xref: /freebsd/sys/fs/nfsserver/nfs_nfsdstate.c (revision 830940567b49bb0c08dfaed40418999e76616909)
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 = 1;
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, int *freedlockp,
69     int cansleep, NFSPROC_T *p);
70 static int nfsrv_freelockowner(struct nfsstate *stp, int *freedlockp,
71     int cansleep, NFSPROC_T *p);
72 static int nfsrv_freeallnfslocks(struct nfsstate *stp, int *freedlockp,
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,
84     struct nfslockfile **new_lfpp, struct nfslockfile **lfpp, fhandle_t *nfhp);
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 void nfsrv_locallocks(vnode_t vp, struct nfslockfile *lfp,
113     NFSPROC_T *p);
114 static int nfsrv_nootherstate(struct nfsstate *stp);
115 
116 /*
117  * Scan the client list for a match and either return the current one,
118  * create a new entry or return an error.
119  * If returning a non-error, the clp structure must either be linked into
120  * the client list or free'd.
121  */
122 APPLESTATIC int
123 nfsrv_setclient(struct nfsrv_descript *nd, struct nfsclient **new_clpp,
124     nfsquad_t *clientidp, nfsquad_t *confirmp, NFSPROC_T *p)
125 {
126 	struct nfsclient *clp = NULL, *new_clp = *new_clpp;
127 	int i;
128 	struct nfsstate *stp, *tstp;
129 	struct sockaddr_in *sad, *rad;
130 	int zapit = 0, gotit, hasstate = 0, igotlock;
131 	static u_int64_t confirm_index = 0;
132 
133 	/*
134 	 * Check for state resource limit exceeded.
135 	 */
136 	if (nfsrv_openpluslock > NFSRV_V4STATELIMIT)
137 		return (NFSERR_RESOURCE);
138 
139 	if ((nd->nd_flag & ND_GSS) && nfsrv_nogsscallback)
140 		/*
141 		 * Don't do callbacks for AUTH_GSS.
142 		 * (Since these aren't yet debugged, they might cause the
143 		 *  server to crap out, if they get past the Init call to
144 		 *  the client.)
145 		 */
146 		new_clp->lc_program = 0;
147 
148 	/* Lock out other nfsd threads */
149 	NFSLOCKV4ROOTMUTEX();
150 	nfsv4_relref(&nfsv4rootfs_lock);
151 	do {
152 		igotlock = nfsv4_lock(&nfsv4rootfs_lock, 1, NULL,
153 		    NFSV4ROOTLOCKMUTEXPTR);
154 	} while (!igotlock);
155 	NFSUNLOCKV4ROOTMUTEX();
156 	NFSLOCKSTATE();	/* to avoid a race with */
157 	NFSUNLOCKSTATE();	/* nfsrv_servertimer() */
158 
159 	/*
160 	 * Search for a match in the client list.
161 	 */
162 	gotit = i = 0;
163 	while (i < NFSCLIENTHASHSIZE && !gotit) {
164 	    LIST_FOREACH(clp, &nfsclienthash[i], lc_hash) {
165 		if (new_clp->lc_idlen == clp->lc_idlen &&
166 		    !NFSBCMP(new_clp->lc_id, clp->lc_id, clp->lc_idlen)) {
167 			gotit = 1;
168 			break;
169 		}
170 	    }
171 	    i++;
172 	}
173 	if (!gotit ||
174 	    (clp->lc_flags & (LCL_NEEDSCONFIRM | LCL_ADMINREVOKED))) {
175 		/*
176 		 * Get rid of the old one.
177 		 */
178 		if (i != NFSCLIENTHASHSIZE) {
179 			LIST_REMOVE(clp, lc_hash);
180 			nfsrv_cleanclient(clp, p);
181 			nfsrv_freedeleglist(&clp->lc_deleg);
182 			nfsrv_freedeleglist(&clp->lc_olddeleg);
183 			zapit = 1;
184 		}
185 		/*
186 		 * Add it after assigning a client id to it.
187 		 */
188 		new_clp->lc_flags |= LCL_NEEDSCONFIRM;
189 		confirmp->qval = new_clp->lc_confirm.qval = ++confirm_index;
190 		clientidp->lval[0] = new_clp->lc_clientid.lval[0] =
191 		    (u_int32_t)nfsrvboottime;
192 		clientidp->lval[1] = new_clp->lc_clientid.lval[1] =
193 		    nfsrv_nextclientindex();
194 		new_clp->lc_stateindex = 0;
195 		new_clp->lc_statemaxindex = 0;
196 		new_clp->lc_cbref = 0;
197 		new_clp->lc_expiry = nfsrv_leaseexpiry();
198 		LIST_INIT(&new_clp->lc_open);
199 		LIST_INIT(&new_clp->lc_deleg);
200 		LIST_INIT(&new_clp->lc_olddeleg);
201 		for (i = 0; i < NFSSTATEHASHSIZE; i++)
202 			LIST_INIT(&new_clp->lc_stateid[i]);
203 		LIST_INSERT_HEAD(NFSCLIENTHASH(new_clp->lc_clientid), new_clp,
204 		    lc_hash);
205 		newnfsstats.srvclients++;
206 		nfsrv_openpluslock++;
207 		nfsrv_clients++;
208 		NFSLOCKV4ROOTMUTEX();
209 		nfsv4_unlock(&nfsv4rootfs_lock, 1);
210 		NFSUNLOCKV4ROOTMUTEX();
211 		if (zapit)
212 			nfsrv_zapclient(clp, p);
213 		*new_clpp = NULL;
214 		return (0);
215 	}
216 
217 	/*
218 	 * Now, handle the cases where the id is already issued.
219 	 */
220 	if (nfsrv_notsamecredname(nd, clp)) {
221 	    /*
222 	     * Check to see if there is expired state that should go away.
223 	     */
224 	    if (clp->lc_expiry < NFSD_MONOSEC &&
225 	        (!LIST_EMPTY(&clp->lc_open) || !LIST_EMPTY(&clp->lc_deleg))) {
226 		nfsrv_cleanclient(clp, p);
227 		nfsrv_freedeleglist(&clp->lc_deleg);
228 	    }
229 
230 	    /*
231 	     * If there is outstanding state, then reply NFSERR_CLIDINUSE per
232 	     * RFC3530 Sec. 8.1.2 last para.
233 	     */
234 	    if (!LIST_EMPTY(&clp->lc_deleg)) {
235 		hasstate = 1;
236 	    } else if (LIST_EMPTY(&clp->lc_open)) {
237 		hasstate = 0;
238 	    } else {
239 		hasstate = 0;
240 		/* Look for an Open on the OpenOwner */
241 		LIST_FOREACH(stp, &clp->lc_open, ls_list) {
242 		    if (!LIST_EMPTY(&stp->ls_open)) {
243 			hasstate = 1;
244 			break;
245 		    }
246 		}
247 	    }
248 	    if (hasstate) {
249 		/*
250 		 * If the uid doesn't match, return NFSERR_CLIDINUSE after
251 		 * filling out the correct ipaddr and portnum.
252 		 */
253 		sad = NFSSOCKADDR(new_clp->lc_req.nr_nam, struct sockaddr_in *);
254 		rad = NFSSOCKADDR(clp->lc_req.nr_nam, struct sockaddr_in *);
255 		sad->sin_addr.s_addr = rad->sin_addr.s_addr;
256 		sad->sin_port = rad->sin_port;
257 		NFSLOCKV4ROOTMUTEX();
258 		nfsv4_unlock(&nfsv4rootfs_lock, 1);
259 		NFSUNLOCKV4ROOTMUTEX();
260 		return (NFSERR_CLIDINUSE);
261 	    }
262 	}
263 
264 	if (NFSBCMP(new_clp->lc_verf, clp->lc_verf, NFSX_VERF)) {
265 		/*
266 		 * If the verifier has changed, the client has rebooted
267 		 * and a new client id is issued. The old state info
268 		 * can be thrown away once the SETCLIENTID_CONFIRM occurs.
269 		 */
270 		LIST_REMOVE(clp, lc_hash);
271 		new_clp->lc_flags |= LCL_NEEDSCONFIRM;
272 		confirmp->qval = new_clp->lc_confirm.qval = ++confirm_index;
273 		clientidp->lval[0] = new_clp->lc_clientid.lval[0] =
274 		    nfsrvboottime;
275 		clientidp->lval[1] = new_clp->lc_clientid.lval[1] =
276 		    nfsrv_nextclientindex();
277 		new_clp->lc_stateindex = 0;
278 		new_clp->lc_statemaxindex = 0;
279 		new_clp->lc_cbref = 0;
280 		new_clp->lc_expiry = nfsrv_leaseexpiry();
281 
282 		/*
283 		 * Save the state until confirmed.
284 		 */
285 		LIST_NEWHEAD(&new_clp->lc_open, &clp->lc_open, ls_list);
286 		LIST_FOREACH(tstp, &new_clp->lc_open, ls_list)
287 			tstp->ls_clp = new_clp;
288 		LIST_NEWHEAD(&new_clp->lc_deleg, &clp->lc_deleg, ls_list);
289 		LIST_FOREACH(tstp, &new_clp->lc_deleg, ls_list)
290 			tstp->ls_clp = new_clp;
291 		LIST_NEWHEAD(&new_clp->lc_olddeleg, &clp->lc_olddeleg,
292 		    ls_list);
293 		LIST_FOREACH(tstp, &new_clp->lc_olddeleg, ls_list)
294 			tstp->ls_clp = new_clp;
295 		for (i = 0; i < NFSSTATEHASHSIZE; i++) {
296 			LIST_NEWHEAD(&new_clp->lc_stateid[i],
297 			    &clp->lc_stateid[i], ls_hash);
298 			LIST_FOREACH(tstp, &new_clp->lc_stateid[i], ls_list)
299 				tstp->ls_clp = new_clp;
300 		}
301 		LIST_INSERT_HEAD(NFSCLIENTHASH(new_clp->lc_clientid), new_clp,
302 		    lc_hash);
303 		newnfsstats.srvclients++;
304 		nfsrv_openpluslock++;
305 		nfsrv_clients++;
306 		NFSLOCKV4ROOTMUTEX();
307 		nfsv4_unlock(&nfsv4rootfs_lock, 1);
308 		NFSUNLOCKV4ROOTMUTEX();
309 
310 		/*
311 		 * Must wait until any outstanding callback on the old clp
312 		 * completes.
313 		 */
314 		while (clp->lc_cbref) {
315 			clp->lc_flags |= LCL_WAKEUPWANTED;
316 			(void) tsleep((caddr_t)clp, PZERO - 1,
317 			    "nfsd clp", 10 * hz);
318 		}
319 		nfsrv_zapclient(clp, p);
320 		*new_clpp = NULL;
321 		return (0);
322 	}
323 	/*
324 	 * id and verifier match, so update the net address info
325 	 * and get rid of any existing callback authentication
326 	 * handle, so a new one will be acquired.
327 	 */
328 	LIST_REMOVE(clp, lc_hash);
329 	new_clp->lc_flags |= (LCL_NEEDSCONFIRM | LCL_DONTCLEAN);
330 	new_clp->lc_expiry = nfsrv_leaseexpiry();
331 	confirmp->qval = new_clp->lc_confirm.qval = ++confirm_index;
332 	clientidp->lval[0] = new_clp->lc_clientid.lval[0] =
333 	    clp->lc_clientid.lval[0];
334 	clientidp->lval[1] = new_clp->lc_clientid.lval[1] =
335 	    clp->lc_clientid.lval[1];
336 	new_clp->lc_delegtime = clp->lc_delegtime;
337 	new_clp->lc_stateindex = clp->lc_stateindex;
338 	new_clp->lc_statemaxindex = clp->lc_statemaxindex;
339 	new_clp->lc_cbref = 0;
340 	LIST_NEWHEAD(&new_clp->lc_open, &clp->lc_open, ls_list);
341 	LIST_FOREACH(tstp, &new_clp->lc_open, ls_list)
342 		tstp->ls_clp = new_clp;
343 	LIST_NEWHEAD(&new_clp->lc_deleg, &clp->lc_deleg, ls_list);
344 	LIST_FOREACH(tstp, &new_clp->lc_deleg, ls_list)
345 		tstp->ls_clp = new_clp;
346 	LIST_NEWHEAD(&new_clp->lc_olddeleg, &clp->lc_olddeleg, ls_list);
347 	LIST_FOREACH(tstp, &new_clp->lc_olddeleg, ls_list)
348 		tstp->ls_clp = new_clp;
349 	for (i = 0; i < NFSSTATEHASHSIZE; i++) {
350 		LIST_NEWHEAD(&new_clp->lc_stateid[i], &clp->lc_stateid[i],
351 		    ls_hash);
352 		LIST_FOREACH(tstp, &new_clp->lc_stateid[i], ls_list)
353 			tstp->ls_clp = new_clp;
354 	}
355 	LIST_INSERT_HEAD(NFSCLIENTHASH(new_clp->lc_clientid), new_clp,
356 	    lc_hash);
357 	newnfsstats.srvclients++;
358 	nfsrv_openpluslock++;
359 	nfsrv_clients++;
360 	NFSLOCKV4ROOTMUTEX();
361 	nfsv4_unlock(&nfsv4rootfs_lock, 1);
362 	NFSUNLOCKV4ROOTMUTEX();
363 
364 	/*
365 	 * Must wait until any outstanding callback on the old clp
366 	 * completes.
367 	 */
368 	while (clp->lc_cbref) {
369 		clp->lc_flags |= LCL_WAKEUPWANTED;
370 		(void) tsleep((caddr_t)clp, PZERO - 1, "nfsd clp", 10 * hz);
371 	}
372 	nfsrv_zapclient(clp, p);
373 	*new_clpp = NULL;
374 	return (0);
375 }
376 
377 /*
378  * Check to see if the client id exists and optionally confirm it.
379  */
380 APPLESTATIC int
381 nfsrv_getclient(nfsquad_t clientid, int opflags, struct nfsclient **clpp,
382     nfsquad_t confirm, struct nfsrv_descript *nd, NFSPROC_T *p)
383 {
384 	struct nfsclient *clp;
385 	struct nfsstate *stp;
386 	int i;
387 	struct nfsclienthashhead *hp;
388 	int error = 0, igotlock, doneok;
389 
390 	if (clpp)
391 		*clpp = NULL;
392 	if (nfsrvboottime != clientid.lval[0])
393 		return (NFSERR_STALECLIENTID);
394 
395 	/*
396 	 * If called with opflags == CLOPS_RENEW, the State Lock is
397 	 * already held. Otherwise, we need to get either that or,
398 	 * for the case of Confirm, lock out the nfsd threads.
399 	 */
400 	if (opflags & CLOPS_CONFIRM) {
401 		NFSLOCKV4ROOTMUTEX();
402 		nfsv4_relref(&nfsv4rootfs_lock);
403 		do {
404 			igotlock = nfsv4_lock(&nfsv4rootfs_lock, 1, NULL,
405 			    NFSV4ROOTLOCKMUTEXPTR);
406 		} while (!igotlock);
407 		NFSUNLOCKV4ROOTMUTEX();
408 		NFSLOCKSTATE();	/* to avoid a race with */
409 		NFSUNLOCKSTATE();	/* nfsrv_servertimer() */
410 	} else if (opflags != CLOPS_RENEW) {
411 		NFSLOCKSTATE();
412 	}
413 
414 	hp = NFSCLIENTHASH(clientid);
415 	LIST_FOREACH(clp, hp, lc_hash) {
416 		if (clp->lc_clientid.lval[1] == clientid.lval[1])
417 			break;
418 	}
419 	if (clp == LIST_END(hp)) {
420 		if (opflags & CLOPS_CONFIRM)
421 			error = NFSERR_STALECLIENTID;
422 		else
423 			error = NFSERR_EXPIRED;
424 	} else if (clp->lc_flags & LCL_ADMINREVOKED) {
425 		/*
426 		 * If marked admin revoked, just return the error.
427 		 */
428 		error = NFSERR_ADMINREVOKED;
429 	}
430 	if (error) {
431 		if (opflags & CLOPS_CONFIRM) {
432 			NFSLOCKV4ROOTMUTEX();
433 			nfsv4_unlock(&nfsv4rootfs_lock, 1);
434 			NFSUNLOCKV4ROOTMUTEX();
435 		} else if (opflags != CLOPS_RENEW) {
436 			NFSUNLOCKSTATE();
437 		}
438 		return (error);
439 	}
440 
441 	/*
442 	 * Perform any operations specified by the opflags.
443 	 */
444 	if (opflags & CLOPS_CONFIRM) {
445 		if (clp->lc_confirm.qval != confirm.qval)
446 			error = NFSERR_STALECLIENTID;
447 		else if (nfsrv_notsamecredname(nd, clp))
448 			error = NFSERR_CLIDINUSE;
449 
450 		if (!error) {
451 		    if ((clp->lc_flags & (LCL_NEEDSCONFIRM | LCL_DONTCLEAN)) ==
452 			LCL_NEEDSCONFIRM) {
453 			/*
454 			 * Hang onto the delegations (as old delegations)
455 			 * for an Open with CLAIM_DELEGATE_PREV unless in
456 			 * grace, but get rid of the rest of the state.
457 			 */
458 			nfsrv_cleanclient(clp, p);
459 			nfsrv_freedeleglist(&clp->lc_olddeleg);
460 			if (nfsrv_checkgrace(0)) {
461 			    /* In grace, so just delete delegations */
462 			    nfsrv_freedeleglist(&clp->lc_deleg);
463 			} else {
464 			    LIST_FOREACH(stp, &clp->lc_deleg, ls_list)
465 				stp->ls_flags |= NFSLCK_OLDDELEG;
466 			    clp->lc_delegtime = NFSD_MONOSEC +
467 				nfsrv_lease + NFSRV_LEASEDELTA;
468 			    LIST_NEWHEAD(&clp->lc_olddeleg, &clp->lc_deleg,
469 				ls_list);
470 			}
471 		    }
472 		    clp->lc_flags &= ~(LCL_NEEDSCONFIRM | LCL_DONTCLEAN);
473 		    if (clp->lc_program)
474 			clp->lc_flags |= LCL_NEEDSCBNULL;
475 		}
476 	} else if (clp->lc_flags & LCL_NEEDSCONFIRM) {
477 		error = NFSERR_EXPIRED;
478 	}
479 
480 	/*
481 	 * If called by the Renew Op, we must check the principal.
482 	 */
483 	if (!error && (opflags & CLOPS_RENEWOP)) {
484 	    if (nfsrv_notsamecredname(nd, clp)) {
485 		doneok = 0;
486 		for (i = 0; i < NFSSTATEHASHSIZE && doneok == 0; i++) {
487 		    LIST_FOREACH(stp, &clp->lc_stateid[i], ls_hash) {
488 			if ((stp->ls_flags & NFSLCK_OPEN) &&
489 			    stp->ls_uid == nd->nd_cred->cr_uid) {
490 				doneok = 1;
491 				break;
492 			}
493 		    }
494 		}
495 		if (!doneok)
496 			error = NFSERR_ACCES;
497 	    }
498 	    if (!error && (clp->lc_flags & LCL_CBDOWN))
499 		error = NFSERR_CBPATHDOWN;
500 	}
501 	if ((!error || error == NFSERR_CBPATHDOWN) &&
502 	     (opflags & CLOPS_RENEW)) {
503 		clp->lc_expiry = nfsrv_leaseexpiry();
504 	}
505 	if (opflags & CLOPS_CONFIRM) {
506 		NFSLOCKV4ROOTMUTEX();
507 		nfsv4_unlock(&nfsv4rootfs_lock, 1);
508 		NFSUNLOCKV4ROOTMUTEX();
509 	} else if (opflags != CLOPS_RENEW) {
510 		NFSUNLOCKSTATE();
511 	}
512 	if (clpp)
513 		*clpp = clp;
514 	return (error);
515 }
516 
517 /*
518  * Called from the new nfssvc syscall to admin revoke a clientid.
519  * Returns 0 for success, error otherwise.
520  */
521 APPLESTATIC int
522 nfsrv_adminrevoke(struct nfsd_clid *revokep, NFSPROC_T *p)
523 {
524 	struct nfsclient *clp = NULL;
525 	int i;
526 	int gotit, igotlock;
527 
528 	/*
529 	 * First, lock out the nfsd so that state won't change while the
530 	 * revocation record is being written to the stable storage restart
531 	 * file.
532 	 */
533 	NFSLOCKV4ROOTMUTEX();
534 	nfsv4_relref(&nfsv4rootfs_lock);
535 	do {
536 		igotlock = nfsv4_lock(&nfsv4rootfs_lock, 1, NULL,
537 		    NFSV4ROOTLOCKMUTEXPTR);
538 	} while (!igotlock);
539 	NFSUNLOCKV4ROOTMUTEX();
540 	NFSLOCKSTATE();	/* to avoid a race with */
541 	NFSUNLOCKSTATE();	/* nfsrv_servertimer() */
542 
543 	/*
544 	 * Search for a match in the client list.
545 	 */
546 	gotit = i = 0;
547 	while (i < NFSCLIENTHASHSIZE && !gotit) {
548 	    LIST_FOREACH(clp, &nfsclienthash[i], lc_hash) {
549 		if (revokep->nclid_idlen == clp->lc_idlen &&
550 		    !NFSBCMP(revokep->nclid_id, clp->lc_id, clp->lc_idlen)) {
551 			gotit = 1;
552 			break;
553 		}
554 	    }
555 	    i++;
556 	}
557 	if (!gotit) {
558 		NFSLOCKV4ROOTMUTEX();
559 		nfsv4_unlock(&nfsv4rootfs_lock, 0);
560 		NFSUNLOCKV4ROOTMUTEX();
561 		return (EPERM);
562 	}
563 
564 	/*
565 	 * Now, write out the revocation record
566 	 */
567 	nfsrv_writestable(clp->lc_id, clp->lc_idlen, NFSNST_REVOKE, p);
568 
569 	/*
570 	 * and clear out the state, marking the clientid revoked.
571 	 */
572 	clp->lc_flags &= ~LCL_CALLBACKSON;
573 	clp->lc_flags |= LCL_ADMINREVOKED;
574 	nfsrv_cleanclient(clp, p);
575 	nfsrv_freedeleglist(&clp->lc_deleg);
576 	nfsrv_freedeleglist(&clp->lc_olddeleg);
577 	NFSLOCKV4ROOTMUTEX();
578 	nfsv4_unlock(&nfsv4rootfs_lock, 0);
579 	NFSUNLOCKV4ROOTMUTEX();
580 	return (0);
581 }
582 
583 /*
584  * Dump out stats for all clients. Called from nfssvc(2), that is used
585  * newnfsstats.
586  */
587 APPLESTATIC void
588 nfsrv_dumpclients(struct nfsd_dumpclients *dumpp, int maxcnt)
589 {
590 	struct nfsclient *clp;
591 	int i = 0, cnt = 0;
592 
593 	NFSLOCKSTATE();
594 	/*
595 	 * Rattle through the client lists until done.
596 	 */
597 	while (i < NFSCLIENTHASHSIZE && cnt < maxcnt) {
598 	    clp = LIST_FIRST(&nfsclienthash[i]);
599 	    while (clp != LIST_END(&nfsclienthash[i]) && cnt < maxcnt) {
600 		nfsrv_dumpaclient(clp, &dumpp[cnt]);
601 		cnt++;
602 		clp = LIST_NEXT(clp, lc_hash);
603 	    }
604 	    i++;
605 	}
606 	if (cnt < maxcnt)
607 	    dumpp[cnt].ndcl_clid.nclid_idlen = 0;
608 	NFSUNLOCKSTATE();
609 }
610 
611 /*
612  * Dump stats for a client. Must be called with the NFSSTATELOCK and spl'd.
613  */
614 static void
615 nfsrv_dumpaclient(struct nfsclient *clp, struct nfsd_dumpclients *dumpp)
616 {
617 	struct nfsstate *stp, *openstp, *lckownstp;
618 	struct nfslock *lop;
619 	struct sockaddr *sad;
620 	struct sockaddr_in *rad;
621 	struct sockaddr_in6 *rad6;
622 
623 	dumpp->ndcl_nopenowners = dumpp->ndcl_nlockowners = 0;
624 	dumpp->ndcl_nopens = dumpp->ndcl_nlocks = 0;
625 	dumpp->ndcl_ndelegs = dumpp->ndcl_nolddelegs = 0;
626 	dumpp->ndcl_flags = clp->lc_flags;
627 	dumpp->ndcl_clid.nclid_idlen = clp->lc_idlen;
628 	NFSBCOPY(clp->lc_id, dumpp->ndcl_clid.nclid_id, clp->lc_idlen);
629 	sad = NFSSOCKADDR(clp->lc_req.nr_nam, struct sockaddr *);
630 	dumpp->ndcl_addrfam = sad->sa_family;
631 	if (sad->sa_family == AF_INET) {
632 		rad = (struct sockaddr_in *)sad;
633 		dumpp->ndcl_cbaddr.sin_addr = rad->sin_addr;
634 	} else {
635 		rad6 = (struct sockaddr_in6 *)sad;
636 		dumpp->ndcl_cbaddr.sin6_addr = rad6->sin6_addr;
637 	}
638 
639 	/*
640 	 * Now, scan the state lists and total up the opens and locks.
641 	 */
642 	LIST_FOREACH(stp, &clp->lc_open, ls_list) {
643 	    dumpp->ndcl_nopenowners++;
644 	    LIST_FOREACH(openstp, &stp->ls_open, ls_list) {
645 		dumpp->ndcl_nopens++;
646 		LIST_FOREACH(lckownstp, &openstp->ls_open, ls_list) {
647 		    dumpp->ndcl_nlockowners++;
648 		    LIST_FOREACH(lop, &lckownstp->ls_lock, lo_lckowner) {
649 			dumpp->ndcl_nlocks++;
650 		    }
651 		}
652 	    }
653 	}
654 
655 	/*
656 	 * and the delegation lists.
657 	 */
658 	LIST_FOREACH(stp, &clp->lc_deleg, ls_list) {
659 	    dumpp->ndcl_ndelegs++;
660 	}
661 	LIST_FOREACH(stp, &clp->lc_olddeleg, ls_list) {
662 	    dumpp->ndcl_nolddelegs++;
663 	}
664 }
665 
666 /*
667  * Dump out lock stats for a file.
668  */
669 APPLESTATIC void
670 nfsrv_dumplocks(vnode_t vp, struct nfsd_dumplocks *ldumpp, int maxcnt,
671     NFSPROC_T *p)
672 {
673 	struct nfsstate *stp;
674 	struct nfslock *lop;
675 	int cnt = 0;
676 	struct nfslockfile *lfp;
677 	struct sockaddr *sad;
678 	struct sockaddr_in *rad;
679 	struct sockaddr_in6 *rad6;
680 	int ret;
681 	fhandle_t nfh;
682 
683 	ret = nfsrv_getlockfh(vp, 0, NULL, &nfh, p);
684 	NFSLOCKSTATE();
685 	if (!ret)
686 		ret = nfsrv_getlockfile(0, NULL, &lfp, &nfh);
687 	if (ret) {
688 		ldumpp[0].ndlck_clid.nclid_idlen = 0;
689 		NFSUNLOCKSTATE();
690 		return;
691 	}
692 
693 	/*
694 	 * For each open share on file, dump it out.
695 	 */
696 	stp = LIST_FIRST(&lfp->lf_open);
697 	while (stp != LIST_END(&lfp->lf_open) && cnt < maxcnt) {
698 		ldumpp[cnt].ndlck_flags = stp->ls_flags;
699 		ldumpp[cnt].ndlck_stateid.seqid = stp->ls_stateid.seqid;
700 		ldumpp[cnt].ndlck_stateid.other[0] = stp->ls_stateid.other[0];
701 		ldumpp[cnt].ndlck_stateid.other[1] = stp->ls_stateid.other[1];
702 		ldumpp[cnt].ndlck_stateid.other[2] = stp->ls_stateid.other[2];
703 		ldumpp[cnt].ndlck_owner.nclid_idlen =
704 		    stp->ls_openowner->ls_ownerlen;
705 		NFSBCOPY(stp->ls_openowner->ls_owner,
706 		    ldumpp[cnt].ndlck_owner.nclid_id,
707 		    stp->ls_openowner->ls_ownerlen);
708 		ldumpp[cnt].ndlck_clid.nclid_idlen = stp->ls_clp->lc_idlen;
709 		NFSBCOPY(stp->ls_clp->lc_id, ldumpp[cnt].ndlck_clid.nclid_id,
710 		    stp->ls_clp->lc_idlen);
711 		sad=NFSSOCKADDR(stp->ls_clp->lc_req.nr_nam, struct sockaddr *);
712 		ldumpp[cnt].ndlck_addrfam = sad->sa_family;
713 		if (sad->sa_family == AF_INET) {
714 			rad = (struct sockaddr_in *)sad;
715 			ldumpp[cnt].ndlck_cbaddr.sin_addr = rad->sin_addr;
716 		} else {
717 			rad6 = (struct sockaddr_in6 *)sad;
718 			ldumpp[cnt].ndlck_cbaddr.sin6_addr = rad6->sin6_addr;
719 		}
720 		stp = LIST_NEXT(stp, ls_file);
721 		cnt++;
722 	}
723 
724 	/*
725 	 * and all locks.
726 	 */
727 	lop = LIST_FIRST(&lfp->lf_lock);
728 	while (lop != LIST_END(&lfp->lf_lock) && cnt < maxcnt) {
729 		stp = lop->lo_stp;
730 		ldumpp[cnt].ndlck_flags = lop->lo_flags;
731 		ldumpp[cnt].ndlck_first = lop->lo_first;
732 		ldumpp[cnt].ndlck_end = lop->lo_end;
733 		ldumpp[cnt].ndlck_stateid.seqid = stp->ls_stateid.seqid;
734 		ldumpp[cnt].ndlck_stateid.other[0] = stp->ls_stateid.other[0];
735 		ldumpp[cnt].ndlck_stateid.other[1] = stp->ls_stateid.other[1];
736 		ldumpp[cnt].ndlck_stateid.other[2] = stp->ls_stateid.other[2];
737 		ldumpp[cnt].ndlck_owner.nclid_idlen = stp->ls_ownerlen;
738 		NFSBCOPY(stp->ls_owner, ldumpp[cnt].ndlck_owner.nclid_id,
739 		    stp->ls_ownerlen);
740 		ldumpp[cnt].ndlck_clid.nclid_idlen = stp->ls_clp->lc_idlen;
741 		NFSBCOPY(stp->ls_clp->lc_id, ldumpp[cnt].ndlck_clid.nclid_id,
742 		    stp->ls_clp->lc_idlen);
743 		sad=NFSSOCKADDR(stp->ls_clp->lc_req.nr_nam, struct sockaddr *);
744 		ldumpp[cnt].ndlck_addrfam = sad->sa_family;
745 		if (sad->sa_family == AF_INET) {
746 			rad = (struct sockaddr_in *)sad;
747 			ldumpp[cnt].ndlck_cbaddr.sin_addr = rad->sin_addr;
748 		} else {
749 			rad6 = (struct sockaddr_in6 *)sad;
750 			ldumpp[cnt].ndlck_cbaddr.sin6_addr = rad6->sin6_addr;
751 		}
752 		lop = LIST_NEXT(lop, lo_lckfile);
753 		cnt++;
754 	}
755 
756 	/*
757 	 * and the delegations.
758 	 */
759 	stp = LIST_FIRST(&lfp->lf_deleg);
760 	while (stp != LIST_END(&lfp->lf_deleg) && cnt < maxcnt) {
761 		ldumpp[cnt].ndlck_flags = stp->ls_flags;
762 		ldumpp[cnt].ndlck_stateid.seqid = stp->ls_stateid.seqid;
763 		ldumpp[cnt].ndlck_stateid.other[0] = stp->ls_stateid.other[0];
764 		ldumpp[cnt].ndlck_stateid.other[1] = stp->ls_stateid.other[1];
765 		ldumpp[cnt].ndlck_stateid.other[2] = stp->ls_stateid.other[2];
766 		ldumpp[cnt].ndlck_owner.nclid_idlen = 0;
767 		ldumpp[cnt].ndlck_clid.nclid_idlen = stp->ls_clp->lc_idlen;
768 		NFSBCOPY(stp->ls_clp->lc_id, ldumpp[cnt].ndlck_clid.nclid_id,
769 		    stp->ls_clp->lc_idlen);
770 		sad=NFSSOCKADDR(stp->ls_clp->lc_req.nr_nam, struct sockaddr *);
771 		ldumpp[cnt].ndlck_addrfam = sad->sa_family;
772 		if (sad->sa_family == AF_INET) {
773 			rad = (struct sockaddr_in *)sad;
774 			ldumpp[cnt].ndlck_cbaddr.sin_addr = rad->sin_addr;
775 		} else {
776 			rad6 = (struct sockaddr_in6 *)sad;
777 			ldumpp[cnt].ndlck_cbaddr.sin6_addr = rad6->sin6_addr;
778 		}
779 		stp = LIST_NEXT(stp, ls_file);
780 		cnt++;
781 	}
782 
783 	/*
784 	 * If list isn't full, mark end of list by setting the client name
785 	 * to zero length.
786 	 */
787 	if (cnt < maxcnt)
788 		ldumpp[cnt].ndlck_clid.nclid_idlen = 0;
789 	NFSUNLOCKSTATE();
790 }
791 
792 /*
793  * Server timer routine. It can scan any linked list, so long
794  * as it holds the spin lock and there is no exclusive lock on
795  * nfsv4rootfs_lock.
796  * Must be called by a kernel thread and not a timer interrupt,
797  * so that it only runs when the nfsd threads are sleeping on a
798  * uniprocessor and uses the State spin lock for an SMP system.
799  * (For OpenBSD, a kthread is ok. For FreeBSD, I think it is ok
800  *  to do this from a callout, since the spin locks work. For
801  *  Darwin, I'm not sure what will work correctly yet.)
802  * Should be called once per second.
803  */
804 APPLESTATIC void
805 nfsrv_servertimer(void)
806 {
807 	struct nfsclient *clp, *nclp;
808 	struct nfsstate *stp, *nstp;
809 	int i;
810 
811 	/*
812 	 * Make sure nfsboottime is set. This is used by V3 as well
813 	 * as V4. Note that nfsboottime is not nfsrvboottime, which is
814 	 * only used by the V4 server for leases.
815 	 */
816 	if (nfsboottime.tv_sec == 0)
817 		NFSSETBOOTTIME(nfsboottime);
818 
819 	/*
820 	 * If server hasn't started yet, just return.
821 	 */
822 	NFSLOCKSTATE();
823 	if (nfsrv_stablefirst.nsf_eograce == 0) {
824 		NFSUNLOCKSTATE();
825 		return;
826 	}
827 	if (!(nfsrv_stablefirst.nsf_flags & NFSNSF_UPDATEDONE)) {
828 		if (!(nfsrv_stablefirst.nsf_flags & NFSNSF_GRACEOVER) &&
829 		    NFSD_MONOSEC > nfsrv_stablefirst.nsf_eograce)
830 			nfsrv_stablefirst.nsf_flags |=
831 			    (NFSNSF_GRACEOVER | NFSNSF_NEEDLOCK);
832 		NFSUNLOCKSTATE();
833 		return;
834 	}
835 
836 	/*
837 	 * Return now if an nfsd thread has the exclusive lock on
838 	 * nfsv4rootfs_lock. The dirty trick here is that we have
839 	 * the spin lock already and the nfsd threads do a:
840 	 * NFSLOCKSTATE, NFSUNLOCKSTATE after getting the exclusive
841 	 * lock, so they won't race with code after this check.
842 	 */
843 	if (nfsv4rootfs_lock.nfslock_lock & NFSV4LOCK_LOCK) {
844 		NFSUNLOCKSTATE();
845 		return;
846 	}
847 
848 	/*
849 	 * For each client...
850 	 */
851 	for (i = 0; i < NFSCLIENTHASHSIZE; i++) {
852 	    clp = LIST_FIRST(&nfsclienthash[i]);
853 	    while (clp != LIST_END(&nfsclienthash[i])) {
854 		nclp = LIST_NEXT(clp, lc_hash);
855 		if (!(clp->lc_flags & LCL_EXPIREIT)) {
856 		    if (((clp->lc_expiry + NFSRV_STALELEASE) < NFSD_MONOSEC
857 			 && ((LIST_EMPTY(&clp->lc_deleg)
858 			      && LIST_EMPTY(&clp->lc_open)) ||
859 			     nfsrv_clients > nfsrv_clienthighwater)) ||
860 			(clp->lc_expiry + NFSRV_MOULDYLEASE) < NFSD_MONOSEC ||
861 			(clp->lc_expiry < NFSD_MONOSEC &&
862 			 (nfsrv_openpluslock * 10 / 9) > NFSRV_V4STATELIMIT)) {
863 			/*
864 			 * Lease has expired several nfsrv_lease times ago:
865 			 * PLUS
866 			 *    - no state is associated with it
867 			 *    OR
868 			 *    - above high water mark for number of clients
869 			 *      (nfsrv_clienthighwater should be large enough
870 			 *       that this only occurs when clients fail to
871 			 *       use the same nfs_client_id4.id. Maybe somewhat
872 			 *       higher that the maximum number of clients that
873 			 *       will mount this server?)
874 			 * OR
875 			 * Lease has expired a very long time ago
876 			 * OR
877 			 * Lease has expired PLUS the number of opens + locks
878 			 * has exceeded 90% of capacity
879 			 *
880 			 * --> Mark for expiry. The actual expiry will be done
881 			 *     by an nfsd sometime soon.
882 			 */
883 			clp->lc_flags |= LCL_EXPIREIT;
884 			nfsrv_stablefirst.nsf_flags |=
885 			    (NFSNSF_NEEDLOCK | NFSNSF_EXPIREDCLIENT);
886 		    } else {
887 			/*
888 			 * If there are no opens, increment no open tick cnt
889 			 * If time exceeds NFSNOOPEN, mark it to be thrown away
890 			 * otherwise, if there is an open, reset no open time
891 			 * Hopefully, this will avoid excessive re-creation
892 			 * of open owners and subsequent open confirms.
893 			 */
894 			stp = LIST_FIRST(&clp->lc_open);
895 			while (stp != LIST_END(&clp->lc_open)) {
896 				nstp = LIST_NEXT(stp, ls_list);
897 				if (LIST_EMPTY(&stp->ls_open)) {
898 					stp->ls_noopens++;
899 					if (stp->ls_noopens > NFSNOOPEN ||
900 					    (nfsrv_openpluslock * 2) >
901 					    NFSRV_V4STATELIMIT)
902 						nfsrv_stablefirst.nsf_flags |=
903 							NFSNSF_NOOPENS;
904 				} else {
905 					stp->ls_noopens = 0;
906 				}
907 				stp = nstp;
908 			}
909 		    }
910 		}
911 		clp = nclp;
912 	    }
913 	}
914 	NFSUNLOCKSTATE();
915 }
916 
917 /*
918  * The following set of functions free up the various data structures.
919  */
920 /*
921  * Clear out all open/lock state related to this nfsclient.
922  * Caller must hold an exclusive lock on nfsv4rootfs_lock, so that
923  * there are no other active nfsd threads.
924  */
925 APPLESTATIC void
926 nfsrv_cleanclient(struct nfsclient *clp, NFSPROC_T *p)
927 {
928 	struct nfsstate *stp, *nstp;
929 
930 	LIST_FOREACH_SAFE(stp, &clp->lc_open, ls_list, nstp) {
931 		nfsrv_freeopenowner(stp, 1, p);
932 	}
933 }
934 
935 /*
936  * Free a client that has been cleaned. It should also already have been
937  * removed from the lists.
938  * (Just to be safe w.r.t. newnfs_disconnect(), call this function when
939  *  softclock interrupts are enabled.)
940  */
941 APPLESTATIC void
942 nfsrv_zapclient(struct nfsclient *clp, NFSPROC_T *p)
943 {
944 
945 #ifdef notyet
946 	if ((clp->lc_flags & (LCL_GSS | LCL_CALLBACKSON)) ==
947 	     (LCL_GSS | LCL_CALLBACKSON) &&
948 	    (clp->lc_hand.nfsh_flag & NFSG_COMPLETE) &&
949 	    clp->lc_handlelen > 0) {
950 		clp->lc_hand.nfsh_flag &= ~NFSG_COMPLETE;
951 		clp->lc_hand.nfsh_flag |= NFSG_DESTROYED;
952 		(void) nfsrv_docallback(clp, NFSV4PROC_CBNULL,
953 			NULL, 0, NULL, NULL, NULL, p);
954 	}
955 #endif
956 	newnfs_disconnect(&clp->lc_req);
957 	NFSSOCKADDRFREE(clp->lc_req.nr_nam);
958 	NFSFREEMUTEX(&clp->lc_req.nr_mtx);
959 	free((caddr_t)clp, M_NFSDCLIENT);
960 	NFSLOCKSTATE();
961 	newnfsstats.srvclients--;
962 	nfsrv_openpluslock--;
963 	nfsrv_clients--;
964 	NFSUNLOCKSTATE();
965 }
966 
967 /*
968  * Free a list of delegation state structures.
969  * (This function will also free all nfslockfile structures that no
970  *  longer have associated state.)
971  */
972 APPLESTATIC void
973 nfsrv_freedeleglist(struct nfsstatehead *sthp)
974 {
975 	struct nfsstate *stp, *nstp;
976 
977 	LIST_FOREACH_SAFE(stp, sthp, ls_list, nstp) {
978 		nfsrv_freedeleg(stp);
979 	}
980 	LIST_INIT(sthp);
981 }
982 
983 /*
984  * Free up a delegation.
985  */
986 static void
987 nfsrv_freedeleg(struct nfsstate *stp)
988 {
989 	struct nfslockfile *lfp;
990 
991 	LIST_REMOVE(stp, ls_hash);
992 	LIST_REMOVE(stp, ls_list);
993 	LIST_REMOVE(stp, ls_file);
994 	lfp = stp->ls_lfp;
995 	if (LIST_EMPTY(&lfp->lf_open) &&
996 	    LIST_EMPTY(&lfp->lf_lock) && LIST_EMPTY(&lfp->lf_deleg))
997 		nfsrv_freenfslockfile(lfp);
998 	FREE((caddr_t)stp, M_NFSDSTATE);
999 	newnfsstats.srvdelegates--;
1000 	nfsrv_openpluslock--;
1001 	nfsrv_delegatecnt--;
1002 }
1003 
1004 /*
1005  * This function frees an open owner and all associated opens.
1006  * Must be called with soft clock interrupts disabled.
1007  */
1008 static void
1009 nfsrv_freeopenowner(struct nfsstate *stp, int cansleep, NFSPROC_T *p)
1010 {
1011 	struct nfsstate *nstp, *tstp;
1012 
1013 	LIST_REMOVE(stp, ls_list);
1014 	/*
1015 	 * Now, free all associated opens.
1016 	 */
1017 	nstp = LIST_FIRST(&stp->ls_open);
1018 	while (nstp != LIST_END(&stp->ls_open)) {
1019 		tstp = nstp;
1020 		nstp = LIST_NEXT(nstp, ls_list);
1021 		(void) nfsrv_freeopen(tstp, NULL, cansleep, p);
1022 	}
1023 	if (stp->ls_op)
1024 		nfsrvd_derefcache(stp->ls_op);
1025 	FREE((caddr_t)stp, M_NFSDSTATE);
1026 	newnfsstats.srvopenowners--;
1027 	nfsrv_openpluslock--;
1028 }
1029 
1030 /*
1031  * This function frees an open (nfsstate open structure) with all associated
1032  * lock_owners and locks. It also frees the nfslockfile structure iff there
1033  * are no other opens on the file.
1034  * Must be called with soft clock interrupts disabled.
1035  * Returns 1 if it free'd the nfslockfile, 0 otherwise.
1036  */
1037 static int
1038 nfsrv_freeopen(struct nfsstate *stp, int *freedlockp, int cansleep,
1039     NFSPROC_T *p)
1040 {
1041 	struct nfsstate *nstp, *tstp;
1042 	struct nfslockfile *lfp;
1043 	int ret = 0, ret2;
1044 
1045 	LIST_REMOVE(stp, ls_hash);
1046 	LIST_REMOVE(stp, ls_list);
1047 	LIST_REMOVE(stp, ls_file);
1048 
1049 	lfp = stp->ls_lfp;
1050 	/*
1051 	 * The nfslockfile is freed here if there are no locks
1052 	 * associated with the open.
1053 	 * If there are locks associated with the open, the
1054 	 * nfslockfile structure can be freed via nfsrv_freelockowner().
1055 	 * (That is why the call must be here instead of after the loop.)
1056 	 */
1057 	if (LIST_EMPTY(&lfp->lf_open) && LIST_EMPTY(&lfp->lf_lock) &&
1058 	    LIST_EMPTY(&lfp->lf_deleg)) {
1059 		nfsrv_freenfslockfile(lfp);
1060 		ret = 1;
1061 	}
1062 	/*
1063 	 * Now, free all lockowners associated with this open.
1064 	 */
1065 	nstp = LIST_FIRST(&stp->ls_open);
1066 	while (nstp != LIST_END(&stp->ls_open)) {
1067 		tstp = nstp;
1068 		nstp = LIST_NEXT(nstp, ls_list);
1069 		ret2 = nfsrv_freelockowner(tstp, freedlockp, cansleep, p);
1070 		if (ret == 0 && ret2 != 0)
1071 			ret = ret2;
1072 	}
1073 	FREE((caddr_t)stp, M_NFSDSTATE);
1074 	newnfsstats.srvopens--;
1075 	nfsrv_openpluslock--;
1076 	return (ret);
1077 }
1078 
1079 /*
1080  * Frees a lockowner and all associated locks.
1081  * It also frees the nfslockfile structure, if there are no more
1082  * references to it.
1083  * Must be called with soft clock interrupts disabled.
1084  * Returns 1 if it free'd the nfslockfile structure, 1 otherwise.
1085  */
1086 static int
1087 nfsrv_freelockowner(struct nfsstate *stp, int *freedlockp, int cansleep,
1088     NFSPROC_T *p)
1089 {
1090 	int ret;
1091 
1092 	LIST_REMOVE(stp, ls_hash);
1093 	LIST_REMOVE(stp, ls_list);
1094 	ret = nfsrv_freeallnfslocks(stp, freedlockp, cansleep, p);
1095 	if (stp->ls_op)
1096 		nfsrvd_derefcache(stp->ls_op);
1097 	FREE((caddr_t)stp, M_NFSDSTATE);
1098 	newnfsstats.srvlockowners--;
1099 	nfsrv_openpluslock--;
1100 	return (ret);
1101 }
1102 
1103 /*
1104  * Free all the nfs locks on a lockowner.
1105  * Returns 1 if it free'd the nfslockfile structure, 0 otherwise.
1106  * If any byte range lock is free'd, *freedlockp is set to 1.
1107  */
1108 static int
1109 nfsrv_freeallnfslocks(struct nfsstate *stp, int *freedlockp, int cansleep,
1110     NFSPROC_T *p)
1111 {
1112 	struct nfslock *lop, *nlop;
1113 	struct nfslockfile *lfp = NULL, *olfp = NULL;
1114 	int ret = 0;
1115 
1116 	lop = LIST_FIRST(&stp->ls_lock);
1117 	while (lop != LIST_END(&stp->ls_lock)) {
1118 		nlop = LIST_NEXT(lop, lo_lckowner);
1119 		/*
1120 		 * Since locks off a lockowner are ordered by
1121 		 * file, you should update the local locks when
1122 		 * you hit the next file OR the end of the lock
1123 		 * list. If there are no locks for other owners,
1124 		 * it must be done before the lockowner is discarded.
1125 		 * (All this only applies if cansleep == 1.)
1126 		 */
1127 		olfp = lfp;
1128 		lfp = lop->lo_lfp;
1129 		nfsrv_freenfslock(lop);
1130 		if (freedlockp)
1131 			*freedlockp = 1;
1132 		if (LIST_EMPTY(&lfp->lf_open) && LIST_EMPTY(&lfp->lf_lock) &&
1133 		    LIST_EMPTY(&lfp->lf_deleg)) {
1134 			if (cansleep)
1135 				nfsrv_locallocks(NULL, lfp, p);
1136 			nfsrv_freenfslockfile(lfp);
1137 			/*
1138 			 * Set the pointer(s) to this lockowner NULL,
1139 			 * to indicate it has been free'd and local
1140 			 * locks discarded already.
1141 			 */
1142 			if (olfp == lfp)
1143 				olfp = NULL;
1144 			lfp = NULL;
1145 			ret = 1;
1146 		}
1147 		if (cansleep && olfp != lfp && olfp != NULL)
1148 			nfsrv_locallocks(NULL, olfp, p);
1149 		lop = nlop;
1150 	}
1151 	if (cansleep && lfp != NULL)
1152 		nfsrv_locallocks(NULL, olfp, p);
1153 	return (ret);
1154 }
1155 
1156 /*
1157  * Free an nfslock structure.
1158  * Must be called with soft clock interrupts disabled.
1159  */
1160 static void
1161 nfsrv_freenfslock(struct nfslock *lop)
1162 {
1163 
1164 	LIST_REMOVE(lop, lo_lckfile);
1165 	LIST_REMOVE(lop, lo_lckowner);
1166 	FREE((caddr_t)lop, M_NFSDLOCK);
1167 	newnfsstats.srvlocks--;
1168 	nfsrv_openpluslock--;
1169 }
1170 
1171 /*
1172  * This function frees an nfslockfile structure.
1173  * Must be called with soft clock interrupts disabled.
1174  */
1175 static void
1176 nfsrv_freenfslockfile(struct nfslockfile *lfp)
1177 {
1178 
1179 	LIST_REMOVE(lfp, lf_hash);
1180 	FREE((caddr_t)lfp, M_NFSDLOCKFILE);
1181 }
1182 
1183 /*
1184  * This function looks up an nfsstate structure via stateid.
1185  */
1186 static int
1187 nfsrv_getstate(struct nfsclient *clp, nfsv4stateid_t *stateidp, __unused u_int32_t flags,
1188     struct nfsstate **stpp)
1189 {
1190 	struct nfsstate *stp;
1191 	struct nfsstatehead *hp;
1192 
1193 	*stpp = NULL;
1194 	hp = NFSSTATEHASH(clp, *stateidp);
1195 	LIST_FOREACH(stp, hp, ls_hash) {
1196 		if (!NFSBCMP(stp->ls_stateid.other, stateidp->other,
1197 			NFSX_STATEIDOTHER))
1198 			break;
1199 	}
1200 
1201 	/*
1202 	 * If no state id in list, return NFSERR_BADSTATEID.
1203 	 */
1204 	if (stp == LIST_END(hp))
1205 		return (NFSERR_BADSTATEID);
1206 	*stpp = stp;
1207 	return (0);
1208 }
1209 
1210 /*
1211  * This function gets an nfsstate structure via owner string.
1212  */
1213 static void
1214 nfsrv_getowner(struct nfsstatehead *hp, struct nfsstate *new_stp,
1215     struct nfsstate **stpp)
1216 {
1217 	struct nfsstate *stp;
1218 
1219 	*stpp = NULL;
1220 	LIST_FOREACH(stp, hp, ls_list) {
1221 		if (new_stp->ls_ownerlen == stp->ls_ownerlen &&
1222 		  !NFSBCMP(new_stp->ls_owner,stp->ls_owner,stp->ls_ownerlen)) {
1223 			*stpp = stp;
1224 			return;
1225 		}
1226 	}
1227 }
1228 
1229 /*
1230  * Lock control function called to update lock status.
1231  * Returns 0 upon success, -1 if there is no lock and the flags indicate
1232  * that one isn't to be created and an NFSERR_xxx for other errors.
1233  * The structures new_stp and new_lop are passed in as pointers that should
1234  * be set to NULL if the structure is used and shouldn't be free'd.
1235  * For the NFSLCK_TEST and NFSLCK_CHECK cases, the structures are
1236  * never used and can safely be allocated on the stack. For all other
1237  * cases, *new_stpp and *new_lopp should be malloc'd before the call,
1238  * in case they are used.
1239  */
1240 APPLESTATIC int
1241 nfsrv_lockctrl(vnode_t vp, struct nfsstate **new_stpp,
1242     struct nfslock **new_lopp, struct nfslockconflict *cfp,
1243     nfsquad_t clientid, nfsv4stateid_t *stateidp, __unused struct nfsexstuff *exp,
1244     struct nfsrv_descript *nd, NFSPROC_T *p)
1245 {
1246 	struct nfslock *lop;
1247 	struct nfsstate *new_stp = *new_stpp;
1248 	struct nfslock *new_lop = *new_lopp;
1249 	struct nfsstate *tstp, *mystp, *nstp;
1250 	int specialid = 0;
1251 	struct nfslockfile *lfp;
1252 	struct nfslock *other_lop = NULL;
1253 	struct nfsstate *stp, *lckstp = NULL;
1254 	struct nfsclient *clp = NULL;
1255 	u_int32_t bits;
1256 	int error = 0, haslock = 0, ret;
1257 	int getlckret, delegation = 0;
1258 	fhandle_t nfh;
1259 
1260 	if (new_stp->ls_flags & (NFSLCK_CHECK | NFSLCK_SETATTR)) {
1261 		/*
1262 		 * Note the special cases of "all 1s" or "all 0s" stateids and
1263 		 * let reads with all 1s go ahead.
1264 		 */
1265 		if (new_stp->ls_stateid.seqid == 0x0 &&
1266 		    new_stp->ls_stateid.other[0] == 0x0 &&
1267 		    new_stp->ls_stateid.other[1] == 0x0 &&
1268 		    new_stp->ls_stateid.other[2] == 0x0)
1269 			specialid = 1;
1270 		else if (new_stp->ls_stateid.seqid == 0xffffffff &&
1271 		    new_stp->ls_stateid.other[0] == 0xffffffff &&
1272 		    new_stp->ls_stateid.other[1] == 0xffffffff &&
1273 		    new_stp->ls_stateid.other[2] == 0xffffffff)
1274 			specialid = 2;
1275 	}
1276 
1277 	/*
1278 	 * Check for restart conditions (client and server).
1279 	 */
1280 	error = nfsrv_checkrestart(clientid, new_stp->ls_flags,
1281 	    &new_stp->ls_stateid, specialid);
1282 	if (error)
1283 		return (error);
1284 
1285 	/*
1286 	 * Check for state resource limit exceeded.
1287 	 */
1288 	if ((new_stp->ls_flags & NFSLCK_LOCK) &&
1289 	    nfsrv_openpluslock > NFSRV_V4STATELIMIT)
1290 		return (NFSERR_RESOURCE);
1291 
1292 	/*
1293 	 * For Lock, check for a conflict with a lock held by
1294 	 * a process running locally on the server now, before
1295 	 * monkeying with nfsd state. Since the vp is locked, any
1296 	 * other local calls are blocked during this Op.
1297 	 */
1298 	if (new_stp->ls_flags & NFSLCK_LOCK) {
1299 		if (new_lop->lo_flags & NFSLCK_WRITE)
1300 			error = nfsvno_localconflict(vp, F_WRLCK,
1301 			    new_lop->lo_first, new_lop->lo_end, cfp, p);
1302 		else
1303 			error = nfsvno_localconflict(vp, F_RDLCK,
1304 			    new_lop->lo_first, new_lop->lo_end, cfp, p);
1305 		if (error)
1306 			return (error);
1307 	}
1308 
1309 	/*
1310 	 * For the lock case, get another nfslock structure,
1311 	 * just in case we need it.
1312 	 * Malloc now, before we start sifting through the linked lists,
1313 	 * in case we have to wait for memory.
1314 	 */
1315 tryagain:
1316 	if (new_stp->ls_flags & NFSLCK_LOCK)
1317 		MALLOC(other_lop, struct nfslock *, sizeof (struct nfslock),
1318 		    M_NFSDLOCK, M_WAITOK);
1319 
1320 	/*
1321 	 * Get the lockfile structure for CFH now, so we can do a sanity
1322 	 * check against the stateid, before incrementing the seqid#, since
1323 	 * we want to return NFSERR_BADSTATEID on failure and the seqid#
1324 	 * shouldn't be incremented for this case.
1325 	 * If nfsrv_getlockfile() returns -1, it means "not found", which
1326 	 * will be handled later.
1327 	 */
1328 	getlckret = nfsrv_getlockfh(vp, new_stp->ls_flags, NULL, &nfh, p);
1329 	NFSLOCKSTATE();
1330 	if (!getlckret)
1331 		getlckret = nfsrv_getlockfile(new_stp->ls_flags, NULL,
1332 		    &lfp, &nfh);
1333 	if (getlckret != 0 && getlckret != -1) {
1334 		NFSUNLOCKSTATE();
1335 		if (other_lop)
1336 			FREE((caddr_t)other_lop, M_NFSDLOCK);
1337 		if (haslock) {
1338 			NFSLOCKV4ROOTMUTEX();
1339 			nfsv4_unlock(&nfsv4rootfs_lock, 1);
1340 			NFSUNLOCKV4ROOTMUTEX();
1341 		}
1342 		return (getlckret);
1343 	}
1344 
1345 	/*
1346 	 * Since the code is manipulating lists that are also
1347 	 * manipulated by nfsrv_servertimer(), soft clock interrupts
1348 	 * must be masked off.
1349 	 */
1350 	if (specialid == 0) {
1351 	    if (new_stp->ls_flags & NFSLCK_TEST) {
1352 		/*
1353 		 * RFC 3530 does not list LockT as an op that renews a
1354 		 * lease, but the concensus seems to be that it is ok
1355 		 * for a server to do so.
1356 		 */
1357 		error = nfsrv_getclient(clientid, CLOPS_RENEW, &clp,
1358 		    (nfsquad_t)((u_quad_t)0), NULL, p);
1359 
1360 		/*
1361 		 * Since NFSERR_EXPIRED, NFSERR_ADMINREVOKED are not valid
1362 		 * error returns for LockT, just go ahead and test for a lock,
1363 		 * since there are no locks for this client, but other locks
1364 		 * can conflict. (ie. same client will always be false)
1365 		 */
1366 		if (error == NFSERR_EXPIRED || error == NFSERR_ADMINREVOKED)
1367 		    error = 0;
1368 		lckstp = new_stp;
1369 	    } else {
1370 	      error = nfsrv_getclient(clientid, CLOPS_RENEW, &clp,
1371 		(nfsquad_t)((u_quad_t)0), NULL, p);
1372 	      if (error == 0)
1373 		/*
1374 		 * Look up the stateid
1375 		 */
1376 		error = nfsrv_getstate(clp, &new_stp->ls_stateid,
1377 		  new_stp->ls_flags, &stp);
1378 	      /*
1379 	       * do some sanity checks for an unconfirmed open or a
1380 	       * stateid that refers to the wrong file, for an open stateid
1381 	       */
1382 	      if (error == 0 && (stp->ls_flags & NFSLCK_OPEN) &&
1383 		  ((stp->ls_openowner->ls_flags & NFSLCK_NEEDSCONFIRM) ||
1384 		   (getlckret != -1 && stp->ls_lfp != lfp)))
1385 			error = NFSERR_BADSTATEID;
1386 	      if (error == 0 &&
1387 		  (stp->ls_flags & (NFSLCK_DELEGREAD | NFSLCK_DELEGWRITE)) &&
1388 		  getlckret != -1 && stp->ls_lfp != lfp)
1389 			error = NFSERR_BADSTATEID;
1390 
1391 	      /*
1392 	       * If the lockowner stateid doesn't refer to the same file,
1393 	       * I believe that is considered ok, since some clients will
1394 	       * only create a single lockowner and use that for all locks
1395 	       * on all files.
1396 	       * For now, log it as a diagnostic, instead of considering it
1397 	       * a BadStateid.
1398 	       */
1399 	      if (error == 0 && (stp->ls_flags &
1400 		  (NFSLCK_OPEN | NFSLCK_DELEGREAD | NFSLCK_DELEGWRITE)) == 0 &&
1401 		  getlckret != -1 && stp->ls_lfp != lfp) {
1402 #ifdef DIAGNOSTIC
1403 		  printf("Got a lock statid for different file open\n");
1404 #endif
1405 		  /*
1406 		  error = NFSERR_BADSTATEID;
1407 		  */
1408 	      }
1409 
1410 	      if (error == 0) {
1411 		    if (new_stp->ls_flags & NFSLCK_OPENTOLOCK) {
1412 			/*
1413 			 * If haslock set, we've already checked the seqid.
1414 			 */
1415 			if (!haslock) {
1416 			    if (stp->ls_flags & NFSLCK_OPEN)
1417 				error = nfsrv_checkseqid(nd, new_stp->ls_seq,
1418 				    stp->ls_openowner, new_stp->ls_op);
1419 			    else
1420 				error = NFSERR_BADSTATEID;
1421 			}
1422 			if (!error)
1423 			    nfsrv_getowner(&stp->ls_open, new_stp, &lckstp);
1424 			if (lckstp)
1425 			    /*
1426 			     * I believe this should be an error, but it
1427 			     * isn't obvious what NFSERR_xxx would be
1428 			     * appropriate, so I'll use NFSERR_INVAL for now.
1429 			     */
1430 			    error = NFSERR_INVAL;
1431 			else
1432 			    lckstp = new_stp;
1433 		    } else if (new_stp->ls_flags&(NFSLCK_LOCK|NFSLCK_UNLOCK)) {
1434 			/*
1435 			 * If haslock set, ditto above.
1436 			 */
1437 			if (!haslock) {
1438 			    if (stp->ls_flags & NFSLCK_OPEN)
1439 				error = NFSERR_BADSTATEID;
1440 			    else
1441 				error = nfsrv_checkseqid(nd, new_stp->ls_seq,
1442 				    stp, new_stp->ls_op);
1443 			}
1444 			lckstp = stp;
1445 		    } else {
1446 			lckstp = stp;
1447 		    }
1448 	      }
1449 	      /*
1450 	       * If the seqid part of the stateid isn't the same, return
1451 	       * NFSERR_OLDSTATEID for cases other than I/O Ops.
1452 	       * For I/O Ops, only return NFSERR_OLDSTATEID if
1453 	       * nfsrv_returnoldstateid is set. (The concensus on the email
1454 	       * list was that most clients would prefer to not receive
1455 	       * NFSERR_OLDSTATEID for I/O Ops, but the RFC suggests that that
1456 	       * is what will happen, so I use the nfsrv_returnoldstateid to
1457 	       * allow for either server configuration.)
1458 	       */
1459 	      if (!error && stp->ls_stateid.seqid!=new_stp->ls_stateid.seqid &&
1460 		  (!(new_stp->ls_flags & NFSLCK_CHECK) ||
1461 		   nfsrv_returnoldstateid))
1462 		    error = NFSERR_OLDSTATEID;
1463 	    }
1464 	}
1465 
1466 	/*
1467 	 * Now we can check for grace.
1468 	 */
1469 	if (!error)
1470 		error = nfsrv_checkgrace(new_stp->ls_flags);
1471 	if ((new_stp->ls_flags & NFSLCK_RECLAIM) && !error &&
1472 		nfsrv_checkstable(clp))
1473 		error = NFSERR_NOGRACE;
1474 	/*
1475 	 * If we successfully Reclaimed state, note that.
1476 	 */
1477 	if ((new_stp->ls_flags & NFSLCK_RECLAIM) && !error)
1478 		nfsrv_markstable(clp);
1479 
1480 	/*
1481 	 * If nd_repstat is set, we can return that now, since the
1482 	 * seqid# has been incremented.
1483 	 */
1484 	if (nd->nd_repstat && !error)
1485 		error = nd->nd_repstat;
1486 	if (error) {
1487 		NFSUNLOCKSTATE();
1488 		if (other_lop)
1489 			FREE((caddr_t)other_lop, M_NFSDLOCK);
1490 		if (haslock) {
1491 			NFSLOCKV4ROOTMUTEX();
1492 			nfsv4_unlock(&nfsv4rootfs_lock, 1);
1493 			NFSUNLOCKV4ROOTMUTEX();
1494 		}
1495 		return (error);
1496 	}
1497 
1498 	/*
1499 	 * Check the nfsrv_getlockfile return.
1500 	 * Returned -1 if no structure found.
1501 	 */
1502 	if (getlckret == -1) {
1503 		error = NFSERR_EXPIRED;
1504 		/*
1505 		 * Called from lockt, so no lock is OK.
1506 		 */
1507 		if (new_stp->ls_flags & NFSLCK_TEST) {
1508 			error = 0;
1509 		} else if (new_stp->ls_flags &
1510 		    (NFSLCK_CHECK | NFSLCK_SETATTR)) {
1511 			/*
1512 			 * Called to check for a lock, OK if the stateid is all
1513 			 * 1s or all 0s, but there should be an nfsstate
1514 			 * otherwise.
1515 			 * (ie. If there is no open, I'll assume no share
1516 			 *  deny bits.)
1517 			 */
1518 			if (specialid)
1519 				error = 0;
1520 			else
1521 				error = NFSERR_BADSTATEID;
1522 		}
1523 		NFSUNLOCKSTATE();
1524 		if (haslock) {
1525 			NFSLOCKV4ROOTMUTEX();
1526 			nfsv4_unlock(&nfsv4rootfs_lock, 1);
1527 			NFSUNLOCKV4ROOTMUTEX();
1528 		}
1529 		/*
1530 		 * Called to lock or unlock, so the lock has gone away.
1531 		 */
1532 		return (error);
1533 	}
1534 
1535 	/*
1536 	 * For NFSLCK_CHECK and NFSLCK_LOCK, test for a share conflict.
1537 	 * For NFSLCK_CHECK, allow a read if write access is granted,
1538 	 * but check for a deny. For NFSLCK_LOCK, require correct access,
1539 	 * which implies a conflicting deny can't exist.
1540 	 */
1541 	if (new_stp->ls_flags & (NFSLCK_CHECK | NFSLCK_LOCK)) {
1542 	    /*
1543 	     * Four kinds of state id:
1544 	     * - specialid (all 0s or all 1s), only for NFSLCK_CHECK
1545 	     * - stateid for an open
1546 	     * - stateid for a delegation
1547 	     * - stateid for a lock owner
1548 	     */
1549 	    if (!specialid) {
1550 		if (stp->ls_flags & (NFSLCK_DELEGREAD | NFSLCK_DELEGWRITE)) {
1551 		    delegation = 1;
1552 		    mystp = stp;
1553 		    nfsrv_delaydelegtimeout(stp);
1554 	        } else if (stp->ls_flags & NFSLCK_OPEN) {
1555 		    mystp = stp;
1556 		} else {
1557 		    mystp = stp->ls_openstp;
1558 		}
1559 		/*
1560 		 * If locking or checking, require correct access
1561 		 * bit set.
1562 		 */
1563 		if (((new_stp->ls_flags & NFSLCK_LOCK) &&
1564 		     !((new_lop->lo_flags >> NFSLCK_LOCKSHIFT) &
1565 		       mystp->ls_flags & NFSLCK_ACCESSBITS)) ||
1566 		    ((new_stp->ls_flags & (NFSLCK_CHECK|NFSLCK_READACCESS)) ==
1567 		      (NFSLCK_CHECK | NFSLCK_READACCESS) &&
1568 		     !(mystp->ls_flags & NFSLCK_READACCESS)) ||
1569 		    ((new_stp->ls_flags & (NFSLCK_CHECK|NFSLCK_WRITEACCESS)) ==
1570 		      (NFSLCK_CHECK | NFSLCK_WRITEACCESS) &&
1571 		     !(mystp->ls_flags & NFSLCK_WRITEACCESS))) {
1572 			NFSUNLOCKSTATE();
1573 			if (other_lop)
1574 				FREE((caddr_t)other_lop, M_NFSDLOCK);
1575 			if (haslock) {
1576 				NFSLOCKV4ROOTMUTEX();
1577 				nfsv4_unlock(&nfsv4rootfs_lock, 1);
1578 				NFSUNLOCKV4ROOTMUTEX();
1579 			}
1580 			return (NFSERR_OPENMODE);
1581 		}
1582 	    } else
1583 		mystp = NULL;
1584 	    if ((new_stp->ls_flags & NFSLCK_CHECK) && !delegation) {
1585 		/*
1586 		 * Check for a conflicting deny bit.
1587 		 */
1588 		LIST_FOREACH(tstp, &lfp->lf_open, ls_file) {
1589 		    if (tstp != mystp) {
1590 			bits = tstp->ls_flags;
1591 			bits >>= NFSLCK_SHIFT;
1592 			if (new_stp->ls_flags & bits & NFSLCK_ACCESSBITS) {
1593 			    ret = nfsrv_clientconflict(tstp->ls_clp, &haslock,
1594 				vp, p);
1595 			    if (ret) {
1596 				/*
1597 				* nfsrv_clientconflict unlocks state
1598 				 * when it returns non-zero.
1599 				 */
1600 				lckstp = NULL;
1601 				goto tryagain;
1602 			    }
1603 			    NFSUNLOCKSTATE();
1604 			    if (haslock) {
1605 				NFSLOCKV4ROOTMUTEX();
1606 				nfsv4_unlock(&nfsv4rootfs_lock, 1);
1607 				NFSUNLOCKV4ROOTMUTEX();
1608 			    }
1609 			    return (NFSERR_OPENMODE);
1610 			}
1611 		    }
1612 		}
1613 
1614 		/* We're outta here */
1615 		NFSUNLOCKSTATE();
1616 		if (haslock) {
1617 			NFSLOCKV4ROOTMUTEX();
1618 			nfsv4_unlock(&nfsv4rootfs_lock, 1);
1619 			NFSUNLOCKV4ROOTMUTEX();
1620 		}
1621 		return (0);
1622 	    }
1623 	}
1624 
1625 	/*
1626 	 * For setattr, just get rid of all the Delegations for other clients.
1627 	 */
1628 	if (new_stp->ls_flags & NFSLCK_SETATTR) {
1629 		ret = nfsrv_cleandeleg(vp, lfp, clp, &haslock, p);
1630 		if (ret) {
1631 			/*
1632 			 * nfsrv_cleandeleg() unlocks state when it
1633 			 * returns non-zero.
1634 			 */
1635 			if (ret == -1) {
1636 				lckstp = NULL;
1637 				goto tryagain;
1638 			}
1639 			return (ret);
1640 		}
1641 		if (!(new_stp->ls_flags & NFSLCK_CHECK) ||
1642 		    (LIST_EMPTY(&lfp->lf_open) && LIST_EMPTY(&lfp->lf_lock) &&
1643 		     LIST_EMPTY(&lfp->lf_deleg))) {
1644 			NFSUNLOCKSTATE();
1645 			if (haslock) {
1646 				NFSLOCKV4ROOTMUTEX();
1647 				nfsv4_unlock(&nfsv4rootfs_lock, 1);
1648 				NFSUNLOCKV4ROOTMUTEX();
1649 			}
1650 			return (0);
1651 		}
1652 	}
1653 
1654 	/*
1655 	 * Check for a conflicting delegation. If one is found, call
1656 	 * nfsrv_delegconflict() to handle it. If the v4root lock hasn't
1657 	 * been set yet, it will get the lock. Otherwise, it will recall
1658 	 * the delegation. Then, we try try again...
1659 	 * I currently believe the conflict algorithm to be:
1660 	 * For Lock Ops (Lock/LockT/LockU)
1661 	 * - there is a conflict iff a different client has a write delegation
1662 	 * For Reading (Read Op)
1663 	 * - there is a conflict iff a different client has a write delegation
1664 	 *   (the specialids are always a different client)
1665 	 * For Writing (Write/Setattr of size)
1666 	 * - there is a conflict if a different client has any delegation
1667 	 * - there is a conflict if the same client has a read delegation
1668 	 *   (I don't understand why this isn't allowed, but that seems to be
1669 	 *    the current concensus?)
1670 	 */
1671 	tstp = LIST_FIRST(&lfp->lf_deleg);
1672 	while (tstp != LIST_END(&lfp->lf_deleg)) {
1673 	    nstp = LIST_NEXT(tstp, ls_file);
1674 	    if ((((new_stp->ls_flags&(NFSLCK_LOCK|NFSLCK_UNLOCK|NFSLCK_TEST))||
1675 		 ((new_stp->ls_flags & NFSLCK_CHECK) &&
1676 		  (new_lop->lo_flags & NFSLCK_READ))) &&
1677 		  clp != tstp->ls_clp &&
1678 		 (tstp->ls_flags & NFSLCK_DELEGWRITE)) ||
1679 		 ((new_stp->ls_flags & NFSLCK_CHECK) &&
1680 		   (new_lop->lo_flags & NFSLCK_WRITE) &&
1681 		  (clp != tstp->ls_clp ||
1682 		   (tstp->ls_flags & NFSLCK_DELEGREAD)))) {
1683 		ret = nfsrv_delegconflict(tstp, &haslock, p, vp);
1684 		if (ret) {
1685 		    /*
1686 		     * nfsrv_delegconflict unlocks state when it
1687 		     * returns non-zero.
1688 		     */
1689 		    if (other_lop) {
1690 			FREE((caddr_t)other_lop, M_NFSDLOCK);
1691 			other_lop = NULL;
1692 		    }
1693 		    if (ret == -1) {
1694 			lckstp = NULL;
1695 			goto tryagain;
1696 		    }
1697 		    return (ret);
1698 		}
1699 	    }
1700 	    tstp = nstp;
1701 	}
1702 
1703 	/*
1704 	 * Handle the unlock case by calling nfsrv_updatelock().
1705 	 * (Should I have done some access checking above for unlock? For now,
1706 	 *  just let it happen.)
1707 	 */
1708 	if (new_stp->ls_flags & NFSLCK_UNLOCK) {
1709 		nfsrv_updatelock(stp, new_lopp, &other_lop, lfp);
1710 		stateidp->seqid = ++(stp->ls_stateid.seqid);
1711 		stateidp->other[0] = stp->ls_stateid.other[0];
1712 		stateidp->other[1] = stp->ls_stateid.other[1];
1713 		stateidp->other[2] = stp->ls_stateid.other[2];
1714 		/*
1715 		 * For a non-empty flp->lf_lock list, I believe
1716 		 * nfsrv_locallocks() can safely traverse the list, including
1717 		 * sleeping, for two reasons:
1718 		 * 1 - The Lock/LockU/Close Ops all require a locked
1719 		 *     vnode for the file and we currently have that.
1720 		 * 2 - The only other thing that modifies a non-empty
1721 		 *     list is nfsrv_cleanclient() and it is always
1722 		 *     done with the exclusive nfsv4rootfs_lock held.
1723 		 *     Since this Op in progress holds either a shared or
1724 		 *     exclusive lock on nfsv4rootfs_lock, that can't
1725 		 *     happen now.
1726 		 * However, the structure pointed to by lfp can go
1727 		 * in many places for an empty list, so that is handled
1728 		 * by passing a NULL pointer to nfsrv_locallocks().
1729 		 * Do that check now, while we are still SMP safe.
1730 		 */
1731 		if (LIST_EMPTY(&lfp->lf_lock))
1732 			lfp = NULL;
1733 		NFSUNLOCKSTATE();
1734 		nfsrv_locallocks(vp, lfp, p);
1735 		if (haslock) {
1736 			NFSLOCKV4ROOTMUTEX();
1737 			nfsv4_unlock(&nfsv4rootfs_lock, 1);
1738 			NFSUNLOCKV4ROOTMUTEX();
1739 		}
1740 		return (0);
1741 	}
1742 
1743 	/*
1744 	 * Search for a conflicting lock. A lock conflicts if:
1745 	 * - the lock range overlaps and
1746 	 * - at least one lock is a write lock and
1747 	 * - it is not owned by the same lock owner
1748 	 */
1749 	if (!delegation) {
1750 	  LIST_FOREACH(lop, &lfp->lf_lock, lo_lckfile) {
1751 	    if (new_lop->lo_end > lop->lo_first &&
1752 		new_lop->lo_first < lop->lo_end &&
1753 		(new_lop->lo_flags == NFSLCK_WRITE ||
1754 		 lop->lo_flags == NFSLCK_WRITE) &&
1755 		lckstp != lop->lo_stp &&
1756 		(lckstp->ls_clp != lop->lo_stp->ls_clp ||
1757 		 lckstp->ls_ownerlen != lop->lo_stp->ls_ownerlen ||
1758 		 NFSBCMP(lckstp->ls_owner, lop->lo_stp->ls_owner,
1759 		    lckstp->ls_ownerlen))) {
1760 		if (other_lop) {
1761 		    FREE((caddr_t)other_lop, M_NFSDLOCK);
1762 		    other_lop = NULL;
1763 		}
1764 		ret = nfsrv_clientconflict(lop->lo_stp->ls_clp,&haslock,vp,p);
1765 		if (ret) {
1766 		    /*
1767 		     * nfsrv_clientconflict() unlocks state when it
1768 		     * returns non-zero.
1769 		     */
1770 		    lckstp = NULL;
1771 		    goto tryagain;
1772 		}
1773 		/*
1774 		 * Found a conflicting lock, so record the conflict and
1775 		 * return the error.
1776 		 */
1777 		if (cfp) {
1778 		    cfp->cl_clientid.lval[0]=lop->lo_stp->ls_stateid.other[0];
1779 		    cfp->cl_clientid.lval[1]=lop->lo_stp->ls_stateid.other[1];
1780 		    cfp->cl_first = lop->lo_first;
1781 		    cfp->cl_end = lop->lo_end;
1782 		    cfp->cl_flags = lop->lo_flags;
1783 		    cfp->cl_ownerlen = lop->lo_stp->ls_ownerlen;
1784 		    NFSBCOPY(lop->lo_stp->ls_owner, cfp->cl_owner,
1785 			cfp->cl_ownerlen);
1786 		}
1787 		if (new_stp->ls_flags & NFSLCK_RECLAIM)
1788 		    error = NFSERR_RECLAIMCONFLICT;
1789 		else if (new_stp->ls_flags & NFSLCK_CHECK)
1790 		    error = NFSERR_LOCKED;
1791 		else
1792 		    error = NFSERR_DENIED;
1793 		NFSUNLOCKSTATE();
1794 		if (haslock) {
1795 			NFSLOCKV4ROOTMUTEX();
1796 			nfsv4_unlock(&nfsv4rootfs_lock, 1);
1797 			NFSUNLOCKV4ROOTMUTEX();
1798 		}
1799 		return (error);
1800 	    }
1801 	  }
1802 	}
1803 
1804 	/*
1805 	 * We only get here if there was no lock that conflicted.
1806 	 */
1807 	if (new_stp->ls_flags & (NFSLCK_TEST | NFSLCK_CHECK)) {
1808 		NFSUNLOCKSTATE();
1809 		if (haslock) {
1810 			NFSLOCKV4ROOTMUTEX();
1811 			nfsv4_unlock(&nfsv4rootfs_lock, 1);
1812 			NFSUNLOCKV4ROOTMUTEX();
1813 		}
1814 		return (0);
1815 	}
1816 
1817 	/*
1818 	 * We only get here when we are creating or modifying a lock.
1819 	 * There are two variants:
1820 	 * - exist_lock_owner where lock_owner exists
1821 	 * - open_to_lock_owner with new lock_owner
1822 	 */
1823 	if (!(new_stp->ls_flags & NFSLCK_OPENTOLOCK)) {
1824 		nfsrv_updatelock(lckstp, new_lopp, &other_lop, lfp);
1825 		stateidp->seqid = ++(lckstp->ls_stateid.seqid);
1826 		stateidp->other[0] = lckstp->ls_stateid.other[0];
1827 		stateidp->other[1] = lckstp->ls_stateid.other[1];
1828 		stateidp->other[2] = lckstp->ls_stateid.other[2];
1829 	} else {
1830 		/*
1831 		 * The new open_to_lock_owner case.
1832 		 * Link the new nfsstate into the lists.
1833 		 */
1834 		new_stp->ls_seq = new_stp->ls_opentolockseq;
1835 		nfsrvd_refcache(new_stp->ls_op);
1836 		stateidp->seqid = new_stp->ls_stateid.seqid = 0;
1837 		stateidp->other[0] = new_stp->ls_stateid.other[0] =
1838 		    clp->lc_clientid.lval[0];
1839 		stateidp->other[1] = new_stp->ls_stateid.other[1] =
1840 		    clp->lc_clientid.lval[1];
1841 		stateidp->other[2] = new_stp->ls_stateid.other[2] =
1842 		    nfsrv_nextstateindex(clp);
1843 		new_stp->ls_clp = clp;
1844 		LIST_INIT(&new_stp->ls_lock);
1845 		new_stp->ls_openstp = stp;
1846 		new_stp->ls_lfp = lfp;
1847 		nfsrv_insertlock(new_lop, (struct nfslock *)new_stp, new_stp,
1848 		    lfp);
1849 		LIST_INSERT_HEAD(NFSSTATEHASH(clp, new_stp->ls_stateid),
1850 		    new_stp, ls_hash);
1851 		LIST_INSERT_HEAD(&stp->ls_open, new_stp, ls_list);
1852 		*new_lopp = NULL;
1853 		*new_stpp = NULL;
1854 		newnfsstats.srvlockowners++;
1855 		nfsrv_openpluslock++;
1856 	}
1857 	/* See comment above, w.r.t. nfsrv_locallocks(). */
1858 	if (LIST_EMPTY(&lfp->lf_lock))
1859 		lfp = NULL;
1860 	NFSUNLOCKSTATE();
1861 	nfsrv_locallocks(vp, lfp, p);
1862 	if (haslock) {
1863 		NFSLOCKV4ROOTMUTEX();
1864 		nfsv4_unlock(&nfsv4rootfs_lock, 1);
1865 		NFSUNLOCKV4ROOTMUTEX();
1866 	}
1867 	if (other_lop)
1868 		FREE((caddr_t)other_lop, M_NFSDLOCK);
1869 	return (0);
1870 }
1871 
1872 /*
1873  * Check for state errors for Open.
1874  * repstat is passed back out as an error if more critical errors
1875  * are not detected.
1876  */
1877 APPLESTATIC int
1878 nfsrv_opencheck(nfsquad_t clientid, nfsv4stateid_t *stateidp,
1879     struct nfsstate *new_stp, vnode_t vp, struct nfsrv_descript *nd,
1880     NFSPROC_T *p, int repstat)
1881 {
1882 	struct nfsstate *stp, *nstp;
1883 	struct nfsclient *clp;
1884 	struct nfsstate *ownerstp;
1885 	struct nfslockfile *lfp, *new_lfp;
1886 	int error, haslock = 0, ret, readonly = 0, getfhret = 0;
1887 
1888 	if ((new_stp->ls_flags & NFSLCK_SHAREBITS) == NFSLCK_READACCESS)
1889 		readonly = 1;
1890 	/*
1891 	 * Check for restart conditions (client and server).
1892 	 */
1893 	error = nfsrv_checkrestart(clientid, new_stp->ls_flags,
1894 		&new_stp->ls_stateid, 0);
1895 	if (error)
1896 		return (error);
1897 
1898 	/*
1899 	 * Check for state resource limit exceeded.
1900 	 * Technically this should be SMP protected, but the worst
1901 	 * case error is "out by one or two" on the count when it
1902 	 * returns NFSERR_RESOURCE and the limit is just a rather
1903 	 * arbitrary high water mark, so no harm is done.
1904 	 */
1905 	if (nfsrv_openpluslock > NFSRV_V4STATELIMIT)
1906 		return (NFSERR_RESOURCE);
1907 
1908 tryagain:
1909 	MALLOC(new_lfp, struct nfslockfile *, sizeof (struct nfslockfile),
1910 	    M_NFSDLOCKFILE, M_WAITOK);
1911 	if (vp)
1912 		getfhret = nfsrv_getlockfh(vp, new_stp->ls_flags, &new_lfp,
1913 		    NULL, p);
1914 	NFSLOCKSTATE();
1915 	/*
1916 	 * Get the nfsclient structure.
1917 	 * Since the code is manipulating lists that are also
1918 	 * manipulated by nfsrv_servertimer(), soft clock interrupts
1919 	 * must be masked off.
1920 	 */
1921 	error = nfsrv_getclient(clientid, CLOPS_RENEW, &clp,
1922 	    (nfsquad_t)((u_quad_t)0), NULL, p);
1923 
1924 	/*
1925 	 * Look up the open owner. See if it needs confirmation and
1926 	 * check the seq#, as required.
1927 	 */
1928 	if (!error)
1929 		nfsrv_getowner(&clp->lc_open, new_stp, &ownerstp);
1930 
1931 	if (!error && ownerstp) {
1932 		error = nfsrv_checkseqid(nd, new_stp->ls_seq, ownerstp,
1933 		    new_stp->ls_op);
1934 		/*
1935 		 * If the OpenOwner hasn't been confirmed, assume the
1936 		 * old one was a replay and this one is ok.
1937 		 * See: RFC3530 Sec. 14.2.18.
1938 		 */
1939 		if (error == NFSERR_BADSEQID &&
1940 		    (ownerstp->ls_flags & NFSLCK_NEEDSCONFIRM))
1941 			error = 0;
1942 	}
1943 
1944 	/*
1945 	 * Check for grace.
1946 	 */
1947 	if (!error)
1948 		error = nfsrv_checkgrace(new_stp->ls_flags);
1949 	if ((new_stp->ls_flags & NFSLCK_RECLAIM) && !error &&
1950 		nfsrv_checkstable(clp))
1951 		error = NFSERR_NOGRACE;
1952 
1953 	/*
1954 	 * If none of the above errors occurred, let repstat be
1955 	 * returned.
1956 	 */
1957 	if (repstat && !error)
1958 		error = repstat;
1959 	if (error) {
1960 		NFSUNLOCKSTATE();
1961 		if (haslock) {
1962 			NFSLOCKV4ROOTMUTEX();
1963 			nfsv4_unlock(&nfsv4rootfs_lock, 1);
1964 			NFSUNLOCKV4ROOTMUTEX();
1965 		}
1966 		free((caddr_t)new_lfp, M_NFSDLOCKFILE);
1967 		return (error);
1968 	}
1969 
1970 	/*
1971 	 * If vp == NULL, the file doesn't exist yet, so return ok.
1972 	 * (This always happens on the first pass, so haslock must be 0.)
1973 	 */
1974 	if (vp == NULL) {
1975 		NFSUNLOCKSTATE();
1976 		FREE((caddr_t)new_lfp, M_NFSDLOCKFILE);
1977 		return (0);
1978 	}
1979 
1980 	/*
1981 	 * Get the structure for the underlying file.
1982 	 */
1983 	if (getfhret)
1984 		error = getfhret;
1985 	else
1986 		error = nfsrv_getlockfile(new_stp->ls_flags, &new_lfp, &lfp,
1987 		    NULL);
1988 	if (new_lfp)
1989 		FREE((caddr_t)new_lfp, M_NFSDLOCKFILE);
1990 	if (error) {
1991 		NFSUNLOCKSTATE();
1992 		if (haslock) {
1993 			NFSLOCKV4ROOTMUTEX();
1994 			nfsv4_unlock(&nfsv4rootfs_lock, 1);
1995 			NFSUNLOCKV4ROOTMUTEX();
1996 		}
1997 		return (error);
1998 	}
1999 
2000 	/*
2001 	 * Search for a conflicting open/share.
2002 	 */
2003 	if (new_stp->ls_flags & NFSLCK_DELEGCUR) {
2004 	    /*
2005 	     * For Delegate_Cur, search for the matching Delegation,
2006 	     * which indicates no conflict.
2007 	     * An old delegation should have been recovered by the
2008 	     * client doing a Claim_DELEGATE_Prev, so I won't let
2009 	     * it match and return NFSERR_EXPIRED. Should I let it
2010 	     * match?
2011 	     */
2012 	    LIST_FOREACH(stp, &lfp->lf_deleg, ls_file) {
2013 		if (!(stp->ls_flags & NFSLCK_OLDDELEG) &&
2014 		    stateidp->seqid == stp->ls_stateid.seqid &&
2015 		    !NFSBCMP(stateidp->other, stp->ls_stateid.other,
2016 			  NFSX_STATEIDOTHER))
2017 			break;
2018 	    }
2019 	    if (stp == LIST_END(&lfp->lf_deleg) ||
2020 		((new_stp->ls_flags & NFSLCK_WRITEACCESS) &&
2021 		 (stp->ls_flags & NFSLCK_DELEGREAD))) {
2022 		NFSUNLOCKSTATE();
2023 		if (haslock) {
2024 			NFSLOCKV4ROOTMUTEX();
2025 			nfsv4_unlock(&nfsv4rootfs_lock, 1);
2026 			NFSUNLOCKV4ROOTMUTEX();
2027 		}
2028 		return (NFSERR_EXPIRED);
2029 	    }
2030 	}
2031 
2032 	/*
2033 	 * Check for access/deny bit conflicts. I check for the same
2034 	 * owner as well, in case the client didn't bother.
2035 	 */
2036 	LIST_FOREACH(stp, &lfp->lf_open, ls_file) {
2037 		if (!(new_stp->ls_flags & NFSLCK_DELEGCUR) &&
2038 		    (((new_stp->ls_flags & NFSLCK_ACCESSBITS) &
2039 		      ((stp->ls_flags>>NFSLCK_SHIFT) & NFSLCK_ACCESSBITS))||
2040 		     ((stp->ls_flags & NFSLCK_ACCESSBITS) &
2041 		      ((new_stp->ls_flags>>NFSLCK_SHIFT)&NFSLCK_ACCESSBITS)))){
2042 			ret = nfsrv_clientconflict(stp->ls_clp,&haslock,vp,p);
2043 			if (ret) {
2044 				/*
2045 				 * nfsrv_clientconflict() unlocks
2046 				 * state when it returns non-zero.
2047 				 */
2048 				goto tryagain;
2049 			}
2050 			if (new_stp->ls_flags & NFSLCK_RECLAIM)
2051 				error = NFSERR_RECLAIMCONFLICT;
2052 			else
2053 				error = NFSERR_SHAREDENIED;
2054 			NFSUNLOCKSTATE();
2055 			if (haslock) {
2056 				NFSLOCKV4ROOTMUTEX();
2057 				nfsv4_unlock(&nfsv4rootfs_lock, 1);
2058 				NFSUNLOCKV4ROOTMUTEX();
2059 			}
2060 			return (error);
2061 		}
2062 	}
2063 
2064 	/*
2065 	 * Check for a conflicting delegation. If one is found, call
2066 	 * nfsrv_delegconflict() to handle it. If the v4root lock hasn't
2067 	 * been set yet, it will get the lock. Otherwise, it will recall
2068 	 * the delegation. Then, we try try again...
2069 	 * (If NFSLCK_DELEGCUR is set, it has a delegation, so there
2070 	 *  isn't a conflict.)
2071 	 * I currently believe the conflict algorithm to be:
2072 	 * For Open with Read Access and Deny None
2073 	 * - there is a conflict iff a different client has a write delegation
2074 	 * For Open with other Write Access or any Deny except None
2075 	 * - there is a conflict if a different client has any delegation
2076 	 * - there is a conflict if the same client has a read delegation
2077 	 *   (The current concensus is that this last case should be
2078 	 *    considered a conflict since the client with a read delegation
2079 	 *    could have done an Open with ReadAccess and WriteDeny
2080 	 *    locally and then not have checked for the WriteDeny.)
2081 	 * Don't check for a Reclaim, since that will be dealt with
2082 	 * by nfsrv_openctrl().
2083 	 */
2084 	if (!(new_stp->ls_flags &
2085 		(NFSLCK_DELEGPREV | NFSLCK_DELEGCUR | NFSLCK_RECLAIM))) {
2086 	    stp = LIST_FIRST(&lfp->lf_deleg);
2087 	    while (stp != LIST_END(&lfp->lf_deleg)) {
2088 		nstp = LIST_NEXT(stp, ls_file);
2089 		if ((readonly && stp->ls_clp != clp &&
2090 		       (stp->ls_flags & NFSLCK_DELEGWRITE)) ||
2091 		    (!readonly && (stp->ls_clp != clp ||
2092 		         (stp->ls_flags & NFSLCK_DELEGREAD)))) {
2093 			ret = nfsrv_delegconflict(stp, &haslock, p, vp);
2094 			if (ret) {
2095 			    /*
2096 			     * nfsrv_delegconflict() unlocks state
2097 			     * when it returns non-zero.
2098 			     */
2099 			    if (ret == -1)
2100 				goto tryagain;
2101 			    return (ret);
2102 			}
2103 		}
2104 		stp = nstp;
2105 	    }
2106 	}
2107 	NFSUNLOCKSTATE();
2108 	if (haslock) {
2109 		NFSLOCKV4ROOTMUTEX();
2110 		nfsv4_unlock(&nfsv4rootfs_lock, 1);
2111 		NFSUNLOCKV4ROOTMUTEX();
2112 	}
2113 	return (0);
2114 }
2115 
2116 /*
2117  * Open control function to create/update open state for an open.
2118  */
2119 APPLESTATIC int
2120 nfsrv_openctrl(struct nfsrv_descript *nd, vnode_t vp,
2121     struct nfsstate **new_stpp, nfsquad_t clientid, nfsv4stateid_t *stateidp,
2122     nfsv4stateid_t *delegstateidp, u_int32_t *rflagsp, struct nfsexstuff *exp,
2123     NFSPROC_T *p, u_quad_t filerev)
2124 {
2125 	struct nfsstate *new_stp = *new_stpp;
2126 	struct nfsstate *stp, *nstp;
2127 	struct nfsstate *openstp = NULL, *new_open, *ownerstp, *new_deleg;
2128 	struct nfslockfile *lfp, *new_lfp;
2129 	struct nfsclient *clp;
2130 	int error, haslock = 0, ret, delegate = 1, writedeleg = 1;
2131 	int readonly = 0, cbret = 1, getfhret = 0;
2132 
2133 	if ((new_stp->ls_flags & NFSLCK_SHAREBITS) == NFSLCK_READACCESS)
2134 		readonly = 1;
2135 	/*
2136 	 * Check for restart conditions (client and server).
2137 	 * (Paranoia, should have been detected by nfsrv_opencheck().)
2138 	 * If an error does show up, return NFSERR_EXPIRED, since the
2139 	 * the seqid# has already been incremented.
2140 	 */
2141 	error = nfsrv_checkrestart(clientid, new_stp->ls_flags,
2142 	    &new_stp->ls_stateid, 0);
2143 	if (error) {
2144 		printf("Nfsd: openctrl unexpected restart err=%d\n",
2145 		    error);
2146 		return (NFSERR_EXPIRED);
2147 	}
2148 
2149 tryagain:
2150 	MALLOC(new_lfp, struct nfslockfile *, sizeof (struct nfslockfile),
2151 	    M_NFSDLOCKFILE, M_WAITOK);
2152 	MALLOC(new_open, struct nfsstate *, sizeof (struct nfsstate),
2153 	    M_NFSDSTATE, M_WAITOK);
2154 	MALLOC(new_deleg, struct nfsstate *, sizeof (struct nfsstate),
2155 	    M_NFSDSTATE, M_WAITOK);
2156 	getfhret = nfsrv_getlockfh(vp, new_stp->ls_flags, &new_lfp,
2157 	    NULL, p);
2158 	NFSLOCKSTATE();
2159 	/*
2160 	 * Get the client structure. Since the linked lists could be changed
2161 	 * by other nfsd processes if this process does a tsleep(), one of
2162 	 * two things must be done.
2163 	 * 1 - don't tsleep()
2164 	 * or
2165 	 * 2 - get the nfsv4_lock() { indicated by haslock == 1 }
2166 	 *     before using the lists, since this lock stops the other
2167 	 *     nfsd. This should only be used for rare cases, since it
2168 	 *     essentially single threads the nfsd.
2169 	 *     At this time, it is only done for cases where the stable
2170 	 *     storage file must be written prior to completion of state
2171 	 *     expiration.
2172 	 */
2173 	error = nfsrv_getclient(clientid, CLOPS_RENEW, &clp,
2174 	    (nfsquad_t)((u_quad_t)0), NULL, p);
2175 	if (!error && (clp->lc_flags & LCL_NEEDSCBNULL) &&
2176 	    clp->lc_program) {
2177 		/*
2178 		 * This happens on the first open for a client
2179 		 * that supports callbacks.
2180 		 */
2181 		NFSUNLOCKSTATE();
2182 		/*
2183 		 * Although nfsrv_docallback() will sleep, clp won't
2184 		 * go away, since they are only removed when the
2185 		 * nfsv4_lock() has blocked the nfsd threads. The
2186 		 * fields in clp can change, but having multiple
2187 		 * threads do this Null callback RPC should be
2188 		 * harmless.
2189 		 */
2190 		cbret = nfsrv_docallback(clp, NFSV4PROC_CBNULL,
2191 		    NULL, 0, NULL, NULL, NULL, p);
2192 		NFSLOCKSTATE();
2193 		clp->lc_flags &= ~LCL_NEEDSCBNULL;
2194 		if (!cbret)
2195 			clp->lc_flags |= LCL_CALLBACKSON;
2196 	}
2197 
2198 	/*
2199 	 * Look up the open owner. See if it needs confirmation and
2200 	 * check the seq#, as required.
2201 	 */
2202 	if (!error)
2203 		nfsrv_getowner(&clp->lc_open, new_stp, &ownerstp);
2204 
2205 	if (error) {
2206 		NFSUNLOCKSTATE();
2207 		printf("Nfsd: openctrl unexpected state err=%d\n",
2208 			error);
2209 		free((caddr_t)new_lfp, M_NFSDLOCKFILE);
2210 		free((caddr_t)new_open, M_NFSDSTATE);
2211 		free((caddr_t)new_deleg, M_NFSDSTATE);
2212 		if (haslock) {
2213 			NFSLOCKV4ROOTMUTEX();
2214 			nfsv4_unlock(&nfsv4rootfs_lock, 1);
2215 			NFSUNLOCKV4ROOTMUTEX();
2216 		}
2217 		return (NFSERR_EXPIRED);
2218 	}
2219 
2220 	if (new_stp->ls_flags & NFSLCK_RECLAIM)
2221 		nfsrv_markstable(clp);
2222 
2223 	/*
2224 	 * Get the structure for the underlying file.
2225 	 */
2226 	if (getfhret)
2227 		error = getfhret;
2228 	else
2229 		error = nfsrv_getlockfile(new_stp->ls_flags, &new_lfp, &lfp,
2230 		    NULL);
2231 	if (new_lfp)
2232 		FREE((caddr_t)new_lfp, M_NFSDLOCKFILE);
2233 	if (error) {
2234 		NFSUNLOCKSTATE();
2235 		printf("Nfsd openctrl unexpected getlockfile err=%d\n",
2236 		    error);
2237 		free((caddr_t)new_open, M_NFSDSTATE);
2238 		free((caddr_t)new_deleg, M_NFSDSTATE);
2239 		if (haslock) {
2240 			NFSLOCKV4ROOTMUTEX();
2241 			nfsv4_unlock(&nfsv4rootfs_lock, 1);
2242 			NFSUNLOCKV4ROOTMUTEX();
2243 		}
2244 		return (error);
2245 	}
2246 
2247 	/*
2248 	 * Search for a conflicting open/share.
2249 	 */
2250 	if (new_stp->ls_flags & NFSLCK_DELEGCUR) {
2251 	    /*
2252 	     * For Delegate_Cur, search for the matching Delegation,
2253 	     * which indicates no conflict.
2254 	     * An old delegation should have been recovered by the
2255 	     * client doing a Claim_DELEGATE_Prev, so I won't let
2256 	     * it match and return NFSERR_EXPIRED. Should I let it
2257 	     * match?
2258 	     */
2259 	    LIST_FOREACH(stp, &lfp->lf_deleg, ls_file) {
2260 		if (!(stp->ls_flags & NFSLCK_OLDDELEG) &&
2261 		    stateidp->seqid == stp->ls_stateid.seqid &&
2262 		    !NFSBCMP(stateidp->other, stp->ls_stateid.other,
2263 			NFSX_STATEIDOTHER))
2264 			break;
2265 	    }
2266 	    if (stp == LIST_END(&lfp->lf_deleg) ||
2267 		((new_stp->ls_flags & NFSLCK_WRITEACCESS) &&
2268 		 (stp->ls_flags & NFSLCK_DELEGREAD))) {
2269 		NFSUNLOCKSTATE();
2270 		printf("Nfsd openctrl unexpected expiry\n");
2271 		free((caddr_t)new_open, M_NFSDSTATE);
2272 		free((caddr_t)new_deleg, M_NFSDSTATE);
2273 		if (haslock) {
2274 			NFSLOCKV4ROOTMUTEX();
2275 			nfsv4_unlock(&nfsv4rootfs_lock, 1);
2276 			NFSUNLOCKV4ROOTMUTEX();
2277 		}
2278 		return (NFSERR_EXPIRED);
2279 	    }
2280 
2281 	    /*
2282 	     * Don't issue a Delegation, since one already exists and
2283 	     * delay delegation timeout, as required.
2284 	     */
2285 	    delegate = 0;
2286 	    nfsrv_delaydelegtimeout(stp);
2287 	}
2288 
2289 	/*
2290 	 * Check for access/deny bit conflicts. I also check for the
2291 	 * same owner, since the client might not have bothered to check.
2292 	 * Also, note an open for the same file and owner, if found,
2293 	 * which is all we do here for Delegate_Cur, since conflict
2294 	 * checking is already done.
2295 	 */
2296 	LIST_FOREACH(stp, &lfp->lf_open, ls_file) {
2297 		if (ownerstp && stp->ls_openowner == ownerstp)
2298 			openstp = stp;
2299 		if (!(new_stp->ls_flags & NFSLCK_DELEGCUR)) {
2300 		    /*
2301 		     * If another client has the file open, the only
2302 		     * delegation that can be issued is a Read delegation
2303 		     * and only if it is a Read open with Deny none.
2304 		     */
2305 		    if (clp != stp->ls_clp) {
2306 			if ((stp->ls_flags & NFSLCK_SHAREBITS) ==
2307 			    NFSLCK_READACCESS)
2308 			    writedeleg = 0;
2309 			else
2310 			    delegate = 0;
2311 		    }
2312 		    if(((new_stp->ls_flags & NFSLCK_ACCESSBITS) &
2313 		        ((stp->ls_flags>>NFSLCK_SHIFT) & NFSLCK_ACCESSBITS))||
2314 		       ((stp->ls_flags & NFSLCK_ACCESSBITS) &
2315 		        ((new_stp->ls_flags>>NFSLCK_SHIFT)&NFSLCK_ACCESSBITS))){
2316 			ret = nfsrv_clientconflict(stp->ls_clp,&haslock,vp,p);
2317 			if (ret) {
2318 				/*
2319 				 * nfsrv_clientconflict() unlocks state
2320 				 * when it returns non-zero.
2321 				 */
2322 				free((caddr_t)new_open, M_NFSDSTATE);
2323 				free((caddr_t)new_deleg, M_NFSDSTATE);
2324 				openstp = NULL;
2325 				goto tryagain;
2326 			}
2327 			if (new_stp->ls_flags & NFSLCK_RECLAIM)
2328 				error = NFSERR_RECLAIMCONFLICT;
2329 			else
2330 				error = NFSERR_SHAREDENIED;
2331 			NFSUNLOCKSTATE();
2332 			if (haslock) {
2333 				NFSLOCKV4ROOTMUTEX();
2334 				nfsv4_unlock(&nfsv4rootfs_lock, 1);
2335 				NFSUNLOCKV4ROOTMUTEX();
2336 			}
2337 			free((caddr_t)new_open, M_NFSDSTATE);
2338 			free((caddr_t)new_deleg, M_NFSDSTATE);
2339 			printf("nfsd openctrl unexpected client cnfl\n");
2340 			return (error);
2341 		    }
2342 		}
2343 	}
2344 
2345 	/*
2346 	 * Check for a conflicting delegation. If one is found, call
2347 	 * nfsrv_delegconflict() to handle it. If the v4root lock hasn't
2348 	 * been set yet, it will get the lock. Otherwise, it will recall
2349 	 * the delegation. Then, we try try again...
2350 	 * (If NFSLCK_DELEGCUR is set, it has a delegation, so there
2351 	 *  isn't a conflict.)
2352 	 * I currently believe the conflict algorithm to be:
2353 	 * For Open with Read Access and Deny None
2354 	 * - there is a conflict iff a different client has a write delegation
2355 	 * For Open with other Write Access or any Deny except None
2356 	 * - there is a conflict if a different client has any delegation
2357 	 * - there is a conflict if the same client has a read delegation
2358 	 *   (The current concensus is that this last case should be
2359 	 *    considered a conflict since the client with a read delegation
2360 	 *    could have done an Open with ReadAccess and WriteDeny
2361 	 *    locally and then not have checked for the WriteDeny.)
2362 	 */
2363 	if (!(new_stp->ls_flags & (NFSLCK_DELEGPREV | NFSLCK_DELEGCUR))) {
2364 	    stp = LIST_FIRST(&lfp->lf_deleg);
2365 	    while (stp != LIST_END(&lfp->lf_deleg)) {
2366 		nstp = LIST_NEXT(stp, ls_file);
2367 		if (stp->ls_clp != clp && (stp->ls_flags & NFSLCK_DELEGREAD))
2368 			writedeleg = 0;
2369 		else
2370 			delegate = 0;
2371 		if ((readonly && stp->ls_clp != clp &&
2372 		       (stp->ls_flags & NFSLCK_DELEGWRITE)) ||
2373 		    (!readonly && (stp->ls_clp != clp ||
2374 		         (stp->ls_flags & NFSLCK_DELEGREAD)))) {
2375 		    if (new_stp->ls_flags & NFSLCK_RECLAIM) {
2376 			delegate = 2;
2377 		    } else {
2378 			ret = nfsrv_delegconflict(stp, &haslock, p, vp);
2379 			if (ret) {
2380 			    /*
2381 			     * nfsrv_delegconflict() unlocks state
2382 			     * when it returns non-zero.
2383 			     */
2384 			    printf("Nfsd openctrl unexpected deleg cnfl\n");
2385 			    free((caddr_t)new_open, M_NFSDSTATE);
2386 			    free((caddr_t)new_deleg, M_NFSDSTATE);
2387 			    if (ret == -1) {
2388 				openstp = NULL;
2389 				goto tryagain;
2390 			    }
2391 			    return (ret);
2392 			}
2393 		    }
2394 		}
2395 		stp = nstp;
2396 	    }
2397 	}
2398 
2399 	/*
2400 	 * We only get here if there was no open that conflicted.
2401 	 * If an open for the owner exists, or in the access/deny bits.
2402 	 * Otherwise it is a new open. If the open_owner hasn't been
2403 	 * confirmed, replace the open with the new one needing confirmation,
2404 	 * otherwise add the open.
2405 	 */
2406 	if (new_stp->ls_flags & NFSLCK_DELEGPREV) {
2407 	    /*
2408 	     * Handle NFSLCK_DELEGPREV by searching the old delegations for
2409 	     * a match. If found, just move the old delegation to the current
2410 	     * delegation list and issue open. If not found, return
2411 	     * NFSERR_EXPIRED.
2412 	     */
2413 	    LIST_FOREACH(stp, &clp->lc_olddeleg, ls_list) {
2414 		if (stp->ls_lfp == lfp) {
2415 		    /* Found it */
2416 		    if (stp->ls_clp != clp)
2417 			panic("olddeleg clp");
2418 		    LIST_REMOVE(stp, ls_list);
2419 		    LIST_REMOVE(stp, ls_hash);
2420 		    stp->ls_flags &= ~NFSLCK_OLDDELEG;
2421 		    stp->ls_stateid.seqid = delegstateidp->seqid = 0;
2422 		    stp->ls_stateid.other[0] = delegstateidp->other[0] =
2423 			clp->lc_clientid.lval[0];
2424 		    stp->ls_stateid.other[1] = delegstateidp->other[1] =
2425 			clp->lc_clientid.lval[1];
2426 		    stp->ls_stateid.other[2] = delegstateidp->other[2] =
2427 			nfsrv_nextstateindex(clp);
2428 		    stp->ls_compref = nd->nd_compref;
2429 		    LIST_INSERT_HEAD(&clp->lc_deleg, stp, ls_list);
2430 		    LIST_INSERT_HEAD(NFSSTATEHASH(clp,
2431 			stp->ls_stateid), stp, ls_hash);
2432 		    if (stp->ls_flags & NFSLCK_DELEGWRITE)
2433 			*rflagsp |= NFSV4OPEN_WRITEDELEGATE;
2434 		    else
2435 			*rflagsp |= NFSV4OPEN_READDELEGATE;
2436 		    clp->lc_delegtime = NFSD_MONOSEC +
2437 			nfsrv_lease + NFSRV_LEASEDELTA;
2438 
2439 		    /*
2440 		     * Now, do the associated open.
2441 		     */
2442 		    new_open->ls_stateid.seqid = 0;
2443 		    new_open->ls_stateid.other[0] = clp->lc_clientid.lval[0];
2444 		    new_open->ls_stateid.other[1] = clp->lc_clientid.lval[1];
2445 		    new_open->ls_stateid.other[2] = nfsrv_nextstateindex(clp);
2446 		    new_open->ls_flags = (new_stp->ls_flags&NFSLCK_DENYBITS)|
2447 			NFSLCK_OPEN;
2448 		    if (stp->ls_flags & NFSLCK_DELEGWRITE)
2449 			new_open->ls_flags |= (NFSLCK_READACCESS |
2450 			    NFSLCK_WRITEACCESS);
2451 		    else
2452 			new_open->ls_flags |= NFSLCK_READACCESS;
2453 		    new_open->ls_uid = new_stp->ls_uid;
2454 		    new_open->ls_lfp = lfp;
2455 		    new_open->ls_clp = clp;
2456 		    LIST_INIT(&new_open->ls_open);
2457 		    LIST_INSERT_HEAD(&lfp->lf_open, new_open, ls_file);
2458 		    LIST_INSERT_HEAD(NFSSTATEHASH(clp, new_open->ls_stateid),
2459 			new_open, ls_hash);
2460 		    /*
2461 		     * and handle the open owner
2462 		     */
2463 		    if (ownerstp) {
2464 			new_open->ls_openowner = ownerstp;
2465 			LIST_INSERT_HEAD(&ownerstp->ls_open,new_open,ls_list);
2466 		    } else {
2467 			new_open->ls_openowner = new_stp;
2468 			new_stp->ls_flags = 0;
2469 			nfsrvd_refcache(new_stp->ls_op);
2470 			new_stp->ls_noopens = 0;
2471 			LIST_INIT(&new_stp->ls_open);
2472 			LIST_INSERT_HEAD(&new_stp->ls_open, new_open, ls_list);
2473 			LIST_INSERT_HEAD(&clp->lc_open, new_stp, ls_list);
2474 			*new_stpp = NULL;
2475 			newnfsstats.srvopenowners++;
2476 			nfsrv_openpluslock++;
2477 		    }
2478 		    openstp = new_open;
2479 		    new_open = NULL;
2480 		    newnfsstats.srvopens++;
2481 		    nfsrv_openpluslock++;
2482 		    break;
2483 		}
2484 	    }
2485 	    if (stp == LIST_END(&clp->lc_olddeleg))
2486 		error = NFSERR_EXPIRED;
2487 	} else if (new_stp->ls_flags & (NFSLCK_DELEGREAD | NFSLCK_DELEGWRITE)) {
2488 	    /*
2489 	     * Scan to see that no delegation for this client and file
2490 	     * doesn't already exist.
2491 	     * There also shouldn't yet be an Open for this file and
2492 	     * openowner.
2493 	     */
2494 	    LIST_FOREACH(stp, &lfp->lf_deleg, ls_file) {
2495 		if (stp->ls_clp == clp)
2496 		    break;
2497 	    }
2498 	    if (stp == LIST_END(&lfp->lf_deleg) && openstp == NULL) {
2499 		/*
2500 		 * This is the Claim_Previous case with a delegation
2501 		 * type != Delegate_None.
2502 		 */
2503 		/*
2504 		 * First, add the delegation. (Although we must issue the
2505 		 * delegation, we can also ask for an immediate return.)
2506 		 */
2507 		new_deleg->ls_stateid.seqid = delegstateidp->seqid = 0;
2508 		new_deleg->ls_stateid.other[0] = delegstateidp->other[0] =
2509 		    clp->lc_clientid.lval[0];
2510 		new_deleg->ls_stateid.other[1] = delegstateidp->other[1] =
2511 		    clp->lc_clientid.lval[1];
2512 		new_deleg->ls_stateid.other[2] = delegstateidp->other[2] =
2513 		    nfsrv_nextstateindex(clp);
2514 		if (new_stp->ls_flags & NFSLCK_DELEGWRITE) {
2515 		    new_deleg->ls_flags = (NFSLCK_DELEGWRITE |
2516 			NFSLCK_READACCESS | NFSLCK_WRITEACCESS);
2517 		    *rflagsp |= NFSV4OPEN_WRITEDELEGATE;
2518 		} else {
2519 		    new_deleg->ls_flags = (NFSLCK_DELEGREAD |
2520 			NFSLCK_READACCESS);
2521 		    *rflagsp |= NFSV4OPEN_READDELEGATE;
2522 		}
2523 		new_deleg->ls_uid = new_stp->ls_uid;
2524 		new_deleg->ls_lfp = lfp;
2525 		new_deleg->ls_clp = clp;
2526 		new_deleg->ls_filerev = filerev;
2527 		new_deleg->ls_compref = nd->nd_compref;
2528 		LIST_INSERT_HEAD(&lfp->lf_deleg, new_deleg, ls_file);
2529 		LIST_INSERT_HEAD(NFSSTATEHASH(clp,
2530 		    new_deleg->ls_stateid), new_deleg, ls_hash);
2531 		LIST_INSERT_HEAD(&clp->lc_deleg, new_deleg, ls_list);
2532 		new_deleg = NULL;
2533 		if (delegate == 2 || nfsrv_issuedelegs == 0 ||
2534 		    (clp->lc_flags & (LCL_CALLBACKSON | LCL_CBDOWN)) !=
2535 		     LCL_CALLBACKSON ||
2536 		    NFSRV_V4DELEGLIMIT(nfsrv_delegatecnt) ||
2537 		    !NFSVNO_DELEGOK(vp))
2538 		    *rflagsp |= NFSV4OPEN_RECALL;
2539 		newnfsstats.srvdelegates++;
2540 		nfsrv_openpluslock++;
2541 		nfsrv_delegatecnt++;
2542 
2543 		/*
2544 		 * Now, do the associated open.
2545 		 */
2546 		new_open->ls_stateid.seqid = 0;
2547 		new_open->ls_stateid.other[0] = clp->lc_clientid.lval[0];
2548 		new_open->ls_stateid.other[1] = clp->lc_clientid.lval[1];
2549 		new_open->ls_stateid.other[2] = nfsrv_nextstateindex(clp);
2550 		new_open->ls_flags = (new_stp->ls_flags & NFSLCK_DENYBITS) |
2551 		    NFSLCK_OPEN;
2552 		if (new_stp->ls_flags & NFSLCK_DELEGWRITE)
2553 			new_open->ls_flags |= (NFSLCK_READACCESS |
2554 			    NFSLCK_WRITEACCESS);
2555 		else
2556 			new_open->ls_flags |= NFSLCK_READACCESS;
2557 		new_open->ls_uid = new_stp->ls_uid;
2558 		new_open->ls_lfp = lfp;
2559 		new_open->ls_clp = clp;
2560 		LIST_INIT(&new_open->ls_open);
2561 		LIST_INSERT_HEAD(&lfp->lf_open, new_open, ls_file);
2562 		LIST_INSERT_HEAD(NFSSTATEHASH(clp, new_open->ls_stateid),
2563 		   new_open, ls_hash);
2564 		/*
2565 		 * and handle the open owner
2566 		 */
2567 		if (ownerstp) {
2568 		    new_open->ls_openowner = ownerstp;
2569 		    LIST_INSERT_HEAD(&ownerstp->ls_open, new_open, ls_list);
2570 		} else {
2571 		    new_open->ls_openowner = new_stp;
2572 		    new_stp->ls_flags = 0;
2573 		    nfsrvd_refcache(new_stp->ls_op);
2574 		    new_stp->ls_noopens = 0;
2575 		    LIST_INIT(&new_stp->ls_open);
2576 		    LIST_INSERT_HEAD(&new_stp->ls_open, new_open, ls_list);
2577 		    LIST_INSERT_HEAD(&clp->lc_open, new_stp, ls_list);
2578 		    *new_stpp = NULL;
2579 		    newnfsstats.srvopenowners++;
2580 		    nfsrv_openpluslock++;
2581 		}
2582 		openstp = new_open;
2583 		new_open = NULL;
2584 		newnfsstats.srvopens++;
2585 		nfsrv_openpluslock++;
2586 	    } else {
2587 		error = NFSERR_RECLAIMCONFLICT;
2588 	    }
2589 	} else if (ownerstp) {
2590 		if (ownerstp->ls_flags & NFSLCK_NEEDSCONFIRM) {
2591 		    /* Replace the open */
2592 		    if (ownerstp->ls_op)
2593 			nfsrvd_derefcache(ownerstp->ls_op);
2594 		    ownerstp->ls_op = new_stp->ls_op;
2595 		    nfsrvd_refcache(ownerstp->ls_op);
2596 		    ownerstp->ls_seq = new_stp->ls_seq;
2597 		    *rflagsp |= NFSV4OPEN_RESULTCONFIRM;
2598 		    stp = LIST_FIRST(&ownerstp->ls_open);
2599 		    stp->ls_flags = (new_stp->ls_flags & NFSLCK_SHAREBITS) |
2600 			NFSLCK_OPEN;
2601 		    stp->ls_stateid.seqid = 0;
2602 		    stp->ls_uid = new_stp->ls_uid;
2603 		    if (lfp != stp->ls_lfp) {
2604 			LIST_REMOVE(stp, ls_file);
2605 			LIST_INSERT_HEAD(&lfp->lf_open, stp, ls_file);
2606 			stp->ls_lfp = lfp;
2607 		    }
2608 		    openstp = stp;
2609 		} else if (openstp) {
2610 		    openstp->ls_flags |= (new_stp->ls_flags & NFSLCK_SHAREBITS);
2611 		    openstp->ls_stateid.seqid++;
2612 
2613 		    /*
2614 		     * This is where we can choose to issue a delegation.
2615 		     */
2616 		    if (delegate && nfsrv_issuedelegs &&
2617 			writedeleg && !NFSVNO_EXRDONLY(exp) &&
2618 			(nfsrv_writedelegifpos || !readonly) &&
2619 			(clp->lc_flags & (LCL_CALLBACKSON | LCL_CBDOWN)) ==
2620 			 LCL_CALLBACKSON &&
2621 			!NFSRV_V4DELEGLIMIT(nfsrv_delegatecnt) &&
2622 			NFSVNO_DELEGOK(vp)) {
2623 			new_deleg->ls_stateid.seqid = delegstateidp->seqid = 0;
2624 			new_deleg->ls_stateid.other[0] = delegstateidp->other[0]
2625 			    = clp->lc_clientid.lval[0];
2626 			new_deleg->ls_stateid.other[1] = delegstateidp->other[1]
2627 			    = clp->lc_clientid.lval[1];
2628 			new_deleg->ls_stateid.other[2] = delegstateidp->other[2]
2629 			    = nfsrv_nextstateindex(clp);
2630 			new_deleg->ls_flags = (NFSLCK_DELEGWRITE |
2631 			    NFSLCK_READACCESS | NFSLCK_WRITEACCESS);
2632 			*rflagsp |= NFSV4OPEN_WRITEDELEGATE;
2633 			new_deleg->ls_uid = new_stp->ls_uid;
2634 			new_deleg->ls_lfp = lfp;
2635 			new_deleg->ls_clp = clp;
2636 			new_deleg->ls_filerev = filerev;
2637 			new_deleg->ls_compref = nd->nd_compref;
2638 			LIST_INSERT_HEAD(&lfp->lf_deleg, new_deleg, ls_file);
2639 			LIST_INSERT_HEAD(NFSSTATEHASH(clp,
2640 			    new_deleg->ls_stateid), new_deleg, ls_hash);
2641 			LIST_INSERT_HEAD(&clp->lc_deleg, new_deleg, ls_list);
2642 			new_deleg = NULL;
2643 			newnfsstats.srvdelegates++;
2644 			nfsrv_openpluslock++;
2645 			nfsrv_delegatecnt++;
2646 		    }
2647 		} else {
2648 		    new_open->ls_stateid.seqid = 0;
2649 		    new_open->ls_stateid.other[0] = clp->lc_clientid.lval[0];
2650 		    new_open->ls_stateid.other[1] = clp->lc_clientid.lval[1];
2651 		    new_open->ls_stateid.other[2] = nfsrv_nextstateindex(clp);
2652 		    new_open->ls_flags = (new_stp->ls_flags & NFSLCK_SHAREBITS)|
2653 			NFSLCK_OPEN;
2654 		    new_open->ls_uid = new_stp->ls_uid;
2655 		    new_open->ls_openowner = ownerstp;
2656 		    new_open->ls_lfp = lfp;
2657 		    new_open->ls_clp = clp;
2658 		    LIST_INIT(&new_open->ls_open);
2659 		    LIST_INSERT_HEAD(&lfp->lf_open, new_open, ls_file);
2660 		    LIST_INSERT_HEAD(&ownerstp->ls_open, new_open, ls_list);
2661 		    LIST_INSERT_HEAD(NFSSTATEHASH(clp, new_open->ls_stateid),
2662 			new_open, ls_hash);
2663 		    openstp = new_open;
2664 		    new_open = NULL;
2665 		    newnfsstats.srvopens++;
2666 		    nfsrv_openpluslock++;
2667 
2668 		    /*
2669 		     * This is where we can choose to issue a delegation.
2670 		     */
2671 		    if (delegate && nfsrv_issuedelegs &&
2672 			(writedeleg || readonly) &&
2673 			(clp->lc_flags & (LCL_CALLBACKSON | LCL_CBDOWN)) ==
2674 			 LCL_CALLBACKSON &&
2675 			!NFSRV_V4DELEGLIMIT(nfsrv_delegatecnt) &&
2676 			NFSVNO_DELEGOK(vp)) {
2677 			new_deleg->ls_stateid.seqid = delegstateidp->seqid = 0;
2678 			new_deleg->ls_stateid.other[0] = delegstateidp->other[0]
2679 			    = clp->lc_clientid.lval[0];
2680 			new_deleg->ls_stateid.other[1] = delegstateidp->other[1]
2681 			    = clp->lc_clientid.lval[1];
2682 			new_deleg->ls_stateid.other[2] = delegstateidp->other[2]
2683 			    = nfsrv_nextstateindex(clp);
2684 			if (writedeleg && !NFSVNO_EXRDONLY(exp) &&
2685 			    (nfsrv_writedelegifpos || !readonly)) {
2686 			    new_deleg->ls_flags = (NFSLCK_DELEGWRITE |
2687 				NFSLCK_READACCESS | NFSLCK_WRITEACCESS);
2688 			    *rflagsp |= NFSV4OPEN_WRITEDELEGATE;
2689 			} else {
2690 			    new_deleg->ls_flags = (NFSLCK_DELEGREAD |
2691 				NFSLCK_READACCESS);
2692 			    *rflagsp |= NFSV4OPEN_READDELEGATE;
2693 			}
2694 			new_deleg->ls_uid = new_stp->ls_uid;
2695 			new_deleg->ls_lfp = lfp;
2696 			new_deleg->ls_clp = clp;
2697 			new_deleg->ls_filerev = filerev;
2698 			new_deleg->ls_compref = nd->nd_compref;
2699 			LIST_INSERT_HEAD(&lfp->lf_deleg, new_deleg, ls_file);
2700 			LIST_INSERT_HEAD(NFSSTATEHASH(clp,
2701 			    new_deleg->ls_stateid), new_deleg, ls_hash);
2702 			LIST_INSERT_HEAD(&clp->lc_deleg, new_deleg, ls_list);
2703 			new_deleg = NULL;
2704 			newnfsstats.srvdelegates++;
2705 			nfsrv_openpluslock++;
2706 			nfsrv_delegatecnt++;
2707 		    }
2708 		}
2709 	} else {
2710 		/*
2711 		 * New owner case. Start the open_owner sequence with a
2712 		 * Needs confirmation (unless a reclaim) and hang the
2713 		 * new open off it.
2714 		 */
2715 		new_open->ls_stateid.seqid = 0;
2716 		new_open->ls_stateid.other[0] = clp->lc_clientid.lval[0];
2717 		new_open->ls_stateid.other[1] = clp->lc_clientid.lval[1];
2718 		new_open->ls_stateid.other[2] = nfsrv_nextstateindex(clp);
2719 		new_open->ls_flags = (new_stp->ls_flags & NFSLCK_SHAREBITS) |
2720 		    NFSLCK_OPEN;
2721 		new_open->ls_uid = new_stp->ls_uid;
2722 		LIST_INIT(&new_open->ls_open);
2723 		new_open->ls_openowner = new_stp;
2724 		new_open->ls_lfp = lfp;
2725 		new_open->ls_clp = clp;
2726 		LIST_INSERT_HEAD(&lfp->lf_open, new_open, ls_file);
2727 		if (new_stp->ls_flags & NFSLCK_RECLAIM) {
2728 			new_stp->ls_flags = 0;
2729 		} else {
2730 			*rflagsp |= NFSV4OPEN_RESULTCONFIRM;
2731 			new_stp->ls_flags = NFSLCK_NEEDSCONFIRM;
2732 		}
2733 		nfsrvd_refcache(new_stp->ls_op);
2734 		new_stp->ls_noopens = 0;
2735 		LIST_INIT(&new_stp->ls_open);
2736 		LIST_INSERT_HEAD(&new_stp->ls_open, new_open, ls_list);
2737 		LIST_INSERT_HEAD(&clp->lc_open, new_stp, ls_list);
2738 		LIST_INSERT_HEAD(NFSSTATEHASH(clp, new_open->ls_stateid),
2739 		    new_open, ls_hash);
2740 		openstp = new_open;
2741 		new_open = NULL;
2742 		*new_stpp = NULL;
2743 		newnfsstats.srvopens++;
2744 		nfsrv_openpluslock++;
2745 		newnfsstats.srvopenowners++;
2746 		nfsrv_openpluslock++;
2747 	}
2748 	if (!error) {
2749 		stateidp->seqid = openstp->ls_stateid.seqid;
2750 		stateidp->other[0] = openstp->ls_stateid.other[0];
2751 		stateidp->other[1] = openstp->ls_stateid.other[1];
2752 		stateidp->other[2] = openstp->ls_stateid.other[2];
2753 	}
2754 	NFSUNLOCKSTATE();
2755 	if (haslock) {
2756 		NFSLOCKV4ROOTMUTEX();
2757 		nfsv4_unlock(&nfsv4rootfs_lock, 1);
2758 		NFSUNLOCKV4ROOTMUTEX();
2759 	}
2760 	if (new_open)
2761 		FREE((caddr_t)new_open, M_NFSDSTATE);
2762 	if (new_deleg)
2763 		FREE((caddr_t)new_deleg, M_NFSDSTATE);
2764 	return (error);
2765 }
2766 
2767 /*
2768  * Open update. Does the confirm, downgrade and close.
2769  */
2770 APPLESTATIC int
2771 nfsrv_openupdate(vnode_t vp, struct nfsstate *new_stp, nfsquad_t clientid,
2772     nfsv4stateid_t *stateidp, struct nfsrv_descript *nd, NFSPROC_T *p)
2773 {
2774 	struct nfsstate *stp, *ownerstp;
2775 	struct nfsclient *clp;
2776 	struct nfslockfile *lfp;
2777 	u_int32_t bits;
2778 	int error, gotstate = 0, len = 0, ret, freedlock;
2779 	u_char client[NFSV4_OPAQUELIMIT];
2780 
2781 	/*
2782 	 * Check for restart conditions (client and server).
2783 	 */
2784 	error = nfsrv_checkrestart(clientid, new_stp->ls_flags,
2785 	    &new_stp->ls_stateid, 0);
2786 	if (error)
2787 		return (error);
2788 
2789 	NFSLOCKSTATE();
2790 	/*
2791 	 * Get the open structure via clientid and stateid.
2792 	 */
2793 	error = nfsrv_getclient(clientid, CLOPS_RENEW, &clp,
2794 	    (nfsquad_t)((u_quad_t)0), NULL, p);
2795 	if (!error)
2796 		error = nfsrv_getstate(clp, &new_stp->ls_stateid,
2797 		    new_stp->ls_flags, &stp);
2798 
2799 	/*
2800 	 * Sanity check the open.
2801 	 */
2802 	if (!error && (!(stp->ls_flags & NFSLCK_OPEN) ||
2803 		(!(new_stp->ls_flags & NFSLCK_CONFIRM) &&
2804 		 (stp->ls_openowner->ls_flags & NFSLCK_NEEDSCONFIRM)) ||
2805 		((new_stp->ls_flags & NFSLCK_CONFIRM) &&
2806 		 (!(stp->ls_openowner->ls_flags & NFSLCK_NEEDSCONFIRM)))))
2807 		error = NFSERR_BADSTATEID;
2808 
2809 	if (!error)
2810 		error = nfsrv_checkseqid(nd, new_stp->ls_seq,
2811 		    stp->ls_openowner, new_stp->ls_op);
2812 	if (!error && stp->ls_stateid.seqid != new_stp->ls_stateid.seqid &&
2813 	    !(new_stp->ls_flags & NFSLCK_CONFIRM))
2814 		error = NFSERR_OLDSTATEID;
2815 	if (!error && vnode_vtype(vp) != VREG) {
2816 		if (vnode_vtype(vp) == VDIR)
2817 			error = NFSERR_ISDIR;
2818 		else
2819 			error = NFSERR_INVAL;
2820 	}
2821 
2822 	if (error) {
2823 		/*
2824 		 * If a client tries to confirm an Open with a bad
2825 		 * seqid# and there are no byte range locks or other Opens
2826 		 * on the openowner, just throw it away, so the next use of the
2827 		 * openowner will start a fresh seq#.
2828 		 */
2829 		if (error == NFSERR_BADSEQID &&
2830 		    (new_stp->ls_flags & NFSLCK_CONFIRM) &&
2831 		    nfsrv_nootherstate(stp))
2832 			nfsrv_freeopenowner(stp->ls_openowner, 0, p);
2833 		NFSUNLOCKSTATE();
2834 		return (error);
2835 	}
2836 
2837 	/*
2838 	 * Set the return stateid.
2839 	 */
2840 	stateidp->seqid = stp->ls_stateid.seqid + 1;
2841 	stateidp->other[0] = stp->ls_stateid.other[0];
2842 	stateidp->other[1] = stp->ls_stateid.other[1];
2843 	stateidp->other[2] = stp->ls_stateid.other[2];
2844 	/*
2845 	 * Now, handle the three cases.
2846 	 */
2847 	if (new_stp->ls_flags & NFSLCK_CONFIRM) {
2848 		/*
2849 		 * If the open doesn't need confirmation, it seems to me that
2850 		 * there is a client error, but I'll just log it and keep going?
2851 		 */
2852 		if (!(stp->ls_openowner->ls_flags & NFSLCK_NEEDSCONFIRM))
2853 			printf("Nfsv4d: stray open confirm\n");
2854 		stp->ls_openowner->ls_flags = 0;
2855 		stp->ls_stateid.seqid++;
2856 		if (!(clp->lc_flags & LCL_STAMPEDSTABLE)) {
2857 			clp->lc_flags |= LCL_STAMPEDSTABLE;
2858 			len = clp->lc_idlen;
2859 			NFSBCOPY(clp->lc_id, client, len);
2860 			gotstate = 1;
2861 		}
2862 		NFSUNLOCKSTATE();
2863 	} else if (new_stp->ls_flags & NFSLCK_CLOSE) {
2864 		ownerstp = stp->ls_openowner;
2865 		lfp = stp->ls_lfp;
2866 		freedlock = 0;
2867 		ret = nfsrv_freeopen(stp, &freedlock, 0, p);
2868 		/* See comment on nfsrv_lockctrl() w.r.t. locallocks. */
2869 		if (ret) {
2870 			lfp = NULL;
2871 		} else {
2872 			if (LIST_EMPTY(&lfp->lf_lock))
2873 				lfp = NULL;
2874 		}
2875 		/*
2876 		 * For now, I won't do this. The openowner should be
2877 		 * free'd in NFSNOOPEN seconds and it will be deref'd then.
2878 		if (LIST_EMPTY(&ownerstp->ls_open) && ownerstp->ls_op) {
2879 			nfsrvd_derefcache(ownerstp->ls_op);
2880 			ownerstp->ls_op = NULL;
2881 		}
2882 		 */
2883 		NFSUNLOCKSTATE();
2884 		if (freedlock && lfp != NULL)
2885 			nfsrv_locallocks(vp, lfp, p);
2886 	} else {
2887 		/*
2888 		 * Update the share bits, making sure that the new set are a
2889 		 * subset of the old ones.
2890 		 */
2891 		bits = (new_stp->ls_flags & NFSLCK_SHAREBITS);
2892 		if (~(stp->ls_flags) & bits) {
2893 			NFSUNLOCKSTATE();
2894 			return (NFSERR_INVAL);
2895 		}
2896 		stp->ls_flags = (bits | NFSLCK_OPEN);
2897 		stp->ls_stateid.seqid++;
2898 		NFSUNLOCKSTATE();
2899 	}
2900 
2901 	/*
2902 	 * If the client just confirmed its first open, write a timestamp
2903 	 * to the stable storage file.
2904 	 */
2905 	if (gotstate)
2906 		nfsrv_writestable(client, len, NFSNST_NEWSTATE, p);
2907 	return (error);
2908 }
2909 
2910 /*
2911  * Delegation update. Does the purge and return.
2912  */
2913 APPLESTATIC int
2914 nfsrv_delegupdate(nfsquad_t clientid, nfsv4stateid_t *stateidp,
2915     vnode_t vp, int op, struct ucred *cred, NFSPROC_T *p)
2916 {
2917 	struct nfsstate *stp;
2918 	struct nfsclient *clp;
2919 	int error;
2920 	fhandle_t fh;
2921 
2922 	/*
2923 	 * Do a sanity check against the file handle for DelegReturn.
2924 	 */
2925 	if (vp) {
2926 		error = nfsvno_getfh(vp, &fh, p);
2927 		if (error)
2928 			return (error);
2929 	}
2930 	/*
2931 	 * Check for restart conditions (client and server).
2932 	 */
2933 	if (op == NFSV4OP_DELEGRETURN)
2934 		error = nfsrv_checkrestart(clientid, NFSLCK_DELEGRETURN,
2935 			stateidp, 0);
2936 	else
2937 		error = nfsrv_checkrestart(clientid, NFSLCK_DELEGPURGE,
2938 			stateidp, 0);
2939 
2940 	NFSLOCKSTATE();
2941 	/*
2942 	 * Get the open structure via clientid and stateid.
2943 	 */
2944 	if (!error)
2945 	    error = nfsrv_getclient(clientid, CLOPS_RENEW, &clp,
2946 		(nfsquad_t)((u_quad_t)0), NULL, p);
2947 	if (error) {
2948 		if (error == NFSERR_CBPATHDOWN)
2949 			error = 0;
2950 		if (error == NFSERR_STALECLIENTID && op == NFSV4OP_DELEGRETURN)
2951 			error = NFSERR_STALESTATEID;
2952 	}
2953 	if (!error && op == NFSV4OP_DELEGRETURN) {
2954 	    error = nfsrv_getstate(clp, stateidp, NFSLCK_DELEGRETURN, &stp);
2955 	    if (!error && stp->ls_stateid.seqid != stateidp->seqid)
2956 		error = NFSERR_OLDSTATEID;
2957 	}
2958 	/*
2959 	 * NFSERR_EXPIRED means that the state has gone away,
2960 	 * so Delegations have been purged. Just return ok.
2961 	 */
2962 	if (error == NFSERR_EXPIRED && op == NFSV4OP_DELEGPURGE) {
2963 		NFSUNLOCKSTATE();
2964 		return (0);
2965 	}
2966 	if (error) {
2967 		NFSUNLOCKSTATE();
2968 		return (error);
2969 	}
2970 
2971 	if (op == NFSV4OP_DELEGRETURN) {
2972 		if (NFSBCMP((caddr_t)&fh, (caddr_t)&stp->ls_lfp->lf_fh,
2973 		    sizeof (fhandle_t))) {
2974 			NFSUNLOCKSTATE();
2975 			return (NFSERR_BADSTATEID);
2976 		}
2977 		nfsrv_freedeleg(stp);
2978 	} else {
2979 		nfsrv_freedeleglist(&clp->lc_olddeleg);
2980 	}
2981 	NFSUNLOCKSTATE();
2982 	return (0);
2983 }
2984 
2985 /*
2986  * Release lock owner.
2987  */
2988 APPLESTATIC int
2989 nfsrv_releaselckown(struct nfsstate *new_stp, nfsquad_t clientid,
2990     NFSPROC_T *p)
2991 {
2992 	struct nfsstate *stp, *nstp, *openstp, *ownstp;
2993 	struct nfsclient *clp;
2994 	int error;
2995 
2996 	/*
2997 	 * Check for restart conditions (client and server).
2998 	 */
2999 	error = nfsrv_checkrestart(clientid, new_stp->ls_flags,
3000 	    &new_stp->ls_stateid, 0);
3001 	if (error)
3002 		return (error);
3003 
3004 	NFSLOCKSTATE();
3005 	/*
3006 	 * Get the lock owner by name.
3007 	 */
3008 	error = nfsrv_getclient(clientid, CLOPS_RENEW, &clp,
3009 	    (nfsquad_t)((u_quad_t)0), NULL, p);
3010 	if (error) {
3011 		NFSUNLOCKSTATE();
3012 		return (error);
3013 	}
3014 	LIST_FOREACH(ownstp, &clp->lc_open, ls_list) {
3015 	    LIST_FOREACH(openstp, &ownstp->ls_open, ls_list) {
3016 		stp = LIST_FIRST(&openstp->ls_open);
3017 		while (stp != LIST_END(&openstp->ls_open)) {
3018 		    nstp = LIST_NEXT(stp, ls_list);
3019 		    /*
3020 		     * If the owner matches, check for locks and
3021 		     * then free or return an error.
3022 		     */
3023 		    if (stp->ls_ownerlen == new_stp->ls_ownerlen &&
3024 			!NFSBCMP(stp->ls_owner, new_stp->ls_owner,
3025 			 stp->ls_ownerlen)){
3026 			if (LIST_EMPTY(&stp->ls_lock)) {
3027 			    (void) nfsrv_freelockowner(stp, NULL, 0, p);
3028 			} else {
3029 			    NFSUNLOCKSTATE();
3030 			    return (NFSERR_LOCKSHELD);
3031 			}
3032 		    }
3033 		    stp = nstp;
3034 		}
3035 	    }
3036 	}
3037 	NFSUNLOCKSTATE();
3038 	return (0);
3039 }
3040 
3041 /*
3042  * Get the file handle for a lock structure.
3043  */
3044 static int
3045 nfsrv_getlockfh(vnode_t vp, u_short flags,
3046     struct nfslockfile **new_lfpp, fhandle_t *nfhp, NFSPROC_T *p)
3047 {
3048 	fhandle_t *fhp = NULL;
3049 	struct nfslockfile *new_lfp;
3050 	int error;
3051 
3052 	/*
3053 	 * For lock, use the new nfslock structure, otherwise just
3054 	 * a fhandle_t on the stack.
3055 	 */
3056 	if (flags & NFSLCK_OPEN) {
3057 		new_lfp = *new_lfpp;
3058 		fhp = &new_lfp->lf_fh;
3059 	} else if (nfhp) {
3060 		fhp = nfhp;
3061 	} else {
3062 		panic("nfsrv_getlockfh");
3063 	}
3064 	error = nfsvno_getfh(vp, fhp, p);
3065 	return (error);
3066 }
3067 
3068 /*
3069  * Get an nfs lock structure. Allocate one, as required, and return a
3070  * pointer to it.
3071  * Returns an NFSERR_xxx upon failure or -1 to indicate no current lock.
3072  */
3073 static int
3074 nfsrv_getlockfile(u_short flags, struct nfslockfile **new_lfpp,
3075     struct nfslockfile **lfpp, fhandle_t *nfhp)
3076 {
3077 	struct nfslockfile *lfp;
3078 	fhandle_t *fhp = NULL, *tfhp;
3079 	struct nfslockhashhead *hp;
3080 	struct nfslockfile *new_lfp = NULL;
3081 
3082 	/*
3083 	 * For lock, use the new nfslock structure, otherwise just
3084 	 * a fhandle_t on the stack.
3085 	 */
3086 	if (flags & NFSLCK_OPEN) {
3087 		new_lfp = *new_lfpp;
3088 		fhp = &new_lfp->lf_fh;
3089 	} else if (nfhp) {
3090 		fhp = nfhp;
3091 	} else {
3092 		panic("nfsrv_getlockfile");
3093 	}
3094 
3095 	hp = NFSLOCKHASH(fhp);
3096 	LIST_FOREACH(lfp, hp, lf_hash) {
3097 		tfhp = &lfp->lf_fh;
3098 		if (NFSVNO_CMPFH(fhp, tfhp)) {
3099 			*lfpp = lfp;
3100 			return (0);
3101 		}
3102 	}
3103 	if (!(flags & NFSLCK_OPEN))
3104 		return (-1);
3105 
3106 	/*
3107 	 * No match, so chain the new one into the list.
3108 	 */
3109 	LIST_INIT(&new_lfp->lf_open);
3110 	LIST_INIT(&new_lfp->lf_lock);
3111 	LIST_INIT(&new_lfp->lf_deleg);
3112 	LIST_INSERT_HEAD(hp, new_lfp, lf_hash);
3113 	*lfpp = new_lfp;
3114 	*new_lfpp = NULL;
3115 	return (0);
3116 }
3117 
3118 /*
3119  * This function adds a nfslock lock structure to the list for the associated
3120  * nfsstate and nfslockfile structures. It will be inserted after the
3121  * entry pointed at by insert_lop.
3122  * Must be called with soft clock interrupts disabled.
3123  */
3124 static void
3125 nfsrv_insertlock(struct nfslock *new_lop, struct nfslock *insert_lop,
3126     struct nfsstate *stp, struct nfslockfile *lfp)
3127 {
3128 	struct nfslock *lop, *nlop;
3129 
3130 	new_lop->lo_stp = stp;
3131 	new_lop->lo_lfp = lfp;
3132 
3133 	/* Insert in increasing lo_first order */
3134 	lop = LIST_FIRST(&lfp->lf_lock);
3135 	if (lop == LIST_END(&lfp->lf_lock) ||
3136 	    new_lop->lo_first <= lop->lo_first) {
3137 		LIST_INSERT_HEAD(&lfp->lf_lock, new_lop, lo_lckfile);
3138 	} else {
3139 		nlop = LIST_NEXT(lop, lo_lckfile);
3140 		while (nlop != LIST_END(&lfp->lf_lock) &&
3141 		       nlop->lo_first < new_lop->lo_first) {
3142 			lop = nlop;
3143 			nlop = LIST_NEXT(lop, lo_lckfile);
3144 		}
3145 		LIST_INSERT_AFTER(lop, new_lop, lo_lckfile);
3146 	}
3147 
3148 	/*
3149 	 * Insert after insert_lop, which is overloaded as stp for
3150 	 * an empty list.
3151 	 */
3152 	if ((struct nfsstate *)insert_lop == stp)
3153 		LIST_INSERT_HEAD(&stp->ls_lock, new_lop, lo_lckowner);
3154 	else
3155 		LIST_INSERT_AFTER(insert_lop, new_lop, lo_lckowner);
3156 	newnfsstats.srvlocks++;
3157 	nfsrv_openpluslock++;
3158 }
3159 
3160 /*
3161  * This function updates the locking for a lock owner and given file. It
3162  * maintains a list of lock ranges ordered on increasing file offset that
3163  * are NFSLCK_READ or NFSLCK_WRITE and non-overlapping (aka POSIX style).
3164  * It always adds new_lop to the list and sometimes uses the one pointed
3165  * at by other_lopp.
3166  * Must be called with soft clock interrupts disabled.
3167  */
3168 static void
3169 nfsrv_updatelock(struct nfsstate *stp, struct nfslock **new_lopp,
3170     struct nfslock **other_lopp, struct nfslockfile *lfp)
3171 {
3172 	struct nfslock *new_lop = *new_lopp;
3173 	struct nfslock *lop, *tlop, *ilop;
3174 	struct nfslock *other_lop = *other_lopp;
3175 	int unlock = 0, myfile = 0;
3176 	u_int64_t tmp;
3177 
3178 	/*
3179 	 * Work down the list until the lock is merged.
3180 	 */
3181 	if (new_lop->lo_flags & NFSLCK_UNLOCK)
3182 		unlock = 1;
3183 	ilop = (struct nfslock *)stp;
3184 	lop = LIST_FIRST(&stp->ls_lock);
3185 	while (lop != LIST_END(&stp->ls_lock)) {
3186 	    /*
3187 	     * Only check locks for this file that aren't before the start of
3188 	     * new lock's range.
3189 	     */
3190 	    if (lop->lo_lfp == lfp) {
3191 	      myfile = 1;
3192 	      if (lop->lo_end >= new_lop->lo_first) {
3193 		if (new_lop->lo_end < lop->lo_first) {
3194 			/*
3195 			 * If the new lock ends before the start of the
3196 			 * current lock's range, no merge, just insert
3197 			 * the new lock.
3198 			 */
3199 			break;
3200 		}
3201 		if (new_lop->lo_flags == lop->lo_flags ||
3202 		    (new_lop->lo_first <= lop->lo_first &&
3203 		     new_lop->lo_end >= lop->lo_end)) {
3204 			/*
3205 			 * This lock can be absorbed by the new lock/unlock.
3206 			 * This happens when it covers the entire range
3207 			 * of the old lock or is contiguous
3208 			 * with the old lock and is of the same type or an
3209 			 * unlock.
3210 			 */
3211 			if (lop->lo_first < new_lop->lo_first)
3212 				new_lop->lo_first = lop->lo_first;
3213 			if (lop->lo_end > new_lop->lo_end)
3214 				new_lop->lo_end = lop->lo_end;
3215 			tlop = lop;
3216 			lop = LIST_NEXT(lop, lo_lckowner);
3217 			nfsrv_freenfslock(tlop);
3218 			continue;
3219 		}
3220 
3221 		/*
3222 		 * All these cases are for contiguous locks that are not the
3223 		 * same type, so they can't be merged.
3224 		 */
3225 		if (new_lop->lo_first <= lop->lo_first) {
3226 			/*
3227 			 * This case is where the new lock overlaps with the
3228 			 * first part of the old lock. Move the start of the
3229 			 * old lock to just past the end of the new lock. The
3230 			 * new lock will be inserted in front of the old, since
3231 			 * ilop hasn't been updated. (We are done now.)
3232 			 */
3233 			lop->lo_first = new_lop->lo_end;
3234 			break;
3235 		}
3236 		if (new_lop->lo_end >= lop->lo_end) {
3237 			/*
3238 			 * This case is where the new lock overlaps with the
3239 			 * end of the old lock's range. Move the old lock's
3240 			 * end to just before the new lock's first and insert
3241 			 * the new lock after the old lock.
3242 			 * Might not be done yet, since the new lock could
3243 			 * overlap further locks with higher ranges.
3244 			 */
3245 			lop->lo_end = new_lop->lo_first;
3246 			ilop = lop;
3247 			lop = LIST_NEXT(lop, lo_lckowner);
3248 			continue;
3249 		}
3250 		/*
3251 		 * The final case is where the new lock's range is in the
3252 		 * middle of the current lock's and splits the current lock
3253 		 * up. Use *other_lopp to handle the second part of the
3254 		 * split old lock range. (We are done now.)
3255 		 * For unlock, we use new_lop as other_lop and tmp, since
3256 		 * other_lop and new_lop are the same for this case.
3257 		 * We noted the unlock case above, so we don't need
3258 		 * new_lop->lo_flags any longer.
3259 		 */
3260 		tmp = new_lop->lo_first;
3261 		if (other_lop == NULL) {
3262 			if (!unlock)
3263 				panic("nfsd srv update unlock");
3264 			other_lop = new_lop;
3265 			*new_lopp = NULL;
3266 		}
3267 		other_lop->lo_first = new_lop->lo_end;
3268 		other_lop->lo_end = lop->lo_end;
3269 		other_lop->lo_flags = lop->lo_flags;
3270 		other_lop->lo_stp = stp;
3271 		other_lop->lo_lfp = lfp;
3272 		lop->lo_end = tmp;
3273 		nfsrv_insertlock(other_lop, lop, stp, lfp);
3274 		*other_lopp = NULL;
3275 		ilop = lop;
3276 		break;
3277 	      }
3278 	    }
3279 	    ilop = lop;
3280 	    lop = LIST_NEXT(lop, lo_lckowner);
3281 	    if (myfile && (lop == LIST_END(&stp->ls_lock) ||
3282 		lop->lo_lfp != lfp))
3283 		break;
3284 	}
3285 
3286 	/*
3287 	 * Insert the new lock in the list at the appropriate place.
3288 	 */
3289 	if (!unlock) {
3290 		nfsrv_insertlock(new_lop, ilop, stp, lfp);
3291 		*new_lopp = NULL;
3292 	}
3293 }
3294 
3295 /*
3296  * This function handles sequencing of locks, etc.
3297  * It returns an error that indicates what the caller should do.
3298  */
3299 static int
3300 nfsrv_checkseqid(struct nfsrv_descript *nd, u_int32_t seqid,
3301     struct nfsstate *stp, struct nfsrvcache *op)
3302 {
3303 
3304 	if (op != nd->nd_rp)
3305 		panic("nfsrvstate checkseqid");
3306 	if (!(op->rc_flag & RC_INPROG))
3307 		panic("nfsrvstate not inprog");
3308 	if (stp->ls_op && stp->ls_op->rc_refcnt <= 0) {
3309 		printf("refcnt=%d\n", stp->ls_op->rc_refcnt);
3310 		panic("nfsrvstate op refcnt");
3311 	}
3312 	if ((stp->ls_seq + 1) == seqid) {
3313 		if (stp->ls_op)
3314 			nfsrvd_derefcache(stp->ls_op);
3315 		stp->ls_op = op;
3316 		nfsrvd_refcache(op);
3317 		stp->ls_seq = seqid;
3318 		return (0);
3319 	} else if (stp->ls_seq == seqid && stp->ls_op &&
3320 		op->rc_xid == stp->ls_op->rc_xid &&
3321 		op->rc_refcnt == 0 &&
3322 		op->rc_reqlen == stp->ls_op->rc_reqlen &&
3323 		op->rc_cksum == stp->ls_op->rc_cksum) {
3324 		if (stp->ls_op->rc_flag & RC_INPROG)
3325 			return (NFSERR_DONTREPLY);
3326 		nd->nd_rp = stp->ls_op;
3327 		nd->nd_rp->rc_flag |= RC_INPROG;
3328 		nfsrvd_delcache(op);
3329 		return (NFSERR_REPLYFROMCACHE);
3330 	}
3331 	return (NFSERR_BADSEQID);
3332 }
3333 
3334 /*
3335  * Get the client ip address for callbacks. If the strings can't be parsed,
3336  * just set lc_program to 0 to indicate no callbacks are possible.
3337  * (For cases where the address can't be parsed or is 0.0.0.0.0.0, set
3338  *  the address to the client's transport address. This won't be used
3339  *  for callbacks, but can be printed out by newnfsstats for info.)
3340  * Return error if the xdr can't be parsed, 0 otherwise.
3341  */
3342 APPLESTATIC int
3343 nfsrv_getclientipaddr(struct nfsrv_descript *nd, struct nfsclient *clp)
3344 {
3345 	u_int32_t *tl;
3346 	u_char *cp, *cp2;
3347 	int i, j;
3348 	struct sockaddr_in *rad, *sad;
3349 	u_char protocol[5], addr[24];
3350 	int error = 0, cantparse = 0;
3351 	union {
3352 		u_long ival;
3353 		u_char cval[4];
3354 	} ip;
3355 	union {
3356 		u_short sval;
3357 		u_char cval[2];
3358 	} port;
3359 
3360 	rad = NFSSOCKADDR(clp->lc_req.nr_nam, struct sockaddr_in *);
3361 	rad->sin_family = AF_INET;
3362 	rad->sin_len = sizeof (struct sockaddr_in);
3363 	rad->sin_addr.s_addr = 0;
3364 	rad->sin_port = 0;
3365 	clp->lc_req.nr_client = NULL;
3366 	clp->lc_req.nr_lock = 0;
3367 	NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
3368 	i = fxdr_unsigned(int, *tl);
3369 	if (i >= 3 && i <= 4) {
3370 		error = nfsrv_mtostr(nd, protocol, i);
3371 		if (error)
3372 			goto nfsmout;
3373 		if (!strcmp(protocol, "tcp")) {
3374 			clp->lc_flags |= LCL_TCPCALLBACK;
3375 			clp->lc_req.nr_sotype = SOCK_STREAM;
3376 			clp->lc_req.nr_soproto = IPPROTO_TCP;
3377 		} else if (!strcmp(protocol, "udp")) {
3378 			clp->lc_req.nr_sotype = SOCK_DGRAM;
3379 			clp->lc_req.nr_soproto = IPPROTO_UDP;
3380 		} else {
3381 			cantparse = 1;
3382 		}
3383 	} else {
3384 		cantparse = 1;
3385 		if (i > 0) {
3386 			error = nfsm_advance(nd, NFSM_RNDUP(i), -1);
3387 			if (error)
3388 				goto nfsmout;
3389 		}
3390 	}
3391 	NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
3392 	i = fxdr_unsigned(int, *tl);
3393 	if (i < 0) {
3394 		error = NFSERR_BADXDR;
3395 		goto nfsmout;
3396 	} else if (i == 0) {
3397 		cantparse = 1;
3398 	} else if (!cantparse && i <= 23 && i >= 11) {
3399 		error = nfsrv_mtostr(nd, addr, i);
3400 		if (error)
3401 			goto nfsmout;
3402 
3403 		/*
3404 		 * Parse out the address fields. We expect 6 decimal numbers
3405 		 * separated by '.'s.
3406 		 */
3407 		cp = addr;
3408 		i = 0;
3409 		while (*cp && i < 6) {
3410 			cp2 = cp;
3411 			while (*cp2 && *cp2 != '.')
3412 				cp2++;
3413 			if (*cp2)
3414 				*cp2++ = '\0';
3415 			else if (i != 5) {
3416 				cantparse = 1;
3417 				break;
3418 			}
3419 			j = nfsrv_getipnumber(cp);
3420 			if (j >= 0) {
3421 				if (i < 4)
3422 					ip.cval[3 - i] = j;
3423 				else
3424 					port.cval[5 - i] = j;
3425 			} else {
3426 				cantparse = 1;
3427 				break;
3428 			}
3429 			cp = cp2;
3430 			i++;
3431 		}
3432 		if (!cantparse) {
3433 			if (ip.ival != 0x0) {
3434 				rad->sin_addr.s_addr = htonl(ip.ival);
3435 				rad->sin_port = htons(port.sval);
3436 			} else {
3437 				cantparse = 1;
3438 			}
3439 		}
3440 	} else {
3441 		cantparse = 1;
3442 		if (i > 0) {
3443 			error = nfsm_advance(nd, NFSM_RNDUP(i), -1);
3444 			if (error)
3445 				goto nfsmout;
3446 		}
3447 	}
3448 	if (cantparse) {
3449 		sad = NFSSOCKADDR(nd->nd_nam, struct sockaddr_in *);
3450 		rad->sin_addr.s_addr = sad->sin_addr.s_addr;
3451 		rad->sin_port = 0x0;
3452 		clp->lc_program = 0;
3453 	}
3454 nfsmout:
3455 	return (error);
3456 }
3457 
3458 /*
3459  * Turn a string of up to three decimal digits into a number. Return -1 upon
3460  * error.
3461  */
3462 static int
3463 nfsrv_getipnumber(u_char *cp)
3464 {
3465 	int i = 0, j = 0;
3466 
3467 	while (*cp) {
3468 		if (j > 2 || *cp < '0' || *cp > '9')
3469 			return (-1);
3470 		i *= 10;
3471 		i += (*cp - '0');
3472 		cp++;
3473 		j++;
3474 	}
3475 	if (i < 256)
3476 		return (i);
3477 	return (-1);
3478 }
3479 
3480 /*
3481  * This function checks for restart conditions.
3482  */
3483 static int
3484 nfsrv_checkrestart(nfsquad_t clientid, u_int32_t flags,
3485     nfsv4stateid_t *stateidp, int specialid)
3486 {
3487 	int ret;
3488 
3489 	/*
3490 	 * First check for a server restart. Open, LockT, ReleaseLockOwner
3491 	 * and DelegPurge have a clientid, the rest a stateid.
3492 	 */
3493 	if (flags &
3494 	    (NFSLCK_OPEN | NFSLCK_TEST | NFSLCK_RELEASE | NFSLCK_DELEGPURGE)) {
3495 		if (clientid.lval[0] != nfsrvboottime)
3496 			return (NFSERR_STALECLIENTID);
3497 	} else if (stateidp->other[0] != nfsrvboottime &&
3498 		specialid == 0)
3499 		return (NFSERR_STALESTATEID);
3500 
3501 	/*
3502 	 * Read, Write, Setattr and LockT can return NFSERR_GRACE and do
3503 	 * not use a lock/open owner seqid#, so the check can be done now.
3504 	 * (The others will be checked, as required, later.)
3505 	 */
3506 	if (!(flags & (NFSLCK_CHECK | NFSLCK_TEST)))
3507 		return (0);
3508 
3509 	NFSLOCKSTATE();
3510 	ret = nfsrv_checkgrace(flags);
3511 	NFSUNLOCKSTATE();
3512 	return (ret);
3513 }
3514 
3515 /*
3516  * Check for grace.
3517  */
3518 static int
3519 nfsrv_checkgrace(u_int32_t flags)
3520 {
3521 
3522 	if (nfsrv_stablefirst.nsf_flags & NFSNSF_GRACEOVER) {
3523 		if (flags & NFSLCK_RECLAIM)
3524 			return (NFSERR_NOGRACE);
3525 	} else {
3526 		if (!(flags & NFSLCK_RECLAIM))
3527 			return (NFSERR_GRACE);
3528 
3529 		/*
3530 		 * If grace is almost over and we are still getting Reclaims,
3531 		 * extend grace a bit.
3532 		 */
3533 		if ((NFSD_MONOSEC + NFSRV_LEASEDELTA) >
3534 		    nfsrv_stablefirst.nsf_eograce)
3535 			nfsrv_stablefirst.nsf_eograce = NFSD_MONOSEC +
3536 				NFSRV_LEASEDELTA;
3537 	}
3538 	return (0);
3539 }
3540 
3541 /*
3542  * Do a server callback.
3543  */
3544 static int
3545 nfsrv_docallback(struct nfsclient *clp, int procnum,
3546     nfsv4stateid_t *stateidp, int trunc, fhandle_t *fhp,
3547     struct nfsvattr *nap, nfsattrbit_t *attrbitp, NFSPROC_T *p)
3548 {
3549 	mbuf_t m;
3550 	u_int32_t *tl;
3551 	struct nfsrv_descript nfsd, *nd = &nfsd;
3552 	struct ucred *cred;
3553 	int error = 0;
3554 	u_int32_t callback;
3555 
3556 	cred = newnfs_getcred();
3557 	NFSLOCKSTATE();	/* mostly for lc_cbref++ */
3558 	if (clp->lc_flags & LCL_NEEDSCONFIRM) {
3559 		NFSUNLOCKSTATE();
3560 		panic("docallb");
3561 	}
3562 	clp->lc_cbref++;
3563 
3564 	/*
3565 	 * Fill the callback program# and version into the request
3566 	 * structure for newnfs_connect() to use.
3567 	 */
3568 	clp->lc_req.nr_prog = clp->lc_program;
3569 	clp->lc_req.nr_vers = NFSV4_CBVERS;
3570 
3571 	/*
3572 	 * First, fill in some of the fields of nd and cr.
3573 	 */
3574 	nd->nd_flag = ND_NFSV4;
3575 	if (clp->lc_flags & LCL_GSS)
3576 		nd->nd_flag |= ND_KERBV;
3577 	nd->nd_repstat = 0;
3578 	cred->cr_uid = clp->lc_uid;
3579 	cred->cr_gid = clp->lc_gid;
3580 	callback = clp->lc_callback;
3581 	NFSUNLOCKSTATE();
3582 	cred->cr_ngroups = 1;
3583 
3584 	/*
3585 	 * Get the first mbuf for the request.
3586 	 */
3587 	MGET(m, M_WAIT, MT_DATA);
3588 	mbuf_setlen(m, 0);
3589 	nd->nd_mreq = nd->nd_mb = m;
3590 	nd->nd_bpos = NFSMTOD(m, caddr_t);
3591 
3592 	/*
3593 	 * and build the callback request.
3594 	 */
3595 	if (procnum == NFSV4OP_CBGETATTR) {
3596 		nd->nd_procnum = NFSV4PROC_CBCOMPOUND;
3597 		(void) nfsm_strtom(nd, "CB Getattr", 10);
3598 		NFSM_BUILD(tl, u_int32_t *, 4 * NFSX_UNSIGNED);
3599 		*tl++ = txdr_unsigned(NFSV4_MINORVERSION);
3600 		*tl++ = txdr_unsigned(callback);
3601 		*tl++ = txdr_unsigned(1);
3602 		*tl = txdr_unsigned(NFSV4OP_CBGETATTR);
3603 		(void) nfsm_fhtom(nd, (u_int8_t *)fhp, NFSX_MYFH, 0);
3604 		(void) nfsrv_putattrbit(nd, attrbitp);
3605 	} else if (procnum == NFSV4OP_CBRECALL) {
3606 		nd->nd_procnum = NFSV4PROC_CBCOMPOUND;
3607 		(void) nfsm_strtom(nd, "CB Recall", 9);
3608 		NFSM_BUILD(tl, u_int32_t *, 5 * NFSX_UNSIGNED + NFSX_STATEID);
3609 		*tl++ = txdr_unsigned(NFSV4_MINORVERSION);
3610 		*tl++ = txdr_unsigned(callback);
3611 		*tl++ = txdr_unsigned(1);
3612 		*tl++ = txdr_unsigned(NFSV4OP_CBRECALL);
3613 		*tl++ = txdr_unsigned(stateidp->seqid);
3614 		NFSBCOPY((caddr_t)stateidp->other, (caddr_t)tl,
3615 		    NFSX_STATEIDOTHER);
3616 		tl += (NFSX_STATEIDOTHER / NFSX_UNSIGNED);
3617 		if (trunc)
3618 			*tl = newnfs_true;
3619 		else
3620 			*tl = newnfs_false;
3621 		(void) nfsm_fhtom(nd, (u_int8_t *)fhp, NFSX_MYFH, 0);
3622 	} else {
3623 		nd->nd_procnum = NFSV4PROC_CBNULL;
3624 	}
3625 
3626 	/*
3627 	 * Call newnfs_connect(), as required, and then newnfs_request().
3628 	 */
3629 	(void) newnfs_sndlock(&clp->lc_req.nr_lock);
3630 	if (clp->lc_req.nr_client == NULL) {
3631 		if (nd->nd_procnum == NFSV4PROC_CBNULL)
3632 			error = newnfs_connect(NULL, &clp->lc_req, cred,
3633 			    NULL, 1);
3634 		else
3635 			error = newnfs_connect(NULL, &clp->lc_req, cred,
3636 			    NULL, 3);
3637 	}
3638 	newnfs_sndunlock(&clp->lc_req.nr_lock);
3639 	if (!error) {
3640 		error = newnfs_request(nd, NULL, clp, &clp->lc_req, NULL,
3641 		    NULL, cred, clp->lc_program, NFSV4_CBVERS, NULL, 1, NULL);
3642 	}
3643 	NFSFREECRED(cred);
3644 
3645 	/*
3646 	 * If error is set here, the Callback path isn't working
3647 	 * properly, so twiddle the appropriate LCL_ flags.
3648 	 * (nd_repstat != 0 indicates the Callback path is working,
3649 	 *  but the callback failed on the client.)
3650 	 */
3651 	if (error) {
3652 		/*
3653 		 * Mark the callback pathway down, which disabled issuing
3654 		 * of delegations and gets Renew to return NFSERR_CBPATHDOWN.
3655 		 */
3656 		NFSLOCKSTATE();
3657 		clp->lc_flags |= LCL_CBDOWN;
3658 		NFSUNLOCKSTATE();
3659 	} else {
3660 		/*
3661 		 * Callback worked. If the callback path was down, disable
3662 		 * callbacks, so no more delegations will be issued. (This
3663 		 * is done on the assumption that the callback pathway is
3664 		 * flakey.)
3665 		 */
3666 		NFSLOCKSTATE();
3667 		if (clp->lc_flags & LCL_CBDOWN)
3668 			clp->lc_flags &= ~(LCL_CBDOWN | LCL_CALLBACKSON);
3669 		NFSUNLOCKSTATE();
3670 		if (nd->nd_repstat)
3671 			error = nd->nd_repstat;
3672 		else if (procnum == NFSV4OP_CBGETATTR)
3673 			error = nfsv4_loadattr(nd, NULL, nap, NULL, NULL, 0,
3674 			    NULL, NULL, NULL, NULL, NULL, 0, NULL, NULL, NULL,
3675 			    p, NULL);
3676 		mbuf_freem(nd->nd_mrep);
3677 	}
3678 	NFSLOCKSTATE();
3679 	clp->lc_cbref--;
3680 	if ((clp->lc_flags & LCL_WAKEUPWANTED) && clp->lc_cbref == 0) {
3681 		clp->lc_flags &= ~LCL_WAKEUPWANTED;
3682 		NFSUNLOCKSTATE();
3683 		wakeup((caddr_t)clp);
3684 	} else {
3685 		NFSUNLOCKSTATE();
3686 	}
3687 	return (error);
3688 }
3689 
3690 /*
3691  * Return the next index# for a clientid. Mostly just increment and return
3692  * the next one, but... if the 32bit unsigned does actually wrap around,
3693  * it should be rebooted.
3694  * At an average rate of one new client per second, it will wrap around in
3695  * approximately 136 years. (I think the server will have been shut
3696  * down or rebooted before then.)
3697  */
3698 static u_int32_t
3699 nfsrv_nextclientindex(void)
3700 {
3701 	static u_int32_t client_index = 0;
3702 
3703 	client_index++;
3704 	if (client_index != 0)
3705 		return (client_index);
3706 
3707 	printf("%s: out of clientids\n", __func__);
3708 	return (client_index);
3709 }
3710 
3711 /*
3712  * Return the next index# for a stateid. Mostly just increment and return
3713  * the next one, but... if the 32bit unsigned does actually wrap around
3714  * (will a BSD server stay up that long?), find
3715  * new start and end values.
3716  */
3717 static u_int32_t
3718 nfsrv_nextstateindex(struct nfsclient *clp)
3719 {
3720 	struct nfsstate *stp;
3721 	int i;
3722 	u_int32_t canuse, min_index, max_index;
3723 
3724 	if (!(clp->lc_flags & LCL_INDEXNOTOK)) {
3725 		clp->lc_stateindex++;
3726 		if (clp->lc_stateindex != clp->lc_statemaxindex)
3727 			return (clp->lc_stateindex);
3728 	}
3729 
3730 	/*
3731 	 * Yuck, we've hit the end.
3732 	 * Look for a new min and max.
3733 	 */
3734 	min_index = 0;
3735 	max_index = 0xffffffff;
3736 	for (i = 0; i < NFSSTATEHASHSIZE; i++) {
3737 	    LIST_FOREACH(stp, &clp->lc_stateid[i], ls_hash) {
3738 		if (stp->ls_stateid.other[2] > 0x80000000) {
3739 		    if (stp->ls_stateid.other[2] < max_index)
3740 			max_index = stp->ls_stateid.other[2];
3741 		} else {
3742 		    if (stp->ls_stateid.other[2] > min_index)
3743 			min_index = stp->ls_stateid.other[2];
3744 		}
3745 	    }
3746 	}
3747 
3748 	/*
3749 	 * Yikes, highly unlikely, but I'll handle it anyhow.
3750 	 */
3751 	if (min_index == 0x80000000 && max_index == 0x80000001) {
3752 	    canuse = 0;
3753 	    /*
3754 	     * Loop around until we find an unused entry. Return that
3755 	     * and set LCL_INDEXNOTOK, so the search will continue next time.
3756 	     * (This is one of those rare cases where a goto is the
3757 	     *  cleanest way to code the loop.)
3758 	     */
3759 tryagain:
3760 	    for (i = 0; i < NFSSTATEHASHSIZE; i++) {
3761 		LIST_FOREACH(stp, &clp->lc_stateid[i], ls_hash) {
3762 		    if (stp->ls_stateid.other[2] == canuse) {
3763 			canuse++;
3764 			goto tryagain;
3765 		    }
3766 		}
3767 	    }
3768 	    clp->lc_flags |= LCL_INDEXNOTOK;
3769 	    return (canuse);
3770 	}
3771 
3772 	/*
3773 	 * Ok to start again from min + 1.
3774 	 */
3775 	clp->lc_stateindex = min_index + 1;
3776 	clp->lc_statemaxindex = max_index;
3777 	clp->lc_flags &= ~LCL_INDEXNOTOK;
3778 	return (clp->lc_stateindex);
3779 }
3780 
3781 /*
3782  * The following functions handle the stable storage file that deals with
3783  * the edge conditions described in RFC3530 Sec. 8.6.3.
3784  * The file is as follows:
3785  * - a single record at the beginning that has the lease time of the
3786  *   previous server instance (before the last reboot) and the nfsrvboottime
3787  *   values for the previous server boots.
3788  *   These previous boot times are used to ensure that the current
3789  *   nfsrvboottime does not, somehow, get set to a previous one.
3790  *   (This is important so that Stale ClientIDs and StateIDs can
3791  *    be recognized.)
3792  *   The number of previous nfsvrboottime values preceeds the list.
3793  * - followed by some number of appended records with:
3794  *   - client id string
3795  *   - flag that indicates it is a record revoking state via lease
3796  *     expiration or similar
3797  *     OR has successfully acquired state.
3798  * These structures vary in length, with the client string at the end, up
3799  * to NFSV4_OPAQUELIMIT in size.
3800  *
3801  * At the end of the grace period, the file is truncated, the first
3802  * record is rewritten with updated information and any acquired state
3803  * records for successful reclaims of state are written.
3804  *
3805  * Subsequent records are appended when the first state is issued to
3806  * a client and when state is revoked for a client.
3807  *
3808  * When reading the file in, state issued records that come later in
3809  * the file override older ones, since the append log is in cronological order.
3810  * If, for some reason, the file can't be read, the grace period is
3811  * immediately terminated and all reclaims get NFSERR_NOGRACE.
3812  */
3813 
3814 /*
3815  * Read in the stable storage file. Called by nfssvc() before the nfsd
3816  * processes start servicing requests.
3817  */
3818 APPLESTATIC void
3819 nfsrv_setupstable(NFSPROC_T *p)
3820 {
3821 	struct nfsrv_stablefirst *sf = &nfsrv_stablefirst;
3822 	struct nfsrv_stable *sp, *nsp;
3823 	struct nfst_rec *tsp;
3824 	int error, i, tryagain;
3825 	off_t off = 0;
3826 	int aresid, len;
3827 	struct timeval curtime;
3828 
3829 	/*
3830 	 * If NFSNSF_UPDATEDONE is set, this is a restart of the nfsds without
3831 	 * a reboot, so state has not been lost.
3832 	 */
3833 	if (sf->nsf_flags & NFSNSF_UPDATEDONE)
3834 		return;
3835 	/*
3836 	 * Set Grace over just until the file reads successfully.
3837 	 */
3838 	NFSGETTIME(&curtime);
3839 	nfsrvboottime = curtime.tv_sec;
3840 	LIST_INIT(&sf->nsf_head);
3841 	sf->nsf_flags = (NFSNSF_GRACEOVER | NFSNSF_NEEDLOCK);
3842 	sf->nsf_eograce = NFSD_MONOSEC + NFSRV_LEASEDELTA;
3843 	if (sf->nsf_fp == NULL)
3844 		return;
3845 	error = NFSD_RDWR(UIO_READ, NFSFPVNODE(sf->nsf_fp),
3846 	    (caddr_t)&sf->nsf_rec, sizeof (struct nfsf_rec), off, UIO_SYSSPACE,
3847 	    0, NFSFPCRED(sf->nsf_fp), &aresid, p);
3848 	if (error || aresid || sf->nsf_numboots == 0 ||
3849 		sf->nsf_numboots > NFSNSF_MAXNUMBOOTS)
3850 		return;
3851 
3852 	/*
3853 	 * Now, read in the boottimes.
3854 	 */
3855 	sf->nsf_bootvals = (time_t *)malloc((sf->nsf_numboots + 1) *
3856 		sizeof (time_t), M_TEMP, M_WAITOK);
3857 	off = sizeof (struct nfsf_rec);
3858 	error = NFSD_RDWR(UIO_READ, NFSFPVNODE(sf->nsf_fp),
3859 	    (caddr_t)sf->nsf_bootvals, sf->nsf_numboots * sizeof (time_t), off,
3860 	    UIO_SYSSPACE, 0, NFSFPCRED(sf->nsf_fp), &aresid, p);
3861 	if (error || aresid) {
3862 		free((caddr_t)sf->nsf_bootvals, M_TEMP);
3863 		sf->nsf_bootvals = NULL;
3864 		return;
3865 	}
3866 
3867 	/*
3868 	 * Make sure this nfsrvboottime is different from all recorded
3869 	 * previous ones.
3870 	 */
3871 	do {
3872 		tryagain = 0;
3873 		for (i = 0; i < sf->nsf_numboots; i++) {
3874 			if (nfsrvboottime == sf->nsf_bootvals[i]) {
3875 				nfsrvboottime++;
3876 				tryagain = 1;
3877 				break;
3878 			}
3879 		}
3880 	} while (tryagain);
3881 
3882 	sf->nsf_flags |= NFSNSF_OK;
3883 	off += (sf->nsf_numboots * sizeof (time_t));
3884 
3885 	/*
3886 	 * Read through the file, building a list of records for grace
3887 	 * checking.
3888 	 * Each record is between sizeof (struct nfst_rec) and
3889 	 * sizeof (struct nfst_rec) + NFSV4_OPAQUELIMIT - 1
3890 	 * and is actually sizeof (struct nfst_rec) + nst_len - 1.
3891 	 */
3892 	tsp = (struct nfst_rec *)malloc(sizeof (struct nfst_rec) +
3893 		NFSV4_OPAQUELIMIT - 1, M_TEMP, M_WAITOK);
3894 	do {
3895 	    error = NFSD_RDWR(UIO_READ, NFSFPVNODE(sf->nsf_fp),
3896 	        (caddr_t)tsp, sizeof (struct nfst_rec) + NFSV4_OPAQUELIMIT - 1,
3897 	        off, UIO_SYSSPACE, 0, NFSFPCRED(sf->nsf_fp), &aresid, p);
3898 	    len = (sizeof (struct nfst_rec) + NFSV4_OPAQUELIMIT - 1) - aresid;
3899 	    if (error || (len > 0 && (len < sizeof (struct nfst_rec) ||
3900 		len < (sizeof (struct nfst_rec) + tsp->len - 1)))) {
3901 		/*
3902 		 * Yuck, the file has been corrupted, so just return
3903 		 * after clearing out any restart state, so the grace period
3904 		 * is over.
3905 		 */
3906 		LIST_FOREACH_SAFE(sp, &sf->nsf_head, nst_list, nsp) {
3907 			LIST_REMOVE(sp, nst_list);
3908 			free((caddr_t)sp, M_TEMP);
3909 		}
3910 		free((caddr_t)tsp, M_TEMP);
3911 		sf->nsf_flags &= ~NFSNSF_OK;
3912 		free((caddr_t)sf->nsf_bootvals, M_TEMP);
3913 		sf->nsf_bootvals = NULL;
3914 		return;
3915 	    }
3916 	    if (len > 0) {
3917 		off += sizeof (struct nfst_rec) + tsp->len - 1;
3918 		/*
3919 		 * Search the list for a matching client.
3920 		 */
3921 		LIST_FOREACH(sp, &sf->nsf_head, nst_list) {
3922 			if (tsp->len == sp->nst_len &&
3923 			    !NFSBCMP(tsp->client, sp->nst_client, tsp->len))
3924 				break;
3925 		}
3926 		if (sp == LIST_END(&sf->nsf_head)) {
3927 			sp = (struct nfsrv_stable *)malloc(tsp->len +
3928 				sizeof (struct nfsrv_stable) - 1, M_TEMP,
3929 				M_WAITOK);
3930 			NFSBCOPY((caddr_t)tsp, (caddr_t)&sp->nst_rec,
3931 				sizeof (struct nfst_rec) + tsp->len - 1);
3932 			LIST_INSERT_HEAD(&sf->nsf_head, sp, nst_list);
3933 		} else {
3934 			if (tsp->flag == NFSNST_REVOKE)
3935 				sp->nst_flag |= NFSNST_REVOKE;
3936 			else
3937 				/*
3938 				 * A subsequent timestamp indicates the client
3939 				 * did a setclientid/confirm and any previous
3940 				 * revoke is no longer relevant.
3941 				 */
3942 				sp->nst_flag &= ~NFSNST_REVOKE;
3943 		}
3944 	    }
3945 	} while (len > 0);
3946 	free((caddr_t)tsp, M_TEMP);
3947 	sf->nsf_flags = NFSNSF_OK;
3948 	sf->nsf_eograce = NFSD_MONOSEC + sf->nsf_lease +
3949 		NFSRV_LEASEDELTA;
3950 }
3951 
3952 /*
3953  * Update the stable storage file, now that the grace period is over.
3954  */
3955 APPLESTATIC void
3956 nfsrv_updatestable(NFSPROC_T *p)
3957 {
3958 	struct nfsrv_stablefirst *sf = &nfsrv_stablefirst;
3959 	struct nfsrv_stable *sp, *nsp;
3960 	int i;
3961 	struct nfsvattr nva;
3962 	vnode_t vp;
3963 #if defined(__FreeBSD_version) && (__FreeBSD_version >= 500000)
3964 	mount_t mp = NULL;
3965 #endif
3966 	int error;
3967 
3968 	if (sf->nsf_fp == NULL || (sf->nsf_flags & NFSNSF_UPDATEDONE))
3969 		return;
3970 	sf->nsf_flags |= NFSNSF_UPDATEDONE;
3971 	/*
3972 	 * Ok, we need to rewrite the stable storage file.
3973 	 * - truncate to 0 length
3974 	 * - write the new first structure
3975 	 * - loop through the data structures, writing out any that
3976 	 *   have timestamps older than the old boot
3977 	 */
3978 	if (sf->nsf_bootvals) {
3979 		sf->nsf_numboots++;
3980 		for (i = sf->nsf_numboots - 2; i >= 0; i--)
3981 			sf->nsf_bootvals[i + 1] = sf->nsf_bootvals[i];
3982 	} else {
3983 		sf->nsf_numboots = 1;
3984 		sf->nsf_bootvals = (time_t *)malloc(sizeof (time_t),
3985 			M_TEMP, M_WAITOK);
3986 	}
3987 	sf->nsf_bootvals[0] = nfsrvboottime;
3988 	sf->nsf_lease = nfsrv_lease;
3989 	NFSVNO_ATTRINIT(&nva);
3990 	NFSVNO_SETATTRVAL(&nva, size, 0);
3991 	vp = NFSFPVNODE(sf->nsf_fp);
3992 	NFS_STARTWRITE(vp, &mp);
3993 	NFSVOPLOCK(vp, LK_EXCLUSIVE | LK_RETRY, p);
3994 	error = nfsvno_setattr(vp, &nva, NFSFPCRED(sf->nsf_fp), p, NULL);
3995 	NFS_ENDWRITE(mp);
3996 	NFSVOPUNLOCK(vp, 0, p);
3997 	if (!error)
3998 	    error = NFSD_RDWR(UIO_WRITE, vp,
3999 		(caddr_t)&sf->nsf_rec, sizeof (struct nfsf_rec), (off_t)0,
4000 		UIO_SYSSPACE, IO_SYNC, NFSFPCRED(sf->nsf_fp), NULL, p);
4001 	if (!error)
4002 	    error = NFSD_RDWR(UIO_WRITE, vp,
4003 		(caddr_t)sf->nsf_bootvals,
4004 		sf->nsf_numboots * sizeof (time_t),
4005 		(off_t)(sizeof (struct nfsf_rec)),
4006 		UIO_SYSSPACE, IO_SYNC, NFSFPCRED(sf->nsf_fp), NULL, p);
4007 	free((caddr_t)sf->nsf_bootvals, M_TEMP);
4008 	sf->nsf_bootvals = NULL;
4009 	if (error) {
4010 		sf->nsf_flags &= ~NFSNSF_OK;
4011 		printf("EEK! Can't write NfsV4 stable storage file\n");
4012 		return;
4013 	}
4014 	sf->nsf_flags |= NFSNSF_OK;
4015 
4016 	/*
4017 	 * Loop through the list and write out timestamp records for
4018 	 * any clients that successfully reclaimed state.
4019 	 */
4020 	LIST_FOREACH_SAFE(sp, &sf->nsf_head, nst_list, nsp) {
4021 		if (sp->nst_flag & NFSNST_GOTSTATE) {
4022 			nfsrv_writestable(sp->nst_client, sp->nst_len,
4023 				NFSNST_NEWSTATE, p);
4024 			sp->nst_clp->lc_flags |= LCL_STAMPEDSTABLE;
4025 		}
4026 		LIST_REMOVE(sp, nst_list);
4027 		free((caddr_t)sp, M_TEMP);
4028 	}
4029 }
4030 
4031 /*
4032  * Append a record to the stable storage file.
4033  */
4034 APPLESTATIC void
4035 nfsrv_writestable(u_char *client, int len, int flag, NFSPROC_T *p)
4036 {
4037 	struct nfsrv_stablefirst *sf = &nfsrv_stablefirst;
4038 	struct nfst_rec *sp;
4039 	int error;
4040 
4041 	if (!(sf->nsf_flags & NFSNSF_OK) || sf->nsf_fp == NULL)
4042 		return;
4043 	sp = (struct nfst_rec *)malloc(sizeof (struct nfst_rec) +
4044 		len - 1, M_TEMP, M_WAITOK);
4045 	sp->len = len;
4046 	NFSBCOPY(client, sp->client, len);
4047 	sp->flag = flag;
4048 	error = NFSD_RDWR(UIO_WRITE, NFSFPVNODE(sf->nsf_fp),
4049 	    (caddr_t)sp, sizeof (struct nfst_rec) + len - 1, (off_t)0,
4050 	    UIO_SYSSPACE, (IO_SYNC | IO_APPEND), NFSFPCRED(sf->nsf_fp), NULL, p);
4051 	free((caddr_t)sp, M_TEMP);
4052 	if (error) {
4053 		sf->nsf_flags &= ~NFSNSF_OK;
4054 		printf("EEK! Can't write NfsV4 stable storage file\n");
4055 	}
4056 }
4057 
4058 /*
4059  * This function is called during the grace period to mark a client
4060  * that successfully reclaimed state.
4061  */
4062 static void
4063 nfsrv_markstable(struct nfsclient *clp)
4064 {
4065 	struct nfsrv_stable *sp;
4066 
4067 	/*
4068 	 * First find the client structure.
4069 	 */
4070 	LIST_FOREACH(sp, &nfsrv_stablefirst.nsf_head, nst_list) {
4071 		if (sp->nst_len == clp->lc_idlen &&
4072 		    !NFSBCMP(sp->nst_client, clp->lc_id, sp->nst_len))
4073 			break;
4074 	}
4075 	if (sp == LIST_END(&nfsrv_stablefirst.nsf_head))
4076 		return;
4077 
4078 	/*
4079 	 * Now, just mark it and set the nfsclient back pointer.
4080 	 */
4081 	sp->nst_flag |= NFSNST_GOTSTATE;
4082 	sp->nst_clp = clp;
4083 }
4084 
4085 /*
4086  * This function is called for a reclaim, to see if it gets grace.
4087  * It returns 0 if a reclaim is allowed, 1 otherwise.
4088  */
4089 static int
4090 nfsrv_checkstable(struct nfsclient *clp)
4091 {
4092 	struct nfsrv_stable *sp;
4093 
4094 	/*
4095 	 * First, find the entry for the client.
4096 	 */
4097 	LIST_FOREACH(sp, &nfsrv_stablefirst.nsf_head, nst_list) {
4098 		if (sp->nst_len == clp->lc_idlen &&
4099 		    !NFSBCMP(sp->nst_client, clp->lc_id, sp->nst_len))
4100 			break;
4101 	}
4102 
4103 	/*
4104 	 * If not in the list, state was revoked or no state was issued
4105 	 * since the previous reboot, a reclaim is denied.
4106 	 */
4107 	if (sp == LIST_END(&nfsrv_stablefirst.nsf_head) ||
4108 	    (sp->nst_flag & NFSNST_REVOKE) ||
4109 	    !(nfsrv_stablefirst.nsf_flags & NFSNSF_OK))
4110 		return (1);
4111 	return (0);
4112 }
4113 
4114 /*
4115  * Test for and try to clear out a conflicting client. This is called by
4116  * nfsrv_lockctrl() and nfsrv_openctrl() when conflicts with other clients
4117  * a found.
4118  * The trick here is that it can't revoke a conflicting client with an
4119  * expired lease unless it holds the v4root lock, so...
4120  * If no v4root lock, get the lock and return 1 to indicate "try again".
4121  * Return 0 to indicate the conflict can't be revoked and 1 to indicate
4122  * the revocation worked and the conflicting client is "bye, bye", so it
4123  * can be tried again.
4124  * Unlocks State before a non-zero value is returned.
4125  */
4126 static int
4127 nfsrv_clientconflict(struct nfsclient *clp, int *haslockp, __unused vnode_t vp,
4128     NFSPROC_T *p)
4129 {
4130 	int gotlock;
4131 
4132 	/*
4133 	 * If lease hasn't expired, we can't fix it.
4134 	 */
4135 	if (clp->lc_expiry >= NFSD_MONOSEC ||
4136 	    !(nfsrv_stablefirst.nsf_flags & NFSNSF_UPDATEDONE))
4137 		return (0);
4138 	if (*haslockp == 0) {
4139 		NFSUNLOCKSTATE();
4140 		NFSVOPUNLOCK(vp, 0, p);
4141 		NFSLOCKV4ROOTMUTEX();
4142 		nfsv4_relref(&nfsv4rootfs_lock);
4143 		do {
4144 			gotlock = nfsv4_lock(&nfsv4rootfs_lock, 1, NULL,
4145 			    NFSV4ROOTLOCKMUTEXPTR);
4146 		} while (!gotlock);
4147 		NFSUNLOCKV4ROOTMUTEX();
4148 		NFSLOCKSTATE();	/* to avoid a race with */
4149 		NFSUNLOCKSTATE();	/* nfsrv_servertimer() */
4150 		*haslockp = 1;
4151 		NFSVOPLOCK(vp, LK_EXCLUSIVE | LK_RETRY, p);
4152 		return (1);
4153 	}
4154 	NFSUNLOCKSTATE();
4155 
4156 	/*
4157 	 * Ok, we can expire the conflicting client.
4158 	 */
4159 	nfsrv_writestable(clp->lc_id, clp->lc_idlen, NFSNST_REVOKE, p);
4160 	nfsrv_cleanclient(clp, p);
4161 	nfsrv_freedeleglist(&clp->lc_deleg);
4162 	nfsrv_freedeleglist(&clp->lc_olddeleg);
4163 	LIST_REMOVE(clp, lc_hash);
4164 	nfsrv_zapclient(clp, p);
4165 	return (1);
4166 }
4167 
4168 
4169 /*
4170  * Resolve a delegation conflict.
4171  * Returns 0 to indicate the conflict was resolved without sleeping.
4172  * Return -1 to indicate that the caller should check for conflicts again.
4173  * Return > 0 for an error that should be returned, normally NFSERR_DELAY.
4174  *
4175  * Also, manipulate the nfsv4root_lock, as required. It isn't changed
4176  * for a return of 0, since there was no sleep and it could be required
4177  * later. It is released for a return of NFSERR_DELAY, since the caller
4178  * will return that error. It is released when a sleep was done waiting
4179  * for the delegation to be returned or expire (so that other nfsds can
4180  * handle ops). Then, it must be acquired for the write to stable storage.
4181  * (This function is somewhat similar to nfsrv_clientconflict(), but
4182  *  the semantics differ in a couple of subtle ways. The return of 0
4183  *  indicates the conflict was resolved without sleeping here, not
4184  *  that the conflict can't be resolved and the handling of nfsv4root_lock
4185  *  differs, as noted above.)
4186  * Unlocks State before returning a non-zero value.
4187  */
4188 static int
4189 nfsrv_delegconflict(struct nfsstate *stp, int *haslockp, NFSPROC_T *p,
4190     __unused vnode_t vp)
4191 {
4192 	struct nfsclient *clp = stp->ls_clp;
4193 	int gotlock, error, retrycnt, zapped_clp;
4194 	nfsv4stateid_t tstateid;
4195 	fhandle_t tfh;
4196 
4197 	/*
4198 	 * If the conflict is with an old delegation...
4199 	 */
4200 	if (stp->ls_flags & NFSLCK_OLDDELEG) {
4201 		/*
4202 		 * You can delete it, if it has expired.
4203 		 */
4204 		if (clp->lc_delegtime < NFSD_MONOSEC) {
4205 			nfsrv_freedeleg(stp);
4206 			NFSUNLOCKSTATE();
4207 			return (-1);
4208 		}
4209 		NFSUNLOCKSTATE();
4210 		/*
4211 		 * During this delay, the old delegation could expire or it
4212 		 * could be recovered by the client via an Open with
4213 		 * CLAIM_DELEGATE_PREV.
4214 		 * Release the nfsv4root_lock, if held.
4215 		 */
4216 		if (*haslockp) {
4217 			*haslockp = 0;
4218 			NFSLOCKV4ROOTMUTEX();
4219 			nfsv4_unlock(&nfsv4rootfs_lock, 1);
4220 			NFSUNLOCKV4ROOTMUTEX();
4221 		}
4222 		return (NFSERR_DELAY);
4223 	}
4224 
4225 	/*
4226 	 * It's a current delegation, so:
4227 	 * - check to see if the delegation has expired
4228 	 *   - if so, get the v4root lock and then expire it
4229 	 */
4230 	if (!(stp->ls_flags & NFSLCK_DELEGRECALL)) {
4231 		/*
4232 		 * - do a recall callback, since not yet done
4233 		 * For now, never allow truncate to be set. To use
4234 		 * truncate safely, it must be guaranteed that the
4235 		 * Remove, Rename or Setattr with size of 0 will
4236 		 * succeed and that would require major changes to
4237 		 * the VFS/Vnode OPs.
4238 		 * Set the expiry time large enough so that it won't expire
4239 		 * until after the callback, then set it correctly, once
4240 		 * the callback is done. (The delegation will now time
4241 		 * out whether or not the Recall worked ok. The timeout
4242 		 * will be extended when ops are done on the delegation
4243 		 * stateid, up to the timelimit.)
4244 		 */
4245 		stp->ls_delegtime = NFSD_MONOSEC + (2 * nfsrv_lease) +
4246 		    NFSRV_LEASEDELTA;
4247 		stp->ls_delegtimelimit = NFSD_MONOSEC + (6 * nfsrv_lease) +
4248 		    NFSRV_LEASEDELTA;
4249 		stp->ls_flags |= NFSLCK_DELEGRECALL;
4250 
4251 		/*
4252 		 * Loop NFSRV_CBRETRYCNT times while the CBRecall replies
4253 		 * NFSERR_BADSTATEID or NFSERR_BADHANDLE. This is done
4254 		 * in order to try and avoid a race that could happen
4255 		 * when a CBRecall request passed the Open reply with
4256 		 * the delegation in it when transitting the network.
4257 		 * Since nfsrv_docallback will sleep, don't use stp after
4258 		 * the call.
4259 		 */
4260 		NFSBCOPY((caddr_t)&stp->ls_stateid, (caddr_t)&tstateid,
4261 		    sizeof (tstateid));
4262 		NFSBCOPY((caddr_t)&stp->ls_lfp->lf_fh, (caddr_t)&tfh,
4263 		    sizeof (tfh));
4264 		NFSUNLOCKSTATE();
4265 		if (*haslockp) {
4266 			*haslockp = 0;
4267 			NFSLOCKV4ROOTMUTEX();
4268 			nfsv4_unlock(&nfsv4rootfs_lock, 1);
4269 			NFSUNLOCKV4ROOTMUTEX();
4270 		}
4271 		retrycnt = 0;
4272 		do {
4273 		    error = nfsrv_docallback(clp, NFSV4OP_CBRECALL,
4274 			&tstateid, 0, &tfh, NULL, NULL, p);
4275 		    retrycnt++;
4276 		} while ((error == NFSERR_BADSTATEID ||
4277 		    error == NFSERR_BADHANDLE) && retrycnt < NFSV4_CBRETRYCNT);
4278 		return (NFSERR_DELAY);
4279 	}
4280 
4281 	if (clp->lc_expiry >= NFSD_MONOSEC &&
4282 	    stp->ls_delegtime >= NFSD_MONOSEC) {
4283 		NFSUNLOCKSTATE();
4284 		/*
4285 		 * A recall has been done, but it has not yet expired.
4286 		 * So, RETURN_DELAY.
4287 		 */
4288 		if (*haslockp) {
4289 			*haslockp = 0;
4290 			NFSLOCKV4ROOTMUTEX();
4291 			nfsv4_unlock(&nfsv4rootfs_lock, 1);
4292 			NFSUNLOCKV4ROOTMUTEX();
4293 		}
4294 		return (NFSERR_DELAY);
4295 	}
4296 
4297 	/*
4298 	 * If we don't yet have the lock, just get it and then return,
4299 	 * since we need that before deleting expired state, such as
4300 	 * this delegation.
4301 	 * When getting the lock, unlock the vnode, so other nfsds that
4302 	 * are in progress, won't get stuck waiting for the vnode lock.
4303 	 */
4304 	if (*haslockp == 0) {
4305 		NFSUNLOCKSTATE();
4306 		NFSVOPUNLOCK(vp, 0, p);
4307 		NFSLOCKV4ROOTMUTEX();
4308 		nfsv4_relref(&nfsv4rootfs_lock);
4309 		do {
4310 			gotlock = nfsv4_lock(&nfsv4rootfs_lock, 1, NULL,
4311 			    NFSV4ROOTLOCKMUTEXPTR);
4312 		} while (!gotlock);
4313 		NFSUNLOCKV4ROOTMUTEX();
4314 		NFSLOCKSTATE();	/* to avoid a race with */
4315 		NFSUNLOCKSTATE();	/* nfsrv_servertimer() */
4316 		*haslockp = 1;
4317 		NFSVOPLOCK(vp, LK_EXCLUSIVE | LK_RETRY, p);
4318 		return (-1);
4319 	}
4320 
4321 	NFSUNLOCKSTATE();
4322 	/*
4323 	 * Ok, we can delete the expired delegation.
4324 	 * First, write the Revoke record to stable storage and then
4325 	 * clear out the conflict.
4326 	 * Since all other nfsd threads are now blocked, we can safely
4327 	 * sleep without the state changing.
4328 	 */
4329 	nfsrv_writestable(clp->lc_id, clp->lc_idlen, NFSNST_REVOKE, p);
4330 	if (clp->lc_expiry < NFSD_MONOSEC) {
4331 		nfsrv_cleanclient(clp, p);
4332 		nfsrv_freedeleglist(&clp->lc_deleg);
4333 		nfsrv_freedeleglist(&clp->lc_olddeleg);
4334 		LIST_REMOVE(clp, lc_hash);
4335 		zapped_clp = 1;
4336 	} else {
4337 		nfsrv_freedeleg(stp);
4338 		zapped_clp = 0;
4339 	}
4340 	if (zapped_clp)
4341 		nfsrv_zapclient(clp, p);
4342 	return (-1);
4343 }
4344 
4345 /*
4346  * Check for a remove allowed, if remove is set to 1 and get rid of
4347  * delegations.
4348  */
4349 APPLESTATIC int
4350 nfsrv_checkremove(vnode_t vp, int remove, NFSPROC_T *p)
4351 {
4352 	struct nfsstate *stp;
4353 	struct nfslockfile *lfp;
4354 	int error, haslock = 0;
4355 	fhandle_t nfh;
4356 
4357 	/*
4358 	 * First, get the lock file structure.
4359 	 * (A return of -1 means no associated state, so remove ok.)
4360 	 */
4361 	error = nfsrv_getlockfh(vp, NFSLCK_CHECK, NULL, &nfh, p);
4362 tryagain:
4363 	NFSLOCKSTATE();
4364 	if (!error)
4365 		error = nfsrv_getlockfile(NFSLCK_CHECK, NULL, &lfp, &nfh);
4366 	if (error) {
4367 		NFSUNLOCKSTATE();
4368 		if (haslock) {
4369 			NFSLOCKV4ROOTMUTEX();
4370 			nfsv4_unlock(&nfsv4rootfs_lock, 1);
4371 			NFSUNLOCKV4ROOTMUTEX();
4372 		}
4373 		if (error == -1)
4374 			return (0);
4375 		return (error);
4376 	}
4377 
4378 	/*
4379 	 * Now, we must Recall any delegations.
4380 	 */
4381 	error = nfsrv_cleandeleg(vp, lfp, NULL, &haslock, p);
4382 	if (error) {
4383 		/*
4384 		 * nfsrv_cleandeleg() unlocks state for non-zero
4385 		 * return.
4386 		 */
4387 		if (error == -1)
4388 			goto tryagain;
4389 		if (haslock) {
4390 			NFSLOCKV4ROOTMUTEX();
4391 			nfsv4_unlock(&nfsv4rootfs_lock, 1);
4392 			NFSUNLOCKV4ROOTMUTEX();
4393 		}
4394 		return (error);
4395 	}
4396 
4397 	/*
4398 	 * Now, look for a conflicting open share.
4399 	 */
4400 	if (remove) {
4401 		LIST_FOREACH(stp, &lfp->lf_open, ls_file) {
4402 			if (stp->ls_flags & NFSLCK_WRITEDENY) {
4403 				error = NFSERR_FILEOPEN;
4404 				break;
4405 			}
4406 		}
4407 	}
4408 
4409 	NFSUNLOCKSTATE();
4410 	if (haslock) {
4411 		NFSLOCKV4ROOTMUTEX();
4412 		nfsv4_unlock(&nfsv4rootfs_lock, 1);
4413 		NFSUNLOCKV4ROOTMUTEX();
4414 	}
4415 	return (error);
4416 }
4417 
4418 /*
4419  * Clear out all delegations for the file referred to by lfp.
4420  * May return NFSERR_DELAY, if there will be a delay waiting for
4421  * delegations to expire.
4422  * Returns -1 to indicate it slept while recalling a delegation.
4423  * This function has the side effect of deleting the nfslockfile structure,
4424  * if it no longer has associated state and didn't have to sleep.
4425  * Unlocks State before a non-zero value is returned.
4426  */
4427 static int
4428 nfsrv_cleandeleg(vnode_t vp, struct nfslockfile *lfp,
4429     struct nfsclient *clp, int *haslockp, NFSPROC_T *p)
4430 {
4431 	struct nfsstate *stp, *nstp;
4432 	int ret;
4433 
4434 	stp = LIST_FIRST(&lfp->lf_deleg);
4435 	while (stp != LIST_END(&lfp->lf_deleg)) {
4436 		nstp = LIST_NEXT(stp, ls_file);
4437 		if (stp->ls_clp != clp) {
4438 			ret = nfsrv_delegconflict(stp, haslockp, p, vp);
4439 			if (ret) {
4440 				/*
4441 				 * nfsrv_delegconflict() unlocks state
4442 				 * when it returns non-zero.
4443 				 */
4444 				return (ret);
4445 			}
4446 		}
4447 		stp = nstp;
4448 	}
4449 	return (0);
4450 }
4451 
4452 /*
4453  * There are certain operations that, when being done outside of NFSv4,
4454  * require that any NFSv4 delegation for the file be recalled.
4455  * This function is to be called for those cases:
4456  * VOP_RENAME() - When a delegation is being recalled for any reason,
4457  *	the client may have to do Opens against the server, using the file's
4458  *	final component name. If the file has been renamed on the server,
4459  *	that component name will be incorrect and the Open will fail.
4460  * VOP_REMOVE() - Theoretically, a client could Open a file after it has
4461  *	been removed on the server, if there is a delegation issued to
4462  *	that client for the file. I say "theoretically" since clients
4463  *	normally do an Access Op before the Open and that Access Op will
4464  *	fail with ESTALE. Note that NFSv2 and 3 don't even do Opens, so
4465  *	they will detect the file's removal in the same manner. (There is
4466  *	one case where RFC3530 allows a client to do an Open without first
4467  *	doing an Access Op, which is passage of a check against the ACE
4468  *	returned with a Write delegation, but current practice is to ignore
4469  *	the ACE and always do an Access Op.)
4470  *	Since the functions can only be called with an unlocked vnode, this
4471  *	can't be done at this time.
4472  * VOP_ADVLOCK() - When a client holds a delegation, it can issue byte range
4473  *	locks locally in the client, which are not visible to the server. To
4474  *	deal with this, issuing of delegations for a vnode must be disabled
4475  *	and all delegations for the vnode recalled. This is done via the
4476  *	second function, using the VV_DISABLEDELEG vflag on the vnode.
4477  */
4478 APPLESTATIC void
4479 nfsd_recalldelegation(vnode_t vp, NFSPROC_T *p)
4480 {
4481 	struct timespec mytime;
4482 	int32_t starttime;
4483 	int error;
4484 
4485 	KASSERT(!VOP_ISLOCKED(vp), ("vp %p is locked", vp));
4486 
4487 	/*
4488 	 * First, check to see if the server is currently running and it has
4489 	 * been called for a regular file when issuing delegations.
4490 	 */
4491 	if (newnfs_numnfsd == 0 || vp->v_type != VREG ||
4492 	    nfsrv_issuedelegs == 0)
4493 		return;
4494 
4495 	/*
4496 	 * Now, call nfsrv_checkremove() in a loop while it returns
4497 	 * NFSERR_DELAY. Return upon any other error or when timed out.
4498 	 */
4499 	NFSGETNANOTIME(&mytime);
4500 	starttime = (u_int32_t)mytime.tv_sec;
4501 	do {
4502 		error = nfsrv_checkremove(vp, 0, p);
4503 		if (error == NFSERR_DELAY) {
4504 			NFSGETNANOTIME(&mytime);
4505 			if (((u_int32_t)mytime.tv_sec - starttime) >
4506 			    NFS_REMOVETIMEO &&
4507 			    ((u_int32_t)mytime.tv_sec - starttime) <
4508 			    100000)
4509 				return;
4510 			/* Sleep for a short period of time */
4511 			(void) nfs_catnap(PZERO, "nfsremove");
4512 		}
4513 	} while (error == NFSERR_DELAY);
4514 }
4515 
4516 APPLESTATIC void
4517 nfsd_disabledelegation(vnode_t vp, NFSPROC_T *p)
4518 {
4519 
4520 #ifdef VV_DISABLEDELEG
4521 	/*
4522 	 * First, flag issuance of delegations disabled.
4523 	 */
4524 	atomic_set_long(&vp->v_vflag, VV_DISABLEDELEG);
4525 #endif
4526 
4527 	/*
4528 	 * Then call nfsd_recalldelegation() to get rid of all extant
4529 	 * delegations.
4530 	 */
4531 	nfsd_recalldelegation(vp, p);
4532 }
4533 
4534 /*
4535  * Check for conflicting locks, etc. and then get rid of delegations.
4536  * (At one point I thought that I should get rid of delegations for any
4537  *  Setattr, since it could potentially disallow the I/O op (read or write)
4538  *  allowed by the delegation. However, Setattr Ops that aren't changing
4539  *  the size get a stateid of all 0s, so you can't tell if it is a delegation
4540  *  for the same client or a different one, so I decided to only get rid
4541  *  of delegations for other clients when the size is being changed.)
4542  * In general, a Setattr can disable NFS I/O Ops that are outstanding, such
4543  * as Write backs, even if there is no delegation, so it really isn't any
4544  * different?)
4545  */
4546 APPLESTATIC int
4547 nfsrv_checksetattr(vnode_t vp, struct nfsrv_descript *nd,
4548     nfsv4stateid_t *stateidp, struct nfsvattr *nvap, nfsattrbit_t *attrbitp,
4549     struct nfsexstuff *exp, NFSPROC_T *p)
4550 {
4551 	struct nfsstate st, *stp = &st;
4552 	struct nfslock lo, *lop = &lo;
4553 	int error = 0;
4554 	nfsquad_t clientid;
4555 
4556 	if (NFSISSET_ATTRBIT(attrbitp, NFSATTRBIT_SIZE)) {
4557 		stp->ls_flags = (NFSLCK_CHECK | NFSLCK_WRITEACCESS);
4558 		lop->lo_first = nvap->na_size;
4559 	} else {
4560 		stp->ls_flags = 0;
4561 		lop->lo_first = 0;
4562 	}
4563 	if (NFSISSET_ATTRBIT(attrbitp, NFSATTRBIT_OWNER) ||
4564 	    NFSISSET_ATTRBIT(attrbitp, NFSATTRBIT_OWNERGROUP) ||
4565 	    NFSISSET_ATTRBIT(attrbitp, NFSATTRBIT_MODE) ||
4566 	    NFSISSET_ATTRBIT(attrbitp, NFSATTRBIT_ACL))
4567 		stp->ls_flags |= NFSLCK_SETATTR;
4568 	if (stp->ls_flags == 0)
4569 		return (0);
4570 	lop->lo_end = NFS64BITSSET;
4571 	lop->lo_flags = NFSLCK_WRITE;
4572 	stp->ls_ownerlen = 0;
4573 	stp->ls_op = NULL;
4574 	stp->ls_uid = nd->nd_cred->cr_uid;
4575 	stp->ls_stateid.seqid = stateidp->seqid;
4576 	clientid.lval[0] = stp->ls_stateid.other[0] = stateidp->other[0];
4577 	clientid.lval[1] = stp->ls_stateid.other[1] = stateidp->other[1];
4578 	stp->ls_stateid.other[2] = stateidp->other[2];
4579 	error = nfsrv_lockctrl(vp, &stp, &lop, NULL, clientid,
4580 	    stateidp, exp, nd, p);
4581 	return (error);
4582 }
4583 
4584 /*
4585  * Check for a write delegation and do a CBGETATTR if there is one, updating
4586  * the attributes, as required.
4587  * Should I return an error if I can't get the attributes? (For now, I'll
4588  * just return ok.
4589  */
4590 APPLESTATIC int
4591 nfsrv_checkgetattr(struct nfsrv_descript *nd, vnode_t vp,
4592     struct nfsvattr *nvap, nfsattrbit_t *attrbitp, struct ucred *cred,
4593     NFSPROC_T *p)
4594 {
4595 	struct nfsstate *stp;
4596 	struct nfslockfile *lfp;
4597 	struct nfsclient *clp;
4598 	struct nfsvattr nva;
4599 	fhandle_t nfh;
4600 	int error;
4601 	nfsattrbit_t cbbits;
4602 	u_quad_t delegfilerev;
4603 
4604 	NFSCBGETATTR_ATTRBIT(attrbitp, &cbbits);
4605 	if (!NFSNONZERO_ATTRBIT(&cbbits))
4606 		return (0);
4607 
4608 	/*
4609 	 * Get the lock file structure.
4610 	 * (A return of -1 means no associated state, so return ok.)
4611 	 */
4612 	error = nfsrv_getlockfh(vp, NFSLCK_CHECK, NULL, &nfh, p);
4613 	NFSLOCKSTATE();
4614 	if (!error)
4615 		error = nfsrv_getlockfile(NFSLCK_CHECK, NULL, &lfp, &nfh);
4616 	if (error) {
4617 		NFSUNLOCKSTATE();
4618 		if (error == -1)
4619 			return (0);
4620 		return (error);
4621 	}
4622 
4623 	/*
4624 	 * Now, look for a write delegation.
4625 	 */
4626 	LIST_FOREACH(stp, &lfp->lf_deleg, ls_file) {
4627 		if (stp->ls_flags & NFSLCK_DELEGWRITE)
4628 			break;
4629 	}
4630 	if (stp == LIST_END(&lfp->lf_deleg)) {
4631 		NFSUNLOCKSTATE();
4632 		return (0);
4633 	}
4634 	clp = stp->ls_clp;
4635 	delegfilerev = stp->ls_filerev;
4636 
4637 	/*
4638 	 * If the Write delegation was issued as a part of this Compound RPC
4639 	 * or if we have an Implied Clientid (used in a previous Op in this
4640 	 * compound) and it is the client the delegation was issued to,
4641 	 * just return ok.
4642 	 * I also assume that it is from the same client iff the network
4643 	 * host IP address is the same as the callback address. (Not
4644 	 * exactly correct by the RFC, but avoids a lot of Getattr
4645 	 * callbacks.)
4646 	 */
4647 	if (nd->nd_compref == stp->ls_compref ||
4648 	    ((nd->nd_flag & ND_IMPLIEDCLID) &&
4649 	     clp->lc_clientid.qval == nd->nd_clientid.qval) ||
4650 	     nfsaddr2_match(clp->lc_req.nr_nam, nd->nd_nam)) {
4651 		NFSUNLOCKSTATE();
4652 		return (0);
4653 	}
4654 
4655 	/*
4656 	 * We are now done with the delegation state structure,
4657 	 * so the statelock can be released and we can now tsleep().
4658 	 */
4659 
4660 	/*
4661 	 * Now, we must do the CB Getattr callback, to see if Change or Size
4662 	 * has changed.
4663 	 */
4664 	if (clp->lc_expiry >= NFSD_MONOSEC) {
4665 		NFSUNLOCKSTATE();
4666 		NFSVNO_ATTRINIT(&nva);
4667 		nva.na_filerev = NFS64BITSSET;
4668 		error = nfsrv_docallback(clp, NFSV4OP_CBGETATTR, NULL,
4669 		    0, &nfh, &nva, &cbbits, p);
4670 		if (!error) {
4671 			if ((nva.na_filerev != NFS64BITSSET &&
4672 			    nva.na_filerev > delegfilerev) ||
4673 			    (NFSVNO_ISSETSIZE(&nva) &&
4674 			     nva.na_size != nvap->na_size)) {
4675 				nfsvno_updfilerev(vp, nvap, cred, p);
4676 				if (NFSVNO_ISSETSIZE(&nva))
4677 					nvap->na_size = nva.na_size;
4678 			}
4679 		}
4680 	} else {
4681 		NFSUNLOCKSTATE();
4682 	}
4683 	return (0);
4684 }
4685 
4686 /*
4687  * This function looks for openowners that haven't had any opens for
4688  * a while and throws them away. Called by an nfsd when NFSNSF_NOOPENS
4689  * is set.
4690  */
4691 APPLESTATIC void
4692 nfsrv_throwawayopens(NFSPROC_T *p)
4693 {
4694 	struct nfsclient *clp, *nclp;
4695 	struct nfsstate *stp, *nstp;
4696 	int i;
4697 
4698 	NFSLOCKSTATE();
4699 	nfsrv_stablefirst.nsf_flags &= ~NFSNSF_NOOPENS;
4700 	/*
4701 	 * For each client...
4702 	 */
4703 	for (i = 0; i < NFSCLIENTHASHSIZE; i++) {
4704 	    LIST_FOREACH_SAFE(clp, &nfsclienthash[i], lc_hash, nclp) {
4705 		LIST_FOREACH_SAFE(stp, &clp->lc_open, ls_list, nstp) {
4706 			if (LIST_EMPTY(&stp->ls_open) &&
4707 			    (stp->ls_noopens > NFSNOOPEN ||
4708 			     (nfsrv_openpluslock * 2) >
4709 			     NFSRV_V4STATELIMIT))
4710 				nfsrv_freeopenowner(stp, 0, p);
4711 		}
4712 	    }
4713 	}
4714 	NFSUNLOCKSTATE();
4715 }
4716 
4717 /*
4718  * This function checks to see if the credentials are the same.
4719  * Returns 1 for not same, 0 otherwise.
4720  */
4721 static int
4722 nfsrv_notsamecredname(struct nfsrv_descript *nd, struct nfsclient *clp)
4723 {
4724 
4725 	if (nd->nd_flag & ND_GSS) {
4726 		if (!(clp->lc_flags & LCL_GSS))
4727 			return (1);
4728 		if (clp->lc_flags & LCL_NAME) {
4729 			if (nd->nd_princlen != clp->lc_namelen ||
4730 			    NFSBCMP(nd->nd_principal, clp->lc_name,
4731 				clp->lc_namelen))
4732 				return (1);
4733 			else
4734 				return (0);
4735 		}
4736 		if (nd->nd_cred->cr_uid == clp->lc_uid)
4737 			return (0);
4738 		else
4739 			return (1);
4740 	} else if (clp->lc_flags & LCL_GSS)
4741 		return (1);
4742 	/*
4743 	 * For AUTH_SYS, allow the same uid or root. (This is underspecified
4744 	 * in RFC3530, which talks about principals, but doesn't say anything
4745 	 * about uids for AUTH_SYS.)
4746 	 */
4747 	if (nd->nd_cred->cr_uid == clp->lc_uid || nd->nd_cred->cr_uid == 0)
4748 		return (0);
4749 	else
4750 		return (1);
4751 }
4752 
4753 /*
4754  * Calculate the lease expiry time.
4755  */
4756 static time_t
4757 nfsrv_leaseexpiry(void)
4758 {
4759 	struct timeval curtime;
4760 
4761 	NFSGETTIME(&curtime);
4762 	if (nfsrv_stablefirst.nsf_eograce > NFSD_MONOSEC)
4763 		return (NFSD_MONOSEC + 2 * (nfsrv_lease + NFSRV_LEASEDELTA));
4764 	return (NFSD_MONOSEC + nfsrv_lease + NFSRV_LEASEDELTA);
4765 }
4766 
4767 /*
4768  * Delay the delegation timeout as far as ls_delegtimelimit, as required.
4769  */
4770 static void
4771 nfsrv_delaydelegtimeout(struct nfsstate *stp)
4772 {
4773 
4774 	if ((stp->ls_flags & NFSLCK_DELEGRECALL) == 0)
4775 		return;
4776 
4777 	if ((stp->ls_delegtime + 15) > NFSD_MONOSEC &&
4778 	    stp->ls_delegtime < stp->ls_delegtimelimit) {
4779 		stp->ls_delegtime += nfsrv_lease;
4780 		if (stp->ls_delegtime > stp->ls_delegtimelimit)
4781 			stp->ls_delegtime = stp->ls_delegtimelimit;
4782 	}
4783 }
4784 
4785 /*
4786  * Go through a lock list and set local locks for all ranges.
4787  * This assumes that the lock list is sorted on increasing
4788  * lo_first and that the list won't change, despite the possibility
4789  * of sleeps.
4790  */
4791 static void
4792 nfsrv_locallocks(vnode_t vp, struct nfslockfile *lfp,
4793     NFSPROC_T *p)
4794 {
4795 	struct nfslock *lop, *nlop;
4796 	vnode_t tvp;
4797 	int newcollate, flags = 0;
4798 	u_int64_t first = 0x0ull, end = 0x0ull;
4799 
4800 	if (!nfsrv_dolocallocks)
4801 		return;
4802 	/*
4803 	 * If vp is NULL, a vnode must be aquired from the file
4804 	 * handle.
4805 	 */
4806 	if (vp == NULL) {
4807 		if (lfp == NULL)
4808 			panic("nfsrv_locallocks");
4809 		tvp = nfsvno_getvp(&lfp->lf_fh);
4810 		if (tvp == NULL)
4811 			return;
4812 	} else {
4813 		tvp = vp;
4814 	}
4815 
4816 	/*
4817 	 * If lfp == NULL, the lock list is empty, so just unlock
4818 	 * everything.
4819 	 */
4820 	if (lfp == NULL) {
4821 		(void) nfsvno_advlock(tvp, F_UNLCK, (u_int64_t)0,
4822 		    NFS64BITSSET, p);
4823 		/* vp can't be NULL */
4824 		return;
4825 	}
4826 
4827 	/* handle whole file case first */
4828 	lop = LIST_FIRST(&lfp->lf_lock);
4829 	if (lop != LIST_END(&lfp->lf_lock) &&
4830 	    lop->lo_first == (u_int64_t)0 &&
4831 	    lop->lo_end == NFS64BITSSET) {
4832 		if (lop->lo_flags & NFSLCK_WRITE)
4833 			(void) nfsvno_advlock(tvp, F_WRLCK, lop->lo_first,
4834 			    lop->lo_end, p);
4835 		else
4836 			(void) nfsvno_advlock(tvp, F_RDLCK, lop->lo_first,
4837 			    lop->lo_end, p);
4838 		if (vp == NULL)
4839 			vput(tvp);
4840 		return;
4841 	}
4842 
4843 	/*
4844 	 * Now, handle the separate byte ranges cases.
4845 	 */
4846 	(void) nfsvno_advlock(tvp, F_UNLCK, (u_int64_t)0,
4847 	    NFS64BITSSET, p);
4848 	newcollate = 1;
4849 	while (lop != LIST_END(&lfp->lf_lock)) {
4850 		nlop = LIST_NEXT(lop, lo_lckfile);
4851 		if (newcollate) {
4852 			first = lop->lo_first;
4853 			end = lop->lo_end;
4854 			flags = lop->lo_flags;
4855 			newcollate = 0;
4856 		}
4857 		if (nlop != LIST_END(&lfp->lf_lock) &&
4858 		    flags == nlop->lo_flags &&
4859 		    end >= nlop->lo_first) {
4860 			/* can collate this one */
4861 			end = nlop->lo_end;
4862 		} else {
4863 			/* do the local lock and start again */
4864 			if (flags & NFSLCK_WRITE)
4865 				(void) nfsvno_advlock(tvp, F_WRLCK, first,
4866 				    end, p);
4867 			else
4868 				(void) nfsvno_advlock(tvp, F_RDLCK, first,
4869 				    end, p);
4870 			newcollate = 1;
4871 		}
4872 		lop = nlop;
4873 	}
4874 	if (vp == NULL)
4875 		vput(tvp);
4876 }
4877 
4878 /*
4879  * This function checks to see if there is any other state associated
4880  * with the openowner for this Open.
4881  * It returns 1 if there is no other state, 0 otherwise.
4882  */
4883 static int
4884 nfsrv_nootherstate(struct nfsstate *stp)
4885 {
4886 	struct nfsstate *tstp;
4887 
4888 	LIST_FOREACH(tstp, &stp->ls_openowner->ls_open, ls_list) {
4889 		if (tstp != stp || !LIST_EMPTY(&tstp->ls_lock))
4890 			return (0);
4891 	}
4892 	return (1);
4893 }
4894 
4895