no space.
6952  */
6953 void
6954 nfsrv_checknospc(void)
6955 {
6956 	struct statfs *tsf;
6957 	struct nfsdevice *ds;
6958 	struct vnode **dvpp, **tdvpp, *dvp;
6959 	char *devid, *tdevid;
6960 	int cnt, error = 0, i;
6961 
6962 	if (nfsrv_devidcnt <= 0)
6963 		return;
6964 	dvpp = mallocarray(nfsrv_devidcnt, sizeof(*dvpp), M_TEMP, M_WAITOK);
6965 	devid = malloc(nfsrv_devidcnt * NFSX_V4DEVICEID, M_TEMP, M_WAITOK);
6966 	tsf = malloc(sizeof(*tsf), M_TEMP, M_WAITOK);
6967 
6968 	/* Get an array of the dvps for the DSs. */
6969 	tdvpp = dvpp;
6970 	tdevid = devid;
6971 	i = 0;
6972 	NFSDDSLOCK();
6973 	/* First, search for matches for same file system. */
6974 	TAILQ_FOREACH(ds, &nfsrv_devidhead, nfsdev_list) {
6975 		if (ds->nfsdev_nmp != NULL && ds->nfsdev_nospc) {
6976 			if (++i > nfsrv_devidcnt)
6977 				break;
6978 			*tdvpp++ = ds->nfsdev_dvp;
6979 			NFSBCOPY(ds->nfsdev_deviceid, tdevid, NFSX_V4DEVICEID);
6980 			tdevid += NFSX_V4DEVICEID;
6981 		}
6982 	}
6983 	NFSDDSUNLOCK();
6984 
6985 	/* Do a VFS_STATFS() for each of the DSs and clear no space. */
6986 	cnt = i;
6987 	tdvpp = dvpp;
6988 	tdevid = devid;
6989 	for (i = 0; i < cnt && error == 0; i++) {
6990 		dvp = *tdvpp++;
6991 		error = VFS_STATFS(dvp->v_mount, tsf);
6992 		if (error == 0 && tsf->f_bavail > 0) {
6993 			NFSD_DEBUG(1, "nfsrv_checknospc: reset nospc\n");
6994 			nfsrv_marknospc(tdevid, false);
6995 		}
6996 		tdevid += NFSX_V4DEVICEID;
6997 	}
6998 	free(tsf, M_TEMP);
6999 	free(dvpp, M_TEMP);
7000 	free(devid, M_TEMP);
7001 }
7002 
7003 extern int (*nfsd_call_nfsd)(struct thread *, struct nfssvc_args *);
7004 
7005 /*
7006  * Called once to initialize data structures...
7007  */
7008 static int
7009 nfsd_modevent(module_t mod, int type, void *data)
7010 {
7011 	int error = 0, i;
7012 	static int loaded = 0;
7013 
7014 	switch (type) {
7015 	case MOD_LOAD:
7016 		if (loaded)
7017 			goto out;
7018 		newnfs_portinit();
7019 		for (i = 0; i < NFSRVCACHE_HASHSIZE; i++) {
7020 			mtx_init(&nfsrchash_table[i].mtx, "nfsrtc", NULL,
7021 			    MTX_DEF);
7022 			mtx_init(&nfsrcahash_table[i].mtx, "nfsrtca", NULL,
7023 			    MTX_DEF);
7024 		}
7025 		mtx_init(&nfsrc_udpmtx, "nfsuc", NULL, MTX_DEF);
7026 		mtx_init(&nfs_v4root_mutex, "nfs4rt", NULL, MTX_DEF);
7027 		mtx_init(&nfsv4root_mnt.mnt_mtx, "nfs4mnt", NULL, MTX_DEF);
7028 		mtx_init(&nfsrv_dontlistlock_mtx, "nfs4dnl", NULL, MTX_DEF);
7029 		mtx_init(&nfsrv_recalllock_mtx, "nfs4rec", NULL, MTX_DEF);
7030 		lockinit(&nfsv4root_mnt.mnt_explock, PVFS, "explock", 0, 0);
7031 		nfsrvd_initcache();
7032 		nfsd_init();
7033 		NFSD_LOCK();
7034 		nfsrvd_init(0);
7035 		NFSD_UNLOCK();
7036 		nfsd_mntinit();
7037 #ifdef VV_DISABLEDELEG
7038 		vn_deleg_ops.vndeleg_recall = nfsd_recalldelegation;
7039 		vn_deleg_ops.vndeleg_disable = nfsd_disabledelegation;
7040 #endif
7041 		nfsd_call_servertimer = nfsrv_servertimer;
7042 		nfsd_call_nfsd = nfssvc_nfsd;
7043 		loaded = 1;
7044 		break;
7045 
7046 	case MOD_UNLOAD:
7047 		if (newnfs_numnfsd != 0) {
7048 			error = EBUSY;
7049 			break;
7050 		}
7051 
7052 #ifdef VV_DISABLEDELEG
7053 		vn_deleg_ops.vndeleg_recall = NULL;
7054 		vn_deleg_ops.vndeleg_disable = NULL;
7055 #endif
7056 		nfsd_call_servertimer = NULL;
7057 		nfsd_call_nfsd = NULL;
7058 
7059 		/* Clean out all NFSv4 state. */
7060 		nfsrv_throwawayallstate(curthread);
7061 
7062 		/* Clean the NFS server reply cache */
7063 		nfsrvd_cleancache();
7064 
7065 		/* Free up the krpc server pool. */
7066 		if (nfsrvd_pool != NULL)
7067 			svcpool_destroy(nfsrvd_pool);
7068 
7069 		/* and get rid of the locks */
7070 		for (i = 0; i < NFSRVCACHE_HASHSIZE; i++) {
7071 			mtx_destroy(&nfsrchash_table[i].mtx);
7072 			mtx_destroy(&nfsrcahash_table[i].mtx);
7073 		}
7074 		mtx_destroy(&nfsrc_udpmtx);
7075 		mtx_destroy(&nfs_v4root_mutex);
7076 		mtx_destroy(&nfsv4root_mnt.mnt_mtx);
7077 		mtx_destroy(&nfsrv_dontlistlock_mtx);
7078 		mtx_destroy(&nfsrv_recalllock_mtx);
7079 		for (i = 0; i < nfsrv_sessionhashsize; i++)
7080 			mtx_destroy(&nfssessionhash[i].mtx);
7081 		if (nfslayouthash != NULL) {
7082 			for (i = 0; i < nfsrv_layouthashsize; i++)
7083 				mtx_destroy(&nfslayouthash[i].mtx);
7084 			free(nfslayouthash, M_NFSDSESSION);
7085 		}
7086 		lockdestroy(&nfsv4root_mnt.mnt_explock);
7087 		free(nfsclienthash, M_NFSDCLIENT);
7088 		free(nfslockhash, M_NFSDLOCKFILE);
7089 		free(nfssessionhash, M_NFSDSESSION);
7090 		loaded = 0;
7091 		break;
7092 	default:
7093 		error = EOPNOTSUPP;
7094 		break;
7095 	}
7096 
7097 out:
7098 	NFSEXITCODE(error);
7099 	return (error);
7100 }
7101 static moduledata_t nfsd_mod = {
7102 	"nfsd",
7103 	nfsd_modevent,
7104 	NULL,
7105 };
7106 DECLARE_MODULE(nfsd, nfsd_mod, SI_SUB_VFS, SI_ORDER_ANY);
7107 
7108 /* So that loader and kldload(2) can find us, wherever we are.. */
7109 MODULE_VERSION(nfsd, 1);
7110 MODULE_DEPEND(nfsd, nfscommon, 1, 1, 1);
7111 MODULE_DEPEND(nfsd, nfslockd, 1, 1, 1);
7112 MODULE_DEPEND(nfsd, krpc, 1, 1, 1);
7113 MODULE_DEPEND(nfsd, nfssvc, 1, 1, 1);
7114