xref: /freebsd/sys/fs/nfsserver/nfs_nfsdport.c (revision ab1e0d2410ece7d391a5b1e2cbc9d1e9857c2fdb)
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 <sys/hash.h>
50 #include <sys/sysctl.h>
51 #include <nlm/nlm_prot.h>
52 #include <nlm/nlm.h>
53 
54 FEATURE(nfsd, "NFSv4 server");
55 
56 extern u_int32_t newnfs_true, newnfs_false, newnfs_xdrneg1;
57 extern int nfsrv_useacl;
58 extern int newnfs_numnfsd;
59 extern struct mount nfsv4root_mnt;
60 extern struct nfsrv_stablefirst nfsrv_stablefirst;
61 extern void (*nfsd_call_servertimer)(void);
62 extern SVCPOOL	*nfsrvd_pool;
63 extern struct nfsv4lock nfsd_suspend_lock;
64 extern struct nfsclienthashhead *nfsclienthash;
65 extern struct nfslockhashhead *nfslockhash;
66 extern struct nfssessionhash *nfssessionhash;
67 extern int nfsrv_sessionhashsize;
68 extern struct nfsstatsv1 nfsstatsv1;
69 extern struct nfslayouthash *nfslayouthash;
70 extern int nfsrv_layouthashsize;
71 extern struct mtx nfsrv_dslock_mtx;
72 extern int nfs_pnfsiothreads;
73 extern struct nfsdontlisthead nfsrv_dontlisthead;
74 extern volatile int nfsrv_dontlistlen;
75 extern volatile int nfsrv_devidcnt;
76 extern int nfsrv_maxpnfsmirror;
77 struct vfsoptlist nfsv4root_opt, nfsv4root_newopt;
78 NFSDLOCKMUTEX;
79 NFSSTATESPINLOCK;
80 struct nfsrchash_bucket nfsrchash_table[NFSRVCACHE_HASHSIZE];
81 struct nfsrchash_bucket nfsrcahash_table[NFSRVCACHE_HASHSIZE];
82 struct mtx nfsrc_udpmtx;
83 struct mtx nfs_v4root_mutex;
84 struct mtx nfsrv_dontlistlock_mtx;
85 struct mtx nfsrv_recalllock_mtx;
86 struct nfsrvfh nfs_rootfh, nfs_pubfh;
87 int nfs_pubfhset = 0, nfs_rootfhset = 0;
88 struct proc *nfsd_master_proc = NULL;
89 int nfsd_debuglevel = 0;
90 static pid_t nfsd_master_pid = (pid_t)-1;
91 static char nfsd_master_comm[MAXCOMLEN + 1];
92 static struct timeval nfsd_master_start;
93 static uint32_t nfsv4_sysid = 0;
94 static fhandle_t zerofh;
95 
96 static int nfssvc_srvcall(struct thread *, struct nfssvc_args *,
97     struct ucred *);
98 
99 int nfsrv_enable_crossmntpt = 1;
100 static int nfs_commit_blks;
101 static int nfs_commit_miss;
102 extern int nfsrv_issuedelegs;
103 extern int nfsrv_dolocallocks;
104 extern int nfsd_enable_stringtouid;
105 extern struct nfsdevicehead nfsrv_devidhead;
106 
107 static void nfsrv_pnfscreate(struct vnode *, struct vattr *, struct ucred *,
108     NFSPROC_T *);
109 static void nfsrv_pnfsremovesetup(struct vnode *, NFSPROC_T *, struct vnode **,
110     int *, char *, fhandle_t *);
111 static void nfsrv_pnfsremove(struct vnode **, int, char *, fhandle_t *,
112     NFSPROC_T *);
113 static int nfsrv_proxyds(struct nfsrv_descript *, struct vnode *, off_t, int,
114     struct ucred *, struct thread *, int, struct mbuf **, char *,
115     struct mbuf **, struct nfsvattr *, struct acl *);
116 static int nfsrv_setextattr(struct vnode *, struct nfsvattr *, NFSPROC_T *);
117 static int nfsrv_readdsrpc(fhandle_t *, off_t, int, struct ucred *,
118     NFSPROC_T *, struct nfsmount *, struct mbuf **, struct mbuf **);
119 static int nfsrv_writedsrpc(fhandle_t *, off_t, int, struct ucred *,
120     NFSPROC_T *, struct vnode *, struct nfsmount **, int, struct mbuf **,
121     char *, int *);
122 static int nfsrv_setacldsrpc(fhandle_t *, struct ucred *, NFSPROC_T *,
123     struct vnode *, struct nfsmount **, int, struct acl *, int *);
124 static int nfsrv_setattrdsrpc(fhandle_t *, struct ucred *, NFSPROC_T *,
125     struct vnode *, struct nfsmount **, int, struct nfsvattr *, int *);
126 static int nfsrv_getattrdsrpc(fhandle_t *, struct ucred *, NFSPROC_T *,
127     struct vnode *, struct nfsmount *, struct nfsvattr *);
128 static int nfsrv_putfhname(fhandle_t *, char *);
129 static int nfsrv_pnfslookupds(struct vnode *, struct vnode *,
130     struct pnfsdsfile *, struct vnode **, NFSPROC_T *);
131 static void nfsrv_pnfssetfh(struct vnode *, struct pnfsdsfile *, char *, char *,
132     struct vnode *, NFSPROC_T *);
133 static int nfsrv_dsremove(struct vnode *, char *, struct ucred *, NFSPROC_T *);
134 static int nfsrv_dssetacl(struct vnode *, struct acl *, struct ucred *,
135     NFSPROC_T *);
136 static int nfsrv_pnfsstatfs(struct statfs *, struct mount *);
137 
138 int nfs_pnfsio(task_fn_t *, void *);
139 
140 SYSCTL_NODE(_vfs, OID_AUTO, nfsd, CTLFLAG_RW, 0, "NFS server");
141 SYSCTL_INT(_vfs_nfsd, OID_AUTO, mirrormnt, CTLFLAG_RW,
142     &nfsrv_enable_crossmntpt, 0, "Enable nfsd to cross mount points");
143 SYSCTL_INT(_vfs_nfsd, OID_AUTO, commit_blks, CTLFLAG_RW, &nfs_commit_blks,
144     0, "");
145 SYSCTL_INT(_vfs_nfsd, OID_AUTO, commit_miss, CTLFLAG_RW, &nfs_commit_miss,
146     0, "");
147 SYSCTL_INT(_vfs_nfsd, OID_AUTO, issue_delegations, CTLFLAG_RW,
148     &nfsrv_issuedelegs, 0, "Enable nfsd to issue delegations");
149 SYSCTL_INT(_vfs_nfsd, OID_AUTO, enable_locallocks, CTLFLAG_RW,
150     &nfsrv_dolocallocks, 0, "Enable nfsd to acquire local locks on files");
151 SYSCTL_INT(_vfs_nfsd, OID_AUTO, debuglevel, CTLFLAG_RW, &nfsd_debuglevel,
152     0, "Debug level for NFS server");
153 SYSCTL_INT(_vfs_nfsd, OID_AUTO, enable_stringtouid, CTLFLAG_RW,
154     &nfsd_enable_stringtouid, 0, "Enable nfsd to accept numeric owner_names");
155 static int nfsrv_pnfsgetdsattr = 1;
156 SYSCTL_INT(_vfs_nfsd, OID_AUTO, pnfsgetdsattr, CTLFLAG_RW,
157     &nfsrv_pnfsgetdsattr, 0, "When set getattr gets DS attributes via RPC");
158 
159 /*
160  * nfsrv_dsdirsize can only be increased and only when the nfsd threads are
161  * not running.
162  * The dsN subdirectories for the increased values must have been created
163  * on all DS servers before this increase is done.
164  */
165 u_int	nfsrv_dsdirsize = 20;
166 static int
167 sysctl_dsdirsize(SYSCTL_HANDLER_ARGS)
168 {
169 	int error, newdsdirsize;
170 
171 	newdsdirsize = nfsrv_dsdirsize;
172 	error = sysctl_handle_int(oidp, &newdsdirsize, 0, req);
173 	if (error != 0 || req->newptr == NULL)
174 		return (error);
175 	if (newdsdirsize <= nfsrv_dsdirsize || newdsdirsize > 10000 ||
176 	    newnfs_numnfsd != 0)
177 		return (EINVAL);
178 	nfsrv_dsdirsize = newdsdirsize;
179 	return (0);
180 }
181 SYSCTL_PROC(_vfs_nfsd, OID_AUTO, dsdirsize, CTLTYPE_UINT | CTLFLAG_RW, 0,
182     sizeof(nfsrv_dsdirsize), sysctl_dsdirsize, "IU",
183     "Number of dsN subdirs on the DS servers");
184 
185 #define	MAX_REORDERED_RPC	16
186 #define	NUM_HEURISTIC		1031
187 #define	NHUSE_INIT		64
188 #define	NHUSE_INC		16
189 #define	NHUSE_MAX		2048
190 
191 static struct nfsheur {
192 	struct vnode *nh_vp;	/* vp to match (unreferenced pointer) */
193 	off_t nh_nextoff;	/* next offset for sequential detection */
194 	int nh_use;		/* use count for selection */
195 	int nh_seqcount;	/* heuristic */
196 } nfsheur[NUM_HEURISTIC];
197 
198 
199 /*
200  * Heuristic to detect sequential operation.
201  */
202 static struct nfsheur *
203 nfsrv_sequential_heuristic(struct uio *uio, struct vnode *vp)
204 {
205 	struct nfsheur *nh;
206 	int hi, try;
207 
208 	/* Locate best candidate. */
209 	try = 32;
210 	hi = ((int)(vm_offset_t)vp / sizeof(struct vnode)) % NUM_HEURISTIC;
211 	nh = &nfsheur[hi];
212 	while (try--) {
213 		if (nfsheur[hi].nh_vp == vp) {
214 			nh = &nfsheur[hi];
215 			break;
216 		}
217 		if (nfsheur[hi].nh_use > 0)
218 			--nfsheur[hi].nh_use;
219 		hi = (hi + 1) % NUM_HEURISTIC;
220 		if (nfsheur[hi].nh_use < nh->nh_use)
221 			nh = &nfsheur[hi];
222 	}
223 
224 	/* Initialize hint if this is a new file. */
225 	if (nh->nh_vp != vp) {
226 		nh->nh_vp = vp;
227 		nh->nh_nextoff = uio->uio_offset;
228 		nh->nh_use = NHUSE_INIT;
229 		if (uio->uio_offset == 0)
230 			nh->nh_seqcount = 4;
231 		else
232 			nh->nh_seqcount = 1;
233 	}
234 
235 	/* Calculate heuristic. */
236 	if ((uio->uio_offset == 0 && nh->nh_seqcount > 0) ||
237 	    uio->uio_offset == nh->nh_nextoff) {
238 		/* See comments in vfs_vnops.c:sequential_heuristic(). */
239 		nh->nh_seqcount += howmany(uio->uio_resid, 16384);
240 		if (nh->nh_seqcount > IO_SEQMAX)
241 			nh->nh_seqcount = IO_SEQMAX;
242 	} else if (qabs(uio->uio_offset - nh->nh_nextoff) <= MAX_REORDERED_RPC *
243 	    imax(vp->v_mount->mnt_stat.f_iosize, uio->uio_resid)) {
244 		/* Probably a reordered RPC, leave seqcount alone. */
245 	} else if (nh->nh_seqcount > 1) {
246 		nh->nh_seqcount /= 2;
247 	} else {
248 		nh->nh_seqcount = 0;
249 	}
250 	nh->nh_use += NHUSE_INC;
251 	if (nh->nh_use > NHUSE_MAX)
252 		nh->nh_use = NHUSE_MAX;
253 	return (nh);
254 }
255 
256 /*
257  * Get attributes into nfsvattr structure.
258  */
259 int
260 nfsvno_getattr(struct vnode *vp, struct nfsvattr *nvap,
261     struct nfsrv_descript *nd, struct thread *p, int vpislocked,
262     nfsattrbit_t *attrbitp)
263 {
264 	int error, gotattr, lockedit = 0;
265 	struct nfsvattr na;
266 
267 	if (vpislocked == 0) {
268 		/*
269 		 * When vpislocked == 0, the vnode is either exclusively
270 		 * locked by this thread or not locked by this thread.
271 		 * As such, shared lock it, if not exclusively locked.
272 		 */
273 		if (NFSVOPISLOCKED(vp) != LK_EXCLUSIVE) {
274 			lockedit = 1;
275 			NFSVOPLOCK(vp, LK_SHARED | LK_RETRY);
276 		}
277 	}
278 
279 	/*
280 	 * Acquire the Change, Size and TimeModify attributes, as required.
281 	 * This needs to be done for regular files if:
282 	 * - non-NFSv4 RPCs or
283 	 * - when attrbitp == NULL or
284 	 * - an NFSv4 RPC with any of the above attributes in attrbitp.
285 	 * A return of 0 for nfsrv_proxyds() indicates that it has acquired
286 	 * these attributes.  nfsrv_proxyds() will return an error if the
287 	 * server is not a pNFS one.
288 	 */
289 	gotattr = 0;
290 	if (vp->v_type == VREG && nfsrv_devidcnt > 0 && (attrbitp == NULL ||
291 	    (nd->nd_flag & ND_NFSV4) == 0 ||
292 	    NFSISSET_ATTRBIT(attrbitp, NFSATTRBIT_CHANGE) ||
293 	    NFSISSET_ATTRBIT(attrbitp, NFSATTRBIT_SIZE) ||
294 	    NFSISSET_ATTRBIT(attrbitp, NFSATTRBIT_TIMEACCESS) ||
295 	    NFSISSET_ATTRBIT(attrbitp, NFSATTRBIT_TIMEMODIFY))) {
296 		error = nfsrv_proxyds(nd, vp, 0, 0, nd->nd_cred, p,
297 		    NFSPROC_GETATTR, NULL, NULL, NULL, &na, NULL);
298 		if (error == 0)
299 			gotattr = 1;
300 	}
301 
302 	error = VOP_GETATTR(vp, &nvap->na_vattr, nd->nd_cred);
303 	if (lockedit != 0)
304 		NFSVOPUNLOCK(vp, 0);
305 
306 	/*
307 	 * If we got the Change, Size and Modify Time from the DS,
308 	 * replace them.
309 	 */
310 	if (gotattr != 0) {
311 		nvap->na_atime = na.na_atime;
312 		nvap->na_mtime = na.na_mtime;
313 		nvap->na_filerev = na.na_filerev;
314 		nvap->na_size = na.na_size;
315 	}
316 	NFSD_DEBUG(4, "nfsvno_getattr: gotattr=%d err=%d chg=%ju\n", gotattr,
317 	    error, (uintmax_t)na.na_filerev);
318 
319 	NFSEXITCODE(error);
320 	return (error);
321 }
322 
323 /*
324  * Get a file handle for a vnode.
325  */
326 int
327 nfsvno_getfh(struct vnode *vp, fhandle_t *fhp, struct thread *p)
328 {
329 	int error;
330 
331 	NFSBZERO((caddr_t)fhp, sizeof(fhandle_t));
332 	fhp->fh_fsid = vp->v_mount->mnt_stat.f_fsid;
333 	error = VOP_VPTOFH(vp, &fhp->fh_fid);
334 
335 	NFSEXITCODE(error);
336 	return (error);
337 }
338 
339 /*
340  * Perform access checking for vnodes obtained from file handles that would
341  * refer to files already opened by a Unix client. You cannot just use
342  * vn_writechk() and VOP_ACCESSX() for two reasons.
343  * 1 - You must check for exported rdonly as well as MNT_RDONLY for the write
344  *     case.
345  * 2 - The owner is to be given access irrespective of mode bits for some
346  *     operations, so that processes that chmod after opening a file don't
347  *     break.
348  */
349 int
350 nfsvno_accchk(struct vnode *vp, accmode_t accmode, struct ucred *cred,
351     struct nfsexstuff *exp, struct thread *p, int override, int vpislocked,
352     u_int32_t *supportedtypep)
353 {
354 	struct vattr vattr;
355 	int error = 0, getret = 0;
356 
357 	if (vpislocked == 0) {
358 		if (NFSVOPLOCK(vp, LK_SHARED) != 0) {
359 			error = EPERM;
360 			goto out;
361 		}
362 	}
363 	if (accmode & VWRITE) {
364 		/* Just vn_writechk() changed to check rdonly */
365 		/*
366 		 * Disallow write attempts on read-only file systems;
367 		 * unless the file is a socket or a block or character
368 		 * device resident on the file system.
369 		 */
370 		if (NFSVNO_EXRDONLY(exp) ||
371 		    (vp->v_mount->mnt_flag & MNT_RDONLY)) {
372 			switch (vp->v_type) {
373 			case VREG:
374 			case VDIR:
375 			case VLNK:
376 				error = EROFS;
377 			default:
378 				break;
379 			}
380 		}
381 		/*
382 		 * If there's shared text associated with
383 		 * the inode, try to free it up once.  If
384 		 * we fail, we can't allow writing.
385 		 */
386 		if (VOP_IS_TEXT(vp) && error == 0)
387 			error = ETXTBSY;
388 	}
389 	if (error != 0) {
390 		if (vpislocked == 0)
391 			NFSVOPUNLOCK(vp, 0);
392 		goto out;
393 	}
394 
395 	/*
396 	 * Should the override still be applied when ACLs are enabled?
397 	 */
398 	error = VOP_ACCESSX(vp, accmode, cred, p);
399 	if (error != 0 && (accmode & (VDELETE | VDELETE_CHILD))) {
400 		/*
401 		 * Try again with VEXPLICIT_DENY, to see if the test for
402 		 * deletion is supported.
403 		 */
404 		error = VOP_ACCESSX(vp, accmode | VEXPLICIT_DENY, cred, p);
405 		if (error == 0) {
406 			if (vp->v_type == VDIR) {
407 				accmode &= ~(VDELETE | VDELETE_CHILD);
408 				accmode |= VWRITE;
409 				error = VOP_ACCESSX(vp, accmode, cred, p);
410 			} else if (supportedtypep != NULL) {
411 				*supportedtypep &= ~NFSACCESS_DELETE;
412 			}
413 		}
414 	}
415 
416 	/*
417 	 * Allow certain operations for the owner (reads and writes
418 	 * on files that are already open).
419 	 */
420 	if (override != NFSACCCHK_NOOVERRIDE &&
421 	    (error == EPERM || error == EACCES)) {
422 		if (cred->cr_uid == 0 && (override & NFSACCCHK_ALLOWROOT))
423 			error = 0;
424 		else if (override & NFSACCCHK_ALLOWOWNER) {
425 			getret = VOP_GETATTR(vp, &vattr, cred);
426 			if (getret == 0 && cred->cr_uid == vattr.va_uid)
427 				error = 0;
428 		}
429 	}
430 	if (vpislocked == 0)
431 		NFSVOPUNLOCK(vp, 0);
432 
433 out:
434 	NFSEXITCODE(error);
435 	return (error);
436 }
437 
438 /*
439  * Set attribute(s) vnop.
440  */
441 int
442 nfsvno_setattr(struct vnode *vp, struct nfsvattr *nvap, struct ucred *cred,
443     struct thread *p, struct nfsexstuff *exp)
444 {
445 	u_quad_t savsize = 0;
446 	int error, savedit;
447 
448 	/*
449 	 * If this is an exported file system and a pNFS service is running,
450 	 * don't VOP_SETATTR() of size for the MDS file system.
451 	 */
452 	savedit = 0;
453 	error = 0;
454 	if (vp->v_type == VREG && (vp->v_mount->mnt_flag & MNT_EXPORTED) != 0 &&
455 	    nfsrv_devidcnt != 0 && nvap->na_vattr.va_size != VNOVAL &&
456 	    nvap->na_vattr.va_size > 0) {
457 		savsize = nvap->na_vattr.va_size;
458 		nvap->na_vattr.va_size = VNOVAL;
459 		if (nvap->na_vattr.va_uid != (uid_t)VNOVAL ||
460 		    nvap->na_vattr.va_gid != (gid_t)VNOVAL ||
461 		    nvap->na_vattr.va_mode != (mode_t)VNOVAL ||
462 		    nvap->na_vattr.va_atime.tv_sec != VNOVAL ||
463 		    nvap->na_vattr.va_mtime.tv_sec != VNOVAL)
464 			savedit = 1;
465 		else
466 			savedit = 2;
467 	}
468 	if (savedit != 2)
469 		error = VOP_SETATTR(vp, &nvap->na_vattr, cred);
470 	if (savedit != 0)
471 		nvap->na_vattr.va_size = savsize;
472 	if (error == 0 && (nvap->na_vattr.va_uid != (uid_t)VNOVAL ||
473 	    nvap->na_vattr.va_gid != (gid_t)VNOVAL ||
474 	    nvap->na_vattr.va_size != VNOVAL ||
475 	    nvap->na_vattr.va_mode != (mode_t)VNOVAL ||
476 	    nvap->na_vattr.va_atime.tv_sec != VNOVAL ||
477 	    nvap->na_vattr.va_mtime.tv_sec != VNOVAL)) {
478 		/* For a pNFS server, set the attributes on the DS file. */
479 		error = nfsrv_proxyds(NULL, vp, 0, 0, cred, p, NFSPROC_SETATTR,
480 		    NULL, NULL, NULL, nvap, NULL);
481 		if (error == ENOENT)
482 			error = 0;
483 	}
484 	NFSEXITCODE(error);
485 	return (error);
486 }
487 
488 /*
489  * Set up nameidata for a lookup() call and do it.
490  */
491 int
492 nfsvno_namei(struct nfsrv_descript *nd, struct nameidata *ndp,
493     struct vnode *dp, int islocked, struct nfsexstuff *exp, struct thread *p,
494     struct vnode **retdirp)
495 {
496 	struct componentname *cnp = &ndp->ni_cnd;
497 	int i;
498 	struct iovec aiov;
499 	struct uio auio;
500 	int lockleaf = (cnp->cn_flags & LOCKLEAF) != 0, linklen;
501 	int error = 0;
502 	char *cp;
503 
504 	*retdirp = NULL;
505 	cnp->cn_nameptr = cnp->cn_pnbuf;
506 	ndp->ni_lcf = 0;
507 	/*
508 	 * Extract and set starting directory.
509 	 */
510 	if (dp->v_type != VDIR) {
511 		if (islocked)
512 			vput(dp);
513 		else
514 			vrele(dp);
515 		nfsvno_relpathbuf(ndp);
516 		error = ENOTDIR;
517 		goto out1;
518 	}
519 	if (islocked)
520 		NFSVOPUNLOCK(dp, 0);
521 	VREF(dp);
522 	*retdirp = dp;
523 	if (NFSVNO_EXRDONLY(exp))
524 		cnp->cn_flags |= RDONLY;
525 	ndp->ni_segflg = UIO_SYSSPACE;
526 
527 	if (nd->nd_flag & ND_PUBLOOKUP) {
528 		ndp->ni_loopcnt = 0;
529 		if (cnp->cn_pnbuf[0] == '/') {
530 			vrele(dp);
531 			/*
532 			 * Check for degenerate pathnames here, since lookup()
533 			 * panics on them.
534 			 */
535 			for (i = 1; i < ndp->ni_pathlen; i++)
536 				if (cnp->cn_pnbuf[i] != '/')
537 					break;
538 			if (i == ndp->ni_pathlen) {
539 				error = NFSERR_ACCES;
540 				goto out;
541 			}
542 			dp = rootvnode;
543 			VREF(dp);
544 		}
545 	} else if ((nfsrv_enable_crossmntpt == 0 && NFSVNO_EXPORTED(exp)) ||
546 	    (nd->nd_flag & ND_NFSV4) == 0) {
547 		/*
548 		 * Only cross mount points for NFSv4 when doing a
549 		 * mount while traversing the file system above
550 		 * the mount point, unless nfsrv_enable_crossmntpt is set.
551 		 */
552 		cnp->cn_flags |= NOCROSSMOUNT;
553 	}
554 
555 	/*
556 	 * Initialize for scan, set ni_startdir and bump ref on dp again
557 	 * because lookup() will dereference ni_startdir.
558 	 */
559 
560 	cnp->cn_thread = p;
561 	ndp->ni_startdir = dp;
562 	ndp->ni_rootdir = rootvnode;
563 	ndp->ni_topdir = NULL;
564 
565 	if (!lockleaf)
566 		cnp->cn_flags |= LOCKLEAF;
567 	for (;;) {
568 		cnp->cn_nameptr = cnp->cn_pnbuf;
569 		/*
570 		 * Call lookup() to do the real work.  If an error occurs,
571 		 * ndp->ni_vp and ni_dvp are left uninitialized or NULL and
572 		 * we do not have to dereference anything before returning.
573 		 * In either case ni_startdir will be dereferenced and NULLed
574 		 * out.
575 		 */
576 		error = lookup(ndp);
577 		if (error)
578 			break;
579 
580 		/*
581 		 * Check for encountering a symbolic link.  Trivial
582 		 * termination occurs if no symlink encountered.
583 		 */
584 		if ((cnp->cn_flags & ISSYMLINK) == 0) {
585 			if ((cnp->cn_flags & (SAVENAME | SAVESTART)) == 0)
586 				nfsvno_relpathbuf(ndp);
587 			if (ndp->ni_vp && !lockleaf)
588 				NFSVOPUNLOCK(ndp->ni_vp, 0);
589 			break;
590 		}
591 
592 		/*
593 		 * Validate symlink
594 		 */
595 		if ((cnp->cn_flags & LOCKPARENT) && ndp->ni_pathlen == 1)
596 			NFSVOPUNLOCK(ndp->ni_dvp, 0);
597 		if (!(nd->nd_flag & ND_PUBLOOKUP)) {
598 			error = EINVAL;
599 			goto badlink2;
600 		}
601 
602 		if (ndp->ni_loopcnt++ >= MAXSYMLINKS) {
603 			error = ELOOP;
604 			goto badlink2;
605 		}
606 		if (ndp->ni_pathlen > 1)
607 			cp = uma_zalloc(namei_zone, M_WAITOK);
608 		else
609 			cp = cnp->cn_pnbuf;
610 		aiov.iov_base = cp;
611 		aiov.iov_len = MAXPATHLEN;
612 		auio.uio_iov = &aiov;
613 		auio.uio_iovcnt = 1;
614 		auio.uio_offset = 0;
615 		auio.uio_rw = UIO_READ;
616 		auio.uio_segflg = UIO_SYSSPACE;
617 		auio.uio_td = NULL;
618 		auio.uio_resid = MAXPATHLEN;
619 		error = VOP_READLINK(ndp->ni_vp, &auio, cnp->cn_cred);
620 		if (error) {
621 		badlink1:
622 			if (ndp->ni_pathlen > 1)
623 				uma_zfree(namei_zone, cp);
624 		badlink2:
625 			vrele(ndp->ni_dvp);
626 			vput(ndp->ni_vp);
627 			break;
628 		}
629 		linklen = MAXPATHLEN - auio.uio_resid;
630 		if (linklen == 0) {
631 			error = ENOENT;
632 			goto badlink1;
633 		}
634 		if (linklen + ndp->ni_pathlen >= MAXPATHLEN) {
635 			error = ENAMETOOLONG;
636 			goto badlink1;
637 		}
638 
639 		/*
640 		 * Adjust or replace path
641 		 */
642 		if (ndp->ni_pathlen > 1) {
643 			NFSBCOPY(ndp->ni_next, cp + linklen, ndp->ni_pathlen);
644 			uma_zfree(namei_zone, cnp->cn_pnbuf);
645 			cnp->cn_pnbuf = cp;
646 		} else
647 			cnp->cn_pnbuf[linklen] = '\0';
648 		ndp->ni_pathlen += linklen;
649 
650 		/*
651 		 * Cleanup refs for next loop and check if root directory
652 		 * should replace current directory.  Normally ni_dvp
653 		 * becomes the new base directory and is cleaned up when
654 		 * we loop.  Explicitly null pointers after invalidation
655 		 * to clarify operation.
656 		 */
657 		vput(ndp->ni_vp);
658 		ndp->ni_vp = NULL;
659 
660 		if (cnp->cn_pnbuf[0] == '/') {
661 			vrele(ndp->ni_dvp);
662 			ndp->ni_dvp = ndp->ni_rootdir;
663 			VREF(ndp->ni_dvp);
664 		}
665 		ndp->ni_startdir = ndp->ni_dvp;
666 		ndp->ni_dvp = NULL;
667 	}
668 	if (!lockleaf)
669 		cnp->cn_flags &= ~LOCKLEAF;
670 
671 out:
672 	if (error) {
673 		nfsvno_relpathbuf(ndp);
674 		ndp->ni_vp = NULL;
675 		ndp->ni_dvp = NULL;
676 		ndp->ni_startdir = NULL;
677 	} else if ((ndp->ni_cnd.cn_flags & (WANTPARENT|LOCKPARENT)) == 0) {
678 		ndp->ni_dvp = NULL;
679 	}
680 
681 out1:
682 	NFSEXITCODE2(error, nd);
683 	return (error);
684 }
685 
686 /*
687  * Set up a pathname buffer and return a pointer to it and, optionally
688  * set a hash pointer.
689  */
690 void
691 nfsvno_setpathbuf(struct nameidata *ndp, char **bufpp, u_long **hashpp)
692 {
693 	struct componentname *cnp = &ndp->ni_cnd;
694 
695 	cnp->cn_flags |= (NOMACCHECK | HASBUF);
696 	cnp->cn_pnbuf = uma_zalloc(namei_zone, M_WAITOK);
697 	if (hashpp != NULL)
698 		*hashpp = NULL;
699 	*bufpp = cnp->cn_pnbuf;
700 }
701 
702 /*
703  * Release the above path buffer, if not released by nfsvno_namei().
704  */
705 void
706 nfsvno_relpathbuf(struct nameidata *ndp)
707 {
708 
709 	if ((ndp->ni_cnd.cn_flags & HASBUF) == 0)
710 		panic("nfsrelpath");
711 	uma_zfree(namei_zone, ndp->ni_cnd.cn_pnbuf);
712 	ndp->ni_cnd.cn_flags &= ~HASBUF;
713 }
714 
715 /*
716  * Readlink vnode op into an mbuf list.
717  */
718 int
719 nfsvno_readlink(struct vnode *vp, struct ucred *cred, struct thread *p,
720     struct mbuf **mpp, struct mbuf **mpendp, int *lenp)
721 {
722 	struct iovec iv[(NFS_MAXPATHLEN+MLEN-1)/MLEN];
723 	struct iovec *ivp = iv;
724 	struct uio io, *uiop = &io;
725 	struct mbuf *mp, *mp2 = NULL, *mp3 = NULL;
726 	int i, len, tlen, error = 0;
727 
728 	len = 0;
729 	i = 0;
730 	while (len < NFS_MAXPATHLEN) {
731 		NFSMGET(mp);
732 		MCLGET(mp, M_WAITOK);
733 		mp->m_len = M_SIZE(mp);
734 		if (len == 0) {
735 			mp3 = mp2 = mp;
736 		} else {
737 			mp2->m_next = mp;
738 			mp2 = mp;
739 		}
740 		if ((len + mp->m_len) > NFS_MAXPATHLEN) {
741 			mp->m_len = NFS_MAXPATHLEN - len;
742 			len = NFS_MAXPATHLEN;
743 		} else {
744 			len += mp->m_len;
745 		}
746 		ivp->iov_base = mtod(mp, caddr_t);
747 		ivp->iov_len = mp->m_len;
748 		i++;
749 		ivp++;
750 	}
751 	uiop->uio_iov = iv;
752 	uiop->uio_iovcnt = i;
753 	uiop->uio_offset = 0;
754 	uiop->uio_resid = len;
755 	uiop->uio_rw = UIO_READ;
756 	uiop->uio_segflg = UIO_SYSSPACE;
757 	uiop->uio_td = NULL;
758 	error = VOP_READLINK(vp, uiop, cred);
759 	if (error) {
760 		m_freem(mp3);
761 		*lenp = 0;
762 		goto out;
763 	}
764 	if (uiop->uio_resid > 0) {
765 		len -= uiop->uio_resid;
766 		tlen = NFSM_RNDUP(len);
767 		nfsrv_adj(mp3, NFS_MAXPATHLEN - tlen, tlen - len);
768 	}
769 	*lenp = len;
770 	*mpp = mp3;
771 	*mpendp = mp;
772 
773 out:
774 	NFSEXITCODE(error);
775 	return (error);
776 }
777 
778 /*
779  * Read vnode op call into mbuf list.
780  */
781 int
782 nfsvno_read(struct vnode *vp, off_t off, int cnt, struct ucred *cred,
783     struct thread *p, struct mbuf **mpp, struct mbuf **mpendp)
784 {
785 	struct mbuf *m;
786 	int i;
787 	struct iovec *iv;
788 	struct iovec *iv2;
789 	int error = 0, len, left, siz, tlen, ioflag = 0;
790 	struct mbuf *m2 = NULL, *m3;
791 	struct uio io, *uiop = &io;
792 	struct nfsheur *nh;
793 
794 	/*
795 	 * Attempt to read from a DS file. A return of ENOENT implies
796 	 * there is no DS file to read.
797 	 */
798 	error = nfsrv_proxyds(NULL, vp, off, cnt, cred, p, NFSPROC_READDS, mpp,
799 	    NULL, mpendp, NULL, NULL);
800 	if (error != ENOENT)
801 		return (error);
802 
803 	len = left = NFSM_RNDUP(cnt);
804 	m3 = NULL;
805 	/*
806 	 * Generate the mbuf list with the uio_iov ref. to it.
807 	 */
808 	i = 0;
809 	while (left > 0) {
810 		NFSMGET(m);
811 		MCLGET(m, M_WAITOK);
812 		m->m_len = 0;
813 		siz = min(M_TRAILINGSPACE(m), left);
814 		left -= siz;
815 		i++;
816 		if (m3)
817 			m2->m_next = m;
818 		else
819 			m3 = m;
820 		m2 = m;
821 	}
822 	iv = malloc(i * sizeof (struct iovec),
823 	    M_TEMP, M_WAITOK);
824 	uiop->uio_iov = iv2 = iv;
825 	m = m3;
826 	left = len;
827 	i = 0;
828 	while (left > 0) {
829 		if (m == NULL)
830 			panic("nfsvno_read iov");
831 		siz = min(M_TRAILINGSPACE(m), left);
832 		if (siz > 0) {
833 			iv->iov_base = mtod(m, caddr_t) + m->m_len;
834 			iv->iov_len = siz;
835 			m->m_len += siz;
836 			left -= siz;
837 			iv++;
838 			i++;
839 		}
840 		m = m->m_next;
841 	}
842 	uiop->uio_iovcnt = i;
843 	uiop->uio_offset = off;
844 	uiop->uio_resid = len;
845 	uiop->uio_rw = UIO_READ;
846 	uiop->uio_segflg = UIO_SYSSPACE;
847 	uiop->uio_td = NULL;
848 	nh = nfsrv_sequential_heuristic(uiop, vp);
849 	ioflag |= nh->nh_seqcount << IO_SEQSHIFT;
850 	/* XXX KDM make this more systematic? */
851 	nfsstatsv1.srvbytes[NFSV4OP_READ] += uiop->uio_resid;
852 	error = VOP_READ(vp, uiop, IO_NODELOCKED | ioflag, cred);
853 	free(iv2, M_TEMP);
854 	if (error) {
855 		m_freem(m3);
856 		*mpp = NULL;
857 		goto out;
858 	}
859 	nh->nh_nextoff = uiop->uio_offset;
860 	tlen = len - uiop->uio_resid;
861 	cnt = cnt < tlen ? cnt : tlen;
862 	tlen = NFSM_RNDUP(cnt);
863 	if (tlen == 0) {
864 		m_freem(m3);
865 		m3 = NULL;
866 	} else if (len != tlen || tlen != cnt)
867 		nfsrv_adj(m3, len - tlen, tlen - cnt);
868 	*mpp = m3;
869 	*mpendp = m2;
870 
871 out:
872 	NFSEXITCODE(error);
873 	return (error);
874 }
875 
876 /*
877  * Write vnode op from an mbuf list.
878  */
879 int
880 nfsvno_write(struct vnode *vp, off_t off, int retlen, int cnt, int *stable,
881     struct mbuf *mp, char *cp, struct ucred *cred, struct thread *p)
882 {
883 	struct iovec *ivp;
884 	int i, len;
885 	struct iovec *iv;
886 	int ioflags, error;
887 	struct uio io, *uiop = &io;
888 	struct nfsheur *nh;
889 
890 	/*
891 	 * Attempt to write to a DS file. A return of ENOENT implies
892 	 * there is no DS file to write.
893 	 */
894 	error = nfsrv_proxyds(NULL, vp, off, retlen, cred, p, NFSPROC_WRITEDS,
895 	    &mp, cp, NULL, NULL, NULL);
896 	if (error != ENOENT) {
897 		*stable = NFSWRITE_FILESYNC;
898 		return (error);
899 	}
900 
901 	ivp = malloc(cnt * sizeof (struct iovec), M_TEMP,
902 	    M_WAITOK);
903 	uiop->uio_iov = iv = ivp;
904 	uiop->uio_iovcnt = cnt;
905 	i = mtod(mp, caddr_t) + mp->m_len - cp;
906 	len = retlen;
907 	while (len > 0) {
908 		if (mp == NULL)
909 			panic("nfsvno_write");
910 		if (i > 0) {
911 			i = min(i, len);
912 			ivp->iov_base = cp;
913 			ivp->iov_len = i;
914 			ivp++;
915 			len -= i;
916 		}
917 		mp = mp->m_next;
918 		if (mp) {
919 			i = mp->m_len;
920 			cp = mtod(mp, caddr_t);
921 		}
922 	}
923 
924 	if (*stable == NFSWRITE_UNSTABLE)
925 		ioflags = IO_NODELOCKED;
926 	else
927 		ioflags = (IO_SYNC | IO_NODELOCKED);
928 	uiop->uio_resid = retlen;
929 	uiop->uio_rw = UIO_WRITE;
930 	uiop->uio_segflg = UIO_SYSSPACE;
931 	NFSUIOPROC(uiop, p);
932 	uiop->uio_offset = off;
933 	nh = nfsrv_sequential_heuristic(uiop, vp);
934 	ioflags |= nh->nh_seqcount << IO_SEQSHIFT;
935 	/* XXX KDM make this more systematic? */
936 	nfsstatsv1.srvbytes[NFSV4OP_WRITE] += uiop->uio_resid;
937 	error = VOP_WRITE(vp, uiop, ioflags, cred);
938 	if (error == 0)
939 		nh->nh_nextoff = uiop->uio_offset;
940 	free(iv, M_TEMP);
941 
942 	NFSEXITCODE(error);
943 	return (error);
944 }
945 
946 /*
947  * Common code for creating a regular file (plus special files for V2).
948  */
949 int
950 nfsvno_createsub(struct nfsrv_descript *nd, struct nameidata *ndp,
951     struct vnode **vpp, struct nfsvattr *nvap, int *exclusive_flagp,
952     int32_t *cverf, NFSDEV_T rdev, struct thread *p, struct nfsexstuff *exp)
953 {
954 	u_quad_t tempsize;
955 	int error;
956 
957 	error = nd->nd_repstat;
958 	if (!error && ndp->ni_vp == NULL) {
959 		if (nvap->na_type == VREG || nvap->na_type == VSOCK) {
960 			vrele(ndp->ni_startdir);
961 			error = VOP_CREATE(ndp->ni_dvp,
962 			    &ndp->ni_vp, &ndp->ni_cnd, &nvap->na_vattr);
963 			/* For a pNFS server, create the data file on a DS. */
964 			if (error == 0 && nvap->na_type == VREG) {
965 				/*
966 				 * Create a data file on a DS for a pNFS server.
967 				 * This function just returns if not
968 				 * running a pNFS DS or the creation fails.
969 				 */
970 				nfsrv_pnfscreate(ndp->ni_vp, &nvap->na_vattr,
971 				    nd->nd_cred, p);
972 			}
973 			vput(ndp->ni_dvp);
974 			nfsvno_relpathbuf(ndp);
975 			if (!error) {
976 				if (*exclusive_flagp) {
977 					*exclusive_flagp = 0;
978 					NFSVNO_ATTRINIT(nvap);
979 					nvap->na_atime.tv_sec = cverf[0];
980 					nvap->na_atime.tv_nsec = cverf[1];
981 					error = VOP_SETATTR(ndp->ni_vp,
982 					    &nvap->na_vattr, nd->nd_cred);
983 					if (error != 0) {
984 						vput(ndp->ni_vp);
985 						ndp->ni_vp = NULL;
986 						error = NFSERR_NOTSUPP;
987 					}
988 				}
989 			}
990 		/*
991 		 * NFS V2 Only. nfsrvd_mknod() does this for V3.
992 		 * (This implies, just get out on an error.)
993 		 */
994 		} else if (nvap->na_type == VCHR || nvap->na_type == VBLK ||
995 			nvap->na_type == VFIFO) {
996 			if (nvap->na_type == VCHR && rdev == 0xffffffff)
997 				nvap->na_type = VFIFO;
998                         if (nvap->na_type != VFIFO &&
999 			    (error = priv_check_cred(nd->nd_cred,
1000 			     PRIV_VFS_MKNOD_DEV, 0))) {
1001 				vrele(ndp->ni_startdir);
1002 				nfsvno_relpathbuf(ndp);
1003 				vput(ndp->ni_dvp);
1004 				goto out;
1005 			}
1006 			nvap->na_rdev = rdev;
1007 			error = VOP_MKNOD(ndp->ni_dvp, &ndp->ni_vp,
1008 			    &ndp->ni_cnd, &nvap->na_vattr);
1009 			vput(ndp->ni_dvp);
1010 			nfsvno_relpathbuf(ndp);
1011 			vrele(ndp->ni_startdir);
1012 			if (error)
1013 				goto out;
1014 		} else {
1015 			vrele(ndp->ni_startdir);
1016 			nfsvno_relpathbuf(ndp);
1017 			vput(ndp->ni_dvp);
1018 			error = ENXIO;
1019 			goto out;
1020 		}
1021 		*vpp = ndp->ni_vp;
1022 	} else {
1023 		/*
1024 		 * Handle cases where error is already set and/or
1025 		 * the file exists.
1026 		 * 1 - clean up the lookup
1027 		 * 2 - iff !error and na_size set, truncate it
1028 		 */
1029 		vrele(ndp->ni_startdir);
1030 		nfsvno_relpathbuf(ndp);
1031 		*vpp = ndp->ni_vp;
1032 		if (ndp->ni_dvp == *vpp)
1033 			vrele(ndp->ni_dvp);
1034 		else
1035 			vput(ndp->ni_dvp);
1036 		if (!error && nvap->na_size != VNOVAL) {
1037 			error = nfsvno_accchk(*vpp, VWRITE,
1038 			    nd->nd_cred, exp, p, NFSACCCHK_NOOVERRIDE,
1039 			    NFSACCCHK_VPISLOCKED, NULL);
1040 			if (!error) {
1041 				tempsize = nvap->na_size;
1042 				NFSVNO_ATTRINIT(nvap);
1043 				nvap->na_size = tempsize;
1044 				error = VOP_SETATTR(*vpp,
1045 				    &nvap->na_vattr, nd->nd_cred);
1046 			}
1047 		}
1048 		if (error)
1049 			vput(*vpp);
1050 	}
1051 
1052 out:
1053 	NFSEXITCODE(error);
1054 	return (error);
1055 }
1056 
1057 /*
1058  * Do a mknod vnode op.
1059  */
1060 int
1061 nfsvno_mknod(struct nameidata *ndp, struct nfsvattr *nvap, struct ucred *cred,
1062     struct thread *p)
1063 {
1064 	int error = 0;
1065 	enum vtype vtyp;
1066 
1067 	vtyp = nvap->na_type;
1068 	/*
1069 	 * Iff doesn't exist, create it.
1070 	 */
1071 	if (ndp->ni_vp) {
1072 		vrele(ndp->ni_startdir);
1073 		nfsvno_relpathbuf(ndp);
1074 		vput(ndp->ni_dvp);
1075 		vrele(ndp->ni_vp);
1076 		error = EEXIST;
1077 		goto out;
1078 	}
1079 	if (vtyp != VCHR && vtyp != VBLK && vtyp != VSOCK && vtyp != VFIFO) {
1080 		vrele(ndp->ni_startdir);
1081 		nfsvno_relpathbuf(ndp);
1082 		vput(ndp->ni_dvp);
1083 		error = NFSERR_BADTYPE;
1084 		goto out;
1085 	}
1086 	if (vtyp == VSOCK) {
1087 		vrele(ndp->ni_startdir);
1088 		error = VOP_CREATE(ndp->ni_dvp, &ndp->ni_vp,
1089 		    &ndp->ni_cnd, &nvap->na_vattr);
1090 		vput(ndp->ni_dvp);
1091 		nfsvno_relpathbuf(ndp);
1092 	} else {
1093 		if (nvap->na_type != VFIFO &&
1094 		    (error = priv_check_cred(cred, PRIV_VFS_MKNOD_DEV, 0))) {
1095 			vrele(ndp->ni_startdir);
1096 			nfsvno_relpathbuf(ndp);
1097 			vput(ndp->ni_dvp);
1098 			goto out;
1099 		}
1100 		error = VOP_MKNOD(ndp->ni_dvp, &ndp->ni_vp,
1101 		    &ndp->ni_cnd, &nvap->na_vattr);
1102 		vput(ndp->ni_dvp);
1103 		nfsvno_relpathbuf(ndp);
1104 		vrele(ndp->ni_startdir);
1105 		/*
1106 		 * Since VOP_MKNOD returns the ni_vp, I can't
1107 		 * see any reason to do the lookup.
1108 		 */
1109 	}
1110 
1111 out:
1112 	NFSEXITCODE(error);
1113 	return (error);
1114 }
1115 
1116 /*
1117  * Mkdir vnode op.
1118  */
1119 int
1120 nfsvno_mkdir(struct nameidata *ndp, struct nfsvattr *nvap, uid_t saved_uid,
1121     struct ucred *cred, struct thread *p, struct nfsexstuff *exp)
1122 {
1123 	int error = 0;
1124 
1125 	if (ndp->ni_vp != NULL) {
1126 		if (ndp->ni_dvp == ndp->ni_vp)
1127 			vrele(ndp->ni_dvp);
1128 		else
1129 			vput(ndp->ni_dvp);
1130 		vrele(ndp->ni_vp);
1131 		nfsvno_relpathbuf(ndp);
1132 		error = EEXIST;
1133 		goto out;
1134 	}
1135 	error = VOP_MKDIR(ndp->ni_dvp, &ndp->ni_vp, &ndp->ni_cnd,
1136 	    &nvap->na_vattr);
1137 	vput(ndp->ni_dvp);
1138 	nfsvno_relpathbuf(ndp);
1139 
1140 out:
1141 	NFSEXITCODE(error);
1142 	return (error);
1143 }
1144 
1145 /*
1146  * symlink vnode op.
1147  */
1148 int
1149 nfsvno_symlink(struct nameidata *ndp, struct nfsvattr *nvap, char *pathcp,
1150     int pathlen, int not_v2, uid_t saved_uid, struct ucred *cred, struct thread *p,
1151     struct nfsexstuff *exp)
1152 {
1153 	int error = 0;
1154 
1155 	if (ndp->ni_vp) {
1156 		vrele(ndp->ni_startdir);
1157 		nfsvno_relpathbuf(ndp);
1158 		if (ndp->ni_dvp == ndp->ni_vp)
1159 			vrele(ndp->ni_dvp);
1160 		else
1161 			vput(ndp->ni_dvp);
1162 		vrele(ndp->ni_vp);
1163 		error = EEXIST;
1164 		goto out;
1165 	}
1166 
1167 	error = VOP_SYMLINK(ndp->ni_dvp, &ndp->ni_vp, &ndp->ni_cnd,
1168 	    &nvap->na_vattr, pathcp);
1169 	vput(ndp->ni_dvp);
1170 	vrele(ndp->ni_startdir);
1171 	nfsvno_relpathbuf(ndp);
1172 	/*
1173 	 * Although FreeBSD still had the lookup code in
1174 	 * it for 7/current, there doesn't seem to be any
1175 	 * point, since VOP_SYMLINK() returns the ni_vp.
1176 	 * Just vput it for v2.
1177 	 */
1178 	if (!not_v2 && !error)
1179 		vput(ndp->ni_vp);
1180 
1181 out:
1182 	NFSEXITCODE(error);
1183 	return (error);
1184 }
1185 
1186 /*
1187  * Parse symbolic link arguments.
1188  * This function has an ugly side effect. It will malloc() an area for
1189  * the symlink and set iov_base to point to it, only if it succeeds.
1190  * So, if it returns with uiop->uio_iov->iov_base != NULL, that must
1191  * be FREE'd later.
1192  */
1193 int
1194 nfsvno_getsymlink(struct nfsrv_descript *nd, struct nfsvattr *nvap,
1195     struct thread *p, char **pathcpp, int *lenp)
1196 {
1197 	u_int32_t *tl;
1198 	char *pathcp = NULL;
1199 	int error = 0, len;
1200 	struct nfsv2_sattr *sp;
1201 
1202 	*pathcpp = NULL;
1203 	*lenp = 0;
1204 	if ((nd->nd_flag & ND_NFSV3) &&
1205 	    (error = nfsrv_sattr(nd, NULL, nvap, NULL, NULL, p)))
1206 		goto nfsmout;
1207 	NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
1208 	len = fxdr_unsigned(int, *tl);
1209 	if (len > NFS_MAXPATHLEN || len <= 0) {
1210 		error = EBADRPC;
1211 		goto nfsmout;
1212 	}
1213 	pathcp = malloc(len + 1, M_TEMP, M_WAITOK);
1214 	error = nfsrv_mtostr(nd, pathcp, len);
1215 	if (error)
1216 		goto nfsmout;
1217 	if (nd->nd_flag & ND_NFSV2) {
1218 		NFSM_DISSECT(sp, struct nfsv2_sattr *, NFSX_V2SATTR);
1219 		nvap->na_mode = fxdr_unsigned(u_int16_t, sp->sa_mode);
1220 	}
1221 	*pathcpp = pathcp;
1222 	*lenp = len;
1223 	NFSEXITCODE2(0, nd);
1224 	return (0);
1225 nfsmout:
1226 	if (pathcp)
1227 		free(pathcp, M_TEMP);
1228 	NFSEXITCODE2(error, nd);
1229 	return (error);
1230 }
1231 
1232 /*
1233  * Remove a non-directory object.
1234  */
1235 int
1236 nfsvno_removesub(struct nameidata *ndp, int is_v4, struct ucred *cred,
1237     struct thread *p, struct nfsexstuff *exp)
1238 {
1239 	struct vnode *vp, *dsdvp[NFSDEV_MAXMIRRORS];
1240 	int error = 0, mirrorcnt;
1241 	char fname[PNFS_FILENAME_LEN + 1];
1242 	fhandle_t fh;
1243 
1244 	vp = ndp->ni_vp;
1245 	dsdvp[0] = NULL;
1246 	if (vp->v_type == VDIR)
1247 		error = NFSERR_ISDIR;
1248 	else if (is_v4)
1249 		error = nfsrv_checkremove(vp, 1, p);
1250 	if (error == 0)
1251 		nfsrv_pnfsremovesetup(vp, p, dsdvp, &mirrorcnt, fname, &fh);
1252 	if (!error)
1253 		error = VOP_REMOVE(ndp->ni_dvp, vp, &ndp->ni_cnd);
1254 	if (error == 0 && dsdvp[0] != NULL)
1255 		nfsrv_pnfsremove(dsdvp, mirrorcnt, fname, &fh, p);
1256 	if (ndp->ni_dvp == vp)
1257 		vrele(ndp->ni_dvp);
1258 	else
1259 		vput(ndp->ni_dvp);
1260 	vput(vp);
1261 	if ((ndp->ni_cnd.cn_flags & SAVENAME) != 0)
1262 		nfsvno_relpathbuf(ndp);
1263 	NFSEXITCODE(error);
1264 	return (error);
1265 }
1266 
1267 /*
1268  * Remove a directory.
1269  */
1270 int
1271 nfsvno_rmdirsub(struct nameidata *ndp, int is_v4, struct ucred *cred,
1272     struct thread *p, struct nfsexstuff *exp)
1273 {
1274 	struct vnode *vp;
1275 	int error = 0;
1276 
1277 	vp = ndp->ni_vp;
1278 	if (vp->v_type != VDIR) {
1279 		error = ENOTDIR;
1280 		goto out;
1281 	}
1282 	/*
1283 	 * No rmdir "." please.
1284 	 */
1285 	if (ndp->ni_dvp == vp) {
1286 		error = EINVAL;
1287 		goto out;
1288 	}
1289 	/*
1290 	 * The root of a mounted filesystem cannot be deleted.
1291 	 */
1292 	if (vp->v_vflag & VV_ROOT)
1293 		error = EBUSY;
1294 out:
1295 	if (!error)
1296 		error = VOP_RMDIR(ndp->ni_dvp, vp, &ndp->ni_cnd);
1297 	if (ndp->ni_dvp == vp)
1298 		vrele(ndp->ni_dvp);
1299 	else
1300 		vput(ndp->ni_dvp);
1301 	vput(vp);
1302 	if ((ndp->ni_cnd.cn_flags & SAVENAME) != 0)
1303 		nfsvno_relpathbuf(ndp);
1304 	NFSEXITCODE(error);
1305 	return (error);
1306 }
1307 
1308 /*
1309  * Rename vnode op.
1310  */
1311 int
1312 nfsvno_rename(struct nameidata *fromndp, struct nameidata *tondp,
1313     u_int32_t ndstat, u_int32_t ndflag, struct ucred *cred, struct thread *p)
1314 {
1315 	struct vnode *fvp, *tvp, *tdvp, *dsdvp[NFSDEV_MAXMIRRORS];
1316 	int error = 0, mirrorcnt;
1317 	char fname[PNFS_FILENAME_LEN + 1];
1318 	fhandle_t fh;
1319 
1320 	dsdvp[0] = NULL;
1321 	fvp = fromndp->ni_vp;
1322 	if (ndstat) {
1323 		vrele(fromndp->ni_dvp);
1324 		vrele(fvp);
1325 		error = ndstat;
1326 		goto out1;
1327 	}
1328 	tdvp = tondp->ni_dvp;
1329 	tvp = tondp->ni_vp;
1330 	if (tvp != NULL) {
1331 		if (fvp->v_type == VDIR && tvp->v_type != VDIR) {
1332 			error = (ndflag & ND_NFSV2) ? EISDIR : EEXIST;
1333 			goto out;
1334 		} else if (fvp->v_type != VDIR && tvp->v_type == VDIR) {
1335 			error = (ndflag & ND_NFSV2) ? ENOTDIR : EEXIST;
1336 			goto out;
1337 		}
1338 		if (tvp->v_type == VDIR && tvp->v_mountedhere) {
1339 			error = (ndflag & ND_NFSV2) ? ENOTEMPTY : EXDEV;
1340 			goto out;
1341 		}
1342 
1343 		/*
1344 		 * A rename to '.' or '..' results in a prematurely
1345 		 * unlocked vnode on FreeBSD5, so I'm just going to fail that
1346 		 * here.
1347 		 */
1348 		if ((tondp->ni_cnd.cn_namelen == 1 &&
1349 		     tondp->ni_cnd.cn_nameptr[0] == '.') ||
1350 		    (tondp->ni_cnd.cn_namelen == 2 &&
1351 		     tondp->ni_cnd.cn_nameptr[0] == '.' &&
1352 		     tondp->ni_cnd.cn_nameptr[1] == '.')) {
1353 			error = EINVAL;
1354 			goto out;
1355 		}
1356 	}
1357 	if (fvp->v_type == VDIR && fvp->v_mountedhere) {
1358 		error = (ndflag & ND_NFSV2) ? ENOTEMPTY : EXDEV;
1359 		goto out;
1360 	}
1361 	if (fvp->v_mount != tdvp->v_mount) {
1362 		error = (ndflag & ND_NFSV2) ? ENOTEMPTY : EXDEV;
1363 		goto out;
1364 	}
1365 	if (fvp == tdvp) {
1366 		error = (ndflag & ND_NFSV2) ? ENOTEMPTY : EINVAL;
1367 		goto out;
1368 	}
1369 	if (fvp == tvp) {
1370 		/*
1371 		 * If source and destination are the same, there is nothing to
1372 		 * do. Set error to -1 to indicate this.
1373 		 */
1374 		error = -1;
1375 		goto out;
1376 	}
1377 	if (ndflag & ND_NFSV4) {
1378 		if (NFSVOPLOCK(fvp, LK_EXCLUSIVE) == 0) {
1379 			error = nfsrv_checkremove(fvp, 0, p);
1380 			NFSVOPUNLOCK(fvp, 0);
1381 		} else
1382 			error = EPERM;
1383 		if (tvp && !error)
1384 			error = nfsrv_checkremove(tvp, 1, p);
1385 	} else {
1386 		/*
1387 		 * For NFSv2 and NFSv3, try to get rid of the delegation, so
1388 		 * that the NFSv4 client won't be confused by the rename.
1389 		 * Since nfsd_recalldelegation() can only be called on an
1390 		 * unlocked vnode at this point and fvp is the file that will
1391 		 * still exist after the rename, just do fvp.
1392 		 */
1393 		nfsd_recalldelegation(fvp, p);
1394 	}
1395 	if (error == 0 && tvp != NULL) {
1396 		nfsrv_pnfsremovesetup(tvp, p, dsdvp, &mirrorcnt, fname, &fh);
1397 		NFSD_DEBUG(4, "nfsvno_rename: pnfsremovesetup"
1398 		    " dsdvp=%p\n", dsdvp[0]);
1399 	}
1400 out:
1401 	if (!error) {
1402 		error = VOP_RENAME(fromndp->ni_dvp, fromndp->ni_vp,
1403 		    &fromndp->ni_cnd, tondp->ni_dvp, tondp->ni_vp,
1404 		    &tondp->ni_cnd);
1405 	} else {
1406 		if (tdvp == tvp)
1407 			vrele(tdvp);
1408 		else
1409 			vput(tdvp);
1410 		if (tvp)
1411 			vput(tvp);
1412 		vrele(fromndp->ni_dvp);
1413 		vrele(fvp);
1414 		if (error == -1)
1415 			error = 0;
1416 	}
1417 
1418 	/*
1419 	 * If dsdvp[0] != NULL, it was set up by nfsrv_pnfsremovesetup() and
1420 	 * if the rename succeeded, the DS file for the tvp needs to be
1421 	 * removed.
1422 	 */
1423 	if (error == 0 && dsdvp[0] != NULL) {
1424 		nfsrv_pnfsremove(dsdvp, mirrorcnt, fname, &fh, p);
1425 		NFSD_DEBUG(4, "nfsvno_rename: pnfsremove\n");
1426 	}
1427 
1428 	vrele(tondp->ni_startdir);
1429 	nfsvno_relpathbuf(tondp);
1430 out1:
1431 	vrele(fromndp->ni_startdir);
1432 	nfsvno_relpathbuf(fromndp);
1433 	NFSEXITCODE(error);
1434 	return (error);
1435 }
1436 
1437 /*
1438  * Link vnode op.
1439  */
1440 int
1441 nfsvno_link(struct nameidata *ndp, struct vnode *vp, struct ucred *cred,
1442     struct thread *p, struct nfsexstuff *exp)
1443 {
1444 	struct vnode *xp;
1445 	int error = 0;
1446 
1447 	xp = ndp->ni_vp;
1448 	if (xp != NULL) {
1449 		error = EEXIST;
1450 	} else {
1451 		xp = ndp->ni_dvp;
1452 		if (vp->v_mount != xp->v_mount)
1453 			error = EXDEV;
1454 	}
1455 	if (!error) {
1456 		NFSVOPLOCK(vp, LK_EXCLUSIVE | LK_RETRY);
1457 		if ((vp->v_iflag & VI_DOOMED) == 0)
1458 			error = VOP_LINK(ndp->ni_dvp, vp, &ndp->ni_cnd);
1459 		else
1460 			error = EPERM;
1461 		if (ndp->ni_dvp == vp)
1462 			vrele(ndp->ni_dvp);
1463 		else
1464 			vput(ndp->ni_dvp);
1465 		NFSVOPUNLOCK(vp, 0);
1466 	} else {
1467 		if (ndp->ni_dvp == ndp->ni_vp)
1468 			vrele(ndp->ni_dvp);
1469 		else
1470 			vput(ndp->ni_dvp);
1471 		if (ndp->ni_vp)
1472 			vrele(ndp->ni_vp);
1473 	}
1474 	nfsvno_relpathbuf(ndp);
1475 	NFSEXITCODE(error);
1476 	return (error);
1477 }
1478 
1479 /*
1480  * Do the fsync() appropriate for the commit.
1481  */
1482 int
1483 nfsvno_fsync(struct vnode *vp, u_int64_t off, int cnt, struct ucred *cred,
1484     struct thread *td)
1485 {
1486 	int error = 0;
1487 
1488 	/*
1489 	 * RFC 1813 3.3.21: if count is 0, a flush from offset to the end of
1490 	 * file is done.  At this time VOP_FSYNC does not accept offset and
1491 	 * byte count parameters so call VOP_FSYNC the whole file for now.
1492 	 * The same is true for NFSv4: RFC 3530 Sec. 14.2.3.
1493 	 * File systems that do not use the buffer cache (as indicated
1494 	 * by MNTK_USES_BCACHE not being set) must use VOP_FSYNC().
1495 	 */
1496 	if (cnt == 0 || cnt > MAX_COMMIT_COUNT ||
1497 	    (vp->v_mount->mnt_kern_flag & MNTK_USES_BCACHE) == 0) {
1498 		/*
1499 		 * Give up and do the whole thing
1500 		 */
1501 		if (vp->v_object &&
1502 		   (vp->v_object->flags & OBJ_MIGHTBEDIRTY)) {
1503 			VM_OBJECT_WLOCK(vp->v_object);
1504 			vm_object_page_clean(vp->v_object, 0, 0, OBJPC_SYNC);
1505 			VM_OBJECT_WUNLOCK(vp->v_object);
1506 		}
1507 		error = VOP_FSYNC(vp, MNT_WAIT, td);
1508 	} else {
1509 		/*
1510 		 * Locate and synchronously write any buffers that fall
1511 		 * into the requested range.  Note:  we are assuming that
1512 		 * f_iosize is a power of 2.
1513 		 */
1514 		int iosize = vp->v_mount->mnt_stat.f_iosize;
1515 		int iomask = iosize - 1;
1516 		struct bufobj *bo;
1517 		daddr_t lblkno;
1518 
1519 		/*
1520 		 * Align to iosize boundary, super-align to page boundary.
1521 		 */
1522 		if (off & iomask) {
1523 			cnt += off & iomask;
1524 			off &= ~(u_quad_t)iomask;
1525 		}
1526 		if (off & PAGE_MASK) {
1527 			cnt += off & PAGE_MASK;
1528 			off &= ~(u_quad_t)PAGE_MASK;
1529 		}
1530 		lblkno = off / iosize;
1531 
1532 		if (vp->v_object &&
1533 		   (vp->v_object->flags & OBJ_MIGHTBEDIRTY)) {
1534 			VM_OBJECT_WLOCK(vp->v_object);
1535 			vm_object_page_clean(vp->v_object, off, off + cnt,
1536 			    OBJPC_SYNC);
1537 			VM_OBJECT_WUNLOCK(vp->v_object);
1538 		}
1539 
1540 		bo = &vp->v_bufobj;
1541 		BO_LOCK(bo);
1542 		while (cnt > 0) {
1543 			struct buf *bp;
1544 
1545 			/*
1546 			 * If we have a buffer and it is marked B_DELWRI we
1547 			 * have to lock and write it.  Otherwise the prior
1548 			 * write is assumed to have already been committed.
1549 			 *
1550 			 * gbincore() can return invalid buffers now so we
1551 			 * have to check that bit as well (though B_DELWRI
1552 			 * should not be set if B_INVAL is set there could be
1553 			 * a race here since we haven't locked the buffer).
1554 			 */
1555 			if ((bp = gbincore(&vp->v_bufobj, lblkno)) != NULL) {
1556 				if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_SLEEPFAIL |
1557 				    LK_INTERLOCK, BO_LOCKPTR(bo)) == ENOLCK) {
1558 					BO_LOCK(bo);
1559 					continue; /* retry */
1560 				}
1561 			    	if ((bp->b_flags & (B_DELWRI|B_INVAL)) ==
1562 				    B_DELWRI) {
1563 					bremfree(bp);
1564 					bp->b_flags &= ~B_ASYNC;
1565 					bwrite(bp);
1566 					++nfs_commit_miss;
1567 				} else
1568 					BUF_UNLOCK(bp);
1569 				BO_LOCK(bo);
1570 			}
1571 			++nfs_commit_blks;
1572 			if (cnt < iosize)
1573 				break;
1574 			cnt -= iosize;
1575 			++lblkno;
1576 		}
1577 		BO_UNLOCK(bo);
1578 	}
1579 	NFSEXITCODE(error);
1580 	return (error);
1581 }
1582 
1583 /*
1584  * Statfs vnode op.
1585  */
1586 int
1587 nfsvno_statfs(struct vnode *vp, struct statfs *sf)
1588 {
1589 	struct statfs *tsf;
1590 	int error;
1591 
1592 	tsf = NULL;
1593 	if (nfsrv_devidcnt > 0) {
1594 		/* For a pNFS service, get the DS numbers. */
1595 		tsf = malloc(sizeof(*tsf), M_TEMP, M_WAITOK | M_ZERO);
1596 		error = nfsrv_pnfsstatfs(tsf, vp->v_mount);
1597 		if (error != 0) {
1598 			free(tsf, M_TEMP);
1599 			tsf = NULL;
1600 		}
1601 	}
1602 	error = VFS_STATFS(vp->v_mount, sf);
1603 	if (error == 0) {
1604 		if (tsf != NULL) {
1605 			sf->f_blocks = tsf->f_blocks;
1606 			sf->f_bavail = tsf->f_bavail;
1607 			sf->f_bfree = tsf->f_bfree;
1608 			sf->f_bsize = tsf->f_bsize;
1609 		}
1610 		/*
1611 		 * Since NFS handles these values as unsigned on the
1612 		 * wire, there is no way to represent negative values,
1613 		 * so set them to 0. Without this, they will appear
1614 		 * to be very large positive values for clients like
1615 		 * Solaris10.
1616 		 */
1617 		if (sf->f_bavail < 0)
1618 			sf->f_bavail = 0;
1619 		if (sf->f_ffree < 0)
1620 			sf->f_ffree = 0;
1621 	}
1622 	free(tsf, M_TEMP);
1623 	NFSEXITCODE(error);
1624 	return (error);
1625 }
1626 
1627 /*
1628  * Do the vnode op stuff for Open. Similar to nfsvno_createsub(), but
1629  * must handle nfsrv_opencheck() calls after any other access checks.
1630  */
1631 void
1632 nfsvno_open(struct nfsrv_descript *nd, struct nameidata *ndp,
1633     nfsquad_t clientid, nfsv4stateid_t *stateidp, struct nfsstate *stp,
1634     int *exclusive_flagp, struct nfsvattr *nvap, int32_t *cverf, int create,
1635     NFSACL_T *aclp, nfsattrbit_t *attrbitp, struct ucred *cred, struct thread *p,
1636     struct nfsexstuff *exp, struct vnode **vpp)
1637 {
1638 	struct vnode *vp = NULL;
1639 	u_quad_t tempsize;
1640 	struct nfsexstuff nes;
1641 
1642 	if (ndp->ni_vp == NULL)
1643 		nd->nd_repstat = nfsrv_opencheck(clientid,
1644 		    stateidp, stp, NULL, nd, p, nd->nd_repstat);
1645 	if (!nd->nd_repstat) {
1646 		if (ndp->ni_vp == NULL) {
1647 			vrele(ndp->ni_startdir);
1648 			nd->nd_repstat = VOP_CREATE(ndp->ni_dvp,
1649 			    &ndp->ni_vp, &ndp->ni_cnd, &nvap->na_vattr);
1650 			/* For a pNFS server, create the data file on a DS. */
1651 			if (nd->nd_repstat == 0) {
1652 				/*
1653 				 * Create a data file on a DS for a pNFS server.
1654 				 * This function just returns if not
1655 				 * running a pNFS DS or the creation fails.
1656 				 */
1657 				nfsrv_pnfscreate(ndp->ni_vp, &nvap->na_vattr,
1658 				    cred, p);
1659 			}
1660 			vput(ndp->ni_dvp);
1661 			nfsvno_relpathbuf(ndp);
1662 			if (!nd->nd_repstat) {
1663 				if (*exclusive_flagp) {
1664 					*exclusive_flagp = 0;
1665 					NFSVNO_ATTRINIT(nvap);
1666 					nvap->na_atime.tv_sec = cverf[0];
1667 					nvap->na_atime.tv_nsec = cverf[1];
1668 					nd->nd_repstat = VOP_SETATTR(ndp->ni_vp,
1669 					    &nvap->na_vattr, cred);
1670 					if (nd->nd_repstat != 0) {
1671 						vput(ndp->ni_vp);
1672 						ndp->ni_vp = NULL;
1673 						nd->nd_repstat = NFSERR_NOTSUPP;
1674 					} else
1675 						NFSSETBIT_ATTRBIT(attrbitp,
1676 						    NFSATTRBIT_TIMEACCESS);
1677 				} else {
1678 					nfsrv_fixattr(nd, ndp->ni_vp, nvap,
1679 					    aclp, p, attrbitp, exp);
1680 				}
1681 			}
1682 			vp = ndp->ni_vp;
1683 		} else {
1684 			if (ndp->ni_startdir)
1685 				vrele(ndp->ni_startdir);
1686 			nfsvno_relpathbuf(ndp);
1687 			vp = ndp->ni_vp;
1688 			if (create == NFSV4OPEN_CREATE) {
1689 				if (ndp->ni_dvp == vp)
1690 					vrele(ndp->ni_dvp);
1691 				else
1692 					vput(ndp->ni_dvp);
1693 			}
1694 			if (NFSVNO_ISSETSIZE(nvap) && vp->v_type == VREG) {
1695 				if (ndp->ni_cnd.cn_flags & RDONLY)
1696 					NFSVNO_SETEXRDONLY(&nes);
1697 				else
1698 					NFSVNO_EXINIT(&nes);
1699 				nd->nd_repstat = nfsvno_accchk(vp,
1700 				    VWRITE, cred, &nes, p,
1701 				    NFSACCCHK_NOOVERRIDE,
1702 				    NFSACCCHK_VPISLOCKED, NULL);
1703 				nd->nd_repstat = nfsrv_opencheck(clientid,
1704 				    stateidp, stp, vp, nd, p, nd->nd_repstat);
1705 				if (!nd->nd_repstat) {
1706 					tempsize = nvap->na_size;
1707 					NFSVNO_ATTRINIT(nvap);
1708 					nvap->na_size = tempsize;
1709 					nd->nd_repstat = VOP_SETATTR(vp,
1710 					    &nvap->na_vattr, cred);
1711 				}
1712 			} else if (vp->v_type == VREG) {
1713 				nd->nd_repstat = nfsrv_opencheck(clientid,
1714 				    stateidp, stp, vp, nd, p, nd->nd_repstat);
1715 			}
1716 		}
1717 	} else {
1718 		if (ndp->ni_cnd.cn_flags & HASBUF)
1719 			nfsvno_relpathbuf(ndp);
1720 		if (ndp->ni_startdir && create == NFSV4OPEN_CREATE) {
1721 			vrele(ndp->ni_startdir);
1722 			if (ndp->ni_dvp == ndp->ni_vp)
1723 				vrele(ndp->ni_dvp);
1724 			else
1725 				vput(ndp->ni_dvp);
1726 			if (ndp->ni_vp)
1727 				vput(ndp->ni_vp);
1728 		}
1729 	}
1730 	*vpp = vp;
1731 
1732 	NFSEXITCODE2(0, nd);
1733 }
1734 
1735 /*
1736  * Updates the file rev and sets the mtime and ctime
1737  * to the current clock time, returning the va_filerev and va_Xtime
1738  * values.
1739  * Return ESTALE to indicate the vnode is VI_DOOMED.
1740  */
1741 int
1742 nfsvno_updfilerev(struct vnode *vp, struct nfsvattr *nvap,
1743     struct nfsrv_descript *nd, struct thread *p)
1744 {
1745 	struct vattr va;
1746 
1747 	VATTR_NULL(&va);
1748 	vfs_timestamp(&va.va_mtime);
1749 	if (NFSVOPISLOCKED(vp) != LK_EXCLUSIVE) {
1750 		NFSVOPLOCK(vp, LK_UPGRADE | LK_RETRY);
1751 		if ((vp->v_iflag & VI_DOOMED) != 0)
1752 			return (ESTALE);
1753 	}
1754 	(void) VOP_SETATTR(vp, &va, nd->nd_cred);
1755 	(void) nfsvno_getattr(vp, nvap, nd, p, 1, NULL);
1756 	return (0);
1757 }
1758 
1759 /*
1760  * Glue routine to nfsv4_fillattr().
1761  */
1762 int
1763 nfsvno_fillattr(struct nfsrv_descript *nd, struct mount *mp, struct vnode *vp,
1764     struct nfsvattr *nvap, fhandle_t *fhp, int rderror, nfsattrbit_t *attrbitp,
1765     struct ucred *cred, struct thread *p, int isdgram, int reterr,
1766     int supports_nfsv4acls, int at_root, uint64_t mounted_on_fileno)
1767 {
1768 	struct statfs *sf;
1769 	int error;
1770 
1771 	sf = NULL;
1772 	if (nfsrv_devidcnt > 0 &&
1773 	    (NFSISSET_ATTRBIT(attrbitp, NFSATTRBIT_SPACEAVAIL) ||
1774 	     NFSISSET_ATTRBIT(attrbitp, NFSATTRBIT_SPACEFREE) ||
1775 	     NFSISSET_ATTRBIT(attrbitp, NFSATTRBIT_SPACETOTAL))) {
1776 		sf = malloc(sizeof(*sf), M_TEMP, M_WAITOK | M_ZERO);
1777 		error = nfsrv_pnfsstatfs(sf, mp);
1778 		if (error != 0) {
1779 			free(sf, M_TEMP);
1780 			sf = NULL;
1781 		}
1782 	}
1783 	error = nfsv4_fillattr(nd, mp, vp, NULL, &nvap->na_vattr, fhp, rderror,
1784 	    attrbitp, cred, p, isdgram, reterr, supports_nfsv4acls, at_root,
1785 	    mounted_on_fileno, sf);
1786 	free(sf, M_TEMP);
1787 	NFSEXITCODE2(0, nd);
1788 	return (error);
1789 }
1790 
1791 /* Since the Readdir vnode ops vary, put the entire functions in here. */
1792 /*
1793  * nfs readdir service
1794  * - mallocs what it thinks is enough to read
1795  *	count rounded up to a multiple of DIRBLKSIZ <= NFS_MAXREADDIR
1796  * - calls VOP_READDIR()
1797  * - loops around building the reply
1798  *	if the output generated exceeds count break out of loop
1799  *	The NFSM_CLGET macro is used here so that the reply will be packed
1800  *	tightly in mbuf clusters.
1801  * - it trims out records with d_fileno == 0
1802  *	this doesn't matter for Unix clients, but they might confuse clients
1803  *	for other os'.
1804  * - it trims out records with d_type == DT_WHT
1805  *	these cannot be seen through NFS (unless we extend the protocol)
1806  *     The alternate call nfsrvd_readdirplus() does lookups as well.
1807  * PS: The NFS protocol spec. does not clarify what the "count" byte
1808  *	argument is a count of.. just name strings and file id's or the
1809  *	entire reply rpc or ...
1810  *	I tried just file name and id sizes and it confused the Sun client,
1811  *	so I am using the full rpc size now. The "paranoia.." comment refers
1812  *	to including the status longwords that are not a part of the dir.
1813  *	"entry" structures, but are in the rpc.
1814  */
1815 int
1816 nfsrvd_readdir(struct nfsrv_descript *nd, int isdgram,
1817     struct vnode *vp, struct thread *p, struct nfsexstuff *exp)
1818 {
1819 	struct dirent *dp;
1820 	u_int32_t *tl;
1821 	int dirlen;
1822 	char *cpos, *cend, *rbuf;
1823 	struct nfsvattr at;
1824 	int nlen, error = 0, getret = 1;
1825 	int siz, cnt, fullsiz, eofflag, ncookies;
1826 	u_int64_t off, toff, verf __unused;
1827 	u_long *cookies = NULL, *cookiep;
1828 	struct uio io;
1829 	struct iovec iv;
1830 	int is_ufs;
1831 
1832 	if (nd->nd_repstat) {
1833 		nfsrv_postopattr(nd, getret, &at);
1834 		goto out;
1835 	}
1836 	if (nd->nd_flag & ND_NFSV2) {
1837 		NFSM_DISSECT(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
1838 		off = fxdr_unsigned(u_quad_t, *tl++);
1839 	} else {
1840 		NFSM_DISSECT(tl, u_int32_t *, 5 * NFSX_UNSIGNED);
1841 		off = fxdr_hyper(tl);
1842 		tl += 2;
1843 		verf = fxdr_hyper(tl);
1844 		tl += 2;
1845 	}
1846 	toff = off;
1847 	cnt = fxdr_unsigned(int, *tl);
1848 	if (cnt > NFS_SRVMAXDATA(nd) || cnt < 0)
1849 		cnt = NFS_SRVMAXDATA(nd);
1850 	siz = ((cnt + DIRBLKSIZ - 1) & ~(DIRBLKSIZ - 1));
1851 	fullsiz = siz;
1852 	if (nd->nd_flag & ND_NFSV3) {
1853 		nd->nd_repstat = getret = nfsvno_getattr(vp, &at, nd, p, 1,
1854 		    NULL);
1855 #if 0
1856 		/*
1857 		 * va_filerev is not sufficient as a cookie verifier,
1858 		 * since it is not supposed to change when entries are
1859 		 * removed/added unless that offset cookies returned to
1860 		 * the client are no longer valid.
1861 		 */
1862 		if (!nd->nd_repstat && toff && verf != at.na_filerev)
1863 			nd->nd_repstat = NFSERR_BAD_COOKIE;
1864 #endif
1865 	}
1866 	if (!nd->nd_repstat && vp->v_type != VDIR)
1867 		nd->nd_repstat = NFSERR_NOTDIR;
1868 	if (nd->nd_repstat == 0 && cnt == 0) {
1869 		if (nd->nd_flag & ND_NFSV2)
1870 			/* NFSv2 does not have NFSERR_TOOSMALL */
1871 			nd->nd_repstat = EPERM;
1872 		else
1873 			nd->nd_repstat = NFSERR_TOOSMALL;
1874 	}
1875 	if (!nd->nd_repstat)
1876 		nd->nd_repstat = nfsvno_accchk(vp, VEXEC,
1877 		    nd->nd_cred, exp, p, NFSACCCHK_NOOVERRIDE,
1878 		    NFSACCCHK_VPISLOCKED, NULL);
1879 	if (nd->nd_repstat) {
1880 		vput(vp);
1881 		if (nd->nd_flag & ND_NFSV3)
1882 			nfsrv_postopattr(nd, getret, &at);
1883 		goto out;
1884 	}
1885 	is_ufs = strcmp(vp->v_mount->mnt_vfc->vfc_name, "ufs") == 0;
1886 	rbuf = malloc(siz, M_TEMP, M_WAITOK);
1887 again:
1888 	eofflag = 0;
1889 	if (cookies) {
1890 		free(cookies, M_TEMP);
1891 		cookies = NULL;
1892 	}
1893 
1894 	iv.iov_base = rbuf;
1895 	iv.iov_len = siz;
1896 	io.uio_iov = &iv;
1897 	io.uio_iovcnt = 1;
1898 	io.uio_offset = (off_t)off;
1899 	io.uio_resid = siz;
1900 	io.uio_segflg = UIO_SYSSPACE;
1901 	io.uio_rw = UIO_READ;
1902 	io.uio_td = NULL;
1903 	nd->nd_repstat = VOP_READDIR(vp, &io, nd->nd_cred, &eofflag, &ncookies,
1904 	    &cookies);
1905 	off = (u_int64_t)io.uio_offset;
1906 	if (io.uio_resid)
1907 		siz -= io.uio_resid;
1908 
1909 	if (!cookies && !nd->nd_repstat)
1910 		nd->nd_repstat = NFSERR_PERM;
1911 	if (nd->nd_flag & ND_NFSV3) {
1912 		getret = nfsvno_getattr(vp, &at, nd, p, 1, NULL);
1913 		if (!nd->nd_repstat)
1914 			nd->nd_repstat = getret;
1915 	}
1916 
1917 	/*
1918 	 * Handles the failed cases. nd->nd_repstat == 0 past here.
1919 	 */
1920 	if (nd->nd_repstat) {
1921 		vput(vp);
1922 		free(rbuf, M_TEMP);
1923 		if (cookies)
1924 			free(cookies, M_TEMP);
1925 		if (nd->nd_flag & ND_NFSV3)
1926 			nfsrv_postopattr(nd, getret, &at);
1927 		goto out;
1928 	}
1929 	/*
1930 	 * If nothing read, return eof
1931 	 * rpc reply
1932 	 */
1933 	if (siz == 0) {
1934 		vput(vp);
1935 		if (nd->nd_flag & ND_NFSV2) {
1936 			NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
1937 		} else {
1938 			nfsrv_postopattr(nd, getret, &at);
1939 			NFSM_BUILD(tl, u_int32_t *, 4 * NFSX_UNSIGNED);
1940 			txdr_hyper(at.na_filerev, tl);
1941 			tl += 2;
1942 		}
1943 		*tl++ = newnfs_false;
1944 		*tl = newnfs_true;
1945 		free(rbuf, M_TEMP);
1946 		free(cookies, M_TEMP);
1947 		goto out;
1948 	}
1949 
1950 	/*
1951 	 * Check for degenerate cases of nothing useful read.
1952 	 * If so go try again
1953 	 */
1954 	cpos = rbuf;
1955 	cend = rbuf + siz;
1956 	dp = (struct dirent *)cpos;
1957 	cookiep = cookies;
1958 
1959 	/*
1960 	 * For some reason FreeBSD's ufs_readdir() chooses to back the
1961 	 * directory offset up to a block boundary, so it is necessary to
1962 	 * skip over the records that precede the requested offset. This
1963 	 * requires the assumption that file offset cookies monotonically
1964 	 * increase.
1965 	 */
1966 	while (cpos < cend && ncookies > 0 &&
1967 	    (dp->d_fileno == 0 || dp->d_type == DT_WHT ||
1968 	     (is_ufs == 1 && ((u_quad_t)(*cookiep)) <= toff))) {
1969 		cpos += dp->d_reclen;
1970 		dp = (struct dirent *)cpos;
1971 		cookiep++;
1972 		ncookies--;
1973 	}
1974 	if (cpos >= cend || ncookies == 0) {
1975 		siz = fullsiz;
1976 		toff = off;
1977 		goto again;
1978 	}
1979 	vput(vp);
1980 
1981 	/*
1982 	 * dirlen is the size of the reply, including all XDR and must
1983 	 * not exceed cnt. For NFSv2, RFC1094 didn't clearly indicate
1984 	 * if the XDR should be included in "count", but to be safe, we do.
1985 	 * (Include the two booleans at the end of the reply in dirlen now.)
1986 	 */
1987 	if (nd->nd_flag & ND_NFSV3) {
1988 		nfsrv_postopattr(nd, getret, &at);
1989 		NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
1990 		txdr_hyper(at.na_filerev, tl);
1991 		dirlen = NFSX_V3POSTOPATTR + NFSX_VERF + 2 * NFSX_UNSIGNED;
1992 	} else {
1993 		dirlen = 2 * NFSX_UNSIGNED;
1994 	}
1995 
1996 	/* Loop through the records and build reply */
1997 	while (cpos < cend && ncookies > 0) {
1998 		nlen = dp->d_namlen;
1999 		if (dp->d_fileno != 0 && dp->d_type != DT_WHT &&
2000 			nlen <= NFS_MAXNAMLEN) {
2001 			if (nd->nd_flag & ND_NFSV3)
2002 				dirlen += (6*NFSX_UNSIGNED + NFSM_RNDUP(nlen));
2003 			else
2004 				dirlen += (4*NFSX_UNSIGNED + NFSM_RNDUP(nlen));
2005 			if (dirlen > cnt) {
2006 				eofflag = 0;
2007 				break;
2008 			}
2009 
2010 			/*
2011 			 * Build the directory record xdr from
2012 			 * the dirent entry.
2013 			 */
2014 			if (nd->nd_flag & ND_NFSV3) {
2015 				NFSM_BUILD(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
2016 				*tl++ = newnfs_true;
2017 				*tl++ = 0;
2018 			} else {
2019 				NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
2020 				*tl++ = newnfs_true;
2021 			}
2022 			*tl = txdr_unsigned(dp->d_fileno);
2023 			(void) nfsm_strtom(nd, dp->d_name, nlen);
2024 			if (nd->nd_flag & ND_NFSV3) {
2025 				NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
2026 				*tl++ = 0;
2027 			} else
2028 				NFSM_BUILD(tl, u_int32_t *, NFSX_UNSIGNED);
2029 			*tl = txdr_unsigned(*cookiep);
2030 		}
2031 		cpos += dp->d_reclen;
2032 		dp = (struct dirent *)cpos;
2033 		cookiep++;
2034 		ncookies--;
2035 	}
2036 	if (cpos < cend)
2037 		eofflag = 0;
2038 	NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
2039 	*tl++ = newnfs_false;
2040 	if (eofflag)
2041 		*tl = newnfs_true;
2042 	else
2043 		*tl = newnfs_false;
2044 	free(rbuf, M_TEMP);
2045 	free(cookies, M_TEMP);
2046 
2047 out:
2048 	NFSEXITCODE2(0, nd);
2049 	return (0);
2050 nfsmout:
2051 	vput(vp);
2052 	NFSEXITCODE2(error, nd);
2053 	return (error);
2054 }
2055 
2056 /*
2057  * Readdirplus for V3 and Readdir for V4.
2058  */
2059 int
2060 nfsrvd_readdirplus(struct nfsrv_descript *nd, int isdgram,
2061     struct vnode *vp, struct thread *p, struct nfsexstuff *exp)
2062 {
2063 	struct dirent *dp;
2064 	u_int32_t *tl;
2065 	int dirlen;
2066 	char *cpos, *cend, *rbuf;
2067 	struct vnode *nvp;
2068 	fhandle_t nfh;
2069 	struct nfsvattr nva, at, *nvap = &nva;
2070 	struct mbuf *mb0, *mb1;
2071 	struct nfsreferral *refp;
2072 	int nlen, r, error = 0, getret = 1, usevget = 1;
2073 	int siz, cnt, fullsiz, eofflag, ncookies, entrycnt;
2074 	caddr_t bpos0, bpos1;
2075 	u_int64_t off, toff, verf;
2076 	u_long *cookies = NULL, *cookiep;
2077 	nfsattrbit_t attrbits, rderrbits, savbits;
2078 	struct uio io;
2079 	struct iovec iv;
2080 	struct componentname cn;
2081 	int at_root, is_ufs, is_zfs, needs_unbusy, supports_nfsv4acls;
2082 	struct mount *mp, *new_mp;
2083 	uint64_t mounted_on_fileno;
2084 
2085 	if (nd->nd_repstat) {
2086 		nfsrv_postopattr(nd, getret, &at);
2087 		goto out;
2088 	}
2089 	NFSM_DISSECT(tl, u_int32_t *, 6 * NFSX_UNSIGNED);
2090 	off = fxdr_hyper(tl);
2091 	toff = off;
2092 	tl += 2;
2093 	verf = fxdr_hyper(tl);
2094 	tl += 2;
2095 	siz = fxdr_unsigned(int, *tl++);
2096 	cnt = fxdr_unsigned(int, *tl);
2097 
2098 	/*
2099 	 * Use the server's maximum data transfer size as the upper bound
2100 	 * on reply datalen.
2101 	 */
2102 	if (cnt > NFS_SRVMAXDATA(nd) || cnt < 0)
2103 		cnt = NFS_SRVMAXDATA(nd);
2104 
2105 	/*
2106 	 * siz is a "hint" of how much directory information (name, fileid,
2107 	 * cookie) should be in the reply. At least one client "hints" 0,
2108 	 * so I set it to cnt for that case. I also round it up to the
2109 	 * next multiple of DIRBLKSIZ.
2110 	 * Since the size of a Readdirplus directory entry reply will always
2111 	 * be greater than a directory entry returned by VOP_READDIR(), it
2112 	 * does not make sense to read more than NFS_SRVMAXDATA() via
2113 	 * VOP_READDIR().
2114 	 */
2115 	if (siz <= 0)
2116 		siz = cnt;
2117 	else if (siz > NFS_SRVMAXDATA(nd))
2118 		siz = NFS_SRVMAXDATA(nd);
2119 	siz = ((siz + DIRBLKSIZ - 1) & ~(DIRBLKSIZ - 1));
2120 
2121 	if (nd->nd_flag & ND_NFSV4) {
2122 		error = nfsrv_getattrbits(nd, &attrbits, NULL, NULL);
2123 		if (error)
2124 			goto nfsmout;
2125 		NFSSET_ATTRBIT(&savbits, &attrbits);
2126 		NFSCLRNOTFILLABLE_ATTRBIT(&attrbits);
2127 		NFSZERO_ATTRBIT(&rderrbits);
2128 		NFSSETBIT_ATTRBIT(&rderrbits, NFSATTRBIT_RDATTRERROR);
2129 	} else {
2130 		NFSZERO_ATTRBIT(&attrbits);
2131 	}
2132 	fullsiz = siz;
2133 	nd->nd_repstat = getret = nfsvno_getattr(vp, &at, nd, p, 1, NULL);
2134 #if 0
2135 	if (!nd->nd_repstat) {
2136 	    if (off && verf != at.na_filerev) {
2137 		/*
2138 		 * va_filerev is not sufficient as a cookie verifier,
2139 		 * since it is not supposed to change when entries are
2140 		 * removed/added unless that offset cookies returned to
2141 		 * the client are no longer valid.
2142 		 */
2143 		if (nd->nd_flag & ND_NFSV4) {
2144 			nd->nd_repstat = NFSERR_NOTSAME;
2145 		} else {
2146 			nd->nd_repstat = NFSERR_BAD_COOKIE;
2147 		}
2148 	    }
2149 	}
2150 #endif
2151 	if (!nd->nd_repstat && vp->v_type != VDIR)
2152 		nd->nd_repstat = NFSERR_NOTDIR;
2153 	if (!nd->nd_repstat && cnt == 0)
2154 		nd->nd_repstat = NFSERR_TOOSMALL;
2155 	if (!nd->nd_repstat)
2156 		nd->nd_repstat = nfsvno_accchk(vp, VEXEC,
2157 		    nd->nd_cred, exp, p, NFSACCCHK_NOOVERRIDE,
2158 		    NFSACCCHK_VPISLOCKED, NULL);
2159 	if (nd->nd_repstat) {
2160 		vput(vp);
2161 		if (nd->nd_flag & ND_NFSV3)
2162 			nfsrv_postopattr(nd, getret, &at);
2163 		goto out;
2164 	}
2165 	is_ufs = strcmp(vp->v_mount->mnt_vfc->vfc_name, "ufs") == 0;
2166 	is_zfs = strcmp(vp->v_mount->mnt_vfc->vfc_name, "zfs") == 0;
2167 
2168 	rbuf = malloc(siz, M_TEMP, M_WAITOK);
2169 again:
2170 	eofflag = 0;
2171 	if (cookies) {
2172 		free(cookies, M_TEMP);
2173 		cookies = NULL;
2174 	}
2175 
2176 	iv.iov_base = rbuf;
2177 	iv.iov_len = siz;
2178 	io.uio_iov = &iv;
2179 	io.uio_iovcnt = 1;
2180 	io.uio_offset = (off_t)off;
2181 	io.uio_resid = siz;
2182 	io.uio_segflg = UIO_SYSSPACE;
2183 	io.uio_rw = UIO_READ;
2184 	io.uio_td = NULL;
2185 	nd->nd_repstat = VOP_READDIR(vp, &io, nd->nd_cred, &eofflag, &ncookies,
2186 	    &cookies);
2187 	off = (u_int64_t)io.uio_offset;
2188 	if (io.uio_resid)
2189 		siz -= io.uio_resid;
2190 
2191 	getret = nfsvno_getattr(vp, &at, nd, p, 1, NULL);
2192 
2193 	if (!cookies && !nd->nd_repstat)
2194 		nd->nd_repstat = NFSERR_PERM;
2195 	if (!nd->nd_repstat)
2196 		nd->nd_repstat = getret;
2197 	if (nd->nd_repstat) {
2198 		vput(vp);
2199 		if (cookies)
2200 			free(cookies, M_TEMP);
2201 		free(rbuf, M_TEMP);
2202 		if (nd->nd_flag & ND_NFSV3)
2203 			nfsrv_postopattr(nd, getret, &at);
2204 		goto out;
2205 	}
2206 	/*
2207 	 * If nothing read, return eof
2208 	 * rpc reply
2209 	 */
2210 	if (siz == 0) {
2211 		vput(vp);
2212 		if (nd->nd_flag & ND_NFSV3)
2213 			nfsrv_postopattr(nd, getret, &at);
2214 		NFSM_BUILD(tl, u_int32_t *, 4 * NFSX_UNSIGNED);
2215 		txdr_hyper(at.na_filerev, tl);
2216 		tl += 2;
2217 		*tl++ = newnfs_false;
2218 		*tl = newnfs_true;
2219 		free(cookies, M_TEMP);
2220 		free(rbuf, M_TEMP);
2221 		goto out;
2222 	}
2223 
2224 	/*
2225 	 * Check for degenerate cases of nothing useful read.
2226 	 * If so go try again
2227 	 */
2228 	cpos = rbuf;
2229 	cend = rbuf + siz;
2230 	dp = (struct dirent *)cpos;
2231 	cookiep = cookies;
2232 
2233 	/*
2234 	 * For some reason FreeBSD's ufs_readdir() chooses to back the
2235 	 * directory offset up to a block boundary, so it is necessary to
2236 	 * skip over the records that precede the requested offset. This
2237 	 * requires the assumption that file offset cookies monotonically
2238 	 * increase.
2239 	 */
2240 	while (cpos < cend && ncookies > 0 &&
2241 	  (dp->d_fileno == 0 || dp->d_type == DT_WHT ||
2242 	   (is_ufs == 1 && ((u_quad_t)(*cookiep)) <= toff) ||
2243 	   ((nd->nd_flag & ND_NFSV4) &&
2244 	    ((dp->d_namlen == 1 && dp->d_name[0] == '.') ||
2245 	     (dp->d_namlen==2 && dp->d_name[0]=='.' && dp->d_name[1]=='.'))))) {
2246 		cpos += dp->d_reclen;
2247 		dp = (struct dirent *)cpos;
2248 		cookiep++;
2249 		ncookies--;
2250 	}
2251 	if (cpos >= cend || ncookies == 0) {
2252 		siz = fullsiz;
2253 		toff = off;
2254 		goto again;
2255 	}
2256 
2257 	/*
2258 	 * Busy the file system so that the mount point won't go away
2259 	 * and, as such, VFS_VGET() can be used safely.
2260 	 */
2261 	mp = vp->v_mount;
2262 	vfs_ref(mp);
2263 	NFSVOPUNLOCK(vp, 0);
2264 	nd->nd_repstat = vfs_busy(mp, 0);
2265 	vfs_rel(mp);
2266 	if (nd->nd_repstat != 0) {
2267 		vrele(vp);
2268 		free(cookies, M_TEMP);
2269 		free(rbuf, M_TEMP);
2270 		if (nd->nd_flag & ND_NFSV3)
2271 			nfsrv_postopattr(nd, getret, &at);
2272 		goto out;
2273 	}
2274 
2275 	/*
2276 	 * Check to see if entries in this directory can be safely acquired
2277 	 * via VFS_VGET() or if a switch to VOP_LOOKUP() is required.
2278 	 * ZFS snapshot directories need VOP_LOOKUP(), so that any
2279 	 * automount of the snapshot directory that is required will
2280 	 * be done.
2281 	 * This needs to be done here for NFSv4, since NFSv4 never does
2282 	 * a VFS_VGET() for "." or "..".
2283 	 */
2284 	if (is_zfs == 1) {
2285 		r = VFS_VGET(mp, at.na_fileid, LK_SHARED, &nvp);
2286 		if (r == EOPNOTSUPP) {
2287 			usevget = 0;
2288 			cn.cn_nameiop = LOOKUP;
2289 			cn.cn_lkflags = LK_SHARED | LK_RETRY;
2290 			cn.cn_cred = nd->nd_cred;
2291 			cn.cn_thread = p;
2292 		} else if (r == 0)
2293 			vput(nvp);
2294 	}
2295 
2296 	/*
2297 	 * Save this position, in case there is an error before one entry
2298 	 * is created.
2299 	 */
2300 	mb0 = nd->nd_mb;
2301 	bpos0 = nd->nd_bpos;
2302 
2303 	/*
2304 	 * Fill in the first part of the reply.
2305 	 * dirlen is the reply length in bytes and cannot exceed cnt.
2306 	 * (Include the two booleans at the end of the reply in dirlen now,
2307 	 *  so we recognize when we have exceeded cnt.)
2308 	 */
2309 	if (nd->nd_flag & ND_NFSV3) {
2310 		dirlen = NFSX_V3POSTOPATTR + NFSX_VERF + 2 * NFSX_UNSIGNED;
2311 		nfsrv_postopattr(nd, getret, &at);
2312 	} else {
2313 		dirlen = NFSX_VERF + 2 * NFSX_UNSIGNED;
2314 	}
2315 	NFSM_BUILD(tl, u_int32_t *, NFSX_VERF);
2316 	txdr_hyper(at.na_filerev, tl);
2317 
2318 	/*
2319 	 * Save this position, in case there is an empty reply needed.
2320 	 */
2321 	mb1 = nd->nd_mb;
2322 	bpos1 = nd->nd_bpos;
2323 
2324 	/* Loop through the records and build reply */
2325 	entrycnt = 0;
2326 	while (cpos < cend && ncookies > 0 && dirlen < cnt) {
2327 		nlen = dp->d_namlen;
2328 		if (dp->d_fileno != 0 && dp->d_type != DT_WHT &&
2329 		    nlen <= NFS_MAXNAMLEN &&
2330 		    ((nd->nd_flag & ND_NFSV3) || nlen > 2 ||
2331 		     (nlen==2 && (dp->d_name[0]!='.' || dp->d_name[1]!='.'))
2332 		      || (nlen == 1 && dp->d_name[0] != '.'))) {
2333 			/*
2334 			 * Save the current position in the reply, in case
2335 			 * this entry exceeds cnt.
2336 			 */
2337 			mb1 = nd->nd_mb;
2338 			bpos1 = nd->nd_bpos;
2339 
2340 			/*
2341 			 * For readdir_and_lookup get the vnode using
2342 			 * the file number.
2343 			 */
2344 			nvp = NULL;
2345 			refp = NULL;
2346 			r = 0;
2347 			at_root = 0;
2348 			needs_unbusy = 0;
2349 			new_mp = mp;
2350 			mounted_on_fileno = (uint64_t)dp->d_fileno;
2351 			if ((nd->nd_flag & ND_NFSV3) ||
2352 			    NFSNONZERO_ATTRBIT(&savbits)) {
2353 				if (nd->nd_flag & ND_NFSV4)
2354 					refp = nfsv4root_getreferral(NULL,
2355 					    vp, dp->d_fileno);
2356 				if (refp == NULL) {
2357 					if (usevget)
2358 						r = VFS_VGET(mp, dp->d_fileno,
2359 						    LK_SHARED, &nvp);
2360 					else
2361 						r = EOPNOTSUPP;
2362 					if (r == EOPNOTSUPP) {
2363 						if (usevget) {
2364 							usevget = 0;
2365 							cn.cn_nameiop = LOOKUP;
2366 							cn.cn_lkflags =
2367 							    LK_SHARED |
2368 							    LK_RETRY;
2369 							cn.cn_cred =
2370 							    nd->nd_cred;
2371 							cn.cn_thread = p;
2372 						}
2373 						cn.cn_nameptr = dp->d_name;
2374 						cn.cn_namelen = nlen;
2375 						cn.cn_flags = ISLASTCN |
2376 						    NOFOLLOW | LOCKLEAF;
2377 						if (nlen == 2 &&
2378 						    dp->d_name[0] == '.' &&
2379 						    dp->d_name[1] == '.')
2380 							cn.cn_flags |=
2381 							    ISDOTDOT;
2382 						if (NFSVOPLOCK(vp, LK_SHARED)
2383 						    != 0) {
2384 							nd->nd_repstat = EPERM;
2385 							break;
2386 						}
2387 						if ((vp->v_vflag & VV_ROOT) != 0
2388 						    && (cn.cn_flags & ISDOTDOT)
2389 						    != 0) {
2390 							vref(vp);
2391 							nvp = vp;
2392 							r = 0;
2393 						} else {
2394 							r = VOP_LOOKUP(vp, &nvp,
2395 							    &cn);
2396 							if (vp != nvp)
2397 								NFSVOPUNLOCK(vp,
2398 								    0);
2399 						}
2400 					}
2401 
2402 					/*
2403 					 * For NFSv4, check to see if nvp is
2404 					 * a mount point and get the mount
2405 					 * point vnode, as required.
2406 					 */
2407 					if (r == 0 &&
2408 					    nfsrv_enable_crossmntpt != 0 &&
2409 					    (nd->nd_flag & ND_NFSV4) != 0 &&
2410 					    nvp->v_type == VDIR &&
2411 					    nvp->v_mountedhere != NULL) {
2412 						new_mp = nvp->v_mountedhere;
2413 						r = vfs_busy(new_mp, 0);
2414 						vput(nvp);
2415 						nvp = NULL;
2416 						if (r == 0) {
2417 							r = VFS_ROOT(new_mp,
2418 							    LK_SHARED, &nvp);
2419 							needs_unbusy = 1;
2420 							if (r == 0)
2421 								at_root = 1;
2422 						}
2423 					}
2424 				}
2425 
2426 				/*
2427 				 * If we failed to look up the entry, then it
2428 				 * has become invalid, most likely removed.
2429 				 */
2430 				if (r != 0) {
2431 					if (needs_unbusy)
2432 						vfs_unbusy(new_mp);
2433 					goto invalid;
2434 				}
2435 				KASSERT(refp != NULL || nvp != NULL,
2436 				    ("%s: undetected lookup error", __func__));
2437 
2438 				if (refp == NULL &&
2439 				    ((nd->nd_flag & ND_NFSV3) ||
2440 				     NFSNONZERO_ATTRBIT(&attrbits))) {
2441 					r = nfsvno_getfh(nvp, &nfh, p);
2442 					if (!r)
2443 					    r = nfsvno_getattr(nvp, nvap, nd, p,
2444 						1, &attrbits);
2445 					if (r == 0 && is_zfs == 1 &&
2446 					    nfsrv_enable_crossmntpt != 0 &&
2447 					    (nd->nd_flag & ND_NFSV4) != 0 &&
2448 					    nvp->v_type == VDIR &&
2449 					    vp->v_mount != nvp->v_mount) {
2450 					    /*
2451 					     * For a ZFS snapshot, there is a
2452 					     * pseudo mount that does not set
2453 					     * v_mountedhere, so it needs to
2454 					     * be detected via a different
2455 					     * mount structure.
2456 					     */
2457 					    at_root = 1;
2458 					    if (new_mp == mp)
2459 						new_mp = nvp->v_mount;
2460 					}
2461 				}
2462 
2463 				/*
2464 				 * If we failed to get attributes of the entry,
2465 				 * then just skip it for NFSv3 (the traditional
2466 				 * behavior in the old NFS server).
2467 				 * For NFSv4 the behavior is controlled by
2468 				 * RDATTRERROR: we either ignore the error or
2469 				 * fail the request.
2470 				 * Note that RDATTRERROR is never set for NFSv3.
2471 				 */
2472 				if (r != 0) {
2473 					if (!NFSISSET_ATTRBIT(&attrbits,
2474 					    NFSATTRBIT_RDATTRERROR)) {
2475 						vput(nvp);
2476 						if (needs_unbusy != 0)
2477 							vfs_unbusy(new_mp);
2478 						if ((nd->nd_flag & ND_NFSV3))
2479 							goto invalid;
2480 						nd->nd_repstat = r;
2481 						break;
2482 					}
2483 				}
2484 			}
2485 
2486 			/*
2487 			 * Build the directory record xdr
2488 			 */
2489 			if (nd->nd_flag & ND_NFSV3) {
2490 				NFSM_BUILD(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
2491 				*tl++ = newnfs_true;
2492 				*tl++ = 0;
2493 				*tl = txdr_unsigned(dp->d_fileno);
2494 				dirlen += nfsm_strtom(nd, dp->d_name, nlen);
2495 				NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
2496 				*tl++ = 0;
2497 				*tl = txdr_unsigned(*cookiep);
2498 				nfsrv_postopattr(nd, 0, nvap);
2499 				dirlen += nfsm_fhtom(nd,(u_int8_t *)&nfh,0,1);
2500 				dirlen += (5*NFSX_UNSIGNED+NFSX_V3POSTOPATTR);
2501 				if (nvp != NULL)
2502 					vput(nvp);
2503 			} else {
2504 				NFSM_BUILD(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
2505 				*tl++ = newnfs_true;
2506 				*tl++ = 0;
2507 				*tl = txdr_unsigned(*cookiep);
2508 				dirlen += nfsm_strtom(nd, dp->d_name, nlen);
2509 				if (nvp != NULL) {
2510 					supports_nfsv4acls =
2511 					    nfs_supportsnfsv4acls(nvp);
2512 					NFSVOPUNLOCK(nvp, 0);
2513 				} else
2514 					supports_nfsv4acls = 0;
2515 				if (refp != NULL) {
2516 					dirlen += nfsrv_putreferralattr(nd,
2517 					    &savbits, refp, 0,
2518 					    &nd->nd_repstat);
2519 					if (nd->nd_repstat) {
2520 						if (nvp != NULL)
2521 							vrele(nvp);
2522 						if (needs_unbusy != 0)
2523 							vfs_unbusy(new_mp);
2524 						break;
2525 					}
2526 				} else if (r) {
2527 					dirlen += nfsvno_fillattr(nd, new_mp,
2528 					    nvp, nvap, &nfh, r, &rderrbits,
2529 					    nd->nd_cred, p, isdgram, 0,
2530 					    supports_nfsv4acls, at_root,
2531 					    mounted_on_fileno);
2532 				} else {
2533 					dirlen += nfsvno_fillattr(nd, new_mp,
2534 					    nvp, nvap, &nfh, r, &attrbits,
2535 					    nd->nd_cred, p, isdgram, 0,
2536 					    supports_nfsv4acls, at_root,
2537 					    mounted_on_fileno);
2538 				}
2539 				if (nvp != NULL)
2540 					vrele(nvp);
2541 				dirlen += (3 * NFSX_UNSIGNED);
2542 			}
2543 			if (needs_unbusy != 0)
2544 				vfs_unbusy(new_mp);
2545 			if (dirlen <= cnt)
2546 				entrycnt++;
2547 		}
2548 invalid:
2549 		cpos += dp->d_reclen;
2550 		dp = (struct dirent *)cpos;
2551 		cookiep++;
2552 		ncookies--;
2553 	}
2554 	vrele(vp);
2555 	vfs_unbusy(mp);
2556 
2557 	/*
2558 	 * If dirlen > cnt, we must strip off the last entry. If that
2559 	 * results in an empty reply, report NFSERR_TOOSMALL.
2560 	 */
2561 	if (dirlen > cnt || nd->nd_repstat) {
2562 		if (!nd->nd_repstat && entrycnt == 0)
2563 			nd->nd_repstat = NFSERR_TOOSMALL;
2564 		if (nd->nd_repstat) {
2565 			newnfs_trimtrailing(nd, mb0, bpos0);
2566 			if (nd->nd_flag & ND_NFSV3)
2567 				nfsrv_postopattr(nd, getret, &at);
2568 		} else
2569 			newnfs_trimtrailing(nd, mb1, bpos1);
2570 		eofflag = 0;
2571 	} else if (cpos < cend)
2572 		eofflag = 0;
2573 	if (!nd->nd_repstat) {
2574 		NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
2575 		*tl++ = newnfs_false;
2576 		if (eofflag)
2577 			*tl = newnfs_true;
2578 		else
2579 			*tl = newnfs_false;
2580 	}
2581 	free(cookies, M_TEMP);
2582 	free(rbuf, M_TEMP);
2583 
2584 out:
2585 	NFSEXITCODE2(0, nd);
2586 	return (0);
2587 nfsmout:
2588 	vput(vp);
2589 	NFSEXITCODE2(error, nd);
2590 	return (error);
2591 }
2592 
2593 /*
2594  * Get the settable attributes out of the mbuf list.
2595  * (Return 0 or EBADRPC)
2596  */
2597 int
2598 nfsrv_sattr(struct nfsrv_descript *nd, vnode_t vp, struct nfsvattr *nvap,
2599     nfsattrbit_t *attrbitp, NFSACL_T *aclp, struct thread *p)
2600 {
2601 	u_int32_t *tl;
2602 	struct nfsv2_sattr *sp;
2603 	int error = 0, toclient = 0;
2604 
2605 	switch (nd->nd_flag & (ND_NFSV2 | ND_NFSV3 | ND_NFSV4)) {
2606 	case ND_NFSV2:
2607 		NFSM_DISSECT(sp, struct nfsv2_sattr *, NFSX_V2SATTR);
2608 		/*
2609 		 * Some old clients didn't fill in the high order 16bits.
2610 		 * --> check the low order 2 bytes for 0xffff
2611 		 */
2612 		if ((fxdr_unsigned(int, sp->sa_mode) & 0xffff) != 0xffff)
2613 			nvap->na_mode = nfstov_mode(sp->sa_mode);
2614 		if (sp->sa_uid != newnfs_xdrneg1)
2615 			nvap->na_uid = fxdr_unsigned(uid_t, sp->sa_uid);
2616 		if (sp->sa_gid != newnfs_xdrneg1)
2617 			nvap->na_gid = fxdr_unsigned(gid_t, sp->sa_gid);
2618 		if (sp->sa_size != newnfs_xdrneg1)
2619 			nvap->na_size = fxdr_unsigned(u_quad_t, sp->sa_size);
2620 		if (sp->sa_atime.nfsv2_sec != newnfs_xdrneg1) {
2621 #ifdef notyet
2622 			fxdr_nfsv2time(&sp->sa_atime, &nvap->na_atime);
2623 #else
2624 			nvap->na_atime.tv_sec =
2625 				fxdr_unsigned(u_int32_t,sp->sa_atime.nfsv2_sec);
2626 			nvap->na_atime.tv_nsec = 0;
2627 #endif
2628 		}
2629 		if (sp->sa_mtime.nfsv2_sec != newnfs_xdrneg1)
2630 			fxdr_nfsv2time(&sp->sa_mtime, &nvap->na_mtime);
2631 		break;
2632 	case ND_NFSV3:
2633 		NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2634 		if (*tl == newnfs_true) {
2635 			NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2636 			nvap->na_mode = nfstov_mode(*tl);
2637 		}
2638 		NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2639 		if (*tl == newnfs_true) {
2640 			NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2641 			nvap->na_uid = fxdr_unsigned(uid_t, *tl);
2642 		}
2643 		NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2644 		if (*tl == newnfs_true) {
2645 			NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2646 			nvap->na_gid = fxdr_unsigned(gid_t, *tl);
2647 		}
2648 		NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2649 		if (*tl == newnfs_true) {
2650 			NFSM_DISSECT(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
2651 			nvap->na_size = fxdr_hyper(tl);
2652 		}
2653 		NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2654 		switch (fxdr_unsigned(int, *tl)) {
2655 		case NFSV3SATTRTIME_TOCLIENT:
2656 			NFSM_DISSECT(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
2657 			fxdr_nfsv3time(tl, &nvap->na_atime);
2658 			toclient = 1;
2659 			break;
2660 		case NFSV3SATTRTIME_TOSERVER:
2661 			vfs_timestamp(&nvap->na_atime);
2662 			nvap->na_vaflags |= VA_UTIMES_NULL;
2663 			break;
2664 		}
2665 		NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2666 		switch (fxdr_unsigned(int, *tl)) {
2667 		case NFSV3SATTRTIME_TOCLIENT:
2668 			NFSM_DISSECT(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
2669 			fxdr_nfsv3time(tl, &nvap->na_mtime);
2670 			nvap->na_vaflags &= ~VA_UTIMES_NULL;
2671 			break;
2672 		case NFSV3SATTRTIME_TOSERVER:
2673 			vfs_timestamp(&nvap->na_mtime);
2674 			if (!toclient)
2675 				nvap->na_vaflags |= VA_UTIMES_NULL;
2676 			break;
2677 		}
2678 		break;
2679 	case ND_NFSV4:
2680 		error = nfsv4_sattr(nd, vp, nvap, attrbitp, aclp, p);
2681 	}
2682 nfsmout:
2683 	NFSEXITCODE2(error, nd);
2684 	return (error);
2685 }
2686 
2687 /*
2688  * Handle the setable attributes for V4.
2689  * Returns NFSERR_BADXDR if it can't be parsed, 0 otherwise.
2690  */
2691 int
2692 nfsv4_sattr(struct nfsrv_descript *nd, vnode_t vp, struct nfsvattr *nvap,
2693     nfsattrbit_t *attrbitp, NFSACL_T *aclp, struct thread *p)
2694 {
2695 	u_int32_t *tl;
2696 	int attrsum = 0;
2697 	int i, j;
2698 	int error, attrsize, bitpos, aclsize, aceerr, retnotsup = 0;
2699 	int toclient = 0;
2700 	u_char *cp, namestr[NFSV4_SMALLSTR + 1];
2701 	uid_t uid;
2702 	gid_t gid;
2703 
2704 	error = nfsrv_getattrbits(nd, attrbitp, NULL, &retnotsup);
2705 	if (error)
2706 		goto nfsmout;
2707 	NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2708 	attrsize = fxdr_unsigned(int, *tl);
2709 
2710 	/*
2711 	 * Loop around getting the setable attributes. If an unsupported
2712 	 * one is found, set nd_repstat == NFSERR_ATTRNOTSUPP and return.
2713 	 */
2714 	if (retnotsup) {
2715 		nd->nd_repstat = NFSERR_ATTRNOTSUPP;
2716 		bitpos = NFSATTRBIT_MAX;
2717 	} else {
2718 		bitpos = 0;
2719 	}
2720 	for (; bitpos < NFSATTRBIT_MAX; bitpos++) {
2721 	    if (attrsum > attrsize) {
2722 		error = NFSERR_BADXDR;
2723 		goto nfsmout;
2724 	    }
2725 	    if (NFSISSET_ATTRBIT(attrbitp, bitpos))
2726 		switch (bitpos) {
2727 		case NFSATTRBIT_SIZE:
2728 			NFSM_DISSECT(tl, u_int32_t *, NFSX_HYPER);
2729                      if (vp != NULL && vp->v_type != VREG) {
2730                             error = (vp->v_type == VDIR) ? NFSERR_ISDIR :
2731                                 NFSERR_INVAL;
2732                             goto nfsmout;
2733 			}
2734 			nvap->na_size = fxdr_hyper(tl);
2735 			attrsum += NFSX_HYPER;
2736 			break;
2737 		case NFSATTRBIT_ACL:
2738 			error = nfsrv_dissectacl(nd, aclp, &aceerr, &aclsize,
2739 			    p);
2740 			if (error)
2741 				goto nfsmout;
2742 			if (aceerr && !nd->nd_repstat)
2743 				nd->nd_repstat = aceerr;
2744 			attrsum += aclsize;
2745 			break;
2746 		case NFSATTRBIT_ARCHIVE:
2747 			NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2748 			if (!nd->nd_repstat)
2749 				nd->nd_repstat = NFSERR_ATTRNOTSUPP;
2750 			attrsum += NFSX_UNSIGNED;
2751 			break;
2752 		case NFSATTRBIT_HIDDEN:
2753 			NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2754 			if (!nd->nd_repstat)
2755 				nd->nd_repstat = NFSERR_ATTRNOTSUPP;
2756 			attrsum += NFSX_UNSIGNED;
2757 			break;
2758 		case NFSATTRBIT_MIMETYPE:
2759 			NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2760 			i = fxdr_unsigned(int, *tl);
2761 			error = nfsm_advance(nd, NFSM_RNDUP(i), -1);
2762 			if (error)
2763 				goto nfsmout;
2764 			if (!nd->nd_repstat)
2765 				nd->nd_repstat = NFSERR_ATTRNOTSUPP;
2766 			attrsum += (NFSX_UNSIGNED + NFSM_RNDUP(i));
2767 			break;
2768 		case NFSATTRBIT_MODE:
2769 			NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2770 			nvap->na_mode = nfstov_mode(*tl);
2771 			attrsum += NFSX_UNSIGNED;
2772 			break;
2773 		case NFSATTRBIT_OWNER:
2774 			NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2775 			j = fxdr_unsigned(int, *tl);
2776 			if (j < 0) {
2777 				error = NFSERR_BADXDR;
2778 				goto nfsmout;
2779 			}
2780 			if (j > NFSV4_SMALLSTR)
2781 				cp = malloc(j + 1, M_NFSSTRING, M_WAITOK);
2782 			else
2783 				cp = namestr;
2784 			error = nfsrv_mtostr(nd, cp, j);
2785 			if (error) {
2786 				if (j > NFSV4_SMALLSTR)
2787 					free(cp, M_NFSSTRING);
2788 				goto nfsmout;
2789 			}
2790 			if (!nd->nd_repstat) {
2791 				nd->nd_repstat = nfsv4_strtouid(nd, cp, j, &uid,
2792 				    p);
2793 				if (!nd->nd_repstat)
2794 					nvap->na_uid = uid;
2795 			}
2796 			if (j > NFSV4_SMALLSTR)
2797 				free(cp, M_NFSSTRING);
2798 			attrsum += (NFSX_UNSIGNED + NFSM_RNDUP(j));
2799 			break;
2800 		case NFSATTRBIT_OWNERGROUP:
2801 			NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2802 			j = fxdr_unsigned(int, *tl);
2803 			if (j < 0) {
2804 				error = NFSERR_BADXDR;
2805 				goto nfsmout;
2806 			}
2807 			if (j > NFSV4_SMALLSTR)
2808 				cp = malloc(j + 1, M_NFSSTRING, M_WAITOK);
2809 			else
2810 				cp = namestr;
2811 			error = nfsrv_mtostr(nd, cp, j);
2812 			if (error) {
2813 				if (j > NFSV4_SMALLSTR)
2814 					free(cp, M_NFSSTRING);
2815 				goto nfsmout;
2816 			}
2817 			if (!nd->nd_repstat) {
2818 				nd->nd_repstat = nfsv4_strtogid(nd, cp, j, &gid,
2819 				    p);
2820 				if (!nd->nd_repstat)
2821 					nvap->na_gid = gid;
2822 			}
2823 			if (j > NFSV4_SMALLSTR)
2824 				free(cp, M_NFSSTRING);
2825 			attrsum += (NFSX_UNSIGNED + NFSM_RNDUP(j));
2826 			break;
2827 		case NFSATTRBIT_SYSTEM:
2828 			NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2829 			if (!nd->nd_repstat)
2830 				nd->nd_repstat = NFSERR_ATTRNOTSUPP;
2831 			attrsum += NFSX_UNSIGNED;
2832 			break;
2833 		case NFSATTRBIT_TIMEACCESSSET:
2834 			NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2835 			attrsum += NFSX_UNSIGNED;
2836 			if (fxdr_unsigned(int, *tl)==NFSV4SATTRTIME_TOCLIENT) {
2837 			    NFSM_DISSECT(tl, u_int32_t *, NFSX_V4TIME);
2838 			    fxdr_nfsv4time(tl, &nvap->na_atime);
2839 			    toclient = 1;
2840 			    attrsum += NFSX_V4TIME;
2841 			} else {
2842 			    vfs_timestamp(&nvap->na_atime);
2843 			    nvap->na_vaflags |= VA_UTIMES_NULL;
2844 			}
2845 			break;
2846 		case NFSATTRBIT_TIMEBACKUP:
2847 			NFSM_DISSECT(tl, u_int32_t *, NFSX_V4TIME);
2848 			if (!nd->nd_repstat)
2849 				nd->nd_repstat = NFSERR_ATTRNOTSUPP;
2850 			attrsum += NFSX_V4TIME;
2851 			break;
2852 		case NFSATTRBIT_TIMECREATE:
2853 			NFSM_DISSECT(tl, u_int32_t *, NFSX_V4TIME);
2854 			if (!nd->nd_repstat)
2855 				nd->nd_repstat = NFSERR_ATTRNOTSUPP;
2856 			attrsum += NFSX_V4TIME;
2857 			break;
2858 		case NFSATTRBIT_TIMEMODIFYSET:
2859 			NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2860 			attrsum += NFSX_UNSIGNED;
2861 			if (fxdr_unsigned(int, *tl)==NFSV4SATTRTIME_TOCLIENT) {
2862 			    NFSM_DISSECT(tl, u_int32_t *, NFSX_V4TIME);
2863 			    fxdr_nfsv4time(tl, &nvap->na_mtime);
2864 			    nvap->na_vaflags &= ~VA_UTIMES_NULL;
2865 			    attrsum += NFSX_V4TIME;
2866 			} else {
2867 			    vfs_timestamp(&nvap->na_mtime);
2868 			    if (!toclient)
2869 				nvap->na_vaflags |= VA_UTIMES_NULL;
2870 			}
2871 			break;
2872 		default:
2873 			nd->nd_repstat = NFSERR_ATTRNOTSUPP;
2874 			/*
2875 			 * set bitpos so we drop out of the loop.
2876 			 */
2877 			bitpos = NFSATTRBIT_MAX;
2878 			break;
2879 		}
2880 	}
2881 
2882 	/*
2883 	 * some clients pad the attrlist, so we need to skip over the
2884 	 * padding.
2885 	 */
2886 	if (attrsum > attrsize) {
2887 		error = NFSERR_BADXDR;
2888 	} else {
2889 		attrsize = NFSM_RNDUP(attrsize);
2890 		if (attrsum < attrsize)
2891 			error = nfsm_advance(nd, attrsize - attrsum, -1);
2892 	}
2893 nfsmout:
2894 	NFSEXITCODE2(error, nd);
2895 	return (error);
2896 }
2897 
2898 /*
2899  * Check/setup export credentials.
2900  */
2901 int
2902 nfsd_excred(struct nfsrv_descript *nd, struct nfsexstuff *exp,
2903     struct ucred *credanon)
2904 {
2905 	int error = 0;
2906 
2907 	/*
2908 	 * Check/setup credentials.
2909 	 */
2910 	if (nd->nd_flag & ND_GSS)
2911 		exp->nes_exflag &= ~MNT_EXPORTANON;
2912 
2913 	/*
2914 	 * Check to see if the operation is allowed for this security flavor.
2915 	 * RFC2623 suggests that the NFSv3 Fsinfo RPC be allowed to
2916 	 * AUTH_NONE or AUTH_SYS for file systems requiring RPCSEC_GSS.
2917 	 * Also, allow Secinfo, so that it can acquire the correct flavor(s).
2918 	 */
2919 	if (nfsvno_testexp(nd, exp) &&
2920 	    nd->nd_procnum != NFSV4OP_SECINFO &&
2921 	    nd->nd_procnum != NFSPROC_FSINFO) {
2922 		if (nd->nd_flag & ND_NFSV4)
2923 			error = NFSERR_WRONGSEC;
2924 		else
2925 			error = (NFSERR_AUTHERR | AUTH_TOOWEAK);
2926 		goto out;
2927 	}
2928 
2929 	/*
2930 	 * Check to see if the file system is exported V4 only.
2931 	 */
2932 	if (NFSVNO_EXV4ONLY(exp) && !(nd->nd_flag & ND_NFSV4)) {
2933 		error = NFSERR_PROGNOTV4;
2934 		goto out;
2935 	}
2936 
2937 	/*
2938 	 * Now, map the user credentials.
2939 	 * (Note that ND_AUTHNONE will only be set for an NFSv3
2940 	 *  Fsinfo RPC. If set for anything else, this code might need
2941 	 *  to change.)
2942 	 */
2943 	if (NFSVNO_EXPORTED(exp)) {
2944 		if (((nd->nd_flag & ND_GSS) == 0 && nd->nd_cred->cr_uid == 0) ||
2945 		     NFSVNO_EXPORTANON(exp) ||
2946 		     (nd->nd_flag & ND_AUTHNONE) != 0) {
2947 			nd->nd_cred->cr_uid = credanon->cr_uid;
2948 			nd->nd_cred->cr_gid = credanon->cr_gid;
2949 			crsetgroups(nd->nd_cred, credanon->cr_ngroups,
2950 			    credanon->cr_groups);
2951 		} else if ((nd->nd_flag & ND_GSS) == 0) {
2952 			/*
2953 			 * If using AUTH_SYS, call nfsrv_getgrpscred() to see
2954 			 * if there is a replacement credential with a group
2955 			 * list set up by "nfsuserd -manage-gids".
2956 			 * If there is no replacement, nfsrv_getgrpscred()
2957 			 * simply returns its argument.
2958 			 */
2959 			nd->nd_cred = nfsrv_getgrpscred(nd->nd_cred);
2960 		}
2961 	}
2962 
2963 out:
2964 	NFSEXITCODE2(error, nd);
2965 	return (error);
2966 }
2967 
2968 /*
2969  * Check exports.
2970  */
2971 int
2972 nfsvno_checkexp(struct mount *mp, struct sockaddr *nam, struct nfsexstuff *exp,
2973     struct ucred **credp)
2974 {
2975 	int i, error, *secflavors;
2976 
2977 	error = VFS_CHECKEXP(mp, nam, &exp->nes_exflag, credp,
2978 	    &exp->nes_numsecflavor, &secflavors);
2979 	if (error) {
2980 		if (nfs_rootfhset) {
2981 			exp->nes_exflag = 0;
2982 			exp->nes_numsecflavor = 0;
2983 			error = 0;
2984 		}
2985 	} else {
2986 		/* Copy the security flavors. */
2987 		for (i = 0; i < exp->nes_numsecflavor; i++)
2988 			exp->nes_secflavors[i] = secflavors[i];
2989 	}
2990 	NFSEXITCODE(error);
2991 	return (error);
2992 }
2993 
2994 /*
2995  * Get a vnode for a file handle and export stuff.
2996  */
2997 int
2998 nfsvno_fhtovp(struct mount *mp, fhandle_t *fhp, struct sockaddr *nam,
2999     int lktype, struct vnode **vpp, struct nfsexstuff *exp,
3000     struct ucred **credp)
3001 {
3002 	int i, error, *secflavors;
3003 
3004 	*credp = NULL;
3005 	exp->nes_numsecflavor = 0;
3006 	error = VFS_FHTOVP(mp, &fhp->fh_fid, lktype, vpp);
3007 	if (error != 0)
3008 		/* Make sure the server replies ESTALE to the client. */
3009 		error = ESTALE;
3010 	if (nam && !error) {
3011 		error = VFS_CHECKEXP(mp, nam, &exp->nes_exflag, credp,
3012 		    &exp->nes_numsecflavor, &secflavors);
3013 		if (error) {
3014 			if (nfs_rootfhset) {
3015 				exp->nes_exflag = 0;
3016 				exp->nes_numsecflavor = 0;
3017 				error = 0;
3018 			} else {
3019 				vput(*vpp);
3020 			}
3021 		} else {
3022 			/* Copy the security flavors. */
3023 			for (i = 0; i < exp->nes_numsecflavor; i++)
3024 				exp->nes_secflavors[i] = secflavors[i];
3025 		}
3026 	}
3027 	NFSEXITCODE(error);
3028 	return (error);
3029 }
3030 
3031 /*
3032  * nfsd_fhtovp() - convert a fh to a vnode ptr
3033  * 	- look up fsid in mount list (if not found ret error)
3034  *	- get vp and export rights by calling nfsvno_fhtovp()
3035  *	- if cred->cr_uid == 0 or MNT_EXPORTANON set it to credanon
3036  *	  for AUTH_SYS
3037  *	- if mpp != NULL, return the mount point so that it can
3038  *	  be used for vn_finished_write() by the caller
3039  */
3040 void
3041 nfsd_fhtovp(struct nfsrv_descript *nd, struct nfsrvfh *nfp, int lktype,
3042     struct vnode **vpp, struct nfsexstuff *exp,
3043     struct mount **mpp, int startwrite, struct thread *p)
3044 {
3045 	struct mount *mp;
3046 	struct ucred *credanon;
3047 	fhandle_t *fhp;
3048 
3049 	fhp = (fhandle_t *)nfp->nfsrvfh_data;
3050 	/*
3051 	 * Check for the special case of the nfsv4root_fh.
3052 	 */
3053 	mp = vfs_busyfs(&fhp->fh_fsid);
3054 	if (mpp != NULL)
3055 		*mpp = mp;
3056 	if (mp == NULL) {
3057 		*vpp = NULL;
3058 		nd->nd_repstat = ESTALE;
3059 		goto out;
3060 	}
3061 
3062 	if (startwrite) {
3063 		vn_start_write(NULL, mpp, V_WAIT);
3064 		if (lktype == LK_SHARED && !(MNT_SHARED_WRITES(mp)))
3065 			lktype = LK_EXCLUSIVE;
3066 	}
3067 	nd->nd_repstat = nfsvno_fhtovp(mp, fhp, nd->nd_nam, lktype, vpp, exp,
3068 	    &credanon);
3069 	vfs_unbusy(mp);
3070 
3071 	/*
3072 	 * For NFSv4 without a pseudo root fs, unexported file handles
3073 	 * can be returned, so that Lookup works everywhere.
3074 	 */
3075 	if (!nd->nd_repstat && exp->nes_exflag == 0 &&
3076 	    !(nd->nd_flag & ND_NFSV4)) {
3077 		vput(*vpp);
3078 		nd->nd_repstat = EACCES;
3079 	}
3080 
3081 	/*
3082 	 * Personally, I've never seen any point in requiring a
3083 	 * reserved port#, since only in the rare case where the
3084 	 * clients are all boxes with secure system privileges,
3085 	 * does it provide any enhanced security, but... some people
3086 	 * believe it to be useful and keep putting this code back in.
3087 	 * (There is also some "security checker" out there that
3088 	 *  complains if the nfs server doesn't enforce this.)
3089 	 * However, note the following:
3090 	 * RFC3530 (NFSv4) specifies that a reserved port# not be
3091 	 *	required.
3092 	 * RFC2623 recommends that, if a reserved port# is checked for,
3093 	 *	that there be a way to turn that off--> ifdef'd.
3094 	 */
3095 #ifdef NFS_REQRSVPORT
3096 	if (!nd->nd_repstat) {
3097 		struct sockaddr_in *saddr;
3098 		struct sockaddr_in6 *saddr6;
3099 
3100 		saddr = NFSSOCKADDR(nd->nd_nam, struct sockaddr_in *);
3101 		saddr6 = NFSSOCKADDR(nd->nd_nam, struct sockaddr_in6 *);
3102 		if (!(nd->nd_flag & ND_NFSV4) &&
3103 		    ((saddr->sin_family == AF_INET &&
3104 		      ntohs(saddr->sin_port) >= IPPORT_RESERVED) ||
3105 		     (saddr6->sin6_family == AF_INET6 &&
3106 		      ntohs(saddr6->sin6_port) >= IPPORT_RESERVED))) {
3107 			vput(*vpp);
3108 			nd->nd_repstat = (NFSERR_AUTHERR | AUTH_TOOWEAK);
3109 		}
3110 	}
3111 #endif	/* NFS_REQRSVPORT */
3112 
3113 	/*
3114 	 * Check/setup credentials.
3115 	 */
3116 	if (!nd->nd_repstat) {
3117 		nd->nd_saveduid = nd->nd_cred->cr_uid;
3118 		nd->nd_repstat = nfsd_excred(nd, exp, credanon);
3119 		if (nd->nd_repstat)
3120 			vput(*vpp);
3121 	}
3122 	if (credanon != NULL)
3123 		crfree(credanon);
3124 	if (nd->nd_repstat) {
3125 		if (startwrite)
3126 			vn_finished_write(mp);
3127 		*vpp = NULL;
3128 		if (mpp != NULL)
3129 			*mpp = NULL;
3130 	}
3131 
3132 out:
3133 	NFSEXITCODE2(0, nd);
3134 }
3135 
3136 /*
3137  * glue for fp.
3138  */
3139 static int
3140 fp_getfvp(struct thread *p, int fd, struct file **fpp, struct vnode **vpp)
3141 {
3142 	struct filedesc *fdp;
3143 	struct file *fp;
3144 	int error = 0;
3145 
3146 	fdp = p->td_proc->p_fd;
3147 	if (fd < 0 || fd >= fdp->fd_nfiles ||
3148 	    (fp = fdp->fd_ofiles[fd].fde_file) == NULL) {
3149 		error = EBADF;
3150 		goto out;
3151 	}
3152 	*fpp = fp;
3153 
3154 out:
3155 	NFSEXITCODE(error);
3156 	return (error);
3157 }
3158 
3159 /*
3160  * Called from nfssvc() to update the exports list. Just call
3161  * vfs_export(). This has to be done, since the v4 root fake fs isn't
3162  * in the mount list.
3163  */
3164 int
3165 nfsrv_v4rootexport(void *argp, struct ucred *cred, struct thread *p)
3166 {
3167 	struct nfsex_args *nfsexargp = (struct nfsex_args *)argp;
3168 	int error = 0;
3169 	struct nameidata nd;
3170 	fhandle_t fh;
3171 
3172 	error = vfs_export(&nfsv4root_mnt, &nfsexargp->export);
3173 	if ((nfsexargp->export.ex_flags & MNT_DELEXPORT) != 0)
3174 		nfs_rootfhset = 0;
3175 	else if (error == 0) {
3176 		if (nfsexargp->fspec == NULL) {
3177 			error = EPERM;
3178 			goto out;
3179 		}
3180 		/*
3181 		 * If fspec != NULL, this is the v4root path.
3182 		 */
3183 		NDINIT(&nd, LOOKUP, FOLLOW, UIO_USERSPACE,
3184 		    nfsexargp->fspec, p);
3185 		if ((error = namei(&nd)) != 0)
3186 			goto out;
3187 		error = nfsvno_getfh(nd.ni_vp, &fh, p);
3188 		vrele(nd.ni_vp);
3189 		if (!error) {
3190 			nfs_rootfh.nfsrvfh_len = NFSX_MYFH;
3191 			NFSBCOPY((caddr_t)&fh,
3192 			    nfs_rootfh.nfsrvfh_data,
3193 			    sizeof (fhandle_t));
3194 			nfs_rootfhset = 1;
3195 		}
3196 	}
3197 
3198 out:
3199 	NFSEXITCODE(error);
3200 	return (error);
3201 }
3202 
3203 /*
3204  * This function needs to test to see if the system is near its limit
3205  * for memory allocation via malloc() or mget() and return True iff
3206  * either of these resources are near their limit.
3207  * XXX (For now, this is just a stub.)
3208  */
3209 int nfsrv_testmalloclimit = 0;
3210 int
3211 nfsrv_mallocmget_limit(void)
3212 {
3213 	static int printmesg = 0;
3214 	static int testval = 1;
3215 
3216 	if (nfsrv_testmalloclimit && (testval++ % 1000) == 0) {
3217 		if ((printmesg++ % 100) == 0)
3218 			printf("nfsd: malloc/mget near limit\n");
3219 		return (1);
3220 	}
3221 	return (0);
3222 }
3223 
3224 /*
3225  * BSD specific initialization of a mount point.
3226  */
3227 void
3228 nfsd_mntinit(void)
3229 {
3230 	static int inited = 0;
3231 
3232 	if (inited)
3233 		return;
3234 	inited = 1;
3235 	nfsv4root_mnt.mnt_flag = (MNT_RDONLY | MNT_EXPORTED);
3236 	TAILQ_INIT(&nfsv4root_mnt.mnt_nvnodelist);
3237 	TAILQ_INIT(&nfsv4root_mnt.mnt_activevnodelist);
3238 	nfsv4root_mnt.mnt_export = NULL;
3239 	TAILQ_INIT(&nfsv4root_opt);
3240 	TAILQ_INIT(&nfsv4root_newopt);
3241 	nfsv4root_mnt.mnt_opt = &nfsv4root_opt;
3242 	nfsv4root_mnt.mnt_optnew = &nfsv4root_newopt;
3243 	nfsv4root_mnt.mnt_nvnodelistsize = 0;
3244 	nfsv4root_mnt.mnt_activevnodelistsize = 0;
3245 }
3246 
3247 /*
3248  * Get a vnode for a file handle, without checking exports, etc.
3249  */
3250 struct vnode *
3251 nfsvno_getvp(fhandle_t *fhp)
3252 {
3253 	struct mount *mp;
3254 	struct vnode *vp;
3255 	int error;
3256 
3257 	mp = vfs_busyfs(&fhp->fh_fsid);
3258 	if (mp == NULL)
3259 		return (NULL);
3260 	error = VFS_FHTOVP(mp, &fhp->fh_fid, LK_EXCLUSIVE, &vp);
3261 	vfs_unbusy(mp);
3262 	if (error)
3263 		return (NULL);
3264 	return (vp);
3265 }
3266 
3267 /*
3268  * Do a local VOP_ADVLOCK().
3269  */
3270 int
3271 nfsvno_advlock(struct vnode *vp, int ftype, u_int64_t first,
3272     u_int64_t end, struct thread *td)
3273 {
3274 	int error = 0;
3275 	struct flock fl;
3276 	u_int64_t tlen;
3277 
3278 	if (nfsrv_dolocallocks == 0)
3279 		goto out;
3280 	ASSERT_VOP_UNLOCKED(vp, "nfsvno_advlock: vp locked");
3281 
3282 	fl.l_whence = SEEK_SET;
3283 	fl.l_type = ftype;
3284 	fl.l_start = (off_t)first;
3285 	if (end == NFS64BITSSET) {
3286 		fl.l_len = 0;
3287 	} else {
3288 		tlen = end - first;
3289 		fl.l_len = (off_t)tlen;
3290 	}
3291 	/*
3292 	 * For FreeBSD8, the l_pid and l_sysid must be set to the same
3293 	 * values for all calls, so that all locks will be held by the
3294 	 * nfsd server. (The nfsd server handles conflicts between the
3295 	 * various clients.)
3296 	 * Since an NFSv4 lockowner is a ClientID plus an array of up to 1024
3297 	 * bytes, so it can't be put in l_sysid.
3298 	 */
3299 	if (nfsv4_sysid == 0)
3300 		nfsv4_sysid = nlm_acquire_next_sysid();
3301 	fl.l_pid = (pid_t)0;
3302 	fl.l_sysid = (int)nfsv4_sysid;
3303 
3304 	if (ftype == F_UNLCK)
3305 		error = VOP_ADVLOCK(vp, (caddr_t)td->td_proc, F_UNLCK, &fl,
3306 		    (F_POSIX | F_REMOTE));
3307 	else
3308 		error = VOP_ADVLOCK(vp, (caddr_t)td->td_proc, F_SETLK, &fl,
3309 		    (F_POSIX | F_REMOTE));
3310 
3311 out:
3312 	NFSEXITCODE(error);
3313 	return (error);
3314 }
3315 
3316 /*
3317  * Check the nfsv4 root exports.
3318  */
3319 int
3320 nfsvno_v4rootexport(struct nfsrv_descript *nd)
3321 {
3322 	struct ucred *credanon;
3323 	int exflags, error = 0, numsecflavor, *secflavors, i;
3324 
3325 	error = vfs_stdcheckexp(&nfsv4root_mnt, nd->nd_nam, &exflags,
3326 	    &credanon, &numsecflavor, &secflavors);
3327 	if (error) {
3328 		error = NFSERR_PROGUNAVAIL;
3329 		goto out;
3330 	}
3331 	if (credanon != NULL)
3332 		crfree(credanon);
3333 	for (i = 0; i < numsecflavor; i++) {
3334 		if (secflavors[i] == AUTH_SYS)
3335 			nd->nd_flag |= ND_EXAUTHSYS;
3336 		else if (secflavors[i] == RPCSEC_GSS_KRB5)
3337 			nd->nd_flag |= ND_EXGSS;
3338 		else if (secflavors[i] == RPCSEC_GSS_KRB5I)
3339 			nd->nd_flag |= ND_EXGSSINTEGRITY;
3340 		else if (secflavors[i] == RPCSEC_GSS_KRB5P)
3341 			nd->nd_flag |= ND_EXGSSPRIVACY;
3342 	}
3343 
3344 out:
3345 	NFSEXITCODE(error);
3346 	return (error);
3347 }
3348 
3349 /*
3350  * Nfs server pseudo system call for the nfsd's
3351  */
3352 /*
3353  * MPSAFE
3354  */
3355 static int
3356 nfssvc_nfsd(struct thread *td, struct nfssvc_args *uap)
3357 {
3358 	struct file *fp;
3359 	struct nfsd_addsock_args sockarg;
3360 	struct nfsd_nfsd_args nfsdarg;
3361 	struct nfsd_nfsd_oargs onfsdarg;
3362 	struct nfsd_pnfsd_args pnfsdarg;
3363 	struct vnode *vp, *nvp, *curdvp;
3364 	struct pnfsdsfile *pf;
3365 	struct nfsdevice *ds, *fds;
3366 	cap_rights_t rights;
3367 	int buflen, error, ret;
3368 	char *buf, *cp, *cp2, *cp3;
3369 	char fname[PNFS_FILENAME_LEN + 1];
3370 
3371 	if (uap->flag & NFSSVC_NFSDADDSOCK) {
3372 		error = copyin(uap->argp, (caddr_t)&sockarg, sizeof (sockarg));
3373 		if (error)
3374 			goto out;
3375 		/*
3376 		 * Since we don't know what rights might be required,
3377 		 * pretend that we need them all. It is better to be too
3378 		 * careful than too reckless.
3379 		 */
3380 		error = fget(td, sockarg.sock,
3381 		    cap_rights_init(&rights, CAP_SOCK_SERVER), &fp);
3382 		if (error != 0)
3383 			goto out;
3384 		if (fp->f_type != DTYPE_SOCKET) {
3385 			fdrop(fp, td);
3386 			error = EPERM;
3387 			goto out;
3388 		}
3389 		error = nfsrvd_addsock(fp);
3390 		fdrop(fp, td);
3391 	} else if (uap->flag & NFSSVC_NFSDNFSD) {
3392 		if (uap->argp == NULL) {
3393 			error = EINVAL;
3394 			goto out;
3395 		}
3396 		if ((uap->flag & NFSSVC_NEWSTRUCT) == 0) {
3397 			error = copyin(uap->argp, &onfsdarg, sizeof(onfsdarg));
3398 			if (error == 0) {
3399 				nfsdarg.principal = onfsdarg.principal;
3400 				nfsdarg.minthreads = onfsdarg.minthreads;
3401 				nfsdarg.maxthreads = onfsdarg.maxthreads;
3402 				nfsdarg.version = 1;
3403 				nfsdarg.addr = NULL;
3404 				nfsdarg.addrlen = 0;
3405 				nfsdarg.dnshost = NULL;
3406 				nfsdarg.dnshostlen = 0;
3407 				nfsdarg.dspath = NULL;
3408 				nfsdarg.dspathlen = 0;
3409 				nfsdarg.mdspath = NULL;
3410 				nfsdarg.mdspathlen = 0;
3411 				nfsdarg.mirrorcnt = 1;
3412 			}
3413 		} else
3414 			error = copyin(uap->argp, &nfsdarg, sizeof(nfsdarg));
3415 		if (error)
3416 			goto out;
3417 		if (nfsdarg.addrlen > 0 && nfsdarg.addrlen < 10000 &&
3418 		    nfsdarg.dnshostlen > 0 && nfsdarg.dnshostlen < 10000 &&
3419 		    nfsdarg.dspathlen > 0 && nfsdarg.dspathlen < 10000 &&
3420 		    nfsdarg.mdspathlen > 0 && nfsdarg.mdspathlen < 10000 &&
3421 		    nfsdarg.mirrorcnt >= 1 &&
3422 		    nfsdarg.mirrorcnt <= NFSDEV_MAXMIRRORS &&
3423 		    nfsdarg.addr != NULL && nfsdarg.dnshost != NULL &&
3424 		    nfsdarg.dspath != NULL && nfsdarg.mdspath != NULL) {
3425 			NFSD_DEBUG(1, "addrlen=%d dspathlen=%d dnslen=%d"
3426 			    " mdspathlen=%d mirrorcnt=%d\n", nfsdarg.addrlen,
3427 			    nfsdarg.dspathlen, nfsdarg.dnshostlen,
3428 			    nfsdarg.mdspathlen, nfsdarg.mirrorcnt);
3429 			cp = malloc(nfsdarg.addrlen + 1, M_TEMP, M_WAITOK);
3430 			error = copyin(nfsdarg.addr, cp, nfsdarg.addrlen);
3431 			if (error != 0) {
3432 				free(cp, M_TEMP);
3433 				goto out;
3434 			}
3435 			cp[nfsdarg.addrlen] = '\0';	/* Ensure nul term. */
3436 			nfsdarg.addr = cp;
3437 			cp = malloc(nfsdarg.dnshostlen + 1, M_TEMP, M_WAITOK);
3438 			error = copyin(nfsdarg.dnshost, cp, nfsdarg.dnshostlen);
3439 			if (error != 0) {
3440 				free(nfsdarg.addr, M_TEMP);
3441 				free(cp, M_TEMP);
3442 				goto out;
3443 			}
3444 			cp[nfsdarg.dnshostlen] = '\0';	/* Ensure nul term. */
3445 			nfsdarg.dnshost = cp;
3446 			cp = malloc(nfsdarg.dspathlen + 1, M_TEMP, M_WAITOK);
3447 			error = copyin(nfsdarg.dspath, cp, nfsdarg.dspathlen);
3448 			if (error != 0) {
3449 				free(nfsdarg.addr, M_TEMP);
3450 				free(nfsdarg.dnshost, M_TEMP);
3451 				free(cp, M_TEMP);
3452 				goto out;
3453 			}
3454 			cp[nfsdarg.dspathlen] = '\0';	/* Ensure nul term. */
3455 			nfsdarg.dspath = cp;
3456 			cp = malloc(nfsdarg.mdspathlen + 1, M_TEMP, M_WAITOK);
3457 			error = copyin(nfsdarg.mdspath, cp, nfsdarg.mdspathlen);
3458 			if (error != 0) {
3459 				free(nfsdarg.addr, M_TEMP);
3460 				free(nfsdarg.dnshost, M_TEMP);
3461 				free(nfsdarg.dspath, M_TEMP);
3462 				free(cp, M_TEMP);
3463 				goto out;
3464 			}
3465 			cp[nfsdarg.mdspathlen] = '\0';	/* Ensure nul term. */
3466 			nfsdarg.mdspath = cp;
3467 		} else {
3468 			nfsdarg.addr = NULL;
3469 			nfsdarg.addrlen = 0;
3470 			nfsdarg.dnshost = NULL;
3471 			nfsdarg.dnshostlen = 0;
3472 			nfsdarg.dspath = NULL;
3473 			nfsdarg.dspathlen = 0;
3474 			nfsdarg.mdspath = NULL;
3475 			nfsdarg.mdspathlen = 0;
3476 			nfsdarg.mirrorcnt = 1;
3477 		}
3478 		error = nfsrvd_nfsd(td, &nfsdarg);
3479 		free(nfsdarg.addr, M_TEMP);
3480 		free(nfsdarg.dnshost, M_TEMP);
3481 		free(nfsdarg.dspath, M_TEMP);
3482 		free(nfsdarg.mdspath, M_TEMP);
3483 	} else if (uap->flag & NFSSVC_PNFSDS) {
3484 		error = copyin(uap->argp, &pnfsdarg, sizeof(pnfsdarg));
3485 		if (error == 0 && (pnfsdarg.op == PNFSDOP_DELDSSERVER ||
3486 		    pnfsdarg.op == PNFSDOP_FORCEDELDS)) {
3487 			cp = malloc(PATH_MAX + 1, M_TEMP, M_WAITOK);
3488 			error = copyinstr(pnfsdarg.dspath, cp, PATH_MAX + 1,
3489 			    NULL);
3490 			if (error == 0)
3491 				error = nfsrv_deldsserver(pnfsdarg.op, cp, td);
3492 			free(cp, M_TEMP);
3493 		} else if (error == 0 && pnfsdarg.op == PNFSDOP_COPYMR) {
3494 			cp = malloc(PATH_MAX + 1, M_TEMP, M_WAITOK);
3495 			buflen = sizeof(*pf) * NFSDEV_MAXMIRRORS;
3496 			buf = malloc(buflen, M_TEMP, M_WAITOK);
3497 			error = copyinstr(pnfsdarg.mdspath, cp, PATH_MAX + 1,
3498 			    NULL);
3499 			NFSD_DEBUG(4, "pnfsdcopymr cp mdspath=%d\n", error);
3500 			if (error == 0 && pnfsdarg.dspath != NULL) {
3501 				cp2 = malloc(PATH_MAX + 1, M_TEMP, M_WAITOK);
3502 				error = copyinstr(pnfsdarg.dspath, cp2,
3503 				    PATH_MAX + 1, NULL);
3504 				NFSD_DEBUG(4, "pnfsdcopymr cp dspath=%d\n",
3505 				    error);
3506 			} else
3507 				cp2 = NULL;
3508 			if (error == 0 && pnfsdarg.curdspath != NULL) {
3509 				cp3 = malloc(PATH_MAX + 1, M_TEMP, M_WAITOK);
3510 				error = copyinstr(pnfsdarg.curdspath, cp3,
3511 				    PATH_MAX + 1, NULL);
3512 				NFSD_DEBUG(4, "pnfsdcopymr cp curdspath=%d\n",
3513 				    error);
3514 			} else
3515 				cp3 = NULL;
3516 			curdvp = NULL;
3517 			fds = NULL;
3518 			if (error == 0)
3519 				error = nfsrv_mdscopymr(cp, cp2, cp3, buf,
3520 				    &buflen, fname, td, &vp, &nvp, &pf, &ds,
3521 				    &fds);
3522 			NFSD_DEBUG(4, "nfsrv_mdscopymr=%d\n", error);
3523 			if (error == 0) {
3524 				if (pf->dsf_dir >= nfsrv_dsdirsize) {
3525 					printf("copymr: dsdir out of range\n");
3526 					pf->dsf_dir = 0;
3527 				}
3528 				NFSD_DEBUG(4, "copymr: buflen=%d\n", buflen);
3529 				error = nfsrv_copymr(vp, nvp,
3530 				    ds->nfsdev_dsdir[pf->dsf_dir], ds, pf,
3531 				    (struct pnfsdsfile *)buf,
3532 				    buflen / sizeof(*pf), td->td_ucred, td);
3533 				vput(vp);
3534 				vput(nvp);
3535 				if (fds != NULL && error == 0) {
3536 					curdvp = fds->nfsdev_dsdir[pf->dsf_dir];
3537 					ret = vn_lock(curdvp, LK_EXCLUSIVE);
3538 					if (ret == 0) {
3539 						nfsrv_dsremove(curdvp, fname,
3540 						    td->td_ucred, td);
3541 						NFSVOPUNLOCK(curdvp, 0);
3542 					}
3543 				}
3544 				NFSD_DEBUG(4, "nfsrv_copymr=%d\n", error);
3545 			}
3546 			free(cp, M_TEMP);
3547 			free(cp2, M_TEMP);
3548 			free(cp3, M_TEMP);
3549 			free(buf, M_TEMP);
3550 		}
3551 	} else {
3552 		error = nfssvc_srvcall(td, uap, td->td_ucred);
3553 	}
3554 
3555 out:
3556 	NFSEXITCODE(error);
3557 	return (error);
3558 }
3559 
3560 static int
3561 nfssvc_srvcall(struct thread *p, struct nfssvc_args *uap, struct ucred *cred)
3562 {
3563 	struct nfsex_args export;
3564 	struct file *fp = NULL;
3565 	int stablefd, len;
3566 	struct nfsd_clid adminrevoke;
3567 	struct nfsd_dumplist dumplist;
3568 	struct nfsd_dumpclients *dumpclients;
3569 	struct nfsd_dumplocklist dumplocklist;
3570 	struct nfsd_dumplocks *dumplocks;
3571 	struct nameidata nd;
3572 	vnode_t vp;
3573 	int error = EINVAL, igotlock;
3574 	struct proc *procp;
3575 	static int suspend_nfsd = 0;
3576 
3577 	if (uap->flag & NFSSVC_PUBLICFH) {
3578 		NFSBZERO((caddr_t)&nfs_pubfh.nfsrvfh_data,
3579 		    sizeof (fhandle_t));
3580 		error = copyin(uap->argp,
3581 		    &nfs_pubfh.nfsrvfh_data, sizeof (fhandle_t));
3582 		if (!error)
3583 			nfs_pubfhset = 1;
3584 	} else if (uap->flag & NFSSVC_V4ROOTEXPORT) {
3585 		error = copyin(uap->argp,(caddr_t)&export,
3586 		    sizeof (struct nfsex_args));
3587 		if (!error)
3588 			error = nfsrv_v4rootexport(&export, cred, p);
3589 	} else if (uap->flag & NFSSVC_NOPUBLICFH) {
3590 		nfs_pubfhset = 0;
3591 		error = 0;
3592 	} else if (uap->flag & NFSSVC_STABLERESTART) {
3593 		error = copyin(uap->argp, (caddr_t)&stablefd,
3594 		    sizeof (int));
3595 		if (!error)
3596 			error = fp_getfvp(p, stablefd, &fp, &vp);
3597 		if (!error && (NFSFPFLAG(fp) & (FREAD | FWRITE)) != (FREAD | FWRITE))
3598 			error = EBADF;
3599 		if (!error && newnfs_numnfsd != 0)
3600 			error = EPERM;
3601 		if (!error) {
3602 			nfsrv_stablefirst.nsf_fp = fp;
3603 			nfsrv_setupstable(p);
3604 		}
3605 	} else if (uap->flag & NFSSVC_ADMINREVOKE) {
3606 		error = copyin(uap->argp, (caddr_t)&adminrevoke,
3607 		    sizeof (struct nfsd_clid));
3608 		if (!error)
3609 			error = nfsrv_adminrevoke(&adminrevoke, p);
3610 	} else if (uap->flag & NFSSVC_DUMPCLIENTS) {
3611 		error = copyin(uap->argp, (caddr_t)&dumplist,
3612 		    sizeof (struct nfsd_dumplist));
3613 		if (!error && (dumplist.ndl_size < 1 ||
3614 			dumplist.ndl_size > NFSRV_MAXDUMPLIST))
3615 			error = EPERM;
3616 		if (!error) {
3617 		    len = sizeof (struct nfsd_dumpclients) * dumplist.ndl_size;
3618 		    dumpclients = (struct nfsd_dumpclients *)malloc(len,
3619 			M_TEMP, M_WAITOK);
3620 		    nfsrv_dumpclients(dumpclients, dumplist.ndl_size);
3621 		    error = copyout(dumpclients,
3622 			CAST_USER_ADDR_T(dumplist.ndl_list), len);
3623 		    free(dumpclients, M_TEMP);
3624 		}
3625 	} else if (uap->flag & NFSSVC_DUMPLOCKS) {
3626 		error = copyin(uap->argp, (caddr_t)&dumplocklist,
3627 		    sizeof (struct nfsd_dumplocklist));
3628 		if (!error && (dumplocklist.ndllck_size < 1 ||
3629 			dumplocklist.ndllck_size > NFSRV_MAXDUMPLIST))
3630 			error = EPERM;
3631 		if (!error)
3632 			error = nfsrv_lookupfilename(&nd,
3633 				dumplocklist.ndllck_fname, p);
3634 		if (!error) {
3635 			len = sizeof (struct nfsd_dumplocks) *
3636 				dumplocklist.ndllck_size;
3637 			dumplocks = (struct nfsd_dumplocks *)malloc(len,
3638 				M_TEMP, M_WAITOK);
3639 			nfsrv_dumplocks(nd.ni_vp, dumplocks,
3640 			    dumplocklist.ndllck_size, p);
3641 			vput(nd.ni_vp);
3642 			error = copyout(dumplocks,
3643 			    CAST_USER_ADDR_T(dumplocklist.ndllck_list), len);
3644 			free(dumplocks, M_TEMP);
3645 		}
3646 	} else if (uap->flag & NFSSVC_BACKUPSTABLE) {
3647 		procp = p->td_proc;
3648 		PROC_LOCK(procp);
3649 		nfsd_master_pid = procp->p_pid;
3650 		bcopy(procp->p_comm, nfsd_master_comm, MAXCOMLEN + 1);
3651 		nfsd_master_start = procp->p_stats->p_start;
3652 		nfsd_master_proc = procp;
3653 		PROC_UNLOCK(procp);
3654 	} else if ((uap->flag & NFSSVC_SUSPENDNFSD) != 0) {
3655 		NFSLOCKV4ROOTMUTEX();
3656 		if (suspend_nfsd == 0) {
3657 			/* Lock out all nfsd threads */
3658 			do {
3659 				igotlock = nfsv4_lock(&nfsd_suspend_lock, 1,
3660 				    NULL, NFSV4ROOTLOCKMUTEXPTR, NULL);
3661 			} while (igotlock == 0 && suspend_nfsd == 0);
3662 			suspend_nfsd = 1;
3663 		}
3664 		NFSUNLOCKV4ROOTMUTEX();
3665 		error = 0;
3666 	} else if ((uap->flag & NFSSVC_RESUMENFSD) != 0) {
3667 		NFSLOCKV4ROOTMUTEX();
3668 		if (suspend_nfsd != 0) {
3669 			nfsv4_unlock(&nfsd_suspend_lock, 0);
3670 			suspend_nfsd = 0;
3671 		}
3672 		NFSUNLOCKV4ROOTMUTEX();
3673 		error = 0;
3674 	}
3675 
3676 	NFSEXITCODE(error);
3677 	return (error);
3678 }
3679 
3680 /*
3681  * Check exports.
3682  * Returns 0 if ok, 1 otherwise.
3683  */
3684 int
3685 nfsvno_testexp(struct nfsrv_descript *nd, struct nfsexstuff *exp)
3686 {
3687 	int i;
3688 
3689 	/*
3690 	 * This seems odd, but allow the case where the security flavor
3691 	 * list is empty. This happens when NFSv4 is traversing non-exported
3692 	 * file systems. Exported file systems should always have a non-empty
3693 	 * security flavor list.
3694 	 */
3695 	if (exp->nes_numsecflavor == 0)
3696 		return (0);
3697 
3698 	for (i = 0; i < exp->nes_numsecflavor; i++) {
3699 		/*
3700 		 * The tests for privacy and integrity must be first,
3701 		 * since ND_GSS is set for everything but AUTH_SYS.
3702 		 */
3703 		if (exp->nes_secflavors[i] == RPCSEC_GSS_KRB5P &&
3704 		    (nd->nd_flag & ND_GSSPRIVACY))
3705 			return (0);
3706 		if (exp->nes_secflavors[i] == RPCSEC_GSS_KRB5I &&
3707 		    (nd->nd_flag & ND_GSSINTEGRITY))
3708 			return (0);
3709 		if (exp->nes_secflavors[i] == RPCSEC_GSS_KRB5 &&
3710 		    (nd->nd_flag & ND_GSS))
3711 			return (0);
3712 		if (exp->nes_secflavors[i] == AUTH_SYS &&
3713 		    (nd->nd_flag & ND_GSS) == 0)
3714 			return (0);
3715 	}
3716 	return (1);
3717 }
3718 
3719 /*
3720  * Calculate a hash value for the fid in a file handle.
3721  */
3722 uint32_t
3723 nfsrv_hashfh(fhandle_t *fhp)
3724 {
3725 	uint32_t hashval;
3726 
3727 	hashval = hash32_buf(&fhp->fh_fid, sizeof(struct fid), 0);
3728 	return (hashval);
3729 }
3730 
3731 /*
3732  * Calculate a hash value for the sessionid.
3733  */
3734 uint32_t
3735 nfsrv_hashsessionid(uint8_t *sessionid)
3736 {
3737 	uint32_t hashval;
3738 
3739 	hashval = hash32_buf(sessionid, NFSX_V4SESSIONID, 0);
3740 	return (hashval);
3741 }
3742 
3743 /*
3744  * Signal the userland master nfsd to backup the stable restart file.
3745  */
3746 void
3747 nfsrv_backupstable(void)
3748 {
3749 	struct proc *procp;
3750 
3751 	if (nfsd_master_proc != NULL) {
3752 		procp = pfind(nfsd_master_pid);
3753 		/* Try to make sure it is the correct process. */
3754 		if (procp == nfsd_master_proc &&
3755 		    procp->p_stats->p_start.tv_sec ==
3756 		    nfsd_master_start.tv_sec &&
3757 		    procp->p_stats->p_start.tv_usec ==
3758 		    nfsd_master_start.tv_usec &&
3759 		    strcmp(procp->p_comm, nfsd_master_comm) == 0)
3760 			kern_psignal(procp, SIGUSR2);
3761 		else
3762 			nfsd_master_proc = NULL;
3763 
3764 		if (procp != NULL)
3765 			PROC_UNLOCK(procp);
3766 	}
3767 }
3768 
3769 /*
3770  * Create a DS data file for nfsrv_pnfscreate(). Called for each mirror.
3771  * The arguments are in a structure, so that they can be passed through
3772  * taskqueue for a kernel process to execute this function.
3773  */
3774 struct nfsrvdscreate {
3775 	int			done;
3776 	int			inprog;
3777 	struct task		tsk;
3778 	struct ucred		*tcred;
3779 	struct vnode		*dvp;
3780 	NFSPROC_T		*p;
3781 	struct pnfsdsfile	*pf;
3782 	int			err;
3783 	fhandle_t		fh;
3784 	struct vattr		va;
3785 	struct vattr		createva;
3786 };
3787 
3788 int
3789 nfsrv_dscreate(struct vnode *dvp, struct vattr *vap, struct vattr *nvap,
3790     fhandle_t *fhp, struct pnfsdsfile *pf, struct pnfsdsattr *dsa,
3791     char *fnamep, struct ucred *tcred, NFSPROC_T *p, struct vnode **nvpp)
3792 {
3793 	struct vnode *nvp;
3794 	struct nameidata named;
3795 	struct vattr va;
3796 	char *bufp;
3797 	u_long *hashp;
3798 	struct nfsnode *np;
3799 	struct nfsmount *nmp;
3800 	int error;
3801 
3802 	NFSNAMEICNDSET(&named.ni_cnd, tcred, CREATE,
3803 	    LOCKPARENT | LOCKLEAF | SAVESTART | NOCACHE);
3804 	nfsvno_setpathbuf(&named, &bufp, &hashp);
3805 	named.ni_cnd.cn_lkflags = LK_EXCLUSIVE;
3806 	named.ni_cnd.cn_thread = p;
3807 	named.ni_cnd.cn_nameptr = bufp;
3808 	if (fnamep != NULL) {
3809 		strlcpy(bufp, fnamep, PNFS_FILENAME_LEN + 1);
3810 		named.ni_cnd.cn_namelen = strlen(bufp);
3811 	} else
3812 		named.ni_cnd.cn_namelen = nfsrv_putfhname(fhp, bufp);
3813 	NFSD_DEBUG(4, "nfsrv_dscreate: dvp=%p fname=%s\n", dvp, bufp);
3814 
3815 	/* Create the date file in the DS mount. */
3816 	error = NFSVOPLOCK(dvp, LK_EXCLUSIVE);
3817 	if (error == 0) {
3818 		error = VOP_CREATE(dvp, &nvp, &named.ni_cnd, vap);
3819 		NFSVOPUNLOCK(dvp, 0);
3820 		if (error == 0) {
3821 			/* Set the ownership of the file. */
3822 			error = VOP_SETATTR(nvp, nvap, tcred);
3823 			NFSD_DEBUG(4, "nfsrv_dscreate:"
3824 			    " setattr-uid=%d\n", error);
3825 			if (error != 0)
3826 				vput(nvp);
3827 		}
3828 		if (error != 0)
3829 			printf("pNFS: pnfscreate failed=%d\n", error);
3830 	} else
3831 		printf("pNFS: pnfscreate vnlock=%d\n", error);
3832 	if (error == 0) {
3833 		np = VTONFS(nvp);
3834 		nmp = VFSTONFS(nvp->v_mount);
3835 		if (strcmp(nvp->v_mount->mnt_vfc->vfc_name, "nfs")
3836 		    != 0 || nmp->nm_nam->sa_len > sizeof(
3837 		    struct sockaddr_in6) ||
3838 		    np->n_fhp->nfh_len != NFSX_MYFH) {
3839 			printf("Bad DS file: fstype=%s salen=%d"
3840 			    " fhlen=%d\n",
3841 			    nvp->v_mount->mnt_vfc->vfc_name,
3842 			    nmp->nm_nam->sa_len, np->n_fhp->nfh_len);
3843 			error = ENOENT;
3844 		}
3845 
3846 		/* Set extattrs for the DS on the MDS file. */
3847 		if (error == 0) {
3848 			if (dsa != NULL) {
3849 				error = VOP_GETATTR(nvp, &va, tcred);
3850 				if (error == 0) {
3851 					dsa->dsa_filerev = va.va_filerev;
3852 					dsa->dsa_size = va.va_size;
3853 					dsa->dsa_atime = va.va_atime;
3854 					dsa->dsa_mtime = va.va_mtime;
3855 				}
3856 			}
3857 			if (error == 0) {
3858 				NFSBCOPY(np->n_fhp->nfh_fh, &pf->dsf_fh,
3859 				    NFSX_MYFH);
3860 				NFSBCOPY(nmp->nm_nam, &pf->dsf_sin,
3861 				    nmp->nm_nam->sa_len);
3862 				NFSBCOPY(named.ni_cnd.cn_nameptr,
3863 				    pf->dsf_filename,
3864 				    sizeof(pf->dsf_filename));
3865 			}
3866 		} else
3867 			printf("pNFS: pnfscreate can't get DS"
3868 			    " attr=%d\n", error);
3869 		if (nvpp != NULL && error == 0)
3870 			*nvpp = nvp;
3871 		else
3872 			vput(nvp);
3873 	}
3874 	nfsvno_relpathbuf(&named);
3875 	return (error);
3876 }
3877 
3878 /*
3879  * Start up the thread that will execute nfsrv_dscreate().
3880  */
3881 static void
3882 start_dscreate(void *arg, int pending)
3883 {
3884 	struct nfsrvdscreate *dsc;
3885 
3886 	dsc = (struct nfsrvdscreate *)arg;
3887 	dsc->err = nfsrv_dscreate(dsc->dvp, &dsc->createva, &dsc->va, &dsc->fh,
3888 	    dsc->pf, NULL, NULL, dsc->tcred, dsc->p, NULL);
3889 	dsc->done = 1;
3890 	NFSD_DEBUG(4, "start_dscreate: err=%d\n", dsc->err);
3891 }
3892 
3893 /*
3894  * Create a pNFS data file on the Data Server(s).
3895  */
3896 static void
3897 nfsrv_pnfscreate(struct vnode *vp, struct vattr *vap, struct ucred *cred,
3898     NFSPROC_T *p)
3899 {
3900 	struct nfsrvdscreate *dsc, *tdsc;
3901 	struct nfsdevice *ds, *tds, *fds;
3902 	struct mount *mp;
3903 	struct pnfsdsfile *pf, *tpf;
3904 	struct pnfsdsattr dsattr;
3905 	struct vattr va;
3906 	struct vnode *dvp[NFSDEV_MAXMIRRORS];
3907 	struct nfsmount *nmp;
3908 	fhandle_t fh;
3909 	uid_t vauid;
3910 	gid_t vagid;
3911 	u_short vamode;
3912 	struct ucred *tcred;
3913 	int dsdir[NFSDEV_MAXMIRRORS], error, i, mirrorcnt, ret;
3914 	int failpos, timo;
3915 
3916 	/* Get a DS server directory in a round-robin order. */
3917 	mirrorcnt = 1;
3918 	mp = vp->v_mount;
3919 	ds = fds = NULL;
3920 	NFSDDSLOCK();
3921 	/*
3922 	 * Search for the first entry that handles this MDS fs, but use the
3923 	 * first entry for all MDS fs's otherwise.
3924 	 */
3925 	TAILQ_FOREACH(tds, &nfsrv_devidhead, nfsdev_list) {
3926 		if (tds->nfsdev_nmp != NULL) {
3927 			if (tds->nfsdev_mdsisset == 0 && ds == NULL)
3928 				ds = tds;
3929 			else if (tds->nfsdev_mdsisset != 0 &&
3930 			    mp->mnt_stat.f_fsid.val[0] ==
3931 			    tds->nfsdev_mdsfsid.val[0] &&
3932 			    mp->mnt_stat.f_fsid.val[1] ==
3933 			    tds->nfsdev_mdsfsid.val[1]) {
3934 				ds = fds = tds;
3935 				break;
3936 			}
3937 		}
3938 	}
3939 	if (ds == NULL) {
3940 		NFSDDSUNLOCK();
3941 		NFSD_DEBUG(4, "nfsrv_pnfscreate: no srv\n");
3942 		return;
3943 	}
3944 	i = dsdir[0] = ds->nfsdev_nextdir;
3945 	ds->nfsdev_nextdir = (ds->nfsdev_nextdir + 1) % nfsrv_dsdirsize;
3946 	dvp[0] = ds->nfsdev_dsdir[i];
3947 	tds = TAILQ_NEXT(ds, nfsdev_list);
3948 	if (nfsrv_maxpnfsmirror > 1 && tds != NULL) {
3949 		TAILQ_FOREACH_FROM(tds, &nfsrv_devidhead, nfsdev_list) {
3950 			if (tds->nfsdev_nmp != NULL &&
3951 			    ((tds->nfsdev_mdsisset == 0 && fds == NULL) ||
3952 			     (tds->nfsdev_mdsisset != 0 && fds != NULL &&
3953 			      mp->mnt_stat.f_fsid.val[0] ==
3954 			      tds->nfsdev_mdsfsid.val[0] &&
3955 			      mp->mnt_stat.f_fsid.val[1] ==
3956 			      tds->nfsdev_mdsfsid.val[1]))) {
3957 				dsdir[mirrorcnt] = i;
3958 				dvp[mirrorcnt] = tds->nfsdev_dsdir[i];
3959 				mirrorcnt++;
3960 				if (mirrorcnt >= nfsrv_maxpnfsmirror)
3961 					break;
3962 			}
3963 		}
3964 	}
3965 	/* Put at end of list to implement round-robin usage. */
3966 	TAILQ_REMOVE(&nfsrv_devidhead, ds, nfsdev_list);
3967 	TAILQ_INSERT_TAIL(&nfsrv_devidhead, ds, nfsdev_list);
3968 	NFSDDSUNLOCK();
3969 	dsc = NULL;
3970 	if (mirrorcnt > 1)
3971 		tdsc = dsc = malloc(sizeof(*dsc) * (mirrorcnt - 1), M_TEMP,
3972 		    M_WAITOK | M_ZERO);
3973 	tpf = pf = malloc(sizeof(*pf) * nfsrv_maxpnfsmirror, M_TEMP, M_WAITOK |
3974 	    M_ZERO);
3975 
3976 	error = nfsvno_getfh(vp, &fh, p);
3977 	if (error == 0)
3978 		error = VOP_GETATTR(vp, &va, cred);
3979 	if (error == 0) {
3980 		/* Set the attributes for "vp" to Setattr the DS vp. */
3981 		vauid = va.va_uid;
3982 		vagid = va.va_gid;
3983 		vamode = va.va_mode;
3984 		VATTR_NULL(&va);
3985 		va.va_uid = vauid;
3986 		va.va_gid = vagid;
3987 		va.va_mode = vamode;
3988 		va.va_size = 0;
3989 	} else
3990 		printf("pNFS: pnfscreate getfh+attr=%d\n", error);
3991 
3992 	NFSD_DEBUG(4, "nfsrv_pnfscreate: cruid=%d crgid=%d\n", cred->cr_uid,
3993 	    cred->cr_gid);
3994 	/* Make data file name based on FH. */
3995 	tcred = newnfs_getcred();
3996 
3997 	/*
3998 	 * Create the file on each DS mirror, using kernel process(es) for the
3999 	 * additional mirrors.
4000 	 */
4001 	failpos = -1;
4002 	for (i = 0; i < mirrorcnt - 1 && error == 0; i++, tpf++, tdsc++) {
4003 		tpf->dsf_dir = dsdir[i];
4004 		tdsc->tcred = tcred;
4005 		tdsc->p = p;
4006 		tdsc->pf = tpf;
4007 		tdsc->createva = *vap;
4008 		NFSBCOPY(&fh, &tdsc->fh, sizeof(fh));
4009 		tdsc->va = va;
4010 		tdsc->dvp = dvp[i];
4011 		tdsc->done = 0;
4012 		tdsc->inprog = 0;
4013 		tdsc->err = 0;
4014 		ret = EIO;
4015 		if (nfs_pnfsiothreads != 0) {
4016 			ret = nfs_pnfsio(start_dscreate, tdsc);
4017 			NFSD_DEBUG(4, "nfsrv_pnfscreate: nfs_pnfsio=%d\n", ret);
4018 		}
4019 		if (ret != 0) {
4020 			ret = nfsrv_dscreate(dvp[i], vap, &va, &fh, tpf, NULL,
4021 			    NULL, tcred, p, NULL);
4022 			if (ret != 0) {
4023 				KASSERT(error == 0, ("nfsrv_dscreate err=%d",
4024 				    error));
4025 				if (failpos == -1 && nfsds_failerr(ret))
4026 					failpos = i;
4027 				else
4028 					error = ret;
4029 			}
4030 		}
4031 	}
4032 	if (error == 0) {
4033 		tpf->dsf_dir = dsdir[mirrorcnt - 1];
4034 		error = nfsrv_dscreate(dvp[mirrorcnt - 1], vap, &va, &fh, tpf,
4035 		    &dsattr, NULL, tcred, p, NULL);
4036 		if (failpos == -1 && mirrorcnt > 1 && nfsds_failerr(error)) {
4037 			failpos = mirrorcnt - 1;
4038 			error = 0;
4039 		}
4040 	}
4041 	timo = hz / 50;		/* Wait for 20msec. */
4042 	if (timo < 1)
4043 		timo = 1;
4044 	/* Wait for kernel task(s) to complete. */
4045 	for (tdsc = dsc, i = 0; i < mirrorcnt - 1; i++, tdsc++) {
4046 		while (tdsc->inprog != 0 && tdsc->done == 0)
4047 			tsleep(&tdsc->tsk, PVFS, "srvdcr", timo);
4048 		if (tdsc->err != 0) {
4049 			if (failpos == -1 && nfsds_failerr(tdsc->err))
4050 				failpos = i;
4051 			else if (error == 0)
4052 				error = tdsc->err;
4053 		}
4054 	}
4055 
4056 	/*
4057 	 * If failpos has been set, that mirror has failed, so it needs
4058 	 * to be disabled.
4059 	 */
4060 	if (failpos >= 0) {
4061 		nmp = VFSTONFS(dvp[failpos]->v_mount);
4062 		NFSLOCKMNT(nmp);
4063 		if ((nmp->nm_privflag & (NFSMNTP_FORCEDISM |
4064 		     NFSMNTP_CANCELRPCS)) == 0) {
4065 			nmp->nm_privflag |= NFSMNTP_CANCELRPCS;
4066 			NFSUNLOCKMNT(nmp);
4067 			ds = nfsrv_deldsnmp(PNFSDOP_DELDSSERVER, nmp, p);
4068 			NFSD_DEBUG(4, "dscreatfail fail=%d ds=%p\n", failpos,
4069 			    ds);
4070 			if (ds != NULL)
4071 				nfsrv_killrpcs(nmp);
4072 			NFSLOCKMNT(nmp);
4073 			nmp->nm_privflag &= ~NFSMNTP_CANCELRPCS;
4074 			wakeup(nmp);
4075 		}
4076 		NFSUNLOCKMNT(nmp);
4077 	}
4078 
4079 	NFSFREECRED(tcred);
4080 	if (error == 0) {
4081 		ASSERT_VOP_ELOCKED(vp, "nfsrv_pnfscreate vp");
4082 
4083 		NFSD_DEBUG(4, "nfsrv_pnfscreate: mirrorcnt=%d maxmirror=%d\n",
4084 		    mirrorcnt, nfsrv_maxpnfsmirror);
4085 		/*
4086 		 * For all mirrors that couldn't be created, fill in the
4087 		 * *pf structure, but with an IP address == 0.0.0.0.
4088 		 */
4089 		tpf = pf + mirrorcnt;
4090 		for (i = mirrorcnt; i < nfsrv_maxpnfsmirror; i++, tpf++) {
4091 			*tpf = *pf;
4092 			tpf->dsf_sin.sin_family = AF_INET;
4093 			tpf->dsf_sin.sin_len = sizeof(struct sockaddr_in);
4094 			tpf->dsf_sin.sin_addr.s_addr = 0;
4095 			tpf->dsf_sin.sin_port = 0;
4096 		}
4097 
4098 		error = vn_extattr_set(vp, IO_NODELOCKED,
4099 		    EXTATTR_NAMESPACE_SYSTEM, "pnfsd.dsfile",
4100 		    sizeof(*pf) * nfsrv_maxpnfsmirror, (char *)pf, p);
4101 		if (error == 0)
4102 			error = vn_extattr_set(vp, IO_NODELOCKED,
4103 			    EXTATTR_NAMESPACE_SYSTEM, "pnfsd.dsattr",
4104 			    sizeof(dsattr), (char *)&dsattr, p);
4105 		if (error != 0)
4106 			printf("pNFS: pnfscreate setextattr=%d\n",
4107 			    error);
4108 	} else
4109 		printf("pNFS: pnfscreate=%d\n", error);
4110 	free(pf, M_TEMP);
4111 	free(dsc, M_TEMP);
4112 }
4113 
4114 /*
4115  * Get the information needed to remove the pNFS Data Server file from the
4116  * Metadata file.  Upon success, ddvp is set non-NULL to the locked
4117  * DS directory vnode.  The caller must unlock *ddvp when done with it.
4118  */
4119 static void
4120 nfsrv_pnfsremovesetup(struct vnode *vp, NFSPROC_T *p, struct vnode **dvpp,
4121     int *mirrorcntp, char *fname, fhandle_t *fhp)
4122 {
4123 	struct vattr va;
4124 	struct ucred *tcred;
4125 	char *buf;
4126 	int buflen, error;
4127 
4128 	dvpp[0] = NULL;
4129 	/* If not an exported regular file or not a pNFS server, just return. */
4130 	if (vp->v_type != VREG || (vp->v_mount->mnt_flag & MNT_EXPORTED) == 0 ||
4131 	    nfsrv_devidcnt == 0)
4132 		return;
4133 
4134 	/* Check to see if this is the last hard link. */
4135 	tcred = newnfs_getcred();
4136 	error = VOP_GETATTR(vp, &va, tcred);
4137 	NFSFREECRED(tcred);
4138 	if (error != 0) {
4139 		printf("pNFS: nfsrv_pnfsremovesetup getattr=%d\n", error);
4140 		return;
4141 	}
4142 	if (va.va_nlink > 1)
4143 		return;
4144 
4145 	error = nfsvno_getfh(vp, fhp, p);
4146 	if (error != 0) {
4147 		printf("pNFS: nfsrv_pnfsremovesetup getfh=%d\n", error);
4148 		return;
4149 	}
4150 
4151 	buflen = 1024;
4152 	buf = malloc(buflen, M_TEMP, M_WAITOK);
4153 	/* Get the directory vnode for the DS mount and the file handle. */
4154 	error = nfsrv_dsgetsockmnt(vp, 0, buf, &buflen, mirrorcntp, p, dvpp,
4155 	    NULL, NULL, fname, NULL, NULL, NULL, NULL, NULL);
4156 	free(buf, M_TEMP);
4157 	if (error != 0)
4158 		printf("pNFS: nfsrv_pnfsremovesetup getsockmnt=%d\n", error);
4159 }
4160 
4161 /*
4162  * Remove a DS data file for nfsrv_pnfsremove(). Called for each mirror.
4163  * The arguments are in a structure, so that they can be passed through
4164  * taskqueue for a kernel process to execute this function.
4165  */
4166 struct nfsrvdsremove {
4167 	int			done;
4168 	int			inprog;
4169 	struct task		tsk;
4170 	struct ucred		*tcred;
4171 	struct vnode		*dvp;
4172 	NFSPROC_T		*p;
4173 	int			err;
4174 	char			fname[PNFS_FILENAME_LEN + 1];
4175 };
4176 
4177 static int
4178 nfsrv_dsremove(struct vnode *dvp, char *fname, struct ucred *tcred,
4179     NFSPROC_T *p)
4180 {
4181 	struct nameidata named;
4182 	struct vnode *nvp;
4183 	char *bufp;
4184 	u_long *hashp;
4185 	int error;
4186 
4187 	error = NFSVOPLOCK(dvp, LK_EXCLUSIVE);
4188 	if (error != 0)
4189 		return (error);
4190 	named.ni_cnd.cn_nameiop = DELETE;
4191 	named.ni_cnd.cn_lkflags = LK_EXCLUSIVE | LK_RETRY;
4192 	named.ni_cnd.cn_cred = tcred;
4193 	named.ni_cnd.cn_thread = p;
4194 	named.ni_cnd.cn_flags = ISLASTCN | LOCKPARENT | LOCKLEAF | SAVENAME;
4195 	nfsvno_setpathbuf(&named, &bufp, &hashp);
4196 	named.ni_cnd.cn_nameptr = bufp;
4197 	named.ni_cnd.cn_namelen = strlen(fname);
4198 	strlcpy(bufp, fname, NAME_MAX);
4199 	NFSD_DEBUG(4, "nfsrv_pnfsremove: filename=%s\n", bufp);
4200 	error = VOP_LOOKUP(dvp, &nvp, &named.ni_cnd);
4201 	NFSD_DEBUG(4, "nfsrv_pnfsremove: aft LOOKUP=%d\n", error);
4202 	if (error == 0) {
4203 		error = VOP_REMOVE(dvp, nvp, &named.ni_cnd);
4204 		vput(nvp);
4205 	}
4206 	NFSVOPUNLOCK(dvp, 0);
4207 	nfsvno_relpathbuf(&named);
4208 	if (error != 0)
4209 		printf("pNFS: nfsrv_pnfsremove failed=%d\n", error);
4210 	return (error);
4211 }
4212 
4213 /*
4214  * Start up the thread that will execute nfsrv_dsremove().
4215  */
4216 static void
4217 start_dsremove(void *arg, int pending)
4218 {
4219 	struct nfsrvdsremove *dsrm;
4220 
4221 	dsrm = (struct nfsrvdsremove *)arg;
4222 	dsrm->err = nfsrv_dsremove(dsrm->dvp, dsrm->fname, dsrm->tcred,
4223 	    dsrm->p);
4224 	dsrm->done = 1;
4225 	NFSD_DEBUG(4, "start_dsremove: err=%d\n", dsrm->err);
4226 }
4227 
4228 /*
4229  * Remove a pNFS data file from a Data Server.
4230  * nfsrv_pnfsremovesetup() must have been called before the MDS file was
4231  * removed to set up the dvp and fill in the FH.
4232  */
4233 static void
4234 nfsrv_pnfsremove(struct vnode **dvp, int mirrorcnt, char *fname, fhandle_t *fhp,
4235     NFSPROC_T *p)
4236 {
4237 	struct ucred *tcred;
4238 	struct nfsrvdsremove *dsrm, *tdsrm;
4239 	struct nfsdevice *ds;
4240 	struct nfsmount *nmp;
4241 	int failpos, i, ret, timo;
4242 
4243 	tcred = newnfs_getcred();
4244 	dsrm = NULL;
4245 	if (mirrorcnt > 1)
4246 		dsrm = malloc(sizeof(*dsrm) * mirrorcnt - 1, M_TEMP, M_WAITOK);
4247 	/*
4248 	 * Remove the file on each DS mirror, using kernel process(es) for the
4249 	 * additional mirrors.
4250 	 */
4251 	failpos = -1;
4252 	for (tdsrm = dsrm, i = 0; i < mirrorcnt - 1; i++, tdsrm++) {
4253 		tdsrm->tcred = tcred;
4254 		tdsrm->p = p;
4255 		tdsrm->dvp = dvp[i];
4256 		strlcpy(tdsrm->fname, fname, PNFS_FILENAME_LEN + 1);
4257 		tdsrm->inprog = 0;
4258 		tdsrm->done = 0;
4259 		tdsrm->err = 0;
4260 		ret = EIO;
4261 		if (nfs_pnfsiothreads != 0) {
4262 			ret = nfs_pnfsio(start_dsremove, tdsrm);
4263 			NFSD_DEBUG(4, "nfsrv_pnfsremove: nfs_pnfsio=%d\n", ret);
4264 		}
4265 		if (ret != 0) {
4266 			ret = nfsrv_dsremove(dvp[i], fname, tcred, p);
4267 			if (failpos == -1 && nfsds_failerr(ret))
4268 				failpos = i;
4269 		}
4270 	}
4271 	ret = nfsrv_dsremove(dvp[mirrorcnt - 1], fname, tcred, p);
4272 	if (failpos == -1 && mirrorcnt > 1 && nfsds_failerr(ret))
4273 		failpos = mirrorcnt - 1;
4274 	timo = hz / 50;		/* Wait for 20msec. */
4275 	if (timo < 1)
4276 		timo = 1;
4277 	/* Wait for kernel task(s) to complete. */
4278 	for (tdsrm = dsrm, i = 0; i < mirrorcnt - 1; i++, tdsrm++) {
4279 		while (tdsrm->inprog != 0 && tdsrm->done == 0)
4280 			tsleep(&tdsrm->tsk, PVFS, "srvdsrm", timo);
4281 		if (failpos == -1 && nfsds_failerr(tdsrm->err))
4282 			failpos = i;
4283 	}
4284 
4285 	/*
4286 	 * If failpos has been set, that mirror has failed, so it needs
4287 	 * to be disabled.
4288 	 */
4289 	if (failpos >= 0) {
4290 		nmp = VFSTONFS(dvp[failpos]->v_mount);
4291 		NFSLOCKMNT(nmp);
4292 		if ((nmp->nm_privflag & (NFSMNTP_FORCEDISM |
4293 		     NFSMNTP_CANCELRPCS)) == 0) {
4294 			nmp->nm_privflag |= NFSMNTP_CANCELRPCS;
4295 			NFSUNLOCKMNT(nmp);
4296 			ds = nfsrv_deldsnmp(PNFSDOP_DELDSSERVER, nmp, p);
4297 			NFSD_DEBUG(4, "dsremovefail fail=%d ds=%p\n", failpos,
4298 			    ds);
4299 			if (ds != NULL)
4300 				nfsrv_killrpcs(nmp);
4301 			NFSLOCKMNT(nmp);
4302 			nmp->nm_privflag &= ~NFSMNTP_CANCELRPCS;
4303 			wakeup(nmp);
4304 		}
4305 		NFSUNLOCKMNT(nmp);
4306 	}
4307 
4308 	/* Get rid all layouts for the file. */
4309 	nfsrv_freefilelayouts(fhp);
4310 
4311 	NFSFREECRED(tcred);
4312 	free(dsrm, M_TEMP);
4313 }
4314 
4315 /*
4316  * Generate a file name based on the file handle and put it in *bufp.
4317  * Return the number of bytes generated.
4318  */
4319 static int
4320 nfsrv_putfhname(fhandle_t *fhp, char *bufp)
4321 {
4322 	int i;
4323 	uint8_t *cp;
4324 	const uint8_t *hexdigits = "0123456789abcdef";
4325 
4326 	cp = (uint8_t *)fhp;
4327 	for (i = 0; i < sizeof(*fhp); i++) {
4328 		bufp[2 * i] = hexdigits[(*cp >> 4) & 0xf];
4329 		bufp[2 * i + 1] = hexdigits[*cp++ & 0xf];
4330 	}
4331 	bufp[2 * i] = '\0';
4332 	return (2 * i);
4333 }
4334 
4335 /*
4336  * Update the Metadata file's attributes from the DS file when a Read/Write
4337  * layout is returned.
4338  * Basically just call nfsrv_proxyds() with procedure == NFSPROC_LAYOUTRETURN
4339  * so that it does a nfsrv_getattrdsrpc() and nfsrv_setextattr() on the DS file.
4340  */
4341 int
4342 nfsrv_updatemdsattr(struct vnode *vp, struct nfsvattr *nap, NFSPROC_T *p)
4343 {
4344 	struct ucred *tcred;
4345 	int error;
4346 
4347 	/* Do this as root so that it won't fail with EACCES. */
4348 	tcred = newnfs_getcred();
4349 	error = nfsrv_proxyds(NULL, vp, 0, 0, tcred, p, NFSPROC_LAYOUTRETURN,
4350 	    NULL, NULL, NULL, nap, NULL);
4351 	NFSFREECRED(tcred);
4352 	return (error);
4353 }
4354 
4355 /*
4356  * Set the NFSv4 ACL on the DS file to the same ACL as the MDS file.
4357  */
4358 static int
4359 nfsrv_dssetacl(struct vnode *vp, struct acl *aclp, struct ucred *cred,
4360     NFSPROC_T *p)
4361 {
4362 	int error;
4363 
4364 	error = nfsrv_proxyds(NULL, vp, 0, 0, cred, p, NFSPROC_SETACL,
4365 	    NULL, NULL, NULL, NULL, aclp);
4366 	return (error);
4367 }
4368 
4369 static int
4370 nfsrv_proxyds(struct nfsrv_descript *nd, struct vnode *vp, off_t off, int cnt,
4371     struct ucred *cred, struct thread *p, int ioproc, struct mbuf **mpp,
4372     char *cp, struct mbuf **mpp2, struct nfsvattr *nap, struct acl *aclp)
4373 {
4374 	struct nfsmount *nmp[NFSDEV_MAXMIRRORS], *failnmp;
4375 	fhandle_t fh[NFSDEV_MAXMIRRORS];
4376 	struct vnode *dvp[NFSDEV_MAXMIRRORS];
4377 	struct nfsdevice *ds;
4378 	struct pnfsdsattr dsattr;
4379 	char *buf;
4380 	int buflen, error, failpos, i, mirrorcnt, origmircnt, trycnt;
4381 
4382 	NFSD_DEBUG(4, "in nfsrv_proxyds\n");
4383 	/*
4384 	 * If not a regular file, not exported or not a pNFS server,
4385 	 * just return ENOENT.
4386 	 */
4387 	if (vp->v_type != VREG || (vp->v_mount->mnt_flag & MNT_EXPORTED) == 0 ||
4388 	    nfsrv_devidcnt == 0)
4389 		return (ENOENT);
4390 
4391 	buflen = 1024;
4392 	buf = malloc(buflen, M_TEMP, M_WAITOK);
4393 	error = 0;
4394 
4395 	/*
4396 	 * For Getattr, get the Change attribute (va_filerev) and size (va_size)
4397 	 * from the MetaData file's extended attribute.
4398 	 */
4399 	if (ioproc == NFSPROC_GETATTR) {
4400 		error = vn_extattr_get(vp, IO_NODELOCKED,
4401 		    EXTATTR_NAMESPACE_SYSTEM, "pnfsd.dsattr", &buflen, buf,
4402 		    p);
4403 		if (error == 0 && buflen != sizeof(dsattr))
4404 			error = ENXIO;
4405 		if (error == 0) {
4406 			NFSBCOPY(buf, &dsattr, buflen);
4407 			nap->na_filerev = dsattr.dsa_filerev;
4408 			nap->na_size = dsattr.dsa_size;
4409 			nap->na_atime = dsattr.dsa_atime;
4410 			nap->na_mtime = dsattr.dsa_mtime;
4411 
4412 			/*
4413 			 * If nfsrv_pnfsgetdsattr is 0 or nfsrv_checkdsattr()
4414 			 * returns 0, just return now.  nfsrv_checkdsattr()
4415 			 * returns 0 if there is no Read/Write layout
4416 			 * plus either an Open/Write_access or Write
4417 			 * delegation issued to a client for the file.
4418 			 */
4419 			if (nfsrv_pnfsgetdsattr == 0 ||
4420 			    nfsrv_checkdsattr(nd, vp, p) == 0) {
4421 				free(buf, M_TEMP);
4422 				return (error);
4423 			}
4424 		}
4425 
4426 		/*
4427 		 * Clear ENOATTR so the code below will attempt to do a
4428 		 * nfsrv_getattrdsrpc() to get the attributes and (re)create
4429 		 * the extended attribute.
4430 		 */
4431 		if (error == ENOATTR)
4432 			error = 0;
4433 	}
4434 
4435 	origmircnt = -1;
4436 	trycnt = 0;
4437 tryagain:
4438 	if (error == 0) {
4439 		buflen = 1024;
4440 		if (ioproc == NFSPROC_READDS && NFSVOPISLOCKED(vp) ==
4441 		    LK_EXCLUSIVE)
4442 			printf("nfsrv_proxyds: Readds vp exclusively locked\n");
4443 		error = nfsrv_dsgetsockmnt(vp, LK_SHARED, buf, &buflen,
4444 		    &mirrorcnt, p, dvp, fh, NULL, NULL, NULL, NULL, NULL,
4445 		    NULL, NULL);
4446 		if (error == 0) {
4447 			for (i = 0; i < mirrorcnt; i++)
4448 				nmp[i] = VFSTONFS(dvp[i]->v_mount);
4449 		} else
4450 			printf("pNFS: proxy getextattr sockaddr=%d\n", error);
4451 	} else
4452 		printf("pNFS: nfsrv_dsgetsockmnt=%d\n", error);
4453 	if (error == 0) {
4454 		failpos = -1;
4455 		if (origmircnt == -1)
4456 			origmircnt = mirrorcnt;
4457 		/*
4458 		 * If failpos is set to a mirror#, then that mirror has
4459 		 * failed and will be disabled. For Read and Getattr, the
4460 		 * function only tries one mirror, so if that mirror has
4461 		 * failed, it will need to be retried. As such, increment
4462 		 * tryitagain for these cases.
4463 		 * For Write, Setattr and Setacl, the function tries all
4464 		 * mirrors and will not return an error for the case where
4465 		 * one mirror has failed. For these cases, the functioning
4466 		 * mirror(s) will have been modified, so a retry isn't
4467 		 * necessary. These functions will set failpos for the
4468 		 * failed mirror#.
4469 		 */
4470 		if (ioproc == NFSPROC_READDS) {
4471 			error = nfsrv_readdsrpc(fh, off, cnt, cred, p, nmp[0],
4472 			    mpp, mpp2);
4473 			if (nfsds_failerr(error) && mirrorcnt > 1) {
4474 				/*
4475 				 * Setting failpos will cause the mirror
4476 				 * to be disabled and then a retry of this
4477 				 * read is required.
4478 				 */
4479 				failpos = 0;
4480 				error = 0;
4481 				trycnt++;
4482 			}
4483 		} else if (ioproc == NFSPROC_WRITEDS)
4484 			error = nfsrv_writedsrpc(fh, off, cnt, cred, p, vp,
4485 			    &nmp[0], mirrorcnt, mpp, cp, &failpos);
4486 		else if (ioproc == NFSPROC_SETATTR)
4487 			error = nfsrv_setattrdsrpc(fh, cred, p, vp, &nmp[0],
4488 			    mirrorcnt, nap, &failpos);
4489 		else if (ioproc == NFSPROC_SETACL)
4490 			error = nfsrv_setacldsrpc(fh, cred, p, vp, &nmp[0],
4491 			    mirrorcnt, aclp, &failpos);
4492 		else {
4493 			error = nfsrv_getattrdsrpc(&fh[mirrorcnt - 1], cred, p,
4494 			    vp, nmp[mirrorcnt - 1], nap);
4495 			if (nfsds_failerr(error) && mirrorcnt > 1) {
4496 				/*
4497 				 * Setting failpos will cause the mirror
4498 				 * to be disabled and then a retry of this
4499 				 * getattr is required.
4500 				 */
4501 				failpos = mirrorcnt - 1;
4502 				error = 0;
4503 				trycnt++;
4504 			}
4505 		}
4506 		ds = NULL;
4507 		if (failpos >= 0) {
4508 			failnmp = nmp[failpos];
4509 			NFSLOCKMNT(failnmp);
4510 			if ((failnmp->nm_privflag & (NFSMNTP_FORCEDISM |
4511 			     NFSMNTP_CANCELRPCS)) == 0) {
4512 				failnmp->nm_privflag |= NFSMNTP_CANCELRPCS;
4513 				NFSUNLOCKMNT(failnmp);
4514 				ds = nfsrv_deldsnmp(PNFSDOP_DELDSSERVER,
4515 				    failnmp, p);
4516 				NFSD_DEBUG(4, "dsldsnmp fail=%d ds=%p\n",
4517 				    failpos, ds);
4518 				if (ds != NULL)
4519 					nfsrv_killrpcs(failnmp);
4520 				NFSLOCKMNT(failnmp);
4521 				failnmp->nm_privflag &= ~NFSMNTP_CANCELRPCS;
4522 				wakeup(failnmp);
4523 			}
4524 			NFSUNLOCKMNT(failnmp);
4525 		}
4526 		for (i = 0; i < mirrorcnt; i++)
4527 			NFSVOPUNLOCK(dvp[i], 0);
4528 		NFSD_DEBUG(4, "nfsrv_proxyds: aft RPC=%d trya=%d\n", error,
4529 		    trycnt);
4530 		/* Try the Read/Getattr again if a mirror was deleted. */
4531 		if (ds != NULL && trycnt > 0 && trycnt < origmircnt)
4532 			goto tryagain;
4533 	} else {
4534 		/* Return ENOENT for any Extended Attribute error. */
4535 		error = ENOENT;
4536 	}
4537 	free(buf, M_TEMP);
4538 	NFSD_DEBUG(4, "nfsrv_proxyds: error=%d\n", error);
4539 	return (error);
4540 }
4541 
4542 /*
4543  * Get the DS mount point, fh and directory from the "pnfsd.dsfile" extended
4544  * attribute.
4545  * newnmpp - If it points to a non-NULL nmp, that is the destination and needs
4546  *           to be checked.  If it points to a NULL nmp, then it returns
4547  *           a suitable destination.
4548  * curnmp - If non-NULL, it is the source mount for the copy.
4549  */
4550 int
4551 nfsrv_dsgetsockmnt(struct vnode *vp, int lktype, char *buf, int *buflenp,
4552     int *mirrorcntp, NFSPROC_T *p, struct vnode **dvpp, fhandle_t *fhp,
4553     char *devid, char *fnamep, struct vnode **nvpp, struct nfsmount **newnmpp,
4554     struct nfsmount *curnmp, int *ippos, int *dsdirp)
4555 {
4556 	struct vnode *dvp, *nvp, **tdvpp;
4557 	struct mount *mp;
4558 	struct nfsmount *nmp, *newnmp;
4559 	struct sockaddr *sad;
4560 	struct sockaddr_in *sin;
4561 	struct nfsdevice *ds, *tds, *fndds;
4562 	struct pnfsdsfile *pf;
4563 	uint32_t dsdir;
4564 	int error, fhiszero, fnd, gotone, i, mirrorcnt;
4565 
4566 	ASSERT_VOP_LOCKED(vp, "nfsrv_dsgetsockmnt vp");
4567 	*mirrorcntp = 1;
4568 	tdvpp = dvpp;
4569 	if (nvpp != NULL)
4570 		*nvpp = NULL;
4571 	if (dvpp != NULL)
4572 		*dvpp = NULL;
4573 	if (ippos != NULL)
4574 		*ippos = -1;
4575 	if (newnmpp != NULL)
4576 		newnmp = *newnmpp;
4577 	else
4578 		newnmp = NULL;
4579 	mp = vp->v_mount;
4580 	error = vn_extattr_get(vp, IO_NODELOCKED, EXTATTR_NAMESPACE_SYSTEM,
4581 	    "pnfsd.dsfile", buflenp, buf, p);
4582 	mirrorcnt = *buflenp / sizeof(*pf);
4583 	if (error == 0 && (mirrorcnt < 1 || mirrorcnt > NFSDEV_MAXMIRRORS ||
4584 	    *buflenp != sizeof(*pf) * mirrorcnt))
4585 		error = ENOATTR;
4586 
4587 	pf = (struct pnfsdsfile *)buf;
4588 	/* If curnmp != NULL, check for a match in the mirror list. */
4589 	if (curnmp != NULL && error == 0) {
4590 		fnd = 0;
4591 		for (i = 0; i < mirrorcnt; i++, pf++) {
4592 			sad = (struct sockaddr *)&pf->dsf_sin;
4593 			if (nfsaddr2_match(sad, curnmp->nm_nam)) {
4594 				if (ippos != NULL)
4595 					*ippos = i;
4596 				fnd = 1;
4597 				break;
4598 			}
4599 		}
4600 		if (fnd == 0)
4601 			error = ENXIO;
4602 	}
4603 
4604 	gotone = 0;
4605 	pf = (struct pnfsdsfile *)buf;
4606 	NFSD_DEBUG(4, "nfsrv_dsgetsockmnt: mirrorcnt=%d err=%d\n", mirrorcnt,
4607 	    error);
4608 	for (i = 0; i < mirrorcnt && error == 0; i++, pf++) {
4609 		fhiszero = 0;
4610 		sad = (struct sockaddr *)&pf->dsf_sin;
4611 		sin = &pf->dsf_sin;
4612 		dsdir = pf->dsf_dir;
4613 		if (dsdir >= nfsrv_dsdirsize) {
4614 			printf("nfsrv_dsgetsockmnt: dsdir=%d\n", dsdir);
4615 			error = ENOATTR;
4616 		} else if (nvpp != NULL && newnmp != NULL &&
4617 		    nfsaddr2_match(sad, newnmp->nm_nam))
4618 			error = EEXIST;
4619 		if (error == 0) {
4620 			if (ippos != NULL && curnmp == NULL &&
4621 			    sad->sa_family == AF_INET &&
4622 			    sin->sin_addr.s_addr == 0)
4623 				*ippos = i;
4624 			if (NFSBCMP(&zerofh, &pf->dsf_fh, sizeof(zerofh)) == 0)
4625 				fhiszero = 1;
4626 			/* Use the socket address to find the mount point. */
4627 			fndds = NULL;
4628 			NFSDDSLOCK();
4629 			/* Find a match for the IP address. */
4630 			TAILQ_FOREACH(ds, &nfsrv_devidhead, nfsdev_list) {
4631 				if (ds->nfsdev_nmp != NULL) {
4632 					dvp = ds->nfsdev_dvp;
4633 					nmp = VFSTONFS(dvp->v_mount);
4634 					if (nmp != ds->nfsdev_nmp)
4635 						printf("different2 nmp %p %p\n",
4636 						    nmp, ds->nfsdev_nmp);
4637 					if (nfsaddr2_match(sad, nmp->nm_nam)) {
4638 						fndds = ds;
4639 						break;
4640 					}
4641 				}
4642 			}
4643 			if (fndds != NULL && newnmpp != NULL &&
4644 			    newnmp == NULL) {
4645 				/* Search for a place to make a mirror copy. */
4646 				TAILQ_FOREACH(tds, &nfsrv_devidhead,
4647 				    nfsdev_list) {
4648 					if (tds->nfsdev_nmp != NULL &&
4649 					    fndds != tds &&
4650 					    ((tds->nfsdev_mdsisset == 0 &&
4651 					      fndds->nfsdev_mdsisset == 0) ||
4652 					     (tds->nfsdev_mdsisset != 0 &&
4653 					      fndds->nfsdev_mdsisset != 0 &&
4654 					      tds->nfsdev_mdsfsid.val[0] ==
4655 					      mp->mnt_stat.f_fsid.val[0] &&
4656 					      tds->nfsdev_mdsfsid.val[1] ==
4657 					      mp->mnt_stat.f_fsid.val[1]))) {
4658 						*newnmpp = tds->nfsdev_nmp;
4659 						break;
4660 					}
4661 				}
4662 				if (tds != NULL) {
4663 					/*
4664 					 * Move this entry to the end of the
4665 					 * list, so it won't be selected as
4666 					 * easily the next time.
4667 					 */
4668 					TAILQ_REMOVE(&nfsrv_devidhead, tds,
4669 					    nfsdev_list);
4670 					TAILQ_INSERT_TAIL(&nfsrv_devidhead, tds,
4671 					    nfsdev_list);
4672 				}
4673 			}
4674 			NFSDDSUNLOCK();
4675 			if (fndds != NULL) {
4676 				dvp = fndds->nfsdev_dsdir[dsdir];
4677 				if (lktype != 0 || fhiszero != 0 ||
4678 				    (nvpp != NULL && *nvpp == NULL)) {
4679 					if (fhiszero != 0)
4680 						error = vn_lock(dvp,
4681 						    LK_EXCLUSIVE);
4682 					else if (lktype != 0)
4683 						error = vn_lock(dvp, lktype);
4684 					else
4685 						error = vn_lock(dvp, LK_SHARED);
4686 					/*
4687 					 * If the file handle is all 0's, try to
4688 					 * do a Lookup against the DS to acquire
4689 					 * it.
4690 					 * If dvpp == NULL or the Lookup fails,
4691 					 * unlock dvp after the call.
4692 					 */
4693 					if (error == 0 && (fhiszero != 0 ||
4694 					    (nvpp != NULL && *nvpp == NULL))) {
4695 						error = nfsrv_pnfslookupds(vp,
4696 						    dvp, pf, &nvp, p);
4697 						if (error == 0) {
4698 							if (fhiszero != 0)
4699 								nfsrv_pnfssetfh(
4700 								    vp, pf,
4701 								    devid,
4702 								    fnamep,
4703 								    nvp, p);
4704 							if (nvpp != NULL &&
4705 							    *nvpp == NULL) {
4706 								*nvpp = nvp;
4707 								*dsdirp = dsdir;
4708 							} else
4709 								vput(nvp);
4710 						}
4711 						if (error != 0 || lktype == 0)
4712 							NFSVOPUNLOCK(dvp, 0);
4713 					}
4714 				}
4715 				if (error == 0) {
4716 					gotone++;
4717 					NFSD_DEBUG(4, "gotone=%d\n", gotone);
4718 					if (devid != NULL) {
4719 						NFSBCOPY(fndds->nfsdev_deviceid,
4720 						    devid, NFSX_V4DEVICEID);
4721 						devid += NFSX_V4DEVICEID;
4722 					}
4723 					if (dvpp != NULL)
4724 						*tdvpp++ = dvp;
4725 					if (fhp != NULL)
4726 						NFSBCOPY(&pf->dsf_fh, fhp++,
4727 						    NFSX_MYFH);
4728 					if (fnamep != NULL && gotone == 1)
4729 						strlcpy(fnamep,
4730 						    pf->dsf_filename,
4731 						    sizeof(pf->dsf_filename));
4732 				} else
4733 					NFSD_DEBUG(4, "nfsrv_dsgetsockmnt "
4734 					    "err=%d\n", error);
4735 			}
4736 		}
4737 	}
4738 	if (error == 0 && gotone == 0)
4739 		error = ENOENT;
4740 
4741 	NFSD_DEBUG(4, "eo nfsrv_dsgetsockmnt: gotone=%d err=%d\n", gotone,
4742 	    error);
4743 	if (error == 0)
4744 		*mirrorcntp = gotone;
4745 	else {
4746 		if (gotone > 0 && dvpp != NULL) {
4747 			/*
4748 			 * If the error didn't occur on the first one and
4749 			 * dvpp != NULL, the one(s) prior to the failure will
4750 			 * have locked dvp's that need to be unlocked.
4751 			 */
4752 			for (i = 0; i < gotone; i++) {
4753 				NFSVOPUNLOCK(*dvpp, 0);
4754 				*dvpp++ = NULL;
4755 			}
4756 		}
4757 		/*
4758 		 * If it found the vnode to be copied from before a failure,
4759 		 * it needs to be vput()'d.
4760 		 */
4761 		if (nvpp != NULL && *nvpp != NULL) {
4762 			vput(*nvpp);
4763 			*nvpp = NULL;
4764 		}
4765 	}
4766 	return (error);
4767 }
4768 
4769 /*
4770  * Set the extended attribute for the Change attribute.
4771  */
4772 static int
4773 nfsrv_setextattr(struct vnode *vp, struct nfsvattr *nap, NFSPROC_T *p)
4774 {
4775 	struct pnfsdsattr dsattr;
4776 	int error;
4777 
4778 	ASSERT_VOP_ELOCKED(vp, "nfsrv_setextattr vp");
4779 	dsattr.dsa_filerev = nap->na_filerev;
4780 	dsattr.dsa_size = nap->na_size;
4781 	dsattr.dsa_atime = nap->na_atime;
4782 	dsattr.dsa_mtime = nap->na_mtime;
4783 	error = vn_extattr_set(vp, IO_NODELOCKED, EXTATTR_NAMESPACE_SYSTEM,
4784 	    "pnfsd.dsattr", sizeof(dsattr), (char *)&dsattr, p);
4785 	if (error != 0)
4786 		printf("pNFS: setextattr=%d\n", error);
4787 	return (error);
4788 }
4789 
4790 static int
4791 nfsrv_readdsrpc(fhandle_t *fhp, off_t off, int len, struct ucred *cred,
4792     NFSPROC_T *p, struct nfsmount *nmp, struct mbuf **mpp, struct mbuf **mpendp)
4793 {
4794 	uint32_t *tl;
4795 	struct nfsrv_descript *nd;
4796 	nfsv4stateid_t st;
4797 	struct mbuf *m, *m2;
4798 	int error = 0, retlen, tlen, trimlen;
4799 
4800 	NFSD_DEBUG(4, "in nfsrv_readdsrpc\n");
4801 	nd = malloc(sizeof(*nd), M_TEMP, M_WAITOK | M_ZERO);
4802 	*mpp = NULL;
4803 	/*
4804 	 * Use a stateid where other is an alternating 01010 pattern and
4805 	 * seqid is 0xffffffff.  This value is not defined as special by
4806 	 * the RFC and is used by the FreeBSD NFS server to indicate an
4807 	 * MDS->DS proxy operation.
4808 	 */
4809 	st.other[0] = 0x55555555;
4810 	st.other[1] = 0x55555555;
4811 	st.other[2] = 0x55555555;
4812 	st.seqid = 0xffffffff;
4813 	nfscl_reqstart(nd, NFSPROC_READDS, nmp, (u_int8_t *)fhp, sizeof(*fhp),
4814 	    NULL, NULL, 0, 0);
4815 	nfsm_stateidtom(nd, &st, NFSSTATEID_PUTSTATEID);
4816 	NFSM_BUILD(tl, uint32_t *, NFSX_UNSIGNED * 3);
4817 	txdr_hyper(off, tl);
4818 	*(tl + 2) = txdr_unsigned(len);
4819 	error = newnfs_request(nd, nmp, NULL, &nmp->nm_sockreq, NULL, p, cred,
4820 	    NFS_PROG, NFS_VER4, NULL, 1, NULL, NULL);
4821 	if (error != 0) {
4822 		free(nd, M_TEMP);
4823 		return (error);
4824 	}
4825 	if (nd->nd_repstat == 0) {
4826 		NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
4827 		NFSM_STRSIZ(retlen, len);
4828 		if (retlen > 0) {
4829 			/* Trim off the pre-data XDR from the mbuf chain. */
4830 			m = nd->nd_mrep;
4831 			while (m != NULL && m != nd->nd_md) {
4832 				if (m->m_next == nd->nd_md) {
4833 					m->m_next = NULL;
4834 					m_freem(nd->nd_mrep);
4835 					nd->nd_mrep = m = nd->nd_md;
4836 				} else
4837 					m = m->m_next;
4838 			}
4839 			if (m == NULL) {
4840 				printf("nfsrv_readdsrpc: busted mbuf list\n");
4841 				error = ENOENT;
4842 				goto nfsmout;
4843 			}
4844 
4845 			/*
4846 			 * Now, adjust first mbuf so that any XDR before the
4847 			 * read data is skipped over.
4848 			 */
4849 			trimlen = nd->nd_dpos - mtod(m, char *);
4850 			if (trimlen > 0) {
4851 				m->m_len -= trimlen;
4852 				NFSM_DATAP(m, trimlen);
4853 			}
4854 
4855 			/*
4856 			 * Truncate the mbuf chain at retlen bytes of data,
4857 			 * plus XDR padding that brings the length up to a
4858 			 * multiple of 4.
4859 			 */
4860 			tlen = NFSM_RNDUP(retlen);
4861 			do {
4862 				if (m->m_len >= tlen) {
4863 					m->m_len = tlen;
4864 					tlen = 0;
4865 					m2 = m->m_next;
4866 					m->m_next = NULL;
4867 					m_freem(m2);
4868 					break;
4869 				}
4870 				tlen -= m->m_len;
4871 				m = m->m_next;
4872 			} while (m != NULL);
4873 			if (tlen > 0) {
4874 				printf("nfsrv_readdsrpc: busted mbuf list\n");
4875 				error = ENOENT;
4876 				goto nfsmout;
4877 			}
4878 			*mpp = nd->nd_mrep;
4879 			*mpendp = m;
4880 			nd->nd_mrep = NULL;
4881 		}
4882 	} else
4883 		error = nd->nd_repstat;
4884 nfsmout:
4885 	/* If nd->nd_mrep is already NULL, this is a no-op. */
4886 	m_freem(nd->nd_mrep);
4887 	free(nd, M_TEMP);
4888 	NFSD_DEBUG(4, "nfsrv_readdsrpc error=%d\n", error);
4889 	return (error);
4890 }
4891 
4892 /*
4893  * Do a write RPC on a DS data file, using this structure for the arguments,
4894  * so that this function can be executed by a separate kernel process.
4895  */
4896 struct nfsrvwritedsdorpc {
4897 	int			done;
4898 	int			inprog;
4899 	struct task		tsk;
4900 	fhandle_t		fh;
4901 	off_t			off;
4902 	int			len;
4903 	struct nfsmount		*nmp;
4904 	struct ucred		*cred;
4905 	NFSPROC_T		*p;
4906 	struct mbuf		*m;
4907 	int			err;
4908 };
4909 
4910 static int
4911 nfsrv_writedsdorpc(struct nfsmount *nmp, fhandle_t *fhp, off_t off, int len,
4912     struct nfsvattr *nap, struct mbuf *m, struct ucred *cred, NFSPROC_T *p)
4913 {
4914 	uint32_t *tl;
4915 	struct nfsrv_descript *nd;
4916 	nfsattrbit_t attrbits;
4917 	nfsv4stateid_t st;
4918 	int commit, error, retlen;
4919 
4920 	nd = malloc(sizeof(*nd), M_TEMP, M_WAITOK | M_ZERO);
4921 	nfscl_reqstart(nd, NFSPROC_WRITE, nmp, (u_int8_t *)fhp,
4922 	    sizeof(fhandle_t), NULL, NULL, 0, 0);
4923 
4924 	/*
4925 	 * Use a stateid where other is an alternating 01010 pattern and
4926 	 * seqid is 0xffffffff.  This value is not defined as special by
4927 	 * the RFC and is used by the FreeBSD NFS server to indicate an
4928 	 * MDS->DS proxy operation.
4929 	 */
4930 	st.other[0] = 0x55555555;
4931 	st.other[1] = 0x55555555;
4932 	st.other[2] = 0x55555555;
4933 	st.seqid = 0xffffffff;
4934 	nfsm_stateidtom(nd, &st, NFSSTATEID_PUTSTATEID);
4935 	NFSM_BUILD(tl, u_int32_t *, NFSX_HYPER + 2 * NFSX_UNSIGNED);
4936 	txdr_hyper(off, tl);
4937 	tl += 2;
4938 	/*
4939 	 * Do all writes FileSync, since the server doesn't hold onto dirty
4940 	 * buffers.  Since clients should be accessing the DS servers directly
4941 	 * using the pNFS layouts, this just needs to work correctly as a
4942 	 * fallback.
4943 	 */
4944 	*tl++ = txdr_unsigned(NFSWRITE_FILESYNC);
4945 	*tl = txdr_unsigned(len);
4946 	NFSD_DEBUG(4, "nfsrv_writedsdorpc: len=%d\n", len);
4947 
4948 	/* Put data in mbuf chain. */
4949 	nd->nd_mb->m_next = m;
4950 
4951 	/* Set nd_mb and nd_bpos to end of data. */
4952 	while (m->m_next != NULL)
4953 		m = m->m_next;
4954 	nd->nd_mb = m;
4955 	nd->nd_bpos = mtod(m, char *) + m->m_len;
4956 	NFSD_DEBUG(4, "nfsrv_writedsdorpc: lastmb len=%d\n", m->m_len);
4957 
4958 	/* Do a Getattr for Size, Change and Modify Time. */
4959 	NFSZERO_ATTRBIT(&attrbits);
4960 	NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_SIZE);
4961 	NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_CHANGE);
4962 	NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_TIMEACCESS);
4963 	NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_TIMEMODIFY);
4964 	NFSM_BUILD(tl, u_int32_t *, NFSX_UNSIGNED);
4965 	*tl = txdr_unsigned(NFSV4OP_GETATTR);
4966 	(void) nfsrv_putattrbit(nd, &attrbits);
4967 	error = newnfs_request(nd, nmp, NULL, &nmp->nm_sockreq, NULL, p,
4968 	    cred, NFS_PROG, NFS_VER4, NULL, 1, NULL, NULL);
4969 	if (error != 0) {
4970 		free(nd, M_TEMP);
4971 		return (error);
4972 	}
4973 	NFSD_DEBUG(4, "nfsrv_writedsdorpc: aft writerpc=%d\n", nd->nd_repstat);
4974 	/* Get rid of weak cache consistency data for now. */
4975 	if ((nd->nd_flag & (ND_NOMOREDATA | ND_NFSV4 | ND_V4WCCATTR)) ==
4976 	    (ND_NFSV4 | ND_V4WCCATTR)) {
4977 		error = nfsv4_loadattr(nd, NULL, nap, NULL, NULL, 0, NULL, NULL,
4978 		    NULL, NULL, NULL, 0, NULL, NULL, NULL, NULL, NULL);
4979 		NFSD_DEBUG(4, "nfsrv_writedsdorpc: wcc attr=%d\n", error);
4980 		if (error != 0)
4981 			goto nfsmout;
4982 		/*
4983 		 * Get rid of Op# and status for next op.
4984 		 */
4985 		NFSM_DISSECT(tl, uint32_t *, 2 * NFSX_UNSIGNED);
4986 		if (*++tl != 0)
4987 			nd->nd_flag |= ND_NOMOREDATA;
4988 	}
4989 	if (nd->nd_repstat == 0) {
4990 		NFSM_DISSECT(tl, uint32_t *, 2 * NFSX_UNSIGNED + NFSX_VERF);
4991 		retlen = fxdr_unsigned(int, *tl++);
4992 		commit = fxdr_unsigned(int, *tl);
4993 		if (commit != NFSWRITE_FILESYNC)
4994 			error = NFSERR_IO;
4995 		NFSD_DEBUG(4, "nfsrv_writedsdorpc:retlen=%d commit=%d err=%d\n",
4996 		    retlen, commit, error);
4997 	} else
4998 		error = nd->nd_repstat;
4999 	/* We have no use for the Write Verifier since we use FileSync. */
5000 
5001 	/*
5002 	 * Get the Change, Size, Access Time and Modify Time attributes and set
5003 	 * on the Metadata file, so its attributes will be what the file's
5004 	 * would be if it had been written.
5005 	 */
5006 	if (error == 0) {
5007 		NFSM_DISSECT(tl, uint32_t *, 2 * NFSX_UNSIGNED);
5008 		error = nfsv4_loadattr(nd, NULL, nap, NULL, NULL, 0, NULL, NULL,
5009 		    NULL, NULL, NULL, 0, NULL, NULL, NULL, NULL, NULL);
5010 	}
5011 	NFSD_DEBUG(4, "nfsrv_writedsdorpc: aft loadattr=%d\n", error);
5012 nfsmout:
5013 	m_freem(nd->nd_mrep);
5014 	free(nd, M_TEMP);
5015 	NFSD_DEBUG(4, "nfsrv_writedsdorpc error=%d\n", error);
5016 	return (error);
5017 }
5018 
5019 /*
5020  * Start up the thread that will execute nfsrv_writedsdorpc().
5021  */
5022 static void
5023 start_writedsdorpc(void *arg, int pending)
5024 {
5025 	struct nfsrvwritedsdorpc *drpc;
5026 
5027 	drpc = (struct nfsrvwritedsdorpc *)arg;
5028 	drpc->err = nfsrv_writedsdorpc(drpc->nmp, &drpc->fh, drpc->off,
5029 	    drpc->len, NULL, drpc->m, drpc->cred, drpc->p);
5030 	drpc->done = 1;
5031 	NFSD_DEBUG(4, "start_writedsdorpc: err=%d\n", drpc->err);
5032 }
5033 
5034 static int
5035 nfsrv_writedsrpc(fhandle_t *fhp, off_t off, int len, struct ucred *cred,
5036     NFSPROC_T *p, struct vnode *vp, struct nfsmount **nmpp, int mirrorcnt,
5037     struct mbuf **mpp, char *cp, int *failposp)
5038 {
5039 	struct nfsrvwritedsdorpc *drpc, *tdrpc;
5040 	struct nfsvattr na;
5041 	struct mbuf *m;
5042 	int error, i, offs, ret, timo;
5043 
5044 	NFSD_DEBUG(4, "in nfsrv_writedsrpc\n");
5045 	KASSERT(*mpp != NULL, ("nfsrv_writedsrpc: NULL mbuf chain"));
5046 	drpc = NULL;
5047 	if (mirrorcnt > 1)
5048 		tdrpc = drpc = malloc(sizeof(*drpc) * (mirrorcnt - 1), M_TEMP,
5049 		    M_WAITOK);
5050 
5051 	/* Calculate offset in mbuf chain that data starts. */
5052 	offs = cp - mtod(*mpp, char *);
5053 	NFSD_DEBUG(4, "nfsrv_writedsrpc: mcopy offs=%d len=%d\n", offs, len);
5054 
5055 	/*
5056 	 * Do the write RPC for every DS, using a separate kernel process
5057 	 * for every DS except the last one.
5058 	 */
5059 	error = 0;
5060 	for (i = 0; i < mirrorcnt - 1; i++, tdrpc++) {
5061 		tdrpc->done = 0;
5062 		NFSBCOPY(fhp, &tdrpc->fh, sizeof(*fhp));
5063 		tdrpc->off = off;
5064 		tdrpc->len = len;
5065 		tdrpc->nmp = *nmpp;
5066 		tdrpc->cred = cred;
5067 		tdrpc->p = p;
5068 		tdrpc->inprog = 0;
5069 		tdrpc->err = 0;
5070 		tdrpc->m = m_copym(*mpp, offs, NFSM_RNDUP(len), M_WAITOK);
5071 		ret = EIO;
5072 		if (nfs_pnfsiothreads != 0) {
5073 			ret = nfs_pnfsio(start_writedsdorpc, tdrpc);
5074 			NFSD_DEBUG(4, "nfsrv_writedsrpc: nfs_pnfsio=%d\n",
5075 			    ret);
5076 		}
5077 		if (ret != 0) {
5078 			ret = nfsrv_writedsdorpc(*nmpp, fhp, off, len, NULL,
5079 			    tdrpc->m, cred, p);
5080 			if (nfsds_failerr(ret) && *failposp == -1)
5081 				*failposp = i;
5082 			else if (error == 0 && ret != 0)
5083 				error = ret;
5084 		}
5085 		nmpp++;
5086 		fhp++;
5087 	}
5088 	m = m_copym(*mpp, offs, NFSM_RNDUP(len), M_WAITOK);
5089 	ret = nfsrv_writedsdorpc(*nmpp, fhp, off, len, &na, m, cred, p);
5090 	if (nfsds_failerr(ret) && *failposp == -1 && mirrorcnt > 1)
5091 		*failposp = mirrorcnt - 1;
5092 	else if (error == 0 && ret != 0)
5093 		error = ret;
5094 	if (error == 0)
5095 		error = nfsrv_setextattr(vp, &na, p);
5096 	NFSD_DEBUG(4, "nfsrv_writedsrpc: aft setextat=%d\n", error);
5097 	tdrpc = drpc;
5098 	timo = hz / 50;		/* Wait for 20msec. */
5099 	if (timo < 1)
5100 		timo = 1;
5101 	for (i = 0; i < mirrorcnt - 1; i++, tdrpc++) {
5102 		/* Wait for RPCs on separate threads to complete. */
5103 		while (tdrpc->inprog != 0 && tdrpc->done == 0)
5104 			tsleep(&tdrpc->tsk, PVFS, "srvwrds", timo);
5105 		if (nfsds_failerr(tdrpc->err) && *failposp == -1)
5106 			*failposp = i;
5107 		else if (error == 0 && tdrpc->err != 0)
5108 			error = tdrpc->err;
5109 	}
5110 	free(drpc, M_TEMP);
5111 	return (error);
5112 }
5113 
5114 static int
5115 nfsrv_setattrdsdorpc(fhandle_t *fhp, struct ucred *cred, NFSPROC_T *p,
5116     struct vnode *vp, struct nfsmount *nmp, struct nfsvattr *nap,
5117     struct nfsvattr *dsnap)
5118 {
5119 	uint32_t *tl;
5120 	struct nfsrv_descript *nd;
5121 	nfsv4stateid_t st;
5122 	nfsattrbit_t attrbits;
5123 	int error;
5124 
5125 	NFSD_DEBUG(4, "in nfsrv_setattrdsdorpc\n");
5126 	nd = malloc(sizeof(*nd), M_TEMP, M_WAITOK | M_ZERO);
5127 	/*
5128 	 * Use a stateid where other is an alternating 01010 pattern and
5129 	 * seqid is 0xffffffff.  This value is not defined as special by
5130 	 * the RFC and is used by the FreeBSD NFS server to indicate an
5131 	 * MDS->DS proxy operation.
5132 	 */
5133 	st.other[0] = 0x55555555;
5134 	st.other[1] = 0x55555555;
5135 	st.other[2] = 0x55555555;
5136 	st.seqid = 0xffffffff;
5137 	nfscl_reqstart(nd, NFSPROC_SETATTR, nmp, (u_int8_t *)fhp, sizeof(*fhp),
5138 	    NULL, NULL, 0, 0);
5139 	nfsm_stateidtom(nd, &st, NFSSTATEID_PUTSTATEID);
5140 	nfscl_fillsattr(nd, &nap->na_vattr, vp, NFSSATTR_FULL, 0);
5141 
5142 	/* Do a Getattr for Size, Change, Access Time and Modify Time. */
5143 	NFSZERO_ATTRBIT(&attrbits);
5144 	NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_SIZE);
5145 	NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_CHANGE);
5146 	NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_TIMEACCESS);
5147 	NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_TIMEMODIFY);
5148 	NFSM_BUILD(tl, u_int32_t *, NFSX_UNSIGNED);
5149 	*tl = txdr_unsigned(NFSV4OP_GETATTR);
5150 	(void) nfsrv_putattrbit(nd, &attrbits);
5151 	error = newnfs_request(nd, nmp, NULL, &nmp->nm_sockreq, NULL, p, cred,
5152 	    NFS_PROG, NFS_VER4, NULL, 1, NULL, NULL);
5153 	if (error != 0) {
5154 		free(nd, M_TEMP);
5155 		return (error);
5156 	}
5157 	NFSD_DEBUG(4, "nfsrv_setattrdsdorpc: aft setattrrpc=%d\n",
5158 	    nd->nd_repstat);
5159 	/* Get rid of weak cache consistency data for now. */
5160 	if ((nd->nd_flag & (ND_NOMOREDATA | ND_NFSV4 | ND_V4WCCATTR)) ==
5161 	    (ND_NFSV4 | ND_V4WCCATTR)) {
5162 		error = nfsv4_loadattr(nd, NULL, dsnap, NULL, NULL, 0, NULL,
5163 		    NULL, NULL, NULL, NULL, 0, NULL, NULL, NULL, NULL, NULL);
5164 		NFSD_DEBUG(4, "nfsrv_setattrdsdorpc: wcc attr=%d\n", error);
5165 		if (error != 0)
5166 			goto nfsmout;
5167 		/*
5168 		 * Get rid of Op# and status for next op.
5169 		 */
5170 		NFSM_DISSECT(tl, uint32_t *, 2 * NFSX_UNSIGNED);
5171 		if (*++tl != 0)
5172 			nd->nd_flag |= ND_NOMOREDATA;
5173 	}
5174 	error = nfsrv_getattrbits(nd, &attrbits, NULL, NULL);
5175 	if (error != 0)
5176 		goto nfsmout;
5177 	if (nd->nd_repstat != 0)
5178 		error = nd->nd_repstat;
5179 	/*
5180 	 * Get the Change, Size, Access Time and Modify Time attributes and set
5181 	 * on the Metadata file, so its attributes will be what the file's
5182 	 * would be if it had been written.
5183 	 */
5184 	if (error == 0) {
5185 		NFSM_DISSECT(tl, uint32_t *, 2 * NFSX_UNSIGNED);
5186 		error = nfsv4_loadattr(nd, NULL, dsnap, NULL, NULL, 0, NULL,
5187 		    NULL, NULL, NULL, NULL, 0, NULL, NULL, NULL, NULL, NULL);
5188 	}
5189 	NFSD_DEBUG(4, "nfsrv_setattrdsdorpc: aft setattr loadattr=%d\n", error);
5190 nfsmout:
5191 	m_freem(nd->nd_mrep);
5192 	free(nd, M_TEMP);
5193 	NFSD_DEBUG(4, "nfsrv_setattrdsdorpc error=%d\n", error);
5194 	return (error);
5195 }
5196 
5197 struct nfsrvsetattrdsdorpc {
5198 	int			done;
5199 	int			inprog;
5200 	struct task		tsk;
5201 	fhandle_t		fh;
5202 	struct nfsmount		*nmp;
5203 	struct vnode		*vp;
5204 	struct ucred		*cred;
5205 	NFSPROC_T		*p;
5206 	struct nfsvattr		na;
5207 	struct nfsvattr		dsna;
5208 	int			err;
5209 };
5210 
5211 /*
5212  * Start up the thread that will execute nfsrv_setattrdsdorpc().
5213  */
5214 static void
5215 start_setattrdsdorpc(void *arg, int pending)
5216 {
5217 	struct nfsrvsetattrdsdorpc *drpc;
5218 
5219 	drpc = (struct nfsrvsetattrdsdorpc *)arg;
5220 	drpc->err = nfsrv_setattrdsdorpc(&drpc->fh, drpc->cred, drpc->p,
5221 	    drpc->vp, drpc->nmp, &drpc->na, &drpc->dsna);
5222 	drpc->done = 1;
5223 }
5224 
5225 static int
5226 nfsrv_setattrdsrpc(fhandle_t *fhp, struct ucred *cred, NFSPROC_T *p,
5227     struct vnode *vp, struct nfsmount **nmpp, int mirrorcnt,
5228     struct nfsvattr *nap, int *failposp)
5229 {
5230 	struct nfsrvsetattrdsdorpc *drpc, *tdrpc;
5231 	struct nfsvattr na;
5232 	int error, i, ret, timo;
5233 
5234 	NFSD_DEBUG(4, "in nfsrv_setattrdsrpc\n");
5235 	drpc = NULL;
5236 	if (mirrorcnt > 1)
5237 		tdrpc = drpc = malloc(sizeof(*drpc) * (mirrorcnt - 1), M_TEMP,
5238 		    M_WAITOK);
5239 
5240 	/*
5241 	 * Do the setattr RPC for every DS, using a separate kernel process
5242 	 * for every DS except the last one.
5243 	 */
5244 	error = 0;
5245 	for (i = 0; i < mirrorcnt - 1; i++, tdrpc++) {
5246 		tdrpc->done = 0;
5247 		tdrpc->inprog = 0;
5248 		NFSBCOPY(fhp, &tdrpc->fh, sizeof(*fhp));
5249 		tdrpc->nmp = *nmpp;
5250 		tdrpc->vp = vp;
5251 		tdrpc->cred = cred;
5252 		tdrpc->p = p;
5253 		tdrpc->na = *nap;
5254 		tdrpc->err = 0;
5255 		ret = EIO;
5256 		if (nfs_pnfsiothreads != 0) {
5257 			ret = nfs_pnfsio(start_setattrdsdorpc, tdrpc);
5258 			NFSD_DEBUG(4, "nfsrv_setattrdsrpc: nfs_pnfsio=%d\n",
5259 			    ret);
5260 		}
5261 		if (ret != 0) {
5262 			ret = nfsrv_setattrdsdorpc(fhp, cred, p, vp, *nmpp, nap,
5263 			    &na);
5264 			if (nfsds_failerr(ret) && *failposp == -1)
5265 				*failposp = i;
5266 			else if (error == 0 && ret != 0)
5267 				error = ret;
5268 		}
5269 		nmpp++;
5270 		fhp++;
5271 	}
5272 	ret = nfsrv_setattrdsdorpc(fhp, cred, p, vp, *nmpp, nap, &na);
5273 	if (nfsds_failerr(ret) && *failposp == -1 && mirrorcnt > 1)
5274 		*failposp = mirrorcnt - 1;
5275 	else if (error == 0 && ret != 0)
5276 		error = ret;
5277 	if (error == 0)
5278 		error = nfsrv_setextattr(vp, &na, p);
5279 	NFSD_DEBUG(4, "nfsrv_setattrdsrpc: aft setextat=%d\n", error);
5280 	tdrpc = drpc;
5281 	timo = hz / 50;		/* Wait for 20msec. */
5282 	if (timo < 1)
5283 		timo = 1;
5284 	for (i = 0; i < mirrorcnt - 1; i++, tdrpc++) {
5285 		/* Wait for RPCs on separate threads to complete. */
5286 		while (tdrpc->inprog != 0 && tdrpc->done == 0)
5287 			tsleep(&tdrpc->tsk, PVFS, "srvsads", timo);
5288 		if (nfsds_failerr(tdrpc->err) && *failposp == -1)
5289 			*failposp = i;
5290 		else if (error == 0 && tdrpc->err != 0)
5291 			error = tdrpc->err;
5292 	}
5293 	free(drpc, M_TEMP);
5294 	return (error);
5295 }
5296 
5297 /*
5298  * Do a Setattr of an NFSv4 ACL on the DS file.
5299  */
5300 static int
5301 nfsrv_setacldsdorpc(fhandle_t *fhp, struct ucred *cred, NFSPROC_T *p,
5302     struct vnode *vp, struct nfsmount *nmp, struct acl *aclp)
5303 {
5304 	struct nfsrv_descript *nd;
5305 	nfsv4stateid_t st;
5306 	nfsattrbit_t attrbits;
5307 	int error;
5308 
5309 	NFSD_DEBUG(4, "in nfsrv_setacldsdorpc\n");
5310 	nd = malloc(sizeof(*nd), M_TEMP, M_WAITOK | M_ZERO);
5311 	/*
5312 	 * Use a stateid where other is an alternating 01010 pattern and
5313 	 * seqid is 0xffffffff.  This value is not defined as special by
5314 	 * the RFC and is used by the FreeBSD NFS server to indicate an
5315 	 * MDS->DS proxy operation.
5316 	 */
5317 	st.other[0] = 0x55555555;
5318 	st.other[1] = 0x55555555;
5319 	st.other[2] = 0x55555555;
5320 	st.seqid = 0xffffffff;
5321 	nfscl_reqstart(nd, NFSPROC_SETACL, nmp, (u_int8_t *)fhp, sizeof(*fhp),
5322 	    NULL, NULL, 0, 0);
5323 	nfsm_stateidtom(nd, &st, NFSSTATEID_PUTSTATEID);
5324 	NFSZERO_ATTRBIT(&attrbits);
5325 	NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_ACL);
5326 	/*
5327 	 * The "vp" argument to nfsv4_fillattr() is only used for vnode_type(),
5328 	 * so passing in the metadata "vp" will be ok, since it is of
5329 	 * the same type (VREG).
5330 	 */
5331 	nfsv4_fillattr(nd, NULL, vp, aclp, NULL, NULL, 0, &attrbits, NULL,
5332 	    NULL, 0, 0, 0, 0, 0, NULL);
5333 	error = newnfs_request(nd, nmp, NULL, &nmp->nm_sockreq, NULL, p, cred,
5334 	    NFS_PROG, NFS_VER4, NULL, 1, NULL, NULL);
5335 	if (error != 0) {
5336 		free(nd, M_TEMP);
5337 		return (error);
5338 	}
5339 	NFSD_DEBUG(4, "nfsrv_setacldsdorpc: aft setaclrpc=%d\n",
5340 	    nd->nd_repstat);
5341 	error = nd->nd_repstat;
5342 	m_freem(nd->nd_mrep);
5343 	free(nd, M_TEMP);
5344 	return (error);
5345 }
5346 
5347 struct nfsrvsetacldsdorpc {
5348 	int			done;
5349 	int			inprog;
5350 	struct task		tsk;
5351 	fhandle_t		fh;
5352 	struct nfsmount		*nmp;
5353 	struct vnode		*vp;
5354 	struct ucred		*cred;
5355 	NFSPROC_T		*p;
5356 	struct acl		*aclp;
5357 	int			err;
5358 };
5359 
5360 /*
5361  * Start up the thread that will execute nfsrv_setacldsdorpc().
5362  */
5363 static void
5364 start_setacldsdorpc(void *arg, int pending)
5365 {
5366 	struct nfsrvsetacldsdorpc *drpc;
5367 
5368 	drpc = (struct nfsrvsetacldsdorpc *)arg;
5369 	drpc->err = nfsrv_setacldsdorpc(&drpc->fh, drpc->cred, drpc->p,
5370 	    drpc->vp, drpc->nmp, drpc->aclp);
5371 	drpc->done = 1;
5372 }
5373 
5374 static int
5375 nfsrv_setacldsrpc(fhandle_t *fhp, struct ucred *cred, NFSPROC_T *p,
5376     struct vnode *vp, struct nfsmount **nmpp, int mirrorcnt, struct acl *aclp,
5377     int *failposp)
5378 {
5379 	struct nfsrvsetacldsdorpc *drpc, *tdrpc;
5380 	int error, i, ret, timo;
5381 
5382 	NFSD_DEBUG(4, "in nfsrv_setacldsrpc\n");
5383 	drpc = NULL;
5384 	if (mirrorcnt > 1)
5385 		tdrpc = drpc = malloc(sizeof(*drpc) * (mirrorcnt - 1), M_TEMP,
5386 		    M_WAITOK);
5387 
5388 	/*
5389 	 * Do the setattr RPC for every DS, using a separate kernel process
5390 	 * for every DS except the last one.
5391 	 */
5392 	error = 0;
5393 	for (i = 0; i < mirrorcnt - 1; i++, tdrpc++) {
5394 		tdrpc->done = 0;
5395 		tdrpc->inprog = 0;
5396 		NFSBCOPY(fhp, &tdrpc->fh, sizeof(*fhp));
5397 		tdrpc->nmp = *nmpp;
5398 		tdrpc->vp = vp;
5399 		tdrpc->cred = cred;
5400 		tdrpc->p = p;
5401 		tdrpc->aclp = aclp;
5402 		tdrpc->err = 0;
5403 		ret = EIO;
5404 		if (nfs_pnfsiothreads != 0) {
5405 			ret = nfs_pnfsio(start_setacldsdorpc, tdrpc);
5406 			NFSD_DEBUG(4, "nfsrv_setacldsrpc: nfs_pnfsio=%d\n",
5407 			    ret);
5408 		}
5409 		if (ret != 0) {
5410 			ret = nfsrv_setacldsdorpc(fhp, cred, p, vp, *nmpp,
5411 			    aclp);
5412 			if (nfsds_failerr(ret) && *failposp == -1)
5413 				*failposp = i;
5414 			else if (error == 0 && ret != 0)
5415 				error = ret;
5416 		}
5417 		nmpp++;
5418 		fhp++;
5419 	}
5420 	ret = nfsrv_setacldsdorpc(fhp, cred, p, vp, *nmpp, aclp);
5421 	if (nfsds_failerr(ret) && *failposp == -1 && mirrorcnt > 1)
5422 		*failposp = mirrorcnt - 1;
5423 	else if (error == 0 && ret != 0)
5424 		error = ret;
5425 	NFSD_DEBUG(4, "nfsrv_setacldsrpc: aft setextat=%d\n", error);
5426 	tdrpc = drpc;
5427 	timo = hz / 50;		/* Wait for 20msec. */
5428 	if (timo < 1)
5429 		timo = 1;
5430 	for (i = 0; i < mirrorcnt - 1; i++, tdrpc++) {
5431 		/* Wait for RPCs on separate threads to complete. */
5432 		while (tdrpc->inprog != 0 && tdrpc->done == 0)
5433 			tsleep(&tdrpc->tsk, PVFS, "srvacds", timo);
5434 		if (nfsds_failerr(tdrpc->err) && *failposp == -1)
5435 			*failposp = i;
5436 		else if (error == 0 && tdrpc->err != 0)
5437 			error = tdrpc->err;
5438 	}
5439 	free(drpc, M_TEMP);
5440 	return (error);
5441 }
5442 
5443 /*
5444  * Getattr call to the DS for the Modify, Size and Change attributes.
5445  */
5446 static int
5447 nfsrv_getattrdsrpc(fhandle_t *fhp, struct ucred *cred, NFSPROC_T *p,
5448     struct vnode *vp, struct nfsmount *nmp, struct nfsvattr *nap)
5449 {
5450 	struct nfsrv_descript *nd;
5451 	int error;
5452 	nfsattrbit_t attrbits;
5453 
5454 	NFSD_DEBUG(4, "in nfsrv_getattrdsrpc\n");
5455 	nd = malloc(sizeof(*nd), M_TEMP, M_WAITOK | M_ZERO);
5456 	nfscl_reqstart(nd, NFSPROC_GETATTR, nmp, (u_int8_t *)fhp,
5457 	    sizeof(fhandle_t), NULL, NULL, 0, 0);
5458 	NFSZERO_ATTRBIT(&attrbits);
5459 	NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_SIZE);
5460 	NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_CHANGE);
5461 	NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_TIMEACCESS);
5462 	NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_TIMEMODIFY);
5463 	(void) nfsrv_putattrbit(nd, &attrbits);
5464 	error = newnfs_request(nd, nmp, NULL, &nmp->nm_sockreq, NULL, p, cred,
5465 	    NFS_PROG, NFS_VER4, NULL, 1, NULL, NULL);
5466 	if (error != 0) {
5467 		free(nd, M_TEMP);
5468 		return (error);
5469 	}
5470 	NFSD_DEBUG(4, "nfsrv_getattrdsrpc: aft getattrrpc=%d\n",
5471 	    nd->nd_repstat);
5472 	if (nd->nd_repstat == 0) {
5473 		error = nfsv4_loadattr(nd, NULL, nap, NULL, NULL, 0,
5474 		    NULL, NULL, NULL, NULL, NULL, 0, NULL, NULL, NULL,
5475 		    NULL, NULL);
5476 		/*
5477 		 * We can only save the updated values in the extended
5478 		 * attribute if the vp is exclusively locked.
5479 		 * This should happen when any of the following operations
5480 		 * occur on the vnode:
5481 		 *    Close, Delegreturn, LayoutCommit, LayoutReturn
5482 		 * As such, the updated extended attribute should get saved
5483 		 * before nfsrv_checkdsattr() returns 0 and allows the cached
5484 		 * attributes to be returned without calling this function.
5485 		 */
5486 		if (error == 0 && VOP_ISLOCKED(vp) == LK_EXCLUSIVE) {
5487 			error = nfsrv_setextattr(vp, nap, p);
5488 			NFSD_DEBUG(4, "nfsrv_getattrdsrpc: aft setextat=%d\n",
5489 			    error);
5490 		}
5491 	} else
5492 		error = nd->nd_repstat;
5493 	m_freem(nd->nd_mrep);
5494 	free(nd, M_TEMP);
5495 	NFSD_DEBUG(4, "nfsrv_getattrdsrpc error=%d\n", error);
5496 	return (error);
5497 }
5498 
5499 /*
5500  * Get the device id and file handle for a DS file.
5501  */
5502 int
5503 nfsrv_dsgetdevandfh(struct vnode *vp, NFSPROC_T *p, int *mirrorcntp,
5504     fhandle_t *fhp, char *devid)
5505 {
5506 	int buflen, error;
5507 	char *buf;
5508 
5509 	buflen = 1024;
5510 	buf = malloc(buflen, M_TEMP, M_WAITOK);
5511 	error = nfsrv_dsgetsockmnt(vp, 0, buf, &buflen, mirrorcntp, p, NULL,
5512 	    fhp, devid, NULL, NULL, NULL, NULL, NULL, NULL);
5513 	free(buf, M_TEMP);
5514 	return (error);
5515 }
5516 
5517 /*
5518  * Do a Lookup against the DS for the filename.
5519  */
5520 static int
5521 nfsrv_pnfslookupds(struct vnode *vp, struct vnode *dvp, struct pnfsdsfile *pf,
5522     struct vnode **nvpp, NFSPROC_T *p)
5523 {
5524 	struct nameidata named;
5525 	struct ucred *tcred;
5526 	char *bufp;
5527 	u_long *hashp;
5528 	struct vnode *nvp;
5529 	int error;
5530 
5531 	tcred = newnfs_getcred();
5532 	named.ni_cnd.cn_nameiop = LOOKUP;
5533 	named.ni_cnd.cn_lkflags = LK_SHARED | LK_RETRY;
5534 	named.ni_cnd.cn_cred = tcred;
5535 	named.ni_cnd.cn_thread = p;
5536 	named.ni_cnd.cn_flags = ISLASTCN | LOCKPARENT | LOCKLEAF | SAVENAME;
5537 	nfsvno_setpathbuf(&named, &bufp, &hashp);
5538 	named.ni_cnd.cn_nameptr = bufp;
5539 	named.ni_cnd.cn_namelen = strlen(pf->dsf_filename);
5540 	strlcpy(bufp, pf->dsf_filename, NAME_MAX);
5541 	NFSD_DEBUG(4, "nfsrv_pnfslookupds: filename=%s\n", bufp);
5542 	error = VOP_LOOKUP(dvp, &nvp, &named.ni_cnd);
5543 	NFSD_DEBUG(4, "nfsrv_pnfslookupds: aft LOOKUP=%d\n", error);
5544 	NFSFREECRED(tcred);
5545 	nfsvno_relpathbuf(&named);
5546 	if (error == 0)
5547 		*nvpp = nvp;
5548 	NFSD_DEBUG(4, "eo nfsrv_pnfslookupds=%d\n", error);
5549 	return (error);
5550 }
5551 
5552 /*
5553  * Set the file handle to the correct one.
5554  */
5555 static void
5556 nfsrv_pnfssetfh(struct vnode *vp, struct pnfsdsfile *pf, char *devid,
5557     char *fnamep, struct vnode *nvp, NFSPROC_T *p)
5558 {
5559 	struct nfsnode *np;
5560 	int ret;
5561 
5562 	np = VTONFS(nvp);
5563 	NFSBCOPY(np->n_fhp->nfh_fh, &pf->dsf_fh, NFSX_MYFH);
5564 	/*
5565 	 * We can only do a vn_set_extattr() if the vnode is exclusively
5566 	 * locked and vn_start_write() has been done.  If devid != NULL or
5567 	 * fnamep != NULL or the vnode is shared locked, vn_start_write()
5568 	 * may not have been done.
5569 	 * If not done now, it will be done on a future call.
5570 	 */
5571 	if (devid == NULL && fnamep == NULL && NFSVOPISLOCKED(vp) ==
5572 	    LK_EXCLUSIVE)
5573 		ret = vn_extattr_set(vp, IO_NODELOCKED,
5574 		    EXTATTR_NAMESPACE_SYSTEM, "pnfsd.dsfile", sizeof(*pf),
5575 		    (char *)pf, p);
5576 	NFSD_DEBUG(4, "eo nfsrv_pnfssetfh=%d\n", ret);
5577 }
5578 
5579 /*
5580  * Cause RPCs waiting on "nmp" to fail.  This is called for a DS mount point
5581  * when the DS has failed.
5582  */
5583 void
5584 nfsrv_killrpcs(struct nfsmount *nmp)
5585 {
5586 
5587 	/*
5588 	 * Call newnfs_nmcancelreqs() to cause
5589 	 * any RPCs in progress on the mount point to
5590 	 * fail.
5591 	 * This will cause any process waiting for an
5592 	 * RPC to complete while holding a vnode lock
5593 	 * on the mounted-on vnode (such as "df" or
5594 	 * a non-forced "umount") to fail.
5595 	 * This will unlock the mounted-on vnode so
5596 	 * a forced dismount can succeed.
5597 	 * The NFSMNTP_CANCELRPCS flag should be set when this function is
5598 	 * called.
5599 	 */
5600 	newnfs_nmcancelreqs(nmp);
5601 }
5602 
5603 /*
5604  * Sum up the statfs info for each of the DSs, so that the client will
5605  * receive the total for all DSs.
5606  */
5607 static int
5608 nfsrv_pnfsstatfs(struct statfs *sf, struct mount *mp)
5609 {
5610 	struct statfs *tsf;
5611 	struct nfsdevice *ds;
5612 	struct vnode **dvpp, **tdvpp, *dvp;
5613 	uint64_t tot;
5614 	int cnt, error = 0, i;
5615 
5616 	if (nfsrv_devidcnt <= 0)
5617 		return (ENXIO);
5618 	dvpp = mallocarray(nfsrv_devidcnt, sizeof(*dvpp), M_TEMP, M_WAITOK);
5619 	tsf = malloc(sizeof(*tsf), M_TEMP, M_WAITOK);
5620 
5621 	/* Get an array of the dvps for the DSs. */
5622 	tdvpp = dvpp;
5623 	i = 0;
5624 	NFSDDSLOCK();
5625 	/* First, search for matches for same file system. */
5626 	TAILQ_FOREACH(ds, &nfsrv_devidhead, nfsdev_list) {
5627 		if (ds->nfsdev_nmp != NULL && ds->nfsdev_mdsisset != 0 &&
5628 		    ds->nfsdev_mdsfsid.val[0] == mp->mnt_stat.f_fsid.val[0] &&
5629 		    ds->nfsdev_mdsfsid.val[1] == mp->mnt_stat.f_fsid.val[1]) {
5630 			if (++i > nfsrv_devidcnt)
5631 				break;
5632 			*tdvpp++ = ds->nfsdev_dvp;
5633 		}
5634 	}
5635 	/*
5636 	 * If no matches for same file system, total all servers not assigned
5637 	 * to a file system.
5638 	 */
5639 	if (i == 0) {
5640 		TAILQ_FOREACH(ds, &nfsrv_devidhead, nfsdev_list) {
5641 			if (ds->nfsdev_nmp != NULL &&
5642 			    ds->nfsdev_mdsisset == 0) {
5643 				if (++i > nfsrv_devidcnt)
5644 					break;
5645 				*tdvpp++ = ds->nfsdev_dvp;
5646 			}
5647 		}
5648 	}
5649 	NFSDDSUNLOCK();
5650 	cnt = i;
5651 
5652 	/* Do a VFS_STATFS() for each of the DSs and sum them up. */
5653 	tdvpp = dvpp;
5654 	for (i = 0; i < cnt && error == 0; i++) {
5655 		dvp = *tdvpp++;
5656 		error = VFS_STATFS(dvp->v_mount, tsf);
5657 		if (error == 0) {
5658 			if (sf->f_bsize == 0) {
5659 				if (tsf->f_bsize > 0)
5660 					sf->f_bsize = tsf->f_bsize;
5661 				else
5662 					sf->f_bsize = 8192;
5663 			}
5664 			if (tsf->f_blocks > 0) {
5665 				if (sf->f_bsize != tsf->f_bsize) {
5666 					tot = tsf->f_blocks * tsf->f_bsize;
5667 					sf->f_blocks += (tot / sf->f_bsize);
5668 				} else
5669 					sf->f_blocks += tsf->f_blocks;
5670 			}
5671 			if (tsf->f_bfree > 0) {
5672 				if (sf->f_bsize != tsf->f_bsize) {
5673 					tot = tsf->f_bfree * tsf->f_bsize;
5674 					sf->f_bfree += (tot / sf->f_bsize);
5675 				} else
5676 					sf->f_bfree += tsf->f_bfree;
5677 			}
5678 			if (tsf->f_bavail > 0) {
5679 				if (sf->f_bsize != tsf->f_bsize) {
5680 					tot = tsf->f_bavail * tsf->f_bsize;
5681 					sf->f_bavail += (tot / sf->f_bsize);
5682 				} else
5683 					sf->f_bavail += tsf->f_bavail;
5684 			}
5685 		}
5686 	}
5687 	free(tsf, M_TEMP);
5688 	free(dvpp, M_TEMP);
5689 	return (error);
5690 }
5691 
5692 /*
5693  * Set an NFSv4 acl.
5694  */
5695 int
5696 nfsrv_setacl(struct vnode *vp, NFSACL_T *aclp, struct ucred *cred, NFSPROC_T *p)
5697 {
5698 	int error;
5699 
5700 	if (nfsrv_useacl == 0 || nfs_supportsnfsv4acls(vp) == 0) {
5701 		error = NFSERR_ATTRNOTSUPP;
5702 		goto out;
5703 	}
5704 	/*
5705 	 * With NFSv4 ACLs, chmod(2) may need to add additional entries.
5706 	 * Make sure it has enough room for that - splitting every entry
5707 	 * into two and appending "canonical six" entries at the end.
5708 	 * Cribbed out of kern/vfs_acl.c - Rick M.
5709 	 */
5710 	if (aclp->acl_cnt > (ACL_MAX_ENTRIES - 6) / 2) {
5711 		error = NFSERR_ATTRNOTSUPP;
5712 		goto out;
5713 	}
5714 	error = VOP_SETACL(vp, ACL_TYPE_NFS4, aclp, cred, p);
5715 	if (error == 0) {
5716 		error = nfsrv_dssetacl(vp, aclp, cred, p);
5717 		if (error == ENOENT)
5718 			error = 0;
5719 	}
5720 
5721 out:
5722 	NFSEXITCODE(error);
5723 	return (error);
5724 }
5725 
5726 extern int (*nfsd_call_nfsd)(struct thread *, struct nfssvc_args *);
5727 
5728 /*
5729  * Called once to initialize data structures...
5730  */
5731 static int
5732 nfsd_modevent(module_t mod, int type, void *data)
5733 {
5734 	int error = 0, i;
5735 	static int loaded = 0;
5736 
5737 	switch (type) {
5738 	case MOD_LOAD:
5739 		if (loaded)
5740 			goto out;
5741 		newnfs_portinit();
5742 		for (i = 0; i < NFSRVCACHE_HASHSIZE; i++) {
5743 			mtx_init(&nfsrchash_table[i].mtx, "nfsrtc", NULL,
5744 			    MTX_DEF);
5745 			mtx_init(&nfsrcahash_table[i].mtx, "nfsrtca", NULL,
5746 			    MTX_DEF);
5747 		}
5748 		mtx_init(&nfsrc_udpmtx, "nfsuc", NULL, MTX_DEF);
5749 		mtx_init(&nfs_v4root_mutex, "nfs4rt", NULL, MTX_DEF);
5750 		mtx_init(&nfsv4root_mnt.mnt_mtx, "nfs4mnt", NULL, MTX_DEF);
5751 		mtx_init(&nfsrv_dontlistlock_mtx, "nfs4dnl", NULL, MTX_DEF);
5752 		mtx_init(&nfsrv_recalllock_mtx, "nfs4rec", NULL, MTX_DEF);
5753 		lockinit(&nfsv4root_mnt.mnt_explock, PVFS, "explock", 0, 0);
5754 		nfsrvd_initcache();
5755 		nfsd_init();
5756 		NFSD_LOCK();
5757 		nfsrvd_init(0);
5758 		NFSD_UNLOCK();
5759 		nfsd_mntinit();
5760 #ifdef VV_DISABLEDELEG
5761 		vn_deleg_ops.vndeleg_recall = nfsd_recalldelegation;
5762 		vn_deleg_ops.vndeleg_disable = nfsd_disabledelegation;
5763 #endif
5764 		nfsd_call_servertimer = nfsrv_servertimer;
5765 		nfsd_call_nfsd = nfssvc_nfsd;
5766 		loaded = 1;
5767 		break;
5768 
5769 	case MOD_UNLOAD:
5770 		if (newnfs_numnfsd != 0) {
5771 			error = EBUSY;
5772 			break;
5773 		}
5774 
5775 #ifdef VV_DISABLEDELEG
5776 		vn_deleg_ops.vndeleg_recall = NULL;
5777 		vn_deleg_ops.vndeleg_disable = NULL;
5778 #endif
5779 		nfsd_call_servertimer = NULL;
5780 		nfsd_call_nfsd = NULL;
5781 
5782 		/* Clean out all NFSv4 state. */
5783 		nfsrv_throwawayallstate(curthread);
5784 
5785 		/* Clean the NFS server reply cache */
5786 		nfsrvd_cleancache();
5787 
5788 		/* Free up the krpc server pool. */
5789 		if (nfsrvd_pool != NULL)
5790 			svcpool_destroy(nfsrvd_pool);
5791 
5792 		/* and get rid of the locks */
5793 		for (i = 0; i < NFSRVCACHE_HASHSIZE; i++) {
5794 			mtx_destroy(&nfsrchash_table[i].mtx);
5795 			mtx_destroy(&nfsrcahash_table[i].mtx);
5796 		}
5797 		mtx_destroy(&nfsrc_udpmtx);
5798 		mtx_destroy(&nfs_v4root_mutex);
5799 		mtx_destroy(&nfsv4root_mnt.mnt_mtx);
5800 		mtx_destroy(&nfsrv_dontlistlock_mtx);
5801 		mtx_destroy(&nfsrv_recalllock_mtx);
5802 		for (i = 0; i < nfsrv_sessionhashsize; i++)
5803 			mtx_destroy(&nfssessionhash[i].mtx);
5804 		if (nfslayouthash != NULL) {
5805 			for (i = 0; i < nfsrv_layouthashsize; i++)
5806 				mtx_destroy(&nfslayouthash[i].mtx);
5807 			free(nfslayouthash, M_NFSDSESSION);
5808 		}
5809 		lockdestroy(&nfsv4root_mnt.mnt_explock);
5810 		free(nfsclienthash, M_NFSDCLIENT);
5811 		free(nfslockhash, M_NFSDLOCKFILE);
5812 		free(nfssessionhash, M_NFSDSESSION);
5813 		loaded = 0;
5814 		break;
5815 	default:
5816 		error = EOPNOTSUPP;
5817 		break;
5818 	}
5819 
5820 out:
5821 	NFSEXITCODE(error);
5822 	return (error);
5823 }
5824 static moduledata_t nfsd_mod = {
5825 	"nfsd",
5826 	nfsd_modevent,
5827 	NULL,
5828 };
5829 DECLARE_MODULE(nfsd, nfsd_mod, SI_SUB_VFS, SI_ORDER_ANY);
5830 
5831 /* So that loader and kldload(2) can find us, wherever we are.. */
5832 MODULE_VERSION(nfsd, 1);
5833 MODULE_DEPEND(nfsd, nfscommon, 1, 1, 1);
5834 MODULE_DEPEND(nfsd, nfslock, 1, 1, 1);
5835 MODULE_DEPEND(nfsd, nfslockd, 1, 1, 1);
5836 MODULE_DEPEND(nfsd, krpc, 1, 1, 1);
5837 MODULE_DEPEND(nfsd, nfssvc, 1, 1, 1);
5838 
5839