xref: /freebsd/sys/fs/nfsserver/nfs_nfsdport.c (revision 058ac3e8063366dafa634d9107642e12b038bf09)
1 /*-
2  * SPDX-License-Identifier: BSD-3-Clause
3  *
4  * Copyright (c) 1989, 1993
5  *	The Regents of the University of California.  All rights reserved.
6  *
7  * This code is derived from software contributed to Berkeley by
8  * Rick Macklem at The University of Guelph.
9  *
10  * Redistribution and use in source and binary forms, with or without
11  * modification, are permitted provided that the following conditions
12  * are met:
13  * 1. Redistributions of source code must retain the above copyright
14  *    notice, this list of conditions and the following disclaimer.
15  * 2. Redistributions in binary form must reproduce the above copyright
16  *    notice, this list of conditions and the following disclaimer in the
17  *    documentation and/or other materials provided with the distribution.
18  * 3. Neither the name of the University nor the names of its contributors
19  *    may be used to endorse or promote products derived from this software
20  *    without specific prior written permission.
21  *
22  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
23  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
24  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
25  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
26  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
27  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
28  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
29  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
30  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
31  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
32  * SUCH DAMAGE.
33  *
34  */
35 
36 #include <sys/cdefs.h>
37 __FBSDID("$FreeBSD$");
38 
39 #include <sys/capsicum.h>
40 #include <sys/extattr.h>
41 
42 /*
43  * Functions that perform the vfs operations required by the routines in
44  * nfsd_serv.c. It is hoped that this change will make the server more
45  * portable.
46  */
47 
48 #include <fs/nfs/nfsport.h>
49 #include <security/mac/mac_framework.h>
50 #include <sys/callout.h>
51 #include <sys/filio.h>
52 #include <sys/hash.h>
53 #include <sys/sysctl.h>
54 #include <nlm/nlm_prot.h>
55 #include <nlm/nlm.h>
56 
57 FEATURE(nfsd, "NFSv4 server");
58 
59 extern u_int32_t newnfs_true, newnfs_false, newnfs_xdrneg1;
60 extern int nfsrv_useacl;
61 extern int newnfs_numnfsd;
62 extern struct mount nfsv4root_mnt;
63 extern struct nfsrv_stablefirst nfsrv_stablefirst;
64 extern SVCPOOL	*nfsrvd_pool;
65 extern struct nfsv4lock nfsd_suspend_lock;
66 extern struct nfsclienthashhead *nfsclienthash;
67 extern struct nfslockhashhead *nfslockhash;
68 extern struct nfssessionhash *nfssessionhash;
69 extern int nfsrv_sessionhashsize;
70 extern struct nfsstatsv1 nfsstatsv1;
71 extern struct nfslayouthash *nfslayouthash;
72 extern int nfsrv_layouthashsize;
73 extern struct mtx nfsrv_dslock_mtx;
74 extern int nfs_pnfsiothreads;
75 extern struct nfsdontlisthead nfsrv_dontlisthead;
76 extern volatile int nfsrv_dontlistlen;
77 extern volatile int nfsrv_devidcnt;
78 extern int nfsrv_maxpnfsmirror;
79 extern uint32_t nfs_srvmaxio;
80 extern int nfs_bufpackets;
81 extern u_long sb_max_adj;
82 struct vfsoptlist nfsv4root_opt, nfsv4root_newopt;
83 NFSDLOCKMUTEX;
84 NFSSTATESPINLOCK;
85 struct nfsrchash_bucket nfsrchash_table[NFSRVCACHE_HASHSIZE];
86 struct nfsrchash_bucket nfsrcahash_table[NFSRVCACHE_HASHSIZE];
87 struct mtx nfsrc_udpmtx;
88 struct mtx nfs_v4root_mutex;
89 struct mtx nfsrv_dontlistlock_mtx;
90 struct mtx nfsrv_recalllock_mtx;
91 struct nfsrvfh nfs_rootfh, nfs_pubfh;
92 int nfs_pubfhset = 0, nfs_rootfhset = 0;
93 struct proc *nfsd_master_proc = NULL;
94 int nfsd_debuglevel = 0;
95 static pid_t nfsd_master_pid = (pid_t)-1;
96 static char nfsd_master_comm[MAXCOMLEN + 1];
97 static struct timeval nfsd_master_start;
98 static uint32_t nfsv4_sysid = 0;
99 static fhandle_t zerofh;
100 struct callout nfsd_callout;
101 
102 static int nfssvc_srvcall(struct thread *, struct nfssvc_args *,
103     struct ucred *);
104 static void nfsvno_updateds(struct vnode *, struct ucred *, struct thread *);
105 
106 int nfsrv_enable_crossmntpt = 1;
107 static int nfs_commit_blks;
108 static int nfs_commit_miss;
109 extern int nfsrv_issuedelegs;
110 extern int nfsrv_dolocallocks;
111 extern int nfsd_enable_stringtouid;
112 extern struct nfsdevicehead nfsrv_devidhead;
113 
114 static int nfsrv_createiovec(int, struct mbuf **, struct mbuf **,
115     struct iovec **);
116 static int nfsrv_createiovec_extpgs(int, int, struct mbuf **,
117     struct mbuf **, struct iovec **);
118 static int nfsrv_createiovecw(int, struct mbuf *, char *, struct iovec **,
119     int *);
120 static void nfsrv_pnfscreate(struct vnode *, struct vattr *, struct ucred *,
121     NFSPROC_T *);
122 static void nfsrv_pnfsremovesetup(struct vnode *, NFSPROC_T *, struct vnode **,
123     int *, char *, fhandle_t *);
124 static void nfsrv_pnfsremove(struct vnode **, int, char *, fhandle_t *,
125     NFSPROC_T *);
126 static int nfsrv_proxyds(struct vnode *, off_t, int, struct ucred *,
127     struct thread *, int, struct mbuf **, char *, struct mbuf **,
128     struct nfsvattr *, struct acl *, off_t *, int, bool *);
129 static int nfsrv_setextattr(struct vnode *, struct nfsvattr *, NFSPROC_T *);
130 static int nfsrv_readdsrpc(fhandle_t *, off_t, int, struct ucred *,
131     NFSPROC_T *, struct nfsmount *, struct mbuf **, struct mbuf **);
132 static int nfsrv_writedsrpc(fhandle_t *, off_t, int, struct ucred *,
133     NFSPROC_T *, struct vnode *, struct nfsmount **, int, struct mbuf **,
134     char *, int *);
135 static int nfsrv_allocatedsrpc(fhandle_t *, off_t, off_t, struct ucred *,
136     NFSPROC_T *, struct vnode *, struct nfsmount **, int, int *);
137 static int nfsrv_deallocatedsrpc(fhandle_t *, off_t, off_t, struct ucred *,
138     NFSPROC_T *, struct vnode *, struct nfsmount **, int, int *);
139 static int nfsrv_setacldsrpc(fhandle_t *, struct ucred *, NFSPROC_T *,
140     struct vnode *, struct nfsmount **, int, struct acl *, int *);
141 static int nfsrv_setattrdsrpc(fhandle_t *, struct ucred *, NFSPROC_T *,
142     struct vnode *, struct nfsmount **, int, struct nfsvattr *, int *);
143 static int nfsrv_getattrdsrpc(fhandle_t *, struct ucred *, NFSPROC_T *,
144     struct vnode *, struct nfsmount *, struct nfsvattr *);
145 static int nfsrv_seekdsrpc(fhandle_t *, off_t *, int, bool *, struct ucred *,
146     NFSPROC_T *, struct nfsmount *);
147 static int nfsrv_putfhname(fhandle_t *, char *);
148 static int nfsrv_pnfslookupds(struct vnode *, struct vnode *,
149     struct pnfsdsfile *, struct vnode **, NFSPROC_T *);
150 static void nfsrv_pnfssetfh(struct vnode *, struct pnfsdsfile *, char *, char *,
151     struct vnode *, NFSPROC_T *);
152 static int nfsrv_dsremove(struct vnode *, char *, struct ucred *, NFSPROC_T *);
153 static int nfsrv_dssetacl(struct vnode *, struct acl *, struct ucred *,
154     NFSPROC_T *);
155 static int nfsrv_pnfsstatfs(struct statfs *, struct mount *);
156 
157 int nfs_pnfsio(task_fn_t *, void *);
158 
159 SYSCTL_NODE(_vfs, OID_AUTO, nfsd, CTLFLAG_RW | CTLFLAG_MPSAFE, 0,
160     "NFS server");
161 SYSCTL_INT(_vfs_nfsd, OID_AUTO, mirrormnt, CTLFLAG_RW,
162     &nfsrv_enable_crossmntpt, 0, "Enable nfsd to cross mount points");
163 SYSCTL_INT(_vfs_nfsd, OID_AUTO, commit_blks, CTLFLAG_RW, &nfs_commit_blks,
164     0, "");
165 SYSCTL_INT(_vfs_nfsd, OID_AUTO, commit_miss, CTLFLAG_RW, &nfs_commit_miss,
166     0, "");
167 SYSCTL_INT(_vfs_nfsd, OID_AUTO, issue_delegations, CTLFLAG_RW,
168     &nfsrv_issuedelegs, 0, "Enable nfsd to issue delegations");
169 SYSCTL_INT(_vfs_nfsd, OID_AUTO, enable_locallocks, CTLFLAG_RW,
170     &nfsrv_dolocallocks, 0, "Enable nfsd to acquire local locks on files");
171 SYSCTL_INT(_vfs_nfsd, OID_AUTO, debuglevel, CTLFLAG_RW, &nfsd_debuglevel,
172     0, "Debug level for NFS server");
173 SYSCTL_INT(_vfs_nfsd, OID_AUTO, enable_stringtouid, CTLFLAG_RW,
174     &nfsd_enable_stringtouid, 0, "Enable nfsd to accept numeric owner_names");
175 static int nfsrv_pnfsgetdsattr = 1;
176 SYSCTL_INT(_vfs_nfsd, OID_AUTO, pnfsgetdsattr, CTLFLAG_RW,
177     &nfsrv_pnfsgetdsattr, 0, "When set getattr gets DS attributes via RPC");
178 
179 /*
180  * nfsrv_dsdirsize can only be increased and only when the nfsd threads are
181  * not running.
182  * The dsN subdirectories for the increased values must have been created
183  * on all DS servers before this increase is done.
184  */
185 u_int	nfsrv_dsdirsize = 20;
186 static int
187 sysctl_dsdirsize(SYSCTL_HANDLER_ARGS)
188 {
189 	int error, newdsdirsize;
190 
191 	newdsdirsize = nfsrv_dsdirsize;
192 	error = sysctl_handle_int(oidp, &newdsdirsize, 0, req);
193 	if (error != 0 || req->newptr == NULL)
194 		return (error);
195 	if (newdsdirsize <= nfsrv_dsdirsize || newdsdirsize > 10000 ||
196 	    newnfs_numnfsd != 0)
197 		return (EINVAL);
198 	nfsrv_dsdirsize = newdsdirsize;
199 	return (0);
200 }
201 SYSCTL_PROC(_vfs_nfsd, OID_AUTO, dsdirsize,
202     CTLTYPE_UINT | CTLFLAG_MPSAFE | CTLFLAG_RW, 0, sizeof(nfsrv_dsdirsize),
203     sysctl_dsdirsize, "IU", "Number of dsN subdirs on the DS servers");
204 
205 /*
206  * nfs_srvmaxio can only be increased and only when the nfsd threads are
207  * not running.  The setting must be a power of 2, with the current limit of
208  * 1Mbyte.
209  */
210 static int
211 sysctl_srvmaxio(SYSCTL_HANDLER_ARGS)
212 {
213 	int error;
214 	u_int newsrvmaxio;
215 	uint64_t tval;
216 
217 	newsrvmaxio = nfs_srvmaxio;
218 	error = sysctl_handle_int(oidp, &newsrvmaxio, 0, req);
219 	if (error != 0 || req->newptr == NULL)
220 		return (error);
221 	if (newsrvmaxio == nfs_srvmaxio)
222 		return (0);
223 	if (newsrvmaxio < nfs_srvmaxio) {
224 		printf("nfsd: vfs.nfsd.srvmaxio can only be increased\n");
225 		return (EINVAL);
226 	}
227 	if (newsrvmaxio > 1048576) {
228 		printf("nfsd: vfs.nfsd.srvmaxio cannot be > 1Mbyte\n");
229 		return (EINVAL);
230 	}
231 	if ((newsrvmaxio & (newsrvmaxio - 1)) != 0) {
232 		printf("nfsd: vfs.nfsd.srvmaxio must be a power of 2\n");
233 		return (EINVAL);
234 	}
235 
236 	/*
237 	 * Check that kern.ipc.maxsockbuf is large enough for
238 	 * newsrviomax, given the setting of vfs.nfs.bufpackets.
239 	 */
240 	if ((newsrvmaxio + NFS_MAXXDR) * nfs_bufpackets >
241 	    sb_max_adj) {
242 		/*
243 		 * Suggest vfs.nfs.bufpackets * maximum RPC message for
244 		 * sb_max_adj.
245 		 */
246 		tval = (newsrvmaxio + NFS_MAXXDR) * nfs_bufpackets;
247 
248 		/*
249 		 * Convert suggested sb_max_adj value to a suggested
250 		 * sb_max value, which is what is set via kern.ipc.maxsockbuf.
251 		 * Perform the inverse calculation of (from uipc_sockbuf.c):
252 		 * sb_max_adj = (u_quad_t)sb_max * MCLBYTES /
253 		 *     (MSIZE + MCLBYTES);
254 		 * XXX If the calculation of sb_max_adj from sb_max changes,
255 		 *     this calculation must be changed as well.
256 		 */
257 		tval *= (MSIZE + MCLBYTES);  /* Brackets for readability. */
258 		tval += MCLBYTES - 1;        /* Round up divide. */
259 		tval /= MCLBYTES;
260 		printf("nfsd: set kern.ipc.maxsockbuf to a minimum of "
261 		    "%ju to support %ubyte NFS I/O\n", (uintmax_t)tval,
262 		    newsrvmaxio);
263 		return (EINVAL);
264 	}
265 
266 	NFSD_LOCK();
267 	if (newnfs_numnfsd != 0) {
268 		NFSD_UNLOCK();
269 		printf("nfsd: cannot set vfs.nfsd.srvmaxio when nfsd "
270 		    "threads are running\n");
271 		return (EINVAL);
272 	}
273 
274 
275 	nfs_srvmaxio = newsrvmaxio;
276 	NFSD_UNLOCK();
277 	return (0);
278 }
279 SYSCTL_PROC(_vfs_nfsd, OID_AUTO, srvmaxio,
280     CTLTYPE_UINT | CTLFLAG_MPSAFE | CTLFLAG_RW, NULL, 0,
281     sysctl_srvmaxio, "IU", "Maximum I/O size in bytes");
282 
283 #define	MAX_REORDERED_RPC	16
284 #define	NUM_HEURISTIC		1031
285 #define	NHUSE_INIT		64
286 #define	NHUSE_INC		16
287 #define	NHUSE_MAX		2048
288 
289 static struct nfsheur {
290 	struct vnode *nh_vp;	/* vp to match (unreferenced pointer) */
291 	off_t nh_nextoff;	/* next offset for sequential detection */
292 	int nh_use;		/* use count for selection */
293 	int nh_seqcount;	/* heuristic */
294 } nfsheur[NUM_HEURISTIC];
295 
296 /*
297  * Heuristic to detect sequential operation.
298  */
299 static struct nfsheur *
300 nfsrv_sequential_heuristic(struct uio *uio, struct vnode *vp)
301 {
302 	struct nfsheur *nh;
303 	int hi, try;
304 
305 	/* Locate best candidate. */
306 	try = 32;
307 	hi = ((int)(vm_offset_t)vp / sizeof(struct vnode)) % NUM_HEURISTIC;
308 	nh = &nfsheur[hi];
309 	while (try--) {
310 		if (nfsheur[hi].nh_vp == vp) {
311 			nh = &nfsheur[hi];
312 			break;
313 		}
314 		if (nfsheur[hi].nh_use > 0)
315 			--nfsheur[hi].nh_use;
316 		hi = (hi + 1) % NUM_HEURISTIC;
317 		if (nfsheur[hi].nh_use < nh->nh_use)
318 			nh = &nfsheur[hi];
319 	}
320 
321 	/* Initialize hint if this is a new file. */
322 	if (nh->nh_vp != vp) {
323 		nh->nh_vp = vp;
324 		nh->nh_nextoff = uio->uio_offset;
325 		nh->nh_use = NHUSE_INIT;
326 		if (uio->uio_offset == 0)
327 			nh->nh_seqcount = 4;
328 		else
329 			nh->nh_seqcount = 1;
330 	}
331 
332 	/* Calculate heuristic. */
333 	if ((uio->uio_offset == 0 && nh->nh_seqcount > 0) ||
334 	    uio->uio_offset == nh->nh_nextoff) {
335 		/* See comments in vfs_vnops.c:sequential_heuristic(). */
336 		nh->nh_seqcount += howmany(uio->uio_resid, 16384);
337 		if (nh->nh_seqcount > IO_SEQMAX)
338 			nh->nh_seqcount = IO_SEQMAX;
339 	} else if (qabs(uio->uio_offset - nh->nh_nextoff) <= MAX_REORDERED_RPC *
340 	    imax(vp->v_mount->mnt_stat.f_iosize, uio->uio_resid)) {
341 		/* Probably a reordered RPC, leave seqcount alone. */
342 	} else if (nh->nh_seqcount > 1) {
343 		nh->nh_seqcount /= 2;
344 	} else {
345 		nh->nh_seqcount = 0;
346 	}
347 	nh->nh_use += NHUSE_INC;
348 	if (nh->nh_use > NHUSE_MAX)
349 		nh->nh_use = NHUSE_MAX;
350 	return (nh);
351 }
352 
353 /*
354  * Get attributes into nfsvattr structure.
355  */
356 int
357 nfsvno_getattr(struct vnode *vp, struct nfsvattr *nvap,
358     struct nfsrv_descript *nd, struct thread *p, int vpislocked,
359     nfsattrbit_t *attrbitp)
360 {
361 	int error, gotattr, lockedit = 0;
362 	struct nfsvattr na;
363 
364 	if (vpislocked == 0) {
365 		/*
366 		 * When vpislocked == 0, the vnode is either exclusively
367 		 * locked by this thread or not locked by this thread.
368 		 * As such, shared lock it, if not exclusively locked.
369 		 */
370 		if (NFSVOPISLOCKED(vp) != LK_EXCLUSIVE) {
371 			lockedit = 1;
372 			NFSVOPLOCK(vp, LK_SHARED | LK_RETRY);
373 		}
374 	}
375 
376 	/*
377 	 * Acquire the Change, Size, TimeAccess, TimeModify and SpaceUsed
378 	 * attributes, as required.
379 	 * This needs to be done for regular files if:
380 	 * - non-NFSv4 RPCs or
381 	 * - when attrbitp == NULL or
382 	 * - an NFSv4 RPC with any of the above attributes in attrbitp.
383 	 * A return of 0 for nfsrv_proxyds() indicates that it has acquired
384 	 * these attributes.  nfsrv_proxyds() will return an error if the
385 	 * server is not a pNFS one.
386 	 */
387 	gotattr = 0;
388 	if (vp->v_type == VREG && nfsrv_devidcnt > 0 && (attrbitp == NULL ||
389 	    (nd->nd_flag & ND_NFSV4) == 0 ||
390 	    NFSISSET_ATTRBIT(attrbitp, NFSATTRBIT_CHANGE) ||
391 	    NFSISSET_ATTRBIT(attrbitp, NFSATTRBIT_SIZE) ||
392 	    NFSISSET_ATTRBIT(attrbitp, NFSATTRBIT_TIMEACCESS) ||
393 	    NFSISSET_ATTRBIT(attrbitp, NFSATTRBIT_TIMEMODIFY) ||
394 	    NFSISSET_ATTRBIT(attrbitp, NFSATTRBIT_SPACEUSED))) {
395 		error = nfsrv_proxyds(vp, 0, 0, nd->nd_cred, p,
396 		    NFSPROC_GETATTR, NULL, NULL, NULL, &na, NULL, NULL, 0,
397 		    NULL);
398 		if (error == 0)
399 			gotattr = 1;
400 	}
401 
402 	error = VOP_GETATTR(vp, &nvap->na_vattr, nd->nd_cred);
403 	if (lockedit != 0)
404 		NFSVOPUNLOCK(vp);
405 
406 	/*
407 	 * If we got the Change, Size and Modify Time from the DS,
408 	 * replace them.
409 	 */
410 	if (gotattr != 0) {
411 		nvap->na_atime = na.na_atime;
412 		nvap->na_mtime = na.na_mtime;
413 		nvap->na_filerev = na.na_filerev;
414 		nvap->na_size = na.na_size;
415 		nvap->na_bytes = na.na_bytes;
416 	}
417 	NFSD_DEBUG(4, "nfsvno_getattr: gotattr=%d err=%d chg=%ju\n", gotattr,
418 	    error, (uintmax_t)na.na_filerev);
419 
420 	NFSEXITCODE(error);
421 	return (error);
422 }
423 
424 /*
425  * Get a file handle for a vnode.
426  */
427 int
428 nfsvno_getfh(struct vnode *vp, fhandle_t *fhp, struct thread *p)
429 {
430 	int error;
431 
432 	NFSBZERO((caddr_t)fhp, sizeof(fhandle_t));
433 	fhp->fh_fsid = vp->v_mount->mnt_stat.f_fsid;
434 	error = VOP_VPTOFH(vp, &fhp->fh_fid);
435 
436 	NFSEXITCODE(error);
437 	return (error);
438 }
439 
440 /*
441  * Perform access checking for vnodes obtained from file handles that would
442  * refer to files already opened by a Unix client. You cannot just use
443  * vn_writechk() and VOP_ACCESSX() for two reasons.
444  * 1 - You must check for exported rdonly as well as MNT_RDONLY for the write
445  *     case.
446  * 2 - The owner is to be given access irrespective of mode bits for some
447  *     operations, so that processes that chmod after opening a file don't
448  *     break.
449  */
450 int
451 nfsvno_accchk(struct vnode *vp, accmode_t accmode, struct ucred *cred,
452     struct nfsexstuff *exp, struct thread *p, int override, int vpislocked,
453     u_int32_t *supportedtypep)
454 {
455 	struct vattr vattr;
456 	int error = 0, getret = 0;
457 
458 	if (vpislocked == 0) {
459 		if (NFSVOPLOCK(vp, LK_SHARED) != 0) {
460 			error = EPERM;
461 			goto out;
462 		}
463 	}
464 	if (accmode & VWRITE) {
465 		/* Just vn_writechk() changed to check rdonly */
466 		/*
467 		 * Disallow write attempts on read-only file systems;
468 		 * unless the file is a socket or a block or character
469 		 * device resident on the file system.
470 		 */
471 		if (NFSVNO_EXRDONLY(exp) ||
472 		    (vp->v_mount->mnt_flag & MNT_RDONLY)) {
473 			switch (vp->v_type) {
474 			case VREG:
475 			case VDIR:
476 			case VLNK:
477 				error = EROFS;
478 			default:
479 				break;
480 			}
481 		}
482 		/*
483 		 * If there's shared text associated with
484 		 * the inode, try to free it up once.  If
485 		 * we fail, we can't allow writing.
486 		 */
487 		if (VOP_IS_TEXT(vp) && error == 0)
488 			error = ETXTBSY;
489 	}
490 	if (error != 0) {
491 		if (vpislocked == 0)
492 			NFSVOPUNLOCK(vp);
493 		goto out;
494 	}
495 
496 	/*
497 	 * Should the override still be applied when ACLs are enabled?
498 	 */
499 	error = VOP_ACCESSX(vp, accmode, cred, p);
500 	if (error != 0 && (accmode & (VDELETE | VDELETE_CHILD))) {
501 		/*
502 		 * Try again with VEXPLICIT_DENY, to see if the test for
503 		 * deletion is supported.
504 		 */
505 		error = VOP_ACCESSX(vp, accmode | VEXPLICIT_DENY, cred, p);
506 		if (error == 0) {
507 			if (vp->v_type == VDIR) {
508 				accmode &= ~(VDELETE | VDELETE_CHILD);
509 				accmode |= VWRITE;
510 				error = VOP_ACCESSX(vp, accmode, cred, p);
511 			} else if (supportedtypep != NULL) {
512 				*supportedtypep &= ~NFSACCESS_DELETE;
513 			}
514 		}
515 	}
516 
517 	/*
518 	 * Allow certain operations for the owner (reads and writes
519 	 * on files that are already open).
520 	 */
521 	if (override != NFSACCCHK_NOOVERRIDE &&
522 	    (error == EPERM || error == EACCES)) {
523 		if (cred->cr_uid == 0 && (override & NFSACCCHK_ALLOWROOT))
524 			error = 0;
525 		else if (override & NFSACCCHK_ALLOWOWNER) {
526 			getret = VOP_GETATTR(vp, &vattr, cred);
527 			if (getret == 0 && cred->cr_uid == vattr.va_uid)
528 				error = 0;
529 		}
530 	}
531 	if (vpislocked == 0)
532 		NFSVOPUNLOCK(vp);
533 
534 out:
535 	NFSEXITCODE(error);
536 	return (error);
537 }
538 
539 /*
540  * Set attribute(s) vnop.
541  */
542 int
543 nfsvno_setattr(struct vnode *vp, struct nfsvattr *nvap, struct ucred *cred,
544     struct thread *p, struct nfsexstuff *exp)
545 {
546 	u_quad_t savsize = 0;
547 	int error, savedit;
548 	time_t savbtime;
549 
550 	/*
551 	 * If this is an exported file system and a pNFS service is running,
552 	 * don't VOP_SETATTR() of size for the MDS file system.
553 	 */
554 	savedit = 0;
555 	error = 0;
556 	if (vp->v_type == VREG && (vp->v_mount->mnt_flag & MNT_EXPORTED) != 0 &&
557 	    nfsrv_devidcnt != 0 && nvap->na_vattr.va_size != VNOVAL &&
558 	    nvap->na_vattr.va_size > 0) {
559 		savsize = nvap->na_vattr.va_size;
560 		nvap->na_vattr.va_size = VNOVAL;
561 		if (nvap->na_vattr.va_uid != (uid_t)VNOVAL ||
562 		    nvap->na_vattr.va_gid != (gid_t)VNOVAL ||
563 		    nvap->na_vattr.va_mode != (mode_t)VNOVAL ||
564 		    nvap->na_vattr.va_atime.tv_sec != VNOVAL ||
565 		    nvap->na_vattr.va_mtime.tv_sec != VNOVAL)
566 			savedit = 1;
567 		else
568 			savedit = 2;
569 	}
570 	if (savedit != 2)
571 		error = VOP_SETATTR(vp, &nvap->na_vattr, cred);
572 	if (savedit != 0)
573 		nvap->na_vattr.va_size = savsize;
574 	if (error == 0 && (nvap->na_vattr.va_uid != (uid_t)VNOVAL ||
575 	    nvap->na_vattr.va_gid != (gid_t)VNOVAL ||
576 	    nvap->na_vattr.va_size != VNOVAL ||
577 	    nvap->na_vattr.va_mode != (mode_t)VNOVAL ||
578 	    nvap->na_vattr.va_atime.tv_sec != VNOVAL ||
579 	    nvap->na_vattr.va_mtime.tv_sec != VNOVAL)) {
580 		/* Never modify birthtime on a DS file. */
581 		savbtime = nvap->na_vattr.va_birthtime.tv_sec;
582 		nvap->na_vattr.va_birthtime.tv_sec = VNOVAL;
583 		/* For a pNFS server, set the attributes on the DS file. */
584 		error = nfsrv_proxyds(vp, 0, 0, cred, p, NFSPROC_SETATTR,
585 		    NULL, NULL, NULL, nvap, NULL, NULL, 0, NULL);
586 		nvap->na_vattr.va_birthtime.tv_sec = savbtime;
587 		if (error == ENOENT)
588 			error = 0;
589 	}
590 	NFSEXITCODE(error);
591 	return (error);
592 }
593 
594 /*
595  * Set up nameidata for a lookup() call and do it.
596  */
597 int
598 nfsvno_namei(struct nfsrv_descript *nd, struct nameidata *ndp,
599     struct vnode *dp, int islocked, struct nfsexstuff *exp,
600     struct vnode **retdirp)
601 {
602 	struct componentname *cnp = &ndp->ni_cnd;
603 	int i;
604 	struct iovec aiov;
605 	struct uio auio;
606 	int lockleaf = (cnp->cn_flags & LOCKLEAF) != 0, linklen;
607 	int error = 0;
608 	char *cp;
609 
610 	*retdirp = NULL;
611 	cnp->cn_nameptr = cnp->cn_pnbuf;
612 	ndp->ni_lcf = 0;
613 	/*
614 	 * Extract and set starting directory.
615 	 */
616 	if (dp->v_type != VDIR) {
617 		if (islocked)
618 			vput(dp);
619 		else
620 			vrele(dp);
621 		nfsvno_relpathbuf(ndp);
622 		error = ENOTDIR;
623 		goto out1;
624 	}
625 	if (islocked)
626 		NFSVOPUNLOCK(dp);
627 	VREF(dp);
628 	*retdirp = dp;
629 	if (NFSVNO_EXRDONLY(exp))
630 		cnp->cn_flags |= RDONLY;
631 	ndp->ni_segflg = UIO_SYSSPACE;
632 
633 	if (nd->nd_flag & ND_PUBLOOKUP) {
634 		ndp->ni_loopcnt = 0;
635 		if (cnp->cn_pnbuf[0] == '/') {
636 			vrele(dp);
637 			/*
638 			 * Check for degenerate pathnames here, since lookup()
639 			 * panics on them.
640 			 */
641 			for (i = 1; i < ndp->ni_pathlen; i++)
642 				if (cnp->cn_pnbuf[i] != '/')
643 					break;
644 			if (i == ndp->ni_pathlen) {
645 				error = NFSERR_ACCES;
646 				goto out;
647 			}
648 			dp = rootvnode;
649 			VREF(dp);
650 		}
651 	} else if ((nfsrv_enable_crossmntpt == 0 && NFSVNO_EXPORTED(exp)) ||
652 	    (nd->nd_flag & ND_NFSV4) == 0) {
653 		/*
654 		 * Only cross mount points for NFSv4 when doing a
655 		 * mount while traversing the file system above
656 		 * the mount point, unless nfsrv_enable_crossmntpt is set.
657 		 */
658 		cnp->cn_flags |= NOCROSSMOUNT;
659 	}
660 
661 	/*
662 	 * Initialize for scan, set ni_startdir and bump ref on dp again
663 	 * because lookup() will dereference ni_startdir.
664 	 */
665 
666 	ndp->ni_startdir = dp;
667 	ndp->ni_rootdir = rootvnode;
668 	ndp->ni_topdir = NULL;
669 
670 	if (!lockleaf)
671 		cnp->cn_flags |= LOCKLEAF;
672 	for (;;) {
673 		cnp->cn_nameptr = cnp->cn_pnbuf;
674 		/*
675 		 * Call lookup() to do the real work.  If an error occurs,
676 		 * ndp->ni_vp and ni_dvp are left uninitialized or NULL and
677 		 * we do not have to dereference anything before returning.
678 		 * In either case ni_startdir will be dereferenced and NULLed
679 		 * out.
680 		 */
681 		error = vfs_lookup(ndp);
682 		if (error)
683 			break;
684 
685 		/*
686 		 * Check for encountering a symbolic link.  Trivial
687 		 * termination occurs if no symlink encountered.
688 		 */
689 		if ((cnp->cn_flags & ISSYMLINK) == 0) {
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);
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 	uma_zfree(namei_zone, ndp->ni_cnd.cn_pnbuf);
813 	ndp->ni_cnd.cn_pnbuf = NULL;
814 }
815 
816 /*
817  * Readlink vnode op into an mbuf list.
818  */
819 int
820 nfsvno_readlink(struct vnode *vp, struct ucred *cred, int maxextsiz,
821     struct thread *p, struct mbuf **mpp, struct mbuf **mpendp, int *lenp)
822 {
823 	struct iovec *iv;
824 	struct uio io, *uiop = &io;
825 	struct mbuf *mp, *mp3;
826 	int len, tlen, error = 0;
827 
828 	len = NFS_MAXPATHLEN;
829 	if (maxextsiz > 0)
830 		uiop->uio_iovcnt = nfsrv_createiovec_extpgs(len, maxextsiz,
831 		    &mp3, &mp, &iv);
832 	else
833 		uiop->uio_iovcnt = nfsrv_createiovec(len, &mp3, &mp, &iv);
834 	uiop->uio_iov = iv;
835 	uiop->uio_offset = 0;
836 	uiop->uio_resid = len;
837 	uiop->uio_rw = UIO_READ;
838 	uiop->uio_segflg = UIO_SYSSPACE;
839 	uiop->uio_td = NULL;
840 	error = VOP_READLINK(vp, uiop, cred);
841 	free(iv, M_TEMP);
842 	if (error) {
843 		m_freem(mp3);
844 		*lenp = 0;
845 		goto out;
846 	}
847 	if (uiop->uio_resid > 0) {
848 		len -= uiop->uio_resid;
849 		tlen = NFSM_RNDUP(len);
850 		if (tlen == 0) {
851 			m_freem(mp3);
852 			mp3 = mp = NULL;
853 		} else if (tlen != NFS_MAXPATHLEN || tlen != len)
854 			mp = nfsrv_adj(mp3, NFS_MAXPATHLEN - tlen,
855 			    tlen - len);
856 	}
857 	*lenp = len;
858 	*mpp = mp3;
859 	*mpendp = mp;
860 
861 out:
862 	NFSEXITCODE(error);
863 	return (error);
864 }
865 
866 /*
867  * Create an mbuf chain and an associated iovec that can be used to Read
868  * or Getextattr of data.
869  * Upon success, return pointers to the first and last mbufs in the chain
870  * plus the malloc'd iovec and its iovlen.
871  */
872 static int
873 nfsrv_createiovec(int len, struct mbuf **mpp, struct mbuf **mpendp,
874     struct iovec **ivp)
875 {
876 	struct mbuf *m, *m2 = NULL, *m3;
877 	struct iovec *iv;
878 	int i, left, siz;
879 
880 	left = len;
881 	m3 = NULL;
882 	/*
883 	 * Generate the mbuf list with the uio_iov ref. to it.
884 	 */
885 	i = 0;
886 	while (left > 0) {
887 		NFSMGET(m);
888 		MCLGET(m, M_WAITOK);
889 		m->m_len = 0;
890 		siz = min(M_TRAILINGSPACE(m), left);
891 		left -= siz;
892 		i++;
893 		if (m3)
894 			m2->m_next = m;
895 		else
896 			m3 = m;
897 		m2 = m;
898 	}
899 	*ivp = iv = malloc(i * sizeof (struct iovec), M_TEMP, M_WAITOK);
900 	m = m3;
901 	left = len;
902 	i = 0;
903 	while (left > 0) {
904 		if (m == NULL)
905 			panic("nfsrv_createiovec iov");
906 		siz = min(M_TRAILINGSPACE(m), left);
907 		if (siz > 0) {
908 			iv->iov_base = mtod(m, caddr_t) + m->m_len;
909 			iv->iov_len = siz;
910 			m->m_len += siz;
911 			left -= siz;
912 			iv++;
913 			i++;
914 		}
915 		m = m->m_next;
916 	}
917 	*mpp = m3;
918 	*mpendp = m2;
919 	return (i);
920 }
921 
922 /*
923  * Create an mbuf chain and an associated iovec that can be used to Read
924  * or Getextattr of data.
925  * Upon success, return pointers to the first and last mbufs in the chain
926  * plus the malloc'd iovec and its iovlen.
927  * Same as above, but creates ext_pgs mbuf(s).
928  */
929 static int
930 nfsrv_createiovec_extpgs(int len, int maxextsiz, struct mbuf **mpp,
931     struct mbuf **mpendp, struct iovec **ivp)
932 {
933 	struct mbuf *m, *m2 = NULL, *m3;
934 	struct iovec *iv;
935 	int i, left, pgno, siz;
936 
937 	left = len;
938 	m3 = NULL;
939 	/*
940 	 * Generate the mbuf list with the uio_iov ref. to it.
941 	 */
942 	i = 0;
943 	while (left > 0) {
944 		siz = min(left, maxextsiz);
945 		m = mb_alloc_ext_plus_pages(siz, M_WAITOK);
946 		left -= siz;
947 		i += m->m_epg_npgs;
948 		if (m3 != NULL)
949 			m2->m_next = m;
950 		else
951 			m3 = m;
952 		m2 = m;
953 	}
954 	*ivp = iv = malloc(i * sizeof (struct iovec), M_TEMP, M_WAITOK);
955 	m = m3;
956 	left = len;
957 	i = 0;
958 	pgno = 0;
959 	while (left > 0) {
960 		if (m == NULL)
961 			panic("nfsvno_createiovec_extpgs iov");
962 		siz = min(PAGE_SIZE, left);
963 		if (siz > 0) {
964 			iv->iov_base = (void *)PHYS_TO_DMAP(m->m_epg_pa[pgno]);
965 			iv->iov_len = siz;
966 			m->m_len += siz;
967 			if (pgno == m->m_epg_npgs - 1)
968 				m->m_epg_last_len = siz;
969 			left -= siz;
970 			iv++;
971 			i++;
972 			pgno++;
973 		}
974 		if (pgno == m->m_epg_npgs && left > 0) {
975 			m = m->m_next;
976 			if (m == NULL)
977 				panic("nfsvno_createiovec_extpgs iov");
978 			pgno = 0;
979 		}
980 	}
981 	*mpp = m3;
982 	*mpendp = m2;
983 	return (i);
984 }
985 
986 /*
987  * Read vnode op call into mbuf list.
988  */
989 int
990 nfsvno_read(struct vnode *vp, off_t off, int cnt, struct ucred *cred,
991     int maxextsiz, struct thread *p, struct mbuf **mpp,
992     struct mbuf **mpendp)
993 {
994 	struct mbuf *m;
995 	struct iovec *iv;
996 	int error = 0, len, tlen, ioflag = 0;
997 	struct mbuf *m3;
998 	struct uio io, *uiop = &io;
999 	struct nfsheur *nh;
1000 
1001 	/*
1002 	 * Attempt to read from a DS file. A return of ENOENT implies
1003 	 * there is no DS file to read.
1004 	 */
1005 	error = nfsrv_proxyds(vp, off, cnt, cred, p, NFSPROC_READDS, mpp,
1006 	    NULL, mpendp, NULL, NULL, NULL, 0, NULL);
1007 	if (error != ENOENT)
1008 		return (error);
1009 
1010 	len = NFSM_RNDUP(cnt);
1011 	if (maxextsiz > 0)
1012 		uiop->uio_iovcnt = nfsrv_createiovec_extpgs(len, maxextsiz,
1013 		    &m3, &m, &iv);
1014 	else
1015 		uiop->uio_iovcnt = nfsrv_createiovec(len, &m3, &m, &iv);
1016 	uiop->uio_iov = iv;
1017 	uiop->uio_offset = off;
1018 	uiop->uio_resid = len;
1019 	uiop->uio_rw = UIO_READ;
1020 	uiop->uio_segflg = UIO_SYSSPACE;
1021 	uiop->uio_td = NULL;
1022 	nh = nfsrv_sequential_heuristic(uiop, vp);
1023 	ioflag |= nh->nh_seqcount << IO_SEQSHIFT;
1024 	/* XXX KDM make this more systematic? */
1025 	nfsstatsv1.srvbytes[NFSV4OP_READ] += uiop->uio_resid;
1026 	error = VOP_READ(vp, uiop, IO_NODELOCKED | ioflag, cred);
1027 	free(iv, M_TEMP);
1028 	if (error) {
1029 		m_freem(m3);
1030 		*mpp = NULL;
1031 		goto out;
1032 	}
1033 	nh->nh_nextoff = uiop->uio_offset;
1034 	tlen = len - uiop->uio_resid;
1035 	cnt = cnt < tlen ? cnt : tlen;
1036 	tlen = NFSM_RNDUP(cnt);
1037 	if (tlen == 0) {
1038 		m_freem(m3);
1039 		m3 = m = NULL;
1040 	} else if (len != tlen || tlen != cnt)
1041 		m = nfsrv_adj(m3, len - tlen, tlen - cnt);
1042 	*mpp = m3;
1043 	*mpendp = m;
1044 
1045 out:
1046 	NFSEXITCODE(error);
1047 	return (error);
1048 }
1049 
1050 /*
1051  * Create the iovec for the mbuf chain passed in as an argument.
1052  * The "cp" argument is where the data starts within the first mbuf in
1053  * the chain. It returns the iovec and the iovcnt.
1054  */
1055 static int
1056 nfsrv_createiovecw(int retlen, struct mbuf *m, char *cp, struct iovec **ivpp,
1057     int *iovcntp)
1058 {
1059 	struct mbuf *mp;
1060 	struct iovec *ivp;
1061 	int cnt, i, len;
1062 
1063 	/*
1064 	 * Loop through the mbuf chain, counting how many mbufs are a
1065 	 * part of this write operation, so the iovec size is known.
1066 	 */
1067 	cnt = 0;
1068 	len = retlen;
1069 	mp = m;
1070 	i = mtod(mp, caddr_t) + mp->m_len - cp;
1071 	while (len > 0) {
1072 		if (i > 0) {
1073 			len -= i;
1074 			cnt++;
1075 		}
1076 		mp = mp->m_next;
1077 		if (!mp) {
1078 			if (len > 0)
1079 				return (EBADRPC);
1080 		} else
1081 			i = mp->m_len;
1082 	}
1083 
1084 	/* Now, create the iovec. */
1085 	mp = m;
1086 	*ivpp = ivp = malloc(cnt * sizeof (struct iovec), M_TEMP,
1087 	    M_WAITOK);
1088 	*iovcntp = cnt;
1089 	i = mtod(mp, caddr_t) + mp->m_len - cp;
1090 	len = retlen;
1091 	while (len > 0) {
1092 		if (mp == NULL)
1093 			panic("nfsrv_createiovecw");
1094 		if (i > 0) {
1095 			i = min(i, len);
1096 			ivp->iov_base = cp;
1097 			ivp->iov_len = i;
1098 			ivp++;
1099 			len -= i;
1100 		}
1101 		mp = mp->m_next;
1102 		if (mp) {
1103 			i = mp->m_len;
1104 			cp = mtod(mp, caddr_t);
1105 		}
1106 	}
1107 	return (0);
1108 }
1109 
1110 /*
1111  * Write vnode op from an mbuf list.
1112  */
1113 int
1114 nfsvno_write(struct vnode *vp, off_t off, int retlen, int *stable,
1115     struct mbuf *mp, char *cp, struct ucred *cred, struct thread *p)
1116 {
1117 	struct iovec *iv;
1118 	int cnt, ioflags, error;
1119 	struct uio io, *uiop = &io;
1120 	struct nfsheur *nh;
1121 
1122 	/*
1123 	 * Attempt to write to a DS file. A return of ENOENT implies
1124 	 * there is no DS file to write.
1125 	 */
1126 	error = nfsrv_proxyds(vp, off, retlen, cred, p, NFSPROC_WRITEDS,
1127 	    &mp, cp, NULL, NULL, NULL, NULL, 0, NULL);
1128 	if (error != ENOENT) {
1129 		*stable = NFSWRITE_FILESYNC;
1130 		return (error);
1131 	}
1132 
1133 	if (*stable == NFSWRITE_UNSTABLE)
1134 		ioflags = IO_NODELOCKED;
1135 	else
1136 		ioflags = (IO_SYNC | IO_NODELOCKED);
1137 	error = nfsrv_createiovecw(retlen, mp, cp, &iv, &cnt);
1138 	if (error != 0)
1139 		return (error);
1140 	uiop->uio_iov = iv;
1141 	uiop->uio_iovcnt = cnt;
1142 	uiop->uio_resid = retlen;
1143 	uiop->uio_rw = UIO_WRITE;
1144 	uiop->uio_segflg = UIO_SYSSPACE;
1145 	NFSUIOPROC(uiop, p);
1146 	uiop->uio_offset = off;
1147 	nh = nfsrv_sequential_heuristic(uiop, vp);
1148 	ioflags |= nh->nh_seqcount << IO_SEQSHIFT;
1149 	/* XXX KDM make this more systematic? */
1150 	nfsstatsv1.srvbytes[NFSV4OP_WRITE] += uiop->uio_resid;
1151 	error = VOP_WRITE(vp, uiop, ioflags, cred);
1152 	if (error == 0)
1153 		nh->nh_nextoff = uiop->uio_offset;
1154 	free(iv, M_TEMP);
1155 
1156 	NFSEXITCODE(error);
1157 	return (error);
1158 }
1159 
1160 /*
1161  * Common code for creating a regular file (plus special files for V2).
1162  */
1163 int
1164 nfsvno_createsub(struct nfsrv_descript *nd, struct nameidata *ndp,
1165     struct vnode **vpp, struct nfsvattr *nvap, int *exclusive_flagp,
1166     int32_t *cverf, NFSDEV_T rdev, struct nfsexstuff *exp)
1167 {
1168 	u_quad_t tempsize;
1169 	int error;
1170 	struct thread *p = curthread;
1171 
1172 	error = nd->nd_repstat;
1173 	if (!error && ndp->ni_vp == NULL) {
1174 		if (nvap->na_type == VREG || nvap->na_type == VSOCK) {
1175 			error = VOP_CREATE(ndp->ni_dvp,
1176 			    &ndp->ni_vp, &ndp->ni_cnd, &nvap->na_vattr);
1177 			/* For a pNFS server, create the data file on a DS. */
1178 			if (error == 0 && nvap->na_type == VREG) {
1179 				/*
1180 				 * Create a data file on a DS for a pNFS server.
1181 				 * This function just returns if not
1182 				 * running a pNFS DS or the creation fails.
1183 				 */
1184 				nfsrv_pnfscreate(ndp->ni_vp, &nvap->na_vattr,
1185 				    nd->nd_cred, p);
1186 			}
1187 			VOP_VPUT_PAIR(ndp->ni_dvp, error == 0 ? &ndp->ni_vp :
1188 			    NULL, false);
1189 			nfsvno_relpathbuf(ndp);
1190 			if (!error) {
1191 				if (*exclusive_flagp) {
1192 					*exclusive_flagp = 0;
1193 					NFSVNO_ATTRINIT(nvap);
1194 					nvap->na_atime.tv_sec = cverf[0];
1195 					nvap->na_atime.tv_nsec = cverf[1];
1196 					error = VOP_SETATTR(ndp->ni_vp,
1197 					    &nvap->na_vattr, nd->nd_cred);
1198 					if (error != 0) {
1199 						vput(ndp->ni_vp);
1200 						ndp->ni_vp = NULL;
1201 						error = NFSERR_NOTSUPP;
1202 					}
1203 				}
1204 			}
1205 		/*
1206 		 * NFS V2 Only. nfsrvd_mknod() does this for V3.
1207 		 * (This implies, just get out on an error.)
1208 		 */
1209 		} else if (nvap->na_type == VCHR || nvap->na_type == VBLK ||
1210 			nvap->na_type == VFIFO) {
1211 			if (nvap->na_type == VCHR && rdev == 0xffffffff)
1212 				nvap->na_type = VFIFO;
1213                         if (nvap->na_type != VFIFO &&
1214 			    (error = priv_check_cred(nd->nd_cred, PRIV_VFS_MKNOD_DEV))) {
1215 				nfsvno_relpathbuf(ndp);
1216 				vput(ndp->ni_dvp);
1217 				goto out;
1218 			}
1219 			nvap->na_rdev = rdev;
1220 			error = VOP_MKNOD(ndp->ni_dvp, &ndp->ni_vp,
1221 			    &ndp->ni_cnd, &nvap->na_vattr);
1222 			VOP_VPUT_PAIR(ndp->ni_dvp, error == 0 ? &ndp->ni_vp :
1223 			    NULL, false);
1224 			nfsvno_relpathbuf(ndp);
1225 			if (error)
1226 				goto out;
1227 		} else {
1228 			nfsvno_relpathbuf(ndp);
1229 			vput(ndp->ni_dvp);
1230 			error = ENXIO;
1231 			goto out;
1232 		}
1233 		*vpp = ndp->ni_vp;
1234 	} else {
1235 		/*
1236 		 * Handle cases where error is already set and/or
1237 		 * the file exists.
1238 		 * 1 - clean up the lookup
1239 		 * 2 - iff !error and na_size set, truncate it
1240 		 */
1241 		nfsvno_relpathbuf(ndp);
1242 		*vpp = ndp->ni_vp;
1243 		if (ndp->ni_dvp == *vpp)
1244 			vrele(ndp->ni_dvp);
1245 		else
1246 			vput(ndp->ni_dvp);
1247 		if (!error && nvap->na_size != VNOVAL) {
1248 			error = nfsvno_accchk(*vpp, VWRITE,
1249 			    nd->nd_cred, exp, p, NFSACCCHK_NOOVERRIDE,
1250 			    NFSACCCHK_VPISLOCKED, NULL);
1251 			if (!error) {
1252 				tempsize = nvap->na_size;
1253 				NFSVNO_ATTRINIT(nvap);
1254 				nvap->na_size = tempsize;
1255 				error = nfsvno_setattr(*vpp, nvap,
1256 				    nd->nd_cred, p, exp);
1257 			}
1258 		}
1259 		if (error)
1260 			vput(*vpp);
1261 	}
1262 
1263 out:
1264 	NFSEXITCODE(error);
1265 	return (error);
1266 }
1267 
1268 /*
1269  * Do a mknod vnode op.
1270  */
1271 int
1272 nfsvno_mknod(struct nameidata *ndp, struct nfsvattr *nvap, struct ucred *cred,
1273     struct thread *p)
1274 {
1275 	int error = 0;
1276 	enum vtype vtyp;
1277 
1278 	vtyp = nvap->na_type;
1279 	/*
1280 	 * Iff doesn't exist, create it.
1281 	 */
1282 	if (ndp->ni_vp) {
1283 		nfsvno_relpathbuf(ndp);
1284 		vput(ndp->ni_dvp);
1285 		vrele(ndp->ni_vp);
1286 		error = EEXIST;
1287 		goto out;
1288 	}
1289 	if (vtyp != VCHR && vtyp != VBLK && vtyp != VSOCK && vtyp != VFIFO) {
1290 		nfsvno_relpathbuf(ndp);
1291 		vput(ndp->ni_dvp);
1292 		error = NFSERR_BADTYPE;
1293 		goto out;
1294 	}
1295 	if (vtyp == VSOCK) {
1296 		error = VOP_CREATE(ndp->ni_dvp, &ndp->ni_vp,
1297 		    &ndp->ni_cnd, &nvap->na_vattr);
1298 		VOP_VPUT_PAIR(ndp->ni_dvp, error == 0 ? &ndp->ni_vp : NULL,
1299 		    false);
1300 		nfsvno_relpathbuf(ndp);
1301 	} else {
1302 		if (nvap->na_type != VFIFO &&
1303 		    (error = priv_check_cred(cred, PRIV_VFS_MKNOD_DEV))) {
1304 			nfsvno_relpathbuf(ndp);
1305 			vput(ndp->ni_dvp);
1306 			goto out;
1307 		}
1308 		error = VOP_MKNOD(ndp->ni_dvp, &ndp->ni_vp,
1309 		    &ndp->ni_cnd, &nvap->na_vattr);
1310 		VOP_VPUT_PAIR(ndp->ni_dvp, error == 0 ? &ndp->ni_vp : NULL,
1311 		    false);
1312 		nfsvno_relpathbuf(ndp);
1313 		/*
1314 		 * Since VOP_MKNOD returns the ni_vp, I can't
1315 		 * see any reason to do the lookup.
1316 		 */
1317 	}
1318 
1319 out:
1320 	NFSEXITCODE(error);
1321 	return (error);
1322 }
1323 
1324 /*
1325  * Mkdir vnode op.
1326  */
1327 int
1328 nfsvno_mkdir(struct nameidata *ndp, struct nfsvattr *nvap, uid_t saved_uid,
1329     struct ucred *cred, struct thread *p, struct nfsexstuff *exp)
1330 {
1331 	int error = 0;
1332 
1333 	if (ndp->ni_vp != NULL) {
1334 		if (ndp->ni_dvp == ndp->ni_vp)
1335 			vrele(ndp->ni_dvp);
1336 		else
1337 			vput(ndp->ni_dvp);
1338 		vrele(ndp->ni_vp);
1339 		nfsvno_relpathbuf(ndp);
1340 		error = EEXIST;
1341 		goto out;
1342 	}
1343 	error = VOP_MKDIR(ndp->ni_dvp, &ndp->ni_vp, &ndp->ni_cnd,
1344 	    &nvap->na_vattr);
1345 	VOP_VPUT_PAIR(ndp->ni_dvp, error == 0 ? &ndp->ni_vp : NULL, false);
1346 	nfsvno_relpathbuf(ndp);
1347 
1348 out:
1349 	NFSEXITCODE(error);
1350 	return (error);
1351 }
1352 
1353 /*
1354  * symlink vnode op.
1355  */
1356 int
1357 nfsvno_symlink(struct nameidata *ndp, struct nfsvattr *nvap, char *pathcp,
1358     int pathlen, int not_v2, uid_t saved_uid, struct ucred *cred, struct thread *p,
1359     struct nfsexstuff *exp)
1360 {
1361 	int error = 0;
1362 
1363 	if (ndp->ni_vp) {
1364 		nfsvno_relpathbuf(ndp);
1365 		if (ndp->ni_dvp == ndp->ni_vp)
1366 			vrele(ndp->ni_dvp);
1367 		else
1368 			vput(ndp->ni_dvp);
1369 		vrele(ndp->ni_vp);
1370 		error = EEXIST;
1371 		goto out;
1372 	}
1373 
1374 	error = VOP_SYMLINK(ndp->ni_dvp, &ndp->ni_vp, &ndp->ni_cnd,
1375 	    &nvap->na_vattr, pathcp);
1376 	/*
1377 	 * Although FreeBSD still had the lookup code in
1378 	 * it for 7/current, there doesn't seem to be any
1379 	 * point, since VOP_SYMLINK() returns the ni_vp.
1380 	 * Just vput it for v2.
1381 	 */
1382 	VOP_VPUT_PAIR(ndp->ni_dvp, &ndp->ni_vp, !not_v2 && error == 0);
1383 	nfsvno_relpathbuf(ndp);
1384 
1385 out:
1386 	NFSEXITCODE(error);
1387 	return (error);
1388 }
1389 
1390 /*
1391  * Parse symbolic link arguments.
1392  * This function has an ugly side effect. It will malloc() an area for
1393  * the symlink and set iov_base to point to it, only if it succeeds.
1394  * So, if it returns with uiop->uio_iov->iov_base != NULL, that must
1395  * be FREE'd later.
1396  */
1397 int
1398 nfsvno_getsymlink(struct nfsrv_descript *nd, struct nfsvattr *nvap,
1399     struct thread *p, char **pathcpp, int *lenp)
1400 {
1401 	u_int32_t *tl;
1402 	char *pathcp = NULL;
1403 	int error = 0, len;
1404 	struct nfsv2_sattr *sp;
1405 
1406 	*pathcpp = NULL;
1407 	*lenp = 0;
1408 	if ((nd->nd_flag & ND_NFSV3) &&
1409 	    (error = nfsrv_sattr(nd, NULL, nvap, NULL, NULL, p)))
1410 		goto nfsmout;
1411 	NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
1412 	len = fxdr_unsigned(int, *tl);
1413 	if (len > NFS_MAXPATHLEN || len <= 0) {
1414 		error = EBADRPC;
1415 		goto nfsmout;
1416 	}
1417 	pathcp = malloc(len + 1, M_TEMP, M_WAITOK);
1418 	error = nfsrv_mtostr(nd, pathcp, len);
1419 	if (error)
1420 		goto nfsmout;
1421 	if (nd->nd_flag & ND_NFSV2) {
1422 		NFSM_DISSECT(sp, struct nfsv2_sattr *, NFSX_V2SATTR);
1423 		nvap->na_mode = fxdr_unsigned(u_int16_t, sp->sa_mode);
1424 	}
1425 	*pathcpp = pathcp;
1426 	*lenp = len;
1427 	NFSEXITCODE2(0, nd);
1428 	return (0);
1429 nfsmout:
1430 	if (pathcp)
1431 		free(pathcp, M_TEMP);
1432 	NFSEXITCODE2(error, nd);
1433 	return (error);
1434 }
1435 
1436 /*
1437  * Remove a non-directory object.
1438  */
1439 int
1440 nfsvno_removesub(struct nameidata *ndp, int is_v4, struct ucred *cred,
1441     struct thread *p, struct nfsexstuff *exp)
1442 {
1443 	struct vnode *vp, *dsdvp[NFSDEV_MAXMIRRORS];
1444 	int error = 0, mirrorcnt;
1445 	char fname[PNFS_FILENAME_LEN + 1];
1446 	fhandle_t fh;
1447 
1448 	vp = ndp->ni_vp;
1449 	dsdvp[0] = NULL;
1450 	if (vp->v_type == VDIR)
1451 		error = NFSERR_ISDIR;
1452 	else if (is_v4)
1453 		error = nfsrv_checkremove(vp, 1, NULL, (nfsquad_t)((u_quad_t)0),
1454 		    p);
1455 	if (error == 0)
1456 		nfsrv_pnfsremovesetup(vp, p, dsdvp, &mirrorcnt, fname, &fh);
1457 	if (!error)
1458 		error = VOP_REMOVE(ndp->ni_dvp, vp, &ndp->ni_cnd);
1459 	if (error == 0 && dsdvp[0] != NULL)
1460 		nfsrv_pnfsremove(dsdvp, mirrorcnt, fname, &fh, p);
1461 	if (ndp->ni_dvp == vp)
1462 		vrele(ndp->ni_dvp);
1463 	else
1464 		vput(ndp->ni_dvp);
1465 	vput(vp);
1466 	nfsvno_relpathbuf(ndp);
1467 	NFSEXITCODE(error);
1468 	return (error);
1469 }
1470 
1471 /*
1472  * Remove a directory.
1473  */
1474 int
1475 nfsvno_rmdirsub(struct nameidata *ndp, int is_v4, struct ucred *cred,
1476     struct thread *p, struct nfsexstuff *exp)
1477 {
1478 	struct vnode *vp;
1479 	int error = 0;
1480 
1481 	vp = ndp->ni_vp;
1482 	if (vp->v_type != VDIR) {
1483 		error = ENOTDIR;
1484 		goto out;
1485 	}
1486 	/*
1487 	 * No rmdir "." please.
1488 	 */
1489 	if (ndp->ni_dvp == vp) {
1490 		error = EINVAL;
1491 		goto out;
1492 	}
1493 	/*
1494 	 * The root of a mounted filesystem cannot be deleted.
1495 	 */
1496 	if (vp->v_vflag & VV_ROOT)
1497 		error = EBUSY;
1498 out:
1499 	if (!error)
1500 		error = VOP_RMDIR(ndp->ni_dvp, vp, &ndp->ni_cnd);
1501 	if (ndp->ni_dvp == vp)
1502 		vrele(ndp->ni_dvp);
1503 	else
1504 		vput(ndp->ni_dvp);
1505 	vput(vp);
1506 	nfsvno_relpathbuf(ndp);
1507 	NFSEXITCODE(error);
1508 	return (error);
1509 }
1510 
1511 /*
1512  * Rename vnode op.
1513  */
1514 int
1515 nfsvno_rename(struct nameidata *fromndp, struct nameidata *tondp,
1516     u_int32_t ndstat, u_int32_t ndflag, struct ucred *cred, struct thread *p)
1517 {
1518 	struct vnode *fvp, *tvp, *tdvp, *dsdvp[NFSDEV_MAXMIRRORS];
1519 	int error = 0, mirrorcnt;
1520 	char fname[PNFS_FILENAME_LEN + 1];
1521 	fhandle_t fh;
1522 
1523 	dsdvp[0] = NULL;
1524 	fvp = fromndp->ni_vp;
1525 	if (ndstat) {
1526 		vrele(fromndp->ni_dvp);
1527 		vrele(fvp);
1528 		error = ndstat;
1529 		goto out1;
1530 	}
1531 	tdvp = tondp->ni_dvp;
1532 	tvp = tondp->ni_vp;
1533 	if (tvp != NULL) {
1534 		if (fvp->v_type == VDIR && tvp->v_type != VDIR) {
1535 			error = (ndflag & ND_NFSV2) ? EISDIR : EEXIST;
1536 			goto out;
1537 		} else if (fvp->v_type != VDIR && tvp->v_type == VDIR) {
1538 			error = (ndflag & ND_NFSV2) ? ENOTDIR : EEXIST;
1539 			goto out;
1540 		}
1541 		if (tvp->v_type == VDIR && tvp->v_mountedhere) {
1542 			error = (ndflag & ND_NFSV2) ? ENOTEMPTY : EXDEV;
1543 			goto out;
1544 		}
1545 
1546 		/*
1547 		 * A rename to '.' or '..' results in a prematurely
1548 		 * unlocked vnode on FreeBSD5, so I'm just going to fail that
1549 		 * here.
1550 		 */
1551 		if ((tondp->ni_cnd.cn_namelen == 1 &&
1552 		     tondp->ni_cnd.cn_nameptr[0] == '.') ||
1553 		    (tondp->ni_cnd.cn_namelen == 2 &&
1554 		     tondp->ni_cnd.cn_nameptr[0] == '.' &&
1555 		     tondp->ni_cnd.cn_nameptr[1] == '.')) {
1556 			error = EINVAL;
1557 			goto out;
1558 		}
1559 	}
1560 	if (fvp->v_type == VDIR && fvp->v_mountedhere) {
1561 		error = (ndflag & ND_NFSV2) ? ENOTEMPTY : EXDEV;
1562 		goto out;
1563 	}
1564 	if (fvp->v_mount != tdvp->v_mount) {
1565 		error = (ndflag & ND_NFSV2) ? ENOTEMPTY : EXDEV;
1566 		goto out;
1567 	}
1568 	if (fvp == tdvp) {
1569 		error = (ndflag & ND_NFSV2) ? ENOTEMPTY : EINVAL;
1570 		goto out;
1571 	}
1572 	if (fvp == tvp) {
1573 		/*
1574 		 * If source and destination are the same, there is nothing to
1575 		 * do. Set error to -1 to indicate this.
1576 		 */
1577 		error = -1;
1578 		goto out;
1579 	}
1580 	if (ndflag & ND_NFSV4) {
1581 		if (NFSVOPLOCK(fvp, LK_EXCLUSIVE) == 0) {
1582 			error = nfsrv_checkremove(fvp, 0, NULL,
1583 			    (nfsquad_t)((u_quad_t)0), p);
1584 			NFSVOPUNLOCK(fvp);
1585 		} else
1586 			error = EPERM;
1587 		if (tvp && !error)
1588 			error = nfsrv_checkremove(tvp, 1, NULL,
1589 			    (nfsquad_t)((u_quad_t)0), p);
1590 	} else {
1591 		/*
1592 		 * For NFSv2 and NFSv3, try to get rid of the delegation, so
1593 		 * that the NFSv4 client won't be confused by the rename.
1594 		 * Since nfsd_recalldelegation() can only be called on an
1595 		 * unlocked vnode at this point and fvp is the file that will
1596 		 * still exist after the rename, just do fvp.
1597 		 */
1598 		nfsd_recalldelegation(fvp, p);
1599 	}
1600 	if (error == 0 && tvp != NULL) {
1601 		nfsrv_pnfsremovesetup(tvp, p, dsdvp, &mirrorcnt, fname, &fh);
1602 		NFSD_DEBUG(4, "nfsvno_rename: pnfsremovesetup"
1603 		    " dsdvp=%p\n", dsdvp[0]);
1604 	}
1605 out:
1606 	if (!error) {
1607 		error = VOP_RENAME(fromndp->ni_dvp, fromndp->ni_vp,
1608 		    &fromndp->ni_cnd, tondp->ni_dvp, tondp->ni_vp,
1609 		    &tondp->ni_cnd);
1610 	} else {
1611 		if (tdvp == tvp)
1612 			vrele(tdvp);
1613 		else
1614 			vput(tdvp);
1615 		if (tvp)
1616 			vput(tvp);
1617 		vrele(fromndp->ni_dvp);
1618 		vrele(fvp);
1619 		if (error == -1)
1620 			error = 0;
1621 	}
1622 
1623 	/*
1624 	 * If dsdvp[0] != NULL, it was set up by nfsrv_pnfsremovesetup() and
1625 	 * if the rename succeeded, the DS file for the tvp needs to be
1626 	 * removed.
1627 	 */
1628 	if (error == 0 && dsdvp[0] != NULL) {
1629 		nfsrv_pnfsremove(dsdvp, mirrorcnt, fname, &fh, p);
1630 		NFSD_DEBUG(4, "nfsvno_rename: pnfsremove\n");
1631 	}
1632 
1633 	nfsvno_relpathbuf(tondp);
1634 out1:
1635 	nfsvno_relpathbuf(fromndp);
1636 	NFSEXITCODE(error);
1637 	return (error);
1638 }
1639 
1640 /*
1641  * Link vnode op.
1642  */
1643 int
1644 nfsvno_link(struct nameidata *ndp, struct vnode *vp, struct ucred *cred,
1645     struct thread *p, struct nfsexstuff *exp)
1646 {
1647 	struct vnode *xp;
1648 	int error = 0;
1649 
1650 	xp = ndp->ni_vp;
1651 	if (xp != NULL) {
1652 		error = EEXIST;
1653 	} else {
1654 		xp = ndp->ni_dvp;
1655 		if (vp->v_mount != xp->v_mount)
1656 			error = EXDEV;
1657 	}
1658 	if (!error) {
1659 		NFSVOPLOCK(vp, LK_EXCLUSIVE | LK_RETRY);
1660 		if (!VN_IS_DOOMED(vp))
1661 			error = VOP_LINK(ndp->ni_dvp, vp, &ndp->ni_cnd);
1662 		else
1663 			error = EPERM;
1664 		if (ndp->ni_dvp == vp) {
1665 			vrele(ndp->ni_dvp);
1666 			NFSVOPUNLOCK(vp);
1667 		} else {
1668 			vref(vp);
1669 			VOP_VPUT_PAIR(ndp->ni_dvp, &vp, true);
1670 		}
1671 	} else {
1672 		if (ndp->ni_dvp == ndp->ni_vp)
1673 			vrele(ndp->ni_dvp);
1674 		else
1675 			vput(ndp->ni_dvp);
1676 		if (ndp->ni_vp)
1677 			vrele(ndp->ni_vp);
1678 	}
1679 	nfsvno_relpathbuf(ndp);
1680 	NFSEXITCODE(error);
1681 	return (error);
1682 }
1683 
1684 /*
1685  * Do the fsync() appropriate for the commit.
1686  */
1687 int
1688 nfsvno_fsync(struct vnode *vp, u_int64_t off, int cnt, struct ucred *cred,
1689     struct thread *td)
1690 {
1691 	int error = 0;
1692 
1693 	/*
1694 	 * RFC 1813 3.3.21: if count is 0, a flush from offset to the end of
1695 	 * file is done.  At this time VOP_FSYNC does not accept offset and
1696 	 * byte count parameters so call VOP_FSYNC the whole file for now.
1697 	 * The same is true for NFSv4: RFC 3530 Sec. 14.2.3.
1698 	 * File systems that do not use the buffer cache (as indicated
1699 	 * by MNTK_USES_BCACHE not being set) must use VOP_FSYNC().
1700 	 */
1701 	if (cnt == 0 || cnt > MAX_COMMIT_COUNT ||
1702 	    (vp->v_mount->mnt_kern_flag & MNTK_USES_BCACHE) == 0) {
1703 		/*
1704 		 * Give up and do the whole thing
1705 		 */
1706 		if (vp->v_object && vm_object_mightbedirty(vp->v_object)) {
1707 			VM_OBJECT_WLOCK(vp->v_object);
1708 			vm_object_page_clean(vp->v_object, 0, 0, OBJPC_SYNC);
1709 			VM_OBJECT_WUNLOCK(vp->v_object);
1710 		}
1711 		error = VOP_FSYNC(vp, MNT_WAIT, td);
1712 	} else {
1713 		/*
1714 		 * Locate and synchronously write any buffers that fall
1715 		 * into the requested range.  Note:  we are assuming that
1716 		 * f_iosize is a power of 2.
1717 		 */
1718 		int iosize = vp->v_mount->mnt_stat.f_iosize;
1719 		int iomask = iosize - 1;
1720 		struct bufobj *bo;
1721 		daddr_t lblkno;
1722 
1723 		/*
1724 		 * Align to iosize boundary, super-align to page boundary.
1725 		 */
1726 		if (off & iomask) {
1727 			cnt += off & iomask;
1728 			off &= ~(u_quad_t)iomask;
1729 		}
1730 		if (off & PAGE_MASK) {
1731 			cnt += off & PAGE_MASK;
1732 			off &= ~(u_quad_t)PAGE_MASK;
1733 		}
1734 		lblkno = off / iosize;
1735 
1736 		if (vp->v_object && vm_object_mightbedirty(vp->v_object)) {
1737 			VM_OBJECT_WLOCK(vp->v_object);
1738 			vm_object_page_clean(vp->v_object, off, off + cnt,
1739 			    OBJPC_SYNC);
1740 			VM_OBJECT_WUNLOCK(vp->v_object);
1741 		}
1742 
1743 		bo = &vp->v_bufobj;
1744 		BO_LOCK(bo);
1745 		while (cnt > 0) {
1746 			struct buf *bp;
1747 
1748 			/*
1749 			 * If we have a buffer and it is marked B_DELWRI we
1750 			 * have to lock and write it.  Otherwise the prior
1751 			 * write is assumed to have already been committed.
1752 			 *
1753 			 * gbincore() can return invalid buffers now so we
1754 			 * have to check that bit as well (though B_DELWRI
1755 			 * should not be set if B_INVAL is set there could be
1756 			 * a race here since we haven't locked the buffer).
1757 			 */
1758 			if ((bp = gbincore(&vp->v_bufobj, lblkno)) != NULL) {
1759 				if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_SLEEPFAIL |
1760 				    LK_INTERLOCK, BO_LOCKPTR(bo)) == ENOLCK) {
1761 					BO_LOCK(bo);
1762 					continue; /* retry */
1763 				}
1764 			    	if ((bp->b_flags & (B_DELWRI|B_INVAL)) ==
1765 				    B_DELWRI) {
1766 					bremfree(bp);
1767 					bp->b_flags &= ~B_ASYNC;
1768 					bwrite(bp);
1769 					++nfs_commit_miss;
1770 				} else
1771 					BUF_UNLOCK(bp);
1772 				BO_LOCK(bo);
1773 			}
1774 			++nfs_commit_blks;
1775 			if (cnt < iosize)
1776 				break;
1777 			cnt -= iosize;
1778 			++lblkno;
1779 		}
1780 		BO_UNLOCK(bo);
1781 	}
1782 	NFSEXITCODE(error);
1783 	return (error);
1784 }
1785 
1786 /*
1787  * Statfs vnode op.
1788  */
1789 int
1790 nfsvno_statfs(struct vnode *vp, struct statfs *sf)
1791 {
1792 	struct statfs *tsf;
1793 	int error;
1794 
1795 	tsf = NULL;
1796 	if (nfsrv_devidcnt > 0) {
1797 		/* For a pNFS service, get the DS numbers. */
1798 		tsf = malloc(sizeof(*tsf), M_TEMP, M_WAITOK | M_ZERO);
1799 		error = nfsrv_pnfsstatfs(tsf, vp->v_mount);
1800 		if (error != 0) {
1801 			free(tsf, M_TEMP);
1802 			tsf = NULL;
1803 		}
1804 	}
1805 	error = VFS_STATFS(vp->v_mount, sf);
1806 	if (error == 0) {
1807 		if (tsf != NULL) {
1808 			sf->f_blocks = tsf->f_blocks;
1809 			sf->f_bavail = tsf->f_bavail;
1810 			sf->f_bfree = tsf->f_bfree;
1811 			sf->f_bsize = tsf->f_bsize;
1812 		}
1813 		/*
1814 		 * Since NFS handles these values as unsigned on the
1815 		 * wire, there is no way to represent negative values,
1816 		 * so set them to 0. Without this, they will appear
1817 		 * to be very large positive values for clients like
1818 		 * Solaris10.
1819 		 */
1820 		if (sf->f_bavail < 0)
1821 			sf->f_bavail = 0;
1822 		if (sf->f_ffree < 0)
1823 			sf->f_ffree = 0;
1824 	}
1825 	free(tsf, M_TEMP);
1826 	NFSEXITCODE(error);
1827 	return (error);
1828 }
1829 
1830 /*
1831  * Do the vnode op stuff for Open. Similar to nfsvno_createsub(), but
1832  * must handle nfsrv_opencheck() calls after any other access checks.
1833  */
1834 void
1835 nfsvno_open(struct nfsrv_descript *nd, struct nameidata *ndp,
1836     nfsquad_t clientid, nfsv4stateid_t *stateidp, struct nfsstate *stp,
1837     int *exclusive_flagp, struct nfsvattr *nvap, int32_t *cverf, int create,
1838     NFSACL_T *aclp, nfsattrbit_t *attrbitp, struct ucred *cred,
1839     struct nfsexstuff *exp, struct vnode **vpp)
1840 {
1841 	struct vnode *vp = NULL;
1842 	u_quad_t tempsize;
1843 	struct nfsexstuff nes;
1844 	struct thread *p = curthread;
1845 
1846 	if (ndp->ni_vp == NULL)
1847 		nd->nd_repstat = nfsrv_opencheck(clientid,
1848 		    stateidp, stp, NULL, nd, p, nd->nd_repstat);
1849 	if (!nd->nd_repstat) {
1850 		if (ndp->ni_vp == NULL) {
1851 			nd->nd_repstat = VOP_CREATE(ndp->ni_dvp,
1852 			    &ndp->ni_vp, &ndp->ni_cnd, &nvap->na_vattr);
1853 			/* For a pNFS server, create the data file on a DS. */
1854 			if (nd->nd_repstat == 0) {
1855 				/*
1856 				 * Create a data file on a DS for a pNFS server.
1857 				 * This function just returns if not
1858 				 * running a pNFS DS or the creation fails.
1859 				 */
1860 				nfsrv_pnfscreate(ndp->ni_vp, &nvap->na_vattr,
1861 				    cred, p);
1862 			}
1863 			VOP_VPUT_PAIR(ndp->ni_dvp, nd->nd_repstat == 0 ?
1864 			    &ndp->ni_vp : NULL, false);
1865 			nfsvno_relpathbuf(ndp);
1866 			if (!nd->nd_repstat) {
1867 				if (*exclusive_flagp) {
1868 					*exclusive_flagp = 0;
1869 					NFSVNO_ATTRINIT(nvap);
1870 					nvap->na_atime.tv_sec = cverf[0];
1871 					nvap->na_atime.tv_nsec = cverf[1];
1872 					nd->nd_repstat = VOP_SETATTR(ndp->ni_vp,
1873 					    &nvap->na_vattr, cred);
1874 					if (nd->nd_repstat != 0) {
1875 						vput(ndp->ni_vp);
1876 						ndp->ni_vp = NULL;
1877 						nd->nd_repstat = NFSERR_NOTSUPP;
1878 					} else
1879 						NFSSETBIT_ATTRBIT(attrbitp,
1880 						    NFSATTRBIT_TIMEACCESS);
1881 				} else {
1882 					nfsrv_fixattr(nd, ndp->ni_vp, nvap,
1883 					    aclp, p, attrbitp, exp);
1884 				}
1885 			}
1886 			vp = ndp->ni_vp;
1887 		} else {
1888 			nfsvno_relpathbuf(ndp);
1889 			vp = ndp->ni_vp;
1890 			if (create == NFSV4OPEN_CREATE) {
1891 				if (ndp->ni_dvp == vp)
1892 					vrele(ndp->ni_dvp);
1893 				else
1894 					vput(ndp->ni_dvp);
1895 			}
1896 			if (NFSVNO_ISSETSIZE(nvap) && vp->v_type == VREG) {
1897 				if (ndp->ni_cnd.cn_flags & RDONLY)
1898 					NFSVNO_SETEXRDONLY(&nes);
1899 				else
1900 					NFSVNO_EXINIT(&nes);
1901 				nd->nd_repstat = nfsvno_accchk(vp,
1902 				    VWRITE, cred, &nes, p,
1903 				    NFSACCCHK_NOOVERRIDE,
1904 				    NFSACCCHK_VPISLOCKED, NULL);
1905 				nd->nd_repstat = nfsrv_opencheck(clientid,
1906 				    stateidp, stp, vp, nd, p, nd->nd_repstat);
1907 				if (!nd->nd_repstat) {
1908 					tempsize = nvap->na_size;
1909 					NFSVNO_ATTRINIT(nvap);
1910 					nvap->na_size = tempsize;
1911 					nd->nd_repstat = nfsvno_setattr(vp,
1912 					    nvap, cred, p, exp);
1913 				}
1914 			} else if (vp->v_type == VREG) {
1915 				nd->nd_repstat = nfsrv_opencheck(clientid,
1916 				    stateidp, stp, vp, nd, p, nd->nd_repstat);
1917 			}
1918 		}
1919 	} else {
1920 		nfsvno_relpathbuf(ndp);
1921 		if (create == NFSV4OPEN_CREATE) {
1922 			if (ndp->ni_dvp == ndp->ni_vp)
1923 				vrele(ndp->ni_dvp);
1924 			else
1925 				vput(ndp->ni_dvp);
1926 			if (ndp->ni_vp)
1927 				vput(ndp->ni_vp);
1928 		}
1929 	}
1930 	*vpp = vp;
1931 
1932 	NFSEXITCODE2(0, nd);
1933 }
1934 
1935 /*
1936  * Updates the file rev and sets the mtime and ctime
1937  * to the current clock time, returning the va_filerev and va_Xtime
1938  * values.
1939  * Return ESTALE to indicate the vnode is VIRF_DOOMED.
1940  */
1941 int
1942 nfsvno_updfilerev(struct vnode *vp, struct nfsvattr *nvap,
1943     struct nfsrv_descript *nd, struct thread *p)
1944 {
1945 	struct vattr va;
1946 
1947 	VATTR_NULL(&va);
1948 	vfs_timestamp(&va.va_mtime);
1949 	if (NFSVOPISLOCKED(vp) != LK_EXCLUSIVE) {
1950 		NFSVOPLOCK(vp, LK_UPGRADE | LK_RETRY);
1951 		if (VN_IS_DOOMED(vp))
1952 			return (ESTALE);
1953 	}
1954 	(void) VOP_SETATTR(vp, &va, nd->nd_cred);
1955 	(void) nfsvno_getattr(vp, nvap, nd, p, 1, NULL);
1956 	return (0);
1957 }
1958 
1959 /*
1960  * Glue routine to nfsv4_fillattr().
1961  */
1962 int
1963 nfsvno_fillattr(struct nfsrv_descript *nd, struct mount *mp, struct vnode *vp,
1964     struct nfsvattr *nvap, fhandle_t *fhp, int rderror, nfsattrbit_t *attrbitp,
1965     struct ucred *cred, struct thread *p, int isdgram, int reterr,
1966     int supports_nfsv4acls, int at_root, uint64_t mounted_on_fileno)
1967 {
1968 	struct statfs *sf;
1969 	int error;
1970 
1971 	sf = NULL;
1972 	if (nfsrv_devidcnt > 0 &&
1973 	    (NFSISSET_ATTRBIT(attrbitp, NFSATTRBIT_SPACEAVAIL) ||
1974 	     NFSISSET_ATTRBIT(attrbitp, NFSATTRBIT_SPACEFREE) ||
1975 	     NFSISSET_ATTRBIT(attrbitp, NFSATTRBIT_SPACETOTAL))) {
1976 		sf = malloc(sizeof(*sf), M_TEMP, M_WAITOK | M_ZERO);
1977 		error = nfsrv_pnfsstatfs(sf, mp);
1978 		if (error != 0) {
1979 			free(sf, M_TEMP);
1980 			sf = NULL;
1981 		}
1982 	}
1983 	error = nfsv4_fillattr(nd, mp, vp, NULL, &nvap->na_vattr, fhp, rderror,
1984 	    attrbitp, cred, p, isdgram, reterr, supports_nfsv4acls, at_root,
1985 	    mounted_on_fileno, sf);
1986 	free(sf, M_TEMP);
1987 	NFSEXITCODE2(0, nd);
1988 	return (error);
1989 }
1990 
1991 /* Since the Readdir vnode ops vary, put the entire functions in here. */
1992 /*
1993  * nfs readdir service
1994  * - mallocs what it thinks is enough to read
1995  *	count rounded up to a multiple of DIRBLKSIZ <= NFS_MAXREADDIR
1996  * - calls VOP_READDIR()
1997  * - loops around building the reply
1998  *	if the output generated exceeds count break out of loop
1999  *	The NFSM_CLGET macro is used here so that the reply will be packed
2000  *	tightly in mbuf clusters.
2001  * - it trims out records with d_fileno == 0
2002  *	this doesn't matter for Unix clients, but they might confuse clients
2003  *	for other os'.
2004  * - it trims out records with d_type == DT_WHT
2005  *	these cannot be seen through NFS (unless we extend the protocol)
2006  *     The alternate call nfsrvd_readdirplus() does lookups as well.
2007  * PS: The NFS protocol spec. does not clarify what the "count" byte
2008  *	argument is a count of.. just name strings and file id's or the
2009  *	entire reply rpc or ...
2010  *	I tried just file name and id sizes and it confused the Sun client,
2011  *	so I am using the full rpc size now. The "paranoia.." comment refers
2012  *	to including the status longwords that are not a part of the dir.
2013  *	"entry" structures, but are in the rpc.
2014  */
2015 int
2016 nfsrvd_readdir(struct nfsrv_descript *nd, int isdgram,
2017     struct vnode *vp, struct nfsexstuff *exp)
2018 {
2019 	struct dirent *dp;
2020 	u_int32_t *tl;
2021 	int dirlen;
2022 	char *cpos, *cend, *rbuf;
2023 	struct nfsvattr at;
2024 	int nlen, error = 0, getret = 1;
2025 	int siz, cnt, fullsiz, eofflag, ncookies;
2026 	u_int64_t off, toff, verf __unused;
2027 	uint64_t *cookies = NULL, *cookiep;
2028 	struct uio io;
2029 	struct iovec iv;
2030 	int is_ufs;
2031 	struct thread *p = curthread;
2032 
2033 	if (nd->nd_repstat) {
2034 		nfsrv_postopattr(nd, getret, &at);
2035 		goto out;
2036 	}
2037 	if (nd->nd_flag & ND_NFSV2) {
2038 		NFSM_DISSECT(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
2039 		off = fxdr_unsigned(u_quad_t, *tl++);
2040 	} else {
2041 		NFSM_DISSECT(tl, u_int32_t *, 5 * NFSX_UNSIGNED);
2042 		off = fxdr_hyper(tl);
2043 		tl += 2;
2044 		verf = fxdr_hyper(tl);
2045 		tl += 2;
2046 	}
2047 	toff = off;
2048 	cnt = fxdr_unsigned(int, *tl);
2049 	if (cnt > NFS_SRVMAXDATA(nd) || cnt < 0)
2050 		cnt = NFS_SRVMAXDATA(nd);
2051 	siz = ((cnt + DIRBLKSIZ - 1) & ~(DIRBLKSIZ - 1));
2052 	fullsiz = siz;
2053 	if (nd->nd_flag & ND_NFSV3) {
2054 		nd->nd_repstat = getret = nfsvno_getattr(vp, &at, nd, p, 1,
2055 		    NULL);
2056 #if 0
2057 		/*
2058 		 * va_filerev is not sufficient as a cookie verifier,
2059 		 * since it is not supposed to change when entries are
2060 		 * removed/added unless that offset cookies returned to
2061 		 * the client are no longer valid.
2062 		 */
2063 		if (!nd->nd_repstat && toff && verf != at.na_filerev)
2064 			nd->nd_repstat = NFSERR_BAD_COOKIE;
2065 #endif
2066 	}
2067 	if (!nd->nd_repstat && vp->v_type != VDIR)
2068 		nd->nd_repstat = NFSERR_NOTDIR;
2069 	if (nd->nd_repstat == 0 && cnt == 0) {
2070 		if (nd->nd_flag & ND_NFSV2)
2071 			/* NFSv2 does not have NFSERR_TOOSMALL */
2072 			nd->nd_repstat = EPERM;
2073 		else
2074 			nd->nd_repstat = NFSERR_TOOSMALL;
2075 	}
2076 	if (!nd->nd_repstat)
2077 		nd->nd_repstat = nfsvno_accchk(vp, VEXEC,
2078 		    nd->nd_cred, exp, p, NFSACCCHK_NOOVERRIDE,
2079 		    NFSACCCHK_VPISLOCKED, NULL);
2080 	if (nd->nd_repstat) {
2081 		vput(vp);
2082 		if (nd->nd_flag & ND_NFSV3)
2083 			nfsrv_postopattr(nd, getret, &at);
2084 		goto out;
2085 	}
2086 	is_ufs = strcmp(vp->v_mount->mnt_vfc->vfc_name, "ufs") == 0;
2087 	rbuf = malloc(siz, M_TEMP, M_WAITOK);
2088 again:
2089 	eofflag = 0;
2090 	if (cookies) {
2091 		free(cookies, M_TEMP);
2092 		cookies = NULL;
2093 	}
2094 
2095 	iv.iov_base = rbuf;
2096 	iv.iov_len = siz;
2097 	io.uio_iov = &iv;
2098 	io.uio_iovcnt = 1;
2099 	io.uio_offset = (off_t)off;
2100 	io.uio_resid = siz;
2101 	io.uio_segflg = UIO_SYSSPACE;
2102 	io.uio_rw = UIO_READ;
2103 	io.uio_td = NULL;
2104 	nd->nd_repstat = VOP_READDIR(vp, &io, nd->nd_cred, &eofflag, &ncookies,
2105 	    &cookies);
2106 	off = (u_int64_t)io.uio_offset;
2107 	if (io.uio_resid)
2108 		siz -= io.uio_resid;
2109 
2110 	if (!cookies && !nd->nd_repstat)
2111 		nd->nd_repstat = NFSERR_PERM;
2112 	if (nd->nd_flag & ND_NFSV3) {
2113 		getret = nfsvno_getattr(vp, &at, nd, p, 1, NULL);
2114 		if (!nd->nd_repstat)
2115 			nd->nd_repstat = getret;
2116 	}
2117 
2118 	/*
2119 	 * Handles the failed cases. nd->nd_repstat == 0 past here.
2120 	 */
2121 	if (nd->nd_repstat) {
2122 		vput(vp);
2123 		free(rbuf, M_TEMP);
2124 		if (cookies)
2125 			free(cookies, M_TEMP);
2126 		if (nd->nd_flag & ND_NFSV3)
2127 			nfsrv_postopattr(nd, getret, &at);
2128 		goto out;
2129 	}
2130 	/*
2131 	 * If nothing read, return eof
2132 	 * rpc reply
2133 	 */
2134 	if (siz == 0) {
2135 		vput(vp);
2136 		if (nd->nd_flag & ND_NFSV2) {
2137 			NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
2138 		} else {
2139 			nfsrv_postopattr(nd, getret, &at);
2140 			NFSM_BUILD(tl, u_int32_t *, 4 * NFSX_UNSIGNED);
2141 			txdr_hyper(at.na_filerev, tl);
2142 			tl += 2;
2143 		}
2144 		*tl++ = newnfs_false;
2145 		*tl = newnfs_true;
2146 		free(rbuf, M_TEMP);
2147 		free(cookies, M_TEMP);
2148 		goto out;
2149 	}
2150 
2151 	/*
2152 	 * Check for degenerate cases of nothing useful read.
2153 	 * If so go try again
2154 	 */
2155 	cpos = rbuf;
2156 	cend = rbuf + siz;
2157 	dp = (struct dirent *)cpos;
2158 	cookiep = cookies;
2159 
2160 	/*
2161 	 * For some reason FreeBSD's ufs_readdir() chooses to back the
2162 	 * directory offset up to a block boundary, so it is necessary to
2163 	 * skip over the records that precede the requested offset. This
2164 	 * requires the assumption that file offset cookies monotonically
2165 	 * increase.
2166 	 */
2167 	while (cpos < cend && ncookies > 0 &&
2168 	    (dp->d_fileno == 0 || dp->d_type == DT_WHT ||
2169 	     (is_ufs == 1 && ((u_quad_t)(*cookiep)) <= toff))) {
2170 		cpos += dp->d_reclen;
2171 		dp = (struct dirent *)cpos;
2172 		cookiep++;
2173 		ncookies--;
2174 	}
2175 	if (cpos >= cend || ncookies == 0) {
2176 		siz = fullsiz;
2177 		toff = off;
2178 		goto again;
2179 	}
2180 	vput(vp);
2181 
2182 	/*
2183 	 * If cnt > MCLBYTES and the reply will not be saved, use
2184 	 * ext_pgs mbufs for TLS.
2185 	 * For NFSv4.0, we do not know for sure if the reply will
2186 	 * be saved, so do not use ext_pgs mbufs for NFSv4.0.
2187 	 */
2188 	if (cnt > MCLBYTES && siz > MCLBYTES &&
2189 	    (nd->nd_flag & (ND_TLS | ND_EXTPG | ND_SAVEREPLY)) == ND_TLS &&
2190 	    (nd->nd_flag & (ND_NFSV4 | ND_NFSV41)) != ND_NFSV4)
2191 		nd->nd_flag |= ND_EXTPG;
2192 
2193 	/*
2194 	 * dirlen is the size of the reply, including all XDR and must
2195 	 * not exceed cnt. For NFSv2, RFC1094 didn't clearly indicate
2196 	 * if the XDR should be included in "count", but to be safe, we do.
2197 	 * (Include the two booleans at the end of the reply in dirlen now.)
2198 	 */
2199 	if (nd->nd_flag & ND_NFSV3) {
2200 		nfsrv_postopattr(nd, getret, &at);
2201 		NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
2202 		txdr_hyper(at.na_filerev, tl);
2203 		dirlen = NFSX_V3POSTOPATTR + NFSX_VERF + 2 * NFSX_UNSIGNED;
2204 	} else {
2205 		dirlen = 2 * NFSX_UNSIGNED;
2206 	}
2207 
2208 	/* Loop through the records and build reply */
2209 	while (cpos < cend && ncookies > 0) {
2210 		nlen = dp->d_namlen;
2211 		if (dp->d_fileno != 0 && dp->d_type != DT_WHT &&
2212 			nlen <= NFS_MAXNAMLEN) {
2213 			if (nd->nd_flag & ND_NFSV3)
2214 				dirlen += (6*NFSX_UNSIGNED + NFSM_RNDUP(nlen));
2215 			else
2216 				dirlen += (4*NFSX_UNSIGNED + NFSM_RNDUP(nlen));
2217 			if (dirlen > cnt) {
2218 				eofflag = 0;
2219 				break;
2220 			}
2221 
2222 			/*
2223 			 * Build the directory record xdr from
2224 			 * the dirent entry.
2225 			 */
2226 			if (nd->nd_flag & ND_NFSV3) {
2227 				NFSM_BUILD(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
2228 				*tl++ = newnfs_true;
2229 				*tl++ = 0;
2230 			} else {
2231 				NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
2232 				*tl++ = newnfs_true;
2233 			}
2234 			*tl = txdr_unsigned(dp->d_fileno);
2235 			(void) nfsm_strtom(nd, dp->d_name, nlen);
2236 			if (nd->nd_flag & ND_NFSV3) {
2237 				NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
2238 				txdr_hyper(*cookiep, tl);
2239 			} else {
2240 				NFSM_BUILD(tl, u_int32_t *, NFSX_UNSIGNED);
2241 				*tl = txdr_unsigned(*cookiep);
2242 			}
2243 		}
2244 		cpos += dp->d_reclen;
2245 		dp = (struct dirent *)cpos;
2246 		cookiep++;
2247 		ncookies--;
2248 	}
2249 	if (cpos < cend)
2250 		eofflag = 0;
2251 	NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
2252 	*tl++ = newnfs_false;
2253 	if (eofflag)
2254 		*tl = newnfs_true;
2255 	else
2256 		*tl = newnfs_false;
2257 	free(rbuf, M_TEMP);
2258 	free(cookies, M_TEMP);
2259 
2260 out:
2261 	NFSEXITCODE2(0, nd);
2262 	return (0);
2263 nfsmout:
2264 	vput(vp);
2265 	NFSEXITCODE2(error, nd);
2266 	return (error);
2267 }
2268 
2269 /*
2270  * Readdirplus for V3 and Readdir for V4.
2271  */
2272 int
2273 nfsrvd_readdirplus(struct nfsrv_descript *nd, int isdgram,
2274     struct vnode *vp, struct nfsexstuff *exp)
2275 {
2276 	struct dirent *dp;
2277 	u_int32_t *tl;
2278 	int dirlen;
2279 	char *cpos, *cend, *rbuf;
2280 	struct vnode *nvp;
2281 	fhandle_t nfh;
2282 	struct nfsvattr nva, at, *nvap = &nva;
2283 	struct mbuf *mb0, *mb1;
2284 	struct nfsreferral *refp;
2285 	int nlen, r, error = 0, getret = 1, usevget = 1;
2286 	int siz, cnt, fullsiz, eofflag, ncookies, entrycnt;
2287 	caddr_t bpos0, bpos1;
2288 	u_int64_t off, toff, verf __unused;
2289 	uint64_t *cookies = NULL, *cookiep;
2290 	nfsattrbit_t attrbits, rderrbits, savbits;
2291 	struct uio io;
2292 	struct iovec iv;
2293 	struct componentname cn;
2294 	int at_root, is_ufs, is_zfs, needs_unbusy, supports_nfsv4acls;
2295 	struct mount *mp, *new_mp;
2296 	uint64_t mounted_on_fileno;
2297 	struct thread *p = curthread;
2298 	int bextpg0, bextpg1, bextpgsiz0, bextpgsiz1;
2299 
2300 	if (nd->nd_repstat) {
2301 		nfsrv_postopattr(nd, getret, &at);
2302 		goto out;
2303 	}
2304 	NFSM_DISSECT(tl, u_int32_t *, 6 * NFSX_UNSIGNED);
2305 	off = fxdr_hyper(tl);
2306 	toff = off;
2307 	tl += 2;
2308 	verf = fxdr_hyper(tl);
2309 	tl += 2;
2310 	siz = fxdr_unsigned(int, *tl++);
2311 	cnt = fxdr_unsigned(int, *tl);
2312 
2313 	/*
2314 	 * Use the server's maximum data transfer size as the upper bound
2315 	 * on reply datalen.
2316 	 */
2317 	if (cnt > NFS_SRVMAXDATA(nd) || cnt < 0)
2318 		cnt = NFS_SRVMAXDATA(nd);
2319 
2320 	/*
2321 	 * siz is a "hint" of how much directory information (name, fileid,
2322 	 * cookie) should be in the reply. At least one client "hints" 0,
2323 	 * so I set it to cnt for that case. I also round it up to the
2324 	 * next multiple of DIRBLKSIZ.
2325 	 * Since the size of a Readdirplus directory entry reply will always
2326 	 * be greater than a directory entry returned by VOP_READDIR(), it
2327 	 * does not make sense to read more than NFS_SRVMAXDATA() via
2328 	 * VOP_READDIR().
2329 	 */
2330 	if (siz <= 0)
2331 		siz = cnt;
2332 	else if (siz > NFS_SRVMAXDATA(nd))
2333 		siz = NFS_SRVMAXDATA(nd);
2334 	siz = ((siz + DIRBLKSIZ - 1) & ~(DIRBLKSIZ - 1));
2335 
2336 	if (nd->nd_flag & ND_NFSV4) {
2337 		error = nfsrv_getattrbits(nd, &attrbits, NULL, NULL);
2338 		if (error)
2339 			goto nfsmout;
2340 		NFSSET_ATTRBIT(&savbits, &attrbits);
2341 		NFSCLRNOTFILLABLE_ATTRBIT(&attrbits, nd);
2342 		NFSZERO_ATTRBIT(&rderrbits);
2343 		NFSSETBIT_ATTRBIT(&rderrbits, NFSATTRBIT_RDATTRERROR);
2344 	} else {
2345 		NFSZERO_ATTRBIT(&attrbits);
2346 	}
2347 	fullsiz = siz;
2348 	nd->nd_repstat = getret = nfsvno_getattr(vp, &at, nd, p, 1, NULL);
2349 #if 0
2350 	if (!nd->nd_repstat) {
2351 	    if (off && verf != at.na_filerev) {
2352 		/*
2353 		 * va_filerev is not sufficient as a cookie verifier,
2354 		 * since it is not supposed to change when entries are
2355 		 * removed/added unless that offset cookies returned to
2356 		 * the client are no longer valid.
2357 		 */
2358 		if (nd->nd_flag & ND_NFSV4) {
2359 			nd->nd_repstat = NFSERR_NOTSAME;
2360 		} else {
2361 			nd->nd_repstat = NFSERR_BAD_COOKIE;
2362 		}
2363 	    }
2364 	}
2365 #endif
2366 	if (!nd->nd_repstat && vp->v_type != VDIR)
2367 		nd->nd_repstat = NFSERR_NOTDIR;
2368 	if (!nd->nd_repstat && cnt == 0)
2369 		nd->nd_repstat = NFSERR_TOOSMALL;
2370 	if (!nd->nd_repstat)
2371 		nd->nd_repstat = nfsvno_accchk(vp, VEXEC,
2372 		    nd->nd_cred, exp, p, NFSACCCHK_NOOVERRIDE,
2373 		    NFSACCCHK_VPISLOCKED, NULL);
2374 	if (nd->nd_repstat) {
2375 		vput(vp);
2376 		if (nd->nd_flag & ND_NFSV3)
2377 			nfsrv_postopattr(nd, getret, &at);
2378 		goto out;
2379 	}
2380 	is_ufs = strcmp(vp->v_mount->mnt_vfc->vfc_name, "ufs") == 0;
2381 	is_zfs = strcmp(vp->v_mount->mnt_vfc->vfc_name, "zfs") == 0;
2382 
2383 	rbuf = malloc(siz, M_TEMP, M_WAITOK);
2384 again:
2385 	eofflag = 0;
2386 	if (cookies) {
2387 		free(cookies, M_TEMP);
2388 		cookies = NULL;
2389 	}
2390 
2391 	iv.iov_base = rbuf;
2392 	iv.iov_len = siz;
2393 	io.uio_iov = &iv;
2394 	io.uio_iovcnt = 1;
2395 	io.uio_offset = (off_t)off;
2396 	io.uio_resid = siz;
2397 	io.uio_segflg = UIO_SYSSPACE;
2398 	io.uio_rw = UIO_READ;
2399 	io.uio_td = NULL;
2400 	nd->nd_repstat = VOP_READDIR(vp, &io, nd->nd_cred, &eofflag, &ncookies,
2401 	    &cookies);
2402 	off = (u_int64_t)io.uio_offset;
2403 	if (io.uio_resid)
2404 		siz -= io.uio_resid;
2405 
2406 	getret = nfsvno_getattr(vp, &at, nd, p, 1, NULL);
2407 
2408 	if (!cookies && !nd->nd_repstat)
2409 		nd->nd_repstat = NFSERR_PERM;
2410 	if (!nd->nd_repstat)
2411 		nd->nd_repstat = getret;
2412 	if (nd->nd_repstat) {
2413 		vput(vp);
2414 		if (cookies)
2415 			free(cookies, M_TEMP);
2416 		free(rbuf, M_TEMP);
2417 		if (nd->nd_flag & ND_NFSV3)
2418 			nfsrv_postopattr(nd, getret, &at);
2419 		goto out;
2420 	}
2421 	/*
2422 	 * If nothing read, return eof
2423 	 * rpc reply
2424 	 */
2425 	if (siz == 0) {
2426 		vput(vp);
2427 		if (nd->nd_flag & ND_NFSV3)
2428 			nfsrv_postopattr(nd, getret, &at);
2429 		NFSM_BUILD(tl, u_int32_t *, 4 * NFSX_UNSIGNED);
2430 		txdr_hyper(at.na_filerev, tl);
2431 		tl += 2;
2432 		*tl++ = newnfs_false;
2433 		*tl = newnfs_true;
2434 		free(cookies, M_TEMP);
2435 		free(rbuf, M_TEMP);
2436 		goto out;
2437 	}
2438 
2439 	/*
2440 	 * Check for degenerate cases of nothing useful read.
2441 	 * If so go try again
2442 	 */
2443 	cpos = rbuf;
2444 	cend = rbuf + siz;
2445 	dp = (struct dirent *)cpos;
2446 	cookiep = cookies;
2447 
2448 	/*
2449 	 * For some reason FreeBSD's ufs_readdir() chooses to back the
2450 	 * directory offset up to a block boundary, so it is necessary to
2451 	 * skip over the records that precede the requested offset. This
2452 	 * requires the assumption that file offset cookies monotonically
2453 	 * increase.
2454 	 */
2455 	while (cpos < cend && ncookies > 0 &&
2456 	  (dp->d_fileno == 0 || dp->d_type == DT_WHT ||
2457 	   (is_ufs == 1 && ((u_quad_t)(*cookiep)) <= toff) ||
2458 	   ((nd->nd_flag & ND_NFSV4) &&
2459 	    ((dp->d_namlen == 1 && dp->d_name[0] == '.') ||
2460 	     (dp->d_namlen==2 && dp->d_name[0]=='.' && dp->d_name[1]=='.'))))) {
2461 		cpos += dp->d_reclen;
2462 		dp = (struct dirent *)cpos;
2463 		cookiep++;
2464 		ncookies--;
2465 	}
2466 	if (cpos >= cend || ncookies == 0) {
2467 		siz = fullsiz;
2468 		toff = off;
2469 		goto again;
2470 	}
2471 
2472 	/*
2473 	 * Busy the file system so that the mount point won't go away
2474 	 * and, as such, VFS_VGET() can be used safely.
2475 	 */
2476 	mp = vp->v_mount;
2477 	vfs_ref(mp);
2478 	NFSVOPUNLOCK(vp);
2479 	nd->nd_repstat = vfs_busy(mp, 0);
2480 	vfs_rel(mp);
2481 	if (nd->nd_repstat != 0) {
2482 		vrele(vp);
2483 		free(cookies, M_TEMP);
2484 		free(rbuf, M_TEMP);
2485 		if (nd->nd_flag & ND_NFSV3)
2486 			nfsrv_postopattr(nd, getret, &at);
2487 		goto out;
2488 	}
2489 
2490 	/*
2491 	 * Check to see if entries in this directory can be safely acquired
2492 	 * via VFS_VGET() or if a switch to VOP_LOOKUP() is required.
2493 	 * ZFS snapshot directories need VOP_LOOKUP(), so that any
2494 	 * automount of the snapshot directory that is required will
2495 	 * be done.
2496 	 * This needs to be done here for NFSv4, since NFSv4 never does
2497 	 * a VFS_VGET() for "." or "..".
2498 	 */
2499 	if (is_zfs == 1) {
2500 		r = VFS_VGET(mp, at.na_fileid, LK_SHARED, &nvp);
2501 		if (r == EOPNOTSUPP) {
2502 			usevget = 0;
2503 			cn.cn_nameiop = LOOKUP;
2504 			cn.cn_lkflags = LK_SHARED | LK_RETRY;
2505 			cn.cn_cred = nd->nd_cred;
2506 		} else if (r == 0)
2507 			vput(nvp);
2508 	}
2509 
2510 	/*
2511 	 * If the reply is likely to exceed MCLBYTES and the reply will
2512 	 * not be saved, use ext_pgs mbufs for TLS.
2513 	 * It is difficult to predict how large each entry will be and
2514 	 * how many entries have been read, so just assume the directory
2515 	 * entries grow by a factor of 4 when attributes are included.
2516 	 * For NFSv4.0, we do not know for sure if the reply will
2517 	 * be saved, so do not use ext_pgs mbufs for NFSv4.0.
2518 	 */
2519 	if (cnt > MCLBYTES && siz > MCLBYTES / 4 &&
2520 	    (nd->nd_flag & (ND_TLS | ND_EXTPG | ND_SAVEREPLY)) == ND_TLS &&
2521 	    (nd->nd_flag & (ND_NFSV4 | ND_NFSV41)) != ND_NFSV4)
2522 		nd->nd_flag |= ND_EXTPG;
2523 
2524 	/*
2525 	 * Save this position, in case there is an error before one entry
2526 	 * is created.
2527 	 */
2528 	mb0 = nd->nd_mb;
2529 	bpos0 = nd->nd_bpos;
2530 	bextpg0 = nd->nd_bextpg;
2531 	bextpgsiz0 = nd->nd_bextpgsiz;
2532 
2533 	/*
2534 	 * Fill in the first part of the reply.
2535 	 * dirlen is the reply length in bytes and cannot exceed cnt.
2536 	 * (Include the two booleans at the end of the reply in dirlen now,
2537 	 *  so we recognize when we have exceeded cnt.)
2538 	 */
2539 	if (nd->nd_flag & ND_NFSV3) {
2540 		dirlen = NFSX_V3POSTOPATTR + NFSX_VERF + 2 * NFSX_UNSIGNED;
2541 		nfsrv_postopattr(nd, getret, &at);
2542 	} else {
2543 		dirlen = NFSX_VERF + 2 * NFSX_UNSIGNED;
2544 	}
2545 	NFSM_BUILD(tl, u_int32_t *, NFSX_VERF);
2546 	txdr_hyper(at.na_filerev, tl);
2547 
2548 	/*
2549 	 * Save this position, in case there is an empty reply needed.
2550 	 */
2551 	mb1 = nd->nd_mb;
2552 	bpos1 = nd->nd_bpos;
2553 	bextpg1 = nd->nd_bextpg;
2554 	bextpgsiz1 = nd->nd_bextpgsiz;
2555 
2556 	/* Loop through the records and build reply */
2557 	entrycnt = 0;
2558 	while (cpos < cend && ncookies > 0 && dirlen < cnt) {
2559 		nlen = dp->d_namlen;
2560 		if (dp->d_fileno != 0 && dp->d_type != DT_WHT &&
2561 		    nlen <= NFS_MAXNAMLEN &&
2562 		    ((nd->nd_flag & ND_NFSV3) || nlen > 2 ||
2563 		     (nlen==2 && (dp->d_name[0]!='.' || dp->d_name[1]!='.'))
2564 		      || (nlen == 1 && dp->d_name[0] != '.'))) {
2565 			/*
2566 			 * Save the current position in the reply, in case
2567 			 * this entry exceeds cnt.
2568 			 */
2569 			mb1 = nd->nd_mb;
2570 			bpos1 = nd->nd_bpos;
2571 			bextpg1 = nd->nd_bextpg;
2572 			bextpgsiz1 = nd->nd_bextpgsiz;
2573 
2574 			/*
2575 			 * For readdir_and_lookup get the vnode using
2576 			 * the file number.
2577 			 */
2578 			nvp = NULL;
2579 			refp = NULL;
2580 			r = 0;
2581 			at_root = 0;
2582 			needs_unbusy = 0;
2583 			new_mp = mp;
2584 			mounted_on_fileno = (uint64_t)dp->d_fileno;
2585 			if ((nd->nd_flag & ND_NFSV3) ||
2586 			    NFSNONZERO_ATTRBIT(&savbits)) {
2587 				if (nd->nd_flag & ND_NFSV4)
2588 					refp = nfsv4root_getreferral(NULL,
2589 					    vp, dp->d_fileno);
2590 				if (refp == NULL) {
2591 					if (usevget)
2592 						r = VFS_VGET(mp, dp->d_fileno,
2593 						    LK_SHARED, &nvp);
2594 					else
2595 						r = EOPNOTSUPP;
2596 					if (r == EOPNOTSUPP) {
2597 						if (usevget) {
2598 							usevget = 0;
2599 							cn.cn_nameiop = LOOKUP;
2600 							cn.cn_lkflags =
2601 							    LK_SHARED |
2602 							    LK_RETRY;
2603 							cn.cn_cred =
2604 							    nd->nd_cred;
2605 						}
2606 						cn.cn_nameptr = dp->d_name;
2607 						cn.cn_namelen = nlen;
2608 						cn.cn_flags = ISLASTCN |
2609 						    NOFOLLOW | LOCKLEAF;
2610 						if (nlen == 2 &&
2611 						    dp->d_name[0] == '.' &&
2612 						    dp->d_name[1] == '.')
2613 							cn.cn_flags |=
2614 							    ISDOTDOT;
2615 						if (NFSVOPLOCK(vp, LK_SHARED)
2616 						    != 0) {
2617 							nd->nd_repstat = EPERM;
2618 							break;
2619 						}
2620 						if ((vp->v_vflag & VV_ROOT) != 0
2621 						    && (cn.cn_flags & ISDOTDOT)
2622 						    != 0) {
2623 							vref(vp);
2624 							nvp = vp;
2625 							r = 0;
2626 						} else {
2627 							r = VOP_LOOKUP(vp, &nvp,
2628 							    &cn);
2629 							if (vp != nvp)
2630 								NFSVOPUNLOCK(vp);
2631 						}
2632 					}
2633 
2634 					/*
2635 					 * For NFSv4, check to see if nvp is
2636 					 * a mount point and get the mount
2637 					 * point vnode, as required.
2638 					 */
2639 					if (r == 0 &&
2640 					    nfsrv_enable_crossmntpt != 0 &&
2641 					    (nd->nd_flag & ND_NFSV4) != 0 &&
2642 					    nvp->v_type == VDIR &&
2643 					    nvp->v_mountedhere != NULL) {
2644 						new_mp = nvp->v_mountedhere;
2645 						r = vfs_busy(new_mp, 0);
2646 						vput(nvp);
2647 						nvp = NULL;
2648 						if (r == 0) {
2649 							r = VFS_ROOT(new_mp,
2650 							    LK_SHARED, &nvp);
2651 							needs_unbusy = 1;
2652 							if (r == 0)
2653 								at_root = 1;
2654 						}
2655 					}
2656 				}
2657 
2658 				/*
2659 				 * If we failed to look up the entry, then it
2660 				 * has become invalid, most likely removed.
2661 				 */
2662 				if (r != 0) {
2663 					if (needs_unbusy)
2664 						vfs_unbusy(new_mp);
2665 					goto invalid;
2666 				}
2667 				KASSERT(refp != NULL || nvp != NULL,
2668 				    ("%s: undetected lookup error", __func__));
2669 
2670 				if (refp == NULL &&
2671 				    ((nd->nd_flag & ND_NFSV3) ||
2672 				     NFSNONZERO_ATTRBIT(&attrbits))) {
2673 					r = nfsvno_getfh(nvp, &nfh, p);
2674 					if (!r)
2675 					    r = nfsvno_getattr(nvp, nvap, nd, p,
2676 						1, &attrbits);
2677 					if (r == 0 && is_zfs == 1 &&
2678 					    nfsrv_enable_crossmntpt != 0 &&
2679 					    (nd->nd_flag & ND_NFSV4) != 0 &&
2680 					    nvp->v_type == VDIR &&
2681 					    vp->v_mount != nvp->v_mount) {
2682 					    /*
2683 					     * For a ZFS snapshot, there is a
2684 					     * pseudo mount that does not set
2685 					     * v_mountedhere, so it needs to
2686 					     * be detected via a different
2687 					     * mount structure.
2688 					     */
2689 					    at_root = 1;
2690 					    if (new_mp == mp)
2691 						new_mp = nvp->v_mount;
2692 					}
2693 				}
2694 
2695 				/*
2696 				 * If we failed to get attributes of the entry,
2697 				 * then just skip it for NFSv3 (the traditional
2698 				 * behavior in the old NFS server).
2699 				 * For NFSv4 the behavior is controlled by
2700 				 * RDATTRERROR: we either ignore the error or
2701 				 * fail the request.
2702 				 * The exception is EOPNOTSUPP, which can be
2703 				 * returned by nfsvno_getfh() for certain
2704 				 * file systems, such as devfs.  This indicates
2705 				 * that the file system cannot be exported,
2706 				 * so just skip over the entry.
2707 				 * Note that RDATTRERROR is never set for NFSv3.
2708 				 */
2709 				if (r != 0) {
2710 					if (!NFSISSET_ATTRBIT(&attrbits,
2711 					    NFSATTRBIT_RDATTRERROR) ||
2712 					    r == EOPNOTSUPP) {
2713 						vput(nvp);
2714 						if (needs_unbusy != 0)
2715 							vfs_unbusy(new_mp);
2716 						if ((nd->nd_flag & ND_NFSV3) ||
2717 						    r == EOPNOTSUPP)
2718 							goto invalid;
2719 						nd->nd_repstat = r;
2720 						break;
2721 					}
2722 				}
2723 			}
2724 
2725 			/*
2726 			 * Build the directory record xdr
2727 			 */
2728 			if (nd->nd_flag & ND_NFSV3) {
2729 				NFSM_BUILD(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
2730 				*tl++ = newnfs_true;
2731 				*tl++ = 0;
2732 				*tl = txdr_unsigned(dp->d_fileno);
2733 				dirlen += nfsm_strtom(nd, dp->d_name, nlen);
2734 				NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
2735 				txdr_hyper(*cookiep, tl);
2736 				nfsrv_postopattr(nd, 0, nvap);
2737 				dirlen += nfsm_fhtom(nd,(u_int8_t *)&nfh,0,1);
2738 				dirlen += (5*NFSX_UNSIGNED+NFSX_V3POSTOPATTR);
2739 				if (nvp != NULL)
2740 					vput(nvp);
2741 			} else {
2742 				NFSM_BUILD(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
2743 				*tl++ = newnfs_true;
2744 				txdr_hyper(*cookiep, tl);
2745 				dirlen += nfsm_strtom(nd, dp->d_name, nlen);
2746 				if (nvp != NULL) {
2747 					supports_nfsv4acls =
2748 					    nfs_supportsnfsv4acls(nvp);
2749 					NFSVOPUNLOCK(nvp);
2750 				} else
2751 					supports_nfsv4acls = 0;
2752 				if (refp != NULL) {
2753 					dirlen += nfsrv_putreferralattr(nd,
2754 					    &savbits, refp, 0,
2755 					    &nd->nd_repstat);
2756 					if (nd->nd_repstat) {
2757 						if (nvp != NULL)
2758 							vrele(nvp);
2759 						if (needs_unbusy != 0)
2760 							vfs_unbusy(new_mp);
2761 						break;
2762 					}
2763 				} else if (r) {
2764 					dirlen += nfsvno_fillattr(nd, new_mp,
2765 					    nvp, nvap, &nfh, r, &rderrbits,
2766 					    nd->nd_cred, p, isdgram, 0,
2767 					    supports_nfsv4acls, at_root,
2768 					    mounted_on_fileno);
2769 				} else {
2770 					dirlen += nfsvno_fillattr(nd, new_mp,
2771 					    nvp, nvap, &nfh, r, &attrbits,
2772 					    nd->nd_cred, p, isdgram, 0,
2773 					    supports_nfsv4acls, at_root,
2774 					    mounted_on_fileno);
2775 				}
2776 				if (nvp != NULL)
2777 					vrele(nvp);
2778 				dirlen += (3 * NFSX_UNSIGNED);
2779 			}
2780 			if (needs_unbusy != 0)
2781 				vfs_unbusy(new_mp);
2782 			if (dirlen <= cnt)
2783 				entrycnt++;
2784 		}
2785 invalid:
2786 		cpos += dp->d_reclen;
2787 		dp = (struct dirent *)cpos;
2788 		cookiep++;
2789 		ncookies--;
2790 	}
2791 	vrele(vp);
2792 	vfs_unbusy(mp);
2793 
2794 	/*
2795 	 * If dirlen > cnt, we must strip off the last entry. If that
2796 	 * results in an empty reply, report NFSERR_TOOSMALL.
2797 	 */
2798 	if (dirlen > cnt || nd->nd_repstat) {
2799 		if (!nd->nd_repstat && entrycnt == 0)
2800 			nd->nd_repstat = NFSERR_TOOSMALL;
2801 		if (nd->nd_repstat) {
2802 			nfsm_trimtrailing(nd, mb0, bpos0, bextpg0, bextpgsiz0);
2803 			if (nd->nd_flag & ND_NFSV3)
2804 				nfsrv_postopattr(nd, getret, &at);
2805 		} else
2806 			nfsm_trimtrailing(nd, mb1, bpos1, bextpg1, bextpgsiz1);
2807 		eofflag = 0;
2808 	} else if (cpos < cend)
2809 		eofflag = 0;
2810 	if (!nd->nd_repstat) {
2811 		NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
2812 		*tl++ = newnfs_false;
2813 		if (eofflag)
2814 			*tl = newnfs_true;
2815 		else
2816 			*tl = newnfs_false;
2817 	}
2818 	free(cookies, M_TEMP);
2819 	free(rbuf, M_TEMP);
2820 
2821 out:
2822 	NFSEXITCODE2(0, nd);
2823 	return (0);
2824 nfsmout:
2825 	vput(vp);
2826 	NFSEXITCODE2(error, nd);
2827 	return (error);
2828 }
2829 
2830 /*
2831  * Get the settable attributes out of the mbuf list.
2832  * (Return 0 or EBADRPC)
2833  */
2834 int
2835 nfsrv_sattr(struct nfsrv_descript *nd, vnode_t vp, struct nfsvattr *nvap,
2836     nfsattrbit_t *attrbitp, NFSACL_T *aclp, struct thread *p)
2837 {
2838 	u_int32_t *tl;
2839 	struct nfsv2_sattr *sp;
2840 	int error = 0, toclient = 0;
2841 
2842 	switch (nd->nd_flag & (ND_NFSV2 | ND_NFSV3 | ND_NFSV4)) {
2843 	case ND_NFSV2:
2844 		NFSM_DISSECT(sp, struct nfsv2_sattr *, NFSX_V2SATTR);
2845 		/*
2846 		 * Some old clients didn't fill in the high order 16bits.
2847 		 * --> check the low order 2 bytes for 0xffff
2848 		 */
2849 		if ((fxdr_unsigned(int, sp->sa_mode) & 0xffff) != 0xffff)
2850 			nvap->na_mode = nfstov_mode(sp->sa_mode);
2851 		if (sp->sa_uid != newnfs_xdrneg1)
2852 			nvap->na_uid = fxdr_unsigned(uid_t, sp->sa_uid);
2853 		if (sp->sa_gid != newnfs_xdrneg1)
2854 			nvap->na_gid = fxdr_unsigned(gid_t, sp->sa_gid);
2855 		if (sp->sa_size != newnfs_xdrneg1)
2856 			nvap->na_size = fxdr_unsigned(u_quad_t, sp->sa_size);
2857 		if (sp->sa_atime.nfsv2_sec != newnfs_xdrneg1) {
2858 #ifdef notyet
2859 			fxdr_nfsv2time(&sp->sa_atime, &nvap->na_atime);
2860 #else
2861 			nvap->na_atime.tv_sec =
2862 				fxdr_unsigned(u_int32_t,sp->sa_atime.nfsv2_sec);
2863 			nvap->na_atime.tv_nsec = 0;
2864 #endif
2865 		}
2866 		if (sp->sa_mtime.nfsv2_sec != newnfs_xdrneg1)
2867 			fxdr_nfsv2time(&sp->sa_mtime, &nvap->na_mtime);
2868 		break;
2869 	case ND_NFSV3:
2870 		NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2871 		if (*tl == newnfs_true) {
2872 			NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2873 			nvap->na_mode = nfstov_mode(*tl);
2874 		}
2875 		NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2876 		if (*tl == newnfs_true) {
2877 			NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2878 			nvap->na_uid = fxdr_unsigned(uid_t, *tl);
2879 		}
2880 		NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2881 		if (*tl == newnfs_true) {
2882 			NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2883 			nvap->na_gid = fxdr_unsigned(gid_t, *tl);
2884 		}
2885 		NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2886 		if (*tl == newnfs_true) {
2887 			NFSM_DISSECT(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
2888 			nvap->na_size = fxdr_hyper(tl);
2889 		}
2890 		NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2891 		switch (fxdr_unsigned(int, *tl)) {
2892 		case NFSV3SATTRTIME_TOCLIENT:
2893 			NFSM_DISSECT(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
2894 			fxdr_nfsv3time(tl, &nvap->na_atime);
2895 			toclient = 1;
2896 			break;
2897 		case NFSV3SATTRTIME_TOSERVER:
2898 			vfs_timestamp(&nvap->na_atime);
2899 			nvap->na_vaflags |= VA_UTIMES_NULL;
2900 			break;
2901 		}
2902 		NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2903 		switch (fxdr_unsigned(int, *tl)) {
2904 		case NFSV3SATTRTIME_TOCLIENT:
2905 			NFSM_DISSECT(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
2906 			fxdr_nfsv3time(tl, &nvap->na_mtime);
2907 			nvap->na_vaflags &= ~VA_UTIMES_NULL;
2908 			break;
2909 		case NFSV3SATTRTIME_TOSERVER:
2910 			vfs_timestamp(&nvap->na_mtime);
2911 			if (!toclient)
2912 				nvap->na_vaflags |= VA_UTIMES_NULL;
2913 			break;
2914 		}
2915 		break;
2916 	case ND_NFSV4:
2917 		error = nfsv4_sattr(nd, vp, nvap, attrbitp, aclp, p);
2918 	}
2919 nfsmout:
2920 	NFSEXITCODE2(error, nd);
2921 	return (error);
2922 }
2923 
2924 /*
2925  * Handle the setable attributes for V4.
2926  * Returns NFSERR_BADXDR if it can't be parsed, 0 otherwise.
2927  */
2928 int
2929 nfsv4_sattr(struct nfsrv_descript *nd, vnode_t vp, struct nfsvattr *nvap,
2930     nfsattrbit_t *attrbitp, NFSACL_T *aclp, struct thread *p)
2931 {
2932 	u_int32_t *tl;
2933 	int attrsum = 0;
2934 	int i, j;
2935 	int error, attrsize, bitpos, aclsize, aceerr, retnotsup = 0;
2936 	int moderet, toclient = 0;
2937 	u_char *cp, namestr[NFSV4_SMALLSTR + 1];
2938 	uid_t uid;
2939 	gid_t gid;
2940 	u_short mode, mask;		/* Same type as va_mode. */
2941 	struct vattr va;
2942 
2943 	error = nfsrv_getattrbits(nd, attrbitp, NULL, &retnotsup);
2944 	if (error)
2945 		goto nfsmout;
2946 	NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2947 	attrsize = fxdr_unsigned(int, *tl);
2948 
2949 	/*
2950 	 * Loop around getting the setable attributes. If an unsupported
2951 	 * one is found, set nd_repstat == NFSERR_ATTRNOTSUPP and return.
2952 	 */
2953 	if (retnotsup) {
2954 		nd->nd_repstat = NFSERR_ATTRNOTSUPP;
2955 		bitpos = NFSATTRBIT_MAX;
2956 	} else {
2957 		bitpos = 0;
2958 	}
2959 	moderet = 0;
2960 	for (; bitpos < NFSATTRBIT_MAX; bitpos++) {
2961 	    if (attrsum > attrsize) {
2962 		error = NFSERR_BADXDR;
2963 		goto nfsmout;
2964 	    }
2965 	    if (NFSISSET_ATTRBIT(attrbitp, bitpos))
2966 		switch (bitpos) {
2967 		case NFSATTRBIT_SIZE:
2968 			NFSM_DISSECT(tl, u_int32_t *, NFSX_HYPER);
2969                      if (vp != NULL && vp->v_type != VREG) {
2970                             error = (vp->v_type == VDIR) ? NFSERR_ISDIR :
2971                                 NFSERR_INVAL;
2972                             goto nfsmout;
2973 			}
2974 			nvap->na_size = fxdr_hyper(tl);
2975 			attrsum += NFSX_HYPER;
2976 			break;
2977 		case NFSATTRBIT_ACL:
2978 			error = nfsrv_dissectacl(nd, aclp, true, &aceerr,
2979 			    &aclsize, p);
2980 			if (error)
2981 				goto nfsmout;
2982 			if (aceerr && !nd->nd_repstat)
2983 				nd->nd_repstat = aceerr;
2984 			attrsum += aclsize;
2985 			break;
2986 		case NFSATTRBIT_ARCHIVE:
2987 			NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2988 			if (!nd->nd_repstat)
2989 				nd->nd_repstat = NFSERR_ATTRNOTSUPP;
2990 			attrsum += NFSX_UNSIGNED;
2991 			break;
2992 		case NFSATTRBIT_HIDDEN:
2993 			NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2994 			if (!nd->nd_repstat)
2995 				nd->nd_repstat = NFSERR_ATTRNOTSUPP;
2996 			attrsum += NFSX_UNSIGNED;
2997 			break;
2998 		case NFSATTRBIT_MIMETYPE:
2999 			NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
3000 			i = fxdr_unsigned(int, *tl);
3001 			error = nfsm_advance(nd, NFSM_RNDUP(i), -1);
3002 			if (error)
3003 				goto nfsmout;
3004 			if (!nd->nd_repstat)
3005 				nd->nd_repstat = NFSERR_ATTRNOTSUPP;
3006 			attrsum += (NFSX_UNSIGNED + NFSM_RNDUP(i));
3007 			break;
3008 		case NFSATTRBIT_MODE:
3009 			moderet = NFSERR_INVAL;	/* Can't do MODESETMASKED. */
3010 			NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
3011 			nvap->na_mode = nfstov_mode(*tl);
3012 			attrsum += NFSX_UNSIGNED;
3013 			break;
3014 		case NFSATTRBIT_OWNER:
3015 			NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
3016 			j = fxdr_unsigned(int, *tl);
3017 			if (j < 0) {
3018 				error = NFSERR_BADXDR;
3019 				goto nfsmout;
3020 			}
3021 			if (j > NFSV4_SMALLSTR)
3022 				cp = malloc(j + 1, M_NFSSTRING, M_WAITOK);
3023 			else
3024 				cp = namestr;
3025 			error = nfsrv_mtostr(nd, cp, j);
3026 			if (error) {
3027 				if (j > NFSV4_SMALLSTR)
3028 					free(cp, M_NFSSTRING);
3029 				goto nfsmout;
3030 			}
3031 			if (!nd->nd_repstat) {
3032 				nd->nd_repstat = nfsv4_strtouid(nd, cp, j,
3033 				    &uid);
3034 				if (!nd->nd_repstat)
3035 					nvap->na_uid = uid;
3036 			}
3037 			if (j > NFSV4_SMALLSTR)
3038 				free(cp, M_NFSSTRING);
3039 			attrsum += (NFSX_UNSIGNED + NFSM_RNDUP(j));
3040 			break;
3041 		case NFSATTRBIT_OWNERGROUP:
3042 			NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
3043 			j = fxdr_unsigned(int, *tl);
3044 			if (j < 0) {
3045 				error = NFSERR_BADXDR;
3046 				goto nfsmout;
3047 			}
3048 			if (j > NFSV4_SMALLSTR)
3049 				cp = malloc(j + 1, M_NFSSTRING, M_WAITOK);
3050 			else
3051 				cp = namestr;
3052 			error = nfsrv_mtostr(nd, cp, j);
3053 			if (error) {
3054 				if (j > NFSV4_SMALLSTR)
3055 					free(cp, M_NFSSTRING);
3056 				goto nfsmout;
3057 			}
3058 			if (!nd->nd_repstat) {
3059 				nd->nd_repstat = nfsv4_strtogid(nd, cp, j,
3060 				    &gid);
3061 				if (!nd->nd_repstat)
3062 					nvap->na_gid = gid;
3063 			}
3064 			if (j > NFSV4_SMALLSTR)
3065 				free(cp, M_NFSSTRING);
3066 			attrsum += (NFSX_UNSIGNED + NFSM_RNDUP(j));
3067 			break;
3068 		case NFSATTRBIT_SYSTEM:
3069 			NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
3070 			if (!nd->nd_repstat)
3071 				nd->nd_repstat = NFSERR_ATTRNOTSUPP;
3072 			attrsum += NFSX_UNSIGNED;
3073 			break;
3074 		case NFSATTRBIT_TIMEACCESSSET:
3075 			NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
3076 			attrsum += NFSX_UNSIGNED;
3077 			if (fxdr_unsigned(int, *tl)==NFSV4SATTRTIME_TOCLIENT) {
3078 			    NFSM_DISSECT(tl, u_int32_t *, NFSX_V4TIME);
3079 			    fxdr_nfsv4time(tl, &nvap->na_atime);
3080 			    toclient = 1;
3081 			    attrsum += NFSX_V4TIME;
3082 			} else {
3083 			    vfs_timestamp(&nvap->na_atime);
3084 			    nvap->na_vaflags |= VA_UTIMES_NULL;
3085 			}
3086 			break;
3087 		case NFSATTRBIT_TIMEBACKUP:
3088 			NFSM_DISSECT(tl, u_int32_t *, NFSX_V4TIME);
3089 			if (!nd->nd_repstat)
3090 				nd->nd_repstat = NFSERR_ATTRNOTSUPP;
3091 			attrsum += NFSX_V4TIME;
3092 			break;
3093 		case NFSATTRBIT_TIMECREATE:
3094 			NFSM_DISSECT(tl, u_int32_t *, NFSX_V4TIME);
3095 			fxdr_nfsv4time(tl, &nvap->na_btime);
3096 			attrsum += NFSX_V4TIME;
3097 			break;
3098 		case NFSATTRBIT_TIMEMODIFYSET:
3099 			NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
3100 			attrsum += NFSX_UNSIGNED;
3101 			if (fxdr_unsigned(int, *tl)==NFSV4SATTRTIME_TOCLIENT) {
3102 			    NFSM_DISSECT(tl, u_int32_t *, NFSX_V4TIME);
3103 			    fxdr_nfsv4time(tl, &nvap->na_mtime);
3104 			    nvap->na_vaflags &= ~VA_UTIMES_NULL;
3105 			    attrsum += NFSX_V4TIME;
3106 			} else {
3107 			    vfs_timestamp(&nvap->na_mtime);
3108 			    if (!toclient)
3109 				nvap->na_vaflags |= VA_UTIMES_NULL;
3110 			}
3111 			break;
3112 		case NFSATTRBIT_MODESETMASKED:
3113 			NFSM_DISSECT(tl, uint32_t *, 2 * NFSX_UNSIGNED);
3114 			mode = fxdr_unsigned(u_short, *tl++);
3115 			mask = fxdr_unsigned(u_short, *tl);
3116 			/*
3117 			 * vp == NULL implies an Open/Create operation.
3118 			 * This attribute can only be used for Setattr and
3119 			 * only for NFSv4.1 or higher.
3120 			 * If moderet != 0, a mode attribute has also been
3121 			 * specified and this attribute cannot be done in the
3122 			 * same Setattr operation.
3123 			 */
3124 			if ((nd->nd_flag & ND_NFSV41) == 0)
3125 				nd->nd_repstat = NFSERR_ATTRNOTSUPP;
3126 			else if ((mode & ~07777) != 0 || (mask & ~07777) != 0 ||
3127 			    vp == NULL)
3128 				nd->nd_repstat = NFSERR_INVAL;
3129 			else if (moderet == 0)
3130 				moderet = VOP_GETATTR(vp, &va, nd->nd_cred);
3131 			if (moderet == 0)
3132 				nvap->na_mode = (mode & mask) |
3133 				    (va.va_mode & ~mask);
3134 			else
3135 				nd->nd_repstat = moderet;
3136 			attrsum += 2 * NFSX_UNSIGNED;
3137 			break;
3138 		default:
3139 			nd->nd_repstat = NFSERR_ATTRNOTSUPP;
3140 			/*
3141 			 * set bitpos so we drop out of the loop.
3142 			 */
3143 			bitpos = NFSATTRBIT_MAX;
3144 			break;
3145 		}
3146 	}
3147 
3148 	/*
3149 	 * some clients pad the attrlist, so we need to skip over the
3150 	 * padding.
3151 	 */
3152 	if (attrsum > attrsize) {
3153 		error = NFSERR_BADXDR;
3154 	} else {
3155 		attrsize = NFSM_RNDUP(attrsize);
3156 		if (attrsum < attrsize)
3157 			error = nfsm_advance(nd, attrsize - attrsum, -1);
3158 	}
3159 nfsmout:
3160 	NFSEXITCODE2(error, nd);
3161 	return (error);
3162 }
3163 
3164 /*
3165  * Check/setup export credentials.
3166  */
3167 int
3168 nfsd_excred(struct nfsrv_descript *nd, struct nfsexstuff *exp,
3169     struct ucred *credanon, bool testsec)
3170 {
3171 	int error;
3172 
3173 	/*
3174 	 * Check/setup credentials.
3175 	 */
3176 	if (nd->nd_flag & ND_GSS)
3177 		exp->nes_exflag &= ~MNT_EXPORTANON;
3178 
3179 	/*
3180 	 * Check to see if the operation is allowed for this security flavor.
3181 	 */
3182 	error = 0;
3183 	if (testsec) {
3184 		error = nfsvno_testexp(nd, exp);
3185 		if (error != 0)
3186 			goto out;
3187 	}
3188 
3189 	/*
3190 	 * Check to see if the file system is exported V4 only.
3191 	 */
3192 	if (NFSVNO_EXV4ONLY(exp) && !(nd->nd_flag & ND_NFSV4)) {
3193 		error = NFSERR_PROGNOTV4;
3194 		goto out;
3195 	}
3196 
3197 	/*
3198 	 * Now, map the user credentials.
3199 	 * (Note that ND_AUTHNONE will only be set for an NFSv3
3200 	 *  Fsinfo RPC. If set for anything else, this code might need
3201 	 *  to change.)
3202 	 */
3203 	if (NFSVNO_EXPORTED(exp)) {
3204 		if (((nd->nd_flag & ND_GSS) == 0 && nd->nd_cred->cr_uid == 0) ||
3205 		     NFSVNO_EXPORTANON(exp) ||
3206 		     (nd->nd_flag & ND_AUTHNONE) != 0) {
3207 			nd->nd_cred->cr_uid = credanon->cr_uid;
3208 			nd->nd_cred->cr_gid = credanon->cr_gid;
3209 			crsetgroups(nd->nd_cred, credanon->cr_ngroups,
3210 			    credanon->cr_groups);
3211 		} else if ((nd->nd_flag & ND_GSS) == 0) {
3212 			/*
3213 			 * If using AUTH_SYS, call nfsrv_getgrpscred() to see
3214 			 * if there is a replacement credential with a group
3215 			 * list set up by "nfsuserd -manage-gids".
3216 			 * If there is no replacement, nfsrv_getgrpscred()
3217 			 * simply returns its argument.
3218 			 */
3219 			nd->nd_cred = nfsrv_getgrpscred(nd->nd_cred);
3220 		}
3221 	}
3222 
3223 out:
3224 	NFSEXITCODE2(error, nd);
3225 	return (error);
3226 }
3227 
3228 /*
3229  * Check exports.
3230  */
3231 int
3232 nfsvno_checkexp(struct mount *mp, struct sockaddr *nam, struct nfsexstuff *exp,
3233     struct ucred **credp)
3234 {
3235 	int error;
3236 
3237 	error = VFS_CHECKEXP(mp, nam, &exp->nes_exflag, credp,
3238 	    &exp->nes_numsecflavor, exp->nes_secflavors);
3239 	if (error) {
3240 		if (nfs_rootfhset) {
3241 			exp->nes_exflag = 0;
3242 			exp->nes_numsecflavor = 0;
3243 			error = 0;
3244 		}
3245 	} else if (exp->nes_numsecflavor < 1 || exp->nes_numsecflavor >
3246 	    MAXSECFLAVORS) {
3247 		printf("nfsvno_checkexp: numsecflavors out of range\n");
3248 		exp->nes_numsecflavor = 0;
3249 		error = EACCES;
3250 	}
3251 	NFSEXITCODE(error);
3252 	return (error);
3253 }
3254 
3255 /*
3256  * Get a vnode for a file handle and export stuff.
3257  */
3258 int
3259 nfsvno_fhtovp(struct mount *mp, fhandle_t *fhp, struct sockaddr *nam,
3260     int lktype, struct vnode **vpp, struct nfsexstuff *exp,
3261     struct ucred **credp)
3262 {
3263 	int error;
3264 
3265 	*credp = NULL;
3266 	exp->nes_numsecflavor = 0;
3267 	error = VFS_FHTOVP(mp, &fhp->fh_fid, lktype, vpp);
3268 	if (error != 0)
3269 		/* Make sure the server replies ESTALE to the client. */
3270 		error = ESTALE;
3271 	if (nam && !error) {
3272 		error = VFS_CHECKEXP(mp, nam, &exp->nes_exflag, credp,
3273 		    &exp->nes_numsecflavor, exp->nes_secflavors);
3274 		if (error) {
3275 			if (nfs_rootfhset) {
3276 				exp->nes_exflag = 0;
3277 				exp->nes_numsecflavor = 0;
3278 				error = 0;
3279 			} else {
3280 				vput(*vpp);
3281 			}
3282 		} else if (exp->nes_numsecflavor < 1 || exp->nes_numsecflavor >
3283 		    MAXSECFLAVORS) {
3284 			printf("nfsvno_fhtovp: numsecflavors out of range\n");
3285 			exp->nes_numsecflavor = 0;
3286 			error = EACCES;
3287 			vput(*vpp);
3288 		}
3289 	}
3290 	NFSEXITCODE(error);
3291 	return (error);
3292 }
3293 
3294 /*
3295  * nfsd_fhtovp() - convert a fh to a vnode ptr
3296  * 	- look up fsid in mount list (if not found ret error)
3297  *	- get vp and export rights by calling nfsvno_fhtovp()
3298  *	- if cred->cr_uid == 0 or MNT_EXPORTANON set it to credanon
3299  *	  for AUTH_SYS
3300  *	- if mpp != NULL, return the mount point so that it can
3301  *	  be used for vn_finished_write() by the caller
3302  */
3303 void
3304 nfsd_fhtovp(struct nfsrv_descript *nd, struct nfsrvfh *nfp, int lktype,
3305     struct vnode **vpp, struct nfsexstuff *exp,
3306     struct mount **mpp, int startwrite, int nextop)
3307 {
3308 	struct mount *mp, *mpw;
3309 	struct ucred *credanon;
3310 	fhandle_t *fhp;
3311 	int error;
3312 
3313 	if (mpp != NULL)
3314 		*mpp = NULL;
3315 	*vpp = NULL;
3316 	fhp = (fhandle_t *)nfp->nfsrvfh_data;
3317 	mp = vfs_busyfs(&fhp->fh_fsid);
3318 	if (mp == NULL) {
3319 		nd->nd_repstat = ESTALE;
3320 		goto out;
3321 	}
3322 
3323 	if (startwrite) {
3324 		mpw = mp;
3325 		error = vn_start_write(NULL, &mpw, V_WAIT);
3326 		if (error != 0) {
3327 			mpw = NULL;
3328 			vfs_unbusy(mp);
3329 			nd->nd_repstat = ESTALE;
3330 			goto out;
3331 		}
3332 		if (lktype == LK_SHARED && !(MNT_SHARED_WRITES(mp)))
3333 			lktype = LK_EXCLUSIVE;
3334 	} else
3335 		mpw = NULL;
3336 
3337 	nd->nd_repstat = nfsvno_fhtovp(mp, fhp, nd->nd_nam, lktype, vpp, exp,
3338 	    &credanon);
3339 	vfs_unbusy(mp);
3340 
3341 	/*
3342 	 * For NFSv4 without a pseudo root fs, unexported file handles
3343 	 * can be returned, so that Lookup works everywhere.
3344 	 */
3345 	if (!nd->nd_repstat && exp->nes_exflag == 0 &&
3346 	    !(nd->nd_flag & ND_NFSV4)) {
3347 		vput(*vpp);
3348 		*vpp = NULL;
3349 		nd->nd_repstat = EACCES;
3350 	}
3351 
3352 	/*
3353 	 * Personally, I've never seen any point in requiring a
3354 	 * reserved port#, since only in the rare case where the
3355 	 * clients are all boxes with secure system privileges,
3356 	 * does it provide any enhanced security, but... some people
3357 	 * believe it to be useful and keep putting this code back in.
3358 	 * (There is also some "security checker" out there that
3359 	 *  complains if the nfs server doesn't enforce this.)
3360 	 * However, note the following:
3361 	 * RFC3530 (NFSv4) specifies that a reserved port# not be
3362 	 *	required.
3363 	 * RFC2623 recommends that, if a reserved port# is checked for,
3364 	 *	that there be a way to turn that off--> ifdef'd.
3365 	 */
3366 #ifdef NFS_REQRSVPORT
3367 	if (!nd->nd_repstat) {
3368 		struct sockaddr_in *saddr;
3369 		struct sockaddr_in6 *saddr6;
3370 
3371 		saddr = NFSSOCKADDR(nd->nd_nam, struct sockaddr_in *);
3372 		saddr6 = NFSSOCKADDR(nd->nd_nam, struct sockaddr_in6 *);
3373 		if (!(nd->nd_flag & ND_NFSV4) &&
3374 		    ((saddr->sin_family == AF_INET &&
3375 		      ntohs(saddr->sin_port) >= IPPORT_RESERVED) ||
3376 		     (saddr6->sin6_family == AF_INET6 &&
3377 		      ntohs(saddr6->sin6_port) >= IPPORT_RESERVED))) {
3378 			vput(*vpp);
3379 			nd->nd_repstat = (NFSERR_AUTHERR | AUTH_TOOWEAK);
3380 		}
3381 	}
3382 #endif	/* NFS_REQRSVPORT */
3383 
3384 	/*
3385 	 * Check/setup credentials.
3386 	 */
3387 	if (!nd->nd_repstat) {
3388 		nd->nd_saveduid = nd->nd_cred->cr_uid;
3389 		nd->nd_repstat = nfsd_excred(nd, exp, credanon,
3390 		    nfsrv_checkwrongsec(nd, nextop, (*vpp)->v_type));
3391 		if (nd->nd_repstat)
3392 			vput(*vpp);
3393 	}
3394 	if (credanon != NULL)
3395 		crfree(credanon);
3396 	if (nd->nd_repstat) {
3397 		vn_finished_write(mpw);
3398 		*vpp = NULL;
3399 	} else if (mpp != NULL) {
3400 		*mpp = mpw;
3401 	}
3402 
3403 out:
3404 	NFSEXITCODE2(0, nd);
3405 }
3406 
3407 /*
3408  * glue for fp.
3409  */
3410 static int
3411 fp_getfvp(struct thread *p, int fd, struct file **fpp, struct vnode **vpp)
3412 {
3413 	struct filedesc *fdp;
3414 	struct file *fp;
3415 	int error = 0;
3416 
3417 	fdp = p->td_proc->p_fd;
3418 	if (fd < 0 || fd >= fdp->fd_nfiles ||
3419 	    (fp = fdp->fd_ofiles[fd].fde_file) == NULL) {
3420 		error = EBADF;
3421 		goto out;
3422 	}
3423 	*fpp = fp;
3424 
3425 out:
3426 	NFSEXITCODE(error);
3427 	return (error);
3428 }
3429 
3430 /*
3431  * Called from nfssvc() to update the exports list. Just call
3432  * vfs_export(). This has to be done, since the v4 root fake fs isn't
3433  * in the mount list.
3434  */
3435 int
3436 nfsrv_v4rootexport(void *argp, struct ucred *cred, struct thread *p)
3437 {
3438 	struct nfsex_args *nfsexargp = (struct nfsex_args *)argp;
3439 	int error = 0;
3440 	struct nameidata nd;
3441 	fhandle_t fh;
3442 
3443 	error = vfs_export(&nfsv4root_mnt, &nfsexargp->export);
3444 	if ((nfsexargp->export.ex_flags & MNT_DELEXPORT) != 0)
3445 		nfs_rootfhset = 0;
3446 	else if (error == 0) {
3447 		if (nfsexargp->fspec == NULL) {
3448 			error = EPERM;
3449 			goto out;
3450 		}
3451 		/*
3452 		 * If fspec != NULL, this is the v4root path.
3453 		 */
3454 		NDINIT(&nd, LOOKUP, FOLLOW, UIO_USERSPACE, nfsexargp->fspec);
3455 		if ((error = namei(&nd)) != 0)
3456 			goto out;
3457 		error = nfsvno_getfh(nd.ni_vp, &fh, p);
3458 		vrele(nd.ni_vp);
3459 		if (!error) {
3460 			nfs_rootfh.nfsrvfh_len = NFSX_MYFH;
3461 			NFSBCOPY((caddr_t)&fh,
3462 			    nfs_rootfh.nfsrvfh_data,
3463 			    sizeof (fhandle_t));
3464 			nfs_rootfhset = 1;
3465 		}
3466 	}
3467 
3468 out:
3469 	NFSEXITCODE(error);
3470 	return (error);
3471 }
3472 
3473 /*
3474  * This function needs to test to see if the system is near its limit
3475  * for memory allocation via malloc() or mget() and return True iff
3476  * either of these resources are near their limit.
3477  * XXX (For now, this is just a stub.)
3478  */
3479 int nfsrv_testmalloclimit = 0;
3480 int
3481 nfsrv_mallocmget_limit(void)
3482 {
3483 	static int printmesg = 0;
3484 	static int testval = 1;
3485 
3486 	if (nfsrv_testmalloclimit && (testval++ % 1000) == 0) {
3487 		if ((printmesg++ % 100) == 0)
3488 			printf("nfsd: malloc/mget near limit\n");
3489 		return (1);
3490 	}
3491 	return (0);
3492 }
3493 
3494 /*
3495  * BSD specific initialization of a mount point.
3496  */
3497 void
3498 nfsd_mntinit(void)
3499 {
3500 	static int inited = 0;
3501 
3502 	if (inited)
3503 		return;
3504 	inited = 1;
3505 	nfsv4root_mnt.mnt_flag = (MNT_RDONLY | MNT_EXPORTED);
3506 	TAILQ_INIT(&nfsv4root_mnt.mnt_nvnodelist);
3507 	TAILQ_INIT(&nfsv4root_mnt.mnt_lazyvnodelist);
3508 	nfsv4root_mnt.mnt_export = NULL;
3509 	TAILQ_INIT(&nfsv4root_opt);
3510 	TAILQ_INIT(&nfsv4root_newopt);
3511 	nfsv4root_mnt.mnt_opt = &nfsv4root_opt;
3512 	nfsv4root_mnt.mnt_optnew = &nfsv4root_newopt;
3513 	nfsv4root_mnt.mnt_nvnodelistsize = 0;
3514 	nfsv4root_mnt.mnt_lazyvnodelistsize = 0;
3515 }
3516 
3517 static void
3518 nfsd_timer(void *arg)
3519 {
3520 
3521 	nfsrv_servertimer();
3522 	callout_reset_sbt(&nfsd_callout, SBT_1S, SBT_1S, nfsd_timer, NULL, 0);
3523 }
3524 
3525 /*
3526  * Get a vnode for a file handle, without checking exports, etc.
3527  */
3528 struct vnode *
3529 nfsvno_getvp(fhandle_t *fhp)
3530 {
3531 	struct mount *mp;
3532 	struct vnode *vp;
3533 	int error;
3534 
3535 	mp = vfs_busyfs(&fhp->fh_fsid);
3536 	if (mp == NULL)
3537 		return (NULL);
3538 	error = VFS_FHTOVP(mp, &fhp->fh_fid, LK_EXCLUSIVE, &vp);
3539 	vfs_unbusy(mp);
3540 	if (error)
3541 		return (NULL);
3542 	return (vp);
3543 }
3544 
3545 /*
3546  * Do a local VOP_ADVLOCK().
3547  */
3548 int
3549 nfsvno_advlock(struct vnode *vp, int ftype, u_int64_t first,
3550     u_int64_t end, struct thread *td)
3551 {
3552 	int error = 0;
3553 	struct flock fl;
3554 	u_int64_t tlen;
3555 
3556 	if (nfsrv_dolocallocks == 0)
3557 		goto out;
3558 	ASSERT_VOP_UNLOCKED(vp, "nfsvno_advlock: vp locked");
3559 
3560 	fl.l_whence = SEEK_SET;
3561 	fl.l_type = ftype;
3562 	fl.l_start = (off_t)first;
3563 	if (end == NFS64BITSSET) {
3564 		fl.l_len = 0;
3565 	} else {
3566 		tlen = end - first;
3567 		fl.l_len = (off_t)tlen;
3568 	}
3569 	/*
3570 	 * For FreeBSD8, the l_pid and l_sysid must be set to the same
3571 	 * values for all calls, so that all locks will be held by the
3572 	 * nfsd server. (The nfsd server handles conflicts between the
3573 	 * various clients.)
3574 	 * Since an NFSv4 lockowner is a ClientID plus an array of up to 1024
3575 	 * bytes, so it can't be put in l_sysid.
3576 	 */
3577 	if (nfsv4_sysid == 0)
3578 		nfsv4_sysid = nlm_acquire_next_sysid();
3579 	fl.l_pid = (pid_t)0;
3580 	fl.l_sysid = (int)nfsv4_sysid;
3581 
3582 	if (ftype == F_UNLCK)
3583 		error = VOP_ADVLOCK(vp, (caddr_t)td->td_proc, F_UNLCK, &fl,
3584 		    (F_POSIX | F_REMOTE));
3585 	else
3586 		error = VOP_ADVLOCK(vp, (caddr_t)td->td_proc, F_SETLK, &fl,
3587 		    (F_POSIX | F_REMOTE));
3588 
3589 out:
3590 	NFSEXITCODE(error);
3591 	return (error);
3592 }
3593 
3594 /*
3595  * Check the nfsv4 root exports.
3596  */
3597 int
3598 nfsvno_v4rootexport(struct nfsrv_descript *nd)
3599 {
3600 	struct ucred *credanon;
3601 	int error = 0, numsecflavor, secflavors[MAXSECFLAVORS], i;
3602 	uint64_t exflags;
3603 
3604 	error = vfs_stdcheckexp(&nfsv4root_mnt, nd->nd_nam, &exflags,
3605 	    &credanon, &numsecflavor, secflavors);
3606 	if (error) {
3607 		error = NFSERR_PROGUNAVAIL;
3608 		goto out;
3609 	}
3610 	if (credanon != NULL)
3611 		crfree(credanon);
3612 	for (i = 0; i < numsecflavor; i++) {
3613 		if (secflavors[i] == AUTH_SYS)
3614 			nd->nd_flag |= ND_EXAUTHSYS;
3615 		else if (secflavors[i] == RPCSEC_GSS_KRB5)
3616 			nd->nd_flag |= ND_EXGSS;
3617 		else if (secflavors[i] == RPCSEC_GSS_KRB5I)
3618 			nd->nd_flag |= ND_EXGSSINTEGRITY;
3619 		else if (secflavors[i] == RPCSEC_GSS_KRB5P)
3620 			nd->nd_flag |= ND_EXGSSPRIVACY;
3621 	}
3622 
3623 	/* And set ND_EXxx flags for TLS. */
3624 	if ((exflags & MNT_EXTLS) != 0) {
3625 		nd->nd_flag |= ND_EXTLS;
3626 		if ((exflags & MNT_EXTLSCERT) != 0)
3627 			nd->nd_flag |= ND_EXTLSCERT;
3628 		if ((exflags & MNT_EXTLSCERTUSER) != 0)
3629 			nd->nd_flag |= ND_EXTLSCERTUSER;
3630 	}
3631 
3632 out:
3633 	NFSEXITCODE(error);
3634 	return (error);
3635 }
3636 
3637 /*
3638  * Nfs server pseudo system call for the nfsd's
3639  */
3640 /*
3641  * MPSAFE
3642  */
3643 static int
3644 nfssvc_nfsd(struct thread *td, struct nfssvc_args *uap)
3645 {
3646 	struct file *fp;
3647 	struct nfsd_addsock_args sockarg;
3648 	struct nfsd_nfsd_args nfsdarg;
3649 	struct nfsd_nfsd_oargs onfsdarg;
3650 	struct nfsd_pnfsd_args pnfsdarg;
3651 	struct vnode *vp, *nvp, *curdvp;
3652 	struct pnfsdsfile *pf;
3653 	struct nfsdevice *ds, *fds;
3654 	cap_rights_t rights;
3655 	int buflen, error, ret;
3656 	char *buf, *cp, *cp2, *cp3;
3657 	char fname[PNFS_FILENAME_LEN + 1];
3658 
3659 	if (uap->flag & NFSSVC_NFSDADDSOCK) {
3660 		error = copyin(uap->argp, (caddr_t)&sockarg, sizeof (sockarg));
3661 		if (error)
3662 			goto out;
3663 		/*
3664 		 * Since we don't know what rights might be required,
3665 		 * pretend that we need them all. It is better to be too
3666 		 * careful than too reckless.
3667 		 */
3668 		error = fget(td, sockarg.sock,
3669 		    cap_rights_init_one(&rights, CAP_SOCK_SERVER), &fp);
3670 		if (error != 0)
3671 			goto out;
3672 		if (fp->f_type != DTYPE_SOCKET) {
3673 			fdrop(fp, td);
3674 			error = EPERM;
3675 			goto out;
3676 		}
3677 		error = nfsrvd_addsock(fp);
3678 		fdrop(fp, td);
3679 	} else if (uap->flag & NFSSVC_NFSDNFSD) {
3680 		if (uap->argp == NULL) {
3681 			error = EINVAL;
3682 			goto out;
3683 		}
3684 		if ((uap->flag & NFSSVC_NEWSTRUCT) == 0) {
3685 			error = copyin(uap->argp, &onfsdarg, sizeof(onfsdarg));
3686 			if (error == 0) {
3687 				nfsdarg.principal = onfsdarg.principal;
3688 				nfsdarg.minthreads = onfsdarg.minthreads;
3689 				nfsdarg.maxthreads = onfsdarg.maxthreads;
3690 				nfsdarg.version = 1;
3691 				nfsdarg.addr = NULL;
3692 				nfsdarg.addrlen = 0;
3693 				nfsdarg.dnshost = NULL;
3694 				nfsdarg.dnshostlen = 0;
3695 				nfsdarg.dspath = NULL;
3696 				nfsdarg.dspathlen = 0;
3697 				nfsdarg.mdspath = NULL;
3698 				nfsdarg.mdspathlen = 0;
3699 				nfsdarg.mirrorcnt = 1;
3700 			}
3701 		} else
3702 			error = copyin(uap->argp, &nfsdarg, sizeof(nfsdarg));
3703 		if (error)
3704 			goto out;
3705 		if (nfsdarg.addrlen > 0 && nfsdarg.addrlen < 10000 &&
3706 		    nfsdarg.dnshostlen > 0 && nfsdarg.dnshostlen < 10000 &&
3707 		    nfsdarg.dspathlen > 0 && nfsdarg.dspathlen < 10000 &&
3708 		    nfsdarg.mdspathlen > 0 && nfsdarg.mdspathlen < 10000 &&
3709 		    nfsdarg.mirrorcnt >= 1 &&
3710 		    nfsdarg.mirrorcnt <= NFSDEV_MAXMIRRORS &&
3711 		    nfsdarg.addr != NULL && nfsdarg.dnshost != NULL &&
3712 		    nfsdarg.dspath != NULL && nfsdarg.mdspath != NULL) {
3713 			NFSD_DEBUG(1, "addrlen=%d dspathlen=%d dnslen=%d"
3714 			    " mdspathlen=%d mirrorcnt=%d\n", nfsdarg.addrlen,
3715 			    nfsdarg.dspathlen, nfsdarg.dnshostlen,
3716 			    nfsdarg.mdspathlen, nfsdarg.mirrorcnt);
3717 			cp = malloc(nfsdarg.addrlen + 1, M_TEMP, M_WAITOK);
3718 			error = copyin(nfsdarg.addr, cp, nfsdarg.addrlen);
3719 			if (error != 0) {
3720 				free(cp, M_TEMP);
3721 				goto out;
3722 			}
3723 			cp[nfsdarg.addrlen] = '\0';	/* Ensure nul term. */
3724 			nfsdarg.addr = cp;
3725 			cp = malloc(nfsdarg.dnshostlen + 1, M_TEMP, M_WAITOK);
3726 			error = copyin(nfsdarg.dnshost, cp, nfsdarg.dnshostlen);
3727 			if (error != 0) {
3728 				free(nfsdarg.addr, M_TEMP);
3729 				free(cp, M_TEMP);
3730 				goto out;
3731 			}
3732 			cp[nfsdarg.dnshostlen] = '\0';	/* Ensure nul term. */
3733 			nfsdarg.dnshost = cp;
3734 			cp = malloc(nfsdarg.dspathlen + 1, M_TEMP, M_WAITOK);
3735 			error = copyin(nfsdarg.dspath, cp, nfsdarg.dspathlen);
3736 			if (error != 0) {
3737 				free(nfsdarg.addr, M_TEMP);
3738 				free(nfsdarg.dnshost, M_TEMP);
3739 				free(cp, M_TEMP);
3740 				goto out;
3741 			}
3742 			cp[nfsdarg.dspathlen] = '\0';	/* Ensure nul term. */
3743 			nfsdarg.dspath = cp;
3744 			cp = malloc(nfsdarg.mdspathlen + 1, M_TEMP, M_WAITOK);
3745 			error = copyin(nfsdarg.mdspath, cp, nfsdarg.mdspathlen);
3746 			if (error != 0) {
3747 				free(nfsdarg.addr, M_TEMP);
3748 				free(nfsdarg.dnshost, M_TEMP);
3749 				free(nfsdarg.dspath, M_TEMP);
3750 				free(cp, M_TEMP);
3751 				goto out;
3752 			}
3753 			cp[nfsdarg.mdspathlen] = '\0';	/* Ensure nul term. */
3754 			nfsdarg.mdspath = cp;
3755 		} else {
3756 			nfsdarg.addr = NULL;
3757 			nfsdarg.addrlen = 0;
3758 			nfsdarg.dnshost = NULL;
3759 			nfsdarg.dnshostlen = 0;
3760 			nfsdarg.dspath = NULL;
3761 			nfsdarg.dspathlen = 0;
3762 			nfsdarg.mdspath = NULL;
3763 			nfsdarg.mdspathlen = 0;
3764 			nfsdarg.mirrorcnt = 1;
3765 		}
3766 		nfsd_timer(NULL);
3767 		error = nfsrvd_nfsd(td, &nfsdarg);
3768 		free(nfsdarg.addr, M_TEMP);
3769 		free(nfsdarg.dnshost, M_TEMP);
3770 		free(nfsdarg.dspath, M_TEMP);
3771 		free(nfsdarg.mdspath, M_TEMP);
3772 	} else if (uap->flag & NFSSVC_PNFSDS) {
3773 		error = copyin(uap->argp, &pnfsdarg, sizeof(pnfsdarg));
3774 		if (error == 0 && (pnfsdarg.op == PNFSDOP_DELDSSERVER ||
3775 		    pnfsdarg.op == PNFSDOP_FORCEDELDS)) {
3776 			cp = malloc(PATH_MAX + 1, M_TEMP, M_WAITOK);
3777 			error = copyinstr(pnfsdarg.dspath, cp, PATH_MAX + 1,
3778 			    NULL);
3779 			if (error == 0)
3780 				error = nfsrv_deldsserver(pnfsdarg.op, cp, td);
3781 			free(cp, M_TEMP);
3782 		} else if (error == 0 && pnfsdarg.op == PNFSDOP_COPYMR) {
3783 			cp = malloc(PATH_MAX + 1, M_TEMP, M_WAITOK);
3784 			buflen = sizeof(*pf) * NFSDEV_MAXMIRRORS;
3785 			buf = malloc(buflen, M_TEMP, M_WAITOK);
3786 			error = copyinstr(pnfsdarg.mdspath, cp, PATH_MAX + 1,
3787 			    NULL);
3788 			NFSD_DEBUG(4, "pnfsdcopymr cp mdspath=%d\n", error);
3789 			if (error == 0 && pnfsdarg.dspath != NULL) {
3790 				cp2 = malloc(PATH_MAX + 1, M_TEMP, M_WAITOK);
3791 				error = copyinstr(pnfsdarg.dspath, cp2,
3792 				    PATH_MAX + 1, NULL);
3793 				NFSD_DEBUG(4, "pnfsdcopymr cp dspath=%d\n",
3794 				    error);
3795 			} else
3796 				cp2 = NULL;
3797 			if (error == 0 && pnfsdarg.curdspath != NULL) {
3798 				cp3 = malloc(PATH_MAX + 1, M_TEMP, M_WAITOK);
3799 				error = copyinstr(pnfsdarg.curdspath, cp3,
3800 				    PATH_MAX + 1, NULL);
3801 				NFSD_DEBUG(4, "pnfsdcopymr cp curdspath=%d\n",
3802 				    error);
3803 			} else
3804 				cp3 = NULL;
3805 			curdvp = NULL;
3806 			fds = NULL;
3807 			if (error == 0)
3808 				error = nfsrv_mdscopymr(cp, cp2, cp3, buf,
3809 				    &buflen, fname, td, &vp, &nvp, &pf, &ds,
3810 				    &fds);
3811 			NFSD_DEBUG(4, "nfsrv_mdscopymr=%d\n", error);
3812 			if (error == 0) {
3813 				if (pf->dsf_dir >= nfsrv_dsdirsize) {
3814 					printf("copymr: dsdir out of range\n");
3815 					pf->dsf_dir = 0;
3816 				}
3817 				NFSD_DEBUG(4, "copymr: buflen=%d\n", buflen);
3818 				error = nfsrv_copymr(vp, nvp,
3819 				    ds->nfsdev_dsdir[pf->dsf_dir], ds, pf,
3820 				    (struct pnfsdsfile *)buf,
3821 				    buflen / sizeof(*pf), td->td_ucred, td);
3822 				vput(vp);
3823 				vput(nvp);
3824 				if (fds != NULL && error == 0) {
3825 					curdvp = fds->nfsdev_dsdir[pf->dsf_dir];
3826 					ret = vn_lock(curdvp, LK_EXCLUSIVE);
3827 					if (ret == 0) {
3828 						nfsrv_dsremove(curdvp, fname,
3829 						    td->td_ucred, td);
3830 						NFSVOPUNLOCK(curdvp);
3831 					}
3832 				}
3833 				NFSD_DEBUG(4, "nfsrv_copymr=%d\n", error);
3834 			}
3835 			free(cp, M_TEMP);
3836 			free(cp2, M_TEMP);
3837 			free(cp3, M_TEMP);
3838 			free(buf, M_TEMP);
3839 		}
3840 	} else {
3841 		error = nfssvc_srvcall(td, uap, td->td_ucred);
3842 	}
3843 
3844 out:
3845 	NFSEXITCODE(error);
3846 	return (error);
3847 }
3848 
3849 static int
3850 nfssvc_srvcall(struct thread *p, struct nfssvc_args *uap, struct ucred *cred)
3851 {
3852 	struct nfsex_args export;
3853 	struct nfsex_oldargs oexp;
3854 	struct file *fp = NULL;
3855 	int stablefd, i, len;
3856 	struct nfsd_clid adminrevoke;
3857 	struct nfsd_dumplist dumplist;
3858 	struct nfsd_dumpclients *dumpclients;
3859 	struct nfsd_dumplocklist dumplocklist;
3860 	struct nfsd_dumplocks *dumplocks;
3861 	struct nameidata nd;
3862 	vnode_t vp;
3863 	int error = EINVAL, igotlock;
3864 	struct proc *procp;
3865 	gid_t *grps;
3866 	static int suspend_nfsd = 0;
3867 
3868 	if (uap->flag & NFSSVC_PUBLICFH) {
3869 		NFSBZERO((caddr_t)&nfs_pubfh.nfsrvfh_data,
3870 		    sizeof (fhandle_t));
3871 		error = copyin(uap->argp,
3872 		    &nfs_pubfh.nfsrvfh_data, sizeof (fhandle_t));
3873 		if (!error)
3874 			nfs_pubfhset = 1;
3875 	} else if ((uap->flag & (NFSSVC_V4ROOTEXPORT | NFSSVC_NEWSTRUCT)) ==
3876 	    (NFSSVC_V4ROOTEXPORT | NFSSVC_NEWSTRUCT)) {
3877 		error = copyin(uap->argp,(caddr_t)&export,
3878 		    sizeof (struct nfsex_args));
3879 		if (!error) {
3880 			grps = NULL;
3881 			if (export.export.ex_ngroups > NGROUPS_MAX ||
3882 			    export.export.ex_ngroups < 0)
3883 				error = EINVAL;
3884 			else if (export.export.ex_ngroups > 0) {
3885 				grps = malloc(export.export.ex_ngroups *
3886 				    sizeof(gid_t), M_TEMP, M_WAITOK);
3887 				error = copyin(export.export.ex_groups, grps,
3888 				    export.export.ex_ngroups * sizeof(gid_t));
3889 				export.export.ex_groups = grps;
3890 			} else
3891 				export.export.ex_groups = NULL;
3892 			if (!error)
3893 				error = nfsrv_v4rootexport(&export, cred, p);
3894 			free(grps, M_TEMP);
3895 		}
3896 	} else if ((uap->flag & (NFSSVC_V4ROOTEXPORT | NFSSVC_NEWSTRUCT)) ==
3897 	    NFSSVC_V4ROOTEXPORT) {
3898 		error = copyin(uap->argp,(caddr_t)&oexp,
3899 		    sizeof (struct nfsex_oldargs));
3900 		if (!error) {
3901 			memset(&export.export, 0, sizeof(export.export));
3902 			export.export.ex_flags = (uint64_t)oexp.export.ex_flags;
3903 			export.export.ex_root = oexp.export.ex_root;
3904 			export.export.ex_uid = oexp.export.ex_anon.cr_uid;
3905 			export.export.ex_ngroups =
3906 			    oexp.export.ex_anon.cr_ngroups;
3907 			export.export.ex_groups = NULL;
3908 			if (export.export.ex_ngroups > XU_NGROUPS ||
3909 			    export.export.ex_ngroups < 0)
3910 				error = EINVAL;
3911 			else if (export.export.ex_ngroups > 0) {
3912 				export.export.ex_groups = malloc(
3913 				    export.export.ex_ngroups * sizeof(gid_t),
3914 				    M_TEMP, M_WAITOK);
3915 				for (i = 0; i < export.export.ex_ngroups; i++)
3916 					export.export.ex_groups[i] =
3917 					    oexp.export.ex_anon.cr_groups[i];
3918 			}
3919 			export.export.ex_addr = oexp.export.ex_addr;
3920 			export.export.ex_addrlen = oexp.export.ex_addrlen;
3921 			export.export.ex_mask = oexp.export.ex_mask;
3922 			export.export.ex_masklen = oexp.export.ex_masklen;
3923 			export.export.ex_indexfile = oexp.export.ex_indexfile;
3924 			export.export.ex_numsecflavors =
3925 			    oexp.export.ex_numsecflavors;
3926 			if (export.export.ex_numsecflavors >= MAXSECFLAVORS ||
3927 			    export.export.ex_numsecflavors < 0)
3928 				error = EINVAL;
3929 			else {
3930 				for (i = 0; i < export.export.ex_numsecflavors;
3931 				    i++)
3932 					export.export.ex_secflavors[i] =
3933 					    oexp.export.ex_secflavors[i];
3934 			}
3935 			export.fspec = oexp.fspec;
3936 			if (error == 0)
3937 				error = nfsrv_v4rootexport(&export, cred, p);
3938 			free(export.export.ex_groups, M_TEMP);
3939 		}
3940 	} else if (uap->flag & NFSSVC_NOPUBLICFH) {
3941 		nfs_pubfhset = 0;
3942 		error = 0;
3943 	} else if (uap->flag & NFSSVC_STABLERESTART) {
3944 		error = copyin(uap->argp, (caddr_t)&stablefd,
3945 		    sizeof (int));
3946 		if (!error)
3947 			error = fp_getfvp(p, stablefd, &fp, &vp);
3948 		if (!error && (NFSFPFLAG(fp) & (FREAD | FWRITE)) != (FREAD | FWRITE))
3949 			error = EBADF;
3950 		if (!error && newnfs_numnfsd != 0)
3951 			error = EPERM;
3952 		if (!error) {
3953 			nfsrv_stablefirst.nsf_fp = fp;
3954 			nfsrv_setupstable(p);
3955 		}
3956 	} else if (uap->flag & NFSSVC_ADMINREVOKE) {
3957 		error = copyin(uap->argp, (caddr_t)&adminrevoke,
3958 		    sizeof (struct nfsd_clid));
3959 		if (!error)
3960 			error = nfsrv_adminrevoke(&adminrevoke, p);
3961 	} else if (uap->flag & NFSSVC_DUMPCLIENTS) {
3962 		error = copyin(uap->argp, (caddr_t)&dumplist,
3963 		    sizeof (struct nfsd_dumplist));
3964 		if (!error && (dumplist.ndl_size < 1 ||
3965 			dumplist.ndl_size > NFSRV_MAXDUMPLIST))
3966 			error = EPERM;
3967 		if (!error) {
3968 		    len = sizeof (struct nfsd_dumpclients) * dumplist.ndl_size;
3969 		    dumpclients = malloc(len, M_TEMP, M_WAITOK | M_ZERO);
3970 		    nfsrv_dumpclients(dumpclients, dumplist.ndl_size);
3971 		    error = copyout(dumpclients, dumplist.ndl_list, len);
3972 		    free(dumpclients, M_TEMP);
3973 		}
3974 	} else if (uap->flag & NFSSVC_DUMPLOCKS) {
3975 		error = copyin(uap->argp, (caddr_t)&dumplocklist,
3976 		    sizeof (struct nfsd_dumplocklist));
3977 		if (!error && (dumplocklist.ndllck_size < 1 ||
3978 			dumplocklist.ndllck_size > NFSRV_MAXDUMPLIST))
3979 			error = EPERM;
3980 		if (!error)
3981 			error = nfsrv_lookupfilename(&nd,
3982 				dumplocklist.ndllck_fname, p);
3983 		if (!error) {
3984 			len = sizeof (struct nfsd_dumplocks) *
3985 				dumplocklist.ndllck_size;
3986 			dumplocks = malloc(len, M_TEMP, M_WAITOK | M_ZERO);
3987 			nfsrv_dumplocks(nd.ni_vp, dumplocks,
3988 			    dumplocklist.ndllck_size, p);
3989 			vput(nd.ni_vp);
3990 			error = copyout(dumplocks, dumplocklist.ndllck_list,
3991 			    len);
3992 			free(dumplocks, M_TEMP);
3993 		}
3994 	} else if (uap->flag & NFSSVC_BACKUPSTABLE) {
3995 		procp = p->td_proc;
3996 		PROC_LOCK(procp);
3997 		nfsd_master_pid = procp->p_pid;
3998 		bcopy(procp->p_comm, nfsd_master_comm, MAXCOMLEN + 1);
3999 		nfsd_master_start = procp->p_stats->p_start;
4000 		nfsd_master_proc = procp;
4001 		PROC_UNLOCK(procp);
4002 	} else if ((uap->flag & NFSSVC_SUSPENDNFSD) != 0) {
4003 		NFSLOCKV4ROOTMUTEX();
4004 		if (suspend_nfsd == 0) {
4005 			/* Lock out all nfsd threads */
4006 			do {
4007 				igotlock = nfsv4_lock(&nfsd_suspend_lock, 1,
4008 				    NULL, NFSV4ROOTLOCKMUTEXPTR, NULL);
4009 			} while (igotlock == 0 && suspend_nfsd == 0);
4010 			suspend_nfsd = 1;
4011 		}
4012 		NFSUNLOCKV4ROOTMUTEX();
4013 		error = 0;
4014 	} else if ((uap->flag & NFSSVC_RESUMENFSD) != 0) {
4015 		NFSLOCKV4ROOTMUTEX();
4016 		if (suspend_nfsd != 0) {
4017 			nfsv4_unlock(&nfsd_suspend_lock, 0);
4018 			suspend_nfsd = 0;
4019 		}
4020 		NFSUNLOCKV4ROOTMUTEX();
4021 		error = 0;
4022 	}
4023 
4024 	NFSEXITCODE(error);
4025 	return (error);
4026 }
4027 
4028 /*
4029  * Check exports.
4030  * Returns 0 if ok, 1 otherwise.
4031  */
4032 int
4033 nfsvno_testexp(struct nfsrv_descript *nd, struct nfsexstuff *exp)
4034 {
4035 	int i;
4036 
4037 	if ((NFSVNO_EXTLS(exp) && (nd->nd_flag & ND_TLS) == 0) ||
4038 	    (NFSVNO_EXTLSCERT(exp) &&
4039 	     (nd->nd_flag & ND_TLSCERT) == 0) ||
4040 	    (NFSVNO_EXTLSCERTUSER(exp) &&
4041 	     (nd->nd_flag & ND_TLSCERTUSER) == 0)) {
4042 		if ((nd->nd_flag & ND_NFSV4) != 0)
4043 			return (NFSERR_WRONGSEC);
4044 #ifdef notnow
4045 		/* There is currently no auth_stat for this. */
4046 		else if ((nd->nd_flag & ND_TLS) == 0)
4047 			return (NFSERR_AUTHERR | AUTH_NEEDS_TLS);
4048 		else
4049 			return (NFSERR_AUTHERR | AUTH_NEEDS_TLS_MUTUAL_HOST);
4050 #endif
4051 		else
4052 			return (NFSERR_AUTHERR | AUTH_TOOWEAK);
4053 	}
4054 
4055 	/*
4056 	 * RFC2623 suggests that the NFSv3 Fsinfo RPC be allowed to use
4057 	 * AUTH_NONE or AUTH_SYS for file systems requiring RPCSEC_GSS.
4058 	 */
4059 	if ((nd->nd_flag & ND_NFSV3) != 0 && nd->nd_procnum == NFSPROC_FSINFO)
4060 		return (0);
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 | 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;
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, cred);
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, cred);
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, cred);
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, cred);
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, cred);
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, cred);
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, cred);
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, cred);
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;
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 		if (error == 0) {
6770 			if (vp->v_type == VREG && nfsrv_devidcnt != 0)
6771 				nfsvno_updateds(vp, cred, p);
6772 			error = VOP_FSYNC(vp, MNT_WAIT, p);
6773 		}
6774 		free(iv, M_TEMP);
6775 	}
6776 
6777 out:
6778 	NFSEXITCODE(error);
6779 	return (error);
6780 }
6781 
6782 /*
6783  * For a pNFS server, the DS file's ctime and
6784  * va_filerev (TimeMetadata and Change) needs to
6785  * be updated.  This is a hack, but works by
6786  * flipping the S_ISGID bit in va_mode and then
6787  * flipping it back.
6788  * It does result in two MDS->DS RPCs, but creating
6789  * a custom RPC just to do this seems overkill, since
6790  * Setxattr/Rmxattr will not be done that frequently.
6791  * If it fails part way through, that is not too
6792  * serious, since the DS file is never executed.
6793  */
6794 static void
6795 nfsvno_updateds(struct vnode *vp, struct ucred *cred, NFSPROC_T *p)
6796 {
6797 	struct nfsvattr nva;
6798 	int ret;
6799 	u_short tmode;
6800 
6801 	ret = VOP_GETATTR(vp, &nva.na_vattr, cred);
6802 	if (ret == 0) {
6803 		tmode = nva.na_mode;
6804 		NFSVNO_ATTRINIT(&nva);
6805 		tmode ^= S_ISGID;
6806 		NFSVNO_SETATTRVAL(&nva, mode, tmode);
6807 		ret = nfsrv_proxyds(vp, 0, 0, cred, p,
6808 		    NFSPROC_SETATTR, NULL, NULL, NULL, &nva,
6809 		    NULL, NULL, 0, NULL);
6810 		if (ret == 0) {
6811 			tmode ^= S_ISGID;
6812 			NFSVNO_SETATTRVAL(&nva, mode, tmode);
6813 			ret = nfsrv_proxyds(vp, 0, 0, cred, p,
6814 			    NFSPROC_SETATTR, NULL, NULL, NULL,
6815 			    &nva, NULL, NULL, 0, NULL);
6816 		}
6817 	}
6818 }
6819 
6820 /*
6821  * Remove Extended attribute vnode op.
6822  */
6823 int
6824 nfsvno_rmxattr(struct nfsrv_descript *nd, struct vnode *vp, char *name,
6825     struct ucred *cred, struct thread *p)
6826 {
6827 	int error;
6828 
6829 	/*
6830 	 * Get rid of any delegations.  I am not sure why this is required,
6831 	 * but RFC-8276 says so.
6832 	 */
6833 	error = nfsrv_checkremove(vp, 0, nd, nd->nd_clientid, p);
6834 	if (error != 0)
6835 		goto out;
6836 #ifdef MAC
6837 	error = mac_vnode_check_deleteextattr(cred, vp, EXTATTR_NAMESPACE_USER,
6838 	    name);
6839 	if (error != 0)
6840 		goto out;
6841 #endif
6842 
6843 	error = VOP_DELETEEXTATTR(vp, EXTATTR_NAMESPACE_USER, name, cred, p);
6844 	if (error == EOPNOTSUPP)
6845 		error = VOP_SETEXTATTR(vp, EXTATTR_NAMESPACE_USER, name, NULL,
6846 		    cred, p);
6847 	if (error == 0) {
6848 		if (vp->v_type == VREG && nfsrv_devidcnt != 0)
6849 			nfsvno_updateds(vp, cred, p);
6850 		error = VOP_FSYNC(vp, MNT_WAIT, p);
6851 	}
6852 out:
6853 	NFSEXITCODE(error);
6854 	return (error);
6855 }
6856 
6857 /*
6858  * List Extended Atribute vnode op into an mbuf list.
6859  */
6860 int
6861 nfsvno_listxattr(struct vnode *vp, uint64_t cookie, struct ucred *cred,
6862     struct thread *p, u_char **bufp, uint32_t *lenp, bool *eofp)
6863 {
6864 	struct iovec iv;
6865 	struct uio io;
6866 	int error;
6867 	size_t siz;
6868 
6869 	*bufp = NULL;
6870 	/* First, find out the size of the extended attribute. */
6871 	error = VOP_LISTEXTATTR(vp, EXTATTR_NAMESPACE_USER, NULL, &siz, cred,
6872 	    p);
6873 	if (error != 0)
6874 		return (NFSERR_NOXATTR);
6875 	if (siz <= cookie) {
6876 		*lenp = 0;
6877 		*eofp = true;
6878 		goto out;
6879 	}
6880 	if (siz > cookie + *lenp) {
6881 		siz = cookie + *lenp;
6882 		*eofp = false;
6883 	} else
6884 		*eofp = true;
6885 	/* Just choose a sanity limit of 10Mbytes for malloc(M_TEMP). */
6886 	if (siz > 10 * 1024 * 1024) {
6887 		error = NFSERR_XATTR2BIG;
6888 		goto out;
6889 	}
6890 	*bufp = malloc(siz, M_TEMP, M_WAITOK);
6891 	iv.iov_base = *bufp;
6892 	iv.iov_len = siz;
6893 	io.uio_iovcnt = 1;
6894 	io.uio_iov = &iv;
6895 	io.uio_offset = 0;
6896 	io.uio_resid = siz;
6897 	io.uio_rw = UIO_READ;
6898 	io.uio_segflg = UIO_SYSSPACE;
6899 	io.uio_td = p;
6900 #ifdef MAC
6901 	error = mac_vnode_check_listextattr(cred, vp, EXTATTR_NAMESPACE_USER);
6902 	if (error != 0)
6903 		goto out;
6904 #endif
6905 
6906 	error = VOP_LISTEXTATTR(vp, EXTATTR_NAMESPACE_USER, &io, NULL, cred,
6907 	    p);
6908 	if (error != 0)
6909 		goto out;
6910 	if (io.uio_resid > 0)
6911 		siz -= io.uio_resid;
6912 	*lenp = siz;
6913 
6914 out:
6915 	if (error != 0) {
6916 		free(*bufp, M_TEMP);
6917 		*bufp = NULL;
6918 	}
6919 	NFSEXITCODE(error);
6920 	return (error);
6921 }
6922 
6923 /*
6924  * Trim trailing data off the mbuf list being built.
6925  */
6926 void
6927 nfsm_trimtrailing(struct nfsrv_descript *nd, struct mbuf *mb, char *bpos,
6928     int bextpg, int bextpgsiz)
6929 {
6930 	vm_page_t pg;
6931 	int fullpgsiz, i;
6932 
6933 	if (mb->m_next != NULL) {
6934 		m_freem(mb->m_next);
6935 		mb->m_next = NULL;
6936 	}
6937 	if ((mb->m_flags & M_EXTPG) != 0) {
6938 		KASSERT(bextpg >= 0 && bextpg < mb->m_epg_npgs,
6939 		    ("nfsm_trimtrailing: bextpg out of range"));
6940 		KASSERT(bpos == (char *)(void *)
6941 		    PHYS_TO_DMAP(mb->m_epg_pa[bextpg]) + PAGE_SIZE - bextpgsiz,
6942 		    ("nfsm_trimtrailing: bextpgsiz bad!"));
6943 
6944 		/* First, get rid of any pages after this position. */
6945 		for (i = mb->m_epg_npgs - 1; i > bextpg; i--) {
6946 			pg = PHYS_TO_VM_PAGE(mb->m_epg_pa[i]);
6947 			vm_page_unwire_noq(pg);
6948 			vm_page_free(pg);
6949 		}
6950 		mb->m_epg_npgs = bextpg + 1;
6951 		if (bextpg == 0)
6952 			fullpgsiz = PAGE_SIZE - mb->m_epg_1st_off;
6953 		else
6954 			fullpgsiz = PAGE_SIZE;
6955 		mb->m_epg_last_len = fullpgsiz - bextpgsiz;
6956 		mb->m_len = m_epg_pagelen(mb, 0, mb->m_epg_1st_off);
6957 		for (i = 1; i < mb->m_epg_npgs; i++)
6958 			mb->m_len += m_epg_pagelen(mb, i, 0);
6959 		nd->nd_bextpgsiz = bextpgsiz;
6960 		nd->nd_bextpg = bextpg;
6961 	} else
6962 		mb->m_len = bpos - mtod(mb, char *);
6963 	nd->nd_mb = mb;
6964 	nd->nd_bpos = bpos;
6965 }
6966 
6967 
6968 /*
6969  * Check to see if a put file handle operation should test for
6970  * NFSERR_WRONGSEC, although NFSv3 actually returns NFSERR_AUTHERR.
6971  * When Open is the next operation, NFSERR_WRONGSEC cannot be
6972  * replied for the Open cases that use a component.  This can
6973  * be identified by the fact that the file handle's type is VDIR.
6974  */
6975 bool
6976 nfsrv_checkwrongsec(struct nfsrv_descript *nd, int nextop, enum vtype vtyp)
6977 {
6978 
6979 	if ((nd->nd_flag & ND_NFSV4) == 0)
6980 		return (true);
6981 
6982 	if ((nd->nd_flag & ND_LASTOP) != 0)
6983 		return (false);
6984 
6985 	if (nextop == NFSV4OP_PUTROOTFH || nextop == NFSV4OP_PUTFH ||
6986 	    nextop == NFSV4OP_PUTPUBFH || nextop == NFSV4OP_RESTOREFH ||
6987 	    nextop == NFSV4OP_LOOKUP || nextop == NFSV4OP_LOOKUPP ||
6988 	    nextop == NFSV4OP_SECINFO || nextop == NFSV4OP_SECINFONONAME)
6989 		return (false);
6990 	if (nextop == NFSV4OP_OPEN && vtyp == VDIR)
6991 		return (false);
6992 	return (true);
6993 }
6994 
6995 /*
6996  * Check DSs marked no space.
6997  */
6998 void
6999 nfsrv_checknospc(void)
7000 {
7001 	struct statfs *tsf;
7002 	struct nfsdevice *ds;
7003 	struct vnode **dvpp, **tdvpp, *dvp;
7004 	char *devid, *tdevid;
7005 	int cnt, error = 0, i;
7006 
7007 	if (nfsrv_devidcnt <= 0)
7008 		return;
7009 	dvpp = mallocarray(nfsrv_devidcnt, sizeof(*dvpp), M_TEMP, M_WAITOK);
7010 	devid = malloc(nfsrv_devidcnt * NFSX_V4DEVICEID, M_TEMP, M_WAITOK);
7011 	tsf = malloc(sizeof(*tsf), M_TEMP, M_WAITOK);
7012 
7013 	/* Get an array of the dvps for the DSs. */
7014 	tdvpp = dvpp;
7015 	tdevid = devid;
7016 	i = 0;
7017 	NFSDDSLOCK();
7018 	/* First, search for matches for same file system. */
7019 	TAILQ_FOREACH(ds, &nfsrv_devidhead, nfsdev_list) {
7020 		if (ds->nfsdev_nmp != NULL && ds->nfsdev_nospc) {
7021 			if (++i > nfsrv_devidcnt)
7022 				break;
7023 			*tdvpp++ = ds->nfsdev_dvp;
7024 			NFSBCOPY(ds->nfsdev_deviceid, tdevid, NFSX_V4DEVICEID);
7025 			tdevid += NFSX_V4DEVICEID;
7026 		}
7027 	}
7028 	NFSDDSUNLOCK();
7029 
7030 	/* Do a VFS_STATFS() for each of the DSs and clear no space. */
7031 	cnt = i;
7032 	tdvpp = dvpp;
7033 	tdevid = devid;
7034 	for (i = 0; i < cnt && error == 0; i++) {
7035 		dvp = *tdvpp++;
7036 		error = VFS_STATFS(dvp->v_mount, tsf);
7037 		if (error == 0 && tsf->f_bavail > 0) {
7038 			NFSD_DEBUG(1, "nfsrv_checknospc: reset nospc\n");
7039 			nfsrv_marknospc(tdevid, false);
7040 		}
7041 		tdevid += NFSX_V4DEVICEID;
7042 	}
7043 	free(tsf, M_TEMP);
7044 	free(dvpp, M_TEMP);
7045 	free(devid, M_TEMP);
7046 }
7047 
7048 extern int (*nfsd_call_nfsd)(struct thread *, struct nfssvc_args *);
7049 
7050 /*
7051  * Called once to initialize data structures...
7052  */
7053 static int
7054 nfsd_modevent(module_t mod, int type, void *data)
7055 {
7056 	int error = 0, i;
7057 	static int loaded = 0;
7058 
7059 	switch (type) {
7060 	case MOD_LOAD:
7061 		if (loaded)
7062 			goto out;
7063 		newnfs_portinit();
7064 		for (i = 0; i < NFSRVCACHE_HASHSIZE; i++) {
7065 			mtx_init(&nfsrchash_table[i].mtx, "nfsrtc", NULL,
7066 			    MTX_DEF);
7067 			mtx_init(&nfsrcahash_table[i].mtx, "nfsrtca", NULL,
7068 			    MTX_DEF);
7069 		}
7070 		mtx_init(&nfsrc_udpmtx, "nfsuc", NULL, MTX_DEF);
7071 		mtx_init(&nfs_v4root_mutex, "nfs4rt", NULL, MTX_DEF);
7072 		mtx_init(&nfsv4root_mnt.mnt_mtx, "nfs4mnt", NULL, MTX_DEF);
7073 		mtx_init(&nfsrv_dontlistlock_mtx, "nfs4dnl", NULL, MTX_DEF);
7074 		mtx_init(&nfsrv_recalllock_mtx, "nfs4rec", NULL, MTX_DEF);
7075 		lockinit(&nfsv4root_mnt.mnt_explock, PVFS, "explock", 0, 0);
7076 		callout_init(&nfsd_callout, 1);
7077 		nfsrvd_initcache();
7078 		nfsd_init();
7079 		NFSD_LOCK();
7080 		nfsrvd_init(0);
7081 		NFSD_UNLOCK();
7082 		nfsd_mntinit();
7083 #ifdef VV_DISABLEDELEG
7084 		vn_deleg_ops.vndeleg_recall = nfsd_recalldelegation;
7085 		vn_deleg_ops.vndeleg_disable = nfsd_disabledelegation;
7086 #endif
7087 		nfsd_call_nfsd = nfssvc_nfsd;
7088 		loaded = 1;
7089 		break;
7090 
7091 	case MOD_UNLOAD:
7092 		if (newnfs_numnfsd != 0) {
7093 			error = EBUSY;
7094 			break;
7095 		}
7096 
7097 #ifdef VV_DISABLEDELEG
7098 		vn_deleg_ops.vndeleg_recall = NULL;
7099 		vn_deleg_ops.vndeleg_disable = NULL;
7100 #endif
7101 		nfsd_call_nfsd = NULL;
7102 		callout_drain(&nfsd_callout);
7103 
7104 		/* Clean out all NFSv4 state. */
7105 		nfsrv_throwawayallstate(curthread);
7106 
7107 		/* Clean the NFS server reply cache */
7108 		nfsrvd_cleancache();
7109 
7110 		/* Free up the krpc server pool. */
7111 		if (nfsrvd_pool != NULL)
7112 			svcpool_destroy(nfsrvd_pool);
7113 
7114 		/* and get rid of the locks */
7115 		for (i = 0; i < NFSRVCACHE_HASHSIZE; i++) {
7116 			mtx_destroy(&nfsrchash_table[i].mtx);
7117 			mtx_destroy(&nfsrcahash_table[i].mtx);
7118 		}
7119 		mtx_destroy(&nfsrc_udpmtx);
7120 		mtx_destroy(&nfs_v4root_mutex);
7121 		mtx_destroy(&nfsv4root_mnt.mnt_mtx);
7122 		mtx_destroy(&nfsrv_dontlistlock_mtx);
7123 		mtx_destroy(&nfsrv_recalllock_mtx);
7124 		for (i = 0; i < nfsrv_sessionhashsize; i++)
7125 			mtx_destroy(&nfssessionhash[i].mtx);
7126 		if (nfslayouthash != NULL) {
7127 			for (i = 0; i < nfsrv_layouthashsize; i++)
7128 				mtx_destroy(&nfslayouthash[i].mtx);
7129 			free(nfslayouthash, M_NFSDSESSION);
7130 		}
7131 		lockdestroy(&nfsv4root_mnt.mnt_explock);
7132 		free(nfsclienthash, M_NFSDCLIENT);
7133 		free(nfslockhash, M_NFSDLOCKFILE);
7134 		free(nfssessionhash, M_NFSDSESSION);
7135 		loaded = 0;
7136 		break;
7137 	default:
7138 		error = EOPNOTSUPP;
7139 		break;
7140 	}
7141 
7142 out:
7143 	NFSEXITCODE(error);
7144 	return (error);
7145 }
7146 static moduledata_t nfsd_mod = {
7147 	"nfsd",
7148 	nfsd_modevent,
7149 	NULL,
7150 };
7151 DECLARE_MODULE(nfsd, nfsd_mod, SI_SUB_VFS, SI_ORDER_ANY);
7152 
7153 /* So that loader and kldload(2) can find us, wherever we are.. */
7154 MODULE_VERSION(nfsd, 1);
7155 MODULE_DEPEND(nfsd, nfscommon, 1, 1, 1);
7156 MODULE_DEPEND(nfsd, nfslockd, 1, 1, 1);
7157 MODULE_DEPEND(nfsd, krpc, 1, 1, 1);
7158 MODULE_DEPEND(nfsd, nfssvc, 1, 1, 1);
7159