xref: /freebsd/sys/fs/nfsserver/nfs_nfsdport.c (revision 3ff01b231dfa83d518854c63e7c9cd1debd1139e)
1 /*-
2  * SPDX-License-Identifier: BSD-3-Clause
3  *
4  * Copyright (c) 1989, 1993
5  *	The Regents of the University of California.  All rights reserved.
6  *
7  * This code is derived from software contributed to Berkeley by
8  * Rick Macklem at The University of Guelph.
9  *
10  * Redistribution and use in source and binary forms, with or without
11  * modification, are permitted provided that the following conditions
12  * are met:
13  * 1. Redistributions of source code must retain the above copyright
14  *    notice, this list of conditions and the following disclaimer.
15  * 2. Redistributions in binary form must reproduce the above copyright
16  *    notice, this list of conditions and the following disclaimer in the
17  *    documentation and/or other materials provided with the distribution.
18  * 3. Neither the name of the University nor the names of its contributors
19  *    may be used to endorse or promote products derived from this software
20  *    without specific prior written permission.
21  *
22  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
23  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
24  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
25  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
26  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
27  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
28  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
29  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
30  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
31  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
32  * SUCH DAMAGE.
33  *
34  */
35 
36 #include <sys/cdefs.h>
37 __FBSDID("$FreeBSD$");
38 
39 #include <sys/capsicum.h>
40 #include <sys/extattr.h>
41 
42 /*
43  * Functions that perform the vfs operations required by the routines in
44  * nfsd_serv.c. It is hoped that this change will make the server more
45  * portable.
46  */
47 
48 #include <fs/nfs/nfsport.h>
49 #include <security/mac/mac_framework.h>
50 #include <sys/filio.h>
51 #include <sys/hash.h>
52 #include <sys/sysctl.h>
53 #include <nlm/nlm_prot.h>
54 #include <nlm/nlm.h>
55 
56 FEATURE(nfsd, "NFSv4 server");
57 
58 extern u_int32_t newnfs_true, newnfs_false, newnfs_xdrneg1;
59 extern int nfsrv_useacl;
60 extern int newnfs_numnfsd;
61 extern struct mount nfsv4root_mnt;
62 extern struct nfsrv_stablefirst nfsrv_stablefirst;
63 extern void (*nfsd_call_servertimer)(void);
64 extern SVCPOOL	*nfsrvd_pool;
65 extern struct nfsv4lock nfsd_suspend_lock;
66 extern struct nfsclienthashhead *nfsclienthash;
67 extern struct nfslockhashhead *nfslockhash;
68 extern struct nfssessionhash *nfssessionhash;
69 extern int nfsrv_sessionhashsize;
70 extern struct nfsstatsv1 nfsstatsv1;
71 extern struct nfslayouthash *nfslayouthash;
72 extern int nfsrv_layouthashsize;
73 extern struct mtx nfsrv_dslock_mtx;
74 extern int nfs_pnfsiothreads;
75 extern struct nfsdontlisthead nfsrv_dontlisthead;
76 extern volatile int nfsrv_dontlistlen;
77 extern volatile int nfsrv_devidcnt;
78 extern int nfsrv_maxpnfsmirror;
79 struct vfsoptlist nfsv4root_opt, nfsv4root_newopt;
80 NFSDLOCKMUTEX;
81 NFSSTATESPINLOCK;
82 struct nfsrchash_bucket nfsrchash_table[NFSRVCACHE_HASHSIZE];
83 struct nfsrchash_bucket nfsrcahash_table[NFSRVCACHE_HASHSIZE];
84 struct mtx nfsrc_udpmtx;
85 struct mtx nfs_v4root_mutex;
86 struct mtx nfsrv_dontlistlock_mtx;
87 struct mtx nfsrv_recalllock_mtx;
88 struct nfsrvfh nfs_rootfh, nfs_pubfh;
89 int nfs_pubfhset = 0, nfs_rootfhset = 0;
90 struct proc *nfsd_master_proc = NULL;
91 int nfsd_debuglevel = 0;
92 static pid_t nfsd_master_pid = (pid_t)-1;
93 static char nfsd_master_comm[MAXCOMLEN + 1];
94 static struct timeval nfsd_master_start;
95 static uint32_t nfsv4_sysid = 0;
96 static fhandle_t zerofh;
97 
98 static int nfssvc_srvcall(struct thread *, struct nfssvc_args *,
99     struct ucred *);
100 
101 int nfsrv_enable_crossmntpt = 1;
102 static int nfs_commit_blks;
103 static int nfs_commit_miss;
104 extern int nfsrv_issuedelegs;
105 extern int nfsrv_dolocallocks;
106 extern int nfsd_enable_stringtouid;
107 extern struct nfsdevicehead nfsrv_devidhead;
108 
109 static int nfsrv_createiovec(int, struct mbuf **, struct mbuf **,
110     struct iovec **);
111 static int nfsrv_createiovec_extpgs(int, int, struct mbuf **,
112     struct mbuf **, struct iovec **);
113 static int nfsrv_createiovecw(int, struct mbuf *, char *, struct iovec **,
114     int *);
115 static void nfsrv_pnfscreate(struct vnode *, struct vattr *, struct ucred *,
116     NFSPROC_T *);
117 static void nfsrv_pnfsremovesetup(struct vnode *, NFSPROC_T *, struct vnode **,
118     int *, char *, fhandle_t *);
119 static void nfsrv_pnfsremove(struct vnode **, int, char *, fhandle_t *,
120     NFSPROC_T *);
121 static int nfsrv_proxyds(struct vnode *, off_t, int, struct ucred *,
122     struct thread *, int, struct mbuf **, char *, struct mbuf **,
123     struct nfsvattr *, struct acl *, off_t *, int, bool *);
124 static int nfsrv_setextattr(struct vnode *, struct nfsvattr *, NFSPROC_T *);
125 static int nfsrv_readdsrpc(fhandle_t *, off_t, int, struct ucred *,
126     NFSPROC_T *, struct nfsmount *, struct mbuf **, struct mbuf **);
127 static int nfsrv_writedsrpc(fhandle_t *, off_t, int, struct ucred *,
128     NFSPROC_T *, struct vnode *, struct nfsmount **, int, struct mbuf **,
129     char *, int *);
130 static int nfsrv_allocatedsrpc(fhandle_t *, off_t, off_t, struct ucred *,
131     NFSPROC_T *, struct vnode *, struct nfsmount **, int, int *);
132 static int nfsrv_setacldsrpc(fhandle_t *, struct ucred *, NFSPROC_T *,
133     struct vnode *, struct nfsmount **, int, struct acl *, int *);
134 static int nfsrv_setattrdsrpc(fhandle_t *, struct ucred *, NFSPROC_T *,
135     struct vnode *, struct nfsmount **, int, struct nfsvattr *, int *);
136 static int nfsrv_getattrdsrpc(fhandle_t *, struct ucred *, NFSPROC_T *,
137     struct vnode *, struct nfsmount *, struct nfsvattr *);
138 static int nfsrv_seekdsrpc(fhandle_t *, off_t *, int, bool *, struct ucred *,
139     NFSPROC_T *, struct nfsmount *);
140 static int nfsrv_putfhname(fhandle_t *, char *);
141 static int nfsrv_pnfslookupds(struct vnode *, struct vnode *,
142     struct pnfsdsfile *, struct vnode **, NFSPROC_T *);
143 static void nfsrv_pnfssetfh(struct vnode *, struct pnfsdsfile *, char *, char *,
144     struct vnode *, NFSPROC_T *);
145 static int nfsrv_dsremove(struct vnode *, char *, struct ucred *, NFSPROC_T *);
146 static int nfsrv_dssetacl(struct vnode *, struct acl *, struct ucred *,
147     NFSPROC_T *);
148 static int nfsrv_pnfsstatfs(struct statfs *, struct mount *);
149 
150 int nfs_pnfsio(task_fn_t *, void *);
151 
152 SYSCTL_NODE(_vfs, OID_AUTO, nfsd, CTLFLAG_RW | CTLFLAG_MPSAFE, 0,
153     "NFS server");
154 SYSCTL_INT(_vfs_nfsd, OID_AUTO, mirrormnt, CTLFLAG_RW,
155     &nfsrv_enable_crossmntpt, 0, "Enable nfsd to cross mount points");
156 SYSCTL_INT(_vfs_nfsd, OID_AUTO, commit_blks, CTLFLAG_RW, &nfs_commit_blks,
157     0, "");
158 SYSCTL_INT(_vfs_nfsd, OID_AUTO, commit_miss, CTLFLAG_RW, &nfs_commit_miss,
159     0, "");
160 SYSCTL_INT(_vfs_nfsd, OID_AUTO, issue_delegations, CTLFLAG_RW,
161     &nfsrv_issuedelegs, 0, "Enable nfsd to issue delegations");
162 SYSCTL_INT(_vfs_nfsd, OID_AUTO, enable_locallocks, CTLFLAG_RW,
163     &nfsrv_dolocallocks, 0, "Enable nfsd to acquire local locks on files");
164 SYSCTL_INT(_vfs_nfsd, OID_AUTO, debuglevel, CTLFLAG_RW, &nfsd_debuglevel,
165     0, "Debug level for NFS server");
166 SYSCTL_INT(_vfs_nfsd, OID_AUTO, enable_stringtouid, CTLFLAG_RW,
167     &nfsd_enable_stringtouid, 0, "Enable nfsd to accept numeric owner_names");
168 static int nfsrv_pnfsgetdsattr = 1;
169 SYSCTL_INT(_vfs_nfsd, OID_AUTO, pnfsgetdsattr, CTLFLAG_RW,
170     &nfsrv_pnfsgetdsattr, 0, "When set getattr gets DS attributes via RPC");
171 
172 /*
173  * nfsrv_dsdirsize can only be increased and only when the nfsd threads are
174  * not running.
175  * The dsN subdirectories for the increased values must have been created
176  * on all DS servers before this increase is done.
177  */
178 u_int	nfsrv_dsdirsize = 20;
179 static int
180 sysctl_dsdirsize(SYSCTL_HANDLER_ARGS)
181 {
182 	int error, newdsdirsize;
183 
184 	newdsdirsize = nfsrv_dsdirsize;
185 	error = sysctl_handle_int(oidp, &newdsdirsize, 0, req);
186 	if (error != 0 || req->newptr == NULL)
187 		return (error);
188 	if (newdsdirsize <= nfsrv_dsdirsize || newdsdirsize > 10000 ||
189 	    newnfs_numnfsd != 0)
190 		return (EINVAL);
191 	nfsrv_dsdirsize = newdsdirsize;
192 	return (0);
193 }
194 SYSCTL_PROC(_vfs_nfsd, OID_AUTO, dsdirsize,
195     CTLTYPE_UINT | CTLFLAG_MPSAFE | CTLFLAG_RW, 0, sizeof(nfsrv_dsdirsize),
196     sysctl_dsdirsize, "IU", "Number of dsN subdirs on the DS servers");
197 
198 #define	MAX_REORDERED_RPC	16
199 #define	NUM_HEURISTIC		1031
200 #define	NHUSE_INIT		64
201 #define	NHUSE_INC		16
202 #define	NHUSE_MAX		2048
203 
204 static struct nfsheur {
205 	struct vnode *nh_vp;	/* vp to match (unreferenced pointer) */
206 	off_t nh_nextoff;	/* next offset for sequential detection */
207 	int nh_use;		/* use count for selection */
208 	int nh_seqcount;	/* heuristic */
209 } nfsheur[NUM_HEURISTIC];
210 
211 /*
212  * Heuristic to detect sequential operation.
213  */
214 static struct nfsheur *
215 nfsrv_sequential_heuristic(struct uio *uio, struct vnode *vp)
216 {
217 	struct nfsheur *nh;
218 	int hi, try;
219 
220 	/* Locate best candidate. */
221 	try = 32;
222 	hi = ((int)(vm_offset_t)vp / sizeof(struct vnode)) % NUM_HEURISTIC;
223 	nh = &nfsheur[hi];
224 	while (try--) {
225 		if (nfsheur[hi].nh_vp == vp) {
226 			nh = &nfsheur[hi];
227 			break;
228 		}
229 		if (nfsheur[hi].nh_use > 0)
230 			--nfsheur[hi].nh_use;
231 		hi = (hi + 1) % NUM_HEURISTIC;
232 		if (nfsheur[hi].nh_use < nh->nh_use)
233 			nh = &nfsheur[hi];
234 	}
235 
236 	/* Initialize hint if this is a new file. */
237 	if (nh->nh_vp != vp) {
238 		nh->nh_vp = vp;
239 		nh->nh_nextoff = uio->uio_offset;
240 		nh->nh_use = NHUSE_INIT;
241 		if (uio->uio_offset == 0)
242 			nh->nh_seqcount = 4;
243 		else
244 			nh->nh_seqcount = 1;
245 	}
246 
247 	/* Calculate heuristic. */
248 	if ((uio->uio_offset == 0 && nh->nh_seqcount > 0) ||
249 	    uio->uio_offset == nh->nh_nextoff) {
250 		/* See comments in vfs_vnops.c:sequential_heuristic(). */
251 		nh->nh_seqcount += howmany(uio->uio_resid, 16384);
252 		if (nh->nh_seqcount > IO_SEQMAX)
253 			nh->nh_seqcount = IO_SEQMAX;
254 	} else if (qabs(uio->uio_offset - nh->nh_nextoff) <= MAX_REORDERED_RPC *
255 	    imax(vp->v_mount->mnt_stat.f_iosize, uio->uio_resid)) {
256 		/* Probably a reordered RPC, leave seqcount alone. */
257 	} else if (nh->nh_seqcount > 1) {
258 		nh->nh_seqcount /= 2;
259 	} else {
260 		nh->nh_seqcount = 0;
261 	}
262 	nh->nh_use += NHUSE_INC;
263 	if (nh->nh_use > NHUSE_MAX)
264 		nh->nh_use = NHUSE_MAX;
265 	return (nh);
266 }
267 
268 /*
269  * Get attributes into nfsvattr structure.
270  */
271 int
272 nfsvno_getattr(struct vnode *vp, struct nfsvattr *nvap,
273     struct nfsrv_descript *nd, struct thread *p, int vpislocked,
274     nfsattrbit_t *attrbitp)
275 {
276 	int error, gotattr, lockedit = 0;
277 	struct nfsvattr na;
278 
279 	if (vpislocked == 0) {
280 		/*
281 		 * When vpislocked == 0, the vnode is either exclusively
282 		 * locked by this thread or not locked by this thread.
283 		 * As such, shared lock it, if not exclusively locked.
284 		 */
285 		if (NFSVOPISLOCKED(vp) != LK_EXCLUSIVE) {
286 			lockedit = 1;
287 			NFSVOPLOCK(vp, LK_SHARED | LK_RETRY);
288 		}
289 	}
290 
291 	/*
292 	 * Acquire the Change, Size, TimeAccess, TimeModify and SpaceUsed
293 	 * attributes, as required.
294 	 * This needs to be done for regular files if:
295 	 * - non-NFSv4 RPCs or
296 	 * - when attrbitp == NULL or
297 	 * - an NFSv4 RPC with any of the above attributes in attrbitp.
298 	 * A return of 0 for nfsrv_proxyds() indicates that it has acquired
299 	 * these attributes.  nfsrv_proxyds() will return an error if the
300 	 * server is not a pNFS one.
301 	 */
302 	gotattr = 0;
303 	if (vp->v_type == VREG && nfsrv_devidcnt > 0 && (attrbitp == NULL ||
304 	    (nd->nd_flag & ND_NFSV4) == 0 ||
305 	    NFSISSET_ATTRBIT(attrbitp, NFSATTRBIT_CHANGE) ||
306 	    NFSISSET_ATTRBIT(attrbitp, NFSATTRBIT_SIZE) ||
307 	    NFSISSET_ATTRBIT(attrbitp, NFSATTRBIT_TIMEACCESS) ||
308 	    NFSISSET_ATTRBIT(attrbitp, NFSATTRBIT_TIMEMODIFY) ||
309 	    NFSISSET_ATTRBIT(attrbitp, NFSATTRBIT_SPACEUSED))) {
310 		error = nfsrv_proxyds(vp, 0, 0, nd->nd_cred, p,
311 		    NFSPROC_GETATTR, NULL, NULL, NULL, &na, NULL, NULL, 0,
312 		    NULL);
313 		if (error == 0)
314 			gotattr = 1;
315 	}
316 
317 	error = VOP_GETATTR(vp, &nvap->na_vattr, nd->nd_cred);
318 	if (lockedit != 0)
319 		NFSVOPUNLOCK(vp);
320 
321 	/*
322 	 * If we got the Change, Size and Modify Time from the DS,
323 	 * replace them.
324 	 */
325 	if (gotattr != 0) {
326 		nvap->na_atime = na.na_atime;
327 		nvap->na_mtime = na.na_mtime;
328 		nvap->na_filerev = na.na_filerev;
329 		nvap->na_size = na.na_size;
330 		nvap->na_bytes = na.na_bytes;
331 	}
332 	NFSD_DEBUG(4, "nfsvno_getattr: gotattr=%d err=%d chg=%ju\n", gotattr,
333 	    error, (uintmax_t)na.na_filerev);
334 
335 	NFSEXITCODE(error);
336 	return (error);
337 }
338 
339 /*
340  * Get a file handle for a vnode.
341  */
342 int
343 nfsvno_getfh(struct vnode *vp, fhandle_t *fhp, struct thread *p)
344 {
345 	int error;
346 
347 	NFSBZERO((caddr_t)fhp, sizeof(fhandle_t));
348 	fhp->fh_fsid = vp->v_mount->mnt_stat.f_fsid;
349 	error = VOP_VPTOFH(vp, &fhp->fh_fid);
350 
351 	NFSEXITCODE(error);
352 	return (error);
353 }
354 
355 /*
356  * Perform access checking for vnodes obtained from file handles that would
357  * refer to files already opened by a Unix client. You cannot just use
358  * vn_writechk() and VOP_ACCESSX() for two reasons.
359  * 1 - You must check for exported rdonly as well as MNT_RDONLY for the write
360  *     case.
361  * 2 - The owner is to be given access irrespective of mode bits for some
362  *     operations, so that processes that chmod after opening a file don't
363  *     break.
364  */
365 int
366 nfsvno_accchk(struct vnode *vp, accmode_t accmode, struct ucred *cred,
367     struct nfsexstuff *exp, struct thread *p, int override, int vpislocked,
368     u_int32_t *supportedtypep)
369 {
370 	struct vattr vattr;
371 	int error = 0, getret = 0;
372 
373 	if (vpislocked == 0) {
374 		if (NFSVOPLOCK(vp, LK_SHARED) != 0) {
375 			error = EPERM;
376 			goto out;
377 		}
378 	}
379 	if (accmode & VWRITE) {
380 		/* Just vn_writechk() changed to check rdonly */
381 		/*
382 		 * Disallow write attempts on read-only file systems;
383 		 * unless the file is a socket or a block or character
384 		 * device resident on the file system.
385 		 */
386 		if (NFSVNO_EXRDONLY(exp) ||
387 		    (vp->v_mount->mnt_flag & MNT_RDONLY)) {
388 			switch (vp->v_type) {
389 			case VREG:
390 			case VDIR:
391 			case VLNK:
392 				error = EROFS;
393 			default:
394 				break;
395 			}
396 		}
397 		/*
398 		 * If there's shared text associated with
399 		 * the inode, try to free it up once.  If
400 		 * we fail, we can't allow writing.
401 		 */
402 		if (VOP_IS_TEXT(vp) && error == 0)
403 			error = ETXTBSY;
404 	}
405 	if (error != 0) {
406 		if (vpislocked == 0)
407 			NFSVOPUNLOCK(vp);
408 		goto out;
409 	}
410 
411 	/*
412 	 * Should the override still be applied when ACLs are enabled?
413 	 */
414 	error = VOP_ACCESSX(vp, accmode, cred, p);
415 	if (error != 0 && (accmode & (VDELETE | VDELETE_CHILD))) {
416 		/*
417 		 * Try again with VEXPLICIT_DENY, to see if the test for
418 		 * deletion is supported.
419 		 */
420 		error = VOP_ACCESSX(vp, accmode | VEXPLICIT_DENY, cred, p);
421 		if (error == 0) {
422 			if (vp->v_type == VDIR) {
423 				accmode &= ~(VDELETE | VDELETE_CHILD);
424 				accmode |= VWRITE;
425 				error = VOP_ACCESSX(vp, accmode, cred, p);
426 			} else if (supportedtypep != NULL) {
427 				*supportedtypep &= ~NFSACCESS_DELETE;
428 			}
429 		}
430 	}
431 
432 	/*
433 	 * Allow certain operations for the owner (reads and writes
434 	 * on files that are already open).
435 	 */
436 	if (override != NFSACCCHK_NOOVERRIDE &&
437 	    (error == EPERM || error == EACCES)) {
438 		if (cred->cr_uid == 0 && (override & NFSACCCHK_ALLOWROOT))
439 			error = 0;
440 		else if (override & NFSACCCHK_ALLOWOWNER) {
441 			getret = VOP_GETATTR(vp, &vattr, cred);
442 			if (getret == 0 && cred->cr_uid == vattr.va_uid)
443 				error = 0;
444 		}
445 	}
446 	if (vpislocked == 0)
447 		NFSVOPUNLOCK(vp);
448 
449 out:
450 	NFSEXITCODE(error);
451 	return (error);
452 }
453 
454 /*
455  * Set attribute(s) vnop.
456  */
457 int
458 nfsvno_setattr(struct vnode *vp, struct nfsvattr *nvap, struct ucred *cred,
459     struct thread *p, struct nfsexstuff *exp)
460 {
461 	u_quad_t savsize = 0;
462 	int error, savedit;
463 	time_t savbtime;
464 
465 	/*
466 	 * If this is an exported file system and a pNFS service is running,
467 	 * don't VOP_SETATTR() of size for the MDS file system.
468 	 */
469 	savedit = 0;
470 	error = 0;
471 	if (vp->v_type == VREG && (vp->v_mount->mnt_flag & MNT_EXPORTED) != 0 &&
472 	    nfsrv_devidcnt != 0 && nvap->na_vattr.va_size != VNOVAL &&
473 	    nvap->na_vattr.va_size > 0) {
474 		savsize = nvap->na_vattr.va_size;
475 		nvap->na_vattr.va_size = VNOVAL;
476 		if (nvap->na_vattr.va_uid != (uid_t)VNOVAL ||
477 		    nvap->na_vattr.va_gid != (gid_t)VNOVAL ||
478 		    nvap->na_vattr.va_mode != (mode_t)VNOVAL ||
479 		    nvap->na_vattr.va_atime.tv_sec != VNOVAL ||
480 		    nvap->na_vattr.va_mtime.tv_sec != VNOVAL)
481 			savedit = 1;
482 		else
483 			savedit = 2;
484 	}
485 	if (savedit != 2)
486 		error = VOP_SETATTR(vp, &nvap->na_vattr, cred);
487 	if (savedit != 0)
488 		nvap->na_vattr.va_size = savsize;
489 	if (error == 0 && (nvap->na_vattr.va_uid != (uid_t)VNOVAL ||
490 	    nvap->na_vattr.va_gid != (gid_t)VNOVAL ||
491 	    nvap->na_vattr.va_size != VNOVAL ||
492 	    nvap->na_vattr.va_mode != (mode_t)VNOVAL ||
493 	    nvap->na_vattr.va_atime.tv_sec != VNOVAL ||
494 	    nvap->na_vattr.va_mtime.tv_sec != VNOVAL)) {
495 		/* Never modify birthtime on a DS file. */
496 		savbtime = nvap->na_vattr.va_birthtime.tv_sec;
497 		nvap->na_vattr.va_birthtime.tv_sec = VNOVAL;
498 		/* For a pNFS server, set the attributes on the DS file. */
499 		error = nfsrv_proxyds(vp, 0, 0, cred, p, NFSPROC_SETATTR,
500 		    NULL, NULL, NULL, nvap, NULL, NULL, 0, NULL);
501 		nvap->na_vattr.va_birthtime.tv_sec = savbtime;
502 		if (error == ENOENT)
503 			error = 0;
504 	}
505 	NFSEXITCODE(error);
506 	return (error);
507 }
508 
509 /*
510  * Set up nameidata for a lookup() call and do it.
511  */
512 int
513 nfsvno_namei(struct nfsrv_descript *nd, struct nameidata *ndp,
514     struct vnode *dp, int islocked, struct nfsexstuff *exp, struct thread *p,
515     struct vnode **retdirp)
516 {
517 	struct componentname *cnp = &ndp->ni_cnd;
518 	int i;
519 	struct iovec aiov;
520 	struct uio auio;
521 	int lockleaf = (cnp->cn_flags & LOCKLEAF) != 0, linklen;
522 	int error = 0;
523 	char *cp;
524 
525 	*retdirp = NULL;
526 	cnp->cn_nameptr = cnp->cn_pnbuf;
527 	ndp->ni_lcf = 0;
528 	/*
529 	 * Extract and set starting directory.
530 	 */
531 	if (dp->v_type != VDIR) {
532 		if (islocked)
533 			vput(dp);
534 		else
535 			vrele(dp);
536 		nfsvno_relpathbuf(ndp);
537 		error = ENOTDIR;
538 		goto out1;
539 	}
540 	if (islocked)
541 		NFSVOPUNLOCK(dp);
542 	VREF(dp);
543 	*retdirp = dp;
544 	if (NFSVNO_EXRDONLY(exp))
545 		cnp->cn_flags |= RDONLY;
546 	ndp->ni_segflg = UIO_SYSSPACE;
547 
548 	if (nd->nd_flag & ND_PUBLOOKUP) {
549 		ndp->ni_loopcnt = 0;
550 		if (cnp->cn_pnbuf[0] == '/') {
551 			vrele(dp);
552 			/*
553 			 * Check for degenerate pathnames here, since lookup()
554 			 * panics on them.
555 			 */
556 			for (i = 1; i < ndp->ni_pathlen; i++)
557 				if (cnp->cn_pnbuf[i] != '/')
558 					break;
559 			if (i == ndp->ni_pathlen) {
560 				error = NFSERR_ACCES;
561 				goto out;
562 			}
563 			dp = rootvnode;
564 			VREF(dp);
565 		}
566 	} else if ((nfsrv_enable_crossmntpt == 0 && NFSVNO_EXPORTED(exp)) ||
567 	    (nd->nd_flag & ND_NFSV4) == 0) {
568 		/*
569 		 * Only cross mount points for NFSv4 when doing a
570 		 * mount while traversing the file system above
571 		 * the mount point, unless nfsrv_enable_crossmntpt is set.
572 		 */
573 		cnp->cn_flags |= NOCROSSMOUNT;
574 	}
575 
576 	/*
577 	 * Initialize for scan, set ni_startdir and bump ref on dp again
578 	 * because lookup() will dereference ni_startdir.
579 	 */
580 
581 	cnp->cn_thread = p;
582 	ndp->ni_startdir = dp;
583 	ndp->ni_rootdir = rootvnode;
584 	ndp->ni_topdir = NULL;
585 
586 	if (!lockleaf)
587 		cnp->cn_flags |= LOCKLEAF;
588 	for (;;) {
589 		cnp->cn_nameptr = cnp->cn_pnbuf;
590 		/*
591 		 * Call lookup() to do the real work.  If an error occurs,
592 		 * ndp->ni_vp and ni_dvp are left uninitialized or NULL and
593 		 * we do not have to dereference anything before returning.
594 		 * In either case ni_startdir will be dereferenced and NULLed
595 		 * out.
596 		 */
597 		error = lookup(ndp);
598 		if (error)
599 			break;
600 
601 		/*
602 		 * Check for encountering a symbolic link.  Trivial
603 		 * termination occurs if no symlink encountered.
604 		 */
605 		if ((cnp->cn_flags & ISSYMLINK) == 0) {
606 			if ((cnp->cn_flags & (SAVENAME | SAVESTART)) == 0)
607 				nfsvno_relpathbuf(ndp);
608 			if (ndp->ni_vp && !lockleaf)
609 				NFSVOPUNLOCK(ndp->ni_vp);
610 			break;
611 		}
612 
613 		/*
614 		 * Validate symlink
615 		 */
616 		if ((cnp->cn_flags & LOCKPARENT) && ndp->ni_pathlen == 1)
617 			NFSVOPUNLOCK(ndp->ni_dvp);
618 		if (!(nd->nd_flag & ND_PUBLOOKUP)) {
619 			error = EINVAL;
620 			goto badlink2;
621 		}
622 
623 		if (ndp->ni_loopcnt++ >= MAXSYMLINKS) {
624 			error = ELOOP;
625 			goto badlink2;
626 		}
627 		if (ndp->ni_pathlen > 1)
628 			cp = uma_zalloc(namei_zone, M_WAITOK);
629 		else
630 			cp = cnp->cn_pnbuf;
631 		aiov.iov_base = cp;
632 		aiov.iov_len = MAXPATHLEN;
633 		auio.uio_iov = &aiov;
634 		auio.uio_iovcnt = 1;
635 		auio.uio_offset = 0;
636 		auio.uio_rw = UIO_READ;
637 		auio.uio_segflg = UIO_SYSSPACE;
638 		auio.uio_td = NULL;
639 		auio.uio_resid = MAXPATHLEN;
640 		error = VOP_READLINK(ndp->ni_vp, &auio, cnp->cn_cred);
641 		if (error) {
642 		badlink1:
643 			if (ndp->ni_pathlen > 1)
644 				uma_zfree(namei_zone, cp);
645 		badlink2:
646 			vrele(ndp->ni_dvp);
647 			vput(ndp->ni_vp);
648 			break;
649 		}
650 		linklen = MAXPATHLEN - auio.uio_resid;
651 		if (linklen == 0) {
652 			error = ENOENT;
653 			goto badlink1;
654 		}
655 		if (linklen + ndp->ni_pathlen >= MAXPATHLEN) {
656 			error = ENAMETOOLONG;
657 			goto badlink1;
658 		}
659 
660 		/*
661 		 * Adjust or replace path
662 		 */
663 		if (ndp->ni_pathlen > 1) {
664 			NFSBCOPY(ndp->ni_next, cp + linklen, ndp->ni_pathlen);
665 			uma_zfree(namei_zone, cnp->cn_pnbuf);
666 			cnp->cn_pnbuf = cp;
667 		} else
668 			cnp->cn_pnbuf[linklen] = '\0';
669 		ndp->ni_pathlen += linklen;
670 
671 		/*
672 		 * Cleanup refs for next loop and check if root directory
673 		 * should replace current directory.  Normally ni_dvp
674 		 * becomes the new base directory and is cleaned up when
675 		 * we loop.  Explicitly null pointers after invalidation
676 		 * to clarify operation.
677 		 */
678 		vput(ndp->ni_vp);
679 		ndp->ni_vp = NULL;
680 
681 		if (cnp->cn_pnbuf[0] == '/') {
682 			vrele(ndp->ni_dvp);
683 			ndp->ni_dvp = ndp->ni_rootdir;
684 			VREF(ndp->ni_dvp);
685 		}
686 		ndp->ni_startdir = ndp->ni_dvp;
687 		ndp->ni_dvp = NULL;
688 	}
689 	if (!lockleaf)
690 		cnp->cn_flags &= ~LOCKLEAF;
691 
692 out:
693 	if (error) {
694 		nfsvno_relpathbuf(ndp);
695 		ndp->ni_vp = NULL;
696 		ndp->ni_dvp = NULL;
697 		ndp->ni_startdir = NULL;
698 	} else if ((ndp->ni_cnd.cn_flags & (WANTPARENT|LOCKPARENT)) == 0) {
699 		ndp->ni_dvp = NULL;
700 	}
701 
702 out1:
703 	NFSEXITCODE2(error, nd);
704 	return (error);
705 }
706 
707 /*
708  * Set up a pathname buffer and return a pointer to it and, optionally
709  * set a hash pointer.
710  */
711 void
712 nfsvno_setpathbuf(struct nameidata *ndp, char **bufpp, u_long **hashpp)
713 {
714 	struct componentname *cnp = &ndp->ni_cnd;
715 
716 	cnp->cn_flags |= (NOMACCHECK | HASBUF);
717 	cnp->cn_pnbuf = uma_zalloc(namei_zone, M_WAITOK);
718 	if (hashpp != NULL)
719 		*hashpp = NULL;
720 	*bufpp = cnp->cn_pnbuf;
721 }
722 
723 /*
724  * Release the above path buffer, if not released by nfsvno_namei().
725  */
726 void
727 nfsvno_relpathbuf(struct nameidata *ndp)
728 {
729 
730 	if ((ndp->ni_cnd.cn_flags & HASBUF) == 0)
731 		panic("nfsrelpath");
732 	uma_zfree(namei_zone, ndp->ni_cnd.cn_pnbuf);
733 	ndp->ni_cnd.cn_flags &= ~HASBUF;
734 }
735 
736 /*
737  * Readlink vnode op into an mbuf list.
738  */
739 int
740 nfsvno_readlink(struct vnode *vp, struct ucred *cred, int maxextsiz,
741     struct thread *p, struct mbuf **mpp, struct mbuf **mpendp, int *lenp)
742 {
743 	struct iovec *iv;
744 	struct uio io, *uiop = &io;
745 	struct mbuf *mp, *mp3;
746 	int len, tlen, error = 0;
747 
748 	len = NFS_MAXPATHLEN;
749 	if (maxextsiz > 0)
750 		uiop->uio_iovcnt = nfsrv_createiovec_extpgs(len, maxextsiz,
751 		    &mp3, &mp, &iv);
752 	else
753 		uiop->uio_iovcnt = nfsrv_createiovec(len, &mp3, &mp, &iv);
754 	uiop->uio_iov = iv;
755 	uiop->uio_offset = 0;
756 	uiop->uio_resid = len;
757 	uiop->uio_rw = UIO_READ;
758 	uiop->uio_segflg = UIO_SYSSPACE;
759 	uiop->uio_td = NULL;
760 	error = VOP_READLINK(vp, uiop, cred);
761 	free(iv, M_TEMP);
762 	if (error) {
763 		m_freem(mp3);
764 		*lenp = 0;
765 		goto out;
766 	}
767 	if (uiop->uio_resid > 0) {
768 		len -= uiop->uio_resid;
769 		tlen = NFSM_RNDUP(len);
770 		if (tlen == 0) {
771 			m_freem(mp3);
772 			mp3 = mp = NULL;
773 		} else if (tlen != NFS_MAXPATHLEN || tlen != len)
774 			mp = nfsrv_adj(mp3, NFS_MAXPATHLEN - tlen,
775 			    tlen - len);
776 	}
777 	*lenp = len;
778 	*mpp = mp3;
779 	*mpendp = mp;
780 
781 out:
782 	NFSEXITCODE(error);
783 	return (error);
784 }
785 
786 /*
787  * Create an mbuf chain and an associated iovec that can be used to Read
788  * or Getextattr of data.
789  * Upon success, return pointers to the first and last mbufs in the chain
790  * plus the malloc'd iovec and its iovlen.
791  */
792 static int
793 nfsrv_createiovec(int len, struct mbuf **mpp, struct mbuf **mpendp,
794     struct iovec **ivp)
795 {
796 	struct mbuf *m, *m2 = NULL, *m3;
797 	struct iovec *iv;
798 	int i, left, siz;
799 
800 	left = len;
801 	m3 = NULL;
802 	/*
803 	 * Generate the mbuf list with the uio_iov ref. to it.
804 	 */
805 	i = 0;
806 	while (left > 0) {
807 		NFSMGET(m);
808 		MCLGET(m, M_WAITOK);
809 		m->m_len = 0;
810 		siz = min(M_TRAILINGSPACE(m), left);
811 		left -= siz;
812 		i++;
813 		if (m3)
814 			m2->m_next = m;
815 		else
816 			m3 = m;
817 		m2 = m;
818 	}
819 	*ivp = iv = malloc(i * sizeof (struct iovec), M_TEMP, M_WAITOK);
820 	m = m3;
821 	left = len;
822 	i = 0;
823 	while (left > 0) {
824 		if (m == NULL)
825 			panic("nfsrv_createiovec iov");
826 		siz = min(M_TRAILINGSPACE(m), left);
827 		if (siz > 0) {
828 			iv->iov_base = mtod(m, caddr_t) + m->m_len;
829 			iv->iov_len = siz;
830 			m->m_len += siz;
831 			left -= siz;
832 			iv++;
833 			i++;
834 		}
835 		m = m->m_next;
836 	}
837 	*mpp = m3;
838 	*mpendp = m2;
839 	return (i);
840 }
841 
842 /*
843  * Create an mbuf chain and an associated iovec that can be used to Read
844  * or Getextattr of data.
845  * Upon success, return pointers to the first and last mbufs in the chain
846  * plus the malloc'd iovec and its iovlen.
847  * Same as above, but creates ext_pgs mbuf(s).
848  */
849 static int
850 nfsrv_createiovec_extpgs(int len, int maxextsiz, struct mbuf **mpp,
851     struct mbuf **mpendp, struct iovec **ivp)
852 {
853 	struct mbuf *m, *m2 = NULL, *m3;
854 	struct iovec *iv;
855 	int i, left, pgno, siz;
856 
857 	left = len;
858 	m3 = NULL;
859 	/*
860 	 * Generate the mbuf list with the uio_iov ref. to it.
861 	 */
862 	i = 0;
863 	while (left > 0) {
864 		siz = min(left, maxextsiz);
865 		m = mb_alloc_ext_plus_pages(siz, M_WAITOK);
866 		left -= siz;
867 		i += m->m_epg_npgs;
868 		if (m3 != NULL)
869 			m2->m_next = m;
870 		else
871 			m3 = m;
872 		m2 = m;
873 	}
874 	*ivp = iv = malloc(i * sizeof (struct iovec), M_TEMP, M_WAITOK);
875 	m = m3;
876 	left = len;
877 	i = 0;
878 	pgno = 0;
879 	while (left > 0) {
880 		if (m == NULL)
881 			panic("nfsvno_createiovec_extpgs iov");
882 		siz = min(PAGE_SIZE, left);
883 		if (siz > 0) {
884 			iv->iov_base = (void *)PHYS_TO_DMAP(m->m_epg_pa[pgno]);
885 			iv->iov_len = siz;
886 			m->m_len += siz;
887 			if (pgno == m->m_epg_npgs - 1)
888 				m->m_epg_last_len = siz;
889 			left -= siz;
890 			iv++;
891 			i++;
892 			pgno++;
893 		}
894 		if (pgno == m->m_epg_npgs && left > 0) {
895 			m = m->m_next;
896 			if (m == NULL)
897 				panic("nfsvno_createiovec_extpgs iov");
898 			pgno = 0;
899 		}
900 	}
901 	*mpp = m3;
902 	*mpendp = m2;
903 	return (i);
904 }
905 
906 /*
907  * Read vnode op call into mbuf list.
908  */
909 int
910 nfsvno_read(struct vnode *vp, off_t off, int cnt, struct ucred *cred,
911     int maxextsiz, struct thread *p, struct mbuf **mpp,
912     struct mbuf **mpendp)
913 {
914 	struct mbuf *m;
915 	struct iovec *iv;
916 	int error = 0, len, tlen, ioflag = 0;
917 	struct mbuf *m3;
918 	struct uio io, *uiop = &io;
919 	struct nfsheur *nh;
920 
921 	/*
922 	 * Attempt to read from a DS file. A return of ENOENT implies
923 	 * there is no DS file to read.
924 	 */
925 	error = nfsrv_proxyds(vp, off, cnt, cred, p, NFSPROC_READDS, mpp,
926 	    NULL, mpendp, NULL, NULL, NULL, 0, NULL);
927 	if (error != ENOENT)
928 		return (error);
929 
930 	len = NFSM_RNDUP(cnt);
931 	if (maxextsiz > 0)
932 		uiop->uio_iovcnt = nfsrv_createiovec_extpgs(len, maxextsiz,
933 		    &m3, &m, &iv);
934 	else
935 		uiop->uio_iovcnt = nfsrv_createiovec(len, &m3, &m, &iv);
936 	uiop->uio_iov = iv;
937 	uiop->uio_offset = off;
938 	uiop->uio_resid = len;
939 	uiop->uio_rw = UIO_READ;
940 	uiop->uio_segflg = UIO_SYSSPACE;
941 	uiop->uio_td = NULL;
942 	nh = nfsrv_sequential_heuristic(uiop, vp);
943 	ioflag |= nh->nh_seqcount << IO_SEQSHIFT;
944 	/* XXX KDM make this more systematic? */
945 	nfsstatsv1.srvbytes[NFSV4OP_READ] += uiop->uio_resid;
946 	error = VOP_READ(vp, uiop, IO_NODELOCKED | ioflag, cred);
947 	free(iv, M_TEMP);
948 	if (error) {
949 		m_freem(m3);
950 		*mpp = NULL;
951 		goto out;
952 	}
953 	nh->nh_nextoff = uiop->uio_offset;
954 	tlen = len - uiop->uio_resid;
955 	cnt = cnt < tlen ? cnt : tlen;
956 	tlen = NFSM_RNDUP(cnt);
957 	if (tlen == 0) {
958 		m_freem(m3);
959 		m3 = m = NULL;
960 	} else if (len != tlen || tlen != cnt)
961 		m = nfsrv_adj(m3, len - tlen, tlen - cnt);
962 	*mpp = m3;
963 	*mpendp = m;
964 
965 out:
966 	NFSEXITCODE(error);
967 	return (error);
968 }
969 
970 /*
971  * Create the iovec for the mbuf chain passed in as an argument.
972  * The "cp" argument is where the data starts within the first mbuf in
973  * the chain. It returns the iovec and the iovcnt.
974  */
975 static int
976 nfsrv_createiovecw(int retlen, struct mbuf *m, char *cp, struct iovec **ivpp,
977     int *iovcntp)
978 {
979 	struct mbuf *mp;
980 	struct iovec *ivp;
981 	int cnt, i, len;
982 
983 	/*
984 	 * Loop through the mbuf chain, counting how many mbufs are a
985 	 * part of this write operation, so the iovec size is known.
986 	 */
987 	cnt = 0;
988 	len = retlen;
989 	mp = m;
990 	i = mtod(mp, caddr_t) + mp->m_len - cp;
991 	while (len > 0) {
992 		if (i > 0) {
993 			len -= i;
994 			cnt++;
995 		}
996 		mp = mp->m_next;
997 		if (!mp) {
998 			if (len > 0)
999 				return (EBADRPC);
1000 		} else
1001 			i = mp->m_len;
1002 	}
1003 
1004 	/* Now, create the iovec. */
1005 	mp = m;
1006 	*ivpp = ivp = malloc(cnt * sizeof (struct iovec), M_TEMP,
1007 	    M_WAITOK);
1008 	*iovcntp = cnt;
1009 	i = mtod(mp, caddr_t) + mp->m_len - cp;
1010 	len = retlen;
1011 	while (len > 0) {
1012 		if (mp == NULL)
1013 			panic("nfsrv_createiovecw");
1014 		if (i > 0) {
1015 			i = min(i, len);
1016 			ivp->iov_base = cp;
1017 			ivp->iov_len = i;
1018 			ivp++;
1019 			len -= i;
1020 		}
1021 		mp = mp->m_next;
1022 		if (mp) {
1023 			i = mp->m_len;
1024 			cp = mtod(mp, caddr_t);
1025 		}
1026 	}
1027 	return (0);
1028 }
1029 
1030 /*
1031  * Write vnode op from an mbuf list.
1032  */
1033 int
1034 nfsvno_write(struct vnode *vp, off_t off, int retlen, int *stable,
1035     struct mbuf *mp, char *cp, struct ucred *cred, struct thread *p)
1036 {
1037 	struct iovec *iv;
1038 	int cnt, ioflags, error;
1039 	struct uio io, *uiop = &io;
1040 	struct nfsheur *nh;
1041 
1042 	/*
1043 	 * Attempt to write to a DS file. A return of ENOENT implies
1044 	 * there is no DS file to write.
1045 	 */
1046 	error = nfsrv_proxyds(vp, off, retlen, cred, p, NFSPROC_WRITEDS,
1047 	    &mp, cp, NULL, NULL, NULL, NULL, 0, NULL);
1048 	if (error != ENOENT) {
1049 		*stable = NFSWRITE_FILESYNC;
1050 		return (error);
1051 	}
1052 
1053 	if (*stable == NFSWRITE_UNSTABLE)
1054 		ioflags = IO_NODELOCKED;
1055 	else
1056 		ioflags = (IO_SYNC | IO_NODELOCKED);
1057 	error = nfsrv_createiovecw(retlen, mp, cp, &iv, &cnt);
1058 	if (error != 0)
1059 		return (error);
1060 	uiop->uio_iov = iv;
1061 	uiop->uio_iovcnt = cnt;
1062 	uiop->uio_resid = retlen;
1063 	uiop->uio_rw = UIO_WRITE;
1064 	uiop->uio_segflg = UIO_SYSSPACE;
1065 	NFSUIOPROC(uiop, p);
1066 	uiop->uio_offset = off;
1067 	nh = nfsrv_sequential_heuristic(uiop, vp);
1068 	ioflags |= nh->nh_seqcount << IO_SEQSHIFT;
1069 	/* XXX KDM make this more systematic? */
1070 	nfsstatsv1.srvbytes[NFSV4OP_WRITE] += uiop->uio_resid;
1071 	error = VOP_WRITE(vp, uiop, ioflags, cred);
1072 	if (error == 0)
1073 		nh->nh_nextoff = uiop->uio_offset;
1074 	free(iv, M_TEMP);
1075 
1076 	NFSEXITCODE(error);
1077 	return (error);
1078 }
1079 
1080 /*
1081  * Common code for creating a regular file (plus special files for V2).
1082  */
1083 int
1084 nfsvno_createsub(struct nfsrv_descript *nd, struct nameidata *ndp,
1085     struct vnode **vpp, struct nfsvattr *nvap, int *exclusive_flagp,
1086     int32_t *cverf, NFSDEV_T rdev, struct nfsexstuff *exp)
1087 {
1088 	u_quad_t tempsize;
1089 	int error;
1090 	struct thread *p = curthread;
1091 
1092 	error = nd->nd_repstat;
1093 	if (!error && ndp->ni_vp == NULL) {
1094 		if (nvap->na_type == VREG || nvap->na_type == VSOCK) {
1095 			vrele(ndp->ni_startdir);
1096 			error = VOP_CREATE(ndp->ni_dvp,
1097 			    &ndp->ni_vp, &ndp->ni_cnd, &nvap->na_vattr);
1098 			/* For a pNFS server, create the data file on a DS. */
1099 			if (error == 0 && nvap->na_type == VREG) {
1100 				/*
1101 				 * Create a data file on a DS for a pNFS server.
1102 				 * This function just returns if not
1103 				 * running a pNFS DS or the creation fails.
1104 				 */
1105 				nfsrv_pnfscreate(ndp->ni_vp, &nvap->na_vattr,
1106 				    nd->nd_cred, p);
1107 			}
1108 			VOP_VPUT_PAIR(ndp->ni_dvp, error == 0 ? &ndp->ni_vp :
1109 			    NULL, false);
1110 			nfsvno_relpathbuf(ndp);
1111 			if (!error) {
1112 				if (*exclusive_flagp) {
1113 					*exclusive_flagp = 0;
1114 					NFSVNO_ATTRINIT(nvap);
1115 					nvap->na_atime.tv_sec = cverf[0];
1116 					nvap->na_atime.tv_nsec = cverf[1];
1117 					error = VOP_SETATTR(ndp->ni_vp,
1118 					    &nvap->na_vattr, nd->nd_cred);
1119 					if (error != 0) {
1120 						vput(ndp->ni_vp);
1121 						ndp->ni_vp = NULL;
1122 						error = NFSERR_NOTSUPP;
1123 					}
1124 				}
1125 			}
1126 		/*
1127 		 * NFS V2 Only. nfsrvd_mknod() does this for V3.
1128 		 * (This implies, just get out on an error.)
1129 		 */
1130 		} else if (nvap->na_type == VCHR || nvap->na_type == VBLK ||
1131 			nvap->na_type == VFIFO) {
1132 			if (nvap->na_type == VCHR && rdev == 0xffffffff)
1133 				nvap->na_type = VFIFO;
1134                         if (nvap->na_type != VFIFO &&
1135 			    (error = priv_check_cred(nd->nd_cred, PRIV_VFS_MKNOD_DEV))) {
1136 				vrele(ndp->ni_startdir);
1137 				nfsvno_relpathbuf(ndp);
1138 				vput(ndp->ni_dvp);
1139 				goto out;
1140 			}
1141 			nvap->na_rdev = rdev;
1142 			error = VOP_MKNOD(ndp->ni_dvp, &ndp->ni_vp,
1143 			    &ndp->ni_cnd, &nvap->na_vattr);
1144 			VOP_VPUT_PAIR(ndp->ni_dvp, error == 0 ? &ndp->ni_vp :
1145 			    NULL, false);
1146 			nfsvno_relpathbuf(ndp);
1147 			vrele(ndp->ni_startdir);
1148 			if (error)
1149 				goto out;
1150 		} else {
1151 			vrele(ndp->ni_startdir);
1152 			nfsvno_relpathbuf(ndp);
1153 			vput(ndp->ni_dvp);
1154 			error = ENXIO;
1155 			goto out;
1156 		}
1157 		*vpp = ndp->ni_vp;
1158 	} else {
1159 		/*
1160 		 * Handle cases where error is already set and/or
1161 		 * the file exists.
1162 		 * 1 - clean up the lookup
1163 		 * 2 - iff !error and na_size set, truncate it
1164 		 */
1165 		vrele(ndp->ni_startdir);
1166 		nfsvno_relpathbuf(ndp);
1167 		*vpp = ndp->ni_vp;
1168 		if (ndp->ni_dvp == *vpp)
1169 			vrele(ndp->ni_dvp);
1170 		else
1171 			vput(ndp->ni_dvp);
1172 		if (!error && nvap->na_size != VNOVAL) {
1173 			error = nfsvno_accchk(*vpp, VWRITE,
1174 			    nd->nd_cred, exp, p, NFSACCCHK_NOOVERRIDE,
1175 			    NFSACCCHK_VPISLOCKED, NULL);
1176 			if (!error) {
1177 				tempsize = nvap->na_size;
1178 				NFSVNO_ATTRINIT(nvap);
1179 				nvap->na_size = tempsize;
1180 				error = VOP_SETATTR(*vpp,
1181 				    &nvap->na_vattr, nd->nd_cred);
1182 			}
1183 		}
1184 		if (error)
1185 			vput(*vpp);
1186 	}
1187 
1188 out:
1189 	NFSEXITCODE(error);
1190 	return (error);
1191 }
1192 
1193 /*
1194  * Do a mknod vnode op.
1195  */
1196 int
1197 nfsvno_mknod(struct nameidata *ndp, struct nfsvattr *nvap, struct ucred *cred,
1198     struct thread *p)
1199 {
1200 	int error = 0;
1201 	enum vtype vtyp;
1202 
1203 	vtyp = nvap->na_type;
1204 	/*
1205 	 * Iff doesn't exist, create it.
1206 	 */
1207 	if (ndp->ni_vp) {
1208 		vrele(ndp->ni_startdir);
1209 		nfsvno_relpathbuf(ndp);
1210 		vput(ndp->ni_dvp);
1211 		vrele(ndp->ni_vp);
1212 		error = EEXIST;
1213 		goto out;
1214 	}
1215 	if (vtyp != VCHR && vtyp != VBLK && vtyp != VSOCK && vtyp != VFIFO) {
1216 		vrele(ndp->ni_startdir);
1217 		nfsvno_relpathbuf(ndp);
1218 		vput(ndp->ni_dvp);
1219 		error = NFSERR_BADTYPE;
1220 		goto out;
1221 	}
1222 	if (vtyp == VSOCK) {
1223 		vrele(ndp->ni_startdir);
1224 		error = VOP_CREATE(ndp->ni_dvp, &ndp->ni_vp,
1225 		    &ndp->ni_cnd, &nvap->na_vattr);
1226 		VOP_VPUT_PAIR(ndp->ni_dvp, error == 0 ? &ndp->ni_vp : NULL,
1227 		    false);
1228 		nfsvno_relpathbuf(ndp);
1229 	} else {
1230 		if (nvap->na_type != VFIFO &&
1231 		    (error = priv_check_cred(cred, PRIV_VFS_MKNOD_DEV))) {
1232 			vrele(ndp->ni_startdir);
1233 			nfsvno_relpathbuf(ndp);
1234 			vput(ndp->ni_dvp);
1235 			goto out;
1236 		}
1237 		error = VOP_MKNOD(ndp->ni_dvp, &ndp->ni_vp,
1238 		    &ndp->ni_cnd, &nvap->na_vattr);
1239 		VOP_VPUT_PAIR(ndp->ni_dvp, error == 0 ? &ndp->ni_vp : NULL,
1240 		    false);
1241 		nfsvno_relpathbuf(ndp);
1242 		vrele(ndp->ni_startdir);
1243 		/*
1244 		 * Since VOP_MKNOD returns the ni_vp, I can't
1245 		 * see any reason to do the lookup.
1246 		 */
1247 	}
1248 
1249 out:
1250 	NFSEXITCODE(error);
1251 	return (error);
1252 }
1253 
1254 /*
1255  * Mkdir vnode op.
1256  */
1257 int
1258 nfsvno_mkdir(struct nameidata *ndp, struct nfsvattr *nvap, uid_t saved_uid,
1259     struct ucred *cred, struct thread *p, struct nfsexstuff *exp)
1260 {
1261 	int error = 0;
1262 
1263 	if (ndp->ni_vp != NULL) {
1264 		if (ndp->ni_dvp == ndp->ni_vp)
1265 			vrele(ndp->ni_dvp);
1266 		else
1267 			vput(ndp->ni_dvp);
1268 		vrele(ndp->ni_vp);
1269 		nfsvno_relpathbuf(ndp);
1270 		error = EEXIST;
1271 		goto out;
1272 	}
1273 	error = VOP_MKDIR(ndp->ni_dvp, &ndp->ni_vp, &ndp->ni_cnd,
1274 	    &nvap->na_vattr);
1275 	VOP_VPUT_PAIR(ndp->ni_dvp, error == 0 ? &ndp->ni_vp : NULL, false);
1276 	nfsvno_relpathbuf(ndp);
1277 
1278 out:
1279 	NFSEXITCODE(error);
1280 	return (error);
1281 }
1282 
1283 /*
1284  * symlink vnode op.
1285  */
1286 int
1287 nfsvno_symlink(struct nameidata *ndp, struct nfsvattr *nvap, char *pathcp,
1288     int pathlen, int not_v2, uid_t saved_uid, struct ucred *cred, struct thread *p,
1289     struct nfsexstuff *exp)
1290 {
1291 	int error = 0;
1292 
1293 	if (ndp->ni_vp) {
1294 		vrele(ndp->ni_startdir);
1295 		nfsvno_relpathbuf(ndp);
1296 		if (ndp->ni_dvp == ndp->ni_vp)
1297 			vrele(ndp->ni_dvp);
1298 		else
1299 			vput(ndp->ni_dvp);
1300 		vrele(ndp->ni_vp);
1301 		error = EEXIST;
1302 		goto out;
1303 	}
1304 
1305 	error = VOP_SYMLINK(ndp->ni_dvp, &ndp->ni_vp, &ndp->ni_cnd,
1306 	    &nvap->na_vattr, pathcp);
1307 	/*
1308 	 * Although FreeBSD still had the lookup code in
1309 	 * it for 7/current, there doesn't seem to be any
1310 	 * point, since VOP_SYMLINK() returns the ni_vp.
1311 	 * Just vput it for v2.
1312 	 */
1313 	VOP_VPUT_PAIR(ndp->ni_dvp, &ndp->ni_vp, !not_v2 && error == 0);
1314 	vrele(ndp->ni_startdir);
1315 	nfsvno_relpathbuf(ndp);
1316 
1317 out:
1318 	NFSEXITCODE(error);
1319 	return (error);
1320 }
1321 
1322 /*
1323  * Parse symbolic link arguments.
1324  * This function has an ugly side effect. It will malloc() an area for
1325  * the symlink and set iov_base to point to it, only if it succeeds.
1326  * So, if it returns with uiop->uio_iov->iov_base != NULL, that must
1327  * be FREE'd later.
1328  */
1329 int
1330 nfsvno_getsymlink(struct nfsrv_descript *nd, struct nfsvattr *nvap,
1331     struct thread *p, char **pathcpp, int *lenp)
1332 {
1333 	u_int32_t *tl;
1334 	char *pathcp = NULL;
1335 	int error = 0, len;
1336 	struct nfsv2_sattr *sp;
1337 
1338 	*pathcpp = NULL;
1339 	*lenp = 0;
1340 	if ((nd->nd_flag & ND_NFSV3) &&
1341 	    (error = nfsrv_sattr(nd, NULL, nvap, NULL, NULL, p)))
1342 		goto nfsmout;
1343 	NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
1344 	len = fxdr_unsigned(int, *tl);
1345 	if (len > NFS_MAXPATHLEN || len <= 0) {
1346 		error = EBADRPC;
1347 		goto nfsmout;
1348 	}
1349 	pathcp = malloc(len + 1, M_TEMP, M_WAITOK);
1350 	error = nfsrv_mtostr(nd, pathcp, len);
1351 	if (error)
1352 		goto nfsmout;
1353 	if (nd->nd_flag & ND_NFSV2) {
1354 		NFSM_DISSECT(sp, struct nfsv2_sattr *, NFSX_V2SATTR);
1355 		nvap->na_mode = fxdr_unsigned(u_int16_t, sp->sa_mode);
1356 	}
1357 	*pathcpp = pathcp;
1358 	*lenp = len;
1359 	NFSEXITCODE2(0, nd);
1360 	return (0);
1361 nfsmout:
1362 	if (pathcp)
1363 		free(pathcp, M_TEMP);
1364 	NFSEXITCODE2(error, nd);
1365 	return (error);
1366 }
1367 
1368 /*
1369  * Remove a non-directory object.
1370  */
1371 int
1372 nfsvno_removesub(struct nameidata *ndp, int is_v4, struct ucred *cred,
1373     struct thread *p, struct nfsexstuff *exp)
1374 {
1375 	struct vnode *vp, *dsdvp[NFSDEV_MAXMIRRORS];
1376 	int error = 0, mirrorcnt;
1377 	char fname[PNFS_FILENAME_LEN + 1];
1378 	fhandle_t fh;
1379 
1380 	vp = ndp->ni_vp;
1381 	dsdvp[0] = NULL;
1382 	if (vp->v_type == VDIR)
1383 		error = NFSERR_ISDIR;
1384 	else if (is_v4)
1385 		error = nfsrv_checkremove(vp, 1, NULL, (nfsquad_t)((u_quad_t)0),
1386 		    p);
1387 	if (error == 0)
1388 		nfsrv_pnfsremovesetup(vp, p, dsdvp, &mirrorcnt, fname, &fh);
1389 	if (!error)
1390 		error = VOP_REMOVE(ndp->ni_dvp, vp, &ndp->ni_cnd);
1391 	if (error == 0 && dsdvp[0] != NULL)
1392 		nfsrv_pnfsremove(dsdvp, mirrorcnt, fname, &fh, p);
1393 	if (ndp->ni_dvp == vp)
1394 		vrele(ndp->ni_dvp);
1395 	else
1396 		vput(ndp->ni_dvp);
1397 	vput(vp);
1398 	if ((ndp->ni_cnd.cn_flags & SAVENAME) != 0)
1399 		nfsvno_relpathbuf(ndp);
1400 	NFSEXITCODE(error);
1401 	return (error);
1402 }
1403 
1404 /*
1405  * Remove a directory.
1406  */
1407 int
1408 nfsvno_rmdirsub(struct nameidata *ndp, int is_v4, struct ucred *cred,
1409     struct thread *p, struct nfsexstuff *exp)
1410 {
1411 	struct vnode *vp;
1412 	int error = 0;
1413 
1414 	vp = ndp->ni_vp;
1415 	if (vp->v_type != VDIR) {
1416 		error = ENOTDIR;
1417 		goto out;
1418 	}
1419 	/*
1420 	 * No rmdir "." please.
1421 	 */
1422 	if (ndp->ni_dvp == vp) {
1423 		error = EINVAL;
1424 		goto out;
1425 	}
1426 	/*
1427 	 * The root of a mounted filesystem cannot be deleted.
1428 	 */
1429 	if (vp->v_vflag & VV_ROOT)
1430 		error = EBUSY;
1431 out:
1432 	if (!error)
1433 		error = VOP_RMDIR(ndp->ni_dvp, vp, &ndp->ni_cnd);
1434 	if (ndp->ni_dvp == vp)
1435 		vrele(ndp->ni_dvp);
1436 	else
1437 		vput(ndp->ni_dvp);
1438 	vput(vp);
1439 	if ((ndp->ni_cnd.cn_flags & SAVENAME) != 0)
1440 		nfsvno_relpathbuf(ndp);
1441 	NFSEXITCODE(error);
1442 	return (error);
1443 }
1444 
1445 /*
1446  * Rename vnode op.
1447  */
1448 int
1449 nfsvno_rename(struct nameidata *fromndp, struct nameidata *tondp,
1450     u_int32_t ndstat, u_int32_t ndflag, struct ucred *cred, struct thread *p)
1451 {
1452 	struct vnode *fvp, *tvp, *tdvp, *dsdvp[NFSDEV_MAXMIRRORS];
1453 	int error = 0, mirrorcnt;
1454 	char fname[PNFS_FILENAME_LEN + 1];
1455 	fhandle_t fh;
1456 
1457 	dsdvp[0] = NULL;
1458 	fvp = fromndp->ni_vp;
1459 	if (ndstat) {
1460 		vrele(fromndp->ni_dvp);
1461 		vrele(fvp);
1462 		error = ndstat;
1463 		goto out1;
1464 	}
1465 	tdvp = tondp->ni_dvp;
1466 	tvp = tondp->ni_vp;
1467 	if (tvp != NULL) {
1468 		if (fvp->v_type == VDIR && tvp->v_type != VDIR) {
1469 			error = (ndflag & ND_NFSV2) ? EISDIR : EEXIST;
1470 			goto out;
1471 		} else if (fvp->v_type != VDIR && tvp->v_type == VDIR) {
1472 			error = (ndflag & ND_NFSV2) ? ENOTDIR : EEXIST;
1473 			goto out;
1474 		}
1475 		if (tvp->v_type == VDIR && tvp->v_mountedhere) {
1476 			error = (ndflag & ND_NFSV2) ? ENOTEMPTY : EXDEV;
1477 			goto out;
1478 		}
1479 
1480 		/*
1481 		 * A rename to '.' or '..' results in a prematurely
1482 		 * unlocked vnode on FreeBSD5, so I'm just going to fail that
1483 		 * here.
1484 		 */
1485 		if ((tondp->ni_cnd.cn_namelen == 1 &&
1486 		     tondp->ni_cnd.cn_nameptr[0] == '.') ||
1487 		    (tondp->ni_cnd.cn_namelen == 2 &&
1488 		     tondp->ni_cnd.cn_nameptr[0] == '.' &&
1489 		     tondp->ni_cnd.cn_nameptr[1] == '.')) {
1490 			error = EINVAL;
1491 			goto out;
1492 		}
1493 	}
1494 	if (fvp->v_type == VDIR && fvp->v_mountedhere) {
1495 		error = (ndflag & ND_NFSV2) ? ENOTEMPTY : EXDEV;
1496 		goto out;
1497 	}
1498 	if (fvp->v_mount != tdvp->v_mount) {
1499 		error = (ndflag & ND_NFSV2) ? ENOTEMPTY : EXDEV;
1500 		goto out;
1501 	}
1502 	if (fvp == tdvp) {
1503 		error = (ndflag & ND_NFSV2) ? ENOTEMPTY : EINVAL;
1504 		goto out;
1505 	}
1506 	if (fvp == tvp) {
1507 		/*
1508 		 * If source and destination are the same, there is nothing to
1509 		 * do. Set error to -1 to indicate this.
1510 		 */
1511 		error = -1;
1512 		goto out;
1513 	}
1514 	if (ndflag & ND_NFSV4) {
1515 		if (NFSVOPLOCK(fvp, LK_EXCLUSIVE) == 0) {
1516 			error = nfsrv_checkremove(fvp, 0, NULL,
1517 			    (nfsquad_t)((u_quad_t)0), p);
1518 			NFSVOPUNLOCK(fvp);
1519 		} else
1520 			error = EPERM;
1521 		if (tvp && !error)
1522 			error = nfsrv_checkremove(tvp, 1, NULL,
1523 			    (nfsquad_t)((u_quad_t)0), p);
1524 	} else {
1525 		/*
1526 		 * For NFSv2 and NFSv3, try to get rid of the delegation, so
1527 		 * that the NFSv4 client won't be confused by the rename.
1528 		 * Since nfsd_recalldelegation() can only be called on an
1529 		 * unlocked vnode at this point and fvp is the file that will
1530 		 * still exist after the rename, just do fvp.
1531 		 */
1532 		nfsd_recalldelegation(fvp, p);
1533 	}
1534 	if (error == 0 && tvp != NULL) {
1535 		nfsrv_pnfsremovesetup(tvp, p, dsdvp, &mirrorcnt, fname, &fh);
1536 		NFSD_DEBUG(4, "nfsvno_rename: pnfsremovesetup"
1537 		    " dsdvp=%p\n", dsdvp[0]);
1538 	}
1539 out:
1540 	if (!error) {
1541 		error = VOP_RENAME(fromndp->ni_dvp, fromndp->ni_vp,
1542 		    &fromndp->ni_cnd, tondp->ni_dvp, tondp->ni_vp,
1543 		    &tondp->ni_cnd);
1544 	} else {
1545 		if (tdvp == tvp)
1546 			vrele(tdvp);
1547 		else
1548 			vput(tdvp);
1549 		if (tvp)
1550 			vput(tvp);
1551 		vrele(fromndp->ni_dvp);
1552 		vrele(fvp);
1553 		if (error == -1)
1554 			error = 0;
1555 	}
1556 
1557 	/*
1558 	 * If dsdvp[0] != NULL, it was set up by nfsrv_pnfsremovesetup() and
1559 	 * if the rename succeeded, the DS file for the tvp needs to be
1560 	 * removed.
1561 	 */
1562 	if (error == 0 && dsdvp[0] != NULL) {
1563 		nfsrv_pnfsremove(dsdvp, mirrorcnt, fname, &fh, p);
1564 		NFSD_DEBUG(4, "nfsvno_rename: pnfsremove\n");
1565 	}
1566 
1567 	vrele(tondp->ni_startdir);
1568 	nfsvno_relpathbuf(tondp);
1569 out1:
1570 	vrele(fromndp->ni_startdir);
1571 	nfsvno_relpathbuf(fromndp);
1572 	NFSEXITCODE(error);
1573 	return (error);
1574 }
1575 
1576 /*
1577  * Link vnode op.
1578  */
1579 int
1580 nfsvno_link(struct nameidata *ndp, struct vnode *vp, struct ucred *cred,
1581     struct thread *p, struct nfsexstuff *exp)
1582 {
1583 	struct vnode *xp;
1584 	int error = 0;
1585 
1586 	xp = ndp->ni_vp;
1587 	if (xp != NULL) {
1588 		error = EEXIST;
1589 	} else {
1590 		xp = ndp->ni_dvp;
1591 		if (vp->v_mount != xp->v_mount)
1592 			error = EXDEV;
1593 	}
1594 	if (!error) {
1595 		NFSVOPLOCK(vp, LK_EXCLUSIVE | LK_RETRY);
1596 		if (!VN_IS_DOOMED(vp))
1597 			error = VOP_LINK(ndp->ni_dvp, vp, &ndp->ni_cnd);
1598 		else
1599 			error = EPERM;
1600 		if (ndp->ni_dvp == vp) {
1601 			vrele(ndp->ni_dvp);
1602 			NFSVOPUNLOCK(vp);
1603 		} else {
1604 			vref(vp);
1605 			VOP_VPUT_PAIR(ndp->ni_dvp, &vp, true);
1606 		}
1607 	} else {
1608 		if (ndp->ni_dvp == ndp->ni_vp)
1609 			vrele(ndp->ni_dvp);
1610 		else
1611 			vput(ndp->ni_dvp);
1612 		if (ndp->ni_vp)
1613 			vrele(ndp->ni_vp);
1614 	}
1615 	nfsvno_relpathbuf(ndp);
1616 	NFSEXITCODE(error);
1617 	return (error);
1618 }
1619 
1620 /*
1621  * Do the fsync() appropriate for the commit.
1622  */
1623 int
1624 nfsvno_fsync(struct vnode *vp, u_int64_t off, int cnt, struct ucred *cred,
1625     struct thread *td)
1626 {
1627 	int error = 0;
1628 
1629 	/*
1630 	 * RFC 1813 3.3.21: if count is 0, a flush from offset to the end of
1631 	 * file is done.  At this time VOP_FSYNC does not accept offset and
1632 	 * byte count parameters so call VOP_FSYNC the whole file for now.
1633 	 * The same is true for NFSv4: RFC 3530 Sec. 14.2.3.
1634 	 * File systems that do not use the buffer cache (as indicated
1635 	 * by MNTK_USES_BCACHE not being set) must use VOP_FSYNC().
1636 	 */
1637 	if (cnt == 0 || cnt > MAX_COMMIT_COUNT ||
1638 	    (vp->v_mount->mnt_kern_flag & MNTK_USES_BCACHE) == 0) {
1639 		/*
1640 		 * Give up and do the whole thing
1641 		 */
1642 		if (vp->v_object && vm_object_mightbedirty(vp->v_object)) {
1643 			VM_OBJECT_WLOCK(vp->v_object);
1644 			vm_object_page_clean(vp->v_object, 0, 0, OBJPC_SYNC);
1645 			VM_OBJECT_WUNLOCK(vp->v_object);
1646 		}
1647 		error = VOP_FSYNC(vp, MNT_WAIT, td);
1648 	} else {
1649 		/*
1650 		 * Locate and synchronously write any buffers that fall
1651 		 * into the requested range.  Note:  we are assuming that
1652 		 * f_iosize is a power of 2.
1653 		 */
1654 		int iosize = vp->v_mount->mnt_stat.f_iosize;
1655 		int iomask = iosize - 1;
1656 		struct bufobj *bo;
1657 		daddr_t lblkno;
1658 
1659 		/*
1660 		 * Align to iosize boundary, super-align to page boundary.
1661 		 */
1662 		if (off & iomask) {
1663 			cnt += off & iomask;
1664 			off &= ~(u_quad_t)iomask;
1665 		}
1666 		if (off & PAGE_MASK) {
1667 			cnt += off & PAGE_MASK;
1668 			off &= ~(u_quad_t)PAGE_MASK;
1669 		}
1670 		lblkno = off / iosize;
1671 
1672 		if (vp->v_object && vm_object_mightbedirty(vp->v_object)) {
1673 			VM_OBJECT_WLOCK(vp->v_object);
1674 			vm_object_page_clean(vp->v_object, off, off + cnt,
1675 			    OBJPC_SYNC);
1676 			VM_OBJECT_WUNLOCK(vp->v_object);
1677 		}
1678 
1679 		bo = &vp->v_bufobj;
1680 		BO_LOCK(bo);
1681 		while (cnt > 0) {
1682 			struct buf *bp;
1683 
1684 			/*
1685 			 * If we have a buffer and it is marked B_DELWRI we
1686 			 * have to lock and write it.  Otherwise the prior
1687 			 * write is assumed to have already been committed.
1688 			 *
1689 			 * gbincore() can return invalid buffers now so we
1690 			 * have to check that bit as well (though B_DELWRI
1691 			 * should not be set if B_INVAL is set there could be
1692 			 * a race here since we haven't locked the buffer).
1693 			 */
1694 			if ((bp = gbincore(&vp->v_bufobj, lblkno)) != NULL) {
1695 				if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_SLEEPFAIL |
1696 				    LK_INTERLOCK, BO_LOCKPTR(bo)) == ENOLCK) {
1697 					BO_LOCK(bo);
1698 					continue; /* retry */
1699 				}
1700 			    	if ((bp->b_flags & (B_DELWRI|B_INVAL)) ==
1701 				    B_DELWRI) {
1702 					bremfree(bp);
1703 					bp->b_flags &= ~B_ASYNC;
1704 					bwrite(bp);
1705 					++nfs_commit_miss;
1706 				} else
1707 					BUF_UNLOCK(bp);
1708 				BO_LOCK(bo);
1709 			}
1710 			++nfs_commit_blks;
1711 			if (cnt < iosize)
1712 				break;
1713 			cnt -= iosize;
1714 			++lblkno;
1715 		}
1716 		BO_UNLOCK(bo);
1717 	}
1718 	NFSEXITCODE(error);
1719 	return (error);
1720 }
1721 
1722 /*
1723  * Statfs vnode op.
1724  */
1725 int
1726 nfsvno_statfs(struct vnode *vp, struct statfs *sf)
1727 {
1728 	struct statfs *tsf;
1729 	int error;
1730 
1731 	tsf = NULL;
1732 	if (nfsrv_devidcnt > 0) {
1733 		/* For a pNFS service, get the DS numbers. */
1734 		tsf = malloc(sizeof(*tsf), M_TEMP, M_WAITOK | M_ZERO);
1735 		error = nfsrv_pnfsstatfs(tsf, vp->v_mount);
1736 		if (error != 0) {
1737 			free(tsf, M_TEMP);
1738 			tsf = NULL;
1739 		}
1740 	}
1741 	error = VFS_STATFS(vp->v_mount, sf);
1742 	if (error == 0) {
1743 		if (tsf != NULL) {
1744 			sf->f_blocks = tsf->f_blocks;
1745 			sf->f_bavail = tsf->f_bavail;
1746 			sf->f_bfree = tsf->f_bfree;
1747 			sf->f_bsize = tsf->f_bsize;
1748 		}
1749 		/*
1750 		 * Since NFS handles these values as unsigned on the
1751 		 * wire, there is no way to represent negative values,
1752 		 * so set them to 0. Without this, they will appear
1753 		 * to be very large positive values for clients like
1754 		 * Solaris10.
1755 		 */
1756 		if (sf->f_bavail < 0)
1757 			sf->f_bavail = 0;
1758 		if (sf->f_ffree < 0)
1759 			sf->f_ffree = 0;
1760 	}
1761 	free(tsf, M_TEMP);
1762 	NFSEXITCODE(error);
1763 	return (error);
1764 }
1765 
1766 /*
1767  * Do the vnode op stuff for Open. Similar to nfsvno_createsub(), but
1768  * must handle nfsrv_opencheck() calls after any other access checks.
1769  */
1770 void
1771 nfsvno_open(struct nfsrv_descript *nd, struct nameidata *ndp,
1772     nfsquad_t clientid, nfsv4stateid_t *stateidp, struct nfsstate *stp,
1773     int *exclusive_flagp, struct nfsvattr *nvap, int32_t *cverf, int create,
1774     NFSACL_T *aclp, nfsattrbit_t *attrbitp, struct ucred *cred,
1775     struct nfsexstuff *exp, struct vnode **vpp)
1776 {
1777 	struct vnode *vp = NULL;
1778 	u_quad_t tempsize;
1779 	struct nfsexstuff nes;
1780 	struct thread *p = curthread;
1781 
1782 	if (ndp->ni_vp == NULL)
1783 		nd->nd_repstat = nfsrv_opencheck(clientid,
1784 		    stateidp, stp, NULL, nd, p, nd->nd_repstat);
1785 	if (!nd->nd_repstat) {
1786 		if (ndp->ni_vp == NULL) {
1787 			vrele(ndp->ni_startdir);
1788 			nd->nd_repstat = VOP_CREATE(ndp->ni_dvp,
1789 			    &ndp->ni_vp, &ndp->ni_cnd, &nvap->na_vattr);
1790 			/* For a pNFS server, create the data file on a DS. */
1791 			if (nd->nd_repstat == 0) {
1792 				/*
1793 				 * Create a data file on a DS for a pNFS server.
1794 				 * This function just returns if not
1795 				 * running a pNFS DS or the creation fails.
1796 				 */
1797 				nfsrv_pnfscreate(ndp->ni_vp, &nvap->na_vattr,
1798 				    cred, p);
1799 			}
1800 			VOP_VPUT_PAIR(ndp->ni_dvp, nd->nd_repstat == 0 ?
1801 			    &ndp->ni_vp : NULL, false);
1802 			nfsvno_relpathbuf(ndp);
1803 			if (!nd->nd_repstat) {
1804 				if (*exclusive_flagp) {
1805 					*exclusive_flagp = 0;
1806 					NFSVNO_ATTRINIT(nvap);
1807 					nvap->na_atime.tv_sec = cverf[0];
1808 					nvap->na_atime.tv_nsec = cverf[1];
1809 					nd->nd_repstat = VOP_SETATTR(ndp->ni_vp,
1810 					    &nvap->na_vattr, cred);
1811 					if (nd->nd_repstat != 0) {
1812 						vput(ndp->ni_vp);
1813 						ndp->ni_vp = NULL;
1814 						nd->nd_repstat = NFSERR_NOTSUPP;
1815 					} else
1816 						NFSSETBIT_ATTRBIT(attrbitp,
1817 						    NFSATTRBIT_TIMEACCESS);
1818 				} else {
1819 					nfsrv_fixattr(nd, ndp->ni_vp, nvap,
1820 					    aclp, p, attrbitp, exp);
1821 				}
1822 			}
1823 			vp = ndp->ni_vp;
1824 		} else {
1825 			if (ndp->ni_startdir)
1826 				vrele(ndp->ni_startdir);
1827 			nfsvno_relpathbuf(ndp);
1828 			vp = ndp->ni_vp;
1829 			if (create == NFSV4OPEN_CREATE) {
1830 				if (ndp->ni_dvp == vp)
1831 					vrele(ndp->ni_dvp);
1832 				else
1833 					vput(ndp->ni_dvp);
1834 			}
1835 			if (NFSVNO_ISSETSIZE(nvap) && vp->v_type == VREG) {
1836 				if (ndp->ni_cnd.cn_flags & RDONLY)
1837 					NFSVNO_SETEXRDONLY(&nes);
1838 				else
1839 					NFSVNO_EXINIT(&nes);
1840 				nd->nd_repstat = nfsvno_accchk(vp,
1841 				    VWRITE, cred, &nes, p,
1842 				    NFSACCCHK_NOOVERRIDE,
1843 				    NFSACCCHK_VPISLOCKED, NULL);
1844 				nd->nd_repstat = nfsrv_opencheck(clientid,
1845 				    stateidp, stp, vp, nd, p, nd->nd_repstat);
1846 				if (!nd->nd_repstat) {
1847 					tempsize = nvap->na_size;
1848 					NFSVNO_ATTRINIT(nvap);
1849 					nvap->na_size = tempsize;
1850 					nd->nd_repstat = VOP_SETATTR(vp,
1851 					    &nvap->na_vattr, cred);
1852 				}
1853 			} else if (vp->v_type == VREG) {
1854 				nd->nd_repstat = nfsrv_opencheck(clientid,
1855 				    stateidp, stp, vp, nd, p, nd->nd_repstat);
1856 			}
1857 		}
1858 	} else {
1859 		if (ndp->ni_cnd.cn_flags & HASBUF)
1860 			nfsvno_relpathbuf(ndp);
1861 		if (ndp->ni_startdir && create == NFSV4OPEN_CREATE) {
1862 			vrele(ndp->ni_startdir);
1863 			if (ndp->ni_dvp == ndp->ni_vp)
1864 				vrele(ndp->ni_dvp);
1865 			else
1866 				vput(ndp->ni_dvp);
1867 			if (ndp->ni_vp)
1868 				vput(ndp->ni_vp);
1869 		}
1870 	}
1871 	*vpp = vp;
1872 
1873 	NFSEXITCODE2(0, nd);
1874 }
1875 
1876 /*
1877  * Updates the file rev and sets the mtime and ctime
1878  * to the current clock time, returning the va_filerev and va_Xtime
1879  * values.
1880  * Return ESTALE to indicate the vnode is VIRF_DOOMED.
1881  */
1882 int
1883 nfsvno_updfilerev(struct vnode *vp, struct nfsvattr *nvap,
1884     struct nfsrv_descript *nd, struct thread *p)
1885 {
1886 	struct vattr va;
1887 
1888 	VATTR_NULL(&va);
1889 	vfs_timestamp(&va.va_mtime);
1890 	if (NFSVOPISLOCKED(vp) != LK_EXCLUSIVE) {
1891 		NFSVOPLOCK(vp, LK_UPGRADE | LK_RETRY);
1892 		if (VN_IS_DOOMED(vp))
1893 			return (ESTALE);
1894 	}
1895 	(void) VOP_SETATTR(vp, &va, nd->nd_cred);
1896 	(void) nfsvno_getattr(vp, nvap, nd, p, 1, NULL);
1897 	return (0);
1898 }
1899 
1900 /*
1901  * Glue routine to nfsv4_fillattr().
1902  */
1903 int
1904 nfsvno_fillattr(struct nfsrv_descript *nd, struct mount *mp, struct vnode *vp,
1905     struct nfsvattr *nvap, fhandle_t *fhp, int rderror, nfsattrbit_t *attrbitp,
1906     struct ucred *cred, struct thread *p, int isdgram, int reterr,
1907     int supports_nfsv4acls, int at_root, uint64_t mounted_on_fileno)
1908 {
1909 	struct statfs *sf;
1910 	int error;
1911 
1912 	sf = NULL;
1913 	if (nfsrv_devidcnt > 0 &&
1914 	    (NFSISSET_ATTRBIT(attrbitp, NFSATTRBIT_SPACEAVAIL) ||
1915 	     NFSISSET_ATTRBIT(attrbitp, NFSATTRBIT_SPACEFREE) ||
1916 	     NFSISSET_ATTRBIT(attrbitp, NFSATTRBIT_SPACETOTAL))) {
1917 		sf = malloc(sizeof(*sf), M_TEMP, M_WAITOK | M_ZERO);
1918 		error = nfsrv_pnfsstatfs(sf, mp);
1919 		if (error != 0) {
1920 			free(sf, M_TEMP);
1921 			sf = NULL;
1922 		}
1923 	}
1924 	error = nfsv4_fillattr(nd, mp, vp, NULL, &nvap->na_vattr, fhp, rderror,
1925 	    attrbitp, cred, p, isdgram, reterr, supports_nfsv4acls, at_root,
1926 	    mounted_on_fileno, sf);
1927 	free(sf, M_TEMP);
1928 	NFSEXITCODE2(0, nd);
1929 	return (error);
1930 }
1931 
1932 /* Since the Readdir vnode ops vary, put the entire functions in here. */
1933 /*
1934  * nfs readdir service
1935  * - mallocs what it thinks is enough to read
1936  *	count rounded up to a multiple of DIRBLKSIZ <= NFS_MAXREADDIR
1937  * - calls VOP_READDIR()
1938  * - loops around building the reply
1939  *	if the output generated exceeds count break out of loop
1940  *	The NFSM_CLGET macro is used here so that the reply will be packed
1941  *	tightly in mbuf clusters.
1942  * - it trims out records with d_fileno == 0
1943  *	this doesn't matter for Unix clients, but they might confuse clients
1944  *	for other os'.
1945  * - it trims out records with d_type == DT_WHT
1946  *	these cannot be seen through NFS (unless we extend the protocol)
1947  *     The alternate call nfsrvd_readdirplus() does lookups as well.
1948  * PS: The NFS protocol spec. does not clarify what the "count" byte
1949  *	argument is a count of.. just name strings and file id's or the
1950  *	entire reply rpc or ...
1951  *	I tried just file name and id sizes and it confused the Sun client,
1952  *	so I am using the full rpc size now. The "paranoia.." comment refers
1953  *	to including the status longwords that are not a part of the dir.
1954  *	"entry" structures, but are in the rpc.
1955  */
1956 int
1957 nfsrvd_readdir(struct nfsrv_descript *nd, int isdgram,
1958     struct vnode *vp, struct nfsexstuff *exp)
1959 {
1960 	struct dirent *dp;
1961 	u_int32_t *tl;
1962 	int dirlen;
1963 	char *cpos, *cend, *rbuf;
1964 	struct nfsvattr at;
1965 	int nlen, error = 0, getret = 1;
1966 	int siz, cnt, fullsiz, eofflag, ncookies;
1967 	u_int64_t off, toff, verf __unused;
1968 	u_long *cookies = NULL, *cookiep;
1969 	struct uio io;
1970 	struct iovec iv;
1971 	int is_ufs;
1972 	struct thread *p = curthread;
1973 
1974 	if (nd->nd_repstat) {
1975 		nfsrv_postopattr(nd, getret, &at);
1976 		goto out;
1977 	}
1978 	if (nd->nd_flag & ND_NFSV2) {
1979 		NFSM_DISSECT(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
1980 		off = fxdr_unsigned(u_quad_t, *tl++);
1981 	} else {
1982 		NFSM_DISSECT(tl, u_int32_t *, 5 * NFSX_UNSIGNED);
1983 		off = fxdr_hyper(tl);
1984 		tl += 2;
1985 		verf = fxdr_hyper(tl);
1986 		tl += 2;
1987 	}
1988 	toff = off;
1989 	cnt = fxdr_unsigned(int, *tl);
1990 	if (cnt > NFS_SRVMAXDATA(nd) || cnt < 0)
1991 		cnt = NFS_SRVMAXDATA(nd);
1992 	siz = ((cnt + DIRBLKSIZ - 1) & ~(DIRBLKSIZ - 1));
1993 	fullsiz = siz;
1994 	if (nd->nd_flag & ND_NFSV3) {
1995 		nd->nd_repstat = getret = nfsvno_getattr(vp, &at, nd, p, 1,
1996 		    NULL);
1997 #if 0
1998 		/*
1999 		 * va_filerev is not sufficient as a cookie verifier,
2000 		 * since it is not supposed to change when entries are
2001 		 * removed/added unless that offset cookies returned to
2002 		 * the client are no longer valid.
2003 		 */
2004 		if (!nd->nd_repstat && toff && verf != at.na_filerev)
2005 			nd->nd_repstat = NFSERR_BAD_COOKIE;
2006 #endif
2007 	}
2008 	if (!nd->nd_repstat && vp->v_type != VDIR)
2009 		nd->nd_repstat = NFSERR_NOTDIR;
2010 	if (nd->nd_repstat == 0 && cnt == 0) {
2011 		if (nd->nd_flag & ND_NFSV2)
2012 			/* NFSv2 does not have NFSERR_TOOSMALL */
2013 			nd->nd_repstat = EPERM;
2014 		else
2015 			nd->nd_repstat = NFSERR_TOOSMALL;
2016 	}
2017 	if (!nd->nd_repstat)
2018 		nd->nd_repstat = nfsvno_accchk(vp, VEXEC,
2019 		    nd->nd_cred, exp, p, NFSACCCHK_NOOVERRIDE,
2020 		    NFSACCCHK_VPISLOCKED, NULL);
2021 	if (nd->nd_repstat) {
2022 		vput(vp);
2023 		if (nd->nd_flag & ND_NFSV3)
2024 			nfsrv_postopattr(nd, getret, &at);
2025 		goto out;
2026 	}
2027 	is_ufs = strcmp(vp->v_mount->mnt_vfc->vfc_name, "ufs") == 0;
2028 	rbuf = malloc(siz, M_TEMP, M_WAITOK);
2029 again:
2030 	eofflag = 0;
2031 	if (cookies) {
2032 		free(cookies, M_TEMP);
2033 		cookies = NULL;
2034 	}
2035 
2036 	iv.iov_base = rbuf;
2037 	iv.iov_len = siz;
2038 	io.uio_iov = &iv;
2039 	io.uio_iovcnt = 1;
2040 	io.uio_offset = (off_t)off;
2041 	io.uio_resid = siz;
2042 	io.uio_segflg = UIO_SYSSPACE;
2043 	io.uio_rw = UIO_READ;
2044 	io.uio_td = NULL;
2045 	nd->nd_repstat = VOP_READDIR(vp, &io, nd->nd_cred, &eofflag, &ncookies,
2046 	    &cookies);
2047 	off = (u_int64_t)io.uio_offset;
2048 	if (io.uio_resid)
2049 		siz -= io.uio_resid;
2050 
2051 	if (!cookies && !nd->nd_repstat)
2052 		nd->nd_repstat = NFSERR_PERM;
2053 	if (nd->nd_flag & ND_NFSV3) {
2054 		getret = nfsvno_getattr(vp, &at, nd, p, 1, NULL);
2055 		if (!nd->nd_repstat)
2056 			nd->nd_repstat = getret;
2057 	}
2058 
2059 	/*
2060 	 * Handles the failed cases. nd->nd_repstat == 0 past here.
2061 	 */
2062 	if (nd->nd_repstat) {
2063 		vput(vp);
2064 		free(rbuf, M_TEMP);
2065 		if (cookies)
2066 			free(cookies, M_TEMP);
2067 		if (nd->nd_flag & ND_NFSV3)
2068 			nfsrv_postopattr(nd, getret, &at);
2069 		goto out;
2070 	}
2071 	/*
2072 	 * If nothing read, return eof
2073 	 * rpc reply
2074 	 */
2075 	if (siz == 0) {
2076 		vput(vp);
2077 		if (nd->nd_flag & ND_NFSV2) {
2078 			NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
2079 		} else {
2080 			nfsrv_postopattr(nd, getret, &at);
2081 			NFSM_BUILD(tl, u_int32_t *, 4 * NFSX_UNSIGNED);
2082 			txdr_hyper(at.na_filerev, tl);
2083 			tl += 2;
2084 		}
2085 		*tl++ = newnfs_false;
2086 		*tl = newnfs_true;
2087 		free(rbuf, M_TEMP);
2088 		free(cookies, M_TEMP);
2089 		goto out;
2090 	}
2091 
2092 	/*
2093 	 * Check for degenerate cases of nothing useful read.
2094 	 * If so go try again
2095 	 */
2096 	cpos = rbuf;
2097 	cend = rbuf + siz;
2098 	dp = (struct dirent *)cpos;
2099 	cookiep = cookies;
2100 
2101 	/*
2102 	 * For some reason FreeBSD's ufs_readdir() chooses to back the
2103 	 * directory offset up to a block boundary, so it is necessary to
2104 	 * skip over the records that precede the requested offset. This
2105 	 * requires the assumption that file offset cookies monotonically
2106 	 * increase.
2107 	 */
2108 	while (cpos < cend && ncookies > 0 &&
2109 	    (dp->d_fileno == 0 || dp->d_type == DT_WHT ||
2110 	     (is_ufs == 1 && ((u_quad_t)(*cookiep)) <= toff))) {
2111 		cpos += dp->d_reclen;
2112 		dp = (struct dirent *)cpos;
2113 		cookiep++;
2114 		ncookies--;
2115 	}
2116 	if (cpos >= cend || ncookies == 0) {
2117 		siz = fullsiz;
2118 		toff = off;
2119 		goto again;
2120 	}
2121 	vput(vp);
2122 
2123 	/*
2124 	 * If cnt > MCLBYTES and the reply will not be saved, use
2125 	 * ext_pgs mbufs for TLS.
2126 	 * For NFSv4.0, we do not know for sure if the reply will
2127 	 * be saved, so do not use ext_pgs mbufs for NFSv4.0.
2128 	 */
2129 	if (cnt > MCLBYTES && siz > MCLBYTES &&
2130 	    (nd->nd_flag & (ND_TLS | ND_EXTPG | ND_SAVEREPLY)) == ND_TLS &&
2131 	    (nd->nd_flag & (ND_NFSV4 | ND_NFSV41)) != ND_NFSV4)
2132 		nd->nd_flag |= ND_EXTPG;
2133 
2134 	/*
2135 	 * dirlen is the size of the reply, including all XDR and must
2136 	 * not exceed cnt. For NFSv2, RFC1094 didn't clearly indicate
2137 	 * if the XDR should be included in "count", but to be safe, we do.
2138 	 * (Include the two booleans at the end of the reply in dirlen now.)
2139 	 */
2140 	if (nd->nd_flag & ND_NFSV3) {
2141 		nfsrv_postopattr(nd, getret, &at);
2142 		NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
2143 		txdr_hyper(at.na_filerev, tl);
2144 		dirlen = NFSX_V3POSTOPATTR + NFSX_VERF + 2 * NFSX_UNSIGNED;
2145 	} else {
2146 		dirlen = 2 * NFSX_UNSIGNED;
2147 	}
2148 
2149 	/* Loop through the records and build reply */
2150 	while (cpos < cend && ncookies > 0) {
2151 		nlen = dp->d_namlen;
2152 		if (dp->d_fileno != 0 && dp->d_type != DT_WHT &&
2153 			nlen <= NFS_MAXNAMLEN) {
2154 			if (nd->nd_flag & ND_NFSV3)
2155 				dirlen += (6*NFSX_UNSIGNED + NFSM_RNDUP(nlen));
2156 			else
2157 				dirlen += (4*NFSX_UNSIGNED + NFSM_RNDUP(nlen));
2158 			if (dirlen > cnt) {
2159 				eofflag = 0;
2160 				break;
2161 			}
2162 
2163 			/*
2164 			 * Build the directory record xdr from
2165 			 * the dirent entry.
2166 			 */
2167 			if (nd->nd_flag & ND_NFSV3) {
2168 				NFSM_BUILD(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
2169 				*tl++ = newnfs_true;
2170 				*tl++ = 0;
2171 			} else {
2172 				NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
2173 				*tl++ = newnfs_true;
2174 			}
2175 			*tl = txdr_unsigned(dp->d_fileno);
2176 			(void) nfsm_strtom(nd, dp->d_name, nlen);
2177 			if (nd->nd_flag & ND_NFSV3) {
2178 				NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
2179 				*tl++ = 0;
2180 			} else
2181 				NFSM_BUILD(tl, u_int32_t *, NFSX_UNSIGNED);
2182 			*tl = txdr_unsigned(*cookiep);
2183 		}
2184 		cpos += dp->d_reclen;
2185 		dp = (struct dirent *)cpos;
2186 		cookiep++;
2187 		ncookies--;
2188 	}
2189 	if (cpos < cend)
2190 		eofflag = 0;
2191 	NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
2192 	*tl++ = newnfs_false;
2193 	if (eofflag)
2194 		*tl = newnfs_true;
2195 	else
2196 		*tl = newnfs_false;
2197 	free(rbuf, M_TEMP);
2198 	free(cookies, M_TEMP);
2199 
2200 out:
2201 	NFSEXITCODE2(0, nd);
2202 	return (0);
2203 nfsmout:
2204 	vput(vp);
2205 	NFSEXITCODE2(error, nd);
2206 	return (error);
2207 }
2208 
2209 /*
2210  * Readdirplus for V3 and Readdir for V4.
2211  */
2212 int
2213 nfsrvd_readdirplus(struct nfsrv_descript *nd, int isdgram,
2214     struct vnode *vp, struct nfsexstuff *exp)
2215 {
2216 	struct dirent *dp;
2217 	u_int32_t *tl;
2218 	int dirlen;
2219 	char *cpos, *cend, *rbuf;
2220 	struct vnode *nvp;
2221 	fhandle_t nfh;
2222 	struct nfsvattr nva, at, *nvap = &nva;
2223 	struct mbuf *mb0, *mb1;
2224 	struct nfsreferral *refp;
2225 	int nlen, r, error = 0, getret = 1, usevget = 1;
2226 	int siz, cnt, fullsiz, eofflag, ncookies, entrycnt;
2227 	caddr_t bpos0, bpos1;
2228 	u_int64_t off, toff, verf;
2229 	u_long *cookies = NULL, *cookiep;
2230 	nfsattrbit_t attrbits, rderrbits, savbits;
2231 	struct uio io;
2232 	struct iovec iv;
2233 	struct componentname cn;
2234 	int at_root, is_ufs, is_zfs, needs_unbusy, supports_nfsv4acls;
2235 	struct mount *mp, *new_mp;
2236 	uint64_t mounted_on_fileno;
2237 	struct thread *p = curthread;
2238 	int bextpg0, bextpg1, bextpgsiz0, bextpgsiz1;
2239 
2240 	if (nd->nd_repstat) {
2241 		nfsrv_postopattr(nd, getret, &at);
2242 		goto out;
2243 	}
2244 	NFSM_DISSECT(tl, u_int32_t *, 6 * NFSX_UNSIGNED);
2245 	off = fxdr_hyper(tl);
2246 	toff = off;
2247 	tl += 2;
2248 	verf = fxdr_hyper(tl);
2249 	tl += 2;
2250 	siz = fxdr_unsigned(int, *tl++);
2251 	cnt = fxdr_unsigned(int, *tl);
2252 
2253 	/*
2254 	 * Use the server's maximum data transfer size as the upper bound
2255 	 * on reply datalen.
2256 	 */
2257 	if (cnt > NFS_SRVMAXDATA(nd) || cnt < 0)
2258 		cnt = NFS_SRVMAXDATA(nd);
2259 
2260 	/*
2261 	 * siz is a "hint" of how much directory information (name, fileid,
2262 	 * cookie) should be in the reply. At least one client "hints" 0,
2263 	 * so I set it to cnt for that case. I also round it up to the
2264 	 * next multiple of DIRBLKSIZ.
2265 	 * Since the size of a Readdirplus directory entry reply will always
2266 	 * be greater than a directory entry returned by VOP_READDIR(), it
2267 	 * does not make sense to read more than NFS_SRVMAXDATA() via
2268 	 * VOP_READDIR().
2269 	 */
2270 	if (siz <= 0)
2271 		siz = cnt;
2272 	else if (siz > NFS_SRVMAXDATA(nd))
2273 		siz = NFS_SRVMAXDATA(nd);
2274 	siz = ((siz + DIRBLKSIZ - 1) & ~(DIRBLKSIZ - 1));
2275 
2276 	if (nd->nd_flag & ND_NFSV4) {
2277 		error = nfsrv_getattrbits(nd, &attrbits, NULL, NULL);
2278 		if (error)
2279 			goto nfsmout;
2280 		NFSSET_ATTRBIT(&savbits, &attrbits);
2281 		NFSCLRNOTFILLABLE_ATTRBIT(&attrbits, nd);
2282 		NFSZERO_ATTRBIT(&rderrbits);
2283 		NFSSETBIT_ATTRBIT(&rderrbits, NFSATTRBIT_RDATTRERROR);
2284 	} else {
2285 		NFSZERO_ATTRBIT(&attrbits);
2286 	}
2287 	fullsiz = siz;
2288 	nd->nd_repstat = getret = nfsvno_getattr(vp, &at, nd, p, 1, NULL);
2289 #if 0
2290 	if (!nd->nd_repstat) {
2291 	    if (off && verf != at.na_filerev) {
2292 		/*
2293 		 * va_filerev is not sufficient as a cookie verifier,
2294 		 * since it is not supposed to change when entries are
2295 		 * removed/added unless that offset cookies returned to
2296 		 * the client are no longer valid.
2297 		 */
2298 		if (nd->nd_flag & ND_NFSV4) {
2299 			nd->nd_repstat = NFSERR_NOTSAME;
2300 		} else {
2301 			nd->nd_repstat = NFSERR_BAD_COOKIE;
2302 		}
2303 	    }
2304 	}
2305 #endif
2306 	if (!nd->nd_repstat && vp->v_type != VDIR)
2307 		nd->nd_repstat = NFSERR_NOTDIR;
2308 	if (!nd->nd_repstat && cnt == 0)
2309 		nd->nd_repstat = NFSERR_TOOSMALL;
2310 	if (!nd->nd_repstat)
2311 		nd->nd_repstat = nfsvno_accchk(vp, VEXEC,
2312 		    nd->nd_cred, exp, p, NFSACCCHK_NOOVERRIDE,
2313 		    NFSACCCHK_VPISLOCKED, NULL);
2314 	if (nd->nd_repstat) {
2315 		vput(vp);
2316 		if (nd->nd_flag & ND_NFSV3)
2317 			nfsrv_postopattr(nd, getret, &at);
2318 		goto out;
2319 	}
2320 	is_ufs = strcmp(vp->v_mount->mnt_vfc->vfc_name, "ufs") == 0;
2321 	is_zfs = strcmp(vp->v_mount->mnt_vfc->vfc_name, "zfs") == 0;
2322 
2323 	rbuf = malloc(siz, M_TEMP, M_WAITOK);
2324 again:
2325 	eofflag = 0;
2326 	if (cookies) {
2327 		free(cookies, M_TEMP);
2328 		cookies = NULL;
2329 	}
2330 
2331 	iv.iov_base = rbuf;
2332 	iv.iov_len = siz;
2333 	io.uio_iov = &iv;
2334 	io.uio_iovcnt = 1;
2335 	io.uio_offset = (off_t)off;
2336 	io.uio_resid = siz;
2337 	io.uio_segflg = UIO_SYSSPACE;
2338 	io.uio_rw = UIO_READ;
2339 	io.uio_td = NULL;
2340 	nd->nd_repstat = VOP_READDIR(vp, &io, nd->nd_cred, &eofflag, &ncookies,
2341 	    &cookies);
2342 	off = (u_int64_t)io.uio_offset;
2343 	if (io.uio_resid)
2344 		siz -= io.uio_resid;
2345 
2346 	getret = nfsvno_getattr(vp, &at, nd, p, 1, NULL);
2347 
2348 	if (!cookies && !nd->nd_repstat)
2349 		nd->nd_repstat = NFSERR_PERM;
2350 	if (!nd->nd_repstat)
2351 		nd->nd_repstat = getret;
2352 	if (nd->nd_repstat) {
2353 		vput(vp);
2354 		if (cookies)
2355 			free(cookies, M_TEMP);
2356 		free(rbuf, M_TEMP);
2357 		if (nd->nd_flag & ND_NFSV3)
2358 			nfsrv_postopattr(nd, getret, &at);
2359 		goto out;
2360 	}
2361 	/*
2362 	 * If nothing read, return eof
2363 	 * rpc reply
2364 	 */
2365 	if (siz == 0) {
2366 		vput(vp);
2367 		if (nd->nd_flag & ND_NFSV3)
2368 			nfsrv_postopattr(nd, getret, &at);
2369 		NFSM_BUILD(tl, u_int32_t *, 4 * NFSX_UNSIGNED);
2370 		txdr_hyper(at.na_filerev, tl);
2371 		tl += 2;
2372 		*tl++ = newnfs_false;
2373 		*tl = newnfs_true;
2374 		free(cookies, M_TEMP);
2375 		free(rbuf, M_TEMP);
2376 		goto out;
2377 	}
2378 
2379 	/*
2380 	 * Check for degenerate cases of nothing useful read.
2381 	 * If so go try again
2382 	 */
2383 	cpos = rbuf;
2384 	cend = rbuf + siz;
2385 	dp = (struct dirent *)cpos;
2386 	cookiep = cookies;
2387 
2388 	/*
2389 	 * For some reason FreeBSD's ufs_readdir() chooses to back the
2390 	 * directory offset up to a block boundary, so it is necessary to
2391 	 * skip over the records that precede the requested offset. This
2392 	 * requires the assumption that file offset cookies monotonically
2393 	 * increase.
2394 	 */
2395 	while (cpos < cend && ncookies > 0 &&
2396 	  (dp->d_fileno == 0 || dp->d_type == DT_WHT ||
2397 	   (is_ufs == 1 && ((u_quad_t)(*cookiep)) <= toff) ||
2398 	   ((nd->nd_flag & ND_NFSV4) &&
2399 	    ((dp->d_namlen == 1 && dp->d_name[0] == '.') ||
2400 	     (dp->d_namlen==2 && dp->d_name[0]=='.' && dp->d_name[1]=='.'))))) {
2401 		cpos += dp->d_reclen;
2402 		dp = (struct dirent *)cpos;
2403 		cookiep++;
2404 		ncookies--;
2405 	}
2406 	if (cpos >= cend || ncookies == 0) {
2407 		siz = fullsiz;
2408 		toff = off;
2409 		goto again;
2410 	}
2411 
2412 	/*
2413 	 * Busy the file system so that the mount point won't go away
2414 	 * and, as such, VFS_VGET() can be used safely.
2415 	 */
2416 	mp = vp->v_mount;
2417 	vfs_ref(mp);
2418 	NFSVOPUNLOCK(vp);
2419 	nd->nd_repstat = vfs_busy(mp, 0);
2420 	vfs_rel(mp);
2421 	if (nd->nd_repstat != 0) {
2422 		vrele(vp);
2423 		free(cookies, M_TEMP);
2424 		free(rbuf, M_TEMP);
2425 		if (nd->nd_flag & ND_NFSV3)
2426 			nfsrv_postopattr(nd, getret, &at);
2427 		goto out;
2428 	}
2429 
2430 	/*
2431 	 * Check to see if entries in this directory can be safely acquired
2432 	 * via VFS_VGET() or if a switch to VOP_LOOKUP() is required.
2433 	 * ZFS snapshot directories need VOP_LOOKUP(), so that any
2434 	 * automount of the snapshot directory that is required will
2435 	 * be done.
2436 	 * This needs to be done here for NFSv4, since NFSv4 never does
2437 	 * a VFS_VGET() for "." or "..".
2438 	 */
2439 	if (is_zfs == 1) {
2440 		r = VFS_VGET(mp, at.na_fileid, LK_SHARED, &nvp);
2441 		if (r == EOPNOTSUPP) {
2442 			usevget = 0;
2443 			cn.cn_nameiop = LOOKUP;
2444 			cn.cn_lkflags = LK_SHARED | LK_RETRY;
2445 			cn.cn_cred = nd->nd_cred;
2446 			cn.cn_thread = p;
2447 		} else if (r == 0)
2448 			vput(nvp);
2449 	}
2450 
2451 	/*
2452 	 * If the reply is likely to exceed MCLBYTES and the reply will
2453 	 * not be saved, use ext_pgs mbufs for TLS.
2454 	 * It is difficult to predict how large each entry will be and
2455 	 * how many entries have been read, so just assume the directory
2456 	 * entries grow by a factor of 4 when attributes are included.
2457 	 * For NFSv4.0, we do not know for sure if the reply will
2458 	 * be saved, so do not use ext_pgs mbufs for NFSv4.0.
2459 	 */
2460 	if (cnt > MCLBYTES && siz > MCLBYTES / 4 &&
2461 	    (nd->nd_flag & (ND_TLS | ND_EXTPG | ND_SAVEREPLY)) == ND_TLS &&
2462 	    (nd->nd_flag & (ND_NFSV4 | ND_NFSV41)) != ND_NFSV4)
2463 		nd->nd_flag |= ND_EXTPG;
2464 
2465 	/*
2466 	 * Save this position, in case there is an error before one entry
2467 	 * is created.
2468 	 */
2469 	mb0 = nd->nd_mb;
2470 	bpos0 = nd->nd_bpos;
2471 	bextpg0 = nd->nd_bextpg;
2472 	bextpgsiz0 = nd->nd_bextpgsiz;
2473 
2474 	/*
2475 	 * Fill in the first part of the reply.
2476 	 * dirlen is the reply length in bytes and cannot exceed cnt.
2477 	 * (Include the two booleans at the end of the reply in dirlen now,
2478 	 *  so we recognize when we have exceeded cnt.)
2479 	 */
2480 	if (nd->nd_flag & ND_NFSV3) {
2481 		dirlen = NFSX_V3POSTOPATTR + NFSX_VERF + 2 * NFSX_UNSIGNED;
2482 		nfsrv_postopattr(nd, getret, &at);
2483 	} else {
2484 		dirlen = NFSX_VERF + 2 * NFSX_UNSIGNED;
2485 	}
2486 	NFSM_BUILD(tl, u_int32_t *, NFSX_VERF);
2487 	txdr_hyper(at.na_filerev, tl);
2488 
2489 	/*
2490 	 * Save this position, in case there is an empty reply needed.
2491 	 */
2492 	mb1 = nd->nd_mb;
2493 	bpos1 = nd->nd_bpos;
2494 	bextpg1 = nd->nd_bextpg;
2495 	bextpgsiz1 = nd->nd_bextpgsiz;
2496 
2497 	/* Loop through the records and build reply */
2498 	entrycnt = 0;
2499 	while (cpos < cend && ncookies > 0 && dirlen < cnt) {
2500 		nlen = dp->d_namlen;
2501 		if (dp->d_fileno != 0 && dp->d_type != DT_WHT &&
2502 		    nlen <= NFS_MAXNAMLEN &&
2503 		    ((nd->nd_flag & ND_NFSV3) || nlen > 2 ||
2504 		     (nlen==2 && (dp->d_name[0]!='.' || dp->d_name[1]!='.'))
2505 		      || (nlen == 1 && dp->d_name[0] != '.'))) {
2506 			/*
2507 			 * Save the current position in the reply, in case
2508 			 * this entry exceeds cnt.
2509 			 */
2510 			mb1 = nd->nd_mb;
2511 			bpos1 = nd->nd_bpos;
2512 			bextpg1 = nd->nd_bextpg;
2513 			bextpgsiz1 = nd->nd_bextpgsiz;
2514 
2515 			/*
2516 			 * For readdir_and_lookup get the vnode using
2517 			 * the file number.
2518 			 */
2519 			nvp = NULL;
2520 			refp = NULL;
2521 			r = 0;
2522 			at_root = 0;
2523 			needs_unbusy = 0;
2524 			new_mp = mp;
2525 			mounted_on_fileno = (uint64_t)dp->d_fileno;
2526 			if ((nd->nd_flag & ND_NFSV3) ||
2527 			    NFSNONZERO_ATTRBIT(&savbits)) {
2528 				if (nd->nd_flag & ND_NFSV4)
2529 					refp = nfsv4root_getreferral(NULL,
2530 					    vp, dp->d_fileno);
2531 				if (refp == NULL) {
2532 					if (usevget)
2533 						r = VFS_VGET(mp, dp->d_fileno,
2534 						    LK_SHARED, &nvp);
2535 					else
2536 						r = EOPNOTSUPP;
2537 					if (r == EOPNOTSUPP) {
2538 						if (usevget) {
2539 							usevget = 0;
2540 							cn.cn_nameiop = LOOKUP;
2541 							cn.cn_lkflags =
2542 							    LK_SHARED |
2543 							    LK_RETRY;
2544 							cn.cn_cred =
2545 							    nd->nd_cred;
2546 							cn.cn_thread = p;
2547 						}
2548 						cn.cn_nameptr = dp->d_name;
2549 						cn.cn_namelen = nlen;
2550 						cn.cn_flags = ISLASTCN |
2551 						    NOFOLLOW | LOCKLEAF;
2552 						if (nlen == 2 &&
2553 						    dp->d_name[0] == '.' &&
2554 						    dp->d_name[1] == '.')
2555 							cn.cn_flags |=
2556 							    ISDOTDOT;
2557 						if (NFSVOPLOCK(vp, LK_SHARED)
2558 						    != 0) {
2559 							nd->nd_repstat = EPERM;
2560 							break;
2561 						}
2562 						if ((vp->v_vflag & VV_ROOT) != 0
2563 						    && (cn.cn_flags & ISDOTDOT)
2564 						    != 0) {
2565 							vref(vp);
2566 							nvp = vp;
2567 							r = 0;
2568 						} else {
2569 							r = VOP_LOOKUP(vp, &nvp,
2570 							    &cn);
2571 							if (vp != nvp)
2572 								NFSVOPUNLOCK(vp);
2573 						}
2574 					}
2575 
2576 					/*
2577 					 * For NFSv4, check to see if nvp is
2578 					 * a mount point and get the mount
2579 					 * point vnode, as required.
2580 					 */
2581 					if (r == 0 &&
2582 					    nfsrv_enable_crossmntpt != 0 &&
2583 					    (nd->nd_flag & ND_NFSV4) != 0 &&
2584 					    nvp->v_type == VDIR &&
2585 					    nvp->v_mountedhere != NULL) {
2586 						new_mp = nvp->v_mountedhere;
2587 						r = vfs_busy(new_mp, 0);
2588 						vput(nvp);
2589 						nvp = NULL;
2590 						if (r == 0) {
2591 							r = VFS_ROOT(new_mp,
2592 							    LK_SHARED, &nvp);
2593 							needs_unbusy = 1;
2594 							if (r == 0)
2595 								at_root = 1;
2596 						}
2597 					}
2598 				}
2599 
2600 				/*
2601 				 * If we failed to look up the entry, then it
2602 				 * has become invalid, most likely removed.
2603 				 */
2604 				if (r != 0) {
2605 					if (needs_unbusy)
2606 						vfs_unbusy(new_mp);
2607 					goto invalid;
2608 				}
2609 				KASSERT(refp != NULL || nvp != NULL,
2610 				    ("%s: undetected lookup error", __func__));
2611 
2612 				if (refp == NULL &&
2613 				    ((nd->nd_flag & ND_NFSV3) ||
2614 				     NFSNONZERO_ATTRBIT(&attrbits))) {
2615 					r = nfsvno_getfh(nvp, &nfh, p);
2616 					if (!r)
2617 					    r = nfsvno_getattr(nvp, nvap, nd, p,
2618 						1, &attrbits);
2619 					if (r == 0 && is_zfs == 1 &&
2620 					    nfsrv_enable_crossmntpt != 0 &&
2621 					    (nd->nd_flag & ND_NFSV4) != 0 &&
2622 					    nvp->v_type == VDIR &&
2623 					    vp->v_mount != nvp->v_mount) {
2624 					    /*
2625 					     * For a ZFS snapshot, there is a
2626 					     * pseudo mount that does not set
2627 					     * v_mountedhere, so it needs to
2628 					     * be detected via a different
2629 					     * mount structure.
2630 					     */
2631 					    at_root = 1;
2632 					    if (new_mp == mp)
2633 						new_mp = nvp->v_mount;
2634 					}
2635 				}
2636 
2637 				/*
2638 				 * If we failed to get attributes of the entry,
2639 				 * then just skip it for NFSv3 (the traditional
2640 				 * behavior in the old NFS server).
2641 				 * For NFSv4 the behavior is controlled by
2642 				 * RDATTRERROR: we either ignore the error or
2643 				 * fail the request.
2644 				 * Note that RDATTRERROR is never set for NFSv3.
2645 				 */
2646 				if (r != 0) {
2647 					if (!NFSISSET_ATTRBIT(&attrbits,
2648 					    NFSATTRBIT_RDATTRERROR)) {
2649 						vput(nvp);
2650 						if (needs_unbusy != 0)
2651 							vfs_unbusy(new_mp);
2652 						if ((nd->nd_flag & ND_NFSV3))
2653 							goto invalid;
2654 						nd->nd_repstat = r;
2655 						break;
2656 					}
2657 				}
2658 			}
2659 
2660 			/*
2661 			 * Build the directory record xdr
2662 			 */
2663 			if (nd->nd_flag & ND_NFSV3) {
2664 				NFSM_BUILD(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
2665 				*tl++ = newnfs_true;
2666 				*tl++ = 0;
2667 				*tl = txdr_unsigned(dp->d_fileno);
2668 				dirlen += nfsm_strtom(nd, dp->d_name, nlen);
2669 				NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
2670 				*tl++ = 0;
2671 				*tl = txdr_unsigned(*cookiep);
2672 				nfsrv_postopattr(nd, 0, nvap);
2673 				dirlen += nfsm_fhtom(nd,(u_int8_t *)&nfh,0,1);
2674 				dirlen += (5*NFSX_UNSIGNED+NFSX_V3POSTOPATTR);
2675 				if (nvp != NULL)
2676 					vput(nvp);
2677 			} else {
2678 				NFSM_BUILD(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
2679 				*tl++ = newnfs_true;
2680 				*tl++ = 0;
2681 				*tl = txdr_unsigned(*cookiep);
2682 				dirlen += nfsm_strtom(nd, dp->d_name, nlen);
2683 				if (nvp != NULL) {
2684 					supports_nfsv4acls =
2685 					    nfs_supportsnfsv4acls(nvp);
2686 					NFSVOPUNLOCK(nvp);
2687 				} else
2688 					supports_nfsv4acls = 0;
2689 				if (refp != NULL) {
2690 					dirlen += nfsrv_putreferralattr(nd,
2691 					    &savbits, refp, 0,
2692 					    &nd->nd_repstat);
2693 					if (nd->nd_repstat) {
2694 						if (nvp != NULL)
2695 							vrele(nvp);
2696 						if (needs_unbusy != 0)
2697 							vfs_unbusy(new_mp);
2698 						break;
2699 					}
2700 				} else if (r) {
2701 					dirlen += nfsvno_fillattr(nd, new_mp,
2702 					    nvp, nvap, &nfh, r, &rderrbits,
2703 					    nd->nd_cred, p, isdgram, 0,
2704 					    supports_nfsv4acls, at_root,
2705 					    mounted_on_fileno);
2706 				} else {
2707 					dirlen += nfsvno_fillattr(nd, new_mp,
2708 					    nvp, nvap, &nfh, r, &attrbits,
2709 					    nd->nd_cred, p, isdgram, 0,
2710 					    supports_nfsv4acls, at_root,
2711 					    mounted_on_fileno);
2712 				}
2713 				if (nvp != NULL)
2714 					vrele(nvp);
2715 				dirlen += (3 * NFSX_UNSIGNED);
2716 			}
2717 			if (needs_unbusy != 0)
2718 				vfs_unbusy(new_mp);
2719 			if (dirlen <= cnt)
2720 				entrycnt++;
2721 		}
2722 invalid:
2723 		cpos += dp->d_reclen;
2724 		dp = (struct dirent *)cpos;
2725 		cookiep++;
2726 		ncookies--;
2727 	}
2728 	vrele(vp);
2729 	vfs_unbusy(mp);
2730 
2731 	/*
2732 	 * If dirlen > cnt, we must strip off the last entry. If that
2733 	 * results in an empty reply, report NFSERR_TOOSMALL.
2734 	 */
2735 	if (dirlen > cnt || nd->nd_repstat) {
2736 		if (!nd->nd_repstat && entrycnt == 0)
2737 			nd->nd_repstat = NFSERR_TOOSMALL;
2738 		if (nd->nd_repstat) {
2739 			nfsm_trimtrailing(nd, mb0, bpos0, bextpg0, bextpgsiz0);
2740 			if (nd->nd_flag & ND_NFSV3)
2741 				nfsrv_postopattr(nd, getret, &at);
2742 		} else
2743 			nfsm_trimtrailing(nd, mb1, bpos1, bextpg1, bextpgsiz1);
2744 		eofflag = 0;
2745 	} else if (cpos < cend)
2746 		eofflag = 0;
2747 	if (!nd->nd_repstat) {
2748 		NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
2749 		*tl++ = newnfs_false;
2750 		if (eofflag)
2751 			*tl = newnfs_true;
2752 		else
2753 			*tl = newnfs_false;
2754 	}
2755 	free(cookies, M_TEMP);
2756 	free(rbuf, M_TEMP);
2757 
2758 out:
2759 	NFSEXITCODE2(0, nd);
2760 	return (0);
2761 nfsmout:
2762 	vput(vp);
2763 	NFSEXITCODE2(error, nd);
2764 	return (error);
2765 }
2766 
2767 /*
2768  * Get the settable attributes out of the mbuf list.
2769  * (Return 0 or EBADRPC)
2770  */
2771 int
2772 nfsrv_sattr(struct nfsrv_descript *nd, vnode_t vp, struct nfsvattr *nvap,
2773     nfsattrbit_t *attrbitp, NFSACL_T *aclp, struct thread *p)
2774 {
2775 	u_int32_t *tl;
2776 	struct nfsv2_sattr *sp;
2777 	int error = 0, toclient = 0;
2778 
2779 	switch (nd->nd_flag & (ND_NFSV2 | ND_NFSV3 | ND_NFSV4)) {
2780 	case ND_NFSV2:
2781 		NFSM_DISSECT(sp, struct nfsv2_sattr *, NFSX_V2SATTR);
2782 		/*
2783 		 * Some old clients didn't fill in the high order 16bits.
2784 		 * --> check the low order 2 bytes for 0xffff
2785 		 */
2786 		if ((fxdr_unsigned(int, sp->sa_mode) & 0xffff) != 0xffff)
2787 			nvap->na_mode = nfstov_mode(sp->sa_mode);
2788 		if (sp->sa_uid != newnfs_xdrneg1)
2789 			nvap->na_uid = fxdr_unsigned(uid_t, sp->sa_uid);
2790 		if (sp->sa_gid != newnfs_xdrneg1)
2791 			nvap->na_gid = fxdr_unsigned(gid_t, sp->sa_gid);
2792 		if (sp->sa_size != newnfs_xdrneg1)
2793 			nvap->na_size = fxdr_unsigned(u_quad_t, sp->sa_size);
2794 		if (sp->sa_atime.nfsv2_sec != newnfs_xdrneg1) {
2795 #ifdef notyet
2796 			fxdr_nfsv2time(&sp->sa_atime, &nvap->na_atime);
2797 #else
2798 			nvap->na_atime.tv_sec =
2799 				fxdr_unsigned(u_int32_t,sp->sa_atime.nfsv2_sec);
2800 			nvap->na_atime.tv_nsec = 0;
2801 #endif
2802 		}
2803 		if (sp->sa_mtime.nfsv2_sec != newnfs_xdrneg1)
2804 			fxdr_nfsv2time(&sp->sa_mtime, &nvap->na_mtime);
2805 		break;
2806 	case ND_NFSV3:
2807 		NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2808 		if (*tl == newnfs_true) {
2809 			NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2810 			nvap->na_mode = nfstov_mode(*tl);
2811 		}
2812 		NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2813 		if (*tl == newnfs_true) {
2814 			NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2815 			nvap->na_uid = fxdr_unsigned(uid_t, *tl);
2816 		}
2817 		NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2818 		if (*tl == newnfs_true) {
2819 			NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2820 			nvap->na_gid = fxdr_unsigned(gid_t, *tl);
2821 		}
2822 		NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2823 		if (*tl == newnfs_true) {
2824 			NFSM_DISSECT(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
2825 			nvap->na_size = fxdr_hyper(tl);
2826 		}
2827 		NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2828 		switch (fxdr_unsigned(int, *tl)) {
2829 		case NFSV3SATTRTIME_TOCLIENT:
2830 			NFSM_DISSECT(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
2831 			fxdr_nfsv3time(tl, &nvap->na_atime);
2832 			toclient = 1;
2833 			break;
2834 		case NFSV3SATTRTIME_TOSERVER:
2835 			vfs_timestamp(&nvap->na_atime);
2836 			nvap->na_vaflags |= VA_UTIMES_NULL;
2837 			break;
2838 		}
2839 		NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2840 		switch (fxdr_unsigned(int, *tl)) {
2841 		case NFSV3SATTRTIME_TOCLIENT:
2842 			NFSM_DISSECT(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
2843 			fxdr_nfsv3time(tl, &nvap->na_mtime);
2844 			nvap->na_vaflags &= ~VA_UTIMES_NULL;
2845 			break;
2846 		case NFSV3SATTRTIME_TOSERVER:
2847 			vfs_timestamp(&nvap->na_mtime);
2848 			if (!toclient)
2849 				nvap->na_vaflags |= VA_UTIMES_NULL;
2850 			break;
2851 		}
2852 		break;
2853 	case ND_NFSV4:
2854 		error = nfsv4_sattr(nd, vp, nvap, attrbitp, aclp, p);
2855 	}
2856 nfsmout:
2857 	NFSEXITCODE2(error, nd);
2858 	return (error);
2859 }
2860 
2861 /*
2862  * Handle the setable attributes for V4.
2863  * Returns NFSERR_BADXDR if it can't be parsed, 0 otherwise.
2864  */
2865 int
2866 nfsv4_sattr(struct nfsrv_descript *nd, vnode_t vp, struct nfsvattr *nvap,
2867     nfsattrbit_t *attrbitp, NFSACL_T *aclp, struct thread *p)
2868 {
2869 	u_int32_t *tl;
2870 	int attrsum = 0;
2871 	int i, j;
2872 	int error, attrsize, bitpos, aclsize, aceerr, retnotsup = 0;
2873 	int moderet, toclient = 0;
2874 	u_char *cp, namestr[NFSV4_SMALLSTR + 1];
2875 	uid_t uid;
2876 	gid_t gid;
2877 	u_short mode, mask;		/* Same type as va_mode. */
2878 	struct vattr va;
2879 
2880 	error = nfsrv_getattrbits(nd, attrbitp, NULL, &retnotsup);
2881 	if (error)
2882 		goto nfsmout;
2883 	NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2884 	attrsize = fxdr_unsigned(int, *tl);
2885 
2886 	/*
2887 	 * Loop around getting the setable attributes. If an unsupported
2888 	 * one is found, set nd_repstat == NFSERR_ATTRNOTSUPP and return.
2889 	 */
2890 	if (retnotsup) {
2891 		nd->nd_repstat = NFSERR_ATTRNOTSUPP;
2892 		bitpos = NFSATTRBIT_MAX;
2893 	} else {
2894 		bitpos = 0;
2895 	}
2896 	moderet = 0;
2897 	for (; bitpos < NFSATTRBIT_MAX; bitpos++) {
2898 	    if (attrsum > attrsize) {
2899 		error = NFSERR_BADXDR;
2900 		goto nfsmout;
2901 	    }
2902 	    if (NFSISSET_ATTRBIT(attrbitp, bitpos))
2903 		switch (bitpos) {
2904 		case NFSATTRBIT_SIZE:
2905 			NFSM_DISSECT(tl, u_int32_t *, NFSX_HYPER);
2906                      if (vp != NULL && vp->v_type != VREG) {
2907                             error = (vp->v_type == VDIR) ? NFSERR_ISDIR :
2908                                 NFSERR_INVAL;
2909                             goto nfsmout;
2910 			}
2911 			nvap->na_size = fxdr_hyper(tl);
2912 			attrsum += NFSX_HYPER;
2913 			break;
2914 		case NFSATTRBIT_ACL:
2915 			error = nfsrv_dissectacl(nd, aclp, &aceerr, &aclsize,
2916 			    p);
2917 			if (error)
2918 				goto nfsmout;
2919 			if (aceerr && !nd->nd_repstat)
2920 				nd->nd_repstat = aceerr;
2921 			attrsum += aclsize;
2922 			break;
2923 		case NFSATTRBIT_ARCHIVE:
2924 			NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2925 			if (!nd->nd_repstat)
2926 				nd->nd_repstat = NFSERR_ATTRNOTSUPP;
2927 			attrsum += NFSX_UNSIGNED;
2928 			break;
2929 		case NFSATTRBIT_HIDDEN:
2930 			NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2931 			if (!nd->nd_repstat)
2932 				nd->nd_repstat = NFSERR_ATTRNOTSUPP;
2933 			attrsum += NFSX_UNSIGNED;
2934 			break;
2935 		case NFSATTRBIT_MIMETYPE:
2936 			NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2937 			i = fxdr_unsigned(int, *tl);
2938 			error = nfsm_advance(nd, NFSM_RNDUP(i), -1);
2939 			if (error)
2940 				goto nfsmout;
2941 			if (!nd->nd_repstat)
2942 				nd->nd_repstat = NFSERR_ATTRNOTSUPP;
2943 			attrsum += (NFSX_UNSIGNED + NFSM_RNDUP(i));
2944 			break;
2945 		case NFSATTRBIT_MODE:
2946 			moderet = NFSERR_INVAL;	/* Can't do MODESETMASKED. */
2947 			NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2948 			nvap->na_mode = nfstov_mode(*tl);
2949 			attrsum += NFSX_UNSIGNED;
2950 			break;
2951 		case NFSATTRBIT_OWNER:
2952 			NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2953 			j = fxdr_unsigned(int, *tl);
2954 			if (j < 0) {
2955 				error = NFSERR_BADXDR;
2956 				goto nfsmout;
2957 			}
2958 			if (j > NFSV4_SMALLSTR)
2959 				cp = malloc(j + 1, M_NFSSTRING, M_WAITOK);
2960 			else
2961 				cp = namestr;
2962 			error = nfsrv_mtostr(nd, cp, j);
2963 			if (error) {
2964 				if (j > NFSV4_SMALLSTR)
2965 					free(cp, M_NFSSTRING);
2966 				goto nfsmout;
2967 			}
2968 			if (!nd->nd_repstat) {
2969 				nd->nd_repstat = nfsv4_strtouid(nd, cp, j,
2970 				    &uid);
2971 				if (!nd->nd_repstat)
2972 					nvap->na_uid = uid;
2973 			}
2974 			if (j > NFSV4_SMALLSTR)
2975 				free(cp, M_NFSSTRING);
2976 			attrsum += (NFSX_UNSIGNED + NFSM_RNDUP(j));
2977 			break;
2978 		case NFSATTRBIT_OWNERGROUP:
2979 			NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2980 			j = fxdr_unsigned(int, *tl);
2981 			if (j < 0) {
2982 				error = NFSERR_BADXDR;
2983 				goto nfsmout;
2984 			}
2985 			if (j > NFSV4_SMALLSTR)
2986 				cp = malloc(j + 1, M_NFSSTRING, M_WAITOK);
2987 			else
2988 				cp = namestr;
2989 			error = nfsrv_mtostr(nd, cp, j);
2990 			if (error) {
2991 				if (j > NFSV4_SMALLSTR)
2992 					free(cp, M_NFSSTRING);
2993 				goto nfsmout;
2994 			}
2995 			if (!nd->nd_repstat) {
2996 				nd->nd_repstat = nfsv4_strtogid(nd, cp, j,
2997 				    &gid);
2998 				if (!nd->nd_repstat)
2999 					nvap->na_gid = gid;
3000 			}
3001 			if (j > NFSV4_SMALLSTR)
3002 				free(cp, M_NFSSTRING);
3003 			attrsum += (NFSX_UNSIGNED + NFSM_RNDUP(j));
3004 			break;
3005 		case NFSATTRBIT_SYSTEM:
3006 			NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
3007 			if (!nd->nd_repstat)
3008 				nd->nd_repstat = NFSERR_ATTRNOTSUPP;
3009 			attrsum += NFSX_UNSIGNED;
3010 			break;
3011 		case NFSATTRBIT_TIMEACCESSSET:
3012 			NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
3013 			attrsum += NFSX_UNSIGNED;
3014 			if (fxdr_unsigned(int, *tl)==NFSV4SATTRTIME_TOCLIENT) {
3015 			    NFSM_DISSECT(tl, u_int32_t *, NFSX_V4TIME);
3016 			    fxdr_nfsv4time(tl, &nvap->na_atime);
3017 			    toclient = 1;
3018 			    attrsum += NFSX_V4TIME;
3019 			} else {
3020 			    vfs_timestamp(&nvap->na_atime);
3021 			    nvap->na_vaflags |= VA_UTIMES_NULL;
3022 			}
3023 			break;
3024 		case NFSATTRBIT_TIMEBACKUP:
3025 			NFSM_DISSECT(tl, u_int32_t *, NFSX_V4TIME);
3026 			if (!nd->nd_repstat)
3027 				nd->nd_repstat = NFSERR_ATTRNOTSUPP;
3028 			attrsum += NFSX_V4TIME;
3029 			break;
3030 		case NFSATTRBIT_TIMECREATE:
3031 			NFSM_DISSECT(tl, u_int32_t *, NFSX_V4TIME);
3032 			fxdr_nfsv4time(tl, &nvap->na_btime);
3033 			attrsum += NFSX_V4TIME;
3034 			break;
3035 		case NFSATTRBIT_TIMEMODIFYSET:
3036 			NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
3037 			attrsum += NFSX_UNSIGNED;
3038 			if (fxdr_unsigned(int, *tl)==NFSV4SATTRTIME_TOCLIENT) {
3039 			    NFSM_DISSECT(tl, u_int32_t *, NFSX_V4TIME);
3040 			    fxdr_nfsv4time(tl, &nvap->na_mtime);
3041 			    nvap->na_vaflags &= ~VA_UTIMES_NULL;
3042 			    attrsum += NFSX_V4TIME;
3043 			} else {
3044 			    vfs_timestamp(&nvap->na_mtime);
3045 			    if (!toclient)
3046 				nvap->na_vaflags |= VA_UTIMES_NULL;
3047 			}
3048 			break;
3049 		case NFSATTRBIT_MODESETMASKED:
3050 			NFSM_DISSECT(tl, uint32_t *, 2 * NFSX_UNSIGNED);
3051 			mode = fxdr_unsigned(u_short, *tl++);
3052 			mask = fxdr_unsigned(u_short, *tl);
3053 			/*
3054 			 * vp == NULL implies an Open/Create operation.
3055 			 * This attribute can only be used for Setattr and
3056 			 * only for NFSv4.1 or higher.
3057 			 * If moderet != 0, a mode attribute has also been
3058 			 * specified and this attribute cannot be done in the
3059 			 * same Setattr operation.
3060 			 */
3061 			if ((nd->nd_flag & ND_NFSV41) == 0)
3062 				nd->nd_repstat = NFSERR_ATTRNOTSUPP;
3063 			else if ((mode & ~07777) != 0 || (mask & ~07777) != 0 ||
3064 			    vp == NULL)
3065 				nd->nd_repstat = NFSERR_INVAL;
3066 			else if (moderet == 0)
3067 				moderet = VOP_GETATTR(vp, &va, nd->nd_cred);
3068 			if (moderet == 0)
3069 				nvap->na_mode = (mode & mask) |
3070 				    (va.va_mode & ~mask);
3071 			else
3072 				nd->nd_repstat = moderet;
3073 			attrsum += 2 * NFSX_UNSIGNED;
3074 			break;
3075 		default:
3076 			nd->nd_repstat = NFSERR_ATTRNOTSUPP;
3077 			/*
3078 			 * set bitpos so we drop out of the loop.
3079 			 */
3080 			bitpos = NFSATTRBIT_MAX;
3081 			break;
3082 		}
3083 	}
3084 
3085 	/*
3086 	 * some clients pad the attrlist, so we need to skip over the
3087 	 * padding.
3088 	 */
3089 	if (attrsum > attrsize) {
3090 		error = NFSERR_BADXDR;
3091 	} else {
3092 		attrsize = NFSM_RNDUP(attrsize);
3093 		if (attrsum < attrsize)
3094 			error = nfsm_advance(nd, attrsize - attrsum, -1);
3095 	}
3096 nfsmout:
3097 	NFSEXITCODE2(error, nd);
3098 	return (error);
3099 }
3100 
3101 /*
3102  * Check/setup export credentials.
3103  */
3104 int
3105 nfsd_excred(struct nfsrv_descript *nd, struct nfsexstuff *exp,
3106     struct ucred *credanon, bool testsec)
3107 {
3108 	int error;
3109 
3110 	/*
3111 	 * Check/setup credentials.
3112 	 */
3113 	if (nd->nd_flag & ND_GSS)
3114 		exp->nes_exflag &= ~MNT_EXPORTANON;
3115 
3116 	/*
3117 	 * Check to see if the operation is allowed for this security flavor.
3118 	 */
3119 	error = 0;
3120 	if (testsec) {
3121 		error = nfsvno_testexp(nd, exp);
3122 		if (error != 0)
3123 			goto out;
3124 	}
3125 
3126 	/*
3127 	 * Check to see if the file system is exported V4 only.
3128 	 */
3129 	if (NFSVNO_EXV4ONLY(exp) && !(nd->nd_flag & ND_NFSV4)) {
3130 		error = NFSERR_PROGNOTV4;
3131 		goto out;
3132 	}
3133 
3134 	/*
3135 	 * Now, map the user credentials.
3136 	 * (Note that ND_AUTHNONE will only be set for an NFSv3
3137 	 *  Fsinfo RPC. If set for anything else, this code might need
3138 	 *  to change.)
3139 	 */
3140 	if (NFSVNO_EXPORTED(exp)) {
3141 		if (((nd->nd_flag & ND_GSS) == 0 && nd->nd_cred->cr_uid == 0) ||
3142 		     NFSVNO_EXPORTANON(exp) ||
3143 		     (nd->nd_flag & ND_AUTHNONE) != 0) {
3144 			nd->nd_cred->cr_uid = credanon->cr_uid;
3145 			nd->nd_cred->cr_gid = credanon->cr_gid;
3146 			crsetgroups(nd->nd_cred, credanon->cr_ngroups,
3147 			    credanon->cr_groups);
3148 		} else if ((nd->nd_flag & ND_GSS) == 0) {
3149 			/*
3150 			 * If using AUTH_SYS, call nfsrv_getgrpscred() to see
3151 			 * if there is a replacement credential with a group
3152 			 * list set up by "nfsuserd -manage-gids".
3153 			 * If there is no replacement, nfsrv_getgrpscred()
3154 			 * simply returns its argument.
3155 			 */
3156 			nd->nd_cred = nfsrv_getgrpscred(nd->nd_cred);
3157 		}
3158 	}
3159 
3160 out:
3161 	NFSEXITCODE2(error, nd);
3162 	return (error);
3163 }
3164 
3165 /*
3166  * Check exports.
3167  */
3168 int
3169 nfsvno_checkexp(struct mount *mp, struct sockaddr *nam, struct nfsexstuff *exp,
3170     struct ucred **credp)
3171 {
3172 	int error;
3173 
3174 	error = VFS_CHECKEXP(mp, nam, &exp->nes_exflag, credp,
3175 	    &exp->nes_numsecflavor, exp->nes_secflavors);
3176 	if (error) {
3177 		if (nfs_rootfhset) {
3178 			exp->nes_exflag = 0;
3179 			exp->nes_numsecflavor = 0;
3180 			error = 0;
3181 		}
3182 	} else if (exp->nes_numsecflavor < 1 || exp->nes_numsecflavor >
3183 	    MAXSECFLAVORS) {
3184 		printf("nfsvno_checkexp: numsecflavors out of range\n");
3185 		exp->nes_numsecflavor = 0;
3186 		error = EACCES;
3187 	}
3188 	NFSEXITCODE(error);
3189 	return (error);
3190 }
3191 
3192 /*
3193  * Get a vnode for a file handle and export stuff.
3194  */
3195 int
3196 nfsvno_fhtovp(struct mount *mp, fhandle_t *fhp, struct sockaddr *nam,
3197     int lktype, struct vnode **vpp, struct nfsexstuff *exp,
3198     struct ucred **credp)
3199 {
3200 	int error;
3201 
3202 	*credp = NULL;
3203 	exp->nes_numsecflavor = 0;
3204 	error = VFS_FHTOVP(mp, &fhp->fh_fid, lktype, vpp);
3205 	if (error != 0)
3206 		/* Make sure the server replies ESTALE to the client. */
3207 		error = ESTALE;
3208 	if (nam && !error) {
3209 		error = VFS_CHECKEXP(mp, nam, &exp->nes_exflag, credp,
3210 		    &exp->nes_numsecflavor, exp->nes_secflavors);
3211 		if (error) {
3212 			if (nfs_rootfhset) {
3213 				exp->nes_exflag = 0;
3214 				exp->nes_numsecflavor = 0;
3215 				error = 0;
3216 			} else {
3217 				vput(*vpp);
3218 			}
3219 		} else if (exp->nes_numsecflavor < 1 || exp->nes_numsecflavor >
3220 		    MAXSECFLAVORS) {
3221 			printf("nfsvno_fhtovp: numsecflavors out of range\n");
3222 			exp->nes_numsecflavor = 0;
3223 			error = EACCES;
3224 			vput(*vpp);
3225 		}
3226 	}
3227 	NFSEXITCODE(error);
3228 	return (error);
3229 }
3230 
3231 /*
3232  * nfsd_fhtovp() - convert a fh to a vnode ptr
3233  * 	- look up fsid in mount list (if not found ret error)
3234  *	- get vp and export rights by calling nfsvno_fhtovp()
3235  *	- if cred->cr_uid == 0 or MNT_EXPORTANON set it to credanon
3236  *	  for AUTH_SYS
3237  *	- if mpp != NULL, return the mount point so that it can
3238  *	  be used for vn_finished_write() by the caller
3239  */
3240 void
3241 nfsd_fhtovp(struct nfsrv_descript *nd, struct nfsrvfh *nfp, int lktype,
3242     struct vnode **vpp, struct nfsexstuff *exp,
3243     struct mount **mpp, int startwrite, int nextop)
3244 {
3245 	struct mount *mp, *mpw;
3246 	struct ucred *credanon;
3247 	fhandle_t *fhp;
3248 	int error;
3249 
3250 	if (mpp != NULL)
3251 		*mpp = NULL;
3252 	*vpp = NULL;
3253 	fhp = (fhandle_t *)nfp->nfsrvfh_data;
3254 	mp = vfs_busyfs(&fhp->fh_fsid);
3255 	if (mp == NULL) {
3256 		nd->nd_repstat = ESTALE;
3257 		goto out;
3258 	}
3259 
3260 	if (startwrite) {
3261 		mpw = mp;
3262 		error = vn_start_write(NULL, &mpw, V_WAIT);
3263 		if (error != 0) {
3264 			mpw = NULL;
3265 			vfs_unbusy(mp);
3266 			nd->nd_repstat = ESTALE;
3267 			goto out;
3268 		}
3269 		if (lktype == LK_SHARED && !(MNT_SHARED_WRITES(mp)))
3270 			lktype = LK_EXCLUSIVE;
3271 	} else
3272 		mpw = NULL;
3273 
3274 	nd->nd_repstat = nfsvno_fhtovp(mp, fhp, nd->nd_nam, lktype, vpp, exp,
3275 	    &credanon);
3276 	vfs_unbusy(mp);
3277 
3278 	/*
3279 	 * For NFSv4 without a pseudo root fs, unexported file handles
3280 	 * can be returned, so that Lookup works everywhere.
3281 	 */
3282 	if (!nd->nd_repstat && exp->nes_exflag == 0 &&
3283 	    !(nd->nd_flag & ND_NFSV4)) {
3284 		vput(*vpp);
3285 		*vpp = NULL;
3286 		nd->nd_repstat = EACCES;
3287 	}
3288 
3289 	/*
3290 	 * Personally, I've never seen any point in requiring a
3291 	 * reserved port#, since only in the rare case where the
3292 	 * clients are all boxes with secure system privileges,
3293 	 * does it provide any enhanced security, but... some people
3294 	 * believe it to be useful and keep putting this code back in.
3295 	 * (There is also some "security checker" out there that
3296 	 *  complains if the nfs server doesn't enforce this.)
3297 	 * However, note the following:
3298 	 * RFC3530 (NFSv4) specifies that a reserved port# not be
3299 	 *	required.
3300 	 * RFC2623 recommends that, if a reserved port# is checked for,
3301 	 *	that there be a way to turn that off--> ifdef'd.
3302 	 */
3303 #ifdef NFS_REQRSVPORT
3304 	if (!nd->nd_repstat) {
3305 		struct sockaddr_in *saddr;
3306 		struct sockaddr_in6 *saddr6;
3307 
3308 		saddr = NFSSOCKADDR(nd->nd_nam, struct sockaddr_in *);
3309 		saddr6 = NFSSOCKADDR(nd->nd_nam, struct sockaddr_in6 *);
3310 		if (!(nd->nd_flag & ND_NFSV4) &&
3311 		    ((saddr->sin_family == AF_INET &&
3312 		      ntohs(saddr->sin_port) >= IPPORT_RESERVED) ||
3313 		     (saddr6->sin6_family == AF_INET6 &&
3314 		      ntohs(saddr6->sin6_port) >= IPPORT_RESERVED))) {
3315 			vput(*vpp);
3316 			nd->nd_repstat = (NFSERR_AUTHERR | AUTH_TOOWEAK);
3317 		}
3318 	}
3319 #endif	/* NFS_REQRSVPORT */
3320 
3321 	/*
3322 	 * Check/setup credentials.
3323 	 */
3324 	if (!nd->nd_repstat) {
3325 		nd->nd_saveduid = nd->nd_cred->cr_uid;
3326 		nd->nd_repstat = nfsd_excred(nd, exp, credanon,
3327 		    nfsrv_checkwrongsec(nd, nextop, (*vpp)->v_type));
3328 		if (nd->nd_repstat)
3329 			vput(*vpp);
3330 	}
3331 	if (credanon != NULL)
3332 		crfree(credanon);
3333 	if (nd->nd_repstat) {
3334 		vn_finished_write(mpw);
3335 		*vpp = NULL;
3336 	} else if (mpp != NULL) {
3337 		*mpp = mpw;
3338 	}
3339 
3340 out:
3341 	NFSEXITCODE2(0, nd);
3342 }
3343 
3344 /*
3345  * glue for fp.
3346  */
3347 static int
3348 fp_getfvp(struct thread *p, int fd, struct file **fpp, struct vnode **vpp)
3349 {
3350 	struct filedesc *fdp;
3351 	struct file *fp;
3352 	int error = 0;
3353 
3354 	fdp = p->td_proc->p_fd;
3355 	if (fd < 0 || fd >= fdp->fd_nfiles ||
3356 	    (fp = fdp->fd_ofiles[fd].fde_file) == NULL) {
3357 		error = EBADF;
3358 		goto out;
3359 	}
3360 	*fpp = fp;
3361 
3362 out:
3363 	NFSEXITCODE(error);
3364 	return (error);
3365 }
3366 
3367 /*
3368  * Called from nfssvc() to update the exports list. Just call
3369  * vfs_export(). This has to be done, since the v4 root fake fs isn't
3370  * in the mount list.
3371  */
3372 int
3373 nfsrv_v4rootexport(void *argp, struct ucred *cred, struct thread *p)
3374 {
3375 	struct nfsex_args *nfsexargp = (struct nfsex_args *)argp;
3376 	int error = 0;
3377 	struct nameidata nd;
3378 	fhandle_t fh;
3379 
3380 	error = vfs_export(&nfsv4root_mnt, &nfsexargp->export);
3381 	if ((nfsexargp->export.ex_flags & MNT_DELEXPORT) != 0)
3382 		nfs_rootfhset = 0;
3383 	else if (error == 0) {
3384 		if (nfsexargp->fspec == NULL) {
3385 			error = EPERM;
3386 			goto out;
3387 		}
3388 		/*
3389 		 * If fspec != NULL, this is the v4root path.
3390 		 */
3391 		NDINIT(&nd, LOOKUP, FOLLOW, UIO_USERSPACE,
3392 		    nfsexargp->fspec, p);
3393 		if ((error = namei(&nd)) != 0)
3394 			goto out;
3395 		error = nfsvno_getfh(nd.ni_vp, &fh, p);
3396 		vrele(nd.ni_vp);
3397 		if (!error) {
3398 			nfs_rootfh.nfsrvfh_len = NFSX_MYFH;
3399 			NFSBCOPY((caddr_t)&fh,
3400 			    nfs_rootfh.nfsrvfh_data,
3401 			    sizeof (fhandle_t));
3402 			nfs_rootfhset = 1;
3403 		}
3404 	}
3405 
3406 out:
3407 	NFSEXITCODE(error);
3408 	return (error);
3409 }
3410 
3411 /*
3412  * This function needs to test to see if the system is near its limit
3413  * for memory allocation via malloc() or mget() and return True iff
3414  * either of these resources are near their limit.
3415  * XXX (For now, this is just a stub.)
3416  */
3417 int nfsrv_testmalloclimit = 0;
3418 int
3419 nfsrv_mallocmget_limit(void)
3420 {
3421 	static int printmesg = 0;
3422 	static int testval = 1;
3423 
3424 	if (nfsrv_testmalloclimit && (testval++ % 1000) == 0) {
3425 		if ((printmesg++ % 100) == 0)
3426 			printf("nfsd: malloc/mget near limit\n");
3427 		return (1);
3428 	}
3429 	return (0);
3430 }
3431 
3432 /*
3433  * BSD specific initialization of a mount point.
3434  */
3435 void
3436 nfsd_mntinit(void)
3437 {
3438 	static int inited = 0;
3439 
3440 	if (inited)
3441 		return;
3442 	inited = 1;
3443 	nfsv4root_mnt.mnt_flag = (MNT_RDONLY | MNT_EXPORTED);
3444 	TAILQ_INIT(&nfsv4root_mnt.mnt_nvnodelist);
3445 	TAILQ_INIT(&nfsv4root_mnt.mnt_lazyvnodelist);
3446 	nfsv4root_mnt.mnt_export = NULL;
3447 	TAILQ_INIT(&nfsv4root_opt);
3448 	TAILQ_INIT(&nfsv4root_newopt);
3449 	nfsv4root_mnt.mnt_opt = &nfsv4root_opt;
3450 	nfsv4root_mnt.mnt_optnew = &nfsv4root_newopt;
3451 	nfsv4root_mnt.mnt_nvnodelistsize = 0;
3452 	nfsv4root_mnt.mnt_lazyvnodelistsize = 0;
3453 }
3454 
3455 /*
3456  * Get a vnode for a file handle, without checking exports, etc.
3457  */
3458 struct vnode *
3459 nfsvno_getvp(fhandle_t *fhp)
3460 {
3461 	struct mount *mp;
3462 	struct vnode *vp;
3463 	int error;
3464 
3465 	mp = vfs_busyfs(&fhp->fh_fsid);
3466 	if (mp == NULL)
3467 		return (NULL);
3468 	error = VFS_FHTOVP(mp, &fhp->fh_fid, LK_EXCLUSIVE, &vp);
3469 	vfs_unbusy(mp);
3470 	if (error)
3471 		return (NULL);
3472 	return (vp);
3473 }
3474 
3475 /*
3476  * Do a local VOP_ADVLOCK().
3477  */
3478 int
3479 nfsvno_advlock(struct vnode *vp, int ftype, u_int64_t first,
3480     u_int64_t end, struct thread *td)
3481 {
3482 	int error = 0;
3483 	struct flock fl;
3484 	u_int64_t tlen;
3485 
3486 	if (nfsrv_dolocallocks == 0)
3487 		goto out;
3488 	ASSERT_VOP_UNLOCKED(vp, "nfsvno_advlock: vp locked");
3489 
3490 	fl.l_whence = SEEK_SET;
3491 	fl.l_type = ftype;
3492 	fl.l_start = (off_t)first;
3493 	if (end == NFS64BITSSET) {
3494 		fl.l_len = 0;
3495 	} else {
3496 		tlen = end - first;
3497 		fl.l_len = (off_t)tlen;
3498 	}
3499 	/*
3500 	 * For FreeBSD8, the l_pid and l_sysid must be set to the same
3501 	 * values for all calls, so that all locks will be held by the
3502 	 * nfsd server. (The nfsd server handles conflicts between the
3503 	 * various clients.)
3504 	 * Since an NFSv4 lockowner is a ClientID plus an array of up to 1024
3505 	 * bytes, so it can't be put in l_sysid.
3506 	 */
3507 	if (nfsv4_sysid == 0)
3508 		nfsv4_sysid = nlm_acquire_next_sysid();
3509 	fl.l_pid = (pid_t)0;
3510 	fl.l_sysid = (int)nfsv4_sysid;
3511 
3512 	if (ftype == F_UNLCK)
3513 		error = VOP_ADVLOCK(vp, (caddr_t)td->td_proc, F_UNLCK, &fl,
3514 		    (F_POSIX | F_REMOTE));
3515 	else
3516 		error = VOP_ADVLOCK(vp, (caddr_t)td->td_proc, F_SETLK, &fl,
3517 		    (F_POSIX | F_REMOTE));
3518 
3519 out:
3520 	NFSEXITCODE(error);
3521 	return (error);
3522 }
3523 
3524 /*
3525  * Check the nfsv4 root exports.
3526  */
3527 int
3528 nfsvno_v4rootexport(struct nfsrv_descript *nd)
3529 {
3530 	struct ucred *credanon;
3531 	int error = 0, numsecflavor, secflavors[MAXSECFLAVORS], i;
3532 	uint64_t exflags;
3533 
3534 	error = vfs_stdcheckexp(&nfsv4root_mnt, nd->nd_nam, &exflags,
3535 	    &credanon, &numsecflavor, secflavors);
3536 	if (error) {
3537 		error = NFSERR_PROGUNAVAIL;
3538 		goto out;
3539 	}
3540 	if (credanon != NULL)
3541 		crfree(credanon);
3542 	for (i = 0; i < numsecflavor; i++) {
3543 		if (secflavors[i] == AUTH_SYS)
3544 			nd->nd_flag |= ND_EXAUTHSYS;
3545 		else if (secflavors[i] == RPCSEC_GSS_KRB5)
3546 			nd->nd_flag |= ND_EXGSS;
3547 		else if (secflavors[i] == RPCSEC_GSS_KRB5I)
3548 			nd->nd_flag |= ND_EXGSSINTEGRITY;
3549 		else if (secflavors[i] == RPCSEC_GSS_KRB5P)
3550 			nd->nd_flag |= ND_EXGSSPRIVACY;
3551 	}
3552 
3553 	/* And set ND_EXxx flags for TLS. */
3554 	if ((exflags & MNT_EXTLS) != 0) {
3555 		nd->nd_flag |= ND_EXTLS;
3556 		if ((exflags & MNT_EXTLSCERT) != 0)
3557 			nd->nd_flag |= ND_EXTLSCERT;
3558 		if ((exflags & MNT_EXTLSCERTUSER) != 0)
3559 			nd->nd_flag |= ND_EXTLSCERTUSER;
3560 	}
3561 
3562 out:
3563 	NFSEXITCODE(error);
3564 	return (error);
3565 }
3566 
3567 /*
3568  * Nfs server pseudo system call for the nfsd's
3569  */
3570 /*
3571  * MPSAFE
3572  */
3573 static int
3574 nfssvc_nfsd(struct thread *td, struct nfssvc_args *uap)
3575 {
3576 	struct file *fp;
3577 	struct nfsd_addsock_args sockarg;
3578 	struct nfsd_nfsd_args nfsdarg;
3579 	struct nfsd_nfsd_oargs onfsdarg;
3580 	struct nfsd_pnfsd_args pnfsdarg;
3581 	struct vnode *vp, *nvp, *curdvp;
3582 	struct pnfsdsfile *pf;
3583 	struct nfsdevice *ds, *fds;
3584 	cap_rights_t rights;
3585 	int buflen, error, ret;
3586 	char *buf, *cp, *cp2, *cp3;
3587 	char fname[PNFS_FILENAME_LEN + 1];
3588 
3589 	if (uap->flag & NFSSVC_NFSDADDSOCK) {
3590 		error = copyin(uap->argp, (caddr_t)&sockarg, sizeof (sockarg));
3591 		if (error)
3592 			goto out;
3593 		/*
3594 		 * Since we don't know what rights might be required,
3595 		 * pretend that we need them all. It is better to be too
3596 		 * careful than too reckless.
3597 		 */
3598 		error = fget(td, sockarg.sock,
3599 		    cap_rights_init_one(&rights, CAP_SOCK_SERVER), &fp);
3600 		if (error != 0)
3601 			goto out;
3602 		if (fp->f_type != DTYPE_SOCKET) {
3603 			fdrop(fp, td);
3604 			error = EPERM;
3605 			goto out;
3606 		}
3607 		error = nfsrvd_addsock(fp);
3608 		fdrop(fp, td);
3609 	} else if (uap->flag & NFSSVC_NFSDNFSD) {
3610 		if (uap->argp == NULL) {
3611 			error = EINVAL;
3612 			goto out;
3613 		}
3614 		if ((uap->flag & NFSSVC_NEWSTRUCT) == 0) {
3615 			error = copyin(uap->argp, &onfsdarg, sizeof(onfsdarg));
3616 			if (error == 0) {
3617 				nfsdarg.principal = onfsdarg.principal;
3618 				nfsdarg.minthreads = onfsdarg.minthreads;
3619 				nfsdarg.maxthreads = onfsdarg.maxthreads;
3620 				nfsdarg.version = 1;
3621 				nfsdarg.addr = NULL;
3622 				nfsdarg.addrlen = 0;
3623 				nfsdarg.dnshost = NULL;
3624 				nfsdarg.dnshostlen = 0;
3625 				nfsdarg.dspath = NULL;
3626 				nfsdarg.dspathlen = 0;
3627 				nfsdarg.mdspath = NULL;
3628 				nfsdarg.mdspathlen = 0;
3629 				nfsdarg.mirrorcnt = 1;
3630 			}
3631 		} else
3632 			error = copyin(uap->argp, &nfsdarg, sizeof(nfsdarg));
3633 		if (error)
3634 			goto out;
3635 		if (nfsdarg.addrlen > 0 && nfsdarg.addrlen < 10000 &&
3636 		    nfsdarg.dnshostlen > 0 && nfsdarg.dnshostlen < 10000 &&
3637 		    nfsdarg.dspathlen > 0 && nfsdarg.dspathlen < 10000 &&
3638 		    nfsdarg.mdspathlen > 0 && nfsdarg.mdspathlen < 10000 &&
3639 		    nfsdarg.mirrorcnt >= 1 &&
3640 		    nfsdarg.mirrorcnt <= NFSDEV_MAXMIRRORS &&
3641 		    nfsdarg.addr != NULL && nfsdarg.dnshost != NULL &&
3642 		    nfsdarg.dspath != NULL && nfsdarg.mdspath != NULL) {
3643 			NFSD_DEBUG(1, "addrlen=%d dspathlen=%d dnslen=%d"
3644 			    " mdspathlen=%d mirrorcnt=%d\n", nfsdarg.addrlen,
3645 			    nfsdarg.dspathlen, nfsdarg.dnshostlen,
3646 			    nfsdarg.mdspathlen, nfsdarg.mirrorcnt);
3647 			cp = malloc(nfsdarg.addrlen + 1, M_TEMP, M_WAITOK);
3648 			error = copyin(nfsdarg.addr, cp, nfsdarg.addrlen);
3649 			if (error != 0) {
3650 				free(cp, M_TEMP);
3651 				goto out;
3652 			}
3653 			cp[nfsdarg.addrlen] = '\0';	/* Ensure nul term. */
3654 			nfsdarg.addr = cp;
3655 			cp = malloc(nfsdarg.dnshostlen + 1, M_TEMP, M_WAITOK);
3656 			error = copyin(nfsdarg.dnshost, cp, nfsdarg.dnshostlen);
3657 			if (error != 0) {
3658 				free(nfsdarg.addr, M_TEMP);
3659 				free(cp, M_TEMP);
3660 				goto out;
3661 			}
3662 			cp[nfsdarg.dnshostlen] = '\0';	/* Ensure nul term. */
3663 			nfsdarg.dnshost = cp;
3664 			cp = malloc(nfsdarg.dspathlen + 1, M_TEMP, M_WAITOK);
3665 			error = copyin(nfsdarg.dspath, cp, nfsdarg.dspathlen);
3666 			if (error != 0) {
3667 				free(nfsdarg.addr, M_TEMP);
3668 				free(nfsdarg.dnshost, M_TEMP);
3669 				free(cp, M_TEMP);
3670 				goto out;
3671 			}
3672 			cp[nfsdarg.dspathlen] = '\0';	/* Ensure nul term. */
3673 			nfsdarg.dspath = cp;
3674 			cp = malloc(nfsdarg.mdspathlen + 1, M_TEMP, M_WAITOK);
3675 			error = copyin(nfsdarg.mdspath, cp, nfsdarg.mdspathlen);
3676 			if (error != 0) {
3677 				free(nfsdarg.addr, M_TEMP);
3678 				free(nfsdarg.dnshost, M_TEMP);
3679 				free(nfsdarg.dspath, M_TEMP);
3680 				free(cp, M_TEMP);
3681 				goto out;
3682 			}
3683 			cp[nfsdarg.mdspathlen] = '\0';	/* Ensure nul term. */
3684 			nfsdarg.mdspath = cp;
3685 		} else {
3686 			nfsdarg.addr = NULL;
3687 			nfsdarg.addrlen = 0;
3688 			nfsdarg.dnshost = NULL;
3689 			nfsdarg.dnshostlen = 0;
3690 			nfsdarg.dspath = NULL;
3691 			nfsdarg.dspathlen = 0;
3692 			nfsdarg.mdspath = NULL;
3693 			nfsdarg.mdspathlen = 0;
3694 			nfsdarg.mirrorcnt = 1;
3695 		}
3696 		error = nfsrvd_nfsd(td, &nfsdarg);
3697 		free(nfsdarg.addr, M_TEMP);
3698 		free(nfsdarg.dnshost, M_TEMP);
3699 		free(nfsdarg.dspath, M_TEMP);
3700 		free(nfsdarg.mdspath, M_TEMP);
3701 	} else if (uap->flag & NFSSVC_PNFSDS) {
3702 		error = copyin(uap->argp, &pnfsdarg, sizeof(pnfsdarg));
3703 		if (error == 0 && (pnfsdarg.op == PNFSDOP_DELDSSERVER ||
3704 		    pnfsdarg.op == PNFSDOP_FORCEDELDS)) {
3705 			cp = malloc(PATH_MAX + 1, M_TEMP, M_WAITOK);
3706 			error = copyinstr(pnfsdarg.dspath, cp, PATH_MAX + 1,
3707 			    NULL);
3708 			if (error == 0)
3709 				error = nfsrv_deldsserver(pnfsdarg.op, cp, td);
3710 			free(cp, M_TEMP);
3711 		} else if (error == 0 && pnfsdarg.op == PNFSDOP_COPYMR) {
3712 			cp = malloc(PATH_MAX + 1, M_TEMP, M_WAITOK);
3713 			buflen = sizeof(*pf) * NFSDEV_MAXMIRRORS;
3714 			buf = malloc(buflen, M_TEMP, M_WAITOK);
3715 			error = copyinstr(pnfsdarg.mdspath, cp, PATH_MAX + 1,
3716 			    NULL);
3717 			NFSD_DEBUG(4, "pnfsdcopymr cp mdspath=%d\n", error);
3718 			if (error == 0 && pnfsdarg.dspath != NULL) {
3719 				cp2 = malloc(PATH_MAX + 1, M_TEMP, M_WAITOK);
3720 				error = copyinstr(pnfsdarg.dspath, cp2,
3721 				    PATH_MAX + 1, NULL);
3722 				NFSD_DEBUG(4, "pnfsdcopymr cp dspath=%d\n",
3723 				    error);
3724 			} else
3725 				cp2 = NULL;
3726 			if (error == 0 && pnfsdarg.curdspath != NULL) {
3727 				cp3 = malloc(PATH_MAX + 1, M_TEMP, M_WAITOK);
3728 				error = copyinstr(pnfsdarg.curdspath, cp3,
3729 				    PATH_MAX + 1, NULL);
3730 				NFSD_DEBUG(4, "pnfsdcopymr cp curdspath=%d\n",
3731 				    error);
3732 			} else
3733 				cp3 = NULL;
3734 			curdvp = NULL;
3735 			fds = NULL;
3736 			if (error == 0)
3737 				error = nfsrv_mdscopymr(cp, cp2, cp3, buf,
3738 				    &buflen, fname, td, &vp, &nvp, &pf, &ds,
3739 				    &fds);
3740 			NFSD_DEBUG(4, "nfsrv_mdscopymr=%d\n", error);
3741 			if (error == 0) {
3742 				if (pf->dsf_dir >= nfsrv_dsdirsize) {
3743 					printf("copymr: dsdir out of range\n");
3744 					pf->dsf_dir = 0;
3745 				}
3746 				NFSD_DEBUG(4, "copymr: buflen=%d\n", buflen);
3747 				error = nfsrv_copymr(vp, nvp,
3748 				    ds->nfsdev_dsdir[pf->dsf_dir], ds, pf,
3749 				    (struct pnfsdsfile *)buf,
3750 				    buflen / sizeof(*pf), td->td_ucred, td);
3751 				vput(vp);
3752 				vput(nvp);
3753 				if (fds != NULL && error == 0) {
3754 					curdvp = fds->nfsdev_dsdir[pf->dsf_dir];
3755 					ret = vn_lock(curdvp, LK_EXCLUSIVE);
3756 					if (ret == 0) {
3757 						nfsrv_dsremove(curdvp, fname,
3758 						    td->td_ucred, td);
3759 						NFSVOPUNLOCK(curdvp);
3760 					}
3761 				}
3762 				NFSD_DEBUG(4, "nfsrv_copymr=%d\n", error);
3763 			}
3764 			free(cp, M_TEMP);
3765 			free(cp2, M_TEMP);
3766 			free(cp3, M_TEMP);
3767 			free(buf, M_TEMP);
3768 		}
3769 	} else {
3770 		error = nfssvc_srvcall(td, uap, td->td_ucred);
3771 	}
3772 
3773 out:
3774 	NFSEXITCODE(error);
3775 	return (error);
3776 }
3777 
3778 static int
3779 nfssvc_srvcall(struct thread *p, struct nfssvc_args *uap, struct ucred *cred)
3780 {
3781 	struct nfsex_args export;
3782 	struct nfsex_oldargs oexp;
3783 	struct file *fp = NULL;
3784 	int stablefd, i, len;
3785 	struct nfsd_clid adminrevoke;
3786 	struct nfsd_dumplist dumplist;
3787 	struct nfsd_dumpclients *dumpclients;
3788 	struct nfsd_dumplocklist dumplocklist;
3789 	struct nfsd_dumplocks *dumplocks;
3790 	struct nameidata nd;
3791 	vnode_t vp;
3792 	int error = EINVAL, igotlock;
3793 	struct proc *procp;
3794 	gid_t *grps;
3795 	static int suspend_nfsd = 0;
3796 
3797 	if (uap->flag & NFSSVC_PUBLICFH) {
3798 		NFSBZERO((caddr_t)&nfs_pubfh.nfsrvfh_data,
3799 		    sizeof (fhandle_t));
3800 		error = copyin(uap->argp,
3801 		    &nfs_pubfh.nfsrvfh_data, sizeof (fhandle_t));
3802 		if (!error)
3803 			nfs_pubfhset = 1;
3804 	} else if ((uap->flag & (NFSSVC_V4ROOTEXPORT | NFSSVC_NEWSTRUCT)) ==
3805 	    (NFSSVC_V4ROOTEXPORT | NFSSVC_NEWSTRUCT)) {
3806 		error = copyin(uap->argp,(caddr_t)&export,
3807 		    sizeof (struct nfsex_args));
3808 		if (!error) {
3809 			grps = NULL;
3810 			if (export.export.ex_ngroups > NGROUPS_MAX ||
3811 			    export.export.ex_ngroups < 0)
3812 				error = EINVAL;
3813 			else if (export.export.ex_ngroups > 0) {
3814 				grps = malloc(export.export.ex_ngroups *
3815 				    sizeof(gid_t), M_TEMP, M_WAITOK);
3816 				error = copyin(export.export.ex_groups, grps,
3817 				    export.export.ex_ngroups * sizeof(gid_t));
3818 				export.export.ex_groups = grps;
3819 			} else
3820 				export.export.ex_groups = NULL;
3821 			if (!error)
3822 				error = nfsrv_v4rootexport(&export, cred, p);
3823 			free(grps, M_TEMP);
3824 		}
3825 	} else if ((uap->flag & (NFSSVC_V4ROOTEXPORT | NFSSVC_NEWSTRUCT)) ==
3826 	    NFSSVC_V4ROOTEXPORT) {
3827 		error = copyin(uap->argp,(caddr_t)&oexp,
3828 		    sizeof (struct nfsex_oldargs));
3829 		if (!error) {
3830 			memset(&export.export, 0, sizeof(export.export));
3831 			export.export.ex_flags = (uint64_t)oexp.export.ex_flags;
3832 			export.export.ex_root = oexp.export.ex_root;
3833 			export.export.ex_uid = oexp.export.ex_anon.cr_uid;
3834 			export.export.ex_ngroups =
3835 			    oexp.export.ex_anon.cr_ngroups;
3836 			export.export.ex_groups = NULL;
3837 			if (export.export.ex_ngroups > XU_NGROUPS ||
3838 			    export.export.ex_ngroups < 0)
3839 				error = EINVAL;
3840 			else if (export.export.ex_ngroups > 0) {
3841 				export.export.ex_groups = malloc(
3842 				    export.export.ex_ngroups * sizeof(gid_t),
3843 				    M_TEMP, M_WAITOK);
3844 				for (i = 0; i < export.export.ex_ngroups; i++)
3845 					export.export.ex_groups[i] =
3846 					    oexp.export.ex_anon.cr_groups[i];
3847 			}
3848 			export.export.ex_addr = oexp.export.ex_addr;
3849 			export.export.ex_addrlen = oexp.export.ex_addrlen;
3850 			export.export.ex_mask = oexp.export.ex_mask;
3851 			export.export.ex_masklen = oexp.export.ex_masklen;
3852 			export.export.ex_indexfile = oexp.export.ex_indexfile;
3853 			export.export.ex_numsecflavors =
3854 			    oexp.export.ex_numsecflavors;
3855 			if (export.export.ex_numsecflavors >= MAXSECFLAVORS ||
3856 			    export.export.ex_numsecflavors < 0)
3857 				error = EINVAL;
3858 			else {
3859 				for (i = 0; i < export.export.ex_numsecflavors;
3860 				    i++)
3861 					export.export.ex_secflavors[i] =
3862 					    oexp.export.ex_secflavors[i];
3863 			}
3864 			export.fspec = oexp.fspec;
3865 			if (error == 0)
3866 				error = nfsrv_v4rootexport(&export, cred, p);
3867 			free(export.export.ex_groups, M_TEMP);
3868 		}
3869 	} else if (uap->flag & NFSSVC_NOPUBLICFH) {
3870 		nfs_pubfhset = 0;
3871 		error = 0;
3872 	} else if (uap->flag & NFSSVC_STABLERESTART) {
3873 		error = copyin(uap->argp, (caddr_t)&stablefd,
3874 		    sizeof (int));
3875 		if (!error)
3876 			error = fp_getfvp(p, stablefd, &fp, &vp);
3877 		if (!error && (NFSFPFLAG(fp) & (FREAD | FWRITE)) != (FREAD | FWRITE))
3878 			error = EBADF;
3879 		if (!error && newnfs_numnfsd != 0)
3880 			error = EPERM;
3881 		if (!error) {
3882 			nfsrv_stablefirst.nsf_fp = fp;
3883 			nfsrv_setupstable(p);
3884 		}
3885 	} else if (uap->flag & NFSSVC_ADMINREVOKE) {
3886 		error = copyin(uap->argp, (caddr_t)&adminrevoke,
3887 		    sizeof (struct nfsd_clid));
3888 		if (!error)
3889 			error = nfsrv_adminrevoke(&adminrevoke, p);
3890 	} else if (uap->flag & NFSSVC_DUMPCLIENTS) {
3891 		error = copyin(uap->argp, (caddr_t)&dumplist,
3892 		    sizeof (struct nfsd_dumplist));
3893 		if (!error && (dumplist.ndl_size < 1 ||
3894 			dumplist.ndl_size > NFSRV_MAXDUMPLIST))
3895 			error = EPERM;
3896 		if (!error) {
3897 		    len = sizeof (struct nfsd_dumpclients) * dumplist.ndl_size;
3898 		    dumpclients = malloc(len, M_TEMP, M_WAITOK | M_ZERO);
3899 		    nfsrv_dumpclients(dumpclients, dumplist.ndl_size);
3900 		    error = copyout(dumpclients, dumplist.ndl_list, len);
3901 		    free(dumpclients, M_TEMP);
3902 		}
3903 	} else if (uap->flag & NFSSVC_DUMPLOCKS) {
3904 		error = copyin(uap->argp, (caddr_t)&dumplocklist,
3905 		    sizeof (struct nfsd_dumplocklist));
3906 		if (!error && (dumplocklist.ndllck_size < 1 ||
3907 			dumplocklist.ndllck_size > NFSRV_MAXDUMPLIST))
3908 			error = EPERM;
3909 		if (!error)
3910 			error = nfsrv_lookupfilename(&nd,
3911 				dumplocklist.ndllck_fname, p);
3912 		if (!error) {
3913 			len = sizeof (struct nfsd_dumplocks) *
3914 				dumplocklist.ndllck_size;
3915 			dumplocks = malloc(len, M_TEMP, M_WAITOK | M_ZERO);
3916 			nfsrv_dumplocks(nd.ni_vp, dumplocks,
3917 			    dumplocklist.ndllck_size, p);
3918 			vput(nd.ni_vp);
3919 			error = copyout(dumplocks, dumplocklist.ndllck_list,
3920 			    len);
3921 			free(dumplocks, M_TEMP);
3922 		}
3923 	} else if (uap->flag & NFSSVC_BACKUPSTABLE) {
3924 		procp = p->td_proc;
3925 		PROC_LOCK(procp);
3926 		nfsd_master_pid = procp->p_pid;
3927 		bcopy(procp->p_comm, nfsd_master_comm, MAXCOMLEN + 1);
3928 		nfsd_master_start = procp->p_stats->p_start;
3929 		nfsd_master_proc = procp;
3930 		PROC_UNLOCK(procp);
3931 	} else if ((uap->flag & NFSSVC_SUSPENDNFSD) != 0) {
3932 		NFSLOCKV4ROOTMUTEX();
3933 		if (suspend_nfsd == 0) {
3934 			/* Lock out all nfsd threads */
3935 			do {
3936 				igotlock = nfsv4_lock(&nfsd_suspend_lock, 1,
3937 				    NULL, NFSV4ROOTLOCKMUTEXPTR, NULL);
3938 			} while (igotlock == 0 && suspend_nfsd == 0);
3939 			suspend_nfsd = 1;
3940 		}
3941 		NFSUNLOCKV4ROOTMUTEX();
3942 		error = 0;
3943 	} else if ((uap->flag & NFSSVC_RESUMENFSD) != 0) {
3944 		NFSLOCKV4ROOTMUTEX();
3945 		if (suspend_nfsd != 0) {
3946 			nfsv4_unlock(&nfsd_suspend_lock, 0);
3947 			suspend_nfsd = 0;
3948 		}
3949 		NFSUNLOCKV4ROOTMUTEX();
3950 		error = 0;
3951 	}
3952 
3953 	NFSEXITCODE(error);
3954 	return (error);
3955 }
3956 
3957 /*
3958  * Check exports.
3959  * Returns 0 if ok, 1 otherwise.
3960  */
3961 int
3962 nfsvno_testexp(struct nfsrv_descript *nd, struct nfsexstuff *exp)
3963 {
3964 	int i;
3965 
3966 	/*
3967 	 * Allow NFSv3 Fsinfo per RFC2623.
3968 	 */
3969 	if (((nd->nd_flag & ND_NFSV4) != 0 ||
3970 	     nd->nd_procnum != NFSPROC_FSINFO) &&
3971 	    ((NFSVNO_EXTLS(exp) && (nd->nd_flag & ND_TLS) == 0) ||
3972 	     (NFSVNO_EXTLSCERT(exp) &&
3973 	      (nd->nd_flag & ND_TLSCERT) == 0) ||
3974 	     (NFSVNO_EXTLSCERTUSER(exp) &&
3975 	      (nd->nd_flag & ND_TLSCERTUSER) == 0))) {
3976 		if ((nd->nd_flag & ND_NFSV4) != 0)
3977 			return (NFSERR_WRONGSEC);
3978 		else if ((nd->nd_flag & ND_TLS) == 0)
3979 			return (NFSERR_AUTHERR | AUTH_NEEDS_TLS);
3980 		else
3981 			return (NFSERR_AUTHERR | AUTH_NEEDS_TLS_MUTUAL_HOST);
3982 	}
3983 
3984 	/*
3985 	 * This seems odd, but allow the case where the security flavor
3986 	 * list is empty. This happens when NFSv4 is traversing non-exported
3987 	 * file systems. Exported file systems should always have a non-empty
3988 	 * security flavor list.
3989 	 */
3990 	if (exp->nes_numsecflavor == 0)
3991 		return (0);
3992 
3993 	for (i = 0; i < exp->nes_numsecflavor; i++) {
3994 		/*
3995 		 * The tests for privacy and integrity must be first,
3996 		 * since ND_GSS is set for everything but AUTH_SYS.
3997 		 */
3998 		if (exp->nes_secflavors[i] == RPCSEC_GSS_KRB5P &&
3999 		    (nd->nd_flag & ND_GSSPRIVACY))
4000 			return (0);
4001 		if (exp->nes_secflavors[i] == RPCSEC_GSS_KRB5I &&
4002 		    (nd->nd_flag & ND_GSSINTEGRITY))
4003 			return (0);
4004 		if (exp->nes_secflavors[i] == RPCSEC_GSS_KRB5 &&
4005 		    (nd->nd_flag & ND_GSS))
4006 			return (0);
4007 		if (exp->nes_secflavors[i] == AUTH_SYS &&
4008 		    (nd->nd_flag & ND_GSS) == 0)
4009 			return (0);
4010 	}
4011 	if ((nd->nd_flag & ND_NFSV4) != 0)
4012 		return (NFSERR_WRONGSEC);
4013 	return (NFSERR_AUTHERR | AUTH_TOOWEAK);
4014 }
4015 
4016 /*
4017  * Calculate a hash value for the fid in a file handle.
4018  */
4019 uint32_t
4020 nfsrv_hashfh(fhandle_t *fhp)
4021 {
4022 	uint32_t hashval;
4023 
4024 	hashval = hash32_buf(&fhp->fh_fid, sizeof(struct fid), 0);
4025 	return (hashval);
4026 }
4027 
4028 /*
4029  * Calculate a hash value for the sessionid.
4030  */
4031 uint32_t
4032 nfsrv_hashsessionid(uint8_t *sessionid)
4033 {
4034 	uint32_t hashval;
4035 
4036 	hashval = hash32_buf(sessionid, NFSX_V4SESSIONID, 0);
4037 	return (hashval);
4038 }
4039 
4040 /*
4041  * Signal the userland master nfsd to backup the stable restart file.
4042  */
4043 void
4044 nfsrv_backupstable(void)
4045 {
4046 	struct proc *procp;
4047 
4048 	if (nfsd_master_proc != NULL) {
4049 		procp = pfind(nfsd_master_pid);
4050 		/* Try to make sure it is the correct process. */
4051 		if (procp == nfsd_master_proc &&
4052 		    procp->p_stats->p_start.tv_sec ==
4053 		    nfsd_master_start.tv_sec &&
4054 		    procp->p_stats->p_start.tv_usec ==
4055 		    nfsd_master_start.tv_usec &&
4056 		    strcmp(procp->p_comm, nfsd_master_comm) == 0)
4057 			kern_psignal(procp, SIGUSR2);
4058 		else
4059 			nfsd_master_proc = NULL;
4060 
4061 		if (procp != NULL)
4062 			PROC_UNLOCK(procp);
4063 	}
4064 }
4065 
4066 /*
4067  * Create a DS data file for nfsrv_pnfscreate(). Called for each mirror.
4068  * The arguments are in a structure, so that they can be passed through
4069  * taskqueue for a kernel process to execute this function.
4070  */
4071 struct nfsrvdscreate {
4072 	int			done;
4073 	int			inprog;
4074 	struct task		tsk;
4075 	struct ucred		*tcred;
4076 	struct vnode		*dvp;
4077 	NFSPROC_T		*p;
4078 	struct pnfsdsfile	*pf;
4079 	int			err;
4080 	fhandle_t		fh;
4081 	struct vattr		va;
4082 	struct vattr		createva;
4083 };
4084 
4085 int
4086 nfsrv_dscreate(struct vnode *dvp, struct vattr *vap, struct vattr *nvap,
4087     fhandle_t *fhp, struct pnfsdsfile *pf, struct pnfsdsattr *dsa,
4088     char *fnamep, struct ucred *tcred, NFSPROC_T *p, struct vnode **nvpp)
4089 {
4090 	struct vnode *nvp;
4091 	struct nameidata named;
4092 	struct vattr va;
4093 	char *bufp;
4094 	u_long *hashp;
4095 	struct nfsnode *np;
4096 	struct nfsmount *nmp;
4097 	int error;
4098 
4099 	NFSNAMEICNDSET(&named.ni_cnd, tcred, CREATE,
4100 	    LOCKPARENT | LOCKLEAF | SAVESTART | NOCACHE);
4101 	nfsvno_setpathbuf(&named, &bufp, &hashp);
4102 	named.ni_cnd.cn_lkflags = LK_EXCLUSIVE;
4103 	named.ni_cnd.cn_thread = p;
4104 	named.ni_cnd.cn_nameptr = bufp;
4105 	if (fnamep != NULL) {
4106 		strlcpy(bufp, fnamep, PNFS_FILENAME_LEN + 1);
4107 		named.ni_cnd.cn_namelen = strlen(bufp);
4108 	} else
4109 		named.ni_cnd.cn_namelen = nfsrv_putfhname(fhp, bufp);
4110 	NFSD_DEBUG(4, "nfsrv_dscreate: dvp=%p fname=%s\n", dvp, bufp);
4111 
4112 	/* Create the date file in the DS mount. */
4113 	error = NFSVOPLOCK(dvp, LK_EXCLUSIVE);
4114 	if (error == 0) {
4115 		error = VOP_CREATE(dvp, &nvp, &named.ni_cnd, vap);
4116 		vref(dvp);
4117 		VOP_VPUT_PAIR(dvp, error == 0 ? &nvp : NULL, false);
4118 		if (error == 0) {
4119 			/* Set the ownership of the file. */
4120 			error = VOP_SETATTR(nvp, nvap, tcred);
4121 			NFSD_DEBUG(4, "nfsrv_dscreate:"
4122 			    " setattr-uid=%d\n", error);
4123 			if (error != 0)
4124 				vput(nvp);
4125 		}
4126 		if (error != 0)
4127 			printf("pNFS: pnfscreate failed=%d\n", error);
4128 	} else
4129 		printf("pNFS: pnfscreate vnlock=%d\n", error);
4130 	if (error == 0) {
4131 		np = VTONFS(nvp);
4132 		nmp = VFSTONFS(nvp->v_mount);
4133 		if (strcmp(nvp->v_mount->mnt_vfc->vfc_name, "nfs")
4134 		    != 0 || nmp->nm_nam->sa_len > sizeof(
4135 		    struct sockaddr_in6) ||
4136 		    np->n_fhp->nfh_len != NFSX_MYFH) {
4137 			printf("Bad DS file: fstype=%s salen=%d"
4138 			    " fhlen=%d\n",
4139 			    nvp->v_mount->mnt_vfc->vfc_name,
4140 			    nmp->nm_nam->sa_len, np->n_fhp->nfh_len);
4141 			error = ENOENT;
4142 		}
4143 
4144 		/* Set extattrs for the DS on the MDS file. */
4145 		if (error == 0) {
4146 			if (dsa != NULL) {
4147 				error = VOP_GETATTR(nvp, &va, tcred);
4148 				if (error == 0) {
4149 					dsa->dsa_filerev = va.va_filerev;
4150 					dsa->dsa_size = va.va_size;
4151 					dsa->dsa_atime = va.va_atime;
4152 					dsa->dsa_mtime = va.va_mtime;
4153 					dsa->dsa_bytes = va.va_bytes;
4154 				}
4155 			}
4156 			if (error == 0) {
4157 				NFSBCOPY(np->n_fhp->nfh_fh, &pf->dsf_fh,
4158 				    NFSX_MYFH);
4159 				NFSBCOPY(nmp->nm_nam, &pf->dsf_sin,
4160 				    nmp->nm_nam->sa_len);
4161 				NFSBCOPY(named.ni_cnd.cn_nameptr,
4162 				    pf->dsf_filename,
4163 				    sizeof(pf->dsf_filename));
4164 			}
4165 		} else
4166 			printf("pNFS: pnfscreate can't get DS"
4167 			    " attr=%d\n", error);
4168 		if (nvpp != NULL && error == 0)
4169 			*nvpp = nvp;
4170 		else
4171 			vput(nvp);
4172 	}
4173 	nfsvno_relpathbuf(&named);
4174 	return (error);
4175 }
4176 
4177 /*
4178  * Start up the thread that will execute nfsrv_dscreate().
4179  */
4180 static void
4181 start_dscreate(void *arg, int pending)
4182 {
4183 	struct nfsrvdscreate *dsc;
4184 
4185 	dsc = (struct nfsrvdscreate *)arg;
4186 	dsc->err = nfsrv_dscreate(dsc->dvp, &dsc->createva, &dsc->va, &dsc->fh,
4187 	    dsc->pf, NULL, NULL, dsc->tcred, dsc->p, NULL);
4188 	dsc->done = 1;
4189 	NFSD_DEBUG(4, "start_dscreate: err=%d\n", dsc->err);
4190 }
4191 
4192 /*
4193  * Create a pNFS data file on the Data Server(s).
4194  */
4195 static void
4196 nfsrv_pnfscreate(struct vnode *vp, struct vattr *vap, struct ucred *cred,
4197     NFSPROC_T *p)
4198 {
4199 	struct nfsrvdscreate *dsc, *tdsc = NULL;
4200 	struct nfsdevice *ds, *tds, *fds;
4201 	struct mount *mp;
4202 	struct pnfsdsfile *pf, *tpf;
4203 	struct pnfsdsattr dsattr;
4204 	struct vattr va;
4205 	struct vnode *dvp[NFSDEV_MAXMIRRORS];
4206 	struct nfsmount *nmp;
4207 	fhandle_t fh;
4208 	uid_t vauid;
4209 	gid_t vagid;
4210 	u_short vamode;
4211 	struct ucred *tcred;
4212 	int dsdir[NFSDEV_MAXMIRRORS], error, i, mirrorcnt, ret;
4213 	int failpos, timo;
4214 
4215 	/* Get a DS server directory in a round-robin order. */
4216 	mirrorcnt = 1;
4217 	mp = vp->v_mount;
4218 	ds = fds = NULL;
4219 	NFSDDSLOCK();
4220 	/*
4221 	 * Search for the first entry that handles this MDS fs, but use the
4222 	 * first entry for all MDS fs's otherwise.
4223 	 */
4224 	TAILQ_FOREACH(tds, &nfsrv_devidhead, nfsdev_list) {
4225 		if (tds->nfsdev_nmp != NULL) {
4226 			if (tds->nfsdev_mdsisset == 0 && ds == NULL)
4227 				ds = tds;
4228 			else if (tds->nfsdev_mdsisset != 0 && fsidcmp(
4229 			    &mp->mnt_stat.f_fsid, &tds->nfsdev_mdsfsid) == 0) {
4230 				ds = fds = tds;
4231 				break;
4232 			}
4233 		}
4234 	}
4235 	if (ds == NULL) {
4236 		NFSDDSUNLOCK();
4237 		NFSD_DEBUG(4, "nfsrv_pnfscreate: no srv\n");
4238 		return;
4239 	}
4240 	i = dsdir[0] = ds->nfsdev_nextdir;
4241 	ds->nfsdev_nextdir = (ds->nfsdev_nextdir + 1) % nfsrv_dsdirsize;
4242 	dvp[0] = ds->nfsdev_dsdir[i];
4243 	tds = TAILQ_NEXT(ds, nfsdev_list);
4244 	if (nfsrv_maxpnfsmirror > 1 && tds != NULL) {
4245 		TAILQ_FOREACH_FROM(tds, &nfsrv_devidhead, nfsdev_list) {
4246 			if (tds->nfsdev_nmp != NULL &&
4247 			    ((tds->nfsdev_mdsisset == 0 && fds == NULL) ||
4248 			     (tds->nfsdev_mdsisset != 0 && fds != NULL &&
4249 			      fsidcmp(&mp->mnt_stat.f_fsid,
4250 			      &tds->nfsdev_mdsfsid) == 0))) {
4251 				dsdir[mirrorcnt] = i;
4252 				dvp[mirrorcnt] = tds->nfsdev_dsdir[i];
4253 				mirrorcnt++;
4254 				if (mirrorcnt >= nfsrv_maxpnfsmirror)
4255 					break;
4256 			}
4257 		}
4258 	}
4259 	/* Put at end of list to implement round-robin usage. */
4260 	TAILQ_REMOVE(&nfsrv_devidhead, ds, nfsdev_list);
4261 	TAILQ_INSERT_TAIL(&nfsrv_devidhead, ds, nfsdev_list);
4262 	NFSDDSUNLOCK();
4263 	dsc = NULL;
4264 	if (mirrorcnt > 1)
4265 		tdsc = dsc = malloc(sizeof(*dsc) * (mirrorcnt - 1), M_TEMP,
4266 		    M_WAITOK | M_ZERO);
4267 	tpf = pf = malloc(sizeof(*pf) * nfsrv_maxpnfsmirror, M_TEMP, M_WAITOK |
4268 	    M_ZERO);
4269 
4270 	error = nfsvno_getfh(vp, &fh, p);
4271 	if (error == 0)
4272 		error = VOP_GETATTR(vp, &va, cred);
4273 	if (error == 0) {
4274 		/* Set the attributes for "vp" to Setattr the DS vp. */
4275 		vauid = va.va_uid;
4276 		vagid = va.va_gid;
4277 		vamode = va.va_mode;
4278 		VATTR_NULL(&va);
4279 		va.va_uid = vauid;
4280 		va.va_gid = vagid;
4281 		va.va_mode = vamode;
4282 		va.va_size = 0;
4283 	} else
4284 		printf("pNFS: pnfscreate getfh+attr=%d\n", error);
4285 
4286 	NFSD_DEBUG(4, "nfsrv_pnfscreate: cruid=%d crgid=%d\n", cred->cr_uid,
4287 	    cred->cr_gid);
4288 	/* Make data file name based on FH. */
4289 	tcred = newnfs_getcred();
4290 
4291 	/*
4292 	 * Create the file on each DS mirror, using kernel process(es) for the
4293 	 * additional mirrors.
4294 	 */
4295 	failpos = -1;
4296 	for (i = 0; i < mirrorcnt - 1 && error == 0; i++, tpf++, tdsc++) {
4297 		tpf->dsf_dir = dsdir[i];
4298 		tdsc->tcred = tcred;
4299 		tdsc->p = p;
4300 		tdsc->pf = tpf;
4301 		tdsc->createva = *vap;
4302 		NFSBCOPY(&fh, &tdsc->fh, sizeof(fh));
4303 		tdsc->va = va;
4304 		tdsc->dvp = dvp[i];
4305 		tdsc->done = 0;
4306 		tdsc->inprog = 0;
4307 		tdsc->err = 0;
4308 		ret = EIO;
4309 		if (nfs_pnfsiothreads != 0) {
4310 			ret = nfs_pnfsio(start_dscreate, tdsc);
4311 			NFSD_DEBUG(4, "nfsrv_pnfscreate: nfs_pnfsio=%d\n", ret);
4312 		}
4313 		if (ret != 0) {
4314 			ret = nfsrv_dscreate(dvp[i], vap, &va, &fh, tpf, NULL,
4315 			    NULL, tcred, p, NULL);
4316 			if (ret != 0) {
4317 				KASSERT(error == 0, ("nfsrv_dscreate err=%d",
4318 				    error));
4319 				if (failpos == -1 && nfsds_failerr(ret))
4320 					failpos = i;
4321 				else
4322 					error = ret;
4323 			}
4324 		}
4325 	}
4326 	if (error == 0) {
4327 		tpf->dsf_dir = dsdir[mirrorcnt - 1];
4328 		error = nfsrv_dscreate(dvp[mirrorcnt - 1], vap, &va, &fh, tpf,
4329 		    &dsattr, NULL, tcred, p, NULL);
4330 		if (failpos == -1 && mirrorcnt > 1 && nfsds_failerr(error)) {
4331 			failpos = mirrorcnt - 1;
4332 			error = 0;
4333 		}
4334 	}
4335 	timo = hz / 50;		/* Wait for 20msec. */
4336 	if (timo < 1)
4337 		timo = 1;
4338 	/* Wait for kernel task(s) to complete. */
4339 	for (tdsc = dsc, i = 0; i < mirrorcnt - 1; i++, tdsc++) {
4340 		while (tdsc->inprog != 0 && tdsc->done == 0)
4341 			tsleep(&tdsc->tsk, PVFS, "srvdcr", timo);
4342 		if (tdsc->err != 0) {
4343 			if (failpos == -1 && nfsds_failerr(tdsc->err))
4344 				failpos = i;
4345 			else if (error == 0)
4346 				error = tdsc->err;
4347 		}
4348 	}
4349 
4350 	/*
4351 	 * If failpos has been set, that mirror has failed, so it needs
4352 	 * to be disabled.
4353 	 */
4354 	if (failpos >= 0) {
4355 		nmp = VFSTONFS(dvp[failpos]->v_mount);
4356 		NFSLOCKMNT(nmp);
4357 		if ((nmp->nm_privflag & (NFSMNTP_FORCEDISM |
4358 		     NFSMNTP_CANCELRPCS)) == 0) {
4359 			nmp->nm_privflag |= NFSMNTP_CANCELRPCS;
4360 			NFSUNLOCKMNT(nmp);
4361 			ds = nfsrv_deldsnmp(PNFSDOP_DELDSSERVER, nmp, p);
4362 			NFSD_DEBUG(4, "dscreatfail fail=%d ds=%p\n", failpos,
4363 			    ds);
4364 			if (ds != NULL)
4365 				nfsrv_killrpcs(nmp);
4366 			NFSLOCKMNT(nmp);
4367 			nmp->nm_privflag &= ~NFSMNTP_CANCELRPCS;
4368 			wakeup(nmp);
4369 		}
4370 		NFSUNLOCKMNT(nmp);
4371 	}
4372 
4373 	NFSFREECRED(tcred);
4374 	if (error == 0) {
4375 		ASSERT_VOP_ELOCKED(vp, "nfsrv_pnfscreate vp");
4376 
4377 		NFSD_DEBUG(4, "nfsrv_pnfscreate: mirrorcnt=%d maxmirror=%d\n",
4378 		    mirrorcnt, nfsrv_maxpnfsmirror);
4379 		/*
4380 		 * For all mirrors that couldn't be created, fill in the
4381 		 * *pf structure, but with an IP address == 0.0.0.0.
4382 		 */
4383 		tpf = pf + mirrorcnt;
4384 		for (i = mirrorcnt; i < nfsrv_maxpnfsmirror; i++, tpf++) {
4385 			*tpf = *pf;
4386 			tpf->dsf_sin.sin_family = AF_INET;
4387 			tpf->dsf_sin.sin_len = sizeof(struct sockaddr_in);
4388 			tpf->dsf_sin.sin_addr.s_addr = 0;
4389 			tpf->dsf_sin.sin_port = 0;
4390 		}
4391 
4392 		error = vn_extattr_set(vp, IO_NODELOCKED,
4393 		    EXTATTR_NAMESPACE_SYSTEM, "pnfsd.dsfile",
4394 		    sizeof(*pf) * nfsrv_maxpnfsmirror, (char *)pf, p);
4395 		if (error == 0)
4396 			error = vn_extattr_set(vp, IO_NODELOCKED,
4397 			    EXTATTR_NAMESPACE_SYSTEM, "pnfsd.dsattr",
4398 			    sizeof(dsattr), (char *)&dsattr, p);
4399 		if (error != 0)
4400 			printf("pNFS: pnfscreate setextattr=%d\n",
4401 			    error);
4402 	} else
4403 		printf("pNFS: pnfscreate=%d\n", error);
4404 	free(pf, M_TEMP);
4405 	free(dsc, M_TEMP);
4406 }
4407 
4408 /*
4409  * Get the information needed to remove the pNFS Data Server file from the
4410  * Metadata file.  Upon success, ddvp is set non-NULL to the locked
4411  * DS directory vnode.  The caller must unlock *ddvp when done with it.
4412  */
4413 static void
4414 nfsrv_pnfsremovesetup(struct vnode *vp, NFSPROC_T *p, struct vnode **dvpp,
4415     int *mirrorcntp, char *fname, fhandle_t *fhp)
4416 {
4417 	struct vattr va;
4418 	struct ucred *tcred;
4419 	char *buf;
4420 	int buflen, error;
4421 
4422 	dvpp[0] = NULL;
4423 	/* If not an exported regular file or not a pNFS server, just return. */
4424 	if (vp->v_type != VREG || (vp->v_mount->mnt_flag & MNT_EXPORTED) == 0 ||
4425 	    nfsrv_devidcnt == 0)
4426 		return;
4427 
4428 	/* Check to see if this is the last hard link. */
4429 	tcred = newnfs_getcred();
4430 	error = VOP_GETATTR(vp, &va, tcred);
4431 	NFSFREECRED(tcred);
4432 	if (error != 0) {
4433 		printf("pNFS: nfsrv_pnfsremovesetup getattr=%d\n", error);
4434 		return;
4435 	}
4436 	if (va.va_nlink > 1)
4437 		return;
4438 
4439 	error = nfsvno_getfh(vp, fhp, p);
4440 	if (error != 0) {
4441 		printf("pNFS: nfsrv_pnfsremovesetup getfh=%d\n", error);
4442 		return;
4443 	}
4444 
4445 	buflen = 1024;
4446 	buf = malloc(buflen, M_TEMP, M_WAITOK);
4447 	/* Get the directory vnode for the DS mount and the file handle. */
4448 	error = nfsrv_dsgetsockmnt(vp, 0, buf, &buflen, mirrorcntp, p, dvpp,
4449 	    NULL, NULL, fname, NULL, NULL, NULL, NULL, NULL);
4450 	free(buf, M_TEMP);
4451 	if (error != 0)
4452 		printf("pNFS: nfsrv_pnfsremovesetup getsockmnt=%d\n", error);
4453 }
4454 
4455 /*
4456  * Remove a DS data file for nfsrv_pnfsremove(). Called for each mirror.
4457  * The arguments are in a structure, so that they can be passed through
4458  * taskqueue for a kernel process to execute this function.
4459  */
4460 struct nfsrvdsremove {
4461 	int			done;
4462 	int			inprog;
4463 	struct task		tsk;
4464 	struct ucred		*tcred;
4465 	struct vnode		*dvp;
4466 	NFSPROC_T		*p;
4467 	int			err;
4468 	char			fname[PNFS_FILENAME_LEN + 1];
4469 };
4470 
4471 static int
4472 nfsrv_dsremove(struct vnode *dvp, char *fname, struct ucred *tcred,
4473     NFSPROC_T *p)
4474 {
4475 	struct nameidata named;
4476 	struct vnode *nvp;
4477 	char *bufp;
4478 	u_long *hashp;
4479 	int error;
4480 
4481 	error = NFSVOPLOCK(dvp, LK_EXCLUSIVE);
4482 	if (error != 0)
4483 		return (error);
4484 	named.ni_cnd.cn_nameiop = DELETE;
4485 	named.ni_cnd.cn_lkflags = LK_EXCLUSIVE | LK_RETRY;
4486 	named.ni_cnd.cn_cred = tcred;
4487 	named.ni_cnd.cn_thread = p;
4488 	named.ni_cnd.cn_flags = ISLASTCN | LOCKPARENT | LOCKLEAF | SAVENAME;
4489 	nfsvno_setpathbuf(&named, &bufp, &hashp);
4490 	named.ni_cnd.cn_nameptr = bufp;
4491 	named.ni_cnd.cn_namelen = strlen(fname);
4492 	strlcpy(bufp, fname, NAME_MAX);
4493 	NFSD_DEBUG(4, "nfsrv_pnfsremove: filename=%s\n", bufp);
4494 	error = VOP_LOOKUP(dvp, &nvp, &named.ni_cnd);
4495 	NFSD_DEBUG(4, "nfsrv_pnfsremove: aft LOOKUP=%d\n", error);
4496 	if (error == 0) {
4497 		error = VOP_REMOVE(dvp, nvp, &named.ni_cnd);
4498 		vput(nvp);
4499 	}
4500 	NFSVOPUNLOCK(dvp);
4501 	nfsvno_relpathbuf(&named);
4502 	if (error != 0)
4503 		printf("pNFS: nfsrv_pnfsremove failed=%d\n", error);
4504 	return (error);
4505 }
4506 
4507 /*
4508  * Start up the thread that will execute nfsrv_dsremove().
4509  */
4510 static void
4511 start_dsremove(void *arg, int pending)
4512 {
4513 	struct nfsrvdsremove *dsrm;
4514 
4515 	dsrm = (struct nfsrvdsremove *)arg;
4516 	dsrm->err = nfsrv_dsremove(dsrm->dvp, dsrm->fname, dsrm->tcred,
4517 	    dsrm->p);
4518 	dsrm->done = 1;
4519 	NFSD_DEBUG(4, "start_dsremove: err=%d\n", dsrm->err);
4520 }
4521 
4522 /*
4523  * Remove a pNFS data file from a Data Server.
4524  * nfsrv_pnfsremovesetup() must have been called before the MDS file was
4525  * removed to set up the dvp and fill in the FH.
4526  */
4527 static void
4528 nfsrv_pnfsremove(struct vnode **dvp, int mirrorcnt, char *fname, fhandle_t *fhp,
4529     NFSPROC_T *p)
4530 {
4531 	struct ucred *tcred;
4532 	struct nfsrvdsremove *dsrm, *tdsrm;
4533 	struct nfsdevice *ds;
4534 	struct nfsmount *nmp;
4535 	int failpos, i, ret, timo;
4536 
4537 	tcred = newnfs_getcred();
4538 	dsrm = NULL;
4539 	if (mirrorcnt > 1)
4540 		dsrm = malloc(sizeof(*dsrm) * mirrorcnt - 1, M_TEMP, M_WAITOK);
4541 	/*
4542 	 * Remove the file on each DS mirror, using kernel process(es) for the
4543 	 * additional mirrors.
4544 	 */
4545 	failpos = -1;
4546 	for (tdsrm = dsrm, i = 0; i < mirrorcnt - 1; i++, tdsrm++) {
4547 		tdsrm->tcred = tcred;
4548 		tdsrm->p = p;
4549 		tdsrm->dvp = dvp[i];
4550 		strlcpy(tdsrm->fname, fname, PNFS_FILENAME_LEN + 1);
4551 		tdsrm->inprog = 0;
4552 		tdsrm->done = 0;
4553 		tdsrm->err = 0;
4554 		ret = EIO;
4555 		if (nfs_pnfsiothreads != 0) {
4556 			ret = nfs_pnfsio(start_dsremove, tdsrm);
4557 			NFSD_DEBUG(4, "nfsrv_pnfsremove: nfs_pnfsio=%d\n", ret);
4558 		}
4559 		if (ret != 0) {
4560 			ret = nfsrv_dsremove(dvp[i], fname, tcred, p);
4561 			if (failpos == -1 && nfsds_failerr(ret))
4562 				failpos = i;
4563 		}
4564 	}
4565 	ret = nfsrv_dsremove(dvp[mirrorcnt - 1], fname, tcred, p);
4566 	if (failpos == -1 && mirrorcnt > 1 && nfsds_failerr(ret))
4567 		failpos = mirrorcnt - 1;
4568 	timo = hz / 50;		/* Wait for 20msec. */
4569 	if (timo < 1)
4570 		timo = 1;
4571 	/* Wait for kernel task(s) to complete. */
4572 	for (tdsrm = dsrm, i = 0; i < mirrorcnt - 1; i++, tdsrm++) {
4573 		while (tdsrm->inprog != 0 && tdsrm->done == 0)
4574 			tsleep(&tdsrm->tsk, PVFS, "srvdsrm", timo);
4575 		if (failpos == -1 && nfsds_failerr(tdsrm->err))
4576 			failpos = i;
4577 	}
4578 
4579 	/*
4580 	 * If failpos has been set, that mirror has failed, so it needs
4581 	 * to be disabled.
4582 	 */
4583 	if (failpos >= 0) {
4584 		nmp = VFSTONFS(dvp[failpos]->v_mount);
4585 		NFSLOCKMNT(nmp);
4586 		if ((nmp->nm_privflag & (NFSMNTP_FORCEDISM |
4587 		     NFSMNTP_CANCELRPCS)) == 0) {
4588 			nmp->nm_privflag |= NFSMNTP_CANCELRPCS;
4589 			NFSUNLOCKMNT(nmp);
4590 			ds = nfsrv_deldsnmp(PNFSDOP_DELDSSERVER, nmp, p);
4591 			NFSD_DEBUG(4, "dsremovefail fail=%d ds=%p\n", failpos,
4592 			    ds);
4593 			if (ds != NULL)
4594 				nfsrv_killrpcs(nmp);
4595 			NFSLOCKMNT(nmp);
4596 			nmp->nm_privflag &= ~NFSMNTP_CANCELRPCS;
4597 			wakeup(nmp);
4598 		}
4599 		NFSUNLOCKMNT(nmp);
4600 	}
4601 
4602 	/* Get rid all layouts for the file. */
4603 	nfsrv_freefilelayouts(fhp);
4604 
4605 	NFSFREECRED(tcred);
4606 	free(dsrm, M_TEMP);
4607 }
4608 
4609 /*
4610  * Generate a file name based on the file handle and put it in *bufp.
4611  * Return the number of bytes generated.
4612  */
4613 static int
4614 nfsrv_putfhname(fhandle_t *fhp, char *bufp)
4615 {
4616 	int i;
4617 	uint8_t *cp;
4618 	const uint8_t *hexdigits = "0123456789abcdef";
4619 
4620 	cp = (uint8_t *)fhp;
4621 	for (i = 0; i < sizeof(*fhp); i++) {
4622 		bufp[2 * i] = hexdigits[(*cp >> 4) & 0xf];
4623 		bufp[2 * i + 1] = hexdigits[*cp++ & 0xf];
4624 	}
4625 	bufp[2 * i] = '\0';
4626 	return (2 * i);
4627 }
4628 
4629 /*
4630  * Update the Metadata file's attributes from the DS file when a Read/Write
4631  * layout is returned.
4632  * Basically just call nfsrv_proxyds() with procedure == NFSPROC_LAYOUTRETURN
4633  * so that it does a nfsrv_getattrdsrpc() and nfsrv_setextattr() on the DS file.
4634  */
4635 int
4636 nfsrv_updatemdsattr(struct vnode *vp, struct nfsvattr *nap, NFSPROC_T *p)
4637 {
4638 	struct ucred *tcred;
4639 	int error;
4640 
4641 	/* Do this as root so that it won't fail with EACCES. */
4642 	tcred = newnfs_getcred();
4643 	error = nfsrv_proxyds(vp, 0, 0, tcred, p, NFSPROC_LAYOUTRETURN,
4644 	    NULL, NULL, NULL, nap, NULL, NULL, 0, NULL);
4645 	NFSFREECRED(tcred);
4646 	return (error);
4647 }
4648 
4649 /*
4650  * Set the NFSv4 ACL on the DS file to the same ACL as the MDS file.
4651  */
4652 static int
4653 nfsrv_dssetacl(struct vnode *vp, struct acl *aclp, struct ucred *cred,
4654     NFSPROC_T *p)
4655 {
4656 	int error;
4657 
4658 	error = nfsrv_proxyds(vp, 0, 0, cred, p, NFSPROC_SETACL,
4659 	    NULL, NULL, NULL, NULL, aclp, NULL, 0, NULL);
4660 	return (error);
4661 }
4662 
4663 static int
4664 nfsrv_proxyds(struct vnode *vp, off_t off, int cnt, struct ucred *cred,
4665     struct thread *p, int ioproc, struct mbuf **mpp, char *cp,
4666     struct mbuf **mpp2, struct nfsvattr *nap, struct acl *aclp,
4667     off_t *offp, int content, bool *eofp)
4668 {
4669 	struct nfsmount *nmp[NFSDEV_MAXMIRRORS], *failnmp;
4670 	fhandle_t fh[NFSDEV_MAXMIRRORS];
4671 	struct vnode *dvp[NFSDEV_MAXMIRRORS];
4672 	struct nfsdevice *ds;
4673 	struct pnfsdsattr dsattr;
4674 	struct opnfsdsattr odsattr;
4675 	char *buf;
4676 	int buflen, error, failpos, i, mirrorcnt, origmircnt, trycnt;
4677 
4678 	NFSD_DEBUG(4, "in nfsrv_proxyds\n");
4679 	/*
4680 	 * If not a regular file, not exported or not a pNFS server,
4681 	 * just return ENOENT.
4682 	 */
4683 	if (vp->v_type != VREG || (vp->v_mount->mnt_flag & MNT_EXPORTED) == 0 ||
4684 	    nfsrv_devidcnt == 0)
4685 		return (ENOENT);
4686 
4687 	buflen = 1024;
4688 	buf = malloc(buflen, M_TEMP, M_WAITOK);
4689 	error = 0;
4690 
4691 	/*
4692 	 * For Getattr, get the Change attribute (va_filerev) and size (va_size)
4693 	 * from the MetaData file's extended attribute.
4694 	 */
4695 	if (ioproc == NFSPROC_GETATTR) {
4696 		error = vn_extattr_get(vp, IO_NODELOCKED,
4697 		    EXTATTR_NAMESPACE_SYSTEM, "pnfsd.dsattr", &buflen, buf,
4698 		    p);
4699 		if (error == 0) {
4700 			if (buflen == sizeof(odsattr)) {
4701 				NFSBCOPY(buf, &odsattr, buflen);
4702 				nap->na_filerev = odsattr.dsa_filerev;
4703 				nap->na_size = odsattr.dsa_size;
4704 				nap->na_atime = odsattr.dsa_atime;
4705 				nap->na_mtime = odsattr.dsa_mtime;
4706 				/*
4707 				 * Fake na_bytes by rounding up na_size.
4708 				 * Since we don't know the block size, just
4709 				 * use BLKDEV_IOSIZE.
4710 				 */
4711 				nap->na_bytes = (odsattr.dsa_size +
4712 				    BLKDEV_IOSIZE - 1) & ~(BLKDEV_IOSIZE - 1);
4713 			} else if (buflen == sizeof(dsattr)) {
4714 				NFSBCOPY(buf, &dsattr, buflen);
4715 				nap->na_filerev = dsattr.dsa_filerev;
4716 				nap->na_size = dsattr.dsa_size;
4717 				nap->na_atime = dsattr.dsa_atime;
4718 				nap->na_mtime = dsattr.dsa_mtime;
4719 				nap->na_bytes = dsattr.dsa_bytes;
4720 			} else
4721 				error = ENXIO;
4722 		}
4723 		if (error == 0) {
4724 			/*
4725 			 * If nfsrv_pnfsgetdsattr is 0 or nfsrv_checkdsattr()
4726 			 * returns 0, just return now.  nfsrv_checkdsattr()
4727 			 * returns 0 if there is no Read/Write layout
4728 			 * plus either an Open/Write_access or Write
4729 			 * delegation issued to a client for the file.
4730 			 */
4731 			if (nfsrv_pnfsgetdsattr == 0 ||
4732 			    nfsrv_checkdsattr(vp, p) == 0) {
4733 				free(buf, M_TEMP);
4734 				return (error);
4735 			}
4736 		}
4737 
4738 		/*
4739 		 * Clear ENOATTR so the code below will attempt to do a
4740 		 * nfsrv_getattrdsrpc() to get the attributes and (re)create
4741 		 * the extended attribute.
4742 		 */
4743 		if (error == ENOATTR)
4744 			error = 0;
4745 	}
4746 
4747 	origmircnt = -1;
4748 	trycnt = 0;
4749 tryagain:
4750 	if (error == 0) {
4751 		buflen = 1024;
4752 		if (ioproc == NFSPROC_READDS && NFSVOPISLOCKED(vp) ==
4753 		    LK_EXCLUSIVE)
4754 			printf("nfsrv_proxyds: Readds vp exclusively locked\n");
4755 		error = nfsrv_dsgetsockmnt(vp, LK_SHARED, buf, &buflen,
4756 		    &mirrorcnt, p, dvp, fh, NULL, NULL, NULL, NULL, NULL,
4757 		    NULL, NULL);
4758 		if (error == 0) {
4759 			for (i = 0; i < mirrorcnt; i++)
4760 				nmp[i] = VFSTONFS(dvp[i]->v_mount);
4761 		} else
4762 			printf("pNFS: proxy getextattr sockaddr=%d\n", error);
4763 	} else
4764 		printf("pNFS: nfsrv_dsgetsockmnt=%d\n", error);
4765 	if (error == 0) {
4766 		failpos = -1;
4767 		if (origmircnt == -1)
4768 			origmircnt = mirrorcnt;
4769 		/*
4770 		 * If failpos is set to a mirror#, then that mirror has
4771 		 * failed and will be disabled. For Read, Getattr and Seek, the
4772 		 * function only tries one mirror, so if that mirror has
4773 		 * failed, it will need to be retried. As such, increment
4774 		 * tryitagain for these cases.
4775 		 * For Write, Setattr and Setacl, the function tries all
4776 		 * mirrors and will not return an error for the case where
4777 		 * one mirror has failed. For these cases, the functioning
4778 		 * mirror(s) will have been modified, so a retry isn't
4779 		 * necessary. These functions will set failpos for the
4780 		 * failed mirror#.
4781 		 */
4782 		if (ioproc == NFSPROC_READDS) {
4783 			error = nfsrv_readdsrpc(fh, off, cnt, cred, p, nmp[0],
4784 			    mpp, mpp2);
4785 			if (nfsds_failerr(error) && mirrorcnt > 1) {
4786 				/*
4787 				 * Setting failpos will cause the mirror
4788 				 * to be disabled and then a retry of this
4789 				 * read is required.
4790 				 */
4791 				failpos = 0;
4792 				error = 0;
4793 				trycnt++;
4794 			}
4795 		} else if (ioproc == NFSPROC_WRITEDS)
4796 			error = nfsrv_writedsrpc(fh, off, cnt, cred, p, vp,
4797 			    &nmp[0], mirrorcnt, mpp, cp, &failpos);
4798 		else if (ioproc == NFSPROC_SETATTR)
4799 			error = nfsrv_setattrdsrpc(fh, cred, p, vp, &nmp[0],
4800 			    mirrorcnt, nap, &failpos);
4801 		else if (ioproc == NFSPROC_SETACL)
4802 			error = nfsrv_setacldsrpc(fh, cred, p, vp, &nmp[0],
4803 			    mirrorcnt, aclp, &failpos);
4804 		else if (ioproc == NFSPROC_SEEKDS) {
4805 			error = nfsrv_seekdsrpc(fh, offp, content, eofp, cred,
4806 			    p, nmp[0]);
4807 			if (nfsds_failerr(error) && mirrorcnt > 1) {
4808 				/*
4809 				 * Setting failpos will cause the mirror
4810 				 * to be disabled and then a retry of this
4811 				 * read is required.
4812 				 */
4813 				failpos = 0;
4814 				error = 0;
4815 				trycnt++;
4816 			}
4817 		} else if (ioproc == NFSPROC_ALLOCATE)
4818 			error = nfsrv_allocatedsrpc(fh, off, *offp, cred, p, vp,
4819 			    &nmp[0], mirrorcnt, &failpos);
4820 		else {
4821 			error = nfsrv_getattrdsrpc(&fh[mirrorcnt - 1], cred, p,
4822 			    vp, nmp[mirrorcnt - 1], nap);
4823 			if (nfsds_failerr(error) && mirrorcnt > 1) {
4824 				/*
4825 				 * Setting failpos will cause the mirror
4826 				 * to be disabled and then a retry of this
4827 				 * getattr is required.
4828 				 */
4829 				failpos = mirrorcnt - 1;
4830 				error = 0;
4831 				trycnt++;
4832 			}
4833 		}
4834 		ds = NULL;
4835 		if (failpos >= 0) {
4836 			failnmp = nmp[failpos];
4837 			NFSLOCKMNT(failnmp);
4838 			if ((failnmp->nm_privflag & (NFSMNTP_FORCEDISM |
4839 			     NFSMNTP_CANCELRPCS)) == 0) {
4840 				failnmp->nm_privflag |= NFSMNTP_CANCELRPCS;
4841 				NFSUNLOCKMNT(failnmp);
4842 				ds = nfsrv_deldsnmp(PNFSDOP_DELDSSERVER,
4843 				    failnmp, p);
4844 				NFSD_DEBUG(4, "dsldsnmp fail=%d ds=%p\n",
4845 				    failpos, ds);
4846 				if (ds != NULL)
4847 					nfsrv_killrpcs(failnmp);
4848 				NFSLOCKMNT(failnmp);
4849 				failnmp->nm_privflag &= ~NFSMNTP_CANCELRPCS;
4850 				wakeup(failnmp);
4851 			}
4852 			NFSUNLOCKMNT(failnmp);
4853 		}
4854 		for (i = 0; i < mirrorcnt; i++)
4855 			NFSVOPUNLOCK(dvp[i]);
4856 		NFSD_DEBUG(4, "nfsrv_proxyds: aft RPC=%d trya=%d\n", error,
4857 		    trycnt);
4858 		/* Try the Read/Getattr again if a mirror was deleted. */
4859 		if (ds != NULL && trycnt > 0 && trycnt < origmircnt)
4860 			goto tryagain;
4861 	} else {
4862 		/* Return ENOENT for any Extended Attribute error. */
4863 		error = ENOENT;
4864 	}
4865 	free(buf, M_TEMP);
4866 	NFSD_DEBUG(4, "nfsrv_proxyds: error=%d\n", error);
4867 	return (error);
4868 }
4869 
4870 /*
4871  * Get the DS mount point, fh and directory from the "pnfsd.dsfile" extended
4872  * attribute.
4873  * newnmpp - If it points to a non-NULL nmp, that is the destination and needs
4874  *           to be checked.  If it points to a NULL nmp, then it returns
4875  *           a suitable destination.
4876  * curnmp - If non-NULL, it is the source mount for the copy.
4877  */
4878 int
4879 nfsrv_dsgetsockmnt(struct vnode *vp, int lktype, char *buf, int *buflenp,
4880     int *mirrorcntp, NFSPROC_T *p, struct vnode **dvpp, fhandle_t *fhp,
4881     char *devid, char *fnamep, struct vnode **nvpp, struct nfsmount **newnmpp,
4882     struct nfsmount *curnmp, int *ippos, int *dsdirp)
4883 {
4884 	struct vnode *dvp, *nvp = NULL, **tdvpp;
4885 	struct mount *mp;
4886 	struct nfsmount *nmp, *newnmp;
4887 	struct sockaddr *sad;
4888 	struct sockaddr_in *sin;
4889 	struct nfsdevice *ds, *tds, *fndds;
4890 	struct pnfsdsfile *pf;
4891 	uint32_t dsdir;
4892 	int error, fhiszero, fnd, gotone, i, mirrorcnt;
4893 
4894 	ASSERT_VOP_LOCKED(vp, "nfsrv_dsgetsockmnt vp");
4895 	*mirrorcntp = 1;
4896 	tdvpp = dvpp;
4897 	if (nvpp != NULL)
4898 		*nvpp = NULL;
4899 	if (dvpp != NULL)
4900 		*dvpp = NULL;
4901 	if (ippos != NULL)
4902 		*ippos = -1;
4903 	if (newnmpp != NULL)
4904 		newnmp = *newnmpp;
4905 	else
4906 		newnmp = NULL;
4907 	mp = vp->v_mount;
4908 	error = vn_extattr_get(vp, IO_NODELOCKED, EXTATTR_NAMESPACE_SYSTEM,
4909 	    "pnfsd.dsfile", buflenp, buf, p);
4910 	mirrorcnt = *buflenp / sizeof(*pf);
4911 	if (error == 0 && (mirrorcnt < 1 || mirrorcnt > NFSDEV_MAXMIRRORS ||
4912 	    *buflenp != sizeof(*pf) * mirrorcnt))
4913 		error = ENOATTR;
4914 
4915 	pf = (struct pnfsdsfile *)buf;
4916 	/* If curnmp != NULL, check for a match in the mirror list. */
4917 	if (curnmp != NULL && error == 0) {
4918 		fnd = 0;
4919 		for (i = 0; i < mirrorcnt; i++, pf++) {
4920 			sad = (struct sockaddr *)&pf->dsf_sin;
4921 			if (nfsaddr2_match(sad, curnmp->nm_nam)) {
4922 				if (ippos != NULL)
4923 					*ippos = i;
4924 				fnd = 1;
4925 				break;
4926 			}
4927 		}
4928 		if (fnd == 0)
4929 			error = ENXIO;
4930 	}
4931 
4932 	gotone = 0;
4933 	pf = (struct pnfsdsfile *)buf;
4934 	NFSD_DEBUG(4, "nfsrv_dsgetsockmnt: mirrorcnt=%d err=%d\n", mirrorcnt,
4935 	    error);
4936 	for (i = 0; i < mirrorcnt && error == 0; i++, pf++) {
4937 		fhiszero = 0;
4938 		sad = (struct sockaddr *)&pf->dsf_sin;
4939 		sin = &pf->dsf_sin;
4940 		dsdir = pf->dsf_dir;
4941 		if (dsdir >= nfsrv_dsdirsize) {
4942 			printf("nfsrv_dsgetsockmnt: dsdir=%d\n", dsdir);
4943 			error = ENOATTR;
4944 		} else if (nvpp != NULL && newnmp != NULL &&
4945 		    nfsaddr2_match(sad, newnmp->nm_nam))
4946 			error = EEXIST;
4947 		if (error == 0) {
4948 			if (ippos != NULL && curnmp == NULL &&
4949 			    sad->sa_family == AF_INET &&
4950 			    sin->sin_addr.s_addr == 0)
4951 				*ippos = i;
4952 			if (NFSBCMP(&zerofh, &pf->dsf_fh, sizeof(zerofh)) == 0)
4953 				fhiszero = 1;
4954 			/* Use the socket address to find the mount point. */
4955 			fndds = NULL;
4956 			NFSDDSLOCK();
4957 			/* Find a match for the IP address. */
4958 			TAILQ_FOREACH(ds, &nfsrv_devidhead, nfsdev_list) {
4959 				if (ds->nfsdev_nmp != NULL) {
4960 					dvp = ds->nfsdev_dvp;
4961 					nmp = VFSTONFS(dvp->v_mount);
4962 					if (nmp != ds->nfsdev_nmp)
4963 						printf("different2 nmp %p %p\n",
4964 						    nmp, ds->nfsdev_nmp);
4965 					if (nfsaddr2_match(sad, nmp->nm_nam)) {
4966 						fndds = ds;
4967 						break;
4968 					}
4969 				}
4970 			}
4971 			if (fndds != NULL && newnmpp != NULL &&
4972 			    newnmp == NULL) {
4973 				/* Search for a place to make a mirror copy. */
4974 				TAILQ_FOREACH(tds, &nfsrv_devidhead,
4975 				    nfsdev_list) {
4976 					if (tds->nfsdev_nmp != NULL &&
4977 					    fndds != tds &&
4978 					    ((tds->nfsdev_mdsisset == 0 &&
4979 					      fndds->nfsdev_mdsisset == 0) ||
4980 					     (tds->nfsdev_mdsisset != 0 &&
4981 					      fndds->nfsdev_mdsisset != 0 &&
4982 					      fsidcmp(&tds->nfsdev_mdsfsid,
4983 					      &mp->mnt_stat.f_fsid) == 0))) {
4984 						*newnmpp = tds->nfsdev_nmp;
4985 						break;
4986 					}
4987 				}
4988 				if (tds != NULL) {
4989 					/*
4990 					 * Move this entry to the end of the
4991 					 * list, so it won't be selected as
4992 					 * easily the next time.
4993 					 */
4994 					TAILQ_REMOVE(&nfsrv_devidhead, tds,
4995 					    nfsdev_list);
4996 					TAILQ_INSERT_TAIL(&nfsrv_devidhead, tds,
4997 					    nfsdev_list);
4998 				}
4999 			}
5000 			NFSDDSUNLOCK();
5001 			if (fndds != NULL) {
5002 				dvp = fndds->nfsdev_dsdir[dsdir];
5003 				if (lktype != 0 || fhiszero != 0 ||
5004 				    (nvpp != NULL && *nvpp == NULL)) {
5005 					if (fhiszero != 0)
5006 						error = vn_lock(dvp,
5007 						    LK_EXCLUSIVE);
5008 					else if (lktype != 0)
5009 						error = vn_lock(dvp, lktype);
5010 					else
5011 						error = vn_lock(dvp, LK_SHARED);
5012 					/*
5013 					 * If the file handle is all 0's, try to
5014 					 * do a Lookup against the DS to acquire
5015 					 * it.
5016 					 * If dvpp == NULL or the Lookup fails,
5017 					 * unlock dvp after the call.
5018 					 */
5019 					if (error == 0 && (fhiszero != 0 ||
5020 					    (nvpp != NULL && *nvpp == NULL))) {
5021 						error = nfsrv_pnfslookupds(vp,
5022 						    dvp, pf, &nvp, p);
5023 						if (error == 0) {
5024 							if (fhiszero != 0)
5025 								nfsrv_pnfssetfh(
5026 								    vp, pf,
5027 								    devid,
5028 								    fnamep,
5029 								    nvp, p);
5030 							if (nvpp != NULL &&
5031 							    *nvpp == NULL) {
5032 								*nvpp = nvp;
5033 								*dsdirp = dsdir;
5034 							} else
5035 								vput(nvp);
5036 						}
5037 						if (error != 0 || lktype == 0)
5038 							NFSVOPUNLOCK(dvp);
5039 					}
5040 				}
5041 				if (error == 0) {
5042 					gotone++;
5043 					NFSD_DEBUG(4, "gotone=%d\n", gotone);
5044 					if (devid != NULL) {
5045 						NFSBCOPY(fndds->nfsdev_deviceid,
5046 						    devid, NFSX_V4DEVICEID);
5047 						devid += NFSX_V4DEVICEID;
5048 					}
5049 					if (dvpp != NULL)
5050 						*tdvpp++ = dvp;
5051 					if (fhp != NULL)
5052 						NFSBCOPY(&pf->dsf_fh, fhp++,
5053 						    NFSX_MYFH);
5054 					if (fnamep != NULL && gotone == 1)
5055 						strlcpy(fnamep,
5056 						    pf->dsf_filename,
5057 						    sizeof(pf->dsf_filename));
5058 				} else
5059 					NFSD_DEBUG(4, "nfsrv_dsgetsockmnt "
5060 					    "err=%d\n", error);
5061 			}
5062 		}
5063 	}
5064 	if (error == 0 && gotone == 0)
5065 		error = ENOENT;
5066 
5067 	NFSD_DEBUG(4, "eo nfsrv_dsgetsockmnt: gotone=%d err=%d\n", gotone,
5068 	    error);
5069 	if (error == 0)
5070 		*mirrorcntp = gotone;
5071 	else {
5072 		if (gotone > 0 && dvpp != NULL) {
5073 			/*
5074 			 * If the error didn't occur on the first one and
5075 			 * dvpp != NULL, the one(s) prior to the failure will
5076 			 * have locked dvp's that need to be unlocked.
5077 			 */
5078 			for (i = 0; i < gotone; i++) {
5079 				NFSVOPUNLOCK(*dvpp);
5080 				*dvpp++ = NULL;
5081 			}
5082 		}
5083 		/*
5084 		 * If it found the vnode to be copied from before a failure,
5085 		 * it needs to be vput()'d.
5086 		 */
5087 		if (nvpp != NULL && *nvpp != NULL) {
5088 			vput(*nvpp);
5089 			*nvpp = NULL;
5090 		}
5091 	}
5092 	return (error);
5093 }
5094 
5095 /*
5096  * Set the extended attribute for the Change attribute.
5097  */
5098 static int
5099 nfsrv_setextattr(struct vnode *vp, struct nfsvattr *nap, NFSPROC_T *p)
5100 {
5101 	struct pnfsdsattr dsattr;
5102 	int error;
5103 
5104 	ASSERT_VOP_ELOCKED(vp, "nfsrv_setextattr vp");
5105 	dsattr.dsa_filerev = nap->na_filerev;
5106 	dsattr.dsa_size = nap->na_size;
5107 	dsattr.dsa_atime = nap->na_atime;
5108 	dsattr.dsa_mtime = nap->na_mtime;
5109 	dsattr.dsa_bytes = nap->na_bytes;
5110 	error = vn_extattr_set(vp, IO_NODELOCKED, EXTATTR_NAMESPACE_SYSTEM,
5111 	    "pnfsd.dsattr", sizeof(dsattr), (char *)&dsattr, p);
5112 	if (error != 0)
5113 		printf("pNFS: setextattr=%d\n", error);
5114 	return (error);
5115 }
5116 
5117 static int
5118 nfsrv_readdsrpc(fhandle_t *fhp, off_t off, int len, struct ucred *cred,
5119     NFSPROC_T *p, struct nfsmount *nmp, struct mbuf **mpp, struct mbuf **mpendp)
5120 {
5121 	uint32_t *tl;
5122 	struct nfsrv_descript *nd;
5123 	nfsv4stateid_t st;
5124 	struct mbuf *m, *m2;
5125 	int error = 0, retlen, tlen, trimlen;
5126 
5127 	NFSD_DEBUG(4, "in nfsrv_readdsrpc\n");
5128 	nd = malloc(sizeof(*nd), M_TEMP, M_WAITOK | M_ZERO);
5129 	*mpp = NULL;
5130 	/*
5131 	 * Use a stateid where other is an alternating 01010 pattern and
5132 	 * seqid is 0xffffffff.  This value is not defined as special by
5133 	 * the RFC and is used by the FreeBSD NFS server to indicate an
5134 	 * MDS->DS proxy operation.
5135 	 */
5136 	st.other[0] = 0x55555555;
5137 	st.other[1] = 0x55555555;
5138 	st.other[2] = 0x55555555;
5139 	st.seqid = 0xffffffff;
5140 	nfscl_reqstart(nd, NFSPROC_READDS, nmp, (u_int8_t *)fhp, sizeof(*fhp),
5141 	    NULL, NULL, 0, 0);
5142 	nfsm_stateidtom(nd, &st, NFSSTATEID_PUTSTATEID);
5143 	NFSM_BUILD(tl, uint32_t *, NFSX_UNSIGNED * 3);
5144 	txdr_hyper(off, tl);
5145 	*(tl + 2) = txdr_unsigned(len);
5146 	error = newnfs_request(nd, nmp, NULL, &nmp->nm_sockreq, NULL, p, cred,
5147 	    NFS_PROG, NFS_VER4, NULL, 1, NULL, NULL);
5148 	if (error != 0) {
5149 		free(nd, M_TEMP);
5150 		return (error);
5151 	}
5152 	if (nd->nd_repstat == 0) {
5153 		NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
5154 		NFSM_STRSIZ(retlen, len);
5155 		if (retlen > 0) {
5156 			/* Trim off the pre-data XDR from the mbuf chain. */
5157 			m = nd->nd_mrep;
5158 			while (m != NULL && m != nd->nd_md) {
5159 				if (m->m_next == nd->nd_md) {
5160 					m->m_next = NULL;
5161 					m_freem(nd->nd_mrep);
5162 					nd->nd_mrep = m = nd->nd_md;
5163 				} else
5164 					m = m->m_next;
5165 			}
5166 			if (m == NULL) {
5167 				printf("nfsrv_readdsrpc: busted mbuf list\n");
5168 				error = ENOENT;
5169 				goto nfsmout;
5170 			}
5171 
5172 			/*
5173 			 * Now, adjust first mbuf so that any XDR before the
5174 			 * read data is skipped over.
5175 			 */
5176 			trimlen = nd->nd_dpos - mtod(m, char *);
5177 			if (trimlen > 0) {
5178 				m->m_len -= trimlen;
5179 				NFSM_DATAP(m, trimlen);
5180 			}
5181 
5182 			/*
5183 			 * Truncate the mbuf chain at retlen bytes of data,
5184 			 * plus XDR padding that brings the length up to a
5185 			 * multiple of 4.
5186 			 */
5187 			tlen = NFSM_RNDUP(retlen);
5188 			do {
5189 				if (m->m_len >= tlen) {
5190 					m->m_len = tlen;
5191 					tlen = 0;
5192 					m2 = m->m_next;
5193 					m->m_next = NULL;
5194 					m_freem(m2);
5195 					break;
5196 				}
5197 				tlen -= m->m_len;
5198 				m = m->m_next;
5199 			} while (m != NULL);
5200 			if (tlen > 0) {
5201 				printf("nfsrv_readdsrpc: busted mbuf list\n");
5202 				error = ENOENT;
5203 				goto nfsmout;
5204 			}
5205 			*mpp = nd->nd_mrep;
5206 			*mpendp = m;
5207 			nd->nd_mrep = NULL;
5208 		}
5209 	} else
5210 		error = nd->nd_repstat;
5211 nfsmout:
5212 	/* If nd->nd_mrep is already NULL, this is a no-op. */
5213 	m_freem(nd->nd_mrep);
5214 	free(nd, M_TEMP);
5215 	NFSD_DEBUG(4, "nfsrv_readdsrpc error=%d\n", error);
5216 	return (error);
5217 }
5218 
5219 /*
5220  * Do a write RPC on a DS data file, using this structure for the arguments,
5221  * so that this function can be executed by a separate kernel process.
5222  */
5223 struct nfsrvwritedsdorpc {
5224 	int			done;
5225 	int			inprog;
5226 	struct task		tsk;
5227 	fhandle_t		fh;
5228 	off_t			off;
5229 	int			len;
5230 	struct nfsmount		*nmp;
5231 	struct ucred		*cred;
5232 	NFSPROC_T		*p;
5233 	struct mbuf		*m;
5234 	int			err;
5235 };
5236 
5237 static int
5238 nfsrv_writedsdorpc(struct nfsmount *nmp, fhandle_t *fhp, off_t off, int len,
5239     struct nfsvattr *nap, struct mbuf *m, struct ucred *cred, NFSPROC_T *p)
5240 {
5241 	uint32_t *tl;
5242 	struct nfsrv_descript *nd;
5243 	nfsattrbit_t attrbits;
5244 	nfsv4stateid_t st;
5245 	int commit, error, retlen;
5246 
5247 	nd = malloc(sizeof(*nd), M_TEMP, M_WAITOK | M_ZERO);
5248 	nfscl_reqstart(nd, NFSPROC_WRITE, nmp, (u_int8_t *)fhp,
5249 	    sizeof(fhandle_t), NULL, NULL, 0, 0);
5250 
5251 	/*
5252 	 * Use a stateid where other is an alternating 01010 pattern and
5253 	 * seqid is 0xffffffff.  This value is not defined as special by
5254 	 * the RFC and is used by the FreeBSD NFS server to indicate an
5255 	 * MDS->DS proxy operation.
5256 	 */
5257 	st.other[0] = 0x55555555;
5258 	st.other[1] = 0x55555555;
5259 	st.other[2] = 0x55555555;
5260 	st.seqid = 0xffffffff;
5261 	nfsm_stateidtom(nd, &st, NFSSTATEID_PUTSTATEID);
5262 	NFSM_BUILD(tl, u_int32_t *, NFSX_HYPER + 2 * NFSX_UNSIGNED);
5263 	txdr_hyper(off, tl);
5264 	tl += 2;
5265 	/*
5266 	 * Do all writes FileSync, since the server doesn't hold onto dirty
5267 	 * buffers.  Since clients should be accessing the DS servers directly
5268 	 * using the pNFS layouts, this just needs to work correctly as a
5269 	 * fallback.
5270 	 */
5271 	*tl++ = txdr_unsigned(NFSWRITE_FILESYNC);
5272 	*tl = txdr_unsigned(len);
5273 	NFSD_DEBUG(4, "nfsrv_writedsdorpc: len=%d\n", len);
5274 
5275 	/* Put data in mbuf chain. */
5276 	nd->nd_mb->m_next = m;
5277 
5278 	/* Set nd_mb and nd_bpos to end of data. */
5279 	while (m->m_next != NULL)
5280 		m = m->m_next;
5281 	nd->nd_mb = m;
5282 	nfsm_set(nd, m->m_len);
5283 	NFSD_DEBUG(4, "nfsrv_writedsdorpc: lastmb len=%d\n", m->m_len);
5284 
5285 	/* Do a Getattr for the attributes that change upon writing. */
5286 	NFSZERO_ATTRBIT(&attrbits);
5287 	NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_SIZE);
5288 	NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_CHANGE);
5289 	NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_TIMEACCESS);
5290 	NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_TIMEMODIFY);
5291 	NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_SPACEUSED);
5292 	NFSM_BUILD(tl, u_int32_t *, NFSX_UNSIGNED);
5293 	*tl = txdr_unsigned(NFSV4OP_GETATTR);
5294 	(void) nfsrv_putattrbit(nd, &attrbits);
5295 	error = newnfs_request(nd, nmp, NULL, &nmp->nm_sockreq, NULL, p,
5296 	    cred, NFS_PROG, NFS_VER4, NULL, 1, NULL, NULL);
5297 	if (error != 0) {
5298 		free(nd, M_TEMP);
5299 		return (error);
5300 	}
5301 	NFSD_DEBUG(4, "nfsrv_writedsdorpc: aft writerpc=%d\n", nd->nd_repstat);
5302 	/* Get rid of weak cache consistency data for now. */
5303 	if ((nd->nd_flag & (ND_NOMOREDATA | ND_NFSV4 | ND_V4WCCATTR)) ==
5304 	    (ND_NFSV4 | ND_V4WCCATTR)) {
5305 		error = nfsv4_loadattr(nd, NULL, nap, NULL, NULL, 0, NULL, NULL,
5306 		    NULL, NULL, NULL, 0, NULL, NULL, NULL, NULL, NULL);
5307 		NFSD_DEBUG(4, "nfsrv_writedsdorpc: wcc attr=%d\n", error);
5308 		if (error != 0)
5309 			goto nfsmout;
5310 		/*
5311 		 * Get rid of Op# and status for next op.
5312 		 */
5313 		NFSM_DISSECT(tl, uint32_t *, 2 * NFSX_UNSIGNED);
5314 		if (*++tl != 0)
5315 			nd->nd_flag |= ND_NOMOREDATA;
5316 	}
5317 	if (nd->nd_repstat == 0) {
5318 		NFSM_DISSECT(tl, uint32_t *, 2 * NFSX_UNSIGNED + NFSX_VERF);
5319 		retlen = fxdr_unsigned(int, *tl++);
5320 		commit = fxdr_unsigned(int, *tl);
5321 		if (commit != NFSWRITE_FILESYNC)
5322 			error = NFSERR_IO;
5323 		NFSD_DEBUG(4, "nfsrv_writedsdorpc:retlen=%d commit=%d err=%d\n",
5324 		    retlen, commit, error);
5325 	} else
5326 		error = nd->nd_repstat;
5327 	/* We have no use for the Write Verifier since we use FileSync. */
5328 
5329 	/*
5330 	 * Get the Change, Size, Access Time and Modify Time attributes and set
5331 	 * on the Metadata file, so its attributes will be what the file's
5332 	 * would be if it had been written.
5333 	 */
5334 	if (error == 0) {
5335 		NFSM_DISSECT(tl, uint32_t *, 2 * NFSX_UNSIGNED);
5336 		error = nfsv4_loadattr(nd, NULL, nap, NULL, NULL, 0, NULL, NULL,
5337 		    NULL, NULL, NULL, 0, NULL, NULL, NULL, NULL, NULL);
5338 	}
5339 	NFSD_DEBUG(4, "nfsrv_writedsdorpc: aft loadattr=%d\n", error);
5340 nfsmout:
5341 	m_freem(nd->nd_mrep);
5342 	free(nd, M_TEMP);
5343 	NFSD_DEBUG(4, "nfsrv_writedsdorpc error=%d\n", error);
5344 	return (error);
5345 }
5346 
5347 /*
5348  * Start up the thread that will execute nfsrv_writedsdorpc().
5349  */
5350 static void
5351 start_writedsdorpc(void *arg, int pending)
5352 {
5353 	struct nfsrvwritedsdorpc *drpc;
5354 
5355 	drpc = (struct nfsrvwritedsdorpc *)arg;
5356 	drpc->err = nfsrv_writedsdorpc(drpc->nmp, &drpc->fh, drpc->off,
5357 	    drpc->len, NULL, drpc->m, drpc->cred, drpc->p);
5358 	drpc->done = 1;
5359 	NFSD_DEBUG(4, "start_writedsdorpc: err=%d\n", drpc->err);
5360 }
5361 
5362 static int
5363 nfsrv_writedsrpc(fhandle_t *fhp, off_t off, int len, struct ucred *cred,
5364     NFSPROC_T *p, struct vnode *vp, struct nfsmount **nmpp, int mirrorcnt,
5365     struct mbuf **mpp, char *cp, int *failposp)
5366 {
5367 	struct nfsrvwritedsdorpc *drpc, *tdrpc = NULL;
5368 	struct nfsvattr na;
5369 	struct mbuf *m;
5370 	int error, i, offs, ret, timo;
5371 
5372 	NFSD_DEBUG(4, "in nfsrv_writedsrpc\n");
5373 	KASSERT(*mpp != NULL, ("nfsrv_writedsrpc: NULL mbuf chain"));
5374 	drpc = NULL;
5375 	if (mirrorcnt > 1)
5376 		tdrpc = drpc = malloc(sizeof(*drpc) * (mirrorcnt - 1), M_TEMP,
5377 		    M_WAITOK);
5378 
5379 	/* Calculate offset in mbuf chain that data starts. */
5380 	offs = cp - mtod(*mpp, char *);
5381 	NFSD_DEBUG(4, "nfsrv_writedsrpc: mcopy offs=%d len=%d\n", offs, len);
5382 
5383 	/*
5384 	 * Do the write RPC for every DS, using a separate kernel process
5385 	 * for every DS except the last one.
5386 	 */
5387 	error = 0;
5388 	for (i = 0; i < mirrorcnt - 1; i++, tdrpc++) {
5389 		tdrpc->done = 0;
5390 		NFSBCOPY(fhp, &tdrpc->fh, sizeof(*fhp));
5391 		tdrpc->off = off;
5392 		tdrpc->len = len;
5393 		tdrpc->nmp = *nmpp;
5394 		tdrpc->cred = cred;
5395 		tdrpc->p = p;
5396 		tdrpc->inprog = 0;
5397 		tdrpc->err = 0;
5398 		tdrpc->m = m_copym(*mpp, offs, NFSM_RNDUP(len), M_WAITOK);
5399 		ret = EIO;
5400 		if (nfs_pnfsiothreads != 0) {
5401 			ret = nfs_pnfsio(start_writedsdorpc, tdrpc);
5402 			NFSD_DEBUG(4, "nfsrv_writedsrpc: nfs_pnfsio=%d\n",
5403 			    ret);
5404 		}
5405 		if (ret != 0) {
5406 			ret = nfsrv_writedsdorpc(*nmpp, fhp, off, len, NULL,
5407 			    tdrpc->m, cred, p);
5408 			if (nfsds_failerr(ret) && *failposp == -1)
5409 				*failposp = i;
5410 			else if (error == 0 && ret != 0)
5411 				error = ret;
5412 		}
5413 		nmpp++;
5414 		fhp++;
5415 	}
5416 	m = m_copym(*mpp, offs, NFSM_RNDUP(len), M_WAITOK);
5417 	ret = nfsrv_writedsdorpc(*nmpp, fhp, off, len, &na, m, cred, p);
5418 	if (nfsds_failerr(ret) && *failposp == -1 && mirrorcnt > 1)
5419 		*failposp = mirrorcnt - 1;
5420 	else if (error == 0 && ret != 0)
5421 		error = ret;
5422 	if (error == 0)
5423 		error = nfsrv_setextattr(vp, &na, p);
5424 	NFSD_DEBUG(4, "nfsrv_writedsrpc: aft setextat=%d\n", error);
5425 	tdrpc = drpc;
5426 	timo = hz / 50;		/* Wait for 20msec. */
5427 	if (timo < 1)
5428 		timo = 1;
5429 	for (i = 0; i < mirrorcnt - 1; i++, tdrpc++) {
5430 		/* Wait for RPCs on separate threads to complete. */
5431 		while (tdrpc->inprog != 0 && tdrpc->done == 0)
5432 			tsleep(&tdrpc->tsk, PVFS, "srvwrds", timo);
5433 		if (nfsds_failerr(tdrpc->err) && *failposp == -1)
5434 			*failposp = i;
5435 		else if (error == 0 && tdrpc->err != 0)
5436 			error = tdrpc->err;
5437 	}
5438 	free(drpc, M_TEMP);
5439 	return (error);
5440 }
5441 
5442 /*
5443  * Do a allocate RPC on a DS data file, using this structure for the arguments,
5444  * so that this function can be executed by a separate kernel process.
5445  */
5446 struct nfsrvallocatedsdorpc {
5447 	int			done;
5448 	int			inprog;
5449 	struct task		tsk;
5450 	fhandle_t		fh;
5451 	off_t			off;
5452 	off_t			len;
5453 	struct nfsmount		*nmp;
5454 	struct ucred		*cred;
5455 	NFSPROC_T		*p;
5456 	int			err;
5457 };
5458 
5459 static int
5460 nfsrv_allocatedsdorpc(struct nfsmount *nmp, fhandle_t *fhp, off_t off,
5461     off_t len, struct nfsvattr *nap, struct ucred *cred, NFSPROC_T *p)
5462 {
5463 	uint32_t *tl;
5464 	struct nfsrv_descript *nd;
5465 	nfsattrbit_t attrbits;
5466 	nfsv4stateid_t st;
5467 	int error;
5468 
5469 	nd = malloc(sizeof(*nd), M_TEMP, M_WAITOK | M_ZERO);
5470 	nfscl_reqstart(nd, NFSPROC_ALLOCATE, nmp, (u_int8_t *)fhp,
5471 	    sizeof(fhandle_t), NULL, NULL, 0, 0);
5472 
5473 	/*
5474 	 * Use a stateid where other is an alternating 01010 pattern and
5475 	 * seqid is 0xffffffff.  This value is not defined as special by
5476 	 * the RFC and is used by the FreeBSD NFS server to indicate an
5477 	 * MDS->DS proxy operation.
5478 	 */
5479 	st.other[0] = 0x55555555;
5480 	st.other[1] = 0x55555555;
5481 	st.other[2] = 0x55555555;
5482 	st.seqid = 0xffffffff;
5483 	nfsm_stateidtom(nd, &st, NFSSTATEID_PUTSTATEID);
5484 	NFSM_BUILD(tl, uint32_t *, 2 * NFSX_HYPER + NFSX_UNSIGNED);
5485 	txdr_hyper(off, tl); tl += 2;
5486 	txdr_hyper(len, tl); tl += 2;
5487 	NFSD_DEBUG(4, "nfsrv_allocatedsdorpc: len=%jd\n", (intmax_t)len);
5488 
5489 	*tl = txdr_unsigned(NFSV4OP_GETATTR);
5490 	NFSGETATTR_ATTRBIT(&attrbits);
5491 	nfsrv_putattrbit(nd, &attrbits);
5492 	error = newnfs_request(nd, nmp, NULL, &nmp->nm_sockreq, NULL, p,
5493 	    cred, NFS_PROG, NFS_VER4, NULL, 1, NULL, NULL);
5494 	if (error != 0) {
5495 		free(nd, M_TEMP);
5496 		return (error);
5497 	}
5498 	NFSD_DEBUG(4, "nfsrv_allocatedsdorpc: aft allocaterpc=%d\n",
5499 	    nd->nd_repstat);
5500 	if (nd->nd_repstat == 0) {
5501 		NFSM_DISSECT(tl, uint32_t *, 2 * NFSX_UNSIGNED);
5502 		error = nfsv4_loadattr(nd, NULL, nap, NULL, NULL, 0, NULL, NULL,
5503 		    NULL, NULL, NULL, 0, NULL, NULL, NULL, NULL, NULL);
5504 	} else
5505 		error = nd->nd_repstat;
5506 	NFSD_DEBUG(4, "nfsrv_allocatedsdorpc: aft loadattr=%d\n", error);
5507 nfsmout:
5508 	m_freem(nd->nd_mrep);
5509 	free(nd, M_TEMP);
5510 	NFSD_DEBUG(4, "nfsrv_allocatedsdorpc error=%d\n", error);
5511 	return (error);
5512 }
5513 
5514 /*
5515  * Start up the thread that will execute nfsrv_allocatedsdorpc().
5516  */
5517 static void
5518 start_allocatedsdorpc(void *arg, int pending)
5519 {
5520 	struct nfsrvallocatedsdorpc *drpc;
5521 
5522 	drpc = (struct nfsrvallocatedsdorpc *)arg;
5523 	drpc->err = nfsrv_allocatedsdorpc(drpc->nmp, &drpc->fh, drpc->off,
5524 	    drpc->len, NULL, drpc->cred, drpc->p);
5525 	drpc->done = 1;
5526 	NFSD_DEBUG(4, "start_allocatedsdorpc: err=%d\n", drpc->err);
5527 }
5528 
5529 static int
5530 nfsrv_allocatedsrpc(fhandle_t *fhp, off_t off, off_t len, struct ucred *cred,
5531     NFSPROC_T *p, struct vnode *vp, struct nfsmount **nmpp, int mirrorcnt,
5532     int *failposp)
5533 {
5534 	struct nfsrvallocatedsdorpc *drpc, *tdrpc = NULL;
5535 	struct nfsvattr na;
5536 	int error, i, ret, timo;
5537 
5538 	NFSD_DEBUG(4, "in nfsrv_allocatedsrpc\n");
5539 	drpc = NULL;
5540 	if (mirrorcnt > 1)
5541 		tdrpc = drpc = malloc(sizeof(*drpc) * (mirrorcnt - 1), M_TEMP,
5542 		    M_WAITOK);
5543 
5544 	/*
5545 	 * Do the allocate RPC for every DS, using a separate kernel process
5546 	 * for every DS except the last one.
5547 	 */
5548 	error = 0;
5549 	for (i = 0; i < mirrorcnt - 1; i++, tdrpc++) {
5550 		tdrpc->done = 0;
5551 		NFSBCOPY(fhp, &tdrpc->fh, sizeof(*fhp));
5552 		tdrpc->off = off;
5553 		tdrpc->len = len;
5554 		tdrpc->nmp = *nmpp;
5555 		tdrpc->cred = cred;
5556 		tdrpc->p = p;
5557 		tdrpc->inprog = 0;
5558 		tdrpc->err = 0;
5559 		ret = EIO;
5560 		if (nfs_pnfsiothreads != 0) {
5561 			ret = nfs_pnfsio(start_allocatedsdorpc, tdrpc);
5562 			NFSD_DEBUG(4, "nfsrv_allocatedsrpc: nfs_pnfsio=%d\n",
5563 			    ret);
5564 		}
5565 		if (ret != 0) {
5566 			ret = nfsrv_allocatedsdorpc(*nmpp, fhp, off, len, NULL,
5567 			    cred, p);
5568 			if (nfsds_failerr(ret) && *failposp == -1)
5569 				*failposp = i;
5570 			else if (error == 0 && ret != 0)
5571 				error = ret;
5572 		}
5573 		nmpp++;
5574 		fhp++;
5575 	}
5576 	ret = nfsrv_allocatedsdorpc(*nmpp, fhp, off, len, &na, cred, p);
5577 	if (nfsds_failerr(ret) && *failposp == -1 && mirrorcnt > 1)
5578 		*failposp = mirrorcnt - 1;
5579 	else if (error == 0 && ret != 0)
5580 		error = ret;
5581 	if (error == 0)
5582 		error = nfsrv_setextattr(vp, &na, p);
5583 	NFSD_DEBUG(4, "nfsrv_allocatedsrpc: aft setextat=%d\n", error);
5584 	tdrpc = drpc;
5585 	timo = hz / 50;		/* Wait for 20msec. */
5586 	if (timo < 1)
5587 		timo = 1;
5588 	for (i = 0; i < mirrorcnt - 1; i++, tdrpc++) {
5589 		/* Wait for RPCs on separate threads to complete. */
5590 		while (tdrpc->inprog != 0 && tdrpc->done == 0)
5591 			tsleep(&tdrpc->tsk, PVFS, "srvalds", timo);
5592 		if (nfsds_failerr(tdrpc->err) && *failposp == -1)
5593 			*failposp = i;
5594 		else if (error == 0 && tdrpc->err != 0)
5595 			error = tdrpc->err;
5596 	}
5597 	free(drpc, M_TEMP);
5598 	return (error);
5599 }
5600 
5601 static int
5602 nfsrv_setattrdsdorpc(fhandle_t *fhp, struct ucred *cred, NFSPROC_T *p,
5603     struct vnode *vp, struct nfsmount *nmp, struct nfsvattr *nap,
5604     struct nfsvattr *dsnap)
5605 {
5606 	uint32_t *tl;
5607 	struct nfsrv_descript *nd;
5608 	nfsv4stateid_t st;
5609 	nfsattrbit_t attrbits;
5610 	int error;
5611 
5612 	NFSD_DEBUG(4, "in nfsrv_setattrdsdorpc\n");
5613 	nd = malloc(sizeof(*nd), M_TEMP, M_WAITOK | M_ZERO);
5614 	/*
5615 	 * Use a stateid where other is an alternating 01010 pattern and
5616 	 * seqid is 0xffffffff.  This value is not defined as special by
5617 	 * the RFC and is used by the FreeBSD NFS server to indicate an
5618 	 * MDS->DS proxy operation.
5619 	 */
5620 	st.other[0] = 0x55555555;
5621 	st.other[1] = 0x55555555;
5622 	st.other[2] = 0x55555555;
5623 	st.seqid = 0xffffffff;
5624 	nfscl_reqstart(nd, NFSPROC_SETATTR, nmp, (u_int8_t *)fhp, sizeof(*fhp),
5625 	    NULL, NULL, 0, 0);
5626 	nfsm_stateidtom(nd, &st, NFSSTATEID_PUTSTATEID);
5627 	nfscl_fillsattr(nd, &nap->na_vattr, vp, NFSSATTR_FULL, 0);
5628 
5629 	/* Do a Getattr for the attributes that change due to writing. */
5630 	NFSZERO_ATTRBIT(&attrbits);
5631 	NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_SIZE);
5632 	NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_CHANGE);
5633 	NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_TIMEACCESS);
5634 	NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_TIMEMODIFY);
5635 	NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_SPACEUSED);
5636 	NFSM_BUILD(tl, u_int32_t *, NFSX_UNSIGNED);
5637 	*tl = txdr_unsigned(NFSV4OP_GETATTR);
5638 	(void) nfsrv_putattrbit(nd, &attrbits);
5639 	error = newnfs_request(nd, nmp, NULL, &nmp->nm_sockreq, NULL, p, cred,
5640 	    NFS_PROG, NFS_VER4, NULL, 1, NULL, NULL);
5641 	if (error != 0) {
5642 		free(nd, M_TEMP);
5643 		return (error);
5644 	}
5645 	NFSD_DEBUG(4, "nfsrv_setattrdsdorpc: aft setattrrpc=%d\n",
5646 	    nd->nd_repstat);
5647 	/* Get rid of weak cache consistency data for now. */
5648 	if ((nd->nd_flag & (ND_NOMOREDATA | ND_NFSV4 | ND_V4WCCATTR)) ==
5649 	    (ND_NFSV4 | ND_V4WCCATTR)) {
5650 		error = nfsv4_loadattr(nd, NULL, dsnap, NULL, NULL, 0, NULL,
5651 		    NULL, NULL, NULL, NULL, 0, NULL, NULL, NULL, NULL, NULL);
5652 		NFSD_DEBUG(4, "nfsrv_setattrdsdorpc: wcc attr=%d\n", error);
5653 		if (error != 0)
5654 			goto nfsmout;
5655 		/*
5656 		 * Get rid of Op# and status for next op.
5657 		 */
5658 		NFSM_DISSECT(tl, uint32_t *, 2 * NFSX_UNSIGNED);
5659 		if (*++tl != 0)
5660 			nd->nd_flag |= ND_NOMOREDATA;
5661 	}
5662 	error = nfsrv_getattrbits(nd, &attrbits, NULL, NULL);
5663 	if (error != 0)
5664 		goto nfsmout;
5665 	if (nd->nd_repstat != 0)
5666 		error = nd->nd_repstat;
5667 	/*
5668 	 * Get the Change, Size, Access Time and Modify Time attributes and set
5669 	 * on the Metadata file, so its attributes will be what the file's
5670 	 * would be if it had been written.
5671 	 */
5672 	if (error == 0) {
5673 		NFSM_DISSECT(tl, uint32_t *, 2 * NFSX_UNSIGNED);
5674 		error = nfsv4_loadattr(nd, NULL, dsnap, NULL, NULL, 0, NULL,
5675 		    NULL, NULL, NULL, NULL, 0, NULL, NULL, NULL, NULL, NULL);
5676 	}
5677 	NFSD_DEBUG(4, "nfsrv_setattrdsdorpc: aft setattr loadattr=%d\n", error);
5678 nfsmout:
5679 	m_freem(nd->nd_mrep);
5680 	free(nd, M_TEMP);
5681 	NFSD_DEBUG(4, "nfsrv_setattrdsdorpc error=%d\n", error);
5682 	return (error);
5683 }
5684 
5685 struct nfsrvsetattrdsdorpc {
5686 	int			done;
5687 	int			inprog;
5688 	struct task		tsk;
5689 	fhandle_t		fh;
5690 	struct nfsmount		*nmp;
5691 	struct vnode		*vp;
5692 	struct ucred		*cred;
5693 	NFSPROC_T		*p;
5694 	struct nfsvattr		na;
5695 	struct nfsvattr		dsna;
5696 	int			err;
5697 };
5698 
5699 /*
5700  * Start up the thread that will execute nfsrv_setattrdsdorpc().
5701  */
5702 static void
5703 start_setattrdsdorpc(void *arg, int pending)
5704 {
5705 	struct nfsrvsetattrdsdorpc *drpc;
5706 
5707 	drpc = (struct nfsrvsetattrdsdorpc *)arg;
5708 	drpc->err = nfsrv_setattrdsdorpc(&drpc->fh, drpc->cred, drpc->p,
5709 	    drpc->vp, drpc->nmp, &drpc->na, &drpc->dsna);
5710 	drpc->done = 1;
5711 }
5712 
5713 static int
5714 nfsrv_setattrdsrpc(fhandle_t *fhp, struct ucred *cred, NFSPROC_T *p,
5715     struct vnode *vp, struct nfsmount **nmpp, int mirrorcnt,
5716     struct nfsvattr *nap, int *failposp)
5717 {
5718 	struct nfsrvsetattrdsdorpc *drpc, *tdrpc = NULL;
5719 	struct nfsvattr na;
5720 	int error, i, ret, timo;
5721 
5722 	NFSD_DEBUG(4, "in nfsrv_setattrdsrpc\n");
5723 	drpc = NULL;
5724 	if (mirrorcnt > 1)
5725 		tdrpc = drpc = malloc(sizeof(*drpc) * (mirrorcnt - 1), M_TEMP,
5726 		    M_WAITOK);
5727 
5728 	/*
5729 	 * Do the setattr RPC for every DS, using a separate kernel process
5730 	 * for every DS except the last one.
5731 	 */
5732 	error = 0;
5733 	for (i = 0; i < mirrorcnt - 1; i++, tdrpc++) {
5734 		tdrpc->done = 0;
5735 		tdrpc->inprog = 0;
5736 		NFSBCOPY(fhp, &tdrpc->fh, sizeof(*fhp));
5737 		tdrpc->nmp = *nmpp;
5738 		tdrpc->vp = vp;
5739 		tdrpc->cred = cred;
5740 		tdrpc->p = p;
5741 		tdrpc->na = *nap;
5742 		tdrpc->err = 0;
5743 		ret = EIO;
5744 		if (nfs_pnfsiothreads != 0) {
5745 			ret = nfs_pnfsio(start_setattrdsdorpc, tdrpc);
5746 			NFSD_DEBUG(4, "nfsrv_setattrdsrpc: nfs_pnfsio=%d\n",
5747 			    ret);
5748 		}
5749 		if (ret != 0) {
5750 			ret = nfsrv_setattrdsdorpc(fhp, cred, p, vp, *nmpp, nap,
5751 			    &na);
5752 			if (nfsds_failerr(ret) && *failposp == -1)
5753 				*failposp = i;
5754 			else if (error == 0 && ret != 0)
5755 				error = ret;
5756 		}
5757 		nmpp++;
5758 		fhp++;
5759 	}
5760 	ret = nfsrv_setattrdsdorpc(fhp, cred, p, vp, *nmpp, nap, &na);
5761 	if (nfsds_failerr(ret) && *failposp == -1 && mirrorcnt > 1)
5762 		*failposp = mirrorcnt - 1;
5763 	else if (error == 0 && ret != 0)
5764 		error = ret;
5765 	if (error == 0)
5766 		error = nfsrv_setextattr(vp, &na, p);
5767 	NFSD_DEBUG(4, "nfsrv_setattrdsrpc: aft setextat=%d\n", error);
5768 	tdrpc = drpc;
5769 	timo = hz / 50;		/* Wait for 20msec. */
5770 	if (timo < 1)
5771 		timo = 1;
5772 	for (i = 0; i < mirrorcnt - 1; i++, tdrpc++) {
5773 		/* Wait for RPCs on separate threads to complete. */
5774 		while (tdrpc->inprog != 0 && tdrpc->done == 0)
5775 			tsleep(&tdrpc->tsk, PVFS, "srvsads", timo);
5776 		if (nfsds_failerr(tdrpc->err) && *failposp == -1)
5777 			*failposp = i;
5778 		else if (error == 0 && tdrpc->err != 0)
5779 			error = tdrpc->err;
5780 	}
5781 	free(drpc, M_TEMP);
5782 	return (error);
5783 }
5784 
5785 /*
5786  * Do a Setattr of an NFSv4 ACL on the DS file.
5787  */
5788 static int
5789 nfsrv_setacldsdorpc(fhandle_t *fhp, struct ucred *cred, NFSPROC_T *p,
5790     struct vnode *vp, struct nfsmount *nmp, struct acl *aclp)
5791 {
5792 	struct nfsrv_descript *nd;
5793 	nfsv4stateid_t st;
5794 	nfsattrbit_t attrbits;
5795 	int error;
5796 
5797 	NFSD_DEBUG(4, "in nfsrv_setacldsdorpc\n");
5798 	nd = malloc(sizeof(*nd), M_TEMP, M_WAITOK | M_ZERO);
5799 	/*
5800 	 * Use a stateid where other is an alternating 01010 pattern and
5801 	 * seqid is 0xffffffff.  This value is not defined as special by
5802 	 * the RFC and is used by the FreeBSD NFS server to indicate an
5803 	 * MDS->DS proxy operation.
5804 	 */
5805 	st.other[0] = 0x55555555;
5806 	st.other[1] = 0x55555555;
5807 	st.other[2] = 0x55555555;
5808 	st.seqid = 0xffffffff;
5809 	nfscl_reqstart(nd, NFSPROC_SETACL, nmp, (u_int8_t *)fhp, sizeof(*fhp),
5810 	    NULL, NULL, 0, 0);
5811 	nfsm_stateidtom(nd, &st, NFSSTATEID_PUTSTATEID);
5812 	NFSZERO_ATTRBIT(&attrbits);
5813 	NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_ACL);
5814 	/*
5815 	 * The "vp" argument to nfsv4_fillattr() is only used for vnode_type(),
5816 	 * so passing in the metadata "vp" will be ok, since it is of
5817 	 * the same type (VREG).
5818 	 */
5819 	nfsv4_fillattr(nd, NULL, vp, aclp, NULL, NULL, 0, &attrbits, NULL,
5820 	    NULL, 0, 0, 0, 0, 0, NULL);
5821 	error = newnfs_request(nd, nmp, NULL, &nmp->nm_sockreq, NULL, p, cred,
5822 	    NFS_PROG, NFS_VER4, NULL, 1, NULL, NULL);
5823 	if (error != 0) {
5824 		free(nd, M_TEMP);
5825 		return (error);
5826 	}
5827 	NFSD_DEBUG(4, "nfsrv_setacldsdorpc: aft setaclrpc=%d\n",
5828 	    nd->nd_repstat);
5829 	error = nd->nd_repstat;
5830 	m_freem(nd->nd_mrep);
5831 	free(nd, M_TEMP);
5832 	return (error);
5833 }
5834 
5835 struct nfsrvsetacldsdorpc {
5836 	int			done;
5837 	int			inprog;
5838 	struct task		tsk;
5839 	fhandle_t		fh;
5840 	struct nfsmount		*nmp;
5841 	struct vnode		*vp;
5842 	struct ucred		*cred;
5843 	NFSPROC_T		*p;
5844 	struct acl		*aclp;
5845 	int			err;
5846 };
5847 
5848 /*
5849  * Start up the thread that will execute nfsrv_setacldsdorpc().
5850  */
5851 static void
5852 start_setacldsdorpc(void *arg, int pending)
5853 {
5854 	struct nfsrvsetacldsdorpc *drpc;
5855 
5856 	drpc = (struct nfsrvsetacldsdorpc *)arg;
5857 	drpc->err = nfsrv_setacldsdorpc(&drpc->fh, drpc->cred, drpc->p,
5858 	    drpc->vp, drpc->nmp, drpc->aclp);
5859 	drpc->done = 1;
5860 }
5861 
5862 static int
5863 nfsrv_setacldsrpc(fhandle_t *fhp, struct ucred *cred, NFSPROC_T *p,
5864     struct vnode *vp, struct nfsmount **nmpp, int mirrorcnt, struct acl *aclp,
5865     int *failposp)
5866 {
5867 	struct nfsrvsetacldsdorpc *drpc, *tdrpc = NULL;
5868 	int error, i, ret, timo;
5869 
5870 	NFSD_DEBUG(4, "in nfsrv_setacldsrpc\n");
5871 	drpc = NULL;
5872 	if (mirrorcnt > 1)
5873 		tdrpc = drpc = malloc(sizeof(*drpc) * (mirrorcnt - 1), M_TEMP,
5874 		    M_WAITOK);
5875 
5876 	/*
5877 	 * Do the setattr RPC for every DS, using a separate kernel process
5878 	 * for every DS except the last one.
5879 	 */
5880 	error = 0;
5881 	for (i = 0; i < mirrorcnt - 1; i++, tdrpc++) {
5882 		tdrpc->done = 0;
5883 		tdrpc->inprog = 0;
5884 		NFSBCOPY(fhp, &tdrpc->fh, sizeof(*fhp));
5885 		tdrpc->nmp = *nmpp;
5886 		tdrpc->vp = vp;
5887 		tdrpc->cred = cred;
5888 		tdrpc->p = p;
5889 		tdrpc->aclp = aclp;
5890 		tdrpc->err = 0;
5891 		ret = EIO;
5892 		if (nfs_pnfsiothreads != 0) {
5893 			ret = nfs_pnfsio(start_setacldsdorpc, tdrpc);
5894 			NFSD_DEBUG(4, "nfsrv_setacldsrpc: nfs_pnfsio=%d\n",
5895 			    ret);
5896 		}
5897 		if (ret != 0) {
5898 			ret = nfsrv_setacldsdorpc(fhp, cred, p, vp, *nmpp,
5899 			    aclp);
5900 			if (nfsds_failerr(ret) && *failposp == -1)
5901 				*failposp = i;
5902 			else if (error == 0 && ret != 0)
5903 				error = ret;
5904 		}
5905 		nmpp++;
5906 		fhp++;
5907 	}
5908 	ret = nfsrv_setacldsdorpc(fhp, cred, p, vp, *nmpp, aclp);
5909 	if (nfsds_failerr(ret) && *failposp == -1 && mirrorcnt > 1)
5910 		*failposp = mirrorcnt - 1;
5911 	else if (error == 0 && ret != 0)
5912 		error = ret;
5913 	NFSD_DEBUG(4, "nfsrv_setacldsrpc: aft setextat=%d\n", error);
5914 	tdrpc = drpc;
5915 	timo = hz / 50;		/* Wait for 20msec. */
5916 	if (timo < 1)
5917 		timo = 1;
5918 	for (i = 0; i < mirrorcnt - 1; i++, tdrpc++) {
5919 		/* Wait for RPCs on separate threads to complete. */
5920 		while (tdrpc->inprog != 0 && tdrpc->done == 0)
5921 			tsleep(&tdrpc->tsk, PVFS, "srvacds", timo);
5922 		if (nfsds_failerr(tdrpc->err) && *failposp == -1)
5923 			*failposp = i;
5924 		else if (error == 0 && tdrpc->err != 0)
5925 			error = tdrpc->err;
5926 	}
5927 	free(drpc, M_TEMP);
5928 	return (error);
5929 }
5930 
5931 /*
5932  * Getattr call to the DS for the attributes that change due to writing.
5933  */
5934 static int
5935 nfsrv_getattrdsrpc(fhandle_t *fhp, struct ucred *cred, NFSPROC_T *p,
5936     struct vnode *vp, struct nfsmount *nmp, struct nfsvattr *nap)
5937 {
5938 	struct nfsrv_descript *nd;
5939 	int error;
5940 	nfsattrbit_t attrbits;
5941 
5942 	NFSD_DEBUG(4, "in nfsrv_getattrdsrpc\n");
5943 	nd = malloc(sizeof(*nd), M_TEMP, M_WAITOK | M_ZERO);
5944 	nfscl_reqstart(nd, NFSPROC_GETATTR, nmp, (u_int8_t *)fhp,
5945 	    sizeof(fhandle_t), NULL, NULL, 0, 0);
5946 	NFSZERO_ATTRBIT(&attrbits);
5947 	NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_SIZE);
5948 	NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_CHANGE);
5949 	NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_TIMEACCESS);
5950 	NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_TIMEMODIFY);
5951 	NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_SPACEUSED);
5952 	(void) nfsrv_putattrbit(nd, &attrbits);
5953 	error = newnfs_request(nd, nmp, NULL, &nmp->nm_sockreq, NULL, p, cred,
5954 	    NFS_PROG, NFS_VER4, NULL, 1, NULL, NULL);
5955 	if (error != 0) {
5956 		free(nd, M_TEMP);
5957 		return (error);
5958 	}
5959 	NFSD_DEBUG(4, "nfsrv_getattrdsrpc: aft getattrrpc=%d\n",
5960 	    nd->nd_repstat);
5961 	if (nd->nd_repstat == 0) {
5962 		error = nfsv4_loadattr(nd, NULL, nap, NULL, NULL, 0,
5963 		    NULL, NULL, NULL, NULL, NULL, 0, NULL, NULL, NULL,
5964 		    NULL, NULL);
5965 		/*
5966 		 * We can only save the updated values in the extended
5967 		 * attribute if the vp is exclusively locked.
5968 		 * This should happen when any of the following operations
5969 		 * occur on the vnode:
5970 		 *    Close, Delegreturn, LayoutCommit, LayoutReturn
5971 		 * As such, the updated extended attribute should get saved
5972 		 * before nfsrv_checkdsattr() returns 0 and allows the cached
5973 		 * attributes to be returned without calling this function.
5974 		 */
5975 		if (error == 0 && VOP_ISLOCKED(vp) == LK_EXCLUSIVE) {
5976 			error = nfsrv_setextattr(vp, nap, p);
5977 			NFSD_DEBUG(4, "nfsrv_getattrdsrpc: aft setextat=%d\n",
5978 			    error);
5979 		}
5980 	} else
5981 		error = nd->nd_repstat;
5982 	m_freem(nd->nd_mrep);
5983 	free(nd, M_TEMP);
5984 	NFSD_DEBUG(4, "nfsrv_getattrdsrpc error=%d\n", error);
5985 	return (error);
5986 }
5987 
5988 /*
5989  * Seek call to a DS.
5990  */
5991 static int
5992 nfsrv_seekdsrpc(fhandle_t *fhp, off_t *offp, int content, bool *eofp,
5993     struct ucred *cred, NFSPROC_T *p, struct nfsmount *nmp)
5994 {
5995 	uint32_t *tl;
5996 	struct nfsrv_descript *nd;
5997 	nfsv4stateid_t st;
5998 	int error;
5999 
6000 	NFSD_DEBUG(4, "in nfsrv_seekdsrpc\n");
6001 	/*
6002 	 * Use a stateid where other is an alternating 01010 pattern and
6003 	 * seqid is 0xffffffff.  This value is not defined as special by
6004 	 * the RFC and is used by the FreeBSD NFS server to indicate an
6005 	 * MDS->DS proxy operation.
6006 	 */
6007 	st.other[0] = 0x55555555;
6008 	st.other[1] = 0x55555555;
6009 	st.other[2] = 0x55555555;
6010 	st.seqid = 0xffffffff;
6011 	nd = malloc(sizeof(*nd), M_TEMP, M_WAITOK | M_ZERO);
6012 	nfscl_reqstart(nd, NFSPROC_SEEKDS, nmp, (u_int8_t *)fhp,
6013 	    sizeof(fhandle_t), NULL, NULL, 0, 0);
6014 	nfsm_stateidtom(nd, &st, NFSSTATEID_PUTSTATEID);
6015 	NFSM_BUILD(tl, uint32_t *, NFSX_HYPER + NFSX_UNSIGNED);
6016 	txdr_hyper(*offp, tl); tl += 2;
6017 	*tl = txdr_unsigned(content);
6018 	error = newnfs_request(nd, nmp, NULL, &nmp->nm_sockreq, NULL, p, cred,
6019 	    NFS_PROG, NFS_VER4, NULL, 1, NULL, NULL);
6020 	if (error != 0) {
6021 		free(nd, M_TEMP);
6022 		return (error);
6023 	}
6024 	NFSD_DEBUG(4, "nfsrv_seekdsrpc: aft seekrpc=%d\n", nd->nd_repstat);
6025 	if (nd->nd_repstat == 0) {
6026 		NFSM_DISSECT(tl, uint32_t *, NFSX_UNSIGNED + NFSX_HYPER);
6027 		if (*tl++ == newnfs_true)
6028 			*eofp = true;
6029 		else
6030 			*eofp = false;
6031 		*offp = fxdr_hyper(tl);
6032 	} else
6033 		error = nd->nd_repstat;
6034 nfsmout:
6035 	m_freem(nd->nd_mrep);
6036 	free(nd, M_TEMP);
6037 	NFSD_DEBUG(4, "nfsrv_seekdsrpc error=%d\n", error);
6038 	return (error);
6039 }
6040 
6041 /*
6042  * Get the device id and file handle for a DS file.
6043  */
6044 int
6045 nfsrv_dsgetdevandfh(struct vnode *vp, NFSPROC_T *p, int *mirrorcntp,
6046     fhandle_t *fhp, char *devid)
6047 {
6048 	int buflen, error;
6049 	char *buf;
6050 
6051 	buflen = 1024;
6052 	buf = malloc(buflen, M_TEMP, M_WAITOK);
6053 	error = nfsrv_dsgetsockmnt(vp, 0, buf, &buflen, mirrorcntp, p, NULL,
6054 	    fhp, devid, NULL, NULL, NULL, NULL, NULL, NULL);
6055 	free(buf, M_TEMP);
6056 	return (error);
6057 }
6058 
6059 /*
6060  * Do a Lookup against the DS for the filename.
6061  */
6062 static int
6063 nfsrv_pnfslookupds(struct vnode *vp, struct vnode *dvp, struct pnfsdsfile *pf,
6064     struct vnode **nvpp, NFSPROC_T *p)
6065 {
6066 	struct nameidata named;
6067 	struct ucred *tcred;
6068 	char *bufp;
6069 	u_long *hashp;
6070 	struct vnode *nvp;
6071 	int error;
6072 
6073 	tcred = newnfs_getcred();
6074 	named.ni_cnd.cn_nameiop = LOOKUP;
6075 	named.ni_cnd.cn_lkflags = LK_SHARED | LK_RETRY;
6076 	named.ni_cnd.cn_cred = tcred;
6077 	named.ni_cnd.cn_thread = p;
6078 	named.ni_cnd.cn_flags = ISLASTCN | LOCKPARENT | LOCKLEAF | SAVENAME;
6079 	nfsvno_setpathbuf(&named, &bufp, &hashp);
6080 	named.ni_cnd.cn_nameptr = bufp;
6081 	named.ni_cnd.cn_namelen = strlen(pf->dsf_filename);
6082 	strlcpy(bufp, pf->dsf_filename, NAME_MAX);
6083 	NFSD_DEBUG(4, "nfsrv_pnfslookupds: filename=%s\n", bufp);
6084 	error = VOP_LOOKUP(dvp, &nvp, &named.ni_cnd);
6085 	NFSD_DEBUG(4, "nfsrv_pnfslookupds: aft LOOKUP=%d\n", error);
6086 	NFSFREECRED(tcred);
6087 	nfsvno_relpathbuf(&named);
6088 	if (error == 0)
6089 		*nvpp = nvp;
6090 	NFSD_DEBUG(4, "eo nfsrv_pnfslookupds=%d\n", error);
6091 	return (error);
6092 }
6093 
6094 /*
6095  * Set the file handle to the correct one.
6096  */
6097 static void
6098 nfsrv_pnfssetfh(struct vnode *vp, struct pnfsdsfile *pf, char *devid,
6099     char *fnamep, struct vnode *nvp, NFSPROC_T *p)
6100 {
6101 	struct nfsnode *np;
6102 	int ret = 0;
6103 
6104 	np = VTONFS(nvp);
6105 	NFSBCOPY(np->n_fhp->nfh_fh, &pf->dsf_fh, NFSX_MYFH);
6106 	/*
6107 	 * We can only do a vn_set_extattr() if the vnode is exclusively
6108 	 * locked and vn_start_write() has been done.  If devid != NULL or
6109 	 * fnamep != NULL or the vnode is shared locked, vn_start_write()
6110 	 * may not have been done.
6111 	 * If not done now, it will be done on a future call.
6112 	 */
6113 	if (devid == NULL && fnamep == NULL && NFSVOPISLOCKED(vp) ==
6114 	    LK_EXCLUSIVE)
6115 		ret = vn_extattr_set(vp, IO_NODELOCKED,
6116 		    EXTATTR_NAMESPACE_SYSTEM, "pnfsd.dsfile", sizeof(*pf),
6117 		    (char *)pf, p);
6118 	NFSD_DEBUG(4, "eo nfsrv_pnfssetfh=%d\n", ret);
6119 }
6120 
6121 /*
6122  * Cause RPCs waiting on "nmp" to fail.  This is called for a DS mount point
6123  * when the DS has failed.
6124  */
6125 void
6126 nfsrv_killrpcs(struct nfsmount *nmp)
6127 {
6128 
6129 	/*
6130 	 * Call newnfs_nmcancelreqs() to cause
6131 	 * any RPCs in progress on the mount point to
6132 	 * fail.
6133 	 * This will cause any process waiting for an
6134 	 * RPC to complete while holding a vnode lock
6135 	 * on the mounted-on vnode (such as "df" or
6136 	 * a non-forced "umount") to fail.
6137 	 * This will unlock the mounted-on vnode so
6138 	 * a forced dismount can succeed.
6139 	 * The NFSMNTP_CANCELRPCS flag should be set when this function is
6140 	 * called.
6141 	 */
6142 	newnfs_nmcancelreqs(nmp);
6143 }
6144 
6145 /*
6146  * Sum up the statfs info for each of the DSs, so that the client will
6147  * receive the total for all DSs.
6148  */
6149 static int
6150 nfsrv_pnfsstatfs(struct statfs *sf, struct mount *mp)
6151 {
6152 	struct statfs *tsf;
6153 	struct nfsdevice *ds;
6154 	struct vnode **dvpp, **tdvpp, *dvp;
6155 	uint64_t tot;
6156 	int cnt, error = 0, i;
6157 
6158 	if (nfsrv_devidcnt <= 0)
6159 		return (ENXIO);
6160 	dvpp = mallocarray(nfsrv_devidcnt, sizeof(*dvpp), M_TEMP, M_WAITOK);
6161 	tsf = malloc(sizeof(*tsf), M_TEMP, M_WAITOK);
6162 
6163 	/* Get an array of the dvps for the DSs. */
6164 	tdvpp = dvpp;
6165 	i = 0;
6166 	NFSDDSLOCK();
6167 	/* First, search for matches for same file system. */
6168 	TAILQ_FOREACH(ds, &nfsrv_devidhead, nfsdev_list) {
6169 		if (ds->nfsdev_nmp != NULL && ds->nfsdev_mdsisset != 0 &&
6170 		    fsidcmp(&ds->nfsdev_mdsfsid, &mp->mnt_stat.f_fsid) == 0) {
6171 			if (++i > nfsrv_devidcnt)
6172 				break;
6173 			*tdvpp++ = ds->nfsdev_dvp;
6174 		}
6175 	}
6176 	/*
6177 	 * If no matches for same file system, total all servers not assigned
6178 	 * to a file system.
6179 	 */
6180 	if (i == 0) {
6181 		TAILQ_FOREACH(ds, &nfsrv_devidhead, nfsdev_list) {
6182 			if (ds->nfsdev_nmp != NULL &&
6183 			    ds->nfsdev_mdsisset == 0) {
6184 				if (++i > nfsrv_devidcnt)
6185 					break;
6186 				*tdvpp++ = ds->nfsdev_dvp;
6187 			}
6188 		}
6189 	}
6190 	NFSDDSUNLOCK();
6191 	cnt = i;
6192 
6193 	/* Do a VFS_STATFS() for each of the DSs and sum them up. */
6194 	tdvpp = dvpp;
6195 	for (i = 0; i < cnt && error == 0; i++) {
6196 		dvp = *tdvpp++;
6197 		error = VFS_STATFS(dvp->v_mount, tsf);
6198 		if (error == 0) {
6199 			if (sf->f_bsize == 0) {
6200 				if (tsf->f_bsize > 0)
6201 					sf->f_bsize = tsf->f_bsize;
6202 				else
6203 					sf->f_bsize = 8192;
6204 			}
6205 			if (tsf->f_blocks > 0) {
6206 				if (sf->f_bsize != tsf->f_bsize) {
6207 					tot = tsf->f_blocks * tsf->f_bsize;
6208 					sf->f_blocks += (tot / sf->f_bsize);
6209 				} else
6210 					sf->f_blocks += tsf->f_blocks;
6211 			}
6212 			if (tsf->f_bfree > 0) {
6213 				if (sf->f_bsize != tsf->f_bsize) {
6214 					tot = tsf->f_bfree * tsf->f_bsize;
6215 					sf->f_bfree += (tot / sf->f_bsize);
6216 				} else
6217 					sf->f_bfree += tsf->f_bfree;
6218 			}
6219 			if (tsf->f_bavail > 0) {
6220 				if (sf->f_bsize != tsf->f_bsize) {
6221 					tot = tsf->f_bavail * tsf->f_bsize;
6222 					sf->f_bavail += (tot / sf->f_bsize);
6223 				} else
6224 					sf->f_bavail += tsf->f_bavail;
6225 			}
6226 		}
6227 	}
6228 	free(tsf, M_TEMP);
6229 	free(dvpp, M_TEMP);
6230 	return (error);
6231 }
6232 
6233 /*
6234  * Set an NFSv4 acl.
6235  */
6236 int
6237 nfsrv_setacl(struct vnode *vp, NFSACL_T *aclp, struct ucred *cred, NFSPROC_T *p)
6238 {
6239 	int error;
6240 
6241 	if (nfsrv_useacl == 0 || nfs_supportsnfsv4acls(vp) == 0) {
6242 		error = NFSERR_ATTRNOTSUPP;
6243 		goto out;
6244 	}
6245 	/*
6246 	 * With NFSv4 ACLs, chmod(2) may need to add additional entries.
6247 	 * Make sure it has enough room for that - splitting every entry
6248 	 * into two and appending "canonical six" entries at the end.
6249 	 * Cribbed out of kern/vfs_acl.c - Rick M.
6250 	 */
6251 	if (aclp->acl_cnt > (ACL_MAX_ENTRIES - 6) / 2) {
6252 		error = NFSERR_ATTRNOTSUPP;
6253 		goto out;
6254 	}
6255 	error = VOP_SETACL(vp, ACL_TYPE_NFS4, aclp, cred, p);
6256 	if (error == 0) {
6257 		error = nfsrv_dssetacl(vp, aclp, cred, p);
6258 		if (error == ENOENT)
6259 			error = 0;
6260 	}
6261 
6262 out:
6263 	NFSEXITCODE(error);
6264 	return (error);
6265 }
6266 
6267 /*
6268  * Seek vnode op call (actually it is a VOP_IOCTL()).
6269  * This function is called with the vnode locked, but unlocks and vrele()s
6270  * the vp before returning.
6271  */
6272 int
6273 nfsvno_seek(struct nfsrv_descript *nd, struct vnode *vp, u_long cmd,
6274     off_t *offp, int content, bool *eofp, struct ucred *cred, NFSPROC_T *p)
6275 {
6276 	struct nfsvattr at;
6277 	int error, ret;
6278 
6279 	ASSERT_VOP_LOCKED(vp, "nfsvno_seek vp");
6280 	/*
6281 	 * Attempt to seek on a DS file. A return of ENOENT implies
6282 	 * there is no DS file to seek on.
6283 	 */
6284 	error = nfsrv_proxyds(vp, 0, 0, cred, p, NFSPROC_SEEKDS, NULL,
6285 	    NULL, NULL, NULL, NULL, offp, content, eofp);
6286 	if (error != ENOENT) {
6287 		vput(vp);
6288 		return (error);
6289 	}
6290 
6291 	/*
6292 	 * Do the VOP_IOCTL() call.  For the case where *offp == file_size,
6293 	 * VOP_IOCTL() will return ENXIO.  However, the correct reply for
6294 	 * NFSv4.2 is *eofp == true and error == 0 for this case.
6295 	 */
6296 	NFSVOPUNLOCK(vp);
6297 	error = VOP_IOCTL(vp, cmd, offp, 0, cred, p);
6298 	*eofp = false;
6299 	if (error == ENXIO || (error == 0 && cmd == FIOSEEKHOLE)) {
6300 		/* Handle the cases where we might be at EOF. */
6301 		ret = nfsvno_getattr(vp, &at, nd, p, 0, NULL);
6302 		if (ret == 0 && *offp == at.na_size) {
6303 			*eofp = true;
6304 			error = 0;
6305 		}
6306 		if (ret != 0 && error == 0)
6307 			error = ret;
6308 	}
6309 	vrele(vp);
6310 	NFSEXITCODE(error);
6311 	return (error);
6312 }
6313 
6314 /*
6315  * Allocate vnode op call.
6316  */
6317 int
6318 nfsvno_allocate(struct vnode *vp, off_t off, off_t len, struct ucred *cred,
6319     NFSPROC_T *p)
6320 {
6321 	int error, trycnt;
6322 
6323 	ASSERT_VOP_ELOCKED(vp, "nfsvno_allocate vp");
6324 	/*
6325 	 * Attempt to allocate on a DS file. A return of ENOENT implies
6326 	 * there is no DS file to allocate on.
6327 	 */
6328 	error = nfsrv_proxyds(vp, off, 0, cred, p, NFSPROC_ALLOCATE, NULL,
6329 	    NULL, NULL, NULL, NULL, &len, 0, NULL);
6330 	if (error != ENOENT)
6331 		return (error);
6332 	error = 0;
6333 
6334 	/*
6335 	 * Do the actual VOP_ALLOCATE(), looping a reasonable number of
6336 	 * times to achieve completion.
6337 	 */
6338 	trycnt = 0;
6339 	while (error == 0 && len > 0 && trycnt++ < 20)
6340 		error = VOP_ALLOCATE(vp, &off, &len);
6341 	if (error == 0 && len > 0)
6342 		error = NFSERR_IO;
6343 	NFSEXITCODE(error);
6344 	return (error);
6345 }
6346 
6347 /*
6348  * Get Extended Atribute vnode op into an mbuf list.
6349  */
6350 int
6351 nfsvno_getxattr(struct vnode *vp, char *name, uint32_t maxresp,
6352     struct ucred *cred, uint64_t flag, int maxextsiz, struct thread *p,
6353     struct mbuf **mpp, struct mbuf **mpendp, int *lenp)
6354 {
6355 	struct iovec *iv;
6356 	struct uio io, *uiop = &io;
6357 	struct mbuf *m, *m2;
6358 	int alen, error, len, tlen;
6359 	size_t siz;
6360 
6361 	/* First, find out the size of the extended attribute. */
6362 	error = VOP_GETEXTATTR(vp, EXTATTR_NAMESPACE_USER, name, NULL,
6363 	    &siz, cred, p);
6364 	if (error != 0)
6365 		return (NFSERR_NOXATTR);
6366 	if (siz > maxresp - NFS_MAXXDR)
6367 		return (NFSERR_XATTR2BIG);
6368 	len = siz;
6369 	tlen = NFSM_RNDUP(len);
6370 	if (tlen > 0) {
6371 		/*
6372 		 * If cnt > MCLBYTES and the reply will not be saved, use
6373 		 * ext_pgs mbufs for TLS.
6374 		 * For NFSv4.0, we do not know for sure if the reply will
6375 		 * be saved, so do not use ext_pgs mbufs for NFSv4.0.
6376 		 * Always use ext_pgs mbufs if ND_EXTPG is set.
6377 		 */
6378 		if ((flag & ND_EXTPG) != 0 || (tlen > MCLBYTES &&
6379 		    (flag & (ND_TLS | ND_SAVEREPLY)) == ND_TLS &&
6380 		    (flag & (ND_NFSV4 | ND_NFSV41)) != ND_NFSV4))
6381 			uiop->uio_iovcnt = nfsrv_createiovec_extpgs(tlen,
6382 			    maxextsiz, &m, &m2, &iv);
6383 		else
6384 			uiop->uio_iovcnt = nfsrv_createiovec(tlen, &m, &m2,
6385 			    &iv);
6386 		uiop->uio_iov = iv;
6387 	} else {
6388 		uiop->uio_iovcnt = 0;
6389 		uiop->uio_iov = iv = NULL;
6390 		m = m2 = NULL;
6391 	}
6392 	uiop->uio_offset = 0;
6393 	uiop->uio_resid = tlen;
6394 	uiop->uio_rw = UIO_READ;
6395 	uiop->uio_segflg = UIO_SYSSPACE;
6396 	uiop->uio_td = p;
6397 #ifdef MAC
6398 	error = mac_vnode_check_getextattr(cred, vp, EXTATTR_NAMESPACE_USER,
6399 	    name);
6400 	if (error != 0)
6401 		goto out;
6402 #endif
6403 
6404 	if (tlen > 0)
6405 		error = VOP_GETEXTATTR(vp, EXTATTR_NAMESPACE_USER, name, uiop,
6406 		    NULL, cred, p);
6407 	if (error != 0)
6408 		goto out;
6409 	if (uiop->uio_resid > 0) {
6410 		alen = tlen;
6411 		len = tlen - uiop->uio_resid;
6412 		tlen = NFSM_RNDUP(len);
6413 		if (alen != tlen)
6414 			printf("nfsvno_getxattr: weird size read\n");
6415 		if (tlen == 0) {
6416 			m_freem(m);
6417 			m = m2 = NULL;
6418 		} else if (alen != tlen || tlen != len)
6419 			m2 = nfsrv_adj(m, alen - tlen, tlen - len);
6420 	}
6421 	*lenp = len;
6422 	*mpp = m;
6423 	*mpendp = m2;
6424 
6425 out:
6426 	if (error != 0) {
6427 		if (m != NULL)
6428 			m_freem(m);
6429 		*lenp = 0;
6430 	}
6431 	free(iv, M_TEMP);
6432 	NFSEXITCODE(error);
6433 	return (error);
6434 }
6435 
6436 /*
6437  * Set Extended attribute vnode op from an mbuf list.
6438  */
6439 int
6440 nfsvno_setxattr(struct vnode *vp, char *name, int len, struct mbuf *m,
6441     char *cp, struct ucred *cred, struct thread *p)
6442 {
6443 	struct iovec *iv;
6444 	struct uio uio, *uiop = &uio;
6445 	int cnt, error;
6446 
6447 	error = 0;
6448 #ifdef MAC
6449 	error = mac_vnode_check_setextattr(cred, vp, EXTATTR_NAMESPACE_USER,
6450 	    name);
6451 #endif
6452 	if (error != 0)
6453 		goto out;
6454 
6455 	uiop->uio_rw = UIO_WRITE;
6456 	uiop->uio_segflg = UIO_SYSSPACE;
6457 	uiop->uio_td = p;
6458 	uiop->uio_offset = 0;
6459 	uiop->uio_resid = len;
6460 	if (len > 0) {
6461 		error = nfsrv_createiovecw(len, m, cp, &iv, &cnt);
6462 		uiop->uio_iov = iv;
6463 		uiop->uio_iovcnt = cnt;
6464 	} else {
6465 		uiop->uio_iov = iv = NULL;
6466 		uiop->uio_iovcnt = 0;
6467 	}
6468 	if (error == 0) {
6469 		error = VOP_SETEXTATTR(vp, EXTATTR_NAMESPACE_USER, name, uiop,
6470 		    cred, p);
6471 		free(iv, M_TEMP);
6472 	}
6473 
6474 out:
6475 	NFSEXITCODE(error);
6476 	return (error);
6477 }
6478 
6479 /*
6480  * Remove Extended attribute vnode op.
6481  */
6482 int
6483 nfsvno_rmxattr(struct nfsrv_descript *nd, struct vnode *vp, char *name,
6484     struct ucred *cred, struct thread *p)
6485 {
6486 	int error;
6487 
6488 	/*
6489 	 * Get rid of any delegations.  I am not sure why this is required,
6490 	 * but RFC-8276 says so.
6491 	 */
6492 	error = nfsrv_checkremove(vp, 0, nd, nd->nd_clientid, p);
6493 	if (error != 0)
6494 		goto out;
6495 #ifdef MAC
6496 	error = mac_vnode_check_deleteextattr(cred, vp, EXTATTR_NAMESPACE_USER,
6497 	    name);
6498 	if (error != 0)
6499 		goto out;
6500 #endif
6501 
6502 	error = VOP_DELETEEXTATTR(vp, EXTATTR_NAMESPACE_USER, name, cred, p);
6503 	if (error == EOPNOTSUPP)
6504 		error = VOP_SETEXTATTR(vp, EXTATTR_NAMESPACE_USER, name, NULL,
6505 		    cred, p);
6506 out:
6507 	NFSEXITCODE(error);
6508 	return (error);
6509 }
6510 
6511 /*
6512  * List Extended Atribute vnode op into an mbuf list.
6513  */
6514 int
6515 nfsvno_listxattr(struct vnode *vp, uint64_t cookie, struct ucred *cred,
6516     struct thread *p, u_char **bufp, uint32_t *lenp, bool *eofp)
6517 {
6518 	struct iovec iv;
6519 	struct uio io;
6520 	int error;
6521 	size_t siz;
6522 
6523 	*bufp = NULL;
6524 	/* First, find out the size of the extended attribute. */
6525 	error = VOP_LISTEXTATTR(vp, EXTATTR_NAMESPACE_USER, NULL, &siz, cred,
6526 	    p);
6527 	if (error != 0)
6528 		return (NFSERR_NOXATTR);
6529 	if (siz <= cookie) {
6530 		*lenp = 0;
6531 		*eofp = true;
6532 		goto out;
6533 	}
6534 	if (siz > cookie + *lenp) {
6535 		siz = cookie + *lenp;
6536 		*eofp = false;
6537 	} else
6538 		*eofp = true;
6539 	/* Just choose a sanity limit of 10Mbytes for malloc(M_TEMP). */
6540 	if (siz > 10 * 1024 * 1024) {
6541 		error = NFSERR_XATTR2BIG;
6542 		goto out;
6543 	}
6544 	*bufp = malloc(siz, M_TEMP, M_WAITOK);
6545 	iv.iov_base = *bufp;
6546 	iv.iov_len = siz;
6547 	io.uio_iovcnt = 1;
6548 	io.uio_iov = &iv;
6549 	io.uio_offset = 0;
6550 	io.uio_resid = siz;
6551 	io.uio_rw = UIO_READ;
6552 	io.uio_segflg = UIO_SYSSPACE;
6553 	io.uio_td = p;
6554 #ifdef MAC
6555 	error = mac_vnode_check_listextattr(cred, vp, EXTATTR_NAMESPACE_USER);
6556 	if (error != 0)
6557 		goto out;
6558 #endif
6559 
6560 	error = VOP_LISTEXTATTR(vp, EXTATTR_NAMESPACE_USER, &io, NULL, cred,
6561 	    p);
6562 	if (error != 0)
6563 		goto out;
6564 	if (io.uio_resid > 0)
6565 		siz -= io.uio_resid;
6566 	*lenp = siz;
6567 
6568 out:
6569 	if (error != 0) {
6570 		free(*bufp, M_TEMP);
6571 		*bufp = NULL;
6572 	}
6573 	NFSEXITCODE(error);
6574 	return (error);
6575 }
6576 
6577 /*
6578  * Trim trailing data off the mbuf list being built.
6579  */
6580 void
6581 nfsm_trimtrailing(struct nfsrv_descript *nd, struct mbuf *mb, char *bpos,
6582     int bextpg, int bextpgsiz)
6583 {
6584 	vm_page_t pg;
6585 	int fullpgsiz, i;
6586 
6587 	if (mb->m_next != NULL) {
6588 		m_freem(mb->m_next);
6589 		mb->m_next = NULL;
6590 	}
6591 	if ((mb->m_flags & M_EXTPG) != 0) {
6592 		KASSERT(bextpg >= 0 && bextpg < mb->m_epg_npgs,
6593 		    ("nfsm_trimtrailing: bextpg out of range"));
6594 		KASSERT(bpos == (char *)(void *)
6595 		    PHYS_TO_DMAP(mb->m_epg_pa[bextpg]) + PAGE_SIZE - bextpgsiz,
6596 		    ("nfsm_trimtrailing: bextpgsiz bad!"));
6597 
6598 		/* First, get rid of any pages after this position. */
6599 		for (i = mb->m_epg_npgs - 1; i > bextpg; i--) {
6600 			pg = PHYS_TO_VM_PAGE(mb->m_epg_pa[i]);
6601 			vm_page_unwire_noq(pg);
6602 			vm_page_free(pg);
6603 		}
6604 		mb->m_epg_npgs = bextpg + 1;
6605 		if (bextpg == 0)
6606 			fullpgsiz = PAGE_SIZE - mb->m_epg_1st_off;
6607 		else
6608 			fullpgsiz = PAGE_SIZE;
6609 		mb->m_epg_last_len = fullpgsiz - bextpgsiz;
6610 		mb->m_len = m_epg_pagelen(mb, 0, mb->m_epg_1st_off);
6611 		for (i = 1; i < mb->m_epg_npgs; i++)
6612 			mb->m_len += m_epg_pagelen(mb, i, 0);
6613 		nd->nd_bextpgsiz = bextpgsiz;
6614 		nd->nd_bextpg = bextpg;
6615 	} else
6616 		mb->m_len = bpos - mtod(mb, char *);
6617 	nd->nd_mb = mb;
6618 	nd->nd_bpos = bpos;
6619 }
6620 
6621 
6622 /*
6623  * Check to see if a put file handle operation should test for
6624  * NFSERR_WRONGSEC, although NFSv3 actually returns NFSERR_AUTHERR.
6625  * When Open is the next operation, NFSERR_WRONGSEC cannot be
6626  * replied for the Open cases that use a component.  Thia can
6627  * be identified by the fact that the file handle's type is VDIR.
6628  */
6629 bool
6630 nfsrv_checkwrongsec(struct nfsrv_descript *nd, int nextop, enum vtype vtyp)
6631 {
6632 
6633 	if ((nd->nd_flag & ND_NFSV4) == 0) {
6634 		if (nd->nd_procnum == NFSPROC_FSINFO)
6635 			return (false);
6636 		return (true);
6637 	}
6638 
6639 	if ((nd->nd_flag & ND_LASTOP) != 0)
6640 		return (false);
6641 
6642 	if (nextop == NFSV4OP_PUTROOTFH || nextop == NFSV4OP_PUTFH ||
6643 	    nextop == NFSV4OP_PUTPUBFH || nextop == NFSV4OP_RESTOREFH ||
6644 	    nextop == NFSV4OP_LOOKUP || nextop == NFSV4OP_LOOKUPP ||
6645 	    nextop == NFSV4OP_SECINFO || nextop == NFSV4OP_SECINFONONAME)
6646 		return (false);
6647 	if (nextop == NFSV4OP_OPEN && vtyp == VDIR)
6648 		return (false);
6649 	return (true);
6650 }
6651 
6652 extern int (*nfsd_call_nfsd)(struct thread *, struct nfssvc_args *);
6653 
6654 /*
6655  * Called once to initialize data structures...
6656  */
6657 static int
6658 nfsd_modevent(module_t mod, int type, void *data)
6659 {
6660 	int error = 0, i;
6661 	static int loaded = 0;
6662 
6663 	switch (type) {
6664 	case MOD_LOAD:
6665 		if (loaded)
6666 			goto out;
6667 		newnfs_portinit();
6668 		for (i = 0; i < NFSRVCACHE_HASHSIZE; i++) {
6669 			mtx_init(&nfsrchash_table[i].mtx, "nfsrtc", NULL,
6670 			    MTX_DEF);
6671 			mtx_init(&nfsrcahash_table[i].mtx, "nfsrtca", NULL,
6672 			    MTX_DEF);
6673 		}
6674 		mtx_init(&nfsrc_udpmtx, "nfsuc", NULL, MTX_DEF);
6675 		mtx_init(&nfs_v4root_mutex, "nfs4rt", NULL, MTX_DEF);
6676 		mtx_init(&nfsv4root_mnt.mnt_mtx, "nfs4mnt", NULL, MTX_DEF);
6677 		mtx_init(&nfsrv_dontlistlock_mtx, "nfs4dnl", NULL, MTX_DEF);
6678 		mtx_init(&nfsrv_recalllock_mtx, "nfs4rec", NULL, MTX_DEF);
6679 		lockinit(&nfsv4root_mnt.mnt_explock, PVFS, "explock", 0, 0);
6680 		nfsrvd_initcache();
6681 		nfsd_init();
6682 		NFSD_LOCK();
6683 		nfsrvd_init(0);
6684 		NFSD_UNLOCK();
6685 		nfsd_mntinit();
6686 #ifdef VV_DISABLEDELEG
6687 		vn_deleg_ops.vndeleg_recall = nfsd_recalldelegation;
6688 		vn_deleg_ops.vndeleg_disable = nfsd_disabledelegation;
6689 #endif
6690 		nfsd_call_servertimer = nfsrv_servertimer;
6691 		nfsd_call_nfsd = nfssvc_nfsd;
6692 		loaded = 1;
6693 		break;
6694 
6695 	case MOD_UNLOAD:
6696 		if (newnfs_numnfsd != 0) {
6697 			error = EBUSY;
6698 			break;
6699 		}
6700 
6701 #ifdef VV_DISABLEDELEG
6702 		vn_deleg_ops.vndeleg_recall = NULL;
6703 		vn_deleg_ops.vndeleg_disable = NULL;
6704 #endif
6705 		nfsd_call_servertimer = NULL;
6706 		nfsd_call_nfsd = NULL;
6707 
6708 		/* Clean out all NFSv4 state. */
6709 		nfsrv_throwawayallstate(curthread);
6710 
6711 		/* Clean the NFS server reply cache */
6712 		nfsrvd_cleancache();
6713 
6714 		/* Free up the krpc server pool. */
6715 		if (nfsrvd_pool != NULL)
6716 			svcpool_destroy(nfsrvd_pool);
6717 
6718 		/* and get rid of the locks */
6719 		for (i = 0; i < NFSRVCACHE_HASHSIZE; i++) {
6720 			mtx_destroy(&nfsrchash_table[i].mtx);
6721 			mtx_destroy(&nfsrcahash_table[i].mtx);
6722 		}
6723 		mtx_destroy(&nfsrc_udpmtx);
6724 		mtx_destroy(&nfs_v4root_mutex);
6725 		mtx_destroy(&nfsv4root_mnt.mnt_mtx);
6726 		mtx_destroy(&nfsrv_dontlistlock_mtx);
6727 		mtx_destroy(&nfsrv_recalllock_mtx);
6728 		for (i = 0; i < nfsrv_sessionhashsize; i++)
6729 			mtx_destroy(&nfssessionhash[i].mtx);
6730 		if (nfslayouthash != NULL) {
6731 			for (i = 0; i < nfsrv_layouthashsize; i++)
6732 				mtx_destroy(&nfslayouthash[i].mtx);
6733 			free(nfslayouthash, M_NFSDSESSION);
6734 		}
6735 		lockdestroy(&nfsv4root_mnt.mnt_explock);
6736 		free(nfsclienthash, M_NFSDCLIENT);
6737 		free(nfslockhash, M_NFSDLOCKFILE);
6738 		free(nfssessionhash, M_NFSDSESSION);
6739 		loaded = 0;
6740 		break;
6741 	default:
6742 		error = EOPNOTSUPP;
6743 		break;
6744 	}
6745 
6746 out:
6747 	NFSEXITCODE(error);
6748 	return (error);
6749 }
6750 static moduledata_t nfsd_mod = {
6751 	"nfsd",
6752 	nfsd_modevent,
6753 	NULL,
6754 };
6755 DECLARE_MODULE(nfsd, nfsd_mod, SI_SUB_VFS, SI_ORDER_ANY);
6756 
6757 /* So that loader and kldload(2) can find us, wherever we are.. */
6758 MODULE_VERSION(nfsd, 1);
6759 MODULE_DEPEND(nfsd, nfscommon, 1, 1, 1);
6760 MODULE_DEPEND(nfsd, nfslockd, 1, 1, 1);
6761 MODULE_DEPEND(nfsd, krpc, 1, 1, 1);
6762 MODULE_DEPEND(nfsd, nfssvc, 1, 1, 1);
6763