xref: /freebsd/sys/fs/nfsserver/nfs_nfsdport.c (revision bc5304a006238115291e7568583632889dffbab9)
1 /*-
2  * SPDX-License-Identifier: BSD-3-Clause
3  *
4  * Copyright (c) 1989, 1993
5  *	The Regents of the University of California.  All rights reserved.
6  *
7  * This code is derived from software contributed to Berkeley by
8  * Rick Macklem at The University of Guelph.
9  *
10  * Redistribution and use in source and binary forms, with or without
11  * modification, are permitted provided that the following conditions
12  * are met:
13  * 1. Redistributions of source code must retain the above copyright
14  *    notice, this list of conditions and the following disclaimer.
15  * 2. Redistributions in binary form must reproduce the above copyright
16  *    notice, this list of conditions and the following disclaimer in the
17  *    documentation and/or other materials provided with the distribution.
18  * 3. Neither the name of the University nor the names of its contributors
19  *    may be used to endorse or promote products derived from this software
20  *    without specific prior written permission.
21  *
22  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
23  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
24  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
25  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
26  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
27  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
28  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
29  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
30  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
31  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
32  * SUCH DAMAGE.
33  *
34  */
35 
36 #include <sys/cdefs.h>
37 __FBSDID("$FreeBSD$");
38 
39 #include <sys/capsicum.h>
40 #include <sys/extattr.h>
41 
42 /*
43  * Functions that perform the vfs operations required by the routines in
44  * nfsd_serv.c. It is hoped that this change will make the server more
45  * portable.
46  */
47 
48 #include <fs/nfs/nfsport.h>
49 #include <security/mac/mac_framework.h>
50 #include <sys/filio.h>
51 #include <sys/hash.h>
52 #include <sys/sysctl.h>
53 #include <nlm/nlm_prot.h>
54 #include <nlm/nlm.h>
55 
56 FEATURE(nfsd, "NFSv4 server");
57 
58 extern u_int32_t newnfs_true, newnfs_false, newnfs_xdrneg1;
59 extern int nfsrv_useacl;
60 extern int newnfs_numnfsd;
61 extern struct mount nfsv4root_mnt;
62 extern struct nfsrv_stablefirst nfsrv_stablefirst;
63 extern void (*nfsd_call_servertimer)(void);
64 extern SVCPOOL	*nfsrvd_pool;
65 extern struct nfsv4lock nfsd_suspend_lock;
66 extern struct nfsclienthashhead *nfsclienthash;
67 extern struct nfslockhashhead *nfslockhash;
68 extern struct nfssessionhash *nfssessionhash;
69 extern int nfsrv_sessionhashsize;
70 extern struct nfsstatsv1 nfsstatsv1;
71 extern struct nfslayouthash *nfslayouthash;
72 extern int nfsrv_layouthashsize;
73 extern struct mtx nfsrv_dslock_mtx;
74 extern int nfs_pnfsiothreads;
75 extern struct nfsdontlisthead nfsrv_dontlisthead;
76 extern volatile int nfsrv_dontlistlen;
77 extern volatile int nfsrv_devidcnt;
78 extern int nfsrv_maxpnfsmirror;
79 extern uint32_t nfs_srvmaxio;
80 extern int nfs_bufpackets;
81 extern u_long sb_max_adj;
82 struct vfsoptlist nfsv4root_opt, nfsv4root_newopt;
83 NFSDLOCKMUTEX;
84 NFSSTATESPINLOCK;
85 struct nfsrchash_bucket nfsrchash_table[NFSRVCACHE_HASHSIZE];
86 struct nfsrchash_bucket nfsrcahash_table[NFSRVCACHE_HASHSIZE];
87 struct mtx nfsrc_udpmtx;
88 struct mtx nfs_v4root_mutex;
89 struct mtx nfsrv_dontlistlock_mtx;
90 struct mtx nfsrv_recalllock_mtx;
91 struct nfsrvfh nfs_rootfh, nfs_pubfh;
92 int nfs_pubfhset = 0, nfs_rootfhset = 0;
93 struct proc *nfsd_master_proc = NULL;
94 int nfsd_debuglevel = 0;
95 static pid_t nfsd_master_pid = (pid_t)-1;
96 static char nfsd_master_comm[MAXCOMLEN + 1];
97 static struct timeval nfsd_master_start;
98 static uint32_t nfsv4_sysid = 0;
99 static fhandle_t zerofh;
100 
101 static int nfssvc_srvcall(struct thread *, struct nfssvc_args *,
102     struct ucred *);
103 
104 int nfsrv_enable_crossmntpt = 1;
105 static int nfs_commit_blks;
106 static int nfs_commit_miss;
107 extern int nfsrv_issuedelegs;
108 extern int nfsrv_dolocallocks;
109 extern int nfsd_enable_stringtouid;
110 extern struct nfsdevicehead nfsrv_devidhead;
111 
112 static int nfsrv_createiovec(int, struct mbuf **, struct mbuf **,
113     struct iovec **);
114 static int nfsrv_createiovec_extpgs(int, int, struct mbuf **,
115     struct mbuf **, struct iovec **);
116 static int nfsrv_createiovecw(int, struct mbuf *, char *, struct iovec **,
117     int *);
118 static void nfsrv_pnfscreate(struct vnode *, struct vattr *, struct ucred *,
119     NFSPROC_T *);
120 static void nfsrv_pnfsremovesetup(struct vnode *, NFSPROC_T *, struct vnode **,
121     int *, char *, fhandle_t *);
122 static void nfsrv_pnfsremove(struct vnode **, int, char *, fhandle_t *,
123     NFSPROC_T *);
124 static int nfsrv_proxyds(struct vnode *, off_t, int, struct ucred *,
125     struct thread *, int, struct mbuf **, char *, struct mbuf **,
126     struct nfsvattr *, struct acl *, off_t *, int, bool *);
127 static int nfsrv_setextattr(struct vnode *, struct nfsvattr *, NFSPROC_T *);
128 static int nfsrv_readdsrpc(fhandle_t *, off_t, int, struct ucred *,
129     NFSPROC_T *, struct nfsmount *, struct mbuf **, struct mbuf **);
130 static int nfsrv_writedsrpc(fhandle_t *, off_t, int, struct ucred *,
131     NFSPROC_T *, struct vnode *, struct nfsmount **, int, struct mbuf **,
132     char *, int *);
133 static int nfsrv_allocatedsrpc(fhandle_t *, off_t, off_t, struct ucred *,
134     NFSPROC_T *, struct vnode *, struct nfsmount **, int, int *);
135 static int nfsrv_setacldsrpc(fhandle_t *, struct ucred *, NFSPROC_T *,
136     struct vnode *, struct nfsmount **, int, struct acl *, int *);
137 static int nfsrv_setattrdsrpc(fhandle_t *, struct ucred *, NFSPROC_T *,
138     struct vnode *, struct nfsmount **, int, struct nfsvattr *, int *);
139 static int nfsrv_getattrdsrpc(fhandle_t *, struct ucred *, NFSPROC_T *,
140     struct vnode *, struct nfsmount *, struct nfsvattr *);
141 static int nfsrv_seekdsrpc(fhandle_t *, off_t *, int, bool *, struct ucred *,
142     NFSPROC_T *, struct nfsmount *);
143 static int nfsrv_putfhname(fhandle_t *, char *);
144 static int nfsrv_pnfslookupds(struct vnode *, struct vnode *,
145     struct pnfsdsfile *, struct vnode **, NFSPROC_T *);
146 static void nfsrv_pnfssetfh(struct vnode *, struct pnfsdsfile *, char *, char *,
147     struct vnode *, NFSPROC_T *);
148 static int nfsrv_dsremove(struct vnode *, char *, struct ucred *, NFSPROC_T *);
149 static int nfsrv_dssetacl(struct vnode *, struct acl *, struct ucred *,
150     NFSPROC_T *);
151 static int nfsrv_pnfsstatfs(struct statfs *, struct mount *);
152 
153 int nfs_pnfsio(task_fn_t *, void *);
154 
155 SYSCTL_NODE(_vfs, OID_AUTO, nfsd, CTLFLAG_RW | CTLFLAG_MPSAFE, 0,
156     "NFS server");
157 SYSCTL_INT(_vfs_nfsd, OID_AUTO, mirrormnt, CTLFLAG_RW,
158     &nfsrv_enable_crossmntpt, 0, "Enable nfsd to cross mount points");
159 SYSCTL_INT(_vfs_nfsd, OID_AUTO, commit_blks, CTLFLAG_RW, &nfs_commit_blks,
160     0, "");
161 SYSCTL_INT(_vfs_nfsd, OID_AUTO, commit_miss, CTLFLAG_RW, &nfs_commit_miss,
162     0, "");
163 SYSCTL_INT(_vfs_nfsd, OID_AUTO, issue_delegations, CTLFLAG_RW,
164     &nfsrv_issuedelegs, 0, "Enable nfsd to issue delegations");
165 SYSCTL_INT(_vfs_nfsd, OID_AUTO, enable_locallocks, CTLFLAG_RW,
166     &nfsrv_dolocallocks, 0, "Enable nfsd to acquire local locks on files");
167 SYSCTL_INT(_vfs_nfsd, OID_AUTO, debuglevel, CTLFLAG_RW, &nfsd_debuglevel,
168     0, "Debug level for NFS server");
169 SYSCTL_INT(_vfs_nfsd, OID_AUTO, enable_stringtouid, CTLFLAG_RW,
170     &nfsd_enable_stringtouid, 0, "Enable nfsd to accept numeric owner_names");
171 static int nfsrv_pnfsgetdsattr = 1;
172 SYSCTL_INT(_vfs_nfsd, OID_AUTO, pnfsgetdsattr, CTLFLAG_RW,
173     &nfsrv_pnfsgetdsattr, 0, "When set getattr gets DS attributes via RPC");
174 
175 /*
176  * nfsrv_dsdirsize can only be increased and only when the nfsd threads are
177  * not running.
178  * The dsN subdirectories for the increased values must have been created
179  * on all DS servers before this increase is done.
180  */
181 u_int	nfsrv_dsdirsize = 20;
182 static int
183 sysctl_dsdirsize(SYSCTL_HANDLER_ARGS)
184 {
185 	int error, newdsdirsize;
186 
187 	newdsdirsize = nfsrv_dsdirsize;
188 	error = sysctl_handle_int(oidp, &newdsdirsize, 0, req);
189 	if (error != 0 || req->newptr == NULL)
190 		return (error);
191 	if (newdsdirsize <= nfsrv_dsdirsize || newdsdirsize > 10000 ||
192 	    newnfs_numnfsd != 0)
193 		return (EINVAL);
194 	nfsrv_dsdirsize = newdsdirsize;
195 	return (0);
196 }
197 SYSCTL_PROC(_vfs_nfsd, OID_AUTO, dsdirsize,
198     CTLTYPE_UINT | CTLFLAG_MPSAFE | CTLFLAG_RW, 0, sizeof(nfsrv_dsdirsize),
199     sysctl_dsdirsize, "IU", "Number of dsN subdirs on the DS servers");
200 
201 /*
202  * nfs_srvmaxio can only be increased and only when the nfsd threads are
203  * not running.  The setting must be a power of 2, with the current limit of
204  * 1Mbyte.
205  */
206 static int
207 sysctl_srvmaxio(SYSCTL_HANDLER_ARGS)
208 {
209 	int error;
210 	u_int newsrvmaxio;
211 	uint64_t tval;
212 
213 	newsrvmaxio = nfs_srvmaxio;
214 	error = sysctl_handle_int(oidp, &newsrvmaxio, 0, req);
215 	if (error != 0 || req->newptr == NULL)
216 		return (error);
217 	if (newsrvmaxio == nfs_srvmaxio)
218 		return (0);
219 	if (newsrvmaxio < nfs_srvmaxio) {
220 		printf("nfsd: vfs.nfsd.srvmaxio can only be increased\n");
221 		return (EINVAL);
222 	}
223 	if (newsrvmaxio > 1048576) {
224 		printf("nfsd: vfs.nfsd.srvmaxio cannot be > 1Mbyte\n");
225 		return (EINVAL);
226 	}
227 	if ((newsrvmaxio & (newsrvmaxio - 1)) != 0) {
228 		printf("nfsd: vfs.nfsd.srvmaxio must be a power of 2\n");
229 		return (EINVAL);
230 	}
231 
232 	/*
233 	 * Check that kern.ipc.maxsockbuf is large enough for
234 	 * newsrviomax, given the setting of vfs.nfs.bufpackets.
235 	 */
236 	if ((newsrvmaxio + NFS_MAXXDR) * nfs_bufpackets >
237 	    sb_max_adj) {
238 		/*
239 		 * Suggest vfs.nfs.bufpackets * maximum RPC message for
240 		 * sb_max_adj.
241 		 */
242 		tval = (newsrvmaxio + NFS_MAXXDR) * nfs_bufpackets;
243 
244 		/*
245 		 * Convert suggested sb_max_adj value to a suggested
246 		 * sb_max value, which is what is set via kern.ipc.maxsockbuf.
247 		 * Perform the inverse calculation of (from uipc_sockbuf.c):
248 		 * sb_max_adj = (u_quad_t)sb_max * MCLBYTES /
249 		 *     (MSIZE + MCLBYTES);
250 		 * XXX If the calculation of sb_max_adj from sb_max changes,
251 		 *     this calculation must be changed as well.
252 		 */
253 		tval *= (MSIZE + MCLBYTES);  /* Brackets for readability. */
254 		tval += MCLBYTES - 1;        /* Round up divide. */
255 		tval /= MCLBYTES;
256 		printf("nfsd: set kern.ipc.maxsockbuf to a minimum of "
257 		    "%ju to support %ubyte NFS I/O\n", (uintmax_t)tval,
258 		    newsrvmaxio);
259 		return (EINVAL);
260 	}
261 
262 	NFSD_LOCK();
263 	if (newnfs_numnfsd != 0) {
264 		NFSD_UNLOCK();
265 		printf("nfsd: cannot set vfs.nfsd.srvmaxio when nfsd "
266 		    "threads are running\n");
267 		return (EINVAL);
268 	}
269 
270 
271 	nfs_srvmaxio = newsrvmaxio;
272 	NFSD_UNLOCK();
273 	return (0);
274 }
275 SYSCTL_PROC(_vfs_nfsd, OID_AUTO, srvmaxio,
276     CTLTYPE_UINT | CTLFLAG_MPSAFE | CTLFLAG_RW, NULL, 0,
277     sysctl_srvmaxio, "IU", "Maximum I/O size in bytes");
278 
279 #define	MAX_REORDERED_RPC	16
280 #define	NUM_HEURISTIC		1031
281 #define	NHUSE_INIT		64
282 #define	NHUSE_INC		16
283 #define	NHUSE_MAX		2048
284 
285 static struct nfsheur {
286 	struct vnode *nh_vp;	/* vp to match (unreferenced pointer) */
287 	off_t nh_nextoff;	/* next offset for sequential detection */
288 	int nh_use;		/* use count for selection */
289 	int nh_seqcount;	/* heuristic */
290 } nfsheur[NUM_HEURISTIC];
291 
292 /*
293  * Heuristic to detect sequential operation.
294  */
295 static struct nfsheur *
296 nfsrv_sequential_heuristic(struct uio *uio, struct vnode *vp)
297 {
298 	struct nfsheur *nh;
299 	int hi, try;
300 
301 	/* Locate best candidate. */
302 	try = 32;
303 	hi = ((int)(vm_offset_t)vp / sizeof(struct vnode)) % NUM_HEURISTIC;
304 	nh = &nfsheur[hi];
305 	while (try--) {
306 		if (nfsheur[hi].nh_vp == vp) {
307 			nh = &nfsheur[hi];
308 			break;
309 		}
310 		if (nfsheur[hi].nh_use > 0)
311 			--nfsheur[hi].nh_use;
312 		hi = (hi + 1) % NUM_HEURISTIC;
313 		if (nfsheur[hi].nh_use < nh->nh_use)
314 			nh = &nfsheur[hi];
315 	}
316 
317 	/* Initialize hint if this is a new file. */
318 	if (nh->nh_vp != vp) {
319 		nh->nh_vp = vp;
320 		nh->nh_nextoff = uio->uio_offset;
321 		nh->nh_use = NHUSE_INIT;
322 		if (uio->uio_offset == 0)
323 			nh->nh_seqcount = 4;
324 		else
325 			nh->nh_seqcount = 1;
326 	}
327 
328 	/* Calculate heuristic. */
329 	if ((uio->uio_offset == 0 && nh->nh_seqcount > 0) ||
330 	    uio->uio_offset == nh->nh_nextoff) {
331 		/* See comments in vfs_vnops.c:sequential_heuristic(). */
332 		nh->nh_seqcount += howmany(uio->uio_resid, 16384);
333 		if (nh->nh_seqcount > IO_SEQMAX)
334 			nh->nh_seqcount = IO_SEQMAX;
335 	} else if (qabs(uio->uio_offset - nh->nh_nextoff) <= MAX_REORDERED_RPC *
336 	    imax(vp->v_mount->mnt_stat.f_iosize, uio->uio_resid)) {
337 		/* Probably a reordered RPC, leave seqcount alone. */
338 	} else if (nh->nh_seqcount > 1) {
339 		nh->nh_seqcount /= 2;
340 	} else {
341 		nh->nh_seqcount = 0;
342 	}
343 	nh->nh_use += NHUSE_INC;
344 	if (nh->nh_use > NHUSE_MAX)
345 		nh->nh_use = NHUSE_MAX;
346 	return (nh);
347 }
348 
349 /*
350  * Get attributes into nfsvattr structure.
351  */
352 int
353 nfsvno_getattr(struct vnode *vp, struct nfsvattr *nvap,
354     struct nfsrv_descript *nd, struct thread *p, int vpislocked,
355     nfsattrbit_t *attrbitp)
356 {
357 	int error, gotattr, lockedit = 0;
358 	struct nfsvattr na;
359 
360 	if (vpislocked == 0) {
361 		/*
362 		 * When vpislocked == 0, the vnode is either exclusively
363 		 * locked by this thread or not locked by this thread.
364 		 * As such, shared lock it, if not exclusively locked.
365 		 */
366 		if (NFSVOPISLOCKED(vp) != LK_EXCLUSIVE) {
367 			lockedit = 1;
368 			NFSVOPLOCK(vp, LK_SHARED | LK_RETRY);
369 		}
370 	}
371 
372 	/*
373 	 * Acquire the Change, Size, TimeAccess, TimeModify and SpaceUsed
374 	 * attributes, as required.
375 	 * This needs to be done for regular files if:
376 	 * - non-NFSv4 RPCs or
377 	 * - when attrbitp == NULL or
378 	 * - an NFSv4 RPC with any of the above attributes in attrbitp.
379 	 * A return of 0 for nfsrv_proxyds() indicates that it has acquired
380 	 * these attributes.  nfsrv_proxyds() will return an error if the
381 	 * server is not a pNFS one.
382 	 */
383 	gotattr = 0;
384 	if (vp->v_type == VREG && nfsrv_devidcnt > 0 && (attrbitp == NULL ||
385 	    (nd->nd_flag & ND_NFSV4) == 0 ||
386 	    NFSISSET_ATTRBIT(attrbitp, NFSATTRBIT_CHANGE) ||
387 	    NFSISSET_ATTRBIT(attrbitp, NFSATTRBIT_SIZE) ||
388 	    NFSISSET_ATTRBIT(attrbitp, NFSATTRBIT_TIMEACCESS) ||
389 	    NFSISSET_ATTRBIT(attrbitp, NFSATTRBIT_TIMEMODIFY) ||
390 	    NFSISSET_ATTRBIT(attrbitp, NFSATTRBIT_SPACEUSED))) {
391 		error = nfsrv_proxyds(vp, 0, 0, nd->nd_cred, p,
392 		    NFSPROC_GETATTR, NULL, NULL, NULL, &na, NULL, NULL, 0,
393 		    NULL);
394 		if (error == 0)
395 			gotattr = 1;
396 	}
397 
398 	error = VOP_GETATTR(vp, &nvap->na_vattr, nd->nd_cred);
399 	if (lockedit != 0)
400 		NFSVOPUNLOCK(vp);
401 
402 	/*
403 	 * If we got the Change, Size and Modify Time from the DS,
404 	 * replace them.
405 	 */
406 	if (gotattr != 0) {
407 		nvap->na_atime = na.na_atime;
408 		nvap->na_mtime = na.na_mtime;
409 		nvap->na_filerev = na.na_filerev;
410 		nvap->na_size = na.na_size;
411 		nvap->na_bytes = na.na_bytes;
412 	}
413 	NFSD_DEBUG(4, "nfsvno_getattr: gotattr=%d err=%d chg=%ju\n", gotattr,
414 	    error, (uintmax_t)na.na_filerev);
415 
416 	NFSEXITCODE(error);
417 	return (error);
418 }
419 
420 /*
421  * Get a file handle for a vnode.
422  */
423 int
424 nfsvno_getfh(struct vnode *vp, fhandle_t *fhp, struct thread *p)
425 {
426 	int error;
427 
428 	NFSBZERO((caddr_t)fhp, sizeof(fhandle_t));
429 	fhp->fh_fsid = vp->v_mount->mnt_stat.f_fsid;
430 	error = VOP_VPTOFH(vp, &fhp->fh_fid);
431 
432 	NFSEXITCODE(error);
433 	return (error);
434 }
435 
436 /*
437  * Perform access checking for vnodes obtained from file handles that would
438  * refer to files already opened by a Unix client. You cannot just use
439  * vn_writechk() and VOP_ACCESSX() for two reasons.
440  * 1 - You must check for exported rdonly as well as MNT_RDONLY for the write
441  *     case.
442  * 2 - The owner is to be given access irrespective of mode bits for some
443  *     operations, so that processes that chmod after opening a file don't
444  *     break.
445  */
446 int
447 nfsvno_accchk(struct vnode *vp, accmode_t accmode, struct ucred *cred,
448     struct nfsexstuff *exp, struct thread *p, int override, int vpislocked,
449     u_int32_t *supportedtypep)
450 {
451 	struct vattr vattr;
452 	int error = 0, getret = 0;
453 
454 	if (vpislocked == 0) {
455 		if (NFSVOPLOCK(vp, LK_SHARED) != 0) {
456 			error = EPERM;
457 			goto out;
458 		}
459 	}
460 	if (accmode & VWRITE) {
461 		/* Just vn_writechk() changed to check rdonly */
462 		/*
463 		 * Disallow write attempts on read-only file systems;
464 		 * unless the file is a socket or a block or character
465 		 * device resident on the file system.
466 		 */
467 		if (NFSVNO_EXRDONLY(exp) ||
468 		    (vp->v_mount->mnt_flag & MNT_RDONLY)) {
469 			switch (vp->v_type) {
470 			case VREG:
471 			case VDIR:
472 			case VLNK:
473 				error = EROFS;
474 			default:
475 				break;
476 			}
477 		}
478 		/*
479 		 * If there's shared text associated with
480 		 * the inode, try to free it up once.  If
481 		 * we fail, we can't allow writing.
482 		 */
483 		if (VOP_IS_TEXT(vp) && error == 0)
484 			error = ETXTBSY;
485 	}
486 	if (error != 0) {
487 		if (vpislocked == 0)
488 			NFSVOPUNLOCK(vp);
489 		goto out;
490 	}
491 
492 	/*
493 	 * Should the override still be applied when ACLs are enabled?
494 	 */
495 	error = VOP_ACCESSX(vp, accmode, cred, p);
496 	if (error != 0 && (accmode & (VDELETE | VDELETE_CHILD))) {
497 		/*
498 		 * Try again with VEXPLICIT_DENY, to see if the test for
499 		 * deletion is supported.
500 		 */
501 		error = VOP_ACCESSX(vp, accmode | VEXPLICIT_DENY, cred, p);
502 		if (error == 0) {
503 			if (vp->v_type == VDIR) {
504 				accmode &= ~(VDELETE | VDELETE_CHILD);
505 				accmode |= VWRITE;
506 				error = VOP_ACCESSX(vp, accmode, cred, p);
507 			} else if (supportedtypep != NULL) {
508 				*supportedtypep &= ~NFSACCESS_DELETE;
509 			}
510 		}
511 	}
512 
513 	/*
514 	 * Allow certain operations for the owner (reads and writes
515 	 * on files that are already open).
516 	 */
517 	if (override != NFSACCCHK_NOOVERRIDE &&
518 	    (error == EPERM || error == EACCES)) {
519 		if (cred->cr_uid == 0 && (override & NFSACCCHK_ALLOWROOT))
520 			error = 0;
521 		else if (override & NFSACCCHK_ALLOWOWNER) {
522 			getret = VOP_GETATTR(vp, &vattr, cred);
523 			if (getret == 0 && cred->cr_uid == vattr.va_uid)
524 				error = 0;
525 		}
526 	}
527 	if (vpislocked == 0)
528 		NFSVOPUNLOCK(vp);
529 
530 out:
531 	NFSEXITCODE(error);
532 	return (error);
533 }
534 
535 /*
536  * Set attribute(s) vnop.
537  */
538 int
539 nfsvno_setattr(struct vnode *vp, struct nfsvattr *nvap, struct ucred *cred,
540     struct thread *p, struct nfsexstuff *exp)
541 {
542 	u_quad_t savsize = 0;
543 	int error, savedit;
544 	time_t savbtime;
545 
546 	/*
547 	 * If this is an exported file system and a pNFS service is running,
548 	 * don't VOP_SETATTR() of size for the MDS file system.
549 	 */
550 	savedit = 0;
551 	error = 0;
552 	if (vp->v_type == VREG && (vp->v_mount->mnt_flag & MNT_EXPORTED) != 0 &&
553 	    nfsrv_devidcnt != 0 && nvap->na_vattr.va_size != VNOVAL &&
554 	    nvap->na_vattr.va_size > 0) {
555 		savsize = nvap->na_vattr.va_size;
556 		nvap->na_vattr.va_size = VNOVAL;
557 		if (nvap->na_vattr.va_uid != (uid_t)VNOVAL ||
558 		    nvap->na_vattr.va_gid != (gid_t)VNOVAL ||
559 		    nvap->na_vattr.va_mode != (mode_t)VNOVAL ||
560 		    nvap->na_vattr.va_atime.tv_sec != VNOVAL ||
561 		    nvap->na_vattr.va_mtime.tv_sec != VNOVAL)
562 			savedit = 1;
563 		else
564 			savedit = 2;
565 	}
566 	if (savedit != 2)
567 		error = VOP_SETATTR(vp, &nvap->na_vattr, cred);
568 	if (savedit != 0)
569 		nvap->na_vattr.va_size = savsize;
570 	if (error == 0 && (nvap->na_vattr.va_uid != (uid_t)VNOVAL ||
571 	    nvap->na_vattr.va_gid != (gid_t)VNOVAL ||
572 	    nvap->na_vattr.va_size != VNOVAL ||
573 	    nvap->na_vattr.va_mode != (mode_t)VNOVAL ||
574 	    nvap->na_vattr.va_atime.tv_sec != VNOVAL ||
575 	    nvap->na_vattr.va_mtime.tv_sec != VNOVAL)) {
576 		/* Never modify birthtime on a DS file. */
577 		savbtime = nvap->na_vattr.va_birthtime.tv_sec;
578 		nvap->na_vattr.va_birthtime.tv_sec = VNOVAL;
579 		/* For a pNFS server, set the attributes on the DS file. */
580 		error = nfsrv_proxyds(vp, 0, 0, cred, p, NFSPROC_SETATTR,
581 		    NULL, NULL, NULL, nvap, NULL, NULL, 0, NULL);
582 		nvap->na_vattr.va_birthtime.tv_sec = savbtime;
583 		if (error == ENOENT)
584 			error = 0;
585 	}
586 	NFSEXITCODE(error);
587 	return (error);
588 }
589 
590 /*
591  * Set up nameidata for a lookup() call and do it.
592  */
593 int
594 nfsvno_namei(struct nfsrv_descript *nd, struct nameidata *ndp,
595     struct vnode *dp, int islocked, struct nfsexstuff *exp, struct thread *p,
596     struct vnode **retdirp)
597 {
598 	struct componentname *cnp = &ndp->ni_cnd;
599 	int i;
600 	struct iovec aiov;
601 	struct uio auio;
602 	int lockleaf = (cnp->cn_flags & LOCKLEAF) != 0, linklen;
603 	int error = 0;
604 	char *cp;
605 
606 	*retdirp = NULL;
607 	cnp->cn_nameptr = cnp->cn_pnbuf;
608 	ndp->ni_lcf = 0;
609 	/*
610 	 * Extract and set starting directory.
611 	 */
612 	if (dp->v_type != VDIR) {
613 		if (islocked)
614 			vput(dp);
615 		else
616 			vrele(dp);
617 		nfsvno_relpathbuf(ndp);
618 		error = ENOTDIR;
619 		goto out1;
620 	}
621 	if (islocked)
622 		NFSVOPUNLOCK(dp);
623 	VREF(dp);
624 	*retdirp = dp;
625 	if (NFSVNO_EXRDONLY(exp))
626 		cnp->cn_flags |= RDONLY;
627 	ndp->ni_segflg = UIO_SYSSPACE;
628 
629 	if (nd->nd_flag & ND_PUBLOOKUP) {
630 		ndp->ni_loopcnt = 0;
631 		if (cnp->cn_pnbuf[0] == '/') {
632 			vrele(dp);
633 			/*
634 			 * Check for degenerate pathnames here, since lookup()
635 			 * panics on them.
636 			 */
637 			for (i = 1; i < ndp->ni_pathlen; i++)
638 				if (cnp->cn_pnbuf[i] != '/')
639 					break;
640 			if (i == ndp->ni_pathlen) {
641 				error = NFSERR_ACCES;
642 				goto out;
643 			}
644 			dp = rootvnode;
645 			VREF(dp);
646 		}
647 	} else if ((nfsrv_enable_crossmntpt == 0 && NFSVNO_EXPORTED(exp)) ||
648 	    (nd->nd_flag & ND_NFSV4) == 0) {
649 		/*
650 		 * Only cross mount points for NFSv4 when doing a
651 		 * mount while traversing the file system above
652 		 * the mount point, unless nfsrv_enable_crossmntpt is set.
653 		 */
654 		cnp->cn_flags |= NOCROSSMOUNT;
655 	}
656 
657 	/*
658 	 * Initialize for scan, set ni_startdir and bump ref on dp again
659 	 * because lookup() will dereference ni_startdir.
660 	 */
661 
662 	cnp->cn_thread = p;
663 	ndp->ni_startdir = dp;
664 	ndp->ni_rootdir = rootvnode;
665 	ndp->ni_topdir = NULL;
666 
667 	if (!lockleaf)
668 		cnp->cn_flags |= LOCKLEAF;
669 	for (;;) {
670 		cnp->cn_nameptr = cnp->cn_pnbuf;
671 		/*
672 		 * Call lookup() to do the real work.  If an error occurs,
673 		 * ndp->ni_vp and ni_dvp are left uninitialized or NULL and
674 		 * we do not have to dereference anything before returning.
675 		 * In either case ni_startdir will be dereferenced and NULLed
676 		 * out.
677 		 */
678 		error = lookup(ndp);
679 		if (error)
680 			break;
681 
682 		/*
683 		 * Check for encountering a symbolic link.  Trivial
684 		 * termination occurs if no symlink encountered.
685 		 */
686 		if ((cnp->cn_flags & ISSYMLINK) == 0) {
687 			if ((cnp->cn_flags & (SAVENAME | SAVESTART)) == 0)
688 				nfsvno_relpathbuf(ndp);
689 			if (ndp->ni_vp && !lockleaf)
690 				NFSVOPUNLOCK(ndp->ni_vp);
691 			break;
692 		}
693 
694 		/*
695 		 * Validate symlink
696 		 */
697 		if ((cnp->cn_flags & LOCKPARENT) && ndp->ni_pathlen == 1)
698 			NFSVOPUNLOCK(ndp->ni_dvp);
699 		if (!(nd->nd_flag & ND_PUBLOOKUP)) {
700 			error = EINVAL;
701 			goto badlink2;
702 		}
703 
704 		if (ndp->ni_loopcnt++ >= MAXSYMLINKS) {
705 			error = ELOOP;
706 			goto badlink2;
707 		}
708 		if (ndp->ni_pathlen > 1)
709 			cp = uma_zalloc(namei_zone, M_WAITOK);
710 		else
711 			cp = cnp->cn_pnbuf;
712 		aiov.iov_base = cp;
713 		aiov.iov_len = MAXPATHLEN;
714 		auio.uio_iov = &aiov;
715 		auio.uio_iovcnt = 1;
716 		auio.uio_offset = 0;
717 		auio.uio_rw = UIO_READ;
718 		auio.uio_segflg = UIO_SYSSPACE;
719 		auio.uio_td = NULL;
720 		auio.uio_resid = MAXPATHLEN;
721 		error = VOP_READLINK(ndp->ni_vp, &auio, cnp->cn_cred);
722 		if (error) {
723 		badlink1:
724 			if (ndp->ni_pathlen > 1)
725 				uma_zfree(namei_zone, cp);
726 		badlink2:
727 			vrele(ndp->ni_dvp);
728 			vput(ndp->ni_vp);
729 			break;
730 		}
731 		linklen = MAXPATHLEN - auio.uio_resid;
732 		if (linklen == 0) {
733 			error = ENOENT;
734 			goto badlink1;
735 		}
736 		if (linklen + ndp->ni_pathlen >= MAXPATHLEN) {
737 			error = ENAMETOOLONG;
738 			goto badlink1;
739 		}
740 
741 		/*
742 		 * Adjust or replace path
743 		 */
744 		if (ndp->ni_pathlen > 1) {
745 			NFSBCOPY(ndp->ni_next, cp + linklen, ndp->ni_pathlen);
746 			uma_zfree(namei_zone, cnp->cn_pnbuf);
747 			cnp->cn_pnbuf = cp;
748 		} else
749 			cnp->cn_pnbuf[linklen] = '\0';
750 		ndp->ni_pathlen += linklen;
751 
752 		/*
753 		 * Cleanup refs for next loop and check if root directory
754 		 * should replace current directory.  Normally ni_dvp
755 		 * becomes the new base directory and is cleaned up when
756 		 * we loop.  Explicitly null pointers after invalidation
757 		 * to clarify operation.
758 		 */
759 		vput(ndp->ni_vp);
760 		ndp->ni_vp = NULL;
761 
762 		if (cnp->cn_pnbuf[0] == '/') {
763 			vrele(ndp->ni_dvp);
764 			ndp->ni_dvp = ndp->ni_rootdir;
765 			VREF(ndp->ni_dvp);
766 		}
767 		ndp->ni_startdir = ndp->ni_dvp;
768 		ndp->ni_dvp = NULL;
769 	}
770 	if (!lockleaf)
771 		cnp->cn_flags &= ~LOCKLEAF;
772 
773 out:
774 	if (error) {
775 		nfsvno_relpathbuf(ndp);
776 		ndp->ni_vp = NULL;
777 		ndp->ni_dvp = NULL;
778 		ndp->ni_startdir = NULL;
779 	} else if ((ndp->ni_cnd.cn_flags & (WANTPARENT|LOCKPARENT)) == 0) {
780 		ndp->ni_dvp = NULL;
781 	}
782 
783 out1:
784 	NFSEXITCODE2(error, nd);
785 	return (error);
786 }
787 
788 /*
789  * Set up a pathname buffer and return a pointer to it and, optionally
790  * set a hash pointer.
791  */
792 void
793 nfsvno_setpathbuf(struct nameidata *ndp, char **bufpp, u_long **hashpp)
794 {
795 	struct componentname *cnp = &ndp->ni_cnd;
796 
797 	cnp->cn_flags |= (NOMACCHECK | HASBUF);
798 	cnp->cn_pnbuf = uma_zalloc(namei_zone, M_WAITOK);
799 	if (hashpp != NULL)
800 		*hashpp = NULL;
801 	*bufpp = cnp->cn_pnbuf;
802 }
803 
804 /*
805  * Release the above path buffer, if not released by nfsvno_namei().
806  */
807 void
808 nfsvno_relpathbuf(struct nameidata *ndp)
809 {
810 
811 	if ((ndp->ni_cnd.cn_flags & HASBUF) == 0)
812 		panic("nfsrelpath");
813 	uma_zfree(namei_zone, ndp->ni_cnd.cn_pnbuf);
814 	ndp->ni_cnd.cn_flags &= ~HASBUF;
815 }
816 
817 /*
818  * Readlink vnode op into an mbuf list.
819  */
820 int
821 nfsvno_readlink(struct vnode *vp, struct ucred *cred, int maxextsiz,
822     struct thread *p, struct mbuf **mpp, struct mbuf **mpendp, int *lenp)
823 {
824 	struct iovec *iv;
825 	struct uio io, *uiop = &io;
826 	struct mbuf *mp, *mp3;
827 	int len, tlen, error = 0;
828 
829 	len = NFS_MAXPATHLEN;
830 	if (maxextsiz > 0)
831 		uiop->uio_iovcnt = nfsrv_createiovec_extpgs(len, maxextsiz,
832 		    &mp3, &mp, &iv);
833 	else
834 		uiop->uio_iovcnt = nfsrv_createiovec(len, &mp3, &mp, &iv);
835 	uiop->uio_iov = iv;
836 	uiop->uio_offset = 0;
837 	uiop->uio_resid = len;
838 	uiop->uio_rw = UIO_READ;
839 	uiop->uio_segflg = UIO_SYSSPACE;
840 	uiop->uio_td = NULL;
841 	error = VOP_READLINK(vp, uiop, cred);
842 	free(iv, M_TEMP);
843 	if (error) {
844 		m_freem(mp3);
845 		*lenp = 0;
846 		goto out;
847 	}
848 	if (uiop->uio_resid > 0) {
849 		len -= uiop->uio_resid;
850 		tlen = NFSM_RNDUP(len);
851 		if (tlen == 0) {
852 			m_freem(mp3);
853 			mp3 = mp = NULL;
854 		} else if (tlen != NFS_MAXPATHLEN || tlen != len)
855 			mp = nfsrv_adj(mp3, NFS_MAXPATHLEN - tlen,
856 			    tlen - len);
857 	}
858 	*lenp = len;
859 	*mpp = mp3;
860 	*mpendp = mp;
861 
862 out:
863 	NFSEXITCODE(error);
864 	return (error);
865 }
866 
867 /*
868  * Create an mbuf chain and an associated iovec that can be used to Read
869  * or Getextattr of data.
870  * Upon success, return pointers to the first and last mbufs in the chain
871  * plus the malloc'd iovec and its iovlen.
872  */
873 static int
874 nfsrv_createiovec(int len, struct mbuf **mpp, struct mbuf **mpendp,
875     struct iovec **ivp)
876 {
877 	struct mbuf *m, *m2 = NULL, *m3;
878 	struct iovec *iv;
879 	int i, left, siz;
880 
881 	left = len;
882 	m3 = NULL;
883 	/*
884 	 * Generate the mbuf list with the uio_iov ref. to it.
885 	 */
886 	i = 0;
887 	while (left > 0) {
888 		NFSMGET(m);
889 		MCLGET(m, M_WAITOK);
890 		m->m_len = 0;
891 		siz = min(M_TRAILINGSPACE(m), left);
892 		left -= siz;
893 		i++;
894 		if (m3)
895 			m2->m_next = m;
896 		else
897 			m3 = m;
898 		m2 = m;
899 	}
900 	*ivp = iv = malloc(i * sizeof (struct iovec), M_TEMP, M_WAITOK);
901 	m = m3;
902 	left = len;
903 	i = 0;
904 	while (left > 0) {
905 		if (m == NULL)
906 			panic("nfsrv_createiovec iov");
907 		siz = min(M_TRAILINGSPACE(m), left);
908 		if (siz > 0) {
909 			iv->iov_base = mtod(m, caddr_t) + m->m_len;
910 			iv->iov_len = siz;
911 			m->m_len += siz;
912 			left -= siz;
913 			iv++;
914 			i++;
915 		}
916 		m = m->m_next;
917 	}
918 	*mpp = m3;
919 	*mpendp = m2;
920 	return (i);
921 }
922 
923 /*
924  * Create an mbuf chain and an associated iovec that can be used to Read
925  * or Getextattr of data.
926  * Upon success, return pointers to the first and last mbufs in the chain
927  * plus the malloc'd iovec and its iovlen.
928  * Same as above, but creates ext_pgs mbuf(s).
929  */
930 static int
931 nfsrv_createiovec_extpgs(int len, int maxextsiz, struct mbuf **mpp,
932     struct mbuf **mpendp, struct iovec **ivp)
933 {
934 	struct mbuf *m, *m2 = NULL, *m3;
935 	struct iovec *iv;
936 	int i, left, pgno, siz;
937 
938 	left = len;
939 	m3 = NULL;
940 	/*
941 	 * Generate the mbuf list with the uio_iov ref. to it.
942 	 */
943 	i = 0;
944 	while (left > 0) {
945 		siz = min(left, maxextsiz);
946 		m = mb_alloc_ext_plus_pages(siz, M_WAITOK);
947 		left -= siz;
948 		i += m->m_epg_npgs;
949 		if (m3 != NULL)
950 			m2->m_next = m;
951 		else
952 			m3 = m;
953 		m2 = m;
954 	}
955 	*ivp = iv = malloc(i * sizeof (struct iovec), M_TEMP, M_WAITOK);
956 	m = m3;
957 	left = len;
958 	i = 0;
959 	pgno = 0;
960 	while (left > 0) {
961 		if (m == NULL)
962 			panic("nfsvno_createiovec_extpgs iov");
963 		siz = min(PAGE_SIZE, left);
964 		if (siz > 0) {
965 			iv->iov_base = (void *)PHYS_TO_DMAP(m->m_epg_pa[pgno]);
966 			iv->iov_len = siz;
967 			m->m_len += siz;
968 			if (pgno == m->m_epg_npgs - 1)
969 				m->m_epg_last_len = siz;
970 			left -= siz;
971 			iv++;
972 			i++;
973 			pgno++;
974 		}
975 		if (pgno == m->m_epg_npgs && left > 0) {
976 			m = m->m_next;
977 			if (m == NULL)
978 				panic("nfsvno_createiovec_extpgs iov");
979 			pgno = 0;
980 		}
981 	}
982 	*mpp = m3;
983 	*mpendp = m2;
984 	return (i);
985 }
986 
987 /*
988  * Read vnode op call into mbuf list.
989  */
990 int
991 nfsvno_read(struct vnode *vp, off_t off, int cnt, struct ucred *cred,
992     int maxextsiz, struct thread *p, struct mbuf **mpp,
993     struct mbuf **mpendp)
994 {
995 	struct mbuf *m;
996 	struct iovec *iv;
997 	int error = 0, len, tlen, ioflag = 0;
998 	struct mbuf *m3;
999 	struct uio io, *uiop = &io;
1000 	struct nfsheur *nh;
1001 
1002 	/*
1003 	 * Attempt to read from a DS file. A return of ENOENT implies
1004 	 * there is no DS file to read.
1005 	 */
1006 	error = nfsrv_proxyds(vp, off, cnt, cred, p, NFSPROC_READDS, mpp,
1007 	    NULL, mpendp, NULL, NULL, NULL, 0, NULL);
1008 	if (error != ENOENT)
1009 		return (error);
1010 
1011 	len = NFSM_RNDUP(cnt);
1012 	if (maxextsiz > 0)
1013 		uiop->uio_iovcnt = nfsrv_createiovec_extpgs(len, maxextsiz,
1014 		    &m3, &m, &iv);
1015 	else
1016 		uiop->uio_iovcnt = nfsrv_createiovec(len, &m3, &m, &iv);
1017 	uiop->uio_iov = iv;
1018 	uiop->uio_offset = off;
1019 	uiop->uio_resid = len;
1020 	uiop->uio_rw = UIO_READ;
1021 	uiop->uio_segflg = UIO_SYSSPACE;
1022 	uiop->uio_td = NULL;
1023 	nh = nfsrv_sequential_heuristic(uiop, vp);
1024 	ioflag |= nh->nh_seqcount << IO_SEQSHIFT;
1025 	/* XXX KDM make this more systematic? */
1026 	nfsstatsv1.srvbytes[NFSV4OP_READ] += uiop->uio_resid;
1027 	error = VOP_READ(vp, uiop, IO_NODELOCKED | ioflag, cred);
1028 	free(iv, M_TEMP);
1029 	if (error) {
1030 		m_freem(m3);
1031 		*mpp = NULL;
1032 		goto out;
1033 	}
1034 	nh->nh_nextoff = uiop->uio_offset;
1035 	tlen = len - uiop->uio_resid;
1036 	cnt = cnt < tlen ? cnt : tlen;
1037 	tlen = NFSM_RNDUP(cnt);
1038 	if (tlen == 0) {
1039 		m_freem(m3);
1040 		m3 = m = NULL;
1041 	} else if (len != tlen || tlen != cnt)
1042 		m = nfsrv_adj(m3, len - tlen, tlen - cnt);
1043 	*mpp = m3;
1044 	*mpendp = m;
1045 
1046 out:
1047 	NFSEXITCODE(error);
1048 	return (error);
1049 }
1050 
1051 /*
1052  * Create the iovec for the mbuf chain passed in as an argument.
1053  * The "cp" argument is where the data starts within the first mbuf in
1054  * the chain. It returns the iovec and the iovcnt.
1055  */
1056 static int
1057 nfsrv_createiovecw(int retlen, struct mbuf *m, char *cp, struct iovec **ivpp,
1058     int *iovcntp)
1059 {
1060 	struct mbuf *mp;
1061 	struct iovec *ivp;
1062 	int cnt, i, len;
1063 
1064 	/*
1065 	 * Loop through the mbuf chain, counting how many mbufs are a
1066 	 * part of this write operation, so the iovec size is known.
1067 	 */
1068 	cnt = 0;
1069 	len = retlen;
1070 	mp = m;
1071 	i = mtod(mp, caddr_t) + mp->m_len - cp;
1072 	while (len > 0) {
1073 		if (i > 0) {
1074 			len -= i;
1075 			cnt++;
1076 		}
1077 		mp = mp->m_next;
1078 		if (!mp) {
1079 			if (len > 0)
1080 				return (EBADRPC);
1081 		} else
1082 			i = mp->m_len;
1083 	}
1084 
1085 	/* Now, create the iovec. */
1086 	mp = m;
1087 	*ivpp = ivp = malloc(cnt * sizeof (struct iovec), M_TEMP,
1088 	    M_WAITOK);
1089 	*iovcntp = cnt;
1090 	i = mtod(mp, caddr_t) + mp->m_len - cp;
1091 	len = retlen;
1092 	while (len > 0) {
1093 		if (mp == NULL)
1094 			panic("nfsrv_createiovecw");
1095 		if (i > 0) {
1096 			i = min(i, len);
1097 			ivp->iov_base = cp;
1098 			ivp->iov_len = i;
1099 			ivp++;
1100 			len -= i;
1101 		}
1102 		mp = mp->m_next;
1103 		if (mp) {
1104 			i = mp->m_len;
1105 			cp = mtod(mp, caddr_t);
1106 		}
1107 	}
1108 	return (0);
1109 }
1110 
1111 /*
1112  * Write vnode op from an mbuf list.
1113  */
1114 int
1115 nfsvno_write(struct vnode *vp, off_t off, int retlen, int *stable,
1116     struct mbuf *mp, char *cp, struct ucred *cred, struct thread *p)
1117 {
1118 	struct iovec *iv;
1119 	int cnt, ioflags, error;
1120 	struct uio io, *uiop = &io;
1121 	struct nfsheur *nh;
1122 
1123 	/*
1124 	 * Attempt to write to a DS file. A return of ENOENT implies
1125 	 * there is no DS file to write.
1126 	 */
1127 	error = nfsrv_proxyds(vp, off, retlen, cred, p, NFSPROC_WRITEDS,
1128 	    &mp, cp, NULL, NULL, NULL, NULL, 0, NULL);
1129 	if (error != ENOENT) {
1130 		*stable = NFSWRITE_FILESYNC;
1131 		return (error);
1132 	}
1133 
1134 	if (*stable == NFSWRITE_UNSTABLE)
1135 		ioflags = IO_NODELOCKED;
1136 	else
1137 		ioflags = (IO_SYNC | IO_NODELOCKED);
1138 	error = nfsrv_createiovecw(retlen, mp, cp, &iv, &cnt);
1139 	if (error != 0)
1140 		return (error);
1141 	uiop->uio_iov = iv;
1142 	uiop->uio_iovcnt = cnt;
1143 	uiop->uio_resid = retlen;
1144 	uiop->uio_rw = UIO_WRITE;
1145 	uiop->uio_segflg = UIO_SYSSPACE;
1146 	NFSUIOPROC(uiop, p);
1147 	uiop->uio_offset = off;
1148 	nh = nfsrv_sequential_heuristic(uiop, vp);
1149 	ioflags |= nh->nh_seqcount << IO_SEQSHIFT;
1150 	/* XXX KDM make this more systematic? */
1151 	nfsstatsv1.srvbytes[NFSV4OP_WRITE] += uiop->uio_resid;
1152 	error = VOP_WRITE(vp, uiop, ioflags, cred);
1153 	if (error == 0)
1154 		nh->nh_nextoff = uiop->uio_offset;
1155 	free(iv, M_TEMP);
1156 
1157 	NFSEXITCODE(error);
1158 	return (error);
1159 }
1160 
1161 /*
1162  * Common code for creating a regular file (plus special files for V2).
1163  */
1164 int
1165 nfsvno_createsub(struct nfsrv_descript *nd, struct nameidata *ndp,
1166     struct vnode **vpp, struct nfsvattr *nvap, int *exclusive_flagp,
1167     int32_t *cverf, NFSDEV_T rdev, struct nfsexstuff *exp)
1168 {
1169 	u_quad_t tempsize;
1170 	int error;
1171 	struct thread *p = curthread;
1172 
1173 	error = nd->nd_repstat;
1174 	if (!error && ndp->ni_vp == NULL) {
1175 		if (nvap->na_type == VREG || nvap->na_type == VSOCK) {
1176 			vrele(ndp->ni_startdir);
1177 			error = VOP_CREATE(ndp->ni_dvp,
1178 			    &ndp->ni_vp, &ndp->ni_cnd, &nvap->na_vattr);
1179 			/* For a pNFS server, create the data file on a DS. */
1180 			if (error == 0 && nvap->na_type == VREG) {
1181 				/*
1182 				 * Create a data file on a DS for a pNFS server.
1183 				 * This function just returns if not
1184 				 * running a pNFS DS or the creation fails.
1185 				 */
1186 				nfsrv_pnfscreate(ndp->ni_vp, &nvap->na_vattr,
1187 				    nd->nd_cred, p);
1188 			}
1189 			VOP_VPUT_PAIR(ndp->ni_dvp, error == 0 ? &ndp->ni_vp :
1190 			    NULL, false);
1191 			nfsvno_relpathbuf(ndp);
1192 			if (!error) {
1193 				if (*exclusive_flagp) {
1194 					*exclusive_flagp = 0;
1195 					NFSVNO_ATTRINIT(nvap);
1196 					nvap->na_atime.tv_sec = cverf[0];
1197 					nvap->na_atime.tv_nsec = cverf[1];
1198 					error = VOP_SETATTR(ndp->ni_vp,
1199 					    &nvap->na_vattr, nd->nd_cred);
1200 					if (error != 0) {
1201 						vput(ndp->ni_vp);
1202 						ndp->ni_vp = NULL;
1203 						error = NFSERR_NOTSUPP;
1204 					}
1205 				}
1206 			}
1207 		/*
1208 		 * NFS V2 Only. nfsrvd_mknod() does this for V3.
1209 		 * (This implies, just get out on an error.)
1210 		 */
1211 		} else if (nvap->na_type == VCHR || nvap->na_type == VBLK ||
1212 			nvap->na_type == VFIFO) {
1213 			if (nvap->na_type == VCHR && rdev == 0xffffffff)
1214 				nvap->na_type = VFIFO;
1215                         if (nvap->na_type != VFIFO &&
1216 			    (error = priv_check_cred(nd->nd_cred, PRIV_VFS_MKNOD_DEV))) {
1217 				vrele(ndp->ni_startdir);
1218 				nfsvno_relpathbuf(ndp);
1219 				vput(ndp->ni_dvp);
1220 				goto out;
1221 			}
1222 			nvap->na_rdev = rdev;
1223 			error = VOP_MKNOD(ndp->ni_dvp, &ndp->ni_vp,
1224 			    &ndp->ni_cnd, &nvap->na_vattr);
1225 			VOP_VPUT_PAIR(ndp->ni_dvp, error == 0 ? &ndp->ni_vp :
1226 			    NULL, false);
1227 			nfsvno_relpathbuf(ndp);
1228 			vrele(ndp->ni_startdir);
1229 			if (error)
1230 				goto out;
1231 		} else {
1232 			vrele(ndp->ni_startdir);
1233 			nfsvno_relpathbuf(ndp);
1234 			vput(ndp->ni_dvp);
1235 			error = ENXIO;
1236 			goto out;
1237 		}
1238 		*vpp = ndp->ni_vp;
1239 	} else {
1240 		/*
1241 		 * Handle cases where error is already set and/or
1242 		 * the file exists.
1243 		 * 1 - clean up the lookup
1244 		 * 2 - iff !error and na_size set, truncate it
1245 		 */
1246 		vrele(ndp->ni_startdir);
1247 		nfsvno_relpathbuf(ndp);
1248 		*vpp = ndp->ni_vp;
1249 		if (ndp->ni_dvp == *vpp)
1250 			vrele(ndp->ni_dvp);
1251 		else
1252 			vput(ndp->ni_dvp);
1253 		if (!error && nvap->na_size != VNOVAL) {
1254 			error = nfsvno_accchk(*vpp, VWRITE,
1255 			    nd->nd_cred, exp, p, NFSACCCHK_NOOVERRIDE,
1256 			    NFSACCCHK_VPISLOCKED, NULL);
1257 			if (!error) {
1258 				tempsize = nvap->na_size;
1259 				NFSVNO_ATTRINIT(nvap);
1260 				nvap->na_size = tempsize;
1261 				error = VOP_SETATTR(*vpp,
1262 				    &nvap->na_vattr, nd->nd_cred);
1263 			}
1264 		}
1265 		if (error)
1266 			vput(*vpp);
1267 	}
1268 
1269 out:
1270 	NFSEXITCODE(error);
1271 	return (error);
1272 }
1273 
1274 /*
1275  * Do a mknod vnode op.
1276  */
1277 int
1278 nfsvno_mknod(struct nameidata *ndp, struct nfsvattr *nvap, struct ucred *cred,
1279     struct thread *p)
1280 {
1281 	int error = 0;
1282 	enum vtype vtyp;
1283 
1284 	vtyp = nvap->na_type;
1285 	/*
1286 	 * Iff doesn't exist, create it.
1287 	 */
1288 	if (ndp->ni_vp) {
1289 		vrele(ndp->ni_startdir);
1290 		nfsvno_relpathbuf(ndp);
1291 		vput(ndp->ni_dvp);
1292 		vrele(ndp->ni_vp);
1293 		error = EEXIST;
1294 		goto out;
1295 	}
1296 	if (vtyp != VCHR && vtyp != VBLK && vtyp != VSOCK && vtyp != VFIFO) {
1297 		vrele(ndp->ni_startdir);
1298 		nfsvno_relpathbuf(ndp);
1299 		vput(ndp->ni_dvp);
1300 		error = NFSERR_BADTYPE;
1301 		goto out;
1302 	}
1303 	if (vtyp == VSOCK) {
1304 		vrele(ndp->ni_startdir);
1305 		error = VOP_CREATE(ndp->ni_dvp, &ndp->ni_vp,
1306 		    &ndp->ni_cnd, &nvap->na_vattr);
1307 		VOP_VPUT_PAIR(ndp->ni_dvp, error == 0 ? &ndp->ni_vp : NULL,
1308 		    false);
1309 		nfsvno_relpathbuf(ndp);
1310 	} else {
1311 		if (nvap->na_type != VFIFO &&
1312 		    (error = priv_check_cred(cred, PRIV_VFS_MKNOD_DEV))) {
1313 			vrele(ndp->ni_startdir);
1314 			nfsvno_relpathbuf(ndp);
1315 			vput(ndp->ni_dvp);
1316 			goto out;
1317 		}
1318 		error = VOP_MKNOD(ndp->ni_dvp, &ndp->ni_vp,
1319 		    &ndp->ni_cnd, &nvap->na_vattr);
1320 		VOP_VPUT_PAIR(ndp->ni_dvp, error == 0 ? &ndp->ni_vp : NULL,
1321 		    false);
1322 		nfsvno_relpathbuf(ndp);
1323 		vrele(ndp->ni_startdir);
1324 		/*
1325 		 * Since VOP_MKNOD returns the ni_vp, I can't
1326 		 * see any reason to do the lookup.
1327 		 */
1328 	}
1329 
1330 out:
1331 	NFSEXITCODE(error);
1332 	return (error);
1333 }
1334 
1335 /*
1336  * Mkdir vnode op.
1337  */
1338 int
1339 nfsvno_mkdir(struct nameidata *ndp, struct nfsvattr *nvap, uid_t saved_uid,
1340     struct ucred *cred, struct thread *p, struct nfsexstuff *exp)
1341 {
1342 	int error = 0;
1343 
1344 	if (ndp->ni_vp != NULL) {
1345 		if (ndp->ni_dvp == ndp->ni_vp)
1346 			vrele(ndp->ni_dvp);
1347 		else
1348 			vput(ndp->ni_dvp);
1349 		vrele(ndp->ni_vp);
1350 		nfsvno_relpathbuf(ndp);
1351 		error = EEXIST;
1352 		goto out;
1353 	}
1354 	error = VOP_MKDIR(ndp->ni_dvp, &ndp->ni_vp, &ndp->ni_cnd,
1355 	    &nvap->na_vattr);
1356 	VOP_VPUT_PAIR(ndp->ni_dvp, error == 0 ? &ndp->ni_vp : NULL, false);
1357 	nfsvno_relpathbuf(ndp);
1358 
1359 out:
1360 	NFSEXITCODE(error);
1361 	return (error);
1362 }
1363 
1364 /*
1365  * symlink vnode op.
1366  */
1367 int
1368 nfsvno_symlink(struct nameidata *ndp, struct nfsvattr *nvap, char *pathcp,
1369     int pathlen, int not_v2, uid_t saved_uid, struct ucred *cred, struct thread *p,
1370     struct nfsexstuff *exp)
1371 {
1372 	int error = 0;
1373 
1374 	if (ndp->ni_vp) {
1375 		vrele(ndp->ni_startdir);
1376 		nfsvno_relpathbuf(ndp);
1377 		if (ndp->ni_dvp == ndp->ni_vp)
1378 			vrele(ndp->ni_dvp);
1379 		else
1380 			vput(ndp->ni_dvp);
1381 		vrele(ndp->ni_vp);
1382 		error = EEXIST;
1383 		goto out;
1384 	}
1385 
1386 	error = VOP_SYMLINK(ndp->ni_dvp, &ndp->ni_vp, &ndp->ni_cnd,
1387 	    &nvap->na_vattr, pathcp);
1388 	/*
1389 	 * Although FreeBSD still had the lookup code in
1390 	 * it for 7/current, there doesn't seem to be any
1391 	 * point, since VOP_SYMLINK() returns the ni_vp.
1392 	 * Just vput it for v2.
1393 	 */
1394 	VOP_VPUT_PAIR(ndp->ni_dvp, &ndp->ni_vp, !not_v2 && error == 0);
1395 	vrele(ndp->ni_startdir);
1396 	nfsvno_relpathbuf(ndp);
1397 
1398 out:
1399 	NFSEXITCODE(error);
1400 	return (error);
1401 }
1402 
1403 /*
1404  * Parse symbolic link arguments.
1405  * This function has an ugly side effect. It will malloc() an area for
1406  * the symlink and set iov_base to point to it, only if it succeeds.
1407  * So, if it returns with uiop->uio_iov->iov_base != NULL, that must
1408  * be FREE'd later.
1409  */
1410 int
1411 nfsvno_getsymlink(struct nfsrv_descript *nd, struct nfsvattr *nvap,
1412     struct thread *p, char **pathcpp, int *lenp)
1413 {
1414 	u_int32_t *tl;
1415 	char *pathcp = NULL;
1416 	int error = 0, len;
1417 	struct nfsv2_sattr *sp;
1418 
1419 	*pathcpp = NULL;
1420 	*lenp = 0;
1421 	if ((nd->nd_flag & ND_NFSV3) &&
1422 	    (error = nfsrv_sattr(nd, NULL, nvap, NULL, NULL, p)))
1423 		goto nfsmout;
1424 	NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
1425 	len = fxdr_unsigned(int, *tl);
1426 	if (len > NFS_MAXPATHLEN || len <= 0) {
1427 		error = EBADRPC;
1428 		goto nfsmout;
1429 	}
1430 	pathcp = malloc(len + 1, M_TEMP, M_WAITOK);
1431 	error = nfsrv_mtostr(nd, pathcp, len);
1432 	if (error)
1433 		goto nfsmout;
1434 	if (nd->nd_flag & ND_NFSV2) {
1435 		NFSM_DISSECT(sp, struct nfsv2_sattr *, NFSX_V2SATTR);
1436 		nvap->na_mode = fxdr_unsigned(u_int16_t, sp->sa_mode);
1437 	}
1438 	*pathcpp = pathcp;
1439 	*lenp = len;
1440 	NFSEXITCODE2(0, nd);
1441 	return (0);
1442 nfsmout:
1443 	if (pathcp)
1444 		free(pathcp, M_TEMP);
1445 	NFSEXITCODE2(error, nd);
1446 	return (error);
1447 }
1448 
1449 /*
1450  * Remove a non-directory object.
1451  */
1452 int
1453 nfsvno_removesub(struct nameidata *ndp, int is_v4, struct ucred *cred,
1454     struct thread *p, struct nfsexstuff *exp)
1455 {
1456 	struct vnode *vp, *dsdvp[NFSDEV_MAXMIRRORS];
1457 	int error = 0, mirrorcnt;
1458 	char fname[PNFS_FILENAME_LEN + 1];
1459 	fhandle_t fh;
1460 
1461 	vp = ndp->ni_vp;
1462 	dsdvp[0] = NULL;
1463 	if (vp->v_type == VDIR)
1464 		error = NFSERR_ISDIR;
1465 	else if (is_v4)
1466 		error = nfsrv_checkremove(vp, 1, NULL, (nfsquad_t)((u_quad_t)0),
1467 		    p);
1468 	if (error == 0)
1469 		nfsrv_pnfsremovesetup(vp, p, dsdvp, &mirrorcnt, fname, &fh);
1470 	if (!error)
1471 		error = VOP_REMOVE(ndp->ni_dvp, vp, &ndp->ni_cnd);
1472 	if (error == 0 && dsdvp[0] != NULL)
1473 		nfsrv_pnfsremove(dsdvp, mirrorcnt, fname, &fh, p);
1474 	if (ndp->ni_dvp == vp)
1475 		vrele(ndp->ni_dvp);
1476 	else
1477 		vput(ndp->ni_dvp);
1478 	vput(vp);
1479 	if ((ndp->ni_cnd.cn_flags & SAVENAME) != 0)
1480 		nfsvno_relpathbuf(ndp);
1481 	NFSEXITCODE(error);
1482 	return (error);
1483 }
1484 
1485 /*
1486  * Remove a directory.
1487  */
1488 int
1489 nfsvno_rmdirsub(struct nameidata *ndp, int is_v4, struct ucred *cred,
1490     struct thread *p, struct nfsexstuff *exp)
1491 {
1492 	struct vnode *vp;
1493 	int error = 0;
1494 
1495 	vp = ndp->ni_vp;
1496 	if (vp->v_type != VDIR) {
1497 		error = ENOTDIR;
1498 		goto out;
1499 	}
1500 	/*
1501 	 * No rmdir "." please.
1502 	 */
1503 	if (ndp->ni_dvp == vp) {
1504 		error = EINVAL;
1505 		goto out;
1506 	}
1507 	/*
1508 	 * The root of a mounted filesystem cannot be deleted.
1509 	 */
1510 	if (vp->v_vflag & VV_ROOT)
1511 		error = EBUSY;
1512 out:
1513 	if (!error)
1514 		error = VOP_RMDIR(ndp->ni_dvp, vp, &ndp->ni_cnd);
1515 	if (ndp->ni_dvp == vp)
1516 		vrele(ndp->ni_dvp);
1517 	else
1518 		vput(ndp->ni_dvp);
1519 	vput(vp);
1520 	if ((ndp->ni_cnd.cn_flags & SAVENAME) != 0)
1521 		nfsvno_relpathbuf(ndp);
1522 	NFSEXITCODE(error);
1523 	return (error);
1524 }
1525 
1526 /*
1527  * Rename vnode op.
1528  */
1529 int
1530 nfsvno_rename(struct nameidata *fromndp, struct nameidata *tondp,
1531     u_int32_t ndstat, u_int32_t ndflag, struct ucred *cred, struct thread *p)
1532 {
1533 	struct vnode *fvp, *tvp, *tdvp, *dsdvp[NFSDEV_MAXMIRRORS];
1534 	int error = 0, mirrorcnt;
1535 	char fname[PNFS_FILENAME_LEN + 1];
1536 	fhandle_t fh;
1537 
1538 	dsdvp[0] = NULL;
1539 	fvp = fromndp->ni_vp;
1540 	if (ndstat) {
1541 		vrele(fromndp->ni_dvp);
1542 		vrele(fvp);
1543 		error = ndstat;
1544 		goto out1;
1545 	}
1546 	tdvp = tondp->ni_dvp;
1547 	tvp = tondp->ni_vp;
1548 	if (tvp != NULL) {
1549 		if (fvp->v_type == VDIR && tvp->v_type != VDIR) {
1550 			error = (ndflag & ND_NFSV2) ? EISDIR : EEXIST;
1551 			goto out;
1552 		} else if (fvp->v_type != VDIR && tvp->v_type == VDIR) {
1553 			error = (ndflag & ND_NFSV2) ? ENOTDIR : EEXIST;
1554 			goto out;
1555 		}
1556 		if (tvp->v_type == VDIR && tvp->v_mountedhere) {
1557 			error = (ndflag & ND_NFSV2) ? ENOTEMPTY : EXDEV;
1558 			goto out;
1559 		}
1560 
1561 		/*
1562 		 * A rename to '.' or '..' results in a prematurely
1563 		 * unlocked vnode on FreeBSD5, so I'm just going to fail that
1564 		 * here.
1565 		 */
1566 		if ((tondp->ni_cnd.cn_namelen == 1 &&
1567 		     tondp->ni_cnd.cn_nameptr[0] == '.') ||
1568 		    (tondp->ni_cnd.cn_namelen == 2 &&
1569 		     tondp->ni_cnd.cn_nameptr[0] == '.' &&
1570 		     tondp->ni_cnd.cn_nameptr[1] == '.')) {
1571 			error = EINVAL;
1572 			goto out;
1573 		}
1574 	}
1575 	if (fvp->v_type == VDIR && fvp->v_mountedhere) {
1576 		error = (ndflag & ND_NFSV2) ? ENOTEMPTY : EXDEV;
1577 		goto out;
1578 	}
1579 	if (fvp->v_mount != tdvp->v_mount) {
1580 		error = (ndflag & ND_NFSV2) ? ENOTEMPTY : EXDEV;
1581 		goto out;
1582 	}
1583 	if (fvp == tdvp) {
1584 		error = (ndflag & ND_NFSV2) ? ENOTEMPTY : EINVAL;
1585 		goto out;
1586 	}
1587 	if (fvp == tvp) {
1588 		/*
1589 		 * If source and destination are the same, there is nothing to
1590 		 * do. Set error to -1 to indicate this.
1591 		 */
1592 		error = -1;
1593 		goto out;
1594 	}
1595 	if (ndflag & ND_NFSV4) {
1596 		if (NFSVOPLOCK(fvp, LK_EXCLUSIVE) == 0) {
1597 			error = nfsrv_checkremove(fvp, 0, NULL,
1598 			    (nfsquad_t)((u_quad_t)0), p);
1599 			NFSVOPUNLOCK(fvp);
1600 		} else
1601 			error = EPERM;
1602 		if (tvp && !error)
1603 			error = nfsrv_checkremove(tvp, 1, NULL,
1604 			    (nfsquad_t)((u_quad_t)0), p);
1605 	} else {
1606 		/*
1607 		 * For NFSv2 and NFSv3, try to get rid of the delegation, so
1608 		 * that the NFSv4 client won't be confused by the rename.
1609 		 * Since nfsd_recalldelegation() can only be called on an
1610 		 * unlocked vnode at this point and fvp is the file that will
1611 		 * still exist after the rename, just do fvp.
1612 		 */
1613 		nfsd_recalldelegation(fvp, p);
1614 	}
1615 	if (error == 0 && tvp != NULL) {
1616 		nfsrv_pnfsremovesetup(tvp, p, dsdvp, &mirrorcnt, fname, &fh);
1617 		NFSD_DEBUG(4, "nfsvno_rename: pnfsremovesetup"
1618 		    " dsdvp=%p\n", dsdvp[0]);
1619 	}
1620 out:
1621 	if (!error) {
1622 		error = VOP_RENAME(fromndp->ni_dvp, fromndp->ni_vp,
1623 		    &fromndp->ni_cnd, tondp->ni_dvp, tondp->ni_vp,
1624 		    &tondp->ni_cnd);
1625 	} else {
1626 		if (tdvp == tvp)
1627 			vrele(tdvp);
1628 		else
1629 			vput(tdvp);
1630 		if (tvp)
1631 			vput(tvp);
1632 		vrele(fromndp->ni_dvp);
1633 		vrele(fvp);
1634 		if (error == -1)
1635 			error = 0;
1636 	}
1637 
1638 	/*
1639 	 * If dsdvp[0] != NULL, it was set up by nfsrv_pnfsremovesetup() and
1640 	 * if the rename succeeded, the DS file for the tvp needs to be
1641 	 * removed.
1642 	 */
1643 	if (error == 0 && dsdvp[0] != NULL) {
1644 		nfsrv_pnfsremove(dsdvp, mirrorcnt, fname, &fh, p);
1645 		NFSD_DEBUG(4, "nfsvno_rename: pnfsremove\n");
1646 	}
1647 
1648 	vrele(tondp->ni_startdir);
1649 	nfsvno_relpathbuf(tondp);
1650 out1:
1651 	vrele(fromndp->ni_startdir);
1652 	nfsvno_relpathbuf(fromndp);
1653 	NFSEXITCODE(error);
1654 	return (error);
1655 }
1656 
1657 /*
1658  * Link vnode op.
1659  */
1660 int
1661 nfsvno_link(struct nameidata *ndp, struct vnode *vp, struct ucred *cred,
1662     struct thread *p, struct nfsexstuff *exp)
1663 {
1664 	struct vnode *xp;
1665 	int error = 0;
1666 
1667 	xp = ndp->ni_vp;
1668 	if (xp != NULL) {
1669 		error = EEXIST;
1670 	} else {
1671 		xp = ndp->ni_dvp;
1672 		if (vp->v_mount != xp->v_mount)
1673 			error = EXDEV;
1674 	}
1675 	if (!error) {
1676 		NFSVOPLOCK(vp, LK_EXCLUSIVE | LK_RETRY);
1677 		if (!VN_IS_DOOMED(vp))
1678 			error = VOP_LINK(ndp->ni_dvp, vp, &ndp->ni_cnd);
1679 		else
1680 			error = EPERM;
1681 		if (ndp->ni_dvp == vp) {
1682 			vrele(ndp->ni_dvp);
1683 			NFSVOPUNLOCK(vp);
1684 		} else {
1685 			vref(vp);
1686 			VOP_VPUT_PAIR(ndp->ni_dvp, &vp, true);
1687 		}
1688 	} else {
1689 		if (ndp->ni_dvp == ndp->ni_vp)
1690 			vrele(ndp->ni_dvp);
1691 		else
1692 			vput(ndp->ni_dvp);
1693 		if (ndp->ni_vp)
1694 			vrele(ndp->ni_vp);
1695 	}
1696 	nfsvno_relpathbuf(ndp);
1697 	NFSEXITCODE(error);
1698 	return (error);
1699 }
1700 
1701 /*
1702  * Do the fsync() appropriate for the commit.
1703  */
1704 int
1705 nfsvno_fsync(struct vnode *vp, u_int64_t off, int cnt, struct ucred *cred,
1706     struct thread *td)
1707 {
1708 	int error = 0;
1709 
1710 	/*
1711 	 * RFC 1813 3.3.21: if count is 0, a flush from offset to the end of
1712 	 * file is done.  At this time VOP_FSYNC does not accept offset and
1713 	 * byte count parameters so call VOP_FSYNC the whole file for now.
1714 	 * The same is true for NFSv4: RFC 3530 Sec. 14.2.3.
1715 	 * File systems that do not use the buffer cache (as indicated
1716 	 * by MNTK_USES_BCACHE not being set) must use VOP_FSYNC().
1717 	 */
1718 	if (cnt == 0 || cnt > MAX_COMMIT_COUNT ||
1719 	    (vp->v_mount->mnt_kern_flag & MNTK_USES_BCACHE) == 0) {
1720 		/*
1721 		 * Give up and do the whole thing
1722 		 */
1723 		if (vp->v_object && vm_object_mightbedirty(vp->v_object)) {
1724 			VM_OBJECT_WLOCK(vp->v_object);
1725 			vm_object_page_clean(vp->v_object, 0, 0, OBJPC_SYNC);
1726 			VM_OBJECT_WUNLOCK(vp->v_object);
1727 		}
1728 		error = VOP_FSYNC(vp, MNT_WAIT, td);
1729 	} else {
1730 		/*
1731 		 * Locate and synchronously write any buffers that fall
1732 		 * into the requested range.  Note:  we are assuming that
1733 		 * f_iosize is a power of 2.
1734 		 */
1735 		int iosize = vp->v_mount->mnt_stat.f_iosize;
1736 		int iomask = iosize - 1;
1737 		struct bufobj *bo;
1738 		daddr_t lblkno;
1739 
1740 		/*
1741 		 * Align to iosize boundary, super-align to page boundary.
1742 		 */
1743 		if (off & iomask) {
1744 			cnt += off & iomask;
1745 			off &= ~(u_quad_t)iomask;
1746 		}
1747 		if (off & PAGE_MASK) {
1748 			cnt += off & PAGE_MASK;
1749 			off &= ~(u_quad_t)PAGE_MASK;
1750 		}
1751 		lblkno = off / iosize;
1752 
1753 		if (vp->v_object && vm_object_mightbedirty(vp->v_object)) {
1754 			VM_OBJECT_WLOCK(vp->v_object);
1755 			vm_object_page_clean(vp->v_object, off, off + cnt,
1756 			    OBJPC_SYNC);
1757 			VM_OBJECT_WUNLOCK(vp->v_object);
1758 		}
1759 
1760 		bo = &vp->v_bufobj;
1761 		BO_LOCK(bo);
1762 		while (cnt > 0) {
1763 			struct buf *bp;
1764 
1765 			/*
1766 			 * If we have a buffer and it is marked B_DELWRI we
1767 			 * have to lock and write it.  Otherwise the prior
1768 			 * write is assumed to have already been committed.
1769 			 *
1770 			 * gbincore() can return invalid buffers now so we
1771 			 * have to check that bit as well (though B_DELWRI
1772 			 * should not be set if B_INVAL is set there could be
1773 			 * a race here since we haven't locked the buffer).
1774 			 */
1775 			if ((bp = gbincore(&vp->v_bufobj, lblkno)) != NULL) {
1776 				if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_SLEEPFAIL |
1777 				    LK_INTERLOCK, BO_LOCKPTR(bo)) == ENOLCK) {
1778 					BO_LOCK(bo);
1779 					continue; /* retry */
1780 				}
1781 			    	if ((bp->b_flags & (B_DELWRI|B_INVAL)) ==
1782 				    B_DELWRI) {
1783 					bremfree(bp);
1784 					bp->b_flags &= ~B_ASYNC;
1785 					bwrite(bp);
1786 					++nfs_commit_miss;
1787 				} else
1788 					BUF_UNLOCK(bp);
1789 				BO_LOCK(bo);
1790 			}
1791 			++nfs_commit_blks;
1792 			if (cnt < iosize)
1793 				break;
1794 			cnt -= iosize;
1795 			++lblkno;
1796 		}
1797 		BO_UNLOCK(bo);
1798 	}
1799 	NFSEXITCODE(error);
1800 	return (error);
1801 }
1802 
1803 /*
1804  * Statfs vnode op.
1805  */
1806 int
1807 nfsvno_statfs(struct vnode *vp, struct statfs *sf)
1808 {
1809 	struct statfs *tsf;
1810 	int error;
1811 
1812 	tsf = NULL;
1813 	if (nfsrv_devidcnt > 0) {
1814 		/* For a pNFS service, get the DS numbers. */
1815 		tsf = malloc(sizeof(*tsf), M_TEMP, M_WAITOK | M_ZERO);
1816 		error = nfsrv_pnfsstatfs(tsf, vp->v_mount);
1817 		if (error != 0) {
1818 			free(tsf, M_TEMP);
1819 			tsf = NULL;
1820 		}
1821 	}
1822 	error = VFS_STATFS(vp->v_mount, sf);
1823 	if (error == 0) {
1824 		if (tsf != NULL) {
1825 			sf->f_blocks = tsf->f_blocks;
1826 			sf->f_bavail = tsf->f_bavail;
1827 			sf->f_bfree = tsf->f_bfree;
1828 			sf->f_bsize = tsf->f_bsize;
1829 		}
1830 		/*
1831 		 * Since NFS handles these values as unsigned on the
1832 		 * wire, there is no way to represent negative values,
1833 		 * so set them to 0. Without this, they will appear
1834 		 * to be very large positive values for clients like
1835 		 * Solaris10.
1836 		 */
1837 		if (sf->f_bavail < 0)
1838 			sf->f_bavail = 0;
1839 		if (sf->f_ffree < 0)
1840 			sf->f_ffree = 0;
1841 	}
1842 	free(tsf, M_TEMP);
1843 	NFSEXITCODE(error);
1844 	return (error);
1845 }
1846 
1847 /*
1848  * Do the vnode op stuff for Open. Similar to nfsvno_createsub(), but
1849  * must handle nfsrv_opencheck() calls after any other access checks.
1850  */
1851 void
1852 nfsvno_open(struct nfsrv_descript *nd, struct nameidata *ndp,
1853     nfsquad_t clientid, nfsv4stateid_t *stateidp, struct nfsstate *stp,
1854     int *exclusive_flagp, struct nfsvattr *nvap, int32_t *cverf, int create,
1855     NFSACL_T *aclp, nfsattrbit_t *attrbitp, struct ucred *cred,
1856     struct nfsexstuff *exp, struct vnode **vpp)
1857 {
1858 	struct vnode *vp = NULL;
1859 	u_quad_t tempsize;
1860 	struct nfsexstuff nes;
1861 	struct thread *p = curthread;
1862 
1863 	if (ndp->ni_vp == NULL)
1864 		nd->nd_repstat = nfsrv_opencheck(clientid,
1865 		    stateidp, stp, NULL, nd, p, nd->nd_repstat);
1866 	if (!nd->nd_repstat) {
1867 		if (ndp->ni_vp == NULL) {
1868 			vrele(ndp->ni_startdir);
1869 			nd->nd_repstat = VOP_CREATE(ndp->ni_dvp,
1870 			    &ndp->ni_vp, &ndp->ni_cnd, &nvap->na_vattr);
1871 			/* For a pNFS server, create the data file on a DS. */
1872 			if (nd->nd_repstat == 0) {
1873 				/*
1874 				 * Create a data file on a DS for a pNFS server.
1875 				 * This function just returns if not
1876 				 * running a pNFS DS or the creation fails.
1877 				 */
1878 				nfsrv_pnfscreate(ndp->ni_vp, &nvap->na_vattr,
1879 				    cred, p);
1880 			}
1881 			VOP_VPUT_PAIR(ndp->ni_dvp, nd->nd_repstat == 0 ?
1882 			    &ndp->ni_vp : NULL, false);
1883 			nfsvno_relpathbuf(ndp);
1884 			if (!nd->nd_repstat) {
1885 				if (*exclusive_flagp) {
1886 					*exclusive_flagp = 0;
1887 					NFSVNO_ATTRINIT(nvap);
1888 					nvap->na_atime.tv_sec = cverf[0];
1889 					nvap->na_atime.tv_nsec = cverf[1];
1890 					nd->nd_repstat = VOP_SETATTR(ndp->ni_vp,
1891 					    &nvap->na_vattr, cred);
1892 					if (nd->nd_repstat != 0) {
1893 						vput(ndp->ni_vp);
1894 						ndp->ni_vp = NULL;
1895 						nd->nd_repstat = NFSERR_NOTSUPP;
1896 					} else
1897 						NFSSETBIT_ATTRBIT(attrbitp,
1898 						    NFSATTRBIT_TIMEACCESS);
1899 				} else {
1900 					nfsrv_fixattr(nd, ndp->ni_vp, nvap,
1901 					    aclp, p, attrbitp, exp);
1902 				}
1903 			}
1904 			vp = ndp->ni_vp;
1905 		} else {
1906 			if (ndp->ni_startdir)
1907 				vrele(ndp->ni_startdir);
1908 			nfsvno_relpathbuf(ndp);
1909 			vp = ndp->ni_vp;
1910 			if (create == NFSV4OPEN_CREATE) {
1911 				if (ndp->ni_dvp == vp)
1912 					vrele(ndp->ni_dvp);
1913 				else
1914 					vput(ndp->ni_dvp);
1915 			}
1916 			if (NFSVNO_ISSETSIZE(nvap) && vp->v_type == VREG) {
1917 				if (ndp->ni_cnd.cn_flags & RDONLY)
1918 					NFSVNO_SETEXRDONLY(&nes);
1919 				else
1920 					NFSVNO_EXINIT(&nes);
1921 				nd->nd_repstat = nfsvno_accchk(vp,
1922 				    VWRITE, cred, &nes, p,
1923 				    NFSACCCHK_NOOVERRIDE,
1924 				    NFSACCCHK_VPISLOCKED, NULL);
1925 				nd->nd_repstat = nfsrv_opencheck(clientid,
1926 				    stateidp, stp, vp, nd, p, nd->nd_repstat);
1927 				if (!nd->nd_repstat) {
1928 					tempsize = nvap->na_size;
1929 					NFSVNO_ATTRINIT(nvap);
1930 					nvap->na_size = tempsize;
1931 					nd->nd_repstat = VOP_SETATTR(vp,
1932 					    &nvap->na_vattr, cred);
1933 				}
1934 			} else if (vp->v_type == VREG) {
1935 				nd->nd_repstat = nfsrv_opencheck(clientid,
1936 				    stateidp, stp, vp, nd, p, nd->nd_repstat);
1937 			}
1938 		}
1939 	} else {
1940 		if (ndp->ni_cnd.cn_flags & HASBUF)
1941 			nfsvno_relpathbuf(ndp);
1942 		if (ndp->ni_startdir && create == NFSV4OPEN_CREATE) {
1943 			vrele(ndp->ni_startdir);
1944 			if (ndp->ni_dvp == ndp->ni_vp)
1945 				vrele(ndp->ni_dvp);
1946 			else
1947 				vput(ndp->ni_dvp);
1948 			if (ndp->ni_vp)
1949 				vput(ndp->ni_vp);
1950 		}
1951 	}
1952 	*vpp = vp;
1953 
1954 	NFSEXITCODE2(0, nd);
1955 }
1956 
1957 /*
1958  * Updates the file rev and sets the mtime and ctime
1959  * to the current clock time, returning the va_filerev and va_Xtime
1960  * values.
1961  * Return ESTALE to indicate the vnode is VIRF_DOOMED.
1962  */
1963 int
1964 nfsvno_updfilerev(struct vnode *vp, struct nfsvattr *nvap,
1965     struct nfsrv_descript *nd, struct thread *p)
1966 {
1967 	struct vattr va;
1968 
1969 	VATTR_NULL(&va);
1970 	vfs_timestamp(&va.va_mtime);
1971 	if (NFSVOPISLOCKED(vp) != LK_EXCLUSIVE) {
1972 		NFSVOPLOCK(vp, LK_UPGRADE | LK_RETRY);
1973 		if (VN_IS_DOOMED(vp))
1974 			return (ESTALE);
1975 	}
1976 	(void) VOP_SETATTR(vp, &va, nd->nd_cred);
1977 	(void) nfsvno_getattr(vp, nvap, nd, p, 1, NULL);
1978 	return (0);
1979 }
1980 
1981 /*
1982  * Glue routine to nfsv4_fillattr().
1983  */
1984 int
1985 nfsvno_fillattr(struct nfsrv_descript *nd, struct mount *mp, struct vnode *vp,
1986     struct nfsvattr *nvap, fhandle_t *fhp, int rderror, nfsattrbit_t *attrbitp,
1987     struct ucred *cred, struct thread *p, int isdgram, int reterr,
1988     int supports_nfsv4acls, int at_root, uint64_t mounted_on_fileno)
1989 {
1990 	struct statfs *sf;
1991 	int error;
1992 
1993 	sf = NULL;
1994 	if (nfsrv_devidcnt > 0 &&
1995 	    (NFSISSET_ATTRBIT(attrbitp, NFSATTRBIT_SPACEAVAIL) ||
1996 	     NFSISSET_ATTRBIT(attrbitp, NFSATTRBIT_SPACEFREE) ||
1997 	     NFSISSET_ATTRBIT(attrbitp, NFSATTRBIT_SPACETOTAL))) {
1998 		sf = malloc(sizeof(*sf), M_TEMP, M_WAITOK | M_ZERO);
1999 		error = nfsrv_pnfsstatfs(sf, mp);
2000 		if (error != 0) {
2001 			free(sf, M_TEMP);
2002 			sf = NULL;
2003 		}
2004 	}
2005 	error = nfsv4_fillattr(nd, mp, vp, NULL, &nvap->na_vattr, fhp, rderror,
2006 	    attrbitp, cred, p, isdgram, reterr, supports_nfsv4acls, at_root,
2007 	    mounted_on_fileno, sf);
2008 	free(sf, M_TEMP);
2009 	NFSEXITCODE2(0, nd);
2010 	return (error);
2011 }
2012 
2013 /* Since the Readdir vnode ops vary, put the entire functions in here. */
2014 /*
2015  * nfs readdir service
2016  * - mallocs what it thinks is enough to read
2017  *	count rounded up to a multiple of DIRBLKSIZ <= NFS_MAXREADDIR
2018  * - calls VOP_READDIR()
2019  * - loops around building the reply
2020  *	if the output generated exceeds count break out of loop
2021  *	The NFSM_CLGET macro is used here so that the reply will be packed
2022  *	tightly in mbuf clusters.
2023  * - it trims out records with d_fileno == 0
2024  *	this doesn't matter for Unix clients, but they might confuse clients
2025  *	for other os'.
2026  * - it trims out records with d_type == DT_WHT
2027  *	these cannot be seen through NFS (unless we extend the protocol)
2028  *     The alternate call nfsrvd_readdirplus() does lookups as well.
2029  * PS: The NFS protocol spec. does not clarify what the "count" byte
2030  *	argument is a count of.. just name strings and file id's or the
2031  *	entire reply rpc or ...
2032  *	I tried just file name and id sizes and it confused the Sun client,
2033  *	so I am using the full rpc size now. The "paranoia.." comment refers
2034  *	to including the status longwords that are not a part of the dir.
2035  *	"entry" structures, but are in the rpc.
2036  */
2037 int
2038 nfsrvd_readdir(struct nfsrv_descript *nd, int isdgram,
2039     struct vnode *vp, struct nfsexstuff *exp)
2040 {
2041 	struct dirent *dp;
2042 	u_int32_t *tl;
2043 	int dirlen;
2044 	char *cpos, *cend, *rbuf;
2045 	struct nfsvattr at;
2046 	int nlen, error = 0, getret = 1;
2047 	int siz, cnt, fullsiz, eofflag, ncookies;
2048 	u_int64_t off, toff, verf __unused;
2049 	u_long *cookies = NULL, *cookiep;
2050 	struct uio io;
2051 	struct iovec iv;
2052 	int is_ufs;
2053 	struct thread *p = curthread;
2054 
2055 	if (nd->nd_repstat) {
2056 		nfsrv_postopattr(nd, getret, &at);
2057 		goto out;
2058 	}
2059 	if (nd->nd_flag & ND_NFSV2) {
2060 		NFSM_DISSECT(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
2061 		off = fxdr_unsigned(u_quad_t, *tl++);
2062 	} else {
2063 		NFSM_DISSECT(tl, u_int32_t *, 5 * NFSX_UNSIGNED);
2064 		off = fxdr_hyper(tl);
2065 		tl += 2;
2066 		verf = fxdr_hyper(tl);
2067 		tl += 2;
2068 	}
2069 	toff = off;
2070 	cnt = fxdr_unsigned(int, *tl);
2071 	if (cnt > NFS_SRVMAXDATA(nd) || cnt < 0)
2072 		cnt = NFS_SRVMAXDATA(nd);
2073 	siz = ((cnt + DIRBLKSIZ - 1) & ~(DIRBLKSIZ - 1));
2074 	fullsiz = siz;
2075 	if (nd->nd_flag & ND_NFSV3) {
2076 		nd->nd_repstat = getret = nfsvno_getattr(vp, &at, nd, p, 1,
2077 		    NULL);
2078 #if 0
2079 		/*
2080 		 * va_filerev is not sufficient as a cookie verifier,
2081 		 * since it is not supposed to change when entries are
2082 		 * removed/added unless that offset cookies returned to
2083 		 * the client are no longer valid.
2084 		 */
2085 		if (!nd->nd_repstat && toff && verf != at.na_filerev)
2086 			nd->nd_repstat = NFSERR_BAD_COOKIE;
2087 #endif
2088 	}
2089 	if (!nd->nd_repstat && vp->v_type != VDIR)
2090 		nd->nd_repstat = NFSERR_NOTDIR;
2091 	if (nd->nd_repstat == 0 && cnt == 0) {
2092 		if (nd->nd_flag & ND_NFSV2)
2093 			/* NFSv2 does not have NFSERR_TOOSMALL */
2094 			nd->nd_repstat = EPERM;
2095 		else
2096 			nd->nd_repstat = NFSERR_TOOSMALL;
2097 	}
2098 	if (!nd->nd_repstat)
2099 		nd->nd_repstat = nfsvno_accchk(vp, VEXEC,
2100 		    nd->nd_cred, exp, p, NFSACCCHK_NOOVERRIDE,
2101 		    NFSACCCHK_VPISLOCKED, NULL);
2102 	if (nd->nd_repstat) {
2103 		vput(vp);
2104 		if (nd->nd_flag & ND_NFSV3)
2105 			nfsrv_postopattr(nd, getret, &at);
2106 		goto out;
2107 	}
2108 	is_ufs = strcmp(vp->v_mount->mnt_vfc->vfc_name, "ufs") == 0;
2109 	rbuf = malloc(siz, M_TEMP, M_WAITOK);
2110 again:
2111 	eofflag = 0;
2112 	if (cookies) {
2113 		free(cookies, M_TEMP);
2114 		cookies = NULL;
2115 	}
2116 
2117 	iv.iov_base = rbuf;
2118 	iv.iov_len = siz;
2119 	io.uio_iov = &iv;
2120 	io.uio_iovcnt = 1;
2121 	io.uio_offset = (off_t)off;
2122 	io.uio_resid = siz;
2123 	io.uio_segflg = UIO_SYSSPACE;
2124 	io.uio_rw = UIO_READ;
2125 	io.uio_td = NULL;
2126 	nd->nd_repstat = VOP_READDIR(vp, &io, nd->nd_cred, &eofflag, &ncookies,
2127 	    &cookies);
2128 	off = (u_int64_t)io.uio_offset;
2129 	if (io.uio_resid)
2130 		siz -= io.uio_resid;
2131 
2132 	if (!cookies && !nd->nd_repstat)
2133 		nd->nd_repstat = NFSERR_PERM;
2134 	if (nd->nd_flag & ND_NFSV3) {
2135 		getret = nfsvno_getattr(vp, &at, nd, p, 1, NULL);
2136 		if (!nd->nd_repstat)
2137 			nd->nd_repstat = getret;
2138 	}
2139 
2140 	/*
2141 	 * Handles the failed cases. nd->nd_repstat == 0 past here.
2142 	 */
2143 	if (nd->nd_repstat) {
2144 		vput(vp);
2145 		free(rbuf, M_TEMP);
2146 		if (cookies)
2147 			free(cookies, M_TEMP);
2148 		if (nd->nd_flag & ND_NFSV3)
2149 			nfsrv_postopattr(nd, getret, &at);
2150 		goto out;
2151 	}
2152 	/*
2153 	 * If nothing read, return eof
2154 	 * rpc reply
2155 	 */
2156 	if (siz == 0) {
2157 		vput(vp);
2158 		if (nd->nd_flag & ND_NFSV2) {
2159 			NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
2160 		} else {
2161 			nfsrv_postopattr(nd, getret, &at);
2162 			NFSM_BUILD(tl, u_int32_t *, 4 * NFSX_UNSIGNED);
2163 			txdr_hyper(at.na_filerev, tl);
2164 			tl += 2;
2165 		}
2166 		*tl++ = newnfs_false;
2167 		*tl = newnfs_true;
2168 		free(rbuf, M_TEMP);
2169 		free(cookies, M_TEMP);
2170 		goto out;
2171 	}
2172 
2173 	/*
2174 	 * Check for degenerate cases of nothing useful read.
2175 	 * If so go try again
2176 	 */
2177 	cpos = rbuf;
2178 	cend = rbuf + siz;
2179 	dp = (struct dirent *)cpos;
2180 	cookiep = cookies;
2181 
2182 	/*
2183 	 * For some reason FreeBSD's ufs_readdir() chooses to back the
2184 	 * directory offset up to a block boundary, so it is necessary to
2185 	 * skip over the records that precede the requested offset. This
2186 	 * requires the assumption that file offset cookies monotonically
2187 	 * increase.
2188 	 */
2189 	while (cpos < cend && ncookies > 0 &&
2190 	    (dp->d_fileno == 0 || dp->d_type == DT_WHT ||
2191 	     (is_ufs == 1 && ((u_quad_t)(*cookiep)) <= toff))) {
2192 		cpos += dp->d_reclen;
2193 		dp = (struct dirent *)cpos;
2194 		cookiep++;
2195 		ncookies--;
2196 	}
2197 	if (cpos >= cend || ncookies == 0) {
2198 		siz = fullsiz;
2199 		toff = off;
2200 		goto again;
2201 	}
2202 	vput(vp);
2203 
2204 	/*
2205 	 * If cnt > MCLBYTES and the reply will not be saved, use
2206 	 * ext_pgs mbufs for TLS.
2207 	 * For NFSv4.0, we do not know for sure if the reply will
2208 	 * be saved, so do not use ext_pgs mbufs for NFSv4.0.
2209 	 */
2210 	if (cnt > MCLBYTES && siz > MCLBYTES &&
2211 	    (nd->nd_flag & (ND_TLS | ND_EXTPG | ND_SAVEREPLY)) == ND_TLS &&
2212 	    (nd->nd_flag & (ND_NFSV4 | ND_NFSV41)) != ND_NFSV4)
2213 		nd->nd_flag |= ND_EXTPG;
2214 
2215 	/*
2216 	 * dirlen is the size of the reply, including all XDR and must
2217 	 * not exceed cnt. For NFSv2, RFC1094 didn't clearly indicate
2218 	 * if the XDR should be included in "count", but to be safe, we do.
2219 	 * (Include the two booleans at the end of the reply in dirlen now.)
2220 	 */
2221 	if (nd->nd_flag & ND_NFSV3) {
2222 		nfsrv_postopattr(nd, getret, &at);
2223 		NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
2224 		txdr_hyper(at.na_filerev, tl);
2225 		dirlen = NFSX_V3POSTOPATTR + NFSX_VERF + 2 * NFSX_UNSIGNED;
2226 	} else {
2227 		dirlen = 2 * NFSX_UNSIGNED;
2228 	}
2229 
2230 	/* Loop through the records and build reply */
2231 	while (cpos < cend && ncookies > 0) {
2232 		nlen = dp->d_namlen;
2233 		if (dp->d_fileno != 0 && dp->d_type != DT_WHT &&
2234 			nlen <= NFS_MAXNAMLEN) {
2235 			if (nd->nd_flag & ND_NFSV3)
2236 				dirlen += (6*NFSX_UNSIGNED + NFSM_RNDUP(nlen));
2237 			else
2238 				dirlen += (4*NFSX_UNSIGNED + NFSM_RNDUP(nlen));
2239 			if (dirlen > cnt) {
2240 				eofflag = 0;
2241 				break;
2242 			}
2243 
2244 			/*
2245 			 * Build the directory record xdr from
2246 			 * the dirent entry.
2247 			 */
2248 			if (nd->nd_flag & ND_NFSV3) {
2249 				NFSM_BUILD(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
2250 				*tl++ = newnfs_true;
2251 				*tl++ = 0;
2252 			} else {
2253 				NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
2254 				*tl++ = newnfs_true;
2255 			}
2256 			*tl = txdr_unsigned(dp->d_fileno);
2257 			(void) nfsm_strtom(nd, dp->d_name, nlen);
2258 			if (nd->nd_flag & ND_NFSV3) {
2259 				NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
2260 				*tl++ = 0;
2261 			} else
2262 				NFSM_BUILD(tl, u_int32_t *, NFSX_UNSIGNED);
2263 			*tl = txdr_unsigned(*cookiep);
2264 		}
2265 		cpos += dp->d_reclen;
2266 		dp = (struct dirent *)cpos;
2267 		cookiep++;
2268 		ncookies--;
2269 	}
2270 	if (cpos < cend)
2271 		eofflag = 0;
2272 	NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
2273 	*tl++ = newnfs_false;
2274 	if (eofflag)
2275 		*tl = newnfs_true;
2276 	else
2277 		*tl = newnfs_false;
2278 	free(rbuf, M_TEMP);
2279 	free(cookies, M_TEMP);
2280 
2281 out:
2282 	NFSEXITCODE2(0, nd);
2283 	return (0);
2284 nfsmout:
2285 	vput(vp);
2286 	NFSEXITCODE2(error, nd);
2287 	return (error);
2288 }
2289 
2290 /*
2291  * Readdirplus for V3 and Readdir for V4.
2292  */
2293 int
2294 nfsrvd_readdirplus(struct nfsrv_descript *nd, int isdgram,
2295     struct vnode *vp, struct nfsexstuff *exp)
2296 {
2297 	struct dirent *dp;
2298 	u_int32_t *tl;
2299 	int dirlen;
2300 	char *cpos, *cend, *rbuf;
2301 	struct vnode *nvp;
2302 	fhandle_t nfh;
2303 	struct nfsvattr nva, at, *nvap = &nva;
2304 	struct mbuf *mb0, *mb1;
2305 	struct nfsreferral *refp;
2306 	int nlen, r, error = 0, getret = 1, usevget = 1;
2307 	int siz, cnt, fullsiz, eofflag, ncookies, entrycnt;
2308 	caddr_t bpos0, bpos1;
2309 	u_int64_t off, toff, verf;
2310 	u_long *cookies = NULL, *cookiep;
2311 	nfsattrbit_t attrbits, rderrbits, savbits;
2312 	struct uio io;
2313 	struct iovec iv;
2314 	struct componentname cn;
2315 	int at_root, is_ufs, is_zfs, needs_unbusy, supports_nfsv4acls;
2316 	struct mount *mp, *new_mp;
2317 	uint64_t mounted_on_fileno;
2318 	struct thread *p = curthread;
2319 	int bextpg0, bextpg1, bextpgsiz0, bextpgsiz1;
2320 
2321 	if (nd->nd_repstat) {
2322 		nfsrv_postopattr(nd, getret, &at);
2323 		goto out;
2324 	}
2325 	NFSM_DISSECT(tl, u_int32_t *, 6 * NFSX_UNSIGNED);
2326 	off = fxdr_hyper(tl);
2327 	toff = off;
2328 	tl += 2;
2329 	verf = fxdr_hyper(tl);
2330 	tl += 2;
2331 	siz = fxdr_unsigned(int, *tl++);
2332 	cnt = fxdr_unsigned(int, *tl);
2333 
2334 	/*
2335 	 * Use the server's maximum data transfer size as the upper bound
2336 	 * on reply datalen.
2337 	 */
2338 	if (cnt > NFS_SRVMAXDATA(nd) || cnt < 0)
2339 		cnt = NFS_SRVMAXDATA(nd);
2340 
2341 	/*
2342 	 * siz is a "hint" of how much directory information (name, fileid,
2343 	 * cookie) should be in the reply. At least one client "hints" 0,
2344 	 * so I set it to cnt for that case. I also round it up to the
2345 	 * next multiple of DIRBLKSIZ.
2346 	 * Since the size of a Readdirplus directory entry reply will always
2347 	 * be greater than a directory entry returned by VOP_READDIR(), it
2348 	 * does not make sense to read more than NFS_SRVMAXDATA() via
2349 	 * VOP_READDIR().
2350 	 */
2351 	if (siz <= 0)
2352 		siz = cnt;
2353 	else if (siz > NFS_SRVMAXDATA(nd))
2354 		siz = NFS_SRVMAXDATA(nd);
2355 	siz = ((siz + DIRBLKSIZ - 1) & ~(DIRBLKSIZ - 1));
2356 
2357 	if (nd->nd_flag & ND_NFSV4) {
2358 		error = nfsrv_getattrbits(nd, &attrbits, NULL, NULL);
2359 		if (error)
2360 			goto nfsmout;
2361 		NFSSET_ATTRBIT(&savbits, &attrbits);
2362 		NFSCLRNOTFILLABLE_ATTRBIT(&attrbits, nd);
2363 		NFSZERO_ATTRBIT(&rderrbits);
2364 		NFSSETBIT_ATTRBIT(&rderrbits, NFSATTRBIT_RDATTRERROR);
2365 	} else {
2366 		NFSZERO_ATTRBIT(&attrbits);
2367 	}
2368 	fullsiz = siz;
2369 	nd->nd_repstat = getret = nfsvno_getattr(vp, &at, nd, p, 1, NULL);
2370 #if 0
2371 	if (!nd->nd_repstat) {
2372 	    if (off && verf != at.na_filerev) {
2373 		/*
2374 		 * va_filerev is not sufficient as a cookie verifier,
2375 		 * since it is not supposed to change when entries are
2376 		 * removed/added unless that offset cookies returned to
2377 		 * the client are no longer valid.
2378 		 */
2379 		if (nd->nd_flag & ND_NFSV4) {
2380 			nd->nd_repstat = NFSERR_NOTSAME;
2381 		} else {
2382 			nd->nd_repstat = NFSERR_BAD_COOKIE;
2383 		}
2384 	    }
2385 	}
2386 #endif
2387 	if (!nd->nd_repstat && vp->v_type != VDIR)
2388 		nd->nd_repstat = NFSERR_NOTDIR;
2389 	if (!nd->nd_repstat && cnt == 0)
2390 		nd->nd_repstat = NFSERR_TOOSMALL;
2391 	if (!nd->nd_repstat)
2392 		nd->nd_repstat = nfsvno_accchk(vp, VEXEC,
2393 		    nd->nd_cred, exp, p, NFSACCCHK_NOOVERRIDE,
2394 		    NFSACCCHK_VPISLOCKED, NULL);
2395 	if (nd->nd_repstat) {
2396 		vput(vp);
2397 		if (nd->nd_flag & ND_NFSV3)
2398 			nfsrv_postopattr(nd, getret, &at);
2399 		goto out;
2400 	}
2401 	is_ufs = strcmp(vp->v_mount->mnt_vfc->vfc_name, "ufs") == 0;
2402 	is_zfs = strcmp(vp->v_mount->mnt_vfc->vfc_name, "zfs") == 0;
2403 
2404 	rbuf = malloc(siz, M_TEMP, M_WAITOK);
2405 again:
2406 	eofflag = 0;
2407 	if (cookies) {
2408 		free(cookies, M_TEMP);
2409 		cookies = NULL;
2410 	}
2411 
2412 	iv.iov_base = rbuf;
2413 	iv.iov_len = siz;
2414 	io.uio_iov = &iv;
2415 	io.uio_iovcnt = 1;
2416 	io.uio_offset = (off_t)off;
2417 	io.uio_resid = siz;
2418 	io.uio_segflg = UIO_SYSSPACE;
2419 	io.uio_rw = UIO_READ;
2420 	io.uio_td = NULL;
2421 	nd->nd_repstat = VOP_READDIR(vp, &io, nd->nd_cred, &eofflag, &ncookies,
2422 	    &cookies);
2423 	off = (u_int64_t)io.uio_offset;
2424 	if (io.uio_resid)
2425 		siz -= io.uio_resid;
2426 
2427 	getret = nfsvno_getattr(vp, &at, nd, p, 1, NULL);
2428 
2429 	if (!cookies && !nd->nd_repstat)
2430 		nd->nd_repstat = NFSERR_PERM;
2431 	if (!nd->nd_repstat)
2432 		nd->nd_repstat = getret;
2433 	if (nd->nd_repstat) {
2434 		vput(vp);
2435 		if (cookies)
2436 			free(cookies, M_TEMP);
2437 		free(rbuf, M_TEMP);
2438 		if (nd->nd_flag & ND_NFSV3)
2439 			nfsrv_postopattr(nd, getret, &at);
2440 		goto out;
2441 	}
2442 	/*
2443 	 * If nothing read, return eof
2444 	 * rpc reply
2445 	 */
2446 	if (siz == 0) {
2447 		vput(vp);
2448 		if (nd->nd_flag & ND_NFSV3)
2449 			nfsrv_postopattr(nd, getret, &at);
2450 		NFSM_BUILD(tl, u_int32_t *, 4 * NFSX_UNSIGNED);
2451 		txdr_hyper(at.na_filerev, tl);
2452 		tl += 2;
2453 		*tl++ = newnfs_false;
2454 		*tl = newnfs_true;
2455 		free(cookies, M_TEMP);
2456 		free(rbuf, M_TEMP);
2457 		goto out;
2458 	}
2459 
2460 	/*
2461 	 * Check for degenerate cases of nothing useful read.
2462 	 * If so go try again
2463 	 */
2464 	cpos = rbuf;
2465 	cend = rbuf + siz;
2466 	dp = (struct dirent *)cpos;
2467 	cookiep = cookies;
2468 
2469 	/*
2470 	 * For some reason FreeBSD's ufs_readdir() chooses to back the
2471 	 * directory offset up to a block boundary, so it is necessary to
2472 	 * skip over the records that precede the requested offset. This
2473 	 * requires the assumption that file offset cookies monotonically
2474 	 * increase.
2475 	 */
2476 	while (cpos < cend && ncookies > 0 &&
2477 	  (dp->d_fileno == 0 || dp->d_type == DT_WHT ||
2478 	   (is_ufs == 1 && ((u_quad_t)(*cookiep)) <= toff) ||
2479 	   ((nd->nd_flag & ND_NFSV4) &&
2480 	    ((dp->d_namlen == 1 && dp->d_name[0] == '.') ||
2481 	     (dp->d_namlen==2 && dp->d_name[0]=='.' && dp->d_name[1]=='.'))))) {
2482 		cpos += dp->d_reclen;
2483 		dp = (struct dirent *)cpos;
2484 		cookiep++;
2485 		ncookies--;
2486 	}
2487 	if (cpos >= cend || ncookies == 0) {
2488 		siz = fullsiz;
2489 		toff = off;
2490 		goto again;
2491 	}
2492 
2493 	/*
2494 	 * Busy the file system so that the mount point won't go away
2495 	 * and, as such, VFS_VGET() can be used safely.
2496 	 */
2497 	mp = vp->v_mount;
2498 	vfs_ref(mp);
2499 	NFSVOPUNLOCK(vp);
2500 	nd->nd_repstat = vfs_busy(mp, 0);
2501 	vfs_rel(mp);
2502 	if (nd->nd_repstat != 0) {
2503 		vrele(vp);
2504 		free(cookies, M_TEMP);
2505 		free(rbuf, M_TEMP);
2506 		if (nd->nd_flag & ND_NFSV3)
2507 			nfsrv_postopattr(nd, getret, &at);
2508 		goto out;
2509 	}
2510 
2511 	/*
2512 	 * Check to see if entries in this directory can be safely acquired
2513 	 * via VFS_VGET() or if a switch to VOP_LOOKUP() is required.
2514 	 * ZFS snapshot directories need VOP_LOOKUP(), so that any
2515 	 * automount of the snapshot directory that is required will
2516 	 * be done.
2517 	 * This needs to be done here for NFSv4, since NFSv4 never does
2518 	 * a VFS_VGET() for "." or "..".
2519 	 */
2520 	if (is_zfs == 1) {
2521 		r = VFS_VGET(mp, at.na_fileid, LK_SHARED, &nvp);
2522 		if (r == EOPNOTSUPP) {
2523 			usevget = 0;
2524 			cn.cn_nameiop = LOOKUP;
2525 			cn.cn_lkflags = LK_SHARED | LK_RETRY;
2526 			cn.cn_cred = nd->nd_cred;
2527 			cn.cn_thread = p;
2528 		} else if (r == 0)
2529 			vput(nvp);
2530 	}
2531 
2532 	/*
2533 	 * If the reply is likely to exceed MCLBYTES and the reply will
2534 	 * not be saved, use ext_pgs mbufs for TLS.
2535 	 * It is difficult to predict how large each entry will be and
2536 	 * how many entries have been read, so just assume the directory
2537 	 * entries grow by a factor of 4 when attributes are included.
2538 	 * For NFSv4.0, we do not know for sure if the reply will
2539 	 * be saved, so do not use ext_pgs mbufs for NFSv4.0.
2540 	 */
2541 	if (cnt > MCLBYTES && siz > MCLBYTES / 4 &&
2542 	    (nd->nd_flag & (ND_TLS | ND_EXTPG | ND_SAVEREPLY)) == ND_TLS &&
2543 	    (nd->nd_flag & (ND_NFSV4 | ND_NFSV41)) != ND_NFSV4)
2544 		nd->nd_flag |= ND_EXTPG;
2545 
2546 	/*
2547 	 * Save this position, in case there is an error before one entry
2548 	 * is created.
2549 	 */
2550 	mb0 = nd->nd_mb;
2551 	bpos0 = nd->nd_bpos;
2552 	bextpg0 = nd->nd_bextpg;
2553 	bextpgsiz0 = nd->nd_bextpgsiz;
2554 
2555 	/*
2556 	 * Fill in the first part of the reply.
2557 	 * dirlen is the reply length in bytes and cannot exceed cnt.
2558 	 * (Include the two booleans at the end of the reply in dirlen now,
2559 	 *  so we recognize when we have exceeded cnt.)
2560 	 */
2561 	if (nd->nd_flag & ND_NFSV3) {
2562 		dirlen = NFSX_V3POSTOPATTR + NFSX_VERF + 2 * NFSX_UNSIGNED;
2563 		nfsrv_postopattr(nd, getret, &at);
2564 	} else {
2565 		dirlen = NFSX_VERF + 2 * NFSX_UNSIGNED;
2566 	}
2567 	NFSM_BUILD(tl, u_int32_t *, NFSX_VERF);
2568 	txdr_hyper(at.na_filerev, tl);
2569 
2570 	/*
2571 	 * Save this position, in case there is an empty reply needed.
2572 	 */
2573 	mb1 = nd->nd_mb;
2574 	bpos1 = nd->nd_bpos;
2575 	bextpg1 = nd->nd_bextpg;
2576 	bextpgsiz1 = nd->nd_bextpgsiz;
2577 
2578 	/* Loop through the records and build reply */
2579 	entrycnt = 0;
2580 	while (cpos < cend && ncookies > 0 && dirlen < cnt) {
2581 		nlen = dp->d_namlen;
2582 		if (dp->d_fileno != 0 && dp->d_type != DT_WHT &&
2583 		    nlen <= NFS_MAXNAMLEN &&
2584 		    ((nd->nd_flag & ND_NFSV3) || nlen > 2 ||
2585 		     (nlen==2 && (dp->d_name[0]!='.' || dp->d_name[1]!='.'))
2586 		      || (nlen == 1 && dp->d_name[0] != '.'))) {
2587 			/*
2588 			 * Save the current position in the reply, in case
2589 			 * this entry exceeds cnt.
2590 			 */
2591 			mb1 = nd->nd_mb;
2592 			bpos1 = nd->nd_bpos;
2593 			bextpg1 = nd->nd_bextpg;
2594 			bextpgsiz1 = nd->nd_bextpgsiz;
2595 
2596 			/*
2597 			 * For readdir_and_lookup get the vnode using
2598 			 * the file number.
2599 			 */
2600 			nvp = NULL;
2601 			refp = NULL;
2602 			r = 0;
2603 			at_root = 0;
2604 			needs_unbusy = 0;
2605 			new_mp = mp;
2606 			mounted_on_fileno = (uint64_t)dp->d_fileno;
2607 			if ((nd->nd_flag & ND_NFSV3) ||
2608 			    NFSNONZERO_ATTRBIT(&savbits)) {
2609 				if (nd->nd_flag & ND_NFSV4)
2610 					refp = nfsv4root_getreferral(NULL,
2611 					    vp, dp->d_fileno);
2612 				if (refp == NULL) {
2613 					if (usevget)
2614 						r = VFS_VGET(mp, dp->d_fileno,
2615 						    LK_SHARED, &nvp);
2616 					else
2617 						r = EOPNOTSUPP;
2618 					if (r == EOPNOTSUPP) {
2619 						if (usevget) {
2620 							usevget = 0;
2621 							cn.cn_nameiop = LOOKUP;
2622 							cn.cn_lkflags =
2623 							    LK_SHARED |
2624 							    LK_RETRY;
2625 							cn.cn_cred =
2626 							    nd->nd_cred;
2627 							cn.cn_thread = p;
2628 						}
2629 						cn.cn_nameptr = dp->d_name;
2630 						cn.cn_namelen = nlen;
2631 						cn.cn_flags = ISLASTCN |
2632 						    NOFOLLOW | LOCKLEAF;
2633 						if (nlen == 2 &&
2634 						    dp->d_name[0] == '.' &&
2635 						    dp->d_name[1] == '.')
2636 							cn.cn_flags |=
2637 							    ISDOTDOT;
2638 						if (NFSVOPLOCK(vp, LK_SHARED)
2639 						    != 0) {
2640 							nd->nd_repstat = EPERM;
2641 							break;
2642 						}
2643 						if ((vp->v_vflag & VV_ROOT) != 0
2644 						    && (cn.cn_flags & ISDOTDOT)
2645 						    != 0) {
2646 							vref(vp);
2647 							nvp = vp;
2648 							r = 0;
2649 						} else {
2650 							r = VOP_LOOKUP(vp, &nvp,
2651 							    &cn);
2652 							if (vp != nvp)
2653 								NFSVOPUNLOCK(vp);
2654 						}
2655 					}
2656 
2657 					/*
2658 					 * For NFSv4, check to see if nvp is
2659 					 * a mount point and get the mount
2660 					 * point vnode, as required.
2661 					 */
2662 					if (r == 0 &&
2663 					    nfsrv_enable_crossmntpt != 0 &&
2664 					    (nd->nd_flag & ND_NFSV4) != 0 &&
2665 					    nvp->v_type == VDIR &&
2666 					    nvp->v_mountedhere != NULL) {
2667 						new_mp = nvp->v_mountedhere;
2668 						r = vfs_busy(new_mp, 0);
2669 						vput(nvp);
2670 						nvp = NULL;
2671 						if (r == 0) {
2672 							r = VFS_ROOT(new_mp,
2673 							    LK_SHARED, &nvp);
2674 							needs_unbusy = 1;
2675 							if (r == 0)
2676 								at_root = 1;
2677 						}
2678 					}
2679 				}
2680 
2681 				/*
2682 				 * If we failed to look up the entry, then it
2683 				 * has become invalid, most likely removed.
2684 				 */
2685 				if (r != 0) {
2686 					if (needs_unbusy)
2687 						vfs_unbusy(new_mp);
2688 					goto invalid;
2689 				}
2690 				KASSERT(refp != NULL || nvp != NULL,
2691 				    ("%s: undetected lookup error", __func__));
2692 
2693 				if (refp == NULL &&
2694 				    ((nd->nd_flag & ND_NFSV3) ||
2695 				     NFSNONZERO_ATTRBIT(&attrbits))) {
2696 					r = nfsvno_getfh(nvp, &nfh, p);
2697 					if (!r)
2698 					    r = nfsvno_getattr(nvp, nvap, nd, p,
2699 						1, &attrbits);
2700 					if (r == 0 && is_zfs == 1 &&
2701 					    nfsrv_enable_crossmntpt != 0 &&
2702 					    (nd->nd_flag & ND_NFSV4) != 0 &&
2703 					    nvp->v_type == VDIR &&
2704 					    vp->v_mount != nvp->v_mount) {
2705 					    /*
2706 					     * For a ZFS snapshot, there is a
2707 					     * pseudo mount that does not set
2708 					     * v_mountedhere, so it needs to
2709 					     * be detected via a different
2710 					     * mount structure.
2711 					     */
2712 					    at_root = 1;
2713 					    if (new_mp == mp)
2714 						new_mp = nvp->v_mount;
2715 					}
2716 				}
2717 
2718 				/*
2719 				 * If we failed to get attributes of the entry,
2720 				 * then just skip it for NFSv3 (the traditional
2721 				 * behavior in the old NFS server).
2722 				 * For NFSv4 the behavior is controlled by
2723 				 * RDATTRERROR: we either ignore the error or
2724 				 * fail the request.
2725 				 * Note that RDATTRERROR is never set for NFSv3.
2726 				 */
2727 				if (r != 0) {
2728 					if (!NFSISSET_ATTRBIT(&attrbits,
2729 					    NFSATTRBIT_RDATTRERROR)) {
2730 						vput(nvp);
2731 						if (needs_unbusy != 0)
2732 							vfs_unbusy(new_mp);
2733 						if ((nd->nd_flag & ND_NFSV3))
2734 							goto invalid;
2735 						nd->nd_repstat = r;
2736 						break;
2737 					}
2738 				}
2739 			}
2740 
2741 			/*
2742 			 * Build the directory record xdr
2743 			 */
2744 			if (nd->nd_flag & ND_NFSV3) {
2745 				NFSM_BUILD(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
2746 				*tl++ = newnfs_true;
2747 				*tl++ = 0;
2748 				*tl = txdr_unsigned(dp->d_fileno);
2749 				dirlen += nfsm_strtom(nd, dp->d_name, nlen);
2750 				NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
2751 				*tl++ = 0;
2752 				*tl = txdr_unsigned(*cookiep);
2753 				nfsrv_postopattr(nd, 0, nvap);
2754 				dirlen += nfsm_fhtom(nd,(u_int8_t *)&nfh,0,1);
2755 				dirlen += (5*NFSX_UNSIGNED+NFSX_V3POSTOPATTR);
2756 				if (nvp != NULL)
2757 					vput(nvp);
2758 			} else {
2759 				NFSM_BUILD(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
2760 				*tl++ = newnfs_true;
2761 				*tl++ = 0;
2762 				*tl = txdr_unsigned(*cookiep);
2763 				dirlen += nfsm_strtom(nd, dp->d_name, nlen);
2764 				if (nvp != NULL) {
2765 					supports_nfsv4acls =
2766 					    nfs_supportsnfsv4acls(nvp);
2767 					NFSVOPUNLOCK(nvp);
2768 				} else
2769 					supports_nfsv4acls = 0;
2770 				if (refp != NULL) {
2771 					dirlen += nfsrv_putreferralattr(nd,
2772 					    &savbits, refp, 0,
2773 					    &nd->nd_repstat);
2774 					if (nd->nd_repstat) {
2775 						if (nvp != NULL)
2776 							vrele(nvp);
2777 						if (needs_unbusy != 0)
2778 							vfs_unbusy(new_mp);
2779 						break;
2780 					}
2781 				} else if (r) {
2782 					dirlen += nfsvno_fillattr(nd, new_mp,
2783 					    nvp, nvap, &nfh, r, &rderrbits,
2784 					    nd->nd_cred, p, isdgram, 0,
2785 					    supports_nfsv4acls, at_root,
2786 					    mounted_on_fileno);
2787 				} else {
2788 					dirlen += nfsvno_fillattr(nd, new_mp,
2789 					    nvp, nvap, &nfh, r, &attrbits,
2790 					    nd->nd_cred, p, isdgram, 0,
2791 					    supports_nfsv4acls, at_root,
2792 					    mounted_on_fileno);
2793 				}
2794 				if (nvp != NULL)
2795 					vrele(nvp);
2796 				dirlen += (3 * NFSX_UNSIGNED);
2797 			}
2798 			if (needs_unbusy != 0)
2799 				vfs_unbusy(new_mp);
2800 			if (dirlen <= cnt)
2801 				entrycnt++;
2802 		}
2803 invalid:
2804 		cpos += dp->d_reclen;
2805 		dp = (struct dirent *)cpos;
2806 		cookiep++;
2807 		ncookies--;
2808 	}
2809 	vrele(vp);
2810 	vfs_unbusy(mp);
2811 
2812 	/*
2813 	 * If dirlen > cnt, we must strip off the last entry. If that
2814 	 * results in an empty reply, report NFSERR_TOOSMALL.
2815 	 */
2816 	if (dirlen > cnt || nd->nd_repstat) {
2817 		if (!nd->nd_repstat && entrycnt == 0)
2818 			nd->nd_repstat = NFSERR_TOOSMALL;
2819 		if (nd->nd_repstat) {
2820 			nfsm_trimtrailing(nd, mb0, bpos0, bextpg0, bextpgsiz0);
2821 			if (nd->nd_flag & ND_NFSV3)
2822 				nfsrv_postopattr(nd, getret, &at);
2823 		} else
2824 			nfsm_trimtrailing(nd, mb1, bpos1, bextpg1, bextpgsiz1);
2825 		eofflag = 0;
2826 	} else if (cpos < cend)
2827 		eofflag = 0;
2828 	if (!nd->nd_repstat) {
2829 		NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
2830 		*tl++ = newnfs_false;
2831 		if (eofflag)
2832 			*tl = newnfs_true;
2833 		else
2834 			*tl = newnfs_false;
2835 	}
2836 	free(cookies, M_TEMP);
2837 	free(rbuf, M_TEMP);
2838 
2839 out:
2840 	NFSEXITCODE2(0, nd);
2841 	return (0);
2842 nfsmout:
2843 	vput(vp);
2844 	NFSEXITCODE2(error, nd);
2845 	return (error);
2846 }
2847 
2848 /*
2849  * Get the settable attributes out of the mbuf list.
2850  * (Return 0 or EBADRPC)
2851  */
2852 int
2853 nfsrv_sattr(struct nfsrv_descript *nd, vnode_t vp, struct nfsvattr *nvap,
2854     nfsattrbit_t *attrbitp, NFSACL_T *aclp, struct thread *p)
2855 {
2856 	u_int32_t *tl;
2857 	struct nfsv2_sattr *sp;
2858 	int error = 0, toclient = 0;
2859 
2860 	switch (nd->nd_flag & (ND_NFSV2 | ND_NFSV3 | ND_NFSV4)) {
2861 	case ND_NFSV2:
2862 		NFSM_DISSECT(sp, struct nfsv2_sattr *, NFSX_V2SATTR);
2863 		/*
2864 		 * Some old clients didn't fill in the high order 16bits.
2865 		 * --> check the low order 2 bytes for 0xffff
2866 		 */
2867 		if ((fxdr_unsigned(int, sp->sa_mode) & 0xffff) != 0xffff)
2868 			nvap->na_mode = nfstov_mode(sp->sa_mode);
2869 		if (sp->sa_uid != newnfs_xdrneg1)
2870 			nvap->na_uid = fxdr_unsigned(uid_t, sp->sa_uid);
2871 		if (sp->sa_gid != newnfs_xdrneg1)
2872 			nvap->na_gid = fxdr_unsigned(gid_t, sp->sa_gid);
2873 		if (sp->sa_size != newnfs_xdrneg1)
2874 			nvap->na_size = fxdr_unsigned(u_quad_t, sp->sa_size);
2875 		if (sp->sa_atime.nfsv2_sec != newnfs_xdrneg1) {
2876 #ifdef notyet
2877 			fxdr_nfsv2time(&sp->sa_atime, &nvap->na_atime);
2878 #else
2879 			nvap->na_atime.tv_sec =
2880 				fxdr_unsigned(u_int32_t,sp->sa_atime.nfsv2_sec);
2881 			nvap->na_atime.tv_nsec = 0;
2882 #endif
2883 		}
2884 		if (sp->sa_mtime.nfsv2_sec != newnfs_xdrneg1)
2885 			fxdr_nfsv2time(&sp->sa_mtime, &nvap->na_mtime);
2886 		break;
2887 	case ND_NFSV3:
2888 		NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2889 		if (*tl == newnfs_true) {
2890 			NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2891 			nvap->na_mode = nfstov_mode(*tl);
2892 		}
2893 		NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2894 		if (*tl == newnfs_true) {
2895 			NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2896 			nvap->na_uid = fxdr_unsigned(uid_t, *tl);
2897 		}
2898 		NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2899 		if (*tl == newnfs_true) {
2900 			NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2901 			nvap->na_gid = fxdr_unsigned(gid_t, *tl);
2902 		}
2903 		NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2904 		if (*tl == newnfs_true) {
2905 			NFSM_DISSECT(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
2906 			nvap->na_size = fxdr_hyper(tl);
2907 		}
2908 		NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2909 		switch (fxdr_unsigned(int, *tl)) {
2910 		case NFSV3SATTRTIME_TOCLIENT:
2911 			NFSM_DISSECT(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
2912 			fxdr_nfsv3time(tl, &nvap->na_atime);
2913 			toclient = 1;
2914 			break;
2915 		case NFSV3SATTRTIME_TOSERVER:
2916 			vfs_timestamp(&nvap->na_atime);
2917 			nvap->na_vaflags |= VA_UTIMES_NULL;
2918 			break;
2919 		}
2920 		NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2921 		switch (fxdr_unsigned(int, *tl)) {
2922 		case NFSV3SATTRTIME_TOCLIENT:
2923 			NFSM_DISSECT(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
2924 			fxdr_nfsv3time(tl, &nvap->na_mtime);
2925 			nvap->na_vaflags &= ~VA_UTIMES_NULL;
2926 			break;
2927 		case NFSV3SATTRTIME_TOSERVER:
2928 			vfs_timestamp(&nvap->na_mtime);
2929 			if (!toclient)
2930 				nvap->na_vaflags |= VA_UTIMES_NULL;
2931 			break;
2932 		}
2933 		break;
2934 	case ND_NFSV4:
2935 		error = nfsv4_sattr(nd, vp, nvap, attrbitp, aclp, p);
2936 	}
2937 nfsmout:
2938 	NFSEXITCODE2(error, nd);
2939 	return (error);
2940 }
2941 
2942 /*
2943  * Handle the setable attributes for V4.
2944  * Returns NFSERR_BADXDR if it can't be parsed, 0 otherwise.
2945  */
2946 int
2947 nfsv4_sattr(struct nfsrv_descript *nd, vnode_t vp, struct nfsvattr *nvap,
2948     nfsattrbit_t *attrbitp, NFSACL_T *aclp, struct thread *p)
2949 {
2950 	u_int32_t *tl;
2951 	int attrsum = 0;
2952 	int i, j;
2953 	int error, attrsize, bitpos, aclsize, aceerr, retnotsup = 0;
2954 	int moderet, toclient = 0;
2955 	u_char *cp, namestr[NFSV4_SMALLSTR + 1];
2956 	uid_t uid;
2957 	gid_t gid;
2958 	u_short mode, mask;		/* Same type as va_mode. */
2959 	struct vattr va;
2960 
2961 	error = nfsrv_getattrbits(nd, attrbitp, NULL, &retnotsup);
2962 	if (error)
2963 		goto nfsmout;
2964 	NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2965 	attrsize = fxdr_unsigned(int, *tl);
2966 
2967 	/*
2968 	 * Loop around getting the setable attributes. If an unsupported
2969 	 * one is found, set nd_repstat == NFSERR_ATTRNOTSUPP and return.
2970 	 */
2971 	if (retnotsup) {
2972 		nd->nd_repstat = NFSERR_ATTRNOTSUPP;
2973 		bitpos = NFSATTRBIT_MAX;
2974 	} else {
2975 		bitpos = 0;
2976 	}
2977 	moderet = 0;
2978 	for (; bitpos < NFSATTRBIT_MAX; bitpos++) {
2979 	    if (attrsum > attrsize) {
2980 		error = NFSERR_BADXDR;
2981 		goto nfsmout;
2982 	    }
2983 	    if (NFSISSET_ATTRBIT(attrbitp, bitpos))
2984 		switch (bitpos) {
2985 		case NFSATTRBIT_SIZE:
2986 			NFSM_DISSECT(tl, u_int32_t *, NFSX_HYPER);
2987                      if (vp != NULL && vp->v_type != VREG) {
2988                             error = (vp->v_type == VDIR) ? NFSERR_ISDIR :
2989                                 NFSERR_INVAL;
2990                             goto nfsmout;
2991 			}
2992 			nvap->na_size = fxdr_hyper(tl);
2993 			attrsum += NFSX_HYPER;
2994 			break;
2995 		case NFSATTRBIT_ACL:
2996 			error = nfsrv_dissectacl(nd, aclp, &aceerr, &aclsize,
2997 			    p);
2998 			if (error)
2999 				goto nfsmout;
3000 			if (aceerr && !nd->nd_repstat)
3001 				nd->nd_repstat = aceerr;
3002 			attrsum += aclsize;
3003 			break;
3004 		case NFSATTRBIT_ARCHIVE:
3005 			NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
3006 			if (!nd->nd_repstat)
3007 				nd->nd_repstat = NFSERR_ATTRNOTSUPP;
3008 			attrsum += NFSX_UNSIGNED;
3009 			break;
3010 		case NFSATTRBIT_HIDDEN:
3011 			NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
3012 			if (!nd->nd_repstat)
3013 				nd->nd_repstat = NFSERR_ATTRNOTSUPP;
3014 			attrsum += NFSX_UNSIGNED;
3015 			break;
3016 		case NFSATTRBIT_MIMETYPE:
3017 			NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
3018 			i = fxdr_unsigned(int, *tl);
3019 			error = nfsm_advance(nd, NFSM_RNDUP(i), -1);
3020 			if (error)
3021 				goto nfsmout;
3022 			if (!nd->nd_repstat)
3023 				nd->nd_repstat = NFSERR_ATTRNOTSUPP;
3024 			attrsum += (NFSX_UNSIGNED + NFSM_RNDUP(i));
3025 			break;
3026 		case NFSATTRBIT_MODE:
3027 			moderet = NFSERR_INVAL;	/* Can't do MODESETMASKED. */
3028 			NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
3029 			nvap->na_mode = nfstov_mode(*tl);
3030 			attrsum += NFSX_UNSIGNED;
3031 			break;
3032 		case NFSATTRBIT_OWNER:
3033 			NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
3034 			j = fxdr_unsigned(int, *tl);
3035 			if (j < 0) {
3036 				error = NFSERR_BADXDR;
3037 				goto nfsmout;
3038 			}
3039 			if (j > NFSV4_SMALLSTR)
3040 				cp = malloc(j + 1, M_NFSSTRING, M_WAITOK);
3041 			else
3042 				cp = namestr;
3043 			error = nfsrv_mtostr(nd, cp, j);
3044 			if (error) {
3045 				if (j > NFSV4_SMALLSTR)
3046 					free(cp, M_NFSSTRING);
3047 				goto nfsmout;
3048 			}
3049 			if (!nd->nd_repstat) {
3050 				nd->nd_repstat = nfsv4_strtouid(nd, cp, j,
3051 				    &uid);
3052 				if (!nd->nd_repstat)
3053 					nvap->na_uid = uid;
3054 			}
3055 			if (j > NFSV4_SMALLSTR)
3056 				free(cp, M_NFSSTRING);
3057 			attrsum += (NFSX_UNSIGNED + NFSM_RNDUP(j));
3058 			break;
3059 		case NFSATTRBIT_OWNERGROUP:
3060 			NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
3061 			j = fxdr_unsigned(int, *tl);
3062 			if (j < 0) {
3063 				error = NFSERR_BADXDR;
3064 				goto nfsmout;
3065 			}
3066 			if (j > NFSV4_SMALLSTR)
3067 				cp = malloc(j + 1, M_NFSSTRING, M_WAITOK);
3068 			else
3069 				cp = namestr;
3070 			error = nfsrv_mtostr(nd, cp, j);
3071 			if (error) {
3072 				if (j > NFSV4_SMALLSTR)
3073 					free(cp, M_NFSSTRING);
3074 				goto nfsmout;
3075 			}
3076 			if (!nd->nd_repstat) {
3077 				nd->nd_repstat = nfsv4_strtogid(nd, cp, j,
3078 				    &gid);
3079 				if (!nd->nd_repstat)
3080 					nvap->na_gid = gid;
3081 			}
3082 			if (j > NFSV4_SMALLSTR)
3083 				free(cp, M_NFSSTRING);
3084 			attrsum += (NFSX_UNSIGNED + NFSM_RNDUP(j));
3085 			break;
3086 		case NFSATTRBIT_SYSTEM:
3087 			NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
3088 			if (!nd->nd_repstat)
3089 				nd->nd_repstat = NFSERR_ATTRNOTSUPP;
3090 			attrsum += NFSX_UNSIGNED;
3091 			break;
3092 		case NFSATTRBIT_TIMEACCESSSET:
3093 			NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
3094 			attrsum += NFSX_UNSIGNED;
3095 			if (fxdr_unsigned(int, *tl)==NFSV4SATTRTIME_TOCLIENT) {
3096 			    NFSM_DISSECT(tl, u_int32_t *, NFSX_V4TIME);
3097 			    fxdr_nfsv4time(tl, &nvap->na_atime);
3098 			    toclient = 1;
3099 			    attrsum += NFSX_V4TIME;
3100 			} else {
3101 			    vfs_timestamp(&nvap->na_atime);
3102 			    nvap->na_vaflags |= VA_UTIMES_NULL;
3103 			}
3104 			break;
3105 		case NFSATTRBIT_TIMEBACKUP:
3106 			NFSM_DISSECT(tl, u_int32_t *, NFSX_V4TIME);
3107 			if (!nd->nd_repstat)
3108 				nd->nd_repstat = NFSERR_ATTRNOTSUPP;
3109 			attrsum += NFSX_V4TIME;
3110 			break;
3111 		case NFSATTRBIT_TIMECREATE:
3112 			NFSM_DISSECT(tl, u_int32_t *, NFSX_V4TIME);
3113 			fxdr_nfsv4time(tl, &nvap->na_btime);
3114 			attrsum += NFSX_V4TIME;
3115 			break;
3116 		case NFSATTRBIT_TIMEMODIFYSET:
3117 			NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
3118 			attrsum += NFSX_UNSIGNED;
3119 			if (fxdr_unsigned(int, *tl)==NFSV4SATTRTIME_TOCLIENT) {
3120 			    NFSM_DISSECT(tl, u_int32_t *, NFSX_V4TIME);
3121 			    fxdr_nfsv4time(tl, &nvap->na_mtime);
3122 			    nvap->na_vaflags &= ~VA_UTIMES_NULL;
3123 			    attrsum += NFSX_V4TIME;
3124 			} else {
3125 			    vfs_timestamp(&nvap->na_mtime);
3126 			    if (!toclient)
3127 				nvap->na_vaflags |= VA_UTIMES_NULL;
3128 			}
3129 			break;
3130 		case NFSATTRBIT_MODESETMASKED:
3131 			NFSM_DISSECT(tl, uint32_t *, 2 * NFSX_UNSIGNED);
3132 			mode = fxdr_unsigned(u_short, *tl++);
3133 			mask = fxdr_unsigned(u_short, *tl);
3134 			/*
3135 			 * vp == NULL implies an Open/Create operation.
3136 			 * This attribute can only be used for Setattr and
3137 			 * only for NFSv4.1 or higher.
3138 			 * If moderet != 0, a mode attribute has also been
3139 			 * specified and this attribute cannot be done in the
3140 			 * same Setattr operation.
3141 			 */
3142 			if ((nd->nd_flag & ND_NFSV41) == 0)
3143 				nd->nd_repstat = NFSERR_ATTRNOTSUPP;
3144 			else if ((mode & ~07777) != 0 || (mask & ~07777) != 0 ||
3145 			    vp == NULL)
3146 				nd->nd_repstat = NFSERR_INVAL;
3147 			else if (moderet == 0)
3148 				moderet = VOP_GETATTR(vp, &va, nd->nd_cred);
3149 			if (moderet == 0)
3150 				nvap->na_mode = (mode & mask) |
3151 				    (va.va_mode & ~mask);
3152 			else
3153 				nd->nd_repstat = moderet;
3154 			attrsum += 2 * NFSX_UNSIGNED;
3155 			break;
3156 		default:
3157 			nd->nd_repstat = NFSERR_ATTRNOTSUPP;
3158 			/*
3159 			 * set bitpos so we drop out of the loop.
3160 			 */
3161 			bitpos = NFSATTRBIT_MAX;
3162 			break;
3163 		}
3164 	}
3165 
3166 	/*
3167 	 * some clients pad the attrlist, so we need to skip over the
3168 	 * padding.
3169 	 */
3170 	if (attrsum > attrsize) {
3171 		error = NFSERR_BADXDR;
3172 	} else {
3173 		attrsize = NFSM_RNDUP(attrsize);
3174 		if (attrsum < attrsize)
3175 			error = nfsm_advance(nd, attrsize - attrsum, -1);
3176 	}
3177 nfsmout:
3178 	NFSEXITCODE2(error, nd);
3179 	return (error);
3180 }
3181 
3182 /*
3183  * Check/setup export credentials.
3184  */
3185 int
3186 nfsd_excred(struct nfsrv_descript *nd, struct nfsexstuff *exp,
3187     struct ucred *credanon, bool testsec)
3188 {
3189 	int error;
3190 
3191 	/*
3192 	 * Check/setup credentials.
3193 	 */
3194 	if (nd->nd_flag & ND_GSS)
3195 		exp->nes_exflag &= ~MNT_EXPORTANON;
3196 
3197 	/*
3198 	 * Check to see if the operation is allowed for this security flavor.
3199 	 */
3200 	error = 0;
3201 	if (testsec) {
3202 		error = nfsvno_testexp(nd, exp);
3203 		if (error != 0)
3204 			goto out;
3205 	}
3206 
3207 	/*
3208 	 * Check to see if the file system is exported V4 only.
3209 	 */
3210 	if (NFSVNO_EXV4ONLY(exp) && !(nd->nd_flag & ND_NFSV4)) {
3211 		error = NFSERR_PROGNOTV4;
3212 		goto out;
3213 	}
3214 
3215 	/*
3216 	 * Now, map the user credentials.
3217 	 * (Note that ND_AUTHNONE will only be set for an NFSv3
3218 	 *  Fsinfo RPC. If set for anything else, this code might need
3219 	 *  to change.)
3220 	 */
3221 	if (NFSVNO_EXPORTED(exp)) {
3222 		if (((nd->nd_flag & ND_GSS) == 0 && nd->nd_cred->cr_uid == 0) ||
3223 		     NFSVNO_EXPORTANON(exp) ||
3224 		     (nd->nd_flag & ND_AUTHNONE) != 0) {
3225 			nd->nd_cred->cr_uid = credanon->cr_uid;
3226 			nd->nd_cred->cr_gid = credanon->cr_gid;
3227 			crsetgroups(nd->nd_cred, credanon->cr_ngroups,
3228 			    credanon->cr_groups);
3229 		} else if ((nd->nd_flag & ND_GSS) == 0) {
3230 			/*
3231 			 * If using AUTH_SYS, call nfsrv_getgrpscred() to see
3232 			 * if there is a replacement credential with a group
3233 			 * list set up by "nfsuserd -manage-gids".
3234 			 * If there is no replacement, nfsrv_getgrpscred()
3235 			 * simply returns its argument.
3236 			 */
3237 			nd->nd_cred = nfsrv_getgrpscred(nd->nd_cred);
3238 		}
3239 	}
3240 
3241 out:
3242 	NFSEXITCODE2(error, nd);
3243 	return (error);
3244 }
3245 
3246 /*
3247  * Check exports.
3248  */
3249 int
3250 nfsvno_checkexp(struct mount *mp, struct sockaddr *nam, struct nfsexstuff *exp,
3251     struct ucred **credp)
3252 {
3253 	int error;
3254 
3255 	error = VFS_CHECKEXP(mp, nam, &exp->nes_exflag, credp,
3256 	    &exp->nes_numsecflavor, exp->nes_secflavors);
3257 	if (error) {
3258 		if (nfs_rootfhset) {
3259 			exp->nes_exflag = 0;
3260 			exp->nes_numsecflavor = 0;
3261 			error = 0;
3262 		}
3263 	} else if (exp->nes_numsecflavor < 1 || exp->nes_numsecflavor >
3264 	    MAXSECFLAVORS) {
3265 		printf("nfsvno_checkexp: numsecflavors out of range\n");
3266 		exp->nes_numsecflavor = 0;
3267 		error = EACCES;
3268 	}
3269 	NFSEXITCODE(error);
3270 	return (error);
3271 }
3272 
3273 /*
3274  * Get a vnode for a file handle and export stuff.
3275  */
3276 int
3277 nfsvno_fhtovp(struct mount *mp, fhandle_t *fhp, struct sockaddr *nam,
3278     int lktype, struct vnode **vpp, struct nfsexstuff *exp,
3279     struct ucred **credp)
3280 {
3281 	int error;
3282 
3283 	*credp = NULL;
3284 	exp->nes_numsecflavor = 0;
3285 	error = VFS_FHTOVP(mp, &fhp->fh_fid, lktype, vpp);
3286 	if (error != 0)
3287 		/* Make sure the server replies ESTALE to the client. */
3288 		error = ESTALE;
3289 	if (nam && !error) {
3290 		error = VFS_CHECKEXP(mp, nam, &exp->nes_exflag, credp,
3291 		    &exp->nes_numsecflavor, exp->nes_secflavors);
3292 		if (error) {
3293 			if (nfs_rootfhset) {
3294 				exp->nes_exflag = 0;
3295 				exp->nes_numsecflavor = 0;
3296 				error = 0;
3297 			} else {
3298 				vput(*vpp);
3299 			}
3300 		} else if (exp->nes_numsecflavor < 1 || exp->nes_numsecflavor >
3301 		    MAXSECFLAVORS) {
3302 			printf("nfsvno_fhtovp: numsecflavors out of range\n");
3303 			exp->nes_numsecflavor = 0;
3304 			error = EACCES;
3305 			vput(*vpp);
3306 		}
3307 	}
3308 	NFSEXITCODE(error);
3309 	return (error);
3310 }
3311 
3312 /*
3313  * nfsd_fhtovp() - convert a fh to a vnode ptr
3314  * 	- look up fsid in mount list (if not found ret error)
3315  *	- get vp and export rights by calling nfsvno_fhtovp()
3316  *	- if cred->cr_uid == 0 or MNT_EXPORTANON set it to credanon
3317  *	  for AUTH_SYS
3318  *	- if mpp != NULL, return the mount point so that it can
3319  *	  be used for vn_finished_write() by the caller
3320  */
3321 void
3322 nfsd_fhtovp(struct nfsrv_descript *nd, struct nfsrvfh *nfp, int lktype,
3323     struct vnode **vpp, struct nfsexstuff *exp,
3324     struct mount **mpp, int startwrite, int nextop)
3325 {
3326 	struct mount *mp, *mpw;
3327 	struct ucred *credanon;
3328 	fhandle_t *fhp;
3329 	int error;
3330 
3331 	if (mpp != NULL)
3332 		*mpp = NULL;
3333 	*vpp = NULL;
3334 	fhp = (fhandle_t *)nfp->nfsrvfh_data;
3335 	mp = vfs_busyfs(&fhp->fh_fsid);
3336 	if (mp == NULL) {
3337 		nd->nd_repstat = ESTALE;
3338 		goto out;
3339 	}
3340 
3341 	if (startwrite) {
3342 		mpw = mp;
3343 		error = vn_start_write(NULL, &mpw, V_WAIT);
3344 		if (error != 0) {
3345 			mpw = NULL;
3346 			vfs_unbusy(mp);
3347 			nd->nd_repstat = ESTALE;
3348 			goto out;
3349 		}
3350 		if (lktype == LK_SHARED && !(MNT_SHARED_WRITES(mp)))
3351 			lktype = LK_EXCLUSIVE;
3352 	} else
3353 		mpw = NULL;
3354 
3355 	nd->nd_repstat = nfsvno_fhtovp(mp, fhp, nd->nd_nam, lktype, vpp, exp,
3356 	    &credanon);
3357 	vfs_unbusy(mp);
3358 
3359 	/*
3360 	 * For NFSv4 without a pseudo root fs, unexported file handles
3361 	 * can be returned, so that Lookup works everywhere.
3362 	 */
3363 	if (!nd->nd_repstat && exp->nes_exflag == 0 &&
3364 	    !(nd->nd_flag & ND_NFSV4)) {
3365 		vput(*vpp);
3366 		*vpp = NULL;
3367 		nd->nd_repstat = EACCES;
3368 	}
3369 
3370 	/*
3371 	 * Personally, I've never seen any point in requiring a
3372 	 * reserved port#, since only in the rare case where the
3373 	 * clients are all boxes with secure system privileges,
3374 	 * does it provide any enhanced security, but... some people
3375 	 * believe it to be useful and keep putting this code back in.
3376 	 * (There is also some "security checker" out there that
3377 	 *  complains if the nfs server doesn't enforce this.)
3378 	 * However, note the following:
3379 	 * RFC3530 (NFSv4) specifies that a reserved port# not be
3380 	 *	required.
3381 	 * RFC2623 recommends that, if a reserved port# is checked for,
3382 	 *	that there be a way to turn that off--> ifdef'd.
3383 	 */
3384 #ifdef NFS_REQRSVPORT
3385 	if (!nd->nd_repstat) {
3386 		struct sockaddr_in *saddr;
3387 		struct sockaddr_in6 *saddr6;
3388 
3389 		saddr = NFSSOCKADDR(nd->nd_nam, struct sockaddr_in *);
3390 		saddr6 = NFSSOCKADDR(nd->nd_nam, struct sockaddr_in6 *);
3391 		if (!(nd->nd_flag & ND_NFSV4) &&
3392 		    ((saddr->sin_family == AF_INET &&
3393 		      ntohs(saddr->sin_port) >= IPPORT_RESERVED) ||
3394 		     (saddr6->sin6_family == AF_INET6 &&
3395 		      ntohs(saddr6->sin6_port) >= IPPORT_RESERVED))) {
3396 			vput(*vpp);
3397 			nd->nd_repstat = (NFSERR_AUTHERR | AUTH_TOOWEAK);
3398 		}
3399 	}
3400 #endif	/* NFS_REQRSVPORT */
3401 
3402 	/*
3403 	 * Check/setup credentials.
3404 	 */
3405 	if (!nd->nd_repstat) {
3406 		nd->nd_saveduid = nd->nd_cred->cr_uid;
3407 		nd->nd_repstat = nfsd_excred(nd, exp, credanon,
3408 		    nfsrv_checkwrongsec(nd, nextop, (*vpp)->v_type));
3409 		if (nd->nd_repstat)
3410 			vput(*vpp);
3411 	}
3412 	if (credanon != NULL)
3413 		crfree(credanon);
3414 	if (nd->nd_repstat) {
3415 		vn_finished_write(mpw);
3416 		*vpp = NULL;
3417 	} else if (mpp != NULL) {
3418 		*mpp = mpw;
3419 	}
3420 
3421 out:
3422 	NFSEXITCODE2(0, nd);
3423 }
3424 
3425 /*
3426  * glue for fp.
3427  */
3428 static int
3429 fp_getfvp(struct thread *p, int fd, struct file **fpp, struct vnode **vpp)
3430 {
3431 	struct filedesc *fdp;
3432 	struct file *fp;
3433 	int error = 0;
3434 
3435 	fdp = p->td_proc->p_fd;
3436 	if (fd < 0 || fd >= fdp->fd_nfiles ||
3437 	    (fp = fdp->fd_ofiles[fd].fde_file) == NULL) {
3438 		error = EBADF;
3439 		goto out;
3440 	}
3441 	*fpp = fp;
3442 
3443 out:
3444 	NFSEXITCODE(error);
3445 	return (error);
3446 }
3447 
3448 /*
3449  * Called from nfssvc() to update the exports list. Just call
3450  * vfs_export(). This has to be done, since the v4 root fake fs isn't
3451  * in the mount list.
3452  */
3453 int
3454 nfsrv_v4rootexport(void *argp, struct ucred *cred, struct thread *p)
3455 {
3456 	struct nfsex_args *nfsexargp = (struct nfsex_args *)argp;
3457 	int error = 0;
3458 	struct nameidata nd;
3459 	fhandle_t fh;
3460 
3461 	error = vfs_export(&nfsv4root_mnt, &nfsexargp->export);
3462 	if ((nfsexargp->export.ex_flags & MNT_DELEXPORT) != 0)
3463 		nfs_rootfhset = 0;
3464 	else if (error == 0) {
3465 		if (nfsexargp->fspec == NULL) {
3466 			error = EPERM;
3467 			goto out;
3468 		}
3469 		/*
3470 		 * If fspec != NULL, this is the v4root path.
3471 		 */
3472 		NDINIT(&nd, LOOKUP, FOLLOW, UIO_USERSPACE,
3473 		    nfsexargp->fspec, p);
3474 		if ((error = namei(&nd)) != 0)
3475 			goto out;
3476 		error = nfsvno_getfh(nd.ni_vp, &fh, p);
3477 		vrele(nd.ni_vp);
3478 		if (!error) {
3479 			nfs_rootfh.nfsrvfh_len = NFSX_MYFH;
3480 			NFSBCOPY((caddr_t)&fh,
3481 			    nfs_rootfh.nfsrvfh_data,
3482 			    sizeof (fhandle_t));
3483 			nfs_rootfhset = 1;
3484 		}
3485 	}
3486 
3487 out:
3488 	NFSEXITCODE(error);
3489 	return (error);
3490 }
3491 
3492 /*
3493  * This function needs to test to see if the system is near its limit
3494  * for memory allocation via malloc() or mget() and return True iff
3495  * either of these resources are near their limit.
3496  * XXX (For now, this is just a stub.)
3497  */
3498 int nfsrv_testmalloclimit = 0;
3499 int
3500 nfsrv_mallocmget_limit(void)
3501 {
3502 	static int printmesg = 0;
3503 	static int testval = 1;
3504 
3505 	if (nfsrv_testmalloclimit && (testval++ % 1000) == 0) {
3506 		if ((printmesg++ % 100) == 0)
3507 			printf("nfsd: malloc/mget near limit\n");
3508 		return (1);
3509 	}
3510 	return (0);
3511 }
3512 
3513 /*
3514  * BSD specific initialization of a mount point.
3515  */
3516 void
3517 nfsd_mntinit(void)
3518 {
3519 	static int inited = 0;
3520 
3521 	if (inited)
3522 		return;
3523 	inited = 1;
3524 	nfsv4root_mnt.mnt_flag = (MNT_RDONLY | MNT_EXPORTED);
3525 	TAILQ_INIT(&nfsv4root_mnt.mnt_nvnodelist);
3526 	TAILQ_INIT(&nfsv4root_mnt.mnt_lazyvnodelist);
3527 	nfsv4root_mnt.mnt_export = NULL;
3528 	TAILQ_INIT(&nfsv4root_opt);
3529 	TAILQ_INIT(&nfsv4root_newopt);
3530 	nfsv4root_mnt.mnt_opt = &nfsv4root_opt;
3531 	nfsv4root_mnt.mnt_optnew = &nfsv4root_newopt;
3532 	nfsv4root_mnt.mnt_nvnodelistsize = 0;
3533 	nfsv4root_mnt.mnt_lazyvnodelistsize = 0;
3534 }
3535 
3536 /*
3537  * Get a vnode for a file handle, without checking exports, etc.
3538  */
3539 struct vnode *
3540 nfsvno_getvp(fhandle_t *fhp)
3541 {
3542 	struct mount *mp;
3543 	struct vnode *vp;
3544 	int error;
3545 
3546 	mp = vfs_busyfs(&fhp->fh_fsid);
3547 	if (mp == NULL)
3548 		return (NULL);
3549 	error = VFS_FHTOVP(mp, &fhp->fh_fid, LK_EXCLUSIVE, &vp);
3550 	vfs_unbusy(mp);
3551 	if (error)
3552 		return (NULL);
3553 	return (vp);
3554 }
3555 
3556 /*
3557  * Do a local VOP_ADVLOCK().
3558  */
3559 int
3560 nfsvno_advlock(struct vnode *vp, int ftype, u_int64_t first,
3561     u_int64_t end, struct thread *td)
3562 {
3563 	int error = 0;
3564 	struct flock fl;
3565 	u_int64_t tlen;
3566 
3567 	if (nfsrv_dolocallocks == 0)
3568 		goto out;
3569 	ASSERT_VOP_UNLOCKED(vp, "nfsvno_advlock: vp locked");
3570 
3571 	fl.l_whence = SEEK_SET;
3572 	fl.l_type = ftype;
3573 	fl.l_start = (off_t)first;
3574 	if (end == NFS64BITSSET) {
3575 		fl.l_len = 0;
3576 	} else {
3577 		tlen = end - first;
3578 		fl.l_len = (off_t)tlen;
3579 	}
3580 	/*
3581 	 * For FreeBSD8, the l_pid and l_sysid must be set to the same
3582 	 * values for all calls, so that all locks will be held by the
3583 	 * nfsd server. (The nfsd server handles conflicts between the
3584 	 * various clients.)
3585 	 * Since an NFSv4 lockowner is a ClientID plus an array of up to 1024
3586 	 * bytes, so it can't be put in l_sysid.
3587 	 */
3588 	if (nfsv4_sysid == 0)
3589 		nfsv4_sysid = nlm_acquire_next_sysid();
3590 	fl.l_pid = (pid_t)0;
3591 	fl.l_sysid = (int)nfsv4_sysid;
3592 
3593 	if (ftype == F_UNLCK)
3594 		error = VOP_ADVLOCK(vp, (caddr_t)td->td_proc, F_UNLCK, &fl,
3595 		    (F_POSIX | F_REMOTE));
3596 	else
3597 		error = VOP_ADVLOCK(vp, (caddr_t)td->td_proc, F_SETLK, &fl,
3598 		    (F_POSIX | F_REMOTE));
3599 
3600 out:
3601 	NFSEXITCODE(error);
3602 	return (error);
3603 }
3604 
3605 /*
3606  * Check the nfsv4 root exports.
3607  */
3608 int
3609 nfsvno_v4rootexport(struct nfsrv_descript *nd)
3610 {
3611 	struct ucred *credanon;
3612 	int error = 0, numsecflavor, secflavors[MAXSECFLAVORS], i;
3613 	uint64_t exflags;
3614 
3615 	error = vfs_stdcheckexp(&nfsv4root_mnt, nd->nd_nam, &exflags,
3616 	    &credanon, &numsecflavor, secflavors);
3617 	if (error) {
3618 		error = NFSERR_PROGUNAVAIL;
3619 		goto out;
3620 	}
3621 	if (credanon != NULL)
3622 		crfree(credanon);
3623 	for (i = 0; i < numsecflavor; i++) {
3624 		if (secflavors[i] == AUTH_SYS)
3625 			nd->nd_flag |= ND_EXAUTHSYS;
3626 		else if (secflavors[i] == RPCSEC_GSS_KRB5)
3627 			nd->nd_flag |= ND_EXGSS;
3628 		else if (secflavors[i] == RPCSEC_GSS_KRB5I)
3629 			nd->nd_flag |= ND_EXGSSINTEGRITY;
3630 		else if (secflavors[i] == RPCSEC_GSS_KRB5P)
3631 			nd->nd_flag |= ND_EXGSSPRIVACY;
3632 	}
3633 
3634 	/* And set ND_EXxx flags for TLS. */
3635 	if ((exflags & MNT_EXTLS) != 0) {
3636 		nd->nd_flag |= ND_EXTLS;
3637 		if ((exflags & MNT_EXTLSCERT) != 0)
3638 			nd->nd_flag |= ND_EXTLSCERT;
3639 		if ((exflags & MNT_EXTLSCERTUSER) != 0)
3640 			nd->nd_flag |= ND_EXTLSCERTUSER;
3641 	}
3642 
3643 out:
3644 	NFSEXITCODE(error);
3645 	return (error);
3646 }
3647 
3648 /*
3649  * Nfs server pseudo system call for the nfsd's
3650  */
3651 /*
3652  * MPSAFE
3653  */
3654 static int
3655 nfssvc_nfsd(struct thread *td, struct nfssvc_args *uap)
3656 {
3657 	struct file *fp;
3658 	struct nfsd_addsock_args sockarg;
3659 	struct nfsd_nfsd_args nfsdarg;
3660 	struct nfsd_nfsd_oargs onfsdarg;
3661 	struct nfsd_pnfsd_args pnfsdarg;
3662 	struct vnode *vp, *nvp, *curdvp;
3663 	struct pnfsdsfile *pf;
3664 	struct nfsdevice *ds, *fds;
3665 	cap_rights_t rights;
3666 	int buflen, error, ret;
3667 	char *buf, *cp, *cp2, *cp3;
3668 	char fname[PNFS_FILENAME_LEN + 1];
3669 
3670 	if (uap->flag & NFSSVC_NFSDADDSOCK) {
3671 		error = copyin(uap->argp, (caddr_t)&sockarg, sizeof (sockarg));
3672 		if (error)
3673 			goto out;
3674 		/*
3675 		 * Since we don't know what rights might be required,
3676 		 * pretend that we need them all. It is better to be too
3677 		 * careful than too reckless.
3678 		 */
3679 		error = fget(td, sockarg.sock,
3680 		    cap_rights_init_one(&rights, CAP_SOCK_SERVER), &fp);
3681 		if (error != 0)
3682 			goto out;
3683 		if (fp->f_type != DTYPE_SOCKET) {
3684 			fdrop(fp, td);
3685 			error = EPERM;
3686 			goto out;
3687 		}
3688 		error = nfsrvd_addsock(fp);
3689 		fdrop(fp, td);
3690 	} else if (uap->flag & NFSSVC_NFSDNFSD) {
3691 		if (uap->argp == NULL) {
3692 			error = EINVAL;
3693 			goto out;
3694 		}
3695 		if ((uap->flag & NFSSVC_NEWSTRUCT) == 0) {
3696 			error = copyin(uap->argp, &onfsdarg, sizeof(onfsdarg));
3697 			if (error == 0) {
3698 				nfsdarg.principal = onfsdarg.principal;
3699 				nfsdarg.minthreads = onfsdarg.minthreads;
3700 				nfsdarg.maxthreads = onfsdarg.maxthreads;
3701 				nfsdarg.version = 1;
3702 				nfsdarg.addr = NULL;
3703 				nfsdarg.addrlen = 0;
3704 				nfsdarg.dnshost = NULL;
3705 				nfsdarg.dnshostlen = 0;
3706 				nfsdarg.dspath = NULL;
3707 				nfsdarg.dspathlen = 0;
3708 				nfsdarg.mdspath = NULL;
3709 				nfsdarg.mdspathlen = 0;
3710 				nfsdarg.mirrorcnt = 1;
3711 			}
3712 		} else
3713 			error = copyin(uap->argp, &nfsdarg, sizeof(nfsdarg));
3714 		if (error)
3715 			goto out;
3716 		if (nfsdarg.addrlen > 0 && nfsdarg.addrlen < 10000 &&
3717 		    nfsdarg.dnshostlen > 0 && nfsdarg.dnshostlen < 10000 &&
3718 		    nfsdarg.dspathlen > 0 && nfsdarg.dspathlen < 10000 &&
3719 		    nfsdarg.mdspathlen > 0 && nfsdarg.mdspathlen < 10000 &&
3720 		    nfsdarg.mirrorcnt >= 1 &&
3721 		    nfsdarg.mirrorcnt <= NFSDEV_MAXMIRRORS &&
3722 		    nfsdarg.addr != NULL && nfsdarg.dnshost != NULL &&
3723 		    nfsdarg.dspath != NULL && nfsdarg.mdspath != NULL) {
3724 			NFSD_DEBUG(1, "addrlen=%d dspathlen=%d dnslen=%d"
3725 			    " mdspathlen=%d mirrorcnt=%d\n", nfsdarg.addrlen,
3726 			    nfsdarg.dspathlen, nfsdarg.dnshostlen,
3727 			    nfsdarg.mdspathlen, nfsdarg.mirrorcnt);
3728 			cp = malloc(nfsdarg.addrlen + 1, M_TEMP, M_WAITOK);
3729 			error = copyin(nfsdarg.addr, cp, nfsdarg.addrlen);
3730 			if (error != 0) {
3731 				free(cp, M_TEMP);
3732 				goto out;
3733 			}
3734 			cp[nfsdarg.addrlen] = '\0';	/* Ensure nul term. */
3735 			nfsdarg.addr = cp;
3736 			cp = malloc(nfsdarg.dnshostlen + 1, M_TEMP, M_WAITOK);
3737 			error = copyin(nfsdarg.dnshost, cp, nfsdarg.dnshostlen);
3738 			if (error != 0) {
3739 				free(nfsdarg.addr, M_TEMP);
3740 				free(cp, M_TEMP);
3741 				goto out;
3742 			}
3743 			cp[nfsdarg.dnshostlen] = '\0';	/* Ensure nul term. */
3744 			nfsdarg.dnshost = cp;
3745 			cp = malloc(nfsdarg.dspathlen + 1, M_TEMP, M_WAITOK);
3746 			error = copyin(nfsdarg.dspath, cp, nfsdarg.dspathlen);
3747 			if (error != 0) {
3748 				free(nfsdarg.addr, M_TEMP);
3749 				free(nfsdarg.dnshost, M_TEMP);
3750 				free(cp, M_TEMP);
3751 				goto out;
3752 			}
3753 			cp[nfsdarg.dspathlen] = '\0';	/* Ensure nul term. */
3754 			nfsdarg.dspath = cp;
3755 			cp = malloc(nfsdarg.mdspathlen + 1, M_TEMP, M_WAITOK);
3756 			error = copyin(nfsdarg.mdspath, cp, nfsdarg.mdspathlen);
3757 			if (error != 0) {
3758 				free(nfsdarg.addr, M_TEMP);
3759 				free(nfsdarg.dnshost, M_TEMP);
3760 				free(nfsdarg.dspath, M_TEMP);
3761 				free(cp, M_TEMP);
3762 				goto out;
3763 			}
3764 			cp[nfsdarg.mdspathlen] = '\0';	/* Ensure nul term. */
3765 			nfsdarg.mdspath = cp;
3766 		} else {
3767 			nfsdarg.addr = NULL;
3768 			nfsdarg.addrlen = 0;
3769 			nfsdarg.dnshost = NULL;
3770 			nfsdarg.dnshostlen = 0;
3771 			nfsdarg.dspath = NULL;
3772 			nfsdarg.dspathlen = 0;
3773 			nfsdarg.mdspath = NULL;
3774 			nfsdarg.mdspathlen = 0;
3775 			nfsdarg.mirrorcnt = 1;
3776 		}
3777 		error = nfsrvd_nfsd(td, &nfsdarg);
3778 		free(nfsdarg.addr, M_TEMP);
3779 		free(nfsdarg.dnshost, M_TEMP);
3780 		free(nfsdarg.dspath, M_TEMP);
3781 		free(nfsdarg.mdspath, M_TEMP);
3782 	} else if (uap->flag & NFSSVC_PNFSDS) {
3783 		error = copyin(uap->argp, &pnfsdarg, sizeof(pnfsdarg));
3784 		if (error == 0 && (pnfsdarg.op == PNFSDOP_DELDSSERVER ||
3785 		    pnfsdarg.op == PNFSDOP_FORCEDELDS)) {
3786 			cp = malloc(PATH_MAX + 1, M_TEMP, M_WAITOK);
3787 			error = copyinstr(pnfsdarg.dspath, cp, PATH_MAX + 1,
3788 			    NULL);
3789 			if (error == 0)
3790 				error = nfsrv_deldsserver(pnfsdarg.op, cp, td);
3791 			free(cp, M_TEMP);
3792 		} else if (error == 0 && pnfsdarg.op == PNFSDOP_COPYMR) {
3793 			cp = malloc(PATH_MAX + 1, M_TEMP, M_WAITOK);
3794 			buflen = sizeof(*pf) * NFSDEV_MAXMIRRORS;
3795 			buf = malloc(buflen, M_TEMP, M_WAITOK);
3796 			error = copyinstr(pnfsdarg.mdspath, cp, PATH_MAX + 1,
3797 			    NULL);
3798 			NFSD_DEBUG(4, "pnfsdcopymr cp mdspath=%d\n", error);
3799 			if (error == 0 && pnfsdarg.dspath != NULL) {
3800 				cp2 = malloc(PATH_MAX + 1, M_TEMP, M_WAITOK);
3801 				error = copyinstr(pnfsdarg.dspath, cp2,
3802 				    PATH_MAX + 1, NULL);
3803 				NFSD_DEBUG(4, "pnfsdcopymr cp dspath=%d\n",
3804 				    error);
3805 			} else
3806 				cp2 = NULL;
3807 			if (error == 0 && pnfsdarg.curdspath != NULL) {
3808 				cp3 = malloc(PATH_MAX + 1, M_TEMP, M_WAITOK);
3809 				error = copyinstr(pnfsdarg.curdspath, cp3,
3810 				    PATH_MAX + 1, NULL);
3811 				NFSD_DEBUG(4, "pnfsdcopymr cp curdspath=%d\n",
3812 				    error);
3813 			} else
3814 				cp3 = NULL;
3815 			curdvp = NULL;
3816 			fds = NULL;
3817 			if (error == 0)
3818 				error = nfsrv_mdscopymr(cp, cp2, cp3, buf,
3819 				    &buflen, fname, td, &vp, &nvp, &pf, &ds,
3820 				    &fds);
3821 			NFSD_DEBUG(4, "nfsrv_mdscopymr=%d\n", error);
3822 			if (error == 0) {
3823 				if (pf->dsf_dir >= nfsrv_dsdirsize) {
3824 					printf("copymr: dsdir out of range\n");
3825 					pf->dsf_dir = 0;
3826 				}
3827 				NFSD_DEBUG(4, "copymr: buflen=%d\n", buflen);
3828 				error = nfsrv_copymr(vp, nvp,
3829 				    ds->nfsdev_dsdir[pf->dsf_dir], ds, pf,
3830 				    (struct pnfsdsfile *)buf,
3831 				    buflen / sizeof(*pf), td->td_ucred, td);
3832 				vput(vp);
3833 				vput(nvp);
3834 				if (fds != NULL && error == 0) {
3835 					curdvp = fds->nfsdev_dsdir[pf->dsf_dir];
3836 					ret = vn_lock(curdvp, LK_EXCLUSIVE);
3837 					if (ret == 0) {
3838 						nfsrv_dsremove(curdvp, fname,
3839 						    td->td_ucred, td);
3840 						NFSVOPUNLOCK(curdvp);
3841 					}
3842 				}
3843 				NFSD_DEBUG(4, "nfsrv_copymr=%d\n", error);
3844 			}
3845 			free(cp, M_TEMP);
3846 			free(cp2, M_TEMP);
3847 			free(cp3, M_TEMP);
3848 			free(buf, M_TEMP);
3849 		}
3850 	} else {
3851 		error = nfssvc_srvcall(td, uap, td->td_ucred);
3852 	}
3853 
3854 out:
3855 	NFSEXITCODE(error);
3856 	return (error);
3857 }
3858 
3859 static int
3860 nfssvc_srvcall(struct thread *p, struct nfssvc_args *uap, struct ucred *cred)
3861 {
3862 	struct nfsex_args export;
3863 	struct nfsex_oldargs oexp;
3864 	struct file *fp = NULL;
3865 	int stablefd, i, len;
3866 	struct nfsd_clid adminrevoke;
3867 	struct nfsd_dumplist dumplist;
3868 	struct nfsd_dumpclients *dumpclients;
3869 	struct nfsd_dumplocklist dumplocklist;
3870 	struct nfsd_dumplocks *dumplocks;
3871 	struct nameidata nd;
3872 	vnode_t vp;
3873 	int error = EINVAL, igotlock;
3874 	struct proc *procp;
3875 	gid_t *grps;
3876 	static int suspend_nfsd = 0;
3877 
3878 	if (uap->flag & NFSSVC_PUBLICFH) {
3879 		NFSBZERO((caddr_t)&nfs_pubfh.nfsrvfh_data,
3880 		    sizeof (fhandle_t));
3881 		error = copyin(uap->argp,
3882 		    &nfs_pubfh.nfsrvfh_data, sizeof (fhandle_t));
3883 		if (!error)
3884 			nfs_pubfhset = 1;
3885 	} else if ((uap->flag & (NFSSVC_V4ROOTEXPORT | NFSSVC_NEWSTRUCT)) ==
3886 	    (NFSSVC_V4ROOTEXPORT | NFSSVC_NEWSTRUCT)) {
3887 		error = copyin(uap->argp,(caddr_t)&export,
3888 		    sizeof (struct nfsex_args));
3889 		if (!error) {
3890 			grps = NULL;
3891 			if (export.export.ex_ngroups > NGROUPS_MAX ||
3892 			    export.export.ex_ngroups < 0)
3893 				error = EINVAL;
3894 			else if (export.export.ex_ngroups > 0) {
3895 				grps = malloc(export.export.ex_ngroups *
3896 				    sizeof(gid_t), M_TEMP, M_WAITOK);
3897 				error = copyin(export.export.ex_groups, grps,
3898 				    export.export.ex_ngroups * sizeof(gid_t));
3899 				export.export.ex_groups = grps;
3900 			} else
3901 				export.export.ex_groups = NULL;
3902 			if (!error)
3903 				error = nfsrv_v4rootexport(&export, cred, p);
3904 			free(grps, M_TEMP);
3905 		}
3906 	} else if ((uap->flag & (NFSSVC_V4ROOTEXPORT | NFSSVC_NEWSTRUCT)) ==
3907 	    NFSSVC_V4ROOTEXPORT) {
3908 		error = copyin(uap->argp,(caddr_t)&oexp,
3909 		    sizeof (struct nfsex_oldargs));
3910 		if (!error) {
3911 			memset(&export.export, 0, sizeof(export.export));
3912 			export.export.ex_flags = (uint64_t)oexp.export.ex_flags;
3913 			export.export.ex_root = oexp.export.ex_root;
3914 			export.export.ex_uid = oexp.export.ex_anon.cr_uid;
3915 			export.export.ex_ngroups =
3916 			    oexp.export.ex_anon.cr_ngroups;
3917 			export.export.ex_groups = NULL;
3918 			if (export.export.ex_ngroups > XU_NGROUPS ||
3919 			    export.export.ex_ngroups < 0)
3920 				error = EINVAL;
3921 			else if (export.export.ex_ngroups > 0) {
3922 				export.export.ex_groups = malloc(
3923 				    export.export.ex_ngroups * sizeof(gid_t),
3924 				    M_TEMP, M_WAITOK);
3925 				for (i = 0; i < export.export.ex_ngroups; i++)
3926 					export.export.ex_groups[i] =
3927 					    oexp.export.ex_anon.cr_groups[i];
3928 			}
3929 			export.export.ex_addr = oexp.export.ex_addr;
3930 			export.export.ex_addrlen = oexp.export.ex_addrlen;
3931 			export.export.ex_mask = oexp.export.ex_mask;
3932 			export.export.ex_masklen = oexp.export.ex_masklen;
3933 			export.export.ex_indexfile = oexp.export.ex_indexfile;
3934 			export.export.ex_numsecflavors =
3935 			    oexp.export.ex_numsecflavors;
3936 			if (export.export.ex_numsecflavors >= MAXSECFLAVORS ||
3937 			    export.export.ex_numsecflavors < 0)
3938 				error = EINVAL;
3939 			else {
3940 				for (i = 0; i < export.export.ex_numsecflavors;
3941 				    i++)
3942 					export.export.ex_secflavors[i] =
3943 					    oexp.export.ex_secflavors[i];
3944 			}
3945 			export.fspec = oexp.fspec;
3946 			if (error == 0)
3947 				error = nfsrv_v4rootexport(&export, cred, p);
3948 			free(export.export.ex_groups, M_TEMP);
3949 		}
3950 	} else if (uap->flag & NFSSVC_NOPUBLICFH) {
3951 		nfs_pubfhset = 0;
3952 		error = 0;
3953 	} else if (uap->flag & NFSSVC_STABLERESTART) {
3954 		error = copyin(uap->argp, (caddr_t)&stablefd,
3955 		    sizeof (int));
3956 		if (!error)
3957 			error = fp_getfvp(p, stablefd, &fp, &vp);
3958 		if (!error && (NFSFPFLAG(fp) & (FREAD | FWRITE)) != (FREAD | FWRITE))
3959 			error = EBADF;
3960 		if (!error && newnfs_numnfsd != 0)
3961 			error = EPERM;
3962 		if (!error) {
3963 			nfsrv_stablefirst.nsf_fp = fp;
3964 			nfsrv_setupstable(p);
3965 		}
3966 	} else if (uap->flag & NFSSVC_ADMINREVOKE) {
3967 		error = copyin(uap->argp, (caddr_t)&adminrevoke,
3968 		    sizeof (struct nfsd_clid));
3969 		if (!error)
3970 			error = nfsrv_adminrevoke(&adminrevoke, p);
3971 	} else if (uap->flag & NFSSVC_DUMPCLIENTS) {
3972 		error = copyin(uap->argp, (caddr_t)&dumplist,
3973 		    sizeof (struct nfsd_dumplist));
3974 		if (!error && (dumplist.ndl_size < 1 ||
3975 			dumplist.ndl_size > NFSRV_MAXDUMPLIST))
3976 			error = EPERM;
3977 		if (!error) {
3978 		    len = sizeof (struct nfsd_dumpclients) * dumplist.ndl_size;
3979 		    dumpclients = malloc(len, M_TEMP, M_WAITOK | M_ZERO);
3980 		    nfsrv_dumpclients(dumpclients, dumplist.ndl_size);
3981 		    error = copyout(dumpclients, dumplist.ndl_list, len);
3982 		    free(dumpclients, M_TEMP);
3983 		}
3984 	} else if (uap->flag & NFSSVC_DUMPLOCKS) {
3985 		error = copyin(uap->argp, (caddr_t)&dumplocklist,
3986 		    sizeof (struct nfsd_dumplocklist));
3987 		if (!error && (dumplocklist.ndllck_size < 1 ||
3988 			dumplocklist.ndllck_size > NFSRV_MAXDUMPLIST))
3989 			error = EPERM;
3990 		if (!error)
3991 			error = nfsrv_lookupfilename(&nd,
3992 				dumplocklist.ndllck_fname, p);
3993 		if (!error) {
3994 			len = sizeof (struct nfsd_dumplocks) *
3995 				dumplocklist.ndllck_size;
3996 			dumplocks = malloc(len, M_TEMP, M_WAITOK | M_ZERO);
3997 			nfsrv_dumplocks(nd.ni_vp, dumplocks,
3998 			    dumplocklist.ndllck_size, p);
3999 			vput(nd.ni_vp);
4000 			error = copyout(dumplocks, dumplocklist.ndllck_list,
4001 			    len);
4002 			free(dumplocks, M_TEMP);
4003 		}
4004 	} else if (uap->flag & NFSSVC_BACKUPSTABLE) {
4005 		procp = p->td_proc;
4006 		PROC_LOCK(procp);
4007 		nfsd_master_pid = procp->p_pid;
4008 		bcopy(procp->p_comm, nfsd_master_comm, MAXCOMLEN + 1);
4009 		nfsd_master_start = procp->p_stats->p_start;
4010 		nfsd_master_proc = procp;
4011 		PROC_UNLOCK(procp);
4012 	} else if ((uap->flag & NFSSVC_SUSPENDNFSD) != 0) {
4013 		NFSLOCKV4ROOTMUTEX();
4014 		if (suspend_nfsd == 0) {
4015 			/* Lock out all nfsd threads */
4016 			do {
4017 				igotlock = nfsv4_lock(&nfsd_suspend_lock, 1,
4018 				    NULL, NFSV4ROOTLOCKMUTEXPTR, NULL);
4019 			} while (igotlock == 0 && suspend_nfsd == 0);
4020 			suspend_nfsd = 1;
4021 		}
4022 		NFSUNLOCKV4ROOTMUTEX();
4023 		error = 0;
4024 	} else if ((uap->flag & NFSSVC_RESUMENFSD) != 0) {
4025 		NFSLOCKV4ROOTMUTEX();
4026 		if (suspend_nfsd != 0) {
4027 			nfsv4_unlock(&nfsd_suspend_lock, 0);
4028 			suspend_nfsd = 0;
4029 		}
4030 		NFSUNLOCKV4ROOTMUTEX();
4031 		error = 0;
4032 	}
4033 
4034 	NFSEXITCODE(error);
4035 	return (error);
4036 }
4037 
4038 /*
4039  * Check exports.
4040  * Returns 0 if ok, 1 otherwise.
4041  */
4042 int
4043 nfsvno_testexp(struct nfsrv_descript *nd, struct nfsexstuff *exp)
4044 {
4045 	int i;
4046 
4047 	/*
4048 	 * Allow NFSv3 Fsinfo per RFC2623.
4049 	 */
4050 	if (((nd->nd_flag & ND_NFSV4) != 0 ||
4051 	     nd->nd_procnum != NFSPROC_FSINFO) &&
4052 	    ((NFSVNO_EXTLS(exp) && (nd->nd_flag & ND_TLS) == 0) ||
4053 	     (NFSVNO_EXTLSCERT(exp) &&
4054 	      (nd->nd_flag & ND_TLSCERT) == 0) ||
4055 	     (NFSVNO_EXTLSCERTUSER(exp) &&
4056 	      (nd->nd_flag & ND_TLSCERTUSER) == 0))) {
4057 		if ((nd->nd_flag & ND_NFSV4) != 0)
4058 			return (NFSERR_WRONGSEC);
4059 		else if ((nd->nd_flag & ND_TLS) == 0)
4060 			return (NFSERR_AUTHERR | AUTH_NEEDS_TLS);
4061 		else
4062 			return (NFSERR_AUTHERR | AUTH_NEEDS_TLS_MUTUAL_HOST);
4063 	}
4064 
4065 	/*
4066 	 * This seems odd, but allow the case where the security flavor
4067 	 * list is empty. This happens when NFSv4 is traversing non-exported
4068 	 * file systems. Exported file systems should always have a non-empty
4069 	 * security flavor list.
4070 	 */
4071 	if (exp->nes_numsecflavor == 0)
4072 		return (0);
4073 
4074 	for (i = 0; i < exp->nes_numsecflavor; i++) {
4075 		/*
4076 		 * The tests for privacy and integrity must be first,
4077 		 * since ND_GSS is set for everything but AUTH_SYS.
4078 		 */
4079 		if (exp->nes_secflavors[i] == RPCSEC_GSS_KRB5P &&
4080 		    (nd->nd_flag & ND_GSSPRIVACY))
4081 			return (0);
4082 		if (exp->nes_secflavors[i] == RPCSEC_GSS_KRB5I &&
4083 		    (nd->nd_flag & ND_GSSINTEGRITY))
4084 			return (0);
4085 		if (exp->nes_secflavors[i] == RPCSEC_GSS_KRB5 &&
4086 		    (nd->nd_flag & ND_GSS))
4087 			return (0);
4088 		if (exp->nes_secflavors[i] == AUTH_SYS &&
4089 		    (nd->nd_flag & ND_GSS) == 0)
4090 			return (0);
4091 	}
4092 	if ((nd->nd_flag & ND_NFSV4) != 0)
4093 		return (NFSERR_WRONGSEC);
4094 	return (NFSERR_AUTHERR | AUTH_TOOWEAK);
4095 }
4096 
4097 /*
4098  * Calculate a hash value for the fid in a file handle.
4099  */
4100 uint32_t
4101 nfsrv_hashfh(fhandle_t *fhp)
4102 {
4103 	uint32_t hashval;
4104 
4105 	hashval = hash32_buf(&fhp->fh_fid, sizeof(struct fid), 0);
4106 	return (hashval);
4107 }
4108 
4109 /*
4110  * Calculate a hash value for the sessionid.
4111  */
4112 uint32_t
4113 nfsrv_hashsessionid(uint8_t *sessionid)
4114 {
4115 	uint32_t hashval;
4116 
4117 	hashval = hash32_buf(sessionid, NFSX_V4SESSIONID, 0);
4118 	return (hashval);
4119 }
4120 
4121 /*
4122  * Signal the userland master nfsd to backup the stable restart file.
4123  */
4124 void
4125 nfsrv_backupstable(void)
4126 {
4127 	struct proc *procp;
4128 
4129 	if (nfsd_master_proc != NULL) {
4130 		procp = pfind(nfsd_master_pid);
4131 		/* Try to make sure it is the correct process. */
4132 		if (procp == nfsd_master_proc &&
4133 		    procp->p_stats->p_start.tv_sec ==
4134 		    nfsd_master_start.tv_sec &&
4135 		    procp->p_stats->p_start.tv_usec ==
4136 		    nfsd_master_start.tv_usec &&
4137 		    strcmp(procp->p_comm, nfsd_master_comm) == 0)
4138 			kern_psignal(procp, SIGUSR2);
4139 		else
4140 			nfsd_master_proc = NULL;
4141 
4142 		if (procp != NULL)
4143 			PROC_UNLOCK(procp);
4144 	}
4145 }
4146 
4147 /*
4148  * Create a DS data file for nfsrv_pnfscreate(). Called for each mirror.
4149  * The arguments are in a structure, so that they can be passed through
4150  * taskqueue for a kernel process to execute this function.
4151  */
4152 struct nfsrvdscreate {
4153 	int			done;
4154 	int			inprog;
4155 	struct task		tsk;
4156 	struct ucred		*tcred;
4157 	struct vnode		*dvp;
4158 	NFSPROC_T		*p;
4159 	struct pnfsdsfile	*pf;
4160 	int			err;
4161 	fhandle_t		fh;
4162 	struct vattr		va;
4163 	struct vattr		createva;
4164 };
4165 
4166 int
4167 nfsrv_dscreate(struct vnode *dvp, struct vattr *vap, struct vattr *nvap,
4168     fhandle_t *fhp, struct pnfsdsfile *pf, struct pnfsdsattr *dsa,
4169     char *fnamep, struct ucred *tcred, NFSPROC_T *p, struct vnode **nvpp)
4170 {
4171 	struct vnode *nvp;
4172 	struct nameidata named;
4173 	struct vattr va;
4174 	char *bufp;
4175 	u_long *hashp;
4176 	struct nfsnode *np;
4177 	struct nfsmount *nmp;
4178 	int error;
4179 
4180 	NFSNAMEICNDSET(&named.ni_cnd, tcred, CREATE,
4181 	    LOCKPARENT | LOCKLEAF | SAVESTART | NOCACHE);
4182 	nfsvno_setpathbuf(&named, &bufp, &hashp);
4183 	named.ni_cnd.cn_lkflags = LK_EXCLUSIVE;
4184 	named.ni_cnd.cn_thread = p;
4185 	named.ni_cnd.cn_nameptr = bufp;
4186 	if (fnamep != NULL) {
4187 		strlcpy(bufp, fnamep, PNFS_FILENAME_LEN + 1);
4188 		named.ni_cnd.cn_namelen = strlen(bufp);
4189 	} else
4190 		named.ni_cnd.cn_namelen = nfsrv_putfhname(fhp, bufp);
4191 	NFSD_DEBUG(4, "nfsrv_dscreate: dvp=%p fname=%s\n", dvp, bufp);
4192 
4193 	/* Create the date file in the DS mount. */
4194 	error = NFSVOPLOCK(dvp, LK_EXCLUSIVE);
4195 	if (error == 0) {
4196 		error = VOP_CREATE(dvp, &nvp, &named.ni_cnd, vap);
4197 		vref(dvp);
4198 		VOP_VPUT_PAIR(dvp, error == 0 ? &nvp : NULL, false);
4199 		if (error == 0) {
4200 			/* Set the ownership of the file. */
4201 			error = VOP_SETATTR(nvp, nvap, tcred);
4202 			NFSD_DEBUG(4, "nfsrv_dscreate:"
4203 			    " setattr-uid=%d\n", error);
4204 			if (error != 0)
4205 				vput(nvp);
4206 		}
4207 		if (error != 0)
4208 			printf("pNFS: pnfscreate failed=%d\n", error);
4209 	} else
4210 		printf("pNFS: pnfscreate vnlock=%d\n", error);
4211 	if (error == 0) {
4212 		np = VTONFS(nvp);
4213 		nmp = VFSTONFS(nvp->v_mount);
4214 		if (strcmp(nvp->v_mount->mnt_vfc->vfc_name, "nfs")
4215 		    != 0 || nmp->nm_nam->sa_len > sizeof(
4216 		    struct sockaddr_in6) ||
4217 		    np->n_fhp->nfh_len != NFSX_MYFH) {
4218 			printf("Bad DS file: fstype=%s salen=%d"
4219 			    " fhlen=%d\n",
4220 			    nvp->v_mount->mnt_vfc->vfc_name,
4221 			    nmp->nm_nam->sa_len, np->n_fhp->nfh_len);
4222 			error = ENOENT;
4223 		}
4224 
4225 		/* Set extattrs for the DS on the MDS file. */
4226 		if (error == 0) {
4227 			if (dsa != NULL) {
4228 				error = VOP_GETATTR(nvp, &va, tcred);
4229 				if (error == 0) {
4230 					dsa->dsa_filerev = va.va_filerev;
4231 					dsa->dsa_size = va.va_size;
4232 					dsa->dsa_atime = va.va_atime;
4233 					dsa->dsa_mtime = va.va_mtime;
4234 					dsa->dsa_bytes = va.va_bytes;
4235 				}
4236 			}
4237 			if (error == 0) {
4238 				NFSBCOPY(np->n_fhp->nfh_fh, &pf->dsf_fh,
4239 				    NFSX_MYFH);
4240 				NFSBCOPY(nmp->nm_nam, &pf->dsf_sin,
4241 				    nmp->nm_nam->sa_len);
4242 				NFSBCOPY(named.ni_cnd.cn_nameptr,
4243 				    pf->dsf_filename,
4244 				    sizeof(pf->dsf_filename));
4245 			}
4246 		} else
4247 			printf("pNFS: pnfscreate can't get DS"
4248 			    " attr=%d\n", error);
4249 		if (nvpp != NULL && error == 0)
4250 			*nvpp = nvp;
4251 		else
4252 			vput(nvp);
4253 	}
4254 	nfsvno_relpathbuf(&named);
4255 	return (error);
4256 }
4257 
4258 /*
4259  * Start up the thread that will execute nfsrv_dscreate().
4260  */
4261 static void
4262 start_dscreate(void *arg, int pending)
4263 {
4264 	struct nfsrvdscreate *dsc;
4265 
4266 	dsc = (struct nfsrvdscreate *)arg;
4267 	dsc->err = nfsrv_dscreate(dsc->dvp, &dsc->createva, &dsc->va, &dsc->fh,
4268 	    dsc->pf, NULL, NULL, dsc->tcred, dsc->p, NULL);
4269 	dsc->done = 1;
4270 	NFSD_DEBUG(4, "start_dscreate: err=%d\n", dsc->err);
4271 }
4272 
4273 /*
4274  * Create a pNFS data file on the Data Server(s).
4275  */
4276 static void
4277 nfsrv_pnfscreate(struct vnode *vp, struct vattr *vap, struct ucred *cred,
4278     NFSPROC_T *p)
4279 {
4280 	struct nfsrvdscreate *dsc, *tdsc = NULL;
4281 	struct nfsdevice *ds, *tds, *fds;
4282 	struct mount *mp;
4283 	struct pnfsdsfile *pf, *tpf;
4284 	struct pnfsdsattr dsattr;
4285 	struct vattr va;
4286 	struct vnode *dvp[NFSDEV_MAXMIRRORS];
4287 	struct nfsmount *nmp;
4288 	fhandle_t fh;
4289 	uid_t vauid;
4290 	gid_t vagid;
4291 	u_short vamode;
4292 	struct ucred *tcred;
4293 	int dsdir[NFSDEV_MAXMIRRORS], error, i, mirrorcnt, ret;
4294 	int failpos, timo;
4295 
4296 	/* Get a DS server directory in a round-robin order. */
4297 	mirrorcnt = 1;
4298 	mp = vp->v_mount;
4299 	ds = fds = NULL;
4300 	NFSDDSLOCK();
4301 	/*
4302 	 * Search for the first entry that handles this MDS fs, but use the
4303 	 * first entry for all MDS fs's otherwise.
4304 	 */
4305 	TAILQ_FOREACH(tds, &nfsrv_devidhead, nfsdev_list) {
4306 		if (tds->nfsdev_nmp != NULL) {
4307 			if (tds->nfsdev_mdsisset == 0 && ds == NULL)
4308 				ds = tds;
4309 			else if (tds->nfsdev_mdsisset != 0 && fsidcmp(
4310 			    &mp->mnt_stat.f_fsid, &tds->nfsdev_mdsfsid) == 0) {
4311 				ds = fds = tds;
4312 				break;
4313 			}
4314 		}
4315 	}
4316 	if (ds == NULL) {
4317 		NFSDDSUNLOCK();
4318 		NFSD_DEBUG(4, "nfsrv_pnfscreate: no srv\n");
4319 		return;
4320 	}
4321 	i = dsdir[0] = ds->nfsdev_nextdir;
4322 	ds->nfsdev_nextdir = (ds->nfsdev_nextdir + 1) % nfsrv_dsdirsize;
4323 	dvp[0] = ds->nfsdev_dsdir[i];
4324 	tds = TAILQ_NEXT(ds, nfsdev_list);
4325 	if (nfsrv_maxpnfsmirror > 1 && tds != NULL) {
4326 		TAILQ_FOREACH_FROM(tds, &nfsrv_devidhead, nfsdev_list) {
4327 			if (tds->nfsdev_nmp != NULL &&
4328 			    ((tds->nfsdev_mdsisset == 0 && fds == NULL) ||
4329 			     (tds->nfsdev_mdsisset != 0 && fds != NULL &&
4330 			      fsidcmp(&mp->mnt_stat.f_fsid,
4331 			      &tds->nfsdev_mdsfsid) == 0))) {
4332 				dsdir[mirrorcnt] = i;
4333 				dvp[mirrorcnt] = tds->nfsdev_dsdir[i];
4334 				mirrorcnt++;
4335 				if (mirrorcnt >= nfsrv_maxpnfsmirror)
4336 					break;
4337 			}
4338 		}
4339 	}
4340 	/* Put at end of list to implement round-robin usage. */
4341 	TAILQ_REMOVE(&nfsrv_devidhead, ds, nfsdev_list);
4342 	TAILQ_INSERT_TAIL(&nfsrv_devidhead, ds, nfsdev_list);
4343 	NFSDDSUNLOCK();
4344 	dsc = NULL;
4345 	if (mirrorcnt > 1)
4346 		tdsc = dsc = malloc(sizeof(*dsc) * (mirrorcnt - 1), M_TEMP,
4347 		    M_WAITOK | M_ZERO);
4348 	tpf = pf = malloc(sizeof(*pf) * nfsrv_maxpnfsmirror, M_TEMP, M_WAITOK |
4349 	    M_ZERO);
4350 
4351 	error = nfsvno_getfh(vp, &fh, p);
4352 	if (error == 0)
4353 		error = VOP_GETATTR(vp, &va, cred);
4354 	if (error == 0) {
4355 		/* Set the attributes for "vp" to Setattr the DS vp. */
4356 		vauid = va.va_uid;
4357 		vagid = va.va_gid;
4358 		vamode = va.va_mode;
4359 		VATTR_NULL(&va);
4360 		va.va_uid = vauid;
4361 		va.va_gid = vagid;
4362 		va.va_mode = vamode;
4363 		va.va_size = 0;
4364 	} else
4365 		printf("pNFS: pnfscreate getfh+attr=%d\n", error);
4366 
4367 	NFSD_DEBUG(4, "nfsrv_pnfscreate: cruid=%d crgid=%d\n", cred->cr_uid,
4368 	    cred->cr_gid);
4369 	/* Make data file name based on FH. */
4370 	tcred = newnfs_getcred();
4371 
4372 	/*
4373 	 * Create the file on each DS mirror, using kernel process(es) for the
4374 	 * additional mirrors.
4375 	 */
4376 	failpos = -1;
4377 	for (i = 0; i < mirrorcnt - 1 && error == 0; i++, tpf++, tdsc++) {
4378 		tpf->dsf_dir = dsdir[i];
4379 		tdsc->tcred = tcred;
4380 		tdsc->p = p;
4381 		tdsc->pf = tpf;
4382 		tdsc->createva = *vap;
4383 		NFSBCOPY(&fh, &tdsc->fh, sizeof(fh));
4384 		tdsc->va = va;
4385 		tdsc->dvp = dvp[i];
4386 		tdsc->done = 0;
4387 		tdsc->inprog = 0;
4388 		tdsc->err = 0;
4389 		ret = EIO;
4390 		if (nfs_pnfsiothreads != 0) {
4391 			ret = nfs_pnfsio(start_dscreate, tdsc);
4392 			NFSD_DEBUG(4, "nfsrv_pnfscreate: nfs_pnfsio=%d\n", ret);
4393 		}
4394 		if (ret != 0) {
4395 			ret = nfsrv_dscreate(dvp[i], vap, &va, &fh, tpf, NULL,
4396 			    NULL, tcred, p, NULL);
4397 			if (ret != 0) {
4398 				KASSERT(error == 0, ("nfsrv_dscreate err=%d",
4399 				    error));
4400 				if (failpos == -1 && nfsds_failerr(ret))
4401 					failpos = i;
4402 				else
4403 					error = ret;
4404 			}
4405 		}
4406 	}
4407 	if (error == 0) {
4408 		tpf->dsf_dir = dsdir[mirrorcnt - 1];
4409 		error = nfsrv_dscreate(dvp[mirrorcnt - 1], vap, &va, &fh, tpf,
4410 		    &dsattr, NULL, tcred, p, NULL);
4411 		if (failpos == -1 && mirrorcnt > 1 && nfsds_failerr(error)) {
4412 			failpos = mirrorcnt - 1;
4413 			error = 0;
4414 		}
4415 	}
4416 	timo = hz / 50;		/* Wait for 20msec. */
4417 	if (timo < 1)
4418 		timo = 1;
4419 	/* Wait for kernel task(s) to complete. */
4420 	for (tdsc = dsc, i = 0; i < mirrorcnt - 1; i++, tdsc++) {
4421 		while (tdsc->inprog != 0 && tdsc->done == 0)
4422 			tsleep(&tdsc->tsk, PVFS, "srvdcr", timo);
4423 		if (tdsc->err != 0) {
4424 			if (failpos == -1 && nfsds_failerr(tdsc->err))
4425 				failpos = i;
4426 			else if (error == 0)
4427 				error = tdsc->err;
4428 		}
4429 	}
4430 
4431 	/*
4432 	 * If failpos has been set, that mirror has failed, so it needs
4433 	 * to be disabled.
4434 	 */
4435 	if (failpos >= 0) {
4436 		nmp = VFSTONFS(dvp[failpos]->v_mount);
4437 		NFSLOCKMNT(nmp);
4438 		if ((nmp->nm_privflag & (NFSMNTP_FORCEDISM |
4439 		     NFSMNTP_CANCELRPCS)) == 0) {
4440 			nmp->nm_privflag |= NFSMNTP_CANCELRPCS;
4441 			NFSUNLOCKMNT(nmp);
4442 			ds = nfsrv_deldsnmp(PNFSDOP_DELDSSERVER, nmp, p);
4443 			NFSD_DEBUG(4, "dscreatfail fail=%d ds=%p\n", failpos,
4444 			    ds);
4445 			if (ds != NULL)
4446 				nfsrv_killrpcs(nmp);
4447 			NFSLOCKMNT(nmp);
4448 			nmp->nm_privflag &= ~NFSMNTP_CANCELRPCS;
4449 			wakeup(nmp);
4450 		}
4451 		NFSUNLOCKMNT(nmp);
4452 	}
4453 
4454 	NFSFREECRED(tcred);
4455 	if (error == 0) {
4456 		ASSERT_VOP_ELOCKED(vp, "nfsrv_pnfscreate vp");
4457 
4458 		NFSD_DEBUG(4, "nfsrv_pnfscreate: mirrorcnt=%d maxmirror=%d\n",
4459 		    mirrorcnt, nfsrv_maxpnfsmirror);
4460 		/*
4461 		 * For all mirrors that couldn't be created, fill in the
4462 		 * *pf structure, but with an IP address == 0.0.0.0.
4463 		 */
4464 		tpf = pf + mirrorcnt;
4465 		for (i = mirrorcnt; i < nfsrv_maxpnfsmirror; i++, tpf++) {
4466 			*tpf = *pf;
4467 			tpf->dsf_sin.sin_family = AF_INET;
4468 			tpf->dsf_sin.sin_len = sizeof(struct sockaddr_in);
4469 			tpf->dsf_sin.sin_addr.s_addr = 0;
4470 			tpf->dsf_sin.sin_port = 0;
4471 		}
4472 
4473 		error = vn_extattr_set(vp, IO_NODELOCKED,
4474 		    EXTATTR_NAMESPACE_SYSTEM, "pnfsd.dsfile",
4475 		    sizeof(*pf) * nfsrv_maxpnfsmirror, (char *)pf, p);
4476 		if (error == 0)
4477 			error = vn_extattr_set(vp, IO_NODELOCKED,
4478 			    EXTATTR_NAMESPACE_SYSTEM, "pnfsd.dsattr",
4479 			    sizeof(dsattr), (char *)&dsattr, p);
4480 		if (error != 0)
4481 			printf("pNFS: pnfscreate setextattr=%d\n",
4482 			    error);
4483 	} else
4484 		printf("pNFS: pnfscreate=%d\n", error);
4485 	free(pf, M_TEMP);
4486 	free(dsc, M_TEMP);
4487 }
4488 
4489 /*
4490  * Get the information needed to remove the pNFS Data Server file from the
4491  * Metadata file.  Upon success, ddvp is set non-NULL to the locked
4492  * DS directory vnode.  The caller must unlock *ddvp when done with it.
4493  */
4494 static void
4495 nfsrv_pnfsremovesetup(struct vnode *vp, NFSPROC_T *p, struct vnode **dvpp,
4496     int *mirrorcntp, char *fname, fhandle_t *fhp)
4497 {
4498 	struct vattr va;
4499 	struct ucred *tcred;
4500 	char *buf;
4501 	int buflen, error;
4502 
4503 	dvpp[0] = NULL;
4504 	/* If not an exported regular file or not a pNFS server, just return. */
4505 	if (vp->v_type != VREG || (vp->v_mount->mnt_flag & MNT_EXPORTED) == 0 ||
4506 	    nfsrv_devidcnt == 0)
4507 		return;
4508 
4509 	/* Check to see if this is the last hard link. */
4510 	tcred = newnfs_getcred();
4511 	error = VOP_GETATTR(vp, &va, tcred);
4512 	NFSFREECRED(tcred);
4513 	if (error != 0) {
4514 		printf("pNFS: nfsrv_pnfsremovesetup getattr=%d\n", error);
4515 		return;
4516 	}
4517 	if (va.va_nlink > 1)
4518 		return;
4519 
4520 	error = nfsvno_getfh(vp, fhp, p);
4521 	if (error != 0) {
4522 		printf("pNFS: nfsrv_pnfsremovesetup getfh=%d\n", error);
4523 		return;
4524 	}
4525 
4526 	buflen = 1024;
4527 	buf = malloc(buflen, M_TEMP, M_WAITOK);
4528 	/* Get the directory vnode for the DS mount and the file handle. */
4529 	error = nfsrv_dsgetsockmnt(vp, 0, buf, &buflen, mirrorcntp, p, dvpp,
4530 	    NULL, NULL, fname, NULL, NULL, NULL, NULL, NULL);
4531 	free(buf, M_TEMP);
4532 	if (error != 0)
4533 		printf("pNFS: nfsrv_pnfsremovesetup getsockmnt=%d\n", error);
4534 }
4535 
4536 /*
4537  * Remove a DS data file for nfsrv_pnfsremove(). Called for each mirror.
4538  * The arguments are in a structure, so that they can be passed through
4539  * taskqueue for a kernel process to execute this function.
4540  */
4541 struct nfsrvdsremove {
4542 	int			done;
4543 	int			inprog;
4544 	struct task		tsk;
4545 	struct ucred		*tcred;
4546 	struct vnode		*dvp;
4547 	NFSPROC_T		*p;
4548 	int			err;
4549 	char			fname[PNFS_FILENAME_LEN + 1];
4550 };
4551 
4552 static int
4553 nfsrv_dsremove(struct vnode *dvp, char *fname, struct ucred *tcred,
4554     NFSPROC_T *p)
4555 {
4556 	struct nameidata named;
4557 	struct vnode *nvp;
4558 	char *bufp;
4559 	u_long *hashp;
4560 	int error;
4561 
4562 	error = NFSVOPLOCK(dvp, LK_EXCLUSIVE);
4563 	if (error != 0)
4564 		return (error);
4565 	named.ni_cnd.cn_nameiop = DELETE;
4566 	named.ni_cnd.cn_lkflags = LK_EXCLUSIVE | LK_RETRY;
4567 	named.ni_cnd.cn_cred = tcred;
4568 	named.ni_cnd.cn_thread = p;
4569 	named.ni_cnd.cn_flags = ISLASTCN | LOCKPARENT | LOCKLEAF | SAVENAME;
4570 	nfsvno_setpathbuf(&named, &bufp, &hashp);
4571 	named.ni_cnd.cn_nameptr = bufp;
4572 	named.ni_cnd.cn_namelen = strlen(fname);
4573 	strlcpy(bufp, fname, NAME_MAX);
4574 	NFSD_DEBUG(4, "nfsrv_pnfsremove: filename=%s\n", bufp);
4575 	error = VOP_LOOKUP(dvp, &nvp, &named.ni_cnd);
4576 	NFSD_DEBUG(4, "nfsrv_pnfsremove: aft LOOKUP=%d\n", error);
4577 	if (error == 0) {
4578 		error = VOP_REMOVE(dvp, nvp, &named.ni_cnd);
4579 		vput(nvp);
4580 	}
4581 	NFSVOPUNLOCK(dvp);
4582 	nfsvno_relpathbuf(&named);
4583 	if (error != 0)
4584 		printf("pNFS: nfsrv_pnfsremove failed=%d\n", error);
4585 	return (error);
4586 }
4587 
4588 /*
4589  * Start up the thread that will execute nfsrv_dsremove().
4590  */
4591 static void
4592 start_dsremove(void *arg, int pending)
4593 {
4594 	struct nfsrvdsremove *dsrm;
4595 
4596 	dsrm = (struct nfsrvdsremove *)arg;
4597 	dsrm->err = nfsrv_dsremove(dsrm->dvp, dsrm->fname, dsrm->tcred,
4598 	    dsrm->p);
4599 	dsrm->done = 1;
4600 	NFSD_DEBUG(4, "start_dsremove: err=%d\n", dsrm->err);
4601 }
4602 
4603 /*
4604  * Remove a pNFS data file from a Data Server.
4605  * nfsrv_pnfsremovesetup() must have been called before the MDS file was
4606  * removed to set up the dvp and fill in the FH.
4607  */
4608 static void
4609 nfsrv_pnfsremove(struct vnode **dvp, int mirrorcnt, char *fname, fhandle_t *fhp,
4610     NFSPROC_T *p)
4611 {
4612 	struct ucred *tcred;
4613 	struct nfsrvdsremove *dsrm, *tdsrm;
4614 	struct nfsdevice *ds;
4615 	struct nfsmount *nmp;
4616 	int failpos, i, ret, timo;
4617 
4618 	tcred = newnfs_getcred();
4619 	dsrm = NULL;
4620 	if (mirrorcnt > 1)
4621 		dsrm = malloc(sizeof(*dsrm) * mirrorcnt - 1, M_TEMP, M_WAITOK);
4622 	/*
4623 	 * Remove the file on each DS mirror, using kernel process(es) for the
4624 	 * additional mirrors.
4625 	 */
4626 	failpos = -1;
4627 	for (tdsrm = dsrm, i = 0; i < mirrorcnt - 1; i++, tdsrm++) {
4628 		tdsrm->tcred = tcred;
4629 		tdsrm->p = p;
4630 		tdsrm->dvp = dvp[i];
4631 		strlcpy(tdsrm->fname, fname, PNFS_FILENAME_LEN + 1);
4632 		tdsrm->inprog = 0;
4633 		tdsrm->done = 0;
4634 		tdsrm->err = 0;
4635 		ret = EIO;
4636 		if (nfs_pnfsiothreads != 0) {
4637 			ret = nfs_pnfsio(start_dsremove, tdsrm);
4638 			NFSD_DEBUG(4, "nfsrv_pnfsremove: nfs_pnfsio=%d\n", ret);
4639 		}
4640 		if (ret != 0) {
4641 			ret = nfsrv_dsremove(dvp[i], fname, tcred, p);
4642 			if (failpos == -1 && nfsds_failerr(ret))
4643 				failpos = i;
4644 		}
4645 	}
4646 	ret = nfsrv_dsremove(dvp[mirrorcnt - 1], fname, tcred, p);
4647 	if (failpos == -1 && mirrorcnt > 1 && nfsds_failerr(ret))
4648 		failpos = mirrorcnt - 1;
4649 	timo = hz / 50;		/* Wait for 20msec. */
4650 	if (timo < 1)
4651 		timo = 1;
4652 	/* Wait for kernel task(s) to complete. */
4653 	for (tdsrm = dsrm, i = 0; i < mirrorcnt - 1; i++, tdsrm++) {
4654 		while (tdsrm->inprog != 0 && tdsrm->done == 0)
4655 			tsleep(&tdsrm->tsk, PVFS, "srvdsrm", timo);
4656 		if (failpos == -1 && nfsds_failerr(tdsrm->err))
4657 			failpos = i;
4658 	}
4659 
4660 	/*
4661 	 * If failpos has been set, that mirror has failed, so it needs
4662 	 * to be disabled.
4663 	 */
4664 	if (failpos >= 0) {
4665 		nmp = VFSTONFS(dvp[failpos]->v_mount);
4666 		NFSLOCKMNT(nmp);
4667 		if ((nmp->nm_privflag & (NFSMNTP_FORCEDISM |
4668 		     NFSMNTP_CANCELRPCS)) == 0) {
4669 			nmp->nm_privflag |= NFSMNTP_CANCELRPCS;
4670 			NFSUNLOCKMNT(nmp);
4671 			ds = nfsrv_deldsnmp(PNFSDOP_DELDSSERVER, nmp, p);
4672 			NFSD_DEBUG(4, "dsremovefail fail=%d ds=%p\n", failpos,
4673 			    ds);
4674 			if (ds != NULL)
4675 				nfsrv_killrpcs(nmp);
4676 			NFSLOCKMNT(nmp);
4677 			nmp->nm_privflag &= ~NFSMNTP_CANCELRPCS;
4678 			wakeup(nmp);
4679 		}
4680 		NFSUNLOCKMNT(nmp);
4681 	}
4682 
4683 	/* Get rid all layouts for the file. */
4684 	nfsrv_freefilelayouts(fhp);
4685 
4686 	NFSFREECRED(tcred);
4687 	free(dsrm, M_TEMP);
4688 }
4689 
4690 /*
4691  * Generate a file name based on the file handle and put it in *bufp.
4692  * Return the number of bytes generated.
4693  */
4694 static int
4695 nfsrv_putfhname(fhandle_t *fhp, char *bufp)
4696 {
4697 	int i;
4698 	uint8_t *cp;
4699 	const uint8_t *hexdigits = "0123456789abcdef";
4700 
4701 	cp = (uint8_t *)fhp;
4702 	for (i = 0; i < sizeof(*fhp); i++) {
4703 		bufp[2 * i] = hexdigits[(*cp >> 4) & 0xf];
4704 		bufp[2 * i + 1] = hexdigits[*cp++ & 0xf];
4705 	}
4706 	bufp[2 * i] = '\0';
4707 	return (2 * i);
4708 }
4709 
4710 /*
4711  * Update the Metadata file's attributes from the DS file when a Read/Write
4712  * layout is returned.
4713  * Basically just call nfsrv_proxyds() with procedure == NFSPROC_LAYOUTRETURN
4714  * so that it does a nfsrv_getattrdsrpc() and nfsrv_setextattr() on the DS file.
4715  */
4716 int
4717 nfsrv_updatemdsattr(struct vnode *vp, struct nfsvattr *nap, NFSPROC_T *p)
4718 {
4719 	struct ucred *tcred;
4720 	int error;
4721 
4722 	/* Do this as root so that it won't fail with EACCES. */
4723 	tcred = newnfs_getcred();
4724 	error = nfsrv_proxyds(vp, 0, 0, tcred, p, NFSPROC_LAYOUTRETURN,
4725 	    NULL, NULL, NULL, nap, NULL, NULL, 0, NULL);
4726 	NFSFREECRED(tcred);
4727 	return (error);
4728 }
4729 
4730 /*
4731  * Set the NFSv4 ACL on the DS file to the same ACL as the MDS file.
4732  */
4733 static int
4734 nfsrv_dssetacl(struct vnode *vp, struct acl *aclp, struct ucred *cred,
4735     NFSPROC_T *p)
4736 {
4737 	int error;
4738 
4739 	error = nfsrv_proxyds(vp, 0, 0, cred, p, NFSPROC_SETACL,
4740 	    NULL, NULL, NULL, NULL, aclp, NULL, 0, NULL);
4741 	return (error);
4742 }
4743 
4744 static int
4745 nfsrv_proxyds(struct vnode *vp, off_t off, int cnt, struct ucred *cred,
4746     struct thread *p, int ioproc, struct mbuf **mpp, char *cp,
4747     struct mbuf **mpp2, struct nfsvattr *nap, struct acl *aclp,
4748     off_t *offp, int content, bool *eofp)
4749 {
4750 	struct nfsmount *nmp[NFSDEV_MAXMIRRORS], *failnmp;
4751 	fhandle_t fh[NFSDEV_MAXMIRRORS];
4752 	struct vnode *dvp[NFSDEV_MAXMIRRORS];
4753 	struct nfsdevice *ds;
4754 	struct pnfsdsattr dsattr;
4755 	struct opnfsdsattr odsattr;
4756 	char *buf;
4757 	int buflen, error, failpos, i, mirrorcnt, origmircnt, trycnt;
4758 
4759 	NFSD_DEBUG(4, "in nfsrv_proxyds\n");
4760 	/*
4761 	 * If not a regular file, not exported or not a pNFS server,
4762 	 * just return ENOENT.
4763 	 */
4764 	if (vp->v_type != VREG || (vp->v_mount->mnt_flag & MNT_EXPORTED) == 0 ||
4765 	    nfsrv_devidcnt == 0)
4766 		return (ENOENT);
4767 
4768 	buflen = 1024;
4769 	buf = malloc(buflen, M_TEMP, M_WAITOK);
4770 	error = 0;
4771 
4772 	/*
4773 	 * For Getattr, get the Change attribute (va_filerev) and size (va_size)
4774 	 * from the MetaData file's extended attribute.
4775 	 */
4776 	if (ioproc == NFSPROC_GETATTR) {
4777 		error = vn_extattr_get(vp, IO_NODELOCKED,
4778 		    EXTATTR_NAMESPACE_SYSTEM, "pnfsd.dsattr", &buflen, buf,
4779 		    p);
4780 		if (error == 0) {
4781 			if (buflen == sizeof(odsattr)) {
4782 				NFSBCOPY(buf, &odsattr, buflen);
4783 				nap->na_filerev = odsattr.dsa_filerev;
4784 				nap->na_size = odsattr.dsa_size;
4785 				nap->na_atime = odsattr.dsa_atime;
4786 				nap->na_mtime = odsattr.dsa_mtime;
4787 				/*
4788 				 * Fake na_bytes by rounding up na_size.
4789 				 * Since we don't know the block size, just
4790 				 * use BLKDEV_IOSIZE.
4791 				 */
4792 				nap->na_bytes = (odsattr.dsa_size +
4793 				    BLKDEV_IOSIZE - 1) & ~(BLKDEV_IOSIZE - 1);
4794 			} else if (buflen == sizeof(dsattr)) {
4795 				NFSBCOPY(buf, &dsattr, buflen);
4796 				nap->na_filerev = dsattr.dsa_filerev;
4797 				nap->na_size = dsattr.dsa_size;
4798 				nap->na_atime = dsattr.dsa_atime;
4799 				nap->na_mtime = dsattr.dsa_mtime;
4800 				nap->na_bytes = dsattr.dsa_bytes;
4801 			} else
4802 				error = ENXIO;
4803 		}
4804 		if (error == 0) {
4805 			/*
4806 			 * If nfsrv_pnfsgetdsattr is 0 or nfsrv_checkdsattr()
4807 			 * returns 0, just return now.  nfsrv_checkdsattr()
4808 			 * returns 0 if there is no Read/Write layout
4809 			 * plus either an Open/Write_access or Write
4810 			 * delegation issued to a client for the file.
4811 			 */
4812 			if (nfsrv_pnfsgetdsattr == 0 ||
4813 			    nfsrv_checkdsattr(vp, p) == 0) {
4814 				free(buf, M_TEMP);
4815 				return (error);
4816 			}
4817 		}
4818 
4819 		/*
4820 		 * Clear ENOATTR so the code below will attempt to do a
4821 		 * nfsrv_getattrdsrpc() to get the attributes and (re)create
4822 		 * the extended attribute.
4823 		 */
4824 		if (error == ENOATTR)
4825 			error = 0;
4826 	}
4827 
4828 	origmircnt = -1;
4829 	trycnt = 0;
4830 tryagain:
4831 	if (error == 0) {
4832 		buflen = 1024;
4833 		if (ioproc == NFSPROC_READDS && NFSVOPISLOCKED(vp) ==
4834 		    LK_EXCLUSIVE)
4835 			printf("nfsrv_proxyds: Readds vp exclusively locked\n");
4836 		error = nfsrv_dsgetsockmnt(vp, LK_SHARED, buf, &buflen,
4837 		    &mirrorcnt, p, dvp, fh, NULL, NULL, NULL, NULL, NULL,
4838 		    NULL, NULL);
4839 		if (error == 0) {
4840 			for (i = 0; i < mirrorcnt; i++)
4841 				nmp[i] = VFSTONFS(dvp[i]->v_mount);
4842 		} else
4843 			printf("pNFS: proxy getextattr sockaddr=%d\n", error);
4844 	} else
4845 		printf("pNFS: nfsrv_dsgetsockmnt=%d\n", error);
4846 	if (error == 0) {
4847 		failpos = -1;
4848 		if (origmircnt == -1)
4849 			origmircnt = mirrorcnt;
4850 		/*
4851 		 * If failpos is set to a mirror#, then that mirror has
4852 		 * failed and will be disabled. For Read, Getattr and Seek, the
4853 		 * function only tries one mirror, so if that mirror has
4854 		 * failed, it will need to be retried. As such, increment
4855 		 * tryitagain for these cases.
4856 		 * For Write, Setattr and Setacl, the function tries all
4857 		 * mirrors and will not return an error for the case where
4858 		 * one mirror has failed. For these cases, the functioning
4859 		 * mirror(s) will have been modified, so a retry isn't
4860 		 * necessary. These functions will set failpos for the
4861 		 * failed mirror#.
4862 		 */
4863 		if (ioproc == NFSPROC_READDS) {
4864 			error = nfsrv_readdsrpc(fh, off, cnt, cred, p, nmp[0],
4865 			    mpp, mpp2);
4866 			if (nfsds_failerr(error) && mirrorcnt > 1) {
4867 				/*
4868 				 * Setting failpos will cause the mirror
4869 				 * to be disabled and then a retry of this
4870 				 * read is required.
4871 				 */
4872 				failpos = 0;
4873 				error = 0;
4874 				trycnt++;
4875 			}
4876 		} else if (ioproc == NFSPROC_WRITEDS)
4877 			error = nfsrv_writedsrpc(fh, off, cnt, cred, p, vp,
4878 			    &nmp[0], mirrorcnt, mpp, cp, &failpos);
4879 		else if (ioproc == NFSPROC_SETATTR)
4880 			error = nfsrv_setattrdsrpc(fh, cred, p, vp, &nmp[0],
4881 			    mirrorcnt, nap, &failpos);
4882 		else if (ioproc == NFSPROC_SETACL)
4883 			error = nfsrv_setacldsrpc(fh, cred, p, vp, &nmp[0],
4884 			    mirrorcnt, aclp, &failpos);
4885 		else if (ioproc == NFSPROC_SEEKDS) {
4886 			error = nfsrv_seekdsrpc(fh, offp, content, eofp, cred,
4887 			    p, nmp[0]);
4888 			if (nfsds_failerr(error) && mirrorcnt > 1) {
4889 				/*
4890 				 * Setting failpos will cause the mirror
4891 				 * to be disabled and then a retry of this
4892 				 * read is required.
4893 				 */
4894 				failpos = 0;
4895 				error = 0;
4896 				trycnt++;
4897 			}
4898 		} else if (ioproc == NFSPROC_ALLOCATE)
4899 			error = nfsrv_allocatedsrpc(fh, off, *offp, cred, p, vp,
4900 			    &nmp[0], mirrorcnt, &failpos);
4901 		else {
4902 			error = nfsrv_getattrdsrpc(&fh[mirrorcnt - 1], cred, p,
4903 			    vp, nmp[mirrorcnt - 1], nap);
4904 			if (nfsds_failerr(error) && mirrorcnt > 1) {
4905 				/*
4906 				 * Setting failpos will cause the mirror
4907 				 * to be disabled and then a retry of this
4908 				 * getattr is required.
4909 				 */
4910 				failpos = mirrorcnt - 1;
4911 				error = 0;
4912 				trycnt++;
4913 			}
4914 		}
4915 		ds = NULL;
4916 		if (failpos >= 0) {
4917 			failnmp = nmp[failpos];
4918 			NFSLOCKMNT(failnmp);
4919 			if ((failnmp->nm_privflag & (NFSMNTP_FORCEDISM |
4920 			     NFSMNTP_CANCELRPCS)) == 0) {
4921 				failnmp->nm_privflag |= NFSMNTP_CANCELRPCS;
4922 				NFSUNLOCKMNT(failnmp);
4923 				ds = nfsrv_deldsnmp(PNFSDOP_DELDSSERVER,
4924 				    failnmp, p);
4925 				NFSD_DEBUG(4, "dsldsnmp fail=%d ds=%p\n",
4926 				    failpos, ds);
4927 				if (ds != NULL)
4928 					nfsrv_killrpcs(failnmp);
4929 				NFSLOCKMNT(failnmp);
4930 				failnmp->nm_privflag &= ~NFSMNTP_CANCELRPCS;
4931 				wakeup(failnmp);
4932 			}
4933 			NFSUNLOCKMNT(failnmp);
4934 		}
4935 		for (i = 0; i < mirrorcnt; i++)
4936 			NFSVOPUNLOCK(dvp[i]);
4937 		NFSD_DEBUG(4, "nfsrv_proxyds: aft RPC=%d trya=%d\n", error,
4938 		    trycnt);
4939 		/* Try the Read/Getattr again if a mirror was deleted. */
4940 		if (ds != NULL && trycnt > 0 && trycnt < origmircnt)
4941 			goto tryagain;
4942 	} else {
4943 		/* Return ENOENT for any Extended Attribute error. */
4944 		error = ENOENT;
4945 	}
4946 	free(buf, M_TEMP);
4947 	NFSD_DEBUG(4, "nfsrv_proxyds: error=%d\n", error);
4948 	return (error);
4949 }
4950 
4951 /*
4952  * Get the DS mount point, fh and directory from the "pnfsd.dsfile" extended
4953  * attribute.
4954  * newnmpp - If it points to a non-NULL nmp, that is the destination and needs
4955  *           to be checked.  If it points to a NULL nmp, then it returns
4956  *           a suitable destination.
4957  * curnmp - If non-NULL, it is the source mount for the copy.
4958  */
4959 int
4960 nfsrv_dsgetsockmnt(struct vnode *vp, int lktype, char *buf, int *buflenp,
4961     int *mirrorcntp, NFSPROC_T *p, struct vnode **dvpp, fhandle_t *fhp,
4962     char *devid, char *fnamep, struct vnode **nvpp, struct nfsmount **newnmpp,
4963     struct nfsmount *curnmp, int *ippos, int *dsdirp)
4964 {
4965 	struct vnode *dvp, *nvp = NULL, **tdvpp;
4966 	struct mount *mp;
4967 	struct nfsmount *nmp, *newnmp;
4968 	struct sockaddr *sad;
4969 	struct sockaddr_in *sin;
4970 	struct nfsdevice *ds, *tds, *fndds;
4971 	struct pnfsdsfile *pf;
4972 	uint32_t dsdir;
4973 	int error, fhiszero, fnd, gotone, i, mirrorcnt;
4974 
4975 	ASSERT_VOP_LOCKED(vp, "nfsrv_dsgetsockmnt vp");
4976 	*mirrorcntp = 1;
4977 	tdvpp = dvpp;
4978 	if (nvpp != NULL)
4979 		*nvpp = NULL;
4980 	if (dvpp != NULL)
4981 		*dvpp = NULL;
4982 	if (ippos != NULL)
4983 		*ippos = -1;
4984 	if (newnmpp != NULL)
4985 		newnmp = *newnmpp;
4986 	else
4987 		newnmp = NULL;
4988 	mp = vp->v_mount;
4989 	error = vn_extattr_get(vp, IO_NODELOCKED, EXTATTR_NAMESPACE_SYSTEM,
4990 	    "pnfsd.dsfile", buflenp, buf, p);
4991 	mirrorcnt = *buflenp / sizeof(*pf);
4992 	if (error == 0 && (mirrorcnt < 1 || mirrorcnt > NFSDEV_MAXMIRRORS ||
4993 	    *buflenp != sizeof(*pf) * mirrorcnt))
4994 		error = ENOATTR;
4995 
4996 	pf = (struct pnfsdsfile *)buf;
4997 	/* If curnmp != NULL, check for a match in the mirror list. */
4998 	if (curnmp != NULL && error == 0) {
4999 		fnd = 0;
5000 		for (i = 0; i < mirrorcnt; i++, pf++) {
5001 			sad = (struct sockaddr *)&pf->dsf_sin;
5002 			if (nfsaddr2_match(sad, curnmp->nm_nam)) {
5003 				if (ippos != NULL)
5004 					*ippos = i;
5005 				fnd = 1;
5006 				break;
5007 			}
5008 		}
5009 		if (fnd == 0)
5010 			error = ENXIO;
5011 	}
5012 
5013 	gotone = 0;
5014 	pf = (struct pnfsdsfile *)buf;
5015 	NFSD_DEBUG(4, "nfsrv_dsgetsockmnt: mirrorcnt=%d err=%d\n", mirrorcnt,
5016 	    error);
5017 	for (i = 0; i < mirrorcnt && error == 0; i++, pf++) {
5018 		fhiszero = 0;
5019 		sad = (struct sockaddr *)&pf->dsf_sin;
5020 		sin = &pf->dsf_sin;
5021 		dsdir = pf->dsf_dir;
5022 		if (dsdir >= nfsrv_dsdirsize) {
5023 			printf("nfsrv_dsgetsockmnt: dsdir=%d\n", dsdir);
5024 			error = ENOATTR;
5025 		} else if (nvpp != NULL && newnmp != NULL &&
5026 		    nfsaddr2_match(sad, newnmp->nm_nam))
5027 			error = EEXIST;
5028 		if (error == 0) {
5029 			if (ippos != NULL && curnmp == NULL &&
5030 			    sad->sa_family == AF_INET &&
5031 			    sin->sin_addr.s_addr == 0)
5032 				*ippos = i;
5033 			if (NFSBCMP(&zerofh, &pf->dsf_fh, sizeof(zerofh)) == 0)
5034 				fhiszero = 1;
5035 			/* Use the socket address to find the mount point. */
5036 			fndds = NULL;
5037 			NFSDDSLOCK();
5038 			/* Find a match for the IP address. */
5039 			TAILQ_FOREACH(ds, &nfsrv_devidhead, nfsdev_list) {
5040 				if (ds->nfsdev_nmp != NULL) {
5041 					dvp = ds->nfsdev_dvp;
5042 					nmp = VFSTONFS(dvp->v_mount);
5043 					if (nmp != ds->nfsdev_nmp)
5044 						printf("different2 nmp %p %p\n",
5045 						    nmp, ds->nfsdev_nmp);
5046 					if (nfsaddr2_match(sad, nmp->nm_nam)) {
5047 						fndds = ds;
5048 						break;
5049 					}
5050 				}
5051 			}
5052 			if (fndds != NULL && newnmpp != NULL &&
5053 			    newnmp == NULL) {
5054 				/* Search for a place to make a mirror copy. */
5055 				TAILQ_FOREACH(tds, &nfsrv_devidhead,
5056 				    nfsdev_list) {
5057 					if (tds->nfsdev_nmp != NULL &&
5058 					    fndds != tds &&
5059 					    ((tds->nfsdev_mdsisset == 0 &&
5060 					      fndds->nfsdev_mdsisset == 0) ||
5061 					     (tds->nfsdev_mdsisset != 0 &&
5062 					      fndds->nfsdev_mdsisset != 0 &&
5063 					      fsidcmp(&tds->nfsdev_mdsfsid,
5064 					      &mp->mnt_stat.f_fsid) == 0))) {
5065 						*newnmpp = tds->nfsdev_nmp;
5066 						break;
5067 					}
5068 				}
5069 				if (tds != NULL) {
5070 					/*
5071 					 * Move this entry to the end of the
5072 					 * list, so it won't be selected as
5073 					 * easily the next time.
5074 					 */
5075 					TAILQ_REMOVE(&nfsrv_devidhead, tds,
5076 					    nfsdev_list);
5077 					TAILQ_INSERT_TAIL(&nfsrv_devidhead, tds,
5078 					    nfsdev_list);
5079 				}
5080 			}
5081 			NFSDDSUNLOCK();
5082 			if (fndds != NULL) {
5083 				dvp = fndds->nfsdev_dsdir[dsdir];
5084 				if (lktype != 0 || fhiszero != 0 ||
5085 				    (nvpp != NULL && *nvpp == NULL)) {
5086 					if (fhiszero != 0)
5087 						error = vn_lock(dvp,
5088 						    LK_EXCLUSIVE);
5089 					else if (lktype != 0)
5090 						error = vn_lock(dvp, lktype);
5091 					else
5092 						error = vn_lock(dvp, LK_SHARED);
5093 					/*
5094 					 * If the file handle is all 0's, try to
5095 					 * do a Lookup against the DS to acquire
5096 					 * it.
5097 					 * If dvpp == NULL or the Lookup fails,
5098 					 * unlock dvp after the call.
5099 					 */
5100 					if (error == 0 && (fhiszero != 0 ||
5101 					    (nvpp != NULL && *nvpp == NULL))) {
5102 						error = nfsrv_pnfslookupds(vp,
5103 						    dvp, pf, &nvp, p);
5104 						if (error == 0) {
5105 							if (fhiszero != 0)
5106 								nfsrv_pnfssetfh(
5107 								    vp, pf,
5108 								    devid,
5109 								    fnamep,
5110 								    nvp, p);
5111 							if (nvpp != NULL &&
5112 							    *nvpp == NULL) {
5113 								*nvpp = nvp;
5114 								*dsdirp = dsdir;
5115 							} else
5116 								vput(nvp);
5117 						}
5118 						if (error != 0 || lktype == 0)
5119 							NFSVOPUNLOCK(dvp);
5120 					}
5121 				}
5122 				if (error == 0) {
5123 					gotone++;
5124 					NFSD_DEBUG(4, "gotone=%d\n", gotone);
5125 					if (devid != NULL) {
5126 						NFSBCOPY(fndds->nfsdev_deviceid,
5127 						    devid, NFSX_V4DEVICEID);
5128 						devid += NFSX_V4DEVICEID;
5129 					}
5130 					if (dvpp != NULL)
5131 						*tdvpp++ = dvp;
5132 					if (fhp != NULL)
5133 						NFSBCOPY(&pf->dsf_fh, fhp++,
5134 						    NFSX_MYFH);
5135 					if (fnamep != NULL && gotone == 1)
5136 						strlcpy(fnamep,
5137 						    pf->dsf_filename,
5138 						    sizeof(pf->dsf_filename));
5139 				} else
5140 					NFSD_DEBUG(4, "nfsrv_dsgetsockmnt "
5141 					    "err=%d\n", error);
5142 			}
5143 		}
5144 	}
5145 	if (error == 0 && gotone == 0)
5146 		error = ENOENT;
5147 
5148 	NFSD_DEBUG(4, "eo nfsrv_dsgetsockmnt: gotone=%d err=%d\n", gotone,
5149 	    error);
5150 	if (error == 0)
5151 		*mirrorcntp = gotone;
5152 	else {
5153 		if (gotone > 0 && dvpp != NULL) {
5154 			/*
5155 			 * If the error didn't occur on the first one and
5156 			 * dvpp != NULL, the one(s) prior to the failure will
5157 			 * have locked dvp's that need to be unlocked.
5158 			 */
5159 			for (i = 0; i < gotone; i++) {
5160 				NFSVOPUNLOCK(*dvpp);
5161 				*dvpp++ = NULL;
5162 			}
5163 		}
5164 		/*
5165 		 * If it found the vnode to be copied from before a failure,
5166 		 * it needs to be vput()'d.
5167 		 */
5168 		if (nvpp != NULL && *nvpp != NULL) {
5169 			vput(*nvpp);
5170 			*nvpp = NULL;
5171 		}
5172 	}
5173 	return (error);
5174 }
5175 
5176 /*
5177  * Set the extended attribute for the Change attribute.
5178  */
5179 static int
5180 nfsrv_setextattr(struct vnode *vp, struct nfsvattr *nap, NFSPROC_T *p)
5181 {
5182 	struct pnfsdsattr dsattr;
5183 	int error;
5184 
5185 	ASSERT_VOP_ELOCKED(vp, "nfsrv_setextattr vp");
5186 	dsattr.dsa_filerev = nap->na_filerev;
5187 	dsattr.dsa_size = nap->na_size;
5188 	dsattr.dsa_atime = nap->na_atime;
5189 	dsattr.dsa_mtime = nap->na_mtime;
5190 	dsattr.dsa_bytes = nap->na_bytes;
5191 	error = vn_extattr_set(vp, IO_NODELOCKED, EXTATTR_NAMESPACE_SYSTEM,
5192 	    "pnfsd.dsattr", sizeof(dsattr), (char *)&dsattr, p);
5193 	if (error != 0)
5194 		printf("pNFS: setextattr=%d\n", error);
5195 	return (error);
5196 }
5197 
5198 static int
5199 nfsrv_readdsrpc(fhandle_t *fhp, off_t off, int len, struct ucred *cred,
5200     NFSPROC_T *p, struct nfsmount *nmp, struct mbuf **mpp, struct mbuf **mpendp)
5201 {
5202 	uint32_t *tl;
5203 	struct nfsrv_descript *nd;
5204 	nfsv4stateid_t st;
5205 	struct mbuf *m, *m2;
5206 	int error = 0, retlen, tlen, trimlen;
5207 
5208 	NFSD_DEBUG(4, "in nfsrv_readdsrpc\n");
5209 	nd = malloc(sizeof(*nd), M_TEMP, M_WAITOK | M_ZERO);
5210 	*mpp = NULL;
5211 	/*
5212 	 * Use a stateid where other is an alternating 01010 pattern and
5213 	 * seqid is 0xffffffff.  This value is not defined as special by
5214 	 * the RFC and is used by the FreeBSD NFS server to indicate an
5215 	 * MDS->DS proxy operation.
5216 	 */
5217 	st.other[0] = 0x55555555;
5218 	st.other[1] = 0x55555555;
5219 	st.other[2] = 0x55555555;
5220 	st.seqid = 0xffffffff;
5221 	nfscl_reqstart(nd, NFSPROC_READDS, nmp, (u_int8_t *)fhp, sizeof(*fhp),
5222 	    NULL, NULL, 0, 0);
5223 	nfsm_stateidtom(nd, &st, NFSSTATEID_PUTSTATEID);
5224 	NFSM_BUILD(tl, uint32_t *, NFSX_UNSIGNED * 3);
5225 	txdr_hyper(off, tl);
5226 	*(tl + 2) = txdr_unsigned(len);
5227 	error = newnfs_request(nd, nmp, NULL, &nmp->nm_sockreq, NULL, p, cred,
5228 	    NFS_PROG, NFS_VER4, NULL, 1, NULL, NULL);
5229 	if (error != 0) {
5230 		free(nd, M_TEMP);
5231 		return (error);
5232 	}
5233 	if (nd->nd_repstat == 0) {
5234 		NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
5235 		NFSM_STRSIZ(retlen, len);
5236 		if (retlen > 0) {
5237 			/* Trim off the pre-data XDR from the mbuf chain. */
5238 			m = nd->nd_mrep;
5239 			while (m != NULL && m != nd->nd_md) {
5240 				if (m->m_next == nd->nd_md) {
5241 					m->m_next = NULL;
5242 					m_freem(nd->nd_mrep);
5243 					nd->nd_mrep = m = nd->nd_md;
5244 				} else
5245 					m = m->m_next;
5246 			}
5247 			if (m == NULL) {
5248 				printf("nfsrv_readdsrpc: busted mbuf list\n");
5249 				error = ENOENT;
5250 				goto nfsmout;
5251 			}
5252 
5253 			/*
5254 			 * Now, adjust first mbuf so that any XDR before the
5255 			 * read data is skipped over.
5256 			 */
5257 			trimlen = nd->nd_dpos - mtod(m, char *);
5258 			if (trimlen > 0) {
5259 				m->m_len -= trimlen;
5260 				NFSM_DATAP(m, trimlen);
5261 			}
5262 
5263 			/*
5264 			 * Truncate the mbuf chain at retlen bytes of data,
5265 			 * plus XDR padding that brings the length up to a
5266 			 * multiple of 4.
5267 			 */
5268 			tlen = NFSM_RNDUP(retlen);
5269 			do {
5270 				if (m->m_len >= tlen) {
5271 					m->m_len = tlen;
5272 					tlen = 0;
5273 					m2 = m->m_next;
5274 					m->m_next = NULL;
5275 					m_freem(m2);
5276 					break;
5277 				}
5278 				tlen -= m->m_len;
5279 				m = m->m_next;
5280 			} while (m != NULL);
5281 			if (tlen > 0) {
5282 				printf("nfsrv_readdsrpc: busted mbuf list\n");
5283 				error = ENOENT;
5284 				goto nfsmout;
5285 			}
5286 			*mpp = nd->nd_mrep;
5287 			*mpendp = m;
5288 			nd->nd_mrep = NULL;
5289 		}
5290 	} else
5291 		error = nd->nd_repstat;
5292 nfsmout:
5293 	/* If nd->nd_mrep is already NULL, this is a no-op. */
5294 	m_freem(nd->nd_mrep);
5295 	free(nd, M_TEMP);
5296 	NFSD_DEBUG(4, "nfsrv_readdsrpc error=%d\n", error);
5297 	return (error);
5298 }
5299 
5300 /*
5301  * Do a write RPC on a DS data file, using this structure for the arguments,
5302  * so that this function can be executed by a separate kernel process.
5303  */
5304 struct nfsrvwritedsdorpc {
5305 	int			done;
5306 	int			inprog;
5307 	struct task		tsk;
5308 	fhandle_t		fh;
5309 	off_t			off;
5310 	int			len;
5311 	struct nfsmount		*nmp;
5312 	struct ucred		*cred;
5313 	NFSPROC_T		*p;
5314 	struct mbuf		*m;
5315 	int			err;
5316 };
5317 
5318 static int
5319 nfsrv_writedsdorpc(struct nfsmount *nmp, fhandle_t *fhp, off_t off, int len,
5320     struct nfsvattr *nap, struct mbuf *m, struct ucred *cred, NFSPROC_T *p)
5321 {
5322 	uint32_t *tl;
5323 	struct nfsrv_descript *nd;
5324 	nfsattrbit_t attrbits;
5325 	nfsv4stateid_t st;
5326 	int commit, error, retlen;
5327 
5328 	nd = malloc(sizeof(*nd), M_TEMP, M_WAITOK | M_ZERO);
5329 	nfscl_reqstart(nd, NFSPROC_WRITE, nmp, (u_int8_t *)fhp,
5330 	    sizeof(fhandle_t), NULL, NULL, 0, 0);
5331 
5332 	/*
5333 	 * Use a stateid where other is an alternating 01010 pattern and
5334 	 * seqid is 0xffffffff.  This value is not defined as special by
5335 	 * the RFC and is used by the FreeBSD NFS server to indicate an
5336 	 * MDS->DS proxy operation.
5337 	 */
5338 	st.other[0] = 0x55555555;
5339 	st.other[1] = 0x55555555;
5340 	st.other[2] = 0x55555555;
5341 	st.seqid = 0xffffffff;
5342 	nfsm_stateidtom(nd, &st, NFSSTATEID_PUTSTATEID);
5343 	NFSM_BUILD(tl, u_int32_t *, NFSX_HYPER + 2 * NFSX_UNSIGNED);
5344 	txdr_hyper(off, tl);
5345 	tl += 2;
5346 	/*
5347 	 * Do all writes FileSync, since the server doesn't hold onto dirty
5348 	 * buffers.  Since clients should be accessing the DS servers directly
5349 	 * using the pNFS layouts, this just needs to work correctly as a
5350 	 * fallback.
5351 	 */
5352 	*tl++ = txdr_unsigned(NFSWRITE_FILESYNC);
5353 	*tl = txdr_unsigned(len);
5354 	NFSD_DEBUG(4, "nfsrv_writedsdorpc: len=%d\n", len);
5355 
5356 	/* Put data in mbuf chain. */
5357 	nd->nd_mb->m_next = m;
5358 
5359 	/* Set nd_mb and nd_bpos to end of data. */
5360 	while (m->m_next != NULL)
5361 		m = m->m_next;
5362 	nd->nd_mb = m;
5363 	nfsm_set(nd, m->m_len);
5364 	NFSD_DEBUG(4, "nfsrv_writedsdorpc: lastmb len=%d\n", m->m_len);
5365 
5366 	/* Do a Getattr for the attributes that change upon writing. */
5367 	NFSZERO_ATTRBIT(&attrbits);
5368 	NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_SIZE);
5369 	NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_CHANGE);
5370 	NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_TIMEACCESS);
5371 	NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_TIMEMODIFY);
5372 	NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_SPACEUSED);
5373 	NFSM_BUILD(tl, u_int32_t *, NFSX_UNSIGNED);
5374 	*tl = txdr_unsigned(NFSV4OP_GETATTR);
5375 	(void) nfsrv_putattrbit(nd, &attrbits);
5376 	error = newnfs_request(nd, nmp, NULL, &nmp->nm_sockreq, NULL, p,
5377 	    cred, NFS_PROG, NFS_VER4, NULL, 1, NULL, NULL);
5378 	if (error != 0) {
5379 		free(nd, M_TEMP);
5380 		return (error);
5381 	}
5382 	NFSD_DEBUG(4, "nfsrv_writedsdorpc: aft writerpc=%d\n", nd->nd_repstat);
5383 	/* Get rid of weak cache consistency data for now. */
5384 	if ((nd->nd_flag & (ND_NOMOREDATA | ND_NFSV4 | ND_V4WCCATTR)) ==
5385 	    (ND_NFSV4 | ND_V4WCCATTR)) {
5386 		error = nfsv4_loadattr(nd, NULL, nap, NULL, NULL, 0, NULL, NULL,
5387 		    NULL, NULL, NULL, 0, NULL, NULL, NULL, NULL, NULL);
5388 		NFSD_DEBUG(4, "nfsrv_writedsdorpc: wcc attr=%d\n", error);
5389 		if (error != 0)
5390 			goto nfsmout;
5391 		/*
5392 		 * Get rid of Op# and status for next op.
5393 		 */
5394 		NFSM_DISSECT(tl, uint32_t *, 2 * NFSX_UNSIGNED);
5395 		if (*++tl != 0)
5396 			nd->nd_flag |= ND_NOMOREDATA;
5397 	}
5398 	if (nd->nd_repstat == 0) {
5399 		NFSM_DISSECT(tl, uint32_t *, 2 * NFSX_UNSIGNED + NFSX_VERF);
5400 		retlen = fxdr_unsigned(int, *tl++);
5401 		commit = fxdr_unsigned(int, *tl);
5402 		if (commit != NFSWRITE_FILESYNC)
5403 			error = NFSERR_IO;
5404 		NFSD_DEBUG(4, "nfsrv_writedsdorpc:retlen=%d commit=%d err=%d\n",
5405 		    retlen, commit, error);
5406 	} else
5407 		error = nd->nd_repstat;
5408 	/* We have no use for the Write Verifier since we use FileSync. */
5409 
5410 	/*
5411 	 * Get the Change, Size, Access Time and Modify Time attributes and set
5412 	 * on the Metadata file, so its attributes will be what the file's
5413 	 * would be if it had been written.
5414 	 */
5415 	if (error == 0) {
5416 		NFSM_DISSECT(tl, uint32_t *, 2 * NFSX_UNSIGNED);
5417 		error = nfsv4_loadattr(nd, NULL, nap, NULL, NULL, 0, NULL, NULL,
5418 		    NULL, NULL, NULL, 0, NULL, NULL, NULL, NULL, NULL);
5419 	}
5420 	NFSD_DEBUG(4, "nfsrv_writedsdorpc: aft loadattr=%d\n", error);
5421 nfsmout:
5422 	m_freem(nd->nd_mrep);
5423 	free(nd, M_TEMP);
5424 	NFSD_DEBUG(4, "nfsrv_writedsdorpc error=%d\n", error);
5425 	return (error);
5426 }
5427 
5428 /*
5429  * Start up the thread that will execute nfsrv_writedsdorpc().
5430  */
5431 static void
5432 start_writedsdorpc(void *arg, int pending)
5433 {
5434 	struct nfsrvwritedsdorpc *drpc;
5435 
5436 	drpc = (struct nfsrvwritedsdorpc *)arg;
5437 	drpc->err = nfsrv_writedsdorpc(drpc->nmp, &drpc->fh, drpc->off,
5438 	    drpc->len, NULL, drpc->m, drpc->cred, drpc->p);
5439 	drpc->done = 1;
5440 	NFSD_DEBUG(4, "start_writedsdorpc: err=%d\n", drpc->err);
5441 }
5442 
5443 static int
5444 nfsrv_writedsrpc(fhandle_t *fhp, off_t off, int len, struct ucred *cred,
5445     NFSPROC_T *p, struct vnode *vp, struct nfsmount **nmpp, int mirrorcnt,
5446     struct mbuf **mpp, char *cp, int *failposp)
5447 {
5448 	struct nfsrvwritedsdorpc *drpc, *tdrpc = NULL;
5449 	struct nfsvattr na;
5450 	struct mbuf *m;
5451 	int error, i, offs, ret, timo;
5452 
5453 	NFSD_DEBUG(4, "in nfsrv_writedsrpc\n");
5454 	KASSERT(*mpp != NULL, ("nfsrv_writedsrpc: NULL mbuf chain"));
5455 	drpc = NULL;
5456 	if (mirrorcnt > 1)
5457 		tdrpc = drpc = malloc(sizeof(*drpc) * (mirrorcnt - 1), M_TEMP,
5458 		    M_WAITOK);
5459 
5460 	/* Calculate offset in mbuf chain that data starts. */
5461 	offs = cp - mtod(*mpp, char *);
5462 	NFSD_DEBUG(4, "nfsrv_writedsrpc: mcopy offs=%d len=%d\n", offs, len);
5463 
5464 	/*
5465 	 * Do the write RPC for every DS, using a separate kernel process
5466 	 * for every DS except the last one.
5467 	 */
5468 	error = 0;
5469 	for (i = 0; i < mirrorcnt - 1; i++, tdrpc++) {
5470 		tdrpc->done = 0;
5471 		NFSBCOPY(fhp, &tdrpc->fh, sizeof(*fhp));
5472 		tdrpc->off = off;
5473 		tdrpc->len = len;
5474 		tdrpc->nmp = *nmpp;
5475 		tdrpc->cred = cred;
5476 		tdrpc->p = p;
5477 		tdrpc->inprog = 0;
5478 		tdrpc->err = 0;
5479 		tdrpc->m = m_copym(*mpp, offs, NFSM_RNDUP(len), M_WAITOK);
5480 		ret = EIO;
5481 		if (nfs_pnfsiothreads != 0) {
5482 			ret = nfs_pnfsio(start_writedsdorpc, tdrpc);
5483 			NFSD_DEBUG(4, "nfsrv_writedsrpc: nfs_pnfsio=%d\n",
5484 			    ret);
5485 		}
5486 		if (ret != 0) {
5487 			ret = nfsrv_writedsdorpc(*nmpp, fhp, off, len, NULL,
5488 			    tdrpc->m, cred, p);
5489 			if (nfsds_failerr(ret) && *failposp == -1)
5490 				*failposp = i;
5491 			else if (error == 0 && ret != 0)
5492 				error = ret;
5493 		}
5494 		nmpp++;
5495 		fhp++;
5496 	}
5497 	m = m_copym(*mpp, offs, NFSM_RNDUP(len), M_WAITOK);
5498 	ret = nfsrv_writedsdorpc(*nmpp, fhp, off, len, &na, m, cred, p);
5499 	if (nfsds_failerr(ret) && *failposp == -1 && mirrorcnt > 1)
5500 		*failposp = mirrorcnt - 1;
5501 	else if (error == 0 && ret != 0)
5502 		error = ret;
5503 	if (error == 0)
5504 		error = nfsrv_setextattr(vp, &na, p);
5505 	NFSD_DEBUG(4, "nfsrv_writedsrpc: aft setextat=%d\n", error);
5506 	tdrpc = drpc;
5507 	timo = hz / 50;		/* Wait for 20msec. */
5508 	if (timo < 1)
5509 		timo = 1;
5510 	for (i = 0; i < mirrorcnt - 1; i++, tdrpc++) {
5511 		/* Wait for RPCs on separate threads to complete. */
5512 		while (tdrpc->inprog != 0 && tdrpc->done == 0)
5513 			tsleep(&tdrpc->tsk, PVFS, "srvwrds", timo);
5514 		if (nfsds_failerr(tdrpc->err) && *failposp == -1)
5515 			*failposp = i;
5516 		else if (error == 0 && tdrpc->err != 0)
5517 			error = tdrpc->err;
5518 	}
5519 	free(drpc, M_TEMP);
5520 	return (error);
5521 }
5522 
5523 /*
5524  * Do a allocate RPC on a DS data file, using this structure for the arguments,
5525  * so that this function can be executed by a separate kernel process.
5526  */
5527 struct nfsrvallocatedsdorpc {
5528 	int			done;
5529 	int			inprog;
5530 	struct task		tsk;
5531 	fhandle_t		fh;
5532 	off_t			off;
5533 	off_t			len;
5534 	struct nfsmount		*nmp;
5535 	struct ucred		*cred;
5536 	NFSPROC_T		*p;
5537 	int			err;
5538 };
5539 
5540 static int
5541 nfsrv_allocatedsdorpc(struct nfsmount *nmp, fhandle_t *fhp, off_t off,
5542     off_t len, struct nfsvattr *nap, struct ucred *cred, NFSPROC_T *p)
5543 {
5544 	uint32_t *tl;
5545 	struct nfsrv_descript *nd;
5546 	nfsattrbit_t attrbits;
5547 	nfsv4stateid_t st;
5548 	int error;
5549 
5550 	nd = malloc(sizeof(*nd), M_TEMP, M_WAITOK | M_ZERO);
5551 	nfscl_reqstart(nd, NFSPROC_ALLOCATE, nmp, (u_int8_t *)fhp,
5552 	    sizeof(fhandle_t), NULL, NULL, 0, 0);
5553 
5554 	/*
5555 	 * Use a stateid where other is an alternating 01010 pattern and
5556 	 * seqid is 0xffffffff.  This value is not defined as special by
5557 	 * the RFC and is used by the FreeBSD NFS server to indicate an
5558 	 * MDS->DS proxy operation.
5559 	 */
5560 	st.other[0] = 0x55555555;
5561 	st.other[1] = 0x55555555;
5562 	st.other[2] = 0x55555555;
5563 	st.seqid = 0xffffffff;
5564 	nfsm_stateidtom(nd, &st, NFSSTATEID_PUTSTATEID);
5565 	NFSM_BUILD(tl, uint32_t *, 2 * NFSX_HYPER + NFSX_UNSIGNED);
5566 	txdr_hyper(off, tl); tl += 2;
5567 	txdr_hyper(len, tl); tl += 2;
5568 	NFSD_DEBUG(4, "nfsrv_allocatedsdorpc: len=%jd\n", (intmax_t)len);
5569 
5570 	*tl = txdr_unsigned(NFSV4OP_GETATTR);
5571 	NFSGETATTR_ATTRBIT(&attrbits);
5572 	nfsrv_putattrbit(nd, &attrbits);
5573 	error = newnfs_request(nd, nmp, NULL, &nmp->nm_sockreq, NULL, p,
5574 	    cred, NFS_PROG, NFS_VER4, NULL, 1, NULL, NULL);
5575 	if (error != 0) {
5576 		free(nd, M_TEMP);
5577 		return (error);
5578 	}
5579 	NFSD_DEBUG(4, "nfsrv_allocatedsdorpc: aft allocaterpc=%d\n",
5580 	    nd->nd_repstat);
5581 	if (nd->nd_repstat == 0) {
5582 		NFSM_DISSECT(tl, uint32_t *, 2 * NFSX_UNSIGNED);
5583 		error = nfsv4_loadattr(nd, NULL, nap, NULL, NULL, 0, NULL, NULL,
5584 		    NULL, NULL, NULL, 0, NULL, NULL, NULL, NULL, NULL);
5585 	} else
5586 		error = nd->nd_repstat;
5587 	NFSD_DEBUG(4, "nfsrv_allocatedsdorpc: aft loadattr=%d\n", error);
5588 nfsmout:
5589 	m_freem(nd->nd_mrep);
5590 	free(nd, M_TEMP);
5591 	NFSD_DEBUG(4, "nfsrv_allocatedsdorpc error=%d\n", error);
5592 	return (error);
5593 }
5594 
5595 /*
5596  * Start up the thread that will execute nfsrv_allocatedsdorpc().
5597  */
5598 static void
5599 start_allocatedsdorpc(void *arg, int pending)
5600 {
5601 	struct nfsrvallocatedsdorpc *drpc;
5602 
5603 	drpc = (struct nfsrvallocatedsdorpc *)arg;
5604 	drpc->err = nfsrv_allocatedsdorpc(drpc->nmp, &drpc->fh, drpc->off,
5605 	    drpc->len, NULL, drpc->cred, drpc->p);
5606 	drpc->done = 1;
5607 	NFSD_DEBUG(4, "start_allocatedsdorpc: err=%d\n", drpc->err);
5608 }
5609 
5610 static int
5611 nfsrv_allocatedsrpc(fhandle_t *fhp, off_t off, off_t len, struct ucred *cred,
5612     NFSPROC_T *p, struct vnode *vp, struct nfsmount **nmpp, int mirrorcnt,
5613     int *failposp)
5614 {
5615 	struct nfsrvallocatedsdorpc *drpc, *tdrpc = NULL;
5616 	struct nfsvattr na;
5617 	int error, i, ret, timo;
5618 
5619 	NFSD_DEBUG(4, "in nfsrv_allocatedsrpc\n");
5620 	drpc = NULL;
5621 	if (mirrorcnt > 1)
5622 		tdrpc = drpc = malloc(sizeof(*drpc) * (mirrorcnt - 1), M_TEMP,
5623 		    M_WAITOK);
5624 
5625 	/*
5626 	 * Do the allocate RPC for every DS, using a separate kernel process
5627 	 * for every DS except the last one.
5628 	 */
5629 	error = 0;
5630 	for (i = 0; i < mirrorcnt - 1; i++, tdrpc++) {
5631 		tdrpc->done = 0;
5632 		NFSBCOPY(fhp, &tdrpc->fh, sizeof(*fhp));
5633 		tdrpc->off = off;
5634 		tdrpc->len = len;
5635 		tdrpc->nmp = *nmpp;
5636 		tdrpc->cred = cred;
5637 		tdrpc->p = p;
5638 		tdrpc->inprog = 0;
5639 		tdrpc->err = 0;
5640 		ret = EIO;
5641 		if (nfs_pnfsiothreads != 0) {
5642 			ret = nfs_pnfsio(start_allocatedsdorpc, tdrpc);
5643 			NFSD_DEBUG(4, "nfsrv_allocatedsrpc: nfs_pnfsio=%d\n",
5644 			    ret);
5645 		}
5646 		if (ret != 0) {
5647 			ret = nfsrv_allocatedsdorpc(*nmpp, fhp, off, len, NULL,
5648 			    cred, p);
5649 			if (nfsds_failerr(ret) && *failposp == -1)
5650 				*failposp = i;
5651 			else if (error == 0 && ret != 0)
5652 				error = ret;
5653 		}
5654 		nmpp++;
5655 		fhp++;
5656 	}
5657 	ret = nfsrv_allocatedsdorpc(*nmpp, fhp, off, len, &na, cred, p);
5658 	if (nfsds_failerr(ret) && *failposp == -1 && mirrorcnt > 1)
5659 		*failposp = mirrorcnt - 1;
5660 	else if (error == 0 && ret != 0)
5661 		error = ret;
5662 	if (error == 0)
5663 		error = nfsrv_setextattr(vp, &na, p);
5664 	NFSD_DEBUG(4, "nfsrv_allocatedsrpc: aft setextat=%d\n", error);
5665 	tdrpc = drpc;
5666 	timo = hz / 50;		/* Wait for 20msec. */
5667 	if (timo < 1)
5668 		timo = 1;
5669 	for (i = 0; i < mirrorcnt - 1; i++, tdrpc++) {
5670 		/* Wait for RPCs on separate threads to complete. */
5671 		while (tdrpc->inprog != 0 && tdrpc->done == 0)
5672 			tsleep(&tdrpc->tsk, PVFS, "srvalds", timo);
5673 		if (nfsds_failerr(tdrpc->err) && *failposp == -1)
5674 			*failposp = i;
5675 		else if (error == 0 && tdrpc->err != 0)
5676 			error = tdrpc->err;
5677 	}
5678 	free(drpc, M_TEMP);
5679 	return (error);
5680 }
5681 
5682 static int
5683 nfsrv_setattrdsdorpc(fhandle_t *fhp, struct ucred *cred, NFSPROC_T *p,
5684     struct vnode *vp, struct nfsmount *nmp, struct nfsvattr *nap,
5685     struct nfsvattr *dsnap)
5686 {
5687 	uint32_t *tl;
5688 	struct nfsrv_descript *nd;
5689 	nfsv4stateid_t st;
5690 	nfsattrbit_t attrbits;
5691 	int error;
5692 
5693 	NFSD_DEBUG(4, "in nfsrv_setattrdsdorpc\n");
5694 	nd = malloc(sizeof(*nd), M_TEMP, M_WAITOK | M_ZERO);
5695 	/*
5696 	 * Use a stateid where other is an alternating 01010 pattern and
5697 	 * seqid is 0xffffffff.  This value is not defined as special by
5698 	 * the RFC and is used by the FreeBSD NFS server to indicate an
5699 	 * MDS->DS proxy operation.
5700 	 */
5701 	st.other[0] = 0x55555555;
5702 	st.other[1] = 0x55555555;
5703 	st.other[2] = 0x55555555;
5704 	st.seqid = 0xffffffff;
5705 	nfscl_reqstart(nd, NFSPROC_SETATTR, nmp, (u_int8_t *)fhp, sizeof(*fhp),
5706 	    NULL, NULL, 0, 0);
5707 	nfsm_stateidtom(nd, &st, NFSSTATEID_PUTSTATEID);
5708 	nfscl_fillsattr(nd, &nap->na_vattr, vp, NFSSATTR_FULL, 0);
5709 
5710 	/* Do a Getattr for the attributes that change due to writing. */
5711 	NFSZERO_ATTRBIT(&attrbits);
5712 	NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_SIZE);
5713 	NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_CHANGE);
5714 	NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_TIMEACCESS);
5715 	NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_TIMEMODIFY);
5716 	NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_SPACEUSED);
5717 	NFSM_BUILD(tl, u_int32_t *, NFSX_UNSIGNED);
5718 	*tl = txdr_unsigned(NFSV4OP_GETATTR);
5719 	(void) nfsrv_putattrbit(nd, &attrbits);
5720 	error = newnfs_request(nd, nmp, NULL, &nmp->nm_sockreq, NULL, p, cred,
5721 	    NFS_PROG, NFS_VER4, NULL, 1, NULL, NULL);
5722 	if (error != 0) {
5723 		free(nd, M_TEMP);
5724 		return (error);
5725 	}
5726 	NFSD_DEBUG(4, "nfsrv_setattrdsdorpc: aft setattrrpc=%d\n",
5727 	    nd->nd_repstat);
5728 	/* Get rid of weak cache consistency data for now. */
5729 	if ((nd->nd_flag & (ND_NOMOREDATA | ND_NFSV4 | ND_V4WCCATTR)) ==
5730 	    (ND_NFSV4 | ND_V4WCCATTR)) {
5731 		error = nfsv4_loadattr(nd, NULL, dsnap, NULL, NULL, 0, NULL,
5732 		    NULL, NULL, NULL, NULL, 0, NULL, NULL, NULL, NULL, NULL);
5733 		NFSD_DEBUG(4, "nfsrv_setattrdsdorpc: wcc attr=%d\n", error);
5734 		if (error != 0)
5735 			goto nfsmout;
5736 		/*
5737 		 * Get rid of Op# and status for next op.
5738 		 */
5739 		NFSM_DISSECT(tl, uint32_t *, 2 * NFSX_UNSIGNED);
5740 		if (*++tl != 0)
5741 			nd->nd_flag |= ND_NOMOREDATA;
5742 	}
5743 	error = nfsrv_getattrbits(nd, &attrbits, NULL, NULL);
5744 	if (error != 0)
5745 		goto nfsmout;
5746 	if (nd->nd_repstat != 0)
5747 		error = nd->nd_repstat;
5748 	/*
5749 	 * Get the Change, Size, Access Time and Modify Time attributes and set
5750 	 * on the Metadata file, so its attributes will be what the file's
5751 	 * would be if it had been written.
5752 	 */
5753 	if (error == 0) {
5754 		NFSM_DISSECT(tl, uint32_t *, 2 * NFSX_UNSIGNED);
5755 		error = nfsv4_loadattr(nd, NULL, dsnap, NULL, NULL, 0, NULL,
5756 		    NULL, NULL, NULL, NULL, 0, NULL, NULL, NULL, NULL, NULL);
5757 	}
5758 	NFSD_DEBUG(4, "nfsrv_setattrdsdorpc: aft setattr loadattr=%d\n", error);
5759 nfsmout:
5760 	m_freem(nd->nd_mrep);
5761 	free(nd, M_TEMP);
5762 	NFSD_DEBUG(4, "nfsrv_setattrdsdorpc error=%d\n", error);
5763 	return (error);
5764 }
5765 
5766 struct nfsrvsetattrdsdorpc {
5767 	int			done;
5768 	int			inprog;
5769 	struct task		tsk;
5770 	fhandle_t		fh;
5771 	struct nfsmount		*nmp;
5772 	struct vnode		*vp;
5773 	struct ucred		*cred;
5774 	NFSPROC_T		*p;
5775 	struct nfsvattr		na;
5776 	struct nfsvattr		dsna;
5777 	int			err;
5778 };
5779 
5780 /*
5781  * Start up the thread that will execute nfsrv_setattrdsdorpc().
5782  */
5783 static void
5784 start_setattrdsdorpc(void *arg, int pending)
5785 {
5786 	struct nfsrvsetattrdsdorpc *drpc;
5787 
5788 	drpc = (struct nfsrvsetattrdsdorpc *)arg;
5789 	drpc->err = nfsrv_setattrdsdorpc(&drpc->fh, drpc->cred, drpc->p,
5790 	    drpc->vp, drpc->nmp, &drpc->na, &drpc->dsna);
5791 	drpc->done = 1;
5792 }
5793 
5794 static int
5795 nfsrv_setattrdsrpc(fhandle_t *fhp, struct ucred *cred, NFSPROC_T *p,
5796     struct vnode *vp, struct nfsmount **nmpp, int mirrorcnt,
5797     struct nfsvattr *nap, int *failposp)
5798 {
5799 	struct nfsrvsetattrdsdorpc *drpc, *tdrpc = NULL;
5800 	struct nfsvattr na;
5801 	int error, i, ret, timo;
5802 
5803 	NFSD_DEBUG(4, "in nfsrv_setattrdsrpc\n");
5804 	drpc = NULL;
5805 	if (mirrorcnt > 1)
5806 		tdrpc = drpc = malloc(sizeof(*drpc) * (mirrorcnt - 1), M_TEMP,
5807 		    M_WAITOK);
5808 
5809 	/*
5810 	 * Do the setattr RPC for every DS, using a separate kernel process
5811 	 * for every DS except the last one.
5812 	 */
5813 	error = 0;
5814 	for (i = 0; i < mirrorcnt - 1; i++, tdrpc++) {
5815 		tdrpc->done = 0;
5816 		tdrpc->inprog = 0;
5817 		NFSBCOPY(fhp, &tdrpc->fh, sizeof(*fhp));
5818 		tdrpc->nmp = *nmpp;
5819 		tdrpc->vp = vp;
5820 		tdrpc->cred = cred;
5821 		tdrpc->p = p;
5822 		tdrpc->na = *nap;
5823 		tdrpc->err = 0;
5824 		ret = EIO;
5825 		if (nfs_pnfsiothreads != 0) {
5826 			ret = nfs_pnfsio(start_setattrdsdorpc, tdrpc);
5827 			NFSD_DEBUG(4, "nfsrv_setattrdsrpc: nfs_pnfsio=%d\n",
5828 			    ret);
5829 		}
5830 		if (ret != 0) {
5831 			ret = nfsrv_setattrdsdorpc(fhp, cred, p, vp, *nmpp, nap,
5832 			    &na);
5833 			if (nfsds_failerr(ret) && *failposp == -1)
5834 				*failposp = i;
5835 			else if (error == 0 && ret != 0)
5836 				error = ret;
5837 		}
5838 		nmpp++;
5839 		fhp++;
5840 	}
5841 	ret = nfsrv_setattrdsdorpc(fhp, cred, p, vp, *nmpp, nap, &na);
5842 	if (nfsds_failerr(ret) && *failposp == -1 && mirrorcnt > 1)
5843 		*failposp = mirrorcnt - 1;
5844 	else if (error == 0 && ret != 0)
5845 		error = ret;
5846 	if (error == 0)
5847 		error = nfsrv_setextattr(vp, &na, p);
5848 	NFSD_DEBUG(4, "nfsrv_setattrdsrpc: aft setextat=%d\n", error);
5849 	tdrpc = drpc;
5850 	timo = hz / 50;		/* Wait for 20msec. */
5851 	if (timo < 1)
5852 		timo = 1;
5853 	for (i = 0; i < mirrorcnt - 1; i++, tdrpc++) {
5854 		/* Wait for RPCs on separate threads to complete. */
5855 		while (tdrpc->inprog != 0 && tdrpc->done == 0)
5856 			tsleep(&tdrpc->tsk, PVFS, "srvsads", timo);
5857 		if (nfsds_failerr(tdrpc->err) && *failposp == -1)
5858 			*failposp = i;
5859 		else if (error == 0 && tdrpc->err != 0)
5860 			error = tdrpc->err;
5861 	}
5862 	free(drpc, M_TEMP);
5863 	return (error);
5864 }
5865 
5866 /*
5867  * Do a Setattr of an NFSv4 ACL on the DS file.
5868  */
5869 static int
5870 nfsrv_setacldsdorpc(fhandle_t *fhp, struct ucred *cred, NFSPROC_T *p,
5871     struct vnode *vp, struct nfsmount *nmp, struct acl *aclp)
5872 {
5873 	struct nfsrv_descript *nd;
5874 	nfsv4stateid_t st;
5875 	nfsattrbit_t attrbits;
5876 	int error;
5877 
5878 	NFSD_DEBUG(4, "in nfsrv_setacldsdorpc\n");
5879 	nd = malloc(sizeof(*nd), M_TEMP, M_WAITOK | M_ZERO);
5880 	/*
5881 	 * Use a stateid where other is an alternating 01010 pattern and
5882 	 * seqid is 0xffffffff.  This value is not defined as special by
5883 	 * the RFC and is used by the FreeBSD NFS server to indicate an
5884 	 * MDS->DS proxy operation.
5885 	 */
5886 	st.other[0] = 0x55555555;
5887 	st.other[1] = 0x55555555;
5888 	st.other[2] = 0x55555555;
5889 	st.seqid = 0xffffffff;
5890 	nfscl_reqstart(nd, NFSPROC_SETACL, nmp, (u_int8_t *)fhp, sizeof(*fhp),
5891 	    NULL, NULL, 0, 0);
5892 	nfsm_stateidtom(nd, &st, NFSSTATEID_PUTSTATEID);
5893 	NFSZERO_ATTRBIT(&attrbits);
5894 	NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_ACL);
5895 	/*
5896 	 * The "vp" argument to nfsv4_fillattr() is only used for vnode_type(),
5897 	 * so passing in the metadata "vp" will be ok, since it is of
5898 	 * the same type (VREG).
5899 	 */
5900 	nfsv4_fillattr(nd, NULL, vp, aclp, NULL, NULL, 0, &attrbits, NULL,
5901 	    NULL, 0, 0, 0, 0, 0, NULL);
5902 	error = newnfs_request(nd, nmp, NULL, &nmp->nm_sockreq, NULL, p, cred,
5903 	    NFS_PROG, NFS_VER4, NULL, 1, NULL, NULL);
5904 	if (error != 0) {
5905 		free(nd, M_TEMP);
5906 		return (error);
5907 	}
5908 	NFSD_DEBUG(4, "nfsrv_setacldsdorpc: aft setaclrpc=%d\n",
5909 	    nd->nd_repstat);
5910 	error = nd->nd_repstat;
5911 	m_freem(nd->nd_mrep);
5912 	free(nd, M_TEMP);
5913 	return (error);
5914 }
5915 
5916 struct nfsrvsetacldsdorpc {
5917 	int			done;
5918 	int			inprog;
5919 	struct task		tsk;
5920 	fhandle_t		fh;
5921 	struct nfsmount		*nmp;
5922 	struct vnode		*vp;
5923 	struct ucred		*cred;
5924 	NFSPROC_T		*p;
5925 	struct acl		*aclp;
5926 	int			err;
5927 };
5928 
5929 /*
5930  * Start up the thread that will execute nfsrv_setacldsdorpc().
5931  */
5932 static void
5933 start_setacldsdorpc(void *arg, int pending)
5934 {
5935 	struct nfsrvsetacldsdorpc *drpc;
5936 
5937 	drpc = (struct nfsrvsetacldsdorpc *)arg;
5938 	drpc->err = nfsrv_setacldsdorpc(&drpc->fh, drpc->cred, drpc->p,
5939 	    drpc->vp, drpc->nmp, drpc->aclp);
5940 	drpc->done = 1;
5941 }
5942 
5943 static int
5944 nfsrv_setacldsrpc(fhandle_t *fhp, struct ucred *cred, NFSPROC_T *p,
5945     struct vnode *vp, struct nfsmount **nmpp, int mirrorcnt, struct acl *aclp,
5946     int *failposp)
5947 {
5948 	struct nfsrvsetacldsdorpc *drpc, *tdrpc = NULL;
5949 	int error, i, ret, timo;
5950 
5951 	NFSD_DEBUG(4, "in nfsrv_setacldsrpc\n");
5952 	drpc = NULL;
5953 	if (mirrorcnt > 1)
5954 		tdrpc = drpc = malloc(sizeof(*drpc) * (mirrorcnt - 1), M_TEMP,
5955 		    M_WAITOK);
5956 
5957 	/*
5958 	 * Do the setattr RPC for every DS, using a separate kernel process
5959 	 * for every DS except the last one.
5960 	 */
5961 	error = 0;
5962 	for (i = 0; i < mirrorcnt - 1; i++, tdrpc++) {
5963 		tdrpc->done = 0;
5964 		tdrpc->inprog = 0;
5965 		NFSBCOPY(fhp, &tdrpc->fh, sizeof(*fhp));
5966 		tdrpc->nmp = *nmpp;
5967 		tdrpc->vp = vp;
5968 		tdrpc->cred = cred;
5969 		tdrpc->p = p;
5970 		tdrpc->aclp = aclp;
5971 		tdrpc->err = 0;
5972 		ret = EIO;
5973 		if (nfs_pnfsiothreads != 0) {
5974 			ret = nfs_pnfsio(start_setacldsdorpc, tdrpc);
5975 			NFSD_DEBUG(4, "nfsrv_setacldsrpc: nfs_pnfsio=%d\n",
5976 			    ret);
5977 		}
5978 		if (ret != 0) {
5979 			ret = nfsrv_setacldsdorpc(fhp, cred, p, vp, *nmpp,
5980 			    aclp);
5981 			if (nfsds_failerr(ret) && *failposp == -1)
5982 				*failposp = i;
5983 			else if (error == 0 && ret != 0)
5984 				error = ret;
5985 		}
5986 		nmpp++;
5987 		fhp++;
5988 	}
5989 	ret = nfsrv_setacldsdorpc(fhp, cred, p, vp, *nmpp, aclp);
5990 	if (nfsds_failerr(ret) && *failposp == -1 && mirrorcnt > 1)
5991 		*failposp = mirrorcnt - 1;
5992 	else if (error == 0 && ret != 0)
5993 		error = ret;
5994 	NFSD_DEBUG(4, "nfsrv_setacldsrpc: aft setextat=%d\n", error);
5995 	tdrpc = drpc;
5996 	timo = hz / 50;		/* Wait for 20msec. */
5997 	if (timo < 1)
5998 		timo = 1;
5999 	for (i = 0; i < mirrorcnt - 1; i++, tdrpc++) {
6000 		/* Wait for RPCs on separate threads to complete. */
6001 		while (tdrpc->inprog != 0 && tdrpc->done == 0)
6002 			tsleep(&tdrpc->tsk, PVFS, "srvacds", timo);
6003 		if (nfsds_failerr(tdrpc->err) && *failposp == -1)
6004 			*failposp = i;
6005 		else if (error == 0 && tdrpc->err != 0)
6006 			error = tdrpc->err;
6007 	}
6008 	free(drpc, M_TEMP);
6009 	return (error);
6010 }
6011 
6012 /*
6013  * Getattr call to the DS for the attributes that change due to writing.
6014  */
6015 static int
6016 nfsrv_getattrdsrpc(fhandle_t *fhp, struct ucred *cred, NFSPROC_T *p,
6017     struct vnode *vp, struct nfsmount *nmp, struct nfsvattr *nap)
6018 {
6019 	struct nfsrv_descript *nd;
6020 	int error;
6021 	nfsattrbit_t attrbits;
6022 
6023 	NFSD_DEBUG(4, "in nfsrv_getattrdsrpc\n");
6024 	nd = malloc(sizeof(*nd), M_TEMP, M_WAITOK | M_ZERO);
6025 	nfscl_reqstart(nd, NFSPROC_GETATTR, nmp, (u_int8_t *)fhp,
6026 	    sizeof(fhandle_t), NULL, NULL, 0, 0);
6027 	NFSZERO_ATTRBIT(&attrbits);
6028 	NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_SIZE);
6029 	NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_CHANGE);
6030 	NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_TIMEACCESS);
6031 	NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_TIMEMODIFY);
6032 	NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_SPACEUSED);
6033 	(void) nfsrv_putattrbit(nd, &attrbits);
6034 	error = newnfs_request(nd, nmp, NULL, &nmp->nm_sockreq, NULL, p, cred,
6035 	    NFS_PROG, NFS_VER4, NULL, 1, NULL, NULL);
6036 	if (error != 0) {
6037 		free(nd, M_TEMP);
6038 		return (error);
6039 	}
6040 	NFSD_DEBUG(4, "nfsrv_getattrdsrpc: aft getattrrpc=%d\n",
6041 	    nd->nd_repstat);
6042 	if (nd->nd_repstat == 0) {
6043 		error = nfsv4_loadattr(nd, NULL, nap, NULL, NULL, 0,
6044 		    NULL, NULL, NULL, NULL, NULL, 0, NULL, NULL, NULL,
6045 		    NULL, NULL);
6046 		/*
6047 		 * We can only save the updated values in the extended
6048 		 * attribute if the vp is exclusively locked.
6049 		 * This should happen when any of the following operations
6050 		 * occur on the vnode:
6051 		 *    Close, Delegreturn, LayoutCommit, LayoutReturn
6052 		 * As such, the updated extended attribute should get saved
6053 		 * before nfsrv_checkdsattr() returns 0 and allows the cached
6054 		 * attributes to be returned without calling this function.
6055 		 */
6056 		if (error == 0 && VOP_ISLOCKED(vp) == LK_EXCLUSIVE) {
6057 			error = nfsrv_setextattr(vp, nap, p);
6058 			NFSD_DEBUG(4, "nfsrv_getattrdsrpc: aft setextat=%d\n",
6059 			    error);
6060 		}
6061 	} else
6062 		error = nd->nd_repstat;
6063 	m_freem(nd->nd_mrep);
6064 	free(nd, M_TEMP);
6065 	NFSD_DEBUG(4, "nfsrv_getattrdsrpc error=%d\n", error);
6066 	return (error);
6067 }
6068 
6069 /*
6070  * Seek call to a DS.
6071  */
6072 static int
6073 nfsrv_seekdsrpc(fhandle_t *fhp, off_t *offp, int content, bool *eofp,
6074     struct ucred *cred, NFSPROC_T *p, struct nfsmount *nmp)
6075 {
6076 	uint32_t *tl;
6077 	struct nfsrv_descript *nd;
6078 	nfsv4stateid_t st;
6079 	int error;
6080 
6081 	NFSD_DEBUG(4, "in nfsrv_seekdsrpc\n");
6082 	/*
6083 	 * Use a stateid where other is an alternating 01010 pattern and
6084 	 * seqid is 0xffffffff.  This value is not defined as special by
6085 	 * the RFC and is used by the FreeBSD NFS server to indicate an
6086 	 * MDS->DS proxy operation.
6087 	 */
6088 	st.other[0] = 0x55555555;
6089 	st.other[1] = 0x55555555;
6090 	st.other[2] = 0x55555555;
6091 	st.seqid = 0xffffffff;
6092 	nd = malloc(sizeof(*nd), M_TEMP, M_WAITOK | M_ZERO);
6093 	nfscl_reqstart(nd, NFSPROC_SEEKDS, nmp, (u_int8_t *)fhp,
6094 	    sizeof(fhandle_t), NULL, NULL, 0, 0);
6095 	nfsm_stateidtom(nd, &st, NFSSTATEID_PUTSTATEID);
6096 	NFSM_BUILD(tl, uint32_t *, NFSX_HYPER + NFSX_UNSIGNED);
6097 	txdr_hyper(*offp, tl); tl += 2;
6098 	*tl = txdr_unsigned(content);
6099 	error = newnfs_request(nd, nmp, NULL, &nmp->nm_sockreq, NULL, p, cred,
6100 	    NFS_PROG, NFS_VER4, NULL, 1, NULL, NULL);
6101 	if (error != 0) {
6102 		free(nd, M_TEMP);
6103 		return (error);
6104 	}
6105 	NFSD_DEBUG(4, "nfsrv_seekdsrpc: aft seekrpc=%d\n", nd->nd_repstat);
6106 	if (nd->nd_repstat == 0) {
6107 		NFSM_DISSECT(tl, uint32_t *, NFSX_UNSIGNED + NFSX_HYPER);
6108 		if (*tl++ == newnfs_true)
6109 			*eofp = true;
6110 		else
6111 			*eofp = false;
6112 		*offp = fxdr_hyper(tl);
6113 	} else
6114 		error = nd->nd_repstat;
6115 nfsmout:
6116 	m_freem(nd->nd_mrep);
6117 	free(nd, M_TEMP);
6118 	NFSD_DEBUG(4, "nfsrv_seekdsrpc error=%d\n", error);
6119 	return (error);
6120 }
6121 
6122 /*
6123  * Get the device id and file handle for a DS file.
6124  */
6125 int
6126 nfsrv_dsgetdevandfh(struct vnode *vp, NFSPROC_T *p, int *mirrorcntp,
6127     fhandle_t *fhp, char *devid)
6128 {
6129 	int buflen, error;
6130 	char *buf;
6131 
6132 	buflen = 1024;
6133 	buf = malloc(buflen, M_TEMP, M_WAITOK);
6134 	error = nfsrv_dsgetsockmnt(vp, 0, buf, &buflen, mirrorcntp, p, NULL,
6135 	    fhp, devid, NULL, NULL, NULL, NULL, NULL, NULL);
6136 	free(buf, M_TEMP);
6137 	return (error);
6138 }
6139 
6140 /*
6141  * Do a Lookup against the DS for the filename.
6142  */
6143 static int
6144 nfsrv_pnfslookupds(struct vnode *vp, struct vnode *dvp, struct pnfsdsfile *pf,
6145     struct vnode **nvpp, NFSPROC_T *p)
6146 {
6147 	struct nameidata named;
6148 	struct ucred *tcred;
6149 	char *bufp;
6150 	u_long *hashp;
6151 	struct vnode *nvp;
6152 	int error;
6153 
6154 	tcred = newnfs_getcred();
6155 	named.ni_cnd.cn_nameiop = LOOKUP;
6156 	named.ni_cnd.cn_lkflags = LK_SHARED | LK_RETRY;
6157 	named.ni_cnd.cn_cred = tcred;
6158 	named.ni_cnd.cn_thread = p;
6159 	named.ni_cnd.cn_flags = ISLASTCN | LOCKPARENT | LOCKLEAF | SAVENAME;
6160 	nfsvno_setpathbuf(&named, &bufp, &hashp);
6161 	named.ni_cnd.cn_nameptr = bufp;
6162 	named.ni_cnd.cn_namelen = strlen(pf->dsf_filename);
6163 	strlcpy(bufp, pf->dsf_filename, NAME_MAX);
6164 	NFSD_DEBUG(4, "nfsrv_pnfslookupds: filename=%s\n", bufp);
6165 	error = VOP_LOOKUP(dvp, &nvp, &named.ni_cnd);
6166 	NFSD_DEBUG(4, "nfsrv_pnfslookupds: aft LOOKUP=%d\n", error);
6167 	NFSFREECRED(tcred);
6168 	nfsvno_relpathbuf(&named);
6169 	if (error == 0)
6170 		*nvpp = nvp;
6171 	NFSD_DEBUG(4, "eo nfsrv_pnfslookupds=%d\n", error);
6172 	return (error);
6173 }
6174 
6175 /*
6176  * Set the file handle to the correct one.
6177  */
6178 static void
6179 nfsrv_pnfssetfh(struct vnode *vp, struct pnfsdsfile *pf, char *devid,
6180     char *fnamep, struct vnode *nvp, NFSPROC_T *p)
6181 {
6182 	struct nfsnode *np;
6183 	int ret = 0;
6184 
6185 	np = VTONFS(nvp);
6186 	NFSBCOPY(np->n_fhp->nfh_fh, &pf->dsf_fh, NFSX_MYFH);
6187 	/*
6188 	 * We can only do a vn_set_extattr() if the vnode is exclusively
6189 	 * locked and vn_start_write() has been done.  If devid != NULL or
6190 	 * fnamep != NULL or the vnode is shared locked, vn_start_write()
6191 	 * may not have been done.
6192 	 * If not done now, it will be done on a future call.
6193 	 */
6194 	if (devid == NULL && fnamep == NULL && NFSVOPISLOCKED(vp) ==
6195 	    LK_EXCLUSIVE)
6196 		ret = vn_extattr_set(vp, IO_NODELOCKED,
6197 		    EXTATTR_NAMESPACE_SYSTEM, "pnfsd.dsfile", sizeof(*pf),
6198 		    (char *)pf, p);
6199 	NFSD_DEBUG(4, "eo nfsrv_pnfssetfh=%d\n", ret);
6200 }
6201 
6202 /*
6203  * Cause RPCs waiting on "nmp" to fail.  This is called for a DS mount point
6204  * when the DS has failed.
6205  */
6206 void
6207 nfsrv_killrpcs(struct nfsmount *nmp)
6208 {
6209 
6210 	/*
6211 	 * Call newnfs_nmcancelreqs() to cause
6212 	 * any RPCs in progress on the mount point to
6213 	 * fail.
6214 	 * This will cause any process waiting for an
6215 	 * RPC to complete while holding a vnode lock
6216 	 * on the mounted-on vnode (such as "df" or
6217 	 * a non-forced "umount") to fail.
6218 	 * This will unlock the mounted-on vnode so
6219 	 * a forced dismount can succeed.
6220 	 * The NFSMNTP_CANCELRPCS flag should be set when this function is
6221 	 * called.
6222 	 */
6223 	newnfs_nmcancelreqs(nmp);
6224 }
6225 
6226 /*
6227  * Sum up the statfs info for each of the DSs, so that the client will
6228  * receive the total for all DSs.
6229  */
6230 static int
6231 nfsrv_pnfsstatfs(struct statfs *sf, struct mount *mp)
6232 {
6233 	struct statfs *tsf;
6234 	struct nfsdevice *ds;
6235 	struct vnode **dvpp, **tdvpp, *dvp;
6236 	uint64_t tot;
6237 	int cnt, error = 0, i;
6238 
6239 	if (nfsrv_devidcnt <= 0)
6240 		return (ENXIO);
6241 	dvpp = mallocarray(nfsrv_devidcnt, sizeof(*dvpp), M_TEMP, M_WAITOK);
6242 	tsf = malloc(sizeof(*tsf), M_TEMP, M_WAITOK);
6243 
6244 	/* Get an array of the dvps for the DSs. */
6245 	tdvpp = dvpp;
6246 	i = 0;
6247 	NFSDDSLOCK();
6248 	/* First, search for matches for same file system. */
6249 	TAILQ_FOREACH(ds, &nfsrv_devidhead, nfsdev_list) {
6250 		if (ds->nfsdev_nmp != NULL && ds->nfsdev_mdsisset != 0 &&
6251 		    fsidcmp(&ds->nfsdev_mdsfsid, &mp->mnt_stat.f_fsid) == 0) {
6252 			if (++i > nfsrv_devidcnt)
6253 				break;
6254 			*tdvpp++ = ds->nfsdev_dvp;
6255 		}
6256 	}
6257 	/*
6258 	 * If no matches for same file system, total all servers not assigned
6259 	 * to a file system.
6260 	 */
6261 	if (i == 0) {
6262 		TAILQ_FOREACH(ds, &nfsrv_devidhead, nfsdev_list) {
6263 			if (ds->nfsdev_nmp != NULL &&
6264 			    ds->nfsdev_mdsisset == 0) {
6265 				if (++i > nfsrv_devidcnt)
6266 					break;
6267 				*tdvpp++ = ds->nfsdev_dvp;
6268 			}
6269 		}
6270 	}
6271 	NFSDDSUNLOCK();
6272 	cnt = i;
6273 
6274 	/* Do a VFS_STATFS() for each of the DSs and sum them up. */
6275 	tdvpp = dvpp;
6276 	for (i = 0; i < cnt && error == 0; i++) {
6277 		dvp = *tdvpp++;
6278 		error = VFS_STATFS(dvp->v_mount, tsf);
6279 		if (error == 0) {
6280 			if (sf->f_bsize == 0) {
6281 				if (tsf->f_bsize > 0)
6282 					sf->f_bsize = tsf->f_bsize;
6283 				else
6284 					sf->f_bsize = 8192;
6285 			}
6286 			if (tsf->f_blocks > 0) {
6287 				if (sf->f_bsize != tsf->f_bsize) {
6288 					tot = tsf->f_blocks * tsf->f_bsize;
6289 					sf->f_blocks += (tot / sf->f_bsize);
6290 				} else
6291 					sf->f_blocks += tsf->f_blocks;
6292 			}
6293 			if (tsf->f_bfree > 0) {
6294 				if (sf->f_bsize != tsf->f_bsize) {
6295 					tot = tsf->f_bfree * tsf->f_bsize;
6296 					sf->f_bfree += (tot / sf->f_bsize);
6297 				} else
6298 					sf->f_bfree += tsf->f_bfree;
6299 			}
6300 			if (tsf->f_bavail > 0) {
6301 				if (sf->f_bsize != tsf->f_bsize) {
6302 					tot = tsf->f_bavail * tsf->f_bsize;
6303 					sf->f_bavail += (tot / sf->f_bsize);
6304 				} else
6305 					sf->f_bavail += tsf->f_bavail;
6306 			}
6307 		}
6308 	}
6309 	free(tsf, M_TEMP);
6310 	free(dvpp, M_TEMP);
6311 	return (error);
6312 }
6313 
6314 /*
6315  * Set an NFSv4 acl.
6316  */
6317 int
6318 nfsrv_setacl(struct vnode *vp, NFSACL_T *aclp, struct ucred *cred, NFSPROC_T *p)
6319 {
6320 	int error;
6321 
6322 	if (nfsrv_useacl == 0 || nfs_supportsnfsv4acls(vp) == 0) {
6323 		error = NFSERR_ATTRNOTSUPP;
6324 		goto out;
6325 	}
6326 	/*
6327 	 * With NFSv4 ACLs, chmod(2) may need to add additional entries.
6328 	 * Make sure it has enough room for that - splitting every entry
6329 	 * into two and appending "canonical six" entries at the end.
6330 	 * Cribbed out of kern/vfs_acl.c - Rick M.
6331 	 */
6332 	if (aclp->acl_cnt > (ACL_MAX_ENTRIES - 6) / 2) {
6333 		error = NFSERR_ATTRNOTSUPP;
6334 		goto out;
6335 	}
6336 	error = VOP_SETACL(vp, ACL_TYPE_NFS4, aclp, cred, p);
6337 	if (error == 0) {
6338 		error = nfsrv_dssetacl(vp, aclp, cred, p);
6339 		if (error == ENOENT)
6340 			error = 0;
6341 	}
6342 
6343 out:
6344 	NFSEXITCODE(error);
6345 	return (error);
6346 }
6347 
6348 /*
6349  * Seek vnode op call (actually it is a VOP_IOCTL()).
6350  * This function is called with the vnode locked, but unlocks and vrele()s
6351  * the vp before returning.
6352  */
6353 int
6354 nfsvno_seek(struct nfsrv_descript *nd, struct vnode *vp, u_long cmd,
6355     off_t *offp, int content, bool *eofp, struct ucred *cred, NFSPROC_T *p)
6356 {
6357 	struct nfsvattr at;
6358 	int error, ret;
6359 
6360 	ASSERT_VOP_LOCKED(vp, "nfsvno_seek vp");
6361 	/*
6362 	 * Attempt to seek on a DS file. A return of ENOENT implies
6363 	 * there is no DS file to seek on.
6364 	 */
6365 	error = nfsrv_proxyds(vp, 0, 0, cred, p, NFSPROC_SEEKDS, NULL,
6366 	    NULL, NULL, NULL, NULL, offp, content, eofp);
6367 	if (error != ENOENT) {
6368 		vput(vp);
6369 		return (error);
6370 	}
6371 
6372 	/*
6373 	 * Do the VOP_IOCTL() call.  For the case where *offp == file_size,
6374 	 * VOP_IOCTL() will return ENXIO.  However, the correct reply for
6375 	 * NFSv4.2 is *eofp == true and error == 0 for this case.
6376 	 */
6377 	NFSVOPUNLOCK(vp);
6378 	error = VOP_IOCTL(vp, cmd, offp, 0, cred, p);
6379 	*eofp = false;
6380 	if (error == ENXIO || (error == 0 && cmd == FIOSEEKHOLE)) {
6381 		/* Handle the cases where we might be at EOF. */
6382 		ret = nfsvno_getattr(vp, &at, nd, p, 0, NULL);
6383 		if (ret == 0 && *offp == at.na_size) {
6384 			*eofp = true;
6385 			error = 0;
6386 		}
6387 		if (ret != 0 && error == 0)
6388 			error = ret;
6389 	}
6390 	vrele(vp);
6391 	NFSEXITCODE(error);
6392 	return (error);
6393 }
6394 
6395 /*
6396  * Allocate vnode op call.
6397  */
6398 int
6399 nfsvno_allocate(struct vnode *vp, off_t off, off_t len, struct ucred *cred,
6400     NFSPROC_T *p)
6401 {
6402 	int error, trycnt;
6403 
6404 	ASSERT_VOP_ELOCKED(vp, "nfsvno_allocate vp");
6405 	/*
6406 	 * Attempt to allocate on a DS file. A return of ENOENT implies
6407 	 * there is no DS file to allocate on.
6408 	 */
6409 	error = nfsrv_proxyds(vp, off, 0, cred, p, NFSPROC_ALLOCATE, NULL,
6410 	    NULL, NULL, NULL, NULL, &len, 0, NULL);
6411 	if (error != ENOENT)
6412 		return (error);
6413 	error = 0;
6414 
6415 	/*
6416 	 * Do the actual VOP_ALLOCATE(), looping a reasonable number of
6417 	 * times to achieve completion.
6418 	 */
6419 	trycnt = 0;
6420 	while (error == 0 && len > 0 && trycnt++ < 20)
6421 		error = VOP_ALLOCATE(vp, &off, &len);
6422 	if (error == 0 && len > 0)
6423 		error = NFSERR_IO;
6424 	NFSEXITCODE(error);
6425 	return (error);
6426 }
6427 
6428 /*
6429  * Get Extended Atribute vnode op into an mbuf list.
6430  */
6431 int
6432 nfsvno_getxattr(struct vnode *vp, char *name, uint32_t maxresp,
6433     struct ucred *cred, uint64_t flag, int maxextsiz, struct thread *p,
6434     struct mbuf **mpp, struct mbuf **mpendp, int *lenp)
6435 {
6436 	struct iovec *iv;
6437 	struct uio io, *uiop = &io;
6438 	struct mbuf *m, *m2;
6439 	int alen, error, len, tlen;
6440 	size_t siz;
6441 
6442 	/* First, find out the size of the extended attribute. */
6443 	error = VOP_GETEXTATTR(vp, EXTATTR_NAMESPACE_USER, name, NULL,
6444 	    &siz, cred, p);
6445 	if (error != 0)
6446 		return (NFSERR_NOXATTR);
6447 	if (siz > maxresp - NFS_MAXXDR)
6448 		return (NFSERR_XATTR2BIG);
6449 	len = siz;
6450 	tlen = NFSM_RNDUP(len);
6451 	if (tlen > 0) {
6452 		/*
6453 		 * If cnt > MCLBYTES and the reply will not be saved, use
6454 		 * ext_pgs mbufs for TLS.
6455 		 * For NFSv4.0, we do not know for sure if the reply will
6456 		 * be saved, so do not use ext_pgs mbufs for NFSv4.0.
6457 		 * Always use ext_pgs mbufs if ND_EXTPG is set.
6458 		 */
6459 		if ((flag & ND_EXTPG) != 0 || (tlen > MCLBYTES &&
6460 		    (flag & (ND_TLS | ND_SAVEREPLY)) == ND_TLS &&
6461 		    (flag & (ND_NFSV4 | ND_NFSV41)) != ND_NFSV4))
6462 			uiop->uio_iovcnt = nfsrv_createiovec_extpgs(tlen,
6463 			    maxextsiz, &m, &m2, &iv);
6464 		else
6465 			uiop->uio_iovcnt = nfsrv_createiovec(tlen, &m, &m2,
6466 			    &iv);
6467 		uiop->uio_iov = iv;
6468 	} else {
6469 		uiop->uio_iovcnt = 0;
6470 		uiop->uio_iov = iv = NULL;
6471 		m = m2 = NULL;
6472 	}
6473 	uiop->uio_offset = 0;
6474 	uiop->uio_resid = tlen;
6475 	uiop->uio_rw = UIO_READ;
6476 	uiop->uio_segflg = UIO_SYSSPACE;
6477 	uiop->uio_td = p;
6478 #ifdef MAC
6479 	error = mac_vnode_check_getextattr(cred, vp, EXTATTR_NAMESPACE_USER,
6480 	    name);
6481 	if (error != 0)
6482 		goto out;
6483 #endif
6484 
6485 	if (tlen > 0)
6486 		error = VOP_GETEXTATTR(vp, EXTATTR_NAMESPACE_USER, name, uiop,
6487 		    NULL, cred, p);
6488 	if (error != 0)
6489 		goto out;
6490 	if (uiop->uio_resid > 0) {
6491 		alen = tlen;
6492 		len = tlen - uiop->uio_resid;
6493 		tlen = NFSM_RNDUP(len);
6494 		if (alen != tlen)
6495 			printf("nfsvno_getxattr: weird size read\n");
6496 		if (tlen == 0) {
6497 			m_freem(m);
6498 			m = m2 = NULL;
6499 		} else if (alen != tlen || tlen != len)
6500 			m2 = nfsrv_adj(m, alen - tlen, tlen - len);
6501 	}
6502 	*lenp = len;
6503 	*mpp = m;
6504 	*mpendp = m2;
6505 
6506 out:
6507 	if (error != 0) {
6508 		if (m != NULL)
6509 			m_freem(m);
6510 		*lenp = 0;
6511 	}
6512 	free(iv, M_TEMP);
6513 	NFSEXITCODE(error);
6514 	return (error);
6515 }
6516 
6517 /*
6518  * Set Extended attribute vnode op from an mbuf list.
6519  */
6520 int
6521 nfsvno_setxattr(struct vnode *vp, char *name, int len, struct mbuf *m,
6522     char *cp, struct ucred *cred, struct thread *p)
6523 {
6524 	struct iovec *iv;
6525 	struct uio uio, *uiop = &uio;
6526 	int cnt, error;
6527 
6528 	error = 0;
6529 #ifdef MAC
6530 	error = mac_vnode_check_setextattr(cred, vp, EXTATTR_NAMESPACE_USER,
6531 	    name);
6532 #endif
6533 	if (error != 0)
6534 		goto out;
6535 
6536 	uiop->uio_rw = UIO_WRITE;
6537 	uiop->uio_segflg = UIO_SYSSPACE;
6538 	uiop->uio_td = p;
6539 	uiop->uio_offset = 0;
6540 	uiop->uio_resid = len;
6541 	if (len > 0) {
6542 		error = nfsrv_createiovecw(len, m, cp, &iv, &cnt);
6543 		uiop->uio_iov = iv;
6544 		uiop->uio_iovcnt = cnt;
6545 	} else {
6546 		uiop->uio_iov = iv = NULL;
6547 		uiop->uio_iovcnt = 0;
6548 	}
6549 	if (error == 0) {
6550 		error = VOP_SETEXTATTR(vp, EXTATTR_NAMESPACE_USER, name, uiop,
6551 		    cred, p);
6552 		free(iv, M_TEMP);
6553 	}
6554 
6555 out:
6556 	NFSEXITCODE(error);
6557 	return (error);
6558 }
6559 
6560 /*
6561  * Remove Extended attribute vnode op.
6562  */
6563 int
6564 nfsvno_rmxattr(struct nfsrv_descript *nd, struct vnode *vp, char *name,
6565     struct ucred *cred, struct thread *p)
6566 {
6567 	int error;
6568 
6569 	/*
6570 	 * Get rid of any delegations.  I am not sure why this is required,
6571 	 * but RFC-8276 says so.
6572 	 */
6573 	error = nfsrv_checkremove(vp, 0, nd, nd->nd_clientid, p);
6574 	if (error != 0)
6575 		goto out;
6576 #ifdef MAC
6577 	error = mac_vnode_check_deleteextattr(cred, vp, EXTATTR_NAMESPACE_USER,
6578 	    name);
6579 	if (error != 0)
6580 		goto out;
6581 #endif
6582 
6583 	error = VOP_DELETEEXTATTR(vp, EXTATTR_NAMESPACE_USER, name, cred, p);
6584 	if (error == EOPNOTSUPP)
6585 		error = VOP_SETEXTATTR(vp, EXTATTR_NAMESPACE_USER, name, NULL,
6586 		    cred, p);
6587 out:
6588 	NFSEXITCODE(error);
6589 	return (error);
6590 }
6591 
6592 /*
6593  * List Extended Atribute vnode op into an mbuf list.
6594  */
6595 int
6596 nfsvno_listxattr(struct vnode *vp, uint64_t cookie, struct ucred *cred,
6597     struct thread *p, u_char **bufp, uint32_t *lenp, bool *eofp)
6598 {
6599 	struct iovec iv;
6600 	struct uio io;
6601 	int error;
6602 	size_t siz;
6603 
6604 	*bufp = NULL;
6605 	/* First, find out the size of the extended attribute. */
6606 	error = VOP_LISTEXTATTR(vp, EXTATTR_NAMESPACE_USER, NULL, &siz, cred,
6607 	    p);
6608 	if (error != 0)
6609 		return (NFSERR_NOXATTR);
6610 	if (siz <= cookie) {
6611 		*lenp = 0;
6612 		*eofp = true;
6613 		goto out;
6614 	}
6615 	if (siz > cookie + *lenp) {
6616 		siz = cookie + *lenp;
6617 		*eofp = false;
6618 	} else
6619 		*eofp = true;
6620 	/* Just choose a sanity limit of 10Mbytes for malloc(M_TEMP). */
6621 	if (siz > 10 * 1024 * 1024) {
6622 		error = NFSERR_XATTR2BIG;
6623 		goto out;
6624 	}
6625 	*bufp = malloc(siz, M_TEMP, M_WAITOK);
6626 	iv.iov_base = *bufp;
6627 	iv.iov_len = siz;
6628 	io.uio_iovcnt = 1;
6629 	io.uio_iov = &iv;
6630 	io.uio_offset = 0;
6631 	io.uio_resid = siz;
6632 	io.uio_rw = UIO_READ;
6633 	io.uio_segflg = UIO_SYSSPACE;
6634 	io.uio_td = p;
6635 #ifdef MAC
6636 	error = mac_vnode_check_listextattr(cred, vp, EXTATTR_NAMESPACE_USER);
6637 	if (error != 0)
6638 		goto out;
6639 #endif
6640 
6641 	error = VOP_LISTEXTATTR(vp, EXTATTR_NAMESPACE_USER, &io, NULL, cred,
6642 	    p);
6643 	if (error != 0)
6644 		goto out;
6645 	if (io.uio_resid > 0)
6646 		siz -= io.uio_resid;
6647 	*lenp = siz;
6648 
6649 out:
6650 	if (error != 0) {
6651 		free(*bufp, M_TEMP);
6652 		*bufp = NULL;
6653 	}
6654 	NFSEXITCODE(error);
6655 	return (error);
6656 }
6657 
6658 /*
6659  * Trim trailing data off the mbuf list being built.
6660  */
6661 void
6662 nfsm_trimtrailing(struct nfsrv_descript *nd, struct mbuf *mb, char *bpos,
6663     int bextpg, int bextpgsiz)
6664 {
6665 	vm_page_t pg;
6666 	int fullpgsiz, i;
6667 
6668 	if (mb->m_next != NULL) {
6669 		m_freem(mb->m_next);
6670 		mb->m_next = NULL;
6671 	}
6672 	if ((mb->m_flags & M_EXTPG) != 0) {
6673 		KASSERT(bextpg >= 0 && bextpg < mb->m_epg_npgs,
6674 		    ("nfsm_trimtrailing: bextpg out of range"));
6675 		KASSERT(bpos == (char *)(void *)
6676 		    PHYS_TO_DMAP(mb->m_epg_pa[bextpg]) + PAGE_SIZE - bextpgsiz,
6677 		    ("nfsm_trimtrailing: bextpgsiz bad!"));
6678 
6679 		/* First, get rid of any pages after this position. */
6680 		for (i = mb->m_epg_npgs - 1; i > bextpg; i--) {
6681 			pg = PHYS_TO_VM_PAGE(mb->m_epg_pa[i]);
6682 			vm_page_unwire_noq(pg);
6683 			vm_page_free(pg);
6684 		}
6685 		mb->m_epg_npgs = bextpg + 1;
6686 		if (bextpg == 0)
6687 			fullpgsiz = PAGE_SIZE - mb->m_epg_1st_off;
6688 		else
6689 			fullpgsiz = PAGE_SIZE;
6690 		mb->m_epg_last_len = fullpgsiz - bextpgsiz;
6691 		mb->m_len = m_epg_pagelen(mb, 0, mb->m_epg_1st_off);
6692 		for (i = 1; i < mb->m_epg_npgs; i++)
6693 			mb->m_len += m_epg_pagelen(mb, i, 0);
6694 		nd->nd_bextpgsiz = bextpgsiz;
6695 		nd->nd_bextpg = bextpg;
6696 	} else
6697 		mb->m_len = bpos - mtod(mb, char *);
6698 	nd->nd_mb = mb;
6699 	nd->nd_bpos = bpos;
6700 }
6701 
6702 
6703 /*
6704  * Check to see if a put file handle operation should test for
6705  * NFSERR_WRONGSEC, although NFSv3 actually returns NFSERR_AUTHERR.
6706  * When Open is the next operation, NFSERR_WRONGSEC cannot be
6707  * replied for the Open cases that use a component.  Thia can
6708  * be identified by the fact that the file handle's type is VDIR.
6709  */
6710 bool
6711 nfsrv_checkwrongsec(struct nfsrv_descript *nd, int nextop, enum vtype vtyp)
6712 {
6713 
6714 	if ((nd->nd_flag & ND_NFSV4) == 0) {
6715 		if (nd->nd_procnum == NFSPROC_FSINFO)
6716 			return (false);
6717 		return (true);
6718 	}
6719 
6720 	if ((nd->nd_flag & ND_LASTOP) != 0)
6721 		return (false);
6722 
6723 	if (nextop == NFSV4OP_PUTROOTFH || nextop == NFSV4OP_PUTFH ||
6724 	    nextop == NFSV4OP_PUTPUBFH || nextop == NFSV4OP_RESTOREFH ||
6725 	    nextop == NFSV4OP_LOOKUP || nextop == NFSV4OP_LOOKUPP ||
6726 	    nextop == NFSV4OP_SECINFO || nextop == NFSV4OP_SECINFONONAME)
6727 		return (false);
6728 	if (nextop == NFSV4OP_OPEN && vtyp == VDIR)
6729 		return (false);
6730 	return (true);
6731 }
6732 
6733 extern int (*nfsd_call_nfsd)(struct thread *, struct nfssvc_args *);
6734 
6735 /*
6736  * Called once to initialize data structures...
6737  */
6738 static int
6739 nfsd_modevent(module_t mod, int type, void *data)
6740 {
6741 	int error = 0, i;
6742 	static int loaded = 0;
6743 
6744 	switch (type) {
6745 	case MOD_LOAD:
6746 		if (loaded)
6747 			goto out;
6748 		newnfs_portinit();
6749 		for (i = 0; i < NFSRVCACHE_HASHSIZE; i++) {
6750 			mtx_init(&nfsrchash_table[i].mtx, "nfsrtc", NULL,
6751 			    MTX_DEF);
6752 			mtx_init(&nfsrcahash_table[i].mtx, "nfsrtca", NULL,
6753 			    MTX_DEF);
6754 		}
6755 		mtx_init(&nfsrc_udpmtx, "nfsuc", NULL, MTX_DEF);
6756 		mtx_init(&nfs_v4root_mutex, "nfs4rt", NULL, MTX_DEF);
6757 		mtx_init(&nfsv4root_mnt.mnt_mtx, "nfs4mnt", NULL, MTX_DEF);
6758 		mtx_init(&nfsrv_dontlistlock_mtx, "nfs4dnl", NULL, MTX_DEF);
6759 		mtx_init(&nfsrv_recalllock_mtx, "nfs4rec", NULL, MTX_DEF);
6760 		lockinit(&nfsv4root_mnt.mnt_explock, PVFS, "explock", 0, 0);
6761 		nfsrvd_initcache();
6762 		nfsd_init();
6763 		NFSD_LOCK();
6764 		nfsrvd_init(0);
6765 		NFSD_UNLOCK();
6766 		nfsd_mntinit();
6767 #ifdef VV_DISABLEDELEG
6768 		vn_deleg_ops.vndeleg_recall = nfsd_recalldelegation;
6769 		vn_deleg_ops.vndeleg_disable = nfsd_disabledelegation;
6770 #endif
6771 		nfsd_call_servertimer = nfsrv_servertimer;
6772 		nfsd_call_nfsd = nfssvc_nfsd;
6773 		loaded = 1;
6774 		break;
6775 
6776 	case MOD_UNLOAD:
6777 		if (newnfs_numnfsd != 0) {
6778 			error = EBUSY;
6779 			break;
6780 		}
6781 
6782 #ifdef VV_DISABLEDELEG
6783 		vn_deleg_ops.vndeleg_recall = NULL;
6784 		vn_deleg_ops.vndeleg_disable = NULL;
6785 #endif
6786 		nfsd_call_servertimer = NULL;
6787 		nfsd_call_nfsd = NULL;
6788 
6789 		/* Clean out all NFSv4 state. */
6790 		nfsrv_throwawayallstate(curthread);
6791 
6792 		/* Clean the NFS server reply cache */
6793 		nfsrvd_cleancache();
6794 
6795 		/* Free up the krpc server pool. */
6796 		if (nfsrvd_pool != NULL)
6797 			svcpool_destroy(nfsrvd_pool);
6798 
6799 		/* and get rid of the locks */
6800 		for (i = 0; i < NFSRVCACHE_HASHSIZE; i++) {
6801 			mtx_destroy(&nfsrchash_table[i].mtx);
6802 			mtx_destroy(&nfsrcahash_table[i].mtx);
6803 		}
6804 		mtx_destroy(&nfsrc_udpmtx);
6805 		mtx_destroy(&nfs_v4root_mutex);
6806 		mtx_destroy(&nfsv4root_mnt.mnt_mtx);
6807 		mtx_destroy(&nfsrv_dontlistlock_mtx);
6808 		mtx_destroy(&nfsrv_recalllock_mtx);
6809 		for (i = 0; i < nfsrv_sessionhashsize; i++)
6810 			mtx_destroy(&nfssessionhash[i].mtx);
6811 		if (nfslayouthash != NULL) {
6812 			for (i = 0; i < nfsrv_layouthashsize; i++)
6813 				mtx_destroy(&nfslayouthash[i].mtx);
6814 			free(nfslayouthash, M_NFSDSESSION);
6815 		}
6816 		lockdestroy(&nfsv4root_mnt.mnt_explock);
6817 		free(nfsclienthash, M_NFSDCLIENT);
6818 		free(nfslockhash, M_NFSDLOCKFILE);
6819 		free(nfssessionhash, M_NFSDSESSION);
6820 		loaded = 0;
6821 		break;
6822 	default:
6823 		error = EOPNOTSUPP;
6824 		break;
6825 	}
6826 
6827 out:
6828 	NFSEXITCODE(error);
6829 	return (error);
6830 }
6831 static moduledata_t nfsd_mod = {
6832 	"nfsd",
6833 	nfsd_modevent,
6834 	NULL,
6835 };
6836 DECLARE_MODULE(nfsd, nfsd_mod, SI_SUB_VFS, SI_ORDER_ANY);
6837 
6838 /* So that loader and kldload(2) can find us, wherever we are.. */
6839 MODULE_VERSION(nfsd, 1);
6840 MODULE_DEPEND(nfsd, nfscommon, 1, 1, 1);
6841 MODULE_DEPEND(nfsd, nfslockd, 1, 1, 1);
6842 MODULE_DEPEND(nfsd, krpc, 1, 1, 1);
6843 MODULE_DEPEND(nfsd, nfssvc, 1, 1, 1);
6844