xref: /freebsd/sys/fs/nfsserver/nfs_nfsdport.c (revision 2d49ff1e21a74b8f641752b63a59dc3ef9e38ddc)
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 #include <sys/capsicum.h>
38 #include <sys/extattr.h>
39 
40 /*
41  * Functions that perform the vfs operations required by the routines in
42  * nfsd_serv.c. It is hoped that this change will make the server more
43  * portable.
44  */
45 
46 #include <fs/nfs/nfsport.h>
47 #include <security/mac/mac_framework.h>
48 #include <sys/callout.h>
49 #include <sys/filio.h>
50 #include <sys/hash.h>
51 #include <sys/osd.h>
52 #include <sys/sysctl.h>
53 #include <nlm/nlm_prot.h>
54 #include <nlm/nlm.h>
55 
56 FEATURE(nfsd, "NFSv4 server");
57 
58 extern u_int32_t newnfs_true, newnfs_false, newnfs_xdrneg1;
59 extern int nfsrv_useacl;
60 extern int newnfs_numnfsd;
61 extern int nfsrv_sessionhashsize;
62 extern struct nfslayouthash *nfslayouthash;
63 extern int nfsrv_layouthashsize;
64 extern struct mtx nfsrv_dslock_mtx;
65 extern int nfs_pnfsiothreads;
66 extern volatile int nfsrv_devidcnt;
67 extern int nfsrv_maxpnfsmirror;
68 extern uint32_t nfs_srvmaxio;
69 extern int nfs_bufpackets;
70 extern u_long sb_max_adj;
71 
72 NFSD_VNET_DECLARE(int, nfsrv_numnfsd);
73 NFSD_VNET_DECLARE(struct nfsrv_stablefirst, nfsrv_stablefirst);
74 NFSD_VNET_DECLARE(SVCPOOL *, nfsrvd_pool);
75 NFSD_VNET_DECLARE(struct nfsclienthashhead *, nfsclienthash);
76 NFSD_VNET_DECLARE(struct nfslockhashhead *, nfslockhash);
77 NFSD_VNET_DECLARE(struct nfssessionhash *, nfssessionhash);
78 NFSD_VNET_DECLARE(struct nfsv4lock, nfsd_suspend_lock);
79 NFSD_VNET_DECLARE(struct nfsstatsv1 *, nfsstatsv1_p);
80 
81 NFSDLOCKMUTEX;
82 NFSSTATESPINLOCK;
83 struct mtx nfsrc_udpmtx;
84 struct mtx nfs_v4root_mutex;
85 struct mtx nfsrv_dontlistlock_mtx;
86 struct mtx nfsrv_recalllock_mtx;
87 struct nfsrvfh nfs_pubfh;
88 int nfs_pubfhset = 0;
89 int nfsd_debuglevel = 0;
90 static pid_t nfsd_master_pid = (pid_t)-1;
91 static char nfsd_master_comm[MAXCOMLEN + 1];
92 static struct timeval nfsd_master_start;
93 static uint32_t nfsv4_sysid = 0;
94 static fhandle_t zerofh;
95 
96 NFSD_VNET_DEFINE(struct proc *, nfsd_master_proc) = NULL;
97 NFSD_VNET_DEFINE(struct nfsrvhashhead *, nfsrvudphashtbl);
98 NFSD_VNET_DEFINE(struct nfsrchash_bucket *, nfsrchash_table);
99 NFSD_VNET_DEFINE(struct nfsrchash_bucket *, nfsrcahash_table);
100 NFSD_VNET_DEFINE(struct nfsrvfh, nfs_rootfh);
101 NFSD_VNET_DEFINE(int, nfs_rootfhset) = 0;
102 NFSD_VNET_DEFINE(struct callout, nfsd_callout);
103 NFSD_VNET_DEFINE_STATIC(struct mount *, nfsv4root_mnt);
104 NFSD_VNET_DEFINE_STATIC(struct vfsoptlist, nfsv4root_opt);
105 NFSD_VNET_DEFINE_STATIC(struct vfsoptlist, nfsv4root_newopt);
106 NFSD_VNET_DEFINE_STATIC(bool, nfsrv_suspend_nfsd) = false;
107 NFSD_VNET_DEFINE_STATIC(bool, nfsrv_mntinited) = false;
108 
109 static int nfssvc_srvcall(struct thread *, struct nfssvc_args *,
110     struct ucred *);
111 static void nfsvno_updateds(struct vnode *, struct ucred *, struct thread *);
112 
113 int nfsrv_enable_crossmntpt = 1;
114 static int nfs_commit_blks;
115 static int nfs_commit_miss;
116 extern int nfsrv_issuedelegs;
117 extern int nfsrv_dolocallocks;
118 extern struct nfsdevicehead nfsrv_devidhead;
119 
120 /* Map d_type to vnode type. */
121 static uint8_t dtype_to_vnode[DT_WHT + 1] = { VNON, VFIFO, VCHR, VNON, VDIR,
122     VNON, VBLK, VNON, VREG, VNON, VLNK, VNON, VSOCK, VNON, VNON };
123 #define	NFS_DTYPETOVTYPE(t)	((t) <= DT_WHT ? dtype_to_vnode[(t)] : VNON)
124 
125 static int nfsrv_createiovec(int, struct mbuf **, struct mbuf **,
126     struct iovec **);
127 static int nfsrv_createiovec_extpgs(int, int, struct mbuf **,
128     struct mbuf **, struct iovec **);
129 static int nfsrv_createiovecw(int, struct mbuf *, char *, struct iovec **,
130     int *);
131 static void nfsrv_pnfscreate(struct vnode *, struct vattr *, struct ucred *,
132     NFSPROC_T *);
133 static void nfsrv_pnfsremovesetup(struct vnode *, NFSPROC_T *, struct vnode **,
134     int *, char *, fhandle_t *);
135 static void nfsrv_pnfsremove(struct vnode **, int, char *, fhandle_t *,
136     NFSPROC_T *);
137 static int nfsrv_proxyds(struct vnode *, off_t, int, struct ucred *,
138     struct thread *, int, struct mbuf **, char *, struct mbuf **,
139     struct nfsvattr *, struct acl *, off_t *, int, bool *);
140 static int nfsrv_setextattr(struct vnode *, struct nfsvattr *, NFSPROC_T *);
141 static int nfsrv_readdsrpc(fhandle_t *, off_t, int, struct ucred *,
142     NFSPROC_T *, struct nfsmount *, struct mbuf **, struct mbuf **);
143 static int nfsrv_writedsrpc(fhandle_t *, off_t, int, struct ucred *,
144     NFSPROC_T *, struct vnode *, struct nfsmount **, int, struct mbuf **,
145     char *, int *);
146 static int nfsrv_allocatedsrpc(fhandle_t *, off_t, off_t, struct ucred *,
147     NFSPROC_T *, struct vnode *, struct nfsmount **, int, int *);
148 static int nfsrv_deallocatedsrpc(fhandle_t *, off_t, off_t, struct ucred *,
149     NFSPROC_T *, struct vnode *, struct nfsmount **, int, int *);
150 static int nfsrv_setacldsrpc(fhandle_t *, struct ucred *, NFSPROC_T *,
151     struct vnode *, struct nfsmount **, int, struct acl *, int *);
152 static int nfsrv_setattrdsrpc(fhandle_t *, struct ucred *, NFSPROC_T *,
153     struct vnode *, struct nfsmount **, int, struct nfsvattr *, int *);
154 static int nfsrv_getattrdsrpc(fhandle_t *, struct ucred *, NFSPROC_T *,
155     struct vnode *, struct nfsmount *, struct nfsvattr *);
156 static int nfsrv_seekdsrpc(fhandle_t *, off_t *, int, bool *, struct ucred *,
157     NFSPROC_T *, struct nfsmount *);
158 static int nfsrv_putfhname(fhandle_t *, char *);
159 static int nfsrv_pnfslookupds(struct vnode *, struct vnode *,
160     struct pnfsdsfile *, struct vnode **, NFSPROC_T *);
161 static void nfsrv_pnfssetfh(struct vnode *, struct pnfsdsfile *, char *, char *,
162     struct vnode *, NFSPROC_T *);
163 static int nfsrv_dsremove(struct vnode *, char *, struct ucred *, NFSPROC_T *);
164 static int nfsrv_dssetacl(struct vnode *, struct acl *, struct ucred *,
165     NFSPROC_T *);
166 static int nfsrv_pnfsstatfs(struct statfs *, struct mount *);
167 
168 int nfs_pnfsio(task_fn_t *, void *);
169 
170 SYSCTL_NODE(_vfs, OID_AUTO, nfsd, CTLFLAG_RW | CTLFLAG_MPSAFE, 0,
171     "NFS server");
172 SYSCTL_INT(_vfs_nfsd, OID_AUTO, mirrormnt, CTLFLAG_RW,
173     &nfsrv_enable_crossmntpt, 0, "Enable nfsd to cross mount points");
174 SYSCTL_INT(_vfs_nfsd, OID_AUTO, commit_blks, CTLFLAG_RW, &nfs_commit_blks,
175     0, "");
176 SYSCTL_INT(_vfs_nfsd, OID_AUTO, commit_miss, CTLFLAG_RW, &nfs_commit_miss,
177     0, "");
178 SYSCTL_INT(_vfs_nfsd, OID_AUTO, issue_delegations, CTLFLAG_RW,
179     &nfsrv_issuedelegs, 0, "Enable nfsd to issue delegations");
180 SYSCTL_INT(_vfs_nfsd, OID_AUTO, enable_locallocks, CTLFLAG_RW,
181     &nfsrv_dolocallocks, 0, "Enable nfsd to acquire local locks on files");
182 SYSCTL_INT(_vfs_nfsd, OID_AUTO, debuglevel, CTLFLAG_RW, &nfsd_debuglevel,
183     0, "Debug level for NFS server");
184 NFSD_VNET_DECLARE(int, nfsd_enable_stringtouid);
185 SYSCTL_INT(_vfs_nfsd, OID_AUTO, enable_stringtouid,
186     CTLFLAG_NFSD_VNET | CTLFLAG_RW, &NFSD_VNET_NAME(nfsd_enable_stringtouid),
187     0, "Enable nfsd to accept numeric owner_names");
188 static int nfsrv_pnfsgetdsattr = 1;
189 SYSCTL_INT(_vfs_nfsd, OID_AUTO, pnfsgetdsattr, CTLFLAG_RW,
190     &nfsrv_pnfsgetdsattr, 0, "When set getattr gets DS attributes via RPC");
191 
192 /*
193  * nfsrv_dsdirsize can only be increased and only when the nfsd threads are
194  * not running.
195  * The dsN subdirectories for the increased values must have been created
196  * on all DS servers before this increase is done.
197  */
198 u_int	nfsrv_dsdirsize = 20;
199 static int
200 sysctl_dsdirsize(SYSCTL_HANDLER_ARGS)
201 {
202 	int error, newdsdirsize;
203 
204 	newdsdirsize = nfsrv_dsdirsize;
205 	error = sysctl_handle_int(oidp, &newdsdirsize, 0, req);
206 	if (error != 0 || req->newptr == NULL)
207 		return (error);
208 	if (newdsdirsize <= nfsrv_dsdirsize || newdsdirsize > 10000 ||
209 	    newnfs_numnfsd != 0)
210 		return (EINVAL);
211 	nfsrv_dsdirsize = newdsdirsize;
212 	return (0);
213 }
214 SYSCTL_PROC(_vfs_nfsd, OID_AUTO, dsdirsize,
215     CTLTYPE_UINT | CTLFLAG_MPSAFE | CTLFLAG_RW, 0, sizeof(nfsrv_dsdirsize),
216     sysctl_dsdirsize, "IU", "Number of dsN subdirs on the DS servers");
217 
218 /*
219  * nfs_srvmaxio can only be increased and only when the nfsd threads are
220  * not running.  The setting must be a power of 2, with the current limit of
221  * 1Mbyte.
222  */
223 static int
224 sysctl_srvmaxio(SYSCTL_HANDLER_ARGS)
225 {
226 	int error;
227 	u_int newsrvmaxio;
228 	uint64_t tval;
229 
230 	newsrvmaxio = nfs_srvmaxio;
231 	error = sysctl_handle_int(oidp, &newsrvmaxio, 0, req);
232 	if (error != 0 || req->newptr == NULL)
233 		return (error);
234 	if (newsrvmaxio == nfs_srvmaxio)
235 		return (0);
236 	if (newsrvmaxio < nfs_srvmaxio) {
237 		printf("nfsd: vfs.nfsd.srvmaxio can only be increased\n");
238 		return (EINVAL);
239 	}
240 	if (newsrvmaxio > 1048576) {
241 		printf("nfsd: vfs.nfsd.srvmaxio cannot be > 1Mbyte\n");
242 		return (EINVAL);
243 	}
244 	if ((newsrvmaxio & (newsrvmaxio - 1)) != 0) {
245 		printf("nfsd: vfs.nfsd.srvmaxio must be a power of 2\n");
246 		return (EINVAL);
247 	}
248 
249 	/*
250 	 * Check that kern.ipc.maxsockbuf is large enough for
251 	 * newsrviomax, given the setting of vfs.nfs.bufpackets.
252 	 */
253 	if ((newsrvmaxio + NFS_MAXXDR) * nfs_bufpackets >
254 	    sb_max_adj) {
255 		/*
256 		 * Suggest vfs.nfs.bufpackets * maximum RPC message for
257 		 * sb_max_adj.
258 		 */
259 		tval = (newsrvmaxio + NFS_MAXXDR) * nfs_bufpackets;
260 
261 		/*
262 		 * Convert suggested sb_max_adj value to a suggested
263 		 * sb_max value, which is what is set via kern.ipc.maxsockbuf.
264 		 * Perform the inverse calculation of (from uipc_sockbuf.c):
265 		 * sb_max_adj = (u_quad_t)sb_max * MCLBYTES /
266 		 *     (MSIZE + MCLBYTES);
267 		 * XXX If the calculation of sb_max_adj from sb_max changes,
268 		 *     this calculation must be changed as well.
269 		 */
270 		tval *= (MSIZE + MCLBYTES);  /* Brackets for readability. */
271 		tval += MCLBYTES - 1;        /* Round up divide. */
272 		tval /= MCLBYTES;
273 		printf("nfsd: set kern.ipc.maxsockbuf to a minimum of "
274 		    "%ju to support %ubyte NFS I/O\n", (uintmax_t)tval,
275 		    newsrvmaxio);
276 		return (EINVAL);
277 	}
278 
279 	NFSD_LOCK();
280 	if (newnfs_numnfsd != 0) {
281 		NFSD_UNLOCK();
282 		printf("nfsd: cannot set vfs.nfsd.srvmaxio when nfsd "
283 		    "threads are running\n");
284 		return (EINVAL);
285 	}
286 
287 
288 	nfs_srvmaxio = newsrvmaxio;
289 	NFSD_UNLOCK();
290 	return (0);
291 }
292 SYSCTL_PROC(_vfs_nfsd, OID_AUTO, srvmaxio,
293     CTLTYPE_UINT | CTLFLAG_MPSAFE | CTLFLAG_RW, NULL, 0,
294     sysctl_srvmaxio, "IU", "Maximum I/O size in bytes");
295 
296 #define	MAX_REORDERED_RPC	16
297 #define	NUM_HEURISTIC		1031
298 #define	NHUSE_INIT		64
299 #define	NHUSE_INC		16
300 #define	NHUSE_MAX		2048
301 
302 static struct nfsheur {
303 	struct vnode *nh_vp;	/* vp to match (unreferenced pointer) */
304 	off_t nh_nextoff;	/* next offset for sequential detection */
305 	int nh_use;		/* use count for selection */
306 	int nh_seqcount;	/* heuristic */
307 } nfsheur[NUM_HEURISTIC];
308 
309 /*
310  * Heuristic to detect sequential operation.
311  */
312 static struct nfsheur *
313 nfsrv_sequential_heuristic(struct uio *uio, struct vnode *vp)
314 {
315 	struct nfsheur *nh;
316 	int hi, try;
317 
318 	/* Locate best candidate. */
319 	try = 32;
320 	hi = ((int)(vm_offset_t)vp / sizeof(struct vnode)) % NUM_HEURISTIC;
321 	nh = &nfsheur[hi];
322 	while (try--) {
323 		if (nfsheur[hi].nh_vp == vp) {
324 			nh = &nfsheur[hi];
325 			break;
326 		}
327 		if (nfsheur[hi].nh_use > 0)
328 			--nfsheur[hi].nh_use;
329 		hi = (hi + 1) % NUM_HEURISTIC;
330 		if (nfsheur[hi].nh_use < nh->nh_use)
331 			nh = &nfsheur[hi];
332 	}
333 
334 	/* Initialize hint if this is a new file. */
335 	if (nh->nh_vp != vp) {
336 		nh->nh_vp = vp;
337 		nh->nh_nextoff = uio->uio_offset;
338 		nh->nh_use = NHUSE_INIT;
339 		if (uio->uio_offset == 0)
340 			nh->nh_seqcount = 4;
341 		else
342 			nh->nh_seqcount = 1;
343 	}
344 
345 	/* Calculate heuristic. */
346 	if ((uio->uio_offset == 0 && nh->nh_seqcount > 0) ||
347 	    uio->uio_offset == nh->nh_nextoff) {
348 		/* See comments in vfs_vnops.c:sequential_heuristic(). */
349 		nh->nh_seqcount += howmany(uio->uio_resid, 16384);
350 		if (nh->nh_seqcount > IO_SEQMAX)
351 			nh->nh_seqcount = IO_SEQMAX;
352 	} else if (qabs(uio->uio_offset - nh->nh_nextoff) <= MAX_REORDERED_RPC *
353 	    imax(vp->v_mount->mnt_stat.f_iosize, uio->uio_resid)) {
354 		/* Probably a reordered RPC, leave seqcount alone. */
355 	} else if (nh->nh_seqcount > 1) {
356 		nh->nh_seqcount /= 2;
357 	} else {
358 		nh->nh_seqcount = 0;
359 	}
360 	nh->nh_use += NHUSE_INC;
361 	if (nh->nh_use > NHUSE_MAX)
362 		nh->nh_use = NHUSE_MAX;
363 	return (nh);
364 }
365 
366 /*
367  * Get attributes into nfsvattr structure.
368  */
369 int
370 nfsvno_getattr(struct vnode *vp, struct nfsvattr *nvap,
371     struct nfsrv_descript *nd, struct thread *p, int vpislocked,
372     nfsattrbit_t *attrbitp)
373 {
374 	int error, gotattr, lockedit = 0;
375 	struct nfsvattr na;
376 
377 	if (vpislocked == 0) {
378 		/*
379 		 * When vpislocked == 0, the vnode is either exclusively
380 		 * locked by this thread or not locked by this thread.
381 		 * As such, shared lock it, if not exclusively locked.
382 		 */
383 		if (NFSVOPISLOCKED(vp) != LK_EXCLUSIVE) {
384 			lockedit = 1;
385 			NFSVOPLOCK(vp, LK_SHARED | LK_RETRY);
386 		}
387 	}
388 
389 	/*
390 	 * Acquire the Change, Size, TimeAccess, TimeModify and SpaceUsed
391 	 * attributes, as required.
392 	 * This needs to be done for regular files if:
393 	 * - non-NFSv4 RPCs or
394 	 * - when attrbitp == NULL or
395 	 * - an NFSv4 RPC with any of the above attributes in attrbitp.
396 	 * A return of 0 for nfsrv_proxyds() indicates that it has acquired
397 	 * these attributes.  nfsrv_proxyds() will return an error if the
398 	 * server is not a pNFS one.
399 	 */
400 	gotattr = 0;
401 	if (vp->v_type == VREG && nfsrv_devidcnt > 0 && (attrbitp == NULL ||
402 	    (nd->nd_flag & ND_NFSV4) == 0 ||
403 	    NFSISSET_ATTRBIT(attrbitp, NFSATTRBIT_CHANGE) ||
404 	    NFSISSET_ATTRBIT(attrbitp, NFSATTRBIT_SIZE) ||
405 	    NFSISSET_ATTRBIT(attrbitp, NFSATTRBIT_TIMEACCESS) ||
406 	    NFSISSET_ATTRBIT(attrbitp, NFSATTRBIT_TIMEMODIFY) ||
407 	    NFSISSET_ATTRBIT(attrbitp, NFSATTRBIT_SPACEUSED))) {
408 		error = nfsrv_proxyds(vp, 0, 0, nd->nd_cred, p,
409 		    NFSPROC_GETATTR, NULL, NULL, NULL, &na, NULL, NULL, 0,
410 		    NULL);
411 		if (error == 0)
412 			gotattr = 1;
413 	}
414 
415 	error = VOP_GETATTR(vp, &nvap->na_vattr, nd->nd_cred);
416 	if (lockedit != 0)
417 		NFSVOPUNLOCK(vp);
418 
419 	/*
420 	 * If we got the Change, Size and Modify Time from the DS,
421 	 * replace them.
422 	 */
423 	if (gotattr != 0) {
424 		nvap->na_atime = na.na_atime;
425 		nvap->na_mtime = na.na_mtime;
426 		nvap->na_filerev = na.na_filerev;
427 		nvap->na_size = na.na_size;
428 		nvap->na_bytes = na.na_bytes;
429 	}
430 	NFSD_DEBUG(4, "nfsvno_getattr: gotattr=%d err=%d chg=%ju\n", gotattr,
431 	    error, (uintmax_t)na.na_filerev);
432 
433 	NFSEXITCODE(error);
434 	return (error);
435 }
436 
437 /*
438  * Get a file handle for a vnode.
439  */
440 int
441 nfsvno_getfh(struct vnode *vp, fhandle_t *fhp, struct thread *p)
442 {
443 	int error;
444 
445 	NFSBZERO((caddr_t)fhp, sizeof(fhandle_t));
446 	fhp->fh_fsid = vp->v_mount->mnt_stat.f_fsid;
447 	error = VOP_VPTOFH(vp, &fhp->fh_fid);
448 
449 	NFSEXITCODE(error);
450 	return (error);
451 }
452 
453 /*
454  * Perform access checking for vnodes obtained from file handles that would
455  * refer to files already opened by a Unix client. You cannot just use
456  * vn_writechk() and VOP_ACCESSX() for two reasons.
457  * 1 - You must check for exported rdonly as well as MNT_RDONLY for the write
458  *     case.
459  * 2 - The owner is to be given access irrespective of mode bits for some
460  *     operations, so that processes that chmod after opening a file don't
461  *     break.
462  */
463 int
464 nfsvno_accchk(struct vnode *vp, accmode_t accmode, struct ucred *cred,
465     struct nfsexstuff *exp, struct thread *p, int override, int vpislocked,
466     u_int32_t *supportedtypep)
467 {
468 	struct vattr vattr;
469 	int error = 0, getret = 0;
470 
471 	if (vpislocked == 0) {
472 		if (NFSVOPLOCK(vp, LK_SHARED) != 0) {
473 			error = EPERM;
474 			goto out;
475 		}
476 	}
477 	if (accmode & VWRITE) {
478 		/* Just vn_writechk() changed to check rdonly */
479 		/*
480 		 * Disallow write attempts on read-only file systems;
481 		 * unless the file is a socket or a block or character
482 		 * device resident on the file system.
483 		 */
484 		if (NFSVNO_EXRDONLY(exp) ||
485 		    (vp->v_mount->mnt_flag & MNT_RDONLY)) {
486 			switch (vp->v_type) {
487 			case VREG:
488 			case VDIR:
489 			case VLNK:
490 				error = EROFS;
491 			default:
492 				break;
493 			}
494 		}
495 		/*
496 		 * If there's shared text associated with
497 		 * the inode, try to free it up once.  If
498 		 * we fail, we can't allow writing.
499 		 */
500 		if (VOP_IS_TEXT(vp) && error == 0)
501 			error = ETXTBSY;
502 	}
503 	if (error != 0) {
504 		if (vpislocked == 0)
505 			NFSVOPUNLOCK(vp);
506 		goto out;
507 	}
508 
509 	/*
510 	 * Should the override still be applied when ACLs are enabled?
511 	 */
512 	error = VOP_ACCESSX(vp, accmode, cred, p);
513 	if (error != 0 && (accmode & (VDELETE | VDELETE_CHILD))) {
514 		/*
515 		 * Try again with VEXPLICIT_DENY, to see if the test for
516 		 * deletion is supported.
517 		 */
518 		error = VOP_ACCESSX(vp, accmode | VEXPLICIT_DENY, cred, p);
519 		if (error == 0) {
520 			if (vp->v_type == VDIR) {
521 				accmode &= ~(VDELETE | VDELETE_CHILD);
522 				accmode |= VWRITE;
523 				error = VOP_ACCESSX(vp, accmode, cred, p);
524 			} else if (supportedtypep != NULL) {
525 				*supportedtypep &= ~NFSACCESS_DELETE;
526 			}
527 		}
528 	}
529 
530 	/*
531 	 * Allow certain operations for the owner (reads and writes
532 	 * on files that are already open).
533 	 */
534 	if (override != NFSACCCHK_NOOVERRIDE &&
535 	    (error == EPERM || error == EACCES)) {
536 		if (cred->cr_uid == 0 && (override & NFSACCCHK_ALLOWROOT))
537 			error = 0;
538 		else if (override & NFSACCCHK_ALLOWOWNER) {
539 			getret = VOP_GETATTR(vp, &vattr, cred);
540 			if (getret == 0 && cred->cr_uid == vattr.va_uid)
541 				error = 0;
542 		}
543 	}
544 	if (vpislocked == 0)
545 		NFSVOPUNLOCK(vp);
546 
547 out:
548 	NFSEXITCODE(error);
549 	return (error);
550 }
551 
552 /*
553  * Set attribute(s) vnop.
554  */
555 int
556 nfsvno_setattr(struct vnode *vp, struct nfsvattr *nvap, struct ucred *cred,
557     struct thread *p, struct nfsexstuff *exp)
558 {
559 	u_quad_t savsize = 0;
560 	int error, savedit;
561 	time_t savbtime;
562 
563 	/*
564 	 * If this is an exported file system and a pNFS service is running,
565 	 * don't VOP_SETATTR() of size for the MDS file system.
566 	 */
567 	savedit = 0;
568 	error = 0;
569 	if (vp->v_type == VREG && (vp->v_mount->mnt_flag & MNT_EXPORTED) != 0 &&
570 	    nfsrv_devidcnt != 0 && nvap->na_vattr.va_size != VNOVAL &&
571 	    nvap->na_vattr.va_size > 0) {
572 		savsize = nvap->na_vattr.va_size;
573 		nvap->na_vattr.va_size = VNOVAL;
574 		if (nvap->na_vattr.va_uid != (uid_t)VNOVAL ||
575 		    nvap->na_vattr.va_gid != (gid_t)VNOVAL ||
576 		    nvap->na_vattr.va_mode != (mode_t)VNOVAL ||
577 		    nvap->na_vattr.va_atime.tv_sec != VNOVAL ||
578 		    nvap->na_vattr.va_mtime.tv_sec != VNOVAL)
579 			savedit = 1;
580 		else
581 			savedit = 2;
582 	}
583 	if (savedit != 2)
584 		error = VOP_SETATTR(vp, &nvap->na_vattr, cred);
585 	if (savedit != 0)
586 		nvap->na_vattr.va_size = savsize;
587 	if (error == 0 && (nvap->na_vattr.va_uid != (uid_t)VNOVAL ||
588 	    nvap->na_vattr.va_gid != (gid_t)VNOVAL ||
589 	    nvap->na_vattr.va_size != VNOVAL ||
590 	    nvap->na_vattr.va_mode != (mode_t)VNOVAL ||
591 	    nvap->na_vattr.va_atime.tv_sec != VNOVAL ||
592 	    nvap->na_vattr.va_mtime.tv_sec != VNOVAL)) {
593 		/* Never modify birthtime on a DS file. */
594 		savbtime = nvap->na_vattr.va_birthtime.tv_sec;
595 		nvap->na_vattr.va_birthtime.tv_sec = VNOVAL;
596 		/* For a pNFS server, set the attributes on the DS file. */
597 		error = nfsrv_proxyds(vp, 0, 0, cred, p, NFSPROC_SETATTR,
598 		    NULL, NULL, NULL, nvap, NULL, NULL, 0, NULL);
599 		nvap->na_vattr.va_birthtime.tv_sec = savbtime;
600 		if (error == ENOENT)
601 			error = 0;
602 	}
603 	NFSEXITCODE(error);
604 	return (error);
605 }
606 
607 /*
608  * Set up nameidata for a lookup() call and do it.
609  */
610 int
611 nfsvno_namei(struct nfsrv_descript *nd, struct nameidata *ndp,
612     struct vnode *dp, int islocked, struct nfsexstuff *exp,
613     struct vnode **retdirp)
614 {
615 	struct componentname *cnp = &ndp->ni_cnd;
616 	int i;
617 	struct iovec aiov;
618 	struct uio auio;
619 	int lockleaf = (cnp->cn_flags & LOCKLEAF) != 0, linklen;
620 	int error = 0;
621 	char *cp;
622 
623 	*retdirp = NULL;
624 	cnp->cn_nameptr = cnp->cn_pnbuf;
625 	ndp->ni_lcf = 0;
626 	/*
627 	 * Extract and set starting directory.
628 	 */
629 	if (dp->v_type != VDIR) {
630 		if (islocked)
631 			vput(dp);
632 		else
633 			vrele(dp);
634 		nfsvno_relpathbuf(ndp);
635 		error = ENOTDIR;
636 		goto out1;
637 	}
638 	if (islocked)
639 		NFSVOPUNLOCK(dp);
640 	VREF(dp);
641 	*retdirp = dp;
642 	if (NFSVNO_EXRDONLY(exp))
643 		cnp->cn_flags |= RDONLY;
644 	ndp->ni_segflg = UIO_SYSSPACE;
645 
646 	if (nd->nd_flag & ND_PUBLOOKUP) {
647 		ndp->ni_loopcnt = 0;
648 		if (cnp->cn_pnbuf[0] == '/') {
649 			vrele(dp);
650 			/*
651 			 * Check for degenerate pathnames here, since lookup()
652 			 * panics on them.
653 			 */
654 			for (i = 1; i < ndp->ni_pathlen; i++)
655 				if (cnp->cn_pnbuf[i] != '/')
656 					break;
657 			if (i == ndp->ni_pathlen) {
658 				error = NFSERR_ACCES;
659 				goto out;
660 			}
661 			dp = rootvnode;
662 			VREF(dp);
663 		}
664 	} else if ((nfsrv_enable_crossmntpt == 0 && NFSVNO_EXPORTED(exp)) ||
665 	    (nd->nd_flag & ND_NFSV4) == 0) {
666 		/*
667 		 * Only cross mount points for NFSv4 when doing a
668 		 * mount while traversing the file system above
669 		 * the mount point, unless nfsrv_enable_crossmntpt is set.
670 		 */
671 		cnp->cn_flags |= NOCROSSMOUNT;
672 	}
673 
674 	/*
675 	 * Initialize for scan, set ni_startdir and bump ref on dp again
676 	 * because lookup() will dereference ni_startdir.
677 	 */
678 
679 	ndp->ni_startdir = dp;
680 	ndp->ni_rootdir = rootvnode;
681 	ndp->ni_topdir = NULL;
682 
683 	if (!lockleaf)
684 		cnp->cn_flags |= LOCKLEAF;
685 	for (;;) {
686 		cnp->cn_nameptr = cnp->cn_pnbuf;
687 		/*
688 		 * Call lookup() to do the real work.  If an error occurs,
689 		 * ndp->ni_vp and ni_dvp are left uninitialized or NULL and
690 		 * we do not have to dereference anything before returning.
691 		 * In either case ni_startdir will be dereferenced and NULLed
692 		 * out.
693 		 */
694 		error = vfs_lookup(ndp);
695 		if (error)
696 			break;
697 
698 		/*
699 		 * Check for encountering a symbolic link.  Trivial
700 		 * termination occurs if no symlink encountered.
701 		 */
702 		if ((cnp->cn_flags & ISSYMLINK) == 0) {
703 			if (ndp->ni_vp && !lockleaf)
704 				NFSVOPUNLOCK(ndp->ni_vp);
705 			break;
706 		}
707 
708 		/*
709 		 * Validate symlink
710 		 */
711 		if ((cnp->cn_flags & LOCKPARENT) && ndp->ni_pathlen == 1)
712 			NFSVOPUNLOCK(ndp->ni_dvp);
713 		if (!(nd->nd_flag & ND_PUBLOOKUP)) {
714 			error = EINVAL;
715 			goto badlink2;
716 		}
717 
718 		if (ndp->ni_loopcnt++ >= MAXSYMLINKS) {
719 			error = ELOOP;
720 			goto badlink2;
721 		}
722 		if (ndp->ni_pathlen > 1)
723 			cp = uma_zalloc(namei_zone, M_WAITOK);
724 		else
725 			cp = cnp->cn_pnbuf;
726 		aiov.iov_base = cp;
727 		aiov.iov_len = MAXPATHLEN;
728 		auio.uio_iov = &aiov;
729 		auio.uio_iovcnt = 1;
730 		auio.uio_offset = 0;
731 		auio.uio_rw = UIO_READ;
732 		auio.uio_segflg = UIO_SYSSPACE;
733 		auio.uio_td = NULL;
734 		auio.uio_resid = MAXPATHLEN;
735 		error = VOP_READLINK(ndp->ni_vp, &auio, cnp->cn_cred);
736 		if (error) {
737 		badlink1:
738 			if (ndp->ni_pathlen > 1)
739 				uma_zfree(namei_zone, cp);
740 		badlink2:
741 			vrele(ndp->ni_dvp);
742 			vput(ndp->ni_vp);
743 			break;
744 		}
745 		linklen = MAXPATHLEN - auio.uio_resid;
746 		if (linklen == 0) {
747 			error = ENOENT;
748 			goto badlink1;
749 		}
750 		if (linklen + ndp->ni_pathlen >= MAXPATHLEN) {
751 			error = ENAMETOOLONG;
752 			goto badlink1;
753 		}
754 
755 		/*
756 		 * Adjust or replace path
757 		 */
758 		if (ndp->ni_pathlen > 1) {
759 			NFSBCOPY(ndp->ni_next, cp + linklen, ndp->ni_pathlen);
760 			uma_zfree(namei_zone, cnp->cn_pnbuf);
761 			cnp->cn_pnbuf = cp;
762 		} else
763 			cnp->cn_pnbuf[linklen] = '\0';
764 		ndp->ni_pathlen += linklen;
765 
766 		/*
767 		 * Cleanup refs for next loop and check if root directory
768 		 * should replace current directory.  Normally ni_dvp
769 		 * becomes the new base directory and is cleaned up when
770 		 * we loop.  Explicitly null pointers after invalidation
771 		 * to clarify operation.
772 		 */
773 		vput(ndp->ni_vp);
774 		ndp->ni_vp = NULL;
775 
776 		if (cnp->cn_pnbuf[0] == '/') {
777 			vrele(ndp->ni_dvp);
778 			ndp->ni_dvp = ndp->ni_rootdir;
779 			VREF(ndp->ni_dvp);
780 		}
781 		ndp->ni_startdir = ndp->ni_dvp;
782 		ndp->ni_dvp = NULL;
783 	}
784 	if (!lockleaf)
785 		cnp->cn_flags &= ~LOCKLEAF;
786 
787 out:
788 	if (error) {
789 		nfsvno_relpathbuf(ndp);
790 		ndp->ni_vp = NULL;
791 		ndp->ni_dvp = NULL;
792 		ndp->ni_startdir = NULL;
793 	} else if ((ndp->ni_cnd.cn_flags & (WANTPARENT|LOCKPARENT)) == 0) {
794 		ndp->ni_dvp = NULL;
795 	}
796 
797 out1:
798 	NFSEXITCODE2(error, nd);
799 	return (error);
800 }
801 
802 /*
803  * Set up a pathname buffer and return a pointer to it and, optionally
804  * set a hash pointer.
805  */
806 void
807 nfsvno_setpathbuf(struct nameidata *ndp, char **bufpp, u_long **hashpp)
808 {
809 	struct componentname *cnp = &ndp->ni_cnd;
810 
811 	cnp->cn_flags |= (NOMACCHECK);
812 	cnp->cn_pnbuf = uma_zalloc(namei_zone, M_WAITOK);
813 	if (hashpp != NULL)
814 		*hashpp = NULL;
815 	*bufpp = cnp->cn_pnbuf;
816 }
817 
818 /*
819  * Release the above path buffer, if not released by nfsvno_namei().
820  */
821 void
822 nfsvno_relpathbuf(struct nameidata *ndp)
823 {
824 
825 	uma_zfree(namei_zone, ndp->ni_cnd.cn_pnbuf);
826 	ndp->ni_cnd.cn_pnbuf = NULL;
827 }
828 
829 /*
830  * Readlink vnode op into an mbuf list.
831  */
832 int
833 nfsvno_readlink(struct vnode *vp, struct ucred *cred, int maxextsiz,
834     struct thread *p, struct mbuf **mpp, struct mbuf **mpendp, int *lenp)
835 {
836 	struct iovec *iv;
837 	struct uio io, *uiop = &io;
838 	struct mbuf *mp, *mp3;
839 	int len, tlen, error = 0;
840 
841 	len = NFS_MAXPATHLEN;
842 	if (maxextsiz > 0)
843 		uiop->uio_iovcnt = nfsrv_createiovec_extpgs(len, maxextsiz,
844 		    &mp3, &mp, &iv);
845 	else
846 		uiop->uio_iovcnt = nfsrv_createiovec(len, &mp3, &mp, &iv);
847 	uiop->uio_iov = iv;
848 	uiop->uio_offset = 0;
849 	uiop->uio_resid = len;
850 	uiop->uio_rw = UIO_READ;
851 	uiop->uio_segflg = UIO_SYSSPACE;
852 	uiop->uio_td = NULL;
853 	error = VOP_READLINK(vp, uiop, cred);
854 	free(iv, M_TEMP);
855 	if (error) {
856 		m_freem(mp3);
857 		*lenp = 0;
858 		goto out;
859 	}
860 	if (uiop->uio_resid > 0) {
861 		len -= uiop->uio_resid;
862 		tlen = NFSM_RNDUP(len);
863 		if (tlen == 0) {
864 			m_freem(mp3);
865 			mp3 = mp = NULL;
866 		} else if (tlen != NFS_MAXPATHLEN || tlen != len)
867 			mp = nfsrv_adj(mp3, NFS_MAXPATHLEN - tlen,
868 			    tlen - len);
869 	}
870 	*lenp = len;
871 	*mpp = mp3;
872 	*mpendp = mp;
873 
874 out:
875 	NFSEXITCODE(error);
876 	return (error);
877 }
878 
879 /*
880  * Create an mbuf chain and an associated iovec that can be used to Read
881  * or Getextattr of data.
882  * Upon success, return pointers to the first and last mbufs in the chain
883  * plus the malloc'd iovec and its iovlen.
884  */
885 static int
886 nfsrv_createiovec(int len, struct mbuf **mpp, struct mbuf **mpendp,
887     struct iovec **ivp)
888 {
889 	struct mbuf *m, *m2 = NULL, *m3;
890 	struct iovec *iv;
891 	int i, left, siz;
892 
893 	left = len;
894 	m3 = NULL;
895 	/*
896 	 * Generate the mbuf list with the uio_iov ref. to it.
897 	 */
898 	i = 0;
899 	while (left > 0) {
900 		NFSMGET(m);
901 		MCLGET(m, M_WAITOK);
902 		m->m_len = 0;
903 		siz = min(M_TRAILINGSPACE(m), left);
904 		left -= siz;
905 		i++;
906 		if (m3)
907 			m2->m_next = m;
908 		else
909 			m3 = m;
910 		m2 = m;
911 	}
912 	*ivp = iv = malloc(i * sizeof (struct iovec), M_TEMP, M_WAITOK);
913 	m = m3;
914 	left = len;
915 	i = 0;
916 	while (left > 0) {
917 		if (m == NULL)
918 			panic("nfsrv_createiovec iov");
919 		siz = min(M_TRAILINGSPACE(m), left);
920 		if (siz > 0) {
921 			iv->iov_base = mtod(m, caddr_t) + m->m_len;
922 			iv->iov_len = siz;
923 			m->m_len += siz;
924 			left -= siz;
925 			iv++;
926 			i++;
927 		}
928 		m = m->m_next;
929 	}
930 	*mpp = m3;
931 	*mpendp = m2;
932 	return (i);
933 }
934 
935 /*
936  * Create an mbuf chain and an associated iovec that can be used to Read
937  * or Getextattr of data.
938  * Upon success, return pointers to the first and last mbufs in the chain
939  * plus the malloc'd iovec and its iovlen.
940  * Same as above, but creates ext_pgs mbuf(s).
941  */
942 static int
943 nfsrv_createiovec_extpgs(int len, int maxextsiz, struct mbuf **mpp,
944     struct mbuf **mpendp, struct iovec **ivp)
945 {
946 	struct mbuf *m, *m2 = NULL, *m3;
947 	struct iovec *iv;
948 	int i, left, pgno, siz;
949 
950 	left = len;
951 	m3 = NULL;
952 	/*
953 	 * Generate the mbuf list with the uio_iov ref. to it.
954 	 */
955 	i = 0;
956 	while (left > 0) {
957 		siz = min(left, maxextsiz);
958 		m = mb_alloc_ext_plus_pages(siz, M_WAITOK);
959 		left -= siz;
960 		i += m->m_epg_npgs;
961 		if (m3 != NULL)
962 			m2->m_next = m;
963 		else
964 			m3 = m;
965 		m2 = m;
966 	}
967 	*ivp = iv = malloc(i * sizeof (struct iovec), M_TEMP, M_WAITOK);
968 	m = m3;
969 	left = len;
970 	i = 0;
971 	pgno = 0;
972 	while (left > 0) {
973 		if (m == NULL)
974 			panic("nfsvno_createiovec_extpgs iov");
975 		siz = min(PAGE_SIZE, left);
976 		if (siz > 0) {
977 			iv->iov_base = (void *)PHYS_TO_DMAP(m->m_epg_pa[pgno]);
978 			iv->iov_len = siz;
979 			m->m_len += siz;
980 			if (pgno == m->m_epg_npgs - 1)
981 				m->m_epg_last_len = siz;
982 			left -= siz;
983 			iv++;
984 			i++;
985 			pgno++;
986 		}
987 		if (pgno == m->m_epg_npgs && left > 0) {
988 			m = m->m_next;
989 			if (m == NULL)
990 				panic("nfsvno_createiovec_extpgs iov");
991 			pgno = 0;
992 		}
993 	}
994 	*mpp = m3;
995 	*mpendp = m2;
996 	return (i);
997 }
998 
999 /*
1000  * Read vnode op call into mbuf list.
1001  */
1002 int
1003 nfsvno_read(struct vnode *vp, off_t off, int cnt, struct ucred *cred,
1004     int maxextsiz, struct thread *p, struct mbuf **mpp,
1005     struct mbuf **mpendp)
1006 {
1007 	struct mbuf *m;
1008 	struct iovec *iv;
1009 	int error = 0, len, tlen, ioflag = 0;
1010 	struct mbuf *m3;
1011 	struct uio io, *uiop = &io;
1012 	struct nfsheur *nh;
1013 
1014 	/*
1015 	 * Attempt to read from a DS file. A return of ENOENT implies
1016 	 * there is no DS file to read.
1017 	 */
1018 	error = nfsrv_proxyds(vp, off, cnt, cred, p, NFSPROC_READDS, mpp,
1019 	    NULL, mpendp, NULL, NULL, NULL, 0, NULL);
1020 	if (error != ENOENT)
1021 		return (error);
1022 
1023 	len = NFSM_RNDUP(cnt);
1024 	if (maxextsiz > 0)
1025 		uiop->uio_iovcnt = nfsrv_createiovec_extpgs(len, maxextsiz,
1026 		    &m3, &m, &iv);
1027 	else
1028 		uiop->uio_iovcnt = nfsrv_createiovec(len, &m3, &m, &iv);
1029 	uiop->uio_iov = iv;
1030 	uiop->uio_offset = off;
1031 	uiop->uio_resid = len;
1032 	uiop->uio_rw = UIO_READ;
1033 	uiop->uio_segflg = UIO_SYSSPACE;
1034 	uiop->uio_td = NULL;
1035 	nh = nfsrv_sequential_heuristic(uiop, vp);
1036 	ioflag |= nh->nh_seqcount << IO_SEQSHIFT;
1037 	/* XXX KDM make this more systematic? */
1038 	NFSD_VNET(nfsstatsv1_p)->srvbytes[NFSV4OP_READ] += uiop->uio_resid;
1039 	error = VOP_READ(vp, uiop, IO_NODELOCKED | ioflag, cred);
1040 	free(iv, M_TEMP);
1041 	if (error) {
1042 		m_freem(m3);
1043 		*mpp = NULL;
1044 		goto out;
1045 	}
1046 	nh->nh_nextoff = uiop->uio_offset;
1047 	tlen = len - uiop->uio_resid;
1048 	cnt = cnt < tlen ? cnt : tlen;
1049 	tlen = NFSM_RNDUP(cnt);
1050 	if (tlen == 0) {
1051 		m_freem(m3);
1052 		m3 = m = NULL;
1053 	} else if (len != tlen || tlen != cnt)
1054 		m = nfsrv_adj(m3, len - tlen, tlen - cnt);
1055 	*mpp = m3;
1056 	*mpendp = m;
1057 
1058 out:
1059 	NFSEXITCODE(error);
1060 	return (error);
1061 }
1062 
1063 /*
1064  * Create the iovec for the mbuf chain passed in as an argument.
1065  * The "cp" argument is where the data starts within the first mbuf in
1066  * the chain. It returns the iovec and the iovcnt.
1067  */
1068 static int
1069 nfsrv_createiovecw(int retlen, struct mbuf *m, char *cp, struct iovec **ivpp,
1070     int *iovcntp)
1071 {
1072 	struct mbuf *mp;
1073 	struct iovec *ivp;
1074 	int cnt, i, len;
1075 
1076 	/*
1077 	 * Loop through the mbuf chain, counting how many mbufs are a
1078 	 * part of this write operation, so the iovec size is known.
1079 	 */
1080 	cnt = 0;
1081 	len = retlen;
1082 	mp = m;
1083 	i = mtod(mp, caddr_t) + mp->m_len - cp;
1084 	while (len > 0) {
1085 		if (i > 0) {
1086 			len -= i;
1087 			cnt++;
1088 		}
1089 		mp = mp->m_next;
1090 		if (!mp) {
1091 			if (len > 0)
1092 				return (EBADRPC);
1093 		} else
1094 			i = mp->m_len;
1095 	}
1096 
1097 	/* Now, create the iovec. */
1098 	mp = m;
1099 	*ivpp = ivp = malloc(cnt * sizeof (struct iovec), M_TEMP,
1100 	    M_WAITOK);
1101 	*iovcntp = cnt;
1102 	i = mtod(mp, caddr_t) + mp->m_len - cp;
1103 	len = retlen;
1104 	while (len > 0) {
1105 		if (mp == NULL)
1106 			panic("nfsrv_createiovecw");
1107 		if (i > 0) {
1108 			i = min(i, len);
1109 			ivp->iov_base = cp;
1110 			ivp->iov_len = i;
1111 			ivp++;
1112 			len -= i;
1113 		}
1114 		mp = mp->m_next;
1115 		if (mp) {
1116 			i = mp->m_len;
1117 			cp = mtod(mp, caddr_t);
1118 		}
1119 	}
1120 	return (0);
1121 }
1122 
1123 /*
1124  * Write vnode op from an mbuf list.
1125  */
1126 int
1127 nfsvno_write(struct vnode *vp, off_t off, int retlen, int *stable,
1128     struct mbuf *mp, char *cp, struct ucred *cred, struct thread *p)
1129 {
1130 	struct iovec *iv;
1131 	int cnt, ioflags, error;
1132 	struct uio io, *uiop = &io;
1133 	struct nfsheur *nh;
1134 
1135 	/*
1136 	 * Attempt to write to a DS file. A return of ENOENT implies
1137 	 * there is no DS file to write.
1138 	 */
1139 	error = nfsrv_proxyds(vp, off, retlen, cred, p, NFSPROC_WRITEDS,
1140 	    &mp, cp, NULL, NULL, NULL, NULL, 0, NULL);
1141 	if (error != ENOENT) {
1142 		*stable = NFSWRITE_FILESYNC;
1143 		return (error);
1144 	}
1145 
1146 	if (*stable == NFSWRITE_UNSTABLE)
1147 		ioflags = IO_NODELOCKED;
1148 	else
1149 		ioflags = (IO_SYNC | IO_NODELOCKED);
1150 	error = nfsrv_createiovecw(retlen, mp, cp, &iv, &cnt);
1151 	if (error != 0)
1152 		return (error);
1153 	uiop->uio_iov = iv;
1154 	uiop->uio_iovcnt = cnt;
1155 	uiop->uio_resid = retlen;
1156 	uiop->uio_rw = UIO_WRITE;
1157 	uiop->uio_segflg = UIO_SYSSPACE;
1158 	NFSUIOPROC(uiop, p);
1159 	uiop->uio_offset = off;
1160 	nh = nfsrv_sequential_heuristic(uiop, vp);
1161 	ioflags |= nh->nh_seqcount << IO_SEQSHIFT;
1162 	/* XXX KDM make this more systematic? */
1163 	NFSD_VNET(nfsstatsv1_p)->srvbytes[NFSV4OP_WRITE] += uiop->uio_resid;
1164 	error = VOP_WRITE(vp, uiop, ioflags, cred);
1165 	if (error == 0)
1166 		nh->nh_nextoff = uiop->uio_offset;
1167 	free(iv, M_TEMP);
1168 
1169 	NFSEXITCODE(error);
1170 	return (error);
1171 }
1172 
1173 /*
1174  * Common code for creating a regular file (plus special files for V2).
1175  */
1176 int
1177 nfsvno_createsub(struct nfsrv_descript *nd, struct nameidata *ndp,
1178     struct vnode **vpp, struct nfsvattr *nvap, int *exclusive_flagp,
1179     int32_t *cverf, NFSDEV_T rdev, struct nfsexstuff *exp)
1180 {
1181 	u_quad_t tempsize;
1182 	int error;
1183 	struct thread *p = curthread;
1184 
1185 	error = nd->nd_repstat;
1186 	if (!error && ndp->ni_vp == NULL) {
1187 		if (nvap->na_type == VREG || nvap->na_type == VSOCK) {
1188 			error = VOP_CREATE(ndp->ni_dvp,
1189 			    &ndp->ni_vp, &ndp->ni_cnd, &nvap->na_vattr);
1190 			/* For a pNFS server, create the data file on a DS. */
1191 			if (error == 0 && nvap->na_type == VREG) {
1192 				/*
1193 				 * Create a data file on a DS for a pNFS server.
1194 				 * This function just returns if not
1195 				 * running a pNFS DS or the creation fails.
1196 				 */
1197 				nfsrv_pnfscreate(ndp->ni_vp, &nvap->na_vattr,
1198 				    nd->nd_cred, p);
1199 			}
1200 			VOP_VPUT_PAIR(ndp->ni_dvp, error == 0 ? &ndp->ni_vp :
1201 			    NULL, false);
1202 			nfsvno_relpathbuf(ndp);
1203 			if (!error) {
1204 				if (*exclusive_flagp) {
1205 					*exclusive_flagp = 0;
1206 					NFSVNO_ATTRINIT(nvap);
1207 					nvap->na_atime.tv_sec = cverf[0];
1208 					nvap->na_atime.tv_nsec = cverf[1];
1209 					error = VOP_SETATTR(ndp->ni_vp,
1210 					    &nvap->na_vattr, nd->nd_cred);
1211 					if (error != 0) {
1212 						vput(ndp->ni_vp);
1213 						ndp->ni_vp = NULL;
1214 						error = NFSERR_NOTSUPP;
1215 					}
1216 				}
1217 			}
1218 		/*
1219 		 * NFS V2 Only. nfsrvd_mknod() does this for V3.
1220 		 * (This implies, just get out on an error.)
1221 		 */
1222 		} else if (nvap->na_type == VCHR || nvap->na_type == VBLK ||
1223 			nvap->na_type == VFIFO) {
1224 			if (nvap->na_type == VCHR && rdev == 0xffffffff)
1225 				nvap->na_type = VFIFO;
1226                         if (nvap->na_type != VFIFO &&
1227 			    (error = priv_check_cred(nd->nd_cred, PRIV_VFS_MKNOD_DEV))) {
1228 				nfsvno_relpathbuf(ndp);
1229 				vput(ndp->ni_dvp);
1230 				goto out;
1231 			}
1232 			nvap->na_rdev = rdev;
1233 			error = VOP_MKNOD(ndp->ni_dvp, &ndp->ni_vp,
1234 			    &ndp->ni_cnd, &nvap->na_vattr);
1235 			VOP_VPUT_PAIR(ndp->ni_dvp, error == 0 ? &ndp->ni_vp :
1236 			    NULL, false);
1237 			nfsvno_relpathbuf(ndp);
1238 			if (error)
1239 				goto out;
1240 		} else {
1241 			nfsvno_relpathbuf(ndp);
1242 			vput(ndp->ni_dvp);
1243 			error = ENXIO;
1244 			goto out;
1245 		}
1246 		*vpp = ndp->ni_vp;
1247 	} else {
1248 		/*
1249 		 * Handle cases where error is already set and/or
1250 		 * the file exists.
1251 		 * 1 - clean up the lookup
1252 		 * 2 - iff !error and na_size set, truncate it
1253 		 */
1254 		nfsvno_relpathbuf(ndp);
1255 		*vpp = ndp->ni_vp;
1256 		if (ndp->ni_dvp == *vpp)
1257 			vrele(ndp->ni_dvp);
1258 		else
1259 			vput(ndp->ni_dvp);
1260 		if (!error && nvap->na_size != VNOVAL) {
1261 			error = nfsvno_accchk(*vpp, VWRITE,
1262 			    nd->nd_cred, exp, p, NFSACCCHK_NOOVERRIDE,
1263 			    NFSACCCHK_VPISLOCKED, NULL);
1264 			if (!error) {
1265 				tempsize = nvap->na_size;
1266 				NFSVNO_ATTRINIT(nvap);
1267 				nvap->na_size = tempsize;
1268 				error = nfsvno_setattr(*vpp, nvap,
1269 				    nd->nd_cred, p, exp);
1270 			}
1271 		}
1272 		if (error)
1273 			vput(*vpp);
1274 	}
1275 
1276 out:
1277 	NFSEXITCODE(error);
1278 	return (error);
1279 }
1280 
1281 /*
1282  * Do a mknod vnode op.
1283  */
1284 int
1285 nfsvno_mknod(struct nameidata *ndp, struct nfsvattr *nvap, struct ucred *cred,
1286     struct thread *p)
1287 {
1288 	int error = 0;
1289 	__enum_uint8(vtype) vtyp;
1290 
1291 	vtyp = nvap->na_type;
1292 	/*
1293 	 * Iff doesn't exist, create it.
1294 	 */
1295 	if (ndp->ni_vp) {
1296 		nfsvno_relpathbuf(ndp);
1297 		vput(ndp->ni_dvp);
1298 		vrele(ndp->ni_vp);
1299 		error = EEXIST;
1300 		goto out;
1301 	}
1302 	if (vtyp != VCHR && vtyp != VBLK && vtyp != VSOCK && vtyp != VFIFO) {
1303 		nfsvno_relpathbuf(ndp);
1304 		vput(ndp->ni_dvp);
1305 		error = NFSERR_BADTYPE;
1306 		goto out;
1307 	}
1308 	if (vtyp == VSOCK) {
1309 		error = VOP_CREATE(ndp->ni_dvp, &ndp->ni_vp,
1310 		    &ndp->ni_cnd, &nvap->na_vattr);
1311 		VOP_VPUT_PAIR(ndp->ni_dvp, error == 0 ? &ndp->ni_vp : NULL,
1312 		    false);
1313 		nfsvno_relpathbuf(ndp);
1314 	} else {
1315 		if (nvap->na_type != VFIFO &&
1316 		    (error = priv_check_cred(cred, PRIV_VFS_MKNOD_DEV))) {
1317 			nfsvno_relpathbuf(ndp);
1318 			vput(ndp->ni_dvp);
1319 			goto out;
1320 		}
1321 		error = VOP_MKNOD(ndp->ni_dvp, &ndp->ni_vp,
1322 		    &ndp->ni_cnd, &nvap->na_vattr);
1323 		VOP_VPUT_PAIR(ndp->ni_dvp, error == 0 ? &ndp->ni_vp : NULL,
1324 		    false);
1325 		nfsvno_relpathbuf(ndp);
1326 		/*
1327 		 * Since VOP_MKNOD returns the ni_vp, I can't
1328 		 * see any reason to do the lookup.
1329 		 */
1330 	}
1331 
1332 out:
1333 	NFSEXITCODE(error);
1334 	return (error);
1335 }
1336 
1337 /*
1338  * Mkdir vnode op.
1339  */
1340 int
1341 nfsvno_mkdir(struct nameidata *ndp, struct nfsvattr *nvap, uid_t saved_uid,
1342     struct ucred *cred, struct thread *p, struct nfsexstuff *exp)
1343 {
1344 	int error = 0;
1345 
1346 	if (ndp->ni_vp != NULL) {
1347 		if (ndp->ni_dvp == ndp->ni_vp)
1348 			vrele(ndp->ni_dvp);
1349 		else
1350 			vput(ndp->ni_dvp);
1351 		vrele(ndp->ni_vp);
1352 		nfsvno_relpathbuf(ndp);
1353 		error = EEXIST;
1354 		goto out;
1355 	}
1356 	error = VOP_MKDIR(ndp->ni_dvp, &ndp->ni_vp, &ndp->ni_cnd,
1357 	    &nvap->na_vattr);
1358 	VOP_VPUT_PAIR(ndp->ni_dvp, error == 0 ? &ndp->ni_vp : NULL, false);
1359 	nfsvno_relpathbuf(ndp);
1360 
1361 out:
1362 	NFSEXITCODE(error);
1363 	return (error);
1364 }
1365 
1366 /*
1367  * symlink vnode op.
1368  */
1369 int
1370 nfsvno_symlink(struct nameidata *ndp, struct nfsvattr *nvap, char *pathcp,
1371     int pathlen, int not_v2, uid_t saved_uid, struct ucred *cred, struct thread *p,
1372     struct nfsexstuff *exp)
1373 {
1374 	int error = 0;
1375 
1376 	if (ndp->ni_vp) {
1377 		nfsvno_relpathbuf(ndp);
1378 		if (ndp->ni_dvp == ndp->ni_vp)
1379 			vrele(ndp->ni_dvp);
1380 		else
1381 			vput(ndp->ni_dvp);
1382 		vrele(ndp->ni_vp);
1383 		error = EEXIST;
1384 		goto out;
1385 	}
1386 
1387 	error = VOP_SYMLINK(ndp->ni_dvp, &ndp->ni_vp, &ndp->ni_cnd,
1388 	    &nvap->na_vattr, pathcp);
1389 	/*
1390 	 * Although FreeBSD still had the lookup code in
1391 	 * it for 7/current, there doesn't seem to be any
1392 	 * point, since VOP_SYMLINK() returns the ni_vp.
1393 	 * Just vput it for v2.
1394 	 */
1395 	VOP_VPUT_PAIR(ndp->ni_dvp, &ndp->ni_vp, !not_v2 && error == 0);
1396 	nfsvno_relpathbuf(ndp);
1397 
1398 out:
1399 	NFSEXITCODE(error);
1400 	return (error);
1401 }
1402 
1403 /*
1404  * Parse symbolic link arguments.
1405  * This function has an ugly side effect. It will malloc() an area for
1406  * the symlink and set iov_base to point to it, only if it succeeds.
1407  * So, if it returns with uiop->uio_iov->iov_base != NULL, that must
1408  * be FREE'd later.
1409  */
1410 int
1411 nfsvno_getsymlink(struct nfsrv_descript *nd, struct nfsvattr *nvap,
1412     struct thread *p, char **pathcpp, int *lenp)
1413 {
1414 	u_int32_t *tl;
1415 	char *pathcp = NULL;
1416 	int error = 0, len;
1417 	struct nfsv2_sattr *sp;
1418 
1419 	*pathcpp = NULL;
1420 	*lenp = 0;
1421 	if ((nd->nd_flag & ND_NFSV3) &&
1422 	    (error = nfsrv_sattr(nd, NULL, nvap, NULL, NULL, p)))
1423 		goto nfsmout;
1424 	NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
1425 	len = fxdr_unsigned(int, *tl);
1426 	if (len > NFS_MAXPATHLEN || len <= 0) {
1427 		error = EBADRPC;
1428 		goto nfsmout;
1429 	}
1430 	pathcp = malloc(len + 1, M_TEMP, M_WAITOK);
1431 	error = nfsrv_mtostr(nd, pathcp, len);
1432 	if (error)
1433 		goto nfsmout;
1434 	if (nd->nd_flag & ND_NFSV2) {
1435 		NFSM_DISSECT(sp, struct nfsv2_sattr *, NFSX_V2SATTR);
1436 		nvap->na_mode = fxdr_unsigned(u_int16_t, sp->sa_mode);
1437 	}
1438 	*pathcpp = pathcp;
1439 	*lenp = len;
1440 	NFSEXITCODE2(0, nd);
1441 	return (0);
1442 nfsmout:
1443 	if (pathcp)
1444 		free(pathcp, M_TEMP);
1445 	NFSEXITCODE2(error, nd);
1446 	return (error);
1447 }
1448 
1449 /*
1450  * Remove a non-directory object.
1451  */
1452 int
1453 nfsvno_removesub(struct nameidata *ndp, int is_v4, struct ucred *cred,
1454     struct thread *p, struct nfsexstuff *exp)
1455 {
1456 	struct vnode *vp, *dsdvp[NFSDEV_MAXMIRRORS];
1457 	int error = 0, mirrorcnt;
1458 	char fname[PNFS_FILENAME_LEN + 1];
1459 	fhandle_t fh;
1460 
1461 	vp = ndp->ni_vp;
1462 	dsdvp[0] = NULL;
1463 	if (vp->v_type == VDIR)
1464 		error = NFSERR_ISDIR;
1465 	else if (is_v4)
1466 		error = nfsrv_checkremove(vp, 1, NULL, (nfsquad_t)((u_quad_t)0),
1467 		    p);
1468 	if (error == 0)
1469 		nfsrv_pnfsremovesetup(vp, p, dsdvp, &mirrorcnt, fname, &fh);
1470 	if (!error)
1471 		error = VOP_REMOVE(ndp->ni_dvp, vp, &ndp->ni_cnd);
1472 	if (error == 0 && dsdvp[0] != NULL)
1473 		nfsrv_pnfsremove(dsdvp, mirrorcnt, fname, &fh, p);
1474 	if (ndp->ni_dvp == vp)
1475 		vrele(ndp->ni_dvp);
1476 	else
1477 		vput(ndp->ni_dvp);
1478 	vput(vp);
1479 	nfsvno_relpathbuf(ndp);
1480 	NFSEXITCODE(error);
1481 	return (error);
1482 }
1483 
1484 /*
1485  * Remove a directory.
1486  */
1487 int
1488 nfsvno_rmdirsub(struct nameidata *ndp, int is_v4, struct ucred *cred,
1489     struct thread *p, struct nfsexstuff *exp)
1490 {
1491 	struct vnode *vp;
1492 	int error = 0;
1493 
1494 	vp = ndp->ni_vp;
1495 	if (vp->v_type != VDIR) {
1496 		error = ENOTDIR;
1497 		goto out;
1498 	}
1499 	/*
1500 	 * No rmdir "." please.
1501 	 */
1502 	if (ndp->ni_dvp == vp) {
1503 		error = EINVAL;
1504 		goto out;
1505 	}
1506 	/*
1507 	 * The root of a mounted filesystem cannot be deleted.
1508 	 */
1509 	if (vp->v_vflag & VV_ROOT)
1510 		error = EBUSY;
1511 out:
1512 	if (!error)
1513 		error = VOP_RMDIR(ndp->ni_dvp, vp, &ndp->ni_cnd);
1514 	if (ndp->ni_dvp == vp)
1515 		vrele(ndp->ni_dvp);
1516 	else
1517 		vput(ndp->ni_dvp);
1518 	vput(vp);
1519 	nfsvno_relpathbuf(ndp);
1520 	NFSEXITCODE(error);
1521 	return (error);
1522 }
1523 
1524 /*
1525  * Rename vnode op.
1526  */
1527 int
1528 nfsvno_rename(struct nameidata *fromndp, struct nameidata *tondp,
1529     u_int32_t ndstat, u_int32_t ndflag, struct ucred *cred, struct thread *p)
1530 {
1531 	struct vnode *fvp, *tvp, *tdvp, *dsdvp[NFSDEV_MAXMIRRORS];
1532 	int error = 0, mirrorcnt;
1533 	char fname[PNFS_FILENAME_LEN + 1];
1534 	fhandle_t fh;
1535 
1536 	dsdvp[0] = NULL;
1537 	fvp = fromndp->ni_vp;
1538 	if (ndstat) {
1539 		vrele(fromndp->ni_dvp);
1540 		vrele(fvp);
1541 		error = ndstat;
1542 		goto out1;
1543 	}
1544 	tdvp = tondp->ni_dvp;
1545 	tvp = tondp->ni_vp;
1546 	if (tvp != NULL) {
1547 		if (fvp->v_type == VDIR && tvp->v_type != VDIR) {
1548 			error = (ndflag & ND_NFSV2) ? EISDIR : EEXIST;
1549 			goto out;
1550 		} else if (fvp->v_type != VDIR && tvp->v_type == VDIR) {
1551 			error = (ndflag & ND_NFSV2) ? ENOTDIR : EEXIST;
1552 			goto out;
1553 		}
1554 		if (tvp->v_type == VDIR && tvp->v_mountedhere) {
1555 			error = (ndflag & ND_NFSV2) ? ENOTEMPTY : EXDEV;
1556 			goto out;
1557 		}
1558 
1559 		/*
1560 		 * A rename to '.' or '..' results in a prematurely
1561 		 * unlocked vnode on FreeBSD5, so I'm just going to fail that
1562 		 * here.
1563 		 */
1564 		if ((tondp->ni_cnd.cn_namelen == 1 &&
1565 		     tondp->ni_cnd.cn_nameptr[0] == '.') ||
1566 		    (tondp->ni_cnd.cn_namelen == 2 &&
1567 		     tondp->ni_cnd.cn_nameptr[0] == '.' &&
1568 		     tondp->ni_cnd.cn_nameptr[1] == '.')) {
1569 			error = EINVAL;
1570 			goto out;
1571 		}
1572 	}
1573 	if (fvp->v_type == VDIR && fvp->v_mountedhere) {
1574 		error = (ndflag & ND_NFSV2) ? ENOTEMPTY : EXDEV;
1575 		goto out;
1576 	}
1577 	if (fvp->v_mount != tdvp->v_mount) {
1578 		error = (ndflag & ND_NFSV2) ? ENOTEMPTY : EXDEV;
1579 		goto out;
1580 	}
1581 	if (fvp == tdvp) {
1582 		error = (ndflag & ND_NFSV2) ? ENOTEMPTY : EINVAL;
1583 		goto out;
1584 	}
1585 	if (fvp == tvp) {
1586 		/*
1587 		 * If source and destination are the same, there is nothing to
1588 		 * do. Set error to -1 to indicate this.
1589 		 */
1590 		error = -1;
1591 		goto out;
1592 	}
1593 	if (ndflag & ND_NFSV4) {
1594 		if (NFSVOPLOCK(fvp, LK_EXCLUSIVE) == 0) {
1595 			error = nfsrv_checkremove(fvp, 0, NULL,
1596 			    (nfsquad_t)((u_quad_t)0), p);
1597 			NFSVOPUNLOCK(fvp);
1598 		} else
1599 			error = EPERM;
1600 		if (tvp && !error)
1601 			error = nfsrv_checkremove(tvp, 1, NULL,
1602 			    (nfsquad_t)((u_quad_t)0), p);
1603 	} else {
1604 		/*
1605 		 * For NFSv2 and NFSv3, try to get rid of the delegation, so
1606 		 * that the NFSv4 client won't be confused by the rename.
1607 		 * Since nfsd_recalldelegation() can only be called on an
1608 		 * unlocked vnode at this point and fvp is the file that will
1609 		 * still exist after the rename, just do fvp.
1610 		 */
1611 		nfsd_recalldelegation(fvp, p);
1612 	}
1613 	if (error == 0 && tvp != NULL) {
1614 		nfsrv_pnfsremovesetup(tvp, p, dsdvp, &mirrorcnt, fname, &fh);
1615 		NFSD_DEBUG(4, "nfsvno_rename: pnfsremovesetup"
1616 		    " dsdvp=%p\n", dsdvp[0]);
1617 	}
1618 out:
1619 	if (!error) {
1620 		error = VOP_RENAME(fromndp->ni_dvp, fromndp->ni_vp,
1621 		    &fromndp->ni_cnd, tondp->ni_dvp, tondp->ni_vp,
1622 		    &tondp->ni_cnd);
1623 	} else {
1624 		if (tdvp == tvp)
1625 			vrele(tdvp);
1626 		else
1627 			vput(tdvp);
1628 		if (tvp)
1629 			vput(tvp);
1630 		vrele(fromndp->ni_dvp);
1631 		vrele(fvp);
1632 		if (error == -1)
1633 			error = 0;
1634 	}
1635 
1636 	/*
1637 	 * If dsdvp[0] != NULL, it was set up by nfsrv_pnfsremovesetup() and
1638 	 * if the rename succeeded, the DS file for the tvp needs to be
1639 	 * removed.
1640 	 */
1641 	if (error == 0 && dsdvp[0] != NULL) {
1642 		nfsrv_pnfsremove(dsdvp, mirrorcnt, fname, &fh, p);
1643 		NFSD_DEBUG(4, "nfsvno_rename: pnfsremove\n");
1644 	}
1645 
1646 	nfsvno_relpathbuf(tondp);
1647 out1:
1648 	nfsvno_relpathbuf(fromndp);
1649 	NFSEXITCODE(error);
1650 	return (error);
1651 }
1652 
1653 /*
1654  * Link vnode op.
1655  */
1656 int
1657 nfsvno_link(struct nameidata *ndp, struct vnode *vp, struct ucred *cred,
1658     struct thread *p, struct nfsexstuff *exp)
1659 {
1660 	struct vnode *xp;
1661 	int error = 0;
1662 
1663 	xp = ndp->ni_vp;
1664 	if (xp != NULL) {
1665 		error = EEXIST;
1666 	} else {
1667 		xp = ndp->ni_dvp;
1668 		if (vp->v_mount != xp->v_mount)
1669 			error = EXDEV;
1670 	}
1671 	if (!error) {
1672 		NFSVOPLOCK(vp, LK_EXCLUSIVE | LK_RETRY);
1673 		if (!VN_IS_DOOMED(vp))
1674 			error = VOP_LINK(ndp->ni_dvp, vp, &ndp->ni_cnd);
1675 		else
1676 			error = EPERM;
1677 		if (ndp->ni_dvp == vp) {
1678 			vrele(ndp->ni_dvp);
1679 			NFSVOPUNLOCK(vp);
1680 		} else {
1681 			vref(vp);
1682 			VOP_VPUT_PAIR(ndp->ni_dvp, &vp, true);
1683 		}
1684 	} else {
1685 		if (ndp->ni_dvp == ndp->ni_vp)
1686 			vrele(ndp->ni_dvp);
1687 		else
1688 			vput(ndp->ni_dvp);
1689 		if (ndp->ni_vp)
1690 			vrele(ndp->ni_vp);
1691 	}
1692 	nfsvno_relpathbuf(ndp);
1693 	NFSEXITCODE(error);
1694 	return (error);
1695 }
1696 
1697 /*
1698  * Do the fsync() appropriate for the commit.
1699  */
1700 int
1701 nfsvno_fsync(struct vnode *vp, u_int64_t off, int cnt, struct ucred *cred,
1702     struct thread *td)
1703 {
1704 	int error = 0;
1705 
1706 	/*
1707 	 * RFC 1813 3.3.21: if count is 0, a flush from offset to the end of
1708 	 * file is done.  At this time VOP_FSYNC does not accept offset and
1709 	 * byte count parameters so call VOP_FSYNC the whole file for now.
1710 	 * The same is true for NFSv4: RFC 3530 Sec. 14.2.3.
1711 	 * File systems that do not use the buffer cache (as indicated
1712 	 * by MNTK_USES_BCACHE not being set) must use VOP_FSYNC().
1713 	 */
1714 	if (cnt == 0 || cnt > MAX_COMMIT_COUNT ||
1715 	    (vp->v_mount->mnt_kern_flag & MNTK_USES_BCACHE) == 0) {
1716 		/*
1717 		 * Give up and do the whole thing
1718 		 */
1719 		if (vp->v_object && vm_object_mightbedirty(vp->v_object)) {
1720 			VM_OBJECT_WLOCK(vp->v_object);
1721 			vm_object_page_clean(vp->v_object, 0, 0, OBJPC_SYNC);
1722 			VM_OBJECT_WUNLOCK(vp->v_object);
1723 		}
1724 		error = VOP_FSYNC(vp, MNT_WAIT, td);
1725 	} else {
1726 		/*
1727 		 * Locate and synchronously write any buffers that fall
1728 		 * into the requested range.  Note:  we are assuming that
1729 		 * f_iosize is a power of 2.
1730 		 */
1731 		int iosize = vp->v_mount->mnt_stat.f_iosize;
1732 		int iomask = iosize - 1;
1733 		struct bufobj *bo;
1734 		daddr_t lblkno;
1735 
1736 		/*
1737 		 * Align to iosize boundary, super-align to page boundary.
1738 		 */
1739 		if (off & iomask) {
1740 			cnt += off & iomask;
1741 			off &= ~(u_quad_t)iomask;
1742 		}
1743 		if (off & PAGE_MASK) {
1744 			cnt += off & PAGE_MASK;
1745 			off &= ~(u_quad_t)PAGE_MASK;
1746 		}
1747 		lblkno = off / iosize;
1748 
1749 		if (vp->v_object && vm_object_mightbedirty(vp->v_object)) {
1750 			VM_OBJECT_WLOCK(vp->v_object);
1751 			vm_object_page_clean(vp->v_object, off, off + cnt,
1752 			    OBJPC_SYNC);
1753 			VM_OBJECT_WUNLOCK(vp->v_object);
1754 		}
1755 
1756 		bo = &vp->v_bufobj;
1757 		BO_LOCK(bo);
1758 		while (cnt > 0) {
1759 			struct buf *bp;
1760 
1761 			/*
1762 			 * If we have a buffer and it is marked B_DELWRI we
1763 			 * have to lock and write it.  Otherwise the prior
1764 			 * write is assumed to have already been committed.
1765 			 *
1766 			 * gbincore() can return invalid buffers now so we
1767 			 * have to check that bit as well (though B_DELWRI
1768 			 * should not be set if B_INVAL is set there could be
1769 			 * a race here since we haven't locked the buffer).
1770 			 */
1771 			if ((bp = gbincore(&vp->v_bufobj, lblkno)) != NULL) {
1772 				if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_SLEEPFAIL |
1773 				    LK_INTERLOCK, BO_LOCKPTR(bo)) == ENOLCK) {
1774 					BO_LOCK(bo);
1775 					continue; /* retry */
1776 				}
1777 			    	if ((bp->b_flags & (B_DELWRI|B_INVAL)) ==
1778 				    B_DELWRI) {
1779 					bremfree(bp);
1780 					bp->b_flags &= ~B_ASYNC;
1781 					bwrite(bp);
1782 					++nfs_commit_miss;
1783 				} else
1784 					BUF_UNLOCK(bp);
1785 				BO_LOCK(bo);
1786 			}
1787 			++nfs_commit_blks;
1788 			if (cnt < iosize)
1789 				break;
1790 			cnt -= iosize;
1791 			++lblkno;
1792 		}
1793 		BO_UNLOCK(bo);
1794 	}
1795 	NFSEXITCODE(error);
1796 	return (error);
1797 }
1798 
1799 /*
1800  * Statfs vnode op.
1801  */
1802 int
1803 nfsvno_statfs(struct vnode *vp, struct statfs *sf)
1804 {
1805 	struct statfs *tsf;
1806 	int error;
1807 
1808 	tsf = NULL;
1809 	if (nfsrv_devidcnt > 0) {
1810 		/* For a pNFS service, get the DS numbers. */
1811 		tsf = malloc(sizeof(*tsf), M_TEMP, M_WAITOK | M_ZERO);
1812 		error = nfsrv_pnfsstatfs(tsf, vp->v_mount);
1813 		if (error != 0) {
1814 			free(tsf, M_TEMP);
1815 			tsf = NULL;
1816 		}
1817 	}
1818 	error = VFS_STATFS(vp->v_mount, sf);
1819 	if (error == 0) {
1820 		if (tsf != NULL) {
1821 			sf->f_blocks = tsf->f_blocks;
1822 			sf->f_bavail = tsf->f_bavail;
1823 			sf->f_bfree = tsf->f_bfree;
1824 			sf->f_bsize = tsf->f_bsize;
1825 		}
1826 		/*
1827 		 * Since NFS handles these values as unsigned on the
1828 		 * wire, there is no way to represent negative values,
1829 		 * so set them to 0. Without this, they will appear
1830 		 * to be very large positive values for clients like
1831 		 * Solaris10.
1832 		 */
1833 		if (sf->f_bavail < 0)
1834 			sf->f_bavail = 0;
1835 		if (sf->f_ffree < 0)
1836 			sf->f_ffree = 0;
1837 	}
1838 	free(tsf, M_TEMP);
1839 	NFSEXITCODE(error);
1840 	return (error);
1841 }
1842 
1843 /*
1844  * Do the vnode op stuff for Open. Similar to nfsvno_createsub(), but
1845  * must handle nfsrv_opencheck() calls after any other access checks.
1846  */
1847 void
1848 nfsvno_open(struct nfsrv_descript *nd, struct nameidata *ndp,
1849     nfsquad_t clientid, nfsv4stateid_t *stateidp, struct nfsstate *stp,
1850     int *exclusive_flagp, struct nfsvattr *nvap, int32_t *cverf, int create,
1851     NFSACL_T *aclp, nfsattrbit_t *attrbitp, struct ucred *cred, bool done_namei,
1852     struct nfsexstuff *exp, struct vnode **vpp)
1853 {
1854 	struct vnode *vp = NULL;
1855 	u_quad_t tempsize;
1856 	struct nfsexstuff nes;
1857 	struct thread *p = curthread;
1858 	uint32_t oldrepstat;
1859 
1860 	if (ndp->ni_vp == NULL) {
1861 		/*
1862 		 * If nfsrv_opencheck() sets nd_repstat, done_namei needs to be
1863 		 * set true, since cleanup after nfsvno_namei() is needed.
1864 		 */
1865 		oldrepstat = nd->nd_repstat;
1866 		nd->nd_repstat = nfsrv_opencheck(clientid,
1867 		    stateidp, stp, NULL, nd, p, nd->nd_repstat);
1868 		if (nd->nd_repstat != 0 && oldrepstat == 0)
1869 			done_namei = true;
1870 	}
1871 	if (!nd->nd_repstat) {
1872 		if (ndp->ni_vp == NULL) {
1873 			nd->nd_repstat = VOP_CREATE(ndp->ni_dvp,
1874 			    &ndp->ni_vp, &ndp->ni_cnd, &nvap->na_vattr);
1875 			/* For a pNFS server, create the data file on a DS. */
1876 			if (nd->nd_repstat == 0) {
1877 				/*
1878 				 * Create a data file on a DS for a pNFS server.
1879 				 * This function just returns if not
1880 				 * running a pNFS DS or the creation fails.
1881 				 */
1882 				nfsrv_pnfscreate(ndp->ni_vp, &nvap->na_vattr,
1883 				    cred, p);
1884 			}
1885 			VOP_VPUT_PAIR(ndp->ni_dvp, nd->nd_repstat == 0 ?
1886 			    &ndp->ni_vp : NULL, false);
1887 			nfsvno_relpathbuf(ndp);
1888 			if (!nd->nd_repstat) {
1889 				if (*exclusive_flagp) {
1890 					*exclusive_flagp = 0;
1891 					NFSVNO_ATTRINIT(nvap);
1892 					nvap->na_atime.tv_sec = cverf[0];
1893 					nvap->na_atime.tv_nsec = cverf[1];
1894 					nd->nd_repstat = VOP_SETATTR(ndp->ni_vp,
1895 					    &nvap->na_vattr, cred);
1896 					if (nd->nd_repstat != 0) {
1897 						vput(ndp->ni_vp);
1898 						ndp->ni_vp = NULL;
1899 						nd->nd_repstat = NFSERR_NOTSUPP;
1900 					} else
1901 						NFSSETBIT_ATTRBIT(attrbitp,
1902 						    NFSATTRBIT_TIMEACCESS);
1903 				} else {
1904 					nfsrv_fixattr(nd, ndp->ni_vp, nvap,
1905 					    aclp, p, attrbitp, exp);
1906 				}
1907 			}
1908 			vp = ndp->ni_vp;
1909 		} else {
1910 			nfsvno_relpathbuf(ndp);
1911 			vp = ndp->ni_vp;
1912 			if (create == NFSV4OPEN_CREATE) {
1913 				if (ndp->ni_dvp == vp)
1914 					vrele(ndp->ni_dvp);
1915 				else
1916 					vput(ndp->ni_dvp);
1917 			}
1918 			if (NFSVNO_ISSETSIZE(nvap) && vp->v_type == VREG) {
1919 				if (ndp->ni_cnd.cn_flags & RDONLY)
1920 					NFSVNO_SETEXRDONLY(&nes);
1921 				else
1922 					NFSVNO_EXINIT(&nes);
1923 				nd->nd_repstat = nfsvno_accchk(vp,
1924 				    VWRITE, cred, &nes, p,
1925 				    NFSACCCHK_NOOVERRIDE,
1926 				    NFSACCCHK_VPISLOCKED, NULL);
1927 				nd->nd_repstat = nfsrv_opencheck(clientid,
1928 				    stateidp, stp, vp, nd, p, nd->nd_repstat);
1929 				if (!nd->nd_repstat) {
1930 					tempsize = nvap->na_size;
1931 					NFSVNO_ATTRINIT(nvap);
1932 					nvap->na_size = tempsize;
1933 					nd->nd_repstat = nfsvno_setattr(vp,
1934 					    nvap, cred, p, exp);
1935 				}
1936 			} else if (vp->v_type == VREG) {
1937 				nd->nd_repstat = nfsrv_opencheck(clientid,
1938 				    stateidp, stp, vp, nd, p, nd->nd_repstat);
1939 			}
1940 		}
1941 	} else if (done_namei) {
1942 		KASSERT(create == NFSV4OPEN_CREATE,
1943 		    ("nfsvno_open: not create"));
1944 		/*
1945 		 * done_namei is set when nfsvno_namei() has completed
1946 		 * successfully, but a subsequent error was set in
1947 		 * nd_repstat.  As such, cleanup of the nfsvno_namei()
1948 		 * results is required.
1949 		 */
1950 		nfsvno_relpathbuf(ndp);
1951 		if (ndp->ni_dvp == ndp->ni_vp)
1952 			vrele(ndp->ni_dvp);
1953 		else
1954 			vput(ndp->ni_dvp);
1955 		if (ndp->ni_vp)
1956 			vput(ndp->ni_vp);
1957 	}
1958 	*vpp = vp;
1959 
1960 	NFSEXITCODE2(0, nd);
1961 }
1962 
1963 /*
1964  * Updates the file rev and sets the mtime and ctime
1965  * to the current clock time, returning the va_filerev and va_Xtime
1966  * values.
1967  * Return ESTALE to indicate the vnode is VIRF_DOOMED.
1968  */
1969 int
1970 nfsvno_updfilerev(struct vnode *vp, struct nfsvattr *nvap,
1971     struct nfsrv_descript *nd, struct thread *p)
1972 {
1973 	struct vattr va;
1974 
1975 	VATTR_NULL(&va);
1976 	vfs_timestamp(&va.va_mtime);
1977 	if (NFSVOPISLOCKED(vp) != LK_EXCLUSIVE) {
1978 		NFSVOPLOCK(vp, LK_UPGRADE | LK_RETRY);
1979 		if (VN_IS_DOOMED(vp))
1980 			return (ESTALE);
1981 	}
1982 	(void) VOP_SETATTR(vp, &va, nd->nd_cred);
1983 	(void) nfsvno_getattr(vp, nvap, nd, p, 1, NULL);
1984 	return (0);
1985 }
1986 
1987 /*
1988  * Glue routine to nfsv4_fillattr().
1989  */
1990 int
1991 nfsvno_fillattr(struct nfsrv_descript *nd, struct mount *mp, struct vnode *vp,
1992     struct nfsvattr *nvap, fhandle_t *fhp, int rderror, nfsattrbit_t *attrbitp,
1993     struct ucred *cred, struct thread *p, int isdgram, int reterr,
1994     int supports_nfsv4acls, int at_root, uint64_t mounted_on_fileno)
1995 {
1996 	struct statfs *sf;
1997 	int error;
1998 
1999 	sf = NULL;
2000 	if (nfsrv_devidcnt > 0 &&
2001 	    (NFSISSET_ATTRBIT(attrbitp, NFSATTRBIT_SPACEAVAIL) ||
2002 	     NFSISSET_ATTRBIT(attrbitp, NFSATTRBIT_SPACEFREE) ||
2003 	     NFSISSET_ATTRBIT(attrbitp, NFSATTRBIT_SPACETOTAL))) {
2004 		sf = malloc(sizeof(*sf), M_TEMP, M_WAITOK | M_ZERO);
2005 		error = nfsrv_pnfsstatfs(sf, mp);
2006 		if (error != 0) {
2007 			free(sf, M_TEMP);
2008 			sf = NULL;
2009 		}
2010 	}
2011 	error = nfsv4_fillattr(nd, mp, vp, NULL, &nvap->na_vattr, fhp, rderror,
2012 	    attrbitp, cred, p, isdgram, reterr, supports_nfsv4acls, at_root,
2013 	    mounted_on_fileno, sf);
2014 	free(sf, M_TEMP);
2015 	NFSEXITCODE2(0, nd);
2016 	return (error);
2017 }
2018 
2019 /* Since the Readdir vnode ops vary, put the entire functions in here. */
2020 /*
2021  * nfs readdir service
2022  * - mallocs what it thinks is enough to read
2023  *	count rounded up to a multiple of DIRBLKSIZ <= NFS_MAXREADDIR
2024  * - calls VOP_READDIR()
2025  * - loops around building the reply
2026  *	if the output generated exceeds count break out of loop
2027  *	The NFSM_CLGET macro is used here so that the reply will be packed
2028  *	tightly in mbuf clusters.
2029  * - it trims out records with d_fileno == 0
2030  *	this doesn't matter for Unix clients, but they might confuse clients
2031  *	for other os'.
2032  * - it trims out records with d_type == DT_WHT
2033  *	these cannot be seen through NFS (unless we extend the protocol)
2034  *     The alternate call nfsrvd_readdirplus() does lookups as well.
2035  * PS: The NFS protocol spec. does not clarify what the "count" byte
2036  *	argument is a count of.. just name strings and file id's or the
2037  *	entire reply rpc or ...
2038  *	I tried just file name and id sizes and it confused the Sun client,
2039  *	so I am using the full rpc size now. The "paranoia.." comment refers
2040  *	to including the status longwords that are not a part of the dir.
2041  *	"entry" structures, but are in the rpc.
2042  */
2043 int
2044 nfsrvd_readdir(struct nfsrv_descript *nd, int isdgram,
2045     struct vnode *vp, struct nfsexstuff *exp)
2046 {
2047 	struct dirent *dp;
2048 	u_int32_t *tl;
2049 	int dirlen;
2050 	char *cpos, *cend, *rbuf;
2051 	struct nfsvattr at;
2052 	int nlen, error = 0, getret = 1;
2053 	int siz, cnt, fullsiz, eofflag, ncookies;
2054 	u_int64_t off, toff, verf __unused;
2055 	uint64_t *cookies = NULL, *cookiep;
2056 	struct uio io;
2057 	struct iovec iv;
2058 	int is_ufs;
2059 	struct thread *p = curthread;
2060 
2061 	if (nd->nd_repstat) {
2062 		nfsrv_postopattr(nd, getret, &at);
2063 		goto out;
2064 	}
2065 	if (nd->nd_flag & ND_NFSV2) {
2066 		NFSM_DISSECT(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
2067 		off = fxdr_unsigned(u_quad_t, *tl++);
2068 	} else {
2069 		NFSM_DISSECT(tl, u_int32_t *, 5 * NFSX_UNSIGNED);
2070 		off = fxdr_hyper(tl);
2071 		tl += 2;
2072 		verf = fxdr_hyper(tl);
2073 		tl += 2;
2074 	}
2075 	toff = off;
2076 	cnt = fxdr_unsigned(int, *tl);
2077 	if (cnt > NFS_SRVMAXDATA(nd) || cnt < 0)
2078 		cnt = NFS_SRVMAXDATA(nd);
2079 	siz = ((cnt + DIRBLKSIZ - 1) & ~(DIRBLKSIZ - 1));
2080 	fullsiz = siz;
2081 	if (nd->nd_flag & ND_NFSV3) {
2082 		nd->nd_repstat = getret = nfsvno_getattr(vp, &at, nd, p, 1,
2083 		    NULL);
2084 #if 0
2085 		/*
2086 		 * va_filerev is not sufficient as a cookie verifier,
2087 		 * since it is not supposed to change when entries are
2088 		 * removed/added unless that offset cookies returned to
2089 		 * the client are no longer valid.
2090 		 */
2091 		if (!nd->nd_repstat && toff && verf != at.na_filerev)
2092 			nd->nd_repstat = NFSERR_BAD_COOKIE;
2093 #endif
2094 	}
2095 	if (!nd->nd_repstat && vp->v_type != VDIR)
2096 		nd->nd_repstat = NFSERR_NOTDIR;
2097 	if (nd->nd_repstat == 0 && cnt == 0) {
2098 		if (nd->nd_flag & ND_NFSV2)
2099 			/* NFSv2 does not have NFSERR_TOOSMALL */
2100 			nd->nd_repstat = EPERM;
2101 		else
2102 			nd->nd_repstat = NFSERR_TOOSMALL;
2103 	}
2104 	if (!nd->nd_repstat)
2105 		nd->nd_repstat = nfsvno_accchk(vp, VEXEC,
2106 		    nd->nd_cred, exp, p, NFSACCCHK_NOOVERRIDE,
2107 		    NFSACCCHK_VPISLOCKED, NULL);
2108 	if (nd->nd_repstat) {
2109 		vput(vp);
2110 		if (nd->nd_flag & ND_NFSV3)
2111 			nfsrv_postopattr(nd, getret, &at);
2112 		goto out;
2113 	}
2114 	is_ufs = strcmp(vp->v_mount->mnt_vfc->vfc_name, "ufs") == 0;
2115 	rbuf = malloc(siz, M_TEMP, M_WAITOK);
2116 again:
2117 	eofflag = 0;
2118 	if (cookies) {
2119 		free(cookies, M_TEMP);
2120 		cookies = NULL;
2121 	}
2122 
2123 	iv.iov_base = rbuf;
2124 	iv.iov_len = siz;
2125 	io.uio_iov = &iv;
2126 	io.uio_iovcnt = 1;
2127 	io.uio_offset = (off_t)off;
2128 	io.uio_resid = siz;
2129 	io.uio_segflg = UIO_SYSSPACE;
2130 	io.uio_rw = UIO_READ;
2131 	io.uio_td = NULL;
2132 	nd->nd_repstat = VOP_READDIR(vp, &io, nd->nd_cred, &eofflag, &ncookies,
2133 	    &cookies);
2134 	off = (u_int64_t)io.uio_offset;
2135 	if (io.uio_resid)
2136 		siz -= io.uio_resid;
2137 
2138 	if (!cookies && !nd->nd_repstat)
2139 		nd->nd_repstat = NFSERR_PERM;
2140 	if (nd->nd_flag & ND_NFSV3) {
2141 		getret = nfsvno_getattr(vp, &at, nd, p, 1, NULL);
2142 		if (!nd->nd_repstat)
2143 			nd->nd_repstat = getret;
2144 	}
2145 
2146 	/*
2147 	 * Handles the failed cases. nd->nd_repstat == 0 past here.
2148 	 */
2149 	if (nd->nd_repstat) {
2150 		vput(vp);
2151 		free(rbuf, M_TEMP);
2152 		if (cookies)
2153 			free(cookies, M_TEMP);
2154 		if (nd->nd_flag & ND_NFSV3)
2155 			nfsrv_postopattr(nd, getret, &at);
2156 		goto out;
2157 	}
2158 	/*
2159 	 * If nothing read, return eof
2160 	 * rpc reply
2161 	 */
2162 	if (siz == 0) {
2163 		vput(vp);
2164 		if (nd->nd_flag & ND_NFSV2) {
2165 			NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
2166 		} else {
2167 			nfsrv_postopattr(nd, getret, &at);
2168 			NFSM_BUILD(tl, u_int32_t *, 4 * NFSX_UNSIGNED);
2169 			txdr_hyper(at.na_filerev, tl);
2170 			tl += 2;
2171 		}
2172 		*tl++ = newnfs_false;
2173 		*tl = newnfs_true;
2174 		free(rbuf, M_TEMP);
2175 		free(cookies, M_TEMP);
2176 		goto out;
2177 	}
2178 
2179 	/*
2180 	 * Check for degenerate cases of nothing useful read.
2181 	 * If so go try again
2182 	 */
2183 	cpos = rbuf;
2184 	cend = rbuf + siz;
2185 	dp = (struct dirent *)cpos;
2186 	cookiep = cookies;
2187 
2188 	/*
2189 	 * For some reason FreeBSD's ufs_readdir() chooses to back the
2190 	 * directory offset up to a block boundary, so it is necessary to
2191 	 * skip over the records that precede the requested offset. This
2192 	 * requires the assumption that file offset cookies monotonically
2193 	 * increase.
2194 	 */
2195 	while (cpos < cend && ncookies > 0 &&
2196 	    (dp->d_fileno == 0 || dp->d_type == DT_WHT ||
2197 	     (is_ufs == 1 && ((u_quad_t)(*cookiep)) <= toff))) {
2198 		cpos += dp->d_reclen;
2199 		dp = (struct dirent *)cpos;
2200 		cookiep++;
2201 		ncookies--;
2202 	}
2203 	if (cpos >= cend || ncookies == 0) {
2204 		siz = fullsiz;
2205 		toff = off;
2206 		goto again;
2207 	}
2208 	vput(vp);
2209 
2210 	/*
2211 	 * If cnt > MCLBYTES and the reply will not be saved, use
2212 	 * ext_pgs mbufs for TLS.
2213 	 * For NFSv4.0, we do not know for sure if the reply will
2214 	 * be saved, so do not use ext_pgs mbufs for NFSv4.0.
2215 	 */
2216 	if (cnt > MCLBYTES && siz > MCLBYTES &&
2217 	    (nd->nd_flag & (ND_TLS | ND_EXTPG | ND_SAVEREPLY)) == ND_TLS &&
2218 	    (nd->nd_flag & (ND_NFSV4 | ND_NFSV41)) != ND_NFSV4)
2219 		nd->nd_flag |= ND_EXTPG;
2220 
2221 	/*
2222 	 * dirlen is the size of the reply, including all XDR and must
2223 	 * not exceed cnt. For NFSv2, RFC1094 didn't clearly indicate
2224 	 * if the XDR should be included in "count", but to be safe, we do.
2225 	 * (Include the two booleans at the end of the reply in dirlen now.)
2226 	 */
2227 	if (nd->nd_flag & ND_NFSV3) {
2228 		nfsrv_postopattr(nd, getret, &at);
2229 		NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
2230 		txdr_hyper(at.na_filerev, tl);
2231 		dirlen = NFSX_V3POSTOPATTR + NFSX_VERF + 2 * NFSX_UNSIGNED;
2232 	} else {
2233 		dirlen = 2 * NFSX_UNSIGNED;
2234 	}
2235 
2236 	/* Loop through the records and build reply */
2237 	while (cpos < cend && ncookies > 0) {
2238 		nlen = dp->d_namlen;
2239 		if (dp->d_fileno != 0 && dp->d_type != DT_WHT &&
2240 			nlen <= NFS_MAXNAMLEN) {
2241 			if (nd->nd_flag & ND_NFSV3)
2242 				dirlen += (6*NFSX_UNSIGNED + NFSM_RNDUP(nlen));
2243 			else
2244 				dirlen += (4*NFSX_UNSIGNED + NFSM_RNDUP(nlen));
2245 			if (dirlen > cnt) {
2246 				eofflag = 0;
2247 				break;
2248 			}
2249 
2250 			/*
2251 			 * Build the directory record xdr from
2252 			 * the dirent entry.
2253 			 */
2254 			if (nd->nd_flag & ND_NFSV3) {
2255 				NFSM_BUILD(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
2256 				*tl++ = newnfs_true;
2257 				txdr_hyper(dp->d_fileno, tl);
2258 			} else {
2259 				NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
2260 				*tl++ = newnfs_true;
2261 				*tl = txdr_unsigned(dp->d_fileno);
2262 			}
2263 			(void) nfsm_strtom(nd, dp->d_name, nlen);
2264 			if (nd->nd_flag & ND_NFSV3) {
2265 				NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
2266 				txdr_hyper(*cookiep, tl);
2267 			} else {
2268 				NFSM_BUILD(tl, u_int32_t *, NFSX_UNSIGNED);
2269 				*tl = txdr_unsigned(*cookiep);
2270 			}
2271 		}
2272 		cpos += dp->d_reclen;
2273 		dp = (struct dirent *)cpos;
2274 		cookiep++;
2275 		ncookies--;
2276 	}
2277 	if (cpos < cend)
2278 		eofflag = 0;
2279 	NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
2280 	*tl++ = newnfs_false;
2281 	if (eofflag)
2282 		*tl = newnfs_true;
2283 	else
2284 		*tl = newnfs_false;
2285 	free(rbuf, M_TEMP);
2286 	free(cookies, M_TEMP);
2287 
2288 out:
2289 	NFSEXITCODE2(0, nd);
2290 	return (0);
2291 nfsmout:
2292 	vput(vp);
2293 	NFSEXITCODE2(error, nd);
2294 	return (error);
2295 }
2296 
2297 /*
2298  * Readdirplus for V3 and Readdir for V4.
2299  */
2300 int
2301 nfsrvd_readdirplus(struct nfsrv_descript *nd, int isdgram,
2302     struct vnode *vp, struct nfsexstuff *exp)
2303 {
2304 	struct dirent *dp;
2305 	u_int32_t *tl;
2306 	int dirlen;
2307 	char *cpos, *cend, *rbuf;
2308 	struct vnode *nvp;
2309 	fhandle_t nfh;
2310 	struct nfsvattr nva, at, *nvap = &nva;
2311 	struct mbuf *mb0, *mb1;
2312 	struct nfsreferral *refp;
2313 	int nlen, r, error = 0, getret = 1, usevget = 1;
2314 	int siz, cnt, fullsiz, eofflag, ncookies, entrycnt;
2315 	caddr_t bpos0, bpos1;
2316 	u_int64_t off, toff, verf __unused;
2317 	uint64_t *cookies = NULL, *cookiep;
2318 	nfsattrbit_t attrbits, rderrbits, savbits, refbits;
2319 	struct uio io;
2320 	struct iovec iv;
2321 	struct componentname cn;
2322 	int at_root, is_ufs, is_zfs, needs_unbusy, supports_nfsv4acls;
2323 	struct mount *mp, *new_mp;
2324 	uint64_t mounted_on_fileno;
2325 	struct thread *p = curthread;
2326 	int bextpg0, bextpg1, bextpgsiz0, bextpgsiz1;
2327 
2328 	if (nd->nd_repstat) {
2329 		nfsrv_postopattr(nd, getret, &at);
2330 		goto out;
2331 	}
2332 	NFSM_DISSECT(tl, u_int32_t *, 6 * NFSX_UNSIGNED);
2333 	off = fxdr_hyper(tl);
2334 	toff = off;
2335 	tl += 2;
2336 	verf = fxdr_hyper(tl);
2337 	tl += 2;
2338 	siz = fxdr_unsigned(int, *tl++);
2339 	cnt = fxdr_unsigned(int, *tl);
2340 
2341 	/*
2342 	 * Use the server's maximum data transfer size as the upper bound
2343 	 * on reply datalen.
2344 	 */
2345 	if (cnt > NFS_SRVMAXDATA(nd) || cnt < 0)
2346 		cnt = NFS_SRVMAXDATA(nd);
2347 
2348 	/*
2349 	 * siz is a "hint" of how much directory information (name, fileid,
2350 	 * cookie) should be in the reply. At least one client "hints" 0,
2351 	 * so I set it to cnt for that case. I also round it up to the
2352 	 * next multiple of DIRBLKSIZ.
2353 	 * Since the size of a Readdirplus directory entry reply will always
2354 	 * be greater than a directory entry returned by VOP_READDIR(), it
2355 	 * does not make sense to read more than NFS_SRVMAXDATA() via
2356 	 * VOP_READDIR().
2357 	 */
2358 	if (siz <= 0)
2359 		siz = cnt;
2360 	else if (siz > NFS_SRVMAXDATA(nd))
2361 		siz = NFS_SRVMAXDATA(nd);
2362 	siz = ((siz + DIRBLKSIZ - 1) & ~(DIRBLKSIZ - 1));
2363 
2364 	if (nd->nd_flag & ND_NFSV4) {
2365 		error = nfsrv_getattrbits(nd, &attrbits, NULL, NULL);
2366 		if (error)
2367 			goto nfsmout;
2368 		NFSSET_ATTRBIT(&savbits, &attrbits);
2369 		NFSSET_ATTRBIT(&refbits, &attrbits);
2370 		NFSCLRNOTFILLABLE_ATTRBIT(&attrbits, nd);
2371 		NFSZERO_ATTRBIT(&rderrbits);
2372 		NFSSETBIT_ATTRBIT(&rderrbits, NFSATTRBIT_RDATTRERROR);
2373 		/*
2374 		 * If these 4 bits are the only attributes requested by the
2375 		 * client, they can be satisfied without acquiring the vnode
2376 		 * for the file object unless it is a directory.
2377 		 * This will be indicated by savbits being all 0s.
2378 		 */
2379 		NFSCLRBIT_ATTRBIT(&savbits, NFSATTRBIT_TYPE);
2380 		NFSCLRBIT_ATTRBIT(&savbits, NFSATTRBIT_FILEID);
2381 		NFSCLRBIT_ATTRBIT(&savbits, NFSATTRBIT_MOUNTEDONFILEID);
2382 		NFSCLRBIT_ATTRBIT(&savbits, NFSATTRBIT_RDATTRERROR);
2383 	} else {
2384 		NFSZERO_ATTRBIT(&attrbits);
2385 	}
2386 	fullsiz = siz;
2387 	nd->nd_repstat = getret = nfsvno_getattr(vp, &at, nd, p, 1, NULL);
2388 #if 0
2389 	if (!nd->nd_repstat) {
2390 	    if (off && verf != at.na_filerev) {
2391 		/*
2392 		 * va_filerev is not sufficient as a cookie verifier,
2393 		 * since it is not supposed to change when entries are
2394 		 * removed/added unless that offset cookies returned to
2395 		 * the client are no longer valid.
2396 		 */
2397 		if (nd->nd_flag & ND_NFSV4) {
2398 			nd->nd_repstat = NFSERR_NOTSAME;
2399 		} else {
2400 			nd->nd_repstat = NFSERR_BAD_COOKIE;
2401 		}
2402 	    }
2403 	}
2404 #endif
2405 	if (!nd->nd_repstat && vp->v_type != VDIR)
2406 		nd->nd_repstat = NFSERR_NOTDIR;
2407 	if (!nd->nd_repstat && cnt == 0)
2408 		nd->nd_repstat = NFSERR_TOOSMALL;
2409 	if (!nd->nd_repstat)
2410 		nd->nd_repstat = nfsvno_accchk(vp, VEXEC,
2411 		    nd->nd_cred, exp, p, NFSACCCHK_NOOVERRIDE,
2412 		    NFSACCCHK_VPISLOCKED, NULL);
2413 	if (nd->nd_repstat) {
2414 		vput(vp);
2415 		if (nd->nd_flag & ND_NFSV3)
2416 			nfsrv_postopattr(nd, getret, &at);
2417 		goto out;
2418 	}
2419 	is_ufs = strcmp(vp->v_mount->mnt_vfc->vfc_name, "ufs") == 0;
2420 	is_zfs = strcmp(vp->v_mount->mnt_vfc->vfc_name, "zfs") == 0;
2421 
2422 	rbuf = malloc(siz, M_TEMP, M_WAITOK);
2423 again:
2424 	eofflag = 0;
2425 	if (cookies) {
2426 		free(cookies, M_TEMP);
2427 		cookies = NULL;
2428 	}
2429 
2430 	iv.iov_base = rbuf;
2431 	iv.iov_len = siz;
2432 	io.uio_iov = &iv;
2433 	io.uio_iovcnt = 1;
2434 	io.uio_offset = (off_t)off;
2435 	io.uio_resid = siz;
2436 	io.uio_segflg = UIO_SYSSPACE;
2437 	io.uio_rw = UIO_READ;
2438 	io.uio_td = NULL;
2439 	nd->nd_repstat = VOP_READDIR(vp, &io, nd->nd_cred, &eofflag, &ncookies,
2440 	    &cookies);
2441 	off = (u_int64_t)io.uio_offset;
2442 	if (io.uio_resid)
2443 		siz -= io.uio_resid;
2444 
2445 	getret = nfsvno_getattr(vp, &at, nd, p, 1, NULL);
2446 
2447 	if (!cookies && !nd->nd_repstat)
2448 		nd->nd_repstat = NFSERR_PERM;
2449 	if (!nd->nd_repstat)
2450 		nd->nd_repstat = getret;
2451 	if (nd->nd_repstat) {
2452 		vput(vp);
2453 		if (cookies)
2454 			free(cookies, M_TEMP);
2455 		free(rbuf, M_TEMP);
2456 		if (nd->nd_flag & ND_NFSV3)
2457 			nfsrv_postopattr(nd, getret, &at);
2458 		goto out;
2459 	}
2460 	/*
2461 	 * If nothing read, return eof
2462 	 * rpc reply
2463 	 */
2464 	if (siz == 0) {
2465 		vput(vp);
2466 		if (nd->nd_flag & ND_NFSV3)
2467 			nfsrv_postopattr(nd, getret, &at);
2468 		NFSM_BUILD(tl, u_int32_t *, 4 * NFSX_UNSIGNED);
2469 		txdr_hyper(at.na_filerev, tl);
2470 		tl += 2;
2471 		*tl++ = newnfs_false;
2472 		*tl = newnfs_true;
2473 		free(cookies, M_TEMP);
2474 		free(rbuf, M_TEMP);
2475 		goto out;
2476 	}
2477 
2478 	/*
2479 	 * Check for degenerate cases of nothing useful read.
2480 	 * If so go try again
2481 	 */
2482 	cpos = rbuf;
2483 	cend = rbuf + siz;
2484 	dp = (struct dirent *)cpos;
2485 	cookiep = cookies;
2486 
2487 	/*
2488 	 * For some reason FreeBSD's ufs_readdir() chooses to back the
2489 	 * directory offset up to a block boundary, so it is necessary to
2490 	 * skip over the records that precede the requested offset. This
2491 	 * requires the assumption that file offset cookies monotonically
2492 	 * increase.
2493 	 */
2494 	while (cpos < cend && ncookies > 0 &&
2495 	  (dp->d_fileno == 0 || dp->d_type == DT_WHT ||
2496 	   (is_ufs == 1 && ((u_quad_t)(*cookiep)) <= toff) ||
2497 	   ((nd->nd_flag & ND_NFSV4) &&
2498 	    ((dp->d_namlen == 1 && dp->d_name[0] == '.') ||
2499 	     (dp->d_namlen==2 && dp->d_name[0]=='.' && dp->d_name[1]=='.'))))) {
2500 		cpos += dp->d_reclen;
2501 		dp = (struct dirent *)cpos;
2502 		cookiep++;
2503 		ncookies--;
2504 	}
2505 	if (cpos >= cend || ncookies == 0) {
2506 		siz = fullsiz;
2507 		toff = off;
2508 		goto again;
2509 	}
2510 
2511 	/*
2512 	 * Busy the file system so that the mount point won't go away
2513 	 * and, as such, VFS_VGET() can be used safely.
2514 	 */
2515 	mp = vp->v_mount;
2516 	vfs_ref(mp);
2517 	NFSVOPUNLOCK(vp);
2518 	nd->nd_repstat = vfs_busy(mp, 0);
2519 	vfs_rel(mp);
2520 	if (nd->nd_repstat != 0) {
2521 		vrele(vp);
2522 		free(cookies, M_TEMP);
2523 		free(rbuf, M_TEMP);
2524 		if (nd->nd_flag & ND_NFSV3)
2525 			nfsrv_postopattr(nd, getret, &at);
2526 		goto out;
2527 	}
2528 
2529 	/*
2530 	 * Check to see if entries in this directory can be safely acquired
2531 	 * via VFS_VGET() or if a switch to VOP_LOOKUP() is required.
2532 	 * ZFS snapshot directories need VOP_LOOKUP(), so that any
2533 	 * automount of the snapshot directory that is required will
2534 	 * be done.
2535 	 * This needs to be done here for NFSv4, since NFSv4 never does
2536 	 * a VFS_VGET() for "." or "..".
2537 	 */
2538 	if (is_zfs == 1) {
2539 		r = VFS_VGET(mp, at.na_fileid, LK_SHARED, &nvp);
2540 		if (r == EOPNOTSUPP) {
2541 			usevget = 0;
2542 			cn.cn_nameiop = LOOKUP;
2543 			cn.cn_lkflags = LK_SHARED | LK_RETRY;
2544 			cn.cn_cred = nd->nd_cred;
2545 		} else if (r == 0)
2546 			vput(nvp);
2547 	}
2548 
2549 	/*
2550 	 * If the reply is likely to exceed MCLBYTES and the reply will
2551 	 * not be saved, use ext_pgs mbufs for TLS.
2552 	 * It is difficult to predict how large each entry will be and
2553 	 * how many entries have been read, so just assume the directory
2554 	 * entries grow by a factor of 4 when attributes are included.
2555 	 * For NFSv4.0, we do not know for sure if the reply will
2556 	 * be saved, so do not use ext_pgs mbufs for NFSv4.0.
2557 	 */
2558 	if (cnt > MCLBYTES && siz > MCLBYTES / 4 &&
2559 	    (nd->nd_flag & (ND_TLS | ND_EXTPG | ND_SAVEREPLY)) == ND_TLS &&
2560 	    (nd->nd_flag & (ND_NFSV4 | ND_NFSV41)) != ND_NFSV4)
2561 		nd->nd_flag |= ND_EXTPG;
2562 
2563 	/*
2564 	 * Save this position, in case there is an error before one entry
2565 	 * is created.
2566 	 */
2567 	mb0 = nd->nd_mb;
2568 	bpos0 = nd->nd_bpos;
2569 	bextpg0 = nd->nd_bextpg;
2570 	bextpgsiz0 = nd->nd_bextpgsiz;
2571 
2572 	/*
2573 	 * Fill in the first part of the reply.
2574 	 * dirlen is the reply length in bytes and cannot exceed cnt.
2575 	 * (Include the two booleans at the end of the reply in dirlen now,
2576 	 *  so we recognize when we have exceeded cnt.)
2577 	 */
2578 	if (nd->nd_flag & ND_NFSV3) {
2579 		dirlen = NFSX_V3POSTOPATTR + NFSX_VERF + 2 * NFSX_UNSIGNED;
2580 		nfsrv_postopattr(nd, getret, &at);
2581 	} else {
2582 		dirlen = NFSX_VERF + 2 * NFSX_UNSIGNED;
2583 	}
2584 	NFSM_BUILD(tl, u_int32_t *, NFSX_VERF);
2585 	txdr_hyper(at.na_filerev, tl);
2586 
2587 	/*
2588 	 * Save this position, in case there is an empty reply needed.
2589 	 */
2590 	mb1 = nd->nd_mb;
2591 	bpos1 = nd->nd_bpos;
2592 	bextpg1 = nd->nd_bextpg;
2593 	bextpgsiz1 = nd->nd_bextpgsiz;
2594 
2595 	/* Loop through the records and build reply */
2596 	entrycnt = 0;
2597 	while (cpos < cend && ncookies > 0 && dirlen < cnt) {
2598 		nlen = dp->d_namlen;
2599 		if (dp->d_fileno != 0 && dp->d_type != DT_WHT &&
2600 		    nlen <= NFS_MAXNAMLEN &&
2601 		    ((nd->nd_flag & ND_NFSV3) || nlen > 2 ||
2602 		     (nlen==2 && (dp->d_name[0]!='.' || dp->d_name[1]!='.'))
2603 		      || (nlen == 1 && dp->d_name[0] != '.'))) {
2604 			/*
2605 			 * Save the current position in the reply, in case
2606 			 * this entry exceeds cnt.
2607 			 */
2608 			mb1 = nd->nd_mb;
2609 			bpos1 = nd->nd_bpos;
2610 			bextpg1 = nd->nd_bextpg;
2611 			bextpgsiz1 = nd->nd_bextpgsiz;
2612 
2613 			/*
2614 			 * For readdir_and_lookup get the vnode using
2615 			 * the file number.
2616 			 */
2617 			nvp = NULL;
2618 			refp = NULL;
2619 			r = 0;
2620 			at_root = 0;
2621 			needs_unbusy = 0;
2622 			new_mp = mp;
2623 			mounted_on_fileno = (uint64_t)dp->d_fileno;
2624 			if ((nd->nd_flag & ND_NFSV3) ||
2625 			    NFSNONZERO_ATTRBIT(&savbits) ||
2626 			    dp->d_type == DT_UNKNOWN ||
2627 			    (dp->d_type == DT_DIR &&
2628 			     nfsrv_enable_crossmntpt != 0)) {
2629 				if (nd->nd_flag & ND_NFSV4)
2630 					refp = nfsv4root_getreferral(NULL,
2631 					    vp, dp->d_fileno);
2632 				if (refp == NULL) {
2633 					if (usevget)
2634 						r = VFS_VGET(mp, dp->d_fileno,
2635 						    LK_SHARED, &nvp);
2636 					else
2637 						r = EOPNOTSUPP;
2638 					if (r == EOPNOTSUPP) {
2639 						if (usevget) {
2640 							usevget = 0;
2641 							cn.cn_nameiop = LOOKUP;
2642 							cn.cn_lkflags =
2643 							    LK_SHARED |
2644 							    LK_RETRY;
2645 							cn.cn_cred =
2646 							    nd->nd_cred;
2647 						}
2648 						cn.cn_nameptr = dp->d_name;
2649 						cn.cn_namelen = nlen;
2650 						cn.cn_flags = ISLASTCN |
2651 						    NOFOLLOW | LOCKLEAF;
2652 						if (nlen == 2 &&
2653 						    dp->d_name[0] == '.' &&
2654 						    dp->d_name[1] == '.')
2655 							cn.cn_flags |=
2656 							    ISDOTDOT;
2657 						if (NFSVOPLOCK(vp, LK_SHARED)
2658 						    != 0) {
2659 							nd->nd_repstat = EPERM;
2660 							break;
2661 						}
2662 						if ((vp->v_vflag & VV_ROOT) != 0
2663 						    && (cn.cn_flags & ISDOTDOT)
2664 						    != 0) {
2665 							vref(vp);
2666 							nvp = vp;
2667 							r = 0;
2668 						} else {
2669 							r = VOP_LOOKUP(vp, &nvp,
2670 							    &cn);
2671 							if (vp != nvp)
2672 								NFSVOPUNLOCK(vp);
2673 						}
2674 					}
2675 
2676 					/*
2677 					 * For NFSv4, check to see if nvp is
2678 					 * a mount point and get the mount
2679 					 * point vnode, as required.
2680 					 */
2681 					if (r == 0 &&
2682 					    nfsrv_enable_crossmntpt != 0 &&
2683 					    (nd->nd_flag & ND_NFSV4) != 0 &&
2684 					    nvp->v_type == VDIR &&
2685 					    nvp->v_mountedhere != NULL) {
2686 						new_mp = nvp->v_mountedhere;
2687 						r = vfs_busy(new_mp, 0);
2688 						vput(nvp);
2689 						nvp = NULL;
2690 						if (r == 0) {
2691 							r = VFS_ROOT(new_mp,
2692 							    LK_SHARED, &nvp);
2693 							needs_unbusy = 1;
2694 							if (r == 0)
2695 								at_root = 1;
2696 						}
2697 					}
2698 				}
2699 
2700 				/*
2701 				 * If we failed to look up the entry, then it
2702 				 * has become invalid, most likely removed.
2703 				 */
2704 				if (r != 0) {
2705 					if (needs_unbusy)
2706 						vfs_unbusy(new_mp);
2707 					goto invalid;
2708 				}
2709 				KASSERT(refp != NULL || nvp != NULL,
2710 				    ("%s: undetected lookup error", __func__));
2711 
2712 				if (refp == NULL &&
2713 				    ((nd->nd_flag & ND_NFSV3) ||
2714 				     NFSNONZERO_ATTRBIT(&attrbits))) {
2715 					r = nfsvno_getfh(nvp, &nfh, p);
2716 					if (!r)
2717 					    r = nfsvno_getattr(nvp, nvap, nd, p,
2718 						1, &attrbits);
2719 					if (r == 0 && is_zfs == 1 &&
2720 					    nfsrv_enable_crossmntpt != 0 &&
2721 					    (nd->nd_flag & ND_NFSV4) != 0 &&
2722 					    nvp->v_type == VDIR &&
2723 					    vp->v_mount != nvp->v_mount) {
2724 					    /*
2725 					     * For a ZFS snapshot, there is a
2726 					     * pseudo mount that does not set
2727 					     * v_mountedhere, so it needs to
2728 					     * be detected via a different
2729 					     * mount structure.
2730 					     */
2731 					    at_root = 1;
2732 					    if (new_mp == mp)
2733 						new_mp = nvp->v_mount;
2734 					}
2735 				}
2736 
2737 				/*
2738 				 * If we failed to get attributes of the entry,
2739 				 * then just skip it for NFSv3 (the traditional
2740 				 * behavior in the old NFS server).
2741 				 * For NFSv4 the behavior is controlled by
2742 				 * RDATTRERROR: we either ignore the error or
2743 				 * fail the request.
2744 				 * The exception is EOPNOTSUPP, which can be
2745 				 * returned by nfsvno_getfh() for certain
2746 				 * file systems, such as devfs.  This indicates
2747 				 * that the file system cannot be exported,
2748 				 * so just skip over the entry.
2749 				 * Note that RDATTRERROR is never set for NFSv3.
2750 				 */
2751 				if (r != 0) {
2752 					if (!NFSISSET_ATTRBIT(&attrbits,
2753 					    NFSATTRBIT_RDATTRERROR) ||
2754 					    r == EOPNOTSUPP) {
2755 						vput(nvp);
2756 						if (needs_unbusy != 0)
2757 							vfs_unbusy(new_mp);
2758 						if ((nd->nd_flag & ND_NFSV3) ||
2759 						    r == EOPNOTSUPP)
2760 							goto invalid;
2761 						nd->nd_repstat = r;
2762 						break;
2763 					}
2764 				}
2765 			} else if (NFSNONZERO_ATTRBIT(&attrbits)) {
2766 				/* Only need Type and/or Fileid. */
2767 				VATTR_NULL(&nvap->na_vattr);
2768 				nvap->na_fileid = dp->d_fileno;
2769 				nvap->na_type = NFS_DTYPETOVTYPE(dp->d_type);
2770 			}
2771 
2772 			/*
2773 			 * Build the directory record xdr
2774 			 */
2775 			if (nd->nd_flag & ND_NFSV3) {
2776 				NFSM_BUILD(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
2777 				*tl++ = newnfs_true;
2778 				txdr_hyper(dp->d_fileno, tl);
2779 				dirlen += nfsm_strtom(nd, dp->d_name, nlen);
2780 				NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
2781 				txdr_hyper(*cookiep, tl);
2782 				nfsrv_postopattr(nd, 0, nvap);
2783 				dirlen += nfsm_fhtom(NULL, nd, (u_int8_t *)&nfh,
2784 				    0, 1);
2785 				dirlen += (5*NFSX_UNSIGNED+NFSX_V3POSTOPATTR);
2786 				if (nvp != NULL)
2787 					vput(nvp);
2788 			} else {
2789 				NFSM_BUILD(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
2790 				*tl++ = newnfs_true;
2791 				txdr_hyper(*cookiep, tl);
2792 				dirlen += nfsm_strtom(nd, dp->d_name, nlen);
2793 				if (nvp != NULL) {
2794 					supports_nfsv4acls =
2795 					    nfs_supportsnfsv4acls(nvp);
2796 					NFSVOPUNLOCK(nvp);
2797 				} else
2798 					supports_nfsv4acls = 0;
2799 				if (refp != NULL) {
2800 					dirlen += nfsrv_putreferralattr(nd,
2801 					    &refbits, refp, 0,
2802 					    &nd->nd_repstat);
2803 					if (nd->nd_repstat) {
2804 						if (nvp != NULL)
2805 							vrele(nvp);
2806 						if (needs_unbusy != 0)
2807 							vfs_unbusy(new_mp);
2808 						break;
2809 					}
2810 				} else if (r) {
2811 					dirlen += nfsvno_fillattr(nd, new_mp,
2812 					    nvp, nvap, &nfh, r, &rderrbits,
2813 					    nd->nd_cred, p, isdgram, 0,
2814 					    supports_nfsv4acls, at_root,
2815 					    mounted_on_fileno);
2816 				} else {
2817 					dirlen += nfsvno_fillattr(nd, new_mp,
2818 					    nvp, nvap, &nfh, r, &attrbits,
2819 					    nd->nd_cred, p, isdgram, 0,
2820 					    supports_nfsv4acls, at_root,
2821 					    mounted_on_fileno);
2822 				}
2823 				if (nvp != NULL)
2824 					vrele(nvp);
2825 				dirlen += (3 * NFSX_UNSIGNED);
2826 			}
2827 			if (needs_unbusy != 0)
2828 				vfs_unbusy(new_mp);
2829 			if (dirlen <= cnt)
2830 				entrycnt++;
2831 		}
2832 invalid:
2833 		cpos += dp->d_reclen;
2834 		dp = (struct dirent *)cpos;
2835 		cookiep++;
2836 		ncookies--;
2837 	}
2838 	vrele(vp);
2839 	vfs_unbusy(mp);
2840 
2841 	/*
2842 	 * If dirlen > cnt, we must strip off the last entry. If that
2843 	 * results in an empty reply, report NFSERR_TOOSMALL.
2844 	 */
2845 	if (dirlen > cnt || nd->nd_repstat) {
2846 		if (!nd->nd_repstat && entrycnt == 0)
2847 			nd->nd_repstat = NFSERR_TOOSMALL;
2848 		if (nd->nd_repstat) {
2849 			nfsm_trimtrailing(nd, mb0, bpos0, bextpg0, bextpgsiz0);
2850 			if (nd->nd_flag & ND_NFSV3)
2851 				nfsrv_postopattr(nd, getret, &at);
2852 		} else
2853 			nfsm_trimtrailing(nd, mb1, bpos1, bextpg1, bextpgsiz1);
2854 		eofflag = 0;
2855 	} else if (cpos < cend)
2856 		eofflag = 0;
2857 	if (!nd->nd_repstat) {
2858 		NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
2859 		*tl++ = newnfs_false;
2860 		if (eofflag)
2861 			*tl = newnfs_true;
2862 		else
2863 			*tl = newnfs_false;
2864 	}
2865 	free(cookies, M_TEMP);
2866 	free(rbuf, M_TEMP);
2867 
2868 out:
2869 	NFSEXITCODE2(0, nd);
2870 	return (0);
2871 nfsmout:
2872 	vput(vp);
2873 	NFSEXITCODE2(error, nd);
2874 	return (error);
2875 }
2876 
2877 /*
2878  * Get the settable attributes out of the mbuf list.
2879  * (Return 0 or EBADRPC)
2880  */
2881 int
2882 nfsrv_sattr(struct nfsrv_descript *nd, vnode_t vp, struct nfsvattr *nvap,
2883     nfsattrbit_t *attrbitp, NFSACL_T *aclp, struct thread *p)
2884 {
2885 	u_int32_t *tl;
2886 	struct nfsv2_sattr *sp;
2887 	int error = 0, toclient = 0;
2888 
2889 	switch (nd->nd_flag & (ND_NFSV2 | ND_NFSV3 | ND_NFSV4)) {
2890 	case ND_NFSV2:
2891 		NFSM_DISSECT(sp, struct nfsv2_sattr *, NFSX_V2SATTR);
2892 		/*
2893 		 * Some old clients didn't fill in the high order 16bits.
2894 		 * --> check the low order 2 bytes for 0xffff
2895 		 */
2896 		if ((fxdr_unsigned(int, sp->sa_mode) & 0xffff) != 0xffff)
2897 			nvap->na_mode = nfstov_mode(sp->sa_mode);
2898 		if (sp->sa_uid != newnfs_xdrneg1)
2899 			nvap->na_uid = fxdr_unsigned(uid_t, sp->sa_uid);
2900 		if (sp->sa_gid != newnfs_xdrneg1)
2901 			nvap->na_gid = fxdr_unsigned(gid_t, sp->sa_gid);
2902 		if (sp->sa_size != newnfs_xdrneg1)
2903 			nvap->na_size = fxdr_unsigned(u_quad_t, sp->sa_size);
2904 		if (sp->sa_atime.nfsv2_sec != newnfs_xdrneg1) {
2905 #ifdef notyet
2906 			fxdr_nfsv2time(&sp->sa_atime, &nvap->na_atime);
2907 #else
2908 			nvap->na_atime.tv_sec =
2909 				fxdr_unsigned(u_int32_t,sp->sa_atime.nfsv2_sec);
2910 			nvap->na_atime.tv_nsec = 0;
2911 #endif
2912 		}
2913 		if (sp->sa_mtime.nfsv2_sec != newnfs_xdrneg1)
2914 			fxdr_nfsv2time(&sp->sa_mtime, &nvap->na_mtime);
2915 		break;
2916 	case ND_NFSV3:
2917 		NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2918 		if (*tl == newnfs_true) {
2919 			NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2920 			nvap->na_mode = nfstov_mode(*tl);
2921 		}
2922 		NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2923 		if (*tl == newnfs_true) {
2924 			NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2925 			nvap->na_uid = fxdr_unsigned(uid_t, *tl);
2926 		}
2927 		NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2928 		if (*tl == newnfs_true) {
2929 			NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2930 			nvap->na_gid = fxdr_unsigned(gid_t, *tl);
2931 		}
2932 		NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2933 		if (*tl == newnfs_true) {
2934 			NFSM_DISSECT(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
2935 			nvap->na_size = fxdr_hyper(tl);
2936 		}
2937 		NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2938 		switch (fxdr_unsigned(int, *tl)) {
2939 		case NFSV3SATTRTIME_TOCLIENT:
2940 			NFSM_DISSECT(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
2941 			fxdr_nfsv3time(tl, &nvap->na_atime);
2942 			toclient = 1;
2943 			break;
2944 		case NFSV3SATTRTIME_TOSERVER:
2945 			vfs_timestamp(&nvap->na_atime);
2946 			nvap->na_vaflags |= VA_UTIMES_NULL;
2947 			break;
2948 		}
2949 		NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2950 		switch (fxdr_unsigned(int, *tl)) {
2951 		case NFSV3SATTRTIME_TOCLIENT:
2952 			NFSM_DISSECT(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
2953 			fxdr_nfsv3time(tl, &nvap->na_mtime);
2954 			nvap->na_vaflags &= ~VA_UTIMES_NULL;
2955 			break;
2956 		case NFSV3SATTRTIME_TOSERVER:
2957 			vfs_timestamp(&nvap->na_mtime);
2958 			if (!toclient)
2959 				nvap->na_vaflags |= VA_UTIMES_NULL;
2960 			break;
2961 		}
2962 		break;
2963 	case ND_NFSV4:
2964 		error = nfsv4_sattr(nd, vp, nvap, attrbitp, aclp, p);
2965 	}
2966 nfsmout:
2967 	NFSEXITCODE2(error, nd);
2968 	return (error);
2969 }
2970 
2971 /*
2972  * Handle the setable attributes for V4.
2973  * Returns NFSERR_BADXDR if it can't be parsed, 0 otherwise.
2974  */
2975 int
2976 nfsv4_sattr(struct nfsrv_descript *nd, vnode_t vp, struct nfsvattr *nvap,
2977     nfsattrbit_t *attrbitp, NFSACL_T *aclp, struct thread *p)
2978 {
2979 	u_int32_t *tl;
2980 	int attrsum = 0;
2981 	int i, j;
2982 	int error, attrsize, bitpos, aclsize, aceerr, retnotsup = 0;
2983 	int moderet, toclient = 0;
2984 	u_char *cp, namestr[NFSV4_SMALLSTR + 1];
2985 	uid_t uid;
2986 	gid_t gid;
2987 	u_short mode, mask;		/* Same type as va_mode. */
2988 	struct vattr va;
2989 
2990 	error = nfsrv_getattrbits(nd, attrbitp, NULL, &retnotsup);
2991 	if (error)
2992 		goto nfsmout;
2993 	NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2994 	attrsize = fxdr_unsigned(int, *tl);
2995 
2996 	/*
2997 	 * Loop around getting the setable attributes. If an unsupported
2998 	 * one is found, set nd_repstat == NFSERR_ATTRNOTSUPP and return.
2999 	 */
3000 	if (retnotsup) {
3001 		nd->nd_repstat = NFSERR_ATTRNOTSUPP;
3002 		bitpos = NFSATTRBIT_MAX;
3003 	} else {
3004 		bitpos = 0;
3005 	}
3006 	moderet = 0;
3007 	for (; bitpos < NFSATTRBIT_MAX; bitpos++) {
3008 	    if (attrsum > attrsize) {
3009 		error = NFSERR_BADXDR;
3010 		goto nfsmout;
3011 	    }
3012 	    if (NFSISSET_ATTRBIT(attrbitp, bitpos))
3013 		switch (bitpos) {
3014 		case NFSATTRBIT_SIZE:
3015 			NFSM_DISSECT(tl, u_int32_t *, NFSX_HYPER);
3016                      if (vp != NULL && vp->v_type != VREG) {
3017                             error = (vp->v_type == VDIR) ? NFSERR_ISDIR :
3018                                 NFSERR_INVAL;
3019                             goto nfsmout;
3020 			}
3021 			nvap->na_size = fxdr_hyper(tl);
3022 			attrsum += NFSX_HYPER;
3023 			break;
3024 		case NFSATTRBIT_ACL:
3025 			error = nfsrv_dissectacl(nd, aclp, true, &aceerr,
3026 			    &aclsize, p);
3027 			if (error)
3028 				goto nfsmout;
3029 			if (aceerr && !nd->nd_repstat)
3030 				nd->nd_repstat = aceerr;
3031 			attrsum += aclsize;
3032 			break;
3033 		case NFSATTRBIT_ARCHIVE:
3034 			NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
3035 			if (!nd->nd_repstat)
3036 				nd->nd_repstat = NFSERR_ATTRNOTSUPP;
3037 			attrsum += NFSX_UNSIGNED;
3038 			break;
3039 		case NFSATTRBIT_HIDDEN:
3040 			NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
3041 			if (!nd->nd_repstat)
3042 				nd->nd_repstat = NFSERR_ATTRNOTSUPP;
3043 			attrsum += NFSX_UNSIGNED;
3044 			break;
3045 		case NFSATTRBIT_MIMETYPE:
3046 			NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
3047 			i = fxdr_unsigned(int, *tl);
3048 			error = nfsm_advance(nd, NFSM_RNDUP(i), -1);
3049 			if (error)
3050 				goto nfsmout;
3051 			if (!nd->nd_repstat)
3052 				nd->nd_repstat = NFSERR_ATTRNOTSUPP;
3053 			attrsum += (NFSX_UNSIGNED + NFSM_RNDUP(i));
3054 			break;
3055 		case NFSATTRBIT_MODE:
3056 			moderet = NFSERR_INVAL;	/* Can't do MODESETMASKED. */
3057 			NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
3058 			nvap->na_mode = nfstov_mode(*tl);
3059 			attrsum += NFSX_UNSIGNED;
3060 			break;
3061 		case NFSATTRBIT_OWNER:
3062 			NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
3063 			j = fxdr_unsigned(int, *tl);
3064 			if (j < 0) {
3065 				error = NFSERR_BADXDR;
3066 				goto nfsmout;
3067 			}
3068 			if (j > NFSV4_SMALLSTR)
3069 				cp = malloc(j + 1, M_NFSSTRING, M_WAITOK);
3070 			else
3071 				cp = namestr;
3072 			error = nfsrv_mtostr(nd, cp, j);
3073 			if (error) {
3074 				if (j > NFSV4_SMALLSTR)
3075 					free(cp, M_NFSSTRING);
3076 				goto nfsmout;
3077 			}
3078 			if (!nd->nd_repstat) {
3079 				nd->nd_repstat = nfsv4_strtouid(nd, cp, j,
3080 				    &uid);
3081 				if (!nd->nd_repstat)
3082 					nvap->na_uid = uid;
3083 			}
3084 			if (j > NFSV4_SMALLSTR)
3085 				free(cp, M_NFSSTRING);
3086 			attrsum += (NFSX_UNSIGNED + NFSM_RNDUP(j));
3087 			break;
3088 		case NFSATTRBIT_OWNERGROUP:
3089 			NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
3090 			j = fxdr_unsigned(int, *tl);
3091 			if (j < 0) {
3092 				error = NFSERR_BADXDR;
3093 				goto nfsmout;
3094 			}
3095 			if (j > NFSV4_SMALLSTR)
3096 				cp = malloc(j + 1, M_NFSSTRING, M_WAITOK);
3097 			else
3098 				cp = namestr;
3099 			error = nfsrv_mtostr(nd, cp, j);
3100 			if (error) {
3101 				if (j > NFSV4_SMALLSTR)
3102 					free(cp, M_NFSSTRING);
3103 				goto nfsmout;
3104 			}
3105 			if (!nd->nd_repstat) {
3106 				nd->nd_repstat = nfsv4_strtogid(nd, cp, j,
3107 				    &gid);
3108 				if (!nd->nd_repstat)
3109 					nvap->na_gid = gid;
3110 			}
3111 			if (j > NFSV4_SMALLSTR)
3112 				free(cp, M_NFSSTRING);
3113 			attrsum += (NFSX_UNSIGNED + NFSM_RNDUP(j));
3114 			break;
3115 		case NFSATTRBIT_SYSTEM:
3116 			NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
3117 			if (!nd->nd_repstat)
3118 				nd->nd_repstat = NFSERR_ATTRNOTSUPP;
3119 			attrsum += NFSX_UNSIGNED;
3120 			break;
3121 		case NFSATTRBIT_TIMEACCESSSET:
3122 			NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
3123 			attrsum += NFSX_UNSIGNED;
3124 			if (fxdr_unsigned(int, *tl)==NFSV4SATTRTIME_TOCLIENT) {
3125 			    NFSM_DISSECT(tl, u_int32_t *, NFSX_V4TIME);
3126 			    fxdr_nfsv4time(tl, &nvap->na_atime);
3127 			    toclient = 1;
3128 			    attrsum += NFSX_V4TIME;
3129 			} else {
3130 			    vfs_timestamp(&nvap->na_atime);
3131 			    nvap->na_vaflags |= VA_UTIMES_NULL;
3132 			}
3133 			break;
3134 		case NFSATTRBIT_TIMEBACKUP:
3135 			NFSM_DISSECT(tl, u_int32_t *, NFSX_V4TIME);
3136 			if (!nd->nd_repstat)
3137 				nd->nd_repstat = NFSERR_ATTRNOTSUPP;
3138 			attrsum += NFSX_V4TIME;
3139 			break;
3140 		case NFSATTRBIT_TIMECREATE:
3141 			NFSM_DISSECT(tl, u_int32_t *, NFSX_V4TIME);
3142 			fxdr_nfsv4time(tl, &nvap->na_btime);
3143 			attrsum += NFSX_V4TIME;
3144 			break;
3145 		case NFSATTRBIT_TIMEMODIFYSET:
3146 			NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
3147 			attrsum += NFSX_UNSIGNED;
3148 			if (fxdr_unsigned(int, *tl)==NFSV4SATTRTIME_TOCLIENT) {
3149 			    NFSM_DISSECT(tl, u_int32_t *, NFSX_V4TIME);
3150 			    fxdr_nfsv4time(tl, &nvap->na_mtime);
3151 			    nvap->na_vaflags &= ~VA_UTIMES_NULL;
3152 			    attrsum += NFSX_V4TIME;
3153 			} else {
3154 			    vfs_timestamp(&nvap->na_mtime);
3155 			    if (!toclient)
3156 				nvap->na_vaflags |= VA_UTIMES_NULL;
3157 			}
3158 			break;
3159 		case NFSATTRBIT_MODESETMASKED:
3160 			NFSM_DISSECT(tl, uint32_t *, 2 * NFSX_UNSIGNED);
3161 			mode = fxdr_unsigned(u_short, *tl++);
3162 			mask = fxdr_unsigned(u_short, *tl);
3163 			/*
3164 			 * vp == NULL implies an Open/Create operation.
3165 			 * This attribute can only be used for Setattr and
3166 			 * only for NFSv4.1 or higher.
3167 			 * If moderet != 0, a mode attribute has also been
3168 			 * specified and this attribute cannot be done in the
3169 			 * same Setattr operation.
3170 			 */
3171 			if ((nd->nd_flag & ND_NFSV41) == 0)
3172 				nd->nd_repstat = NFSERR_ATTRNOTSUPP;
3173 			else if ((mode & ~07777) != 0 || (mask & ~07777) != 0 ||
3174 			    vp == NULL)
3175 				nd->nd_repstat = NFSERR_INVAL;
3176 			else if (moderet == 0)
3177 				moderet = VOP_GETATTR(vp, &va, nd->nd_cred);
3178 			if (moderet == 0)
3179 				nvap->na_mode = (mode & mask) |
3180 				    (va.va_mode & ~mask);
3181 			else
3182 				nd->nd_repstat = moderet;
3183 			attrsum += 2 * NFSX_UNSIGNED;
3184 			break;
3185 		default:
3186 			nd->nd_repstat = NFSERR_ATTRNOTSUPP;
3187 			/*
3188 			 * set bitpos so we drop out of the loop.
3189 			 */
3190 			bitpos = NFSATTRBIT_MAX;
3191 			break;
3192 		}
3193 	}
3194 
3195 	/*
3196 	 * some clients pad the attrlist, so we need to skip over the
3197 	 * padding.
3198 	 */
3199 	if (attrsum > attrsize) {
3200 		error = NFSERR_BADXDR;
3201 	} else {
3202 		attrsize = NFSM_RNDUP(attrsize);
3203 		if (attrsum < attrsize)
3204 			error = nfsm_advance(nd, attrsize - attrsum, -1);
3205 	}
3206 nfsmout:
3207 	NFSEXITCODE2(error, nd);
3208 	return (error);
3209 }
3210 
3211 /*
3212  * Check/setup export credentials.
3213  */
3214 int
3215 nfsd_excred(struct nfsrv_descript *nd, struct nfsexstuff *exp,
3216     struct ucred *credanon, bool testsec)
3217 {
3218 	int error;
3219 
3220 	/*
3221 	 * Check/setup credentials.
3222 	 */
3223 	if (nd->nd_flag & ND_GSS)
3224 		exp->nes_exflag &= ~MNT_EXPORTANON;
3225 
3226 	/*
3227 	 * Check to see if the operation is allowed for this security flavor.
3228 	 */
3229 	error = 0;
3230 	if (testsec) {
3231 		error = nfsvno_testexp(nd, exp);
3232 		if (error != 0)
3233 			goto out;
3234 	}
3235 
3236 	/*
3237 	 * Check to see if the file system is exported V4 only.
3238 	 */
3239 	if (NFSVNO_EXV4ONLY(exp) && !(nd->nd_flag & ND_NFSV4)) {
3240 		error = NFSERR_PROGNOTV4;
3241 		goto out;
3242 	}
3243 
3244 	/*
3245 	 * Now, map the user credentials.
3246 	 * (Note that ND_AUTHNONE will only be set for an NFSv3
3247 	 *  Fsinfo RPC. If set for anything else, this code might need
3248 	 *  to change.)
3249 	 */
3250 	if (NFSVNO_EXPORTED(exp)) {
3251 		if (((nd->nd_flag & ND_GSS) == 0 && nd->nd_cred->cr_uid == 0) ||
3252 		     NFSVNO_EXPORTANON(exp) ||
3253 		     (nd->nd_flag & ND_AUTHNONE) != 0) {
3254 			nd->nd_cred->cr_uid = credanon->cr_uid;
3255 			nd->nd_cred->cr_gid = credanon->cr_gid;
3256 			crsetgroups(nd->nd_cred, credanon->cr_ngroups,
3257 			    credanon->cr_groups);
3258 		} else if ((nd->nd_flag & ND_GSS) == 0) {
3259 			/*
3260 			 * If using AUTH_SYS, call nfsrv_getgrpscred() to see
3261 			 * if there is a replacement credential with a group
3262 			 * list set up by "nfsuserd -manage-gids".
3263 			 * If there is no replacement, nfsrv_getgrpscred()
3264 			 * simply returns its argument.
3265 			 */
3266 			nd->nd_cred = nfsrv_getgrpscred(nd->nd_cred);
3267 		}
3268 	}
3269 
3270 out:
3271 	NFSEXITCODE2(error, nd);
3272 	return (error);
3273 }
3274 
3275 /*
3276  * Check exports.
3277  */
3278 int
3279 nfsvno_checkexp(struct mount *mp, struct sockaddr *nam, struct nfsexstuff *exp,
3280     struct ucred **credp)
3281 {
3282 	int error;
3283 
3284 	error = 0;
3285 	*credp = NULL;
3286 	MNT_ILOCK(mp);
3287 	if (mp->mnt_exjail == NULL ||
3288 	    mp->mnt_exjail->cr_prison != curthread->td_ucred->cr_prison)
3289 		error = EACCES;
3290 	MNT_IUNLOCK(mp);
3291 	if (error == 0)
3292 		error = VFS_CHECKEXP(mp, nam, &exp->nes_exflag, credp,
3293 		    &exp->nes_numsecflavor, exp->nes_secflavors);
3294 	if (error) {
3295 		if (NFSD_VNET(nfs_rootfhset)) {
3296 			exp->nes_exflag = 0;
3297 			exp->nes_numsecflavor = 0;
3298 			error = 0;
3299 		}
3300 	} else if (exp->nes_numsecflavor < 1 || exp->nes_numsecflavor >
3301 	    MAXSECFLAVORS) {
3302 		printf("nfsvno_checkexp: numsecflavors out of range\n");
3303 		exp->nes_numsecflavor = 0;
3304 		error = EACCES;
3305 	}
3306 	NFSEXITCODE(error);
3307 	return (error);
3308 }
3309 
3310 /*
3311  * Get a vnode for a file handle and export stuff.
3312  */
3313 int
3314 nfsvno_fhtovp(struct mount *mp, fhandle_t *fhp, struct sockaddr *nam,
3315     int lktype, struct vnode **vpp, struct nfsexstuff *exp,
3316     struct ucred **credp)
3317 {
3318 	int error;
3319 
3320 	*credp = NULL;
3321 	exp->nes_numsecflavor = 0;
3322 	error = VFS_FHTOVP(mp, &fhp->fh_fid, lktype, vpp);
3323 	if (error != 0)
3324 		/* Make sure the server replies ESTALE to the client. */
3325 		error = ESTALE;
3326 	if (nam && !error) {
3327 		MNT_ILOCK(mp);
3328 		if (mp->mnt_exjail == NULL ||
3329 		    mp->mnt_exjail->cr_prison != curthread->td_ucred->cr_prison)
3330 			error = EACCES;
3331 		MNT_IUNLOCK(mp);
3332 		if (error == 0)
3333 			error = VFS_CHECKEXP(mp, nam, &exp->nes_exflag, credp,
3334 			    &exp->nes_numsecflavor, exp->nes_secflavors);
3335 		if (error) {
3336 			if (NFSD_VNET(nfs_rootfhset)) {
3337 				exp->nes_exflag = 0;
3338 				exp->nes_numsecflavor = 0;
3339 				error = 0;
3340 			} else {
3341 				vput(*vpp);
3342 			}
3343 		} else if (exp->nes_numsecflavor < 1 || exp->nes_numsecflavor >
3344 		    MAXSECFLAVORS) {
3345 			printf("nfsvno_fhtovp: numsecflavors out of range\n");
3346 			exp->nes_numsecflavor = 0;
3347 			error = EACCES;
3348 			vput(*vpp);
3349 		}
3350 	}
3351 	NFSEXITCODE(error);
3352 	return (error);
3353 }
3354 
3355 /*
3356  * nfsd_fhtovp() - convert a fh to a vnode ptr
3357  * 	- look up fsid in mount list (if not found ret error)
3358  *	- get vp and export rights by calling nfsvno_fhtovp()
3359  *	- if cred->cr_uid == 0 or MNT_EXPORTANON set it to credanon
3360  *	  for AUTH_SYS
3361  *	- if mpp != NULL, return the mount point so that it can
3362  *	  be used for vn_finished_write() by the caller
3363  */
3364 void
3365 nfsd_fhtovp(struct nfsrv_descript *nd, struct nfsrvfh *nfp, int lktype,
3366     struct vnode **vpp, struct nfsexstuff *exp,
3367     struct mount **mpp, int startwrite, int nextop)
3368 {
3369 	struct mount *mp, *mpw;
3370 	struct ucred *credanon;
3371 	fhandle_t *fhp;
3372 	int error;
3373 
3374 	if (mpp != NULL)
3375 		*mpp = NULL;
3376 	*vpp = NULL;
3377 	fhp = (fhandle_t *)nfp->nfsrvfh_data;
3378 	mp = vfs_busyfs(&fhp->fh_fsid);
3379 	if (mp == NULL) {
3380 		nd->nd_repstat = ESTALE;
3381 		goto out;
3382 	}
3383 
3384 	if (startwrite) {
3385 		mpw = mp;
3386 		error = vn_start_write(NULL, &mpw, V_WAIT);
3387 		if (error != 0) {
3388 			mpw = NULL;
3389 			vfs_unbusy(mp);
3390 			nd->nd_repstat = ESTALE;
3391 			goto out;
3392 		}
3393 		if (lktype == LK_SHARED && !(MNT_SHARED_WRITES(mp)))
3394 			lktype = LK_EXCLUSIVE;
3395 	} else
3396 		mpw = NULL;
3397 
3398 	nd->nd_repstat = nfsvno_fhtovp(mp, fhp, nd->nd_nam, lktype, vpp, exp,
3399 	    &credanon);
3400 	vfs_unbusy(mp);
3401 
3402 	/*
3403 	 * For NFSv4 without a pseudo root fs, unexported file handles
3404 	 * can be returned, so that Lookup works everywhere.
3405 	 */
3406 	if (!nd->nd_repstat && exp->nes_exflag == 0 &&
3407 	    !(nd->nd_flag & ND_NFSV4)) {
3408 		vput(*vpp);
3409 		*vpp = NULL;
3410 		nd->nd_repstat = EACCES;
3411 	}
3412 
3413 	/*
3414 	 * Personally, I've never seen any point in requiring a
3415 	 * reserved port#, since only in the rare case where the
3416 	 * clients are all boxes with secure system privileges,
3417 	 * does it provide any enhanced security, but... some people
3418 	 * believe it to be useful and keep putting this code back in.
3419 	 * (There is also some "security checker" out there that
3420 	 *  complains if the nfs server doesn't enforce this.)
3421 	 * However, note the following:
3422 	 * RFC3530 (NFSv4) specifies that a reserved port# not be
3423 	 *	required.
3424 	 * RFC2623 recommends that, if a reserved port# is checked for,
3425 	 *	that there be a way to turn that off--> ifdef'd.
3426 	 */
3427 #ifdef NFS_REQRSVPORT
3428 	if (!nd->nd_repstat) {
3429 		struct sockaddr_in *saddr;
3430 		struct sockaddr_in6 *saddr6;
3431 
3432 		saddr = NFSSOCKADDR(nd->nd_nam, struct sockaddr_in *);
3433 		saddr6 = NFSSOCKADDR(nd->nd_nam, struct sockaddr_in6 *);
3434 		if (!(nd->nd_flag & ND_NFSV4) &&
3435 		    ((saddr->sin_family == AF_INET &&
3436 		      ntohs(saddr->sin_port) >= IPPORT_RESERVED) ||
3437 		     (saddr6->sin6_family == AF_INET6 &&
3438 		      ntohs(saddr6->sin6_port) >= IPPORT_RESERVED))) {
3439 			vput(*vpp);
3440 			nd->nd_repstat = (NFSERR_AUTHERR | AUTH_TOOWEAK);
3441 		}
3442 	}
3443 #endif	/* NFS_REQRSVPORT */
3444 
3445 	/*
3446 	 * Check/setup credentials.
3447 	 */
3448 	if (!nd->nd_repstat) {
3449 		nd->nd_saveduid = nd->nd_cred->cr_uid;
3450 		nd->nd_repstat = nfsd_excred(nd, exp, credanon,
3451 		    nfsrv_checkwrongsec(nd, nextop, (*vpp)->v_type));
3452 		if (nd->nd_repstat)
3453 			vput(*vpp);
3454 	}
3455 	if (credanon != NULL)
3456 		crfree(credanon);
3457 	if (nd->nd_repstat) {
3458 		vn_finished_write(mpw);
3459 		*vpp = NULL;
3460 	} else if (mpp != NULL) {
3461 		*mpp = mpw;
3462 	}
3463 
3464 out:
3465 	NFSEXITCODE2(0, nd);
3466 }
3467 
3468 /*
3469  * glue for fp.
3470  */
3471 static int
3472 fp_getfvp(struct thread *p, int fd, struct file **fpp, struct vnode **vpp)
3473 {
3474 	struct filedesc *fdp;
3475 	struct file *fp;
3476 	int error = 0;
3477 
3478 	fdp = p->td_proc->p_fd;
3479 	if (fd < 0 || fd >= fdp->fd_nfiles ||
3480 	    (fp = fdp->fd_ofiles[fd].fde_file) == NULL) {
3481 		error = EBADF;
3482 		goto out;
3483 	}
3484 	*fpp = fp;
3485 
3486 out:
3487 	NFSEXITCODE(error);
3488 	return (error);
3489 }
3490 
3491 /*
3492  * Called from nfssvc() to update the exports list. Just call
3493  * vfs_export(). This has to be done, since the v4 root fake fs isn't
3494  * in the mount list.
3495  */
3496 int
3497 nfsrv_v4rootexport(void *argp, struct ucred *cred, struct thread *p)
3498 {
3499 	struct nfsex_args *nfsexargp = (struct nfsex_args *)argp;
3500 	int error = 0;
3501 	struct nameidata nd;
3502 	fhandle_t fh;
3503 
3504 	error = vfs_export(NFSD_VNET(nfsv4root_mnt), &nfsexargp->export, false);
3505 	if ((nfsexargp->export.ex_flags & MNT_DELEXPORT) != 0)
3506 		NFSD_VNET(nfs_rootfhset) = 0;
3507 	else if (error == 0) {
3508 		if (nfsexargp->fspec == NULL) {
3509 			error = EPERM;
3510 			goto out;
3511 		}
3512 		/*
3513 		 * If fspec != NULL, this is the v4root path.
3514 		 */
3515 		NDINIT(&nd, LOOKUP, FOLLOW, UIO_USERSPACE, nfsexargp->fspec);
3516 		if ((error = namei(&nd)) != 0)
3517 			goto out;
3518 		error = nfsvno_getfh(nd.ni_vp, &fh, p);
3519 		vrele(nd.ni_vp);
3520 		if (!error) {
3521 			NFSD_VNET(nfs_rootfh).nfsrvfh_len = NFSX_MYFH;
3522 			NFSBCOPY((caddr_t)&fh,
3523 			    NFSD_VNET(nfs_rootfh).nfsrvfh_data,
3524 			    sizeof (fhandle_t));
3525 			NFSD_VNET(nfs_rootfhset) = 1;
3526 		}
3527 	}
3528 
3529 out:
3530 	NFSEXITCODE(error);
3531 	return (error);
3532 }
3533 
3534 /*
3535  * This function needs to test to see if the system is near its limit
3536  * for memory allocation via malloc() or mget() and return True iff
3537  * either of these resources are near their limit.
3538  * XXX (For now, this is just a stub.)
3539  */
3540 int nfsrv_testmalloclimit = 0;
3541 int
3542 nfsrv_mallocmget_limit(void)
3543 {
3544 	static int printmesg = 0;
3545 	static int testval = 1;
3546 
3547 	if (nfsrv_testmalloclimit && (testval++ % 1000) == 0) {
3548 		if ((printmesg++ % 100) == 0)
3549 			printf("nfsd: malloc/mget near limit\n");
3550 		return (1);
3551 	}
3552 	return (0);
3553 }
3554 
3555 /*
3556  * BSD specific initialization of a mount point.
3557  */
3558 void
3559 nfsd_mntinit(void)
3560 {
3561 
3562 	NFSD_LOCK();
3563 	if (NFSD_VNET(nfsrv_mntinited)) {
3564 		NFSD_UNLOCK();
3565 		return;
3566 	}
3567 	NFSD_VNET(nfsrv_mntinited) = true;
3568 	nfsrvd_init(0);
3569 	NFSD_UNLOCK();
3570 
3571 	NFSD_VNET(nfsv4root_mnt) = malloc(sizeof(struct mount), M_TEMP,
3572 	    M_WAITOK | M_ZERO);
3573 	NFSD_VNET(nfsv4root_mnt)->mnt_flag = (MNT_RDONLY | MNT_EXPORTED);
3574 	mtx_init(&NFSD_VNET(nfsv4root_mnt)->mnt_mtx, "nfs4mnt", NULL, MTX_DEF);
3575 	lockinit(&NFSD_VNET(nfsv4root_mnt)->mnt_explock, PVFS, "explock", 0, 0);
3576 	TAILQ_INIT(&NFSD_VNET(nfsv4root_mnt)->mnt_nvnodelist);
3577 	TAILQ_INIT(&NFSD_VNET(nfsv4root_mnt)->mnt_lazyvnodelist);
3578 	NFSD_VNET(nfsv4root_mnt)->mnt_export = NULL;
3579 	TAILQ_INIT(&NFSD_VNET(nfsv4root_opt));
3580 	TAILQ_INIT(&NFSD_VNET(nfsv4root_newopt));
3581 	NFSD_VNET(nfsv4root_mnt)->mnt_opt = &NFSD_VNET(nfsv4root_opt);
3582 	NFSD_VNET(nfsv4root_mnt)->mnt_optnew = &NFSD_VNET(nfsv4root_newopt);
3583 	NFSD_VNET(nfsv4root_mnt)->mnt_nvnodelistsize = 0;
3584 	NFSD_VNET(nfsv4root_mnt)->mnt_lazyvnodelistsize = 0;
3585 	callout_init(&NFSD_VNET(nfsd_callout), 1);
3586 
3587 	nfsrvd_initcache();
3588 	nfsd_init();
3589 }
3590 
3591 static void
3592 nfsd_timer(void *arg)
3593 {
3594 	struct vnet *vnetp;
3595 
3596 	vnetp = (struct vnet *)arg;
3597 	NFSD_CURVNET_SET_QUIET(vnetp);
3598 	nfsrv_servertimer(vnetp);
3599 	callout_reset_sbt(&NFSD_VNET(nfsd_callout), SBT_1S, SBT_1S, nfsd_timer,
3600 	    arg, 0);
3601 	NFSD_CURVNET_RESTORE();
3602 }
3603 
3604 /*
3605  * Get a vnode for a file handle, without checking exports, etc.
3606  */
3607 struct vnode *
3608 nfsvno_getvp(fhandle_t *fhp)
3609 {
3610 	struct mount *mp;
3611 	struct vnode *vp;
3612 	int error;
3613 
3614 	mp = vfs_busyfs(&fhp->fh_fsid);
3615 	if (mp == NULL)
3616 		return (NULL);
3617 	error = VFS_FHTOVP(mp, &fhp->fh_fid, LK_EXCLUSIVE, &vp);
3618 	vfs_unbusy(mp);
3619 	if (error)
3620 		return (NULL);
3621 	return (vp);
3622 }
3623 
3624 /*
3625  * Do a local VOP_ADVLOCK().
3626  */
3627 int
3628 nfsvno_advlock(struct vnode *vp, int ftype, u_int64_t first,
3629     u_int64_t end, struct thread *td)
3630 {
3631 	int error = 0;
3632 	struct flock fl;
3633 	u_int64_t tlen;
3634 
3635 	if (nfsrv_dolocallocks == 0)
3636 		goto out;
3637 	ASSERT_VOP_UNLOCKED(vp, "nfsvno_advlock: vp locked");
3638 
3639 	fl.l_whence = SEEK_SET;
3640 	fl.l_type = ftype;
3641 	fl.l_start = (off_t)first;
3642 	if (end == NFS64BITSSET) {
3643 		fl.l_len = 0;
3644 	} else {
3645 		tlen = end - first;
3646 		fl.l_len = (off_t)tlen;
3647 	}
3648 	/*
3649 	 * For FreeBSD8, the l_pid and l_sysid must be set to the same
3650 	 * values for all calls, so that all locks will be held by the
3651 	 * nfsd server. (The nfsd server handles conflicts between the
3652 	 * various clients.)
3653 	 * Since an NFSv4 lockowner is a ClientID plus an array of up to 1024
3654 	 * bytes, so it can't be put in l_sysid.
3655 	 */
3656 	if (nfsv4_sysid == 0)
3657 		nfsv4_sysid = nlm_acquire_next_sysid();
3658 	fl.l_pid = (pid_t)0;
3659 	fl.l_sysid = (int)nfsv4_sysid;
3660 
3661 	if (ftype == F_UNLCK)
3662 		error = VOP_ADVLOCK(vp, (caddr_t)td->td_proc, F_UNLCK, &fl,
3663 		    (F_POSIX | F_REMOTE));
3664 	else
3665 		error = VOP_ADVLOCK(vp, (caddr_t)td->td_proc, F_SETLK, &fl,
3666 		    (F_POSIX | F_REMOTE));
3667 
3668 out:
3669 	NFSEXITCODE(error);
3670 	return (error);
3671 }
3672 
3673 /*
3674  * Check the nfsv4 root exports.
3675  */
3676 int
3677 nfsvno_v4rootexport(struct nfsrv_descript *nd)
3678 {
3679 	struct ucred *credanon;
3680 	int error = 0, numsecflavor, secflavors[MAXSECFLAVORS], i;
3681 	uint64_t exflags;
3682 
3683 	error = vfs_stdcheckexp(NFSD_VNET(nfsv4root_mnt), nd->nd_nam, &exflags,
3684 	    &credanon, &numsecflavor, secflavors);
3685 	if (error) {
3686 		error = NFSERR_PROGUNAVAIL;
3687 		goto out;
3688 	}
3689 	if (credanon != NULL)
3690 		crfree(credanon);
3691 	for (i = 0; i < numsecflavor; i++) {
3692 		if (secflavors[i] == AUTH_SYS)
3693 			nd->nd_flag |= ND_EXAUTHSYS;
3694 		else if (secflavors[i] == RPCSEC_GSS_KRB5)
3695 			nd->nd_flag |= ND_EXGSS;
3696 		else if (secflavors[i] == RPCSEC_GSS_KRB5I)
3697 			nd->nd_flag |= ND_EXGSSINTEGRITY;
3698 		else if (secflavors[i] == RPCSEC_GSS_KRB5P)
3699 			nd->nd_flag |= ND_EXGSSPRIVACY;
3700 	}
3701 
3702 	/* And set ND_EXxx flags for TLS. */
3703 	if ((exflags & MNT_EXTLS) != 0) {
3704 		nd->nd_flag |= ND_EXTLS;
3705 		if ((exflags & MNT_EXTLSCERT) != 0)
3706 			nd->nd_flag |= ND_EXTLSCERT;
3707 		if ((exflags & MNT_EXTLSCERTUSER) != 0)
3708 			nd->nd_flag |= ND_EXTLSCERTUSER;
3709 	}
3710 
3711 out:
3712 	NFSEXITCODE(error);
3713 	return (error);
3714 }
3715 
3716 /*
3717  * Nfs server pseudo system call for the nfsd's
3718  */
3719 /*
3720  * MPSAFE
3721  */
3722 static int
3723 nfssvc_nfsd(struct thread *td, struct nfssvc_args *uap)
3724 {
3725 	struct file *fp;
3726 	struct nfsd_addsock_args sockarg;
3727 	struct nfsd_nfsd_args nfsdarg;
3728 	struct nfsd_nfsd_oargs onfsdarg;
3729 	struct nfsd_pnfsd_args pnfsdarg;
3730 	struct vnode *vp, *nvp, *curdvp;
3731 	struct pnfsdsfile *pf;
3732 	struct nfsdevice *ds, *fds;
3733 	cap_rights_t rights;
3734 	int buflen, error, ret;
3735 	char *buf, *cp, *cp2, *cp3;
3736 	char fname[PNFS_FILENAME_LEN + 1];
3737 
3738 	NFSD_CURVNET_SET(NFSD_TD_TO_VNET(td));
3739 	if (uap->flag & NFSSVC_NFSDADDSOCK) {
3740 		error = copyin(uap->argp, (caddr_t)&sockarg, sizeof (sockarg));
3741 		if (error)
3742 			goto out;
3743 		/*
3744 		 * Since we don't know what rights might be required,
3745 		 * pretend that we need them all. It is better to be too
3746 		 * careful than too reckless.
3747 		 */
3748 		error = fget(td, sockarg.sock,
3749 		    cap_rights_init_one(&rights, CAP_SOCK_SERVER), &fp);
3750 		if (error != 0)
3751 			goto out;
3752 		if (fp->f_type != DTYPE_SOCKET) {
3753 			fdrop(fp, td);
3754 			error = EPERM;
3755 			goto out;
3756 		}
3757 		error = nfsrvd_addsock(fp);
3758 		fdrop(fp, td);
3759 	} else if (uap->flag & NFSSVC_NFSDNFSD) {
3760 		if (uap->argp == NULL) {
3761 			error = EINVAL;
3762 			goto out;
3763 		}
3764 		if ((uap->flag & NFSSVC_NEWSTRUCT) == 0) {
3765 			error = copyin(uap->argp, &onfsdarg, sizeof(onfsdarg));
3766 			if (error == 0) {
3767 				nfsdarg.principal = onfsdarg.principal;
3768 				nfsdarg.minthreads = onfsdarg.minthreads;
3769 				nfsdarg.maxthreads = onfsdarg.maxthreads;
3770 				nfsdarg.version = 1;
3771 				nfsdarg.addr = NULL;
3772 				nfsdarg.addrlen = 0;
3773 				nfsdarg.dnshost = NULL;
3774 				nfsdarg.dnshostlen = 0;
3775 				nfsdarg.dspath = NULL;
3776 				nfsdarg.dspathlen = 0;
3777 				nfsdarg.mdspath = NULL;
3778 				nfsdarg.mdspathlen = 0;
3779 				nfsdarg.mirrorcnt = 1;
3780 			}
3781 		} else
3782 			error = copyin(uap->argp, &nfsdarg, sizeof(nfsdarg));
3783 		if (error)
3784 			goto out;
3785 		if (nfsdarg.addrlen > 0 && nfsdarg.addrlen < 10000 &&
3786 		    nfsdarg.dnshostlen > 0 && nfsdarg.dnshostlen < 10000 &&
3787 		    nfsdarg.dspathlen > 0 && nfsdarg.dspathlen < 10000 &&
3788 		    nfsdarg.mdspathlen > 0 && nfsdarg.mdspathlen < 10000 &&
3789 		    nfsdarg.mirrorcnt >= 1 &&
3790 		    nfsdarg.mirrorcnt <= NFSDEV_MAXMIRRORS &&
3791 		    nfsdarg.addr != NULL && nfsdarg.dnshost != NULL &&
3792 		    nfsdarg.dspath != NULL && nfsdarg.mdspath != NULL) {
3793 			NFSD_DEBUG(1, "addrlen=%d dspathlen=%d dnslen=%d"
3794 			    " mdspathlen=%d mirrorcnt=%d\n", nfsdarg.addrlen,
3795 			    nfsdarg.dspathlen, nfsdarg.dnshostlen,
3796 			    nfsdarg.mdspathlen, nfsdarg.mirrorcnt);
3797 			cp = malloc(nfsdarg.addrlen + 1, M_TEMP, M_WAITOK);
3798 			error = copyin(nfsdarg.addr, cp, nfsdarg.addrlen);
3799 			if (error != 0) {
3800 				free(cp, M_TEMP);
3801 				goto out;
3802 			}
3803 			cp[nfsdarg.addrlen] = '\0';	/* Ensure nul term. */
3804 			nfsdarg.addr = cp;
3805 			cp = malloc(nfsdarg.dnshostlen + 1, M_TEMP, M_WAITOK);
3806 			error = copyin(nfsdarg.dnshost, cp, nfsdarg.dnshostlen);
3807 			if (error != 0) {
3808 				free(nfsdarg.addr, M_TEMP);
3809 				free(cp, M_TEMP);
3810 				goto out;
3811 			}
3812 			cp[nfsdarg.dnshostlen] = '\0';	/* Ensure nul term. */
3813 			nfsdarg.dnshost = cp;
3814 			cp = malloc(nfsdarg.dspathlen + 1, M_TEMP, M_WAITOK);
3815 			error = copyin(nfsdarg.dspath, cp, nfsdarg.dspathlen);
3816 			if (error != 0) {
3817 				free(nfsdarg.addr, M_TEMP);
3818 				free(nfsdarg.dnshost, M_TEMP);
3819 				free(cp, M_TEMP);
3820 				goto out;
3821 			}
3822 			cp[nfsdarg.dspathlen] = '\0';	/* Ensure nul term. */
3823 			nfsdarg.dspath = cp;
3824 			cp = malloc(nfsdarg.mdspathlen + 1, M_TEMP, M_WAITOK);
3825 			error = copyin(nfsdarg.mdspath, cp, nfsdarg.mdspathlen);
3826 			if (error != 0) {
3827 				free(nfsdarg.addr, M_TEMP);
3828 				free(nfsdarg.dnshost, M_TEMP);
3829 				free(nfsdarg.dspath, M_TEMP);
3830 				free(cp, M_TEMP);
3831 				goto out;
3832 			}
3833 			cp[nfsdarg.mdspathlen] = '\0';	/* Ensure nul term. */
3834 			nfsdarg.mdspath = cp;
3835 		} else {
3836 			nfsdarg.addr = NULL;
3837 			nfsdarg.addrlen = 0;
3838 			nfsdarg.dnshost = NULL;
3839 			nfsdarg.dnshostlen = 0;
3840 			nfsdarg.dspath = NULL;
3841 			nfsdarg.dspathlen = 0;
3842 			nfsdarg.mdspath = NULL;
3843 			nfsdarg.mdspathlen = 0;
3844 			nfsdarg.mirrorcnt = 1;
3845 		}
3846 		nfsd_timer(NFSD_TD_TO_VNET(td));
3847 		error = nfsrvd_nfsd(td, &nfsdarg);
3848 		callout_drain(&NFSD_VNET(nfsd_callout));
3849 		free(nfsdarg.addr, M_TEMP);
3850 		free(nfsdarg.dnshost, M_TEMP);
3851 		free(nfsdarg.dspath, M_TEMP);
3852 		free(nfsdarg.mdspath, M_TEMP);
3853 	} else if (uap->flag & NFSSVC_PNFSDS) {
3854 		error = copyin(uap->argp, &pnfsdarg, sizeof(pnfsdarg));
3855 		if (error == 0 && (pnfsdarg.op == PNFSDOP_DELDSSERVER ||
3856 		    pnfsdarg.op == PNFSDOP_FORCEDELDS)) {
3857 			cp = malloc(PATH_MAX + 1, M_TEMP, M_WAITOK);
3858 			error = copyinstr(pnfsdarg.dspath, cp, PATH_MAX + 1,
3859 			    NULL);
3860 			if (error == 0)
3861 				error = nfsrv_deldsserver(pnfsdarg.op, cp, td);
3862 			free(cp, M_TEMP);
3863 		} else if (error == 0 && pnfsdarg.op == PNFSDOP_COPYMR) {
3864 			cp = malloc(PATH_MAX + 1, M_TEMP, M_WAITOK);
3865 			buflen = sizeof(*pf) * NFSDEV_MAXMIRRORS;
3866 			buf = malloc(buflen, M_TEMP, M_WAITOK);
3867 			error = copyinstr(pnfsdarg.mdspath, cp, PATH_MAX + 1,
3868 			    NULL);
3869 			NFSD_DEBUG(4, "pnfsdcopymr cp mdspath=%d\n", error);
3870 			if (error == 0 && pnfsdarg.dspath != NULL) {
3871 				cp2 = malloc(PATH_MAX + 1, M_TEMP, M_WAITOK);
3872 				error = copyinstr(pnfsdarg.dspath, cp2,
3873 				    PATH_MAX + 1, NULL);
3874 				NFSD_DEBUG(4, "pnfsdcopymr cp dspath=%d\n",
3875 				    error);
3876 			} else
3877 				cp2 = NULL;
3878 			if (error == 0 && pnfsdarg.curdspath != NULL) {
3879 				cp3 = malloc(PATH_MAX + 1, M_TEMP, M_WAITOK);
3880 				error = copyinstr(pnfsdarg.curdspath, cp3,
3881 				    PATH_MAX + 1, NULL);
3882 				NFSD_DEBUG(4, "pnfsdcopymr cp curdspath=%d\n",
3883 				    error);
3884 			} else
3885 				cp3 = NULL;
3886 			curdvp = NULL;
3887 			fds = NULL;
3888 			if (error == 0)
3889 				error = nfsrv_mdscopymr(cp, cp2, cp3, buf,
3890 				    &buflen, fname, td, &vp, &nvp, &pf, &ds,
3891 				    &fds);
3892 			NFSD_DEBUG(4, "nfsrv_mdscopymr=%d\n", error);
3893 			if (error == 0) {
3894 				if (pf->dsf_dir >= nfsrv_dsdirsize) {
3895 					printf("copymr: dsdir out of range\n");
3896 					pf->dsf_dir = 0;
3897 				}
3898 				NFSD_DEBUG(4, "copymr: buflen=%d\n", buflen);
3899 				error = nfsrv_copymr(vp, nvp,
3900 				    ds->nfsdev_dsdir[pf->dsf_dir], ds, pf,
3901 				    (struct pnfsdsfile *)buf,
3902 				    buflen / sizeof(*pf), td->td_ucred, td);
3903 				vput(vp);
3904 				vput(nvp);
3905 				if (fds != NULL && error == 0) {
3906 					curdvp = fds->nfsdev_dsdir[pf->dsf_dir];
3907 					ret = vn_lock(curdvp, LK_EXCLUSIVE);
3908 					if (ret == 0) {
3909 						nfsrv_dsremove(curdvp, fname,
3910 						    td->td_ucred, td);
3911 						NFSVOPUNLOCK(curdvp);
3912 					}
3913 				}
3914 				NFSD_DEBUG(4, "nfsrv_copymr=%d\n", error);
3915 			}
3916 			free(cp, M_TEMP);
3917 			free(cp2, M_TEMP);
3918 			free(cp3, M_TEMP);
3919 			free(buf, M_TEMP);
3920 		}
3921 	} else {
3922 		error = nfssvc_srvcall(td, uap, td->td_ucred);
3923 	}
3924 
3925 out:
3926 	NFSD_CURVNET_RESTORE();
3927 	NFSEXITCODE(error);
3928 	return (error);
3929 }
3930 
3931 static int
3932 nfssvc_srvcall(struct thread *p, struct nfssvc_args *uap, struct ucred *cred)
3933 {
3934 	struct nfsex_args export;
3935 	struct nfsex_oldargs oexp;
3936 	struct file *fp = NULL;
3937 	int stablefd, i, len;
3938 	struct nfsd_clid adminrevoke;
3939 	struct nfsd_dumplist dumplist;
3940 	struct nfsd_dumpclients *dumpclients;
3941 	struct nfsd_dumplocklist dumplocklist;
3942 	struct nfsd_dumplocks *dumplocks;
3943 	struct nameidata nd;
3944 	vnode_t vp;
3945 	int error = EINVAL, igotlock;
3946 	struct proc *procp;
3947 	gid_t *grps;
3948 
3949 	if (uap->flag & NFSSVC_PUBLICFH) {
3950 		NFSBZERO((caddr_t)&nfs_pubfh.nfsrvfh_data,
3951 		    sizeof (fhandle_t));
3952 		error = copyin(uap->argp,
3953 		    &nfs_pubfh.nfsrvfh_data, sizeof (fhandle_t));
3954 		if (!error)
3955 			nfs_pubfhset = 1;
3956 	} else if ((uap->flag & (NFSSVC_V4ROOTEXPORT | NFSSVC_NEWSTRUCT)) ==
3957 	    (NFSSVC_V4ROOTEXPORT | NFSSVC_NEWSTRUCT)) {
3958 		error = copyin(uap->argp,(caddr_t)&export,
3959 		    sizeof (struct nfsex_args));
3960 		if (!error) {
3961 			grps = NULL;
3962 			if (export.export.ex_ngroups > NGROUPS_MAX ||
3963 			    export.export.ex_ngroups < 0)
3964 				error = EINVAL;
3965 			else if (export.export.ex_ngroups > 0) {
3966 				grps = malloc(export.export.ex_ngroups *
3967 				    sizeof(gid_t), M_TEMP, M_WAITOK);
3968 				error = copyin(export.export.ex_groups, grps,
3969 				    export.export.ex_ngroups * sizeof(gid_t));
3970 				export.export.ex_groups = grps;
3971 			} else
3972 				export.export.ex_groups = NULL;
3973 			if (!error)
3974 				error = nfsrv_v4rootexport(&export, cred, p);
3975 			free(grps, M_TEMP);
3976 		}
3977 	} else if ((uap->flag & (NFSSVC_V4ROOTEXPORT | NFSSVC_NEWSTRUCT)) ==
3978 	    NFSSVC_V4ROOTEXPORT) {
3979 		error = copyin(uap->argp,(caddr_t)&oexp,
3980 		    sizeof (struct nfsex_oldargs));
3981 		if (!error) {
3982 			memset(&export.export, 0, sizeof(export.export));
3983 			export.export.ex_flags = (uint64_t)oexp.export.ex_flags;
3984 			export.export.ex_root = oexp.export.ex_root;
3985 			export.export.ex_uid = oexp.export.ex_anon.cr_uid;
3986 			export.export.ex_ngroups =
3987 			    oexp.export.ex_anon.cr_ngroups;
3988 			export.export.ex_groups = NULL;
3989 			if (export.export.ex_ngroups > XU_NGROUPS ||
3990 			    export.export.ex_ngroups < 0)
3991 				error = EINVAL;
3992 			else if (export.export.ex_ngroups > 0) {
3993 				export.export.ex_groups = malloc(
3994 				    export.export.ex_ngroups * sizeof(gid_t),
3995 				    M_TEMP, M_WAITOK);
3996 				for (i = 0; i < export.export.ex_ngroups; i++)
3997 					export.export.ex_groups[i] =
3998 					    oexp.export.ex_anon.cr_groups[i];
3999 			}
4000 			export.export.ex_addr = oexp.export.ex_addr;
4001 			export.export.ex_addrlen = oexp.export.ex_addrlen;
4002 			export.export.ex_mask = oexp.export.ex_mask;
4003 			export.export.ex_masklen = oexp.export.ex_masklen;
4004 			export.export.ex_indexfile = oexp.export.ex_indexfile;
4005 			export.export.ex_numsecflavors =
4006 			    oexp.export.ex_numsecflavors;
4007 			if (export.export.ex_numsecflavors >= MAXSECFLAVORS ||
4008 			    export.export.ex_numsecflavors < 0)
4009 				error = EINVAL;
4010 			else {
4011 				for (i = 0; i < export.export.ex_numsecflavors;
4012 				    i++)
4013 					export.export.ex_secflavors[i] =
4014 					    oexp.export.ex_secflavors[i];
4015 			}
4016 			export.fspec = oexp.fspec;
4017 			if (error == 0)
4018 				error = nfsrv_v4rootexport(&export, cred, p);
4019 			free(export.export.ex_groups, M_TEMP);
4020 		}
4021 	} else if (uap->flag & NFSSVC_NOPUBLICFH) {
4022 		nfs_pubfhset = 0;
4023 		error = 0;
4024 	} else if (uap->flag & NFSSVC_STABLERESTART) {
4025 		error = copyin(uap->argp, (caddr_t)&stablefd,
4026 		    sizeof (int));
4027 		if (!error)
4028 			error = fp_getfvp(p, stablefd, &fp, &vp);
4029 		if (!error && (NFSFPFLAG(fp) & (FREAD | FWRITE)) != (FREAD | FWRITE))
4030 			error = EBADF;
4031 		if (!error && NFSD_VNET(nfsrv_numnfsd) != 0)
4032 			error = ENXIO;
4033 		if (!error) {
4034 			NFSD_VNET(nfsrv_stablefirst).nsf_fp = fp;
4035 			nfsrv_setupstable(p);
4036 		}
4037 	} else if (uap->flag & NFSSVC_ADMINREVOKE) {
4038 		error = copyin(uap->argp, (caddr_t)&adminrevoke,
4039 		    sizeof (struct nfsd_clid));
4040 		if (!error)
4041 			error = nfsrv_adminrevoke(&adminrevoke, p);
4042 	} else if (uap->flag & NFSSVC_DUMPCLIENTS) {
4043 		error = copyin(uap->argp, (caddr_t)&dumplist,
4044 		    sizeof (struct nfsd_dumplist));
4045 		if (!error && (dumplist.ndl_size < 1 ||
4046 			dumplist.ndl_size > NFSRV_MAXDUMPLIST))
4047 			error = EPERM;
4048 		if (!error) {
4049 		    len = sizeof (struct nfsd_dumpclients) * dumplist.ndl_size;
4050 		    dumpclients = malloc(len, M_TEMP, M_WAITOK | M_ZERO);
4051 		    nfsrv_dumpclients(dumpclients, dumplist.ndl_size);
4052 		    error = copyout(dumpclients, dumplist.ndl_list, len);
4053 		    free(dumpclients, M_TEMP);
4054 		}
4055 	} else if (uap->flag & NFSSVC_DUMPLOCKS) {
4056 		error = copyin(uap->argp, (caddr_t)&dumplocklist,
4057 		    sizeof (struct nfsd_dumplocklist));
4058 		if (!error && (dumplocklist.ndllck_size < 1 ||
4059 			dumplocklist.ndllck_size > NFSRV_MAXDUMPLIST))
4060 			error = EPERM;
4061 		if (!error)
4062 			error = nfsrv_lookupfilename(&nd,
4063 				dumplocklist.ndllck_fname, p);
4064 		if (!error) {
4065 			len = sizeof (struct nfsd_dumplocks) *
4066 				dumplocklist.ndllck_size;
4067 			dumplocks = malloc(len, M_TEMP, M_WAITOK | M_ZERO);
4068 			nfsrv_dumplocks(nd.ni_vp, dumplocks,
4069 			    dumplocklist.ndllck_size, p);
4070 			vput(nd.ni_vp);
4071 			error = copyout(dumplocks, dumplocklist.ndllck_list,
4072 			    len);
4073 			free(dumplocks, M_TEMP);
4074 		}
4075 	} else if (uap->flag & NFSSVC_BACKUPSTABLE) {
4076 		procp = p->td_proc;
4077 		PROC_LOCK(procp);
4078 		nfsd_master_pid = procp->p_pid;
4079 		bcopy(procp->p_comm, nfsd_master_comm, MAXCOMLEN + 1);
4080 		nfsd_master_start = procp->p_stats->p_start;
4081 		NFSD_VNET(nfsd_master_proc) = procp;
4082 		PROC_UNLOCK(procp);
4083 	} else if ((uap->flag & NFSSVC_SUSPENDNFSD) != 0) {
4084 		NFSLOCKV4ROOTMUTEX();
4085 		if (!NFSD_VNET(nfsrv_suspend_nfsd)) {
4086 			/* Lock out all nfsd threads */
4087 			do {
4088 				igotlock = nfsv4_lock(
4089 				    &NFSD_VNET(nfsd_suspend_lock), 1, NULL,
4090 				    NFSV4ROOTLOCKMUTEXPTR, NULL);
4091 			} while (igotlock == 0 &&
4092 			    !NFSD_VNET(nfsrv_suspend_nfsd));
4093 			NFSD_VNET(nfsrv_suspend_nfsd) = true;
4094 		}
4095 		NFSUNLOCKV4ROOTMUTEX();
4096 		error = 0;
4097 	} else if ((uap->flag & NFSSVC_RESUMENFSD) != 0) {
4098 		NFSLOCKV4ROOTMUTEX();
4099 		if (NFSD_VNET(nfsrv_suspend_nfsd)) {
4100 			nfsv4_unlock(&NFSD_VNET(nfsd_suspend_lock), 0);
4101 			NFSD_VNET(nfsrv_suspend_nfsd) = false;
4102 		}
4103 		NFSUNLOCKV4ROOTMUTEX();
4104 		error = 0;
4105 	}
4106 
4107 	NFSEXITCODE(error);
4108 	return (error);
4109 }
4110 
4111 /*
4112  * Check exports.
4113  * Returns 0 if ok, 1 otherwise.
4114  */
4115 int
4116 nfsvno_testexp(struct nfsrv_descript *nd, struct nfsexstuff *exp)
4117 {
4118 	int i;
4119 
4120 	if ((NFSVNO_EXTLS(exp) && (nd->nd_flag & ND_TLS) == 0) ||
4121 	    (NFSVNO_EXTLSCERT(exp) &&
4122 	     (nd->nd_flag & ND_TLSCERT) == 0) ||
4123 	    (NFSVNO_EXTLSCERTUSER(exp) &&
4124 	     (nd->nd_flag & ND_TLSCERTUSER) == 0)) {
4125 		if ((nd->nd_flag & ND_NFSV4) != 0)
4126 			return (NFSERR_WRONGSEC);
4127 #ifdef notnow
4128 		/* There is currently no auth_stat for this. */
4129 		else if ((nd->nd_flag & ND_TLS) == 0)
4130 			return (NFSERR_AUTHERR | AUTH_NEEDS_TLS);
4131 		else
4132 			return (NFSERR_AUTHERR | AUTH_NEEDS_TLS_MUTUAL_HOST);
4133 #endif
4134 		else
4135 			return (NFSERR_AUTHERR | AUTH_TOOWEAK);
4136 	}
4137 
4138 	/*
4139 	 * RFC2623 suggests that the NFSv3 Fsinfo RPC be allowed to use
4140 	 * AUTH_NONE or AUTH_SYS for file systems requiring RPCSEC_GSS.
4141 	 */
4142 	if ((nd->nd_flag & ND_NFSV3) != 0 && nd->nd_procnum == NFSPROC_FSINFO)
4143 		return (0);
4144 
4145 	/*
4146 	 * This seems odd, but allow the case where the security flavor
4147 	 * list is empty. This happens when NFSv4 is traversing non-exported
4148 	 * file systems. Exported file systems should always have a non-empty
4149 	 * security flavor list.
4150 	 */
4151 	if (exp->nes_numsecflavor == 0)
4152 		return (0);
4153 
4154 	for (i = 0; i < exp->nes_numsecflavor; i++) {
4155 		/*
4156 		 * The tests for privacy and integrity must be first,
4157 		 * since ND_GSS is set for everything but AUTH_SYS.
4158 		 */
4159 		if (exp->nes_secflavors[i] == RPCSEC_GSS_KRB5P &&
4160 		    (nd->nd_flag & ND_GSSPRIVACY))
4161 			return (0);
4162 		if (exp->nes_secflavors[i] == RPCSEC_GSS_KRB5I &&
4163 		    (nd->nd_flag & ND_GSSINTEGRITY))
4164 			return (0);
4165 		if (exp->nes_secflavors[i] == RPCSEC_GSS_KRB5 &&
4166 		    (nd->nd_flag & ND_GSS))
4167 			return (0);
4168 		if (exp->nes_secflavors[i] == AUTH_SYS &&
4169 		    (nd->nd_flag & ND_GSS) == 0)
4170 			return (0);
4171 	}
4172 	if ((nd->nd_flag & ND_NFSV4) != 0)
4173 		return (NFSERR_WRONGSEC);
4174 	return (NFSERR_AUTHERR | AUTH_TOOWEAK);
4175 }
4176 
4177 /*
4178  * Calculate a hash value for the fid in a file handle.
4179  */
4180 uint32_t
4181 nfsrv_hashfh(fhandle_t *fhp)
4182 {
4183 	uint32_t hashval;
4184 
4185 	hashval = hash32_buf(&fhp->fh_fid, sizeof(struct fid), 0);
4186 	return (hashval);
4187 }
4188 
4189 /*
4190  * Calculate a hash value for the sessionid.
4191  */
4192 uint32_t
4193 nfsrv_hashsessionid(uint8_t *sessionid)
4194 {
4195 	uint32_t hashval;
4196 
4197 	hashval = hash32_buf(sessionid, NFSX_V4SESSIONID, 0);
4198 	return (hashval);
4199 }
4200 
4201 /*
4202  * Signal the userland master nfsd to backup the stable restart file.
4203  */
4204 void
4205 nfsrv_backupstable(void)
4206 {
4207 	struct proc *procp;
4208 
4209 	if (NFSD_VNET(nfsd_master_proc) != NULL) {
4210 		procp = pfind(nfsd_master_pid);
4211 		/* Try to make sure it is the correct process. */
4212 		if (procp == NFSD_VNET(nfsd_master_proc) &&
4213 		    procp->p_stats->p_start.tv_sec ==
4214 		    nfsd_master_start.tv_sec &&
4215 		    procp->p_stats->p_start.tv_usec ==
4216 		    nfsd_master_start.tv_usec &&
4217 		    strcmp(procp->p_comm, nfsd_master_comm) == 0)
4218 			kern_psignal(procp, SIGUSR2);
4219 		else
4220 			NFSD_VNET(nfsd_master_proc) = NULL;
4221 
4222 		if (procp != NULL)
4223 			PROC_UNLOCK(procp);
4224 	}
4225 }
4226 
4227 /*
4228  * Create a DS data file for nfsrv_pnfscreate(). Called for each mirror.
4229  * The arguments are in a structure, so that they can be passed through
4230  * taskqueue for a kernel process to execute this function.
4231  */
4232 struct nfsrvdscreate {
4233 	int			done;
4234 	int			inprog;
4235 	struct task		tsk;
4236 	struct ucred		*tcred;
4237 	struct vnode		*dvp;
4238 	NFSPROC_T		*p;
4239 	struct pnfsdsfile	*pf;
4240 	int			err;
4241 	fhandle_t		fh;
4242 	struct vattr		va;
4243 	struct vattr		createva;
4244 };
4245 
4246 int
4247 nfsrv_dscreate(struct vnode *dvp, struct vattr *vap, struct vattr *nvap,
4248     fhandle_t *fhp, struct pnfsdsfile *pf, struct pnfsdsattr *dsa,
4249     char *fnamep, struct ucred *tcred, NFSPROC_T *p, struct vnode **nvpp)
4250 {
4251 	struct vnode *nvp;
4252 	struct nameidata named;
4253 	struct vattr va;
4254 	char *bufp;
4255 	u_long *hashp;
4256 	struct nfsnode *np;
4257 	struct nfsmount *nmp;
4258 	int error;
4259 
4260 	NFSNAMEICNDSET(&named.ni_cnd, tcred, CREATE,
4261 	    LOCKPARENT | LOCKLEAF | NOCACHE);
4262 	nfsvno_setpathbuf(&named, &bufp, &hashp);
4263 	named.ni_cnd.cn_lkflags = LK_EXCLUSIVE;
4264 	named.ni_cnd.cn_nameptr = bufp;
4265 	if (fnamep != NULL) {
4266 		strlcpy(bufp, fnamep, PNFS_FILENAME_LEN + 1);
4267 		named.ni_cnd.cn_namelen = strlen(bufp);
4268 	} else
4269 		named.ni_cnd.cn_namelen = nfsrv_putfhname(fhp, bufp);
4270 	NFSD_DEBUG(4, "nfsrv_dscreate: dvp=%p fname=%s\n", dvp, bufp);
4271 
4272 	/* Create the date file in the DS mount. */
4273 	error = NFSVOPLOCK(dvp, LK_EXCLUSIVE);
4274 	if (error == 0) {
4275 		error = VOP_CREATE(dvp, &nvp, &named.ni_cnd, vap);
4276 		vref(dvp);
4277 		VOP_VPUT_PAIR(dvp, error == 0 ? &nvp : NULL, false);
4278 		if (error == 0) {
4279 			/* Set the ownership of the file. */
4280 			error = VOP_SETATTR(nvp, nvap, tcred);
4281 			NFSD_DEBUG(4, "nfsrv_dscreate:"
4282 			    " setattr-uid=%d\n", error);
4283 			if (error != 0)
4284 				vput(nvp);
4285 		}
4286 		if (error != 0)
4287 			printf("pNFS: pnfscreate failed=%d\n", error);
4288 	} else
4289 		printf("pNFS: pnfscreate vnlock=%d\n", error);
4290 	if (error == 0) {
4291 		np = VTONFS(nvp);
4292 		nmp = VFSTONFS(nvp->v_mount);
4293 		if (strcmp(nvp->v_mount->mnt_vfc->vfc_name, "nfs")
4294 		    != 0 || nmp->nm_nam->sa_len > sizeof(
4295 		    struct sockaddr_in6) ||
4296 		    np->n_fhp->nfh_len != NFSX_MYFH) {
4297 			printf("Bad DS file: fstype=%s salen=%d"
4298 			    " fhlen=%d\n",
4299 			    nvp->v_mount->mnt_vfc->vfc_name,
4300 			    nmp->nm_nam->sa_len, np->n_fhp->nfh_len);
4301 			error = ENOENT;
4302 		}
4303 
4304 		/* Set extattrs for the DS on the MDS file. */
4305 		if (error == 0) {
4306 			if (dsa != NULL) {
4307 				error = VOP_GETATTR(nvp, &va, tcred);
4308 				if (error == 0) {
4309 					dsa->dsa_filerev = va.va_filerev;
4310 					dsa->dsa_size = va.va_size;
4311 					dsa->dsa_atime = va.va_atime;
4312 					dsa->dsa_mtime = va.va_mtime;
4313 					dsa->dsa_bytes = va.va_bytes;
4314 				}
4315 			}
4316 			if (error == 0) {
4317 				NFSBCOPY(np->n_fhp->nfh_fh, &pf->dsf_fh,
4318 				    NFSX_MYFH);
4319 				NFSBCOPY(nmp->nm_nam, &pf->dsf_sin,
4320 				    nmp->nm_nam->sa_len);
4321 				NFSBCOPY(named.ni_cnd.cn_nameptr,
4322 				    pf->dsf_filename,
4323 				    sizeof(pf->dsf_filename));
4324 			}
4325 		} else
4326 			printf("pNFS: pnfscreate can't get DS"
4327 			    " attr=%d\n", error);
4328 		if (nvpp != NULL && error == 0)
4329 			*nvpp = nvp;
4330 		else
4331 			vput(nvp);
4332 	}
4333 	nfsvno_relpathbuf(&named);
4334 	return (error);
4335 }
4336 
4337 /*
4338  * Start up the thread that will execute nfsrv_dscreate().
4339  */
4340 static void
4341 start_dscreate(void *arg, int pending)
4342 {
4343 	struct nfsrvdscreate *dsc;
4344 
4345 	dsc = (struct nfsrvdscreate *)arg;
4346 	dsc->err = nfsrv_dscreate(dsc->dvp, &dsc->createva, &dsc->va, &dsc->fh,
4347 	    dsc->pf, NULL, NULL, dsc->tcred, dsc->p, NULL);
4348 	dsc->done = 1;
4349 	NFSD_DEBUG(4, "start_dscreate: err=%d\n", dsc->err);
4350 }
4351 
4352 /*
4353  * Create a pNFS data file on the Data Server(s).
4354  */
4355 static void
4356 nfsrv_pnfscreate(struct vnode *vp, struct vattr *vap, struct ucred *cred,
4357     NFSPROC_T *p)
4358 {
4359 	struct nfsrvdscreate *dsc, *tdsc = NULL;
4360 	struct nfsdevice *ds, *tds, *fds;
4361 	struct mount *mp;
4362 	struct pnfsdsfile *pf, *tpf;
4363 	struct pnfsdsattr dsattr;
4364 	struct vattr va;
4365 	struct vnode *dvp[NFSDEV_MAXMIRRORS];
4366 	struct nfsmount *nmp;
4367 	fhandle_t fh;
4368 	uid_t vauid;
4369 	gid_t vagid;
4370 	u_short vamode;
4371 	struct ucred *tcred;
4372 	int dsdir[NFSDEV_MAXMIRRORS], error, i, mirrorcnt, ret;
4373 	int failpos, timo;
4374 
4375 	/* Get a DS server directory in a round-robin order. */
4376 	mirrorcnt = 1;
4377 	mp = vp->v_mount;
4378 	ds = fds = NULL;
4379 	NFSDDSLOCK();
4380 	/*
4381 	 * Search for the first entry that handles this MDS fs, but use the
4382 	 * first entry for all MDS fs's otherwise.
4383 	 */
4384 	TAILQ_FOREACH(tds, &nfsrv_devidhead, nfsdev_list) {
4385 		if (tds->nfsdev_nmp != NULL) {
4386 			if (tds->nfsdev_mdsisset == 0 && ds == NULL)
4387 				ds = tds;
4388 			else if (tds->nfsdev_mdsisset != 0 && fsidcmp(
4389 			    &mp->mnt_stat.f_fsid, &tds->nfsdev_mdsfsid) == 0) {
4390 				ds = fds = tds;
4391 				break;
4392 			}
4393 		}
4394 	}
4395 	if (ds == NULL) {
4396 		NFSDDSUNLOCK();
4397 		NFSD_DEBUG(4, "nfsrv_pnfscreate: no srv\n");
4398 		return;
4399 	}
4400 	i = dsdir[0] = ds->nfsdev_nextdir;
4401 	ds->nfsdev_nextdir = (ds->nfsdev_nextdir + 1) % nfsrv_dsdirsize;
4402 	dvp[0] = ds->nfsdev_dsdir[i];
4403 	tds = TAILQ_NEXT(ds, nfsdev_list);
4404 	if (nfsrv_maxpnfsmirror > 1 && tds != NULL) {
4405 		TAILQ_FOREACH_FROM(tds, &nfsrv_devidhead, nfsdev_list) {
4406 			if (tds->nfsdev_nmp != NULL &&
4407 			    ((tds->nfsdev_mdsisset == 0 && fds == NULL) ||
4408 			     (tds->nfsdev_mdsisset != 0 && fds != NULL &&
4409 			      fsidcmp(&mp->mnt_stat.f_fsid,
4410 			      &tds->nfsdev_mdsfsid) == 0))) {
4411 				dsdir[mirrorcnt] = i;
4412 				dvp[mirrorcnt] = tds->nfsdev_dsdir[i];
4413 				mirrorcnt++;
4414 				if (mirrorcnt >= nfsrv_maxpnfsmirror)
4415 					break;
4416 			}
4417 		}
4418 	}
4419 	/* Put at end of list to implement round-robin usage. */
4420 	TAILQ_REMOVE(&nfsrv_devidhead, ds, nfsdev_list);
4421 	TAILQ_INSERT_TAIL(&nfsrv_devidhead, ds, nfsdev_list);
4422 	NFSDDSUNLOCK();
4423 	dsc = NULL;
4424 	if (mirrorcnt > 1)
4425 		tdsc = dsc = malloc(sizeof(*dsc) * (mirrorcnt - 1), M_TEMP,
4426 		    M_WAITOK | M_ZERO);
4427 	tpf = pf = malloc(sizeof(*pf) * nfsrv_maxpnfsmirror, M_TEMP, M_WAITOK |
4428 	    M_ZERO);
4429 
4430 	error = nfsvno_getfh(vp, &fh, p);
4431 	if (error == 0)
4432 		error = VOP_GETATTR(vp, &va, cred);
4433 	if (error == 0) {
4434 		/* Set the attributes for "vp" to Setattr the DS vp. */
4435 		vauid = va.va_uid;
4436 		vagid = va.va_gid;
4437 		vamode = va.va_mode;
4438 		VATTR_NULL(&va);
4439 		va.va_uid = vauid;
4440 		va.va_gid = vagid;
4441 		va.va_mode = vamode;
4442 		va.va_size = 0;
4443 	} else
4444 		printf("pNFS: pnfscreate getfh+attr=%d\n", error);
4445 
4446 	NFSD_DEBUG(4, "nfsrv_pnfscreate: cruid=%d crgid=%d\n", cred->cr_uid,
4447 	    cred->cr_gid);
4448 	/* Make data file name based on FH. */
4449 	tcred = newnfs_getcred();
4450 
4451 	/*
4452 	 * Create the file on each DS mirror, using kernel process(es) for the
4453 	 * additional mirrors.
4454 	 */
4455 	failpos = -1;
4456 	for (i = 0; i < mirrorcnt - 1 && error == 0; i++, tpf++, tdsc++) {
4457 		tpf->dsf_dir = dsdir[i];
4458 		tdsc->tcred = tcred;
4459 		tdsc->p = p;
4460 		tdsc->pf = tpf;
4461 		tdsc->createva = *vap;
4462 		NFSBCOPY(&fh, &tdsc->fh, sizeof(fh));
4463 		tdsc->va = va;
4464 		tdsc->dvp = dvp[i];
4465 		tdsc->done = 0;
4466 		tdsc->inprog = 0;
4467 		tdsc->err = 0;
4468 		ret = EIO;
4469 		if (nfs_pnfsiothreads != 0) {
4470 			ret = nfs_pnfsio(start_dscreate, tdsc);
4471 			NFSD_DEBUG(4, "nfsrv_pnfscreate: nfs_pnfsio=%d\n", ret);
4472 		}
4473 		if (ret != 0) {
4474 			ret = nfsrv_dscreate(dvp[i], vap, &va, &fh, tpf, NULL,
4475 			    NULL, tcred, p, NULL);
4476 			if (ret != 0) {
4477 				KASSERT(error == 0, ("nfsrv_dscreate err=%d",
4478 				    error));
4479 				if (failpos == -1 && nfsds_failerr(ret))
4480 					failpos = i;
4481 				else
4482 					error = ret;
4483 			}
4484 		}
4485 	}
4486 	if (error == 0) {
4487 		tpf->dsf_dir = dsdir[mirrorcnt - 1];
4488 		error = nfsrv_dscreate(dvp[mirrorcnt - 1], vap, &va, &fh, tpf,
4489 		    &dsattr, NULL, tcred, p, NULL);
4490 		if (failpos == -1 && mirrorcnt > 1 && nfsds_failerr(error)) {
4491 			failpos = mirrorcnt - 1;
4492 			error = 0;
4493 		}
4494 	}
4495 	timo = hz / 50;		/* Wait for 20msec. */
4496 	if (timo < 1)
4497 		timo = 1;
4498 	/* Wait for kernel task(s) to complete. */
4499 	for (tdsc = dsc, i = 0; i < mirrorcnt - 1; i++, tdsc++) {
4500 		while (tdsc->inprog != 0 && tdsc->done == 0)
4501 			tsleep(&tdsc->tsk, PVFS, "srvdcr", timo);
4502 		if (tdsc->err != 0) {
4503 			if (failpos == -1 && nfsds_failerr(tdsc->err))
4504 				failpos = i;
4505 			else if (error == 0)
4506 				error = tdsc->err;
4507 		}
4508 	}
4509 
4510 	/*
4511 	 * If failpos has been set, that mirror has failed, so it needs
4512 	 * to be disabled.
4513 	 */
4514 	if (failpos >= 0) {
4515 		nmp = VFSTONFS(dvp[failpos]->v_mount);
4516 		NFSLOCKMNT(nmp);
4517 		if ((nmp->nm_privflag & (NFSMNTP_FORCEDISM |
4518 		     NFSMNTP_CANCELRPCS)) == 0) {
4519 			nmp->nm_privflag |= NFSMNTP_CANCELRPCS;
4520 			NFSUNLOCKMNT(nmp);
4521 			ds = nfsrv_deldsnmp(PNFSDOP_DELDSSERVER, nmp, p);
4522 			NFSD_DEBUG(4, "dscreatfail fail=%d ds=%p\n", failpos,
4523 			    ds);
4524 			if (ds != NULL)
4525 				nfsrv_killrpcs(nmp);
4526 			NFSLOCKMNT(nmp);
4527 			nmp->nm_privflag &= ~NFSMNTP_CANCELRPCS;
4528 			wakeup(nmp);
4529 		}
4530 		NFSUNLOCKMNT(nmp);
4531 	}
4532 
4533 	NFSFREECRED(tcred);
4534 	if (error == 0) {
4535 		ASSERT_VOP_ELOCKED(vp, "nfsrv_pnfscreate vp");
4536 
4537 		NFSD_DEBUG(4, "nfsrv_pnfscreate: mirrorcnt=%d maxmirror=%d\n",
4538 		    mirrorcnt, nfsrv_maxpnfsmirror);
4539 		/*
4540 		 * For all mirrors that couldn't be created, fill in the
4541 		 * *pf structure, but with an IP address == 0.0.0.0.
4542 		 */
4543 		tpf = pf + mirrorcnt;
4544 		for (i = mirrorcnt; i < nfsrv_maxpnfsmirror; i++, tpf++) {
4545 			*tpf = *pf;
4546 			tpf->dsf_sin.sin_family = AF_INET;
4547 			tpf->dsf_sin.sin_len = sizeof(struct sockaddr_in);
4548 			tpf->dsf_sin.sin_addr.s_addr = 0;
4549 			tpf->dsf_sin.sin_port = 0;
4550 		}
4551 
4552 		error = vn_extattr_set(vp, IO_NODELOCKED,
4553 		    EXTATTR_NAMESPACE_SYSTEM, "pnfsd.dsfile",
4554 		    sizeof(*pf) * nfsrv_maxpnfsmirror, (char *)pf, p);
4555 		if (error == 0)
4556 			error = vn_extattr_set(vp, IO_NODELOCKED,
4557 			    EXTATTR_NAMESPACE_SYSTEM, "pnfsd.dsattr",
4558 			    sizeof(dsattr), (char *)&dsattr, p);
4559 		if (error != 0)
4560 			printf("pNFS: pnfscreate setextattr=%d\n",
4561 			    error);
4562 	} else
4563 		printf("pNFS: pnfscreate=%d\n", error);
4564 	free(pf, M_TEMP);
4565 	free(dsc, M_TEMP);
4566 }
4567 
4568 /*
4569  * Get the information needed to remove the pNFS Data Server file from the
4570  * Metadata file.  Upon success, ddvp is set non-NULL to the locked
4571  * DS directory vnode.  The caller must unlock *ddvp when done with it.
4572  */
4573 static void
4574 nfsrv_pnfsremovesetup(struct vnode *vp, NFSPROC_T *p, struct vnode **dvpp,
4575     int *mirrorcntp, char *fname, fhandle_t *fhp)
4576 {
4577 	struct vattr va;
4578 	struct ucred *tcred;
4579 	char *buf;
4580 	int buflen, error;
4581 
4582 	dvpp[0] = NULL;
4583 	/* If not an exported regular file or not a pNFS server, just return. */
4584 	if (vp->v_type != VREG || (vp->v_mount->mnt_flag & MNT_EXPORTED) == 0 ||
4585 	    nfsrv_devidcnt == 0)
4586 		return;
4587 
4588 	/* Check to see if this is the last hard link. */
4589 	tcred = newnfs_getcred();
4590 	error = VOP_GETATTR(vp, &va, tcred);
4591 	NFSFREECRED(tcred);
4592 	if (error != 0) {
4593 		printf("pNFS: nfsrv_pnfsremovesetup getattr=%d\n", error);
4594 		return;
4595 	}
4596 	if (va.va_nlink > 1)
4597 		return;
4598 
4599 	error = nfsvno_getfh(vp, fhp, p);
4600 	if (error != 0) {
4601 		printf("pNFS: nfsrv_pnfsremovesetup getfh=%d\n", error);
4602 		return;
4603 	}
4604 
4605 	buflen = 1024;
4606 	buf = malloc(buflen, M_TEMP, M_WAITOK);
4607 	/* Get the directory vnode for the DS mount and the file handle. */
4608 	error = nfsrv_dsgetsockmnt(vp, 0, buf, &buflen, mirrorcntp, p, dvpp,
4609 	    NULL, NULL, fname, NULL, NULL, NULL, NULL, NULL);
4610 	free(buf, M_TEMP);
4611 	if (error != 0)
4612 		printf("pNFS: nfsrv_pnfsremovesetup getsockmnt=%d\n", error);
4613 }
4614 
4615 /*
4616  * Remove a DS data file for nfsrv_pnfsremove(). Called for each mirror.
4617  * The arguments are in a structure, so that they can be passed through
4618  * taskqueue for a kernel process to execute this function.
4619  */
4620 struct nfsrvdsremove {
4621 	int			done;
4622 	int			inprog;
4623 	struct task		tsk;
4624 	struct ucred		*tcred;
4625 	struct vnode		*dvp;
4626 	NFSPROC_T		*p;
4627 	int			err;
4628 	char			fname[PNFS_FILENAME_LEN + 1];
4629 };
4630 
4631 static int
4632 nfsrv_dsremove(struct vnode *dvp, char *fname, struct ucred *tcred,
4633     NFSPROC_T *p)
4634 {
4635 	struct nameidata named;
4636 	struct vnode *nvp;
4637 	char *bufp;
4638 	u_long *hashp;
4639 	int error;
4640 
4641 	error = NFSVOPLOCK(dvp, LK_EXCLUSIVE);
4642 	if (error != 0)
4643 		return (error);
4644 	named.ni_cnd.cn_nameiop = DELETE;
4645 	named.ni_cnd.cn_lkflags = LK_EXCLUSIVE | LK_RETRY;
4646 	named.ni_cnd.cn_cred = tcred;
4647 	named.ni_cnd.cn_flags = ISLASTCN | LOCKPARENT | LOCKLEAF;
4648 	nfsvno_setpathbuf(&named, &bufp, &hashp);
4649 	named.ni_cnd.cn_nameptr = bufp;
4650 	named.ni_cnd.cn_namelen = strlen(fname);
4651 	strlcpy(bufp, fname, NAME_MAX);
4652 	NFSD_DEBUG(4, "nfsrv_pnfsremove: filename=%s\n", bufp);
4653 	error = VOP_LOOKUP(dvp, &nvp, &named.ni_cnd);
4654 	NFSD_DEBUG(4, "nfsrv_pnfsremove: aft LOOKUP=%d\n", error);
4655 	if (error == 0) {
4656 		error = VOP_REMOVE(dvp, nvp, &named.ni_cnd);
4657 		vput(nvp);
4658 	}
4659 	NFSVOPUNLOCK(dvp);
4660 	nfsvno_relpathbuf(&named);
4661 	if (error != 0)
4662 		printf("pNFS: nfsrv_pnfsremove failed=%d\n", error);
4663 	return (error);
4664 }
4665 
4666 /*
4667  * Start up the thread that will execute nfsrv_dsremove().
4668  */
4669 static void
4670 start_dsremove(void *arg, int pending)
4671 {
4672 	struct nfsrvdsremove *dsrm;
4673 
4674 	dsrm = (struct nfsrvdsremove *)arg;
4675 	dsrm->err = nfsrv_dsremove(dsrm->dvp, dsrm->fname, dsrm->tcred,
4676 	    dsrm->p);
4677 	dsrm->done = 1;
4678 	NFSD_DEBUG(4, "start_dsremove: err=%d\n", dsrm->err);
4679 }
4680 
4681 /*
4682  * Remove a pNFS data file from a Data Server.
4683  * nfsrv_pnfsremovesetup() must have been called before the MDS file was
4684  * removed to set up the dvp and fill in the FH.
4685  */
4686 static void
4687 nfsrv_pnfsremove(struct vnode **dvp, int mirrorcnt, char *fname, fhandle_t *fhp,
4688     NFSPROC_T *p)
4689 {
4690 	struct ucred *tcred;
4691 	struct nfsrvdsremove *dsrm, *tdsrm;
4692 	struct nfsdevice *ds;
4693 	struct nfsmount *nmp;
4694 	int failpos, i, ret, timo;
4695 
4696 	tcred = newnfs_getcred();
4697 	dsrm = NULL;
4698 	if (mirrorcnt > 1)
4699 		dsrm = malloc(sizeof(*dsrm) * mirrorcnt - 1, M_TEMP, M_WAITOK);
4700 	/*
4701 	 * Remove the file on each DS mirror, using kernel process(es) for the
4702 	 * additional mirrors.
4703 	 */
4704 	failpos = -1;
4705 	for (tdsrm = dsrm, i = 0; i < mirrorcnt - 1; i++, tdsrm++) {
4706 		tdsrm->tcred = tcred;
4707 		tdsrm->p = p;
4708 		tdsrm->dvp = dvp[i];
4709 		strlcpy(tdsrm->fname, fname, PNFS_FILENAME_LEN + 1);
4710 		tdsrm->inprog = 0;
4711 		tdsrm->done = 0;
4712 		tdsrm->err = 0;
4713 		ret = EIO;
4714 		if (nfs_pnfsiothreads != 0) {
4715 			ret = nfs_pnfsio(start_dsremove, tdsrm);
4716 			NFSD_DEBUG(4, "nfsrv_pnfsremove: nfs_pnfsio=%d\n", ret);
4717 		}
4718 		if (ret != 0) {
4719 			ret = nfsrv_dsremove(dvp[i], fname, tcred, p);
4720 			if (failpos == -1 && nfsds_failerr(ret))
4721 				failpos = i;
4722 		}
4723 	}
4724 	ret = nfsrv_dsremove(dvp[mirrorcnt - 1], fname, tcred, p);
4725 	if (failpos == -1 && mirrorcnt > 1 && nfsds_failerr(ret))
4726 		failpos = mirrorcnt - 1;
4727 	timo = hz / 50;		/* Wait for 20msec. */
4728 	if (timo < 1)
4729 		timo = 1;
4730 	/* Wait for kernel task(s) to complete. */
4731 	for (tdsrm = dsrm, i = 0; i < mirrorcnt - 1; i++, tdsrm++) {
4732 		while (tdsrm->inprog != 0 && tdsrm->done == 0)
4733 			tsleep(&tdsrm->tsk, PVFS, "srvdsrm", timo);
4734 		if (failpos == -1 && nfsds_failerr(tdsrm->err))
4735 			failpos = i;
4736 	}
4737 
4738 	/*
4739 	 * If failpos has been set, that mirror has failed, so it needs
4740 	 * to be disabled.
4741 	 */
4742 	if (failpos >= 0) {
4743 		nmp = VFSTONFS(dvp[failpos]->v_mount);
4744 		NFSLOCKMNT(nmp);
4745 		if ((nmp->nm_privflag & (NFSMNTP_FORCEDISM |
4746 		     NFSMNTP_CANCELRPCS)) == 0) {
4747 			nmp->nm_privflag |= NFSMNTP_CANCELRPCS;
4748 			NFSUNLOCKMNT(nmp);
4749 			ds = nfsrv_deldsnmp(PNFSDOP_DELDSSERVER, nmp, p);
4750 			NFSD_DEBUG(4, "dsremovefail fail=%d ds=%p\n", failpos,
4751 			    ds);
4752 			if (ds != NULL)
4753 				nfsrv_killrpcs(nmp);
4754 			NFSLOCKMNT(nmp);
4755 			nmp->nm_privflag &= ~NFSMNTP_CANCELRPCS;
4756 			wakeup(nmp);
4757 		}
4758 		NFSUNLOCKMNT(nmp);
4759 	}
4760 
4761 	/* Get rid all layouts for the file. */
4762 	nfsrv_freefilelayouts(fhp);
4763 
4764 	NFSFREECRED(tcred);
4765 	free(dsrm, M_TEMP);
4766 }
4767 
4768 /*
4769  * Generate a file name based on the file handle and put it in *bufp.
4770  * Return the number of bytes generated.
4771  */
4772 static int
4773 nfsrv_putfhname(fhandle_t *fhp, char *bufp)
4774 {
4775 	int i;
4776 	uint8_t *cp;
4777 	const uint8_t *hexdigits = "0123456789abcdef";
4778 
4779 	cp = (uint8_t *)fhp;
4780 	for (i = 0; i < sizeof(*fhp); i++) {
4781 		bufp[2 * i] = hexdigits[(*cp >> 4) & 0xf];
4782 		bufp[2 * i + 1] = hexdigits[*cp++ & 0xf];
4783 	}
4784 	bufp[2 * i] = '\0';
4785 	return (2 * i);
4786 }
4787 
4788 /*
4789  * Update the Metadata file's attributes from the DS file when a Read/Write
4790  * layout is returned.
4791  * Basically just call nfsrv_proxyds() with procedure == NFSPROC_LAYOUTRETURN
4792  * so that it does a nfsrv_getattrdsrpc() and nfsrv_setextattr() on the DS file.
4793  */
4794 int
4795 nfsrv_updatemdsattr(struct vnode *vp, struct nfsvattr *nap, NFSPROC_T *p)
4796 {
4797 	struct ucred *tcred;
4798 	int error;
4799 
4800 	/* Do this as root so that it won't fail with EACCES. */
4801 	tcred = newnfs_getcred();
4802 	error = nfsrv_proxyds(vp, 0, 0, tcred, p, NFSPROC_LAYOUTRETURN,
4803 	    NULL, NULL, NULL, nap, NULL, NULL, 0, NULL);
4804 	NFSFREECRED(tcred);
4805 	return (error);
4806 }
4807 
4808 /*
4809  * Set the NFSv4 ACL on the DS file to the same ACL as the MDS file.
4810  */
4811 static int
4812 nfsrv_dssetacl(struct vnode *vp, struct acl *aclp, struct ucred *cred,
4813     NFSPROC_T *p)
4814 {
4815 	int error;
4816 
4817 	error = nfsrv_proxyds(vp, 0, 0, cred, p, NFSPROC_SETACL,
4818 	    NULL, NULL, NULL, NULL, aclp, NULL, 0, NULL);
4819 	return (error);
4820 }
4821 
4822 static int
4823 nfsrv_proxyds(struct vnode *vp, off_t off, int cnt, struct ucred *cred,
4824     struct thread *p, int ioproc, struct mbuf **mpp, char *cp,
4825     struct mbuf **mpp2, struct nfsvattr *nap, struct acl *aclp,
4826     off_t *offp, int content, bool *eofp)
4827 {
4828 	struct nfsmount *nmp[NFSDEV_MAXMIRRORS], *failnmp;
4829 	fhandle_t fh[NFSDEV_MAXMIRRORS];
4830 	struct vnode *dvp[NFSDEV_MAXMIRRORS];
4831 	struct nfsdevice *ds;
4832 	struct pnfsdsattr dsattr;
4833 	struct opnfsdsattr odsattr;
4834 	char *buf;
4835 	int buflen, error, failpos, i, mirrorcnt, origmircnt, trycnt;
4836 
4837 	NFSD_DEBUG(4, "in nfsrv_proxyds\n");
4838 	/*
4839 	 * If not a regular file, not exported or not a pNFS server,
4840 	 * just return ENOENT.
4841 	 */
4842 	if (vp->v_type != VREG || (vp->v_mount->mnt_flag & MNT_EXPORTED) == 0 ||
4843 	    nfsrv_devidcnt == 0)
4844 		return (ENOENT);
4845 
4846 	buflen = 1024;
4847 	buf = malloc(buflen, M_TEMP, M_WAITOK);
4848 	error = 0;
4849 
4850 	/*
4851 	 * For Getattr, get the Change attribute (va_filerev) and size (va_size)
4852 	 * from the MetaData file's extended attribute.
4853 	 */
4854 	if (ioproc == NFSPROC_GETATTR) {
4855 		error = vn_extattr_get(vp, IO_NODELOCKED,
4856 		    EXTATTR_NAMESPACE_SYSTEM, "pnfsd.dsattr", &buflen, buf,
4857 		    p);
4858 		if (error == 0) {
4859 			if (buflen == sizeof(odsattr)) {
4860 				NFSBCOPY(buf, &odsattr, buflen);
4861 				nap->na_filerev = odsattr.dsa_filerev;
4862 				nap->na_size = odsattr.dsa_size;
4863 				nap->na_atime = odsattr.dsa_atime;
4864 				nap->na_mtime = odsattr.dsa_mtime;
4865 				/*
4866 				 * Fake na_bytes by rounding up na_size.
4867 				 * Since we don't know the block size, just
4868 				 * use BLKDEV_IOSIZE.
4869 				 */
4870 				nap->na_bytes = (odsattr.dsa_size +
4871 				    BLKDEV_IOSIZE - 1) & ~(BLKDEV_IOSIZE - 1);
4872 			} else if (buflen == sizeof(dsattr)) {
4873 				NFSBCOPY(buf, &dsattr, buflen);
4874 				nap->na_filerev = dsattr.dsa_filerev;
4875 				nap->na_size = dsattr.dsa_size;
4876 				nap->na_atime = dsattr.dsa_atime;
4877 				nap->na_mtime = dsattr.dsa_mtime;
4878 				nap->na_bytes = dsattr.dsa_bytes;
4879 			} else
4880 				error = ENXIO;
4881 		}
4882 		if (error == 0) {
4883 			/*
4884 			 * If nfsrv_pnfsgetdsattr is 0 or nfsrv_checkdsattr()
4885 			 * returns 0, just return now.  nfsrv_checkdsattr()
4886 			 * returns 0 if there is no Read/Write layout
4887 			 * plus either an Open/Write_access or Write
4888 			 * delegation issued to a client for the file.
4889 			 */
4890 			if (nfsrv_pnfsgetdsattr == 0 ||
4891 			    nfsrv_checkdsattr(vp, p) == 0) {
4892 				free(buf, M_TEMP);
4893 				return (error);
4894 			}
4895 		}
4896 
4897 		/*
4898 		 * Clear ENOATTR so the code below will attempt to do a
4899 		 * nfsrv_getattrdsrpc() to get the attributes and (re)create
4900 		 * the extended attribute.
4901 		 */
4902 		if (error == ENOATTR)
4903 			error = 0;
4904 	}
4905 
4906 	origmircnt = -1;
4907 	trycnt = 0;
4908 tryagain:
4909 	if (error == 0) {
4910 		buflen = 1024;
4911 		if (ioproc == NFSPROC_READDS && NFSVOPISLOCKED(vp) ==
4912 		    LK_EXCLUSIVE)
4913 			printf("nfsrv_proxyds: Readds vp exclusively locked\n");
4914 		error = nfsrv_dsgetsockmnt(vp, LK_SHARED, buf, &buflen,
4915 		    &mirrorcnt, p, dvp, fh, NULL, NULL, NULL, NULL, NULL,
4916 		    NULL, NULL);
4917 		if (error == 0) {
4918 			for (i = 0; i < mirrorcnt; i++)
4919 				nmp[i] = VFSTONFS(dvp[i]->v_mount);
4920 		} else
4921 			printf("pNFS: proxy getextattr sockaddr=%d\n", error);
4922 	} else
4923 		printf("pNFS: nfsrv_dsgetsockmnt=%d\n", error);
4924 	if (error == 0) {
4925 		failpos = -1;
4926 		if (origmircnt == -1)
4927 			origmircnt = mirrorcnt;
4928 		/*
4929 		 * If failpos is set to a mirror#, then that mirror has
4930 		 * failed and will be disabled. For Read, Getattr and Seek, the
4931 		 * function only tries one mirror, so if that mirror has
4932 		 * failed, it will need to be retried. As such, increment
4933 		 * tryitagain for these cases.
4934 		 * For Write, Setattr and Setacl, the function tries all
4935 		 * mirrors and will not return an error for the case where
4936 		 * one mirror has failed. For these cases, the functioning
4937 		 * mirror(s) will have been modified, so a retry isn't
4938 		 * necessary. These functions will set failpos for the
4939 		 * failed mirror#.
4940 		 */
4941 		if (ioproc == NFSPROC_READDS) {
4942 			error = nfsrv_readdsrpc(fh, off, cnt, cred, p, nmp[0],
4943 			    mpp, mpp2);
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_WRITEDS)
4955 			error = nfsrv_writedsrpc(fh, off, cnt, cred, p, vp,
4956 			    &nmp[0], mirrorcnt, mpp, cp, &failpos);
4957 		else if (ioproc == NFSPROC_SETATTR)
4958 			error = nfsrv_setattrdsrpc(fh, cred, p, vp, &nmp[0],
4959 			    mirrorcnt, nap, &failpos);
4960 		else if (ioproc == NFSPROC_SETACL)
4961 			error = nfsrv_setacldsrpc(fh, cred, p, vp, &nmp[0],
4962 			    mirrorcnt, aclp, &failpos);
4963 		else if (ioproc == NFSPROC_SEEKDS) {
4964 			error = nfsrv_seekdsrpc(fh, offp, content, eofp, cred,
4965 			    p, nmp[0]);
4966 			if (nfsds_failerr(error) && mirrorcnt > 1) {
4967 				/*
4968 				 * Setting failpos will cause the mirror
4969 				 * to be disabled and then a retry of this
4970 				 * read is required.
4971 				 */
4972 				failpos = 0;
4973 				error = 0;
4974 				trycnt++;
4975 			}
4976 		} else if (ioproc == NFSPROC_ALLOCATE)
4977 			error = nfsrv_allocatedsrpc(fh, off, *offp, cred, p, vp,
4978 			    &nmp[0], mirrorcnt, &failpos);
4979 		else if (ioproc == NFSPROC_DEALLOCATE)
4980 			error = nfsrv_deallocatedsrpc(fh, off, *offp, cred, p,
4981 			    vp, &nmp[0], mirrorcnt, &failpos);
4982 		else {
4983 			error = nfsrv_getattrdsrpc(&fh[mirrorcnt - 1], cred, p,
4984 			    vp, nmp[mirrorcnt - 1], nap);
4985 			if (nfsds_failerr(error) && mirrorcnt > 1) {
4986 				/*
4987 				 * Setting failpos will cause the mirror
4988 				 * to be disabled and then a retry of this
4989 				 * getattr is required.
4990 				 */
4991 				failpos = mirrorcnt - 1;
4992 				error = 0;
4993 				trycnt++;
4994 			}
4995 		}
4996 		ds = NULL;
4997 		if (failpos >= 0) {
4998 			failnmp = nmp[failpos];
4999 			NFSLOCKMNT(failnmp);
5000 			if ((failnmp->nm_privflag & (NFSMNTP_FORCEDISM |
5001 			     NFSMNTP_CANCELRPCS)) == 0) {
5002 				failnmp->nm_privflag |= NFSMNTP_CANCELRPCS;
5003 				NFSUNLOCKMNT(failnmp);
5004 				ds = nfsrv_deldsnmp(PNFSDOP_DELDSSERVER,
5005 				    failnmp, p);
5006 				NFSD_DEBUG(4, "dsldsnmp fail=%d ds=%p\n",
5007 				    failpos, ds);
5008 				if (ds != NULL)
5009 					nfsrv_killrpcs(failnmp);
5010 				NFSLOCKMNT(failnmp);
5011 				failnmp->nm_privflag &= ~NFSMNTP_CANCELRPCS;
5012 				wakeup(failnmp);
5013 			}
5014 			NFSUNLOCKMNT(failnmp);
5015 		}
5016 		for (i = 0; i < mirrorcnt; i++)
5017 			NFSVOPUNLOCK(dvp[i]);
5018 		NFSD_DEBUG(4, "nfsrv_proxyds: aft RPC=%d trya=%d\n", error,
5019 		    trycnt);
5020 		/* Try the Read/Getattr again if a mirror was deleted. */
5021 		if (ds != NULL && trycnt > 0 && trycnt < origmircnt)
5022 			goto tryagain;
5023 	} else {
5024 		/* Return ENOENT for any Extended Attribute error. */
5025 		error = ENOENT;
5026 	}
5027 	free(buf, M_TEMP);
5028 	NFSD_DEBUG(4, "nfsrv_proxyds: error=%d\n", error);
5029 	return (error);
5030 }
5031 
5032 /*
5033  * Get the DS mount point, fh and directory from the "pnfsd.dsfile" extended
5034  * attribute.
5035  * newnmpp - If it points to a non-NULL nmp, that is the destination and needs
5036  *           to be checked.  If it points to a NULL nmp, then it returns
5037  *           a suitable destination.
5038  * curnmp - If non-NULL, it is the source mount for the copy.
5039  */
5040 int
5041 nfsrv_dsgetsockmnt(struct vnode *vp, int lktype, char *buf, int *buflenp,
5042     int *mirrorcntp, NFSPROC_T *p, struct vnode **dvpp, fhandle_t *fhp,
5043     char *devid, char *fnamep, struct vnode **nvpp, struct nfsmount **newnmpp,
5044     struct nfsmount *curnmp, int *ippos, int *dsdirp)
5045 {
5046 	struct vnode *dvp, *nvp = NULL, **tdvpp;
5047 	struct mount *mp;
5048 	struct nfsmount *nmp, *newnmp;
5049 	struct sockaddr *sad;
5050 	struct sockaddr_in *sin;
5051 	struct nfsdevice *ds, *tds, *fndds;
5052 	struct pnfsdsfile *pf;
5053 	uint32_t dsdir;
5054 	int error, fhiszero, fnd, gotone, i, mirrorcnt;
5055 
5056 	ASSERT_VOP_LOCKED(vp, "nfsrv_dsgetsockmnt vp");
5057 	*mirrorcntp = 1;
5058 	tdvpp = dvpp;
5059 	if (nvpp != NULL)
5060 		*nvpp = NULL;
5061 	if (dvpp != NULL)
5062 		*dvpp = NULL;
5063 	if (ippos != NULL)
5064 		*ippos = -1;
5065 	if (newnmpp != NULL)
5066 		newnmp = *newnmpp;
5067 	else
5068 		newnmp = NULL;
5069 	mp = vp->v_mount;
5070 	error = vn_extattr_get(vp, IO_NODELOCKED, EXTATTR_NAMESPACE_SYSTEM,
5071 	    "pnfsd.dsfile", buflenp, buf, p);
5072 	mirrorcnt = *buflenp / sizeof(*pf);
5073 	if (error == 0 && (mirrorcnt < 1 || mirrorcnt > NFSDEV_MAXMIRRORS ||
5074 	    *buflenp != sizeof(*pf) * mirrorcnt))
5075 		error = ENOATTR;
5076 
5077 	pf = (struct pnfsdsfile *)buf;
5078 	/* If curnmp != NULL, check for a match in the mirror list. */
5079 	if (curnmp != NULL && error == 0) {
5080 		fnd = 0;
5081 		for (i = 0; i < mirrorcnt; i++, pf++) {
5082 			sad = (struct sockaddr *)&pf->dsf_sin;
5083 			if (nfsaddr2_match(sad, curnmp->nm_nam)) {
5084 				if (ippos != NULL)
5085 					*ippos = i;
5086 				fnd = 1;
5087 				break;
5088 			}
5089 		}
5090 		if (fnd == 0)
5091 			error = ENXIO;
5092 	}
5093 
5094 	gotone = 0;
5095 	pf = (struct pnfsdsfile *)buf;
5096 	NFSD_DEBUG(4, "nfsrv_dsgetsockmnt: mirrorcnt=%d err=%d\n", mirrorcnt,
5097 	    error);
5098 	for (i = 0; i < mirrorcnt && error == 0; i++, pf++) {
5099 		fhiszero = 0;
5100 		sad = (struct sockaddr *)&pf->dsf_sin;
5101 		sin = &pf->dsf_sin;
5102 		dsdir = pf->dsf_dir;
5103 		if (dsdir >= nfsrv_dsdirsize) {
5104 			printf("nfsrv_dsgetsockmnt: dsdir=%d\n", dsdir);
5105 			error = ENOATTR;
5106 		} else if (nvpp != NULL && newnmp != NULL &&
5107 		    nfsaddr2_match(sad, newnmp->nm_nam))
5108 			error = EEXIST;
5109 		if (error == 0) {
5110 			if (ippos != NULL && curnmp == NULL &&
5111 			    sad->sa_family == AF_INET &&
5112 			    sin->sin_addr.s_addr == 0)
5113 				*ippos = i;
5114 			if (NFSBCMP(&zerofh, &pf->dsf_fh, sizeof(zerofh)) == 0)
5115 				fhiszero = 1;
5116 			/* Use the socket address to find the mount point. */
5117 			fndds = NULL;
5118 			NFSDDSLOCK();
5119 			/* Find a match for the IP address. */
5120 			TAILQ_FOREACH(ds, &nfsrv_devidhead, nfsdev_list) {
5121 				if (ds->nfsdev_nmp != NULL) {
5122 					dvp = ds->nfsdev_dvp;
5123 					nmp = VFSTONFS(dvp->v_mount);
5124 					if (nmp != ds->nfsdev_nmp)
5125 						printf("different2 nmp %p %p\n",
5126 						    nmp, ds->nfsdev_nmp);
5127 					if (nfsaddr2_match(sad, nmp->nm_nam)) {
5128 						fndds = ds;
5129 						break;
5130 					}
5131 				}
5132 			}
5133 			if (fndds != NULL && newnmpp != NULL &&
5134 			    newnmp == NULL) {
5135 				/* Search for a place to make a mirror copy. */
5136 				TAILQ_FOREACH(tds, &nfsrv_devidhead,
5137 				    nfsdev_list) {
5138 					if (tds->nfsdev_nmp != NULL &&
5139 					    fndds != tds &&
5140 					    ((tds->nfsdev_mdsisset == 0 &&
5141 					      fndds->nfsdev_mdsisset == 0) ||
5142 					     (tds->nfsdev_mdsisset != 0 &&
5143 					      fndds->nfsdev_mdsisset != 0 &&
5144 					      fsidcmp(&tds->nfsdev_mdsfsid,
5145 					      &mp->mnt_stat.f_fsid) == 0))) {
5146 						*newnmpp = tds->nfsdev_nmp;
5147 						break;
5148 					}
5149 				}
5150 				if (tds != NULL) {
5151 					/*
5152 					 * Move this entry to the end of the
5153 					 * list, so it won't be selected as
5154 					 * easily the next time.
5155 					 */
5156 					TAILQ_REMOVE(&nfsrv_devidhead, tds,
5157 					    nfsdev_list);
5158 					TAILQ_INSERT_TAIL(&nfsrv_devidhead, tds,
5159 					    nfsdev_list);
5160 				}
5161 			}
5162 			NFSDDSUNLOCK();
5163 			if (fndds != NULL) {
5164 				dvp = fndds->nfsdev_dsdir[dsdir];
5165 				if (lktype != 0 || fhiszero != 0 ||
5166 				    (nvpp != NULL && *nvpp == NULL)) {
5167 					if (fhiszero != 0)
5168 						error = vn_lock(dvp,
5169 						    LK_EXCLUSIVE);
5170 					else if (lktype != 0)
5171 						error = vn_lock(dvp, lktype);
5172 					else
5173 						error = vn_lock(dvp, LK_SHARED);
5174 					/*
5175 					 * If the file handle is all 0's, try to
5176 					 * do a Lookup against the DS to acquire
5177 					 * it.
5178 					 * If dvpp == NULL or the Lookup fails,
5179 					 * unlock dvp after the call.
5180 					 */
5181 					if (error == 0 && (fhiszero != 0 ||
5182 					    (nvpp != NULL && *nvpp == NULL))) {
5183 						error = nfsrv_pnfslookupds(vp,
5184 						    dvp, pf, &nvp, p);
5185 						if (error == 0) {
5186 							if (fhiszero != 0)
5187 								nfsrv_pnfssetfh(
5188 								    vp, pf,
5189 								    devid,
5190 								    fnamep,
5191 								    nvp, p);
5192 							if (nvpp != NULL &&
5193 							    *nvpp == NULL) {
5194 								*nvpp = nvp;
5195 								*dsdirp = dsdir;
5196 							} else
5197 								vput(nvp);
5198 						}
5199 						if (error != 0 || lktype == 0)
5200 							NFSVOPUNLOCK(dvp);
5201 					}
5202 				}
5203 				if (error == 0) {
5204 					gotone++;
5205 					NFSD_DEBUG(4, "gotone=%d\n", gotone);
5206 					if (devid != NULL) {
5207 						NFSBCOPY(fndds->nfsdev_deviceid,
5208 						    devid, NFSX_V4DEVICEID);
5209 						devid += NFSX_V4DEVICEID;
5210 					}
5211 					if (dvpp != NULL)
5212 						*tdvpp++ = dvp;
5213 					if (fhp != NULL)
5214 						NFSBCOPY(&pf->dsf_fh, fhp++,
5215 						    NFSX_MYFH);
5216 					if (fnamep != NULL && gotone == 1)
5217 						strlcpy(fnamep,
5218 						    pf->dsf_filename,
5219 						    sizeof(pf->dsf_filename));
5220 				} else
5221 					NFSD_DEBUG(4, "nfsrv_dsgetsockmnt "
5222 					    "err=%d\n", error);
5223 			}
5224 		}
5225 	}
5226 	if (error == 0 && gotone == 0)
5227 		error = ENOENT;
5228 
5229 	NFSD_DEBUG(4, "eo nfsrv_dsgetsockmnt: gotone=%d err=%d\n", gotone,
5230 	    error);
5231 	if (error == 0)
5232 		*mirrorcntp = gotone;
5233 	else {
5234 		if (gotone > 0 && dvpp != NULL) {
5235 			/*
5236 			 * If the error didn't occur on the first one and
5237 			 * dvpp != NULL, the one(s) prior to the failure will
5238 			 * have locked dvp's that need to be unlocked.
5239 			 */
5240 			for (i = 0; i < gotone; i++) {
5241 				NFSVOPUNLOCK(*dvpp);
5242 				*dvpp++ = NULL;
5243 			}
5244 		}
5245 		/*
5246 		 * If it found the vnode to be copied from before a failure,
5247 		 * it needs to be vput()'d.
5248 		 */
5249 		if (nvpp != NULL && *nvpp != NULL) {
5250 			vput(*nvpp);
5251 			*nvpp = NULL;
5252 		}
5253 	}
5254 	return (error);
5255 }
5256 
5257 /*
5258  * Set the extended attribute for the Change attribute.
5259  */
5260 static int
5261 nfsrv_setextattr(struct vnode *vp, struct nfsvattr *nap, NFSPROC_T *p)
5262 {
5263 	struct pnfsdsattr dsattr;
5264 	int error;
5265 
5266 	ASSERT_VOP_ELOCKED(vp, "nfsrv_setextattr vp");
5267 	dsattr.dsa_filerev = nap->na_filerev;
5268 	dsattr.dsa_size = nap->na_size;
5269 	dsattr.dsa_atime = nap->na_atime;
5270 	dsattr.dsa_mtime = nap->na_mtime;
5271 	dsattr.dsa_bytes = nap->na_bytes;
5272 	error = vn_extattr_set(vp, IO_NODELOCKED, EXTATTR_NAMESPACE_SYSTEM,
5273 	    "pnfsd.dsattr", sizeof(dsattr), (char *)&dsattr, p);
5274 	if (error != 0)
5275 		printf("pNFS: setextattr=%d\n", error);
5276 	return (error);
5277 }
5278 
5279 static int
5280 nfsrv_readdsrpc(fhandle_t *fhp, off_t off, int len, struct ucred *cred,
5281     NFSPROC_T *p, struct nfsmount *nmp, struct mbuf **mpp, struct mbuf **mpendp)
5282 {
5283 	uint32_t *tl;
5284 	struct nfsrv_descript *nd;
5285 	nfsv4stateid_t st;
5286 	struct mbuf *m, *m2;
5287 	int error = 0, retlen, tlen, trimlen;
5288 
5289 	NFSD_DEBUG(4, "in nfsrv_readdsrpc\n");
5290 	nd = malloc(sizeof(*nd), M_TEMP, M_WAITOK | M_ZERO);
5291 	*mpp = NULL;
5292 	/*
5293 	 * Use a stateid where other is an alternating 01010 pattern and
5294 	 * seqid is 0xffffffff.  This value is not defined as special by
5295 	 * the RFC and is used by the FreeBSD NFS server to indicate an
5296 	 * MDS->DS proxy operation.
5297 	 */
5298 	st.other[0] = 0x55555555;
5299 	st.other[1] = 0x55555555;
5300 	st.other[2] = 0x55555555;
5301 	st.seqid = 0xffffffff;
5302 	nfscl_reqstart(nd, NFSPROC_READDS, nmp, (u_int8_t *)fhp, sizeof(*fhp),
5303 	    NULL, NULL, 0, 0, cred);
5304 	nfsm_stateidtom(nd, &st, NFSSTATEID_PUTSTATEID);
5305 	NFSM_BUILD(tl, uint32_t *, NFSX_UNSIGNED * 3);
5306 	txdr_hyper(off, tl);
5307 	*(tl + 2) = txdr_unsigned(len);
5308 	error = newnfs_request(nd, nmp, NULL, &nmp->nm_sockreq, NULL, p, cred,
5309 	    NFS_PROG, NFS_VER4, NULL, 1, NULL, NULL);
5310 	if (error != 0) {
5311 		free(nd, M_TEMP);
5312 		return (error);
5313 	}
5314 	if (nd->nd_repstat == 0) {
5315 		NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
5316 		NFSM_STRSIZ(retlen, len);
5317 		if (retlen > 0) {
5318 			/* Trim off the pre-data XDR from the mbuf chain. */
5319 			m = nd->nd_mrep;
5320 			while (m != NULL && m != nd->nd_md) {
5321 				if (m->m_next == nd->nd_md) {
5322 					m->m_next = NULL;
5323 					m_freem(nd->nd_mrep);
5324 					nd->nd_mrep = m = nd->nd_md;
5325 				} else
5326 					m = m->m_next;
5327 			}
5328 			if (m == NULL) {
5329 				printf("nfsrv_readdsrpc: busted mbuf list\n");
5330 				error = ENOENT;
5331 				goto nfsmout;
5332 			}
5333 
5334 			/*
5335 			 * Now, adjust first mbuf so that any XDR before the
5336 			 * read data is skipped over.
5337 			 */
5338 			trimlen = nd->nd_dpos - mtod(m, char *);
5339 			if (trimlen > 0) {
5340 				m->m_len -= trimlen;
5341 				NFSM_DATAP(m, trimlen);
5342 			}
5343 
5344 			/*
5345 			 * Truncate the mbuf chain at retlen bytes of data,
5346 			 * plus XDR padding that brings the length up to a
5347 			 * multiple of 4.
5348 			 */
5349 			tlen = NFSM_RNDUP(retlen);
5350 			do {
5351 				if (m->m_len >= tlen) {
5352 					m->m_len = tlen;
5353 					tlen = 0;
5354 					m2 = m->m_next;
5355 					m->m_next = NULL;
5356 					m_freem(m2);
5357 					break;
5358 				}
5359 				tlen -= m->m_len;
5360 				m = m->m_next;
5361 			} while (m != NULL);
5362 			if (tlen > 0) {
5363 				printf("nfsrv_readdsrpc: busted mbuf list\n");
5364 				error = ENOENT;
5365 				goto nfsmout;
5366 			}
5367 			*mpp = nd->nd_mrep;
5368 			*mpendp = m;
5369 			nd->nd_mrep = NULL;
5370 		}
5371 	} else
5372 		error = nd->nd_repstat;
5373 nfsmout:
5374 	/* If nd->nd_mrep is already NULL, this is a no-op. */
5375 	m_freem(nd->nd_mrep);
5376 	free(nd, M_TEMP);
5377 	NFSD_DEBUG(4, "nfsrv_readdsrpc error=%d\n", error);
5378 	return (error);
5379 }
5380 
5381 /*
5382  * Do a write RPC on a DS data file, using this structure for the arguments,
5383  * so that this function can be executed by a separate kernel process.
5384  */
5385 struct nfsrvwritedsdorpc {
5386 	int			done;
5387 	int			inprog;
5388 	struct task		tsk;
5389 	fhandle_t		fh;
5390 	off_t			off;
5391 	int			len;
5392 	struct nfsmount		*nmp;
5393 	struct ucred		*cred;
5394 	NFSPROC_T		*p;
5395 	struct mbuf		*m;
5396 	int			err;
5397 };
5398 
5399 static int
5400 nfsrv_writedsdorpc(struct nfsmount *nmp, fhandle_t *fhp, off_t off, int len,
5401     struct nfsvattr *nap, struct mbuf *m, struct ucred *cred, NFSPROC_T *p)
5402 {
5403 	uint32_t *tl;
5404 	struct nfsrv_descript *nd;
5405 	nfsattrbit_t attrbits;
5406 	nfsv4stateid_t st;
5407 	int commit, error, retlen;
5408 
5409 	nd = malloc(sizeof(*nd), M_TEMP, M_WAITOK | M_ZERO);
5410 	nfscl_reqstart(nd, NFSPROC_WRITE, nmp, (u_int8_t *)fhp,
5411 	    sizeof(fhandle_t), NULL, NULL, 0, 0, cred);
5412 
5413 	/*
5414 	 * Use a stateid where other is an alternating 01010 pattern and
5415 	 * seqid is 0xffffffff.  This value is not defined as special by
5416 	 * the RFC and is used by the FreeBSD NFS server to indicate an
5417 	 * MDS->DS proxy operation.
5418 	 */
5419 	st.other[0] = 0x55555555;
5420 	st.other[1] = 0x55555555;
5421 	st.other[2] = 0x55555555;
5422 	st.seqid = 0xffffffff;
5423 	nfsm_stateidtom(nd, &st, NFSSTATEID_PUTSTATEID);
5424 	NFSM_BUILD(tl, u_int32_t *, NFSX_HYPER + 2 * NFSX_UNSIGNED);
5425 	txdr_hyper(off, tl);
5426 	tl += 2;
5427 	/*
5428 	 * Do all writes FileSync, since the server doesn't hold onto dirty
5429 	 * buffers.  Since clients should be accessing the DS servers directly
5430 	 * using the pNFS layouts, this just needs to work correctly as a
5431 	 * fallback.
5432 	 */
5433 	*tl++ = txdr_unsigned(NFSWRITE_FILESYNC);
5434 	*tl = txdr_unsigned(len);
5435 	NFSD_DEBUG(4, "nfsrv_writedsdorpc: len=%d\n", len);
5436 
5437 	/* Put data in mbuf chain. */
5438 	nd->nd_mb->m_next = m;
5439 
5440 	/* Set nd_mb and nd_bpos to end of data. */
5441 	while (m->m_next != NULL)
5442 		m = m->m_next;
5443 	nd->nd_mb = m;
5444 	nfsm_set(nd, m->m_len);
5445 	NFSD_DEBUG(4, "nfsrv_writedsdorpc: lastmb len=%d\n", m->m_len);
5446 
5447 	/* Do a Getattr for the attributes that change upon writing. */
5448 	NFSZERO_ATTRBIT(&attrbits);
5449 	NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_SIZE);
5450 	NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_CHANGE);
5451 	NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_TIMEACCESS);
5452 	NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_TIMEMODIFY);
5453 	NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_SPACEUSED);
5454 	NFSM_BUILD(tl, u_int32_t *, NFSX_UNSIGNED);
5455 	*tl = txdr_unsigned(NFSV4OP_GETATTR);
5456 	(void) nfsrv_putattrbit(nd, &attrbits);
5457 	error = newnfs_request(nd, nmp, NULL, &nmp->nm_sockreq, NULL, p,
5458 	    cred, NFS_PROG, NFS_VER4, NULL, 1, NULL, NULL);
5459 	if (error != 0) {
5460 		free(nd, M_TEMP);
5461 		return (error);
5462 	}
5463 	NFSD_DEBUG(4, "nfsrv_writedsdorpc: aft writerpc=%d\n", nd->nd_repstat);
5464 	/* Get rid of weak cache consistency data for now. */
5465 	if ((nd->nd_flag & (ND_NOMOREDATA | ND_NFSV4 | ND_V4WCCATTR)) ==
5466 	    (ND_NFSV4 | ND_V4WCCATTR)) {
5467 		error = nfsv4_loadattr(nd, NULL, nap, NULL, NULL, 0, NULL, NULL,
5468 		    NULL, NULL, NULL, 0, NULL, NULL, NULL, NULL, NULL);
5469 		NFSD_DEBUG(4, "nfsrv_writedsdorpc: wcc attr=%d\n", error);
5470 		if (error != 0)
5471 			goto nfsmout;
5472 		/*
5473 		 * Get rid of Op# and status for next op.
5474 		 */
5475 		NFSM_DISSECT(tl, uint32_t *, 2 * NFSX_UNSIGNED);
5476 		if (*++tl != 0)
5477 			nd->nd_flag |= ND_NOMOREDATA;
5478 	}
5479 	if (nd->nd_repstat == 0) {
5480 		NFSM_DISSECT(tl, uint32_t *, 2 * NFSX_UNSIGNED + NFSX_VERF);
5481 		retlen = fxdr_unsigned(int, *tl++);
5482 		commit = fxdr_unsigned(int, *tl);
5483 		if (commit != NFSWRITE_FILESYNC)
5484 			error = NFSERR_IO;
5485 		NFSD_DEBUG(4, "nfsrv_writedsdorpc:retlen=%d commit=%d err=%d\n",
5486 		    retlen, commit, error);
5487 	} else
5488 		error = nd->nd_repstat;
5489 	/* We have no use for the Write Verifier since we use FileSync. */
5490 
5491 	/*
5492 	 * Get the Change, Size, Access Time and Modify Time attributes and set
5493 	 * on the Metadata file, so its attributes will be what the file's
5494 	 * would be if it had been written.
5495 	 */
5496 	if (error == 0) {
5497 		NFSM_DISSECT(tl, uint32_t *, 2 * NFSX_UNSIGNED);
5498 		error = nfsv4_loadattr(nd, NULL, nap, NULL, NULL, 0, NULL, NULL,
5499 		    NULL, NULL, NULL, 0, NULL, NULL, NULL, NULL, NULL);
5500 	}
5501 	NFSD_DEBUG(4, "nfsrv_writedsdorpc: aft loadattr=%d\n", error);
5502 nfsmout:
5503 	m_freem(nd->nd_mrep);
5504 	free(nd, M_TEMP);
5505 	NFSD_DEBUG(4, "nfsrv_writedsdorpc error=%d\n", error);
5506 	return (error);
5507 }
5508 
5509 /*
5510  * Start up the thread that will execute nfsrv_writedsdorpc().
5511  */
5512 static void
5513 start_writedsdorpc(void *arg, int pending)
5514 {
5515 	struct nfsrvwritedsdorpc *drpc;
5516 
5517 	drpc = (struct nfsrvwritedsdorpc *)arg;
5518 	drpc->err = nfsrv_writedsdorpc(drpc->nmp, &drpc->fh, drpc->off,
5519 	    drpc->len, NULL, drpc->m, drpc->cred, drpc->p);
5520 	drpc->done = 1;
5521 	NFSD_DEBUG(4, "start_writedsdorpc: err=%d\n", drpc->err);
5522 }
5523 
5524 static int
5525 nfsrv_writedsrpc(fhandle_t *fhp, off_t off, int len, struct ucred *cred,
5526     NFSPROC_T *p, struct vnode *vp, struct nfsmount **nmpp, int mirrorcnt,
5527     struct mbuf **mpp, char *cp, int *failposp)
5528 {
5529 	struct nfsrvwritedsdorpc *drpc, *tdrpc = NULL;
5530 	struct nfsvattr na;
5531 	struct mbuf *m;
5532 	int error, i, offs, ret, timo;
5533 
5534 	NFSD_DEBUG(4, "in nfsrv_writedsrpc\n");
5535 	KASSERT(*mpp != NULL, ("nfsrv_writedsrpc: NULL mbuf chain"));
5536 	drpc = NULL;
5537 	if (mirrorcnt > 1)
5538 		tdrpc = drpc = malloc(sizeof(*drpc) * (mirrorcnt - 1), M_TEMP,
5539 		    M_WAITOK);
5540 
5541 	/* Calculate offset in mbuf chain that data starts. */
5542 	offs = cp - mtod(*mpp, char *);
5543 	NFSD_DEBUG(4, "nfsrv_writedsrpc: mcopy offs=%d len=%d\n", offs, len);
5544 
5545 	/*
5546 	 * Do the write RPC for every DS, using a separate kernel process
5547 	 * for every DS except the last one.
5548 	 */
5549 	error = 0;
5550 	for (i = 0; i < mirrorcnt - 1; i++, tdrpc++) {
5551 		tdrpc->done = 0;
5552 		NFSBCOPY(fhp, &tdrpc->fh, sizeof(*fhp));
5553 		tdrpc->off = off;
5554 		tdrpc->len = len;
5555 		tdrpc->nmp = *nmpp;
5556 		tdrpc->cred = cred;
5557 		tdrpc->p = p;
5558 		tdrpc->inprog = 0;
5559 		tdrpc->err = 0;
5560 		tdrpc->m = m_copym(*mpp, offs, NFSM_RNDUP(len), M_WAITOK);
5561 		ret = EIO;
5562 		if (nfs_pnfsiothreads != 0) {
5563 			ret = nfs_pnfsio(start_writedsdorpc, tdrpc);
5564 			NFSD_DEBUG(4, "nfsrv_writedsrpc: nfs_pnfsio=%d\n",
5565 			    ret);
5566 		}
5567 		if (ret != 0) {
5568 			ret = nfsrv_writedsdorpc(*nmpp, fhp, off, len, NULL,
5569 			    tdrpc->m, cred, p);
5570 			if (nfsds_failerr(ret) && *failposp == -1)
5571 				*failposp = i;
5572 			else if (error == 0 && ret != 0)
5573 				error = ret;
5574 		}
5575 		nmpp++;
5576 		fhp++;
5577 	}
5578 	m = m_copym(*mpp, offs, NFSM_RNDUP(len), M_WAITOK);
5579 	ret = nfsrv_writedsdorpc(*nmpp, fhp, off, len, &na, m, cred, p);
5580 	if (nfsds_failerr(ret) && *failposp == -1 && mirrorcnt > 1)
5581 		*failposp = mirrorcnt - 1;
5582 	else if (error == 0 && ret != 0)
5583 		error = ret;
5584 	if (error == 0)
5585 		error = nfsrv_setextattr(vp, &na, p);
5586 	NFSD_DEBUG(4, "nfsrv_writedsrpc: aft setextat=%d\n", error);
5587 	tdrpc = drpc;
5588 	timo = hz / 50;		/* Wait for 20msec. */
5589 	if (timo < 1)
5590 		timo = 1;
5591 	for (i = 0; i < mirrorcnt - 1; i++, tdrpc++) {
5592 		/* Wait for RPCs on separate threads to complete. */
5593 		while (tdrpc->inprog != 0 && tdrpc->done == 0)
5594 			tsleep(&tdrpc->tsk, PVFS, "srvwrds", timo);
5595 		if (nfsds_failerr(tdrpc->err) && *failposp == -1)
5596 			*failposp = i;
5597 		else if (error == 0 && tdrpc->err != 0)
5598 			error = tdrpc->err;
5599 	}
5600 	free(drpc, M_TEMP);
5601 	return (error);
5602 }
5603 
5604 /*
5605  * Do a allocate RPC on a DS data file, using this structure for the arguments,
5606  * so that this function can be executed by a separate kernel process.
5607  */
5608 struct nfsrvallocatedsdorpc {
5609 	int			done;
5610 	int			inprog;
5611 	struct task		tsk;
5612 	fhandle_t		fh;
5613 	off_t			off;
5614 	off_t			len;
5615 	struct nfsmount		*nmp;
5616 	struct ucred		*cred;
5617 	NFSPROC_T		*p;
5618 	int			err;
5619 };
5620 
5621 static int
5622 nfsrv_allocatedsdorpc(struct nfsmount *nmp, fhandle_t *fhp, off_t off,
5623     off_t len, struct nfsvattr *nap, struct ucred *cred, NFSPROC_T *p)
5624 {
5625 	uint32_t *tl;
5626 	struct nfsrv_descript *nd;
5627 	nfsattrbit_t attrbits;
5628 	nfsv4stateid_t st;
5629 	int error;
5630 
5631 	nd = malloc(sizeof(*nd), M_TEMP, M_WAITOK | M_ZERO);
5632 	nfscl_reqstart(nd, NFSPROC_ALLOCATE, nmp, (u_int8_t *)fhp,
5633 	    sizeof(fhandle_t), NULL, NULL, 0, 0, cred);
5634 
5635 	/*
5636 	 * Use a stateid where other is an alternating 01010 pattern and
5637 	 * seqid is 0xffffffff.  This value is not defined as special by
5638 	 * the RFC and is used by the FreeBSD NFS server to indicate an
5639 	 * MDS->DS proxy operation.
5640 	 */
5641 	st.other[0] = 0x55555555;
5642 	st.other[1] = 0x55555555;
5643 	st.other[2] = 0x55555555;
5644 	st.seqid = 0xffffffff;
5645 	nfsm_stateidtom(nd, &st, NFSSTATEID_PUTSTATEID);
5646 	NFSM_BUILD(tl, uint32_t *, 2 * NFSX_HYPER + NFSX_UNSIGNED);
5647 	txdr_hyper(off, tl); tl += 2;
5648 	txdr_hyper(len, tl); tl += 2;
5649 	NFSD_DEBUG(4, "nfsrv_allocatedsdorpc: len=%jd\n", (intmax_t)len);
5650 
5651 	*tl = txdr_unsigned(NFSV4OP_GETATTR);
5652 	NFSGETATTR_ATTRBIT(&attrbits);
5653 	nfsrv_putattrbit(nd, &attrbits);
5654 	error = newnfs_request(nd, nmp, NULL, &nmp->nm_sockreq, NULL, p,
5655 	    cred, NFS_PROG, NFS_VER4, NULL, 1, NULL, NULL);
5656 	if (error != 0) {
5657 		free(nd, M_TEMP);
5658 		return (error);
5659 	}
5660 	NFSD_DEBUG(4, "nfsrv_allocatedsdorpc: aft allocaterpc=%d\n",
5661 	    nd->nd_repstat);
5662 	if (nd->nd_repstat == 0) {
5663 		NFSM_DISSECT(tl, uint32_t *, 2 * NFSX_UNSIGNED);
5664 		error = nfsv4_loadattr(nd, NULL, nap, NULL, NULL, 0, NULL, NULL,
5665 		    NULL, NULL, NULL, 0, NULL, NULL, NULL, NULL, NULL);
5666 	} else
5667 		error = nd->nd_repstat;
5668 	NFSD_DEBUG(4, "nfsrv_allocatedsdorpc: aft loadattr=%d\n", error);
5669 nfsmout:
5670 	m_freem(nd->nd_mrep);
5671 	free(nd, M_TEMP);
5672 	NFSD_DEBUG(4, "nfsrv_allocatedsdorpc error=%d\n", error);
5673 	return (error);
5674 }
5675 
5676 /*
5677  * Start up the thread that will execute nfsrv_allocatedsdorpc().
5678  */
5679 static void
5680 start_allocatedsdorpc(void *arg, int pending)
5681 {
5682 	struct nfsrvallocatedsdorpc *drpc;
5683 
5684 	drpc = (struct nfsrvallocatedsdorpc *)arg;
5685 	drpc->err = nfsrv_allocatedsdorpc(drpc->nmp, &drpc->fh, drpc->off,
5686 	    drpc->len, NULL, drpc->cred, drpc->p);
5687 	drpc->done = 1;
5688 	NFSD_DEBUG(4, "start_allocatedsdorpc: err=%d\n", drpc->err);
5689 }
5690 
5691 static int
5692 nfsrv_allocatedsrpc(fhandle_t *fhp, off_t off, off_t len, struct ucred *cred,
5693     NFSPROC_T *p, struct vnode *vp, struct nfsmount **nmpp, int mirrorcnt,
5694     int *failposp)
5695 {
5696 	struct nfsrvallocatedsdorpc *drpc, *tdrpc = NULL;
5697 	struct nfsvattr na;
5698 	int error, i, ret, timo;
5699 
5700 	NFSD_DEBUG(4, "in nfsrv_allocatedsrpc\n");
5701 	drpc = NULL;
5702 	if (mirrorcnt > 1)
5703 		tdrpc = drpc = malloc(sizeof(*drpc) * (mirrorcnt - 1), M_TEMP,
5704 		    M_WAITOK);
5705 
5706 	/*
5707 	 * Do the allocate RPC for every DS, using a separate kernel process
5708 	 * for every DS except the last one.
5709 	 */
5710 	error = 0;
5711 	for (i = 0; i < mirrorcnt - 1; i++, tdrpc++) {
5712 		tdrpc->done = 0;
5713 		NFSBCOPY(fhp, &tdrpc->fh, sizeof(*fhp));
5714 		tdrpc->off = off;
5715 		tdrpc->len = len;
5716 		tdrpc->nmp = *nmpp;
5717 		tdrpc->cred = cred;
5718 		tdrpc->p = p;
5719 		tdrpc->inprog = 0;
5720 		tdrpc->err = 0;
5721 		ret = EIO;
5722 		if (nfs_pnfsiothreads != 0) {
5723 			ret = nfs_pnfsio(start_allocatedsdorpc, tdrpc);
5724 			NFSD_DEBUG(4, "nfsrv_allocatedsrpc: nfs_pnfsio=%d\n",
5725 			    ret);
5726 		}
5727 		if (ret != 0) {
5728 			ret = nfsrv_allocatedsdorpc(*nmpp, fhp, off, len, NULL,
5729 			    cred, p);
5730 			if (nfsds_failerr(ret) && *failposp == -1)
5731 				*failposp = i;
5732 			else if (error == 0 && ret != 0)
5733 				error = ret;
5734 		}
5735 		nmpp++;
5736 		fhp++;
5737 	}
5738 	ret = nfsrv_allocatedsdorpc(*nmpp, fhp, off, len, &na, cred, p);
5739 	if (nfsds_failerr(ret) && *failposp == -1 && mirrorcnt > 1)
5740 		*failposp = mirrorcnt - 1;
5741 	else if (error == 0 && ret != 0)
5742 		error = ret;
5743 	if (error == 0)
5744 		error = nfsrv_setextattr(vp, &na, p);
5745 	NFSD_DEBUG(4, "nfsrv_allocatedsrpc: aft setextat=%d\n", error);
5746 	tdrpc = drpc;
5747 	timo = hz / 50;		/* Wait for 20msec. */
5748 	if (timo < 1)
5749 		timo = 1;
5750 	for (i = 0; i < mirrorcnt - 1; i++, tdrpc++) {
5751 		/* Wait for RPCs on separate threads to complete. */
5752 		while (tdrpc->inprog != 0 && tdrpc->done == 0)
5753 			tsleep(&tdrpc->tsk, PVFS, "srvalds", timo);
5754 		if (nfsds_failerr(tdrpc->err) && *failposp == -1)
5755 			*failposp = i;
5756 		else if (error == 0 && tdrpc->err != 0)
5757 			error = tdrpc->err;
5758 	}
5759 	free(drpc, M_TEMP);
5760 	return (error);
5761 }
5762 
5763 /*
5764  * Do a deallocate RPC on a DS data file, using this structure for the
5765  * arguments, so that this function can be executed by a separate kernel
5766  * process.
5767  */
5768 struct nfsrvdeallocatedsdorpc {
5769 	int			done;
5770 	int			inprog;
5771 	struct task		tsk;
5772 	fhandle_t		fh;
5773 	off_t			off;
5774 	off_t			len;
5775 	struct nfsmount		*nmp;
5776 	struct ucred		*cred;
5777 	NFSPROC_T		*p;
5778 	int			err;
5779 };
5780 
5781 static int
5782 nfsrv_deallocatedsdorpc(struct nfsmount *nmp, fhandle_t *fhp, off_t off,
5783     off_t len, struct nfsvattr *nap, struct ucred *cred, NFSPROC_T *p)
5784 {
5785 	uint32_t *tl;
5786 	struct nfsrv_descript *nd;
5787 	nfsattrbit_t attrbits;
5788 	nfsv4stateid_t st;
5789 	int error;
5790 
5791 	nd = malloc(sizeof(*nd), M_TEMP, M_WAITOK | M_ZERO);
5792 	nfscl_reqstart(nd, NFSPROC_DEALLOCATE, nmp, (u_int8_t *)fhp,
5793 	    sizeof(fhandle_t), NULL, NULL, 0, 0, cred);
5794 
5795 	/*
5796 	 * Use a stateid where other is an alternating 01010 pattern and
5797 	 * seqid is 0xffffffff.  This value is not defined as special by
5798 	 * the RFC and is used by the FreeBSD NFS server to indicate an
5799 	 * MDS->DS proxy operation.
5800 	 */
5801 	st.other[0] = 0x55555555;
5802 	st.other[1] = 0x55555555;
5803 	st.other[2] = 0x55555555;
5804 	st.seqid = 0xffffffff;
5805 	nfsm_stateidtom(nd, &st, NFSSTATEID_PUTSTATEID);
5806 	NFSM_BUILD(tl, uint32_t *, 2 * NFSX_HYPER + NFSX_UNSIGNED);
5807 	txdr_hyper(off, tl); tl += 2;
5808 	txdr_hyper(len, tl); tl += 2;
5809 	NFSD_DEBUG(4, "nfsrv_deallocatedsdorpc: len=%jd\n", (intmax_t)len);
5810 
5811 	/* Do a Getattr for the attributes that change upon writing. */
5812 	NFSZERO_ATTRBIT(&attrbits);
5813 	NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_SIZE);
5814 	NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_CHANGE);
5815 	NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_TIMEACCESS);
5816 	NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_TIMEMODIFY);
5817 	NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_SPACEUSED);
5818 	*tl = txdr_unsigned(NFSV4OP_GETATTR);
5819 	nfsrv_putattrbit(nd, &attrbits);
5820 	error = newnfs_request(nd, nmp, NULL, &nmp->nm_sockreq, NULL, p,
5821 	    cred, NFS_PROG, NFS_VER4, NULL, 1, NULL, NULL);
5822 	if (error != 0) {
5823 		free(nd, M_TEMP);
5824 		return (error);
5825 	}
5826 	NFSD_DEBUG(4, "nfsrv_deallocatedsdorpc: aft deallocaterpc=%d\n",
5827 	    nd->nd_repstat);
5828 	/* Get rid of weak cache consistency data for now. */
5829 	if ((nd->nd_flag & (ND_NOMOREDATA | ND_NFSV4 | ND_V4WCCATTR)) ==
5830 	    (ND_NFSV4 | ND_V4WCCATTR)) {
5831 		error = nfsv4_loadattr(nd, NULL, nap, NULL, NULL, 0, NULL, NULL,
5832 		    NULL, NULL, NULL, 0, NULL, NULL, NULL, NULL, NULL);
5833 		NFSD_DEBUG(4, "nfsrv_deallocatedsdorpc: wcc attr=%d\n", error);
5834 		if (error != 0)
5835 			goto nfsmout;
5836 		/*
5837 		 * Get rid of Op# and status for next op.
5838 		 */
5839 		NFSM_DISSECT(tl, uint32_t *, 2 * NFSX_UNSIGNED);
5840 		if (*++tl != 0)
5841 			nd->nd_flag |= ND_NOMOREDATA;
5842 	}
5843 	if (nd->nd_repstat == 0) {
5844 		NFSM_DISSECT(tl, uint32_t *, 2 * NFSX_UNSIGNED);
5845 		error = nfsv4_loadattr(nd, NULL, nap, NULL, NULL, 0, NULL, NULL,
5846 		    NULL, NULL, NULL, 0, NULL, NULL, NULL, NULL, NULL);
5847 	} else
5848 		error = nd->nd_repstat;
5849 	NFSD_DEBUG(4, "nfsrv_deallocatedsdorpc: aft loadattr=%d\n", error);
5850 nfsmout:
5851 	m_freem(nd->nd_mrep);
5852 	free(nd, M_TEMP);
5853 	NFSD_DEBUG(4, "nfsrv_deallocatedsdorpc error=%d\n", error);
5854 	return (error);
5855 }
5856 
5857 /*
5858  * Start up the thread that will execute nfsrv_deallocatedsdorpc().
5859  */
5860 static void
5861 start_deallocatedsdorpc(void *arg, int pending)
5862 {
5863 	struct nfsrvdeallocatedsdorpc *drpc;
5864 
5865 	drpc = (struct nfsrvdeallocatedsdorpc *)arg;
5866 	drpc->err = nfsrv_deallocatedsdorpc(drpc->nmp, &drpc->fh, drpc->off,
5867 	    drpc->len, NULL, drpc->cred, drpc->p);
5868 	drpc->done = 1;
5869 	NFSD_DEBUG(4, "start_deallocatedsdorpc: err=%d\n", drpc->err);
5870 }
5871 
5872 static int
5873 nfsrv_deallocatedsrpc(fhandle_t *fhp, off_t off, off_t len, struct ucred *cred,
5874     NFSPROC_T *p, struct vnode *vp, struct nfsmount **nmpp, int mirrorcnt,
5875     int *failposp)
5876 {
5877 	struct nfsrvdeallocatedsdorpc *drpc, *tdrpc = NULL;
5878 	struct nfsvattr na;
5879 	int error, i, ret, timo;
5880 
5881 	NFSD_DEBUG(4, "in nfsrv_deallocatedsrpc\n");
5882 	drpc = NULL;
5883 	if (mirrorcnt > 1)
5884 		tdrpc = drpc = malloc(sizeof(*drpc) * (mirrorcnt - 1), M_TEMP,
5885 		    M_WAITOK);
5886 
5887 	/*
5888 	 * Do the deallocate RPC for every DS, using a separate kernel process
5889 	 * for every DS except the last one.
5890 	 */
5891 	error = 0;
5892 	for (i = 0; i < mirrorcnt - 1; i++, tdrpc++) {
5893 		tdrpc->done = 0;
5894 		NFSBCOPY(fhp, &tdrpc->fh, sizeof(*fhp));
5895 		tdrpc->off = off;
5896 		tdrpc->len = len;
5897 		tdrpc->nmp = *nmpp;
5898 		tdrpc->cred = cred;
5899 		tdrpc->p = p;
5900 		tdrpc->inprog = 0;
5901 		tdrpc->err = 0;
5902 		ret = EIO;
5903 		if (nfs_pnfsiothreads != 0) {
5904 			ret = nfs_pnfsio(start_deallocatedsdorpc, tdrpc);
5905 			NFSD_DEBUG(4, "nfsrv_deallocatedsrpc: nfs_pnfsio=%d\n",
5906 			    ret);
5907 		}
5908 		if (ret != 0) {
5909 			ret = nfsrv_deallocatedsdorpc(*nmpp, fhp, off, len,
5910 			    NULL, cred, p);
5911 			if (nfsds_failerr(ret) && *failposp == -1)
5912 				*failposp = i;
5913 			else if (error == 0 && ret != 0)
5914 				error = ret;
5915 		}
5916 		nmpp++;
5917 		fhp++;
5918 	}
5919 	ret = nfsrv_deallocatedsdorpc(*nmpp, fhp, off, len, &na, cred, p);
5920 	if (nfsds_failerr(ret) && *failposp == -1 && mirrorcnt > 1)
5921 		*failposp = mirrorcnt - 1;
5922 	else if (error == 0 && ret != 0)
5923 		error = ret;
5924 	if (error == 0)
5925 		error = nfsrv_setextattr(vp, &na, p);
5926 	NFSD_DEBUG(4, "nfsrv_deallocatedsrpc: aft setextat=%d\n", error);
5927 	tdrpc = drpc;
5928 	timo = hz / 50;		/* Wait for 20msec. */
5929 	if (timo < 1)
5930 		timo = 1;
5931 	for (i = 0; i < mirrorcnt - 1; i++, tdrpc++) {
5932 		/* Wait for RPCs on separate threads to complete. */
5933 		while (tdrpc->inprog != 0 && tdrpc->done == 0)
5934 			tsleep(&tdrpc->tsk, PVFS, "srvalds", timo);
5935 		if (nfsds_failerr(tdrpc->err) && *failposp == -1)
5936 			*failposp = i;
5937 		else if (error == 0 && tdrpc->err != 0)
5938 			error = tdrpc->err;
5939 	}
5940 	free(drpc, M_TEMP);
5941 	return (error);
5942 }
5943 
5944 static int
5945 nfsrv_setattrdsdorpc(fhandle_t *fhp, struct ucred *cred, NFSPROC_T *p,
5946     struct vnode *vp, struct nfsmount *nmp, struct nfsvattr *nap,
5947     struct nfsvattr *dsnap)
5948 {
5949 	uint32_t *tl;
5950 	struct nfsrv_descript *nd;
5951 	nfsv4stateid_t st;
5952 	nfsattrbit_t attrbits;
5953 	int error;
5954 
5955 	NFSD_DEBUG(4, "in nfsrv_setattrdsdorpc\n");
5956 	nd = malloc(sizeof(*nd), M_TEMP, M_WAITOK | M_ZERO);
5957 	/*
5958 	 * Use a stateid where other is an alternating 01010 pattern and
5959 	 * seqid is 0xffffffff.  This value is not defined as special by
5960 	 * the RFC and is used by the FreeBSD NFS server to indicate an
5961 	 * MDS->DS proxy operation.
5962 	 */
5963 	st.other[0] = 0x55555555;
5964 	st.other[1] = 0x55555555;
5965 	st.other[2] = 0x55555555;
5966 	st.seqid = 0xffffffff;
5967 	nfscl_reqstart(nd, NFSPROC_SETATTR, nmp, (u_int8_t *)fhp, sizeof(*fhp),
5968 	    NULL, NULL, 0, 0, cred);
5969 	nfsm_stateidtom(nd, &st, NFSSTATEID_PUTSTATEID);
5970 	nfscl_fillsattr(nd, &nap->na_vattr, vp, NFSSATTR_FULL, 0);
5971 
5972 	/* Do a Getattr for the attributes that change due to writing. */
5973 	NFSZERO_ATTRBIT(&attrbits);
5974 	NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_SIZE);
5975 	NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_CHANGE);
5976 	NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_TIMEACCESS);
5977 	NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_TIMEMODIFY);
5978 	NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_SPACEUSED);
5979 	NFSM_BUILD(tl, u_int32_t *, NFSX_UNSIGNED);
5980 	*tl = txdr_unsigned(NFSV4OP_GETATTR);
5981 	(void) nfsrv_putattrbit(nd, &attrbits);
5982 	error = newnfs_request(nd, nmp, NULL, &nmp->nm_sockreq, NULL, p, cred,
5983 	    NFS_PROG, NFS_VER4, NULL, 1, NULL, NULL);
5984 	if (error != 0) {
5985 		free(nd, M_TEMP);
5986 		return (error);
5987 	}
5988 	NFSD_DEBUG(4, "nfsrv_setattrdsdorpc: aft setattrrpc=%d\n",
5989 	    nd->nd_repstat);
5990 	/* Get rid of weak cache consistency data for now. */
5991 	if ((nd->nd_flag & (ND_NOMOREDATA | ND_NFSV4 | ND_V4WCCATTR)) ==
5992 	    (ND_NFSV4 | ND_V4WCCATTR)) {
5993 		error = nfsv4_loadattr(nd, NULL, dsnap, NULL, NULL, 0, NULL,
5994 		    NULL, NULL, NULL, NULL, 0, NULL, NULL, NULL, NULL, NULL);
5995 		NFSD_DEBUG(4, "nfsrv_setattrdsdorpc: wcc attr=%d\n", error);
5996 		if (error != 0)
5997 			goto nfsmout;
5998 		/*
5999 		 * Get rid of Op# and status for next op.
6000 		 */
6001 		NFSM_DISSECT(tl, uint32_t *, 2 * NFSX_UNSIGNED);
6002 		if (*++tl != 0)
6003 			nd->nd_flag |= ND_NOMOREDATA;
6004 	}
6005 	error = nfsrv_getattrbits(nd, &attrbits, NULL, NULL);
6006 	if (error != 0)
6007 		goto nfsmout;
6008 	if (nd->nd_repstat != 0)
6009 		error = nd->nd_repstat;
6010 	/*
6011 	 * Get the Change, Size, Access Time and Modify Time attributes and set
6012 	 * on the Metadata file, so its attributes will be what the file's
6013 	 * would be if it had been written.
6014 	 */
6015 	if (error == 0) {
6016 		NFSM_DISSECT(tl, uint32_t *, 2 * NFSX_UNSIGNED);
6017 		error = nfsv4_loadattr(nd, NULL, dsnap, NULL, NULL, 0, NULL,
6018 		    NULL, NULL, NULL, NULL, 0, NULL, NULL, NULL, NULL, NULL);
6019 	}
6020 	NFSD_DEBUG(4, "nfsrv_setattrdsdorpc: aft setattr loadattr=%d\n", error);
6021 nfsmout:
6022 	m_freem(nd->nd_mrep);
6023 	free(nd, M_TEMP);
6024 	NFSD_DEBUG(4, "nfsrv_setattrdsdorpc error=%d\n", error);
6025 	return (error);
6026 }
6027 
6028 struct nfsrvsetattrdsdorpc {
6029 	int			done;
6030 	int			inprog;
6031 	struct task		tsk;
6032 	fhandle_t		fh;
6033 	struct nfsmount		*nmp;
6034 	struct vnode		*vp;
6035 	struct ucred		*cred;
6036 	NFSPROC_T		*p;
6037 	struct nfsvattr		na;
6038 	struct nfsvattr		dsna;
6039 	int			err;
6040 };
6041 
6042 /*
6043  * Start up the thread that will execute nfsrv_setattrdsdorpc().
6044  */
6045 static void
6046 start_setattrdsdorpc(void *arg, int pending)
6047 {
6048 	struct nfsrvsetattrdsdorpc *drpc;
6049 
6050 	drpc = (struct nfsrvsetattrdsdorpc *)arg;
6051 	drpc->err = nfsrv_setattrdsdorpc(&drpc->fh, drpc->cred, drpc->p,
6052 	    drpc->vp, drpc->nmp, &drpc->na, &drpc->dsna);
6053 	drpc->done = 1;
6054 }
6055 
6056 static int
6057 nfsrv_setattrdsrpc(fhandle_t *fhp, struct ucred *cred, NFSPROC_T *p,
6058     struct vnode *vp, struct nfsmount **nmpp, int mirrorcnt,
6059     struct nfsvattr *nap, int *failposp)
6060 {
6061 	struct nfsrvsetattrdsdorpc *drpc, *tdrpc = NULL;
6062 	struct nfsvattr na;
6063 	int error, i, ret, timo;
6064 
6065 	NFSD_DEBUG(4, "in nfsrv_setattrdsrpc\n");
6066 	drpc = NULL;
6067 	if (mirrorcnt > 1)
6068 		tdrpc = drpc = malloc(sizeof(*drpc) * (mirrorcnt - 1), M_TEMP,
6069 		    M_WAITOK);
6070 
6071 	/*
6072 	 * Do the setattr RPC for every DS, using a separate kernel process
6073 	 * for every DS except the last one.
6074 	 */
6075 	error = 0;
6076 	for (i = 0; i < mirrorcnt - 1; i++, tdrpc++) {
6077 		tdrpc->done = 0;
6078 		tdrpc->inprog = 0;
6079 		NFSBCOPY(fhp, &tdrpc->fh, sizeof(*fhp));
6080 		tdrpc->nmp = *nmpp;
6081 		tdrpc->vp = vp;
6082 		tdrpc->cred = cred;
6083 		tdrpc->p = p;
6084 		tdrpc->na = *nap;
6085 		tdrpc->err = 0;
6086 		ret = EIO;
6087 		if (nfs_pnfsiothreads != 0) {
6088 			ret = nfs_pnfsio(start_setattrdsdorpc, tdrpc);
6089 			NFSD_DEBUG(4, "nfsrv_setattrdsrpc: nfs_pnfsio=%d\n",
6090 			    ret);
6091 		}
6092 		if (ret != 0) {
6093 			ret = nfsrv_setattrdsdorpc(fhp, cred, p, vp, *nmpp, nap,
6094 			    &na);
6095 			if (nfsds_failerr(ret) && *failposp == -1)
6096 				*failposp = i;
6097 			else if (error == 0 && ret != 0)
6098 				error = ret;
6099 		}
6100 		nmpp++;
6101 		fhp++;
6102 	}
6103 	ret = nfsrv_setattrdsdorpc(fhp, cred, p, vp, *nmpp, nap, &na);
6104 	if (nfsds_failerr(ret) && *failposp == -1 && mirrorcnt > 1)
6105 		*failposp = mirrorcnt - 1;
6106 	else if (error == 0 && ret != 0)
6107 		error = ret;
6108 	if (error == 0)
6109 		error = nfsrv_setextattr(vp, &na, p);
6110 	NFSD_DEBUG(4, "nfsrv_setattrdsrpc: aft setextat=%d\n", error);
6111 	tdrpc = drpc;
6112 	timo = hz / 50;		/* Wait for 20msec. */
6113 	if (timo < 1)
6114 		timo = 1;
6115 	for (i = 0; i < mirrorcnt - 1; i++, tdrpc++) {
6116 		/* Wait for RPCs on separate threads to complete. */
6117 		while (tdrpc->inprog != 0 && tdrpc->done == 0)
6118 			tsleep(&tdrpc->tsk, PVFS, "srvsads", timo);
6119 		if (nfsds_failerr(tdrpc->err) && *failposp == -1)
6120 			*failposp = i;
6121 		else if (error == 0 && tdrpc->err != 0)
6122 			error = tdrpc->err;
6123 	}
6124 	free(drpc, M_TEMP);
6125 	return (error);
6126 }
6127 
6128 /*
6129  * Do a Setattr of an NFSv4 ACL on the DS file.
6130  */
6131 static int
6132 nfsrv_setacldsdorpc(fhandle_t *fhp, struct ucred *cred, NFSPROC_T *p,
6133     struct vnode *vp, struct nfsmount *nmp, struct acl *aclp)
6134 {
6135 	struct nfsrv_descript *nd;
6136 	nfsv4stateid_t st;
6137 	nfsattrbit_t attrbits;
6138 	int error;
6139 
6140 	NFSD_DEBUG(4, "in nfsrv_setacldsdorpc\n");
6141 	nd = malloc(sizeof(*nd), M_TEMP, M_WAITOK | M_ZERO);
6142 	/*
6143 	 * Use a stateid where other is an alternating 01010 pattern and
6144 	 * seqid is 0xffffffff.  This value is not defined as special by
6145 	 * the RFC and is used by the FreeBSD NFS server to indicate an
6146 	 * MDS->DS proxy operation.
6147 	 */
6148 	st.other[0] = 0x55555555;
6149 	st.other[1] = 0x55555555;
6150 	st.other[2] = 0x55555555;
6151 	st.seqid = 0xffffffff;
6152 	nfscl_reqstart(nd, NFSPROC_SETACL, nmp, (u_int8_t *)fhp, sizeof(*fhp),
6153 	    NULL, NULL, 0, 0, cred);
6154 	nfsm_stateidtom(nd, &st, NFSSTATEID_PUTSTATEID);
6155 	NFSZERO_ATTRBIT(&attrbits);
6156 	NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_ACL);
6157 	/*
6158 	 * The "vp" argument to nfsv4_fillattr() is only used for vnode_type(),
6159 	 * so passing in the metadata "vp" will be ok, since it is of
6160 	 * the same type (VREG).
6161 	 */
6162 	nfsv4_fillattr(nd, NULL, vp, aclp, NULL, NULL, 0, &attrbits, NULL,
6163 	    NULL, 0, 0, 0, 0, 0, NULL);
6164 	error = newnfs_request(nd, nmp, NULL, &nmp->nm_sockreq, NULL, p, cred,
6165 	    NFS_PROG, NFS_VER4, NULL, 1, NULL, NULL);
6166 	if (error != 0) {
6167 		free(nd, M_TEMP);
6168 		return (error);
6169 	}
6170 	NFSD_DEBUG(4, "nfsrv_setacldsdorpc: aft setaclrpc=%d\n",
6171 	    nd->nd_repstat);
6172 	error = nd->nd_repstat;
6173 	m_freem(nd->nd_mrep);
6174 	free(nd, M_TEMP);
6175 	return (error);
6176 }
6177 
6178 struct nfsrvsetacldsdorpc {
6179 	int			done;
6180 	int			inprog;
6181 	struct task		tsk;
6182 	fhandle_t		fh;
6183 	struct nfsmount		*nmp;
6184 	struct vnode		*vp;
6185 	struct ucred		*cred;
6186 	NFSPROC_T		*p;
6187 	struct acl		*aclp;
6188 	int			err;
6189 };
6190 
6191 /*
6192  * Start up the thread that will execute nfsrv_setacldsdorpc().
6193  */
6194 static void
6195 start_setacldsdorpc(void *arg, int pending)
6196 {
6197 	struct nfsrvsetacldsdorpc *drpc;
6198 
6199 	drpc = (struct nfsrvsetacldsdorpc *)arg;
6200 	drpc->err = nfsrv_setacldsdorpc(&drpc->fh, drpc->cred, drpc->p,
6201 	    drpc->vp, drpc->nmp, drpc->aclp);
6202 	drpc->done = 1;
6203 }
6204 
6205 static int
6206 nfsrv_setacldsrpc(fhandle_t *fhp, struct ucred *cred, NFSPROC_T *p,
6207     struct vnode *vp, struct nfsmount **nmpp, int mirrorcnt, struct acl *aclp,
6208     int *failposp)
6209 {
6210 	struct nfsrvsetacldsdorpc *drpc, *tdrpc = NULL;
6211 	int error, i, ret, timo;
6212 
6213 	NFSD_DEBUG(4, "in nfsrv_setacldsrpc\n");
6214 	drpc = NULL;
6215 	if (mirrorcnt > 1)
6216 		tdrpc = drpc = malloc(sizeof(*drpc) * (mirrorcnt - 1), M_TEMP,
6217 		    M_WAITOK);
6218 
6219 	/*
6220 	 * Do the setattr RPC for every DS, using a separate kernel process
6221 	 * for every DS except the last one.
6222 	 */
6223 	error = 0;
6224 	for (i = 0; i < mirrorcnt - 1; i++, tdrpc++) {
6225 		tdrpc->done = 0;
6226 		tdrpc->inprog = 0;
6227 		NFSBCOPY(fhp, &tdrpc->fh, sizeof(*fhp));
6228 		tdrpc->nmp = *nmpp;
6229 		tdrpc->vp = vp;
6230 		tdrpc->cred = cred;
6231 		tdrpc->p = p;
6232 		tdrpc->aclp = aclp;
6233 		tdrpc->err = 0;
6234 		ret = EIO;
6235 		if (nfs_pnfsiothreads != 0) {
6236 			ret = nfs_pnfsio(start_setacldsdorpc, tdrpc);
6237 			NFSD_DEBUG(4, "nfsrv_setacldsrpc: nfs_pnfsio=%d\n",
6238 			    ret);
6239 		}
6240 		if (ret != 0) {
6241 			ret = nfsrv_setacldsdorpc(fhp, cred, p, vp, *nmpp,
6242 			    aclp);
6243 			if (nfsds_failerr(ret) && *failposp == -1)
6244 				*failposp = i;
6245 			else if (error == 0 && ret != 0)
6246 				error = ret;
6247 		}
6248 		nmpp++;
6249 		fhp++;
6250 	}
6251 	ret = nfsrv_setacldsdorpc(fhp, cred, p, vp, *nmpp, aclp);
6252 	if (nfsds_failerr(ret) && *failposp == -1 && mirrorcnt > 1)
6253 		*failposp = mirrorcnt - 1;
6254 	else if (error == 0 && ret != 0)
6255 		error = ret;
6256 	NFSD_DEBUG(4, "nfsrv_setacldsrpc: aft setextat=%d\n", error);
6257 	tdrpc = drpc;
6258 	timo = hz / 50;		/* Wait for 20msec. */
6259 	if (timo < 1)
6260 		timo = 1;
6261 	for (i = 0; i < mirrorcnt - 1; i++, tdrpc++) {
6262 		/* Wait for RPCs on separate threads to complete. */
6263 		while (tdrpc->inprog != 0 && tdrpc->done == 0)
6264 			tsleep(&tdrpc->tsk, PVFS, "srvacds", timo);
6265 		if (nfsds_failerr(tdrpc->err) && *failposp == -1)
6266 			*failposp = i;
6267 		else if (error == 0 && tdrpc->err != 0)
6268 			error = tdrpc->err;
6269 	}
6270 	free(drpc, M_TEMP);
6271 	return (error);
6272 }
6273 
6274 /*
6275  * Getattr call to the DS for the attributes that change due to writing.
6276  */
6277 static int
6278 nfsrv_getattrdsrpc(fhandle_t *fhp, struct ucred *cred, NFSPROC_T *p,
6279     struct vnode *vp, struct nfsmount *nmp, struct nfsvattr *nap)
6280 {
6281 	struct nfsrv_descript *nd;
6282 	int error;
6283 	nfsattrbit_t attrbits;
6284 
6285 	NFSD_DEBUG(4, "in nfsrv_getattrdsrpc\n");
6286 	nd = malloc(sizeof(*nd), M_TEMP, M_WAITOK | M_ZERO);
6287 	nfscl_reqstart(nd, NFSPROC_GETATTR, nmp, (u_int8_t *)fhp,
6288 	    sizeof(fhandle_t), NULL, NULL, 0, 0, cred);
6289 	NFSZERO_ATTRBIT(&attrbits);
6290 	NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_SIZE);
6291 	NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_CHANGE);
6292 	NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_TIMEACCESS);
6293 	NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_TIMEMODIFY);
6294 	NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_SPACEUSED);
6295 	(void) nfsrv_putattrbit(nd, &attrbits);
6296 	error = newnfs_request(nd, nmp, NULL, &nmp->nm_sockreq, NULL, p, cred,
6297 	    NFS_PROG, NFS_VER4, NULL, 1, NULL, NULL);
6298 	if (error != 0) {
6299 		free(nd, M_TEMP);
6300 		return (error);
6301 	}
6302 	NFSD_DEBUG(4, "nfsrv_getattrdsrpc: aft getattrrpc=%d\n",
6303 	    nd->nd_repstat);
6304 	if (nd->nd_repstat == 0) {
6305 		error = nfsv4_loadattr(nd, NULL, nap, NULL, NULL, 0,
6306 		    NULL, NULL, NULL, NULL, NULL, 0, NULL, NULL, NULL,
6307 		    NULL, NULL);
6308 		/*
6309 		 * We can only save the updated values in the extended
6310 		 * attribute if the vp is exclusively locked.
6311 		 * This should happen when any of the following operations
6312 		 * occur on the vnode:
6313 		 *    Close, Delegreturn, LayoutCommit, LayoutReturn
6314 		 * As such, the updated extended attribute should get saved
6315 		 * before nfsrv_checkdsattr() returns 0 and allows the cached
6316 		 * attributes to be returned without calling this function.
6317 		 */
6318 		if (error == 0 && VOP_ISLOCKED(vp) == LK_EXCLUSIVE) {
6319 			error = nfsrv_setextattr(vp, nap, p);
6320 			NFSD_DEBUG(4, "nfsrv_getattrdsrpc: aft setextat=%d\n",
6321 			    error);
6322 		}
6323 	} else
6324 		error = nd->nd_repstat;
6325 	m_freem(nd->nd_mrep);
6326 	free(nd, M_TEMP);
6327 	NFSD_DEBUG(4, "nfsrv_getattrdsrpc error=%d\n", error);
6328 	return (error);
6329 }
6330 
6331 /*
6332  * Seek call to a DS.
6333  */
6334 static int
6335 nfsrv_seekdsrpc(fhandle_t *fhp, off_t *offp, int content, bool *eofp,
6336     struct ucred *cred, NFSPROC_T *p, struct nfsmount *nmp)
6337 {
6338 	uint32_t *tl;
6339 	struct nfsrv_descript *nd;
6340 	nfsv4stateid_t st;
6341 	int error;
6342 
6343 	NFSD_DEBUG(4, "in nfsrv_seekdsrpc\n");
6344 	/*
6345 	 * Use a stateid where other is an alternating 01010 pattern and
6346 	 * seqid is 0xffffffff.  This value is not defined as special by
6347 	 * the RFC and is used by the FreeBSD NFS server to indicate an
6348 	 * MDS->DS proxy operation.
6349 	 */
6350 	st.other[0] = 0x55555555;
6351 	st.other[1] = 0x55555555;
6352 	st.other[2] = 0x55555555;
6353 	st.seqid = 0xffffffff;
6354 	nd = malloc(sizeof(*nd), M_TEMP, M_WAITOK | M_ZERO);
6355 	nfscl_reqstart(nd, NFSPROC_SEEKDS, nmp, (u_int8_t *)fhp,
6356 	    sizeof(fhandle_t), NULL, NULL, 0, 0, cred);
6357 	nfsm_stateidtom(nd, &st, NFSSTATEID_PUTSTATEID);
6358 	NFSM_BUILD(tl, uint32_t *, NFSX_HYPER + NFSX_UNSIGNED);
6359 	txdr_hyper(*offp, tl); tl += 2;
6360 	*tl = txdr_unsigned(content);
6361 	error = newnfs_request(nd, nmp, NULL, &nmp->nm_sockreq, NULL, p, cred,
6362 	    NFS_PROG, NFS_VER4, NULL, 1, NULL, NULL);
6363 	if (error != 0) {
6364 		free(nd, M_TEMP);
6365 		return (error);
6366 	}
6367 	NFSD_DEBUG(4, "nfsrv_seekdsrpc: aft seekrpc=%d\n", nd->nd_repstat);
6368 	if (nd->nd_repstat == 0) {
6369 		NFSM_DISSECT(tl, uint32_t *, NFSX_UNSIGNED + NFSX_HYPER);
6370 		if (*tl++ == newnfs_true)
6371 			*eofp = true;
6372 		else
6373 			*eofp = false;
6374 		*offp = fxdr_hyper(tl);
6375 	} else
6376 		error = nd->nd_repstat;
6377 nfsmout:
6378 	m_freem(nd->nd_mrep);
6379 	free(nd, M_TEMP);
6380 	NFSD_DEBUG(4, "nfsrv_seekdsrpc error=%d\n", error);
6381 	return (error);
6382 }
6383 
6384 /*
6385  * Get the device id and file handle for a DS file.
6386  */
6387 int
6388 nfsrv_dsgetdevandfh(struct vnode *vp, NFSPROC_T *p, int *mirrorcntp,
6389     fhandle_t *fhp, char *devid)
6390 {
6391 	int buflen, error;
6392 	char *buf;
6393 
6394 	buflen = 1024;
6395 	buf = malloc(buflen, M_TEMP, M_WAITOK);
6396 	error = nfsrv_dsgetsockmnt(vp, 0, buf, &buflen, mirrorcntp, p, NULL,
6397 	    fhp, devid, NULL, NULL, NULL, NULL, NULL, NULL);
6398 	free(buf, M_TEMP);
6399 	return (error);
6400 }
6401 
6402 /*
6403  * Do a Lookup against the DS for the filename.
6404  */
6405 static int
6406 nfsrv_pnfslookupds(struct vnode *vp, struct vnode *dvp, struct pnfsdsfile *pf,
6407     struct vnode **nvpp, NFSPROC_T *p)
6408 {
6409 	struct nameidata named;
6410 	struct ucred *tcred;
6411 	char *bufp;
6412 	u_long *hashp;
6413 	struct vnode *nvp;
6414 	int error;
6415 
6416 	tcred = newnfs_getcred();
6417 	named.ni_cnd.cn_nameiop = LOOKUP;
6418 	named.ni_cnd.cn_lkflags = LK_SHARED | LK_RETRY;
6419 	named.ni_cnd.cn_cred = tcred;
6420 	named.ni_cnd.cn_flags = ISLASTCN | LOCKPARENT | LOCKLEAF;
6421 	nfsvno_setpathbuf(&named, &bufp, &hashp);
6422 	named.ni_cnd.cn_nameptr = bufp;
6423 	named.ni_cnd.cn_namelen = strlen(pf->dsf_filename);
6424 	strlcpy(bufp, pf->dsf_filename, NAME_MAX);
6425 	NFSD_DEBUG(4, "nfsrv_pnfslookupds: filename=%s\n", bufp);
6426 	error = VOP_LOOKUP(dvp, &nvp, &named.ni_cnd);
6427 	NFSD_DEBUG(4, "nfsrv_pnfslookupds: aft LOOKUP=%d\n", error);
6428 	NFSFREECRED(tcred);
6429 	nfsvno_relpathbuf(&named);
6430 	if (error == 0)
6431 		*nvpp = nvp;
6432 	NFSD_DEBUG(4, "eo nfsrv_pnfslookupds=%d\n", error);
6433 	return (error);
6434 }
6435 
6436 /*
6437  * Set the file handle to the correct one.
6438  */
6439 static void
6440 nfsrv_pnfssetfh(struct vnode *vp, struct pnfsdsfile *pf, char *devid,
6441     char *fnamep, struct vnode *nvp, NFSPROC_T *p)
6442 {
6443 	struct nfsnode *np;
6444 	int ret = 0;
6445 
6446 	np = VTONFS(nvp);
6447 	NFSBCOPY(np->n_fhp->nfh_fh, &pf->dsf_fh, NFSX_MYFH);
6448 	/*
6449 	 * We can only do a vn_set_extattr() if the vnode is exclusively
6450 	 * locked and vn_start_write() has been done.  If devid != NULL or
6451 	 * fnamep != NULL or the vnode is shared locked, vn_start_write()
6452 	 * may not have been done.
6453 	 * If not done now, it will be done on a future call.
6454 	 */
6455 	if (devid == NULL && fnamep == NULL && NFSVOPISLOCKED(vp) ==
6456 	    LK_EXCLUSIVE)
6457 		ret = vn_extattr_set(vp, IO_NODELOCKED,
6458 		    EXTATTR_NAMESPACE_SYSTEM, "pnfsd.dsfile", sizeof(*pf),
6459 		    (char *)pf, p);
6460 	NFSD_DEBUG(4, "eo nfsrv_pnfssetfh=%d\n", ret);
6461 }
6462 
6463 /*
6464  * Cause RPCs waiting on "nmp" to fail.  This is called for a DS mount point
6465  * when the DS has failed.
6466  */
6467 void
6468 nfsrv_killrpcs(struct nfsmount *nmp)
6469 {
6470 
6471 	/*
6472 	 * Call newnfs_nmcancelreqs() to cause
6473 	 * any RPCs in progress on the mount point to
6474 	 * fail.
6475 	 * This will cause any process waiting for an
6476 	 * RPC to complete while holding a vnode lock
6477 	 * on the mounted-on vnode (such as "df" or
6478 	 * a non-forced "umount") to fail.
6479 	 * This will unlock the mounted-on vnode so
6480 	 * a forced dismount can succeed.
6481 	 * The NFSMNTP_CANCELRPCS flag should be set when this function is
6482 	 * called.
6483 	 */
6484 	newnfs_nmcancelreqs(nmp);
6485 }
6486 
6487 /*
6488  * Sum up the statfs info for each of the DSs, so that the client will
6489  * receive the total for all DSs.
6490  */
6491 static int
6492 nfsrv_pnfsstatfs(struct statfs *sf, struct mount *mp)
6493 {
6494 	struct statfs *tsf;
6495 	struct nfsdevice *ds;
6496 	struct vnode **dvpp, **tdvpp, *dvp;
6497 	uint64_t tot;
6498 	int cnt, error = 0, i;
6499 
6500 	if (nfsrv_devidcnt <= 0)
6501 		return (ENXIO);
6502 	dvpp = mallocarray(nfsrv_devidcnt, sizeof(*dvpp), M_TEMP, M_WAITOK);
6503 	tsf = malloc(sizeof(*tsf), M_TEMP, M_WAITOK);
6504 
6505 	/* Get an array of the dvps for the DSs. */
6506 	tdvpp = dvpp;
6507 	i = 0;
6508 	NFSDDSLOCK();
6509 	/* First, search for matches for same file system. */
6510 	TAILQ_FOREACH(ds, &nfsrv_devidhead, nfsdev_list) {
6511 		if (ds->nfsdev_nmp != NULL && ds->nfsdev_mdsisset != 0 &&
6512 		    fsidcmp(&ds->nfsdev_mdsfsid, &mp->mnt_stat.f_fsid) == 0) {
6513 			if (++i > nfsrv_devidcnt)
6514 				break;
6515 			*tdvpp++ = ds->nfsdev_dvp;
6516 		}
6517 	}
6518 	/*
6519 	 * If no matches for same file system, total all servers not assigned
6520 	 * to a file system.
6521 	 */
6522 	if (i == 0) {
6523 		TAILQ_FOREACH(ds, &nfsrv_devidhead, nfsdev_list) {
6524 			if (ds->nfsdev_nmp != NULL &&
6525 			    ds->nfsdev_mdsisset == 0) {
6526 				if (++i > nfsrv_devidcnt)
6527 					break;
6528 				*tdvpp++ = ds->nfsdev_dvp;
6529 			}
6530 		}
6531 	}
6532 	NFSDDSUNLOCK();
6533 	cnt = i;
6534 
6535 	/* Do a VFS_STATFS() for each of the DSs and sum them up. */
6536 	tdvpp = dvpp;
6537 	for (i = 0; i < cnt && error == 0; i++) {
6538 		dvp = *tdvpp++;
6539 		error = VFS_STATFS(dvp->v_mount, tsf);
6540 		if (error == 0) {
6541 			if (sf->f_bsize == 0) {
6542 				if (tsf->f_bsize > 0)
6543 					sf->f_bsize = tsf->f_bsize;
6544 				else
6545 					sf->f_bsize = 8192;
6546 			}
6547 			if (tsf->f_blocks > 0) {
6548 				if (sf->f_bsize != tsf->f_bsize) {
6549 					tot = tsf->f_blocks * tsf->f_bsize;
6550 					sf->f_blocks += (tot / sf->f_bsize);
6551 				} else
6552 					sf->f_blocks += tsf->f_blocks;
6553 			}
6554 			if (tsf->f_bfree > 0) {
6555 				if (sf->f_bsize != tsf->f_bsize) {
6556 					tot = tsf->f_bfree * tsf->f_bsize;
6557 					sf->f_bfree += (tot / sf->f_bsize);
6558 				} else
6559 					sf->f_bfree += tsf->f_bfree;
6560 			}
6561 			if (tsf->f_bavail > 0) {
6562 				if (sf->f_bsize != tsf->f_bsize) {
6563 					tot = tsf->f_bavail * tsf->f_bsize;
6564 					sf->f_bavail += (tot / sf->f_bsize);
6565 				} else
6566 					sf->f_bavail += tsf->f_bavail;
6567 			}
6568 		}
6569 	}
6570 	free(tsf, M_TEMP);
6571 	free(dvpp, M_TEMP);
6572 	return (error);
6573 }
6574 
6575 /*
6576  * Set an NFSv4 acl.
6577  */
6578 int
6579 nfsrv_setacl(struct vnode *vp, NFSACL_T *aclp, struct ucred *cred, NFSPROC_T *p)
6580 {
6581 	int error;
6582 
6583 	if (nfsrv_useacl == 0 || nfs_supportsnfsv4acls(vp) == 0) {
6584 		error = NFSERR_ATTRNOTSUPP;
6585 		goto out;
6586 	}
6587 	/*
6588 	 * With NFSv4 ACLs, chmod(2) may need to add additional entries.
6589 	 * Make sure it has enough room for that - splitting every entry
6590 	 * into two and appending "canonical six" entries at the end.
6591 	 * Cribbed out of kern/vfs_acl.c - Rick M.
6592 	 */
6593 	if (aclp->acl_cnt > (ACL_MAX_ENTRIES - 6) / 2) {
6594 		error = NFSERR_ATTRNOTSUPP;
6595 		goto out;
6596 	}
6597 	error = VOP_SETACL(vp, ACL_TYPE_NFS4, aclp, cred, p);
6598 	if (error == 0) {
6599 		error = nfsrv_dssetacl(vp, aclp, cred, p);
6600 		if (error == ENOENT)
6601 			error = 0;
6602 	}
6603 
6604 out:
6605 	NFSEXITCODE(error);
6606 	return (error);
6607 }
6608 
6609 /*
6610  * Seek vnode op call (actually it is a VOP_IOCTL()).
6611  * This function is called with the vnode locked, but unlocks and vrele()s
6612  * the vp before returning.
6613  */
6614 int
6615 nfsvno_seek(struct nfsrv_descript *nd, struct vnode *vp, u_long cmd,
6616     off_t *offp, int content, bool *eofp, struct ucred *cred, NFSPROC_T *p)
6617 {
6618 	struct nfsvattr at;
6619 	int error, ret;
6620 
6621 	ASSERT_VOP_LOCKED(vp, "nfsvno_seek vp");
6622 	/*
6623 	 * Attempt to seek on a DS file. A return of ENOENT implies
6624 	 * there is no DS file to seek on.
6625 	 */
6626 	error = nfsrv_proxyds(vp, 0, 0, cred, p, NFSPROC_SEEKDS, NULL,
6627 	    NULL, NULL, NULL, NULL, offp, content, eofp);
6628 	if (error != ENOENT) {
6629 		vput(vp);
6630 		return (error);
6631 	}
6632 
6633 	/*
6634 	 * Do the VOP_IOCTL() call.  For the case where *offp == file_size,
6635 	 * VOP_IOCTL() will return ENXIO.  However, the correct reply for
6636 	 * NFSv4.2 is *eofp == true and error == 0 for this case.
6637 	 */
6638 	NFSVOPUNLOCK(vp);
6639 	error = VOP_IOCTL(vp, cmd, offp, 0, cred, p);
6640 	*eofp = false;
6641 	if (error == ENXIO || (error == 0 && cmd == FIOSEEKHOLE)) {
6642 		/* Handle the cases where we might be at EOF. */
6643 		ret = nfsvno_getattr(vp, &at, nd, p, 0, NULL);
6644 		if (ret == 0 && *offp == at.na_size) {
6645 			*eofp = true;
6646 			error = 0;
6647 		}
6648 		if (ret != 0 && error == 0)
6649 			error = ret;
6650 	}
6651 	vrele(vp);
6652 	NFSEXITCODE(error);
6653 	return (error);
6654 }
6655 
6656 /*
6657  * Allocate vnode op call.
6658  */
6659 int
6660 nfsvno_allocate(struct vnode *vp, off_t off, off_t len, struct ucred *cred,
6661     NFSPROC_T *p)
6662 {
6663 	int error;
6664 	off_t olen;
6665 
6666 	ASSERT_VOP_ELOCKED(vp, "nfsvno_allocate vp");
6667 	/*
6668 	 * Attempt to allocate on a DS file. A return of ENOENT implies
6669 	 * there is no DS file to allocate on.
6670 	 */
6671 	error = nfsrv_proxyds(vp, off, 0, cred, p, NFSPROC_ALLOCATE, NULL,
6672 	    NULL, NULL, NULL, NULL, &len, 0, NULL);
6673 	if (error != ENOENT)
6674 		return (error);
6675 
6676 	/*
6677 	 * Do the actual VOP_ALLOCATE(), looping so long as
6678 	 * progress is being made, to achieve completion.
6679 	 */
6680 	do {
6681 		olen = len;
6682 		error = VOP_ALLOCATE(vp, &off, &len, IO_SYNC, cred);
6683 		if (error == 0 && len > 0 && olen > len)
6684 			maybe_yield();
6685 	} while (error == 0 && len > 0 && olen > len);
6686 	if (error == 0 && len > 0)
6687 		error = NFSERR_IO;
6688 	NFSEXITCODE(error);
6689 	return (error);
6690 }
6691 
6692 /*
6693  * Deallocate vnode op call.
6694  */
6695 int
6696 nfsvno_deallocate(struct vnode *vp, off_t off, off_t len, struct ucred *cred,
6697     NFSPROC_T *p)
6698 {
6699 	int error;
6700 	off_t olen;
6701 
6702 	ASSERT_VOP_ELOCKED(vp, "nfsvno_deallocate vp");
6703 	/*
6704 	 * Attempt to deallocate on a DS file. A return of ENOENT implies
6705 	 * there is no DS file to deallocate on.
6706 	 */
6707 	error = nfsrv_proxyds(vp, off, 0, cred, p, NFSPROC_DEALLOCATE, NULL,
6708 	    NULL, NULL, NULL, NULL, &len, 0, NULL);
6709 	if (error != ENOENT)
6710 		return (error);
6711 
6712 	/*
6713 	 * Do the actual VOP_DEALLOCATE(), looping so long as
6714 	 * progress is being made, to achieve completion.
6715 	 */
6716 	do {
6717 		olen = len;
6718 		error = VOP_DEALLOCATE(vp, &off, &len, 0, IO_SYNC, cred);
6719 		if (error == 0 && len > 0 && olen > len)
6720 			maybe_yield();
6721 	} while (error == 0 && len > 0 && olen > len);
6722 	if (error == 0 && len > 0)
6723 		error = NFSERR_IO;
6724 	NFSEXITCODE(error);
6725 	return (error);
6726 }
6727 
6728 /*
6729  * Get Extended Atribute vnode op into an mbuf list.
6730  */
6731 int
6732 nfsvno_getxattr(struct vnode *vp, char *name, uint32_t maxresp,
6733     struct ucred *cred, uint64_t flag, int maxextsiz, struct thread *p,
6734     struct mbuf **mpp, struct mbuf **mpendp, int *lenp)
6735 {
6736 	struct iovec *iv;
6737 	struct uio io, *uiop = &io;
6738 	struct mbuf *m, *m2;
6739 	int alen, error, len, tlen;
6740 	size_t siz;
6741 
6742 	/* First, find out the size of the extended attribute. */
6743 	error = VOP_GETEXTATTR(vp, EXTATTR_NAMESPACE_USER, name, NULL,
6744 	    &siz, cred, p);
6745 	if (error != 0)
6746 		return (NFSERR_NOXATTR);
6747 	if (siz > maxresp - NFS_MAXXDR)
6748 		return (NFSERR_XATTR2BIG);
6749 	len = siz;
6750 	tlen = NFSM_RNDUP(len);
6751 	if (tlen > 0) {
6752 		/*
6753 		 * If cnt > MCLBYTES and the reply will not be saved, use
6754 		 * ext_pgs mbufs for TLS.
6755 		 * For NFSv4.0, we do not know for sure if the reply will
6756 		 * be saved, so do not use ext_pgs mbufs for NFSv4.0.
6757 		 * Always use ext_pgs mbufs if ND_EXTPG is set.
6758 		 */
6759 		if ((flag & ND_EXTPG) != 0 || (tlen > MCLBYTES &&
6760 		    (flag & (ND_TLS | ND_SAVEREPLY)) == ND_TLS &&
6761 		    (flag & (ND_NFSV4 | ND_NFSV41)) != ND_NFSV4))
6762 			uiop->uio_iovcnt = nfsrv_createiovec_extpgs(tlen,
6763 			    maxextsiz, &m, &m2, &iv);
6764 		else
6765 			uiop->uio_iovcnt = nfsrv_createiovec(tlen, &m, &m2,
6766 			    &iv);
6767 		uiop->uio_iov = iv;
6768 	} else {
6769 		uiop->uio_iovcnt = 0;
6770 		uiop->uio_iov = iv = NULL;
6771 		m = m2 = NULL;
6772 	}
6773 	uiop->uio_offset = 0;
6774 	uiop->uio_resid = tlen;
6775 	uiop->uio_rw = UIO_READ;
6776 	uiop->uio_segflg = UIO_SYSSPACE;
6777 	uiop->uio_td = p;
6778 #ifdef MAC
6779 	error = mac_vnode_check_getextattr(cred, vp, EXTATTR_NAMESPACE_USER,
6780 	    name);
6781 	if (error != 0)
6782 		goto out;
6783 #endif
6784 
6785 	if (tlen > 0)
6786 		error = VOP_GETEXTATTR(vp, EXTATTR_NAMESPACE_USER, name, uiop,
6787 		    NULL, cred, p);
6788 	if (error != 0)
6789 		goto out;
6790 	if (uiop->uio_resid > 0) {
6791 		alen = tlen;
6792 		len = tlen - uiop->uio_resid;
6793 		tlen = NFSM_RNDUP(len);
6794 		if (alen != tlen)
6795 			printf("nfsvno_getxattr: weird size read\n");
6796 		if (tlen == 0) {
6797 			m_freem(m);
6798 			m = m2 = NULL;
6799 		} else if (alen != tlen || tlen != len)
6800 			m2 = nfsrv_adj(m, alen - tlen, tlen - len);
6801 	}
6802 	*lenp = len;
6803 	*mpp = m;
6804 	*mpendp = m2;
6805 
6806 out:
6807 	if (error != 0) {
6808 		if (m != NULL)
6809 			m_freem(m);
6810 		*lenp = 0;
6811 	}
6812 	free(iv, M_TEMP);
6813 	NFSEXITCODE(error);
6814 	return (error);
6815 }
6816 
6817 /*
6818  * Set Extended attribute vnode op from an mbuf list.
6819  */
6820 int
6821 nfsvno_setxattr(struct vnode *vp, char *name, int len, struct mbuf *m,
6822     char *cp, struct ucred *cred, struct thread *p)
6823 {
6824 	struct iovec *iv;
6825 	struct uio uio, *uiop = &uio;
6826 	int cnt, error;
6827 
6828 	error = 0;
6829 #ifdef MAC
6830 	error = mac_vnode_check_setextattr(cred, vp, EXTATTR_NAMESPACE_USER,
6831 	    name);
6832 #endif
6833 	if (error != 0)
6834 		goto out;
6835 
6836 	uiop->uio_rw = UIO_WRITE;
6837 	uiop->uio_segflg = UIO_SYSSPACE;
6838 	uiop->uio_td = p;
6839 	uiop->uio_offset = 0;
6840 	uiop->uio_resid = len;
6841 	if (len > 0) {
6842 		error = nfsrv_createiovecw(len, m, cp, &iv, &cnt);
6843 		uiop->uio_iov = iv;
6844 		uiop->uio_iovcnt = cnt;
6845 	} else {
6846 		uiop->uio_iov = iv = NULL;
6847 		uiop->uio_iovcnt = 0;
6848 	}
6849 	if (error == 0) {
6850 		error = VOP_SETEXTATTR(vp, EXTATTR_NAMESPACE_USER, name, uiop,
6851 		    cred, p);
6852 		if (error == 0) {
6853 			if (vp->v_type == VREG && nfsrv_devidcnt != 0)
6854 				nfsvno_updateds(vp, cred, p);
6855 			error = VOP_FSYNC(vp, MNT_WAIT, p);
6856 		}
6857 		free(iv, M_TEMP);
6858 	}
6859 
6860 out:
6861 	NFSEXITCODE(error);
6862 	return (error);
6863 }
6864 
6865 /*
6866  * For a pNFS server, the DS file's ctime and
6867  * va_filerev (TimeMetadata and Change) needs to
6868  * be updated.  This is a hack, but works by
6869  * flipping the S_ISGID bit in va_mode and then
6870  * flipping it back.
6871  * It does result in two MDS->DS RPCs, but creating
6872  * a custom RPC just to do this seems overkill, since
6873  * Setxattr/Rmxattr will not be done that frequently.
6874  * If it fails part way through, that is not too
6875  * serious, since the DS file is never executed.
6876  */
6877 static void
6878 nfsvno_updateds(struct vnode *vp, struct ucred *cred, NFSPROC_T *p)
6879 {
6880 	struct nfsvattr nva;
6881 	int ret;
6882 	u_short tmode;
6883 
6884 	ret = VOP_GETATTR(vp, &nva.na_vattr, cred);
6885 	if (ret == 0) {
6886 		tmode = nva.na_mode;
6887 		NFSVNO_ATTRINIT(&nva);
6888 		tmode ^= S_ISGID;
6889 		NFSVNO_SETATTRVAL(&nva, mode, tmode);
6890 		ret = nfsrv_proxyds(vp, 0, 0, cred, p,
6891 		    NFSPROC_SETATTR, NULL, NULL, NULL, &nva,
6892 		    NULL, NULL, 0, NULL);
6893 		if (ret == 0) {
6894 			tmode ^= S_ISGID;
6895 			NFSVNO_SETATTRVAL(&nva, mode, tmode);
6896 			ret = nfsrv_proxyds(vp, 0, 0, cred, p,
6897 			    NFSPROC_SETATTR, NULL, NULL, NULL,
6898 			    &nva, NULL, NULL, 0, NULL);
6899 		}
6900 	}
6901 }
6902 
6903 /*
6904  * Remove Extended attribute vnode op.
6905  */
6906 int
6907 nfsvno_rmxattr(struct nfsrv_descript *nd, struct vnode *vp, char *name,
6908     struct ucred *cred, struct thread *p)
6909 {
6910 	int error;
6911 
6912 	/*
6913 	 * Get rid of any delegations.  I am not sure why this is required,
6914 	 * but RFC-8276 says so.
6915 	 */
6916 	error = nfsrv_checkremove(vp, 0, nd, nd->nd_clientid, p);
6917 	if (error != 0)
6918 		goto out;
6919 #ifdef MAC
6920 	error = mac_vnode_check_deleteextattr(cred, vp, EXTATTR_NAMESPACE_USER,
6921 	    name);
6922 	if (error != 0)
6923 		goto out;
6924 #endif
6925 
6926 	error = VOP_DELETEEXTATTR(vp, EXTATTR_NAMESPACE_USER, name, cred, p);
6927 	if (error == EOPNOTSUPP)
6928 		error = VOP_SETEXTATTR(vp, EXTATTR_NAMESPACE_USER, name, NULL,
6929 		    cred, p);
6930 	if (error == 0) {
6931 		if (vp->v_type == VREG && nfsrv_devidcnt != 0)
6932 			nfsvno_updateds(vp, cred, p);
6933 		error = VOP_FSYNC(vp, MNT_WAIT, p);
6934 	}
6935 out:
6936 	NFSEXITCODE(error);
6937 	return (error);
6938 }
6939 
6940 /*
6941  * List Extended Atribute vnode op into an mbuf list.
6942  */
6943 int
6944 nfsvno_listxattr(struct vnode *vp, uint64_t cookie, struct ucred *cred,
6945     struct thread *p, u_char **bufp, uint32_t *lenp, bool *eofp)
6946 {
6947 	struct iovec iv;
6948 	struct uio io;
6949 	int error;
6950 	size_t siz;
6951 
6952 	*bufp = NULL;
6953 	/* First, find out the size of the extended attribute. */
6954 	error = VOP_LISTEXTATTR(vp, EXTATTR_NAMESPACE_USER, NULL, &siz, cred,
6955 	    p);
6956 	if (error != 0)
6957 		return (NFSERR_NOXATTR);
6958 	if (siz <= cookie) {
6959 		*lenp = 0;
6960 		*eofp = true;
6961 		goto out;
6962 	}
6963 	if (siz > cookie + *lenp) {
6964 		siz = cookie + *lenp;
6965 		*eofp = false;
6966 	} else
6967 		*eofp = true;
6968 	/* Just choose a sanity limit of 10Mbytes for malloc(M_TEMP). */
6969 	if (siz > 10 * 1024 * 1024) {
6970 		error = NFSERR_XATTR2BIG;
6971 		goto out;
6972 	}
6973 	*bufp = malloc(siz, M_TEMP, M_WAITOK);
6974 	iv.iov_base = *bufp;
6975 	iv.iov_len = siz;
6976 	io.uio_iovcnt = 1;
6977 	io.uio_iov = &iv;
6978 	io.uio_offset = 0;
6979 	io.uio_resid = siz;
6980 	io.uio_rw = UIO_READ;
6981 	io.uio_segflg = UIO_SYSSPACE;
6982 	io.uio_td = p;
6983 #ifdef MAC
6984 	error = mac_vnode_check_listextattr(cred, vp, EXTATTR_NAMESPACE_USER);
6985 	if (error != 0)
6986 		goto out;
6987 #endif
6988 
6989 	error = VOP_LISTEXTATTR(vp, EXTATTR_NAMESPACE_USER, &io, NULL, cred,
6990 	    p);
6991 	if (error != 0)
6992 		goto out;
6993 	if (io.uio_resid > 0)
6994 		siz -= io.uio_resid;
6995 	*lenp = siz;
6996 
6997 out:
6998 	if (error != 0) {
6999 		free(*bufp, M_TEMP);
7000 		*bufp = NULL;
7001 	}
7002 	NFSEXITCODE(error);
7003 	return (error);
7004 }
7005 
7006 /*
7007  * Trim trailing data off the mbuf list being built.
7008  */
7009 void
7010 nfsm_trimtrailing(struct nfsrv_descript *nd, struct mbuf *mb, char *bpos,
7011     int bextpg, int bextpgsiz)
7012 {
7013 	vm_page_t pg;
7014 	int fullpgsiz, i;
7015 
7016 	if (mb->m_next != NULL) {
7017 		m_freem(mb->m_next);
7018 		mb->m_next = NULL;
7019 	}
7020 	if ((mb->m_flags & M_EXTPG) != 0) {
7021 		KASSERT(bextpg >= 0 && bextpg < mb->m_epg_npgs,
7022 		    ("nfsm_trimtrailing: bextpg out of range"));
7023 		KASSERT(bpos == (char *)(void *)
7024 		    PHYS_TO_DMAP(mb->m_epg_pa[bextpg]) + PAGE_SIZE - bextpgsiz,
7025 		    ("nfsm_trimtrailing: bextpgsiz bad!"));
7026 
7027 		/* First, get rid of any pages after this position. */
7028 		for (i = mb->m_epg_npgs - 1; i > bextpg; i--) {
7029 			pg = PHYS_TO_VM_PAGE(mb->m_epg_pa[i]);
7030 			vm_page_unwire_noq(pg);
7031 			vm_page_free(pg);
7032 		}
7033 		mb->m_epg_npgs = bextpg + 1;
7034 		if (bextpg == 0)
7035 			fullpgsiz = PAGE_SIZE - mb->m_epg_1st_off;
7036 		else
7037 			fullpgsiz = PAGE_SIZE;
7038 		mb->m_epg_last_len = fullpgsiz - bextpgsiz;
7039 		mb->m_len = m_epg_pagelen(mb, 0, mb->m_epg_1st_off);
7040 		for (i = 1; i < mb->m_epg_npgs; i++)
7041 			mb->m_len += m_epg_pagelen(mb, i, 0);
7042 		nd->nd_bextpgsiz = bextpgsiz;
7043 		nd->nd_bextpg = bextpg;
7044 	} else
7045 		mb->m_len = bpos - mtod(mb, char *);
7046 	nd->nd_mb = mb;
7047 	nd->nd_bpos = bpos;
7048 }
7049 
7050 
7051 /*
7052  * Check to see if a put file handle operation should test for
7053  * NFSERR_WRONGSEC, although NFSv3 actually returns NFSERR_AUTHERR.
7054  * When Open is the next operation, NFSERR_WRONGSEC cannot be
7055  * replied for the Open cases that use a component.  This can
7056  * be identified by the fact that the file handle's type is VDIR.
7057  */
7058 bool
7059 nfsrv_checkwrongsec(struct nfsrv_descript *nd, int nextop, __enum_uint8(vtype) vtyp)
7060 {
7061 
7062 	if ((nd->nd_flag & ND_NFSV4) == 0)
7063 		return (true);
7064 
7065 	if ((nd->nd_flag & ND_LASTOP) != 0)
7066 		return (false);
7067 
7068 	if (nextop == NFSV4OP_PUTROOTFH || nextop == NFSV4OP_PUTFH ||
7069 	    nextop == NFSV4OP_PUTPUBFH || nextop == NFSV4OP_RESTOREFH ||
7070 	    nextop == NFSV4OP_LOOKUP || nextop == NFSV4OP_LOOKUPP ||
7071 	    nextop == NFSV4OP_SECINFO || nextop == NFSV4OP_SECINFONONAME)
7072 		return (false);
7073 	if (nextop == NFSV4OP_OPEN && vtyp == VDIR)
7074 		return (false);
7075 	return (true);
7076 }
7077 
7078 /*
7079  * Check DSs marked no space.
7080  */
7081 void
7082 nfsrv_checknospc(void)
7083 {
7084 	struct statfs *tsf;
7085 	struct nfsdevice *ds;
7086 	struct vnode **dvpp, **tdvpp, *dvp;
7087 	char *devid, *tdevid;
7088 	int cnt, error = 0, i;
7089 
7090 	if (nfsrv_devidcnt <= 0)
7091 		return;
7092 	dvpp = mallocarray(nfsrv_devidcnt, sizeof(*dvpp), M_TEMP, M_WAITOK);
7093 	devid = malloc(nfsrv_devidcnt * NFSX_V4DEVICEID, M_TEMP, M_WAITOK);
7094 	tsf = malloc(sizeof(*tsf), M_TEMP, M_WAITOK);
7095 
7096 	/* Get an array of the dvps for the DSs. */
7097 	tdvpp = dvpp;
7098 	tdevid = devid;
7099 	i = 0;
7100 	NFSDDSLOCK();
7101 	/* First, search for matches for same file system. */
7102 	TAILQ_FOREACH(ds, &nfsrv_devidhead, nfsdev_list) {
7103 		if (ds->nfsdev_nmp != NULL && ds->nfsdev_nospc) {
7104 			if (++i > nfsrv_devidcnt)
7105 				break;
7106 			*tdvpp++ = ds->nfsdev_dvp;
7107 			NFSBCOPY(ds->nfsdev_deviceid, tdevid, NFSX_V4DEVICEID);
7108 			tdevid += NFSX_V4DEVICEID;
7109 		}
7110 	}
7111 	NFSDDSUNLOCK();
7112 
7113 	/* Do a VFS_STATFS() for each of the DSs and clear no space. */
7114 	cnt = i;
7115 	tdvpp = dvpp;
7116 	tdevid = devid;
7117 	for (i = 0; i < cnt && error == 0; i++) {
7118 		dvp = *tdvpp++;
7119 		error = VFS_STATFS(dvp->v_mount, tsf);
7120 		if (error == 0 && tsf->f_bavail > 0) {
7121 			NFSD_DEBUG(1, "nfsrv_checknospc: reset nospc\n");
7122 			nfsrv_marknospc(tdevid, false);
7123 		}
7124 		tdevid += NFSX_V4DEVICEID;
7125 	}
7126 	free(tsf, M_TEMP);
7127 	free(dvpp, M_TEMP);
7128 	free(devid, M_TEMP);
7129 }
7130 
7131 /*
7132  * Initialize everything that needs to be initialized for a vnet.
7133  */
7134 static void
7135 nfsrv_vnetinit(const void *unused __unused)
7136 {
7137 
7138 	nfsd_mntinit();
7139 }
7140 VNET_SYSINIT(nfsrv_vnetinit, SI_SUB_VNET_DONE, SI_ORDER_ANY,
7141     nfsrv_vnetinit, NULL);
7142 
7143 /*
7144  * Clean up everything that is in a vnet and needs to be
7145  * done when the jail is destroyed or the module unloaded.
7146  */
7147 static void
7148 nfsrv_cleanup(const void *unused __unused)
7149 {
7150 	int i;
7151 
7152 	NFSD_LOCK();
7153 	if (!NFSD_VNET(nfsrv_mntinited)) {
7154 		NFSD_UNLOCK();
7155 		return;
7156 	}
7157 	NFSD_VNET(nfsrv_mntinited) = false;
7158 	NFSD_UNLOCK();
7159 
7160 	/* Clean out all NFSv4 state. */
7161 	nfsrv_throwawayallstate(curthread);
7162 
7163 	/* Clean the NFS server reply cache */
7164 	nfsrvd_cleancache();
7165 
7166 	/* Clean out v4root exports. */
7167 	if (NFSD_VNET(nfsv4root_mnt)->mnt_export != NULL) {
7168 		vfs_free_addrlist(NFSD_VNET(nfsv4root_mnt)->mnt_export);
7169 		free(NFSD_VNET(nfsv4root_mnt)->mnt_export, M_MOUNT);
7170 		NFSD_VNET(nfsv4root_mnt)->mnt_export = NULL;
7171 	}
7172 
7173 	/* Free up the krpc server pool. */
7174 	if (NFSD_VNET(nfsrvd_pool) != NULL)
7175 		svcpool_destroy(NFSD_VNET(nfsrvd_pool));
7176 
7177 	/* and get rid of the locks */
7178 	for (i = 0; i < NFSRVCACHE_HASHSIZE; i++) {
7179 		mtx_destroy(&NFSD_VNET(nfsrchash_table)[i].mtx);
7180 		mtx_destroy(&NFSD_VNET(nfsrcahash_table)[i].mtx);
7181 	}
7182 	mtx_destroy(&NFSD_VNET(nfsv4root_mnt)->mnt_mtx);
7183 	for (i = 0; i < nfsrv_sessionhashsize; i++)
7184 		mtx_destroy(&NFSD_VNET(nfssessionhash)[i].mtx);
7185 	lockdestroy(&NFSD_VNET(nfsv4root_mnt)->mnt_explock);
7186 	free(NFSD_VNET(nfsrvudphashtbl), M_NFSRVCACHE);
7187 	free(NFSD_VNET(nfsrchash_table), M_NFSRVCACHE);
7188 	free(NFSD_VNET(nfsrcahash_table), M_NFSRVCACHE);
7189 	free(NFSD_VNET(nfsclienthash), M_NFSDCLIENT);
7190 	free(NFSD_VNET(nfslockhash), M_NFSDLOCKFILE);
7191 	free(NFSD_VNET(nfssessionhash), M_NFSDSESSION);
7192 	free(NFSD_VNET(nfsv4root_mnt), M_TEMP);
7193 	NFSD_VNET(nfsv4root_mnt) = NULL;
7194 }
7195 VNET_SYSUNINIT(nfsrv_cleanup, SI_SUB_VNET_DONE, SI_ORDER_ANY,
7196     nfsrv_cleanup, NULL);
7197 
7198 extern int (*nfsd_call_nfsd)(struct thread *, struct nfssvc_args *);
7199 
7200 /*
7201  * Called once to initialize data structures...
7202  */
7203 static int
7204 nfsd_modevent(module_t mod, int type, void *data)
7205 {
7206 	int error = 0, i;
7207 	static int loaded = 0;
7208 
7209 	switch (type) {
7210 	case MOD_LOAD:
7211 		if (loaded)
7212 			goto out;
7213 		newnfs_portinit();
7214 		mtx_init(&nfsrc_udpmtx, "nfsuc", NULL, MTX_DEF);
7215 		mtx_init(&nfs_v4root_mutex, "nfs4rt", NULL, MTX_DEF);
7216 		mtx_init(&nfsrv_dontlistlock_mtx, "nfs4dnl", NULL, MTX_DEF);
7217 		mtx_init(&nfsrv_recalllock_mtx, "nfs4rec", NULL, MTX_DEF);
7218 #ifdef VV_DISABLEDELEG
7219 		vn_deleg_ops.vndeleg_recall = nfsd_recalldelegation;
7220 		vn_deleg_ops.vndeleg_disable = nfsd_disabledelegation;
7221 #endif
7222 		nfsd_call_nfsd = nfssvc_nfsd;
7223 		loaded = 1;
7224 		break;
7225 
7226 	case MOD_UNLOAD:
7227 		if (newnfs_numnfsd != 0) {
7228 			error = EBUSY;
7229 			break;
7230 		}
7231 
7232 #ifdef VV_DISABLEDELEG
7233 		vn_deleg_ops.vndeleg_recall = NULL;
7234 		vn_deleg_ops.vndeleg_disable = NULL;
7235 #endif
7236 		nfsd_call_nfsd = NULL;
7237 		mtx_destroy(&nfsrc_udpmtx);
7238 		mtx_destroy(&nfs_v4root_mutex);
7239 		mtx_destroy(&nfsrv_dontlistlock_mtx);
7240 		mtx_destroy(&nfsrv_recalllock_mtx);
7241 		if (nfslayouthash != NULL) {
7242 			for (i = 0; i < nfsrv_layouthashsize; i++)
7243 				mtx_destroy(&nfslayouthash[i].mtx);
7244 			free(nfslayouthash, M_NFSDSESSION);
7245 		}
7246 		loaded = 0;
7247 		break;
7248 	default:
7249 		error = EOPNOTSUPP;
7250 		break;
7251 	}
7252 
7253 out:
7254 	NFSEXITCODE(error);
7255 	return (error);
7256 }
7257 static moduledata_t nfsd_mod = {
7258 	"nfsd",
7259 	nfsd_modevent,
7260 	NULL,
7261 };
7262 DECLARE_MODULE(nfsd, nfsd_mod, SI_SUB_VFS, SI_ORDER_ANY);
7263 
7264 /* So that loader and kldload(2) can find us, wherever we are.. */
7265 MODULE_VERSION(nfsd, 1);
7266 MODULE_DEPEND(nfsd, nfscommon, 1, 1, 1);
7267 MODULE_DEPEND(nfsd, nfslockd, 1, 1, 1);
7268 MODULE_DEPEND(nfsd, krpc, 1, 1, 1);
7269 MODULE_DEPEND(nfsd, nfssvc, 1, 1, 1);
7270