xref: /freebsd/sys/fs/nfsserver/nfs_nfsdstate.c (revision 51c8a9c1be57b6750e7c64e1379e8c33bd0f02c1)
1 /*-
2  * SPDX-License-Identifier: BSD-2-Clause
3  *
4  * Copyright (c) 2009 Rick Macklem, University of Guelph
5  * All rights reserved.
6  *
7  * Redistribution and use in source and binary forms, with or without
8  * modification, are permitted provided that the following conditions
9  * are met:
10  * 1. Redistributions of source code must retain the above copyright
11  *    notice, this list of conditions and the following disclaimer.
12  * 2. Redistributions in binary form must reproduce the above copyright
13  *    notice, this list of conditions and the following disclaimer in the
14  *    documentation and/or other materials provided with the distribution.
15  *
16  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
17  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
18  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
19  * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
20  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
21  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
22  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
23  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
24  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
25  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
26  * SUCH DAMAGE.
27  *
28  */
29 
30 #include <sys/cdefs.h>
31 #include "opt_inet.h"
32 #include "opt_inet6.h"
33 #include <sys/extattr.h>
34 #include <fs/nfs/nfsport.h>
35 
36 int nfsrv_issuedelegs = 0;
37 int nfsrv_dolocallocks = 0;
38 struct nfsv4lock nfsv4rootfs_lock;
39 time_t nfsdev_time = 0;
40 int nfsrv_layouthashsize;
41 volatile int nfsrv_layoutcnt = 0;
42 
43 NFSD_VNET_DEFINE(struct nfsrv_stablefirst, nfsrv_stablefirst);
44 
45 NFSD_VNET_DECLARE(int, nfsrv_numnfsd);
46 NFSD_VNET_DECLARE(struct nfsstatsv1 *, nfsstatsv1_p);
47 
48 extern uint32_t nfs_srvmaxio;
49 extern int nfsrv_lease;
50 extern struct timeval nfsboottime;
51 extern u_int32_t newnfs_true, newnfs_false;
52 extern struct mtx nfsrv_dslock_mtx;
53 extern struct mtx nfsrv_recalllock_mtx;
54 extern struct mtx nfsrv_dontlistlock_mtx;
55 extern int nfsd_debuglevel;
56 extern u_int nfsrv_dsdirsize;
57 extern struct nfsdevicehead nfsrv_devidhead;
58 extern int nfsrv_doflexfile;
59 extern int nfsrv_maxpnfsmirror;
60 NFSV4ROOTLOCKMUTEX;
61 NFSSTATESPINLOCK;
62 extern struct nfsdontlisthead nfsrv_dontlisthead;
63 extern volatile int nfsrv_devidcnt;
64 extern struct nfslayouthead nfsrv_recalllisthead;
65 extern char *nfsrv_zeropnfsdat;
66 
67 SYSCTL_DECL(_vfs_nfsd);
68 int	nfsrv_statehashsize = NFSSTATEHASHSIZE;
69 SYSCTL_INT(_vfs_nfsd, OID_AUTO, statehashsize, CTLFLAG_RDTUN,
70     &nfsrv_statehashsize, 0,
71     "Size of state hash table set via loader.conf");
72 
73 int	nfsrv_clienthashsize = NFSCLIENTHASHSIZE;
74 SYSCTL_INT(_vfs_nfsd, OID_AUTO, clienthashsize, CTLFLAG_RDTUN,
75     &nfsrv_clienthashsize, 0,
76     "Size of client hash table set via loader.conf");
77 
78 int	nfsrv_lockhashsize = NFSLOCKHASHSIZE;
79 SYSCTL_INT(_vfs_nfsd, OID_AUTO, fhhashsize, CTLFLAG_RDTUN,
80     &nfsrv_lockhashsize, 0,
81     "Size of file handle hash table set via loader.conf");
82 
83 int	nfsrv_sessionhashsize = NFSSESSIONHASHSIZE;
84 SYSCTL_INT(_vfs_nfsd, OID_AUTO, sessionhashsize, CTLFLAG_RDTUN,
85     &nfsrv_sessionhashsize, 0,
86     "Size of session hash table set via loader.conf");
87 
88 int	nfsrv_layouthighwater = NFSLAYOUTHIGHWATER;
89 SYSCTL_INT(_vfs_nfsd, OID_AUTO, layouthighwater, CTLFLAG_RDTUN,
90     &nfsrv_layouthighwater, 0,
91     "High water mark for number of layouts set via loader.conf");
92 
93 static int	nfsrv_v4statelimit = NFSRV_V4STATELIMIT;
94 SYSCTL_INT(_vfs_nfsd, OID_AUTO, v4statelimit, CTLFLAG_RWTUN,
95     &nfsrv_v4statelimit, 0,
96     "High water limit for NFSv4 opens+locks+delegations");
97 
98 static int	nfsrv_writedelegifpos = 0;
99 SYSCTL_INT(_vfs_nfsd, OID_AUTO, writedelegifpos, CTLFLAG_RW,
100     &nfsrv_writedelegifpos, 0,
101     "Issue a write delegation for read opens if possible");
102 
103 static int	nfsrv_allowreadforwriteopen = 1;
104 SYSCTL_INT(_vfs_nfsd, OID_AUTO, allowreadforwriteopen, CTLFLAG_RW,
105     &nfsrv_allowreadforwriteopen, 0,
106     "Allow Reads to be done with Write Access StateIDs");
107 
108 int	nfsrv_pnfsatime = 0;
109 SYSCTL_INT(_vfs_nfsd, OID_AUTO, pnfsstrictatime, CTLFLAG_RW,
110     &nfsrv_pnfsatime, 0,
111     "For pNFS service, do Getattr ops to keep atime up-to-date");
112 
113 int	nfsrv_flexlinuxhack = 0;
114 SYSCTL_INT(_vfs_nfsd, OID_AUTO, flexlinuxhack, CTLFLAG_RW,
115     &nfsrv_flexlinuxhack, 0,
116     "For Linux clients, hack around Flex File Layout bug");
117 
118 /*
119  * Hash lists for nfs V4.
120  */
121 NFSD_VNET_DEFINE(struct nfsclienthashhead *, nfsclienthash);
122 NFSD_VNET_DEFINE(struct nfslockhashhead *, nfslockhash);
123 NFSD_VNET_DEFINE(struct nfssessionhash *, nfssessionhash);
124 
125 struct nfslayouthash		*nfslayouthash;
126 volatile int nfsrv_dontlistlen = 0;
127 
128 static u_int32_t nfsrv_openpluslock = 0, nfsrv_delegatecnt = 0;
129 static int nfsrv_returnoldstateid = 0, nfsrv_clients = 0;
130 static int nfsrv_clienthighwater = NFSRV_CLIENTHIGHWATER;
131 static int nfsrv_nogsscallback = 0;
132 static volatile int nfsrv_writedelegcnt = 0;
133 static int nfsrv_faildscnt;
134 
135 NFSD_VNET_DEFINE_STATIC(time_t, nfsrvboottime);
136 
137 /* local functions */
138 static void nfsrv_dumpaclient(struct nfsclient *clp,
139     struct nfsd_dumpclients *dumpp);
140 static void nfsrv_freeopenowner(struct nfsstate *stp, int cansleep,
141     NFSPROC_T *p);
142 static int nfsrv_freeopen(struct nfsstate *stp, vnode_t vp, int cansleep,
143     NFSPROC_T *p);
144 static void nfsrv_freelockowner(struct nfsstate *stp, vnode_t vp, int cansleep,
145     NFSPROC_T *p);
146 static void nfsrv_freeallnfslocks(struct nfsstate *stp, vnode_t vp,
147     int cansleep, NFSPROC_T *p);
148 static void nfsrv_freenfslock(struct nfslock *lop);
149 static void nfsrv_freenfslockfile(struct nfslockfile *lfp);
150 static void nfsrv_freedeleg(struct nfsstate *);
151 static int nfsrv_getstate(struct nfsclient *clp, nfsv4stateid_t *stateidp,
152     u_int32_t flags, struct nfsstate **stpp);
153 static void nfsrv_getowner(struct nfsstatehead *hp, struct nfsstate *new_stp,
154     struct nfsstate **stpp);
155 static int nfsrv_getlockfh(vnode_t vp, u_short flags,
156     struct nfslockfile *new_lfp, fhandle_t *nfhp, NFSPROC_T *p);
157 static int nfsrv_getlockfile(u_short flags, struct nfslockfile **new_lfpp,
158     struct nfslockfile **lfpp, fhandle_t *nfhp, int lockit);
159 static void nfsrv_insertlock(struct nfslock *new_lop,
160     struct nfslock *insert_lop, struct nfsstate *stp, struct nfslockfile *lfp);
161 static void nfsrv_updatelock(struct nfsstate *stp, struct nfslock **new_lopp,
162     struct nfslock **other_lopp, struct nfslockfile *lfp);
163 static int nfsrv_getipnumber(u_char *cp);
164 static int nfsrv_checkrestart(nfsquad_t clientid, u_int32_t flags,
165     nfsv4stateid_t *stateidp, int specialid);
166 static int nfsrv_checkgrace(struct nfsrv_descript *nd, struct nfsclient *clp,
167     u_int32_t flags);
168 static int nfsrv_docallback(struct nfsclient *clp, int procnum,
169     nfsv4stateid_t *stateidp, int trunc, fhandle_t *fhp,
170     struct nfsvattr *nap, nfsattrbit_t *attrbitp, int laytype, NFSPROC_T *p);
171 static int nfsrv_cbcallargs(struct nfsrv_descript *nd, struct nfsclient *clp,
172     uint32_t callback, int op, const char *optag, struct nfsdsession **sepp,
173     int *slotposp);
174 static u_int32_t nfsrv_nextclientindex(void);
175 static u_int32_t nfsrv_nextstateindex(struct nfsclient *clp);
176 static void nfsrv_markstable(struct nfsclient *clp);
177 static void nfsrv_markreclaim(struct nfsclient *clp);
178 static int nfsrv_checkstable(struct nfsclient *clp);
179 static int nfsrv_clientconflict(struct nfsclient *clp, int *haslockp, struct
180     vnode *vp, NFSPROC_T *p);
181 static int nfsrv_delegconflict(struct nfsstate *stp, int *haslockp,
182     NFSPROC_T *p, vnode_t vp);
183 static int nfsrv_cleandeleg(vnode_t vp, struct nfslockfile *lfp,
184     struct nfsclient *clp, int *haslockp, NFSPROC_T *p);
185 static int nfsrv_notsamecredname(int op, struct nfsrv_descript *nd,
186     struct nfsclient *clp);
187 static time_t nfsrv_leaseexpiry(void);
188 static void nfsrv_delaydelegtimeout(struct nfsstate *stp);
189 static int nfsrv_checkseqid(struct nfsrv_descript *nd, u_int32_t seqid,
190     struct nfsstate *stp, struct nfsrvcache *op);
191 static int nfsrv_nootherstate(struct nfsstate *stp);
192 static int nfsrv_locallock(vnode_t vp, struct nfslockfile *lfp, int flags,
193     uint64_t first, uint64_t end, struct nfslockconflict *cfp, NFSPROC_T *p);
194 static void nfsrv_localunlock(vnode_t vp, struct nfslockfile *lfp,
195     uint64_t init_first, uint64_t init_end, NFSPROC_T *p);
196 static int nfsrv_dolocal(vnode_t vp, struct nfslockfile *lfp, int flags,
197     int oldflags, uint64_t first, uint64_t end, struct nfslockconflict *cfp,
198     NFSPROC_T *p);
199 static void nfsrv_locallock_rollback(vnode_t vp, struct nfslockfile *lfp,
200     NFSPROC_T *p);
201 static void nfsrv_locallock_commit(struct nfslockfile *lfp, int flags,
202     uint64_t first, uint64_t end);
203 static void nfsrv_locklf(struct nfslockfile *lfp);
204 static void nfsrv_unlocklf(struct nfslockfile *lfp);
205 static struct nfsdsession *nfsrv_findsession(uint8_t *sessionid);
206 static int nfsrv_freesession(struct nfsrv_descript *nd, struct nfsdsession *sep,
207     uint8_t *sessionid, bool locked, SVCXPRT **old_xprtp);
208 static int nfsv4_setcbsequence(struct nfsrv_descript *nd, struct nfsclient *clp,
209     int dont_replycache, struct nfsdsession **sepp, int *slotposp);
210 static int nfsv4_getcbsession(struct nfsclient *clp, struct nfsdsession **sepp);
211 static int nfsrv_addlayout(struct nfsrv_descript *nd, struct nfslayout **lypp,
212     nfsv4stateid_t *stateidp, char *layp, int *layoutlenp, NFSPROC_T *p);
213 static void nfsrv_freelayout(struct nfslayouthead *lhp, struct nfslayout *lyp);
214 static void nfsrv_freelayoutlist(nfsquad_t clientid);
215 static void nfsrv_freelayouts(nfsquad_t *clid, fsid_t *fs, int laytype,
216     int iomode);
217 static void nfsrv_freealllayouts(void);
218 static void nfsrv_freedevid(struct nfsdevice *ds);
219 static int nfsrv_setdsserver(char *dspathp, char *mdspathp, NFSPROC_T *p,
220     struct nfsdevice **dsp);
221 static void nfsrv_deleteds(struct nfsdevice *fndds);
222 static void nfsrv_allocdevid(struct nfsdevice *ds, char *addr, char *dnshost);
223 static void nfsrv_freealldevids(void);
224 static void nfsrv_flexlayouterr(struct nfsrv_descript *nd, uint32_t *layp,
225     int maxcnt, NFSPROC_T *p);
226 static int nfsrv_recalllayout(nfsquad_t clid, nfsv4stateid_t *stateidp,
227     fhandle_t *fhp, struct nfslayout *lyp, int changed, int laytype,
228     NFSPROC_T *p);
229 static int nfsrv_findlayout(nfsquad_t *clientidp, fhandle_t *fhp, int laytype,
230     NFSPROC_T *, struct nfslayout **lypp);
231 static int nfsrv_fndclid(nfsquad_t *clidvec, nfsquad_t clid, int clidcnt);
232 static struct nfslayout *nfsrv_filelayout(struct nfsrv_descript *nd, int iomode,
233     fhandle_t *fhp, fhandle_t *dsfhp, char *devid, fsid_t fs);
234 static struct nfslayout *nfsrv_flexlayout(struct nfsrv_descript *nd, int iomode,
235     int mirrorcnt, fhandle_t *fhp, fhandle_t *dsfhp, char *devid, fsid_t fs);
236 static int nfsrv_dontlayout(fhandle_t *fhp);
237 static int nfsrv_createdsfile(vnode_t vp, fhandle_t *fhp, struct pnfsdsfile *pf,
238     vnode_t dvp, struct nfsdevice *ds, struct ucred *cred, NFSPROC_T *p,
239     vnode_t *tvpp);
240 static struct nfsdevice *nfsrv_findmirroredds(struct nfsmount *nmp);
241 static int nfsrv_checkmachcred(int op, struct nfsrv_descript *nd,
242     struct nfsclient *clp);
243 static void nfsrv_issuedelegation(struct vnode *vp, struct nfsclient *clp,
244     struct nfsrv_descript *nd, int delegate, int writedeleg, int readonly,
245     u_quad_t filerev, uint64_t rdonly, struct nfsstate **new_delegp,
246     struct nfsstate *new_stp, struct nfslockfile *lfp, uint32_t *rflagsp,
247     nfsv4stateid_t *delegstateidp);
248 static void nfsrv_clientlock(bool mlocked);
249 static void nfsrv_clientunlock(bool mlocked);
250 
251 /*
252  * Lock the client structure, either with the mutex or the exclusive nfsd lock.
253  */
254 static void
255 nfsrv_clientlock(bool mlocked)
256 {
257 	int igotlock;
258 
259 	if (mlocked) {
260 		NFSLOCKSTATE();
261 	} else {
262 		NFSLOCKV4ROOTMUTEX();
263 		nfsv4_relref(&nfsv4rootfs_lock);
264 		do {
265 			igotlock = nfsv4_lock(&nfsv4rootfs_lock, 1, NULL,
266 			    NFSV4ROOTLOCKMUTEXPTR, NULL);
267 		} while (!igotlock);
268 		NFSUNLOCKV4ROOTMUTEX();
269 	}
270 }
271 
272 /*
273  * Unlock the client structure.
274  */
275 static void
276 nfsrv_clientunlock(bool mlocked)
277 {
278 
279 	if (mlocked) {
280 		NFSUNLOCKSTATE();
281 	} else {
282 		NFSLOCKV4ROOTMUTEX();
283 		nfsv4_unlock(&nfsv4rootfs_lock, 1);
284 		NFSUNLOCKV4ROOTMUTEX();
285 	}
286 }
287 
288 /*
289  * Scan the client list for a match and either return the current one,
290  * create a new entry or return an error.
291  * If returning a non-error, the clp structure must either be linked into
292  * the client list or free'd.
293  */
294 int
295 nfsrv_setclient(struct nfsrv_descript *nd, struct nfsclient **new_clpp,
296     nfsquad_t *clientidp, nfsquad_t *confirmp, NFSPROC_T *p)
297 {
298 	struct nfsclient *clp = NULL, *new_clp = *new_clpp;
299 	int i, error = 0, ret;
300 	struct nfsstate *stp, *tstp;
301 #ifdef INET
302 	struct sockaddr_in *sin, *rin;
303 #endif
304 #ifdef INET6
305 	struct sockaddr_in6 *sin6, *rin6;
306 #endif
307 	struct nfsdsession *sep, *nsep;
308 	SVCXPRT *old_xprt;
309 	struct nfssessionhead old_sess;
310 	int zapit = 0, gotit, hasstate = 0;
311 	bool mlocked;
312 	static u_int64_t confirm_index = 0;
313 
314 	/*
315 	 * Check for state resource limit exceeded.
316 	 */
317 	if (nfsrv_openpluslock > nfsrv_v4statelimit) {
318 		error = NFSERR_RESOURCE;
319 		goto out;
320 	}
321 
322 	if (nfsrv_issuedelegs == 0 ||
323 	    ((nd->nd_flag & ND_GSS) != 0 && nfsrv_nogsscallback != 0))
324 		/*
325 		 * Don't do callbacks when delegations are disabled or
326 		 * for AUTH_GSS unless enabled via nfsrv_nogsscallback.
327 		 * If establishing a callback connection is attempted
328 		 * when a firewall is blocking the callback path, the
329 		 * server may wait too long for the connect attempt to
330 		 * succeed during the Open. Some clients, such as Linux,
331 		 * may timeout and give up on the Open before the server
332 		 * replies. Also, since AUTH_GSS callbacks are not
333 		 * yet interoperability tested, they might cause the
334 		 * server to crap out, if they get past the Init call to
335 		 * the client.
336 		 */
337 		new_clp->lc_program = 0;
338 
339 	mlocked = true;
340 	if (nfsrv_dolocallocks != 0)
341 		mlocked = false;
342 	/* Lock out other nfsd threads */
343 	nfsrv_clientlock(mlocked);
344 
345 	/*
346 	 * Search for a match in the client list.
347 	 */
348 	gotit = i = 0;
349 	while (i < nfsrv_clienthashsize && !gotit) {
350 	    LIST_FOREACH(clp, &NFSD_VNET(nfsclienthash)[i], lc_hash) {
351 		if (new_clp->lc_idlen == clp->lc_idlen &&
352 		    !NFSBCMP(new_clp->lc_id, clp->lc_id, clp->lc_idlen)) {
353 			gotit = 1;
354 			break;
355 		}
356 	    }
357 	    if (gotit == 0)
358 		i++;
359 	}
360 	old_xprt = NULL;
361 	if (!gotit ||
362 	    (clp->lc_flags & (LCL_NEEDSCONFIRM | LCL_ADMINREVOKED))) {
363 		if ((nd->nd_flag & ND_NFSV41) != 0 && confirmp->lval[1] != 0) {
364 			/*
365 			 * For NFSv4.1, if confirmp->lval[1] is non-zero, the
366 			 * client is trying to update a confirmed clientid.
367 			 */
368 			nfsrv_clientunlock(mlocked);
369 			confirmp->lval[1] = 0;
370 			error = NFSERR_NOENT;
371 			goto out;
372 		}
373 		/*
374 		 * Get rid of the old one.
375 		 */
376 		if (i != nfsrv_clienthashsize) {
377 			LIST_REMOVE(clp, lc_hash);
378 			if (mlocked)
379 				nfsrv_cleanclient(clp, p, true, &old_xprt);
380 			else
381 				nfsrv_cleanclient(clp, p, false, NULL);
382 			nfsrv_freedeleglist(&clp->lc_deleg);
383 			nfsrv_freedeleglist(&clp->lc_olddeleg);
384 			zapit = 1;
385 		}
386 		/*
387 		 * Add it after assigning a client id to it.
388 		 */
389 		new_clp->lc_flags |= LCL_NEEDSCONFIRM;
390 		if ((nd->nd_flag & ND_NFSV41) != 0) {
391 			confirmp->lval[0] = ++confirm_index;
392 			new_clp->lc_confirm.lval[0] = confirmp->lval[0] - 1;
393 		} else
394 			confirmp->qval = new_clp->lc_confirm.qval =
395 			    ++confirm_index;
396 		clientidp->lval[0] = new_clp->lc_clientid.lval[0] =
397 		    NFSD_VNET(nfsrvboottime);
398 		clientidp->lval[1] = new_clp->lc_clientid.lval[1] =
399 		    nfsrv_nextclientindex();
400 		new_clp->lc_stateindex = 0;
401 		new_clp->lc_statemaxindex = 0;
402 		new_clp->lc_prevsess = 0;
403 		new_clp->lc_cbref = 0;
404 		new_clp->lc_expiry = nfsrv_leaseexpiry();
405 		LIST_INIT(&new_clp->lc_open);
406 		LIST_INIT(&new_clp->lc_deleg);
407 		LIST_INIT(&new_clp->lc_olddeleg);
408 		LIST_INIT(&new_clp->lc_session);
409 		for (i = 0; i < nfsrv_statehashsize; i++)
410 			LIST_INIT(&new_clp->lc_stateid[i]);
411 		LIST_INSERT_HEAD(NFSCLIENTHASH(new_clp->lc_clientid), new_clp,
412 		    lc_hash);
413 		NFSD_VNET(nfsstatsv1_p)->srvclients++;
414 		nfsrv_openpluslock++;
415 		nfsrv_clients++;
416 		nfsrv_clientunlock(mlocked);
417 		if (zapit != 0) {
418 			if (old_xprt != NULL)
419 				SVC_RELEASE(old_xprt);
420 			nfsrv_zapclient(clp, p);
421 		}
422 		*new_clpp = NULL;
423 		goto out;
424 	}
425 
426 	/*
427 	 * Now, handle the cases where the id is already issued.
428 	 */
429 	if (nfsrv_notsamecredname(NFSV4OP_EXCHANGEID, nd, clp)) {
430 	    /*
431 	     * Check to see if there is expired state that should go away.
432 	     */
433 	    if (clp->lc_expiry < NFSD_MONOSEC &&
434 	        (!LIST_EMPTY(&clp->lc_open) || !LIST_EMPTY(&clp->lc_deleg))) {
435 		if (mlocked)
436 		    nfsrv_cleanclient(clp, p, true, &old_xprt);
437 		else
438 		    nfsrv_cleanclient(clp, p, false, NULL);
439 		nfsrv_freedeleglist(&clp->lc_deleg);
440 	    }
441 
442 	    /*
443 	     * If there is outstanding state, then reply NFSERR_CLIDINUSE per
444 	     * RFC3530 Sec. 8.1.2 last para.
445 	     */
446 	    if (!LIST_EMPTY(&clp->lc_deleg)) {
447 		hasstate = 1;
448 	    } else if (LIST_EMPTY(&clp->lc_open)) {
449 		hasstate = 0;
450 	    } else {
451 		hasstate = 0;
452 		/* Look for an Open on the OpenOwner */
453 		LIST_FOREACH(stp, &clp->lc_open, ls_list) {
454 		    if (!LIST_EMPTY(&stp->ls_open)) {
455 			hasstate = 1;
456 			break;
457 		    }
458 		}
459 	    }
460 	    if (hasstate) {
461 		/*
462 		 * If the uid doesn't match, return NFSERR_CLIDINUSE after
463 		 * filling out the correct ipaddr and portnum.
464 		 */
465 		switch (clp->lc_req.nr_nam->sa_family) {
466 #ifdef INET
467 		case AF_INET:
468 			sin = (struct sockaddr_in *)new_clp->lc_req.nr_nam;
469 			rin = (struct sockaddr_in *)clp->lc_req.nr_nam;
470 			sin->sin_addr.s_addr = rin->sin_addr.s_addr;
471 			sin->sin_port = rin->sin_port;
472 			break;
473 #endif
474 #ifdef INET6
475 		case AF_INET6:
476 			sin6 = (struct sockaddr_in6 *)new_clp->lc_req.nr_nam;
477 			rin6 = (struct sockaddr_in6 *)clp->lc_req.nr_nam;
478 			sin6->sin6_addr = rin6->sin6_addr;
479 			sin6->sin6_port = rin6->sin6_port;
480 			break;
481 #endif
482 		}
483 		nfsrv_clientunlock(mlocked);
484 		if (old_xprt != NULL)
485 			SVC_RELEASE(old_xprt);
486 		error = NFSERR_CLIDINUSE;
487 		goto out;
488 	    }
489 	}
490 
491 	if (NFSBCMP(new_clp->lc_verf, clp->lc_verf, NFSX_VERF)) {
492 		/*
493 		 * If the verifier has changed, the client has rebooted
494 		 * and a new client id is issued. The old state info
495 		 * can be thrown away once the SetClientID_Confirm or
496 		 * Create_Session that confirms the clientid occurs.
497 		 */
498 		LIST_REMOVE(clp, lc_hash);
499 
500 		LIST_NEWHEAD(&old_sess, &clp->lc_session, sess_list);
501 
502 		new_clp->lc_flags |= LCL_NEEDSCONFIRM;
503 		if ((nd->nd_flag & ND_NFSV41) != 0) {
504 			confirmp->lval[0] = ++confirm_index;
505 			new_clp->lc_confirm.lval[0] = confirmp->lval[0] - 1;
506 		} else
507 			confirmp->qval = new_clp->lc_confirm.qval =
508 			    ++confirm_index;
509 		clientidp->lval[0] = new_clp->lc_clientid.lval[0] =
510 		    NFSD_VNET(nfsrvboottime);
511 		clientidp->lval[1] = new_clp->lc_clientid.lval[1] =
512 		    nfsrv_nextclientindex();
513 		new_clp->lc_stateindex = 0;
514 		new_clp->lc_statemaxindex = 0;
515 		new_clp->lc_prevsess = 0;
516 		new_clp->lc_cbref = 0;
517 		new_clp->lc_expiry = nfsrv_leaseexpiry();
518 
519 		/*
520 		 * Save the state until confirmed.
521 		 */
522 		LIST_NEWHEAD(&new_clp->lc_open, &clp->lc_open, ls_list);
523 		LIST_FOREACH(tstp, &new_clp->lc_open, ls_list)
524 			tstp->ls_clp = new_clp;
525 		LIST_NEWHEAD(&new_clp->lc_deleg, &clp->lc_deleg, ls_list);
526 		LIST_FOREACH(tstp, &new_clp->lc_deleg, ls_list)
527 			tstp->ls_clp = new_clp;
528 		LIST_NEWHEAD(&new_clp->lc_olddeleg, &clp->lc_olddeleg,
529 		    ls_list);
530 		LIST_FOREACH(tstp, &new_clp->lc_olddeleg, ls_list)
531 			tstp->ls_clp = new_clp;
532 		for (i = 0; i < nfsrv_statehashsize; i++) {
533 			LIST_NEWHEAD(&new_clp->lc_stateid[i],
534 			    &clp->lc_stateid[i], ls_hash);
535 			LIST_FOREACH(tstp, &new_clp->lc_stateid[i], ls_hash)
536 				tstp->ls_clp = new_clp;
537 		}
538 		LIST_INIT(&new_clp->lc_session);
539 		LIST_INSERT_HEAD(NFSCLIENTHASH(new_clp->lc_clientid), new_clp,
540 		    lc_hash);
541 		NFSD_VNET(nfsstatsv1_p)->srvclients++;
542 		nfsrv_openpluslock++;
543 		nfsrv_clients++;
544 		if (!mlocked) {
545 			nfsrv_clientunlock(mlocked);
546 			NFSLOCKSTATE();
547 		}
548 
549 		/*
550 		 * Must wait until any outstanding callback on the old clp
551 		 * completes.
552 		 */
553 		while (clp->lc_cbref) {
554 			clp->lc_flags |= LCL_WAKEUPWANTED;
555 			(void)mtx_sleep(clp, NFSSTATEMUTEXPTR, PZERO - 1,
556 			    "nfsd clp", 10 * hz);
557 		}
558 		NFSUNLOCKSTATE();
559 		if (old_xprt != NULL)
560 			SVC_RELEASE(old_xprt);
561 		/* Get rid of all sessions on this clientid. */
562 		LIST_FOREACH_SAFE(sep, &old_sess, sess_list, nsep) {
563 			ret = nfsrv_freesession(NULL, sep, NULL, false, NULL);
564 			if (ret != 0)
565 				printf("nfsrv_setclient: verifier changed free"
566 				    " session failed=%d\n", ret);
567 		}
568 
569 		nfsrv_zapclient(clp, p);
570 		*new_clpp = NULL;
571 		goto out;
572 	}
573 
574 	/* For NFSv4.1, mark that we found a confirmed clientid. */
575 	if ((nd->nd_flag & ND_NFSV41) != 0) {
576 		clientidp->lval[0] = clp->lc_clientid.lval[0];
577 		clientidp->lval[1] = clp->lc_clientid.lval[1];
578 		confirmp->lval[0] = 0;	/* Ignored by client */
579 		confirmp->lval[1] = 1;
580 	} else {
581 		/*
582 		 * id and verifier match, so update the net address info
583 		 * and get rid of any existing callback authentication
584 		 * handle, so a new one will be acquired.
585 		 */
586 		LIST_REMOVE(clp, lc_hash);
587 		new_clp->lc_flags |= (LCL_NEEDSCONFIRM | LCL_DONTCLEAN);
588 		new_clp->lc_expiry = nfsrv_leaseexpiry();
589 		confirmp->qval = new_clp->lc_confirm.qval = ++confirm_index;
590 		clientidp->lval[0] = new_clp->lc_clientid.lval[0] =
591 		    clp->lc_clientid.lval[0];
592 		clientidp->lval[1] = new_clp->lc_clientid.lval[1] =
593 		    clp->lc_clientid.lval[1];
594 		new_clp->lc_delegtime = clp->lc_delegtime;
595 		new_clp->lc_stateindex = clp->lc_stateindex;
596 		new_clp->lc_statemaxindex = clp->lc_statemaxindex;
597 		new_clp->lc_cbref = 0;
598 		LIST_NEWHEAD(&new_clp->lc_open, &clp->lc_open, ls_list);
599 		LIST_FOREACH(tstp, &new_clp->lc_open, ls_list)
600 			tstp->ls_clp = new_clp;
601 		LIST_NEWHEAD(&new_clp->lc_deleg, &clp->lc_deleg, ls_list);
602 		LIST_FOREACH(tstp, &new_clp->lc_deleg, ls_list)
603 			tstp->ls_clp = new_clp;
604 		LIST_NEWHEAD(&new_clp->lc_olddeleg, &clp->lc_olddeleg, ls_list);
605 		LIST_FOREACH(tstp, &new_clp->lc_olddeleg, ls_list)
606 			tstp->ls_clp = new_clp;
607 		for (i = 0; i < nfsrv_statehashsize; i++) {
608 			LIST_NEWHEAD(&new_clp->lc_stateid[i],
609 			    &clp->lc_stateid[i], ls_hash);
610 			LIST_FOREACH(tstp, &new_clp->lc_stateid[i], ls_hash)
611 				tstp->ls_clp = new_clp;
612 		}
613 		LIST_INIT(&new_clp->lc_session);
614 		LIST_INSERT_HEAD(NFSCLIENTHASH(new_clp->lc_clientid), new_clp,
615 		    lc_hash);
616 		NFSD_VNET(nfsstatsv1_p)->srvclients++;
617 		nfsrv_openpluslock++;
618 		nfsrv_clients++;
619 	}
620 	if (!mlocked)
621 		nfsrv_clientunlock(mlocked);
622 
623 	if ((nd->nd_flag & ND_NFSV41) == 0) {
624 		/*
625 		 * Must wait until any outstanding callback on the old clp
626 		 * completes.
627 		 */
628 		if (!mlocked)
629 			NFSLOCKSTATE();
630 		while (clp->lc_cbref) {
631 			clp->lc_flags |= LCL_WAKEUPWANTED;
632 			(void)mtx_sleep(clp, NFSSTATEMUTEXPTR, PZERO - 1,
633 			    "nfsdclp", 10 * hz);
634 		}
635 		NFSUNLOCKSTATE();
636 		if (old_xprt != NULL)
637 			SVC_RELEASE(old_xprt);
638 		nfsrv_zapclient(clp, p);
639 		*new_clpp = NULL;
640 	} else {
641 		if (mlocked)
642 			NFSUNLOCKSTATE();
643 		if (old_xprt != NULL)
644 			SVC_RELEASE(old_xprt);
645 	}
646 
647 out:
648 	NFSEXITCODE2(error, nd);
649 	return (error);
650 }
651 
652 /*
653  * Check to see if the client id exists and optionally confirm it.
654  */
655 int
656 nfsrv_getclient(nfsquad_t clientid, int opflags, struct nfsclient **clpp,
657     struct nfsdsession *nsep, nfsquad_t confirm, uint32_t cbprogram,
658     struct nfsrv_descript *nd, NFSPROC_T *p)
659 {
660 	struct nfsclient *clp;
661 	struct nfsstate *stp;
662 	int i;
663 	struct nfsclienthashhead *hp;
664 	int error = 0, doneok, igotlock;
665 	struct nfssessionhash *shp;
666 	struct nfsdsession *sep;
667 	uint64_t sessid[2];
668 	CLIENT *client;
669 	SVCXPRT *old_xprt;
670 	bool mlocked, sess_replay;
671 	static uint64_t next_sess = 0;
672 
673 	if (clpp)
674 		*clpp = NULL;
675 	if ((nd == NULL || (nd->nd_flag & ND_NFSV41) == 0 ||
676 	    opflags != CLOPS_RENEW) && NFSD_VNET(nfsrvboottime) !=
677 	    clientid.lval[0]) {
678 		error = NFSERR_STALECLIENTID;
679 		goto out;
680 	}
681 
682 	/*
683 	 * If called with opflags == CLOPS_RENEW, the State Lock is
684 	 * already held. Otherwise, we need to get either that or,
685 	 * for the case of Confirm, lock out the nfsd threads.
686 	 */
687 	client = NULL;
688 	old_xprt = NULL;
689 	mlocked = true;
690 	if (nfsrv_dolocallocks != 0)
691 		mlocked = false;
692 	if (opflags & CLOPS_CONFIRM) {
693 		if (nsep != NULL &&
694 		    (nsep->sess_crflags & NFSV4CRSESS_CONNBACKCHAN) != 0)
695 			client = (struct __rpc_client *)
696 			    clnt_bck_create(nd->nd_xprt->xp_socket,
697 			    cbprogram, NFSV4_CBVERS);
698 		if (mlocked) {
699 			nfsrv_clientlock(mlocked);
700 		} else {
701 			NFSLOCKV4ROOTMUTEX();
702 			nfsv4_relref(&nfsv4rootfs_lock);
703 			do {
704 				igotlock = nfsv4_lock(&nfsv4rootfs_lock, 1,
705 				    NULL, NFSV4ROOTLOCKMUTEXPTR, NULL);
706 			} while (!igotlock);
707 		}
708 		/*
709 		 * Create a new sessionid here, since we need to do it where
710 		 * there is a mutex held to serialize update of next_sess.
711 		 */
712 		if ((nd->nd_flag & ND_NFSV41) != 0) {
713 			sessid[0] = ++next_sess;
714 			sessid[1] = clientid.qval;
715 		}
716 		if (!mlocked)
717 			NFSUNLOCKV4ROOTMUTEX();
718 	} else if (opflags != CLOPS_RENEW) {
719 		NFSLOCKSTATE();
720 	}
721 
722 	/* For NFSv4.1, the clp is acquired from the associated session. */
723 	if (nd != NULL && (nd->nd_flag & ND_NFSV41) != 0 &&
724 	    opflags == CLOPS_RENEW) {
725 		clp = NULL;
726 		if ((nd->nd_flag & ND_HASSEQUENCE) != 0) {
727 			shp = NFSSESSIONHASH(nd->nd_sessionid);
728 			NFSLOCKSESSION(shp);
729 			sep = nfsrv_findsession(nd->nd_sessionid);
730 			if (sep != NULL)
731 				clp = sep->sess_clp;
732 			NFSUNLOCKSESSION(shp);
733 		}
734 	} else {
735 		hp = NFSCLIENTHASH(clientid);
736 		LIST_FOREACH(clp, hp, lc_hash) {
737 			if (clp->lc_clientid.lval[1] == clientid.lval[1])
738 				break;
739 		}
740 	}
741 	if (clp == NULL) {
742 		if (opflags & CLOPS_CONFIRM)
743 			error = NFSERR_STALECLIENTID;
744 		else
745 			error = NFSERR_EXPIRED;
746 	} else if (clp->lc_flags & LCL_ADMINREVOKED) {
747 		/*
748 		 * If marked admin revoked, just return the error.
749 		 */
750 		error = NFSERR_ADMINREVOKED;
751 	}
752 	if (error) {
753 		if (opflags & CLOPS_CONFIRM) {
754 			nfsrv_clientunlock(mlocked);
755 			if (client != NULL)
756 				CLNT_RELEASE(client);
757 		} else if (opflags != CLOPS_RENEW) {
758 			NFSUNLOCKSTATE();
759 		}
760 		goto out;
761 	}
762 
763 	/*
764 	 * Perform any operations specified by the opflags.
765 	 */
766 	if (opflags & CLOPS_CONFIRM) {
767 		sess_replay = false;
768 		if ((nd->nd_flag & ND_NFSV41) != 0) {
769 		    /*
770 		     * For the case where lc_confirm.lval[0] == confirm.lval[0],
771 		     * use the new session, but with the previous sessionid.
772 		     * This is not exactly what the RFC describes, but should
773 		     * result in the same reply as the previous CreateSession.
774 		     */
775 		    if (clp->lc_confirm.lval[0] + 1 == confirm.lval[0]) {
776 			clp->lc_confirm.lval[0] = confirm.lval[0];
777 			clp->lc_prevsess = sessid[0];
778 		    } else if (clp->lc_confirm.lval[0] == confirm.lval[0]) {
779 			if (clp->lc_prevsess == 0)
780 			    error = NFSERR_SEQMISORDERED;
781 			else
782 			    sessid[0] = clp->lc_prevsess;
783 			sess_replay = true;
784 		    } else
785 			error = NFSERR_SEQMISORDERED;
786 		} else if ((nd->nd_flag & ND_NFSV41) == 0 &&
787 		     clp->lc_confirm.qval != confirm.qval)
788 			error = NFSERR_STALECLIENTID;
789 		if (error == 0 && nfsrv_notsamecredname(NFSV4OP_CREATESESSION,
790 		    nd, clp))
791 			error = NFSERR_CLIDINUSE;
792 
793 		if (!error) {
794 		    if ((clp->lc_flags & (LCL_NEEDSCONFIRM | LCL_DONTCLEAN)) ==
795 			LCL_NEEDSCONFIRM) {
796 			/*
797 			 * Hang onto the delegations (as old delegations)
798 			 * for an Open with CLAIM_DELEGATE_PREV unless in
799 			 * grace, but get rid of the rest of the state.
800 			 */
801 			if (mlocked)
802 				nfsrv_cleanclient(clp, p, true, &old_xprt);
803 			else
804 				nfsrv_cleanclient(clp, p, false, NULL);
805 			nfsrv_freedeleglist(&clp->lc_olddeleg);
806 			if (nfsrv_checkgrace(nd, clp, 0)) {
807 			    /* In grace, so just delete delegations */
808 			    nfsrv_freedeleglist(&clp->lc_deleg);
809 			} else {
810 			    LIST_FOREACH(stp, &clp->lc_deleg, ls_list)
811 				stp->ls_flags |= NFSLCK_OLDDELEG;
812 			    clp->lc_delegtime = NFSD_MONOSEC +
813 				nfsrv_lease + NFSRV_LEASEDELTA;
814 			    LIST_NEWHEAD(&clp->lc_olddeleg, &clp->lc_deleg,
815 				ls_list);
816 			}
817 			if ((nd->nd_flag & ND_NFSV41) != 0)
818 			    clp->lc_program = cbprogram;
819 		    }
820 		    clp->lc_flags &= ~(LCL_NEEDSCONFIRM | LCL_DONTCLEAN);
821 		    if (clp->lc_program)
822 			clp->lc_flags |= LCL_NEEDSCBNULL;
823 		    /* For NFSv4.1, link the session onto the client. */
824 		    if (nsep != NULL) {
825 			/* Hold a reference on the xprt for a backchannel. */
826 			if ((nsep->sess_crflags & NFSV4CRSESS_CONNBACKCHAN)
827 			    != 0 && !sess_replay) {
828 			    if (clp->lc_req.nr_client == NULL) {
829 				clp->lc_req.nr_client = client;
830 				client = NULL;
831 			    }
832 			    if (clp->lc_req.nr_client != NULL) {
833 				SVC_ACQUIRE(nd->nd_xprt);
834 				CLNT_ACQUIRE(clp->lc_req.nr_client);
835 				nd->nd_xprt->xp_p2 = clp->lc_req.nr_client;
836 				/* Disable idle timeout. */
837 				nd->nd_xprt->xp_idletimeout = 0;
838 				nsep->sess_cbsess.nfsess_xprt = nd->nd_xprt;
839 			    } else
840 				nsep->sess_crflags &= ~NFSV4CRSESS_CONNBACKCHAN;
841 			}
842 			NFSBCOPY(sessid, nsep->sess_sessionid,
843 			    NFSX_V4SESSIONID);
844 			NFSBCOPY(sessid, nsep->sess_cbsess.nfsess_sessionid,
845 			    NFSX_V4SESSIONID);
846 			if (!sess_replay) {
847 			    shp = NFSSESSIONHASH(nsep->sess_sessionid);
848 			    if (!mlocked)
849 				NFSLOCKSTATE();
850 			    NFSLOCKSESSION(shp);
851 			    LIST_INSERT_HEAD(&shp->list, nsep, sess_hash);
852 			    LIST_INSERT_HEAD(&clp->lc_session, nsep, sess_list);
853 			    nsep->sess_clp = clp;
854 			    NFSUNLOCKSESSION(shp);
855 			    if (!mlocked)
856 				NFSUNLOCKSTATE();
857 			}
858 		    }
859 		}
860 	} else if (clp->lc_flags & LCL_NEEDSCONFIRM) {
861 		error = NFSERR_EXPIRED;
862 	}
863 
864 	/*
865 	 * If called by the Renew Op, we must check the principal.
866 	 */
867 	if (!error && (opflags & CLOPS_RENEWOP)) {
868 	    if (nfsrv_notsamecredname(0, nd, clp)) {
869 		doneok = 0;
870 		for (i = 0; i < nfsrv_statehashsize && doneok == 0; i++) {
871 		    LIST_FOREACH(stp, &clp->lc_stateid[i], ls_hash) {
872 			if ((stp->ls_flags & NFSLCK_OPEN) &&
873 			    stp->ls_uid == nd->nd_cred->cr_uid) {
874 				doneok = 1;
875 				break;
876 			}
877 		    }
878 		}
879 		if (!doneok)
880 			error = NFSERR_ACCES;
881 	    }
882 	    if (!error && (clp->lc_flags & LCL_CBDOWN))
883 		error = NFSERR_CBPATHDOWN;
884 	}
885 	if ((!error || error == NFSERR_CBPATHDOWN) &&
886 	     (opflags & CLOPS_RENEW)) {
887 		clp->lc_expiry = nfsrv_leaseexpiry();
888 	}
889 	if (opflags & CLOPS_CONFIRM) {
890 		nfsrv_clientunlock(mlocked);
891 		if (client != NULL)
892 			CLNT_RELEASE(client);
893 		if (old_xprt != NULL)
894 			SVC_RELEASE(old_xprt);
895 	} else if (opflags != CLOPS_RENEW) {
896 		NFSUNLOCKSTATE();
897 	}
898 	if (clpp)
899 		*clpp = clp;
900 
901 out:
902 	NFSEXITCODE2(error, nd);
903 	return (error);
904 }
905 
906 /*
907  * Perform the NFSv4.1 destroy clientid.
908  */
909 int
910 nfsrv_destroyclient(struct nfsrv_descript *nd, nfsquad_t clientid, NFSPROC_T *p)
911 {
912 	struct nfsclient *clp;
913 	struct nfsclienthashhead *hp;
914 	SVCXPRT *old_xprt;
915 	int error = 0, i;
916 	bool mlocked;
917 
918 	if (NFSD_VNET(nfsrvboottime) != clientid.lval[0]) {
919 		error = NFSERR_STALECLIENTID;
920 		goto out;
921 	}
922 
923 	mlocked = true;
924 	if (nfsrv_dolocallocks != 0)
925 		mlocked = false;
926 	/* Lock out other nfsd threads */
927 	nfsrv_clientlock(mlocked);
928 
929 	hp = NFSCLIENTHASH(clientid);
930 	LIST_FOREACH(clp, hp, lc_hash) {
931 		if (clp->lc_clientid.lval[1] == clientid.lval[1])
932 			break;
933 	}
934 	if (clp == NULL) {
935 		nfsrv_clientunlock(mlocked);
936 		/* Just return ok, since it is gone. */
937 		goto out;
938 	}
939 
940 	/* Check for the SP4_MACH_CRED case. */
941 	error = nfsrv_checkmachcred(NFSV4OP_DESTROYCLIENTID, nd, clp);
942 	if (error != 0) {
943 		nfsrv_clientunlock(mlocked);
944 		goto out;
945 	}
946 
947 	/*
948 	 * Free up all layouts on the clientid.  Should the client return the
949 	 * layouts?
950 	 */
951 	nfsrv_freelayoutlist(clientid);
952 
953 	/* Scan for state on the clientid. */
954 	for (i = 0; i < nfsrv_statehashsize; i++)
955 		if (!LIST_EMPTY(&clp->lc_stateid[i])) {
956 			nfsrv_clientunlock(mlocked);
957 			error = NFSERR_CLIENTIDBUSY;
958 			goto out;
959 		}
960 	if (!LIST_EMPTY(&clp->lc_session) || !LIST_EMPTY(&clp->lc_deleg)) {
961 		nfsrv_clientunlock(mlocked);
962 		error = NFSERR_CLIENTIDBUSY;
963 		goto out;
964 	}
965 
966 	/* Destroy the clientid and return ok. */
967 	old_xprt = NULL;
968 	if (mlocked)
969 		nfsrv_cleanclient(clp, p, true, &old_xprt);
970 	else
971 		nfsrv_cleanclient(clp, p, false, NULL);
972 	nfsrv_freedeleglist(&clp->lc_deleg);
973 	nfsrv_freedeleglist(&clp->lc_olddeleg);
974 	LIST_REMOVE(clp, lc_hash);
975 	nfsrv_clientunlock(mlocked);
976 	if (old_xprt != NULL)
977 		SVC_RELEASE(old_xprt);
978 	nfsrv_zapclient(clp, p);
979 out:
980 	NFSEXITCODE2(error, nd);
981 	return (error);
982 }
983 
984 /*
985  * Called from the new nfssvc syscall to admin revoke a clientid.
986  * Returns 0 for success, error otherwise.
987  */
988 int
989 nfsrv_adminrevoke(struct nfsd_clid *revokep, NFSPROC_T *p)
990 {
991 	struct nfsclient *clp = NULL;
992 	int i, error = 0;
993 	int gotit, igotlock;
994 
995 	/*
996 	 * First, lock out the nfsd so that state won't change while the
997 	 * revocation record is being written to the stable storage restart
998 	 * file.
999 	 */
1000 	NFSLOCKV4ROOTMUTEX();
1001 	do {
1002 		igotlock = nfsv4_lock(&nfsv4rootfs_lock, 1, NULL,
1003 		    NFSV4ROOTLOCKMUTEXPTR, NULL);
1004 	} while (!igotlock);
1005 	NFSUNLOCKV4ROOTMUTEX();
1006 
1007 	/*
1008 	 * Search for a match in the client list.
1009 	 */
1010 	gotit = i = 0;
1011 	while (i < nfsrv_clienthashsize && !gotit) {
1012 	    LIST_FOREACH(clp, &NFSD_VNET(nfsclienthash)[i], lc_hash) {
1013 		if (revokep->nclid_idlen == clp->lc_idlen &&
1014 		    !NFSBCMP(revokep->nclid_id, clp->lc_id, clp->lc_idlen)) {
1015 			gotit = 1;
1016 			break;
1017 		}
1018 	    }
1019 	    i++;
1020 	}
1021 	if (!gotit) {
1022 		NFSLOCKV4ROOTMUTEX();
1023 		nfsv4_unlock(&nfsv4rootfs_lock, 0);
1024 		NFSUNLOCKV4ROOTMUTEX();
1025 		error = EPERM;
1026 		goto out;
1027 	}
1028 
1029 	/*
1030 	 * Now, write out the revocation record
1031 	 */
1032 	nfsrv_writestable(clp->lc_id, clp->lc_idlen, NFSNST_REVOKE, p);
1033 	nfsrv_backupstable();
1034 
1035 	/*
1036 	 * and clear out the state, marking the clientid revoked.
1037 	 */
1038 	clp->lc_flags &= ~LCL_CALLBACKSON;
1039 	clp->lc_flags |= LCL_ADMINREVOKED;
1040 	nfsrv_cleanclient(clp, p, false, NULL);
1041 	nfsrv_freedeleglist(&clp->lc_deleg);
1042 	nfsrv_freedeleglist(&clp->lc_olddeleg);
1043 	NFSLOCKV4ROOTMUTEX();
1044 	nfsv4_unlock(&nfsv4rootfs_lock, 0);
1045 	NFSUNLOCKV4ROOTMUTEX();
1046 
1047 out:
1048 	NFSEXITCODE(error);
1049 	return (error);
1050 }
1051 
1052 /*
1053  * Dump out stats for all clients. Called from nfssvc(2), that is used
1054  * nfsstatsv1.
1055  */
1056 void
1057 nfsrv_dumpclients(struct nfsd_dumpclients *dumpp, int maxcnt)
1058 {
1059 	struct nfsclient *clp;
1060 	int i = 0, cnt = 0;
1061 
1062 	/*
1063 	 * First, get a reference on the nfsv4rootfs_lock so that an
1064 	 * exclusive lock cannot be acquired while dumping the clients.
1065 	 */
1066 	NFSLOCKV4ROOTMUTEX();
1067 	nfsv4_getref(&nfsv4rootfs_lock, NULL, NFSV4ROOTLOCKMUTEXPTR, NULL);
1068 	NFSUNLOCKV4ROOTMUTEX();
1069 	NFSLOCKSTATE();
1070 	/*
1071 	 * Rattle through the client lists until done.
1072 	 */
1073 	while (i < nfsrv_clienthashsize && cnt < maxcnt) {
1074 	    clp = LIST_FIRST(&NFSD_VNET(nfsclienthash)[i]);
1075 	    while (clp != LIST_END(&NFSD_VNET(nfsclienthash)[i]) && cnt <
1076 		maxcnt) {
1077 		nfsrv_dumpaclient(clp, &dumpp[cnt]);
1078 		cnt++;
1079 		clp = LIST_NEXT(clp, lc_hash);
1080 	    }
1081 	    i++;
1082 	}
1083 	if (cnt < maxcnt)
1084 	    dumpp[cnt].ndcl_clid.nclid_idlen = 0;
1085 	NFSUNLOCKSTATE();
1086 	NFSLOCKV4ROOTMUTEX();
1087 	nfsv4_relref(&nfsv4rootfs_lock);
1088 	NFSUNLOCKV4ROOTMUTEX();
1089 }
1090 
1091 /*
1092  * Dump stats for a client. Must be called with the NFSSTATELOCK and spl'd.
1093  */
1094 static void
1095 nfsrv_dumpaclient(struct nfsclient *clp, struct nfsd_dumpclients *dumpp)
1096 {
1097 	struct nfsstate *stp, *openstp, *lckownstp;
1098 	struct nfslock *lop;
1099 	sa_family_t af;
1100 #ifdef INET
1101 	struct sockaddr_in *rin;
1102 #endif
1103 #ifdef INET6
1104 	struct sockaddr_in6 *rin6;
1105 #endif
1106 
1107 	dumpp->ndcl_nopenowners = dumpp->ndcl_nlockowners = 0;
1108 	dumpp->ndcl_nopens = dumpp->ndcl_nlocks = 0;
1109 	dumpp->ndcl_ndelegs = dumpp->ndcl_nolddelegs = 0;
1110 	dumpp->ndcl_flags = clp->lc_flags;
1111 	dumpp->ndcl_clid.nclid_idlen = clp->lc_idlen;
1112 	NFSBCOPY(clp->lc_id, dumpp->ndcl_clid.nclid_id, clp->lc_idlen);
1113 	af = clp->lc_req.nr_nam->sa_family;
1114 	dumpp->ndcl_addrfam = af;
1115 	switch (af) {
1116 #ifdef INET
1117 	case AF_INET:
1118 		rin = (struct sockaddr_in *)clp->lc_req.nr_nam;
1119 		dumpp->ndcl_cbaddr.sin_addr = rin->sin_addr;
1120 		break;
1121 #endif
1122 #ifdef INET6
1123 	case AF_INET6:
1124 		rin6 = (struct sockaddr_in6 *)clp->lc_req.nr_nam;
1125 		dumpp->ndcl_cbaddr.sin6_addr = rin6->sin6_addr;
1126 		break;
1127 #endif
1128 	}
1129 
1130 	/*
1131 	 * Now, scan the state lists and total up the opens and locks.
1132 	 */
1133 	LIST_FOREACH(stp, &clp->lc_open, ls_list) {
1134 	    dumpp->ndcl_nopenowners++;
1135 	    LIST_FOREACH(openstp, &stp->ls_open, ls_list) {
1136 		dumpp->ndcl_nopens++;
1137 		LIST_FOREACH(lckownstp, &openstp->ls_open, ls_list) {
1138 		    dumpp->ndcl_nlockowners++;
1139 		    LIST_FOREACH(lop, &lckownstp->ls_lock, lo_lckowner) {
1140 			dumpp->ndcl_nlocks++;
1141 		    }
1142 		}
1143 	    }
1144 	}
1145 
1146 	/*
1147 	 * and the delegation lists.
1148 	 */
1149 	LIST_FOREACH(stp, &clp->lc_deleg, ls_list) {
1150 	    dumpp->ndcl_ndelegs++;
1151 	}
1152 	LIST_FOREACH(stp, &clp->lc_olddeleg, ls_list) {
1153 	    dumpp->ndcl_nolddelegs++;
1154 	}
1155 }
1156 
1157 /*
1158  * Dump out lock stats for a file.
1159  */
1160 void
1161 nfsrv_dumplocks(vnode_t vp, struct nfsd_dumplocks *ldumpp, int maxcnt,
1162     NFSPROC_T *p)
1163 {
1164 	struct nfsstate *stp;
1165 	struct nfslock *lop;
1166 	int cnt = 0;
1167 	struct nfslockfile *lfp;
1168 	sa_family_t af;
1169 #ifdef INET
1170 	struct sockaddr_in *rin;
1171 #endif
1172 #ifdef INET6
1173 	struct sockaddr_in6 *rin6;
1174 #endif
1175 	int ret;
1176 	fhandle_t nfh;
1177 
1178 	ret = nfsrv_getlockfh(vp, 0, NULL, &nfh, p);
1179 	/*
1180 	 * First, get a reference on the nfsv4rootfs_lock so that an
1181 	 * exclusive lock on it cannot be acquired while dumping the locks.
1182 	 */
1183 	NFSLOCKV4ROOTMUTEX();
1184 	nfsv4_getref(&nfsv4rootfs_lock, NULL, NFSV4ROOTLOCKMUTEXPTR, NULL);
1185 	NFSUNLOCKV4ROOTMUTEX();
1186 	NFSLOCKSTATE();
1187 	if (!ret)
1188 		ret = nfsrv_getlockfile(0, NULL, &lfp, &nfh, 0);
1189 	if (ret) {
1190 		ldumpp[0].ndlck_clid.nclid_idlen = 0;
1191 		NFSUNLOCKSTATE();
1192 		NFSLOCKV4ROOTMUTEX();
1193 		nfsv4_relref(&nfsv4rootfs_lock);
1194 		NFSUNLOCKV4ROOTMUTEX();
1195 		return;
1196 	}
1197 
1198 	/*
1199 	 * For each open share on file, dump it out.
1200 	 */
1201 	stp = LIST_FIRST(&lfp->lf_open);
1202 	while (stp != LIST_END(&lfp->lf_open) && cnt < maxcnt) {
1203 		ldumpp[cnt].ndlck_flags = stp->ls_flags;
1204 		ldumpp[cnt].ndlck_stateid.seqid = stp->ls_stateid.seqid;
1205 		ldumpp[cnt].ndlck_stateid.other[0] = stp->ls_stateid.other[0];
1206 		ldumpp[cnt].ndlck_stateid.other[1] = stp->ls_stateid.other[1];
1207 		ldumpp[cnt].ndlck_stateid.other[2] = stp->ls_stateid.other[2];
1208 		ldumpp[cnt].ndlck_owner.nclid_idlen =
1209 		    stp->ls_openowner->ls_ownerlen;
1210 		NFSBCOPY(stp->ls_openowner->ls_owner,
1211 		    ldumpp[cnt].ndlck_owner.nclid_id,
1212 		    stp->ls_openowner->ls_ownerlen);
1213 		ldumpp[cnt].ndlck_clid.nclid_idlen = stp->ls_clp->lc_idlen;
1214 		NFSBCOPY(stp->ls_clp->lc_id, ldumpp[cnt].ndlck_clid.nclid_id,
1215 		    stp->ls_clp->lc_idlen);
1216 		af = stp->ls_clp->lc_req.nr_nam->sa_family;
1217 		ldumpp[cnt].ndlck_addrfam = af;
1218 		switch (af) {
1219 #ifdef INET
1220 		case AF_INET:
1221 			rin = (struct sockaddr_in *)stp->ls_clp->lc_req.nr_nam;
1222 			ldumpp[cnt].ndlck_cbaddr.sin_addr = rin->sin_addr;
1223 			break;
1224 #endif
1225 #ifdef INET6
1226 		case AF_INET6:
1227 			rin6 = (struct sockaddr_in6 *)
1228 			    stp->ls_clp->lc_req.nr_nam;
1229 			ldumpp[cnt].ndlck_cbaddr.sin6_addr = rin6->sin6_addr;
1230 			break;
1231 #endif
1232 		}
1233 		stp = LIST_NEXT(stp, ls_file);
1234 		cnt++;
1235 	}
1236 
1237 	/*
1238 	 * and all locks.
1239 	 */
1240 	lop = LIST_FIRST(&lfp->lf_lock);
1241 	while (lop != LIST_END(&lfp->lf_lock) && cnt < maxcnt) {
1242 		stp = lop->lo_stp;
1243 		ldumpp[cnt].ndlck_flags = lop->lo_flags;
1244 		ldumpp[cnt].ndlck_first = lop->lo_first;
1245 		ldumpp[cnt].ndlck_end = lop->lo_end;
1246 		ldumpp[cnt].ndlck_stateid.seqid = stp->ls_stateid.seqid;
1247 		ldumpp[cnt].ndlck_stateid.other[0] = stp->ls_stateid.other[0];
1248 		ldumpp[cnt].ndlck_stateid.other[1] = stp->ls_stateid.other[1];
1249 		ldumpp[cnt].ndlck_stateid.other[2] = stp->ls_stateid.other[2];
1250 		ldumpp[cnt].ndlck_owner.nclid_idlen = stp->ls_ownerlen;
1251 		NFSBCOPY(stp->ls_owner, ldumpp[cnt].ndlck_owner.nclid_id,
1252 		    stp->ls_ownerlen);
1253 		ldumpp[cnt].ndlck_clid.nclid_idlen = stp->ls_clp->lc_idlen;
1254 		NFSBCOPY(stp->ls_clp->lc_id, ldumpp[cnt].ndlck_clid.nclid_id,
1255 		    stp->ls_clp->lc_idlen);
1256 		af = stp->ls_clp->lc_req.nr_nam->sa_family;
1257 		ldumpp[cnt].ndlck_addrfam = af;
1258 		switch (af) {
1259 #ifdef INET
1260 		case AF_INET:
1261 			rin = (struct sockaddr_in *)stp->ls_clp->lc_req.nr_nam;
1262 			ldumpp[cnt].ndlck_cbaddr.sin_addr = rin->sin_addr;
1263 			break;
1264 #endif
1265 #ifdef INET6
1266 		case AF_INET6:
1267 			rin6 = (struct sockaddr_in6 *)
1268 			    stp->ls_clp->lc_req.nr_nam;
1269 			ldumpp[cnt].ndlck_cbaddr.sin6_addr = rin6->sin6_addr;
1270 			break;
1271 #endif
1272 		}
1273 		lop = LIST_NEXT(lop, lo_lckfile);
1274 		cnt++;
1275 	}
1276 
1277 	/*
1278 	 * and the delegations.
1279 	 */
1280 	stp = LIST_FIRST(&lfp->lf_deleg);
1281 	while (stp != LIST_END(&lfp->lf_deleg) && cnt < maxcnt) {
1282 		ldumpp[cnt].ndlck_flags = stp->ls_flags;
1283 		ldumpp[cnt].ndlck_stateid.seqid = stp->ls_stateid.seqid;
1284 		ldumpp[cnt].ndlck_stateid.other[0] = stp->ls_stateid.other[0];
1285 		ldumpp[cnt].ndlck_stateid.other[1] = stp->ls_stateid.other[1];
1286 		ldumpp[cnt].ndlck_stateid.other[2] = stp->ls_stateid.other[2];
1287 		ldumpp[cnt].ndlck_owner.nclid_idlen = 0;
1288 		ldumpp[cnt].ndlck_clid.nclid_idlen = stp->ls_clp->lc_idlen;
1289 		NFSBCOPY(stp->ls_clp->lc_id, ldumpp[cnt].ndlck_clid.nclid_id,
1290 		    stp->ls_clp->lc_idlen);
1291 		af = stp->ls_clp->lc_req.nr_nam->sa_family;
1292 		ldumpp[cnt].ndlck_addrfam = af;
1293 		switch (af) {
1294 #ifdef INET
1295 		case AF_INET:
1296 			rin = (struct sockaddr_in *)stp->ls_clp->lc_req.nr_nam;
1297 			ldumpp[cnt].ndlck_cbaddr.sin_addr = rin->sin_addr;
1298 			break;
1299 #endif
1300 #ifdef INET6
1301 		case AF_INET6:
1302 			rin6 = (struct sockaddr_in6 *)
1303 			    stp->ls_clp->lc_req.nr_nam;
1304 			ldumpp[cnt].ndlck_cbaddr.sin6_addr = rin6->sin6_addr;
1305 			break;
1306 #endif
1307 		}
1308 		stp = LIST_NEXT(stp, ls_file);
1309 		cnt++;
1310 	}
1311 
1312 	/*
1313 	 * If list isn't full, mark end of list by setting the client name
1314 	 * to zero length.
1315 	 */
1316 	if (cnt < maxcnt)
1317 		ldumpp[cnt].ndlck_clid.nclid_idlen = 0;
1318 	NFSUNLOCKSTATE();
1319 	NFSLOCKV4ROOTMUTEX();
1320 	nfsv4_relref(&nfsv4rootfs_lock);
1321 	NFSUNLOCKV4ROOTMUTEX();
1322 }
1323 
1324 /*
1325  * Server timer routine. It can scan any linked list, so long
1326  * as it holds the spin/mutex lock and there is no exclusive lock on
1327  * nfsv4rootfs_lock.
1328  * (For OpenBSD, a kthread is ok. For FreeBSD, I think it is ok
1329  *  to do this from a callout, since the spin locks work. For
1330  *  Darwin, I'm not sure what will work correctly yet.)
1331  * Should be called once per second.
1332  */
1333 void
1334 nfsrv_servertimer(void *arg __unused)
1335 {
1336 	struct nfsclient *clp, *nclp;
1337 	struct nfsstate *stp, *nstp;
1338 	int got_ref, i;
1339 
1340 	/*
1341 	 * Make sure nfsboottime is set. This is used by V3 as well
1342 	 * as V4. Note that nfsboottime is not nfsrvboottime, which is
1343 	 * only used by the V4 server for leases.
1344 	 */
1345 	if (nfsboottime.tv_sec == 0)
1346 		NFSSETBOOTTIME(nfsboottime);
1347 
1348 	/*
1349 	 * If server hasn't started yet, just return.
1350 	 */
1351 	NFSLOCKSTATE();
1352 	if (NFSD_VNET(nfsrv_stablefirst).nsf_eograce == 0) {
1353 		NFSUNLOCKSTATE();
1354 		return;
1355 	}
1356 	if (!(NFSD_VNET(nfsrv_stablefirst).nsf_flags & NFSNSF_UPDATEDONE)) {
1357 		if (!(NFSD_VNET(nfsrv_stablefirst).nsf_flags &
1358 		      NFSNSF_GRACEOVER) &&
1359 		    NFSD_MONOSEC > NFSD_VNET(nfsrv_stablefirst).nsf_eograce)
1360 			NFSD_VNET(nfsrv_stablefirst).nsf_flags |=
1361 			    (NFSNSF_GRACEOVER | NFSNSF_NEEDLOCK);
1362 		NFSUNLOCKSTATE();
1363 		return;
1364 	}
1365 
1366 	/*
1367 	 * Try and get a reference count on the nfsv4rootfs_lock so that
1368 	 * no nfsd thread can acquire an exclusive lock on it before this
1369 	 * call is done. If it is already exclusively locked, just return.
1370 	 */
1371 	NFSLOCKV4ROOTMUTEX();
1372 	got_ref = nfsv4_getref_nonblock(&nfsv4rootfs_lock);
1373 	NFSUNLOCKV4ROOTMUTEX();
1374 	if (got_ref == 0) {
1375 		NFSUNLOCKSTATE();
1376 		return;
1377 	}
1378 
1379 	/*
1380 	 * For each client...
1381 	 */
1382 	for (i = 0; i < nfsrv_clienthashsize; i++) {
1383 	    clp = LIST_FIRST(&NFSD_VNET(nfsclienthash)[i]);
1384 	    while (clp != LIST_END(&NFSD_VNET(nfsclienthash)[i])) {
1385 		nclp = LIST_NEXT(clp, lc_hash);
1386 		if (!(clp->lc_flags & LCL_EXPIREIT)) {
1387 		    if (((clp->lc_expiry + NFSRV_STALELEASE) < NFSD_MONOSEC
1388 			 && ((LIST_EMPTY(&clp->lc_deleg)
1389 			      && LIST_EMPTY(&clp->lc_open)) ||
1390 			     nfsrv_clients > nfsrv_clienthighwater)) ||
1391 			(clp->lc_expiry + NFSRV_MOULDYLEASE) < NFSD_MONOSEC ||
1392 			(clp->lc_expiry < NFSD_MONOSEC &&
1393 			 (nfsrv_openpluslock * 10 / 9) > nfsrv_v4statelimit)) {
1394 			/*
1395 			 * Lease has expired several nfsrv_lease times ago:
1396 			 * PLUS
1397 			 *    - no state is associated with it
1398 			 *    OR
1399 			 *    - above high water mark for number of clients
1400 			 *      (nfsrv_clienthighwater should be large enough
1401 			 *       that this only occurs when clients fail to
1402 			 *       use the same nfs_client_id4.id. Maybe somewhat
1403 			 *       higher that the maximum number of clients that
1404 			 *       will mount this server?)
1405 			 * OR
1406 			 * Lease has expired a very long time ago
1407 			 * OR
1408 			 * Lease has expired PLUS the number of opens + locks
1409 			 * has exceeded 90% of capacity
1410 			 *
1411 			 * --> Mark for expiry. The actual expiry will be done
1412 			 *     by an nfsd sometime soon.
1413 			 */
1414 			clp->lc_flags |= LCL_EXPIREIT;
1415 			NFSD_VNET(nfsrv_stablefirst).nsf_flags |=
1416 			    (NFSNSF_NEEDLOCK | NFSNSF_EXPIREDCLIENT);
1417 		    } else {
1418 			/*
1419 			 * If there are no opens, increment no open tick cnt
1420 			 * If time exceeds NFSNOOPEN, mark it to be thrown away
1421 			 * otherwise, if there is an open, reset no open time
1422 			 * Hopefully, this will avoid excessive re-creation
1423 			 * of open owners and subsequent open confirms.
1424 			 */
1425 			stp = LIST_FIRST(&clp->lc_open);
1426 			while (stp != LIST_END(&clp->lc_open)) {
1427 				nstp = LIST_NEXT(stp, ls_list);
1428 				if (LIST_EMPTY(&stp->ls_open)) {
1429 					stp->ls_noopens++;
1430 					if (stp->ls_noopens > NFSNOOPEN ||
1431 					    (nfsrv_openpluslock * 2) >
1432 					    nfsrv_v4statelimit)
1433 						NFSD_VNET(nfsrv_stablefirst).nsf_flags |=
1434 							NFSNSF_NOOPENS;
1435 				} else {
1436 					stp->ls_noopens = 0;
1437 				}
1438 				stp = nstp;
1439 			}
1440 		    }
1441 		}
1442 		clp = nclp;
1443 	    }
1444 	}
1445 	NFSUNLOCKSTATE();
1446 	NFSLOCKV4ROOTMUTEX();
1447 	nfsv4_relref(&nfsv4rootfs_lock);
1448 	NFSUNLOCKV4ROOTMUTEX();
1449 }
1450 
1451 /*
1452  * The following set of functions free up the various data structures.
1453  */
1454 /*
1455  * Clear out all open/lock state related to this nfsclient.
1456  * Caller must hold an exclusive lock on nfsv4rootfs_lock, so that
1457  * there are no other active nfsd threads.
1458  */
1459 void
1460 nfsrv_cleanclient(struct nfsclient *clp, NFSPROC_T *p, bool locked,
1461     SVCXPRT **old_xprtp)
1462 {
1463 	struct nfsstate *stp, *nstp;
1464 	struct nfsdsession *sep, *nsep;
1465 
1466 	LIST_FOREACH_SAFE(stp, &clp->lc_open, ls_list, nstp) {
1467 		if (locked)
1468 			nfsrv_freeopenowner(stp, 0, p);
1469 		else
1470 			nfsrv_freeopenowner(stp, 1, p);
1471 	}
1472 	if ((clp->lc_flags & LCL_ADMINREVOKED) == 0)
1473 		LIST_FOREACH_SAFE(sep, &clp->lc_session, sess_list, nsep)
1474 			(void)nfsrv_freesession(NULL, sep, NULL, locked,
1475 			    old_xprtp);
1476 }
1477 
1478 /*
1479  * Free a client that has been cleaned. It should also already have been
1480  * removed from the lists.
1481  * (Just to be safe w.r.t. newnfs_disconnect(), call this function when
1482  *  softclock interrupts are enabled.)
1483  */
1484 void
1485 nfsrv_zapclient(struct nfsclient *clp, NFSPROC_T *p)
1486 {
1487 
1488 #ifdef notyet
1489 	if ((clp->lc_flags & (LCL_GSS | LCL_CALLBACKSON)) ==
1490 	     (LCL_GSS | LCL_CALLBACKSON) &&
1491 	    (clp->lc_hand.nfsh_flag & NFSG_COMPLETE) &&
1492 	    clp->lc_handlelen > 0) {
1493 		clp->lc_hand.nfsh_flag &= ~NFSG_COMPLETE;
1494 		clp->lc_hand.nfsh_flag |= NFSG_DESTROYED;
1495 		(void) nfsrv_docallback(clp, NFSV4PROC_CBNULL,
1496 			NULL, 0, NULL, NULL, NULL, 0, p);
1497 	}
1498 #endif
1499 	newnfs_disconnect(NULL, &clp->lc_req);
1500 	free(clp->lc_req.nr_nam, M_SONAME);
1501 	NFSFREEMUTEX(&clp->lc_req.nr_mtx);
1502 	free(clp->lc_stateid, M_NFSDCLIENT);
1503 	free(clp, M_NFSDCLIENT);
1504 	NFSLOCKSTATE();
1505 	NFSD_VNET(nfsstatsv1_p)->srvclients--;
1506 	nfsrv_openpluslock--;
1507 	nfsrv_clients--;
1508 	NFSUNLOCKSTATE();
1509 }
1510 
1511 /*
1512  * Free a list of delegation state structures.
1513  * (This function will also free all nfslockfile structures that no
1514  *  longer have associated state.)
1515  */
1516 void
1517 nfsrv_freedeleglist(struct nfsstatehead *sthp)
1518 {
1519 	struct nfsstate *stp, *nstp;
1520 
1521 	LIST_FOREACH_SAFE(stp, sthp, ls_list, nstp) {
1522 		nfsrv_freedeleg(stp);
1523 	}
1524 	LIST_INIT(sthp);
1525 }
1526 
1527 /*
1528  * Free up a delegation.
1529  */
1530 static void
1531 nfsrv_freedeleg(struct nfsstate *stp)
1532 {
1533 	struct nfslockfile *lfp;
1534 
1535 	LIST_REMOVE(stp, ls_hash);
1536 	LIST_REMOVE(stp, ls_list);
1537 	LIST_REMOVE(stp, ls_file);
1538 	if ((stp->ls_flags & NFSLCK_DELEGWRITE) != 0)
1539 		nfsrv_writedelegcnt--;
1540 	lfp = stp->ls_lfp;
1541 	if (LIST_EMPTY(&lfp->lf_open) &&
1542 	    LIST_EMPTY(&lfp->lf_lock) && LIST_EMPTY(&lfp->lf_deleg) &&
1543 	    LIST_EMPTY(&lfp->lf_locallock) && LIST_EMPTY(&lfp->lf_rollback) &&
1544 	    lfp->lf_usecount == 0 &&
1545 	    nfsv4_testlock(&lfp->lf_locallock_lck) == 0)
1546 		nfsrv_freenfslockfile(lfp);
1547 	free(stp, M_NFSDSTATE);
1548 	NFSD_VNET(nfsstatsv1_p)->srvdelegates--;
1549 	nfsrv_openpluslock--;
1550 	nfsrv_delegatecnt--;
1551 }
1552 
1553 /*
1554  * This function frees an open owner and all associated opens.
1555  */
1556 static void
1557 nfsrv_freeopenowner(struct nfsstate *stp, int cansleep, NFSPROC_T *p)
1558 {
1559 	struct nfsstate *nstp, *tstp;
1560 
1561 	LIST_REMOVE(stp, ls_list);
1562 	/*
1563 	 * Now, free all associated opens.
1564 	 */
1565 	nstp = LIST_FIRST(&stp->ls_open);
1566 	while (nstp != LIST_END(&stp->ls_open)) {
1567 		tstp = nstp;
1568 		nstp = LIST_NEXT(nstp, ls_list);
1569 		(void) nfsrv_freeopen(tstp, NULL, cansleep, p);
1570 	}
1571 	if (stp->ls_op)
1572 		nfsrvd_derefcache(stp->ls_op);
1573 	free(stp, M_NFSDSTATE);
1574 	NFSD_VNET(nfsstatsv1_p)->srvopenowners--;
1575 	nfsrv_openpluslock--;
1576 }
1577 
1578 /*
1579  * This function frees an open (nfsstate open structure) with all associated
1580  * lock_owners and locks. It also frees the nfslockfile structure iff there
1581  * are no other opens on the file.
1582  * Returns 1 if it free'd the nfslockfile, 0 otherwise.
1583  */
1584 static int
1585 nfsrv_freeopen(struct nfsstate *stp, vnode_t vp, int cansleep, NFSPROC_T *p)
1586 {
1587 	struct nfsstate *nstp, *tstp;
1588 	struct nfslockfile *lfp;
1589 	int ret;
1590 
1591 	LIST_REMOVE(stp, ls_hash);
1592 	LIST_REMOVE(stp, ls_list);
1593 	LIST_REMOVE(stp, ls_file);
1594 
1595 	lfp = stp->ls_lfp;
1596 	/*
1597 	 * Now, free all lockowners associated with this open.
1598 	 */
1599 	LIST_FOREACH_SAFE(tstp, &stp->ls_open, ls_list, nstp)
1600 		nfsrv_freelockowner(tstp, vp, cansleep, p);
1601 
1602 	/*
1603 	 * The nfslockfile is freed here if there are no locks
1604 	 * associated with the open.
1605 	 * If there are locks associated with the open, the
1606 	 * nfslockfile structure can be freed via nfsrv_freelockowner().
1607 	 * Acquire the state mutex to avoid races with calls to
1608 	 * nfsrv_getlockfile().
1609 	 */
1610 	if (cansleep != 0)
1611 		NFSLOCKSTATE();
1612 	if (lfp != NULL && LIST_EMPTY(&lfp->lf_open) &&
1613 	    LIST_EMPTY(&lfp->lf_deleg) && LIST_EMPTY(&lfp->lf_lock) &&
1614 	    LIST_EMPTY(&lfp->lf_locallock) && LIST_EMPTY(&lfp->lf_rollback) &&
1615 	    lfp->lf_usecount == 0 &&
1616 	    (cansleep != 0 || nfsv4_testlock(&lfp->lf_locallock_lck) == 0)) {
1617 		nfsrv_freenfslockfile(lfp);
1618 		ret = 1;
1619 	} else
1620 		ret = 0;
1621 	if (cansleep != 0)
1622 		NFSUNLOCKSTATE();
1623 	free(stp, M_NFSDSTATE);
1624 	NFSD_VNET(nfsstatsv1_p)->srvopens--;
1625 	nfsrv_openpluslock--;
1626 	return (ret);
1627 }
1628 
1629 /*
1630  * Frees a lockowner and all associated locks.
1631  */
1632 static void
1633 nfsrv_freelockowner(struct nfsstate *stp, vnode_t vp, int cansleep,
1634     NFSPROC_T *p)
1635 {
1636 
1637 	LIST_REMOVE(stp, ls_hash);
1638 	LIST_REMOVE(stp, ls_list);
1639 	nfsrv_freeallnfslocks(stp, vp, cansleep, p);
1640 	if (stp->ls_op)
1641 		nfsrvd_derefcache(stp->ls_op);
1642 	free(stp, M_NFSDSTATE);
1643 	NFSD_VNET(nfsstatsv1_p)->srvlockowners--;
1644 	nfsrv_openpluslock--;
1645 }
1646 
1647 /*
1648  * Free all the nfs locks on a lockowner.
1649  */
1650 static void
1651 nfsrv_freeallnfslocks(struct nfsstate *stp, vnode_t vp, int cansleep,
1652     NFSPROC_T *p)
1653 {
1654 	struct nfslock *lop, *nlop;
1655 	struct nfsrollback *rlp, *nrlp;
1656 	struct nfslockfile *lfp = NULL;
1657 	int gottvp = 0;
1658 	vnode_t tvp = NULL;
1659 	uint64_t first, end;
1660 
1661 	if (vp != NULL)
1662 		ASSERT_VOP_UNLOCKED(vp, "nfsrv_freeallnfslocks: vnode locked");
1663 	lop = LIST_FIRST(&stp->ls_lock);
1664 	while (lop != LIST_END(&stp->ls_lock)) {
1665 		nlop = LIST_NEXT(lop, lo_lckowner);
1666 		/*
1667 		 * Since all locks should be for the same file, lfp should
1668 		 * not change.
1669 		 */
1670 		if (lfp == NULL)
1671 			lfp = lop->lo_lfp;
1672 		else if (lfp != lop->lo_lfp)
1673 			panic("allnfslocks");
1674 		/*
1675 		 * If vp is NULL and cansleep != 0, a vnode must be acquired
1676 		 * from the file handle. This only occurs when called from
1677 		 * nfsrv_cleanclient().
1678 		 */
1679 		if (gottvp == 0) {
1680 			if (nfsrv_dolocallocks == 0)
1681 				tvp = NULL;
1682 			else if (vp == NULL && cansleep != 0) {
1683 				tvp = nfsvno_getvp(&lfp->lf_fh);
1684 				if (tvp != NULL)
1685 					NFSVOPUNLOCK(tvp);
1686 			} else
1687 				tvp = vp;
1688 			gottvp = 1;
1689 		}
1690 
1691 		if (tvp != NULL) {
1692 			if (cansleep == 0)
1693 				panic("allnfs2");
1694 			first = lop->lo_first;
1695 			end = lop->lo_end;
1696 			nfsrv_freenfslock(lop);
1697 			nfsrv_localunlock(tvp, lfp, first, end, p);
1698 			LIST_FOREACH_SAFE(rlp, &lfp->lf_rollback, rlck_list,
1699 			    nrlp)
1700 				free(rlp, M_NFSDROLLBACK);
1701 			LIST_INIT(&lfp->lf_rollback);
1702 		} else
1703 			nfsrv_freenfslock(lop);
1704 		lop = nlop;
1705 	}
1706 	if (vp == NULL && tvp != NULL)
1707 		vrele(tvp);
1708 }
1709 
1710 /*
1711  * Free an nfslock structure.
1712  */
1713 static void
1714 nfsrv_freenfslock(struct nfslock *lop)
1715 {
1716 
1717 	if (lop->lo_lckfile.le_prev != NULL) {
1718 		LIST_REMOVE(lop, lo_lckfile);
1719 		NFSD_VNET(nfsstatsv1_p)->srvlocks--;
1720 		nfsrv_openpluslock--;
1721 	}
1722 	LIST_REMOVE(lop, lo_lckowner);
1723 	free(lop, M_NFSDLOCK);
1724 }
1725 
1726 /*
1727  * This function frees an nfslockfile structure.
1728  */
1729 static void
1730 nfsrv_freenfslockfile(struct nfslockfile *lfp)
1731 {
1732 
1733 	LIST_REMOVE(lfp, lf_hash);
1734 	free(lfp, M_NFSDLOCKFILE);
1735 }
1736 
1737 /*
1738  * This function looks up an nfsstate structure via stateid.
1739  */
1740 static int
1741 nfsrv_getstate(struct nfsclient *clp, nfsv4stateid_t *stateidp, __unused u_int32_t flags,
1742     struct nfsstate **stpp)
1743 {
1744 	struct nfsstate *stp;
1745 	struct nfsstatehead *hp;
1746 	int error = 0;
1747 
1748 	*stpp = NULL;
1749 	hp = NFSSTATEHASH(clp, *stateidp);
1750 	LIST_FOREACH(stp, hp, ls_hash) {
1751 		if (!NFSBCMP(stp->ls_stateid.other, stateidp->other,
1752 			NFSX_STATEIDOTHER))
1753 			break;
1754 	}
1755 
1756 	/*
1757 	 * If no state id in list, return NFSERR_BADSTATEID.
1758 	 */
1759 	if (stp == LIST_END(hp)) {
1760 		error = NFSERR_BADSTATEID;
1761 		goto out;
1762 	}
1763 	*stpp = stp;
1764 
1765 out:
1766 	NFSEXITCODE(error);
1767 	return (error);
1768 }
1769 
1770 /*
1771  * This function gets an nfsstate structure via owner string.
1772  */
1773 static void
1774 nfsrv_getowner(struct nfsstatehead *hp, struct nfsstate *new_stp,
1775     struct nfsstate **stpp)
1776 {
1777 	struct nfsstate *stp;
1778 
1779 	*stpp = NULL;
1780 	LIST_FOREACH(stp, hp, ls_list) {
1781 		if (new_stp->ls_ownerlen == stp->ls_ownerlen &&
1782 		  !NFSBCMP(new_stp->ls_owner,stp->ls_owner,stp->ls_ownerlen)) {
1783 			*stpp = stp;
1784 			return;
1785 		}
1786 	}
1787 }
1788 
1789 /*
1790  * Lock control function called to update lock status.
1791  * Returns 0 upon success, -1 if there is no lock and the flags indicate
1792  * that one isn't to be created and an NFSERR_xxx for other errors.
1793  * The structures new_stp and new_lop are passed in as pointers that should
1794  * be set to NULL if the structure is used and shouldn't be free'd.
1795  * For the NFSLCK_TEST and NFSLCK_CHECK cases, the structures are
1796  * never used and can safely be allocated on the stack. For all other
1797  * cases, *new_stpp and *new_lopp should be malloc'd before the call,
1798  * in case they are used.
1799  */
1800 int
1801 nfsrv_lockctrl(vnode_t vp, struct nfsstate **new_stpp,
1802     struct nfslock **new_lopp, struct nfslockconflict *cfp,
1803     nfsquad_t clientid, nfsv4stateid_t *stateidp,
1804     __unused struct nfsexstuff *exp,
1805     struct nfsrv_descript *nd, NFSPROC_T *p)
1806 {
1807 	struct nfslock *lop;
1808 	struct nfsstate *new_stp = *new_stpp;
1809 	struct nfslock *new_lop = *new_lopp;
1810 	struct nfsstate *tstp, *mystp, *nstp;
1811 	int specialid = 0;
1812 	struct nfslockfile *lfp;
1813 	struct nfslock *other_lop = NULL;
1814 	struct nfsstate *stp, *lckstp = NULL;
1815 	struct nfsclient *clp = NULL;
1816 	u_int32_t bits;
1817 	int error = 0, haslock = 0, ret, reterr;
1818 	int getlckret, delegation = 0, filestruct_locked, vnode_unlocked = 0;
1819 	fhandle_t nfh;
1820 	uint64_t first, end;
1821 	uint32_t lock_flags;
1822 
1823 	if (new_stp->ls_flags & (NFSLCK_CHECK | NFSLCK_SETATTR)) {
1824 		/*
1825 		 * Note the special cases of "all 1s" or "all 0s" stateids and
1826 		 * let reads with all 1s go ahead.
1827 		 */
1828 		if (new_stp->ls_stateid.seqid == 0x0 &&
1829 		    new_stp->ls_stateid.other[0] == 0x0 &&
1830 		    new_stp->ls_stateid.other[1] == 0x0 &&
1831 		    new_stp->ls_stateid.other[2] == 0x0)
1832 			specialid = 1;
1833 		else if (new_stp->ls_stateid.seqid == 0xffffffff &&
1834 		    new_stp->ls_stateid.other[0] == 0xffffffff &&
1835 		    new_stp->ls_stateid.other[1] == 0xffffffff &&
1836 		    new_stp->ls_stateid.other[2] == 0xffffffff)
1837 			specialid = 2;
1838 	}
1839 
1840 	/*
1841 	 * Check for restart conditions (client and server).
1842 	 */
1843 	error = nfsrv_checkrestart(clientid, new_stp->ls_flags,
1844 	    &new_stp->ls_stateid, specialid);
1845 	if (error)
1846 		goto out;
1847 
1848 	/*
1849 	 * Check for state resource limit exceeded.
1850 	 */
1851 	if ((new_stp->ls_flags & NFSLCK_LOCK) &&
1852 	    nfsrv_openpluslock > nfsrv_v4statelimit) {
1853 		error = NFSERR_RESOURCE;
1854 		goto out;
1855 	}
1856 
1857 	/*
1858 	 * For the lock case, get another nfslock structure,
1859 	 * just in case we need it.
1860 	 * Malloc now, before we start sifting through the linked lists,
1861 	 * in case we have to wait for memory.
1862 	 */
1863 tryagain:
1864 	if (new_stp->ls_flags & NFSLCK_LOCK)
1865 		other_lop = malloc(sizeof (struct nfslock),
1866 		    M_NFSDLOCK, M_WAITOK);
1867 	filestruct_locked = 0;
1868 	reterr = 0;
1869 	lfp = NULL;
1870 
1871 	/*
1872 	 * Get the lockfile structure for CFH now, so we can do a sanity
1873 	 * check against the stateid, before incrementing the seqid#, since
1874 	 * we want to return NFSERR_BADSTATEID on failure and the seqid#
1875 	 * shouldn't be incremented for this case.
1876 	 * If nfsrv_getlockfile() returns -1, it means "not found", which
1877 	 * will be handled later.
1878 	 * If we are doing Lock/LockU and local locking is enabled, sleep
1879 	 * lock the nfslockfile structure.
1880 	 */
1881 	getlckret = nfsrv_getlockfh(vp, new_stp->ls_flags, NULL, &nfh, p);
1882 	NFSLOCKSTATE();
1883 	if (getlckret == 0) {
1884 		if ((new_stp->ls_flags & (NFSLCK_LOCK | NFSLCK_UNLOCK)) != 0 &&
1885 		    nfsrv_dolocallocks != 0 && nd->nd_repstat == 0) {
1886 			getlckret = nfsrv_getlockfile(new_stp->ls_flags, NULL,
1887 			    &lfp, &nfh, 1);
1888 			if (getlckret == 0)
1889 				filestruct_locked = 1;
1890 		} else
1891 			getlckret = nfsrv_getlockfile(new_stp->ls_flags, NULL,
1892 			    &lfp, &nfh, 0);
1893 	}
1894 	if (getlckret != 0 && getlckret != -1)
1895 		reterr = getlckret;
1896 
1897 	if (filestruct_locked != 0) {
1898 		LIST_INIT(&lfp->lf_rollback);
1899 		if ((new_stp->ls_flags & NFSLCK_LOCK)) {
1900 			/*
1901 			 * For local locking, do the advisory locking now, so
1902 			 * that any conflict can be detected. A failure later
1903 			 * can be rolled back locally. If an error is returned,
1904 			 * struct nfslockfile has been unlocked and any local
1905 			 * locking rolled back.
1906 			 */
1907 			NFSUNLOCKSTATE();
1908 			if (vnode_unlocked == 0) {
1909 				ASSERT_VOP_ELOCKED(vp, "nfsrv_lockctrl1");
1910 				vnode_unlocked = 1;
1911 				NFSVOPUNLOCK(vp);
1912 			}
1913 			reterr = nfsrv_locallock(vp, lfp,
1914 			    (new_lop->lo_flags & (NFSLCK_READ | NFSLCK_WRITE)),
1915 			    new_lop->lo_first, new_lop->lo_end, cfp, p);
1916 			NFSLOCKSTATE();
1917 		}
1918 	}
1919 
1920 	if (specialid == 0) {
1921 	    if (new_stp->ls_flags & NFSLCK_TEST) {
1922 		/*
1923 		 * RFC 3530 does not list LockT as an op that renews a
1924 		 * lease, but the consensus seems to be that it is ok
1925 		 * for a server to do so.
1926 		 */
1927 		error = nfsrv_getclient(clientid, CLOPS_RENEW, &clp, NULL,
1928 		    (nfsquad_t)((u_quad_t)0), 0, nd, p);
1929 
1930 		/*
1931 		 * Since NFSERR_EXPIRED, NFSERR_ADMINREVOKED are not valid
1932 		 * error returns for LockT, just go ahead and test for a lock,
1933 		 * since there are no locks for this client, but other locks
1934 		 * can conflict. (ie. same client will always be false)
1935 		 */
1936 		if (error == NFSERR_EXPIRED || error == NFSERR_ADMINREVOKED)
1937 		    error = 0;
1938 		lckstp = new_stp;
1939 	    } else {
1940 	      error = nfsrv_getclient(clientid, CLOPS_RENEW, &clp, NULL,
1941 		(nfsquad_t)((u_quad_t)0), 0, nd, p);
1942 	      if (error == 0)
1943 		/*
1944 		 * Look up the stateid
1945 		 */
1946 		error = nfsrv_getstate(clp, &new_stp->ls_stateid,
1947 		  new_stp->ls_flags, &stp);
1948 	      /*
1949 	       * do some sanity checks for an unconfirmed open or a
1950 	       * stateid that refers to the wrong file, for an open stateid
1951 	       */
1952 	      if (error == 0 && (stp->ls_flags & NFSLCK_OPEN) &&
1953 		  ((stp->ls_openowner->ls_flags & NFSLCK_NEEDSCONFIRM) ||
1954 		   (getlckret == 0 && stp->ls_lfp != lfp))){
1955 		      /*
1956 		       * NFSLCK_SETATTR should return OK rather than NFSERR_BADSTATEID
1957 		       * The only exception is using SETATTR with SIZE.
1958 		       * */
1959                     if ((new_stp->ls_flags &
1960                          (NFSLCK_SETATTR | NFSLCK_CHECK)) != NFSLCK_SETATTR)
1961 			     error = NFSERR_BADSTATEID;
1962 	      }
1963 
1964 		if (error == 0 &&
1965 		  (stp->ls_flags & (NFSLCK_DELEGREAD | NFSLCK_DELEGWRITE)) &&
1966 		  getlckret == 0 && stp->ls_lfp != lfp)
1967 			error = NFSERR_BADSTATEID;
1968 
1969 	      /*
1970 	       * If the lockowner stateid doesn't refer to the same file,
1971 	       * I believe that is considered ok, since some clients will
1972 	       * only create a single lockowner and use that for all locks
1973 	       * on all files.
1974 	       * For now, log it as a diagnostic, instead of considering it
1975 	       * a BadStateid.
1976 	       */
1977 	      if (error == 0 && (stp->ls_flags &
1978 		  (NFSLCK_OPEN | NFSLCK_DELEGREAD | NFSLCK_DELEGWRITE)) == 0 &&
1979 		  getlckret == 0 && stp->ls_lfp != lfp) {
1980 #ifdef DIAGNOSTIC
1981 		  printf("Got a lock statid for different file open\n");
1982 #endif
1983 		  /*
1984 		  error = NFSERR_BADSTATEID;
1985 		  */
1986 	      }
1987 
1988 	      if (error == 0) {
1989 		    if (new_stp->ls_flags & NFSLCK_OPENTOLOCK) {
1990 			/*
1991 			 * If haslock set, we've already checked the seqid.
1992 			 */
1993 			if (!haslock) {
1994 			    if (stp->ls_flags & NFSLCK_OPEN)
1995 				error = nfsrv_checkseqid(nd, new_stp->ls_seq,
1996 				    stp->ls_openowner, new_stp->ls_op);
1997 			    else
1998 				error = NFSERR_BADSTATEID;
1999 			}
2000 			if (!error)
2001 			    nfsrv_getowner(&stp->ls_open, new_stp, &lckstp);
2002 			if (lckstp) {
2003 			    /*
2004 			     * For NFSv4.1 and NFSv4.2 allow an
2005 			     * open_to_lock_owner when the lock_owner already
2006 			     * exists.  Just clear NFSLCK_OPENTOLOCK so that
2007 			     * a new lock_owner will not be created.
2008 			     * RFC7530 states that the error for NFSv4.0
2009 			     * is NFS4ERR_BAD_SEQID.
2010 			     */
2011 			    if ((nd->nd_flag & ND_NFSV41) != 0)
2012 				new_stp->ls_flags &= ~NFSLCK_OPENTOLOCK;
2013 			    else
2014 				error = NFSERR_BADSEQID;
2015 			} else
2016 			    lckstp = new_stp;
2017 		    } else if (new_stp->ls_flags&(NFSLCK_LOCK|NFSLCK_UNLOCK)) {
2018 			/*
2019 			 * If haslock set, ditto above.
2020 			 */
2021 			if (!haslock) {
2022 			    if (stp->ls_flags & NFSLCK_OPEN)
2023 				error = NFSERR_BADSTATEID;
2024 			    else
2025 				error = nfsrv_checkseqid(nd, new_stp->ls_seq,
2026 				    stp, new_stp->ls_op);
2027 			}
2028 			lckstp = stp;
2029 		    } else {
2030 			lckstp = stp;
2031 		    }
2032 	      }
2033 	      /*
2034 	       * If the seqid part of the stateid isn't the same, return
2035 	       * NFSERR_OLDSTATEID for cases other than I/O Ops.
2036 	       * For I/O Ops, only return NFSERR_OLDSTATEID if
2037 	       * nfsrv_returnoldstateid is set. (The consensus on the email
2038 	       * list was that most clients would prefer to not receive
2039 	       * NFSERR_OLDSTATEID for I/O Ops, but the RFC suggests that that
2040 	       * is what will happen, so I use the nfsrv_returnoldstateid to
2041 	       * allow for either server configuration.)
2042 	       */
2043 	      if (!error && stp->ls_stateid.seqid!=new_stp->ls_stateid.seqid &&
2044 		  (((nd->nd_flag & ND_NFSV41) == 0 &&
2045 		   (!(new_stp->ls_flags & NFSLCK_CHECK) ||
2046 		    nfsrv_returnoldstateid)) ||
2047 		   ((nd->nd_flag & ND_NFSV41) != 0 &&
2048 		    new_stp->ls_stateid.seqid != 0)))
2049 		    error = NFSERR_OLDSTATEID;
2050 	    }
2051 	}
2052 
2053 	/*
2054 	 * Now we can check for grace.
2055 	 */
2056 	if (!error)
2057 		error = nfsrv_checkgrace(nd, clp, new_stp->ls_flags);
2058 	if ((new_stp->ls_flags & NFSLCK_RECLAIM) && !error &&
2059 		nfsrv_checkstable(clp))
2060 		error = NFSERR_NOGRACE;
2061 	/*
2062 	 * If we successfully Reclaimed state, note that.
2063 	 */
2064 	if ((new_stp->ls_flags & NFSLCK_RECLAIM) && !error)
2065 		nfsrv_markstable(clp);
2066 
2067 	/*
2068 	 * At this point, either error == NFSERR_BADSTATEID or the
2069 	 * seqid# has been updated, so we can return any error.
2070 	 * If error == 0, there may be an error in:
2071 	 *    nd_repstat - Set by the calling function.
2072 	 *    reterr - Set above, if getting the nfslockfile structure
2073 	 *       or acquiring the local lock failed.
2074 	 *    (If both of these are set, nd_repstat should probably be
2075 	 *     returned, since that error was detected before this
2076 	 *     function call.)
2077 	 */
2078 	if (error != 0 || nd->nd_repstat != 0 || reterr != 0) {
2079 		if (error == 0) {
2080 			if (nd->nd_repstat != 0)
2081 				error = nd->nd_repstat;
2082 			else
2083 				error = reterr;
2084 		}
2085 		if (filestruct_locked != 0) {
2086 			/* Roll back local locks. */
2087 			NFSUNLOCKSTATE();
2088 			if (vnode_unlocked == 0) {
2089 				ASSERT_VOP_ELOCKED(vp, "nfsrv_lockctrl2");
2090 				vnode_unlocked = 1;
2091 				NFSVOPUNLOCK(vp);
2092 			}
2093 			nfsrv_locallock_rollback(vp, lfp, p);
2094 			NFSLOCKSTATE();
2095 			nfsrv_unlocklf(lfp);
2096 		}
2097 		NFSUNLOCKSTATE();
2098 		goto out;
2099 	}
2100 
2101 	/*
2102 	 * Check the nfsrv_getlockfile return.
2103 	 * Returned -1 if no structure found.
2104 	 */
2105 	if (getlckret == -1) {
2106 		error = NFSERR_EXPIRED;
2107 		/*
2108 		 * Called from lockt, so no lock is OK.
2109 		 */
2110 		if (new_stp->ls_flags & NFSLCK_TEST) {
2111 			error = 0;
2112 		} else if (new_stp->ls_flags &
2113 		    (NFSLCK_CHECK | NFSLCK_SETATTR)) {
2114 			/*
2115 			 * Called to check for a lock, OK if the stateid is all
2116 			 * 1s or all 0s, but there should be an nfsstate
2117 			 * otherwise.
2118 			 * (ie. If there is no open, I'll assume no share
2119 			 *  deny bits.)
2120 			 */
2121 			if (specialid)
2122 				error = 0;
2123 			else
2124 				error = NFSERR_BADSTATEID;
2125 		}
2126 		NFSUNLOCKSTATE();
2127 		goto out;
2128 	}
2129 
2130 	/*
2131 	 * For NFSLCK_CHECK and NFSLCK_LOCK, test for a share conflict.
2132 	 * For NFSLCK_CHECK, allow a read if write access is granted,
2133 	 * but check for a deny. For NFSLCK_LOCK, require correct access,
2134 	 * which implies a conflicting deny can't exist.
2135 	 */
2136 	if (new_stp->ls_flags & (NFSLCK_CHECK | NFSLCK_LOCK)) {
2137 	    /*
2138 	     * Four kinds of state id:
2139 	     * - specialid (all 0s or all 1s), only for NFSLCK_CHECK
2140 	     * - stateid for an open
2141 	     * - stateid for a delegation
2142 	     * - stateid for a lock owner
2143 	     */
2144 	    if (!specialid) {
2145 		if (stp->ls_flags & (NFSLCK_DELEGREAD | NFSLCK_DELEGWRITE)) {
2146 		    delegation = 1;
2147 		    mystp = stp;
2148 		    nfsrv_delaydelegtimeout(stp);
2149 	        } else if (stp->ls_flags & NFSLCK_OPEN) {
2150 		    mystp = stp;
2151 		} else {
2152 		    mystp = stp->ls_openstp;
2153 		}
2154 		/*
2155 		 * If locking or checking, require correct access
2156 		 * bit set.
2157 		 */
2158 		if (((new_stp->ls_flags & NFSLCK_LOCK) &&
2159 		     !((new_lop->lo_flags >> NFSLCK_LOCKSHIFT) &
2160 		       mystp->ls_flags & NFSLCK_ACCESSBITS)) ||
2161 		    ((new_stp->ls_flags & (NFSLCK_CHECK|NFSLCK_READACCESS)) ==
2162 		      (NFSLCK_CHECK | NFSLCK_READACCESS) &&
2163 		     !(mystp->ls_flags & NFSLCK_READACCESS) &&
2164 		     nfsrv_allowreadforwriteopen == 0) ||
2165 		    ((new_stp->ls_flags & (NFSLCK_CHECK|NFSLCK_WRITEACCESS)) ==
2166 		      (NFSLCK_CHECK | NFSLCK_WRITEACCESS) &&
2167 		     !(mystp->ls_flags & NFSLCK_WRITEACCESS))) {
2168 			if (filestruct_locked != 0) {
2169 				/* Roll back local locks. */
2170 				NFSUNLOCKSTATE();
2171 				if (vnode_unlocked == 0) {
2172 					ASSERT_VOP_ELOCKED(vp,
2173 					    "nfsrv_lockctrl3");
2174 					vnode_unlocked = 1;
2175 					NFSVOPUNLOCK(vp);
2176 				}
2177 				nfsrv_locallock_rollback(vp, lfp, p);
2178 				NFSLOCKSTATE();
2179 				nfsrv_unlocklf(lfp);
2180 			}
2181 			NFSUNLOCKSTATE();
2182 			error = NFSERR_OPENMODE;
2183 			goto out;
2184 		}
2185 	    } else
2186 		mystp = NULL;
2187 	    if ((new_stp->ls_flags & NFSLCK_CHECK) && !delegation) {
2188 		/*
2189 		 * Check for a conflicting deny bit.
2190 		 */
2191 		LIST_FOREACH(tstp, &lfp->lf_open, ls_file) {
2192 		    if (tstp != mystp) {
2193 			bits = tstp->ls_flags;
2194 			bits >>= NFSLCK_SHIFT;
2195 			if (new_stp->ls_flags & bits & NFSLCK_ACCESSBITS) {
2196 			    KASSERT(vnode_unlocked == 0,
2197 				("nfsrv_lockctrl: vnode unlocked1"));
2198 			    ret = nfsrv_clientconflict(tstp->ls_clp, &haslock,
2199 				vp, p);
2200 			    if (ret == 1) {
2201 				/*
2202 				* nfsrv_clientconflict unlocks state
2203 				 * when it returns non-zero.
2204 				 */
2205 				lckstp = NULL;
2206 				goto tryagain;
2207 			    }
2208 			    if (ret == 0)
2209 				NFSUNLOCKSTATE();
2210 			    if (ret == 2)
2211 				error = NFSERR_PERM;
2212 			    else
2213 				error = NFSERR_OPENMODE;
2214 			    goto out;
2215 			}
2216 		    }
2217 		}
2218 
2219 		/* We're outta here */
2220 		NFSUNLOCKSTATE();
2221 		goto out;
2222 	    }
2223 	}
2224 
2225 	/*
2226 	 * For setattr, just get rid of all the Delegations for other clients.
2227 	 */
2228 	if (new_stp->ls_flags & NFSLCK_SETATTR) {
2229 		KASSERT(vnode_unlocked == 0,
2230 		    ("nfsrv_lockctrl: vnode unlocked2"));
2231 		ret = nfsrv_cleandeleg(vp, lfp, clp, &haslock, p);
2232 		if (ret) {
2233 			/*
2234 			 * nfsrv_cleandeleg() unlocks state when it
2235 			 * returns non-zero.
2236 			 */
2237 			if (ret == -1) {
2238 				lckstp = NULL;
2239 				goto tryagain;
2240 			}
2241 			error = ret;
2242 			goto out;
2243 		}
2244 		if (!(new_stp->ls_flags & NFSLCK_CHECK) ||
2245 		    (LIST_EMPTY(&lfp->lf_open) && LIST_EMPTY(&lfp->lf_lock) &&
2246 		     LIST_EMPTY(&lfp->lf_deleg))) {
2247 			NFSUNLOCKSTATE();
2248 			goto out;
2249 		}
2250 	}
2251 
2252 	/*
2253 	 * Check for a conflicting delegation. If one is found, call
2254 	 * nfsrv_delegconflict() to handle it. If the v4root lock hasn't
2255 	 * been set yet, it will get the lock. Otherwise, it will recall
2256 	 * the delegation. Then, we try try again...
2257 	 * I currently believe the conflict algorithm to be:
2258 	 * For Lock Ops (Lock/LockT/LockU)
2259 	 * - there is a conflict iff a different client has a write delegation
2260 	 * For Reading (Read Op)
2261 	 * - there is a conflict iff a different client has a write delegation
2262 	 *   (the specialids are always a different client)
2263 	 * For Writing (Write/Setattr of size)
2264 	 * - there is a conflict if a different client has any delegation
2265 	 * - there is a conflict if the same client has a read delegation
2266 	 *   (I don't understand why this isn't allowed, but that seems to be
2267 	 *    the current consensus?)
2268 	 */
2269 	tstp = LIST_FIRST(&lfp->lf_deleg);
2270 	while (tstp != LIST_END(&lfp->lf_deleg)) {
2271 	    nstp = LIST_NEXT(tstp, ls_file);
2272 	    if ((((new_stp->ls_flags&(NFSLCK_LOCK|NFSLCK_UNLOCK|NFSLCK_TEST))||
2273 		 ((new_stp->ls_flags & NFSLCK_CHECK) &&
2274 		  (new_lop->lo_flags & NFSLCK_READ))) &&
2275 		  clp != tstp->ls_clp &&
2276 		 (tstp->ls_flags & NFSLCK_DELEGWRITE)) ||
2277 		 ((new_stp->ls_flags & NFSLCK_CHECK) &&
2278 		   (new_lop->lo_flags & NFSLCK_WRITE) &&
2279 		  (clp != tstp->ls_clp ||
2280 		   (tstp->ls_flags & NFSLCK_DELEGREAD)))) {
2281 		ret = 0;
2282 		if (filestruct_locked != 0) {
2283 			/* Roll back local locks. */
2284 			NFSUNLOCKSTATE();
2285 			if (vnode_unlocked == 0) {
2286 				ASSERT_VOP_ELOCKED(vp, "nfsrv_lockctrl4");
2287 				NFSVOPUNLOCK(vp);
2288 			}
2289 			nfsrv_locallock_rollback(vp, lfp, p);
2290 			NFSLOCKSTATE();
2291 			nfsrv_unlocklf(lfp);
2292 			NFSUNLOCKSTATE();
2293 			NFSVOPLOCK(vp, LK_EXCLUSIVE | LK_RETRY);
2294 			vnode_unlocked = 0;
2295 			if (VN_IS_DOOMED(vp))
2296 				ret = NFSERR_SERVERFAULT;
2297 			NFSLOCKSTATE();
2298 		}
2299 		if (ret == 0)
2300 			ret = nfsrv_delegconflict(tstp, &haslock, p, vp);
2301 		if (ret) {
2302 		    /*
2303 		     * nfsrv_delegconflict unlocks state when it
2304 		     * returns non-zero, which it always does.
2305 		     */
2306 		    if (other_lop) {
2307 			free(other_lop, M_NFSDLOCK);
2308 			other_lop = NULL;
2309 		    }
2310 		    if (ret == -1) {
2311 			lckstp = NULL;
2312 			goto tryagain;
2313 		    }
2314 		    error = ret;
2315 		    goto out;
2316 		}
2317 		/* Never gets here. */
2318 	    }
2319 	    tstp = nstp;
2320 	}
2321 
2322 	/*
2323 	 * Handle the unlock case by calling nfsrv_updatelock().
2324 	 * (Should I have done some access checking above for unlock? For now,
2325 	 *  just let it happen.)
2326 	 */
2327 	if (new_stp->ls_flags & NFSLCK_UNLOCK) {
2328 		first = new_lop->lo_first;
2329 		end = new_lop->lo_end;
2330 		nfsrv_updatelock(stp, new_lopp, &other_lop, lfp);
2331 		stateidp->seqid = ++(stp->ls_stateid.seqid);
2332 		if ((nd->nd_flag & ND_NFSV41) != 0 && stateidp->seqid == 0)
2333 			stateidp->seqid = stp->ls_stateid.seqid = 1;
2334 		stateidp->other[0] = stp->ls_stateid.other[0];
2335 		stateidp->other[1] = stp->ls_stateid.other[1];
2336 		stateidp->other[2] = stp->ls_stateid.other[2];
2337 		if (filestruct_locked != 0) {
2338 			NFSUNLOCKSTATE();
2339 			if (vnode_unlocked == 0) {
2340 				ASSERT_VOP_ELOCKED(vp, "nfsrv_lockctrl5");
2341 				vnode_unlocked = 1;
2342 				NFSVOPUNLOCK(vp);
2343 			}
2344 			/* Update the local locks. */
2345 			nfsrv_localunlock(vp, lfp, first, end, p);
2346 			NFSLOCKSTATE();
2347 			nfsrv_unlocklf(lfp);
2348 		}
2349 		NFSUNLOCKSTATE();
2350 		goto out;
2351 	}
2352 
2353 	/*
2354 	 * Search for a conflicting lock. A lock conflicts if:
2355 	 * - the lock range overlaps and
2356 	 * - at least one lock is a write lock and
2357 	 * - it is not owned by the same lock owner
2358 	 */
2359 	if (!delegation) {
2360 	  LIST_FOREACH(lop, &lfp->lf_lock, lo_lckfile) {
2361 	    if (new_lop->lo_end > lop->lo_first &&
2362 		new_lop->lo_first < lop->lo_end &&
2363 		(new_lop->lo_flags == NFSLCK_WRITE ||
2364 		 lop->lo_flags == NFSLCK_WRITE) &&
2365 		lckstp != lop->lo_stp &&
2366 		(clp != lop->lo_stp->ls_clp ||
2367 		 lckstp->ls_ownerlen != lop->lo_stp->ls_ownerlen ||
2368 		 NFSBCMP(lckstp->ls_owner, lop->lo_stp->ls_owner,
2369 		    lckstp->ls_ownerlen))) {
2370 		if (other_lop) {
2371 		    free(other_lop, M_NFSDLOCK);
2372 		    other_lop = NULL;
2373 		}
2374 		if (vnode_unlocked != 0)
2375 		    ret = nfsrv_clientconflict(lop->lo_stp->ls_clp, &haslock,
2376 			NULL, p);
2377 		else
2378 		    ret = nfsrv_clientconflict(lop->lo_stp->ls_clp, &haslock,
2379 			vp, p);
2380 		if (ret == 1) {
2381 		    if (filestruct_locked != 0) {
2382 			if (vnode_unlocked == 0) {
2383 				ASSERT_VOP_ELOCKED(vp, "nfsrv_lockctrl6");
2384 				NFSVOPUNLOCK(vp);
2385 			}
2386 			/* Roll back local locks. */
2387 			nfsrv_locallock_rollback(vp, lfp, p);
2388 			NFSLOCKSTATE();
2389 			nfsrv_unlocklf(lfp);
2390 			NFSUNLOCKSTATE();
2391 			NFSVOPLOCK(vp, LK_EXCLUSIVE | LK_RETRY);
2392 			vnode_unlocked = 0;
2393 			if (VN_IS_DOOMED(vp)) {
2394 				error = NFSERR_SERVERFAULT;
2395 				goto out;
2396 			}
2397 		    }
2398 		    /*
2399 		     * nfsrv_clientconflict() unlocks state when it
2400 		     * returns non-zero.
2401 		     */
2402 		    lckstp = NULL;
2403 		    goto tryagain;
2404 		}
2405 		/*
2406 		 * Found a conflicting lock, so record the conflict and
2407 		 * return the error.
2408 		 */
2409 		if (cfp != NULL && ret == 0) {
2410 		    cfp->cl_clientid.lval[0]=lop->lo_stp->ls_stateid.other[0];
2411 		    cfp->cl_clientid.lval[1]=lop->lo_stp->ls_stateid.other[1];
2412 		    cfp->cl_first = lop->lo_first;
2413 		    cfp->cl_end = lop->lo_end;
2414 		    cfp->cl_flags = lop->lo_flags;
2415 		    cfp->cl_ownerlen = lop->lo_stp->ls_ownerlen;
2416 		    NFSBCOPY(lop->lo_stp->ls_owner, cfp->cl_owner,
2417 			cfp->cl_ownerlen);
2418 		}
2419 		if (ret == 2)
2420 		    error = NFSERR_PERM;
2421 		else if (new_stp->ls_flags & NFSLCK_RECLAIM)
2422 		    error = NFSERR_RECLAIMCONFLICT;
2423 		else if (new_stp->ls_flags & NFSLCK_CHECK)
2424 		    error = NFSERR_LOCKED;
2425 		else
2426 		    error = NFSERR_DENIED;
2427 		if (filestruct_locked != 0 && ret == 0) {
2428 			/* Roll back local locks. */
2429 			NFSUNLOCKSTATE();
2430 			if (vnode_unlocked == 0) {
2431 				ASSERT_VOP_ELOCKED(vp, "nfsrv_lockctrl7");
2432 				vnode_unlocked = 1;
2433 				NFSVOPUNLOCK(vp);
2434 			}
2435 			nfsrv_locallock_rollback(vp, lfp, p);
2436 			NFSLOCKSTATE();
2437 			nfsrv_unlocklf(lfp);
2438 		}
2439 		if (ret == 0)
2440 			NFSUNLOCKSTATE();
2441 		goto out;
2442 	    }
2443 	  }
2444 	}
2445 
2446 	/*
2447 	 * We only get here if there was no lock that conflicted.
2448 	 */
2449 	if (new_stp->ls_flags & (NFSLCK_TEST | NFSLCK_CHECK)) {
2450 		NFSUNLOCKSTATE();
2451 		goto out;
2452 	}
2453 
2454 	/*
2455 	 * We only get here when we are creating or modifying a lock.
2456 	 * There are two variants:
2457 	 * - exist_lock_owner where lock_owner exists
2458 	 * - open_to_lock_owner with new lock_owner
2459 	 */
2460 	first = new_lop->lo_first;
2461 	end = new_lop->lo_end;
2462 	lock_flags = new_lop->lo_flags;
2463 	if (!(new_stp->ls_flags & NFSLCK_OPENTOLOCK)) {
2464 		nfsrv_updatelock(lckstp, new_lopp, &other_lop, lfp);
2465 		stateidp->seqid = ++(lckstp->ls_stateid.seqid);
2466 		if ((nd->nd_flag & ND_NFSV41) != 0 && stateidp->seqid == 0)
2467 			stateidp->seqid = lckstp->ls_stateid.seqid = 1;
2468 		stateidp->other[0] = lckstp->ls_stateid.other[0];
2469 		stateidp->other[1] = lckstp->ls_stateid.other[1];
2470 		stateidp->other[2] = lckstp->ls_stateid.other[2];
2471 	} else {
2472 		/*
2473 		 * The new open_to_lock_owner case.
2474 		 * Link the new nfsstate into the lists.
2475 		 */
2476 		new_stp->ls_seq = new_stp->ls_opentolockseq;
2477 		nfsrvd_refcache(new_stp->ls_op);
2478 		stateidp->seqid = new_stp->ls_stateid.seqid = 1;
2479 		stateidp->other[0] = new_stp->ls_stateid.other[0] =
2480 		    clp->lc_clientid.lval[0];
2481 		stateidp->other[1] = new_stp->ls_stateid.other[1] =
2482 		    clp->lc_clientid.lval[1];
2483 		stateidp->other[2] = new_stp->ls_stateid.other[2] =
2484 		    nfsrv_nextstateindex(clp);
2485 		new_stp->ls_clp = clp;
2486 		LIST_INIT(&new_stp->ls_lock);
2487 		new_stp->ls_openstp = stp;
2488 		new_stp->ls_lfp = lfp;
2489 		nfsrv_insertlock(new_lop, (struct nfslock *)new_stp, new_stp,
2490 		    lfp);
2491 		LIST_INSERT_HEAD(NFSSTATEHASH(clp, new_stp->ls_stateid),
2492 		    new_stp, ls_hash);
2493 		LIST_INSERT_HEAD(&stp->ls_open, new_stp, ls_list);
2494 		*new_lopp = NULL;
2495 		*new_stpp = NULL;
2496 		NFSD_VNET(nfsstatsv1_p)->srvlockowners++;
2497 		nfsrv_openpluslock++;
2498 	}
2499 	if (filestruct_locked != 0) {
2500 		NFSUNLOCKSTATE();
2501 		nfsrv_locallock_commit(lfp, lock_flags, first, end);
2502 		NFSLOCKSTATE();
2503 		nfsrv_unlocklf(lfp);
2504 	}
2505 	NFSUNLOCKSTATE();
2506 
2507 out:
2508 	if (haslock) {
2509 		NFSLOCKV4ROOTMUTEX();
2510 		nfsv4_unlock(&nfsv4rootfs_lock, 1);
2511 		NFSUNLOCKV4ROOTMUTEX();
2512 	}
2513 	if (vnode_unlocked != 0) {
2514 		NFSVOPLOCK(vp, LK_EXCLUSIVE | LK_RETRY);
2515 		if (error == 0 && VN_IS_DOOMED(vp))
2516 			error = NFSERR_SERVERFAULT;
2517 	}
2518 	if (other_lop)
2519 		free(other_lop, M_NFSDLOCK);
2520 	NFSEXITCODE2(error, nd);
2521 	return (error);
2522 }
2523 
2524 /*
2525  * Check for state errors for Open.
2526  * repstat is passed back out as an error if more critical errors
2527  * are not detected.
2528  */
2529 int
2530 nfsrv_opencheck(nfsquad_t clientid, nfsv4stateid_t *stateidp,
2531     struct nfsstate *new_stp, vnode_t vp, struct nfsrv_descript *nd,
2532     NFSPROC_T *p, int repstat)
2533 {
2534 	struct nfsstate *stp, *nstp;
2535 	struct nfsclient *clp;
2536 	struct nfsstate *ownerstp;
2537 	struct nfslockfile *lfp, *new_lfp;
2538 	int error = 0, haslock = 0, ret, readonly = 0, getfhret = 0;
2539 
2540 	if ((new_stp->ls_flags & NFSLCK_SHAREBITS) == NFSLCK_READACCESS)
2541 		readonly = 1;
2542 	/*
2543 	 * Check for restart conditions (client and server).
2544 	 */
2545 	error = nfsrv_checkrestart(clientid, new_stp->ls_flags,
2546 		&new_stp->ls_stateid, 0);
2547 	if (error)
2548 		goto out;
2549 
2550 	/*
2551 	 * Check for state resource limit exceeded.
2552 	 * Technically this should be SMP protected, but the worst
2553 	 * case error is "out by one or two" on the count when it
2554 	 * returns NFSERR_RESOURCE and the limit is just a rather
2555 	 * arbitrary high water mark, so no harm is done.
2556 	 */
2557 	if (nfsrv_openpluslock > nfsrv_v4statelimit) {
2558 		error = NFSERR_RESOURCE;
2559 		goto out;
2560 	}
2561 
2562 tryagain:
2563 	new_lfp = malloc(sizeof (struct nfslockfile),
2564 	    M_NFSDLOCKFILE, M_WAITOK);
2565 	if (vp)
2566 		getfhret = nfsrv_getlockfh(vp, new_stp->ls_flags, new_lfp,
2567 		    NULL, p);
2568 	NFSLOCKSTATE();
2569 	/*
2570 	 * Get the nfsclient structure.
2571 	 */
2572 	error = nfsrv_getclient(clientid, CLOPS_RENEW, &clp, NULL,
2573 	    (nfsquad_t)((u_quad_t)0), 0, nd, p);
2574 
2575 	/*
2576 	 * Look up the open owner. See if it needs confirmation and
2577 	 * check the seq#, as required.
2578 	 */
2579 	if (!error)
2580 		nfsrv_getowner(&clp->lc_open, new_stp, &ownerstp);
2581 
2582 	if (!error && ownerstp) {
2583 		error = nfsrv_checkseqid(nd, new_stp->ls_seq, ownerstp,
2584 		    new_stp->ls_op);
2585 		/*
2586 		 * If the OpenOwner hasn't been confirmed, assume the
2587 		 * old one was a replay and this one is ok.
2588 		 * See: RFC3530 Sec. 14.2.18.
2589 		 */
2590 		if (error == NFSERR_BADSEQID &&
2591 		    (ownerstp->ls_flags & NFSLCK_NEEDSCONFIRM))
2592 			error = 0;
2593 	}
2594 
2595 	/*
2596 	 * Check for grace.
2597 	 */
2598 	if (!error)
2599 		error = nfsrv_checkgrace(nd, clp, new_stp->ls_flags);
2600 	if ((new_stp->ls_flags & NFSLCK_RECLAIM) && !error &&
2601 		nfsrv_checkstable(clp))
2602 		error = NFSERR_NOGRACE;
2603 
2604 	/*
2605 	 * If none of the above errors occurred, let repstat be
2606 	 * returned.
2607 	 */
2608 	if (repstat && !error)
2609 		error = repstat;
2610 	if (error) {
2611 		NFSUNLOCKSTATE();
2612 		if (haslock) {
2613 			NFSLOCKV4ROOTMUTEX();
2614 			nfsv4_unlock(&nfsv4rootfs_lock, 1);
2615 			NFSUNLOCKV4ROOTMUTEX();
2616 		}
2617 		free(new_lfp, M_NFSDLOCKFILE);
2618 		goto out;
2619 	}
2620 
2621 	/*
2622 	 * If vp == NULL, the file doesn't exist yet, so return ok.
2623 	 * (This always happens on the first pass, so haslock must be 0.)
2624 	 */
2625 	if (vp == NULL) {
2626 		NFSUNLOCKSTATE();
2627 		free(new_lfp, M_NFSDLOCKFILE);
2628 		goto out;
2629 	}
2630 
2631 	/*
2632 	 * Get the structure for the underlying file.
2633 	 */
2634 	if (getfhret)
2635 		error = getfhret;
2636 	else
2637 		error = nfsrv_getlockfile(new_stp->ls_flags, &new_lfp, &lfp,
2638 		    NULL, 0);
2639 	if (new_lfp)
2640 		free(new_lfp, M_NFSDLOCKFILE);
2641 	if (error) {
2642 		NFSUNLOCKSTATE();
2643 		if (haslock) {
2644 			NFSLOCKV4ROOTMUTEX();
2645 			nfsv4_unlock(&nfsv4rootfs_lock, 1);
2646 			NFSUNLOCKV4ROOTMUTEX();
2647 		}
2648 		goto out;
2649 	}
2650 
2651 	/*
2652 	 * Search for a conflicting open/share.
2653 	 */
2654 	if (new_stp->ls_flags & NFSLCK_DELEGCUR) {
2655 	    /*
2656 	     * For Delegate_Cur, search for the matching Delegation,
2657 	     * which indicates no conflict.
2658 	     * An old delegation should have been recovered by the
2659 	     * client doing a Claim_DELEGATE_Prev, so I won't let
2660 	     * it match and return NFSERR_EXPIRED. Should I let it
2661 	     * match?
2662 	     */
2663 	    LIST_FOREACH(stp, &lfp->lf_deleg, ls_file) {
2664 		if (!(stp->ls_flags & NFSLCK_OLDDELEG) &&
2665 		    (((nd->nd_flag & ND_NFSV41) != 0 &&
2666 		    stateidp->seqid == 0) ||
2667 		    stateidp->seqid == stp->ls_stateid.seqid) &&
2668 		    !NFSBCMP(stateidp->other, stp->ls_stateid.other,
2669 			  NFSX_STATEIDOTHER))
2670 			break;
2671 	    }
2672 	    if (stp == LIST_END(&lfp->lf_deleg) ||
2673 		((new_stp->ls_flags & NFSLCK_WRITEACCESS) &&
2674 		 (stp->ls_flags & NFSLCK_DELEGREAD))) {
2675 		NFSUNLOCKSTATE();
2676 		if (haslock) {
2677 			NFSLOCKV4ROOTMUTEX();
2678 			nfsv4_unlock(&nfsv4rootfs_lock, 1);
2679 			NFSUNLOCKV4ROOTMUTEX();
2680 		}
2681 		error = NFSERR_EXPIRED;
2682 		goto out;
2683 	    }
2684 	}
2685 
2686 	/*
2687 	 * Check for access/deny bit conflicts. I check for the same
2688 	 * owner as well, in case the client didn't bother.
2689 	 */
2690 	LIST_FOREACH(stp, &lfp->lf_open, ls_file) {
2691 		if (!(new_stp->ls_flags & NFSLCK_DELEGCUR) &&
2692 		    (((new_stp->ls_flags & NFSLCK_ACCESSBITS) &
2693 		      ((stp->ls_flags>>NFSLCK_SHIFT) & NFSLCK_ACCESSBITS))||
2694 		     ((stp->ls_flags & NFSLCK_ACCESSBITS) &
2695 		      ((new_stp->ls_flags>>NFSLCK_SHIFT)&NFSLCK_ACCESSBITS)))){
2696 			ret = nfsrv_clientconflict(stp->ls_clp,&haslock,vp,p);
2697 			if (ret == 1) {
2698 				/*
2699 				 * nfsrv_clientconflict() unlocks
2700 				 * state when it returns non-zero.
2701 				 */
2702 				goto tryagain;
2703 			}
2704 			if (ret == 2)
2705 				error = NFSERR_PERM;
2706 			else if (new_stp->ls_flags & NFSLCK_RECLAIM)
2707 				error = NFSERR_RECLAIMCONFLICT;
2708 			else
2709 				error = NFSERR_SHAREDENIED;
2710 			if (ret == 0)
2711 				NFSUNLOCKSTATE();
2712 			if (haslock) {
2713 				NFSLOCKV4ROOTMUTEX();
2714 				nfsv4_unlock(&nfsv4rootfs_lock, 1);
2715 				NFSUNLOCKV4ROOTMUTEX();
2716 			}
2717 			goto out;
2718 		}
2719 	}
2720 
2721 	/*
2722 	 * Check for a conflicting delegation. If one is found, call
2723 	 * nfsrv_delegconflict() to handle it. If the v4root lock hasn't
2724 	 * been set yet, it will get the lock. Otherwise, it will recall
2725 	 * the delegation. Then, we try try again...
2726 	 * (If NFSLCK_DELEGCUR is set, it has a delegation, so there
2727 	 *  isn't a conflict.)
2728 	 * I currently believe the conflict algorithm to be:
2729 	 * For Open with Read Access and Deny None
2730 	 * - there is a conflict iff a different client has a write delegation
2731 	 * For Open with other Write Access or any Deny except None
2732 	 * - there is a conflict if a different client has any delegation
2733 	 * - there is a conflict if the same client has a read delegation
2734 	 *   (The current consensus is that this last case should be
2735 	 *    considered a conflict since the client with a read delegation
2736 	 *    could have done an Open with ReadAccess and WriteDeny
2737 	 *    locally and then not have checked for the WriteDeny.)
2738 	 *    The exception is a NFSv4.1/4.2 client that has requested
2739 	 *    an atomic upgrade to a write delegation.
2740 	 * Don't check for a Reclaim, since that will be dealt with
2741 	 * by nfsrv_openctrl().
2742 	 */
2743 	if (!(new_stp->ls_flags &
2744 		(NFSLCK_DELEGPREV | NFSLCK_DELEGCUR | NFSLCK_RECLAIM))) {
2745 	    stp = LIST_FIRST(&lfp->lf_deleg);
2746 	    while (stp != LIST_END(&lfp->lf_deleg)) {
2747 		nstp = LIST_NEXT(stp, ls_file);
2748 		if ((readonly && stp->ls_clp != clp &&
2749 		     (stp->ls_flags & NFSLCK_DELEGWRITE) != 0) ||
2750 		    (!readonly && (stp->ls_clp != clp ||
2751 		     ((stp->ls_flags & NFSLCK_DELEGREAD) != 0 &&
2752 		      (new_stp->ls_flags & NFSLCK_WANTWDELEG) == 0)))) {
2753 			ret = nfsrv_delegconflict(stp, &haslock, p, vp);
2754 			if (ret) {
2755 			    /*
2756 			     * nfsrv_delegconflict() unlocks state
2757 			     * when it returns non-zero.
2758 			     */
2759 			    if (ret == -1)
2760 				goto tryagain;
2761 			    error = ret;
2762 			    goto out;
2763 			}
2764 		}
2765 		stp = nstp;
2766 	    }
2767 	}
2768 	NFSUNLOCKSTATE();
2769 	if (haslock) {
2770 		NFSLOCKV4ROOTMUTEX();
2771 		nfsv4_unlock(&nfsv4rootfs_lock, 1);
2772 		NFSUNLOCKV4ROOTMUTEX();
2773 	}
2774 
2775 out:
2776 	NFSEXITCODE2(error, nd);
2777 	return (error);
2778 }
2779 
2780 /*
2781  * Open control function to create/update open state for an open.
2782  */
2783 int
2784 nfsrv_openctrl(struct nfsrv_descript *nd, vnode_t vp,
2785     struct nfsstate **new_stpp, nfsquad_t clientid, nfsv4stateid_t *stateidp,
2786     nfsv4stateid_t *delegstateidp, u_int32_t *rflagsp, struct nfsexstuff *exp,
2787     NFSPROC_T *p, u_quad_t filerev)
2788 {
2789 	struct nfsstate *new_stp = *new_stpp;
2790 	struct nfsstate *stp, *nstp;
2791 	struct nfsstate *openstp = NULL, *new_open, *ownerstp, *new_deleg;
2792 	struct nfslockfile *lfp, *new_lfp;
2793 	struct nfsclient *clp;
2794 	int error = 0, haslock = 0, ret, delegate = 1, writedeleg = 1;
2795 	int readonly = 0, cbret = 1, getfhret = 0;
2796 	int gotstate = 0, len = 0;
2797 	u_char *clidp = NULL;
2798 
2799 	if ((new_stp->ls_flags & NFSLCK_SHAREBITS) == NFSLCK_READACCESS)
2800 		readonly = 1;
2801 	/*
2802 	 * Check for restart conditions (client and server).
2803 	 * (Paranoia, should have been detected by nfsrv_opencheck().)
2804 	 * If an error does show up, return NFSERR_EXPIRED, since the
2805 	 * the seqid# has already been incremented.
2806 	 */
2807 	error = nfsrv_checkrestart(clientid, new_stp->ls_flags,
2808 	    &new_stp->ls_stateid, 0);
2809 	if (error) {
2810 		printf("Nfsd: openctrl unexpected restart err=%d\n",
2811 		    error);
2812 		error = NFSERR_EXPIRED;
2813 		goto out;
2814 	}
2815 
2816 	clidp = malloc(NFSV4_OPAQUELIMIT, M_TEMP, M_WAITOK);
2817 tryagain:
2818 	new_lfp = malloc(sizeof (struct nfslockfile),
2819 	    M_NFSDLOCKFILE, M_WAITOK);
2820 	new_open = malloc(sizeof (struct nfsstate),
2821 	    M_NFSDSTATE, M_WAITOK);
2822 	new_deleg = malloc(sizeof (struct nfsstate),
2823 	    M_NFSDSTATE, M_WAITOK);
2824 	getfhret = nfsrv_getlockfh(vp, new_stp->ls_flags, new_lfp,
2825 	    NULL, p);
2826 	NFSLOCKSTATE();
2827 	/*
2828 	 * Get the client structure. Since the linked lists could be changed
2829 	 * by other nfsd processes if this process does a tsleep(), one of
2830 	 * two things must be done.
2831 	 * 1 - don't tsleep()
2832 	 * or
2833 	 * 2 - get the nfsv4_lock() { indicated by haslock == 1 }
2834 	 *     before using the lists, since this lock stops the other
2835 	 *     nfsd. This should only be used for rare cases, since it
2836 	 *     essentially single threads the nfsd.
2837 	 *     At this time, it is only done for cases where the stable
2838 	 *     storage file must be written prior to completion of state
2839 	 *     expiration.
2840 	 */
2841 	error = nfsrv_getclient(clientid, CLOPS_RENEW, &clp, NULL,
2842 	    (nfsquad_t)((u_quad_t)0), 0, nd, p);
2843 	if (!error && (clp->lc_flags & LCL_NEEDSCBNULL) &&
2844 	    clp->lc_program) {
2845 		/*
2846 		 * This happens on the first open for a client
2847 		 * that supports callbacks.
2848 		 */
2849 		NFSUNLOCKSTATE();
2850 		/*
2851 		 * Although nfsrv_docallback() will sleep, clp won't
2852 		 * go away, since they are only removed when the
2853 		 * nfsv4_lock() has blocked the nfsd threads. The
2854 		 * fields in clp can change, but having multiple
2855 		 * threads do this Null callback RPC should be
2856 		 * harmless.
2857 		 */
2858 		cbret = nfsrv_docallback(clp, NFSV4PROC_CBNULL,
2859 		    NULL, 0, NULL, NULL, NULL, 0, p);
2860 		NFSLOCKSTATE();
2861 		clp->lc_flags &= ~LCL_NEEDSCBNULL;
2862 		if (!cbret)
2863 			clp->lc_flags |= LCL_CALLBACKSON;
2864 	}
2865 
2866 	/*
2867 	 * Look up the open owner. See if it needs confirmation and
2868 	 * check the seq#, as required.
2869 	 */
2870 	if (!error)
2871 		nfsrv_getowner(&clp->lc_open, new_stp, &ownerstp);
2872 
2873 	if (error) {
2874 		NFSUNLOCKSTATE();
2875 		printf("Nfsd: openctrl unexpected state err=%d\n",
2876 			error);
2877 		free(new_lfp, M_NFSDLOCKFILE);
2878 		free(new_open, M_NFSDSTATE);
2879 		free(new_deleg, M_NFSDSTATE);
2880 		if (haslock) {
2881 			NFSLOCKV4ROOTMUTEX();
2882 			nfsv4_unlock(&nfsv4rootfs_lock, 1);
2883 			NFSUNLOCKV4ROOTMUTEX();
2884 		}
2885 		error = NFSERR_EXPIRED;
2886 		goto out;
2887 	}
2888 
2889 	if (new_stp->ls_flags & NFSLCK_RECLAIM)
2890 		nfsrv_markstable(clp);
2891 
2892 	/*
2893 	 * Get the structure for the underlying file.
2894 	 */
2895 	if (getfhret)
2896 		error = getfhret;
2897 	else
2898 		error = nfsrv_getlockfile(new_stp->ls_flags, &new_lfp, &lfp,
2899 		    NULL, 0);
2900 	if (new_lfp)
2901 		free(new_lfp, M_NFSDLOCKFILE);
2902 	if (error) {
2903 		NFSUNLOCKSTATE();
2904 		printf("Nfsd openctrl unexpected getlockfile err=%d\n",
2905 		    error);
2906 		free(new_open, M_NFSDSTATE);
2907 		free(new_deleg, M_NFSDSTATE);
2908 		if (haslock) {
2909 			NFSLOCKV4ROOTMUTEX();
2910 			nfsv4_unlock(&nfsv4rootfs_lock, 1);
2911 			NFSUNLOCKV4ROOTMUTEX();
2912 		}
2913 		goto out;
2914 	}
2915 
2916 	/*
2917 	 * Search for a conflicting open/share.
2918 	 */
2919 	if (new_stp->ls_flags & NFSLCK_DELEGCUR) {
2920 	    /*
2921 	     * For Delegate_Cur, search for the matching Delegation,
2922 	     * which indicates no conflict.
2923 	     * An old delegation should have been recovered by the
2924 	     * client doing a Claim_DELEGATE_Prev, so I won't let
2925 	     * it match and return NFSERR_EXPIRED. Should I let it
2926 	     * match?
2927 	     */
2928 	    LIST_FOREACH(stp, &lfp->lf_deleg, ls_file) {
2929 		if (!(stp->ls_flags & NFSLCK_OLDDELEG) &&
2930 		    (((nd->nd_flag & ND_NFSV41) != 0 &&
2931 		    stateidp->seqid == 0) ||
2932 		    stateidp->seqid == stp->ls_stateid.seqid) &&
2933 		    !NFSBCMP(stateidp->other, stp->ls_stateid.other,
2934 			NFSX_STATEIDOTHER))
2935 			break;
2936 	    }
2937 	    if (stp == LIST_END(&lfp->lf_deleg) ||
2938 		((new_stp->ls_flags & NFSLCK_WRITEACCESS) &&
2939 		 (stp->ls_flags & NFSLCK_DELEGREAD))) {
2940 		NFSUNLOCKSTATE();
2941 		printf("Nfsd openctrl unexpected expiry\n");
2942 		free(new_open, M_NFSDSTATE);
2943 		free(new_deleg, M_NFSDSTATE);
2944 		if (haslock) {
2945 			NFSLOCKV4ROOTMUTEX();
2946 			nfsv4_unlock(&nfsv4rootfs_lock, 1);
2947 			NFSUNLOCKV4ROOTMUTEX();
2948 		}
2949 		error = NFSERR_EXPIRED;
2950 		goto out;
2951 	    }
2952 
2953 	    /*
2954 	     * Don't issue a Delegation, since one already exists and
2955 	     * delay delegation timeout, as required.
2956 	     */
2957 	    delegate = 0;
2958 	    nfsrv_delaydelegtimeout(stp);
2959 	}
2960 
2961 	/*
2962 	 * Check for access/deny bit conflicts. I also check for the
2963 	 * same owner, since the client might not have bothered to check.
2964 	 * Also, note an open for the same file and owner, if found,
2965 	 * which is all we do here for Delegate_Cur, since conflict
2966 	 * checking is already done.
2967 	 */
2968 	LIST_FOREACH(stp, &lfp->lf_open, ls_file) {
2969 		if (ownerstp && stp->ls_openowner == ownerstp)
2970 			openstp = stp;
2971 		if (!(new_stp->ls_flags & NFSLCK_DELEGCUR)) {
2972 		    /*
2973 		     * If another client has the file open, the only
2974 		     * delegation that can be issued is a Read delegation
2975 		     * and only if it is a Read open with Deny none.
2976 		     */
2977 		    if (clp != stp->ls_clp) {
2978 			if ((stp->ls_flags & NFSLCK_SHAREBITS) ==
2979 			    NFSLCK_READACCESS)
2980 			    writedeleg = 0;
2981 			else
2982 			    delegate = 0;
2983 		    }
2984 		    if(((new_stp->ls_flags & NFSLCK_ACCESSBITS) &
2985 		        ((stp->ls_flags>>NFSLCK_SHIFT) & NFSLCK_ACCESSBITS))||
2986 		       ((stp->ls_flags & NFSLCK_ACCESSBITS) &
2987 		        ((new_stp->ls_flags>>NFSLCK_SHIFT)&NFSLCK_ACCESSBITS))){
2988 			ret = nfsrv_clientconflict(stp->ls_clp,&haslock,vp,p);
2989 			if (ret == 1) {
2990 				/*
2991 				 * nfsrv_clientconflict() unlocks state
2992 				 * when it returns non-zero.
2993 				 */
2994 				free(new_open, M_NFSDSTATE);
2995 				free(new_deleg, M_NFSDSTATE);
2996 				openstp = NULL;
2997 				goto tryagain;
2998 			}
2999 			if (ret == 2)
3000 				error = NFSERR_PERM;
3001 			else if (new_stp->ls_flags & NFSLCK_RECLAIM)
3002 				error = NFSERR_RECLAIMCONFLICT;
3003 			else
3004 				error = NFSERR_SHAREDENIED;
3005 			if (ret == 0)
3006 				NFSUNLOCKSTATE();
3007 			if (haslock) {
3008 				NFSLOCKV4ROOTMUTEX();
3009 				nfsv4_unlock(&nfsv4rootfs_lock, 1);
3010 				NFSUNLOCKV4ROOTMUTEX();
3011 			}
3012 			free(new_open, M_NFSDSTATE);
3013 			free(new_deleg, M_NFSDSTATE);
3014 			printf("nfsd openctrl unexpected client cnfl\n");
3015 			goto out;
3016 		    }
3017 		}
3018 	}
3019 
3020 	/*
3021 	 * Check for a conflicting delegation. If one is found, call
3022 	 * nfsrv_delegconflict() to handle it. If the v4root lock hasn't
3023 	 * been set yet, it will get the lock. Otherwise, it will recall
3024 	 * the delegation. Then, we try try again...
3025 	 * (If NFSLCK_DELEGCUR is set, it has a delegation, so there
3026 	 *  isn't a conflict.)
3027 	 * I currently believe the conflict algorithm to be:
3028 	 * For Open with Read Access and Deny None
3029 	 * - there is a conflict iff a different client has a write delegation
3030 	 * For Open with other Write Access or any Deny except None
3031 	 * - there is a conflict if a different client has any delegation
3032 	 * - there is a conflict if the same client has a read delegation
3033 	 *   (The current consensus is that this last case should be
3034 	 *    considered a conflict since the client with a read delegation
3035 	 *    could have done an Open with ReadAccess and WriteDeny
3036 	 *    locally and then not have checked for the WriteDeny.)
3037 	 *    The exception is a NFSv4.1/4.2 client that has requested
3038 	 *    an atomic upgrade to a write delegation.
3039 	 */
3040 	if (!(new_stp->ls_flags & (NFSLCK_DELEGPREV | NFSLCK_DELEGCUR))) {
3041 	    stp = LIST_FIRST(&lfp->lf_deleg);
3042 	    while (stp != LIST_END(&lfp->lf_deleg)) {
3043 		nstp = LIST_NEXT(stp, ls_file);
3044 		if (stp->ls_clp != clp && (stp->ls_flags & NFSLCK_DELEGREAD))
3045 			writedeleg = 0;
3046 		else if (stp->ls_clp != clp ||
3047 		    (stp->ls_flags & NFSLCK_DELEGWRITE) != 0 ||
3048 		    (new_stp->ls_flags & NFSLCK_WANTWDELEG) == 0)
3049 			delegate = 0;
3050 		if ((readonly && stp->ls_clp != clp &&
3051 		     (stp->ls_flags & NFSLCK_DELEGWRITE) != 0) ||
3052 		    (!readonly && (stp->ls_clp != clp ||
3053 		     ((stp->ls_flags & NFSLCK_DELEGREAD) != 0 &&
3054 		      (new_stp->ls_flags & NFSLCK_WANTWDELEG) == 0)))) {
3055 		    if (new_stp->ls_flags & NFSLCK_RECLAIM) {
3056 			delegate = 2;
3057 		    } else {
3058 			ret = nfsrv_delegconflict(stp, &haslock, p, vp);
3059 			if (ret) {
3060 			    /*
3061 			     * nfsrv_delegconflict() unlocks state
3062 			     * when it returns non-zero.
3063 			     */
3064 			    printf("Nfsd openctrl unexpected deleg cnfl\n");
3065 			    free(new_open, M_NFSDSTATE);
3066 			    free(new_deleg, M_NFSDSTATE);
3067 			    if (ret == -1) {
3068 				openstp = NULL;
3069 				goto tryagain;
3070 			    }
3071 			    error = ret;
3072 			    goto out;
3073 			}
3074 		    }
3075 		}
3076 		stp = nstp;
3077 	    }
3078 	}
3079 
3080 	/*
3081 	 * We only get here if there was no open that conflicted.
3082 	 * If an open for the owner exists, or in the access/deny bits.
3083 	 * Otherwise it is a new open. If the open_owner hasn't been
3084 	 * confirmed, replace the open with the new one needing confirmation,
3085 	 * otherwise add the open.
3086 	 */
3087 	if (new_stp->ls_flags & NFSLCK_DELEGPREV) {
3088 	    /*
3089 	     * Handle NFSLCK_DELEGPREV by searching the old delegations for
3090 	     * a match. If found, just move the old delegation to the current
3091 	     * delegation list and issue open. If not found, return
3092 	     * NFSERR_EXPIRED.
3093 	     */
3094 	    LIST_FOREACH(stp, &clp->lc_olddeleg, ls_list) {
3095 		if (stp->ls_lfp == lfp) {
3096 		    /* Found it */
3097 		    if (stp->ls_clp != clp)
3098 			panic("olddeleg clp");
3099 		    LIST_REMOVE(stp, ls_list);
3100 		    LIST_REMOVE(stp, ls_hash);
3101 		    stp->ls_flags &= ~NFSLCK_OLDDELEG;
3102 		    stp->ls_stateid.seqid = delegstateidp->seqid = 1;
3103 		    stp->ls_stateid.other[0] = delegstateidp->other[0] =
3104 			clp->lc_clientid.lval[0];
3105 		    stp->ls_stateid.other[1] = delegstateidp->other[1] =
3106 			clp->lc_clientid.lval[1];
3107 		    stp->ls_stateid.other[2] = delegstateidp->other[2] =
3108 			nfsrv_nextstateindex(clp);
3109 		    stp->ls_compref = nd->nd_compref;
3110 		    LIST_INSERT_HEAD(&clp->lc_deleg, stp, ls_list);
3111 		    LIST_INSERT_HEAD(NFSSTATEHASH(clp,
3112 			stp->ls_stateid), stp, ls_hash);
3113 		    if (stp->ls_flags & NFSLCK_DELEGWRITE)
3114 			*rflagsp |= NFSV4OPEN_WRITEDELEGATE;
3115 		    else
3116 			*rflagsp |= NFSV4OPEN_READDELEGATE;
3117 		    clp->lc_delegtime = NFSD_MONOSEC +
3118 			nfsrv_lease + NFSRV_LEASEDELTA;
3119 
3120 		    /*
3121 		     * Now, do the associated open.
3122 		     */
3123 		    new_open->ls_stateid.seqid = 1;
3124 		    new_open->ls_stateid.other[0] = clp->lc_clientid.lval[0];
3125 		    new_open->ls_stateid.other[1] = clp->lc_clientid.lval[1];
3126 		    new_open->ls_stateid.other[2] = nfsrv_nextstateindex(clp);
3127 		    new_open->ls_flags = (new_stp->ls_flags&NFSLCK_DENYBITS)|
3128 			NFSLCK_OPEN;
3129 		    if (stp->ls_flags & NFSLCK_DELEGWRITE)
3130 			new_open->ls_flags |= (NFSLCK_READACCESS |
3131 			    NFSLCK_WRITEACCESS);
3132 		    else
3133 			new_open->ls_flags |= NFSLCK_READACCESS;
3134 		    new_open->ls_uid = new_stp->ls_uid;
3135 		    new_open->ls_lfp = lfp;
3136 		    new_open->ls_clp = clp;
3137 		    LIST_INIT(&new_open->ls_open);
3138 		    LIST_INSERT_HEAD(&lfp->lf_open, new_open, ls_file);
3139 		    LIST_INSERT_HEAD(NFSSTATEHASH(clp, new_open->ls_stateid),
3140 			new_open, ls_hash);
3141 		    /*
3142 		     * and handle the open owner
3143 		     */
3144 		    if (ownerstp) {
3145 			new_open->ls_openowner = ownerstp;
3146 			LIST_INSERT_HEAD(&ownerstp->ls_open,new_open,ls_list);
3147 		    } else {
3148 			new_open->ls_openowner = new_stp;
3149 			new_stp->ls_flags = 0;
3150 			nfsrvd_refcache(new_stp->ls_op);
3151 			new_stp->ls_noopens = 0;
3152 			LIST_INIT(&new_stp->ls_open);
3153 			LIST_INSERT_HEAD(&new_stp->ls_open, new_open, ls_list);
3154 			LIST_INSERT_HEAD(&clp->lc_open, new_stp, ls_list);
3155 			*new_stpp = NULL;
3156 			NFSD_VNET(nfsstatsv1_p)->srvopenowners++;
3157 			nfsrv_openpluslock++;
3158 		    }
3159 		    openstp = new_open;
3160 		    new_open = NULL;
3161 		    NFSD_VNET(nfsstatsv1_p)->srvopens++;
3162 		    nfsrv_openpluslock++;
3163 		    break;
3164 		}
3165 	    }
3166 	    if (stp == LIST_END(&clp->lc_olddeleg))
3167 		error = NFSERR_EXPIRED;
3168 	} else if (new_stp->ls_flags & (NFSLCK_DELEGREAD | NFSLCK_DELEGWRITE)) {
3169 	    /*
3170 	     * Scan to see that no delegation for this client and file
3171 	     * doesn't already exist.
3172 	     * There also shouldn't yet be an Open for this file and
3173 	     * openowner.
3174 	     */
3175 	    LIST_FOREACH(stp, &lfp->lf_deleg, ls_file) {
3176 		if (stp->ls_clp == clp)
3177 		    break;
3178 	    }
3179 	    if (stp == LIST_END(&lfp->lf_deleg) && openstp == NULL) {
3180 		/*
3181 		 * This is the Claim_Previous case with a delegation
3182 		 * type != Delegate_None.
3183 		 */
3184 		/*
3185 		 * First, add the delegation. (Although we must issue the
3186 		 * delegation, we can also ask for an immediate return.)
3187 		 */
3188 		new_deleg->ls_stateid.seqid = delegstateidp->seqid = 1;
3189 		new_deleg->ls_stateid.other[0] = delegstateidp->other[0] =
3190 		    clp->lc_clientid.lval[0];
3191 		new_deleg->ls_stateid.other[1] = delegstateidp->other[1] =
3192 		    clp->lc_clientid.lval[1];
3193 		new_deleg->ls_stateid.other[2] = delegstateidp->other[2] =
3194 		    nfsrv_nextstateindex(clp);
3195 		if (new_stp->ls_flags & NFSLCK_DELEGWRITE) {
3196 		    new_deleg->ls_flags = (NFSLCK_DELEGWRITE |
3197 			NFSLCK_READACCESS | NFSLCK_WRITEACCESS);
3198 		    *rflagsp |= NFSV4OPEN_WRITEDELEGATE;
3199 		    nfsrv_writedelegcnt++;
3200 		} else {
3201 		    new_deleg->ls_flags = (NFSLCK_DELEGREAD |
3202 			NFSLCK_READACCESS);
3203 		    *rflagsp |= NFSV4OPEN_READDELEGATE;
3204 		}
3205 		new_deleg->ls_uid = new_stp->ls_uid;
3206 		new_deleg->ls_lfp = lfp;
3207 		new_deleg->ls_clp = clp;
3208 		new_deleg->ls_filerev = filerev;
3209 		new_deleg->ls_compref = nd->nd_compref;
3210 		new_deleg->ls_lastrecall = 0;
3211 		LIST_INSERT_HEAD(&lfp->lf_deleg, new_deleg, ls_file);
3212 		LIST_INSERT_HEAD(NFSSTATEHASH(clp,
3213 		    new_deleg->ls_stateid), new_deleg, ls_hash);
3214 		LIST_INSERT_HEAD(&clp->lc_deleg, new_deleg, ls_list);
3215 		new_deleg = NULL;
3216 		if (delegate == 2 || nfsrv_issuedelegs == 0 ||
3217 		    (clp->lc_flags & (LCL_CALLBACKSON | LCL_CBDOWN)) !=
3218 		     LCL_CALLBACKSON ||
3219 		    NFSRV_V4DELEGLIMIT(nfsrv_delegatecnt) ||
3220 		    !NFSVNO_DELEGOK(vp))
3221 		    *rflagsp |= NFSV4OPEN_RECALL;
3222 		NFSD_VNET(nfsstatsv1_p)->srvdelegates++;
3223 		nfsrv_openpluslock++;
3224 		nfsrv_delegatecnt++;
3225 
3226 		/*
3227 		 * Now, do the associated open.
3228 		 */
3229 		new_open->ls_stateid.seqid = 1;
3230 		new_open->ls_stateid.other[0] = clp->lc_clientid.lval[0];
3231 		new_open->ls_stateid.other[1] = clp->lc_clientid.lval[1];
3232 		new_open->ls_stateid.other[2] = nfsrv_nextstateindex(clp);
3233 		new_open->ls_flags = (new_stp->ls_flags & NFSLCK_DENYBITS) |
3234 		    NFSLCK_OPEN;
3235 		if (new_stp->ls_flags & NFSLCK_DELEGWRITE)
3236 			new_open->ls_flags |= (NFSLCK_READACCESS |
3237 			    NFSLCK_WRITEACCESS);
3238 		else
3239 			new_open->ls_flags |= NFSLCK_READACCESS;
3240 		new_open->ls_uid = new_stp->ls_uid;
3241 		new_open->ls_lfp = lfp;
3242 		new_open->ls_clp = clp;
3243 		LIST_INIT(&new_open->ls_open);
3244 		LIST_INSERT_HEAD(&lfp->lf_open, new_open, ls_file);
3245 		LIST_INSERT_HEAD(NFSSTATEHASH(clp, new_open->ls_stateid),
3246 		   new_open, ls_hash);
3247 		/*
3248 		 * and handle the open owner
3249 		 */
3250 		if (ownerstp) {
3251 		    new_open->ls_openowner = ownerstp;
3252 		    LIST_INSERT_HEAD(&ownerstp->ls_open, new_open, ls_list);
3253 		} else {
3254 		    new_open->ls_openowner = new_stp;
3255 		    new_stp->ls_flags = 0;
3256 		    nfsrvd_refcache(new_stp->ls_op);
3257 		    new_stp->ls_noopens = 0;
3258 		    LIST_INIT(&new_stp->ls_open);
3259 		    LIST_INSERT_HEAD(&new_stp->ls_open, new_open, ls_list);
3260 		    LIST_INSERT_HEAD(&clp->lc_open, new_stp, ls_list);
3261 		    *new_stpp = NULL;
3262 		    NFSD_VNET(nfsstatsv1_p)->srvopenowners++;
3263 		    nfsrv_openpluslock++;
3264 		}
3265 		openstp = new_open;
3266 		new_open = NULL;
3267 		NFSD_VNET(nfsstatsv1_p)->srvopens++;
3268 		nfsrv_openpluslock++;
3269 	    } else {
3270 		error = NFSERR_RECLAIMCONFLICT;
3271 	    }
3272 	} else if (ownerstp) {
3273 		if (ownerstp->ls_flags & NFSLCK_NEEDSCONFIRM) {
3274 		    /* Replace the open */
3275 		    if (ownerstp->ls_op)
3276 			nfsrvd_derefcache(ownerstp->ls_op);
3277 		    ownerstp->ls_op = new_stp->ls_op;
3278 		    nfsrvd_refcache(ownerstp->ls_op);
3279 		    ownerstp->ls_seq = new_stp->ls_seq;
3280 		    *rflagsp |= NFSV4OPEN_RESULTCONFIRM;
3281 		    stp = LIST_FIRST(&ownerstp->ls_open);
3282 		    stp->ls_flags = (new_stp->ls_flags & NFSLCK_SHAREBITS) |
3283 			NFSLCK_OPEN;
3284 		    stp->ls_stateid.seqid = 1;
3285 		    stp->ls_uid = new_stp->ls_uid;
3286 		    if (lfp != stp->ls_lfp) {
3287 			LIST_REMOVE(stp, ls_file);
3288 			LIST_INSERT_HEAD(&lfp->lf_open, stp, ls_file);
3289 			stp->ls_lfp = lfp;
3290 		    }
3291 		    openstp = stp;
3292 		} else if (openstp) {
3293 		    openstp->ls_flags |= (new_stp->ls_flags & NFSLCK_SHAREBITS);
3294 		    openstp->ls_stateid.seqid++;
3295 		    if ((nd->nd_flag & ND_NFSV41) != 0 &&
3296 			openstp->ls_stateid.seqid == 0)
3297 			openstp->ls_stateid.seqid = 1;
3298 
3299 		    /*
3300 		     * This is where we can choose to issue a delegation.
3301 		     */
3302 		    nfsrv_issuedelegation(vp, clp, nd, delegate, writedeleg,
3303 			readonly, filerev, NFSVNO_EXRDONLY(exp), &new_deleg,
3304 			new_stp, lfp, rflagsp, delegstateidp);
3305 		} else {
3306 		    new_open->ls_stateid.seqid = 1;
3307 		    new_open->ls_stateid.other[0] = clp->lc_clientid.lval[0];
3308 		    new_open->ls_stateid.other[1] = clp->lc_clientid.lval[1];
3309 		    new_open->ls_stateid.other[2] = nfsrv_nextstateindex(clp);
3310 		    new_open->ls_flags = (new_stp->ls_flags & NFSLCK_SHAREBITS)|
3311 			NFSLCK_OPEN;
3312 		    new_open->ls_uid = new_stp->ls_uid;
3313 		    new_open->ls_openowner = ownerstp;
3314 		    new_open->ls_lfp = lfp;
3315 		    new_open->ls_clp = clp;
3316 		    LIST_INIT(&new_open->ls_open);
3317 		    LIST_INSERT_HEAD(&lfp->lf_open, new_open, ls_file);
3318 		    LIST_INSERT_HEAD(&ownerstp->ls_open, new_open, ls_list);
3319 		    LIST_INSERT_HEAD(NFSSTATEHASH(clp, new_open->ls_stateid),
3320 			new_open, ls_hash);
3321 		    openstp = new_open;
3322 		    new_open = NULL;
3323 		    NFSD_VNET(nfsstatsv1_p)->srvopens++;
3324 		    nfsrv_openpluslock++;
3325 
3326 		    /*
3327 		     * This is where we can choose to issue a delegation.
3328 		     */
3329 		    nfsrv_issuedelegation(vp, clp, nd, delegate, writedeleg,
3330 			readonly, filerev, NFSVNO_EXRDONLY(exp), &new_deleg,
3331 			new_stp, lfp, rflagsp, delegstateidp);
3332 		}
3333 	} else {
3334 		/*
3335 		 * New owner case. Start the open_owner sequence with a
3336 		 * Needs confirmation (unless a reclaim) and hang the
3337 		 * new open off it.
3338 		 */
3339 		new_open->ls_stateid.seqid = 1;
3340 		new_open->ls_stateid.other[0] = clp->lc_clientid.lval[0];
3341 		new_open->ls_stateid.other[1] = clp->lc_clientid.lval[1];
3342 		new_open->ls_stateid.other[2] = nfsrv_nextstateindex(clp);
3343 		new_open->ls_flags = (new_stp->ls_flags & NFSLCK_SHAREBITS) |
3344 		    NFSLCK_OPEN;
3345 		new_open->ls_uid = new_stp->ls_uid;
3346 		LIST_INIT(&new_open->ls_open);
3347 		new_open->ls_openowner = new_stp;
3348 		new_open->ls_lfp = lfp;
3349 		new_open->ls_clp = clp;
3350 		LIST_INSERT_HEAD(&lfp->lf_open, new_open, ls_file);
3351 		if (new_stp->ls_flags & NFSLCK_RECLAIM) {
3352 			new_stp->ls_flags = 0;
3353 		} else if ((nd->nd_flag & ND_NFSV41) != 0) {
3354 		    /*
3355 		     * This is where we can choose to issue a delegation.
3356 		     */
3357 		    nfsrv_issuedelegation(vp, clp, nd, delegate, writedeleg,
3358 			readonly, filerev, NFSVNO_EXRDONLY(exp), &new_deleg,
3359 			new_stp, lfp, rflagsp, delegstateidp);
3360 		    /* NFSv4.1 never needs confirmation. */
3361 		    new_stp->ls_flags = 0;
3362 
3363 		    /*
3364 		     * Since NFSv4.1 never does an OpenConfirm, the first
3365 		     * open state will be acquired here.
3366 		     */
3367 		    if (!(clp->lc_flags & LCL_STAMPEDSTABLE)) {
3368 			clp->lc_flags |= LCL_STAMPEDSTABLE;
3369 			len = clp->lc_idlen;
3370 			NFSBCOPY(clp->lc_id, clidp, len);
3371 			gotstate = 1;
3372 		    }
3373 		} else {
3374 		    *rflagsp |= NFSV4OPEN_RESULTCONFIRM;
3375 		    new_stp->ls_flags = NFSLCK_NEEDSCONFIRM;
3376 		}
3377 		nfsrvd_refcache(new_stp->ls_op);
3378 		new_stp->ls_noopens = 0;
3379 		LIST_INIT(&new_stp->ls_open);
3380 		LIST_INSERT_HEAD(&new_stp->ls_open, new_open, ls_list);
3381 		LIST_INSERT_HEAD(&clp->lc_open, new_stp, ls_list);
3382 		LIST_INSERT_HEAD(NFSSTATEHASH(clp, new_open->ls_stateid),
3383 		    new_open, ls_hash);
3384 		openstp = new_open;
3385 		new_open = NULL;
3386 		*new_stpp = NULL;
3387 		NFSD_VNET(nfsstatsv1_p)->srvopens++;
3388 		nfsrv_openpluslock++;
3389 		NFSD_VNET(nfsstatsv1_p)->srvopenowners++;
3390 		nfsrv_openpluslock++;
3391 	}
3392 	if (!error) {
3393 		stateidp->seqid = openstp->ls_stateid.seqid;
3394 		stateidp->other[0] = openstp->ls_stateid.other[0];
3395 		stateidp->other[1] = openstp->ls_stateid.other[1];
3396 		stateidp->other[2] = openstp->ls_stateid.other[2];
3397 	}
3398 	NFSUNLOCKSTATE();
3399 	if (haslock) {
3400 		NFSLOCKV4ROOTMUTEX();
3401 		nfsv4_unlock(&nfsv4rootfs_lock, 1);
3402 		NFSUNLOCKV4ROOTMUTEX();
3403 	}
3404 	if (new_open)
3405 		free(new_open, M_NFSDSTATE);
3406 	if (new_deleg)
3407 		free(new_deleg, M_NFSDSTATE);
3408 
3409 	/*
3410 	 * If the NFSv4.1 client just acquired its first open, write a timestamp
3411 	 * to the stable storage file.
3412 	 */
3413 	if (gotstate != 0) {
3414 		nfsrv_writestable(clidp, len, NFSNST_NEWSTATE, p);
3415 		nfsrv_backupstable();
3416 	}
3417 
3418 out:
3419 	free(clidp, M_TEMP);
3420 	NFSEXITCODE2(error, nd);
3421 	return (error);
3422 }
3423 
3424 /*
3425  * Open update. Does the confirm, downgrade and close.
3426  */
3427 int
3428 nfsrv_openupdate(vnode_t vp, struct nfsstate *new_stp, nfsquad_t clientid,
3429     nfsv4stateid_t *stateidp, struct nfsrv_descript *nd, NFSPROC_T *p,
3430     int *retwriteaccessp)
3431 {
3432 	struct nfsstate *stp;
3433 	struct nfsclient *clp;
3434 	struct nfslockfile *lfp;
3435 	u_int32_t bits;
3436 	int error = 0, gotstate = 0, len = 0;
3437 	u_char *clidp = NULL;
3438 
3439 	/*
3440 	 * Check for restart conditions (client and server).
3441 	 */
3442 	error = nfsrv_checkrestart(clientid, new_stp->ls_flags,
3443 	    &new_stp->ls_stateid, 0);
3444 	if (error)
3445 		goto out;
3446 
3447 	clidp = malloc(NFSV4_OPAQUELIMIT, M_TEMP, M_WAITOK);
3448 	NFSLOCKSTATE();
3449 	/*
3450 	 * Get the open structure via clientid and stateid.
3451 	 */
3452 	error = nfsrv_getclient(clientid, CLOPS_RENEW, &clp, NULL,
3453 	    (nfsquad_t)((u_quad_t)0), 0, nd, p);
3454 	if (!error)
3455 		error = nfsrv_getstate(clp, &new_stp->ls_stateid,
3456 		    new_stp->ls_flags, &stp);
3457 
3458 	/*
3459 	 * Sanity check the open.
3460 	 */
3461 	if (!error && (!(stp->ls_flags & NFSLCK_OPEN) ||
3462 		(!(new_stp->ls_flags & NFSLCK_CONFIRM) &&
3463 		 (stp->ls_openowner->ls_flags & NFSLCK_NEEDSCONFIRM)) ||
3464 		((new_stp->ls_flags & NFSLCK_CONFIRM) &&
3465 		 (!(stp->ls_openowner->ls_flags & NFSLCK_NEEDSCONFIRM)))))
3466 		error = NFSERR_BADSTATEID;
3467 
3468 	if (!error)
3469 		error = nfsrv_checkseqid(nd, new_stp->ls_seq,
3470 		    stp->ls_openowner, new_stp->ls_op);
3471 	if (!error && stp->ls_stateid.seqid != new_stp->ls_stateid.seqid &&
3472 	    (((nd->nd_flag & ND_NFSV41) == 0 &&
3473 	      !(new_stp->ls_flags & NFSLCK_CONFIRM)) ||
3474 	     ((nd->nd_flag & ND_NFSV41) != 0 &&
3475 	      new_stp->ls_stateid.seqid != 0)))
3476 		error = NFSERR_OLDSTATEID;
3477 	if (!error && vp->v_type != VREG) {
3478 		if (vp->v_type == VDIR)
3479 			error = NFSERR_ISDIR;
3480 		else
3481 			error = NFSERR_INVAL;
3482 	}
3483 
3484 	if (error) {
3485 		/*
3486 		 * If a client tries to confirm an Open with a bad
3487 		 * seqid# and there are no byte range locks or other Opens
3488 		 * on the openowner, just throw it away, so the next use of the
3489 		 * openowner will start a fresh seq#.
3490 		 */
3491 		if (error == NFSERR_BADSEQID &&
3492 		    (new_stp->ls_flags & NFSLCK_CONFIRM) &&
3493 		    nfsrv_nootherstate(stp))
3494 			nfsrv_freeopenowner(stp->ls_openowner, 0, p);
3495 		NFSUNLOCKSTATE();
3496 		goto out;
3497 	}
3498 
3499 	/*
3500 	 * Set the return stateid.
3501 	 */
3502 	stateidp->seqid = stp->ls_stateid.seqid + 1;
3503 	if ((nd->nd_flag & ND_NFSV41) != 0 && stateidp->seqid == 0)
3504 		stateidp->seqid = 1;
3505 	stateidp->other[0] = stp->ls_stateid.other[0];
3506 	stateidp->other[1] = stp->ls_stateid.other[1];
3507 	stateidp->other[2] = stp->ls_stateid.other[2];
3508 	/*
3509 	 * Now, handle the three cases.
3510 	 */
3511 	if (new_stp->ls_flags & NFSLCK_CONFIRM) {
3512 		/*
3513 		 * If the open doesn't need confirmation, it seems to me that
3514 		 * there is a client error, but I'll just log it and keep going?
3515 		 */
3516 		if (!(stp->ls_openowner->ls_flags & NFSLCK_NEEDSCONFIRM))
3517 			printf("Nfsv4d: stray open confirm\n");
3518 		stp->ls_openowner->ls_flags = 0;
3519 		stp->ls_stateid.seqid++;
3520 		if ((nd->nd_flag & ND_NFSV41) != 0 &&
3521 		    stp->ls_stateid.seqid == 0)
3522 			stp->ls_stateid.seqid = 1;
3523 		if (!(clp->lc_flags & LCL_STAMPEDSTABLE)) {
3524 			clp->lc_flags |= LCL_STAMPEDSTABLE;
3525 			len = clp->lc_idlen;
3526 			NFSBCOPY(clp->lc_id, clidp, len);
3527 			gotstate = 1;
3528 		}
3529 		NFSUNLOCKSTATE();
3530 	} else if (new_stp->ls_flags & NFSLCK_CLOSE) {
3531 		lfp = stp->ls_lfp;
3532 		if (retwriteaccessp != NULL) {
3533 			if ((stp->ls_flags & NFSLCK_WRITEACCESS) != 0)
3534 				*retwriteaccessp = 1;
3535 			else
3536 				*retwriteaccessp = 0;
3537 		}
3538 		if (nfsrv_dolocallocks != 0 && !LIST_EMPTY(&stp->ls_open)) {
3539 			/* Get the lf lock */
3540 			nfsrv_locklf(lfp);
3541 			NFSUNLOCKSTATE();
3542 			ASSERT_VOP_ELOCKED(vp, "nfsrv_openupdate");
3543 			NFSVOPUNLOCK(vp);
3544 			if (nfsrv_freeopen(stp, vp, 1, p) == 0) {
3545 				NFSLOCKSTATE();
3546 				nfsrv_unlocklf(lfp);
3547 				NFSUNLOCKSTATE();
3548 			}
3549 			NFSVOPLOCK(vp, LK_EXCLUSIVE | LK_RETRY);
3550 		} else {
3551 			(void) nfsrv_freeopen(stp, NULL, 0, p);
3552 			NFSUNLOCKSTATE();
3553 		}
3554 	} else {
3555 		/*
3556 		 * Update the share bits, making sure that the new set are a
3557 		 * subset of the old ones.
3558 		 */
3559 		bits = (new_stp->ls_flags & NFSLCK_SHAREBITS);
3560 		if (~(stp->ls_flags) & bits) {
3561 			NFSUNLOCKSTATE();
3562 			error = NFSERR_INVAL;
3563 			goto out;
3564 		}
3565 		stp->ls_flags = (bits | NFSLCK_OPEN);
3566 		stp->ls_stateid.seqid++;
3567 		if ((nd->nd_flag & ND_NFSV41) != 0 &&
3568 		    stp->ls_stateid.seqid == 0)
3569 			stp->ls_stateid.seqid = 1;
3570 		NFSUNLOCKSTATE();
3571 	}
3572 
3573 	/*
3574 	 * If the client just confirmed its first open, write a timestamp
3575 	 * to the stable storage file.
3576 	 */
3577 	if (gotstate != 0) {
3578 		nfsrv_writestable(clidp, len, NFSNST_NEWSTATE, p);
3579 		nfsrv_backupstable();
3580 	}
3581 
3582 out:
3583 	free(clidp, M_TEMP);
3584 	NFSEXITCODE2(error, nd);
3585 	return (error);
3586 }
3587 
3588 /*
3589  * Delegation update. Does the purge and return.
3590  */
3591 int
3592 nfsrv_delegupdate(struct nfsrv_descript *nd, nfsquad_t clientid,
3593     nfsv4stateid_t *stateidp, vnode_t vp, int op, struct ucred *cred,
3594     NFSPROC_T *p, int *retwriteaccessp)
3595 {
3596 	struct nfsstate *stp;
3597 	struct nfsclient *clp;
3598 	int error = 0;
3599 	fhandle_t fh;
3600 
3601 	/*
3602 	 * Do a sanity check against the file handle for DelegReturn.
3603 	 */
3604 	if (vp) {
3605 		error = nfsvno_getfh(vp, &fh, p);
3606 		if (error)
3607 			goto out;
3608 	}
3609 	/*
3610 	 * Check for restart conditions (client and server).
3611 	 */
3612 	if (op == NFSV4OP_DELEGRETURN)
3613 		error = nfsrv_checkrestart(clientid, NFSLCK_DELEGRETURN,
3614 			stateidp, 0);
3615 	else
3616 		error = nfsrv_checkrestart(clientid, NFSLCK_DELEGPURGE,
3617 			stateidp, 0);
3618 
3619 	NFSLOCKSTATE();
3620 	/*
3621 	 * Get the open structure via clientid and stateid.
3622 	 */
3623 	if (!error)
3624 	    error = nfsrv_getclient(clientid, CLOPS_RENEW, &clp, NULL,
3625 		(nfsquad_t)((u_quad_t)0), 0, nd, p);
3626 	if (error) {
3627 		if (error == NFSERR_CBPATHDOWN)
3628 			error = 0;
3629 		if (error == NFSERR_STALECLIENTID && op == NFSV4OP_DELEGRETURN)
3630 			error = NFSERR_STALESTATEID;
3631 	}
3632 	if (!error && op == NFSV4OP_DELEGRETURN) {
3633 	    error = nfsrv_getstate(clp, stateidp, NFSLCK_DELEGRETURN, &stp);
3634 	    if (!error && stp->ls_stateid.seqid != stateidp->seqid &&
3635 		((nd->nd_flag & ND_NFSV41) == 0 || stateidp->seqid != 0))
3636 		error = NFSERR_OLDSTATEID;
3637 	}
3638 	/*
3639 	 * NFSERR_EXPIRED means that the state has gone away,
3640 	 * so Delegations have been purged. Just return ok.
3641 	 */
3642 	if (error == NFSERR_EXPIRED && op == NFSV4OP_DELEGPURGE) {
3643 		NFSUNLOCKSTATE();
3644 		error = 0;
3645 		goto out;
3646 	}
3647 	if (error) {
3648 		NFSUNLOCKSTATE();
3649 		goto out;
3650 	}
3651 
3652 	if (op == NFSV4OP_DELEGRETURN) {
3653 		if (NFSBCMP((caddr_t)&fh, (caddr_t)&stp->ls_lfp->lf_fh,
3654 		    sizeof (fhandle_t))) {
3655 			NFSUNLOCKSTATE();
3656 			error = NFSERR_BADSTATEID;
3657 			goto out;
3658 		}
3659 		if (retwriteaccessp != NULL) {
3660 			if ((stp->ls_flags & NFSLCK_DELEGWRITE) != 0)
3661 				*retwriteaccessp = 1;
3662 			else
3663 				*retwriteaccessp = 0;
3664 		}
3665 		nfsrv_freedeleg(stp);
3666 	} else {
3667 		nfsrv_freedeleglist(&clp->lc_olddeleg);
3668 	}
3669 	NFSUNLOCKSTATE();
3670 	error = 0;
3671 
3672 out:
3673 	NFSEXITCODE(error);
3674 	return (error);
3675 }
3676 
3677 /*
3678  * Release lock owner.
3679  */
3680 int
3681 nfsrv_releaselckown(struct nfsstate *new_stp, nfsquad_t clientid,
3682     NFSPROC_T *p)
3683 {
3684 	struct nfsstate *stp, *nstp, *openstp, *ownstp;
3685 	struct nfsclient *clp;
3686 	int error = 0;
3687 
3688 	/*
3689 	 * Check for restart conditions (client and server).
3690 	 */
3691 	error = nfsrv_checkrestart(clientid, new_stp->ls_flags,
3692 	    &new_stp->ls_stateid, 0);
3693 	if (error)
3694 		goto out;
3695 
3696 	NFSLOCKSTATE();
3697 	/*
3698 	 * Get the lock owner by name.
3699 	 */
3700 	error = nfsrv_getclient(clientid, CLOPS_RENEW, &clp, NULL,
3701 	    (nfsquad_t)((u_quad_t)0), 0, NULL, p);
3702 	if (error) {
3703 		NFSUNLOCKSTATE();
3704 		goto out;
3705 	}
3706 	LIST_FOREACH(ownstp, &clp->lc_open, ls_list) {
3707 	    LIST_FOREACH(openstp, &ownstp->ls_open, ls_list) {
3708 		stp = LIST_FIRST(&openstp->ls_open);
3709 		while (stp != LIST_END(&openstp->ls_open)) {
3710 		    nstp = LIST_NEXT(stp, ls_list);
3711 		    /*
3712 		     * If the owner matches, check for locks and
3713 		     * then free or return an error.
3714 		     */
3715 		    if (stp->ls_ownerlen == new_stp->ls_ownerlen &&
3716 			!NFSBCMP(stp->ls_owner, new_stp->ls_owner,
3717 			 stp->ls_ownerlen)){
3718 			if (LIST_EMPTY(&stp->ls_lock)) {
3719 			    nfsrv_freelockowner(stp, NULL, 0, p);
3720 			} else {
3721 			    NFSUNLOCKSTATE();
3722 			    error = NFSERR_LOCKSHELD;
3723 			    goto out;
3724 			}
3725 		    }
3726 		    stp = nstp;
3727 		}
3728 	    }
3729 	}
3730 	NFSUNLOCKSTATE();
3731 
3732 out:
3733 	NFSEXITCODE(error);
3734 	return (error);
3735 }
3736 
3737 /*
3738  * Get the file handle for a lock structure.
3739  */
3740 static int
3741 nfsrv_getlockfh(vnode_t vp, u_short flags, struct nfslockfile *new_lfp,
3742     fhandle_t *nfhp, NFSPROC_T *p)
3743 {
3744 	fhandle_t *fhp = NULL;
3745 	int error;
3746 
3747 	/*
3748 	 * For lock, use the new nfslock structure, otherwise just
3749 	 * a fhandle_t on the stack.
3750 	 */
3751 	if (flags & NFSLCK_OPEN) {
3752 		KASSERT(new_lfp != NULL, ("nfsrv_getlockfh: new_lfp NULL"));
3753 		fhp = &new_lfp->lf_fh;
3754 	} else if (nfhp) {
3755 		fhp = nfhp;
3756 	} else {
3757 		panic("nfsrv_getlockfh");
3758 	}
3759 	error = nfsvno_getfh(vp, fhp, p);
3760 	NFSEXITCODE(error);
3761 	return (error);
3762 }
3763 
3764 /*
3765  * Get an nfs lock structure. Allocate one, as required, and return a
3766  * pointer to it.
3767  * Returns an NFSERR_xxx upon failure or -1 to indicate no current lock.
3768  */
3769 static int
3770 nfsrv_getlockfile(u_short flags, struct nfslockfile **new_lfpp,
3771     struct nfslockfile **lfpp, fhandle_t *nfhp, int lockit)
3772 {
3773 	struct nfslockfile *lfp;
3774 	fhandle_t *fhp = NULL, *tfhp;
3775 	struct nfslockhashhead *hp;
3776 	struct nfslockfile *new_lfp = NULL;
3777 
3778 	/*
3779 	 * For lock, use the new nfslock structure, otherwise just
3780 	 * a fhandle_t on the stack.
3781 	 */
3782 	if (flags & NFSLCK_OPEN) {
3783 		new_lfp = *new_lfpp;
3784 		fhp = &new_lfp->lf_fh;
3785 	} else if (nfhp) {
3786 		fhp = nfhp;
3787 	} else {
3788 		panic("nfsrv_getlockfile");
3789 	}
3790 
3791 	hp = NFSLOCKHASH(fhp);
3792 	LIST_FOREACH(lfp, hp, lf_hash) {
3793 		tfhp = &lfp->lf_fh;
3794 		if (NFSVNO_CMPFH(fhp, tfhp)) {
3795 			if (lockit)
3796 				nfsrv_locklf(lfp);
3797 			*lfpp = lfp;
3798 			return (0);
3799 		}
3800 	}
3801 	if (!(flags & NFSLCK_OPEN))
3802 		return (-1);
3803 
3804 	/*
3805 	 * No match, so chain the new one into the list.
3806 	 */
3807 	LIST_INIT(&new_lfp->lf_open);
3808 	LIST_INIT(&new_lfp->lf_lock);
3809 	LIST_INIT(&new_lfp->lf_deleg);
3810 	LIST_INIT(&new_lfp->lf_locallock);
3811 	LIST_INIT(&new_lfp->lf_rollback);
3812 	new_lfp->lf_locallock_lck.nfslock_usecnt = 0;
3813 	new_lfp->lf_locallock_lck.nfslock_lock = 0;
3814 	new_lfp->lf_usecount = 0;
3815 	LIST_INSERT_HEAD(hp, new_lfp, lf_hash);
3816 	*lfpp = new_lfp;
3817 	*new_lfpp = NULL;
3818 	return (0);
3819 }
3820 
3821 /*
3822  * This function adds a nfslock lock structure to the list for the associated
3823  * nfsstate and nfslockfile structures. It will be inserted after the
3824  * entry pointed at by insert_lop.
3825  */
3826 static void
3827 nfsrv_insertlock(struct nfslock *new_lop, struct nfslock *insert_lop,
3828     struct nfsstate *stp, struct nfslockfile *lfp)
3829 {
3830 	struct nfslock *lop, *nlop;
3831 
3832 	new_lop->lo_stp = stp;
3833 	new_lop->lo_lfp = lfp;
3834 
3835 	if (stp != NULL) {
3836 		/* Insert in increasing lo_first order */
3837 		lop = LIST_FIRST(&lfp->lf_lock);
3838 		if (lop == LIST_END(&lfp->lf_lock) ||
3839 		    new_lop->lo_first <= lop->lo_first) {
3840 			LIST_INSERT_HEAD(&lfp->lf_lock, new_lop, lo_lckfile);
3841 		} else {
3842 			nlop = LIST_NEXT(lop, lo_lckfile);
3843 			while (nlop != LIST_END(&lfp->lf_lock) &&
3844 			       nlop->lo_first < new_lop->lo_first) {
3845 				lop = nlop;
3846 				nlop = LIST_NEXT(lop, lo_lckfile);
3847 			}
3848 			LIST_INSERT_AFTER(lop, new_lop, lo_lckfile);
3849 		}
3850 	} else {
3851 		new_lop->lo_lckfile.le_prev = NULL;	/* list not used */
3852 	}
3853 
3854 	/*
3855 	 * Insert after insert_lop, which is overloaded as stp or lfp for
3856 	 * an empty list.
3857 	 */
3858 	if (stp == NULL && (struct nfslockfile *)insert_lop == lfp)
3859 		LIST_INSERT_HEAD(&lfp->lf_locallock, new_lop, lo_lckowner);
3860 	else if ((struct nfsstate *)insert_lop == stp)
3861 		LIST_INSERT_HEAD(&stp->ls_lock, new_lop, lo_lckowner);
3862 	else
3863 		LIST_INSERT_AFTER(insert_lop, new_lop, lo_lckowner);
3864 	if (stp != NULL) {
3865 		NFSD_VNET(nfsstatsv1_p)->srvlocks++;
3866 		nfsrv_openpluslock++;
3867 	}
3868 }
3869 
3870 /*
3871  * This function updates the locking for a lock owner and given file. It
3872  * maintains a list of lock ranges ordered on increasing file offset that
3873  * are NFSLCK_READ or NFSLCK_WRITE and non-overlapping (aka POSIX style).
3874  * It always adds new_lop to the list and sometimes uses the one pointed
3875  * at by other_lopp.
3876  */
3877 static void
3878 nfsrv_updatelock(struct nfsstate *stp, struct nfslock **new_lopp,
3879     struct nfslock **other_lopp, struct nfslockfile *lfp)
3880 {
3881 	struct nfslock *new_lop = *new_lopp;
3882 	struct nfslock *lop, *tlop, *ilop;
3883 	struct nfslock *other_lop = *other_lopp;
3884 	int unlock = 0, myfile = 0;
3885 	u_int64_t tmp;
3886 
3887 	/*
3888 	 * Work down the list until the lock is merged.
3889 	 */
3890 	if (new_lop->lo_flags & NFSLCK_UNLOCK)
3891 		unlock = 1;
3892 	if (stp != NULL) {
3893 		ilop = (struct nfslock *)stp;
3894 		lop = LIST_FIRST(&stp->ls_lock);
3895 	} else {
3896 		ilop = (struct nfslock *)lfp;
3897 		lop = LIST_FIRST(&lfp->lf_locallock);
3898 	}
3899 	while (lop != NULL) {
3900 	    /*
3901 	     * Only check locks for this file that aren't before the start of
3902 	     * new lock's range.
3903 	     */
3904 	    if (lop->lo_lfp == lfp) {
3905 	      myfile = 1;
3906 	      if (lop->lo_end >= new_lop->lo_first) {
3907 		if (new_lop->lo_end < lop->lo_first) {
3908 			/*
3909 			 * If the new lock ends before the start of the
3910 			 * current lock's range, no merge, just insert
3911 			 * the new lock.
3912 			 */
3913 			break;
3914 		}
3915 		if (new_lop->lo_flags == lop->lo_flags ||
3916 		    (new_lop->lo_first <= lop->lo_first &&
3917 		     new_lop->lo_end >= lop->lo_end)) {
3918 			/*
3919 			 * This lock can be absorbed by the new lock/unlock.
3920 			 * This happens when it covers the entire range
3921 			 * of the old lock or is contiguous
3922 			 * with the old lock and is of the same type or an
3923 			 * unlock.
3924 			 */
3925 			if (lop->lo_first < new_lop->lo_first)
3926 				new_lop->lo_first = lop->lo_first;
3927 			if (lop->lo_end > new_lop->lo_end)
3928 				new_lop->lo_end = lop->lo_end;
3929 			tlop = lop;
3930 			lop = LIST_NEXT(lop, lo_lckowner);
3931 			nfsrv_freenfslock(tlop);
3932 			continue;
3933 		}
3934 
3935 		/*
3936 		 * All these cases are for contiguous locks that are not the
3937 		 * same type, so they can't be merged.
3938 		 */
3939 		if (new_lop->lo_first <= lop->lo_first) {
3940 			/*
3941 			 * This case is where the new lock overlaps with the
3942 			 * first part of the old lock. Move the start of the
3943 			 * old lock to just past the end of the new lock. The
3944 			 * new lock will be inserted in front of the old, since
3945 			 * ilop hasn't been updated. (We are done now.)
3946 			 */
3947 			lop->lo_first = new_lop->lo_end;
3948 			break;
3949 		}
3950 		if (new_lop->lo_end >= lop->lo_end) {
3951 			/*
3952 			 * This case is where the new lock overlaps with the
3953 			 * end of the old lock's range. Move the old lock's
3954 			 * end to just before the new lock's first and insert
3955 			 * the new lock after the old lock.
3956 			 * Might not be done yet, since the new lock could
3957 			 * overlap further locks with higher ranges.
3958 			 */
3959 			lop->lo_end = new_lop->lo_first;
3960 			ilop = lop;
3961 			lop = LIST_NEXT(lop, lo_lckowner);
3962 			continue;
3963 		}
3964 		/*
3965 		 * The final case is where the new lock's range is in the
3966 		 * middle of the current lock's and splits the current lock
3967 		 * up. Use *other_lopp to handle the second part of the
3968 		 * split old lock range. (We are done now.)
3969 		 * For unlock, we use new_lop as other_lop and tmp, since
3970 		 * other_lop and new_lop are the same for this case.
3971 		 * We noted the unlock case above, so we don't need
3972 		 * new_lop->lo_flags any longer.
3973 		 */
3974 		tmp = new_lop->lo_first;
3975 		if (other_lop == NULL) {
3976 			if (!unlock)
3977 				panic("nfsd srv update unlock");
3978 			other_lop = new_lop;
3979 			*new_lopp = NULL;
3980 		}
3981 		other_lop->lo_first = new_lop->lo_end;
3982 		other_lop->lo_end = lop->lo_end;
3983 		other_lop->lo_flags = lop->lo_flags;
3984 		other_lop->lo_stp = stp;
3985 		other_lop->lo_lfp = lfp;
3986 		lop->lo_end = tmp;
3987 		nfsrv_insertlock(other_lop, lop, stp, lfp);
3988 		*other_lopp = NULL;
3989 		ilop = lop;
3990 		break;
3991 	      }
3992 	    }
3993 	    ilop = lop;
3994 	    lop = LIST_NEXT(lop, lo_lckowner);
3995 	    if (myfile && (lop == NULL || lop->lo_lfp != lfp))
3996 		break;
3997 	}
3998 
3999 	/*
4000 	 * Insert the new lock in the list at the appropriate place.
4001 	 */
4002 	if (!unlock) {
4003 		nfsrv_insertlock(new_lop, ilop, stp, lfp);
4004 		*new_lopp = NULL;
4005 	}
4006 }
4007 
4008 /*
4009  * This function handles sequencing of locks, etc.
4010  * It returns an error that indicates what the caller should do.
4011  */
4012 static int
4013 nfsrv_checkseqid(struct nfsrv_descript *nd, u_int32_t seqid,
4014     struct nfsstate *stp, struct nfsrvcache *op)
4015 {
4016 	int error = 0;
4017 
4018 	if ((nd->nd_flag & ND_NFSV41) != 0)
4019 		/* NFSv4.1 ignores the open_seqid and lock_seqid. */
4020 		goto out;
4021 	if (op != nd->nd_rp)
4022 		panic("nfsrvstate checkseqid");
4023 	if (!(op->rc_flag & RC_INPROG))
4024 		panic("nfsrvstate not inprog");
4025 	if (stp->ls_op && stp->ls_op->rc_refcnt <= 0) {
4026 		printf("refcnt=%d\n", stp->ls_op->rc_refcnt);
4027 		panic("nfsrvstate op refcnt");
4028 	}
4029 
4030 	/* If ND_ERELOOKUP is set, the seqid has already been handled. */
4031 	if ((nd->nd_flag & ND_ERELOOKUP) != 0)
4032 		goto out;
4033 
4034 	if ((stp->ls_seq + 1) == seqid) {
4035 		if (stp->ls_op)
4036 			nfsrvd_derefcache(stp->ls_op);
4037 		stp->ls_op = op;
4038 		nfsrvd_refcache(op);
4039 		stp->ls_seq = seqid;
4040 		goto out;
4041 	} else if (stp->ls_seq == seqid && stp->ls_op &&
4042 		op->rc_xid == stp->ls_op->rc_xid &&
4043 		op->rc_refcnt == 0 &&
4044 		op->rc_reqlen == stp->ls_op->rc_reqlen &&
4045 		op->rc_cksum == stp->ls_op->rc_cksum) {
4046 		if (stp->ls_op->rc_flag & RC_INPROG) {
4047 			error = NFSERR_DONTREPLY;
4048 			goto out;
4049 		}
4050 		nd->nd_rp = stp->ls_op;
4051 		nd->nd_rp->rc_flag |= RC_INPROG;
4052 		nfsrvd_delcache(op);
4053 		error = NFSERR_REPLYFROMCACHE;
4054 		goto out;
4055 	}
4056 	error = NFSERR_BADSEQID;
4057 
4058 out:
4059 	NFSEXITCODE2(error, nd);
4060 	return (error);
4061 }
4062 
4063 /*
4064  * Get the client ip address for callbacks. If the strings can't be parsed,
4065  * just set lc_program to 0 to indicate no callbacks are possible.
4066  * (For cases where the address can't be parsed or is 0.0.0.0.0.0, set
4067  *  the address to the client's transport address. This won't be used
4068  *  for callbacks, but can be printed out by nfsstats for info.)
4069  * Return error if the xdr can't be parsed, 0 otherwise.
4070  */
4071 int
4072 nfsrv_getclientipaddr(struct nfsrv_descript *nd, struct nfsclient *clp)
4073 {
4074 	u_int32_t *tl;
4075 	u_char *cp, *cp2;
4076 	int i, j, maxalen = 0, minalen = 0;
4077 	sa_family_t af;
4078 #ifdef INET
4079 	struct sockaddr_in *rin = NULL, *sin;
4080 #endif
4081 #ifdef INET6
4082 	struct sockaddr_in6 *rin6 = NULL, *sin6;
4083 #endif
4084 	u_char *addr;
4085 	int error = 0, cantparse = 0;
4086 	union {
4087 		in_addr_t ival;
4088 		u_char cval[4];
4089 	} ip;
4090 	union {
4091 		in_port_t sval;
4092 		u_char cval[2];
4093 	} port;
4094 
4095 	/* 8 is the maximum length of the port# string. */
4096 	addr = malloc(INET6_ADDRSTRLEN + 8, M_TEMP, M_WAITOK);
4097 	clp->lc_req.nr_client = NULL;
4098 	clp->lc_req.nr_lock = 0;
4099 	af = AF_UNSPEC;
4100 	NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
4101 	i = fxdr_unsigned(int, *tl);
4102 	if (i >= 3 && i <= 4) {
4103 		error = nfsrv_mtostr(nd, addr, i);
4104 		if (error)
4105 			goto nfsmout;
4106 #ifdef INET
4107 		if (!strcmp(addr, "tcp")) {
4108 			clp->lc_flags |= LCL_TCPCALLBACK;
4109 			clp->lc_req.nr_sotype = SOCK_STREAM;
4110 			clp->lc_req.nr_soproto = IPPROTO_TCP;
4111 			af = AF_INET;
4112 		} else if (!strcmp(addr, "udp")) {
4113 			clp->lc_req.nr_sotype = SOCK_DGRAM;
4114 			clp->lc_req.nr_soproto = IPPROTO_UDP;
4115 			af = AF_INET;
4116 		}
4117 #endif
4118 #ifdef INET6
4119 		if (af == AF_UNSPEC) {
4120 			if (!strcmp(addr, "tcp6")) {
4121 				clp->lc_flags |= LCL_TCPCALLBACK;
4122 				clp->lc_req.nr_sotype = SOCK_STREAM;
4123 				clp->lc_req.nr_soproto = IPPROTO_TCP;
4124 				af = AF_INET6;
4125 			} else if (!strcmp(addr, "udp6")) {
4126 				clp->lc_req.nr_sotype = SOCK_DGRAM;
4127 				clp->lc_req.nr_soproto = IPPROTO_UDP;
4128 				af = AF_INET6;
4129 			}
4130 		}
4131 #endif
4132 		if (af == AF_UNSPEC) {
4133 			cantparse = 1;
4134 		}
4135 	} else {
4136 		cantparse = 1;
4137 		if (i > 0) {
4138 			error = nfsm_advance(nd, NFSM_RNDUP(i), -1);
4139 			if (error)
4140 				goto nfsmout;
4141 		}
4142 	}
4143 	/*
4144 	 * The caller has allocated clp->lc_req.nr_nam to be large enough
4145 	 * for either AF_INET or AF_INET6 and zeroed out the contents.
4146 	 * maxalen is set to the maximum length of the host IP address string
4147 	 * plus 8 for the maximum length of the port#.
4148 	 * minalen is set to the minimum length of the host IP address string
4149 	 * plus 4 for the minimum length of the port#.
4150 	 * These lengths do not include NULL termination,
4151 	 * so INET[6]_ADDRSTRLEN - 1 is used in the calculations.
4152 	 */
4153 	switch (af) {
4154 #ifdef INET
4155 	case AF_INET:
4156 		rin = (struct sockaddr_in *)clp->lc_req.nr_nam;
4157 		rin->sin_family = AF_INET;
4158 		rin->sin_len = sizeof(struct sockaddr_in);
4159 		maxalen = INET_ADDRSTRLEN - 1 + 8;
4160 		minalen = 7 + 4;
4161 		break;
4162 #endif
4163 #ifdef INET6
4164 	case AF_INET6:
4165 		rin6 = (struct sockaddr_in6 *)clp->lc_req.nr_nam;
4166 		rin6->sin6_family = AF_INET6;
4167 		rin6->sin6_len = sizeof(struct sockaddr_in6);
4168 		maxalen = INET6_ADDRSTRLEN - 1 + 8;
4169 		minalen = 3 + 4;
4170 		break;
4171 #endif
4172 	}
4173 	NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
4174 	i = fxdr_unsigned(int, *tl);
4175 	if (i < 0) {
4176 		error = NFSERR_BADXDR;
4177 		goto nfsmout;
4178 	} else if (i == 0) {
4179 		cantparse = 1;
4180 	} else if (!cantparse && i <= maxalen && i >= minalen) {
4181 		error = nfsrv_mtostr(nd, addr, i);
4182 		if (error)
4183 			goto nfsmout;
4184 
4185 		/*
4186 		 * Parse out the address fields. We expect 6 decimal numbers
4187 		 * separated by '.'s for AF_INET and two decimal numbers
4188 		 * preceeded by '.'s for AF_INET6.
4189 		 */
4190 		cp = NULL;
4191 		switch (af) {
4192 #ifdef INET6
4193 		/*
4194 		 * For AF_INET6, first parse the host address.
4195 		 */
4196 		case AF_INET6:
4197 			cp = strchr(addr, '.');
4198 			if (cp != NULL) {
4199 				*cp++ = '\0';
4200 				if (inet_pton(af, addr, &rin6->sin6_addr) == 1)
4201 					i = 4;
4202 				else {
4203 					cp = NULL;
4204 					cantparse = 1;
4205 				}
4206 			}
4207 			break;
4208 #endif
4209 #ifdef INET
4210 		case AF_INET:
4211 			cp = addr;
4212 			i = 0;
4213 			break;
4214 #endif
4215 		}
4216 		while (cp != NULL && *cp && i < 6) {
4217 			cp2 = cp;
4218 			while (*cp2 && *cp2 != '.')
4219 				cp2++;
4220 			if (*cp2)
4221 				*cp2++ = '\0';
4222 			else if (i != 5) {
4223 				cantparse = 1;
4224 				break;
4225 			}
4226 			j = nfsrv_getipnumber(cp);
4227 			if (j >= 0) {
4228 				if (i < 4)
4229 					ip.cval[3 - i] = j;
4230 				else
4231 					port.cval[5 - i] = j;
4232 			} else {
4233 				cantparse = 1;
4234 				break;
4235 			}
4236 			cp = cp2;
4237 			i++;
4238 		}
4239 		if (!cantparse) {
4240 			/*
4241 			 * The host address INADDR_ANY is (mis)used to indicate
4242 			 * "there is no valid callback address".
4243 			 */
4244 			switch (af) {
4245 #ifdef INET6
4246 			case AF_INET6:
4247 				if (!IN6_ARE_ADDR_EQUAL(&rin6->sin6_addr,
4248 				    &in6addr_any))
4249 					rin6->sin6_port = htons(port.sval);
4250 				else
4251 					cantparse = 1;
4252 				break;
4253 #endif
4254 #ifdef INET
4255 			case AF_INET:
4256 				if (ip.ival != INADDR_ANY) {
4257 					rin->sin_addr.s_addr = htonl(ip.ival);
4258 					rin->sin_port = htons(port.sval);
4259 				} else {
4260 					cantparse = 1;
4261 				}
4262 				break;
4263 #endif
4264 			}
4265 		}
4266 	} else {
4267 		cantparse = 1;
4268 		if (i > 0) {
4269 			error = nfsm_advance(nd, NFSM_RNDUP(i), -1);
4270 			if (error)
4271 				goto nfsmout;
4272 		}
4273 	}
4274 	if (cantparse) {
4275 		switch (nd->nd_nam->sa_family) {
4276 #ifdef INET
4277 		case AF_INET:
4278 			sin = (struct sockaddr_in *)nd->nd_nam;
4279 			rin = (struct sockaddr_in *)clp->lc_req.nr_nam;
4280 			rin->sin_family = AF_INET;
4281 			rin->sin_len = sizeof(struct sockaddr_in);
4282 			rin->sin_addr.s_addr = sin->sin_addr.s_addr;
4283 			rin->sin_port = 0x0;
4284 			break;
4285 #endif
4286 #ifdef INET6
4287 		case AF_INET6:
4288 			sin6 = (struct sockaddr_in6 *)nd->nd_nam;
4289 			rin6 = (struct sockaddr_in6 *)clp->lc_req.nr_nam;
4290 			rin6->sin6_family = AF_INET6;
4291 			rin6->sin6_len = sizeof(struct sockaddr_in6);
4292 			rin6->sin6_addr = sin6->sin6_addr;
4293 			rin6->sin6_port = 0x0;
4294 			break;
4295 #endif
4296 		}
4297 		clp->lc_program = 0;
4298 	}
4299 nfsmout:
4300 	free(addr, M_TEMP);
4301 	NFSEXITCODE2(error, nd);
4302 	return (error);
4303 }
4304 
4305 /*
4306  * Turn a string of up to three decimal digits into a number. Return -1 upon
4307  * error.
4308  */
4309 static int
4310 nfsrv_getipnumber(u_char *cp)
4311 {
4312 	int i = 0, j = 0;
4313 
4314 	while (*cp) {
4315 		if (j > 2 || *cp < '0' || *cp > '9')
4316 			return (-1);
4317 		i *= 10;
4318 		i += (*cp - '0');
4319 		cp++;
4320 		j++;
4321 	}
4322 	if (i < 256)
4323 		return (i);
4324 	return (-1);
4325 }
4326 
4327 /*
4328  * This function checks for restart conditions.
4329  */
4330 static int
4331 nfsrv_checkrestart(nfsquad_t clientid, u_int32_t flags,
4332     nfsv4stateid_t *stateidp, int specialid)
4333 {
4334 	int ret = 0;
4335 
4336 	/*
4337 	 * First check for a server restart. Open, LockT, ReleaseLockOwner
4338 	 * and DelegPurge have a clientid, the rest a stateid.
4339 	 */
4340 	if (flags &
4341 	    (NFSLCK_OPEN | NFSLCK_TEST | NFSLCK_RELEASE | NFSLCK_DELEGPURGE)) {
4342 		if (clientid.lval[0] != NFSD_VNET(nfsrvboottime)) {
4343 			ret = NFSERR_STALECLIENTID;
4344 			goto out;
4345 		}
4346 	} else if (stateidp->other[0] != NFSD_VNET(nfsrvboottime) &&
4347 		specialid == 0) {
4348 		ret = NFSERR_STALESTATEID;
4349 		goto out;
4350 	}
4351 
4352 	/*
4353 	 * Read, Write, Setattr and LockT can return NFSERR_GRACE and do
4354 	 * not use a lock/open owner seqid#, so the check can be done now.
4355 	 * (The others will be checked, as required, later.)
4356 	 */
4357 	if (!(flags & (NFSLCK_CHECK | NFSLCK_TEST)))
4358 		goto out;
4359 
4360 	NFSLOCKSTATE();
4361 	ret = nfsrv_checkgrace(NULL, NULL, flags);
4362 	NFSUNLOCKSTATE();
4363 
4364 out:
4365 	NFSEXITCODE(ret);
4366 	return (ret);
4367 }
4368 
4369 /*
4370  * Check for grace.
4371  */
4372 static int
4373 nfsrv_checkgrace(struct nfsrv_descript *nd, struct nfsclient *clp,
4374     u_int32_t flags)
4375 {
4376 	int error = 0, notreclaimed;
4377 	struct nfsrv_stable *sp;
4378 
4379 	if ((NFSD_VNET(nfsrv_stablefirst).nsf_flags & (NFSNSF_UPDATEDONE |
4380 	     NFSNSF_GRACEOVER)) == 0) {
4381 		/*
4382 		 * First, check to see if all of the clients have done a
4383 		 * ReclaimComplete.  If so, grace can end now.
4384 		 */
4385 		notreclaimed = 0;
4386 		LIST_FOREACH(sp, &NFSD_VNET(nfsrv_stablefirst).nsf_head,
4387 		    nst_list) {
4388 			if ((sp->nst_flag & NFSNST_RECLAIMED) == 0) {
4389 				notreclaimed = 1;
4390 				break;
4391 			}
4392 		}
4393 		if (notreclaimed == 0)
4394 			NFSD_VNET(nfsrv_stablefirst).nsf_flags |=
4395 			    (NFSNSF_GRACEOVER | NFSNSF_NEEDLOCK);
4396 	}
4397 
4398 	if ((NFSD_VNET(nfsrv_stablefirst).nsf_flags & NFSNSF_GRACEOVER) != 0) {
4399 		if (flags & NFSLCK_RECLAIM) {
4400 			error = NFSERR_NOGRACE;
4401 			goto out;
4402 		}
4403 	} else {
4404 		if (!(flags & NFSLCK_RECLAIM)) {
4405 			error = NFSERR_GRACE;
4406 			goto out;
4407 		}
4408 		if (nd != NULL && clp != NULL &&
4409 		    (nd->nd_flag & ND_NFSV41) != 0 &&
4410 		    (clp->lc_flags & LCL_RECLAIMCOMPLETE) != 0) {
4411 			error = NFSERR_NOGRACE;
4412 			goto out;
4413 		}
4414 
4415 		/*
4416 		 * If grace is almost over and we are still getting Reclaims,
4417 		 * extend grace a bit.
4418 		 */
4419 		if ((NFSD_MONOSEC + NFSRV_LEASEDELTA) >
4420 		    NFSD_VNET(nfsrv_stablefirst).nsf_eograce)
4421 			NFSD_VNET(nfsrv_stablefirst).nsf_eograce =
4422 				NFSD_MONOSEC + NFSRV_LEASEDELTA;
4423 	}
4424 
4425 out:
4426 	NFSEXITCODE(error);
4427 	return (error);
4428 }
4429 
4430 /*
4431  * Do a server callback.
4432  * The "trunc" argument is slightly overloaded and refers to different
4433  * boolean arguments for CBRECALL and CBLAYOUTRECALL.
4434  */
4435 static int
4436 nfsrv_docallback(struct nfsclient *clp, int procnum, nfsv4stateid_t *stateidp,
4437     int trunc, fhandle_t *fhp, struct nfsvattr *nap, nfsattrbit_t *attrbitp,
4438     int laytype, NFSPROC_T *p)
4439 {
4440 	struct mbuf *m;
4441 	u_int32_t *tl;
4442 	struct nfsrv_descript *nd;
4443 	struct ucred *cred;
4444 	int error = 0, slotpos;
4445 	u_int32_t callback;
4446 	struct nfsdsession *sep = NULL;
4447 	uint64_t tval;
4448 	bool dotls;
4449 
4450 	nd = malloc(sizeof(*nd), M_TEMP, M_WAITOK | M_ZERO);
4451 	cred = newnfs_getcred();
4452 	NFSLOCKSTATE();	/* mostly for lc_cbref++ */
4453 	if (clp->lc_flags & LCL_NEEDSCONFIRM) {
4454 		NFSUNLOCKSTATE();
4455 		panic("docallb");
4456 	}
4457 	clp->lc_cbref++;
4458 
4459 	/*
4460 	 * Fill the callback program# and version into the request
4461 	 * structure for newnfs_connect() to use.
4462 	 */
4463 	clp->lc_req.nr_prog = clp->lc_program;
4464 #ifdef notnow
4465 	if ((clp->lc_flags & LCL_NFSV41) != 0)
4466 		clp->lc_req.nr_vers = NFSV41_CBVERS;
4467 	else
4468 #endif
4469 		clp->lc_req.nr_vers = NFSV4_CBVERS;
4470 
4471 	/*
4472 	 * First, fill in some of the fields of nd and cr.
4473 	 */
4474 	nd->nd_flag = ND_NFSV4;
4475 	if (clp->lc_flags & LCL_GSS)
4476 		nd->nd_flag |= ND_KERBV;
4477 	if ((clp->lc_flags & LCL_NFSV41) != 0)
4478 		nd->nd_flag |= ND_NFSV41;
4479 	if ((clp->lc_flags & LCL_NFSV42) != 0)
4480 		nd->nd_flag |= ND_NFSV42;
4481 	nd->nd_repstat = 0;
4482 	cred->cr_uid = clp->lc_uid;
4483 	cred->cr_gid = clp->lc_gid;
4484 	callback = clp->lc_callback;
4485 	NFSUNLOCKSTATE();
4486 	cred->cr_ngroups = 1;
4487 
4488 	/*
4489 	 * Get the first mbuf for the request.
4490 	 */
4491 	MGET(m, M_WAITOK, MT_DATA);
4492 	m->m_len = 0;
4493 	nd->nd_mreq = nd->nd_mb = m;
4494 	nd->nd_bpos = mtod(m, caddr_t);
4495 
4496 	/*
4497 	 * and build the callback request.
4498 	 */
4499 	if (procnum == NFSV4OP_CBGETATTR) {
4500 		nd->nd_procnum = NFSV4PROC_CBCOMPOUND;
4501 		error = nfsrv_cbcallargs(nd, clp, callback, NFSV4OP_CBGETATTR,
4502 		    "CB Getattr", &sep, &slotpos);
4503 		if (error != 0) {
4504 			m_freem(nd->nd_mreq);
4505 			goto errout;
4506 		}
4507 		(void)nfsm_fhtom(NULL, nd, (u_int8_t *)fhp, NFSX_MYFH, 0);
4508 		(void)nfsrv_putattrbit(nd, attrbitp);
4509 	} else if (procnum == NFSV4OP_CBRECALL) {
4510 		nd->nd_procnum = NFSV4PROC_CBCOMPOUND;
4511 		error = nfsrv_cbcallargs(nd, clp, callback, NFSV4OP_CBRECALL,
4512 		    "CB Recall", &sep, &slotpos);
4513 		if (error != 0) {
4514 			m_freem(nd->nd_mreq);
4515 			goto errout;
4516 		}
4517 		NFSM_BUILD(tl, u_int32_t *, NFSX_UNSIGNED + NFSX_STATEID);
4518 		*tl++ = txdr_unsigned(stateidp->seqid);
4519 		NFSBCOPY((caddr_t)stateidp->other, (caddr_t)tl,
4520 		    NFSX_STATEIDOTHER);
4521 		tl += (NFSX_STATEIDOTHER / NFSX_UNSIGNED);
4522 		if (trunc)
4523 			*tl = newnfs_true;
4524 		else
4525 			*tl = newnfs_false;
4526 		(void)nfsm_fhtom(NULL, nd, (u_int8_t *)fhp, NFSX_MYFH, 0);
4527 	} else if (procnum == NFSV4OP_CBLAYOUTRECALL) {
4528 		NFSD_DEBUG(4, "docallback layout recall\n");
4529 		nd->nd_procnum = NFSV4PROC_CBCOMPOUND;
4530 		error = nfsrv_cbcallargs(nd, clp, callback,
4531 		    NFSV4OP_CBLAYOUTRECALL, "CB Reclayout", &sep, &slotpos);
4532 		NFSD_DEBUG(4, "aft cbcallargs=%d\n", error);
4533 		if (error != 0) {
4534 			m_freem(nd->nd_mreq);
4535 			goto errout;
4536 		}
4537 		NFSM_BUILD(tl, u_int32_t *, 4 * NFSX_UNSIGNED);
4538 		*tl++ = txdr_unsigned(laytype);
4539 		*tl++ = txdr_unsigned(NFSLAYOUTIOMODE_ANY);
4540 		if (trunc)
4541 			*tl++ = newnfs_true;
4542 		else
4543 			*tl++ = newnfs_false;
4544 		*tl = txdr_unsigned(NFSV4LAYOUTRET_FILE);
4545 		(void)nfsm_fhtom(NULL, nd, (uint8_t *)fhp, NFSX_MYFH, 0);
4546 		NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_HYPER + NFSX_STATEID);
4547 		tval = 0;
4548 		txdr_hyper(tval, tl); tl += 2;
4549 		tval = UINT64_MAX;
4550 		txdr_hyper(tval, tl); tl += 2;
4551 		*tl++ = txdr_unsigned(stateidp->seqid);
4552 		NFSBCOPY(stateidp->other, tl, NFSX_STATEIDOTHER);
4553 		tl += (NFSX_STATEIDOTHER / NFSX_UNSIGNED);
4554 		NFSD_DEBUG(4, "aft args\n");
4555 	} else if (procnum == NFSV4PROC_CBNULL) {
4556 		nd->nd_procnum = NFSV4PROC_CBNULL;
4557 		if ((clp->lc_flags & LCL_NFSV41) != 0) {
4558 			error = nfsv4_getcbsession(clp, &sep);
4559 			if (error != 0) {
4560 				m_freem(nd->nd_mreq);
4561 				goto errout;
4562 			}
4563 		}
4564 	} else {
4565 		error = NFSERR_SERVERFAULT;
4566 		m_freem(nd->nd_mreq);
4567 		goto errout;
4568 	}
4569 
4570 	/*
4571 	 * Call newnfs_connect(), as required, and then newnfs_request().
4572 	 */
4573 	dotls = false;
4574 	if ((clp->lc_flags & LCL_TLSCB) != 0)
4575 		dotls = true;
4576 	(void) newnfs_sndlock(&clp->lc_req.nr_lock);
4577 	if (clp->lc_req.nr_client == NULL) {
4578 		if ((clp->lc_flags & LCL_NFSV41) != 0) {
4579 			error = ECONNREFUSED;
4580 			if (procnum != NFSV4PROC_CBNULL)
4581 				nfsv4_freeslot(&sep->sess_cbsess, slotpos,
4582 				    true);
4583 			nfsrv_freesession(NULL, sep, NULL, false, NULL);
4584 		} else if (nd->nd_procnum == NFSV4PROC_CBNULL)
4585 			error = newnfs_connect(NULL, &clp->lc_req, cred,
4586 			    NULL, 1, dotls, &clp->lc_req.nr_client);
4587 		else
4588 			error = newnfs_connect(NULL, &clp->lc_req, cred,
4589 			    NULL, 3, dotls, &clp->lc_req.nr_client);
4590 	}
4591 	newnfs_sndunlock(&clp->lc_req.nr_lock);
4592 	NFSD_DEBUG(4, "aft sndunlock=%d\n", error);
4593 	if (!error) {
4594 		if ((nd->nd_flag & ND_NFSV41) != 0) {
4595 			KASSERT(sep != NULL, ("sep NULL"));
4596 			if (sep->sess_cbsess.nfsess_xprt != NULL)
4597 				error = newnfs_request(nd, NULL, clp,
4598 				    &clp->lc_req, NULL, NULL, cred,
4599 				    clp->lc_program, clp->lc_req.nr_vers, NULL,
4600 				    1, NULL, &sep->sess_cbsess);
4601 			else {
4602 				/*
4603 				 * This should probably never occur, but if a
4604 				 * client somehow does an RPC without a
4605 				 * SequenceID Op that causes a callback just
4606 				 * after the nfsd threads have been terminated
4607 				 * and restarted we could conceivably get here
4608 				 * without a backchannel xprt.
4609 				 */
4610 				printf("nfsrv_docallback: no xprt\n");
4611 				error = ECONNREFUSED;
4612 			}
4613 			NFSD_DEBUG(4, "aft newnfs_request=%d\n", error);
4614 			if (error != 0 && procnum != NFSV4PROC_CBNULL) {
4615 				/*
4616 				 * It is likely that the callback was never
4617 				 * processed by the client and, as such,
4618 				 * the sequence# for the session slot needs
4619 				 * to be backed up by one to avoid a
4620 				 * NFSERR_SEQMISORDERED error reply.
4621 				 * For the unlikely case where the callback
4622 				 * was processed by the client, this will
4623 				 * make the next callback on the slot
4624 				 * appear to be a retry.
4625 				 * Since callbacks never specify that the
4626 				 * reply be cached, this "apparent retry"
4627 				 * should not be a problem.
4628 				 */
4629 				nfsv4_freeslot(&sep->sess_cbsess, slotpos,
4630 				    true);
4631 			}
4632 			nfsrv_freesession(NULL, sep, NULL, false, NULL);
4633 		} else
4634 			error = newnfs_request(nd, NULL, clp, &clp->lc_req,
4635 			    NULL, NULL, cred, clp->lc_program,
4636 			    clp->lc_req.nr_vers, NULL, 1, NULL, NULL);
4637 	}
4638 errout:
4639 	NFSFREECRED(cred);
4640 
4641 	/*
4642 	 * If error is set here, the Callback path isn't working
4643 	 * properly, so twiddle the appropriate LCL_ flags.
4644 	 * (nd_repstat != 0 indicates the Callback path is working,
4645 	 *  but the callback failed on the client.)
4646 	 */
4647 	if (error) {
4648 		/*
4649 		 * Mark the callback pathway down, which disabled issuing
4650 		 * of delegations and gets Renew to return NFSERR_CBPATHDOWN.
4651 		 */
4652 		NFSLOCKSTATE();
4653 		clp->lc_flags |= LCL_CBDOWN;
4654 		NFSUNLOCKSTATE();
4655 	} else {
4656 		/*
4657 		 * Callback worked. If the callback path was down, disable
4658 		 * callbacks, so no more delegations will be issued. (This
4659 		 * is done on the assumption that the callback pathway is
4660 		 * flakey.)
4661 		 */
4662 		NFSLOCKSTATE();
4663 		if (clp->lc_flags & LCL_CBDOWN)
4664 			clp->lc_flags &= ~(LCL_CBDOWN | LCL_CALLBACKSON);
4665 		NFSUNLOCKSTATE();
4666 		if (nd->nd_repstat) {
4667 			error = nd->nd_repstat;
4668 			NFSD_DEBUG(1, "nfsrv_docallback op=%d err=%d\n",
4669 			    procnum, error);
4670 		} else if (error == 0 && procnum == NFSV4OP_CBGETATTR)
4671 			error = nfsv4_loadattr(nd, NULL, nap, NULL, NULL, 0,
4672 			    NULL, NULL, NULL, NULL, NULL, 0, NULL, NULL, NULL,
4673 			    p, NULL);
4674 		m_freem(nd->nd_mrep);
4675 	}
4676 	NFSLOCKSTATE();
4677 	clp->lc_cbref--;
4678 	if ((clp->lc_flags & LCL_WAKEUPWANTED) && clp->lc_cbref == 0) {
4679 		clp->lc_flags &= ~LCL_WAKEUPWANTED;
4680 		wakeup(clp);
4681 	}
4682 	NFSUNLOCKSTATE();
4683 
4684 	free(nd, M_TEMP);
4685 	NFSEXITCODE(error);
4686 	return (error);
4687 }
4688 
4689 /*
4690  * Set up the compound RPC for the callback.
4691  */
4692 static int
4693 nfsrv_cbcallargs(struct nfsrv_descript *nd, struct nfsclient *clp,
4694     uint32_t callback, int op, const char *optag, struct nfsdsession **sepp,
4695     int *slotposp)
4696 {
4697 	uint32_t *tl;
4698 	int error, len;
4699 
4700 	len = strlen(optag);
4701 	(void)nfsm_strtom(nd, optag, len);
4702 	NFSM_BUILD(tl, uint32_t *, 4 * NFSX_UNSIGNED);
4703 	if ((nd->nd_flag & ND_NFSV41) != 0) {
4704 		if ((nd->nd_flag & ND_NFSV42) != 0)
4705 			*tl++ = txdr_unsigned(NFSV42_MINORVERSION);
4706 		else
4707 			*tl++ = txdr_unsigned(NFSV41_MINORVERSION);
4708 		*tl++ = txdr_unsigned(callback);
4709 		*tl++ = txdr_unsigned(2);
4710 		*tl = txdr_unsigned(NFSV4OP_CBSEQUENCE);
4711 		error = nfsv4_setcbsequence(nd, clp, 1, sepp, slotposp);
4712 		if (error != 0)
4713 			return (error);
4714 		NFSM_BUILD(tl, u_int32_t *, NFSX_UNSIGNED);
4715 		*tl = txdr_unsigned(op);
4716 	} else {
4717 		*tl++ = txdr_unsigned(NFSV4_MINORVERSION);
4718 		*tl++ = txdr_unsigned(callback);
4719 		*tl++ = txdr_unsigned(1);
4720 		*tl = txdr_unsigned(op);
4721 	}
4722 	return (0);
4723 }
4724 
4725 /*
4726  * Return the next index# for a clientid. Mostly just increment and return
4727  * the next one, but... if the 32bit unsigned does actually wrap around,
4728  * it should be rebooted.
4729  * At an average rate of one new client per second, it will wrap around in
4730  * approximately 136 years. (I think the server will have been shut
4731  * down or rebooted before then.)
4732  */
4733 static u_int32_t
4734 nfsrv_nextclientindex(void)
4735 {
4736 	static u_int32_t client_index = 0;
4737 
4738 	client_index++;
4739 	if (client_index != 0)
4740 		return (client_index);
4741 
4742 	printf("%s: out of clientids\n", __func__);
4743 	return (client_index);
4744 }
4745 
4746 /*
4747  * Return the next index# for a stateid. Mostly just increment and return
4748  * the next one, but... if the 32bit unsigned does actually wrap around
4749  * (will a BSD server stay up that long?), find
4750  * new start and end values.
4751  */
4752 static u_int32_t
4753 nfsrv_nextstateindex(struct nfsclient *clp)
4754 {
4755 	struct nfsstate *stp;
4756 	int i;
4757 	u_int32_t canuse, min_index, max_index;
4758 
4759 	if (!(clp->lc_flags & LCL_INDEXNOTOK)) {
4760 		clp->lc_stateindex++;
4761 		if (clp->lc_stateindex != clp->lc_statemaxindex)
4762 			return (clp->lc_stateindex);
4763 	}
4764 
4765 	/*
4766 	 * Yuck, we've hit the end.
4767 	 * Look for a new min and max.
4768 	 */
4769 	min_index = 0;
4770 	max_index = 0xffffffff;
4771 	for (i = 0; i < nfsrv_statehashsize; i++) {
4772 	    LIST_FOREACH(stp, &clp->lc_stateid[i], ls_hash) {
4773 		if (stp->ls_stateid.other[2] > 0x80000000) {
4774 		    if (stp->ls_stateid.other[2] < max_index)
4775 			max_index = stp->ls_stateid.other[2];
4776 		} else {
4777 		    if (stp->ls_stateid.other[2] > min_index)
4778 			min_index = stp->ls_stateid.other[2];
4779 		}
4780 	    }
4781 	}
4782 
4783 	/*
4784 	 * Yikes, highly unlikely, but I'll handle it anyhow.
4785 	 */
4786 	if (min_index == 0x80000000 && max_index == 0x80000001) {
4787 	    canuse = 0;
4788 	    /*
4789 	     * Loop around until we find an unused entry. Return that
4790 	     * and set LCL_INDEXNOTOK, so the search will continue next time.
4791 	     * (This is one of those rare cases where a goto is the
4792 	     *  cleanest way to code the loop.)
4793 	     */
4794 tryagain:
4795 	    for (i = 0; i < nfsrv_statehashsize; i++) {
4796 		LIST_FOREACH(stp, &clp->lc_stateid[i], ls_hash) {
4797 		    if (stp->ls_stateid.other[2] == canuse) {
4798 			canuse++;
4799 			goto tryagain;
4800 		    }
4801 		}
4802 	    }
4803 	    clp->lc_flags |= LCL_INDEXNOTOK;
4804 	    return (canuse);
4805 	}
4806 
4807 	/*
4808 	 * Ok to start again from min + 1.
4809 	 */
4810 	clp->lc_stateindex = min_index + 1;
4811 	clp->lc_statemaxindex = max_index;
4812 	clp->lc_flags &= ~LCL_INDEXNOTOK;
4813 	return (clp->lc_stateindex);
4814 }
4815 
4816 /*
4817  * The following functions handle the stable storage file that deals with
4818  * the edge conditions described in RFC3530 Sec. 8.6.3.
4819  * The file is as follows:
4820  * - a single record at the beginning that has the lease time of the
4821  *   previous server instance (before the last reboot) and the nfsrvboottime
4822  *   values for the previous server boots.
4823  *   These previous boot times are used to ensure that the current
4824  *   nfsrvboottime does not, somehow, get set to a previous one.
4825  *   (This is important so that Stale ClientIDs and StateIDs can
4826  *    be recognized.)
4827  *   The number of previous nfsvrboottime values precedes the list.
4828  * - followed by some number of appended records with:
4829  *   - client id string
4830  *   - flag that indicates it is a record revoking state via lease
4831  *     expiration or similar
4832  *     OR has successfully acquired state.
4833  * These structures vary in length, with the client string at the end, up
4834  * to NFSV4_OPAQUELIMIT in size.
4835  *
4836  * At the end of the grace period, the file is truncated, the first
4837  * record is rewritten with updated information and any acquired state
4838  * records for successful reclaims of state are written.
4839  *
4840  * Subsequent records are appended when the first state is issued to
4841  * a client and when state is revoked for a client.
4842  *
4843  * When reading the file in, state issued records that come later in
4844  * the file override older ones, since the append log is in cronological order.
4845  * If, for some reason, the file can't be read, the grace period is
4846  * immediately terminated and all reclaims get NFSERR_NOGRACE.
4847  */
4848 
4849 /*
4850  * Read in the stable storage file. Called by nfssvc() before the nfsd
4851  * processes start servicing requests.
4852  */
4853 void
4854 nfsrv_setupstable(NFSPROC_T *p)
4855 {
4856 	struct nfsrv_stablefirst *sf = &NFSD_VNET(nfsrv_stablefirst);
4857 	struct nfsrv_stable *sp, *nsp;
4858 	struct nfst_rec *tsp;
4859 	int error, i, tryagain;
4860 	off_t off = 0;
4861 	ssize_t aresid, len;
4862 
4863 	/*
4864 	 * If NFSNSF_UPDATEDONE is set, this is a restart of the nfsds without
4865 	 * a reboot, so state has not been lost.
4866 	 */
4867 	if (sf->nsf_flags & NFSNSF_UPDATEDONE)
4868 		return;
4869 	/*
4870 	 * Set Grace over just until the file reads successfully.
4871 	 */
4872 	NFSD_VNET(nfsrvboottime) = time_second;
4873 	LIST_INIT(&sf->nsf_head);
4874 	sf->nsf_flags = (NFSNSF_GRACEOVER | NFSNSF_NEEDLOCK);
4875 	sf->nsf_eograce = NFSD_MONOSEC + NFSRV_LEASEDELTA;
4876 	if (sf->nsf_fp == NULL)
4877 		return;
4878 	error = NFSD_RDWR(UIO_READ, NFSFPVNODE(sf->nsf_fp),
4879 	    (caddr_t)&sf->nsf_rec, sizeof (struct nfsf_rec), off, UIO_SYSSPACE,
4880 	    0, NFSFPCRED(sf->nsf_fp), &aresid, p);
4881 	if (error || aresid || sf->nsf_numboots == 0 ||
4882 		sf->nsf_numboots > NFSNSF_MAXNUMBOOTS)
4883 		return;
4884 
4885 	/*
4886 	 * Now, read in the boottimes.
4887 	 */
4888 	sf->nsf_bootvals = (time_t *)malloc((sf->nsf_numboots + 1) *
4889 		sizeof(time_t), M_TEMP, M_WAITOK);
4890 	off = sizeof (struct nfsf_rec);
4891 	error = NFSD_RDWR(UIO_READ, NFSFPVNODE(sf->nsf_fp),
4892 	    (caddr_t)sf->nsf_bootvals, sf->nsf_numboots * sizeof (time_t), off,
4893 	    UIO_SYSSPACE, 0, NFSFPCRED(sf->nsf_fp), &aresid, p);
4894 	if (error || aresid) {
4895 		free(sf->nsf_bootvals, M_TEMP);
4896 		sf->nsf_bootvals = NULL;
4897 		return;
4898 	}
4899 
4900 	/*
4901 	 * Make sure this nfsrvboottime is different from all recorded
4902 	 * previous ones.
4903 	 */
4904 	do {
4905 		tryagain = 0;
4906 		for (i = 0; i < sf->nsf_numboots; i++) {
4907 			if (NFSD_VNET(nfsrvboottime) == sf->nsf_bootvals[i]) {
4908 				NFSD_VNET(nfsrvboottime)++;
4909 				tryagain = 1;
4910 				break;
4911 			}
4912 		}
4913 	} while (tryagain);
4914 
4915 	sf->nsf_flags |= NFSNSF_OK;
4916 	off += (sf->nsf_numboots * sizeof (time_t));
4917 
4918 	/*
4919 	 * Read through the file, building a list of records for grace
4920 	 * checking.
4921 	 * Each record is between sizeof (struct nfst_rec) and
4922 	 * sizeof (struct nfst_rec) + NFSV4_OPAQUELIMIT - 1
4923 	 * and is actually sizeof (struct nfst_rec) + nst_len - 1.
4924 	 */
4925 	tsp = (struct nfst_rec *)malloc(sizeof (struct nfst_rec) +
4926 		NFSV4_OPAQUELIMIT - 1, M_TEMP, M_WAITOK);
4927 	do {
4928 	    error = NFSD_RDWR(UIO_READ, NFSFPVNODE(sf->nsf_fp),
4929 	        (caddr_t)tsp, sizeof (struct nfst_rec) + NFSV4_OPAQUELIMIT - 1,
4930 	        off, UIO_SYSSPACE, 0, NFSFPCRED(sf->nsf_fp), &aresid, p);
4931 	    len = (sizeof (struct nfst_rec) + NFSV4_OPAQUELIMIT - 1) - aresid;
4932 	    if (error || (len > 0 && (len < sizeof (struct nfst_rec) ||
4933 		len < (sizeof (struct nfst_rec) + tsp->len - 1)))) {
4934 		/*
4935 		 * Yuck, the file has been corrupted, so just return
4936 		 * after clearing out any restart state, so the grace period
4937 		 * is over.
4938 		 */
4939 		LIST_FOREACH_SAFE(sp, &sf->nsf_head, nst_list, nsp) {
4940 			LIST_REMOVE(sp, nst_list);
4941 			free(sp, M_TEMP);
4942 		}
4943 		free(tsp, M_TEMP);
4944 		sf->nsf_flags &= ~NFSNSF_OK;
4945 		free(sf->nsf_bootvals, M_TEMP);
4946 		sf->nsf_bootvals = NULL;
4947 		return;
4948 	    }
4949 	    if (len > 0) {
4950 		off += sizeof (struct nfst_rec) + tsp->len - 1;
4951 		/*
4952 		 * Search the list for a matching client.
4953 		 */
4954 		LIST_FOREACH(sp, &sf->nsf_head, nst_list) {
4955 			if (tsp->len == sp->nst_len &&
4956 			    !NFSBCMP(tsp->client, sp->nst_client, tsp->len))
4957 				break;
4958 		}
4959 		if (sp == LIST_END(&sf->nsf_head)) {
4960 			sp = (struct nfsrv_stable *)malloc(tsp->len +
4961 				sizeof (struct nfsrv_stable) - 1, M_TEMP,
4962 				M_WAITOK);
4963 			NFSBCOPY((caddr_t)tsp, (caddr_t)&sp->nst_rec,
4964 				sizeof (struct nfst_rec) + tsp->len - 1);
4965 			LIST_INSERT_HEAD(&sf->nsf_head, sp, nst_list);
4966 		} else {
4967 			if (tsp->flag == NFSNST_REVOKE)
4968 				sp->nst_flag |= NFSNST_REVOKE;
4969 			else
4970 				/*
4971 				 * A subsequent timestamp indicates the client
4972 				 * did a setclientid/confirm and any previous
4973 				 * revoke is no longer relevant.
4974 				 */
4975 				sp->nst_flag &= ~NFSNST_REVOKE;
4976 		}
4977 	    }
4978 	} while (len > 0);
4979 	free(tsp, M_TEMP);
4980 	sf->nsf_flags = NFSNSF_OK;
4981 	sf->nsf_eograce = NFSD_MONOSEC + sf->nsf_lease +
4982 		NFSRV_LEASEDELTA;
4983 }
4984 
4985 /*
4986  * Update the stable storage file, now that the grace period is over.
4987  */
4988 void
4989 nfsrv_updatestable(NFSPROC_T *p)
4990 {
4991 	struct nfsrv_stablefirst *sf = &NFSD_VNET(nfsrv_stablefirst);
4992 	struct nfsrv_stable *sp, *nsp;
4993 	int i;
4994 	struct nfsvattr nva;
4995 	vnode_t vp;
4996 #if defined(__FreeBSD_version) && (__FreeBSD_version >= 500000)
4997 	mount_t mp = NULL;
4998 #endif
4999 	int error;
5000 
5001 	if (sf->nsf_fp == NULL || (sf->nsf_flags & NFSNSF_UPDATEDONE))
5002 		return;
5003 	sf->nsf_flags |= NFSNSF_UPDATEDONE;
5004 	/*
5005 	 * Ok, we need to rewrite the stable storage file.
5006 	 * - truncate to 0 length
5007 	 * - write the new first structure
5008 	 * - loop through the data structures, writing out any that
5009 	 *   have timestamps older than the old boot
5010 	 */
5011 	if (sf->nsf_bootvals) {
5012 		sf->nsf_numboots++;
5013 		for (i = sf->nsf_numboots - 2; i >= 0; i--)
5014 			sf->nsf_bootvals[i + 1] = sf->nsf_bootvals[i];
5015 	} else {
5016 		sf->nsf_numboots = 1;
5017 		sf->nsf_bootvals = (time_t *)malloc(sizeof(time_t),
5018 			M_TEMP, M_WAITOK);
5019 	}
5020 	sf->nsf_bootvals[0] = NFSD_VNET(nfsrvboottime);
5021 	sf->nsf_lease = nfsrv_lease;
5022 	NFSVNO_ATTRINIT(&nva);
5023 	NFSVNO_SETATTRVAL(&nva, size, 0);
5024 	vp = NFSFPVNODE(sf->nsf_fp);
5025 	vn_start_write(vp, &mp, V_WAIT);
5026 	if (NFSVOPLOCK(vp, LK_EXCLUSIVE) == 0) {
5027 		error = nfsvno_setattr(vp, &nva, NFSFPCRED(sf->nsf_fp), p,
5028 		    NULL);
5029 		NFSVOPUNLOCK(vp);
5030 	} else
5031 		error = EPERM;
5032 	vn_finished_write(mp);
5033 	if (!error)
5034 	    error = NFSD_RDWR(UIO_WRITE, vp,
5035 		(caddr_t)&sf->nsf_rec, sizeof (struct nfsf_rec), (off_t)0,
5036 		UIO_SYSSPACE, IO_SYNC, NFSFPCRED(sf->nsf_fp), NULL, p);
5037 	if (!error)
5038 	    error = NFSD_RDWR(UIO_WRITE, vp,
5039 		(caddr_t)sf->nsf_bootvals,
5040 		sf->nsf_numboots * sizeof (time_t),
5041 		(off_t)(sizeof (struct nfsf_rec)),
5042 		UIO_SYSSPACE, IO_SYNC, NFSFPCRED(sf->nsf_fp), NULL, p);
5043 	free(sf->nsf_bootvals, M_TEMP);
5044 	sf->nsf_bootvals = NULL;
5045 	if (error) {
5046 		sf->nsf_flags &= ~NFSNSF_OK;
5047 		printf("EEK! Can't write NfsV4 stable storage file\n");
5048 		return;
5049 	}
5050 	sf->nsf_flags |= NFSNSF_OK;
5051 
5052 	/*
5053 	 * Loop through the list and write out timestamp records for
5054 	 * any clients that successfully reclaimed state.
5055 	 */
5056 	LIST_FOREACH_SAFE(sp, &sf->nsf_head, nst_list, nsp) {
5057 		if (sp->nst_flag & NFSNST_GOTSTATE) {
5058 			nfsrv_writestable(sp->nst_client, sp->nst_len,
5059 				NFSNST_NEWSTATE, p);
5060 			sp->nst_clp->lc_flags |= LCL_STAMPEDSTABLE;
5061 		}
5062 		LIST_REMOVE(sp, nst_list);
5063 		free(sp, M_TEMP);
5064 	}
5065 	nfsrv_backupstable();
5066 }
5067 
5068 /*
5069  * Append a record to the stable storage file.
5070  */
5071 void
5072 nfsrv_writestable(u_char *client, int len, int flag, NFSPROC_T *p)
5073 {
5074 	struct nfsrv_stablefirst *sf = &NFSD_VNET(nfsrv_stablefirst);
5075 	struct nfst_rec *sp;
5076 	int error;
5077 
5078 	if (!(sf->nsf_flags & NFSNSF_OK) || sf->nsf_fp == NULL)
5079 		return;
5080 	sp = (struct nfst_rec *)malloc(sizeof (struct nfst_rec) +
5081 		len - 1, M_TEMP, M_WAITOK);
5082 	sp->len = len;
5083 	NFSBCOPY(client, sp->client, len);
5084 	sp->flag = flag;
5085 	error = NFSD_RDWR(UIO_WRITE, NFSFPVNODE(sf->nsf_fp),
5086 	    (caddr_t)sp, sizeof (struct nfst_rec) + len - 1, (off_t)0,
5087 	    UIO_SYSSPACE, (IO_SYNC | IO_APPEND), NFSFPCRED(sf->nsf_fp), NULL, p);
5088 	free(sp, M_TEMP);
5089 	if (error) {
5090 		sf->nsf_flags &= ~NFSNSF_OK;
5091 		printf("EEK! Can't write NfsV4 stable storage file\n");
5092 	}
5093 }
5094 
5095 /*
5096  * This function is called during the grace period to mark a client
5097  * that successfully reclaimed state.
5098  */
5099 static void
5100 nfsrv_markstable(struct nfsclient *clp)
5101 {
5102 	struct nfsrv_stable *sp;
5103 
5104 	/*
5105 	 * First find the client structure.
5106 	 */
5107 	LIST_FOREACH(sp, &NFSD_VNET(nfsrv_stablefirst).nsf_head, nst_list) {
5108 		if (sp->nst_len == clp->lc_idlen &&
5109 		    !NFSBCMP(sp->nst_client, clp->lc_id, sp->nst_len))
5110 			break;
5111 	}
5112 	if (sp == LIST_END(&NFSD_VNET(nfsrv_stablefirst).nsf_head))
5113 		return;
5114 
5115 	/*
5116 	 * Now, just mark it and set the nfsclient back pointer.
5117 	 */
5118 	sp->nst_flag |= NFSNST_GOTSTATE;
5119 	sp->nst_clp = clp;
5120 }
5121 
5122 /*
5123  * This function is called when a NFSv4.1 client does a ReclaimComplete.
5124  * Very similar to nfsrv_markstable(), except for the flag being set.
5125  */
5126 static void
5127 nfsrv_markreclaim(struct nfsclient *clp)
5128 {
5129 	struct nfsrv_stable *sp;
5130 
5131 	/*
5132 	 * First find the client structure.
5133 	 */
5134 	LIST_FOREACH(sp, &NFSD_VNET(nfsrv_stablefirst).nsf_head, nst_list) {
5135 		if (sp->nst_len == clp->lc_idlen &&
5136 		    !NFSBCMP(sp->nst_client, clp->lc_id, sp->nst_len))
5137 			break;
5138 	}
5139 	if (sp == LIST_END(&NFSD_VNET(nfsrv_stablefirst).nsf_head))
5140 		return;
5141 
5142 	/*
5143 	 * Now, just set the flag.
5144 	 */
5145 	sp->nst_flag |= NFSNST_RECLAIMED;
5146 
5147 	/*
5148 	 * Free up any old delegations.
5149 	 */
5150 	nfsrv_freedeleglist(&clp->lc_olddeleg);
5151 }
5152 
5153 /*
5154  * This function is called for a reclaim, to see if it gets grace.
5155  * It returns 0 if a reclaim is allowed, 1 otherwise.
5156  */
5157 static int
5158 nfsrv_checkstable(struct nfsclient *clp)
5159 {
5160 	struct nfsrv_stable *sp;
5161 
5162 	/*
5163 	 * First, find the entry for the client.
5164 	 */
5165 	LIST_FOREACH(sp, &NFSD_VNET(nfsrv_stablefirst).nsf_head, nst_list) {
5166 		if (sp->nst_len == clp->lc_idlen &&
5167 		    !NFSBCMP(sp->nst_client, clp->lc_id, sp->nst_len))
5168 			break;
5169 	}
5170 
5171 	/*
5172 	 * If not in the list, state was revoked or no state was issued
5173 	 * since the previous reboot, a reclaim is denied.
5174 	 */
5175 	if (sp == LIST_END(&NFSD_VNET(nfsrv_stablefirst).nsf_head) ||
5176 	    (sp->nst_flag & NFSNST_REVOKE) ||
5177 	    !(NFSD_VNET(nfsrv_stablefirst).nsf_flags & NFSNSF_OK))
5178 		return (1);
5179 	return (0);
5180 }
5181 
5182 /*
5183  * Test for and try to clear out a conflicting client. This is called by
5184  * nfsrv_lockctrl() and nfsrv_openctrl() when conflicts with other clients
5185  * a found.
5186  * The trick here is that it can't revoke a conflicting client with an
5187  * expired lease unless it holds the v4root lock, so...
5188  * If no v4root lock, get the lock and return 1 to indicate "try again".
5189  * Return 0 to indicate the conflict can't be revoked and 1 to indicate
5190  * the revocation worked and the conflicting client is "bye, bye", so it
5191  * can be tried again.
5192  * Return 2 to indicate that the vnode is VIRF_DOOMED after NFSVOPLOCK().
5193  * Unlocks State before a non-zero value is returned.
5194  */
5195 static int
5196 nfsrv_clientconflict(struct nfsclient *clp, int *haslockp, vnode_t vp,
5197     NFSPROC_T *p)
5198 {
5199 	int gotlock, lktype = 0;
5200 
5201 	/*
5202 	 * If lease hasn't expired, we can't fix it.
5203 	 */
5204 	if (clp->lc_expiry >= NFSD_MONOSEC ||
5205 	    !(NFSD_VNET(nfsrv_stablefirst).nsf_flags & NFSNSF_UPDATEDONE))
5206 		return (0);
5207 	if (*haslockp == 0) {
5208 		NFSUNLOCKSTATE();
5209 		if (vp != NULL) {
5210 			lktype = NFSVOPISLOCKED(vp);
5211 			NFSVOPUNLOCK(vp);
5212 		}
5213 		NFSLOCKV4ROOTMUTEX();
5214 		nfsv4_relref(&nfsv4rootfs_lock);
5215 		do {
5216 			gotlock = nfsv4_lock(&nfsv4rootfs_lock, 1, NULL,
5217 			    NFSV4ROOTLOCKMUTEXPTR, NULL);
5218 		} while (!gotlock);
5219 		NFSUNLOCKV4ROOTMUTEX();
5220 		*haslockp = 1;
5221 		if (vp != NULL) {
5222 			NFSVOPLOCK(vp, lktype | LK_RETRY);
5223 			if (VN_IS_DOOMED(vp))
5224 				return (2);
5225 		}
5226 		return (1);
5227 	}
5228 	NFSUNLOCKSTATE();
5229 
5230 	/*
5231 	 * Ok, we can expire the conflicting client.
5232 	 */
5233 	nfsrv_writestable(clp->lc_id, clp->lc_idlen, NFSNST_REVOKE, p);
5234 	nfsrv_backupstable();
5235 	nfsrv_cleanclient(clp, p, false, NULL);
5236 	nfsrv_freedeleglist(&clp->lc_deleg);
5237 	nfsrv_freedeleglist(&clp->lc_olddeleg);
5238 	LIST_REMOVE(clp, lc_hash);
5239 	nfsrv_zapclient(clp, p);
5240 	return (1);
5241 }
5242 
5243 /*
5244  * Resolve a delegation conflict.
5245  * Returns 0 to indicate the conflict was resolved without sleeping.
5246  * Return -1 to indicate that the caller should check for conflicts again.
5247  * Return > 0 for an error that should be returned, normally NFSERR_DELAY.
5248  *
5249  * Also, manipulate the nfsv4root_lock, as required. It isn't changed
5250  * for a return of 0, since there was no sleep and it could be required
5251  * later. It is released for a return of NFSERR_DELAY, since the caller
5252  * will return that error. It is released when a sleep was done waiting
5253  * for the delegation to be returned or expire (so that other nfsds can
5254  * handle ops). Then, it must be acquired for the write to stable storage.
5255  * (This function is somewhat similar to nfsrv_clientconflict(), but
5256  *  the semantics differ in a couple of subtle ways. The return of 0
5257  *  indicates the conflict was resolved without sleeping here, not
5258  *  that the conflict can't be resolved and the handling of nfsv4root_lock
5259  *  differs, as noted above.)
5260  * Unlocks State before returning a non-zero value.
5261  */
5262 static int
5263 nfsrv_delegconflict(struct nfsstate *stp, int *haslockp, NFSPROC_T *p,
5264     vnode_t vp)
5265 {
5266 	struct nfsclient *clp = stp->ls_clp;
5267 	int gotlock, error, lktype = 0, retrycnt, zapped_clp;
5268 	nfsv4stateid_t tstateid;
5269 	fhandle_t tfh;
5270 
5271 	/*
5272 	 * If the conflict is with an old delegation...
5273 	 */
5274 	if (stp->ls_flags & NFSLCK_OLDDELEG) {
5275 		/*
5276 		 * You can delete it, if it has expired.
5277 		 */
5278 		if (clp->lc_delegtime < NFSD_MONOSEC) {
5279 			nfsrv_freedeleg(stp);
5280 			NFSUNLOCKSTATE();
5281 			error = -1;
5282 			goto out;
5283 		}
5284 		NFSUNLOCKSTATE();
5285 		/*
5286 		 * During this delay, the old delegation could expire or it
5287 		 * could be recovered by the client via an Open with
5288 		 * CLAIM_DELEGATE_PREV.
5289 		 * Release the nfsv4root_lock, if held.
5290 		 */
5291 		if (*haslockp) {
5292 			*haslockp = 0;
5293 			NFSLOCKV4ROOTMUTEX();
5294 			nfsv4_unlock(&nfsv4rootfs_lock, 1);
5295 			NFSUNLOCKV4ROOTMUTEX();
5296 		}
5297 		error = NFSERR_DELAY;
5298 		goto out;
5299 	}
5300 
5301 	/*
5302 	 * It's a current delegation, so:
5303 	 * - check to see if the delegation has expired
5304 	 *   - if so, get the v4root lock and then expire it
5305 	 */
5306 	if ((stp->ls_flags & NFSLCK_DELEGRECALL) == 0 || (stp->ls_lastrecall <
5307 	    NFSD_MONOSEC && clp->lc_expiry >= NFSD_MONOSEC &&
5308 	    stp->ls_delegtime >= NFSD_MONOSEC)) {
5309 		/*
5310 		 * - do a recall callback, since not yet done
5311 		 * For now, never allow truncate to be set. To use
5312 		 * truncate safely, it must be guaranteed that the
5313 		 * Remove, Rename or Setattr with size of 0 will
5314 		 * succeed and that would require major changes to
5315 		 * the VFS/Vnode OPs.
5316 		 * Set the expiry time large enough so that it won't expire
5317 		 * until after the callback, then set it correctly, once
5318 		 * the callback is done. (The delegation will now time
5319 		 * out whether or not the Recall worked ok. The timeout
5320 		 * will be extended when ops are done on the delegation
5321 		 * stateid, up to the timelimit.)
5322 		 */
5323 		if ((stp->ls_flags & NFSLCK_DELEGRECALL) == 0) {
5324 			stp->ls_delegtime = NFSD_MONOSEC + (2 * nfsrv_lease) +
5325 			    NFSRV_LEASEDELTA;
5326 			stp->ls_delegtimelimit = NFSD_MONOSEC + (6 *
5327 			    nfsrv_lease) + NFSRV_LEASEDELTA;
5328 			stp->ls_flags |= NFSLCK_DELEGRECALL;
5329 		}
5330 		stp->ls_lastrecall = time_uptime + 1;
5331 
5332 		/*
5333 		 * Loop NFSRV_CBRETRYCNT times while the CBRecall replies
5334 		 * NFSERR_BADSTATEID or NFSERR_BADHANDLE. This is done
5335 		 * in order to try and avoid a race that could happen
5336 		 * when a CBRecall request passed the Open reply with
5337 		 * the delegation in it when transitting the network.
5338 		 * Since nfsrv_docallback will sleep, don't use stp after
5339 		 * the call.
5340 		 */
5341 		NFSBCOPY((caddr_t)&stp->ls_stateid, (caddr_t)&tstateid,
5342 		    sizeof (tstateid));
5343 		NFSBCOPY((caddr_t)&stp->ls_lfp->lf_fh, (caddr_t)&tfh,
5344 		    sizeof (tfh));
5345 		NFSUNLOCKSTATE();
5346 		if (*haslockp) {
5347 			*haslockp = 0;
5348 			NFSLOCKV4ROOTMUTEX();
5349 			nfsv4_unlock(&nfsv4rootfs_lock, 1);
5350 			NFSUNLOCKV4ROOTMUTEX();
5351 		}
5352 		retrycnt = 0;
5353 		do {
5354 		    error = nfsrv_docallback(clp, NFSV4OP_CBRECALL,
5355 			&tstateid, 0, &tfh, NULL, NULL, 0, p);
5356 		    retrycnt++;
5357 		} while ((error == NFSERR_BADSTATEID ||
5358 		    error == NFSERR_BADHANDLE) && retrycnt < NFSV4_CBRETRYCNT);
5359 		error = NFSERR_DELAY;
5360 		goto out;
5361 	}
5362 
5363 	if (clp->lc_expiry >= NFSD_MONOSEC &&
5364 	    stp->ls_delegtime >= NFSD_MONOSEC) {
5365 		NFSUNLOCKSTATE();
5366 		/*
5367 		 * A recall has been done, but it has not yet expired.
5368 		 * So, RETURN_DELAY.
5369 		 */
5370 		if (*haslockp) {
5371 			*haslockp = 0;
5372 			NFSLOCKV4ROOTMUTEX();
5373 			nfsv4_unlock(&nfsv4rootfs_lock, 1);
5374 			NFSUNLOCKV4ROOTMUTEX();
5375 		}
5376 		error = NFSERR_DELAY;
5377 		goto out;
5378 	}
5379 
5380 	/*
5381 	 * If we don't yet have the lock, just get it and then return,
5382 	 * since we need that before deleting expired state, such as
5383 	 * this delegation.
5384 	 * When getting the lock, unlock the vnode, so other nfsds that
5385 	 * are in progress, won't get stuck waiting for the vnode lock.
5386 	 */
5387 	if (*haslockp == 0) {
5388 		NFSUNLOCKSTATE();
5389 		if (vp != NULL) {
5390 			lktype = NFSVOPISLOCKED(vp);
5391 			NFSVOPUNLOCK(vp);
5392 		}
5393 		NFSLOCKV4ROOTMUTEX();
5394 		nfsv4_relref(&nfsv4rootfs_lock);
5395 		do {
5396 			gotlock = nfsv4_lock(&nfsv4rootfs_lock, 1, NULL,
5397 			    NFSV4ROOTLOCKMUTEXPTR, NULL);
5398 		} while (!gotlock);
5399 		NFSUNLOCKV4ROOTMUTEX();
5400 		*haslockp = 1;
5401 		if (vp != NULL) {
5402 			NFSVOPLOCK(vp, lktype | LK_RETRY);
5403 			if (VN_IS_DOOMED(vp)) {
5404 				*haslockp = 0;
5405 				NFSLOCKV4ROOTMUTEX();
5406 				nfsv4_unlock(&nfsv4rootfs_lock, 1);
5407 				NFSUNLOCKV4ROOTMUTEX();
5408 				error = NFSERR_PERM;
5409 				goto out;
5410 			}
5411 		}
5412 		error = -1;
5413 		goto out;
5414 	}
5415 
5416 	NFSUNLOCKSTATE();
5417 	/*
5418 	 * Ok, we can delete the expired delegation.
5419 	 * First, write the Revoke record to stable storage and then
5420 	 * clear out the conflict.
5421 	 * Since all other nfsd threads are now blocked, we can safely
5422 	 * sleep without the state changing.
5423 	 */
5424 	nfsrv_writestable(clp->lc_id, clp->lc_idlen, NFSNST_REVOKE, p);
5425 	nfsrv_backupstable();
5426 	if (clp->lc_expiry < NFSD_MONOSEC) {
5427 		nfsrv_cleanclient(clp, p, false, NULL);
5428 		nfsrv_freedeleglist(&clp->lc_deleg);
5429 		nfsrv_freedeleglist(&clp->lc_olddeleg);
5430 		LIST_REMOVE(clp, lc_hash);
5431 		zapped_clp = 1;
5432 	} else {
5433 		nfsrv_freedeleg(stp);
5434 		zapped_clp = 0;
5435 	}
5436 	if (zapped_clp)
5437 		nfsrv_zapclient(clp, p);
5438 	error = -1;
5439 
5440 out:
5441 	NFSEXITCODE(error);
5442 	return (error);
5443 }
5444 
5445 /*
5446  * Check for a remove allowed, if remove is set to 1 and get rid of
5447  * delegations.
5448  */
5449 int
5450 nfsrv_checkremove(vnode_t vp, int remove, struct nfsrv_descript *nd,
5451     nfsquad_t clientid, NFSPROC_T *p)
5452 {
5453 	struct nfsclient *clp;
5454 	struct nfsstate *stp;
5455 	struct nfslockfile *lfp;
5456 	int error, haslock = 0;
5457 	fhandle_t nfh;
5458 
5459 	clp = NULL;
5460 	/*
5461 	 * First, get the lock file structure.
5462 	 * (A return of -1 means no associated state, so remove ok.)
5463 	 */
5464 	error = nfsrv_getlockfh(vp, NFSLCK_CHECK, NULL, &nfh, p);
5465 tryagain:
5466 	NFSLOCKSTATE();
5467 	if (error == 0 && clientid.qval != 0)
5468 		error = nfsrv_getclient(clientid, CLOPS_RENEW, &clp, NULL,
5469 		    (nfsquad_t)((u_quad_t)0), 0, nd, p);
5470 	if (!error)
5471 		error = nfsrv_getlockfile(NFSLCK_CHECK, NULL, &lfp, &nfh, 0);
5472 	if (error) {
5473 		NFSUNLOCKSTATE();
5474 		if (haslock) {
5475 			NFSLOCKV4ROOTMUTEX();
5476 			nfsv4_unlock(&nfsv4rootfs_lock, 1);
5477 			NFSUNLOCKV4ROOTMUTEX();
5478 		}
5479 		if (error == -1)
5480 			error = 0;
5481 		goto out;
5482 	}
5483 
5484 	/*
5485 	 * Now, we must Recall any delegations.
5486 	 */
5487 	error = nfsrv_cleandeleg(vp, lfp, clp, &haslock, p);
5488 	if (error) {
5489 		/*
5490 		 * nfsrv_cleandeleg() unlocks state for non-zero
5491 		 * return.
5492 		 */
5493 		if (error == -1)
5494 			goto tryagain;
5495 		if (haslock) {
5496 			NFSLOCKV4ROOTMUTEX();
5497 			nfsv4_unlock(&nfsv4rootfs_lock, 1);
5498 			NFSUNLOCKV4ROOTMUTEX();
5499 		}
5500 		goto out;
5501 	}
5502 
5503 	/*
5504 	 * Now, look for a conflicting open share.
5505 	 */
5506 	if (remove) {
5507 		/*
5508 		 * If the entry in the directory was the last reference to the
5509 		 * corresponding filesystem object, the object can be destroyed
5510 		 * */
5511 		if(lfp->lf_usecount>1)
5512 			LIST_FOREACH(stp, &lfp->lf_open, ls_file) {
5513 				if (stp->ls_flags & NFSLCK_WRITEDENY) {
5514 					error = NFSERR_FILEOPEN;
5515 					break;
5516 				}
5517 			}
5518 	}
5519 
5520 	NFSUNLOCKSTATE();
5521 	if (haslock) {
5522 		NFSLOCKV4ROOTMUTEX();
5523 		nfsv4_unlock(&nfsv4rootfs_lock, 1);
5524 		NFSUNLOCKV4ROOTMUTEX();
5525 	}
5526 
5527 out:
5528 	NFSEXITCODE(error);
5529 	return (error);
5530 }
5531 
5532 /*
5533  * Clear out all delegations for the file referred to by lfp.
5534  * May return NFSERR_DELAY, if there will be a delay waiting for
5535  * delegations to expire.
5536  * Returns -1 to indicate it slept while recalling a delegation.
5537  * This function has the side effect of deleting the nfslockfile structure,
5538  * if it no longer has associated state and didn't have to sleep.
5539  * Unlocks State before a non-zero value is returned.
5540  */
5541 static int
5542 nfsrv_cleandeleg(vnode_t vp, struct nfslockfile *lfp,
5543     struct nfsclient *clp, int *haslockp, NFSPROC_T *p)
5544 {
5545 	struct nfsstate *stp, *nstp;
5546 	int ret = 0;
5547 
5548 	stp = LIST_FIRST(&lfp->lf_deleg);
5549 	while (stp != LIST_END(&lfp->lf_deleg)) {
5550 		nstp = LIST_NEXT(stp, ls_file);
5551 		if (stp->ls_clp != clp) {
5552 			ret = nfsrv_delegconflict(stp, haslockp, p, vp);
5553 			if (ret) {
5554 				/*
5555 				 * nfsrv_delegconflict() unlocks state
5556 				 * when it returns non-zero.
5557 				 */
5558 				goto out;
5559 			}
5560 		}
5561 		stp = nstp;
5562 	}
5563 out:
5564 	NFSEXITCODE(ret);
5565 	return (ret);
5566 }
5567 
5568 /*
5569  * There are certain operations that, when being done outside of NFSv4,
5570  * require that any NFSv4 delegation for the file be recalled.
5571  * This function is to be called for those cases:
5572  * VOP_RENAME() - When a delegation is being recalled for any reason,
5573  *	the client may have to do Opens against the server, using the file's
5574  *	final component name. If the file has been renamed on the server,
5575  *	that component name will be incorrect and the Open will fail.
5576  * VOP_REMOVE() - Theoretically, a client could Open a file after it has
5577  *	been removed on the server, if there is a delegation issued to
5578  *	that client for the file. I say "theoretically" since clients
5579  *	normally do an Access Op before the Open and that Access Op will
5580  *	fail with ESTALE. Note that NFSv2 and 3 don't even do Opens, so
5581  *	they will detect the file's removal in the same manner. (There is
5582  *	one case where RFC3530 allows a client to do an Open without first
5583  *	doing an Access Op, which is passage of a check against the ACE
5584  *	returned with a Write delegation, but current practice is to ignore
5585  *	the ACE and always do an Access Op.)
5586  *	Since the functions can only be called with an unlocked vnode, this
5587  *	can't be done at this time.
5588  * VOP_ADVLOCK() - When a client holds a delegation, it can issue byte range
5589  *	locks locally in the client, which are not visible to the server. To
5590  *	deal with this, issuing of delegations for a vnode must be disabled
5591  *	and all delegations for the vnode recalled. This is done via the
5592  *	second function, using the VV_DISABLEDELEG vflag on the vnode.
5593  */
5594 void
5595 nfsd_recalldelegation(vnode_t vp, NFSPROC_T *p)
5596 {
5597 	time_t starttime;
5598 	int error;
5599 
5600 	/*
5601 	 * First, check to see if the server is currently running and it has
5602 	 * been called for a regular file when issuing delegations.
5603 	 */
5604 	if (NFSD_VNET(nfsrv_numnfsd) == 0 || vp->v_type != VREG ||
5605 	    nfsrv_issuedelegs == 0)
5606 		return;
5607 
5608 	KASSERT((NFSVOPISLOCKED(vp) != LK_EXCLUSIVE), ("vp %p is locked", vp));
5609 	/*
5610 	 * First, get a reference on the nfsv4rootfs_lock so that an
5611 	 * exclusive lock cannot be acquired by another thread.
5612 	 */
5613 	NFSLOCKV4ROOTMUTEX();
5614 	nfsv4_getref(&nfsv4rootfs_lock, NULL, NFSV4ROOTLOCKMUTEXPTR, NULL);
5615 	NFSUNLOCKV4ROOTMUTEX();
5616 
5617 	/*
5618 	 * Now, call nfsrv_checkremove() in a loop while it returns
5619 	 * NFSERR_DELAY. Return upon any other error or when timed out.
5620 	 */
5621 	starttime = NFSD_MONOSEC;
5622 	do {
5623 		if (NFSVOPLOCK(vp, LK_EXCLUSIVE) == 0) {
5624 			error = nfsrv_checkremove(vp, 0, NULL,
5625 			    (nfsquad_t)((u_quad_t)0), p);
5626 			NFSVOPUNLOCK(vp);
5627 		} else
5628 			error = EPERM;
5629 		if (error == NFSERR_DELAY) {
5630 			if (NFSD_MONOSEC - starttime > NFS_REMOVETIMEO)
5631 				break;
5632 			/* Sleep for a short period of time */
5633 			(void) nfs_catnap(PZERO, 0, "nfsremove");
5634 		}
5635 	} while (error == NFSERR_DELAY);
5636 	NFSLOCKV4ROOTMUTEX();
5637 	nfsv4_relref(&nfsv4rootfs_lock);
5638 	NFSUNLOCKV4ROOTMUTEX();
5639 }
5640 
5641 void
5642 nfsd_disabledelegation(vnode_t vp, NFSPROC_T *p)
5643 {
5644 
5645 #ifdef VV_DISABLEDELEG
5646 	/*
5647 	 * First, flag issuance of delegations disabled.
5648 	 */
5649 	atomic_set_long(&vp->v_vflag, VV_DISABLEDELEG);
5650 #endif
5651 
5652 	/*
5653 	 * Then call nfsd_recalldelegation() to get rid of all extant
5654 	 * delegations.
5655 	 */
5656 	nfsd_recalldelegation(vp, p);
5657 }
5658 
5659 /*
5660  * Check for conflicting locks, etc. and then get rid of delegations.
5661  * (At one point I thought that I should get rid of delegations for any
5662  *  Setattr, since it could potentially disallow the I/O op (read or write)
5663  *  allowed by the delegation. However, Setattr Ops that aren't changing
5664  *  the size get a stateid of all 0s, so you can't tell if it is a delegation
5665  *  for the same client or a different one, so I decided to only get rid
5666  *  of delegations for other clients when the size is being changed.)
5667  * In general, a Setattr can disable NFS I/O Ops that are outstanding, such
5668  * as Write backs, even if there is no delegation, so it really isn't any
5669  * different?)
5670  */
5671 int
5672 nfsrv_checksetattr(vnode_t vp, struct nfsrv_descript *nd,
5673     nfsv4stateid_t *stateidp, struct nfsvattr *nvap, nfsattrbit_t *attrbitp,
5674     struct nfsexstuff *exp, NFSPROC_T *p)
5675 {
5676 	struct nfsstate st, *stp = &st;
5677 	struct nfslock lo, *lop = &lo;
5678 	int error = 0;
5679 	nfsquad_t clientid;
5680 
5681 	if (NFSISSET_ATTRBIT(attrbitp, NFSATTRBIT_SIZE)) {
5682 		stp->ls_flags = (NFSLCK_CHECK | NFSLCK_WRITEACCESS);
5683 		lop->lo_first = nvap->na_size;
5684 	} else {
5685 		stp->ls_flags = 0;
5686 		lop->lo_first = 0;
5687 	}
5688 	if (NFSISSET_ATTRBIT(attrbitp, NFSATTRBIT_OWNER) ||
5689 	    NFSISSET_ATTRBIT(attrbitp, NFSATTRBIT_OWNERGROUP) ||
5690 	    NFSISSET_ATTRBIT(attrbitp, NFSATTRBIT_MODE) ||
5691 	    NFSISSET_ATTRBIT(attrbitp, NFSATTRBIT_ACL))
5692 		stp->ls_flags |= NFSLCK_SETATTR;
5693 	if (stp->ls_flags == 0)
5694 		goto out;
5695 	lop->lo_end = NFS64BITSSET;
5696 	lop->lo_flags = NFSLCK_WRITE;
5697 	stp->ls_ownerlen = 0;
5698 	stp->ls_op = NULL;
5699 	stp->ls_uid = nd->nd_cred->cr_uid;
5700 	stp->ls_stateid.seqid = stateidp->seqid;
5701 	clientid.lval[0] = stp->ls_stateid.other[0] = stateidp->other[0];
5702 	clientid.lval[1] = stp->ls_stateid.other[1] = stateidp->other[1];
5703 	stp->ls_stateid.other[2] = stateidp->other[2];
5704 	error = nfsrv_lockctrl(vp, &stp, &lop, NULL, clientid,
5705 	    stateidp, exp, nd, p);
5706 
5707 out:
5708 	NFSEXITCODE2(error, nd);
5709 	return (error);
5710 }
5711 
5712 /*
5713  * Check for a write delegation and do a CBGETATTR if there is one, updating
5714  * the attributes, as required.
5715  * Should I return an error if I can't get the attributes? (For now, I'll
5716  * just return ok.
5717  */
5718 int
5719 nfsrv_checkgetattr(struct nfsrv_descript *nd, vnode_t vp,
5720     struct nfsvattr *nvap, nfsattrbit_t *attrbitp, NFSPROC_T *p)
5721 {
5722 	struct nfsstate *stp;
5723 	struct nfslockfile *lfp;
5724 	struct nfsclient *clp;
5725 	struct nfsvattr nva;
5726 	fhandle_t nfh;
5727 	int error = 0;
5728 	nfsattrbit_t cbbits;
5729 	u_quad_t delegfilerev;
5730 
5731 	NFSCBGETATTR_ATTRBIT(attrbitp, &cbbits);
5732 	if (!NFSNONZERO_ATTRBIT(&cbbits))
5733 		goto out;
5734 	if (nfsrv_writedelegcnt == 0)
5735 		goto out;
5736 
5737 	/*
5738 	 * Get the lock file structure.
5739 	 * (A return of -1 means no associated state, so return ok.)
5740 	 */
5741 	error = nfsrv_getlockfh(vp, NFSLCK_CHECK, NULL, &nfh, p);
5742 	NFSLOCKSTATE();
5743 	if (!error)
5744 		error = nfsrv_getlockfile(NFSLCK_CHECK, NULL, &lfp, &nfh, 0);
5745 	if (error) {
5746 		NFSUNLOCKSTATE();
5747 		if (error == -1)
5748 			error = 0;
5749 		goto out;
5750 	}
5751 
5752 	/*
5753 	 * Now, look for a write delegation.
5754 	 */
5755 	LIST_FOREACH(stp, &lfp->lf_deleg, ls_file) {
5756 		if (stp->ls_flags & NFSLCK_DELEGWRITE)
5757 			break;
5758 	}
5759 	if (stp == LIST_END(&lfp->lf_deleg)) {
5760 		NFSUNLOCKSTATE();
5761 		goto out;
5762 	}
5763 	clp = stp->ls_clp;
5764 
5765 	/* If the clientid is not confirmed, ignore the delegation. */
5766 	if (clp->lc_flags & LCL_NEEDSCONFIRM) {
5767 		NFSUNLOCKSTATE();
5768 		goto out;
5769 	}
5770 
5771 	delegfilerev = stp->ls_filerev;
5772 	/*
5773 	 * If the Write delegation was issued as a part of this Compound RPC
5774 	 * or if we have an Implied Clientid (used in a previous Op in this
5775 	 * compound) and it is the client the delegation was issued to,
5776 	 * just return ok.
5777 	 * I also assume that it is from the same client iff the network
5778 	 * host IP address is the same as the callback address. (Not
5779 	 * exactly correct by the RFC, but avoids a lot of Getattr
5780 	 * callbacks.)
5781 	 */
5782 	if (nd->nd_compref == stp->ls_compref ||
5783 	    ((nd->nd_flag & ND_IMPLIEDCLID) &&
5784 	     clp->lc_clientid.qval == nd->nd_clientid.qval) ||
5785 	     nfsaddr2_match(clp->lc_req.nr_nam, nd->nd_nam)) {
5786 		NFSUNLOCKSTATE();
5787 		goto out;
5788 	}
5789 
5790 	/*
5791 	 * We are now done with the delegation state structure,
5792 	 * so the statelock can be released and we can now tsleep().
5793 	 */
5794 
5795 	/*
5796 	 * Now, we must do the CB Getattr callback, to see if Change or Size
5797 	 * has changed.
5798 	 */
5799 	if (clp->lc_expiry >= NFSD_MONOSEC) {
5800 		NFSUNLOCKSTATE();
5801 		NFSVNO_ATTRINIT(&nva);
5802 		nva.na_filerev = NFS64BITSSET;
5803 		error = nfsrv_docallback(clp, NFSV4OP_CBGETATTR, NULL,
5804 		    0, &nfh, &nva, &cbbits, 0, p);
5805 		if (!error) {
5806 			if ((nva.na_filerev != NFS64BITSSET &&
5807 			    nva.na_filerev > delegfilerev) ||
5808 			    (NFSVNO_ISSETSIZE(&nva) &&
5809 			     nva.na_size != nvap->na_size)) {
5810 				error = nfsvno_updfilerev(vp, nvap, nd, p);
5811 				if (NFSVNO_ISSETSIZE(&nva))
5812 					nvap->na_size = nva.na_size;
5813 			}
5814 		} else
5815 			error = 0;	/* Ignore callback errors for now. */
5816 	} else {
5817 		NFSUNLOCKSTATE();
5818 	}
5819 
5820 out:
5821 	NFSEXITCODE2(error, nd);
5822 	return (error);
5823 }
5824 
5825 /*
5826  * This function looks for openowners that haven't had any opens for
5827  * a while and throws them away. Called by an nfsd when NFSNSF_NOOPENS
5828  * is set.
5829  */
5830 void
5831 nfsrv_throwawayopens(NFSPROC_T *p)
5832 {
5833 	struct nfsclient *clp, *nclp;
5834 	struct nfsstate *stp, *nstp;
5835 	int i;
5836 
5837 	NFSLOCKSTATE();
5838 	NFSD_VNET(nfsrv_stablefirst).nsf_flags &= ~NFSNSF_NOOPENS;
5839 	/*
5840 	 * For each client...
5841 	 */
5842 	for (i = 0; i < nfsrv_clienthashsize; i++) {
5843 	    LIST_FOREACH_SAFE(clp, &NFSD_VNET(nfsclienthash)[i], lc_hash,
5844 		nclp) {
5845 		LIST_FOREACH_SAFE(stp, &clp->lc_open, ls_list, nstp) {
5846 			if (LIST_EMPTY(&stp->ls_open) &&
5847 			    (stp->ls_noopens > NFSNOOPEN ||
5848 			     (nfsrv_openpluslock * 2) >
5849 			     nfsrv_v4statelimit))
5850 				nfsrv_freeopenowner(stp, 0, p);
5851 		}
5852 	    }
5853 	}
5854 	NFSUNLOCKSTATE();
5855 }
5856 
5857 /*
5858  * This function checks to see if the credentials are the same.
5859  * The check for same credentials is needed for state management operations
5860  * for NFSv4.0 or NFSv4.1/4.2 when SP4_MACH_CRED is configured via
5861  * ExchangeID.
5862  * Returns 1 for not same, 0 otherwise.
5863  */
5864 static int
5865 nfsrv_notsamecredname(int op, struct nfsrv_descript *nd, struct nfsclient *clp)
5866 {
5867 
5868 	/* Check for the SP4_MACH_CRED case. */
5869 	if (op != 0 && nfsrv_checkmachcred(op, nd, clp) != 0)
5870 		return (1);
5871 
5872 	/* For NFSv4.1/4.2, SP4_NONE always allows this. */
5873 	if ((nd->nd_flag & ND_NFSV41) != 0)
5874 		return (0);
5875 
5876 	if (nd->nd_flag & ND_GSS) {
5877 		if (!(clp->lc_flags & LCL_GSS))
5878 			return (1);
5879 		if (clp->lc_flags & LCL_NAME) {
5880 			if (nd->nd_princlen != clp->lc_namelen ||
5881 			    NFSBCMP(nd->nd_principal, clp->lc_name,
5882 				clp->lc_namelen))
5883 				return (1);
5884 			else
5885 				return (0);
5886 		}
5887 		if (nd->nd_cred->cr_uid == clp->lc_uid)
5888 			return (0);
5889 		else
5890 			return (1);
5891 	} else if (clp->lc_flags & LCL_GSS)
5892 		return (1);
5893 	/*
5894 	 * For AUTH_SYS, allow the same uid or root. (This is underspecified
5895 	 * in RFC3530, which talks about principals, but doesn't say anything
5896 	 * about uids for AUTH_SYS.)
5897 	 */
5898 	if (nd->nd_cred->cr_uid == clp->lc_uid || nd->nd_cred->cr_uid == 0)
5899 		return (0);
5900 	else
5901 		return (1);
5902 }
5903 
5904 /*
5905  * Calculate the lease expiry time.
5906  */
5907 static time_t
5908 nfsrv_leaseexpiry(void)
5909 {
5910 
5911 	if (NFSD_VNET(nfsrv_stablefirst).nsf_eograce > NFSD_MONOSEC)
5912 		return (NFSD_MONOSEC + 2 * (nfsrv_lease + NFSRV_LEASEDELTA));
5913 	return (NFSD_MONOSEC + nfsrv_lease + NFSRV_LEASEDELTA);
5914 }
5915 
5916 /*
5917  * Delay the delegation timeout as far as ls_delegtimelimit, as required.
5918  */
5919 static void
5920 nfsrv_delaydelegtimeout(struct nfsstate *stp)
5921 {
5922 
5923 	if ((stp->ls_flags & NFSLCK_DELEGRECALL) == 0)
5924 		return;
5925 
5926 	if ((stp->ls_delegtime + 15) > NFSD_MONOSEC &&
5927 	    stp->ls_delegtime < stp->ls_delegtimelimit) {
5928 		stp->ls_delegtime += nfsrv_lease;
5929 		if (stp->ls_delegtime > stp->ls_delegtimelimit)
5930 			stp->ls_delegtime = stp->ls_delegtimelimit;
5931 	}
5932 }
5933 
5934 /*
5935  * This function checks to see if there is any other state associated
5936  * with the openowner for this Open.
5937  * It returns 1 if there is no other state, 0 otherwise.
5938  */
5939 static int
5940 nfsrv_nootherstate(struct nfsstate *stp)
5941 {
5942 	struct nfsstate *tstp;
5943 
5944 	LIST_FOREACH(tstp, &stp->ls_openowner->ls_open, ls_list) {
5945 		if (tstp != stp || !LIST_EMPTY(&tstp->ls_lock))
5946 			return (0);
5947 	}
5948 	return (1);
5949 }
5950 
5951 /*
5952  * Create a list of lock deltas (changes to local byte range locking
5953  * that can be rolled back using the list) and apply the changes via
5954  * nfsvno_advlock(). Optionally, lock the list. It is expected that either
5955  * the rollback or update function will be called after this.
5956  * It returns an error (and rolls back, as required), if any nfsvno_advlock()
5957  * call fails. If it returns an error, it will unlock the list.
5958  */
5959 static int
5960 nfsrv_locallock(vnode_t vp, struct nfslockfile *lfp, int flags,
5961     uint64_t first, uint64_t end, struct nfslockconflict *cfp, NFSPROC_T *p)
5962 {
5963 	struct nfslock *lop, *nlop;
5964 	int error = 0;
5965 
5966 	/* Loop through the list of locks. */
5967 	lop = LIST_FIRST(&lfp->lf_locallock);
5968 	while (first < end && lop != NULL) {
5969 		nlop = LIST_NEXT(lop, lo_lckowner);
5970 		if (first >= lop->lo_end) {
5971 			/* not there yet */
5972 			lop = nlop;
5973 		} else if (first < lop->lo_first) {
5974 			/* new one starts before entry in list */
5975 			if (end <= lop->lo_first) {
5976 				/* no overlap between old and new */
5977 				error = nfsrv_dolocal(vp, lfp, flags,
5978 				    NFSLCK_UNLOCK, first, end, cfp, p);
5979 				if (error != 0)
5980 					break;
5981 				first = end;
5982 			} else {
5983 				/* handle fragment overlapped with new one */
5984 				error = nfsrv_dolocal(vp, lfp, flags,
5985 				    NFSLCK_UNLOCK, first, lop->lo_first, cfp,
5986 				    p);
5987 				if (error != 0)
5988 					break;
5989 				first = lop->lo_first;
5990 			}
5991 		} else {
5992 			/* new one overlaps this entry in list */
5993 			if (end <= lop->lo_end) {
5994 				/* overlaps all of new one */
5995 				error = nfsrv_dolocal(vp, lfp, flags,
5996 				    lop->lo_flags, first, end, cfp, p);
5997 				if (error != 0)
5998 					break;
5999 				first = end;
6000 			} else {
6001 				/* handle fragment overlapped with new one */
6002 				error = nfsrv_dolocal(vp, lfp, flags,
6003 				    lop->lo_flags, first, lop->lo_end, cfp, p);
6004 				if (error != 0)
6005 					break;
6006 				first = lop->lo_end;
6007 				lop = nlop;
6008 			}
6009 		}
6010 	}
6011 	if (first < end && error == 0)
6012 		/* handle fragment past end of list */
6013 		error = nfsrv_dolocal(vp, lfp, flags, NFSLCK_UNLOCK, first,
6014 		    end, cfp, p);
6015 
6016 	NFSEXITCODE(error);
6017 	return (error);
6018 }
6019 
6020 /*
6021  * Local lock unlock. Unlock all byte ranges that are no longer locked
6022  * by NFSv4. To do this, unlock any subranges of first-->end that
6023  * do not overlap with the byte ranges of any lock in the lfp->lf_lock
6024  * list. This list has all locks for the file held by other
6025  * <clientid, lockowner> tuples. The list is ordered by increasing
6026  * lo_first value, but may have entries that overlap each other, for
6027  * the case of read locks.
6028  */
6029 static void
6030 nfsrv_localunlock(vnode_t vp, struct nfslockfile *lfp, uint64_t init_first,
6031     uint64_t init_end, NFSPROC_T *p)
6032 {
6033 	struct nfslock *lop;
6034 	uint64_t first, end, prevfirst __unused;
6035 
6036 	first = init_first;
6037 	end = init_end;
6038 	while (first < init_end) {
6039 		/* Loop through all nfs locks, adjusting first and end */
6040 		prevfirst = 0;
6041 		LIST_FOREACH(lop, &lfp->lf_lock, lo_lckfile) {
6042 			KASSERT(prevfirst <= lop->lo_first,
6043 			    ("nfsv4 locks out of order"));
6044 			KASSERT(lop->lo_first < lop->lo_end,
6045 			    ("nfsv4 bogus lock"));
6046 			prevfirst = lop->lo_first;
6047 			if (first >= lop->lo_first &&
6048 			    first < lop->lo_end)
6049 				/*
6050 				 * Overlaps with initial part, so trim
6051 				 * off that initial part by moving first past
6052 				 * it.
6053 				 */
6054 				first = lop->lo_end;
6055 			else if (end > lop->lo_first &&
6056 			    lop->lo_first > first) {
6057 				/*
6058 				 * This lock defines the end of the
6059 				 * segment to unlock, so set end to the
6060 				 * start of it and break out of the loop.
6061 				 */
6062 				end = lop->lo_first;
6063 				break;
6064 			}
6065 			if (first >= end)
6066 				/*
6067 				 * There is no segment left to do, so
6068 				 * break out of this loop and then exit
6069 				 * the outer while() since first will be set
6070 				 * to end, which must equal init_end here.
6071 				 */
6072 				break;
6073 		}
6074 		if (first < end) {
6075 			/* Unlock this segment */
6076 			(void) nfsrv_dolocal(vp, lfp, NFSLCK_UNLOCK,
6077 			    NFSLCK_READ, first, end, NULL, p);
6078 			nfsrv_locallock_commit(lfp, NFSLCK_UNLOCK,
6079 			    first, end);
6080 		}
6081 		/*
6082 		 * Now move past this segment and look for any further
6083 		 * segment in the range, if there is one.
6084 		 */
6085 		first = end;
6086 		end = init_end;
6087 	}
6088 }
6089 
6090 /*
6091  * Do the local lock operation and update the rollback list, as required.
6092  * Perform the rollback and return the error if nfsvno_advlock() fails.
6093  */
6094 static int
6095 nfsrv_dolocal(vnode_t vp, struct nfslockfile *lfp, int flags, int oldflags,
6096     uint64_t first, uint64_t end, struct nfslockconflict *cfp, NFSPROC_T *p)
6097 {
6098 	struct nfsrollback *rlp;
6099 	int error = 0, ltype, oldltype;
6100 
6101 	if (flags & NFSLCK_WRITE)
6102 		ltype = F_WRLCK;
6103 	else if (flags & NFSLCK_READ)
6104 		ltype = F_RDLCK;
6105 	else
6106 		ltype = F_UNLCK;
6107 	if (oldflags & NFSLCK_WRITE)
6108 		oldltype = F_WRLCK;
6109 	else if (oldflags & NFSLCK_READ)
6110 		oldltype = F_RDLCK;
6111 	else
6112 		oldltype = F_UNLCK;
6113 	if (ltype == oldltype || (oldltype == F_WRLCK && ltype == F_RDLCK))
6114 		/* nothing to do */
6115 		goto out;
6116 	error = nfsvno_advlock(vp, ltype, first, end, p);
6117 	if (error != 0) {
6118 		if (cfp != NULL) {
6119 			cfp->cl_clientid.lval[0] = 0;
6120 			cfp->cl_clientid.lval[1] = 0;
6121 			cfp->cl_first = 0;
6122 			cfp->cl_end = NFS64BITSSET;
6123 			cfp->cl_flags = NFSLCK_WRITE;
6124 			cfp->cl_ownerlen = 5;
6125 			NFSBCOPY("LOCAL", cfp->cl_owner, 5);
6126 		}
6127 		nfsrv_locallock_rollback(vp, lfp, p);
6128 	} else if (ltype != F_UNLCK) {
6129 		rlp = malloc(sizeof (struct nfsrollback), M_NFSDROLLBACK,
6130 		    M_WAITOK);
6131 		rlp->rlck_first = first;
6132 		rlp->rlck_end = end;
6133 		rlp->rlck_type = oldltype;
6134 		LIST_INSERT_HEAD(&lfp->lf_rollback, rlp, rlck_list);
6135 	}
6136 
6137 out:
6138 	NFSEXITCODE(error);
6139 	return (error);
6140 }
6141 
6142 /*
6143  * Roll back local lock changes and free up the rollback list.
6144  */
6145 static void
6146 nfsrv_locallock_rollback(vnode_t vp, struct nfslockfile *lfp, NFSPROC_T *p)
6147 {
6148 	struct nfsrollback *rlp, *nrlp;
6149 
6150 	LIST_FOREACH_SAFE(rlp, &lfp->lf_rollback, rlck_list, nrlp) {
6151 		(void) nfsvno_advlock(vp, rlp->rlck_type, rlp->rlck_first,
6152 		    rlp->rlck_end, p);
6153 		free(rlp, M_NFSDROLLBACK);
6154 	}
6155 	LIST_INIT(&lfp->lf_rollback);
6156 }
6157 
6158 /*
6159  * Update local lock list and delete rollback list (ie now committed to the
6160  * local locks). Most of the work is done by the internal function.
6161  */
6162 static void
6163 nfsrv_locallock_commit(struct nfslockfile *lfp, int flags, uint64_t first,
6164     uint64_t end)
6165 {
6166 	struct nfsrollback *rlp, *nrlp;
6167 	struct nfslock *new_lop, *other_lop;
6168 
6169 	new_lop = malloc(sizeof (struct nfslock), M_NFSDLOCK, M_WAITOK);
6170 	if (flags & (NFSLCK_READ | NFSLCK_WRITE))
6171 		other_lop = malloc(sizeof (struct nfslock), M_NFSDLOCK,
6172 		    M_WAITOK);
6173 	else
6174 		other_lop = NULL;
6175 	new_lop->lo_flags = flags;
6176 	new_lop->lo_first = first;
6177 	new_lop->lo_end = end;
6178 	nfsrv_updatelock(NULL, &new_lop, &other_lop, lfp);
6179 	if (new_lop != NULL)
6180 		free(new_lop, M_NFSDLOCK);
6181 	if (other_lop != NULL)
6182 		free(other_lop, M_NFSDLOCK);
6183 
6184 	/* and get rid of the rollback list */
6185 	LIST_FOREACH_SAFE(rlp, &lfp->lf_rollback, rlck_list, nrlp)
6186 		free(rlp, M_NFSDROLLBACK);
6187 	LIST_INIT(&lfp->lf_rollback);
6188 }
6189 
6190 /*
6191  * Lock the struct nfslockfile for local lock updating.
6192  */
6193 static void
6194 nfsrv_locklf(struct nfslockfile *lfp)
6195 {
6196 	int gotlock;
6197 
6198 	/* lf_usecount ensures *lfp won't be free'd */
6199 	lfp->lf_usecount++;
6200 	do {
6201 		gotlock = nfsv4_lock(&lfp->lf_locallock_lck, 1, NULL,
6202 		    NFSSTATEMUTEXPTR, NULL);
6203 	} while (gotlock == 0);
6204 	lfp->lf_usecount--;
6205 }
6206 
6207 /*
6208  * Unlock the struct nfslockfile after local lock updating.
6209  */
6210 static void
6211 nfsrv_unlocklf(struct nfslockfile *lfp)
6212 {
6213 
6214 	nfsv4_unlock(&lfp->lf_locallock_lck, 0);
6215 }
6216 
6217 /*
6218  * Clear out all state for the NFSv4 server.
6219  * Must be called by a thread that can sleep when no nfsds are running.
6220  */
6221 void
6222 nfsrv_throwawayallstate(NFSPROC_T *p)
6223 {
6224 	struct nfsclient *clp, *nclp;
6225 	struct nfslockfile *lfp, *nlfp;
6226 	int i;
6227 
6228 	/*
6229 	 * For each client, clean out the state and then free the structure.
6230 	 */
6231 	for (i = 0; i < nfsrv_clienthashsize; i++) {
6232 		LIST_FOREACH_SAFE(clp, &NFSD_VNET(nfsclienthash)[i], lc_hash,
6233 		    nclp) {
6234 			nfsrv_cleanclient(clp, p, false, NULL);
6235 			nfsrv_freedeleglist(&clp->lc_deleg);
6236 			nfsrv_freedeleglist(&clp->lc_olddeleg);
6237 			free(clp->lc_stateid, M_NFSDCLIENT);
6238 			free(clp, M_NFSDCLIENT);
6239 		}
6240 	}
6241 
6242 	/*
6243 	 * Also, free up any remaining lock file structures.
6244 	 */
6245 	for (i = 0; i < nfsrv_lockhashsize; i++) {
6246 		LIST_FOREACH_SAFE(lfp, &NFSD_VNET(nfslockhash)[i], lf_hash,
6247 		    nlfp) {
6248 			printf("nfsd unload: fnd a lock file struct\n");
6249 			nfsrv_freenfslockfile(lfp);
6250 		}
6251 	}
6252 
6253 	/* And get rid of the deviceid structures and layouts. */
6254 	nfsrv_freealllayoutsanddevids();
6255 }
6256 
6257 /*
6258  * Check the sequence# for the session and slot provided as an argument.
6259  * Also, renew the lease if the session will return NFS_OK.
6260  */
6261 int
6262 nfsrv_checksequence(struct nfsrv_descript *nd, uint32_t sequenceid,
6263     uint32_t *highest_slotidp, uint32_t *target_highest_slotidp, int cache_this,
6264     uint32_t *sflagsp, NFSPROC_T *p)
6265 {
6266 	struct nfsdsession *sep;
6267 	struct nfssessionhash *shp;
6268 	int error;
6269 
6270 	shp = NFSSESSIONHASH(nd->nd_sessionid);
6271 	NFSLOCKSESSION(shp);
6272 	sep = nfsrv_findsession(nd->nd_sessionid);
6273 	if (sep == NULL) {
6274 		NFSUNLOCKSESSION(shp);
6275 		return (NFSERR_BADSESSION);
6276 	}
6277 	error = nfsv4_seqsession(sequenceid, nd->nd_slotid, *highest_slotidp,
6278 	    sep->sess_slots, NULL, NFSV4_SLOTS - 1);
6279 	if (error != 0) {
6280 		NFSUNLOCKSESSION(shp);
6281 		return (error);
6282 	}
6283 	if (cache_this != 0)
6284 		nd->nd_flag |= ND_SAVEREPLY;
6285 	/* Renew the lease. */
6286 	sep->sess_clp->lc_expiry = nfsrv_leaseexpiry();
6287 	nd->nd_clientid.qval = sep->sess_clp->lc_clientid.qval;
6288 	nd->nd_flag |= ND_IMPLIEDCLID;
6289 
6290 	/* Handle the SP4_MECH_CRED case for NFSv4.1/4.2. */
6291 	if ((sep->sess_clp->lc_flags & LCL_MACHCRED) != 0 &&
6292 	    (nd->nd_flag & (ND_GSSINTEGRITY | ND_GSSPRIVACY)) != 0 &&
6293 	    nd->nd_princlen == sep->sess_clp->lc_namelen &&
6294 	    !NFSBCMP(sep->sess_clp->lc_name, nd->nd_principal,
6295 	    nd->nd_princlen)) {
6296 		nd->nd_flag |= ND_MACHCRED;
6297 		NFSSET_OPBIT(&nd->nd_allowops, &sep->sess_clp->lc_allowops);
6298 	}
6299 
6300 	/* Save maximum request and reply sizes. */
6301 	nd->nd_maxreq = sep->sess_maxreq;
6302 	nd->nd_maxresp = sep->sess_maxresp;
6303 
6304 	*sflagsp = 0;
6305 	if (sep->sess_clp->lc_req.nr_client == NULL ||
6306 	    (sep->sess_clp->lc_flags & LCL_CBDOWN) != 0)
6307 		*sflagsp |= NFSV4SEQ_CBPATHDOWN;
6308 	NFSUNLOCKSESSION(shp);
6309 	if (error == NFSERR_EXPIRED) {
6310 		*sflagsp |= NFSV4SEQ_EXPIREDALLSTATEREVOKED;
6311 		error = 0;
6312 	} else if (error == NFSERR_ADMINREVOKED) {
6313 		*sflagsp |= NFSV4SEQ_ADMINSTATEREVOKED;
6314 		error = 0;
6315 	}
6316 	*highest_slotidp = *target_highest_slotidp = NFSV4_SLOTS - 1;
6317 	return (0);
6318 }
6319 
6320 /*
6321  * Check/set reclaim complete for this session/clientid.
6322  */
6323 int
6324 nfsrv_checkreclaimcomplete(struct nfsrv_descript *nd, int onefs)
6325 {
6326 	struct nfsdsession *sep;
6327 	struct nfssessionhash *shp;
6328 	int error = 0;
6329 
6330 	shp = NFSSESSIONHASH(nd->nd_sessionid);
6331 	NFSLOCKSTATE();
6332 	NFSLOCKSESSION(shp);
6333 	sep = nfsrv_findsession(nd->nd_sessionid);
6334 	if (sep == NULL) {
6335 		NFSUNLOCKSESSION(shp);
6336 		NFSUNLOCKSTATE();
6337 		return (NFSERR_BADSESSION);
6338 	}
6339 
6340 	if (onefs != 0)
6341 		sep->sess_clp->lc_flags |= LCL_RECLAIMONEFS;
6342 		/* Check to see if reclaim complete has already happened. */
6343 	else if ((sep->sess_clp->lc_flags & LCL_RECLAIMCOMPLETE) != 0)
6344 		error = NFSERR_COMPLETEALREADY;
6345 	else {
6346 		sep->sess_clp->lc_flags |= LCL_RECLAIMCOMPLETE;
6347 		nfsrv_markreclaim(sep->sess_clp);
6348 	}
6349 	NFSUNLOCKSESSION(shp);
6350 	NFSUNLOCKSTATE();
6351 	return (error);
6352 }
6353 
6354 /*
6355  * Cache the reply in a session slot.
6356  */
6357 void
6358 nfsrv_cache_session(struct nfsrv_descript *nd, struct mbuf **m)
6359 {
6360 	struct nfsdsession *sep;
6361 	struct nfssessionhash *shp;
6362 	char *buf, *cp;
6363 #ifdef INET
6364 	struct sockaddr_in *sin;
6365 #endif
6366 #ifdef INET6
6367 	struct sockaddr_in6 *sin6;
6368 #endif
6369 
6370 	shp = NFSSESSIONHASH(nd->nd_sessionid);
6371 	NFSLOCKSESSION(shp);
6372 	sep = nfsrv_findsession(nd->nd_sessionid);
6373 	if (sep == NULL) {
6374 		NFSUNLOCKSESSION(shp);
6375 		if ((NFSD_VNET(nfsrv_stablefirst).nsf_flags &
6376 		     NFSNSF_GRACEOVER) != 0) {
6377 			buf = malloc(INET6_ADDRSTRLEN, M_TEMP, M_WAITOK);
6378 			switch (nd->nd_nam->sa_family) {
6379 #ifdef INET
6380 			case AF_INET:
6381 				sin = (struct sockaddr_in *)nd->nd_nam;
6382 				cp = inet_ntop(sin->sin_family,
6383 				    &sin->sin_addr.s_addr, buf,
6384 				    INET6_ADDRSTRLEN);
6385 				break;
6386 #endif
6387 #ifdef INET6
6388 			case AF_INET6:
6389 				sin6 = (struct sockaddr_in6 *)nd->nd_nam;
6390 				cp = inet_ntop(sin6->sin6_family,
6391 				    &sin6->sin6_addr, buf, INET6_ADDRSTRLEN);
6392 				break;
6393 #endif
6394 			default:
6395 				cp = NULL;
6396 			}
6397 			if (cp != NULL)
6398 				printf("nfsrv_cache_session: no session "
6399 				    "IPaddr=%s, check NFS clients for unique "
6400 				    "/etc/hostid's\n", cp);
6401 			else
6402 				printf("nfsrv_cache_session: no session, "
6403 				    "check NFS clients for unique "
6404 				    "/etc/hostid's\n");
6405 			free(buf, M_TEMP);
6406 		}
6407 		m_freem(*m);
6408 		return;
6409 	}
6410 	nfsv4_seqsess_cacherep(nd->nd_slotid, sep->sess_slots, nd->nd_repstat,
6411 	    m);
6412 	NFSUNLOCKSESSION(shp);
6413 }
6414 
6415 /*
6416  * Search for a session that matches the sessionid.
6417  */
6418 static struct nfsdsession *
6419 nfsrv_findsession(uint8_t *sessionid)
6420 {
6421 	struct nfsdsession *sep;
6422 	struct nfssessionhash *shp;
6423 
6424 	shp = NFSSESSIONHASH(sessionid);
6425 	LIST_FOREACH(sep, &shp->list, sess_hash) {
6426 		if (!NFSBCMP(sessionid, sep->sess_sessionid, NFSX_V4SESSIONID))
6427 			break;
6428 	}
6429 	return (sep);
6430 }
6431 
6432 /*
6433  * Destroy a session.
6434  */
6435 int
6436 nfsrv_destroysession(struct nfsrv_descript *nd, uint8_t *sessionid)
6437 {
6438 	int error, igotlock, samesess;
6439 
6440 	samesess = 0;
6441 	if (!NFSBCMP(sessionid, nd->nd_sessionid, NFSX_V4SESSIONID) &&
6442 	    (nd->nd_flag & ND_HASSEQUENCE) != 0) {
6443 		samesess = 1;
6444 		if ((nd->nd_flag & ND_LASTOP) == 0)
6445 			return (NFSERR_BADSESSION);
6446 	}
6447 
6448 	/* Lock out other nfsd threads */
6449 	NFSLOCKV4ROOTMUTEX();
6450 	nfsv4_relref(&nfsv4rootfs_lock);
6451 	do {
6452 		igotlock = nfsv4_lock(&nfsv4rootfs_lock, 1, NULL,
6453 		    NFSV4ROOTLOCKMUTEXPTR, NULL);
6454 	} while (igotlock == 0);
6455 	NFSUNLOCKV4ROOTMUTEX();
6456 
6457 	error = nfsrv_freesession(nd, NULL, sessionid, false, NULL);
6458 	if (error == 0 && samesess != 0)
6459 		nd->nd_flag &= ~ND_HASSEQUENCE;
6460 
6461 	NFSLOCKV4ROOTMUTEX();
6462 	nfsv4_unlock(&nfsv4rootfs_lock, 1);
6463 	NFSUNLOCKV4ROOTMUTEX();
6464 	return (error);
6465 }
6466 
6467 /*
6468  * Bind a connection to a session.
6469  * For now, only certain variants are supported, since the current session
6470  * structure can only handle a single backchannel entry, which will be
6471  * applied to all connections if it is set.
6472  */
6473 int
6474 nfsrv_bindconnsess(struct nfsrv_descript *nd, uint8_t *sessionid, int *foreaftp)
6475 {
6476 	struct nfssessionhash *shp;
6477 	struct nfsdsession *sep;
6478 	struct nfsclient *clp;
6479 	SVCXPRT *savxprt;
6480 	int error;
6481 
6482 	error = 0;
6483 	savxprt = NULL;
6484 	shp = NFSSESSIONHASH(sessionid);
6485 	NFSLOCKSTATE();
6486 	NFSLOCKSESSION(shp);
6487 	sep = nfsrv_findsession(sessionid);
6488 	if (sep != NULL) {
6489 		clp = sep->sess_clp;
6490 		error = nfsrv_checkmachcred(NFSV4OP_BINDCONNTOSESS, nd, clp);
6491 		if (error != 0)
6492 			goto out;
6493 		if (*foreaftp == NFSCDFC4_BACK ||
6494 		    *foreaftp == NFSCDFC4_BACK_OR_BOTH ||
6495 		    *foreaftp == NFSCDFC4_FORE_OR_BOTH) {
6496 			/* Try to set up a backchannel. */
6497 			if (clp->lc_req.nr_client == NULL) {
6498 				NFSD_DEBUG(2, "nfsrv_bindconnsess: acquire "
6499 				    "backchannel\n");
6500 				clp->lc_req.nr_client = (struct __rpc_client *)
6501 				    clnt_bck_create(nd->nd_xprt->xp_socket,
6502 				    sep->sess_cbprogram, NFSV4_CBVERS);
6503 			}
6504 			if (clp->lc_req.nr_client != NULL) {
6505 				NFSD_DEBUG(2, "nfsrv_bindconnsess: set up "
6506 				    "backchannel\n");
6507 				savxprt = sep->sess_cbsess.nfsess_xprt;
6508 				SVC_ACQUIRE(nd->nd_xprt);
6509 				CLNT_ACQUIRE(clp->lc_req.nr_client);
6510 				nd->nd_xprt->xp_p2 = clp->lc_req.nr_client;
6511 				/* Disable idle timeout. */
6512 				nd->nd_xprt->xp_idletimeout = 0;
6513 				sep->sess_cbsess.nfsess_xprt = nd->nd_xprt;
6514 				sep->sess_crflags |= NFSV4CRSESS_CONNBACKCHAN;
6515 				clp->lc_flags |= LCL_DONEBINDCONN |
6516 				    LCL_NEEDSCBNULL;
6517 				clp->lc_flags &= ~LCL_CBDOWN;
6518 				if (*foreaftp == NFSCDFS4_BACK)
6519 					*foreaftp = NFSCDFS4_BACK;
6520 				else
6521 					*foreaftp = NFSCDFS4_BOTH;
6522 			} else if (*foreaftp != NFSCDFC4_BACK) {
6523 				NFSD_DEBUG(2, "nfsrv_bindconnsess: can't set "
6524 				    "up backchannel\n");
6525 				sep->sess_crflags &= ~NFSV4CRSESS_CONNBACKCHAN;
6526 				clp->lc_flags |= LCL_DONEBINDCONN;
6527 				*foreaftp = NFSCDFS4_FORE;
6528 			} else {
6529 				error = NFSERR_NOTSUPP;
6530 				printf("nfsrv_bindconnsess: Can't add "
6531 				    "backchannel\n");
6532 			}
6533 		} else {
6534 			NFSD_DEBUG(2, "nfsrv_bindconnsess: Set forechannel\n");
6535 			clp->lc_flags |= LCL_DONEBINDCONN;
6536 			*foreaftp = NFSCDFS4_FORE;
6537 		}
6538 	} else
6539 		error = NFSERR_BADSESSION;
6540 out:
6541 	NFSUNLOCKSESSION(shp);
6542 	NFSUNLOCKSTATE();
6543 	if (savxprt != NULL)
6544 		SVC_RELEASE(savxprt);
6545 	return (error);
6546 }
6547 
6548 /*
6549  * Free up a session structure.
6550  */
6551 static int
6552 nfsrv_freesession(struct nfsrv_descript *nd, struct nfsdsession *sep,
6553     uint8_t *sessionid, bool locked, SVCXPRT **old_xprtp)
6554 {
6555 	struct nfssessionhash *shp;
6556 	int i;
6557 
6558 	if (!locked)
6559 		NFSLOCKSTATE();
6560 	if (sep == NULL) {
6561 		shp = NFSSESSIONHASH(sessionid);
6562 		NFSLOCKSESSION(shp);
6563 		sep = nfsrv_findsession(sessionid);
6564 	} else {
6565 		shp = NFSSESSIONHASH(sep->sess_sessionid);
6566 		NFSLOCKSESSION(shp);
6567 	}
6568 	if (sep != NULL) {
6569 		/* Check for the SP4_MACH_CRED case. */
6570 		if (nd != NULL && nfsrv_checkmachcred(NFSV4OP_DESTROYSESSION,
6571 		    nd, sep->sess_clp) != 0) {
6572 			NFSUNLOCKSESSION(shp);
6573 			if (!locked)
6574 				NFSUNLOCKSTATE();
6575 			return (NFSERR_AUTHERR | AUTH_TOOWEAK);
6576 		}
6577 
6578 		sep->sess_refcnt--;
6579 		if (sep->sess_refcnt > 0) {
6580 			NFSUNLOCKSESSION(shp);
6581 			if (!locked)
6582 				NFSUNLOCKSTATE();
6583 			return (NFSERR_BACKCHANBUSY);
6584 		}
6585 		LIST_REMOVE(sep, sess_hash);
6586 		LIST_REMOVE(sep, sess_list);
6587 	}
6588 	NFSUNLOCKSESSION(shp);
6589 	if (!locked)
6590 		NFSUNLOCKSTATE();
6591 	if (sep == NULL)
6592 		return (NFSERR_BADSESSION);
6593 	for (i = 0; i < NFSV4_SLOTS; i++)
6594 		if (sep->sess_slots[i].nfssl_reply != NULL)
6595 			m_freem(sep->sess_slots[i].nfssl_reply);
6596 	if (!locked) {
6597 		if (sep->sess_cbsess.nfsess_xprt != NULL)
6598 			SVC_RELEASE(sep->sess_cbsess.nfsess_xprt);
6599 		if (old_xprtp != NULL)
6600 			*old_xprtp = NULL;
6601 	} else if (old_xprtp != NULL)
6602 		*old_xprtp = sep->sess_cbsess.nfsess_xprt;
6603 	free(sep, M_NFSDSESSION);
6604 	return (0);
6605 }
6606 
6607 /*
6608  * Free a stateid.
6609  * RFC5661 says that it should fail when there are associated opens, locks
6610  * or delegations. Since stateids represent opens, I don't see how you can
6611  * free an open stateid (it will be free'd when closed), so this function
6612  * only works for lock stateids (freeing the lock_owner) or delegations.
6613  */
6614 int
6615 nfsrv_freestateid(struct nfsrv_descript *nd, nfsv4stateid_t *stateidp,
6616     NFSPROC_T *p)
6617 {
6618 	struct nfsclient *clp;
6619 	struct nfsstate *stp;
6620 	int error;
6621 
6622 	NFSLOCKSTATE();
6623 	/*
6624 	 * Look up the stateid
6625 	 */
6626 	error = nfsrv_getclient((nfsquad_t)((u_quad_t)0), CLOPS_RENEW, &clp,
6627 	    NULL, (nfsquad_t)((u_quad_t)0), 0, nd, p);
6628 	if (error == 0) {
6629 		/* First, check for a delegation. */
6630 		LIST_FOREACH(stp, &clp->lc_deleg, ls_list) {
6631 			if (!NFSBCMP(stp->ls_stateid.other, stateidp->other,
6632 			    NFSX_STATEIDOTHER))
6633 				break;
6634 		}
6635 		if (stp != NULL) {
6636 			nfsrv_freedeleg(stp);
6637 			NFSUNLOCKSTATE();
6638 			return (error);
6639 		}
6640 	}
6641 	/* Not a delegation, try for a lock_owner. */
6642 	if (error == 0)
6643 		error = nfsrv_getstate(clp, stateidp, 0, &stp);
6644 	if (error == 0 && ((stp->ls_flags & (NFSLCK_OPEN | NFSLCK_DELEGREAD |
6645 	    NFSLCK_DELEGWRITE)) != 0 || (stp->ls_flags & NFSLCK_LOCK) == 0))
6646 		/* Not a lock_owner stateid. */
6647 		error = NFSERR_LOCKSHELD;
6648 	if (error == 0 && !LIST_EMPTY(&stp->ls_lock))
6649 		error = NFSERR_LOCKSHELD;
6650 	if (error == 0)
6651 		nfsrv_freelockowner(stp, NULL, 0, p);
6652 	NFSUNLOCKSTATE();
6653 	return (error);
6654 }
6655 
6656 /*
6657  * Test a stateid.
6658  */
6659 int
6660 nfsrv_teststateid(struct nfsrv_descript *nd, nfsv4stateid_t *stateidp,
6661     NFSPROC_T *p)
6662 {
6663 	struct nfsclient *clp;
6664 	struct nfsstate *stp;
6665 	int error;
6666 
6667 	NFSLOCKSTATE();
6668 	/*
6669 	 * Look up the stateid
6670 	 */
6671 	error = nfsrv_getclient((nfsquad_t)((u_quad_t)0), CLOPS_RENEW, &clp,
6672 	    NULL, (nfsquad_t)((u_quad_t)0), 0, nd, p);
6673 	if (error == 0)
6674 		error = nfsrv_getstate(clp, stateidp, 0, &stp);
6675 	if (error == 0 && stateidp->seqid != 0 &&
6676 	    SEQ_LT(stateidp->seqid, stp->ls_stateid.seqid))
6677 		error = NFSERR_OLDSTATEID;
6678 	NFSUNLOCKSTATE();
6679 	return (error);
6680 }
6681 
6682 /*
6683  * Generate the xdr for an NFSv4.1 CBSequence Operation.
6684  */
6685 static int
6686 nfsv4_setcbsequence(struct nfsrv_descript *nd, struct nfsclient *clp,
6687     int dont_replycache, struct nfsdsession **sepp, int *slotposp)
6688 {
6689 	struct nfsdsession *sep;
6690 	uint32_t *tl, slotseq = 0;
6691 	int maxslot;
6692 	uint8_t sessionid[NFSX_V4SESSIONID];
6693 	int error;
6694 
6695 	error = nfsv4_getcbsession(clp, sepp);
6696 	if (error != 0)
6697 		return (error);
6698 	sep = *sepp;
6699 	nfsv4_sequencelookup(NULL, &sep->sess_cbsess, slotposp, &maxslot,
6700 	    &slotseq, sessionid, true);
6701 	KASSERT(maxslot >= 0, ("nfsv4_setcbsequence neg maxslot"));
6702 
6703 	/* Build the Sequence arguments. */
6704 	NFSM_BUILD(tl, uint32_t *, NFSX_V4SESSIONID + 5 * NFSX_UNSIGNED);
6705 	bcopy(sessionid, tl, NFSX_V4SESSIONID);
6706 	tl += NFSX_V4SESSIONID / NFSX_UNSIGNED;
6707 	nd->nd_slotseq = tl;
6708 	nd->nd_slotid = *slotposp;
6709 	nd->nd_flag |= ND_HASSLOTID;
6710 	*tl++ = txdr_unsigned(slotseq);
6711 	*tl++ = txdr_unsigned(*slotposp);
6712 	*tl++ = txdr_unsigned(maxslot);
6713 	if (dont_replycache == 0)
6714 		*tl++ = newnfs_true;
6715 	else
6716 		*tl++ = newnfs_false;
6717 	*tl = 0;			/* No referring call list, for now. */
6718 	nd->nd_flag |= ND_HASSEQUENCE;
6719 	return (0);
6720 }
6721 
6722 /*
6723  * Get a session for the callback.
6724  */
6725 static int
6726 nfsv4_getcbsession(struct nfsclient *clp, struct nfsdsession **sepp)
6727 {
6728 	struct nfsdsession *sep;
6729 
6730 	NFSLOCKSTATE();
6731 	LIST_FOREACH(sep, &clp->lc_session, sess_list) {
6732 		if ((sep->sess_crflags & NFSV4CRSESS_CONNBACKCHAN) != 0)
6733 			break;
6734 	}
6735 	if (sep == NULL) {
6736 		NFSUNLOCKSTATE();
6737 		return (NFSERR_BADSESSION);
6738 	}
6739 	sep->sess_refcnt++;
6740 	*sepp = sep;
6741 	NFSUNLOCKSTATE();
6742 	return (0);
6743 }
6744 
6745 /*
6746  * Free up all backchannel xprts.  This needs to be done when the nfsd threads
6747  * exit, since those transports will all be going away.
6748  * This is only called after all the nfsd threads are done performing RPCs,
6749  * so locking shouldn't be an issue.
6750  */
6751 void
6752 nfsrv_freeallbackchannel_xprts(void)
6753 {
6754 	struct nfsdsession *sep;
6755 	struct nfsclient *clp;
6756 	SVCXPRT *xprt;
6757 	int i;
6758 
6759 	for (i = 0; i < nfsrv_clienthashsize; i++) {
6760 		LIST_FOREACH(clp, &NFSD_VNET(nfsclienthash)[i], lc_hash) {
6761 			LIST_FOREACH(sep, &clp->lc_session, sess_list) {
6762 				xprt = sep->sess_cbsess.nfsess_xprt;
6763 				sep->sess_cbsess.nfsess_xprt = NULL;
6764 				if (xprt != NULL)
6765 					SVC_RELEASE(xprt);
6766 			}
6767 		}
6768 	}
6769 }
6770 
6771 /*
6772  * Do a layout commit.  Actually just call nfsrv_updatemdsattr().
6773  * I have no idea if the rest of these arguments will ever be useful?
6774  */
6775 int
6776 nfsrv_layoutcommit(struct nfsrv_descript *nd, vnode_t vp, int layouttype,
6777     int hasnewoff, uint64_t newoff, uint64_t offset, uint64_t len,
6778     int hasnewmtime, struct timespec *newmtimep, int reclaim,
6779     nfsv4stateid_t *stateidp, int maxcnt, char *layp, int *hasnewsizep,
6780     uint64_t *newsizep, struct ucred *cred, NFSPROC_T *p)
6781 {
6782 	struct nfsvattr na;
6783 	int error;
6784 
6785 	error = nfsrv_updatemdsattr(vp, &na, p);
6786 	if (error == 0) {
6787 		*hasnewsizep = 1;
6788 		*newsizep = na.na_size;
6789 	}
6790 	return (error);
6791 }
6792 
6793 /*
6794  * Try and get a layout.
6795  */
6796 int
6797 nfsrv_layoutget(struct nfsrv_descript *nd, vnode_t vp, struct nfsexstuff *exp,
6798     int layouttype, int *iomode, uint64_t *offset, uint64_t *len,
6799     uint64_t minlen, nfsv4stateid_t *stateidp, int maxcnt, int *retonclose,
6800     int *layoutlenp, char *layp, struct ucred *cred, NFSPROC_T *p)
6801 {
6802 	struct nfslayouthash *lhyp;
6803 	struct nfslayout *lyp;
6804 	char *devid;
6805 	fhandle_t fh, *dsfhp;
6806 	int error, mirrorcnt;
6807 
6808 	if (nfsrv_devidcnt == 0)
6809 		return (NFSERR_UNKNLAYOUTTYPE);
6810 
6811 	if (*offset != 0)
6812 		printf("nfsrv_layoutget: off=%ju len=%ju\n", (uintmax_t)*offset,
6813 		    (uintmax_t)*len);
6814 	error = nfsvno_getfh(vp, &fh, p);
6815 	NFSD_DEBUG(4, "layoutget getfh=%d\n", error);
6816 	if (error != 0)
6817 		return (error);
6818 
6819 	/*
6820 	 * For now, all layouts are for entire files.
6821 	 * Only issue Read/Write layouts if requested for a non-readonly fs.
6822 	 */
6823 	if (NFSVNO_EXRDONLY(exp)) {
6824 		if (*iomode == NFSLAYOUTIOMODE_RW)
6825 			return (NFSERR_LAYOUTTRYLATER);
6826 		*iomode = NFSLAYOUTIOMODE_READ;
6827 	}
6828 	if (*iomode != NFSLAYOUTIOMODE_RW)
6829 		*iomode = NFSLAYOUTIOMODE_READ;
6830 
6831 	/*
6832 	 * Check to see if a write layout can be issued for this file.
6833 	 * This is used during mirror recovery to avoid RW layouts being
6834 	 * issued for a file while it is being copied to the recovered
6835 	 * mirror.
6836 	 */
6837 	if (*iomode == NFSLAYOUTIOMODE_RW && nfsrv_dontlayout(&fh) != 0)
6838 		return (NFSERR_LAYOUTTRYLATER);
6839 
6840 	*retonclose = 0;
6841 	*offset = 0;
6842 	*len = UINT64_MAX;
6843 
6844 	/* First, see if a layout already exists and return if found. */
6845 	lhyp = NFSLAYOUTHASH(&fh);
6846 	NFSLOCKLAYOUT(lhyp);
6847 	error = nfsrv_findlayout(&nd->nd_clientid, &fh, layouttype, p, &lyp);
6848 	NFSD_DEBUG(4, "layoutget findlay=%d\n", error);
6849 	/*
6850 	 * Not sure if the seqid must be the same, so I won't check it.
6851 	 */
6852 	if (error == 0 && (stateidp->other[0] != lyp->lay_stateid.other[0] ||
6853 	    stateidp->other[1] != lyp->lay_stateid.other[1] ||
6854 	    stateidp->other[2] != lyp->lay_stateid.other[2])) {
6855 		if ((lyp->lay_flags & NFSLAY_CALLB) == 0) {
6856 			NFSUNLOCKLAYOUT(lhyp);
6857 			NFSD_DEBUG(1, "ret bad stateid\n");
6858 			return (NFSERR_BADSTATEID);
6859 		}
6860 		/*
6861 		 * I believe we get here because there is a race between
6862 		 * the client processing the CBLAYOUTRECALL and the layout
6863 		 * being deleted here on the server.
6864 		 * The client has now done a LayoutGet with a non-layout
6865 		 * stateid, as it would when there is no layout.
6866 		 * As such, free this layout and set error == NFSERR_BADSTATEID
6867 		 * so the code below will create a new layout structure as
6868 		 * would happen if no layout was found.
6869 		 * "lyp" will be set before being used below, but set it NULL
6870 		 * as a safety belt.
6871 		 */
6872 		nfsrv_freelayout(&lhyp->list, lyp);
6873 		lyp = NULL;
6874 		error = NFSERR_BADSTATEID;
6875 	}
6876 	if (error == 0) {
6877 		if (lyp->lay_layoutlen > maxcnt) {
6878 			NFSUNLOCKLAYOUT(lhyp);
6879 			NFSD_DEBUG(1, "ret layout too small\n");
6880 			return (NFSERR_TOOSMALL);
6881 		}
6882 		if (*iomode == NFSLAYOUTIOMODE_RW) {
6883 			if ((lyp->lay_flags & NFSLAY_NOSPC) != 0) {
6884 				NFSUNLOCKLAYOUT(lhyp);
6885 				NFSD_DEBUG(1, "ret layout nospace\n");
6886 				return (NFSERR_NOSPC);
6887 			}
6888 			lyp->lay_flags |= NFSLAY_RW;
6889 		} else
6890 			lyp->lay_flags |= NFSLAY_READ;
6891 		NFSBCOPY(lyp->lay_xdr, layp, lyp->lay_layoutlen);
6892 		*layoutlenp = lyp->lay_layoutlen;
6893 		if (++lyp->lay_stateid.seqid == 0)
6894 			lyp->lay_stateid.seqid = 1;
6895 		stateidp->seqid = lyp->lay_stateid.seqid;
6896 		NFSUNLOCKLAYOUT(lhyp);
6897 		NFSD_DEBUG(4, "ret fnd layout\n");
6898 		return (0);
6899 	}
6900 	NFSUNLOCKLAYOUT(lhyp);
6901 
6902 	/* Find the device id and file handle. */
6903 	dsfhp = malloc(sizeof(fhandle_t) * NFSDEV_MAXMIRRORS, M_TEMP, M_WAITOK);
6904 	devid = malloc(NFSX_V4DEVICEID * NFSDEV_MAXMIRRORS, M_TEMP, M_WAITOK);
6905 	error = nfsrv_dsgetdevandfh(vp, p, &mirrorcnt, dsfhp, devid);
6906 	NFSD_DEBUG(4, "layoutget devandfh=%d\n", error);
6907 	if (error == 0) {
6908 		if (layouttype == NFSLAYOUT_NFSV4_1_FILES) {
6909 			if (NFSX_V4FILELAYOUT > maxcnt)
6910 				error = NFSERR_TOOSMALL;
6911 			else
6912 				lyp = nfsrv_filelayout(nd, *iomode, &fh, dsfhp,
6913 				    devid, vp->v_mount->mnt_stat.f_fsid);
6914 		} else {
6915 			if (NFSX_V4FLEXLAYOUT(mirrorcnt) > maxcnt)
6916 				error = NFSERR_TOOSMALL;
6917 			else
6918 				lyp = nfsrv_flexlayout(nd, *iomode, mirrorcnt,
6919 				    &fh, dsfhp, devid,
6920 				    vp->v_mount->mnt_stat.f_fsid);
6921 		}
6922 	}
6923 	free(dsfhp, M_TEMP);
6924 	free(devid, M_TEMP);
6925 	if (error != 0)
6926 		return (error);
6927 
6928 	/*
6929 	 * Now, add this layout to the list.
6930 	 */
6931 	error = nfsrv_addlayout(nd, &lyp, stateidp, layp, layoutlenp, p);
6932 	NFSD_DEBUG(4, "layoutget addl=%d\n", error);
6933 	/*
6934 	 * The lyp will be set to NULL by nfsrv_addlayout() if it
6935 	 * linked the new structure into the lists.
6936 	 */
6937 	free(lyp, M_NFSDSTATE);
6938 	return (error);
6939 }
6940 
6941 /*
6942  * Generate a File Layout.
6943  */
6944 static struct nfslayout *
6945 nfsrv_filelayout(struct nfsrv_descript *nd, int iomode, fhandle_t *fhp,
6946     fhandle_t *dsfhp, char *devid, fsid_t fs)
6947 {
6948 	uint32_t *tl;
6949 	struct nfslayout *lyp;
6950 	uint64_t pattern_offset;
6951 
6952 	lyp = malloc(sizeof(struct nfslayout) + NFSX_V4FILELAYOUT, M_NFSDSTATE,
6953 	    M_WAITOK | M_ZERO);
6954 	lyp->lay_type = NFSLAYOUT_NFSV4_1_FILES;
6955 	if (iomode == NFSLAYOUTIOMODE_RW)
6956 		lyp->lay_flags = NFSLAY_RW;
6957 	else
6958 		lyp->lay_flags = NFSLAY_READ;
6959 	NFSBCOPY(fhp, &lyp->lay_fh, sizeof(*fhp));
6960 	lyp->lay_clientid.qval = nd->nd_clientid.qval;
6961 	lyp->lay_fsid = fs;
6962 	NFSBCOPY(devid, lyp->lay_deviceid, NFSX_V4DEVICEID);
6963 
6964 	/* Fill in the xdr for the files layout. */
6965 	tl = (uint32_t *)lyp->lay_xdr;
6966 	NFSBCOPY(devid, tl, NFSX_V4DEVICEID);		/* Device ID. */
6967 	tl += (NFSX_V4DEVICEID / NFSX_UNSIGNED);
6968 
6969 	/* Set the stripe size to the maximum I/O size. */
6970 	*tl++ = txdr_unsigned(nfs_srvmaxio & NFSFLAYUTIL_STRIPE_MASK);
6971 	*tl++ = 0;					/* 1st stripe index. */
6972 	pattern_offset = 0;
6973 	txdr_hyper(pattern_offset, tl); tl += 2;	/* Pattern offset. */
6974 	*tl++ = txdr_unsigned(1);			/* 1 file handle. */
6975 	*tl++ = txdr_unsigned(NFSX_V4PNFSFH);
6976 	NFSBCOPY(dsfhp, tl, sizeof(*dsfhp));
6977 	lyp->lay_layoutlen = NFSX_V4FILELAYOUT;
6978 	return (lyp);
6979 }
6980 
6981 #define	FLEX_OWNERID	"999"
6982 #define	FLEX_UID0	"0"
6983 /*
6984  * Generate a Flex File Layout.
6985  * The FLEX_OWNERID can be any string of 3 decimal digits. Although this
6986  * string goes on the wire, it isn't supposed to be used by the client,
6987  * since this server uses tight coupling.
6988  * Although not recommended by the spec., if vfs.nfsd.flexlinuxhack=1 use
6989  * a string of "0". This works around the Linux Flex File Layout driver bug
6990  * which uses the synthetic uid/gid strings for the "tightly coupled" case.
6991  */
6992 static struct nfslayout *
6993 nfsrv_flexlayout(struct nfsrv_descript *nd, int iomode, int mirrorcnt,
6994     fhandle_t *fhp, fhandle_t *dsfhp, char *devid, fsid_t fs)
6995 {
6996 	uint32_t *tl;
6997 	struct nfslayout *lyp;
6998 	uint64_t lenval;
6999 	int i;
7000 
7001 	lyp = malloc(sizeof(struct nfslayout) + NFSX_V4FLEXLAYOUT(mirrorcnt),
7002 	    M_NFSDSTATE, M_WAITOK | M_ZERO);
7003 	lyp->lay_type = NFSLAYOUT_FLEXFILE;
7004 	if (iomode == NFSLAYOUTIOMODE_RW)
7005 		lyp->lay_flags = NFSLAY_RW;
7006 	else
7007 		lyp->lay_flags = NFSLAY_READ;
7008 	NFSBCOPY(fhp, &lyp->lay_fh, sizeof(*fhp));
7009 	lyp->lay_clientid.qval = nd->nd_clientid.qval;
7010 	lyp->lay_fsid = fs;
7011 	lyp->lay_mirrorcnt = mirrorcnt;
7012 	NFSBCOPY(devid, lyp->lay_deviceid, NFSX_V4DEVICEID);
7013 
7014 	/* Fill in the xdr for the files layout. */
7015 	tl = (uint32_t *)lyp->lay_xdr;
7016 	lenval = 0;
7017 	txdr_hyper(lenval, tl); tl += 2;		/* Stripe unit. */
7018 	*tl++ = txdr_unsigned(mirrorcnt);		/* # of mirrors. */
7019 	for (i = 0; i < mirrorcnt; i++) {
7020 		*tl++ = txdr_unsigned(1);		/* One stripe. */
7021 		NFSBCOPY(devid, tl, NFSX_V4DEVICEID);	/* Device ID. */
7022 		tl += (NFSX_V4DEVICEID / NFSX_UNSIGNED);
7023 		devid += NFSX_V4DEVICEID;
7024 		*tl++ = txdr_unsigned(1);		/* Efficiency. */
7025 		*tl++ = 0;				/* Proxy Stateid. */
7026 		*tl++ = 0x55555555;
7027 		*tl++ = 0x55555555;
7028 		*tl++ = 0x55555555;
7029 		*tl++ = txdr_unsigned(1);		/* 1 file handle. */
7030 		*tl++ = txdr_unsigned(NFSX_V4PNFSFH);
7031 		NFSBCOPY(dsfhp, tl, sizeof(*dsfhp));
7032 		tl += (NFSM_RNDUP(NFSX_V4PNFSFH) / NFSX_UNSIGNED);
7033 		dsfhp++;
7034 		if (nfsrv_flexlinuxhack != 0) {
7035 			*tl++ = txdr_unsigned(strlen(FLEX_UID0));
7036 			*tl = 0;		/* 0 pad string. */
7037 			NFSBCOPY(FLEX_UID0, tl++, strlen(FLEX_UID0));
7038 			*tl++ = txdr_unsigned(strlen(FLEX_UID0));
7039 			*tl = 0;		/* 0 pad string. */
7040 			NFSBCOPY(FLEX_UID0, tl++, strlen(FLEX_UID0));
7041 		} else {
7042 			*tl++ = txdr_unsigned(strlen(FLEX_OWNERID));
7043 			NFSBCOPY(FLEX_OWNERID, tl++, NFSX_UNSIGNED);
7044 			*tl++ = txdr_unsigned(strlen(FLEX_OWNERID));
7045 			NFSBCOPY(FLEX_OWNERID, tl++, NFSX_UNSIGNED);
7046 		}
7047 	}
7048 	*tl++ = txdr_unsigned(0);		/* ff_flags. */
7049 	*tl = txdr_unsigned(60);		/* Status interval hint. */
7050 	lyp->lay_layoutlen = NFSX_V4FLEXLAYOUT(mirrorcnt);
7051 	return (lyp);
7052 }
7053 
7054 /*
7055  * Parse and process Flex File errors returned via LayoutReturn.
7056  */
7057 static void
7058 nfsrv_flexlayouterr(struct nfsrv_descript *nd, uint32_t *layp, int maxcnt,
7059     NFSPROC_T *p)
7060 {
7061 	uint32_t *tl;
7062 	int cnt, errcnt, i, j, opnum, stat;
7063 	char devid[NFSX_V4DEVICEID];
7064 
7065 	tl = layp;
7066 	maxcnt -= NFSX_UNSIGNED;
7067 	if (maxcnt > 0)
7068 		cnt = fxdr_unsigned(int, *tl++);
7069 	else
7070 		cnt = 0;
7071 	NFSD_DEBUG(4, "flexlayouterr cnt=%d\n", cnt);
7072 	for (i = 0; i < cnt; i++) {
7073 		maxcnt -= NFSX_STATEID + 2 * NFSX_HYPER +
7074 		    NFSX_UNSIGNED;
7075 		if (maxcnt <= 0)
7076 			break;
7077 		/* Skip offset, length and stateid for now. */
7078 		tl += (4 + NFSX_STATEID / NFSX_UNSIGNED);
7079 		errcnt = fxdr_unsigned(int, *tl++);
7080 		NFSD_DEBUG(4, "flexlayouterr errcnt=%d\n", errcnt);
7081 		for (j = 0; j < errcnt; j++) {
7082 			maxcnt -= NFSX_V4DEVICEID + 2 * NFSX_UNSIGNED;
7083 			if (maxcnt < 0)
7084 				break;
7085 			NFSBCOPY(tl, devid, NFSX_V4DEVICEID);
7086 			tl += (NFSX_V4DEVICEID / NFSX_UNSIGNED);
7087 			stat = fxdr_unsigned(int, *tl++);
7088 			opnum = fxdr_unsigned(int, *tl++);
7089 			NFSD_DEBUG(4, "flexlayouterr op=%d stat=%d\n", opnum,
7090 			    stat);
7091 			/*
7092 			 * Except for NFSERR_ACCES, NFSERR_STALE and
7093 			 * NFSERR_NOSPC errors, disable the mirror.
7094 			 */
7095 			if (stat != NFSERR_ACCES && stat != NFSERR_STALE &&
7096 			    stat != NFSERR_NOSPC)
7097 				nfsrv_delds(devid, p);
7098 
7099 			/* For NFSERR_NOSPC, mark all devids and layouts. */
7100 			if (stat == NFSERR_NOSPC)
7101 				nfsrv_marknospc(devid, true);
7102 		}
7103 	}
7104 }
7105 
7106 /*
7107  * This function removes all flex file layouts which has a mirror with
7108  * a device id that matches the argument.
7109  * Called when the DS represented by the device id has failed.
7110  */
7111 void
7112 nfsrv_flexmirrordel(char *devid, NFSPROC_T *p)
7113 {
7114 	uint32_t *tl;
7115 	struct nfslayout *lyp, *nlyp;
7116 	struct nfslayouthash *lhyp;
7117 	struct nfslayouthead loclyp;
7118 	int i, j;
7119 
7120 	NFSD_DEBUG(4, "flexmirrordel\n");
7121 	/* Move all layouts found onto a local list. */
7122 	TAILQ_INIT(&loclyp);
7123 	for (i = 0; i < nfsrv_layouthashsize; i++) {
7124 		lhyp = &nfslayouthash[i];
7125 		NFSLOCKLAYOUT(lhyp);
7126 		TAILQ_FOREACH_SAFE(lyp, &lhyp->list, lay_list, nlyp) {
7127 			if (lyp->lay_type == NFSLAYOUT_FLEXFILE &&
7128 			    lyp->lay_mirrorcnt > 1) {
7129 				NFSD_DEBUG(4, "possible match\n");
7130 				tl = lyp->lay_xdr;
7131 				tl += 3;
7132 				for (j = 0; j < lyp->lay_mirrorcnt; j++) {
7133 					tl++;
7134 					if (NFSBCMP(devid, tl, NFSX_V4DEVICEID)
7135 					    == 0) {
7136 						/* Found one. */
7137 						NFSD_DEBUG(4, "fnd one\n");
7138 						TAILQ_REMOVE(&lhyp->list, lyp,
7139 						    lay_list);
7140 						TAILQ_INSERT_HEAD(&loclyp, lyp,
7141 						    lay_list);
7142 						break;
7143 					}
7144 					tl += (NFSX_V4DEVICEID / NFSX_UNSIGNED +
7145 					    NFSM_RNDUP(NFSX_V4PNFSFH) /
7146 					    NFSX_UNSIGNED + 11 * NFSX_UNSIGNED);
7147 				}
7148 			}
7149 		}
7150 		NFSUNLOCKLAYOUT(lhyp);
7151 	}
7152 
7153 	/* Now, try to do a Layout recall for each one found. */
7154 	TAILQ_FOREACH_SAFE(lyp, &loclyp, lay_list, nlyp) {
7155 		NFSD_DEBUG(4, "do layout recall\n");
7156 		/*
7157 		 * The layout stateid.seqid needs to be incremented
7158 		 * before doing a LAYOUT_RECALL callback.
7159 		 */
7160 		if (++lyp->lay_stateid.seqid == 0)
7161 			lyp->lay_stateid.seqid = 1;
7162 		nfsrv_recalllayout(lyp->lay_clientid, &lyp->lay_stateid,
7163 		    &lyp->lay_fh, lyp, 1, lyp->lay_type, p);
7164 		nfsrv_freelayout(&loclyp, lyp);
7165 	}
7166 }
7167 
7168 /*
7169  * Do a recall callback to the client for this layout.
7170  */
7171 static int
7172 nfsrv_recalllayout(nfsquad_t clid, nfsv4stateid_t *stateidp, fhandle_t *fhp,
7173     struct nfslayout *lyp, int changed, int laytype, NFSPROC_T *p)
7174 {
7175 	struct nfsclient *clp;
7176 	int error;
7177 
7178 	NFSD_DEBUG(4, "nfsrv_recalllayout\n");
7179 	error = nfsrv_getclient(clid, 0, &clp, NULL, (nfsquad_t)((u_quad_t)0),
7180 	    0, NULL, p);
7181 	NFSD_DEBUG(4, "aft nfsrv_getclient=%d\n", error);
7182 	if (error != 0) {
7183 		printf("nfsrv_recalllayout: getclient err=%d\n", error);
7184 		return (error);
7185 	}
7186 	if ((clp->lc_flags & LCL_NFSV41) != 0) {
7187 		error = nfsrv_docallback(clp, NFSV4OP_CBLAYOUTRECALL,
7188 		    stateidp, changed, fhp, NULL, NULL, laytype, p);
7189 		/* If lyp != NULL, handle an error return here. */
7190 		if (error != 0 && lyp != NULL) {
7191 			NFSDRECALLLOCK();
7192 			/*
7193 			 * Mark it returned, since no layout recall
7194 			 * has been done.
7195 			 * All errors seem to be non-recoverable, although
7196 			 * NFSERR_NOMATCHLAYOUT is a normal event.
7197 			 */
7198 			if ((lyp->lay_flags & NFSLAY_RECALL) != 0) {
7199 				lyp->lay_flags |= NFSLAY_RETURNED;
7200 				wakeup(lyp);
7201 			}
7202 			NFSDRECALLUNLOCK();
7203 			if (error != NFSERR_NOMATCHLAYOUT)
7204 				printf("nfsrv_recalllayout: err=%d\n", error);
7205 		}
7206 	} else
7207 		printf("nfsrv_recalllayout: clp not NFSv4.1\n");
7208 	return (error);
7209 }
7210 
7211 /*
7212  * Find a layout to recall when we exceed our high water mark.
7213  */
7214 void
7215 nfsrv_recalloldlayout(NFSPROC_T *p)
7216 {
7217 	struct nfslayouthash *lhyp;
7218 	struct nfslayout *lyp;
7219 	nfsquad_t clientid;
7220 	nfsv4stateid_t stateid;
7221 	fhandle_t fh;
7222 	int error, laytype = 0, ret;
7223 
7224 	lhyp = &nfslayouthash[arc4random() % nfsrv_layouthashsize];
7225 	NFSLOCKLAYOUT(lhyp);
7226 	TAILQ_FOREACH_REVERSE(lyp, &lhyp->list, nfslayouthead, lay_list) {
7227 		if ((lyp->lay_flags & NFSLAY_CALLB) == 0) {
7228 			lyp->lay_flags |= NFSLAY_CALLB;
7229 			/*
7230 			 * The layout stateid.seqid needs to be incremented
7231 			 * before doing a LAYOUT_RECALL callback.
7232 			 */
7233 			if (++lyp->lay_stateid.seqid == 0)
7234 				lyp->lay_stateid.seqid = 1;
7235 			clientid = lyp->lay_clientid;
7236 			stateid = lyp->lay_stateid;
7237 			NFSBCOPY(&lyp->lay_fh, &fh, sizeof(fh));
7238 			laytype = lyp->lay_type;
7239 			break;
7240 		}
7241 	}
7242 	NFSUNLOCKLAYOUT(lhyp);
7243 	if (lyp != NULL) {
7244 		error = nfsrv_recalllayout(clientid, &stateid, &fh, NULL, 0,
7245 		    laytype, p);
7246 		if (error != 0 && error != NFSERR_NOMATCHLAYOUT)
7247 			NFSD_DEBUG(4, "recallold=%d\n", error);
7248 		if (error != 0) {
7249 			NFSLOCKLAYOUT(lhyp);
7250 			/*
7251 			 * Since the hash list was unlocked, we need to
7252 			 * find it again.
7253 			 */
7254 			ret = nfsrv_findlayout(&clientid, &fh, laytype, p,
7255 			    &lyp);
7256 			if (ret == 0 &&
7257 			    (lyp->lay_flags & NFSLAY_CALLB) != 0 &&
7258 			    lyp->lay_stateid.other[0] == stateid.other[0] &&
7259 			    lyp->lay_stateid.other[1] == stateid.other[1] &&
7260 			    lyp->lay_stateid.other[2] == stateid.other[2]) {
7261 				/*
7262 				 * The client no longer knows this layout, so
7263 				 * it can be free'd now.
7264 				 */
7265 				if (error == NFSERR_NOMATCHLAYOUT)
7266 					nfsrv_freelayout(&lhyp->list, lyp);
7267 				else {
7268 					/*
7269 					 * Leave it to be tried later by
7270 					 * clearing NFSLAY_CALLB and moving
7271 					 * it to the head of the list, so it
7272 					 * won't be tried again for a while.
7273 					 */
7274 					lyp->lay_flags &= ~NFSLAY_CALLB;
7275 					TAILQ_REMOVE(&lhyp->list, lyp,
7276 					    lay_list);
7277 					TAILQ_INSERT_HEAD(&lhyp->list, lyp,
7278 					    lay_list);
7279 				}
7280 			}
7281 			NFSUNLOCKLAYOUT(lhyp);
7282 		}
7283 	}
7284 }
7285 
7286 /*
7287  * Try and return layout(s).
7288  */
7289 int
7290 nfsrv_layoutreturn(struct nfsrv_descript *nd, vnode_t vp,
7291     int layouttype, int iomode, uint64_t offset, uint64_t len, int reclaim,
7292     int kind, nfsv4stateid_t *stateidp, int maxcnt, uint32_t *layp, int *fndp,
7293     struct ucred *cred, NFSPROC_T *p)
7294 {
7295 	struct nfsvattr na;
7296 	struct nfslayouthash *lhyp;
7297 	struct nfslayout *lyp;
7298 	fhandle_t fh;
7299 	int error = 0;
7300 
7301 	*fndp = 0;
7302 	if (kind == NFSV4LAYOUTRET_FILE) {
7303 		error = nfsvno_getfh(vp, &fh, p);
7304 		if (error == 0) {
7305 			error = nfsrv_updatemdsattr(vp, &na, p);
7306 			if (error != 0)
7307 				printf("nfsrv_layoutreturn: updatemdsattr"
7308 				    " failed=%d\n", error);
7309 		}
7310 		if (error == 0) {
7311 			if (reclaim == newnfs_true) {
7312 				error = nfsrv_checkgrace(NULL, NULL,
7313 				    NFSLCK_RECLAIM);
7314 				if (error != NFSERR_NOGRACE)
7315 					error = 0;
7316 				return (error);
7317 			}
7318 			lhyp = NFSLAYOUTHASH(&fh);
7319 			NFSDRECALLLOCK();
7320 			NFSLOCKLAYOUT(lhyp);
7321 			error = nfsrv_findlayout(&nd->nd_clientid, &fh,
7322 			    layouttype, p, &lyp);
7323 			NFSD_DEBUG(4, "layoutret findlay=%d\n", error);
7324 			if (error == 0 &&
7325 			    stateidp->other[0] == lyp->lay_stateid.other[0] &&
7326 			    stateidp->other[1] == lyp->lay_stateid.other[1] &&
7327 			    stateidp->other[2] == lyp->lay_stateid.other[2]) {
7328 				NFSD_DEBUG(4, "nfsrv_layoutreturn: stateid %d"
7329 				    " %x %x %x laystateid %d %x %x %x"
7330 				    " off=%ju len=%ju flgs=0x%x\n",
7331 				    stateidp->seqid, stateidp->other[0],
7332 				    stateidp->other[1], stateidp->other[2],
7333 				    lyp->lay_stateid.seqid,
7334 				    lyp->lay_stateid.other[0],
7335 				    lyp->lay_stateid.other[1],
7336 				    lyp->lay_stateid.other[2],
7337 				    (uintmax_t)offset, (uintmax_t)len,
7338 				    lyp->lay_flags);
7339 				if (++lyp->lay_stateid.seqid == 0)
7340 					lyp->lay_stateid.seqid = 1;
7341 				stateidp->seqid = lyp->lay_stateid.seqid;
7342 				if (offset == 0 && len == UINT64_MAX) {
7343 					if ((iomode & NFSLAYOUTIOMODE_READ) !=
7344 					    0)
7345 						lyp->lay_flags &= ~NFSLAY_READ;
7346 					if ((iomode & NFSLAYOUTIOMODE_RW) != 0)
7347 						lyp->lay_flags &= ~NFSLAY_RW;
7348 					if ((lyp->lay_flags & (NFSLAY_READ |
7349 					    NFSLAY_RW)) == 0)
7350 						nfsrv_freelayout(&lhyp->list,
7351 						    lyp);
7352 					else
7353 						*fndp = 1;
7354 				} else
7355 					*fndp = 1;
7356 			}
7357 			NFSUNLOCKLAYOUT(lhyp);
7358 			/* Search the nfsrv_recalllist for a match. */
7359 			TAILQ_FOREACH(lyp, &nfsrv_recalllisthead, lay_list) {
7360 				if (NFSBCMP(&lyp->lay_fh, &fh,
7361 				    sizeof(fh)) == 0 &&
7362 				    lyp->lay_clientid.qval ==
7363 				    nd->nd_clientid.qval &&
7364 				    stateidp->other[0] ==
7365 				    lyp->lay_stateid.other[0] &&
7366 				    stateidp->other[1] ==
7367 				    lyp->lay_stateid.other[1] &&
7368 				    stateidp->other[2] ==
7369 				    lyp->lay_stateid.other[2]) {
7370 					lyp->lay_flags |= NFSLAY_RETURNED;
7371 					wakeup(lyp);
7372 					error = 0;
7373 				}
7374 			}
7375 			NFSDRECALLUNLOCK();
7376 		}
7377 		if (layouttype == NFSLAYOUT_FLEXFILE && layp != NULL)
7378 			nfsrv_flexlayouterr(nd, layp, maxcnt, p);
7379 	} else if (kind == NFSV4LAYOUTRET_FSID)
7380 		nfsrv_freelayouts(&nd->nd_clientid,
7381 		    &vp->v_mount->mnt_stat.f_fsid, layouttype, iomode);
7382 	else if (kind == NFSV4LAYOUTRET_ALL)
7383 		nfsrv_freelayouts(&nd->nd_clientid, NULL, layouttype, iomode);
7384 	else
7385 		error = NFSERR_INVAL;
7386 	if (error == -1)
7387 		error = 0;
7388 	return (error);
7389 }
7390 
7391 /*
7392  * Look for an existing layout.
7393  */
7394 static int
7395 nfsrv_findlayout(nfsquad_t *clientidp, fhandle_t *fhp, int laytype,
7396     NFSPROC_T *p, struct nfslayout **lypp)
7397 {
7398 	struct nfslayouthash *lhyp;
7399 	struct nfslayout *lyp;
7400 	int ret;
7401 
7402 	*lypp = NULL;
7403 	ret = 0;
7404 	lhyp = NFSLAYOUTHASH(fhp);
7405 	TAILQ_FOREACH(lyp, &lhyp->list, lay_list) {
7406 		if (NFSBCMP(&lyp->lay_fh, fhp, sizeof(*fhp)) == 0 &&
7407 		    lyp->lay_clientid.qval == clientidp->qval &&
7408 		    lyp->lay_type == laytype)
7409 			break;
7410 	}
7411 	if (lyp != NULL)
7412 		*lypp = lyp;
7413 	else
7414 		ret = -1;
7415 	return (ret);
7416 }
7417 
7418 /*
7419  * Add the new layout, as required.
7420  */
7421 static int
7422 nfsrv_addlayout(struct nfsrv_descript *nd, struct nfslayout **lypp,
7423     nfsv4stateid_t *stateidp, char *layp, int *layoutlenp, NFSPROC_T *p)
7424 {
7425 	struct nfsclient *clp;
7426 	struct nfslayouthash *lhyp;
7427 	struct nfslayout *lyp, *nlyp;
7428 	fhandle_t *fhp;
7429 	int error;
7430 
7431 	KASSERT((nd->nd_flag & ND_IMPLIEDCLID) != 0,
7432 	    ("nfsrv_layoutget: no nd_clientid\n"));
7433 	lyp = *lypp;
7434 	fhp = &lyp->lay_fh;
7435 	NFSLOCKSTATE();
7436 	error = nfsrv_getclient((nfsquad_t)((u_quad_t)0), CLOPS_RENEW, &clp,
7437 	    NULL, (nfsquad_t)((u_quad_t)0), 0, nd, p);
7438 	if (error != 0) {
7439 		NFSUNLOCKSTATE();
7440 		return (error);
7441 	}
7442 	lyp->lay_stateid.seqid = stateidp->seqid = 1;
7443 	lyp->lay_stateid.other[0] = stateidp->other[0] =
7444 	    clp->lc_clientid.lval[0];
7445 	lyp->lay_stateid.other[1] = stateidp->other[1] =
7446 	    clp->lc_clientid.lval[1];
7447 	lyp->lay_stateid.other[2] = stateidp->other[2] =
7448 	    nfsrv_nextstateindex(clp);
7449 	NFSUNLOCKSTATE();
7450 
7451 	lhyp = NFSLAYOUTHASH(fhp);
7452 	NFSLOCKLAYOUT(lhyp);
7453 	TAILQ_FOREACH(nlyp, &lhyp->list, lay_list) {
7454 		if (NFSBCMP(&nlyp->lay_fh, fhp, sizeof(*fhp)) == 0 &&
7455 		    nlyp->lay_clientid.qval == nd->nd_clientid.qval)
7456 			break;
7457 	}
7458 	if (nlyp != NULL) {
7459 		/* A layout already exists, so use it. */
7460 		nlyp->lay_flags |= (lyp->lay_flags & (NFSLAY_READ | NFSLAY_RW));
7461 		NFSBCOPY(nlyp->lay_xdr, layp, nlyp->lay_layoutlen);
7462 		*layoutlenp = nlyp->lay_layoutlen;
7463 		if (++nlyp->lay_stateid.seqid == 0)
7464 			nlyp->lay_stateid.seqid = 1;
7465 		stateidp->seqid = nlyp->lay_stateid.seqid;
7466 		stateidp->other[0] = nlyp->lay_stateid.other[0];
7467 		stateidp->other[1] = nlyp->lay_stateid.other[1];
7468 		stateidp->other[2] = nlyp->lay_stateid.other[2];
7469 		NFSUNLOCKLAYOUT(lhyp);
7470 		return (0);
7471 	}
7472 
7473 	/* Insert the new layout in the lists. */
7474 	*lypp = NULL;
7475 	atomic_add_int(&nfsrv_layoutcnt, 1);
7476 	NFSD_VNET(nfsstatsv1_p)->srvlayouts++;
7477 	NFSBCOPY(lyp->lay_xdr, layp, lyp->lay_layoutlen);
7478 	*layoutlenp = lyp->lay_layoutlen;
7479 	TAILQ_INSERT_HEAD(&lhyp->list, lyp, lay_list);
7480 	NFSUNLOCKLAYOUT(lhyp);
7481 	return (0);
7482 }
7483 
7484 /*
7485  * Get the devinfo for a deviceid.
7486  */
7487 int
7488 nfsrv_getdevinfo(char *devid, int layouttype, uint32_t *maxcnt,
7489     uint32_t *notify, int *devaddrlen, char **devaddr)
7490 {
7491 	struct nfsdevice *ds;
7492 
7493 	if ((layouttype != NFSLAYOUT_NFSV4_1_FILES && layouttype !=
7494 	     NFSLAYOUT_FLEXFILE) ||
7495 	    (nfsrv_maxpnfsmirror > 1 && layouttype == NFSLAYOUT_NFSV4_1_FILES))
7496 		return (NFSERR_UNKNLAYOUTTYPE);
7497 
7498 	/*
7499 	 * Now, search for the device id.  Note that the structures won't go
7500 	 * away, but the order changes in the list.  As such, the lock only
7501 	 * needs to be held during the search through the list.
7502 	 */
7503 	NFSDDSLOCK();
7504 	TAILQ_FOREACH(ds, &nfsrv_devidhead, nfsdev_list) {
7505 		if (NFSBCMP(devid, ds->nfsdev_deviceid, NFSX_V4DEVICEID) == 0 &&
7506 		    ds->nfsdev_nmp != NULL)
7507 			break;
7508 	}
7509 	NFSDDSUNLOCK();
7510 	if (ds == NULL)
7511 		return (NFSERR_NOENT);
7512 
7513 	/* If the correct nfsdev_XXXXaddrlen is > 0, we have the device info. */
7514 	*devaddrlen = 0;
7515 	if (layouttype == NFSLAYOUT_NFSV4_1_FILES) {
7516 		*devaddrlen = ds->nfsdev_fileaddrlen;
7517 		*devaddr = ds->nfsdev_fileaddr;
7518 	} else if (layouttype == NFSLAYOUT_FLEXFILE) {
7519 		*devaddrlen = ds->nfsdev_flexaddrlen;
7520 		*devaddr = ds->nfsdev_flexaddr;
7521 	}
7522 	if (*devaddrlen == 0)
7523 		return (NFSERR_UNKNLAYOUTTYPE);
7524 
7525 	/*
7526 	 * The XDR overhead is 3 unsigned values: layout_type,
7527 	 * length_of_address and notify bitmap.
7528 	 * If the notify array is changed to not all zeros, the
7529 	 * count of unsigned values must be increased.
7530 	 */
7531 	if (*maxcnt > 0 && *maxcnt < NFSM_RNDUP(*devaddrlen) +
7532 	    3 * NFSX_UNSIGNED) {
7533 		*maxcnt = NFSM_RNDUP(*devaddrlen) + 3 * NFSX_UNSIGNED;
7534 		return (NFSERR_TOOSMALL);
7535 	}
7536 	return (0);
7537 }
7538 
7539 /*
7540  * Free a list of layout state structures.
7541  */
7542 static void
7543 nfsrv_freelayoutlist(nfsquad_t clientid)
7544 {
7545 	struct nfslayouthash *lhyp;
7546 	struct nfslayout *lyp, *nlyp;
7547 	int i;
7548 
7549 	for (i = 0; i < nfsrv_layouthashsize; i++) {
7550 		lhyp = &nfslayouthash[i];
7551 		NFSLOCKLAYOUT(lhyp);
7552 		TAILQ_FOREACH_SAFE(lyp, &lhyp->list, lay_list, nlyp) {
7553 			if (lyp->lay_clientid.qval == clientid.qval)
7554 				nfsrv_freelayout(&lhyp->list, lyp);
7555 		}
7556 		NFSUNLOCKLAYOUT(lhyp);
7557 	}
7558 }
7559 
7560 /*
7561  * Free up a layout.
7562  */
7563 static void
7564 nfsrv_freelayout(struct nfslayouthead *lhp, struct nfslayout *lyp)
7565 {
7566 
7567 	NFSD_DEBUG(4, "Freelayout=%p\n", lyp);
7568 	atomic_add_int(&nfsrv_layoutcnt, -1);
7569 	NFSD_VNET(nfsstatsv1_p)->srvlayouts--;
7570 	TAILQ_REMOVE(lhp, lyp, lay_list);
7571 	free(lyp, M_NFSDSTATE);
7572 }
7573 
7574 /*
7575  * Free up a device id.
7576  */
7577 void
7578 nfsrv_freeonedevid(struct nfsdevice *ds)
7579 {
7580 	int i;
7581 
7582 	atomic_add_int(&nfsrv_devidcnt, -1);
7583 	vrele(ds->nfsdev_dvp);
7584 	for (i = 0; i < nfsrv_dsdirsize; i++)
7585 		if (ds->nfsdev_dsdir[i] != NULL)
7586 			vrele(ds->nfsdev_dsdir[i]);
7587 	free(ds->nfsdev_fileaddr, M_NFSDSTATE);
7588 	free(ds->nfsdev_flexaddr, M_NFSDSTATE);
7589 	free(ds->nfsdev_host, M_NFSDSTATE);
7590 	free(ds, M_NFSDSTATE);
7591 }
7592 
7593 /*
7594  * Free up a device id and its mirrors.
7595  */
7596 static void
7597 nfsrv_freedevid(struct nfsdevice *ds)
7598 {
7599 
7600 	TAILQ_REMOVE(&nfsrv_devidhead, ds, nfsdev_list);
7601 	nfsrv_freeonedevid(ds);
7602 }
7603 
7604 /*
7605  * Free all layouts and device ids.
7606  * Done when the nfsd threads are shut down since there may be a new
7607  * modified device id list created when the nfsd is restarted.
7608  */
7609 void
7610 nfsrv_freealllayoutsanddevids(void)
7611 {
7612 	struct nfsdontlist *mrp, *nmrp;
7613 	struct nfslayout *lyp, *nlyp;
7614 
7615 	/* Get rid of the deviceid structures. */
7616 	nfsrv_freealldevids();
7617 	TAILQ_INIT(&nfsrv_devidhead);
7618 	nfsrv_devidcnt = 0;
7619 
7620 	/* Get rid of all layouts. */
7621 	nfsrv_freealllayouts();
7622 
7623 	/* Get rid of any nfsdontlist entries. */
7624 	LIST_FOREACH_SAFE(mrp, &nfsrv_dontlisthead, nfsmr_list, nmrp)
7625 		free(mrp, M_NFSDSTATE);
7626 	LIST_INIT(&nfsrv_dontlisthead);
7627 	nfsrv_dontlistlen = 0;
7628 
7629 	/* Free layouts in the recall list. */
7630 	TAILQ_FOREACH_SAFE(lyp, &nfsrv_recalllisthead, lay_list, nlyp)
7631 		nfsrv_freelayout(&nfsrv_recalllisthead, lyp);
7632 	TAILQ_INIT(&nfsrv_recalllisthead);
7633 }
7634 
7635 /*
7636  * Free layouts that match the arguments.
7637  */
7638 static void
7639 nfsrv_freelayouts(nfsquad_t *clid, fsid_t *fs, int laytype, int iomode)
7640 {
7641 	struct nfslayouthash *lhyp;
7642 	struct nfslayout *lyp, *nlyp;
7643 	int i;
7644 
7645 	for (i = 0; i < nfsrv_layouthashsize; i++) {
7646 		lhyp = &nfslayouthash[i];
7647 		NFSLOCKLAYOUT(lhyp);
7648 		TAILQ_FOREACH_SAFE(lyp, &lhyp->list, lay_list, nlyp) {
7649 			if (clid->qval != lyp->lay_clientid.qval)
7650 				continue;
7651 			if (fs != NULL && fsidcmp(fs, &lyp->lay_fsid) != 0)
7652 				continue;
7653 			if (laytype != lyp->lay_type)
7654 				continue;
7655 			if ((iomode & NFSLAYOUTIOMODE_READ) != 0)
7656 				lyp->lay_flags &= ~NFSLAY_READ;
7657 			if ((iomode & NFSLAYOUTIOMODE_RW) != 0)
7658 				lyp->lay_flags &= ~NFSLAY_RW;
7659 			if ((lyp->lay_flags & (NFSLAY_READ | NFSLAY_RW)) == 0)
7660 				nfsrv_freelayout(&lhyp->list, lyp);
7661 		}
7662 		NFSUNLOCKLAYOUT(lhyp);
7663 	}
7664 }
7665 
7666 /*
7667  * Free all layouts for the argument file.
7668  */
7669 void
7670 nfsrv_freefilelayouts(fhandle_t *fhp)
7671 {
7672 	struct nfslayouthash *lhyp;
7673 	struct nfslayout *lyp, *nlyp;
7674 
7675 	lhyp = NFSLAYOUTHASH(fhp);
7676 	NFSLOCKLAYOUT(lhyp);
7677 	TAILQ_FOREACH_SAFE(lyp, &lhyp->list, lay_list, nlyp) {
7678 		if (NFSBCMP(&lyp->lay_fh, fhp, sizeof(*fhp)) == 0)
7679 			nfsrv_freelayout(&lhyp->list, lyp);
7680 	}
7681 	NFSUNLOCKLAYOUT(lhyp);
7682 }
7683 
7684 /*
7685  * Free all layouts.
7686  */
7687 static void
7688 nfsrv_freealllayouts(void)
7689 {
7690 	struct nfslayouthash *lhyp;
7691 	struct nfslayout *lyp, *nlyp;
7692 	int i;
7693 
7694 	for (i = 0; i < nfsrv_layouthashsize; i++) {
7695 		lhyp = &nfslayouthash[i];
7696 		NFSLOCKLAYOUT(lhyp);
7697 		TAILQ_FOREACH_SAFE(lyp, &lhyp->list, lay_list, nlyp)
7698 			nfsrv_freelayout(&lhyp->list, lyp);
7699 		NFSUNLOCKLAYOUT(lhyp);
7700 	}
7701 }
7702 
7703 /*
7704  * Look up the mount path for the DS server.
7705  */
7706 static int
7707 nfsrv_setdsserver(char *dspathp, char *mdspathp, NFSPROC_T *p,
7708     struct nfsdevice **dsp)
7709 {
7710 	struct nameidata nd;
7711 	struct nfsdevice *ds;
7712 	struct mount *mp;
7713 	int error, i;
7714 	char *dsdirpath;
7715 	size_t dsdirsize;
7716 
7717 	NFSD_DEBUG(4, "setdssrv path=%s\n", dspathp);
7718 	*dsp = NULL;
7719 	if (jailed(p->td_ucred)) {
7720 		printf("A pNFS nfsd cannot run in a jail\n");
7721 		return (EPERM);
7722 	}
7723 	NDINIT(&nd, LOOKUP, FOLLOW | LOCKSHARED | LOCKLEAF, UIO_SYSSPACE,
7724 	    dspathp);
7725 	error = namei(&nd);
7726 	NFSD_DEBUG(4, "lookup=%d\n", error);
7727 	if (error != 0)
7728 		return (error);
7729 	if (nd.ni_vp->v_type != VDIR) {
7730 		vput(nd.ni_vp);
7731 		NFSD_DEBUG(4, "dspath not dir\n");
7732 		return (ENOTDIR);
7733 	}
7734 	if (strcmp(nd.ni_vp->v_mount->mnt_vfc->vfc_name, "nfs") != 0) {
7735 		vput(nd.ni_vp);
7736 		NFSD_DEBUG(4, "dspath not an NFS mount\n");
7737 		return (ENXIO);
7738 	}
7739 
7740 	/*
7741 	 * Allocate a DS server structure with the NFS mounted directory
7742 	 * vnode reference counted, so that a non-forced dismount will
7743 	 * fail with EBUSY.
7744 	 * This structure is always linked into the list, even if an error
7745 	 * is being returned.  The caller will free the entire list upon
7746 	 * an error return.
7747 	 */
7748 	*dsp = ds = malloc(sizeof(*ds) + nfsrv_dsdirsize * sizeof(vnode_t),
7749 	    M_NFSDSTATE, M_WAITOK | M_ZERO);
7750 	ds->nfsdev_dvp = nd.ni_vp;
7751 	ds->nfsdev_nmp = VFSTONFS(nd.ni_vp->v_mount);
7752 	NFSVOPUNLOCK(nd.ni_vp);
7753 
7754 	dsdirsize = strlen(dspathp) + 16;
7755 	dsdirpath = malloc(dsdirsize, M_TEMP, M_WAITOK);
7756 	/* Now, create the DS directory structures. */
7757 	for (i = 0; i < nfsrv_dsdirsize; i++) {
7758 		snprintf(dsdirpath, dsdirsize, "%s/ds%d", dspathp, i);
7759 		NDINIT(&nd, LOOKUP, FOLLOW | LOCKSHARED | LOCKLEAF,
7760 		    UIO_SYSSPACE, dsdirpath);
7761 		error = namei(&nd);
7762 		NFSD_DEBUG(4, "dsdirpath=%s lookup=%d\n", dsdirpath, error);
7763 		if (error != 0)
7764 			break;
7765 		if (nd.ni_vp->v_type != VDIR) {
7766 			vput(nd.ni_vp);
7767 			error = ENOTDIR;
7768 			NFSD_DEBUG(4, "dsdirpath not a VDIR\n");
7769 			break;
7770 		}
7771 		if (strcmp(nd.ni_vp->v_mount->mnt_vfc->vfc_name, "nfs") != 0) {
7772 			vput(nd.ni_vp);
7773 			error = ENXIO;
7774 			NFSD_DEBUG(4, "dsdirpath not an NFS mount\n");
7775 			break;
7776 		}
7777 		ds->nfsdev_dsdir[i] = nd.ni_vp;
7778 		NFSVOPUNLOCK(nd.ni_vp);
7779 	}
7780 	free(dsdirpath, M_TEMP);
7781 
7782 	if (strlen(mdspathp) > 0) {
7783 		/*
7784 		 * This DS stores file for a specific MDS exported file
7785 		 * system.
7786 		 */
7787 		NDINIT(&nd, LOOKUP, FOLLOW | LOCKSHARED | LOCKLEAF,
7788 		    UIO_SYSSPACE, mdspathp);
7789 		error = namei(&nd);
7790 		NFSD_DEBUG(4, "mds lookup=%d\n", error);
7791 		if (error != 0)
7792 			goto out;
7793 		if (nd.ni_vp->v_type != VDIR) {
7794 			vput(nd.ni_vp);
7795 			error = ENOTDIR;
7796 			NFSD_DEBUG(4, "mdspath not dir\n");
7797 			goto out;
7798 		}
7799 		mp = nd.ni_vp->v_mount;
7800 		if ((mp->mnt_flag & MNT_EXPORTED) == 0) {
7801 			vput(nd.ni_vp);
7802 			error = ENXIO;
7803 			NFSD_DEBUG(4, "mdspath not an exported fs\n");
7804 			goto out;
7805 		}
7806 		ds->nfsdev_mdsfsid = mp->mnt_stat.f_fsid;
7807 		ds->nfsdev_mdsisset = 1;
7808 		vput(nd.ni_vp);
7809 	}
7810 
7811 out:
7812 	TAILQ_INSERT_TAIL(&nfsrv_devidhead, ds, nfsdev_list);
7813 	atomic_add_int(&nfsrv_devidcnt, 1);
7814 	return (error);
7815 }
7816 
7817 /*
7818  * Look up the mount path for the DS server and delete it.
7819  */
7820 int
7821 nfsrv_deldsserver(int op, char *dspathp, NFSPROC_T *p)
7822 {
7823 	struct mount *mp;
7824 	struct nfsmount *nmp;
7825 	struct nfsdevice *ds;
7826 	int error;
7827 
7828 	NFSD_DEBUG(4, "deldssrv path=%s\n", dspathp);
7829 	/*
7830 	 * Search for the path in the mount list.  Avoid looking the path
7831 	 * up, since this mount point may be hung, with associated locked
7832 	 * vnodes, etc.
7833 	 * Set NFSMNTP_CANCELRPCS so that any forced dismount will be blocked
7834 	 * until this completes.
7835 	 * As noted in the man page, this should be done before any forced
7836 	 * dismount on the mount point, but at least the handshake on
7837 	 * NFSMNTP_CANCELRPCS should make it safe.
7838 	 */
7839 	error = 0;
7840 	ds = NULL;
7841 	nmp = NULL;
7842 	mtx_lock(&mountlist_mtx);
7843 	TAILQ_FOREACH(mp, &mountlist, mnt_list) {
7844 		if (strcmp(mp->mnt_stat.f_mntonname, dspathp) == 0 &&
7845 		    strcmp(mp->mnt_stat.f_fstypename, "nfs") == 0 &&
7846 		    mp->mnt_data != NULL) {
7847 			nmp = VFSTONFS(mp);
7848 			NFSLOCKMNT(nmp);
7849 			if ((nmp->nm_privflag & (NFSMNTP_FORCEDISM |
7850 			     NFSMNTP_CANCELRPCS)) == 0) {
7851 				nmp->nm_privflag |= NFSMNTP_CANCELRPCS;
7852 				NFSUNLOCKMNT(nmp);
7853 			} else {
7854 				NFSUNLOCKMNT(nmp);
7855 				nmp = NULL;
7856 			}
7857 			break;
7858 		}
7859 	}
7860 	mtx_unlock(&mountlist_mtx);
7861 
7862 	if (nmp != NULL) {
7863 		ds = nfsrv_deldsnmp(op, nmp, p);
7864 		NFSD_DEBUG(4, "deldsnmp=%p\n", ds);
7865 		if (ds != NULL) {
7866 			nfsrv_killrpcs(nmp);
7867 			NFSD_DEBUG(4, "aft killrpcs\n");
7868 		} else
7869 			error = ENXIO;
7870 		NFSLOCKMNT(nmp);
7871 		nmp->nm_privflag &= ~NFSMNTP_CANCELRPCS;
7872 		wakeup(nmp);
7873 		NFSUNLOCKMNT(nmp);
7874 	} else
7875 		error = EINVAL;
7876 	return (error);
7877 }
7878 
7879 /*
7880  * Search for and remove a DS entry which matches the "nmp" argument.
7881  * The nfsdevice structure pointer is returned so that the caller can
7882  * free it via nfsrv_freeonedevid().
7883  * For the forced case, do not try to do LayoutRecalls, since the server
7884  * must be shut down now anyhow.
7885  */
7886 struct nfsdevice *
7887 nfsrv_deldsnmp(int op, struct nfsmount *nmp, NFSPROC_T *p)
7888 {
7889 	struct nfsdevice *fndds;
7890 
7891 	NFSD_DEBUG(4, "deldsdvp\n");
7892 	NFSDDSLOCK();
7893 	if (op == PNFSDOP_FORCEDELDS)
7894 		fndds = nfsv4_findmirror(nmp);
7895 	else
7896 		fndds = nfsrv_findmirroredds(nmp);
7897 	if (fndds != NULL)
7898 		nfsrv_deleteds(fndds);
7899 	NFSDDSUNLOCK();
7900 	if (fndds != NULL) {
7901 		if (op != PNFSDOP_FORCEDELDS)
7902 			nfsrv_flexmirrordel(fndds->nfsdev_deviceid, p);
7903 		printf("pNFS server: mirror %s failed\n", fndds->nfsdev_host);
7904 	}
7905 	return (fndds);
7906 }
7907 
7908 /*
7909  * Similar to nfsrv_deldsnmp(), except that the DS is indicated by deviceid.
7910  * This function also calls nfsrv_killrpcs() to unblock RPCs on the mount
7911  * point.
7912  * Also, returns an error instead of the nfsdevice found.
7913  */
7914 int
7915 nfsrv_delds(char *devid, NFSPROC_T *p)
7916 {
7917 	struct nfsdevice *ds, *fndds;
7918 	struct nfsmount *nmp;
7919 	int fndmirror;
7920 
7921 	NFSD_DEBUG(4, "delds\n");
7922 	/*
7923 	 * Search the DS server list for a match with devid.
7924 	 * Remove the DS entry if found and there is a mirror.
7925 	 */
7926 	fndds = NULL;
7927 	nmp = NULL;
7928 	fndmirror = 0;
7929 	NFSDDSLOCK();
7930 	TAILQ_FOREACH(ds, &nfsrv_devidhead, nfsdev_list) {
7931 		if (NFSBCMP(ds->nfsdev_deviceid, devid, NFSX_V4DEVICEID) == 0 &&
7932 		    ds->nfsdev_nmp != NULL) {
7933 			NFSD_DEBUG(4, "fnd main ds\n");
7934 			fndds = ds;
7935 			break;
7936 		}
7937 	}
7938 	if (fndds == NULL) {
7939 		NFSDDSUNLOCK();
7940 		return (ENXIO);
7941 	}
7942 	if (fndds->nfsdev_mdsisset == 0 && nfsrv_faildscnt > 0)
7943 		fndmirror = 1;
7944 	else if (fndds->nfsdev_mdsisset != 0) {
7945 		/* For the fsid is set case, search for a mirror. */
7946 		TAILQ_FOREACH(ds, &nfsrv_devidhead, nfsdev_list) {
7947 			if (ds != fndds && ds->nfsdev_nmp != NULL &&
7948 			    ds->nfsdev_mdsisset != 0 &&
7949 			    fsidcmp(&ds->nfsdev_mdsfsid,
7950 			    &fndds->nfsdev_mdsfsid) == 0) {
7951 				fndmirror = 1;
7952 				break;
7953 			}
7954 		}
7955 	}
7956 	if (fndmirror != 0) {
7957 		nmp = fndds->nfsdev_nmp;
7958 		NFSLOCKMNT(nmp);
7959 		if ((nmp->nm_privflag & (NFSMNTP_FORCEDISM |
7960 		     NFSMNTP_CANCELRPCS)) == 0) {
7961 			nmp->nm_privflag |= NFSMNTP_CANCELRPCS;
7962 			NFSUNLOCKMNT(nmp);
7963 			nfsrv_deleteds(fndds);
7964 		} else {
7965 			NFSUNLOCKMNT(nmp);
7966 			nmp = NULL;
7967 		}
7968 	}
7969 	NFSDDSUNLOCK();
7970 	if (nmp != NULL) {
7971 		nfsrv_flexmirrordel(fndds->nfsdev_deviceid, p);
7972 		printf("pNFS server: mirror %s failed\n", fndds->nfsdev_host);
7973 		nfsrv_killrpcs(nmp);
7974 		NFSLOCKMNT(nmp);
7975 		nmp->nm_privflag &= ~NFSMNTP_CANCELRPCS;
7976 		wakeup(nmp);
7977 		NFSUNLOCKMNT(nmp);
7978 		return (0);
7979 	}
7980 	return (ENXIO);
7981 }
7982 
7983 /*
7984  * Mark a DS as disabled by setting nfsdev_nmp = NULL.
7985  */
7986 static void
7987 nfsrv_deleteds(struct nfsdevice *fndds)
7988 {
7989 
7990 	NFSD_DEBUG(4, "deleteds: deleting a mirror\n");
7991 	fndds->nfsdev_nmp = NULL;
7992 	if (fndds->nfsdev_mdsisset == 0)
7993 		nfsrv_faildscnt--;
7994 }
7995 
7996 /*
7997  * Fill in the addr structures for the File and Flex File layouts.
7998  */
7999 static void
8000 nfsrv_allocdevid(struct nfsdevice *ds, char *addr, char *dnshost)
8001 {
8002 	uint32_t *tl;
8003 	char *netprot;
8004 	int addrlen;
8005 	static uint64_t new_devid = 0;
8006 
8007 	if (strchr(addr, ':') != NULL)
8008 		netprot = "tcp6";
8009 	else
8010 		netprot = "tcp";
8011 
8012 	/* Fill in the device id. */
8013 	NFSBCOPY(&nfsdev_time, ds->nfsdev_deviceid, sizeof(nfsdev_time));
8014 	new_devid++;
8015 	NFSBCOPY(&new_devid, &ds->nfsdev_deviceid[sizeof(nfsdev_time)],
8016 	    sizeof(new_devid));
8017 
8018 	/*
8019 	 * Fill in the file addr (actually the nfsv4_file_layout_ds_addr4
8020 	 * as defined in RFC5661) in XDR.
8021 	 */
8022 	addrlen = NFSM_RNDUP(strlen(addr)) + NFSM_RNDUP(strlen(netprot)) +
8023 	    6 * NFSX_UNSIGNED;
8024 	NFSD_DEBUG(4, "hn=%s addr=%s netprot=%s\n", dnshost, addr, netprot);
8025 	ds->nfsdev_fileaddrlen = addrlen;
8026 	tl = malloc(addrlen, M_NFSDSTATE, M_WAITOK | M_ZERO);
8027 	ds->nfsdev_fileaddr = (char *)tl;
8028 	*tl++ = txdr_unsigned(1);		/* One stripe with index 0. */
8029 	*tl++ = 0;
8030 	*tl++ = txdr_unsigned(1);		/* One multipath list */
8031 	*tl++ = txdr_unsigned(1);		/* with one entry in it. */
8032 	/* The netaddr for this one entry. */
8033 	*tl++ = txdr_unsigned(strlen(netprot));
8034 	NFSBCOPY(netprot, tl, strlen(netprot));
8035 	tl += (NFSM_RNDUP(strlen(netprot)) / NFSX_UNSIGNED);
8036 	*tl++ = txdr_unsigned(strlen(addr));
8037 	NFSBCOPY(addr, tl, strlen(addr));
8038 
8039 	/*
8040 	 * Fill in the flex file addr (actually the ff_device_addr4
8041 	 * as defined for Flexible File Layout) in XDR.
8042 	 */
8043 	addrlen = NFSM_RNDUP(strlen(addr)) + NFSM_RNDUP(strlen(netprot)) +
8044 	    14 * NFSX_UNSIGNED;
8045 	ds->nfsdev_flexaddrlen = addrlen;
8046 	tl = malloc(addrlen, M_NFSDSTATE, M_WAITOK | M_ZERO);
8047 	ds->nfsdev_flexaddr = (char *)tl;
8048 	*tl++ = txdr_unsigned(1);		/* One multipath entry. */
8049 	/* The netaddr for this one entry. */
8050 	*tl++ = txdr_unsigned(strlen(netprot));
8051 	NFSBCOPY(netprot, tl, strlen(netprot));
8052 	tl += (NFSM_RNDUP(strlen(netprot)) / NFSX_UNSIGNED);
8053 	*tl++ = txdr_unsigned(strlen(addr));
8054 	NFSBCOPY(addr, tl, strlen(addr));
8055 	tl += (NFSM_RNDUP(strlen(addr)) / NFSX_UNSIGNED);
8056 	*tl++ = txdr_unsigned(2);		/* Two NFS Versions. */
8057 	*tl++ = txdr_unsigned(NFS_VER4);	/* NFSv4. */
8058 	*tl++ = txdr_unsigned(NFSV42_MINORVERSION); /* Minor version 2. */
8059 	*tl++ = txdr_unsigned(nfs_srvmaxio);	/* DS max rsize. */
8060 	*tl++ = txdr_unsigned(nfs_srvmaxio);	/* DS max wsize. */
8061 	*tl++ = newnfs_true;			/* Tightly coupled. */
8062 	*tl++ = txdr_unsigned(NFS_VER4);	/* NFSv4. */
8063 	*tl++ = txdr_unsigned(NFSV41_MINORVERSION); /* Minor version 1. */
8064 	*tl++ = txdr_unsigned(nfs_srvmaxio);	/* DS max rsize. */
8065 	*tl++ = txdr_unsigned(nfs_srvmaxio);	/* DS max wsize. */
8066 	*tl = newnfs_true;			/* Tightly coupled. */
8067 
8068 	ds->nfsdev_hostnamelen = strlen(dnshost);
8069 	ds->nfsdev_host = malloc(ds->nfsdev_hostnamelen + 1, M_NFSDSTATE,
8070 	    M_WAITOK);
8071 	NFSBCOPY(dnshost, ds->nfsdev_host, ds->nfsdev_hostnamelen + 1);
8072 }
8073 
8074 /*
8075  * Create the device id list.
8076  * Return 0 if the nfsd threads are to run and ENXIO if the "-p" argument
8077  * is misconfigured.
8078  */
8079 int
8080 nfsrv_createdevids(struct nfsd_nfsd_args *args, NFSPROC_T *p)
8081 {
8082 	struct nfsdevice *ds;
8083 	char *addrp, *dnshostp, *dspathp, *mdspathp;
8084 	int error, i;
8085 
8086 	addrp = args->addr;
8087 	dnshostp = args->dnshost;
8088 	dspathp = args->dspath;
8089 	mdspathp = args->mdspath;
8090 	nfsrv_maxpnfsmirror = args->mirrorcnt;
8091 	if (addrp == NULL || dnshostp == NULL || dspathp == NULL ||
8092 	    mdspathp == NULL)
8093 		return (0);
8094 
8095 	/*
8096 	 * Loop around for each nul-terminated string in args->addr,
8097 	 * args->dnshost, args->dnspath and args->mdspath.
8098 	 */
8099 	while (addrp < (args->addr + args->addrlen) &&
8100 	    dnshostp < (args->dnshost + args->dnshostlen) &&
8101 	    dspathp < (args->dspath + args->dspathlen) &&
8102 	    mdspathp < (args->mdspath + args->mdspathlen)) {
8103 		error = nfsrv_setdsserver(dspathp, mdspathp, p, &ds);
8104 		if (error != 0) {
8105 			/* Free all DS servers. */
8106 			nfsrv_freealldevids();
8107 			nfsrv_devidcnt = 0;
8108 			return (ENXIO);
8109 		}
8110 		nfsrv_allocdevid(ds, addrp, dnshostp);
8111 		addrp += (strlen(addrp) + 1);
8112 		dnshostp += (strlen(dnshostp) + 1);
8113 		dspathp += (strlen(dspathp) + 1);
8114 		mdspathp += (strlen(mdspathp) + 1);
8115 	}
8116 	if (nfsrv_devidcnt < nfsrv_maxpnfsmirror) {
8117 		/* Free all DS servers. */
8118 		nfsrv_freealldevids();
8119 		nfsrv_devidcnt = 0;
8120 		nfsrv_maxpnfsmirror = 1;
8121 		return (ENXIO);
8122 	}
8123 	/* We can fail at most one less DS than the mirror level. */
8124 	nfsrv_faildscnt = nfsrv_maxpnfsmirror - 1;
8125 
8126 	/*
8127 	 * Allocate the nfslayout hash table now, since this is a pNFS server.
8128 	 * Make it 1% of the high water mark and at least 100.
8129 	 */
8130 	if (nfslayouthash == NULL) {
8131 		nfsrv_layouthashsize = nfsrv_layouthighwater / 100;
8132 		if (nfsrv_layouthashsize < 100)
8133 			nfsrv_layouthashsize = 100;
8134 		nfslayouthash = mallocarray(nfsrv_layouthashsize,
8135 		    sizeof(struct nfslayouthash), M_NFSDSESSION, M_WAITOK |
8136 		    M_ZERO);
8137 		for (i = 0; i < nfsrv_layouthashsize; i++) {
8138 			mtx_init(&nfslayouthash[i].mtx, "nfslm", NULL, MTX_DEF);
8139 			TAILQ_INIT(&nfslayouthash[i].list);
8140 		}
8141 	}
8142 	return (0);
8143 }
8144 
8145 /*
8146  * Free all device ids.
8147  */
8148 static void
8149 nfsrv_freealldevids(void)
8150 {
8151 	struct nfsdevice *ds, *nds;
8152 
8153 	TAILQ_FOREACH_SAFE(ds, &nfsrv_devidhead, nfsdev_list, nds)
8154 		nfsrv_freedevid(ds);
8155 }
8156 
8157 /*
8158  * Check to see if there is a Read/Write Layout plus either:
8159  * - A Write Delegation
8160  * or
8161  * - An Open with Write_access.
8162  * Return 1 if this is the case and 0 otherwise.
8163  * This function is used by nfsrv_proxyds() to decide if doing a Proxy
8164  * Getattr RPC to the Data Server (DS) is necessary.
8165  */
8166 #define	NFSCLIDVECSIZE	6
8167 int
8168 nfsrv_checkdsattr(vnode_t vp, NFSPROC_T *p)
8169 {
8170 	fhandle_t fh, *tfhp;
8171 	struct nfsstate *stp;
8172 	struct nfslayout *lyp;
8173 	struct nfslayouthash *lhyp;
8174 	struct nfslockhashhead *hp;
8175 	struct nfslockfile *lfp;
8176 	nfsquad_t clid[NFSCLIDVECSIZE];
8177 	int clidcnt, ret;
8178 
8179 	ret = nfsvno_getfh(vp, &fh, p);
8180 	if (ret != 0)
8181 		return (0);
8182 
8183 	/* First check for a Read/Write Layout. */
8184 	clidcnt = 0;
8185 	lhyp = NFSLAYOUTHASH(&fh);
8186 	NFSLOCKLAYOUT(lhyp);
8187 	TAILQ_FOREACH(lyp, &lhyp->list, lay_list) {
8188 		if (NFSBCMP(&lyp->lay_fh, &fh, sizeof(fh)) == 0 &&
8189 		    ((lyp->lay_flags & NFSLAY_RW) != 0 ||
8190 		     ((lyp->lay_flags & NFSLAY_READ) != 0 &&
8191 		      nfsrv_pnfsatime != 0))) {
8192 			if (clidcnt < NFSCLIDVECSIZE)
8193 				clid[clidcnt].qval = lyp->lay_clientid.qval;
8194 			clidcnt++;
8195 		}
8196 	}
8197 	NFSUNLOCKLAYOUT(lhyp);
8198 	if (clidcnt == 0) {
8199 		/* None found, so return 0. */
8200 		return (0);
8201 	}
8202 
8203 	/* Get the nfslockfile for this fh. */
8204 	NFSLOCKSTATE();
8205 	hp = NFSLOCKHASH(&fh);
8206 	LIST_FOREACH(lfp, hp, lf_hash) {
8207 		tfhp = &lfp->lf_fh;
8208 		if (NFSVNO_CMPFH(&fh, tfhp))
8209 			break;
8210 	}
8211 	if (lfp == NULL) {
8212 		/* None found, so return 0. */
8213 		NFSUNLOCKSTATE();
8214 		return (0);
8215 	}
8216 
8217 	/* Now, look for a Write delegation for this clientid. */
8218 	LIST_FOREACH(stp, &lfp->lf_deleg, ls_file) {
8219 		if ((stp->ls_flags & NFSLCK_DELEGWRITE) != 0 &&
8220 		    nfsrv_fndclid(clid, stp->ls_clp->lc_clientid, clidcnt) != 0)
8221 			break;
8222 	}
8223 	if (stp != NULL) {
8224 		/* Found one, so return 1. */
8225 		NFSUNLOCKSTATE();
8226 		return (1);
8227 	}
8228 
8229 	/* No Write delegation, so look for an Open with Write_access. */
8230 	LIST_FOREACH(stp, &lfp->lf_open, ls_file) {
8231 		KASSERT((stp->ls_flags & NFSLCK_OPEN) != 0,
8232 		    ("nfsrv_checkdsattr: Non-open in Open list\n"));
8233 		if ((stp->ls_flags & NFSLCK_WRITEACCESS) != 0 &&
8234 		    nfsrv_fndclid(clid, stp->ls_clp->lc_clientid, clidcnt) != 0)
8235 			break;
8236 	}
8237 	NFSUNLOCKSTATE();
8238 	if (stp != NULL)
8239 		return (1);
8240 	return (0);
8241 }
8242 
8243 /*
8244  * Look for a matching clientid in the vector. Return 1 if one might match.
8245  */
8246 static int
8247 nfsrv_fndclid(nfsquad_t *clidvec, nfsquad_t clid, int clidcnt)
8248 {
8249 	int i;
8250 
8251 	/* If too many for the vector, return 1 since there might be a match. */
8252 	if (clidcnt > NFSCLIDVECSIZE)
8253 		return (1);
8254 
8255 	for (i = 0; i < clidcnt; i++)
8256 		if (clidvec[i].qval == clid.qval)
8257 			return (1);
8258 	return (0);
8259 }
8260 
8261 /*
8262  * Check the don't list for "vp" and see if issuing an rw layout is allowed.
8263  * Return 1 if issuing an rw layout isn't allowed, 0 otherwise.
8264  */
8265 static int
8266 nfsrv_dontlayout(fhandle_t *fhp)
8267 {
8268 	struct nfsdontlist *mrp;
8269 	int ret;
8270 
8271 	if (nfsrv_dontlistlen == 0)
8272 		return (0);
8273 	ret = 0;
8274 	NFSDDONTLISTLOCK();
8275 	LIST_FOREACH(mrp, &nfsrv_dontlisthead, nfsmr_list) {
8276 		if (NFSBCMP(fhp, &mrp->nfsmr_fh, sizeof(*fhp)) == 0 &&
8277 		    (mrp->nfsmr_flags & NFSMR_DONTLAYOUT) != 0) {
8278 			ret = 1;
8279 			break;
8280 		}
8281 	}
8282 	NFSDDONTLISTUNLOCK();
8283 	return (ret);
8284 }
8285 
8286 #define	PNFSDS_COPYSIZ	65536
8287 /*
8288  * Create a new file on a DS and copy the contents of an extant DS file to it.
8289  * This can be used for recovery of a DS file onto a recovered DS.
8290  * The steps are:
8291  * - When called, the MDS file's vnode is locked, blocking LayoutGet operations.
8292  * - Disable issuing of read/write layouts for the file via the nfsdontlist,
8293  *   so that they will be disabled after the MDS file's vnode is unlocked.
8294  * - Set up the nfsrv_recalllist so that recall of read/write layouts can
8295  *   be done.
8296  * - Unlock the MDS file's vnode, so that the client(s) can perform proxied
8297  *   writes, LayoutCommits and LayoutReturns for the file when completing the
8298  *   LayoutReturn requested by the LayoutRecall callback.
8299  * - Issue a LayoutRecall callback for all read/write layouts and wait for
8300  *   them to be returned. (If the LayoutRecall callback replies
8301  *   NFSERR_NOMATCHLAYOUT, they are gone and no LayoutReturn is needed.)
8302  * - Exclusively lock the MDS file's vnode.  This ensures that no proxied
8303  *   writes are in progress or can occur during the DS file copy.
8304  *   It also blocks Setattr operations.
8305  * - Create the file on the recovered mirror.
8306  * - Copy the file from the operational DS.
8307  * - Copy any ACL from the MDS file to the new DS file.
8308  * - Set the modify time of the new DS file to that of the MDS file.
8309  * - Update the extended attribute for the MDS file.
8310  * - Enable issuing of rw layouts by deleting the nfsdontlist entry.
8311  * - The caller will unlock the MDS file's vnode allowing operations
8312  *   to continue normally, since it is now on the mirror again.
8313  */
8314 int
8315 nfsrv_copymr(vnode_t vp, vnode_t fvp, vnode_t dvp, struct nfsdevice *ds,
8316     struct pnfsdsfile *pf, struct pnfsdsfile *wpf, int mirrorcnt,
8317     struct ucred *cred, NFSPROC_T *p)
8318 {
8319 	struct nfsdontlist *mrp, *nmrp;
8320 	struct nfslayouthash *lhyp;
8321 	struct nfslayout *lyp, *nlyp;
8322 	struct nfslayouthead thl;
8323 	struct mount *mp, *tvmp;
8324 	struct acl *aclp;
8325 	struct vattr va;
8326 	struct timespec mtime;
8327 	fhandle_t fh;
8328 	vnode_t tvp;
8329 	off_t rdpos, wrpos;
8330 	ssize_t aresid;
8331 	char *dat;
8332 	int didprintf, ret, retacl, xfer;
8333 
8334 	ASSERT_VOP_LOCKED(fvp, "nfsrv_copymr fvp");
8335 	ASSERT_VOP_LOCKED(vp, "nfsrv_copymr vp");
8336 	/*
8337 	 * Allocate a nfsdontlist entry and set the NFSMR_DONTLAYOUT flag
8338 	 * so that no more RW layouts will get issued.
8339 	 */
8340 	ret = nfsvno_getfh(vp, &fh, p);
8341 	if (ret != 0) {
8342 		NFSD_DEBUG(4, "nfsrv_copymr: getfh=%d\n", ret);
8343 		return (ret);
8344 	}
8345 	nmrp = malloc(sizeof(*nmrp), M_NFSDSTATE, M_WAITOK);
8346 	nmrp->nfsmr_flags = NFSMR_DONTLAYOUT;
8347 	NFSBCOPY(&fh, &nmrp->nfsmr_fh, sizeof(fh));
8348 	NFSDDONTLISTLOCK();
8349 	LIST_FOREACH(mrp, &nfsrv_dontlisthead, nfsmr_list) {
8350 		if (NFSBCMP(&fh, &mrp->nfsmr_fh, sizeof(fh)) == 0)
8351 			break;
8352 	}
8353 	if (mrp == NULL) {
8354 		LIST_INSERT_HEAD(&nfsrv_dontlisthead, nmrp, nfsmr_list);
8355 		mrp = nmrp;
8356 		nmrp = NULL;
8357 		nfsrv_dontlistlen++;
8358 		NFSD_DEBUG(4, "nfsrv_copymr: in dontlist\n");
8359 	} else {
8360 		NFSDDONTLISTUNLOCK();
8361 		free(nmrp, M_NFSDSTATE);
8362 		NFSD_DEBUG(4, "nfsrv_copymr: dup dontlist\n");
8363 		return (ENXIO);
8364 	}
8365 	NFSDDONTLISTUNLOCK();
8366 
8367 	/*
8368 	 * Search for all RW layouts for this file.  Move them to the
8369 	 * recall list, so they can be recalled and their return noted.
8370 	 */
8371 	lhyp = NFSLAYOUTHASH(&fh);
8372 	NFSDRECALLLOCK();
8373 	NFSLOCKLAYOUT(lhyp);
8374 	TAILQ_FOREACH_SAFE(lyp, &lhyp->list, lay_list, nlyp) {
8375 		if (NFSBCMP(&lyp->lay_fh, &fh, sizeof(fh)) == 0 &&
8376 		    (lyp->lay_flags & NFSLAY_RW) != 0) {
8377 			TAILQ_REMOVE(&lhyp->list, lyp, lay_list);
8378 			TAILQ_INSERT_HEAD(&nfsrv_recalllisthead, lyp, lay_list);
8379 			lyp->lay_trycnt = 0;
8380 		}
8381 	}
8382 	NFSUNLOCKLAYOUT(lhyp);
8383 	NFSDRECALLUNLOCK();
8384 
8385 	ret = 0;
8386 	mp = tvmp = NULL;
8387 	didprintf = 0;
8388 	TAILQ_INIT(&thl);
8389 	/* Unlock the MDS vp, so that a LayoutReturn can be done on it. */
8390 	NFSVOPUNLOCK(vp);
8391 	/* Now, do a recall for all layouts not yet recalled. */
8392 tryagain:
8393 	NFSDRECALLLOCK();
8394 	TAILQ_FOREACH(lyp, &nfsrv_recalllisthead, lay_list) {
8395 		if (NFSBCMP(&lyp->lay_fh, &fh, sizeof(fh)) == 0 &&
8396 		    (lyp->lay_flags & NFSLAY_RECALL) == 0) {
8397 			lyp->lay_flags |= NFSLAY_RECALL;
8398 			/*
8399 			 * The layout stateid.seqid needs to be incremented
8400 			 * before doing a LAYOUT_RECALL callback.
8401 			 */
8402 			if (++lyp->lay_stateid.seqid == 0)
8403 				lyp->lay_stateid.seqid = 1;
8404 			NFSDRECALLUNLOCK();
8405 			nfsrv_recalllayout(lyp->lay_clientid, &lyp->lay_stateid,
8406 			    &lyp->lay_fh, lyp, 0, lyp->lay_type, p);
8407 			NFSD_DEBUG(4, "nfsrv_copymr: recalled layout\n");
8408 			goto tryagain;
8409 		}
8410 	}
8411 
8412 	/* Now wait for them to be returned. */
8413 tryagain2:
8414 	TAILQ_FOREACH(lyp, &nfsrv_recalllisthead, lay_list) {
8415 		if (NFSBCMP(&lyp->lay_fh, &fh, sizeof(fh)) == 0) {
8416 			if ((lyp->lay_flags & NFSLAY_RETURNED) != 0) {
8417 				TAILQ_REMOVE(&nfsrv_recalllisthead, lyp,
8418 				    lay_list);
8419 				TAILQ_INSERT_HEAD(&thl, lyp, lay_list);
8420 				NFSD_DEBUG(4,
8421 				    "nfsrv_copymr: layout returned\n");
8422 			} else {
8423 				lyp->lay_trycnt++;
8424 				ret = mtx_sleep(lyp, NFSDRECALLMUTEXPTR,
8425 				    PVFS | PCATCH, "nfsmrl", hz);
8426 				NFSD_DEBUG(4, "nfsrv_copymr: aft sleep=%d\n",
8427 				    ret);
8428 				if (ret == EINTR || ret == ERESTART)
8429 					break;
8430 				if ((lyp->lay_flags & NFSLAY_RETURNED) == 0) {
8431 					/*
8432 					 * Give up after 60sec and return
8433 					 * ENXIO, failing the copymr.
8434 					 * This layout will remain on the
8435 					 * recalllist.  It can only be cleared
8436 					 * by restarting the nfsd.
8437 					 * This seems the safe way to handle
8438 					 * it, since it cannot be safely copied
8439 					 * with an outstanding RW layout.
8440 					 */
8441 					if (lyp->lay_trycnt >= 60) {
8442 						ret = ENXIO;
8443 						break;
8444 					}
8445 					if (didprintf == 0) {
8446 						printf("nfsrv_copymr: layout "
8447 						    "not returned\n");
8448 						didprintf = 1;
8449 					}
8450 				}
8451 			}
8452 			goto tryagain2;
8453 		}
8454 	}
8455 	NFSDRECALLUNLOCK();
8456 	/* We can now get rid of the layouts that have been returned. */
8457 	TAILQ_FOREACH_SAFE(lyp, &thl, lay_list, nlyp)
8458 		nfsrv_freelayout(&thl, lyp);
8459 
8460 	/*
8461 	 * Do the vn_start_write() calls here, before the MDS vnode is
8462 	 * locked and the tvp is created (locked) in the NFS file system
8463 	 * that dvp is in.
8464 	 * For tvmp, this probably isn't necessary, since it will be an
8465 	 * NFS mount and they are not suspendable at this time.
8466 	 */
8467 	if (ret == 0)
8468 		ret = vn_start_write(vp, &mp, V_WAIT | V_PCATCH);
8469 	if (ret == 0) {
8470 		tvmp = dvp->v_mount;
8471 		ret = vn_start_write(NULL, &tvmp, V_WAIT | V_PCATCH);
8472 	}
8473 
8474 	/*
8475 	 * LK_EXCLUSIVE lock the MDS vnode, so that any
8476 	 * proxied writes through the MDS will be blocked until we have
8477 	 * completed the copy and update of the extended attributes.
8478 	 * This will also ensure that any attributes and ACL will not be
8479 	 * changed until the copy is complete.
8480 	 */
8481 	NFSVOPLOCK(vp, LK_EXCLUSIVE | LK_RETRY);
8482 	if (ret == 0 && VN_IS_DOOMED(vp)) {
8483 		NFSD_DEBUG(4, "nfsrv_copymr: lk_exclusive doomed\n");
8484 		ret = ESTALE;
8485 	}
8486 
8487 	/* Create the data file on the recovered DS. */
8488 	if (ret == 0)
8489 		ret = nfsrv_createdsfile(vp, &fh, pf, dvp, ds, cred, p, &tvp);
8490 
8491 	/* Copy the DS file, if created successfully. */
8492 	if (ret == 0) {
8493 		/*
8494 		 * Get any NFSv4 ACL on the MDS file, so that it can be set
8495 		 * on the new DS file.
8496 		 */
8497 		aclp = acl_alloc(M_WAITOK | M_ZERO);
8498 		retacl = VOP_GETACL(vp, ACL_TYPE_NFS4, aclp, cred, p);
8499 		if (retacl != 0 && retacl != ENOATTR)
8500 			NFSD_DEBUG(1, "nfsrv_copymr: vop_getacl=%d\n", retacl);
8501 		dat = malloc(PNFSDS_COPYSIZ, M_TEMP, M_WAITOK);
8502 		/* Malloc a block of 0s used to check for holes. */
8503 		if (nfsrv_zeropnfsdat == NULL)
8504 			nfsrv_zeropnfsdat = malloc(PNFSDS_COPYSIZ, M_TEMP,
8505 			    M_WAITOK | M_ZERO);
8506 		rdpos = wrpos = 0;
8507 		ret = VOP_GETATTR(fvp, &va, cred);
8508 		aresid = 0;
8509 		while (ret == 0 && aresid == 0) {
8510 			ret = vn_rdwr(UIO_READ, fvp, dat, PNFSDS_COPYSIZ,
8511 			    rdpos, UIO_SYSSPACE, IO_NODELOCKED, cred, NULL,
8512 			    &aresid, p);
8513 			xfer = PNFSDS_COPYSIZ - aresid;
8514 			if (ret == 0 && xfer > 0) {
8515 				rdpos += xfer;
8516 				/*
8517 				 * Skip the write for holes, except for the
8518 				 * last block.
8519 				 */
8520 				if (xfer < PNFSDS_COPYSIZ || rdpos ==
8521 				    va.va_size || NFSBCMP(dat,
8522 				    nfsrv_zeropnfsdat, PNFSDS_COPYSIZ) != 0)
8523 					ret = vn_rdwr(UIO_WRITE, tvp, dat, xfer,
8524 					    wrpos, UIO_SYSSPACE, IO_NODELOCKED,
8525 					    cred, NULL, NULL, p);
8526 				if (ret == 0)
8527 					wrpos += xfer;
8528 			}
8529 		}
8530 
8531 		/* If there is an ACL and the copy succeeded, set the ACL. */
8532 		if (ret == 0 && retacl == 0) {
8533 			ret = VOP_SETACL(tvp, ACL_TYPE_NFS4, aclp, cred, p);
8534 			/*
8535 			 * Don't consider these as errors, since VOP_GETACL()
8536 			 * can return an ACL when they are not actually
8537 			 * supported.  For example, for UFS, VOP_GETACL()
8538 			 * will return a trivial ACL based on the uid/gid/mode
8539 			 * when there is no ACL on the file.
8540 			 * This case should be recognized as a trivial ACL
8541 			 * by UFS's VOP_SETACL() and succeed, but...
8542 			 */
8543 			if (ret == ENOATTR || ret == EOPNOTSUPP || ret == EPERM)
8544 				ret = 0;
8545 		}
8546 
8547 		if (ret == 0)
8548 			ret = VOP_FSYNC(tvp, MNT_WAIT, p);
8549 
8550 		/* Set the DS data file's modify time that of the MDS file. */
8551 		if (ret == 0)
8552 			ret = VOP_GETATTR(vp, &va, cred);
8553 		if (ret == 0) {
8554 			mtime = va.va_mtime;
8555 			VATTR_NULL(&va);
8556 			va.va_mtime = mtime;
8557 			ret = VOP_SETATTR(tvp, &va, cred);
8558 		}
8559 
8560 		vput(tvp);
8561 		acl_free(aclp);
8562 		free(dat, M_TEMP);
8563 	}
8564 	if (tvmp != NULL)
8565 		vn_finished_write(tvmp);
8566 
8567 	/* Update the extended attributes for the newly created DS file. */
8568 	if (ret == 0)
8569 		ret = vn_extattr_set(vp, IO_NODELOCKED,
8570 		    EXTATTR_NAMESPACE_SYSTEM, "pnfsd.dsfile",
8571 		    sizeof(*wpf) * mirrorcnt, (char *)wpf, p);
8572 	if (mp != NULL)
8573 		vn_finished_write(mp);
8574 
8575 	/* Get rid of the dontlist entry, so that Layouts can be issued. */
8576 	NFSDDONTLISTLOCK();
8577 	LIST_REMOVE(mrp, nfsmr_list);
8578 	NFSDDONTLISTUNLOCK();
8579 	free(mrp, M_NFSDSTATE);
8580 	return (ret);
8581 }
8582 
8583 /*
8584  * Create a data storage file on the recovered DS.
8585  */
8586 static int
8587 nfsrv_createdsfile(vnode_t vp, fhandle_t *fhp, struct pnfsdsfile *pf,
8588     vnode_t dvp, struct nfsdevice *ds, struct ucred *cred, NFSPROC_T *p,
8589     vnode_t *tvpp)
8590 {
8591 	struct vattr va, nva;
8592 	int error;
8593 
8594 	/* Make data file name based on FH. */
8595 	error = VOP_GETATTR(vp, &va, cred);
8596 	if (error == 0) {
8597 		/* Set the attributes for "vp" to Setattr the DS vp. */
8598 		VATTR_NULL(&nva);
8599 		nva.va_uid = va.va_uid;
8600 		nva.va_gid = va.va_gid;
8601 		nva.va_mode = va.va_mode;
8602 		nva.va_size = 0;
8603 		VATTR_NULL(&va);
8604 		va.va_type = VREG;
8605 		va.va_mode = nva.va_mode;
8606 		NFSD_DEBUG(4, "nfsrv_dscreatefile: dvp=%p pf=%p\n", dvp, pf);
8607 		error = nfsrv_dscreate(dvp, &va, &nva, fhp, pf, NULL,
8608 		    pf->dsf_filename, cred, p, tvpp);
8609 	}
8610 	return (error);
8611 }
8612 
8613 /*
8614  * Look up the MDS file shared locked, and then get the extended attribute
8615  * to find the extant DS file to be copied to the new mirror.
8616  * If successful, *vpp is set to the MDS file's vp and *nvpp is
8617  * set to a DS data file for the MDS file, both exclusively locked.
8618  * The "buf" argument has the pnfsdsfile structure from the MDS file
8619  * in it and buflen is set to its length.
8620  */
8621 int
8622 nfsrv_mdscopymr(char *mdspathp, char *dspathp, char *curdspathp, char *buf,
8623     int *buflenp, char *fname, NFSPROC_T *p, struct vnode **vpp,
8624     struct vnode **nvpp, struct pnfsdsfile **pfp, struct nfsdevice **dsp,
8625     struct nfsdevice **fdsp)
8626 {
8627 	struct nameidata nd;
8628 	struct vnode *vp, *curvp;
8629 	struct pnfsdsfile *pf;
8630 	struct nfsmount *nmp, *curnmp;
8631 	int dsdir, error, mirrorcnt, ippos;
8632 
8633 	vp = NULL;
8634 	curvp = NULL;
8635 	curnmp = NULL;
8636 	*dsp = NULL;
8637 	*fdsp = NULL;
8638 	if (dspathp == NULL && curdspathp != NULL)
8639 		return (EPERM);
8640 
8641 	/*
8642 	 * Look up the MDS file shared locked.  The lock will be upgraded
8643 	 * to an exclusive lock after any rw layouts have been returned.
8644 	 */
8645 	NFSD_DEBUG(4, "mdsopen path=%s\n", mdspathp);
8646 	NDINIT(&nd, LOOKUP, FOLLOW | LOCKSHARED | LOCKLEAF, UIO_SYSSPACE,
8647 	    mdspathp);
8648 	error = namei(&nd);
8649 	NFSD_DEBUG(4, "lookup=%d\n", error);
8650 	if (error != 0)
8651 		return (error);
8652 	if (nd.ni_vp->v_type != VREG) {
8653 		vput(nd.ni_vp);
8654 		NFSD_DEBUG(4, "mdspath not reg\n");
8655 		return (EISDIR);
8656 	}
8657 	vp = nd.ni_vp;
8658 
8659 	if (curdspathp != NULL) {
8660 		/*
8661 		 * Look up the current DS path and find the nfsdev structure for
8662 		 * it.
8663 		 */
8664 		NFSD_DEBUG(4, "curmdsdev path=%s\n", curdspathp);
8665 		NDINIT(&nd, LOOKUP, FOLLOW | LOCKSHARED | LOCKLEAF,
8666 		    UIO_SYSSPACE, curdspathp);
8667 		error = namei(&nd);
8668 		NFSD_DEBUG(4, "ds lookup=%d\n", error);
8669 		if (error != 0) {
8670 			vput(vp);
8671 			return (error);
8672 		}
8673 		if (nd.ni_vp->v_type != VDIR) {
8674 			vput(nd.ni_vp);
8675 			vput(vp);
8676 			NFSD_DEBUG(4, "curdspath not dir\n");
8677 			return (ENOTDIR);
8678 		}
8679 		if (strcmp(nd.ni_vp->v_mount->mnt_vfc->vfc_name, "nfs") != 0) {
8680 			vput(nd.ni_vp);
8681 			vput(vp);
8682 			NFSD_DEBUG(4, "curdspath not an NFS mount\n");
8683 			return (ENXIO);
8684 		}
8685 		curnmp = VFSTONFS(nd.ni_vp->v_mount);
8686 
8687 		/* Search the nfsdev list for a match. */
8688 		NFSDDSLOCK();
8689 		*fdsp = nfsv4_findmirror(curnmp);
8690 		NFSDDSUNLOCK();
8691 		if (*fdsp == NULL)
8692 			curnmp = NULL;
8693 		if (curnmp == NULL) {
8694 			vput(nd.ni_vp);
8695 			vput(vp);
8696 			NFSD_DEBUG(4, "mdscopymr: no current ds\n");
8697 			return (ENXIO);
8698 		}
8699 		curvp = nd.ni_vp;
8700 	}
8701 
8702 	if (dspathp != NULL) {
8703 		/* Look up the nfsdev path and find the nfsdev structure. */
8704 		NFSD_DEBUG(4, "mdsdev path=%s\n", dspathp);
8705 		NDINIT(&nd, LOOKUP, FOLLOW | LOCKSHARED | LOCKLEAF,
8706 		    UIO_SYSSPACE, dspathp);
8707 		error = namei(&nd);
8708 		NFSD_DEBUG(4, "ds lookup=%d\n", error);
8709 		if (error != 0) {
8710 			vput(vp);
8711 			if (curvp != NULL)
8712 				vput(curvp);
8713 			return (error);
8714 		}
8715 		if (nd.ni_vp->v_type != VDIR || nd.ni_vp == curvp) {
8716 			vput(nd.ni_vp);
8717 			vput(vp);
8718 			if (curvp != NULL)
8719 				vput(curvp);
8720 			NFSD_DEBUG(4, "dspath not dir\n");
8721 			if (nd.ni_vp == curvp)
8722 				return (EPERM);
8723 			return (ENOTDIR);
8724 		}
8725 		if (strcmp(nd.ni_vp->v_mount->mnt_vfc->vfc_name, "nfs") != 0) {
8726 			vput(nd.ni_vp);
8727 			vput(vp);
8728 			if (curvp != NULL)
8729 				vput(curvp);
8730 			NFSD_DEBUG(4, "dspath not an NFS mount\n");
8731 			return (ENXIO);
8732 		}
8733 		nmp = VFSTONFS(nd.ni_vp->v_mount);
8734 
8735 		/*
8736 		 * Search the nfsdevice list for a match.  If curnmp == NULL,
8737 		 * this is a recovery and there must be a mirror.
8738 		 */
8739 		NFSDDSLOCK();
8740 		if (curnmp == NULL)
8741 			*dsp = nfsrv_findmirroredds(nmp);
8742 		else
8743 			*dsp = nfsv4_findmirror(nmp);
8744 		NFSDDSUNLOCK();
8745 		if (*dsp == NULL) {
8746 			vput(nd.ni_vp);
8747 			vput(vp);
8748 			if (curvp != NULL)
8749 				vput(curvp);
8750 			NFSD_DEBUG(4, "mdscopymr: no ds\n");
8751 			return (ENXIO);
8752 		}
8753 	} else {
8754 		nd.ni_vp = NULL;
8755 		nmp = NULL;
8756 	}
8757 
8758 	/*
8759 	 * Get a vp for an available DS data file using the extended
8760 	 * attribute on the MDS file.
8761 	 * If there is a valid entry for the new DS in the extended attribute
8762 	 * on the MDS file (as checked via the nmp argument),
8763 	 * nfsrv_dsgetsockmnt() returns EEXIST, so no copying will occur.
8764 	 */
8765 	error = nfsrv_dsgetsockmnt(vp, 0, buf, buflenp, &mirrorcnt, p,
8766 	    NULL, NULL, NULL, fname, nvpp, &nmp, curnmp, &ippos, &dsdir);
8767 	if (curvp != NULL)
8768 		vput(curvp);
8769 	if (nd.ni_vp == NULL) {
8770 		if (error == 0 && nmp != NULL) {
8771 			/* Search the nfsdev list for a match. */
8772 			NFSDDSLOCK();
8773 			*dsp = nfsrv_findmirroredds(nmp);
8774 			NFSDDSUNLOCK();
8775 		}
8776 		if (error == 0 && (nmp == NULL || *dsp == NULL)) {
8777 			if (nvpp != NULL && *nvpp != NULL) {
8778 				vput(*nvpp);
8779 				*nvpp = NULL;
8780 			}
8781 			error = ENXIO;
8782 		}
8783 	} else
8784 		vput(nd.ni_vp);
8785 
8786 	/*
8787 	 * When dspathp != NULL and curdspathp == NULL, this is a recovery
8788 	 * and is only allowed if there is a 0.0.0.0 IP address entry.
8789 	 * When curdspathp != NULL, the ippos will be set to that entry.
8790 	 */
8791 	if (error == 0 && dspathp != NULL && ippos == -1) {
8792 		if (nvpp != NULL && *nvpp != NULL) {
8793 			vput(*nvpp);
8794 			*nvpp = NULL;
8795 		}
8796 		error = ENXIO;
8797 	}
8798 	if (error == 0) {
8799 		*vpp = vp;
8800 
8801 		pf = (struct pnfsdsfile *)buf;
8802 		if (ippos == -1) {
8803 			/* If no zeroip pnfsdsfile, add one. */
8804 			ippos = *buflenp / sizeof(*pf);
8805 			*buflenp += sizeof(*pf);
8806 			pf += ippos;
8807 			pf->dsf_dir = dsdir;
8808 			strlcpy(pf->dsf_filename, fname,
8809 			    sizeof(pf->dsf_filename));
8810 		} else
8811 			pf += ippos;
8812 		*pfp = pf;
8813 	} else
8814 		vput(vp);
8815 	return (error);
8816 }
8817 
8818 /*
8819  * Search for a matching pnfsd mirror device structure, base on the nmp arg.
8820  * Return one if found, NULL otherwise.
8821  */
8822 static struct nfsdevice *
8823 nfsrv_findmirroredds(struct nfsmount *nmp)
8824 {
8825 	struct nfsdevice *ds, *fndds;
8826 	int fndmirror;
8827 
8828 	mtx_assert(NFSDDSMUTEXPTR, MA_OWNED);
8829 	/*
8830 	 * Search the DS server list for a match with nmp.
8831 	 * Remove the DS entry if found and there is a mirror.
8832 	 */
8833 	fndds = NULL;
8834 	fndmirror = 0;
8835 	if (nfsrv_devidcnt == 0)
8836 		return (fndds);
8837 	TAILQ_FOREACH(ds, &nfsrv_devidhead, nfsdev_list) {
8838 		if (ds->nfsdev_nmp == nmp) {
8839 			NFSD_DEBUG(4, "nfsrv_findmirroredds: fnd main ds\n");
8840 			fndds = ds;
8841 			break;
8842 		}
8843 	}
8844 	if (fndds == NULL)
8845 		return (fndds);
8846 	if (fndds->nfsdev_mdsisset == 0 && nfsrv_faildscnt > 0)
8847 		fndmirror = 1;
8848 	else if (fndds->nfsdev_mdsisset != 0) {
8849 		/* For the fsid is set case, search for a mirror. */
8850 		TAILQ_FOREACH(ds, &nfsrv_devidhead, nfsdev_list) {
8851 			if (ds != fndds && ds->nfsdev_nmp != NULL &&
8852 			    ds->nfsdev_mdsisset != 0 &&
8853 			    fsidcmp(&ds->nfsdev_mdsfsid,
8854 			    &fndds->nfsdev_mdsfsid) == 0) {
8855 				fndmirror = 1;
8856 				break;
8857 			}
8858 		}
8859 	}
8860 	if (fndmirror == 0) {
8861 		NFSD_DEBUG(4, "nfsrv_findmirroredds: no mirror for DS\n");
8862 		return (NULL);
8863 	}
8864 	return (fndds);
8865 }
8866 
8867 /*
8868  * Mark the appropriate devid and all associated layout as "out of space".
8869  */
8870 void
8871 nfsrv_marknospc(char *devid, bool setit)
8872 {
8873 	struct nfsdevice *ds;
8874 	struct nfslayout *lyp;
8875 	struct nfslayouthash *lhyp;
8876 	int i;
8877 
8878 	NFSDDSLOCK();
8879 	TAILQ_FOREACH(ds, &nfsrv_devidhead, nfsdev_list) {
8880 		if (NFSBCMP(ds->nfsdev_deviceid, devid, NFSX_V4DEVICEID) == 0) {
8881 			NFSD_DEBUG(1, "nfsrv_marknospc: devid %d\n", setit);
8882 			ds->nfsdev_nospc = setit;
8883 		}
8884 	}
8885 	NFSDDSUNLOCK();
8886 
8887 	for (i = 0; i < nfsrv_layouthashsize; i++) {
8888 		lhyp = &nfslayouthash[i];
8889 		NFSLOCKLAYOUT(lhyp);
8890 		TAILQ_FOREACH(lyp, &lhyp->list, lay_list) {
8891 			if (NFSBCMP(lyp->lay_deviceid, devid,
8892 			    NFSX_V4DEVICEID) == 0) {
8893 				NFSD_DEBUG(1, "nfsrv_marknospc: layout %d\n",
8894 				    setit);
8895 				if (setit)
8896 					lyp->lay_flags |= NFSLAY_NOSPC;
8897 				else
8898 					lyp->lay_flags &= ~NFSLAY_NOSPC;
8899 			}
8900 		}
8901 		NFSUNLOCKLAYOUT(lhyp);
8902 	}
8903 }
8904 
8905 /*
8906  * Check to see if SP4_MACH_CRED is in use and, if it is, check that the
8907  * correct machine credential is being used.
8908  */
8909 static int
8910 nfsrv_checkmachcred(int op, struct nfsrv_descript *nd, struct nfsclient *clp)
8911 {
8912 
8913 	if ((clp->lc_flags & LCL_MACHCRED) == 0 ||
8914 	    !NFSISSET_OPBIT(&clp->lc_mustops, op))
8915 		return (0);
8916 	KASSERT((nd->nd_flag & ND_NFSV41) != 0,
8917 	    ("nfsrv_checkmachcred: MachCred for NFSv4.0"));
8918 	if ((nd->nd_flag & (ND_GSSINTEGRITY | ND_GSSPRIVACY)) != 0 &&
8919 	    nd->nd_princlen == clp->lc_namelen &&
8920 	    !NFSBCMP(nd->nd_principal, clp->lc_name, nd->nd_princlen))
8921 		return (0);
8922 	return (NFSERR_AUTHERR | AUTH_TOOWEAK);
8923 }
8924 
8925 /*
8926  * Issue a delegation and, optionally set rflagsp for why not.
8927  */
8928 static void
8929 nfsrv_issuedelegation(struct vnode *vp, struct nfsclient *clp,
8930     struct nfsrv_descript *nd, int delegate, int writedeleg, int readonly,
8931     u_quad_t filerev, uint64_t rdonly, struct nfsstate **new_delegp,
8932     struct nfsstate *new_stp, struct nfslockfile *lfp, uint32_t *rflagsp,
8933     nfsv4stateid_t *delegstateidp)
8934 {
8935 	struct nfsstate *up_deleg, *new_deleg;
8936 
8937 	new_deleg = *new_delegp;
8938 	up_deleg = LIST_FIRST(&lfp->lf_deleg);
8939 	if ((new_stp->ls_flags & NFSLCK_WANTNODELEG) != 0)
8940 		*rflagsp |= NFSV4OPEN_WDNOTWANTED;
8941 	else if (nfsrv_issuedelegs == 0)
8942 		*rflagsp |= NFSV4OPEN_WDSUPPFTYPE;
8943 	else if (NFSRV_V4DELEGLIMIT(nfsrv_delegatecnt))
8944 		*rflagsp |= NFSV4OPEN_WDRESOURCE;
8945 	else if (delegate == 0 || !NFSVNO_DELEGOK(vp) ||
8946 	    (writedeleg == 0 && (readonly == 0 ||
8947 	    (new_stp->ls_flags & NFSLCK_WANTWDELEG) != 0)) ||
8948 	    (clp->lc_flags & (LCL_CALLBACKSON | LCL_CBDOWN)) !=
8949 	     LCL_CALLBACKSON) {
8950 		/* Is this a downgrade attempt? */
8951 		if (up_deleg != NULL && up_deleg->ls_clp == clp &&
8952 		    (up_deleg->ls_flags & NFSLCK_DELEGWRITE) != 0 &&
8953 		    (new_stp->ls_flags & NFSLCK_WANTRDELEG) != 0)
8954 			*rflagsp |= NFSV4OPEN_WDNOTSUPPDOWNGRADE;
8955 		else
8956 			*rflagsp |= NFSV4OPEN_WDCONTENTION;
8957 	} else if (up_deleg != NULL &&
8958 	    (up_deleg->ls_flags & NFSLCK_DELEGREAD) != 0 &&
8959 	    (new_stp->ls_flags & NFSLCK_WANTWDELEG) != 0) {
8960 		/* This is an atomic upgrade. */
8961 		up_deleg->ls_stateid.seqid++;
8962 		delegstateidp->seqid = up_deleg->ls_stateid.seqid;
8963 		delegstateidp->other[0] = up_deleg->ls_stateid.other[0];
8964 		delegstateidp->other[1] = up_deleg->ls_stateid.other[1];
8965 		delegstateidp->other[2] = up_deleg->ls_stateid.other[2];
8966 		up_deleg->ls_flags = (NFSLCK_DELEGWRITE |
8967 		    NFSLCK_READACCESS | NFSLCK_WRITEACCESS);
8968 		*rflagsp |= NFSV4OPEN_WRITEDELEGATE;
8969 		nfsrv_writedelegcnt++;
8970 	} else {
8971 		new_deleg->ls_stateid.seqid = delegstateidp->seqid = 1;
8972 		new_deleg->ls_stateid.other[0] = delegstateidp->other[0]
8973 		    = clp->lc_clientid.lval[0];
8974 		new_deleg->ls_stateid.other[1] = delegstateidp->other[1]
8975 		    = clp->lc_clientid.lval[1];
8976 		new_deleg->ls_stateid.other[2] = delegstateidp->other[2]
8977 		    = nfsrv_nextstateindex(clp);
8978 		if (writedeleg && !rdonly &&
8979 		    (nfsrv_writedelegifpos || !readonly) &&
8980 		    (new_stp->ls_flags & (NFSLCK_WANTRDELEG |
8981 		     NFSLCK_WANTWDELEG)) != NFSLCK_WANTRDELEG) {
8982 			new_deleg->ls_flags = (NFSLCK_DELEGWRITE |
8983 			    NFSLCK_READACCESS | NFSLCK_WRITEACCESS);
8984 			*rflagsp |= NFSV4OPEN_WRITEDELEGATE;
8985 			nfsrv_writedelegcnt++;
8986 		} else {
8987 			new_deleg->ls_flags = (NFSLCK_DELEGREAD |
8988 			    NFSLCK_READACCESS);
8989 			*rflagsp |= NFSV4OPEN_READDELEGATE;
8990 		}
8991 		new_deleg->ls_uid = new_stp->ls_uid;
8992 		new_deleg->ls_lfp = lfp;
8993 		new_deleg->ls_clp = clp;
8994 		new_deleg->ls_filerev = filerev;
8995 		new_deleg->ls_compref = nd->nd_compref;
8996 		new_deleg->ls_lastrecall = 0;
8997 		LIST_INSERT_HEAD(&lfp->lf_deleg, new_deleg, ls_file);
8998 		LIST_INSERT_HEAD(NFSSTATEHASH(clp, new_deleg->ls_stateid),
8999 		    new_deleg, ls_hash);
9000 		LIST_INSERT_HEAD(&clp->lc_deleg, new_deleg, ls_list);
9001 		*new_delegp = NULL;
9002 		NFSD_VNET(nfsstatsv1_p)->srvdelegates++;
9003 		nfsrv_openpluslock++;
9004 		nfsrv_delegatecnt++;
9005 	}
9006 }
9007