xref: /freebsd/sys/fs/nfsserver/nfs_nfsdstate.c (revision fcb560670601b2a4d87bb31d7531c8dcc37ee71b)
1 /*-
2  * Copyright (c) 2009 Rick Macklem, University of Guelph
3  * All rights reserved.
4  *
5  * Redistribution and use in source and binary forms, with or without
6  * modification, are permitted provided that the following conditions
7  * are met:
8  * 1. Redistributions of source code must retain the above copyright
9  *    notice, this list of conditions and the following disclaimer.
10  * 2. Redistributions in binary form must reproduce the above copyright
11  *    notice, this list of conditions and the following disclaimer in the
12  *    documentation and/or other materials provided with the distribution.
13  *
14  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
15  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
16  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
17  * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
18  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
19  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
20  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
21  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
22  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
23  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
24  * SUCH DAMAGE.
25  *
26  */
27 
28 #include <sys/cdefs.h>
29 __FBSDID("$FreeBSD$");
30 
31 #ifndef APPLEKEXT
32 #include <fs/nfs/nfsport.h>
33 
34 struct nfsrv_stablefirst nfsrv_stablefirst;
35 int nfsrv_issuedelegs = 0;
36 int nfsrv_dolocallocks = 0;
37 struct nfsv4lock nfsv4rootfs_lock;
38 
39 extern int newnfs_numnfsd;
40 extern struct nfsstats newnfsstats;
41 extern int nfsrv_lease;
42 extern struct timeval nfsboottime;
43 extern u_int32_t newnfs_true, newnfs_false;
44 NFSV4ROOTLOCKMUTEX;
45 NFSSTATESPINLOCK;
46 
47 /*
48  * Hash lists for nfs V4.
49  * (Some would put them in the .h file, but I don't like declaring storage
50  *  in a .h)
51  */
52 struct nfsclienthashhead nfsclienthash[NFSCLIENTHASHSIZE];
53 struct nfslockhashhead nfslockhash[NFSLOCKHASHSIZE];
54 struct nfssessionhash nfssessionhash[NFSSESSIONHASHSIZE];
55 #endif	/* !APPLEKEXT */
56 
57 static u_int32_t nfsrv_openpluslock = 0, nfsrv_delegatecnt = 0;
58 static time_t nfsrvboottime;
59 static int nfsrv_writedelegifpos = 1;
60 static int nfsrv_returnoldstateid = 0, nfsrv_clients = 0;
61 static int nfsrv_clienthighwater = NFSRV_CLIENTHIGHWATER;
62 static int nfsrv_nogsscallback = 0;
63 
64 /* local functions */
65 static void nfsrv_dumpaclient(struct nfsclient *clp,
66     struct nfsd_dumpclients *dumpp);
67 static void nfsrv_freeopenowner(struct nfsstate *stp, int cansleep,
68     NFSPROC_T *p);
69 static int nfsrv_freeopen(struct nfsstate *stp, vnode_t vp, int cansleep,
70     NFSPROC_T *p);
71 static void nfsrv_freelockowner(struct nfsstate *stp, vnode_t vp, int cansleep,
72     NFSPROC_T *p);
73 static void nfsrv_freeallnfslocks(struct nfsstate *stp, vnode_t vp,
74     int cansleep, NFSPROC_T *p);
75 static void nfsrv_freenfslock(struct nfslock *lop);
76 static void nfsrv_freenfslockfile(struct nfslockfile *lfp);
77 static void nfsrv_freedeleg(struct nfsstate *);
78 static int nfsrv_getstate(struct nfsclient *clp, nfsv4stateid_t *stateidp,
79     u_int32_t flags, struct nfsstate **stpp);
80 static void nfsrv_getowner(struct nfsstatehead *hp, struct nfsstate *new_stp,
81     struct nfsstate **stpp);
82 static int nfsrv_getlockfh(vnode_t vp, u_short flags,
83     struct nfslockfile *new_lfp, fhandle_t *nfhp, NFSPROC_T *p);
84 static int nfsrv_getlockfile(u_short flags, struct nfslockfile **new_lfpp,
85     struct nfslockfile **lfpp, fhandle_t *nfhp, int lockit);
86 static void nfsrv_insertlock(struct nfslock *new_lop,
87     struct nfslock *insert_lop, struct nfsstate *stp, struct nfslockfile *lfp);
88 static void nfsrv_updatelock(struct nfsstate *stp, struct nfslock **new_lopp,
89     struct nfslock **other_lopp, struct nfslockfile *lfp);
90 static int nfsrv_getipnumber(u_char *cp);
91 static int nfsrv_checkrestart(nfsquad_t clientid, u_int32_t flags,
92     nfsv4stateid_t *stateidp, int specialid);
93 static int nfsrv_checkgrace(struct nfsrv_descript *nd, struct nfsclient *clp,
94     u_int32_t flags);
95 static int nfsrv_docallback(struct nfsclient *clp, int procnum,
96     nfsv4stateid_t *stateidp, int trunc, fhandle_t *fhp,
97     struct nfsvattr *nap, nfsattrbit_t *attrbitp, NFSPROC_T *p);
98 static int nfsrv_cbcallargs(struct nfsrv_descript *nd, struct nfsclient *clp,
99     uint32_t callback, int op, const char *optag, struct nfsdsession **sepp);
100 static u_int32_t nfsrv_nextclientindex(void);
101 static u_int32_t nfsrv_nextstateindex(struct nfsclient *clp);
102 static void nfsrv_markstable(struct nfsclient *clp);
103 static int nfsrv_checkstable(struct nfsclient *clp);
104 static int nfsrv_clientconflict(struct nfsclient *clp, int *haslockp, struct
105     vnode *vp, NFSPROC_T *p);
106 static int nfsrv_delegconflict(struct nfsstate *stp, int *haslockp,
107     NFSPROC_T *p, vnode_t vp);
108 static int nfsrv_cleandeleg(vnode_t vp, struct nfslockfile *lfp,
109     struct nfsclient *clp, int *haslockp, NFSPROC_T *p);
110 static int nfsrv_notsamecredname(struct nfsrv_descript *nd,
111     struct nfsclient *clp);
112 static time_t nfsrv_leaseexpiry(void);
113 static void nfsrv_delaydelegtimeout(struct nfsstate *stp);
114 static int nfsrv_checkseqid(struct nfsrv_descript *nd, u_int32_t seqid,
115     struct nfsstate *stp, struct nfsrvcache *op);
116 static int nfsrv_nootherstate(struct nfsstate *stp);
117 static int nfsrv_locallock(vnode_t vp, struct nfslockfile *lfp, int flags,
118     uint64_t first, uint64_t end, struct nfslockconflict *cfp, NFSPROC_T *p);
119 static void nfsrv_localunlock(vnode_t vp, struct nfslockfile *lfp,
120     uint64_t init_first, uint64_t init_end, NFSPROC_T *p);
121 static int nfsrv_dolocal(vnode_t vp, struct nfslockfile *lfp, int flags,
122     int oldflags, uint64_t first, uint64_t end, struct nfslockconflict *cfp,
123     NFSPROC_T *p);
124 static void nfsrv_locallock_rollback(vnode_t vp, struct nfslockfile *lfp,
125     NFSPROC_T *p);
126 static void nfsrv_locallock_commit(struct nfslockfile *lfp, int flags,
127     uint64_t first, uint64_t end);
128 static void nfsrv_locklf(struct nfslockfile *lfp);
129 static void nfsrv_unlocklf(struct nfslockfile *lfp);
130 static struct nfsdsession *nfsrv_findsession(uint8_t *sessionid);
131 static int nfsrv_freesession(struct nfsdsession *sep, uint8_t *sessionid);
132 static int nfsv4_setcbsequence(struct nfsrv_descript *nd, struct nfsclient *clp,
133     int dont_replycache, struct nfsdsession **sepp);
134 static int nfsv4_getcbsession(struct nfsclient *clp, struct nfsdsession **sepp);
135 
136 /*
137  * Scan the client list for a match and either return the current one,
138  * create a new entry or return an error.
139  * If returning a non-error, the clp structure must either be linked into
140  * the client list or free'd.
141  */
142 APPLESTATIC int
143 nfsrv_setclient(struct nfsrv_descript *nd, struct nfsclient **new_clpp,
144     nfsquad_t *clientidp, nfsquad_t *confirmp, NFSPROC_T *p)
145 {
146 	struct nfsclient *clp = NULL, *new_clp = *new_clpp;
147 	int i, error = 0;
148 	struct nfsstate *stp, *tstp;
149 	struct sockaddr_in *sad, *rad;
150 	int zapit = 0, gotit, hasstate = 0, igotlock;
151 	static u_int64_t confirm_index = 0;
152 
153 	/*
154 	 * Check for state resource limit exceeded.
155 	 */
156 	if (nfsrv_openpluslock > NFSRV_V4STATELIMIT) {
157 		error = NFSERR_RESOURCE;
158 		goto out;
159 	}
160 
161 	if (nfsrv_issuedelegs == 0 ||
162 	    ((nd->nd_flag & ND_GSS) != 0 && nfsrv_nogsscallback != 0))
163 		/*
164 		 * Don't do callbacks when delegations are disabled or
165 		 * for AUTH_GSS unless enabled via nfsrv_nogsscallback.
166 		 * If establishing a callback connection is attempted
167 		 * when a firewall is blocking the callback path, the
168 		 * server may wait too long for the connect attempt to
169 		 * succeed during the Open. Some clients, such as Linux,
170 		 * may timeout and give up on the Open before the server
171 		 * replies. Also, since AUTH_GSS callbacks are not
172 		 * yet interoperability tested, they might cause the
173 		 * server to crap out, if they get past the Init call to
174 		 * the client.
175 		 */
176 		new_clp->lc_program = 0;
177 
178 	/* Lock out other nfsd threads */
179 	NFSLOCKV4ROOTMUTEX();
180 	nfsv4_relref(&nfsv4rootfs_lock);
181 	do {
182 		igotlock = nfsv4_lock(&nfsv4rootfs_lock, 1, NULL,
183 		    NFSV4ROOTLOCKMUTEXPTR, NULL);
184 	} while (!igotlock);
185 	NFSUNLOCKV4ROOTMUTEX();
186 
187 	/*
188 	 * Search for a match in the client list.
189 	 */
190 	gotit = i = 0;
191 	while (i < NFSCLIENTHASHSIZE && !gotit) {
192 	    LIST_FOREACH(clp, &nfsclienthash[i], lc_hash) {
193 		if (new_clp->lc_idlen == clp->lc_idlen &&
194 		    !NFSBCMP(new_clp->lc_id, clp->lc_id, clp->lc_idlen)) {
195 			gotit = 1;
196 			break;
197 		}
198 	    }
199 	    i++;
200 	}
201 	if (!gotit ||
202 	    (clp->lc_flags & (LCL_NEEDSCONFIRM | LCL_ADMINREVOKED))) {
203 		if ((nd->nd_flag & ND_NFSV41) != 0 && confirmp->lval[1] != 0) {
204 			/*
205 			 * For NFSv4.1, if confirmp->lval[1] is non-zero, the
206 			 * client is trying to update a confirmed clientid.
207 			 */
208 			NFSLOCKV4ROOTMUTEX();
209 			nfsv4_unlock(&nfsv4rootfs_lock, 1);
210 			NFSUNLOCKV4ROOTMUTEX();
211 			confirmp->lval[1] = 0;
212 			error = NFSERR_NOENT;
213 			goto out;
214 		}
215 		/*
216 		 * Get rid of the old one.
217 		 */
218 		if (i != NFSCLIENTHASHSIZE) {
219 			LIST_REMOVE(clp, lc_hash);
220 			nfsrv_cleanclient(clp, p);
221 			nfsrv_freedeleglist(&clp->lc_deleg);
222 			nfsrv_freedeleglist(&clp->lc_olddeleg);
223 			zapit = 1;
224 		}
225 		/*
226 		 * Add it after assigning a client id to it.
227 		 */
228 		new_clp->lc_flags |= LCL_NEEDSCONFIRM;
229 		if ((nd->nd_flag & ND_NFSV41) != 0)
230 			new_clp->lc_confirm.lval[0] = confirmp->lval[0] =
231 			    ++confirm_index;
232 		else
233 			confirmp->qval = new_clp->lc_confirm.qval =
234 			    ++confirm_index;
235 		clientidp->lval[0] = new_clp->lc_clientid.lval[0] =
236 		    (u_int32_t)nfsrvboottime;
237 		clientidp->lval[1] = new_clp->lc_clientid.lval[1] =
238 		    nfsrv_nextclientindex();
239 		new_clp->lc_stateindex = 0;
240 		new_clp->lc_statemaxindex = 0;
241 		new_clp->lc_cbref = 0;
242 		new_clp->lc_expiry = nfsrv_leaseexpiry();
243 		LIST_INIT(&new_clp->lc_open);
244 		LIST_INIT(&new_clp->lc_deleg);
245 		LIST_INIT(&new_clp->lc_olddeleg);
246 		LIST_INIT(&new_clp->lc_session);
247 		for (i = 0; i < NFSSTATEHASHSIZE; i++)
248 			LIST_INIT(&new_clp->lc_stateid[i]);
249 		LIST_INSERT_HEAD(NFSCLIENTHASH(new_clp->lc_clientid), new_clp,
250 		    lc_hash);
251 		newnfsstats.srvclients++;
252 		nfsrv_openpluslock++;
253 		nfsrv_clients++;
254 		NFSLOCKV4ROOTMUTEX();
255 		nfsv4_unlock(&nfsv4rootfs_lock, 1);
256 		NFSUNLOCKV4ROOTMUTEX();
257 		if (zapit)
258 			nfsrv_zapclient(clp, p);
259 		*new_clpp = NULL;
260 		goto out;
261 	}
262 
263 	/*
264 	 * Now, handle the cases where the id is already issued.
265 	 */
266 	if (nfsrv_notsamecredname(nd, clp)) {
267 	    /*
268 	     * Check to see if there is expired state that should go away.
269 	     */
270 	    if (clp->lc_expiry < NFSD_MONOSEC &&
271 	        (!LIST_EMPTY(&clp->lc_open) || !LIST_EMPTY(&clp->lc_deleg))) {
272 		nfsrv_cleanclient(clp, p);
273 		nfsrv_freedeleglist(&clp->lc_deleg);
274 	    }
275 
276 	    /*
277 	     * If there is outstanding state, then reply NFSERR_CLIDINUSE per
278 	     * RFC3530 Sec. 8.1.2 last para.
279 	     */
280 	    if (!LIST_EMPTY(&clp->lc_deleg)) {
281 		hasstate = 1;
282 	    } else if (LIST_EMPTY(&clp->lc_open)) {
283 		hasstate = 0;
284 	    } else {
285 		hasstate = 0;
286 		/* Look for an Open on the OpenOwner */
287 		LIST_FOREACH(stp, &clp->lc_open, ls_list) {
288 		    if (!LIST_EMPTY(&stp->ls_open)) {
289 			hasstate = 1;
290 			break;
291 		    }
292 		}
293 	    }
294 	    if (hasstate) {
295 		/*
296 		 * If the uid doesn't match, return NFSERR_CLIDINUSE after
297 		 * filling out the correct ipaddr and portnum.
298 		 */
299 		sad = NFSSOCKADDR(new_clp->lc_req.nr_nam, struct sockaddr_in *);
300 		rad = NFSSOCKADDR(clp->lc_req.nr_nam, struct sockaddr_in *);
301 		sad->sin_addr.s_addr = rad->sin_addr.s_addr;
302 		sad->sin_port = rad->sin_port;
303 		NFSLOCKV4ROOTMUTEX();
304 		nfsv4_unlock(&nfsv4rootfs_lock, 1);
305 		NFSUNLOCKV4ROOTMUTEX();
306 		error = NFSERR_CLIDINUSE;
307 		goto out;
308 	    }
309 	}
310 
311 	if (NFSBCMP(new_clp->lc_verf, clp->lc_verf, NFSX_VERF)) {
312 		/*
313 		 * If the verifier has changed, the client has rebooted
314 		 * and a new client id is issued. The old state info
315 		 * can be thrown away once the SETCLIENTID_CONFIRM occurs.
316 		 */
317 		LIST_REMOVE(clp, lc_hash);
318 		new_clp->lc_flags |= LCL_NEEDSCONFIRM;
319 		if ((nd->nd_flag & ND_NFSV41) != 0)
320 			new_clp->lc_confirm.lval[0] = confirmp->lval[0] =
321 			    ++confirm_index;
322 		else
323 			confirmp->qval = new_clp->lc_confirm.qval =
324 			    ++confirm_index;
325 		clientidp->lval[0] = new_clp->lc_clientid.lval[0] =
326 		    nfsrvboottime;
327 		clientidp->lval[1] = new_clp->lc_clientid.lval[1] =
328 		    nfsrv_nextclientindex();
329 		new_clp->lc_stateindex = 0;
330 		new_clp->lc_statemaxindex = 0;
331 		new_clp->lc_cbref = 0;
332 		new_clp->lc_expiry = nfsrv_leaseexpiry();
333 
334 		/*
335 		 * Save the state until confirmed.
336 		 */
337 		LIST_NEWHEAD(&new_clp->lc_open, &clp->lc_open, ls_list);
338 		LIST_FOREACH(tstp, &new_clp->lc_open, ls_list)
339 			tstp->ls_clp = new_clp;
340 		LIST_NEWHEAD(&new_clp->lc_deleg, &clp->lc_deleg, ls_list);
341 		LIST_FOREACH(tstp, &new_clp->lc_deleg, ls_list)
342 			tstp->ls_clp = new_clp;
343 		LIST_NEWHEAD(&new_clp->lc_olddeleg, &clp->lc_olddeleg,
344 		    ls_list);
345 		LIST_FOREACH(tstp, &new_clp->lc_olddeleg, ls_list)
346 			tstp->ls_clp = new_clp;
347 		for (i = 0; i < NFSSTATEHASHSIZE; i++) {
348 			LIST_NEWHEAD(&new_clp->lc_stateid[i],
349 			    &clp->lc_stateid[i], ls_hash);
350 			LIST_FOREACH(tstp, &new_clp->lc_stateid[i], ls_hash)
351 				tstp->ls_clp = new_clp;
352 		}
353 		LIST_INSERT_HEAD(NFSCLIENTHASH(new_clp->lc_clientid), new_clp,
354 		    lc_hash);
355 		newnfsstats.srvclients++;
356 		nfsrv_openpluslock++;
357 		nfsrv_clients++;
358 		NFSLOCKV4ROOTMUTEX();
359 		nfsv4_unlock(&nfsv4rootfs_lock, 1);
360 		NFSUNLOCKV4ROOTMUTEX();
361 
362 		/*
363 		 * Must wait until any outstanding callback on the old clp
364 		 * completes.
365 		 */
366 		NFSLOCKSTATE();
367 		while (clp->lc_cbref) {
368 			clp->lc_flags |= LCL_WAKEUPWANTED;
369 			(void)mtx_sleep(clp, NFSSTATEMUTEXPTR, PZERO - 1,
370 			    "nfsd clp", 10 * hz);
371 		}
372 		NFSUNLOCKSTATE();
373 		nfsrv_zapclient(clp, p);
374 		*new_clpp = NULL;
375 		goto out;
376 	}
377 
378 	/* For NFSv4.1, mark that we found a confirmed clientid. */
379 	if ((nd->nd_flag & ND_NFSV41) != 0)
380 		confirmp->lval[1] = 1;
381 	else {
382 		/*
383 		 * id and verifier match, so update the net address info
384 		 * and get rid of any existing callback authentication
385 		 * handle, so a new one will be acquired.
386 		 */
387 		LIST_REMOVE(clp, lc_hash);
388 		new_clp->lc_flags |= (LCL_NEEDSCONFIRM | LCL_DONTCLEAN);
389 		new_clp->lc_expiry = nfsrv_leaseexpiry();
390 		confirmp->qval = new_clp->lc_confirm.qval = ++confirm_index;
391 		clientidp->lval[0] = new_clp->lc_clientid.lval[0] =
392 		    clp->lc_clientid.lval[0];
393 		clientidp->lval[1] = new_clp->lc_clientid.lval[1] =
394 		    clp->lc_clientid.lval[1];
395 		new_clp->lc_delegtime = clp->lc_delegtime;
396 		new_clp->lc_stateindex = clp->lc_stateindex;
397 		new_clp->lc_statemaxindex = clp->lc_statemaxindex;
398 		new_clp->lc_cbref = 0;
399 		LIST_NEWHEAD(&new_clp->lc_open, &clp->lc_open, ls_list);
400 		LIST_FOREACH(tstp, &new_clp->lc_open, ls_list)
401 			tstp->ls_clp = new_clp;
402 		LIST_NEWHEAD(&new_clp->lc_deleg, &clp->lc_deleg, ls_list);
403 		LIST_FOREACH(tstp, &new_clp->lc_deleg, ls_list)
404 			tstp->ls_clp = new_clp;
405 		LIST_NEWHEAD(&new_clp->lc_olddeleg, &clp->lc_olddeleg, ls_list);
406 		LIST_FOREACH(tstp, &new_clp->lc_olddeleg, ls_list)
407 			tstp->ls_clp = new_clp;
408 		for (i = 0; i < NFSSTATEHASHSIZE; i++) {
409 			LIST_NEWHEAD(&new_clp->lc_stateid[i],
410 			    &clp->lc_stateid[i], ls_hash);
411 			LIST_FOREACH(tstp, &new_clp->lc_stateid[i], ls_hash)
412 				tstp->ls_clp = new_clp;
413 		}
414 		LIST_INSERT_HEAD(NFSCLIENTHASH(new_clp->lc_clientid), new_clp,
415 		    lc_hash);
416 		newnfsstats.srvclients++;
417 		nfsrv_openpluslock++;
418 		nfsrv_clients++;
419 	}
420 	NFSLOCKV4ROOTMUTEX();
421 	nfsv4_unlock(&nfsv4rootfs_lock, 1);
422 	NFSUNLOCKV4ROOTMUTEX();
423 
424 	if ((nd->nd_flag & ND_NFSV41) == 0) {
425 		/*
426 		 * Must wait until any outstanding callback on the old clp
427 		 * completes.
428 		 */
429 		NFSLOCKSTATE();
430 		while (clp->lc_cbref) {
431 			clp->lc_flags |= LCL_WAKEUPWANTED;
432 			(void)mtx_sleep(clp, NFSSTATEMUTEXPTR, PZERO - 1,
433 			    "nfsdclp", 10 * hz);
434 		}
435 		NFSUNLOCKSTATE();
436 		nfsrv_zapclient(clp, p);
437 		*new_clpp = NULL;
438 	}
439 
440 out:
441 	NFSEXITCODE2(error, nd);
442 	return (error);
443 }
444 
445 /*
446  * Check to see if the client id exists and optionally confirm it.
447  */
448 APPLESTATIC int
449 nfsrv_getclient(nfsquad_t clientid, int opflags, struct nfsclient **clpp,
450     struct nfsdsession *nsep, nfsquad_t confirm, uint32_t cbprogram,
451     struct nfsrv_descript *nd, NFSPROC_T *p)
452 {
453 	struct nfsclient *clp;
454 	struct nfsstate *stp;
455 	int i;
456 	struct nfsclienthashhead *hp;
457 	int error = 0, igotlock, doneok;
458 	struct nfssessionhash *shp;
459 	struct nfsdsession *sep;
460 	uint64_t sessid[2];
461 	static uint64_t next_sess = 0;
462 
463 	if (clpp)
464 		*clpp = NULL;
465 	if ((nd == NULL || (nd->nd_flag & ND_NFSV41) == 0 ||
466 	    opflags != CLOPS_RENEW) && nfsrvboottime != clientid.lval[0]) {
467 		error = NFSERR_STALECLIENTID;
468 		goto out;
469 	}
470 
471 	/*
472 	 * If called with opflags == CLOPS_RENEW, the State Lock is
473 	 * already held. Otherwise, we need to get either that or,
474 	 * for the case of Confirm, lock out the nfsd threads.
475 	 */
476 	if (opflags & CLOPS_CONFIRM) {
477 		NFSLOCKV4ROOTMUTEX();
478 		nfsv4_relref(&nfsv4rootfs_lock);
479 		do {
480 			igotlock = nfsv4_lock(&nfsv4rootfs_lock, 1, NULL,
481 			    NFSV4ROOTLOCKMUTEXPTR, NULL);
482 		} while (!igotlock);
483 		/*
484 		 * Create a new sessionid here, since we need to do it where
485 		 * there is a mutex held to serialize update of next_sess.
486 		 */
487 		if ((nd->nd_flag & ND_NFSV41) != 0) {
488 			sessid[0] = ++next_sess;
489 			sessid[1] = clientid.qval;
490 		}
491 		NFSUNLOCKV4ROOTMUTEX();
492 	} else if (opflags != CLOPS_RENEW) {
493 		NFSLOCKSTATE();
494 	}
495 
496 	/* For NFSv4.1, the clp is acquired from the associated session. */
497 	if (nd != NULL && (nd->nd_flag & ND_NFSV41) != 0 &&
498 	    opflags == CLOPS_RENEW) {
499 		clp = NULL;
500 		if ((nd->nd_flag & ND_HASSEQUENCE) != 0) {
501 			shp = NFSSESSIONHASH(nd->nd_sessionid);
502 			NFSLOCKSESSION(shp);
503 			sep = nfsrv_findsession(nd->nd_sessionid);
504 			if (sep != NULL)
505 				clp = sep->sess_clp;
506 			NFSUNLOCKSESSION(shp);
507 		}
508 	} else {
509 		hp = NFSCLIENTHASH(clientid);
510 		LIST_FOREACH(clp, hp, lc_hash) {
511 			if (clp->lc_clientid.lval[1] == clientid.lval[1])
512 				break;
513 		}
514 	}
515 	if (clp == NULL) {
516 		if (opflags & CLOPS_CONFIRM)
517 			error = NFSERR_STALECLIENTID;
518 		else
519 			error = NFSERR_EXPIRED;
520 	} else if (clp->lc_flags & LCL_ADMINREVOKED) {
521 		/*
522 		 * If marked admin revoked, just return the error.
523 		 */
524 		error = NFSERR_ADMINREVOKED;
525 	}
526 	if (error) {
527 		if (opflags & CLOPS_CONFIRM) {
528 			NFSLOCKV4ROOTMUTEX();
529 			nfsv4_unlock(&nfsv4rootfs_lock, 1);
530 			NFSUNLOCKV4ROOTMUTEX();
531 		} else if (opflags != CLOPS_RENEW) {
532 			NFSUNLOCKSTATE();
533 		}
534 		goto out;
535 	}
536 
537 	/*
538 	 * Perform any operations specified by the opflags.
539 	 */
540 	if (opflags & CLOPS_CONFIRM) {
541 		if (((nd->nd_flag & ND_NFSV41) != 0 &&
542 		     clp->lc_confirm.lval[0] != confirm.lval[0]) ||
543 		    ((nd->nd_flag & ND_NFSV41) == 0 &&
544 		     clp->lc_confirm.qval != confirm.qval))
545 			error = NFSERR_STALECLIENTID;
546 		else if (nfsrv_notsamecredname(nd, clp))
547 			error = NFSERR_CLIDINUSE;
548 
549 		if (!error) {
550 		    if ((clp->lc_flags & (LCL_NEEDSCONFIRM | LCL_DONTCLEAN)) ==
551 			LCL_NEEDSCONFIRM) {
552 			/*
553 			 * Hang onto the delegations (as old delegations)
554 			 * for an Open with CLAIM_DELEGATE_PREV unless in
555 			 * grace, but get rid of the rest of the state.
556 			 */
557 			nfsrv_cleanclient(clp, p);
558 			nfsrv_freedeleglist(&clp->lc_olddeleg);
559 			if (nfsrv_checkgrace(nd, clp, 0)) {
560 			    /* In grace, so just delete delegations */
561 			    nfsrv_freedeleglist(&clp->lc_deleg);
562 			} else {
563 			    LIST_FOREACH(stp, &clp->lc_deleg, ls_list)
564 				stp->ls_flags |= NFSLCK_OLDDELEG;
565 			    clp->lc_delegtime = NFSD_MONOSEC +
566 				nfsrv_lease + NFSRV_LEASEDELTA;
567 			    LIST_NEWHEAD(&clp->lc_olddeleg, &clp->lc_deleg,
568 				ls_list);
569 			}
570 			if ((nd->nd_flag & ND_NFSV41) != 0)
571 			    clp->lc_program = cbprogram;
572 		    }
573 		    clp->lc_flags &= ~(LCL_NEEDSCONFIRM | LCL_DONTCLEAN);
574 		    if (clp->lc_program)
575 			clp->lc_flags |= LCL_NEEDSCBNULL;
576 		    /* For NFSv4.1, link the session onto the client. */
577 		    if (nsep != NULL) {
578 			/* Hold a reference on the xprt for a backchannel. */
579 			if ((nsep->sess_crflags & NFSV4CRSESS_CONNBACKCHAN)
580 			    != 0 && clp->lc_req.nr_client == NULL) {
581 			    clp->lc_req.nr_client = (struct __rpc_client *)
582 				clnt_bck_create(nd->nd_xprt->xp_socket,
583 				cbprogram, NFSV4_CBVERS);
584 			    if (clp->lc_req.nr_client != NULL) {
585 				SVC_ACQUIRE(nd->nd_xprt);
586 				nd->nd_xprt->xp_p2 =
587 				    clp->lc_req.nr_client->cl_private;
588 				/* Disable idle timeout. */
589 				nd->nd_xprt->xp_idletimeout = 0;
590 				nsep->sess_cbsess.nfsess_xprt = nd->nd_xprt;
591 			    } else
592 				nsep->sess_crflags &= ~NFSV4CRSESS_CONNBACKCHAN;
593 			}
594 			NFSBCOPY(sessid, nsep->sess_sessionid,
595 			    NFSX_V4SESSIONID);
596 			NFSBCOPY(sessid, nsep->sess_cbsess.nfsess_sessionid,
597 			    NFSX_V4SESSIONID);
598 			shp = NFSSESSIONHASH(nsep->sess_sessionid);
599 			NFSLOCKSESSION(shp);
600 			LIST_INSERT_HEAD(&shp->list, nsep, sess_hash);
601 			NFSLOCKSTATE();
602 			LIST_INSERT_HEAD(&clp->lc_session, nsep, sess_list);
603 			nsep->sess_clp = clp;
604 			NFSUNLOCKSTATE();
605 			NFSUNLOCKSESSION(shp);
606 		    }
607 		}
608 	} else if (clp->lc_flags & LCL_NEEDSCONFIRM) {
609 		error = NFSERR_EXPIRED;
610 	}
611 
612 	/*
613 	 * If called by the Renew Op, we must check the principal.
614 	 */
615 	if (!error && (opflags & CLOPS_RENEWOP)) {
616 	    if (nfsrv_notsamecredname(nd, clp)) {
617 		doneok = 0;
618 		for (i = 0; i < NFSSTATEHASHSIZE && doneok == 0; i++) {
619 		    LIST_FOREACH(stp, &clp->lc_stateid[i], ls_hash) {
620 			if ((stp->ls_flags & NFSLCK_OPEN) &&
621 			    stp->ls_uid == nd->nd_cred->cr_uid) {
622 				doneok = 1;
623 				break;
624 			}
625 		    }
626 		}
627 		if (!doneok)
628 			error = NFSERR_ACCES;
629 	    }
630 	    if (!error && (clp->lc_flags & LCL_CBDOWN))
631 		error = NFSERR_CBPATHDOWN;
632 	}
633 	if ((!error || error == NFSERR_CBPATHDOWN) &&
634 	     (opflags & CLOPS_RENEW)) {
635 		clp->lc_expiry = nfsrv_leaseexpiry();
636 	}
637 	if (opflags & CLOPS_CONFIRM) {
638 		NFSLOCKV4ROOTMUTEX();
639 		nfsv4_unlock(&nfsv4rootfs_lock, 1);
640 		NFSUNLOCKV4ROOTMUTEX();
641 	} else if (opflags != CLOPS_RENEW) {
642 		NFSUNLOCKSTATE();
643 	}
644 	if (clpp)
645 		*clpp = clp;
646 
647 out:
648 	NFSEXITCODE2(error, nd);
649 	return (error);
650 }
651 
652 /*
653  * Perform the NFSv4.1 destroy clientid.
654  */
655 int
656 nfsrv_destroyclient(nfsquad_t clientid, NFSPROC_T *p)
657 {
658 	struct nfsclient *clp;
659 	struct nfsclienthashhead *hp;
660 	int error = 0, i, igotlock;
661 
662 	if (nfsrvboottime != clientid.lval[0]) {
663 		error = NFSERR_STALECLIENTID;
664 		goto out;
665 	}
666 
667 	/* Lock out other nfsd threads */
668 	NFSLOCKV4ROOTMUTEX();
669 	nfsv4_relref(&nfsv4rootfs_lock);
670 	do {
671 		igotlock = nfsv4_lock(&nfsv4rootfs_lock, 1, NULL,
672 		    NFSV4ROOTLOCKMUTEXPTR, NULL);
673 	} while (igotlock == 0);
674 	NFSUNLOCKV4ROOTMUTEX();
675 
676 	hp = NFSCLIENTHASH(clientid);
677 	LIST_FOREACH(clp, hp, lc_hash) {
678 		if (clp->lc_clientid.lval[1] == clientid.lval[1])
679 			break;
680 	}
681 	if (clp == NULL) {
682 		NFSLOCKV4ROOTMUTEX();
683 		nfsv4_unlock(&nfsv4rootfs_lock, 1);
684 		NFSUNLOCKV4ROOTMUTEX();
685 		/* Just return ok, since it is gone. */
686 		goto out;
687 	}
688 
689 	/* Scan for state on the clientid. */
690 	for (i = 0; i < NFSSTATEHASHSIZE; i++)
691 		if (!LIST_EMPTY(&clp->lc_stateid[i])) {
692 			NFSLOCKV4ROOTMUTEX();
693 			nfsv4_unlock(&nfsv4rootfs_lock, 1);
694 			NFSUNLOCKV4ROOTMUTEX();
695 			error = NFSERR_CLIENTIDBUSY;
696 			goto out;
697 		}
698 	if (!LIST_EMPTY(&clp->lc_session) || !LIST_EMPTY(&clp->lc_deleg)) {
699 		NFSLOCKV4ROOTMUTEX();
700 		nfsv4_unlock(&nfsv4rootfs_lock, 1);
701 		NFSUNLOCKV4ROOTMUTEX();
702 		error = NFSERR_CLIENTIDBUSY;
703 		goto out;
704 	}
705 
706 	/* Destroy the clientid and return ok. */
707 	nfsrv_cleanclient(clp, p);
708 	nfsrv_freedeleglist(&clp->lc_deleg);
709 	nfsrv_freedeleglist(&clp->lc_olddeleg);
710 	LIST_REMOVE(clp, lc_hash);
711 	NFSLOCKV4ROOTMUTEX();
712 	nfsv4_unlock(&nfsv4rootfs_lock, 1);
713 	NFSUNLOCKV4ROOTMUTEX();
714 	nfsrv_zapclient(clp, p);
715 out:
716 	NFSEXITCODE2(error, nd);
717 	return (error);
718 }
719 
720 /*
721  * Called from the new nfssvc syscall to admin revoke a clientid.
722  * Returns 0 for success, error otherwise.
723  */
724 APPLESTATIC int
725 nfsrv_adminrevoke(struct nfsd_clid *revokep, NFSPROC_T *p)
726 {
727 	struct nfsclient *clp = NULL;
728 	int i, error = 0;
729 	int gotit, igotlock;
730 
731 	/*
732 	 * First, lock out the nfsd so that state won't change while the
733 	 * revocation record is being written to the stable storage restart
734 	 * file.
735 	 */
736 	NFSLOCKV4ROOTMUTEX();
737 	do {
738 		igotlock = nfsv4_lock(&nfsv4rootfs_lock, 1, NULL,
739 		    NFSV4ROOTLOCKMUTEXPTR, NULL);
740 	} while (!igotlock);
741 	NFSUNLOCKV4ROOTMUTEX();
742 
743 	/*
744 	 * Search for a match in the client list.
745 	 */
746 	gotit = i = 0;
747 	while (i < NFSCLIENTHASHSIZE && !gotit) {
748 	    LIST_FOREACH(clp, &nfsclienthash[i], lc_hash) {
749 		if (revokep->nclid_idlen == clp->lc_idlen &&
750 		    !NFSBCMP(revokep->nclid_id, clp->lc_id, clp->lc_idlen)) {
751 			gotit = 1;
752 			break;
753 		}
754 	    }
755 	    i++;
756 	}
757 	if (!gotit) {
758 		NFSLOCKV4ROOTMUTEX();
759 		nfsv4_unlock(&nfsv4rootfs_lock, 0);
760 		NFSUNLOCKV4ROOTMUTEX();
761 		error = EPERM;
762 		goto out;
763 	}
764 
765 	/*
766 	 * Now, write out the revocation record
767 	 */
768 	nfsrv_writestable(clp->lc_id, clp->lc_idlen, NFSNST_REVOKE, p);
769 	nfsrv_backupstable();
770 
771 	/*
772 	 * and clear out the state, marking the clientid revoked.
773 	 */
774 	clp->lc_flags &= ~LCL_CALLBACKSON;
775 	clp->lc_flags |= LCL_ADMINREVOKED;
776 	nfsrv_cleanclient(clp, p);
777 	nfsrv_freedeleglist(&clp->lc_deleg);
778 	nfsrv_freedeleglist(&clp->lc_olddeleg);
779 	NFSLOCKV4ROOTMUTEX();
780 	nfsv4_unlock(&nfsv4rootfs_lock, 0);
781 	NFSUNLOCKV4ROOTMUTEX();
782 
783 out:
784 	NFSEXITCODE(error);
785 	return (error);
786 }
787 
788 /*
789  * Dump out stats for all clients. Called from nfssvc(2), that is used
790  * newnfsstats.
791  */
792 APPLESTATIC void
793 nfsrv_dumpclients(struct nfsd_dumpclients *dumpp, int maxcnt)
794 {
795 	struct nfsclient *clp;
796 	int i = 0, cnt = 0;
797 
798 	/*
799 	 * First, get a reference on the nfsv4rootfs_lock so that an
800 	 * exclusive lock cannot be acquired while dumping the clients.
801 	 */
802 	NFSLOCKV4ROOTMUTEX();
803 	nfsv4_getref(&nfsv4rootfs_lock, NULL, NFSV4ROOTLOCKMUTEXPTR, NULL);
804 	NFSUNLOCKV4ROOTMUTEX();
805 	NFSLOCKSTATE();
806 	/*
807 	 * Rattle through the client lists until done.
808 	 */
809 	while (i < NFSCLIENTHASHSIZE && cnt < maxcnt) {
810 	    clp = LIST_FIRST(&nfsclienthash[i]);
811 	    while (clp != LIST_END(&nfsclienthash[i]) && cnt < maxcnt) {
812 		nfsrv_dumpaclient(clp, &dumpp[cnt]);
813 		cnt++;
814 		clp = LIST_NEXT(clp, lc_hash);
815 	    }
816 	    i++;
817 	}
818 	if (cnt < maxcnt)
819 	    dumpp[cnt].ndcl_clid.nclid_idlen = 0;
820 	NFSUNLOCKSTATE();
821 	NFSLOCKV4ROOTMUTEX();
822 	nfsv4_relref(&nfsv4rootfs_lock);
823 	NFSUNLOCKV4ROOTMUTEX();
824 }
825 
826 /*
827  * Dump stats for a client. Must be called with the NFSSTATELOCK and spl'd.
828  */
829 static void
830 nfsrv_dumpaclient(struct nfsclient *clp, struct nfsd_dumpclients *dumpp)
831 {
832 	struct nfsstate *stp, *openstp, *lckownstp;
833 	struct nfslock *lop;
834 	struct sockaddr *sad;
835 	struct sockaddr_in *rad;
836 	struct sockaddr_in6 *rad6;
837 
838 	dumpp->ndcl_nopenowners = dumpp->ndcl_nlockowners = 0;
839 	dumpp->ndcl_nopens = dumpp->ndcl_nlocks = 0;
840 	dumpp->ndcl_ndelegs = dumpp->ndcl_nolddelegs = 0;
841 	dumpp->ndcl_flags = clp->lc_flags;
842 	dumpp->ndcl_clid.nclid_idlen = clp->lc_idlen;
843 	NFSBCOPY(clp->lc_id, dumpp->ndcl_clid.nclid_id, clp->lc_idlen);
844 	sad = NFSSOCKADDR(clp->lc_req.nr_nam, struct sockaddr *);
845 	dumpp->ndcl_addrfam = sad->sa_family;
846 	if (sad->sa_family == AF_INET) {
847 		rad = (struct sockaddr_in *)sad;
848 		dumpp->ndcl_cbaddr.sin_addr = rad->sin_addr;
849 	} else {
850 		rad6 = (struct sockaddr_in6 *)sad;
851 		dumpp->ndcl_cbaddr.sin6_addr = rad6->sin6_addr;
852 	}
853 
854 	/*
855 	 * Now, scan the state lists and total up the opens and locks.
856 	 */
857 	LIST_FOREACH(stp, &clp->lc_open, ls_list) {
858 	    dumpp->ndcl_nopenowners++;
859 	    LIST_FOREACH(openstp, &stp->ls_open, ls_list) {
860 		dumpp->ndcl_nopens++;
861 		LIST_FOREACH(lckownstp, &openstp->ls_open, ls_list) {
862 		    dumpp->ndcl_nlockowners++;
863 		    LIST_FOREACH(lop, &lckownstp->ls_lock, lo_lckowner) {
864 			dumpp->ndcl_nlocks++;
865 		    }
866 		}
867 	    }
868 	}
869 
870 	/*
871 	 * and the delegation lists.
872 	 */
873 	LIST_FOREACH(stp, &clp->lc_deleg, ls_list) {
874 	    dumpp->ndcl_ndelegs++;
875 	}
876 	LIST_FOREACH(stp, &clp->lc_olddeleg, ls_list) {
877 	    dumpp->ndcl_nolddelegs++;
878 	}
879 }
880 
881 /*
882  * Dump out lock stats for a file.
883  */
884 APPLESTATIC void
885 nfsrv_dumplocks(vnode_t vp, struct nfsd_dumplocks *ldumpp, int maxcnt,
886     NFSPROC_T *p)
887 {
888 	struct nfsstate *stp;
889 	struct nfslock *lop;
890 	int cnt = 0;
891 	struct nfslockfile *lfp;
892 	struct sockaddr *sad;
893 	struct sockaddr_in *rad;
894 	struct sockaddr_in6 *rad6;
895 	int ret;
896 	fhandle_t nfh;
897 
898 	ret = nfsrv_getlockfh(vp, 0, NULL, &nfh, p);
899 	/*
900 	 * First, get a reference on the nfsv4rootfs_lock so that an
901 	 * exclusive lock on it cannot be acquired while dumping the locks.
902 	 */
903 	NFSLOCKV4ROOTMUTEX();
904 	nfsv4_getref(&nfsv4rootfs_lock, NULL, NFSV4ROOTLOCKMUTEXPTR, NULL);
905 	NFSUNLOCKV4ROOTMUTEX();
906 	NFSLOCKSTATE();
907 	if (!ret)
908 		ret = nfsrv_getlockfile(0, NULL, &lfp, &nfh, 0);
909 	if (ret) {
910 		ldumpp[0].ndlck_clid.nclid_idlen = 0;
911 		NFSUNLOCKSTATE();
912 		NFSLOCKV4ROOTMUTEX();
913 		nfsv4_relref(&nfsv4rootfs_lock);
914 		NFSUNLOCKV4ROOTMUTEX();
915 		return;
916 	}
917 
918 	/*
919 	 * For each open share on file, dump it out.
920 	 */
921 	stp = LIST_FIRST(&lfp->lf_open);
922 	while (stp != LIST_END(&lfp->lf_open) && cnt < maxcnt) {
923 		ldumpp[cnt].ndlck_flags = stp->ls_flags;
924 		ldumpp[cnt].ndlck_stateid.seqid = stp->ls_stateid.seqid;
925 		ldumpp[cnt].ndlck_stateid.other[0] = stp->ls_stateid.other[0];
926 		ldumpp[cnt].ndlck_stateid.other[1] = stp->ls_stateid.other[1];
927 		ldumpp[cnt].ndlck_stateid.other[2] = stp->ls_stateid.other[2];
928 		ldumpp[cnt].ndlck_owner.nclid_idlen =
929 		    stp->ls_openowner->ls_ownerlen;
930 		NFSBCOPY(stp->ls_openowner->ls_owner,
931 		    ldumpp[cnt].ndlck_owner.nclid_id,
932 		    stp->ls_openowner->ls_ownerlen);
933 		ldumpp[cnt].ndlck_clid.nclid_idlen = stp->ls_clp->lc_idlen;
934 		NFSBCOPY(stp->ls_clp->lc_id, ldumpp[cnt].ndlck_clid.nclid_id,
935 		    stp->ls_clp->lc_idlen);
936 		sad=NFSSOCKADDR(stp->ls_clp->lc_req.nr_nam, struct sockaddr *);
937 		ldumpp[cnt].ndlck_addrfam = sad->sa_family;
938 		if (sad->sa_family == AF_INET) {
939 			rad = (struct sockaddr_in *)sad;
940 			ldumpp[cnt].ndlck_cbaddr.sin_addr = rad->sin_addr;
941 		} else {
942 			rad6 = (struct sockaddr_in6 *)sad;
943 			ldumpp[cnt].ndlck_cbaddr.sin6_addr = rad6->sin6_addr;
944 		}
945 		stp = LIST_NEXT(stp, ls_file);
946 		cnt++;
947 	}
948 
949 	/*
950 	 * and all locks.
951 	 */
952 	lop = LIST_FIRST(&lfp->lf_lock);
953 	while (lop != LIST_END(&lfp->lf_lock) && cnt < maxcnt) {
954 		stp = lop->lo_stp;
955 		ldumpp[cnt].ndlck_flags = lop->lo_flags;
956 		ldumpp[cnt].ndlck_first = lop->lo_first;
957 		ldumpp[cnt].ndlck_end = lop->lo_end;
958 		ldumpp[cnt].ndlck_stateid.seqid = stp->ls_stateid.seqid;
959 		ldumpp[cnt].ndlck_stateid.other[0] = stp->ls_stateid.other[0];
960 		ldumpp[cnt].ndlck_stateid.other[1] = stp->ls_stateid.other[1];
961 		ldumpp[cnt].ndlck_stateid.other[2] = stp->ls_stateid.other[2];
962 		ldumpp[cnt].ndlck_owner.nclid_idlen = stp->ls_ownerlen;
963 		NFSBCOPY(stp->ls_owner, ldumpp[cnt].ndlck_owner.nclid_id,
964 		    stp->ls_ownerlen);
965 		ldumpp[cnt].ndlck_clid.nclid_idlen = stp->ls_clp->lc_idlen;
966 		NFSBCOPY(stp->ls_clp->lc_id, ldumpp[cnt].ndlck_clid.nclid_id,
967 		    stp->ls_clp->lc_idlen);
968 		sad=NFSSOCKADDR(stp->ls_clp->lc_req.nr_nam, struct sockaddr *);
969 		ldumpp[cnt].ndlck_addrfam = sad->sa_family;
970 		if (sad->sa_family == AF_INET) {
971 			rad = (struct sockaddr_in *)sad;
972 			ldumpp[cnt].ndlck_cbaddr.sin_addr = rad->sin_addr;
973 		} else {
974 			rad6 = (struct sockaddr_in6 *)sad;
975 			ldumpp[cnt].ndlck_cbaddr.sin6_addr = rad6->sin6_addr;
976 		}
977 		lop = LIST_NEXT(lop, lo_lckfile);
978 		cnt++;
979 	}
980 
981 	/*
982 	 * and the delegations.
983 	 */
984 	stp = LIST_FIRST(&lfp->lf_deleg);
985 	while (stp != LIST_END(&lfp->lf_deleg) && cnt < maxcnt) {
986 		ldumpp[cnt].ndlck_flags = stp->ls_flags;
987 		ldumpp[cnt].ndlck_stateid.seqid = stp->ls_stateid.seqid;
988 		ldumpp[cnt].ndlck_stateid.other[0] = stp->ls_stateid.other[0];
989 		ldumpp[cnt].ndlck_stateid.other[1] = stp->ls_stateid.other[1];
990 		ldumpp[cnt].ndlck_stateid.other[2] = stp->ls_stateid.other[2];
991 		ldumpp[cnt].ndlck_owner.nclid_idlen = 0;
992 		ldumpp[cnt].ndlck_clid.nclid_idlen = stp->ls_clp->lc_idlen;
993 		NFSBCOPY(stp->ls_clp->lc_id, ldumpp[cnt].ndlck_clid.nclid_id,
994 		    stp->ls_clp->lc_idlen);
995 		sad=NFSSOCKADDR(stp->ls_clp->lc_req.nr_nam, struct sockaddr *);
996 		ldumpp[cnt].ndlck_addrfam = sad->sa_family;
997 		if (sad->sa_family == AF_INET) {
998 			rad = (struct sockaddr_in *)sad;
999 			ldumpp[cnt].ndlck_cbaddr.sin_addr = rad->sin_addr;
1000 		} else {
1001 			rad6 = (struct sockaddr_in6 *)sad;
1002 			ldumpp[cnt].ndlck_cbaddr.sin6_addr = rad6->sin6_addr;
1003 		}
1004 		stp = LIST_NEXT(stp, ls_file);
1005 		cnt++;
1006 	}
1007 
1008 	/*
1009 	 * If list isn't full, mark end of list by setting the client name
1010 	 * to zero length.
1011 	 */
1012 	if (cnt < maxcnt)
1013 		ldumpp[cnt].ndlck_clid.nclid_idlen = 0;
1014 	NFSUNLOCKSTATE();
1015 	NFSLOCKV4ROOTMUTEX();
1016 	nfsv4_relref(&nfsv4rootfs_lock);
1017 	NFSUNLOCKV4ROOTMUTEX();
1018 }
1019 
1020 /*
1021  * Server timer routine. It can scan any linked list, so long
1022  * as it holds the spin/mutex lock and there is no exclusive lock on
1023  * nfsv4rootfs_lock.
1024  * (For OpenBSD, a kthread is ok. For FreeBSD, I think it is ok
1025  *  to do this from a callout, since the spin locks work. For
1026  *  Darwin, I'm not sure what will work correctly yet.)
1027  * Should be called once per second.
1028  */
1029 APPLESTATIC void
1030 nfsrv_servertimer(void)
1031 {
1032 	struct nfsclient *clp, *nclp;
1033 	struct nfsstate *stp, *nstp;
1034 	int got_ref, i;
1035 
1036 	/*
1037 	 * Make sure nfsboottime is set. This is used by V3 as well
1038 	 * as V4. Note that nfsboottime is not nfsrvboottime, which is
1039 	 * only used by the V4 server for leases.
1040 	 */
1041 	if (nfsboottime.tv_sec == 0)
1042 		NFSSETBOOTTIME(nfsboottime);
1043 
1044 	/*
1045 	 * If server hasn't started yet, just return.
1046 	 */
1047 	NFSLOCKSTATE();
1048 	if (nfsrv_stablefirst.nsf_eograce == 0) {
1049 		NFSUNLOCKSTATE();
1050 		return;
1051 	}
1052 	if (!(nfsrv_stablefirst.nsf_flags & NFSNSF_UPDATEDONE)) {
1053 		if (!(nfsrv_stablefirst.nsf_flags & NFSNSF_GRACEOVER) &&
1054 		    NFSD_MONOSEC > nfsrv_stablefirst.nsf_eograce)
1055 			nfsrv_stablefirst.nsf_flags |=
1056 			    (NFSNSF_GRACEOVER | NFSNSF_NEEDLOCK);
1057 		NFSUNLOCKSTATE();
1058 		return;
1059 	}
1060 
1061 	/*
1062 	 * Try and get a reference count on the nfsv4rootfs_lock so that
1063 	 * no nfsd thread can acquire an exclusive lock on it before this
1064 	 * call is done. If it is already exclusively locked, just return.
1065 	 */
1066 	NFSLOCKV4ROOTMUTEX();
1067 	got_ref = nfsv4_getref_nonblock(&nfsv4rootfs_lock);
1068 	NFSUNLOCKV4ROOTMUTEX();
1069 	if (got_ref == 0) {
1070 		NFSUNLOCKSTATE();
1071 		return;
1072 	}
1073 
1074 	/*
1075 	 * For each client...
1076 	 */
1077 	for (i = 0; i < NFSCLIENTHASHSIZE; i++) {
1078 	    clp = LIST_FIRST(&nfsclienthash[i]);
1079 	    while (clp != LIST_END(&nfsclienthash[i])) {
1080 		nclp = LIST_NEXT(clp, lc_hash);
1081 		if (!(clp->lc_flags & LCL_EXPIREIT)) {
1082 		    if (((clp->lc_expiry + NFSRV_STALELEASE) < NFSD_MONOSEC
1083 			 && ((LIST_EMPTY(&clp->lc_deleg)
1084 			      && LIST_EMPTY(&clp->lc_open)) ||
1085 			     nfsrv_clients > nfsrv_clienthighwater)) ||
1086 			(clp->lc_expiry + NFSRV_MOULDYLEASE) < NFSD_MONOSEC ||
1087 			(clp->lc_expiry < NFSD_MONOSEC &&
1088 			 (nfsrv_openpluslock * 10 / 9) > NFSRV_V4STATELIMIT)) {
1089 			/*
1090 			 * Lease has expired several nfsrv_lease times ago:
1091 			 * PLUS
1092 			 *    - no state is associated with it
1093 			 *    OR
1094 			 *    - above high water mark for number of clients
1095 			 *      (nfsrv_clienthighwater should be large enough
1096 			 *       that this only occurs when clients fail to
1097 			 *       use the same nfs_client_id4.id. Maybe somewhat
1098 			 *       higher that the maximum number of clients that
1099 			 *       will mount this server?)
1100 			 * OR
1101 			 * Lease has expired a very long time ago
1102 			 * OR
1103 			 * Lease has expired PLUS the number of opens + locks
1104 			 * has exceeded 90% of capacity
1105 			 *
1106 			 * --> Mark for expiry. The actual expiry will be done
1107 			 *     by an nfsd sometime soon.
1108 			 */
1109 			clp->lc_flags |= LCL_EXPIREIT;
1110 			nfsrv_stablefirst.nsf_flags |=
1111 			    (NFSNSF_NEEDLOCK | NFSNSF_EXPIREDCLIENT);
1112 		    } else {
1113 			/*
1114 			 * If there are no opens, increment no open tick cnt
1115 			 * If time exceeds NFSNOOPEN, mark it to be thrown away
1116 			 * otherwise, if there is an open, reset no open time
1117 			 * Hopefully, this will avoid excessive re-creation
1118 			 * of open owners and subsequent open confirms.
1119 			 */
1120 			stp = LIST_FIRST(&clp->lc_open);
1121 			while (stp != LIST_END(&clp->lc_open)) {
1122 				nstp = LIST_NEXT(stp, ls_list);
1123 				if (LIST_EMPTY(&stp->ls_open)) {
1124 					stp->ls_noopens++;
1125 					if (stp->ls_noopens > NFSNOOPEN ||
1126 					    (nfsrv_openpluslock * 2) >
1127 					    NFSRV_V4STATELIMIT)
1128 						nfsrv_stablefirst.nsf_flags |=
1129 							NFSNSF_NOOPENS;
1130 				} else {
1131 					stp->ls_noopens = 0;
1132 				}
1133 				stp = nstp;
1134 			}
1135 		    }
1136 		}
1137 		clp = nclp;
1138 	    }
1139 	}
1140 	NFSUNLOCKSTATE();
1141 	NFSLOCKV4ROOTMUTEX();
1142 	nfsv4_relref(&nfsv4rootfs_lock);
1143 	NFSUNLOCKV4ROOTMUTEX();
1144 }
1145 
1146 /*
1147  * The following set of functions free up the various data structures.
1148  */
1149 /*
1150  * Clear out all open/lock state related to this nfsclient.
1151  * Caller must hold an exclusive lock on nfsv4rootfs_lock, so that
1152  * there are no other active nfsd threads.
1153  */
1154 APPLESTATIC void
1155 nfsrv_cleanclient(struct nfsclient *clp, NFSPROC_T *p)
1156 {
1157 	struct nfsstate *stp, *nstp;
1158 	struct nfsdsession *sep, *nsep;
1159 
1160 	LIST_FOREACH_SAFE(stp, &clp->lc_open, ls_list, nstp)
1161 		nfsrv_freeopenowner(stp, 1, p);
1162 	if ((clp->lc_flags & LCL_ADMINREVOKED) == 0)
1163 		LIST_FOREACH_SAFE(sep, &clp->lc_session, sess_list, nsep)
1164 			(void)nfsrv_freesession(sep, NULL);
1165 }
1166 
1167 /*
1168  * Free a client that has been cleaned. It should also already have been
1169  * removed from the lists.
1170  * (Just to be safe w.r.t. newnfs_disconnect(), call this function when
1171  *  softclock interrupts are enabled.)
1172  */
1173 APPLESTATIC void
1174 nfsrv_zapclient(struct nfsclient *clp, NFSPROC_T *p)
1175 {
1176 
1177 #ifdef notyet
1178 	if ((clp->lc_flags & (LCL_GSS | LCL_CALLBACKSON)) ==
1179 	     (LCL_GSS | LCL_CALLBACKSON) &&
1180 	    (clp->lc_hand.nfsh_flag & NFSG_COMPLETE) &&
1181 	    clp->lc_handlelen > 0) {
1182 		clp->lc_hand.nfsh_flag &= ~NFSG_COMPLETE;
1183 		clp->lc_hand.nfsh_flag |= NFSG_DESTROYED;
1184 		(void) nfsrv_docallback(clp, NFSV4PROC_CBNULL,
1185 			NULL, 0, NULL, NULL, NULL, p);
1186 	}
1187 #endif
1188 	newnfs_disconnect(&clp->lc_req);
1189 	NFSSOCKADDRFREE(clp->lc_req.nr_nam);
1190 	NFSFREEMUTEX(&clp->lc_req.nr_mtx);
1191 	free((caddr_t)clp, M_NFSDCLIENT);
1192 	NFSLOCKSTATE();
1193 	newnfsstats.srvclients--;
1194 	nfsrv_openpluslock--;
1195 	nfsrv_clients--;
1196 	NFSUNLOCKSTATE();
1197 }
1198 
1199 /*
1200  * Free a list of delegation state structures.
1201  * (This function will also free all nfslockfile structures that no
1202  *  longer have associated state.)
1203  */
1204 APPLESTATIC void
1205 nfsrv_freedeleglist(struct nfsstatehead *sthp)
1206 {
1207 	struct nfsstate *stp, *nstp;
1208 
1209 	LIST_FOREACH_SAFE(stp, sthp, ls_list, nstp) {
1210 		nfsrv_freedeleg(stp);
1211 	}
1212 	LIST_INIT(sthp);
1213 }
1214 
1215 /*
1216  * Free up a delegation.
1217  */
1218 static void
1219 nfsrv_freedeleg(struct nfsstate *stp)
1220 {
1221 	struct nfslockfile *lfp;
1222 
1223 	LIST_REMOVE(stp, ls_hash);
1224 	LIST_REMOVE(stp, ls_list);
1225 	LIST_REMOVE(stp, ls_file);
1226 	lfp = stp->ls_lfp;
1227 	if (LIST_EMPTY(&lfp->lf_open) &&
1228 	    LIST_EMPTY(&lfp->lf_lock) && LIST_EMPTY(&lfp->lf_deleg) &&
1229 	    LIST_EMPTY(&lfp->lf_locallock) && LIST_EMPTY(&lfp->lf_rollback) &&
1230 	    lfp->lf_usecount == 0 &&
1231 	    nfsv4_testlock(&lfp->lf_locallock_lck) == 0)
1232 		nfsrv_freenfslockfile(lfp);
1233 	FREE((caddr_t)stp, M_NFSDSTATE);
1234 	newnfsstats.srvdelegates--;
1235 	nfsrv_openpluslock--;
1236 	nfsrv_delegatecnt--;
1237 }
1238 
1239 /*
1240  * This function frees an open owner and all associated opens.
1241  */
1242 static void
1243 nfsrv_freeopenowner(struct nfsstate *stp, int cansleep, NFSPROC_T *p)
1244 {
1245 	struct nfsstate *nstp, *tstp;
1246 
1247 	LIST_REMOVE(stp, ls_list);
1248 	/*
1249 	 * Now, free all associated opens.
1250 	 */
1251 	nstp = LIST_FIRST(&stp->ls_open);
1252 	while (nstp != LIST_END(&stp->ls_open)) {
1253 		tstp = nstp;
1254 		nstp = LIST_NEXT(nstp, ls_list);
1255 		(void) nfsrv_freeopen(tstp, NULL, cansleep, p);
1256 	}
1257 	if (stp->ls_op)
1258 		nfsrvd_derefcache(stp->ls_op);
1259 	FREE((caddr_t)stp, M_NFSDSTATE);
1260 	newnfsstats.srvopenowners--;
1261 	nfsrv_openpluslock--;
1262 }
1263 
1264 /*
1265  * This function frees an open (nfsstate open structure) with all associated
1266  * lock_owners and locks. It also frees the nfslockfile structure iff there
1267  * are no other opens on the file.
1268  * Returns 1 if it free'd the nfslockfile, 0 otherwise.
1269  */
1270 static int
1271 nfsrv_freeopen(struct nfsstate *stp, vnode_t vp, int cansleep, NFSPROC_T *p)
1272 {
1273 	struct nfsstate *nstp, *tstp;
1274 	struct nfslockfile *lfp;
1275 	int ret;
1276 
1277 	LIST_REMOVE(stp, ls_hash);
1278 	LIST_REMOVE(stp, ls_list);
1279 	LIST_REMOVE(stp, ls_file);
1280 
1281 	lfp = stp->ls_lfp;
1282 	/*
1283 	 * Now, free all lockowners associated with this open.
1284 	 */
1285 	LIST_FOREACH_SAFE(tstp, &stp->ls_open, ls_list, nstp)
1286 		nfsrv_freelockowner(tstp, vp, cansleep, p);
1287 
1288 	/*
1289 	 * The nfslockfile is freed here if there are no locks
1290 	 * associated with the open.
1291 	 * If there are locks associated with the open, the
1292 	 * nfslockfile structure can be freed via nfsrv_freelockowner().
1293 	 * Acquire the state mutex to avoid races with calls to
1294 	 * nfsrv_getlockfile().
1295 	 */
1296 	if (cansleep != 0)
1297 		NFSLOCKSTATE();
1298 	if (lfp != NULL && LIST_EMPTY(&lfp->lf_open) &&
1299 	    LIST_EMPTY(&lfp->lf_deleg) && LIST_EMPTY(&lfp->lf_lock) &&
1300 	    LIST_EMPTY(&lfp->lf_locallock) && LIST_EMPTY(&lfp->lf_rollback) &&
1301 	    lfp->lf_usecount == 0 &&
1302 	    (cansleep != 0 || nfsv4_testlock(&lfp->lf_locallock_lck) == 0)) {
1303 		nfsrv_freenfslockfile(lfp);
1304 		ret = 1;
1305 	} else
1306 		ret = 0;
1307 	if (cansleep != 0)
1308 		NFSUNLOCKSTATE();
1309 	FREE((caddr_t)stp, M_NFSDSTATE);
1310 	newnfsstats.srvopens--;
1311 	nfsrv_openpluslock--;
1312 	return (ret);
1313 }
1314 
1315 /*
1316  * Frees a lockowner and all associated locks.
1317  */
1318 static void
1319 nfsrv_freelockowner(struct nfsstate *stp, vnode_t vp, int cansleep,
1320     NFSPROC_T *p)
1321 {
1322 
1323 	LIST_REMOVE(stp, ls_hash);
1324 	LIST_REMOVE(stp, ls_list);
1325 	nfsrv_freeallnfslocks(stp, vp, cansleep, p);
1326 	if (stp->ls_op)
1327 		nfsrvd_derefcache(stp->ls_op);
1328 	FREE((caddr_t)stp, M_NFSDSTATE);
1329 	newnfsstats.srvlockowners--;
1330 	nfsrv_openpluslock--;
1331 }
1332 
1333 /*
1334  * Free all the nfs locks on a lockowner.
1335  */
1336 static void
1337 nfsrv_freeallnfslocks(struct nfsstate *stp, vnode_t vp, int cansleep,
1338     NFSPROC_T *p)
1339 {
1340 	struct nfslock *lop, *nlop;
1341 	struct nfsrollback *rlp, *nrlp;
1342 	struct nfslockfile *lfp = NULL;
1343 	int gottvp = 0;
1344 	vnode_t tvp = NULL;
1345 	uint64_t first, end;
1346 
1347 	if (vp != NULL)
1348 		ASSERT_VOP_UNLOCKED(vp, "nfsrv_freeallnfslocks: vnode locked");
1349 	lop = LIST_FIRST(&stp->ls_lock);
1350 	while (lop != LIST_END(&stp->ls_lock)) {
1351 		nlop = LIST_NEXT(lop, lo_lckowner);
1352 		/*
1353 		 * Since all locks should be for the same file, lfp should
1354 		 * not change.
1355 		 */
1356 		if (lfp == NULL)
1357 			lfp = lop->lo_lfp;
1358 		else if (lfp != lop->lo_lfp)
1359 			panic("allnfslocks");
1360 		/*
1361 		 * If vp is NULL and cansleep != 0, a vnode must be acquired
1362 		 * from the file handle. This only occurs when called from
1363 		 * nfsrv_cleanclient().
1364 		 */
1365 		if (gottvp == 0) {
1366 			if (nfsrv_dolocallocks == 0)
1367 				tvp = NULL;
1368 			else if (vp == NULL && cansleep != 0) {
1369 				tvp = nfsvno_getvp(&lfp->lf_fh);
1370 				NFSVOPUNLOCK(tvp, 0);
1371 			} else
1372 				tvp = vp;
1373 			gottvp = 1;
1374 		}
1375 
1376 		if (tvp != NULL) {
1377 			if (cansleep == 0)
1378 				panic("allnfs2");
1379 			first = lop->lo_first;
1380 			end = lop->lo_end;
1381 			nfsrv_freenfslock(lop);
1382 			nfsrv_localunlock(tvp, lfp, first, end, p);
1383 			LIST_FOREACH_SAFE(rlp, &lfp->lf_rollback, rlck_list,
1384 			    nrlp)
1385 				free(rlp, M_NFSDROLLBACK);
1386 			LIST_INIT(&lfp->lf_rollback);
1387 		} else
1388 			nfsrv_freenfslock(lop);
1389 		lop = nlop;
1390 	}
1391 	if (vp == NULL && tvp != NULL)
1392 		vrele(tvp);
1393 }
1394 
1395 /*
1396  * Free an nfslock structure.
1397  */
1398 static void
1399 nfsrv_freenfslock(struct nfslock *lop)
1400 {
1401 
1402 	if (lop->lo_lckfile.le_prev != NULL) {
1403 		LIST_REMOVE(lop, lo_lckfile);
1404 		newnfsstats.srvlocks--;
1405 		nfsrv_openpluslock--;
1406 	}
1407 	LIST_REMOVE(lop, lo_lckowner);
1408 	FREE((caddr_t)lop, M_NFSDLOCK);
1409 }
1410 
1411 /*
1412  * This function frees an nfslockfile structure.
1413  */
1414 static void
1415 nfsrv_freenfslockfile(struct nfslockfile *lfp)
1416 {
1417 
1418 	LIST_REMOVE(lfp, lf_hash);
1419 	FREE((caddr_t)lfp, M_NFSDLOCKFILE);
1420 }
1421 
1422 /*
1423  * This function looks up an nfsstate structure via stateid.
1424  */
1425 static int
1426 nfsrv_getstate(struct nfsclient *clp, nfsv4stateid_t *stateidp, __unused u_int32_t flags,
1427     struct nfsstate **stpp)
1428 {
1429 	struct nfsstate *stp;
1430 	struct nfsstatehead *hp;
1431 	int error = 0;
1432 
1433 	*stpp = NULL;
1434 	hp = NFSSTATEHASH(clp, *stateidp);
1435 	LIST_FOREACH(stp, hp, ls_hash) {
1436 		if (!NFSBCMP(stp->ls_stateid.other, stateidp->other,
1437 			NFSX_STATEIDOTHER))
1438 			break;
1439 	}
1440 
1441 	/*
1442 	 * If no state id in list, return NFSERR_BADSTATEID.
1443 	 */
1444 	if (stp == LIST_END(hp)) {
1445 		error = NFSERR_BADSTATEID;
1446 		goto out;
1447 	}
1448 	*stpp = stp;
1449 
1450 out:
1451 	NFSEXITCODE(error);
1452 	return (error);
1453 }
1454 
1455 /*
1456  * This function gets an nfsstate structure via owner string.
1457  */
1458 static void
1459 nfsrv_getowner(struct nfsstatehead *hp, struct nfsstate *new_stp,
1460     struct nfsstate **stpp)
1461 {
1462 	struct nfsstate *stp;
1463 
1464 	*stpp = NULL;
1465 	LIST_FOREACH(stp, hp, ls_list) {
1466 		if (new_stp->ls_ownerlen == stp->ls_ownerlen &&
1467 		  !NFSBCMP(new_stp->ls_owner,stp->ls_owner,stp->ls_ownerlen)) {
1468 			*stpp = stp;
1469 			return;
1470 		}
1471 	}
1472 }
1473 
1474 /*
1475  * Lock control function called to update lock status.
1476  * Returns 0 upon success, -1 if there is no lock and the flags indicate
1477  * that one isn't to be created and an NFSERR_xxx for other errors.
1478  * The structures new_stp and new_lop are passed in as pointers that should
1479  * be set to NULL if the structure is used and shouldn't be free'd.
1480  * For the NFSLCK_TEST and NFSLCK_CHECK cases, the structures are
1481  * never used and can safely be allocated on the stack. For all other
1482  * cases, *new_stpp and *new_lopp should be malloc'd before the call,
1483  * in case they are used.
1484  */
1485 APPLESTATIC int
1486 nfsrv_lockctrl(vnode_t vp, struct nfsstate **new_stpp,
1487     struct nfslock **new_lopp, struct nfslockconflict *cfp,
1488     nfsquad_t clientid, nfsv4stateid_t *stateidp,
1489     __unused struct nfsexstuff *exp,
1490     struct nfsrv_descript *nd, NFSPROC_T *p)
1491 {
1492 	struct nfslock *lop;
1493 	struct nfsstate *new_stp = *new_stpp;
1494 	struct nfslock *new_lop = *new_lopp;
1495 	struct nfsstate *tstp, *mystp, *nstp;
1496 	int specialid = 0;
1497 	struct nfslockfile *lfp;
1498 	struct nfslock *other_lop = NULL;
1499 	struct nfsstate *stp, *lckstp = NULL;
1500 	struct nfsclient *clp = NULL;
1501 	u_int32_t bits;
1502 	int error = 0, haslock = 0, ret, reterr;
1503 	int getlckret, delegation = 0, filestruct_locked, vnode_unlocked = 0;
1504 	fhandle_t nfh;
1505 	uint64_t first, end;
1506 	uint32_t lock_flags;
1507 
1508 	if (new_stp->ls_flags & (NFSLCK_CHECK | NFSLCK_SETATTR)) {
1509 		/*
1510 		 * Note the special cases of "all 1s" or "all 0s" stateids and
1511 		 * let reads with all 1s go ahead.
1512 		 */
1513 		if (new_stp->ls_stateid.seqid == 0x0 &&
1514 		    new_stp->ls_stateid.other[0] == 0x0 &&
1515 		    new_stp->ls_stateid.other[1] == 0x0 &&
1516 		    new_stp->ls_stateid.other[2] == 0x0)
1517 			specialid = 1;
1518 		else if (new_stp->ls_stateid.seqid == 0xffffffff &&
1519 		    new_stp->ls_stateid.other[0] == 0xffffffff &&
1520 		    new_stp->ls_stateid.other[1] == 0xffffffff &&
1521 		    new_stp->ls_stateid.other[2] == 0xffffffff)
1522 			specialid = 2;
1523 	}
1524 
1525 	/*
1526 	 * Check for restart conditions (client and server).
1527 	 */
1528 	error = nfsrv_checkrestart(clientid, new_stp->ls_flags,
1529 	    &new_stp->ls_stateid, specialid);
1530 	if (error)
1531 		goto out;
1532 
1533 	/*
1534 	 * Check for state resource limit exceeded.
1535 	 */
1536 	if ((new_stp->ls_flags & NFSLCK_LOCK) &&
1537 	    nfsrv_openpluslock > NFSRV_V4STATELIMIT) {
1538 		error = NFSERR_RESOURCE;
1539 		goto out;
1540 	}
1541 
1542 	/*
1543 	 * For the lock case, get another nfslock structure,
1544 	 * just in case we need it.
1545 	 * Malloc now, before we start sifting through the linked lists,
1546 	 * in case we have to wait for memory.
1547 	 */
1548 tryagain:
1549 	if (new_stp->ls_flags & NFSLCK_LOCK)
1550 		MALLOC(other_lop, struct nfslock *, sizeof (struct nfslock),
1551 		    M_NFSDLOCK, M_WAITOK);
1552 	filestruct_locked = 0;
1553 	reterr = 0;
1554 	lfp = NULL;
1555 
1556 	/*
1557 	 * Get the lockfile structure for CFH now, so we can do a sanity
1558 	 * check against the stateid, before incrementing the seqid#, since
1559 	 * we want to return NFSERR_BADSTATEID on failure and the seqid#
1560 	 * shouldn't be incremented for this case.
1561 	 * If nfsrv_getlockfile() returns -1, it means "not found", which
1562 	 * will be handled later.
1563 	 * If we are doing Lock/LockU and local locking is enabled, sleep
1564 	 * lock the nfslockfile structure.
1565 	 */
1566 	getlckret = nfsrv_getlockfh(vp, new_stp->ls_flags, NULL, &nfh, p);
1567 	NFSLOCKSTATE();
1568 	if (getlckret == 0) {
1569 		if ((new_stp->ls_flags & (NFSLCK_LOCK | NFSLCK_UNLOCK)) != 0 &&
1570 		    nfsrv_dolocallocks != 0 && nd->nd_repstat == 0) {
1571 			getlckret = nfsrv_getlockfile(new_stp->ls_flags, NULL,
1572 			    &lfp, &nfh, 1);
1573 			if (getlckret == 0)
1574 				filestruct_locked = 1;
1575 		} else
1576 			getlckret = nfsrv_getlockfile(new_stp->ls_flags, NULL,
1577 			    &lfp, &nfh, 0);
1578 	}
1579 	if (getlckret != 0 && getlckret != -1)
1580 		reterr = getlckret;
1581 
1582 	if (filestruct_locked != 0) {
1583 		LIST_INIT(&lfp->lf_rollback);
1584 		if ((new_stp->ls_flags & NFSLCK_LOCK)) {
1585 			/*
1586 			 * For local locking, do the advisory locking now, so
1587 			 * that any conflict can be detected. A failure later
1588 			 * can be rolled back locally. If an error is returned,
1589 			 * struct nfslockfile has been unlocked and any local
1590 			 * locking rolled back.
1591 			 */
1592 			NFSUNLOCKSTATE();
1593 			if (vnode_unlocked == 0) {
1594 				ASSERT_VOP_ELOCKED(vp, "nfsrv_lockctrl1");
1595 				vnode_unlocked = 1;
1596 				NFSVOPUNLOCK(vp, 0);
1597 			}
1598 			reterr = nfsrv_locallock(vp, lfp,
1599 			    (new_lop->lo_flags & (NFSLCK_READ | NFSLCK_WRITE)),
1600 			    new_lop->lo_first, new_lop->lo_end, cfp, p);
1601 			NFSLOCKSTATE();
1602 		}
1603 	}
1604 
1605 	if (specialid == 0) {
1606 	    if (new_stp->ls_flags & NFSLCK_TEST) {
1607 		/*
1608 		 * RFC 3530 does not list LockT as an op that renews a
1609 		 * lease, but the concensus seems to be that it is ok
1610 		 * for a server to do so.
1611 		 */
1612 		error = nfsrv_getclient(clientid, CLOPS_RENEW, &clp, NULL,
1613 		    (nfsquad_t)((u_quad_t)0), 0, nd, p);
1614 
1615 		/*
1616 		 * Since NFSERR_EXPIRED, NFSERR_ADMINREVOKED are not valid
1617 		 * error returns for LockT, just go ahead and test for a lock,
1618 		 * since there are no locks for this client, but other locks
1619 		 * can conflict. (ie. same client will always be false)
1620 		 */
1621 		if (error == NFSERR_EXPIRED || error == NFSERR_ADMINREVOKED)
1622 		    error = 0;
1623 		lckstp = new_stp;
1624 	    } else {
1625 	      error = nfsrv_getclient(clientid, CLOPS_RENEW, &clp, NULL,
1626 		(nfsquad_t)((u_quad_t)0), 0, nd, p);
1627 	      if (error == 0)
1628 		/*
1629 		 * Look up the stateid
1630 		 */
1631 		error = nfsrv_getstate(clp, &new_stp->ls_stateid,
1632 		  new_stp->ls_flags, &stp);
1633 	      /*
1634 	       * do some sanity checks for an unconfirmed open or a
1635 	       * stateid that refers to the wrong file, for an open stateid
1636 	       */
1637 	      if (error == 0 && (stp->ls_flags & NFSLCK_OPEN) &&
1638 		  ((stp->ls_openowner->ls_flags & NFSLCK_NEEDSCONFIRM) ||
1639 		   (getlckret == 0 && stp->ls_lfp != lfp))){
1640 		      /*
1641 		       * NFSLCK_SETATTR should return OK rather than NFSERR_BADSTATEID
1642 		       * The only exception is using SETATTR with SIZE.
1643 		       * */
1644                     if ((new_stp->ls_flags &
1645                          (NFSLCK_SETATTR | NFSLCK_CHECK)) != NFSLCK_SETATTR)
1646 			     error = NFSERR_BADSTATEID;
1647 	      }
1648 
1649 		if (error == 0 &&
1650 		  (stp->ls_flags & (NFSLCK_DELEGREAD | NFSLCK_DELEGWRITE)) &&
1651 		  getlckret == 0 && stp->ls_lfp != lfp)
1652 			error = NFSERR_BADSTATEID;
1653 
1654 	      /*
1655 	       * If the lockowner stateid doesn't refer to the same file,
1656 	       * I believe that is considered ok, since some clients will
1657 	       * only create a single lockowner and use that for all locks
1658 	       * on all files.
1659 	       * For now, log it as a diagnostic, instead of considering it
1660 	       * a BadStateid.
1661 	       */
1662 	      if (error == 0 && (stp->ls_flags &
1663 		  (NFSLCK_OPEN | NFSLCK_DELEGREAD | NFSLCK_DELEGWRITE)) == 0 &&
1664 		  getlckret == 0 && stp->ls_lfp != lfp) {
1665 #ifdef DIAGNOSTIC
1666 		  printf("Got a lock statid for different file open\n");
1667 #endif
1668 		  /*
1669 		  error = NFSERR_BADSTATEID;
1670 		  */
1671 	      }
1672 
1673 	      if (error == 0) {
1674 		    if (new_stp->ls_flags & NFSLCK_OPENTOLOCK) {
1675 			/*
1676 			 * If haslock set, we've already checked the seqid.
1677 			 */
1678 			if (!haslock) {
1679 			    if (stp->ls_flags & NFSLCK_OPEN)
1680 				error = nfsrv_checkseqid(nd, new_stp->ls_seq,
1681 				    stp->ls_openowner, new_stp->ls_op);
1682 			    else
1683 				error = NFSERR_BADSTATEID;
1684 			}
1685 			if (!error)
1686 			    nfsrv_getowner(&stp->ls_open, new_stp, &lckstp);
1687 			if (lckstp)
1688 			    /*
1689 			     * I believe this should be an error, but it
1690 			     * isn't obvious what NFSERR_xxx would be
1691 			     * appropriate, so I'll use NFSERR_INVAL for now.
1692 			     */
1693 			    error = NFSERR_INVAL;
1694 			else
1695 			    lckstp = new_stp;
1696 		    } else if (new_stp->ls_flags&(NFSLCK_LOCK|NFSLCK_UNLOCK)) {
1697 			/*
1698 			 * If haslock set, ditto above.
1699 			 */
1700 			if (!haslock) {
1701 			    if (stp->ls_flags & NFSLCK_OPEN)
1702 				error = NFSERR_BADSTATEID;
1703 			    else
1704 				error = nfsrv_checkseqid(nd, new_stp->ls_seq,
1705 				    stp, new_stp->ls_op);
1706 			}
1707 			lckstp = stp;
1708 		    } else {
1709 			lckstp = stp;
1710 		    }
1711 	      }
1712 	      /*
1713 	       * If the seqid part of the stateid isn't the same, return
1714 	       * NFSERR_OLDSTATEID for cases other than I/O Ops.
1715 	       * For I/O Ops, only return NFSERR_OLDSTATEID if
1716 	       * nfsrv_returnoldstateid is set. (The concensus on the email
1717 	       * list was that most clients would prefer to not receive
1718 	       * NFSERR_OLDSTATEID for I/O Ops, but the RFC suggests that that
1719 	       * is what will happen, so I use the nfsrv_returnoldstateid to
1720 	       * allow for either server configuration.)
1721 	       */
1722 	      if (!error && stp->ls_stateid.seqid!=new_stp->ls_stateid.seqid &&
1723 		  (((nd->nd_flag & ND_NFSV41) == 0 &&
1724 		   (!(new_stp->ls_flags & NFSLCK_CHECK) ||
1725 		    nfsrv_returnoldstateid)) ||
1726 		   ((nd->nd_flag & ND_NFSV41) != 0 &&
1727 		    new_stp->ls_stateid.seqid != 0)))
1728 		    error = NFSERR_OLDSTATEID;
1729 	    }
1730 	}
1731 
1732 	/*
1733 	 * Now we can check for grace.
1734 	 */
1735 	if (!error)
1736 		error = nfsrv_checkgrace(nd, clp, new_stp->ls_flags);
1737 	if ((new_stp->ls_flags & NFSLCK_RECLAIM) && !error &&
1738 		nfsrv_checkstable(clp))
1739 		error = NFSERR_NOGRACE;
1740 	/*
1741 	 * If we successfully Reclaimed state, note that.
1742 	 */
1743 	if ((new_stp->ls_flags & NFSLCK_RECLAIM) && !error)
1744 		nfsrv_markstable(clp);
1745 
1746 	/*
1747 	 * At this point, either error == NFSERR_BADSTATEID or the
1748 	 * seqid# has been updated, so we can return any error.
1749 	 * If error == 0, there may be an error in:
1750 	 *    nd_repstat - Set by the calling function.
1751 	 *    reterr - Set above, if getting the nfslockfile structure
1752 	 *       or acquiring the local lock failed.
1753 	 *    (If both of these are set, nd_repstat should probably be
1754 	 *     returned, since that error was detected before this
1755 	 *     function call.)
1756 	 */
1757 	if (error != 0 || nd->nd_repstat != 0 || reterr != 0) {
1758 		if (error == 0) {
1759 			if (nd->nd_repstat != 0)
1760 				error = nd->nd_repstat;
1761 			else
1762 				error = reterr;
1763 		}
1764 		if (filestruct_locked != 0) {
1765 			/* Roll back local locks. */
1766 			NFSUNLOCKSTATE();
1767 			if (vnode_unlocked == 0) {
1768 				ASSERT_VOP_ELOCKED(vp, "nfsrv_lockctrl2");
1769 				vnode_unlocked = 1;
1770 				NFSVOPUNLOCK(vp, 0);
1771 			}
1772 			nfsrv_locallock_rollback(vp, lfp, p);
1773 			NFSLOCKSTATE();
1774 			nfsrv_unlocklf(lfp);
1775 		}
1776 		NFSUNLOCKSTATE();
1777 		goto out;
1778 	}
1779 
1780 	/*
1781 	 * Check the nfsrv_getlockfile return.
1782 	 * Returned -1 if no structure found.
1783 	 */
1784 	if (getlckret == -1) {
1785 		error = NFSERR_EXPIRED;
1786 		/*
1787 		 * Called from lockt, so no lock is OK.
1788 		 */
1789 		if (new_stp->ls_flags & NFSLCK_TEST) {
1790 			error = 0;
1791 		} else if (new_stp->ls_flags &
1792 		    (NFSLCK_CHECK | NFSLCK_SETATTR)) {
1793 			/*
1794 			 * Called to check for a lock, OK if the stateid is all
1795 			 * 1s or all 0s, but there should be an nfsstate
1796 			 * otherwise.
1797 			 * (ie. If there is no open, I'll assume no share
1798 			 *  deny bits.)
1799 			 */
1800 			if (specialid)
1801 				error = 0;
1802 			else
1803 				error = NFSERR_BADSTATEID;
1804 		}
1805 		NFSUNLOCKSTATE();
1806 		goto out;
1807 	}
1808 
1809 	/*
1810 	 * For NFSLCK_CHECK and NFSLCK_LOCK, test for a share conflict.
1811 	 * For NFSLCK_CHECK, allow a read if write access is granted,
1812 	 * but check for a deny. For NFSLCK_LOCK, require correct access,
1813 	 * which implies a conflicting deny can't exist.
1814 	 */
1815 	if (new_stp->ls_flags & (NFSLCK_CHECK | NFSLCK_LOCK)) {
1816 	    /*
1817 	     * Four kinds of state id:
1818 	     * - specialid (all 0s or all 1s), only for NFSLCK_CHECK
1819 	     * - stateid for an open
1820 	     * - stateid for a delegation
1821 	     * - stateid for a lock owner
1822 	     */
1823 	    if (!specialid) {
1824 		if (stp->ls_flags & (NFSLCK_DELEGREAD | NFSLCK_DELEGWRITE)) {
1825 		    delegation = 1;
1826 		    mystp = stp;
1827 		    nfsrv_delaydelegtimeout(stp);
1828 	        } else if (stp->ls_flags & NFSLCK_OPEN) {
1829 		    mystp = stp;
1830 		} else {
1831 		    mystp = stp->ls_openstp;
1832 		}
1833 		/*
1834 		 * If locking or checking, require correct access
1835 		 * bit set.
1836 		 */
1837 		if (((new_stp->ls_flags & NFSLCK_LOCK) &&
1838 		     !((new_lop->lo_flags >> NFSLCK_LOCKSHIFT) &
1839 		       mystp->ls_flags & NFSLCK_ACCESSBITS)) ||
1840 		    ((new_stp->ls_flags & (NFSLCK_CHECK|NFSLCK_READACCESS)) ==
1841 		      (NFSLCK_CHECK | NFSLCK_READACCESS) &&
1842 		     !(mystp->ls_flags & NFSLCK_READACCESS)) ||
1843 		    ((new_stp->ls_flags & (NFSLCK_CHECK|NFSLCK_WRITEACCESS)) ==
1844 		      (NFSLCK_CHECK | NFSLCK_WRITEACCESS) &&
1845 		     !(mystp->ls_flags & NFSLCK_WRITEACCESS))) {
1846 			if (filestruct_locked != 0) {
1847 				/* Roll back local locks. */
1848 				NFSUNLOCKSTATE();
1849 				if (vnode_unlocked == 0) {
1850 					ASSERT_VOP_ELOCKED(vp,
1851 					    "nfsrv_lockctrl3");
1852 					vnode_unlocked = 1;
1853 					NFSVOPUNLOCK(vp, 0);
1854 				}
1855 				nfsrv_locallock_rollback(vp, lfp, p);
1856 				NFSLOCKSTATE();
1857 				nfsrv_unlocklf(lfp);
1858 			}
1859 			NFSUNLOCKSTATE();
1860 			error = NFSERR_OPENMODE;
1861 			goto out;
1862 		}
1863 	    } else
1864 		mystp = NULL;
1865 	    if ((new_stp->ls_flags & NFSLCK_CHECK) && !delegation) {
1866 		/*
1867 		 * Check for a conflicting deny bit.
1868 		 */
1869 		LIST_FOREACH(tstp, &lfp->lf_open, ls_file) {
1870 		    if (tstp != mystp) {
1871 			bits = tstp->ls_flags;
1872 			bits >>= NFSLCK_SHIFT;
1873 			if (new_stp->ls_flags & bits & NFSLCK_ACCESSBITS) {
1874 			    KASSERT(vnode_unlocked == 0,
1875 				("nfsrv_lockctrl: vnode unlocked1"));
1876 			    ret = nfsrv_clientconflict(tstp->ls_clp, &haslock,
1877 				vp, p);
1878 			    if (ret == 1) {
1879 				/*
1880 				* nfsrv_clientconflict unlocks state
1881 				 * when it returns non-zero.
1882 				 */
1883 				lckstp = NULL;
1884 				goto tryagain;
1885 			    }
1886 			    if (ret == 0)
1887 				NFSUNLOCKSTATE();
1888 			    if (ret == 2)
1889 				error = NFSERR_PERM;
1890 			    else
1891 				error = NFSERR_OPENMODE;
1892 			    goto out;
1893 			}
1894 		    }
1895 		}
1896 
1897 		/* We're outta here */
1898 		NFSUNLOCKSTATE();
1899 		goto out;
1900 	    }
1901 	}
1902 
1903 	/*
1904 	 * For setattr, just get rid of all the Delegations for other clients.
1905 	 */
1906 	if (new_stp->ls_flags & NFSLCK_SETATTR) {
1907 		KASSERT(vnode_unlocked == 0,
1908 		    ("nfsrv_lockctrl: vnode unlocked2"));
1909 		ret = nfsrv_cleandeleg(vp, lfp, clp, &haslock, p);
1910 		if (ret) {
1911 			/*
1912 			 * nfsrv_cleandeleg() unlocks state when it
1913 			 * returns non-zero.
1914 			 */
1915 			if (ret == -1) {
1916 				lckstp = NULL;
1917 				goto tryagain;
1918 			}
1919 			error = ret;
1920 			goto out;
1921 		}
1922 		if (!(new_stp->ls_flags & NFSLCK_CHECK) ||
1923 		    (LIST_EMPTY(&lfp->lf_open) && LIST_EMPTY(&lfp->lf_lock) &&
1924 		     LIST_EMPTY(&lfp->lf_deleg))) {
1925 			NFSUNLOCKSTATE();
1926 			goto out;
1927 		}
1928 	}
1929 
1930 	/*
1931 	 * Check for a conflicting delegation. If one is found, call
1932 	 * nfsrv_delegconflict() to handle it. If the v4root lock hasn't
1933 	 * been set yet, it will get the lock. Otherwise, it will recall
1934 	 * the delegation. Then, we try try again...
1935 	 * I currently believe the conflict algorithm to be:
1936 	 * For Lock Ops (Lock/LockT/LockU)
1937 	 * - there is a conflict iff a different client has a write delegation
1938 	 * For Reading (Read Op)
1939 	 * - there is a conflict iff a different client has a write delegation
1940 	 *   (the specialids are always a different client)
1941 	 * For Writing (Write/Setattr of size)
1942 	 * - there is a conflict if a different client has any delegation
1943 	 * - there is a conflict if the same client has a read delegation
1944 	 *   (I don't understand why this isn't allowed, but that seems to be
1945 	 *    the current concensus?)
1946 	 */
1947 	tstp = LIST_FIRST(&lfp->lf_deleg);
1948 	while (tstp != LIST_END(&lfp->lf_deleg)) {
1949 	    nstp = LIST_NEXT(tstp, ls_file);
1950 	    if ((((new_stp->ls_flags&(NFSLCK_LOCK|NFSLCK_UNLOCK|NFSLCK_TEST))||
1951 		 ((new_stp->ls_flags & NFSLCK_CHECK) &&
1952 		  (new_lop->lo_flags & NFSLCK_READ))) &&
1953 		  clp != tstp->ls_clp &&
1954 		 (tstp->ls_flags & NFSLCK_DELEGWRITE)) ||
1955 		 ((new_stp->ls_flags & NFSLCK_CHECK) &&
1956 		   (new_lop->lo_flags & NFSLCK_WRITE) &&
1957 		  (clp != tstp->ls_clp ||
1958 		   (tstp->ls_flags & NFSLCK_DELEGREAD)))) {
1959 		ret = 0;
1960 		if (filestruct_locked != 0) {
1961 			/* Roll back local locks. */
1962 			NFSUNLOCKSTATE();
1963 			if (vnode_unlocked == 0) {
1964 				ASSERT_VOP_ELOCKED(vp, "nfsrv_lockctrl4");
1965 				NFSVOPUNLOCK(vp, 0);
1966 			}
1967 			nfsrv_locallock_rollback(vp, lfp, p);
1968 			NFSLOCKSTATE();
1969 			nfsrv_unlocklf(lfp);
1970 			NFSUNLOCKSTATE();
1971 			NFSVOPLOCK(vp, LK_EXCLUSIVE | LK_RETRY);
1972 			vnode_unlocked = 0;
1973 			if ((vp->v_iflag & VI_DOOMED) != 0)
1974 				ret = NFSERR_SERVERFAULT;
1975 			NFSLOCKSTATE();
1976 		}
1977 		if (ret == 0)
1978 			ret = nfsrv_delegconflict(tstp, &haslock, p, vp);
1979 		if (ret) {
1980 		    /*
1981 		     * nfsrv_delegconflict unlocks state when it
1982 		     * returns non-zero, which it always does.
1983 		     */
1984 		    if (other_lop) {
1985 			FREE((caddr_t)other_lop, M_NFSDLOCK);
1986 			other_lop = NULL;
1987 		    }
1988 		    if (ret == -1) {
1989 			lckstp = NULL;
1990 			goto tryagain;
1991 		    }
1992 		    error = ret;
1993 		    goto out;
1994 		}
1995 		/* Never gets here. */
1996 	    }
1997 	    tstp = nstp;
1998 	}
1999 
2000 	/*
2001 	 * Handle the unlock case by calling nfsrv_updatelock().
2002 	 * (Should I have done some access checking above for unlock? For now,
2003 	 *  just let it happen.)
2004 	 */
2005 	if (new_stp->ls_flags & NFSLCK_UNLOCK) {
2006 		first = new_lop->lo_first;
2007 		end = new_lop->lo_end;
2008 		nfsrv_updatelock(stp, new_lopp, &other_lop, lfp);
2009 		stateidp->seqid = ++(stp->ls_stateid.seqid);
2010 		if ((nd->nd_flag & ND_NFSV41) != 0 && stateidp->seqid == 0)
2011 			stateidp->seqid = stp->ls_stateid.seqid = 1;
2012 		stateidp->other[0] = stp->ls_stateid.other[0];
2013 		stateidp->other[1] = stp->ls_stateid.other[1];
2014 		stateidp->other[2] = stp->ls_stateid.other[2];
2015 		if (filestruct_locked != 0) {
2016 			NFSUNLOCKSTATE();
2017 			if (vnode_unlocked == 0) {
2018 				ASSERT_VOP_ELOCKED(vp, "nfsrv_lockctrl5");
2019 				vnode_unlocked = 1;
2020 				NFSVOPUNLOCK(vp, 0);
2021 			}
2022 			/* Update the local locks. */
2023 			nfsrv_localunlock(vp, lfp, first, end, p);
2024 			NFSLOCKSTATE();
2025 			nfsrv_unlocklf(lfp);
2026 		}
2027 		NFSUNLOCKSTATE();
2028 		goto out;
2029 	}
2030 
2031 	/*
2032 	 * Search for a conflicting lock. A lock conflicts if:
2033 	 * - the lock range overlaps and
2034 	 * - at least one lock is a write lock and
2035 	 * - it is not owned by the same lock owner
2036 	 */
2037 	if (!delegation) {
2038 	  LIST_FOREACH(lop, &lfp->lf_lock, lo_lckfile) {
2039 	    if (new_lop->lo_end > lop->lo_first &&
2040 		new_lop->lo_first < lop->lo_end &&
2041 		(new_lop->lo_flags == NFSLCK_WRITE ||
2042 		 lop->lo_flags == NFSLCK_WRITE) &&
2043 		lckstp != lop->lo_stp &&
2044 		(clp != lop->lo_stp->ls_clp ||
2045 		 lckstp->ls_ownerlen != lop->lo_stp->ls_ownerlen ||
2046 		 NFSBCMP(lckstp->ls_owner, lop->lo_stp->ls_owner,
2047 		    lckstp->ls_ownerlen))) {
2048 		if (other_lop) {
2049 		    FREE((caddr_t)other_lop, M_NFSDLOCK);
2050 		    other_lop = NULL;
2051 		}
2052 		if (vnode_unlocked != 0)
2053 		    ret = nfsrv_clientconflict(lop->lo_stp->ls_clp, &haslock,
2054 			NULL, p);
2055 		else
2056 		    ret = nfsrv_clientconflict(lop->lo_stp->ls_clp, &haslock,
2057 			vp, p);
2058 		if (ret == 1) {
2059 		    if (filestruct_locked != 0) {
2060 			if (vnode_unlocked == 0) {
2061 				ASSERT_VOP_ELOCKED(vp, "nfsrv_lockctrl6");
2062 				NFSVOPUNLOCK(vp, 0);
2063 			}
2064 			/* Roll back local locks. */
2065 			nfsrv_locallock_rollback(vp, lfp, p);
2066 			NFSLOCKSTATE();
2067 			nfsrv_unlocklf(lfp);
2068 			NFSUNLOCKSTATE();
2069 			NFSVOPLOCK(vp, LK_EXCLUSIVE | LK_RETRY);
2070 			vnode_unlocked = 0;
2071 			if ((vp->v_iflag & VI_DOOMED) != 0) {
2072 				error = NFSERR_SERVERFAULT;
2073 				goto out;
2074 			}
2075 		    }
2076 		    /*
2077 		     * nfsrv_clientconflict() unlocks state when it
2078 		     * returns non-zero.
2079 		     */
2080 		    lckstp = NULL;
2081 		    goto tryagain;
2082 		}
2083 		/*
2084 		 * Found a conflicting lock, so record the conflict and
2085 		 * return the error.
2086 		 */
2087 		if (cfp != NULL && ret == 0) {
2088 		    cfp->cl_clientid.lval[0]=lop->lo_stp->ls_stateid.other[0];
2089 		    cfp->cl_clientid.lval[1]=lop->lo_stp->ls_stateid.other[1];
2090 		    cfp->cl_first = lop->lo_first;
2091 		    cfp->cl_end = lop->lo_end;
2092 		    cfp->cl_flags = lop->lo_flags;
2093 		    cfp->cl_ownerlen = lop->lo_stp->ls_ownerlen;
2094 		    NFSBCOPY(lop->lo_stp->ls_owner, cfp->cl_owner,
2095 			cfp->cl_ownerlen);
2096 		}
2097 		if (ret == 2)
2098 		    error = NFSERR_PERM;
2099 		else if (new_stp->ls_flags & NFSLCK_RECLAIM)
2100 		    error = NFSERR_RECLAIMCONFLICT;
2101 		else if (new_stp->ls_flags & NFSLCK_CHECK)
2102 		    error = NFSERR_LOCKED;
2103 		else
2104 		    error = NFSERR_DENIED;
2105 		if (filestruct_locked != 0 && ret == 0) {
2106 			/* Roll back local locks. */
2107 			NFSUNLOCKSTATE();
2108 			if (vnode_unlocked == 0) {
2109 				ASSERT_VOP_ELOCKED(vp, "nfsrv_lockctrl7");
2110 				vnode_unlocked = 1;
2111 				NFSVOPUNLOCK(vp, 0);
2112 			}
2113 			nfsrv_locallock_rollback(vp, lfp, p);
2114 			NFSLOCKSTATE();
2115 			nfsrv_unlocklf(lfp);
2116 		}
2117 		if (ret == 0)
2118 			NFSUNLOCKSTATE();
2119 		goto out;
2120 	    }
2121 	  }
2122 	}
2123 
2124 	/*
2125 	 * We only get here if there was no lock that conflicted.
2126 	 */
2127 	if (new_stp->ls_flags & (NFSLCK_TEST | NFSLCK_CHECK)) {
2128 		NFSUNLOCKSTATE();
2129 		goto out;
2130 	}
2131 
2132 	/*
2133 	 * We only get here when we are creating or modifying a lock.
2134 	 * There are two variants:
2135 	 * - exist_lock_owner where lock_owner exists
2136 	 * - open_to_lock_owner with new lock_owner
2137 	 */
2138 	first = new_lop->lo_first;
2139 	end = new_lop->lo_end;
2140 	lock_flags = new_lop->lo_flags;
2141 	if (!(new_stp->ls_flags & NFSLCK_OPENTOLOCK)) {
2142 		nfsrv_updatelock(lckstp, new_lopp, &other_lop, lfp);
2143 		stateidp->seqid = ++(lckstp->ls_stateid.seqid);
2144 		if ((nd->nd_flag & ND_NFSV41) != 0 && stateidp->seqid == 0)
2145 			stateidp->seqid = lckstp->ls_stateid.seqid = 1;
2146 		stateidp->other[0] = lckstp->ls_stateid.other[0];
2147 		stateidp->other[1] = lckstp->ls_stateid.other[1];
2148 		stateidp->other[2] = lckstp->ls_stateid.other[2];
2149 	} else {
2150 		/*
2151 		 * The new open_to_lock_owner case.
2152 		 * Link the new nfsstate into the lists.
2153 		 */
2154 		new_stp->ls_seq = new_stp->ls_opentolockseq;
2155 		nfsrvd_refcache(new_stp->ls_op);
2156 		stateidp->seqid = new_stp->ls_stateid.seqid = 1;
2157 		stateidp->other[0] = new_stp->ls_stateid.other[0] =
2158 		    clp->lc_clientid.lval[0];
2159 		stateidp->other[1] = new_stp->ls_stateid.other[1] =
2160 		    clp->lc_clientid.lval[1];
2161 		stateidp->other[2] = new_stp->ls_stateid.other[2] =
2162 		    nfsrv_nextstateindex(clp);
2163 		new_stp->ls_clp = clp;
2164 		LIST_INIT(&new_stp->ls_lock);
2165 		new_stp->ls_openstp = stp;
2166 		new_stp->ls_lfp = lfp;
2167 		nfsrv_insertlock(new_lop, (struct nfslock *)new_stp, new_stp,
2168 		    lfp);
2169 		LIST_INSERT_HEAD(NFSSTATEHASH(clp, new_stp->ls_stateid),
2170 		    new_stp, ls_hash);
2171 		LIST_INSERT_HEAD(&stp->ls_open, new_stp, ls_list);
2172 		*new_lopp = NULL;
2173 		*new_stpp = NULL;
2174 		newnfsstats.srvlockowners++;
2175 		nfsrv_openpluslock++;
2176 	}
2177 	if (filestruct_locked != 0) {
2178 		NFSUNLOCKSTATE();
2179 		nfsrv_locallock_commit(lfp, lock_flags, first, end);
2180 		NFSLOCKSTATE();
2181 		nfsrv_unlocklf(lfp);
2182 	}
2183 	NFSUNLOCKSTATE();
2184 
2185 out:
2186 	if (haslock) {
2187 		NFSLOCKV4ROOTMUTEX();
2188 		nfsv4_unlock(&nfsv4rootfs_lock, 1);
2189 		NFSUNLOCKV4ROOTMUTEX();
2190 	}
2191 	if (vnode_unlocked != 0) {
2192 		NFSVOPLOCK(vp, LK_EXCLUSIVE | LK_RETRY);
2193 		if (error == 0 && (vp->v_iflag & VI_DOOMED) != 0)
2194 			error = NFSERR_SERVERFAULT;
2195 	}
2196 	if (other_lop)
2197 		FREE((caddr_t)other_lop, M_NFSDLOCK);
2198 	NFSEXITCODE2(error, nd);
2199 	return (error);
2200 }
2201 
2202 /*
2203  * Check for state errors for Open.
2204  * repstat is passed back out as an error if more critical errors
2205  * are not detected.
2206  */
2207 APPLESTATIC int
2208 nfsrv_opencheck(nfsquad_t clientid, nfsv4stateid_t *stateidp,
2209     struct nfsstate *new_stp, vnode_t vp, struct nfsrv_descript *nd,
2210     NFSPROC_T *p, int repstat)
2211 {
2212 	struct nfsstate *stp, *nstp;
2213 	struct nfsclient *clp;
2214 	struct nfsstate *ownerstp;
2215 	struct nfslockfile *lfp, *new_lfp;
2216 	int error = 0, haslock = 0, ret, readonly = 0, getfhret = 0;
2217 
2218 	if ((new_stp->ls_flags & NFSLCK_SHAREBITS) == NFSLCK_READACCESS)
2219 		readonly = 1;
2220 	/*
2221 	 * Check for restart conditions (client and server).
2222 	 */
2223 	error = nfsrv_checkrestart(clientid, new_stp->ls_flags,
2224 		&new_stp->ls_stateid, 0);
2225 	if (error)
2226 		goto out;
2227 
2228 	/*
2229 	 * Check for state resource limit exceeded.
2230 	 * Technically this should be SMP protected, but the worst
2231 	 * case error is "out by one or two" on the count when it
2232 	 * returns NFSERR_RESOURCE and the limit is just a rather
2233 	 * arbitrary high water mark, so no harm is done.
2234 	 */
2235 	if (nfsrv_openpluslock > NFSRV_V4STATELIMIT) {
2236 		error = NFSERR_RESOURCE;
2237 		goto out;
2238 	}
2239 
2240 tryagain:
2241 	MALLOC(new_lfp, struct nfslockfile *, sizeof (struct nfslockfile),
2242 	    M_NFSDLOCKFILE, M_WAITOK);
2243 	if (vp)
2244 		getfhret = nfsrv_getlockfh(vp, new_stp->ls_flags, new_lfp,
2245 		    NULL, p);
2246 	NFSLOCKSTATE();
2247 	/*
2248 	 * Get the nfsclient structure.
2249 	 */
2250 	error = nfsrv_getclient(clientid, CLOPS_RENEW, &clp, NULL,
2251 	    (nfsquad_t)((u_quad_t)0), 0, nd, p);
2252 
2253 	/*
2254 	 * Look up the open owner. See if it needs confirmation and
2255 	 * check the seq#, as required.
2256 	 */
2257 	if (!error)
2258 		nfsrv_getowner(&clp->lc_open, new_stp, &ownerstp);
2259 
2260 	if (!error && ownerstp) {
2261 		error = nfsrv_checkseqid(nd, new_stp->ls_seq, ownerstp,
2262 		    new_stp->ls_op);
2263 		/*
2264 		 * If the OpenOwner hasn't been confirmed, assume the
2265 		 * old one was a replay and this one is ok.
2266 		 * See: RFC3530 Sec. 14.2.18.
2267 		 */
2268 		if (error == NFSERR_BADSEQID &&
2269 		    (ownerstp->ls_flags & NFSLCK_NEEDSCONFIRM))
2270 			error = 0;
2271 	}
2272 
2273 	/*
2274 	 * Check for grace.
2275 	 */
2276 	if (!error)
2277 		error = nfsrv_checkgrace(nd, clp, new_stp->ls_flags);
2278 	if ((new_stp->ls_flags & NFSLCK_RECLAIM) && !error &&
2279 		nfsrv_checkstable(clp))
2280 		error = NFSERR_NOGRACE;
2281 
2282 	/*
2283 	 * If none of the above errors occurred, let repstat be
2284 	 * returned.
2285 	 */
2286 	if (repstat && !error)
2287 		error = repstat;
2288 	if (error) {
2289 		NFSUNLOCKSTATE();
2290 		if (haslock) {
2291 			NFSLOCKV4ROOTMUTEX();
2292 			nfsv4_unlock(&nfsv4rootfs_lock, 1);
2293 			NFSUNLOCKV4ROOTMUTEX();
2294 		}
2295 		free((caddr_t)new_lfp, M_NFSDLOCKFILE);
2296 		goto out;
2297 	}
2298 
2299 	/*
2300 	 * If vp == NULL, the file doesn't exist yet, so return ok.
2301 	 * (This always happens on the first pass, so haslock must be 0.)
2302 	 */
2303 	if (vp == NULL) {
2304 		NFSUNLOCKSTATE();
2305 		FREE((caddr_t)new_lfp, M_NFSDLOCKFILE);
2306 		goto out;
2307 	}
2308 
2309 	/*
2310 	 * Get the structure for the underlying file.
2311 	 */
2312 	if (getfhret)
2313 		error = getfhret;
2314 	else
2315 		error = nfsrv_getlockfile(new_stp->ls_flags, &new_lfp, &lfp,
2316 		    NULL, 0);
2317 	if (new_lfp)
2318 		FREE((caddr_t)new_lfp, M_NFSDLOCKFILE);
2319 	if (error) {
2320 		NFSUNLOCKSTATE();
2321 		if (haslock) {
2322 			NFSLOCKV4ROOTMUTEX();
2323 			nfsv4_unlock(&nfsv4rootfs_lock, 1);
2324 			NFSUNLOCKV4ROOTMUTEX();
2325 		}
2326 		goto out;
2327 	}
2328 
2329 	/*
2330 	 * Search for a conflicting open/share.
2331 	 */
2332 	if (new_stp->ls_flags & NFSLCK_DELEGCUR) {
2333 	    /*
2334 	     * For Delegate_Cur, search for the matching Delegation,
2335 	     * which indicates no conflict.
2336 	     * An old delegation should have been recovered by the
2337 	     * client doing a Claim_DELEGATE_Prev, so I won't let
2338 	     * it match and return NFSERR_EXPIRED. Should I let it
2339 	     * match?
2340 	     */
2341 	    LIST_FOREACH(stp, &lfp->lf_deleg, ls_file) {
2342 		if (!(stp->ls_flags & NFSLCK_OLDDELEG) &&
2343 		    (((nd->nd_flag & ND_NFSV41) != 0 &&
2344 		    stateidp->seqid == 0) ||
2345 		    stateidp->seqid == stp->ls_stateid.seqid) &&
2346 		    !NFSBCMP(stateidp->other, stp->ls_stateid.other,
2347 			  NFSX_STATEIDOTHER))
2348 			break;
2349 	    }
2350 	    if (stp == LIST_END(&lfp->lf_deleg) ||
2351 		((new_stp->ls_flags & NFSLCK_WRITEACCESS) &&
2352 		 (stp->ls_flags & NFSLCK_DELEGREAD))) {
2353 		NFSUNLOCKSTATE();
2354 		if (haslock) {
2355 			NFSLOCKV4ROOTMUTEX();
2356 			nfsv4_unlock(&nfsv4rootfs_lock, 1);
2357 			NFSUNLOCKV4ROOTMUTEX();
2358 		}
2359 		error = NFSERR_EXPIRED;
2360 		goto out;
2361 	    }
2362 	}
2363 
2364 	/*
2365 	 * Check for access/deny bit conflicts. I check for the same
2366 	 * owner as well, in case the client didn't bother.
2367 	 */
2368 	LIST_FOREACH(stp, &lfp->lf_open, ls_file) {
2369 		if (!(new_stp->ls_flags & NFSLCK_DELEGCUR) &&
2370 		    (((new_stp->ls_flags & NFSLCK_ACCESSBITS) &
2371 		      ((stp->ls_flags>>NFSLCK_SHIFT) & NFSLCK_ACCESSBITS))||
2372 		     ((stp->ls_flags & NFSLCK_ACCESSBITS) &
2373 		      ((new_stp->ls_flags>>NFSLCK_SHIFT)&NFSLCK_ACCESSBITS)))){
2374 			ret = nfsrv_clientconflict(stp->ls_clp,&haslock,vp,p);
2375 			if (ret == 1) {
2376 				/*
2377 				 * nfsrv_clientconflict() unlocks
2378 				 * state when it returns non-zero.
2379 				 */
2380 				goto tryagain;
2381 			}
2382 			if (ret == 2)
2383 				error = NFSERR_PERM;
2384 			else if (new_stp->ls_flags & NFSLCK_RECLAIM)
2385 				error = NFSERR_RECLAIMCONFLICT;
2386 			else
2387 				error = NFSERR_SHAREDENIED;
2388 			if (ret == 0)
2389 				NFSUNLOCKSTATE();
2390 			if (haslock) {
2391 				NFSLOCKV4ROOTMUTEX();
2392 				nfsv4_unlock(&nfsv4rootfs_lock, 1);
2393 				NFSUNLOCKV4ROOTMUTEX();
2394 			}
2395 			goto out;
2396 		}
2397 	}
2398 
2399 	/*
2400 	 * Check for a conflicting delegation. If one is found, call
2401 	 * nfsrv_delegconflict() to handle it. If the v4root lock hasn't
2402 	 * been set yet, it will get the lock. Otherwise, it will recall
2403 	 * the delegation. Then, we try try again...
2404 	 * (If NFSLCK_DELEGCUR is set, it has a delegation, so there
2405 	 *  isn't a conflict.)
2406 	 * I currently believe the conflict algorithm to be:
2407 	 * For Open with Read Access and Deny None
2408 	 * - there is a conflict iff a different client has a write delegation
2409 	 * For Open with other Write Access or any Deny except None
2410 	 * - there is a conflict if a different client has any delegation
2411 	 * - there is a conflict if the same client has a read delegation
2412 	 *   (The current concensus is that this last case should be
2413 	 *    considered a conflict since the client with a read delegation
2414 	 *    could have done an Open with ReadAccess and WriteDeny
2415 	 *    locally and then not have checked for the WriteDeny.)
2416 	 * Don't check for a Reclaim, since that will be dealt with
2417 	 * by nfsrv_openctrl().
2418 	 */
2419 	if (!(new_stp->ls_flags &
2420 		(NFSLCK_DELEGPREV | NFSLCK_DELEGCUR | NFSLCK_RECLAIM))) {
2421 	    stp = LIST_FIRST(&lfp->lf_deleg);
2422 	    while (stp != LIST_END(&lfp->lf_deleg)) {
2423 		nstp = LIST_NEXT(stp, ls_file);
2424 		if ((readonly && stp->ls_clp != clp &&
2425 		       (stp->ls_flags & NFSLCK_DELEGWRITE)) ||
2426 		    (!readonly && (stp->ls_clp != clp ||
2427 		         (stp->ls_flags & NFSLCK_DELEGREAD)))) {
2428 			ret = nfsrv_delegconflict(stp, &haslock, p, vp);
2429 			if (ret) {
2430 			    /*
2431 			     * nfsrv_delegconflict() unlocks state
2432 			     * when it returns non-zero.
2433 			     */
2434 			    if (ret == -1)
2435 				goto tryagain;
2436 			    error = ret;
2437 			    goto out;
2438 			}
2439 		}
2440 		stp = nstp;
2441 	    }
2442 	}
2443 	NFSUNLOCKSTATE();
2444 	if (haslock) {
2445 		NFSLOCKV4ROOTMUTEX();
2446 		nfsv4_unlock(&nfsv4rootfs_lock, 1);
2447 		NFSUNLOCKV4ROOTMUTEX();
2448 	}
2449 
2450 out:
2451 	NFSEXITCODE2(error, nd);
2452 	return (error);
2453 }
2454 
2455 /*
2456  * Open control function to create/update open state for an open.
2457  */
2458 APPLESTATIC int
2459 nfsrv_openctrl(struct nfsrv_descript *nd, vnode_t vp,
2460     struct nfsstate **new_stpp, nfsquad_t clientid, nfsv4stateid_t *stateidp,
2461     nfsv4stateid_t *delegstateidp, u_int32_t *rflagsp, struct nfsexstuff *exp,
2462     NFSPROC_T *p, u_quad_t filerev)
2463 {
2464 	struct nfsstate *new_stp = *new_stpp;
2465 	struct nfsstate *stp, *nstp;
2466 	struct nfsstate *openstp = NULL, *new_open, *ownerstp, *new_deleg;
2467 	struct nfslockfile *lfp, *new_lfp;
2468 	struct nfsclient *clp;
2469 	int error = 0, haslock = 0, ret, delegate = 1, writedeleg = 1;
2470 	int readonly = 0, cbret = 1, getfhret = 0;
2471 
2472 	if ((new_stp->ls_flags & NFSLCK_SHAREBITS) == NFSLCK_READACCESS)
2473 		readonly = 1;
2474 	/*
2475 	 * Check for restart conditions (client and server).
2476 	 * (Paranoia, should have been detected by nfsrv_opencheck().)
2477 	 * If an error does show up, return NFSERR_EXPIRED, since the
2478 	 * the seqid# has already been incremented.
2479 	 */
2480 	error = nfsrv_checkrestart(clientid, new_stp->ls_flags,
2481 	    &new_stp->ls_stateid, 0);
2482 	if (error) {
2483 		printf("Nfsd: openctrl unexpected restart err=%d\n",
2484 		    error);
2485 		error = NFSERR_EXPIRED;
2486 		goto out;
2487 	}
2488 
2489 tryagain:
2490 	MALLOC(new_lfp, struct nfslockfile *, sizeof (struct nfslockfile),
2491 	    M_NFSDLOCKFILE, M_WAITOK);
2492 	MALLOC(new_open, struct nfsstate *, sizeof (struct nfsstate),
2493 	    M_NFSDSTATE, M_WAITOK);
2494 	MALLOC(new_deleg, struct nfsstate *, sizeof (struct nfsstate),
2495 	    M_NFSDSTATE, M_WAITOK);
2496 	getfhret = nfsrv_getlockfh(vp, new_stp->ls_flags, new_lfp,
2497 	    NULL, p);
2498 	NFSLOCKSTATE();
2499 	/*
2500 	 * Get the client structure. Since the linked lists could be changed
2501 	 * by other nfsd processes if this process does a tsleep(), one of
2502 	 * two things must be done.
2503 	 * 1 - don't tsleep()
2504 	 * or
2505 	 * 2 - get the nfsv4_lock() { indicated by haslock == 1 }
2506 	 *     before using the lists, since this lock stops the other
2507 	 *     nfsd. This should only be used for rare cases, since it
2508 	 *     essentially single threads the nfsd.
2509 	 *     At this time, it is only done for cases where the stable
2510 	 *     storage file must be written prior to completion of state
2511 	 *     expiration.
2512 	 */
2513 	error = nfsrv_getclient(clientid, CLOPS_RENEW, &clp, NULL,
2514 	    (nfsquad_t)((u_quad_t)0), 0, nd, p);
2515 	if (!error && (clp->lc_flags & LCL_NEEDSCBNULL) &&
2516 	    clp->lc_program) {
2517 		/*
2518 		 * This happens on the first open for a client
2519 		 * that supports callbacks.
2520 		 */
2521 		NFSUNLOCKSTATE();
2522 		/*
2523 		 * Although nfsrv_docallback() will sleep, clp won't
2524 		 * go away, since they are only removed when the
2525 		 * nfsv4_lock() has blocked the nfsd threads. The
2526 		 * fields in clp can change, but having multiple
2527 		 * threads do this Null callback RPC should be
2528 		 * harmless.
2529 		 */
2530 		cbret = nfsrv_docallback(clp, NFSV4PROC_CBNULL,
2531 		    NULL, 0, NULL, NULL, NULL, p);
2532 		NFSLOCKSTATE();
2533 		clp->lc_flags &= ~LCL_NEEDSCBNULL;
2534 		if (!cbret)
2535 			clp->lc_flags |= LCL_CALLBACKSON;
2536 	}
2537 
2538 	/*
2539 	 * Look up the open owner. See if it needs confirmation and
2540 	 * check the seq#, as required.
2541 	 */
2542 	if (!error)
2543 		nfsrv_getowner(&clp->lc_open, new_stp, &ownerstp);
2544 
2545 	if (error) {
2546 		NFSUNLOCKSTATE();
2547 		printf("Nfsd: openctrl unexpected state err=%d\n",
2548 			error);
2549 		free((caddr_t)new_lfp, M_NFSDLOCKFILE);
2550 		free((caddr_t)new_open, M_NFSDSTATE);
2551 		free((caddr_t)new_deleg, M_NFSDSTATE);
2552 		if (haslock) {
2553 			NFSLOCKV4ROOTMUTEX();
2554 			nfsv4_unlock(&nfsv4rootfs_lock, 1);
2555 			NFSUNLOCKV4ROOTMUTEX();
2556 		}
2557 		error = NFSERR_EXPIRED;
2558 		goto out;
2559 	}
2560 
2561 	if (new_stp->ls_flags & NFSLCK_RECLAIM)
2562 		nfsrv_markstable(clp);
2563 
2564 	/*
2565 	 * Get the structure for the underlying file.
2566 	 */
2567 	if (getfhret)
2568 		error = getfhret;
2569 	else
2570 		error = nfsrv_getlockfile(new_stp->ls_flags, &new_lfp, &lfp,
2571 		    NULL, 0);
2572 	if (new_lfp)
2573 		FREE((caddr_t)new_lfp, M_NFSDLOCKFILE);
2574 	if (error) {
2575 		NFSUNLOCKSTATE();
2576 		printf("Nfsd openctrl unexpected getlockfile err=%d\n",
2577 		    error);
2578 		free((caddr_t)new_open, M_NFSDSTATE);
2579 		free((caddr_t)new_deleg, M_NFSDSTATE);
2580 		if (haslock) {
2581 			NFSLOCKV4ROOTMUTEX();
2582 			nfsv4_unlock(&nfsv4rootfs_lock, 1);
2583 			NFSUNLOCKV4ROOTMUTEX();
2584 		}
2585 		goto out;
2586 	}
2587 
2588 	/*
2589 	 * Search for a conflicting open/share.
2590 	 */
2591 	if (new_stp->ls_flags & NFSLCK_DELEGCUR) {
2592 	    /*
2593 	     * For Delegate_Cur, search for the matching Delegation,
2594 	     * which indicates no conflict.
2595 	     * An old delegation should have been recovered by the
2596 	     * client doing a Claim_DELEGATE_Prev, so I won't let
2597 	     * it match and return NFSERR_EXPIRED. Should I let it
2598 	     * match?
2599 	     */
2600 	    LIST_FOREACH(stp, &lfp->lf_deleg, ls_file) {
2601 		if (!(stp->ls_flags & NFSLCK_OLDDELEG) &&
2602 		    (((nd->nd_flag & ND_NFSV41) != 0 &&
2603 		    stateidp->seqid == 0) ||
2604 		    stateidp->seqid == stp->ls_stateid.seqid) &&
2605 		    !NFSBCMP(stateidp->other, stp->ls_stateid.other,
2606 			NFSX_STATEIDOTHER))
2607 			break;
2608 	    }
2609 	    if (stp == LIST_END(&lfp->lf_deleg) ||
2610 		((new_stp->ls_flags & NFSLCK_WRITEACCESS) &&
2611 		 (stp->ls_flags & NFSLCK_DELEGREAD))) {
2612 		NFSUNLOCKSTATE();
2613 		printf("Nfsd openctrl unexpected expiry\n");
2614 		free((caddr_t)new_open, M_NFSDSTATE);
2615 		free((caddr_t)new_deleg, M_NFSDSTATE);
2616 		if (haslock) {
2617 			NFSLOCKV4ROOTMUTEX();
2618 			nfsv4_unlock(&nfsv4rootfs_lock, 1);
2619 			NFSUNLOCKV4ROOTMUTEX();
2620 		}
2621 		error = NFSERR_EXPIRED;
2622 		goto out;
2623 	    }
2624 
2625 	    /*
2626 	     * Don't issue a Delegation, since one already exists and
2627 	     * delay delegation timeout, as required.
2628 	     */
2629 	    delegate = 0;
2630 	    nfsrv_delaydelegtimeout(stp);
2631 	}
2632 
2633 	/*
2634 	 * Check for access/deny bit conflicts. I also check for the
2635 	 * same owner, since the client might not have bothered to check.
2636 	 * Also, note an open for the same file and owner, if found,
2637 	 * which is all we do here for Delegate_Cur, since conflict
2638 	 * checking is already done.
2639 	 */
2640 	LIST_FOREACH(stp, &lfp->lf_open, ls_file) {
2641 		if (ownerstp && stp->ls_openowner == ownerstp)
2642 			openstp = stp;
2643 		if (!(new_stp->ls_flags & NFSLCK_DELEGCUR)) {
2644 		    /*
2645 		     * If another client has the file open, the only
2646 		     * delegation that can be issued is a Read delegation
2647 		     * and only if it is a Read open with Deny none.
2648 		     */
2649 		    if (clp != stp->ls_clp) {
2650 			if ((stp->ls_flags & NFSLCK_SHAREBITS) ==
2651 			    NFSLCK_READACCESS)
2652 			    writedeleg = 0;
2653 			else
2654 			    delegate = 0;
2655 		    }
2656 		    if(((new_stp->ls_flags & NFSLCK_ACCESSBITS) &
2657 		        ((stp->ls_flags>>NFSLCK_SHIFT) & NFSLCK_ACCESSBITS))||
2658 		       ((stp->ls_flags & NFSLCK_ACCESSBITS) &
2659 		        ((new_stp->ls_flags>>NFSLCK_SHIFT)&NFSLCK_ACCESSBITS))){
2660 			ret = nfsrv_clientconflict(stp->ls_clp,&haslock,vp,p);
2661 			if (ret == 1) {
2662 				/*
2663 				 * nfsrv_clientconflict() unlocks state
2664 				 * when it returns non-zero.
2665 				 */
2666 				free((caddr_t)new_open, M_NFSDSTATE);
2667 				free((caddr_t)new_deleg, M_NFSDSTATE);
2668 				openstp = NULL;
2669 				goto tryagain;
2670 			}
2671 			if (ret == 2)
2672 				error = NFSERR_PERM;
2673 			else if (new_stp->ls_flags & NFSLCK_RECLAIM)
2674 				error = NFSERR_RECLAIMCONFLICT;
2675 			else
2676 				error = NFSERR_SHAREDENIED;
2677 			if (ret == 0)
2678 				NFSUNLOCKSTATE();
2679 			if (haslock) {
2680 				NFSLOCKV4ROOTMUTEX();
2681 				nfsv4_unlock(&nfsv4rootfs_lock, 1);
2682 				NFSUNLOCKV4ROOTMUTEX();
2683 			}
2684 			free((caddr_t)new_open, M_NFSDSTATE);
2685 			free((caddr_t)new_deleg, M_NFSDSTATE);
2686 			printf("nfsd openctrl unexpected client cnfl\n");
2687 			goto out;
2688 		    }
2689 		}
2690 	}
2691 
2692 	/*
2693 	 * Check for a conflicting delegation. If one is found, call
2694 	 * nfsrv_delegconflict() to handle it. If the v4root lock hasn't
2695 	 * been set yet, it will get the lock. Otherwise, it will recall
2696 	 * the delegation. Then, we try try again...
2697 	 * (If NFSLCK_DELEGCUR is set, it has a delegation, so there
2698 	 *  isn't a conflict.)
2699 	 * I currently believe the conflict algorithm to be:
2700 	 * For Open with Read Access and Deny None
2701 	 * - there is a conflict iff a different client has a write delegation
2702 	 * For Open with other Write Access or any Deny except None
2703 	 * - there is a conflict if a different client has any delegation
2704 	 * - there is a conflict if the same client has a read delegation
2705 	 *   (The current concensus is that this last case should be
2706 	 *    considered a conflict since the client with a read delegation
2707 	 *    could have done an Open with ReadAccess and WriteDeny
2708 	 *    locally and then not have checked for the WriteDeny.)
2709 	 */
2710 	if (!(new_stp->ls_flags & (NFSLCK_DELEGPREV | NFSLCK_DELEGCUR))) {
2711 	    stp = LIST_FIRST(&lfp->lf_deleg);
2712 	    while (stp != LIST_END(&lfp->lf_deleg)) {
2713 		nstp = LIST_NEXT(stp, ls_file);
2714 		if (stp->ls_clp != clp && (stp->ls_flags & NFSLCK_DELEGREAD))
2715 			writedeleg = 0;
2716 		else
2717 			delegate = 0;
2718 		if ((readonly && stp->ls_clp != clp &&
2719 		       (stp->ls_flags & NFSLCK_DELEGWRITE)) ||
2720 		    (!readonly && (stp->ls_clp != clp ||
2721 		         (stp->ls_flags & NFSLCK_DELEGREAD)))) {
2722 		    if (new_stp->ls_flags & NFSLCK_RECLAIM) {
2723 			delegate = 2;
2724 		    } else {
2725 			ret = nfsrv_delegconflict(stp, &haslock, p, vp);
2726 			if (ret) {
2727 			    /*
2728 			     * nfsrv_delegconflict() unlocks state
2729 			     * when it returns non-zero.
2730 			     */
2731 			    printf("Nfsd openctrl unexpected deleg cnfl\n");
2732 			    free((caddr_t)new_open, M_NFSDSTATE);
2733 			    free((caddr_t)new_deleg, M_NFSDSTATE);
2734 			    if (ret == -1) {
2735 				openstp = NULL;
2736 				goto tryagain;
2737 			    }
2738 			    error = ret;
2739 			    goto out;
2740 			}
2741 		    }
2742 		}
2743 		stp = nstp;
2744 	    }
2745 	}
2746 
2747 	/*
2748 	 * We only get here if there was no open that conflicted.
2749 	 * If an open for the owner exists, or in the access/deny bits.
2750 	 * Otherwise it is a new open. If the open_owner hasn't been
2751 	 * confirmed, replace the open with the new one needing confirmation,
2752 	 * otherwise add the open.
2753 	 */
2754 	if (new_stp->ls_flags & NFSLCK_DELEGPREV) {
2755 	    /*
2756 	     * Handle NFSLCK_DELEGPREV by searching the old delegations for
2757 	     * a match. If found, just move the old delegation to the current
2758 	     * delegation list and issue open. If not found, return
2759 	     * NFSERR_EXPIRED.
2760 	     */
2761 	    LIST_FOREACH(stp, &clp->lc_olddeleg, ls_list) {
2762 		if (stp->ls_lfp == lfp) {
2763 		    /* Found it */
2764 		    if (stp->ls_clp != clp)
2765 			panic("olddeleg clp");
2766 		    LIST_REMOVE(stp, ls_list);
2767 		    LIST_REMOVE(stp, ls_hash);
2768 		    stp->ls_flags &= ~NFSLCK_OLDDELEG;
2769 		    stp->ls_stateid.seqid = delegstateidp->seqid = 1;
2770 		    stp->ls_stateid.other[0] = delegstateidp->other[0] =
2771 			clp->lc_clientid.lval[0];
2772 		    stp->ls_stateid.other[1] = delegstateidp->other[1] =
2773 			clp->lc_clientid.lval[1];
2774 		    stp->ls_stateid.other[2] = delegstateidp->other[2] =
2775 			nfsrv_nextstateindex(clp);
2776 		    stp->ls_compref = nd->nd_compref;
2777 		    LIST_INSERT_HEAD(&clp->lc_deleg, stp, ls_list);
2778 		    LIST_INSERT_HEAD(NFSSTATEHASH(clp,
2779 			stp->ls_stateid), stp, ls_hash);
2780 		    if (stp->ls_flags & NFSLCK_DELEGWRITE)
2781 			*rflagsp |= NFSV4OPEN_WRITEDELEGATE;
2782 		    else
2783 			*rflagsp |= NFSV4OPEN_READDELEGATE;
2784 		    clp->lc_delegtime = NFSD_MONOSEC +
2785 			nfsrv_lease + NFSRV_LEASEDELTA;
2786 
2787 		    /*
2788 		     * Now, do the associated open.
2789 		     */
2790 		    new_open->ls_stateid.seqid = 1;
2791 		    new_open->ls_stateid.other[0] = clp->lc_clientid.lval[0];
2792 		    new_open->ls_stateid.other[1] = clp->lc_clientid.lval[1];
2793 		    new_open->ls_stateid.other[2] = nfsrv_nextstateindex(clp);
2794 		    new_open->ls_flags = (new_stp->ls_flags&NFSLCK_DENYBITS)|
2795 			NFSLCK_OPEN;
2796 		    if (stp->ls_flags & NFSLCK_DELEGWRITE)
2797 			new_open->ls_flags |= (NFSLCK_READACCESS |
2798 			    NFSLCK_WRITEACCESS);
2799 		    else
2800 			new_open->ls_flags |= NFSLCK_READACCESS;
2801 		    new_open->ls_uid = new_stp->ls_uid;
2802 		    new_open->ls_lfp = lfp;
2803 		    new_open->ls_clp = clp;
2804 		    LIST_INIT(&new_open->ls_open);
2805 		    LIST_INSERT_HEAD(&lfp->lf_open, new_open, ls_file);
2806 		    LIST_INSERT_HEAD(NFSSTATEHASH(clp, new_open->ls_stateid),
2807 			new_open, ls_hash);
2808 		    /*
2809 		     * and handle the open owner
2810 		     */
2811 		    if (ownerstp) {
2812 			new_open->ls_openowner = ownerstp;
2813 			LIST_INSERT_HEAD(&ownerstp->ls_open,new_open,ls_list);
2814 		    } else {
2815 			new_open->ls_openowner = new_stp;
2816 			new_stp->ls_flags = 0;
2817 			nfsrvd_refcache(new_stp->ls_op);
2818 			new_stp->ls_noopens = 0;
2819 			LIST_INIT(&new_stp->ls_open);
2820 			LIST_INSERT_HEAD(&new_stp->ls_open, new_open, ls_list);
2821 			LIST_INSERT_HEAD(&clp->lc_open, new_stp, ls_list);
2822 			*new_stpp = NULL;
2823 			newnfsstats.srvopenowners++;
2824 			nfsrv_openpluslock++;
2825 		    }
2826 		    openstp = new_open;
2827 		    new_open = NULL;
2828 		    newnfsstats.srvopens++;
2829 		    nfsrv_openpluslock++;
2830 		    break;
2831 		}
2832 	    }
2833 	    if (stp == LIST_END(&clp->lc_olddeleg))
2834 		error = NFSERR_EXPIRED;
2835 	} else if (new_stp->ls_flags & (NFSLCK_DELEGREAD | NFSLCK_DELEGWRITE)) {
2836 	    /*
2837 	     * Scan to see that no delegation for this client and file
2838 	     * doesn't already exist.
2839 	     * There also shouldn't yet be an Open for this file and
2840 	     * openowner.
2841 	     */
2842 	    LIST_FOREACH(stp, &lfp->lf_deleg, ls_file) {
2843 		if (stp->ls_clp == clp)
2844 		    break;
2845 	    }
2846 	    if (stp == LIST_END(&lfp->lf_deleg) && openstp == NULL) {
2847 		/*
2848 		 * This is the Claim_Previous case with a delegation
2849 		 * type != Delegate_None.
2850 		 */
2851 		/*
2852 		 * First, add the delegation. (Although we must issue the
2853 		 * delegation, we can also ask for an immediate return.)
2854 		 */
2855 		new_deleg->ls_stateid.seqid = delegstateidp->seqid = 1;
2856 		new_deleg->ls_stateid.other[0] = delegstateidp->other[0] =
2857 		    clp->lc_clientid.lval[0];
2858 		new_deleg->ls_stateid.other[1] = delegstateidp->other[1] =
2859 		    clp->lc_clientid.lval[1];
2860 		new_deleg->ls_stateid.other[2] = delegstateidp->other[2] =
2861 		    nfsrv_nextstateindex(clp);
2862 		if (new_stp->ls_flags & NFSLCK_DELEGWRITE) {
2863 		    new_deleg->ls_flags = (NFSLCK_DELEGWRITE |
2864 			NFSLCK_READACCESS | NFSLCK_WRITEACCESS);
2865 		    *rflagsp |= NFSV4OPEN_WRITEDELEGATE;
2866 		} else {
2867 		    new_deleg->ls_flags = (NFSLCK_DELEGREAD |
2868 			NFSLCK_READACCESS);
2869 		    *rflagsp |= NFSV4OPEN_READDELEGATE;
2870 		}
2871 		new_deleg->ls_uid = new_stp->ls_uid;
2872 		new_deleg->ls_lfp = lfp;
2873 		new_deleg->ls_clp = clp;
2874 		new_deleg->ls_filerev = filerev;
2875 		new_deleg->ls_compref = nd->nd_compref;
2876 		LIST_INSERT_HEAD(&lfp->lf_deleg, new_deleg, ls_file);
2877 		LIST_INSERT_HEAD(NFSSTATEHASH(clp,
2878 		    new_deleg->ls_stateid), new_deleg, ls_hash);
2879 		LIST_INSERT_HEAD(&clp->lc_deleg, new_deleg, ls_list);
2880 		new_deleg = NULL;
2881 		if (delegate == 2 || nfsrv_issuedelegs == 0 ||
2882 		    (clp->lc_flags & (LCL_CALLBACKSON | LCL_CBDOWN)) !=
2883 		     LCL_CALLBACKSON ||
2884 		    NFSRV_V4DELEGLIMIT(nfsrv_delegatecnt) ||
2885 		    !NFSVNO_DELEGOK(vp))
2886 		    *rflagsp |= NFSV4OPEN_RECALL;
2887 		newnfsstats.srvdelegates++;
2888 		nfsrv_openpluslock++;
2889 		nfsrv_delegatecnt++;
2890 
2891 		/*
2892 		 * Now, do the associated open.
2893 		 */
2894 		new_open->ls_stateid.seqid = 1;
2895 		new_open->ls_stateid.other[0] = clp->lc_clientid.lval[0];
2896 		new_open->ls_stateid.other[1] = clp->lc_clientid.lval[1];
2897 		new_open->ls_stateid.other[2] = nfsrv_nextstateindex(clp);
2898 		new_open->ls_flags = (new_stp->ls_flags & NFSLCK_DENYBITS) |
2899 		    NFSLCK_OPEN;
2900 		if (new_stp->ls_flags & NFSLCK_DELEGWRITE)
2901 			new_open->ls_flags |= (NFSLCK_READACCESS |
2902 			    NFSLCK_WRITEACCESS);
2903 		else
2904 			new_open->ls_flags |= NFSLCK_READACCESS;
2905 		new_open->ls_uid = new_stp->ls_uid;
2906 		new_open->ls_lfp = lfp;
2907 		new_open->ls_clp = clp;
2908 		LIST_INIT(&new_open->ls_open);
2909 		LIST_INSERT_HEAD(&lfp->lf_open, new_open, ls_file);
2910 		LIST_INSERT_HEAD(NFSSTATEHASH(clp, new_open->ls_stateid),
2911 		   new_open, ls_hash);
2912 		/*
2913 		 * and handle the open owner
2914 		 */
2915 		if (ownerstp) {
2916 		    new_open->ls_openowner = ownerstp;
2917 		    LIST_INSERT_HEAD(&ownerstp->ls_open, new_open, ls_list);
2918 		} else {
2919 		    new_open->ls_openowner = new_stp;
2920 		    new_stp->ls_flags = 0;
2921 		    nfsrvd_refcache(new_stp->ls_op);
2922 		    new_stp->ls_noopens = 0;
2923 		    LIST_INIT(&new_stp->ls_open);
2924 		    LIST_INSERT_HEAD(&new_stp->ls_open, new_open, ls_list);
2925 		    LIST_INSERT_HEAD(&clp->lc_open, new_stp, ls_list);
2926 		    *new_stpp = NULL;
2927 		    newnfsstats.srvopenowners++;
2928 		    nfsrv_openpluslock++;
2929 		}
2930 		openstp = new_open;
2931 		new_open = NULL;
2932 		newnfsstats.srvopens++;
2933 		nfsrv_openpluslock++;
2934 	    } else {
2935 		error = NFSERR_RECLAIMCONFLICT;
2936 	    }
2937 	} else if (ownerstp) {
2938 		if (ownerstp->ls_flags & NFSLCK_NEEDSCONFIRM) {
2939 		    /* Replace the open */
2940 		    if (ownerstp->ls_op)
2941 			nfsrvd_derefcache(ownerstp->ls_op);
2942 		    ownerstp->ls_op = new_stp->ls_op;
2943 		    nfsrvd_refcache(ownerstp->ls_op);
2944 		    ownerstp->ls_seq = new_stp->ls_seq;
2945 		    *rflagsp |= NFSV4OPEN_RESULTCONFIRM;
2946 		    stp = LIST_FIRST(&ownerstp->ls_open);
2947 		    stp->ls_flags = (new_stp->ls_flags & NFSLCK_SHAREBITS) |
2948 			NFSLCK_OPEN;
2949 		    stp->ls_stateid.seqid = 1;
2950 		    stp->ls_uid = new_stp->ls_uid;
2951 		    if (lfp != stp->ls_lfp) {
2952 			LIST_REMOVE(stp, ls_file);
2953 			LIST_INSERT_HEAD(&lfp->lf_open, stp, ls_file);
2954 			stp->ls_lfp = lfp;
2955 		    }
2956 		    openstp = stp;
2957 		} else if (openstp) {
2958 		    openstp->ls_flags |= (new_stp->ls_flags & NFSLCK_SHAREBITS);
2959 		    openstp->ls_stateid.seqid++;
2960 		    if ((nd->nd_flag & ND_NFSV41) != 0 &&
2961 			openstp->ls_stateid.seqid == 0)
2962 			openstp->ls_stateid.seqid = 1;
2963 
2964 		    /*
2965 		     * This is where we can choose to issue a delegation.
2966 		     */
2967 		    if (delegate == 0 || writedeleg == 0 ||
2968 			NFSVNO_EXRDONLY(exp) || (readonly != 0 &&
2969 			nfsrv_writedelegifpos == 0) ||
2970 			!NFSVNO_DELEGOK(vp) ||
2971 			(new_stp->ls_flags & NFSLCK_WANTRDELEG) != 0 ||
2972 			(clp->lc_flags & (LCL_CALLBACKSON | LCL_CBDOWN)) !=
2973 			 LCL_CALLBACKSON)
2974 			*rflagsp |= NFSV4OPEN_WDCONTENTION;
2975 		    else if (nfsrv_issuedelegs == 0 ||
2976 			NFSRV_V4DELEGLIMIT(nfsrv_delegatecnt))
2977 			*rflagsp |= NFSV4OPEN_WDRESOURCE;
2978 		    else if ((new_stp->ls_flags & NFSLCK_WANTNODELEG) != 0)
2979 			*rflagsp |= NFSV4OPEN_WDNOTWANTED;
2980 		    else {
2981 			new_deleg->ls_stateid.seqid = delegstateidp->seqid = 1;
2982 			new_deleg->ls_stateid.other[0] = delegstateidp->other[0]
2983 			    = clp->lc_clientid.lval[0];
2984 			new_deleg->ls_stateid.other[1] = delegstateidp->other[1]
2985 			    = clp->lc_clientid.lval[1];
2986 			new_deleg->ls_stateid.other[2] = delegstateidp->other[2]
2987 			    = nfsrv_nextstateindex(clp);
2988 			new_deleg->ls_flags = (NFSLCK_DELEGWRITE |
2989 			    NFSLCK_READACCESS | NFSLCK_WRITEACCESS);
2990 			*rflagsp |= NFSV4OPEN_WRITEDELEGATE;
2991 			new_deleg->ls_uid = new_stp->ls_uid;
2992 			new_deleg->ls_lfp = lfp;
2993 			new_deleg->ls_clp = clp;
2994 			new_deleg->ls_filerev = filerev;
2995 			new_deleg->ls_compref = nd->nd_compref;
2996 			LIST_INSERT_HEAD(&lfp->lf_deleg, new_deleg, ls_file);
2997 			LIST_INSERT_HEAD(NFSSTATEHASH(clp,
2998 			    new_deleg->ls_stateid), new_deleg, ls_hash);
2999 			LIST_INSERT_HEAD(&clp->lc_deleg, new_deleg, ls_list);
3000 			new_deleg = NULL;
3001 			newnfsstats.srvdelegates++;
3002 			nfsrv_openpluslock++;
3003 			nfsrv_delegatecnt++;
3004 		    }
3005 		} else {
3006 		    new_open->ls_stateid.seqid = 1;
3007 		    new_open->ls_stateid.other[0] = clp->lc_clientid.lval[0];
3008 		    new_open->ls_stateid.other[1] = clp->lc_clientid.lval[1];
3009 		    new_open->ls_stateid.other[2] = nfsrv_nextstateindex(clp);
3010 		    new_open->ls_flags = (new_stp->ls_flags & NFSLCK_SHAREBITS)|
3011 			NFSLCK_OPEN;
3012 		    new_open->ls_uid = new_stp->ls_uid;
3013 		    new_open->ls_openowner = ownerstp;
3014 		    new_open->ls_lfp = lfp;
3015 		    new_open->ls_clp = clp;
3016 		    LIST_INIT(&new_open->ls_open);
3017 		    LIST_INSERT_HEAD(&lfp->lf_open, new_open, ls_file);
3018 		    LIST_INSERT_HEAD(&ownerstp->ls_open, new_open, ls_list);
3019 		    LIST_INSERT_HEAD(NFSSTATEHASH(clp, new_open->ls_stateid),
3020 			new_open, ls_hash);
3021 		    openstp = new_open;
3022 		    new_open = NULL;
3023 		    newnfsstats.srvopens++;
3024 		    nfsrv_openpluslock++;
3025 
3026 		    /*
3027 		     * This is where we can choose to issue a delegation.
3028 		     */
3029 		    if (delegate == 0 || (writedeleg == 0 && readonly == 0) ||
3030 			!NFSVNO_DELEGOK(vp) ||
3031 			(clp->lc_flags & (LCL_CALLBACKSON | LCL_CBDOWN)) !=
3032 			 LCL_CALLBACKSON)
3033 			*rflagsp |= NFSV4OPEN_WDCONTENTION;
3034 		    else if (nfsrv_issuedelegs == 0 ||
3035 			NFSRV_V4DELEGLIMIT(nfsrv_delegatecnt))
3036 			*rflagsp |= NFSV4OPEN_WDRESOURCE;
3037 		    else if ((new_stp->ls_flags & NFSLCK_WANTNODELEG) != 0)
3038 			*rflagsp |= NFSV4OPEN_WDNOTWANTED;
3039 		    else {
3040 			new_deleg->ls_stateid.seqid = delegstateidp->seqid = 1;
3041 			new_deleg->ls_stateid.other[0] = delegstateidp->other[0]
3042 			    = clp->lc_clientid.lval[0];
3043 			new_deleg->ls_stateid.other[1] = delegstateidp->other[1]
3044 			    = clp->lc_clientid.lval[1];
3045 			new_deleg->ls_stateid.other[2] = delegstateidp->other[2]
3046 			    = nfsrv_nextstateindex(clp);
3047 			if (writedeleg && !NFSVNO_EXRDONLY(exp) &&
3048 			    (nfsrv_writedelegifpos || !readonly) &&
3049 			    (new_stp->ls_flags & NFSLCK_WANTRDELEG) == 0) {
3050 			    new_deleg->ls_flags = (NFSLCK_DELEGWRITE |
3051 				NFSLCK_READACCESS | NFSLCK_WRITEACCESS);
3052 			    *rflagsp |= NFSV4OPEN_WRITEDELEGATE;
3053 			} else {
3054 			    new_deleg->ls_flags = (NFSLCK_DELEGREAD |
3055 				NFSLCK_READACCESS);
3056 			    *rflagsp |= NFSV4OPEN_READDELEGATE;
3057 			}
3058 			new_deleg->ls_uid = new_stp->ls_uid;
3059 			new_deleg->ls_lfp = lfp;
3060 			new_deleg->ls_clp = clp;
3061 			new_deleg->ls_filerev = filerev;
3062 			new_deleg->ls_compref = nd->nd_compref;
3063 			LIST_INSERT_HEAD(&lfp->lf_deleg, new_deleg, ls_file);
3064 			LIST_INSERT_HEAD(NFSSTATEHASH(clp,
3065 			    new_deleg->ls_stateid), new_deleg, ls_hash);
3066 			LIST_INSERT_HEAD(&clp->lc_deleg, new_deleg, ls_list);
3067 			new_deleg = NULL;
3068 			newnfsstats.srvdelegates++;
3069 			nfsrv_openpluslock++;
3070 			nfsrv_delegatecnt++;
3071 		    }
3072 		}
3073 	} else {
3074 		/*
3075 		 * New owner case. Start the open_owner sequence with a
3076 		 * Needs confirmation (unless a reclaim) and hang the
3077 		 * new open off it.
3078 		 */
3079 		new_open->ls_stateid.seqid = 1;
3080 		new_open->ls_stateid.other[0] = clp->lc_clientid.lval[0];
3081 		new_open->ls_stateid.other[1] = clp->lc_clientid.lval[1];
3082 		new_open->ls_stateid.other[2] = nfsrv_nextstateindex(clp);
3083 		new_open->ls_flags = (new_stp->ls_flags & NFSLCK_SHAREBITS) |
3084 		    NFSLCK_OPEN;
3085 		new_open->ls_uid = new_stp->ls_uid;
3086 		LIST_INIT(&new_open->ls_open);
3087 		new_open->ls_openowner = new_stp;
3088 		new_open->ls_lfp = lfp;
3089 		new_open->ls_clp = clp;
3090 		LIST_INSERT_HEAD(&lfp->lf_open, new_open, ls_file);
3091 		if (new_stp->ls_flags & NFSLCK_RECLAIM) {
3092 			new_stp->ls_flags = 0;
3093 		} else if ((nd->nd_flag & ND_NFSV41) != 0) {
3094 			/* NFSv4.1 never needs confirmation. */
3095 			new_stp->ls_flags = 0;
3096 
3097 			/*
3098 			 * This is where we can choose to issue a delegation.
3099 			 */
3100 			if (delegate && nfsrv_issuedelegs &&
3101 			    (writedeleg || readonly) &&
3102 			    (clp->lc_flags & (LCL_CALLBACKSON | LCL_CBDOWN)) ==
3103 			     LCL_CALLBACKSON &&
3104 			    !NFSRV_V4DELEGLIMIT(nfsrv_delegatecnt) &&
3105 			    NFSVNO_DELEGOK(vp) &&
3106 			    ((nd->nd_flag & ND_NFSV41) == 0 ||
3107 			     (new_stp->ls_flags & NFSLCK_WANTNODELEG) == 0)) {
3108 				new_deleg->ls_stateid.seqid =
3109 				    delegstateidp->seqid = 1;
3110 				new_deleg->ls_stateid.other[0] =
3111 				    delegstateidp->other[0]
3112 				    = clp->lc_clientid.lval[0];
3113 				new_deleg->ls_stateid.other[1] =
3114 				    delegstateidp->other[1]
3115 				    = clp->lc_clientid.lval[1];
3116 				new_deleg->ls_stateid.other[2] =
3117 				    delegstateidp->other[2]
3118 				    = nfsrv_nextstateindex(clp);
3119 				if (writedeleg && !NFSVNO_EXRDONLY(exp) &&
3120 				    (nfsrv_writedelegifpos || !readonly) &&
3121 				    ((nd->nd_flag & ND_NFSV41) == 0 ||
3122 				     (new_stp->ls_flags & NFSLCK_WANTRDELEG) ==
3123 				     0)) {
3124 					new_deleg->ls_flags =
3125 					    (NFSLCK_DELEGWRITE |
3126 					     NFSLCK_READACCESS |
3127 					     NFSLCK_WRITEACCESS);
3128 					*rflagsp |= NFSV4OPEN_WRITEDELEGATE;
3129 				} else {
3130 					new_deleg->ls_flags =
3131 					    (NFSLCK_DELEGREAD |
3132 					     NFSLCK_READACCESS);
3133 					*rflagsp |= NFSV4OPEN_READDELEGATE;
3134 				}
3135 				new_deleg->ls_uid = new_stp->ls_uid;
3136 				new_deleg->ls_lfp = lfp;
3137 				new_deleg->ls_clp = clp;
3138 				new_deleg->ls_filerev = filerev;
3139 				new_deleg->ls_compref = nd->nd_compref;
3140 				LIST_INSERT_HEAD(&lfp->lf_deleg, new_deleg,
3141 				    ls_file);
3142 				LIST_INSERT_HEAD(NFSSTATEHASH(clp,
3143 				    new_deleg->ls_stateid), new_deleg, ls_hash);
3144 				LIST_INSERT_HEAD(&clp->lc_deleg, new_deleg,
3145 				    ls_list);
3146 				new_deleg = NULL;
3147 				newnfsstats.srvdelegates++;
3148 				nfsrv_openpluslock++;
3149 				nfsrv_delegatecnt++;
3150 			}
3151 		} else {
3152 			*rflagsp |= NFSV4OPEN_RESULTCONFIRM;
3153 			new_stp->ls_flags = NFSLCK_NEEDSCONFIRM;
3154 		}
3155 		nfsrvd_refcache(new_stp->ls_op);
3156 		new_stp->ls_noopens = 0;
3157 		LIST_INIT(&new_stp->ls_open);
3158 		LIST_INSERT_HEAD(&new_stp->ls_open, new_open, ls_list);
3159 		LIST_INSERT_HEAD(&clp->lc_open, new_stp, ls_list);
3160 		LIST_INSERT_HEAD(NFSSTATEHASH(clp, new_open->ls_stateid),
3161 		    new_open, ls_hash);
3162 		openstp = new_open;
3163 		new_open = NULL;
3164 		*new_stpp = NULL;
3165 		newnfsstats.srvopens++;
3166 		nfsrv_openpluslock++;
3167 		newnfsstats.srvopenowners++;
3168 		nfsrv_openpluslock++;
3169 	}
3170 	if (!error) {
3171 		stateidp->seqid = openstp->ls_stateid.seqid;
3172 		stateidp->other[0] = openstp->ls_stateid.other[0];
3173 		stateidp->other[1] = openstp->ls_stateid.other[1];
3174 		stateidp->other[2] = openstp->ls_stateid.other[2];
3175 	}
3176 	NFSUNLOCKSTATE();
3177 	if (haslock) {
3178 		NFSLOCKV4ROOTMUTEX();
3179 		nfsv4_unlock(&nfsv4rootfs_lock, 1);
3180 		NFSUNLOCKV4ROOTMUTEX();
3181 	}
3182 	if (new_open)
3183 		FREE((caddr_t)new_open, M_NFSDSTATE);
3184 	if (new_deleg)
3185 		FREE((caddr_t)new_deleg, M_NFSDSTATE);
3186 
3187 out:
3188 	NFSEXITCODE2(error, nd);
3189 	return (error);
3190 }
3191 
3192 /*
3193  * Open update. Does the confirm, downgrade and close.
3194  */
3195 APPLESTATIC int
3196 nfsrv_openupdate(vnode_t vp, struct nfsstate *new_stp, nfsquad_t clientid,
3197     nfsv4stateid_t *stateidp, struct nfsrv_descript *nd, NFSPROC_T *p)
3198 {
3199 	struct nfsstate *stp, *ownerstp;
3200 	struct nfsclient *clp;
3201 	struct nfslockfile *lfp;
3202 	u_int32_t bits;
3203 	int error = 0, gotstate = 0, len = 0;
3204 	u_char client[NFSV4_OPAQUELIMIT];
3205 
3206 	/*
3207 	 * Check for restart conditions (client and server).
3208 	 */
3209 	error = nfsrv_checkrestart(clientid, new_stp->ls_flags,
3210 	    &new_stp->ls_stateid, 0);
3211 	if (error)
3212 		goto out;
3213 
3214 	NFSLOCKSTATE();
3215 	/*
3216 	 * Get the open structure via clientid and stateid.
3217 	 */
3218 	error = nfsrv_getclient(clientid, CLOPS_RENEW, &clp, NULL,
3219 	    (nfsquad_t)((u_quad_t)0), 0, nd, p);
3220 	if (!error)
3221 		error = nfsrv_getstate(clp, &new_stp->ls_stateid,
3222 		    new_stp->ls_flags, &stp);
3223 
3224 	/*
3225 	 * Sanity check the open.
3226 	 */
3227 	if (!error && (!(stp->ls_flags & NFSLCK_OPEN) ||
3228 		(!(new_stp->ls_flags & NFSLCK_CONFIRM) &&
3229 		 (stp->ls_openowner->ls_flags & NFSLCK_NEEDSCONFIRM)) ||
3230 		((new_stp->ls_flags & NFSLCK_CONFIRM) &&
3231 		 (!(stp->ls_openowner->ls_flags & NFSLCK_NEEDSCONFIRM)))))
3232 		error = NFSERR_BADSTATEID;
3233 
3234 	if (!error)
3235 		error = nfsrv_checkseqid(nd, new_stp->ls_seq,
3236 		    stp->ls_openowner, new_stp->ls_op);
3237 	if (!error && stp->ls_stateid.seqid != new_stp->ls_stateid.seqid &&
3238 	    (((nd->nd_flag & ND_NFSV41) == 0 &&
3239 	      !(new_stp->ls_flags & NFSLCK_CONFIRM)) ||
3240 	     ((nd->nd_flag & ND_NFSV41) != 0 &&
3241 	      new_stp->ls_stateid.seqid != 0)))
3242 		error = NFSERR_OLDSTATEID;
3243 	if (!error && vnode_vtype(vp) != VREG) {
3244 		if (vnode_vtype(vp) == VDIR)
3245 			error = NFSERR_ISDIR;
3246 		else
3247 			error = NFSERR_INVAL;
3248 	}
3249 
3250 	if (error) {
3251 		/*
3252 		 * If a client tries to confirm an Open with a bad
3253 		 * seqid# and there are no byte range locks or other Opens
3254 		 * on the openowner, just throw it away, so the next use of the
3255 		 * openowner will start a fresh seq#.
3256 		 */
3257 		if (error == NFSERR_BADSEQID &&
3258 		    (new_stp->ls_flags & NFSLCK_CONFIRM) &&
3259 		    nfsrv_nootherstate(stp))
3260 			nfsrv_freeopenowner(stp->ls_openowner, 0, p);
3261 		NFSUNLOCKSTATE();
3262 		goto out;
3263 	}
3264 
3265 	/*
3266 	 * Set the return stateid.
3267 	 */
3268 	stateidp->seqid = stp->ls_stateid.seqid + 1;
3269 	if ((nd->nd_flag & ND_NFSV41) != 0 && stateidp->seqid == 0)
3270 		stateidp->seqid = 1;
3271 	stateidp->other[0] = stp->ls_stateid.other[0];
3272 	stateidp->other[1] = stp->ls_stateid.other[1];
3273 	stateidp->other[2] = stp->ls_stateid.other[2];
3274 	/*
3275 	 * Now, handle the three cases.
3276 	 */
3277 	if (new_stp->ls_flags & NFSLCK_CONFIRM) {
3278 		/*
3279 		 * If the open doesn't need confirmation, it seems to me that
3280 		 * there is a client error, but I'll just log it and keep going?
3281 		 */
3282 		if (!(stp->ls_openowner->ls_flags & NFSLCK_NEEDSCONFIRM))
3283 			printf("Nfsv4d: stray open confirm\n");
3284 		stp->ls_openowner->ls_flags = 0;
3285 		stp->ls_stateid.seqid++;
3286 		if ((nd->nd_flag & ND_NFSV41) != 0 &&
3287 		    stp->ls_stateid.seqid == 0)
3288 			stp->ls_stateid.seqid = 1;
3289 		if (!(clp->lc_flags & LCL_STAMPEDSTABLE)) {
3290 			clp->lc_flags |= LCL_STAMPEDSTABLE;
3291 			len = clp->lc_idlen;
3292 			NFSBCOPY(clp->lc_id, client, len);
3293 			gotstate = 1;
3294 		}
3295 		NFSUNLOCKSTATE();
3296 	} else if (new_stp->ls_flags & NFSLCK_CLOSE) {
3297 		ownerstp = stp->ls_openowner;
3298 		lfp = stp->ls_lfp;
3299 		if (nfsrv_dolocallocks != 0 && !LIST_EMPTY(&stp->ls_open)) {
3300 			/* Get the lf lock */
3301 			nfsrv_locklf(lfp);
3302 			NFSUNLOCKSTATE();
3303 			ASSERT_VOP_ELOCKED(vp, "nfsrv_openupdate");
3304 			NFSVOPUNLOCK(vp, 0);
3305 			if (nfsrv_freeopen(stp, vp, 1, p) == 0) {
3306 				NFSLOCKSTATE();
3307 				nfsrv_unlocklf(lfp);
3308 				NFSUNLOCKSTATE();
3309 			}
3310 			NFSVOPLOCK(vp, LK_EXCLUSIVE | LK_RETRY);
3311 		} else {
3312 			(void) nfsrv_freeopen(stp, NULL, 0, p);
3313 			NFSUNLOCKSTATE();
3314 		}
3315 	} else {
3316 		/*
3317 		 * Update the share bits, making sure that the new set are a
3318 		 * subset of the old ones.
3319 		 */
3320 		bits = (new_stp->ls_flags & NFSLCK_SHAREBITS);
3321 		if (~(stp->ls_flags) & bits) {
3322 			NFSUNLOCKSTATE();
3323 			error = NFSERR_INVAL;
3324 			goto out;
3325 		}
3326 		stp->ls_flags = (bits | NFSLCK_OPEN);
3327 		stp->ls_stateid.seqid++;
3328 		if ((nd->nd_flag & ND_NFSV41) != 0 &&
3329 		    stp->ls_stateid.seqid == 0)
3330 			stp->ls_stateid.seqid = 1;
3331 		NFSUNLOCKSTATE();
3332 	}
3333 
3334 	/*
3335 	 * If the client just confirmed its first open, write a timestamp
3336 	 * to the stable storage file.
3337 	 */
3338 	if (gotstate != 0) {
3339 		nfsrv_writestable(client, len, NFSNST_NEWSTATE, p);
3340 		nfsrv_backupstable();
3341 	}
3342 
3343 out:
3344 	NFSEXITCODE2(error, nd);
3345 	return (error);
3346 }
3347 
3348 /*
3349  * Delegation update. Does the purge and return.
3350  */
3351 APPLESTATIC int
3352 nfsrv_delegupdate(struct nfsrv_descript *nd, nfsquad_t clientid,
3353     nfsv4stateid_t *stateidp, vnode_t vp, int op, struct ucred *cred,
3354     NFSPROC_T *p)
3355 {
3356 	struct nfsstate *stp;
3357 	struct nfsclient *clp;
3358 	int error = 0;
3359 	fhandle_t fh;
3360 
3361 	/*
3362 	 * Do a sanity check against the file handle for DelegReturn.
3363 	 */
3364 	if (vp) {
3365 		error = nfsvno_getfh(vp, &fh, p);
3366 		if (error)
3367 			goto out;
3368 	}
3369 	/*
3370 	 * Check for restart conditions (client and server).
3371 	 */
3372 	if (op == NFSV4OP_DELEGRETURN)
3373 		error = nfsrv_checkrestart(clientid, NFSLCK_DELEGRETURN,
3374 			stateidp, 0);
3375 	else
3376 		error = nfsrv_checkrestart(clientid, NFSLCK_DELEGPURGE,
3377 			stateidp, 0);
3378 
3379 	NFSLOCKSTATE();
3380 	/*
3381 	 * Get the open structure via clientid and stateid.
3382 	 */
3383 	if (!error)
3384 	    error = nfsrv_getclient(clientid, CLOPS_RENEW, &clp, NULL,
3385 		(nfsquad_t)((u_quad_t)0), 0, nd, p);
3386 	if (error) {
3387 		if (error == NFSERR_CBPATHDOWN)
3388 			error = 0;
3389 		if (error == NFSERR_STALECLIENTID && op == NFSV4OP_DELEGRETURN)
3390 			error = NFSERR_STALESTATEID;
3391 	}
3392 	if (!error && op == NFSV4OP_DELEGRETURN) {
3393 	    error = nfsrv_getstate(clp, stateidp, NFSLCK_DELEGRETURN, &stp);
3394 	    if (!error && stp->ls_stateid.seqid != stateidp->seqid &&
3395 		((nd->nd_flag & ND_NFSV41) == 0 || stateidp->seqid != 0))
3396 		error = NFSERR_OLDSTATEID;
3397 	}
3398 	/*
3399 	 * NFSERR_EXPIRED means that the state has gone away,
3400 	 * so Delegations have been purged. Just return ok.
3401 	 */
3402 	if (error == NFSERR_EXPIRED && op == NFSV4OP_DELEGPURGE) {
3403 		NFSUNLOCKSTATE();
3404 		error = 0;
3405 		goto out;
3406 	}
3407 	if (error) {
3408 		NFSUNLOCKSTATE();
3409 		goto out;
3410 	}
3411 
3412 	if (op == NFSV4OP_DELEGRETURN) {
3413 		if (NFSBCMP((caddr_t)&fh, (caddr_t)&stp->ls_lfp->lf_fh,
3414 		    sizeof (fhandle_t))) {
3415 			NFSUNLOCKSTATE();
3416 			error = NFSERR_BADSTATEID;
3417 			goto out;
3418 		}
3419 		nfsrv_freedeleg(stp);
3420 	} else {
3421 		nfsrv_freedeleglist(&clp->lc_olddeleg);
3422 	}
3423 	NFSUNLOCKSTATE();
3424 	error = 0;
3425 
3426 out:
3427 	NFSEXITCODE(error);
3428 	return (error);
3429 }
3430 
3431 /*
3432  * Release lock owner.
3433  */
3434 APPLESTATIC int
3435 nfsrv_releaselckown(struct nfsstate *new_stp, nfsquad_t clientid,
3436     NFSPROC_T *p)
3437 {
3438 	struct nfsstate *stp, *nstp, *openstp, *ownstp;
3439 	struct nfsclient *clp;
3440 	int error = 0;
3441 
3442 	/*
3443 	 * Check for restart conditions (client and server).
3444 	 */
3445 	error = nfsrv_checkrestart(clientid, new_stp->ls_flags,
3446 	    &new_stp->ls_stateid, 0);
3447 	if (error)
3448 		goto out;
3449 
3450 	NFSLOCKSTATE();
3451 	/*
3452 	 * Get the lock owner by name.
3453 	 */
3454 	error = nfsrv_getclient(clientid, CLOPS_RENEW, &clp, NULL,
3455 	    (nfsquad_t)((u_quad_t)0), 0, NULL, p);
3456 	if (error) {
3457 		NFSUNLOCKSTATE();
3458 		goto out;
3459 	}
3460 	LIST_FOREACH(ownstp, &clp->lc_open, ls_list) {
3461 	    LIST_FOREACH(openstp, &ownstp->ls_open, ls_list) {
3462 		stp = LIST_FIRST(&openstp->ls_open);
3463 		while (stp != LIST_END(&openstp->ls_open)) {
3464 		    nstp = LIST_NEXT(stp, ls_list);
3465 		    /*
3466 		     * If the owner matches, check for locks and
3467 		     * then free or return an error.
3468 		     */
3469 		    if (stp->ls_ownerlen == new_stp->ls_ownerlen &&
3470 			!NFSBCMP(stp->ls_owner, new_stp->ls_owner,
3471 			 stp->ls_ownerlen)){
3472 			if (LIST_EMPTY(&stp->ls_lock)) {
3473 			    nfsrv_freelockowner(stp, NULL, 0, p);
3474 			} else {
3475 			    NFSUNLOCKSTATE();
3476 			    error = NFSERR_LOCKSHELD;
3477 			    goto out;
3478 			}
3479 		    }
3480 		    stp = nstp;
3481 		}
3482 	    }
3483 	}
3484 	NFSUNLOCKSTATE();
3485 
3486 out:
3487 	NFSEXITCODE(error);
3488 	return (error);
3489 }
3490 
3491 /*
3492  * Get the file handle for a lock structure.
3493  */
3494 static int
3495 nfsrv_getlockfh(vnode_t vp, u_short flags, struct nfslockfile *new_lfp,
3496     fhandle_t *nfhp, NFSPROC_T *p)
3497 {
3498 	fhandle_t *fhp = NULL;
3499 	int error;
3500 
3501 	/*
3502 	 * For lock, use the new nfslock structure, otherwise just
3503 	 * a fhandle_t on the stack.
3504 	 */
3505 	if (flags & NFSLCK_OPEN) {
3506 		KASSERT(new_lfp != NULL, ("nfsrv_getlockfh: new_lfp NULL"));
3507 		fhp = &new_lfp->lf_fh;
3508 	} else if (nfhp) {
3509 		fhp = nfhp;
3510 	} else {
3511 		panic("nfsrv_getlockfh");
3512 	}
3513 	error = nfsvno_getfh(vp, fhp, p);
3514 	NFSEXITCODE(error);
3515 	return (error);
3516 }
3517 
3518 /*
3519  * Get an nfs lock structure. Allocate one, as required, and return a
3520  * pointer to it.
3521  * Returns an NFSERR_xxx upon failure or -1 to indicate no current lock.
3522  */
3523 static int
3524 nfsrv_getlockfile(u_short flags, struct nfslockfile **new_lfpp,
3525     struct nfslockfile **lfpp, fhandle_t *nfhp, int lockit)
3526 {
3527 	struct nfslockfile *lfp;
3528 	fhandle_t *fhp = NULL, *tfhp;
3529 	struct nfslockhashhead *hp;
3530 	struct nfslockfile *new_lfp = NULL;
3531 
3532 	/*
3533 	 * For lock, use the new nfslock structure, otherwise just
3534 	 * a fhandle_t on the stack.
3535 	 */
3536 	if (flags & NFSLCK_OPEN) {
3537 		new_lfp = *new_lfpp;
3538 		fhp = &new_lfp->lf_fh;
3539 	} else if (nfhp) {
3540 		fhp = nfhp;
3541 	} else {
3542 		panic("nfsrv_getlockfile");
3543 	}
3544 
3545 	hp = NFSLOCKHASH(fhp);
3546 	LIST_FOREACH(lfp, hp, lf_hash) {
3547 		tfhp = &lfp->lf_fh;
3548 		if (NFSVNO_CMPFH(fhp, tfhp)) {
3549 			if (lockit)
3550 				nfsrv_locklf(lfp);
3551 			*lfpp = lfp;
3552 			return (0);
3553 		}
3554 	}
3555 	if (!(flags & NFSLCK_OPEN))
3556 		return (-1);
3557 
3558 	/*
3559 	 * No match, so chain the new one into the list.
3560 	 */
3561 	LIST_INIT(&new_lfp->lf_open);
3562 	LIST_INIT(&new_lfp->lf_lock);
3563 	LIST_INIT(&new_lfp->lf_deleg);
3564 	LIST_INIT(&new_lfp->lf_locallock);
3565 	LIST_INIT(&new_lfp->lf_rollback);
3566 	new_lfp->lf_locallock_lck.nfslock_usecnt = 0;
3567 	new_lfp->lf_locallock_lck.nfslock_lock = 0;
3568 	new_lfp->lf_usecount = 0;
3569 	LIST_INSERT_HEAD(hp, new_lfp, lf_hash);
3570 	*lfpp = new_lfp;
3571 	*new_lfpp = NULL;
3572 	return (0);
3573 }
3574 
3575 /*
3576  * This function adds a nfslock lock structure to the list for the associated
3577  * nfsstate and nfslockfile structures. It will be inserted after the
3578  * entry pointed at by insert_lop.
3579  */
3580 static void
3581 nfsrv_insertlock(struct nfslock *new_lop, struct nfslock *insert_lop,
3582     struct nfsstate *stp, struct nfslockfile *lfp)
3583 {
3584 	struct nfslock *lop, *nlop;
3585 
3586 	new_lop->lo_stp = stp;
3587 	new_lop->lo_lfp = lfp;
3588 
3589 	if (stp != NULL) {
3590 		/* Insert in increasing lo_first order */
3591 		lop = LIST_FIRST(&lfp->lf_lock);
3592 		if (lop == LIST_END(&lfp->lf_lock) ||
3593 		    new_lop->lo_first <= lop->lo_first) {
3594 			LIST_INSERT_HEAD(&lfp->lf_lock, new_lop, lo_lckfile);
3595 		} else {
3596 			nlop = LIST_NEXT(lop, lo_lckfile);
3597 			while (nlop != LIST_END(&lfp->lf_lock) &&
3598 			       nlop->lo_first < new_lop->lo_first) {
3599 				lop = nlop;
3600 				nlop = LIST_NEXT(lop, lo_lckfile);
3601 			}
3602 			LIST_INSERT_AFTER(lop, new_lop, lo_lckfile);
3603 		}
3604 	} else {
3605 		new_lop->lo_lckfile.le_prev = NULL;	/* list not used */
3606 	}
3607 
3608 	/*
3609 	 * Insert after insert_lop, which is overloaded as stp or lfp for
3610 	 * an empty list.
3611 	 */
3612 	if (stp == NULL && (struct nfslockfile *)insert_lop == lfp)
3613 		LIST_INSERT_HEAD(&lfp->lf_locallock, new_lop, lo_lckowner);
3614 	else if ((struct nfsstate *)insert_lop == stp)
3615 		LIST_INSERT_HEAD(&stp->ls_lock, new_lop, lo_lckowner);
3616 	else
3617 		LIST_INSERT_AFTER(insert_lop, new_lop, lo_lckowner);
3618 	if (stp != NULL) {
3619 		newnfsstats.srvlocks++;
3620 		nfsrv_openpluslock++;
3621 	}
3622 }
3623 
3624 /*
3625  * This function updates the locking for a lock owner and given file. It
3626  * maintains a list of lock ranges ordered on increasing file offset that
3627  * are NFSLCK_READ or NFSLCK_WRITE and non-overlapping (aka POSIX style).
3628  * It always adds new_lop to the list and sometimes uses the one pointed
3629  * at by other_lopp.
3630  */
3631 static void
3632 nfsrv_updatelock(struct nfsstate *stp, struct nfslock **new_lopp,
3633     struct nfslock **other_lopp, struct nfslockfile *lfp)
3634 {
3635 	struct nfslock *new_lop = *new_lopp;
3636 	struct nfslock *lop, *tlop, *ilop;
3637 	struct nfslock *other_lop = *other_lopp;
3638 	int unlock = 0, myfile = 0;
3639 	u_int64_t tmp;
3640 
3641 	/*
3642 	 * Work down the list until the lock is merged.
3643 	 */
3644 	if (new_lop->lo_flags & NFSLCK_UNLOCK)
3645 		unlock = 1;
3646 	if (stp != NULL) {
3647 		ilop = (struct nfslock *)stp;
3648 		lop = LIST_FIRST(&stp->ls_lock);
3649 	} else {
3650 		ilop = (struct nfslock *)lfp;
3651 		lop = LIST_FIRST(&lfp->lf_locallock);
3652 	}
3653 	while (lop != NULL) {
3654 	    /*
3655 	     * Only check locks for this file that aren't before the start of
3656 	     * new lock's range.
3657 	     */
3658 	    if (lop->lo_lfp == lfp) {
3659 	      myfile = 1;
3660 	      if (lop->lo_end >= new_lop->lo_first) {
3661 		if (new_lop->lo_end < lop->lo_first) {
3662 			/*
3663 			 * If the new lock ends before the start of the
3664 			 * current lock's range, no merge, just insert
3665 			 * the new lock.
3666 			 */
3667 			break;
3668 		}
3669 		if (new_lop->lo_flags == lop->lo_flags ||
3670 		    (new_lop->lo_first <= lop->lo_first &&
3671 		     new_lop->lo_end >= lop->lo_end)) {
3672 			/*
3673 			 * This lock can be absorbed by the new lock/unlock.
3674 			 * This happens when it covers the entire range
3675 			 * of the old lock or is contiguous
3676 			 * with the old lock and is of the same type or an
3677 			 * unlock.
3678 			 */
3679 			if (lop->lo_first < new_lop->lo_first)
3680 				new_lop->lo_first = lop->lo_first;
3681 			if (lop->lo_end > new_lop->lo_end)
3682 				new_lop->lo_end = lop->lo_end;
3683 			tlop = lop;
3684 			lop = LIST_NEXT(lop, lo_lckowner);
3685 			nfsrv_freenfslock(tlop);
3686 			continue;
3687 		}
3688 
3689 		/*
3690 		 * All these cases are for contiguous locks that are not the
3691 		 * same type, so they can't be merged.
3692 		 */
3693 		if (new_lop->lo_first <= lop->lo_first) {
3694 			/*
3695 			 * This case is where the new lock overlaps with the
3696 			 * first part of the old lock. Move the start of the
3697 			 * old lock to just past the end of the new lock. The
3698 			 * new lock will be inserted in front of the old, since
3699 			 * ilop hasn't been updated. (We are done now.)
3700 			 */
3701 			lop->lo_first = new_lop->lo_end;
3702 			break;
3703 		}
3704 		if (new_lop->lo_end >= lop->lo_end) {
3705 			/*
3706 			 * This case is where the new lock overlaps with the
3707 			 * end of the old lock's range. Move the old lock's
3708 			 * end to just before the new lock's first and insert
3709 			 * the new lock after the old lock.
3710 			 * Might not be done yet, since the new lock could
3711 			 * overlap further locks with higher ranges.
3712 			 */
3713 			lop->lo_end = new_lop->lo_first;
3714 			ilop = lop;
3715 			lop = LIST_NEXT(lop, lo_lckowner);
3716 			continue;
3717 		}
3718 		/*
3719 		 * The final case is where the new lock's range is in the
3720 		 * middle of the current lock's and splits the current lock
3721 		 * up. Use *other_lopp to handle the second part of the
3722 		 * split old lock range. (We are done now.)
3723 		 * For unlock, we use new_lop as other_lop and tmp, since
3724 		 * other_lop and new_lop are the same for this case.
3725 		 * We noted the unlock case above, so we don't need
3726 		 * new_lop->lo_flags any longer.
3727 		 */
3728 		tmp = new_lop->lo_first;
3729 		if (other_lop == NULL) {
3730 			if (!unlock)
3731 				panic("nfsd srv update unlock");
3732 			other_lop = new_lop;
3733 			*new_lopp = NULL;
3734 		}
3735 		other_lop->lo_first = new_lop->lo_end;
3736 		other_lop->lo_end = lop->lo_end;
3737 		other_lop->lo_flags = lop->lo_flags;
3738 		other_lop->lo_stp = stp;
3739 		other_lop->lo_lfp = lfp;
3740 		lop->lo_end = tmp;
3741 		nfsrv_insertlock(other_lop, lop, stp, lfp);
3742 		*other_lopp = NULL;
3743 		ilop = lop;
3744 		break;
3745 	      }
3746 	    }
3747 	    ilop = lop;
3748 	    lop = LIST_NEXT(lop, lo_lckowner);
3749 	    if (myfile && (lop == NULL || lop->lo_lfp != lfp))
3750 		break;
3751 	}
3752 
3753 	/*
3754 	 * Insert the new lock in the list at the appropriate place.
3755 	 */
3756 	if (!unlock) {
3757 		nfsrv_insertlock(new_lop, ilop, stp, lfp);
3758 		*new_lopp = NULL;
3759 	}
3760 }
3761 
3762 /*
3763  * This function handles sequencing of locks, etc.
3764  * It returns an error that indicates what the caller should do.
3765  */
3766 static int
3767 nfsrv_checkseqid(struct nfsrv_descript *nd, u_int32_t seqid,
3768     struct nfsstate *stp, struct nfsrvcache *op)
3769 {
3770 	int error = 0;
3771 
3772 	if ((nd->nd_flag & ND_NFSV41) != 0)
3773 		/* NFSv4.1 ignores the open_seqid and lock_seqid. */
3774 		goto out;
3775 	if (op != nd->nd_rp)
3776 		panic("nfsrvstate checkseqid");
3777 	if (!(op->rc_flag & RC_INPROG))
3778 		panic("nfsrvstate not inprog");
3779 	if (stp->ls_op && stp->ls_op->rc_refcnt <= 0) {
3780 		printf("refcnt=%d\n", stp->ls_op->rc_refcnt);
3781 		panic("nfsrvstate op refcnt");
3782 	}
3783 	if ((stp->ls_seq + 1) == seqid) {
3784 		if (stp->ls_op)
3785 			nfsrvd_derefcache(stp->ls_op);
3786 		stp->ls_op = op;
3787 		nfsrvd_refcache(op);
3788 		stp->ls_seq = seqid;
3789 		goto out;
3790 	} else if (stp->ls_seq == seqid && stp->ls_op &&
3791 		op->rc_xid == stp->ls_op->rc_xid &&
3792 		op->rc_refcnt == 0 &&
3793 		op->rc_reqlen == stp->ls_op->rc_reqlen &&
3794 		op->rc_cksum == stp->ls_op->rc_cksum) {
3795 		if (stp->ls_op->rc_flag & RC_INPROG) {
3796 			error = NFSERR_DONTREPLY;
3797 			goto out;
3798 		}
3799 		nd->nd_rp = stp->ls_op;
3800 		nd->nd_rp->rc_flag |= RC_INPROG;
3801 		nfsrvd_delcache(op);
3802 		error = NFSERR_REPLYFROMCACHE;
3803 		goto out;
3804 	}
3805 	error = NFSERR_BADSEQID;
3806 
3807 out:
3808 	NFSEXITCODE2(error, nd);
3809 	return (error);
3810 }
3811 
3812 /*
3813  * Get the client ip address for callbacks. If the strings can't be parsed,
3814  * just set lc_program to 0 to indicate no callbacks are possible.
3815  * (For cases where the address can't be parsed or is 0.0.0.0.0.0, set
3816  *  the address to the client's transport address. This won't be used
3817  *  for callbacks, but can be printed out by newnfsstats for info.)
3818  * Return error if the xdr can't be parsed, 0 otherwise.
3819  */
3820 APPLESTATIC int
3821 nfsrv_getclientipaddr(struct nfsrv_descript *nd, struct nfsclient *clp)
3822 {
3823 	u_int32_t *tl;
3824 	u_char *cp, *cp2;
3825 	int i, j;
3826 	struct sockaddr_in *rad, *sad;
3827 	u_char protocol[5], addr[24];
3828 	int error = 0, cantparse = 0;
3829 	union {
3830 		u_long ival;
3831 		u_char cval[4];
3832 	} ip;
3833 	union {
3834 		u_short sval;
3835 		u_char cval[2];
3836 	} port;
3837 
3838 	rad = NFSSOCKADDR(clp->lc_req.nr_nam, struct sockaddr_in *);
3839 	rad->sin_family = AF_INET;
3840 	rad->sin_len = sizeof (struct sockaddr_in);
3841 	rad->sin_addr.s_addr = 0;
3842 	rad->sin_port = 0;
3843 	clp->lc_req.nr_client = NULL;
3844 	clp->lc_req.nr_lock = 0;
3845 	NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
3846 	i = fxdr_unsigned(int, *tl);
3847 	if (i >= 3 && i <= 4) {
3848 		error = nfsrv_mtostr(nd, protocol, i);
3849 		if (error)
3850 			goto nfsmout;
3851 		if (!strcmp(protocol, "tcp")) {
3852 			clp->lc_flags |= LCL_TCPCALLBACK;
3853 			clp->lc_req.nr_sotype = SOCK_STREAM;
3854 			clp->lc_req.nr_soproto = IPPROTO_TCP;
3855 		} else if (!strcmp(protocol, "udp")) {
3856 			clp->lc_req.nr_sotype = SOCK_DGRAM;
3857 			clp->lc_req.nr_soproto = IPPROTO_UDP;
3858 		} else {
3859 			cantparse = 1;
3860 		}
3861 	} else {
3862 		cantparse = 1;
3863 		if (i > 0) {
3864 			error = nfsm_advance(nd, NFSM_RNDUP(i), -1);
3865 			if (error)
3866 				goto nfsmout;
3867 		}
3868 	}
3869 	NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
3870 	i = fxdr_unsigned(int, *tl);
3871 	if (i < 0) {
3872 		error = NFSERR_BADXDR;
3873 		goto nfsmout;
3874 	} else if (i == 0) {
3875 		cantparse = 1;
3876 	} else if (!cantparse && i <= 23 && i >= 11) {
3877 		error = nfsrv_mtostr(nd, addr, i);
3878 		if (error)
3879 			goto nfsmout;
3880 
3881 		/*
3882 		 * Parse out the address fields. We expect 6 decimal numbers
3883 		 * separated by '.'s.
3884 		 */
3885 		cp = addr;
3886 		i = 0;
3887 		while (*cp && i < 6) {
3888 			cp2 = cp;
3889 			while (*cp2 && *cp2 != '.')
3890 				cp2++;
3891 			if (*cp2)
3892 				*cp2++ = '\0';
3893 			else if (i != 5) {
3894 				cantparse = 1;
3895 				break;
3896 			}
3897 			j = nfsrv_getipnumber(cp);
3898 			if (j >= 0) {
3899 				if (i < 4)
3900 					ip.cval[3 - i] = j;
3901 				else
3902 					port.cval[5 - i] = j;
3903 			} else {
3904 				cantparse = 1;
3905 				break;
3906 			}
3907 			cp = cp2;
3908 			i++;
3909 		}
3910 		if (!cantparse) {
3911 			if (ip.ival != 0x0) {
3912 				rad->sin_addr.s_addr = htonl(ip.ival);
3913 				rad->sin_port = htons(port.sval);
3914 			} else {
3915 				cantparse = 1;
3916 			}
3917 		}
3918 	} else {
3919 		cantparse = 1;
3920 		if (i > 0) {
3921 			error = nfsm_advance(nd, NFSM_RNDUP(i), -1);
3922 			if (error)
3923 				goto nfsmout;
3924 		}
3925 	}
3926 	if (cantparse) {
3927 		sad = NFSSOCKADDR(nd->nd_nam, struct sockaddr_in *);
3928 		rad->sin_addr.s_addr = sad->sin_addr.s_addr;
3929 		rad->sin_port = 0x0;
3930 		clp->lc_program = 0;
3931 	}
3932 nfsmout:
3933 	NFSEXITCODE2(error, nd);
3934 	return (error);
3935 }
3936 
3937 /*
3938  * Turn a string of up to three decimal digits into a number. Return -1 upon
3939  * error.
3940  */
3941 static int
3942 nfsrv_getipnumber(u_char *cp)
3943 {
3944 	int i = 0, j = 0;
3945 
3946 	while (*cp) {
3947 		if (j > 2 || *cp < '0' || *cp > '9')
3948 			return (-1);
3949 		i *= 10;
3950 		i += (*cp - '0');
3951 		cp++;
3952 		j++;
3953 	}
3954 	if (i < 256)
3955 		return (i);
3956 	return (-1);
3957 }
3958 
3959 /*
3960  * This function checks for restart conditions.
3961  */
3962 static int
3963 nfsrv_checkrestart(nfsquad_t clientid, u_int32_t flags,
3964     nfsv4stateid_t *stateidp, int specialid)
3965 {
3966 	int ret = 0;
3967 
3968 	/*
3969 	 * First check for a server restart. Open, LockT, ReleaseLockOwner
3970 	 * and DelegPurge have a clientid, the rest a stateid.
3971 	 */
3972 	if (flags &
3973 	    (NFSLCK_OPEN | NFSLCK_TEST | NFSLCK_RELEASE | NFSLCK_DELEGPURGE)) {
3974 		if (clientid.lval[0] != nfsrvboottime) {
3975 			ret = NFSERR_STALECLIENTID;
3976 			goto out;
3977 		}
3978 	} else if (stateidp->other[0] != nfsrvboottime &&
3979 		specialid == 0) {
3980 		ret = NFSERR_STALESTATEID;
3981 		goto out;
3982 	}
3983 
3984 	/*
3985 	 * Read, Write, Setattr and LockT can return NFSERR_GRACE and do
3986 	 * not use a lock/open owner seqid#, so the check can be done now.
3987 	 * (The others will be checked, as required, later.)
3988 	 */
3989 	if (!(flags & (NFSLCK_CHECK | NFSLCK_TEST)))
3990 		goto out;
3991 
3992 	NFSLOCKSTATE();
3993 	ret = nfsrv_checkgrace(NULL, NULL, flags);
3994 	NFSUNLOCKSTATE();
3995 
3996 out:
3997 	NFSEXITCODE(ret);
3998 	return (ret);
3999 }
4000 
4001 /*
4002  * Check for grace.
4003  */
4004 static int
4005 nfsrv_checkgrace(struct nfsrv_descript *nd, struct nfsclient *clp,
4006     u_int32_t flags)
4007 {
4008 	int error = 0;
4009 
4010 	if ((nfsrv_stablefirst.nsf_flags & NFSNSF_GRACEOVER) != 0) {
4011 		if (flags & NFSLCK_RECLAIM) {
4012 			error = NFSERR_NOGRACE;
4013 			goto out;
4014 		}
4015 	} else {
4016 		if (!(flags & NFSLCK_RECLAIM)) {
4017 			error = NFSERR_GRACE;
4018 			goto out;
4019 		}
4020 		if (nd != NULL && clp != NULL &&
4021 		    (nd->nd_flag & ND_NFSV41) != 0 &&
4022 		    (clp->lc_flags & LCL_RECLAIMCOMPLETE) != 0) {
4023 			error = NFSERR_NOGRACE;
4024 			goto out;
4025 		}
4026 
4027 		/*
4028 		 * If grace is almost over and we are still getting Reclaims,
4029 		 * extend grace a bit.
4030 		 */
4031 		if ((NFSD_MONOSEC + NFSRV_LEASEDELTA) >
4032 		    nfsrv_stablefirst.nsf_eograce)
4033 			nfsrv_stablefirst.nsf_eograce = NFSD_MONOSEC +
4034 				NFSRV_LEASEDELTA;
4035 	}
4036 
4037 out:
4038 	NFSEXITCODE(error);
4039 	return (error);
4040 }
4041 
4042 /*
4043  * Do a server callback.
4044  */
4045 static int
4046 nfsrv_docallback(struct nfsclient *clp, int procnum,
4047     nfsv4stateid_t *stateidp, int trunc, fhandle_t *fhp,
4048     struct nfsvattr *nap, nfsattrbit_t *attrbitp, NFSPROC_T *p)
4049 {
4050 	mbuf_t m;
4051 	u_int32_t *tl;
4052 	struct nfsrv_descript nfsd, *nd = &nfsd;
4053 	struct ucred *cred;
4054 	int error = 0;
4055 	u_int32_t callback;
4056 	struct nfsdsession *sep = NULL;
4057 
4058 	cred = newnfs_getcred();
4059 	NFSLOCKSTATE();	/* mostly for lc_cbref++ */
4060 	if (clp->lc_flags & LCL_NEEDSCONFIRM) {
4061 		NFSUNLOCKSTATE();
4062 		panic("docallb");
4063 	}
4064 	clp->lc_cbref++;
4065 
4066 	/*
4067 	 * Fill the callback program# and version into the request
4068 	 * structure for newnfs_connect() to use.
4069 	 */
4070 	clp->lc_req.nr_prog = clp->lc_program;
4071 #ifdef notnow
4072 	if ((clp->lc_flags & LCL_NFSV41) != 0)
4073 		clp->lc_req.nr_vers = NFSV41_CBVERS;
4074 	else
4075 #endif
4076 		clp->lc_req.nr_vers = NFSV4_CBVERS;
4077 
4078 	/*
4079 	 * First, fill in some of the fields of nd and cr.
4080 	 */
4081 	nd->nd_flag = ND_NFSV4;
4082 	if (clp->lc_flags & LCL_GSS)
4083 		nd->nd_flag |= ND_KERBV;
4084 	if ((clp->lc_flags & LCL_NFSV41) != 0)
4085 		nd->nd_flag |= ND_NFSV41;
4086 	nd->nd_repstat = 0;
4087 	cred->cr_uid = clp->lc_uid;
4088 	cred->cr_gid = clp->lc_gid;
4089 	callback = clp->lc_callback;
4090 	NFSUNLOCKSTATE();
4091 	cred->cr_ngroups = 1;
4092 
4093 	/*
4094 	 * Get the first mbuf for the request.
4095 	 */
4096 	MGET(m, M_WAITOK, MT_DATA);
4097 	mbuf_setlen(m, 0);
4098 	nd->nd_mreq = nd->nd_mb = m;
4099 	nd->nd_bpos = NFSMTOD(m, caddr_t);
4100 
4101 	/*
4102 	 * and build the callback request.
4103 	 */
4104 	if (procnum == NFSV4OP_CBGETATTR) {
4105 		nd->nd_procnum = NFSV4PROC_CBCOMPOUND;
4106 		error = nfsrv_cbcallargs(nd, clp, callback, NFSV4OP_CBGETATTR,
4107 		    "CB Getattr", &sep);
4108 		if (error != 0) {
4109 			mbuf_freem(nd->nd_mreq);
4110 			goto errout;
4111 		}
4112 		(void)nfsm_fhtom(nd, (u_int8_t *)fhp, NFSX_MYFH, 0);
4113 		(void)nfsrv_putattrbit(nd, attrbitp);
4114 	} else if (procnum == NFSV4OP_CBRECALL) {
4115 		nd->nd_procnum = NFSV4PROC_CBCOMPOUND;
4116 		error = nfsrv_cbcallargs(nd, clp, callback, NFSV4OP_CBRECALL,
4117 		    "CB Recall", &sep);
4118 		if (error != 0) {
4119 			mbuf_freem(nd->nd_mreq);
4120 			goto errout;
4121 		}
4122 		NFSM_BUILD(tl, u_int32_t *, NFSX_UNSIGNED + NFSX_STATEID);
4123 		*tl++ = txdr_unsigned(stateidp->seqid);
4124 		NFSBCOPY((caddr_t)stateidp->other, (caddr_t)tl,
4125 		    NFSX_STATEIDOTHER);
4126 		tl += (NFSX_STATEIDOTHER / NFSX_UNSIGNED);
4127 		if (trunc)
4128 			*tl = newnfs_true;
4129 		else
4130 			*tl = newnfs_false;
4131 		(void)nfsm_fhtom(nd, (u_int8_t *)fhp, NFSX_MYFH, 0);
4132 	} else if (procnum == NFSV4PROC_CBNULL) {
4133 		nd->nd_procnum = NFSV4PROC_CBNULL;
4134 		if ((clp->lc_flags & LCL_NFSV41) != 0) {
4135 			error = nfsv4_getcbsession(clp, &sep);
4136 			if (error != 0) {
4137 				mbuf_freem(nd->nd_mreq);
4138 				goto errout;
4139 			}
4140 		}
4141 	} else {
4142 		error = NFSERR_SERVERFAULT;
4143 		mbuf_freem(nd->nd_mreq);
4144 		goto errout;
4145 	}
4146 
4147 	/*
4148 	 * Call newnfs_connect(), as required, and then newnfs_request().
4149 	 */
4150 	(void) newnfs_sndlock(&clp->lc_req.nr_lock);
4151 	if (clp->lc_req.nr_client == NULL) {
4152 		if ((clp->lc_flags & LCL_NFSV41) != 0)
4153 			error = ECONNREFUSED;
4154 		else if (nd->nd_procnum == NFSV4PROC_CBNULL)
4155 			error = newnfs_connect(NULL, &clp->lc_req, cred,
4156 			    NULL, 1);
4157 		else
4158 			error = newnfs_connect(NULL, &clp->lc_req, cred,
4159 			    NULL, 3);
4160 	}
4161 	newnfs_sndunlock(&clp->lc_req.nr_lock);
4162 	if (!error) {
4163 		if ((nd->nd_flag & ND_NFSV41) != 0) {
4164 			KASSERT(sep != NULL, ("sep NULL"));
4165 			error = newnfs_request(nd, NULL, clp, &clp->lc_req,
4166 			    NULL, NULL, cred, clp->lc_program,
4167 			    clp->lc_req.nr_vers, NULL, 1, NULL,
4168 			    &sep->sess_cbsess);
4169 			nfsrv_freesession(sep, NULL);
4170 		} else
4171 			error = newnfs_request(nd, NULL, clp, &clp->lc_req,
4172 			    NULL, NULL, cred, clp->lc_program,
4173 			    clp->lc_req.nr_vers, NULL, 1, NULL, NULL);
4174 	}
4175 errout:
4176 	NFSFREECRED(cred);
4177 
4178 	/*
4179 	 * If error is set here, the Callback path isn't working
4180 	 * properly, so twiddle the appropriate LCL_ flags.
4181 	 * (nd_repstat != 0 indicates the Callback path is working,
4182 	 *  but the callback failed on the client.)
4183 	 */
4184 	if (error) {
4185 		/*
4186 		 * Mark the callback pathway down, which disabled issuing
4187 		 * of delegations and gets Renew to return NFSERR_CBPATHDOWN.
4188 		 */
4189 		NFSLOCKSTATE();
4190 		clp->lc_flags |= LCL_CBDOWN;
4191 		NFSUNLOCKSTATE();
4192 	} else {
4193 		/*
4194 		 * Callback worked. If the callback path was down, disable
4195 		 * callbacks, so no more delegations will be issued. (This
4196 		 * is done on the assumption that the callback pathway is
4197 		 * flakey.)
4198 		 */
4199 		NFSLOCKSTATE();
4200 		if (clp->lc_flags & LCL_CBDOWN)
4201 			clp->lc_flags &= ~(LCL_CBDOWN | LCL_CALLBACKSON);
4202 		NFSUNLOCKSTATE();
4203 		if (nd->nd_repstat)
4204 			error = nd->nd_repstat;
4205 		else if (error == 0 && procnum == NFSV4OP_CBGETATTR)
4206 			error = nfsv4_loadattr(nd, NULL, nap, NULL, NULL, 0,
4207 			    NULL, NULL, NULL, NULL, NULL, 0, NULL, NULL, NULL,
4208 			    p, NULL);
4209 		mbuf_freem(nd->nd_mrep);
4210 	}
4211 	NFSLOCKSTATE();
4212 	clp->lc_cbref--;
4213 	if ((clp->lc_flags & LCL_WAKEUPWANTED) && clp->lc_cbref == 0) {
4214 		clp->lc_flags &= ~LCL_WAKEUPWANTED;
4215 		wakeup(clp);
4216 	}
4217 	NFSUNLOCKSTATE();
4218 
4219 	NFSEXITCODE(error);
4220 	return (error);
4221 }
4222 
4223 /*
4224  * Set up the compound RPC for the callback.
4225  */
4226 static int
4227 nfsrv_cbcallargs(struct nfsrv_descript *nd, struct nfsclient *clp,
4228     uint32_t callback, int op, const char *optag, struct nfsdsession **sepp)
4229 {
4230 	uint32_t *tl;
4231 	int error, len;
4232 
4233 	len = strlen(optag);
4234 	(void)nfsm_strtom(nd, optag, len);
4235 	NFSM_BUILD(tl, uint32_t *, 4 * NFSX_UNSIGNED);
4236 	if ((nd->nd_flag & ND_NFSV41) != 0) {
4237 		*tl++ = txdr_unsigned(NFSV41_MINORVERSION);
4238 		*tl++ = txdr_unsigned(callback);
4239 		*tl++ = txdr_unsigned(2);
4240 		*tl = txdr_unsigned(NFSV4OP_CBSEQUENCE);
4241 		error = nfsv4_setcbsequence(nd, clp, 1, sepp);
4242 		if (error != 0)
4243 			return (error);
4244 		NFSM_BUILD(tl, u_int32_t *, NFSX_UNSIGNED);
4245 		*tl = txdr_unsigned(op);
4246 	} else {
4247 		*tl++ = txdr_unsigned(NFSV4_MINORVERSION);
4248 		*tl++ = txdr_unsigned(callback);
4249 		*tl++ = txdr_unsigned(1);
4250 		*tl = txdr_unsigned(op);
4251 	}
4252 	return (0);
4253 }
4254 
4255 /*
4256  * Return the next index# for a clientid. Mostly just increment and return
4257  * the next one, but... if the 32bit unsigned does actually wrap around,
4258  * it should be rebooted.
4259  * At an average rate of one new client per second, it will wrap around in
4260  * approximately 136 years. (I think the server will have been shut
4261  * down or rebooted before then.)
4262  */
4263 static u_int32_t
4264 nfsrv_nextclientindex(void)
4265 {
4266 	static u_int32_t client_index = 0;
4267 
4268 	client_index++;
4269 	if (client_index != 0)
4270 		return (client_index);
4271 
4272 	printf("%s: out of clientids\n", __func__);
4273 	return (client_index);
4274 }
4275 
4276 /*
4277  * Return the next index# for a stateid. Mostly just increment and return
4278  * the next one, but... if the 32bit unsigned does actually wrap around
4279  * (will a BSD server stay up that long?), find
4280  * new start and end values.
4281  */
4282 static u_int32_t
4283 nfsrv_nextstateindex(struct nfsclient *clp)
4284 {
4285 	struct nfsstate *stp;
4286 	int i;
4287 	u_int32_t canuse, min_index, max_index;
4288 
4289 	if (!(clp->lc_flags & LCL_INDEXNOTOK)) {
4290 		clp->lc_stateindex++;
4291 		if (clp->lc_stateindex != clp->lc_statemaxindex)
4292 			return (clp->lc_stateindex);
4293 	}
4294 
4295 	/*
4296 	 * Yuck, we've hit the end.
4297 	 * Look for a new min and max.
4298 	 */
4299 	min_index = 0;
4300 	max_index = 0xffffffff;
4301 	for (i = 0; i < NFSSTATEHASHSIZE; i++) {
4302 	    LIST_FOREACH(stp, &clp->lc_stateid[i], ls_hash) {
4303 		if (stp->ls_stateid.other[2] > 0x80000000) {
4304 		    if (stp->ls_stateid.other[2] < max_index)
4305 			max_index = stp->ls_stateid.other[2];
4306 		} else {
4307 		    if (stp->ls_stateid.other[2] > min_index)
4308 			min_index = stp->ls_stateid.other[2];
4309 		}
4310 	    }
4311 	}
4312 
4313 	/*
4314 	 * Yikes, highly unlikely, but I'll handle it anyhow.
4315 	 */
4316 	if (min_index == 0x80000000 && max_index == 0x80000001) {
4317 	    canuse = 0;
4318 	    /*
4319 	     * Loop around until we find an unused entry. Return that
4320 	     * and set LCL_INDEXNOTOK, so the search will continue next time.
4321 	     * (This is one of those rare cases where a goto is the
4322 	     *  cleanest way to code the loop.)
4323 	     */
4324 tryagain:
4325 	    for (i = 0; i < NFSSTATEHASHSIZE; i++) {
4326 		LIST_FOREACH(stp, &clp->lc_stateid[i], ls_hash) {
4327 		    if (stp->ls_stateid.other[2] == canuse) {
4328 			canuse++;
4329 			goto tryagain;
4330 		    }
4331 		}
4332 	    }
4333 	    clp->lc_flags |= LCL_INDEXNOTOK;
4334 	    return (canuse);
4335 	}
4336 
4337 	/*
4338 	 * Ok to start again from min + 1.
4339 	 */
4340 	clp->lc_stateindex = min_index + 1;
4341 	clp->lc_statemaxindex = max_index;
4342 	clp->lc_flags &= ~LCL_INDEXNOTOK;
4343 	return (clp->lc_stateindex);
4344 }
4345 
4346 /*
4347  * The following functions handle the stable storage file that deals with
4348  * the edge conditions described in RFC3530 Sec. 8.6.3.
4349  * The file is as follows:
4350  * - a single record at the beginning that has the lease time of the
4351  *   previous server instance (before the last reboot) and the nfsrvboottime
4352  *   values for the previous server boots.
4353  *   These previous boot times are used to ensure that the current
4354  *   nfsrvboottime does not, somehow, get set to a previous one.
4355  *   (This is important so that Stale ClientIDs and StateIDs can
4356  *    be recognized.)
4357  *   The number of previous nfsvrboottime values preceeds the list.
4358  * - followed by some number of appended records with:
4359  *   - client id string
4360  *   - flag that indicates it is a record revoking state via lease
4361  *     expiration or similar
4362  *     OR has successfully acquired state.
4363  * These structures vary in length, with the client string at the end, up
4364  * to NFSV4_OPAQUELIMIT in size.
4365  *
4366  * At the end of the grace period, the file is truncated, the first
4367  * record is rewritten with updated information and any acquired state
4368  * records for successful reclaims of state are written.
4369  *
4370  * Subsequent records are appended when the first state is issued to
4371  * a client and when state is revoked for a client.
4372  *
4373  * When reading the file in, state issued records that come later in
4374  * the file override older ones, since the append log is in cronological order.
4375  * If, for some reason, the file can't be read, the grace period is
4376  * immediately terminated and all reclaims get NFSERR_NOGRACE.
4377  */
4378 
4379 /*
4380  * Read in the stable storage file. Called by nfssvc() before the nfsd
4381  * processes start servicing requests.
4382  */
4383 APPLESTATIC void
4384 nfsrv_setupstable(NFSPROC_T *p)
4385 {
4386 	struct nfsrv_stablefirst *sf = &nfsrv_stablefirst;
4387 	struct nfsrv_stable *sp, *nsp;
4388 	struct nfst_rec *tsp;
4389 	int error, i, tryagain;
4390 	off_t off = 0;
4391 	ssize_t aresid, len;
4392 
4393 	/*
4394 	 * If NFSNSF_UPDATEDONE is set, this is a restart of the nfsds without
4395 	 * a reboot, so state has not been lost.
4396 	 */
4397 	if (sf->nsf_flags & NFSNSF_UPDATEDONE)
4398 		return;
4399 	/*
4400 	 * Set Grace over just until the file reads successfully.
4401 	 */
4402 	nfsrvboottime = time_second;
4403 	LIST_INIT(&sf->nsf_head);
4404 	sf->nsf_flags = (NFSNSF_GRACEOVER | NFSNSF_NEEDLOCK);
4405 	sf->nsf_eograce = NFSD_MONOSEC + NFSRV_LEASEDELTA;
4406 	if (sf->nsf_fp == NULL)
4407 		return;
4408 	error = NFSD_RDWR(UIO_READ, NFSFPVNODE(sf->nsf_fp),
4409 	    (caddr_t)&sf->nsf_rec, sizeof (struct nfsf_rec), off, UIO_SYSSPACE,
4410 	    0, NFSFPCRED(sf->nsf_fp), &aresid, p);
4411 	if (error || aresid || sf->nsf_numboots == 0 ||
4412 		sf->nsf_numboots > NFSNSF_MAXNUMBOOTS)
4413 		return;
4414 
4415 	/*
4416 	 * Now, read in the boottimes.
4417 	 */
4418 	sf->nsf_bootvals = (time_t *)malloc((sf->nsf_numboots + 1) *
4419 		sizeof (time_t), M_TEMP, M_WAITOK);
4420 	off = sizeof (struct nfsf_rec);
4421 	error = NFSD_RDWR(UIO_READ, NFSFPVNODE(sf->nsf_fp),
4422 	    (caddr_t)sf->nsf_bootvals, sf->nsf_numboots * sizeof (time_t), off,
4423 	    UIO_SYSSPACE, 0, NFSFPCRED(sf->nsf_fp), &aresid, p);
4424 	if (error || aresid) {
4425 		free((caddr_t)sf->nsf_bootvals, M_TEMP);
4426 		sf->nsf_bootvals = NULL;
4427 		return;
4428 	}
4429 
4430 	/*
4431 	 * Make sure this nfsrvboottime is different from all recorded
4432 	 * previous ones.
4433 	 */
4434 	do {
4435 		tryagain = 0;
4436 		for (i = 0; i < sf->nsf_numboots; i++) {
4437 			if (nfsrvboottime == sf->nsf_bootvals[i]) {
4438 				nfsrvboottime++;
4439 				tryagain = 1;
4440 				break;
4441 			}
4442 		}
4443 	} while (tryagain);
4444 
4445 	sf->nsf_flags |= NFSNSF_OK;
4446 	off += (sf->nsf_numboots * sizeof (time_t));
4447 
4448 	/*
4449 	 * Read through the file, building a list of records for grace
4450 	 * checking.
4451 	 * Each record is between sizeof (struct nfst_rec) and
4452 	 * sizeof (struct nfst_rec) + NFSV4_OPAQUELIMIT - 1
4453 	 * and is actually sizeof (struct nfst_rec) + nst_len - 1.
4454 	 */
4455 	tsp = (struct nfst_rec *)malloc(sizeof (struct nfst_rec) +
4456 		NFSV4_OPAQUELIMIT - 1, M_TEMP, M_WAITOK);
4457 	do {
4458 	    error = NFSD_RDWR(UIO_READ, NFSFPVNODE(sf->nsf_fp),
4459 	        (caddr_t)tsp, sizeof (struct nfst_rec) + NFSV4_OPAQUELIMIT - 1,
4460 	        off, UIO_SYSSPACE, 0, NFSFPCRED(sf->nsf_fp), &aresid, p);
4461 	    len = (sizeof (struct nfst_rec) + NFSV4_OPAQUELIMIT - 1) - aresid;
4462 	    if (error || (len > 0 && (len < sizeof (struct nfst_rec) ||
4463 		len < (sizeof (struct nfst_rec) + tsp->len - 1)))) {
4464 		/*
4465 		 * Yuck, the file has been corrupted, so just return
4466 		 * after clearing out any restart state, so the grace period
4467 		 * is over.
4468 		 */
4469 		LIST_FOREACH_SAFE(sp, &sf->nsf_head, nst_list, nsp) {
4470 			LIST_REMOVE(sp, nst_list);
4471 			free((caddr_t)sp, M_TEMP);
4472 		}
4473 		free((caddr_t)tsp, M_TEMP);
4474 		sf->nsf_flags &= ~NFSNSF_OK;
4475 		free((caddr_t)sf->nsf_bootvals, M_TEMP);
4476 		sf->nsf_bootvals = NULL;
4477 		return;
4478 	    }
4479 	    if (len > 0) {
4480 		off += sizeof (struct nfst_rec) + tsp->len - 1;
4481 		/*
4482 		 * Search the list for a matching client.
4483 		 */
4484 		LIST_FOREACH(sp, &sf->nsf_head, nst_list) {
4485 			if (tsp->len == sp->nst_len &&
4486 			    !NFSBCMP(tsp->client, sp->nst_client, tsp->len))
4487 				break;
4488 		}
4489 		if (sp == LIST_END(&sf->nsf_head)) {
4490 			sp = (struct nfsrv_stable *)malloc(tsp->len +
4491 				sizeof (struct nfsrv_stable) - 1, M_TEMP,
4492 				M_WAITOK);
4493 			NFSBCOPY((caddr_t)tsp, (caddr_t)&sp->nst_rec,
4494 				sizeof (struct nfst_rec) + tsp->len - 1);
4495 			LIST_INSERT_HEAD(&sf->nsf_head, sp, nst_list);
4496 		} else {
4497 			if (tsp->flag == NFSNST_REVOKE)
4498 				sp->nst_flag |= NFSNST_REVOKE;
4499 			else
4500 				/*
4501 				 * A subsequent timestamp indicates the client
4502 				 * did a setclientid/confirm and any previous
4503 				 * revoke is no longer relevant.
4504 				 */
4505 				sp->nst_flag &= ~NFSNST_REVOKE;
4506 		}
4507 	    }
4508 	} while (len > 0);
4509 	free((caddr_t)tsp, M_TEMP);
4510 	sf->nsf_flags = NFSNSF_OK;
4511 	sf->nsf_eograce = NFSD_MONOSEC + sf->nsf_lease +
4512 		NFSRV_LEASEDELTA;
4513 }
4514 
4515 /*
4516  * Update the stable storage file, now that the grace period is over.
4517  */
4518 APPLESTATIC void
4519 nfsrv_updatestable(NFSPROC_T *p)
4520 {
4521 	struct nfsrv_stablefirst *sf = &nfsrv_stablefirst;
4522 	struct nfsrv_stable *sp, *nsp;
4523 	int i;
4524 	struct nfsvattr nva;
4525 	vnode_t vp;
4526 #if defined(__FreeBSD_version) && (__FreeBSD_version >= 500000)
4527 	mount_t mp = NULL;
4528 #endif
4529 	int error;
4530 
4531 	if (sf->nsf_fp == NULL || (sf->nsf_flags & NFSNSF_UPDATEDONE))
4532 		return;
4533 	sf->nsf_flags |= NFSNSF_UPDATEDONE;
4534 	/*
4535 	 * Ok, we need to rewrite the stable storage file.
4536 	 * - truncate to 0 length
4537 	 * - write the new first structure
4538 	 * - loop through the data structures, writing out any that
4539 	 *   have timestamps older than the old boot
4540 	 */
4541 	if (sf->nsf_bootvals) {
4542 		sf->nsf_numboots++;
4543 		for (i = sf->nsf_numboots - 2; i >= 0; i--)
4544 			sf->nsf_bootvals[i + 1] = sf->nsf_bootvals[i];
4545 	} else {
4546 		sf->nsf_numboots = 1;
4547 		sf->nsf_bootvals = (time_t *)malloc(sizeof (time_t),
4548 			M_TEMP, M_WAITOK);
4549 	}
4550 	sf->nsf_bootvals[0] = nfsrvboottime;
4551 	sf->nsf_lease = nfsrv_lease;
4552 	NFSVNO_ATTRINIT(&nva);
4553 	NFSVNO_SETATTRVAL(&nva, size, 0);
4554 	vp = NFSFPVNODE(sf->nsf_fp);
4555 	vn_start_write(vp, &mp, V_WAIT);
4556 	if (NFSVOPLOCK(vp, LK_EXCLUSIVE) == 0) {
4557 		error = nfsvno_setattr(vp, &nva, NFSFPCRED(sf->nsf_fp), p,
4558 		    NULL);
4559 		NFSVOPUNLOCK(vp, 0);
4560 	} else
4561 		error = EPERM;
4562 	vn_finished_write(mp);
4563 	if (!error)
4564 	    error = NFSD_RDWR(UIO_WRITE, vp,
4565 		(caddr_t)&sf->nsf_rec, sizeof (struct nfsf_rec), (off_t)0,
4566 		UIO_SYSSPACE, IO_SYNC, NFSFPCRED(sf->nsf_fp), NULL, p);
4567 	if (!error)
4568 	    error = NFSD_RDWR(UIO_WRITE, vp,
4569 		(caddr_t)sf->nsf_bootvals,
4570 		sf->nsf_numboots * sizeof (time_t),
4571 		(off_t)(sizeof (struct nfsf_rec)),
4572 		UIO_SYSSPACE, IO_SYNC, NFSFPCRED(sf->nsf_fp), NULL, p);
4573 	free((caddr_t)sf->nsf_bootvals, M_TEMP);
4574 	sf->nsf_bootvals = NULL;
4575 	if (error) {
4576 		sf->nsf_flags &= ~NFSNSF_OK;
4577 		printf("EEK! Can't write NfsV4 stable storage file\n");
4578 		return;
4579 	}
4580 	sf->nsf_flags |= NFSNSF_OK;
4581 
4582 	/*
4583 	 * Loop through the list and write out timestamp records for
4584 	 * any clients that successfully reclaimed state.
4585 	 */
4586 	LIST_FOREACH_SAFE(sp, &sf->nsf_head, nst_list, nsp) {
4587 		if (sp->nst_flag & NFSNST_GOTSTATE) {
4588 			nfsrv_writestable(sp->nst_client, sp->nst_len,
4589 				NFSNST_NEWSTATE, p);
4590 			sp->nst_clp->lc_flags |= LCL_STAMPEDSTABLE;
4591 		}
4592 		LIST_REMOVE(sp, nst_list);
4593 		free((caddr_t)sp, M_TEMP);
4594 	}
4595 	nfsrv_backupstable();
4596 }
4597 
4598 /*
4599  * Append a record to the stable storage file.
4600  */
4601 APPLESTATIC void
4602 nfsrv_writestable(u_char *client, int len, int flag, NFSPROC_T *p)
4603 {
4604 	struct nfsrv_stablefirst *sf = &nfsrv_stablefirst;
4605 	struct nfst_rec *sp;
4606 	int error;
4607 
4608 	if (!(sf->nsf_flags & NFSNSF_OK) || sf->nsf_fp == NULL)
4609 		return;
4610 	sp = (struct nfst_rec *)malloc(sizeof (struct nfst_rec) +
4611 		len - 1, M_TEMP, M_WAITOK);
4612 	sp->len = len;
4613 	NFSBCOPY(client, sp->client, len);
4614 	sp->flag = flag;
4615 	error = NFSD_RDWR(UIO_WRITE, NFSFPVNODE(sf->nsf_fp),
4616 	    (caddr_t)sp, sizeof (struct nfst_rec) + len - 1, (off_t)0,
4617 	    UIO_SYSSPACE, (IO_SYNC | IO_APPEND), NFSFPCRED(sf->nsf_fp), NULL, p);
4618 	free((caddr_t)sp, M_TEMP);
4619 	if (error) {
4620 		sf->nsf_flags &= ~NFSNSF_OK;
4621 		printf("EEK! Can't write NfsV4 stable storage file\n");
4622 	}
4623 }
4624 
4625 /*
4626  * This function is called during the grace period to mark a client
4627  * that successfully reclaimed state.
4628  */
4629 static void
4630 nfsrv_markstable(struct nfsclient *clp)
4631 {
4632 	struct nfsrv_stable *sp;
4633 
4634 	/*
4635 	 * First find the client structure.
4636 	 */
4637 	LIST_FOREACH(sp, &nfsrv_stablefirst.nsf_head, nst_list) {
4638 		if (sp->nst_len == clp->lc_idlen &&
4639 		    !NFSBCMP(sp->nst_client, clp->lc_id, sp->nst_len))
4640 			break;
4641 	}
4642 	if (sp == LIST_END(&nfsrv_stablefirst.nsf_head))
4643 		return;
4644 
4645 	/*
4646 	 * Now, just mark it and set the nfsclient back pointer.
4647 	 */
4648 	sp->nst_flag |= NFSNST_GOTSTATE;
4649 	sp->nst_clp = clp;
4650 }
4651 
4652 /*
4653  * This function is called for a reclaim, to see if it gets grace.
4654  * It returns 0 if a reclaim is allowed, 1 otherwise.
4655  */
4656 static int
4657 nfsrv_checkstable(struct nfsclient *clp)
4658 {
4659 	struct nfsrv_stable *sp;
4660 
4661 	/*
4662 	 * First, find the entry for the client.
4663 	 */
4664 	LIST_FOREACH(sp, &nfsrv_stablefirst.nsf_head, nst_list) {
4665 		if (sp->nst_len == clp->lc_idlen &&
4666 		    !NFSBCMP(sp->nst_client, clp->lc_id, sp->nst_len))
4667 			break;
4668 	}
4669 
4670 	/*
4671 	 * If not in the list, state was revoked or no state was issued
4672 	 * since the previous reboot, a reclaim is denied.
4673 	 */
4674 	if (sp == LIST_END(&nfsrv_stablefirst.nsf_head) ||
4675 	    (sp->nst_flag & NFSNST_REVOKE) ||
4676 	    !(nfsrv_stablefirst.nsf_flags & NFSNSF_OK))
4677 		return (1);
4678 	return (0);
4679 }
4680 
4681 /*
4682  * Test for and try to clear out a conflicting client. This is called by
4683  * nfsrv_lockctrl() and nfsrv_openctrl() when conflicts with other clients
4684  * a found.
4685  * The trick here is that it can't revoke a conflicting client with an
4686  * expired lease unless it holds the v4root lock, so...
4687  * If no v4root lock, get the lock and return 1 to indicate "try again".
4688  * Return 0 to indicate the conflict can't be revoked and 1 to indicate
4689  * the revocation worked and the conflicting client is "bye, bye", so it
4690  * can be tried again.
4691  * Return 2 to indicate that the vnode is VI_DOOMED after NFSVOPLOCK().
4692  * Unlocks State before a non-zero value is returned.
4693  */
4694 static int
4695 nfsrv_clientconflict(struct nfsclient *clp, int *haslockp, vnode_t vp,
4696     NFSPROC_T *p)
4697 {
4698 	int gotlock, lktype = 0;
4699 
4700 	/*
4701 	 * If lease hasn't expired, we can't fix it.
4702 	 */
4703 	if (clp->lc_expiry >= NFSD_MONOSEC ||
4704 	    !(nfsrv_stablefirst.nsf_flags & NFSNSF_UPDATEDONE))
4705 		return (0);
4706 	if (*haslockp == 0) {
4707 		NFSUNLOCKSTATE();
4708 		if (vp != NULL) {
4709 			lktype = NFSVOPISLOCKED(vp);
4710 			NFSVOPUNLOCK(vp, 0);
4711 		}
4712 		NFSLOCKV4ROOTMUTEX();
4713 		nfsv4_relref(&nfsv4rootfs_lock);
4714 		do {
4715 			gotlock = nfsv4_lock(&nfsv4rootfs_lock, 1, NULL,
4716 			    NFSV4ROOTLOCKMUTEXPTR, NULL);
4717 		} while (!gotlock);
4718 		NFSUNLOCKV4ROOTMUTEX();
4719 		*haslockp = 1;
4720 		if (vp != NULL) {
4721 			NFSVOPLOCK(vp, lktype | LK_RETRY);
4722 			if ((vp->v_iflag & VI_DOOMED) != 0)
4723 				return (2);
4724 		}
4725 		return (1);
4726 	}
4727 	NFSUNLOCKSTATE();
4728 
4729 	/*
4730 	 * Ok, we can expire the conflicting client.
4731 	 */
4732 	nfsrv_writestable(clp->lc_id, clp->lc_idlen, NFSNST_REVOKE, p);
4733 	nfsrv_backupstable();
4734 	nfsrv_cleanclient(clp, p);
4735 	nfsrv_freedeleglist(&clp->lc_deleg);
4736 	nfsrv_freedeleglist(&clp->lc_olddeleg);
4737 	LIST_REMOVE(clp, lc_hash);
4738 	nfsrv_zapclient(clp, p);
4739 	return (1);
4740 }
4741 
4742 /*
4743  * Resolve a delegation conflict.
4744  * Returns 0 to indicate the conflict was resolved without sleeping.
4745  * Return -1 to indicate that the caller should check for conflicts again.
4746  * Return > 0 for an error that should be returned, normally NFSERR_DELAY.
4747  *
4748  * Also, manipulate the nfsv4root_lock, as required. It isn't changed
4749  * for a return of 0, since there was no sleep and it could be required
4750  * later. It is released for a return of NFSERR_DELAY, since the caller
4751  * will return that error. It is released when a sleep was done waiting
4752  * for the delegation to be returned or expire (so that other nfsds can
4753  * handle ops). Then, it must be acquired for the write to stable storage.
4754  * (This function is somewhat similar to nfsrv_clientconflict(), but
4755  *  the semantics differ in a couple of subtle ways. The return of 0
4756  *  indicates the conflict was resolved without sleeping here, not
4757  *  that the conflict can't be resolved and the handling of nfsv4root_lock
4758  *  differs, as noted above.)
4759  * Unlocks State before returning a non-zero value.
4760  */
4761 static int
4762 nfsrv_delegconflict(struct nfsstate *stp, int *haslockp, NFSPROC_T *p,
4763     vnode_t vp)
4764 {
4765 	struct nfsclient *clp = stp->ls_clp;
4766 	int gotlock, error, lktype = 0, retrycnt, zapped_clp;
4767 	nfsv4stateid_t tstateid;
4768 	fhandle_t tfh;
4769 
4770 	/*
4771 	 * If the conflict is with an old delegation...
4772 	 */
4773 	if (stp->ls_flags & NFSLCK_OLDDELEG) {
4774 		/*
4775 		 * You can delete it, if it has expired.
4776 		 */
4777 		if (clp->lc_delegtime < NFSD_MONOSEC) {
4778 			nfsrv_freedeleg(stp);
4779 			NFSUNLOCKSTATE();
4780 			error = -1;
4781 			goto out;
4782 		}
4783 		NFSUNLOCKSTATE();
4784 		/*
4785 		 * During this delay, the old delegation could expire or it
4786 		 * could be recovered by the client via an Open with
4787 		 * CLAIM_DELEGATE_PREV.
4788 		 * Release the nfsv4root_lock, if held.
4789 		 */
4790 		if (*haslockp) {
4791 			*haslockp = 0;
4792 			NFSLOCKV4ROOTMUTEX();
4793 			nfsv4_unlock(&nfsv4rootfs_lock, 1);
4794 			NFSUNLOCKV4ROOTMUTEX();
4795 		}
4796 		error = NFSERR_DELAY;
4797 		goto out;
4798 	}
4799 
4800 	/*
4801 	 * It's a current delegation, so:
4802 	 * - check to see if the delegation has expired
4803 	 *   - if so, get the v4root lock and then expire it
4804 	 */
4805 	if (!(stp->ls_flags & NFSLCK_DELEGRECALL)) {
4806 		/*
4807 		 * - do a recall callback, since not yet done
4808 		 * For now, never allow truncate to be set. To use
4809 		 * truncate safely, it must be guaranteed that the
4810 		 * Remove, Rename or Setattr with size of 0 will
4811 		 * succeed and that would require major changes to
4812 		 * the VFS/Vnode OPs.
4813 		 * Set the expiry time large enough so that it won't expire
4814 		 * until after the callback, then set it correctly, once
4815 		 * the callback is done. (The delegation will now time
4816 		 * out whether or not the Recall worked ok. The timeout
4817 		 * will be extended when ops are done on the delegation
4818 		 * stateid, up to the timelimit.)
4819 		 */
4820 		stp->ls_delegtime = NFSD_MONOSEC + (2 * nfsrv_lease) +
4821 		    NFSRV_LEASEDELTA;
4822 		stp->ls_delegtimelimit = NFSD_MONOSEC + (6 * nfsrv_lease) +
4823 		    NFSRV_LEASEDELTA;
4824 		stp->ls_flags |= NFSLCK_DELEGRECALL;
4825 
4826 		/*
4827 		 * Loop NFSRV_CBRETRYCNT times while the CBRecall replies
4828 		 * NFSERR_BADSTATEID or NFSERR_BADHANDLE. This is done
4829 		 * in order to try and avoid a race that could happen
4830 		 * when a CBRecall request passed the Open reply with
4831 		 * the delegation in it when transitting the network.
4832 		 * Since nfsrv_docallback will sleep, don't use stp after
4833 		 * the call.
4834 		 */
4835 		NFSBCOPY((caddr_t)&stp->ls_stateid, (caddr_t)&tstateid,
4836 		    sizeof (tstateid));
4837 		NFSBCOPY((caddr_t)&stp->ls_lfp->lf_fh, (caddr_t)&tfh,
4838 		    sizeof (tfh));
4839 		NFSUNLOCKSTATE();
4840 		if (*haslockp) {
4841 			*haslockp = 0;
4842 			NFSLOCKV4ROOTMUTEX();
4843 			nfsv4_unlock(&nfsv4rootfs_lock, 1);
4844 			NFSUNLOCKV4ROOTMUTEX();
4845 		}
4846 		retrycnt = 0;
4847 		do {
4848 		    error = nfsrv_docallback(clp, NFSV4OP_CBRECALL,
4849 			&tstateid, 0, &tfh, NULL, NULL, p);
4850 		    retrycnt++;
4851 		} while ((error == NFSERR_BADSTATEID ||
4852 		    error == NFSERR_BADHANDLE) && retrycnt < NFSV4_CBRETRYCNT);
4853 		error = NFSERR_DELAY;
4854 		goto out;
4855 	}
4856 
4857 	if (clp->lc_expiry >= NFSD_MONOSEC &&
4858 	    stp->ls_delegtime >= NFSD_MONOSEC) {
4859 		NFSUNLOCKSTATE();
4860 		/*
4861 		 * A recall has been done, but it has not yet expired.
4862 		 * So, RETURN_DELAY.
4863 		 */
4864 		if (*haslockp) {
4865 			*haslockp = 0;
4866 			NFSLOCKV4ROOTMUTEX();
4867 			nfsv4_unlock(&nfsv4rootfs_lock, 1);
4868 			NFSUNLOCKV4ROOTMUTEX();
4869 		}
4870 		error = NFSERR_DELAY;
4871 		goto out;
4872 	}
4873 
4874 	/*
4875 	 * If we don't yet have the lock, just get it and then return,
4876 	 * since we need that before deleting expired state, such as
4877 	 * this delegation.
4878 	 * When getting the lock, unlock the vnode, so other nfsds that
4879 	 * are in progress, won't get stuck waiting for the vnode lock.
4880 	 */
4881 	if (*haslockp == 0) {
4882 		NFSUNLOCKSTATE();
4883 		if (vp != NULL) {
4884 			lktype = NFSVOPISLOCKED(vp);
4885 			NFSVOPUNLOCK(vp, 0);
4886 		}
4887 		NFSLOCKV4ROOTMUTEX();
4888 		nfsv4_relref(&nfsv4rootfs_lock);
4889 		do {
4890 			gotlock = nfsv4_lock(&nfsv4rootfs_lock, 1, NULL,
4891 			    NFSV4ROOTLOCKMUTEXPTR, NULL);
4892 		} while (!gotlock);
4893 		NFSUNLOCKV4ROOTMUTEX();
4894 		*haslockp = 1;
4895 		if (vp != NULL) {
4896 			NFSVOPLOCK(vp, lktype | LK_RETRY);
4897 			if ((vp->v_iflag & VI_DOOMED) != 0) {
4898 				*haslockp = 0;
4899 				NFSLOCKV4ROOTMUTEX();
4900 				nfsv4_unlock(&nfsv4rootfs_lock, 1);
4901 				NFSUNLOCKV4ROOTMUTEX();
4902 				error = NFSERR_PERM;
4903 				goto out;
4904 			}
4905 		}
4906 		error = -1;
4907 		goto out;
4908 	}
4909 
4910 	NFSUNLOCKSTATE();
4911 	/*
4912 	 * Ok, we can delete the expired delegation.
4913 	 * First, write the Revoke record to stable storage and then
4914 	 * clear out the conflict.
4915 	 * Since all other nfsd threads are now blocked, we can safely
4916 	 * sleep without the state changing.
4917 	 */
4918 	nfsrv_writestable(clp->lc_id, clp->lc_idlen, NFSNST_REVOKE, p);
4919 	nfsrv_backupstable();
4920 	if (clp->lc_expiry < NFSD_MONOSEC) {
4921 		nfsrv_cleanclient(clp, p);
4922 		nfsrv_freedeleglist(&clp->lc_deleg);
4923 		nfsrv_freedeleglist(&clp->lc_olddeleg);
4924 		LIST_REMOVE(clp, lc_hash);
4925 		zapped_clp = 1;
4926 	} else {
4927 		nfsrv_freedeleg(stp);
4928 		zapped_clp = 0;
4929 	}
4930 	if (zapped_clp)
4931 		nfsrv_zapclient(clp, p);
4932 	error = -1;
4933 
4934 out:
4935 	NFSEXITCODE(error);
4936 	return (error);
4937 }
4938 
4939 /*
4940  * Check for a remove allowed, if remove is set to 1 and get rid of
4941  * delegations.
4942  */
4943 APPLESTATIC int
4944 nfsrv_checkremove(vnode_t vp, int remove, NFSPROC_T *p)
4945 {
4946 	struct nfsstate *stp;
4947 	struct nfslockfile *lfp;
4948 	int error, haslock = 0;
4949 	fhandle_t nfh;
4950 
4951 	/*
4952 	 * First, get the lock file structure.
4953 	 * (A return of -1 means no associated state, so remove ok.)
4954 	 */
4955 	error = nfsrv_getlockfh(vp, NFSLCK_CHECK, NULL, &nfh, p);
4956 tryagain:
4957 	NFSLOCKSTATE();
4958 	if (!error)
4959 		error = nfsrv_getlockfile(NFSLCK_CHECK, NULL, &lfp, &nfh, 0);
4960 	if (error) {
4961 		NFSUNLOCKSTATE();
4962 		if (haslock) {
4963 			NFSLOCKV4ROOTMUTEX();
4964 			nfsv4_unlock(&nfsv4rootfs_lock, 1);
4965 			NFSUNLOCKV4ROOTMUTEX();
4966 		}
4967 		if (error == -1)
4968 			error = 0;
4969 		goto out;
4970 	}
4971 
4972 	/*
4973 	 * Now, we must Recall any delegations.
4974 	 */
4975 	error = nfsrv_cleandeleg(vp, lfp, NULL, &haslock, p);
4976 	if (error) {
4977 		/*
4978 		 * nfsrv_cleandeleg() unlocks state for non-zero
4979 		 * return.
4980 		 */
4981 		if (error == -1)
4982 			goto tryagain;
4983 		if (haslock) {
4984 			NFSLOCKV4ROOTMUTEX();
4985 			nfsv4_unlock(&nfsv4rootfs_lock, 1);
4986 			NFSUNLOCKV4ROOTMUTEX();
4987 		}
4988 		goto out;
4989 	}
4990 
4991 	/*
4992 	 * Now, look for a conflicting open share.
4993 	 */
4994 	if (remove) {
4995 		/*
4996 		 * If the entry in the directory was the last reference to the
4997 		 * corresponding filesystem object, the object can be destroyed
4998 		 * */
4999 		if(lfp->lf_usecount>1)
5000 			LIST_FOREACH(stp, &lfp->lf_open, ls_file) {
5001 				if (stp->ls_flags & NFSLCK_WRITEDENY) {
5002 					error = NFSERR_FILEOPEN;
5003 					break;
5004 				}
5005 			}
5006 	}
5007 
5008 	NFSUNLOCKSTATE();
5009 	if (haslock) {
5010 		NFSLOCKV4ROOTMUTEX();
5011 		nfsv4_unlock(&nfsv4rootfs_lock, 1);
5012 		NFSUNLOCKV4ROOTMUTEX();
5013 	}
5014 
5015 out:
5016 	NFSEXITCODE(error);
5017 	return (error);
5018 }
5019 
5020 /*
5021  * Clear out all delegations for the file referred to by lfp.
5022  * May return NFSERR_DELAY, if there will be a delay waiting for
5023  * delegations to expire.
5024  * Returns -1 to indicate it slept while recalling a delegation.
5025  * This function has the side effect of deleting the nfslockfile structure,
5026  * if it no longer has associated state and didn't have to sleep.
5027  * Unlocks State before a non-zero value is returned.
5028  */
5029 static int
5030 nfsrv_cleandeleg(vnode_t vp, struct nfslockfile *lfp,
5031     struct nfsclient *clp, int *haslockp, NFSPROC_T *p)
5032 {
5033 	struct nfsstate *stp, *nstp;
5034 	int ret = 0;
5035 
5036 	stp = LIST_FIRST(&lfp->lf_deleg);
5037 	while (stp != LIST_END(&lfp->lf_deleg)) {
5038 		nstp = LIST_NEXT(stp, ls_file);
5039 		if (stp->ls_clp != clp) {
5040 			ret = nfsrv_delegconflict(stp, haslockp, p, vp);
5041 			if (ret) {
5042 				/*
5043 				 * nfsrv_delegconflict() unlocks state
5044 				 * when it returns non-zero.
5045 				 */
5046 				goto out;
5047 			}
5048 		}
5049 		stp = nstp;
5050 	}
5051 out:
5052 	NFSEXITCODE(ret);
5053 	return (ret);
5054 }
5055 
5056 /*
5057  * There are certain operations that, when being done outside of NFSv4,
5058  * require that any NFSv4 delegation for the file be recalled.
5059  * This function is to be called for those cases:
5060  * VOP_RENAME() - When a delegation is being recalled for any reason,
5061  *	the client may have to do Opens against the server, using the file's
5062  *	final component name. If the file has been renamed on the server,
5063  *	that component name will be incorrect and the Open will fail.
5064  * VOP_REMOVE() - Theoretically, a client could Open a file after it has
5065  *	been removed on the server, if there is a delegation issued to
5066  *	that client for the file. I say "theoretically" since clients
5067  *	normally do an Access Op before the Open and that Access Op will
5068  *	fail with ESTALE. Note that NFSv2 and 3 don't even do Opens, so
5069  *	they will detect the file's removal in the same manner. (There is
5070  *	one case where RFC3530 allows a client to do an Open without first
5071  *	doing an Access Op, which is passage of a check against the ACE
5072  *	returned with a Write delegation, but current practice is to ignore
5073  *	the ACE and always do an Access Op.)
5074  *	Since the functions can only be called with an unlocked vnode, this
5075  *	can't be done at this time.
5076  * VOP_ADVLOCK() - When a client holds a delegation, it can issue byte range
5077  *	locks locally in the client, which are not visible to the server. To
5078  *	deal with this, issuing of delegations for a vnode must be disabled
5079  *	and all delegations for the vnode recalled. This is done via the
5080  *	second function, using the VV_DISABLEDELEG vflag on the vnode.
5081  */
5082 APPLESTATIC void
5083 nfsd_recalldelegation(vnode_t vp, NFSPROC_T *p)
5084 {
5085 	time_t starttime;
5086 	int error;
5087 
5088 	/*
5089 	 * First, check to see if the server is currently running and it has
5090 	 * been called for a regular file when issuing delegations.
5091 	 */
5092 	if (newnfs_numnfsd == 0 || vp->v_type != VREG ||
5093 	    nfsrv_issuedelegs == 0)
5094 		return;
5095 
5096 	KASSERT((NFSVOPISLOCKED(vp) != LK_EXCLUSIVE), ("vp %p is locked", vp));
5097 	/*
5098 	 * First, get a reference on the nfsv4rootfs_lock so that an
5099 	 * exclusive lock cannot be acquired by another thread.
5100 	 */
5101 	NFSLOCKV4ROOTMUTEX();
5102 	nfsv4_getref(&nfsv4rootfs_lock, NULL, NFSV4ROOTLOCKMUTEXPTR, NULL);
5103 	NFSUNLOCKV4ROOTMUTEX();
5104 
5105 	/*
5106 	 * Now, call nfsrv_checkremove() in a loop while it returns
5107 	 * NFSERR_DELAY. Return upon any other error or when timed out.
5108 	 */
5109 	starttime = NFSD_MONOSEC;
5110 	do {
5111 		if (NFSVOPLOCK(vp, LK_EXCLUSIVE) == 0) {
5112 			error = nfsrv_checkremove(vp, 0, p);
5113 			NFSVOPUNLOCK(vp, 0);
5114 		} else
5115 			error = EPERM;
5116 		if (error == NFSERR_DELAY) {
5117 			if (NFSD_MONOSEC - starttime > NFS_REMOVETIMEO)
5118 				break;
5119 			/* Sleep for a short period of time */
5120 			(void) nfs_catnap(PZERO, 0, "nfsremove");
5121 		}
5122 	} while (error == NFSERR_DELAY);
5123 	NFSLOCKV4ROOTMUTEX();
5124 	nfsv4_relref(&nfsv4rootfs_lock);
5125 	NFSUNLOCKV4ROOTMUTEX();
5126 }
5127 
5128 APPLESTATIC void
5129 nfsd_disabledelegation(vnode_t vp, NFSPROC_T *p)
5130 {
5131 
5132 #ifdef VV_DISABLEDELEG
5133 	/*
5134 	 * First, flag issuance of delegations disabled.
5135 	 */
5136 	atomic_set_long(&vp->v_vflag, VV_DISABLEDELEG);
5137 #endif
5138 
5139 	/*
5140 	 * Then call nfsd_recalldelegation() to get rid of all extant
5141 	 * delegations.
5142 	 */
5143 	nfsd_recalldelegation(vp, p);
5144 }
5145 
5146 /*
5147  * Check for conflicting locks, etc. and then get rid of delegations.
5148  * (At one point I thought that I should get rid of delegations for any
5149  *  Setattr, since it could potentially disallow the I/O op (read or write)
5150  *  allowed by the delegation. However, Setattr Ops that aren't changing
5151  *  the size get a stateid of all 0s, so you can't tell if it is a delegation
5152  *  for the same client or a different one, so I decided to only get rid
5153  *  of delegations for other clients when the size is being changed.)
5154  * In general, a Setattr can disable NFS I/O Ops that are outstanding, such
5155  * as Write backs, even if there is no delegation, so it really isn't any
5156  * different?)
5157  */
5158 APPLESTATIC int
5159 nfsrv_checksetattr(vnode_t vp, struct nfsrv_descript *nd,
5160     nfsv4stateid_t *stateidp, struct nfsvattr *nvap, nfsattrbit_t *attrbitp,
5161     struct nfsexstuff *exp, NFSPROC_T *p)
5162 {
5163 	struct nfsstate st, *stp = &st;
5164 	struct nfslock lo, *lop = &lo;
5165 	int error = 0;
5166 	nfsquad_t clientid;
5167 
5168 	if (NFSISSET_ATTRBIT(attrbitp, NFSATTRBIT_SIZE)) {
5169 		stp->ls_flags = (NFSLCK_CHECK | NFSLCK_WRITEACCESS);
5170 		lop->lo_first = nvap->na_size;
5171 	} else {
5172 		stp->ls_flags = 0;
5173 		lop->lo_first = 0;
5174 	}
5175 	if (NFSISSET_ATTRBIT(attrbitp, NFSATTRBIT_OWNER) ||
5176 	    NFSISSET_ATTRBIT(attrbitp, NFSATTRBIT_OWNERGROUP) ||
5177 	    NFSISSET_ATTRBIT(attrbitp, NFSATTRBIT_MODE) ||
5178 	    NFSISSET_ATTRBIT(attrbitp, NFSATTRBIT_ACL))
5179 		stp->ls_flags |= NFSLCK_SETATTR;
5180 	if (stp->ls_flags == 0)
5181 		goto out;
5182 	lop->lo_end = NFS64BITSSET;
5183 	lop->lo_flags = NFSLCK_WRITE;
5184 	stp->ls_ownerlen = 0;
5185 	stp->ls_op = NULL;
5186 	stp->ls_uid = nd->nd_cred->cr_uid;
5187 	stp->ls_stateid.seqid = stateidp->seqid;
5188 	clientid.lval[0] = stp->ls_stateid.other[0] = stateidp->other[0];
5189 	clientid.lval[1] = stp->ls_stateid.other[1] = stateidp->other[1];
5190 	stp->ls_stateid.other[2] = stateidp->other[2];
5191 	error = nfsrv_lockctrl(vp, &stp, &lop, NULL, clientid,
5192 	    stateidp, exp, nd, p);
5193 
5194 out:
5195 	NFSEXITCODE2(error, nd);
5196 	return (error);
5197 }
5198 
5199 /*
5200  * Check for a write delegation and do a CBGETATTR if there is one, updating
5201  * the attributes, as required.
5202  * Should I return an error if I can't get the attributes? (For now, I'll
5203  * just return ok.
5204  */
5205 APPLESTATIC int
5206 nfsrv_checkgetattr(struct nfsrv_descript *nd, vnode_t vp,
5207     struct nfsvattr *nvap, nfsattrbit_t *attrbitp, struct ucred *cred,
5208     NFSPROC_T *p)
5209 {
5210 	struct nfsstate *stp;
5211 	struct nfslockfile *lfp;
5212 	struct nfsclient *clp;
5213 	struct nfsvattr nva;
5214 	fhandle_t nfh;
5215 	int error = 0;
5216 	nfsattrbit_t cbbits;
5217 	u_quad_t delegfilerev;
5218 
5219 	NFSCBGETATTR_ATTRBIT(attrbitp, &cbbits);
5220 	if (!NFSNONZERO_ATTRBIT(&cbbits))
5221 		goto out;
5222 
5223 	/*
5224 	 * Get the lock file structure.
5225 	 * (A return of -1 means no associated state, so return ok.)
5226 	 */
5227 	error = nfsrv_getlockfh(vp, NFSLCK_CHECK, NULL, &nfh, p);
5228 	NFSLOCKSTATE();
5229 	if (!error)
5230 		error = nfsrv_getlockfile(NFSLCK_CHECK, NULL, &lfp, &nfh, 0);
5231 	if (error) {
5232 		NFSUNLOCKSTATE();
5233 		if (error == -1)
5234 			error = 0;
5235 		goto out;
5236 	}
5237 
5238 	/*
5239 	 * Now, look for a write delegation.
5240 	 */
5241 	LIST_FOREACH(stp, &lfp->lf_deleg, ls_file) {
5242 		if (stp->ls_flags & NFSLCK_DELEGWRITE)
5243 			break;
5244 	}
5245 	if (stp == LIST_END(&lfp->lf_deleg)) {
5246 		NFSUNLOCKSTATE();
5247 		goto out;
5248 	}
5249 	clp = stp->ls_clp;
5250 	delegfilerev = stp->ls_filerev;
5251 
5252 	/*
5253 	 * If the Write delegation was issued as a part of this Compound RPC
5254 	 * or if we have an Implied Clientid (used in a previous Op in this
5255 	 * compound) and it is the client the delegation was issued to,
5256 	 * just return ok.
5257 	 * I also assume that it is from the same client iff the network
5258 	 * host IP address is the same as the callback address. (Not
5259 	 * exactly correct by the RFC, but avoids a lot of Getattr
5260 	 * callbacks.)
5261 	 */
5262 	if (nd->nd_compref == stp->ls_compref ||
5263 	    ((nd->nd_flag & ND_IMPLIEDCLID) &&
5264 	     clp->lc_clientid.qval == nd->nd_clientid.qval) ||
5265 	     nfsaddr2_match(clp->lc_req.nr_nam, nd->nd_nam)) {
5266 		NFSUNLOCKSTATE();
5267 		goto out;
5268 	}
5269 
5270 	/*
5271 	 * We are now done with the delegation state structure,
5272 	 * so the statelock can be released and we can now tsleep().
5273 	 */
5274 
5275 	/*
5276 	 * Now, we must do the CB Getattr callback, to see if Change or Size
5277 	 * has changed.
5278 	 */
5279 	if (clp->lc_expiry >= NFSD_MONOSEC) {
5280 		NFSUNLOCKSTATE();
5281 		NFSVNO_ATTRINIT(&nva);
5282 		nva.na_filerev = NFS64BITSSET;
5283 		error = nfsrv_docallback(clp, NFSV4OP_CBGETATTR, NULL,
5284 		    0, &nfh, &nva, &cbbits, p);
5285 		if (!error) {
5286 			if ((nva.na_filerev != NFS64BITSSET &&
5287 			    nva.na_filerev > delegfilerev) ||
5288 			    (NFSVNO_ISSETSIZE(&nva) &&
5289 			     nva.na_size != nvap->na_size)) {
5290 				error = nfsvno_updfilerev(vp, nvap, cred, p);
5291 				if (NFSVNO_ISSETSIZE(&nva))
5292 					nvap->na_size = nva.na_size;
5293 			}
5294 		} else
5295 			error = 0;	/* Ignore callback errors for now. */
5296 	} else {
5297 		NFSUNLOCKSTATE();
5298 	}
5299 
5300 out:
5301 	NFSEXITCODE2(error, nd);
5302 	return (error);
5303 }
5304 
5305 /*
5306  * This function looks for openowners that haven't had any opens for
5307  * a while and throws them away. Called by an nfsd when NFSNSF_NOOPENS
5308  * is set.
5309  */
5310 APPLESTATIC void
5311 nfsrv_throwawayopens(NFSPROC_T *p)
5312 {
5313 	struct nfsclient *clp, *nclp;
5314 	struct nfsstate *stp, *nstp;
5315 	int i;
5316 
5317 	NFSLOCKSTATE();
5318 	nfsrv_stablefirst.nsf_flags &= ~NFSNSF_NOOPENS;
5319 	/*
5320 	 * For each client...
5321 	 */
5322 	for (i = 0; i < NFSCLIENTHASHSIZE; i++) {
5323 	    LIST_FOREACH_SAFE(clp, &nfsclienthash[i], lc_hash, nclp) {
5324 		LIST_FOREACH_SAFE(stp, &clp->lc_open, ls_list, nstp) {
5325 			if (LIST_EMPTY(&stp->ls_open) &&
5326 			    (stp->ls_noopens > NFSNOOPEN ||
5327 			     (nfsrv_openpluslock * 2) >
5328 			     NFSRV_V4STATELIMIT))
5329 				nfsrv_freeopenowner(stp, 0, p);
5330 		}
5331 	    }
5332 	}
5333 	NFSUNLOCKSTATE();
5334 }
5335 
5336 /*
5337  * This function checks to see if the credentials are the same.
5338  * Returns 1 for not same, 0 otherwise.
5339  */
5340 static int
5341 nfsrv_notsamecredname(struct nfsrv_descript *nd, struct nfsclient *clp)
5342 {
5343 
5344 	if (nd->nd_flag & ND_GSS) {
5345 		if (!(clp->lc_flags & LCL_GSS))
5346 			return (1);
5347 		if (clp->lc_flags & LCL_NAME) {
5348 			if (nd->nd_princlen != clp->lc_namelen ||
5349 			    NFSBCMP(nd->nd_principal, clp->lc_name,
5350 				clp->lc_namelen))
5351 				return (1);
5352 			else
5353 				return (0);
5354 		}
5355 		if (nd->nd_cred->cr_uid == clp->lc_uid)
5356 			return (0);
5357 		else
5358 			return (1);
5359 	} else if (clp->lc_flags & LCL_GSS)
5360 		return (1);
5361 	/*
5362 	 * For AUTH_SYS, allow the same uid or root. (This is underspecified
5363 	 * in RFC3530, which talks about principals, but doesn't say anything
5364 	 * about uids for AUTH_SYS.)
5365 	 */
5366 	if (nd->nd_cred->cr_uid == clp->lc_uid || nd->nd_cred->cr_uid == 0)
5367 		return (0);
5368 	else
5369 		return (1);
5370 }
5371 
5372 /*
5373  * Calculate the lease expiry time.
5374  */
5375 static time_t
5376 nfsrv_leaseexpiry(void)
5377 {
5378 
5379 	if (nfsrv_stablefirst.nsf_eograce > NFSD_MONOSEC)
5380 		return (NFSD_MONOSEC + 2 * (nfsrv_lease + NFSRV_LEASEDELTA));
5381 	return (NFSD_MONOSEC + nfsrv_lease + NFSRV_LEASEDELTA);
5382 }
5383 
5384 /*
5385  * Delay the delegation timeout as far as ls_delegtimelimit, as required.
5386  */
5387 static void
5388 nfsrv_delaydelegtimeout(struct nfsstate *stp)
5389 {
5390 
5391 	if ((stp->ls_flags & NFSLCK_DELEGRECALL) == 0)
5392 		return;
5393 
5394 	if ((stp->ls_delegtime + 15) > NFSD_MONOSEC &&
5395 	    stp->ls_delegtime < stp->ls_delegtimelimit) {
5396 		stp->ls_delegtime += nfsrv_lease;
5397 		if (stp->ls_delegtime > stp->ls_delegtimelimit)
5398 			stp->ls_delegtime = stp->ls_delegtimelimit;
5399 	}
5400 }
5401 
5402 /*
5403  * This function checks to see if there is any other state associated
5404  * with the openowner for this Open.
5405  * It returns 1 if there is no other state, 0 otherwise.
5406  */
5407 static int
5408 nfsrv_nootherstate(struct nfsstate *stp)
5409 {
5410 	struct nfsstate *tstp;
5411 
5412 	LIST_FOREACH(tstp, &stp->ls_openowner->ls_open, ls_list) {
5413 		if (tstp != stp || !LIST_EMPTY(&tstp->ls_lock))
5414 			return (0);
5415 	}
5416 	return (1);
5417 }
5418 
5419 /*
5420  * Create a list of lock deltas (changes to local byte range locking
5421  * that can be rolled back using the list) and apply the changes via
5422  * nfsvno_advlock(). Optionally, lock the list. It is expected that either
5423  * the rollback or update function will be called after this.
5424  * It returns an error (and rolls back, as required), if any nfsvno_advlock()
5425  * call fails. If it returns an error, it will unlock the list.
5426  */
5427 static int
5428 nfsrv_locallock(vnode_t vp, struct nfslockfile *lfp, int flags,
5429     uint64_t first, uint64_t end, struct nfslockconflict *cfp, NFSPROC_T *p)
5430 {
5431 	struct nfslock *lop, *nlop;
5432 	int error = 0;
5433 
5434 	/* Loop through the list of locks. */
5435 	lop = LIST_FIRST(&lfp->lf_locallock);
5436 	while (first < end && lop != NULL) {
5437 		nlop = LIST_NEXT(lop, lo_lckowner);
5438 		if (first >= lop->lo_end) {
5439 			/* not there yet */
5440 			lop = nlop;
5441 		} else if (first < lop->lo_first) {
5442 			/* new one starts before entry in list */
5443 			if (end <= lop->lo_first) {
5444 				/* no overlap between old and new */
5445 				error = nfsrv_dolocal(vp, lfp, flags,
5446 				    NFSLCK_UNLOCK, first, end, cfp, p);
5447 				if (error != 0)
5448 					break;
5449 				first = end;
5450 			} else {
5451 				/* handle fragment overlapped with new one */
5452 				error = nfsrv_dolocal(vp, lfp, flags,
5453 				    NFSLCK_UNLOCK, first, lop->lo_first, cfp,
5454 				    p);
5455 				if (error != 0)
5456 					break;
5457 				first = lop->lo_first;
5458 			}
5459 		} else {
5460 			/* new one overlaps this entry in list */
5461 			if (end <= lop->lo_end) {
5462 				/* overlaps all of new one */
5463 				error = nfsrv_dolocal(vp, lfp, flags,
5464 				    lop->lo_flags, first, end, cfp, p);
5465 				if (error != 0)
5466 					break;
5467 				first = end;
5468 			} else {
5469 				/* handle fragment overlapped with new one */
5470 				error = nfsrv_dolocal(vp, lfp, flags,
5471 				    lop->lo_flags, first, lop->lo_end, cfp, p);
5472 				if (error != 0)
5473 					break;
5474 				first = lop->lo_end;
5475 				lop = nlop;
5476 			}
5477 		}
5478 	}
5479 	if (first < end && error == 0)
5480 		/* handle fragment past end of list */
5481 		error = nfsrv_dolocal(vp, lfp, flags, NFSLCK_UNLOCK, first,
5482 		    end, cfp, p);
5483 
5484 	NFSEXITCODE(error);
5485 	return (error);
5486 }
5487 
5488 /*
5489  * Local lock unlock. Unlock all byte ranges that are no longer locked
5490  * by NFSv4. To do this, unlock any subranges of first-->end that
5491  * do not overlap with the byte ranges of any lock in the lfp->lf_lock
5492  * list. This list has all locks for the file held by other
5493  * <clientid, lockowner> tuples. The list is ordered by increasing
5494  * lo_first value, but may have entries that overlap each other, for
5495  * the case of read locks.
5496  */
5497 static void
5498 nfsrv_localunlock(vnode_t vp, struct nfslockfile *lfp, uint64_t init_first,
5499     uint64_t init_end, NFSPROC_T *p)
5500 {
5501 	struct nfslock *lop;
5502 	uint64_t first, end, prevfirst;
5503 
5504 	first = init_first;
5505 	end = init_end;
5506 	while (first < init_end) {
5507 		/* Loop through all nfs locks, adjusting first and end */
5508 		prevfirst = 0;
5509 		LIST_FOREACH(lop, &lfp->lf_lock, lo_lckfile) {
5510 			KASSERT(prevfirst <= lop->lo_first,
5511 			    ("nfsv4 locks out of order"));
5512 			KASSERT(lop->lo_first < lop->lo_end,
5513 			    ("nfsv4 bogus lock"));
5514 			prevfirst = lop->lo_first;
5515 			if (first >= lop->lo_first &&
5516 			    first < lop->lo_end)
5517 				/*
5518 				 * Overlaps with initial part, so trim
5519 				 * off that initial part by moving first past
5520 				 * it.
5521 				 */
5522 				first = lop->lo_end;
5523 			else if (end > lop->lo_first &&
5524 			    lop->lo_first > first) {
5525 				/*
5526 				 * This lock defines the end of the
5527 				 * segment to unlock, so set end to the
5528 				 * start of it and break out of the loop.
5529 				 */
5530 				end = lop->lo_first;
5531 				break;
5532 			}
5533 			if (first >= end)
5534 				/*
5535 				 * There is no segment left to do, so
5536 				 * break out of this loop and then exit
5537 				 * the outer while() since first will be set
5538 				 * to end, which must equal init_end here.
5539 				 */
5540 				break;
5541 		}
5542 		if (first < end) {
5543 			/* Unlock this segment */
5544 			(void) nfsrv_dolocal(vp, lfp, NFSLCK_UNLOCK,
5545 			    NFSLCK_READ, first, end, NULL, p);
5546 			nfsrv_locallock_commit(lfp, NFSLCK_UNLOCK,
5547 			    first, end);
5548 		}
5549 		/*
5550 		 * Now move past this segment and look for any further
5551 		 * segment in the range, if there is one.
5552 		 */
5553 		first = end;
5554 		end = init_end;
5555 	}
5556 }
5557 
5558 /*
5559  * Do the local lock operation and update the rollback list, as required.
5560  * Perform the rollback and return the error if nfsvno_advlock() fails.
5561  */
5562 static int
5563 nfsrv_dolocal(vnode_t vp, struct nfslockfile *lfp, int flags, int oldflags,
5564     uint64_t first, uint64_t end, struct nfslockconflict *cfp, NFSPROC_T *p)
5565 {
5566 	struct nfsrollback *rlp;
5567 	int error = 0, ltype, oldltype;
5568 
5569 	if (flags & NFSLCK_WRITE)
5570 		ltype = F_WRLCK;
5571 	else if (flags & NFSLCK_READ)
5572 		ltype = F_RDLCK;
5573 	else
5574 		ltype = F_UNLCK;
5575 	if (oldflags & NFSLCK_WRITE)
5576 		oldltype = F_WRLCK;
5577 	else if (oldflags & NFSLCK_READ)
5578 		oldltype = F_RDLCK;
5579 	else
5580 		oldltype = F_UNLCK;
5581 	if (ltype == oldltype || (oldltype == F_WRLCK && ltype == F_RDLCK))
5582 		/* nothing to do */
5583 		goto out;
5584 	error = nfsvno_advlock(vp, ltype, first, end, p);
5585 	if (error != 0) {
5586 		if (cfp != NULL) {
5587 			cfp->cl_clientid.lval[0] = 0;
5588 			cfp->cl_clientid.lval[1] = 0;
5589 			cfp->cl_first = 0;
5590 			cfp->cl_end = NFS64BITSSET;
5591 			cfp->cl_flags = NFSLCK_WRITE;
5592 			cfp->cl_ownerlen = 5;
5593 			NFSBCOPY("LOCAL", cfp->cl_owner, 5);
5594 		}
5595 		nfsrv_locallock_rollback(vp, lfp, p);
5596 	} else if (ltype != F_UNLCK) {
5597 		rlp = malloc(sizeof (struct nfsrollback), M_NFSDROLLBACK,
5598 		    M_WAITOK);
5599 		rlp->rlck_first = first;
5600 		rlp->rlck_end = end;
5601 		rlp->rlck_type = oldltype;
5602 		LIST_INSERT_HEAD(&lfp->lf_rollback, rlp, rlck_list);
5603 	}
5604 
5605 out:
5606 	NFSEXITCODE(error);
5607 	return (error);
5608 }
5609 
5610 /*
5611  * Roll back local lock changes and free up the rollback list.
5612  */
5613 static void
5614 nfsrv_locallock_rollback(vnode_t vp, struct nfslockfile *lfp, NFSPROC_T *p)
5615 {
5616 	struct nfsrollback *rlp, *nrlp;
5617 
5618 	LIST_FOREACH_SAFE(rlp, &lfp->lf_rollback, rlck_list, nrlp) {
5619 		(void) nfsvno_advlock(vp, rlp->rlck_type, rlp->rlck_first,
5620 		    rlp->rlck_end, p);
5621 		free(rlp, M_NFSDROLLBACK);
5622 	}
5623 	LIST_INIT(&lfp->lf_rollback);
5624 }
5625 
5626 /*
5627  * Update local lock list and delete rollback list (ie now committed to the
5628  * local locks). Most of the work is done by the internal function.
5629  */
5630 static void
5631 nfsrv_locallock_commit(struct nfslockfile *lfp, int flags, uint64_t first,
5632     uint64_t end)
5633 {
5634 	struct nfsrollback *rlp, *nrlp;
5635 	struct nfslock *new_lop, *other_lop;
5636 
5637 	new_lop = malloc(sizeof (struct nfslock), M_NFSDLOCK, M_WAITOK);
5638 	if (flags & (NFSLCK_READ | NFSLCK_WRITE))
5639 		other_lop = malloc(sizeof (struct nfslock), M_NFSDLOCK,
5640 		    M_WAITOK);
5641 	else
5642 		other_lop = NULL;
5643 	new_lop->lo_flags = flags;
5644 	new_lop->lo_first = first;
5645 	new_lop->lo_end = end;
5646 	nfsrv_updatelock(NULL, &new_lop, &other_lop, lfp);
5647 	if (new_lop != NULL)
5648 		free(new_lop, M_NFSDLOCK);
5649 	if (other_lop != NULL)
5650 		free(other_lop, M_NFSDLOCK);
5651 
5652 	/* and get rid of the rollback list */
5653 	LIST_FOREACH_SAFE(rlp, &lfp->lf_rollback, rlck_list, nrlp)
5654 		free(rlp, M_NFSDROLLBACK);
5655 	LIST_INIT(&lfp->lf_rollback);
5656 }
5657 
5658 /*
5659  * Lock the struct nfslockfile for local lock updating.
5660  */
5661 static void
5662 nfsrv_locklf(struct nfslockfile *lfp)
5663 {
5664 	int gotlock;
5665 
5666 	/* lf_usecount ensures *lfp won't be free'd */
5667 	lfp->lf_usecount++;
5668 	do {
5669 		gotlock = nfsv4_lock(&lfp->lf_locallock_lck, 1, NULL,
5670 		    NFSSTATEMUTEXPTR, NULL);
5671 	} while (gotlock == 0);
5672 	lfp->lf_usecount--;
5673 }
5674 
5675 /*
5676  * Unlock the struct nfslockfile after local lock updating.
5677  */
5678 static void
5679 nfsrv_unlocklf(struct nfslockfile *lfp)
5680 {
5681 
5682 	nfsv4_unlock(&lfp->lf_locallock_lck, 0);
5683 }
5684 
5685 /*
5686  * Clear out all state for the NFSv4 server.
5687  * Must be called by a thread that can sleep when no nfsds are running.
5688  */
5689 void
5690 nfsrv_throwawayallstate(NFSPROC_T *p)
5691 {
5692 	struct nfsclient *clp, *nclp;
5693 	struct nfslockfile *lfp, *nlfp;
5694 	int i;
5695 
5696 	/*
5697 	 * For each client, clean out the state and then free the structure.
5698 	 */
5699 	for (i = 0; i < NFSCLIENTHASHSIZE; i++) {
5700 		LIST_FOREACH_SAFE(clp, &nfsclienthash[i], lc_hash, nclp) {
5701 			nfsrv_cleanclient(clp, p);
5702 			nfsrv_freedeleglist(&clp->lc_deleg);
5703 			nfsrv_freedeleglist(&clp->lc_olddeleg);
5704 			free(clp, M_NFSDCLIENT);
5705 		}
5706 	}
5707 
5708 	/*
5709 	 * Also, free up any remaining lock file structures.
5710 	 */
5711 	for (i = 0; i < NFSLOCKHASHSIZE; i++) {
5712 		LIST_FOREACH_SAFE(lfp, &nfslockhash[i], lf_hash, nlfp) {
5713 			printf("nfsd unload: fnd a lock file struct\n");
5714 			nfsrv_freenfslockfile(lfp);
5715 		}
5716 	}
5717 }
5718 
5719 /*
5720  * Check the sequence# for the session and slot provided as an argument.
5721  * Also, renew the lease if the session will return NFS_OK.
5722  */
5723 int
5724 nfsrv_checksequence(struct nfsrv_descript *nd, uint32_t sequenceid,
5725     uint32_t *highest_slotidp, uint32_t *target_highest_slotidp, int cache_this,
5726     uint32_t *sflagsp, NFSPROC_T *p)
5727 {
5728 	struct nfsdsession *sep;
5729 	struct nfssessionhash *shp;
5730 	int error;
5731 	SVCXPRT *savxprt;
5732 
5733 	shp = NFSSESSIONHASH(nd->nd_sessionid);
5734 	NFSLOCKSESSION(shp);
5735 	sep = nfsrv_findsession(nd->nd_sessionid);
5736 	if (sep == NULL) {
5737 		NFSUNLOCKSESSION(shp);
5738 		return (NFSERR_BADSESSION);
5739 	}
5740 	error = nfsv4_seqsession(sequenceid, nd->nd_slotid, *highest_slotidp,
5741 	    sep->sess_slots, NULL, NFSV4_SLOTS - 1);
5742 	if (error != 0) {
5743 		NFSUNLOCKSESSION(shp);
5744 		return (error);
5745 	}
5746 	if (cache_this != 0)
5747 		nd->nd_flag |= ND_SAVEREPLY;
5748 	/* Renew the lease. */
5749 	sep->sess_clp->lc_expiry = nfsrv_leaseexpiry();
5750 	nd->nd_clientid.qval = sep->sess_clp->lc_clientid.qval;
5751 	nd->nd_flag |= ND_IMPLIEDCLID;
5752 
5753 	/*
5754 	 * If this session handles the backchannel, save the nd_xprt for this
5755 	 * RPC, since this is the one being used.
5756 	 */
5757 	if (sep->sess_cbsess.nfsess_xprt != NULL &&
5758 	    (sep->sess_crflags & NFSV4CRSESS_CONNBACKCHAN) != 0) {
5759 		savxprt = sep->sess_cbsess.nfsess_xprt;
5760 		SVC_ACQUIRE(nd->nd_xprt);
5761 		nd->nd_xprt->xp_p2 = savxprt->xp_p2;
5762 		nd->nd_xprt->xp_idletimeout = 0;	/* Disable timeout. */
5763 		sep->sess_cbsess.nfsess_xprt = nd->nd_xprt;
5764 		SVC_RELEASE(savxprt);
5765 	}
5766 
5767 	*sflagsp = 0;
5768 	if (sep->sess_clp->lc_req.nr_client == NULL)
5769 		*sflagsp |= NFSV4SEQ_CBPATHDOWN;
5770 	NFSUNLOCKSESSION(shp);
5771 	if (error == NFSERR_EXPIRED) {
5772 		*sflagsp |= NFSV4SEQ_EXPIREDALLSTATEREVOKED;
5773 		error = 0;
5774 	} else if (error == NFSERR_ADMINREVOKED) {
5775 		*sflagsp |= NFSV4SEQ_ADMINSTATEREVOKED;
5776 		error = 0;
5777 	}
5778 	*highest_slotidp = *target_highest_slotidp = NFSV4_SLOTS - 1;
5779 	return (0);
5780 }
5781 
5782 /*
5783  * Check/set reclaim complete for this session/clientid.
5784  */
5785 int
5786 nfsrv_checkreclaimcomplete(struct nfsrv_descript *nd)
5787 {
5788 	struct nfsdsession *sep;
5789 	struct nfssessionhash *shp;
5790 	int error = 0;
5791 
5792 	shp = NFSSESSIONHASH(nd->nd_sessionid);
5793 	NFSLOCKSTATE();
5794 	NFSLOCKSESSION(shp);
5795 	sep = nfsrv_findsession(nd->nd_sessionid);
5796 	if (sep == NULL) {
5797 		NFSUNLOCKSESSION(shp);
5798 		NFSUNLOCKSTATE();
5799 		return (NFSERR_BADSESSION);
5800 	}
5801 
5802 	/* Check to see if reclaim complete has already happened. */
5803 	if ((sep->sess_clp->lc_flags & LCL_RECLAIMCOMPLETE) != 0)
5804 		error = NFSERR_COMPLETEALREADY;
5805 	else
5806 		sep->sess_clp->lc_flags |= LCL_RECLAIMCOMPLETE;
5807 	NFSUNLOCKSESSION(shp);
5808 	NFSUNLOCKSTATE();
5809 	return (error);
5810 }
5811 
5812 /*
5813  * Cache the reply in a session slot.
5814  */
5815 void
5816 nfsrv_cache_session(uint8_t *sessionid, uint32_t slotid, int repstat,
5817    struct mbuf **m)
5818 {
5819 	struct nfsdsession *sep;
5820 	struct nfssessionhash *shp;
5821 
5822 	shp = NFSSESSIONHASH(sessionid);
5823 	NFSLOCKSESSION(shp);
5824 	sep = nfsrv_findsession(sessionid);
5825 	if (sep == NULL) {
5826 		NFSUNLOCKSESSION(shp);
5827 		printf("nfsrv_cache_session: no session\n");
5828 		m_freem(*m);
5829 		return;
5830 	}
5831 	nfsv4_seqsess_cacherep(slotid, sep->sess_slots, repstat, m);
5832 	NFSUNLOCKSESSION(shp);
5833 }
5834 
5835 /*
5836  * Search for a session that matches the sessionid.
5837  */
5838 static struct nfsdsession *
5839 nfsrv_findsession(uint8_t *sessionid)
5840 {
5841 	struct nfsdsession *sep;
5842 	struct nfssessionhash *shp;
5843 
5844 	shp = NFSSESSIONHASH(sessionid);
5845 	LIST_FOREACH(sep, &shp->list, sess_hash) {
5846 		if (!NFSBCMP(sessionid, sep->sess_sessionid, NFSX_V4SESSIONID))
5847 			break;
5848 	}
5849 	return (sep);
5850 }
5851 
5852 /*
5853  * Destroy a session.
5854  */
5855 int
5856 nfsrv_destroysession(struct nfsrv_descript *nd, uint8_t *sessionid)
5857 {
5858 	int error, samesess;
5859 
5860 	samesess = 0;
5861 	if (!NFSBCMP(sessionid, nd->nd_sessionid, NFSX_V4SESSIONID)) {
5862 		samesess = 1;
5863 		if ((nd->nd_flag & ND_LASTOP) == 0)
5864 			return (NFSERR_BADSESSION);
5865 	}
5866 	error = nfsrv_freesession(NULL, sessionid);
5867 	if (error == 0 && samesess != 0)
5868 		nd->nd_flag &= ~ND_HASSEQUENCE;
5869 	return (error);
5870 }
5871 
5872 /*
5873  * Free up a session structure.
5874  */
5875 static int
5876 nfsrv_freesession(struct nfsdsession *sep, uint8_t *sessionid)
5877 {
5878 	struct nfssessionhash *shp;
5879 	int i;
5880 
5881 	if (sep == NULL) {
5882 		shp = NFSSESSIONHASH(sessionid);
5883 		NFSLOCKSESSION(shp);
5884 		sep = nfsrv_findsession(sessionid);
5885 	} else {
5886 		shp = NFSSESSIONHASH(sep->sess_sessionid);
5887 		NFSLOCKSESSION(shp);
5888 	}
5889 	if (sep != NULL) {
5890 		NFSLOCKSTATE();
5891 		sep->sess_refcnt--;
5892 		if (sep->sess_refcnt > 0) {
5893 			NFSUNLOCKSTATE();
5894 			NFSUNLOCKSESSION(shp);
5895 			return (0);
5896 		}
5897 		LIST_REMOVE(sep, sess_hash);
5898 		LIST_REMOVE(sep, sess_list);
5899 		NFSUNLOCKSTATE();
5900 	}
5901 	NFSUNLOCKSESSION(shp);
5902 	if (sep == NULL)
5903 		return (NFSERR_BADSESSION);
5904 	for (i = 0; i < NFSV4_SLOTS; i++)
5905 		if (sep->sess_slots[i].nfssl_reply != NULL)
5906 			m_freem(sep->sess_slots[i].nfssl_reply);
5907 	if (sep->sess_cbsess.nfsess_xprt != NULL)
5908 		SVC_RELEASE(sep->sess_cbsess.nfsess_xprt);
5909 	free(sep, M_NFSDSESSION);
5910 	return (0);
5911 }
5912 
5913 /*
5914  * Free a stateid.
5915  * RFC5661 says that it should fail when there are associated opens, locks
5916  * or delegations. Since stateids represent opens, I don't see how you can
5917  * free an open stateid (it will be free'd when closed), so this function
5918  * only works for lock stateids (freeing the lock_owner) or delegations.
5919  */
5920 int
5921 nfsrv_freestateid(struct nfsrv_descript *nd, nfsv4stateid_t *stateidp,
5922     NFSPROC_T *p)
5923 {
5924 	struct nfsclient *clp;
5925 	struct nfsstate *stp;
5926 	int error;
5927 
5928 	NFSLOCKSTATE();
5929 	/*
5930 	 * Look up the stateid
5931 	 */
5932 	error = nfsrv_getclient((nfsquad_t)((u_quad_t)0), CLOPS_RENEW, &clp,
5933 	    NULL, (nfsquad_t)((u_quad_t)0), 0, nd, p);
5934 	if (error == 0) {
5935 		/* First, check for a delegation. */
5936 		LIST_FOREACH(stp, &clp->lc_deleg, ls_list) {
5937 			if (!NFSBCMP(stp->ls_stateid.other, stateidp->other,
5938 			    NFSX_STATEIDOTHER))
5939 				break;
5940 		}
5941 		if (stp != NULL) {
5942 			nfsrv_freedeleg(stp);
5943 			NFSUNLOCKSTATE();
5944 			return (error);
5945 		}
5946 	}
5947 	/* Not a delegation, try for a lock_owner. */
5948 	if (error == 0)
5949 		error = nfsrv_getstate(clp, stateidp, 0, &stp);
5950 	if (error == 0 && ((stp->ls_flags & (NFSLCK_OPEN | NFSLCK_DELEGREAD |
5951 	    NFSLCK_DELEGWRITE)) != 0 || (stp->ls_flags & NFSLCK_LOCK) == 0))
5952 		/* Not a lock_owner stateid. */
5953 		error = NFSERR_LOCKSHELD;
5954 	if (error == 0 && !LIST_EMPTY(&stp->ls_lock))
5955 		error = NFSERR_LOCKSHELD;
5956 	if (error == 0)
5957 		nfsrv_freelockowner(stp, NULL, 0, p);
5958 	NFSUNLOCKSTATE();
5959 	return (error);
5960 }
5961 
5962 /*
5963  * Generate the xdr for an NFSv4.1 CBSequence Operation.
5964  */
5965 static int
5966 nfsv4_setcbsequence(struct nfsrv_descript *nd, struct nfsclient *clp,
5967     int dont_replycache, struct nfsdsession **sepp)
5968 {
5969 	struct nfsdsession *sep;
5970 	uint32_t *tl, slotseq = 0;
5971 	int maxslot, slotpos;
5972 	uint8_t sessionid[NFSX_V4SESSIONID];
5973 	int error;
5974 
5975 	error = nfsv4_getcbsession(clp, sepp);
5976 	if (error != 0)
5977 		return (error);
5978 	sep = *sepp;
5979 	(void)nfsv4_sequencelookup(NULL, &sep->sess_cbsess, &slotpos, &maxslot,
5980 	    &slotseq, sessionid);
5981 	KASSERT(maxslot >= 0, ("nfsv4_setcbsequence neg maxslot"));
5982 
5983 	/* Build the Sequence arguments. */
5984 	NFSM_BUILD(tl, uint32_t *, NFSX_V4SESSIONID + 5 * NFSX_UNSIGNED);
5985 	bcopy(sessionid, tl, NFSX_V4SESSIONID);
5986 	tl += NFSX_V4SESSIONID / NFSX_UNSIGNED;
5987 	nd->nd_slotseq = tl;
5988 	*tl++ = txdr_unsigned(slotseq);
5989 	*tl++ = txdr_unsigned(slotpos);
5990 	*tl++ = txdr_unsigned(maxslot);
5991 	if (dont_replycache == 0)
5992 		*tl++ = newnfs_true;
5993 	else
5994 		*tl++ = newnfs_false;
5995 	*tl = 0;			/* No referring call list, for now. */
5996 	nd->nd_flag |= ND_HASSEQUENCE;
5997 	return (0);
5998 }
5999 
6000 /*
6001  * Get a session for the callback.
6002  */
6003 static int
6004 nfsv4_getcbsession(struct nfsclient *clp, struct nfsdsession **sepp)
6005 {
6006 	struct nfsdsession *sep;
6007 
6008 	NFSLOCKSTATE();
6009 	LIST_FOREACH(sep, &clp->lc_session, sess_list) {
6010 		if ((sep->sess_crflags & NFSV4CRSESS_CONNBACKCHAN) != 0)
6011 			break;
6012 	}
6013 	if (sep == NULL) {
6014 		NFSUNLOCKSTATE();
6015 		return (NFSERR_BADSESSION);
6016 	}
6017 	sep->sess_refcnt++;
6018 	*sepp = sep;
6019 	NFSUNLOCKSTATE();
6020 	return (0);
6021 }
6022 
6023