xref: /freebsd/sys/fs/nfsserver/nfs_nfsdstate.c (revision dd21556857e8d40f66bf5ad54754d9d52669ebf7)
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 NFSD_VNET_DEFINE_STATIC(bool, nfsd_disable_grace) = false;
119 SYSCTL_BOOL(_vfs_nfsd, OID_AUTO, testing_disable_grace,
120     CTLFLAG_NFSD_VNET | CTLFLAG_RW, &NFSD_VNET_NAME(nfsd_disable_grace),
121     0, "Disable grace for testing");
122 
123 /*
124  * Hash lists for nfs V4.
125  */
126 NFSD_VNET_DEFINE(struct nfsclienthashhead *, nfsclienthash);
127 NFSD_VNET_DEFINE(struct nfslockhashhead *, nfslockhash);
128 NFSD_VNET_DEFINE(struct nfssessionhash *, nfssessionhash);
129 
130 struct nfslayouthash		*nfslayouthash;
131 volatile int nfsrv_dontlistlen = 0;
132 
133 static u_int32_t nfsrv_openpluslock = 0, nfsrv_delegatecnt = 0;
134 static int nfsrv_returnoldstateid = 0, nfsrv_clients = 0;
135 static int nfsrv_clienthighwater = NFSRV_CLIENTHIGHWATER;
136 static int nfsrv_nogsscallback = 0;
137 static volatile int nfsrv_writedelegcnt = 0;
138 static int nfsrv_faildscnt;
139 
140 NFSD_VNET_DEFINE_STATIC(time_t, nfsrvboottime);
141 
142 /* local functions */
143 static void nfsrv_dumpaclient(struct nfsclient *clp,
144     struct nfsd_dumpclients *dumpp);
145 static void nfsrv_freeopenowner(struct nfsstate *stp, int cansleep,
146     NFSPROC_T *p);
147 static void nfsrv_freeopen(struct nfsstate *stp, vnode_t vp, int cansleep,
148     NFSPROC_T *p);
149 static void nfsrv_freelockowner(struct nfsstate *stp, vnode_t vp, int cansleep,
150     NFSPROC_T *p);
151 static void nfsrv_freeallnfslocks(struct nfsstate *stp, vnode_t vp,
152     int cansleep, NFSPROC_T *p);
153 static void nfsrv_freenfslock(struct nfslock *lop);
154 static void nfsrv_freenfslockfile(struct nfslockfile *lfp);
155 static void nfsrv_freedeleg(struct nfsstate *);
156 static int nfsrv_getstate(struct nfsclient *clp, nfsv4stateid_t *stateidp,
157     u_int32_t flags, struct nfsstate **stpp);
158 static void nfsrv_getowner(struct nfsstatehead *hp, struct nfsstate *new_stp,
159     struct nfsstate **stpp);
160 static int nfsrv_getlockfh(vnode_t vp, u_short flags,
161     struct nfslockfile *new_lfp, fhandle_t *nfhp, NFSPROC_T *p);
162 static int nfsrv_getlockfile(u_short flags, struct nfslockfile **new_lfpp,
163     struct nfslockfile **lfpp, fhandle_t *nfhp, int lockit);
164 static void nfsrv_insertlock(struct nfslock *new_lop,
165     struct nfslock *insert_lop, struct nfsstate *stp, struct nfslockfile *lfp);
166 static void nfsrv_updatelock(struct nfsstate *stp, struct nfslock **new_lopp,
167     struct nfslock **other_lopp, struct nfslockfile *lfp);
168 static int nfsrv_getipnumber(u_char *cp);
169 static int nfsrv_checkrestart(nfsquad_t clientid, u_int32_t flags,
170     nfsv4stateid_t *stateidp, int specialid);
171 static int nfsrv_checkgrace(struct nfsrv_descript *nd, struct nfsclient *clp,
172     u_int32_t flags);
173 static int nfsrv_docallback(struct nfsclient *clp, int procnum,
174     nfsv4stateid_t *stateidp, int trunc, fhandle_t *fhp,
175     struct nfsvattr *nap, nfsattrbit_t *attrbitp, int laytype, NFSPROC_T *p);
176 static int nfsrv_cbcallargs(struct nfsrv_descript *nd, struct nfsclient *clp,
177     uint32_t callback, int op, const char *optag, struct nfsdsession **sepp,
178     int *slotposp);
179 static u_int32_t nfsrv_nextclientindex(void);
180 static u_int32_t nfsrv_nextstateindex(struct nfsclient *clp);
181 static void nfsrv_markstable(struct nfsclient *clp);
182 static void nfsrv_markreclaim(struct nfsclient *clp);
183 static int nfsrv_checkstable(struct nfsclient *clp);
184 static int nfsrv_clientconflict(struct nfsclient *clp, int *haslockp, struct
185     vnode *vp, NFSPROC_T *p);
186 static int nfsrv_delegconflict(struct nfsstate *stp, int *haslockp,
187     NFSPROC_T *p, vnode_t vp);
188 static int nfsrv_cleandeleg(vnode_t vp, struct nfslockfile *lfp,
189     struct nfsclient *clp, int *haslockp, NFSPROC_T *p);
190 static int nfsrv_notsamecredname(int op, struct nfsrv_descript *nd,
191     struct nfsclient *clp);
192 static time_t nfsrv_leaseexpiry(void);
193 static void nfsrv_delaydelegtimeout(struct nfsstate *stp);
194 static int nfsrv_checkseqid(struct nfsrv_descript *nd, u_int32_t seqid,
195     struct nfsstate *stp, struct nfsrvcache *op);
196 static int nfsrv_nootherstate(struct nfsstate *stp);
197 static int nfsrv_locallock(vnode_t vp, struct nfslockfile *lfp, int flags,
198     uint64_t first, uint64_t end, struct nfslockconflict *cfp, NFSPROC_T *p);
199 static void nfsrv_localunlock(vnode_t vp, struct nfslockfile *lfp,
200     uint64_t init_first, uint64_t init_end, NFSPROC_T *p);
201 static int nfsrv_dolocal(vnode_t vp, struct nfslockfile *lfp, int flags,
202     int oldflags, uint64_t first, uint64_t end, struct nfslockconflict *cfp,
203     NFSPROC_T *p);
204 static void nfsrv_locallock_rollback(vnode_t vp, struct nfslockfile *lfp,
205     NFSPROC_T *p);
206 static void nfsrv_locallock_commit(struct nfslockfile *lfp, int flags,
207     uint64_t first, uint64_t end);
208 static void nfsrv_locklf(struct nfslockfile *lfp);
209 static void nfsrv_unlocklf(struct nfslockfile *lfp);
210 static struct nfsdsession *nfsrv_findsession(uint8_t *sessionid);
211 static int nfsrv_freesession(struct nfsrv_descript *nd, struct nfsdsession *sep,
212     uint8_t *sessionid, bool locked, SVCXPRT **old_xprtp);
213 static int nfsv4_setcbsequence(struct nfsrv_descript *nd, struct nfsclient *clp,
214     int dont_replycache, struct nfsdsession **sepp, int *slotposp);
215 static int nfsv4_getcbsession(struct nfsclient *clp, struct nfsdsession **sepp);
216 static int nfsrv_addlayout(struct nfsrv_descript *nd, struct nfslayout **lypp,
217     nfsv4stateid_t *stateidp, char *layp, int *layoutlenp, NFSPROC_T *p);
218 static void nfsrv_freelayout(struct nfslayouthead *lhp, struct nfslayout *lyp);
219 static void nfsrv_freelayoutlist(nfsquad_t clientid);
220 static void nfsrv_freelayouts(nfsquad_t *clid, fsid_t *fs, int laytype,
221     int iomode);
222 static void nfsrv_freealllayouts(void);
223 static void nfsrv_freedevid(struct nfsdevice *ds);
224 static int nfsrv_setdsserver(char *dspathp, char *mdspathp, NFSPROC_T *p,
225     struct nfsdevice **dsp);
226 static void nfsrv_deleteds(struct nfsdevice *fndds);
227 static void nfsrv_allocdevid(struct nfsdevice *ds, char *addr, char *dnshost);
228 static void nfsrv_freealldevids(void);
229 static void nfsrv_flexlayouterr(struct nfsrv_descript *nd, uint32_t *layp,
230     int maxcnt, NFSPROC_T *p);
231 static int nfsrv_recalllayout(nfsquad_t clid, nfsv4stateid_t *stateidp,
232     fhandle_t *fhp, struct nfslayout *lyp, int changed, int laytype,
233     NFSPROC_T *p);
234 static int nfsrv_findlayout(nfsquad_t *clientidp, fhandle_t *fhp, int laytype,
235     NFSPROC_T *, struct nfslayout **lypp);
236 static int nfsrv_fndclid(nfsquad_t *clidvec, nfsquad_t clid, int clidcnt);
237 static struct nfslayout *nfsrv_filelayout(struct nfsrv_descript *nd, int iomode,
238     fhandle_t *fhp, fhandle_t *dsfhp, char *devid, fsid_t fs);
239 static struct nfslayout *nfsrv_flexlayout(struct nfsrv_descript *nd, int iomode,
240     int mirrorcnt, fhandle_t *fhp, fhandle_t *dsfhp, char *devid, fsid_t fs);
241 static int nfsrv_dontlayout(fhandle_t *fhp);
242 static int nfsrv_createdsfile(vnode_t vp, fhandle_t *fhp, struct pnfsdsfile *pf,
243     vnode_t dvp, struct nfsdevice *ds, struct ucred *cred, NFSPROC_T *p,
244     vnode_t *tvpp);
245 static struct nfsdevice *nfsrv_findmirroredds(struct nfsmount *nmp);
246 static int nfsrv_checkmachcred(int op, struct nfsrv_descript *nd,
247     struct nfsclient *clp);
248 static void nfsrv_issuedelegation(struct vnode *vp, struct nfsclient *clp,
249     struct nfsrv_descript *nd, int delegate, int writedeleg, int readonly,
250     u_quad_t filerev, uint64_t rdonly, struct nfsstate **new_delegp,
251     struct nfsstate *new_stp, struct nfslockfile *lfp, uint32_t *rflagsp,
252     nfsv4stateid_t *delegstateidp);
253 static void nfsrv_clientlock(bool mlocked);
254 static void nfsrv_clientunlock(bool mlocked);
255 
256 /*
257  * Lock the client structure, either with the mutex or the exclusive nfsd lock.
258  */
259 static void
260 nfsrv_clientlock(bool mlocked)
261 {
262 	int igotlock;
263 
264 	if (mlocked) {
265 		NFSLOCKSTATE();
266 	} else {
267 		NFSLOCKV4ROOTMUTEX();
268 		nfsv4_relref(&nfsv4rootfs_lock);
269 		do {
270 			igotlock = nfsv4_lock(&nfsv4rootfs_lock, 1, NULL,
271 			    NFSV4ROOTLOCKMUTEXPTR, NULL);
272 		} while (!igotlock);
273 		NFSUNLOCKV4ROOTMUTEX();
274 	}
275 }
276 
277 /*
278  * Unlock the client structure.
279  */
280 static void
281 nfsrv_clientunlock(bool mlocked)
282 {
283 
284 	if (mlocked) {
285 		NFSUNLOCKSTATE();
286 	} else {
287 		NFSLOCKV4ROOTMUTEX();
288 		nfsv4_unlock(&nfsv4rootfs_lock, 1);
289 		NFSUNLOCKV4ROOTMUTEX();
290 	}
291 }
292 
293 /*
294  * Scan the client list for a match and either return the current one,
295  * create a new entry or return an error.
296  * If returning a non-error, the clp structure must either be linked into
297  * the client list or free'd.
298  */
299 int
300 nfsrv_setclient(struct nfsrv_descript *nd, struct nfsclient **new_clpp,
301     nfsquad_t *clientidp, nfsquad_t *confirmp, NFSPROC_T *p)
302 {
303 	struct nfsclient *clp = NULL, *new_clp = *new_clpp;
304 	int i, error = 0, ret;
305 	struct nfsstate *stp, *tstp;
306 #ifdef INET
307 	struct sockaddr_in *sin, *rin;
308 #endif
309 #ifdef INET6
310 	struct sockaddr_in6 *sin6, *rin6;
311 #endif
312 	struct nfsdsession *sep, *nsep;
313 	SVCXPRT *old_xprt;
314 	struct nfssessionhead old_sess;
315 	int zapit = 0, gotit, hasstate = 0;
316 	bool mlocked;
317 	static u_int64_t confirm_index = 0;
318 
319 	/*
320 	 * Check for state resource limit exceeded.
321 	 */
322 	if (nfsrv_openpluslock > nfsrv_v4statelimit) {
323 		error = NFSERR_RESOURCE;
324 		goto out;
325 	}
326 
327 	if (nfsrv_issuedelegs == 0 ||
328 	    ((nd->nd_flag & ND_GSS) != 0 && nfsrv_nogsscallback != 0))
329 		/*
330 		 * Don't do callbacks when delegations are disabled or
331 		 * for AUTH_GSS unless enabled via nfsrv_nogsscallback.
332 		 * If establishing a callback connection is attempted
333 		 * when a firewall is blocking the callback path, the
334 		 * server may wait too long for the connect attempt to
335 		 * succeed during the Open. Some clients, such as Linux,
336 		 * may timeout and give up on the Open before the server
337 		 * replies. Also, since AUTH_GSS callbacks are not
338 		 * yet interoperability tested, they might cause the
339 		 * server to crap out, if they get past the Init call to
340 		 * the client.
341 		 */
342 		new_clp->lc_program = 0;
343 
344 	mlocked = true;
345 	if (nfsrv_dolocallocks != 0)
346 		mlocked = false;
347 	/* Lock out other nfsd threads */
348 	nfsrv_clientlock(mlocked);
349 
350 	/*
351 	 * Search for a match in the client list.
352 	 */
353 	gotit = i = 0;
354 	while (i < nfsrv_clienthashsize && !gotit) {
355 	    LIST_FOREACH(clp, &NFSD_VNET(nfsclienthash)[i], lc_hash) {
356 		if (new_clp->lc_idlen == clp->lc_idlen &&
357 		    !NFSBCMP(new_clp->lc_id, clp->lc_id, clp->lc_idlen)) {
358 			gotit = 1;
359 			break;
360 		}
361 	    }
362 	    if (gotit == 0)
363 		i++;
364 	}
365 	old_xprt = NULL;
366 	if (!gotit ||
367 	    (clp->lc_flags & (LCL_NEEDSCONFIRM | LCL_ADMINREVOKED))) {
368 		if ((nd->nd_flag & ND_NFSV41) != 0 && confirmp->lval[1] != 0) {
369 			/*
370 			 * For NFSv4.1, if confirmp->lval[1] is non-zero, the
371 			 * client is trying to update a confirmed clientid.
372 			 */
373 			nfsrv_clientunlock(mlocked);
374 			confirmp->lval[1] = 0;
375 			error = NFSERR_NOENT;
376 			goto out;
377 		}
378 		/*
379 		 * Get rid of the old one.
380 		 */
381 		if (i != nfsrv_clienthashsize) {
382 			LIST_REMOVE(clp, lc_hash);
383 			if (mlocked)
384 				nfsrv_cleanclient(clp, p, true, &old_xprt);
385 			else
386 				nfsrv_cleanclient(clp, p, false, NULL);
387 			nfsrv_freedeleglist(&clp->lc_deleg);
388 			nfsrv_freedeleglist(&clp->lc_olddeleg);
389 			zapit = 1;
390 		}
391 		/*
392 		 * Add it after assigning a client id to it.
393 		 */
394 		new_clp->lc_flags |= LCL_NEEDSCONFIRM;
395 		if ((nd->nd_flag & ND_NFSV41) != 0) {
396 			confirmp->lval[0] = ++confirm_index;
397 			new_clp->lc_confirm.lval[0] = confirmp->lval[0] - 1;
398 		} else
399 			confirmp->qval = new_clp->lc_confirm.qval =
400 			    ++confirm_index;
401 		clientidp->lval[0] = new_clp->lc_clientid.lval[0] =
402 		    NFSD_VNET(nfsrvboottime);
403 		clientidp->lval[1] = new_clp->lc_clientid.lval[1] =
404 		    nfsrv_nextclientindex();
405 		new_clp->lc_stateindex = 0;
406 		new_clp->lc_statemaxindex = 0;
407 		new_clp->lc_prevsess = 0;
408 		new_clp->lc_cbref = 0;
409 		new_clp->lc_expiry = nfsrv_leaseexpiry();
410 		LIST_INIT(&new_clp->lc_open);
411 		LIST_INIT(&new_clp->lc_deleg);
412 		LIST_INIT(&new_clp->lc_olddeleg);
413 		LIST_INIT(&new_clp->lc_session);
414 		for (i = 0; i < nfsrv_statehashsize; i++)
415 			LIST_INIT(&new_clp->lc_stateid[i]);
416 		LIST_INSERT_HEAD(NFSCLIENTHASH(new_clp->lc_clientid), new_clp,
417 		    lc_hash);
418 		NFSD_VNET(nfsstatsv1_p)->srvclients++;
419 		nfsrv_openpluslock++;
420 		nfsrv_clients++;
421 		nfsrv_clientunlock(mlocked);
422 		if (zapit != 0) {
423 			if (old_xprt != NULL)
424 				SVC_RELEASE(old_xprt);
425 			nfsrv_zapclient(clp, p);
426 		}
427 		*new_clpp = NULL;
428 		goto out;
429 	}
430 
431 	/*
432 	 * Now, handle the cases where the id is already issued.
433 	 */
434 	if (nfsrv_notsamecredname(NFSV4OP_EXCHANGEID, nd, clp)) {
435 	    /*
436 	     * Check to see if there is expired state that should go away.
437 	     */
438 	    if (clp->lc_expiry < NFSD_MONOSEC &&
439 	        (!LIST_EMPTY(&clp->lc_open) || !LIST_EMPTY(&clp->lc_deleg))) {
440 		if (mlocked)
441 		    nfsrv_cleanclient(clp, p, true, &old_xprt);
442 		else
443 		    nfsrv_cleanclient(clp, p, false, NULL);
444 		nfsrv_freedeleglist(&clp->lc_deleg);
445 	    }
446 
447 	    /*
448 	     * If there is outstanding state, then reply NFSERR_CLIDINUSE per
449 	     * RFC3530 Sec. 8.1.2 last para.
450 	     */
451 	    if (!LIST_EMPTY(&clp->lc_deleg)) {
452 		hasstate = 1;
453 	    } else if (LIST_EMPTY(&clp->lc_open)) {
454 		hasstate = 0;
455 	    } else {
456 		hasstate = 0;
457 		/* Look for an Open on the OpenOwner */
458 		LIST_FOREACH(stp, &clp->lc_open, ls_list) {
459 		    if (!LIST_EMPTY(&stp->ls_open)) {
460 			hasstate = 1;
461 			break;
462 		    }
463 		}
464 	    }
465 	    if (hasstate) {
466 		/*
467 		 * If the uid doesn't match, return NFSERR_CLIDINUSE after
468 		 * filling out the correct ipaddr and portnum.
469 		 */
470 		switch (clp->lc_req.nr_nam->sa_family) {
471 #ifdef INET
472 		case AF_INET:
473 			sin = (struct sockaddr_in *)new_clp->lc_req.nr_nam;
474 			rin = (struct sockaddr_in *)clp->lc_req.nr_nam;
475 			sin->sin_addr.s_addr = rin->sin_addr.s_addr;
476 			sin->sin_port = rin->sin_port;
477 			break;
478 #endif
479 #ifdef INET6
480 		case AF_INET6:
481 			sin6 = (struct sockaddr_in6 *)new_clp->lc_req.nr_nam;
482 			rin6 = (struct sockaddr_in6 *)clp->lc_req.nr_nam;
483 			sin6->sin6_addr = rin6->sin6_addr;
484 			sin6->sin6_port = rin6->sin6_port;
485 			break;
486 #endif
487 		}
488 		nfsrv_clientunlock(mlocked);
489 		if (old_xprt != NULL)
490 			SVC_RELEASE(old_xprt);
491 		error = NFSERR_CLIDINUSE;
492 		goto out;
493 	    }
494 	}
495 
496 	if (NFSBCMP(new_clp->lc_verf, clp->lc_verf, NFSX_VERF)) {
497 		/*
498 		 * If the verifier has changed, the client has rebooted
499 		 * and a new client id is issued. The old state info
500 		 * can be thrown away once the SetClientID_Confirm or
501 		 * Create_Session that confirms the clientid occurs.
502 		 */
503 		LIST_REMOVE(clp, lc_hash);
504 
505 		LIST_NEWHEAD(&old_sess, &clp->lc_session, sess_list);
506 
507 		new_clp->lc_flags |= LCL_NEEDSCONFIRM;
508 		if ((nd->nd_flag & ND_NFSV41) != 0) {
509 			confirmp->lval[0] = ++confirm_index;
510 			new_clp->lc_confirm.lval[0] = confirmp->lval[0] - 1;
511 		} else
512 			confirmp->qval = new_clp->lc_confirm.qval =
513 			    ++confirm_index;
514 		clientidp->lval[0] = new_clp->lc_clientid.lval[0] =
515 		    NFSD_VNET(nfsrvboottime);
516 		clientidp->lval[1] = new_clp->lc_clientid.lval[1] =
517 		    nfsrv_nextclientindex();
518 		new_clp->lc_stateindex = 0;
519 		new_clp->lc_statemaxindex = 0;
520 		new_clp->lc_prevsess = 0;
521 		new_clp->lc_cbref = 0;
522 		new_clp->lc_expiry = nfsrv_leaseexpiry();
523 
524 		/*
525 		 * Save the state until confirmed.
526 		 */
527 		LIST_NEWHEAD(&new_clp->lc_open, &clp->lc_open, ls_list);
528 		LIST_FOREACH(tstp, &new_clp->lc_open, ls_list)
529 			tstp->ls_clp = new_clp;
530 		LIST_NEWHEAD(&new_clp->lc_deleg, &clp->lc_deleg, ls_list);
531 		LIST_FOREACH(tstp, &new_clp->lc_deleg, ls_list)
532 			tstp->ls_clp = new_clp;
533 		LIST_NEWHEAD(&new_clp->lc_olddeleg, &clp->lc_olddeleg,
534 		    ls_list);
535 		LIST_FOREACH(tstp, &new_clp->lc_olddeleg, ls_list)
536 			tstp->ls_clp = new_clp;
537 		for (i = 0; i < nfsrv_statehashsize; i++) {
538 			LIST_NEWHEAD(&new_clp->lc_stateid[i],
539 			    &clp->lc_stateid[i], ls_hash);
540 			LIST_FOREACH(tstp, &new_clp->lc_stateid[i], ls_hash)
541 				tstp->ls_clp = new_clp;
542 		}
543 		LIST_INIT(&new_clp->lc_session);
544 		LIST_INSERT_HEAD(NFSCLIENTHASH(new_clp->lc_clientid), new_clp,
545 		    lc_hash);
546 		NFSD_VNET(nfsstatsv1_p)->srvclients++;
547 		nfsrv_openpluslock++;
548 		nfsrv_clients++;
549 		if (!mlocked) {
550 			nfsrv_clientunlock(mlocked);
551 			NFSLOCKSTATE();
552 		}
553 
554 		/*
555 		 * Must wait until any outstanding callback on the old clp
556 		 * completes.
557 		 */
558 		while (clp->lc_cbref) {
559 			clp->lc_flags |= LCL_WAKEUPWANTED;
560 			(void)mtx_sleep(clp, NFSSTATEMUTEXPTR, PZERO - 1,
561 			    "nfsd clp", 10 * hz);
562 		}
563 		NFSUNLOCKSTATE();
564 		if (old_xprt != NULL)
565 			SVC_RELEASE(old_xprt);
566 		/* Get rid of all sessions on this clientid. */
567 		LIST_FOREACH_SAFE(sep, &old_sess, sess_list, nsep) {
568 			ret = nfsrv_freesession(NULL, sep, NULL, false, NULL);
569 			if (ret != 0)
570 				printf("nfsrv_setclient: verifier changed free"
571 				    " session failed=%d\n", ret);
572 		}
573 
574 		nfsrv_zapclient(clp, p);
575 		*new_clpp = NULL;
576 		goto out;
577 	}
578 
579 	/* For NFSv4.1, mark that we found a confirmed clientid. */
580 	if ((nd->nd_flag & ND_NFSV41) != 0) {
581 		clientidp->lval[0] = clp->lc_clientid.lval[0];
582 		clientidp->lval[1] = clp->lc_clientid.lval[1];
583 		confirmp->lval[0] = 0;	/* Ignored by client */
584 		confirmp->lval[1] = 1;
585 	} else {
586 		/*
587 		 * id and verifier match, so update the net address info
588 		 * and get rid of any existing callback authentication
589 		 * handle, so a new one will be acquired.
590 		 */
591 		LIST_REMOVE(clp, lc_hash);
592 		new_clp->lc_flags |= (LCL_NEEDSCONFIRM | LCL_DONTCLEAN);
593 		new_clp->lc_expiry = nfsrv_leaseexpiry();
594 		confirmp->qval = new_clp->lc_confirm.qval = ++confirm_index;
595 		clientidp->lval[0] = new_clp->lc_clientid.lval[0] =
596 		    clp->lc_clientid.lval[0];
597 		clientidp->lval[1] = new_clp->lc_clientid.lval[1] =
598 		    clp->lc_clientid.lval[1];
599 		new_clp->lc_delegtime = clp->lc_delegtime;
600 		new_clp->lc_stateindex = clp->lc_stateindex;
601 		new_clp->lc_statemaxindex = clp->lc_statemaxindex;
602 		new_clp->lc_cbref = 0;
603 		LIST_NEWHEAD(&new_clp->lc_open, &clp->lc_open, ls_list);
604 		LIST_FOREACH(tstp, &new_clp->lc_open, ls_list)
605 			tstp->ls_clp = new_clp;
606 		LIST_NEWHEAD(&new_clp->lc_deleg, &clp->lc_deleg, ls_list);
607 		LIST_FOREACH(tstp, &new_clp->lc_deleg, ls_list)
608 			tstp->ls_clp = new_clp;
609 		LIST_NEWHEAD(&new_clp->lc_olddeleg, &clp->lc_olddeleg, ls_list);
610 		LIST_FOREACH(tstp, &new_clp->lc_olddeleg, ls_list)
611 			tstp->ls_clp = new_clp;
612 		for (i = 0; i < nfsrv_statehashsize; i++) {
613 			LIST_NEWHEAD(&new_clp->lc_stateid[i],
614 			    &clp->lc_stateid[i], ls_hash);
615 			LIST_FOREACH(tstp, &new_clp->lc_stateid[i], ls_hash)
616 				tstp->ls_clp = new_clp;
617 		}
618 		LIST_INIT(&new_clp->lc_session);
619 		LIST_INSERT_HEAD(NFSCLIENTHASH(new_clp->lc_clientid), new_clp,
620 		    lc_hash);
621 		NFSD_VNET(nfsstatsv1_p)->srvclients++;
622 		nfsrv_openpluslock++;
623 		nfsrv_clients++;
624 	}
625 	if (!mlocked)
626 		nfsrv_clientunlock(mlocked);
627 
628 	if ((nd->nd_flag & ND_NFSV41) == 0) {
629 		/*
630 		 * Must wait until any outstanding callback on the old clp
631 		 * completes.
632 		 */
633 		if (!mlocked)
634 			NFSLOCKSTATE();
635 		while (clp->lc_cbref) {
636 			clp->lc_flags |= LCL_WAKEUPWANTED;
637 			(void)mtx_sleep(clp, NFSSTATEMUTEXPTR, PZERO - 1,
638 			    "nfsdclp", 10 * hz);
639 		}
640 		NFSUNLOCKSTATE();
641 		if (old_xprt != NULL)
642 			SVC_RELEASE(old_xprt);
643 		nfsrv_zapclient(clp, p);
644 		*new_clpp = NULL;
645 	} else {
646 		if (mlocked)
647 			NFSUNLOCKSTATE();
648 		if (old_xprt != NULL)
649 			SVC_RELEASE(old_xprt);
650 	}
651 
652 out:
653 	NFSEXITCODE2(error, nd);
654 	return (error);
655 }
656 
657 /*
658  * Check to see if the client id exists and optionally confirm it.
659  */
660 int
661 nfsrv_getclient(nfsquad_t clientid, int opflags, struct nfsclient **clpp,
662     struct nfsdsession *nsep, nfsquad_t confirm, uint32_t cbprogram,
663     struct nfsrv_descript *nd, NFSPROC_T *p)
664 {
665 	struct nfsclient *clp;
666 	struct nfsstate *stp;
667 	int i;
668 	struct nfsclienthashhead *hp;
669 	int error = 0, doneok, igotlock;
670 	struct nfssessionhash *shp;
671 	struct nfsdsession *sep;
672 	uint64_t sessid[2];
673 	CLIENT *client;
674 	SVCXPRT *old_xprt;
675 	bool mlocked, sess_replay;
676 	static uint64_t next_sess = 0;
677 
678 	if (clpp)
679 		*clpp = NULL;
680 	if ((nd == NULL || (nd->nd_flag & ND_NFSV41) == 0 ||
681 	    opflags != CLOPS_RENEW) && NFSD_VNET(nfsrvboottime) !=
682 	    clientid.lval[0]) {
683 		error = NFSERR_STALECLIENTID;
684 		goto out;
685 	}
686 
687 	/*
688 	 * If called with opflags == CLOPS_RENEW, the State Lock is
689 	 * already held. Otherwise, we need to get either that or,
690 	 * for the case of Confirm, lock out the nfsd threads.
691 	 */
692 	client = NULL;
693 	old_xprt = NULL;
694 	mlocked = true;
695 	if (nfsrv_dolocallocks != 0)
696 		mlocked = false;
697 	if (opflags & CLOPS_CONFIRM) {
698 		if (nsep != NULL &&
699 		    (nsep->sess_crflags & NFSV4CRSESS_CONNBACKCHAN) != 0)
700 			client = (struct __rpc_client *)
701 			    clnt_bck_create(nd->nd_xprt->xp_socket,
702 			    cbprogram, NFSV4_CBVERS);
703 		if (mlocked) {
704 			nfsrv_clientlock(mlocked);
705 		} else {
706 			NFSLOCKV4ROOTMUTEX();
707 			nfsv4_relref(&nfsv4rootfs_lock);
708 			do {
709 				igotlock = nfsv4_lock(&nfsv4rootfs_lock, 1,
710 				    NULL, NFSV4ROOTLOCKMUTEXPTR, NULL);
711 			} while (!igotlock);
712 		}
713 		/*
714 		 * Create a new sessionid here, since we need to do it where
715 		 * there is a mutex held to serialize update of next_sess.
716 		 */
717 		if ((nd->nd_flag & ND_NFSV41) != 0) {
718 			sessid[0] = ++next_sess;
719 			sessid[1] = clientid.qval;
720 		}
721 		if (!mlocked)
722 			NFSUNLOCKV4ROOTMUTEX();
723 	} else if (opflags != CLOPS_RENEW) {
724 		NFSLOCKSTATE();
725 	}
726 
727 	/* For NFSv4.1, the clp is acquired from the associated session. */
728 	if (nd != NULL && (nd->nd_flag & ND_NFSV41) != 0 &&
729 	    opflags == CLOPS_RENEW) {
730 		clp = NULL;
731 		if ((nd->nd_flag & ND_HASSEQUENCE) != 0) {
732 			shp = NFSSESSIONHASH(nd->nd_sessionid);
733 			NFSLOCKSESSION(shp);
734 			sep = nfsrv_findsession(nd->nd_sessionid);
735 			if (sep != NULL)
736 				clp = sep->sess_clp;
737 			NFSUNLOCKSESSION(shp);
738 		}
739 	} else {
740 		hp = NFSCLIENTHASH(clientid);
741 		LIST_FOREACH(clp, hp, lc_hash) {
742 			if (clp->lc_clientid.lval[1] == clientid.lval[1])
743 				break;
744 		}
745 	}
746 	if (clp == NULL) {
747 		if (opflags & CLOPS_CONFIRM)
748 			error = NFSERR_STALECLIENTID;
749 		else
750 			error = NFSERR_EXPIRED;
751 	} else if (clp->lc_flags & LCL_ADMINREVOKED) {
752 		/*
753 		 * If marked admin revoked, just return the error.
754 		 */
755 		error = NFSERR_ADMINREVOKED;
756 	}
757 	if (error) {
758 		if (opflags & CLOPS_CONFIRM) {
759 			nfsrv_clientunlock(mlocked);
760 			if (client != NULL)
761 				CLNT_RELEASE(client);
762 		} else if (opflags != CLOPS_RENEW) {
763 			NFSUNLOCKSTATE();
764 		}
765 		goto out;
766 	}
767 
768 	/*
769 	 * Perform any operations specified by the opflags.
770 	 */
771 	if (opflags & CLOPS_CONFIRM) {
772 		sess_replay = false;
773 		if ((nd->nd_flag & ND_NFSV41) != 0) {
774 		    /*
775 		     * For the case where lc_confirm.lval[0] == confirm.lval[0],
776 		     * use the new session, but with the previous sessionid.
777 		     * This is not exactly what the RFC describes, but should
778 		     * result in the same reply as the previous CreateSession.
779 		     */
780 		    if (clp->lc_confirm.lval[0] + 1 == confirm.lval[0]) {
781 			clp->lc_confirm.lval[0] = confirm.lval[0];
782 			clp->lc_prevsess = sessid[0];
783 		    } else if (clp->lc_confirm.lval[0] == confirm.lval[0]) {
784 			if (clp->lc_prevsess == 0)
785 			    error = NFSERR_SEQMISORDERED;
786 			else
787 			    sessid[0] = clp->lc_prevsess;
788 			sess_replay = true;
789 		    } else
790 			error = NFSERR_SEQMISORDERED;
791 		} else if ((nd->nd_flag & ND_NFSV41) == 0 &&
792 		     clp->lc_confirm.qval != confirm.qval)
793 			error = NFSERR_STALECLIENTID;
794 		if (error == 0 && nfsrv_notsamecredname(NFSV4OP_CREATESESSION,
795 		    nd, clp))
796 			error = NFSERR_CLIDINUSE;
797 
798 		if (!error) {
799 		    if ((clp->lc_flags & (LCL_NEEDSCONFIRM | LCL_DONTCLEAN)) ==
800 			LCL_NEEDSCONFIRM) {
801 			/*
802 			 * Hang onto the delegations (as old delegations)
803 			 * for an Open with CLAIM_DELEGATE_PREV unless in
804 			 * grace, but get rid of the rest of the state.
805 			 */
806 			if (mlocked)
807 				nfsrv_cleanclient(clp, p, true, &old_xprt);
808 			else
809 				nfsrv_cleanclient(clp, p, false, NULL);
810 			nfsrv_freedeleglist(&clp->lc_olddeleg);
811 			if (nfsrv_checkgrace(nd, clp, 0)) {
812 			    /* In grace, so just delete delegations */
813 			    nfsrv_freedeleglist(&clp->lc_deleg);
814 			} else {
815 			    LIST_FOREACH(stp, &clp->lc_deleg, ls_list)
816 				stp->ls_flags |= NFSLCK_OLDDELEG;
817 			    clp->lc_delegtime = NFSD_MONOSEC +
818 				nfsrv_lease + NFSRV_LEASEDELTA;
819 			    LIST_NEWHEAD(&clp->lc_olddeleg, &clp->lc_deleg,
820 				ls_list);
821 			}
822 			if ((nd->nd_flag & ND_NFSV41) != 0)
823 			    clp->lc_program = cbprogram;
824 		    }
825 		    clp->lc_flags &= ~(LCL_NEEDSCONFIRM | LCL_DONTCLEAN);
826 		    if (clp->lc_program)
827 			clp->lc_flags |= LCL_NEEDSCBNULL;
828 		    /* For NFSv4.1, link the session onto the client. */
829 		    if (nsep != NULL) {
830 			/* Hold a reference on the xprt for a backchannel. */
831 			if ((nsep->sess_crflags & NFSV4CRSESS_CONNBACKCHAN)
832 			    != 0 && !sess_replay) {
833 			    if (clp->lc_req.nr_client == NULL) {
834 				clp->lc_req.nr_client = client;
835 				client = NULL;
836 			    }
837 			    if (clp->lc_req.nr_client != NULL) {
838 				SVC_ACQUIRE(nd->nd_xprt);
839 				CLNT_ACQUIRE(clp->lc_req.nr_client);
840 				nd->nd_xprt->xp_p2 = clp->lc_req.nr_client;
841 				/* Disable idle timeout. */
842 				nd->nd_xprt->xp_idletimeout = 0;
843 				nsep->sess_cbsess.nfsess_xprt = nd->nd_xprt;
844 			    } else
845 				nsep->sess_crflags &= ~NFSV4CRSESS_CONNBACKCHAN;
846 			}
847 			NFSBCOPY(sessid, nsep->sess_sessionid,
848 			    NFSX_V4SESSIONID);
849 			NFSBCOPY(sessid, nsep->sess_cbsess.nfsess_sessionid,
850 			    NFSX_V4SESSIONID);
851 			if (!sess_replay) {
852 			    shp = NFSSESSIONHASH(nsep->sess_sessionid);
853 			    if (!mlocked)
854 				NFSLOCKSTATE();
855 			    NFSLOCKSESSION(shp);
856 			    LIST_INSERT_HEAD(&shp->list, nsep, sess_hash);
857 			    LIST_INSERT_HEAD(&clp->lc_session, nsep, sess_list);
858 			    nsep->sess_clp = clp;
859 			    NFSUNLOCKSESSION(shp);
860 			    if (!mlocked)
861 				NFSUNLOCKSTATE();
862 			}
863 		    }
864 		}
865 	} else if (clp->lc_flags & LCL_NEEDSCONFIRM) {
866 		error = NFSERR_EXPIRED;
867 	}
868 
869 	/*
870 	 * If called by the Renew Op, we must check the principal.
871 	 */
872 	if (!error && (opflags & CLOPS_RENEWOP)) {
873 	    if (nfsrv_notsamecredname(0, nd, clp)) {
874 		doneok = 0;
875 		for (i = 0; i < nfsrv_statehashsize && doneok == 0; i++) {
876 		    LIST_FOREACH(stp, &clp->lc_stateid[i], ls_hash) {
877 			if ((stp->ls_flags & NFSLCK_OPEN) &&
878 			    stp->ls_uid == nd->nd_cred->cr_uid) {
879 				doneok = 1;
880 				break;
881 			}
882 		    }
883 		}
884 		if (!doneok)
885 			error = NFSERR_ACCES;
886 	    }
887 	    if (!error && (clp->lc_flags & LCL_CBDOWN))
888 		error = NFSERR_CBPATHDOWN;
889 	}
890 	if ((!error || error == NFSERR_CBPATHDOWN) &&
891 	     (opflags & CLOPS_RENEW)) {
892 		clp->lc_expiry = nfsrv_leaseexpiry();
893 	}
894 	if (opflags & CLOPS_CONFIRM) {
895 		nfsrv_clientunlock(mlocked);
896 		if (client != NULL)
897 			CLNT_RELEASE(client);
898 		if (old_xprt != NULL)
899 			SVC_RELEASE(old_xprt);
900 	} else if (opflags != CLOPS_RENEW) {
901 		NFSUNLOCKSTATE();
902 	}
903 	if (clpp)
904 		*clpp = clp;
905 
906 out:
907 	NFSEXITCODE2(error, nd);
908 	return (error);
909 }
910 
911 /*
912  * Perform the NFSv4.1 destroy clientid.
913  */
914 int
915 nfsrv_destroyclient(struct nfsrv_descript *nd, nfsquad_t clientid, NFSPROC_T *p)
916 {
917 	struct nfsclient *clp;
918 	struct nfsclienthashhead *hp;
919 	SVCXPRT *old_xprt;
920 	int error = 0, i;
921 	bool mlocked;
922 
923 	if (NFSD_VNET(nfsrvboottime) != clientid.lval[0]) {
924 		error = NFSERR_STALECLIENTID;
925 		goto out;
926 	}
927 
928 	mlocked = true;
929 	if (nfsrv_dolocallocks != 0)
930 		mlocked = false;
931 	/* Lock out other nfsd threads */
932 	nfsrv_clientlock(mlocked);
933 
934 	hp = NFSCLIENTHASH(clientid);
935 	LIST_FOREACH(clp, hp, lc_hash) {
936 		if (clp->lc_clientid.lval[1] == clientid.lval[1])
937 			break;
938 	}
939 	if (clp == NULL) {
940 		nfsrv_clientunlock(mlocked);
941 		/* Just return ok, since it is gone. */
942 		goto out;
943 	}
944 
945 	/* Check for the SP4_MACH_CRED case. */
946 	error = nfsrv_checkmachcred(NFSV4OP_DESTROYCLIENTID, nd, clp);
947 	if (error != 0) {
948 		nfsrv_clientunlock(mlocked);
949 		goto out;
950 	}
951 
952 	/*
953 	 * Free up all layouts on the clientid.  Should the client return the
954 	 * layouts?
955 	 */
956 	nfsrv_freelayoutlist(clientid);
957 
958 	/* Scan for state on the clientid. */
959 	for (i = 0; i < nfsrv_statehashsize; i++)
960 		if (!LIST_EMPTY(&clp->lc_stateid[i])) {
961 			nfsrv_clientunlock(mlocked);
962 			error = NFSERR_CLIENTIDBUSY;
963 			goto out;
964 		}
965 	if (!LIST_EMPTY(&clp->lc_session) || !LIST_EMPTY(&clp->lc_deleg)) {
966 		nfsrv_clientunlock(mlocked);
967 		error = NFSERR_CLIENTIDBUSY;
968 		goto out;
969 	}
970 
971 	/* Destroy the clientid and return ok. */
972 	old_xprt = NULL;
973 	if (mlocked)
974 		nfsrv_cleanclient(clp, p, true, &old_xprt);
975 	else
976 		nfsrv_cleanclient(clp, p, false, NULL);
977 	nfsrv_freedeleglist(&clp->lc_deleg);
978 	nfsrv_freedeleglist(&clp->lc_olddeleg);
979 	LIST_REMOVE(clp, lc_hash);
980 	nfsrv_clientunlock(mlocked);
981 	if (old_xprt != NULL)
982 		SVC_RELEASE(old_xprt);
983 	nfsrv_zapclient(clp, p);
984 out:
985 	NFSEXITCODE2(error, nd);
986 	return (error);
987 }
988 
989 /*
990  * Called from the new nfssvc syscall to admin revoke a clientid.
991  * Returns 0 for success, error otherwise.
992  */
993 int
994 nfsrv_adminrevoke(struct nfsd_clid *revokep, NFSPROC_T *p)
995 {
996 	struct nfsclient *clp = NULL;
997 	int i, error = 0;
998 	int gotit, igotlock;
999 
1000 	/*
1001 	 * First, lock out the nfsd so that state won't change while the
1002 	 * revocation record is being written to the stable storage restart
1003 	 * file.
1004 	 */
1005 	NFSLOCKV4ROOTMUTEX();
1006 	do {
1007 		igotlock = nfsv4_lock(&nfsv4rootfs_lock, 1, NULL,
1008 		    NFSV4ROOTLOCKMUTEXPTR, NULL);
1009 	} while (!igotlock);
1010 	NFSUNLOCKV4ROOTMUTEX();
1011 
1012 	/*
1013 	 * Search for a match in the client list.
1014 	 */
1015 	gotit = i = 0;
1016 	while (i < nfsrv_clienthashsize && !gotit) {
1017 	    LIST_FOREACH(clp, &NFSD_VNET(nfsclienthash)[i], lc_hash) {
1018 		if (revokep->nclid_idlen == clp->lc_idlen &&
1019 		    !NFSBCMP(revokep->nclid_id, clp->lc_id, clp->lc_idlen)) {
1020 			gotit = 1;
1021 			break;
1022 		}
1023 	    }
1024 	    i++;
1025 	}
1026 	if (!gotit) {
1027 		NFSLOCKV4ROOTMUTEX();
1028 		nfsv4_unlock(&nfsv4rootfs_lock, 0);
1029 		NFSUNLOCKV4ROOTMUTEX();
1030 		error = EPERM;
1031 		goto out;
1032 	}
1033 
1034 	/*
1035 	 * Now, write out the revocation record
1036 	 */
1037 	nfsrv_writestable(clp->lc_id, clp->lc_idlen, NFSNST_REVOKE, p);
1038 	nfsrv_backupstable();
1039 
1040 	/*
1041 	 * and clear out the state, marking the clientid revoked.
1042 	 */
1043 	clp->lc_flags &= ~LCL_CALLBACKSON;
1044 	clp->lc_flags |= LCL_ADMINREVOKED;
1045 	nfsrv_cleanclient(clp, p, false, NULL);
1046 	nfsrv_freedeleglist(&clp->lc_deleg);
1047 	nfsrv_freedeleglist(&clp->lc_olddeleg);
1048 	NFSLOCKV4ROOTMUTEX();
1049 	nfsv4_unlock(&nfsv4rootfs_lock, 0);
1050 	NFSUNLOCKV4ROOTMUTEX();
1051 
1052 out:
1053 	NFSEXITCODE(error);
1054 	return (error);
1055 }
1056 
1057 /*
1058  * Dump out stats for all clients. Called from nfssvc(2), that is used
1059  * nfsstatsv1.
1060  */
1061 void
1062 nfsrv_dumpclients(struct nfsd_dumpclients *dumpp, int maxcnt)
1063 {
1064 	struct nfsclient *clp;
1065 	int i = 0, cnt = 0;
1066 
1067 	/*
1068 	 * First, get a reference on the nfsv4rootfs_lock so that an
1069 	 * exclusive lock cannot be acquired while dumping the clients.
1070 	 */
1071 	NFSLOCKV4ROOTMUTEX();
1072 	nfsv4_getref(&nfsv4rootfs_lock, NULL, NFSV4ROOTLOCKMUTEXPTR, NULL);
1073 	NFSUNLOCKV4ROOTMUTEX();
1074 	NFSLOCKSTATE();
1075 	/*
1076 	 * Rattle through the client lists until done.
1077 	 */
1078 	while (i < nfsrv_clienthashsize && cnt < maxcnt) {
1079 	    clp = LIST_FIRST(&NFSD_VNET(nfsclienthash)[i]);
1080 	    while (clp != LIST_END(&NFSD_VNET(nfsclienthash)[i]) && cnt <
1081 		maxcnt) {
1082 		nfsrv_dumpaclient(clp, &dumpp[cnt]);
1083 		cnt++;
1084 		clp = LIST_NEXT(clp, lc_hash);
1085 	    }
1086 	    i++;
1087 	}
1088 	if (cnt < maxcnt)
1089 	    dumpp[cnt].ndcl_clid.nclid_idlen = 0;
1090 	NFSUNLOCKSTATE();
1091 	NFSLOCKV4ROOTMUTEX();
1092 	nfsv4_relref(&nfsv4rootfs_lock);
1093 	NFSUNLOCKV4ROOTMUTEX();
1094 }
1095 
1096 /*
1097  * Dump stats for a client. Must be called with the NFSSTATELOCK and spl'd.
1098  */
1099 static void
1100 nfsrv_dumpaclient(struct nfsclient *clp, struct nfsd_dumpclients *dumpp)
1101 {
1102 	struct nfsstate *stp, *openstp, *lckownstp;
1103 	struct nfslock *lop;
1104 	sa_family_t af;
1105 #ifdef INET
1106 	struct sockaddr_in *rin;
1107 #endif
1108 #ifdef INET6
1109 	struct sockaddr_in6 *rin6;
1110 #endif
1111 
1112 	dumpp->ndcl_nopenowners = dumpp->ndcl_nlockowners = 0;
1113 	dumpp->ndcl_nopens = dumpp->ndcl_nlocks = 0;
1114 	dumpp->ndcl_ndelegs = dumpp->ndcl_nolddelegs = 0;
1115 	dumpp->ndcl_flags = clp->lc_flags;
1116 	dumpp->ndcl_clid.nclid_idlen = clp->lc_idlen;
1117 	NFSBCOPY(clp->lc_id, dumpp->ndcl_clid.nclid_id, clp->lc_idlen);
1118 	af = clp->lc_req.nr_nam->sa_family;
1119 	dumpp->ndcl_addrfam = af;
1120 	switch (af) {
1121 #ifdef INET
1122 	case AF_INET:
1123 		rin = (struct sockaddr_in *)clp->lc_req.nr_nam;
1124 		dumpp->ndcl_cbaddr.sin_addr = rin->sin_addr;
1125 		break;
1126 #endif
1127 #ifdef INET6
1128 	case AF_INET6:
1129 		rin6 = (struct sockaddr_in6 *)clp->lc_req.nr_nam;
1130 		dumpp->ndcl_cbaddr.sin6_addr = rin6->sin6_addr;
1131 		break;
1132 #endif
1133 	}
1134 
1135 	/*
1136 	 * Now, scan the state lists and total up the opens and locks.
1137 	 */
1138 	LIST_FOREACH(stp, &clp->lc_open, ls_list) {
1139 	    dumpp->ndcl_nopenowners++;
1140 	    LIST_FOREACH(openstp, &stp->ls_open, ls_list) {
1141 		dumpp->ndcl_nopens++;
1142 		LIST_FOREACH(lckownstp, &openstp->ls_open, ls_list) {
1143 		    dumpp->ndcl_nlockowners++;
1144 		    LIST_FOREACH(lop, &lckownstp->ls_lock, lo_lckowner) {
1145 			dumpp->ndcl_nlocks++;
1146 		    }
1147 		}
1148 	    }
1149 	}
1150 
1151 	/*
1152 	 * and the delegation lists.
1153 	 */
1154 	LIST_FOREACH(stp, &clp->lc_deleg, ls_list) {
1155 	    dumpp->ndcl_ndelegs++;
1156 	}
1157 	LIST_FOREACH(stp, &clp->lc_olddeleg, ls_list) {
1158 	    dumpp->ndcl_nolddelegs++;
1159 	}
1160 }
1161 
1162 /*
1163  * Dump out lock stats for a file.
1164  */
1165 void
1166 nfsrv_dumplocks(vnode_t vp, struct nfsd_dumplocks *ldumpp, int maxcnt,
1167     NFSPROC_T *p)
1168 {
1169 	struct nfsstate *stp;
1170 	struct nfslock *lop;
1171 	int cnt = 0;
1172 	struct nfslockfile *lfp;
1173 	sa_family_t af;
1174 #ifdef INET
1175 	struct sockaddr_in *rin;
1176 #endif
1177 #ifdef INET6
1178 	struct sockaddr_in6 *rin6;
1179 #endif
1180 	int ret;
1181 	fhandle_t nfh;
1182 
1183 	ret = nfsrv_getlockfh(vp, 0, NULL, &nfh, p);
1184 	/*
1185 	 * First, get a reference on the nfsv4rootfs_lock so that an
1186 	 * exclusive lock on it cannot be acquired while dumping the locks.
1187 	 */
1188 	NFSLOCKV4ROOTMUTEX();
1189 	nfsv4_getref(&nfsv4rootfs_lock, NULL, NFSV4ROOTLOCKMUTEXPTR, NULL);
1190 	NFSUNLOCKV4ROOTMUTEX();
1191 	NFSLOCKSTATE();
1192 	if (!ret)
1193 		ret = nfsrv_getlockfile(0, NULL, &lfp, &nfh, 0);
1194 	if (ret) {
1195 		ldumpp[0].ndlck_clid.nclid_idlen = 0;
1196 		NFSUNLOCKSTATE();
1197 		NFSLOCKV4ROOTMUTEX();
1198 		nfsv4_relref(&nfsv4rootfs_lock);
1199 		NFSUNLOCKV4ROOTMUTEX();
1200 		return;
1201 	}
1202 
1203 	/*
1204 	 * For each open share on file, dump it out.
1205 	 */
1206 	stp = LIST_FIRST(&lfp->lf_open);
1207 	while (stp != LIST_END(&lfp->lf_open) && cnt < maxcnt) {
1208 		ldumpp[cnt].ndlck_flags = stp->ls_flags;
1209 		ldumpp[cnt].ndlck_stateid.seqid = stp->ls_stateid.seqid;
1210 		ldumpp[cnt].ndlck_stateid.other[0] = stp->ls_stateid.other[0];
1211 		ldumpp[cnt].ndlck_stateid.other[1] = stp->ls_stateid.other[1];
1212 		ldumpp[cnt].ndlck_stateid.other[2] = stp->ls_stateid.other[2];
1213 		ldumpp[cnt].ndlck_owner.nclid_idlen =
1214 		    stp->ls_openowner->ls_ownerlen;
1215 		NFSBCOPY(stp->ls_openowner->ls_owner,
1216 		    ldumpp[cnt].ndlck_owner.nclid_id,
1217 		    stp->ls_openowner->ls_ownerlen);
1218 		ldumpp[cnt].ndlck_clid.nclid_idlen = stp->ls_clp->lc_idlen;
1219 		NFSBCOPY(stp->ls_clp->lc_id, ldumpp[cnt].ndlck_clid.nclid_id,
1220 		    stp->ls_clp->lc_idlen);
1221 		af = stp->ls_clp->lc_req.nr_nam->sa_family;
1222 		ldumpp[cnt].ndlck_addrfam = af;
1223 		switch (af) {
1224 #ifdef INET
1225 		case AF_INET:
1226 			rin = (struct sockaddr_in *)stp->ls_clp->lc_req.nr_nam;
1227 			ldumpp[cnt].ndlck_cbaddr.sin_addr = rin->sin_addr;
1228 			break;
1229 #endif
1230 #ifdef INET6
1231 		case AF_INET6:
1232 			rin6 = (struct sockaddr_in6 *)
1233 			    stp->ls_clp->lc_req.nr_nam;
1234 			ldumpp[cnt].ndlck_cbaddr.sin6_addr = rin6->sin6_addr;
1235 			break;
1236 #endif
1237 		}
1238 		stp = LIST_NEXT(stp, ls_file);
1239 		cnt++;
1240 	}
1241 
1242 	/*
1243 	 * and all locks.
1244 	 */
1245 	lop = LIST_FIRST(&lfp->lf_lock);
1246 	while (lop != LIST_END(&lfp->lf_lock) && cnt < maxcnt) {
1247 		stp = lop->lo_stp;
1248 		ldumpp[cnt].ndlck_flags = lop->lo_flags;
1249 		ldumpp[cnt].ndlck_first = lop->lo_first;
1250 		ldumpp[cnt].ndlck_end = lop->lo_end;
1251 		ldumpp[cnt].ndlck_stateid.seqid = stp->ls_stateid.seqid;
1252 		ldumpp[cnt].ndlck_stateid.other[0] = stp->ls_stateid.other[0];
1253 		ldumpp[cnt].ndlck_stateid.other[1] = stp->ls_stateid.other[1];
1254 		ldumpp[cnt].ndlck_stateid.other[2] = stp->ls_stateid.other[2];
1255 		ldumpp[cnt].ndlck_owner.nclid_idlen = stp->ls_ownerlen;
1256 		NFSBCOPY(stp->ls_owner, ldumpp[cnt].ndlck_owner.nclid_id,
1257 		    stp->ls_ownerlen);
1258 		ldumpp[cnt].ndlck_clid.nclid_idlen = stp->ls_clp->lc_idlen;
1259 		NFSBCOPY(stp->ls_clp->lc_id, ldumpp[cnt].ndlck_clid.nclid_id,
1260 		    stp->ls_clp->lc_idlen);
1261 		af = stp->ls_clp->lc_req.nr_nam->sa_family;
1262 		ldumpp[cnt].ndlck_addrfam = af;
1263 		switch (af) {
1264 #ifdef INET
1265 		case AF_INET:
1266 			rin = (struct sockaddr_in *)stp->ls_clp->lc_req.nr_nam;
1267 			ldumpp[cnt].ndlck_cbaddr.sin_addr = rin->sin_addr;
1268 			break;
1269 #endif
1270 #ifdef INET6
1271 		case AF_INET6:
1272 			rin6 = (struct sockaddr_in6 *)
1273 			    stp->ls_clp->lc_req.nr_nam;
1274 			ldumpp[cnt].ndlck_cbaddr.sin6_addr = rin6->sin6_addr;
1275 			break;
1276 #endif
1277 		}
1278 		lop = LIST_NEXT(lop, lo_lckfile);
1279 		cnt++;
1280 	}
1281 
1282 	/*
1283 	 * and the delegations.
1284 	 */
1285 	stp = LIST_FIRST(&lfp->lf_deleg);
1286 	while (stp != LIST_END(&lfp->lf_deleg) && cnt < maxcnt) {
1287 		ldumpp[cnt].ndlck_flags = stp->ls_flags;
1288 		ldumpp[cnt].ndlck_stateid.seqid = stp->ls_stateid.seqid;
1289 		ldumpp[cnt].ndlck_stateid.other[0] = stp->ls_stateid.other[0];
1290 		ldumpp[cnt].ndlck_stateid.other[1] = stp->ls_stateid.other[1];
1291 		ldumpp[cnt].ndlck_stateid.other[2] = stp->ls_stateid.other[2];
1292 		ldumpp[cnt].ndlck_owner.nclid_idlen = 0;
1293 		ldumpp[cnt].ndlck_clid.nclid_idlen = stp->ls_clp->lc_idlen;
1294 		NFSBCOPY(stp->ls_clp->lc_id, ldumpp[cnt].ndlck_clid.nclid_id,
1295 		    stp->ls_clp->lc_idlen);
1296 		af = stp->ls_clp->lc_req.nr_nam->sa_family;
1297 		ldumpp[cnt].ndlck_addrfam = af;
1298 		switch (af) {
1299 #ifdef INET
1300 		case AF_INET:
1301 			rin = (struct sockaddr_in *)stp->ls_clp->lc_req.nr_nam;
1302 			ldumpp[cnt].ndlck_cbaddr.sin_addr = rin->sin_addr;
1303 			break;
1304 #endif
1305 #ifdef INET6
1306 		case AF_INET6:
1307 			rin6 = (struct sockaddr_in6 *)
1308 			    stp->ls_clp->lc_req.nr_nam;
1309 			ldumpp[cnt].ndlck_cbaddr.sin6_addr = rin6->sin6_addr;
1310 			break;
1311 #endif
1312 		}
1313 		stp = LIST_NEXT(stp, ls_file);
1314 		cnt++;
1315 	}
1316 
1317 	/*
1318 	 * If list isn't full, mark end of list by setting the client name
1319 	 * to zero length.
1320 	 */
1321 	if (cnt < maxcnt)
1322 		ldumpp[cnt].ndlck_clid.nclid_idlen = 0;
1323 	NFSUNLOCKSTATE();
1324 	NFSLOCKV4ROOTMUTEX();
1325 	nfsv4_relref(&nfsv4rootfs_lock);
1326 	NFSUNLOCKV4ROOTMUTEX();
1327 }
1328 
1329 /*
1330  * Server timer routine. It can scan any linked list, so long
1331  * as it holds the spin/mutex lock and there is no exclusive lock on
1332  * nfsv4rootfs_lock.
1333  * (For OpenBSD, a kthread is ok. For FreeBSD, I think it is ok
1334  *  to do this from a callout, since the spin locks work. For
1335  *  Darwin, I'm not sure what will work correctly yet.)
1336  * Should be called once per second.
1337  */
1338 void
1339 nfsrv_servertimer(void *arg __unused)
1340 {
1341 	struct nfsclient *clp, *nclp;
1342 	struct nfsstate *stp, *nstp;
1343 	int got_ref, i;
1344 
1345 	/*
1346 	 * Make sure nfsboottime is set. This is used by V3 as well
1347 	 * as V4. Note that nfsboottime is not nfsrvboottime, which is
1348 	 * only used by the V4 server for leases.
1349 	 */
1350 	if (nfsboottime.tv_sec == 0)
1351 		NFSSETBOOTTIME(nfsboottime);
1352 
1353 	/*
1354 	 * If server hasn't started yet, just return.
1355 	 */
1356 	NFSLOCKSTATE();
1357 	if (NFSD_VNET(nfsrv_stablefirst).nsf_eograce == 0) {
1358 		NFSUNLOCKSTATE();
1359 		return;
1360 	}
1361 	if (!(NFSD_VNET(nfsrv_stablefirst).nsf_flags & NFSNSF_UPDATEDONE)) {
1362 		if (!(NFSD_VNET(nfsrv_stablefirst).nsf_flags &
1363 		      NFSNSF_GRACEOVER) &&
1364 		    NFSD_MONOSEC > NFSD_VNET(nfsrv_stablefirst).nsf_eograce)
1365 			NFSD_VNET(nfsrv_stablefirst).nsf_flags |=
1366 			    (NFSNSF_GRACEOVER | NFSNSF_NEEDLOCK);
1367 		NFSUNLOCKSTATE();
1368 		return;
1369 	}
1370 
1371 	/*
1372 	 * Try and get a reference count on the nfsv4rootfs_lock so that
1373 	 * no nfsd thread can acquire an exclusive lock on it before this
1374 	 * call is done. If it is already exclusively locked, just return.
1375 	 */
1376 	NFSLOCKV4ROOTMUTEX();
1377 	got_ref = nfsv4_getref_nonblock(&nfsv4rootfs_lock);
1378 	NFSUNLOCKV4ROOTMUTEX();
1379 	if (got_ref == 0) {
1380 		NFSUNLOCKSTATE();
1381 		return;
1382 	}
1383 
1384 	/*
1385 	 * For each client...
1386 	 */
1387 	for (i = 0; i < nfsrv_clienthashsize; i++) {
1388 	    clp = LIST_FIRST(&NFSD_VNET(nfsclienthash)[i]);
1389 	    while (clp != LIST_END(&NFSD_VNET(nfsclienthash)[i])) {
1390 		nclp = LIST_NEXT(clp, lc_hash);
1391 		if (!(clp->lc_flags & LCL_EXPIREIT)) {
1392 		    if (((clp->lc_expiry + NFSRV_STALELEASE) < NFSD_MONOSEC
1393 			 && ((LIST_EMPTY(&clp->lc_deleg)
1394 			      && LIST_EMPTY(&clp->lc_open)) ||
1395 			     nfsrv_clients > nfsrv_clienthighwater)) ||
1396 			(clp->lc_expiry + NFSRV_MOULDYLEASE) < NFSD_MONOSEC ||
1397 			(clp->lc_expiry < NFSD_MONOSEC &&
1398 			 (nfsrv_openpluslock * 10 / 9) > nfsrv_v4statelimit)) {
1399 			/*
1400 			 * Lease has expired several nfsrv_lease times ago:
1401 			 * PLUS
1402 			 *    - no state is associated with it
1403 			 *    OR
1404 			 *    - above high water mark for number of clients
1405 			 *      (nfsrv_clienthighwater should be large enough
1406 			 *       that this only occurs when clients fail to
1407 			 *       use the same nfs_client_id4.id. Maybe somewhat
1408 			 *       higher that the maximum number of clients that
1409 			 *       will mount this server?)
1410 			 * OR
1411 			 * Lease has expired a very long time ago
1412 			 * OR
1413 			 * Lease has expired PLUS the number of opens + locks
1414 			 * has exceeded 90% of capacity
1415 			 *
1416 			 * --> Mark for expiry. The actual expiry will be done
1417 			 *     by an nfsd sometime soon.
1418 			 */
1419 			clp->lc_flags |= LCL_EXPIREIT;
1420 			NFSD_VNET(nfsrv_stablefirst).nsf_flags |=
1421 			    (NFSNSF_NEEDLOCK | NFSNSF_EXPIREDCLIENT);
1422 		    } else {
1423 			/*
1424 			 * If there are no opens, increment no open tick cnt
1425 			 * If time exceeds NFSNOOPEN, mark it to be thrown away
1426 			 * otherwise, if there is an open, reset no open time
1427 			 * Hopefully, this will avoid excessive re-creation
1428 			 * of open owners and subsequent open confirms.
1429 			 */
1430 			stp = LIST_FIRST(&clp->lc_open);
1431 			while (stp != LIST_END(&clp->lc_open)) {
1432 				nstp = LIST_NEXT(stp, ls_list);
1433 				if (LIST_EMPTY(&stp->ls_open)) {
1434 					stp->ls_noopens++;
1435 					if (stp->ls_noopens > NFSNOOPEN ||
1436 					    (nfsrv_openpluslock * 2) >
1437 					    nfsrv_v4statelimit)
1438 						NFSD_VNET(nfsrv_stablefirst).nsf_flags |=
1439 							NFSNSF_NOOPENS;
1440 				} else {
1441 					stp->ls_noopens = 0;
1442 				}
1443 				stp = nstp;
1444 			}
1445 		    }
1446 		}
1447 		clp = nclp;
1448 	    }
1449 	}
1450 	NFSUNLOCKSTATE();
1451 	NFSLOCKV4ROOTMUTEX();
1452 	nfsv4_relref(&nfsv4rootfs_lock);
1453 	NFSUNLOCKV4ROOTMUTEX();
1454 }
1455 
1456 /*
1457  * The following set of functions free up the various data structures.
1458  */
1459 /*
1460  * Clear out all open/lock state related to this nfsclient.
1461  * Caller must hold an exclusive lock on nfsv4rootfs_lock, so that
1462  * there are no other active nfsd threads.
1463  */
1464 void
1465 nfsrv_cleanclient(struct nfsclient *clp, NFSPROC_T *p, bool locked,
1466     SVCXPRT **old_xprtp)
1467 {
1468 	struct nfsstate *stp, *nstp;
1469 	struct nfsdsession *sep, *nsep;
1470 
1471 	LIST_FOREACH_SAFE(stp, &clp->lc_open, ls_list, nstp) {
1472 		if (locked)
1473 			nfsrv_freeopenowner(stp, 0, p);
1474 		else
1475 			nfsrv_freeopenowner(stp, 1, p);
1476 	}
1477 	if ((clp->lc_flags & LCL_ADMINREVOKED) == 0)
1478 		LIST_FOREACH_SAFE(sep, &clp->lc_session, sess_list, nsep)
1479 			(void)nfsrv_freesession(NULL, sep, NULL, locked,
1480 			    old_xprtp);
1481 }
1482 
1483 /*
1484  * Free a client that has been cleaned. It should also already have been
1485  * removed from the lists.
1486  * (Just to be safe w.r.t. newnfs_disconnect(), call this function when
1487  *  softclock interrupts are enabled.)
1488  */
1489 void
1490 nfsrv_zapclient(struct nfsclient *clp, NFSPROC_T *p)
1491 {
1492 
1493 #ifdef notyet
1494 	if ((clp->lc_flags & (LCL_GSS | LCL_CALLBACKSON)) ==
1495 	     (LCL_GSS | LCL_CALLBACKSON) &&
1496 	    (clp->lc_hand.nfsh_flag & NFSG_COMPLETE) &&
1497 	    clp->lc_handlelen > 0) {
1498 		clp->lc_hand.nfsh_flag &= ~NFSG_COMPLETE;
1499 		clp->lc_hand.nfsh_flag |= NFSG_DESTROYED;
1500 		(void) nfsrv_docallback(clp, NFSV4PROC_CBNULL,
1501 			NULL, 0, NULL, NULL, NULL, 0, p);
1502 	}
1503 #endif
1504 	newnfs_disconnect(NULL, &clp->lc_req);
1505 	free(clp->lc_req.nr_nam, M_SONAME);
1506 	NFSFREEMUTEX(&clp->lc_req.nr_mtx);
1507 	free(clp->lc_stateid, M_NFSDCLIENT);
1508 	free(clp, M_NFSDCLIENT);
1509 	NFSLOCKSTATE();
1510 	NFSD_VNET(nfsstatsv1_p)->srvclients--;
1511 	nfsrv_openpluslock--;
1512 	nfsrv_clients--;
1513 	NFSUNLOCKSTATE();
1514 }
1515 
1516 /*
1517  * Free a list of delegation state structures.
1518  * (This function will also free all nfslockfile structures that no
1519  *  longer have associated state.)
1520  */
1521 void
1522 nfsrv_freedeleglist(struct nfsstatehead *sthp)
1523 {
1524 	struct nfsstate *stp, *nstp;
1525 
1526 	LIST_FOREACH_SAFE(stp, sthp, ls_list, nstp) {
1527 		nfsrv_freedeleg(stp);
1528 	}
1529 	LIST_INIT(sthp);
1530 }
1531 
1532 /*
1533  * Free up a delegation.
1534  */
1535 static void
1536 nfsrv_freedeleg(struct nfsstate *stp)
1537 {
1538 	struct nfslockfile *lfp;
1539 
1540 	LIST_REMOVE(stp, ls_hash);
1541 	LIST_REMOVE(stp, ls_list);
1542 	LIST_REMOVE(stp, ls_file);
1543 	if ((stp->ls_flags & NFSLCK_DELEGWRITE) != 0)
1544 		nfsrv_writedelegcnt--;
1545 	lfp = stp->ls_lfp;
1546 	if (LIST_EMPTY(&lfp->lf_open) &&
1547 	    LIST_EMPTY(&lfp->lf_lock) && LIST_EMPTY(&lfp->lf_deleg) &&
1548 	    LIST_EMPTY(&lfp->lf_locallock) && LIST_EMPTY(&lfp->lf_rollback) &&
1549 	    lfp->lf_usecount == 0 &&
1550 	    nfsv4_testlock(&lfp->lf_locallock_lck) == 0)
1551 		nfsrv_freenfslockfile(lfp);
1552 	free(stp, M_NFSDSTATE);
1553 	NFSD_VNET(nfsstatsv1_p)->srvdelegates--;
1554 	nfsrv_openpluslock--;
1555 	nfsrv_delegatecnt--;
1556 }
1557 
1558 /*
1559  * This function frees an open owner and all associated opens.
1560  */
1561 static void
1562 nfsrv_freeopenowner(struct nfsstate *stp, int cansleep, NFSPROC_T *p)
1563 {
1564 	struct nfsstate *nstp, *tstp;
1565 
1566 	LIST_REMOVE(stp, ls_list);
1567 	/*
1568 	 * Now, free all associated opens.
1569 	 */
1570 	nstp = LIST_FIRST(&stp->ls_open);
1571 	while (nstp != LIST_END(&stp->ls_open)) {
1572 		tstp = nstp;
1573 		nstp = LIST_NEXT(nstp, ls_list);
1574 		nfsrv_freeopen(tstp, NULL, cansleep, p);
1575 	}
1576 	if (stp->ls_op)
1577 		nfsrvd_derefcache(stp->ls_op);
1578 	free(stp, M_NFSDSTATE);
1579 	NFSD_VNET(nfsstatsv1_p)->srvopenowners--;
1580 	nfsrv_openpluslock--;
1581 }
1582 
1583 /*
1584  * This function frees an open (nfsstate open structure) with all associated
1585  * lock_owners and locks. It also frees the nfslockfile structure iff there
1586  * are no other opens on the file.
1587  * Returns 1 if it free'd the nfslockfile, 0 otherwise.
1588  */
1589 static void
1590 nfsrv_freeopen(struct nfsstate *stp, vnode_t vp, int cansleep, NFSPROC_T *p)
1591 {
1592 	struct nfsstate *nstp, *tstp;
1593 	struct nfslockfile *lfp;
1594 
1595 	LIST_REMOVE(stp, ls_hash);
1596 	LIST_REMOVE(stp, ls_list);
1597 	LIST_REMOVE(stp, ls_file);
1598 
1599 	lfp = stp->ls_lfp;
1600 	/*
1601 	 * Now, free all lockowners associated with this open.
1602 	 * Note that, if vp != NULL, nfsrv_freelockowner() will
1603 	 * not call nfsrv_freeallnfslocks(), so it needs to be called, below.
1604 	 */
1605 	LIST_FOREACH_SAFE(tstp, &stp->ls_open, ls_list, nstp)
1606 		nfsrv_freelockowner(tstp, vp, cansleep, p);
1607 
1608 	if (vp != NULL) {
1609 		KASSERT(cansleep != 0, ("nfsrv_freeopen: cansleep == 0"));
1610 		mtx_assert(NFSSTATEMUTEXPTR, MA_OWNED);
1611 		/*
1612 		 * Only called with vp != NULL for Close when
1613 		 * vfs.nfsd.enable_locallocks != 0.
1614 		 * Lock the lfp so that it will not go away and do the
1615 		 * nfsrv_freeallnfslocks() call that was not done by
1616 		 * nfsrv_freelockowner().
1617 		 */
1618 		nfsrv_locklf(lfp);
1619 		NFSUNLOCKSTATE();
1620 		NFSVOPUNLOCK(vp);
1621 		nfsrv_freeallnfslocks(stp, vp, cansleep, p);
1622 		NFSVOPLOCK(vp, LK_EXCLUSIVE | LK_RETRY);
1623 		NFSLOCKSTATE();
1624 		nfsrv_unlocklf(lfp);
1625 	}
1626 
1627 	/*
1628 	 * The nfslockfile is freed here if there are no locks
1629 	 * associated with the open.
1630 	 * If there are locks associated with the open, the
1631 	 * nfslockfile structure can be freed via nfsrv_freelockowner().
1632 	 */
1633 	if (lfp != NULL && LIST_EMPTY(&lfp->lf_open) &&
1634 	    LIST_EMPTY(&lfp->lf_deleg) && LIST_EMPTY(&lfp->lf_lock) &&
1635 	    LIST_EMPTY(&lfp->lf_locallock) && LIST_EMPTY(&lfp->lf_rollback) &&
1636 	    lfp->lf_usecount == 0 &&
1637 	    nfsv4_testlock(&lfp->lf_locallock_lck) == 0)
1638 		nfsrv_freenfslockfile(lfp);
1639 	free(stp, M_NFSDSTATE);
1640 	NFSD_VNET(nfsstatsv1_p)->srvopens--;
1641 	nfsrv_openpluslock--;
1642 }
1643 
1644 /*
1645  * Frees a lockowner and all associated locks.
1646  */
1647 static void
1648 nfsrv_freelockowner(struct nfsstate *stp, vnode_t vp, int cansleep,
1649     NFSPROC_T *p)
1650 {
1651 
1652 	LIST_REMOVE(stp, ls_hash);
1653 	LIST_REMOVE(stp, ls_list);
1654 	if (vp == NULL)
1655 		nfsrv_freeallnfslocks(stp, vp, cansleep, p);
1656 	if (stp->ls_op)
1657 		nfsrvd_derefcache(stp->ls_op);
1658 	free(stp, M_NFSDSTATE);
1659 	NFSD_VNET(nfsstatsv1_p)->srvlockowners--;
1660 	nfsrv_openpluslock--;
1661 }
1662 
1663 /*
1664  * Free all the nfs locks on a lockowner.
1665  */
1666 static void
1667 nfsrv_freeallnfslocks(struct nfsstate *stp, vnode_t vp, int cansleep,
1668     NFSPROC_T *p)
1669 {
1670 	struct nfslock *lop, *nlop;
1671 	struct nfsrollback *rlp, *nrlp;
1672 	struct nfslockfile *lfp = NULL;
1673 	int gottvp = 0;
1674 	vnode_t tvp = NULL;
1675 	uint64_t first, end;
1676 
1677 	if (vp != NULL)
1678 		ASSERT_VOP_UNLOCKED(vp, "nfsrv_freeallnfslocks: vnode locked");
1679 	lop = LIST_FIRST(&stp->ls_lock);
1680 	while (lop != LIST_END(&stp->ls_lock)) {
1681 		nlop = LIST_NEXT(lop, lo_lckowner);
1682 		/*
1683 		 * Since all locks should be for the same file, lfp should
1684 		 * not change.
1685 		 */
1686 		if (lfp == NULL)
1687 			lfp = lop->lo_lfp;
1688 		else if (lfp != lop->lo_lfp)
1689 			panic("allnfslocks");
1690 		/*
1691 		 * If vp is NULL and cansleep != 0, a vnode must be acquired
1692 		 * from the file handle. This only occurs when called from
1693 		 * nfsrv_cleanclient().
1694 		 */
1695 		if (gottvp == 0) {
1696 			if (nfsrv_dolocallocks == 0)
1697 				tvp = NULL;
1698 			else if (vp == NULL && cansleep != 0) {
1699 				tvp = nfsvno_getvp(&lfp->lf_fh);
1700 				if (tvp != NULL)
1701 					NFSVOPUNLOCK(tvp);
1702 			} else
1703 				tvp = vp;
1704 			gottvp = 1;
1705 		}
1706 
1707 		if (tvp != NULL) {
1708 			if (cansleep == 0)
1709 				panic("allnfs2");
1710 			first = lop->lo_first;
1711 			end = lop->lo_end;
1712 			nfsrv_freenfslock(lop);
1713 			nfsrv_localunlock(tvp, lfp, first, end, p);
1714 			LIST_FOREACH_SAFE(rlp, &lfp->lf_rollback, rlck_list,
1715 			    nrlp)
1716 				free(rlp, M_NFSDROLLBACK);
1717 			LIST_INIT(&lfp->lf_rollback);
1718 		} else
1719 			nfsrv_freenfslock(lop);
1720 		lop = nlop;
1721 	}
1722 	if (vp == NULL && tvp != NULL)
1723 		vrele(tvp);
1724 }
1725 
1726 /*
1727  * Free an nfslock structure.
1728  */
1729 static void
1730 nfsrv_freenfslock(struct nfslock *lop)
1731 {
1732 
1733 	if (lop->lo_lckfile.le_prev != NULL) {
1734 		LIST_REMOVE(lop, lo_lckfile);
1735 		NFSD_VNET(nfsstatsv1_p)->srvlocks--;
1736 		nfsrv_openpluslock--;
1737 	}
1738 	LIST_REMOVE(lop, lo_lckowner);
1739 	free(lop, M_NFSDLOCK);
1740 }
1741 
1742 /*
1743  * This function frees an nfslockfile structure.
1744  */
1745 static void
1746 nfsrv_freenfslockfile(struct nfslockfile *lfp)
1747 {
1748 
1749 	LIST_REMOVE(lfp, lf_hash);
1750 	free(lfp, M_NFSDLOCKFILE);
1751 }
1752 
1753 /*
1754  * This function looks up an nfsstate structure via stateid.
1755  */
1756 static int
1757 nfsrv_getstate(struct nfsclient *clp, nfsv4stateid_t *stateidp, __unused u_int32_t flags,
1758     struct nfsstate **stpp)
1759 {
1760 	struct nfsstate *stp;
1761 	struct nfsstatehead *hp;
1762 	int error = 0;
1763 
1764 	*stpp = NULL;
1765 	hp = NFSSTATEHASH(clp, *stateidp);
1766 	LIST_FOREACH(stp, hp, ls_hash) {
1767 		if (!NFSBCMP(stp->ls_stateid.other, stateidp->other,
1768 			NFSX_STATEIDOTHER))
1769 			break;
1770 	}
1771 
1772 	/*
1773 	 * If no state id in list, return NFSERR_BADSTATEID.
1774 	 */
1775 	if (stp == LIST_END(hp)) {
1776 		error = NFSERR_BADSTATEID;
1777 		goto out;
1778 	}
1779 	*stpp = stp;
1780 
1781 out:
1782 	NFSEXITCODE(error);
1783 	return (error);
1784 }
1785 
1786 /*
1787  * This function gets an nfsstate structure via owner string.
1788  */
1789 static void
1790 nfsrv_getowner(struct nfsstatehead *hp, struct nfsstate *new_stp,
1791     struct nfsstate **stpp)
1792 {
1793 	struct nfsstate *stp;
1794 
1795 	*stpp = NULL;
1796 	LIST_FOREACH(stp, hp, ls_list) {
1797 		if (new_stp->ls_ownerlen == stp->ls_ownerlen &&
1798 		  !NFSBCMP(new_stp->ls_owner,stp->ls_owner,stp->ls_ownerlen)) {
1799 			*stpp = stp;
1800 			return;
1801 		}
1802 	}
1803 }
1804 
1805 /*
1806  * Lock control function called to update lock status.
1807  * Returns 0 upon success, -1 if there is no lock and the flags indicate
1808  * that one isn't to be created and an NFSERR_xxx for other errors.
1809  * The structures new_stp and new_lop are passed in as pointers that should
1810  * be set to NULL if the structure is used and shouldn't be free'd.
1811  * For the NFSLCK_TEST and NFSLCK_CHECK cases, the structures are
1812  * never used and can safely be allocated on the stack. For all other
1813  * cases, *new_stpp and *new_lopp should be malloc'd before the call,
1814  * in case they are used.
1815  */
1816 int
1817 nfsrv_lockctrl(vnode_t vp, struct nfsstate **new_stpp,
1818     struct nfslock **new_lopp, struct nfslockconflict *cfp,
1819     nfsquad_t clientid, nfsv4stateid_t *stateidp,
1820     __unused struct nfsexstuff *exp,
1821     struct nfsrv_descript *nd, NFSPROC_T *p)
1822 {
1823 	struct nfslock *lop;
1824 	struct nfsstate *new_stp = *new_stpp;
1825 	struct nfslock *new_lop = *new_lopp;
1826 	struct nfsstate *tstp, *mystp, *nstp;
1827 	int specialid = 0;
1828 	struct nfslockfile *lfp;
1829 	struct nfslock *other_lop = NULL;
1830 	struct nfsstate *stp, *lckstp = NULL;
1831 	struct nfsclient *clp = NULL;
1832 	u_int32_t bits;
1833 	int error = 0, haslock = 0, ret, reterr;
1834 	int getlckret, delegation = 0, filestruct_locked, vnode_unlocked = 0;
1835 	fhandle_t nfh;
1836 	uint64_t first, end;
1837 	uint32_t lock_flags;
1838 
1839 	if (new_stp->ls_flags & (NFSLCK_CHECK | NFSLCK_SETATTR)) {
1840 		/*
1841 		 * Note the special cases of "all 1s" or "all 0s" stateids and
1842 		 * let reads with all 1s go ahead.
1843 		 */
1844 		if (new_stp->ls_stateid.seqid == 0x0 &&
1845 		    new_stp->ls_stateid.other[0] == 0x0 &&
1846 		    new_stp->ls_stateid.other[1] == 0x0 &&
1847 		    new_stp->ls_stateid.other[2] == 0x0)
1848 			specialid = 1;
1849 		else if (new_stp->ls_stateid.seqid == 0xffffffff &&
1850 		    new_stp->ls_stateid.other[0] == 0xffffffff &&
1851 		    new_stp->ls_stateid.other[1] == 0xffffffff &&
1852 		    new_stp->ls_stateid.other[2] == 0xffffffff)
1853 			specialid = 2;
1854 	}
1855 
1856 	/*
1857 	 * Check for restart conditions (client and server).
1858 	 */
1859 	error = nfsrv_checkrestart(clientid, new_stp->ls_flags,
1860 	    &new_stp->ls_stateid, specialid);
1861 	if (error)
1862 		goto out;
1863 
1864 	/*
1865 	 * Check for state resource limit exceeded.
1866 	 */
1867 	if ((new_stp->ls_flags & NFSLCK_LOCK) &&
1868 	    nfsrv_openpluslock > nfsrv_v4statelimit) {
1869 		error = NFSERR_RESOURCE;
1870 		goto out;
1871 	}
1872 
1873 	/*
1874 	 * For the lock case, get another nfslock structure,
1875 	 * just in case we need it.
1876 	 * Malloc now, before we start sifting through the linked lists,
1877 	 * in case we have to wait for memory.
1878 	 */
1879 tryagain:
1880 	if (new_stp->ls_flags & NFSLCK_LOCK)
1881 		other_lop = malloc(sizeof (struct nfslock),
1882 		    M_NFSDLOCK, M_WAITOK);
1883 	filestruct_locked = 0;
1884 	reterr = 0;
1885 	lfp = NULL;
1886 
1887 	/*
1888 	 * Get the lockfile structure for CFH now, so we can do a sanity
1889 	 * check against the stateid, before incrementing the seqid#, since
1890 	 * we want to return NFSERR_BADSTATEID on failure and the seqid#
1891 	 * shouldn't be incremented for this case.
1892 	 * If nfsrv_getlockfile() returns -1, it means "not found", which
1893 	 * will be handled later.
1894 	 * If we are doing Lock/LockU and local locking is enabled, sleep
1895 	 * lock the nfslockfile structure.
1896 	 */
1897 	getlckret = nfsrv_getlockfh(vp, new_stp->ls_flags, NULL, &nfh, p);
1898 	NFSLOCKSTATE();
1899 	if (getlckret == 0) {
1900 		if ((new_stp->ls_flags & (NFSLCK_LOCK | NFSLCK_UNLOCK)) != 0 &&
1901 		    nfsrv_dolocallocks != 0 && nd->nd_repstat == 0) {
1902 			getlckret = nfsrv_getlockfile(new_stp->ls_flags, NULL,
1903 			    &lfp, &nfh, 1);
1904 			if (getlckret == 0)
1905 				filestruct_locked = 1;
1906 		} else
1907 			getlckret = nfsrv_getlockfile(new_stp->ls_flags, NULL,
1908 			    &lfp, &nfh, 0);
1909 	}
1910 	if (getlckret != 0 && getlckret != -1)
1911 		reterr = getlckret;
1912 
1913 	if (filestruct_locked != 0) {
1914 		LIST_INIT(&lfp->lf_rollback);
1915 		if ((new_stp->ls_flags & NFSLCK_LOCK)) {
1916 			/*
1917 			 * For local locking, do the advisory locking now, so
1918 			 * that any conflict can be detected. A failure later
1919 			 * can be rolled back locally. If an error is returned,
1920 			 * struct nfslockfile has been unlocked and any local
1921 			 * locking rolled back.
1922 			 */
1923 			NFSUNLOCKSTATE();
1924 			if (vnode_unlocked == 0) {
1925 				ASSERT_VOP_ELOCKED(vp, "nfsrv_lockctrl1");
1926 				vnode_unlocked = 1;
1927 				NFSVOPUNLOCK(vp);
1928 			}
1929 			reterr = nfsrv_locallock(vp, lfp,
1930 			    (new_lop->lo_flags & (NFSLCK_READ | NFSLCK_WRITE)),
1931 			    new_lop->lo_first, new_lop->lo_end, cfp, p);
1932 			NFSLOCKSTATE();
1933 		}
1934 	}
1935 
1936 	if (specialid == 0) {
1937 	    if (new_stp->ls_flags & NFSLCK_TEST) {
1938 		/*
1939 		 * RFC 3530 does not list LockT as an op that renews a
1940 		 * lease, but the consensus seems to be that it is ok
1941 		 * for a server to do so.
1942 		 */
1943 		error = nfsrv_getclient(clientid, CLOPS_RENEW, &clp, NULL,
1944 		    (nfsquad_t)((u_quad_t)0), 0, nd, p);
1945 
1946 		/*
1947 		 * Since NFSERR_EXPIRED, NFSERR_ADMINREVOKED are not valid
1948 		 * error returns for LockT, just go ahead and test for a lock,
1949 		 * since there are no locks for this client, but other locks
1950 		 * can conflict. (ie. same client will always be false)
1951 		 */
1952 		if (error == NFSERR_EXPIRED || error == NFSERR_ADMINREVOKED)
1953 		    error = 0;
1954 		lckstp = new_stp;
1955 	    } else {
1956 	      error = nfsrv_getclient(clientid, CLOPS_RENEW, &clp, NULL,
1957 		(nfsquad_t)((u_quad_t)0), 0, nd, p);
1958 	      if (error == 0)
1959 		/*
1960 		 * Look up the stateid
1961 		 */
1962 		error = nfsrv_getstate(clp, &new_stp->ls_stateid,
1963 		  new_stp->ls_flags, &stp);
1964 	      /*
1965 	       * do some sanity checks for an unconfirmed open or a
1966 	       * stateid that refers to the wrong file, for an open stateid
1967 	       */
1968 	      if (error == 0 && (stp->ls_flags & NFSLCK_OPEN) &&
1969 		  ((stp->ls_openowner->ls_flags & NFSLCK_NEEDSCONFIRM) ||
1970 		   (getlckret == 0 && stp->ls_lfp != lfp))){
1971 		      /*
1972 		       * NFSLCK_SETATTR should return OK rather than NFSERR_BADSTATEID
1973 		       * The only exception is using SETATTR with SIZE.
1974 		       * */
1975                     if ((new_stp->ls_flags &
1976                          (NFSLCK_SETATTR | NFSLCK_CHECK)) != NFSLCK_SETATTR)
1977 			     error = NFSERR_BADSTATEID;
1978 	      }
1979 
1980 		if (error == 0 &&
1981 		  (stp->ls_flags & (NFSLCK_DELEGREAD | NFSLCK_DELEGWRITE)) &&
1982 		  getlckret == 0 && stp->ls_lfp != lfp)
1983 			error = NFSERR_BADSTATEID;
1984 
1985 	      /*
1986 	       * If the lockowner stateid doesn't refer to the same file,
1987 	       * I believe that is considered ok, since some clients will
1988 	       * only create a single lockowner and use that for all locks
1989 	       * on all files.
1990 	       * For now, log it as a diagnostic, instead of considering it
1991 	       * a BadStateid.
1992 	       */
1993 	      if (error == 0 && (stp->ls_flags &
1994 		  (NFSLCK_OPEN | NFSLCK_DELEGREAD | NFSLCK_DELEGWRITE)) == 0 &&
1995 		  getlckret == 0 && stp->ls_lfp != lfp) {
1996 #ifdef DIAGNOSTIC
1997 		  printf("Got a lock statid for different file open\n");
1998 #endif
1999 		  /*
2000 		  error = NFSERR_BADSTATEID;
2001 		  */
2002 	      }
2003 
2004 	      if (error == 0) {
2005 		    if (new_stp->ls_flags & NFSLCK_OPENTOLOCK) {
2006 			/*
2007 			 * If haslock set, we've already checked the seqid.
2008 			 */
2009 			if (!haslock) {
2010 			    if (stp->ls_flags & NFSLCK_OPEN)
2011 				error = nfsrv_checkseqid(nd, new_stp->ls_seq,
2012 				    stp->ls_openowner, new_stp->ls_op);
2013 			    else
2014 				error = NFSERR_BADSTATEID;
2015 			}
2016 			if (!error)
2017 			    nfsrv_getowner(&stp->ls_open, new_stp, &lckstp);
2018 			if (lckstp) {
2019 			    /*
2020 			     * For NFSv4.1 and NFSv4.2 allow an
2021 			     * open_to_lock_owner when the lock_owner already
2022 			     * exists.  Just clear NFSLCK_OPENTOLOCK so that
2023 			     * a new lock_owner will not be created.
2024 			     * RFC7530 states that the error for NFSv4.0
2025 			     * is NFS4ERR_BAD_SEQID.
2026 			     */
2027 			    if ((nd->nd_flag & ND_NFSV41) != 0)
2028 				new_stp->ls_flags &= ~NFSLCK_OPENTOLOCK;
2029 			    else
2030 				error = NFSERR_BADSEQID;
2031 			} else
2032 			    lckstp = new_stp;
2033 		    } else if (new_stp->ls_flags&(NFSLCK_LOCK|NFSLCK_UNLOCK)) {
2034 			/*
2035 			 * If haslock set, ditto above.
2036 			 */
2037 			if (!haslock) {
2038 			    if (stp->ls_flags & NFSLCK_OPEN)
2039 				error = NFSERR_BADSTATEID;
2040 			    else
2041 				error = nfsrv_checkseqid(nd, new_stp->ls_seq,
2042 				    stp, new_stp->ls_op);
2043 			}
2044 			lckstp = stp;
2045 		    } else {
2046 			lckstp = stp;
2047 		    }
2048 	      }
2049 	      /*
2050 	       * If the seqid part of the stateid isn't the same, return
2051 	       * NFSERR_OLDSTATEID for cases other than I/O Ops.
2052 	       * For I/O Ops, only return NFSERR_OLDSTATEID if
2053 	       * nfsrv_returnoldstateid is set. (The consensus on the email
2054 	       * list was that most clients would prefer to not receive
2055 	       * NFSERR_OLDSTATEID for I/O Ops, but the RFC suggests that that
2056 	       * is what will happen, so I use the nfsrv_returnoldstateid to
2057 	       * allow for either server configuration.)
2058 	       */
2059 	      if (!error && stp->ls_stateid.seqid!=new_stp->ls_stateid.seqid &&
2060 		  (((nd->nd_flag & ND_NFSV41) == 0 &&
2061 		   (!(new_stp->ls_flags & NFSLCK_CHECK) ||
2062 		    nfsrv_returnoldstateid)) ||
2063 		   ((nd->nd_flag & ND_NFSV41) != 0 &&
2064 		    new_stp->ls_stateid.seqid != 0)))
2065 		    error = NFSERR_OLDSTATEID;
2066 	    }
2067 	}
2068 
2069 	/*
2070 	 * Now we can check for grace.
2071 	 */
2072 	if (!error)
2073 		error = nfsrv_checkgrace(nd, clp, new_stp->ls_flags);
2074 	if ((new_stp->ls_flags & NFSLCK_RECLAIM) && !error &&
2075 		nfsrv_checkstable(clp))
2076 		error = NFSERR_NOGRACE;
2077 	/*
2078 	 * If we successfully Reclaimed state, note that.
2079 	 */
2080 	if ((new_stp->ls_flags & NFSLCK_RECLAIM) && !error)
2081 		nfsrv_markstable(clp);
2082 
2083 	/*
2084 	 * At this point, either error == NFSERR_BADSTATEID or the
2085 	 * seqid# has been updated, so we can return any error.
2086 	 * If error == 0, there may be an error in:
2087 	 *    nd_repstat - Set by the calling function.
2088 	 *    reterr - Set above, if getting the nfslockfile structure
2089 	 *       or acquiring the local lock failed.
2090 	 *    (If both of these are set, nd_repstat should probably be
2091 	 *     returned, since that error was detected before this
2092 	 *     function call.)
2093 	 */
2094 	if (error != 0 || nd->nd_repstat != 0 || reterr != 0) {
2095 		if (error == 0) {
2096 			if (nd->nd_repstat != 0)
2097 				error = nd->nd_repstat;
2098 			else
2099 				error = reterr;
2100 		}
2101 		if (filestruct_locked != 0) {
2102 			/* Roll back local locks. */
2103 			NFSUNLOCKSTATE();
2104 			if (vnode_unlocked == 0) {
2105 				ASSERT_VOP_ELOCKED(vp, "nfsrv_lockctrl2");
2106 				vnode_unlocked = 1;
2107 				NFSVOPUNLOCK(vp);
2108 			}
2109 			nfsrv_locallock_rollback(vp, lfp, p);
2110 			NFSLOCKSTATE();
2111 			nfsrv_unlocklf(lfp);
2112 		}
2113 		NFSUNLOCKSTATE();
2114 		goto out;
2115 	}
2116 
2117 	/*
2118 	 * Check the nfsrv_getlockfile return.
2119 	 * Returned -1 if no structure found.
2120 	 */
2121 	if (getlckret == -1) {
2122 		error = NFSERR_EXPIRED;
2123 		/*
2124 		 * Called from lockt, so no lock is OK.
2125 		 */
2126 		if (new_stp->ls_flags & NFSLCK_TEST) {
2127 			error = 0;
2128 		} else if (new_stp->ls_flags &
2129 		    (NFSLCK_CHECK | NFSLCK_SETATTR)) {
2130 			/*
2131 			 * Called to check for a lock, OK if the stateid is all
2132 			 * 1s or all 0s, but there should be an nfsstate
2133 			 * otherwise.
2134 			 * (ie. If there is no open, I'll assume no share
2135 			 *  deny bits.)
2136 			 */
2137 			if (specialid)
2138 				error = 0;
2139 			else
2140 				error = NFSERR_BADSTATEID;
2141 		}
2142 		NFSUNLOCKSTATE();
2143 		goto out;
2144 	}
2145 
2146 	/*
2147 	 * For NFSLCK_CHECK and NFSLCK_LOCK, test for a share conflict.
2148 	 * For NFSLCK_CHECK, allow a read if write access is granted,
2149 	 * but check for a deny. For NFSLCK_LOCK, require correct access,
2150 	 * which implies a conflicting deny can't exist.
2151 	 */
2152 	if (new_stp->ls_flags & (NFSLCK_CHECK | NFSLCK_LOCK)) {
2153 	    /*
2154 	     * Four kinds of state id:
2155 	     * - specialid (all 0s or all 1s), only for NFSLCK_CHECK
2156 	     * - stateid for an open
2157 	     * - stateid for a delegation
2158 	     * - stateid for a lock owner
2159 	     */
2160 	    if (!specialid) {
2161 		if (stp->ls_flags & (NFSLCK_DELEGREAD | NFSLCK_DELEGWRITE)) {
2162 		    delegation = 1;
2163 		    mystp = stp;
2164 		    nfsrv_delaydelegtimeout(stp);
2165 	        } else if (stp->ls_flags & NFSLCK_OPEN) {
2166 		    mystp = stp;
2167 		} else {
2168 		    mystp = stp->ls_openstp;
2169 		}
2170 		/*
2171 		 * If locking or checking, require correct access
2172 		 * bit set.
2173 		 */
2174 		if (((new_stp->ls_flags & NFSLCK_LOCK) &&
2175 		     !((new_lop->lo_flags >> NFSLCK_LOCKSHIFT) &
2176 		       mystp->ls_flags & NFSLCK_ACCESSBITS)) ||
2177 		    ((new_stp->ls_flags & (NFSLCK_CHECK|NFSLCK_READACCESS)) ==
2178 		      (NFSLCK_CHECK | NFSLCK_READACCESS) &&
2179 		     !(mystp->ls_flags & NFSLCK_READACCESS) &&
2180 		     nfsrv_allowreadforwriteopen == 0) ||
2181 		    ((new_stp->ls_flags & (NFSLCK_CHECK|NFSLCK_WRITEACCESS)) ==
2182 		      (NFSLCK_CHECK | NFSLCK_WRITEACCESS) &&
2183 		     !(mystp->ls_flags & NFSLCK_WRITEACCESS))) {
2184 			if (filestruct_locked != 0) {
2185 				/* Roll back local locks. */
2186 				NFSUNLOCKSTATE();
2187 				if (vnode_unlocked == 0) {
2188 					ASSERT_VOP_ELOCKED(vp,
2189 					    "nfsrv_lockctrl3");
2190 					vnode_unlocked = 1;
2191 					NFSVOPUNLOCK(vp);
2192 				}
2193 				nfsrv_locallock_rollback(vp, lfp, p);
2194 				NFSLOCKSTATE();
2195 				nfsrv_unlocklf(lfp);
2196 			}
2197 			NFSUNLOCKSTATE();
2198 			error = NFSERR_OPENMODE;
2199 			goto out;
2200 		}
2201 	    } else
2202 		mystp = NULL;
2203 	    if ((new_stp->ls_flags & NFSLCK_CHECK) && !delegation) {
2204 		/*
2205 		 * Check for a conflicting deny bit.
2206 		 */
2207 		LIST_FOREACH(tstp, &lfp->lf_open, ls_file) {
2208 		    if (tstp != mystp) {
2209 			bits = tstp->ls_flags;
2210 			bits >>= NFSLCK_SHIFT;
2211 			if (new_stp->ls_flags & bits & NFSLCK_ACCESSBITS) {
2212 			    KASSERT(vnode_unlocked == 0,
2213 				("nfsrv_lockctrl: vnode unlocked1"));
2214 			    ret = nfsrv_clientconflict(tstp->ls_clp, &haslock,
2215 				vp, p);
2216 			    if (ret == 1) {
2217 				/*
2218 				* nfsrv_clientconflict unlocks state
2219 				 * when it returns non-zero.
2220 				 */
2221 				lckstp = NULL;
2222 				goto tryagain;
2223 			    }
2224 			    if (ret == 0)
2225 				NFSUNLOCKSTATE();
2226 			    if (ret == 2)
2227 				error = NFSERR_PERM;
2228 			    else
2229 				error = NFSERR_OPENMODE;
2230 			    goto out;
2231 			}
2232 		    }
2233 		}
2234 
2235 		/* We're outta here */
2236 		NFSUNLOCKSTATE();
2237 		goto out;
2238 	    }
2239 	}
2240 
2241 	/*
2242 	 * For setattr, just get rid of all the Delegations for other clients.
2243 	 */
2244 	if (new_stp->ls_flags & NFSLCK_SETATTR) {
2245 		KASSERT(vnode_unlocked == 0,
2246 		    ("nfsrv_lockctrl: vnode unlocked2"));
2247 		ret = nfsrv_cleandeleg(vp, lfp, clp, &haslock, p);
2248 		if (ret) {
2249 			/*
2250 			 * nfsrv_cleandeleg() unlocks state when it
2251 			 * returns non-zero.
2252 			 */
2253 			if (ret == -1) {
2254 				lckstp = NULL;
2255 				goto tryagain;
2256 			}
2257 			error = ret;
2258 			goto out;
2259 		}
2260 		if (!(new_stp->ls_flags & NFSLCK_CHECK) ||
2261 		    (LIST_EMPTY(&lfp->lf_open) && LIST_EMPTY(&lfp->lf_lock) &&
2262 		     LIST_EMPTY(&lfp->lf_deleg))) {
2263 			NFSUNLOCKSTATE();
2264 			goto out;
2265 		}
2266 	}
2267 
2268 	/*
2269 	 * Check for a conflicting delegation. If one is found, call
2270 	 * nfsrv_delegconflict() to handle it. If the v4root lock hasn't
2271 	 * been set yet, it will get the lock. Otherwise, it will recall
2272 	 * the delegation. Then, we try try again...
2273 	 * I currently believe the conflict algorithm to be:
2274 	 * For Lock Ops (Lock/LockT/LockU)
2275 	 * - there is a conflict iff a different client has a write delegation
2276 	 * For Reading (Read Op)
2277 	 * - there is a conflict iff a different client has a write delegation
2278 	 *   (the specialids are always a different client)
2279 	 * For Writing (Write/Setattr of size)
2280 	 * - there is a conflict if a different client has any delegation
2281 	 * - there is a conflict if the same client has a read delegation
2282 	 *   (I don't understand why this isn't allowed, but that seems to be
2283 	 *    the current consensus?)
2284 	 */
2285 	tstp = LIST_FIRST(&lfp->lf_deleg);
2286 	while (tstp != LIST_END(&lfp->lf_deleg)) {
2287 	    nstp = LIST_NEXT(tstp, ls_file);
2288 	    if ((((new_stp->ls_flags&(NFSLCK_LOCK|NFSLCK_UNLOCK|NFSLCK_TEST))||
2289 		 ((new_stp->ls_flags & NFSLCK_CHECK) &&
2290 		  (new_lop->lo_flags & NFSLCK_READ))) &&
2291 		  clp != tstp->ls_clp &&
2292 		 (tstp->ls_flags & NFSLCK_DELEGWRITE)) ||
2293 		 ((new_stp->ls_flags & NFSLCK_CHECK) &&
2294 		   (new_lop->lo_flags & NFSLCK_WRITE) &&
2295 		  (clp != tstp->ls_clp ||
2296 		   (tstp->ls_flags & NFSLCK_DELEGREAD)))) {
2297 		ret = 0;
2298 		if (filestruct_locked != 0) {
2299 			/* Roll back local locks. */
2300 			NFSUNLOCKSTATE();
2301 			if (vnode_unlocked == 0) {
2302 				ASSERT_VOP_ELOCKED(vp, "nfsrv_lockctrl4");
2303 				NFSVOPUNLOCK(vp);
2304 			}
2305 			nfsrv_locallock_rollback(vp, lfp, p);
2306 			NFSLOCKSTATE();
2307 			nfsrv_unlocklf(lfp);
2308 			NFSUNLOCKSTATE();
2309 			NFSVOPLOCK(vp, LK_EXCLUSIVE | LK_RETRY);
2310 			vnode_unlocked = 0;
2311 			if (VN_IS_DOOMED(vp))
2312 				ret = NFSERR_SERVERFAULT;
2313 			NFSLOCKSTATE();
2314 		}
2315 		if (ret == 0)
2316 			ret = nfsrv_delegconflict(tstp, &haslock, p, vp);
2317 		if (ret) {
2318 		    /*
2319 		     * nfsrv_delegconflict unlocks state when it
2320 		     * returns non-zero, which it always does.
2321 		     */
2322 		    if (other_lop) {
2323 			free(other_lop, M_NFSDLOCK);
2324 			other_lop = NULL;
2325 		    }
2326 		    if (ret == -1) {
2327 			lckstp = NULL;
2328 			goto tryagain;
2329 		    }
2330 		    error = ret;
2331 		    goto out;
2332 		}
2333 		/* Never gets here. */
2334 	    }
2335 	    tstp = nstp;
2336 	}
2337 
2338 	/*
2339 	 * Handle the unlock case by calling nfsrv_updatelock().
2340 	 * (Should I have done some access checking above for unlock? For now,
2341 	 *  just let it happen.)
2342 	 */
2343 	if (new_stp->ls_flags & NFSLCK_UNLOCK) {
2344 		first = new_lop->lo_first;
2345 		end = new_lop->lo_end;
2346 		nfsrv_updatelock(stp, new_lopp, &other_lop, lfp);
2347 		stateidp->seqid = ++(stp->ls_stateid.seqid);
2348 		if ((nd->nd_flag & ND_NFSV41) != 0 && stateidp->seqid == 0)
2349 			stateidp->seqid = stp->ls_stateid.seqid = 1;
2350 		stateidp->other[0] = stp->ls_stateid.other[0];
2351 		stateidp->other[1] = stp->ls_stateid.other[1];
2352 		stateidp->other[2] = stp->ls_stateid.other[2];
2353 		if (filestruct_locked != 0) {
2354 			NFSUNLOCKSTATE();
2355 			if (vnode_unlocked == 0) {
2356 				ASSERT_VOP_ELOCKED(vp, "nfsrv_lockctrl5");
2357 				vnode_unlocked = 1;
2358 				NFSVOPUNLOCK(vp);
2359 			}
2360 			/* Update the local locks. */
2361 			nfsrv_localunlock(vp, lfp, first, end, p);
2362 			NFSLOCKSTATE();
2363 			nfsrv_unlocklf(lfp);
2364 		}
2365 		NFSUNLOCKSTATE();
2366 		goto out;
2367 	}
2368 
2369 	/*
2370 	 * Search for a conflicting lock. A lock conflicts if:
2371 	 * - the lock range overlaps and
2372 	 * - at least one lock is a write lock and
2373 	 * - it is not owned by the same lock owner
2374 	 */
2375 	if (!delegation) {
2376 	  LIST_FOREACH(lop, &lfp->lf_lock, lo_lckfile) {
2377 	    if (new_lop->lo_end > lop->lo_first &&
2378 		new_lop->lo_first < lop->lo_end &&
2379 		(new_lop->lo_flags == NFSLCK_WRITE ||
2380 		 lop->lo_flags == NFSLCK_WRITE) &&
2381 		lckstp != lop->lo_stp &&
2382 		(clp != lop->lo_stp->ls_clp ||
2383 		 lckstp->ls_ownerlen != lop->lo_stp->ls_ownerlen ||
2384 		 NFSBCMP(lckstp->ls_owner, lop->lo_stp->ls_owner,
2385 		    lckstp->ls_ownerlen))) {
2386 		if (other_lop) {
2387 		    free(other_lop, M_NFSDLOCK);
2388 		    other_lop = NULL;
2389 		}
2390 		if (vnode_unlocked != 0)
2391 		    ret = nfsrv_clientconflict(lop->lo_stp->ls_clp, &haslock,
2392 			NULL, p);
2393 		else
2394 		    ret = nfsrv_clientconflict(lop->lo_stp->ls_clp, &haslock,
2395 			vp, p);
2396 		if (ret == 1) {
2397 		    if (filestruct_locked != 0) {
2398 			if (vnode_unlocked == 0) {
2399 				ASSERT_VOP_ELOCKED(vp, "nfsrv_lockctrl6");
2400 				NFSVOPUNLOCK(vp);
2401 			}
2402 			/* Roll back local locks. */
2403 			nfsrv_locallock_rollback(vp, lfp, p);
2404 			NFSLOCKSTATE();
2405 			nfsrv_unlocklf(lfp);
2406 			NFSUNLOCKSTATE();
2407 			NFSVOPLOCK(vp, LK_EXCLUSIVE | LK_RETRY);
2408 			vnode_unlocked = 0;
2409 			if (VN_IS_DOOMED(vp)) {
2410 				error = NFSERR_SERVERFAULT;
2411 				goto out;
2412 			}
2413 		    }
2414 		    /*
2415 		     * nfsrv_clientconflict() unlocks state when it
2416 		     * returns non-zero.
2417 		     */
2418 		    lckstp = NULL;
2419 		    goto tryagain;
2420 		}
2421 		/*
2422 		 * Found a conflicting lock, so record the conflict and
2423 		 * return the error.
2424 		 */
2425 		if (cfp != NULL && ret == 0) {
2426 		    cfp->cl_clientid.lval[0]=lop->lo_stp->ls_stateid.other[0];
2427 		    cfp->cl_clientid.lval[1]=lop->lo_stp->ls_stateid.other[1];
2428 		    cfp->cl_first = lop->lo_first;
2429 		    cfp->cl_end = lop->lo_end;
2430 		    cfp->cl_flags = lop->lo_flags;
2431 		    cfp->cl_ownerlen = lop->lo_stp->ls_ownerlen;
2432 		    NFSBCOPY(lop->lo_stp->ls_owner, cfp->cl_owner,
2433 			cfp->cl_ownerlen);
2434 		}
2435 		if (ret == 2)
2436 		    error = NFSERR_PERM;
2437 		else if (new_stp->ls_flags & NFSLCK_RECLAIM)
2438 		    error = NFSERR_RECLAIMCONFLICT;
2439 		else if (new_stp->ls_flags & NFSLCK_CHECK)
2440 		    error = NFSERR_LOCKED;
2441 		else
2442 		    error = NFSERR_DENIED;
2443 		if (filestruct_locked != 0 && ret == 0) {
2444 			/* Roll back local locks. */
2445 			NFSUNLOCKSTATE();
2446 			if (vnode_unlocked == 0) {
2447 				ASSERT_VOP_ELOCKED(vp, "nfsrv_lockctrl7");
2448 				vnode_unlocked = 1;
2449 				NFSVOPUNLOCK(vp);
2450 			}
2451 			nfsrv_locallock_rollback(vp, lfp, p);
2452 			NFSLOCKSTATE();
2453 			nfsrv_unlocklf(lfp);
2454 		}
2455 		if (ret == 0)
2456 			NFSUNLOCKSTATE();
2457 		goto out;
2458 	    }
2459 	  }
2460 	}
2461 
2462 	/*
2463 	 * We only get here if there was no lock that conflicted.
2464 	 */
2465 	if (new_stp->ls_flags & (NFSLCK_TEST | NFSLCK_CHECK)) {
2466 		NFSUNLOCKSTATE();
2467 		goto out;
2468 	}
2469 
2470 	/*
2471 	 * We only get here when we are creating or modifying a lock.
2472 	 * There are two variants:
2473 	 * - exist_lock_owner where lock_owner exists
2474 	 * - open_to_lock_owner with new lock_owner
2475 	 */
2476 	first = new_lop->lo_first;
2477 	end = new_lop->lo_end;
2478 	lock_flags = new_lop->lo_flags;
2479 	if (!(new_stp->ls_flags & NFSLCK_OPENTOLOCK)) {
2480 		nfsrv_updatelock(lckstp, new_lopp, &other_lop, lfp);
2481 		stateidp->seqid = ++(lckstp->ls_stateid.seqid);
2482 		if ((nd->nd_flag & ND_NFSV41) != 0 && stateidp->seqid == 0)
2483 			stateidp->seqid = lckstp->ls_stateid.seqid = 1;
2484 		stateidp->other[0] = lckstp->ls_stateid.other[0];
2485 		stateidp->other[1] = lckstp->ls_stateid.other[1];
2486 		stateidp->other[2] = lckstp->ls_stateid.other[2];
2487 	} else {
2488 		/*
2489 		 * The new open_to_lock_owner case.
2490 		 * Link the new nfsstate into the lists.
2491 		 */
2492 		new_stp->ls_seq = new_stp->ls_opentolockseq;
2493 		nfsrvd_refcache(new_stp->ls_op);
2494 		stateidp->seqid = new_stp->ls_stateid.seqid = 1;
2495 		stateidp->other[0] = new_stp->ls_stateid.other[0] =
2496 		    clp->lc_clientid.lval[0];
2497 		stateidp->other[1] = new_stp->ls_stateid.other[1] =
2498 		    clp->lc_clientid.lval[1];
2499 		stateidp->other[2] = new_stp->ls_stateid.other[2] =
2500 		    nfsrv_nextstateindex(clp);
2501 		new_stp->ls_clp = clp;
2502 		LIST_INIT(&new_stp->ls_lock);
2503 		new_stp->ls_openstp = stp;
2504 		new_stp->ls_lfp = lfp;
2505 		nfsrv_insertlock(new_lop, (struct nfslock *)new_stp, new_stp,
2506 		    lfp);
2507 		LIST_INSERT_HEAD(NFSSTATEHASH(clp, new_stp->ls_stateid),
2508 		    new_stp, ls_hash);
2509 		LIST_INSERT_HEAD(&stp->ls_open, new_stp, ls_list);
2510 		*new_lopp = NULL;
2511 		*new_stpp = NULL;
2512 		NFSD_VNET(nfsstatsv1_p)->srvlockowners++;
2513 		nfsrv_openpluslock++;
2514 	}
2515 	if (filestruct_locked != 0) {
2516 		NFSUNLOCKSTATE();
2517 		nfsrv_locallock_commit(lfp, lock_flags, first, end);
2518 		NFSLOCKSTATE();
2519 		nfsrv_unlocklf(lfp);
2520 	}
2521 	NFSUNLOCKSTATE();
2522 
2523 out:
2524 	if (haslock) {
2525 		NFSLOCKV4ROOTMUTEX();
2526 		nfsv4_unlock(&nfsv4rootfs_lock, 1);
2527 		NFSUNLOCKV4ROOTMUTEX();
2528 	}
2529 	if (vnode_unlocked != 0) {
2530 		NFSVOPLOCK(vp, LK_EXCLUSIVE | LK_RETRY);
2531 		if (error == 0 && VN_IS_DOOMED(vp))
2532 			error = NFSERR_SERVERFAULT;
2533 	}
2534 	if (other_lop)
2535 		free(other_lop, M_NFSDLOCK);
2536 	NFSEXITCODE2(error, nd);
2537 	return (error);
2538 }
2539 
2540 /*
2541  * Check for state errors for Open.
2542  * repstat is passed back out as an error if more critical errors
2543  * are not detected.
2544  */
2545 int
2546 nfsrv_opencheck(nfsquad_t clientid, nfsv4stateid_t *stateidp,
2547     struct nfsstate *new_stp, vnode_t vp, struct nfsrv_descript *nd,
2548     NFSPROC_T *p, int repstat)
2549 {
2550 	struct nfsstate *stp, *nstp;
2551 	struct nfsclient *clp;
2552 	struct nfsstate *ownerstp;
2553 	struct nfslockfile *lfp, *new_lfp;
2554 	int error = 0, haslock = 0, ret, readonly = 0, getfhret = 0;
2555 
2556 	if ((new_stp->ls_flags & NFSLCK_SHAREBITS) == NFSLCK_READACCESS)
2557 		readonly = 1;
2558 	/*
2559 	 * Check for restart conditions (client and server).
2560 	 */
2561 	error = nfsrv_checkrestart(clientid, new_stp->ls_flags,
2562 		&new_stp->ls_stateid, 0);
2563 	if (error)
2564 		goto out;
2565 
2566 	/*
2567 	 * Check for state resource limit exceeded.
2568 	 * Technically this should be SMP protected, but the worst
2569 	 * case error is "out by one or two" on the count when it
2570 	 * returns NFSERR_RESOURCE and the limit is just a rather
2571 	 * arbitrary high water mark, so no harm is done.
2572 	 */
2573 	if (nfsrv_openpluslock > nfsrv_v4statelimit) {
2574 		error = NFSERR_RESOURCE;
2575 		goto out;
2576 	}
2577 
2578 tryagain:
2579 	new_lfp = malloc(sizeof (struct nfslockfile),
2580 	    M_NFSDLOCKFILE, M_WAITOK);
2581 	if (vp)
2582 		getfhret = nfsrv_getlockfh(vp, new_stp->ls_flags, new_lfp,
2583 		    NULL, p);
2584 	NFSLOCKSTATE();
2585 	/*
2586 	 * Get the nfsclient structure.
2587 	 */
2588 	error = nfsrv_getclient(clientid, CLOPS_RENEW, &clp, NULL,
2589 	    (nfsquad_t)((u_quad_t)0), 0, nd, p);
2590 
2591 	/*
2592 	 * Look up the open owner. See if it needs confirmation and
2593 	 * check the seq#, as required.
2594 	 */
2595 	if (!error)
2596 		nfsrv_getowner(&clp->lc_open, new_stp, &ownerstp);
2597 
2598 	if (!error && ownerstp) {
2599 		error = nfsrv_checkseqid(nd, new_stp->ls_seq, ownerstp,
2600 		    new_stp->ls_op);
2601 		/*
2602 		 * If the OpenOwner hasn't been confirmed, assume the
2603 		 * old one was a replay and this one is ok.
2604 		 * See: RFC3530 Sec. 14.2.18.
2605 		 */
2606 		if (error == NFSERR_BADSEQID &&
2607 		    (ownerstp->ls_flags & NFSLCK_NEEDSCONFIRM))
2608 			error = 0;
2609 	}
2610 
2611 	/*
2612 	 * Check for grace.
2613 	 */
2614 	if (!error)
2615 		error = nfsrv_checkgrace(nd, clp, new_stp->ls_flags);
2616 	if ((new_stp->ls_flags & NFSLCK_RECLAIM) && !error &&
2617 		nfsrv_checkstable(clp))
2618 		error = NFSERR_NOGRACE;
2619 
2620 	/*
2621 	 * If none of the above errors occurred, let repstat be
2622 	 * returned.
2623 	 */
2624 	if (repstat && !error)
2625 		error = repstat;
2626 	if (error) {
2627 		NFSUNLOCKSTATE();
2628 		if (haslock) {
2629 			NFSLOCKV4ROOTMUTEX();
2630 			nfsv4_unlock(&nfsv4rootfs_lock, 1);
2631 			NFSUNLOCKV4ROOTMUTEX();
2632 		}
2633 		free(new_lfp, M_NFSDLOCKFILE);
2634 		goto out;
2635 	}
2636 
2637 	/*
2638 	 * If vp == NULL, the file doesn't exist yet, so return ok.
2639 	 * (This always happens on the first pass, so haslock must be 0.)
2640 	 */
2641 	if (vp == NULL) {
2642 		NFSUNLOCKSTATE();
2643 		free(new_lfp, M_NFSDLOCKFILE);
2644 		goto out;
2645 	}
2646 
2647 	/*
2648 	 * Get the structure for the underlying file.
2649 	 */
2650 	if (getfhret)
2651 		error = getfhret;
2652 	else
2653 		error = nfsrv_getlockfile(new_stp->ls_flags, &new_lfp, &lfp,
2654 		    NULL, 0);
2655 	if (new_lfp)
2656 		free(new_lfp, M_NFSDLOCKFILE);
2657 	if (error) {
2658 		NFSUNLOCKSTATE();
2659 		if (haslock) {
2660 			NFSLOCKV4ROOTMUTEX();
2661 			nfsv4_unlock(&nfsv4rootfs_lock, 1);
2662 			NFSUNLOCKV4ROOTMUTEX();
2663 		}
2664 		goto out;
2665 	}
2666 
2667 	/*
2668 	 * Search for a conflicting open/share.
2669 	 */
2670 	if (new_stp->ls_flags & NFSLCK_DELEGCUR) {
2671 	    /*
2672 	     * For Delegate_Cur, search for the matching Delegation,
2673 	     * which indicates no conflict.
2674 	     * An old delegation should have been recovered by the
2675 	     * client doing a Claim_DELEGATE_Prev, so I won't let
2676 	     * it match and return NFSERR_EXPIRED. Should I let it
2677 	     * match?
2678 	     */
2679 	    LIST_FOREACH(stp, &lfp->lf_deleg, ls_file) {
2680 		if (!(stp->ls_flags & NFSLCK_OLDDELEG) &&
2681 		    (((nd->nd_flag & ND_NFSV41) != 0 &&
2682 		    stateidp->seqid == 0) ||
2683 		    stateidp->seqid == stp->ls_stateid.seqid) &&
2684 		    !NFSBCMP(stateidp->other, stp->ls_stateid.other,
2685 			  NFSX_STATEIDOTHER))
2686 			break;
2687 	    }
2688 	    if (stp == LIST_END(&lfp->lf_deleg) ||
2689 		((new_stp->ls_flags & NFSLCK_WRITEACCESS) &&
2690 		 (stp->ls_flags & NFSLCK_DELEGREAD))) {
2691 		NFSUNLOCKSTATE();
2692 		if (haslock) {
2693 			NFSLOCKV4ROOTMUTEX();
2694 			nfsv4_unlock(&nfsv4rootfs_lock, 1);
2695 			NFSUNLOCKV4ROOTMUTEX();
2696 		}
2697 		error = NFSERR_EXPIRED;
2698 		goto out;
2699 	    }
2700 	}
2701 
2702 	/*
2703 	 * Check for access/deny bit conflicts. I check for the same
2704 	 * owner as well, in case the client didn't bother.
2705 	 */
2706 	LIST_FOREACH(stp, &lfp->lf_open, ls_file) {
2707 		if (!(new_stp->ls_flags & NFSLCK_DELEGCUR) &&
2708 		    (((new_stp->ls_flags & NFSLCK_ACCESSBITS) &
2709 		      ((stp->ls_flags>>NFSLCK_SHIFT) & NFSLCK_ACCESSBITS))||
2710 		     ((stp->ls_flags & NFSLCK_ACCESSBITS) &
2711 		      ((new_stp->ls_flags>>NFSLCK_SHIFT)&NFSLCK_ACCESSBITS)))){
2712 			ret = nfsrv_clientconflict(stp->ls_clp,&haslock,vp,p);
2713 			if (ret == 1) {
2714 				/*
2715 				 * nfsrv_clientconflict() unlocks
2716 				 * state when it returns non-zero.
2717 				 */
2718 				goto tryagain;
2719 			}
2720 			if (ret == 2)
2721 				error = NFSERR_PERM;
2722 			else if (new_stp->ls_flags & NFSLCK_RECLAIM)
2723 				error = NFSERR_RECLAIMCONFLICT;
2724 			else
2725 				error = NFSERR_SHAREDENIED;
2726 			if (ret == 0)
2727 				NFSUNLOCKSTATE();
2728 			if (haslock) {
2729 				NFSLOCKV4ROOTMUTEX();
2730 				nfsv4_unlock(&nfsv4rootfs_lock, 1);
2731 				NFSUNLOCKV4ROOTMUTEX();
2732 			}
2733 			goto out;
2734 		}
2735 	}
2736 
2737 	/*
2738 	 * Check for a conflicting delegation. If one is found, call
2739 	 * nfsrv_delegconflict() to handle it. If the v4root lock hasn't
2740 	 * been set yet, it will get the lock. Otherwise, it will recall
2741 	 * the delegation. Then, we try try again...
2742 	 * (If NFSLCK_DELEGCUR is set, it has a delegation, so there
2743 	 *  isn't a conflict.)
2744 	 * I currently believe the conflict algorithm to be:
2745 	 * For Open with Read Access and Deny None
2746 	 * - there is a conflict iff a different client has a write delegation
2747 	 * For Open with other Write Access or any Deny except None
2748 	 * - there is a conflict if a different client has any delegation
2749 	 * - there is a conflict if the same client has a read delegation
2750 	 *   (The current consensus is that this last case should be
2751 	 *    considered a conflict since the client with a read delegation
2752 	 *    could have done an Open with ReadAccess and WriteDeny
2753 	 *    locally and then not have checked for the WriteDeny.)
2754 	 *    The exception is a NFSv4.1/4.2 client that has requested
2755 	 *    an atomic upgrade to a write delegation.
2756 	 * Don't check for a Reclaim, since that will be dealt with
2757 	 * by nfsrv_openctrl().
2758 	 */
2759 	if (!(new_stp->ls_flags &
2760 		(NFSLCK_DELEGPREV | NFSLCK_DELEGCUR | NFSLCK_RECLAIM))) {
2761 	    stp = LIST_FIRST(&lfp->lf_deleg);
2762 	    while (stp != LIST_END(&lfp->lf_deleg)) {
2763 		nstp = LIST_NEXT(stp, ls_file);
2764 		if ((readonly && stp->ls_clp != clp &&
2765 		     (stp->ls_flags & NFSLCK_DELEGWRITE) != 0) ||
2766 		    (!readonly && (stp->ls_clp != clp ||
2767 		     ((stp->ls_flags & NFSLCK_DELEGREAD) != 0 &&
2768 		      (new_stp->ls_flags & NFSLCK_WANTWDELEG) == 0)))) {
2769 			ret = nfsrv_delegconflict(stp, &haslock, p, vp);
2770 			if (ret) {
2771 			    /*
2772 			     * nfsrv_delegconflict() unlocks state
2773 			     * when it returns non-zero.
2774 			     */
2775 			    if (ret == -1)
2776 				goto tryagain;
2777 			    error = ret;
2778 			    goto out;
2779 			}
2780 		}
2781 		stp = nstp;
2782 	    }
2783 	}
2784 	NFSUNLOCKSTATE();
2785 	if (haslock) {
2786 		NFSLOCKV4ROOTMUTEX();
2787 		nfsv4_unlock(&nfsv4rootfs_lock, 1);
2788 		NFSUNLOCKV4ROOTMUTEX();
2789 	}
2790 
2791 out:
2792 	NFSEXITCODE2(error, nd);
2793 	return (error);
2794 }
2795 
2796 /*
2797  * Open control function to create/update open state for an open.
2798  */
2799 int
2800 nfsrv_openctrl(struct nfsrv_descript *nd, vnode_t vp,
2801     struct nfsstate **new_stpp, nfsquad_t clientid, nfsv4stateid_t *stateidp,
2802     nfsv4stateid_t *delegstateidp, u_int32_t *rflagsp, struct nfsexstuff *exp,
2803     NFSPROC_T *p, u_quad_t filerev)
2804 {
2805 	struct nfsstate *new_stp = *new_stpp;
2806 	struct nfsstate *stp, *nstp;
2807 	struct nfsstate *openstp = NULL, *new_open, *ownerstp, *new_deleg;
2808 	struct nfslockfile *lfp, *new_lfp;
2809 	struct nfsclient *clp;
2810 	int error = 0, haslock = 0, ret, delegate = 1, writedeleg = 1;
2811 	int readonly = 0, cbret = 1, getfhret = 0;
2812 	int gotstate = 0, len = 0;
2813 	u_char *clidp = NULL;
2814 
2815 	if ((new_stp->ls_flags & NFSLCK_SHAREBITS) == NFSLCK_READACCESS)
2816 		readonly = 1;
2817 	/*
2818 	 * Check for restart conditions (client and server).
2819 	 * (Paranoia, should have been detected by nfsrv_opencheck().)
2820 	 * If an error does show up, return NFSERR_EXPIRED, since the
2821 	 * the seqid# has already been incremented.
2822 	 */
2823 	error = nfsrv_checkrestart(clientid, new_stp->ls_flags,
2824 	    &new_stp->ls_stateid, 0);
2825 	if (error) {
2826 		printf("Nfsd: openctrl unexpected restart err=%d\n",
2827 		    error);
2828 		error = NFSERR_EXPIRED;
2829 		goto out;
2830 	}
2831 
2832 	clidp = malloc(NFSV4_OPAQUELIMIT, M_TEMP, M_WAITOK);
2833 tryagain:
2834 	new_lfp = malloc(sizeof (struct nfslockfile),
2835 	    M_NFSDLOCKFILE, M_WAITOK);
2836 	new_open = malloc(sizeof (struct nfsstate),
2837 	    M_NFSDSTATE, M_WAITOK);
2838 	new_deleg = malloc(sizeof (struct nfsstate),
2839 	    M_NFSDSTATE, M_WAITOK);
2840 	getfhret = nfsrv_getlockfh(vp, new_stp->ls_flags, new_lfp,
2841 	    NULL, p);
2842 	NFSLOCKSTATE();
2843 	/*
2844 	 * Get the client structure. Since the linked lists could be changed
2845 	 * by other nfsd processes if this process does a tsleep(), one of
2846 	 * two things must be done.
2847 	 * 1 - don't tsleep()
2848 	 * or
2849 	 * 2 - get the nfsv4_lock() { indicated by haslock == 1 }
2850 	 *     before using the lists, since this lock stops the other
2851 	 *     nfsd. This should only be used for rare cases, since it
2852 	 *     essentially single threads the nfsd.
2853 	 *     At this time, it is only done for cases where the stable
2854 	 *     storage file must be written prior to completion of state
2855 	 *     expiration.
2856 	 */
2857 	error = nfsrv_getclient(clientid, CLOPS_RENEW, &clp, NULL,
2858 	    (nfsquad_t)((u_quad_t)0), 0, nd, p);
2859 	if (!error && (clp->lc_flags & LCL_NEEDSCBNULL) &&
2860 	    clp->lc_program) {
2861 		/*
2862 		 * This happens on the first open for a client
2863 		 * that supports callbacks.
2864 		 */
2865 		NFSUNLOCKSTATE();
2866 		/*
2867 		 * Although nfsrv_docallback() will sleep, clp won't
2868 		 * go away, since they are only removed when the
2869 		 * nfsv4_lock() has blocked the nfsd threads. The
2870 		 * fields in clp can change, but having multiple
2871 		 * threads do this Null callback RPC should be
2872 		 * harmless.
2873 		 */
2874 		cbret = nfsrv_docallback(clp, NFSV4PROC_CBNULL,
2875 		    NULL, 0, NULL, NULL, NULL, 0, p);
2876 		NFSLOCKSTATE();
2877 		clp->lc_flags &= ~LCL_NEEDSCBNULL;
2878 		if (!cbret)
2879 			clp->lc_flags |= LCL_CALLBACKSON;
2880 	}
2881 
2882 	/*
2883 	 * Look up the open owner. See if it needs confirmation and
2884 	 * check the seq#, as required.
2885 	 */
2886 	if (!error)
2887 		nfsrv_getowner(&clp->lc_open, new_stp, &ownerstp);
2888 
2889 	if (error) {
2890 		NFSUNLOCKSTATE();
2891 		printf("Nfsd: openctrl unexpected state err=%d\n",
2892 			error);
2893 		free(new_lfp, M_NFSDLOCKFILE);
2894 		free(new_open, M_NFSDSTATE);
2895 		free(new_deleg, M_NFSDSTATE);
2896 		if (haslock) {
2897 			NFSLOCKV4ROOTMUTEX();
2898 			nfsv4_unlock(&nfsv4rootfs_lock, 1);
2899 			NFSUNLOCKV4ROOTMUTEX();
2900 		}
2901 		error = NFSERR_EXPIRED;
2902 		goto out;
2903 	}
2904 
2905 	if (new_stp->ls_flags & NFSLCK_RECLAIM)
2906 		nfsrv_markstable(clp);
2907 
2908 	/*
2909 	 * Get the structure for the underlying file.
2910 	 */
2911 	if (getfhret)
2912 		error = getfhret;
2913 	else
2914 		error = nfsrv_getlockfile(new_stp->ls_flags, &new_lfp, &lfp,
2915 		    NULL, 0);
2916 	if (new_lfp)
2917 		free(new_lfp, M_NFSDLOCKFILE);
2918 	if (error) {
2919 		NFSUNLOCKSTATE();
2920 		printf("Nfsd openctrl unexpected getlockfile err=%d\n",
2921 		    error);
2922 		free(new_open, M_NFSDSTATE);
2923 		free(new_deleg, M_NFSDSTATE);
2924 		if (haslock) {
2925 			NFSLOCKV4ROOTMUTEX();
2926 			nfsv4_unlock(&nfsv4rootfs_lock, 1);
2927 			NFSUNLOCKV4ROOTMUTEX();
2928 		}
2929 		goto out;
2930 	}
2931 
2932 	/*
2933 	 * Search for a conflicting open/share.
2934 	 */
2935 	if (new_stp->ls_flags & NFSLCK_DELEGCUR) {
2936 	    /*
2937 	     * For Delegate_Cur, search for the matching Delegation,
2938 	     * which indicates no conflict.
2939 	     * An old delegation should have been recovered by the
2940 	     * client doing a Claim_DELEGATE_Prev, so I won't let
2941 	     * it match and return NFSERR_EXPIRED. Should I let it
2942 	     * match?
2943 	     */
2944 	    LIST_FOREACH(stp, &lfp->lf_deleg, ls_file) {
2945 		if (!(stp->ls_flags & NFSLCK_OLDDELEG) &&
2946 		    (((nd->nd_flag & ND_NFSV41) != 0 &&
2947 		    stateidp->seqid == 0) ||
2948 		    stateidp->seqid == stp->ls_stateid.seqid) &&
2949 		    !NFSBCMP(stateidp->other, stp->ls_stateid.other,
2950 			NFSX_STATEIDOTHER))
2951 			break;
2952 	    }
2953 	    if (stp == LIST_END(&lfp->lf_deleg) ||
2954 		((new_stp->ls_flags & NFSLCK_WRITEACCESS) &&
2955 		 (stp->ls_flags & NFSLCK_DELEGREAD))) {
2956 		NFSUNLOCKSTATE();
2957 		printf("Nfsd openctrl unexpected expiry\n");
2958 		free(new_open, M_NFSDSTATE);
2959 		free(new_deleg, M_NFSDSTATE);
2960 		if (haslock) {
2961 			NFSLOCKV4ROOTMUTEX();
2962 			nfsv4_unlock(&nfsv4rootfs_lock, 1);
2963 			NFSUNLOCKV4ROOTMUTEX();
2964 		}
2965 		error = NFSERR_EXPIRED;
2966 		goto out;
2967 	    }
2968 
2969 	    /*
2970 	     * Don't issue a Delegation, since one already exists and
2971 	     * delay delegation timeout, as required.
2972 	     */
2973 	    delegate = 0;
2974 	    nfsrv_delaydelegtimeout(stp);
2975 	}
2976 
2977 	/*
2978 	 * Check for access/deny bit conflicts. I also check for the
2979 	 * same owner, since the client might not have bothered to check.
2980 	 * Also, note an open for the same file and owner, if found,
2981 	 * which is all we do here for Delegate_Cur, since conflict
2982 	 * checking is already done.
2983 	 */
2984 	LIST_FOREACH(stp, &lfp->lf_open, ls_file) {
2985 		if (ownerstp && stp->ls_openowner == ownerstp)
2986 			openstp = stp;
2987 		if (!(new_stp->ls_flags & NFSLCK_DELEGCUR)) {
2988 		    /*
2989 		     * If another client has the file open, the only
2990 		     * delegation that can be issued is a Read delegation
2991 		     * and only if it is a Read open with Deny none.
2992 		     */
2993 		    if (clp != stp->ls_clp) {
2994 			if ((stp->ls_flags & NFSLCK_SHAREBITS) ==
2995 			    NFSLCK_READACCESS)
2996 			    writedeleg = 0;
2997 			else
2998 			    delegate = 0;
2999 		    }
3000 		    if(((new_stp->ls_flags & NFSLCK_ACCESSBITS) &
3001 		        ((stp->ls_flags>>NFSLCK_SHIFT) & NFSLCK_ACCESSBITS))||
3002 		       ((stp->ls_flags & NFSLCK_ACCESSBITS) &
3003 		        ((new_stp->ls_flags>>NFSLCK_SHIFT)&NFSLCK_ACCESSBITS))){
3004 			ret = nfsrv_clientconflict(stp->ls_clp,&haslock,vp,p);
3005 			if (ret == 1) {
3006 				/*
3007 				 * nfsrv_clientconflict() unlocks state
3008 				 * when it returns non-zero.
3009 				 */
3010 				free(new_open, M_NFSDSTATE);
3011 				free(new_deleg, M_NFSDSTATE);
3012 				openstp = NULL;
3013 				goto tryagain;
3014 			}
3015 			if (ret == 2)
3016 				error = NFSERR_PERM;
3017 			else if (new_stp->ls_flags & NFSLCK_RECLAIM)
3018 				error = NFSERR_RECLAIMCONFLICT;
3019 			else
3020 				error = NFSERR_SHAREDENIED;
3021 			if (ret == 0)
3022 				NFSUNLOCKSTATE();
3023 			if (haslock) {
3024 				NFSLOCKV4ROOTMUTEX();
3025 				nfsv4_unlock(&nfsv4rootfs_lock, 1);
3026 				NFSUNLOCKV4ROOTMUTEX();
3027 			}
3028 			free(new_open, M_NFSDSTATE);
3029 			free(new_deleg, M_NFSDSTATE);
3030 			printf("nfsd openctrl unexpected client cnfl\n");
3031 			goto out;
3032 		    }
3033 		}
3034 	}
3035 
3036 	/*
3037 	 * Check for a conflicting delegation. If one is found, call
3038 	 * nfsrv_delegconflict() to handle it. If the v4root lock hasn't
3039 	 * been set yet, it will get the lock. Otherwise, it will recall
3040 	 * the delegation. Then, we try try again...
3041 	 * (If NFSLCK_DELEGCUR is set, it has a delegation, so there
3042 	 *  isn't a conflict.)
3043 	 * I currently believe the conflict algorithm to be:
3044 	 * For Open with Read Access and Deny None
3045 	 * - there is a conflict iff a different client has a write delegation
3046 	 * For Open with other Write Access or any Deny except None
3047 	 * - there is a conflict if a different client has any delegation
3048 	 * - there is a conflict if the same client has a read delegation
3049 	 *   (The current consensus is that this last case should be
3050 	 *    considered a conflict since the client with a read delegation
3051 	 *    could have done an Open with ReadAccess and WriteDeny
3052 	 *    locally and then not have checked for the WriteDeny.)
3053 	 *    The exception is a NFSv4.1/4.2 client that has requested
3054 	 *    an atomic upgrade to a write delegation.
3055 	 */
3056 	if (!(new_stp->ls_flags & (NFSLCK_DELEGPREV | NFSLCK_DELEGCUR))) {
3057 	    stp = LIST_FIRST(&lfp->lf_deleg);
3058 	    while (stp != LIST_END(&lfp->lf_deleg)) {
3059 		nstp = LIST_NEXT(stp, ls_file);
3060 		if (stp->ls_clp != clp && (stp->ls_flags & NFSLCK_DELEGREAD))
3061 			writedeleg = 0;
3062 		else if (stp->ls_clp != clp ||
3063 		    (stp->ls_flags & NFSLCK_DELEGWRITE) != 0 ||
3064 		    (new_stp->ls_flags & NFSLCK_WANTWDELEG) == 0)
3065 			delegate = 0;
3066 		if ((readonly && stp->ls_clp != clp &&
3067 		     (stp->ls_flags & NFSLCK_DELEGWRITE) != 0) ||
3068 		    (!readonly && (stp->ls_clp != clp ||
3069 		     ((stp->ls_flags & NFSLCK_DELEGREAD) != 0 &&
3070 		      (new_stp->ls_flags & NFSLCK_WANTWDELEG) == 0)))) {
3071 		    if (new_stp->ls_flags & NFSLCK_RECLAIM) {
3072 			delegate = 2;
3073 		    } else {
3074 			ret = nfsrv_delegconflict(stp, &haslock, p, vp);
3075 			if (ret) {
3076 			    /*
3077 			     * nfsrv_delegconflict() unlocks state
3078 			     * when it returns non-zero.
3079 			     */
3080 			    printf("Nfsd openctrl unexpected deleg cnfl\n");
3081 			    free(new_open, M_NFSDSTATE);
3082 			    free(new_deleg, M_NFSDSTATE);
3083 			    if (ret == -1) {
3084 				openstp = NULL;
3085 				goto tryagain;
3086 			    }
3087 			    error = ret;
3088 			    goto out;
3089 			}
3090 		    }
3091 		}
3092 		stp = nstp;
3093 	    }
3094 	}
3095 
3096 	/*
3097 	 * We only get here if there was no open that conflicted.
3098 	 * If an open for the owner exists, or in the access/deny bits.
3099 	 * Otherwise it is a new open. If the open_owner hasn't been
3100 	 * confirmed, replace the open with the new one needing confirmation,
3101 	 * otherwise add the open.
3102 	 */
3103 	if (new_stp->ls_flags & NFSLCK_DELEGPREV) {
3104 	    /*
3105 	     * Handle NFSLCK_DELEGPREV by searching the old delegations for
3106 	     * a match. If found, just move the old delegation to the current
3107 	     * delegation list and issue open. If not found, return
3108 	     * NFSERR_EXPIRED.
3109 	     */
3110 	    LIST_FOREACH(stp, &clp->lc_olddeleg, ls_list) {
3111 		if (stp->ls_lfp == lfp) {
3112 		    /* Found it */
3113 		    if (stp->ls_clp != clp)
3114 			panic("olddeleg clp");
3115 		    LIST_REMOVE(stp, ls_list);
3116 		    LIST_REMOVE(stp, ls_hash);
3117 		    stp->ls_flags &= ~NFSLCK_OLDDELEG;
3118 		    stp->ls_stateid.seqid = delegstateidp->seqid = 1;
3119 		    stp->ls_stateid.other[0] = delegstateidp->other[0] =
3120 			clp->lc_clientid.lval[0];
3121 		    stp->ls_stateid.other[1] = delegstateidp->other[1] =
3122 			clp->lc_clientid.lval[1];
3123 		    stp->ls_stateid.other[2] = delegstateidp->other[2] =
3124 			nfsrv_nextstateindex(clp);
3125 		    stp->ls_compref = nd->nd_compref;
3126 		    LIST_INSERT_HEAD(&clp->lc_deleg, stp, ls_list);
3127 		    LIST_INSERT_HEAD(NFSSTATEHASH(clp,
3128 			stp->ls_stateid), stp, ls_hash);
3129 		    if (stp->ls_flags & NFSLCK_DELEGWRITE)
3130 			*rflagsp |= NFSV4OPEN_WRITEDELEGATE;
3131 		    else
3132 			*rflagsp |= NFSV4OPEN_READDELEGATE;
3133 		    clp->lc_delegtime = NFSD_MONOSEC +
3134 			nfsrv_lease + NFSRV_LEASEDELTA;
3135 
3136 		    /*
3137 		     * Now, do the associated open.
3138 		     */
3139 		    new_open->ls_stateid.seqid = 1;
3140 		    new_open->ls_stateid.other[0] = clp->lc_clientid.lval[0];
3141 		    new_open->ls_stateid.other[1] = clp->lc_clientid.lval[1];
3142 		    new_open->ls_stateid.other[2] = nfsrv_nextstateindex(clp);
3143 		    new_open->ls_flags = (new_stp->ls_flags&NFSLCK_DENYBITS)|
3144 			NFSLCK_OPEN;
3145 		    if (stp->ls_flags & NFSLCK_DELEGWRITE)
3146 			new_open->ls_flags |= (NFSLCK_READACCESS |
3147 			    NFSLCK_WRITEACCESS);
3148 		    else
3149 			new_open->ls_flags |= NFSLCK_READACCESS;
3150 		    new_open->ls_uid = new_stp->ls_uid;
3151 		    new_open->ls_lfp = lfp;
3152 		    new_open->ls_clp = clp;
3153 		    LIST_INIT(&new_open->ls_open);
3154 		    LIST_INSERT_HEAD(&lfp->lf_open, new_open, ls_file);
3155 		    LIST_INSERT_HEAD(NFSSTATEHASH(clp, new_open->ls_stateid),
3156 			new_open, ls_hash);
3157 		    /*
3158 		     * and handle the open owner
3159 		     */
3160 		    if (ownerstp) {
3161 			new_open->ls_openowner = ownerstp;
3162 			LIST_INSERT_HEAD(&ownerstp->ls_open,new_open,ls_list);
3163 		    } else {
3164 			new_open->ls_openowner = new_stp;
3165 			new_stp->ls_flags = 0;
3166 			nfsrvd_refcache(new_stp->ls_op);
3167 			new_stp->ls_noopens = 0;
3168 			LIST_INIT(&new_stp->ls_open);
3169 			LIST_INSERT_HEAD(&new_stp->ls_open, new_open, ls_list);
3170 			LIST_INSERT_HEAD(&clp->lc_open, new_stp, ls_list);
3171 			*new_stpp = NULL;
3172 			NFSD_VNET(nfsstatsv1_p)->srvopenowners++;
3173 			nfsrv_openpluslock++;
3174 		    }
3175 		    openstp = new_open;
3176 		    new_open = NULL;
3177 		    NFSD_VNET(nfsstatsv1_p)->srvopens++;
3178 		    nfsrv_openpluslock++;
3179 		    break;
3180 		}
3181 	    }
3182 	    if (stp == LIST_END(&clp->lc_olddeleg))
3183 		error = NFSERR_EXPIRED;
3184 	} else if (new_stp->ls_flags & (NFSLCK_DELEGREAD | NFSLCK_DELEGWRITE)) {
3185 	    /*
3186 	     * Scan to see that no delegation for this client and file
3187 	     * doesn't already exist.
3188 	     * There also shouldn't yet be an Open for this file and
3189 	     * openowner.
3190 	     */
3191 	    LIST_FOREACH(stp, &lfp->lf_deleg, ls_file) {
3192 		if (stp->ls_clp == clp)
3193 		    break;
3194 	    }
3195 	    if (stp == LIST_END(&lfp->lf_deleg) && openstp == NULL) {
3196 		/*
3197 		 * This is the Claim_Previous case with a delegation
3198 		 * type != Delegate_None.
3199 		 */
3200 		/*
3201 		 * First, add the delegation. (Although we must issue the
3202 		 * delegation, we can also ask for an immediate return.)
3203 		 */
3204 		new_deleg->ls_stateid.seqid = delegstateidp->seqid = 1;
3205 		new_deleg->ls_stateid.other[0] = delegstateidp->other[0] =
3206 		    clp->lc_clientid.lval[0];
3207 		new_deleg->ls_stateid.other[1] = delegstateidp->other[1] =
3208 		    clp->lc_clientid.lval[1];
3209 		new_deleg->ls_stateid.other[2] = delegstateidp->other[2] =
3210 		    nfsrv_nextstateindex(clp);
3211 		if (new_stp->ls_flags & NFSLCK_DELEGWRITE) {
3212 		    new_deleg->ls_flags = (NFSLCK_DELEGWRITE |
3213 			NFSLCK_READACCESS | NFSLCK_WRITEACCESS);
3214 		    *rflagsp |= NFSV4OPEN_WRITEDELEGATE;
3215 		    nfsrv_writedelegcnt++;
3216 		} else {
3217 		    new_deleg->ls_flags = (NFSLCK_DELEGREAD |
3218 			NFSLCK_READACCESS);
3219 		    *rflagsp |= NFSV4OPEN_READDELEGATE;
3220 		}
3221 		new_deleg->ls_uid = new_stp->ls_uid;
3222 		new_deleg->ls_lfp = lfp;
3223 		new_deleg->ls_clp = clp;
3224 		new_deleg->ls_filerev = filerev;
3225 		new_deleg->ls_compref = nd->nd_compref;
3226 		new_deleg->ls_lastrecall = 0;
3227 		LIST_INSERT_HEAD(&lfp->lf_deleg, new_deleg, ls_file);
3228 		LIST_INSERT_HEAD(NFSSTATEHASH(clp,
3229 		    new_deleg->ls_stateid), new_deleg, ls_hash);
3230 		LIST_INSERT_HEAD(&clp->lc_deleg, new_deleg, ls_list);
3231 		new_deleg = NULL;
3232 		if (delegate == 2 || nfsrv_issuedelegs == 0 ||
3233 		    (clp->lc_flags & (LCL_CALLBACKSON | LCL_CBDOWN)) !=
3234 		     LCL_CALLBACKSON ||
3235 		    NFSRV_V4DELEGLIMIT(nfsrv_delegatecnt) ||
3236 		    !NFSVNO_DELEGOK(vp))
3237 		    *rflagsp |= NFSV4OPEN_RECALL;
3238 		NFSD_VNET(nfsstatsv1_p)->srvdelegates++;
3239 		nfsrv_openpluslock++;
3240 		nfsrv_delegatecnt++;
3241 
3242 		/*
3243 		 * Now, do the associated open.
3244 		 */
3245 		new_open->ls_stateid.seqid = 1;
3246 		new_open->ls_stateid.other[0] = clp->lc_clientid.lval[0];
3247 		new_open->ls_stateid.other[1] = clp->lc_clientid.lval[1];
3248 		new_open->ls_stateid.other[2] = nfsrv_nextstateindex(clp);
3249 		new_open->ls_flags = (new_stp->ls_flags & NFSLCK_DENYBITS) |
3250 		    NFSLCK_OPEN;
3251 		if (new_stp->ls_flags & NFSLCK_DELEGWRITE)
3252 			new_open->ls_flags |= (NFSLCK_READACCESS |
3253 			    NFSLCK_WRITEACCESS);
3254 		else
3255 			new_open->ls_flags |= NFSLCK_READACCESS;
3256 		new_open->ls_uid = new_stp->ls_uid;
3257 		new_open->ls_lfp = lfp;
3258 		new_open->ls_clp = clp;
3259 		LIST_INIT(&new_open->ls_open);
3260 		LIST_INSERT_HEAD(&lfp->lf_open, new_open, ls_file);
3261 		LIST_INSERT_HEAD(NFSSTATEHASH(clp, new_open->ls_stateid),
3262 		   new_open, ls_hash);
3263 		/*
3264 		 * and handle the open owner
3265 		 */
3266 		if (ownerstp) {
3267 		    new_open->ls_openowner = ownerstp;
3268 		    LIST_INSERT_HEAD(&ownerstp->ls_open, new_open, ls_list);
3269 		} else {
3270 		    new_open->ls_openowner = new_stp;
3271 		    new_stp->ls_flags = 0;
3272 		    nfsrvd_refcache(new_stp->ls_op);
3273 		    new_stp->ls_noopens = 0;
3274 		    LIST_INIT(&new_stp->ls_open);
3275 		    LIST_INSERT_HEAD(&new_stp->ls_open, new_open, ls_list);
3276 		    LIST_INSERT_HEAD(&clp->lc_open, new_stp, ls_list);
3277 		    *new_stpp = NULL;
3278 		    NFSD_VNET(nfsstatsv1_p)->srvopenowners++;
3279 		    nfsrv_openpluslock++;
3280 		}
3281 		openstp = new_open;
3282 		new_open = NULL;
3283 		NFSD_VNET(nfsstatsv1_p)->srvopens++;
3284 		nfsrv_openpluslock++;
3285 	    } else {
3286 		error = NFSERR_RECLAIMCONFLICT;
3287 	    }
3288 	} else if (ownerstp) {
3289 		if (ownerstp->ls_flags & NFSLCK_NEEDSCONFIRM) {
3290 		    /* Replace the open */
3291 		    if (ownerstp->ls_op)
3292 			nfsrvd_derefcache(ownerstp->ls_op);
3293 		    ownerstp->ls_op = new_stp->ls_op;
3294 		    nfsrvd_refcache(ownerstp->ls_op);
3295 		    ownerstp->ls_seq = new_stp->ls_seq;
3296 		    *rflagsp |= NFSV4OPEN_RESULTCONFIRM;
3297 		    stp = LIST_FIRST(&ownerstp->ls_open);
3298 		    stp->ls_flags = (new_stp->ls_flags & NFSLCK_SHAREBITS) |
3299 			NFSLCK_OPEN;
3300 		    stp->ls_stateid.seqid = 1;
3301 		    stp->ls_uid = new_stp->ls_uid;
3302 		    if (lfp != stp->ls_lfp) {
3303 			LIST_REMOVE(stp, ls_file);
3304 			LIST_INSERT_HEAD(&lfp->lf_open, stp, ls_file);
3305 			stp->ls_lfp = lfp;
3306 		    }
3307 		    openstp = stp;
3308 		} else if (openstp) {
3309 		    openstp->ls_flags |= (new_stp->ls_flags & NFSLCK_SHAREBITS);
3310 		    openstp->ls_stateid.seqid++;
3311 		    if ((nd->nd_flag & ND_NFSV41) != 0 &&
3312 			openstp->ls_stateid.seqid == 0)
3313 			openstp->ls_stateid.seqid = 1;
3314 
3315 		    /*
3316 		     * This is where we can choose to issue a delegation.
3317 		     */
3318 		    nfsrv_issuedelegation(vp, clp, nd, delegate, writedeleg,
3319 			readonly, filerev, NFSVNO_EXRDONLY(exp), &new_deleg,
3320 			new_stp, lfp, rflagsp, delegstateidp);
3321 		} else {
3322 		    new_open->ls_stateid.seqid = 1;
3323 		    new_open->ls_stateid.other[0] = clp->lc_clientid.lval[0];
3324 		    new_open->ls_stateid.other[1] = clp->lc_clientid.lval[1];
3325 		    new_open->ls_stateid.other[2] = nfsrv_nextstateindex(clp);
3326 		    new_open->ls_flags = (new_stp->ls_flags & NFSLCK_SHAREBITS)|
3327 			NFSLCK_OPEN;
3328 		    new_open->ls_uid = new_stp->ls_uid;
3329 		    new_open->ls_openowner = ownerstp;
3330 		    new_open->ls_lfp = lfp;
3331 		    new_open->ls_clp = clp;
3332 		    LIST_INIT(&new_open->ls_open);
3333 		    LIST_INSERT_HEAD(&lfp->lf_open, new_open, ls_file);
3334 		    LIST_INSERT_HEAD(&ownerstp->ls_open, new_open, ls_list);
3335 		    LIST_INSERT_HEAD(NFSSTATEHASH(clp, new_open->ls_stateid),
3336 			new_open, ls_hash);
3337 		    openstp = new_open;
3338 		    new_open = NULL;
3339 		    NFSD_VNET(nfsstatsv1_p)->srvopens++;
3340 		    nfsrv_openpluslock++;
3341 
3342 		    /*
3343 		     * This is where we can choose to issue a delegation.
3344 		     */
3345 		    nfsrv_issuedelegation(vp, clp, nd, delegate, writedeleg,
3346 			readonly, filerev, NFSVNO_EXRDONLY(exp), &new_deleg,
3347 			new_stp, lfp, rflagsp, delegstateidp);
3348 		}
3349 	} else {
3350 		/*
3351 		 * New owner case. Start the open_owner sequence with a
3352 		 * Needs confirmation (unless a reclaim) and hang the
3353 		 * new open off it.
3354 		 */
3355 		new_open->ls_stateid.seqid = 1;
3356 		new_open->ls_stateid.other[0] = clp->lc_clientid.lval[0];
3357 		new_open->ls_stateid.other[1] = clp->lc_clientid.lval[1];
3358 		new_open->ls_stateid.other[2] = nfsrv_nextstateindex(clp);
3359 		new_open->ls_flags = (new_stp->ls_flags & NFSLCK_SHAREBITS) |
3360 		    NFSLCK_OPEN;
3361 		new_open->ls_uid = new_stp->ls_uid;
3362 		LIST_INIT(&new_open->ls_open);
3363 		new_open->ls_openowner = new_stp;
3364 		new_open->ls_lfp = lfp;
3365 		new_open->ls_clp = clp;
3366 		LIST_INSERT_HEAD(&lfp->lf_open, new_open, ls_file);
3367 		if (new_stp->ls_flags & NFSLCK_RECLAIM) {
3368 			new_stp->ls_flags = 0;
3369 		} else if ((nd->nd_flag & ND_NFSV41) != 0) {
3370 		    /*
3371 		     * This is where we can choose to issue a delegation.
3372 		     */
3373 		    nfsrv_issuedelegation(vp, clp, nd, delegate, writedeleg,
3374 			readonly, filerev, NFSVNO_EXRDONLY(exp), &new_deleg,
3375 			new_stp, lfp, rflagsp, delegstateidp);
3376 		    /* NFSv4.1 never needs confirmation. */
3377 		    new_stp->ls_flags = 0;
3378 
3379 		    /*
3380 		     * Since NFSv4.1 never does an OpenConfirm, the first
3381 		     * open state will be acquired here.
3382 		     */
3383 		    if (!(clp->lc_flags & LCL_STAMPEDSTABLE)) {
3384 			clp->lc_flags |= LCL_STAMPEDSTABLE;
3385 			len = clp->lc_idlen;
3386 			NFSBCOPY(clp->lc_id, clidp, len);
3387 			gotstate = 1;
3388 		    }
3389 		} else {
3390 		    *rflagsp |= NFSV4OPEN_RESULTCONFIRM;
3391 		    new_stp->ls_flags = NFSLCK_NEEDSCONFIRM;
3392 		}
3393 		nfsrvd_refcache(new_stp->ls_op);
3394 		new_stp->ls_noopens = 0;
3395 		LIST_INIT(&new_stp->ls_open);
3396 		LIST_INSERT_HEAD(&new_stp->ls_open, new_open, ls_list);
3397 		LIST_INSERT_HEAD(&clp->lc_open, new_stp, ls_list);
3398 		LIST_INSERT_HEAD(NFSSTATEHASH(clp, new_open->ls_stateid),
3399 		    new_open, ls_hash);
3400 		openstp = new_open;
3401 		new_open = NULL;
3402 		*new_stpp = NULL;
3403 		NFSD_VNET(nfsstatsv1_p)->srvopens++;
3404 		nfsrv_openpluslock++;
3405 		NFSD_VNET(nfsstatsv1_p)->srvopenowners++;
3406 		nfsrv_openpluslock++;
3407 	}
3408 	if (!error) {
3409 		stateidp->seqid = openstp->ls_stateid.seqid;
3410 		stateidp->other[0] = openstp->ls_stateid.other[0];
3411 		stateidp->other[1] = openstp->ls_stateid.other[1];
3412 		stateidp->other[2] = openstp->ls_stateid.other[2];
3413 	}
3414 	NFSUNLOCKSTATE();
3415 	if (haslock) {
3416 		NFSLOCKV4ROOTMUTEX();
3417 		nfsv4_unlock(&nfsv4rootfs_lock, 1);
3418 		NFSUNLOCKV4ROOTMUTEX();
3419 	}
3420 	if (new_open)
3421 		free(new_open, M_NFSDSTATE);
3422 	if (new_deleg)
3423 		free(new_deleg, M_NFSDSTATE);
3424 
3425 	/*
3426 	 * If the NFSv4.1 client just acquired its first open, write a timestamp
3427 	 * to the stable storage file.
3428 	 */
3429 	if (gotstate != 0) {
3430 		nfsrv_writestable(clidp, len, NFSNST_NEWSTATE, p);
3431 		nfsrv_backupstable();
3432 	}
3433 
3434 out:
3435 	free(clidp, M_TEMP);
3436 	NFSEXITCODE2(error, nd);
3437 	return (error);
3438 }
3439 
3440 /*
3441  * Open update. Does the confirm, downgrade and close.
3442  */
3443 int
3444 nfsrv_openupdate(vnode_t vp, struct nfsstate *new_stp, nfsquad_t clientid,
3445     nfsv4stateid_t *stateidp, struct nfsrv_descript *nd, NFSPROC_T *p,
3446     int *retwriteaccessp)
3447 {
3448 	struct nfsstate *stp;
3449 	struct nfsclient *clp;
3450 	u_int32_t bits;
3451 	int error = 0, gotstate = 0, len = 0;
3452 	u_char *clidp = NULL;
3453 
3454 	/*
3455 	 * Check for restart conditions (client and server).
3456 	 */
3457 	error = nfsrv_checkrestart(clientid, new_stp->ls_flags,
3458 	    &new_stp->ls_stateid, 0);
3459 	if (error)
3460 		goto out;
3461 
3462 	clidp = malloc(NFSV4_OPAQUELIMIT, M_TEMP, M_WAITOK);
3463 	NFSLOCKSTATE();
3464 	/*
3465 	 * Get the open structure via clientid and stateid.
3466 	 */
3467 	error = nfsrv_getclient(clientid, CLOPS_RENEW, &clp, NULL,
3468 	    (nfsquad_t)((u_quad_t)0), 0, nd, p);
3469 	if (!error)
3470 		error = nfsrv_getstate(clp, &new_stp->ls_stateid,
3471 		    new_stp->ls_flags, &stp);
3472 
3473 	/*
3474 	 * Sanity check the open.
3475 	 */
3476 	if (!error && (!(stp->ls_flags & NFSLCK_OPEN) ||
3477 		(!(new_stp->ls_flags & NFSLCK_CONFIRM) &&
3478 		 (stp->ls_openowner->ls_flags & NFSLCK_NEEDSCONFIRM)) ||
3479 		((new_stp->ls_flags & NFSLCK_CONFIRM) &&
3480 		 (!(stp->ls_openowner->ls_flags & NFSLCK_NEEDSCONFIRM)))))
3481 		error = NFSERR_BADSTATEID;
3482 
3483 	if (!error)
3484 		error = nfsrv_checkseqid(nd, new_stp->ls_seq,
3485 		    stp->ls_openowner, new_stp->ls_op);
3486 	if (!error && stp->ls_stateid.seqid != new_stp->ls_stateid.seqid &&
3487 	    (((nd->nd_flag & ND_NFSV41) == 0 &&
3488 	      !(new_stp->ls_flags & NFSLCK_CONFIRM)) ||
3489 	     ((nd->nd_flag & ND_NFSV41) != 0 &&
3490 	      new_stp->ls_stateid.seqid != 0)))
3491 		error = NFSERR_OLDSTATEID;
3492 	if (!error && vp->v_type != VREG) {
3493 		if (vp->v_type == VDIR)
3494 			error = NFSERR_ISDIR;
3495 		else
3496 			error = NFSERR_INVAL;
3497 	}
3498 
3499 	if (error) {
3500 		/*
3501 		 * If a client tries to confirm an Open with a bad
3502 		 * seqid# and there are no byte range locks or other Opens
3503 		 * on the openowner, just throw it away, so the next use of the
3504 		 * openowner will start a fresh seq#.
3505 		 */
3506 		if (error == NFSERR_BADSEQID &&
3507 		    (new_stp->ls_flags & NFSLCK_CONFIRM) &&
3508 		    nfsrv_nootherstate(stp))
3509 			nfsrv_freeopenowner(stp->ls_openowner, 0, p);
3510 		NFSUNLOCKSTATE();
3511 		goto out;
3512 	}
3513 
3514 	/*
3515 	 * Set the return stateid.
3516 	 */
3517 	stateidp->seqid = stp->ls_stateid.seqid + 1;
3518 	if ((nd->nd_flag & ND_NFSV41) != 0 && stateidp->seqid == 0)
3519 		stateidp->seqid = 1;
3520 	stateidp->other[0] = stp->ls_stateid.other[0];
3521 	stateidp->other[1] = stp->ls_stateid.other[1];
3522 	stateidp->other[2] = stp->ls_stateid.other[2];
3523 	/*
3524 	 * Now, handle the three cases.
3525 	 */
3526 	if (new_stp->ls_flags & NFSLCK_CONFIRM) {
3527 		/*
3528 		 * If the open doesn't need confirmation, it seems to me that
3529 		 * there is a client error, but I'll just log it and keep going?
3530 		 */
3531 		if (!(stp->ls_openowner->ls_flags & NFSLCK_NEEDSCONFIRM))
3532 			printf("Nfsv4d: stray open confirm\n");
3533 		stp->ls_openowner->ls_flags = 0;
3534 		stp->ls_stateid.seqid++;
3535 		if ((nd->nd_flag & ND_NFSV41) != 0 &&
3536 		    stp->ls_stateid.seqid == 0)
3537 			stp->ls_stateid.seqid = 1;
3538 		if (!(clp->lc_flags & LCL_STAMPEDSTABLE)) {
3539 			clp->lc_flags |= LCL_STAMPEDSTABLE;
3540 			len = clp->lc_idlen;
3541 			NFSBCOPY(clp->lc_id, clidp, len);
3542 			gotstate = 1;
3543 		}
3544 	} else if (new_stp->ls_flags & NFSLCK_CLOSE) {
3545 		if (retwriteaccessp != NULL) {
3546 			if ((stp->ls_flags & NFSLCK_WRITEACCESS) != 0)
3547 				*retwriteaccessp = 1;
3548 			else
3549 				*retwriteaccessp = 0;
3550 		}
3551 		if (nfsrv_dolocallocks != 0 && !LIST_EMPTY(&stp->ls_open)) {
3552 			ASSERT_VOP_ELOCKED(vp, "nfsrv_openupdate");
3553 			nfsrv_freeopen(stp, vp, 1, p);
3554 		} else {
3555 			nfsrv_freeopen(stp, NULL, 0, p);
3556 		}
3557 	} else {
3558 		/*
3559 		 * Update the share bits, making sure that the new set are a
3560 		 * subset of the old ones.
3561 		 */
3562 		bits = (new_stp->ls_flags & NFSLCK_SHAREBITS);
3563 		if (~(stp->ls_flags) & bits) {
3564 			NFSUNLOCKSTATE();
3565 			error = NFSERR_INVAL;
3566 			goto out;
3567 		}
3568 		stp->ls_flags = (bits | NFSLCK_OPEN);
3569 		stp->ls_stateid.seqid++;
3570 		if ((nd->nd_flag & ND_NFSV41) != 0 &&
3571 		    stp->ls_stateid.seqid == 0)
3572 			stp->ls_stateid.seqid = 1;
3573 	}
3574 	NFSUNLOCKSTATE();
3575 
3576 	/*
3577 	 * If the client just confirmed its first open, write a timestamp
3578 	 * to the stable storage file.
3579 	 */
3580 	if (gotstate != 0) {
3581 		nfsrv_writestable(clidp, len, NFSNST_NEWSTATE, p);
3582 		nfsrv_backupstable();
3583 	}
3584 
3585 out:
3586 	free(clidp, M_TEMP);
3587 	NFSEXITCODE2(error, nd);
3588 	return (error);
3589 }
3590 
3591 /*
3592  * Delegation update. Does the purge and return.
3593  */
3594 int
3595 nfsrv_delegupdate(struct nfsrv_descript *nd, nfsquad_t clientid,
3596     nfsv4stateid_t *stateidp, vnode_t vp, int op, struct ucred *cred,
3597     NFSPROC_T *p, int *retwriteaccessp)
3598 {
3599 	struct nfsstate *stp;
3600 	struct nfsclient *clp;
3601 	int error = 0;
3602 	fhandle_t fh;
3603 
3604 	/*
3605 	 * Do a sanity check against the file handle for DelegReturn.
3606 	 */
3607 	if (vp) {
3608 		error = nfsvno_getfh(vp, &fh, p);
3609 		if (error)
3610 			goto out;
3611 	}
3612 	/*
3613 	 * Check for restart conditions (client and server).
3614 	 */
3615 	if (op == NFSV4OP_DELEGRETURN)
3616 		error = nfsrv_checkrestart(clientid, NFSLCK_DELEGRETURN,
3617 			stateidp, 0);
3618 	else
3619 		error = nfsrv_checkrestart(clientid, NFSLCK_DELEGPURGE,
3620 			stateidp, 0);
3621 
3622 	NFSLOCKSTATE();
3623 	/*
3624 	 * Get the open structure via clientid and stateid.
3625 	 */
3626 	if (!error)
3627 	    error = nfsrv_getclient(clientid, CLOPS_RENEW, &clp, NULL,
3628 		(nfsquad_t)((u_quad_t)0), 0, nd, p);
3629 	if (error) {
3630 		if (error == NFSERR_CBPATHDOWN)
3631 			error = 0;
3632 		if (error == NFSERR_STALECLIENTID && op == NFSV4OP_DELEGRETURN)
3633 			error = NFSERR_STALESTATEID;
3634 	}
3635 	if (!error && op == NFSV4OP_DELEGRETURN) {
3636 	    error = nfsrv_getstate(clp, stateidp, NFSLCK_DELEGRETURN, &stp);
3637 	    if (!error && stp->ls_stateid.seqid != stateidp->seqid &&
3638 		((nd->nd_flag & ND_NFSV41) == 0 || stateidp->seqid != 0))
3639 		error = NFSERR_OLDSTATEID;
3640 	}
3641 	/*
3642 	 * NFSERR_EXPIRED means that the state has gone away,
3643 	 * so Delegations have been purged. Just return ok.
3644 	 */
3645 	if (error == NFSERR_EXPIRED && op == NFSV4OP_DELEGPURGE) {
3646 		NFSUNLOCKSTATE();
3647 		error = 0;
3648 		goto out;
3649 	}
3650 	if (error) {
3651 		NFSUNLOCKSTATE();
3652 		goto out;
3653 	}
3654 
3655 	if (op == NFSV4OP_DELEGRETURN) {
3656 		if (NFSBCMP((caddr_t)&fh, (caddr_t)&stp->ls_lfp->lf_fh,
3657 		    sizeof (fhandle_t))) {
3658 			NFSUNLOCKSTATE();
3659 			error = NFSERR_BADSTATEID;
3660 			goto out;
3661 		}
3662 		if (retwriteaccessp != NULL) {
3663 			if ((stp->ls_flags & NFSLCK_DELEGWRITE) != 0)
3664 				*retwriteaccessp = 1;
3665 			else
3666 				*retwriteaccessp = 0;
3667 		}
3668 		nfsrv_freedeleg(stp);
3669 	} else {
3670 		nfsrv_freedeleglist(&clp->lc_olddeleg);
3671 	}
3672 	NFSUNLOCKSTATE();
3673 	error = 0;
3674 
3675 out:
3676 	NFSEXITCODE(error);
3677 	return (error);
3678 }
3679 
3680 /*
3681  * Release lock owner.
3682  */
3683 int
3684 nfsrv_releaselckown(struct nfsstate *new_stp, nfsquad_t clientid,
3685     NFSPROC_T *p)
3686 {
3687 	struct nfsstate *stp, *nstp, *openstp, *ownstp;
3688 	struct nfsclient *clp;
3689 	int error = 0;
3690 
3691 	/*
3692 	 * Check for restart conditions (client and server).
3693 	 */
3694 	error = nfsrv_checkrestart(clientid, new_stp->ls_flags,
3695 	    &new_stp->ls_stateid, 0);
3696 	if (error)
3697 		goto out;
3698 
3699 	NFSLOCKSTATE();
3700 	/*
3701 	 * Get the lock owner by name.
3702 	 */
3703 	error = nfsrv_getclient(clientid, CLOPS_RENEW, &clp, NULL,
3704 	    (nfsquad_t)((u_quad_t)0), 0, NULL, p);
3705 	if (error) {
3706 		NFSUNLOCKSTATE();
3707 		goto out;
3708 	}
3709 	LIST_FOREACH(ownstp, &clp->lc_open, ls_list) {
3710 	    LIST_FOREACH(openstp, &ownstp->ls_open, ls_list) {
3711 		stp = LIST_FIRST(&openstp->ls_open);
3712 		while (stp != LIST_END(&openstp->ls_open)) {
3713 		    nstp = LIST_NEXT(stp, ls_list);
3714 		    /*
3715 		     * If the owner matches, check for locks and
3716 		     * then free or return an error.
3717 		     */
3718 		    if (stp->ls_ownerlen == new_stp->ls_ownerlen &&
3719 			!NFSBCMP(stp->ls_owner, new_stp->ls_owner,
3720 			 stp->ls_ownerlen)){
3721 			if (LIST_EMPTY(&stp->ls_lock)) {
3722 			    nfsrv_freelockowner(stp, NULL, 0, p);
3723 			} else {
3724 			    NFSUNLOCKSTATE();
3725 			    error = NFSERR_LOCKSHELD;
3726 			    goto out;
3727 			}
3728 		    }
3729 		    stp = nstp;
3730 		}
3731 	    }
3732 	}
3733 	NFSUNLOCKSTATE();
3734 
3735 out:
3736 	NFSEXITCODE(error);
3737 	return (error);
3738 }
3739 
3740 /*
3741  * Get the file handle for a lock structure.
3742  */
3743 static int
3744 nfsrv_getlockfh(vnode_t vp, u_short flags, struct nfslockfile *new_lfp,
3745     fhandle_t *nfhp, NFSPROC_T *p)
3746 {
3747 	fhandle_t *fhp = NULL;
3748 	int error;
3749 
3750 	/*
3751 	 * For lock, use the new nfslock structure, otherwise just
3752 	 * a fhandle_t on the stack.
3753 	 */
3754 	if (flags & NFSLCK_OPEN) {
3755 		KASSERT(new_lfp != NULL, ("nfsrv_getlockfh: new_lfp NULL"));
3756 		fhp = &new_lfp->lf_fh;
3757 	} else if (nfhp) {
3758 		fhp = nfhp;
3759 	} else {
3760 		panic("nfsrv_getlockfh");
3761 	}
3762 	error = nfsvno_getfh(vp, fhp, p);
3763 	NFSEXITCODE(error);
3764 	return (error);
3765 }
3766 
3767 /*
3768  * Get an nfs lock structure. Allocate one, as required, and return a
3769  * pointer to it.
3770  * Returns an NFSERR_xxx upon failure or -1 to indicate no current lock.
3771  */
3772 static int
3773 nfsrv_getlockfile(u_short flags, struct nfslockfile **new_lfpp,
3774     struct nfslockfile **lfpp, fhandle_t *nfhp, int lockit)
3775 {
3776 	struct nfslockfile *lfp;
3777 	fhandle_t *fhp = NULL, *tfhp;
3778 	struct nfslockhashhead *hp;
3779 	struct nfslockfile *new_lfp = NULL;
3780 
3781 	/*
3782 	 * For lock, use the new nfslock structure, otherwise just
3783 	 * a fhandle_t on the stack.
3784 	 */
3785 	if (flags & NFSLCK_OPEN) {
3786 		new_lfp = *new_lfpp;
3787 		fhp = &new_lfp->lf_fh;
3788 	} else if (nfhp) {
3789 		fhp = nfhp;
3790 	} else {
3791 		panic("nfsrv_getlockfile");
3792 	}
3793 
3794 	hp = NFSLOCKHASH(fhp);
3795 	LIST_FOREACH(lfp, hp, lf_hash) {
3796 		tfhp = &lfp->lf_fh;
3797 		if (NFSVNO_CMPFH(fhp, tfhp)) {
3798 			if (lockit)
3799 				nfsrv_locklf(lfp);
3800 			*lfpp = lfp;
3801 			return (0);
3802 		}
3803 	}
3804 	if (!(flags & NFSLCK_OPEN))
3805 		return (-1);
3806 
3807 	/*
3808 	 * No match, so chain the new one into the list.
3809 	 */
3810 	LIST_INIT(&new_lfp->lf_open);
3811 	LIST_INIT(&new_lfp->lf_lock);
3812 	LIST_INIT(&new_lfp->lf_deleg);
3813 	LIST_INIT(&new_lfp->lf_locallock);
3814 	LIST_INIT(&new_lfp->lf_rollback);
3815 	new_lfp->lf_locallock_lck.nfslock_usecnt = 0;
3816 	new_lfp->lf_locallock_lck.nfslock_lock = 0;
3817 	new_lfp->lf_usecount = 0;
3818 	LIST_INSERT_HEAD(hp, new_lfp, lf_hash);
3819 	*lfpp = new_lfp;
3820 	*new_lfpp = NULL;
3821 	return (0);
3822 }
3823 
3824 /*
3825  * This function adds a nfslock lock structure to the list for the associated
3826  * nfsstate and nfslockfile structures. It will be inserted after the
3827  * entry pointed at by insert_lop.
3828  */
3829 static void
3830 nfsrv_insertlock(struct nfslock *new_lop, struct nfslock *insert_lop,
3831     struct nfsstate *stp, struct nfslockfile *lfp)
3832 {
3833 	struct nfslock *lop, *nlop;
3834 
3835 	new_lop->lo_stp = stp;
3836 	new_lop->lo_lfp = lfp;
3837 
3838 	if (stp != NULL) {
3839 		/* Insert in increasing lo_first order */
3840 		lop = LIST_FIRST(&lfp->lf_lock);
3841 		if (lop == LIST_END(&lfp->lf_lock) ||
3842 		    new_lop->lo_first <= lop->lo_first) {
3843 			LIST_INSERT_HEAD(&lfp->lf_lock, new_lop, lo_lckfile);
3844 		} else {
3845 			nlop = LIST_NEXT(lop, lo_lckfile);
3846 			while (nlop != LIST_END(&lfp->lf_lock) &&
3847 			       nlop->lo_first < new_lop->lo_first) {
3848 				lop = nlop;
3849 				nlop = LIST_NEXT(lop, lo_lckfile);
3850 			}
3851 			LIST_INSERT_AFTER(lop, new_lop, lo_lckfile);
3852 		}
3853 	} else {
3854 		new_lop->lo_lckfile.le_prev = NULL;	/* list not used */
3855 	}
3856 
3857 	/*
3858 	 * Insert after insert_lop, which is overloaded as stp or lfp for
3859 	 * an empty list.
3860 	 */
3861 	if (stp == NULL && (struct nfslockfile *)insert_lop == lfp)
3862 		LIST_INSERT_HEAD(&lfp->lf_locallock, new_lop, lo_lckowner);
3863 	else if ((struct nfsstate *)insert_lop == stp)
3864 		LIST_INSERT_HEAD(&stp->ls_lock, new_lop, lo_lckowner);
3865 	else
3866 		LIST_INSERT_AFTER(insert_lop, new_lop, lo_lckowner);
3867 	if (stp != NULL) {
3868 		NFSD_VNET(nfsstatsv1_p)->srvlocks++;
3869 		nfsrv_openpluslock++;
3870 	}
3871 }
3872 
3873 /*
3874  * This function updates the locking for a lock owner and given file. It
3875  * maintains a list of lock ranges ordered on increasing file offset that
3876  * are NFSLCK_READ or NFSLCK_WRITE and non-overlapping (aka POSIX style).
3877  * It always adds new_lop to the list and sometimes uses the one pointed
3878  * at by other_lopp.
3879  */
3880 static void
3881 nfsrv_updatelock(struct nfsstate *stp, struct nfslock **new_lopp,
3882     struct nfslock **other_lopp, struct nfslockfile *lfp)
3883 {
3884 	struct nfslock *new_lop = *new_lopp;
3885 	struct nfslock *lop, *tlop, *ilop;
3886 	struct nfslock *other_lop = *other_lopp;
3887 	int unlock = 0, myfile = 0;
3888 	u_int64_t tmp;
3889 
3890 	/*
3891 	 * Work down the list until the lock is merged.
3892 	 */
3893 	if (new_lop->lo_flags & NFSLCK_UNLOCK)
3894 		unlock = 1;
3895 	if (stp != NULL) {
3896 		ilop = (struct nfslock *)stp;
3897 		lop = LIST_FIRST(&stp->ls_lock);
3898 	} else {
3899 		ilop = (struct nfslock *)lfp;
3900 		lop = LIST_FIRST(&lfp->lf_locallock);
3901 	}
3902 	while (lop != NULL) {
3903 	    /*
3904 	     * Only check locks for this file that aren't before the start of
3905 	     * new lock's range.
3906 	     */
3907 	    if (lop->lo_lfp == lfp) {
3908 	      myfile = 1;
3909 	      if (lop->lo_end >= new_lop->lo_first) {
3910 		if (new_lop->lo_end < lop->lo_first) {
3911 			/*
3912 			 * If the new lock ends before the start of the
3913 			 * current lock's range, no merge, just insert
3914 			 * the new lock.
3915 			 */
3916 			break;
3917 		}
3918 		if (new_lop->lo_flags == lop->lo_flags ||
3919 		    (new_lop->lo_first <= lop->lo_first &&
3920 		     new_lop->lo_end >= lop->lo_end)) {
3921 			/*
3922 			 * This lock can be absorbed by the new lock/unlock.
3923 			 * This happens when it covers the entire range
3924 			 * of the old lock or is contiguous
3925 			 * with the old lock and is of the same type or an
3926 			 * unlock.
3927 			 */
3928 			if (lop->lo_first < new_lop->lo_first)
3929 				new_lop->lo_first = lop->lo_first;
3930 			if (lop->lo_end > new_lop->lo_end)
3931 				new_lop->lo_end = lop->lo_end;
3932 			tlop = lop;
3933 			lop = LIST_NEXT(lop, lo_lckowner);
3934 			nfsrv_freenfslock(tlop);
3935 			continue;
3936 		}
3937 
3938 		/*
3939 		 * All these cases are for contiguous locks that are not the
3940 		 * same type, so they can't be merged.
3941 		 */
3942 		if (new_lop->lo_first <= lop->lo_first) {
3943 			/*
3944 			 * This case is where the new lock overlaps with the
3945 			 * first part of the old lock. Move the start of the
3946 			 * old lock to just past the end of the new lock. The
3947 			 * new lock will be inserted in front of the old, since
3948 			 * ilop hasn't been updated. (We are done now.)
3949 			 */
3950 			lop->lo_first = new_lop->lo_end;
3951 			break;
3952 		}
3953 		if (new_lop->lo_end >= lop->lo_end) {
3954 			/*
3955 			 * This case is where the new lock overlaps with the
3956 			 * end of the old lock's range. Move the old lock's
3957 			 * end to just before the new lock's first and insert
3958 			 * the new lock after the old lock.
3959 			 * Might not be done yet, since the new lock could
3960 			 * overlap further locks with higher ranges.
3961 			 */
3962 			lop->lo_end = new_lop->lo_first;
3963 			ilop = lop;
3964 			lop = LIST_NEXT(lop, lo_lckowner);
3965 			continue;
3966 		}
3967 		/*
3968 		 * The final case is where the new lock's range is in the
3969 		 * middle of the current lock's and splits the current lock
3970 		 * up. Use *other_lopp to handle the second part of the
3971 		 * split old lock range. (We are done now.)
3972 		 * For unlock, we use new_lop as other_lop and tmp, since
3973 		 * other_lop and new_lop are the same for this case.
3974 		 * We noted the unlock case above, so we don't need
3975 		 * new_lop->lo_flags any longer.
3976 		 */
3977 		tmp = new_lop->lo_first;
3978 		if (other_lop == NULL) {
3979 			if (!unlock)
3980 				panic("nfsd srv update unlock");
3981 			other_lop = new_lop;
3982 			*new_lopp = NULL;
3983 		}
3984 		other_lop->lo_first = new_lop->lo_end;
3985 		other_lop->lo_end = lop->lo_end;
3986 		other_lop->lo_flags = lop->lo_flags;
3987 		other_lop->lo_stp = stp;
3988 		other_lop->lo_lfp = lfp;
3989 		lop->lo_end = tmp;
3990 		nfsrv_insertlock(other_lop, lop, stp, lfp);
3991 		*other_lopp = NULL;
3992 		ilop = lop;
3993 		break;
3994 	      }
3995 	    }
3996 	    ilop = lop;
3997 	    lop = LIST_NEXT(lop, lo_lckowner);
3998 	    if (myfile && (lop == NULL || lop->lo_lfp != lfp))
3999 		break;
4000 	}
4001 
4002 	/*
4003 	 * Insert the new lock in the list at the appropriate place.
4004 	 */
4005 	if (!unlock) {
4006 		nfsrv_insertlock(new_lop, ilop, stp, lfp);
4007 		*new_lopp = NULL;
4008 	}
4009 }
4010 
4011 /*
4012  * This function handles sequencing of locks, etc.
4013  * It returns an error that indicates what the caller should do.
4014  */
4015 static int
4016 nfsrv_checkseqid(struct nfsrv_descript *nd, u_int32_t seqid,
4017     struct nfsstate *stp, struct nfsrvcache *op)
4018 {
4019 	int error = 0;
4020 
4021 	if ((nd->nd_flag & ND_NFSV41) != 0)
4022 		/* NFSv4.1 ignores the open_seqid and lock_seqid. */
4023 		goto out;
4024 	if (op != nd->nd_rp)
4025 		panic("nfsrvstate checkseqid");
4026 	if (!(op->rc_flag & RC_INPROG))
4027 		panic("nfsrvstate not inprog");
4028 	if (stp->ls_op && stp->ls_op->rc_refcnt <= 0) {
4029 		printf("refcnt=%d\n", stp->ls_op->rc_refcnt);
4030 		panic("nfsrvstate op refcnt");
4031 	}
4032 
4033 	/* If ND_ERELOOKUP is set, the seqid has already been handled. */
4034 	if ((nd->nd_flag & ND_ERELOOKUP) != 0)
4035 		goto out;
4036 
4037 	if ((stp->ls_seq + 1) == seqid) {
4038 		if (stp->ls_op)
4039 			nfsrvd_derefcache(stp->ls_op);
4040 		stp->ls_op = op;
4041 		nfsrvd_refcache(op);
4042 		stp->ls_seq = seqid;
4043 		goto out;
4044 	} else if (stp->ls_seq == seqid && stp->ls_op &&
4045 		op->rc_xid == stp->ls_op->rc_xid &&
4046 		op->rc_refcnt == 0 &&
4047 		op->rc_reqlen == stp->ls_op->rc_reqlen &&
4048 		op->rc_cksum == stp->ls_op->rc_cksum) {
4049 		if (stp->ls_op->rc_flag & RC_INPROG) {
4050 			error = NFSERR_DONTREPLY;
4051 			goto out;
4052 		}
4053 		nd->nd_rp = stp->ls_op;
4054 		nd->nd_rp->rc_flag |= RC_INPROG;
4055 		nfsrvd_delcache(op);
4056 		error = NFSERR_REPLYFROMCACHE;
4057 		goto out;
4058 	}
4059 	error = NFSERR_BADSEQID;
4060 
4061 out:
4062 	NFSEXITCODE2(error, nd);
4063 	return (error);
4064 }
4065 
4066 /*
4067  * Get the client ip address for callbacks. If the strings can't be parsed,
4068  * just set lc_program to 0 to indicate no callbacks are possible.
4069  * (For cases where the address can't be parsed or is 0.0.0.0.0.0, set
4070  *  the address to the client's transport address. This won't be used
4071  *  for callbacks, but can be printed out by nfsstats for info.)
4072  * Return error if the xdr can't be parsed, 0 otherwise.
4073  */
4074 int
4075 nfsrv_getclientipaddr(struct nfsrv_descript *nd, struct nfsclient *clp)
4076 {
4077 	u_int32_t *tl;
4078 	u_char *cp, *cp2;
4079 	int i, j, maxalen = 0, minalen = 0;
4080 	sa_family_t af;
4081 #ifdef INET
4082 	struct sockaddr_in *rin = NULL, *sin;
4083 #endif
4084 #ifdef INET6
4085 	struct sockaddr_in6 *rin6 = NULL, *sin6;
4086 #endif
4087 	u_char *addr;
4088 	int error = 0, cantparse = 0;
4089 	union {
4090 		in_addr_t ival;
4091 		u_char cval[4];
4092 	} ip;
4093 	union {
4094 		in_port_t sval;
4095 		u_char cval[2];
4096 	} port;
4097 
4098 	/* 8 is the maximum length of the port# string. */
4099 	addr = malloc(INET6_ADDRSTRLEN + 8, M_TEMP, M_WAITOK);
4100 	clp->lc_req.nr_client = NULL;
4101 	clp->lc_req.nr_lock = 0;
4102 	af = AF_UNSPEC;
4103 	NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
4104 	i = fxdr_unsigned(int, *tl);
4105 	if (i >= 3 && i <= 4) {
4106 		error = nfsrv_mtostr(nd, addr, i);
4107 		if (error)
4108 			goto nfsmout;
4109 #ifdef INET
4110 		if (!strcmp(addr, "tcp")) {
4111 			clp->lc_flags |= LCL_TCPCALLBACK;
4112 			clp->lc_req.nr_sotype = SOCK_STREAM;
4113 			clp->lc_req.nr_soproto = IPPROTO_TCP;
4114 			af = AF_INET;
4115 		} else if (!strcmp(addr, "udp")) {
4116 			clp->lc_req.nr_sotype = SOCK_DGRAM;
4117 			clp->lc_req.nr_soproto = IPPROTO_UDP;
4118 			af = AF_INET;
4119 		}
4120 #endif
4121 #ifdef INET6
4122 		if (af == AF_UNSPEC) {
4123 			if (!strcmp(addr, "tcp6")) {
4124 				clp->lc_flags |= LCL_TCPCALLBACK;
4125 				clp->lc_req.nr_sotype = SOCK_STREAM;
4126 				clp->lc_req.nr_soproto = IPPROTO_TCP;
4127 				af = AF_INET6;
4128 			} else if (!strcmp(addr, "udp6")) {
4129 				clp->lc_req.nr_sotype = SOCK_DGRAM;
4130 				clp->lc_req.nr_soproto = IPPROTO_UDP;
4131 				af = AF_INET6;
4132 			}
4133 		}
4134 #endif
4135 		if (af == AF_UNSPEC) {
4136 			cantparse = 1;
4137 		}
4138 	} else {
4139 		cantparse = 1;
4140 		if (i > 0) {
4141 			error = nfsm_advance(nd, NFSM_RNDUP(i), -1);
4142 			if (error)
4143 				goto nfsmout;
4144 		}
4145 	}
4146 	/*
4147 	 * The caller has allocated clp->lc_req.nr_nam to be large enough
4148 	 * for either AF_INET or AF_INET6 and zeroed out the contents.
4149 	 * maxalen is set to the maximum length of the host IP address string
4150 	 * plus 8 for the maximum length of the port#.
4151 	 * minalen is set to the minimum length of the host IP address string
4152 	 * plus 4 for the minimum length of the port#.
4153 	 * These lengths do not include NULL termination,
4154 	 * so INET[6]_ADDRSTRLEN - 1 is used in the calculations.
4155 	 */
4156 	switch (af) {
4157 #ifdef INET
4158 	case AF_INET:
4159 		rin = (struct sockaddr_in *)clp->lc_req.nr_nam;
4160 		rin->sin_family = AF_INET;
4161 		rin->sin_len = sizeof(struct sockaddr_in);
4162 		maxalen = INET_ADDRSTRLEN - 1 + 8;
4163 		minalen = 7 + 4;
4164 		break;
4165 #endif
4166 #ifdef INET6
4167 	case AF_INET6:
4168 		rin6 = (struct sockaddr_in6 *)clp->lc_req.nr_nam;
4169 		rin6->sin6_family = AF_INET6;
4170 		rin6->sin6_len = sizeof(struct sockaddr_in6);
4171 		maxalen = INET6_ADDRSTRLEN - 1 + 8;
4172 		minalen = 3 + 4;
4173 		break;
4174 #endif
4175 	}
4176 	NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
4177 	i = fxdr_unsigned(int, *tl);
4178 	if (i < 0) {
4179 		error = NFSERR_BADXDR;
4180 		goto nfsmout;
4181 	} else if (i == 0) {
4182 		cantparse = 1;
4183 	} else if (!cantparse && i <= maxalen && i >= minalen) {
4184 		error = nfsrv_mtostr(nd, addr, i);
4185 		if (error)
4186 			goto nfsmout;
4187 
4188 		/*
4189 		 * Parse out the address fields. We expect 6 decimal numbers
4190 		 * separated by '.'s for AF_INET and two decimal numbers
4191 		 * preceeded by '.'s for AF_INET6.
4192 		 */
4193 		cp = NULL;
4194 		switch (af) {
4195 #ifdef INET6
4196 		/*
4197 		 * For AF_INET6, first parse the host address.
4198 		 */
4199 		case AF_INET6:
4200 			cp = strchr(addr, '.');
4201 			if (cp != NULL) {
4202 				*cp++ = '\0';
4203 				if (inet_pton(af, addr, &rin6->sin6_addr) == 1)
4204 					i = 4;
4205 				else {
4206 					cp = NULL;
4207 					cantparse = 1;
4208 				}
4209 			}
4210 			break;
4211 #endif
4212 #ifdef INET
4213 		case AF_INET:
4214 			cp = addr;
4215 			i = 0;
4216 			break;
4217 #endif
4218 		}
4219 		while (cp != NULL && *cp && i < 6) {
4220 			cp2 = cp;
4221 			while (*cp2 && *cp2 != '.')
4222 				cp2++;
4223 			if (*cp2)
4224 				*cp2++ = '\0';
4225 			else if (i != 5) {
4226 				cantparse = 1;
4227 				break;
4228 			}
4229 			j = nfsrv_getipnumber(cp);
4230 			if (j >= 0) {
4231 				if (i < 4)
4232 					ip.cval[3 - i] = j;
4233 				else
4234 					port.cval[5 - i] = j;
4235 			} else {
4236 				cantparse = 1;
4237 				break;
4238 			}
4239 			cp = cp2;
4240 			i++;
4241 		}
4242 		if (!cantparse) {
4243 			/*
4244 			 * The host address INADDR_ANY is (mis)used to indicate
4245 			 * "there is no valid callback address".
4246 			 */
4247 			switch (af) {
4248 #ifdef INET6
4249 			case AF_INET6:
4250 				if (!IN6_ARE_ADDR_EQUAL(&rin6->sin6_addr,
4251 				    &in6addr_any))
4252 					rin6->sin6_port = htons(port.sval);
4253 				else
4254 					cantparse = 1;
4255 				break;
4256 #endif
4257 #ifdef INET
4258 			case AF_INET:
4259 				if (ip.ival != INADDR_ANY) {
4260 					rin->sin_addr.s_addr = htonl(ip.ival);
4261 					rin->sin_port = htons(port.sval);
4262 				} else {
4263 					cantparse = 1;
4264 				}
4265 				break;
4266 #endif
4267 			}
4268 		}
4269 	} else {
4270 		cantparse = 1;
4271 		if (i > 0) {
4272 			error = nfsm_advance(nd, NFSM_RNDUP(i), -1);
4273 			if (error)
4274 				goto nfsmout;
4275 		}
4276 	}
4277 	if (cantparse) {
4278 		switch (nd->nd_nam->sa_family) {
4279 #ifdef INET
4280 		case AF_INET:
4281 			sin = (struct sockaddr_in *)nd->nd_nam;
4282 			rin = (struct sockaddr_in *)clp->lc_req.nr_nam;
4283 			rin->sin_family = AF_INET;
4284 			rin->sin_len = sizeof(struct sockaddr_in);
4285 			rin->sin_addr.s_addr = sin->sin_addr.s_addr;
4286 			rin->sin_port = 0x0;
4287 			break;
4288 #endif
4289 #ifdef INET6
4290 		case AF_INET6:
4291 			sin6 = (struct sockaddr_in6 *)nd->nd_nam;
4292 			rin6 = (struct sockaddr_in6 *)clp->lc_req.nr_nam;
4293 			rin6->sin6_family = AF_INET6;
4294 			rin6->sin6_len = sizeof(struct sockaddr_in6);
4295 			rin6->sin6_addr = sin6->sin6_addr;
4296 			rin6->sin6_port = 0x0;
4297 			break;
4298 #endif
4299 		}
4300 		clp->lc_program = 0;
4301 	}
4302 nfsmout:
4303 	free(addr, M_TEMP);
4304 	NFSEXITCODE2(error, nd);
4305 	return (error);
4306 }
4307 
4308 /*
4309  * Turn a string of up to three decimal digits into a number. Return -1 upon
4310  * error.
4311  */
4312 static int
4313 nfsrv_getipnumber(u_char *cp)
4314 {
4315 	int i = 0, j = 0;
4316 
4317 	while (*cp) {
4318 		if (j > 2 || *cp < '0' || *cp > '9')
4319 			return (-1);
4320 		i *= 10;
4321 		i += (*cp - '0');
4322 		cp++;
4323 		j++;
4324 	}
4325 	if (i < 256)
4326 		return (i);
4327 	return (-1);
4328 }
4329 
4330 /*
4331  * This function checks for restart conditions.
4332  */
4333 static int
4334 nfsrv_checkrestart(nfsquad_t clientid, u_int32_t flags,
4335     nfsv4stateid_t *stateidp, int specialid)
4336 {
4337 	int ret = 0;
4338 
4339 	/*
4340 	 * First check for a server restart. Open, LockT, ReleaseLockOwner
4341 	 * and DelegPurge have a clientid, the rest a stateid.
4342 	 */
4343 	if (flags &
4344 	    (NFSLCK_OPEN | NFSLCK_TEST | NFSLCK_RELEASE | NFSLCK_DELEGPURGE)) {
4345 		if (clientid.lval[0] != NFSD_VNET(nfsrvboottime)) {
4346 			ret = NFSERR_STALECLIENTID;
4347 			goto out;
4348 		}
4349 	} else if (stateidp->other[0] != NFSD_VNET(nfsrvboottime) &&
4350 		specialid == 0) {
4351 		ret = NFSERR_STALESTATEID;
4352 		goto out;
4353 	}
4354 
4355 	/*
4356 	 * Read, Write, Setattr and LockT can return NFSERR_GRACE and do
4357 	 * not use a lock/open owner seqid#, so the check can be done now.
4358 	 * (The others will be checked, as required, later.)
4359 	 */
4360 	if (!(flags & (NFSLCK_CHECK | NFSLCK_TEST)))
4361 		goto out;
4362 
4363 	NFSLOCKSTATE();
4364 	ret = nfsrv_checkgrace(NULL, NULL, flags);
4365 	NFSUNLOCKSTATE();
4366 
4367 out:
4368 	NFSEXITCODE(ret);
4369 	return (ret);
4370 }
4371 
4372 /*
4373  * Check for grace.
4374  */
4375 static int
4376 nfsrv_checkgrace(struct nfsrv_descript *nd, struct nfsclient *clp,
4377     u_int32_t flags)
4378 {
4379 	int error = 0, notreclaimed;
4380 	struct nfsrv_stable *sp;
4381 
4382 	if ((NFSD_VNET(nfsrv_stablefirst).nsf_flags & (NFSNSF_UPDATEDONE |
4383 	     NFSNSF_GRACEOVER)) == 0) {
4384 		/*
4385 		 * First, check to see if all of the clients have done a
4386 		 * ReclaimComplete.  If so, grace can end now.
4387 		 */
4388 		notreclaimed = 0;
4389 		if (!NFSD_VNET(nfsd_disable_grace)) {
4390 			LIST_FOREACH(sp, &NFSD_VNET(nfsrv_stablefirst).nsf_head,
4391 			    nst_list) {
4392 				if ((sp->nst_flag & NFSNST_RECLAIMED) == 0) {
4393 					notreclaimed = 1;
4394 					break;
4395 				}
4396 			}
4397 		}
4398 		if (notreclaimed == 0)
4399 			NFSD_VNET(nfsrv_stablefirst).nsf_flags |=
4400 			    (NFSNSF_GRACEOVER | NFSNSF_NEEDLOCK);
4401 	}
4402 
4403 	if ((NFSD_VNET(nfsrv_stablefirst).nsf_flags & NFSNSF_GRACEOVER) != 0) {
4404 		if (flags & NFSLCK_RECLAIM) {
4405 			error = NFSERR_NOGRACE;
4406 			goto out;
4407 		}
4408 	} else {
4409 		if (!(flags & NFSLCK_RECLAIM)) {
4410 			error = NFSERR_GRACE;
4411 			goto out;
4412 		}
4413 		if (nd != NULL && clp != NULL &&
4414 		    (nd->nd_flag & ND_NFSV41) != 0 &&
4415 		    (clp->lc_flags & LCL_RECLAIMCOMPLETE) != 0) {
4416 			error = NFSERR_NOGRACE;
4417 			goto out;
4418 		}
4419 
4420 		/*
4421 		 * If grace is almost over and we are still getting Reclaims,
4422 		 * extend grace a bit.
4423 		 */
4424 		if ((NFSD_MONOSEC + NFSRV_LEASEDELTA) >
4425 		    NFSD_VNET(nfsrv_stablefirst).nsf_eograce)
4426 			NFSD_VNET(nfsrv_stablefirst).nsf_eograce =
4427 				NFSD_MONOSEC + NFSRV_LEASEDELTA;
4428 	}
4429 
4430 out:
4431 	NFSEXITCODE(error);
4432 	return (error);
4433 }
4434 
4435 /*
4436  * Do a server callback.
4437  * The "trunc" argument is slightly overloaded and refers to different
4438  * boolean arguments for CBRECALL and CBLAYOUTRECALL.
4439  */
4440 static int
4441 nfsrv_docallback(struct nfsclient *clp, int procnum, nfsv4stateid_t *stateidp,
4442     int trunc, fhandle_t *fhp, struct nfsvattr *nap, nfsattrbit_t *attrbitp,
4443     int laytype, NFSPROC_T *p)
4444 {
4445 	struct mbuf *m;
4446 	u_int32_t *tl;
4447 	struct nfsrv_descript *nd;
4448 	struct ucred *cred;
4449 	int error = 0, slotpos;
4450 	u_int32_t callback;
4451 	struct nfsdsession *sep = NULL;
4452 	uint64_t tval;
4453 	bool dotls;
4454 
4455 	nd = malloc(sizeof(*nd), M_TEMP, M_WAITOK | M_ZERO);
4456 	cred = newnfs_getcred();
4457 	NFSLOCKSTATE();	/* mostly for lc_cbref++ */
4458 	if (clp->lc_flags & LCL_NEEDSCONFIRM) {
4459 		NFSUNLOCKSTATE();
4460 		panic("docallb");
4461 	}
4462 	clp->lc_cbref++;
4463 
4464 	/*
4465 	 * Fill the callback program# and version into the request
4466 	 * structure for newnfs_connect() to use.
4467 	 */
4468 	clp->lc_req.nr_prog = clp->lc_program;
4469 #ifdef notnow
4470 	if ((clp->lc_flags & LCL_NFSV41) != 0)
4471 		clp->lc_req.nr_vers = NFSV41_CBVERS;
4472 	else
4473 #endif
4474 		clp->lc_req.nr_vers = NFSV4_CBVERS;
4475 
4476 	/*
4477 	 * First, fill in some of the fields of nd and cr.
4478 	 */
4479 	nd->nd_flag = ND_NFSV4;
4480 	if (clp->lc_flags & LCL_GSS)
4481 		nd->nd_flag |= ND_KERBV;
4482 	if ((clp->lc_flags & LCL_NFSV41) != 0)
4483 		nd->nd_flag |= ND_NFSV41;
4484 	if ((clp->lc_flags & LCL_NFSV42) != 0)
4485 		nd->nd_flag |= ND_NFSV42;
4486 	nd->nd_repstat = 0;
4487 	cred->cr_uid = clp->lc_uid;
4488 	cred->cr_gid = clp->lc_gid;
4489 	callback = clp->lc_callback;
4490 	NFSUNLOCKSTATE();
4491 	cred->cr_ngroups = 1;
4492 
4493 	/*
4494 	 * Get the first mbuf for the request.
4495 	 */
4496 	MGET(m, M_WAITOK, MT_DATA);
4497 	m->m_len = 0;
4498 	nd->nd_mreq = nd->nd_mb = m;
4499 	nd->nd_bpos = mtod(m, caddr_t);
4500 
4501 	/*
4502 	 * and build the callback request.
4503 	 */
4504 	if (procnum == NFSV4OP_CBGETATTR) {
4505 		nd->nd_procnum = NFSV4PROC_CBCOMPOUND;
4506 		error = nfsrv_cbcallargs(nd, clp, callback, NFSV4OP_CBGETATTR,
4507 		    "CB Getattr", &sep, &slotpos);
4508 		if (error != 0) {
4509 			m_freem(nd->nd_mreq);
4510 			goto errout;
4511 		}
4512 		(void)nfsm_fhtom(NULL, nd, (u_int8_t *)fhp, NFSX_MYFH, 0);
4513 		(void)nfsrv_putattrbit(nd, attrbitp);
4514 	} else if (procnum == NFSV4OP_CBRECALL) {
4515 		nd->nd_procnum = NFSV4PROC_CBCOMPOUND;
4516 		error = nfsrv_cbcallargs(nd, clp, callback, NFSV4OP_CBRECALL,
4517 		    "CB Recall", &sep, &slotpos);
4518 		if (error != 0) {
4519 			m_freem(nd->nd_mreq);
4520 			goto errout;
4521 		}
4522 		NFSM_BUILD(tl, u_int32_t *, NFSX_UNSIGNED + NFSX_STATEID);
4523 		*tl++ = txdr_unsigned(stateidp->seqid);
4524 		NFSBCOPY((caddr_t)stateidp->other, (caddr_t)tl,
4525 		    NFSX_STATEIDOTHER);
4526 		tl += (NFSX_STATEIDOTHER / NFSX_UNSIGNED);
4527 		if (trunc)
4528 			*tl = newnfs_true;
4529 		else
4530 			*tl = newnfs_false;
4531 		(void)nfsm_fhtom(NULL, nd, (u_int8_t *)fhp, NFSX_MYFH, 0);
4532 	} else if (procnum == NFSV4OP_CBLAYOUTRECALL) {
4533 		NFSD_DEBUG(4, "docallback layout recall\n");
4534 		nd->nd_procnum = NFSV4PROC_CBCOMPOUND;
4535 		error = nfsrv_cbcallargs(nd, clp, callback,
4536 		    NFSV4OP_CBLAYOUTRECALL, "CB Reclayout", &sep, &slotpos);
4537 		NFSD_DEBUG(4, "aft cbcallargs=%d\n", error);
4538 		if (error != 0) {
4539 			m_freem(nd->nd_mreq);
4540 			goto errout;
4541 		}
4542 		NFSM_BUILD(tl, u_int32_t *, 4 * NFSX_UNSIGNED);
4543 		*tl++ = txdr_unsigned(laytype);
4544 		*tl++ = txdr_unsigned(NFSLAYOUTIOMODE_ANY);
4545 		if (trunc)
4546 			*tl++ = newnfs_true;
4547 		else
4548 			*tl++ = newnfs_false;
4549 		*tl = txdr_unsigned(NFSV4LAYOUTRET_FILE);
4550 		(void)nfsm_fhtom(NULL, nd, (uint8_t *)fhp, NFSX_MYFH, 0);
4551 		NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_HYPER + NFSX_STATEID);
4552 		tval = 0;
4553 		txdr_hyper(tval, tl); tl += 2;
4554 		tval = UINT64_MAX;
4555 		txdr_hyper(tval, tl); tl += 2;
4556 		*tl++ = txdr_unsigned(stateidp->seqid);
4557 		NFSBCOPY(stateidp->other, tl, NFSX_STATEIDOTHER);
4558 		tl += (NFSX_STATEIDOTHER / NFSX_UNSIGNED);
4559 		NFSD_DEBUG(4, "aft args\n");
4560 	} else if (procnum == NFSV4PROC_CBNULL) {
4561 		nd->nd_procnum = NFSV4PROC_CBNULL;
4562 		if ((clp->lc_flags & LCL_NFSV41) != 0) {
4563 			error = nfsv4_getcbsession(clp, &sep);
4564 			if (error != 0) {
4565 				m_freem(nd->nd_mreq);
4566 				goto errout;
4567 			}
4568 		}
4569 	} else {
4570 		error = NFSERR_SERVERFAULT;
4571 		m_freem(nd->nd_mreq);
4572 		goto errout;
4573 	}
4574 
4575 	/*
4576 	 * Call newnfs_connect(), as required, and then newnfs_request().
4577 	 */
4578 	dotls = false;
4579 	if ((clp->lc_flags & LCL_TLSCB) != 0)
4580 		dotls = true;
4581 	(void) newnfs_sndlock(&clp->lc_req.nr_lock);
4582 	if (clp->lc_req.nr_client == NULL) {
4583 		if ((clp->lc_flags & LCL_NFSV41) != 0) {
4584 			error = ECONNREFUSED;
4585 			if (procnum != NFSV4PROC_CBNULL)
4586 				nfsv4_freeslot(&sep->sess_cbsess, slotpos,
4587 				    true);
4588 			nfsrv_freesession(NULL, sep, NULL, false, NULL);
4589 		} else if (nd->nd_procnum == NFSV4PROC_CBNULL)
4590 			error = newnfs_connect(NULL, &clp->lc_req, cred,
4591 			    NULL, 1, dotls, &clp->lc_req.nr_client);
4592 		else
4593 			error = newnfs_connect(NULL, &clp->lc_req, cred,
4594 			    NULL, 3, dotls, &clp->lc_req.nr_client);
4595 	}
4596 	newnfs_sndunlock(&clp->lc_req.nr_lock);
4597 	NFSD_DEBUG(4, "aft sndunlock=%d\n", error);
4598 	if (!error) {
4599 		if ((nd->nd_flag & ND_NFSV41) != 0) {
4600 			KASSERT(sep != NULL, ("sep NULL"));
4601 			if (sep->sess_cbsess.nfsess_xprt != NULL)
4602 				error = newnfs_request(nd, NULL, clp,
4603 				    &clp->lc_req, NULL, NULL, cred,
4604 				    clp->lc_program, clp->lc_req.nr_vers, NULL,
4605 				    1, NULL, &sep->sess_cbsess);
4606 			else {
4607 				/*
4608 				 * This should probably never occur, but if a
4609 				 * client somehow does an RPC without a
4610 				 * SequenceID Op that causes a callback just
4611 				 * after the nfsd threads have been terminated
4612 				 * and restarted we could conceivably get here
4613 				 * without a backchannel xprt.
4614 				 */
4615 				printf("nfsrv_docallback: no xprt\n");
4616 				error = ECONNREFUSED;
4617 			}
4618 			NFSD_DEBUG(4, "aft newnfs_request=%d\n", error);
4619 			if (error != 0 && procnum != NFSV4PROC_CBNULL) {
4620 				/*
4621 				 * It is likely that the callback was never
4622 				 * processed by the client and, as such,
4623 				 * the sequence# for the session slot needs
4624 				 * to be backed up by one to avoid a
4625 				 * NFSERR_SEQMISORDERED error reply.
4626 				 * For the unlikely case where the callback
4627 				 * was processed by the client, this will
4628 				 * make the next callback on the slot
4629 				 * appear to be a retry.
4630 				 * Since callbacks never specify that the
4631 				 * reply be cached, this "apparent retry"
4632 				 * should not be a problem.
4633 				 */
4634 				nfsv4_freeslot(&sep->sess_cbsess, slotpos,
4635 				    true);
4636 			}
4637 			nfsrv_freesession(NULL, sep, NULL, false, NULL);
4638 		} else
4639 			error = newnfs_request(nd, NULL, clp, &clp->lc_req,
4640 			    NULL, NULL, cred, clp->lc_program,
4641 			    clp->lc_req.nr_vers, NULL, 1, NULL, NULL);
4642 	}
4643 errout:
4644 	NFSFREECRED(cred);
4645 
4646 	/*
4647 	 * If error is set here, the Callback path isn't working
4648 	 * properly, so twiddle the appropriate LCL_ flags.
4649 	 * (nd_repstat != 0 indicates the Callback path is working,
4650 	 *  but the callback failed on the client.)
4651 	 */
4652 	if (error) {
4653 		/*
4654 		 * Mark the callback pathway down, which disabled issuing
4655 		 * of delegations and gets Renew to return NFSERR_CBPATHDOWN.
4656 		 */
4657 		NFSLOCKSTATE();
4658 		clp->lc_flags |= LCL_CBDOWN;
4659 		NFSUNLOCKSTATE();
4660 	} else {
4661 		/*
4662 		 * Callback worked. If the callback path was down, disable
4663 		 * callbacks, so no more delegations will be issued. (This
4664 		 * is done on the assumption that the callback pathway is
4665 		 * flakey.)
4666 		 */
4667 		NFSLOCKSTATE();
4668 		if (clp->lc_flags & LCL_CBDOWN)
4669 			clp->lc_flags &= ~(LCL_CBDOWN | LCL_CALLBACKSON);
4670 		NFSUNLOCKSTATE();
4671 		if (nd->nd_repstat) {
4672 			error = nd->nd_repstat;
4673 			NFSD_DEBUG(1, "nfsrv_docallback op=%d err=%d\n",
4674 			    procnum, error);
4675 		} else if (error == 0 && procnum == NFSV4OP_CBGETATTR)
4676 			error = nfsv4_loadattr(nd, NULL, nap, NULL, NULL, 0,
4677 			    NULL, NULL, NULL, NULL, NULL, 0, NULL, NULL, NULL,
4678 			    p, NULL);
4679 		m_freem(nd->nd_mrep);
4680 	}
4681 	NFSLOCKSTATE();
4682 	clp->lc_cbref--;
4683 	if ((clp->lc_flags & LCL_WAKEUPWANTED) && clp->lc_cbref == 0) {
4684 		clp->lc_flags &= ~LCL_WAKEUPWANTED;
4685 		wakeup(clp);
4686 	}
4687 	NFSUNLOCKSTATE();
4688 
4689 	free(nd, M_TEMP);
4690 	NFSEXITCODE(error);
4691 	return (error);
4692 }
4693 
4694 /*
4695  * Set up the compound RPC for the callback.
4696  */
4697 static int
4698 nfsrv_cbcallargs(struct nfsrv_descript *nd, struct nfsclient *clp,
4699     uint32_t callback, int op, const char *optag, struct nfsdsession **sepp,
4700     int *slotposp)
4701 {
4702 	uint32_t *tl;
4703 	int error, len;
4704 
4705 	len = strlen(optag);
4706 	(void)nfsm_strtom(nd, optag, len);
4707 	NFSM_BUILD(tl, uint32_t *, 4 * NFSX_UNSIGNED);
4708 	if ((nd->nd_flag & ND_NFSV41) != 0) {
4709 		if ((nd->nd_flag & ND_NFSV42) != 0)
4710 			*tl++ = txdr_unsigned(NFSV42_MINORVERSION);
4711 		else
4712 			*tl++ = txdr_unsigned(NFSV41_MINORVERSION);
4713 		*tl++ = txdr_unsigned(callback);
4714 		*tl++ = txdr_unsigned(2);
4715 		*tl = txdr_unsigned(NFSV4OP_CBSEQUENCE);
4716 		error = nfsv4_setcbsequence(nd, clp, 1, sepp, slotposp);
4717 		if (error != 0)
4718 			return (error);
4719 		NFSM_BUILD(tl, u_int32_t *, NFSX_UNSIGNED);
4720 		*tl = txdr_unsigned(op);
4721 	} else {
4722 		*tl++ = txdr_unsigned(NFSV4_MINORVERSION);
4723 		*tl++ = txdr_unsigned(callback);
4724 		*tl++ = txdr_unsigned(1);
4725 		*tl = txdr_unsigned(op);
4726 	}
4727 	return (0);
4728 }
4729 
4730 /*
4731  * Return the next index# for a clientid. Mostly just increment and return
4732  * the next one, but... if the 32bit unsigned does actually wrap around,
4733  * it should be rebooted.
4734  * At an average rate of one new client per second, it will wrap around in
4735  * approximately 136 years. (I think the server will have been shut
4736  * down or rebooted before then.)
4737  */
4738 static u_int32_t
4739 nfsrv_nextclientindex(void)
4740 {
4741 	static u_int32_t client_index = 0;
4742 
4743 	client_index++;
4744 	if (client_index != 0)
4745 		return (client_index);
4746 
4747 	printf("%s: out of clientids\n", __func__);
4748 	return (client_index);
4749 }
4750 
4751 /*
4752  * Return the next index# for a stateid. Mostly just increment and return
4753  * the next one, but... if the 32bit unsigned does actually wrap around
4754  * (will a BSD server stay up that long?), find
4755  * new start and end values.
4756  */
4757 static u_int32_t
4758 nfsrv_nextstateindex(struct nfsclient *clp)
4759 {
4760 	struct nfsstate *stp;
4761 	int i;
4762 	u_int32_t canuse, min_index, max_index;
4763 
4764 	if (!(clp->lc_flags & LCL_INDEXNOTOK)) {
4765 		clp->lc_stateindex++;
4766 		if (clp->lc_stateindex != clp->lc_statemaxindex)
4767 			return (clp->lc_stateindex);
4768 	}
4769 
4770 	/*
4771 	 * Yuck, we've hit the end.
4772 	 * Look for a new min and max.
4773 	 */
4774 	min_index = 0;
4775 	max_index = 0xffffffff;
4776 	for (i = 0; i < nfsrv_statehashsize; i++) {
4777 	    LIST_FOREACH(stp, &clp->lc_stateid[i], ls_hash) {
4778 		if (stp->ls_stateid.other[2] > 0x80000000) {
4779 		    if (stp->ls_stateid.other[2] < max_index)
4780 			max_index = stp->ls_stateid.other[2];
4781 		} else {
4782 		    if (stp->ls_stateid.other[2] > min_index)
4783 			min_index = stp->ls_stateid.other[2];
4784 		}
4785 	    }
4786 	}
4787 
4788 	/*
4789 	 * Yikes, highly unlikely, but I'll handle it anyhow.
4790 	 */
4791 	if (min_index == 0x80000000 && max_index == 0x80000001) {
4792 	    canuse = 0;
4793 	    /*
4794 	     * Loop around until we find an unused entry. Return that
4795 	     * and set LCL_INDEXNOTOK, so the search will continue next time.
4796 	     * (This is one of those rare cases where a goto is the
4797 	     *  cleanest way to code the loop.)
4798 	     */
4799 tryagain:
4800 	    for (i = 0; i < nfsrv_statehashsize; i++) {
4801 		LIST_FOREACH(stp, &clp->lc_stateid[i], ls_hash) {
4802 		    if (stp->ls_stateid.other[2] == canuse) {
4803 			canuse++;
4804 			goto tryagain;
4805 		    }
4806 		}
4807 	    }
4808 	    clp->lc_flags |= LCL_INDEXNOTOK;
4809 	    return (canuse);
4810 	}
4811 
4812 	/*
4813 	 * Ok to start again from min + 1.
4814 	 */
4815 	clp->lc_stateindex = min_index + 1;
4816 	clp->lc_statemaxindex = max_index;
4817 	clp->lc_flags &= ~LCL_INDEXNOTOK;
4818 	return (clp->lc_stateindex);
4819 }
4820 
4821 /*
4822  * The following functions handle the stable storage file that deals with
4823  * the edge conditions described in RFC3530 Sec. 8.6.3.
4824  * The file is as follows:
4825  * - a single record at the beginning that has the lease time of the
4826  *   previous server instance (before the last reboot) and the nfsrvboottime
4827  *   values for the previous server boots.
4828  *   These previous boot times are used to ensure that the current
4829  *   nfsrvboottime does not, somehow, get set to a previous one.
4830  *   (This is important so that Stale ClientIDs and StateIDs can
4831  *    be recognized.)
4832  *   The number of previous nfsvrboottime values precedes the list.
4833  * - followed by some number of appended records with:
4834  *   - client id string
4835  *   - flag that indicates it is a record revoking state via lease
4836  *     expiration or similar
4837  *     OR has successfully acquired state.
4838  * These structures vary in length, with the client string at the end, up
4839  * to NFSV4_OPAQUELIMIT in size.
4840  *
4841  * At the end of the grace period, the file is truncated, the first
4842  * record is rewritten with updated information and any acquired state
4843  * records for successful reclaims of state are written.
4844  *
4845  * Subsequent records are appended when the first state is issued to
4846  * a client and when state is revoked for a client.
4847  *
4848  * When reading the file in, state issued records that come later in
4849  * the file override older ones, since the append log is in cronological order.
4850  * If, for some reason, the file can't be read, the grace period is
4851  * immediately terminated and all reclaims get NFSERR_NOGRACE.
4852  */
4853 
4854 /*
4855  * Read in the stable storage file. Called by nfssvc() before the nfsd
4856  * processes start servicing requests.
4857  */
4858 void
4859 nfsrv_setupstable(NFSPROC_T *p)
4860 {
4861 	struct nfsrv_stablefirst *sf = &NFSD_VNET(nfsrv_stablefirst);
4862 	struct nfsrv_stable *sp, *nsp;
4863 	struct nfst_rec *tsp;
4864 	int error, i, tryagain;
4865 	off_t off = 0;
4866 	ssize_t aresid, len;
4867 
4868 	/*
4869 	 * If NFSNSF_UPDATEDONE is set, this is a restart of the nfsds without
4870 	 * a reboot, so state has not been lost.
4871 	 */
4872 	if (sf->nsf_flags & NFSNSF_UPDATEDONE)
4873 		return;
4874 	/*
4875 	 * Set Grace over just until the file reads successfully.
4876 	 */
4877 	NFSD_VNET(nfsrvboottime) = time_second;
4878 	LIST_INIT(&sf->nsf_head);
4879 	sf->nsf_flags = (NFSNSF_GRACEOVER | NFSNSF_NEEDLOCK);
4880 	sf->nsf_eograce = NFSD_MONOSEC + NFSRV_LEASEDELTA;
4881 	if (sf->nsf_fp == NULL)
4882 		return;
4883 	error = NFSD_RDWR(UIO_READ, NFSFPVNODE(sf->nsf_fp),
4884 	    (caddr_t)&sf->nsf_rec, sizeof (struct nfsf_rec), off, UIO_SYSSPACE,
4885 	    0, NFSFPCRED(sf->nsf_fp), &aresid, p);
4886 	if (error || aresid || sf->nsf_numboots == 0 ||
4887 		sf->nsf_numboots > NFSNSF_MAXNUMBOOTS)
4888 		return;
4889 
4890 	/*
4891 	 * Now, read in the boottimes.
4892 	 */
4893 	sf->nsf_bootvals = (time_t *)malloc((sf->nsf_numboots + 1) *
4894 		sizeof(time_t), M_TEMP, M_WAITOK);
4895 	off = sizeof (struct nfsf_rec);
4896 	error = NFSD_RDWR(UIO_READ, NFSFPVNODE(sf->nsf_fp),
4897 	    (caddr_t)sf->nsf_bootvals, sf->nsf_numboots * sizeof (time_t), off,
4898 	    UIO_SYSSPACE, 0, NFSFPCRED(sf->nsf_fp), &aresid, p);
4899 	if (error || aresid) {
4900 		free(sf->nsf_bootvals, M_TEMP);
4901 		sf->nsf_bootvals = NULL;
4902 		return;
4903 	}
4904 
4905 	/*
4906 	 * Make sure this nfsrvboottime is different from all recorded
4907 	 * previous ones.
4908 	 */
4909 	do {
4910 		tryagain = 0;
4911 		for (i = 0; i < sf->nsf_numboots; i++) {
4912 			if (NFSD_VNET(nfsrvboottime) == sf->nsf_bootvals[i]) {
4913 				NFSD_VNET(nfsrvboottime)++;
4914 				tryagain = 1;
4915 				break;
4916 			}
4917 		}
4918 	} while (tryagain);
4919 
4920 	sf->nsf_flags |= NFSNSF_OK;
4921 	off += (sf->nsf_numboots * sizeof (time_t));
4922 
4923 	/*
4924 	 * Read through the file, building a list of records for grace
4925 	 * checking.
4926 	 * Each record is between sizeof (struct nfst_rec) and
4927 	 * sizeof (struct nfst_rec) + NFSV4_OPAQUELIMIT - 1
4928 	 * and is actually sizeof (struct nfst_rec) + nst_len - 1.
4929 	 */
4930 	tsp = (struct nfst_rec *)malloc(sizeof (struct nfst_rec) +
4931 		NFSV4_OPAQUELIMIT - 1, M_TEMP, M_WAITOK);
4932 	do {
4933 	    error = NFSD_RDWR(UIO_READ, NFSFPVNODE(sf->nsf_fp),
4934 	        (caddr_t)tsp, sizeof (struct nfst_rec) + NFSV4_OPAQUELIMIT - 1,
4935 	        off, UIO_SYSSPACE, 0, NFSFPCRED(sf->nsf_fp), &aresid, p);
4936 	    len = (sizeof (struct nfst_rec) + NFSV4_OPAQUELIMIT - 1) - aresid;
4937 	    if (error || (len > 0 && (len < sizeof (struct nfst_rec) ||
4938 		len < (sizeof (struct nfst_rec) + tsp->len - 1)))) {
4939 		/*
4940 		 * Yuck, the file has been corrupted, so just return
4941 		 * after clearing out any restart state, so the grace period
4942 		 * is over.
4943 		 */
4944 		LIST_FOREACH_SAFE(sp, &sf->nsf_head, nst_list, nsp) {
4945 			LIST_REMOVE(sp, nst_list);
4946 			free(sp, M_TEMP);
4947 		}
4948 		free(tsp, M_TEMP);
4949 		sf->nsf_flags &= ~NFSNSF_OK;
4950 		free(sf->nsf_bootvals, M_TEMP);
4951 		sf->nsf_bootvals = NULL;
4952 		return;
4953 	    }
4954 	    if (len > 0) {
4955 		off += sizeof (struct nfst_rec) + tsp->len - 1;
4956 		/*
4957 		 * Search the list for a matching client.
4958 		 */
4959 		LIST_FOREACH(sp, &sf->nsf_head, nst_list) {
4960 			if (tsp->len == sp->nst_len &&
4961 			    !NFSBCMP(tsp->client, sp->nst_client, tsp->len))
4962 				break;
4963 		}
4964 		if (sp == LIST_END(&sf->nsf_head)) {
4965 			sp = (struct nfsrv_stable *)malloc(tsp->len +
4966 				sizeof (struct nfsrv_stable) - 1, M_TEMP,
4967 				M_WAITOK);
4968 			NFSBCOPY((caddr_t)tsp, (caddr_t)&sp->nst_rec,
4969 				sizeof (struct nfst_rec) + tsp->len - 1);
4970 			LIST_INSERT_HEAD(&sf->nsf_head, sp, nst_list);
4971 		} else {
4972 			if (tsp->flag == NFSNST_REVOKE)
4973 				sp->nst_flag |= NFSNST_REVOKE;
4974 			else
4975 				/*
4976 				 * A subsequent timestamp indicates the client
4977 				 * did a setclientid/confirm and any previous
4978 				 * revoke is no longer relevant.
4979 				 */
4980 				sp->nst_flag &= ~NFSNST_REVOKE;
4981 		}
4982 	    }
4983 	} while (len > 0);
4984 	free(tsp, M_TEMP);
4985 	sf->nsf_flags = NFSNSF_OK;
4986 	sf->nsf_eograce = NFSD_MONOSEC + sf->nsf_lease +
4987 		NFSRV_LEASEDELTA;
4988 }
4989 
4990 /*
4991  * Update the stable storage file, now that the grace period is over.
4992  */
4993 void
4994 nfsrv_updatestable(NFSPROC_T *p)
4995 {
4996 	struct nfsrv_stablefirst *sf = &NFSD_VNET(nfsrv_stablefirst);
4997 	struct nfsrv_stable *sp, *nsp;
4998 	int i;
4999 	struct nfsvattr nva;
5000 	vnode_t vp;
5001 #if defined(__FreeBSD_version) && (__FreeBSD_version >= 500000)
5002 	mount_t mp = NULL;
5003 #endif
5004 	int error;
5005 
5006 	if (sf->nsf_fp == NULL || (sf->nsf_flags & NFSNSF_UPDATEDONE))
5007 		return;
5008 	sf->nsf_flags |= NFSNSF_UPDATEDONE;
5009 	/*
5010 	 * Ok, we need to rewrite the stable storage file.
5011 	 * - truncate to 0 length
5012 	 * - write the new first structure
5013 	 * - loop through the data structures, writing out any that
5014 	 *   have timestamps older than the old boot
5015 	 */
5016 	if (sf->nsf_bootvals) {
5017 		sf->nsf_numboots++;
5018 		for (i = sf->nsf_numboots - 2; i >= 0; i--)
5019 			sf->nsf_bootvals[i + 1] = sf->nsf_bootvals[i];
5020 	} else {
5021 		sf->nsf_numboots = 1;
5022 		sf->nsf_bootvals = (time_t *)malloc(sizeof(time_t),
5023 			M_TEMP, M_WAITOK);
5024 	}
5025 	sf->nsf_bootvals[0] = NFSD_VNET(nfsrvboottime);
5026 	sf->nsf_lease = nfsrv_lease;
5027 	NFSVNO_ATTRINIT(&nva);
5028 	NFSVNO_SETATTRVAL(&nva, size, 0);
5029 	vp = NFSFPVNODE(sf->nsf_fp);
5030 	vn_start_write(vp, &mp, V_WAIT);
5031 	if (NFSVOPLOCK(vp, LK_EXCLUSIVE) == 0) {
5032 		error = nfsvno_setattr(vp, &nva, NFSFPCRED(sf->nsf_fp), p,
5033 		    NULL);
5034 		NFSVOPUNLOCK(vp);
5035 	} else
5036 		error = EPERM;
5037 	vn_finished_write(mp);
5038 	if (!error)
5039 	    error = NFSD_RDWR(UIO_WRITE, vp,
5040 		(caddr_t)&sf->nsf_rec, sizeof (struct nfsf_rec), (off_t)0,
5041 		UIO_SYSSPACE, IO_SYNC, NFSFPCRED(sf->nsf_fp), NULL, p);
5042 	if (!error)
5043 	    error = NFSD_RDWR(UIO_WRITE, vp,
5044 		(caddr_t)sf->nsf_bootvals,
5045 		sf->nsf_numboots * sizeof (time_t),
5046 		(off_t)(sizeof (struct nfsf_rec)),
5047 		UIO_SYSSPACE, IO_SYNC, NFSFPCRED(sf->nsf_fp), NULL, p);
5048 	free(sf->nsf_bootvals, M_TEMP);
5049 	sf->nsf_bootvals = NULL;
5050 	if (error) {
5051 		sf->nsf_flags &= ~NFSNSF_OK;
5052 		printf("EEK! Can't write NfsV4 stable storage file\n");
5053 		return;
5054 	}
5055 	sf->nsf_flags |= NFSNSF_OK;
5056 
5057 	/*
5058 	 * Loop through the list and write out timestamp records for
5059 	 * any clients that successfully reclaimed state.
5060 	 */
5061 	LIST_FOREACH_SAFE(sp, &sf->nsf_head, nst_list, nsp) {
5062 		if (sp->nst_flag & NFSNST_GOTSTATE) {
5063 			nfsrv_writestable(sp->nst_client, sp->nst_len,
5064 				NFSNST_NEWSTATE, p);
5065 			sp->nst_clp->lc_flags |= LCL_STAMPEDSTABLE;
5066 		}
5067 		LIST_REMOVE(sp, nst_list);
5068 		free(sp, M_TEMP);
5069 	}
5070 	nfsrv_backupstable();
5071 }
5072 
5073 /*
5074  * Append a record to the stable storage file.
5075  */
5076 void
5077 nfsrv_writestable(u_char *client, int len, int flag, NFSPROC_T *p)
5078 {
5079 	struct nfsrv_stablefirst *sf = &NFSD_VNET(nfsrv_stablefirst);
5080 	struct nfst_rec *sp;
5081 	int error;
5082 
5083 	if (!(sf->nsf_flags & NFSNSF_OK) || sf->nsf_fp == NULL)
5084 		return;
5085 	sp = (struct nfst_rec *)malloc(sizeof (struct nfst_rec) +
5086 		len - 1, M_TEMP, M_WAITOK);
5087 	sp->len = len;
5088 	NFSBCOPY(client, sp->client, len);
5089 	sp->flag = flag;
5090 	error = NFSD_RDWR(UIO_WRITE, NFSFPVNODE(sf->nsf_fp),
5091 	    (caddr_t)sp, sizeof (struct nfst_rec) + len - 1, (off_t)0,
5092 	    UIO_SYSSPACE, (IO_SYNC | IO_APPEND), NFSFPCRED(sf->nsf_fp), NULL, p);
5093 	free(sp, M_TEMP);
5094 	if (error) {
5095 		sf->nsf_flags &= ~NFSNSF_OK;
5096 		printf("EEK! Can't write NfsV4 stable storage file\n");
5097 	}
5098 }
5099 
5100 /*
5101  * This function is called during the grace period to mark a client
5102  * that successfully reclaimed state.
5103  */
5104 static void
5105 nfsrv_markstable(struct nfsclient *clp)
5106 {
5107 	struct nfsrv_stable *sp;
5108 
5109 	/*
5110 	 * First find the client structure.
5111 	 */
5112 	LIST_FOREACH(sp, &NFSD_VNET(nfsrv_stablefirst).nsf_head, nst_list) {
5113 		if (sp->nst_len == clp->lc_idlen &&
5114 		    !NFSBCMP(sp->nst_client, clp->lc_id, sp->nst_len))
5115 			break;
5116 	}
5117 	if (sp == LIST_END(&NFSD_VNET(nfsrv_stablefirst).nsf_head))
5118 		return;
5119 
5120 	/*
5121 	 * Now, just mark it and set the nfsclient back pointer.
5122 	 */
5123 	sp->nst_flag |= NFSNST_GOTSTATE;
5124 	sp->nst_clp = clp;
5125 }
5126 
5127 /*
5128  * This function is called when a NFSv4.1 client does a ReclaimComplete.
5129  * Very similar to nfsrv_markstable(), except for the flag being set.
5130  */
5131 static void
5132 nfsrv_markreclaim(struct nfsclient *clp)
5133 {
5134 	struct nfsrv_stable *sp;
5135 
5136 	/*
5137 	 * First find the client structure.
5138 	 */
5139 	LIST_FOREACH(sp, &NFSD_VNET(nfsrv_stablefirst).nsf_head, nst_list) {
5140 		if (sp->nst_len == clp->lc_idlen &&
5141 		    !NFSBCMP(sp->nst_client, clp->lc_id, sp->nst_len))
5142 			break;
5143 	}
5144 	if (sp == LIST_END(&NFSD_VNET(nfsrv_stablefirst).nsf_head))
5145 		return;
5146 
5147 	/*
5148 	 * Now, just set the flag.
5149 	 */
5150 	sp->nst_flag |= NFSNST_RECLAIMED;
5151 
5152 	/*
5153 	 * Free up any old delegations.
5154 	 */
5155 	nfsrv_freedeleglist(&clp->lc_olddeleg);
5156 }
5157 
5158 /*
5159  * This function is called for a reclaim, to see if it gets grace.
5160  * It returns 0 if a reclaim is allowed, 1 otherwise.
5161  */
5162 static int
5163 nfsrv_checkstable(struct nfsclient *clp)
5164 {
5165 	struct nfsrv_stable *sp;
5166 
5167 	/*
5168 	 * First, find the entry for the client.
5169 	 */
5170 	LIST_FOREACH(sp, &NFSD_VNET(nfsrv_stablefirst).nsf_head, nst_list) {
5171 		if (sp->nst_len == clp->lc_idlen &&
5172 		    !NFSBCMP(sp->nst_client, clp->lc_id, sp->nst_len))
5173 			break;
5174 	}
5175 
5176 	/*
5177 	 * If not in the list, state was revoked or no state was issued
5178 	 * since the previous reboot, a reclaim is denied.
5179 	 */
5180 	if (sp == LIST_END(&NFSD_VNET(nfsrv_stablefirst).nsf_head) ||
5181 	    (sp->nst_flag & NFSNST_REVOKE) ||
5182 	    !(NFSD_VNET(nfsrv_stablefirst).nsf_flags & NFSNSF_OK))
5183 		return (1);
5184 	return (0);
5185 }
5186 
5187 /*
5188  * Test for and try to clear out a conflicting client. This is called by
5189  * nfsrv_lockctrl() and nfsrv_openctrl() when conflicts with other clients
5190  * a found.
5191  * The trick here is that it can't revoke a conflicting client with an
5192  * expired lease unless it holds the v4root lock, so...
5193  * If no v4root lock, get the lock and return 1 to indicate "try again".
5194  * Return 0 to indicate the conflict can't be revoked and 1 to indicate
5195  * the revocation worked and the conflicting client is "bye, bye", so it
5196  * can be tried again.
5197  * Return 2 to indicate that the vnode is VIRF_DOOMED after NFSVOPLOCK().
5198  * Unlocks State before a non-zero value is returned.
5199  */
5200 static int
5201 nfsrv_clientconflict(struct nfsclient *clp, int *haslockp, vnode_t vp,
5202     NFSPROC_T *p)
5203 {
5204 	int gotlock, lktype = 0;
5205 
5206 	/*
5207 	 * If lease hasn't expired, we can't fix it.
5208 	 */
5209 	if (clp->lc_expiry >= NFSD_MONOSEC ||
5210 	    !(NFSD_VNET(nfsrv_stablefirst).nsf_flags & NFSNSF_UPDATEDONE))
5211 		return (0);
5212 	if (*haslockp == 0) {
5213 		NFSUNLOCKSTATE();
5214 		if (vp != NULL) {
5215 			lktype = NFSVOPISLOCKED(vp);
5216 			NFSVOPUNLOCK(vp);
5217 		}
5218 		NFSLOCKV4ROOTMUTEX();
5219 		nfsv4_relref(&nfsv4rootfs_lock);
5220 		do {
5221 			gotlock = nfsv4_lock(&nfsv4rootfs_lock, 1, NULL,
5222 			    NFSV4ROOTLOCKMUTEXPTR, NULL);
5223 		} while (!gotlock);
5224 		NFSUNLOCKV4ROOTMUTEX();
5225 		*haslockp = 1;
5226 		if (vp != NULL) {
5227 			NFSVOPLOCK(vp, lktype | LK_RETRY);
5228 			if (VN_IS_DOOMED(vp))
5229 				return (2);
5230 		}
5231 		return (1);
5232 	}
5233 	NFSUNLOCKSTATE();
5234 
5235 	/*
5236 	 * Ok, we can expire the conflicting client.
5237 	 */
5238 	nfsrv_writestable(clp->lc_id, clp->lc_idlen, NFSNST_REVOKE, p);
5239 	nfsrv_backupstable();
5240 	nfsrv_cleanclient(clp, p, false, NULL);
5241 	nfsrv_freedeleglist(&clp->lc_deleg);
5242 	nfsrv_freedeleglist(&clp->lc_olddeleg);
5243 	LIST_REMOVE(clp, lc_hash);
5244 	nfsrv_zapclient(clp, p);
5245 	return (1);
5246 }
5247 
5248 /*
5249  * Resolve a delegation conflict.
5250  * Returns 0 to indicate the conflict was resolved without sleeping.
5251  * Return -1 to indicate that the caller should check for conflicts again.
5252  * Return > 0 for an error that should be returned, normally NFSERR_DELAY.
5253  *
5254  * Also, manipulate the nfsv4root_lock, as required. It isn't changed
5255  * for a return of 0, since there was no sleep and it could be required
5256  * later. It is released for a return of NFSERR_DELAY, since the caller
5257  * will return that error. It is released when a sleep was done waiting
5258  * for the delegation to be returned or expire (so that other nfsds can
5259  * handle ops). Then, it must be acquired for the write to stable storage.
5260  * (This function is somewhat similar to nfsrv_clientconflict(), but
5261  *  the semantics differ in a couple of subtle ways. The return of 0
5262  *  indicates the conflict was resolved without sleeping here, not
5263  *  that the conflict can't be resolved and the handling of nfsv4root_lock
5264  *  differs, as noted above.)
5265  * Unlocks State before returning a non-zero value.
5266  */
5267 static int
5268 nfsrv_delegconflict(struct nfsstate *stp, int *haslockp, NFSPROC_T *p,
5269     vnode_t vp)
5270 {
5271 	struct nfsclient *clp = stp->ls_clp;
5272 	int gotlock, error, lktype = 0, retrycnt, zapped_clp;
5273 	nfsv4stateid_t tstateid;
5274 	fhandle_t tfh;
5275 
5276 	/*
5277 	 * If the conflict is with an old delegation...
5278 	 */
5279 	if (stp->ls_flags & NFSLCK_OLDDELEG) {
5280 		/*
5281 		 * You can delete it, if it has expired.
5282 		 */
5283 		if (clp->lc_delegtime < NFSD_MONOSEC) {
5284 			nfsrv_freedeleg(stp);
5285 			NFSUNLOCKSTATE();
5286 			error = -1;
5287 			goto out;
5288 		}
5289 		NFSUNLOCKSTATE();
5290 		/*
5291 		 * During this delay, the old delegation could expire or it
5292 		 * could be recovered by the client via an Open with
5293 		 * CLAIM_DELEGATE_PREV.
5294 		 * Release the nfsv4root_lock, if held.
5295 		 */
5296 		if (*haslockp) {
5297 			*haslockp = 0;
5298 			NFSLOCKV4ROOTMUTEX();
5299 			nfsv4_unlock(&nfsv4rootfs_lock, 1);
5300 			NFSUNLOCKV4ROOTMUTEX();
5301 		}
5302 		error = NFSERR_DELAY;
5303 		goto out;
5304 	}
5305 
5306 	/*
5307 	 * It's a current delegation, so:
5308 	 * - check to see if the delegation has expired
5309 	 *   - if so, get the v4root lock and then expire it
5310 	 */
5311 	if ((stp->ls_flags & NFSLCK_DELEGRECALL) == 0 || (stp->ls_lastrecall <
5312 	    NFSD_MONOSEC && clp->lc_expiry >= NFSD_MONOSEC &&
5313 	    stp->ls_delegtime >= NFSD_MONOSEC)) {
5314 		/*
5315 		 * - do a recall callback, since not yet done
5316 		 * For now, never allow truncate to be set. To use
5317 		 * truncate safely, it must be guaranteed that the
5318 		 * Remove, Rename or Setattr with size of 0 will
5319 		 * succeed and that would require major changes to
5320 		 * the VFS/Vnode OPs.
5321 		 * Set the expiry time large enough so that it won't expire
5322 		 * until after the callback, then set it correctly, once
5323 		 * the callback is done. (The delegation will now time
5324 		 * out whether or not the Recall worked ok. The timeout
5325 		 * will be extended when ops are done on the delegation
5326 		 * stateid, up to the timelimit.)
5327 		 */
5328 		if ((stp->ls_flags & NFSLCK_DELEGRECALL) == 0) {
5329 			stp->ls_delegtime = NFSD_MONOSEC + (2 * nfsrv_lease) +
5330 			    NFSRV_LEASEDELTA;
5331 			stp->ls_delegtimelimit = NFSD_MONOSEC + (6 *
5332 			    nfsrv_lease) + NFSRV_LEASEDELTA;
5333 			stp->ls_flags |= NFSLCK_DELEGRECALL;
5334 		}
5335 		stp->ls_lastrecall = time_uptime + 1;
5336 
5337 		/*
5338 		 * Loop NFSRV_CBRETRYCNT times while the CBRecall replies
5339 		 * NFSERR_BADSTATEID or NFSERR_BADHANDLE. This is done
5340 		 * in order to try and avoid a race that could happen
5341 		 * when a CBRecall request passed the Open reply with
5342 		 * the delegation in it when transitting the network.
5343 		 * Since nfsrv_docallback will sleep, don't use stp after
5344 		 * the call.
5345 		 */
5346 		NFSBCOPY((caddr_t)&stp->ls_stateid, (caddr_t)&tstateid,
5347 		    sizeof (tstateid));
5348 		NFSBCOPY((caddr_t)&stp->ls_lfp->lf_fh, (caddr_t)&tfh,
5349 		    sizeof (tfh));
5350 		NFSUNLOCKSTATE();
5351 		if (*haslockp) {
5352 			*haslockp = 0;
5353 			NFSLOCKV4ROOTMUTEX();
5354 			nfsv4_unlock(&nfsv4rootfs_lock, 1);
5355 			NFSUNLOCKV4ROOTMUTEX();
5356 		}
5357 		retrycnt = 0;
5358 		do {
5359 		    error = nfsrv_docallback(clp, NFSV4OP_CBRECALL,
5360 			&tstateid, 0, &tfh, NULL, NULL, 0, p);
5361 		    retrycnt++;
5362 		} while ((error == NFSERR_BADSTATEID ||
5363 		    error == NFSERR_BADHANDLE) && retrycnt < NFSV4_CBRETRYCNT);
5364 		error = NFSERR_DELAY;
5365 		goto out;
5366 	}
5367 
5368 	if (clp->lc_expiry >= NFSD_MONOSEC &&
5369 	    stp->ls_delegtime >= NFSD_MONOSEC) {
5370 		NFSUNLOCKSTATE();
5371 		/*
5372 		 * A recall has been done, but it has not yet expired.
5373 		 * So, RETURN_DELAY.
5374 		 */
5375 		if (*haslockp) {
5376 			*haslockp = 0;
5377 			NFSLOCKV4ROOTMUTEX();
5378 			nfsv4_unlock(&nfsv4rootfs_lock, 1);
5379 			NFSUNLOCKV4ROOTMUTEX();
5380 		}
5381 		error = NFSERR_DELAY;
5382 		goto out;
5383 	}
5384 
5385 	/*
5386 	 * If we don't yet have the lock, just get it and then return,
5387 	 * since we need that before deleting expired state, such as
5388 	 * this delegation.
5389 	 * When getting the lock, unlock the vnode, so other nfsds that
5390 	 * are in progress, won't get stuck waiting for the vnode lock.
5391 	 */
5392 	if (*haslockp == 0) {
5393 		NFSUNLOCKSTATE();
5394 		if (vp != NULL) {
5395 			lktype = NFSVOPISLOCKED(vp);
5396 			NFSVOPUNLOCK(vp);
5397 		}
5398 		NFSLOCKV4ROOTMUTEX();
5399 		nfsv4_relref(&nfsv4rootfs_lock);
5400 		do {
5401 			gotlock = nfsv4_lock(&nfsv4rootfs_lock, 1, NULL,
5402 			    NFSV4ROOTLOCKMUTEXPTR, NULL);
5403 		} while (!gotlock);
5404 		NFSUNLOCKV4ROOTMUTEX();
5405 		*haslockp = 1;
5406 		if (vp != NULL) {
5407 			NFSVOPLOCK(vp, lktype | LK_RETRY);
5408 			if (VN_IS_DOOMED(vp)) {
5409 				*haslockp = 0;
5410 				NFSLOCKV4ROOTMUTEX();
5411 				nfsv4_unlock(&nfsv4rootfs_lock, 1);
5412 				NFSUNLOCKV4ROOTMUTEX();
5413 				error = NFSERR_PERM;
5414 				goto out;
5415 			}
5416 		}
5417 		error = -1;
5418 		goto out;
5419 	}
5420 
5421 	NFSUNLOCKSTATE();
5422 	/*
5423 	 * Ok, we can delete the expired delegation.
5424 	 * First, write the Revoke record to stable storage and then
5425 	 * clear out the conflict.
5426 	 * Since all other nfsd threads are now blocked, we can safely
5427 	 * sleep without the state changing.
5428 	 */
5429 	nfsrv_writestable(clp->lc_id, clp->lc_idlen, NFSNST_REVOKE, p);
5430 	nfsrv_backupstable();
5431 	if (clp->lc_expiry < NFSD_MONOSEC) {
5432 		nfsrv_cleanclient(clp, p, false, NULL);
5433 		nfsrv_freedeleglist(&clp->lc_deleg);
5434 		nfsrv_freedeleglist(&clp->lc_olddeleg);
5435 		LIST_REMOVE(clp, lc_hash);
5436 		zapped_clp = 1;
5437 	} else {
5438 		nfsrv_freedeleg(stp);
5439 		zapped_clp = 0;
5440 	}
5441 	if (zapped_clp)
5442 		nfsrv_zapclient(clp, p);
5443 	error = -1;
5444 
5445 out:
5446 	NFSEXITCODE(error);
5447 	return (error);
5448 }
5449 
5450 /*
5451  * Check for a remove allowed, if remove is set to 1 and get rid of
5452  * delegations.
5453  */
5454 int
5455 nfsrv_checkremove(vnode_t vp, int remove, struct nfsrv_descript *nd,
5456     nfsquad_t clientid, NFSPROC_T *p)
5457 {
5458 	struct nfsclient *clp;
5459 	struct nfsstate *stp;
5460 	struct nfslockfile *lfp;
5461 	int error, haslock = 0;
5462 	fhandle_t nfh;
5463 
5464 	clp = NULL;
5465 	/*
5466 	 * First, get the lock file structure.
5467 	 * (A return of -1 means no associated state, so remove ok.)
5468 	 */
5469 	error = nfsrv_getlockfh(vp, NFSLCK_CHECK, NULL, &nfh, p);
5470 tryagain:
5471 	NFSLOCKSTATE();
5472 	if (error == 0 && clientid.qval != 0)
5473 		error = nfsrv_getclient(clientid, CLOPS_RENEW, &clp, NULL,
5474 		    (nfsquad_t)((u_quad_t)0), 0, nd, p);
5475 	if (!error)
5476 		error = nfsrv_getlockfile(NFSLCK_CHECK, NULL, &lfp, &nfh, 0);
5477 	if (error) {
5478 		NFSUNLOCKSTATE();
5479 		if (haslock) {
5480 			NFSLOCKV4ROOTMUTEX();
5481 			nfsv4_unlock(&nfsv4rootfs_lock, 1);
5482 			NFSUNLOCKV4ROOTMUTEX();
5483 		}
5484 		if (error == -1)
5485 			error = 0;
5486 		goto out;
5487 	}
5488 
5489 	/*
5490 	 * Now, we must Recall any delegations.
5491 	 */
5492 	error = nfsrv_cleandeleg(vp, lfp, clp, &haslock, p);
5493 	if (error) {
5494 		/*
5495 		 * nfsrv_cleandeleg() unlocks state for non-zero
5496 		 * return.
5497 		 */
5498 		if (error == -1)
5499 			goto tryagain;
5500 		if (haslock) {
5501 			NFSLOCKV4ROOTMUTEX();
5502 			nfsv4_unlock(&nfsv4rootfs_lock, 1);
5503 			NFSUNLOCKV4ROOTMUTEX();
5504 		}
5505 		goto out;
5506 	}
5507 
5508 	/*
5509 	 * Now, look for a conflicting open share.
5510 	 */
5511 	if (remove) {
5512 		/*
5513 		 * If the entry in the directory was the last reference to the
5514 		 * corresponding filesystem object, the object can be destroyed
5515 		 * */
5516 		if(lfp->lf_usecount>1)
5517 			LIST_FOREACH(stp, &lfp->lf_open, ls_file) {
5518 				if (stp->ls_flags & NFSLCK_WRITEDENY) {
5519 					error = NFSERR_FILEOPEN;
5520 					break;
5521 				}
5522 			}
5523 	}
5524 
5525 	NFSUNLOCKSTATE();
5526 	if (haslock) {
5527 		NFSLOCKV4ROOTMUTEX();
5528 		nfsv4_unlock(&nfsv4rootfs_lock, 1);
5529 		NFSUNLOCKV4ROOTMUTEX();
5530 	}
5531 
5532 out:
5533 	NFSEXITCODE(error);
5534 	return (error);
5535 }
5536 
5537 /*
5538  * Clear out all delegations for the file referred to by lfp.
5539  * May return NFSERR_DELAY, if there will be a delay waiting for
5540  * delegations to expire.
5541  * Returns -1 to indicate it slept while recalling a delegation.
5542  * This function has the side effect of deleting the nfslockfile structure,
5543  * if it no longer has associated state and didn't have to sleep.
5544  * Unlocks State before a non-zero value is returned.
5545  */
5546 static int
5547 nfsrv_cleandeleg(vnode_t vp, struct nfslockfile *lfp,
5548     struct nfsclient *clp, int *haslockp, NFSPROC_T *p)
5549 {
5550 	struct nfsstate *stp, *nstp;
5551 	int ret = 0;
5552 
5553 	stp = LIST_FIRST(&lfp->lf_deleg);
5554 	while (stp != LIST_END(&lfp->lf_deleg)) {
5555 		nstp = LIST_NEXT(stp, ls_file);
5556 		if (stp->ls_clp != clp) {
5557 			ret = nfsrv_delegconflict(stp, haslockp, p, vp);
5558 			if (ret) {
5559 				/*
5560 				 * nfsrv_delegconflict() unlocks state
5561 				 * when it returns non-zero.
5562 				 */
5563 				goto out;
5564 			}
5565 		}
5566 		stp = nstp;
5567 	}
5568 out:
5569 	NFSEXITCODE(ret);
5570 	return (ret);
5571 }
5572 
5573 /*
5574  * There are certain operations that, when being done outside of NFSv4,
5575  * require that any NFSv4 delegation for the file be recalled.
5576  * This function is to be called for those cases:
5577  * VOP_RENAME() - When a delegation is being recalled for any reason,
5578  *	the client may have to do Opens against the server, using the file's
5579  *	final component name. If the file has been renamed on the server,
5580  *	that component name will be incorrect and the Open will fail.
5581  * VOP_REMOVE() - Theoretically, a client could Open a file after it has
5582  *	been removed on the server, if there is a delegation issued to
5583  *	that client for the file. I say "theoretically" since clients
5584  *	normally do an Access Op before the Open and that Access Op will
5585  *	fail with ESTALE. Note that NFSv2 and 3 don't even do Opens, so
5586  *	they will detect the file's removal in the same manner. (There is
5587  *	one case where RFC3530 allows a client to do an Open without first
5588  *	doing an Access Op, which is passage of a check against the ACE
5589  *	returned with a Write delegation, but current practice is to ignore
5590  *	the ACE and always do an Access Op.)
5591  *	Since the functions can only be called with an unlocked vnode, this
5592  *	can't be done at this time.
5593  * VOP_ADVLOCK() - When a client holds a delegation, it can issue byte range
5594  *	locks locally in the client, which are not visible to the server. To
5595  *	deal with this, issuing of delegations for a vnode must be disabled
5596  *	and all delegations for the vnode recalled. This is done via the
5597  *	second function, using the VV_DISABLEDELEG vflag on the vnode.
5598  */
5599 void
5600 nfsd_recalldelegation(vnode_t vp, NFSPROC_T *p)
5601 {
5602 	time_t starttime;
5603 	int error;
5604 
5605 	/*
5606 	 * First, check to see if the server is currently running and it has
5607 	 * been called for a regular file when issuing delegations.
5608 	 */
5609 	if (NFSD_VNET(nfsrv_numnfsd) == 0 || vp->v_type != VREG ||
5610 	    nfsrv_issuedelegs == 0)
5611 		return;
5612 
5613 	KASSERT((NFSVOPISLOCKED(vp) != LK_EXCLUSIVE), ("vp %p is locked", vp));
5614 	/*
5615 	 * First, get a reference on the nfsv4rootfs_lock so that an
5616 	 * exclusive lock cannot be acquired by another thread.
5617 	 */
5618 	NFSLOCKV4ROOTMUTEX();
5619 	nfsv4_getref(&nfsv4rootfs_lock, NULL, NFSV4ROOTLOCKMUTEXPTR, NULL);
5620 	NFSUNLOCKV4ROOTMUTEX();
5621 
5622 	/*
5623 	 * Now, call nfsrv_checkremove() in a loop while it returns
5624 	 * NFSERR_DELAY. Return upon any other error or when timed out.
5625 	 */
5626 	starttime = NFSD_MONOSEC;
5627 	do {
5628 		if (NFSVOPLOCK(vp, LK_EXCLUSIVE) == 0) {
5629 			error = nfsrv_checkremove(vp, 0, NULL,
5630 			    (nfsquad_t)((u_quad_t)0), p);
5631 			NFSVOPUNLOCK(vp);
5632 		} else
5633 			error = EPERM;
5634 		if (error == NFSERR_DELAY) {
5635 			if (NFSD_MONOSEC - starttime > NFS_REMOVETIMEO)
5636 				break;
5637 			/* Sleep for a short period of time */
5638 			(void) nfs_catnap(PZERO, 0, "nfsremove");
5639 		}
5640 	} while (error == NFSERR_DELAY);
5641 	NFSLOCKV4ROOTMUTEX();
5642 	nfsv4_relref(&nfsv4rootfs_lock);
5643 	NFSUNLOCKV4ROOTMUTEX();
5644 }
5645 
5646 void
5647 nfsd_disabledelegation(vnode_t vp, NFSPROC_T *p)
5648 {
5649 
5650 #ifdef VV_DISABLEDELEG
5651 	/*
5652 	 * First, flag issuance of delegations disabled.
5653 	 */
5654 	atomic_set_long(&vp->v_vflag, VV_DISABLEDELEG);
5655 #endif
5656 
5657 	/*
5658 	 * Then call nfsd_recalldelegation() to get rid of all extant
5659 	 * delegations.
5660 	 */
5661 	nfsd_recalldelegation(vp, p);
5662 }
5663 
5664 /*
5665  * Check for conflicting locks, etc. and then get rid of delegations.
5666  * (At one point I thought that I should get rid of delegations for any
5667  *  Setattr, since it could potentially disallow the I/O op (read or write)
5668  *  allowed by the delegation. However, Setattr Ops that aren't changing
5669  *  the size get a stateid of all 0s, so you can't tell if it is a delegation
5670  *  for the same client or a different one, so I decided to only get rid
5671  *  of delegations for other clients when the size is being changed.)
5672  * In general, a Setattr can disable NFS I/O Ops that are outstanding, such
5673  * as Write backs, even if there is no delegation, so it really isn't any
5674  * different?)
5675  */
5676 int
5677 nfsrv_checksetattr(vnode_t vp, struct nfsrv_descript *nd,
5678     nfsv4stateid_t *stateidp, struct nfsvattr *nvap, nfsattrbit_t *attrbitp,
5679     struct nfsexstuff *exp, NFSPROC_T *p)
5680 {
5681 	struct nfsstate st, *stp = &st;
5682 	struct nfslock lo, *lop = &lo;
5683 	int error = 0;
5684 	nfsquad_t clientid;
5685 
5686 	if (NFSISSET_ATTRBIT(attrbitp, NFSATTRBIT_SIZE)) {
5687 		stp->ls_flags = (NFSLCK_CHECK | NFSLCK_WRITEACCESS);
5688 		lop->lo_first = nvap->na_size;
5689 	} else {
5690 		stp->ls_flags = 0;
5691 		lop->lo_first = 0;
5692 	}
5693 	if (NFSISSET_ATTRBIT(attrbitp, NFSATTRBIT_OWNER) ||
5694 	    NFSISSET_ATTRBIT(attrbitp, NFSATTRBIT_OWNERGROUP) ||
5695 	    NFSISSET_ATTRBIT(attrbitp, NFSATTRBIT_MODE) ||
5696 	    NFSISSET_ATTRBIT(attrbitp, NFSATTRBIT_ACL))
5697 		stp->ls_flags |= NFSLCK_SETATTR;
5698 	if (stp->ls_flags == 0)
5699 		goto out;
5700 	lop->lo_end = NFS64BITSSET;
5701 	lop->lo_flags = NFSLCK_WRITE;
5702 	stp->ls_ownerlen = 0;
5703 	stp->ls_op = NULL;
5704 	stp->ls_uid = nd->nd_cred->cr_uid;
5705 	stp->ls_stateid.seqid = stateidp->seqid;
5706 	clientid.lval[0] = stp->ls_stateid.other[0] = stateidp->other[0];
5707 	clientid.lval[1] = stp->ls_stateid.other[1] = stateidp->other[1];
5708 	stp->ls_stateid.other[2] = stateidp->other[2];
5709 	error = nfsrv_lockctrl(vp, &stp, &lop, NULL, clientid,
5710 	    stateidp, exp, nd, p);
5711 
5712 out:
5713 	NFSEXITCODE2(error, nd);
5714 	return (error);
5715 }
5716 
5717 /*
5718  * Check for a write delegation and do a CBGETATTR if there is one, updating
5719  * the attributes, as required.
5720  * Should I return an error if I can't get the attributes? (For now, I'll
5721  * just return ok.
5722  */
5723 int
5724 nfsrv_checkgetattr(struct nfsrv_descript *nd, vnode_t vp,
5725     struct nfsvattr *nvap, nfsattrbit_t *attrbitp, NFSPROC_T *p)
5726 {
5727 	struct nfsstate *stp;
5728 	struct nfslockfile *lfp;
5729 	struct nfsclient *clp;
5730 	struct nfsvattr nva;
5731 	fhandle_t nfh;
5732 	int error = 0;
5733 	nfsattrbit_t cbbits;
5734 	u_quad_t delegfilerev;
5735 
5736 	NFSCBGETATTR_ATTRBIT(attrbitp, &cbbits);
5737 	if (!NFSNONZERO_ATTRBIT(&cbbits))
5738 		goto out;
5739 	if (nfsrv_writedelegcnt == 0)
5740 		goto out;
5741 
5742 	/*
5743 	 * Get the lock file structure.
5744 	 * (A return of -1 means no associated state, so return ok.)
5745 	 */
5746 	error = nfsrv_getlockfh(vp, NFSLCK_CHECK, NULL, &nfh, p);
5747 	NFSLOCKSTATE();
5748 	if (!error)
5749 		error = nfsrv_getlockfile(NFSLCK_CHECK, NULL, &lfp, &nfh, 0);
5750 	if (error) {
5751 		NFSUNLOCKSTATE();
5752 		if (error == -1)
5753 			error = 0;
5754 		goto out;
5755 	}
5756 
5757 	/*
5758 	 * Now, look for a write delegation.
5759 	 */
5760 	LIST_FOREACH(stp, &lfp->lf_deleg, ls_file) {
5761 		if (stp->ls_flags & NFSLCK_DELEGWRITE)
5762 			break;
5763 	}
5764 	if (stp == LIST_END(&lfp->lf_deleg)) {
5765 		NFSUNLOCKSTATE();
5766 		goto out;
5767 	}
5768 	clp = stp->ls_clp;
5769 
5770 	/* If the clientid is not confirmed, ignore the delegation. */
5771 	if (clp->lc_flags & LCL_NEEDSCONFIRM) {
5772 		NFSUNLOCKSTATE();
5773 		goto out;
5774 	}
5775 
5776 	delegfilerev = stp->ls_filerev;
5777 	/*
5778 	 * If the Write delegation was issued as a part of this Compound RPC
5779 	 * or if we have an Implied Clientid (used in a previous Op in this
5780 	 * compound) and it is the client the delegation was issued to,
5781 	 * just return ok.
5782 	 * I also assume that it is from the same client iff the network
5783 	 * host IP address is the same as the callback address. (Not
5784 	 * exactly correct by the RFC, but avoids a lot of Getattr
5785 	 * callbacks.)
5786 	 */
5787 	if (nd->nd_compref == stp->ls_compref ||
5788 	    ((nd->nd_flag & ND_IMPLIEDCLID) &&
5789 	     clp->lc_clientid.qval == nd->nd_clientid.qval) ||
5790 	     nfsaddr2_match(clp->lc_req.nr_nam, nd->nd_nam)) {
5791 		NFSUNLOCKSTATE();
5792 		goto out;
5793 	}
5794 
5795 	/*
5796 	 * We are now done with the delegation state structure,
5797 	 * so the statelock can be released and we can now tsleep().
5798 	 */
5799 
5800 	/*
5801 	 * Now, we must do the CB Getattr callback, to see if Change or Size
5802 	 * has changed.
5803 	 */
5804 	if (clp->lc_expiry >= NFSD_MONOSEC) {
5805 		NFSUNLOCKSTATE();
5806 		NFSVNO_ATTRINIT(&nva);
5807 		nva.na_filerev = NFS64BITSSET;
5808 		error = nfsrv_docallback(clp, NFSV4OP_CBGETATTR, NULL,
5809 		    0, &nfh, &nva, &cbbits, 0, p);
5810 		if (!error) {
5811 			if ((nva.na_filerev != NFS64BITSSET &&
5812 			    nva.na_filerev > delegfilerev) ||
5813 			    (NFSVNO_ISSETSIZE(&nva) &&
5814 			     nva.na_size != nvap->na_size)) {
5815 				error = nfsvno_updfilerev(vp, nvap, nd, p);
5816 				if (NFSVNO_ISSETSIZE(&nva))
5817 					nvap->na_size = nva.na_size;
5818 			}
5819 		} else
5820 			error = 0;	/* Ignore callback errors for now. */
5821 	} else {
5822 		NFSUNLOCKSTATE();
5823 	}
5824 
5825 out:
5826 	NFSEXITCODE2(error, nd);
5827 	return (error);
5828 }
5829 
5830 /*
5831  * This function looks for openowners that haven't had any opens for
5832  * a while and throws them away. Called by an nfsd when NFSNSF_NOOPENS
5833  * is set.
5834  */
5835 void
5836 nfsrv_throwawayopens(NFSPROC_T *p)
5837 {
5838 	struct nfsclient *clp, *nclp;
5839 	struct nfsstate *stp, *nstp;
5840 	int i;
5841 
5842 	NFSLOCKSTATE();
5843 	NFSD_VNET(nfsrv_stablefirst).nsf_flags &= ~NFSNSF_NOOPENS;
5844 	/*
5845 	 * For each client...
5846 	 */
5847 	for (i = 0; i < nfsrv_clienthashsize; i++) {
5848 	    LIST_FOREACH_SAFE(clp, &NFSD_VNET(nfsclienthash)[i], lc_hash,
5849 		nclp) {
5850 		LIST_FOREACH_SAFE(stp, &clp->lc_open, ls_list, nstp) {
5851 			if (LIST_EMPTY(&stp->ls_open) &&
5852 			    (stp->ls_noopens > NFSNOOPEN ||
5853 			     (nfsrv_openpluslock * 2) >
5854 			     nfsrv_v4statelimit))
5855 				nfsrv_freeopenowner(stp, 0, p);
5856 		}
5857 	    }
5858 	}
5859 	NFSUNLOCKSTATE();
5860 }
5861 
5862 /*
5863  * This function checks to see if the credentials are the same.
5864  * The check for same credentials is needed for state management operations
5865  * for NFSv4.0 or NFSv4.1/4.2 when SP4_MACH_CRED is configured via
5866  * ExchangeID.
5867  * Returns 1 for not same, 0 otherwise.
5868  */
5869 static int
5870 nfsrv_notsamecredname(int op, struct nfsrv_descript *nd, struct nfsclient *clp)
5871 {
5872 
5873 	/* Check for the SP4_MACH_CRED case. */
5874 	if (op != 0 && nfsrv_checkmachcred(op, nd, clp) != 0)
5875 		return (1);
5876 
5877 	/* For NFSv4.1/4.2, SP4_NONE always allows this. */
5878 	if ((nd->nd_flag & ND_NFSV41) != 0)
5879 		return (0);
5880 
5881 	if (nd->nd_flag & ND_GSS) {
5882 		if (!(clp->lc_flags & LCL_GSS))
5883 			return (1);
5884 		if (clp->lc_flags & LCL_NAME) {
5885 			if (nd->nd_princlen != clp->lc_namelen ||
5886 			    NFSBCMP(nd->nd_principal, clp->lc_name,
5887 				clp->lc_namelen))
5888 				return (1);
5889 			else
5890 				return (0);
5891 		}
5892 		if (nd->nd_cred->cr_uid == clp->lc_uid)
5893 			return (0);
5894 		else
5895 			return (1);
5896 	} else if (clp->lc_flags & LCL_GSS)
5897 		return (1);
5898 	/*
5899 	 * For AUTH_SYS, allow the same uid or root. (This is underspecified
5900 	 * in RFC3530, which talks about principals, but doesn't say anything
5901 	 * about uids for AUTH_SYS.)
5902 	 */
5903 	if (nd->nd_cred->cr_uid == clp->lc_uid || nd->nd_cred->cr_uid == 0)
5904 		return (0);
5905 	else
5906 		return (1);
5907 }
5908 
5909 /*
5910  * Calculate the lease expiry time.
5911  */
5912 static time_t
5913 nfsrv_leaseexpiry(void)
5914 {
5915 
5916 	if (NFSD_VNET(nfsrv_stablefirst).nsf_eograce > NFSD_MONOSEC)
5917 		return (NFSD_MONOSEC + 2 * (nfsrv_lease + NFSRV_LEASEDELTA));
5918 	return (NFSD_MONOSEC + nfsrv_lease + NFSRV_LEASEDELTA);
5919 }
5920 
5921 /*
5922  * Delay the delegation timeout as far as ls_delegtimelimit, as required.
5923  */
5924 static void
5925 nfsrv_delaydelegtimeout(struct nfsstate *stp)
5926 {
5927 
5928 	if ((stp->ls_flags & NFSLCK_DELEGRECALL) == 0)
5929 		return;
5930 
5931 	if ((stp->ls_delegtime + 15) > NFSD_MONOSEC &&
5932 	    stp->ls_delegtime < stp->ls_delegtimelimit) {
5933 		stp->ls_delegtime += nfsrv_lease;
5934 		if (stp->ls_delegtime > stp->ls_delegtimelimit)
5935 			stp->ls_delegtime = stp->ls_delegtimelimit;
5936 	}
5937 }
5938 
5939 /*
5940  * This function checks to see if there is any other state associated
5941  * with the openowner for this Open.
5942  * It returns 1 if there is no other state, 0 otherwise.
5943  */
5944 static int
5945 nfsrv_nootherstate(struct nfsstate *stp)
5946 {
5947 	struct nfsstate *tstp;
5948 
5949 	LIST_FOREACH(tstp, &stp->ls_openowner->ls_open, ls_list) {
5950 		if (tstp != stp || !LIST_EMPTY(&tstp->ls_lock))
5951 			return (0);
5952 	}
5953 	return (1);
5954 }
5955 
5956 /*
5957  * Create a list of lock deltas (changes to local byte range locking
5958  * that can be rolled back using the list) and apply the changes via
5959  * nfsvno_advlock(). Optionally, lock the list. It is expected that either
5960  * the rollback or update function will be called after this.
5961  * It returns an error (and rolls back, as required), if any nfsvno_advlock()
5962  * call fails. If it returns an error, it will unlock the list.
5963  */
5964 static int
5965 nfsrv_locallock(vnode_t vp, struct nfslockfile *lfp, int flags,
5966     uint64_t first, uint64_t end, struct nfslockconflict *cfp, NFSPROC_T *p)
5967 {
5968 	struct nfslock *lop, *nlop;
5969 	int error = 0;
5970 
5971 	/* Loop through the list of locks. */
5972 	lop = LIST_FIRST(&lfp->lf_locallock);
5973 	while (first < end && lop != NULL) {
5974 		nlop = LIST_NEXT(lop, lo_lckowner);
5975 		if (first >= lop->lo_end) {
5976 			/* not there yet */
5977 			lop = nlop;
5978 		} else if (first < lop->lo_first) {
5979 			/* new one starts before entry in list */
5980 			if (end <= lop->lo_first) {
5981 				/* no overlap between old and new */
5982 				error = nfsrv_dolocal(vp, lfp, flags,
5983 				    NFSLCK_UNLOCK, first, end, cfp, p);
5984 				if (error != 0)
5985 					break;
5986 				first = end;
5987 			} else {
5988 				/* handle fragment overlapped with new one */
5989 				error = nfsrv_dolocal(vp, lfp, flags,
5990 				    NFSLCK_UNLOCK, first, lop->lo_first, cfp,
5991 				    p);
5992 				if (error != 0)
5993 					break;
5994 				first = lop->lo_first;
5995 			}
5996 		} else {
5997 			/* new one overlaps this entry in list */
5998 			if (end <= lop->lo_end) {
5999 				/* overlaps all of new one */
6000 				error = nfsrv_dolocal(vp, lfp, flags,
6001 				    lop->lo_flags, first, end, cfp, p);
6002 				if (error != 0)
6003 					break;
6004 				first = end;
6005 			} else {
6006 				/* handle fragment overlapped with new one */
6007 				error = nfsrv_dolocal(vp, lfp, flags,
6008 				    lop->lo_flags, first, lop->lo_end, cfp, p);
6009 				if (error != 0)
6010 					break;
6011 				first = lop->lo_end;
6012 				lop = nlop;
6013 			}
6014 		}
6015 	}
6016 	if (first < end && error == 0)
6017 		/* handle fragment past end of list */
6018 		error = nfsrv_dolocal(vp, lfp, flags, NFSLCK_UNLOCK, first,
6019 		    end, cfp, p);
6020 
6021 	NFSEXITCODE(error);
6022 	return (error);
6023 }
6024 
6025 /*
6026  * Local lock unlock. Unlock all byte ranges that are no longer locked
6027  * by NFSv4. To do this, unlock any subranges of first-->end that
6028  * do not overlap with the byte ranges of any lock in the lfp->lf_lock
6029  * list. This list has all locks for the file held by other
6030  * <clientid, lockowner> tuples. The list is ordered by increasing
6031  * lo_first value, but may have entries that overlap each other, for
6032  * the case of read locks.
6033  */
6034 static void
6035 nfsrv_localunlock(vnode_t vp, struct nfslockfile *lfp, uint64_t init_first,
6036     uint64_t init_end, NFSPROC_T *p)
6037 {
6038 	struct nfslock *lop;
6039 	uint64_t first, end, prevfirst __unused;
6040 
6041 	first = init_first;
6042 	end = init_end;
6043 	while (first < init_end) {
6044 		/* Loop through all nfs locks, adjusting first and end */
6045 		prevfirst = 0;
6046 		LIST_FOREACH(lop, &lfp->lf_lock, lo_lckfile) {
6047 			KASSERT(prevfirst <= lop->lo_first,
6048 			    ("nfsv4 locks out of order"));
6049 			KASSERT(lop->lo_first < lop->lo_end,
6050 			    ("nfsv4 bogus lock"));
6051 			prevfirst = lop->lo_first;
6052 			if (first >= lop->lo_first &&
6053 			    first < lop->lo_end)
6054 				/*
6055 				 * Overlaps with initial part, so trim
6056 				 * off that initial part by moving first past
6057 				 * it.
6058 				 */
6059 				first = lop->lo_end;
6060 			else if (end > lop->lo_first &&
6061 			    lop->lo_first > first) {
6062 				/*
6063 				 * This lock defines the end of the
6064 				 * segment to unlock, so set end to the
6065 				 * start of it and break out of the loop.
6066 				 */
6067 				end = lop->lo_first;
6068 				break;
6069 			}
6070 			if (first >= end)
6071 				/*
6072 				 * There is no segment left to do, so
6073 				 * break out of this loop and then exit
6074 				 * the outer while() since first will be set
6075 				 * to end, which must equal init_end here.
6076 				 */
6077 				break;
6078 		}
6079 		if (first < end) {
6080 			/* Unlock this segment */
6081 			(void) nfsrv_dolocal(vp, lfp, NFSLCK_UNLOCK,
6082 			    NFSLCK_READ, first, end, NULL, p);
6083 			nfsrv_locallock_commit(lfp, NFSLCK_UNLOCK,
6084 			    first, end);
6085 		}
6086 		/*
6087 		 * Now move past this segment and look for any further
6088 		 * segment in the range, if there is one.
6089 		 */
6090 		first = end;
6091 		end = init_end;
6092 	}
6093 }
6094 
6095 /*
6096  * Do the local lock operation and update the rollback list, as required.
6097  * Perform the rollback and return the error if nfsvno_advlock() fails.
6098  */
6099 static int
6100 nfsrv_dolocal(vnode_t vp, struct nfslockfile *lfp, int flags, int oldflags,
6101     uint64_t first, uint64_t end, struct nfslockconflict *cfp, NFSPROC_T *p)
6102 {
6103 	struct nfsrollback *rlp;
6104 	int error = 0, ltype, oldltype;
6105 
6106 	if (flags & NFSLCK_WRITE)
6107 		ltype = F_WRLCK;
6108 	else if (flags & NFSLCK_READ)
6109 		ltype = F_RDLCK;
6110 	else
6111 		ltype = F_UNLCK;
6112 	if (oldflags & NFSLCK_WRITE)
6113 		oldltype = F_WRLCK;
6114 	else if (oldflags & NFSLCK_READ)
6115 		oldltype = F_RDLCK;
6116 	else
6117 		oldltype = F_UNLCK;
6118 	if (ltype == oldltype || (oldltype == F_WRLCK && ltype == F_RDLCK))
6119 		/* nothing to do */
6120 		goto out;
6121 	error = nfsvno_advlock(vp, ltype, first, end, p);
6122 	if (error != 0) {
6123 		if (cfp != NULL) {
6124 			cfp->cl_clientid.lval[0] = 0;
6125 			cfp->cl_clientid.lval[1] = 0;
6126 			cfp->cl_first = 0;
6127 			cfp->cl_end = NFS64BITSSET;
6128 			cfp->cl_flags = NFSLCK_WRITE;
6129 			cfp->cl_ownerlen = 5;
6130 			NFSBCOPY("LOCAL", cfp->cl_owner, 5);
6131 		}
6132 		nfsrv_locallock_rollback(vp, lfp, p);
6133 	} else if (ltype != F_UNLCK) {
6134 		rlp = malloc(sizeof (struct nfsrollback), M_NFSDROLLBACK,
6135 		    M_WAITOK);
6136 		rlp->rlck_first = first;
6137 		rlp->rlck_end = end;
6138 		rlp->rlck_type = oldltype;
6139 		LIST_INSERT_HEAD(&lfp->lf_rollback, rlp, rlck_list);
6140 	}
6141 
6142 out:
6143 	NFSEXITCODE(error);
6144 	return (error);
6145 }
6146 
6147 /*
6148  * Roll back local lock changes and free up the rollback list.
6149  */
6150 static void
6151 nfsrv_locallock_rollback(vnode_t vp, struct nfslockfile *lfp, NFSPROC_T *p)
6152 {
6153 	struct nfsrollback *rlp, *nrlp;
6154 
6155 	LIST_FOREACH_SAFE(rlp, &lfp->lf_rollback, rlck_list, nrlp) {
6156 		(void) nfsvno_advlock(vp, rlp->rlck_type, rlp->rlck_first,
6157 		    rlp->rlck_end, p);
6158 		free(rlp, M_NFSDROLLBACK);
6159 	}
6160 	LIST_INIT(&lfp->lf_rollback);
6161 }
6162 
6163 /*
6164  * Update local lock list and delete rollback list (ie now committed to the
6165  * local locks). Most of the work is done by the internal function.
6166  */
6167 static void
6168 nfsrv_locallock_commit(struct nfslockfile *lfp, int flags, uint64_t first,
6169     uint64_t end)
6170 {
6171 	struct nfsrollback *rlp, *nrlp;
6172 	struct nfslock *new_lop, *other_lop;
6173 
6174 	new_lop = malloc(sizeof (struct nfslock), M_NFSDLOCK, M_WAITOK);
6175 	if (flags & (NFSLCK_READ | NFSLCK_WRITE))
6176 		other_lop = malloc(sizeof (struct nfslock), M_NFSDLOCK,
6177 		    M_WAITOK);
6178 	else
6179 		other_lop = NULL;
6180 	new_lop->lo_flags = flags;
6181 	new_lop->lo_first = first;
6182 	new_lop->lo_end = end;
6183 	nfsrv_updatelock(NULL, &new_lop, &other_lop, lfp);
6184 	if (new_lop != NULL)
6185 		free(new_lop, M_NFSDLOCK);
6186 	if (other_lop != NULL)
6187 		free(other_lop, M_NFSDLOCK);
6188 
6189 	/* and get rid of the rollback list */
6190 	LIST_FOREACH_SAFE(rlp, &lfp->lf_rollback, rlck_list, nrlp)
6191 		free(rlp, M_NFSDROLLBACK);
6192 	LIST_INIT(&lfp->lf_rollback);
6193 }
6194 
6195 /*
6196  * Lock the struct nfslockfile for local lock updating.
6197  */
6198 static void
6199 nfsrv_locklf(struct nfslockfile *lfp)
6200 {
6201 	int gotlock;
6202 
6203 	/* lf_usecount ensures *lfp won't be free'd */
6204 	lfp->lf_usecount++;
6205 	do {
6206 		gotlock = nfsv4_lock(&lfp->lf_locallock_lck, 1, NULL,
6207 		    NFSSTATEMUTEXPTR, NULL);
6208 	} while (gotlock == 0);
6209 	lfp->lf_usecount--;
6210 }
6211 
6212 /*
6213  * Unlock the struct nfslockfile after local lock updating.
6214  */
6215 static void
6216 nfsrv_unlocklf(struct nfslockfile *lfp)
6217 {
6218 
6219 	nfsv4_unlock(&lfp->lf_locallock_lck, 0);
6220 }
6221 
6222 /*
6223  * Clear out all state for the NFSv4 server.
6224  * Must be called by a thread that can sleep when no nfsds are running.
6225  */
6226 void
6227 nfsrv_throwawayallstate(NFSPROC_T *p)
6228 {
6229 	struct nfsclient *clp, *nclp;
6230 	struct nfslockfile *lfp, *nlfp;
6231 	int i;
6232 
6233 	/*
6234 	 * For each client, clean out the state and then free the structure.
6235 	 */
6236 	for (i = 0; i < nfsrv_clienthashsize; i++) {
6237 		LIST_FOREACH_SAFE(clp, &NFSD_VNET(nfsclienthash)[i], lc_hash,
6238 		    nclp) {
6239 			nfsrv_cleanclient(clp, p, false, NULL);
6240 			nfsrv_freedeleglist(&clp->lc_deleg);
6241 			nfsrv_freedeleglist(&clp->lc_olddeleg);
6242 			free(clp->lc_stateid, M_NFSDCLIENT);
6243 			free(clp, M_NFSDCLIENT);
6244 		}
6245 	}
6246 
6247 	/*
6248 	 * Also, free up any remaining lock file structures.
6249 	 */
6250 	for (i = 0; i < nfsrv_lockhashsize; i++) {
6251 		LIST_FOREACH_SAFE(lfp, &NFSD_VNET(nfslockhash)[i], lf_hash,
6252 		    nlfp) {
6253 			printf("nfsd unload: fnd a lock file struct\n");
6254 			nfsrv_freenfslockfile(lfp);
6255 		}
6256 	}
6257 
6258 	/* And get rid of the deviceid structures and layouts. */
6259 	nfsrv_freealllayoutsanddevids();
6260 }
6261 
6262 /*
6263  * Check the sequence# for the session and slot provided as an argument.
6264  * Also, renew the lease if the session will return NFS_OK.
6265  */
6266 int
6267 nfsrv_checksequence(struct nfsrv_descript *nd, uint32_t sequenceid,
6268     uint32_t *highest_slotidp, uint32_t *target_highest_slotidp, int cache_this,
6269     uint32_t *sflagsp, NFSPROC_T *p)
6270 {
6271 	struct nfsdsession *sep;
6272 	struct nfssessionhash *shp;
6273 	int error;
6274 
6275 	shp = NFSSESSIONHASH(nd->nd_sessionid);
6276 	NFSLOCKSESSION(shp);
6277 	sep = nfsrv_findsession(nd->nd_sessionid);
6278 	if (sep == NULL) {
6279 		NFSUNLOCKSESSION(shp);
6280 		return (NFSERR_BADSESSION);
6281 	}
6282 	error = nfsv4_seqsession(sequenceid, nd->nd_slotid, *highest_slotidp,
6283 	    sep->sess_slots, NULL, NFSV4_SLOTS - 1);
6284 	if (error != 0) {
6285 		NFSUNLOCKSESSION(shp);
6286 		return (error);
6287 	}
6288 	if (cache_this != 0)
6289 		nd->nd_flag |= ND_SAVEREPLY;
6290 	/* Renew the lease. */
6291 	sep->sess_clp->lc_expiry = nfsrv_leaseexpiry();
6292 	nd->nd_clientid.qval = sep->sess_clp->lc_clientid.qval;
6293 	nd->nd_flag |= ND_IMPLIEDCLID;
6294 
6295 	/* Handle the SP4_MECH_CRED case for NFSv4.1/4.2. */
6296 	if ((sep->sess_clp->lc_flags & LCL_MACHCRED) != 0 &&
6297 	    (nd->nd_flag & (ND_GSSINTEGRITY | ND_GSSPRIVACY)) != 0 &&
6298 	    nd->nd_princlen == sep->sess_clp->lc_namelen &&
6299 	    !NFSBCMP(sep->sess_clp->lc_name, nd->nd_principal,
6300 	    nd->nd_princlen)) {
6301 		nd->nd_flag |= ND_MACHCRED;
6302 		NFSSET_OPBIT(&nd->nd_allowops, &sep->sess_clp->lc_allowops);
6303 	}
6304 
6305 	/* Save maximum request and reply sizes. */
6306 	nd->nd_maxreq = sep->sess_maxreq;
6307 	nd->nd_maxresp = sep->sess_maxresp;
6308 
6309 	*sflagsp = 0;
6310 	if (sep->sess_clp->lc_req.nr_client == NULL ||
6311 	    (sep->sess_clp->lc_flags & LCL_CBDOWN) != 0)
6312 		*sflagsp |= NFSV4SEQ_CBPATHDOWN;
6313 	NFSUNLOCKSESSION(shp);
6314 	if (error == NFSERR_EXPIRED) {
6315 		*sflagsp |= NFSV4SEQ_EXPIREDALLSTATEREVOKED;
6316 		error = 0;
6317 	} else if (error == NFSERR_ADMINREVOKED) {
6318 		*sflagsp |= NFSV4SEQ_ADMINSTATEREVOKED;
6319 		error = 0;
6320 	}
6321 	*highest_slotidp = *target_highest_slotidp = NFSV4_SLOTS - 1;
6322 	return (0);
6323 }
6324 
6325 /*
6326  * Check/set reclaim complete for this session/clientid.
6327  */
6328 int
6329 nfsrv_checkreclaimcomplete(struct nfsrv_descript *nd, int onefs)
6330 {
6331 	struct nfsdsession *sep;
6332 	struct nfssessionhash *shp;
6333 	int error = 0;
6334 
6335 	shp = NFSSESSIONHASH(nd->nd_sessionid);
6336 	NFSLOCKSTATE();
6337 	NFSLOCKSESSION(shp);
6338 	sep = nfsrv_findsession(nd->nd_sessionid);
6339 	if (sep == NULL) {
6340 		NFSUNLOCKSESSION(shp);
6341 		NFSUNLOCKSTATE();
6342 		return (NFSERR_BADSESSION);
6343 	}
6344 
6345 	if (onefs != 0)
6346 		sep->sess_clp->lc_flags |= LCL_RECLAIMONEFS;
6347 		/* Check to see if reclaim complete has already happened. */
6348 	else if ((sep->sess_clp->lc_flags & LCL_RECLAIMCOMPLETE) != 0)
6349 		error = NFSERR_COMPLETEALREADY;
6350 	else {
6351 		sep->sess_clp->lc_flags |= LCL_RECLAIMCOMPLETE;
6352 		nfsrv_markreclaim(sep->sess_clp);
6353 	}
6354 	NFSUNLOCKSESSION(shp);
6355 	NFSUNLOCKSTATE();
6356 	return (error);
6357 }
6358 
6359 /*
6360  * Cache the reply in a session slot.
6361  */
6362 void
6363 nfsrv_cache_session(struct nfsrv_descript *nd, struct mbuf **m)
6364 {
6365 	struct nfsdsession *sep;
6366 	struct nfssessionhash *shp;
6367 	char *buf, *cp;
6368 #ifdef INET
6369 	struct sockaddr_in *sin;
6370 #endif
6371 #ifdef INET6
6372 	struct sockaddr_in6 *sin6;
6373 #endif
6374 
6375 	shp = NFSSESSIONHASH(nd->nd_sessionid);
6376 	NFSLOCKSESSION(shp);
6377 	sep = nfsrv_findsession(nd->nd_sessionid);
6378 	if (sep == NULL) {
6379 		NFSUNLOCKSESSION(shp);
6380 		if ((NFSD_VNET(nfsrv_stablefirst).nsf_flags &
6381 		     NFSNSF_GRACEOVER) != 0) {
6382 			buf = malloc(INET6_ADDRSTRLEN, M_TEMP, M_WAITOK);
6383 			switch (nd->nd_nam->sa_family) {
6384 #ifdef INET
6385 			case AF_INET:
6386 				sin = (struct sockaddr_in *)nd->nd_nam;
6387 				cp = inet_ntop(sin->sin_family,
6388 				    &sin->sin_addr.s_addr, buf,
6389 				    INET6_ADDRSTRLEN);
6390 				break;
6391 #endif
6392 #ifdef INET6
6393 			case AF_INET6:
6394 				sin6 = (struct sockaddr_in6 *)nd->nd_nam;
6395 				cp = inet_ntop(sin6->sin6_family,
6396 				    &sin6->sin6_addr, buf, INET6_ADDRSTRLEN);
6397 				break;
6398 #endif
6399 			default:
6400 				cp = NULL;
6401 			}
6402 			if (cp != NULL)
6403 				printf("nfsrv_cache_session: no session "
6404 				    "IPaddr=%s, check NFS clients for unique "
6405 				    "/etc/hostid's\n", cp);
6406 			else
6407 				printf("nfsrv_cache_session: no session, "
6408 				    "check NFS clients for unique "
6409 				    "/etc/hostid's\n");
6410 			free(buf, M_TEMP);
6411 		}
6412 		m_freem(*m);
6413 		return;
6414 	}
6415 	nfsv4_seqsess_cacherep(nd->nd_slotid, sep->sess_slots, nd->nd_repstat,
6416 	    m);
6417 	NFSUNLOCKSESSION(shp);
6418 }
6419 
6420 /*
6421  * Search for a session that matches the sessionid.
6422  */
6423 static struct nfsdsession *
6424 nfsrv_findsession(uint8_t *sessionid)
6425 {
6426 	struct nfsdsession *sep;
6427 	struct nfssessionhash *shp;
6428 
6429 	shp = NFSSESSIONHASH(sessionid);
6430 	LIST_FOREACH(sep, &shp->list, sess_hash) {
6431 		if (!NFSBCMP(sessionid, sep->sess_sessionid, NFSX_V4SESSIONID))
6432 			break;
6433 	}
6434 	return (sep);
6435 }
6436 
6437 /*
6438  * Destroy a session.
6439  */
6440 int
6441 nfsrv_destroysession(struct nfsrv_descript *nd, uint8_t *sessionid)
6442 {
6443 	int error, igotlock, samesess;
6444 
6445 	samesess = 0;
6446 	if (!NFSBCMP(sessionid, nd->nd_sessionid, NFSX_V4SESSIONID) &&
6447 	    (nd->nd_flag & ND_HASSEQUENCE) != 0) {
6448 		samesess = 1;
6449 		if ((nd->nd_flag & ND_LASTOP) == 0)
6450 			return (NFSERR_BADSESSION);
6451 	}
6452 
6453 	/* Lock out other nfsd threads */
6454 	NFSLOCKV4ROOTMUTEX();
6455 	nfsv4_relref(&nfsv4rootfs_lock);
6456 	do {
6457 		igotlock = nfsv4_lock(&nfsv4rootfs_lock, 1, NULL,
6458 		    NFSV4ROOTLOCKMUTEXPTR, NULL);
6459 	} while (igotlock == 0);
6460 	NFSUNLOCKV4ROOTMUTEX();
6461 
6462 	error = nfsrv_freesession(nd, NULL, sessionid, false, NULL);
6463 	if (error == 0 && samesess != 0)
6464 		nd->nd_flag &= ~ND_HASSEQUENCE;
6465 
6466 	NFSLOCKV4ROOTMUTEX();
6467 	nfsv4_unlock(&nfsv4rootfs_lock, 1);
6468 	NFSUNLOCKV4ROOTMUTEX();
6469 	return (error);
6470 }
6471 
6472 /*
6473  * Bind a connection to a session.
6474  * For now, only certain variants are supported, since the current session
6475  * structure can only handle a single backchannel entry, which will be
6476  * applied to all connections if it is set.
6477  */
6478 int
6479 nfsrv_bindconnsess(struct nfsrv_descript *nd, uint8_t *sessionid, int *foreaftp)
6480 {
6481 	struct nfssessionhash *shp;
6482 	struct nfsdsession *sep;
6483 	struct nfsclient *clp;
6484 	SVCXPRT *savxprt;
6485 	int error;
6486 
6487 	error = 0;
6488 	savxprt = NULL;
6489 	shp = NFSSESSIONHASH(sessionid);
6490 	NFSLOCKSTATE();
6491 	NFSLOCKSESSION(shp);
6492 	sep = nfsrv_findsession(sessionid);
6493 	if (sep != NULL) {
6494 		clp = sep->sess_clp;
6495 		error = nfsrv_checkmachcred(NFSV4OP_BINDCONNTOSESS, nd, clp);
6496 		if (error != 0)
6497 			goto out;
6498 		if (*foreaftp == NFSCDFC4_BACK ||
6499 		    *foreaftp == NFSCDFC4_BACK_OR_BOTH ||
6500 		    *foreaftp == NFSCDFC4_FORE_OR_BOTH) {
6501 			/* Try to set up a backchannel. */
6502 			if (clp->lc_req.nr_client == NULL) {
6503 				NFSD_DEBUG(2, "nfsrv_bindconnsess: acquire "
6504 				    "backchannel\n");
6505 				clp->lc_req.nr_client = (struct __rpc_client *)
6506 				    clnt_bck_create(nd->nd_xprt->xp_socket,
6507 				    sep->sess_cbprogram, NFSV4_CBVERS);
6508 			}
6509 			if (clp->lc_req.nr_client != NULL) {
6510 				NFSD_DEBUG(2, "nfsrv_bindconnsess: set up "
6511 				    "backchannel\n");
6512 				savxprt = sep->sess_cbsess.nfsess_xprt;
6513 				SVC_ACQUIRE(nd->nd_xprt);
6514 				CLNT_ACQUIRE(clp->lc_req.nr_client);
6515 				nd->nd_xprt->xp_p2 = clp->lc_req.nr_client;
6516 				/* Disable idle timeout. */
6517 				nd->nd_xprt->xp_idletimeout = 0;
6518 				sep->sess_cbsess.nfsess_xprt = nd->nd_xprt;
6519 				sep->sess_crflags |= NFSV4CRSESS_CONNBACKCHAN;
6520 				clp->lc_flags |= LCL_DONEBINDCONN |
6521 				    LCL_NEEDSCBNULL;
6522 				clp->lc_flags &= ~LCL_CBDOWN;
6523 				if (*foreaftp == NFSCDFS4_BACK)
6524 					*foreaftp = NFSCDFS4_BACK;
6525 				else
6526 					*foreaftp = NFSCDFS4_BOTH;
6527 			} else if (*foreaftp != NFSCDFC4_BACK) {
6528 				NFSD_DEBUG(2, "nfsrv_bindconnsess: can't set "
6529 				    "up backchannel\n");
6530 				sep->sess_crflags &= ~NFSV4CRSESS_CONNBACKCHAN;
6531 				clp->lc_flags |= LCL_DONEBINDCONN;
6532 				*foreaftp = NFSCDFS4_FORE;
6533 			} else {
6534 				error = NFSERR_NOTSUPP;
6535 				printf("nfsrv_bindconnsess: Can't add "
6536 				    "backchannel\n");
6537 			}
6538 		} else {
6539 			NFSD_DEBUG(2, "nfsrv_bindconnsess: Set forechannel\n");
6540 			clp->lc_flags |= LCL_DONEBINDCONN;
6541 			*foreaftp = NFSCDFS4_FORE;
6542 		}
6543 	} else
6544 		error = NFSERR_BADSESSION;
6545 out:
6546 	NFSUNLOCKSESSION(shp);
6547 	NFSUNLOCKSTATE();
6548 	if (savxprt != NULL)
6549 		SVC_RELEASE(savxprt);
6550 	return (error);
6551 }
6552 
6553 /*
6554  * Free up a session structure.
6555  */
6556 static int
6557 nfsrv_freesession(struct nfsrv_descript *nd, struct nfsdsession *sep,
6558     uint8_t *sessionid, bool locked, SVCXPRT **old_xprtp)
6559 {
6560 	struct nfssessionhash *shp;
6561 	int i;
6562 
6563 	if (!locked)
6564 		NFSLOCKSTATE();
6565 	if (sep == NULL) {
6566 		shp = NFSSESSIONHASH(sessionid);
6567 		NFSLOCKSESSION(shp);
6568 		sep = nfsrv_findsession(sessionid);
6569 	} else {
6570 		shp = NFSSESSIONHASH(sep->sess_sessionid);
6571 		NFSLOCKSESSION(shp);
6572 	}
6573 	if (sep != NULL) {
6574 		/* Check for the SP4_MACH_CRED case. */
6575 		if (nd != NULL && nfsrv_checkmachcred(NFSV4OP_DESTROYSESSION,
6576 		    nd, sep->sess_clp) != 0) {
6577 			NFSUNLOCKSESSION(shp);
6578 			if (!locked)
6579 				NFSUNLOCKSTATE();
6580 			return (NFSERR_AUTHERR | AUTH_TOOWEAK);
6581 		}
6582 
6583 		sep->sess_refcnt--;
6584 		if (sep->sess_refcnt > 0) {
6585 			NFSUNLOCKSESSION(shp);
6586 			if (!locked)
6587 				NFSUNLOCKSTATE();
6588 			return (NFSERR_BACKCHANBUSY);
6589 		}
6590 		LIST_REMOVE(sep, sess_hash);
6591 		LIST_REMOVE(sep, sess_list);
6592 	}
6593 	NFSUNLOCKSESSION(shp);
6594 	if (!locked)
6595 		NFSUNLOCKSTATE();
6596 	if (sep == NULL)
6597 		return (NFSERR_BADSESSION);
6598 	for (i = 0; i < NFSV4_SLOTS; i++)
6599 		if (sep->sess_slots[i].nfssl_reply != NULL)
6600 			m_freem(sep->sess_slots[i].nfssl_reply);
6601 	if (!locked) {
6602 		if (sep->sess_cbsess.nfsess_xprt != NULL)
6603 			SVC_RELEASE(sep->sess_cbsess.nfsess_xprt);
6604 		if (old_xprtp != NULL)
6605 			*old_xprtp = NULL;
6606 	} else if (old_xprtp != NULL)
6607 		*old_xprtp = sep->sess_cbsess.nfsess_xprt;
6608 	free(sep, M_NFSDSESSION);
6609 	return (0);
6610 }
6611 
6612 /*
6613  * Free a stateid.
6614  * RFC5661 says that it should fail when there are associated opens, locks
6615  * or delegations. Since stateids represent opens, I don't see how you can
6616  * free an open stateid (it will be free'd when closed), so this function
6617  * only works for lock stateids (freeing the lock_owner) or delegations.
6618  */
6619 int
6620 nfsrv_freestateid(struct nfsrv_descript *nd, nfsv4stateid_t *stateidp,
6621     NFSPROC_T *p)
6622 {
6623 	struct nfsclient *clp;
6624 	struct nfsstate *stp;
6625 	int error;
6626 
6627 	NFSLOCKSTATE();
6628 	/*
6629 	 * Look up the stateid
6630 	 */
6631 	error = nfsrv_getclient((nfsquad_t)((u_quad_t)0), CLOPS_RENEW, &clp,
6632 	    NULL, (nfsquad_t)((u_quad_t)0), 0, nd, p);
6633 	if (error == 0) {
6634 		/* First, check for a delegation. */
6635 		LIST_FOREACH(stp, &clp->lc_deleg, ls_list) {
6636 			if (!NFSBCMP(stp->ls_stateid.other, stateidp->other,
6637 			    NFSX_STATEIDOTHER))
6638 				break;
6639 		}
6640 		if (stp != NULL) {
6641 			nfsrv_freedeleg(stp);
6642 			NFSUNLOCKSTATE();
6643 			return (error);
6644 		}
6645 	}
6646 	/* Not a delegation, try for a lock_owner. */
6647 	if (error == 0)
6648 		error = nfsrv_getstate(clp, stateidp, 0, &stp);
6649 	if (error == 0 && ((stp->ls_flags & (NFSLCK_OPEN | NFSLCK_DELEGREAD |
6650 	    NFSLCK_DELEGWRITE)) != 0 || (stp->ls_flags & NFSLCK_LOCK) == 0))
6651 		/* Not a lock_owner stateid. */
6652 		error = NFSERR_LOCKSHELD;
6653 	if (error == 0 && !LIST_EMPTY(&stp->ls_lock))
6654 		error = NFSERR_LOCKSHELD;
6655 	if (error == 0)
6656 		nfsrv_freelockowner(stp, NULL, 0, p);
6657 	NFSUNLOCKSTATE();
6658 	return (error);
6659 }
6660 
6661 /*
6662  * Test a stateid.
6663  */
6664 int
6665 nfsrv_teststateid(struct nfsrv_descript *nd, nfsv4stateid_t *stateidp,
6666     NFSPROC_T *p)
6667 {
6668 	struct nfsclient *clp;
6669 	struct nfsstate *stp;
6670 	int error;
6671 
6672 	NFSLOCKSTATE();
6673 	/*
6674 	 * Look up the stateid
6675 	 */
6676 	error = nfsrv_getclient((nfsquad_t)((u_quad_t)0), CLOPS_RENEW, &clp,
6677 	    NULL, (nfsquad_t)((u_quad_t)0), 0, nd, p);
6678 	if (error == 0)
6679 		error = nfsrv_getstate(clp, stateidp, 0, &stp);
6680 	if (error == 0 && stateidp->seqid != 0 &&
6681 	    SEQ_LT(stateidp->seqid, stp->ls_stateid.seqid))
6682 		error = NFSERR_OLDSTATEID;
6683 	NFSUNLOCKSTATE();
6684 	return (error);
6685 }
6686 
6687 /*
6688  * Generate the xdr for an NFSv4.1 CBSequence Operation.
6689  */
6690 static int
6691 nfsv4_setcbsequence(struct nfsrv_descript *nd, struct nfsclient *clp,
6692     int dont_replycache, struct nfsdsession **sepp, int *slotposp)
6693 {
6694 	struct nfsdsession *sep;
6695 	uint32_t *tl, slotseq = 0;
6696 	int maxslot;
6697 	uint8_t sessionid[NFSX_V4SESSIONID];
6698 	int error;
6699 
6700 	error = nfsv4_getcbsession(clp, sepp);
6701 	if (error != 0)
6702 		return (error);
6703 	sep = *sepp;
6704 	nfsv4_sequencelookup(NULL, &sep->sess_cbsess, slotposp, &maxslot,
6705 	    &slotseq, sessionid, true);
6706 	KASSERT(maxslot >= 0, ("nfsv4_setcbsequence neg maxslot"));
6707 
6708 	/* Build the Sequence arguments. */
6709 	NFSM_BUILD(tl, uint32_t *, NFSX_V4SESSIONID + 5 * NFSX_UNSIGNED);
6710 	bcopy(sessionid, tl, NFSX_V4SESSIONID);
6711 	tl += NFSX_V4SESSIONID / NFSX_UNSIGNED;
6712 	nd->nd_slotseq = tl;
6713 	nd->nd_slotid = *slotposp;
6714 	nd->nd_flag |= ND_HASSLOTID;
6715 	*tl++ = txdr_unsigned(slotseq);
6716 	*tl++ = txdr_unsigned(*slotposp);
6717 	*tl++ = txdr_unsigned(maxslot);
6718 	if (dont_replycache == 0)
6719 		*tl++ = newnfs_true;
6720 	else
6721 		*tl++ = newnfs_false;
6722 	*tl = 0;			/* No referring call list, for now. */
6723 	nd->nd_flag |= ND_HASSEQUENCE;
6724 	return (0);
6725 }
6726 
6727 /*
6728  * Get a session for the callback.
6729  */
6730 static int
6731 nfsv4_getcbsession(struct nfsclient *clp, struct nfsdsession **sepp)
6732 {
6733 	struct nfsdsession *sep;
6734 
6735 	NFSLOCKSTATE();
6736 	LIST_FOREACH(sep, &clp->lc_session, sess_list) {
6737 		if ((sep->sess_crflags & NFSV4CRSESS_CONNBACKCHAN) != 0)
6738 			break;
6739 	}
6740 	if (sep == NULL) {
6741 		NFSUNLOCKSTATE();
6742 		return (NFSERR_BADSESSION);
6743 	}
6744 	sep->sess_refcnt++;
6745 	*sepp = sep;
6746 	NFSUNLOCKSTATE();
6747 	return (0);
6748 }
6749 
6750 /*
6751  * Free up all backchannel xprts.  This needs to be done when the nfsd threads
6752  * exit, since those transports will all be going away.
6753  * This is only called after all the nfsd threads are done performing RPCs,
6754  * so locking shouldn't be an issue.
6755  */
6756 void
6757 nfsrv_freeallbackchannel_xprts(void)
6758 {
6759 	struct nfsdsession *sep;
6760 	struct nfsclient *clp;
6761 	SVCXPRT *xprt;
6762 	int i;
6763 
6764 	for (i = 0; i < nfsrv_clienthashsize; i++) {
6765 		LIST_FOREACH(clp, &NFSD_VNET(nfsclienthash)[i], lc_hash) {
6766 			LIST_FOREACH(sep, &clp->lc_session, sess_list) {
6767 				xprt = sep->sess_cbsess.nfsess_xprt;
6768 				sep->sess_cbsess.nfsess_xprt = NULL;
6769 				if (xprt != NULL)
6770 					SVC_RELEASE(xprt);
6771 			}
6772 		}
6773 	}
6774 }
6775 
6776 /*
6777  * Do a layout commit.  Actually just call nfsrv_updatemdsattr().
6778  * I have no idea if the rest of these arguments will ever be useful?
6779  */
6780 int
6781 nfsrv_layoutcommit(struct nfsrv_descript *nd, vnode_t vp, int layouttype,
6782     int hasnewoff, uint64_t newoff, uint64_t offset, uint64_t len,
6783     int hasnewmtime, struct timespec *newmtimep, int reclaim,
6784     nfsv4stateid_t *stateidp, int maxcnt, char *layp, int *hasnewsizep,
6785     uint64_t *newsizep, struct ucred *cred, NFSPROC_T *p)
6786 {
6787 	struct nfsvattr na;
6788 	int error;
6789 
6790 	error = nfsrv_updatemdsattr(vp, &na, p);
6791 	if (error == 0) {
6792 		*hasnewsizep = 1;
6793 		*newsizep = na.na_size;
6794 	}
6795 	return (error);
6796 }
6797 
6798 /*
6799  * Try and get a layout.
6800  */
6801 int
6802 nfsrv_layoutget(struct nfsrv_descript *nd, vnode_t vp, struct nfsexstuff *exp,
6803     int layouttype, int *iomode, uint64_t *offset, uint64_t *len,
6804     uint64_t minlen, nfsv4stateid_t *stateidp, int maxcnt, int *retonclose,
6805     int *layoutlenp, char *layp, struct ucred *cred, NFSPROC_T *p)
6806 {
6807 	struct nfslayouthash *lhyp;
6808 	struct nfslayout *lyp;
6809 	char *devid;
6810 	fhandle_t fh, *dsfhp;
6811 	int error, mirrorcnt;
6812 
6813 	if (nfsrv_devidcnt == 0)
6814 		return (NFSERR_UNKNLAYOUTTYPE);
6815 
6816 	if (*offset != 0)
6817 		printf("nfsrv_layoutget: off=%ju len=%ju\n", (uintmax_t)*offset,
6818 		    (uintmax_t)*len);
6819 	error = nfsvno_getfh(vp, &fh, p);
6820 	NFSD_DEBUG(4, "layoutget getfh=%d\n", error);
6821 	if (error != 0)
6822 		return (error);
6823 
6824 	/*
6825 	 * For now, all layouts are for entire files.
6826 	 * Only issue Read/Write layouts if requested for a non-readonly fs.
6827 	 */
6828 	if (NFSVNO_EXRDONLY(exp)) {
6829 		if (*iomode == NFSLAYOUTIOMODE_RW)
6830 			return (NFSERR_LAYOUTTRYLATER);
6831 		*iomode = NFSLAYOUTIOMODE_READ;
6832 	}
6833 	if (*iomode != NFSLAYOUTIOMODE_RW)
6834 		*iomode = NFSLAYOUTIOMODE_READ;
6835 
6836 	/*
6837 	 * Check to see if a write layout can be issued for this file.
6838 	 * This is used during mirror recovery to avoid RW layouts being
6839 	 * issued for a file while it is being copied to the recovered
6840 	 * mirror.
6841 	 */
6842 	if (*iomode == NFSLAYOUTIOMODE_RW && nfsrv_dontlayout(&fh) != 0)
6843 		return (NFSERR_LAYOUTTRYLATER);
6844 
6845 	*retonclose = 0;
6846 	*offset = 0;
6847 	*len = UINT64_MAX;
6848 
6849 	/* First, see if a layout already exists and return if found. */
6850 	lhyp = NFSLAYOUTHASH(&fh);
6851 	NFSLOCKLAYOUT(lhyp);
6852 	error = nfsrv_findlayout(&nd->nd_clientid, &fh, layouttype, p, &lyp);
6853 	NFSD_DEBUG(4, "layoutget findlay=%d\n", error);
6854 	/*
6855 	 * Not sure if the seqid must be the same, so I won't check it.
6856 	 */
6857 	if (error == 0 && (stateidp->other[0] != lyp->lay_stateid.other[0] ||
6858 	    stateidp->other[1] != lyp->lay_stateid.other[1] ||
6859 	    stateidp->other[2] != lyp->lay_stateid.other[2])) {
6860 		if ((lyp->lay_flags & NFSLAY_CALLB) == 0) {
6861 			NFSUNLOCKLAYOUT(lhyp);
6862 			NFSD_DEBUG(1, "ret bad stateid\n");
6863 			return (NFSERR_BADSTATEID);
6864 		}
6865 		/*
6866 		 * I believe we get here because there is a race between
6867 		 * the client processing the CBLAYOUTRECALL and the layout
6868 		 * being deleted here on the server.
6869 		 * The client has now done a LayoutGet with a non-layout
6870 		 * stateid, as it would when there is no layout.
6871 		 * As such, free this layout and set error == NFSERR_BADSTATEID
6872 		 * so the code below will create a new layout structure as
6873 		 * would happen if no layout was found.
6874 		 * "lyp" will be set before being used below, but set it NULL
6875 		 * as a safety belt.
6876 		 */
6877 		nfsrv_freelayout(&lhyp->list, lyp);
6878 		lyp = NULL;
6879 		error = NFSERR_BADSTATEID;
6880 	}
6881 	if (error == 0) {
6882 		if (lyp->lay_layoutlen > maxcnt) {
6883 			NFSUNLOCKLAYOUT(lhyp);
6884 			NFSD_DEBUG(1, "ret layout too small\n");
6885 			return (NFSERR_TOOSMALL);
6886 		}
6887 		if (*iomode == NFSLAYOUTIOMODE_RW) {
6888 			if ((lyp->lay_flags & NFSLAY_NOSPC) != 0) {
6889 				NFSUNLOCKLAYOUT(lhyp);
6890 				NFSD_DEBUG(1, "ret layout nospace\n");
6891 				return (NFSERR_NOSPC);
6892 			}
6893 			lyp->lay_flags |= NFSLAY_RW;
6894 		} else
6895 			lyp->lay_flags |= NFSLAY_READ;
6896 		NFSBCOPY(lyp->lay_xdr, layp, lyp->lay_layoutlen);
6897 		*layoutlenp = lyp->lay_layoutlen;
6898 		if (++lyp->lay_stateid.seqid == 0)
6899 			lyp->lay_stateid.seqid = 1;
6900 		stateidp->seqid = lyp->lay_stateid.seqid;
6901 		NFSUNLOCKLAYOUT(lhyp);
6902 		NFSD_DEBUG(4, "ret fnd layout\n");
6903 		return (0);
6904 	}
6905 	NFSUNLOCKLAYOUT(lhyp);
6906 
6907 	/* Find the device id and file handle. */
6908 	dsfhp = malloc(sizeof(fhandle_t) * NFSDEV_MAXMIRRORS, M_TEMP, M_WAITOK);
6909 	devid = malloc(NFSX_V4DEVICEID * NFSDEV_MAXMIRRORS, M_TEMP, M_WAITOK);
6910 	error = nfsrv_dsgetdevandfh(vp, p, &mirrorcnt, dsfhp, devid);
6911 	NFSD_DEBUG(4, "layoutget devandfh=%d\n", error);
6912 	if (error == 0) {
6913 		if (layouttype == NFSLAYOUT_NFSV4_1_FILES) {
6914 			if (NFSX_V4FILELAYOUT > maxcnt)
6915 				error = NFSERR_TOOSMALL;
6916 			else
6917 				lyp = nfsrv_filelayout(nd, *iomode, &fh, dsfhp,
6918 				    devid, vp->v_mount->mnt_stat.f_fsid);
6919 		} else {
6920 			if (NFSX_V4FLEXLAYOUT(mirrorcnt) > maxcnt)
6921 				error = NFSERR_TOOSMALL;
6922 			else
6923 				lyp = nfsrv_flexlayout(nd, *iomode, mirrorcnt,
6924 				    &fh, dsfhp, devid,
6925 				    vp->v_mount->mnt_stat.f_fsid);
6926 		}
6927 	}
6928 	free(dsfhp, M_TEMP);
6929 	free(devid, M_TEMP);
6930 	if (error != 0)
6931 		return (error);
6932 
6933 	/*
6934 	 * Now, add this layout to the list.
6935 	 */
6936 	error = nfsrv_addlayout(nd, &lyp, stateidp, layp, layoutlenp, p);
6937 	NFSD_DEBUG(4, "layoutget addl=%d\n", error);
6938 	/*
6939 	 * The lyp will be set to NULL by nfsrv_addlayout() if it
6940 	 * linked the new structure into the lists.
6941 	 */
6942 	free(lyp, M_NFSDSTATE);
6943 	return (error);
6944 }
6945 
6946 /*
6947  * Generate a File Layout.
6948  */
6949 static struct nfslayout *
6950 nfsrv_filelayout(struct nfsrv_descript *nd, int iomode, fhandle_t *fhp,
6951     fhandle_t *dsfhp, char *devid, fsid_t fs)
6952 {
6953 	uint32_t *tl;
6954 	struct nfslayout *lyp;
6955 	uint64_t pattern_offset;
6956 
6957 	lyp = malloc(sizeof(struct nfslayout) + NFSX_V4FILELAYOUT, M_NFSDSTATE,
6958 	    M_WAITOK | M_ZERO);
6959 	lyp->lay_type = NFSLAYOUT_NFSV4_1_FILES;
6960 	if (iomode == NFSLAYOUTIOMODE_RW)
6961 		lyp->lay_flags = NFSLAY_RW;
6962 	else
6963 		lyp->lay_flags = NFSLAY_READ;
6964 	NFSBCOPY(fhp, &lyp->lay_fh, sizeof(*fhp));
6965 	lyp->lay_clientid.qval = nd->nd_clientid.qval;
6966 	lyp->lay_fsid = fs;
6967 	NFSBCOPY(devid, lyp->lay_deviceid, NFSX_V4DEVICEID);
6968 
6969 	/* Fill in the xdr for the files layout. */
6970 	tl = (uint32_t *)lyp->lay_xdr;
6971 	NFSBCOPY(devid, tl, NFSX_V4DEVICEID);		/* Device ID. */
6972 	tl += (NFSX_V4DEVICEID / NFSX_UNSIGNED);
6973 
6974 	/* Set the stripe size to the maximum I/O size. */
6975 	*tl++ = txdr_unsigned(nfs_srvmaxio & NFSFLAYUTIL_STRIPE_MASK);
6976 	*tl++ = 0;					/* 1st stripe index. */
6977 	pattern_offset = 0;
6978 	txdr_hyper(pattern_offset, tl); tl += 2;	/* Pattern offset. */
6979 	*tl++ = txdr_unsigned(1);			/* 1 file handle. */
6980 	*tl++ = txdr_unsigned(NFSX_V4PNFSFH);
6981 	NFSBCOPY(dsfhp, tl, sizeof(*dsfhp));
6982 	lyp->lay_layoutlen = NFSX_V4FILELAYOUT;
6983 	return (lyp);
6984 }
6985 
6986 #define	FLEX_OWNERID	"999"
6987 #define	FLEX_UID0	"0"
6988 /*
6989  * Generate a Flex File Layout.
6990  * The FLEX_OWNERID can be any string of 3 decimal digits. Although this
6991  * string goes on the wire, it isn't supposed to be used by the client,
6992  * since this server uses tight coupling.
6993  * Although not recommended by the spec., if vfs.nfsd.flexlinuxhack=1 use
6994  * a string of "0". This works around the Linux Flex File Layout driver bug
6995  * which uses the synthetic uid/gid strings for the "tightly coupled" case.
6996  */
6997 static struct nfslayout *
6998 nfsrv_flexlayout(struct nfsrv_descript *nd, int iomode, int mirrorcnt,
6999     fhandle_t *fhp, fhandle_t *dsfhp, char *devid, fsid_t fs)
7000 {
7001 	uint32_t *tl;
7002 	struct nfslayout *lyp;
7003 	uint64_t lenval;
7004 	int i;
7005 
7006 	lyp = malloc(sizeof(struct nfslayout) + NFSX_V4FLEXLAYOUT(mirrorcnt),
7007 	    M_NFSDSTATE, M_WAITOK | M_ZERO);
7008 	lyp->lay_type = NFSLAYOUT_FLEXFILE;
7009 	if (iomode == NFSLAYOUTIOMODE_RW)
7010 		lyp->lay_flags = NFSLAY_RW;
7011 	else
7012 		lyp->lay_flags = NFSLAY_READ;
7013 	NFSBCOPY(fhp, &lyp->lay_fh, sizeof(*fhp));
7014 	lyp->lay_clientid.qval = nd->nd_clientid.qval;
7015 	lyp->lay_fsid = fs;
7016 	lyp->lay_mirrorcnt = mirrorcnt;
7017 	NFSBCOPY(devid, lyp->lay_deviceid, NFSX_V4DEVICEID);
7018 
7019 	/* Fill in the xdr for the files layout. */
7020 	tl = (uint32_t *)lyp->lay_xdr;
7021 	lenval = 0;
7022 	txdr_hyper(lenval, tl); tl += 2;		/* Stripe unit. */
7023 	*tl++ = txdr_unsigned(mirrorcnt);		/* # of mirrors. */
7024 	for (i = 0; i < mirrorcnt; i++) {
7025 		*tl++ = txdr_unsigned(1);		/* One stripe. */
7026 		NFSBCOPY(devid, tl, NFSX_V4DEVICEID);	/* Device ID. */
7027 		tl += (NFSX_V4DEVICEID / NFSX_UNSIGNED);
7028 		devid += NFSX_V4DEVICEID;
7029 		*tl++ = txdr_unsigned(1);		/* Efficiency. */
7030 		*tl++ = 0;				/* Proxy Stateid. */
7031 		*tl++ = 0x55555555;
7032 		*tl++ = 0x55555555;
7033 		*tl++ = 0x55555555;
7034 		*tl++ = txdr_unsigned(1);		/* 1 file handle. */
7035 		*tl++ = txdr_unsigned(NFSX_V4PNFSFH);
7036 		NFSBCOPY(dsfhp, tl, sizeof(*dsfhp));
7037 		tl += (NFSM_RNDUP(NFSX_V4PNFSFH) / NFSX_UNSIGNED);
7038 		dsfhp++;
7039 		if (nfsrv_flexlinuxhack != 0) {
7040 			*tl++ = txdr_unsigned(strlen(FLEX_UID0));
7041 			*tl = 0;		/* 0 pad string. */
7042 			NFSBCOPY(FLEX_UID0, tl++, strlen(FLEX_UID0));
7043 			*tl++ = txdr_unsigned(strlen(FLEX_UID0));
7044 			*tl = 0;		/* 0 pad string. */
7045 			NFSBCOPY(FLEX_UID0, tl++, strlen(FLEX_UID0));
7046 		} else {
7047 			*tl++ = txdr_unsigned(strlen(FLEX_OWNERID));
7048 			NFSBCOPY(FLEX_OWNERID, tl++, NFSX_UNSIGNED);
7049 			*tl++ = txdr_unsigned(strlen(FLEX_OWNERID));
7050 			NFSBCOPY(FLEX_OWNERID, tl++, NFSX_UNSIGNED);
7051 		}
7052 	}
7053 	*tl++ = txdr_unsigned(0);		/* ff_flags. */
7054 	*tl = txdr_unsigned(60);		/* Status interval hint. */
7055 	lyp->lay_layoutlen = NFSX_V4FLEXLAYOUT(mirrorcnt);
7056 	return (lyp);
7057 }
7058 
7059 /*
7060  * Parse and process Flex File errors returned via LayoutReturn.
7061  */
7062 static void
7063 nfsrv_flexlayouterr(struct nfsrv_descript *nd, uint32_t *layp, int maxcnt,
7064     NFSPROC_T *p)
7065 {
7066 	uint32_t *tl;
7067 	int cnt, errcnt, i, j, opnum, stat;
7068 	char devid[NFSX_V4DEVICEID];
7069 
7070 	tl = layp;
7071 	maxcnt -= NFSX_UNSIGNED;
7072 	if (maxcnt > 0)
7073 		cnt = fxdr_unsigned(int, *tl++);
7074 	else
7075 		cnt = 0;
7076 	NFSD_DEBUG(4, "flexlayouterr cnt=%d\n", cnt);
7077 	for (i = 0; i < cnt; i++) {
7078 		maxcnt -= NFSX_STATEID + 2 * NFSX_HYPER +
7079 		    NFSX_UNSIGNED;
7080 		if (maxcnt <= 0)
7081 			break;
7082 		/* Skip offset, length and stateid for now. */
7083 		tl += (4 + NFSX_STATEID / NFSX_UNSIGNED);
7084 		errcnt = fxdr_unsigned(int, *tl++);
7085 		NFSD_DEBUG(4, "flexlayouterr errcnt=%d\n", errcnt);
7086 		for (j = 0; j < errcnt; j++) {
7087 			maxcnt -= NFSX_V4DEVICEID + 2 * NFSX_UNSIGNED;
7088 			if (maxcnt < 0)
7089 				break;
7090 			NFSBCOPY(tl, devid, NFSX_V4DEVICEID);
7091 			tl += (NFSX_V4DEVICEID / NFSX_UNSIGNED);
7092 			stat = fxdr_unsigned(int, *tl++);
7093 			opnum = fxdr_unsigned(int, *tl++);
7094 			NFSD_DEBUG(4, "flexlayouterr op=%d stat=%d\n", opnum,
7095 			    stat);
7096 			/*
7097 			 * Except for NFSERR_ACCES, NFSERR_STALE and
7098 			 * NFSERR_NOSPC errors, disable the mirror.
7099 			 */
7100 			if (stat != NFSERR_ACCES && stat != NFSERR_STALE &&
7101 			    stat != NFSERR_NOSPC)
7102 				nfsrv_delds(devid, p);
7103 
7104 			/* For NFSERR_NOSPC, mark all devids and layouts. */
7105 			if (stat == NFSERR_NOSPC)
7106 				nfsrv_marknospc(devid, true);
7107 		}
7108 	}
7109 }
7110 
7111 /*
7112  * This function removes all flex file layouts which has a mirror with
7113  * a device id that matches the argument.
7114  * Called when the DS represented by the device id has failed.
7115  */
7116 void
7117 nfsrv_flexmirrordel(char *devid, NFSPROC_T *p)
7118 {
7119 	uint32_t *tl;
7120 	struct nfslayout *lyp, *nlyp;
7121 	struct nfslayouthash *lhyp;
7122 	struct nfslayouthead loclyp;
7123 	int i, j;
7124 
7125 	NFSD_DEBUG(4, "flexmirrordel\n");
7126 	/* Move all layouts found onto a local list. */
7127 	TAILQ_INIT(&loclyp);
7128 	for (i = 0; i < nfsrv_layouthashsize; i++) {
7129 		lhyp = &nfslayouthash[i];
7130 		NFSLOCKLAYOUT(lhyp);
7131 		TAILQ_FOREACH_SAFE(lyp, &lhyp->list, lay_list, nlyp) {
7132 			if (lyp->lay_type == NFSLAYOUT_FLEXFILE &&
7133 			    lyp->lay_mirrorcnt > 1) {
7134 				NFSD_DEBUG(4, "possible match\n");
7135 				tl = lyp->lay_xdr;
7136 				tl += 3;
7137 				for (j = 0; j < lyp->lay_mirrorcnt; j++) {
7138 					tl++;
7139 					if (NFSBCMP(devid, tl, NFSX_V4DEVICEID)
7140 					    == 0) {
7141 						/* Found one. */
7142 						NFSD_DEBUG(4, "fnd one\n");
7143 						TAILQ_REMOVE(&lhyp->list, lyp,
7144 						    lay_list);
7145 						TAILQ_INSERT_HEAD(&loclyp, lyp,
7146 						    lay_list);
7147 						break;
7148 					}
7149 					tl += (NFSX_V4DEVICEID / NFSX_UNSIGNED +
7150 					    NFSM_RNDUP(NFSX_V4PNFSFH) /
7151 					    NFSX_UNSIGNED + 11 * NFSX_UNSIGNED);
7152 				}
7153 			}
7154 		}
7155 		NFSUNLOCKLAYOUT(lhyp);
7156 	}
7157 
7158 	/* Now, try to do a Layout recall for each one found. */
7159 	TAILQ_FOREACH_SAFE(lyp, &loclyp, lay_list, nlyp) {
7160 		NFSD_DEBUG(4, "do layout recall\n");
7161 		/*
7162 		 * The layout stateid.seqid needs to be incremented
7163 		 * before doing a LAYOUT_RECALL callback.
7164 		 */
7165 		if (++lyp->lay_stateid.seqid == 0)
7166 			lyp->lay_stateid.seqid = 1;
7167 		nfsrv_recalllayout(lyp->lay_clientid, &lyp->lay_stateid,
7168 		    &lyp->lay_fh, lyp, 1, lyp->lay_type, p);
7169 		nfsrv_freelayout(&loclyp, lyp);
7170 	}
7171 }
7172 
7173 /*
7174  * Do a recall callback to the client for this layout.
7175  */
7176 static int
7177 nfsrv_recalllayout(nfsquad_t clid, nfsv4stateid_t *stateidp, fhandle_t *fhp,
7178     struct nfslayout *lyp, int changed, int laytype, NFSPROC_T *p)
7179 {
7180 	struct nfsclient *clp;
7181 	int error;
7182 
7183 	NFSD_DEBUG(4, "nfsrv_recalllayout\n");
7184 	error = nfsrv_getclient(clid, 0, &clp, NULL, (nfsquad_t)((u_quad_t)0),
7185 	    0, NULL, p);
7186 	NFSD_DEBUG(4, "aft nfsrv_getclient=%d\n", error);
7187 	if (error != 0) {
7188 		printf("nfsrv_recalllayout: getclient err=%d\n", error);
7189 		return (error);
7190 	}
7191 	if ((clp->lc_flags & LCL_NFSV41) != 0) {
7192 		error = nfsrv_docallback(clp, NFSV4OP_CBLAYOUTRECALL,
7193 		    stateidp, changed, fhp, NULL, NULL, laytype, p);
7194 		/* If lyp != NULL, handle an error return here. */
7195 		if (error != 0 && lyp != NULL) {
7196 			NFSDRECALLLOCK();
7197 			/*
7198 			 * Mark it returned, since no layout recall
7199 			 * has been done.
7200 			 * All errors seem to be non-recoverable, although
7201 			 * NFSERR_NOMATCHLAYOUT is a normal event.
7202 			 */
7203 			if ((lyp->lay_flags & NFSLAY_RECALL) != 0) {
7204 				lyp->lay_flags |= NFSLAY_RETURNED;
7205 				wakeup(lyp);
7206 			}
7207 			NFSDRECALLUNLOCK();
7208 			if (error != NFSERR_NOMATCHLAYOUT)
7209 				printf("nfsrv_recalllayout: err=%d\n", error);
7210 		}
7211 	} else
7212 		printf("nfsrv_recalllayout: clp not NFSv4.1\n");
7213 	return (error);
7214 }
7215 
7216 /*
7217  * Find a layout to recall when we exceed our high water mark.
7218  */
7219 void
7220 nfsrv_recalloldlayout(NFSPROC_T *p)
7221 {
7222 	struct nfslayouthash *lhyp;
7223 	struct nfslayout *lyp;
7224 	nfsquad_t clientid;
7225 	nfsv4stateid_t stateid;
7226 	fhandle_t fh;
7227 	int error, laytype = 0, ret;
7228 
7229 	lhyp = &nfslayouthash[arc4random() % nfsrv_layouthashsize];
7230 	NFSLOCKLAYOUT(lhyp);
7231 	TAILQ_FOREACH_REVERSE(lyp, &lhyp->list, nfslayouthead, lay_list) {
7232 		if ((lyp->lay_flags & NFSLAY_CALLB) == 0) {
7233 			lyp->lay_flags |= NFSLAY_CALLB;
7234 			/*
7235 			 * The layout stateid.seqid needs to be incremented
7236 			 * before doing a LAYOUT_RECALL callback.
7237 			 */
7238 			if (++lyp->lay_stateid.seqid == 0)
7239 				lyp->lay_stateid.seqid = 1;
7240 			clientid = lyp->lay_clientid;
7241 			stateid = lyp->lay_stateid;
7242 			NFSBCOPY(&lyp->lay_fh, &fh, sizeof(fh));
7243 			laytype = lyp->lay_type;
7244 			break;
7245 		}
7246 	}
7247 	NFSUNLOCKLAYOUT(lhyp);
7248 	if (lyp != NULL) {
7249 		error = nfsrv_recalllayout(clientid, &stateid, &fh, NULL, 0,
7250 		    laytype, p);
7251 		if (error != 0 && error != NFSERR_NOMATCHLAYOUT)
7252 			NFSD_DEBUG(4, "recallold=%d\n", error);
7253 		if (error != 0) {
7254 			NFSLOCKLAYOUT(lhyp);
7255 			/*
7256 			 * Since the hash list was unlocked, we need to
7257 			 * find it again.
7258 			 */
7259 			ret = nfsrv_findlayout(&clientid, &fh, laytype, p,
7260 			    &lyp);
7261 			if (ret == 0 &&
7262 			    (lyp->lay_flags & NFSLAY_CALLB) != 0 &&
7263 			    lyp->lay_stateid.other[0] == stateid.other[0] &&
7264 			    lyp->lay_stateid.other[1] == stateid.other[1] &&
7265 			    lyp->lay_stateid.other[2] == stateid.other[2]) {
7266 				/*
7267 				 * The client no longer knows this layout, so
7268 				 * it can be free'd now.
7269 				 */
7270 				if (error == NFSERR_NOMATCHLAYOUT)
7271 					nfsrv_freelayout(&lhyp->list, lyp);
7272 				else {
7273 					/*
7274 					 * Leave it to be tried later by
7275 					 * clearing NFSLAY_CALLB and moving
7276 					 * it to the head of the list, so it
7277 					 * won't be tried again for a while.
7278 					 */
7279 					lyp->lay_flags &= ~NFSLAY_CALLB;
7280 					TAILQ_REMOVE(&lhyp->list, lyp,
7281 					    lay_list);
7282 					TAILQ_INSERT_HEAD(&lhyp->list, lyp,
7283 					    lay_list);
7284 				}
7285 			}
7286 			NFSUNLOCKLAYOUT(lhyp);
7287 		}
7288 	}
7289 }
7290 
7291 /*
7292  * Try and return layout(s).
7293  */
7294 int
7295 nfsrv_layoutreturn(struct nfsrv_descript *nd, vnode_t vp,
7296     int layouttype, int iomode, uint64_t offset, uint64_t len, int reclaim,
7297     int kind, nfsv4stateid_t *stateidp, int maxcnt, uint32_t *layp, int *fndp,
7298     struct ucred *cred, NFSPROC_T *p)
7299 {
7300 	struct nfsvattr na;
7301 	struct nfslayouthash *lhyp;
7302 	struct nfslayout *lyp;
7303 	fhandle_t fh;
7304 	int error = 0;
7305 
7306 	*fndp = 0;
7307 	if (kind == NFSV4LAYOUTRET_FILE) {
7308 		error = nfsvno_getfh(vp, &fh, p);
7309 		if (error == 0) {
7310 			error = nfsrv_updatemdsattr(vp, &na, p);
7311 			if (error != 0)
7312 				printf("nfsrv_layoutreturn: updatemdsattr"
7313 				    " failed=%d\n", error);
7314 		}
7315 		if (error == 0) {
7316 			if (reclaim == newnfs_true) {
7317 				error = nfsrv_checkgrace(NULL, NULL,
7318 				    NFSLCK_RECLAIM);
7319 				if (error != NFSERR_NOGRACE)
7320 					error = 0;
7321 				return (error);
7322 			}
7323 			lhyp = NFSLAYOUTHASH(&fh);
7324 			NFSDRECALLLOCK();
7325 			NFSLOCKLAYOUT(lhyp);
7326 			error = nfsrv_findlayout(&nd->nd_clientid, &fh,
7327 			    layouttype, p, &lyp);
7328 			NFSD_DEBUG(4, "layoutret findlay=%d\n", error);
7329 			if (error == 0 &&
7330 			    stateidp->other[0] == lyp->lay_stateid.other[0] &&
7331 			    stateidp->other[1] == lyp->lay_stateid.other[1] &&
7332 			    stateidp->other[2] == lyp->lay_stateid.other[2]) {
7333 				NFSD_DEBUG(4, "nfsrv_layoutreturn: stateid %d"
7334 				    " %x %x %x laystateid %d %x %x %x"
7335 				    " off=%ju len=%ju flgs=0x%x\n",
7336 				    stateidp->seqid, stateidp->other[0],
7337 				    stateidp->other[1], stateidp->other[2],
7338 				    lyp->lay_stateid.seqid,
7339 				    lyp->lay_stateid.other[0],
7340 				    lyp->lay_stateid.other[1],
7341 				    lyp->lay_stateid.other[2],
7342 				    (uintmax_t)offset, (uintmax_t)len,
7343 				    lyp->lay_flags);
7344 				if (++lyp->lay_stateid.seqid == 0)
7345 					lyp->lay_stateid.seqid = 1;
7346 				stateidp->seqid = lyp->lay_stateid.seqid;
7347 				if (offset == 0 && len == UINT64_MAX) {
7348 					if ((iomode & NFSLAYOUTIOMODE_READ) !=
7349 					    0)
7350 						lyp->lay_flags &= ~NFSLAY_READ;
7351 					if ((iomode & NFSLAYOUTIOMODE_RW) != 0)
7352 						lyp->lay_flags &= ~NFSLAY_RW;
7353 					if ((lyp->lay_flags & (NFSLAY_READ |
7354 					    NFSLAY_RW)) == 0)
7355 						nfsrv_freelayout(&lhyp->list,
7356 						    lyp);
7357 					else
7358 						*fndp = 1;
7359 				} else
7360 					*fndp = 1;
7361 			}
7362 			NFSUNLOCKLAYOUT(lhyp);
7363 			/* Search the nfsrv_recalllist for a match. */
7364 			TAILQ_FOREACH(lyp, &nfsrv_recalllisthead, lay_list) {
7365 				if (NFSBCMP(&lyp->lay_fh, &fh,
7366 				    sizeof(fh)) == 0 &&
7367 				    lyp->lay_clientid.qval ==
7368 				    nd->nd_clientid.qval &&
7369 				    stateidp->other[0] ==
7370 				    lyp->lay_stateid.other[0] &&
7371 				    stateidp->other[1] ==
7372 				    lyp->lay_stateid.other[1] &&
7373 				    stateidp->other[2] ==
7374 				    lyp->lay_stateid.other[2]) {
7375 					lyp->lay_flags |= NFSLAY_RETURNED;
7376 					wakeup(lyp);
7377 					error = 0;
7378 				}
7379 			}
7380 			NFSDRECALLUNLOCK();
7381 		}
7382 		if (layouttype == NFSLAYOUT_FLEXFILE && layp != NULL)
7383 			nfsrv_flexlayouterr(nd, layp, maxcnt, p);
7384 	} else if (kind == NFSV4LAYOUTRET_FSID)
7385 		nfsrv_freelayouts(&nd->nd_clientid,
7386 		    &vp->v_mount->mnt_stat.f_fsid, layouttype, iomode);
7387 	else if (kind == NFSV4LAYOUTRET_ALL)
7388 		nfsrv_freelayouts(&nd->nd_clientid, NULL, layouttype, iomode);
7389 	else
7390 		error = NFSERR_INVAL;
7391 	if (error == -1)
7392 		error = 0;
7393 	return (error);
7394 }
7395 
7396 /*
7397  * Look for an existing layout.
7398  */
7399 static int
7400 nfsrv_findlayout(nfsquad_t *clientidp, fhandle_t *fhp, int laytype,
7401     NFSPROC_T *p, struct nfslayout **lypp)
7402 {
7403 	struct nfslayouthash *lhyp;
7404 	struct nfslayout *lyp;
7405 	int ret;
7406 
7407 	*lypp = NULL;
7408 	ret = 0;
7409 	lhyp = NFSLAYOUTHASH(fhp);
7410 	TAILQ_FOREACH(lyp, &lhyp->list, lay_list) {
7411 		if (NFSBCMP(&lyp->lay_fh, fhp, sizeof(*fhp)) == 0 &&
7412 		    lyp->lay_clientid.qval == clientidp->qval &&
7413 		    lyp->lay_type == laytype)
7414 			break;
7415 	}
7416 	if (lyp != NULL)
7417 		*lypp = lyp;
7418 	else
7419 		ret = -1;
7420 	return (ret);
7421 }
7422 
7423 /*
7424  * Add the new layout, as required.
7425  */
7426 static int
7427 nfsrv_addlayout(struct nfsrv_descript *nd, struct nfslayout **lypp,
7428     nfsv4stateid_t *stateidp, char *layp, int *layoutlenp, NFSPROC_T *p)
7429 {
7430 	struct nfsclient *clp;
7431 	struct nfslayouthash *lhyp;
7432 	struct nfslayout *lyp, *nlyp;
7433 	fhandle_t *fhp;
7434 	int error;
7435 
7436 	KASSERT((nd->nd_flag & ND_IMPLIEDCLID) != 0,
7437 	    ("nfsrv_layoutget: no nd_clientid\n"));
7438 	lyp = *lypp;
7439 	fhp = &lyp->lay_fh;
7440 	NFSLOCKSTATE();
7441 	error = nfsrv_getclient((nfsquad_t)((u_quad_t)0), CLOPS_RENEW, &clp,
7442 	    NULL, (nfsquad_t)((u_quad_t)0), 0, nd, p);
7443 	if (error != 0) {
7444 		NFSUNLOCKSTATE();
7445 		return (error);
7446 	}
7447 	lyp->lay_stateid.seqid = stateidp->seqid = 1;
7448 	lyp->lay_stateid.other[0] = stateidp->other[0] =
7449 	    clp->lc_clientid.lval[0];
7450 	lyp->lay_stateid.other[1] = stateidp->other[1] =
7451 	    clp->lc_clientid.lval[1];
7452 	lyp->lay_stateid.other[2] = stateidp->other[2] =
7453 	    nfsrv_nextstateindex(clp);
7454 	NFSUNLOCKSTATE();
7455 
7456 	lhyp = NFSLAYOUTHASH(fhp);
7457 	NFSLOCKLAYOUT(lhyp);
7458 	TAILQ_FOREACH(nlyp, &lhyp->list, lay_list) {
7459 		if (NFSBCMP(&nlyp->lay_fh, fhp, sizeof(*fhp)) == 0 &&
7460 		    nlyp->lay_clientid.qval == nd->nd_clientid.qval)
7461 			break;
7462 	}
7463 	if (nlyp != NULL) {
7464 		/* A layout already exists, so use it. */
7465 		nlyp->lay_flags |= (lyp->lay_flags & (NFSLAY_READ | NFSLAY_RW));
7466 		NFSBCOPY(nlyp->lay_xdr, layp, nlyp->lay_layoutlen);
7467 		*layoutlenp = nlyp->lay_layoutlen;
7468 		if (++nlyp->lay_stateid.seqid == 0)
7469 			nlyp->lay_stateid.seqid = 1;
7470 		stateidp->seqid = nlyp->lay_stateid.seqid;
7471 		stateidp->other[0] = nlyp->lay_stateid.other[0];
7472 		stateidp->other[1] = nlyp->lay_stateid.other[1];
7473 		stateidp->other[2] = nlyp->lay_stateid.other[2];
7474 		NFSUNLOCKLAYOUT(lhyp);
7475 		return (0);
7476 	}
7477 
7478 	/* Insert the new layout in the lists. */
7479 	*lypp = NULL;
7480 	atomic_add_int(&nfsrv_layoutcnt, 1);
7481 	NFSD_VNET(nfsstatsv1_p)->srvlayouts++;
7482 	NFSBCOPY(lyp->lay_xdr, layp, lyp->lay_layoutlen);
7483 	*layoutlenp = lyp->lay_layoutlen;
7484 	TAILQ_INSERT_HEAD(&lhyp->list, lyp, lay_list);
7485 	NFSUNLOCKLAYOUT(lhyp);
7486 	return (0);
7487 }
7488 
7489 /*
7490  * Get the devinfo for a deviceid.
7491  */
7492 int
7493 nfsrv_getdevinfo(char *devid, int layouttype, uint32_t *maxcnt,
7494     uint32_t *notify, int *devaddrlen, char **devaddr)
7495 {
7496 	struct nfsdevice *ds;
7497 
7498 	if ((layouttype != NFSLAYOUT_NFSV4_1_FILES && layouttype !=
7499 	     NFSLAYOUT_FLEXFILE) ||
7500 	    (nfsrv_maxpnfsmirror > 1 && layouttype == NFSLAYOUT_NFSV4_1_FILES))
7501 		return (NFSERR_UNKNLAYOUTTYPE);
7502 
7503 	/*
7504 	 * Now, search for the device id.  Note that the structures won't go
7505 	 * away, but the order changes in the list.  As such, the lock only
7506 	 * needs to be held during the search through the list.
7507 	 */
7508 	NFSDDSLOCK();
7509 	TAILQ_FOREACH(ds, &nfsrv_devidhead, nfsdev_list) {
7510 		if (NFSBCMP(devid, ds->nfsdev_deviceid, NFSX_V4DEVICEID) == 0 &&
7511 		    ds->nfsdev_nmp != NULL)
7512 			break;
7513 	}
7514 	NFSDDSUNLOCK();
7515 	if (ds == NULL)
7516 		return (NFSERR_NOENT);
7517 
7518 	/* If the correct nfsdev_XXXXaddrlen is > 0, we have the device info. */
7519 	*devaddrlen = 0;
7520 	if (layouttype == NFSLAYOUT_NFSV4_1_FILES) {
7521 		*devaddrlen = ds->nfsdev_fileaddrlen;
7522 		*devaddr = ds->nfsdev_fileaddr;
7523 	} else if (layouttype == NFSLAYOUT_FLEXFILE) {
7524 		*devaddrlen = ds->nfsdev_flexaddrlen;
7525 		*devaddr = ds->nfsdev_flexaddr;
7526 	}
7527 	if (*devaddrlen == 0)
7528 		return (NFSERR_UNKNLAYOUTTYPE);
7529 
7530 	/*
7531 	 * The XDR overhead is 3 unsigned values: layout_type,
7532 	 * length_of_address and notify bitmap.
7533 	 * If the notify array is changed to not all zeros, the
7534 	 * count of unsigned values must be increased.
7535 	 */
7536 	if (*maxcnt > 0 && *maxcnt < NFSM_RNDUP(*devaddrlen) +
7537 	    3 * NFSX_UNSIGNED) {
7538 		*maxcnt = NFSM_RNDUP(*devaddrlen) + 3 * NFSX_UNSIGNED;
7539 		return (NFSERR_TOOSMALL);
7540 	}
7541 	return (0);
7542 }
7543 
7544 /*
7545  * Free a list of layout state structures.
7546  */
7547 static void
7548 nfsrv_freelayoutlist(nfsquad_t clientid)
7549 {
7550 	struct nfslayouthash *lhyp;
7551 	struct nfslayout *lyp, *nlyp;
7552 	int i;
7553 
7554 	for (i = 0; i < nfsrv_layouthashsize; i++) {
7555 		lhyp = &nfslayouthash[i];
7556 		NFSLOCKLAYOUT(lhyp);
7557 		TAILQ_FOREACH_SAFE(lyp, &lhyp->list, lay_list, nlyp) {
7558 			if (lyp->lay_clientid.qval == clientid.qval)
7559 				nfsrv_freelayout(&lhyp->list, lyp);
7560 		}
7561 		NFSUNLOCKLAYOUT(lhyp);
7562 	}
7563 }
7564 
7565 /*
7566  * Free up a layout.
7567  */
7568 static void
7569 nfsrv_freelayout(struct nfslayouthead *lhp, struct nfslayout *lyp)
7570 {
7571 
7572 	NFSD_DEBUG(4, "Freelayout=%p\n", lyp);
7573 	atomic_add_int(&nfsrv_layoutcnt, -1);
7574 	NFSD_VNET(nfsstatsv1_p)->srvlayouts--;
7575 	TAILQ_REMOVE(lhp, lyp, lay_list);
7576 	free(lyp, M_NFSDSTATE);
7577 }
7578 
7579 /*
7580  * Free up a device id.
7581  */
7582 void
7583 nfsrv_freeonedevid(struct nfsdevice *ds)
7584 {
7585 	int i;
7586 
7587 	atomic_add_int(&nfsrv_devidcnt, -1);
7588 	vrele(ds->nfsdev_dvp);
7589 	for (i = 0; i < nfsrv_dsdirsize; i++)
7590 		if (ds->nfsdev_dsdir[i] != NULL)
7591 			vrele(ds->nfsdev_dsdir[i]);
7592 	free(ds->nfsdev_fileaddr, M_NFSDSTATE);
7593 	free(ds->nfsdev_flexaddr, M_NFSDSTATE);
7594 	free(ds->nfsdev_host, M_NFSDSTATE);
7595 	free(ds, M_NFSDSTATE);
7596 }
7597 
7598 /*
7599  * Free up a device id and its mirrors.
7600  */
7601 static void
7602 nfsrv_freedevid(struct nfsdevice *ds)
7603 {
7604 
7605 	TAILQ_REMOVE(&nfsrv_devidhead, ds, nfsdev_list);
7606 	nfsrv_freeonedevid(ds);
7607 }
7608 
7609 /*
7610  * Free all layouts and device ids.
7611  * Done when the nfsd threads are shut down since there may be a new
7612  * modified device id list created when the nfsd is restarted.
7613  */
7614 void
7615 nfsrv_freealllayoutsanddevids(void)
7616 {
7617 	struct nfsdontlist *mrp, *nmrp;
7618 	struct nfslayout *lyp, *nlyp;
7619 
7620 	/* Get rid of the deviceid structures. */
7621 	nfsrv_freealldevids();
7622 	TAILQ_INIT(&nfsrv_devidhead);
7623 	nfsrv_devidcnt = 0;
7624 
7625 	/* Get rid of all layouts. */
7626 	nfsrv_freealllayouts();
7627 
7628 	/* Get rid of any nfsdontlist entries. */
7629 	LIST_FOREACH_SAFE(mrp, &nfsrv_dontlisthead, nfsmr_list, nmrp)
7630 		free(mrp, M_NFSDSTATE);
7631 	LIST_INIT(&nfsrv_dontlisthead);
7632 	nfsrv_dontlistlen = 0;
7633 
7634 	/* Free layouts in the recall list. */
7635 	TAILQ_FOREACH_SAFE(lyp, &nfsrv_recalllisthead, lay_list, nlyp)
7636 		nfsrv_freelayout(&nfsrv_recalllisthead, lyp);
7637 	TAILQ_INIT(&nfsrv_recalllisthead);
7638 }
7639 
7640 /*
7641  * Free layouts that match the arguments.
7642  */
7643 static void
7644 nfsrv_freelayouts(nfsquad_t *clid, fsid_t *fs, int laytype, int iomode)
7645 {
7646 	struct nfslayouthash *lhyp;
7647 	struct nfslayout *lyp, *nlyp;
7648 	int i;
7649 
7650 	for (i = 0; i < nfsrv_layouthashsize; i++) {
7651 		lhyp = &nfslayouthash[i];
7652 		NFSLOCKLAYOUT(lhyp);
7653 		TAILQ_FOREACH_SAFE(lyp, &lhyp->list, lay_list, nlyp) {
7654 			if (clid->qval != lyp->lay_clientid.qval)
7655 				continue;
7656 			if (fs != NULL && fsidcmp(fs, &lyp->lay_fsid) != 0)
7657 				continue;
7658 			if (laytype != lyp->lay_type)
7659 				continue;
7660 			if ((iomode & NFSLAYOUTIOMODE_READ) != 0)
7661 				lyp->lay_flags &= ~NFSLAY_READ;
7662 			if ((iomode & NFSLAYOUTIOMODE_RW) != 0)
7663 				lyp->lay_flags &= ~NFSLAY_RW;
7664 			if ((lyp->lay_flags & (NFSLAY_READ | NFSLAY_RW)) == 0)
7665 				nfsrv_freelayout(&lhyp->list, lyp);
7666 		}
7667 		NFSUNLOCKLAYOUT(lhyp);
7668 	}
7669 }
7670 
7671 /*
7672  * Free all layouts for the argument file.
7673  */
7674 void
7675 nfsrv_freefilelayouts(fhandle_t *fhp)
7676 {
7677 	struct nfslayouthash *lhyp;
7678 	struct nfslayout *lyp, *nlyp;
7679 
7680 	lhyp = NFSLAYOUTHASH(fhp);
7681 	NFSLOCKLAYOUT(lhyp);
7682 	TAILQ_FOREACH_SAFE(lyp, &lhyp->list, lay_list, nlyp) {
7683 		if (NFSBCMP(&lyp->lay_fh, fhp, sizeof(*fhp)) == 0)
7684 			nfsrv_freelayout(&lhyp->list, lyp);
7685 	}
7686 	NFSUNLOCKLAYOUT(lhyp);
7687 }
7688 
7689 /*
7690  * Free all layouts.
7691  */
7692 static void
7693 nfsrv_freealllayouts(void)
7694 {
7695 	struct nfslayouthash *lhyp;
7696 	struct nfslayout *lyp, *nlyp;
7697 	int i;
7698 
7699 	for (i = 0; i < nfsrv_layouthashsize; i++) {
7700 		lhyp = &nfslayouthash[i];
7701 		NFSLOCKLAYOUT(lhyp);
7702 		TAILQ_FOREACH_SAFE(lyp, &lhyp->list, lay_list, nlyp)
7703 			nfsrv_freelayout(&lhyp->list, lyp);
7704 		NFSUNLOCKLAYOUT(lhyp);
7705 	}
7706 }
7707 
7708 /*
7709  * Look up the mount path for the DS server.
7710  */
7711 static int
7712 nfsrv_setdsserver(char *dspathp, char *mdspathp, NFSPROC_T *p,
7713     struct nfsdevice **dsp)
7714 {
7715 	struct nameidata nd;
7716 	struct nfsdevice *ds;
7717 	struct mount *mp;
7718 	int error, i;
7719 	char *dsdirpath;
7720 	size_t dsdirsize;
7721 
7722 	NFSD_DEBUG(4, "setdssrv path=%s\n", dspathp);
7723 	*dsp = NULL;
7724 	if (jailed(p->td_ucred)) {
7725 		printf("A pNFS nfsd cannot run in a jail\n");
7726 		return (EPERM);
7727 	}
7728 	NDINIT(&nd, LOOKUP, FOLLOW | LOCKSHARED | LOCKLEAF, UIO_SYSSPACE,
7729 	    dspathp);
7730 	error = namei(&nd);
7731 	NFSD_DEBUG(4, "lookup=%d\n", error);
7732 	if (error != 0)
7733 		return (error);
7734 	if (nd.ni_vp->v_type != VDIR) {
7735 		vput(nd.ni_vp);
7736 		NFSD_DEBUG(4, "dspath not dir\n");
7737 		return (ENOTDIR);
7738 	}
7739 	if (strcmp(nd.ni_vp->v_mount->mnt_vfc->vfc_name, "nfs") != 0) {
7740 		vput(nd.ni_vp);
7741 		NFSD_DEBUG(4, "dspath not an NFS mount\n");
7742 		return (ENXIO);
7743 	}
7744 
7745 	/*
7746 	 * Allocate a DS server structure with the NFS mounted directory
7747 	 * vnode reference counted, so that a non-forced dismount will
7748 	 * fail with EBUSY.
7749 	 * This structure is always linked into the list, even if an error
7750 	 * is being returned.  The caller will free the entire list upon
7751 	 * an error return.
7752 	 */
7753 	*dsp = ds = malloc(sizeof(*ds) + nfsrv_dsdirsize * sizeof(vnode_t),
7754 	    M_NFSDSTATE, M_WAITOK | M_ZERO);
7755 	ds->nfsdev_dvp = nd.ni_vp;
7756 	ds->nfsdev_nmp = VFSTONFS(nd.ni_vp->v_mount);
7757 	NFSVOPUNLOCK(nd.ni_vp);
7758 
7759 	dsdirsize = strlen(dspathp) + 16;
7760 	dsdirpath = malloc(dsdirsize, M_TEMP, M_WAITOK);
7761 	/* Now, create the DS directory structures. */
7762 	for (i = 0; i < nfsrv_dsdirsize; i++) {
7763 		snprintf(dsdirpath, dsdirsize, "%s/ds%d", dspathp, i);
7764 		NDINIT(&nd, LOOKUP, FOLLOW | LOCKSHARED | LOCKLEAF,
7765 		    UIO_SYSSPACE, dsdirpath);
7766 		error = namei(&nd);
7767 		NFSD_DEBUG(4, "dsdirpath=%s lookup=%d\n", dsdirpath, error);
7768 		if (error != 0)
7769 			break;
7770 		if (nd.ni_vp->v_type != VDIR) {
7771 			vput(nd.ni_vp);
7772 			error = ENOTDIR;
7773 			NFSD_DEBUG(4, "dsdirpath not a VDIR\n");
7774 			break;
7775 		}
7776 		if (strcmp(nd.ni_vp->v_mount->mnt_vfc->vfc_name, "nfs") != 0) {
7777 			vput(nd.ni_vp);
7778 			error = ENXIO;
7779 			NFSD_DEBUG(4, "dsdirpath not an NFS mount\n");
7780 			break;
7781 		}
7782 		ds->nfsdev_dsdir[i] = nd.ni_vp;
7783 		NFSVOPUNLOCK(nd.ni_vp);
7784 	}
7785 	free(dsdirpath, M_TEMP);
7786 
7787 	if (strlen(mdspathp) > 0) {
7788 		/*
7789 		 * This DS stores file for a specific MDS exported file
7790 		 * system.
7791 		 */
7792 		NDINIT(&nd, LOOKUP, FOLLOW | LOCKSHARED | LOCKLEAF,
7793 		    UIO_SYSSPACE, mdspathp);
7794 		error = namei(&nd);
7795 		NFSD_DEBUG(4, "mds lookup=%d\n", error);
7796 		if (error != 0)
7797 			goto out;
7798 		if (nd.ni_vp->v_type != VDIR) {
7799 			vput(nd.ni_vp);
7800 			error = ENOTDIR;
7801 			NFSD_DEBUG(4, "mdspath not dir\n");
7802 			goto out;
7803 		}
7804 		mp = nd.ni_vp->v_mount;
7805 		if ((mp->mnt_flag & MNT_EXPORTED) == 0) {
7806 			vput(nd.ni_vp);
7807 			error = ENXIO;
7808 			NFSD_DEBUG(4, "mdspath not an exported fs\n");
7809 			goto out;
7810 		}
7811 		ds->nfsdev_mdsfsid = mp->mnt_stat.f_fsid;
7812 		ds->nfsdev_mdsisset = 1;
7813 		vput(nd.ni_vp);
7814 	}
7815 
7816 out:
7817 	TAILQ_INSERT_TAIL(&nfsrv_devidhead, ds, nfsdev_list);
7818 	atomic_add_int(&nfsrv_devidcnt, 1);
7819 	return (error);
7820 }
7821 
7822 /*
7823  * Look up the mount path for the DS server and delete it.
7824  */
7825 int
7826 nfsrv_deldsserver(int op, char *dspathp, NFSPROC_T *p)
7827 {
7828 	struct mount *mp;
7829 	struct nfsmount *nmp;
7830 	struct nfsdevice *ds;
7831 	int error;
7832 
7833 	NFSD_DEBUG(4, "deldssrv path=%s\n", dspathp);
7834 	/*
7835 	 * Search for the path in the mount list.  Avoid looking the path
7836 	 * up, since this mount point may be hung, with associated locked
7837 	 * vnodes, etc.
7838 	 * Set NFSMNTP_CANCELRPCS so that any forced dismount will be blocked
7839 	 * until this completes.
7840 	 * As noted in the man page, this should be done before any forced
7841 	 * dismount on the mount point, but at least the handshake on
7842 	 * NFSMNTP_CANCELRPCS should make it safe.
7843 	 */
7844 	error = 0;
7845 	ds = NULL;
7846 	nmp = NULL;
7847 	mtx_lock(&mountlist_mtx);
7848 	TAILQ_FOREACH(mp, &mountlist, mnt_list) {
7849 		if (strcmp(mp->mnt_stat.f_mntonname, dspathp) == 0 &&
7850 		    strcmp(mp->mnt_stat.f_fstypename, "nfs") == 0 &&
7851 		    mp->mnt_data != NULL) {
7852 			nmp = VFSTONFS(mp);
7853 			NFSLOCKMNT(nmp);
7854 			if ((nmp->nm_privflag & (NFSMNTP_FORCEDISM |
7855 			     NFSMNTP_CANCELRPCS)) == 0) {
7856 				nmp->nm_privflag |= NFSMNTP_CANCELRPCS;
7857 				NFSUNLOCKMNT(nmp);
7858 			} else {
7859 				NFSUNLOCKMNT(nmp);
7860 				nmp = NULL;
7861 			}
7862 			break;
7863 		}
7864 	}
7865 	mtx_unlock(&mountlist_mtx);
7866 
7867 	if (nmp != NULL) {
7868 		ds = nfsrv_deldsnmp(op, nmp, p);
7869 		NFSD_DEBUG(4, "deldsnmp=%p\n", ds);
7870 		if (ds != NULL) {
7871 			nfsrv_killrpcs(nmp);
7872 			NFSD_DEBUG(4, "aft killrpcs\n");
7873 		} else
7874 			error = ENXIO;
7875 		NFSLOCKMNT(nmp);
7876 		nmp->nm_privflag &= ~NFSMNTP_CANCELRPCS;
7877 		wakeup(nmp);
7878 		NFSUNLOCKMNT(nmp);
7879 	} else
7880 		error = EINVAL;
7881 	return (error);
7882 }
7883 
7884 /*
7885  * Search for and remove a DS entry which matches the "nmp" argument.
7886  * The nfsdevice structure pointer is returned so that the caller can
7887  * free it via nfsrv_freeonedevid().
7888  * For the forced case, do not try to do LayoutRecalls, since the server
7889  * must be shut down now anyhow.
7890  */
7891 struct nfsdevice *
7892 nfsrv_deldsnmp(int op, struct nfsmount *nmp, NFSPROC_T *p)
7893 {
7894 	struct nfsdevice *fndds;
7895 
7896 	NFSD_DEBUG(4, "deldsdvp\n");
7897 	NFSDDSLOCK();
7898 	if (op == PNFSDOP_FORCEDELDS)
7899 		fndds = nfsv4_findmirror(nmp);
7900 	else
7901 		fndds = nfsrv_findmirroredds(nmp);
7902 	if (fndds != NULL)
7903 		nfsrv_deleteds(fndds);
7904 	NFSDDSUNLOCK();
7905 	if (fndds != NULL) {
7906 		if (op != PNFSDOP_FORCEDELDS)
7907 			nfsrv_flexmirrordel(fndds->nfsdev_deviceid, p);
7908 		printf("pNFS server: mirror %s failed\n", fndds->nfsdev_host);
7909 	}
7910 	return (fndds);
7911 }
7912 
7913 /*
7914  * Similar to nfsrv_deldsnmp(), except that the DS is indicated by deviceid.
7915  * This function also calls nfsrv_killrpcs() to unblock RPCs on the mount
7916  * point.
7917  * Also, returns an error instead of the nfsdevice found.
7918  */
7919 int
7920 nfsrv_delds(char *devid, NFSPROC_T *p)
7921 {
7922 	struct nfsdevice *ds, *fndds;
7923 	struct nfsmount *nmp;
7924 	int fndmirror;
7925 
7926 	NFSD_DEBUG(4, "delds\n");
7927 	/*
7928 	 * Search the DS server list for a match with devid.
7929 	 * Remove the DS entry if found and there is a mirror.
7930 	 */
7931 	fndds = NULL;
7932 	nmp = NULL;
7933 	fndmirror = 0;
7934 	NFSDDSLOCK();
7935 	TAILQ_FOREACH(ds, &nfsrv_devidhead, nfsdev_list) {
7936 		if (NFSBCMP(ds->nfsdev_deviceid, devid, NFSX_V4DEVICEID) == 0 &&
7937 		    ds->nfsdev_nmp != NULL) {
7938 			NFSD_DEBUG(4, "fnd main ds\n");
7939 			fndds = ds;
7940 			break;
7941 		}
7942 	}
7943 	if (fndds == NULL) {
7944 		NFSDDSUNLOCK();
7945 		return (ENXIO);
7946 	}
7947 	if (fndds->nfsdev_mdsisset == 0 && nfsrv_faildscnt > 0)
7948 		fndmirror = 1;
7949 	else if (fndds->nfsdev_mdsisset != 0) {
7950 		/* For the fsid is set case, search for a mirror. */
7951 		TAILQ_FOREACH(ds, &nfsrv_devidhead, nfsdev_list) {
7952 			if (ds != fndds && ds->nfsdev_nmp != NULL &&
7953 			    ds->nfsdev_mdsisset != 0 &&
7954 			    fsidcmp(&ds->nfsdev_mdsfsid,
7955 			    &fndds->nfsdev_mdsfsid) == 0) {
7956 				fndmirror = 1;
7957 				break;
7958 			}
7959 		}
7960 	}
7961 	if (fndmirror != 0) {
7962 		nmp = fndds->nfsdev_nmp;
7963 		NFSLOCKMNT(nmp);
7964 		if ((nmp->nm_privflag & (NFSMNTP_FORCEDISM |
7965 		     NFSMNTP_CANCELRPCS)) == 0) {
7966 			nmp->nm_privflag |= NFSMNTP_CANCELRPCS;
7967 			NFSUNLOCKMNT(nmp);
7968 			nfsrv_deleteds(fndds);
7969 		} else {
7970 			NFSUNLOCKMNT(nmp);
7971 			nmp = NULL;
7972 		}
7973 	}
7974 	NFSDDSUNLOCK();
7975 	if (nmp != NULL) {
7976 		nfsrv_flexmirrordel(fndds->nfsdev_deviceid, p);
7977 		printf("pNFS server: mirror %s failed\n", fndds->nfsdev_host);
7978 		nfsrv_killrpcs(nmp);
7979 		NFSLOCKMNT(nmp);
7980 		nmp->nm_privflag &= ~NFSMNTP_CANCELRPCS;
7981 		wakeup(nmp);
7982 		NFSUNLOCKMNT(nmp);
7983 		return (0);
7984 	}
7985 	return (ENXIO);
7986 }
7987 
7988 /*
7989  * Mark a DS as disabled by setting nfsdev_nmp = NULL.
7990  */
7991 static void
7992 nfsrv_deleteds(struct nfsdevice *fndds)
7993 {
7994 
7995 	NFSD_DEBUG(4, "deleteds: deleting a mirror\n");
7996 	fndds->nfsdev_nmp = NULL;
7997 	if (fndds->nfsdev_mdsisset == 0)
7998 		nfsrv_faildscnt--;
7999 }
8000 
8001 /*
8002  * Fill in the addr structures for the File and Flex File layouts.
8003  */
8004 static void
8005 nfsrv_allocdevid(struct nfsdevice *ds, char *addr, char *dnshost)
8006 {
8007 	uint32_t *tl;
8008 	char *netprot;
8009 	int addrlen;
8010 	static uint64_t new_devid = 0;
8011 
8012 	if (strchr(addr, ':') != NULL)
8013 		netprot = "tcp6";
8014 	else
8015 		netprot = "tcp";
8016 
8017 	/* Fill in the device id. */
8018 	NFSBCOPY(&nfsdev_time, ds->nfsdev_deviceid, sizeof(nfsdev_time));
8019 	new_devid++;
8020 	NFSBCOPY(&new_devid, &ds->nfsdev_deviceid[sizeof(nfsdev_time)],
8021 	    sizeof(new_devid));
8022 
8023 	/*
8024 	 * Fill in the file addr (actually the nfsv4_file_layout_ds_addr4
8025 	 * as defined in RFC5661) in XDR.
8026 	 */
8027 	addrlen = NFSM_RNDUP(strlen(addr)) + NFSM_RNDUP(strlen(netprot)) +
8028 	    6 * NFSX_UNSIGNED;
8029 	NFSD_DEBUG(4, "hn=%s addr=%s netprot=%s\n", dnshost, addr, netprot);
8030 	ds->nfsdev_fileaddrlen = addrlen;
8031 	tl = malloc(addrlen, M_NFSDSTATE, M_WAITOK | M_ZERO);
8032 	ds->nfsdev_fileaddr = (char *)tl;
8033 	*tl++ = txdr_unsigned(1);		/* One stripe with index 0. */
8034 	*tl++ = 0;
8035 	*tl++ = txdr_unsigned(1);		/* One multipath list */
8036 	*tl++ = txdr_unsigned(1);		/* with one entry in it. */
8037 	/* The netaddr for this one entry. */
8038 	*tl++ = txdr_unsigned(strlen(netprot));
8039 	NFSBCOPY(netprot, tl, strlen(netprot));
8040 	tl += (NFSM_RNDUP(strlen(netprot)) / NFSX_UNSIGNED);
8041 	*tl++ = txdr_unsigned(strlen(addr));
8042 	NFSBCOPY(addr, tl, strlen(addr));
8043 
8044 	/*
8045 	 * Fill in the flex file addr (actually the ff_device_addr4
8046 	 * as defined for Flexible File Layout) in XDR.
8047 	 */
8048 	addrlen = NFSM_RNDUP(strlen(addr)) + NFSM_RNDUP(strlen(netprot)) +
8049 	    14 * NFSX_UNSIGNED;
8050 	ds->nfsdev_flexaddrlen = addrlen;
8051 	tl = malloc(addrlen, M_NFSDSTATE, M_WAITOK | M_ZERO);
8052 	ds->nfsdev_flexaddr = (char *)tl;
8053 	*tl++ = txdr_unsigned(1);		/* One multipath entry. */
8054 	/* The netaddr for this one entry. */
8055 	*tl++ = txdr_unsigned(strlen(netprot));
8056 	NFSBCOPY(netprot, tl, strlen(netprot));
8057 	tl += (NFSM_RNDUP(strlen(netprot)) / NFSX_UNSIGNED);
8058 	*tl++ = txdr_unsigned(strlen(addr));
8059 	NFSBCOPY(addr, tl, strlen(addr));
8060 	tl += (NFSM_RNDUP(strlen(addr)) / NFSX_UNSIGNED);
8061 	*tl++ = txdr_unsigned(2);		/* Two NFS Versions. */
8062 	*tl++ = txdr_unsigned(NFS_VER4);	/* NFSv4. */
8063 	*tl++ = txdr_unsigned(NFSV42_MINORVERSION); /* Minor version 2. */
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 	*tl++ = txdr_unsigned(NFS_VER4);	/* NFSv4. */
8068 	*tl++ = txdr_unsigned(NFSV41_MINORVERSION); /* Minor version 1. */
8069 	*tl++ = txdr_unsigned(nfs_srvmaxio);	/* DS max rsize. */
8070 	*tl++ = txdr_unsigned(nfs_srvmaxio);	/* DS max wsize. */
8071 	*tl = newnfs_true;			/* Tightly coupled. */
8072 
8073 	ds->nfsdev_hostnamelen = strlen(dnshost);
8074 	ds->nfsdev_host = malloc(ds->nfsdev_hostnamelen + 1, M_NFSDSTATE,
8075 	    M_WAITOK);
8076 	NFSBCOPY(dnshost, ds->nfsdev_host, ds->nfsdev_hostnamelen + 1);
8077 }
8078 
8079 /*
8080  * Create the device id list.
8081  * Return 0 if the nfsd threads are to run and ENXIO if the "-p" argument
8082  * is misconfigured.
8083  */
8084 int
8085 nfsrv_createdevids(struct nfsd_nfsd_args *args, NFSPROC_T *p)
8086 {
8087 	struct nfsdevice *ds;
8088 	char *addrp, *dnshostp, *dspathp, *mdspathp;
8089 	int error, i;
8090 
8091 	addrp = args->addr;
8092 	dnshostp = args->dnshost;
8093 	dspathp = args->dspath;
8094 	mdspathp = args->mdspath;
8095 	nfsrv_maxpnfsmirror = args->mirrorcnt;
8096 	if (addrp == NULL || dnshostp == NULL || dspathp == NULL ||
8097 	    mdspathp == NULL)
8098 		return (0);
8099 
8100 	/*
8101 	 * Loop around for each nul-terminated string in args->addr,
8102 	 * args->dnshost, args->dnspath and args->mdspath.
8103 	 */
8104 	while (addrp < (args->addr + args->addrlen) &&
8105 	    dnshostp < (args->dnshost + args->dnshostlen) &&
8106 	    dspathp < (args->dspath + args->dspathlen) &&
8107 	    mdspathp < (args->mdspath + args->mdspathlen)) {
8108 		error = nfsrv_setdsserver(dspathp, mdspathp, p, &ds);
8109 		if (error != 0) {
8110 			/* Free all DS servers. */
8111 			nfsrv_freealldevids();
8112 			nfsrv_devidcnt = 0;
8113 			return (ENXIO);
8114 		}
8115 		nfsrv_allocdevid(ds, addrp, dnshostp);
8116 		addrp += (strlen(addrp) + 1);
8117 		dnshostp += (strlen(dnshostp) + 1);
8118 		dspathp += (strlen(dspathp) + 1);
8119 		mdspathp += (strlen(mdspathp) + 1);
8120 	}
8121 	if (nfsrv_devidcnt < nfsrv_maxpnfsmirror) {
8122 		/* Free all DS servers. */
8123 		nfsrv_freealldevids();
8124 		nfsrv_devidcnt = 0;
8125 		nfsrv_maxpnfsmirror = 1;
8126 		return (ENXIO);
8127 	}
8128 	/* We can fail at most one less DS than the mirror level. */
8129 	nfsrv_faildscnt = nfsrv_maxpnfsmirror - 1;
8130 
8131 	/*
8132 	 * Allocate the nfslayout hash table now, since this is a pNFS server.
8133 	 * Make it 1% of the high water mark and at least 100.
8134 	 */
8135 	if (nfslayouthash == NULL) {
8136 		nfsrv_layouthashsize = nfsrv_layouthighwater / 100;
8137 		if (nfsrv_layouthashsize < 100)
8138 			nfsrv_layouthashsize = 100;
8139 		nfslayouthash = mallocarray(nfsrv_layouthashsize,
8140 		    sizeof(struct nfslayouthash), M_NFSDSESSION, M_WAITOK |
8141 		    M_ZERO);
8142 		for (i = 0; i < nfsrv_layouthashsize; i++) {
8143 			mtx_init(&nfslayouthash[i].mtx, "nfslm", NULL, MTX_DEF);
8144 			TAILQ_INIT(&nfslayouthash[i].list);
8145 		}
8146 	}
8147 	return (0);
8148 }
8149 
8150 /*
8151  * Free all device ids.
8152  */
8153 static void
8154 nfsrv_freealldevids(void)
8155 {
8156 	struct nfsdevice *ds, *nds;
8157 
8158 	TAILQ_FOREACH_SAFE(ds, &nfsrv_devidhead, nfsdev_list, nds)
8159 		nfsrv_freedevid(ds);
8160 }
8161 
8162 /*
8163  * Check to see if there is a Read/Write Layout plus either:
8164  * - A Write Delegation
8165  * or
8166  * - An Open with Write_access.
8167  * Return 1 if this is the case and 0 otherwise.
8168  * This function is used by nfsrv_proxyds() to decide if doing a Proxy
8169  * Getattr RPC to the Data Server (DS) is necessary.
8170  */
8171 #define	NFSCLIDVECSIZE	6
8172 int
8173 nfsrv_checkdsattr(vnode_t vp, NFSPROC_T *p)
8174 {
8175 	fhandle_t fh, *tfhp;
8176 	struct nfsstate *stp;
8177 	struct nfslayout *lyp;
8178 	struct nfslayouthash *lhyp;
8179 	struct nfslockhashhead *hp;
8180 	struct nfslockfile *lfp;
8181 	nfsquad_t clid[NFSCLIDVECSIZE];
8182 	int clidcnt, ret;
8183 
8184 	ret = nfsvno_getfh(vp, &fh, p);
8185 	if (ret != 0)
8186 		return (0);
8187 
8188 	/* First check for a Read/Write Layout. */
8189 	clidcnt = 0;
8190 	lhyp = NFSLAYOUTHASH(&fh);
8191 	NFSLOCKLAYOUT(lhyp);
8192 	TAILQ_FOREACH(lyp, &lhyp->list, lay_list) {
8193 		if (NFSBCMP(&lyp->lay_fh, &fh, sizeof(fh)) == 0 &&
8194 		    ((lyp->lay_flags & NFSLAY_RW) != 0 ||
8195 		     ((lyp->lay_flags & NFSLAY_READ) != 0 &&
8196 		      nfsrv_pnfsatime != 0))) {
8197 			if (clidcnt < NFSCLIDVECSIZE)
8198 				clid[clidcnt].qval = lyp->lay_clientid.qval;
8199 			clidcnt++;
8200 		}
8201 	}
8202 	NFSUNLOCKLAYOUT(lhyp);
8203 	if (clidcnt == 0) {
8204 		/* None found, so return 0. */
8205 		return (0);
8206 	}
8207 
8208 	/* Get the nfslockfile for this fh. */
8209 	NFSLOCKSTATE();
8210 	hp = NFSLOCKHASH(&fh);
8211 	LIST_FOREACH(lfp, hp, lf_hash) {
8212 		tfhp = &lfp->lf_fh;
8213 		if (NFSVNO_CMPFH(&fh, tfhp))
8214 			break;
8215 	}
8216 	if (lfp == NULL) {
8217 		/* None found, so return 0. */
8218 		NFSUNLOCKSTATE();
8219 		return (0);
8220 	}
8221 
8222 	/* Now, look for a Write delegation for this clientid. */
8223 	LIST_FOREACH(stp, &lfp->lf_deleg, ls_file) {
8224 		if ((stp->ls_flags & NFSLCK_DELEGWRITE) != 0 &&
8225 		    nfsrv_fndclid(clid, stp->ls_clp->lc_clientid, clidcnt) != 0)
8226 			break;
8227 	}
8228 	if (stp != NULL) {
8229 		/* Found one, so return 1. */
8230 		NFSUNLOCKSTATE();
8231 		return (1);
8232 	}
8233 
8234 	/* No Write delegation, so look for an Open with Write_access. */
8235 	LIST_FOREACH(stp, &lfp->lf_open, ls_file) {
8236 		KASSERT((stp->ls_flags & NFSLCK_OPEN) != 0,
8237 		    ("nfsrv_checkdsattr: Non-open in Open list\n"));
8238 		if ((stp->ls_flags & NFSLCK_WRITEACCESS) != 0 &&
8239 		    nfsrv_fndclid(clid, stp->ls_clp->lc_clientid, clidcnt) != 0)
8240 			break;
8241 	}
8242 	NFSUNLOCKSTATE();
8243 	if (stp != NULL)
8244 		return (1);
8245 	return (0);
8246 }
8247 
8248 /*
8249  * Look for a matching clientid in the vector. Return 1 if one might match.
8250  */
8251 static int
8252 nfsrv_fndclid(nfsquad_t *clidvec, nfsquad_t clid, int clidcnt)
8253 {
8254 	int i;
8255 
8256 	/* If too many for the vector, return 1 since there might be a match. */
8257 	if (clidcnt > NFSCLIDVECSIZE)
8258 		return (1);
8259 
8260 	for (i = 0; i < clidcnt; i++)
8261 		if (clidvec[i].qval == clid.qval)
8262 			return (1);
8263 	return (0);
8264 }
8265 
8266 /*
8267  * Check the don't list for "vp" and see if issuing an rw layout is allowed.
8268  * Return 1 if issuing an rw layout isn't allowed, 0 otherwise.
8269  */
8270 static int
8271 nfsrv_dontlayout(fhandle_t *fhp)
8272 {
8273 	struct nfsdontlist *mrp;
8274 	int ret;
8275 
8276 	if (nfsrv_dontlistlen == 0)
8277 		return (0);
8278 	ret = 0;
8279 	NFSDDONTLISTLOCK();
8280 	LIST_FOREACH(mrp, &nfsrv_dontlisthead, nfsmr_list) {
8281 		if (NFSBCMP(fhp, &mrp->nfsmr_fh, sizeof(*fhp)) == 0 &&
8282 		    (mrp->nfsmr_flags & NFSMR_DONTLAYOUT) != 0) {
8283 			ret = 1;
8284 			break;
8285 		}
8286 	}
8287 	NFSDDONTLISTUNLOCK();
8288 	return (ret);
8289 }
8290 
8291 #define	PNFSDS_COPYSIZ	65536
8292 /*
8293  * Create a new file on a DS and copy the contents of an extant DS file to it.
8294  * This can be used for recovery of a DS file onto a recovered DS.
8295  * The steps are:
8296  * - When called, the MDS file's vnode is locked, blocking LayoutGet operations.
8297  * - Disable issuing of read/write layouts for the file via the nfsdontlist,
8298  *   so that they will be disabled after the MDS file's vnode is unlocked.
8299  * - Set up the nfsrv_recalllist so that recall of read/write layouts can
8300  *   be done.
8301  * - Unlock the MDS file's vnode, so that the client(s) can perform proxied
8302  *   writes, LayoutCommits and LayoutReturns for the file when completing the
8303  *   LayoutReturn requested by the LayoutRecall callback.
8304  * - Issue a LayoutRecall callback for all read/write layouts and wait for
8305  *   them to be returned. (If the LayoutRecall callback replies
8306  *   NFSERR_NOMATCHLAYOUT, they are gone and no LayoutReturn is needed.)
8307  * - Exclusively lock the MDS file's vnode.  This ensures that no proxied
8308  *   writes are in progress or can occur during the DS file copy.
8309  *   It also blocks Setattr operations.
8310  * - Create the file on the recovered mirror.
8311  * - Copy the file from the operational DS.
8312  * - Copy any ACL from the MDS file to the new DS file.
8313  * - Set the modify time of the new DS file to that of the MDS file.
8314  * - Update the extended attribute for the MDS file.
8315  * - Enable issuing of rw layouts by deleting the nfsdontlist entry.
8316  * - The caller will unlock the MDS file's vnode allowing operations
8317  *   to continue normally, since it is now on the mirror again.
8318  */
8319 int
8320 nfsrv_copymr(vnode_t vp, vnode_t fvp, vnode_t dvp, struct nfsdevice *ds,
8321     struct pnfsdsfile *pf, struct pnfsdsfile *wpf, int mirrorcnt,
8322     struct ucred *cred, NFSPROC_T *p)
8323 {
8324 	struct nfsdontlist *mrp, *nmrp;
8325 	struct nfslayouthash *lhyp;
8326 	struct nfslayout *lyp, *nlyp;
8327 	struct nfslayouthead thl;
8328 	struct mount *mp, *tvmp;
8329 	struct acl *aclp;
8330 	struct vattr va;
8331 	struct timespec mtime;
8332 	fhandle_t fh;
8333 	vnode_t tvp;
8334 	off_t rdpos, wrpos;
8335 	ssize_t aresid;
8336 	char *dat;
8337 	int didprintf, ret, retacl, xfer;
8338 
8339 	ASSERT_VOP_LOCKED(fvp, "nfsrv_copymr fvp");
8340 	ASSERT_VOP_LOCKED(vp, "nfsrv_copymr vp");
8341 	/*
8342 	 * Allocate a nfsdontlist entry and set the NFSMR_DONTLAYOUT flag
8343 	 * so that no more RW layouts will get issued.
8344 	 */
8345 	ret = nfsvno_getfh(vp, &fh, p);
8346 	if (ret != 0) {
8347 		NFSD_DEBUG(4, "nfsrv_copymr: getfh=%d\n", ret);
8348 		return (ret);
8349 	}
8350 	nmrp = malloc(sizeof(*nmrp), M_NFSDSTATE, M_WAITOK);
8351 	nmrp->nfsmr_flags = NFSMR_DONTLAYOUT;
8352 	NFSBCOPY(&fh, &nmrp->nfsmr_fh, sizeof(fh));
8353 	NFSDDONTLISTLOCK();
8354 	LIST_FOREACH(mrp, &nfsrv_dontlisthead, nfsmr_list) {
8355 		if (NFSBCMP(&fh, &mrp->nfsmr_fh, sizeof(fh)) == 0)
8356 			break;
8357 	}
8358 	if (mrp == NULL) {
8359 		LIST_INSERT_HEAD(&nfsrv_dontlisthead, nmrp, nfsmr_list);
8360 		mrp = nmrp;
8361 		nmrp = NULL;
8362 		nfsrv_dontlistlen++;
8363 		NFSD_DEBUG(4, "nfsrv_copymr: in dontlist\n");
8364 	} else {
8365 		NFSDDONTLISTUNLOCK();
8366 		free(nmrp, M_NFSDSTATE);
8367 		NFSD_DEBUG(4, "nfsrv_copymr: dup dontlist\n");
8368 		return (ENXIO);
8369 	}
8370 	NFSDDONTLISTUNLOCK();
8371 
8372 	/*
8373 	 * Search for all RW layouts for this file.  Move them to the
8374 	 * recall list, so they can be recalled and their return noted.
8375 	 */
8376 	lhyp = NFSLAYOUTHASH(&fh);
8377 	NFSDRECALLLOCK();
8378 	NFSLOCKLAYOUT(lhyp);
8379 	TAILQ_FOREACH_SAFE(lyp, &lhyp->list, lay_list, nlyp) {
8380 		if (NFSBCMP(&lyp->lay_fh, &fh, sizeof(fh)) == 0 &&
8381 		    (lyp->lay_flags & NFSLAY_RW) != 0) {
8382 			TAILQ_REMOVE(&lhyp->list, lyp, lay_list);
8383 			TAILQ_INSERT_HEAD(&nfsrv_recalllisthead, lyp, lay_list);
8384 			lyp->lay_trycnt = 0;
8385 		}
8386 	}
8387 	NFSUNLOCKLAYOUT(lhyp);
8388 	NFSDRECALLUNLOCK();
8389 
8390 	ret = 0;
8391 	mp = tvmp = NULL;
8392 	didprintf = 0;
8393 	TAILQ_INIT(&thl);
8394 	/* Unlock the MDS vp, so that a LayoutReturn can be done on it. */
8395 	NFSVOPUNLOCK(vp);
8396 	/* Now, do a recall for all layouts not yet recalled. */
8397 tryagain:
8398 	NFSDRECALLLOCK();
8399 	TAILQ_FOREACH(lyp, &nfsrv_recalllisthead, lay_list) {
8400 		if (NFSBCMP(&lyp->lay_fh, &fh, sizeof(fh)) == 0 &&
8401 		    (lyp->lay_flags & NFSLAY_RECALL) == 0) {
8402 			lyp->lay_flags |= NFSLAY_RECALL;
8403 			/*
8404 			 * The layout stateid.seqid needs to be incremented
8405 			 * before doing a LAYOUT_RECALL callback.
8406 			 */
8407 			if (++lyp->lay_stateid.seqid == 0)
8408 				lyp->lay_stateid.seqid = 1;
8409 			NFSDRECALLUNLOCK();
8410 			nfsrv_recalllayout(lyp->lay_clientid, &lyp->lay_stateid,
8411 			    &lyp->lay_fh, lyp, 0, lyp->lay_type, p);
8412 			NFSD_DEBUG(4, "nfsrv_copymr: recalled layout\n");
8413 			goto tryagain;
8414 		}
8415 	}
8416 
8417 	/* Now wait for them to be returned. */
8418 tryagain2:
8419 	TAILQ_FOREACH(lyp, &nfsrv_recalllisthead, lay_list) {
8420 		if (NFSBCMP(&lyp->lay_fh, &fh, sizeof(fh)) == 0) {
8421 			if ((lyp->lay_flags & NFSLAY_RETURNED) != 0) {
8422 				TAILQ_REMOVE(&nfsrv_recalllisthead, lyp,
8423 				    lay_list);
8424 				TAILQ_INSERT_HEAD(&thl, lyp, lay_list);
8425 				NFSD_DEBUG(4,
8426 				    "nfsrv_copymr: layout returned\n");
8427 			} else {
8428 				lyp->lay_trycnt++;
8429 				ret = mtx_sleep(lyp, NFSDRECALLMUTEXPTR,
8430 				    PVFS | PCATCH, "nfsmrl", hz);
8431 				NFSD_DEBUG(4, "nfsrv_copymr: aft sleep=%d\n",
8432 				    ret);
8433 				if (ret == EINTR || ret == ERESTART)
8434 					break;
8435 				if ((lyp->lay_flags & NFSLAY_RETURNED) == 0) {
8436 					/*
8437 					 * Give up after 60sec and return
8438 					 * ENXIO, failing the copymr.
8439 					 * This layout will remain on the
8440 					 * recalllist.  It can only be cleared
8441 					 * by restarting the nfsd.
8442 					 * This seems the safe way to handle
8443 					 * it, since it cannot be safely copied
8444 					 * with an outstanding RW layout.
8445 					 */
8446 					if (lyp->lay_trycnt >= 60) {
8447 						ret = ENXIO;
8448 						break;
8449 					}
8450 					if (didprintf == 0) {
8451 						printf("nfsrv_copymr: layout "
8452 						    "not returned\n");
8453 						didprintf = 1;
8454 					}
8455 				}
8456 			}
8457 			goto tryagain2;
8458 		}
8459 	}
8460 	NFSDRECALLUNLOCK();
8461 	/* We can now get rid of the layouts that have been returned. */
8462 	TAILQ_FOREACH_SAFE(lyp, &thl, lay_list, nlyp)
8463 		nfsrv_freelayout(&thl, lyp);
8464 
8465 	/*
8466 	 * Do the vn_start_write() calls here, before the MDS vnode is
8467 	 * locked and the tvp is created (locked) in the NFS file system
8468 	 * that dvp is in.
8469 	 * For tvmp, this probably isn't necessary, since it will be an
8470 	 * NFS mount and they are not suspendable at this time.
8471 	 */
8472 	if (ret == 0)
8473 		ret = vn_start_write(vp, &mp, V_WAIT | V_PCATCH);
8474 	if (ret == 0) {
8475 		tvmp = dvp->v_mount;
8476 		ret = vn_start_write(NULL, &tvmp, V_WAIT | V_PCATCH);
8477 	}
8478 
8479 	/*
8480 	 * LK_EXCLUSIVE lock the MDS vnode, so that any
8481 	 * proxied writes through the MDS will be blocked until we have
8482 	 * completed the copy and update of the extended attributes.
8483 	 * This will also ensure that any attributes and ACL will not be
8484 	 * changed until the copy is complete.
8485 	 */
8486 	NFSVOPLOCK(vp, LK_EXCLUSIVE | LK_RETRY);
8487 	if (ret == 0 && VN_IS_DOOMED(vp)) {
8488 		NFSD_DEBUG(4, "nfsrv_copymr: lk_exclusive doomed\n");
8489 		ret = ESTALE;
8490 	}
8491 
8492 	/* Create the data file on the recovered DS. */
8493 	if (ret == 0)
8494 		ret = nfsrv_createdsfile(vp, &fh, pf, dvp, ds, cred, p, &tvp);
8495 
8496 	/* Copy the DS file, if created successfully. */
8497 	if (ret == 0) {
8498 		/*
8499 		 * Get any NFSv4 ACL on the MDS file, so that it can be set
8500 		 * on the new DS file.
8501 		 */
8502 		aclp = acl_alloc(M_WAITOK | M_ZERO);
8503 		retacl = VOP_GETACL(vp, ACL_TYPE_NFS4, aclp, cred, p);
8504 		if (retacl != 0 && retacl != ENOATTR)
8505 			NFSD_DEBUG(1, "nfsrv_copymr: vop_getacl=%d\n", retacl);
8506 		dat = malloc(PNFSDS_COPYSIZ, M_TEMP, M_WAITOK);
8507 		/* Malloc a block of 0s used to check for holes. */
8508 		if (nfsrv_zeropnfsdat == NULL)
8509 			nfsrv_zeropnfsdat = malloc(PNFSDS_COPYSIZ, M_TEMP,
8510 			    M_WAITOK | M_ZERO);
8511 		rdpos = wrpos = 0;
8512 		ret = VOP_GETATTR(fvp, &va, cred);
8513 		aresid = 0;
8514 		while (ret == 0 && aresid == 0) {
8515 			ret = vn_rdwr(UIO_READ, fvp, dat, PNFSDS_COPYSIZ,
8516 			    rdpos, UIO_SYSSPACE, IO_NODELOCKED, cred, NULL,
8517 			    &aresid, p);
8518 			xfer = PNFSDS_COPYSIZ - aresid;
8519 			if (ret == 0 && xfer > 0) {
8520 				rdpos += xfer;
8521 				/*
8522 				 * Skip the write for holes, except for the
8523 				 * last block.
8524 				 */
8525 				if (xfer < PNFSDS_COPYSIZ || rdpos ==
8526 				    va.va_size || NFSBCMP(dat,
8527 				    nfsrv_zeropnfsdat, PNFSDS_COPYSIZ) != 0)
8528 					ret = vn_rdwr(UIO_WRITE, tvp, dat, xfer,
8529 					    wrpos, UIO_SYSSPACE, IO_NODELOCKED,
8530 					    cred, NULL, NULL, p);
8531 				if (ret == 0)
8532 					wrpos += xfer;
8533 			}
8534 		}
8535 
8536 		/* If there is an ACL and the copy succeeded, set the ACL. */
8537 		if (ret == 0 && retacl == 0) {
8538 			ret = VOP_SETACL(tvp, ACL_TYPE_NFS4, aclp, cred, p);
8539 			/*
8540 			 * Don't consider these as errors, since VOP_GETACL()
8541 			 * can return an ACL when they are not actually
8542 			 * supported.  For example, for UFS, VOP_GETACL()
8543 			 * will return a trivial ACL based on the uid/gid/mode
8544 			 * when there is no ACL on the file.
8545 			 * This case should be recognized as a trivial ACL
8546 			 * by UFS's VOP_SETACL() and succeed, but...
8547 			 */
8548 			if (ret == ENOATTR || ret == EOPNOTSUPP || ret == EPERM)
8549 				ret = 0;
8550 		}
8551 
8552 		if (ret == 0)
8553 			ret = VOP_FSYNC(tvp, MNT_WAIT, p);
8554 
8555 		/* Set the DS data file's modify time that of the MDS file. */
8556 		if (ret == 0)
8557 			ret = VOP_GETATTR(vp, &va, cred);
8558 		if (ret == 0) {
8559 			mtime = va.va_mtime;
8560 			VATTR_NULL(&va);
8561 			va.va_mtime = mtime;
8562 			ret = VOP_SETATTR(tvp, &va, cred);
8563 		}
8564 
8565 		vput(tvp);
8566 		acl_free(aclp);
8567 		free(dat, M_TEMP);
8568 	}
8569 	if (tvmp != NULL)
8570 		vn_finished_write(tvmp);
8571 
8572 	/* Update the extended attributes for the newly created DS file. */
8573 	if (ret == 0)
8574 		ret = vn_extattr_set(vp, IO_NODELOCKED,
8575 		    EXTATTR_NAMESPACE_SYSTEM, "pnfsd.dsfile",
8576 		    sizeof(*wpf) * mirrorcnt, (char *)wpf, p);
8577 	if (mp != NULL)
8578 		vn_finished_write(mp);
8579 
8580 	/* Get rid of the dontlist entry, so that Layouts can be issued. */
8581 	NFSDDONTLISTLOCK();
8582 	LIST_REMOVE(mrp, nfsmr_list);
8583 	NFSDDONTLISTUNLOCK();
8584 	free(mrp, M_NFSDSTATE);
8585 	return (ret);
8586 }
8587 
8588 /*
8589  * Create a data storage file on the recovered DS.
8590  */
8591 static int
8592 nfsrv_createdsfile(vnode_t vp, fhandle_t *fhp, struct pnfsdsfile *pf,
8593     vnode_t dvp, struct nfsdevice *ds, struct ucred *cred, NFSPROC_T *p,
8594     vnode_t *tvpp)
8595 {
8596 	struct vattr va, nva;
8597 	int error;
8598 
8599 	/* Make data file name based on FH. */
8600 	error = VOP_GETATTR(vp, &va, cred);
8601 	if (error == 0) {
8602 		/* Set the attributes for "vp" to Setattr the DS vp. */
8603 		VATTR_NULL(&nva);
8604 		nva.va_uid = va.va_uid;
8605 		nva.va_gid = va.va_gid;
8606 		nva.va_mode = va.va_mode;
8607 		nva.va_size = 0;
8608 		VATTR_NULL(&va);
8609 		va.va_type = VREG;
8610 		va.va_mode = nva.va_mode;
8611 		NFSD_DEBUG(4, "nfsrv_dscreatefile: dvp=%p pf=%p\n", dvp, pf);
8612 		error = nfsrv_dscreate(dvp, &va, &nva, fhp, pf, NULL,
8613 		    pf->dsf_filename, cred, p, tvpp);
8614 	}
8615 	return (error);
8616 }
8617 
8618 /*
8619  * Look up the MDS file shared locked, and then get the extended attribute
8620  * to find the extant DS file to be copied to the new mirror.
8621  * If successful, *vpp is set to the MDS file's vp and *nvpp is
8622  * set to a DS data file for the MDS file, both exclusively locked.
8623  * The "buf" argument has the pnfsdsfile structure from the MDS file
8624  * in it and buflen is set to its length.
8625  */
8626 int
8627 nfsrv_mdscopymr(char *mdspathp, char *dspathp, char *curdspathp, char *buf,
8628     int *buflenp, char *fname, NFSPROC_T *p, struct vnode **vpp,
8629     struct vnode **nvpp, struct pnfsdsfile **pfp, struct nfsdevice **dsp,
8630     struct nfsdevice **fdsp)
8631 {
8632 	struct nameidata nd;
8633 	struct vnode *vp, *curvp;
8634 	struct pnfsdsfile *pf;
8635 	struct nfsmount *nmp, *curnmp;
8636 	int dsdir, error, mirrorcnt, ippos;
8637 
8638 	vp = NULL;
8639 	curvp = NULL;
8640 	curnmp = NULL;
8641 	*dsp = NULL;
8642 	*fdsp = NULL;
8643 	if (dspathp == NULL && curdspathp != NULL)
8644 		return (EPERM);
8645 
8646 	/*
8647 	 * Look up the MDS file shared locked.  The lock will be upgraded
8648 	 * to an exclusive lock after any rw layouts have been returned.
8649 	 */
8650 	NFSD_DEBUG(4, "mdsopen path=%s\n", mdspathp);
8651 	NDINIT(&nd, LOOKUP, FOLLOW | LOCKSHARED | LOCKLEAF, UIO_SYSSPACE,
8652 	    mdspathp);
8653 	error = namei(&nd);
8654 	NFSD_DEBUG(4, "lookup=%d\n", error);
8655 	if (error != 0)
8656 		return (error);
8657 	if (nd.ni_vp->v_type != VREG) {
8658 		vput(nd.ni_vp);
8659 		NFSD_DEBUG(4, "mdspath not reg\n");
8660 		return (EISDIR);
8661 	}
8662 	vp = nd.ni_vp;
8663 
8664 	if (curdspathp != NULL) {
8665 		/*
8666 		 * Look up the current DS path and find the nfsdev structure for
8667 		 * it.
8668 		 */
8669 		NFSD_DEBUG(4, "curmdsdev path=%s\n", curdspathp);
8670 		NDINIT(&nd, LOOKUP, FOLLOW | LOCKSHARED | LOCKLEAF,
8671 		    UIO_SYSSPACE, curdspathp);
8672 		error = namei(&nd);
8673 		NFSD_DEBUG(4, "ds lookup=%d\n", error);
8674 		if (error != 0) {
8675 			vput(vp);
8676 			return (error);
8677 		}
8678 		if (nd.ni_vp->v_type != VDIR) {
8679 			vput(nd.ni_vp);
8680 			vput(vp);
8681 			NFSD_DEBUG(4, "curdspath not dir\n");
8682 			return (ENOTDIR);
8683 		}
8684 		if (strcmp(nd.ni_vp->v_mount->mnt_vfc->vfc_name, "nfs") != 0) {
8685 			vput(nd.ni_vp);
8686 			vput(vp);
8687 			NFSD_DEBUG(4, "curdspath not an NFS mount\n");
8688 			return (ENXIO);
8689 		}
8690 		curnmp = VFSTONFS(nd.ni_vp->v_mount);
8691 
8692 		/* Search the nfsdev list for a match. */
8693 		NFSDDSLOCK();
8694 		*fdsp = nfsv4_findmirror(curnmp);
8695 		NFSDDSUNLOCK();
8696 		if (*fdsp == NULL)
8697 			curnmp = NULL;
8698 		if (curnmp == NULL) {
8699 			vput(nd.ni_vp);
8700 			vput(vp);
8701 			NFSD_DEBUG(4, "mdscopymr: no current ds\n");
8702 			return (ENXIO);
8703 		}
8704 		curvp = nd.ni_vp;
8705 	}
8706 
8707 	if (dspathp != NULL) {
8708 		/* Look up the nfsdev path and find the nfsdev structure. */
8709 		NFSD_DEBUG(4, "mdsdev path=%s\n", dspathp);
8710 		NDINIT(&nd, LOOKUP, FOLLOW | LOCKSHARED | LOCKLEAF,
8711 		    UIO_SYSSPACE, dspathp);
8712 		error = namei(&nd);
8713 		NFSD_DEBUG(4, "ds lookup=%d\n", error);
8714 		if (error != 0) {
8715 			vput(vp);
8716 			if (curvp != NULL)
8717 				vput(curvp);
8718 			return (error);
8719 		}
8720 		if (nd.ni_vp->v_type != VDIR || nd.ni_vp == curvp) {
8721 			vput(nd.ni_vp);
8722 			vput(vp);
8723 			if (curvp != NULL)
8724 				vput(curvp);
8725 			NFSD_DEBUG(4, "dspath not dir\n");
8726 			if (nd.ni_vp == curvp)
8727 				return (EPERM);
8728 			return (ENOTDIR);
8729 		}
8730 		if (strcmp(nd.ni_vp->v_mount->mnt_vfc->vfc_name, "nfs") != 0) {
8731 			vput(nd.ni_vp);
8732 			vput(vp);
8733 			if (curvp != NULL)
8734 				vput(curvp);
8735 			NFSD_DEBUG(4, "dspath not an NFS mount\n");
8736 			return (ENXIO);
8737 		}
8738 		nmp = VFSTONFS(nd.ni_vp->v_mount);
8739 
8740 		/*
8741 		 * Search the nfsdevice list for a match.  If curnmp == NULL,
8742 		 * this is a recovery and there must be a mirror.
8743 		 */
8744 		NFSDDSLOCK();
8745 		if (curnmp == NULL)
8746 			*dsp = nfsrv_findmirroredds(nmp);
8747 		else
8748 			*dsp = nfsv4_findmirror(nmp);
8749 		NFSDDSUNLOCK();
8750 		if (*dsp == NULL) {
8751 			vput(nd.ni_vp);
8752 			vput(vp);
8753 			if (curvp != NULL)
8754 				vput(curvp);
8755 			NFSD_DEBUG(4, "mdscopymr: no ds\n");
8756 			return (ENXIO);
8757 		}
8758 	} else {
8759 		nd.ni_vp = NULL;
8760 		nmp = NULL;
8761 	}
8762 
8763 	/*
8764 	 * Get a vp for an available DS data file using the extended
8765 	 * attribute on the MDS file.
8766 	 * If there is a valid entry for the new DS in the extended attribute
8767 	 * on the MDS file (as checked via the nmp argument),
8768 	 * nfsrv_dsgetsockmnt() returns EEXIST, so no copying will occur.
8769 	 */
8770 	error = nfsrv_dsgetsockmnt(vp, 0, buf, buflenp, &mirrorcnt, p,
8771 	    NULL, NULL, NULL, fname, nvpp, &nmp, curnmp, &ippos, &dsdir);
8772 	if (curvp != NULL)
8773 		vput(curvp);
8774 	if (nd.ni_vp == NULL) {
8775 		if (error == 0 && nmp != NULL) {
8776 			/* Search the nfsdev list for a match. */
8777 			NFSDDSLOCK();
8778 			*dsp = nfsrv_findmirroredds(nmp);
8779 			NFSDDSUNLOCK();
8780 		}
8781 		if (error == 0 && (nmp == NULL || *dsp == NULL)) {
8782 			if (nvpp != NULL && *nvpp != NULL) {
8783 				vput(*nvpp);
8784 				*nvpp = NULL;
8785 			}
8786 			error = ENXIO;
8787 		}
8788 	} else
8789 		vput(nd.ni_vp);
8790 
8791 	/*
8792 	 * When dspathp != NULL and curdspathp == NULL, this is a recovery
8793 	 * and is only allowed if there is a 0.0.0.0 IP address entry.
8794 	 * When curdspathp != NULL, the ippos will be set to that entry.
8795 	 */
8796 	if (error == 0 && dspathp != NULL && ippos == -1) {
8797 		if (nvpp != NULL && *nvpp != NULL) {
8798 			vput(*nvpp);
8799 			*nvpp = NULL;
8800 		}
8801 		error = ENXIO;
8802 	}
8803 	if (error == 0) {
8804 		*vpp = vp;
8805 
8806 		pf = (struct pnfsdsfile *)buf;
8807 		if (ippos == -1) {
8808 			/* If no zeroip pnfsdsfile, add one. */
8809 			ippos = *buflenp / sizeof(*pf);
8810 			*buflenp += sizeof(*pf);
8811 			pf += ippos;
8812 			pf->dsf_dir = dsdir;
8813 			strlcpy(pf->dsf_filename, fname,
8814 			    sizeof(pf->dsf_filename));
8815 		} else
8816 			pf += ippos;
8817 		*pfp = pf;
8818 	} else
8819 		vput(vp);
8820 	return (error);
8821 }
8822 
8823 /*
8824  * Search for a matching pnfsd mirror device structure, base on the nmp arg.
8825  * Return one if found, NULL otherwise.
8826  */
8827 static struct nfsdevice *
8828 nfsrv_findmirroredds(struct nfsmount *nmp)
8829 {
8830 	struct nfsdevice *ds, *fndds;
8831 	int fndmirror;
8832 
8833 	mtx_assert(NFSDDSMUTEXPTR, MA_OWNED);
8834 	/*
8835 	 * Search the DS server list for a match with nmp.
8836 	 * Remove the DS entry if found and there is a mirror.
8837 	 */
8838 	fndds = NULL;
8839 	fndmirror = 0;
8840 	if (nfsrv_devidcnt == 0)
8841 		return (fndds);
8842 	TAILQ_FOREACH(ds, &nfsrv_devidhead, nfsdev_list) {
8843 		if (ds->nfsdev_nmp == nmp) {
8844 			NFSD_DEBUG(4, "nfsrv_findmirroredds: fnd main ds\n");
8845 			fndds = ds;
8846 			break;
8847 		}
8848 	}
8849 	if (fndds == NULL)
8850 		return (fndds);
8851 	if (fndds->nfsdev_mdsisset == 0 && nfsrv_faildscnt > 0)
8852 		fndmirror = 1;
8853 	else if (fndds->nfsdev_mdsisset != 0) {
8854 		/* For the fsid is set case, search for a mirror. */
8855 		TAILQ_FOREACH(ds, &nfsrv_devidhead, nfsdev_list) {
8856 			if (ds != fndds && ds->nfsdev_nmp != NULL &&
8857 			    ds->nfsdev_mdsisset != 0 &&
8858 			    fsidcmp(&ds->nfsdev_mdsfsid,
8859 			    &fndds->nfsdev_mdsfsid) == 0) {
8860 				fndmirror = 1;
8861 				break;
8862 			}
8863 		}
8864 	}
8865 	if (fndmirror == 0) {
8866 		NFSD_DEBUG(4, "nfsrv_findmirroredds: no mirror for DS\n");
8867 		return (NULL);
8868 	}
8869 	return (fndds);
8870 }
8871 
8872 /*
8873  * Mark the appropriate devid and all associated layout as "out of space".
8874  */
8875 void
8876 nfsrv_marknospc(char *devid, bool setit)
8877 {
8878 	struct nfsdevice *ds;
8879 	struct nfslayout *lyp;
8880 	struct nfslayouthash *lhyp;
8881 	int i;
8882 
8883 	NFSDDSLOCK();
8884 	TAILQ_FOREACH(ds, &nfsrv_devidhead, nfsdev_list) {
8885 		if (NFSBCMP(ds->nfsdev_deviceid, devid, NFSX_V4DEVICEID) == 0) {
8886 			NFSD_DEBUG(1, "nfsrv_marknospc: devid %d\n", setit);
8887 			ds->nfsdev_nospc = setit;
8888 		}
8889 	}
8890 	NFSDDSUNLOCK();
8891 
8892 	for (i = 0; i < nfsrv_layouthashsize; i++) {
8893 		lhyp = &nfslayouthash[i];
8894 		NFSLOCKLAYOUT(lhyp);
8895 		TAILQ_FOREACH(lyp, &lhyp->list, lay_list) {
8896 			if (NFSBCMP(lyp->lay_deviceid, devid,
8897 			    NFSX_V4DEVICEID) == 0) {
8898 				NFSD_DEBUG(1, "nfsrv_marknospc: layout %d\n",
8899 				    setit);
8900 				if (setit)
8901 					lyp->lay_flags |= NFSLAY_NOSPC;
8902 				else
8903 					lyp->lay_flags &= ~NFSLAY_NOSPC;
8904 			}
8905 		}
8906 		NFSUNLOCKLAYOUT(lhyp);
8907 	}
8908 }
8909 
8910 /*
8911  * Check to see if SP4_MACH_CRED is in use and, if it is, check that the
8912  * correct machine credential is being used.
8913  */
8914 static int
8915 nfsrv_checkmachcred(int op, struct nfsrv_descript *nd, struct nfsclient *clp)
8916 {
8917 
8918 	if ((clp->lc_flags & LCL_MACHCRED) == 0 ||
8919 	    !NFSISSET_OPBIT(&clp->lc_mustops, op))
8920 		return (0);
8921 	KASSERT((nd->nd_flag & ND_NFSV41) != 0,
8922 	    ("nfsrv_checkmachcred: MachCred for NFSv4.0"));
8923 	if ((nd->nd_flag & (ND_GSSINTEGRITY | ND_GSSPRIVACY)) != 0 &&
8924 	    nd->nd_princlen == clp->lc_namelen &&
8925 	    !NFSBCMP(nd->nd_principal, clp->lc_name, nd->nd_princlen))
8926 		return (0);
8927 	return (NFSERR_AUTHERR | AUTH_TOOWEAK);
8928 }
8929 
8930 /*
8931  * Issue a delegation and, optionally set rflagsp for why not.
8932  */
8933 static void
8934 nfsrv_issuedelegation(struct vnode *vp, struct nfsclient *clp,
8935     struct nfsrv_descript *nd, int delegate, int writedeleg, int readonly,
8936     u_quad_t filerev, uint64_t rdonly, struct nfsstate **new_delegp,
8937     struct nfsstate *new_stp, struct nfslockfile *lfp, uint32_t *rflagsp,
8938     nfsv4stateid_t *delegstateidp)
8939 {
8940 	struct nfsstate *up_deleg, *new_deleg;
8941 
8942 	new_deleg = *new_delegp;
8943 	up_deleg = LIST_FIRST(&lfp->lf_deleg);
8944 	if ((new_stp->ls_flags & NFSLCK_WANTNODELEG) != 0)
8945 		*rflagsp |= NFSV4OPEN_WDNOTWANTED;
8946 	else if (nfsrv_issuedelegs == 0)
8947 		*rflagsp |= NFSV4OPEN_WDSUPPFTYPE;
8948 	else if (NFSRV_V4DELEGLIMIT(nfsrv_delegatecnt))
8949 		*rflagsp |= NFSV4OPEN_WDRESOURCE;
8950 	else if (delegate == 0 || !NFSVNO_DELEGOK(vp) ||
8951 	    (writedeleg == 0 && (readonly == 0 ||
8952 	    (new_stp->ls_flags & NFSLCK_WANTWDELEG) != 0)) ||
8953 	    (clp->lc_flags & (LCL_CALLBACKSON | LCL_CBDOWN)) !=
8954 	     LCL_CALLBACKSON) {
8955 		/* Is this a downgrade attempt? */
8956 		if (up_deleg != NULL && up_deleg->ls_clp == clp &&
8957 		    (up_deleg->ls_flags & NFSLCK_DELEGWRITE) != 0 &&
8958 		    (new_stp->ls_flags & NFSLCK_WANTRDELEG) != 0)
8959 			*rflagsp |= NFSV4OPEN_WDNOTSUPPDOWNGRADE;
8960 		else
8961 			*rflagsp |= NFSV4OPEN_WDCONTENTION;
8962 	} else if (up_deleg != NULL &&
8963 	    (up_deleg->ls_flags & NFSLCK_DELEGREAD) != 0 &&
8964 	    (new_stp->ls_flags & NFSLCK_WANTWDELEG) != 0) {
8965 		/* This is an atomic upgrade. */
8966 		up_deleg->ls_stateid.seqid++;
8967 		delegstateidp->seqid = up_deleg->ls_stateid.seqid;
8968 		delegstateidp->other[0] = up_deleg->ls_stateid.other[0];
8969 		delegstateidp->other[1] = up_deleg->ls_stateid.other[1];
8970 		delegstateidp->other[2] = up_deleg->ls_stateid.other[2];
8971 		up_deleg->ls_flags = (NFSLCK_DELEGWRITE |
8972 		    NFSLCK_READACCESS | NFSLCK_WRITEACCESS);
8973 		*rflagsp |= NFSV4OPEN_WRITEDELEGATE;
8974 		nfsrv_writedelegcnt++;
8975 	} else {
8976 		new_deleg->ls_stateid.seqid = delegstateidp->seqid = 1;
8977 		new_deleg->ls_stateid.other[0] = delegstateidp->other[0]
8978 		    = clp->lc_clientid.lval[0];
8979 		new_deleg->ls_stateid.other[1] = delegstateidp->other[1]
8980 		    = clp->lc_clientid.lval[1];
8981 		new_deleg->ls_stateid.other[2] = delegstateidp->other[2]
8982 		    = nfsrv_nextstateindex(clp);
8983 		if (writedeleg && !rdonly &&
8984 		    (nfsrv_writedelegifpos || !readonly) &&
8985 		    (new_stp->ls_flags & (NFSLCK_WANTRDELEG |
8986 		     NFSLCK_WANTWDELEG)) != NFSLCK_WANTRDELEG) {
8987 			new_deleg->ls_flags = (NFSLCK_DELEGWRITE |
8988 			    NFSLCK_READACCESS | NFSLCK_WRITEACCESS);
8989 			*rflagsp |= NFSV4OPEN_WRITEDELEGATE;
8990 			nfsrv_writedelegcnt++;
8991 		} else {
8992 			new_deleg->ls_flags = (NFSLCK_DELEGREAD |
8993 			    NFSLCK_READACCESS);
8994 			*rflagsp |= NFSV4OPEN_READDELEGATE;
8995 		}
8996 		new_deleg->ls_uid = new_stp->ls_uid;
8997 		new_deleg->ls_lfp = lfp;
8998 		new_deleg->ls_clp = clp;
8999 		new_deleg->ls_filerev = filerev;
9000 		new_deleg->ls_compref = nd->nd_compref;
9001 		new_deleg->ls_lastrecall = 0;
9002 		LIST_INSERT_HEAD(&lfp->lf_deleg, new_deleg, ls_file);
9003 		LIST_INSERT_HEAD(NFSSTATEHASH(clp, new_deleg->ls_stateid),
9004 		    new_deleg, ls_hash);
9005 		LIST_INSERT_HEAD(&clp->lc_deleg, new_deleg, ls_list);
9006 		*new_delegp = NULL;
9007 		NFSD_VNET(nfsstatsv1_p)->srvdelegates++;
9008 		nfsrv_openpluslock++;
9009 		nfsrv_delegatecnt++;
9010 	}
9011 }
9012