xref: /freebsd/sys/fs/nfsserver/nfs_nfsdstate.c (revision aa64588d28258aef88cc33b8043112e8856948d0)
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 and lru 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 #endif	/* !APPLEKEXT */
55 
56 static u_int32_t nfsrv_openpluslock = 0, nfsrv_delegatecnt = 0;
57 static time_t nfsrvboottime;
58 static int nfsrv_writedelegifpos = 1;
59 static int nfsrv_returnoldstateid = 0, nfsrv_clients = 0;
60 static int nfsrv_clienthighwater = NFSRV_CLIENTHIGHWATER;
61 static int nfsrv_nogsscallback = 0;
62 
63 /* local functions */
64 static void nfsrv_dumpaclient(struct nfsclient *clp,
65     struct nfsd_dumpclients *dumpp);
66 static void nfsrv_freeopenowner(struct nfsstate *stp, int cansleep,
67     NFSPROC_T *p);
68 static int nfsrv_freeopen(struct nfsstate *stp, vnode_t vp, int cansleep,
69     NFSPROC_T *p);
70 static void nfsrv_freelockowner(struct nfsstate *stp, vnode_t vp, int cansleep,
71     NFSPROC_T *p);
72 static void nfsrv_freeallnfslocks(struct nfsstate *stp, vnode_t vp,
73     int cansleep, NFSPROC_T *p);
74 static void nfsrv_freenfslock(struct nfslock *lop);
75 static void nfsrv_freenfslockfile(struct nfslockfile *lfp);
76 static void nfsrv_freedeleg(struct nfsstate *);
77 static int nfsrv_getstate(struct nfsclient *clp, nfsv4stateid_t *stateidp,
78     u_int32_t flags, struct nfsstate **stpp);
79 static void nfsrv_getowner(struct nfsstatehead *hp, struct nfsstate *new_stp,
80     struct nfsstate **stpp);
81 static int nfsrv_getlockfh(vnode_t vp, u_short flags,
82     struct nfslockfile **new_lfpp, fhandle_t *nfhp, NFSPROC_T *p);
83 static int nfsrv_getlockfile(u_short flags, struct nfslockfile **new_lfpp,
84     struct nfslockfile **lfpp, fhandle_t *nfhp, int lockit);
85 static void nfsrv_insertlock(struct nfslock *new_lop,
86     struct nfslock *insert_lop, struct nfsstate *stp, struct nfslockfile *lfp);
87 static void nfsrv_updatelock(struct nfsstate *stp, struct nfslock **new_lopp,
88     struct nfslock **other_lopp, struct nfslockfile *lfp);
89 static int nfsrv_getipnumber(u_char *cp);
90 static int nfsrv_checkrestart(nfsquad_t clientid, u_int32_t flags,
91     nfsv4stateid_t *stateidp, int specialid);
92 static int nfsrv_checkgrace(u_int32_t flags);
93 static int nfsrv_docallback(struct nfsclient *clp, int procnum,
94     nfsv4stateid_t *stateidp, int trunc, fhandle_t *fhp,
95     struct nfsvattr *nap, nfsattrbit_t *attrbitp, NFSPROC_T *p);
96 static u_int32_t nfsrv_nextclientindex(void);
97 static u_int32_t nfsrv_nextstateindex(struct nfsclient *clp);
98 static void nfsrv_markstable(struct nfsclient *clp);
99 static int nfsrv_checkstable(struct nfsclient *clp);
100 static int nfsrv_clientconflict(struct nfsclient *clp, int *haslockp, struct
101     vnode *vp, NFSPROC_T *p);
102 static int nfsrv_delegconflict(struct nfsstate *stp, int *haslockp,
103     NFSPROC_T *p, vnode_t vp);
104 static int nfsrv_cleandeleg(vnode_t vp, struct nfslockfile *lfp,
105     struct nfsclient *clp, int *haslockp, NFSPROC_T *p);
106 static int nfsrv_notsamecredname(struct nfsrv_descript *nd,
107     struct nfsclient *clp);
108 static time_t nfsrv_leaseexpiry(void);
109 static void nfsrv_delaydelegtimeout(struct nfsstate *stp);
110 static int nfsrv_checkseqid(struct nfsrv_descript *nd, u_int32_t seqid,
111     struct nfsstate *stp, struct nfsrvcache *op);
112 static int nfsrv_nootherstate(struct nfsstate *stp);
113 static int nfsrv_locallock(vnode_t vp, struct nfslockfile *lfp, int flags,
114     uint64_t first, uint64_t end, struct nfslockconflict *cfp, NFSPROC_T *p);
115 static void nfsrv_localunlock(vnode_t vp, struct nfslockfile *lfp,
116     uint64_t init_first, uint64_t init_end, NFSPROC_T *p);
117 static int nfsrv_dolocal(vnode_t vp, struct nfslockfile *lfp, int flags,
118     int oldflags, uint64_t first, uint64_t end, struct nfslockconflict *cfp,
119     NFSPROC_T *p);
120 static void nfsrv_locallock_rollback(vnode_t vp, struct nfslockfile *lfp,
121     NFSPROC_T *p);
122 static void nfsrv_locallock_commit(struct nfslockfile *lfp, int flags,
123     uint64_t first, uint64_t end);
124 static void nfsrv_locklf(struct nfslockfile *lfp);
125 static void nfsrv_unlocklf(struct nfslockfile *lfp);
126 
127 /*
128  * Scan the client list for a match and either return the current one,
129  * create a new entry or return an error.
130  * If returning a non-error, the clp structure must either be linked into
131  * the client list or free'd.
132  */
133 APPLESTATIC int
134 nfsrv_setclient(struct nfsrv_descript *nd, struct nfsclient **new_clpp,
135     nfsquad_t *clientidp, nfsquad_t *confirmp, NFSPROC_T *p)
136 {
137 	struct nfsclient *clp = NULL, *new_clp = *new_clpp;
138 	int i;
139 	struct nfsstate *stp, *tstp;
140 	struct sockaddr_in *sad, *rad;
141 	int zapit = 0, gotit, hasstate = 0, igotlock;
142 	static u_int64_t confirm_index = 0;
143 
144 	/*
145 	 * Check for state resource limit exceeded.
146 	 */
147 	if (nfsrv_openpluslock > NFSRV_V4STATELIMIT)
148 		return (NFSERR_RESOURCE);
149 
150 	if ((nd->nd_flag & ND_GSS) && nfsrv_nogsscallback)
151 		/*
152 		 * Don't do callbacks for AUTH_GSS.
153 		 * (Since these aren't yet debugged, they might cause the
154 		 *  server to crap out, if they get past the Init call to
155 		 *  the client.)
156 		 */
157 		new_clp->lc_program = 0;
158 
159 	/* Lock out other nfsd threads */
160 	NFSLOCKV4ROOTMUTEX();
161 	nfsv4_relref(&nfsv4rootfs_lock);
162 	do {
163 		igotlock = nfsv4_lock(&nfsv4rootfs_lock, 1, NULL,
164 		    NFSV4ROOTLOCKMUTEXPTR);
165 	} while (!igotlock);
166 	NFSUNLOCKV4ROOTMUTEX();
167 	NFSLOCKSTATE();	/* to avoid a race with */
168 	NFSUNLOCKSTATE();	/* nfsrv_servertimer() */
169 
170 	/*
171 	 * Search for a match in the client list.
172 	 */
173 	gotit = i = 0;
174 	while (i < NFSCLIENTHASHSIZE && !gotit) {
175 	    LIST_FOREACH(clp, &nfsclienthash[i], lc_hash) {
176 		if (new_clp->lc_idlen == clp->lc_idlen &&
177 		    !NFSBCMP(new_clp->lc_id, clp->lc_id, clp->lc_idlen)) {
178 			gotit = 1;
179 			break;
180 		}
181 	    }
182 	    i++;
183 	}
184 	if (!gotit ||
185 	    (clp->lc_flags & (LCL_NEEDSCONFIRM | LCL_ADMINREVOKED))) {
186 		/*
187 		 * Get rid of the old one.
188 		 */
189 		if (i != NFSCLIENTHASHSIZE) {
190 			LIST_REMOVE(clp, lc_hash);
191 			nfsrv_cleanclient(clp, p);
192 			nfsrv_freedeleglist(&clp->lc_deleg);
193 			nfsrv_freedeleglist(&clp->lc_olddeleg);
194 			zapit = 1;
195 		}
196 		/*
197 		 * Add it after assigning a client id to it.
198 		 */
199 		new_clp->lc_flags |= LCL_NEEDSCONFIRM;
200 		confirmp->qval = new_clp->lc_confirm.qval = ++confirm_index;
201 		clientidp->lval[0] = new_clp->lc_clientid.lval[0] =
202 		    (u_int32_t)nfsrvboottime;
203 		clientidp->lval[1] = new_clp->lc_clientid.lval[1] =
204 		    nfsrv_nextclientindex();
205 		new_clp->lc_stateindex = 0;
206 		new_clp->lc_statemaxindex = 0;
207 		new_clp->lc_cbref = 0;
208 		new_clp->lc_expiry = nfsrv_leaseexpiry();
209 		LIST_INIT(&new_clp->lc_open);
210 		LIST_INIT(&new_clp->lc_deleg);
211 		LIST_INIT(&new_clp->lc_olddeleg);
212 		for (i = 0; i < NFSSTATEHASHSIZE; i++)
213 			LIST_INIT(&new_clp->lc_stateid[i]);
214 		LIST_INSERT_HEAD(NFSCLIENTHASH(new_clp->lc_clientid), new_clp,
215 		    lc_hash);
216 		newnfsstats.srvclients++;
217 		nfsrv_openpluslock++;
218 		nfsrv_clients++;
219 		NFSLOCKV4ROOTMUTEX();
220 		nfsv4_unlock(&nfsv4rootfs_lock, 1);
221 		NFSUNLOCKV4ROOTMUTEX();
222 		if (zapit)
223 			nfsrv_zapclient(clp, p);
224 		*new_clpp = NULL;
225 		return (0);
226 	}
227 
228 	/*
229 	 * Now, handle the cases where the id is already issued.
230 	 */
231 	if (nfsrv_notsamecredname(nd, clp)) {
232 	    /*
233 	     * Check to see if there is expired state that should go away.
234 	     */
235 	    if (clp->lc_expiry < NFSD_MONOSEC &&
236 	        (!LIST_EMPTY(&clp->lc_open) || !LIST_EMPTY(&clp->lc_deleg))) {
237 		nfsrv_cleanclient(clp, p);
238 		nfsrv_freedeleglist(&clp->lc_deleg);
239 	    }
240 
241 	    /*
242 	     * If there is outstanding state, then reply NFSERR_CLIDINUSE per
243 	     * RFC3530 Sec. 8.1.2 last para.
244 	     */
245 	    if (!LIST_EMPTY(&clp->lc_deleg)) {
246 		hasstate = 1;
247 	    } else if (LIST_EMPTY(&clp->lc_open)) {
248 		hasstate = 0;
249 	    } else {
250 		hasstate = 0;
251 		/* Look for an Open on the OpenOwner */
252 		LIST_FOREACH(stp, &clp->lc_open, ls_list) {
253 		    if (!LIST_EMPTY(&stp->ls_open)) {
254 			hasstate = 1;
255 			break;
256 		    }
257 		}
258 	    }
259 	    if (hasstate) {
260 		/*
261 		 * If the uid doesn't match, return NFSERR_CLIDINUSE after
262 		 * filling out the correct ipaddr and portnum.
263 		 */
264 		sad = NFSSOCKADDR(new_clp->lc_req.nr_nam, struct sockaddr_in *);
265 		rad = NFSSOCKADDR(clp->lc_req.nr_nam, struct sockaddr_in *);
266 		sad->sin_addr.s_addr = rad->sin_addr.s_addr;
267 		sad->sin_port = rad->sin_port;
268 		NFSLOCKV4ROOTMUTEX();
269 		nfsv4_unlock(&nfsv4rootfs_lock, 1);
270 		NFSUNLOCKV4ROOTMUTEX();
271 		return (NFSERR_CLIDINUSE);
272 	    }
273 	}
274 
275 	if (NFSBCMP(new_clp->lc_verf, clp->lc_verf, NFSX_VERF)) {
276 		/*
277 		 * If the verifier has changed, the client has rebooted
278 		 * and a new client id is issued. The old state info
279 		 * can be thrown away once the SETCLIENTID_CONFIRM occurs.
280 		 */
281 		LIST_REMOVE(clp, lc_hash);
282 		new_clp->lc_flags |= LCL_NEEDSCONFIRM;
283 		confirmp->qval = new_clp->lc_confirm.qval = ++confirm_index;
284 		clientidp->lval[0] = new_clp->lc_clientid.lval[0] =
285 		    nfsrvboottime;
286 		clientidp->lval[1] = new_clp->lc_clientid.lval[1] =
287 		    nfsrv_nextclientindex();
288 		new_clp->lc_stateindex = 0;
289 		new_clp->lc_statemaxindex = 0;
290 		new_clp->lc_cbref = 0;
291 		new_clp->lc_expiry = nfsrv_leaseexpiry();
292 
293 		/*
294 		 * Save the state until confirmed.
295 		 */
296 		LIST_NEWHEAD(&new_clp->lc_open, &clp->lc_open, ls_list);
297 		LIST_FOREACH(tstp, &new_clp->lc_open, ls_list)
298 			tstp->ls_clp = new_clp;
299 		LIST_NEWHEAD(&new_clp->lc_deleg, &clp->lc_deleg, ls_list);
300 		LIST_FOREACH(tstp, &new_clp->lc_deleg, ls_list)
301 			tstp->ls_clp = new_clp;
302 		LIST_NEWHEAD(&new_clp->lc_olddeleg, &clp->lc_olddeleg,
303 		    ls_list);
304 		LIST_FOREACH(tstp, &new_clp->lc_olddeleg, ls_list)
305 			tstp->ls_clp = new_clp;
306 		for (i = 0; i < NFSSTATEHASHSIZE; i++) {
307 			LIST_NEWHEAD(&new_clp->lc_stateid[i],
308 			    &clp->lc_stateid[i], ls_hash);
309 			LIST_FOREACH(tstp, &new_clp->lc_stateid[i], ls_list)
310 				tstp->ls_clp = new_clp;
311 		}
312 		LIST_INSERT_HEAD(NFSCLIENTHASH(new_clp->lc_clientid), new_clp,
313 		    lc_hash);
314 		newnfsstats.srvclients++;
315 		nfsrv_openpluslock++;
316 		nfsrv_clients++;
317 		NFSLOCKV4ROOTMUTEX();
318 		nfsv4_unlock(&nfsv4rootfs_lock, 1);
319 		NFSUNLOCKV4ROOTMUTEX();
320 
321 		/*
322 		 * Must wait until any outstanding callback on the old clp
323 		 * completes.
324 		 */
325 		while (clp->lc_cbref) {
326 			clp->lc_flags |= LCL_WAKEUPWANTED;
327 			(void) tsleep((caddr_t)clp, PZERO - 1,
328 			    "nfsd clp", 10 * hz);
329 		}
330 		nfsrv_zapclient(clp, p);
331 		*new_clpp = NULL;
332 		return (0);
333 	}
334 	/*
335 	 * id and verifier match, so update the net address info
336 	 * and get rid of any existing callback authentication
337 	 * handle, so a new one will be acquired.
338 	 */
339 	LIST_REMOVE(clp, lc_hash);
340 	new_clp->lc_flags |= (LCL_NEEDSCONFIRM | LCL_DONTCLEAN);
341 	new_clp->lc_expiry = nfsrv_leaseexpiry();
342 	confirmp->qval = new_clp->lc_confirm.qval = ++confirm_index;
343 	clientidp->lval[0] = new_clp->lc_clientid.lval[0] =
344 	    clp->lc_clientid.lval[0];
345 	clientidp->lval[1] = new_clp->lc_clientid.lval[1] =
346 	    clp->lc_clientid.lval[1];
347 	new_clp->lc_delegtime = clp->lc_delegtime;
348 	new_clp->lc_stateindex = clp->lc_stateindex;
349 	new_clp->lc_statemaxindex = clp->lc_statemaxindex;
350 	new_clp->lc_cbref = 0;
351 	LIST_NEWHEAD(&new_clp->lc_open, &clp->lc_open, ls_list);
352 	LIST_FOREACH(tstp, &new_clp->lc_open, ls_list)
353 		tstp->ls_clp = new_clp;
354 	LIST_NEWHEAD(&new_clp->lc_deleg, &clp->lc_deleg, ls_list);
355 	LIST_FOREACH(tstp, &new_clp->lc_deleg, ls_list)
356 		tstp->ls_clp = new_clp;
357 	LIST_NEWHEAD(&new_clp->lc_olddeleg, &clp->lc_olddeleg, ls_list);
358 	LIST_FOREACH(tstp, &new_clp->lc_olddeleg, ls_list)
359 		tstp->ls_clp = new_clp;
360 	for (i = 0; i < NFSSTATEHASHSIZE; i++) {
361 		LIST_NEWHEAD(&new_clp->lc_stateid[i], &clp->lc_stateid[i],
362 		    ls_hash);
363 		LIST_FOREACH(tstp, &new_clp->lc_stateid[i], ls_list)
364 			tstp->ls_clp = new_clp;
365 	}
366 	LIST_INSERT_HEAD(NFSCLIENTHASH(new_clp->lc_clientid), new_clp,
367 	    lc_hash);
368 	newnfsstats.srvclients++;
369 	nfsrv_openpluslock++;
370 	nfsrv_clients++;
371 	NFSLOCKV4ROOTMUTEX();
372 	nfsv4_unlock(&nfsv4rootfs_lock, 1);
373 	NFSUNLOCKV4ROOTMUTEX();
374 
375 	/*
376 	 * Must wait until any outstanding callback on the old clp
377 	 * completes.
378 	 */
379 	while (clp->lc_cbref) {
380 		clp->lc_flags |= LCL_WAKEUPWANTED;
381 		(void) tsleep((caddr_t)clp, PZERO - 1, "nfsd clp", 10 * hz);
382 	}
383 	nfsrv_zapclient(clp, p);
384 	*new_clpp = NULL;
385 	return (0);
386 }
387 
388 /*
389  * Check to see if the client id exists and optionally confirm it.
390  */
391 APPLESTATIC int
392 nfsrv_getclient(nfsquad_t clientid, int opflags, struct nfsclient **clpp,
393     nfsquad_t confirm, struct nfsrv_descript *nd, NFSPROC_T *p)
394 {
395 	struct nfsclient *clp;
396 	struct nfsstate *stp;
397 	int i;
398 	struct nfsclienthashhead *hp;
399 	int error = 0, igotlock, doneok;
400 
401 	if (clpp)
402 		*clpp = NULL;
403 	if (nfsrvboottime != clientid.lval[0])
404 		return (NFSERR_STALECLIENTID);
405 
406 	/*
407 	 * If called with opflags == CLOPS_RENEW, the State Lock is
408 	 * already held. Otherwise, we need to get either that or,
409 	 * for the case of Confirm, lock out the nfsd threads.
410 	 */
411 	if (opflags & CLOPS_CONFIRM) {
412 		NFSLOCKV4ROOTMUTEX();
413 		nfsv4_relref(&nfsv4rootfs_lock);
414 		do {
415 			igotlock = nfsv4_lock(&nfsv4rootfs_lock, 1, NULL,
416 			    NFSV4ROOTLOCKMUTEXPTR);
417 		} while (!igotlock);
418 		NFSUNLOCKV4ROOTMUTEX();
419 		NFSLOCKSTATE();	/* to avoid a race with */
420 		NFSUNLOCKSTATE();	/* nfsrv_servertimer() */
421 	} else if (opflags != CLOPS_RENEW) {
422 		NFSLOCKSTATE();
423 	}
424 
425 	hp = NFSCLIENTHASH(clientid);
426 	LIST_FOREACH(clp, hp, lc_hash) {
427 		if (clp->lc_clientid.lval[1] == clientid.lval[1])
428 			break;
429 	}
430 	if (clp == LIST_END(hp)) {
431 		if (opflags & CLOPS_CONFIRM)
432 			error = NFSERR_STALECLIENTID;
433 		else
434 			error = NFSERR_EXPIRED;
435 	} else if (clp->lc_flags & LCL_ADMINREVOKED) {
436 		/*
437 		 * If marked admin revoked, just return the error.
438 		 */
439 		error = NFSERR_ADMINREVOKED;
440 	}
441 	if (error) {
442 		if (opflags & CLOPS_CONFIRM) {
443 			NFSLOCKV4ROOTMUTEX();
444 			nfsv4_unlock(&nfsv4rootfs_lock, 1);
445 			NFSUNLOCKV4ROOTMUTEX();
446 		} else if (opflags != CLOPS_RENEW) {
447 			NFSUNLOCKSTATE();
448 		}
449 		return (error);
450 	}
451 
452 	/*
453 	 * Perform any operations specified by the opflags.
454 	 */
455 	if (opflags & CLOPS_CONFIRM) {
456 		if (clp->lc_confirm.qval != confirm.qval)
457 			error = NFSERR_STALECLIENTID;
458 		else if (nfsrv_notsamecredname(nd, clp))
459 			error = NFSERR_CLIDINUSE;
460 
461 		if (!error) {
462 		    if ((clp->lc_flags & (LCL_NEEDSCONFIRM | LCL_DONTCLEAN)) ==
463 			LCL_NEEDSCONFIRM) {
464 			/*
465 			 * Hang onto the delegations (as old delegations)
466 			 * for an Open with CLAIM_DELEGATE_PREV unless in
467 			 * grace, but get rid of the rest of the state.
468 			 */
469 			nfsrv_cleanclient(clp, p);
470 			nfsrv_freedeleglist(&clp->lc_olddeleg);
471 			if (nfsrv_checkgrace(0)) {
472 			    /* In grace, so just delete delegations */
473 			    nfsrv_freedeleglist(&clp->lc_deleg);
474 			} else {
475 			    LIST_FOREACH(stp, &clp->lc_deleg, ls_list)
476 				stp->ls_flags |= NFSLCK_OLDDELEG;
477 			    clp->lc_delegtime = NFSD_MONOSEC +
478 				nfsrv_lease + NFSRV_LEASEDELTA;
479 			    LIST_NEWHEAD(&clp->lc_olddeleg, &clp->lc_deleg,
480 				ls_list);
481 			}
482 		    }
483 		    clp->lc_flags &= ~(LCL_NEEDSCONFIRM | LCL_DONTCLEAN);
484 		    if (clp->lc_program)
485 			clp->lc_flags |= LCL_NEEDSCBNULL;
486 		}
487 	} else if (clp->lc_flags & LCL_NEEDSCONFIRM) {
488 		error = NFSERR_EXPIRED;
489 	}
490 
491 	/*
492 	 * If called by the Renew Op, we must check the principal.
493 	 */
494 	if (!error && (opflags & CLOPS_RENEWOP)) {
495 	    if (nfsrv_notsamecredname(nd, clp)) {
496 		doneok = 0;
497 		for (i = 0; i < NFSSTATEHASHSIZE && doneok == 0; i++) {
498 		    LIST_FOREACH(stp, &clp->lc_stateid[i], ls_hash) {
499 			if ((stp->ls_flags & NFSLCK_OPEN) &&
500 			    stp->ls_uid == nd->nd_cred->cr_uid) {
501 				doneok = 1;
502 				break;
503 			}
504 		    }
505 		}
506 		if (!doneok)
507 			error = NFSERR_ACCES;
508 	    }
509 	    if (!error && (clp->lc_flags & LCL_CBDOWN))
510 		error = NFSERR_CBPATHDOWN;
511 	}
512 	if ((!error || error == NFSERR_CBPATHDOWN) &&
513 	     (opflags & CLOPS_RENEW)) {
514 		clp->lc_expiry = nfsrv_leaseexpiry();
515 	}
516 	if (opflags & CLOPS_CONFIRM) {
517 		NFSLOCKV4ROOTMUTEX();
518 		nfsv4_unlock(&nfsv4rootfs_lock, 1);
519 		NFSUNLOCKV4ROOTMUTEX();
520 	} else if (opflags != CLOPS_RENEW) {
521 		NFSUNLOCKSTATE();
522 	}
523 	if (clpp)
524 		*clpp = clp;
525 	return (error);
526 }
527 
528 /*
529  * Called from the new nfssvc syscall to admin revoke a clientid.
530  * Returns 0 for success, error otherwise.
531  */
532 APPLESTATIC int
533 nfsrv_adminrevoke(struct nfsd_clid *revokep, NFSPROC_T *p)
534 {
535 	struct nfsclient *clp = NULL;
536 	int i;
537 	int gotit, igotlock;
538 
539 	/*
540 	 * First, lock out the nfsd so that state won't change while the
541 	 * revocation record is being written to the stable storage restart
542 	 * file.
543 	 */
544 	NFSLOCKV4ROOTMUTEX();
545 	nfsv4_relref(&nfsv4rootfs_lock);
546 	do {
547 		igotlock = nfsv4_lock(&nfsv4rootfs_lock, 1, NULL,
548 		    NFSV4ROOTLOCKMUTEXPTR);
549 	} while (!igotlock);
550 	NFSUNLOCKV4ROOTMUTEX();
551 	NFSLOCKSTATE();	/* to avoid a race with */
552 	NFSUNLOCKSTATE();	/* nfsrv_servertimer() */
553 
554 	/*
555 	 * Search for a match in the client list.
556 	 */
557 	gotit = i = 0;
558 	while (i < NFSCLIENTHASHSIZE && !gotit) {
559 	    LIST_FOREACH(clp, &nfsclienthash[i], lc_hash) {
560 		if (revokep->nclid_idlen == clp->lc_idlen &&
561 		    !NFSBCMP(revokep->nclid_id, clp->lc_id, clp->lc_idlen)) {
562 			gotit = 1;
563 			break;
564 		}
565 	    }
566 	    i++;
567 	}
568 	if (!gotit) {
569 		NFSLOCKV4ROOTMUTEX();
570 		nfsv4_unlock(&nfsv4rootfs_lock, 0);
571 		NFSUNLOCKV4ROOTMUTEX();
572 		return (EPERM);
573 	}
574 
575 	/*
576 	 * Now, write out the revocation record
577 	 */
578 	nfsrv_writestable(clp->lc_id, clp->lc_idlen, NFSNST_REVOKE, p);
579 
580 	/*
581 	 * and clear out the state, marking the clientid revoked.
582 	 */
583 	clp->lc_flags &= ~LCL_CALLBACKSON;
584 	clp->lc_flags |= LCL_ADMINREVOKED;
585 	nfsrv_cleanclient(clp, p);
586 	nfsrv_freedeleglist(&clp->lc_deleg);
587 	nfsrv_freedeleglist(&clp->lc_olddeleg);
588 	NFSLOCKV4ROOTMUTEX();
589 	nfsv4_unlock(&nfsv4rootfs_lock, 0);
590 	NFSUNLOCKV4ROOTMUTEX();
591 	return (0);
592 }
593 
594 /*
595  * Dump out stats for all clients. Called from nfssvc(2), that is used
596  * newnfsstats.
597  */
598 APPLESTATIC void
599 nfsrv_dumpclients(struct nfsd_dumpclients *dumpp, int maxcnt)
600 {
601 	struct nfsclient *clp;
602 	int i = 0, cnt = 0;
603 
604 	NFSLOCKSTATE();
605 	/*
606 	 * Rattle through the client lists until done.
607 	 */
608 	while (i < NFSCLIENTHASHSIZE && cnt < maxcnt) {
609 	    clp = LIST_FIRST(&nfsclienthash[i]);
610 	    while (clp != LIST_END(&nfsclienthash[i]) && cnt < maxcnt) {
611 		nfsrv_dumpaclient(clp, &dumpp[cnt]);
612 		cnt++;
613 		clp = LIST_NEXT(clp, lc_hash);
614 	    }
615 	    i++;
616 	}
617 	if (cnt < maxcnt)
618 	    dumpp[cnt].ndcl_clid.nclid_idlen = 0;
619 	NFSUNLOCKSTATE();
620 }
621 
622 /*
623  * Dump stats for a client. Must be called with the NFSSTATELOCK and spl'd.
624  */
625 static void
626 nfsrv_dumpaclient(struct nfsclient *clp, struct nfsd_dumpclients *dumpp)
627 {
628 	struct nfsstate *stp, *openstp, *lckownstp;
629 	struct nfslock *lop;
630 	struct sockaddr *sad;
631 	struct sockaddr_in *rad;
632 	struct sockaddr_in6 *rad6;
633 
634 	dumpp->ndcl_nopenowners = dumpp->ndcl_nlockowners = 0;
635 	dumpp->ndcl_nopens = dumpp->ndcl_nlocks = 0;
636 	dumpp->ndcl_ndelegs = dumpp->ndcl_nolddelegs = 0;
637 	dumpp->ndcl_flags = clp->lc_flags;
638 	dumpp->ndcl_clid.nclid_idlen = clp->lc_idlen;
639 	NFSBCOPY(clp->lc_id, dumpp->ndcl_clid.nclid_id, clp->lc_idlen);
640 	sad = NFSSOCKADDR(clp->lc_req.nr_nam, struct sockaddr *);
641 	dumpp->ndcl_addrfam = sad->sa_family;
642 	if (sad->sa_family == AF_INET) {
643 		rad = (struct sockaddr_in *)sad;
644 		dumpp->ndcl_cbaddr.sin_addr = rad->sin_addr;
645 	} else {
646 		rad6 = (struct sockaddr_in6 *)sad;
647 		dumpp->ndcl_cbaddr.sin6_addr = rad6->sin6_addr;
648 	}
649 
650 	/*
651 	 * Now, scan the state lists and total up the opens and locks.
652 	 */
653 	LIST_FOREACH(stp, &clp->lc_open, ls_list) {
654 	    dumpp->ndcl_nopenowners++;
655 	    LIST_FOREACH(openstp, &stp->ls_open, ls_list) {
656 		dumpp->ndcl_nopens++;
657 		LIST_FOREACH(lckownstp, &openstp->ls_open, ls_list) {
658 		    dumpp->ndcl_nlockowners++;
659 		    LIST_FOREACH(lop, &lckownstp->ls_lock, lo_lckowner) {
660 			dumpp->ndcl_nlocks++;
661 		    }
662 		}
663 	    }
664 	}
665 
666 	/*
667 	 * and the delegation lists.
668 	 */
669 	LIST_FOREACH(stp, &clp->lc_deleg, ls_list) {
670 	    dumpp->ndcl_ndelegs++;
671 	}
672 	LIST_FOREACH(stp, &clp->lc_olddeleg, ls_list) {
673 	    dumpp->ndcl_nolddelegs++;
674 	}
675 }
676 
677 /*
678  * Dump out lock stats for a file.
679  */
680 APPLESTATIC void
681 nfsrv_dumplocks(vnode_t vp, struct nfsd_dumplocks *ldumpp, int maxcnt,
682     NFSPROC_T *p)
683 {
684 	struct nfsstate *stp;
685 	struct nfslock *lop;
686 	int cnt = 0;
687 	struct nfslockfile *lfp;
688 	struct sockaddr *sad;
689 	struct sockaddr_in *rad;
690 	struct sockaddr_in6 *rad6;
691 	int ret;
692 	fhandle_t nfh;
693 
694 	ret = nfsrv_getlockfh(vp, 0, NULL, &nfh, p);
695 	NFSLOCKSTATE();
696 	if (!ret)
697 		ret = nfsrv_getlockfile(0, NULL, &lfp, &nfh, 0);
698 	if (ret) {
699 		ldumpp[0].ndlck_clid.nclid_idlen = 0;
700 		NFSUNLOCKSTATE();
701 		return;
702 	}
703 
704 	/*
705 	 * For each open share on file, dump it out.
706 	 */
707 	stp = LIST_FIRST(&lfp->lf_open);
708 	while (stp != LIST_END(&lfp->lf_open) && cnt < maxcnt) {
709 		ldumpp[cnt].ndlck_flags = stp->ls_flags;
710 		ldumpp[cnt].ndlck_stateid.seqid = stp->ls_stateid.seqid;
711 		ldumpp[cnt].ndlck_stateid.other[0] = stp->ls_stateid.other[0];
712 		ldumpp[cnt].ndlck_stateid.other[1] = stp->ls_stateid.other[1];
713 		ldumpp[cnt].ndlck_stateid.other[2] = stp->ls_stateid.other[2];
714 		ldumpp[cnt].ndlck_owner.nclid_idlen =
715 		    stp->ls_openowner->ls_ownerlen;
716 		NFSBCOPY(stp->ls_openowner->ls_owner,
717 		    ldumpp[cnt].ndlck_owner.nclid_id,
718 		    stp->ls_openowner->ls_ownerlen);
719 		ldumpp[cnt].ndlck_clid.nclid_idlen = stp->ls_clp->lc_idlen;
720 		NFSBCOPY(stp->ls_clp->lc_id, ldumpp[cnt].ndlck_clid.nclid_id,
721 		    stp->ls_clp->lc_idlen);
722 		sad=NFSSOCKADDR(stp->ls_clp->lc_req.nr_nam, struct sockaddr *);
723 		ldumpp[cnt].ndlck_addrfam = sad->sa_family;
724 		if (sad->sa_family == AF_INET) {
725 			rad = (struct sockaddr_in *)sad;
726 			ldumpp[cnt].ndlck_cbaddr.sin_addr = rad->sin_addr;
727 		} else {
728 			rad6 = (struct sockaddr_in6 *)sad;
729 			ldumpp[cnt].ndlck_cbaddr.sin6_addr = rad6->sin6_addr;
730 		}
731 		stp = LIST_NEXT(stp, ls_file);
732 		cnt++;
733 	}
734 
735 	/*
736 	 * and all locks.
737 	 */
738 	lop = LIST_FIRST(&lfp->lf_lock);
739 	while (lop != LIST_END(&lfp->lf_lock) && cnt < maxcnt) {
740 		stp = lop->lo_stp;
741 		ldumpp[cnt].ndlck_flags = lop->lo_flags;
742 		ldumpp[cnt].ndlck_first = lop->lo_first;
743 		ldumpp[cnt].ndlck_end = lop->lo_end;
744 		ldumpp[cnt].ndlck_stateid.seqid = stp->ls_stateid.seqid;
745 		ldumpp[cnt].ndlck_stateid.other[0] = stp->ls_stateid.other[0];
746 		ldumpp[cnt].ndlck_stateid.other[1] = stp->ls_stateid.other[1];
747 		ldumpp[cnt].ndlck_stateid.other[2] = stp->ls_stateid.other[2];
748 		ldumpp[cnt].ndlck_owner.nclid_idlen = stp->ls_ownerlen;
749 		NFSBCOPY(stp->ls_owner, ldumpp[cnt].ndlck_owner.nclid_id,
750 		    stp->ls_ownerlen);
751 		ldumpp[cnt].ndlck_clid.nclid_idlen = stp->ls_clp->lc_idlen;
752 		NFSBCOPY(stp->ls_clp->lc_id, ldumpp[cnt].ndlck_clid.nclid_id,
753 		    stp->ls_clp->lc_idlen);
754 		sad=NFSSOCKADDR(stp->ls_clp->lc_req.nr_nam, struct sockaddr *);
755 		ldumpp[cnt].ndlck_addrfam = sad->sa_family;
756 		if (sad->sa_family == AF_INET) {
757 			rad = (struct sockaddr_in *)sad;
758 			ldumpp[cnt].ndlck_cbaddr.sin_addr = rad->sin_addr;
759 		} else {
760 			rad6 = (struct sockaddr_in6 *)sad;
761 			ldumpp[cnt].ndlck_cbaddr.sin6_addr = rad6->sin6_addr;
762 		}
763 		lop = LIST_NEXT(lop, lo_lckfile);
764 		cnt++;
765 	}
766 
767 	/*
768 	 * and the delegations.
769 	 */
770 	stp = LIST_FIRST(&lfp->lf_deleg);
771 	while (stp != LIST_END(&lfp->lf_deleg) && cnt < maxcnt) {
772 		ldumpp[cnt].ndlck_flags = stp->ls_flags;
773 		ldumpp[cnt].ndlck_stateid.seqid = stp->ls_stateid.seqid;
774 		ldumpp[cnt].ndlck_stateid.other[0] = stp->ls_stateid.other[0];
775 		ldumpp[cnt].ndlck_stateid.other[1] = stp->ls_stateid.other[1];
776 		ldumpp[cnt].ndlck_stateid.other[2] = stp->ls_stateid.other[2];
777 		ldumpp[cnt].ndlck_owner.nclid_idlen = 0;
778 		ldumpp[cnt].ndlck_clid.nclid_idlen = stp->ls_clp->lc_idlen;
779 		NFSBCOPY(stp->ls_clp->lc_id, ldumpp[cnt].ndlck_clid.nclid_id,
780 		    stp->ls_clp->lc_idlen);
781 		sad=NFSSOCKADDR(stp->ls_clp->lc_req.nr_nam, struct sockaddr *);
782 		ldumpp[cnt].ndlck_addrfam = sad->sa_family;
783 		if (sad->sa_family == AF_INET) {
784 			rad = (struct sockaddr_in *)sad;
785 			ldumpp[cnt].ndlck_cbaddr.sin_addr = rad->sin_addr;
786 		} else {
787 			rad6 = (struct sockaddr_in6 *)sad;
788 			ldumpp[cnt].ndlck_cbaddr.sin6_addr = rad6->sin6_addr;
789 		}
790 		stp = LIST_NEXT(stp, ls_file);
791 		cnt++;
792 	}
793 
794 	/*
795 	 * If list isn't full, mark end of list by setting the client name
796 	 * to zero length.
797 	 */
798 	if (cnt < maxcnt)
799 		ldumpp[cnt].ndlck_clid.nclid_idlen = 0;
800 	NFSUNLOCKSTATE();
801 }
802 
803 /*
804  * Server timer routine. It can scan any linked list, so long
805  * as it holds the spin lock and there is no exclusive lock on
806  * nfsv4rootfs_lock.
807  * Must be called by a kernel thread and not a timer interrupt,
808  * so that it only runs when the nfsd threads are sleeping on a
809  * uniprocessor and uses the State spin lock for an SMP system.
810  * (For OpenBSD, a kthread is ok. For FreeBSD, I think it is ok
811  *  to do this from a callout, since the spin locks work. For
812  *  Darwin, I'm not sure what will work correctly yet.)
813  * Should be called once per second.
814  */
815 APPLESTATIC void
816 nfsrv_servertimer(void)
817 {
818 	struct nfsclient *clp, *nclp;
819 	struct nfsstate *stp, *nstp;
820 	int i;
821 
822 	/*
823 	 * Make sure nfsboottime is set. This is used by V3 as well
824 	 * as V4. Note that nfsboottime is not nfsrvboottime, which is
825 	 * only used by the V4 server for leases.
826 	 */
827 	if (nfsboottime.tv_sec == 0)
828 		NFSSETBOOTTIME(nfsboottime);
829 
830 	/*
831 	 * If server hasn't started yet, just return.
832 	 */
833 	NFSLOCKSTATE();
834 	if (nfsrv_stablefirst.nsf_eograce == 0) {
835 		NFSUNLOCKSTATE();
836 		return;
837 	}
838 	if (!(nfsrv_stablefirst.nsf_flags & NFSNSF_UPDATEDONE)) {
839 		if (!(nfsrv_stablefirst.nsf_flags & NFSNSF_GRACEOVER) &&
840 		    NFSD_MONOSEC > nfsrv_stablefirst.nsf_eograce)
841 			nfsrv_stablefirst.nsf_flags |=
842 			    (NFSNSF_GRACEOVER | NFSNSF_NEEDLOCK);
843 		NFSUNLOCKSTATE();
844 		return;
845 	}
846 
847 	/*
848 	 * Return now if an nfsd thread has the exclusive lock on
849 	 * nfsv4rootfs_lock. The dirty trick here is that we have
850 	 * the spin lock already and the nfsd threads do a:
851 	 * NFSLOCKSTATE, NFSUNLOCKSTATE after getting the exclusive
852 	 * lock, so they won't race with code after this check.
853 	 */
854 	if (nfsv4rootfs_lock.nfslock_lock & NFSV4LOCK_LOCK) {
855 		NFSUNLOCKSTATE();
856 		return;
857 	}
858 
859 	/*
860 	 * For each client...
861 	 */
862 	for (i = 0; i < NFSCLIENTHASHSIZE; i++) {
863 	    clp = LIST_FIRST(&nfsclienthash[i]);
864 	    while (clp != LIST_END(&nfsclienthash[i])) {
865 		nclp = LIST_NEXT(clp, lc_hash);
866 		if (!(clp->lc_flags & LCL_EXPIREIT)) {
867 		    if (((clp->lc_expiry + NFSRV_STALELEASE) < NFSD_MONOSEC
868 			 && ((LIST_EMPTY(&clp->lc_deleg)
869 			      && LIST_EMPTY(&clp->lc_open)) ||
870 			     nfsrv_clients > nfsrv_clienthighwater)) ||
871 			(clp->lc_expiry + NFSRV_MOULDYLEASE) < NFSD_MONOSEC ||
872 			(clp->lc_expiry < NFSD_MONOSEC &&
873 			 (nfsrv_openpluslock * 10 / 9) > NFSRV_V4STATELIMIT)) {
874 			/*
875 			 * Lease has expired several nfsrv_lease times ago:
876 			 * PLUS
877 			 *    - no state is associated with it
878 			 *    OR
879 			 *    - above high water mark for number of clients
880 			 *      (nfsrv_clienthighwater should be large enough
881 			 *       that this only occurs when clients fail to
882 			 *       use the same nfs_client_id4.id. Maybe somewhat
883 			 *       higher that the maximum number of clients that
884 			 *       will mount this server?)
885 			 * OR
886 			 * Lease has expired a very long time ago
887 			 * OR
888 			 * Lease has expired PLUS the number of opens + locks
889 			 * has exceeded 90% of capacity
890 			 *
891 			 * --> Mark for expiry. The actual expiry will be done
892 			 *     by an nfsd sometime soon.
893 			 */
894 			clp->lc_flags |= LCL_EXPIREIT;
895 			nfsrv_stablefirst.nsf_flags |=
896 			    (NFSNSF_NEEDLOCK | NFSNSF_EXPIREDCLIENT);
897 		    } else {
898 			/*
899 			 * If there are no opens, increment no open tick cnt
900 			 * If time exceeds NFSNOOPEN, mark it to be thrown away
901 			 * otherwise, if there is an open, reset no open time
902 			 * Hopefully, this will avoid excessive re-creation
903 			 * of open owners and subsequent open confirms.
904 			 */
905 			stp = LIST_FIRST(&clp->lc_open);
906 			while (stp != LIST_END(&clp->lc_open)) {
907 				nstp = LIST_NEXT(stp, ls_list);
908 				if (LIST_EMPTY(&stp->ls_open)) {
909 					stp->ls_noopens++;
910 					if (stp->ls_noopens > NFSNOOPEN ||
911 					    (nfsrv_openpluslock * 2) >
912 					    NFSRV_V4STATELIMIT)
913 						nfsrv_stablefirst.nsf_flags |=
914 							NFSNSF_NOOPENS;
915 				} else {
916 					stp->ls_noopens = 0;
917 				}
918 				stp = nstp;
919 			}
920 		    }
921 		}
922 		clp = nclp;
923 	    }
924 	}
925 	NFSUNLOCKSTATE();
926 }
927 
928 /*
929  * The following set of functions free up the various data structures.
930  */
931 /*
932  * Clear out all open/lock state related to this nfsclient.
933  * Caller must hold an exclusive lock on nfsv4rootfs_lock, so that
934  * there are no other active nfsd threads.
935  */
936 APPLESTATIC void
937 nfsrv_cleanclient(struct nfsclient *clp, NFSPROC_T *p)
938 {
939 	struct nfsstate *stp, *nstp;
940 
941 	LIST_FOREACH_SAFE(stp, &clp->lc_open, ls_list, nstp)
942 		nfsrv_freeopenowner(stp, 1, p);
943 }
944 
945 /*
946  * Free a client that has been cleaned. It should also already have been
947  * removed from the lists.
948  * (Just to be safe w.r.t. newnfs_disconnect(), call this function when
949  *  softclock interrupts are enabled.)
950  */
951 APPLESTATIC void
952 nfsrv_zapclient(struct nfsclient *clp, NFSPROC_T *p)
953 {
954 
955 #ifdef notyet
956 	if ((clp->lc_flags & (LCL_GSS | LCL_CALLBACKSON)) ==
957 	     (LCL_GSS | LCL_CALLBACKSON) &&
958 	    (clp->lc_hand.nfsh_flag & NFSG_COMPLETE) &&
959 	    clp->lc_handlelen > 0) {
960 		clp->lc_hand.nfsh_flag &= ~NFSG_COMPLETE;
961 		clp->lc_hand.nfsh_flag |= NFSG_DESTROYED;
962 		(void) nfsrv_docallback(clp, NFSV4PROC_CBNULL,
963 			NULL, 0, NULL, NULL, NULL, p);
964 	}
965 #endif
966 	newnfs_disconnect(&clp->lc_req);
967 	NFSSOCKADDRFREE(clp->lc_req.nr_nam);
968 	NFSFREEMUTEX(&clp->lc_req.nr_mtx);
969 	free((caddr_t)clp, M_NFSDCLIENT);
970 	NFSLOCKSTATE();
971 	newnfsstats.srvclients--;
972 	nfsrv_openpluslock--;
973 	nfsrv_clients--;
974 	NFSUNLOCKSTATE();
975 }
976 
977 /*
978  * Free a list of delegation state structures.
979  * (This function will also free all nfslockfile structures that no
980  *  longer have associated state.)
981  */
982 APPLESTATIC void
983 nfsrv_freedeleglist(struct nfsstatehead *sthp)
984 {
985 	struct nfsstate *stp, *nstp;
986 
987 	LIST_FOREACH_SAFE(stp, sthp, ls_list, nstp) {
988 		nfsrv_freedeleg(stp);
989 	}
990 	LIST_INIT(sthp);
991 }
992 
993 /*
994  * Free up a delegation.
995  */
996 static void
997 nfsrv_freedeleg(struct nfsstate *stp)
998 {
999 	struct nfslockfile *lfp;
1000 
1001 	LIST_REMOVE(stp, ls_hash);
1002 	LIST_REMOVE(stp, ls_list);
1003 	LIST_REMOVE(stp, ls_file);
1004 	lfp = stp->ls_lfp;
1005 	if (LIST_EMPTY(&lfp->lf_open) &&
1006 	    LIST_EMPTY(&lfp->lf_lock) && LIST_EMPTY(&lfp->lf_deleg) &&
1007 	    LIST_EMPTY(&lfp->lf_locallock) && LIST_EMPTY(&lfp->lf_rollback) &&
1008 	    lfp->lf_usecount == 0 &&
1009 	    nfsv4_testlock(&lfp->lf_locallock_lck) == 0)
1010 		nfsrv_freenfslockfile(lfp);
1011 	FREE((caddr_t)stp, M_NFSDSTATE);
1012 	newnfsstats.srvdelegates--;
1013 	nfsrv_openpluslock--;
1014 	nfsrv_delegatecnt--;
1015 }
1016 
1017 /*
1018  * This function frees an open owner and all associated opens.
1019  * Must be called with soft clock interrupts disabled.
1020  */
1021 static void
1022 nfsrv_freeopenowner(struct nfsstate *stp, int cansleep, NFSPROC_T *p)
1023 {
1024 	struct nfsstate *nstp, *tstp;
1025 
1026 	LIST_REMOVE(stp, ls_list);
1027 	/*
1028 	 * Now, free all associated opens.
1029 	 */
1030 	nstp = LIST_FIRST(&stp->ls_open);
1031 	while (nstp != LIST_END(&stp->ls_open)) {
1032 		tstp = nstp;
1033 		nstp = LIST_NEXT(nstp, ls_list);
1034 		(void) nfsrv_freeopen(tstp, NULL, cansleep, p);
1035 	}
1036 	if (stp->ls_op)
1037 		nfsrvd_derefcache(stp->ls_op);
1038 	FREE((caddr_t)stp, M_NFSDSTATE);
1039 	newnfsstats.srvopenowners--;
1040 	nfsrv_openpluslock--;
1041 }
1042 
1043 /*
1044  * This function frees an open (nfsstate open structure) with all associated
1045  * lock_owners and locks. It also frees the nfslockfile structure iff there
1046  * are no other opens on the file.
1047  * Returns 1 if it free'd the nfslockfile, 0 otherwise.
1048  */
1049 static int
1050 nfsrv_freeopen(struct nfsstate *stp, vnode_t vp, int cansleep, NFSPROC_T *p)
1051 {
1052 	struct nfsstate *nstp, *tstp;
1053 	struct nfslockfile *lfp;
1054 	int ret;
1055 
1056 	LIST_REMOVE(stp, ls_hash);
1057 	LIST_REMOVE(stp, ls_list);
1058 	LIST_REMOVE(stp, ls_file);
1059 
1060 	lfp = stp->ls_lfp;
1061 	/*
1062 	 * Now, free all lockowners associated with this open.
1063 	 */
1064 	LIST_FOREACH_SAFE(tstp, &stp->ls_open, ls_list, nstp)
1065 		nfsrv_freelockowner(tstp, vp, cansleep, p);
1066 
1067 	/*
1068 	 * The nfslockfile is freed here if there are no locks
1069 	 * associated with the open.
1070 	 * If there are locks associated with the open, the
1071 	 * nfslockfile structure can be freed via nfsrv_freelockowner().
1072 	 * (That is why the call must be here instead of after the loop.)
1073 	 */
1074 	if (lfp != NULL && LIST_EMPTY(&lfp->lf_open) &&
1075 	    LIST_EMPTY(&lfp->lf_deleg) && LIST_EMPTY(&lfp->lf_lock) &&
1076 	    LIST_EMPTY(&lfp->lf_locallock) && LIST_EMPTY(&lfp->lf_rollback) &&
1077 	    lfp->lf_usecount == 0 &&
1078 	    (cansleep != 0 || nfsv4_testlock(&lfp->lf_locallock_lck) == 0)) {
1079 		nfsrv_freenfslockfile(lfp);
1080 		ret = 1;
1081 	} else
1082 		ret = 0;
1083 	FREE((caddr_t)stp, M_NFSDSTATE);
1084 	newnfsstats.srvopens--;
1085 	nfsrv_openpluslock--;
1086 	return (ret);
1087 }
1088 
1089 /*
1090  * Frees a lockowner and all associated locks.
1091  */
1092 static void
1093 nfsrv_freelockowner(struct nfsstate *stp, vnode_t vp, int cansleep,
1094     NFSPROC_T *p)
1095 {
1096 
1097 	LIST_REMOVE(stp, ls_hash);
1098 	LIST_REMOVE(stp, ls_list);
1099 	nfsrv_freeallnfslocks(stp, vp, cansleep, p);
1100 	if (stp->ls_op)
1101 		nfsrvd_derefcache(stp->ls_op);
1102 	FREE((caddr_t)stp, M_NFSDSTATE);
1103 	newnfsstats.srvlockowners--;
1104 	nfsrv_openpluslock--;
1105 }
1106 
1107 /*
1108  * Free all the nfs locks on a lockowner.
1109  */
1110 static void
1111 nfsrv_freeallnfslocks(struct nfsstate *stp, vnode_t vp, int cansleep,
1112     NFSPROC_T *p)
1113 {
1114 	struct nfslock *lop, *nlop;
1115 	struct nfsrollback *rlp, *nrlp;
1116 	struct nfslockfile *lfp = NULL;
1117 	int gottvp = 0;
1118 	vnode_t tvp = NULL;
1119 
1120 	lop = LIST_FIRST(&stp->ls_lock);
1121 	while (lop != LIST_END(&stp->ls_lock)) {
1122 		nlop = LIST_NEXT(lop, lo_lckowner);
1123 		/*
1124 		 * Since all locks should be for the same file, lfp should
1125 		 * not change.
1126 		 */
1127 		if (lfp == NULL)
1128 			lfp = lop->lo_lfp;
1129 		else if (lfp != lop->lo_lfp)
1130 			panic("allnfslocks");
1131 		/*
1132 		 * If vp is NULL and cansleep != 0, a vnode must be acquired
1133 		 * from the file handle. This only occurs when called from
1134 		 * nfsrv_cleanclient().
1135 		 */
1136 		if (gottvp == 0) {
1137 			if (nfsrv_dolocallocks == 0)
1138 				tvp = NULL;
1139 			else if (vp == NULL && cansleep != 0)
1140 				tvp = nfsvno_getvp(&lfp->lf_fh);
1141 			else
1142 				tvp = vp;
1143 			gottvp = 1;
1144 		}
1145 
1146 		if (tvp != NULL) {
1147 			if (cansleep == 0)
1148 				panic("allnfs2");
1149 			nfsrv_localunlock(tvp, lfp, lop->lo_first,
1150 			    lop->lo_end, p);
1151 			LIST_FOREACH_SAFE(rlp, &lfp->lf_rollback, rlck_list,
1152 			    nrlp)
1153 				free(rlp, M_NFSDROLLBACK);
1154 			LIST_INIT(&lfp->lf_rollback);
1155 		}
1156 		nfsrv_freenfslock(lop);
1157 		lop = nlop;
1158 	}
1159 	if (vp == NULL && tvp != NULL)
1160 		vput(tvp);
1161 }
1162 
1163 /*
1164  * Free an nfslock structure.
1165  * Must be called with soft clock interrupts disabled.
1166  */
1167 static void
1168 nfsrv_freenfslock(struct nfslock *lop)
1169 {
1170 
1171 	if (lop->lo_lckfile.le_prev != NULL) {
1172 		LIST_REMOVE(lop, lo_lckfile);
1173 		newnfsstats.srvlocks--;
1174 		nfsrv_openpluslock--;
1175 	}
1176 	LIST_REMOVE(lop, lo_lckowner);
1177 	FREE((caddr_t)lop, M_NFSDLOCK);
1178 }
1179 
1180 /*
1181  * This function frees an nfslockfile structure.
1182  * Must be called with soft clock interrupts disabled.
1183  */
1184 static void
1185 nfsrv_freenfslockfile(struct nfslockfile *lfp)
1186 {
1187 
1188 	LIST_REMOVE(lfp, lf_hash);
1189 	FREE((caddr_t)lfp, M_NFSDLOCKFILE);
1190 }
1191 
1192 /*
1193  * This function looks up an nfsstate structure via stateid.
1194  */
1195 static int
1196 nfsrv_getstate(struct nfsclient *clp, nfsv4stateid_t *stateidp, __unused u_int32_t flags,
1197     struct nfsstate **stpp)
1198 {
1199 	struct nfsstate *stp;
1200 	struct nfsstatehead *hp;
1201 
1202 	*stpp = NULL;
1203 	hp = NFSSTATEHASH(clp, *stateidp);
1204 	LIST_FOREACH(stp, hp, ls_hash) {
1205 		if (!NFSBCMP(stp->ls_stateid.other, stateidp->other,
1206 			NFSX_STATEIDOTHER))
1207 			break;
1208 	}
1209 
1210 	/*
1211 	 * If no state id in list, return NFSERR_BADSTATEID.
1212 	 */
1213 	if (stp == LIST_END(hp))
1214 		return (NFSERR_BADSTATEID);
1215 	*stpp = stp;
1216 	return (0);
1217 }
1218 
1219 /*
1220  * This function gets an nfsstate structure via owner string.
1221  */
1222 static void
1223 nfsrv_getowner(struct nfsstatehead *hp, struct nfsstate *new_stp,
1224     struct nfsstate **stpp)
1225 {
1226 	struct nfsstate *stp;
1227 
1228 	*stpp = NULL;
1229 	LIST_FOREACH(stp, hp, ls_list) {
1230 		if (new_stp->ls_ownerlen == stp->ls_ownerlen &&
1231 		  !NFSBCMP(new_stp->ls_owner,stp->ls_owner,stp->ls_ownerlen)) {
1232 			*stpp = stp;
1233 			return;
1234 		}
1235 	}
1236 }
1237 
1238 /*
1239  * Lock control function called to update lock status.
1240  * Returns 0 upon success, -1 if there is no lock and the flags indicate
1241  * that one isn't to be created and an NFSERR_xxx for other errors.
1242  * The structures new_stp and new_lop are passed in as pointers that should
1243  * be set to NULL if the structure is used and shouldn't be free'd.
1244  * For the NFSLCK_TEST and NFSLCK_CHECK cases, the structures are
1245  * never used and can safely be allocated on the stack. For all other
1246  * cases, *new_stpp and *new_lopp should be malloc'd before the call,
1247  * in case they are used.
1248  */
1249 APPLESTATIC int
1250 nfsrv_lockctrl(vnode_t vp, struct nfsstate **new_stpp,
1251     struct nfslock **new_lopp, struct nfslockconflict *cfp,
1252     nfsquad_t clientid, nfsv4stateid_t *stateidp,
1253     __unused struct nfsexstuff *exp,
1254     struct nfsrv_descript *nd, NFSPROC_T *p)
1255 {
1256 	struct nfslock *lop;
1257 	struct nfsstate *new_stp = *new_stpp;
1258 	struct nfslock *new_lop = *new_lopp;
1259 	struct nfsstate *tstp, *mystp, *nstp;
1260 	int specialid = 0;
1261 	struct nfslockfile *lfp;
1262 	struct nfslock *other_lop = NULL;
1263 	struct nfsstate *stp, *lckstp = NULL;
1264 	struct nfsclient *clp = NULL;
1265 	u_int32_t bits;
1266 	int error = 0, haslock = 0, ret, reterr;
1267 	int getlckret, delegation = 0, filestruct_locked;
1268 	fhandle_t nfh;
1269 	uint64_t first, end;
1270 	uint32_t lock_flags;
1271 
1272 	if (new_stp->ls_flags & (NFSLCK_CHECK | NFSLCK_SETATTR)) {
1273 		/*
1274 		 * Note the special cases of "all 1s" or "all 0s" stateids and
1275 		 * let reads with all 1s go ahead.
1276 		 */
1277 		if (new_stp->ls_stateid.seqid == 0x0 &&
1278 		    new_stp->ls_stateid.other[0] == 0x0 &&
1279 		    new_stp->ls_stateid.other[1] == 0x0 &&
1280 		    new_stp->ls_stateid.other[2] == 0x0)
1281 			specialid = 1;
1282 		else if (new_stp->ls_stateid.seqid == 0xffffffff &&
1283 		    new_stp->ls_stateid.other[0] == 0xffffffff &&
1284 		    new_stp->ls_stateid.other[1] == 0xffffffff &&
1285 		    new_stp->ls_stateid.other[2] == 0xffffffff)
1286 			specialid = 2;
1287 	}
1288 
1289 	/*
1290 	 * Check for restart conditions (client and server).
1291 	 */
1292 	error = nfsrv_checkrestart(clientid, new_stp->ls_flags,
1293 	    &new_stp->ls_stateid, specialid);
1294 	if (error)
1295 		return (error);
1296 
1297 	/*
1298 	 * Check for state resource limit exceeded.
1299 	 */
1300 	if ((new_stp->ls_flags & NFSLCK_LOCK) &&
1301 	    nfsrv_openpluslock > NFSRV_V4STATELIMIT)
1302 		return (NFSERR_RESOURCE);
1303 
1304 	/*
1305 	 * For the lock case, get another nfslock structure,
1306 	 * just in case we need it.
1307 	 * Malloc now, before we start sifting through the linked lists,
1308 	 * in case we have to wait for memory.
1309 	 */
1310 tryagain:
1311 	if (new_stp->ls_flags & NFSLCK_LOCK)
1312 		MALLOC(other_lop, struct nfslock *, sizeof (struct nfslock),
1313 		    M_NFSDLOCK, M_WAITOK);
1314 	filestruct_locked = 0;
1315 	reterr = 0;
1316 	lfp = NULL;
1317 
1318 	/*
1319 	 * Get the lockfile structure for CFH now, so we can do a sanity
1320 	 * check against the stateid, before incrementing the seqid#, since
1321 	 * we want to return NFSERR_BADSTATEID on failure and the seqid#
1322 	 * shouldn't be incremented for this case.
1323 	 * If nfsrv_getlockfile() returns -1, it means "not found", which
1324 	 * will be handled later.
1325 	 * If we are doing Lock/LockU and local locking is enabled, sleep
1326 	 * lock the nfslockfile structure.
1327 	 */
1328 	getlckret = nfsrv_getlockfh(vp, new_stp->ls_flags, NULL, &nfh, p);
1329 	NFSLOCKSTATE();
1330 	if (getlckret == 0) {
1331 		if ((new_stp->ls_flags & (NFSLCK_LOCK | NFSLCK_UNLOCK)) != 0 &&
1332 		    nfsrv_dolocallocks != 0 && nd->nd_repstat == 0) {
1333 			getlckret = nfsrv_getlockfile(new_stp->ls_flags, NULL,
1334 			    &lfp, &nfh, 1);
1335 			if (getlckret == 0)
1336 				filestruct_locked = 1;
1337 		} else
1338 			getlckret = nfsrv_getlockfile(new_stp->ls_flags, NULL,
1339 			    &lfp, &nfh, 0);
1340 	}
1341 	if (getlckret != 0 && getlckret != -1)
1342 		reterr = getlckret;
1343 
1344 	if (filestruct_locked != 0) {
1345 		LIST_INIT(&lfp->lf_rollback);
1346 		if ((new_stp->ls_flags & NFSLCK_LOCK)) {
1347 			/*
1348 			 * For local locking, do the advisory locking now, so
1349 			 * that any conflict can be detected. A failure later
1350 			 * can be rolled back locally. If an error is returned,
1351 			 * struct nfslockfile has been unlocked and any local
1352 			 * locking rolled back.
1353 			 */
1354 			NFSUNLOCKSTATE();
1355 			reterr = nfsrv_locallock(vp, lfp,
1356 			    (new_lop->lo_flags & (NFSLCK_READ | NFSLCK_WRITE)),
1357 			    new_lop->lo_first, new_lop->lo_end, cfp, p);
1358 			NFSLOCKSTATE();
1359 		}
1360 	}
1361 
1362 	/*
1363 	 * Since the code is manipulating lists that are also
1364 	 * manipulated by nfsrv_servertimer(), soft clock interrupts
1365 	 * must be masked off.
1366 	 */
1367 	if (specialid == 0) {
1368 	    if (new_stp->ls_flags & NFSLCK_TEST) {
1369 		/*
1370 		 * RFC 3530 does not list LockT as an op that renews a
1371 		 * lease, but the concensus seems to be that it is ok
1372 		 * for a server to do so.
1373 		 */
1374 		error = nfsrv_getclient(clientid, CLOPS_RENEW, &clp,
1375 		    (nfsquad_t)((u_quad_t)0), NULL, p);
1376 
1377 		/*
1378 		 * Since NFSERR_EXPIRED, NFSERR_ADMINREVOKED are not valid
1379 		 * error returns for LockT, just go ahead and test for a lock,
1380 		 * since there are no locks for this client, but other locks
1381 		 * can conflict. (ie. same client will always be false)
1382 		 */
1383 		if (error == NFSERR_EXPIRED || error == NFSERR_ADMINREVOKED)
1384 		    error = 0;
1385 		lckstp = new_stp;
1386 	    } else {
1387 	      error = nfsrv_getclient(clientid, CLOPS_RENEW, &clp,
1388 		(nfsquad_t)((u_quad_t)0), NULL, p);
1389 	      if (error == 0)
1390 		/*
1391 		 * Look up the stateid
1392 		 */
1393 		error = nfsrv_getstate(clp, &new_stp->ls_stateid,
1394 		  new_stp->ls_flags, &stp);
1395 	      /*
1396 	       * do some sanity checks for an unconfirmed open or a
1397 	       * stateid that refers to the wrong file, for an open stateid
1398 	       */
1399 	      if (error == 0 && (stp->ls_flags & NFSLCK_OPEN) &&
1400 		  ((stp->ls_openowner->ls_flags & NFSLCK_NEEDSCONFIRM) ||
1401 		   (getlckret == 0 && stp->ls_lfp != lfp)))
1402 			error = NFSERR_BADSTATEID;
1403 	      if (error == 0 &&
1404 		  (stp->ls_flags & (NFSLCK_DELEGREAD | NFSLCK_DELEGWRITE)) &&
1405 		  getlckret == 0 && stp->ls_lfp != lfp)
1406 			error = NFSERR_BADSTATEID;
1407 
1408 	      /*
1409 	       * If the lockowner stateid doesn't refer to the same file,
1410 	       * I believe that is considered ok, since some clients will
1411 	       * only create a single lockowner and use that for all locks
1412 	       * on all files.
1413 	       * For now, log it as a diagnostic, instead of considering it
1414 	       * a BadStateid.
1415 	       */
1416 	      if (error == 0 && (stp->ls_flags &
1417 		  (NFSLCK_OPEN | NFSLCK_DELEGREAD | NFSLCK_DELEGWRITE)) == 0 &&
1418 		  getlckret == 0 && stp->ls_lfp != lfp) {
1419 #ifdef DIAGNOSTIC
1420 		  printf("Got a lock statid for different file open\n");
1421 #endif
1422 		  /*
1423 		  error = NFSERR_BADSTATEID;
1424 		  */
1425 	      }
1426 
1427 	      if (error == 0) {
1428 		    if (new_stp->ls_flags & NFSLCK_OPENTOLOCK) {
1429 			/*
1430 			 * If haslock set, we've already checked the seqid.
1431 			 */
1432 			if (!haslock) {
1433 			    if (stp->ls_flags & NFSLCK_OPEN)
1434 				error = nfsrv_checkseqid(nd, new_stp->ls_seq,
1435 				    stp->ls_openowner, new_stp->ls_op);
1436 			    else
1437 				error = NFSERR_BADSTATEID;
1438 			}
1439 			if (!error)
1440 			    nfsrv_getowner(&stp->ls_open, new_stp, &lckstp);
1441 			if (lckstp)
1442 			    /*
1443 			     * I believe this should be an error, but it
1444 			     * isn't obvious what NFSERR_xxx would be
1445 			     * appropriate, so I'll use NFSERR_INVAL for now.
1446 			     */
1447 			    error = NFSERR_INVAL;
1448 			else
1449 			    lckstp = new_stp;
1450 		    } else if (new_stp->ls_flags&(NFSLCK_LOCK|NFSLCK_UNLOCK)) {
1451 			/*
1452 			 * If haslock set, ditto above.
1453 			 */
1454 			if (!haslock) {
1455 			    if (stp->ls_flags & NFSLCK_OPEN)
1456 				error = NFSERR_BADSTATEID;
1457 			    else
1458 				error = nfsrv_checkseqid(nd, new_stp->ls_seq,
1459 				    stp, new_stp->ls_op);
1460 			}
1461 			lckstp = stp;
1462 		    } else {
1463 			lckstp = stp;
1464 		    }
1465 	      }
1466 	      /*
1467 	       * If the seqid part of the stateid isn't the same, return
1468 	       * NFSERR_OLDSTATEID for cases other than I/O Ops.
1469 	       * For I/O Ops, only return NFSERR_OLDSTATEID if
1470 	       * nfsrv_returnoldstateid is set. (The concensus on the email
1471 	       * list was that most clients would prefer to not receive
1472 	       * NFSERR_OLDSTATEID for I/O Ops, but the RFC suggests that that
1473 	       * is what will happen, so I use the nfsrv_returnoldstateid to
1474 	       * allow for either server configuration.)
1475 	       */
1476 	      if (!error && stp->ls_stateid.seqid!=new_stp->ls_stateid.seqid &&
1477 		  (!(new_stp->ls_flags & NFSLCK_CHECK) ||
1478 		   nfsrv_returnoldstateid))
1479 		    error = NFSERR_OLDSTATEID;
1480 	    }
1481 	}
1482 
1483 	/*
1484 	 * Now we can check for grace.
1485 	 */
1486 	if (!error)
1487 		error = nfsrv_checkgrace(new_stp->ls_flags);
1488 	if ((new_stp->ls_flags & NFSLCK_RECLAIM) && !error &&
1489 		nfsrv_checkstable(clp))
1490 		error = NFSERR_NOGRACE;
1491 	/*
1492 	 * If we successfully Reclaimed state, note that.
1493 	 */
1494 	if ((new_stp->ls_flags & NFSLCK_RECLAIM) && !error)
1495 		nfsrv_markstable(clp);
1496 
1497 	/*
1498 	 * At this point, either error == NFSERR_BADSTATEID or the
1499 	 * seqid# has been updated, so we can return any error.
1500 	 * If error == 0, there may be an error in:
1501 	 *    nd_repstat - Set by the calling function.
1502 	 *    reterr - Set above, if getting the nfslockfile structure
1503 	 *       or acquiring the local lock failed.
1504 	 *    (If both of these are set, nd_repstat should probably be
1505 	 *     returned, since that error was detected before this
1506 	 *     function call.)
1507 	 */
1508 	if (error != 0 || nd->nd_repstat != 0 || reterr != 0) {
1509 		if (error == 0) {
1510 			if (nd->nd_repstat != 0)
1511 				error = nd->nd_repstat;
1512 			else
1513 				error = reterr;
1514 		}
1515 		if (filestruct_locked != 0) {
1516 			/* Roll back local locks. */
1517 			NFSUNLOCKSTATE();
1518 			nfsrv_locallock_rollback(vp, lfp, p);
1519 			NFSLOCKSTATE();
1520 			nfsrv_unlocklf(lfp);
1521 		}
1522 		NFSUNLOCKSTATE();
1523 		if (other_lop)
1524 			FREE((caddr_t)other_lop, M_NFSDLOCK);
1525 		if (haslock) {
1526 			NFSLOCKV4ROOTMUTEX();
1527 			nfsv4_unlock(&nfsv4rootfs_lock, 1);
1528 			NFSUNLOCKV4ROOTMUTEX();
1529 		}
1530 		return (error);
1531 	}
1532 
1533 	/*
1534 	 * Check the nfsrv_getlockfile return.
1535 	 * Returned -1 if no structure found.
1536 	 */
1537 	if (getlckret == -1) {
1538 		error = NFSERR_EXPIRED;
1539 		/*
1540 		 * Called from lockt, so no lock is OK.
1541 		 */
1542 		if (new_stp->ls_flags & NFSLCK_TEST) {
1543 			error = 0;
1544 		} else if (new_stp->ls_flags &
1545 		    (NFSLCK_CHECK | NFSLCK_SETATTR)) {
1546 			/*
1547 			 * Called to check for a lock, OK if the stateid is all
1548 			 * 1s or all 0s, but there should be an nfsstate
1549 			 * otherwise.
1550 			 * (ie. If there is no open, I'll assume no share
1551 			 *  deny bits.)
1552 			 */
1553 			if (specialid)
1554 				error = 0;
1555 			else
1556 				error = NFSERR_BADSTATEID;
1557 		}
1558 		NFSUNLOCKSTATE();
1559 		if (haslock) {
1560 			NFSLOCKV4ROOTMUTEX();
1561 			nfsv4_unlock(&nfsv4rootfs_lock, 1);
1562 			NFSUNLOCKV4ROOTMUTEX();
1563 		}
1564 		/*
1565 		 * Called to lock or unlock, so the lock has gone away.
1566 		 */
1567 		return (error);
1568 	}
1569 
1570 	/*
1571 	 * For NFSLCK_CHECK and NFSLCK_LOCK, test for a share conflict.
1572 	 * For NFSLCK_CHECK, allow a read if write access is granted,
1573 	 * but check for a deny. For NFSLCK_LOCK, require correct access,
1574 	 * which implies a conflicting deny can't exist.
1575 	 */
1576 	if (new_stp->ls_flags & (NFSLCK_CHECK | NFSLCK_LOCK)) {
1577 	    /*
1578 	     * Four kinds of state id:
1579 	     * - specialid (all 0s or all 1s), only for NFSLCK_CHECK
1580 	     * - stateid for an open
1581 	     * - stateid for a delegation
1582 	     * - stateid for a lock owner
1583 	     */
1584 	    if (!specialid) {
1585 		if (stp->ls_flags & (NFSLCK_DELEGREAD | NFSLCK_DELEGWRITE)) {
1586 		    delegation = 1;
1587 		    mystp = stp;
1588 		    nfsrv_delaydelegtimeout(stp);
1589 	        } else if (stp->ls_flags & NFSLCK_OPEN) {
1590 		    mystp = stp;
1591 		} else {
1592 		    mystp = stp->ls_openstp;
1593 		}
1594 		/*
1595 		 * If locking or checking, require correct access
1596 		 * bit set.
1597 		 */
1598 		if (((new_stp->ls_flags & NFSLCK_LOCK) &&
1599 		     !((new_lop->lo_flags >> NFSLCK_LOCKSHIFT) &
1600 		       mystp->ls_flags & NFSLCK_ACCESSBITS)) ||
1601 		    ((new_stp->ls_flags & (NFSLCK_CHECK|NFSLCK_READACCESS)) ==
1602 		      (NFSLCK_CHECK | NFSLCK_READACCESS) &&
1603 		     !(mystp->ls_flags & NFSLCK_READACCESS)) ||
1604 		    ((new_stp->ls_flags & (NFSLCK_CHECK|NFSLCK_WRITEACCESS)) ==
1605 		      (NFSLCK_CHECK | NFSLCK_WRITEACCESS) &&
1606 		     !(mystp->ls_flags & NFSLCK_WRITEACCESS))) {
1607 			if (filestruct_locked != 0) {
1608 				/* Roll back local locks. */
1609 				NFSUNLOCKSTATE();
1610 				nfsrv_locallock_rollback(vp, lfp, p);
1611 				NFSLOCKSTATE();
1612 				nfsrv_unlocklf(lfp);
1613 			}
1614 			NFSUNLOCKSTATE();
1615 			if (other_lop)
1616 				FREE((caddr_t)other_lop, M_NFSDLOCK);
1617 			if (haslock) {
1618 				NFSLOCKV4ROOTMUTEX();
1619 				nfsv4_unlock(&nfsv4rootfs_lock, 1);
1620 				NFSUNLOCKV4ROOTMUTEX();
1621 			}
1622 			return (NFSERR_OPENMODE);
1623 		}
1624 	    } else
1625 		mystp = NULL;
1626 	    if ((new_stp->ls_flags & NFSLCK_CHECK) && !delegation) {
1627 		/*
1628 		 * Check for a conflicting deny bit.
1629 		 */
1630 		LIST_FOREACH(tstp, &lfp->lf_open, ls_file) {
1631 		    if (tstp != mystp) {
1632 			bits = tstp->ls_flags;
1633 			bits >>= NFSLCK_SHIFT;
1634 			if (new_stp->ls_flags & bits & NFSLCK_ACCESSBITS) {
1635 			    ret = nfsrv_clientconflict(tstp->ls_clp, &haslock,
1636 				vp, p);
1637 			    if (ret) {
1638 				/*
1639 				* nfsrv_clientconflict unlocks state
1640 				 * when it returns non-zero.
1641 				 */
1642 				lckstp = NULL;
1643 				goto tryagain;
1644 			    }
1645 			    NFSUNLOCKSTATE();
1646 			    if (haslock) {
1647 				NFSLOCKV4ROOTMUTEX();
1648 				nfsv4_unlock(&nfsv4rootfs_lock, 1);
1649 				NFSUNLOCKV4ROOTMUTEX();
1650 			    }
1651 			    return (NFSERR_OPENMODE);
1652 			}
1653 		    }
1654 		}
1655 
1656 		/* We're outta here */
1657 		NFSUNLOCKSTATE();
1658 		if (haslock) {
1659 			NFSLOCKV4ROOTMUTEX();
1660 			nfsv4_unlock(&nfsv4rootfs_lock, 1);
1661 			NFSUNLOCKV4ROOTMUTEX();
1662 		}
1663 		return (0);
1664 	    }
1665 	}
1666 
1667 	/*
1668 	 * For setattr, just get rid of all the Delegations for other clients.
1669 	 */
1670 	if (new_stp->ls_flags & NFSLCK_SETATTR) {
1671 		ret = nfsrv_cleandeleg(vp, lfp, clp, &haslock, p);
1672 		if (ret) {
1673 			/*
1674 			 * nfsrv_cleandeleg() unlocks state when it
1675 			 * returns non-zero.
1676 			 */
1677 			if (ret == -1) {
1678 				lckstp = NULL;
1679 				goto tryagain;
1680 			}
1681 			return (ret);
1682 		}
1683 		if (!(new_stp->ls_flags & NFSLCK_CHECK) ||
1684 		    (LIST_EMPTY(&lfp->lf_open) && LIST_EMPTY(&lfp->lf_lock) &&
1685 		     LIST_EMPTY(&lfp->lf_deleg))) {
1686 			NFSUNLOCKSTATE();
1687 			if (haslock) {
1688 				NFSLOCKV4ROOTMUTEX();
1689 				nfsv4_unlock(&nfsv4rootfs_lock, 1);
1690 				NFSUNLOCKV4ROOTMUTEX();
1691 			}
1692 			return (0);
1693 		}
1694 	}
1695 
1696 	/*
1697 	 * Check for a conflicting delegation. If one is found, call
1698 	 * nfsrv_delegconflict() to handle it. If the v4root lock hasn't
1699 	 * been set yet, it will get the lock. Otherwise, it will recall
1700 	 * the delegation. Then, we try try again...
1701 	 * I currently believe the conflict algorithm to be:
1702 	 * For Lock Ops (Lock/LockT/LockU)
1703 	 * - there is a conflict iff a different client has a write delegation
1704 	 * For Reading (Read Op)
1705 	 * - there is a conflict iff a different client has a write delegation
1706 	 *   (the specialids are always a different client)
1707 	 * For Writing (Write/Setattr of size)
1708 	 * - there is a conflict if a different client has any delegation
1709 	 * - there is a conflict if the same client has a read delegation
1710 	 *   (I don't understand why this isn't allowed, but that seems to be
1711 	 *    the current concensus?)
1712 	 */
1713 	tstp = LIST_FIRST(&lfp->lf_deleg);
1714 	while (tstp != LIST_END(&lfp->lf_deleg)) {
1715 	    nstp = LIST_NEXT(tstp, ls_file);
1716 	    if ((((new_stp->ls_flags&(NFSLCK_LOCK|NFSLCK_UNLOCK|NFSLCK_TEST))||
1717 		 ((new_stp->ls_flags & NFSLCK_CHECK) &&
1718 		  (new_lop->lo_flags & NFSLCK_READ))) &&
1719 		  clp != tstp->ls_clp &&
1720 		 (tstp->ls_flags & NFSLCK_DELEGWRITE)) ||
1721 		 ((new_stp->ls_flags & NFSLCK_CHECK) &&
1722 		   (new_lop->lo_flags & NFSLCK_WRITE) &&
1723 		  (clp != tstp->ls_clp ||
1724 		   (tstp->ls_flags & NFSLCK_DELEGREAD)))) {
1725 		if (filestruct_locked != 0) {
1726 			/* Roll back local locks. */
1727 			NFSUNLOCKSTATE();
1728 			nfsrv_locallock_rollback(vp, lfp, p);
1729 			NFSLOCKSTATE();
1730 			nfsrv_unlocklf(lfp);
1731 		}
1732 		ret = nfsrv_delegconflict(tstp, &haslock, p, vp);
1733 		if (ret) {
1734 		    /*
1735 		     * nfsrv_delegconflict unlocks state when it
1736 		     * returns non-zero, which it always does.
1737 		     */
1738 		    if (other_lop) {
1739 			FREE((caddr_t)other_lop, M_NFSDLOCK);
1740 			other_lop = NULL;
1741 		    }
1742 		    if (ret == -1) {
1743 			lckstp = NULL;
1744 			goto tryagain;
1745 		    }
1746 		    return (ret);
1747 		}
1748 		/* Never gets here. */
1749 	    }
1750 	    tstp = nstp;
1751 	}
1752 
1753 	/*
1754 	 * Handle the unlock case by calling nfsrv_updatelock().
1755 	 * (Should I have done some access checking above for unlock? For now,
1756 	 *  just let it happen.)
1757 	 */
1758 	if (new_stp->ls_flags & NFSLCK_UNLOCK) {
1759 		first = new_lop->lo_first;
1760 		end = new_lop->lo_end;
1761 		nfsrv_updatelock(stp, new_lopp, &other_lop, lfp);
1762 		stateidp->seqid = ++(stp->ls_stateid.seqid);
1763 		stateidp->other[0] = stp->ls_stateid.other[0];
1764 		stateidp->other[1] = stp->ls_stateid.other[1];
1765 		stateidp->other[2] = stp->ls_stateid.other[2];
1766 		if (filestruct_locked != 0) {
1767 			NFSUNLOCKSTATE();
1768 			/* Update the local locks. */
1769 			nfsrv_localunlock(vp, lfp, first, end, p);
1770 			NFSLOCKSTATE();
1771 			nfsrv_unlocklf(lfp);
1772 		}
1773 		NFSUNLOCKSTATE();
1774 		if (haslock) {
1775 			NFSLOCKV4ROOTMUTEX();
1776 			nfsv4_unlock(&nfsv4rootfs_lock, 1);
1777 			NFSUNLOCKV4ROOTMUTEX();
1778 		}
1779 		return (0);
1780 	}
1781 
1782 	/*
1783 	 * Search for a conflicting lock. A lock conflicts if:
1784 	 * - the lock range overlaps and
1785 	 * - at least one lock is a write lock and
1786 	 * - it is not owned by the same lock owner
1787 	 */
1788 	if (!delegation) {
1789 	  LIST_FOREACH(lop, &lfp->lf_lock, lo_lckfile) {
1790 	    if (new_lop->lo_end > lop->lo_first &&
1791 		new_lop->lo_first < lop->lo_end &&
1792 		(new_lop->lo_flags == NFSLCK_WRITE ||
1793 		 lop->lo_flags == NFSLCK_WRITE) &&
1794 		lckstp != lop->lo_stp &&
1795 		(clp != lop->lo_stp->ls_clp ||
1796 		 lckstp->ls_ownerlen != lop->lo_stp->ls_ownerlen ||
1797 		 NFSBCMP(lckstp->ls_owner, lop->lo_stp->ls_owner,
1798 		    lckstp->ls_ownerlen))) {
1799 		if (other_lop) {
1800 		    FREE((caddr_t)other_lop, M_NFSDLOCK);
1801 		    other_lop = NULL;
1802 		}
1803 		ret = nfsrv_clientconflict(lop->lo_stp->ls_clp,&haslock,vp,p);
1804 		if (ret) {
1805 		    if (filestruct_locked != 0) {
1806 			/* Roll back local locks. */
1807 			nfsrv_locallock_rollback(vp, lfp, p);
1808 			NFSLOCKSTATE();
1809 			nfsrv_unlocklf(lfp);
1810 			NFSUNLOCKSTATE();
1811 		    }
1812 		    /*
1813 		     * nfsrv_clientconflict() unlocks state when it
1814 		     * returns non-zero.
1815 		     */
1816 		    lckstp = NULL;
1817 		    goto tryagain;
1818 		}
1819 		/*
1820 		 * Found a conflicting lock, so record the conflict and
1821 		 * return the error.
1822 		 */
1823 		if (cfp) {
1824 		    cfp->cl_clientid.lval[0]=lop->lo_stp->ls_stateid.other[0];
1825 		    cfp->cl_clientid.lval[1]=lop->lo_stp->ls_stateid.other[1];
1826 		    cfp->cl_first = lop->lo_first;
1827 		    cfp->cl_end = lop->lo_end;
1828 		    cfp->cl_flags = lop->lo_flags;
1829 		    cfp->cl_ownerlen = lop->lo_stp->ls_ownerlen;
1830 		    NFSBCOPY(lop->lo_stp->ls_owner, cfp->cl_owner,
1831 			cfp->cl_ownerlen);
1832 		}
1833 		if (new_stp->ls_flags & NFSLCK_RECLAIM)
1834 		    error = NFSERR_RECLAIMCONFLICT;
1835 		else if (new_stp->ls_flags & NFSLCK_CHECK)
1836 		    error = NFSERR_LOCKED;
1837 		else
1838 		    error = NFSERR_DENIED;
1839 		if (filestruct_locked != 0) {
1840 			/* Roll back local locks. */
1841 			NFSUNLOCKSTATE();
1842 			nfsrv_locallock_rollback(vp, lfp, p);
1843 			NFSLOCKSTATE();
1844 			nfsrv_unlocklf(lfp);
1845 		}
1846 		NFSUNLOCKSTATE();
1847 		if (haslock) {
1848 			NFSLOCKV4ROOTMUTEX();
1849 			nfsv4_unlock(&nfsv4rootfs_lock, 1);
1850 			NFSUNLOCKV4ROOTMUTEX();
1851 		}
1852 		return (error);
1853 	    }
1854 	  }
1855 	}
1856 
1857 	/*
1858 	 * We only get here if there was no lock that conflicted.
1859 	 */
1860 	if (new_stp->ls_flags & (NFSLCK_TEST | NFSLCK_CHECK)) {
1861 		NFSUNLOCKSTATE();
1862 		if (haslock) {
1863 			NFSLOCKV4ROOTMUTEX();
1864 			nfsv4_unlock(&nfsv4rootfs_lock, 1);
1865 			NFSUNLOCKV4ROOTMUTEX();
1866 		}
1867 		return (0);
1868 	}
1869 
1870 	/*
1871 	 * We only get here when we are creating or modifying a lock.
1872 	 * There are two variants:
1873 	 * - exist_lock_owner where lock_owner exists
1874 	 * - open_to_lock_owner with new lock_owner
1875 	 */
1876 	first = new_lop->lo_first;
1877 	end = new_lop->lo_end;
1878 	lock_flags = new_lop->lo_flags;
1879 	if (!(new_stp->ls_flags & NFSLCK_OPENTOLOCK)) {
1880 		nfsrv_updatelock(lckstp, new_lopp, &other_lop, lfp);
1881 		stateidp->seqid = ++(lckstp->ls_stateid.seqid);
1882 		stateidp->other[0] = lckstp->ls_stateid.other[0];
1883 		stateidp->other[1] = lckstp->ls_stateid.other[1];
1884 		stateidp->other[2] = lckstp->ls_stateid.other[2];
1885 	} else {
1886 		/*
1887 		 * The new open_to_lock_owner case.
1888 		 * Link the new nfsstate into the lists.
1889 		 */
1890 		new_stp->ls_seq = new_stp->ls_opentolockseq;
1891 		nfsrvd_refcache(new_stp->ls_op);
1892 		stateidp->seqid = new_stp->ls_stateid.seqid = 0;
1893 		stateidp->other[0] = new_stp->ls_stateid.other[0] =
1894 		    clp->lc_clientid.lval[0];
1895 		stateidp->other[1] = new_stp->ls_stateid.other[1] =
1896 		    clp->lc_clientid.lval[1];
1897 		stateidp->other[2] = new_stp->ls_stateid.other[2] =
1898 		    nfsrv_nextstateindex(clp);
1899 		new_stp->ls_clp = clp;
1900 		LIST_INIT(&new_stp->ls_lock);
1901 		new_stp->ls_openstp = stp;
1902 		new_stp->ls_lfp = lfp;
1903 		nfsrv_insertlock(new_lop, (struct nfslock *)new_stp, new_stp,
1904 		    lfp);
1905 		LIST_INSERT_HEAD(NFSSTATEHASH(clp, new_stp->ls_stateid),
1906 		    new_stp, ls_hash);
1907 		LIST_INSERT_HEAD(&stp->ls_open, new_stp, ls_list);
1908 		*new_lopp = NULL;
1909 		*new_stpp = NULL;
1910 		newnfsstats.srvlockowners++;
1911 		nfsrv_openpluslock++;
1912 	}
1913 	if (filestruct_locked != 0) {
1914 		NFSUNLOCKSTATE();
1915 		nfsrv_locallock_commit(lfp, lock_flags, first, end);
1916 		NFSLOCKSTATE();
1917 		nfsrv_unlocklf(lfp);
1918 	}
1919 	NFSUNLOCKSTATE();
1920 	if (haslock) {
1921 		NFSLOCKV4ROOTMUTEX();
1922 		nfsv4_unlock(&nfsv4rootfs_lock, 1);
1923 		NFSUNLOCKV4ROOTMUTEX();
1924 	}
1925 	if (other_lop)
1926 		FREE((caddr_t)other_lop, M_NFSDLOCK);
1927 	return (0);
1928 }
1929 
1930 /*
1931  * Check for state errors for Open.
1932  * repstat is passed back out as an error if more critical errors
1933  * are not detected.
1934  */
1935 APPLESTATIC int
1936 nfsrv_opencheck(nfsquad_t clientid, nfsv4stateid_t *stateidp,
1937     struct nfsstate *new_stp, vnode_t vp, struct nfsrv_descript *nd,
1938     NFSPROC_T *p, int repstat)
1939 {
1940 	struct nfsstate *stp, *nstp;
1941 	struct nfsclient *clp;
1942 	struct nfsstate *ownerstp;
1943 	struct nfslockfile *lfp, *new_lfp;
1944 	int error, haslock = 0, ret, readonly = 0, getfhret = 0;
1945 
1946 	if ((new_stp->ls_flags & NFSLCK_SHAREBITS) == NFSLCK_READACCESS)
1947 		readonly = 1;
1948 	/*
1949 	 * Check for restart conditions (client and server).
1950 	 */
1951 	error = nfsrv_checkrestart(clientid, new_stp->ls_flags,
1952 		&new_stp->ls_stateid, 0);
1953 	if (error)
1954 		return (error);
1955 
1956 	/*
1957 	 * Check for state resource limit exceeded.
1958 	 * Technically this should be SMP protected, but the worst
1959 	 * case error is "out by one or two" on the count when it
1960 	 * returns NFSERR_RESOURCE and the limit is just a rather
1961 	 * arbitrary high water mark, so no harm is done.
1962 	 */
1963 	if (nfsrv_openpluslock > NFSRV_V4STATELIMIT)
1964 		return (NFSERR_RESOURCE);
1965 
1966 tryagain:
1967 	MALLOC(new_lfp, struct nfslockfile *, sizeof (struct nfslockfile),
1968 	    M_NFSDLOCKFILE, M_WAITOK);
1969 	if (vp)
1970 		getfhret = nfsrv_getlockfh(vp, new_stp->ls_flags, &new_lfp,
1971 		    NULL, p);
1972 	NFSLOCKSTATE();
1973 	/*
1974 	 * Get the nfsclient structure.
1975 	 * Since the code is manipulating lists that are also
1976 	 * manipulated by nfsrv_servertimer(), soft clock interrupts
1977 	 * must be masked off.
1978 	 */
1979 	error = nfsrv_getclient(clientid, CLOPS_RENEW, &clp,
1980 	    (nfsquad_t)((u_quad_t)0), NULL, p);
1981 
1982 	/*
1983 	 * Look up the open owner. See if it needs confirmation and
1984 	 * check the seq#, as required.
1985 	 */
1986 	if (!error)
1987 		nfsrv_getowner(&clp->lc_open, new_stp, &ownerstp);
1988 
1989 	if (!error && ownerstp) {
1990 		error = nfsrv_checkseqid(nd, new_stp->ls_seq, ownerstp,
1991 		    new_stp->ls_op);
1992 		/*
1993 		 * If the OpenOwner hasn't been confirmed, assume the
1994 		 * old one was a replay and this one is ok.
1995 		 * See: RFC3530 Sec. 14.2.18.
1996 		 */
1997 		if (error == NFSERR_BADSEQID &&
1998 		    (ownerstp->ls_flags & NFSLCK_NEEDSCONFIRM))
1999 			error = 0;
2000 	}
2001 
2002 	/*
2003 	 * Check for grace.
2004 	 */
2005 	if (!error)
2006 		error = nfsrv_checkgrace(new_stp->ls_flags);
2007 	if ((new_stp->ls_flags & NFSLCK_RECLAIM) && !error &&
2008 		nfsrv_checkstable(clp))
2009 		error = NFSERR_NOGRACE;
2010 
2011 	/*
2012 	 * If none of the above errors occurred, let repstat be
2013 	 * returned.
2014 	 */
2015 	if (repstat && !error)
2016 		error = repstat;
2017 	if (error) {
2018 		NFSUNLOCKSTATE();
2019 		if (haslock) {
2020 			NFSLOCKV4ROOTMUTEX();
2021 			nfsv4_unlock(&nfsv4rootfs_lock, 1);
2022 			NFSUNLOCKV4ROOTMUTEX();
2023 		}
2024 		free((caddr_t)new_lfp, M_NFSDLOCKFILE);
2025 		return (error);
2026 	}
2027 
2028 	/*
2029 	 * If vp == NULL, the file doesn't exist yet, so return ok.
2030 	 * (This always happens on the first pass, so haslock must be 0.)
2031 	 */
2032 	if (vp == NULL) {
2033 		NFSUNLOCKSTATE();
2034 		FREE((caddr_t)new_lfp, M_NFSDLOCKFILE);
2035 		return (0);
2036 	}
2037 
2038 	/*
2039 	 * Get the structure for the underlying file.
2040 	 */
2041 	if (getfhret)
2042 		error = getfhret;
2043 	else
2044 		error = nfsrv_getlockfile(new_stp->ls_flags, &new_lfp, &lfp,
2045 		    NULL, 0);
2046 	if (new_lfp)
2047 		FREE((caddr_t)new_lfp, M_NFSDLOCKFILE);
2048 	if (error) {
2049 		NFSUNLOCKSTATE();
2050 		if (haslock) {
2051 			NFSLOCKV4ROOTMUTEX();
2052 			nfsv4_unlock(&nfsv4rootfs_lock, 1);
2053 			NFSUNLOCKV4ROOTMUTEX();
2054 		}
2055 		return (error);
2056 	}
2057 
2058 	/*
2059 	 * Search for a conflicting open/share.
2060 	 */
2061 	if (new_stp->ls_flags & NFSLCK_DELEGCUR) {
2062 	    /*
2063 	     * For Delegate_Cur, search for the matching Delegation,
2064 	     * which indicates no conflict.
2065 	     * An old delegation should have been recovered by the
2066 	     * client doing a Claim_DELEGATE_Prev, so I won't let
2067 	     * it match and return NFSERR_EXPIRED. Should I let it
2068 	     * match?
2069 	     */
2070 	    LIST_FOREACH(stp, &lfp->lf_deleg, ls_file) {
2071 		if (!(stp->ls_flags & NFSLCK_OLDDELEG) &&
2072 		    stateidp->seqid == stp->ls_stateid.seqid &&
2073 		    !NFSBCMP(stateidp->other, stp->ls_stateid.other,
2074 			  NFSX_STATEIDOTHER))
2075 			break;
2076 	    }
2077 	    if (stp == LIST_END(&lfp->lf_deleg) ||
2078 		((new_stp->ls_flags & NFSLCK_WRITEACCESS) &&
2079 		 (stp->ls_flags & NFSLCK_DELEGREAD))) {
2080 		NFSUNLOCKSTATE();
2081 		if (haslock) {
2082 			NFSLOCKV4ROOTMUTEX();
2083 			nfsv4_unlock(&nfsv4rootfs_lock, 1);
2084 			NFSUNLOCKV4ROOTMUTEX();
2085 		}
2086 		return (NFSERR_EXPIRED);
2087 	    }
2088 	}
2089 
2090 	/*
2091 	 * Check for access/deny bit conflicts. I check for the same
2092 	 * owner as well, in case the client didn't bother.
2093 	 */
2094 	LIST_FOREACH(stp, &lfp->lf_open, ls_file) {
2095 		if (!(new_stp->ls_flags & NFSLCK_DELEGCUR) &&
2096 		    (((new_stp->ls_flags & NFSLCK_ACCESSBITS) &
2097 		      ((stp->ls_flags>>NFSLCK_SHIFT) & NFSLCK_ACCESSBITS))||
2098 		     ((stp->ls_flags & NFSLCK_ACCESSBITS) &
2099 		      ((new_stp->ls_flags>>NFSLCK_SHIFT)&NFSLCK_ACCESSBITS)))){
2100 			ret = nfsrv_clientconflict(stp->ls_clp,&haslock,vp,p);
2101 			if (ret) {
2102 				/*
2103 				 * nfsrv_clientconflict() unlocks
2104 				 * state when it returns non-zero.
2105 				 */
2106 				goto tryagain;
2107 			}
2108 			if (new_stp->ls_flags & NFSLCK_RECLAIM)
2109 				error = NFSERR_RECLAIMCONFLICT;
2110 			else
2111 				error = NFSERR_SHAREDENIED;
2112 			NFSUNLOCKSTATE();
2113 			if (haslock) {
2114 				NFSLOCKV4ROOTMUTEX();
2115 				nfsv4_unlock(&nfsv4rootfs_lock, 1);
2116 				NFSUNLOCKV4ROOTMUTEX();
2117 			}
2118 			return (error);
2119 		}
2120 	}
2121 
2122 	/*
2123 	 * Check for a conflicting delegation. If one is found, call
2124 	 * nfsrv_delegconflict() to handle it. If the v4root lock hasn't
2125 	 * been set yet, it will get the lock. Otherwise, it will recall
2126 	 * the delegation. Then, we try try again...
2127 	 * (If NFSLCK_DELEGCUR is set, it has a delegation, so there
2128 	 *  isn't a conflict.)
2129 	 * I currently believe the conflict algorithm to be:
2130 	 * For Open with Read Access and Deny None
2131 	 * - there is a conflict iff a different client has a write delegation
2132 	 * For Open with other Write Access or any Deny except None
2133 	 * - there is a conflict if a different client has any delegation
2134 	 * - there is a conflict if the same client has a read delegation
2135 	 *   (The current concensus is that this last case should be
2136 	 *    considered a conflict since the client with a read delegation
2137 	 *    could have done an Open with ReadAccess and WriteDeny
2138 	 *    locally and then not have checked for the WriteDeny.)
2139 	 * Don't check for a Reclaim, since that will be dealt with
2140 	 * by nfsrv_openctrl().
2141 	 */
2142 	if (!(new_stp->ls_flags &
2143 		(NFSLCK_DELEGPREV | NFSLCK_DELEGCUR | NFSLCK_RECLAIM))) {
2144 	    stp = LIST_FIRST(&lfp->lf_deleg);
2145 	    while (stp != LIST_END(&lfp->lf_deleg)) {
2146 		nstp = LIST_NEXT(stp, ls_file);
2147 		if ((readonly && stp->ls_clp != clp &&
2148 		       (stp->ls_flags & NFSLCK_DELEGWRITE)) ||
2149 		    (!readonly && (stp->ls_clp != clp ||
2150 		         (stp->ls_flags & NFSLCK_DELEGREAD)))) {
2151 			ret = nfsrv_delegconflict(stp, &haslock, p, vp);
2152 			if (ret) {
2153 			    /*
2154 			     * nfsrv_delegconflict() unlocks state
2155 			     * when it returns non-zero.
2156 			     */
2157 			    if (ret == -1)
2158 				goto tryagain;
2159 			    return (ret);
2160 			}
2161 		}
2162 		stp = nstp;
2163 	    }
2164 	}
2165 	NFSUNLOCKSTATE();
2166 	if (haslock) {
2167 		NFSLOCKV4ROOTMUTEX();
2168 		nfsv4_unlock(&nfsv4rootfs_lock, 1);
2169 		NFSUNLOCKV4ROOTMUTEX();
2170 	}
2171 	return (0);
2172 }
2173 
2174 /*
2175  * Open control function to create/update open state for an open.
2176  */
2177 APPLESTATIC int
2178 nfsrv_openctrl(struct nfsrv_descript *nd, vnode_t vp,
2179     struct nfsstate **new_stpp, nfsquad_t clientid, nfsv4stateid_t *stateidp,
2180     nfsv4stateid_t *delegstateidp, u_int32_t *rflagsp, struct nfsexstuff *exp,
2181     NFSPROC_T *p, u_quad_t filerev)
2182 {
2183 	struct nfsstate *new_stp = *new_stpp;
2184 	struct nfsstate *stp, *nstp;
2185 	struct nfsstate *openstp = NULL, *new_open, *ownerstp, *new_deleg;
2186 	struct nfslockfile *lfp, *new_lfp;
2187 	struct nfsclient *clp;
2188 	int error, haslock = 0, ret, delegate = 1, writedeleg = 1;
2189 	int readonly = 0, cbret = 1, getfhret = 0;
2190 
2191 	if ((new_stp->ls_flags & NFSLCK_SHAREBITS) == NFSLCK_READACCESS)
2192 		readonly = 1;
2193 	/*
2194 	 * Check for restart conditions (client and server).
2195 	 * (Paranoia, should have been detected by nfsrv_opencheck().)
2196 	 * If an error does show up, return NFSERR_EXPIRED, since the
2197 	 * the seqid# has already been incremented.
2198 	 */
2199 	error = nfsrv_checkrestart(clientid, new_stp->ls_flags,
2200 	    &new_stp->ls_stateid, 0);
2201 	if (error) {
2202 		printf("Nfsd: openctrl unexpected restart err=%d\n",
2203 		    error);
2204 		return (NFSERR_EXPIRED);
2205 	}
2206 
2207 tryagain:
2208 	MALLOC(new_lfp, struct nfslockfile *, sizeof (struct nfslockfile),
2209 	    M_NFSDLOCKFILE, M_WAITOK);
2210 	MALLOC(new_open, struct nfsstate *, sizeof (struct nfsstate),
2211 	    M_NFSDSTATE, M_WAITOK);
2212 	MALLOC(new_deleg, struct nfsstate *, sizeof (struct nfsstate),
2213 	    M_NFSDSTATE, M_WAITOK);
2214 	getfhret = nfsrv_getlockfh(vp, new_stp->ls_flags, &new_lfp,
2215 	    NULL, p);
2216 	NFSLOCKSTATE();
2217 	/*
2218 	 * Get the client structure. Since the linked lists could be changed
2219 	 * by other nfsd processes if this process does a tsleep(), one of
2220 	 * two things must be done.
2221 	 * 1 - don't tsleep()
2222 	 * or
2223 	 * 2 - get the nfsv4_lock() { indicated by haslock == 1 }
2224 	 *     before using the lists, since this lock stops the other
2225 	 *     nfsd. This should only be used for rare cases, since it
2226 	 *     essentially single threads the nfsd.
2227 	 *     At this time, it is only done for cases where the stable
2228 	 *     storage file must be written prior to completion of state
2229 	 *     expiration.
2230 	 */
2231 	error = nfsrv_getclient(clientid, CLOPS_RENEW, &clp,
2232 	    (nfsquad_t)((u_quad_t)0), NULL, p);
2233 	if (!error && (clp->lc_flags & LCL_NEEDSCBNULL) &&
2234 	    clp->lc_program) {
2235 		/*
2236 		 * This happens on the first open for a client
2237 		 * that supports callbacks.
2238 		 */
2239 		NFSUNLOCKSTATE();
2240 		/*
2241 		 * Although nfsrv_docallback() will sleep, clp won't
2242 		 * go away, since they are only removed when the
2243 		 * nfsv4_lock() has blocked the nfsd threads. The
2244 		 * fields in clp can change, but having multiple
2245 		 * threads do this Null callback RPC should be
2246 		 * harmless.
2247 		 */
2248 		cbret = nfsrv_docallback(clp, NFSV4PROC_CBNULL,
2249 		    NULL, 0, NULL, NULL, NULL, p);
2250 		NFSLOCKSTATE();
2251 		clp->lc_flags &= ~LCL_NEEDSCBNULL;
2252 		if (!cbret)
2253 			clp->lc_flags |= LCL_CALLBACKSON;
2254 	}
2255 
2256 	/*
2257 	 * Look up the open owner. See if it needs confirmation and
2258 	 * check the seq#, as required.
2259 	 */
2260 	if (!error)
2261 		nfsrv_getowner(&clp->lc_open, new_stp, &ownerstp);
2262 
2263 	if (error) {
2264 		NFSUNLOCKSTATE();
2265 		printf("Nfsd: openctrl unexpected state err=%d\n",
2266 			error);
2267 		free((caddr_t)new_lfp, M_NFSDLOCKFILE);
2268 		free((caddr_t)new_open, M_NFSDSTATE);
2269 		free((caddr_t)new_deleg, M_NFSDSTATE);
2270 		if (haslock) {
2271 			NFSLOCKV4ROOTMUTEX();
2272 			nfsv4_unlock(&nfsv4rootfs_lock, 1);
2273 			NFSUNLOCKV4ROOTMUTEX();
2274 		}
2275 		return (NFSERR_EXPIRED);
2276 	}
2277 
2278 	if (new_stp->ls_flags & NFSLCK_RECLAIM)
2279 		nfsrv_markstable(clp);
2280 
2281 	/*
2282 	 * Get the structure for the underlying file.
2283 	 */
2284 	if (getfhret)
2285 		error = getfhret;
2286 	else
2287 		error = nfsrv_getlockfile(new_stp->ls_flags, &new_lfp, &lfp,
2288 		    NULL, 0);
2289 	if (new_lfp)
2290 		FREE((caddr_t)new_lfp, M_NFSDLOCKFILE);
2291 	if (error) {
2292 		NFSUNLOCKSTATE();
2293 		printf("Nfsd openctrl unexpected getlockfile err=%d\n",
2294 		    error);
2295 		free((caddr_t)new_open, M_NFSDSTATE);
2296 		free((caddr_t)new_deleg, M_NFSDSTATE);
2297 		if (haslock) {
2298 			NFSLOCKV4ROOTMUTEX();
2299 			nfsv4_unlock(&nfsv4rootfs_lock, 1);
2300 			NFSUNLOCKV4ROOTMUTEX();
2301 		}
2302 		return (error);
2303 	}
2304 
2305 	/*
2306 	 * Search for a conflicting open/share.
2307 	 */
2308 	if (new_stp->ls_flags & NFSLCK_DELEGCUR) {
2309 	    /*
2310 	     * For Delegate_Cur, search for the matching Delegation,
2311 	     * which indicates no conflict.
2312 	     * An old delegation should have been recovered by the
2313 	     * client doing a Claim_DELEGATE_Prev, so I won't let
2314 	     * it match and return NFSERR_EXPIRED. Should I let it
2315 	     * match?
2316 	     */
2317 	    LIST_FOREACH(stp, &lfp->lf_deleg, ls_file) {
2318 		if (!(stp->ls_flags & NFSLCK_OLDDELEG) &&
2319 		    stateidp->seqid == stp->ls_stateid.seqid &&
2320 		    !NFSBCMP(stateidp->other, stp->ls_stateid.other,
2321 			NFSX_STATEIDOTHER))
2322 			break;
2323 	    }
2324 	    if (stp == LIST_END(&lfp->lf_deleg) ||
2325 		((new_stp->ls_flags & NFSLCK_WRITEACCESS) &&
2326 		 (stp->ls_flags & NFSLCK_DELEGREAD))) {
2327 		NFSUNLOCKSTATE();
2328 		printf("Nfsd openctrl unexpected expiry\n");
2329 		free((caddr_t)new_open, M_NFSDSTATE);
2330 		free((caddr_t)new_deleg, M_NFSDSTATE);
2331 		if (haslock) {
2332 			NFSLOCKV4ROOTMUTEX();
2333 			nfsv4_unlock(&nfsv4rootfs_lock, 1);
2334 			NFSUNLOCKV4ROOTMUTEX();
2335 		}
2336 		return (NFSERR_EXPIRED);
2337 	    }
2338 
2339 	    /*
2340 	     * Don't issue a Delegation, since one already exists and
2341 	     * delay delegation timeout, as required.
2342 	     */
2343 	    delegate = 0;
2344 	    nfsrv_delaydelegtimeout(stp);
2345 	}
2346 
2347 	/*
2348 	 * Check for access/deny bit conflicts. I also check for the
2349 	 * same owner, since the client might not have bothered to check.
2350 	 * Also, note an open for the same file and owner, if found,
2351 	 * which is all we do here for Delegate_Cur, since conflict
2352 	 * checking is already done.
2353 	 */
2354 	LIST_FOREACH(stp, &lfp->lf_open, ls_file) {
2355 		if (ownerstp && stp->ls_openowner == ownerstp)
2356 			openstp = stp;
2357 		if (!(new_stp->ls_flags & NFSLCK_DELEGCUR)) {
2358 		    /*
2359 		     * If another client has the file open, the only
2360 		     * delegation that can be issued is a Read delegation
2361 		     * and only if it is a Read open with Deny none.
2362 		     */
2363 		    if (clp != stp->ls_clp) {
2364 			if ((stp->ls_flags & NFSLCK_SHAREBITS) ==
2365 			    NFSLCK_READACCESS)
2366 			    writedeleg = 0;
2367 			else
2368 			    delegate = 0;
2369 		    }
2370 		    if(((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) {
2376 				/*
2377 				 * nfsrv_clientconflict() unlocks state
2378 				 * when it returns non-zero.
2379 				 */
2380 				free((caddr_t)new_open, M_NFSDSTATE);
2381 				free((caddr_t)new_deleg, M_NFSDSTATE);
2382 				openstp = NULL;
2383 				goto tryagain;
2384 			}
2385 			if (new_stp->ls_flags & NFSLCK_RECLAIM)
2386 				error = NFSERR_RECLAIMCONFLICT;
2387 			else
2388 				error = NFSERR_SHAREDENIED;
2389 			NFSUNLOCKSTATE();
2390 			if (haslock) {
2391 				NFSLOCKV4ROOTMUTEX();
2392 				nfsv4_unlock(&nfsv4rootfs_lock, 1);
2393 				NFSUNLOCKV4ROOTMUTEX();
2394 			}
2395 			free((caddr_t)new_open, M_NFSDSTATE);
2396 			free((caddr_t)new_deleg, M_NFSDSTATE);
2397 			printf("nfsd openctrl unexpected client cnfl\n");
2398 			return (error);
2399 		    }
2400 		}
2401 	}
2402 
2403 	/*
2404 	 * Check for a conflicting delegation. If one is found, call
2405 	 * nfsrv_delegconflict() to handle it. If the v4root lock hasn't
2406 	 * been set yet, it will get the lock. Otherwise, it will recall
2407 	 * the delegation. Then, we try try again...
2408 	 * (If NFSLCK_DELEGCUR is set, it has a delegation, so there
2409 	 *  isn't a conflict.)
2410 	 * I currently believe the conflict algorithm to be:
2411 	 * For Open with Read Access and Deny None
2412 	 * - there is a conflict iff a different client has a write delegation
2413 	 * For Open with other Write Access or any Deny except None
2414 	 * - there is a conflict if a different client has any delegation
2415 	 * - there is a conflict if the same client has a read delegation
2416 	 *   (The current concensus is that this last case should be
2417 	 *    considered a conflict since the client with a read delegation
2418 	 *    could have done an Open with ReadAccess and WriteDeny
2419 	 *    locally and then not have checked for the WriteDeny.)
2420 	 */
2421 	if (!(new_stp->ls_flags & (NFSLCK_DELEGPREV | NFSLCK_DELEGCUR))) {
2422 	    stp = LIST_FIRST(&lfp->lf_deleg);
2423 	    while (stp != LIST_END(&lfp->lf_deleg)) {
2424 		nstp = LIST_NEXT(stp, ls_file);
2425 		if (stp->ls_clp != clp && (stp->ls_flags & NFSLCK_DELEGREAD))
2426 			writedeleg = 0;
2427 		else
2428 			delegate = 0;
2429 		if ((readonly && stp->ls_clp != clp &&
2430 		       (stp->ls_flags & NFSLCK_DELEGWRITE)) ||
2431 		    (!readonly && (stp->ls_clp != clp ||
2432 		         (stp->ls_flags & NFSLCK_DELEGREAD)))) {
2433 		    if (new_stp->ls_flags & NFSLCK_RECLAIM) {
2434 			delegate = 2;
2435 		    } else {
2436 			ret = nfsrv_delegconflict(stp, &haslock, p, vp);
2437 			if (ret) {
2438 			    /*
2439 			     * nfsrv_delegconflict() unlocks state
2440 			     * when it returns non-zero.
2441 			     */
2442 			    printf("Nfsd openctrl unexpected deleg cnfl\n");
2443 			    free((caddr_t)new_open, M_NFSDSTATE);
2444 			    free((caddr_t)new_deleg, M_NFSDSTATE);
2445 			    if (ret == -1) {
2446 				openstp = NULL;
2447 				goto tryagain;
2448 			    }
2449 			    return (ret);
2450 			}
2451 		    }
2452 		}
2453 		stp = nstp;
2454 	    }
2455 	}
2456 
2457 	/*
2458 	 * We only get here if there was no open that conflicted.
2459 	 * If an open for the owner exists, or in the access/deny bits.
2460 	 * Otherwise it is a new open. If the open_owner hasn't been
2461 	 * confirmed, replace the open with the new one needing confirmation,
2462 	 * otherwise add the open.
2463 	 */
2464 	if (new_stp->ls_flags & NFSLCK_DELEGPREV) {
2465 	    /*
2466 	     * Handle NFSLCK_DELEGPREV by searching the old delegations for
2467 	     * a match. If found, just move the old delegation to the current
2468 	     * delegation list and issue open. If not found, return
2469 	     * NFSERR_EXPIRED.
2470 	     */
2471 	    LIST_FOREACH(stp, &clp->lc_olddeleg, ls_list) {
2472 		if (stp->ls_lfp == lfp) {
2473 		    /* Found it */
2474 		    if (stp->ls_clp != clp)
2475 			panic("olddeleg clp");
2476 		    LIST_REMOVE(stp, ls_list);
2477 		    LIST_REMOVE(stp, ls_hash);
2478 		    stp->ls_flags &= ~NFSLCK_OLDDELEG;
2479 		    stp->ls_stateid.seqid = delegstateidp->seqid = 0;
2480 		    stp->ls_stateid.other[0] = delegstateidp->other[0] =
2481 			clp->lc_clientid.lval[0];
2482 		    stp->ls_stateid.other[1] = delegstateidp->other[1] =
2483 			clp->lc_clientid.lval[1];
2484 		    stp->ls_stateid.other[2] = delegstateidp->other[2] =
2485 			nfsrv_nextstateindex(clp);
2486 		    stp->ls_compref = nd->nd_compref;
2487 		    LIST_INSERT_HEAD(&clp->lc_deleg, stp, ls_list);
2488 		    LIST_INSERT_HEAD(NFSSTATEHASH(clp,
2489 			stp->ls_stateid), stp, ls_hash);
2490 		    if (stp->ls_flags & NFSLCK_DELEGWRITE)
2491 			*rflagsp |= NFSV4OPEN_WRITEDELEGATE;
2492 		    else
2493 			*rflagsp |= NFSV4OPEN_READDELEGATE;
2494 		    clp->lc_delegtime = NFSD_MONOSEC +
2495 			nfsrv_lease + NFSRV_LEASEDELTA;
2496 
2497 		    /*
2498 		     * Now, do the associated open.
2499 		     */
2500 		    new_open->ls_stateid.seqid = 0;
2501 		    new_open->ls_stateid.other[0] = clp->lc_clientid.lval[0];
2502 		    new_open->ls_stateid.other[1] = clp->lc_clientid.lval[1];
2503 		    new_open->ls_stateid.other[2] = nfsrv_nextstateindex(clp);
2504 		    new_open->ls_flags = (new_stp->ls_flags&NFSLCK_DENYBITS)|
2505 			NFSLCK_OPEN;
2506 		    if (stp->ls_flags & NFSLCK_DELEGWRITE)
2507 			new_open->ls_flags |= (NFSLCK_READACCESS |
2508 			    NFSLCK_WRITEACCESS);
2509 		    else
2510 			new_open->ls_flags |= NFSLCK_READACCESS;
2511 		    new_open->ls_uid = new_stp->ls_uid;
2512 		    new_open->ls_lfp = lfp;
2513 		    new_open->ls_clp = clp;
2514 		    LIST_INIT(&new_open->ls_open);
2515 		    LIST_INSERT_HEAD(&lfp->lf_open, new_open, ls_file);
2516 		    LIST_INSERT_HEAD(NFSSTATEHASH(clp, new_open->ls_stateid),
2517 			new_open, ls_hash);
2518 		    /*
2519 		     * and handle the open owner
2520 		     */
2521 		    if (ownerstp) {
2522 			new_open->ls_openowner = ownerstp;
2523 			LIST_INSERT_HEAD(&ownerstp->ls_open,new_open,ls_list);
2524 		    } else {
2525 			new_open->ls_openowner = new_stp;
2526 			new_stp->ls_flags = 0;
2527 			nfsrvd_refcache(new_stp->ls_op);
2528 			new_stp->ls_noopens = 0;
2529 			LIST_INIT(&new_stp->ls_open);
2530 			LIST_INSERT_HEAD(&new_stp->ls_open, new_open, ls_list);
2531 			LIST_INSERT_HEAD(&clp->lc_open, new_stp, ls_list);
2532 			*new_stpp = NULL;
2533 			newnfsstats.srvopenowners++;
2534 			nfsrv_openpluslock++;
2535 		    }
2536 		    openstp = new_open;
2537 		    new_open = NULL;
2538 		    newnfsstats.srvopens++;
2539 		    nfsrv_openpluslock++;
2540 		    break;
2541 		}
2542 	    }
2543 	    if (stp == LIST_END(&clp->lc_olddeleg))
2544 		error = NFSERR_EXPIRED;
2545 	} else if (new_stp->ls_flags & (NFSLCK_DELEGREAD | NFSLCK_DELEGWRITE)) {
2546 	    /*
2547 	     * Scan to see that no delegation for this client and file
2548 	     * doesn't already exist.
2549 	     * There also shouldn't yet be an Open for this file and
2550 	     * openowner.
2551 	     */
2552 	    LIST_FOREACH(stp, &lfp->lf_deleg, ls_file) {
2553 		if (stp->ls_clp == clp)
2554 		    break;
2555 	    }
2556 	    if (stp == LIST_END(&lfp->lf_deleg) && openstp == NULL) {
2557 		/*
2558 		 * This is the Claim_Previous case with a delegation
2559 		 * type != Delegate_None.
2560 		 */
2561 		/*
2562 		 * First, add the delegation. (Although we must issue the
2563 		 * delegation, we can also ask for an immediate return.)
2564 		 */
2565 		new_deleg->ls_stateid.seqid = delegstateidp->seqid = 0;
2566 		new_deleg->ls_stateid.other[0] = delegstateidp->other[0] =
2567 		    clp->lc_clientid.lval[0];
2568 		new_deleg->ls_stateid.other[1] = delegstateidp->other[1] =
2569 		    clp->lc_clientid.lval[1];
2570 		new_deleg->ls_stateid.other[2] = delegstateidp->other[2] =
2571 		    nfsrv_nextstateindex(clp);
2572 		if (new_stp->ls_flags & NFSLCK_DELEGWRITE) {
2573 		    new_deleg->ls_flags = (NFSLCK_DELEGWRITE |
2574 			NFSLCK_READACCESS | NFSLCK_WRITEACCESS);
2575 		    *rflagsp |= NFSV4OPEN_WRITEDELEGATE;
2576 		} else {
2577 		    new_deleg->ls_flags = (NFSLCK_DELEGREAD |
2578 			NFSLCK_READACCESS);
2579 		    *rflagsp |= NFSV4OPEN_READDELEGATE;
2580 		}
2581 		new_deleg->ls_uid = new_stp->ls_uid;
2582 		new_deleg->ls_lfp = lfp;
2583 		new_deleg->ls_clp = clp;
2584 		new_deleg->ls_filerev = filerev;
2585 		new_deleg->ls_compref = nd->nd_compref;
2586 		LIST_INSERT_HEAD(&lfp->lf_deleg, new_deleg, ls_file);
2587 		LIST_INSERT_HEAD(NFSSTATEHASH(clp,
2588 		    new_deleg->ls_stateid), new_deleg, ls_hash);
2589 		LIST_INSERT_HEAD(&clp->lc_deleg, new_deleg, ls_list);
2590 		new_deleg = NULL;
2591 		if (delegate == 2 || nfsrv_issuedelegs == 0 ||
2592 		    (clp->lc_flags & (LCL_CALLBACKSON | LCL_CBDOWN)) !=
2593 		     LCL_CALLBACKSON ||
2594 		    NFSRV_V4DELEGLIMIT(nfsrv_delegatecnt) ||
2595 		    !NFSVNO_DELEGOK(vp))
2596 		    *rflagsp |= NFSV4OPEN_RECALL;
2597 		newnfsstats.srvdelegates++;
2598 		nfsrv_openpluslock++;
2599 		nfsrv_delegatecnt++;
2600 
2601 		/*
2602 		 * Now, do the associated open.
2603 		 */
2604 		new_open->ls_stateid.seqid = 0;
2605 		new_open->ls_stateid.other[0] = clp->lc_clientid.lval[0];
2606 		new_open->ls_stateid.other[1] = clp->lc_clientid.lval[1];
2607 		new_open->ls_stateid.other[2] = nfsrv_nextstateindex(clp);
2608 		new_open->ls_flags = (new_stp->ls_flags & NFSLCK_DENYBITS) |
2609 		    NFSLCK_OPEN;
2610 		if (new_stp->ls_flags & NFSLCK_DELEGWRITE)
2611 			new_open->ls_flags |= (NFSLCK_READACCESS |
2612 			    NFSLCK_WRITEACCESS);
2613 		else
2614 			new_open->ls_flags |= NFSLCK_READACCESS;
2615 		new_open->ls_uid = new_stp->ls_uid;
2616 		new_open->ls_lfp = lfp;
2617 		new_open->ls_clp = clp;
2618 		LIST_INIT(&new_open->ls_open);
2619 		LIST_INSERT_HEAD(&lfp->lf_open, new_open, ls_file);
2620 		LIST_INSERT_HEAD(NFSSTATEHASH(clp, new_open->ls_stateid),
2621 		   new_open, ls_hash);
2622 		/*
2623 		 * and handle the open owner
2624 		 */
2625 		if (ownerstp) {
2626 		    new_open->ls_openowner = ownerstp;
2627 		    LIST_INSERT_HEAD(&ownerstp->ls_open, new_open, ls_list);
2628 		} else {
2629 		    new_open->ls_openowner = new_stp;
2630 		    new_stp->ls_flags = 0;
2631 		    nfsrvd_refcache(new_stp->ls_op);
2632 		    new_stp->ls_noopens = 0;
2633 		    LIST_INIT(&new_stp->ls_open);
2634 		    LIST_INSERT_HEAD(&new_stp->ls_open, new_open, ls_list);
2635 		    LIST_INSERT_HEAD(&clp->lc_open, new_stp, ls_list);
2636 		    *new_stpp = NULL;
2637 		    newnfsstats.srvopenowners++;
2638 		    nfsrv_openpluslock++;
2639 		}
2640 		openstp = new_open;
2641 		new_open = NULL;
2642 		newnfsstats.srvopens++;
2643 		nfsrv_openpluslock++;
2644 	    } else {
2645 		error = NFSERR_RECLAIMCONFLICT;
2646 	    }
2647 	} else if (ownerstp) {
2648 		if (ownerstp->ls_flags & NFSLCK_NEEDSCONFIRM) {
2649 		    /* Replace the open */
2650 		    if (ownerstp->ls_op)
2651 			nfsrvd_derefcache(ownerstp->ls_op);
2652 		    ownerstp->ls_op = new_stp->ls_op;
2653 		    nfsrvd_refcache(ownerstp->ls_op);
2654 		    ownerstp->ls_seq = new_stp->ls_seq;
2655 		    *rflagsp |= NFSV4OPEN_RESULTCONFIRM;
2656 		    stp = LIST_FIRST(&ownerstp->ls_open);
2657 		    stp->ls_flags = (new_stp->ls_flags & NFSLCK_SHAREBITS) |
2658 			NFSLCK_OPEN;
2659 		    stp->ls_stateid.seqid = 0;
2660 		    stp->ls_uid = new_stp->ls_uid;
2661 		    if (lfp != stp->ls_lfp) {
2662 			LIST_REMOVE(stp, ls_file);
2663 			LIST_INSERT_HEAD(&lfp->lf_open, stp, ls_file);
2664 			stp->ls_lfp = lfp;
2665 		    }
2666 		    openstp = stp;
2667 		} else if (openstp) {
2668 		    openstp->ls_flags |= (new_stp->ls_flags & NFSLCK_SHAREBITS);
2669 		    openstp->ls_stateid.seqid++;
2670 
2671 		    /*
2672 		     * This is where we can choose to issue a delegation.
2673 		     */
2674 		    if (delegate && nfsrv_issuedelegs &&
2675 			writedeleg && !NFSVNO_EXRDONLY(exp) &&
2676 			(nfsrv_writedelegifpos || !readonly) &&
2677 			(clp->lc_flags & (LCL_CALLBACKSON | LCL_CBDOWN)) ==
2678 			 LCL_CALLBACKSON &&
2679 			!NFSRV_V4DELEGLIMIT(nfsrv_delegatecnt) &&
2680 			NFSVNO_DELEGOK(vp)) {
2681 			new_deleg->ls_stateid.seqid = delegstateidp->seqid = 0;
2682 			new_deleg->ls_stateid.other[0] = delegstateidp->other[0]
2683 			    = clp->lc_clientid.lval[0];
2684 			new_deleg->ls_stateid.other[1] = delegstateidp->other[1]
2685 			    = clp->lc_clientid.lval[1];
2686 			new_deleg->ls_stateid.other[2] = delegstateidp->other[2]
2687 			    = nfsrv_nextstateindex(clp);
2688 			new_deleg->ls_flags = (NFSLCK_DELEGWRITE |
2689 			    NFSLCK_READACCESS | NFSLCK_WRITEACCESS);
2690 			*rflagsp |= NFSV4OPEN_WRITEDELEGATE;
2691 			new_deleg->ls_uid = new_stp->ls_uid;
2692 			new_deleg->ls_lfp = lfp;
2693 			new_deleg->ls_clp = clp;
2694 			new_deleg->ls_filerev = filerev;
2695 			new_deleg->ls_compref = nd->nd_compref;
2696 			LIST_INSERT_HEAD(&lfp->lf_deleg, new_deleg, ls_file);
2697 			LIST_INSERT_HEAD(NFSSTATEHASH(clp,
2698 			    new_deleg->ls_stateid), new_deleg, ls_hash);
2699 			LIST_INSERT_HEAD(&clp->lc_deleg, new_deleg, ls_list);
2700 			new_deleg = NULL;
2701 			newnfsstats.srvdelegates++;
2702 			nfsrv_openpluslock++;
2703 			nfsrv_delegatecnt++;
2704 		    }
2705 		} else {
2706 		    new_open->ls_stateid.seqid = 0;
2707 		    new_open->ls_stateid.other[0] = clp->lc_clientid.lval[0];
2708 		    new_open->ls_stateid.other[1] = clp->lc_clientid.lval[1];
2709 		    new_open->ls_stateid.other[2] = nfsrv_nextstateindex(clp);
2710 		    new_open->ls_flags = (new_stp->ls_flags & NFSLCK_SHAREBITS)|
2711 			NFSLCK_OPEN;
2712 		    new_open->ls_uid = new_stp->ls_uid;
2713 		    new_open->ls_openowner = ownerstp;
2714 		    new_open->ls_lfp = lfp;
2715 		    new_open->ls_clp = clp;
2716 		    LIST_INIT(&new_open->ls_open);
2717 		    LIST_INSERT_HEAD(&lfp->lf_open, new_open, ls_file);
2718 		    LIST_INSERT_HEAD(&ownerstp->ls_open, new_open, ls_list);
2719 		    LIST_INSERT_HEAD(NFSSTATEHASH(clp, new_open->ls_stateid),
2720 			new_open, ls_hash);
2721 		    openstp = new_open;
2722 		    new_open = NULL;
2723 		    newnfsstats.srvopens++;
2724 		    nfsrv_openpluslock++;
2725 
2726 		    /*
2727 		     * This is where we can choose to issue a delegation.
2728 		     */
2729 		    if (delegate && nfsrv_issuedelegs &&
2730 			(writedeleg || readonly) &&
2731 			(clp->lc_flags & (LCL_CALLBACKSON | LCL_CBDOWN)) ==
2732 			 LCL_CALLBACKSON &&
2733 			!NFSRV_V4DELEGLIMIT(nfsrv_delegatecnt) &&
2734 			NFSVNO_DELEGOK(vp)) {
2735 			new_deleg->ls_stateid.seqid = delegstateidp->seqid = 0;
2736 			new_deleg->ls_stateid.other[0] = delegstateidp->other[0]
2737 			    = clp->lc_clientid.lval[0];
2738 			new_deleg->ls_stateid.other[1] = delegstateidp->other[1]
2739 			    = clp->lc_clientid.lval[1];
2740 			new_deleg->ls_stateid.other[2] = delegstateidp->other[2]
2741 			    = nfsrv_nextstateindex(clp);
2742 			if (writedeleg && !NFSVNO_EXRDONLY(exp) &&
2743 			    (nfsrv_writedelegifpos || !readonly)) {
2744 			    new_deleg->ls_flags = (NFSLCK_DELEGWRITE |
2745 				NFSLCK_READACCESS | NFSLCK_WRITEACCESS);
2746 			    *rflagsp |= NFSV4OPEN_WRITEDELEGATE;
2747 			} else {
2748 			    new_deleg->ls_flags = (NFSLCK_DELEGREAD |
2749 				NFSLCK_READACCESS);
2750 			    *rflagsp |= NFSV4OPEN_READDELEGATE;
2751 			}
2752 			new_deleg->ls_uid = new_stp->ls_uid;
2753 			new_deleg->ls_lfp = lfp;
2754 			new_deleg->ls_clp = clp;
2755 			new_deleg->ls_filerev = filerev;
2756 			new_deleg->ls_compref = nd->nd_compref;
2757 			LIST_INSERT_HEAD(&lfp->lf_deleg, new_deleg, ls_file);
2758 			LIST_INSERT_HEAD(NFSSTATEHASH(clp,
2759 			    new_deleg->ls_stateid), new_deleg, ls_hash);
2760 			LIST_INSERT_HEAD(&clp->lc_deleg, new_deleg, ls_list);
2761 			new_deleg = NULL;
2762 			newnfsstats.srvdelegates++;
2763 			nfsrv_openpluslock++;
2764 			nfsrv_delegatecnt++;
2765 		    }
2766 		}
2767 	} else {
2768 		/*
2769 		 * New owner case. Start the open_owner sequence with a
2770 		 * Needs confirmation (unless a reclaim) and hang the
2771 		 * new open off it.
2772 		 */
2773 		new_open->ls_stateid.seqid = 0;
2774 		new_open->ls_stateid.other[0] = clp->lc_clientid.lval[0];
2775 		new_open->ls_stateid.other[1] = clp->lc_clientid.lval[1];
2776 		new_open->ls_stateid.other[2] = nfsrv_nextstateindex(clp);
2777 		new_open->ls_flags = (new_stp->ls_flags & NFSLCK_SHAREBITS) |
2778 		    NFSLCK_OPEN;
2779 		new_open->ls_uid = new_stp->ls_uid;
2780 		LIST_INIT(&new_open->ls_open);
2781 		new_open->ls_openowner = new_stp;
2782 		new_open->ls_lfp = lfp;
2783 		new_open->ls_clp = clp;
2784 		LIST_INSERT_HEAD(&lfp->lf_open, new_open, ls_file);
2785 		if (new_stp->ls_flags & NFSLCK_RECLAIM) {
2786 			new_stp->ls_flags = 0;
2787 		} else {
2788 			*rflagsp |= NFSV4OPEN_RESULTCONFIRM;
2789 			new_stp->ls_flags = NFSLCK_NEEDSCONFIRM;
2790 		}
2791 		nfsrvd_refcache(new_stp->ls_op);
2792 		new_stp->ls_noopens = 0;
2793 		LIST_INIT(&new_stp->ls_open);
2794 		LIST_INSERT_HEAD(&new_stp->ls_open, new_open, ls_list);
2795 		LIST_INSERT_HEAD(&clp->lc_open, new_stp, ls_list);
2796 		LIST_INSERT_HEAD(NFSSTATEHASH(clp, new_open->ls_stateid),
2797 		    new_open, ls_hash);
2798 		openstp = new_open;
2799 		new_open = NULL;
2800 		*new_stpp = NULL;
2801 		newnfsstats.srvopens++;
2802 		nfsrv_openpluslock++;
2803 		newnfsstats.srvopenowners++;
2804 		nfsrv_openpluslock++;
2805 	}
2806 	if (!error) {
2807 		stateidp->seqid = openstp->ls_stateid.seqid;
2808 		stateidp->other[0] = openstp->ls_stateid.other[0];
2809 		stateidp->other[1] = openstp->ls_stateid.other[1];
2810 		stateidp->other[2] = openstp->ls_stateid.other[2];
2811 	}
2812 	NFSUNLOCKSTATE();
2813 	if (haslock) {
2814 		NFSLOCKV4ROOTMUTEX();
2815 		nfsv4_unlock(&nfsv4rootfs_lock, 1);
2816 		NFSUNLOCKV4ROOTMUTEX();
2817 	}
2818 	if (new_open)
2819 		FREE((caddr_t)new_open, M_NFSDSTATE);
2820 	if (new_deleg)
2821 		FREE((caddr_t)new_deleg, M_NFSDSTATE);
2822 	return (error);
2823 }
2824 
2825 /*
2826  * Open update. Does the confirm, downgrade and close.
2827  */
2828 APPLESTATIC int
2829 nfsrv_openupdate(vnode_t vp, struct nfsstate *new_stp, nfsquad_t clientid,
2830     nfsv4stateid_t *stateidp, struct nfsrv_descript *nd, NFSPROC_T *p)
2831 {
2832 	struct nfsstate *stp, *ownerstp;
2833 	struct nfsclient *clp;
2834 	struct nfslockfile *lfp;
2835 	u_int32_t bits;
2836 	int error, gotstate = 0, len = 0;
2837 	u_char client[NFSV4_OPAQUELIMIT];
2838 
2839 	/*
2840 	 * Check for restart conditions (client and server).
2841 	 */
2842 	error = nfsrv_checkrestart(clientid, new_stp->ls_flags,
2843 	    &new_stp->ls_stateid, 0);
2844 	if (error)
2845 		return (error);
2846 
2847 	NFSLOCKSTATE();
2848 	/*
2849 	 * Get the open structure via clientid and stateid.
2850 	 */
2851 	error = nfsrv_getclient(clientid, CLOPS_RENEW, &clp,
2852 	    (nfsquad_t)((u_quad_t)0), NULL, p);
2853 	if (!error)
2854 		error = nfsrv_getstate(clp, &new_stp->ls_stateid,
2855 		    new_stp->ls_flags, &stp);
2856 
2857 	/*
2858 	 * Sanity check the open.
2859 	 */
2860 	if (!error && (!(stp->ls_flags & NFSLCK_OPEN) ||
2861 		(!(new_stp->ls_flags & NFSLCK_CONFIRM) &&
2862 		 (stp->ls_openowner->ls_flags & NFSLCK_NEEDSCONFIRM)) ||
2863 		((new_stp->ls_flags & NFSLCK_CONFIRM) &&
2864 		 (!(stp->ls_openowner->ls_flags & NFSLCK_NEEDSCONFIRM)))))
2865 		error = NFSERR_BADSTATEID;
2866 
2867 	if (!error)
2868 		error = nfsrv_checkseqid(nd, new_stp->ls_seq,
2869 		    stp->ls_openowner, new_stp->ls_op);
2870 	if (!error && stp->ls_stateid.seqid != new_stp->ls_stateid.seqid &&
2871 	    !(new_stp->ls_flags & NFSLCK_CONFIRM))
2872 		error = NFSERR_OLDSTATEID;
2873 	if (!error && vnode_vtype(vp) != VREG) {
2874 		if (vnode_vtype(vp) == VDIR)
2875 			error = NFSERR_ISDIR;
2876 		else
2877 			error = NFSERR_INVAL;
2878 	}
2879 
2880 	if (error) {
2881 		/*
2882 		 * If a client tries to confirm an Open with a bad
2883 		 * seqid# and there are no byte range locks or other Opens
2884 		 * on the openowner, just throw it away, so the next use of the
2885 		 * openowner will start a fresh seq#.
2886 		 */
2887 		if (error == NFSERR_BADSEQID &&
2888 		    (new_stp->ls_flags & NFSLCK_CONFIRM) &&
2889 		    nfsrv_nootherstate(stp))
2890 			nfsrv_freeopenowner(stp->ls_openowner, 0, p);
2891 		NFSUNLOCKSTATE();
2892 		return (error);
2893 	}
2894 
2895 	/*
2896 	 * Set the return stateid.
2897 	 */
2898 	stateidp->seqid = stp->ls_stateid.seqid + 1;
2899 	stateidp->other[0] = stp->ls_stateid.other[0];
2900 	stateidp->other[1] = stp->ls_stateid.other[1];
2901 	stateidp->other[2] = stp->ls_stateid.other[2];
2902 	/*
2903 	 * Now, handle the three cases.
2904 	 */
2905 	if (new_stp->ls_flags & NFSLCK_CONFIRM) {
2906 		/*
2907 		 * If the open doesn't need confirmation, it seems to me that
2908 		 * there is a client error, but I'll just log it and keep going?
2909 		 */
2910 		if (!(stp->ls_openowner->ls_flags & NFSLCK_NEEDSCONFIRM))
2911 			printf("Nfsv4d: stray open confirm\n");
2912 		stp->ls_openowner->ls_flags = 0;
2913 		stp->ls_stateid.seqid++;
2914 		if (!(clp->lc_flags & LCL_STAMPEDSTABLE)) {
2915 			clp->lc_flags |= LCL_STAMPEDSTABLE;
2916 			len = clp->lc_idlen;
2917 			NFSBCOPY(clp->lc_id, client, len);
2918 			gotstate = 1;
2919 		}
2920 		NFSUNLOCKSTATE();
2921 	} else if (new_stp->ls_flags & NFSLCK_CLOSE) {
2922 		ownerstp = stp->ls_openowner;
2923 		lfp = stp->ls_lfp;
2924 		if (nfsrv_dolocallocks != 0 && !LIST_EMPTY(&stp->ls_open)) {
2925 			/* Get the lf lock */
2926 			nfsrv_locklf(lfp);
2927 			NFSUNLOCKSTATE();
2928 			if (nfsrv_freeopen(stp, vp, 1, p) == 0) {
2929 				NFSLOCKSTATE();
2930 				nfsrv_unlocklf(lfp);
2931 				NFSUNLOCKSTATE();
2932 			}
2933 		} else {
2934 			(void) nfsrv_freeopen(stp, NULL, 0, p);
2935 			NFSUNLOCKSTATE();
2936 		}
2937 	} else {
2938 		/*
2939 		 * Update the share bits, making sure that the new set are a
2940 		 * subset of the old ones.
2941 		 */
2942 		bits = (new_stp->ls_flags & NFSLCK_SHAREBITS);
2943 		if (~(stp->ls_flags) & bits) {
2944 			NFSUNLOCKSTATE();
2945 			return (NFSERR_INVAL);
2946 		}
2947 		stp->ls_flags = (bits | NFSLCK_OPEN);
2948 		stp->ls_stateid.seqid++;
2949 		NFSUNLOCKSTATE();
2950 	}
2951 
2952 	/*
2953 	 * If the client just confirmed its first open, write a timestamp
2954 	 * to the stable storage file.
2955 	 */
2956 	if (gotstate)
2957 		nfsrv_writestable(client, len, NFSNST_NEWSTATE, p);
2958 	return (error);
2959 }
2960 
2961 /*
2962  * Delegation update. Does the purge and return.
2963  */
2964 APPLESTATIC int
2965 nfsrv_delegupdate(nfsquad_t clientid, nfsv4stateid_t *stateidp,
2966     vnode_t vp, int op, struct ucred *cred, NFSPROC_T *p)
2967 {
2968 	struct nfsstate *stp;
2969 	struct nfsclient *clp;
2970 	int error;
2971 	fhandle_t fh;
2972 
2973 	/*
2974 	 * Do a sanity check against the file handle for DelegReturn.
2975 	 */
2976 	if (vp) {
2977 		error = nfsvno_getfh(vp, &fh, p);
2978 		if (error)
2979 			return (error);
2980 	}
2981 	/*
2982 	 * Check for restart conditions (client and server).
2983 	 */
2984 	if (op == NFSV4OP_DELEGRETURN)
2985 		error = nfsrv_checkrestart(clientid, NFSLCK_DELEGRETURN,
2986 			stateidp, 0);
2987 	else
2988 		error = nfsrv_checkrestart(clientid, NFSLCK_DELEGPURGE,
2989 			stateidp, 0);
2990 
2991 	NFSLOCKSTATE();
2992 	/*
2993 	 * Get the open structure via clientid and stateid.
2994 	 */
2995 	if (!error)
2996 	    error = nfsrv_getclient(clientid, CLOPS_RENEW, &clp,
2997 		(nfsquad_t)((u_quad_t)0), NULL, p);
2998 	if (error) {
2999 		if (error == NFSERR_CBPATHDOWN)
3000 			error = 0;
3001 		if (error == NFSERR_STALECLIENTID && op == NFSV4OP_DELEGRETURN)
3002 			error = NFSERR_STALESTATEID;
3003 	}
3004 	if (!error && op == NFSV4OP_DELEGRETURN) {
3005 	    error = nfsrv_getstate(clp, stateidp, NFSLCK_DELEGRETURN, &stp);
3006 	    if (!error && stp->ls_stateid.seqid != stateidp->seqid)
3007 		error = NFSERR_OLDSTATEID;
3008 	}
3009 	/*
3010 	 * NFSERR_EXPIRED means that the state has gone away,
3011 	 * so Delegations have been purged. Just return ok.
3012 	 */
3013 	if (error == NFSERR_EXPIRED && op == NFSV4OP_DELEGPURGE) {
3014 		NFSUNLOCKSTATE();
3015 		return (0);
3016 	}
3017 	if (error) {
3018 		NFSUNLOCKSTATE();
3019 		return (error);
3020 	}
3021 
3022 	if (op == NFSV4OP_DELEGRETURN) {
3023 		if (NFSBCMP((caddr_t)&fh, (caddr_t)&stp->ls_lfp->lf_fh,
3024 		    sizeof (fhandle_t))) {
3025 			NFSUNLOCKSTATE();
3026 			return (NFSERR_BADSTATEID);
3027 		}
3028 		nfsrv_freedeleg(stp);
3029 	} else {
3030 		nfsrv_freedeleglist(&clp->lc_olddeleg);
3031 	}
3032 	NFSUNLOCKSTATE();
3033 	return (0);
3034 }
3035 
3036 /*
3037  * Release lock owner.
3038  */
3039 APPLESTATIC int
3040 nfsrv_releaselckown(struct nfsstate *new_stp, nfsquad_t clientid,
3041     NFSPROC_T *p)
3042 {
3043 	struct nfsstate *stp, *nstp, *openstp, *ownstp;
3044 	struct nfsclient *clp;
3045 	int error;
3046 
3047 	/*
3048 	 * Check for restart conditions (client and server).
3049 	 */
3050 	error = nfsrv_checkrestart(clientid, new_stp->ls_flags,
3051 	    &new_stp->ls_stateid, 0);
3052 	if (error)
3053 		return (error);
3054 
3055 	NFSLOCKSTATE();
3056 	/*
3057 	 * Get the lock owner by name.
3058 	 */
3059 	error = nfsrv_getclient(clientid, CLOPS_RENEW, &clp,
3060 	    (nfsquad_t)((u_quad_t)0), NULL, p);
3061 	if (error) {
3062 		NFSUNLOCKSTATE();
3063 		return (error);
3064 	}
3065 	LIST_FOREACH(ownstp, &clp->lc_open, ls_list) {
3066 	    LIST_FOREACH(openstp, &ownstp->ls_open, ls_list) {
3067 		stp = LIST_FIRST(&openstp->ls_open);
3068 		while (stp != LIST_END(&openstp->ls_open)) {
3069 		    nstp = LIST_NEXT(stp, ls_list);
3070 		    /*
3071 		     * If the owner matches, check for locks and
3072 		     * then free or return an error.
3073 		     */
3074 		    if (stp->ls_ownerlen == new_stp->ls_ownerlen &&
3075 			!NFSBCMP(stp->ls_owner, new_stp->ls_owner,
3076 			 stp->ls_ownerlen)){
3077 			if (LIST_EMPTY(&stp->ls_lock)) {
3078 			    nfsrv_freelockowner(stp, NULL, 0, p);
3079 			} else {
3080 			    NFSUNLOCKSTATE();
3081 			    return (NFSERR_LOCKSHELD);
3082 			}
3083 		    }
3084 		    stp = nstp;
3085 		}
3086 	    }
3087 	}
3088 	NFSUNLOCKSTATE();
3089 	return (0);
3090 }
3091 
3092 /*
3093  * Get the file handle for a lock structure.
3094  */
3095 static int
3096 nfsrv_getlockfh(vnode_t vp, u_short flags,
3097     struct nfslockfile **new_lfpp, fhandle_t *nfhp, NFSPROC_T *p)
3098 {
3099 	fhandle_t *fhp = NULL;
3100 	struct nfslockfile *new_lfp;
3101 	int error;
3102 
3103 	/*
3104 	 * For lock, use the new nfslock structure, otherwise just
3105 	 * a fhandle_t on the stack.
3106 	 */
3107 	if (flags & NFSLCK_OPEN) {
3108 		new_lfp = *new_lfpp;
3109 		fhp = &new_lfp->lf_fh;
3110 	} else if (nfhp) {
3111 		fhp = nfhp;
3112 	} else {
3113 		panic("nfsrv_getlockfh");
3114 	}
3115 	error = nfsvno_getfh(vp, fhp, p);
3116 	return (error);
3117 }
3118 
3119 /*
3120  * Get an nfs lock structure. Allocate one, as required, and return a
3121  * pointer to it.
3122  * Returns an NFSERR_xxx upon failure or -1 to indicate no current lock.
3123  */
3124 static int
3125 nfsrv_getlockfile(u_short flags, struct nfslockfile **new_lfpp,
3126     struct nfslockfile **lfpp, fhandle_t *nfhp, int lockit)
3127 {
3128 	struct nfslockfile *lfp;
3129 	fhandle_t *fhp = NULL, *tfhp;
3130 	struct nfslockhashhead *hp;
3131 	struct nfslockfile *new_lfp = NULL;
3132 
3133 	/*
3134 	 * For lock, use the new nfslock structure, otherwise just
3135 	 * a fhandle_t on the stack.
3136 	 */
3137 	if (flags & NFSLCK_OPEN) {
3138 		new_lfp = *new_lfpp;
3139 		fhp = &new_lfp->lf_fh;
3140 	} else if (nfhp) {
3141 		fhp = nfhp;
3142 	} else {
3143 		panic("nfsrv_getlockfile");
3144 	}
3145 
3146 	hp = NFSLOCKHASH(fhp);
3147 	LIST_FOREACH(lfp, hp, lf_hash) {
3148 		tfhp = &lfp->lf_fh;
3149 		if (NFSVNO_CMPFH(fhp, tfhp)) {
3150 			if (lockit)
3151 				nfsrv_locklf(lfp);
3152 			*lfpp = lfp;
3153 			return (0);
3154 		}
3155 	}
3156 	if (!(flags & NFSLCK_OPEN))
3157 		return (-1);
3158 
3159 	/*
3160 	 * No match, so chain the new one into the list.
3161 	 */
3162 	LIST_INIT(&new_lfp->lf_open);
3163 	LIST_INIT(&new_lfp->lf_lock);
3164 	LIST_INIT(&new_lfp->lf_deleg);
3165 	LIST_INIT(&new_lfp->lf_locallock);
3166 	LIST_INIT(&new_lfp->lf_rollback);
3167 	new_lfp->lf_locallock_lck.nfslock_usecnt = 0;
3168 	new_lfp->lf_locallock_lck.nfslock_lock = 0;
3169 	new_lfp->lf_usecount = 0;
3170 	LIST_INSERT_HEAD(hp, new_lfp, lf_hash);
3171 	*lfpp = new_lfp;
3172 	*new_lfpp = NULL;
3173 	return (0);
3174 }
3175 
3176 /*
3177  * This function adds a nfslock lock structure to the list for the associated
3178  * nfsstate and nfslockfile structures. It will be inserted after the
3179  * entry pointed at by insert_lop.
3180  * Must be called with soft clock interrupts disabled.
3181  */
3182 static void
3183 nfsrv_insertlock(struct nfslock *new_lop, struct nfslock *insert_lop,
3184     struct nfsstate *stp, struct nfslockfile *lfp)
3185 {
3186 	struct nfslock *lop, *nlop;
3187 
3188 	new_lop->lo_stp = stp;
3189 	new_lop->lo_lfp = lfp;
3190 
3191 	if (stp != NULL) {
3192 		/* Insert in increasing lo_first order */
3193 		lop = LIST_FIRST(&lfp->lf_lock);
3194 		if (lop == LIST_END(&lfp->lf_lock) ||
3195 		    new_lop->lo_first <= lop->lo_first) {
3196 			LIST_INSERT_HEAD(&lfp->lf_lock, new_lop, lo_lckfile);
3197 		} else {
3198 			nlop = LIST_NEXT(lop, lo_lckfile);
3199 			while (nlop != LIST_END(&lfp->lf_lock) &&
3200 			       nlop->lo_first < new_lop->lo_first) {
3201 				lop = nlop;
3202 				nlop = LIST_NEXT(lop, lo_lckfile);
3203 			}
3204 			LIST_INSERT_AFTER(lop, new_lop, lo_lckfile);
3205 		}
3206 	} else {
3207 		new_lop->lo_lckfile.le_prev = NULL;	/* list not used */
3208 	}
3209 
3210 	/*
3211 	 * Insert after insert_lop, which is overloaded as stp or lfp for
3212 	 * an empty list.
3213 	 */
3214 	if (stp == NULL && (struct nfslockfile *)insert_lop == lfp)
3215 		LIST_INSERT_HEAD(&lfp->lf_locallock, new_lop, lo_lckowner);
3216 	else if ((struct nfsstate *)insert_lop == stp)
3217 		LIST_INSERT_HEAD(&stp->ls_lock, new_lop, lo_lckowner);
3218 	else
3219 		LIST_INSERT_AFTER(insert_lop, new_lop, lo_lckowner);
3220 	if (stp != NULL) {
3221 		newnfsstats.srvlocks++;
3222 		nfsrv_openpluslock++;
3223 	}
3224 }
3225 
3226 /*
3227  * This function updates the locking for a lock owner and given file. It
3228  * maintains a list of lock ranges ordered on increasing file offset that
3229  * are NFSLCK_READ or NFSLCK_WRITE and non-overlapping (aka POSIX style).
3230  * It always adds new_lop to the list and sometimes uses the one pointed
3231  * at by other_lopp.
3232  * Must be called with soft clock interrupts disabled.
3233  */
3234 static void
3235 nfsrv_updatelock(struct nfsstate *stp, struct nfslock **new_lopp,
3236     struct nfslock **other_lopp, struct nfslockfile *lfp)
3237 {
3238 	struct nfslock *new_lop = *new_lopp;
3239 	struct nfslock *lop, *tlop, *ilop;
3240 	struct nfslock *other_lop = *other_lopp;
3241 	int unlock = 0, myfile = 0;
3242 	u_int64_t tmp;
3243 
3244 	/*
3245 	 * Work down the list until the lock is merged.
3246 	 */
3247 	if (new_lop->lo_flags & NFSLCK_UNLOCK)
3248 		unlock = 1;
3249 	if (stp != NULL) {
3250 		ilop = (struct nfslock *)stp;
3251 		lop = LIST_FIRST(&stp->ls_lock);
3252 	} else {
3253 		ilop = (struct nfslock *)lfp;
3254 		lop = LIST_FIRST(&lfp->lf_locallock);
3255 	}
3256 	while (lop != NULL) {
3257 	    /*
3258 	     * Only check locks for this file that aren't before the start of
3259 	     * new lock's range.
3260 	     */
3261 	    if (lop->lo_lfp == lfp) {
3262 	      myfile = 1;
3263 	      if (lop->lo_end >= new_lop->lo_first) {
3264 		if (new_lop->lo_end < lop->lo_first) {
3265 			/*
3266 			 * If the new lock ends before the start of the
3267 			 * current lock's range, no merge, just insert
3268 			 * the new lock.
3269 			 */
3270 			break;
3271 		}
3272 		if (new_lop->lo_flags == lop->lo_flags ||
3273 		    (new_lop->lo_first <= lop->lo_first &&
3274 		     new_lop->lo_end >= lop->lo_end)) {
3275 			/*
3276 			 * This lock can be absorbed by the new lock/unlock.
3277 			 * This happens when it covers the entire range
3278 			 * of the old lock or is contiguous
3279 			 * with the old lock and is of the same type or an
3280 			 * unlock.
3281 			 */
3282 			if (lop->lo_first < new_lop->lo_first)
3283 				new_lop->lo_first = lop->lo_first;
3284 			if (lop->lo_end > new_lop->lo_end)
3285 				new_lop->lo_end = lop->lo_end;
3286 			tlop = lop;
3287 			lop = LIST_NEXT(lop, lo_lckowner);
3288 			nfsrv_freenfslock(tlop);
3289 			continue;
3290 		}
3291 
3292 		/*
3293 		 * All these cases are for contiguous locks that are not the
3294 		 * same type, so they can't be merged.
3295 		 */
3296 		if (new_lop->lo_first <= lop->lo_first) {
3297 			/*
3298 			 * This case is where the new lock overlaps with the
3299 			 * first part of the old lock. Move the start of the
3300 			 * old lock to just past the end of the new lock. The
3301 			 * new lock will be inserted in front of the old, since
3302 			 * ilop hasn't been updated. (We are done now.)
3303 			 */
3304 			lop->lo_first = new_lop->lo_end;
3305 			break;
3306 		}
3307 		if (new_lop->lo_end >= lop->lo_end) {
3308 			/*
3309 			 * This case is where the new lock overlaps with the
3310 			 * end of the old lock's range. Move the old lock's
3311 			 * end to just before the new lock's first and insert
3312 			 * the new lock after the old lock.
3313 			 * Might not be done yet, since the new lock could
3314 			 * overlap further locks with higher ranges.
3315 			 */
3316 			lop->lo_end = new_lop->lo_first;
3317 			ilop = lop;
3318 			lop = LIST_NEXT(lop, lo_lckowner);
3319 			continue;
3320 		}
3321 		/*
3322 		 * The final case is where the new lock's range is in the
3323 		 * middle of the current lock's and splits the current lock
3324 		 * up. Use *other_lopp to handle the second part of the
3325 		 * split old lock range. (We are done now.)
3326 		 * For unlock, we use new_lop as other_lop and tmp, since
3327 		 * other_lop and new_lop are the same for this case.
3328 		 * We noted the unlock case above, so we don't need
3329 		 * new_lop->lo_flags any longer.
3330 		 */
3331 		tmp = new_lop->lo_first;
3332 		if (other_lop == NULL) {
3333 			if (!unlock)
3334 				panic("nfsd srv update unlock");
3335 			other_lop = new_lop;
3336 			*new_lopp = NULL;
3337 		}
3338 		other_lop->lo_first = new_lop->lo_end;
3339 		other_lop->lo_end = lop->lo_end;
3340 		other_lop->lo_flags = lop->lo_flags;
3341 		other_lop->lo_stp = stp;
3342 		other_lop->lo_lfp = lfp;
3343 		lop->lo_end = tmp;
3344 		nfsrv_insertlock(other_lop, lop, stp, lfp);
3345 		*other_lopp = NULL;
3346 		ilop = lop;
3347 		break;
3348 	      }
3349 	    }
3350 	    ilop = lop;
3351 	    lop = LIST_NEXT(lop, lo_lckowner);
3352 	    if (myfile && (lop == NULL || lop->lo_lfp != lfp))
3353 		break;
3354 	}
3355 
3356 	/*
3357 	 * Insert the new lock in the list at the appropriate place.
3358 	 */
3359 	if (!unlock) {
3360 		nfsrv_insertlock(new_lop, ilop, stp, lfp);
3361 		*new_lopp = NULL;
3362 	}
3363 }
3364 
3365 /*
3366  * This function handles sequencing of locks, etc.
3367  * It returns an error that indicates what the caller should do.
3368  */
3369 static int
3370 nfsrv_checkseqid(struct nfsrv_descript *nd, u_int32_t seqid,
3371     struct nfsstate *stp, struct nfsrvcache *op)
3372 {
3373 
3374 	if (op != nd->nd_rp)
3375 		panic("nfsrvstate checkseqid");
3376 	if (!(op->rc_flag & RC_INPROG))
3377 		panic("nfsrvstate not inprog");
3378 	if (stp->ls_op && stp->ls_op->rc_refcnt <= 0) {
3379 		printf("refcnt=%d\n", stp->ls_op->rc_refcnt);
3380 		panic("nfsrvstate op refcnt");
3381 	}
3382 	if ((stp->ls_seq + 1) == seqid) {
3383 		if (stp->ls_op)
3384 			nfsrvd_derefcache(stp->ls_op);
3385 		stp->ls_op = op;
3386 		nfsrvd_refcache(op);
3387 		stp->ls_seq = seqid;
3388 		return (0);
3389 	} else if (stp->ls_seq == seqid && stp->ls_op &&
3390 		op->rc_xid == stp->ls_op->rc_xid &&
3391 		op->rc_refcnt == 0 &&
3392 		op->rc_reqlen == stp->ls_op->rc_reqlen &&
3393 		op->rc_cksum == stp->ls_op->rc_cksum) {
3394 		if (stp->ls_op->rc_flag & RC_INPROG)
3395 			return (NFSERR_DONTREPLY);
3396 		nd->nd_rp = stp->ls_op;
3397 		nd->nd_rp->rc_flag |= RC_INPROG;
3398 		nfsrvd_delcache(op);
3399 		return (NFSERR_REPLYFROMCACHE);
3400 	}
3401 	return (NFSERR_BADSEQID);
3402 }
3403 
3404 /*
3405  * Get the client ip address for callbacks. If the strings can't be parsed,
3406  * just set lc_program to 0 to indicate no callbacks are possible.
3407  * (For cases where the address can't be parsed or is 0.0.0.0.0.0, set
3408  *  the address to the client's transport address. This won't be used
3409  *  for callbacks, but can be printed out by newnfsstats for info.)
3410  * Return error if the xdr can't be parsed, 0 otherwise.
3411  */
3412 APPLESTATIC int
3413 nfsrv_getclientipaddr(struct nfsrv_descript *nd, struct nfsclient *clp)
3414 {
3415 	u_int32_t *tl;
3416 	u_char *cp, *cp2;
3417 	int i, j;
3418 	struct sockaddr_in *rad, *sad;
3419 	u_char protocol[5], addr[24];
3420 	int error = 0, cantparse = 0;
3421 	union {
3422 		u_long ival;
3423 		u_char cval[4];
3424 	} ip;
3425 	union {
3426 		u_short sval;
3427 		u_char cval[2];
3428 	} port;
3429 
3430 	rad = NFSSOCKADDR(clp->lc_req.nr_nam, struct sockaddr_in *);
3431 	rad->sin_family = AF_INET;
3432 	rad->sin_len = sizeof (struct sockaddr_in);
3433 	rad->sin_addr.s_addr = 0;
3434 	rad->sin_port = 0;
3435 	clp->lc_req.nr_client = NULL;
3436 	clp->lc_req.nr_lock = 0;
3437 	NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
3438 	i = fxdr_unsigned(int, *tl);
3439 	if (i >= 3 && i <= 4) {
3440 		error = nfsrv_mtostr(nd, protocol, i);
3441 		if (error)
3442 			goto nfsmout;
3443 		if (!strcmp(protocol, "tcp")) {
3444 			clp->lc_flags |= LCL_TCPCALLBACK;
3445 			clp->lc_req.nr_sotype = SOCK_STREAM;
3446 			clp->lc_req.nr_soproto = IPPROTO_TCP;
3447 		} else if (!strcmp(protocol, "udp")) {
3448 			clp->lc_req.nr_sotype = SOCK_DGRAM;
3449 			clp->lc_req.nr_soproto = IPPROTO_UDP;
3450 		} else {
3451 			cantparse = 1;
3452 		}
3453 	} else {
3454 		cantparse = 1;
3455 		if (i > 0) {
3456 			error = nfsm_advance(nd, NFSM_RNDUP(i), -1);
3457 			if (error)
3458 				goto nfsmout;
3459 		}
3460 	}
3461 	NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
3462 	i = fxdr_unsigned(int, *tl);
3463 	if (i < 0) {
3464 		error = NFSERR_BADXDR;
3465 		goto nfsmout;
3466 	} else if (i == 0) {
3467 		cantparse = 1;
3468 	} else if (!cantparse && i <= 23 && i >= 11) {
3469 		error = nfsrv_mtostr(nd, addr, i);
3470 		if (error)
3471 			goto nfsmout;
3472 
3473 		/*
3474 		 * Parse out the address fields. We expect 6 decimal numbers
3475 		 * separated by '.'s.
3476 		 */
3477 		cp = addr;
3478 		i = 0;
3479 		while (*cp && i < 6) {
3480 			cp2 = cp;
3481 			while (*cp2 && *cp2 != '.')
3482 				cp2++;
3483 			if (*cp2)
3484 				*cp2++ = '\0';
3485 			else if (i != 5) {
3486 				cantparse = 1;
3487 				break;
3488 			}
3489 			j = nfsrv_getipnumber(cp);
3490 			if (j >= 0) {
3491 				if (i < 4)
3492 					ip.cval[3 - i] = j;
3493 				else
3494 					port.cval[5 - i] = j;
3495 			} else {
3496 				cantparse = 1;
3497 				break;
3498 			}
3499 			cp = cp2;
3500 			i++;
3501 		}
3502 		if (!cantparse) {
3503 			if (ip.ival != 0x0) {
3504 				rad->sin_addr.s_addr = htonl(ip.ival);
3505 				rad->sin_port = htons(port.sval);
3506 			} else {
3507 				cantparse = 1;
3508 			}
3509 		}
3510 	} else {
3511 		cantparse = 1;
3512 		if (i > 0) {
3513 			error = nfsm_advance(nd, NFSM_RNDUP(i), -1);
3514 			if (error)
3515 				goto nfsmout;
3516 		}
3517 	}
3518 	if (cantparse) {
3519 		sad = NFSSOCKADDR(nd->nd_nam, struct sockaddr_in *);
3520 		rad->sin_addr.s_addr = sad->sin_addr.s_addr;
3521 		rad->sin_port = 0x0;
3522 		clp->lc_program = 0;
3523 	}
3524 nfsmout:
3525 	return (error);
3526 }
3527 
3528 /*
3529  * Turn a string of up to three decimal digits into a number. Return -1 upon
3530  * error.
3531  */
3532 static int
3533 nfsrv_getipnumber(u_char *cp)
3534 {
3535 	int i = 0, j = 0;
3536 
3537 	while (*cp) {
3538 		if (j > 2 || *cp < '0' || *cp > '9')
3539 			return (-1);
3540 		i *= 10;
3541 		i += (*cp - '0');
3542 		cp++;
3543 		j++;
3544 	}
3545 	if (i < 256)
3546 		return (i);
3547 	return (-1);
3548 }
3549 
3550 /*
3551  * This function checks for restart conditions.
3552  */
3553 static int
3554 nfsrv_checkrestart(nfsquad_t clientid, u_int32_t flags,
3555     nfsv4stateid_t *stateidp, int specialid)
3556 {
3557 	int ret;
3558 
3559 	/*
3560 	 * First check for a server restart. Open, LockT, ReleaseLockOwner
3561 	 * and DelegPurge have a clientid, the rest a stateid.
3562 	 */
3563 	if (flags &
3564 	    (NFSLCK_OPEN | NFSLCK_TEST | NFSLCK_RELEASE | NFSLCK_DELEGPURGE)) {
3565 		if (clientid.lval[0] != nfsrvboottime)
3566 			return (NFSERR_STALECLIENTID);
3567 	} else if (stateidp->other[0] != nfsrvboottime &&
3568 		specialid == 0)
3569 		return (NFSERR_STALESTATEID);
3570 
3571 	/*
3572 	 * Read, Write, Setattr and LockT can return NFSERR_GRACE and do
3573 	 * not use a lock/open owner seqid#, so the check can be done now.
3574 	 * (The others will be checked, as required, later.)
3575 	 */
3576 	if (!(flags & (NFSLCK_CHECK | NFSLCK_TEST)))
3577 		return (0);
3578 
3579 	NFSLOCKSTATE();
3580 	ret = nfsrv_checkgrace(flags);
3581 	NFSUNLOCKSTATE();
3582 	return (ret);
3583 }
3584 
3585 /*
3586  * Check for grace.
3587  */
3588 static int
3589 nfsrv_checkgrace(u_int32_t flags)
3590 {
3591 
3592 	if (nfsrv_stablefirst.nsf_flags & NFSNSF_GRACEOVER) {
3593 		if (flags & NFSLCK_RECLAIM)
3594 			return (NFSERR_NOGRACE);
3595 	} else {
3596 		if (!(flags & NFSLCK_RECLAIM))
3597 			return (NFSERR_GRACE);
3598 
3599 		/*
3600 		 * If grace is almost over and we are still getting Reclaims,
3601 		 * extend grace a bit.
3602 		 */
3603 		if ((NFSD_MONOSEC + NFSRV_LEASEDELTA) >
3604 		    nfsrv_stablefirst.nsf_eograce)
3605 			nfsrv_stablefirst.nsf_eograce = NFSD_MONOSEC +
3606 				NFSRV_LEASEDELTA;
3607 	}
3608 	return (0);
3609 }
3610 
3611 /*
3612  * Do a server callback.
3613  */
3614 static int
3615 nfsrv_docallback(struct nfsclient *clp, int procnum,
3616     nfsv4stateid_t *stateidp, int trunc, fhandle_t *fhp,
3617     struct nfsvattr *nap, nfsattrbit_t *attrbitp, NFSPROC_T *p)
3618 {
3619 	mbuf_t m;
3620 	u_int32_t *tl;
3621 	struct nfsrv_descript nfsd, *nd = &nfsd;
3622 	struct ucred *cred;
3623 	int error = 0;
3624 	u_int32_t callback;
3625 
3626 	cred = newnfs_getcred();
3627 	NFSLOCKSTATE();	/* mostly for lc_cbref++ */
3628 	if (clp->lc_flags & LCL_NEEDSCONFIRM) {
3629 		NFSUNLOCKSTATE();
3630 		panic("docallb");
3631 	}
3632 	clp->lc_cbref++;
3633 
3634 	/*
3635 	 * Fill the callback program# and version into the request
3636 	 * structure for newnfs_connect() to use.
3637 	 */
3638 	clp->lc_req.nr_prog = clp->lc_program;
3639 	clp->lc_req.nr_vers = NFSV4_CBVERS;
3640 
3641 	/*
3642 	 * First, fill in some of the fields of nd and cr.
3643 	 */
3644 	nd->nd_flag = ND_NFSV4;
3645 	if (clp->lc_flags & LCL_GSS)
3646 		nd->nd_flag |= ND_KERBV;
3647 	nd->nd_repstat = 0;
3648 	cred->cr_uid = clp->lc_uid;
3649 	cred->cr_gid = clp->lc_gid;
3650 	callback = clp->lc_callback;
3651 	NFSUNLOCKSTATE();
3652 	cred->cr_ngroups = 1;
3653 
3654 	/*
3655 	 * Get the first mbuf for the request.
3656 	 */
3657 	MGET(m, M_WAIT, MT_DATA);
3658 	mbuf_setlen(m, 0);
3659 	nd->nd_mreq = nd->nd_mb = m;
3660 	nd->nd_bpos = NFSMTOD(m, caddr_t);
3661 
3662 	/*
3663 	 * and build the callback request.
3664 	 */
3665 	if (procnum == NFSV4OP_CBGETATTR) {
3666 		nd->nd_procnum = NFSV4PROC_CBCOMPOUND;
3667 		(void) nfsm_strtom(nd, "CB Getattr", 10);
3668 		NFSM_BUILD(tl, u_int32_t *, 4 * NFSX_UNSIGNED);
3669 		*tl++ = txdr_unsigned(NFSV4_MINORVERSION);
3670 		*tl++ = txdr_unsigned(callback);
3671 		*tl++ = txdr_unsigned(1);
3672 		*tl = txdr_unsigned(NFSV4OP_CBGETATTR);
3673 		(void) nfsm_fhtom(nd, (u_int8_t *)fhp, NFSX_MYFH, 0);
3674 		(void) nfsrv_putattrbit(nd, attrbitp);
3675 	} else if (procnum == NFSV4OP_CBRECALL) {
3676 		nd->nd_procnum = NFSV4PROC_CBCOMPOUND;
3677 		(void) nfsm_strtom(nd, "CB Recall", 9);
3678 		NFSM_BUILD(tl, u_int32_t *, 5 * NFSX_UNSIGNED + NFSX_STATEID);
3679 		*tl++ = txdr_unsigned(NFSV4_MINORVERSION);
3680 		*tl++ = txdr_unsigned(callback);
3681 		*tl++ = txdr_unsigned(1);
3682 		*tl++ = txdr_unsigned(NFSV4OP_CBRECALL);
3683 		*tl++ = txdr_unsigned(stateidp->seqid);
3684 		NFSBCOPY((caddr_t)stateidp->other, (caddr_t)tl,
3685 		    NFSX_STATEIDOTHER);
3686 		tl += (NFSX_STATEIDOTHER / NFSX_UNSIGNED);
3687 		if (trunc)
3688 			*tl = newnfs_true;
3689 		else
3690 			*tl = newnfs_false;
3691 		(void) nfsm_fhtom(nd, (u_int8_t *)fhp, NFSX_MYFH, 0);
3692 	} else {
3693 		nd->nd_procnum = NFSV4PROC_CBNULL;
3694 	}
3695 
3696 	/*
3697 	 * Call newnfs_connect(), as required, and then newnfs_request().
3698 	 */
3699 	(void) newnfs_sndlock(&clp->lc_req.nr_lock);
3700 	if (clp->lc_req.nr_client == NULL) {
3701 		if (nd->nd_procnum == NFSV4PROC_CBNULL)
3702 			error = newnfs_connect(NULL, &clp->lc_req, cred,
3703 			    NULL, 1);
3704 		else
3705 			error = newnfs_connect(NULL, &clp->lc_req, cred,
3706 			    NULL, 3);
3707 	}
3708 	newnfs_sndunlock(&clp->lc_req.nr_lock);
3709 	if (!error) {
3710 		error = newnfs_request(nd, NULL, clp, &clp->lc_req, NULL,
3711 		    NULL, cred, clp->lc_program, NFSV4_CBVERS, NULL, 1, NULL);
3712 	}
3713 	NFSFREECRED(cred);
3714 
3715 	/*
3716 	 * If error is set here, the Callback path isn't working
3717 	 * properly, so twiddle the appropriate LCL_ flags.
3718 	 * (nd_repstat != 0 indicates the Callback path is working,
3719 	 *  but the callback failed on the client.)
3720 	 */
3721 	if (error) {
3722 		/*
3723 		 * Mark the callback pathway down, which disabled issuing
3724 		 * of delegations and gets Renew to return NFSERR_CBPATHDOWN.
3725 		 */
3726 		NFSLOCKSTATE();
3727 		clp->lc_flags |= LCL_CBDOWN;
3728 		NFSUNLOCKSTATE();
3729 	} else {
3730 		/*
3731 		 * Callback worked. If the callback path was down, disable
3732 		 * callbacks, so no more delegations will be issued. (This
3733 		 * is done on the assumption that the callback pathway is
3734 		 * flakey.)
3735 		 */
3736 		NFSLOCKSTATE();
3737 		if (clp->lc_flags & LCL_CBDOWN)
3738 			clp->lc_flags &= ~(LCL_CBDOWN | LCL_CALLBACKSON);
3739 		NFSUNLOCKSTATE();
3740 		if (nd->nd_repstat)
3741 			error = nd->nd_repstat;
3742 		else if (procnum == NFSV4OP_CBGETATTR)
3743 			error = nfsv4_loadattr(nd, NULL, nap, NULL, NULL, 0,
3744 			    NULL, NULL, NULL, NULL, NULL, 0, NULL, NULL, NULL,
3745 			    p, NULL);
3746 		mbuf_freem(nd->nd_mrep);
3747 	}
3748 	NFSLOCKSTATE();
3749 	clp->lc_cbref--;
3750 	if ((clp->lc_flags & LCL_WAKEUPWANTED) && clp->lc_cbref == 0) {
3751 		clp->lc_flags &= ~LCL_WAKEUPWANTED;
3752 		NFSUNLOCKSTATE();
3753 		wakeup((caddr_t)clp);
3754 	} else {
3755 		NFSUNLOCKSTATE();
3756 	}
3757 	return (error);
3758 }
3759 
3760 /*
3761  * Return the next index# for a clientid. Mostly just increment and return
3762  * the next one, but... if the 32bit unsigned does actually wrap around,
3763  * it should be rebooted.
3764  * At an average rate of one new client per second, it will wrap around in
3765  * approximately 136 years. (I think the server will have been shut
3766  * down or rebooted before then.)
3767  */
3768 static u_int32_t
3769 nfsrv_nextclientindex(void)
3770 {
3771 	static u_int32_t client_index = 0;
3772 
3773 	client_index++;
3774 	if (client_index != 0)
3775 		return (client_index);
3776 
3777 	printf("%s: out of clientids\n", __func__);
3778 	return (client_index);
3779 }
3780 
3781 /*
3782  * Return the next index# for a stateid. Mostly just increment and return
3783  * the next one, but... if the 32bit unsigned does actually wrap around
3784  * (will a BSD server stay up that long?), find
3785  * new start and end values.
3786  */
3787 static u_int32_t
3788 nfsrv_nextstateindex(struct nfsclient *clp)
3789 {
3790 	struct nfsstate *stp;
3791 	int i;
3792 	u_int32_t canuse, min_index, max_index;
3793 
3794 	if (!(clp->lc_flags & LCL_INDEXNOTOK)) {
3795 		clp->lc_stateindex++;
3796 		if (clp->lc_stateindex != clp->lc_statemaxindex)
3797 			return (clp->lc_stateindex);
3798 	}
3799 
3800 	/*
3801 	 * Yuck, we've hit the end.
3802 	 * Look for a new min and max.
3803 	 */
3804 	min_index = 0;
3805 	max_index = 0xffffffff;
3806 	for (i = 0; i < NFSSTATEHASHSIZE; i++) {
3807 	    LIST_FOREACH(stp, &clp->lc_stateid[i], ls_hash) {
3808 		if (stp->ls_stateid.other[2] > 0x80000000) {
3809 		    if (stp->ls_stateid.other[2] < max_index)
3810 			max_index = stp->ls_stateid.other[2];
3811 		} else {
3812 		    if (stp->ls_stateid.other[2] > min_index)
3813 			min_index = stp->ls_stateid.other[2];
3814 		}
3815 	    }
3816 	}
3817 
3818 	/*
3819 	 * Yikes, highly unlikely, but I'll handle it anyhow.
3820 	 */
3821 	if (min_index == 0x80000000 && max_index == 0x80000001) {
3822 	    canuse = 0;
3823 	    /*
3824 	     * Loop around until we find an unused entry. Return that
3825 	     * and set LCL_INDEXNOTOK, so the search will continue next time.
3826 	     * (This is one of those rare cases where a goto is the
3827 	     *  cleanest way to code the loop.)
3828 	     */
3829 tryagain:
3830 	    for (i = 0; i < NFSSTATEHASHSIZE; i++) {
3831 		LIST_FOREACH(stp, &clp->lc_stateid[i], ls_hash) {
3832 		    if (stp->ls_stateid.other[2] == canuse) {
3833 			canuse++;
3834 			goto tryagain;
3835 		    }
3836 		}
3837 	    }
3838 	    clp->lc_flags |= LCL_INDEXNOTOK;
3839 	    return (canuse);
3840 	}
3841 
3842 	/*
3843 	 * Ok to start again from min + 1.
3844 	 */
3845 	clp->lc_stateindex = min_index + 1;
3846 	clp->lc_statemaxindex = max_index;
3847 	clp->lc_flags &= ~LCL_INDEXNOTOK;
3848 	return (clp->lc_stateindex);
3849 }
3850 
3851 /*
3852  * The following functions handle the stable storage file that deals with
3853  * the edge conditions described in RFC3530 Sec. 8.6.3.
3854  * The file is as follows:
3855  * - a single record at the beginning that has the lease time of the
3856  *   previous server instance (before the last reboot) and the nfsrvboottime
3857  *   values for the previous server boots.
3858  *   These previous boot times are used to ensure that the current
3859  *   nfsrvboottime does not, somehow, get set to a previous one.
3860  *   (This is important so that Stale ClientIDs and StateIDs can
3861  *    be recognized.)
3862  *   The number of previous nfsvrboottime values preceeds the list.
3863  * - followed by some number of appended records with:
3864  *   - client id string
3865  *   - flag that indicates it is a record revoking state via lease
3866  *     expiration or similar
3867  *     OR has successfully acquired state.
3868  * These structures vary in length, with the client string at the end, up
3869  * to NFSV4_OPAQUELIMIT in size.
3870  *
3871  * At the end of the grace period, the file is truncated, the first
3872  * record is rewritten with updated information and any acquired state
3873  * records for successful reclaims of state are written.
3874  *
3875  * Subsequent records are appended when the first state is issued to
3876  * a client and when state is revoked for a client.
3877  *
3878  * When reading the file in, state issued records that come later in
3879  * the file override older ones, since the append log is in cronological order.
3880  * If, for some reason, the file can't be read, the grace period is
3881  * immediately terminated and all reclaims get NFSERR_NOGRACE.
3882  */
3883 
3884 /*
3885  * Read in the stable storage file. Called by nfssvc() before the nfsd
3886  * processes start servicing requests.
3887  */
3888 APPLESTATIC void
3889 nfsrv_setupstable(NFSPROC_T *p)
3890 {
3891 	struct nfsrv_stablefirst *sf = &nfsrv_stablefirst;
3892 	struct nfsrv_stable *sp, *nsp;
3893 	struct nfst_rec *tsp;
3894 	int error, i, tryagain;
3895 	off_t off = 0;
3896 	int aresid, len;
3897 	struct timeval curtime;
3898 
3899 	/*
3900 	 * If NFSNSF_UPDATEDONE is set, this is a restart of the nfsds without
3901 	 * a reboot, so state has not been lost.
3902 	 */
3903 	if (sf->nsf_flags & NFSNSF_UPDATEDONE)
3904 		return;
3905 	/*
3906 	 * Set Grace over just until the file reads successfully.
3907 	 */
3908 	NFSGETTIME(&curtime);
3909 	nfsrvboottime = curtime.tv_sec;
3910 	LIST_INIT(&sf->nsf_head);
3911 	sf->nsf_flags = (NFSNSF_GRACEOVER | NFSNSF_NEEDLOCK);
3912 	sf->nsf_eograce = NFSD_MONOSEC + NFSRV_LEASEDELTA;
3913 	if (sf->nsf_fp == NULL)
3914 		return;
3915 	error = NFSD_RDWR(UIO_READ, NFSFPVNODE(sf->nsf_fp),
3916 	    (caddr_t)&sf->nsf_rec, sizeof (struct nfsf_rec), off, UIO_SYSSPACE,
3917 	    0, NFSFPCRED(sf->nsf_fp), &aresid, p);
3918 	if (error || aresid || sf->nsf_numboots == 0 ||
3919 		sf->nsf_numboots > NFSNSF_MAXNUMBOOTS)
3920 		return;
3921 
3922 	/*
3923 	 * Now, read in the boottimes.
3924 	 */
3925 	sf->nsf_bootvals = (time_t *)malloc((sf->nsf_numboots + 1) *
3926 		sizeof (time_t), M_TEMP, M_WAITOK);
3927 	off = sizeof (struct nfsf_rec);
3928 	error = NFSD_RDWR(UIO_READ, NFSFPVNODE(sf->nsf_fp),
3929 	    (caddr_t)sf->nsf_bootvals, sf->nsf_numboots * sizeof (time_t), off,
3930 	    UIO_SYSSPACE, 0, NFSFPCRED(sf->nsf_fp), &aresid, p);
3931 	if (error || aresid) {
3932 		free((caddr_t)sf->nsf_bootvals, M_TEMP);
3933 		sf->nsf_bootvals = NULL;
3934 		return;
3935 	}
3936 
3937 	/*
3938 	 * Make sure this nfsrvboottime is different from all recorded
3939 	 * previous ones.
3940 	 */
3941 	do {
3942 		tryagain = 0;
3943 		for (i = 0; i < sf->nsf_numboots; i++) {
3944 			if (nfsrvboottime == sf->nsf_bootvals[i]) {
3945 				nfsrvboottime++;
3946 				tryagain = 1;
3947 				break;
3948 			}
3949 		}
3950 	} while (tryagain);
3951 
3952 	sf->nsf_flags |= NFSNSF_OK;
3953 	off += (sf->nsf_numboots * sizeof (time_t));
3954 
3955 	/*
3956 	 * Read through the file, building a list of records for grace
3957 	 * checking.
3958 	 * Each record is between sizeof (struct nfst_rec) and
3959 	 * sizeof (struct nfst_rec) + NFSV4_OPAQUELIMIT - 1
3960 	 * and is actually sizeof (struct nfst_rec) + nst_len - 1.
3961 	 */
3962 	tsp = (struct nfst_rec *)malloc(sizeof (struct nfst_rec) +
3963 		NFSV4_OPAQUELIMIT - 1, M_TEMP, M_WAITOK);
3964 	do {
3965 	    error = NFSD_RDWR(UIO_READ, NFSFPVNODE(sf->nsf_fp),
3966 	        (caddr_t)tsp, sizeof (struct nfst_rec) + NFSV4_OPAQUELIMIT - 1,
3967 	        off, UIO_SYSSPACE, 0, NFSFPCRED(sf->nsf_fp), &aresid, p);
3968 	    len = (sizeof (struct nfst_rec) + NFSV4_OPAQUELIMIT - 1) - aresid;
3969 	    if (error || (len > 0 && (len < sizeof (struct nfst_rec) ||
3970 		len < (sizeof (struct nfst_rec) + tsp->len - 1)))) {
3971 		/*
3972 		 * Yuck, the file has been corrupted, so just return
3973 		 * after clearing out any restart state, so the grace period
3974 		 * is over.
3975 		 */
3976 		LIST_FOREACH_SAFE(sp, &sf->nsf_head, nst_list, nsp) {
3977 			LIST_REMOVE(sp, nst_list);
3978 			free((caddr_t)sp, M_TEMP);
3979 		}
3980 		free((caddr_t)tsp, M_TEMP);
3981 		sf->nsf_flags &= ~NFSNSF_OK;
3982 		free((caddr_t)sf->nsf_bootvals, M_TEMP);
3983 		sf->nsf_bootvals = NULL;
3984 		return;
3985 	    }
3986 	    if (len > 0) {
3987 		off += sizeof (struct nfst_rec) + tsp->len - 1;
3988 		/*
3989 		 * Search the list for a matching client.
3990 		 */
3991 		LIST_FOREACH(sp, &sf->nsf_head, nst_list) {
3992 			if (tsp->len == sp->nst_len &&
3993 			    !NFSBCMP(tsp->client, sp->nst_client, tsp->len))
3994 				break;
3995 		}
3996 		if (sp == LIST_END(&sf->nsf_head)) {
3997 			sp = (struct nfsrv_stable *)malloc(tsp->len +
3998 				sizeof (struct nfsrv_stable) - 1, M_TEMP,
3999 				M_WAITOK);
4000 			NFSBCOPY((caddr_t)tsp, (caddr_t)&sp->nst_rec,
4001 				sizeof (struct nfst_rec) + tsp->len - 1);
4002 			LIST_INSERT_HEAD(&sf->nsf_head, sp, nst_list);
4003 		} else {
4004 			if (tsp->flag == NFSNST_REVOKE)
4005 				sp->nst_flag |= NFSNST_REVOKE;
4006 			else
4007 				/*
4008 				 * A subsequent timestamp indicates the client
4009 				 * did a setclientid/confirm and any previous
4010 				 * revoke is no longer relevant.
4011 				 */
4012 				sp->nst_flag &= ~NFSNST_REVOKE;
4013 		}
4014 	    }
4015 	} while (len > 0);
4016 	free((caddr_t)tsp, M_TEMP);
4017 	sf->nsf_flags = NFSNSF_OK;
4018 	sf->nsf_eograce = NFSD_MONOSEC + sf->nsf_lease +
4019 		NFSRV_LEASEDELTA;
4020 }
4021 
4022 /*
4023  * Update the stable storage file, now that the grace period is over.
4024  */
4025 APPLESTATIC void
4026 nfsrv_updatestable(NFSPROC_T *p)
4027 {
4028 	struct nfsrv_stablefirst *sf = &nfsrv_stablefirst;
4029 	struct nfsrv_stable *sp, *nsp;
4030 	int i;
4031 	struct nfsvattr nva;
4032 	vnode_t vp;
4033 #if defined(__FreeBSD_version) && (__FreeBSD_version >= 500000)
4034 	mount_t mp = NULL;
4035 #endif
4036 	int error;
4037 
4038 	if (sf->nsf_fp == NULL || (sf->nsf_flags & NFSNSF_UPDATEDONE))
4039 		return;
4040 	sf->nsf_flags |= NFSNSF_UPDATEDONE;
4041 	/*
4042 	 * Ok, we need to rewrite the stable storage file.
4043 	 * - truncate to 0 length
4044 	 * - write the new first structure
4045 	 * - loop through the data structures, writing out any that
4046 	 *   have timestamps older than the old boot
4047 	 */
4048 	if (sf->nsf_bootvals) {
4049 		sf->nsf_numboots++;
4050 		for (i = sf->nsf_numboots - 2; i >= 0; i--)
4051 			sf->nsf_bootvals[i + 1] = sf->nsf_bootvals[i];
4052 	} else {
4053 		sf->nsf_numboots = 1;
4054 		sf->nsf_bootvals = (time_t *)malloc(sizeof (time_t),
4055 			M_TEMP, M_WAITOK);
4056 	}
4057 	sf->nsf_bootvals[0] = nfsrvboottime;
4058 	sf->nsf_lease = nfsrv_lease;
4059 	NFSVNO_ATTRINIT(&nva);
4060 	NFSVNO_SETATTRVAL(&nva, size, 0);
4061 	vp = NFSFPVNODE(sf->nsf_fp);
4062 	NFS_STARTWRITE(vp, &mp);
4063 	NFSVOPLOCK(vp, LK_EXCLUSIVE | LK_RETRY, p);
4064 	error = nfsvno_setattr(vp, &nva, NFSFPCRED(sf->nsf_fp), p, NULL);
4065 	NFS_ENDWRITE(mp);
4066 	NFSVOPUNLOCK(vp, 0, p);
4067 	if (!error)
4068 	    error = NFSD_RDWR(UIO_WRITE, vp,
4069 		(caddr_t)&sf->nsf_rec, sizeof (struct nfsf_rec), (off_t)0,
4070 		UIO_SYSSPACE, IO_SYNC, NFSFPCRED(sf->nsf_fp), NULL, p);
4071 	if (!error)
4072 	    error = NFSD_RDWR(UIO_WRITE, vp,
4073 		(caddr_t)sf->nsf_bootvals,
4074 		sf->nsf_numboots * sizeof (time_t),
4075 		(off_t)(sizeof (struct nfsf_rec)),
4076 		UIO_SYSSPACE, IO_SYNC, NFSFPCRED(sf->nsf_fp), NULL, p);
4077 	free((caddr_t)sf->nsf_bootvals, M_TEMP);
4078 	sf->nsf_bootvals = NULL;
4079 	if (error) {
4080 		sf->nsf_flags &= ~NFSNSF_OK;
4081 		printf("EEK! Can't write NfsV4 stable storage file\n");
4082 		return;
4083 	}
4084 	sf->nsf_flags |= NFSNSF_OK;
4085 
4086 	/*
4087 	 * Loop through the list and write out timestamp records for
4088 	 * any clients that successfully reclaimed state.
4089 	 */
4090 	LIST_FOREACH_SAFE(sp, &sf->nsf_head, nst_list, nsp) {
4091 		if (sp->nst_flag & NFSNST_GOTSTATE) {
4092 			nfsrv_writestable(sp->nst_client, sp->nst_len,
4093 				NFSNST_NEWSTATE, p);
4094 			sp->nst_clp->lc_flags |= LCL_STAMPEDSTABLE;
4095 		}
4096 		LIST_REMOVE(sp, nst_list);
4097 		free((caddr_t)sp, M_TEMP);
4098 	}
4099 }
4100 
4101 /*
4102  * Append a record to the stable storage file.
4103  */
4104 APPLESTATIC void
4105 nfsrv_writestable(u_char *client, int len, int flag, NFSPROC_T *p)
4106 {
4107 	struct nfsrv_stablefirst *sf = &nfsrv_stablefirst;
4108 	struct nfst_rec *sp;
4109 	int error;
4110 
4111 	if (!(sf->nsf_flags & NFSNSF_OK) || sf->nsf_fp == NULL)
4112 		return;
4113 	sp = (struct nfst_rec *)malloc(sizeof (struct nfst_rec) +
4114 		len - 1, M_TEMP, M_WAITOK);
4115 	sp->len = len;
4116 	NFSBCOPY(client, sp->client, len);
4117 	sp->flag = flag;
4118 	error = NFSD_RDWR(UIO_WRITE, NFSFPVNODE(sf->nsf_fp),
4119 	    (caddr_t)sp, sizeof (struct nfst_rec) + len - 1, (off_t)0,
4120 	    UIO_SYSSPACE, (IO_SYNC | IO_APPEND), NFSFPCRED(sf->nsf_fp), NULL, p);
4121 	free((caddr_t)sp, M_TEMP);
4122 	if (error) {
4123 		sf->nsf_flags &= ~NFSNSF_OK;
4124 		printf("EEK! Can't write NfsV4 stable storage file\n");
4125 	}
4126 }
4127 
4128 /*
4129  * This function is called during the grace period to mark a client
4130  * that successfully reclaimed state.
4131  */
4132 static void
4133 nfsrv_markstable(struct nfsclient *clp)
4134 {
4135 	struct nfsrv_stable *sp;
4136 
4137 	/*
4138 	 * First find the client structure.
4139 	 */
4140 	LIST_FOREACH(sp, &nfsrv_stablefirst.nsf_head, nst_list) {
4141 		if (sp->nst_len == clp->lc_idlen &&
4142 		    !NFSBCMP(sp->nst_client, clp->lc_id, sp->nst_len))
4143 			break;
4144 	}
4145 	if (sp == LIST_END(&nfsrv_stablefirst.nsf_head))
4146 		return;
4147 
4148 	/*
4149 	 * Now, just mark it and set the nfsclient back pointer.
4150 	 */
4151 	sp->nst_flag |= NFSNST_GOTSTATE;
4152 	sp->nst_clp = clp;
4153 }
4154 
4155 /*
4156  * This function is called for a reclaim, to see if it gets grace.
4157  * It returns 0 if a reclaim is allowed, 1 otherwise.
4158  */
4159 static int
4160 nfsrv_checkstable(struct nfsclient *clp)
4161 {
4162 	struct nfsrv_stable *sp;
4163 
4164 	/*
4165 	 * First, find the entry for the client.
4166 	 */
4167 	LIST_FOREACH(sp, &nfsrv_stablefirst.nsf_head, nst_list) {
4168 		if (sp->nst_len == clp->lc_idlen &&
4169 		    !NFSBCMP(sp->nst_client, clp->lc_id, sp->nst_len))
4170 			break;
4171 	}
4172 
4173 	/*
4174 	 * If not in the list, state was revoked or no state was issued
4175 	 * since the previous reboot, a reclaim is denied.
4176 	 */
4177 	if (sp == LIST_END(&nfsrv_stablefirst.nsf_head) ||
4178 	    (sp->nst_flag & NFSNST_REVOKE) ||
4179 	    !(nfsrv_stablefirst.nsf_flags & NFSNSF_OK))
4180 		return (1);
4181 	return (0);
4182 }
4183 
4184 /*
4185  * Test for and try to clear out a conflicting client. This is called by
4186  * nfsrv_lockctrl() and nfsrv_openctrl() when conflicts with other clients
4187  * a found.
4188  * The trick here is that it can't revoke a conflicting client with an
4189  * expired lease unless it holds the v4root lock, so...
4190  * If no v4root lock, get the lock and return 1 to indicate "try again".
4191  * Return 0 to indicate the conflict can't be revoked and 1 to indicate
4192  * the revocation worked and the conflicting client is "bye, bye", so it
4193  * can be tried again.
4194  * Unlocks State before a non-zero value is returned.
4195  */
4196 static int
4197 nfsrv_clientconflict(struct nfsclient *clp, int *haslockp, __unused vnode_t vp,
4198     NFSPROC_T *p)
4199 {
4200 	int gotlock;
4201 
4202 	/*
4203 	 * If lease hasn't expired, we can't fix it.
4204 	 */
4205 	if (clp->lc_expiry >= NFSD_MONOSEC ||
4206 	    !(nfsrv_stablefirst.nsf_flags & NFSNSF_UPDATEDONE))
4207 		return (0);
4208 	if (*haslockp == 0) {
4209 		NFSUNLOCKSTATE();
4210 		NFSVOPUNLOCK(vp, 0, p);
4211 		NFSLOCKV4ROOTMUTEX();
4212 		nfsv4_relref(&nfsv4rootfs_lock);
4213 		do {
4214 			gotlock = nfsv4_lock(&nfsv4rootfs_lock, 1, NULL,
4215 			    NFSV4ROOTLOCKMUTEXPTR);
4216 		} while (!gotlock);
4217 		NFSUNLOCKV4ROOTMUTEX();
4218 		NFSLOCKSTATE();	/* to avoid a race with */
4219 		NFSUNLOCKSTATE();	/* nfsrv_servertimer() */
4220 		*haslockp = 1;
4221 		NFSVOPLOCK(vp, LK_EXCLUSIVE | LK_RETRY, p);
4222 		return (1);
4223 	}
4224 	NFSUNLOCKSTATE();
4225 
4226 	/*
4227 	 * Ok, we can expire the conflicting client.
4228 	 */
4229 	nfsrv_writestable(clp->lc_id, clp->lc_idlen, NFSNST_REVOKE, p);
4230 	nfsrv_cleanclient(clp, p);
4231 	nfsrv_freedeleglist(&clp->lc_deleg);
4232 	nfsrv_freedeleglist(&clp->lc_olddeleg);
4233 	LIST_REMOVE(clp, lc_hash);
4234 	nfsrv_zapclient(clp, p);
4235 	return (1);
4236 }
4237 
4238 
4239 /*
4240  * Resolve a delegation conflict.
4241  * Returns 0 to indicate the conflict was resolved without sleeping.
4242  * Return -1 to indicate that the caller should check for conflicts again.
4243  * Return > 0 for an error that should be returned, normally NFSERR_DELAY.
4244  *
4245  * Also, manipulate the nfsv4root_lock, as required. It isn't changed
4246  * for a return of 0, since there was no sleep and it could be required
4247  * later. It is released for a return of NFSERR_DELAY, since the caller
4248  * will return that error. It is released when a sleep was done waiting
4249  * for the delegation to be returned or expire (so that other nfsds can
4250  * handle ops). Then, it must be acquired for the write to stable storage.
4251  * (This function is somewhat similar to nfsrv_clientconflict(), but
4252  *  the semantics differ in a couple of subtle ways. The return of 0
4253  *  indicates the conflict was resolved without sleeping here, not
4254  *  that the conflict can't be resolved and the handling of nfsv4root_lock
4255  *  differs, as noted above.)
4256  * Unlocks State before returning a non-zero value.
4257  */
4258 static int
4259 nfsrv_delegconflict(struct nfsstate *stp, int *haslockp, NFSPROC_T *p,
4260     __unused vnode_t vp)
4261 {
4262 	struct nfsclient *clp = stp->ls_clp;
4263 	int gotlock, error, retrycnt, zapped_clp;
4264 	nfsv4stateid_t tstateid;
4265 	fhandle_t tfh;
4266 
4267 	/*
4268 	 * If the conflict is with an old delegation...
4269 	 */
4270 	if (stp->ls_flags & NFSLCK_OLDDELEG) {
4271 		/*
4272 		 * You can delete it, if it has expired.
4273 		 */
4274 		if (clp->lc_delegtime < NFSD_MONOSEC) {
4275 			nfsrv_freedeleg(stp);
4276 			NFSUNLOCKSTATE();
4277 			return (-1);
4278 		}
4279 		NFSUNLOCKSTATE();
4280 		/*
4281 		 * During this delay, the old delegation could expire or it
4282 		 * could be recovered by the client via an Open with
4283 		 * CLAIM_DELEGATE_PREV.
4284 		 * Release the nfsv4root_lock, if held.
4285 		 */
4286 		if (*haslockp) {
4287 			*haslockp = 0;
4288 			NFSLOCKV4ROOTMUTEX();
4289 			nfsv4_unlock(&nfsv4rootfs_lock, 1);
4290 			NFSUNLOCKV4ROOTMUTEX();
4291 		}
4292 		return (NFSERR_DELAY);
4293 	}
4294 
4295 	/*
4296 	 * It's a current delegation, so:
4297 	 * - check to see if the delegation has expired
4298 	 *   - if so, get the v4root lock and then expire it
4299 	 */
4300 	if (!(stp->ls_flags & NFSLCK_DELEGRECALL)) {
4301 		/*
4302 		 * - do a recall callback, since not yet done
4303 		 * For now, never allow truncate to be set. To use
4304 		 * truncate safely, it must be guaranteed that the
4305 		 * Remove, Rename or Setattr with size of 0 will
4306 		 * succeed and that would require major changes to
4307 		 * the VFS/Vnode OPs.
4308 		 * Set the expiry time large enough so that it won't expire
4309 		 * until after the callback, then set it correctly, once
4310 		 * the callback is done. (The delegation will now time
4311 		 * out whether or not the Recall worked ok. The timeout
4312 		 * will be extended when ops are done on the delegation
4313 		 * stateid, up to the timelimit.)
4314 		 */
4315 		stp->ls_delegtime = NFSD_MONOSEC + (2 * nfsrv_lease) +
4316 		    NFSRV_LEASEDELTA;
4317 		stp->ls_delegtimelimit = NFSD_MONOSEC + (6 * nfsrv_lease) +
4318 		    NFSRV_LEASEDELTA;
4319 		stp->ls_flags |= NFSLCK_DELEGRECALL;
4320 
4321 		/*
4322 		 * Loop NFSRV_CBRETRYCNT times while the CBRecall replies
4323 		 * NFSERR_BADSTATEID or NFSERR_BADHANDLE. This is done
4324 		 * in order to try and avoid a race that could happen
4325 		 * when a CBRecall request passed the Open reply with
4326 		 * the delegation in it when transitting the network.
4327 		 * Since nfsrv_docallback will sleep, don't use stp after
4328 		 * the call.
4329 		 */
4330 		NFSBCOPY((caddr_t)&stp->ls_stateid, (caddr_t)&tstateid,
4331 		    sizeof (tstateid));
4332 		NFSBCOPY((caddr_t)&stp->ls_lfp->lf_fh, (caddr_t)&tfh,
4333 		    sizeof (tfh));
4334 		NFSUNLOCKSTATE();
4335 		if (*haslockp) {
4336 			*haslockp = 0;
4337 			NFSLOCKV4ROOTMUTEX();
4338 			nfsv4_unlock(&nfsv4rootfs_lock, 1);
4339 			NFSUNLOCKV4ROOTMUTEX();
4340 		}
4341 		retrycnt = 0;
4342 		do {
4343 		    error = nfsrv_docallback(clp, NFSV4OP_CBRECALL,
4344 			&tstateid, 0, &tfh, NULL, NULL, p);
4345 		    retrycnt++;
4346 		} while ((error == NFSERR_BADSTATEID ||
4347 		    error == NFSERR_BADHANDLE) && retrycnt < NFSV4_CBRETRYCNT);
4348 		return (NFSERR_DELAY);
4349 	}
4350 
4351 	if (clp->lc_expiry >= NFSD_MONOSEC &&
4352 	    stp->ls_delegtime >= NFSD_MONOSEC) {
4353 		NFSUNLOCKSTATE();
4354 		/*
4355 		 * A recall has been done, but it has not yet expired.
4356 		 * So, RETURN_DELAY.
4357 		 */
4358 		if (*haslockp) {
4359 			*haslockp = 0;
4360 			NFSLOCKV4ROOTMUTEX();
4361 			nfsv4_unlock(&nfsv4rootfs_lock, 1);
4362 			NFSUNLOCKV4ROOTMUTEX();
4363 		}
4364 		return (NFSERR_DELAY);
4365 	}
4366 
4367 	/*
4368 	 * If we don't yet have the lock, just get it and then return,
4369 	 * since we need that before deleting expired state, such as
4370 	 * this delegation.
4371 	 * When getting the lock, unlock the vnode, so other nfsds that
4372 	 * are in progress, won't get stuck waiting for the vnode lock.
4373 	 */
4374 	if (*haslockp == 0) {
4375 		NFSUNLOCKSTATE();
4376 		NFSVOPUNLOCK(vp, 0, p);
4377 		NFSLOCKV4ROOTMUTEX();
4378 		nfsv4_relref(&nfsv4rootfs_lock);
4379 		do {
4380 			gotlock = nfsv4_lock(&nfsv4rootfs_lock, 1, NULL,
4381 			    NFSV4ROOTLOCKMUTEXPTR);
4382 		} while (!gotlock);
4383 		NFSUNLOCKV4ROOTMUTEX();
4384 		NFSLOCKSTATE();	/* to avoid a race with */
4385 		NFSUNLOCKSTATE();	/* nfsrv_servertimer() */
4386 		*haslockp = 1;
4387 		NFSVOPLOCK(vp, LK_EXCLUSIVE | LK_RETRY, p);
4388 		return (-1);
4389 	}
4390 
4391 	NFSUNLOCKSTATE();
4392 	/*
4393 	 * Ok, we can delete the expired delegation.
4394 	 * First, write the Revoke record to stable storage and then
4395 	 * clear out the conflict.
4396 	 * Since all other nfsd threads are now blocked, we can safely
4397 	 * sleep without the state changing.
4398 	 */
4399 	nfsrv_writestable(clp->lc_id, clp->lc_idlen, NFSNST_REVOKE, p);
4400 	if (clp->lc_expiry < NFSD_MONOSEC) {
4401 		nfsrv_cleanclient(clp, p);
4402 		nfsrv_freedeleglist(&clp->lc_deleg);
4403 		nfsrv_freedeleglist(&clp->lc_olddeleg);
4404 		LIST_REMOVE(clp, lc_hash);
4405 		zapped_clp = 1;
4406 	} else {
4407 		nfsrv_freedeleg(stp);
4408 		zapped_clp = 0;
4409 	}
4410 	if (zapped_clp)
4411 		nfsrv_zapclient(clp, p);
4412 	return (-1);
4413 }
4414 
4415 /*
4416  * Check for a remove allowed, if remove is set to 1 and get rid of
4417  * delegations.
4418  */
4419 APPLESTATIC int
4420 nfsrv_checkremove(vnode_t vp, int remove, NFSPROC_T *p)
4421 {
4422 	struct nfsstate *stp;
4423 	struct nfslockfile *lfp;
4424 	int error, haslock = 0;
4425 	fhandle_t nfh;
4426 
4427 	/*
4428 	 * First, get the lock file structure.
4429 	 * (A return of -1 means no associated state, so remove ok.)
4430 	 */
4431 	error = nfsrv_getlockfh(vp, NFSLCK_CHECK, NULL, &nfh, p);
4432 tryagain:
4433 	NFSLOCKSTATE();
4434 	if (!error)
4435 		error = nfsrv_getlockfile(NFSLCK_CHECK, NULL, &lfp, &nfh, 0);
4436 	if (error) {
4437 		NFSUNLOCKSTATE();
4438 		if (haslock) {
4439 			NFSLOCKV4ROOTMUTEX();
4440 			nfsv4_unlock(&nfsv4rootfs_lock, 1);
4441 			NFSUNLOCKV4ROOTMUTEX();
4442 		}
4443 		if (error == -1)
4444 			return (0);
4445 		return (error);
4446 	}
4447 
4448 	/*
4449 	 * Now, we must Recall any delegations.
4450 	 */
4451 	error = nfsrv_cleandeleg(vp, lfp, NULL, &haslock, p);
4452 	if (error) {
4453 		/*
4454 		 * nfsrv_cleandeleg() unlocks state for non-zero
4455 		 * return.
4456 		 */
4457 		if (error == -1)
4458 			goto tryagain;
4459 		if (haslock) {
4460 			NFSLOCKV4ROOTMUTEX();
4461 			nfsv4_unlock(&nfsv4rootfs_lock, 1);
4462 			NFSUNLOCKV4ROOTMUTEX();
4463 		}
4464 		return (error);
4465 	}
4466 
4467 	/*
4468 	 * Now, look for a conflicting open share.
4469 	 */
4470 	if (remove) {
4471 		LIST_FOREACH(stp, &lfp->lf_open, ls_file) {
4472 			if (stp->ls_flags & NFSLCK_WRITEDENY) {
4473 				error = NFSERR_FILEOPEN;
4474 				break;
4475 			}
4476 		}
4477 	}
4478 
4479 	NFSUNLOCKSTATE();
4480 	if (haslock) {
4481 		NFSLOCKV4ROOTMUTEX();
4482 		nfsv4_unlock(&nfsv4rootfs_lock, 1);
4483 		NFSUNLOCKV4ROOTMUTEX();
4484 	}
4485 	return (error);
4486 }
4487 
4488 /*
4489  * Clear out all delegations for the file referred to by lfp.
4490  * May return NFSERR_DELAY, if there will be a delay waiting for
4491  * delegations to expire.
4492  * Returns -1 to indicate it slept while recalling a delegation.
4493  * This function has the side effect of deleting the nfslockfile structure,
4494  * if it no longer has associated state and didn't have to sleep.
4495  * Unlocks State before a non-zero value is returned.
4496  */
4497 static int
4498 nfsrv_cleandeleg(vnode_t vp, struct nfslockfile *lfp,
4499     struct nfsclient *clp, int *haslockp, NFSPROC_T *p)
4500 {
4501 	struct nfsstate *stp, *nstp;
4502 	int ret;
4503 
4504 	stp = LIST_FIRST(&lfp->lf_deleg);
4505 	while (stp != LIST_END(&lfp->lf_deleg)) {
4506 		nstp = LIST_NEXT(stp, ls_file);
4507 		if (stp->ls_clp != clp) {
4508 			ret = nfsrv_delegconflict(stp, haslockp, p, vp);
4509 			if (ret) {
4510 				/*
4511 				 * nfsrv_delegconflict() unlocks state
4512 				 * when it returns non-zero.
4513 				 */
4514 				return (ret);
4515 			}
4516 		}
4517 		stp = nstp;
4518 	}
4519 	return (0);
4520 }
4521 
4522 /*
4523  * There are certain operations that, when being done outside of NFSv4,
4524  * require that any NFSv4 delegation for the file be recalled.
4525  * This function is to be called for those cases:
4526  * VOP_RENAME() - When a delegation is being recalled for any reason,
4527  *	the client may have to do Opens against the server, using the file's
4528  *	final component name. If the file has been renamed on the server,
4529  *	that component name will be incorrect and the Open will fail.
4530  * VOP_REMOVE() - Theoretically, a client could Open a file after it has
4531  *	been removed on the server, if there is a delegation issued to
4532  *	that client for the file. I say "theoretically" since clients
4533  *	normally do an Access Op before the Open and that Access Op will
4534  *	fail with ESTALE. Note that NFSv2 and 3 don't even do Opens, so
4535  *	they will detect the file's removal in the same manner. (There is
4536  *	one case where RFC3530 allows a client to do an Open without first
4537  *	doing an Access Op, which is passage of a check against the ACE
4538  *	returned with a Write delegation, but current practice is to ignore
4539  *	the ACE and always do an Access Op.)
4540  *	Since the functions can only be called with an unlocked vnode, this
4541  *	can't be done at this time.
4542  * VOP_ADVLOCK() - When a client holds a delegation, it can issue byte range
4543  *	locks locally in the client, which are not visible to the server. To
4544  *	deal with this, issuing of delegations for a vnode must be disabled
4545  *	and all delegations for the vnode recalled. This is done via the
4546  *	second function, using the VV_DISABLEDELEG vflag on the vnode.
4547  */
4548 APPLESTATIC void
4549 nfsd_recalldelegation(vnode_t vp, NFSPROC_T *p)
4550 {
4551 	struct timespec mytime;
4552 	int32_t starttime;
4553 	int error;
4554 
4555 	KASSERT(!VOP_ISLOCKED(vp), ("vp %p is locked", vp));
4556 
4557 	/*
4558 	 * First, check to see if the server is currently running and it has
4559 	 * been called for a regular file when issuing delegations.
4560 	 */
4561 	if (newnfs_numnfsd == 0 || vp->v_type != VREG ||
4562 	    nfsrv_issuedelegs == 0)
4563 		return;
4564 
4565 	/*
4566 	 * Now, call nfsrv_checkremove() in a loop while it returns
4567 	 * NFSERR_DELAY. Return upon any other error or when timed out.
4568 	 */
4569 	NFSGETNANOTIME(&mytime);
4570 	starttime = (u_int32_t)mytime.tv_sec;
4571 	do {
4572 		error = nfsrv_checkremove(vp, 0, p);
4573 		if (error == NFSERR_DELAY) {
4574 			NFSGETNANOTIME(&mytime);
4575 			if (((u_int32_t)mytime.tv_sec - starttime) >
4576 			    NFS_REMOVETIMEO &&
4577 			    ((u_int32_t)mytime.tv_sec - starttime) <
4578 			    100000)
4579 				return;
4580 			/* Sleep for a short period of time */
4581 			(void) nfs_catnap(PZERO, 0, "nfsremove");
4582 		}
4583 	} while (error == NFSERR_DELAY);
4584 }
4585 
4586 APPLESTATIC void
4587 nfsd_disabledelegation(vnode_t vp, NFSPROC_T *p)
4588 {
4589 
4590 #ifdef VV_DISABLEDELEG
4591 	/*
4592 	 * First, flag issuance of delegations disabled.
4593 	 */
4594 	atomic_set_long(&vp->v_vflag, VV_DISABLEDELEG);
4595 #endif
4596 
4597 	/*
4598 	 * Then call nfsd_recalldelegation() to get rid of all extant
4599 	 * delegations.
4600 	 */
4601 	nfsd_recalldelegation(vp, p);
4602 }
4603 
4604 /*
4605  * Check for conflicting locks, etc. and then get rid of delegations.
4606  * (At one point I thought that I should get rid of delegations for any
4607  *  Setattr, since it could potentially disallow the I/O op (read or write)
4608  *  allowed by the delegation. However, Setattr Ops that aren't changing
4609  *  the size get a stateid of all 0s, so you can't tell if it is a delegation
4610  *  for the same client or a different one, so I decided to only get rid
4611  *  of delegations for other clients when the size is being changed.)
4612  * In general, a Setattr can disable NFS I/O Ops that are outstanding, such
4613  * as Write backs, even if there is no delegation, so it really isn't any
4614  * different?)
4615  */
4616 APPLESTATIC int
4617 nfsrv_checksetattr(vnode_t vp, struct nfsrv_descript *nd,
4618     nfsv4stateid_t *stateidp, struct nfsvattr *nvap, nfsattrbit_t *attrbitp,
4619     struct nfsexstuff *exp, NFSPROC_T *p)
4620 {
4621 	struct nfsstate st, *stp = &st;
4622 	struct nfslock lo, *lop = &lo;
4623 	int error = 0;
4624 	nfsquad_t clientid;
4625 
4626 	if (NFSISSET_ATTRBIT(attrbitp, NFSATTRBIT_SIZE)) {
4627 		stp->ls_flags = (NFSLCK_CHECK | NFSLCK_WRITEACCESS);
4628 		lop->lo_first = nvap->na_size;
4629 	} else {
4630 		stp->ls_flags = 0;
4631 		lop->lo_first = 0;
4632 	}
4633 	if (NFSISSET_ATTRBIT(attrbitp, NFSATTRBIT_OWNER) ||
4634 	    NFSISSET_ATTRBIT(attrbitp, NFSATTRBIT_OWNERGROUP) ||
4635 	    NFSISSET_ATTRBIT(attrbitp, NFSATTRBIT_MODE) ||
4636 	    NFSISSET_ATTRBIT(attrbitp, NFSATTRBIT_ACL))
4637 		stp->ls_flags |= NFSLCK_SETATTR;
4638 	if (stp->ls_flags == 0)
4639 		return (0);
4640 	lop->lo_end = NFS64BITSSET;
4641 	lop->lo_flags = NFSLCK_WRITE;
4642 	stp->ls_ownerlen = 0;
4643 	stp->ls_op = NULL;
4644 	stp->ls_uid = nd->nd_cred->cr_uid;
4645 	stp->ls_stateid.seqid = stateidp->seqid;
4646 	clientid.lval[0] = stp->ls_stateid.other[0] = stateidp->other[0];
4647 	clientid.lval[1] = stp->ls_stateid.other[1] = stateidp->other[1];
4648 	stp->ls_stateid.other[2] = stateidp->other[2];
4649 	error = nfsrv_lockctrl(vp, &stp, &lop, NULL, clientid,
4650 	    stateidp, exp, nd, p);
4651 	return (error);
4652 }
4653 
4654 /*
4655  * Check for a write delegation and do a CBGETATTR if there is one, updating
4656  * the attributes, as required.
4657  * Should I return an error if I can't get the attributes? (For now, I'll
4658  * just return ok.
4659  */
4660 APPLESTATIC int
4661 nfsrv_checkgetattr(struct nfsrv_descript *nd, vnode_t vp,
4662     struct nfsvattr *nvap, nfsattrbit_t *attrbitp, struct ucred *cred,
4663     NFSPROC_T *p)
4664 {
4665 	struct nfsstate *stp;
4666 	struct nfslockfile *lfp;
4667 	struct nfsclient *clp;
4668 	struct nfsvattr nva;
4669 	fhandle_t nfh;
4670 	int error;
4671 	nfsattrbit_t cbbits;
4672 	u_quad_t delegfilerev;
4673 
4674 	NFSCBGETATTR_ATTRBIT(attrbitp, &cbbits);
4675 	if (!NFSNONZERO_ATTRBIT(&cbbits))
4676 		return (0);
4677 
4678 	/*
4679 	 * Get the lock file structure.
4680 	 * (A return of -1 means no associated state, so return ok.)
4681 	 */
4682 	error = nfsrv_getlockfh(vp, NFSLCK_CHECK, NULL, &nfh, p);
4683 	NFSLOCKSTATE();
4684 	if (!error)
4685 		error = nfsrv_getlockfile(NFSLCK_CHECK, NULL, &lfp, &nfh, 0);
4686 	if (error) {
4687 		NFSUNLOCKSTATE();
4688 		if (error == -1)
4689 			return (0);
4690 		return (error);
4691 	}
4692 
4693 	/*
4694 	 * Now, look for a write delegation.
4695 	 */
4696 	LIST_FOREACH(stp, &lfp->lf_deleg, ls_file) {
4697 		if (stp->ls_flags & NFSLCK_DELEGWRITE)
4698 			break;
4699 	}
4700 	if (stp == LIST_END(&lfp->lf_deleg)) {
4701 		NFSUNLOCKSTATE();
4702 		return (0);
4703 	}
4704 	clp = stp->ls_clp;
4705 	delegfilerev = stp->ls_filerev;
4706 
4707 	/*
4708 	 * If the Write delegation was issued as a part of this Compound RPC
4709 	 * or if we have an Implied Clientid (used in a previous Op in this
4710 	 * compound) and it is the client the delegation was issued to,
4711 	 * just return ok.
4712 	 * I also assume that it is from the same client iff the network
4713 	 * host IP address is the same as the callback address. (Not
4714 	 * exactly correct by the RFC, but avoids a lot of Getattr
4715 	 * callbacks.)
4716 	 */
4717 	if (nd->nd_compref == stp->ls_compref ||
4718 	    ((nd->nd_flag & ND_IMPLIEDCLID) &&
4719 	     clp->lc_clientid.qval == nd->nd_clientid.qval) ||
4720 	     nfsaddr2_match(clp->lc_req.nr_nam, nd->nd_nam)) {
4721 		NFSUNLOCKSTATE();
4722 		return (0);
4723 	}
4724 
4725 	/*
4726 	 * We are now done with the delegation state structure,
4727 	 * so the statelock can be released and we can now tsleep().
4728 	 */
4729 
4730 	/*
4731 	 * Now, we must do the CB Getattr callback, to see if Change or Size
4732 	 * has changed.
4733 	 */
4734 	if (clp->lc_expiry >= NFSD_MONOSEC) {
4735 		NFSUNLOCKSTATE();
4736 		NFSVNO_ATTRINIT(&nva);
4737 		nva.na_filerev = NFS64BITSSET;
4738 		error = nfsrv_docallback(clp, NFSV4OP_CBGETATTR, NULL,
4739 		    0, &nfh, &nva, &cbbits, p);
4740 		if (!error) {
4741 			if ((nva.na_filerev != NFS64BITSSET &&
4742 			    nva.na_filerev > delegfilerev) ||
4743 			    (NFSVNO_ISSETSIZE(&nva) &&
4744 			     nva.na_size != nvap->na_size)) {
4745 				nfsvno_updfilerev(vp, nvap, cred, p);
4746 				if (NFSVNO_ISSETSIZE(&nva))
4747 					nvap->na_size = nva.na_size;
4748 			}
4749 		}
4750 	} else {
4751 		NFSUNLOCKSTATE();
4752 	}
4753 	return (0);
4754 }
4755 
4756 /*
4757  * This function looks for openowners that haven't had any opens for
4758  * a while and throws them away. Called by an nfsd when NFSNSF_NOOPENS
4759  * is set.
4760  */
4761 APPLESTATIC void
4762 nfsrv_throwawayopens(NFSPROC_T *p)
4763 {
4764 	struct nfsclient *clp, *nclp;
4765 	struct nfsstate *stp, *nstp;
4766 	int i;
4767 
4768 	NFSLOCKSTATE();
4769 	nfsrv_stablefirst.nsf_flags &= ~NFSNSF_NOOPENS;
4770 	/*
4771 	 * For each client...
4772 	 */
4773 	for (i = 0; i < NFSCLIENTHASHSIZE; i++) {
4774 	    LIST_FOREACH_SAFE(clp, &nfsclienthash[i], lc_hash, nclp) {
4775 		LIST_FOREACH_SAFE(stp, &clp->lc_open, ls_list, nstp) {
4776 			if (LIST_EMPTY(&stp->ls_open) &&
4777 			    (stp->ls_noopens > NFSNOOPEN ||
4778 			     (nfsrv_openpluslock * 2) >
4779 			     NFSRV_V4STATELIMIT))
4780 				nfsrv_freeopenowner(stp, 0, p);
4781 		}
4782 	    }
4783 	}
4784 	NFSUNLOCKSTATE();
4785 }
4786 
4787 /*
4788  * This function checks to see if the credentials are the same.
4789  * Returns 1 for not same, 0 otherwise.
4790  */
4791 static int
4792 nfsrv_notsamecredname(struct nfsrv_descript *nd, struct nfsclient *clp)
4793 {
4794 
4795 	if (nd->nd_flag & ND_GSS) {
4796 		if (!(clp->lc_flags & LCL_GSS))
4797 			return (1);
4798 		if (clp->lc_flags & LCL_NAME) {
4799 			if (nd->nd_princlen != clp->lc_namelen ||
4800 			    NFSBCMP(nd->nd_principal, clp->lc_name,
4801 				clp->lc_namelen))
4802 				return (1);
4803 			else
4804 				return (0);
4805 		}
4806 		if (nd->nd_cred->cr_uid == clp->lc_uid)
4807 			return (0);
4808 		else
4809 			return (1);
4810 	} else if (clp->lc_flags & LCL_GSS)
4811 		return (1);
4812 	/*
4813 	 * For AUTH_SYS, allow the same uid or root. (This is underspecified
4814 	 * in RFC3530, which talks about principals, but doesn't say anything
4815 	 * about uids for AUTH_SYS.)
4816 	 */
4817 	if (nd->nd_cred->cr_uid == clp->lc_uid || nd->nd_cred->cr_uid == 0)
4818 		return (0);
4819 	else
4820 		return (1);
4821 }
4822 
4823 /*
4824  * Calculate the lease expiry time.
4825  */
4826 static time_t
4827 nfsrv_leaseexpiry(void)
4828 {
4829 	struct timeval curtime;
4830 
4831 	NFSGETTIME(&curtime);
4832 	if (nfsrv_stablefirst.nsf_eograce > NFSD_MONOSEC)
4833 		return (NFSD_MONOSEC + 2 * (nfsrv_lease + NFSRV_LEASEDELTA));
4834 	return (NFSD_MONOSEC + nfsrv_lease + NFSRV_LEASEDELTA);
4835 }
4836 
4837 /*
4838  * Delay the delegation timeout as far as ls_delegtimelimit, as required.
4839  */
4840 static void
4841 nfsrv_delaydelegtimeout(struct nfsstate *stp)
4842 {
4843 
4844 	if ((stp->ls_flags & NFSLCK_DELEGRECALL) == 0)
4845 		return;
4846 
4847 	if ((stp->ls_delegtime + 15) > NFSD_MONOSEC &&
4848 	    stp->ls_delegtime < stp->ls_delegtimelimit) {
4849 		stp->ls_delegtime += nfsrv_lease;
4850 		if (stp->ls_delegtime > stp->ls_delegtimelimit)
4851 			stp->ls_delegtime = stp->ls_delegtimelimit;
4852 	}
4853 }
4854 
4855 /*
4856  * This function checks to see if there is any other state associated
4857  * with the openowner for this Open.
4858  * It returns 1 if there is no other state, 0 otherwise.
4859  */
4860 static int
4861 nfsrv_nootherstate(struct nfsstate *stp)
4862 {
4863 	struct nfsstate *tstp;
4864 
4865 	LIST_FOREACH(tstp, &stp->ls_openowner->ls_open, ls_list) {
4866 		if (tstp != stp || !LIST_EMPTY(&tstp->ls_lock))
4867 			return (0);
4868 	}
4869 	return (1);
4870 }
4871 
4872 /*
4873  * Create a list of lock deltas (changes to local byte range locking
4874  * that can be rolled back using the list) and apply the changes via
4875  * nfsvno_advlock(). Optionally, lock the list. It is expected that either
4876  * the rollback or update function will be called after this.
4877  * It returns an error (and rolls back, as required), if any nfsvno_advlock()
4878  * call fails. If it returns an error, it will unlock the list.
4879  */
4880 static int
4881 nfsrv_locallock(vnode_t vp, struct nfslockfile *lfp, int flags,
4882     uint64_t first, uint64_t end, struct nfslockconflict *cfp, NFSPROC_T *p)
4883 {
4884 	struct nfslock *lop, *nlop;
4885 	int error = 0;
4886 
4887 	/* Loop through the list of locks. */
4888 	lop = LIST_FIRST(&lfp->lf_locallock);
4889 	while (first < end && lop != NULL) {
4890 		nlop = LIST_NEXT(lop, lo_lckowner);
4891 		if (first >= lop->lo_end) {
4892 			/* not there yet */
4893 			lop = nlop;
4894 		} else if (first < lop->lo_first) {
4895 			/* new one starts before entry in list */
4896 			if (end <= lop->lo_first) {
4897 				/* no overlap between old and new */
4898 				error = nfsrv_dolocal(vp, lfp, flags,
4899 				    NFSLCK_UNLOCK, first, end, cfp, p);
4900 				if (error != 0)
4901 					break;
4902 				first = end;
4903 			} else {
4904 				/* handle fragment overlapped with new one */
4905 				error = nfsrv_dolocal(vp, lfp, flags,
4906 				    NFSLCK_UNLOCK, first, lop->lo_first, cfp,
4907 				    p);
4908 				if (error != 0)
4909 					break;
4910 				first = lop->lo_first;
4911 			}
4912 		} else {
4913 			/* new one overlaps this entry in list */
4914 			if (end <= lop->lo_end) {
4915 				/* overlaps all of new one */
4916 				error = nfsrv_dolocal(vp, lfp, flags,
4917 				    lop->lo_flags, first, end, cfp, p);
4918 				if (error != 0)
4919 					break;
4920 				first = end;
4921 			} else {
4922 				/* handle fragment overlapped with new one */
4923 				error = nfsrv_dolocal(vp, lfp, flags,
4924 				    lop->lo_flags, first, lop->lo_end, cfp, p);
4925 				if (error != 0)
4926 					break;
4927 				first = lop->lo_end;
4928 				lop = nlop;
4929 			}
4930 		}
4931 	}
4932 	if (first < end && error == 0)
4933 		/* handle fragment past end of list */
4934 		error = nfsrv_dolocal(vp, lfp, flags, NFSLCK_UNLOCK, first,
4935 		    end, cfp, p);
4936 	return (error);
4937 }
4938 
4939 /*
4940  * Local lock unlock. Unlock all byte ranges that are no longer locked
4941  * by NFSv4.
4942  */
4943 static void
4944 nfsrv_localunlock(vnode_t vp, struct nfslockfile *lfp, uint64_t init_first,
4945     uint64_t init_end, NFSPROC_T *p)
4946 {
4947 	struct nfslock *lop;
4948 
4949 	uint64_t first, end;
4950 
4951 	first = init_first;
4952 	end = init_end;
4953 	while (first < init_end) {
4954 		/* Loop through all nfs locks, adjusting first and end */
4955 		LIST_FOREACH(lop, &lfp->lf_lock, lo_lckfile) {
4956 			if (first >= lop->lo_first &&
4957 			    first < lop->lo_end)
4958 				/* Overlaps initial part */
4959 				first = lop->lo_end;
4960 			else if (end > lop->lo_first &&
4961 			    lop->lo_first >= first)
4962 				/* Begins before end and past first */
4963 				end = lop->lo_first;
4964 			if (first >= end)
4965 				/* shrunk to 0 so this iteration is done */
4966 				break;
4967 		}
4968 		if (first < end) {
4969 			/* Unlock this segment */
4970 			(void) nfsrv_dolocal(vp, lfp, NFSLCK_UNLOCK,
4971 			    NFSLCK_READ, first, end, NULL, p);
4972 			nfsrv_locallock_commit(lfp, NFSLCK_UNLOCK,
4973 			    first, end);
4974 		}
4975 		/* and move on to the rest of the range */
4976 		first = end;
4977 		end = init_end;
4978 	}
4979 }
4980 
4981 /*
4982  * Do the local lock operation and update the rollback list, as required.
4983  * Perform the rollback and return the error if nfsvno_advlock() fails.
4984  */
4985 static int
4986 nfsrv_dolocal(vnode_t vp, struct nfslockfile *lfp, int flags, int oldflags,
4987     uint64_t first, uint64_t end, struct nfslockconflict *cfp, NFSPROC_T *p)
4988 {
4989 	struct nfsrollback *rlp;
4990 	int error, ltype, oldltype;
4991 
4992 	if (flags & NFSLCK_WRITE)
4993 		ltype = F_WRLCK;
4994 	else if (flags & NFSLCK_READ)
4995 		ltype = F_RDLCK;
4996 	else
4997 		ltype = F_UNLCK;
4998 	if (oldflags & NFSLCK_WRITE)
4999 		oldltype = F_WRLCK;
5000 	else if (oldflags & NFSLCK_READ)
5001 		oldltype = F_RDLCK;
5002 	else
5003 		oldltype = F_UNLCK;
5004 	if (ltype == oldltype || (oldltype == F_WRLCK && ltype == F_RDLCK))
5005 		/* nothing to do */
5006 		return (0);
5007 	error = nfsvno_advlock(vp, ltype, first, end, p);
5008 	if (error != 0) {
5009 		if (cfp != NULL) {
5010 			cfp->cl_clientid.lval[0] = 0;
5011 			cfp->cl_clientid.lval[1] = 0;
5012 			cfp->cl_first = 0;
5013 			cfp->cl_end = NFS64BITSSET;
5014 			cfp->cl_flags = NFSLCK_WRITE;
5015 			cfp->cl_ownerlen = 5;
5016 			NFSBCOPY("LOCAL", cfp->cl_owner, 5);
5017 		}
5018 		nfsrv_locallock_rollback(vp, lfp, p);
5019 	} else if (ltype != F_UNLCK) {
5020 		rlp = malloc(sizeof (struct nfsrollback), M_NFSDROLLBACK,
5021 		    M_WAITOK);
5022 		rlp->rlck_first = first;
5023 		rlp->rlck_end = end;
5024 		rlp->rlck_type = oldltype;
5025 		LIST_INSERT_HEAD(&lfp->lf_rollback, rlp, rlck_list);
5026 	}
5027 	return (error);
5028 }
5029 
5030 /*
5031  * Roll back local lock changes and free up the rollback list.
5032  */
5033 static void
5034 nfsrv_locallock_rollback(vnode_t vp, struct nfslockfile *lfp, NFSPROC_T *p)
5035 {
5036 	struct nfsrollback *rlp, *nrlp;
5037 
5038 	LIST_FOREACH_SAFE(rlp, &lfp->lf_rollback, rlck_list, nrlp) {
5039 		(void) nfsvno_advlock(vp, rlp->rlck_type, rlp->rlck_first,
5040 		    rlp->rlck_end, p);
5041 		free(rlp, M_NFSDROLLBACK);
5042 	}
5043 	LIST_INIT(&lfp->lf_rollback);
5044 }
5045 
5046 /*
5047  * Update local lock list and delete rollback list (ie now committed to the
5048  * local locks). Most of the work is done by the internal function.
5049  */
5050 static void
5051 nfsrv_locallock_commit(struct nfslockfile *lfp, int flags, uint64_t first,
5052     uint64_t end)
5053 {
5054 	struct nfsrollback *rlp, *nrlp;
5055 	struct nfslock *new_lop, *other_lop;
5056 
5057 	new_lop = malloc(sizeof (struct nfslock), M_NFSDLOCK, M_WAITOK);
5058 	if (flags & (NFSLCK_READ | NFSLCK_WRITE))
5059 		other_lop = malloc(sizeof (struct nfslock), M_NFSDLOCK,
5060 		    M_WAITOK);
5061 	else
5062 		other_lop = NULL;
5063 	new_lop->lo_flags = flags;
5064 	new_lop->lo_first = first;
5065 	new_lop->lo_end = end;
5066 	nfsrv_updatelock(NULL, &new_lop, &other_lop, lfp);
5067 	if (new_lop != NULL)
5068 		free(new_lop, M_NFSDLOCK);
5069 	if (other_lop != NULL)
5070 		free(other_lop, M_NFSDLOCK);
5071 
5072 	/* and get rid of the rollback list */
5073 	LIST_FOREACH_SAFE(rlp, &lfp->lf_rollback, rlck_list, nrlp)
5074 		free(rlp, M_NFSDROLLBACK);
5075 	LIST_INIT(&lfp->lf_rollback);
5076 }
5077 
5078 /*
5079  * Lock the struct nfslockfile for local lock updating.
5080  */
5081 static void
5082 nfsrv_locklf(struct nfslockfile *lfp)
5083 {
5084 	int gotlock;
5085 
5086 	/* lf_usecount ensures *lfp won't be free'd */
5087 	lfp->lf_usecount++;
5088 	do {
5089 		gotlock = nfsv4_lock(&lfp->lf_locallock_lck, 1, NULL,
5090 		    NFSSTATEMUTEXPTR);
5091 	} while (gotlock == 0);
5092 	lfp->lf_usecount--;
5093 }
5094 
5095 /*
5096  * Unlock the struct nfslockfile after local lock updating.
5097  */
5098 static void
5099 nfsrv_unlocklf(struct nfslockfile *lfp)
5100 {
5101 
5102 	nfsv4_unlock(&lfp->lf_locallock_lck, 0);
5103 }
5104 
5105