xref: /freebsd/sys/fs/nfsserver/nfs_nfsdport.c (revision ce9f95bd835ec7d3ac6def5a910282509b1413f0)
1 /*-
2  * SPDX-License-Identifier: BSD-3-Clause
3  *
4  * Copyright (c) 1989, 1993
5  *	The Regents of the University of California.  All rights reserved.
6  *
7  * This code is derived from software contributed to Berkeley by
8  * Rick Macklem at The University of Guelph.
9  *
10  * Redistribution and use in source and binary forms, with or without
11  * modification, are permitted provided that the following conditions
12  * are met:
13  * 1. Redistributions of source code must retain the above copyright
14  *    notice, this list of conditions and the following disclaimer.
15  * 2. Redistributions in binary form must reproduce the above copyright
16  *    notice, this list of conditions and the following disclaimer in the
17  *    documentation and/or other materials provided with the distribution.
18  * 3. Neither the name of the University nor the names of its contributors
19  *    may be used to endorse or promote products derived from this software
20  *    without specific prior written permission.
21  *
22  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
23  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
24  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
25  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
26  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
27  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
28  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
29  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
30  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
31  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
32  * SUCH DAMAGE.
33  *
34  */
35 
36 #include <sys/cdefs.h>
37 __FBSDID("$FreeBSD$");
38 
39 #include <sys/capsicum.h>
40 #include <sys/extattr.h>
41 
42 /*
43  * Functions that perform the vfs operations required by the routines in
44  * nfsd_serv.c. It is hoped that this change will make the server more
45  * portable.
46  */
47 
48 #include <fs/nfs/nfsport.h>
49 #include <security/mac/mac_framework.h>
50 #include <sys/callout.h>
51 #include <sys/filio.h>
52 #include <sys/hash.h>
53 #include <sys/osd.h>
54 #include <sys/sysctl.h>
55 #include <nlm/nlm_prot.h>
56 #include <nlm/nlm.h>
57 
58 FEATURE(nfsd, "NFSv4 server");
59 
60 extern u_int32_t newnfs_true, newnfs_false, newnfs_xdrneg1;
61 extern int nfsrv_useacl;
62 extern int newnfs_numnfsd;
63 extern int nfsrv_sessionhashsize;
64 extern struct nfslayouthash *nfslayouthash;
65 extern int nfsrv_layouthashsize;
66 extern struct mtx nfsrv_dslock_mtx;
67 extern int nfs_pnfsiothreads;
68 extern volatile int nfsrv_devidcnt;
69 extern int nfsrv_maxpnfsmirror;
70 extern uint32_t nfs_srvmaxio;
71 extern int nfs_bufpackets;
72 extern u_long sb_max_adj;
73 
74 NFSD_VNET_DECLARE(int, nfsrv_numnfsd);
75 NFSD_VNET_DECLARE(struct nfsrv_stablefirst, nfsrv_stablefirst);
76 NFSD_VNET_DECLARE(SVCPOOL *, nfsrvd_pool);
77 NFSD_VNET_DECLARE(struct nfsclienthashhead *, nfsclienthash);
78 NFSD_VNET_DECLARE(struct nfslockhashhead *, nfslockhash);
79 NFSD_VNET_DECLARE(struct nfssessionhash *, nfssessionhash);
80 NFSD_VNET_DECLARE(struct nfsv4lock, nfsd_suspend_lock);
81 NFSD_VNET_DECLARE(struct nfsstatsv1 *, nfsstatsv1_p);
82 
83 NFSDLOCKMUTEX;
84 NFSSTATESPINLOCK;
85 struct mtx nfsrc_udpmtx;
86 struct mtx nfs_v4root_mutex;
87 struct mtx nfsrv_dontlistlock_mtx;
88 struct mtx nfsrv_recalllock_mtx;
89 struct nfsrvfh nfs_pubfh;
90 int nfs_pubfhset = 0;
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 NFSD_VNET_DEFINE(struct proc *, nfsd_master_proc) = NULL;
99 NFSD_VNET_DEFINE(struct nfsrvhashhead *, nfsrvudphashtbl);
100 NFSD_VNET_DEFINE(struct nfsrchash_bucket *, nfsrchash_table);
101 NFSD_VNET_DEFINE(struct nfsrchash_bucket *, nfsrcahash_table);
102 NFSD_VNET_DEFINE(struct nfsrvfh, nfs_rootfh);
103 NFSD_VNET_DEFINE(int, nfs_rootfhset) = 0;
104 NFSD_VNET_DEFINE(struct callout, nfsd_callout);
105 NFSD_VNET_DEFINE_STATIC(struct mount *, nfsv4root_mnt);
106 NFSD_VNET_DEFINE_STATIC(struct vfsoptlist, nfsv4root_opt);
107 NFSD_VNET_DEFINE_STATIC(struct vfsoptlist, nfsv4root_newopt);
108 NFSD_VNET_DEFINE_STATIC(bool, nfsrv_suspend_nfsd) = false;
109 NFSD_VNET_DEFINE_STATIC(bool, nfsrv_mntinited) = false;
110 
111 static int nfssvc_srvcall(struct thread *, struct nfssvc_args *,
112     struct ucred *);
113 static void nfsvno_updateds(struct vnode *, struct ucred *, struct thread *);
114 
115 int nfsrv_enable_crossmntpt = 1;
116 static int nfs_commit_blks;
117 static int nfs_commit_miss;
118 extern int nfsrv_issuedelegs;
119 extern int nfsrv_dolocallocks;
120 extern struct nfsdevicehead nfsrv_devidhead;
121 
122 static int nfsrv_createiovec(int, struct mbuf **, struct mbuf **,
123     struct iovec **);
124 static int nfsrv_createiovec_extpgs(int, int, struct mbuf **,
125     struct mbuf **, struct iovec **);
126 static int nfsrv_createiovecw(int, struct mbuf *, char *, struct iovec **,
127     int *);
128 static void nfsrv_pnfscreate(struct vnode *, struct vattr *, struct ucred *,
129     NFSPROC_T *);
130 static void nfsrv_pnfsremovesetup(struct vnode *, NFSPROC_T *, struct vnode **,
131     int *, char *, fhandle_t *);
132 static void nfsrv_pnfsremove(struct vnode **, int, char *, fhandle_t *,
133     NFSPROC_T *);
134 static int nfsrv_proxyds(struct vnode *, off_t, int, struct ucred *,
135     struct thread *, int, struct mbuf **, char *, struct mbuf **,
136     struct nfsvattr *, struct acl *, off_t *, int, bool *);
137 static int nfsrv_setextattr(struct vnode *, struct nfsvattr *, NFSPROC_T *);
138 static int nfsrv_readdsrpc(fhandle_t *, off_t, int, struct ucred *,
139     NFSPROC_T *, struct nfsmount *, struct mbuf **, struct mbuf **);
140 static int nfsrv_writedsrpc(fhandle_t *, off_t, int, struct ucred *,
141     NFSPROC_T *, struct vnode *, struct nfsmount **, int, struct mbuf **,
142     char *, int *);
143 static int nfsrv_allocatedsrpc(fhandle_t *, off_t, off_t, struct ucred *,
144     NFSPROC_T *, struct vnode *, struct nfsmount **, int, int *);
145 static int nfsrv_deallocatedsrpc(fhandle_t *, off_t, off_t, struct ucred *,
146     NFSPROC_T *, struct vnode *, struct nfsmount **, int, int *);
147 static int nfsrv_setacldsrpc(fhandle_t *, struct ucred *, NFSPROC_T *,
148     struct vnode *, struct nfsmount **, int, struct acl *, int *);
149 static int nfsrv_setattrdsrpc(fhandle_t *, struct ucred *, NFSPROC_T *,
150     struct vnode *, struct nfsmount **, int, struct nfsvattr *, int *);
151 static int nfsrv_getattrdsrpc(fhandle_t *, struct ucred *, NFSPROC_T *,
152     struct vnode *, struct nfsmount *, struct nfsvattr *);
153 static int nfsrv_seekdsrpc(fhandle_t *, off_t *, int, bool *, struct ucred *,
154     NFSPROC_T *, struct nfsmount *);
155 static int nfsrv_putfhname(fhandle_t *, char *);
156 static int nfsrv_pnfslookupds(struct vnode *, struct vnode *,
157     struct pnfsdsfile *, struct vnode **, NFSPROC_T *);
158 static void nfsrv_pnfssetfh(struct vnode *, struct pnfsdsfile *, char *, char *,
159     struct vnode *, NFSPROC_T *);
160 static int nfsrv_dsremove(struct vnode *, char *, struct ucred *, NFSPROC_T *);
161 static int nfsrv_dssetacl(struct vnode *, struct acl *, struct ucred *,
162     NFSPROC_T *);
163 static int nfsrv_pnfsstatfs(struct statfs *, struct mount *);
164 
165 int nfs_pnfsio(task_fn_t *, void *);
166 
167 SYSCTL_NODE(_vfs, OID_AUTO, nfsd, CTLFLAG_RW | CTLFLAG_MPSAFE, 0,
168     "NFS server");
169 SYSCTL_INT(_vfs_nfsd, OID_AUTO, mirrormnt, CTLFLAG_RW,
170     &nfsrv_enable_crossmntpt, 0, "Enable nfsd to cross mount points");
171 SYSCTL_INT(_vfs_nfsd, OID_AUTO, commit_blks, CTLFLAG_RW, &nfs_commit_blks,
172     0, "");
173 SYSCTL_INT(_vfs_nfsd, OID_AUTO, commit_miss, CTLFLAG_RW, &nfs_commit_miss,
174     0, "");
175 SYSCTL_INT(_vfs_nfsd, OID_AUTO, issue_delegations, CTLFLAG_RW,
176     &nfsrv_issuedelegs, 0, "Enable nfsd to issue delegations");
177 SYSCTL_INT(_vfs_nfsd, OID_AUTO, enable_locallocks, CTLFLAG_RW,
178     &nfsrv_dolocallocks, 0, "Enable nfsd to acquire local locks on files");
179 SYSCTL_INT(_vfs_nfsd, OID_AUTO, debuglevel, CTLFLAG_RW, &nfsd_debuglevel,
180     0, "Debug level for NFS server");
181 NFSD_VNET_DECLARE(int, nfsd_enable_stringtouid);
182 SYSCTL_INT(_vfs_nfsd, OID_AUTO, enable_stringtouid,
183     CTLFLAG_NFSD_VNET | CTLFLAG_RW, &NFSD_VNET_NAME(nfsd_enable_stringtouid),
184     0, "Enable nfsd to accept numeric owner_names");
185 static int nfsrv_pnfsgetdsattr = 1;
186 SYSCTL_INT(_vfs_nfsd, OID_AUTO, pnfsgetdsattr, CTLFLAG_RW,
187     &nfsrv_pnfsgetdsattr, 0, "When set getattr gets DS attributes via RPC");
188 
189 /*
190  * nfsrv_dsdirsize can only be increased and only when the nfsd threads are
191  * not running.
192  * The dsN subdirectories for the increased values must have been created
193  * on all DS servers before this increase is done.
194  */
195 u_int	nfsrv_dsdirsize = 20;
196 static int
197 sysctl_dsdirsize(SYSCTL_HANDLER_ARGS)
198 {
199 	int error, newdsdirsize;
200 
201 	newdsdirsize = nfsrv_dsdirsize;
202 	error = sysctl_handle_int(oidp, &newdsdirsize, 0, req);
203 	if (error != 0 || req->newptr == NULL)
204 		return (error);
205 	if (newdsdirsize <= nfsrv_dsdirsize || newdsdirsize > 10000 ||
206 	    newnfs_numnfsd != 0)
207 		return (EINVAL);
208 	nfsrv_dsdirsize = newdsdirsize;
209 	return (0);
210 }
211 SYSCTL_PROC(_vfs_nfsd, OID_AUTO, dsdirsize,
212     CTLTYPE_UINT | CTLFLAG_MPSAFE | CTLFLAG_RW, 0, sizeof(nfsrv_dsdirsize),
213     sysctl_dsdirsize, "IU", "Number of dsN subdirs on the DS servers");
214 
215 /*
216  * nfs_srvmaxio can only be increased and only when the nfsd threads are
217  * not running.  The setting must be a power of 2, with the current limit of
218  * 1Mbyte.
219  */
220 static int
221 sysctl_srvmaxio(SYSCTL_HANDLER_ARGS)
222 {
223 	int error;
224 	u_int newsrvmaxio;
225 	uint64_t tval;
226 
227 	newsrvmaxio = nfs_srvmaxio;
228 	error = sysctl_handle_int(oidp, &newsrvmaxio, 0, req);
229 	if (error != 0 || req->newptr == NULL)
230 		return (error);
231 	if (newsrvmaxio == nfs_srvmaxio)
232 		return (0);
233 	if (newsrvmaxio < nfs_srvmaxio) {
234 		printf("nfsd: vfs.nfsd.srvmaxio can only be increased\n");
235 		return (EINVAL);
236 	}
237 	if (newsrvmaxio > 1048576) {
238 		printf("nfsd: vfs.nfsd.srvmaxio cannot be > 1Mbyte\n");
239 		return (EINVAL);
240 	}
241 	if ((newsrvmaxio & (newsrvmaxio - 1)) != 0) {
242 		printf("nfsd: vfs.nfsd.srvmaxio must be a power of 2\n");
243 		return (EINVAL);
244 	}
245 
246 	/*
247 	 * Check that kern.ipc.maxsockbuf is large enough for
248 	 * newsrviomax, given the setting of vfs.nfs.bufpackets.
249 	 */
250 	if ((newsrvmaxio + NFS_MAXXDR) * nfs_bufpackets >
251 	    sb_max_adj) {
252 		/*
253 		 * Suggest vfs.nfs.bufpackets * maximum RPC message for
254 		 * sb_max_adj.
255 		 */
256 		tval = (newsrvmaxio + NFS_MAXXDR) * nfs_bufpackets;
257 
258 		/*
259 		 * Convert suggested sb_max_adj value to a suggested
260 		 * sb_max value, which is what is set via kern.ipc.maxsockbuf.
261 		 * Perform the inverse calculation of (from uipc_sockbuf.c):
262 		 * sb_max_adj = (u_quad_t)sb_max * MCLBYTES /
263 		 *     (MSIZE + MCLBYTES);
264 		 * XXX If the calculation of sb_max_adj from sb_max changes,
265 		 *     this calculation must be changed as well.
266 		 */
267 		tval *= (MSIZE + MCLBYTES);  /* Brackets for readability. */
268 		tval += MCLBYTES - 1;        /* Round up divide. */
269 		tval /= MCLBYTES;
270 		printf("nfsd: set kern.ipc.maxsockbuf to a minimum of "
271 		    "%ju to support %ubyte NFS I/O\n", (uintmax_t)tval,
272 		    newsrvmaxio);
273 		return (EINVAL);
274 	}
275 
276 	NFSD_LOCK();
277 	if (newnfs_numnfsd != 0) {
278 		NFSD_UNLOCK();
279 		printf("nfsd: cannot set vfs.nfsd.srvmaxio when nfsd "
280 		    "threads are running\n");
281 		return (EINVAL);
282 	}
283 
284 
285 	nfs_srvmaxio = newsrvmaxio;
286 	NFSD_UNLOCK();
287 	return (0);
288 }
289 SYSCTL_PROC(_vfs_nfsd, OID_AUTO, srvmaxio,
290     CTLTYPE_UINT | CTLFLAG_MPSAFE | CTLFLAG_RW, NULL, 0,
291     sysctl_srvmaxio, "IU", "Maximum I/O size in bytes");
292 
293 #define	MAX_REORDERED_RPC	16
294 #define	NUM_HEURISTIC		1031
295 #define	NHUSE_INIT		64
296 #define	NHUSE_INC		16
297 #define	NHUSE_MAX		2048
298 
299 static struct nfsheur {
300 	struct vnode *nh_vp;	/* vp to match (unreferenced pointer) */
301 	off_t nh_nextoff;	/* next offset for sequential detection */
302 	int nh_use;		/* use count for selection */
303 	int nh_seqcount;	/* heuristic */
304 } nfsheur[NUM_HEURISTIC];
305 
306 /*
307  * Heuristic to detect sequential operation.
308  */
309 static struct nfsheur *
310 nfsrv_sequential_heuristic(struct uio *uio, struct vnode *vp)
311 {
312 	struct nfsheur *nh;
313 	int hi, try;
314 
315 	/* Locate best candidate. */
316 	try = 32;
317 	hi = ((int)(vm_offset_t)vp / sizeof(struct vnode)) % NUM_HEURISTIC;
318 	nh = &nfsheur[hi];
319 	while (try--) {
320 		if (nfsheur[hi].nh_vp == vp) {
321 			nh = &nfsheur[hi];
322 			break;
323 		}
324 		if (nfsheur[hi].nh_use > 0)
325 			--nfsheur[hi].nh_use;
326 		hi = (hi + 1) % NUM_HEURISTIC;
327 		if (nfsheur[hi].nh_use < nh->nh_use)
328 			nh = &nfsheur[hi];
329 	}
330 
331 	/* Initialize hint if this is a new file. */
332 	if (nh->nh_vp != vp) {
333 		nh->nh_vp = vp;
334 		nh->nh_nextoff = uio->uio_offset;
335 		nh->nh_use = NHUSE_INIT;
336 		if (uio->uio_offset == 0)
337 			nh->nh_seqcount = 4;
338 		else
339 			nh->nh_seqcount = 1;
340 	}
341 
342 	/* Calculate heuristic. */
343 	if ((uio->uio_offset == 0 && nh->nh_seqcount > 0) ||
344 	    uio->uio_offset == nh->nh_nextoff) {
345 		/* See comments in vfs_vnops.c:sequential_heuristic(). */
346 		nh->nh_seqcount += howmany(uio->uio_resid, 16384);
347 		if (nh->nh_seqcount > IO_SEQMAX)
348 			nh->nh_seqcount = IO_SEQMAX;
349 	} else if (qabs(uio->uio_offset - nh->nh_nextoff) <= MAX_REORDERED_RPC *
350 	    imax(vp->v_mount->mnt_stat.f_iosize, uio->uio_resid)) {
351 		/* Probably a reordered RPC, leave seqcount alone. */
352 	} else if (nh->nh_seqcount > 1) {
353 		nh->nh_seqcount /= 2;
354 	} else {
355 		nh->nh_seqcount = 0;
356 	}
357 	nh->nh_use += NHUSE_INC;
358 	if (nh->nh_use > NHUSE_MAX)
359 		nh->nh_use = NHUSE_MAX;
360 	return (nh);
361 }
362 
363 /*
364  * Get attributes into nfsvattr structure.
365  */
366 int
367 nfsvno_getattr(struct vnode *vp, struct nfsvattr *nvap,
368     struct nfsrv_descript *nd, struct thread *p, int vpislocked,
369     nfsattrbit_t *attrbitp)
370 {
371 	int error, gotattr, lockedit = 0;
372 	struct nfsvattr na;
373 
374 	if (vpislocked == 0) {
375 		/*
376 		 * When vpislocked == 0, the vnode is either exclusively
377 		 * locked by this thread or not locked by this thread.
378 		 * As such, shared lock it, if not exclusively locked.
379 		 */
380 		if (NFSVOPISLOCKED(vp) != LK_EXCLUSIVE) {
381 			lockedit = 1;
382 			NFSVOPLOCK(vp, LK_SHARED | LK_RETRY);
383 		}
384 	}
385 
386 	/*
387 	 * Acquire the Change, Size, TimeAccess, TimeModify and SpaceUsed
388 	 * attributes, as required.
389 	 * This needs to be done for regular files if:
390 	 * - non-NFSv4 RPCs or
391 	 * - when attrbitp == NULL or
392 	 * - an NFSv4 RPC with any of the above attributes in attrbitp.
393 	 * A return of 0 for nfsrv_proxyds() indicates that it has acquired
394 	 * these attributes.  nfsrv_proxyds() will return an error if the
395 	 * server is not a pNFS one.
396 	 */
397 	gotattr = 0;
398 	if (vp->v_type == VREG && nfsrv_devidcnt > 0 && (attrbitp == NULL ||
399 	    (nd->nd_flag & ND_NFSV4) == 0 ||
400 	    NFSISSET_ATTRBIT(attrbitp, NFSATTRBIT_CHANGE) ||
401 	    NFSISSET_ATTRBIT(attrbitp, NFSATTRBIT_SIZE) ||
402 	    NFSISSET_ATTRBIT(attrbitp, NFSATTRBIT_TIMEACCESS) ||
403 	    NFSISSET_ATTRBIT(attrbitp, NFSATTRBIT_TIMEMODIFY) ||
404 	    NFSISSET_ATTRBIT(attrbitp, NFSATTRBIT_SPACEUSED))) {
405 		error = nfsrv_proxyds(vp, 0, 0, nd->nd_cred, p,
406 		    NFSPROC_GETATTR, NULL, NULL, NULL, &na, NULL, NULL, 0,
407 		    NULL);
408 		if (error == 0)
409 			gotattr = 1;
410 	}
411 
412 	error = VOP_GETATTR(vp, &nvap->na_vattr, nd->nd_cred);
413 	if (lockedit != 0)
414 		NFSVOPUNLOCK(vp);
415 
416 	/*
417 	 * If we got the Change, Size and Modify Time from the DS,
418 	 * replace them.
419 	 */
420 	if (gotattr != 0) {
421 		nvap->na_atime = na.na_atime;
422 		nvap->na_mtime = na.na_mtime;
423 		nvap->na_filerev = na.na_filerev;
424 		nvap->na_size = na.na_size;
425 		nvap->na_bytes = na.na_bytes;
426 	}
427 	NFSD_DEBUG(4, "nfsvno_getattr: gotattr=%d err=%d chg=%ju\n", gotattr,
428 	    error, (uintmax_t)na.na_filerev);
429 
430 	NFSEXITCODE(error);
431 	return (error);
432 }
433 
434 /*
435  * Get a file handle for a vnode.
436  */
437 int
438 nfsvno_getfh(struct vnode *vp, fhandle_t *fhp, struct thread *p)
439 {
440 	int error;
441 
442 	NFSBZERO((caddr_t)fhp, sizeof(fhandle_t));
443 	fhp->fh_fsid = vp->v_mount->mnt_stat.f_fsid;
444 	error = VOP_VPTOFH(vp, &fhp->fh_fid);
445 
446 	NFSEXITCODE(error);
447 	return (error);
448 }
449 
450 /*
451  * Perform access checking for vnodes obtained from file handles that would
452  * refer to files already opened by a Unix client. You cannot just use
453  * vn_writechk() and VOP_ACCESSX() for two reasons.
454  * 1 - You must check for exported rdonly as well as MNT_RDONLY for the write
455  *     case.
456  * 2 - The owner is to be given access irrespective of mode bits for some
457  *     operations, so that processes that chmod after opening a file don't
458  *     break.
459  */
460 int
461 nfsvno_accchk(struct vnode *vp, accmode_t accmode, struct ucred *cred,
462     struct nfsexstuff *exp, struct thread *p, int override, int vpislocked,
463     u_int32_t *supportedtypep)
464 {
465 	struct vattr vattr;
466 	int error = 0, getret = 0;
467 
468 	if (vpislocked == 0) {
469 		if (NFSVOPLOCK(vp, LK_SHARED) != 0) {
470 			error = EPERM;
471 			goto out;
472 		}
473 	}
474 	if (accmode & VWRITE) {
475 		/* Just vn_writechk() changed to check rdonly */
476 		/*
477 		 * Disallow write attempts on read-only file systems;
478 		 * unless the file is a socket or a block or character
479 		 * device resident on the file system.
480 		 */
481 		if (NFSVNO_EXRDONLY(exp) ||
482 		    (vp->v_mount->mnt_flag & MNT_RDONLY)) {
483 			switch (vp->v_type) {
484 			case VREG:
485 			case VDIR:
486 			case VLNK:
487 				error = EROFS;
488 			default:
489 				break;
490 			}
491 		}
492 		/*
493 		 * If there's shared text associated with
494 		 * the inode, try to free it up once.  If
495 		 * we fail, we can't allow writing.
496 		 */
497 		if (VOP_IS_TEXT(vp) && error == 0)
498 			error = ETXTBSY;
499 	}
500 	if (error != 0) {
501 		if (vpislocked == 0)
502 			NFSVOPUNLOCK(vp);
503 		goto out;
504 	}
505 
506 	/*
507 	 * Should the override still be applied when ACLs are enabled?
508 	 */
509 	error = VOP_ACCESSX(vp, accmode, cred, p);
510 	if (error != 0 && (accmode & (VDELETE | VDELETE_CHILD))) {
511 		/*
512 		 * Try again with VEXPLICIT_DENY, to see if the test for
513 		 * deletion is supported.
514 		 */
515 		error = VOP_ACCESSX(vp, accmode | VEXPLICIT_DENY, cred, p);
516 		if (error == 0) {
517 			if (vp->v_type == VDIR) {
518 				accmode &= ~(VDELETE | VDELETE_CHILD);
519 				accmode |= VWRITE;
520 				error = VOP_ACCESSX(vp, accmode, cred, p);
521 			} else if (supportedtypep != NULL) {
522 				*supportedtypep &= ~NFSACCESS_DELETE;
523 			}
524 		}
525 	}
526 
527 	/*
528 	 * Allow certain operations for the owner (reads and writes
529 	 * on files that are already open).
530 	 */
531 	if (override != NFSACCCHK_NOOVERRIDE &&
532 	    (error == EPERM || error == EACCES)) {
533 		if (cred->cr_uid == 0 && (override & NFSACCCHK_ALLOWROOT))
534 			error = 0;
535 		else if (override & NFSACCCHK_ALLOWOWNER) {
536 			getret = VOP_GETATTR(vp, &vattr, cred);
537 			if (getret == 0 && cred->cr_uid == vattr.va_uid)
538 				error = 0;
539 		}
540 	}
541 	if (vpislocked == 0)
542 		NFSVOPUNLOCK(vp);
543 
544 out:
545 	NFSEXITCODE(error);
546 	return (error);
547 }
548 
549 /*
550  * Set attribute(s) vnop.
551  */
552 int
553 nfsvno_setattr(struct vnode *vp, struct nfsvattr *nvap, struct ucred *cred,
554     struct thread *p, struct nfsexstuff *exp)
555 {
556 	u_quad_t savsize = 0;
557 	int error, savedit;
558 	time_t savbtime;
559 
560 	/*
561 	 * If this is an exported file system and a pNFS service is running,
562 	 * don't VOP_SETATTR() of size for the MDS file system.
563 	 */
564 	savedit = 0;
565 	error = 0;
566 	if (vp->v_type == VREG && (vp->v_mount->mnt_flag & MNT_EXPORTED) != 0 &&
567 	    nfsrv_devidcnt != 0 && nvap->na_vattr.va_size != VNOVAL &&
568 	    nvap->na_vattr.va_size > 0) {
569 		savsize = nvap->na_vattr.va_size;
570 		nvap->na_vattr.va_size = VNOVAL;
571 		if (nvap->na_vattr.va_uid != (uid_t)VNOVAL ||
572 		    nvap->na_vattr.va_gid != (gid_t)VNOVAL ||
573 		    nvap->na_vattr.va_mode != (mode_t)VNOVAL ||
574 		    nvap->na_vattr.va_atime.tv_sec != VNOVAL ||
575 		    nvap->na_vattr.va_mtime.tv_sec != VNOVAL)
576 			savedit = 1;
577 		else
578 			savedit = 2;
579 	}
580 	if (savedit != 2)
581 		error = VOP_SETATTR(vp, &nvap->na_vattr, cred);
582 	if (savedit != 0)
583 		nvap->na_vattr.va_size = savsize;
584 	if (error == 0 && (nvap->na_vattr.va_uid != (uid_t)VNOVAL ||
585 	    nvap->na_vattr.va_gid != (gid_t)VNOVAL ||
586 	    nvap->na_vattr.va_size != VNOVAL ||
587 	    nvap->na_vattr.va_mode != (mode_t)VNOVAL ||
588 	    nvap->na_vattr.va_atime.tv_sec != VNOVAL ||
589 	    nvap->na_vattr.va_mtime.tv_sec != VNOVAL)) {
590 		/* Never modify birthtime on a DS file. */
591 		savbtime = nvap->na_vattr.va_birthtime.tv_sec;
592 		nvap->na_vattr.va_birthtime.tv_sec = VNOVAL;
593 		/* For a pNFS server, set the attributes on the DS file. */
594 		error = nfsrv_proxyds(vp, 0, 0, cred, p, NFSPROC_SETATTR,
595 		    NULL, NULL, NULL, nvap, NULL, NULL, 0, NULL);
596 		nvap->na_vattr.va_birthtime.tv_sec = savbtime;
597 		if (error == ENOENT)
598 			error = 0;
599 	}
600 	NFSEXITCODE(error);
601 	return (error);
602 }
603 
604 /*
605  * Set up nameidata for a lookup() call and do it.
606  */
607 int
608 nfsvno_namei(struct nfsrv_descript *nd, struct nameidata *ndp,
609     struct vnode *dp, int islocked, struct nfsexstuff *exp,
610     struct vnode **retdirp)
611 {
612 	struct componentname *cnp = &ndp->ni_cnd;
613 	int i;
614 	struct iovec aiov;
615 	struct uio auio;
616 	int lockleaf = (cnp->cn_flags & LOCKLEAF) != 0, linklen;
617 	int error = 0;
618 	char *cp;
619 
620 	*retdirp = NULL;
621 	cnp->cn_nameptr = cnp->cn_pnbuf;
622 	ndp->ni_lcf = 0;
623 	/*
624 	 * Extract and set starting directory.
625 	 */
626 	if (dp->v_type != VDIR) {
627 		if (islocked)
628 			vput(dp);
629 		else
630 			vrele(dp);
631 		nfsvno_relpathbuf(ndp);
632 		error = ENOTDIR;
633 		goto out1;
634 	}
635 	if (islocked)
636 		NFSVOPUNLOCK(dp);
637 	VREF(dp);
638 	*retdirp = dp;
639 	if (NFSVNO_EXRDONLY(exp))
640 		cnp->cn_flags |= RDONLY;
641 	ndp->ni_segflg = UIO_SYSSPACE;
642 
643 	if (nd->nd_flag & ND_PUBLOOKUP) {
644 		ndp->ni_loopcnt = 0;
645 		if (cnp->cn_pnbuf[0] == '/') {
646 			vrele(dp);
647 			/*
648 			 * Check for degenerate pathnames here, since lookup()
649 			 * panics on them.
650 			 */
651 			for (i = 1; i < ndp->ni_pathlen; i++)
652 				if (cnp->cn_pnbuf[i] != '/')
653 					break;
654 			if (i == ndp->ni_pathlen) {
655 				error = NFSERR_ACCES;
656 				goto out;
657 			}
658 			dp = rootvnode;
659 			VREF(dp);
660 		}
661 	} else if ((nfsrv_enable_crossmntpt == 0 && NFSVNO_EXPORTED(exp)) ||
662 	    (nd->nd_flag & ND_NFSV4) == 0) {
663 		/*
664 		 * Only cross mount points for NFSv4 when doing a
665 		 * mount while traversing the file system above
666 		 * the mount point, unless nfsrv_enable_crossmntpt is set.
667 		 */
668 		cnp->cn_flags |= NOCROSSMOUNT;
669 	}
670 
671 	/*
672 	 * Initialize for scan, set ni_startdir and bump ref on dp again
673 	 * because lookup() will dereference ni_startdir.
674 	 */
675 
676 	ndp->ni_startdir = dp;
677 	ndp->ni_rootdir = rootvnode;
678 	ndp->ni_topdir = NULL;
679 
680 	if (!lockleaf)
681 		cnp->cn_flags |= LOCKLEAF;
682 	for (;;) {
683 		cnp->cn_nameptr = cnp->cn_pnbuf;
684 		/*
685 		 * Call lookup() to do the real work.  If an error occurs,
686 		 * ndp->ni_vp and ni_dvp are left uninitialized or NULL and
687 		 * we do not have to dereference anything before returning.
688 		 * In either case ni_startdir will be dereferenced and NULLed
689 		 * out.
690 		 */
691 		error = vfs_lookup(ndp);
692 		if (error)
693 			break;
694 
695 		/*
696 		 * Check for encountering a symbolic link.  Trivial
697 		 * termination occurs if no symlink encountered.
698 		 */
699 		if ((cnp->cn_flags & ISSYMLINK) == 0) {
700 			if (ndp->ni_vp && !lockleaf)
701 				NFSVOPUNLOCK(ndp->ni_vp);
702 			break;
703 		}
704 
705 		/*
706 		 * Validate symlink
707 		 */
708 		if ((cnp->cn_flags & LOCKPARENT) && ndp->ni_pathlen == 1)
709 			NFSVOPUNLOCK(ndp->ni_dvp);
710 		if (!(nd->nd_flag & ND_PUBLOOKUP)) {
711 			error = EINVAL;
712 			goto badlink2;
713 		}
714 
715 		if (ndp->ni_loopcnt++ >= MAXSYMLINKS) {
716 			error = ELOOP;
717 			goto badlink2;
718 		}
719 		if (ndp->ni_pathlen > 1)
720 			cp = uma_zalloc(namei_zone, M_WAITOK);
721 		else
722 			cp = cnp->cn_pnbuf;
723 		aiov.iov_base = cp;
724 		aiov.iov_len = MAXPATHLEN;
725 		auio.uio_iov = &aiov;
726 		auio.uio_iovcnt = 1;
727 		auio.uio_offset = 0;
728 		auio.uio_rw = UIO_READ;
729 		auio.uio_segflg = UIO_SYSSPACE;
730 		auio.uio_td = NULL;
731 		auio.uio_resid = MAXPATHLEN;
732 		error = VOP_READLINK(ndp->ni_vp, &auio, cnp->cn_cred);
733 		if (error) {
734 		badlink1:
735 			if (ndp->ni_pathlen > 1)
736 				uma_zfree(namei_zone, cp);
737 		badlink2:
738 			vrele(ndp->ni_dvp);
739 			vput(ndp->ni_vp);
740 			break;
741 		}
742 		linklen = MAXPATHLEN - auio.uio_resid;
743 		if (linklen == 0) {
744 			error = ENOENT;
745 			goto badlink1;
746 		}
747 		if (linklen + ndp->ni_pathlen >= MAXPATHLEN) {
748 			error = ENAMETOOLONG;
749 			goto badlink1;
750 		}
751 
752 		/*
753 		 * Adjust or replace path
754 		 */
755 		if (ndp->ni_pathlen > 1) {
756 			NFSBCOPY(ndp->ni_next, cp + linklen, ndp->ni_pathlen);
757 			uma_zfree(namei_zone, cnp->cn_pnbuf);
758 			cnp->cn_pnbuf = cp;
759 		} else
760 			cnp->cn_pnbuf[linklen] = '\0';
761 		ndp->ni_pathlen += linklen;
762 
763 		/*
764 		 * Cleanup refs for next loop and check if root directory
765 		 * should replace current directory.  Normally ni_dvp
766 		 * becomes the new base directory and is cleaned up when
767 		 * we loop.  Explicitly null pointers after invalidation
768 		 * to clarify operation.
769 		 */
770 		vput(ndp->ni_vp);
771 		ndp->ni_vp = NULL;
772 
773 		if (cnp->cn_pnbuf[0] == '/') {
774 			vrele(ndp->ni_dvp);
775 			ndp->ni_dvp = ndp->ni_rootdir;
776 			VREF(ndp->ni_dvp);
777 		}
778 		ndp->ni_startdir = ndp->ni_dvp;
779 		ndp->ni_dvp = NULL;
780 	}
781 	if (!lockleaf)
782 		cnp->cn_flags &= ~LOCKLEAF;
783 
784 out:
785 	if (error) {
786 		nfsvno_relpathbuf(ndp);
787 		ndp->ni_vp = NULL;
788 		ndp->ni_dvp = NULL;
789 		ndp->ni_startdir = NULL;
790 	} else if ((ndp->ni_cnd.cn_flags & (WANTPARENT|LOCKPARENT)) == 0) {
791 		ndp->ni_dvp = NULL;
792 	}
793 
794 out1:
795 	NFSEXITCODE2(error, nd);
796 	return (error);
797 }
798 
799 /*
800  * Set up a pathname buffer and return a pointer to it and, optionally
801  * set a hash pointer.
802  */
803 void
804 nfsvno_setpathbuf(struct nameidata *ndp, char **bufpp, u_long **hashpp)
805 {
806 	struct componentname *cnp = &ndp->ni_cnd;
807 
808 	cnp->cn_flags |= (NOMACCHECK);
809 	cnp->cn_pnbuf = uma_zalloc(namei_zone, M_WAITOK);
810 	if (hashpp != NULL)
811 		*hashpp = NULL;
812 	*bufpp = cnp->cn_pnbuf;
813 }
814 
815 /*
816  * Release the above path buffer, if not released by nfsvno_namei().
817  */
818 void
819 nfsvno_relpathbuf(struct nameidata *ndp)
820 {
821 
822 	uma_zfree(namei_zone, ndp->ni_cnd.cn_pnbuf);
823 	ndp->ni_cnd.cn_pnbuf = NULL;
824 }
825 
826 /*
827  * Readlink vnode op into an mbuf list.
828  */
829 int
830 nfsvno_readlink(struct vnode *vp, struct ucred *cred, int maxextsiz,
831     struct thread *p, struct mbuf **mpp, struct mbuf **mpendp, int *lenp)
832 {
833 	struct iovec *iv;
834 	struct uio io, *uiop = &io;
835 	struct mbuf *mp, *mp3;
836 	int len, tlen, error = 0;
837 
838 	len = NFS_MAXPATHLEN;
839 	if (maxextsiz > 0)
840 		uiop->uio_iovcnt = nfsrv_createiovec_extpgs(len, maxextsiz,
841 		    &mp3, &mp, &iv);
842 	else
843 		uiop->uio_iovcnt = nfsrv_createiovec(len, &mp3, &mp, &iv);
844 	uiop->uio_iov = iv;
845 	uiop->uio_offset = 0;
846 	uiop->uio_resid = len;
847 	uiop->uio_rw = UIO_READ;
848 	uiop->uio_segflg = UIO_SYSSPACE;
849 	uiop->uio_td = NULL;
850 	error = VOP_READLINK(vp, uiop, cred);
851 	free(iv, M_TEMP);
852 	if (error) {
853 		m_freem(mp3);
854 		*lenp = 0;
855 		goto out;
856 	}
857 	if (uiop->uio_resid > 0) {
858 		len -= uiop->uio_resid;
859 		tlen = NFSM_RNDUP(len);
860 		if (tlen == 0) {
861 			m_freem(mp3);
862 			mp3 = mp = NULL;
863 		} else if (tlen != NFS_MAXPATHLEN || tlen != len)
864 			mp = nfsrv_adj(mp3, NFS_MAXPATHLEN - tlen,
865 			    tlen - len);
866 	}
867 	*lenp = len;
868 	*mpp = mp3;
869 	*mpendp = mp;
870 
871 out:
872 	NFSEXITCODE(error);
873 	return (error);
874 }
875 
876 /*
877  * Create an mbuf chain and an associated iovec that can be used to Read
878  * or Getextattr of data.
879  * Upon success, return pointers to the first and last mbufs in the chain
880  * plus the malloc'd iovec and its iovlen.
881  */
882 static int
883 nfsrv_createiovec(int len, struct mbuf **mpp, struct mbuf **mpendp,
884     struct iovec **ivp)
885 {
886 	struct mbuf *m, *m2 = NULL, *m3;
887 	struct iovec *iv;
888 	int i, left, siz;
889 
890 	left = len;
891 	m3 = NULL;
892 	/*
893 	 * Generate the mbuf list with the uio_iov ref. to it.
894 	 */
895 	i = 0;
896 	while (left > 0) {
897 		NFSMGET(m);
898 		MCLGET(m, M_WAITOK);
899 		m->m_len = 0;
900 		siz = min(M_TRAILINGSPACE(m), left);
901 		left -= siz;
902 		i++;
903 		if (m3)
904 			m2->m_next = m;
905 		else
906 			m3 = m;
907 		m2 = m;
908 	}
909 	*ivp = iv = malloc(i * sizeof (struct iovec), M_TEMP, M_WAITOK);
910 	m = m3;
911 	left = len;
912 	i = 0;
913 	while (left > 0) {
914 		if (m == NULL)
915 			panic("nfsrv_createiovec iov");
916 		siz = min(M_TRAILINGSPACE(m), left);
917 		if (siz > 0) {
918 			iv->iov_base = mtod(m, caddr_t) + m->m_len;
919 			iv->iov_len = siz;
920 			m->m_len += siz;
921 			left -= siz;
922 			iv++;
923 			i++;
924 		}
925 		m = m->m_next;
926 	}
927 	*mpp = m3;
928 	*mpendp = m2;
929 	return (i);
930 }
931 
932 /*
933  * Create an mbuf chain and an associated iovec that can be used to Read
934  * or Getextattr of data.
935  * Upon success, return pointers to the first and last mbufs in the chain
936  * plus the malloc'd iovec and its iovlen.
937  * Same as above, but creates ext_pgs mbuf(s).
938  */
939 static int
940 nfsrv_createiovec_extpgs(int len, int maxextsiz, struct mbuf **mpp,
941     struct mbuf **mpendp, struct iovec **ivp)
942 {
943 	struct mbuf *m, *m2 = NULL, *m3;
944 	struct iovec *iv;
945 	int i, left, pgno, siz;
946 
947 	left = len;
948 	m3 = NULL;
949 	/*
950 	 * Generate the mbuf list with the uio_iov ref. to it.
951 	 */
952 	i = 0;
953 	while (left > 0) {
954 		siz = min(left, maxextsiz);
955 		m = mb_alloc_ext_plus_pages(siz, M_WAITOK);
956 		left -= siz;
957 		i += m->m_epg_npgs;
958 		if (m3 != NULL)
959 			m2->m_next = m;
960 		else
961 			m3 = m;
962 		m2 = m;
963 	}
964 	*ivp = iv = malloc(i * sizeof (struct iovec), M_TEMP, M_WAITOK);
965 	m = m3;
966 	left = len;
967 	i = 0;
968 	pgno = 0;
969 	while (left > 0) {
970 		if (m == NULL)
971 			panic("nfsvno_createiovec_extpgs iov");
972 		siz = min(PAGE_SIZE, left);
973 		if (siz > 0) {
974 			iv->iov_base = (void *)PHYS_TO_DMAP(m->m_epg_pa[pgno]);
975 			iv->iov_len = siz;
976 			m->m_len += siz;
977 			if (pgno == m->m_epg_npgs - 1)
978 				m->m_epg_last_len = siz;
979 			left -= siz;
980 			iv++;
981 			i++;
982 			pgno++;
983 		}
984 		if (pgno == m->m_epg_npgs && left > 0) {
985 			m = m->m_next;
986 			if (m == NULL)
987 				panic("nfsvno_createiovec_extpgs iov");
988 			pgno = 0;
989 		}
990 	}
991 	*mpp = m3;
992 	*mpendp = m2;
993 	return (i);
994 }
995 
996 /*
997  * Read vnode op call into mbuf list.
998  */
999 int
1000 nfsvno_read(struct vnode *vp, off_t off, int cnt, struct ucred *cred,
1001     int maxextsiz, struct thread *p, struct mbuf **mpp,
1002     struct mbuf **mpendp)
1003 {
1004 	struct mbuf *m;
1005 	struct iovec *iv;
1006 	int error = 0, len, tlen, ioflag = 0;
1007 	struct mbuf *m3;
1008 	struct uio io, *uiop = &io;
1009 	struct nfsheur *nh;
1010 
1011 	/*
1012 	 * Attempt to read from a DS file. A return of ENOENT implies
1013 	 * there is no DS file to read.
1014 	 */
1015 	error = nfsrv_proxyds(vp, off, cnt, cred, p, NFSPROC_READDS, mpp,
1016 	    NULL, mpendp, NULL, NULL, NULL, 0, NULL);
1017 	if (error != ENOENT)
1018 		return (error);
1019 
1020 	len = NFSM_RNDUP(cnt);
1021 	if (maxextsiz > 0)
1022 		uiop->uio_iovcnt = nfsrv_createiovec_extpgs(len, maxextsiz,
1023 		    &m3, &m, &iv);
1024 	else
1025 		uiop->uio_iovcnt = nfsrv_createiovec(len, &m3, &m, &iv);
1026 	uiop->uio_iov = iv;
1027 	uiop->uio_offset = off;
1028 	uiop->uio_resid = len;
1029 	uiop->uio_rw = UIO_READ;
1030 	uiop->uio_segflg = UIO_SYSSPACE;
1031 	uiop->uio_td = NULL;
1032 	nh = nfsrv_sequential_heuristic(uiop, vp);
1033 	ioflag |= nh->nh_seqcount << IO_SEQSHIFT;
1034 	/* XXX KDM make this more systematic? */
1035 	NFSD_VNET(nfsstatsv1_p)->srvbytes[NFSV4OP_READ] += uiop->uio_resid;
1036 	error = VOP_READ(vp, uiop, IO_NODELOCKED | ioflag, cred);
1037 	free(iv, M_TEMP);
1038 	if (error) {
1039 		m_freem(m3);
1040 		*mpp = NULL;
1041 		goto out;
1042 	}
1043 	nh->nh_nextoff = uiop->uio_offset;
1044 	tlen = len - uiop->uio_resid;
1045 	cnt = cnt < tlen ? cnt : tlen;
1046 	tlen = NFSM_RNDUP(cnt);
1047 	if (tlen == 0) {
1048 		m_freem(m3);
1049 		m3 = m = NULL;
1050 	} else if (len != tlen || tlen != cnt)
1051 		m = nfsrv_adj(m3, len - tlen, tlen - cnt);
1052 	*mpp = m3;
1053 	*mpendp = m;
1054 
1055 out:
1056 	NFSEXITCODE(error);
1057 	return (error);
1058 }
1059 
1060 /*
1061  * Create the iovec for the mbuf chain passed in as an argument.
1062  * The "cp" argument is where the data starts within the first mbuf in
1063  * the chain. It returns the iovec and the iovcnt.
1064  */
1065 static int
1066 nfsrv_createiovecw(int retlen, struct mbuf *m, char *cp, struct iovec **ivpp,
1067     int *iovcntp)
1068 {
1069 	struct mbuf *mp;
1070 	struct iovec *ivp;
1071 	int cnt, i, len;
1072 
1073 	/*
1074 	 * Loop through the mbuf chain, counting how many mbufs are a
1075 	 * part of this write operation, so the iovec size is known.
1076 	 */
1077 	cnt = 0;
1078 	len = retlen;
1079 	mp = m;
1080 	i = mtod(mp, caddr_t) + mp->m_len - cp;
1081 	while (len > 0) {
1082 		if (i > 0) {
1083 			len -= i;
1084 			cnt++;
1085 		}
1086 		mp = mp->m_next;
1087 		if (!mp) {
1088 			if (len > 0)
1089 				return (EBADRPC);
1090 		} else
1091 			i = mp->m_len;
1092 	}
1093 
1094 	/* Now, create the iovec. */
1095 	mp = m;
1096 	*ivpp = ivp = malloc(cnt * sizeof (struct iovec), M_TEMP,
1097 	    M_WAITOK);
1098 	*iovcntp = cnt;
1099 	i = mtod(mp, caddr_t) + mp->m_len - cp;
1100 	len = retlen;
1101 	while (len > 0) {
1102 		if (mp == NULL)
1103 			panic("nfsrv_createiovecw");
1104 		if (i > 0) {
1105 			i = min(i, len);
1106 			ivp->iov_base = cp;
1107 			ivp->iov_len = i;
1108 			ivp++;
1109 			len -= i;
1110 		}
1111 		mp = mp->m_next;
1112 		if (mp) {
1113 			i = mp->m_len;
1114 			cp = mtod(mp, caddr_t);
1115 		}
1116 	}
1117 	return (0);
1118 }
1119 
1120 /*
1121  * Write vnode op from an mbuf list.
1122  */
1123 int
1124 nfsvno_write(struct vnode *vp, off_t off, int retlen, int *stable,
1125     struct mbuf *mp, char *cp, struct ucred *cred, struct thread *p)
1126 {
1127 	struct iovec *iv;
1128 	int cnt, ioflags, error;
1129 	struct uio io, *uiop = &io;
1130 	struct nfsheur *nh;
1131 
1132 	/*
1133 	 * Attempt to write to a DS file. A return of ENOENT implies
1134 	 * there is no DS file to write.
1135 	 */
1136 	error = nfsrv_proxyds(vp, off, retlen, cred, p, NFSPROC_WRITEDS,
1137 	    &mp, cp, NULL, NULL, NULL, NULL, 0, NULL);
1138 	if (error != ENOENT) {
1139 		*stable = NFSWRITE_FILESYNC;
1140 		return (error);
1141 	}
1142 
1143 	if (*stable == NFSWRITE_UNSTABLE)
1144 		ioflags = IO_NODELOCKED;
1145 	else
1146 		ioflags = (IO_SYNC | IO_NODELOCKED);
1147 	error = nfsrv_createiovecw(retlen, mp, cp, &iv, &cnt);
1148 	if (error != 0)
1149 		return (error);
1150 	uiop->uio_iov = iv;
1151 	uiop->uio_iovcnt = cnt;
1152 	uiop->uio_resid = retlen;
1153 	uiop->uio_rw = UIO_WRITE;
1154 	uiop->uio_segflg = UIO_SYSSPACE;
1155 	NFSUIOPROC(uiop, p);
1156 	uiop->uio_offset = off;
1157 	nh = nfsrv_sequential_heuristic(uiop, vp);
1158 	ioflags |= nh->nh_seqcount << IO_SEQSHIFT;
1159 	/* XXX KDM make this more systematic? */
1160 	NFSD_VNET(nfsstatsv1_p)->srvbytes[NFSV4OP_WRITE] += uiop->uio_resid;
1161 	error = VOP_WRITE(vp, uiop, ioflags, cred);
1162 	if (error == 0)
1163 		nh->nh_nextoff = uiop->uio_offset;
1164 	free(iv, M_TEMP);
1165 
1166 	NFSEXITCODE(error);
1167 	return (error);
1168 }
1169 
1170 /*
1171  * Common code for creating a regular file (plus special files for V2).
1172  */
1173 int
1174 nfsvno_createsub(struct nfsrv_descript *nd, struct nameidata *ndp,
1175     struct vnode **vpp, struct nfsvattr *nvap, int *exclusive_flagp,
1176     int32_t *cverf, NFSDEV_T rdev, struct nfsexstuff *exp)
1177 {
1178 	u_quad_t tempsize;
1179 	int error;
1180 	struct thread *p = curthread;
1181 
1182 	error = nd->nd_repstat;
1183 	if (!error && ndp->ni_vp == NULL) {
1184 		if (nvap->na_type == VREG || nvap->na_type == VSOCK) {
1185 			error = VOP_CREATE(ndp->ni_dvp,
1186 			    &ndp->ni_vp, &ndp->ni_cnd, &nvap->na_vattr);
1187 			/* For a pNFS server, create the data file on a DS. */
1188 			if (error == 0 && nvap->na_type == VREG) {
1189 				/*
1190 				 * Create a data file on a DS for a pNFS server.
1191 				 * This function just returns if not
1192 				 * running a pNFS DS or the creation fails.
1193 				 */
1194 				nfsrv_pnfscreate(ndp->ni_vp, &nvap->na_vattr,
1195 				    nd->nd_cred, p);
1196 			}
1197 			VOP_VPUT_PAIR(ndp->ni_dvp, error == 0 ? &ndp->ni_vp :
1198 			    NULL, false);
1199 			nfsvno_relpathbuf(ndp);
1200 			if (!error) {
1201 				if (*exclusive_flagp) {
1202 					*exclusive_flagp = 0;
1203 					NFSVNO_ATTRINIT(nvap);
1204 					nvap->na_atime.tv_sec = cverf[0];
1205 					nvap->na_atime.tv_nsec = cverf[1];
1206 					error = VOP_SETATTR(ndp->ni_vp,
1207 					    &nvap->na_vattr, nd->nd_cred);
1208 					if (error != 0) {
1209 						vput(ndp->ni_vp);
1210 						ndp->ni_vp = NULL;
1211 						error = NFSERR_NOTSUPP;
1212 					}
1213 				}
1214 			}
1215 		/*
1216 		 * NFS V2 Only. nfsrvd_mknod() does this for V3.
1217 		 * (This implies, just get out on an error.)
1218 		 */
1219 		} else if (nvap->na_type == VCHR || nvap->na_type == VBLK ||
1220 			nvap->na_type == VFIFO) {
1221 			if (nvap->na_type == VCHR && rdev == 0xffffffff)
1222 				nvap->na_type = VFIFO;
1223                         if (nvap->na_type != VFIFO &&
1224 			    (error = priv_check_cred(nd->nd_cred, PRIV_VFS_MKNOD_DEV))) {
1225 				nfsvno_relpathbuf(ndp);
1226 				vput(ndp->ni_dvp);
1227 				goto out;
1228 			}
1229 			nvap->na_rdev = rdev;
1230 			error = VOP_MKNOD(ndp->ni_dvp, &ndp->ni_vp,
1231 			    &ndp->ni_cnd, &nvap->na_vattr);
1232 			VOP_VPUT_PAIR(ndp->ni_dvp, error == 0 ? &ndp->ni_vp :
1233 			    NULL, false);
1234 			nfsvno_relpathbuf(ndp);
1235 			if (error)
1236 				goto out;
1237 		} else {
1238 			nfsvno_relpathbuf(ndp);
1239 			vput(ndp->ni_dvp);
1240 			error = ENXIO;
1241 			goto out;
1242 		}
1243 		*vpp = ndp->ni_vp;
1244 	} else {
1245 		/*
1246 		 * Handle cases where error is already set and/or
1247 		 * the file exists.
1248 		 * 1 - clean up the lookup
1249 		 * 2 - iff !error and na_size set, truncate it
1250 		 */
1251 		nfsvno_relpathbuf(ndp);
1252 		*vpp = ndp->ni_vp;
1253 		if (ndp->ni_dvp == *vpp)
1254 			vrele(ndp->ni_dvp);
1255 		else
1256 			vput(ndp->ni_dvp);
1257 		if (!error && nvap->na_size != VNOVAL) {
1258 			error = nfsvno_accchk(*vpp, VWRITE,
1259 			    nd->nd_cred, exp, p, NFSACCCHK_NOOVERRIDE,
1260 			    NFSACCCHK_VPISLOCKED, NULL);
1261 			if (!error) {
1262 				tempsize = nvap->na_size;
1263 				NFSVNO_ATTRINIT(nvap);
1264 				nvap->na_size = tempsize;
1265 				error = nfsvno_setattr(*vpp, nvap,
1266 				    nd->nd_cred, p, exp);
1267 			}
1268 		}
1269 		if (error)
1270 			vput(*vpp);
1271 	}
1272 
1273 out:
1274 	NFSEXITCODE(error);
1275 	return (error);
1276 }
1277 
1278 /*
1279  * Do a mknod vnode op.
1280  */
1281 int
1282 nfsvno_mknod(struct nameidata *ndp, struct nfsvattr *nvap, struct ucred *cred,
1283     struct thread *p)
1284 {
1285 	int error = 0;
1286 	enum vtype vtyp;
1287 
1288 	vtyp = nvap->na_type;
1289 	/*
1290 	 * Iff doesn't exist, create it.
1291 	 */
1292 	if (ndp->ni_vp) {
1293 		nfsvno_relpathbuf(ndp);
1294 		vput(ndp->ni_dvp);
1295 		vrele(ndp->ni_vp);
1296 		error = EEXIST;
1297 		goto out;
1298 	}
1299 	if (vtyp != VCHR && vtyp != VBLK && vtyp != VSOCK && vtyp != VFIFO) {
1300 		nfsvno_relpathbuf(ndp);
1301 		vput(ndp->ni_dvp);
1302 		error = NFSERR_BADTYPE;
1303 		goto out;
1304 	}
1305 	if (vtyp == VSOCK) {
1306 		error = VOP_CREATE(ndp->ni_dvp, &ndp->ni_vp,
1307 		    &ndp->ni_cnd, &nvap->na_vattr);
1308 		VOP_VPUT_PAIR(ndp->ni_dvp, error == 0 ? &ndp->ni_vp : NULL,
1309 		    false);
1310 		nfsvno_relpathbuf(ndp);
1311 	} else {
1312 		if (nvap->na_type != VFIFO &&
1313 		    (error = priv_check_cred(cred, PRIV_VFS_MKNOD_DEV))) {
1314 			nfsvno_relpathbuf(ndp);
1315 			vput(ndp->ni_dvp);
1316 			goto out;
1317 		}
1318 		error = VOP_MKNOD(ndp->ni_dvp, &ndp->ni_vp,
1319 		    &ndp->ni_cnd, &nvap->na_vattr);
1320 		VOP_VPUT_PAIR(ndp->ni_dvp, error == 0 ? &ndp->ni_vp : NULL,
1321 		    false);
1322 		nfsvno_relpathbuf(ndp);
1323 		/*
1324 		 * Since VOP_MKNOD returns the ni_vp, I can't
1325 		 * see any reason to do the lookup.
1326 		 */
1327 	}
1328 
1329 out:
1330 	NFSEXITCODE(error);
1331 	return (error);
1332 }
1333 
1334 /*
1335  * Mkdir vnode op.
1336  */
1337 int
1338 nfsvno_mkdir(struct nameidata *ndp, struct nfsvattr *nvap, uid_t saved_uid,
1339     struct ucred *cred, struct thread *p, struct nfsexstuff *exp)
1340 {
1341 	int error = 0;
1342 
1343 	if (ndp->ni_vp != NULL) {
1344 		if (ndp->ni_dvp == ndp->ni_vp)
1345 			vrele(ndp->ni_dvp);
1346 		else
1347 			vput(ndp->ni_dvp);
1348 		vrele(ndp->ni_vp);
1349 		nfsvno_relpathbuf(ndp);
1350 		error = EEXIST;
1351 		goto out;
1352 	}
1353 	error = VOP_MKDIR(ndp->ni_dvp, &ndp->ni_vp, &ndp->ni_cnd,
1354 	    &nvap->na_vattr);
1355 	VOP_VPUT_PAIR(ndp->ni_dvp, error == 0 ? &ndp->ni_vp : NULL, false);
1356 	nfsvno_relpathbuf(ndp);
1357 
1358 out:
1359 	NFSEXITCODE(error);
1360 	return (error);
1361 }
1362 
1363 /*
1364  * symlink vnode op.
1365  */
1366 int
1367 nfsvno_symlink(struct nameidata *ndp, struct nfsvattr *nvap, char *pathcp,
1368     int pathlen, int not_v2, uid_t saved_uid, struct ucred *cred, struct thread *p,
1369     struct nfsexstuff *exp)
1370 {
1371 	int error = 0;
1372 
1373 	if (ndp->ni_vp) {
1374 		nfsvno_relpathbuf(ndp);
1375 		if (ndp->ni_dvp == ndp->ni_vp)
1376 			vrele(ndp->ni_dvp);
1377 		else
1378 			vput(ndp->ni_dvp);
1379 		vrele(ndp->ni_vp);
1380 		error = EEXIST;
1381 		goto out;
1382 	}
1383 
1384 	error = VOP_SYMLINK(ndp->ni_dvp, &ndp->ni_vp, &ndp->ni_cnd,
1385 	    &nvap->na_vattr, pathcp);
1386 	/*
1387 	 * Although FreeBSD still had the lookup code in
1388 	 * it for 7/current, there doesn't seem to be any
1389 	 * point, since VOP_SYMLINK() returns the ni_vp.
1390 	 * Just vput it for v2.
1391 	 */
1392 	VOP_VPUT_PAIR(ndp->ni_dvp, &ndp->ni_vp, !not_v2 && error == 0);
1393 	nfsvno_relpathbuf(ndp);
1394 
1395 out:
1396 	NFSEXITCODE(error);
1397 	return (error);
1398 }
1399 
1400 /*
1401  * Parse symbolic link arguments.
1402  * This function has an ugly side effect. It will malloc() an area for
1403  * the symlink and set iov_base to point to it, only if it succeeds.
1404  * So, if it returns with uiop->uio_iov->iov_base != NULL, that must
1405  * be FREE'd later.
1406  */
1407 int
1408 nfsvno_getsymlink(struct nfsrv_descript *nd, struct nfsvattr *nvap,
1409     struct thread *p, char **pathcpp, int *lenp)
1410 {
1411 	u_int32_t *tl;
1412 	char *pathcp = NULL;
1413 	int error = 0, len;
1414 	struct nfsv2_sattr *sp;
1415 
1416 	*pathcpp = NULL;
1417 	*lenp = 0;
1418 	if ((nd->nd_flag & ND_NFSV3) &&
1419 	    (error = nfsrv_sattr(nd, NULL, nvap, NULL, NULL, p)))
1420 		goto nfsmout;
1421 	NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
1422 	len = fxdr_unsigned(int, *tl);
1423 	if (len > NFS_MAXPATHLEN || len <= 0) {
1424 		error = EBADRPC;
1425 		goto nfsmout;
1426 	}
1427 	pathcp = malloc(len + 1, M_TEMP, M_WAITOK);
1428 	error = nfsrv_mtostr(nd, pathcp, len);
1429 	if (error)
1430 		goto nfsmout;
1431 	if (nd->nd_flag & ND_NFSV2) {
1432 		NFSM_DISSECT(sp, struct nfsv2_sattr *, NFSX_V2SATTR);
1433 		nvap->na_mode = fxdr_unsigned(u_int16_t, sp->sa_mode);
1434 	}
1435 	*pathcpp = pathcp;
1436 	*lenp = len;
1437 	NFSEXITCODE2(0, nd);
1438 	return (0);
1439 nfsmout:
1440 	if (pathcp)
1441 		free(pathcp, M_TEMP);
1442 	NFSEXITCODE2(error, nd);
1443 	return (error);
1444 }
1445 
1446 /*
1447  * Remove a non-directory object.
1448  */
1449 int
1450 nfsvno_removesub(struct nameidata *ndp, int is_v4, struct ucred *cred,
1451     struct thread *p, struct nfsexstuff *exp)
1452 {
1453 	struct vnode *vp, *dsdvp[NFSDEV_MAXMIRRORS];
1454 	int error = 0, mirrorcnt;
1455 	char fname[PNFS_FILENAME_LEN + 1];
1456 	fhandle_t fh;
1457 
1458 	vp = ndp->ni_vp;
1459 	dsdvp[0] = NULL;
1460 	if (vp->v_type == VDIR)
1461 		error = NFSERR_ISDIR;
1462 	else if (is_v4)
1463 		error = nfsrv_checkremove(vp, 1, NULL, (nfsquad_t)((u_quad_t)0),
1464 		    p);
1465 	if (error == 0)
1466 		nfsrv_pnfsremovesetup(vp, p, dsdvp, &mirrorcnt, fname, &fh);
1467 	if (!error)
1468 		error = VOP_REMOVE(ndp->ni_dvp, vp, &ndp->ni_cnd);
1469 	if (error == 0 && dsdvp[0] != NULL)
1470 		nfsrv_pnfsremove(dsdvp, mirrorcnt, fname, &fh, p);
1471 	if (ndp->ni_dvp == vp)
1472 		vrele(ndp->ni_dvp);
1473 	else
1474 		vput(ndp->ni_dvp);
1475 	vput(vp);
1476 	nfsvno_relpathbuf(ndp);
1477 	NFSEXITCODE(error);
1478 	return (error);
1479 }
1480 
1481 /*
1482  * Remove a directory.
1483  */
1484 int
1485 nfsvno_rmdirsub(struct nameidata *ndp, int is_v4, struct ucred *cred,
1486     struct thread *p, struct nfsexstuff *exp)
1487 {
1488 	struct vnode *vp;
1489 	int error = 0;
1490 
1491 	vp = ndp->ni_vp;
1492 	if (vp->v_type != VDIR) {
1493 		error = ENOTDIR;
1494 		goto out;
1495 	}
1496 	/*
1497 	 * No rmdir "." please.
1498 	 */
1499 	if (ndp->ni_dvp == vp) {
1500 		error = EINVAL;
1501 		goto out;
1502 	}
1503 	/*
1504 	 * The root of a mounted filesystem cannot be deleted.
1505 	 */
1506 	if (vp->v_vflag & VV_ROOT)
1507 		error = EBUSY;
1508 out:
1509 	if (!error)
1510 		error = VOP_RMDIR(ndp->ni_dvp, vp, &ndp->ni_cnd);
1511 	if (ndp->ni_dvp == vp)
1512 		vrele(ndp->ni_dvp);
1513 	else
1514 		vput(ndp->ni_dvp);
1515 	vput(vp);
1516 	nfsvno_relpathbuf(ndp);
1517 	NFSEXITCODE(error);
1518 	return (error);
1519 }
1520 
1521 /*
1522  * Rename vnode op.
1523  */
1524 int
1525 nfsvno_rename(struct nameidata *fromndp, struct nameidata *tondp,
1526     u_int32_t ndstat, u_int32_t ndflag, struct ucred *cred, struct thread *p)
1527 {
1528 	struct vnode *fvp, *tvp, *tdvp, *dsdvp[NFSDEV_MAXMIRRORS];
1529 	int error = 0, mirrorcnt;
1530 	char fname[PNFS_FILENAME_LEN + 1];
1531 	fhandle_t fh;
1532 
1533 	dsdvp[0] = NULL;
1534 	fvp = fromndp->ni_vp;
1535 	if (ndstat) {
1536 		vrele(fromndp->ni_dvp);
1537 		vrele(fvp);
1538 		error = ndstat;
1539 		goto out1;
1540 	}
1541 	tdvp = tondp->ni_dvp;
1542 	tvp = tondp->ni_vp;
1543 	if (tvp != NULL) {
1544 		if (fvp->v_type == VDIR && tvp->v_type != VDIR) {
1545 			error = (ndflag & ND_NFSV2) ? EISDIR : EEXIST;
1546 			goto out;
1547 		} else if (fvp->v_type != VDIR && tvp->v_type == VDIR) {
1548 			error = (ndflag & ND_NFSV2) ? ENOTDIR : EEXIST;
1549 			goto out;
1550 		}
1551 		if (tvp->v_type == VDIR && tvp->v_mountedhere) {
1552 			error = (ndflag & ND_NFSV2) ? ENOTEMPTY : EXDEV;
1553 			goto out;
1554 		}
1555 
1556 		/*
1557 		 * A rename to '.' or '..' results in a prematurely
1558 		 * unlocked vnode on FreeBSD5, so I'm just going to fail that
1559 		 * here.
1560 		 */
1561 		if ((tondp->ni_cnd.cn_namelen == 1 &&
1562 		     tondp->ni_cnd.cn_nameptr[0] == '.') ||
1563 		    (tondp->ni_cnd.cn_namelen == 2 &&
1564 		     tondp->ni_cnd.cn_nameptr[0] == '.' &&
1565 		     tondp->ni_cnd.cn_nameptr[1] == '.')) {
1566 			error = EINVAL;
1567 			goto out;
1568 		}
1569 	}
1570 	if (fvp->v_type == VDIR && fvp->v_mountedhere) {
1571 		error = (ndflag & ND_NFSV2) ? ENOTEMPTY : EXDEV;
1572 		goto out;
1573 	}
1574 	if (fvp->v_mount != tdvp->v_mount) {
1575 		error = (ndflag & ND_NFSV2) ? ENOTEMPTY : EXDEV;
1576 		goto out;
1577 	}
1578 	if (fvp == tdvp) {
1579 		error = (ndflag & ND_NFSV2) ? ENOTEMPTY : EINVAL;
1580 		goto out;
1581 	}
1582 	if (fvp == tvp) {
1583 		/*
1584 		 * If source and destination are the same, there is nothing to
1585 		 * do. Set error to -1 to indicate this.
1586 		 */
1587 		error = -1;
1588 		goto out;
1589 	}
1590 	if (ndflag & ND_NFSV4) {
1591 		if (NFSVOPLOCK(fvp, LK_EXCLUSIVE) == 0) {
1592 			error = nfsrv_checkremove(fvp, 0, NULL,
1593 			    (nfsquad_t)((u_quad_t)0), p);
1594 			NFSVOPUNLOCK(fvp);
1595 		} else
1596 			error = EPERM;
1597 		if (tvp && !error)
1598 			error = nfsrv_checkremove(tvp, 1, NULL,
1599 			    (nfsquad_t)((u_quad_t)0), p);
1600 	} else {
1601 		/*
1602 		 * For NFSv2 and NFSv3, try to get rid of the delegation, so
1603 		 * that the NFSv4 client won't be confused by the rename.
1604 		 * Since nfsd_recalldelegation() can only be called on an
1605 		 * unlocked vnode at this point and fvp is the file that will
1606 		 * still exist after the rename, just do fvp.
1607 		 */
1608 		nfsd_recalldelegation(fvp, p);
1609 	}
1610 	if (error == 0 && tvp != NULL) {
1611 		nfsrv_pnfsremovesetup(tvp, p, dsdvp, &mirrorcnt, fname, &fh);
1612 		NFSD_DEBUG(4, "nfsvno_rename: pnfsremovesetup"
1613 		    " dsdvp=%p\n", dsdvp[0]);
1614 	}
1615 out:
1616 	if (!error) {
1617 		error = VOP_RENAME(fromndp->ni_dvp, fromndp->ni_vp,
1618 		    &fromndp->ni_cnd, tondp->ni_dvp, tondp->ni_vp,
1619 		    &tondp->ni_cnd);
1620 	} else {
1621 		if (tdvp == tvp)
1622 			vrele(tdvp);
1623 		else
1624 			vput(tdvp);
1625 		if (tvp)
1626 			vput(tvp);
1627 		vrele(fromndp->ni_dvp);
1628 		vrele(fvp);
1629 		if (error == -1)
1630 			error = 0;
1631 	}
1632 
1633 	/*
1634 	 * If dsdvp[0] != NULL, it was set up by nfsrv_pnfsremovesetup() and
1635 	 * if the rename succeeded, the DS file for the tvp needs to be
1636 	 * removed.
1637 	 */
1638 	if (error == 0 && dsdvp[0] != NULL) {
1639 		nfsrv_pnfsremove(dsdvp, mirrorcnt, fname, &fh, p);
1640 		NFSD_DEBUG(4, "nfsvno_rename: pnfsremove\n");
1641 	}
1642 
1643 	nfsvno_relpathbuf(tondp);
1644 out1:
1645 	nfsvno_relpathbuf(fromndp);
1646 	NFSEXITCODE(error);
1647 	return (error);
1648 }
1649 
1650 /*
1651  * Link vnode op.
1652  */
1653 int
1654 nfsvno_link(struct nameidata *ndp, struct vnode *vp, struct ucred *cred,
1655     struct thread *p, struct nfsexstuff *exp)
1656 {
1657 	struct vnode *xp;
1658 	int error = 0;
1659 
1660 	xp = ndp->ni_vp;
1661 	if (xp != NULL) {
1662 		error = EEXIST;
1663 	} else {
1664 		xp = ndp->ni_dvp;
1665 		if (vp->v_mount != xp->v_mount)
1666 			error = EXDEV;
1667 	}
1668 	if (!error) {
1669 		NFSVOPLOCK(vp, LK_EXCLUSIVE | LK_RETRY);
1670 		if (!VN_IS_DOOMED(vp))
1671 			error = VOP_LINK(ndp->ni_dvp, vp, &ndp->ni_cnd);
1672 		else
1673 			error = EPERM;
1674 		if (ndp->ni_dvp == vp) {
1675 			vrele(ndp->ni_dvp);
1676 			NFSVOPUNLOCK(vp);
1677 		} else {
1678 			vref(vp);
1679 			VOP_VPUT_PAIR(ndp->ni_dvp, &vp, true);
1680 		}
1681 	} else {
1682 		if (ndp->ni_dvp == ndp->ni_vp)
1683 			vrele(ndp->ni_dvp);
1684 		else
1685 			vput(ndp->ni_dvp);
1686 		if (ndp->ni_vp)
1687 			vrele(ndp->ni_vp);
1688 	}
1689 	nfsvno_relpathbuf(ndp);
1690 	NFSEXITCODE(error);
1691 	return (error);
1692 }
1693 
1694 /*
1695  * Do the fsync() appropriate for the commit.
1696  */
1697 int
1698 nfsvno_fsync(struct vnode *vp, u_int64_t off, int cnt, struct ucred *cred,
1699     struct thread *td)
1700 {
1701 	int error = 0;
1702 
1703 	/*
1704 	 * RFC 1813 3.3.21: if count is 0, a flush from offset to the end of
1705 	 * file is done.  At this time VOP_FSYNC does not accept offset and
1706 	 * byte count parameters so call VOP_FSYNC the whole file for now.
1707 	 * The same is true for NFSv4: RFC 3530 Sec. 14.2.3.
1708 	 * File systems that do not use the buffer cache (as indicated
1709 	 * by MNTK_USES_BCACHE not being set) must use VOP_FSYNC().
1710 	 */
1711 	if (cnt == 0 || cnt > MAX_COMMIT_COUNT ||
1712 	    (vp->v_mount->mnt_kern_flag & MNTK_USES_BCACHE) == 0) {
1713 		/*
1714 		 * Give up and do the whole thing
1715 		 */
1716 		if (vp->v_object && vm_object_mightbedirty(vp->v_object)) {
1717 			VM_OBJECT_WLOCK(vp->v_object);
1718 			vm_object_page_clean(vp->v_object, 0, 0, OBJPC_SYNC);
1719 			VM_OBJECT_WUNLOCK(vp->v_object);
1720 		}
1721 		error = VOP_FSYNC(vp, MNT_WAIT, td);
1722 	} else {
1723 		/*
1724 		 * Locate and synchronously write any buffers that fall
1725 		 * into the requested range.  Note:  we are assuming that
1726 		 * f_iosize is a power of 2.
1727 		 */
1728 		int iosize = vp->v_mount->mnt_stat.f_iosize;
1729 		int iomask = iosize - 1;
1730 		struct bufobj *bo;
1731 		daddr_t lblkno;
1732 
1733 		/*
1734 		 * Align to iosize boundary, super-align to page boundary.
1735 		 */
1736 		if (off & iomask) {
1737 			cnt += off & iomask;
1738 			off &= ~(u_quad_t)iomask;
1739 		}
1740 		if (off & PAGE_MASK) {
1741 			cnt += off & PAGE_MASK;
1742 			off &= ~(u_quad_t)PAGE_MASK;
1743 		}
1744 		lblkno = off / iosize;
1745 
1746 		if (vp->v_object && vm_object_mightbedirty(vp->v_object)) {
1747 			VM_OBJECT_WLOCK(vp->v_object);
1748 			vm_object_page_clean(vp->v_object, off, off + cnt,
1749 			    OBJPC_SYNC);
1750 			VM_OBJECT_WUNLOCK(vp->v_object);
1751 		}
1752 
1753 		bo = &vp->v_bufobj;
1754 		BO_LOCK(bo);
1755 		while (cnt > 0) {
1756 			struct buf *bp;
1757 
1758 			/*
1759 			 * If we have a buffer and it is marked B_DELWRI we
1760 			 * have to lock and write it.  Otherwise the prior
1761 			 * write is assumed to have already been committed.
1762 			 *
1763 			 * gbincore() can return invalid buffers now so we
1764 			 * have to check that bit as well (though B_DELWRI
1765 			 * should not be set if B_INVAL is set there could be
1766 			 * a race here since we haven't locked the buffer).
1767 			 */
1768 			if ((bp = gbincore(&vp->v_bufobj, lblkno)) != NULL) {
1769 				if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_SLEEPFAIL |
1770 				    LK_INTERLOCK, BO_LOCKPTR(bo)) == ENOLCK) {
1771 					BO_LOCK(bo);
1772 					continue; /* retry */
1773 				}
1774 			    	if ((bp->b_flags & (B_DELWRI|B_INVAL)) ==
1775 				    B_DELWRI) {
1776 					bremfree(bp);
1777 					bp->b_flags &= ~B_ASYNC;
1778 					bwrite(bp);
1779 					++nfs_commit_miss;
1780 				} else
1781 					BUF_UNLOCK(bp);
1782 				BO_LOCK(bo);
1783 			}
1784 			++nfs_commit_blks;
1785 			if (cnt < iosize)
1786 				break;
1787 			cnt -= iosize;
1788 			++lblkno;
1789 		}
1790 		BO_UNLOCK(bo);
1791 	}
1792 	NFSEXITCODE(error);
1793 	return (error);
1794 }
1795 
1796 /*
1797  * Statfs vnode op.
1798  */
1799 int
1800 nfsvno_statfs(struct vnode *vp, struct statfs *sf)
1801 {
1802 	struct statfs *tsf;
1803 	int error;
1804 
1805 	tsf = NULL;
1806 	if (nfsrv_devidcnt > 0) {
1807 		/* For a pNFS service, get the DS numbers. */
1808 		tsf = malloc(sizeof(*tsf), M_TEMP, M_WAITOK | M_ZERO);
1809 		error = nfsrv_pnfsstatfs(tsf, vp->v_mount);
1810 		if (error != 0) {
1811 			free(tsf, M_TEMP);
1812 			tsf = NULL;
1813 		}
1814 	}
1815 	error = VFS_STATFS(vp->v_mount, sf);
1816 	if (error == 0) {
1817 		if (tsf != NULL) {
1818 			sf->f_blocks = tsf->f_blocks;
1819 			sf->f_bavail = tsf->f_bavail;
1820 			sf->f_bfree = tsf->f_bfree;
1821 			sf->f_bsize = tsf->f_bsize;
1822 		}
1823 		/*
1824 		 * Since NFS handles these values as unsigned on the
1825 		 * wire, there is no way to represent negative values,
1826 		 * so set them to 0. Without this, they will appear
1827 		 * to be very large positive values for clients like
1828 		 * Solaris10.
1829 		 */
1830 		if (sf->f_bavail < 0)
1831 			sf->f_bavail = 0;
1832 		if (sf->f_ffree < 0)
1833 			sf->f_ffree = 0;
1834 	}
1835 	free(tsf, M_TEMP);
1836 	NFSEXITCODE(error);
1837 	return (error);
1838 }
1839 
1840 /*
1841  * Do the vnode op stuff for Open. Similar to nfsvno_createsub(), but
1842  * must handle nfsrv_opencheck() calls after any other access checks.
1843  */
1844 void
1845 nfsvno_open(struct nfsrv_descript *nd, struct nameidata *ndp,
1846     nfsquad_t clientid, nfsv4stateid_t *stateidp, struct nfsstate *stp,
1847     int *exclusive_flagp, struct nfsvattr *nvap, int32_t *cverf, int create,
1848     NFSACL_T *aclp, nfsattrbit_t *attrbitp, struct ucred *cred, bool done_namei,
1849     struct nfsexstuff *exp, struct vnode **vpp)
1850 {
1851 	struct vnode *vp = NULL;
1852 	u_quad_t tempsize;
1853 	struct nfsexstuff nes;
1854 	struct thread *p = curthread;
1855 	uint32_t oldrepstat;
1856 
1857 	if (ndp->ni_vp == NULL) {
1858 		/*
1859 		 * If nfsrv_opencheck() sets nd_repstat, done_namei needs to be
1860 		 * set true, since cleanup after nfsvno_namei() is needed.
1861 		 */
1862 		oldrepstat = nd->nd_repstat;
1863 		nd->nd_repstat = nfsrv_opencheck(clientid,
1864 		    stateidp, stp, NULL, nd, p, nd->nd_repstat);
1865 		if (nd->nd_repstat != 0 && oldrepstat == 0)
1866 			done_namei = true;
1867 	}
1868 	if (!nd->nd_repstat) {
1869 		if (ndp->ni_vp == NULL) {
1870 			nd->nd_repstat = VOP_CREATE(ndp->ni_dvp,
1871 			    &ndp->ni_vp, &ndp->ni_cnd, &nvap->na_vattr);
1872 			/* For a pNFS server, create the data file on a DS. */
1873 			if (nd->nd_repstat == 0) {
1874 				/*
1875 				 * Create a data file on a DS for a pNFS server.
1876 				 * This function just returns if not
1877 				 * running a pNFS DS or the creation fails.
1878 				 */
1879 				nfsrv_pnfscreate(ndp->ni_vp, &nvap->na_vattr,
1880 				    cred, p);
1881 			}
1882 			VOP_VPUT_PAIR(ndp->ni_dvp, nd->nd_repstat == 0 ?
1883 			    &ndp->ni_vp : NULL, false);
1884 			nfsvno_relpathbuf(ndp);
1885 			if (!nd->nd_repstat) {
1886 				if (*exclusive_flagp) {
1887 					*exclusive_flagp = 0;
1888 					NFSVNO_ATTRINIT(nvap);
1889 					nvap->na_atime.tv_sec = cverf[0];
1890 					nvap->na_atime.tv_nsec = cverf[1];
1891 					nd->nd_repstat = VOP_SETATTR(ndp->ni_vp,
1892 					    &nvap->na_vattr, cred);
1893 					if (nd->nd_repstat != 0) {
1894 						vput(ndp->ni_vp);
1895 						ndp->ni_vp = NULL;
1896 						nd->nd_repstat = NFSERR_NOTSUPP;
1897 					} else
1898 						NFSSETBIT_ATTRBIT(attrbitp,
1899 						    NFSATTRBIT_TIMEACCESS);
1900 				} else {
1901 					nfsrv_fixattr(nd, ndp->ni_vp, nvap,
1902 					    aclp, p, attrbitp, exp);
1903 				}
1904 			}
1905 			vp = ndp->ni_vp;
1906 		} else {
1907 			nfsvno_relpathbuf(ndp);
1908 			vp = ndp->ni_vp;
1909 			if (create == NFSV4OPEN_CREATE) {
1910 				if (ndp->ni_dvp == vp)
1911 					vrele(ndp->ni_dvp);
1912 				else
1913 					vput(ndp->ni_dvp);
1914 			}
1915 			if (NFSVNO_ISSETSIZE(nvap) && vp->v_type == VREG) {
1916 				if (ndp->ni_cnd.cn_flags & RDONLY)
1917 					NFSVNO_SETEXRDONLY(&nes);
1918 				else
1919 					NFSVNO_EXINIT(&nes);
1920 				nd->nd_repstat = nfsvno_accchk(vp,
1921 				    VWRITE, cred, &nes, p,
1922 				    NFSACCCHK_NOOVERRIDE,
1923 				    NFSACCCHK_VPISLOCKED, NULL);
1924 				nd->nd_repstat = nfsrv_opencheck(clientid,
1925 				    stateidp, stp, vp, nd, p, nd->nd_repstat);
1926 				if (!nd->nd_repstat) {
1927 					tempsize = nvap->na_size;
1928 					NFSVNO_ATTRINIT(nvap);
1929 					nvap->na_size = tempsize;
1930 					nd->nd_repstat = nfsvno_setattr(vp,
1931 					    nvap, cred, p, exp);
1932 				}
1933 			} else if (vp->v_type == VREG) {
1934 				nd->nd_repstat = nfsrv_opencheck(clientid,
1935 				    stateidp, stp, vp, nd, p, nd->nd_repstat);
1936 			}
1937 		}
1938 	} else if (done_namei) {
1939 		KASSERT(create == NFSV4OPEN_CREATE,
1940 		    ("nfsvno_open: not create"));
1941 		/*
1942 		 * done_namei is set when nfsvno_namei() has completed
1943 		 * successfully, but a subsequent error was set in
1944 		 * nd_repstat.  As such, cleanup of the nfsvno_namei()
1945 		 * results is required.
1946 		 */
1947 		nfsvno_relpathbuf(ndp);
1948 		if (ndp->ni_dvp == ndp->ni_vp)
1949 			vrele(ndp->ni_dvp);
1950 		else
1951 			vput(ndp->ni_dvp);
1952 		if (ndp->ni_vp)
1953 			vput(ndp->ni_vp);
1954 	}
1955 	*vpp = vp;
1956 
1957 	NFSEXITCODE2(0, nd);
1958 }
1959 
1960 /*
1961  * Updates the file rev and sets the mtime and ctime
1962  * to the current clock time, returning the va_filerev and va_Xtime
1963  * values.
1964  * Return ESTALE to indicate the vnode is VIRF_DOOMED.
1965  */
1966 int
1967 nfsvno_updfilerev(struct vnode *vp, struct nfsvattr *nvap,
1968     struct nfsrv_descript *nd, struct thread *p)
1969 {
1970 	struct vattr va;
1971 
1972 	VATTR_NULL(&va);
1973 	vfs_timestamp(&va.va_mtime);
1974 	if (NFSVOPISLOCKED(vp) != LK_EXCLUSIVE) {
1975 		NFSVOPLOCK(vp, LK_UPGRADE | LK_RETRY);
1976 		if (VN_IS_DOOMED(vp))
1977 			return (ESTALE);
1978 	}
1979 	(void) VOP_SETATTR(vp, &va, nd->nd_cred);
1980 	(void) nfsvno_getattr(vp, nvap, nd, p, 1, NULL);
1981 	return (0);
1982 }
1983 
1984 /*
1985  * Glue routine to nfsv4_fillattr().
1986  */
1987 int
1988 nfsvno_fillattr(struct nfsrv_descript *nd, struct mount *mp, struct vnode *vp,
1989     struct nfsvattr *nvap, fhandle_t *fhp, int rderror, nfsattrbit_t *attrbitp,
1990     struct ucred *cred, struct thread *p, int isdgram, int reterr,
1991     int supports_nfsv4acls, int at_root, uint64_t mounted_on_fileno)
1992 {
1993 	struct statfs *sf;
1994 	int error;
1995 
1996 	sf = NULL;
1997 	if (nfsrv_devidcnt > 0 &&
1998 	    (NFSISSET_ATTRBIT(attrbitp, NFSATTRBIT_SPACEAVAIL) ||
1999 	     NFSISSET_ATTRBIT(attrbitp, NFSATTRBIT_SPACEFREE) ||
2000 	     NFSISSET_ATTRBIT(attrbitp, NFSATTRBIT_SPACETOTAL))) {
2001 		sf = malloc(sizeof(*sf), M_TEMP, M_WAITOK | M_ZERO);
2002 		error = nfsrv_pnfsstatfs(sf, mp);
2003 		if (error != 0) {
2004 			free(sf, M_TEMP);
2005 			sf = NULL;
2006 		}
2007 	}
2008 	error = nfsv4_fillattr(nd, mp, vp, NULL, &nvap->na_vattr, fhp, rderror,
2009 	    attrbitp, cred, p, isdgram, reterr, supports_nfsv4acls, at_root,
2010 	    mounted_on_fileno, sf);
2011 	free(sf, M_TEMP);
2012 	NFSEXITCODE2(0, nd);
2013 	return (error);
2014 }
2015 
2016 /* Since the Readdir vnode ops vary, put the entire functions in here. */
2017 /*
2018  * nfs readdir service
2019  * - mallocs what it thinks is enough to read
2020  *	count rounded up to a multiple of DIRBLKSIZ <= NFS_MAXREADDIR
2021  * - calls VOP_READDIR()
2022  * - loops around building the reply
2023  *	if the output generated exceeds count break out of loop
2024  *	The NFSM_CLGET macro is used here so that the reply will be packed
2025  *	tightly in mbuf clusters.
2026  * - it trims out records with d_fileno == 0
2027  *	this doesn't matter for Unix clients, but they might confuse clients
2028  *	for other os'.
2029  * - it trims out records with d_type == DT_WHT
2030  *	these cannot be seen through NFS (unless we extend the protocol)
2031  *     The alternate call nfsrvd_readdirplus() does lookups as well.
2032  * PS: The NFS protocol spec. does not clarify what the "count" byte
2033  *	argument is a count of.. just name strings and file id's or the
2034  *	entire reply rpc or ...
2035  *	I tried just file name and id sizes and it confused the Sun client,
2036  *	so I am using the full rpc size now. The "paranoia.." comment refers
2037  *	to including the status longwords that are not a part of the dir.
2038  *	"entry" structures, but are in the rpc.
2039  */
2040 int
2041 nfsrvd_readdir(struct nfsrv_descript *nd, int isdgram,
2042     struct vnode *vp, struct nfsexstuff *exp)
2043 {
2044 	struct dirent *dp;
2045 	u_int32_t *tl;
2046 	int dirlen;
2047 	char *cpos, *cend, *rbuf;
2048 	struct nfsvattr at;
2049 	int nlen, error = 0, getret = 1;
2050 	int siz, cnt, fullsiz, eofflag, ncookies;
2051 	u_int64_t off, toff, verf __unused;
2052 	uint64_t *cookies = NULL, *cookiep;
2053 	struct uio io;
2054 	struct iovec iv;
2055 	int is_ufs;
2056 	struct thread *p = curthread;
2057 
2058 	if (nd->nd_repstat) {
2059 		nfsrv_postopattr(nd, getret, &at);
2060 		goto out;
2061 	}
2062 	if (nd->nd_flag & ND_NFSV2) {
2063 		NFSM_DISSECT(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
2064 		off = fxdr_unsigned(u_quad_t, *tl++);
2065 	} else {
2066 		NFSM_DISSECT(tl, u_int32_t *, 5 * NFSX_UNSIGNED);
2067 		off = fxdr_hyper(tl);
2068 		tl += 2;
2069 		verf = fxdr_hyper(tl);
2070 		tl += 2;
2071 	}
2072 	toff = off;
2073 	cnt = fxdr_unsigned(int, *tl);
2074 	if (cnt > NFS_SRVMAXDATA(nd) || cnt < 0)
2075 		cnt = NFS_SRVMAXDATA(nd);
2076 	siz = ((cnt + DIRBLKSIZ - 1) & ~(DIRBLKSIZ - 1));
2077 	fullsiz = siz;
2078 	if (nd->nd_flag & ND_NFSV3) {
2079 		nd->nd_repstat = getret = nfsvno_getattr(vp, &at, nd, p, 1,
2080 		    NULL);
2081 #if 0
2082 		/*
2083 		 * va_filerev is not sufficient as a cookie verifier,
2084 		 * since it is not supposed to change when entries are
2085 		 * removed/added unless that offset cookies returned to
2086 		 * the client are no longer valid.
2087 		 */
2088 		if (!nd->nd_repstat && toff && verf != at.na_filerev)
2089 			nd->nd_repstat = NFSERR_BAD_COOKIE;
2090 #endif
2091 	}
2092 	if (!nd->nd_repstat && vp->v_type != VDIR)
2093 		nd->nd_repstat = NFSERR_NOTDIR;
2094 	if (nd->nd_repstat == 0 && cnt == 0) {
2095 		if (nd->nd_flag & ND_NFSV2)
2096 			/* NFSv2 does not have NFSERR_TOOSMALL */
2097 			nd->nd_repstat = EPERM;
2098 		else
2099 			nd->nd_repstat = NFSERR_TOOSMALL;
2100 	}
2101 	if (!nd->nd_repstat)
2102 		nd->nd_repstat = nfsvno_accchk(vp, VEXEC,
2103 		    nd->nd_cred, exp, p, NFSACCCHK_NOOVERRIDE,
2104 		    NFSACCCHK_VPISLOCKED, NULL);
2105 	if (nd->nd_repstat) {
2106 		vput(vp);
2107 		if (nd->nd_flag & ND_NFSV3)
2108 			nfsrv_postopattr(nd, getret, &at);
2109 		goto out;
2110 	}
2111 	is_ufs = strcmp(vp->v_mount->mnt_vfc->vfc_name, "ufs") == 0;
2112 	rbuf = malloc(siz, M_TEMP, M_WAITOK);
2113 again:
2114 	eofflag = 0;
2115 	if (cookies) {
2116 		free(cookies, M_TEMP);
2117 		cookies = NULL;
2118 	}
2119 
2120 	iv.iov_base = rbuf;
2121 	iv.iov_len = siz;
2122 	io.uio_iov = &iv;
2123 	io.uio_iovcnt = 1;
2124 	io.uio_offset = (off_t)off;
2125 	io.uio_resid = siz;
2126 	io.uio_segflg = UIO_SYSSPACE;
2127 	io.uio_rw = UIO_READ;
2128 	io.uio_td = NULL;
2129 	nd->nd_repstat = VOP_READDIR(vp, &io, nd->nd_cred, &eofflag, &ncookies,
2130 	    &cookies);
2131 	off = (u_int64_t)io.uio_offset;
2132 	if (io.uio_resid)
2133 		siz -= io.uio_resid;
2134 
2135 	if (!cookies && !nd->nd_repstat)
2136 		nd->nd_repstat = NFSERR_PERM;
2137 	if (nd->nd_flag & ND_NFSV3) {
2138 		getret = nfsvno_getattr(vp, &at, nd, p, 1, NULL);
2139 		if (!nd->nd_repstat)
2140 			nd->nd_repstat = getret;
2141 	}
2142 
2143 	/*
2144 	 * Handles the failed cases. nd->nd_repstat == 0 past here.
2145 	 */
2146 	if (nd->nd_repstat) {
2147 		vput(vp);
2148 		free(rbuf, M_TEMP);
2149 		if (cookies)
2150 			free(cookies, M_TEMP);
2151 		if (nd->nd_flag & ND_NFSV3)
2152 			nfsrv_postopattr(nd, getret, &at);
2153 		goto out;
2154 	}
2155 	/*
2156 	 * If nothing read, return eof
2157 	 * rpc reply
2158 	 */
2159 	if (siz == 0) {
2160 		vput(vp);
2161 		if (nd->nd_flag & ND_NFSV2) {
2162 			NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
2163 		} else {
2164 			nfsrv_postopattr(nd, getret, &at);
2165 			NFSM_BUILD(tl, u_int32_t *, 4 * NFSX_UNSIGNED);
2166 			txdr_hyper(at.na_filerev, tl);
2167 			tl += 2;
2168 		}
2169 		*tl++ = newnfs_false;
2170 		*tl = newnfs_true;
2171 		free(rbuf, M_TEMP);
2172 		free(cookies, M_TEMP);
2173 		goto out;
2174 	}
2175 
2176 	/*
2177 	 * Check for degenerate cases of nothing useful read.
2178 	 * If so go try again
2179 	 */
2180 	cpos = rbuf;
2181 	cend = rbuf + siz;
2182 	dp = (struct dirent *)cpos;
2183 	cookiep = cookies;
2184 
2185 	/*
2186 	 * For some reason FreeBSD's ufs_readdir() chooses to back the
2187 	 * directory offset up to a block boundary, so it is necessary to
2188 	 * skip over the records that precede the requested offset. This
2189 	 * requires the assumption that file offset cookies monotonically
2190 	 * increase.
2191 	 */
2192 	while (cpos < cend && ncookies > 0 &&
2193 	    (dp->d_fileno == 0 || dp->d_type == DT_WHT ||
2194 	     (is_ufs == 1 && ((u_quad_t)(*cookiep)) <= toff))) {
2195 		cpos += dp->d_reclen;
2196 		dp = (struct dirent *)cpos;
2197 		cookiep++;
2198 		ncookies--;
2199 	}
2200 	if (cpos >= cend || ncookies == 0) {
2201 		siz = fullsiz;
2202 		toff = off;
2203 		goto again;
2204 	}
2205 	vput(vp);
2206 
2207 	/*
2208 	 * If cnt > MCLBYTES and the reply will not be saved, use
2209 	 * ext_pgs mbufs for TLS.
2210 	 * For NFSv4.0, we do not know for sure if the reply will
2211 	 * be saved, so do not use ext_pgs mbufs for NFSv4.0.
2212 	 */
2213 	if (cnt > MCLBYTES && siz > MCLBYTES &&
2214 	    (nd->nd_flag & (ND_TLS | ND_EXTPG | ND_SAVEREPLY)) == ND_TLS &&
2215 	    (nd->nd_flag & (ND_NFSV4 | ND_NFSV41)) != ND_NFSV4)
2216 		nd->nd_flag |= ND_EXTPG;
2217 
2218 	/*
2219 	 * dirlen is the size of the reply, including all XDR and must
2220 	 * not exceed cnt. For NFSv2, RFC1094 didn't clearly indicate
2221 	 * if the XDR should be included in "count", but to be safe, we do.
2222 	 * (Include the two booleans at the end of the reply in dirlen now.)
2223 	 */
2224 	if (nd->nd_flag & ND_NFSV3) {
2225 		nfsrv_postopattr(nd, getret, &at);
2226 		NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
2227 		txdr_hyper(at.na_filerev, tl);
2228 		dirlen = NFSX_V3POSTOPATTR + NFSX_VERF + 2 * NFSX_UNSIGNED;
2229 	} else {
2230 		dirlen = 2 * NFSX_UNSIGNED;
2231 	}
2232 
2233 	/* Loop through the records and build reply */
2234 	while (cpos < cend && ncookies > 0) {
2235 		nlen = dp->d_namlen;
2236 		if (dp->d_fileno != 0 && dp->d_type != DT_WHT &&
2237 			nlen <= NFS_MAXNAMLEN) {
2238 			if (nd->nd_flag & ND_NFSV3)
2239 				dirlen += (6*NFSX_UNSIGNED + NFSM_RNDUP(nlen));
2240 			else
2241 				dirlen += (4*NFSX_UNSIGNED + NFSM_RNDUP(nlen));
2242 			if (dirlen > cnt) {
2243 				eofflag = 0;
2244 				break;
2245 			}
2246 
2247 			/*
2248 			 * Build the directory record xdr from
2249 			 * the dirent entry.
2250 			 */
2251 			if (nd->nd_flag & ND_NFSV3) {
2252 				NFSM_BUILD(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
2253 				*tl++ = newnfs_true;
2254 				*tl++ = 0;
2255 			} else {
2256 				NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
2257 				*tl++ = newnfs_true;
2258 			}
2259 			*tl = txdr_unsigned(dp->d_fileno);
2260 			(void) nfsm_strtom(nd, dp->d_name, nlen);
2261 			if (nd->nd_flag & ND_NFSV3) {
2262 				NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
2263 				txdr_hyper(*cookiep, tl);
2264 			} else {
2265 				NFSM_BUILD(tl, u_int32_t *, NFSX_UNSIGNED);
2266 				*tl = txdr_unsigned(*cookiep);
2267 			}
2268 		}
2269 		cpos += dp->d_reclen;
2270 		dp = (struct dirent *)cpos;
2271 		cookiep++;
2272 		ncookies--;
2273 	}
2274 	if (cpos < cend)
2275 		eofflag = 0;
2276 	NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
2277 	*tl++ = newnfs_false;
2278 	if (eofflag)
2279 		*tl = newnfs_true;
2280 	else
2281 		*tl = newnfs_false;
2282 	free(rbuf, M_TEMP);
2283 	free(cookies, M_TEMP);
2284 
2285 out:
2286 	NFSEXITCODE2(0, nd);
2287 	return (0);
2288 nfsmout:
2289 	vput(vp);
2290 	NFSEXITCODE2(error, nd);
2291 	return (error);
2292 }
2293 
2294 /*
2295  * Readdirplus for V3 and Readdir for V4.
2296  */
2297 int
2298 nfsrvd_readdirplus(struct nfsrv_descript *nd, int isdgram,
2299     struct vnode *vp, struct nfsexstuff *exp)
2300 {
2301 	struct dirent *dp;
2302 	u_int32_t *tl;
2303 	int dirlen;
2304 	char *cpos, *cend, *rbuf;
2305 	struct vnode *nvp;
2306 	fhandle_t nfh;
2307 	struct nfsvattr nva, at, *nvap = &nva;
2308 	struct mbuf *mb0, *mb1;
2309 	struct nfsreferral *refp;
2310 	int nlen, r, error = 0, getret = 1, usevget = 1;
2311 	int siz, cnt, fullsiz, eofflag, ncookies, entrycnt;
2312 	caddr_t bpos0, bpos1;
2313 	u_int64_t off, toff, verf __unused;
2314 	uint64_t *cookies = NULL, *cookiep;
2315 	nfsattrbit_t attrbits, rderrbits, savbits;
2316 	struct uio io;
2317 	struct iovec iv;
2318 	struct componentname cn;
2319 	int at_root, is_ufs, is_zfs, needs_unbusy, supports_nfsv4acls;
2320 	struct mount *mp, *new_mp;
2321 	uint64_t mounted_on_fileno;
2322 	struct thread *p = curthread;
2323 	int bextpg0, bextpg1, bextpgsiz0, bextpgsiz1;
2324 
2325 	if (nd->nd_repstat) {
2326 		nfsrv_postopattr(nd, getret, &at);
2327 		goto out;
2328 	}
2329 	NFSM_DISSECT(tl, u_int32_t *, 6 * NFSX_UNSIGNED);
2330 	off = fxdr_hyper(tl);
2331 	toff = off;
2332 	tl += 2;
2333 	verf = fxdr_hyper(tl);
2334 	tl += 2;
2335 	siz = fxdr_unsigned(int, *tl++);
2336 	cnt = fxdr_unsigned(int, *tl);
2337 
2338 	/*
2339 	 * Use the server's maximum data transfer size as the upper bound
2340 	 * on reply datalen.
2341 	 */
2342 	if (cnt > NFS_SRVMAXDATA(nd) || cnt < 0)
2343 		cnt = NFS_SRVMAXDATA(nd);
2344 
2345 	/*
2346 	 * siz is a "hint" of how much directory information (name, fileid,
2347 	 * cookie) should be in the reply. At least one client "hints" 0,
2348 	 * so I set it to cnt for that case. I also round it up to the
2349 	 * next multiple of DIRBLKSIZ.
2350 	 * Since the size of a Readdirplus directory entry reply will always
2351 	 * be greater than a directory entry returned by VOP_READDIR(), it
2352 	 * does not make sense to read more than NFS_SRVMAXDATA() via
2353 	 * VOP_READDIR().
2354 	 */
2355 	if (siz <= 0)
2356 		siz = cnt;
2357 	else if (siz > NFS_SRVMAXDATA(nd))
2358 		siz = NFS_SRVMAXDATA(nd);
2359 	siz = ((siz + DIRBLKSIZ - 1) & ~(DIRBLKSIZ - 1));
2360 
2361 	if (nd->nd_flag & ND_NFSV4) {
2362 		error = nfsrv_getattrbits(nd, &attrbits, NULL, NULL);
2363 		if (error)
2364 			goto nfsmout;
2365 		NFSSET_ATTRBIT(&savbits, &attrbits);
2366 		NFSCLRNOTFILLABLE_ATTRBIT(&attrbits, nd);
2367 		NFSZERO_ATTRBIT(&rderrbits);
2368 		NFSSETBIT_ATTRBIT(&rderrbits, NFSATTRBIT_RDATTRERROR);
2369 	} else {
2370 		NFSZERO_ATTRBIT(&attrbits);
2371 	}
2372 	fullsiz = siz;
2373 	nd->nd_repstat = getret = nfsvno_getattr(vp, &at, nd, p, 1, NULL);
2374 #if 0
2375 	if (!nd->nd_repstat) {
2376 	    if (off && verf != at.na_filerev) {
2377 		/*
2378 		 * va_filerev is not sufficient as a cookie verifier,
2379 		 * since it is not supposed to change when entries are
2380 		 * removed/added unless that offset cookies returned to
2381 		 * the client are no longer valid.
2382 		 */
2383 		if (nd->nd_flag & ND_NFSV4) {
2384 			nd->nd_repstat = NFSERR_NOTSAME;
2385 		} else {
2386 			nd->nd_repstat = NFSERR_BAD_COOKIE;
2387 		}
2388 	    }
2389 	}
2390 #endif
2391 	if (!nd->nd_repstat && vp->v_type != VDIR)
2392 		nd->nd_repstat = NFSERR_NOTDIR;
2393 	if (!nd->nd_repstat && cnt == 0)
2394 		nd->nd_repstat = NFSERR_TOOSMALL;
2395 	if (!nd->nd_repstat)
2396 		nd->nd_repstat = nfsvno_accchk(vp, VEXEC,
2397 		    nd->nd_cred, exp, p, NFSACCCHK_NOOVERRIDE,
2398 		    NFSACCCHK_VPISLOCKED, NULL);
2399 	if (nd->nd_repstat) {
2400 		vput(vp);
2401 		if (nd->nd_flag & ND_NFSV3)
2402 			nfsrv_postopattr(nd, getret, &at);
2403 		goto out;
2404 	}
2405 	is_ufs = strcmp(vp->v_mount->mnt_vfc->vfc_name, "ufs") == 0;
2406 	is_zfs = strcmp(vp->v_mount->mnt_vfc->vfc_name, "zfs") == 0;
2407 
2408 	rbuf = malloc(siz, M_TEMP, M_WAITOK);
2409 again:
2410 	eofflag = 0;
2411 	if (cookies) {
2412 		free(cookies, M_TEMP);
2413 		cookies = NULL;
2414 	}
2415 
2416 	iv.iov_base = rbuf;
2417 	iv.iov_len = siz;
2418 	io.uio_iov = &iv;
2419 	io.uio_iovcnt = 1;
2420 	io.uio_offset = (off_t)off;
2421 	io.uio_resid = siz;
2422 	io.uio_segflg = UIO_SYSSPACE;
2423 	io.uio_rw = UIO_READ;
2424 	io.uio_td = NULL;
2425 	nd->nd_repstat = VOP_READDIR(vp, &io, nd->nd_cred, &eofflag, &ncookies,
2426 	    &cookies);
2427 	off = (u_int64_t)io.uio_offset;
2428 	if (io.uio_resid)
2429 		siz -= io.uio_resid;
2430 
2431 	getret = nfsvno_getattr(vp, &at, nd, p, 1, NULL);
2432 
2433 	if (!cookies && !nd->nd_repstat)
2434 		nd->nd_repstat = NFSERR_PERM;
2435 	if (!nd->nd_repstat)
2436 		nd->nd_repstat = getret;
2437 	if (nd->nd_repstat) {
2438 		vput(vp);
2439 		if (cookies)
2440 			free(cookies, M_TEMP);
2441 		free(rbuf, M_TEMP);
2442 		if (nd->nd_flag & ND_NFSV3)
2443 			nfsrv_postopattr(nd, getret, &at);
2444 		goto out;
2445 	}
2446 	/*
2447 	 * If nothing read, return eof
2448 	 * rpc reply
2449 	 */
2450 	if (siz == 0) {
2451 		vput(vp);
2452 		if (nd->nd_flag & ND_NFSV3)
2453 			nfsrv_postopattr(nd, getret, &at);
2454 		NFSM_BUILD(tl, u_int32_t *, 4 * NFSX_UNSIGNED);
2455 		txdr_hyper(at.na_filerev, tl);
2456 		tl += 2;
2457 		*tl++ = newnfs_false;
2458 		*tl = newnfs_true;
2459 		free(cookies, M_TEMP);
2460 		free(rbuf, M_TEMP);
2461 		goto out;
2462 	}
2463 
2464 	/*
2465 	 * Check for degenerate cases of nothing useful read.
2466 	 * If so go try again
2467 	 */
2468 	cpos = rbuf;
2469 	cend = rbuf + siz;
2470 	dp = (struct dirent *)cpos;
2471 	cookiep = cookies;
2472 
2473 	/*
2474 	 * For some reason FreeBSD's ufs_readdir() chooses to back the
2475 	 * directory offset up to a block boundary, so it is necessary to
2476 	 * skip over the records that precede the requested offset. This
2477 	 * requires the assumption that file offset cookies monotonically
2478 	 * increase.
2479 	 */
2480 	while (cpos < cend && ncookies > 0 &&
2481 	  (dp->d_fileno == 0 || dp->d_type == DT_WHT ||
2482 	   (is_ufs == 1 && ((u_quad_t)(*cookiep)) <= toff) ||
2483 	   ((nd->nd_flag & ND_NFSV4) &&
2484 	    ((dp->d_namlen == 1 && dp->d_name[0] == '.') ||
2485 	     (dp->d_namlen==2 && dp->d_name[0]=='.' && dp->d_name[1]=='.'))))) {
2486 		cpos += dp->d_reclen;
2487 		dp = (struct dirent *)cpos;
2488 		cookiep++;
2489 		ncookies--;
2490 	}
2491 	if (cpos >= cend || ncookies == 0) {
2492 		siz = fullsiz;
2493 		toff = off;
2494 		goto again;
2495 	}
2496 
2497 	/*
2498 	 * Busy the file system so that the mount point won't go away
2499 	 * and, as such, VFS_VGET() can be used safely.
2500 	 */
2501 	mp = vp->v_mount;
2502 	vfs_ref(mp);
2503 	NFSVOPUNLOCK(vp);
2504 	nd->nd_repstat = vfs_busy(mp, 0);
2505 	vfs_rel(mp);
2506 	if (nd->nd_repstat != 0) {
2507 		vrele(vp);
2508 		free(cookies, M_TEMP);
2509 		free(rbuf, M_TEMP);
2510 		if (nd->nd_flag & ND_NFSV3)
2511 			nfsrv_postopattr(nd, getret, &at);
2512 		goto out;
2513 	}
2514 
2515 	/*
2516 	 * Check to see if entries in this directory can be safely acquired
2517 	 * via VFS_VGET() or if a switch to VOP_LOOKUP() is required.
2518 	 * ZFS snapshot directories need VOP_LOOKUP(), so that any
2519 	 * automount of the snapshot directory that is required will
2520 	 * be done.
2521 	 * This needs to be done here for NFSv4, since NFSv4 never does
2522 	 * a VFS_VGET() for "." or "..".
2523 	 */
2524 	if (is_zfs == 1) {
2525 		r = VFS_VGET(mp, at.na_fileid, LK_SHARED, &nvp);
2526 		if (r == EOPNOTSUPP) {
2527 			usevget = 0;
2528 			cn.cn_nameiop = LOOKUP;
2529 			cn.cn_lkflags = LK_SHARED | LK_RETRY;
2530 			cn.cn_cred = nd->nd_cred;
2531 		} else if (r == 0)
2532 			vput(nvp);
2533 	}
2534 
2535 	/*
2536 	 * If the reply is likely to exceed MCLBYTES and the reply will
2537 	 * not be saved, use ext_pgs mbufs for TLS.
2538 	 * It is difficult to predict how large each entry will be and
2539 	 * how many entries have been read, so just assume the directory
2540 	 * entries grow by a factor of 4 when attributes are included.
2541 	 * For NFSv4.0, we do not know for sure if the reply will
2542 	 * be saved, so do not use ext_pgs mbufs for NFSv4.0.
2543 	 */
2544 	if (cnt > MCLBYTES && siz > MCLBYTES / 4 &&
2545 	    (nd->nd_flag & (ND_TLS | ND_EXTPG | ND_SAVEREPLY)) == ND_TLS &&
2546 	    (nd->nd_flag & (ND_NFSV4 | ND_NFSV41)) != ND_NFSV4)
2547 		nd->nd_flag |= ND_EXTPG;
2548 
2549 	/*
2550 	 * Save this position, in case there is an error before one entry
2551 	 * is created.
2552 	 */
2553 	mb0 = nd->nd_mb;
2554 	bpos0 = nd->nd_bpos;
2555 	bextpg0 = nd->nd_bextpg;
2556 	bextpgsiz0 = nd->nd_bextpgsiz;
2557 
2558 	/*
2559 	 * Fill in the first part of the reply.
2560 	 * dirlen is the reply length in bytes and cannot exceed cnt.
2561 	 * (Include the two booleans at the end of the reply in dirlen now,
2562 	 *  so we recognize when we have exceeded cnt.)
2563 	 */
2564 	if (nd->nd_flag & ND_NFSV3) {
2565 		dirlen = NFSX_V3POSTOPATTR + NFSX_VERF + 2 * NFSX_UNSIGNED;
2566 		nfsrv_postopattr(nd, getret, &at);
2567 	} else {
2568 		dirlen = NFSX_VERF + 2 * NFSX_UNSIGNED;
2569 	}
2570 	NFSM_BUILD(tl, u_int32_t *, NFSX_VERF);
2571 	txdr_hyper(at.na_filerev, tl);
2572 
2573 	/*
2574 	 * Save this position, in case there is an empty reply needed.
2575 	 */
2576 	mb1 = nd->nd_mb;
2577 	bpos1 = nd->nd_bpos;
2578 	bextpg1 = nd->nd_bextpg;
2579 	bextpgsiz1 = nd->nd_bextpgsiz;
2580 
2581 	/* Loop through the records and build reply */
2582 	entrycnt = 0;
2583 	while (cpos < cend && ncookies > 0 && dirlen < cnt) {
2584 		nlen = dp->d_namlen;
2585 		if (dp->d_fileno != 0 && dp->d_type != DT_WHT &&
2586 		    nlen <= NFS_MAXNAMLEN &&
2587 		    ((nd->nd_flag & ND_NFSV3) || nlen > 2 ||
2588 		     (nlen==2 && (dp->d_name[0]!='.' || dp->d_name[1]!='.'))
2589 		      || (nlen == 1 && dp->d_name[0] != '.'))) {
2590 			/*
2591 			 * Save the current position in the reply, in case
2592 			 * this entry exceeds cnt.
2593 			 */
2594 			mb1 = nd->nd_mb;
2595 			bpos1 = nd->nd_bpos;
2596 			bextpg1 = nd->nd_bextpg;
2597 			bextpgsiz1 = nd->nd_bextpgsiz;
2598 
2599 			/*
2600 			 * For readdir_and_lookup get the vnode using
2601 			 * the file number.
2602 			 */
2603 			nvp = NULL;
2604 			refp = NULL;
2605 			r = 0;
2606 			at_root = 0;
2607 			needs_unbusy = 0;
2608 			new_mp = mp;
2609 			mounted_on_fileno = (uint64_t)dp->d_fileno;
2610 			if ((nd->nd_flag & ND_NFSV3) ||
2611 			    NFSNONZERO_ATTRBIT(&savbits)) {
2612 				if (nd->nd_flag & ND_NFSV4)
2613 					refp = nfsv4root_getreferral(NULL,
2614 					    vp, dp->d_fileno);
2615 				if (refp == NULL) {
2616 					if (usevget)
2617 						r = VFS_VGET(mp, dp->d_fileno,
2618 						    LK_SHARED, &nvp);
2619 					else
2620 						r = EOPNOTSUPP;
2621 					if (r == EOPNOTSUPP) {
2622 						if (usevget) {
2623 							usevget = 0;
2624 							cn.cn_nameiop = LOOKUP;
2625 							cn.cn_lkflags =
2626 							    LK_SHARED |
2627 							    LK_RETRY;
2628 							cn.cn_cred =
2629 							    nd->nd_cred;
2630 						}
2631 						cn.cn_nameptr = dp->d_name;
2632 						cn.cn_namelen = nlen;
2633 						cn.cn_flags = ISLASTCN |
2634 						    NOFOLLOW | LOCKLEAF;
2635 						if (nlen == 2 &&
2636 						    dp->d_name[0] == '.' &&
2637 						    dp->d_name[1] == '.')
2638 							cn.cn_flags |=
2639 							    ISDOTDOT;
2640 						if (NFSVOPLOCK(vp, LK_SHARED)
2641 						    != 0) {
2642 							nd->nd_repstat = EPERM;
2643 							break;
2644 						}
2645 						if ((vp->v_vflag & VV_ROOT) != 0
2646 						    && (cn.cn_flags & ISDOTDOT)
2647 						    != 0) {
2648 							vref(vp);
2649 							nvp = vp;
2650 							r = 0;
2651 						} else {
2652 							r = VOP_LOOKUP(vp, &nvp,
2653 							    &cn);
2654 							if (vp != nvp)
2655 								NFSVOPUNLOCK(vp);
2656 						}
2657 					}
2658 
2659 					/*
2660 					 * For NFSv4, check to see if nvp is
2661 					 * a mount point and get the mount
2662 					 * point vnode, as required.
2663 					 */
2664 					if (r == 0 &&
2665 					    nfsrv_enable_crossmntpt != 0 &&
2666 					    (nd->nd_flag & ND_NFSV4) != 0 &&
2667 					    nvp->v_type == VDIR &&
2668 					    nvp->v_mountedhere != NULL) {
2669 						new_mp = nvp->v_mountedhere;
2670 						r = vfs_busy(new_mp, 0);
2671 						vput(nvp);
2672 						nvp = NULL;
2673 						if (r == 0) {
2674 							r = VFS_ROOT(new_mp,
2675 							    LK_SHARED, &nvp);
2676 							needs_unbusy = 1;
2677 							if (r == 0)
2678 								at_root = 1;
2679 						}
2680 					}
2681 				}
2682 
2683 				/*
2684 				 * If we failed to look up the entry, then it
2685 				 * has become invalid, most likely removed.
2686 				 */
2687 				if (r != 0) {
2688 					if (needs_unbusy)
2689 						vfs_unbusy(new_mp);
2690 					goto invalid;
2691 				}
2692 				KASSERT(refp != NULL || nvp != NULL,
2693 				    ("%s: undetected lookup error", __func__));
2694 
2695 				if (refp == NULL &&
2696 				    ((nd->nd_flag & ND_NFSV3) ||
2697 				     NFSNONZERO_ATTRBIT(&attrbits))) {
2698 					r = nfsvno_getfh(nvp, &nfh, p);
2699 					if (!r)
2700 					    r = nfsvno_getattr(nvp, nvap, nd, p,
2701 						1, &attrbits);
2702 					if (r == 0 && is_zfs == 1 &&
2703 					    nfsrv_enable_crossmntpt != 0 &&
2704 					    (nd->nd_flag & ND_NFSV4) != 0 &&
2705 					    nvp->v_type == VDIR &&
2706 					    vp->v_mount != nvp->v_mount) {
2707 					    /*
2708 					     * For a ZFS snapshot, there is a
2709 					     * pseudo mount that does not set
2710 					     * v_mountedhere, so it needs to
2711 					     * be detected via a different
2712 					     * mount structure.
2713 					     */
2714 					    at_root = 1;
2715 					    if (new_mp == mp)
2716 						new_mp = nvp->v_mount;
2717 					}
2718 				}
2719 
2720 				/*
2721 				 * If we failed to get attributes of the entry,
2722 				 * then just skip it for NFSv3 (the traditional
2723 				 * behavior in the old NFS server).
2724 				 * For NFSv4 the behavior is controlled by
2725 				 * RDATTRERROR: we either ignore the error or
2726 				 * fail the request.
2727 				 * The exception is EOPNOTSUPP, which can be
2728 				 * returned by nfsvno_getfh() for certain
2729 				 * file systems, such as devfs.  This indicates
2730 				 * that the file system cannot be exported,
2731 				 * so just skip over the entry.
2732 				 * Note that RDATTRERROR is never set for NFSv3.
2733 				 */
2734 				if (r != 0) {
2735 					if (!NFSISSET_ATTRBIT(&attrbits,
2736 					    NFSATTRBIT_RDATTRERROR) ||
2737 					    r == EOPNOTSUPP) {
2738 						vput(nvp);
2739 						if (needs_unbusy != 0)
2740 							vfs_unbusy(new_mp);
2741 						if ((nd->nd_flag & ND_NFSV3) ||
2742 						    r == EOPNOTSUPP)
2743 							goto invalid;
2744 						nd->nd_repstat = r;
2745 						break;
2746 					}
2747 				}
2748 			}
2749 
2750 			/*
2751 			 * Build the directory record xdr
2752 			 */
2753 			if (nd->nd_flag & ND_NFSV3) {
2754 				NFSM_BUILD(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
2755 				*tl++ = newnfs_true;
2756 				*tl++ = 0;
2757 				*tl = txdr_unsigned(dp->d_fileno);
2758 				dirlen += nfsm_strtom(nd, dp->d_name, nlen);
2759 				NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
2760 				txdr_hyper(*cookiep, tl);
2761 				nfsrv_postopattr(nd, 0, nvap);
2762 				dirlen += nfsm_fhtom(nd,(u_int8_t *)&nfh,0,1);
2763 				dirlen += (5*NFSX_UNSIGNED+NFSX_V3POSTOPATTR);
2764 				if (nvp != NULL)
2765 					vput(nvp);
2766 			} else {
2767 				NFSM_BUILD(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
2768 				*tl++ = newnfs_true;
2769 				txdr_hyper(*cookiep, tl);
2770 				dirlen += nfsm_strtom(nd, dp->d_name, nlen);
2771 				if (nvp != NULL) {
2772 					supports_nfsv4acls =
2773 					    nfs_supportsnfsv4acls(nvp);
2774 					NFSVOPUNLOCK(nvp);
2775 				} else
2776 					supports_nfsv4acls = 0;
2777 				if (refp != NULL) {
2778 					dirlen += nfsrv_putreferralattr(nd,
2779 					    &savbits, refp, 0,
2780 					    &nd->nd_repstat);
2781 					if (nd->nd_repstat) {
2782 						if (nvp != NULL)
2783 							vrele(nvp);
2784 						if (needs_unbusy != 0)
2785 							vfs_unbusy(new_mp);
2786 						break;
2787 					}
2788 				} else if (r) {
2789 					dirlen += nfsvno_fillattr(nd, new_mp,
2790 					    nvp, nvap, &nfh, r, &rderrbits,
2791 					    nd->nd_cred, p, isdgram, 0,
2792 					    supports_nfsv4acls, at_root,
2793 					    mounted_on_fileno);
2794 				} else {
2795 					dirlen += nfsvno_fillattr(nd, new_mp,
2796 					    nvp, nvap, &nfh, r, &attrbits,
2797 					    nd->nd_cred, p, isdgram, 0,
2798 					    supports_nfsv4acls, at_root,
2799 					    mounted_on_fileno);
2800 				}
2801 				if (nvp != NULL)
2802 					vrele(nvp);
2803 				dirlen += (3 * NFSX_UNSIGNED);
2804 			}
2805 			if (needs_unbusy != 0)
2806 				vfs_unbusy(new_mp);
2807 			if (dirlen <= cnt)
2808 				entrycnt++;
2809 		}
2810 invalid:
2811 		cpos += dp->d_reclen;
2812 		dp = (struct dirent *)cpos;
2813 		cookiep++;
2814 		ncookies--;
2815 	}
2816 	vrele(vp);
2817 	vfs_unbusy(mp);
2818 
2819 	/*
2820 	 * If dirlen > cnt, we must strip off the last entry. If that
2821 	 * results in an empty reply, report NFSERR_TOOSMALL.
2822 	 */
2823 	if (dirlen > cnt || nd->nd_repstat) {
2824 		if (!nd->nd_repstat && entrycnt == 0)
2825 			nd->nd_repstat = NFSERR_TOOSMALL;
2826 		if (nd->nd_repstat) {
2827 			nfsm_trimtrailing(nd, mb0, bpos0, bextpg0, bextpgsiz0);
2828 			if (nd->nd_flag & ND_NFSV3)
2829 				nfsrv_postopattr(nd, getret, &at);
2830 		} else
2831 			nfsm_trimtrailing(nd, mb1, bpos1, bextpg1, bextpgsiz1);
2832 		eofflag = 0;
2833 	} else if (cpos < cend)
2834 		eofflag = 0;
2835 	if (!nd->nd_repstat) {
2836 		NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
2837 		*tl++ = newnfs_false;
2838 		if (eofflag)
2839 			*tl = newnfs_true;
2840 		else
2841 			*tl = newnfs_false;
2842 	}
2843 	free(cookies, M_TEMP);
2844 	free(rbuf, M_TEMP);
2845 
2846 out:
2847 	NFSEXITCODE2(0, nd);
2848 	return (0);
2849 nfsmout:
2850 	vput(vp);
2851 	NFSEXITCODE2(error, nd);
2852 	return (error);
2853 }
2854 
2855 /*
2856  * Get the settable attributes out of the mbuf list.
2857  * (Return 0 or EBADRPC)
2858  */
2859 int
2860 nfsrv_sattr(struct nfsrv_descript *nd, vnode_t vp, struct nfsvattr *nvap,
2861     nfsattrbit_t *attrbitp, NFSACL_T *aclp, struct thread *p)
2862 {
2863 	u_int32_t *tl;
2864 	struct nfsv2_sattr *sp;
2865 	int error = 0, toclient = 0;
2866 
2867 	switch (nd->nd_flag & (ND_NFSV2 | ND_NFSV3 | ND_NFSV4)) {
2868 	case ND_NFSV2:
2869 		NFSM_DISSECT(sp, struct nfsv2_sattr *, NFSX_V2SATTR);
2870 		/*
2871 		 * Some old clients didn't fill in the high order 16bits.
2872 		 * --> check the low order 2 bytes for 0xffff
2873 		 */
2874 		if ((fxdr_unsigned(int, sp->sa_mode) & 0xffff) != 0xffff)
2875 			nvap->na_mode = nfstov_mode(sp->sa_mode);
2876 		if (sp->sa_uid != newnfs_xdrneg1)
2877 			nvap->na_uid = fxdr_unsigned(uid_t, sp->sa_uid);
2878 		if (sp->sa_gid != newnfs_xdrneg1)
2879 			nvap->na_gid = fxdr_unsigned(gid_t, sp->sa_gid);
2880 		if (sp->sa_size != newnfs_xdrneg1)
2881 			nvap->na_size = fxdr_unsigned(u_quad_t, sp->sa_size);
2882 		if (sp->sa_atime.nfsv2_sec != newnfs_xdrneg1) {
2883 #ifdef notyet
2884 			fxdr_nfsv2time(&sp->sa_atime, &nvap->na_atime);
2885 #else
2886 			nvap->na_atime.tv_sec =
2887 				fxdr_unsigned(u_int32_t,sp->sa_atime.nfsv2_sec);
2888 			nvap->na_atime.tv_nsec = 0;
2889 #endif
2890 		}
2891 		if (sp->sa_mtime.nfsv2_sec != newnfs_xdrneg1)
2892 			fxdr_nfsv2time(&sp->sa_mtime, &nvap->na_mtime);
2893 		break;
2894 	case ND_NFSV3:
2895 		NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2896 		if (*tl == newnfs_true) {
2897 			NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2898 			nvap->na_mode = nfstov_mode(*tl);
2899 		}
2900 		NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2901 		if (*tl == newnfs_true) {
2902 			NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2903 			nvap->na_uid = fxdr_unsigned(uid_t, *tl);
2904 		}
2905 		NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2906 		if (*tl == newnfs_true) {
2907 			NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2908 			nvap->na_gid = fxdr_unsigned(gid_t, *tl);
2909 		}
2910 		NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2911 		if (*tl == newnfs_true) {
2912 			NFSM_DISSECT(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
2913 			nvap->na_size = fxdr_hyper(tl);
2914 		}
2915 		NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2916 		switch (fxdr_unsigned(int, *tl)) {
2917 		case NFSV3SATTRTIME_TOCLIENT:
2918 			NFSM_DISSECT(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
2919 			fxdr_nfsv3time(tl, &nvap->na_atime);
2920 			toclient = 1;
2921 			break;
2922 		case NFSV3SATTRTIME_TOSERVER:
2923 			vfs_timestamp(&nvap->na_atime);
2924 			nvap->na_vaflags |= VA_UTIMES_NULL;
2925 			break;
2926 		}
2927 		NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2928 		switch (fxdr_unsigned(int, *tl)) {
2929 		case NFSV3SATTRTIME_TOCLIENT:
2930 			NFSM_DISSECT(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
2931 			fxdr_nfsv3time(tl, &nvap->na_mtime);
2932 			nvap->na_vaflags &= ~VA_UTIMES_NULL;
2933 			break;
2934 		case NFSV3SATTRTIME_TOSERVER:
2935 			vfs_timestamp(&nvap->na_mtime);
2936 			if (!toclient)
2937 				nvap->na_vaflags |= VA_UTIMES_NULL;
2938 			break;
2939 		}
2940 		break;
2941 	case ND_NFSV4:
2942 		error = nfsv4_sattr(nd, vp, nvap, attrbitp, aclp, p);
2943 	}
2944 nfsmout:
2945 	NFSEXITCODE2(error, nd);
2946 	return (error);
2947 }
2948 
2949 /*
2950  * Handle the setable attributes for V4.
2951  * Returns NFSERR_BADXDR if it can't be parsed, 0 otherwise.
2952  */
2953 int
2954 nfsv4_sattr(struct nfsrv_descript *nd, vnode_t vp, struct nfsvattr *nvap,
2955     nfsattrbit_t *attrbitp, NFSACL_T *aclp, struct thread *p)
2956 {
2957 	u_int32_t *tl;
2958 	int attrsum = 0;
2959 	int i, j;
2960 	int error, attrsize, bitpos, aclsize, aceerr, retnotsup = 0;
2961 	int moderet, toclient = 0;
2962 	u_char *cp, namestr[NFSV4_SMALLSTR + 1];
2963 	uid_t uid;
2964 	gid_t gid;
2965 	u_short mode, mask;		/* Same type as va_mode. */
2966 	struct vattr va;
2967 
2968 	error = nfsrv_getattrbits(nd, attrbitp, NULL, &retnotsup);
2969 	if (error)
2970 		goto nfsmout;
2971 	NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2972 	attrsize = fxdr_unsigned(int, *tl);
2973 
2974 	/*
2975 	 * Loop around getting the setable attributes. If an unsupported
2976 	 * one is found, set nd_repstat == NFSERR_ATTRNOTSUPP and return.
2977 	 */
2978 	if (retnotsup) {
2979 		nd->nd_repstat = NFSERR_ATTRNOTSUPP;
2980 		bitpos = NFSATTRBIT_MAX;
2981 	} else {
2982 		bitpos = 0;
2983 	}
2984 	moderet = 0;
2985 	for (; bitpos < NFSATTRBIT_MAX; bitpos++) {
2986 	    if (attrsum > attrsize) {
2987 		error = NFSERR_BADXDR;
2988 		goto nfsmout;
2989 	    }
2990 	    if (NFSISSET_ATTRBIT(attrbitp, bitpos))
2991 		switch (bitpos) {
2992 		case NFSATTRBIT_SIZE:
2993 			NFSM_DISSECT(tl, u_int32_t *, NFSX_HYPER);
2994                      if (vp != NULL && vp->v_type != VREG) {
2995                             error = (vp->v_type == VDIR) ? NFSERR_ISDIR :
2996                                 NFSERR_INVAL;
2997                             goto nfsmout;
2998 			}
2999 			nvap->na_size = fxdr_hyper(tl);
3000 			attrsum += NFSX_HYPER;
3001 			break;
3002 		case NFSATTRBIT_ACL:
3003 			error = nfsrv_dissectacl(nd, aclp, true, &aceerr,
3004 			    &aclsize, p);
3005 			if (error)
3006 				goto nfsmout;
3007 			if (aceerr && !nd->nd_repstat)
3008 				nd->nd_repstat = aceerr;
3009 			attrsum += aclsize;
3010 			break;
3011 		case NFSATTRBIT_ARCHIVE:
3012 			NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
3013 			if (!nd->nd_repstat)
3014 				nd->nd_repstat = NFSERR_ATTRNOTSUPP;
3015 			attrsum += NFSX_UNSIGNED;
3016 			break;
3017 		case NFSATTRBIT_HIDDEN:
3018 			NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
3019 			if (!nd->nd_repstat)
3020 				nd->nd_repstat = NFSERR_ATTRNOTSUPP;
3021 			attrsum += NFSX_UNSIGNED;
3022 			break;
3023 		case NFSATTRBIT_MIMETYPE:
3024 			NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
3025 			i = fxdr_unsigned(int, *tl);
3026 			error = nfsm_advance(nd, NFSM_RNDUP(i), -1);
3027 			if (error)
3028 				goto nfsmout;
3029 			if (!nd->nd_repstat)
3030 				nd->nd_repstat = NFSERR_ATTRNOTSUPP;
3031 			attrsum += (NFSX_UNSIGNED + NFSM_RNDUP(i));
3032 			break;
3033 		case NFSATTRBIT_MODE:
3034 			moderet = NFSERR_INVAL;	/* Can't do MODESETMASKED. */
3035 			NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
3036 			nvap->na_mode = nfstov_mode(*tl);
3037 			attrsum += NFSX_UNSIGNED;
3038 			break;
3039 		case NFSATTRBIT_OWNER:
3040 			NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
3041 			j = fxdr_unsigned(int, *tl);
3042 			if (j < 0) {
3043 				error = NFSERR_BADXDR;
3044 				goto nfsmout;
3045 			}
3046 			if (j > NFSV4_SMALLSTR)
3047 				cp = malloc(j + 1, M_NFSSTRING, M_WAITOK);
3048 			else
3049 				cp = namestr;
3050 			error = nfsrv_mtostr(nd, cp, j);
3051 			if (error) {
3052 				if (j > NFSV4_SMALLSTR)
3053 					free(cp, M_NFSSTRING);
3054 				goto nfsmout;
3055 			}
3056 			if (!nd->nd_repstat) {
3057 				nd->nd_repstat = nfsv4_strtouid(nd, cp, j,
3058 				    &uid);
3059 				if (!nd->nd_repstat)
3060 					nvap->na_uid = uid;
3061 			}
3062 			if (j > NFSV4_SMALLSTR)
3063 				free(cp, M_NFSSTRING);
3064 			attrsum += (NFSX_UNSIGNED + NFSM_RNDUP(j));
3065 			break;
3066 		case NFSATTRBIT_OWNERGROUP:
3067 			NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
3068 			j = fxdr_unsigned(int, *tl);
3069 			if (j < 0) {
3070 				error = NFSERR_BADXDR;
3071 				goto nfsmout;
3072 			}
3073 			if (j > NFSV4_SMALLSTR)
3074 				cp = malloc(j + 1, M_NFSSTRING, M_WAITOK);
3075 			else
3076 				cp = namestr;
3077 			error = nfsrv_mtostr(nd, cp, j);
3078 			if (error) {
3079 				if (j > NFSV4_SMALLSTR)
3080 					free(cp, M_NFSSTRING);
3081 				goto nfsmout;
3082 			}
3083 			if (!nd->nd_repstat) {
3084 				nd->nd_repstat = nfsv4_strtogid(nd, cp, j,
3085 				    &gid);
3086 				if (!nd->nd_repstat)
3087 					nvap->na_gid = gid;
3088 			}
3089 			if (j > NFSV4_SMALLSTR)
3090 				free(cp, M_NFSSTRING);
3091 			attrsum += (NFSX_UNSIGNED + NFSM_RNDUP(j));
3092 			break;
3093 		case NFSATTRBIT_SYSTEM:
3094 			NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
3095 			if (!nd->nd_repstat)
3096 				nd->nd_repstat = NFSERR_ATTRNOTSUPP;
3097 			attrsum += NFSX_UNSIGNED;
3098 			break;
3099 		case NFSATTRBIT_TIMEACCESSSET:
3100 			NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
3101 			attrsum += NFSX_UNSIGNED;
3102 			if (fxdr_unsigned(int, *tl)==NFSV4SATTRTIME_TOCLIENT) {
3103 			    NFSM_DISSECT(tl, u_int32_t *, NFSX_V4TIME);
3104 			    fxdr_nfsv4time(tl, &nvap->na_atime);
3105 			    toclient = 1;
3106 			    attrsum += NFSX_V4TIME;
3107 			} else {
3108 			    vfs_timestamp(&nvap->na_atime);
3109 			    nvap->na_vaflags |= VA_UTIMES_NULL;
3110 			}
3111 			break;
3112 		case NFSATTRBIT_TIMEBACKUP:
3113 			NFSM_DISSECT(tl, u_int32_t *, NFSX_V4TIME);
3114 			if (!nd->nd_repstat)
3115 				nd->nd_repstat = NFSERR_ATTRNOTSUPP;
3116 			attrsum += NFSX_V4TIME;
3117 			break;
3118 		case NFSATTRBIT_TIMECREATE:
3119 			NFSM_DISSECT(tl, u_int32_t *, NFSX_V4TIME);
3120 			fxdr_nfsv4time(tl, &nvap->na_btime);
3121 			attrsum += NFSX_V4TIME;
3122 			break;
3123 		case NFSATTRBIT_TIMEMODIFYSET:
3124 			NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
3125 			attrsum += NFSX_UNSIGNED;
3126 			if (fxdr_unsigned(int, *tl)==NFSV4SATTRTIME_TOCLIENT) {
3127 			    NFSM_DISSECT(tl, u_int32_t *, NFSX_V4TIME);
3128 			    fxdr_nfsv4time(tl, &nvap->na_mtime);
3129 			    nvap->na_vaflags &= ~VA_UTIMES_NULL;
3130 			    attrsum += NFSX_V4TIME;
3131 			} else {
3132 			    vfs_timestamp(&nvap->na_mtime);
3133 			    if (!toclient)
3134 				nvap->na_vaflags |= VA_UTIMES_NULL;
3135 			}
3136 			break;
3137 		case NFSATTRBIT_MODESETMASKED:
3138 			NFSM_DISSECT(tl, uint32_t *, 2 * NFSX_UNSIGNED);
3139 			mode = fxdr_unsigned(u_short, *tl++);
3140 			mask = fxdr_unsigned(u_short, *tl);
3141 			/*
3142 			 * vp == NULL implies an Open/Create operation.
3143 			 * This attribute can only be used for Setattr and
3144 			 * only for NFSv4.1 or higher.
3145 			 * If moderet != 0, a mode attribute has also been
3146 			 * specified and this attribute cannot be done in the
3147 			 * same Setattr operation.
3148 			 */
3149 			if ((nd->nd_flag & ND_NFSV41) == 0)
3150 				nd->nd_repstat = NFSERR_ATTRNOTSUPP;
3151 			else if ((mode & ~07777) != 0 || (mask & ~07777) != 0 ||
3152 			    vp == NULL)
3153 				nd->nd_repstat = NFSERR_INVAL;
3154 			else if (moderet == 0)
3155 				moderet = VOP_GETATTR(vp, &va, nd->nd_cred);
3156 			if (moderet == 0)
3157 				nvap->na_mode = (mode & mask) |
3158 				    (va.va_mode & ~mask);
3159 			else
3160 				nd->nd_repstat = moderet;
3161 			attrsum += 2 * NFSX_UNSIGNED;
3162 			break;
3163 		default:
3164 			nd->nd_repstat = NFSERR_ATTRNOTSUPP;
3165 			/*
3166 			 * set bitpos so we drop out of the loop.
3167 			 */
3168 			bitpos = NFSATTRBIT_MAX;
3169 			break;
3170 		}
3171 	}
3172 
3173 	/*
3174 	 * some clients pad the attrlist, so we need to skip over the
3175 	 * padding.
3176 	 */
3177 	if (attrsum > attrsize) {
3178 		error = NFSERR_BADXDR;
3179 	} else {
3180 		attrsize = NFSM_RNDUP(attrsize);
3181 		if (attrsum < attrsize)
3182 			error = nfsm_advance(nd, attrsize - attrsum, -1);
3183 	}
3184 nfsmout:
3185 	NFSEXITCODE2(error, nd);
3186 	return (error);
3187 }
3188 
3189 /*
3190  * Check/setup export credentials.
3191  */
3192 int
3193 nfsd_excred(struct nfsrv_descript *nd, struct nfsexstuff *exp,
3194     struct ucred *credanon, bool testsec)
3195 {
3196 	int error;
3197 
3198 	/*
3199 	 * Check/setup credentials.
3200 	 */
3201 	if (nd->nd_flag & ND_GSS)
3202 		exp->nes_exflag &= ~MNT_EXPORTANON;
3203 
3204 	/*
3205 	 * Check to see if the operation is allowed for this security flavor.
3206 	 */
3207 	error = 0;
3208 	if (testsec) {
3209 		error = nfsvno_testexp(nd, exp);
3210 		if (error != 0)
3211 			goto out;
3212 	}
3213 
3214 	/*
3215 	 * Check to see if the file system is exported V4 only.
3216 	 */
3217 	if (NFSVNO_EXV4ONLY(exp) && !(nd->nd_flag & ND_NFSV4)) {
3218 		error = NFSERR_PROGNOTV4;
3219 		goto out;
3220 	}
3221 
3222 	/*
3223 	 * Now, map the user credentials.
3224 	 * (Note that ND_AUTHNONE will only be set for an NFSv3
3225 	 *  Fsinfo RPC. If set for anything else, this code might need
3226 	 *  to change.)
3227 	 */
3228 	if (NFSVNO_EXPORTED(exp)) {
3229 		if (((nd->nd_flag & ND_GSS) == 0 && nd->nd_cred->cr_uid == 0) ||
3230 		     NFSVNO_EXPORTANON(exp) ||
3231 		     (nd->nd_flag & ND_AUTHNONE) != 0) {
3232 			nd->nd_cred->cr_uid = credanon->cr_uid;
3233 			nd->nd_cred->cr_gid = credanon->cr_gid;
3234 			crsetgroups(nd->nd_cred, credanon->cr_ngroups,
3235 			    credanon->cr_groups);
3236 		} else if ((nd->nd_flag & ND_GSS) == 0) {
3237 			/*
3238 			 * If using AUTH_SYS, call nfsrv_getgrpscred() to see
3239 			 * if there is a replacement credential with a group
3240 			 * list set up by "nfsuserd -manage-gids".
3241 			 * If there is no replacement, nfsrv_getgrpscred()
3242 			 * simply returns its argument.
3243 			 */
3244 			nd->nd_cred = nfsrv_getgrpscred(nd->nd_cred);
3245 		}
3246 	}
3247 
3248 out:
3249 	NFSEXITCODE2(error, nd);
3250 	return (error);
3251 }
3252 
3253 /*
3254  * Check exports.
3255  */
3256 int
3257 nfsvno_checkexp(struct mount *mp, struct sockaddr *nam, struct nfsexstuff *exp,
3258     struct ucred **credp)
3259 {
3260 	int error;
3261 
3262 	error = 0;
3263 	*credp = NULL;
3264 	MNT_ILOCK(mp);
3265 	if (mp->mnt_exjail == NULL ||
3266 	    mp->mnt_exjail->cr_prison != curthread->td_ucred->cr_prison)
3267 		error = EACCES;
3268 	MNT_IUNLOCK(mp);
3269 	if (error == 0)
3270 		error = VFS_CHECKEXP(mp, nam, &exp->nes_exflag, credp,
3271 		    &exp->nes_numsecflavor, exp->nes_secflavors);
3272 	if (error) {
3273 		if (NFSD_VNET(nfs_rootfhset)) {
3274 			exp->nes_exflag = 0;
3275 			exp->nes_numsecflavor = 0;
3276 			error = 0;
3277 		}
3278 	} else if (exp->nes_numsecflavor < 1 || exp->nes_numsecflavor >
3279 	    MAXSECFLAVORS) {
3280 		printf("nfsvno_checkexp: numsecflavors out of range\n");
3281 		exp->nes_numsecflavor = 0;
3282 		error = EACCES;
3283 	}
3284 	NFSEXITCODE(error);
3285 	return (error);
3286 }
3287 
3288 /*
3289  * Get a vnode for a file handle and export stuff.
3290  */
3291 int
3292 nfsvno_fhtovp(struct mount *mp, fhandle_t *fhp, struct sockaddr *nam,
3293     int lktype, struct vnode **vpp, struct nfsexstuff *exp,
3294     struct ucred **credp)
3295 {
3296 	int error;
3297 
3298 	*credp = NULL;
3299 	exp->nes_numsecflavor = 0;
3300 	error = VFS_FHTOVP(mp, &fhp->fh_fid, lktype, vpp);
3301 	if (error != 0)
3302 		/* Make sure the server replies ESTALE to the client. */
3303 		error = ESTALE;
3304 	if (nam && !error) {
3305 		MNT_ILOCK(mp);
3306 		if (mp->mnt_exjail == NULL ||
3307 		    mp->mnt_exjail->cr_prison != curthread->td_ucred->cr_prison)
3308 			error = EACCES;
3309 		MNT_IUNLOCK(mp);
3310 		if (error == 0)
3311 			error = VFS_CHECKEXP(mp, nam, &exp->nes_exflag, credp,
3312 			    &exp->nes_numsecflavor, exp->nes_secflavors);
3313 		if (error) {
3314 			if (NFSD_VNET(nfs_rootfhset)) {
3315 				exp->nes_exflag = 0;
3316 				exp->nes_numsecflavor = 0;
3317 				error = 0;
3318 			} else {
3319 				vput(*vpp);
3320 			}
3321 		} else if (exp->nes_numsecflavor < 1 || exp->nes_numsecflavor >
3322 		    MAXSECFLAVORS) {
3323 			printf("nfsvno_fhtovp: numsecflavors out of range\n");
3324 			exp->nes_numsecflavor = 0;
3325 			error = EACCES;
3326 			vput(*vpp);
3327 		}
3328 	}
3329 	NFSEXITCODE(error);
3330 	return (error);
3331 }
3332 
3333 /*
3334  * nfsd_fhtovp() - convert a fh to a vnode ptr
3335  * 	- look up fsid in mount list (if not found ret error)
3336  *	- get vp and export rights by calling nfsvno_fhtovp()
3337  *	- if cred->cr_uid == 0 or MNT_EXPORTANON set it to credanon
3338  *	  for AUTH_SYS
3339  *	- if mpp != NULL, return the mount point so that it can
3340  *	  be used for vn_finished_write() by the caller
3341  */
3342 void
3343 nfsd_fhtovp(struct nfsrv_descript *nd, struct nfsrvfh *nfp, int lktype,
3344     struct vnode **vpp, struct nfsexstuff *exp,
3345     struct mount **mpp, int startwrite, int nextop)
3346 {
3347 	struct mount *mp, *mpw;
3348 	struct ucred *credanon;
3349 	fhandle_t *fhp;
3350 	int error;
3351 
3352 	if (mpp != NULL)
3353 		*mpp = NULL;
3354 	*vpp = NULL;
3355 	fhp = (fhandle_t *)nfp->nfsrvfh_data;
3356 	mp = vfs_busyfs(&fhp->fh_fsid);
3357 	if (mp == NULL) {
3358 		nd->nd_repstat = ESTALE;
3359 		goto out;
3360 	}
3361 
3362 	if (startwrite) {
3363 		mpw = mp;
3364 		error = vn_start_write(NULL, &mpw, V_WAIT);
3365 		if (error != 0) {
3366 			mpw = NULL;
3367 			vfs_unbusy(mp);
3368 			nd->nd_repstat = ESTALE;
3369 			goto out;
3370 		}
3371 		if (lktype == LK_SHARED && !(MNT_SHARED_WRITES(mp)))
3372 			lktype = LK_EXCLUSIVE;
3373 	} else
3374 		mpw = NULL;
3375 
3376 	nd->nd_repstat = nfsvno_fhtovp(mp, fhp, nd->nd_nam, lktype, vpp, exp,
3377 	    &credanon);
3378 	vfs_unbusy(mp);
3379 
3380 	/*
3381 	 * For NFSv4 without a pseudo root fs, unexported file handles
3382 	 * can be returned, so that Lookup works everywhere.
3383 	 */
3384 	if (!nd->nd_repstat && exp->nes_exflag == 0 &&
3385 	    !(nd->nd_flag & ND_NFSV4)) {
3386 		vput(*vpp);
3387 		*vpp = NULL;
3388 		nd->nd_repstat = EACCES;
3389 	}
3390 
3391 	/*
3392 	 * Personally, I've never seen any point in requiring a
3393 	 * reserved port#, since only in the rare case where the
3394 	 * clients are all boxes with secure system privileges,
3395 	 * does it provide any enhanced security, but... some people
3396 	 * believe it to be useful and keep putting this code back in.
3397 	 * (There is also some "security checker" out there that
3398 	 *  complains if the nfs server doesn't enforce this.)
3399 	 * However, note the following:
3400 	 * RFC3530 (NFSv4) specifies that a reserved port# not be
3401 	 *	required.
3402 	 * RFC2623 recommends that, if a reserved port# is checked for,
3403 	 *	that there be a way to turn that off--> ifdef'd.
3404 	 */
3405 #ifdef NFS_REQRSVPORT
3406 	if (!nd->nd_repstat) {
3407 		struct sockaddr_in *saddr;
3408 		struct sockaddr_in6 *saddr6;
3409 
3410 		saddr = NFSSOCKADDR(nd->nd_nam, struct sockaddr_in *);
3411 		saddr6 = NFSSOCKADDR(nd->nd_nam, struct sockaddr_in6 *);
3412 		if (!(nd->nd_flag & ND_NFSV4) &&
3413 		    ((saddr->sin_family == AF_INET &&
3414 		      ntohs(saddr->sin_port) >= IPPORT_RESERVED) ||
3415 		     (saddr6->sin6_family == AF_INET6 &&
3416 		      ntohs(saddr6->sin6_port) >= IPPORT_RESERVED))) {
3417 			vput(*vpp);
3418 			nd->nd_repstat = (NFSERR_AUTHERR | AUTH_TOOWEAK);
3419 		}
3420 	}
3421 #endif	/* NFS_REQRSVPORT */
3422 
3423 	/*
3424 	 * Check/setup credentials.
3425 	 */
3426 	if (!nd->nd_repstat) {
3427 		nd->nd_saveduid = nd->nd_cred->cr_uid;
3428 		nd->nd_repstat = nfsd_excred(nd, exp, credanon,
3429 		    nfsrv_checkwrongsec(nd, nextop, (*vpp)->v_type));
3430 		if (nd->nd_repstat)
3431 			vput(*vpp);
3432 	}
3433 	if (credanon != NULL)
3434 		crfree(credanon);
3435 	if (nd->nd_repstat) {
3436 		vn_finished_write(mpw);
3437 		*vpp = NULL;
3438 	} else if (mpp != NULL) {
3439 		*mpp = mpw;
3440 	}
3441 
3442 out:
3443 	NFSEXITCODE2(0, nd);
3444 }
3445 
3446 /*
3447  * glue for fp.
3448  */
3449 static int
3450 fp_getfvp(struct thread *p, int fd, struct file **fpp, struct vnode **vpp)
3451 {
3452 	struct filedesc *fdp;
3453 	struct file *fp;
3454 	int error = 0;
3455 
3456 	fdp = p->td_proc->p_fd;
3457 	if (fd < 0 || fd >= fdp->fd_nfiles ||
3458 	    (fp = fdp->fd_ofiles[fd].fde_file) == NULL) {
3459 		error = EBADF;
3460 		goto out;
3461 	}
3462 	*fpp = fp;
3463 
3464 out:
3465 	NFSEXITCODE(error);
3466 	return (error);
3467 }
3468 
3469 /*
3470  * Called from nfssvc() to update the exports list. Just call
3471  * vfs_export(). This has to be done, since the v4 root fake fs isn't
3472  * in the mount list.
3473  */
3474 int
3475 nfsrv_v4rootexport(void *argp, struct ucred *cred, struct thread *p)
3476 {
3477 	struct nfsex_args *nfsexargp = (struct nfsex_args *)argp;
3478 	int error = 0;
3479 	struct nameidata nd;
3480 	fhandle_t fh;
3481 
3482 	error = vfs_export(NFSD_VNET(nfsv4root_mnt), &nfsexargp->export, false);
3483 	if ((nfsexargp->export.ex_flags & MNT_DELEXPORT) != 0)
3484 		NFSD_VNET(nfs_rootfhset) = 0;
3485 	else if (error == 0) {
3486 		if (nfsexargp->fspec == NULL) {
3487 			error = EPERM;
3488 			goto out;
3489 		}
3490 		/*
3491 		 * If fspec != NULL, this is the v4root path.
3492 		 */
3493 		NDINIT(&nd, LOOKUP, FOLLOW, UIO_USERSPACE, nfsexargp->fspec);
3494 		if ((error = namei(&nd)) != 0)
3495 			goto out;
3496 		error = nfsvno_getfh(nd.ni_vp, &fh, p);
3497 		vrele(nd.ni_vp);
3498 		if (!error) {
3499 			NFSD_VNET(nfs_rootfh).nfsrvfh_len = NFSX_MYFH;
3500 			NFSBCOPY((caddr_t)&fh,
3501 			    NFSD_VNET(nfs_rootfh).nfsrvfh_data,
3502 			    sizeof (fhandle_t));
3503 			NFSD_VNET(nfs_rootfhset) = 1;
3504 		}
3505 	}
3506 
3507 out:
3508 	NFSEXITCODE(error);
3509 	return (error);
3510 }
3511 
3512 /*
3513  * This function needs to test to see if the system is near its limit
3514  * for memory allocation via malloc() or mget() and return True iff
3515  * either of these resources are near their limit.
3516  * XXX (For now, this is just a stub.)
3517  */
3518 int nfsrv_testmalloclimit = 0;
3519 int
3520 nfsrv_mallocmget_limit(void)
3521 {
3522 	static int printmesg = 0;
3523 	static int testval = 1;
3524 
3525 	if (nfsrv_testmalloclimit && (testval++ % 1000) == 0) {
3526 		if ((printmesg++ % 100) == 0)
3527 			printf("nfsd: malloc/mget near limit\n");
3528 		return (1);
3529 	}
3530 	return (0);
3531 }
3532 
3533 /*
3534  * BSD specific initialization of a mount point.
3535  */
3536 void
3537 nfsd_mntinit(void)
3538 {
3539 
3540 	NFSD_LOCK();
3541 	if (NFSD_VNET(nfsrv_mntinited)) {
3542 		NFSD_UNLOCK();
3543 		return;
3544 	}
3545 	NFSD_VNET(nfsrv_mntinited) = true;
3546 	nfsrvd_init(0);
3547 	NFSD_UNLOCK();
3548 
3549 	NFSD_VNET(nfsv4root_mnt) = malloc(sizeof(struct mount), M_TEMP,
3550 	    M_WAITOK | M_ZERO);
3551 	NFSD_VNET(nfsv4root_mnt)->mnt_flag = (MNT_RDONLY | MNT_EXPORTED);
3552 	mtx_init(&NFSD_VNET(nfsv4root_mnt)->mnt_mtx, "nfs4mnt", NULL, MTX_DEF);
3553 	lockinit(&NFSD_VNET(nfsv4root_mnt)->mnt_explock, PVFS, "explock", 0, 0);
3554 	TAILQ_INIT(&NFSD_VNET(nfsv4root_mnt)->mnt_nvnodelist);
3555 	TAILQ_INIT(&NFSD_VNET(nfsv4root_mnt)->mnt_lazyvnodelist);
3556 	NFSD_VNET(nfsv4root_mnt)->mnt_export = NULL;
3557 	TAILQ_INIT(&NFSD_VNET(nfsv4root_opt));
3558 	TAILQ_INIT(&NFSD_VNET(nfsv4root_newopt));
3559 	NFSD_VNET(nfsv4root_mnt)->mnt_opt = &NFSD_VNET(nfsv4root_opt);
3560 	NFSD_VNET(nfsv4root_mnt)->mnt_optnew = &NFSD_VNET(nfsv4root_newopt);
3561 	NFSD_VNET(nfsv4root_mnt)->mnt_nvnodelistsize = 0;
3562 	NFSD_VNET(nfsv4root_mnt)->mnt_lazyvnodelistsize = 0;
3563 	callout_init(&NFSD_VNET(nfsd_callout), 1);
3564 
3565 	nfsrvd_initcache();
3566 	nfsd_init();
3567 }
3568 
3569 static void
3570 nfsd_timer(void *arg)
3571 {
3572 	struct vnet *vnetp;
3573 
3574 	vnetp = (struct vnet *)arg;
3575 	NFSD_CURVNET_SET_QUIET(vnetp);
3576 	nfsrv_servertimer(vnetp);
3577 	callout_reset_sbt(&NFSD_VNET(nfsd_callout), SBT_1S, SBT_1S, nfsd_timer,
3578 	    arg, 0);
3579 	NFSD_CURVNET_RESTORE();
3580 }
3581 
3582 /*
3583  * Get a vnode for a file handle, without checking exports, etc.
3584  */
3585 struct vnode *
3586 nfsvno_getvp(fhandle_t *fhp)
3587 {
3588 	struct mount *mp;
3589 	struct vnode *vp;
3590 	int error;
3591 
3592 	mp = vfs_busyfs(&fhp->fh_fsid);
3593 	if (mp == NULL)
3594 		return (NULL);
3595 	error = VFS_FHTOVP(mp, &fhp->fh_fid, LK_EXCLUSIVE, &vp);
3596 	vfs_unbusy(mp);
3597 	if (error)
3598 		return (NULL);
3599 	return (vp);
3600 }
3601 
3602 /*
3603  * Do a local VOP_ADVLOCK().
3604  */
3605 int
3606 nfsvno_advlock(struct vnode *vp, int ftype, u_int64_t first,
3607     u_int64_t end, struct thread *td)
3608 {
3609 	int error = 0;
3610 	struct flock fl;
3611 	u_int64_t tlen;
3612 
3613 	if (nfsrv_dolocallocks == 0)
3614 		goto out;
3615 	ASSERT_VOP_UNLOCKED(vp, "nfsvno_advlock: vp locked");
3616 
3617 	fl.l_whence = SEEK_SET;
3618 	fl.l_type = ftype;
3619 	fl.l_start = (off_t)first;
3620 	if (end == NFS64BITSSET) {
3621 		fl.l_len = 0;
3622 	} else {
3623 		tlen = end - first;
3624 		fl.l_len = (off_t)tlen;
3625 	}
3626 	/*
3627 	 * For FreeBSD8, the l_pid and l_sysid must be set to the same
3628 	 * values for all calls, so that all locks will be held by the
3629 	 * nfsd server. (The nfsd server handles conflicts between the
3630 	 * various clients.)
3631 	 * Since an NFSv4 lockowner is a ClientID plus an array of up to 1024
3632 	 * bytes, so it can't be put in l_sysid.
3633 	 */
3634 	if (nfsv4_sysid == 0)
3635 		nfsv4_sysid = nlm_acquire_next_sysid();
3636 	fl.l_pid = (pid_t)0;
3637 	fl.l_sysid = (int)nfsv4_sysid;
3638 
3639 	if (ftype == F_UNLCK)
3640 		error = VOP_ADVLOCK(vp, (caddr_t)td->td_proc, F_UNLCK, &fl,
3641 		    (F_POSIX | F_REMOTE));
3642 	else
3643 		error = VOP_ADVLOCK(vp, (caddr_t)td->td_proc, F_SETLK, &fl,
3644 		    (F_POSIX | F_REMOTE));
3645 
3646 out:
3647 	NFSEXITCODE(error);
3648 	return (error);
3649 }
3650 
3651 /*
3652  * Check the nfsv4 root exports.
3653  */
3654 int
3655 nfsvno_v4rootexport(struct nfsrv_descript *nd)
3656 {
3657 	struct ucred *credanon;
3658 	int error = 0, numsecflavor, secflavors[MAXSECFLAVORS], i;
3659 	uint64_t exflags;
3660 
3661 	error = vfs_stdcheckexp(NFSD_VNET(nfsv4root_mnt), nd->nd_nam, &exflags,
3662 	    &credanon, &numsecflavor, secflavors);
3663 	if (error) {
3664 		error = NFSERR_PROGUNAVAIL;
3665 		goto out;
3666 	}
3667 	if (credanon != NULL)
3668 		crfree(credanon);
3669 	for (i = 0; i < numsecflavor; i++) {
3670 		if (secflavors[i] == AUTH_SYS)
3671 			nd->nd_flag |= ND_EXAUTHSYS;
3672 		else if (secflavors[i] == RPCSEC_GSS_KRB5)
3673 			nd->nd_flag |= ND_EXGSS;
3674 		else if (secflavors[i] == RPCSEC_GSS_KRB5I)
3675 			nd->nd_flag |= ND_EXGSSINTEGRITY;
3676 		else if (secflavors[i] == RPCSEC_GSS_KRB5P)
3677 			nd->nd_flag |= ND_EXGSSPRIVACY;
3678 	}
3679 
3680 	/* And set ND_EXxx flags for TLS. */
3681 	if ((exflags & MNT_EXTLS) != 0) {
3682 		nd->nd_flag |= ND_EXTLS;
3683 		if ((exflags & MNT_EXTLSCERT) != 0)
3684 			nd->nd_flag |= ND_EXTLSCERT;
3685 		if ((exflags & MNT_EXTLSCERTUSER) != 0)
3686 			nd->nd_flag |= ND_EXTLSCERTUSER;
3687 	}
3688 
3689 out:
3690 	NFSEXITCODE(error);
3691 	return (error);
3692 }
3693 
3694 /*
3695  * Nfs server pseudo system call for the nfsd's
3696  */
3697 /*
3698  * MPSAFE
3699  */
3700 static int
3701 nfssvc_nfsd(struct thread *td, struct nfssvc_args *uap)
3702 {
3703 	struct file *fp;
3704 	struct nfsd_addsock_args sockarg;
3705 	struct nfsd_nfsd_args nfsdarg;
3706 	struct nfsd_nfsd_oargs onfsdarg;
3707 	struct nfsd_pnfsd_args pnfsdarg;
3708 	struct vnode *vp, *nvp, *curdvp;
3709 	struct pnfsdsfile *pf;
3710 	struct nfsdevice *ds, *fds;
3711 	cap_rights_t rights;
3712 	int buflen, error, ret;
3713 	char *buf, *cp, *cp2, *cp3;
3714 	char fname[PNFS_FILENAME_LEN + 1];
3715 
3716 	NFSD_CURVNET_SET(NFSD_TD_TO_VNET(td));
3717 	if (uap->flag & NFSSVC_NFSDADDSOCK) {
3718 		error = copyin(uap->argp, (caddr_t)&sockarg, sizeof (sockarg));
3719 		if (error)
3720 			goto out;
3721 		/*
3722 		 * Since we don't know what rights might be required,
3723 		 * pretend that we need them all. It is better to be too
3724 		 * careful than too reckless.
3725 		 */
3726 		error = fget(td, sockarg.sock,
3727 		    cap_rights_init_one(&rights, CAP_SOCK_SERVER), &fp);
3728 		if (error != 0)
3729 			goto out;
3730 		if (fp->f_type != DTYPE_SOCKET) {
3731 			fdrop(fp, td);
3732 			error = EPERM;
3733 			goto out;
3734 		}
3735 		error = nfsrvd_addsock(fp);
3736 		fdrop(fp, td);
3737 	} else if (uap->flag & NFSSVC_NFSDNFSD) {
3738 		if (uap->argp == NULL) {
3739 			error = EINVAL;
3740 			goto out;
3741 		}
3742 		if ((uap->flag & NFSSVC_NEWSTRUCT) == 0) {
3743 			error = copyin(uap->argp, &onfsdarg, sizeof(onfsdarg));
3744 			if (error == 0) {
3745 				nfsdarg.principal = onfsdarg.principal;
3746 				nfsdarg.minthreads = onfsdarg.minthreads;
3747 				nfsdarg.maxthreads = onfsdarg.maxthreads;
3748 				nfsdarg.version = 1;
3749 				nfsdarg.addr = NULL;
3750 				nfsdarg.addrlen = 0;
3751 				nfsdarg.dnshost = NULL;
3752 				nfsdarg.dnshostlen = 0;
3753 				nfsdarg.dspath = NULL;
3754 				nfsdarg.dspathlen = 0;
3755 				nfsdarg.mdspath = NULL;
3756 				nfsdarg.mdspathlen = 0;
3757 				nfsdarg.mirrorcnt = 1;
3758 			}
3759 		} else
3760 			error = copyin(uap->argp, &nfsdarg, sizeof(nfsdarg));
3761 		if (error)
3762 			goto out;
3763 		if (nfsdarg.addrlen > 0 && nfsdarg.addrlen < 10000 &&
3764 		    nfsdarg.dnshostlen > 0 && nfsdarg.dnshostlen < 10000 &&
3765 		    nfsdarg.dspathlen > 0 && nfsdarg.dspathlen < 10000 &&
3766 		    nfsdarg.mdspathlen > 0 && nfsdarg.mdspathlen < 10000 &&
3767 		    nfsdarg.mirrorcnt >= 1 &&
3768 		    nfsdarg.mirrorcnt <= NFSDEV_MAXMIRRORS &&
3769 		    nfsdarg.addr != NULL && nfsdarg.dnshost != NULL &&
3770 		    nfsdarg.dspath != NULL && nfsdarg.mdspath != NULL) {
3771 			NFSD_DEBUG(1, "addrlen=%d dspathlen=%d dnslen=%d"
3772 			    " mdspathlen=%d mirrorcnt=%d\n", nfsdarg.addrlen,
3773 			    nfsdarg.dspathlen, nfsdarg.dnshostlen,
3774 			    nfsdarg.mdspathlen, nfsdarg.mirrorcnt);
3775 			cp = malloc(nfsdarg.addrlen + 1, M_TEMP, M_WAITOK);
3776 			error = copyin(nfsdarg.addr, cp, nfsdarg.addrlen);
3777 			if (error != 0) {
3778 				free(cp, M_TEMP);
3779 				goto out;
3780 			}
3781 			cp[nfsdarg.addrlen] = '\0';	/* Ensure nul term. */
3782 			nfsdarg.addr = cp;
3783 			cp = malloc(nfsdarg.dnshostlen + 1, M_TEMP, M_WAITOK);
3784 			error = copyin(nfsdarg.dnshost, cp, nfsdarg.dnshostlen);
3785 			if (error != 0) {
3786 				free(nfsdarg.addr, M_TEMP);
3787 				free(cp, M_TEMP);
3788 				goto out;
3789 			}
3790 			cp[nfsdarg.dnshostlen] = '\0';	/* Ensure nul term. */
3791 			nfsdarg.dnshost = cp;
3792 			cp = malloc(nfsdarg.dspathlen + 1, M_TEMP, M_WAITOK);
3793 			error = copyin(nfsdarg.dspath, cp, nfsdarg.dspathlen);
3794 			if (error != 0) {
3795 				free(nfsdarg.addr, M_TEMP);
3796 				free(nfsdarg.dnshost, M_TEMP);
3797 				free(cp, M_TEMP);
3798 				goto out;
3799 			}
3800 			cp[nfsdarg.dspathlen] = '\0';	/* Ensure nul term. */
3801 			nfsdarg.dspath = cp;
3802 			cp = malloc(nfsdarg.mdspathlen + 1, M_TEMP, M_WAITOK);
3803 			error = copyin(nfsdarg.mdspath, cp, nfsdarg.mdspathlen);
3804 			if (error != 0) {
3805 				free(nfsdarg.addr, M_TEMP);
3806 				free(nfsdarg.dnshost, M_TEMP);
3807 				free(nfsdarg.dspath, M_TEMP);
3808 				free(cp, M_TEMP);
3809 				goto out;
3810 			}
3811 			cp[nfsdarg.mdspathlen] = '\0';	/* Ensure nul term. */
3812 			nfsdarg.mdspath = cp;
3813 		} else {
3814 			nfsdarg.addr = NULL;
3815 			nfsdarg.addrlen = 0;
3816 			nfsdarg.dnshost = NULL;
3817 			nfsdarg.dnshostlen = 0;
3818 			nfsdarg.dspath = NULL;
3819 			nfsdarg.dspathlen = 0;
3820 			nfsdarg.mdspath = NULL;
3821 			nfsdarg.mdspathlen = 0;
3822 			nfsdarg.mirrorcnt = 1;
3823 		}
3824 		nfsd_timer(NFSD_TD_TO_VNET(td));
3825 		error = nfsrvd_nfsd(td, &nfsdarg);
3826 		callout_drain(&NFSD_VNET(nfsd_callout));
3827 		free(nfsdarg.addr, M_TEMP);
3828 		free(nfsdarg.dnshost, M_TEMP);
3829 		free(nfsdarg.dspath, M_TEMP);
3830 		free(nfsdarg.mdspath, M_TEMP);
3831 	} else if (uap->flag & NFSSVC_PNFSDS) {
3832 		error = copyin(uap->argp, &pnfsdarg, sizeof(pnfsdarg));
3833 		if (error == 0 && (pnfsdarg.op == PNFSDOP_DELDSSERVER ||
3834 		    pnfsdarg.op == PNFSDOP_FORCEDELDS)) {
3835 			cp = malloc(PATH_MAX + 1, M_TEMP, M_WAITOK);
3836 			error = copyinstr(pnfsdarg.dspath, cp, PATH_MAX + 1,
3837 			    NULL);
3838 			if (error == 0)
3839 				error = nfsrv_deldsserver(pnfsdarg.op, cp, td);
3840 			free(cp, M_TEMP);
3841 		} else if (error == 0 && pnfsdarg.op == PNFSDOP_COPYMR) {
3842 			cp = malloc(PATH_MAX + 1, M_TEMP, M_WAITOK);
3843 			buflen = sizeof(*pf) * NFSDEV_MAXMIRRORS;
3844 			buf = malloc(buflen, M_TEMP, M_WAITOK);
3845 			error = copyinstr(pnfsdarg.mdspath, cp, PATH_MAX + 1,
3846 			    NULL);
3847 			NFSD_DEBUG(4, "pnfsdcopymr cp mdspath=%d\n", error);
3848 			if (error == 0 && pnfsdarg.dspath != NULL) {
3849 				cp2 = malloc(PATH_MAX + 1, M_TEMP, M_WAITOK);
3850 				error = copyinstr(pnfsdarg.dspath, cp2,
3851 				    PATH_MAX + 1, NULL);
3852 				NFSD_DEBUG(4, "pnfsdcopymr cp dspath=%d\n",
3853 				    error);
3854 			} else
3855 				cp2 = NULL;
3856 			if (error == 0 && pnfsdarg.curdspath != NULL) {
3857 				cp3 = malloc(PATH_MAX + 1, M_TEMP, M_WAITOK);
3858 				error = copyinstr(pnfsdarg.curdspath, cp3,
3859 				    PATH_MAX + 1, NULL);
3860 				NFSD_DEBUG(4, "pnfsdcopymr cp curdspath=%d\n",
3861 				    error);
3862 			} else
3863 				cp3 = NULL;
3864 			curdvp = NULL;
3865 			fds = NULL;
3866 			if (error == 0)
3867 				error = nfsrv_mdscopymr(cp, cp2, cp3, buf,
3868 				    &buflen, fname, td, &vp, &nvp, &pf, &ds,
3869 				    &fds);
3870 			NFSD_DEBUG(4, "nfsrv_mdscopymr=%d\n", error);
3871 			if (error == 0) {
3872 				if (pf->dsf_dir >= nfsrv_dsdirsize) {
3873 					printf("copymr: dsdir out of range\n");
3874 					pf->dsf_dir = 0;
3875 				}
3876 				NFSD_DEBUG(4, "copymr: buflen=%d\n", buflen);
3877 				error = nfsrv_copymr(vp, nvp,
3878 				    ds->nfsdev_dsdir[pf->dsf_dir], ds, pf,
3879 				    (struct pnfsdsfile *)buf,
3880 				    buflen / sizeof(*pf), td->td_ucred, td);
3881 				vput(vp);
3882 				vput(nvp);
3883 				if (fds != NULL && error == 0) {
3884 					curdvp = fds->nfsdev_dsdir[pf->dsf_dir];
3885 					ret = vn_lock(curdvp, LK_EXCLUSIVE);
3886 					if (ret == 0) {
3887 						nfsrv_dsremove(curdvp, fname,
3888 						    td->td_ucred, td);
3889 						NFSVOPUNLOCK(curdvp);
3890 					}
3891 				}
3892 				NFSD_DEBUG(4, "nfsrv_copymr=%d\n", error);
3893 			}
3894 			free(cp, M_TEMP);
3895 			free(cp2, M_TEMP);
3896 			free(cp3, M_TEMP);
3897 			free(buf, M_TEMP);
3898 		}
3899 	} else {
3900 		error = nfssvc_srvcall(td, uap, td->td_ucred);
3901 	}
3902 
3903 out:
3904 	NFSD_CURVNET_RESTORE();
3905 	NFSEXITCODE(error);
3906 	return (error);
3907 }
3908 
3909 static int
3910 nfssvc_srvcall(struct thread *p, struct nfssvc_args *uap, struct ucred *cred)
3911 {
3912 	struct nfsex_args export;
3913 	struct nfsex_oldargs oexp;
3914 	struct file *fp = NULL;
3915 	int stablefd, i, len;
3916 	struct nfsd_clid adminrevoke;
3917 	struct nfsd_dumplist dumplist;
3918 	struct nfsd_dumpclients *dumpclients;
3919 	struct nfsd_dumplocklist dumplocklist;
3920 	struct nfsd_dumplocks *dumplocks;
3921 	struct nameidata nd;
3922 	vnode_t vp;
3923 	int error = EINVAL, igotlock;
3924 	struct proc *procp;
3925 	gid_t *grps;
3926 
3927 	if (uap->flag & NFSSVC_PUBLICFH) {
3928 		NFSBZERO((caddr_t)&nfs_pubfh.nfsrvfh_data,
3929 		    sizeof (fhandle_t));
3930 		error = copyin(uap->argp,
3931 		    &nfs_pubfh.nfsrvfh_data, sizeof (fhandle_t));
3932 		if (!error)
3933 			nfs_pubfhset = 1;
3934 	} else if ((uap->flag & (NFSSVC_V4ROOTEXPORT | NFSSVC_NEWSTRUCT)) ==
3935 	    (NFSSVC_V4ROOTEXPORT | NFSSVC_NEWSTRUCT)) {
3936 		error = copyin(uap->argp,(caddr_t)&export,
3937 		    sizeof (struct nfsex_args));
3938 		if (!error) {
3939 			grps = NULL;
3940 			if (export.export.ex_ngroups > NGROUPS_MAX ||
3941 			    export.export.ex_ngroups < 0)
3942 				error = EINVAL;
3943 			else if (export.export.ex_ngroups > 0) {
3944 				grps = malloc(export.export.ex_ngroups *
3945 				    sizeof(gid_t), M_TEMP, M_WAITOK);
3946 				error = copyin(export.export.ex_groups, grps,
3947 				    export.export.ex_ngroups * sizeof(gid_t));
3948 				export.export.ex_groups = grps;
3949 			} else
3950 				export.export.ex_groups = NULL;
3951 			if (!error)
3952 				error = nfsrv_v4rootexport(&export, cred, p);
3953 			free(grps, M_TEMP);
3954 		}
3955 	} else if ((uap->flag & (NFSSVC_V4ROOTEXPORT | NFSSVC_NEWSTRUCT)) ==
3956 	    NFSSVC_V4ROOTEXPORT) {
3957 		error = copyin(uap->argp,(caddr_t)&oexp,
3958 		    sizeof (struct nfsex_oldargs));
3959 		if (!error) {
3960 			memset(&export.export, 0, sizeof(export.export));
3961 			export.export.ex_flags = (uint64_t)oexp.export.ex_flags;
3962 			export.export.ex_root = oexp.export.ex_root;
3963 			export.export.ex_uid = oexp.export.ex_anon.cr_uid;
3964 			export.export.ex_ngroups =
3965 			    oexp.export.ex_anon.cr_ngroups;
3966 			export.export.ex_groups = NULL;
3967 			if (export.export.ex_ngroups > XU_NGROUPS ||
3968 			    export.export.ex_ngroups < 0)
3969 				error = EINVAL;
3970 			else if (export.export.ex_ngroups > 0) {
3971 				export.export.ex_groups = malloc(
3972 				    export.export.ex_ngroups * sizeof(gid_t),
3973 				    M_TEMP, M_WAITOK);
3974 				for (i = 0; i < export.export.ex_ngroups; i++)
3975 					export.export.ex_groups[i] =
3976 					    oexp.export.ex_anon.cr_groups[i];
3977 			}
3978 			export.export.ex_addr = oexp.export.ex_addr;
3979 			export.export.ex_addrlen = oexp.export.ex_addrlen;
3980 			export.export.ex_mask = oexp.export.ex_mask;
3981 			export.export.ex_masklen = oexp.export.ex_masklen;
3982 			export.export.ex_indexfile = oexp.export.ex_indexfile;
3983 			export.export.ex_numsecflavors =
3984 			    oexp.export.ex_numsecflavors;
3985 			if (export.export.ex_numsecflavors >= MAXSECFLAVORS ||
3986 			    export.export.ex_numsecflavors < 0)
3987 				error = EINVAL;
3988 			else {
3989 				for (i = 0; i < export.export.ex_numsecflavors;
3990 				    i++)
3991 					export.export.ex_secflavors[i] =
3992 					    oexp.export.ex_secflavors[i];
3993 			}
3994 			export.fspec = oexp.fspec;
3995 			if (error == 0)
3996 				error = nfsrv_v4rootexport(&export, cred, p);
3997 			free(export.export.ex_groups, M_TEMP);
3998 		}
3999 	} else if (uap->flag & NFSSVC_NOPUBLICFH) {
4000 		nfs_pubfhset = 0;
4001 		error = 0;
4002 	} else if (uap->flag & NFSSVC_STABLERESTART) {
4003 		error = copyin(uap->argp, (caddr_t)&stablefd,
4004 		    sizeof (int));
4005 		if (!error)
4006 			error = fp_getfvp(p, stablefd, &fp, &vp);
4007 		if (!error && (NFSFPFLAG(fp) & (FREAD | FWRITE)) != (FREAD | FWRITE))
4008 			error = EBADF;
4009 		if (!error && NFSD_VNET(nfsrv_numnfsd) != 0)
4010 			error = ENXIO;
4011 		if (!error) {
4012 			NFSD_VNET(nfsrv_stablefirst).nsf_fp = fp;
4013 			nfsrv_setupstable(p);
4014 		}
4015 	} else if (uap->flag & NFSSVC_ADMINREVOKE) {
4016 		error = copyin(uap->argp, (caddr_t)&adminrevoke,
4017 		    sizeof (struct nfsd_clid));
4018 		if (!error)
4019 			error = nfsrv_adminrevoke(&adminrevoke, p);
4020 	} else if (uap->flag & NFSSVC_DUMPCLIENTS) {
4021 		error = copyin(uap->argp, (caddr_t)&dumplist,
4022 		    sizeof (struct nfsd_dumplist));
4023 		if (!error && (dumplist.ndl_size < 1 ||
4024 			dumplist.ndl_size > NFSRV_MAXDUMPLIST))
4025 			error = EPERM;
4026 		if (!error) {
4027 		    len = sizeof (struct nfsd_dumpclients) * dumplist.ndl_size;
4028 		    dumpclients = malloc(len, M_TEMP, M_WAITOK | M_ZERO);
4029 		    nfsrv_dumpclients(dumpclients, dumplist.ndl_size);
4030 		    error = copyout(dumpclients, dumplist.ndl_list, len);
4031 		    free(dumpclients, M_TEMP);
4032 		}
4033 	} else if (uap->flag & NFSSVC_DUMPLOCKS) {
4034 		error = copyin(uap->argp, (caddr_t)&dumplocklist,
4035 		    sizeof (struct nfsd_dumplocklist));
4036 		if (!error && (dumplocklist.ndllck_size < 1 ||
4037 			dumplocklist.ndllck_size > NFSRV_MAXDUMPLIST))
4038 			error = EPERM;
4039 		if (!error)
4040 			error = nfsrv_lookupfilename(&nd,
4041 				dumplocklist.ndllck_fname, p);
4042 		if (!error) {
4043 			len = sizeof (struct nfsd_dumplocks) *
4044 				dumplocklist.ndllck_size;
4045 			dumplocks = malloc(len, M_TEMP, M_WAITOK | M_ZERO);
4046 			nfsrv_dumplocks(nd.ni_vp, dumplocks,
4047 			    dumplocklist.ndllck_size, p);
4048 			vput(nd.ni_vp);
4049 			error = copyout(dumplocks, dumplocklist.ndllck_list,
4050 			    len);
4051 			free(dumplocks, M_TEMP);
4052 		}
4053 	} else if (uap->flag & NFSSVC_BACKUPSTABLE) {
4054 		procp = p->td_proc;
4055 		PROC_LOCK(procp);
4056 		nfsd_master_pid = procp->p_pid;
4057 		bcopy(procp->p_comm, nfsd_master_comm, MAXCOMLEN + 1);
4058 		nfsd_master_start = procp->p_stats->p_start;
4059 		NFSD_VNET(nfsd_master_proc) = procp;
4060 		PROC_UNLOCK(procp);
4061 	} else if ((uap->flag & NFSSVC_SUSPENDNFSD) != 0) {
4062 		NFSLOCKV4ROOTMUTEX();
4063 		if (!NFSD_VNET(nfsrv_suspend_nfsd)) {
4064 			/* Lock out all nfsd threads */
4065 			do {
4066 				igotlock = nfsv4_lock(
4067 				    &NFSD_VNET(nfsd_suspend_lock), 1, NULL,
4068 				    NFSV4ROOTLOCKMUTEXPTR, NULL);
4069 			} while (igotlock == 0 &&
4070 			    !NFSD_VNET(nfsrv_suspend_nfsd));
4071 			NFSD_VNET(nfsrv_suspend_nfsd) = true;
4072 		}
4073 		NFSUNLOCKV4ROOTMUTEX();
4074 		error = 0;
4075 	} else if ((uap->flag & NFSSVC_RESUMENFSD) != 0) {
4076 		NFSLOCKV4ROOTMUTEX();
4077 		if (NFSD_VNET(nfsrv_suspend_nfsd)) {
4078 			nfsv4_unlock(&NFSD_VNET(nfsd_suspend_lock), 0);
4079 			NFSD_VNET(nfsrv_suspend_nfsd) = false;
4080 		}
4081 		NFSUNLOCKV4ROOTMUTEX();
4082 		error = 0;
4083 	}
4084 
4085 	NFSEXITCODE(error);
4086 	return (error);
4087 }
4088 
4089 /*
4090  * Check exports.
4091  * Returns 0 if ok, 1 otherwise.
4092  */
4093 int
4094 nfsvno_testexp(struct nfsrv_descript *nd, struct nfsexstuff *exp)
4095 {
4096 	int i;
4097 
4098 	if ((NFSVNO_EXTLS(exp) && (nd->nd_flag & ND_TLS) == 0) ||
4099 	    (NFSVNO_EXTLSCERT(exp) &&
4100 	     (nd->nd_flag & ND_TLSCERT) == 0) ||
4101 	    (NFSVNO_EXTLSCERTUSER(exp) &&
4102 	     (nd->nd_flag & ND_TLSCERTUSER) == 0)) {
4103 		if ((nd->nd_flag & ND_NFSV4) != 0)
4104 			return (NFSERR_WRONGSEC);
4105 #ifdef notnow
4106 		/* There is currently no auth_stat for this. */
4107 		else if ((nd->nd_flag & ND_TLS) == 0)
4108 			return (NFSERR_AUTHERR | AUTH_NEEDS_TLS);
4109 		else
4110 			return (NFSERR_AUTHERR | AUTH_NEEDS_TLS_MUTUAL_HOST);
4111 #endif
4112 		else
4113 			return (NFSERR_AUTHERR | AUTH_TOOWEAK);
4114 	}
4115 
4116 	/*
4117 	 * RFC2623 suggests that the NFSv3 Fsinfo RPC be allowed to use
4118 	 * AUTH_NONE or AUTH_SYS for file systems requiring RPCSEC_GSS.
4119 	 */
4120 	if ((nd->nd_flag & ND_NFSV3) != 0 && nd->nd_procnum == NFSPROC_FSINFO)
4121 		return (0);
4122 
4123 	/*
4124 	 * This seems odd, but allow the case where the security flavor
4125 	 * list is empty. This happens when NFSv4 is traversing non-exported
4126 	 * file systems. Exported file systems should always have a non-empty
4127 	 * security flavor list.
4128 	 */
4129 	if (exp->nes_numsecflavor == 0)
4130 		return (0);
4131 
4132 	for (i = 0; i < exp->nes_numsecflavor; i++) {
4133 		/*
4134 		 * The tests for privacy and integrity must be first,
4135 		 * since ND_GSS is set for everything but AUTH_SYS.
4136 		 */
4137 		if (exp->nes_secflavors[i] == RPCSEC_GSS_KRB5P &&
4138 		    (nd->nd_flag & ND_GSSPRIVACY))
4139 			return (0);
4140 		if (exp->nes_secflavors[i] == RPCSEC_GSS_KRB5I &&
4141 		    (nd->nd_flag & ND_GSSINTEGRITY))
4142 			return (0);
4143 		if (exp->nes_secflavors[i] == RPCSEC_GSS_KRB5 &&
4144 		    (nd->nd_flag & ND_GSS))
4145 			return (0);
4146 		if (exp->nes_secflavors[i] == AUTH_SYS &&
4147 		    (nd->nd_flag & ND_GSS) == 0)
4148 			return (0);
4149 	}
4150 	if ((nd->nd_flag & ND_NFSV4) != 0)
4151 		return (NFSERR_WRONGSEC);
4152 	return (NFSERR_AUTHERR | AUTH_TOOWEAK);
4153 }
4154 
4155 /*
4156  * Calculate a hash value for the fid in a file handle.
4157  */
4158 uint32_t
4159 nfsrv_hashfh(fhandle_t *fhp)
4160 {
4161 	uint32_t hashval;
4162 
4163 	hashval = hash32_buf(&fhp->fh_fid, sizeof(struct fid), 0);
4164 	return (hashval);
4165 }
4166 
4167 /*
4168  * Calculate a hash value for the sessionid.
4169  */
4170 uint32_t
4171 nfsrv_hashsessionid(uint8_t *sessionid)
4172 {
4173 	uint32_t hashval;
4174 
4175 	hashval = hash32_buf(sessionid, NFSX_V4SESSIONID, 0);
4176 	return (hashval);
4177 }
4178 
4179 /*
4180  * Signal the userland master nfsd to backup the stable restart file.
4181  */
4182 void
4183 nfsrv_backupstable(void)
4184 {
4185 	struct proc *procp;
4186 
4187 	if (NFSD_VNET(nfsd_master_proc) != NULL) {
4188 		procp = pfind(nfsd_master_pid);
4189 		/* Try to make sure it is the correct process. */
4190 		if (procp == NFSD_VNET(nfsd_master_proc) &&
4191 		    procp->p_stats->p_start.tv_sec ==
4192 		    nfsd_master_start.tv_sec &&
4193 		    procp->p_stats->p_start.tv_usec ==
4194 		    nfsd_master_start.tv_usec &&
4195 		    strcmp(procp->p_comm, nfsd_master_comm) == 0)
4196 			kern_psignal(procp, SIGUSR2);
4197 		else
4198 			NFSD_VNET(nfsd_master_proc) = NULL;
4199 
4200 		if (procp != NULL)
4201 			PROC_UNLOCK(procp);
4202 	}
4203 }
4204 
4205 /*
4206  * Create a DS data file for nfsrv_pnfscreate(). Called for each mirror.
4207  * The arguments are in a structure, so that they can be passed through
4208  * taskqueue for a kernel process to execute this function.
4209  */
4210 struct nfsrvdscreate {
4211 	int			done;
4212 	int			inprog;
4213 	struct task		tsk;
4214 	struct ucred		*tcred;
4215 	struct vnode		*dvp;
4216 	NFSPROC_T		*p;
4217 	struct pnfsdsfile	*pf;
4218 	int			err;
4219 	fhandle_t		fh;
4220 	struct vattr		va;
4221 	struct vattr		createva;
4222 };
4223 
4224 int
4225 nfsrv_dscreate(struct vnode *dvp, struct vattr *vap, struct vattr *nvap,
4226     fhandle_t *fhp, struct pnfsdsfile *pf, struct pnfsdsattr *dsa,
4227     char *fnamep, struct ucred *tcred, NFSPROC_T *p, struct vnode **nvpp)
4228 {
4229 	struct vnode *nvp;
4230 	struct nameidata named;
4231 	struct vattr va;
4232 	char *bufp;
4233 	u_long *hashp;
4234 	struct nfsnode *np;
4235 	struct nfsmount *nmp;
4236 	int error;
4237 
4238 	NFSNAMEICNDSET(&named.ni_cnd, tcred, CREATE,
4239 	    LOCKPARENT | LOCKLEAF | NOCACHE);
4240 	nfsvno_setpathbuf(&named, &bufp, &hashp);
4241 	named.ni_cnd.cn_lkflags = LK_EXCLUSIVE;
4242 	named.ni_cnd.cn_nameptr = bufp;
4243 	if (fnamep != NULL) {
4244 		strlcpy(bufp, fnamep, PNFS_FILENAME_LEN + 1);
4245 		named.ni_cnd.cn_namelen = strlen(bufp);
4246 	} else
4247 		named.ni_cnd.cn_namelen = nfsrv_putfhname(fhp, bufp);
4248 	NFSD_DEBUG(4, "nfsrv_dscreate: dvp=%p fname=%s\n", dvp, bufp);
4249 
4250 	/* Create the date file in the DS mount. */
4251 	error = NFSVOPLOCK(dvp, LK_EXCLUSIVE);
4252 	if (error == 0) {
4253 		error = VOP_CREATE(dvp, &nvp, &named.ni_cnd, vap);
4254 		vref(dvp);
4255 		VOP_VPUT_PAIR(dvp, error == 0 ? &nvp : NULL, false);
4256 		if (error == 0) {
4257 			/* Set the ownership of the file. */
4258 			error = VOP_SETATTR(nvp, nvap, tcred);
4259 			NFSD_DEBUG(4, "nfsrv_dscreate:"
4260 			    " setattr-uid=%d\n", error);
4261 			if (error != 0)
4262 				vput(nvp);
4263 		}
4264 		if (error != 0)
4265 			printf("pNFS: pnfscreate failed=%d\n", error);
4266 	} else
4267 		printf("pNFS: pnfscreate vnlock=%d\n", error);
4268 	if (error == 0) {
4269 		np = VTONFS(nvp);
4270 		nmp = VFSTONFS(nvp->v_mount);
4271 		if (strcmp(nvp->v_mount->mnt_vfc->vfc_name, "nfs")
4272 		    != 0 || nmp->nm_nam->sa_len > sizeof(
4273 		    struct sockaddr_in6) ||
4274 		    np->n_fhp->nfh_len != NFSX_MYFH) {
4275 			printf("Bad DS file: fstype=%s salen=%d"
4276 			    " fhlen=%d\n",
4277 			    nvp->v_mount->mnt_vfc->vfc_name,
4278 			    nmp->nm_nam->sa_len, np->n_fhp->nfh_len);
4279 			error = ENOENT;
4280 		}
4281 
4282 		/* Set extattrs for the DS on the MDS file. */
4283 		if (error == 0) {
4284 			if (dsa != NULL) {
4285 				error = VOP_GETATTR(nvp, &va, tcred);
4286 				if (error == 0) {
4287 					dsa->dsa_filerev = va.va_filerev;
4288 					dsa->dsa_size = va.va_size;
4289 					dsa->dsa_atime = va.va_atime;
4290 					dsa->dsa_mtime = va.va_mtime;
4291 					dsa->dsa_bytes = va.va_bytes;
4292 				}
4293 			}
4294 			if (error == 0) {
4295 				NFSBCOPY(np->n_fhp->nfh_fh, &pf->dsf_fh,
4296 				    NFSX_MYFH);
4297 				NFSBCOPY(nmp->nm_nam, &pf->dsf_sin,
4298 				    nmp->nm_nam->sa_len);
4299 				NFSBCOPY(named.ni_cnd.cn_nameptr,
4300 				    pf->dsf_filename,
4301 				    sizeof(pf->dsf_filename));
4302 			}
4303 		} else
4304 			printf("pNFS: pnfscreate can't get DS"
4305 			    " attr=%d\n", error);
4306 		if (nvpp != NULL && error == 0)
4307 			*nvpp = nvp;
4308 		else
4309 			vput(nvp);
4310 	}
4311 	nfsvno_relpathbuf(&named);
4312 	return (error);
4313 }
4314 
4315 /*
4316  * Start up the thread that will execute nfsrv_dscreate().
4317  */
4318 static void
4319 start_dscreate(void *arg, int pending)
4320 {
4321 	struct nfsrvdscreate *dsc;
4322 
4323 	dsc = (struct nfsrvdscreate *)arg;
4324 	dsc->err = nfsrv_dscreate(dsc->dvp, &dsc->createva, &dsc->va, &dsc->fh,
4325 	    dsc->pf, NULL, NULL, dsc->tcred, dsc->p, NULL);
4326 	dsc->done = 1;
4327 	NFSD_DEBUG(4, "start_dscreate: err=%d\n", dsc->err);
4328 }
4329 
4330 /*
4331  * Create a pNFS data file on the Data Server(s).
4332  */
4333 static void
4334 nfsrv_pnfscreate(struct vnode *vp, struct vattr *vap, struct ucred *cred,
4335     NFSPROC_T *p)
4336 {
4337 	struct nfsrvdscreate *dsc, *tdsc = NULL;
4338 	struct nfsdevice *ds, *tds, *fds;
4339 	struct mount *mp;
4340 	struct pnfsdsfile *pf, *tpf;
4341 	struct pnfsdsattr dsattr;
4342 	struct vattr va;
4343 	struct vnode *dvp[NFSDEV_MAXMIRRORS];
4344 	struct nfsmount *nmp;
4345 	fhandle_t fh;
4346 	uid_t vauid;
4347 	gid_t vagid;
4348 	u_short vamode;
4349 	struct ucred *tcred;
4350 	int dsdir[NFSDEV_MAXMIRRORS], error, i, mirrorcnt, ret;
4351 	int failpos, timo;
4352 
4353 	/* Get a DS server directory in a round-robin order. */
4354 	mirrorcnt = 1;
4355 	mp = vp->v_mount;
4356 	ds = fds = NULL;
4357 	NFSDDSLOCK();
4358 	/*
4359 	 * Search for the first entry that handles this MDS fs, but use the
4360 	 * first entry for all MDS fs's otherwise.
4361 	 */
4362 	TAILQ_FOREACH(tds, &nfsrv_devidhead, nfsdev_list) {
4363 		if (tds->nfsdev_nmp != NULL) {
4364 			if (tds->nfsdev_mdsisset == 0 && ds == NULL)
4365 				ds = tds;
4366 			else if (tds->nfsdev_mdsisset != 0 && fsidcmp(
4367 			    &mp->mnt_stat.f_fsid, &tds->nfsdev_mdsfsid) == 0) {
4368 				ds = fds = tds;
4369 				break;
4370 			}
4371 		}
4372 	}
4373 	if (ds == NULL) {
4374 		NFSDDSUNLOCK();
4375 		NFSD_DEBUG(4, "nfsrv_pnfscreate: no srv\n");
4376 		return;
4377 	}
4378 	i = dsdir[0] = ds->nfsdev_nextdir;
4379 	ds->nfsdev_nextdir = (ds->nfsdev_nextdir + 1) % nfsrv_dsdirsize;
4380 	dvp[0] = ds->nfsdev_dsdir[i];
4381 	tds = TAILQ_NEXT(ds, nfsdev_list);
4382 	if (nfsrv_maxpnfsmirror > 1 && tds != NULL) {
4383 		TAILQ_FOREACH_FROM(tds, &nfsrv_devidhead, nfsdev_list) {
4384 			if (tds->nfsdev_nmp != NULL &&
4385 			    ((tds->nfsdev_mdsisset == 0 && fds == NULL) ||
4386 			     (tds->nfsdev_mdsisset != 0 && fds != NULL &&
4387 			      fsidcmp(&mp->mnt_stat.f_fsid,
4388 			      &tds->nfsdev_mdsfsid) == 0))) {
4389 				dsdir[mirrorcnt] = i;
4390 				dvp[mirrorcnt] = tds->nfsdev_dsdir[i];
4391 				mirrorcnt++;
4392 				if (mirrorcnt >= nfsrv_maxpnfsmirror)
4393 					break;
4394 			}
4395 		}
4396 	}
4397 	/* Put at end of list to implement round-robin usage. */
4398 	TAILQ_REMOVE(&nfsrv_devidhead, ds, nfsdev_list);
4399 	TAILQ_INSERT_TAIL(&nfsrv_devidhead, ds, nfsdev_list);
4400 	NFSDDSUNLOCK();
4401 	dsc = NULL;
4402 	if (mirrorcnt > 1)
4403 		tdsc = dsc = malloc(sizeof(*dsc) * (mirrorcnt - 1), M_TEMP,
4404 		    M_WAITOK | M_ZERO);
4405 	tpf = pf = malloc(sizeof(*pf) * nfsrv_maxpnfsmirror, M_TEMP, M_WAITOK |
4406 	    M_ZERO);
4407 
4408 	error = nfsvno_getfh(vp, &fh, p);
4409 	if (error == 0)
4410 		error = VOP_GETATTR(vp, &va, cred);
4411 	if (error == 0) {
4412 		/* Set the attributes for "vp" to Setattr the DS vp. */
4413 		vauid = va.va_uid;
4414 		vagid = va.va_gid;
4415 		vamode = va.va_mode;
4416 		VATTR_NULL(&va);
4417 		va.va_uid = vauid;
4418 		va.va_gid = vagid;
4419 		va.va_mode = vamode;
4420 		va.va_size = 0;
4421 	} else
4422 		printf("pNFS: pnfscreate getfh+attr=%d\n", error);
4423 
4424 	NFSD_DEBUG(4, "nfsrv_pnfscreate: cruid=%d crgid=%d\n", cred->cr_uid,
4425 	    cred->cr_gid);
4426 	/* Make data file name based on FH. */
4427 	tcred = newnfs_getcred();
4428 
4429 	/*
4430 	 * Create the file on each DS mirror, using kernel process(es) for the
4431 	 * additional mirrors.
4432 	 */
4433 	failpos = -1;
4434 	for (i = 0; i < mirrorcnt - 1 && error == 0; i++, tpf++, tdsc++) {
4435 		tpf->dsf_dir = dsdir[i];
4436 		tdsc->tcred = tcred;
4437 		tdsc->p = p;
4438 		tdsc->pf = tpf;
4439 		tdsc->createva = *vap;
4440 		NFSBCOPY(&fh, &tdsc->fh, sizeof(fh));
4441 		tdsc->va = va;
4442 		tdsc->dvp = dvp[i];
4443 		tdsc->done = 0;
4444 		tdsc->inprog = 0;
4445 		tdsc->err = 0;
4446 		ret = EIO;
4447 		if (nfs_pnfsiothreads != 0) {
4448 			ret = nfs_pnfsio(start_dscreate, tdsc);
4449 			NFSD_DEBUG(4, "nfsrv_pnfscreate: nfs_pnfsio=%d\n", ret);
4450 		}
4451 		if (ret != 0) {
4452 			ret = nfsrv_dscreate(dvp[i], vap, &va, &fh, tpf, NULL,
4453 			    NULL, tcred, p, NULL);
4454 			if (ret != 0) {
4455 				KASSERT(error == 0, ("nfsrv_dscreate err=%d",
4456 				    error));
4457 				if (failpos == -1 && nfsds_failerr(ret))
4458 					failpos = i;
4459 				else
4460 					error = ret;
4461 			}
4462 		}
4463 	}
4464 	if (error == 0) {
4465 		tpf->dsf_dir = dsdir[mirrorcnt - 1];
4466 		error = nfsrv_dscreate(dvp[mirrorcnt - 1], vap, &va, &fh, tpf,
4467 		    &dsattr, NULL, tcred, p, NULL);
4468 		if (failpos == -1 && mirrorcnt > 1 && nfsds_failerr(error)) {
4469 			failpos = mirrorcnt - 1;
4470 			error = 0;
4471 		}
4472 	}
4473 	timo = hz / 50;		/* Wait for 20msec. */
4474 	if (timo < 1)
4475 		timo = 1;
4476 	/* Wait for kernel task(s) to complete. */
4477 	for (tdsc = dsc, i = 0; i < mirrorcnt - 1; i++, tdsc++) {
4478 		while (tdsc->inprog != 0 && tdsc->done == 0)
4479 			tsleep(&tdsc->tsk, PVFS, "srvdcr", timo);
4480 		if (tdsc->err != 0) {
4481 			if (failpos == -1 && nfsds_failerr(tdsc->err))
4482 				failpos = i;
4483 			else if (error == 0)
4484 				error = tdsc->err;
4485 		}
4486 	}
4487 
4488 	/*
4489 	 * If failpos has been set, that mirror has failed, so it needs
4490 	 * to be disabled.
4491 	 */
4492 	if (failpos >= 0) {
4493 		nmp = VFSTONFS(dvp[failpos]->v_mount);
4494 		NFSLOCKMNT(nmp);
4495 		if ((nmp->nm_privflag & (NFSMNTP_FORCEDISM |
4496 		     NFSMNTP_CANCELRPCS)) == 0) {
4497 			nmp->nm_privflag |= NFSMNTP_CANCELRPCS;
4498 			NFSUNLOCKMNT(nmp);
4499 			ds = nfsrv_deldsnmp(PNFSDOP_DELDSSERVER, nmp, p);
4500 			NFSD_DEBUG(4, "dscreatfail fail=%d ds=%p\n", failpos,
4501 			    ds);
4502 			if (ds != NULL)
4503 				nfsrv_killrpcs(nmp);
4504 			NFSLOCKMNT(nmp);
4505 			nmp->nm_privflag &= ~NFSMNTP_CANCELRPCS;
4506 			wakeup(nmp);
4507 		}
4508 		NFSUNLOCKMNT(nmp);
4509 	}
4510 
4511 	NFSFREECRED(tcred);
4512 	if (error == 0) {
4513 		ASSERT_VOP_ELOCKED(vp, "nfsrv_pnfscreate vp");
4514 
4515 		NFSD_DEBUG(4, "nfsrv_pnfscreate: mirrorcnt=%d maxmirror=%d\n",
4516 		    mirrorcnt, nfsrv_maxpnfsmirror);
4517 		/*
4518 		 * For all mirrors that couldn't be created, fill in the
4519 		 * *pf structure, but with an IP address == 0.0.0.0.
4520 		 */
4521 		tpf = pf + mirrorcnt;
4522 		for (i = mirrorcnt; i < nfsrv_maxpnfsmirror; i++, tpf++) {
4523 			*tpf = *pf;
4524 			tpf->dsf_sin.sin_family = AF_INET;
4525 			tpf->dsf_sin.sin_len = sizeof(struct sockaddr_in);
4526 			tpf->dsf_sin.sin_addr.s_addr = 0;
4527 			tpf->dsf_sin.sin_port = 0;
4528 		}
4529 
4530 		error = vn_extattr_set(vp, IO_NODELOCKED,
4531 		    EXTATTR_NAMESPACE_SYSTEM, "pnfsd.dsfile",
4532 		    sizeof(*pf) * nfsrv_maxpnfsmirror, (char *)pf, p);
4533 		if (error == 0)
4534 			error = vn_extattr_set(vp, IO_NODELOCKED,
4535 			    EXTATTR_NAMESPACE_SYSTEM, "pnfsd.dsattr",
4536 			    sizeof(dsattr), (char *)&dsattr, p);
4537 		if (error != 0)
4538 			printf("pNFS: pnfscreate setextattr=%d\n",
4539 			    error);
4540 	} else
4541 		printf("pNFS: pnfscreate=%d\n", error);
4542 	free(pf, M_TEMP);
4543 	free(dsc, M_TEMP);
4544 }
4545 
4546 /*
4547  * Get the information needed to remove the pNFS Data Server file from the
4548  * Metadata file.  Upon success, ddvp is set non-NULL to the locked
4549  * DS directory vnode.  The caller must unlock *ddvp when done with it.
4550  */
4551 static void
4552 nfsrv_pnfsremovesetup(struct vnode *vp, NFSPROC_T *p, struct vnode **dvpp,
4553     int *mirrorcntp, char *fname, fhandle_t *fhp)
4554 {
4555 	struct vattr va;
4556 	struct ucred *tcred;
4557 	char *buf;
4558 	int buflen, error;
4559 
4560 	dvpp[0] = NULL;
4561 	/* If not an exported regular file or not a pNFS server, just return. */
4562 	if (vp->v_type != VREG || (vp->v_mount->mnt_flag & MNT_EXPORTED) == 0 ||
4563 	    nfsrv_devidcnt == 0)
4564 		return;
4565 
4566 	/* Check to see if this is the last hard link. */
4567 	tcred = newnfs_getcred();
4568 	error = VOP_GETATTR(vp, &va, tcred);
4569 	NFSFREECRED(tcred);
4570 	if (error != 0) {
4571 		printf("pNFS: nfsrv_pnfsremovesetup getattr=%d\n", error);
4572 		return;
4573 	}
4574 	if (va.va_nlink > 1)
4575 		return;
4576 
4577 	error = nfsvno_getfh(vp, fhp, p);
4578 	if (error != 0) {
4579 		printf("pNFS: nfsrv_pnfsremovesetup getfh=%d\n", error);
4580 		return;
4581 	}
4582 
4583 	buflen = 1024;
4584 	buf = malloc(buflen, M_TEMP, M_WAITOK);
4585 	/* Get the directory vnode for the DS mount and the file handle. */
4586 	error = nfsrv_dsgetsockmnt(vp, 0, buf, &buflen, mirrorcntp, p, dvpp,
4587 	    NULL, NULL, fname, NULL, NULL, NULL, NULL, NULL);
4588 	free(buf, M_TEMP);
4589 	if (error != 0)
4590 		printf("pNFS: nfsrv_pnfsremovesetup getsockmnt=%d\n", error);
4591 }
4592 
4593 /*
4594  * Remove a DS data file for nfsrv_pnfsremove(). Called for each mirror.
4595  * The arguments are in a structure, so that they can be passed through
4596  * taskqueue for a kernel process to execute this function.
4597  */
4598 struct nfsrvdsremove {
4599 	int			done;
4600 	int			inprog;
4601 	struct task		tsk;
4602 	struct ucred		*tcred;
4603 	struct vnode		*dvp;
4604 	NFSPROC_T		*p;
4605 	int			err;
4606 	char			fname[PNFS_FILENAME_LEN + 1];
4607 };
4608 
4609 static int
4610 nfsrv_dsremove(struct vnode *dvp, char *fname, struct ucred *tcred,
4611     NFSPROC_T *p)
4612 {
4613 	struct nameidata named;
4614 	struct vnode *nvp;
4615 	char *bufp;
4616 	u_long *hashp;
4617 	int error;
4618 
4619 	error = NFSVOPLOCK(dvp, LK_EXCLUSIVE);
4620 	if (error != 0)
4621 		return (error);
4622 	named.ni_cnd.cn_nameiop = DELETE;
4623 	named.ni_cnd.cn_lkflags = LK_EXCLUSIVE | LK_RETRY;
4624 	named.ni_cnd.cn_cred = tcred;
4625 	named.ni_cnd.cn_flags = ISLASTCN | LOCKPARENT | LOCKLEAF;
4626 	nfsvno_setpathbuf(&named, &bufp, &hashp);
4627 	named.ni_cnd.cn_nameptr = bufp;
4628 	named.ni_cnd.cn_namelen = strlen(fname);
4629 	strlcpy(bufp, fname, NAME_MAX);
4630 	NFSD_DEBUG(4, "nfsrv_pnfsremove: filename=%s\n", bufp);
4631 	error = VOP_LOOKUP(dvp, &nvp, &named.ni_cnd);
4632 	NFSD_DEBUG(4, "nfsrv_pnfsremove: aft LOOKUP=%d\n", error);
4633 	if (error == 0) {
4634 		error = VOP_REMOVE(dvp, nvp, &named.ni_cnd);
4635 		vput(nvp);
4636 	}
4637 	NFSVOPUNLOCK(dvp);
4638 	nfsvno_relpathbuf(&named);
4639 	if (error != 0)
4640 		printf("pNFS: nfsrv_pnfsremove failed=%d\n", error);
4641 	return (error);
4642 }
4643 
4644 /*
4645  * Start up the thread that will execute nfsrv_dsremove().
4646  */
4647 static void
4648 start_dsremove(void *arg, int pending)
4649 {
4650 	struct nfsrvdsremove *dsrm;
4651 
4652 	dsrm = (struct nfsrvdsremove *)arg;
4653 	dsrm->err = nfsrv_dsremove(dsrm->dvp, dsrm->fname, dsrm->tcred,
4654 	    dsrm->p);
4655 	dsrm->done = 1;
4656 	NFSD_DEBUG(4, "start_dsremove: err=%d\n", dsrm->err);
4657 }
4658 
4659 /*
4660  * Remove a pNFS data file from a Data Server.
4661  * nfsrv_pnfsremovesetup() must have been called before the MDS file was
4662  * removed to set up the dvp and fill in the FH.
4663  */
4664 static void
4665 nfsrv_pnfsremove(struct vnode **dvp, int mirrorcnt, char *fname, fhandle_t *fhp,
4666     NFSPROC_T *p)
4667 {
4668 	struct ucred *tcred;
4669 	struct nfsrvdsremove *dsrm, *tdsrm;
4670 	struct nfsdevice *ds;
4671 	struct nfsmount *nmp;
4672 	int failpos, i, ret, timo;
4673 
4674 	tcred = newnfs_getcred();
4675 	dsrm = NULL;
4676 	if (mirrorcnt > 1)
4677 		dsrm = malloc(sizeof(*dsrm) * mirrorcnt - 1, M_TEMP, M_WAITOK);
4678 	/*
4679 	 * Remove the file on each DS mirror, using kernel process(es) for the
4680 	 * additional mirrors.
4681 	 */
4682 	failpos = -1;
4683 	for (tdsrm = dsrm, i = 0; i < mirrorcnt - 1; i++, tdsrm++) {
4684 		tdsrm->tcred = tcred;
4685 		tdsrm->p = p;
4686 		tdsrm->dvp = dvp[i];
4687 		strlcpy(tdsrm->fname, fname, PNFS_FILENAME_LEN + 1);
4688 		tdsrm->inprog = 0;
4689 		tdsrm->done = 0;
4690 		tdsrm->err = 0;
4691 		ret = EIO;
4692 		if (nfs_pnfsiothreads != 0) {
4693 			ret = nfs_pnfsio(start_dsremove, tdsrm);
4694 			NFSD_DEBUG(4, "nfsrv_pnfsremove: nfs_pnfsio=%d\n", ret);
4695 		}
4696 		if (ret != 0) {
4697 			ret = nfsrv_dsremove(dvp[i], fname, tcred, p);
4698 			if (failpos == -1 && nfsds_failerr(ret))
4699 				failpos = i;
4700 		}
4701 	}
4702 	ret = nfsrv_dsremove(dvp[mirrorcnt - 1], fname, tcred, p);
4703 	if (failpos == -1 && mirrorcnt > 1 && nfsds_failerr(ret))
4704 		failpos = mirrorcnt - 1;
4705 	timo = hz / 50;		/* Wait for 20msec. */
4706 	if (timo < 1)
4707 		timo = 1;
4708 	/* Wait for kernel task(s) to complete. */
4709 	for (tdsrm = dsrm, i = 0; i < mirrorcnt - 1; i++, tdsrm++) {
4710 		while (tdsrm->inprog != 0 && tdsrm->done == 0)
4711 			tsleep(&tdsrm->tsk, PVFS, "srvdsrm", timo);
4712 		if (failpos == -1 && nfsds_failerr(tdsrm->err))
4713 			failpos = i;
4714 	}
4715 
4716 	/*
4717 	 * If failpos has been set, that mirror has failed, so it needs
4718 	 * to be disabled.
4719 	 */
4720 	if (failpos >= 0) {
4721 		nmp = VFSTONFS(dvp[failpos]->v_mount);
4722 		NFSLOCKMNT(nmp);
4723 		if ((nmp->nm_privflag & (NFSMNTP_FORCEDISM |
4724 		     NFSMNTP_CANCELRPCS)) == 0) {
4725 			nmp->nm_privflag |= NFSMNTP_CANCELRPCS;
4726 			NFSUNLOCKMNT(nmp);
4727 			ds = nfsrv_deldsnmp(PNFSDOP_DELDSSERVER, nmp, p);
4728 			NFSD_DEBUG(4, "dsremovefail fail=%d ds=%p\n", failpos,
4729 			    ds);
4730 			if (ds != NULL)
4731 				nfsrv_killrpcs(nmp);
4732 			NFSLOCKMNT(nmp);
4733 			nmp->nm_privflag &= ~NFSMNTP_CANCELRPCS;
4734 			wakeup(nmp);
4735 		}
4736 		NFSUNLOCKMNT(nmp);
4737 	}
4738 
4739 	/* Get rid all layouts for the file. */
4740 	nfsrv_freefilelayouts(fhp);
4741 
4742 	NFSFREECRED(tcred);
4743 	free(dsrm, M_TEMP);
4744 }
4745 
4746 /*
4747  * Generate a file name based on the file handle and put it in *bufp.
4748  * Return the number of bytes generated.
4749  */
4750 static int
4751 nfsrv_putfhname(fhandle_t *fhp, char *bufp)
4752 {
4753 	int i;
4754 	uint8_t *cp;
4755 	const uint8_t *hexdigits = "0123456789abcdef";
4756 
4757 	cp = (uint8_t *)fhp;
4758 	for (i = 0; i < sizeof(*fhp); i++) {
4759 		bufp[2 * i] = hexdigits[(*cp >> 4) & 0xf];
4760 		bufp[2 * i + 1] = hexdigits[*cp++ & 0xf];
4761 	}
4762 	bufp[2 * i] = '\0';
4763 	return (2 * i);
4764 }
4765 
4766 /*
4767  * Update the Metadata file's attributes from the DS file when a Read/Write
4768  * layout is returned.
4769  * Basically just call nfsrv_proxyds() with procedure == NFSPROC_LAYOUTRETURN
4770  * so that it does a nfsrv_getattrdsrpc() and nfsrv_setextattr() on the DS file.
4771  */
4772 int
4773 nfsrv_updatemdsattr(struct vnode *vp, struct nfsvattr *nap, NFSPROC_T *p)
4774 {
4775 	struct ucred *tcred;
4776 	int error;
4777 
4778 	/* Do this as root so that it won't fail with EACCES. */
4779 	tcred = newnfs_getcred();
4780 	error = nfsrv_proxyds(vp, 0, 0, tcred, p, NFSPROC_LAYOUTRETURN,
4781 	    NULL, NULL, NULL, nap, NULL, NULL, 0, NULL);
4782 	NFSFREECRED(tcred);
4783 	return (error);
4784 }
4785 
4786 /*
4787  * Set the NFSv4 ACL on the DS file to the same ACL as the MDS file.
4788  */
4789 static int
4790 nfsrv_dssetacl(struct vnode *vp, struct acl *aclp, struct ucred *cred,
4791     NFSPROC_T *p)
4792 {
4793 	int error;
4794 
4795 	error = nfsrv_proxyds(vp, 0, 0, cred, p, NFSPROC_SETACL,
4796 	    NULL, NULL, NULL, NULL, aclp, NULL, 0, NULL);
4797 	return (error);
4798 }
4799 
4800 static int
4801 nfsrv_proxyds(struct vnode *vp, off_t off, int cnt, struct ucred *cred,
4802     struct thread *p, int ioproc, struct mbuf **mpp, char *cp,
4803     struct mbuf **mpp2, struct nfsvattr *nap, struct acl *aclp,
4804     off_t *offp, int content, bool *eofp)
4805 {
4806 	struct nfsmount *nmp[NFSDEV_MAXMIRRORS], *failnmp;
4807 	fhandle_t fh[NFSDEV_MAXMIRRORS];
4808 	struct vnode *dvp[NFSDEV_MAXMIRRORS];
4809 	struct nfsdevice *ds;
4810 	struct pnfsdsattr dsattr;
4811 	struct opnfsdsattr odsattr;
4812 	char *buf;
4813 	int buflen, error, failpos, i, mirrorcnt, origmircnt, trycnt;
4814 
4815 	NFSD_DEBUG(4, "in nfsrv_proxyds\n");
4816 	/*
4817 	 * If not a regular file, not exported or not a pNFS server,
4818 	 * just return ENOENT.
4819 	 */
4820 	if (vp->v_type != VREG || (vp->v_mount->mnt_flag & MNT_EXPORTED) == 0 ||
4821 	    nfsrv_devidcnt == 0)
4822 		return (ENOENT);
4823 
4824 	buflen = 1024;
4825 	buf = malloc(buflen, M_TEMP, M_WAITOK);
4826 	error = 0;
4827 
4828 	/*
4829 	 * For Getattr, get the Change attribute (va_filerev) and size (va_size)
4830 	 * from the MetaData file's extended attribute.
4831 	 */
4832 	if (ioproc == NFSPROC_GETATTR) {
4833 		error = vn_extattr_get(vp, IO_NODELOCKED,
4834 		    EXTATTR_NAMESPACE_SYSTEM, "pnfsd.dsattr", &buflen, buf,
4835 		    p);
4836 		if (error == 0) {
4837 			if (buflen == sizeof(odsattr)) {
4838 				NFSBCOPY(buf, &odsattr, buflen);
4839 				nap->na_filerev = odsattr.dsa_filerev;
4840 				nap->na_size = odsattr.dsa_size;
4841 				nap->na_atime = odsattr.dsa_atime;
4842 				nap->na_mtime = odsattr.dsa_mtime;
4843 				/*
4844 				 * Fake na_bytes by rounding up na_size.
4845 				 * Since we don't know the block size, just
4846 				 * use BLKDEV_IOSIZE.
4847 				 */
4848 				nap->na_bytes = (odsattr.dsa_size +
4849 				    BLKDEV_IOSIZE - 1) & ~(BLKDEV_IOSIZE - 1);
4850 			} else if (buflen == sizeof(dsattr)) {
4851 				NFSBCOPY(buf, &dsattr, buflen);
4852 				nap->na_filerev = dsattr.dsa_filerev;
4853 				nap->na_size = dsattr.dsa_size;
4854 				nap->na_atime = dsattr.dsa_atime;
4855 				nap->na_mtime = dsattr.dsa_mtime;
4856 				nap->na_bytes = dsattr.dsa_bytes;
4857 			} else
4858 				error = ENXIO;
4859 		}
4860 		if (error == 0) {
4861 			/*
4862 			 * If nfsrv_pnfsgetdsattr is 0 or nfsrv_checkdsattr()
4863 			 * returns 0, just return now.  nfsrv_checkdsattr()
4864 			 * returns 0 if there is no Read/Write layout
4865 			 * plus either an Open/Write_access or Write
4866 			 * delegation issued to a client for the file.
4867 			 */
4868 			if (nfsrv_pnfsgetdsattr == 0 ||
4869 			    nfsrv_checkdsattr(vp, p) == 0) {
4870 				free(buf, M_TEMP);
4871 				return (error);
4872 			}
4873 		}
4874 
4875 		/*
4876 		 * Clear ENOATTR so the code below will attempt to do a
4877 		 * nfsrv_getattrdsrpc() to get the attributes and (re)create
4878 		 * the extended attribute.
4879 		 */
4880 		if (error == ENOATTR)
4881 			error = 0;
4882 	}
4883 
4884 	origmircnt = -1;
4885 	trycnt = 0;
4886 tryagain:
4887 	if (error == 0) {
4888 		buflen = 1024;
4889 		if (ioproc == NFSPROC_READDS && NFSVOPISLOCKED(vp) ==
4890 		    LK_EXCLUSIVE)
4891 			printf("nfsrv_proxyds: Readds vp exclusively locked\n");
4892 		error = nfsrv_dsgetsockmnt(vp, LK_SHARED, buf, &buflen,
4893 		    &mirrorcnt, p, dvp, fh, NULL, NULL, NULL, NULL, NULL,
4894 		    NULL, NULL);
4895 		if (error == 0) {
4896 			for (i = 0; i < mirrorcnt; i++)
4897 				nmp[i] = VFSTONFS(dvp[i]->v_mount);
4898 		} else
4899 			printf("pNFS: proxy getextattr sockaddr=%d\n", error);
4900 	} else
4901 		printf("pNFS: nfsrv_dsgetsockmnt=%d\n", error);
4902 	if (error == 0) {
4903 		failpos = -1;
4904 		if (origmircnt == -1)
4905 			origmircnt = mirrorcnt;
4906 		/*
4907 		 * If failpos is set to a mirror#, then that mirror has
4908 		 * failed and will be disabled. For Read, Getattr and Seek, the
4909 		 * function only tries one mirror, so if that mirror has
4910 		 * failed, it will need to be retried. As such, increment
4911 		 * tryitagain for these cases.
4912 		 * For Write, Setattr and Setacl, the function tries all
4913 		 * mirrors and will not return an error for the case where
4914 		 * one mirror has failed. For these cases, the functioning
4915 		 * mirror(s) will have been modified, so a retry isn't
4916 		 * necessary. These functions will set failpos for the
4917 		 * failed mirror#.
4918 		 */
4919 		if (ioproc == NFSPROC_READDS) {
4920 			error = nfsrv_readdsrpc(fh, off, cnt, cred, p, nmp[0],
4921 			    mpp, mpp2);
4922 			if (nfsds_failerr(error) && mirrorcnt > 1) {
4923 				/*
4924 				 * Setting failpos will cause the mirror
4925 				 * to be disabled and then a retry of this
4926 				 * read is required.
4927 				 */
4928 				failpos = 0;
4929 				error = 0;
4930 				trycnt++;
4931 			}
4932 		} else if (ioproc == NFSPROC_WRITEDS)
4933 			error = nfsrv_writedsrpc(fh, off, cnt, cred, p, vp,
4934 			    &nmp[0], mirrorcnt, mpp, cp, &failpos);
4935 		else if (ioproc == NFSPROC_SETATTR)
4936 			error = nfsrv_setattrdsrpc(fh, cred, p, vp, &nmp[0],
4937 			    mirrorcnt, nap, &failpos);
4938 		else if (ioproc == NFSPROC_SETACL)
4939 			error = nfsrv_setacldsrpc(fh, cred, p, vp, &nmp[0],
4940 			    mirrorcnt, aclp, &failpos);
4941 		else if (ioproc == NFSPROC_SEEKDS) {
4942 			error = nfsrv_seekdsrpc(fh, offp, content, eofp, cred,
4943 			    p, nmp[0]);
4944 			if (nfsds_failerr(error) && mirrorcnt > 1) {
4945 				/*
4946 				 * Setting failpos will cause the mirror
4947 				 * to be disabled and then a retry of this
4948 				 * read is required.
4949 				 */
4950 				failpos = 0;
4951 				error = 0;
4952 				trycnt++;
4953 			}
4954 		} else if (ioproc == NFSPROC_ALLOCATE)
4955 			error = nfsrv_allocatedsrpc(fh, off, *offp, cred, p, vp,
4956 			    &nmp[0], mirrorcnt, &failpos);
4957 		else if (ioproc == NFSPROC_DEALLOCATE)
4958 			error = nfsrv_deallocatedsrpc(fh, off, *offp, cred, p,
4959 			    vp, &nmp[0], mirrorcnt, &failpos);
4960 		else {
4961 			error = nfsrv_getattrdsrpc(&fh[mirrorcnt - 1], cred, p,
4962 			    vp, nmp[mirrorcnt - 1], nap);
4963 			if (nfsds_failerr(error) && mirrorcnt > 1) {
4964 				/*
4965 				 * Setting failpos will cause the mirror
4966 				 * to be disabled and then a retry of this
4967 				 * getattr is required.
4968 				 */
4969 				failpos = mirrorcnt - 1;
4970 				error = 0;
4971 				trycnt++;
4972 			}
4973 		}
4974 		ds = NULL;
4975 		if (failpos >= 0) {
4976 			failnmp = nmp[failpos];
4977 			NFSLOCKMNT(failnmp);
4978 			if ((failnmp->nm_privflag & (NFSMNTP_FORCEDISM |
4979 			     NFSMNTP_CANCELRPCS)) == 0) {
4980 				failnmp->nm_privflag |= NFSMNTP_CANCELRPCS;
4981 				NFSUNLOCKMNT(failnmp);
4982 				ds = nfsrv_deldsnmp(PNFSDOP_DELDSSERVER,
4983 				    failnmp, p);
4984 				NFSD_DEBUG(4, "dsldsnmp fail=%d ds=%p\n",
4985 				    failpos, ds);
4986 				if (ds != NULL)
4987 					nfsrv_killrpcs(failnmp);
4988 				NFSLOCKMNT(failnmp);
4989 				failnmp->nm_privflag &= ~NFSMNTP_CANCELRPCS;
4990 				wakeup(failnmp);
4991 			}
4992 			NFSUNLOCKMNT(failnmp);
4993 		}
4994 		for (i = 0; i < mirrorcnt; i++)
4995 			NFSVOPUNLOCK(dvp[i]);
4996 		NFSD_DEBUG(4, "nfsrv_proxyds: aft RPC=%d trya=%d\n", error,
4997 		    trycnt);
4998 		/* Try the Read/Getattr again if a mirror was deleted. */
4999 		if (ds != NULL && trycnt > 0 && trycnt < origmircnt)
5000 			goto tryagain;
5001 	} else {
5002 		/* Return ENOENT for any Extended Attribute error. */
5003 		error = ENOENT;
5004 	}
5005 	free(buf, M_TEMP);
5006 	NFSD_DEBUG(4, "nfsrv_proxyds: error=%d\n", error);
5007 	return (error);
5008 }
5009 
5010 /*
5011  * Get the DS mount point, fh and directory from the "pnfsd.dsfile" extended
5012  * attribute.
5013  * newnmpp - If it points to a non-NULL nmp, that is the destination and needs
5014  *           to be checked.  If it points to a NULL nmp, then it returns
5015  *           a suitable destination.
5016  * curnmp - If non-NULL, it is the source mount for the copy.
5017  */
5018 int
5019 nfsrv_dsgetsockmnt(struct vnode *vp, int lktype, char *buf, int *buflenp,
5020     int *mirrorcntp, NFSPROC_T *p, struct vnode **dvpp, fhandle_t *fhp,
5021     char *devid, char *fnamep, struct vnode **nvpp, struct nfsmount **newnmpp,
5022     struct nfsmount *curnmp, int *ippos, int *dsdirp)
5023 {
5024 	struct vnode *dvp, *nvp = NULL, **tdvpp;
5025 	struct mount *mp;
5026 	struct nfsmount *nmp, *newnmp;
5027 	struct sockaddr *sad;
5028 	struct sockaddr_in *sin;
5029 	struct nfsdevice *ds, *tds, *fndds;
5030 	struct pnfsdsfile *pf;
5031 	uint32_t dsdir;
5032 	int error, fhiszero, fnd, gotone, i, mirrorcnt;
5033 
5034 	ASSERT_VOP_LOCKED(vp, "nfsrv_dsgetsockmnt vp");
5035 	*mirrorcntp = 1;
5036 	tdvpp = dvpp;
5037 	if (nvpp != NULL)
5038 		*nvpp = NULL;
5039 	if (dvpp != NULL)
5040 		*dvpp = NULL;
5041 	if (ippos != NULL)
5042 		*ippos = -1;
5043 	if (newnmpp != NULL)
5044 		newnmp = *newnmpp;
5045 	else
5046 		newnmp = NULL;
5047 	mp = vp->v_mount;
5048 	error = vn_extattr_get(vp, IO_NODELOCKED, EXTATTR_NAMESPACE_SYSTEM,
5049 	    "pnfsd.dsfile", buflenp, buf, p);
5050 	mirrorcnt = *buflenp / sizeof(*pf);
5051 	if (error == 0 && (mirrorcnt < 1 || mirrorcnt > NFSDEV_MAXMIRRORS ||
5052 	    *buflenp != sizeof(*pf) * mirrorcnt))
5053 		error = ENOATTR;
5054 
5055 	pf = (struct pnfsdsfile *)buf;
5056 	/* If curnmp != NULL, check for a match in the mirror list. */
5057 	if (curnmp != NULL && error == 0) {
5058 		fnd = 0;
5059 		for (i = 0; i < mirrorcnt; i++, pf++) {
5060 			sad = (struct sockaddr *)&pf->dsf_sin;
5061 			if (nfsaddr2_match(sad, curnmp->nm_nam)) {
5062 				if (ippos != NULL)
5063 					*ippos = i;
5064 				fnd = 1;
5065 				break;
5066 			}
5067 		}
5068 		if (fnd == 0)
5069 			error = ENXIO;
5070 	}
5071 
5072 	gotone = 0;
5073 	pf = (struct pnfsdsfile *)buf;
5074 	NFSD_DEBUG(4, "nfsrv_dsgetsockmnt: mirrorcnt=%d err=%d\n", mirrorcnt,
5075 	    error);
5076 	for (i = 0; i < mirrorcnt && error == 0; i++, pf++) {
5077 		fhiszero = 0;
5078 		sad = (struct sockaddr *)&pf->dsf_sin;
5079 		sin = &pf->dsf_sin;
5080 		dsdir = pf->dsf_dir;
5081 		if (dsdir >= nfsrv_dsdirsize) {
5082 			printf("nfsrv_dsgetsockmnt: dsdir=%d\n", dsdir);
5083 			error = ENOATTR;
5084 		} else if (nvpp != NULL && newnmp != NULL &&
5085 		    nfsaddr2_match(sad, newnmp->nm_nam))
5086 			error = EEXIST;
5087 		if (error == 0) {
5088 			if (ippos != NULL && curnmp == NULL &&
5089 			    sad->sa_family == AF_INET &&
5090 			    sin->sin_addr.s_addr == 0)
5091 				*ippos = i;
5092 			if (NFSBCMP(&zerofh, &pf->dsf_fh, sizeof(zerofh)) == 0)
5093 				fhiszero = 1;
5094 			/* Use the socket address to find the mount point. */
5095 			fndds = NULL;
5096 			NFSDDSLOCK();
5097 			/* Find a match for the IP address. */
5098 			TAILQ_FOREACH(ds, &nfsrv_devidhead, nfsdev_list) {
5099 				if (ds->nfsdev_nmp != NULL) {
5100 					dvp = ds->nfsdev_dvp;
5101 					nmp = VFSTONFS(dvp->v_mount);
5102 					if (nmp != ds->nfsdev_nmp)
5103 						printf("different2 nmp %p %p\n",
5104 						    nmp, ds->nfsdev_nmp);
5105 					if (nfsaddr2_match(sad, nmp->nm_nam)) {
5106 						fndds = ds;
5107 						break;
5108 					}
5109 				}
5110 			}
5111 			if (fndds != NULL && newnmpp != NULL &&
5112 			    newnmp == NULL) {
5113 				/* Search for a place to make a mirror copy. */
5114 				TAILQ_FOREACH(tds, &nfsrv_devidhead,
5115 				    nfsdev_list) {
5116 					if (tds->nfsdev_nmp != NULL &&
5117 					    fndds != tds &&
5118 					    ((tds->nfsdev_mdsisset == 0 &&
5119 					      fndds->nfsdev_mdsisset == 0) ||
5120 					     (tds->nfsdev_mdsisset != 0 &&
5121 					      fndds->nfsdev_mdsisset != 0 &&
5122 					      fsidcmp(&tds->nfsdev_mdsfsid,
5123 					      &mp->mnt_stat.f_fsid) == 0))) {
5124 						*newnmpp = tds->nfsdev_nmp;
5125 						break;
5126 					}
5127 				}
5128 				if (tds != NULL) {
5129 					/*
5130 					 * Move this entry to the end of the
5131 					 * list, so it won't be selected as
5132 					 * easily the next time.
5133 					 */
5134 					TAILQ_REMOVE(&nfsrv_devidhead, tds,
5135 					    nfsdev_list);
5136 					TAILQ_INSERT_TAIL(&nfsrv_devidhead, tds,
5137 					    nfsdev_list);
5138 				}
5139 			}
5140 			NFSDDSUNLOCK();
5141 			if (fndds != NULL) {
5142 				dvp = fndds->nfsdev_dsdir[dsdir];
5143 				if (lktype != 0 || fhiszero != 0 ||
5144 				    (nvpp != NULL && *nvpp == NULL)) {
5145 					if (fhiszero != 0)
5146 						error = vn_lock(dvp,
5147 						    LK_EXCLUSIVE);
5148 					else if (lktype != 0)
5149 						error = vn_lock(dvp, lktype);
5150 					else
5151 						error = vn_lock(dvp, LK_SHARED);
5152 					/*
5153 					 * If the file handle is all 0's, try to
5154 					 * do a Lookup against the DS to acquire
5155 					 * it.
5156 					 * If dvpp == NULL or the Lookup fails,
5157 					 * unlock dvp after the call.
5158 					 */
5159 					if (error == 0 && (fhiszero != 0 ||
5160 					    (nvpp != NULL && *nvpp == NULL))) {
5161 						error = nfsrv_pnfslookupds(vp,
5162 						    dvp, pf, &nvp, p);
5163 						if (error == 0) {
5164 							if (fhiszero != 0)
5165 								nfsrv_pnfssetfh(
5166 								    vp, pf,
5167 								    devid,
5168 								    fnamep,
5169 								    nvp, p);
5170 							if (nvpp != NULL &&
5171 							    *nvpp == NULL) {
5172 								*nvpp = nvp;
5173 								*dsdirp = dsdir;
5174 							} else
5175 								vput(nvp);
5176 						}
5177 						if (error != 0 || lktype == 0)
5178 							NFSVOPUNLOCK(dvp);
5179 					}
5180 				}
5181 				if (error == 0) {
5182 					gotone++;
5183 					NFSD_DEBUG(4, "gotone=%d\n", gotone);
5184 					if (devid != NULL) {
5185 						NFSBCOPY(fndds->nfsdev_deviceid,
5186 						    devid, NFSX_V4DEVICEID);
5187 						devid += NFSX_V4DEVICEID;
5188 					}
5189 					if (dvpp != NULL)
5190 						*tdvpp++ = dvp;
5191 					if (fhp != NULL)
5192 						NFSBCOPY(&pf->dsf_fh, fhp++,
5193 						    NFSX_MYFH);
5194 					if (fnamep != NULL && gotone == 1)
5195 						strlcpy(fnamep,
5196 						    pf->dsf_filename,
5197 						    sizeof(pf->dsf_filename));
5198 				} else
5199 					NFSD_DEBUG(4, "nfsrv_dsgetsockmnt "
5200 					    "err=%d\n", error);
5201 			}
5202 		}
5203 	}
5204 	if (error == 0 && gotone == 0)
5205 		error = ENOENT;
5206 
5207 	NFSD_DEBUG(4, "eo nfsrv_dsgetsockmnt: gotone=%d err=%d\n", gotone,
5208 	    error);
5209 	if (error == 0)
5210 		*mirrorcntp = gotone;
5211 	else {
5212 		if (gotone > 0 && dvpp != NULL) {
5213 			/*
5214 			 * If the error didn't occur on the first one and
5215 			 * dvpp != NULL, the one(s) prior to the failure will
5216 			 * have locked dvp's that need to be unlocked.
5217 			 */
5218 			for (i = 0; i < gotone; i++) {
5219 				NFSVOPUNLOCK(*dvpp);
5220 				*dvpp++ = NULL;
5221 			}
5222 		}
5223 		/*
5224 		 * If it found the vnode to be copied from before a failure,
5225 		 * it needs to be vput()'d.
5226 		 */
5227 		if (nvpp != NULL && *nvpp != NULL) {
5228 			vput(*nvpp);
5229 			*nvpp = NULL;
5230 		}
5231 	}
5232 	return (error);
5233 }
5234 
5235 /*
5236  * Set the extended attribute for the Change attribute.
5237  */
5238 static int
5239 nfsrv_setextattr(struct vnode *vp, struct nfsvattr *nap, NFSPROC_T *p)
5240 {
5241 	struct pnfsdsattr dsattr;
5242 	int error;
5243 
5244 	ASSERT_VOP_ELOCKED(vp, "nfsrv_setextattr vp");
5245 	dsattr.dsa_filerev = nap->na_filerev;
5246 	dsattr.dsa_size = nap->na_size;
5247 	dsattr.dsa_atime = nap->na_atime;
5248 	dsattr.dsa_mtime = nap->na_mtime;
5249 	dsattr.dsa_bytes = nap->na_bytes;
5250 	error = vn_extattr_set(vp, IO_NODELOCKED, EXTATTR_NAMESPACE_SYSTEM,
5251 	    "pnfsd.dsattr", sizeof(dsattr), (char *)&dsattr, p);
5252 	if (error != 0)
5253 		printf("pNFS: setextattr=%d\n", error);
5254 	return (error);
5255 }
5256 
5257 static int
5258 nfsrv_readdsrpc(fhandle_t *fhp, off_t off, int len, struct ucred *cred,
5259     NFSPROC_T *p, struct nfsmount *nmp, struct mbuf **mpp, struct mbuf **mpendp)
5260 {
5261 	uint32_t *tl;
5262 	struct nfsrv_descript *nd;
5263 	nfsv4stateid_t st;
5264 	struct mbuf *m, *m2;
5265 	int error = 0, retlen, tlen, trimlen;
5266 
5267 	NFSD_DEBUG(4, "in nfsrv_readdsrpc\n");
5268 	nd = malloc(sizeof(*nd), M_TEMP, M_WAITOK | M_ZERO);
5269 	*mpp = NULL;
5270 	/*
5271 	 * Use a stateid where other is an alternating 01010 pattern and
5272 	 * seqid is 0xffffffff.  This value is not defined as special by
5273 	 * the RFC and is used by the FreeBSD NFS server to indicate an
5274 	 * MDS->DS proxy operation.
5275 	 */
5276 	st.other[0] = 0x55555555;
5277 	st.other[1] = 0x55555555;
5278 	st.other[2] = 0x55555555;
5279 	st.seqid = 0xffffffff;
5280 	nfscl_reqstart(nd, NFSPROC_READDS, nmp, (u_int8_t *)fhp, sizeof(*fhp),
5281 	    NULL, NULL, 0, 0, cred);
5282 	nfsm_stateidtom(nd, &st, NFSSTATEID_PUTSTATEID);
5283 	NFSM_BUILD(tl, uint32_t *, NFSX_UNSIGNED * 3);
5284 	txdr_hyper(off, tl);
5285 	*(tl + 2) = txdr_unsigned(len);
5286 	error = newnfs_request(nd, nmp, NULL, &nmp->nm_sockreq, NULL, p, cred,
5287 	    NFS_PROG, NFS_VER4, NULL, 1, NULL, NULL);
5288 	if (error != 0) {
5289 		free(nd, M_TEMP);
5290 		return (error);
5291 	}
5292 	if (nd->nd_repstat == 0) {
5293 		NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
5294 		NFSM_STRSIZ(retlen, len);
5295 		if (retlen > 0) {
5296 			/* Trim off the pre-data XDR from the mbuf chain. */
5297 			m = nd->nd_mrep;
5298 			while (m != NULL && m != nd->nd_md) {
5299 				if (m->m_next == nd->nd_md) {
5300 					m->m_next = NULL;
5301 					m_freem(nd->nd_mrep);
5302 					nd->nd_mrep = m = nd->nd_md;
5303 				} else
5304 					m = m->m_next;
5305 			}
5306 			if (m == NULL) {
5307 				printf("nfsrv_readdsrpc: busted mbuf list\n");
5308 				error = ENOENT;
5309 				goto nfsmout;
5310 			}
5311 
5312 			/*
5313 			 * Now, adjust first mbuf so that any XDR before the
5314 			 * read data is skipped over.
5315 			 */
5316 			trimlen = nd->nd_dpos - mtod(m, char *);
5317 			if (trimlen > 0) {
5318 				m->m_len -= trimlen;
5319 				NFSM_DATAP(m, trimlen);
5320 			}
5321 
5322 			/*
5323 			 * Truncate the mbuf chain at retlen bytes of data,
5324 			 * plus XDR padding that brings the length up to a
5325 			 * multiple of 4.
5326 			 */
5327 			tlen = NFSM_RNDUP(retlen);
5328 			do {
5329 				if (m->m_len >= tlen) {
5330 					m->m_len = tlen;
5331 					tlen = 0;
5332 					m2 = m->m_next;
5333 					m->m_next = NULL;
5334 					m_freem(m2);
5335 					break;
5336 				}
5337 				tlen -= m->m_len;
5338 				m = m->m_next;
5339 			} while (m != NULL);
5340 			if (tlen > 0) {
5341 				printf("nfsrv_readdsrpc: busted mbuf list\n");
5342 				error = ENOENT;
5343 				goto nfsmout;
5344 			}
5345 			*mpp = nd->nd_mrep;
5346 			*mpendp = m;
5347 			nd->nd_mrep = NULL;
5348 		}
5349 	} else
5350 		error = nd->nd_repstat;
5351 nfsmout:
5352 	/* If nd->nd_mrep is already NULL, this is a no-op. */
5353 	m_freem(nd->nd_mrep);
5354 	free(nd, M_TEMP);
5355 	NFSD_DEBUG(4, "nfsrv_readdsrpc error=%d\n", error);
5356 	return (error);
5357 }
5358 
5359 /*
5360  * Do a write RPC on a DS data file, using this structure for the arguments,
5361  * so that this function can be executed by a separate kernel process.
5362  */
5363 struct nfsrvwritedsdorpc {
5364 	int			done;
5365 	int			inprog;
5366 	struct task		tsk;
5367 	fhandle_t		fh;
5368 	off_t			off;
5369 	int			len;
5370 	struct nfsmount		*nmp;
5371 	struct ucred		*cred;
5372 	NFSPROC_T		*p;
5373 	struct mbuf		*m;
5374 	int			err;
5375 };
5376 
5377 static int
5378 nfsrv_writedsdorpc(struct nfsmount *nmp, fhandle_t *fhp, off_t off, int len,
5379     struct nfsvattr *nap, struct mbuf *m, struct ucred *cred, NFSPROC_T *p)
5380 {
5381 	uint32_t *tl;
5382 	struct nfsrv_descript *nd;
5383 	nfsattrbit_t attrbits;
5384 	nfsv4stateid_t st;
5385 	int commit, error, retlen;
5386 
5387 	nd = malloc(sizeof(*nd), M_TEMP, M_WAITOK | M_ZERO);
5388 	nfscl_reqstart(nd, NFSPROC_WRITE, nmp, (u_int8_t *)fhp,
5389 	    sizeof(fhandle_t), NULL, NULL, 0, 0, cred);
5390 
5391 	/*
5392 	 * Use a stateid where other is an alternating 01010 pattern and
5393 	 * seqid is 0xffffffff.  This value is not defined as special by
5394 	 * the RFC and is used by the FreeBSD NFS server to indicate an
5395 	 * MDS->DS proxy operation.
5396 	 */
5397 	st.other[0] = 0x55555555;
5398 	st.other[1] = 0x55555555;
5399 	st.other[2] = 0x55555555;
5400 	st.seqid = 0xffffffff;
5401 	nfsm_stateidtom(nd, &st, NFSSTATEID_PUTSTATEID);
5402 	NFSM_BUILD(tl, u_int32_t *, NFSX_HYPER + 2 * NFSX_UNSIGNED);
5403 	txdr_hyper(off, tl);
5404 	tl += 2;
5405 	/*
5406 	 * Do all writes FileSync, since the server doesn't hold onto dirty
5407 	 * buffers.  Since clients should be accessing the DS servers directly
5408 	 * using the pNFS layouts, this just needs to work correctly as a
5409 	 * fallback.
5410 	 */
5411 	*tl++ = txdr_unsigned(NFSWRITE_FILESYNC);
5412 	*tl = txdr_unsigned(len);
5413 	NFSD_DEBUG(4, "nfsrv_writedsdorpc: len=%d\n", len);
5414 
5415 	/* Put data in mbuf chain. */
5416 	nd->nd_mb->m_next = m;
5417 
5418 	/* Set nd_mb and nd_bpos to end of data. */
5419 	while (m->m_next != NULL)
5420 		m = m->m_next;
5421 	nd->nd_mb = m;
5422 	nfsm_set(nd, m->m_len);
5423 	NFSD_DEBUG(4, "nfsrv_writedsdorpc: lastmb len=%d\n", m->m_len);
5424 
5425 	/* Do a Getattr for the attributes that change upon writing. */
5426 	NFSZERO_ATTRBIT(&attrbits);
5427 	NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_SIZE);
5428 	NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_CHANGE);
5429 	NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_TIMEACCESS);
5430 	NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_TIMEMODIFY);
5431 	NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_SPACEUSED);
5432 	NFSM_BUILD(tl, u_int32_t *, NFSX_UNSIGNED);
5433 	*tl = txdr_unsigned(NFSV4OP_GETATTR);
5434 	(void) nfsrv_putattrbit(nd, &attrbits);
5435 	error = newnfs_request(nd, nmp, NULL, &nmp->nm_sockreq, NULL, p,
5436 	    cred, NFS_PROG, NFS_VER4, NULL, 1, NULL, NULL);
5437 	if (error != 0) {
5438 		free(nd, M_TEMP);
5439 		return (error);
5440 	}
5441 	NFSD_DEBUG(4, "nfsrv_writedsdorpc: aft writerpc=%d\n", nd->nd_repstat);
5442 	/* Get rid of weak cache consistency data for now. */
5443 	if ((nd->nd_flag & (ND_NOMOREDATA | ND_NFSV4 | ND_V4WCCATTR)) ==
5444 	    (ND_NFSV4 | ND_V4WCCATTR)) {
5445 		error = nfsv4_loadattr(nd, NULL, nap, NULL, NULL, 0, NULL, NULL,
5446 		    NULL, NULL, NULL, 0, NULL, NULL, NULL, NULL, NULL);
5447 		NFSD_DEBUG(4, "nfsrv_writedsdorpc: wcc attr=%d\n", error);
5448 		if (error != 0)
5449 			goto nfsmout;
5450 		/*
5451 		 * Get rid of Op# and status for next op.
5452 		 */
5453 		NFSM_DISSECT(tl, uint32_t *, 2 * NFSX_UNSIGNED);
5454 		if (*++tl != 0)
5455 			nd->nd_flag |= ND_NOMOREDATA;
5456 	}
5457 	if (nd->nd_repstat == 0) {
5458 		NFSM_DISSECT(tl, uint32_t *, 2 * NFSX_UNSIGNED + NFSX_VERF);
5459 		retlen = fxdr_unsigned(int, *tl++);
5460 		commit = fxdr_unsigned(int, *tl);
5461 		if (commit != NFSWRITE_FILESYNC)
5462 			error = NFSERR_IO;
5463 		NFSD_DEBUG(4, "nfsrv_writedsdorpc:retlen=%d commit=%d err=%d\n",
5464 		    retlen, commit, error);
5465 	} else
5466 		error = nd->nd_repstat;
5467 	/* We have no use for the Write Verifier since we use FileSync. */
5468 
5469 	/*
5470 	 * Get the Change, Size, Access Time and Modify Time attributes and set
5471 	 * on the Metadata file, so its attributes will be what the file's
5472 	 * would be if it had been written.
5473 	 */
5474 	if (error == 0) {
5475 		NFSM_DISSECT(tl, uint32_t *, 2 * NFSX_UNSIGNED);
5476 		error = nfsv4_loadattr(nd, NULL, nap, NULL, NULL, 0, NULL, NULL,
5477 		    NULL, NULL, NULL, 0, NULL, NULL, NULL, NULL, NULL);
5478 	}
5479 	NFSD_DEBUG(4, "nfsrv_writedsdorpc: aft loadattr=%d\n", error);
5480 nfsmout:
5481 	m_freem(nd->nd_mrep);
5482 	free(nd, M_TEMP);
5483 	NFSD_DEBUG(4, "nfsrv_writedsdorpc error=%d\n", error);
5484 	return (error);
5485 }
5486 
5487 /*
5488  * Start up the thread that will execute nfsrv_writedsdorpc().
5489  */
5490 static void
5491 start_writedsdorpc(void *arg, int pending)
5492 {
5493 	struct nfsrvwritedsdorpc *drpc;
5494 
5495 	drpc = (struct nfsrvwritedsdorpc *)arg;
5496 	drpc->err = nfsrv_writedsdorpc(drpc->nmp, &drpc->fh, drpc->off,
5497 	    drpc->len, NULL, drpc->m, drpc->cred, drpc->p);
5498 	drpc->done = 1;
5499 	NFSD_DEBUG(4, "start_writedsdorpc: err=%d\n", drpc->err);
5500 }
5501 
5502 static int
5503 nfsrv_writedsrpc(fhandle_t *fhp, off_t off, int len, struct ucred *cred,
5504     NFSPROC_T *p, struct vnode *vp, struct nfsmount **nmpp, int mirrorcnt,
5505     struct mbuf **mpp, char *cp, int *failposp)
5506 {
5507 	struct nfsrvwritedsdorpc *drpc, *tdrpc = NULL;
5508 	struct nfsvattr na;
5509 	struct mbuf *m;
5510 	int error, i, offs, ret, timo;
5511 
5512 	NFSD_DEBUG(4, "in nfsrv_writedsrpc\n");
5513 	KASSERT(*mpp != NULL, ("nfsrv_writedsrpc: NULL mbuf chain"));
5514 	drpc = NULL;
5515 	if (mirrorcnt > 1)
5516 		tdrpc = drpc = malloc(sizeof(*drpc) * (mirrorcnt - 1), M_TEMP,
5517 		    M_WAITOK);
5518 
5519 	/* Calculate offset in mbuf chain that data starts. */
5520 	offs = cp - mtod(*mpp, char *);
5521 	NFSD_DEBUG(4, "nfsrv_writedsrpc: mcopy offs=%d len=%d\n", offs, len);
5522 
5523 	/*
5524 	 * Do the write RPC for every DS, using a separate kernel process
5525 	 * for every DS except the last one.
5526 	 */
5527 	error = 0;
5528 	for (i = 0; i < mirrorcnt - 1; i++, tdrpc++) {
5529 		tdrpc->done = 0;
5530 		NFSBCOPY(fhp, &tdrpc->fh, sizeof(*fhp));
5531 		tdrpc->off = off;
5532 		tdrpc->len = len;
5533 		tdrpc->nmp = *nmpp;
5534 		tdrpc->cred = cred;
5535 		tdrpc->p = p;
5536 		tdrpc->inprog = 0;
5537 		tdrpc->err = 0;
5538 		tdrpc->m = m_copym(*mpp, offs, NFSM_RNDUP(len), M_WAITOK);
5539 		ret = EIO;
5540 		if (nfs_pnfsiothreads != 0) {
5541 			ret = nfs_pnfsio(start_writedsdorpc, tdrpc);
5542 			NFSD_DEBUG(4, "nfsrv_writedsrpc: nfs_pnfsio=%d\n",
5543 			    ret);
5544 		}
5545 		if (ret != 0) {
5546 			ret = nfsrv_writedsdorpc(*nmpp, fhp, off, len, NULL,
5547 			    tdrpc->m, cred, p);
5548 			if (nfsds_failerr(ret) && *failposp == -1)
5549 				*failposp = i;
5550 			else if (error == 0 && ret != 0)
5551 				error = ret;
5552 		}
5553 		nmpp++;
5554 		fhp++;
5555 	}
5556 	m = m_copym(*mpp, offs, NFSM_RNDUP(len), M_WAITOK);
5557 	ret = nfsrv_writedsdorpc(*nmpp, fhp, off, len, &na, m, cred, p);
5558 	if (nfsds_failerr(ret) && *failposp == -1 && mirrorcnt > 1)
5559 		*failposp = mirrorcnt - 1;
5560 	else if (error == 0 && ret != 0)
5561 		error = ret;
5562 	if (error == 0)
5563 		error = nfsrv_setextattr(vp, &na, p);
5564 	NFSD_DEBUG(4, "nfsrv_writedsrpc: aft setextat=%d\n", error);
5565 	tdrpc = drpc;
5566 	timo = hz / 50;		/* Wait for 20msec. */
5567 	if (timo < 1)
5568 		timo = 1;
5569 	for (i = 0; i < mirrorcnt - 1; i++, tdrpc++) {
5570 		/* Wait for RPCs on separate threads to complete. */
5571 		while (tdrpc->inprog != 0 && tdrpc->done == 0)
5572 			tsleep(&tdrpc->tsk, PVFS, "srvwrds", timo);
5573 		if (nfsds_failerr(tdrpc->err) && *failposp == -1)
5574 			*failposp = i;
5575 		else if (error == 0 && tdrpc->err != 0)
5576 			error = tdrpc->err;
5577 	}
5578 	free(drpc, M_TEMP);
5579 	return (error);
5580 }
5581 
5582 /*
5583  * Do a allocate RPC on a DS data file, using this structure for the arguments,
5584  * so that this function can be executed by a separate kernel process.
5585  */
5586 struct nfsrvallocatedsdorpc {
5587 	int			done;
5588 	int			inprog;
5589 	struct task		tsk;
5590 	fhandle_t		fh;
5591 	off_t			off;
5592 	off_t			len;
5593 	struct nfsmount		*nmp;
5594 	struct ucred		*cred;
5595 	NFSPROC_T		*p;
5596 	int			err;
5597 };
5598 
5599 static int
5600 nfsrv_allocatedsdorpc(struct nfsmount *nmp, fhandle_t *fhp, off_t off,
5601     off_t len, struct nfsvattr *nap, struct ucred *cred, NFSPROC_T *p)
5602 {
5603 	uint32_t *tl;
5604 	struct nfsrv_descript *nd;
5605 	nfsattrbit_t attrbits;
5606 	nfsv4stateid_t st;
5607 	int error;
5608 
5609 	nd = malloc(sizeof(*nd), M_TEMP, M_WAITOK | M_ZERO);
5610 	nfscl_reqstart(nd, NFSPROC_ALLOCATE, nmp, (u_int8_t *)fhp,
5611 	    sizeof(fhandle_t), NULL, NULL, 0, 0, cred);
5612 
5613 	/*
5614 	 * Use a stateid where other is an alternating 01010 pattern and
5615 	 * seqid is 0xffffffff.  This value is not defined as special by
5616 	 * the RFC and is used by the FreeBSD NFS server to indicate an
5617 	 * MDS->DS proxy operation.
5618 	 */
5619 	st.other[0] = 0x55555555;
5620 	st.other[1] = 0x55555555;
5621 	st.other[2] = 0x55555555;
5622 	st.seqid = 0xffffffff;
5623 	nfsm_stateidtom(nd, &st, NFSSTATEID_PUTSTATEID);
5624 	NFSM_BUILD(tl, uint32_t *, 2 * NFSX_HYPER + NFSX_UNSIGNED);
5625 	txdr_hyper(off, tl); tl += 2;
5626 	txdr_hyper(len, tl); tl += 2;
5627 	NFSD_DEBUG(4, "nfsrv_allocatedsdorpc: len=%jd\n", (intmax_t)len);
5628 
5629 	*tl = txdr_unsigned(NFSV4OP_GETATTR);
5630 	NFSGETATTR_ATTRBIT(&attrbits);
5631 	nfsrv_putattrbit(nd, &attrbits);
5632 	error = newnfs_request(nd, nmp, NULL, &nmp->nm_sockreq, NULL, p,
5633 	    cred, NFS_PROG, NFS_VER4, NULL, 1, NULL, NULL);
5634 	if (error != 0) {
5635 		free(nd, M_TEMP);
5636 		return (error);
5637 	}
5638 	NFSD_DEBUG(4, "nfsrv_allocatedsdorpc: aft allocaterpc=%d\n",
5639 	    nd->nd_repstat);
5640 	if (nd->nd_repstat == 0) {
5641 		NFSM_DISSECT(tl, uint32_t *, 2 * NFSX_UNSIGNED);
5642 		error = nfsv4_loadattr(nd, NULL, nap, NULL, NULL, 0, NULL, NULL,
5643 		    NULL, NULL, NULL, 0, NULL, NULL, NULL, NULL, NULL);
5644 	} else
5645 		error = nd->nd_repstat;
5646 	NFSD_DEBUG(4, "nfsrv_allocatedsdorpc: aft loadattr=%d\n", error);
5647 nfsmout:
5648 	m_freem(nd->nd_mrep);
5649 	free(nd, M_TEMP);
5650 	NFSD_DEBUG(4, "nfsrv_allocatedsdorpc error=%d\n", error);
5651 	return (error);
5652 }
5653 
5654 /*
5655  * Start up the thread that will execute nfsrv_allocatedsdorpc().
5656  */
5657 static void
5658 start_allocatedsdorpc(void *arg, int pending)
5659 {
5660 	struct nfsrvallocatedsdorpc *drpc;
5661 
5662 	drpc = (struct nfsrvallocatedsdorpc *)arg;
5663 	drpc->err = nfsrv_allocatedsdorpc(drpc->nmp, &drpc->fh, drpc->off,
5664 	    drpc->len, NULL, drpc->cred, drpc->p);
5665 	drpc->done = 1;
5666 	NFSD_DEBUG(4, "start_allocatedsdorpc: err=%d\n", drpc->err);
5667 }
5668 
5669 static int
5670 nfsrv_allocatedsrpc(fhandle_t *fhp, off_t off, off_t len, struct ucred *cred,
5671     NFSPROC_T *p, struct vnode *vp, struct nfsmount **nmpp, int mirrorcnt,
5672     int *failposp)
5673 {
5674 	struct nfsrvallocatedsdorpc *drpc, *tdrpc = NULL;
5675 	struct nfsvattr na;
5676 	int error, i, ret, timo;
5677 
5678 	NFSD_DEBUG(4, "in nfsrv_allocatedsrpc\n");
5679 	drpc = NULL;
5680 	if (mirrorcnt > 1)
5681 		tdrpc = drpc = malloc(sizeof(*drpc) * (mirrorcnt - 1), M_TEMP,
5682 		    M_WAITOK);
5683 
5684 	/*
5685 	 * Do the allocate RPC for every DS, using a separate kernel process
5686 	 * for every DS except the last one.
5687 	 */
5688 	error = 0;
5689 	for (i = 0; i < mirrorcnt - 1; i++, tdrpc++) {
5690 		tdrpc->done = 0;
5691 		NFSBCOPY(fhp, &tdrpc->fh, sizeof(*fhp));
5692 		tdrpc->off = off;
5693 		tdrpc->len = len;
5694 		tdrpc->nmp = *nmpp;
5695 		tdrpc->cred = cred;
5696 		tdrpc->p = p;
5697 		tdrpc->inprog = 0;
5698 		tdrpc->err = 0;
5699 		ret = EIO;
5700 		if (nfs_pnfsiothreads != 0) {
5701 			ret = nfs_pnfsio(start_allocatedsdorpc, tdrpc);
5702 			NFSD_DEBUG(4, "nfsrv_allocatedsrpc: nfs_pnfsio=%d\n",
5703 			    ret);
5704 		}
5705 		if (ret != 0) {
5706 			ret = nfsrv_allocatedsdorpc(*nmpp, fhp, off, len, NULL,
5707 			    cred, p);
5708 			if (nfsds_failerr(ret) && *failposp == -1)
5709 				*failposp = i;
5710 			else if (error == 0 && ret != 0)
5711 				error = ret;
5712 		}
5713 		nmpp++;
5714 		fhp++;
5715 	}
5716 	ret = nfsrv_allocatedsdorpc(*nmpp, fhp, off, len, &na, cred, p);
5717 	if (nfsds_failerr(ret) && *failposp == -1 && mirrorcnt > 1)
5718 		*failposp = mirrorcnt - 1;
5719 	else if (error == 0 && ret != 0)
5720 		error = ret;
5721 	if (error == 0)
5722 		error = nfsrv_setextattr(vp, &na, p);
5723 	NFSD_DEBUG(4, "nfsrv_allocatedsrpc: aft setextat=%d\n", error);
5724 	tdrpc = drpc;
5725 	timo = hz / 50;		/* Wait for 20msec. */
5726 	if (timo < 1)
5727 		timo = 1;
5728 	for (i = 0; i < mirrorcnt - 1; i++, tdrpc++) {
5729 		/* Wait for RPCs on separate threads to complete. */
5730 		while (tdrpc->inprog != 0 && tdrpc->done == 0)
5731 			tsleep(&tdrpc->tsk, PVFS, "srvalds", timo);
5732 		if (nfsds_failerr(tdrpc->err) && *failposp == -1)
5733 			*failposp = i;
5734 		else if (error == 0 && tdrpc->err != 0)
5735 			error = tdrpc->err;
5736 	}
5737 	free(drpc, M_TEMP);
5738 	return (error);
5739 }
5740 
5741 /*
5742  * Do a deallocate RPC on a DS data file, using this structure for the
5743  * arguments, so that this function can be executed by a separate kernel
5744  * process.
5745  */
5746 struct nfsrvdeallocatedsdorpc {
5747 	int			done;
5748 	int			inprog;
5749 	struct task		tsk;
5750 	fhandle_t		fh;
5751 	off_t			off;
5752 	off_t			len;
5753 	struct nfsmount		*nmp;
5754 	struct ucred		*cred;
5755 	NFSPROC_T		*p;
5756 	int			err;
5757 };
5758 
5759 static int
5760 nfsrv_deallocatedsdorpc(struct nfsmount *nmp, fhandle_t *fhp, off_t off,
5761     off_t len, struct nfsvattr *nap, struct ucred *cred, NFSPROC_T *p)
5762 {
5763 	uint32_t *tl;
5764 	struct nfsrv_descript *nd;
5765 	nfsattrbit_t attrbits;
5766 	nfsv4stateid_t st;
5767 	int error;
5768 
5769 	nd = malloc(sizeof(*nd), M_TEMP, M_WAITOK | M_ZERO);
5770 	nfscl_reqstart(nd, NFSPROC_DEALLOCATE, nmp, (u_int8_t *)fhp,
5771 	    sizeof(fhandle_t), NULL, NULL, 0, 0, cred);
5772 
5773 	/*
5774 	 * Use a stateid where other is an alternating 01010 pattern and
5775 	 * seqid is 0xffffffff.  This value is not defined as special by
5776 	 * the RFC and is used by the FreeBSD NFS server to indicate an
5777 	 * MDS->DS proxy operation.
5778 	 */
5779 	st.other[0] = 0x55555555;
5780 	st.other[1] = 0x55555555;
5781 	st.other[2] = 0x55555555;
5782 	st.seqid = 0xffffffff;
5783 	nfsm_stateidtom(nd, &st, NFSSTATEID_PUTSTATEID);
5784 	NFSM_BUILD(tl, uint32_t *, 2 * NFSX_HYPER + NFSX_UNSIGNED);
5785 	txdr_hyper(off, tl); tl += 2;
5786 	txdr_hyper(len, tl); tl += 2;
5787 	NFSD_DEBUG(4, "nfsrv_deallocatedsdorpc: len=%jd\n", (intmax_t)len);
5788 
5789 	/* Do a Getattr for the attributes that change upon writing. */
5790 	NFSZERO_ATTRBIT(&attrbits);
5791 	NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_SIZE);
5792 	NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_CHANGE);
5793 	NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_TIMEACCESS);
5794 	NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_TIMEMODIFY);
5795 	NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_SPACEUSED);
5796 	*tl = txdr_unsigned(NFSV4OP_GETATTR);
5797 	nfsrv_putattrbit(nd, &attrbits);
5798 	error = newnfs_request(nd, nmp, NULL, &nmp->nm_sockreq, NULL, p,
5799 	    cred, NFS_PROG, NFS_VER4, NULL, 1, NULL, NULL);
5800 	if (error != 0) {
5801 		free(nd, M_TEMP);
5802 		return (error);
5803 	}
5804 	NFSD_DEBUG(4, "nfsrv_deallocatedsdorpc: aft deallocaterpc=%d\n",
5805 	    nd->nd_repstat);
5806 	/* Get rid of weak cache consistency data for now. */
5807 	if ((nd->nd_flag & (ND_NOMOREDATA | ND_NFSV4 | ND_V4WCCATTR)) ==
5808 	    (ND_NFSV4 | ND_V4WCCATTR)) {
5809 		error = nfsv4_loadattr(nd, NULL, nap, NULL, NULL, 0, NULL, NULL,
5810 		    NULL, NULL, NULL, 0, NULL, NULL, NULL, NULL, NULL);
5811 		NFSD_DEBUG(4, "nfsrv_deallocatedsdorpc: wcc attr=%d\n", error);
5812 		if (error != 0)
5813 			goto nfsmout;
5814 		/*
5815 		 * Get rid of Op# and status for next op.
5816 		 */
5817 		NFSM_DISSECT(tl, uint32_t *, 2 * NFSX_UNSIGNED);
5818 		if (*++tl != 0)
5819 			nd->nd_flag |= ND_NOMOREDATA;
5820 	}
5821 	if (nd->nd_repstat == 0) {
5822 		NFSM_DISSECT(tl, uint32_t *, 2 * NFSX_UNSIGNED);
5823 		error = nfsv4_loadattr(nd, NULL, nap, NULL, NULL, 0, NULL, NULL,
5824 		    NULL, NULL, NULL, 0, NULL, NULL, NULL, NULL, NULL);
5825 	} else
5826 		error = nd->nd_repstat;
5827 	NFSD_DEBUG(4, "nfsrv_deallocatedsdorpc: aft loadattr=%d\n", error);
5828 nfsmout:
5829 	m_freem(nd->nd_mrep);
5830 	free(nd, M_TEMP);
5831 	NFSD_DEBUG(4, "nfsrv_deallocatedsdorpc error=%d\n", error);
5832 	return (error);
5833 }
5834 
5835 /*
5836  * Start up the thread that will execute nfsrv_deallocatedsdorpc().
5837  */
5838 static void
5839 start_deallocatedsdorpc(void *arg, int pending)
5840 {
5841 	struct nfsrvdeallocatedsdorpc *drpc;
5842 
5843 	drpc = (struct nfsrvdeallocatedsdorpc *)arg;
5844 	drpc->err = nfsrv_deallocatedsdorpc(drpc->nmp, &drpc->fh, drpc->off,
5845 	    drpc->len, NULL, drpc->cred, drpc->p);
5846 	drpc->done = 1;
5847 	NFSD_DEBUG(4, "start_deallocatedsdorpc: err=%d\n", drpc->err);
5848 }
5849 
5850 static int
5851 nfsrv_deallocatedsrpc(fhandle_t *fhp, off_t off, off_t len, struct ucred *cred,
5852     NFSPROC_T *p, struct vnode *vp, struct nfsmount **nmpp, int mirrorcnt,
5853     int *failposp)
5854 {
5855 	struct nfsrvdeallocatedsdorpc *drpc, *tdrpc = NULL;
5856 	struct nfsvattr na;
5857 	int error, i, ret, timo;
5858 
5859 	NFSD_DEBUG(4, "in nfsrv_deallocatedsrpc\n");
5860 	drpc = NULL;
5861 	if (mirrorcnt > 1)
5862 		tdrpc = drpc = malloc(sizeof(*drpc) * (mirrorcnt - 1), M_TEMP,
5863 		    M_WAITOK);
5864 
5865 	/*
5866 	 * Do the deallocate RPC for every DS, using a separate kernel process
5867 	 * for every DS except the last one.
5868 	 */
5869 	error = 0;
5870 	for (i = 0; i < mirrorcnt - 1; i++, tdrpc++) {
5871 		tdrpc->done = 0;
5872 		NFSBCOPY(fhp, &tdrpc->fh, sizeof(*fhp));
5873 		tdrpc->off = off;
5874 		tdrpc->len = len;
5875 		tdrpc->nmp = *nmpp;
5876 		tdrpc->cred = cred;
5877 		tdrpc->p = p;
5878 		tdrpc->inprog = 0;
5879 		tdrpc->err = 0;
5880 		ret = EIO;
5881 		if (nfs_pnfsiothreads != 0) {
5882 			ret = nfs_pnfsio(start_deallocatedsdorpc, tdrpc);
5883 			NFSD_DEBUG(4, "nfsrv_deallocatedsrpc: nfs_pnfsio=%d\n",
5884 			    ret);
5885 		}
5886 		if (ret != 0) {
5887 			ret = nfsrv_deallocatedsdorpc(*nmpp, fhp, off, len,
5888 			    NULL, cred, p);
5889 			if (nfsds_failerr(ret) && *failposp == -1)
5890 				*failposp = i;
5891 			else if (error == 0 && ret != 0)
5892 				error = ret;
5893 		}
5894 		nmpp++;
5895 		fhp++;
5896 	}
5897 	ret = nfsrv_deallocatedsdorpc(*nmpp, fhp, off, len, &na, cred, p);
5898 	if (nfsds_failerr(ret) && *failposp == -1 && mirrorcnt > 1)
5899 		*failposp = mirrorcnt - 1;
5900 	else if (error == 0 && ret != 0)
5901 		error = ret;
5902 	if (error == 0)
5903 		error = nfsrv_setextattr(vp, &na, p);
5904 	NFSD_DEBUG(4, "nfsrv_deallocatedsrpc: aft setextat=%d\n", error);
5905 	tdrpc = drpc;
5906 	timo = hz / 50;		/* Wait for 20msec. */
5907 	if (timo < 1)
5908 		timo = 1;
5909 	for (i = 0; i < mirrorcnt - 1; i++, tdrpc++) {
5910 		/* Wait for RPCs on separate threads to complete. */
5911 		while (tdrpc->inprog != 0 && tdrpc->done == 0)
5912 			tsleep(&tdrpc->tsk, PVFS, "srvalds", timo);
5913 		if (nfsds_failerr(tdrpc->err) && *failposp == -1)
5914 			*failposp = i;
5915 		else if (error == 0 && tdrpc->err != 0)
5916 			error = tdrpc->err;
5917 	}
5918 	free(drpc, M_TEMP);
5919 	return (error);
5920 }
5921 
5922 static int
5923 nfsrv_setattrdsdorpc(fhandle_t *fhp, struct ucred *cred, NFSPROC_T *p,
5924     struct vnode *vp, struct nfsmount *nmp, struct nfsvattr *nap,
5925     struct nfsvattr *dsnap)
5926 {
5927 	uint32_t *tl;
5928 	struct nfsrv_descript *nd;
5929 	nfsv4stateid_t st;
5930 	nfsattrbit_t attrbits;
5931 	int error;
5932 
5933 	NFSD_DEBUG(4, "in nfsrv_setattrdsdorpc\n");
5934 	nd = malloc(sizeof(*nd), M_TEMP, M_WAITOK | M_ZERO);
5935 	/*
5936 	 * Use a stateid where other is an alternating 01010 pattern and
5937 	 * seqid is 0xffffffff.  This value is not defined as special by
5938 	 * the RFC and is used by the FreeBSD NFS server to indicate an
5939 	 * MDS->DS proxy operation.
5940 	 */
5941 	st.other[0] = 0x55555555;
5942 	st.other[1] = 0x55555555;
5943 	st.other[2] = 0x55555555;
5944 	st.seqid = 0xffffffff;
5945 	nfscl_reqstart(nd, NFSPROC_SETATTR, nmp, (u_int8_t *)fhp, sizeof(*fhp),
5946 	    NULL, NULL, 0, 0, cred);
5947 	nfsm_stateidtom(nd, &st, NFSSTATEID_PUTSTATEID);
5948 	nfscl_fillsattr(nd, &nap->na_vattr, vp, NFSSATTR_FULL, 0);
5949 
5950 	/* Do a Getattr for the attributes that change due to writing. */
5951 	NFSZERO_ATTRBIT(&attrbits);
5952 	NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_SIZE);
5953 	NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_CHANGE);
5954 	NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_TIMEACCESS);
5955 	NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_TIMEMODIFY);
5956 	NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_SPACEUSED);
5957 	NFSM_BUILD(tl, u_int32_t *, NFSX_UNSIGNED);
5958 	*tl = txdr_unsigned(NFSV4OP_GETATTR);
5959 	(void) nfsrv_putattrbit(nd, &attrbits);
5960 	error = newnfs_request(nd, nmp, NULL, &nmp->nm_sockreq, NULL, p, cred,
5961 	    NFS_PROG, NFS_VER4, NULL, 1, NULL, NULL);
5962 	if (error != 0) {
5963 		free(nd, M_TEMP);
5964 		return (error);
5965 	}
5966 	NFSD_DEBUG(4, "nfsrv_setattrdsdorpc: aft setattrrpc=%d\n",
5967 	    nd->nd_repstat);
5968 	/* Get rid of weak cache consistency data for now. */
5969 	if ((nd->nd_flag & (ND_NOMOREDATA | ND_NFSV4 | ND_V4WCCATTR)) ==
5970 	    (ND_NFSV4 | ND_V4WCCATTR)) {
5971 		error = nfsv4_loadattr(nd, NULL, dsnap, NULL, NULL, 0, NULL,
5972 		    NULL, NULL, NULL, NULL, 0, NULL, NULL, NULL, NULL, NULL);
5973 		NFSD_DEBUG(4, "nfsrv_setattrdsdorpc: wcc attr=%d\n", error);
5974 		if (error != 0)
5975 			goto nfsmout;
5976 		/*
5977 		 * Get rid of Op# and status for next op.
5978 		 */
5979 		NFSM_DISSECT(tl, uint32_t *, 2 * NFSX_UNSIGNED);
5980 		if (*++tl != 0)
5981 			nd->nd_flag |= ND_NOMOREDATA;
5982 	}
5983 	error = nfsrv_getattrbits(nd, &attrbits, NULL, NULL);
5984 	if (error != 0)
5985 		goto nfsmout;
5986 	if (nd->nd_repstat != 0)
5987 		error = nd->nd_repstat;
5988 	/*
5989 	 * Get the Change, Size, Access Time and Modify Time attributes and set
5990 	 * on the Metadata file, so its attributes will be what the file's
5991 	 * would be if it had been written.
5992 	 */
5993 	if (error == 0) {
5994 		NFSM_DISSECT(tl, uint32_t *, 2 * NFSX_UNSIGNED);
5995 		error = nfsv4_loadattr(nd, NULL, dsnap, NULL, NULL, 0, NULL,
5996 		    NULL, NULL, NULL, NULL, 0, NULL, NULL, NULL, NULL, NULL);
5997 	}
5998 	NFSD_DEBUG(4, "nfsrv_setattrdsdorpc: aft setattr loadattr=%d\n", error);
5999 nfsmout:
6000 	m_freem(nd->nd_mrep);
6001 	free(nd, M_TEMP);
6002 	NFSD_DEBUG(4, "nfsrv_setattrdsdorpc error=%d\n", error);
6003 	return (error);
6004 }
6005 
6006 struct nfsrvsetattrdsdorpc {
6007 	int			done;
6008 	int			inprog;
6009 	struct task		tsk;
6010 	fhandle_t		fh;
6011 	struct nfsmount		*nmp;
6012 	struct vnode		*vp;
6013 	struct ucred		*cred;
6014 	NFSPROC_T		*p;
6015 	struct nfsvattr		na;
6016 	struct nfsvattr		dsna;
6017 	int			err;
6018 };
6019 
6020 /*
6021  * Start up the thread that will execute nfsrv_setattrdsdorpc().
6022  */
6023 static void
6024 start_setattrdsdorpc(void *arg, int pending)
6025 {
6026 	struct nfsrvsetattrdsdorpc *drpc;
6027 
6028 	drpc = (struct nfsrvsetattrdsdorpc *)arg;
6029 	drpc->err = nfsrv_setattrdsdorpc(&drpc->fh, drpc->cred, drpc->p,
6030 	    drpc->vp, drpc->nmp, &drpc->na, &drpc->dsna);
6031 	drpc->done = 1;
6032 }
6033 
6034 static int
6035 nfsrv_setattrdsrpc(fhandle_t *fhp, struct ucred *cred, NFSPROC_T *p,
6036     struct vnode *vp, struct nfsmount **nmpp, int mirrorcnt,
6037     struct nfsvattr *nap, int *failposp)
6038 {
6039 	struct nfsrvsetattrdsdorpc *drpc, *tdrpc = NULL;
6040 	struct nfsvattr na;
6041 	int error, i, ret, timo;
6042 
6043 	NFSD_DEBUG(4, "in nfsrv_setattrdsrpc\n");
6044 	drpc = NULL;
6045 	if (mirrorcnt > 1)
6046 		tdrpc = drpc = malloc(sizeof(*drpc) * (mirrorcnt - 1), M_TEMP,
6047 		    M_WAITOK);
6048 
6049 	/*
6050 	 * Do the setattr RPC for every DS, using a separate kernel process
6051 	 * for every DS except the last one.
6052 	 */
6053 	error = 0;
6054 	for (i = 0; i < mirrorcnt - 1; i++, tdrpc++) {
6055 		tdrpc->done = 0;
6056 		tdrpc->inprog = 0;
6057 		NFSBCOPY(fhp, &tdrpc->fh, sizeof(*fhp));
6058 		tdrpc->nmp = *nmpp;
6059 		tdrpc->vp = vp;
6060 		tdrpc->cred = cred;
6061 		tdrpc->p = p;
6062 		tdrpc->na = *nap;
6063 		tdrpc->err = 0;
6064 		ret = EIO;
6065 		if (nfs_pnfsiothreads != 0) {
6066 			ret = nfs_pnfsio(start_setattrdsdorpc, tdrpc);
6067 			NFSD_DEBUG(4, "nfsrv_setattrdsrpc: nfs_pnfsio=%d\n",
6068 			    ret);
6069 		}
6070 		if (ret != 0) {
6071 			ret = nfsrv_setattrdsdorpc(fhp, cred, p, vp, *nmpp, nap,
6072 			    &na);
6073 			if (nfsds_failerr(ret) && *failposp == -1)
6074 				*failposp = i;
6075 			else if (error == 0 && ret != 0)
6076 				error = ret;
6077 		}
6078 		nmpp++;
6079 		fhp++;
6080 	}
6081 	ret = nfsrv_setattrdsdorpc(fhp, cred, p, vp, *nmpp, nap, &na);
6082 	if (nfsds_failerr(ret) && *failposp == -1 && mirrorcnt > 1)
6083 		*failposp = mirrorcnt - 1;
6084 	else if (error == 0 && ret != 0)
6085 		error = ret;
6086 	if (error == 0)
6087 		error = nfsrv_setextattr(vp, &na, p);
6088 	NFSD_DEBUG(4, "nfsrv_setattrdsrpc: aft setextat=%d\n", error);
6089 	tdrpc = drpc;
6090 	timo = hz / 50;		/* Wait for 20msec. */
6091 	if (timo < 1)
6092 		timo = 1;
6093 	for (i = 0; i < mirrorcnt - 1; i++, tdrpc++) {
6094 		/* Wait for RPCs on separate threads to complete. */
6095 		while (tdrpc->inprog != 0 && tdrpc->done == 0)
6096 			tsleep(&tdrpc->tsk, PVFS, "srvsads", timo);
6097 		if (nfsds_failerr(tdrpc->err) && *failposp == -1)
6098 			*failposp = i;
6099 		else if (error == 0 && tdrpc->err != 0)
6100 			error = tdrpc->err;
6101 	}
6102 	free(drpc, M_TEMP);
6103 	return (error);
6104 }
6105 
6106 /*
6107  * Do a Setattr of an NFSv4 ACL on the DS file.
6108  */
6109 static int
6110 nfsrv_setacldsdorpc(fhandle_t *fhp, struct ucred *cred, NFSPROC_T *p,
6111     struct vnode *vp, struct nfsmount *nmp, struct acl *aclp)
6112 {
6113 	struct nfsrv_descript *nd;
6114 	nfsv4stateid_t st;
6115 	nfsattrbit_t attrbits;
6116 	int error;
6117 
6118 	NFSD_DEBUG(4, "in nfsrv_setacldsdorpc\n");
6119 	nd = malloc(sizeof(*nd), M_TEMP, M_WAITOK | M_ZERO);
6120 	/*
6121 	 * Use a stateid where other is an alternating 01010 pattern and
6122 	 * seqid is 0xffffffff.  This value is not defined as special by
6123 	 * the RFC and is used by the FreeBSD NFS server to indicate an
6124 	 * MDS->DS proxy operation.
6125 	 */
6126 	st.other[0] = 0x55555555;
6127 	st.other[1] = 0x55555555;
6128 	st.other[2] = 0x55555555;
6129 	st.seqid = 0xffffffff;
6130 	nfscl_reqstart(nd, NFSPROC_SETACL, nmp, (u_int8_t *)fhp, sizeof(*fhp),
6131 	    NULL, NULL, 0, 0, cred);
6132 	nfsm_stateidtom(nd, &st, NFSSTATEID_PUTSTATEID);
6133 	NFSZERO_ATTRBIT(&attrbits);
6134 	NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_ACL);
6135 	/*
6136 	 * The "vp" argument to nfsv4_fillattr() is only used for vnode_type(),
6137 	 * so passing in the metadata "vp" will be ok, since it is of
6138 	 * the same type (VREG).
6139 	 */
6140 	nfsv4_fillattr(nd, NULL, vp, aclp, NULL, NULL, 0, &attrbits, NULL,
6141 	    NULL, 0, 0, 0, 0, 0, NULL);
6142 	error = newnfs_request(nd, nmp, NULL, &nmp->nm_sockreq, NULL, p, cred,
6143 	    NFS_PROG, NFS_VER4, NULL, 1, NULL, NULL);
6144 	if (error != 0) {
6145 		free(nd, M_TEMP);
6146 		return (error);
6147 	}
6148 	NFSD_DEBUG(4, "nfsrv_setacldsdorpc: aft setaclrpc=%d\n",
6149 	    nd->nd_repstat);
6150 	error = nd->nd_repstat;
6151 	m_freem(nd->nd_mrep);
6152 	free(nd, M_TEMP);
6153 	return (error);
6154 }
6155 
6156 struct nfsrvsetacldsdorpc {
6157 	int			done;
6158 	int			inprog;
6159 	struct task		tsk;
6160 	fhandle_t		fh;
6161 	struct nfsmount		*nmp;
6162 	struct vnode		*vp;
6163 	struct ucred		*cred;
6164 	NFSPROC_T		*p;
6165 	struct acl		*aclp;
6166 	int			err;
6167 };
6168 
6169 /*
6170  * Start up the thread that will execute nfsrv_setacldsdorpc().
6171  */
6172 static void
6173 start_setacldsdorpc(void *arg, int pending)
6174 {
6175 	struct nfsrvsetacldsdorpc *drpc;
6176 
6177 	drpc = (struct nfsrvsetacldsdorpc *)arg;
6178 	drpc->err = nfsrv_setacldsdorpc(&drpc->fh, drpc->cred, drpc->p,
6179 	    drpc->vp, drpc->nmp, drpc->aclp);
6180 	drpc->done = 1;
6181 }
6182 
6183 static int
6184 nfsrv_setacldsrpc(fhandle_t *fhp, struct ucred *cred, NFSPROC_T *p,
6185     struct vnode *vp, struct nfsmount **nmpp, int mirrorcnt, struct acl *aclp,
6186     int *failposp)
6187 {
6188 	struct nfsrvsetacldsdorpc *drpc, *tdrpc = NULL;
6189 	int error, i, ret, timo;
6190 
6191 	NFSD_DEBUG(4, "in nfsrv_setacldsrpc\n");
6192 	drpc = NULL;
6193 	if (mirrorcnt > 1)
6194 		tdrpc = drpc = malloc(sizeof(*drpc) * (mirrorcnt - 1), M_TEMP,
6195 		    M_WAITOK);
6196 
6197 	/*
6198 	 * Do the setattr RPC for every DS, using a separate kernel process
6199 	 * for every DS except the last one.
6200 	 */
6201 	error = 0;
6202 	for (i = 0; i < mirrorcnt - 1; i++, tdrpc++) {
6203 		tdrpc->done = 0;
6204 		tdrpc->inprog = 0;
6205 		NFSBCOPY(fhp, &tdrpc->fh, sizeof(*fhp));
6206 		tdrpc->nmp = *nmpp;
6207 		tdrpc->vp = vp;
6208 		tdrpc->cred = cred;
6209 		tdrpc->p = p;
6210 		tdrpc->aclp = aclp;
6211 		tdrpc->err = 0;
6212 		ret = EIO;
6213 		if (nfs_pnfsiothreads != 0) {
6214 			ret = nfs_pnfsio(start_setacldsdorpc, tdrpc);
6215 			NFSD_DEBUG(4, "nfsrv_setacldsrpc: nfs_pnfsio=%d\n",
6216 			    ret);
6217 		}
6218 		if (ret != 0) {
6219 			ret = nfsrv_setacldsdorpc(fhp, cred, p, vp, *nmpp,
6220 			    aclp);
6221 			if (nfsds_failerr(ret) && *failposp == -1)
6222 				*failposp = i;
6223 			else if (error == 0 && ret != 0)
6224 				error = ret;
6225 		}
6226 		nmpp++;
6227 		fhp++;
6228 	}
6229 	ret = nfsrv_setacldsdorpc(fhp, cred, p, vp, *nmpp, aclp);
6230 	if (nfsds_failerr(ret) && *failposp == -1 && mirrorcnt > 1)
6231 		*failposp = mirrorcnt - 1;
6232 	else if (error == 0 && ret != 0)
6233 		error = ret;
6234 	NFSD_DEBUG(4, "nfsrv_setacldsrpc: aft setextat=%d\n", error);
6235 	tdrpc = drpc;
6236 	timo = hz / 50;		/* Wait for 20msec. */
6237 	if (timo < 1)
6238 		timo = 1;
6239 	for (i = 0; i < mirrorcnt - 1; i++, tdrpc++) {
6240 		/* Wait for RPCs on separate threads to complete. */
6241 		while (tdrpc->inprog != 0 && tdrpc->done == 0)
6242 			tsleep(&tdrpc->tsk, PVFS, "srvacds", timo);
6243 		if (nfsds_failerr(tdrpc->err) && *failposp == -1)
6244 			*failposp = i;
6245 		else if (error == 0 && tdrpc->err != 0)
6246 			error = tdrpc->err;
6247 	}
6248 	free(drpc, M_TEMP);
6249 	return (error);
6250 }
6251 
6252 /*
6253  * Getattr call to the DS for the attributes that change due to writing.
6254  */
6255 static int
6256 nfsrv_getattrdsrpc(fhandle_t *fhp, struct ucred *cred, NFSPROC_T *p,
6257     struct vnode *vp, struct nfsmount *nmp, struct nfsvattr *nap)
6258 {
6259 	struct nfsrv_descript *nd;
6260 	int error;
6261 	nfsattrbit_t attrbits;
6262 
6263 	NFSD_DEBUG(4, "in nfsrv_getattrdsrpc\n");
6264 	nd = malloc(sizeof(*nd), M_TEMP, M_WAITOK | M_ZERO);
6265 	nfscl_reqstart(nd, NFSPROC_GETATTR, nmp, (u_int8_t *)fhp,
6266 	    sizeof(fhandle_t), NULL, NULL, 0, 0, cred);
6267 	NFSZERO_ATTRBIT(&attrbits);
6268 	NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_SIZE);
6269 	NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_CHANGE);
6270 	NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_TIMEACCESS);
6271 	NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_TIMEMODIFY);
6272 	NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_SPACEUSED);
6273 	(void) nfsrv_putattrbit(nd, &attrbits);
6274 	error = newnfs_request(nd, nmp, NULL, &nmp->nm_sockreq, NULL, p, cred,
6275 	    NFS_PROG, NFS_VER4, NULL, 1, NULL, NULL);
6276 	if (error != 0) {
6277 		free(nd, M_TEMP);
6278 		return (error);
6279 	}
6280 	NFSD_DEBUG(4, "nfsrv_getattrdsrpc: aft getattrrpc=%d\n",
6281 	    nd->nd_repstat);
6282 	if (nd->nd_repstat == 0) {
6283 		error = nfsv4_loadattr(nd, NULL, nap, NULL, NULL, 0,
6284 		    NULL, NULL, NULL, NULL, NULL, 0, NULL, NULL, NULL,
6285 		    NULL, NULL);
6286 		/*
6287 		 * We can only save the updated values in the extended
6288 		 * attribute if the vp is exclusively locked.
6289 		 * This should happen when any of the following operations
6290 		 * occur on the vnode:
6291 		 *    Close, Delegreturn, LayoutCommit, LayoutReturn
6292 		 * As such, the updated extended attribute should get saved
6293 		 * before nfsrv_checkdsattr() returns 0 and allows the cached
6294 		 * attributes to be returned without calling this function.
6295 		 */
6296 		if (error == 0 && VOP_ISLOCKED(vp) == LK_EXCLUSIVE) {
6297 			error = nfsrv_setextattr(vp, nap, p);
6298 			NFSD_DEBUG(4, "nfsrv_getattrdsrpc: aft setextat=%d\n",
6299 			    error);
6300 		}
6301 	} else
6302 		error = nd->nd_repstat;
6303 	m_freem(nd->nd_mrep);
6304 	free(nd, M_TEMP);
6305 	NFSD_DEBUG(4, "nfsrv_getattrdsrpc error=%d\n", error);
6306 	return (error);
6307 }
6308 
6309 /*
6310  * Seek call to a DS.
6311  */
6312 static int
6313 nfsrv_seekdsrpc(fhandle_t *fhp, off_t *offp, int content, bool *eofp,
6314     struct ucred *cred, NFSPROC_T *p, struct nfsmount *nmp)
6315 {
6316 	uint32_t *tl;
6317 	struct nfsrv_descript *nd;
6318 	nfsv4stateid_t st;
6319 	int error;
6320 
6321 	NFSD_DEBUG(4, "in nfsrv_seekdsrpc\n");
6322 	/*
6323 	 * Use a stateid where other is an alternating 01010 pattern and
6324 	 * seqid is 0xffffffff.  This value is not defined as special by
6325 	 * the RFC and is used by the FreeBSD NFS server to indicate an
6326 	 * MDS->DS proxy operation.
6327 	 */
6328 	st.other[0] = 0x55555555;
6329 	st.other[1] = 0x55555555;
6330 	st.other[2] = 0x55555555;
6331 	st.seqid = 0xffffffff;
6332 	nd = malloc(sizeof(*nd), M_TEMP, M_WAITOK | M_ZERO);
6333 	nfscl_reqstart(nd, NFSPROC_SEEKDS, nmp, (u_int8_t *)fhp,
6334 	    sizeof(fhandle_t), NULL, NULL, 0, 0, cred);
6335 	nfsm_stateidtom(nd, &st, NFSSTATEID_PUTSTATEID);
6336 	NFSM_BUILD(tl, uint32_t *, NFSX_HYPER + NFSX_UNSIGNED);
6337 	txdr_hyper(*offp, tl); tl += 2;
6338 	*tl = txdr_unsigned(content);
6339 	error = newnfs_request(nd, nmp, NULL, &nmp->nm_sockreq, NULL, p, cred,
6340 	    NFS_PROG, NFS_VER4, NULL, 1, NULL, NULL);
6341 	if (error != 0) {
6342 		free(nd, M_TEMP);
6343 		return (error);
6344 	}
6345 	NFSD_DEBUG(4, "nfsrv_seekdsrpc: aft seekrpc=%d\n", nd->nd_repstat);
6346 	if (nd->nd_repstat == 0) {
6347 		NFSM_DISSECT(tl, uint32_t *, NFSX_UNSIGNED + NFSX_HYPER);
6348 		if (*tl++ == newnfs_true)
6349 			*eofp = true;
6350 		else
6351 			*eofp = false;
6352 		*offp = fxdr_hyper(tl);
6353 	} else
6354 		error = nd->nd_repstat;
6355 nfsmout:
6356 	m_freem(nd->nd_mrep);
6357 	free(nd, M_TEMP);
6358 	NFSD_DEBUG(4, "nfsrv_seekdsrpc error=%d\n", error);
6359 	return (error);
6360 }
6361 
6362 /*
6363  * Get the device id and file handle for a DS file.
6364  */
6365 int
6366 nfsrv_dsgetdevandfh(struct vnode *vp, NFSPROC_T *p, int *mirrorcntp,
6367     fhandle_t *fhp, char *devid)
6368 {
6369 	int buflen, error;
6370 	char *buf;
6371 
6372 	buflen = 1024;
6373 	buf = malloc(buflen, M_TEMP, M_WAITOK);
6374 	error = nfsrv_dsgetsockmnt(vp, 0, buf, &buflen, mirrorcntp, p, NULL,
6375 	    fhp, devid, NULL, NULL, NULL, NULL, NULL, NULL);
6376 	free(buf, M_TEMP);
6377 	return (error);
6378 }
6379 
6380 /*
6381  * Do a Lookup against the DS for the filename.
6382  */
6383 static int
6384 nfsrv_pnfslookupds(struct vnode *vp, struct vnode *dvp, struct pnfsdsfile *pf,
6385     struct vnode **nvpp, NFSPROC_T *p)
6386 {
6387 	struct nameidata named;
6388 	struct ucred *tcred;
6389 	char *bufp;
6390 	u_long *hashp;
6391 	struct vnode *nvp;
6392 	int error;
6393 
6394 	tcred = newnfs_getcred();
6395 	named.ni_cnd.cn_nameiop = LOOKUP;
6396 	named.ni_cnd.cn_lkflags = LK_SHARED | LK_RETRY;
6397 	named.ni_cnd.cn_cred = tcred;
6398 	named.ni_cnd.cn_flags = ISLASTCN | LOCKPARENT | LOCKLEAF;
6399 	nfsvno_setpathbuf(&named, &bufp, &hashp);
6400 	named.ni_cnd.cn_nameptr = bufp;
6401 	named.ni_cnd.cn_namelen = strlen(pf->dsf_filename);
6402 	strlcpy(bufp, pf->dsf_filename, NAME_MAX);
6403 	NFSD_DEBUG(4, "nfsrv_pnfslookupds: filename=%s\n", bufp);
6404 	error = VOP_LOOKUP(dvp, &nvp, &named.ni_cnd);
6405 	NFSD_DEBUG(4, "nfsrv_pnfslookupds: aft LOOKUP=%d\n", error);
6406 	NFSFREECRED(tcred);
6407 	nfsvno_relpathbuf(&named);
6408 	if (error == 0)
6409 		*nvpp = nvp;
6410 	NFSD_DEBUG(4, "eo nfsrv_pnfslookupds=%d\n", error);
6411 	return (error);
6412 }
6413 
6414 /*
6415  * Set the file handle to the correct one.
6416  */
6417 static void
6418 nfsrv_pnfssetfh(struct vnode *vp, struct pnfsdsfile *pf, char *devid,
6419     char *fnamep, struct vnode *nvp, NFSPROC_T *p)
6420 {
6421 	struct nfsnode *np;
6422 	int ret = 0;
6423 
6424 	np = VTONFS(nvp);
6425 	NFSBCOPY(np->n_fhp->nfh_fh, &pf->dsf_fh, NFSX_MYFH);
6426 	/*
6427 	 * We can only do a vn_set_extattr() if the vnode is exclusively
6428 	 * locked and vn_start_write() has been done.  If devid != NULL or
6429 	 * fnamep != NULL or the vnode is shared locked, vn_start_write()
6430 	 * may not have been done.
6431 	 * If not done now, it will be done on a future call.
6432 	 */
6433 	if (devid == NULL && fnamep == NULL && NFSVOPISLOCKED(vp) ==
6434 	    LK_EXCLUSIVE)
6435 		ret = vn_extattr_set(vp, IO_NODELOCKED,
6436 		    EXTATTR_NAMESPACE_SYSTEM, "pnfsd.dsfile", sizeof(*pf),
6437 		    (char *)pf, p);
6438 	NFSD_DEBUG(4, "eo nfsrv_pnfssetfh=%d\n", ret);
6439 }
6440 
6441 /*
6442  * Cause RPCs waiting on "nmp" to fail.  This is called for a DS mount point
6443  * when the DS has failed.
6444  */
6445 void
6446 nfsrv_killrpcs(struct nfsmount *nmp)
6447 {
6448 
6449 	/*
6450 	 * Call newnfs_nmcancelreqs() to cause
6451 	 * any RPCs in progress on the mount point to
6452 	 * fail.
6453 	 * This will cause any process waiting for an
6454 	 * RPC to complete while holding a vnode lock
6455 	 * on the mounted-on vnode (such as "df" or
6456 	 * a non-forced "umount") to fail.
6457 	 * This will unlock the mounted-on vnode so
6458 	 * a forced dismount can succeed.
6459 	 * The NFSMNTP_CANCELRPCS flag should be set when this function is
6460 	 * called.
6461 	 */
6462 	newnfs_nmcancelreqs(nmp);
6463 }
6464 
6465 /*
6466  * Sum up the statfs info for each of the DSs, so that the client will
6467  * receive the total for all DSs.
6468  */
6469 static int
6470 nfsrv_pnfsstatfs(struct statfs *sf, struct mount *mp)
6471 {
6472 	struct statfs *tsf;
6473 	struct nfsdevice *ds;
6474 	struct vnode **dvpp, **tdvpp, *dvp;
6475 	uint64_t tot;
6476 	int cnt, error = 0, i;
6477 
6478 	if (nfsrv_devidcnt <= 0)
6479 		return (ENXIO);
6480 	dvpp = mallocarray(nfsrv_devidcnt, sizeof(*dvpp), M_TEMP, M_WAITOK);
6481 	tsf = malloc(sizeof(*tsf), M_TEMP, M_WAITOK);
6482 
6483 	/* Get an array of the dvps for the DSs. */
6484 	tdvpp = dvpp;
6485 	i = 0;
6486 	NFSDDSLOCK();
6487 	/* First, search for matches for same file system. */
6488 	TAILQ_FOREACH(ds, &nfsrv_devidhead, nfsdev_list) {
6489 		if (ds->nfsdev_nmp != NULL && ds->nfsdev_mdsisset != 0 &&
6490 		    fsidcmp(&ds->nfsdev_mdsfsid, &mp->mnt_stat.f_fsid) == 0) {
6491 			if (++i > nfsrv_devidcnt)
6492 				break;
6493 			*tdvpp++ = ds->nfsdev_dvp;
6494 		}
6495 	}
6496 	/*
6497 	 * If no matches for same file system, total all servers not assigned
6498 	 * to a file system.
6499 	 */
6500 	if (i == 0) {
6501 		TAILQ_FOREACH(ds, &nfsrv_devidhead, nfsdev_list) {
6502 			if (ds->nfsdev_nmp != NULL &&
6503 			    ds->nfsdev_mdsisset == 0) {
6504 				if (++i > nfsrv_devidcnt)
6505 					break;
6506 				*tdvpp++ = ds->nfsdev_dvp;
6507 			}
6508 		}
6509 	}
6510 	NFSDDSUNLOCK();
6511 	cnt = i;
6512 
6513 	/* Do a VFS_STATFS() for each of the DSs and sum them up. */
6514 	tdvpp = dvpp;
6515 	for (i = 0; i < cnt && error == 0; i++) {
6516 		dvp = *tdvpp++;
6517 		error = VFS_STATFS(dvp->v_mount, tsf);
6518 		if (error == 0) {
6519 			if (sf->f_bsize == 0) {
6520 				if (tsf->f_bsize > 0)
6521 					sf->f_bsize = tsf->f_bsize;
6522 				else
6523 					sf->f_bsize = 8192;
6524 			}
6525 			if (tsf->f_blocks > 0) {
6526 				if (sf->f_bsize != tsf->f_bsize) {
6527 					tot = tsf->f_blocks * tsf->f_bsize;
6528 					sf->f_blocks += (tot / sf->f_bsize);
6529 				} else
6530 					sf->f_blocks += tsf->f_blocks;
6531 			}
6532 			if (tsf->f_bfree > 0) {
6533 				if (sf->f_bsize != tsf->f_bsize) {
6534 					tot = tsf->f_bfree * tsf->f_bsize;
6535 					sf->f_bfree += (tot / sf->f_bsize);
6536 				} else
6537 					sf->f_bfree += tsf->f_bfree;
6538 			}
6539 			if (tsf->f_bavail > 0) {
6540 				if (sf->f_bsize != tsf->f_bsize) {
6541 					tot = tsf->f_bavail * tsf->f_bsize;
6542 					sf->f_bavail += (tot / sf->f_bsize);
6543 				} else
6544 					sf->f_bavail += tsf->f_bavail;
6545 			}
6546 		}
6547 	}
6548 	free(tsf, M_TEMP);
6549 	free(dvpp, M_TEMP);
6550 	return (error);
6551 }
6552 
6553 /*
6554  * Set an NFSv4 acl.
6555  */
6556 int
6557 nfsrv_setacl(struct vnode *vp, NFSACL_T *aclp, struct ucred *cred, NFSPROC_T *p)
6558 {
6559 	int error;
6560 
6561 	if (nfsrv_useacl == 0 || nfs_supportsnfsv4acls(vp) == 0) {
6562 		error = NFSERR_ATTRNOTSUPP;
6563 		goto out;
6564 	}
6565 	/*
6566 	 * With NFSv4 ACLs, chmod(2) may need to add additional entries.
6567 	 * Make sure it has enough room for that - splitting every entry
6568 	 * into two and appending "canonical six" entries at the end.
6569 	 * Cribbed out of kern/vfs_acl.c - Rick M.
6570 	 */
6571 	if (aclp->acl_cnt > (ACL_MAX_ENTRIES - 6) / 2) {
6572 		error = NFSERR_ATTRNOTSUPP;
6573 		goto out;
6574 	}
6575 	error = VOP_SETACL(vp, ACL_TYPE_NFS4, aclp, cred, p);
6576 	if (error == 0) {
6577 		error = nfsrv_dssetacl(vp, aclp, cred, p);
6578 		if (error == ENOENT)
6579 			error = 0;
6580 	}
6581 
6582 out:
6583 	NFSEXITCODE(error);
6584 	return (error);
6585 }
6586 
6587 /*
6588  * Seek vnode op call (actually it is a VOP_IOCTL()).
6589  * This function is called with the vnode locked, but unlocks and vrele()s
6590  * the vp before returning.
6591  */
6592 int
6593 nfsvno_seek(struct nfsrv_descript *nd, struct vnode *vp, u_long cmd,
6594     off_t *offp, int content, bool *eofp, struct ucred *cred, NFSPROC_T *p)
6595 {
6596 	struct nfsvattr at;
6597 	int error, ret;
6598 
6599 	ASSERT_VOP_LOCKED(vp, "nfsvno_seek vp");
6600 	/*
6601 	 * Attempt to seek on a DS file. A return of ENOENT implies
6602 	 * there is no DS file to seek on.
6603 	 */
6604 	error = nfsrv_proxyds(vp, 0, 0, cred, p, NFSPROC_SEEKDS, NULL,
6605 	    NULL, NULL, NULL, NULL, offp, content, eofp);
6606 	if (error != ENOENT) {
6607 		vput(vp);
6608 		return (error);
6609 	}
6610 
6611 	/*
6612 	 * Do the VOP_IOCTL() call.  For the case where *offp == file_size,
6613 	 * VOP_IOCTL() will return ENXIO.  However, the correct reply for
6614 	 * NFSv4.2 is *eofp == true and error == 0 for this case.
6615 	 */
6616 	NFSVOPUNLOCK(vp);
6617 	error = VOP_IOCTL(vp, cmd, offp, 0, cred, p);
6618 	*eofp = false;
6619 	if (error == ENXIO || (error == 0 && cmd == FIOSEEKHOLE)) {
6620 		/* Handle the cases where we might be at EOF. */
6621 		ret = nfsvno_getattr(vp, &at, nd, p, 0, NULL);
6622 		if (ret == 0 && *offp == at.na_size) {
6623 			*eofp = true;
6624 			error = 0;
6625 		}
6626 		if (ret != 0 && error == 0)
6627 			error = ret;
6628 	}
6629 	vrele(vp);
6630 	NFSEXITCODE(error);
6631 	return (error);
6632 }
6633 
6634 /*
6635  * Allocate vnode op call.
6636  */
6637 int
6638 nfsvno_allocate(struct vnode *vp, off_t off, off_t len, struct ucred *cred,
6639     NFSPROC_T *p)
6640 {
6641 	int error;
6642 	off_t olen;
6643 
6644 	ASSERT_VOP_ELOCKED(vp, "nfsvno_allocate vp");
6645 	/*
6646 	 * Attempt to allocate on a DS file. A return of ENOENT implies
6647 	 * there is no DS file to allocate on.
6648 	 */
6649 	error = nfsrv_proxyds(vp, off, 0, cred, p, NFSPROC_ALLOCATE, NULL,
6650 	    NULL, NULL, NULL, NULL, &len, 0, NULL);
6651 	if (error != ENOENT)
6652 		return (error);
6653 
6654 	/*
6655 	 * Do the actual VOP_ALLOCATE(), looping so long as
6656 	 * progress is being made, to achieve completion.
6657 	 */
6658 	do {
6659 		olen = len;
6660 		error = VOP_ALLOCATE(vp, &off, &len, IO_SYNC, cred);
6661 		if (error == 0 && len > 0 && olen > len)
6662 			maybe_yield();
6663 	} while (error == 0 && len > 0 && olen > len);
6664 	if (error == 0 && len > 0)
6665 		error = NFSERR_IO;
6666 	NFSEXITCODE(error);
6667 	return (error);
6668 }
6669 
6670 /*
6671  * Deallocate vnode op call.
6672  */
6673 int
6674 nfsvno_deallocate(struct vnode *vp, off_t off, off_t len, struct ucred *cred,
6675     NFSPROC_T *p)
6676 {
6677 	int error;
6678 	off_t olen;
6679 
6680 	ASSERT_VOP_ELOCKED(vp, "nfsvno_deallocate vp");
6681 	/*
6682 	 * Attempt to deallocate on a DS file. A return of ENOENT implies
6683 	 * there is no DS file to deallocate on.
6684 	 */
6685 	error = nfsrv_proxyds(vp, off, 0, cred, p, NFSPROC_DEALLOCATE, NULL,
6686 	    NULL, NULL, NULL, NULL, &len, 0, NULL);
6687 	if (error != ENOENT)
6688 		return (error);
6689 
6690 	/*
6691 	 * Do the actual VOP_DEALLOCATE(), looping so long as
6692 	 * progress is being made, to achieve completion.
6693 	 */
6694 	do {
6695 		olen = len;
6696 		error = VOP_DEALLOCATE(vp, &off, &len, 0, IO_SYNC, cred);
6697 		if (error == 0 && len > 0 && olen > len)
6698 			maybe_yield();
6699 	} while (error == 0 && len > 0 && olen > len);
6700 	if (error == 0 && len > 0)
6701 		error = NFSERR_IO;
6702 	NFSEXITCODE(error);
6703 	return (error);
6704 }
6705 
6706 /*
6707  * Get Extended Atribute vnode op into an mbuf list.
6708  */
6709 int
6710 nfsvno_getxattr(struct vnode *vp, char *name, uint32_t maxresp,
6711     struct ucred *cred, uint64_t flag, int maxextsiz, struct thread *p,
6712     struct mbuf **mpp, struct mbuf **mpendp, int *lenp)
6713 {
6714 	struct iovec *iv;
6715 	struct uio io, *uiop = &io;
6716 	struct mbuf *m, *m2;
6717 	int alen, error, len, tlen;
6718 	size_t siz;
6719 
6720 	/* First, find out the size of the extended attribute. */
6721 	error = VOP_GETEXTATTR(vp, EXTATTR_NAMESPACE_USER, name, NULL,
6722 	    &siz, cred, p);
6723 	if (error != 0)
6724 		return (NFSERR_NOXATTR);
6725 	if (siz > maxresp - NFS_MAXXDR)
6726 		return (NFSERR_XATTR2BIG);
6727 	len = siz;
6728 	tlen = NFSM_RNDUP(len);
6729 	if (tlen > 0) {
6730 		/*
6731 		 * If cnt > MCLBYTES and the reply will not be saved, use
6732 		 * ext_pgs mbufs for TLS.
6733 		 * For NFSv4.0, we do not know for sure if the reply will
6734 		 * be saved, so do not use ext_pgs mbufs for NFSv4.0.
6735 		 * Always use ext_pgs mbufs if ND_EXTPG is set.
6736 		 */
6737 		if ((flag & ND_EXTPG) != 0 || (tlen > MCLBYTES &&
6738 		    (flag & (ND_TLS | ND_SAVEREPLY)) == ND_TLS &&
6739 		    (flag & (ND_NFSV4 | ND_NFSV41)) != ND_NFSV4))
6740 			uiop->uio_iovcnt = nfsrv_createiovec_extpgs(tlen,
6741 			    maxextsiz, &m, &m2, &iv);
6742 		else
6743 			uiop->uio_iovcnt = nfsrv_createiovec(tlen, &m, &m2,
6744 			    &iv);
6745 		uiop->uio_iov = iv;
6746 	} else {
6747 		uiop->uio_iovcnt = 0;
6748 		uiop->uio_iov = iv = NULL;
6749 		m = m2 = NULL;
6750 	}
6751 	uiop->uio_offset = 0;
6752 	uiop->uio_resid = tlen;
6753 	uiop->uio_rw = UIO_READ;
6754 	uiop->uio_segflg = UIO_SYSSPACE;
6755 	uiop->uio_td = p;
6756 #ifdef MAC
6757 	error = mac_vnode_check_getextattr(cred, vp, EXTATTR_NAMESPACE_USER,
6758 	    name);
6759 	if (error != 0)
6760 		goto out;
6761 #endif
6762 
6763 	if (tlen > 0)
6764 		error = VOP_GETEXTATTR(vp, EXTATTR_NAMESPACE_USER, name, uiop,
6765 		    NULL, cred, p);
6766 	if (error != 0)
6767 		goto out;
6768 	if (uiop->uio_resid > 0) {
6769 		alen = tlen;
6770 		len = tlen - uiop->uio_resid;
6771 		tlen = NFSM_RNDUP(len);
6772 		if (alen != tlen)
6773 			printf("nfsvno_getxattr: weird size read\n");
6774 		if (tlen == 0) {
6775 			m_freem(m);
6776 			m = m2 = NULL;
6777 		} else if (alen != tlen || tlen != len)
6778 			m2 = nfsrv_adj(m, alen - tlen, tlen - len);
6779 	}
6780 	*lenp = len;
6781 	*mpp = m;
6782 	*mpendp = m2;
6783 
6784 out:
6785 	if (error != 0) {
6786 		if (m != NULL)
6787 			m_freem(m);
6788 		*lenp = 0;
6789 	}
6790 	free(iv, M_TEMP);
6791 	NFSEXITCODE(error);
6792 	return (error);
6793 }
6794 
6795 /*
6796  * Set Extended attribute vnode op from an mbuf list.
6797  */
6798 int
6799 nfsvno_setxattr(struct vnode *vp, char *name, int len, struct mbuf *m,
6800     char *cp, struct ucred *cred, struct thread *p)
6801 {
6802 	struct iovec *iv;
6803 	struct uio uio, *uiop = &uio;
6804 	int cnt, error;
6805 
6806 	error = 0;
6807 #ifdef MAC
6808 	error = mac_vnode_check_setextattr(cred, vp, EXTATTR_NAMESPACE_USER,
6809 	    name);
6810 #endif
6811 	if (error != 0)
6812 		goto out;
6813 
6814 	uiop->uio_rw = UIO_WRITE;
6815 	uiop->uio_segflg = UIO_SYSSPACE;
6816 	uiop->uio_td = p;
6817 	uiop->uio_offset = 0;
6818 	uiop->uio_resid = len;
6819 	if (len > 0) {
6820 		error = nfsrv_createiovecw(len, m, cp, &iv, &cnt);
6821 		uiop->uio_iov = iv;
6822 		uiop->uio_iovcnt = cnt;
6823 	} else {
6824 		uiop->uio_iov = iv = NULL;
6825 		uiop->uio_iovcnt = 0;
6826 	}
6827 	if (error == 0) {
6828 		error = VOP_SETEXTATTR(vp, EXTATTR_NAMESPACE_USER, name, uiop,
6829 		    cred, p);
6830 		if (error == 0) {
6831 			if (vp->v_type == VREG && nfsrv_devidcnt != 0)
6832 				nfsvno_updateds(vp, cred, p);
6833 			error = VOP_FSYNC(vp, MNT_WAIT, p);
6834 		}
6835 		free(iv, M_TEMP);
6836 	}
6837 
6838 out:
6839 	NFSEXITCODE(error);
6840 	return (error);
6841 }
6842 
6843 /*
6844  * For a pNFS server, the DS file's ctime and
6845  * va_filerev (TimeMetadata and Change) needs to
6846  * be updated.  This is a hack, but works by
6847  * flipping the S_ISGID bit in va_mode and then
6848  * flipping it back.
6849  * It does result in two MDS->DS RPCs, but creating
6850  * a custom RPC just to do this seems overkill, since
6851  * Setxattr/Rmxattr will not be done that frequently.
6852  * If it fails part way through, that is not too
6853  * serious, since the DS file is never executed.
6854  */
6855 static void
6856 nfsvno_updateds(struct vnode *vp, struct ucred *cred, NFSPROC_T *p)
6857 {
6858 	struct nfsvattr nva;
6859 	int ret;
6860 	u_short tmode;
6861 
6862 	ret = VOP_GETATTR(vp, &nva.na_vattr, cred);
6863 	if (ret == 0) {
6864 		tmode = nva.na_mode;
6865 		NFSVNO_ATTRINIT(&nva);
6866 		tmode ^= S_ISGID;
6867 		NFSVNO_SETATTRVAL(&nva, mode, tmode);
6868 		ret = nfsrv_proxyds(vp, 0, 0, cred, p,
6869 		    NFSPROC_SETATTR, NULL, NULL, NULL, &nva,
6870 		    NULL, NULL, 0, NULL);
6871 		if (ret == 0) {
6872 			tmode ^= S_ISGID;
6873 			NFSVNO_SETATTRVAL(&nva, mode, tmode);
6874 			ret = nfsrv_proxyds(vp, 0, 0, cred, p,
6875 			    NFSPROC_SETATTR, NULL, NULL, NULL,
6876 			    &nva, NULL, NULL, 0, NULL);
6877 		}
6878 	}
6879 }
6880 
6881 /*
6882  * Remove Extended attribute vnode op.
6883  */
6884 int
6885 nfsvno_rmxattr(struct nfsrv_descript *nd, struct vnode *vp, char *name,
6886     struct ucred *cred, struct thread *p)
6887 {
6888 	int error;
6889 
6890 	/*
6891 	 * Get rid of any delegations.  I am not sure why this is required,
6892 	 * but RFC-8276 says so.
6893 	 */
6894 	error = nfsrv_checkremove(vp, 0, nd, nd->nd_clientid, p);
6895 	if (error != 0)
6896 		goto out;
6897 #ifdef MAC
6898 	error = mac_vnode_check_deleteextattr(cred, vp, EXTATTR_NAMESPACE_USER,
6899 	    name);
6900 	if (error != 0)
6901 		goto out;
6902 #endif
6903 
6904 	error = VOP_DELETEEXTATTR(vp, EXTATTR_NAMESPACE_USER, name, cred, p);
6905 	if (error == EOPNOTSUPP)
6906 		error = VOP_SETEXTATTR(vp, EXTATTR_NAMESPACE_USER, name, NULL,
6907 		    cred, p);
6908 	if (error == 0) {
6909 		if (vp->v_type == VREG && nfsrv_devidcnt != 0)
6910 			nfsvno_updateds(vp, cred, p);
6911 		error = VOP_FSYNC(vp, MNT_WAIT, p);
6912 	}
6913 out:
6914 	NFSEXITCODE(error);
6915 	return (error);
6916 }
6917 
6918 /*
6919  * List Extended Atribute vnode op into an mbuf list.
6920  */
6921 int
6922 nfsvno_listxattr(struct vnode *vp, uint64_t cookie, struct ucred *cred,
6923     struct thread *p, u_char **bufp, uint32_t *lenp, bool *eofp)
6924 {
6925 	struct iovec iv;
6926 	struct uio io;
6927 	int error;
6928 	size_t siz;
6929 
6930 	*bufp = NULL;
6931 	/* First, find out the size of the extended attribute. */
6932 	error = VOP_LISTEXTATTR(vp, EXTATTR_NAMESPACE_USER, NULL, &siz, cred,
6933 	    p);
6934 	if (error != 0)
6935 		return (NFSERR_NOXATTR);
6936 	if (siz <= cookie) {
6937 		*lenp = 0;
6938 		*eofp = true;
6939 		goto out;
6940 	}
6941 	if (siz > cookie + *lenp) {
6942 		siz = cookie + *lenp;
6943 		*eofp = false;
6944 	} else
6945 		*eofp = true;
6946 	/* Just choose a sanity limit of 10Mbytes for malloc(M_TEMP). */
6947 	if (siz > 10 * 1024 * 1024) {
6948 		error = NFSERR_XATTR2BIG;
6949 		goto out;
6950 	}
6951 	*bufp = malloc(siz, M_TEMP, M_WAITOK);
6952 	iv.iov_base = *bufp;
6953 	iv.iov_len = siz;
6954 	io.uio_iovcnt = 1;
6955 	io.uio_iov = &iv;
6956 	io.uio_offset = 0;
6957 	io.uio_resid = siz;
6958 	io.uio_rw = UIO_READ;
6959 	io.uio_segflg = UIO_SYSSPACE;
6960 	io.uio_td = p;
6961 #ifdef MAC
6962 	error = mac_vnode_check_listextattr(cred, vp, EXTATTR_NAMESPACE_USER);
6963 	if (error != 0)
6964 		goto out;
6965 #endif
6966 
6967 	error = VOP_LISTEXTATTR(vp, EXTATTR_NAMESPACE_USER, &io, NULL, cred,
6968 	    p);
6969 	if (error != 0)
6970 		goto out;
6971 	if (io.uio_resid > 0)
6972 		siz -= io.uio_resid;
6973 	*lenp = siz;
6974 
6975 out:
6976 	if (error != 0) {
6977 		free(*bufp, M_TEMP);
6978 		*bufp = NULL;
6979 	}
6980 	NFSEXITCODE(error);
6981 	return (error);
6982 }
6983 
6984 /*
6985  * Trim trailing data off the mbuf list being built.
6986  */
6987 void
6988 nfsm_trimtrailing(struct nfsrv_descript *nd, struct mbuf *mb, char *bpos,
6989     int bextpg, int bextpgsiz)
6990 {
6991 	vm_page_t pg;
6992 	int fullpgsiz, i;
6993 
6994 	if (mb->m_next != NULL) {
6995 		m_freem(mb->m_next);
6996 		mb->m_next = NULL;
6997 	}
6998 	if ((mb->m_flags & M_EXTPG) != 0) {
6999 		KASSERT(bextpg >= 0 && bextpg < mb->m_epg_npgs,
7000 		    ("nfsm_trimtrailing: bextpg out of range"));
7001 		KASSERT(bpos == (char *)(void *)
7002 		    PHYS_TO_DMAP(mb->m_epg_pa[bextpg]) + PAGE_SIZE - bextpgsiz,
7003 		    ("nfsm_trimtrailing: bextpgsiz bad!"));
7004 
7005 		/* First, get rid of any pages after this position. */
7006 		for (i = mb->m_epg_npgs - 1; i > bextpg; i--) {
7007 			pg = PHYS_TO_VM_PAGE(mb->m_epg_pa[i]);
7008 			vm_page_unwire_noq(pg);
7009 			vm_page_free(pg);
7010 		}
7011 		mb->m_epg_npgs = bextpg + 1;
7012 		if (bextpg == 0)
7013 			fullpgsiz = PAGE_SIZE - mb->m_epg_1st_off;
7014 		else
7015 			fullpgsiz = PAGE_SIZE;
7016 		mb->m_epg_last_len = fullpgsiz - bextpgsiz;
7017 		mb->m_len = m_epg_pagelen(mb, 0, mb->m_epg_1st_off);
7018 		for (i = 1; i < mb->m_epg_npgs; i++)
7019 			mb->m_len += m_epg_pagelen(mb, i, 0);
7020 		nd->nd_bextpgsiz = bextpgsiz;
7021 		nd->nd_bextpg = bextpg;
7022 	} else
7023 		mb->m_len = bpos - mtod(mb, char *);
7024 	nd->nd_mb = mb;
7025 	nd->nd_bpos = bpos;
7026 }
7027 
7028 
7029 /*
7030  * Check to see if a put file handle operation should test for
7031  * NFSERR_WRONGSEC, although NFSv3 actually returns NFSERR_AUTHERR.
7032  * When Open is the next operation, NFSERR_WRONGSEC cannot be
7033  * replied for the Open cases that use a component.  This can
7034  * be identified by the fact that the file handle's type is VDIR.
7035  */
7036 bool
7037 nfsrv_checkwrongsec(struct nfsrv_descript *nd, int nextop, enum vtype vtyp)
7038 {
7039 
7040 	if ((nd->nd_flag & ND_NFSV4) == 0)
7041 		return (true);
7042 
7043 	if ((nd->nd_flag & ND_LASTOP) != 0)
7044 		return (false);
7045 
7046 	if (nextop == NFSV4OP_PUTROOTFH || nextop == NFSV4OP_PUTFH ||
7047 	    nextop == NFSV4OP_PUTPUBFH || nextop == NFSV4OP_RESTOREFH ||
7048 	    nextop == NFSV4OP_LOOKUP || nextop == NFSV4OP_LOOKUPP ||
7049 	    nextop == NFSV4OP_SECINFO || nextop == NFSV4OP_SECINFONONAME)
7050 		return (false);
7051 	if (nextop == NFSV4OP_OPEN && vtyp == VDIR)
7052 		return (false);
7053 	return (true);
7054 }
7055 
7056 /*
7057  * Check DSs marked no space.
7058  */
7059 void
7060 nfsrv_checknospc(void)
7061 {
7062 	struct statfs *tsf;
7063 	struct nfsdevice *ds;
7064 	struct vnode **dvpp, **tdvpp, *dvp;
7065 	char *devid, *tdevid;
7066 	int cnt, error = 0, i;
7067 
7068 	if (nfsrv_devidcnt <= 0)
7069 		return;
7070 	dvpp = mallocarray(nfsrv_devidcnt, sizeof(*dvpp), M_TEMP, M_WAITOK);
7071 	devid = malloc(nfsrv_devidcnt * NFSX_V4DEVICEID, M_TEMP, M_WAITOK);
7072 	tsf = malloc(sizeof(*tsf), M_TEMP, M_WAITOK);
7073 
7074 	/* Get an array of the dvps for the DSs. */
7075 	tdvpp = dvpp;
7076 	tdevid = devid;
7077 	i = 0;
7078 	NFSDDSLOCK();
7079 	/* First, search for matches for same file system. */
7080 	TAILQ_FOREACH(ds, &nfsrv_devidhead, nfsdev_list) {
7081 		if (ds->nfsdev_nmp != NULL && ds->nfsdev_nospc) {
7082 			if (++i > nfsrv_devidcnt)
7083 				break;
7084 			*tdvpp++ = ds->nfsdev_dvp;
7085 			NFSBCOPY(ds->nfsdev_deviceid, tdevid, NFSX_V4DEVICEID);
7086 			tdevid += NFSX_V4DEVICEID;
7087 		}
7088 	}
7089 	NFSDDSUNLOCK();
7090 
7091 	/* Do a VFS_STATFS() for each of the DSs and clear no space. */
7092 	cnt = i;
7093 	tdvpp = dvpp;
7094 	tdevid = devid;
7095 	for (i = 0; i < cnt && error == 0; i++) {
7096 		dvp = *tdvpp++;
7097 		error = VFS_STATFS(dvp->v_mount, tsf);
7098 		if (error == 0 && tsf->f_bavail > 0) {
7099 			NFSD_DEBUG(1, "nfsrv_checknospc: reset nospc\n");
7100 			nfsrv_marknospc(tdevid, false);
7101 		}
7102 		tdevid += NFSX_V4DEVICEID;
7103 	}
7104 	free(tsf, M_TEMP);
7105 	free(dvpp, M_TEMP);
7106 	free(devid, M_TEMP);
7107 }
7108 
7109 /*
7110  * Initialize everything that needs to be initialized for a vnet.
7111  */
7112 static void
7113 nfsrv_vnetinit(const void *unused __unused)
7114 {
7115 
7116 	nfsd_mntinit();
7117 }
7118 VNET_SYSINIT(nfsrv_vnetinit, SI_SUB_VNET_DONE, SI_ORDER_ANY,
7119     nfsrv_vnetinit, NULL);
7120 
7121 /*
7122  * Clean up everything that is in a vnet and needs to be
7123  * done when the jail is destroyed or the module unloaded.
7124  */
7125 static void
7126 nfsrv_cleanup(const void *unused __unused)
7127 {
7128 	int i;
7129 
7130 	NFSD_LOCK();
7131 	if (!NFSD_VNET(nfsrv_mntinited)) {
7132 		NFSD_UNLOCK();
7133 		return;
7134 	}
7135 	NFSD_VNET(nfsrv_mntinited) = false;
7136 	NFSD_UNLOCK();
7137 
7138 	/* Clean out all NFSv4 state. */
7139 	nfsrv_throwawayallstate(curthread);
7140 
7141 	/* Clean the NFS server reply cache */
7142 	nfsrvd_cleancache();
7143 
7144 	/* Clean out v4root exports. */
7145 	if (NFSD_VNET(nfsv4root_mnt)->mnt_export != NULL) {
7146 		vfs_free_addrlist(NFSD_VNET(nfsv4root_mnt)->mnt_export);
7147 		free(NFSD_VNET(nfsv4root_mnt)->mnt_export, M_MOUNT);
7148 		NFSD_VNET(nfsv4root_mnt)->mnt_export = NULL;
7149 	}
7150 
7151 	/* Free up the krpc server pool. */
7152 	if (NFSD_VNET(nfsrvd_pool) != NULL)
7153 		svcpool_destroy(NFSD_VNET(nfsrvd_pool));
7154 
7155 	/* and get rid of the locks */
7156 	for (i = 0; i < NFSRVCACHE_HASHSIZE; i++) {
7157 		mtx_destroy(&NFSD_VNET(nfsrchash_table)[i].mtx);
7158 		mtx_destroy(&NFSD_VNET(nfsrcahash_table)[i].mtx);
7159 	}
7160 	mtx_destroy(&NFSD_VNET(nfsv4root_mnt)->mnt_mtx);
7161 	for (i = 0; i < nfsrv_sessionhashsize; i++)
7162 		mtx_destroy(&NFSD_VNET(nfssessionhash)[i].mtx);
7163 	lockdestroy(&NFSD_VNET(nfsv4root_mnt)->mnt_explock);
7164 	free(NFSD_VNET(nfsrvudphashtbl), M_NFSRVCACHE);
7165 	free(NFSD_VNET(nfsrchash_table), M_NFSRVCACHE);
7166 	free(NFSD_VNET(nfsrcahash_table), M_NFSRVCACHE);
7167 	free(NFSD_VNET(nfsclienthash), M_NFSDCLIENT);
7168 	free(NFSD_VNET(nfslockhash), M_NFSDLOCKFILE);
7169 	free(NFSD_VNET(nfssessionhash), M_NFSDSESSION);
7170 	free(NFSD_VNET(nfsv4root_mnt), M_TEMP);
7171 	NFSD_VNET(nfsv4root_mnt) = NULL;
7172 }
7173 VNET_SYSUNINIT(nfsrv_cleanup, SI_SUB_VNET_DONE, SI_ORDER_ANY,
7174     nfsrv_cleanup, NULL);
7175 
7176 extern int (*nfsd_call_nfsd)(struct thread *, struct nfssvc_args *);
7177 
7178 /*
7179  * Called once to initialize data structures...
7180  */
7181 static int
7182 nfsd_modevent(module_t mod, int type, void *data)
7183 {
7184 	int error = 0, i;
7185 	static int loaded = 0;
7186 
7187 	switch (type) {
7188 	case MOD_LOAD:
7189 		if (loaded)
7190 			goto out;
7191 		newnfs_portinit();
7192 		mtx_init(&nfsrc_udpmtx, "nfsuc", NULL, MTX_DEF);
7193 		mtx_init(&nfs_v4root_mutex, "nfs4rt", NULL, MTX_DEF);
7194 		mtx_init(&nfsrv_dontlistlock_mtx, "nfs4dnl", NULL, MTX_DEF);
7195 		mtx_init(&nfsrv_recalllock_mtx, "nfs4rec", NULL, MTX_DEF);
7196 #ifdef VV_DISABLEDELEG
7197 		vn_deleg_ops.vndeleg_recall = nfsd_recalldelegation;
7198 		vn_deleg_ops.vndeleg_disable = nfsd_disabledelegation;
7199 #endif
7200 		nfsd_call_nfsd = nfssvc_nfsd;
7201 		loaded = 1;
7202 		break;
7203 
7204 	case MOD_UNLOAD:
7205 		if (newnfs_numnfsd != 0) {
7206 			error = EBUSY;
7207 			break;
7208 		}
7209 
7210 #ifdef VV_DISABLEDELEG
7211 		vn_deleg_ops.vndeleg_recall = NULL;
7212 		vn_deleg_ops.vndeleg_disable = NULL;
7213 #endif
7214 		nfsd_call_nfsd = NULL;
7215 		mtx_destroy(&nfsrc_udpmtx);
7216 		mtx_destroy(&nfs_v4root_mutex);
7217 		mtx_destroy(&nfsrv_dontlistlock_mtx);
7218 		mtx_destroy(&nfsrv_recalllock_mtx);
7219 		if (nfslayouthash != NULL) {
7220 			for (i = 0; i < nfsrv_layouthashsize; i++)
7221 				mtx_destroy(&nfslayouthash[i].mtx);
7222 			free(nfslayouthash, M_NFSDSESSION);
7223 		}
7224 		loaded = 0;
7225 		break;
7226 	default:
7227 		error = EOPNOTSUPP;
7228 		break;
7229 	}
7230 
7231 out:
7232 	NFSEXITCODE(error);
7233 	return (error);
7234 }
7235 static moduledata_t nfsd_mod = {
7236 	"nfsd",
7237 	nfsd_modevent,
7238 	NULL,
7239 };
7240 DECLARE_MODULE(nfsd, nfsd_mod, SI_SUB_VFS, SI_ORDER_ANY);
7241 
7242 /* So that loader and kldload(2) can find us, wherever we are.. */
7243 MODULE_VERSION(nfsd, 1);
7244 MODULE_DEPEND(nfsd, nfscommon, 1, 1, 1);
7245 MODULE_DEPEND(nfsd, nfslockd, 1, 1, 1);
7246 MODULE_DEPEND(nfsd, krpc, 1, 1, 1);
7247 MODULE_DEPEND(nfsd, nfssvc, 1, 1, 1);
7248