xref: /freebsd/sys/fs/nfsserver/nfs_nfsdstate.c (revision a3cf0ef5a295c885c895fabfd56470c0d1db322d)
1 /*-
2  * Copyright (c) 2009 Rick Macklem, University of Guelph
3  * All rights reserved.
4  *
5  * Redistribution and use in source and binary forms, with or without
6  * modification, are permitted provided that the following conditions
7  * are met:
8  * 1. Redistributions of source code must retain the above copyright
9  *    notice, this list of conditions and the following disclaimer.
10  * 2. Redistributions in binary form must reproduce the above copyright
11  *    notice, this list of conditions and the following disclaimer in the
12  *    documentation and/or other materials provided with the distribution.
13  *
14  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
15  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
16  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
17  * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
18  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
19  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
20  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
21  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
22  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
23  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
24  * SUCH DAMAGE.
25  *
26  */
27 
28 #include <sys/cdefs.h>
29 __FBSDID("$FreeBSD$");
30 
31 #ifndef APPLEKEXT
32 #include <fs/nfs/nfsport.h>
33 
34 struct nfsrv_stablefirst nfsrv_stablefirst;
35 int nfsrv_issuedelegs = 0;
36 int nfsrv_dolocallocks = 0;
37 struct nfsv4lock nfsv4rootfs_lock;
38 
39 extern int newnfs_numnfsd;
40 extern struct nfsstats newnfsstats;
41 extern int nfsrv_lease;
42 extern struct timeval nfsboottime;
43 extern u_int32_t newnfs_true, newnfs_false;
44 NFSV4ROOTLOCKMUTEX;
45 NFSSTATESPINLOCK;
46 
47 /*
48  * Hash lists for nfs V4.
49  * (Some would put them in the .h file, but I don't like declaring storage
50  *  in a .h)
51  */
52 struct nfsclienthashhead nfsclienthash[NFSCLIENTHASHSIZE];
53 struct nfslockhashhead nfslockhash[NFSLOCKHASHSIZE];
54 #endif	/* !APPLEKEXT */
55 
56 static u_int32_t nfsrv_openpluslock = 0, nfsrv_delegatecnt = 0;
57 static time_t nfsrvboottime;
58 static int nfsrv_writedelegifpos = 1;
59 static int nfsrv_returnoldstateid = 0, nfsrv_clients = 0;
60 static int nfsrv_clienthighwater = NFSRV_CLIENTHIGHWATER;
61 static int nfsrv_nogsscallback = 0;
62 
63 /* local functions */
64 static void nfsrv_dumpaclient(struct nfsclient *clp,
65     struct nfsd_dumpclients *dumpp);
66 static void nfsrv_freeopenowner(struct nfsstate *stp, int cansleep,
67     NFSPROC_T *p);
68 static int nfsrv_freeopen(struct nfsstate *stp, vnode_t vp, int cansleep,
69     NFSPROC_T *p);
70 static void nfsrv_freelockowner(struct nfsstate *stp, vnode_t vp, int cansleep,
71     NFSPROC_T *p);
72 static void nfsrv_freeallnfslocks(struct nfsstate *stp, vnode_t vp,
73     int cansleep, NFSPROC_T *p);
74 static void nfsrv_freenfslock(struct nfslock *lop);
75 static void nfsrv_freenfslockfile(struct nfslockfile *lfp);
76 static void nfsrv_freedeleg(struct nfsstate *);
77 static int nfsrv_getstate(struct nfsclient *clp, nfsv4stateid_t *stateidp,
78     u_int32_t flags, struct nfsstate **stpp);
79 static void nfsrv_getowner(struct nfsstatehead *hp, struct nfsstate *new_stp,
80     struct nfsstate **stpp);
81 static int nfsrv_getlockfh(vnode_t vp, u_short flags,
82     struct nfslockfile **new_lfpp, fhandle_t *nfhp, NFSPROC_T *p);
83 static int nfsrv_getlockfile(u_short flags, struct nfslockfile **new_lfpp,
84     struct nfslockfile **lfpp, fhandle_t *nfhp, int lockit);
85 static void nfsrv_insertlock(struct nfslock *new_lop,
86     struct nfslock *insert_lop, struct nfsstate *stp, struct nfslockfile *lfp);
87 static void nfsrv_updatelock(struct nfsstate *stp, struct nfslock **new_lopp,
88     struct nfslock **other_lopp, struct nfslockfile *lfp);
89 static int nfsrv_getipnumber(u_char *cp);
90 static int nfsrv_checkrestart(nfsquad_t clientid, u_int32_t flags,
91     nfsv4stateid_t *stateidp, int specialid);
92 static int nfsrv_checkgrace(u_int32_t flags);
93 static int nfsrv_docallback(struct nfsclient *clp, int procnum,
94     nfsv4stateid_t *stateidp, int trunc, fhandle_t *fhp,
95     struct nfsvattr *nap, nfsattrbit_t *attrbitp, NFSPROC_T *p);
96 static u_int32_t nfsrv_nextclientindex(void);
97 static u_int32_t nfsrv_nextstateindex(struct nfsclient *clp);
98 static void nfsrv_markstable(struct nfsclient *clp);
99 static int nfsrv_checkstable(struct nfsclient *clp);
100 static int nfsrv_clientconflict(struct nfsclient *clp, int *haslockp, struct
101     vnode *vp, NFSPROC_T *p);
102 static int nfsrv_delegconflict(struct nfsstate *stp, int *haslockp,
103     NFSPROC_T *p, vnode_t vp);
104 static int nfsrv_cleandeleg(vnode_t vp, struct nfslockfile *lfp,
105     struct nfsclient *clp, int *haslockp, NFSPROC_T *p);
106 static int nfsrv_notsamecredname(struct nfsrv_descript *nd,
107     struct nfsclient *clp);
108 static time_t nfsrv_leaseexpiry(void);
109 static void nfsrv_delaydelegtimeout(struct nfsstate *stp);
110 static int nfsrv_checkseqid(struct nfsrv_descript *nd, u_int32_t seqid,
111     struct nfsstate *stp, struct nfsrvcache *op);
112 static int nfsrv_nootherstate(struct nfsstate *stp);
113 static int nfsrv_locallock(vnode_t vp, struct nfslockfile *lfp, int flags,
114     uint64_t first, uint64_t end, struct nfslockconflict *cfp, NFSPROC_T *p);
115 static void nfsrv_localunlock(vnode_t vp, struct nfslockfile *lfp,
116     uint64_t init_first, uint64_t init_end, NFSPROC_T *p);
117 static int nfsrv_dolocal(vnode_t vp, struct nfslockfile *lfp, int flags,
118     int oldflags, uint64_t first, uint64_t end, struct nfslockconflict *cfp,
119     NFSPROC_T *p);
120 static void nfsrv_locallock_rollback(vnode_t vp, struct nfslockfile *lfp,
121     NFSPROC_T *p);
122 static void nfsrv_locallock_commit(struct nfslockfile *lfp, int flags,
123     uint64_t first, uint64_t end);
124 static void nfsrv_locklf(struct nfslockfile *lfp);
125 static void nfsrv_unlocklf(struct nfslockfile *lfp);
126 
127 /*
128  * Scan the client list for a match and either return the current one,
129  * create a new entry or return an error.
130  * If returning a non-error, the clp structure must either be linked into
131  * the client list or free'd.
132  */
133 APPLESTATIC int
134 nfsrv_setclient(struct nfsrv_descript *nd, struct nfsclient **new_clpp,
135     nfsquad_t *clientidp, nfsquad_t *confirmp, NFSPROC_T *p)
136 {
137 	struct nfsclient *clp = NULL, *new_clp = *new_clpp;
138 	int i;
139 	struct nfsstate *stp, *tstp;
140 	struct sockaddr_in *sad, *rad;
141 	int zapit = 0, gotit, hasstate = 0, igotlock;
142 	static u_int64_t confirm_index = 0;
143 
144 	/*
145 	 * Check for state resource limit exceeded.
146 	 */
147 	if (nfsrv_openpluslock > NFSRV_V4STATELIMIT)
148 		return (NFSERR_RESOURCE);
149 
150 	if ((nd->nd_flag & ND_GSS) && nfsrv_nogsscallback)
151 		/*
152 		 * Don't do callbacks for AUTH_GSS.
153 		 * (Since these aren't yet debugged, they might cause the
154 		 *  server to crap out, if they get past the Init call to
155 		 *  the client.)
156 		 */
157 		new_clp->lc_program = 0;
158 
159 	/* Lock out other nfsd threads */
160 	NFSLOCKV4ROOTMUTEX();
161 	nfsv4_relref(&nfsv4rootfs_lock);
162 	do {
163 		igotlock = nfsv4_lock(&nfsv4rootfs_lock, 1, NULL,
164 		    NFSV4ROOTLOCKMUTEXPTR);
165 	} while (!igotlock);
166 	NFSUNLOCKV4ROOTMUTEX();
167 
168 	/*
169 	 * Search for a match in the client list.
170 	 */
171 	gotit = i = 0;
172 	while (i < NFSCLIENTHASHSIZE && !gotit) {
173 	    LIST_FOREACH(clp, &nfsclienthash[i], lc_hash) {
174 		if (new_clp->lc_idlen == clp->lc_idlen &&
175 		    !NFSBCMP(new_clp->lc_id, clp->lc_id, clp->lc_idlen)) {
176 			gotit = 1;
177 			break;
178 		}
179 	    }
180 	    i++;
181 	}
182 	if (!gotit ||
183 	    (clp->lc_flags & (LCL_NEEDSCONFIRM | LCL_ADMINREVOKED))) {
184 		/*
185 		 * Get rid of the old one.
186 		 */
187 		if (i != NFSCLIENTHASHSIZE) {
188 			LIST_REMOVE(clp, lc_hash);
189 			nfsrv_cleanclient(clp, p);
190 			nfsrv_freedeleglist(&clp->lc_deleg);
191 			nfsrv_freedeleglist(&clp->lc_olddeleg);
192 			zapit = 1;
193 		}
194 		/*
195 		 * Add it after assigning a client id to it.
196 		 */
197 		new_clp->lc_flags |= LCL_NEEDSCONFIRM;
198 		confirmp->qval = new_clp->lc_confirm.qval = ++confirm_index;
199 		clientidp->lval[0] = new_clp->lc_clientid.lval[0] =
200 		    (u_int32_t)nfsrvboottime;
201 		clientidp->lval[1] = new_clp->lc_clientid.lval[1] =
202 		    nfsrv_nextclientindex();
203 		new_clp->lc_stateindex = 0;
204 		new_clp->lc_statemaxindex = 0;
205 		new_clp->lc_cbref = 0;
206 		new_clp->lc_expiry = nfsrv_leaseexpiry();
207 		LIST_INIT(&new_clp->lc_open);
208 		LIST_INIT(&new_clp->lc_deleg);
209 		LIST_INIT(&new_clp->lc_olddeleg);
210 		for (i = 0; i < NFSSTATEHASHSIZE; i++)
211 			LIST_INIT(&new_clp->lc_stateid[i]);
212 		LIST_INSERT_HEAD(NFSCLIENTHASH(new_clp->lc_clientid), new_clp,
213 		    lc_hash);
214 		newnfsstats.srvclients++;
215 		nfsrv_openpluslock++;
216 		nfsrv_clients++;
217 		NFSLOCKV4ROOTMUTEX();
218 		nfsv4_unlock(&nfsv4rootfs_lock, 1);
219 		NFSUNLOCKV4ROOTMUTEX();
220 		if (zapit)
221 			nfsrv_zapclient(clp, p);
222 		*new_clpp = NULL;
223 		return (0);
224 	}
225 
226 	/*
227 	 * Now, handle the cases where the id is already issued.
228 	 */
229 	if (nfsrv_notsamecredname(nd, clp)) {
230 	    /*
231 	     * Check to see if there is expired state that should go away.
232 	     */
233 	    if (clp->lc_expiry < NFSD_MONOSEC &&
234 	        (!LIST_EMPTY(&clp->lc_open) || !LIST_EMPTY(&clp->lc_deleg))) {
235 		nfsrv_cleanclient(clp, p);
236 		nfsrv_freedeleglist(&clp->lc_deleg);
237 	    }
238 
239 	    /*
240 	     * If there is outstanding state, then reply NFSERR_CLIDINUSE per
241 	     * RFC3530 Sec. 8.1.2 last para.
242 	     */
243 	    if (!LIST_EMPTY(&clp->lc_deleg)) {
244 		hasstate = 1;
245 	    } else if (LIST_EMPTY(&clp->lc_open)) {
246 		hasstate = 0;
247 	    } else {
248 		hasstate = 0;
249 		/* Look for an Open on the OpenOwner */
250 		LIST_FOREACH(stp, &clp->lc_open, ls_list) {
251 		    if (!LIST_EMPTY(&stp->ls_open)) {
252 			hasstate = 1;
253 			break;
254 		    }
255 		}
256 	    }
257 	    if (hasstate) {
258 		/*
259 		 * If the uid doesn't match, return NFSERR_CLIDINUSE after
260 		 * filling out the correct ipaddr and portnum.
261 		 */
262 		sad = NFSSOCKADDR(new_clp->lc_req.nr_nam, struct sockaddr_in *);
263 		rad = NFSSOCKADDR(clp->lc_req.nr_nam, struct sockaddr_in *);
264 		sad->sin_addr.s_addr = rad->sin_addr.s_addr;
265 		sad->sin_port = rad->sin_port;
266 		NFSLOCKV4ROOTMUTEX();
267 		nfsv4_unlock(&nfsv4rootfs_lock, 1);
268 		NFSUNLOCKV4ROOTMUTEX();
269 		return (NFSERR_CLIDINUSE);
270 	    }
271 	}
272 
273 	if (NFSBCMP(new_clp->lc_verf, clp->lc_verf, NFSX_VERF)) {
274 		/*
275 		 * If the verifier has changed, the client has rebooted
276 		 * and a new client id is issued. The old state info
277 		 * can be thrown away once the SETCLIENTID_CONFIRM occurs.
278 		 */
279 		LIST_REMOVE(clp, lc_hash);
280 		new_clp->lc_flags |= LCL_NEEDSCONFIRM;
281 		confirmp->qval = new_clp->lc_confirm.qval = ++confirm_index;
282 		clientidp->lval[0] = new_clp->lc_clientid.lval[0] =
283 		    nfsrvboottime;
284 		clientidp->lval[1] = new_clp->lc_clientid.lval[1] =
285 		    nfsrv_nextclientindex();
286 		new_clp->lc_stateindex = 0;
287 		new_clp->lc_statemaxindex = 0;
288 		new_clp->lc_cbref = 0;
289 		new_clp->lc_expiry = nfsrv_leaseexpiry();
290 
291 		/*
292 		 * Save the state until confirmed.
293 		 */
294 		LIST_NEWHEAD(&new_clp->lc_open, &clp->lc_open, ls_list);
295 		LIST_FOREACH(tstp, &new_clp->lc_open, ls_list)
296 			tstp->ls_clp = new_clp;
297 		LIST_NEWHEAD(&new_clp->lc_deleg, &clp->lc_deleg, ls_list);
298 		LIST_FOREACH(tstp, &new_clp->lc_deleg, ls_list)
299 			tstp->ls_clp = new_clp;
300 		LIST_NEWHEAD(&new_clp->lc_olddeleg, &clp->lc_olddeleg,
301 		    ls_list);
302 		LIST_FOREACH(tstp, &new_clp->lc_olddeleg, ls_list)
303 			tstp->ls_clp = new_clp;
304 		for (i = 0; i < NFSSTATEHASHSIZE; i++) {
305 			LIST_NEWHEAD(&new_clp->lc_stateid[i],
306 			    &clp->lc_stateid[i], ls_hash);
307 			LIST_FOREACH(tstp, &new_clp->lc_stateid[i], ls_list)
308 				tstp->ls_clp = new_clp;
309 		}
310 		LIST_INSERT_HEAD(NFSCLIENTHASH(new_clp->lc_clientid), new_clp,
311 		    lc_hash);
312 		newnfsstats.srvclients++;
313 		nfsrv_openpluslock++;
314 		nfsrv_clients++;
315 		NFSLOCKV4ROOTMUTEX();
316 		nfsv4_unlock(&nfsv4rootfs_lock, 1);
317 		NFSUNLOCKV4ROOTMUTEX();
318 
319 		/*
320 		 * Must wait until any outstanding callback on the old clp
321 		 * completes.
322 		 */
323 		while (clp->lc_cbref) {
324 			clp->lc_flags |= LCL_WAKEUPWANTED;
325 			(void) tsleep((caddr_t)clp, PZERO - 1,
326 			    "nfsd clp", 10 * hz);
327 		}
328 		nfsrv_zapclient(clp, p);
329 		*new_clpp = NULL;
330 		return (0);
331 	}
332 	/*
333 	 * id and verifier match, so update the net address info
334 	 * and get rid of any existing callback authentication
335 	 * handle, so a new one will be acquired.
336 	 */
337 	LIST_REMOVE(clp, lc_hash);
338 	new_clp->lc_flags |= (LCL_NEEDSCONFIRM | LCL_DONTCLEAN);
339 	new_clp->lc_expiry = nfsrv_leaseexpiry();
340 	confirmp->qval = new_clp->lc_confirm.qval = ++confirm_index;
341 	clientidp->lval[0] = new_clp->lc_clientid.lval[0] =
342 	    clp->lc_clientid.lval[0];
343 	clientidp->lval[1] = new_clp->lc_clientid.lval[1] =
344 	    clp->lc_clientid.lval[1];
345 	new_clp->lc_delegtime = clp->lc_delegtime;
346 	new_clp->lc_stateindex = clp->lc_stateindex;
347 	new_clp->lc_statemaxindex = clp->lc_statemaxindex;
348 	new_clp->lc_cbref = 0;
349 	LIST_NEWHEAD(&new_clp->lc_open, &clp->lc_open, ls_list);
350 	LIST_FOREACH(tstp, &new_clp->lc_open, ls_list)
351 		tstp->ls_clp = new_clp;
352 	LIST_NEWHEAD(&new_clp->lc_deleg, &clp->lc_deleg, ls_list);
353 	LIST_FOREACH(tstp, &new_clp->lc_deleg, ls_list)
354 		tstp->ls_clp = new_clp;
355 	LIST_NEWHEAD(&new_clp->lc_olddeleg, &clp->lc_olddeleg, ls_list);
356 	LIST_FOREACH(tstp, &new_clp->lc_olddeleg, ls_list)
357 		tstp->ls_clp = new_clp;
358 	for (i = 0; i < NFSSTATEHASHSIZE; i++) {
359 		LIST_NEWHEAD(&new_clp->lc_stateid[i], &clp->lc_stateid[i],
360 		    ls_hash);
361 		LIST_FOREACH(tstp, &new_clp->lc_stateid[i], ls_list)
362 			tstp->ls_clp = new_clp;
363 	}
364 	LIST_INSERT_HEAD(NFSCLIENTHASH(new_clp->lc_clientid), new_clp,
365 	    lc_hash);
366 	newnfsstats.srvclients++;
367 	nfsrv_openpluslock++;
368 	nfsrv_clients++;
369 	NFSLOCKV4ROOTMUTEX();
370 	nfsv4_unlock(&nfsv4rootfs_lock, 1);
371 	NFSUNLOCKV4ROOTMUTEX();
372 
373 	/*
374 	 * Must wait until any outstanding callback on the old clp
375 	 * completes.
376 	 */
377 	while (clp->lc_cbref) {
378 		clp->lc_flags |= LCL_WAKEUPWANTED;
379 		(void) tsleep((caddr_t)clp, PZERO - 1, "nfsd clp", 10 * hz);
380 	}
381 	nfsrv_zapclient(clp, p);
382 	*new_clpp = NULL;
383 	return (0);
384 }
385 
386 /*
387  * Check to see if the client id exists and optionally confirm it.
388  */
389 APPLESTATIC int
390 nfsrv_getclient(nfsquad_t clientid, int opflags, struct nfsclient **clpp,
391     nfsquad_t confirm, struct nfsrv_descript *nd, NFSPROC_T *p)
392 {
393 	struct nfsclient *clp;
394 	struct nfsstate *stp;
395 	int i;
396 	struct nfsclienthashhead *hp;
397 	int error = 0, igotlock, doneok;
398 
399 	if (clpp)
400 		*clpp = NULL;
401 	if (nfsrvboottime != clientid.lval[0])
402 		return (NFSERR_STALECLIENTID);
403 
404 	/*
405 	 * If called with opflags == CLOPS_RENEW, the State Lock is
406 	 * already held. Otherwise, we need to get either that or,
407 	 * for the case of Confirm, lock out the nfsd threads.
408 	 */
409 	if (opflags & CLOPS_CONFIRM) {
410 		NFSLOCKV4ROOTMUTEX();
411 		nfsv4_relref(&nfsv4rootfs_lock);
412 		do {
413 			igotlock = nfsv4_lock(&nfsv4rootfs_lock, 1, NULL,
414 			    NFSV4ROOTLOCKMUTEXPTR);
415 		} while (!igotlock);
416 		NFSUNLOCKV4ROOTMUTEX();
417 	} else if (opflags != CLOPS_RENEW) {
418 		NFSLOCKSTATE();
419 	}
420 
421 	hp = NFSCLIENTHASH(clientid);
422 	LIST_FOREACH(clp, hp, lc_hash) {
423 		if (clp->lc_clientid.lval[1] == clientid.lval[1])
424 			break;
425 	}
426 	if (clp == LIST_END(hp)) {
427 		if (opflags & CLOPS_CONFIRM)
428 			error = NFSERR_STALECLIENTID;
429 		else
430 			error = NFSERR_EXPIRED;
431 	} else if (clp->lc_flags & LCL_ADMINREVOKED) {
432 		/*
433 		 * If marked admin revoked, just return the error.
434 		 */
435 		error = NFSERR_ADMINREVOKED;
436 	}
437 	if (error) {
438 		if (opflags & CLOPS_CONFIRM) {
439 			NFSLOCKV4ROOTMUTEX();
440 			nfsv4_unlock(&nfsv4rootfs_lock, 1);
441 			NFSUNLOCKV4ROOTMUTEX();
442 		} else if (opflags != CLOPS_RENEW) {
443 			NFSUNLOCKSTATE();
444 		}
445 		return (error);
446 	}
447 
448 	/*
449 	 * Perform any operations specified by the opflags.
450 	 */
451 	if (opflags & CLOPS_CONFIRM) {
452 		if (clp->lc_confirm.qval != confirm.qval)
453 			error = NFSERR_STALECLIENTID;
454 		else if (nfsrv_notsamecredname(nd, clp))
455 			error = NFSERR_CLIDINUSE;
456 
457 		if (!error) {
458 		    if ((clp->lc_flags & (LCL_NEEDSCONFIRM | LCL_DONTCLEAN)) ==
459 			LCL_NEEDSCONFIRM) {
460 			/*
461 			 * Hang onto the delegations (as old delegations)
462 			 * for an Open with CLAIM_DELEGATE_PREV unless in
463 			 * grace, but get rid of the rest of the state.
464 			 */
465 			nfsrv_cleanclient(clp, p);
466 			nfsrv_freedeleglist(&clp->lc_olddeleg);
467 			if (nfsrv_checkgrace(0)) {
468 			    /* In grace, so just delete delegations */
469 			    nfsrv_freedeleglist(&clp->lc_deleg);
470 			} else {
471 			    LIST_FOREACH(stp, &clp->lc_deleg, ls_list)
472 				stp->ls_flags |= NFSLCK_OLDDELEG;
473 			    clp->lc_delegtime = NFSD_MONOSEC +
474 				nfsrv_lease + NFSRV_LEASEDELTA;
475 			    LIST_NEWHEAD(&clp->lc_olddeleg, &clp->lc_deleg,
476 				ls_list);
477 			}
478 		    }
479 		    clp->lc_flags &= ~(LCL_NEEDSCONFIRM | LCL_DONTCLEAN);
480 		    if (clp->lc_program)
481 			clp->lc_flags |= LCL_NEEDSCBNULL;
482 		}
483 	} else if (clp->lc_flags & LCL_NEEDSCONFIRM) {
484 		error = NFSERR_EXPIRED;
485 	}
486 
487 	/*
488 	 * If called by the Renew Op, we must check the principal.
489 	 */
490 	if (!error && (opflags & CLOPS_RENEWOP)) {
491 	    if (nfsrv_notsamecredname(nd, clp)) {
492 		doneok = 0;
493 		for (i = 0; i < NFSSTATEHASHSIZE && doneok == 0; i++) {
494 		    LIST_FOREACH(stp, &clp->lc_stateid[i], ls_hash) {
495 			if ((stp->ls_flags & NFSLCK_OPEN) &&
496 			    stp->ls_uid == nd->nd_cred->cr_uid) {
497 				doneok = 1;
498 				break;
499 			}
500 		    }
501 		}
502 		if (!doneok)
503 			error = NFSERR_ACCES;
504 	    }
505 	    if (!error && (clp->lc_flags & LCL_CBDOWN))
506 		error = NFSERR_CBPATHDOWN;
507 	}
508 	if ((!error || error == NFSERR_CBPATHDOWN) &&
509 	     (opflags & CLOPS_RENEW)) {
510 		clp->lc_expiry = nfsrv_leaseexpiry();
511 	}
512 	if (opflags & CLOPS_CONFIRM) {
513 		NFSLOCKV4ROOTMUTEX();
514 		nfsv4_unlock(&nfsv4rootfs_lock, 1);
515 		NFSUNLOCKV4ROOTMUTEX();
516 	} else if (opflags != CLOPS_RENEW) {
517 		NFSUNLOCKSTATE();
518 	}
519 	if (clpp)
520 		*clpp = clp;
521 	return (error);
522 }
523 
524 /*
525  * Called from the new nfssvc syscall to admin revoke a clientid.
526  * Returns 0 for success, error otherwise.
527  */
528 APPLESTATIC int
529 nfsrv_adminrevoke(struct nfsd_clid *revokep, NFSPROC_T *p)
530 {
531 	struct nfsclient *clp = NULL;
532 	int i;
533 	int gotit, igotlock;
534 
535 	/*
536 	 * First, lock out the nfsd so that state won't change while the
537 	 * revocation record is being written to the stable storage restart
538 	 * file.
539 	 */
540 	NFSLOCKV4ROOTMUTEX();
541 	do {
542 		igotlock = nfsv4_lock(&nfsv4rootfs_lock, 1, NULL,
543 		    NFSV4ROOTLOCKMUTEXPTR);
544 	} while (!igotlock);
545 	NFSUNLOCKV4ROOTMUTEX();
546 
547 	/*
548 	 * Search for a match in the client list.
549 	 */
550 	gotit = i = 0;
551 	while (i < NFSCLIENTHASHSIZE && !gotit) {
552 	    LIST_FOREACH(clp, &nfsclienthash[i], lc_hash) {
553 		if (revokep->nclid_idlen == clp->lc_idlen &&
554 		    !NFSBCMP(revokep->nclid_id, clp->lc_id, clp->lc_idlen)) {
555 			gotit = 1;
556 			break;
557 		}
558 	    }
559 	    i++;
560 	}
561 	if (!gotit) {
562 		NFSLOCKV4ROOTMUTEX();
563 		nfsv4_unlock(&nfsv4rootfs_lock, 0);
564 		NFSUNLOCKV4ROOTMUTEX();
565 		return (EPERM);
566 	}
567 
568 	/*
569 	 * Now, write out the revocation record
570 	 */
571 	nfsrv_writestable(clp->lc_id, clp->lc_idlen, NFSNST_REVOKE, p);
572 
573 	/*
574 	 * and clear out the state, marking the clientid revoked.
575 	 */
576 	clp->lc_flags &= ~LCL_CALLBACKSON;
577 	clp->lc_flags |= LCL_ADMINREVOKED;
578 	nfsrv_cleanclient(clp, p);
579 	nfsrv_freedeleglist(&clp->lc_deleg);
580 	nfsrv_freedeleglist(&clp->lc_olddeleg);
581 	NFSLOCKV4ROOTMUTEX();
582 	nfsv4_unlock(&nfsv4rootfs_lock, 0);
583 	NFSUNLOCKV4ROOTMUTEX();
584 	return (0);
585 }
586 
587 /*
588  * Dump out stats for all clients. Called from nfssvc(2), that is used
589  * newnfsstats.
590  */
591 APPLESTATIC void
592 nfsrv_dumpclients(struct nfsd_dumpclients *dumpp, int maxcnt)
593 {
594 	struct nfsclient *clp;
595 	int i = 0, cnt = 0;
596 
597 	/*
598 	 * First, get a reference on the nfsv4rootfs_lock so that an
599 	 * exclusive lock cannot be acquired while dumping the clients.
600 	 */
601 	NFSLOCKV4ROOTMUTEX();
602 	nfsv4_getref(&nfsv4rootfs_lock, NULL, NFSV4ROOTLOCKMUTEXPTR);
603 	NFSUNLOCKV4ROOTMUTEX();
604 	NFSLOCKSTATE();
605 	/*
606 	 * Rattle through the client lists until done.
607 	 */
608 	while (i < NFSCLIENTHASHSIZE && cnt < maxcnt) {
609 	    clp = LIST_FIRST(&nfsclienthash[i]);
610 	    while (clp != LIST_END(&nfsclienthash[i]) && cnt < maxcnt) {
611 		nfsrv_dumpaclient(clp, &dumpp[cnt]);
612 		cnt++;
613 		clp = LIST_NEXT(clp, lc_hash);
614 	    }
615 	    i++;
616 	}
617 	if (cnt < maxcnt)
618 	    dumpp[cnt].ndcl_clid.nclid_idlen = 0;
619 	NFSUNLOCKSTATE();
620 	NFSLOCKV4ROOTMUTEX();
621 	nfsv4_relref(&nfsv4rootfs_lock);
622 	NFSUNLOCKV4ROOTMUTEX();
623 }
624 
625 /*
626  * Dump stats for a client. Must be called with the NFSSTATELOCK and spl'd.
627  */
628 static void
629 nfsrv_dumpaclient(struct nfsclient *clp, struct nfsd_dumpclients *dumpp)
630 {
631 	struct nfsstate *stp, *openstp, *lckownstp;
632 	struct nfslock *lop;
633 	struct sockaddr *sad;
634 	struct sockaddr_in *rad;
635 	struct sockaddr_in6 *rad6;
636 
637 	dumpp->ndcl_nopenowners = dumpp->ndcl_nlockowners = 0;
638 	dumpp->ndcl_nopens = dumpp->ndcl_nlocks = 0;
639 	dumpp->ndcl_ndelegs = dumpp->ndcl_nolddelegs = 0;
640 	dumpp->ndcl_flags = clp->lc_flags;
641 	dumpp->ndcl_clid.nclid_idlen = clp->lc_idlen;
642 	NFSBCOPY(clp->lc_id, dumpp->ndcl_clid.nclid_id, clp->lc_idlen);
643 	sad = NFSSOCKADDR(clp->lc_req.nr_nam, struct sockaddr *);
644 	dumpp->ndcl_addrfam = sad->sa_family;
645 	if (sad->sa_family == AF_INET) {
646 		rad = (struct sockaddr_in *)sad;
647 		dumpp->ndcl_cbaddr.sin_addr = rad->sin_addr;
648 	} else {
649 		rad6 = (struct sockaddr_in6 *)sad;
650 		dumpp->ndcl_cbaddr.sin6_addr = rad6->sin6_addr;
651 	}
652 
653 	/*
654 	 * Now, scan the state lists and total up the opens and locks.
655 	 */
656 	LIST_FOREACH(stp, &clp->lc_open, ls_list) {
657 	    dumpp->ndcl_nopenowners++;
658 	    LIST_FOREACH(openstp, &stp->ls_open, ls_list) {
659 		dumpp->ndcl_nopens++;
660 		LIST_FOREACH(lckownstp, &openstp->ls_open, ls_list) {
661 		    dumpp->ndcl_nlockowners++;
662 		    LIST_FOREACH(lop, &lckownstp->ls_lock, lo_lckowner) {
663 			dumpp->ndcl_nlocks++;
664 		    }
665 		}
666 	    }
667 	}
668 
669 	/*
670 	 * and the delegation lists.
671 	 */
672 	LIST_FOREACH(stp, &clp->lc_deleg, ls_list) {
673 	    dumpp->ndcl_ndelegs++;
674 	}
675 	LIST_FOREACH(stp, &clp->lc_olddeleg, ls_list) {
676 	    dumpp->ndcl_nolddelegs++;
677 	}
678 }
679 
680 /*
681  * Dump out lock stats for a file.
682  */
683 APPLESTATIC void
684 nfsrv_dumplocks(vnode_t vp, struct nfsd_dumplocks *ldumpp, int maxcnt,
685     NFSPROC_T *p)
686 {
687 	struct nfsstate *stp;
688 	struct nfslock *lop;
689 	int cnt = 0;
690 	struct nfslockfile *lfp;
691 	struct sockaddr *sad;
692 	struct sockaddr_in *rad;
693 	struct sockaddr_in6 *rad6;
694 	int ret;
695 	fhandle_t nfh;
696 
697 	ret = nfsrv_getlockfh(vp, 0, NULL, &nfh, p);
698 	/*
699 	 * First, get a reference on the nfsv4rootfs_lock so that an
700 	 * exclusive lock on it cannot be acquired while dumping the locks.
701 	 */
702 	NFSLOCKV4ROOTMUTEX();
703 	nfsv4_getref(&nfsv4rootfs_lock, NULL, NFSV4ROOTLOCKMUTEXPTR);
704 	NFSUNLOCKV4ROOTMUTEX();
705 	NFSLOCKSTATE();
706 	if (!ret)
707 		ret = nfsrv_getlockfile(0, NULL, &lfp, &nfh, 0);
708 	if (ret) {
709 		ldumpp[0].ndlck_clid.nclid_idlen = 0;
710 		NFSUNLOCKSTATE();
711 		NFSLOCKV4ROOTMUTEX();
712 		nfsv4_relref(&nfsv4rootfs_lock);
713 		NFSUNLOCKV4ROOTMUTEX();
714 		return;
715 	}
716 
717 	/*
718 	 * For each open share on file, dump it out.
719 	 */
720 	stp = LIST_FIRST(&lfp->lf_open);
721 	while (stp != LIST_END(&lfp->lf_open) && cnt < maxcnt) {
722 		ldumpp[cnt].ndlck_flags = stp->ls_flags;
723 		ldumpp[cnt].ndlck_stateid.seqid = stp->ls_stateid.seqid;
724 		ldumpp[cnt].ndlck_stateid.other[0] = stp->ls_stateid.other[0];
725 		ldumpp[cnt].ndlck_stateid.other[1] = stp->ls_stateid.other[1];
726 		ldumpp[cnt].ndlck_stateid.other[2] = stp->ls_stateid.other[2];
727 		ldumpp[cnt].ndlck_owner.nclid_idlen =
728 		    stp->ls_openowner->ls_ownerlen;
729 		NFSBCOPY(stp->ls_openowner->ls_owner,
730 		    ldumpp[cnt].ndlck_owner.nclid_id,
731 		    stp->ls_openowner->ls_ownerlen);
732 		ldumpp[cnt].ndlck_clid.nclid_idlen = stp->ls_clp->lc_idlen;
733 		NFSBCOPY(stp->ls_clp->lc_id, ldumpp[cnt].ndlck_clid.nclid_id,
734 		    stp->ls_clp->lc_idlen);
735 		sad=NFSSOCKADDR(stp->ls_clp->lc_req.nr_nam, struct sockaddr *);
736 		ldumpp[cnt].ndlck_addrfam = sad->sa_family;
737 		if (sad->sa_family == AF_INET) {
738 			rad = (struct sockaddr_in *)sad;
739 			ldumpp[cnt].ndlck_cbaddr.sin_addr = rad->sin_addr;
740 		} else {
741 			rad6 = (struct sockaddr_in6 *)sad;
742 			ldumpp[cnt].ndlck_cbaddr.sin6_addr = rad6->sin6_addr;
743 		}
744 		stp = LIST_NEXT(stp, ls_file);
745 		cnt++;
746 	}
747 
748 	/*
749 	 * and all locks.
750 	 */
751 	lop = LIST_FIRST(&lfp->lf_lock);
752 	while (lop != LIST_END(&lfp->lf_lock) && cnt < maxcnt) {
753 		stp = lop->lo_stp;
754 		ldumpp[cnt].ndlck_flags = lop->lo_flags;
755 		ldumpp[cnt].ndlck_first = lop->lo_first;
756 		ldumpp[cnt].ndlck_end = lop->lo_end;
757 		ldumpp[cnt].ndlck_stateid.seqid = stp->ls_stateid.seqid;
758 		ldumpp[cnt].ndlck_stateid.other[0] = stp->ls_stateid.other[0];
759 		ldumpp[cnt].ndlck_stateid.other[1] = stp->ls_stateid.other[1];
760 		ldumpp[cnt].ndlck_stateid.other[2] = stp->ls_stateid.other[2];
761 		ldumpp[cnt].ndlck_owner.nclid_idlen = stp->ls_ownerlen;
762 		NFSBCOPY(stp->ls_owner, ldumpp[cnt].ndlck_owner.nclid_id,
763 		    stp->ls_ownerlen);
764 		ldumpp[cnt].ndlck_clid.nclid_idlen = stp->ls_clp->lc_idlen;
765 		NFSBCOPY(stp->ls_clp->lc_id, ldumpp[cnt].ndlck_clid.nclid_id,
766 		    stp->ls_clp->lc_idlen);
767 		sad=NFSSOCKADDR(stp->ls_clp->lc_req.nr_nam, struct sockaddr *);
768 		ldumpp[cnt].ndlck_addrfam = sad->sa_family;
769 		if (sad->sa_family == AF_INET) {
770 			rad = (struct sockaddr_in *)sad;
771 			ldumpp[cnt].ndlck_cbaddr.sin_addr = rad->sin_addr;
772 		} else {
773 			rad6 = (struct sockaddr_in6 *)sad;
774 			ldumpp[cnt].ndlck_cbaddr.sin6_addr = rad6->sin6_addr;
775 		}
776 		lop = LIST_NEXT(lop, lo_lckfile);
777 		cnt++;
778 	}
779 
780 	/*
781 	 * and the delegations.
782 	 */
783 	stp = LIST_FIRST(&lfp->lf_deleg);
784 	while (stp != LIST_END(&lfp->lf_deleg) && cnt < maxcnt) {
785 		ldumpp[cnt].ndlck_flags = stp->ls_flags;
786 		ldumpp[cnt].ndlck_stateid.seqid = stp->ls_stateid.seqid;
787 		ldumpp[cnt].ndlck_stateid.other[0] = stp->ls_stateid.other[0];
788 		ldumpp[cnt].ndlck_stateid.other[1] = stp->ls_stateid.other[1];
789 		ldumpp[cnt].ndlck_stateid.other[2] = stp->ls_stateid.other[2];
790 		ldumpp[cnt].ndlck_owner.nclid_idlen = 0;
791 		ldumpp[cnt].ndlck_clid.nclid_idlen = stp->ls_clp->lc_idlen;
792 		NFSBCOPY(stp->ls_clp->lc_id, ldumpp[cnt].ndlck_clid.nclid_id,
793 		    stp->ls_clp->lc_idlen);
794 		sad=NFSSOCKADDR(stp->ls_clp->lc_req.nr_nam, struct sockaddr *);
795 		ldumpp[cnt].ndlck_addrfam = sad->sa_family;
796 		if (sad->sa_family == AF_INET) {
797 			rad = (struct sockaddr_in *)sad;
798 			ldumpp[cnt].ndlck_cbaddr.sin_addr = rad->sin_addr;
799 		} else {
800 			rad6 = (struct sockaddr_in6 *)sad;
801 			ldumpp[cnt].ndlck_cbaddr.sin6_addr = rad6->sin6_addr;
802 		}
803 		stp = LIST_NEXT(stp, ls_file);
804 		cnt++;
805 	}
806 
807 	/*
808 	 * If list isn't full, mark end of list by setting the client name
809 	 * to zero length.
810 	 */
811 	if (cnt < maxcnt)
812 		ldumpp[cnt].ndlck_clid.nclid_idlen = 0;
813 	NFSUNLOCKSTATE();
814 	NFSLOCKV4ROOTMUTEX();
815 	nfsv4_relref(&nfsv4rootfs_lock);
816 	NFSUNLOCKV4ROOTMUTEX();
817 }
818 
819 /*
820  * Server timer routine. It can scan any linked list, so long
821  * as it holds the spin/mutex lock and there is no exclusive lock on
822  * nfsv4rootfs_lock.
823  * (For OpenBSD, a kthread is ok. For FreeBSD, I think it is ok
824  *  to do this from a callout, since the spin locks work. For
825  *  Darwin, I'm not sure what will work correctly yet.)
826  * Should be called once per second.
827  */
828 APPLESTATIC void
829 nfsrv_servertimer(void)
830 {
831 	struct nfsclient *clp, *nclp;
832 	struct nfsstate *stp, *nstp;
833 	int got_ref, i;
834 
835 	/*
836 	 * Make sure nfsboottime is set. This is used by V3 as well
837 	 * as V4. Note that nfsboottime is not nfsrvboottime, which is
838 	 * only used by the V4 server for leases.
839 	 */
840 	if (nfsboottime.tv_sec == 0)
841 		NFSSETBOOTTIME(nfsboottime);
842 
843 	/*
844 	 * If server hasn't started yet, just return.
845 	 */
846 	NFSLOCKSTATE();
847 	if (nfsrv_stablefirst.nsf_eograce == 0) {
848 		NFSUNLOCKSTATE();
849 		return;
850 	}
851 	if (!(nfsrv_stablefirst.nsf_flags & NFSNSF_UPDATEDONE)) {
852 		if (!(nfsrv_stablefirst.nsf_flags & NFSNSF_GRACEOVER) &&
853 		    NFSD_MONOSEC > nfsrv_stablefirst.nsf_eograce)
854 			nfsrv_stablefirst.nsf_flags |=
855 			    (NFSNSF_GRACEOVER | NFSNSF_NEEDLOCK);
856 		NFSUNLOCKSTATE();
857 		return;
858 	}
859 
860 	/*
861 	 * Try and get a reference count on the nfsv4rootfs_lock so that
862 	 * no nfsd thread can acquire an exclusive lock on it before this
863 	 * call is done. If it is already exclusively locked, just return.
864 	 */
865 	NFSLOCKV4ROOTMUTEX();
866 	got_ref = nfsv4_getref_nonblock(&nfsv4rootfs_lock);
867 	NFSUNLOCKV4ROOTMUTEX();
868 	if (got_ref == 0) {
869 		NFSUNLOCKSTATE();
870 		return;
871 	}
872 
873 	/*
874 	 * For each client...
875 	 */
876 	for (i = 0; i < NFSCLIENTHASHSIZE; i++) {
877 	    clp = LIST_FIRST(&nfsclienthash[i]);
878 	    while (clp != LIST_END(&nfsclienthash[i])) {
879 		nclp = LIST_NEXT(clp, lc_hash);
880 		if (!(clp->lc_flags & LCL_EXPIREIT)) {
881 		    if (((clp->lc_expiry + NFSRV_STALELEASE) < NFSD_MONOSEC
882 			 && ((LIST_EMPTY(&clp->lc_deleg)
883 			      && LIST_EMPTY(&clp->lc_open)) ||
884 			     nfsrv_clients > nfsrv_clienthighwater)) ||
885 			(clp->lc_expiry + NFSRV_MOULDYLEASE) < NFSD_MONOSEC ||
886 			(clp->lc_expiry < NFSD_MONOSEC &&
887 			 (nfsrv_openpluslock * 10 / 9) > NFSRV_V4STATELIMIT)) {
888 			/*
889 			 * Lease has expired several nfsrv_lease times ago:
890 			 * PLUS
891 			 *    - no state is associated with it
892 			 *    OR
893 			 *    - above high water mark for number of clients
894 			 *      (nfsrv_clienthighwater should be large enough
895 			 *       that this only occurs when clients fail to
896 			 *       use the same nfs_client_id4.id. Maybe somewhat
897 			 *       higher that the maximum number of clients that
898 			 *       will mount this server?)
899 			 * OR
900 			 * Lease has expired a very long time ago
901 			 * OR
902 			 * Lease has expired PLUS the number of opens + locks
903 			 * has exceeded 90% of capacity
904 			 *
905 			 * --> Mark for expiry. The actual expiry will be done
906 			 *     by an nfsd sometime soon.
907 			 */
908 			clp->lc_flags |= LCL_EXPIREIT;
909 			nfsrv_stablefirst.nsf_flags |=
910 			    (NFSNSF_NEEDLOCK | NFSNSF_EXPIREDCLIENT);
911 		    } else {
912 			/*
913 			 * If there are no opens, increment no open tick cnt
914 			 * If time exceeds NFSNOOPEN, mark it to be thrown away
915 			 * otherwise, if there is an open, reset no open time
916 			 * Hopefully, this will avoid excessive re-creation
917 			 * of open owners and subsequent open confirms.
918 			 */
919 			stp = LIST_FIRST(&clp->lc_open);
920 			while (stp != LIST_END(&clp->lc_open)) {
921 				nstp = LIST_NEXT(stp, ls_list);
922 				if (LIST_EMPTY(&stp->ls_open)) {
923 					stp->ls_noopens++;
924 					if (stp->ls_noopens > NFSNOOPEN ||
925 					    (nfsrv_openpluslock * 2) >
926 					    NFSRV_V4STATELIMIT)
927 						nfsrv_stablefirst.nsf_flags |=
928 							NFSNSF_NOOPENS;
929 				} else {
930 					stp->ls_noopens = 0;
931 				}
932 				stp = nstp;
933 			}
934 		    }
935 		}
936 		clp = nclp;
937 	    }
938 	}
939 	NFSUNLOCKSTATE();
940 	NFSLOCKV4ROOTMUTEX();
941 	nfsv4_relref(&nfsv4rootfs_lock);
942 	NFSUNLOCKV4ROOTMUTEX();
943 }
944 
945 /*
946  * The following set of functions free up the various data structures.
947  */
948 /*
949  * Clear out all open/lock state related to this nfsclient.
950  * Caller must hold an exclusive lock on nfsv4rootfs_lock, so that
951  * there are no other active nfsd threads.
952  */
953 APPLESTATIC void
954 nfsrv_cleanclient(struct nfsclient *clp, NFSPROC_T *p)
955 {
956 	struct nfsstate *stp, *nstp;
957 
958 	LIST_FOREACH_SAFE(stp, &clp->lc_open, ls_list, nstp)
959 		nfsrv_freeopenowner(stp, 1, p);
960 }
961 
962 /*
963  * Free a client that has been cleaned. It should also already have been
964  * removed from the lists.
965  * (Just to be safe w.r.t. newnfs_disconnect(), call this function when
966  *  softclock interrupts are enabled.)
967  */
968 APPLESTATIC void
969 nfsrv_zapclient(struct nfsclient *clp, NFSPROC_T *p)
970 {
971 
972 #ifdef notyet
973 	if ((clp->lc_flags & (LCL_GSS | LCL_CALLBACKSON)) ==
974 	     (LCL_GSS | LCL_CALLBACKSON) &&
975 	    (clp->lc_hand.nfsh_flag & NFSG_COMPLETE) &&
976 	    clp->lc_handlelen > 0) {
977 		clp->lc_hand.nfsh_flag &= ~NFSG_COMPLETE;
978 		clp->lc_hand.nfsh_flag |= NFSG_DESTROYED;
979 		(void) nfsrv_docallback(clp, NFSV4PROC_CBNULL,
980 			NULL, 0, NULL, NULL, NULL, p);
981 	}
982 #endif
983 	newnfs_disconnect(&clp->lc_req);
984 	NFSSOCKADDRFREE(clp->lc_req.nr_nam);
985 	NFSFREEMUTEX(&clp->lc_req.nr_mtx);
986 	free((caddr_t)clp, M_NFSDCLIENT);
987 	NFSLOCKSTATE();
988 	newnfsstats.srvclients--;
989 	nfsrv_openpluslock--;
990 	nfsrv_clients--;
991 	NFSUNLOCKSTATE();
992 }
993 
994 /*
995  * Free a list of delegation state structures.
996  * (This function will also free all nfslockfile structures that no
997  *  longer have associated state.)
998  */
999 APPLESTATIC void
1000 nfsrv_freedeleglist(struct nfsstatehead *sthp)
1001 {
1002 	struct nfsstate *stp, *nstp;
1003 
1004 	LIST_FOREACH_SAFE(stp, sthp, ls_list, nstp) {
1005 		nfsrv_freedeleg(stp);
1006 	}
1007 	LIST_INIT(sthp);
1008 }
1009 
1010 /*
1011  * Free up a delegation.
1012  */
1013 static void
1014 nfsrv_freedeleg(struct nfsstate *stp)
1015 {
1016 	struct nfslockfile *lfp;
1017 
1018 	LIST_REMOVE(stp, ls_hash);
1019 	LIST_REMOVE(stp, ls_list);
1020 	LIST_REMOVE(stp, ls_file);
1021 	lfp = stp->ls_lfp;
1022 	if (LIST_EMPTY(&lfp->lf_open) &&
1023 	    LIST_EMPTY(&lfp->lf_lock) && LIST_EMPTY(&lfp->lf_deleg) &&
1024 	    LIST_EMPTY(&lfp->lf_locallock) && LIST_EMPTY(&lfp->lf_rollback) &&
1025 	    lfp->lf_usecount == 0 &&
1026 	    nfsv4_testlock(&lfp->lf_locallock_lck) == 0)
1027 		nfsrv_freenfslockfile(lfp);
1028 	FREE((caddr_t)stp, M_NFSDSTATE);
1029 	newnfsstats.srvdelegates--;
1030 	nfsrv_openpluslock--;
1031 	nfsrv_delegatecnt--;
1032 }
1033 
1034 /*
1035  * This function frees an open owner and all associated opens.
1036  */
1037 static void
1038 nfsrv_freeopenowner(struct nfsstate *stp, int cansleep, NFSPROC_T *p)
1039 {
1040 	struct nfsstate *nstp, *tstp;
1041 
1042 	LIST_REMOVE(stp, ls_list);
1043 	/*
1044 	 * Now, free all associated opens.
1045 	 */
1046 	nstp = LIST_FIRST(&stp->ls_open);
1047 	while (nstp != LIST_END(&stp->ls_open)) {
1048 		tstp = nstp;
1049 		nstp = LIST_NEXT(nstp, ls_list);
1050 		(void) nfsrv_freeopen(tstp, NULL, cansleep, p);
1051 	}
1052 	if (stp->ls_op)
1053 		nfsrvd_derefcache(stp->ls_op);
1054 	FREE((caddr_t)stp, M_NFSDSTATE);
1055 	newnfsstats.srvopenowners--;
1056 	nfsrv_openpluslock--;
1057 }
1058 
1059 /*
1060  * This function frees an open (nfsstate open structure) with all associated
1061  * lock_owners and locks. It also frees the nfslockfile structure iff there
1062  * are no other opens on the file.
1063  * Returns 1 if it free'd the nfslockfile, 0 otherwise.
1064  */
1065 static int
1066 nfsrv_freeopen(struct nfsstate *stp, vnode_t vp, int cansleep, NFSPROC_T *p)
1067 {
1068 	struct nfsstate *nstp, *tstp;
1069 	struct nfslockfile *lfp;
1070 	int ret;
1071 
1072 	LIST_REMOVE(stp, ls_hash);
1073 	LIST_REMOVE(stp, ls_list);
1074 	LIST_REMOVE(stp, ls_file);
1075 
1076 	lfp = stp->ls_lfp;
1077 	/*
1078 	 * Now, free all lockowners associated with this open.
1079 	 */
1080 	LIST_FOREACH_SAFE(tstp, &stp->ls_open, ls_list, nstp)
1081 		nfsrv_freelockowner(tstp, vp, cansleep, p);
1082 
1083 	/*
1084 	 * The nfslockfile is freed here if there are no locks
1085 	 * associated with the open.
1086 	 * If there are locks associated with the open, the
1087 	 * nfslockfile structure can be freed via nfsrv_freelockowner().
1088 	 * Acquire the state mutex to avoid races with calls to
1089 	 * nfsrv_getlockfile().
1090 	 */
1091 	if (cansleep != 0)
1092 		NFSLOCKSTATE();
1093 	if (lfp != NULL && LIST_EMPTY(&lfp->lf_open) &&
1094 	    LIST_EMPTY(&lfp->lf_deleg) && LIST_EMPTY(&lfp->lf_lock) &&
1095 	    LIST_EMPTY(&lfp->lf_locallock) && LIST_EMPTY(&lfp->lf_rollback) &&
1096 	    lfp->lf_usecount == 0 &&
1097 	    (cansleep != 0 || nfsv4_testlock(&lfp->lf_locallock_lck) == 0)) {
1098 		nfsrv_freenfslockfile(lfp);
1099 		ret = 1;
1100 	} else
1101 		ret = 0;
1102 	if (cansleep != 0)
1103 		NFSUNLOCKSTATE();
1104 	FREE((caddr_t)stp, M_NFSDSTATE);
1105 	newnfsstats.srvopens--;
1106 	nfsrv_openpluslock--;
1107 	return (ret);
1108 }
1109 
1110 /*
1111  * Frees a lockowner and all associated locks.
1112  */
1113 static void
1114 nfsrv_freelockowner(struct nfsstate *stp, vnode_t vp, int cansleep,
1115     NFSPROC_T *p)
1116 {
1117 
1118 	LIST_REMOVE(stp, ls_hash);
1119 	LIST_REMOVE(stp, ls_list);
1120 	nfsrv_freeallnfslocks(stp, vp, cansleep, p);
1121 	if (stp->ls_op)
1122 		nfsrvd_derefcache(stp->ls_op);
1123 	FREE((caddr_t)stp, M_NFSDSTATE);
1124 	newnfsstats.srvlockowners--;
1125 	nfsrv_openpluslock--;
1126 }
1127 
1128 /*
1129  * Free all the nfs locks on a lockowner.
1130  */
1131 static void
1132 nfsrv_freeallnfslocks(struct nfsstate *stp, vnode_t vp, int cansleep,
1133     NFSPROC_T *p)
1134 {
1135 	struct nfslock *lop, *nlop;
1136 	struct nfsrollback *rlp, *nrlp;
1137 	struct nfslockfile *lfp = NULL;
1138 	int gottvp = 0;
1139 	vnode_t tvp = NULL;
1140 	uint64_t first, end;
1141 
1142 	lop = LIST_FIRST(&stp->ls_lock);
1143 	while (lop != LIST_END(&stp->ls_lock)) {
1144 		nlop = LIST_NEXT(lop, lo_lckowner);
1145 		/*
1146 		 * Since all locks should be for the same file, lfp should
1147 		 * not change.
1148 		 */
1149 		if (lfp == NULL)
1150 			lfp = lop->lo_lfp;
1151 		else if (lfp != lop->lo_lfp)
1152 			panic("allnfslocks");
1153 		/*
1154 		 * If vp is NULL and cansleep != 0, a vnode must be acquired
1155 		 * from the file handle. This only occurs when called from
1156 		 * nfsrv_cleanclient().
1157 		 */
1158 		if (gottvp == 0) {
1159 			if (nfsrv_dolocallocks == 0)
1160 				tvp = NULL;
1161 			else if (vp == NULL && cansleep != 0)
1162 				tvp = nfsvno_getvp(&lfp->lf_fh);
1163 			else
1164 				tvp = vp;
1165 			gottvp = 1;
1166 		}
1167 
1168 		if (tvp != NULL) {
1169 			if (cansleep == 0)
1170 				panic("allnfs2");
1171 			first = lop->lo_first;
1172 			end = lop->lo_end;
1173 			nfsrv_freenfslock(lop);
1174 			nfsrv_localunlock(tvp, lfp, first, end, p);
1175 			LIST_FOREACH_SAFE(rlp, &lfp->lf_rollback, rlck_list,
1176 			    nrlp)
1177 				free(rlp, M_NFSDROLLBACK);
1178 			LIST_INIT(&lfp->lf_rollback);
1179 		} else
1180 			nfsrv_freenfslock(lop);
1181 		lop = nlop;
1182 	}
1183 	if (vp == NULL && tvp != NULL)
1184 		vput(tvp);
1185 }
1186 
1187 /*
1188  * Free an nfslock structure.
1189  */
1190 static void
1191 nfsrv_freenfslock(struct nfslock *lop)
1192 {
1193 
1194 	if (lop->lo_lckfile.le_prev != NULL) {
1195 		LIST_REMOVE(lop, lo_lckfile);
1196 		newnfsstats.srvlocks--;
1197 		nfsrv_openpluslock--;
1198 	}
1199 	LIST_REMOVE(lop, lo_lckowner);
1200 	FREE((caddr_t)lop, M_NFSDLOCK);
1201 }
1202 
1203 /*
1204  * This function frees an nfslockfile structure.
1205  */
1206 static void
1207 nfsrv_freenfslockfile(struct nfslockfile *lfp)
1208 {
1209 
1210 	LIST_REMOVE(lfp, lf_hash);
1211 	FREE((caddr_t)lfp, M_NFSDLOCKFILE);
1212 }
1213 
1214 /*
1215  * This function looks up an nfsstate structure via stateid.
1216  */
1217 static int
1218 nfsrv_getstate(struct nfsclient *clp, nfsv4stateid_t *stateidp, __unused u_int32_t flags,
1219     struct nfsstate **stpp)
1220 {
1221 	struct nfsstate *stp;
1222 	struct nfsstatehead *hp;
1223 
1224 	*stpp = NULL;
1225 	hp = NFSSTATEHASH(clp, *stateidp);
1226 	LIST_FOREACH(stp, hp, ls_hash) {
1227 		if (!NFSBCMP(stp->ls_stateid.other, stateidp->other,
1228 			NFSX_STATEIDOTHER))
1229 			break;
1230 	}
1231 
1232 	/*
1233 	 * If no state id in list, return NFSERR_BADSTATEID.
1234 	 */
1235 	if (stp == LIST_END(hp))
1236 		return (NFSERR_BADSTATEID);
1237 	*stpp = stp;
1238 	return (0);
1239 }
1240 
1241 /*
1242  * This function gets an nfsstate structure via owner string.
1243  */
1244 static void
1245 nfsrv_getowner(struct nfsstatehead *hp, struct nfsstate *new_stp,
1246     struct nfsstate **stpp)
1247 {
1248 	struct nfsstate *stp;
1249 
1250 	*stpp = NULL;
1251 	LIST_FOREACH(stp, hp, ls_list) {
1252 		if (new_stp->ls_ownerlen == stp->ls_ownerlen &&
1253 		  !NFSBCMP(new_stp->ls_owner,stp->ls_owner,stp->ls_ownerlen)) {
1254 			*stpp = stp;
1255 			return;
1256 		}
1257 	}
1258 }
1259 
1260 /*
1261  * Lock control function called to update lock status.
1262  * Returns 0 upon success, -1 if there is no lock and the flags indicate
1263  * that one isn't to be created and an NFSERR_xxx for other errors.
1264  * The structures new_stp and new_lop are passed in as pointers that should
1265  * be set to NULL if the structure is used and shouldn't be free'd.
1266  * For the NFSLCK_TEST and NFSLCK_CHECK cases, the structures are
1267  * never used and can safely be allocated on the stack. For all other
1268  * cases, *new_stpp and *new_lopp should be malloc'd before the call,
1269  * in case they are used.
1270  */
1271 APPLESTATIC int
1272 nfsrv_lockctrl(vnode_t vp, struct nfsstate **new_stpp,
1273     struct nfslock **new_lopp, struct nfslockconflict *cfp,
1274     nfsquad_t clientid, nfsv4stateid_t *stateidp,
1275     __unused struct nfsexstuff *exp,
1276     struct nfsrv_descript *nd, NFSPROC_T *p)
1277 {
1278 	struct nfslock *lop;
1279 	struct nfsstate *new_stp = *new_stpp;
1280 	struct nfslock *new_lop = *new_lopp;
1281 	struct nfsstate *tstp, *mystp, *nstp;
1282 	int specialid = 0;
1283 	struct nfslockfile *lfp;
1284 	struct nfslock *other_lop = NULL;
1285 	struct nfsstate *stp, *lckstp = NULL;
1286 	struct nfsclient *clp = NULL;
1287 	u_int32_t bits;
1288 	int error = 0, haslock = 0, ret, reterr;
1289 	int getlckret, delegation = 0, filestruct_locked;
1290 	fhandle_t nfh;
1291 	uint64_t first, end;
1292 	uint32_t lock_flags;
1293 
1294 	if (new_stp->ls_flags & (NFSLCK_CHECK | NFSLCK_SETATTR)) {
1295 		/*
1296 		 * Note the special cases of "all 1s" or "all 0s" stateids and
1297 		 * let reads with all 1s go ahead.
1298 		 */
1299 		if (new_stp->ls_stateid.seqid == 0x0 &&
1300 		    new_stp->ls_stateid.other[0] == 0x0 &&
1301 		    new_stp->ls_stateid.other[1] == 0x0 &&
1302 		    new_stp->ls_stateid.other[2] == 0x0)
1303 			specialid = 1;
1304 		else if (new_stp->ls_stateid.seqid == 0xffffffff &&
1305 		    new_stp->ls_stateid.other[0] == 0xffffffff &&
1306 		    new_stp->ls_stateid.other[1] == 0xffffffff &&
1307 		    new_stp->ls_stateid.other[2] == 0xffffffff)
1308 			specialid = 2;
1309 	}
1310 
1311 	/*
1312 	 * Check for restart conditions (client and server).
1313 	 */
1314 	error = nfsrv_checkrestart(clientid, new_stp->ls_flags,
1315 	    &new_stp->ls_stateid, specialid);
1316 	if (error)
1317 		return (error);
1318 
1319 	/*
1320 	 * Check for state resource limit exceeded.
1321 	 */
1322 	if ((new_stp->ls_flags & NFSLCK_LOCK) &&
1323 	    nfsrv_openpluslock > NFSRV_V4STATELIMIT)
1324 		return (NFSERR_RESOURCE);
1325 
1326 	/*
1327 	 * For the lock case, get another nfslock structure,
1328 	 * just in case we need it.
1329 	 * Malloc now, before we start sifting through the linked lists,
1330 	 * in case we have to wait for memory.
1331 	 */
1332 tryagain:
1333 	if (new_stp->ls_flags & NFSLCK_LOCK)
1334 		MALLOC(other_lop, struct nfslock *, sizeof (struct nfslock),
1335 		    M_NFSDLOCK, M_WAITOK);
1336 	filestruct_locked = 0;
1337 	reterr = 0;
1338 	lfp = NULL;
1339 
1340 	/*
1341 	 * Get the lockfile structure for CFH now, so we can do a sanity
1342 	 * check against the stateid, before incrementing the seqid#, since
1343 	 * we want to return NFSERR_BADSTATEID on failure and the seqid#
1344 	 * shouldn't be incremented for this case.
1345 	 * If nfsrv_getlockfile() returns -1, it means "not found", which
1346 	 * will be handled later.
1347 	 * If we are doing Lock/LockU and local locking is enabled, sleep
1348 	 * lock the nfslockfile structure.
1349 	 */
1350 	getlckret = nfsrv_getlockfh(vp, new_stp->ls_flags, NULL, &nfh, p);
1351 	NFSLOCKSTATE();
1352 	if (getlckret == 0) {
1353 		if ((new_stp->ls_flags & (NFSLCK_LOCK | NFSLCK_UNLOCK)) != 0 &&
1354 		    nfsrv_dolocallocks != 0 && nd->nd_repstat == 0) {
1355 			getlckret = nfsrv_getlockfile(new_stp->ls_flags, NULL,
1356 			    &lfp, &nfh, 1);
1357 			if (getlckret == 0)
1358 				filestruct_locked = 1;
1359 		} else
1360 			getlckret = nfsrv_getlockfile(new_stp->ls_flags, NULL,
1361 			    &lfp, &nfh, 0);
1362 	}
1363 	if (getlckret != 0 && getlckret != -1)
1364 		reterr = getlckret;
1365 
1366 	if (filestruct_locked != 0) {
1367 		LIST_INIT(&lfp->lf_rollback);
1368 		if ((new_stp->ls_flags & NFSLCK_LOCK)) {
1369 			/*
1370 			 * For local locking, do the advisory locking now, so
1371 			 * that any conflict can be detected. A failure later
1372 			 * can be rolled back locally. If an error is returned,
1373 			 * struct nfslockfile has been unlocked and any local
1374 			 * locking rolled back.
1375 			 */
1376 			NFSUNLOCKSTATE();
1377 			reterr = nfsrv_locallock(vp, lfp,
1378 			    (new_lop->lo_flags & (NFSLCK_READ | NFSLCK_WRITE)),
1379 			    new_lop->lo_first, new_lop->lo_end, cfp, p);
1380 			NFSLOCKSTATE();
1381 		}
1382 	}
1383 
1384 	if (specialid == 0) {
1385 	    if (new_stp->ls_flags & NFSLCK_TEST) {
1386 		/*
1387 		 * RFC 3530 does not list LockT as an op that renews a
1388 		 * lease, but the concensus seems to be that it is ok
1389 		 * for a server to do so.
1390 		 */
1391 		error = nfsrv_getclient(clientid, CLOPS_RENEW, &clp,
1392 		    (nfsquad_t)((u_quad_t)0), NULL, p);
1393 
1394 		/*
1395 		 * Since NFSERR_EXPIRED, NFSERR_ADMINREVOKED are not valid
1396 		 * error returns for LockT, just go ahead and test for a lock,
1397 		 * since there are no locks for this client, but other locks
1398 		 * can conflict. (ie. same client will always be false)
1399 		 */
1400 		if (error == NFSERR_EXPIRED || error == NFSERR_ADMINREVOKED)
1401 		    error = 0;
1402 		lckstp = new_stp;
1403 	    } else {
1404 	      error = nfsrv_getclient(clientid, CLOPS_RENEW, &clp,
1405 		(nfsquad_t)((u_quad_t)0), NULL, p);
1406 	      if (error == 0)
1407 		/*
1408 		 * Look up the stateid
1409 		 */
1410 		error = nfsrv_getstate(clp, &new_stp->ls_stateid,
1411 		  new_stp->ls_flags, &stp);
1412 	      /*
1413 	       * do some sanity checks for an unconfirmed open or a
1414 	       * stateid that refers to the wrong file, for an open stateid
1415 	       */
1416 	      if (error == 0 && (stp->ls_flags & NFSLCK_OPEN) &&
1417 		  ((stp->ls_openowner->ls_flags & NFSLCK_NEEDSCONFIRM) ||
1418 		   (getlckret == 0 && stp->ls_lfp != lfp)))
1419 			error = NFSERR_BADSTATEID;
1420 	      if (error == 0 &&
1421 		  (stp->ls_flags & (NFSLCK_DELEGREAD | NFSLCK_DELEGWRITE)) &&
1422 		  getlckret == 0 && stp->ls_lfp != lfp)
1423 			error = NFSERR_BADSTATEID;
1424 
1425 	      /*
1426 	       * If the lockowner stateid doesn't refer to the same file,
1427 	       * I believe that is considered ok, since some clients will
1428 	       * only create a single lockowner and use that for all locks
1429 	       * on all files.
1430 	       * For now, log it as a diagnostic, instead of considering it
1431 	       * a BadStateid.
1432 	       */
1433 	      if (error == 0 && (stp->ls_flags &
1434 		  (NFSLCK_OPEN | NFSLCK_DELEGREAD | NFSLCK_DELEGWRITE)) == 0 &&
1435 		  getlckret == 0 && stp->ls_lfp != lfp) {
1436 #ifdef DIAGNOSTIC
1437 		  printf("Got a lock statid for different file open\n");
1438 #endif
1439 		  /*
1440 		  error = NFSERR_BADSTATEID;
1441 		  */
1442 	      }
1443 
1444 	      if (error == 0) {
1445 		    if (new_stp->ls_flags & NFSLCK_OPENTOLOCK) {
1446 			/*
1447 			 * If haslock set, we've already checked the seqid.
1448 			 */
1449 			if (!haslock) {
1450 			    if (stp->ls_flags & NFSLCK_OPEN)
1451 				error = nfsrv_checkseqid(nd, new_stp->ls_seq,
1452 				    stp->ls_openowner, new_stp->ls_op);
1453 			    else
1454 				error = NFSERR_BADSTATEID;
1455 			}
1456 			if (!error)
1457 			    nfsrv_getowner(&stp->ls_open, new_stp, &lckstp);
1458 			if (lckstp)
1459 			    /*
1460 			     * I believe this should be an error, but it
1461 			     * isn't obvious what NFSERR_xxx would be
1462 			     * appropriate, so I'll use NFSERR_INVAL for now.
1463 			     */
1464 			    error = NFSERR_INVAL;
1465 			else
1466 			    lckstp = new_stp;
1467 		    } else if (new_stp->ls_flags&(NFSLCK_LOCK|NFSLCK_UNLOCK)) {
1468 			/*
1469 			 * If haslock set, ditto above.
1470 			 */
1471 			if (!haslock) {
1472 			    if (stp->ls_flags & NFSLCK_OPEN)
1473 				error = NFSERR_BADSTATEID;
1474 			    else
1475 				error = nfsrv_checkseqid(nd, new_stp->ls_seq,
1476 				    stp, new_stp->ls_op);
1477 			}
1478 			lckstp = stp;
1479 		    } else {
1480 			lckstp = stp;
1481 		    }
1482 	      }
1483 	      /*
1484 	       * If the seqid part of the stateid isn't the same, return
1485 	       * NFSERR_OLDSTATEID for cases other than I/O Ops.
1486 	       * For I/O Ops, only return NFSERR_OLDSTATEID if
1487 	       * nfsrv_returnoldstateid is set. (The concensus on the email
1488 	       * list was that most clients would prefer to not receive
1489 	       * NFSERR_OLDSTATEID for I/O Ops, but the RFC suggests that that
1490 	       * is what will happen, so I use the nfsrv_returnoldstateid to
1491 	       * allow for either server configuration.)
1492 	       */
1493 	      if (!error && stp->ls_stateid.seqid!=new_stp->ls_stateid.seqid &&
1494 		  (!(new_stp->ls_flags & NFSLCK_CHECK) ||
1495 		   nfsrv_returnoldstateid))
1496 		    error = NFSERR_OLDSTATEID;
1497 	    }
1498 	}
1499 
1500 	/*
1501 	 * Now we can check for grace.
1502 	 */
1503 	if (!error)
1504 		error = nfsrv_checkgrace(new_stp->ls_flags);
1505 	if ((new_stp->ls_flags & NFSLCK_RECLAIM) && !error &&
1506 		nfsrv_checkstable(clp))
1507 		error = NFSERR_NOGRACE;
1508 	/*
1509 	 * If we successfully Reclaimed state, note that.
1510 	 */
1511 	if ((new_stp->ls_flags & NFSLCK_RECLAIM) && !error)
1512 		nfsrv_markstable(clp);
1513 
1514 	/*
1515 	 * At this point, either error == NFSERR_BADSTATEID or the
1516 	 * seqid# has been updated, so we can return any error.
1517 	 * If error == 0, there may be an error in:
1518 	 *    nd_repstat - Set by the calling function.
1519 	 *    reterr - Set above, if getting the nfslockfile structure
1520 	 *       or acquiring the local lock failed.
1521 	 *    (If both of these are set, nd_repstat should probably be
1522 	 *     returned, since that error was detected before this
1523 	 *     function call.)
1524 	 */
1525 	if (error != 0 || nd->nd_repstat != 0 || reterr != 0) {
1526 		if (error == 0) {
1527 			if (nd->nd_repstat != 0)
1528 				error = nd->nd_repstat;
1529 			else
1530 				error = reterr;
1531 		}
1532 		if (filestruct_locked != 0) {
1533 			/* Roll back local locks. */
1534 			NFSUNLOCKSTATE();
1535 			nfsrv_locallock_rollback(vp, lfp, p);
1536 			NFSLOCKSTATE();
1537 			nfsrv_unlocklf(lfp);
1538 		}
1539 		NFSUNLOCKSTATE();
1540 		if (other_lop)
1541 			FREE((caddr_t)other_lop, M_NFSDLOCK);
1542 		if (haslock) {
1543 			NFSLOCKV4ROOTMUTEX();
1544 			nfsv4_unlock(&nfsv4rootfs_lock, 1);
1545 			NFSUNLOCKV4ROOTMUTEX();
1546 		}
1547 		return (error);
1548 	}
1549 
1550 	/*
1551 	 * Check the nfsrv_getlockfile return.
1552 	 * Returned -1 if no structure found.
1553 	 */
1554 	if (getlckret == -1) {
1555 		error = NFSERR_EXPIRED;
1556 		/*
1557 		 * Called from lockt, so no lock is OK.
1558 		 */
1559 		if (new_stp->ls_flags & NFSLCK_TEST) {
1560 			error = 0;
1561 		} else if (new_stp->ls_flags &
1562 		    (NFSLCK_CHECK | NFSLCK_SETATTR)) {
1563 			/*
1564 			 * Called to check for a lock, OK if the stateid is all
1565 			 * 1s or all 0s, but there should be an nfsstate
1566 			 * otherwise.
1567 			 * (ie. If there is no open, I'll assume no share
1568 			 *  deny bits.)
1569 			 */
1570 			if (specialid)
1571 				error = 0;
1572 			else
1573 				error = NFSERR_BADSTATEID;
1574 		}
1575 		NFSUNLOCKSTATE();
1576 		if (haslock) {
1577 			NFSLOCKV4ROOTMUTEX();
1578 			nfsv4_unlock(&nfsv4rootfs_lock, 1);
1579 			NFSUNLOCKV4ROOTMUTEX();
1580 		}
1581 		/*
1582 		 * Called to lock or unlock, so the lock has gone away.
1583 		 */
1584 		return (error);
1585 	}
1586 
1587 	/*
1588 	 * For NFSLCK_CHECK and NFSLCK_LOCK, test for a share conflict.
1589 	 * For NFSLCK_CHECK, allow a read if write access is granted,
1590 	 * but check for a deny. For NFSLCK_LOCK, require correct access,
1591 	 * which implies a conflicting deny can't exist.
1592 	 */
1593 	if (new_stp->ls_flags & (NFSLCK_CHECK | NFSLCK_LOCK)) {
1594 	    /*
1595 	     * Four kinds of state id:
1596 	     * - specialid (all 0s or all 1s), only for NFSLCK_CHECK
1597 	     * - stateid for an open
1598 	     * - stateid for a delegation
1599 	     * - stateid for a lock owner
1600 	     */
1601 	    if (!specialid) {
1602 		if (stp->ls_flags & (NFSLCK_DELEGREAD | NFSLCK_DELEGWRITE)) {
1603 		    delegation = 1;
1604 		    mystp = stp;
1605 		    nfsrv_delaydelegtimeout(stp);
1606 	        } else if (stp->ls_flags & NFSLCK_OPEN) {
1607 		    mystp = stp;
1608 		} else {
1609 		    mystp = stp->ls_openstp;
1610 		}
1611 		/*
1612 		 * If locking or checking, require correct access
1613 		 * bit set.
1614 		 */
1615 		if (((new_stp->ls_flags & NFSLCK_LOCK) &&
1616 		     !((new_lop->lo_flags >> NFSLCK_LOCKSHIFT) &
1617 		       mystp->ls_flags & NFSLCK_ACCESSBITS)) ||
1618 		    ((new_stp->ls_flags & (NFSLCK_CHECK|NFSLCK_READACCESS)) ==
1619 		      (NFSLCK_CHECK | NFSLCK_READACCESS) &&
1620 		     !(mystp->ls_flags & NFSLCK_READACCESS)) ||
1621 		    ((new_stp->ls_flags & (NFSLCK_CHECK|NFSLCK_WRITEACCESS)) ==
1622 		      (NFSLCK_CHECK | NFSLCK_WRITEACCESS) &&
1623 		     !(mystp->ls_flags & NFSLCK_WRITEACCESS))) {
1624 			if (filestruct_locked != 0) {
1625 				/* Roll back local locks. */
1626 				NFSUNLOCKSTATE();
1627 				nfsrv_locallock_rollback(vp, lfp, p);
1628 				NFSLOCKSTATE();
1629 				nfsrv_unlocklf(lfp);
1630 			}
1631 			NFSUNLOCKSTATE();
1632 			if (other_lop)
1633 				FREE((caddr_t)other_lop, M_NFSDLOCK);
1634 			if (haslock) {
1635 				NFSLOCKV4ROOTMUTEX();
1636 				nfsv4_unlock(&nfsv4rootfs_lock, 1);
1637 				NFSUNLOCKV4ROOTMUTEX();
1638 			}
1639 			return (NFSERR_OPENMODE);
1640 		}
1641 	    } else
1642 		mystp = NULL;
1643 	    if ((new_stp->ls_flags & NFSLCK_CHECK) && !delegation) {
1644 		/*
1645 		 * Check for a conflicting deny bit.
1646 		 */
1647 		LIST_FOREACH(tstp, &lfp->lf_open, ls_file) {
1648 		    if (tstp != mystp) {
1649 			bits = tstp->ls_flags;
1650 			bits >>= NFSLCK_SHIFT;
1651 			if (new_stp->ls_flags & bits & NFSLCK_ACCESSBITS) {
1652 			    ret = nfsrv_clientconflict(tstp->ls_clp, &haslock,
1653 				vp, p);
1654 			    if (ret) {
1655 				/*
1656 				* nfsrv_clientconflict unlocks state
1657 				 * when it returns non-zero.
1658 				 */
1659 				lckstp = NULL;
1660 				goto tryagain;
1661 			    }
1662 			    NFSUNLOCKSTATE();
1663 			    if (haslock) {
1664 				NFSLOCKV4ROOTMUTEX();
1665 				nfsv4_unlock(&nfsv4rootfs_lock, 1);
1666 				NFSUNLOCKV4ROOTMUTEX();
1667 			    }
1668 			    return (NFSERR_OPENMODE);
1669 			}
1670 		    }
1671 		}
1672 
1673 		/* We're outta here */
1674 		NFSUNLOCKSTATE();
1675 		if (haslock) {
1676 			NFSLOCKV4ROOTMUTEX();
1677 			nfsv4_unlock(&nfsv4rootfs_lock, 1);
1678 			NFSUNLOCKV4ROOTMUTEX();
1679 		}
1680 		return (0);
1681 	    }
1682 	}
1683 
1684 	/*
1685 	 * For setattr, just get rid of all the Delegations for other clients.
1686 	 */
1687 	if (new_stp->ls_flags & NFSLCK_SETATTR) {
1688 		ret = nfsrv_cleandeleg(vp, lfp, clp, &haslock, p);
1689 		if (ret) {
1690 			/*
1691 			 * nfsrv_cleandeleg() unlocks state when it
1692 			 * returns non-zero.
1693 			 */
1694 			if (ret == -1) {
1695 				lckstp = NULL;
1696 				goto tryagain;
1697 			}
1698 			return (ret);
1699 		}
1700 		if (!(new_stp->ls_flags & NFSLCK_CHECK) ||
1701 		    (LIST_EMPTY(&lfp->lf_open) && LIST_EMPTY(&lfp->lf_lock) &&
1702 		     LIST_EMPTY(&lfp->lf_deleg))) {
1703 			NFSUNLOCKSTATE();
1704 			if (haslock) {
1705 				NFSLOCKV4ROOTMUTEX();
1706 				nfsv4_unlock(&nfsv4rootfs_lock, 1);
1707 				NFSUNLOCKV4ROOTMUTEX();
1708 			}
1709 			return (0);
1710 		}
1711 	}
1712 
1713 	/*
1714 	 * Check for a conflicting delegation. If one is found, call
1715 	 * nfsrv_delegconflict() to handle it. If the v4root lock hasn't
1716 	 * been set yet, it will get the lock. Otherwise, it will recall
1717 	 * the delegation. Then, we try try again...
1718 	 * I currently believe the conflict algorithm to be:
1719 	 * For Lock Ops (Lock/LockT/LockU)
1720 	 * - there is a conflict iff a different client has a write delegation
1721 	 * For Reading (Read Op)
1722 	 * - there is a conflict iff a different client has a write delegation
1723 	 *   (the specialids are always a different client)
1724 	 * For Writing (Write/Setattr of size)
1725 	 * - there is a conflict if a different client has any delegation
1726 	 * - there is a conflict if the same client has a read delegation
1727 	 *   (I don't understand why this isn't allowed, but that seems to be
1728 	 *    the current concensus?)
1729 	 */
1730 	tstp = LIST_FIRST(&lfp->lf_deleg);
1731 	while (tstp != LIST_END(&lfp->lf_deleg)) {
1732 	    nstp = LIST_NEXT(tstp, ls_file);
1733 	    if ((((new_stp->ls_flags&(NFSLCK_LOCK|NFSLCK_UNLOCK|NFSLCK_TEST))||
1734 		 ((new_stp->ls_flags & NFSLCK_CHECK) &&
1735 		  (new_lop->lo_flags & NFSLCK_READ))) &&
1736 		  clp != tstp->ls_clp &&
1737 		 (tstp->ls_flags & NFSLCK_DELEGWRITE)) ||
1738 		 ((new_stp->ls_flags & NFSLCK_CHECK) &&
1739 		   (new_lop->lo_flags & NFSLCK_WRITE) &&
1740 		  (clp != tstp->ls_clp ||
1741 		   (tstp->ls_flags & NFSLCK_DELEGREAD)))) {
1742 		if (filestruct_locked != 0) {
1743 			/* Roll back local locks. */
1744 			NFSUNLOCKSTATE();
1745 			nfsrv_locallock_rollback(vp, lfp, p);
1746 			NFSLOCKSTATE();
1747 			nfsrv_unlocklf(lfp);
1748 		}
1749 		ret = nfsrv_delegconflict(tstp, &haslock, p, vp);
1750 		if (ret) {
1751 		    /*
1752 		     * nfsrv_delegconflict unlocks state when it
1753 		     * returns non-zero, which it always does.
1754 		     */
1755 		    if (other_lop) {
1756 			FREE((caddr_t)other_lop, M_NFSDLOCK);
1757 			other_lop = NULL;
1758 		    }
1759 		    if (ret == -1) {
1760 			lckstp = NULL;
1761 			goto tryagain;
1762 		    }
1763 		    return (ret);
1764 		}
1765 		/* Never gets here. */
1766 	    }
1767 	    tstp = nstp;
1768 	}
1769 
1770 	/*
1771 	 * Handle the unlock case by calling nfsrv_updatelock().
1772 	 * (Should I have done some access checking above for unlock? For now,
1773 	 *  just let it happen.)
1774 	 */
1775 	if (new_stp->ls_flags & NFSLCK_UNLOCK) {
1776 		first = new_lop->lo_first;
1777 		end = new_lop->lo_end;
1778 		nfsrv_updatelock(stp, new_lopp, &other_lop, lfp);
1779 		stateidp->seqid = ++(stp->ls_stateid.seqid);
1780 		stateidp->other[0] = stp->ls_stateid.other[0];
1781 		stateidp->other[1] = stp->ls_stateid.other[1];
1782 		stateidp->other[2] = stp->ls_stateid.other[2];
1783 		if (filestruct_locked != 0) {
1784 			NFSUNLOCKSTATE();
1785 			/* Update the local locks. */
1786 			nfsrv_localunlock(vp, lfp, first, end, p);
1787 			NFSLOCKSTATE();
1788 			nfsrv_unlocklf(lfp);
1789 		}
1790 		NFSUNLOCKSTATE();
1791 		if (haslock) {
1792 			NFSLOCKV4ROOTMUTEX();
1793 			nfsv4_unlock(&nfsv4rootfs_lock, 1);
1794 			NFSUNLOCKV4ROOTMUTEX();
1795 		}
1796 		return (0);
1797 	}
1798 
1799 	/*
1800 	 * Search for a conflicting lock. A lock conflicts if:
1801 	 * - the lock range overlaps and
1802 	 * - at least one lock is a write lock and
1803 	 * - it is not owned by the same lock owner
1804 	 */
1805 	if (!delegation) {
1806 	  LIST_FOREACH(lop, &lfp->lf_lock, lo_lckfile) {
1807 	    if (new_lop->lo_end > lop->lo_first &&
1808 		new_lop->lo_first < lop->lo_end &&
1809 		(new_lop->lo_flags == NFSLCK_WRITE ||
1810 		 lop->lo_flags == NFSLCK_WRITE) &&
1811 		lckstp != lop->lo_stp &&
1812 		(clp != lop->lo_stp->ls_clp ||
1813 		 lckstp->ls_ownerlen != lop->lo_stp->ls_ownerlen ||
1814 		 NFSBCMP(lckstp->ls_owner, lop->lo_stp->ls_owner,
1815 		    lckstp->ls_ownerlen))) {
1816 		if (other_lop) {
1817 		    FREE((caddr_t)other_lop, M_NFSDLOCK);
1818 		    other_lop = NULL;
1819 		}
1820 		ret = nfsrv_clientconflict(lop->lo_stp->ls_clp,&haslock,vp,p);
1821 		if (ret) {
1822 		    if (filestruct_locked != 0) {
1823 			/* Roll back local locks. */
1824 			nfsrv_locallock_rollback(vp, lfp, p);
1825 			NFSLOCKSTATE();
1826 			nfsrv_unlocklf(lfp);
1827 			NFSUNLOCKSTATE();
1828 		    }
1829 		    /*
1830 		     * nfsrv_clientconflict() unlocks state when it
1831 		     * returns non-zero.
1832 		     */
1833 		    lckstp = NULL;
1834 		    goto tryagain;
1835 		}
1836 		/*
1837 		 * Found a conflicting lock, so record the conflict and
1838 		 * return the error.
1839 		 */
1840 		if (cfp) {
1841 		    cfp->cl_clientid.lval[0]=lop->lo_stp->ls_stateid.other[0];
1842 		    cfp->cl_clientid.lval[1]=lop->lo_stp->ls_stateid.other[1];
1843 		    cfp->cl_first = lop->lo_first;
1844 		    cfp->cl_end = lop->lo_end;
1845 		    cfp->cl_flags = lop->lo_flags;
1846 		    cfp->cl_ownerlen = lop->lo_stp->ls_ownerlen;
1847 		    NFSBCOPY(lop->lo_stp->ls_owner, cfp->cl_owner,
1848 			cfp->cl_ownerlen);
1849 		}
1850 		if (new_stp->ls_flags & NFSLCK_RECLAIM)
1851 		    error = NFSERR_RECLAIMCONFLICT;
1852 		else if (new_stp->ls_flags & NFSLCK_CHECK)
1853 		    error = NFSERR_LOCKED;
1854 		else
1855 		    error = NFSERR_DENIED;
1856 		if (filestruct_locked != 0) {
1857 			/* Roll back local locks. */
1858 			NFSUNLOCKSTATE();
1859 			nfsrv_locallock_rollback(vp, lfp, p);
1860 			NFSLOCKSTATE();
1861 			nfsrv_unlocklf(lfp);
1862 		}
1863 		NFSUNLOCKSTATE();
1864 		if (haslock) {
1865 			NFSLOCKV4ROOTMUTEX();
1866 			nfsv4_unlock(&nfsv4rootfs_lock, 1);
1867 			NFSUNLOCKV4ROOTMUTEX();
1868 		}
1869 		return (error);
1870 	    }
1871 	  }
1872 	}
1873 
1874 	/*
1875 	 * We only get here if there was no lock that conflicted.
1876 	 */
1877 	if (new_stp->ls_flags & (NFSLCK_TEST | NFSLCK_CHECK)) {
1878 		NFSUNLOCKSTATE();
1879 		if (haslock) {
1880 			NFSLOCKV4ROOTMUTEX();
1881 			nfsv4_unlock(&nfsv4rootfs_lock, 1);
1882 			NFSUNLOCKV4ROOTMUTEX();
1883 		}
1884 		return (0);
1885 	}
1886 
1887 	/*
1888 	 * We only get here when we are creating or modifying a lock.
1889 	 * There are two variants:
1890 	 * - exist_lock_owner where lock_owner exists
1891 	 * - open_to_lock_owner with new lock_owner
1892 	 */
1893 	first = new_lop->lo_first;
1894 	end = new_lop->lo_end;
1895 	lock_flags = new_lop->lo_flags;
1896 	if (!(new_stp->ls_flags & NFSLCK_OPENTOLOCK)) {
1897 		nfsrv_updatelock(lckstp, new_lopp, &other_lop, lfp);
1898 		stateidp->seqid = ++(lckstp->ls_stateid.seqid);
1899 		stateidp->other[0] = lckstp->ls_stateid.other[0];
1900 		stateidp->other[1] = lckstp->ls_stateid.other[1];
1901 		stateidp->other[2] = lckstp->ls_stateid.other[2];
1902 	} else {
1903 		/*
1904 		 * The new open_to_lock_owner case.
1905 		 * Link the new nfsstate into the lists.
1906 		 */
1907 		new_stp->ls_seq = new_stp->ls_opentolockseq;
1908 		nfsrvd_refcache(new_stp->ls_op);
1909 		stateidp->seqid = new_stp->ls_stateid.seqid = 0;
1910 		stateidp->other[0] = new_stp->ls_stateid.other[0] =
1911 		    clp->lc_clientid.lval[0];
1912 		stateidp->other[1] = new_stp->ls_stateid.other[1] =
1913 		    clp->lc_clientid.lval[1];
1914 		stateidp->other[2] = new_stp->ls_stateid.other[2] =
1915 		    nfsrv_nextstateindex(clp);
1916 		new_stp->ls_clp = clp;
1917 		LIST_INIT(&new_stp->ls_lock);
1918 		new_stp->ls_openstp = stp;
1919 		new_stp->ls_lfp = lfp;
1920 		nfsrv_insertlock(new_lop, (struct nfslock *)new_stp, new_stp,
1921 		    lfp);
1922 		LIST_INSERT_HEAD(NFSSTATEHASH(clp, new_stp->ls_stateid),
1923 		    new_stp, ls_hash);
1924 		LIST_INSERT_HEAD(&stp->ls_open, new_stp, ls_list);
1925 		*new_lopp = NULL;
1926 		*new_stpp = NULL;
1927 		newnfsstats.srvlockowners++;
1928 		nfsrv_openpluslock++;
1929 	}
1930 	if (filestruct_locked != 0) {
1931 		NFSUNLOCKSTATE();
1932 		nfsrv_locallock_commit(lfp, lock_flags, first, end);
1933 		NFSLOCKSTATE();
1934 		nfsrv_unlocklf(lfp);
1935 	}
1936 	NFSUNLOCKSTATE();
1937 	if (haslock) {
1938 		NFSLOCKV4ROOTMUTEX();
1939 		nfsv4_unlock(&nfsv4rootfs_lock, 1);
1940 		NFSUNLOCKV4ROOTMUTEX();
1941 	}
1942 	if (other_lop)
1943 		FREE((caddr_t)other_lop, M_NFSDLOCK);
1944 	return (0);
1945 }
1946 
1947 /*
1948  * Check for state errors for Open.
1949  * repstat is passed back out as an error if more critical errors
1950  * are not detected.
1951  */
1952 APPLESTATIC int
1953 nfsrv_opencheck(nfsquad_t clientid, nfsv4stateid_t *stateidp,
1954     struct nfsstate *new_stp, vnode_t vp, struct nfsrv_descript *nd,
1955     NFSPROC_T *p, int repstat)
1956 {
1957 	struct nfsstate *stp, *nstp;
1958 	struct nfsclient *clp;
1959 	struct nfsstate *ownerstp;
1960 	struct nfslockfile *lfp, *new_lfp;
1961 	int error, haslock = 0, ret, readonly = 0, getfhret = 0;
1962 
1963 	if ((new_stp->ls_flags & NFSLCK_SHAREBITS) == NFSLCK_READACCESS)
1964 		readonly = 1;
1965 	/*
1966 	 * Check for restart conditions (client and server).
1967 	 */
1968 	error = nfsrv_checkrestart(clientid, new_stp->ls_flags,
1969 		&new_stp->ls_stateid, 0);
1970 	if (error)
1971 		return (error);
1972 
1973 	/*
1974 	 * Check for state resource limit exceeded.
1975 	 * Technically this should be SMP protected, but the worst
1976 	 * case error is "out by one or two" on the count when it
1977 	 * returns NFSERR_RESOURCE and the limit is just a rather
1978 	 * arbitrary high water mark, so no harm is done.
1979 	 */
1980 	if (nfsrv_openpluslock > NFSRV_V4STATELIMIT)
1981 		return (NFSERR_RESOURCE);
1982 
1983 tryagain:
1984 	MALLOC(new_lfp, struct nfslockfile *, sizeof (struct nfslockfile),
1985 	    M_NFSDLOCKFILE, M_WAITOK);
1986 	if (vp)
1987 		getfhret = nfsrv_getlockfh(vp, new_stp->ls_flags, &new_lfp,
1988 		    NULL, p);
1989 	NFSLOCKSTATE();
1990 	/*
1991 	 * Get the nfsclient structure.
1992 	 */
1993 	error = nfsrv_getclient(clientid, CLOPS_RENEW, &clp,
1994 	    (nfsquad_t)((u_quad_t)0), NULL, p);
1995 
1996 	/*
1997 	 * Look up the open owner. See if it needs confirmation and
1998 	 * check the seq#, as required.
1999 	 */
2000 	if (!error)
2001 		nfsrv_getowner(&clp->lc_open, new_stp, &ownerstp);
2002 
2003 	if (!error && ownerstp) {
2004 		error = nfsrv_checkseqid(nd, new_stp->ls_seq, ownerstp,
2005 		    new_stp->ls_op);
2006 		/*
2007 		 * If the OpenOwner hasn't been confirmed, assume the
2008 		 * old one was a replay and this one is ok.
2009 		 * See: RFC3530 Sec. 14.2.18.
2010 		 */
2011 		if (error == NFSERR_BADSEQID &&
2012 		    (ownerstp->ls_flags & NFSLCK_NEEDSCONFIRM))
2013 			error = 0;
2014 	}
2015 
2016 	/*
2017 	 * Check for grace.
2018 	 */
2019 	if (!error)
2020 		error = nfsrv_checkgrace(new_stp->ls_flags);
2021 	if ((new_stp->ls_flags & NFSLCK_RECLAIM) && !error &&
2022 		nfsrv_checkstable(clp))
2023 		error = NFSERR_NOGRACE;
2024 
2025 	/*
2026 	 * If none of the above errors occurred, let repstat be
2027 	 * returned.
2028 	 */
2029 	if (repstat && !error)
2030 		error = repstat;
2031 	if (error) {
2032 		NFSUNLOCKSTATE();
2033 		if (haslock) {
2034 			NFSLOCKV4ROOTMUTEX();
2035 			nfsv4_unlock(&nfsv4rootfs_lock, 1);
2036 			NFSUNLOCKV4ROOTMUTEX();
2037 		}
2038 		free((caddr_t)new_lfp, M_NFSDLOCKFILE);
2039 		return (error);
2040 	}
2041 
2042 	/*
2043 	 * If vp == NULL, the file doesn't exist yet, so return ok.
2044 	 * (This always happens on the first pass, so haslock must be 0.)
2045 	 */
2046 	if (vp == NULL) {
2047 		NFSUNLOCKSTATE();
2048 		FREE((caddr_t)new_lfp, M_NFSDLOCKFILE);
2049 		return (0);
2050 	}
2051 
2052 	/*
2053 	 * Get the structure for the underlying file.
2054 	 */
2055 	if (getfhret)
2056 		error = getfhret;
2057 	else
2058 		error = nfsrv_getlockfile(new_stp->ls_flags, &new_lfp, &lfp,
2059 		    NULL, 0);
2060 	if (new_lfp)
2061 		FREE((caddr_t)new_lfp, M_NFSDLOCKFILE);
2062 	if (error) {
2063 		NFSUNLOCKSTATE();
2064 		if (haslock) {
2065 			NFSLOCKV4ROOTMUTEX();
2066 			nfsv4_unlock(&nfsv4rootfs_lock, 1);
2067 			NFSUNLOCKV4ROOTMUTEX();
2068 		}
2069 		return (error);
2070 	}
2071 
2072 	/*
2073 	 * Search for a conflicting open/share.
2074 	 */
2075 	if (new_stp->ls_flags & NFSLCK_DELEGCUR) {
2076 	    /*
2077 	     * For Delegate_Cur, search for the matching Delegation,
2078 	     * which indicates no conflict.
2079 	     * An old delegation should have been recovered by the
2080 	     * client doing a Claim_DELEGATE_Prev, so I won't let
2081 	     * it match and return NFSERR_EXPIRED. Should I let it
2082 	     * match?
2083 	     */
2084 	    LIST_FOREACH(stp, &lfp->lf_deleg, ls_file) {
2085 		if (!(stp->ls_flags & NFSLCK_OLDDELEG) &&
2086 		    stateidp->seqid == stp->ls_stateid.seqid &&
2087 		    !NFSBCMP(stateidp->other, stp->ls_stateid.other,
2088 			  NFSX_STATEIDOTHER))
2089 			break;
2090 	    }
2091 	    if (stp == LIST_END(&lfp->lf_deleg) ||
2092 		((new_stp->ls_flags & NFSLCK_WRITEACCESS) &&
2093 		 (stp->ls_flags & NFSLCK_DELEGREAD))) {
2094 		NFSUNLOCKSTATE();
2095 		if (haslock) {
2096 			NFSLOCKV4ROOTMUTEX();
2097 			nfsv4_unlock(&nfsv4rootfs_lock, 1);
2098 			NFSUNLOCKV4ROOTMUTEX();
2099 		}
2100 		return (NFSERR_EXPIRED);
2101 	    }
2102 	}
2103 
2104 	/*
2105 	 * Check for access/deny bit conflicts. I check for the same
2106 	 * owner as well, in case the client didn't bother.
2107 	 */
2108 	LIST_FOREACH(stp, &lfp->lf_open, ls_file) {
2109 		if (!(new_stp->ls_flags & NFSLCK_DELEGCUR) &&
2110 		    (((new_stp->ls_flags & NFSLCK_ACCESSBITS) &
2111 		      ((stp->ls_flags>>NFSLCK_SHIFT) & NFSLCK_ACCESSBITS))||
2112 		     ((stp->ls_flags & NFSLCK_ACCESSBITS) &
2113 		      ((new_stp->ls_flags>>NFSLCK_SHIFT)&NFSLCK_ACCESSBITS)))){
2114 			ret = nfsrv_clientconflict(stp->ls_clp,&haslock,vp,p);
2115 			if (ret) {
2116 				/*
2117 				 * nfsrv_clientconflict() unlocks
2118 				 * state when it returns non-zero.
2119 				 */
2120 				goto tryagain;
2121 			}
2122 			if (new_stp->ls_flags & NFSLCK_RECLAIM)
2123 				error = NFSERR_RECLAIMCONFLICT;
2124 			else
2125 				error = NFSERR_SHAREDENIED;
2126 			NFSUNLOCKSTATE();
2127 			if (haslock) {
2128 				NFSLOCKV4ROOTMUTEX();
2129 				nfsv4_unlock(&nfsv4rootfs_lock, 1);
2130 				NFSUNLOCKV4ROOTMUTEX();
2131 			}
2132 			return (error);
2133 		}
2134 	}
2135 
2136 	/*
2137 	 * Check for a conflicting delegation. If one is found, call
2138 	 * nfsrv_delegconflict() to handle it. If the v4root lock hasn't
2139 	 * been set yet, it will get the lock. Otherwise, it will recall
2140 	 * the delegation. Then, we try try again...
2141 	 * (If NFSLCK_DELEGCUR is set, it has a delegation, so there
2142 	 *  isn't a conflict.)
2143 	 * I currently believe the conflict algorithm to be:
2144 	 * For Open with Read Access and Deny None
2145 	 * - there is a conflict iff a different client has a write delegation
2146 	 * For Open with other Write Access or any Deny except None
2147 	 * - there is a conflict if a different client has any delegation
2148 	 * - there is a conflict if the same client has a read delegation
2149 	 *   (The current concensus is that this last case should be
2150 	 *    considered a conflict since the client with a read delegation
2151 	 *    could have done an Open with ReadAccess and WriteDeny
2152 	 *    locally and then not have checked for the WriteDeny.)
2153 	 * Don't check for a Reclaim, since that will be dealt with
2154 	 * by nfsrv_openctrl().
2155 	 */
2156 	if (!(new_stp->ls_flags &
2157 		(NFSLCK_DELEGPREV | NFSLCK_DELEGCUR | NFSLCK_RECLAIM))) {
2158 	    stp = LIST_FIRST(&lfp->lf_deleg);
2159 	    while (stp != LIST_END(&lfp->lf_deleg)) {
2160 		nstp = LIST_NEXT(stp, ls_file);
2161 		if ((readonly && stp->ls_clp != clp &&
2162 		       (stp->ls_flags & NFSLCK_DELEGWRITE)) ||
2163 		    (!readonly && (stp->ls_clp != clp ||
2164 		         (stp->ls_flags & NFSLCK_DELEGREAD)))) {
2165 			ret = nfsrv_delegconflict(stp, &haslock, p, vp);
2166 			if (ret) {
2167 			    /*
2168 			     * nfsrv_delegconflict() unlocks state
2169 			     * when it returns non-zero.
2170 			     */
2171 			    if (ret == -1)
2172 				goto tryagain;
2173 			    return (ret);
2174 			}
2175 		}
2176 		stp = nstp;
2177 	    }
2178 	}
2179 	NFSUNLOCKSTATE();
2180 	if (haslock) {
2181 		NFSLOCKV4ROOTMUTEX();
2182 		nfsv4_unlock(&nfsv4rootfs_lock, 1);
2183 		NFSUNLOCKV4ROOTMUTEX();
2184 	}
2185 	return (0);
2186 }
2187 
2188 /*
2189  * Open control function to create/update open state for an open.
2190  */
2191 APPLESTATIC int
2192 nfsrv_openctrl(struct nfsrv_descript *nd, vnode_t vp,
2193     struct nfsstate **new_stpp, nfsquad_t clientid, nfsv4stateid_t *stateidp,
2194     nfsv4stateid_t *delegstateidp, u_int32_t *rflagsp, struct nfsexstuff *exp,
2195     NFSPROC_T *p, u_quad_t filerev)
2196 {
2197 	struct nfsstate *new_stp = *new_stpp;
2198 	struct nfsstate *stp, *nstp;
2199 	struct nfsstate *openstp = NULL, *new_open, *ownerstp, *new_deleg;
2200 	struct nfslockfile *lfp, *new_lfp;
2201 	struct nfsclient *clp;
2202 	int error, haslock = 0, ret, delegate = 1, writedeleg = 1;
2203 	int readonly = 0, cbret = 1, getfhret = 0;
2204 
2205 	if ((new_stp->ls_flags & NFSLCK_SHAREBITS) == NFSLCK_READACCESS)
2206 		readonly = 1;
2207 	/*
2208 	 * Check for restart conditions (client and server).
2209 	 * (Paranoia, should have been detected by nfsrv_opencheck().)
2210 	 * If an error does show up, return NFSERR_EXPIRED, since the
2211 	 * the seqid# has already been incremented.
2212 	 */
2213 	error = nfsrv_checkrestart(clientid, new_stp->ls_flags,
2214 	    &new_stp->ls_stateid, 0);
2215 	if (error) {
2216 		printf("Nfsd: openctrl unexpected restart err=%d\n",
2217 		    error);
2218 		return (NFSERR_EXPIRED);
2219 	}
2220 
2221 tryagain:
2222 	MALLOC(new_lfp, struct nfslockfile *, sizeof (struct nfslockfile),
2223 	    M_NFSDLOCKFILE, M_WAITOK);
2224 	MALLOC(new_open, struct nfsstate *, sizeof (struct nfsstate),
2225 	    M_NFSDSTATE, M_WAITOK);
2226 	MALLOC(new_deleg, struct nfsstate *, sizeof (struct nfsstate),
2227 	    M_NFSDSTATE, M_WAITOK);
2228 	getfhret = nfsrv_getlockfh(vp, new_stp->ls_flags, &new_lfp,
2229 	    NULL, p);
2230 	NFSLOCKSTATE();
2231 	/*
2232 	 * Get the client structure. Since the linked lists could be changed
2233 	 * by other nfsd processes if this process does a tsleep(), one of
2234 	 * two things must be done.
2235 	 * 1 - don't tsleep()
2236 	 * or
2237 	 * 2 - get the nfsv4_lock() { indicated by haslock == 1 }
2238 	 *     before using the lists, since this lock stops the other
2239 	 *     nfsd. This should only be used for rare cases, since it
2240 	 *     essentially single threads the nfsd.
2241 	 *     At this time, it is only done for cases where the stable
2242 	 *     storage file must be written prior to completion of state
2243 	 *     expiration.
2244 	 */
2245 	error = nfsrv_getclient(clientid, CLOPS_RENEW, &clp,
2246 	    (nfsquad_t)((u_quad_t)0), NULL, p);
2247 	if (!error && (clp->lc_flags & LCL_NEEDSCBNULL) &&
2248 	    clp->lc_program) {
2249 		/*
2250 		 * This happens on the first open for a client
2251 		 * that supports callbacks.
2252 		 */
2253 		NFSUNLOCKSTATE();
2254 		/*
2255 		 * Although nfsrv_docallback() will sleep, clp won't
2256 		 * go away, since they are only removed when the
2257 		 * nfsv4_lock() has blocked the nfsd threads. The
2258 		 * fields in clp can change, but having multiple
2259 		 * threads do this Null callback RPC should be
2260 		 * harmless.
2261 		 */
2262 		cbret = nfsrv_docallback(clp, NFSV4PROC_CBNULL,
2263 		    NULL, 0, NULL, NULL, NULL, p);
2264 		NFSLOCKSTATE();
2265 		clp->lc_flags &= ~LCL_NEEDSCBNULL;
2266 		if (!cbret)
2267 			clp->lc_flags |= LCL_CALLBACKSON;
2268 	}
2269 
2270 	/*
2271 	 * Look up the open owner. See if it needs confirmation and
2272 	 * check the seq#, as required.
2273 	 */
2274 	if (!error)
2275 		nfsrv_getowner(&clp->lc_open, new_stp, &ownerstp);
2276 
2277 	if (error) {
2278 		NFSUNLOCKSTATE();
2279 		printf("Nfsd: openctrl unexpected state err=%d\n",
2280 			error);
2281 		free((caddr_t)new_lfp, M_NFSDLOCKFILE);
2282 		free((caddr_t)new_open, M_NFSDSTATE);
2283 		free((caddr_t)new_deleg, M_NFSDSTATE);
2284 		if (haslock) {
2285 			NFSLOCKV4ROOTMUTEX();
2286 			nfsv4_unlock(&nfsv4rootfs_lock, 1);
2287 			NFSUNLOCKV4ROOTMUTEX();
2288 		}
2289 		return (NFSERR_EXPIRED);
2290 	}
2291 
2292 	if (new_stp->ls_flags & NFSLCK_RECLAIM)
2293 		nfsrv_markstable(clp);
2294 
2295 	/*
2296 	 * Get the structure for the underlying file.
2297 	 */
2298 	if (getfhret)
2299 		error = getfhret;
2300 	else
2301 		error = nfsrv_getlockfile(new_stp->ls_flags, &new_lfp, &lfp,
2302 		    NULL, 0);
2303 	if (new_lfp)
2304 		FREE((caddr_t)new_lfp, M_NFSDLOCKFILE);
2305 	if (error) {
2306 		NFSUNLOCKSTATE();
2307 		printf("Nfsd openctrl unexpected getlockfile err=%d\n",
2308 		    error);
2309 		free((caddr_t)new_open, M_NFSDSTATE);
2310 		free((caddr_t)new_deleg, M_NFSDSTATE);
2311 		if (haslock) {
2312 			NFSLOCKV4ROOTMUTEX();
2313 			nfsv4_unlock(&nfsv4rootfs_lock, 1);
2314 			NFSUNLOCKV4ROOTMUTEX();
2315 		}
2316 		return (error);
2317 	}
2318 
2319 	/*
2320 	 * Search for a conflicting open/share.
2321 	 */
2322 	if (new_stp->ls_flags & NFSLCK_DELEGCUR) {
2323 	    /*
2324 	     * For Delegate_Cur, search for the matching Delegation,
2325 	     * which indicates no conflict.
2326 	     * An old delegation should have been recovered by the
2327 	     * client doing a Claim_DELEGATE_Prev, so I won't let
2328 	     * it match and return NFSERR_EXPIRED. Should I let it
2329 	     * match?
2330 	     */
2331 	    LIST_FOREACH(stp, &lfp->lf_deleg, ls_file) {
2332 		if (!(stp->ls_flags & NFSLCK_OLDDELEG) &&
2333 		    stateidp->seqid == stp->ls_stateid.seqid &&
2334 		    !NFSBCMP(stateidp->other, stp->ls_stateid.other,
2335 			NFSX_STATEIDOTHER))
2336 			break;
2337 	    }
2338 	    if (stp == LIST_END(&lfp->lf_deleg) ||
2339 		((new_stp->ls_flags & NFSLCK_WRITEACCESS) &&
2340 		 (stp->ls_flags & NFSLCK_DELEGREAD))) {
2341 		NFSUNLOCKSTATE();
2342 		printf("Nfsd openctrl unexpected expiry\n");
2343 		free((caddr_t)new_open, M_NFSDSTATE);
2344 		free((caddr_t)new_deleg, M_NFSDSTATE);
2345 		if (haslock) {
2346 			NFSLOCKV4ROOTMUTEX();
2347 			nfsv4_unlock(&nfsv4rootfs_lock, 1);
2348 			NFSUNLOCKV4ROOTMUTEX();
2349 		}
2350 		return (NFSERR_EXPIRED);
2351 	    }
2352 
2353 	    /*
2354 	     * Don't issue a Delegation, since one already exists and
2355 	     * delay delegation timeout, as required.
2356 	     */
2357 	    delegate = 0;
2358 	    nfsrv_delaydelegtimeout(stp);
2359 	}
2360 
2361 	/*
2362 	 * Check for access/deny bit conflicts. I also check for the
2363 	 * same owner, since the client might not have bothered to check.
2364 	 * Also, note an open for the same file and owner, if found,
2365 	 * which is all we do here for Delegate_Cur, since conflict
2366 	 * checking is already done.
2367 	 */
2368 	LIST_FOREACH(stp, &lfp->lf_open, ls_file) {
2369 		if (ownerstp && stp->ls_openowner == ownerstp)
2370 			openstp = stp;
2371 		if (!(new_stp->ls_flags & NFSLCK_DELEGCUR)) {
2372 		    /*
2373 		     * If another client has the file open, the only
2374 		     * delegation that can be issued is a Read delegation
2375 		     * and only if it is a Read open with Deny none.
2376 		     */
2377 		    if (clp != stp->ls_clp) {
2378 			if ((stp->ls_flags & NFSLCK_SHAREBITS) ==
2379 			    NFSLCK_READACCESS)
2380 			    writedeleg = 0;
2381 			else
2382 			    delegate = 0;
2383 		    }
2384 		    if(((new_stp->ls_flags & NFSLCK_ACCESSBITS) &
2385 		        ((stp->ls_flags>>NFSLCK_SHIFT) & NFSLCK_ACCESSBITS))||
2386 		       ((stp->ls_flags & NFSLCK_ACCESSBITS) &
2387 		        ((new_stp->ls_flags>>NFSLCK_SHIFT)&NFSLCK_ACCESSBITS))){
2388 			ret = nfsrv_clientconflict(stp->ls_clp,&haslock,vp,p);
2389 			if (ret) {
2390 				/*
2391 				 * nfsrv_clientconflict() unlocks state
2392 				 * when it returns non-zero.
2393 				 */
2394 				free((caddr_t)new_open, M_NFSDSTATE);
2395 				free((caddr_t)new_deleg, M_NFSDSTATE);
2396 				openstp = NULL;
2397 				goto tryagain;
2398 			}
2399 			if (new_stp->ls_flags & NFSLCK_RECLAIM)
2400 				error = NFSERR_RECLAIMCONFLICT;
2401 			else
2402 				error = NFSERR_SHAREDENIED;
2403 			NFSUNLOCKSTATE();
2404 			if (haslock) {
2405 				NFSLOCKV4ROOTMUTEX();
2406 				nfsv4_unlock(&nfsv4rootfs_lock, 1);
2407 				NFSUNLOCKV4ROOTMUTEX();
2408 			}
2409 			free((caddr_t)new_open, M_NFSDSTATE);
2410 			free((caddr_t)new_deleg, M_NFSDSTATE);
2411 			printf("nfsd openctrl unexpected client cnfl\n");
2412 			return (error);
2413 		    }
2414 		}
2415 	}
2416 
2417 	/*
2418 	 * Check for a conflicting delegation. If one is found, call
2419 	 * nfsrv_delegconflict() to handle it. If the v4root lock hasn't
2420 	 * been set yet, it will get the lock. Otherwise, it will recall
2421 	 * the delegation. Then, we try try again...
2422 	 * (If NFSLCK_DELEGCUR is set, it has a delegation, so there
2423 	 *  isn't a conflict.)
2424 	 * I currently believe the conflict algorithm to be:
2425 	 * For Open with Read Access and Deny None
2426 	 * - there is a conflict iff a different client has a write delegation
2427 	 * For Open with other Write Access or any Deny except None
2428 	 * - there is a conflict if a different client has any delegation
2429 	 * - there is a conflict if the same client has a read delegation
2430 	 *   (The current concensus is that this last case should be
2431 	 *    considered a conflict since the client with a read delegation
2432 	 *    could have done an Open with ReadAccess and WriteDeny
2433 	 *    locally and then not have checked for the WriteDeny.)
2434 	 */
2435 	if (!(new_stp->ls_flags & (NFSLCK_DELEGPREV | NFSLCK_DELEGCUR))) {
2436 	    stp = LIST_FIRST(&lfp->lf_deleg);
2437 	    while (stp != LIST_END(&lfp->lf_deleg)) {
2438 		nstp = LIST_NEXT(stp, ls_file);
2439 		if (stp->ls_clp != clp && (stp->ls_flags & NFSLCK_DELEGREAD))
2440 			writedeleg = 0;
2441 		else
2442 			delegate = 0;
2443 		if ((readonly && stp->ls_clp != clp &&
2444 		       (stp->ls_flags & NFSLCK_DELEGWRITE)) ||
2445 		    (!readonly && (stp->ls_clp != clp ||
2446 		         (stp->ls_flags & NFSLCK_DELEGREAD)))) {
2447 		    if (new_stp->ls_flags & NFSLCK_RECLAIM) {
2448 			delegate = 2;
2449 		    } else {
2450 			ret = nfsrv_delegconflict(stp, &haslock, p, vp);
2451 			if (ret) {
2452 			    /*
2453 			     * nfsrv_delegconflict() unlocks state
2454 			     * when it returns non-zero.
2455 			     */
2456 			    printf("Nfsd openctrl unexpected deleg cnfl\n");
2457 			    free((caddr_t)new_open, M_NFSDSTATE);
2458 			    free((caddr_t)new_deleg, M_NFSDSTATE);
2459 			    if (ret == -1) {
2460 				openstp = NULL;
2461 				goto tryagain;
2462 			    }
2463 			    return (ret);
2464 			}
2465 		    }
2466 		}
2467 		stp = nstp;
2468 	    }
2469 	}
2470 
2471 	/*
2472 	 * We only get here if there was no open that conflicted.
2473 	 * If an open for the owner exists, or in the access/deny bits.
2474 	 * Otherwise it is a new open. If the open_owner hasn't been
2475 	 * confirmed, replace the open with the new one needing confirmation,
2476 	 * otherwise add the open.
2477 	 */
2478 	if (new_stp->ls_flags & NFSLCK_DELEGPREV) {
2479 	    /*
2480 	     * Handle NFSLCK_DELEGPREV by searching the old delegations for
2481 	     * a match. If found, just move the old delegation to the current
2482 	     * delegation list and issue open. If not found, return
2483 	     * NFSERR_EXPIRED.
2484 	     */
2485 	    LIST_FOREACH(stp, &clp->lc_olddeleg, ls_list) {
2486 		if (stp->ls_lfp == lfp) {
2487 		    /* Found it */
2488 		    if (stp->ls_clp != clp)
2489 			panic("olddeleg clp");
2490 		    LIST_REMOVE(stp, ls_list);
2491 		    LIST_REMOVE(stp, ls_hash);
2492 		    stp->ls_flags &= ~NFSLCK_OLDDELEG;
2493 		    stp->ls_stateid.seqid = delegstateidp->seqid = 0;
2494 		    stp->ls_stateid.other[0] = delegstateidp->other[0] =
2495 			clp->lc_clientid.lval[0];
2496 		    stp->ls_stateid.other[1] = delegstateidp->other[1] =
2497 			clp->lc_clientid.lval[1];
2498 		    stp->ls_stateid.other[2] = delegstateidp->other[2] =
2499 			nfsrv_nextstateindex(clp);
2500 		    stp->ls_compref = nd->nd_compref;
2501 		    LIST_INSERT_HEAD(&clp->lc_deleg, stp, ls_list);
2502 		    LIST_INSERT_HEAD(NFSSTATEHASH(clp,
2503 			stp->ls_stateid), stp, ls_hash);
2504 		    if (stp->ls_flags & NFSLCK_DELEGWRITE)
2505 			*rflagsp |= NFSV4OPEN_WRITEDELEGATE;
2506 		    else
2507 			*rflagsp |= NFSV4OPEN_READDELEGATE;
2508 		    clp->lc_delegtime = NFSD_MONOSEC +
2509 			nfsrv_lease + NFSRV_LEASEDELTA;
2510 
2511 		    /*
2512 		     * Now, do the associated open.
2513 		     */
2514 		    new_open->ls_stateid.seqid = 0;
2515 		    new_open->ls_stateid.other[0] = clp->lc_clientid.lval[0];
2516 		    new_open->ls_stateid.other[1] = clp->lc_clientid.lval[1];
2517 		    new_open->ls_stateid.other[2] = nfsrv_nextstateindex(clp);
2518 		    new_open->ls_flags = (new_stp->ls_flags&NFSLCK_DENYBITS)|
2519 			NFSLCK_OPEN;
2520 		    if (stp->ls_flags & NFSLCK_DELEGWRITE)
2521 			new_open->ls_flags |= (NFSLCK_READACCESS |
2522 			    NFSLCK_WRITEACCESS);
2523 		    else
2524 			new_open->ls_flags |= NFSLCK_READACCESS;
2525 		    new_open->ls_uid = new_stp->ls_uid;
2526 		    new_open->ls_lfp = lfp;
2527 		    new_open->ls_clp = clp;
2528 		    LIST_INIT(&new_open->ls_open);
2529 		    LIST_INSERT_HEAD(&lfp->lf_open, new_open, ls_file);
2530 		    LIST_INSERT_HEAD(NFSSTATEHASH(clp, new_open->ls_stateid),
2531 			new_open, ls_hash);
2532 		    /*
2533 		     * and handle the open owner
2534 		     */
2535 		    if (ownerstp) {
2536 			new_open->ls_openowner = ownerstp;
2537 			LIST_INSERT_HEAD(&ownerstp->ls_open,new_open,ls_list);
2538 		    } else {
2539 			new_open->ls_openowner = new_stp;
2540 			new_stp->ls_flags = 0;
2541 			nfsrvd_refcache(new_stp->ls_op);
2542 			new_stp->ls_noopens = 0;
2543 			LIST_INIT(&new_stp->ls_open);
2544 			LIST_INSERT_HEAD(&new_stp->ls_open, new_open, ls_list);
2545 			LIST_INSERT_HEAD(&clp->lc_open, new_stp, ls_list);
2546 			*new_stpp = NULL;
2547 			newnfsstats.srvopenowners++;
2548 			nfsrv_openpluslock++;
2549 		    }
2550 		    openstp = new_open;
2551 		    new_open = NULL;
2552 		    newnfsstats.srvopens++;
2553 		    nfsrv_openpluslock++;
2554 		    break;
2555 		}
2556 	    }
2557 	    if (stp == LIST_END(&clp->lc_olddeleg))
2558 		error = NFSERR_EXPIRED;
2559 	} else if (new_stp->ls_flags & (NFSLCK_DELEGREAD | NFSLCK_DELEGWRITE)) {
2560 	    /*
2561 	     * Scan to see that no delegation for this client and file
2562 	     * doesn't already exist.
2563 	     * There also shouldn't yet be an Open for this file and
2564 	     * openowner.
2565 	     */
2566 	    LIST_FOREACH(stp, &lfp->lf_deleg, ls_file) {
2567 		if (stp->ls_clp == clp)
2568 		    break;
2569 	    }
2570 	    if (stp == LIST_END(&lfp->lf_deleg) && openstp == NULL) {
2571 		/*
2572 		 * This is the Claim_Previous case with a delegation
2573 		 * type != Delegate_None.
2574 		 */
2575 		/*
2576 		 * First, add the delegation. (Although we must issue the
2577 		 * delegation, we can also ask for an immediate return.)
2578 		 */
2579 		new_deleg->ls_stateid.seqid = delegstateidp->seqid = 0;
2580 		new_deleg->ls_stateid.other[0] = delegstateidp->other[0] =
2581 		    clp->lc_clientid.lval[0];
2582 		new_deleg->ls_stateid.other[1] = delegstateidp->other[1] =
2583 		    clp->lc_clientid.lval[1];
2584 		new_deleg->ls_stateid.other[2] = delegstateidp->other[2] =
2585 		    nfsrv_nextstateindex(clp);
2586 		if (new_stp->ls_flags & NFSLCK_DELEGWRITE) {
2587 		    new_deleg->ls_flags = (NFSLCK_DELEGWRITE |
2588 			NFSLCK_READACCESS | NFSLCK_WRITEACCESS);
2589 		    *rflagsp |= NFSV4OPEN_WRITEDELEGATE;
2590 		} else {
2591 		    new_deleg->ls_flags = (NFSLCK_DELEGREAD |
2592 			NFSLCK_READACCESS);
2593 		    *rflagsp |= NFSV4OPEN_READDELEGATE;
2594 		}
2595 		new_deleg->ls_uid = new_stp->ls_uid;
2596 		new_deleg->ls_lfp = lfp;
2597 		new_deleg->ls_clp = clp;
2598 		new_deleg->ls_filerev = filerev;
2599 		new_deleg->ls_compref = nd->nd_compref;
2600 		LIST_INSERT_HEAD(&lfp->lf_deleg, new_deleg, ls_file);
2601 		LIST_INSERT_HEAD(NFSSTATEHASH(clp,
2602 		    new_deleg->ls_stateid), new_deleg, ls_hash);
2603 		LIST_INSERT_HEAD(&clp->lc_deleg, new_deleg, ls_list);
2604 		new_deleg = NULL;
2605 		if (delegate == 2 || nfsrv_issuedelegs == 0 ||
2606 		    (clp->lc_flags & (LCL_CALLBACKSON | LCL_CBDOWN)) !=
2607 		     LCL_CALLBACKSON ||
2608 		    NFSRV_V4DELEGLIMIT(nfsrv_delegatecnt) ||
2609 		    !NFSVNO_DELEGOK(vp))
2610 		    *rflagsp |= NFSV4OPEN_RECALL;
2611 		newnfsstats.srvdelegates++;
2612 		nfsrv_openpluslock++;
2613 		nfsrv_delegatecnt++;
2614 
2615 		/*
2616 		 * Now, do the associated open.
2617 		 */
2618 		new_open->ls_stateid.seqid = 0;
2619 		new_open->ls_stateid.other[0] = clp->lc_clientid.lval[0];
2620 		new_open->ls_stateid.other[1] = clp->lc_clientid.lval[1];
2621 		new_open->ls_stateid.other[2] = nfsrv_nextstateindex(clp);
2622 		new_open->ls_flags = (new_stp->ls_flags & NFSLCK_DENYBITS) |
2623 		    NFSLCK_OPEN;
2624 		if (new_stp->ls_flags & NFSLCK_DELEGWRITE)
2625 			new_open->ls_flags |= (NFSLCK_READACCESS |
2626 			    NFSLCK_WRITEACCESS);
2627 		else
2628 			new_open->ls_flags |= NFSLCK_READACCESS;
2629 		new_open->ls_uid = new_stp->ls_uid;
2630 		new_open->ls_lfp = lfp;
2631 		new_open->ls_clp = clp;
2632 		LIST_INIT(&new_open->ls_open);
2633 		LIST_INSERT_HEAD(&lfp->lf_open, new_open, ls_file);
2634 		LIST_INSERT_HEAD(NFSSTATEHASH(clp, new_open->ls_stateid),
2635 		   new_open, ls_hash);
2636 		/*
2637 		 * and handle the open owner
2638 		 */
2639 		if (ownerstp) {
2640 		    new_open->ls_openowner = ownerstp;
2641 		    LIST_INSERT_HEAD(&ownerstp->ls_open, new_open, ls_list);
2642 		} else {
2643 		    new_open->ls_openowner = new_stp;
2644 		    new_stp->ls_flags = 0;
2645 		    nfsrvd_refcache(new_stp->ls_op);
2646 		    new_stp->ls_noopens = 0;
2647 		    LIST_INIT(&new_stp->ls_open);
2648 		    LIST_INSERT_HEAD(&new_stp->ls_open, new_open, ls_list);
2649 		    LIST_INSERT_HEAD(&clp->lc_open, new_stp, ls_list);
2650 		    *new_stpp = NULL;
2651 		    newnfsstats.srvopenowners++;
2652 		    nfsrv_openpluslock++;
2653 		}
2654 		openstp = new_open;
2655 		new_open = NULL;
2656 		newnfsstats.srvopens++;
2657 		nfsrv_openpluslock++;
2658 	    } else {
2659 		error = NFSERR_RECLAIMCONFLICT;
2660 	    }
2661 	} else if (ownerstp) {
2662 		if (ownerstp->ls_flags & NFSLCK_NEEDSCONFIRM) {
2663 		    /* Replace the open */
2664 		    if (ownerstp->ls_op)
2665 			nfsrvd_derefcache(ownerstp->ls_op);
2666 		    ownerstp->ls_op = new_stp->ls_op;
2667 		    nfsrvd_refcache(ownerstp->ls_op);
2668 		    ownerstp->ls_seq = new_stp->ls_seq;
2669 		    *rflagsp |= NFSV4OPEN_RESULTCONFIRM;
2670 		    stp = LIST_FIRST(&ownerstp->ls_open);
2671 		    stp->ls_flags = (new_stp->ls_flags & NFSLCK_SHAREBITS) |
2672 			NFSLCK_OPEN;
2673 		    stp->ls_stateid.seqid = 0;
2674 		    stp->ls_uid = new_stp->ls_uid;
2675 		    if (lfp != stp->ls_lfp) {
2676 			LIST_REMOVE(stp, ls_file);
2677 			LIST_INSERT_HEAD(&lfp->lf_open, stp, ls_file);
2678 			stp->ls_lfp = lfp;
2679 		    }
2680 		    openstp = stp;
2681 		} else if (openstp) {
2682 		    openstp->ls_flags |= (new_stp->ls_flags & NFSLCK_SHAREBITS);
2683 		    openstp->ls_stateid.seqid++;
2684 
2685 		    /*
2686 		     * This is where we can choose to issue a delegation.
2687 		     */
2688 		    if (delegate && nfsrv_issuedelegs &&
2689 			writedeleg && !NFSVNO_EXRDONLY(exp) &&
2690 			(nfsrv_writedelegifpos || !readonly) &&
2691 			(clp->lc_flags & (LCL_CALLBACKSON | LCL_CBDOWN)) ==
2692 			 LCL_CALLBACKSON &&
2693 			!NFSRV_V4DELEGLIMIT(nfsrv_delegatecnt) &&
2694 			NFSVNO_DELEGOK(vp)) {
2695 			new_deleg->ls_stateid.seqid = delegstateidp->seqid = 0;
2696 			new_deleg->ls_stateid.other[0] = delegstateidp->other[0]
2697 			    = clp->lc_clientid.lval[0];
2698 			new_deleg->ls_stateid.other[1] = delegstateidp->other[1]
2699 			    = clp->lc_clientid.lval[1];
2700 			new_deleg->ls_stateid.other[2] = delegstateidp->other[2]
2701 			    = nfsrv_nextstateindex(clp);
2702 			new_deleg->ls_flags = (NFSLCK_DELEGWRITE |
2703 			    NFSLCK_READACCESS | NFSLCK_WRITEACCESS);
2704 			*rflagsp |= NFSV4OPEN_WRITEDELEGATE;
2705 			new_deleg->ls_uid = new_stp->ls_uid;
2706 			new_deleg->ls_lfp = lfp;
2707 			new_deleg->ls_clp = clp;
2708 			new_deleg->ls_filerev = filerev;
2709 			new_deleg->ls_compref = nd->nd_compref;
2710 			LIST_INSERT_HEAD(&lfp->lf_deleg, new_deleg, ls_file);
2711 			LIST_INSERT_HEAD(NFSSTATEHASH(clp,
2712 			    new_deleg->ls_stateid), new_deleg, ls_hash);
2713 			LIST_INSERT_HEAD(&clp->lc_deleg, new_deleg, ls_list);
2714 			new_deleg = NULL;
2715 			newnfsstats.srvdelegates++;
2716 			nfsrv_openpluslock++;
2717 			nfsrv_delegatecnt++;
2718 		    }
2719 		} else {
2720 		    new_open->ls_stateid.seqid = 0;
2721 		    new_open->ls_stateid.other[0] = clp->lc_clientid.lval[0];
2722 		    new_open->ls_stateid.other[1] = clp->lc_clientid.lval[1];
2723 		    new_open->ls_stateid.other[2] = nfsrv_nextstateindex(clp);
2724 		    new_open->ls_flags = (new_stp->ls_flags & NFSLCK_SHAREBITS)|
2725 			NFSLCK_OPEN;
2726 		    new_open->ls_uid = new_stp->ls_uid;
2727 		    new_open->ls_openowner = ownerstp;
2728 		    new_open->ls_lfp = lfp;
2729 		    new_open->ls_clp = clp;
2730 		    LIST_INIT(&new_open->ls_open);
2731 		    LIST_INSERT_HEAD(&lfp->lf_open, new_open, ls_file);
2732 		    LIST_INSERT_HEAD(&ownerstp->ls_open, new_open, ls_list);
2733 		    LIST_INSERT_HEAD(NFSSTATEHASH(clp, new_open->ls_stateid),
2734 			new_open, ls_hash);
2735 		    openstp = new_open;
2736 		    new_open = NULL;
2737 		    newnfsstats.srvopens++;
2738 		    nfsrv_openpluslock++;
2739 
2740 		    /*
2741 		     * This is where we can choose to issue a delegation.
2742 		     */
2743 		    if (delegate && nfsrv_issuedelegs &&
2744 			(writedeleg || readonly) &&
2745 			(clp->lc_flags & (LCL_CALLBACKSON | LCL_CBDOWN)) ==
2746 			 LCL_CALLBACKSON &&
2747 			!NFSRV_V4DELEGLIMIT(nfsrv_delegatecnt) &&
2748 			NFSVNO_DELEGOK(vp)) {
2749 			new_deleg->ls_stateid.seqid = delegstateidp->seqid = 0;
2750 			new_deleg->ls_stateid.other[0] = delegstateidp->other[0]
2751 			    = clp->lc_clientid.lval[0];
2752 			new_deleg->ls_stateid.other[1] = delegstateidp->other[1]
2753 			    = clp->lc_clientid.lval[1];
2754 			new_deleg->ls_stateid.other[2] = delegstateidp->other[2]
2755 			    = nfsrv_nextstateindex(clp);
2756 			if (writedeleg && !NFSVNO_EXRDONLY(exp) &&
2757 			    (nfsrv_writedelegifpos || !readonly)) {
2758 			    new_deleg->ls_flags = (NFSLCK_DELEGWRITE |
2759 				NFSLCK_READACCESS | NFSLCK_WRITEACCESS);
2760 			    *rflagsp |= NFSV4OPEN_WRITEDELEGATE;
2761 			} else {
2762 			    new_deleg->ls_flags = (NFSLCK_DELEGREAD |
2763 				NFSLCK_READACCESS);
2764 			    *rflagsp |= NFSV4OPEN_READDELEGATE;
2765 			}
2766 			new_deleg->ls_uid = new_stp->ls_uid;
2767 			new_deleg->ls_lfp = lfp;
2768 			new_deleg->ls_clp = clp;
2769 			new_deleg->ls_filerev = filerev;
2770 			new_deleg->ls_compref = nd->nd_compref;
2771 			LIST_INSERT_HEAD(&lfp->lf_deleg, new_deleg, ls_file);
2772 			LIST_INSERT_HEAD(NFSSTATEHASH(clp,
2773 			    new_deleg->ls_stateid), new_deleg, ls_hash);
2774 			LIST_INSERT_HEAD(&clp->lc_deleg, new_deleg, ls_list);
2775 			new_deleg = NULL;
2776 			newnfsstats.srvdelegates++;
2777 			nfsrv_openpluslock++;
2778 			nfsrv_delegatecnt++;
2779 		    }
2780 		}
2781 	} else {
2782 		/*
2783 		 * New owner case. Start the open_owner sequence with a
2784 		 * Needs confirmation (unless a reclaim) and hang the
2785 		 * new open off it.
2786 		 */
2787 		new_open->ls_stateid.seqid = 0;
2788 		new_open->ls_stateid.other[0] = clp->lc_clientid.lval[0];
2789 		new_open->ls_stateid.other[1] = clp->lc_clientid.lval[1];
2790 		new_open->ls_stateid.other[2] = nfsrv_nextstateindex(clp);
2791 		new_open->ls_flags = (new_stp->ls_flags & NFSLCK_SHAREBITS) |
2792 		    NFSLCK_OPEN;
2793 		new_open->ls_uid = new_stp->ls_uid;
2794 		LIST_INIT(&new_open->ls_open);
2795 		new_open->ls_openowner = new_stp;
2796 		new_open->ls_lfp = lfp;
2797 		new_open->ls_clp = clp;
2798 		LIST_INSERT_HEAD(&lfp->lf_open, new_open, ls_file);
2799 		if (new_stp->ls_flags & NFSLCK_RECLAIM) {
2800 			new_stp->ls_flags = 0;
2801 		} else {
2802 			*rflagsp |= NFSV4OPEN_RESULTCONFIRM;
2803 			new_stp->ls_flags = NFSLCK_NEEDSCONFIRM;
2804 		}
2805 		nfsrvd_refcache(new_stp->ls_op);
2806 		new_stp->ls_noopens = 0;
2807 		LIST_INIT(&new_stp->ls_open);
2808 		LIST_INSERT_HEAD(&new_stp->ls_open, new_open, ls_list);
2809 		LIST_INSERT_HEAD(&clp->lc_open, new_stp, ls_list);
2810 		LIST_INSERT_HEAD(NFSSTATEHASH(clp, new_open->ls_stateid),
2811 		    new_open, ls_hash);
2812 		openstp = new_open;
2813 		new_open = NULL;
2814 		*new_stpp = NULL;
2815 		newnfsstats.srvopens++;
2816 		nfsrv_openpluslock++;
2817 		newnfsstats.srvopenowners++;
2818 		nfsrv_openpluslock++;
2819 	}
2820 	if (!error) {
2821 		stateidp->seqid = openstp->ls_stateid.seqid;
2822 		stateidp->other[0] = openstp->ls_stateid.other[0];
2823 		stateidp->other[1] = openstp->ls_stateid.other[1];
2824 		stateidp->other[2] = openstp->ls_stateid.other[2];
2825 	}
2826 	NFSUNLOCKSTATE();
2827 	if (haslock) {
2828 		NFSLOCKV4ROOTMUTEX();
2829 		nfsv4_unlock(&nfsv4rootfs_lock, 1);
2830 		NFSUNLOCKV4ROOTMUTEX();
2831 	}
2832 	if (new_open)
2833 		FREE((caddr_t)new_open, M_NFSDSTATE);
2834 	if (new_deleg)
2835 		FREE((caddr_t)new_deleg, M_NFSDSTATE);
2836 	return (error);
2837 }
2838 
2839 /*
2840  * Open update. Does the confirm, downgrade and close.
2841  */
2842 APPLESTATIC int
2843 nfsrv_openupdate(vnode_t vp, struct nfsstate *new_stp, nfsquad_t clientid,
2844     nfsv4stateid_t *stateidp, struct nfsrv_descript *nd, NFSPROC_T *p)
2845 {
2846 	struct nfsstate *stp, *ownerstp;
2847 	struct nfsclient *clp;
2848 	struct nfslockfile *lfp;
2849 	u_int32_t bits;
2850 	int error, gotstate = 0, len = 0;
2851 	u_char client[NFSV4_OPAQUELIMIT];
2852 
2853 	/*
2854 	 * Check for restart conditions (client and server).
2855 	 */
2856 	error = nfsrv_checkrestart(clientid, new_stp->ls_flags,
2857 	    &new_stp->ls_stateid, 0);
2858 	if (error)
2859 		return (error);
2860 
2861 	NFSLOCKSTATE();
2862 	/*
2863 	 * Get the open structure via clientid and stateid.
2864 	 */
2865 	error = nfsrv_getclient(clientid, CLOPS_RENEW, &clp,
2866 	    (nfsquad_t)((u_quad_t)0), NULL, p);
2867 	if (!error)
2868 		error = nfsrv_getstate(clp, &new_stp->ls_stateid,
2869 		    new_stp->ls_flags, &stp);
2870 
2871 	/*
2872 	 * Sanity check the open.
2873 	 */
2874 	if (!error && (!(stp->ls_flags & NFSLCK_OPEN) ||
2875 		(!(new_stp->ls_flags & NFSLCK_CONFIRM) &&
2876 		 (stp->ls_openowner->ls_flags & NFSLCK_NEEDSCONFIRM)) ||
2877 		((new_stp->ls_flags & NFSLCK_CONFIRM) &&
2878 		 (!(stp->ls_openowner->ls_flags & NFSLCK_NEEDSCONFIRM)))))
2879 		error = NFSERR_BADSTATEID;
2880 
2881 	if (!error)
2882 		error = nfsrv_checkseqid(nd, new_stp->ls_seq,
2883 		    stp->ls_openowner, new_stp->ls_op);
2884 	if (!error && stp->ls_stateid.seqid != new_stp->ls_stateid.seqid &&
2885 	    !(new_stp->ls_flags & NFSLCK_CONFIRM))
2886 		error = NFSERR_OLDSTATEID;
2887 	if (!error && vnode_vtype(vp) != VREG) {
2888 		if (vnode_vtype(vp) == VDIR)
2889 			error = NFSERR_ISDIR;
2890 		else
2891 			error = NFSERR_INVAL;
2892 	}
2893 
2894 	if (error) {
2895 		/*
2896 		 * If a client tries to confirm an Open with a bad
2897 		 * seqid# and there are no byte range locks or other Opens
2898 		 * on the openowner, just throw it away, so the next use of the
2899 		 * openowner will start a fresh seq#.
2900 		 */
2901 		if (error == NFSERR_BADSEQID &&
2902 		    (new_stp->ls_flags & NFSLCK_CONFIRM) &&
2903 		    nfsrv_nootherstate(stp))
2904 			nfsrv_freeopenowner(stp->ls_openowner, 0, p);
2905 		NFSUNLOCKSTATE();
2906 		return (error);
2907 	}
2908 
2909 	/*
2910 	 * Set the return stateid.
2911 	 */
2912 	stateidp->seqid = stp->ls_stateid.seqid + 1;
2913 	stateidp->other[0] = stp->ls_stateid.other[0];
2914 	stateidp->other[1] = stp->ls_stateid.other[1];
2915 	stateidp->other[2] = stp->ls_stateid.other[2];
2916 	/*
2917 	 * Now, handle the three cases.
2918 	 */
2919 	if (new_stp->ls_flags & NFSLCK_CONFIRM) {
2920 		/*
2921 		 * If the open doesn't need confirmation, it seems to me that
2922 		 * there is a client error, but I'll just log it and keep going?
2923 		 */
2924 		if (!(stp->ls_openowner->ls_flags & NFSLCK_NEEDSCONFIRM))
2925 			printf("Nfsv4d: stray open confirm\n");
2926 		stp->ls_openowner->ls_flags = 0;
2927 		stp->ls_stateid.seqid++;
2928 		if (!(clp->lc_flags & LCL_STAMPEDSTABLE)) {
2929 			clp->lc_flags |= LCL_STAMPEDSTABLE;
2930 			len = clp->lc_idlen;
2931 			NFSBCOPY(clp->lc_id, client, len);
2932 			gotstate = 1;
2933 		}
2934 		NFSUNLOCKSTATE();
2935 	} else if (new_stp->ls_flags & NFSLCK_CLOSE) {
2936 		ownerstp = stp->ls_openowner;
2937 		lfp = stp->ls_lfp;
2938 		if (nfsrv_dolocallocks != 0 && !LIST_EMPTY(&stp->ls_open)) {
2939 			/* Get the lf lock */
2940 			nfsrv_locklf(lfp);
2941 			NFSUNLOCKSTATE();
2942 			if (nfsrv_freeopen(stp, vp, 1, p) == 0) {
2943 				NFSLOCKSTATE();
2944 				nfsrv_unlocklf(lfp);
2945 				NFSUNLOCKSTATE();
2946 			}
2947 		} else {
2948 			(void) nfsrv_freeopen(stp, NULL, 0, p);
2949 			NFSUNLOCKSTATE();
2950 		}
2951 	} else {
2952 		/*
2953 		 * Update the share bits, making sure that the new set are a
2954 		 * subset of the old ones.
2955 		 */
2956 		bits = (new_stp->ls_flags & NFSLCK_SHAREBITS);
2957 		if (~(stp->ls_flags) & bits) {
2958 			NFSUNLOCKSTATE();
2959 			return (NFSERR_INVAL);
2960 		}
2961 		stp->ls_flags = (bits | NFSLCK_OPEN);
2962 		stp->ls_stateid.seqid++;
2963 		NFSUNLOCKSTATE();
2964 	}
2965 
2966 	/*
2967 	 * If the client just confirmed its first open, write a timestamp
2968 	 * to the stable storage file.
2969 	 */
2970 	if (gotstate)
2971 		nfsrv_writestable(client, len, NFSNST_NEWSTATE, p);
2972 	return (error);
2973 }
2974 
2975 /*
2976  * Delegation update. Does the purge and return.
2977  */
2978 APPLESTATIC int
2979 nfsrv_delegupdate(nfsquad_t clientid, nfsv4stateid_t *stateidp,
2980     vnode_t vp, int op, struct ucred *cred, NFSPROC_T *p)
2981 {
2982 	struct nfsstate *stp;
2983 	struct nfsclient *clp;
2984 	int error;
2985 	fhandle_t fh;
2986 
2987 	/*
2988 	 * Do a sanity check against the file handle for DelegReturn.
2989 	 */
2990 	if (vp) {
2991 		error = nfsvno_getfh(vp, &fh, p);
2992 		if (error)
2993 			return (error);
2994 	}
2995 	/*
2996 	 * Check for restart conditions (client and server).
2997 	 */
2998 	if (op == NFSV4OP_DELEGRETURN)
2999 		error = nfsrv_checkrestart(clientid, NFSLCK_DELEGRETURN,
3000 			stateidp, 0);
3001 	else
3002 		error = nfsrv_checkrestart(clientid, NFSLCK_DELEGPURGE,
3003 			stateidp, 0);
3004 
3005 	NFSLOCKSTATE();
3006 	/*
3007 	 * Get the open structure via clientid and stateid.
3008 	 */
3009 	if (!error)
3010 	    error = nfsrv_getclient(clientid, CLOPS_RENEW, &clp,
3011 		(nfsquad_t)((u_quad_t)0), NULL, p);
3012 	if (error) {
3013 		if (error == NFSERR_CBPATHDOWN)
3014 			error = 0;
3015 		if (error == NFSERR_STALECLIENTID && op == NFSV4OP_DELEGRETURN)
3016 			error = NFSERR_STALESTATEID;
3017 	}
3018 	if (!error && op == NFSV4OP_DELEGRETURN) {
3019 	    error = nfsrv_getstate(clp, stateidp, NFSLCK_DELEGRETURN, &stp);
3020 	    if (!error && stp->ls_stateid.seqid != stateidp->seqid)
3021 		error = NFSERR_OLDSTATEID;
3022 	}
3023 	/*
3024 	 * NFSERR_EXPIRED means that the state has gone away,
3025 	 * so Delegations have been purged. Just return ok.
3026 	 */
3027 	if (error == NFSERR_EXPIRED && op == NFSV4OP_DELEGPURGE) {
3028 		NFSUNLOCKSTATE();
3029 		return (0);
3030 	}
3031 	if (error) {
3032 		NFSUNLOCKSTATE();
3033 		return (error);
3034 	}
3035 
3036 	if (op == NFSV4OP_DELEGRETURN) {
3037 		if (NFSBCMP((caddr_t)&fh, (caddr_t)&stp->ls_lfp->lf_fh,
3038 		    sizeof (fhandle_t))) {
3039 			NFSUNLOCKSTATE();
3040 			return (NFSERR_BADSTATEID);
3041 		}
3042 		nfsrv_freedeleg(stp);
3043 	} else {
3044 		nfsrv_freedeleglist(&clp->lc_olddeleg);
3045 	}
3046 	NFSUNLOCKSTATE();
3047 	return (0);
3048 }
3049 
3050 /*
3051  * Release lock owner.
3052  */
3053 APPLESTATIC int
3054 nfsrv_releaselckown(struct nfsstate *new_stp, nfsquad_t clientid,
3055     NFSPROC_T *p)
3056 {
3057 	struct nfsstate *stp, *nstp, *openstp, *ownstp;
3058 	struct nfsclient *clp;
3059 	int error;
3060 
3061 	/*
3062 	 * Check for restart conditions (client and server).
3063 	 */
3064 	error = nfsrv_checkrestart(clientid, new_stp->ls_flags,
3065 	    &new_stp->ls_stateid, 0);
3066 	if (error)
3067 		return (error);
3068 
3069 	NFSLOCKSTATE();
3070 	/*
3071 	 * Get the lock owner by name.
3072 	 */
3073 	error = nfsrv_getclient(clientid, CLOPS_RENEW, &clp,
3074 	    (nfsquad_t)((u_quad_t)0), NULL, p);
3075 	if (error) {
3076 		NFSUNLOCKSTATE();
3077 		return (error);
3078 	}
3079 	LIST_FOREACH(ownstp, &clp->lc_open, ls_list) {
3080 	    LIST_FOREACH(openstp, &ownstp->ls_open, ls_list) {
3081 		stp = LIST_FIRST(&openstp->ls_open);
3082 		while (stp != LIST_END(&openstp->ls_open)) {
3083 		    nstp = LIST_NEXT(stp, ls_list);
3084 		    /*
3085 		     * If the owner matches, check for locks and
3086 		     * then free or return an error.
3087 		     */
3088 		    if (stp->ls_ownerlen == new_stp->ls_ownerlen &&
3089 			!NFSBCMP(stp->ls_owner, new_stp->ls_owner,
3090 			 stp->ls_ownerlen)){
3091 			if (LIST_EMPTY(&stp->ls_lock)) {
3092 			    nfsrv_freelockowner(stp, NULL, 0, p);
3093 			} else {
3094 			    NFSUNLOCKSTATE();
3095 			    return (NFSERR_LOCKSHELD);
3096 			}
3097 		    }
3098 		    stp = nstp;
3099 		}
3100 	    }
3101 	}
3102 	NFSUNLOCKSTATE();
3103 	return (0);
3104 }
3105 
3106 /*
3107  * Get the file handle for a lock structure.
3108  */
3109 static int
3110 nfsrv_getlockfh(vnode_t vp, u_short flags,
3111     struct nfslockfile **new_lfpp, fhandle_t *nfhp, NFSPROC_T *p)
3112 {
3113 	fhandle_t *fhp = NULL;
3114 	struct nfslockfile *new_lfp;
3115 	int error;
3116 
3117 	/*
3118 	 * For lock, use the new nfslock structure, otherwise just
3119 	 * a fhandle_t on the stack.
3120 	 */
3121 	if (flags & NFSLCK_OPEN) {
3122 		new_lfp = *new_lfpp;
3123 		fhp = &new_lfp->lf_fh;
3124 	} else if (nfhp) {
3125 		fhp = nfhp;
3126 	} else {
3127 		panic("nfsrv_getlockfh");
3128 	}
3129 	error = nfsvno_getfh(vp, fhp, p);
3130 	return (error);
3131 }
3132 
3133 /*
3134  * Get an nfs lock structure. Allocate one, as required, and return a
3135  * pointer to it.
3136  * Returns an NFSERR_xxx upon failure or -1 to indicate no current lock.
3137  */
3138 static int
3139 nfsrv_getlockfile(u_short flags, struct nfslockfile **new_lfpp,
3140     struct nfslockfile **lfpp, fhandle_t *nfhp, int lockit)
3141 {
3142 	struct nfslockfile *lfp;
3143 	fhandle_t *fhp = NULL, *tfhp;
3144 	struct nfslockhashhead *hp;
3145 	struct nfslockfile *new_lfp = NULL;
3146 
3147 	/*
3148 	 * For lock, use the new nfslock structure, otherwise just
3149 	 * a fhandle_t on the stack.
3150 	 */
3151 	if (flags & NFSLCK_OPEN) {
3152 		new_lfp = *new_lfpp;
3153 		fhp = &new_lfp->lf_fh;
3154 	} else if (nfhp) {
3155 		fhp = nfhp;
3156 	} else {
3157 		panic("nfsrv_getlockfile");
3158 	}
3159 
3160 	hp = NFSLOCKHASH(fhp);
3161 	LIST_FOREACH(lfp, hp, lf_hash) {
3162 		tfhp = &lfp->lf_fh;
3163 		if (NFSVNO_CMPFH(fhp, tfhp)) {
3164 			if (lockit)
3165 				nfsrv_locklf(lfp);
3166 			*lfpp = lfp;
3167 			return (0);
3168 		}
3169 	}
3170 	if (!(flags & NFSLCK_OPEN))
3171 		return (-1);
3172 
3173 	/*
3174 	 * No match, so chain the new one into the list.
3175 	 */
3176 	LIST_INIT(&new_lfp->lf_open);
3177 	LIST_INIT(&new_lfp->lf_lock);
3178 	LIST_INIT(&new_lfp->lf_deleg);
3179 	LIST_INIT(&new_lfp->lf_locallock);
3180 	LIST_INIT(&new_lfp->lf_rollback);
3181 	new_lfp->lf_locallock_lck.nfslock_usecnt = 0;
3182 	new_lfp->lf_locallock_lck.nfslock_lock = 0;
3183 	new_lfp->lf_usecount = 0;
3184 	LIST_INSERT_HEAD(hp, new_lfp, lf_hash);
3185 	*lfpp = new_lfp;
3186 	*new_lfpp = NULL;
3187 	return (0);
3188 }
3189 
3190 /*
3191  * This function adds a nfslock lock structure to the list for the associated
3192  * nfsstate and nfslockfile structures. It will be inserted after the
3193  * entry pointed at by insert_lop.
3194  */
3195 static void
3196 nfsrv_insertlock(struct nfslock *new_lop, struct nfslock *insert_lop,
3197     struct nfsstate *stp, struct nfslockfile *lfp)
3198 {
3199 	struct nfslock *lop, *nlop;
3200 
3201 	new_lop->lo_stp = stp;
3202 	new_lop->lo_lfp = lfp;
3203 
3204 	if (stp != NULL) {
3205 		/* Insert in increasing lo_first order */
3206 		lop = LIST_FIRST(&lfp->lf_lock);
3207 		if (lop == LIST_END(&lfp->lf_lock) ||
3208 		    new_lop->lo_first <= lop->lo_first) {
3209 			LIST_INSERT_HEAD(&lfp->lf_lock, new_lop, lo_lckfile);
3210 		} else {
3211 			nlop = LIST_NEXT(lop, lo_lckfile);
3212 			while (nlop != LIST_END(&lfp->lf_lock) &&
3213 			       nlop->lo_first < new_lop->lo_first) {
3214 				lop = nlop;
3215 				nlop = LIST_NEXT(lop, lo_lckfile);
3216 			}
3217 			LIST_INSERT_AFTER(lop, new_lop, lo_lckfile);
3218 		}
3219 	} else {
3220 		new_lop->lo_lckfile.le_prev = NULL;	/* list not used */
3221 	}
3222 
3223 	/*
3224 	 * Insert after insert_lop, which is overloaded as stp or lfp for
3225 	 * an empty list.
3226 	 */
3227 	if (stp == NULL && (struct nfslockfile *)insert_lop == lfp)
3228 		LIST_INSERT_HEAD(&lfp->lf_locallock, new_lop, lo_lckowner);
3229 	else if ((struct nfsstate *)insert_lop == stp)
3230 		LIST_INSERT_HEAD(&stp->ls_lock, new_lop, lo_lckowner);
3231 	else
3232 		LIST_INSERT_AFTER(insert_lop, new_lop, lo_lckowner);
3233 	if (stp != NULL) {
3234 		newnfsstats.srvlocks++;
3235 		nfsrv_openpluslock++;
3236 	}
3237 }
3238 
3239 /*
3240  * This function updates the locking for a lock owner and given file. It
3241  * maintains a list of lock ranges ordered on increasing file offset that
3242  * are NFSLCK_READ or NFSLCK_WRITE and non-overlapping (aka POSIX style).
3243  * It always adds new_lop to the list and sometimes uses the one pointed
3244  * at by other_lopp.
3245  */
3246 static void
3247 nfsrv_updatelock(struct nfsstate *stp, struct nfslock **new_lopp,
3248     struct nfslock **other_lopp, struct nfslockfile *lfp)
3249 {
3250 	struct nfslock *new_lop = *new_lopp;
3251 	struct nfslock *lop, *tlop, *ilop;
3252 	struct nfslock *other_lop = *other_lopp;
3253 	int unlock = 0, myfile = 0;
3254 	u_int64_t tmp;
3255 
3256 	/*
3257 	 * Work down the list until the lock is merged.
3258 	 */
3259 	if (new_lop->lo_flags & NFSLCK_UNLOCK)
3260 		unlock = 1;
3261 	if (stp != NULL) {
3262 		ilop = (struct nfslock *)stp;
3263 		lop = LIST_FIRST(&stp->ls_lock);
3264 	} else {
3265 		ilop = (struct nfslock *)lfp;
3266 		lop = LIST_FIRST(&lfp->lf_locallock);
3267 	}
3268 	while (lop != NULL) {
3269 	    /*
3270 	     * Only check locks for this file that aren't before the start of
3271 	     * new lock's range.
3272 	     */
3273 	    if (lop->lo_lfp == lfp) {
3274 	      myfile = 1;
3275 	      if (lop->lo_end >= new_lop->lo_first) {
3276 		if (new_lop->lo_end < lop->lo_first) {
3277 			/*
3278 			 * If the new lock ends before the start of the
3279 			 * current lock's range, no merge, just insert
3280 			 * the new lock.
3281 			 */
3282 			break;
3283 		}
3284 		if (new_lop->lo_flags == lop->lo_flags ||
3285 		    (new_lop->lo_first <= lop->lo_first &&
3286 		     new_lop->lo_end >= lop->lo_end)) {
3287 			/*
3288 			 * This lock can be absorbed by the new lock/unlock.
3289 			 * This happens when it covers the entire range
3290 			 * of the old lock or is contiguous
3291 			 * with the old lock and is of the same type or an
3292 			 * unlock.
3293 			 */
3294 			if (lop->lo_first < new_lop->lo_first)
3295 				new_lop->lo_first = lop->lo_first;
3296 			if (lop->lo_end > new_lop->lo_end)
3297 				new_lop->lo_end = lop->lo_end;
3298 			tlop = lop;
3299 			lop = LIST_NEXT(lop, lo_lckowner);
3300 			nfsrv_freenfslock(tlop);
3301 			continue;
3302 		}
3303 
3304 		/*
3305 		 * All these cases are for contiguous locks that are not the
3306 		 * same type, so they can't be merged.
3307 		 */
3308 		if (new_lop->lo_first <= lop->lo_first) {
3309 			/*
3310 			 * This case is where the new lock overlaps with the
3311 			 * first part of the old lock. Move the start of the
3312 			 * old lock to just past the end of the new lock. The
3313 			 * new lock will be inserted in front of the old, since
3314 			 * ilop hasn't been updated. (We are done now.)
3315 			 */
3316 			lop->lo_first = new_lop->lo_end;
3317 			break;
3318 		}
3319 		if (new_lop->lo_end >= lop->lo_end) {
3320 			/*
3321 			 * This case is where the new lock overlaps with the
3322 			 * end of the old lock's range. Move the old lock's
3323 			 * end to just before the new lock's first and insert
3324 			 * the new lock after the old lock.
3325 			 * Might not be done yet, since the new lock could
3326 			 * overlap further locks with higher ranges.
3327 			 */
3328 			lop->lo_end = new_lop->lo_first;
3329 			ilop = lop;
3330 			lop = LIST_NEXT(lop, lo_lckowner);
3331 			continue;
3332 		}
3333 		/*
3334 		 * The final case is where the new lock's range is in the
3335 		 * middle of the current lock's and splits the current lock
3336 		 * up. Use *other_lopp to handle the second part of the
3337 		 * split old lock range. (We are done now.)
3338 		 * For unlock, we use new_lop as other_lop and tmp, since
3339 		 * other_lop and new_lop are the same for this case.
3340 		 * We noted the unlock case above, so we don't need
3341 		 * new_lop->lo_flags any longer.
3342 		 */
3343 		tmp = new_lop->lo_first;
3344 		if (other_lop == NULL) {
3345 			if (!unlock)
3346 				panic("nfsd srv update unlock");
3347 			other_lop = new_lop;
3348 			*new_lopp = NULL;
3349 		}
3350 		other_lop->lo_first = new_lop->lo_end;
3351 		other_lop->lo_end = lop->lo_end;
3352 		other_lop->lo_flags = lop->lo_flags;
3353 		other_lop->lo_stp = stp;
3354 		other_lop->lo_lfp = lfp;
3355 		lop->lo_end = tmp;
3356 		nfsrv_insertlock(other_lop, lop, stp, lfp);
3357 		*other_lopp = NULL;
3358 		ilop = lop;
3359 		break;
3360 	      }
3361 	    }
3362 	    ilop = lop;
3363 	    lop = LIST_NEXT(lop, lo_lckowner);
3364 	    if (myfile && (lop == NULL || lop->lo_lfp != lfp))
3365 		break;
3366 	}
3367 
3368 	/*
3369 	 * Insert the new lock in the list at the appropriate place.
3370 	 */
3371 	if (!unlock) {
3372 		nfsrv_insertlock(new_lop, ilop, stp, lfp);
3373 		*new_lopp = NULL;
3374 	}
3375 }
3376 
3377 /*
3378  * This function handles sequencing of locks, etc.
3379  * It returns an error that indicates what the caller should do.
3380  */
3381 static int
3382 nfsrv_checkseqid(struct nfsrv_descript *nd, u_int32_t seqid,
3383     struct nfsstate *stp, struct nfsrvcache *op)
3384 {
3385 
3386 	if (op != nd->nd_rp)
3387 		panic("nfsrvstate checkseqid");
3388 	if (!(op->rc_flag & RC_INPROG))
3389 		panic("nfsrvstate not inprog");
3390 	if (stp->ls_op && stp->ls_op->rc_refcnt <= 0) {
3391 		printf("refcnt=%d\n", stp->ls_op->rc_refcnt);
3392 		panic("nfsrvstate op refcnt");
3393 	}
3394 	if ((stp->ls_seq + 1) == seqid) {
3395 		if (stp->ls_op)
3396 			nfsrvd_derefcache(stp->ls_op);
3397 		stp->ls_op = op;
3398 		nfsrvd_refcache(op);
3399 		stp->ls_seq = seqid;
3400 		return (0);
3401 	} else if (stp->ls_seq == seqid && stp->ls_op &&
3402 		op->rc_xid == stp->ls_op->rc_xid &&
3403 		op->rc_refcnt == 0 &&
3404 		op->rc_reqlen == stp->ls_op->rc_reqlen &&
3405 		op->rc_cksum == stp->ls_op->rc_cksum) {
3406 		if (stp->ls_op->rc_flag & RC_INPROG)
3407 			return (NFSERR_DONTREPLY);
3408 		nd->nd_rp = stp->ls_op;
3409 		nd->nd_rp->rc_flag |= RC_INPROG;
3410 		nfsrvd_delcache(op);
3411 		return (NFSERR_REPLYFROMCACHE);
3412 	}
3413 	return (NFSERR_BADSEQID);
3414 }
3415 
3416 /*
3417  * Get the client ip address for callbacks. If the strings can't be parsed,
3418  * just set lc_program to 0 to indicate no callbacks are possible.
3419  * (For cases where the address can't be parsed or is 0.0.0.0.0.0, set
3420  *  the address to the client's transport address. This won't be used
3421  *  for callbacks, but can be printed out by newnfsstats for info.)
3422  * Return error if the xdr can't be parsed, 0 otherwise.
3423  */
3424 APPLESTATIC int
3425 nfsrv_getclientipaddr(struct nfsrv_descript *nd, struct nfsclient *clp)
3426 {
3427 	u_int32_t *tl;
3428 	u_char *cp, *cp2;
3429 	int i, j;
3430 	struct sockaddr_in *rad, *sad;
3431 	u_char protocol[5], addr[24];
3432 	int error = 0, cantparse = 0;
3433 	union {
3434 		u_long ival;
3435 		u_char cval[4];
3436 	} ip;
3437 	union {
3438 		u_short sval;
3439 		u_char cval[2];
3440 	} port;
3441 
3442 	rad = NFSSOCKADDR(clp->lc_req.nr_nam, struct sockaddr_in *);
3443 	rad->sin_family = AF_INET;
3444 	rad->sin_len = sizeof (struct sockaddr_in);
3445 	rad->sin_addr.s_addr = 0;
3446 	rad->sin_port = 0;
3447 	clp->lc_req.nr_client = NULL;
3448 	clp->lc_req.nr_lock = 0;
3449 	NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
3450 	i = fxdr_unsigned(int, *tl);
3451 	if (i >= 3 && i <= 4) {
3452 		error = nfsrv_mtostr(nd, protocol, i);
3453 		if (error)
3454 			goto nfsmout;
3455 		if (!strcmp(protocol, "tcp")) {
3456 			clp->lc_flags |= LCL_TCPCALLBACK;
3457 			clp->lc_req.nr_sotype = SOCK_STREAM;
3458 			clp->lc_req.nr_soproto = IPPROTO_TCP;
3459 		} else if (!strcmp(protocol, "udp")) {
3460 			clp->lc_req.nr_sotype = SOCK_DGRAM;
3461 			clp->lc_req.nr_soproto = IPPROTO_UDP;
3462 		} else {
3463 			cantparse = 1;
3464 		}
3465 	} else {
3466 		cantparse = 1;
3467 		if (i > 0) {
3468 			error = nfsm_advance(nd, NFSM_RNDUP(i), -1);
3469 			if (error)
3470 				goto nfsmout;
3471 		}
3472 	}
3473 	NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
3474 	i = fxdr_unsigned(int, *tl);
3475 	if (i < 0) {
3476 		error = NFSERR_BADXDR;
3477 		goto nfsmout;
3478 	} else if (i == 0) {
3479 		cantparse = 1;
3480 	} else if (!cantparse && i <= 23 && i >= 11) {
3481 		error = nfsrv_mtostr(nd, addr, i);
3482 		if (error)
3483 			goto nfsmout;
3484 
3485 		/*
3486 		 * Parse out the address fields. We expect 6 decimal numbers
3487 		 * separated by '.'s.
3488 		 */
3489 		cp = addr;
3490 		i = 0;
3491 		while (*cp && i < 6) {
3492 			cp2 = cp;
3493 			while (*cp2 && *cp2 != '.')
3494 				cp2++;
3495 			if (*cp2)
3496 				*cp2++ = '\0';
3497 			else if (i != 5) {
3498 				cantparse = 1;
3499 				break;
3500 			}
3501 			j = nfsrv_getipnumber(cp);
3502 			if (j >= 0) {
3503 				if (i < 4)
3504 					ip.cval[3 - i] = j;
3505 				else
3506 					port.cval[5 - i] = j;
3507 			} else {
3508 				cantparse = 1;
3509 				break;
3510 			}
3511 			cp = cp2;
3512 			i++;
3513 		}
3514 		if (!cantparse) {
3515 			if (ip.ival != 0x0) {
3516 				rad->sin_addr.s_addr = htonl(ip.ival);
3517 				rad->sin_port = htons(port.sval);
3518 			} else {
3519 				cantparse = 1;
3520 			}
3521 		}
3522 	} else {
3523 		cantparse = 1;
3524 		if (i > 0) {
3525 			error = nfsm_advance(nd, NFSM_RNDUP(i), -1);
3526 			if (error)
3527 				goto nfsmout;
3528 		}
3529 	}
3530 	if (cantparse) {
3531 		sad = NFSSOCKADDR(nd->nd_nam, struct sockaddr_in *);
3532 		rad->sin_addr.s_addr = sad->sin_addr.s_addr;
3533 		rad->sin_port = 0x0;
3534 		clp->lc_program = 0;
3535 	}
3536 nfsmout:
3537 	return (error);
3538 }
3539 
3540 /*
3541  * Turn a string of up to three decimal digits into a number. Return -1 upon
3542  * error.
3543  */
3544 static int
3545 nfsrv_getipnumber(u_char *cp)
3546 {
3547 	int i = 0, j = 0;
3548 
3549 	while (*cp) {
3550 		if (j > 2 || *cp < '0' || *cp > '9')
3551 			return (-1);
3552 		i *= 10;
3553 		i += (*cp - '0');
3554 		cp++;
3555 		j++;
3556 	}
3557 	if (i < 256)
3558 		return (i);
3559 	return (-1);
3560 }
3561 
3562 /*
3563  * This function checks for restart conditions.
3564  */
3565 static int
3566 nfsrv_checkrestart(nfsquad_t clientid, u_int32_t flags,
3567     nfsv4stateid_t *stateidp, int specialid)
3568 {
3569 	int ret;
3570 
3571 	/*
3572 	 * First check for a server restart. Open, LockT, ReleaseLockOwner
3573 	 * and DelegPurge have a clientid, the rest a stateid.
3574 	 */
3575 	if (flags &
3576 	    (NFSLCK_OPEN | NFSLCK_TEST | NFSLCK_RELEASE | NFSLCK_DELEGPURGE)) {
3577 		if (clientid.lval[0] != nfsrvboottime)
3578 			return (NFSERR_STALECLIENTID);
3579 	} else if (stateidp->other[0] != nfsrvboottime &&
3580 		specialid == 0)
3581 		return (NFSERR_STALESTATEID);
3582 
3583 	/*
3584 	 * Read, Write, Setattr and LockT can return NFSERR_GRACE and do
3585 	 * not use a lock/open owner seqid#, so the check can be done now.
3586 	 * (The others will be checked, as required, later.)
3587 	 */
3588 	if (!(flags & (NFSLCK_CHECK | NFSLCK_TEST)))
3589 		return (0);
3590 
3591 	NFSLOCKSTATE();
3592 	ret = nfsrv_checkgrace(flags);
3593 	NFSUNLOCKSTATE();
3594 	return (ret);
3595 }
3596 
3597 /*
3598  * Check for grace.
3599  */
3600 static int
3601 nfsrv_checkgrace(u_int32_t flags)
3602 {
3603 
3604 	if (nfsrv_stablefirst.nsf_flags & NFSNSF_GRACEOVER) {
3605 		if (flags & NFSLCK_RECLAIM)
3606 			return (NFSERR_NOGRACE);
3607 	} else {
3608 		if (!(flags & NFSLCK_RECLAIM))
3609 			return (NFSERR_GRACE);
3610 
3611 		/*
3612 		 * If grace is almost over and we are still getting Reclaims,
3613 		 * extend grace a bit.
3614 		 */
3615 		if ((NFSD_MONOSEC + NFSRV_LEASEDELTA) >
3616 		    nfsrv_stablefirst.nsf_eograce)
3617 			nfsrv_stablefirst.nsf_eograce = NFSD_MONOSEC +
3618 				NFSRV_LEASEDELTA;
3619 	}
3620 	return (0);
3621 }
3622 
3623 /*
3624  * Do a server callback.
3625  */
3626 static int
3627 nfsrv_docallback(struct nfsclient *clp, int procnum,
3628     nfsv4stateid_t *stateidp, int trunc, fhandle_t *fhp,
3629     struct nfsvattr *nap, nfsattrbit_t *attrbitp, NFSPROC_T *p)
3630 {
3631 	mbuf_t m;
3632 	u_int32_t *tl;
3633 	struct nfsrv_descript nfsd, *nd = &nfsd;
3634 	struct ucred *cred;
3635 	int error = 0;
3636 	u_int32_t callback;
3637 
3638 	cred = newnfs_getcred();
3639 	NFSLOCKSTATE();	/* mostly for lc_cbref++ */
3640 	if (clp->lc_flags & LCL_NEEDSCONFIRM) {
3641 		NFSUNLOCKSTATE();
3642 		panic("docallb");
3643 	}
3644 	clp->lc_cbref++;
3645 
3646 	/*
3647 	 * Fill the callback program# and version into the request
3648 	 * structure for newnfs_connect() to use.
3649 	 */
3650 	clp->lc_req.nr_prog = clp->lc_program;
3651 	clp->lc_req.nr_vers = NFSV4_CBVERS;
3652 
3653 	/*
3654 	 * First, fill in some of the fields of nd and cr.
3655 	 */
3656 	nd->nd_flag = ND_NFSV4;
3657 	if (clp->lc_flags & LCL_GSS)
3658 		nd->nd_flag |= ND_KERBV;
3659 	nd->nd_repstat = 0;
3660 	cred->cr_uid = clp->lc_uid;
3661 	cred->cr_gid = clp->lc_gid;
3662 	callback = clp->lc_callback;
3663 	NFSUNLOCKSTATE();
3664 	cred->cr_ngroups = 1;
3665 
3666 	/*
3667 	 * Get the first mbuf for the request.
3668 	 */
3669 	MGET(m, M_WAIT, MT_DATA);
3670 	mbuf_setlen(m, 0);
3671 	nd->nd_mreq = nd->nd_mb = m;
3672 	nd->nd_bpos = NFSMTOD(m, caddr_t);
3673 
3674 	/*
3675 	 * and build the callback request.
3676 	 */
3677 	if (procnum == NFSV4OP_CBGETATTR) {
3678 		nd->nd_procnum = NFSV4PROC_CBCOMPOUND;
3679 		(void) nfsm_strtom(nd, "CB Getattr", 10);
3680 		NFSM_BUILD(tl, u_int32_t *, 4 * NFSX_UNSIGNED);
3681 		*tl++ = txdr_unsigned(NFSV4_MINORVERSION);
3682 		*tl++ = txdr_unsigned(callback);
3683 		*tl++ = txdr_unsigned(1);
3684 		*tl = txdr_unsigned(NFSV4OP_CBGETATTR);
3685 		(void) nfsm_fhtom(nd, (u_int8_t *)fhp, NFSX_MYFH, 0);
3686 		(void) nfsrv_putattrbit(nd, attrbitp);
3687 	} else if (procnum == NFSV4OP_CBRECALL) {
3688 		nd->nd_procnum = NFSV4PROC_CBCOMPOUND;
3689 		(void) nfsm_strtom(nd, "CB Recall", 9);
3690 		NFSM_BUILD(tl, u_int32_t *, 5 * NFSX_UNSIGNED + NFSX_STATEID);
3691 		*tl++ = txdr_unsigned(NFSV4_MINORVERSION);
3692 		*tl++ = txdr_unsigned(callback);
3693 		*tl++ = txdr_unsigned(1);
3694 		*tl++ = txdr_unsigned(NFSV4OP_CBRECALL);
3695 		*tl++ = txdr_unsigned(stateidp->seqid);
3696 		NFSBCOPY((caddr_t)stateidp->other, (caddr_t)tl,
3697 		    NFSX_STATEIDOTHER);
3698 		tl += (NFSX_STATEIDOTHER / NFSX_UNSIGNED);
3699 		if (trunc)
3700 			*tl = newnfs_true;
3701 		else
3702 			*tl = newnfs_false;
3703 		(void) nfsm_fhtom(nd, (u_int8_t *)fhp, NFSX_MYFH, 0);
3704 	} else {
3705 		nd->nd_procnum = NFSV4PROC_CBNULL;
3706 	}
3707 
3708 	/*
3709 	 * Call newnfs_connect(), as required, and then newnfs_request().
3710 	 */
3711 	(void) newnfs_sndlock(&clp->lc_req.nr_lock);
3712 	if (clp->lc_req.nr_client == NULL) {
3713 		if (nd->nd_procnum == NFSV4PROC_CBNULL)
3714 			error = newnfs_connect(NULL, &clp->lc_req, cred,
3715 			    NULL, 1);
3716 		else
3717 			error = newnfs_connect(NULL, &clp->lc_req, cred,
3718 			    NULL, 3);
3719 	}
3720 	newnfs_sndunlock(&clp->lc_req.nr_lock);
3721 	if (!error) {
3722 		error = newnfs_request(nd, NULL, clp, &clp->lc_req, NULL,
3723 		    NULL, cred, clp->lc_program, NFSV4_CBVERS, NULL, 1, NULL);
3724 	}
3725 	NFSFREECRED(cred);
3726 
3727 	/*
3728 	 * If error is set here, the Callback path isn't working
3729 	 * properly, so twiddle the appropriate LCL_ flags.
3730 	 * (nd_repstat != 0 indicates the Callback path is working,
3731 	 *  but the callback failed on the client.)
3732 	 */
3733 	if (error) {
3734 		/*
3735 		 * Mark the callback pathway down, which disabled issuing
3736 		 * of delegations and gets Renew to return NFSERR_CBPATHDOWN.
3737 		 */
3738 		NFSLOCKSTATE();
3739 		clp->lc_flags |= LCL_CBDOWN;
3740 		NFSUNLOCKSTATE();
3741 	} else {
3742 		/*
3743 		 * Callback worked. If the callback path was down, disable
3744 		 * callbacks, so no more delegations will be issued. (This
3745 		 * is done on the assumption that the callback pathway is
3746 		 * flakey.)
3747 		 */
3748 		NFSLOCKSTATE();
3749 		if (clp->lc_flags & LCL_CBDOWN)
3750 			clp->lc_flags &= ~(LCL_CBDOWN | LCL_CALLBACKSON);
3751 		NFSUNLOCKSTATE();
3752 		if (nd->nd_repstat)
3753 			error = nd->nd_repstat;
3754 		else if (procnum == NFSV4OP_CBGETATTR)
3755 			error = nfsv4_loadattr(nd, NULL, nap, NULL, NULL, 0,
3756 			    NULL, NULL, NULL, NULL, NULL, 0, NULL, NULL, NULL,
3757 			    p, NULL);
3758 		mbuf_freem(nd->nd_mrep);
3759 	}
3760 	NFSLOCKSTATE();
3761 	clp->lc_cbref--;
3762 	if ((clp->lc_flags & LCL_WAKEUPWANTED) && clp->lc_cbref == 0) {
3763 		clp->lc_flags &= ~LCL_WAKEUPWANTED;
3764 		NFSUNLOCKSTATE();
3765 		wakeup((caddr_t)clp);
3766 	} else {
3767 		NFSUNLOCKSTATE();
3768 	}
3769 	return (error);
3770 }
3771 
3772 /*
3773  * Return the next index# for a clientid. Mostly just increment and return
3774  * the next one, but... if the 32bit unsigned does actually wrap around,
3775  * it should be rebooted.
3776  * At an average rate of one new client per second, it will wrap around in
3777  * approximately 136 years. (I think the server will have been shut
3778  * down or rebooted before then.)
3779  */
3780 static u_int32_t
3781 nfsrv_nextclientindex(void)
3782 {
3783 	static u_int32_t client_index = 0;
3784 
3785 	client_index++;
3786 	if (client_index != 0)
3787 		return (client_index);
3788 
3789 	printf("%s: out of clientids\n", __func__);
3790 	return (client_index);
3791 }
3792 
3793 /*
3794  * Return the next index# for a stateid. Mostly just increment and return
3795  * the next one, but... if the 32bit unsigned does actually wrap around
3796  * (will a BSD server stay up that long?), find
3797  * new start and end values.
3798  */
3799 static u_int32_t
3800 nfsrv_nextstateindex(struct nfsclient *clp)
3801 {
3802 	struct nfsstate *stp;
3803 	int i;
3804 	u_int32_t canuse, min_index, max_index;
3805 
3806 	if (!(clp->lc_flags & LCL_INDEXNOTOK)) {
3807 		clp->lc_stateindex++;
3808 		if (clp->lc_stateindex != clp->lc_statemaxindex)
3809 			return (clp->lc_stateindex);
3810 	}
3811 
3812 	/*
3813 	 * Yuck, we've hit the end.
3814 	 * Look for a new min and max.
3815 	 */
3816 	min_index = 0;
3817 	max_index = 0xffffffff;
3818 	for (i = 0; i < NFSSTATEHASHSIZE; i++) {
3819 	    LIST_FOREACH(stp, &clp->lc_stateid[i], ls_hash) {
3820 		if (stp->ls_stateid.other[2] > 0x80000000) {
3821 		    if (stp->ls_stateid.other[2] < max_index)
3822 			max_index = stp->ls_stateid.other[2];
3823 		} else {
3824 		    if (stp->ls_stateid.other[2] > min_index)
3825 			min_index = stp->ls_stateid.other[2];
3826 		}
3827 	    }
3828 	}
3829 
3830 	/*
3831 	 * Yikes, highly unlikely, but I'll handle it anyhow.
3832 	 */
3833 	if (min_index == 0x80000000 && max_index == 0x80000001) {
3834 	    canuse = 0;
3835 	    /*
3836 	     * Loop around until we find an unused entry. Return that
3837 	     * and set LCL_INDEXNOTOK, so the search will continue next time.
3838 	     * (This is one of those rare cases where a goto is the
3839 	     *  cleanest way to code the loop.)
3840 	     */
3841 tryagain:
3842 	    for (i = 0; i < NFSSTATEHASHSIZE; i++) {
3843 		LIST_FOREACH(stp, &clp->lc_stateid[i], ls_hash) {
3844 		    if (stp->ls_stateid.other[2] == canuse) {
3845 			canuse++;
3846 			goto tryagain;
3847 		    }
3848 		}
3849 	    }
3850 	    clp->lc_flags |= LCL_INDEXNOTOK;
3851 	    return (canuse);
3852 	}
3853 
3854 	/*
3855 	 * Ok to start again from min + 1.
3856 	 */
3857 	clp->lc_stateindex = min_index + 1;
3858 	clp->lc_statemaxindex = max_index;
3859 	clp->lc_flags &= ~LCL_INDEXNOTOK;
3860 	return (clp->lc_stateindex);
3861 }
3862 
3863 /*
3864  * The following functions handle the stable storage file that deals with
3865  * the edge conditions described in RFC3530 Sec. 8.6.3.
3866  * The file is as follows:
3867  * - a single record at the beginning that has the lease time of the
3868  *   previous server instance (before the last reboot) and the nfsrvboottime
3869  *   values for the previous server boots.
3870  *   These previous boot times are used to ensure that the current
3871  *   nfsrvboottime does not, somehow, get set to a previous one.
3872  *   (This is important so that Stale ClientIDs and StateIDs can
3873  *    be recognized.)
3874  *   The number of previous nfsvrboottime values preceeds the list.
3875  * - followed by some number of appended records with:
3876  *   - client id string
3877  *   - flag that indicates it is a record revoking state via lease
3878  *     expiration or similar
3879  *     OR has successfully acquired state.
3880  * These structures vary in length, with the client string at the end, up
3881  * to NFSV4_OPAQUELIMIT in size.
3882  *
3883  * At the end of the grace period, the file is truncated, the first
3884  * record is rewritten with updated information and any acquired state
3885  * records for successful reclaims of state are written.
3886  *
3887  * Subsequent records are appended when the first state is issued to
3888  * a client and when state is revoked for a client.
3889  *
3890  * When reading the file in, state issued records that come later in
3891  * the file override older ones, since the append log is in cronological order.
3892  * If, for some reason, the file can't be read, the grace period is
3893  * immediately terminated and all reclaims get NFSERR_NOGRACE.
3894  */
3895 
3896 /*
3897  * Read in the stable storage file. Called by nfssvc() before the nfsd
3898  * processes start servicing requests.
3899  */
3900 APPLESTATIC void
3901 nfsrv_setupstable(NFSPROC_T *p)
3902 {
3903 	struct nfsrv_stablefirst *sf = &nfsrv_stablefirst;
3904 	struct nfsrv_stable *sp, *nsp;
3905 	struct nfst_rec *tsp;
3906 	int error, i, tryagain;
3907 	off_t off = 0;
3908 	int aresid, len;
3909 	struct timeval curtime;
3910 
3911 	/*
3912 	 * If NFSNSF_UPDATEDONE is set, this is a restart of the nfsds without
3913 	 * a reboot, so state has not been lost.
3914 	 */
3915 	if (sf->nsf_flags & NFSNSF_UPDATEDONE)
3916 		return;
3917 	/*
3918 	 * Set Grace over just until the file reads successfully.
3919 	 */
3920 	NFSGETTIME(&curtime);
3921 	nfsrvboottime = curtime.tv_sec;
3922 	LIST_INIT(&sf->nsf_head);
3923 	sf->nsf_flags = (NFSNSF_GRACEOVER | NFSNSF_NEEDLOCK);
3924 	sf->nsf_eograce = NFSD_MONOSEC + NFSRV_LEASEDELTA;
3925 	if (sf->nsf_fp == NULL)
3926 		return;
3927 	error = NFSD_RDWR(UIO_READ, NFSFPVNODE(sf->nsf_fp),
3928 	    (caddr_t)&sf->nsf_rec, sizeof (struct nfsf_rec), off, UIO_SYSSPACE,
3929 	    0, NFSFPCRED(sf->nsf_fp), &aresid, p);
3930 	if (error || aresid || sf->nsf_numboots == 0 ||
3931 		sf->nsf_numboots > NFSNSF_MAXNUMBOOTS)
3932 		return;
3933 
3934 	/*
3935 	 * Now, read in the boottimes.
3936 	 */
3937 	sf->nsf_bootvals = (time_t *)malloc((sf->nsf_numboots + 1) *
3938 		sizeof (time_t), M_TEMP, M_WAITOK);
3939 	off = sizeof (struct nfsf_rec);
3940 	error = NFSD_RDWR(UIO_READ, NFSFPVNODE(sf->nsf_fp),
3941 	    (caddr_t)sf->nsf_bootvals, sf->nsf_numboots * sizeof (time_t), off,
3942 	    UIO_SYSSPACE, 0, NFSFPCRED(sf->nsf_fp), &aresid, p);
3943 	if (error || aresid) {
3944 		free((caddr_t)sf->nsf_bootvals, M_TEMP);
3945 		sf->nsf_bootvals = NULL;
3946 		return;
3947 	}
3948 
3949 	/*
3950 	 * Make sure this nfsrvboottime is different from all recorded
3951 	 * previous ones.
3952 	 */
3953 	do {
3954 		tryagain = 0;
3955 		for (i = 0; i < sf->nsf_numboots; i++) {
3956 			if (nfsrvboottime == sf->nsf_bootvals[i]) {
3957 				nfsrvboottime++;
3958 				tryagain = 1;
3959 				break;
3960 			}
3961 		}
3962 	} while (tryagain);
3963 
3964 	sf->nsf_flags |= NFSNSF_OK;
3965 	off += (sf->nsf_numboots * sizeof (time_t));
3966 
3967 	/*
3968 	 * Read through the file, building a list of records for grace
3969 	 * checking.
3970 	 * Each record is between sizeof (struct nfst_rec) and
3971 	 * sizeof (struct nfst_rec) + NFSV4_OPAQUELIMIT - 1
3972 	 * and is actually sizeof (struct nfst_rec) + nst_len - 1.
3973 	 */
3974 	tsp = (struct nfst_rec *)malloc(sizeof (struct nfst_rec) +
3975 		NFSV4_OPAQUELIMIT - 1, M_TEMP, M_WAITOK);
3976 	do {
3977 	    error = NFSD_RDWR(UIO_READ, NFSFPVNODE(sf->nsf_fp),
3978 	        (caddr_t)tsp, sizeof (struct nfst_rec) + NFSV4_OPAQUELIMIT - 1,
3979 	        off, UIO_SYSSPACE, 0, NFSFPCRED(sf->nsf_fp), &aresid, p);
3980 	    len = (sizeof (struct nfst_rec) + NFSV4_OPAQUELIMIT - 1) - aresid;
3981 	    if (error || (len > 0 && (len < sizeof (struct nfst_rec) ||
3982 		len < (sizeof (struct nfst_rec) + tsp->len - 1)))) {
3983 		/*
3984 		 * Yuck, the file has been corrupted, so just return
3985 		 * after clearing out any restart state, so the grace period
3986 		 * is over.
3987 		 */
3988 		LIST_FOREACH_SAFE(sp, &sf->nsf_head, nst_list, nsp) {
3989 			LIST_REMOVE(sp, nst_list);
3990 			free((caddr_t)sp, M_TEMP);
3991 		}
3992 		free((caddr_t)tsp, M_TEMP);
3993 		sf->nsf_flags &= ~NFSNSF_OK;
3994 		free((caddr_t)sf->nsf_bootvals, M_TEMP);
3995 		sf->nsf_bootvals = NULL;
3996 		return;
3997 	    }
3998 	    if (len > 0) {
3999 		off += sizeof (struct nfst_rec) + tsp->len - 1;
4000 		/*
4001 		 * Search the list for a matching client.
4002 		 */
4003 		LIST_FOREACH(sp, &sf->nsf_head, nst_list) {
4004 			if (tsp->len == sp->nst_len &&
4005 			    !NFSBCMP(tsp->client, sp->nst_client, tsp->len))
4006 				break;
4007 		}
4008 		if (sp == LIST_END(&sf->nsf_head)) {
4009 			sp = (struct nfsrv_stable *)malloc(tsp->len +
4010 				sizeof (struct nfsrv_stable) - 1, M_TEMP,
4011 				M_WAITOK);
4012 			NFSBCOPY((caddr_t)tsp, (caddr_t)&sp->nst_rec,
4013 				sizeof (struct nfst_rec) + tsp->len - 1);
4014 			LIST_INSERT_HEAD(&sf->nsf_head, sp, nst_list);
4015 		} else {
4016 			if (tsp->flag == NFSNST_REVOKE)
4017 				sp->nst_flag |= NFSNST_REVOKE;
4018 			else
4019 				/*
4020 				 * A subsequent timestamp indicates the client
4021 				 * did a setclientid/confirm and any previous
4022 				 * revoke is no longer relevant.
4023 				 */
4024 				sp->nst_flag &= ~NFSNST_REVOKE;
4025 		}
4026 	    }
4027 	} while (len > 0);
4028 	free((caddr_t)tsp, M_TEMP);
4029 	sf->nsf_flags = NFSNSF_OK;
4030 	sf->nsf_eograce = NFSD_MONOSEC + sf->nsf_lease +
4031 		NFSRV_LEASEDELTA;
4032 }
4033 
4034 /*
4035  * Update the stable storage file, now that the grace period is over.
4036  */
4037 APPLESTATIC void
4038 nfsrv_updatestable(NFSPROC_T *p)
4039 {
4040 	struct nfsrv_stablefirst *sf = &nfsrv_stablefirst;
4041 	struct nfsrv_stable *sp, *nsp;
4042 	int i;
4043 	struct nfsvattr nva;
4044 	vnode_t vp;
4045 #if defined(__FreeBSD_version) && (__FreeBSD_version >= 500000)
4046 	mount_t mp = NULL;
4047 #endif
4048 	int error;
4049 
4050 	if (sf->nsf_fp == NULL || (sf->nsf_flags & NFSNSF_UPDATEDONE))
4051 		return;
4052 	sf->nsf_flags |= NFSNSF_UPDATEDONE;
4053 	/*
4054 	 * Ok, we need to rewrite the stable storage file.
4055 	 * - truncate to 0 length
4056 	 * - write the new first structure
4057 	 * - loop through the data structures, writing out any that
4058 	 *   have timestamps older than the old boot
4059 	 */
4060 	if (sf->nsf_bootvals) {
4061 		sf->nsf_numboots++;
4062 		for (i = sf->nsf_numboots - 2; i >= 0; i--)
4063 			sf->nsf_bootvals[i + 1] = sf->nsf_bootvals[i];
4064 	} else {
4065 		sf->nsf_numboots = 1;
4066 		sf->nsf_bootvals = (time_t *)malloc(sizeof (time_t),
4067 			M_TEMP, M_WAITOK);
4068 	}
4069 	sf->nsf_bootvals[0] = nfsrvboottime;
4070 	sf->nsf_lease = nfsrv_lease;
4071 	NFSVNO_ATTRINIT(&nva);
4072 	NFSVNO_SETATTRVAL(&nva, size, 0);
4073 	vp = NFSFPVNODE(sf->nsf_fp);
4074 	NFS_STARTWRITE(vp, &mp);
4075 	NFSVOPLOCK(vp, LK_EXCLUSIVE | LK_RETRY, p);
4076 	error = nfsvno_setattr(vp, &nva, NFSFPCRED(sf->nsf_fp), p, NULL);
4077 	NFS_ENDWRITE(mp);
4078 	NFSVOPUNLOCK(vp, 0, p);
4079 	if (!error)
4080 	    error = NFSD_RDWR(UIO_WRITE, vp,
4081 		(caddr_t)&sf->nsf_rec, sizeof (struct nfsf_rec), (off_t)0,
4082 		UIO_SYSSPACE, IO_SYNC, NFSFPCRED(sf->nsf_fp), NULL, p);
4083 	if (!error)
4084 	    error = NFSD_RDWR(UIO_WRITE, vp,
4085 		(caddr_t)sf->nsf_bootvals,
4086 		sf->nsf_numboots * sizeof (time_t),
4087 		(off_t)(sizeof (struct nfsf_rec)),
4088 		UIO_SYSSPACE, IO_SYNC, NFSFPCRED(sf->nsf_fp), NULL, p);
4089 	free((caddr_t)sf->nsf_bootvals, M_TEMP);
4090 	sf->nsf_bootvals = NULL;
4091 	if (error) {
4092 		sf->nsf_flags &= ~NFSNSF_OK;
4093 		printf("EEK! Can't write NfsV4 stable storage file\n");
4094 		return;
4095 	}
4096 	sf->nsf_flags |= NFSNSF_OK;
4097 
4098 	/*
4099 	 * Loop through the list and write out timestamp records for
4100 	 * any clients that successfully reclaimed state.
4101 	 */
4102 	LIST_FOREACH_SAFE(sp, &sf->nsf_head, nst_list, nsp) {
4103 		if (sp->nst_flag & NFSNST_GOTSTATE) {
4104 			nfsrv_writestable(sp->nst_client, sp->nst_len,
4105 				NFSNST_NEWSTATE, p);
4106 			sp->nst_clp->lc_flags |= LCL_STAMPEDSTABLE;
4107 		}
4108 		LIST_REMOVE(sp, nst_list);
4109 		free((caddr_t)sp, M_TEMP);
4110 	}
4111 }
4112 
4113 /*
4114  * Append a record to the stable storage file.
4115  */
4116 APPLESTATIC void
4117 nfsrv_writestable(u_char *client, int len, int flag, NFSPROC_T *p)
4118 {
4119 	struct nfsrv_stablefirst *sf = &nfsrv_stablefirst;
4120 	struct nfst_rec *sp;
4121 	int error;
4122 
4123 	if (!(sf->nsf_flags & NFSNSF_OK) || sf->nsf_fp == NULL)
4124 		return;
4125 	sp = (struct nfst_rec *)malloc(sizeof (struct nfst_rec) +
4126 		len - 1, M_TEMP, M_WAITOK);
4127 	sp->len = len;
4128 	NFSBCOPY(client, sp->client, len);
4129 	sp->flag = flag;
4130 	error = NFSD_RDWR(UIO_WRITE, NFSFPVNODE(sf->nsf_fp),
4131 	    (caddr_t)sp, sizeof (struct nfst_rec) + len - 1, (off_t)0,
4132 	    UIO_SYSSPACE, (IO_SYNC | IO_APPEND), NFSFPCRED(sf->nsf_fp), NULL, p);
4133 	free((caddr_t)sp, M_TEMP);
4134 	if (error) {
4135 		sf->nsf_flags &= ~NFSNSF_OK;
4136 		printf("EEK! Can't write NfsV4 stable storage file\n");
4137 	}
4138 }
4139 
4140 /*
4141  * This function is called during the grace period to mark a client
4142  * that successfully reclaimed state.
4143  */
4144 static void
4145 nfsrv_markstable(struct nfsclient *clp)
4146 {
4147 	struct nfsrv_stable *sp;
4148 
4149 	/*
4150 	 * First find the client structure.
4151 	 */
4152 	LIST_FOREACH(sp, &nfsrv_stablefirst.nsf_head, nst_list) {
4153 		if (sp->nst_len == clp->lc_idlen &&
4154 		    !NFSBCMP(sp->nst_client, clp->lc_id, sp->nst_len))
4155 			break;
4156 	}
4157 	if (sp == LIST_END(&nfsrv_stablefirst.nsf_head))
4158 		return;
4159 
4160 	/*
4161 	 * Now, just mark it and set the nfsclient back pointer.
4162 	 */
4163 	sp->nst_flag |= NFSNST_GOTSTATE;
4164 	sp->nst_clp = clp;
4165 }
4166 
4167 /*
4168  * This function is called for a reclaim, to see if it gets grace.
4169  * It returns 0 if a reclaim is allowed, 1 otherwise.
4170  */
4171 static int
4172 nfsrv_checkstable(struct nfsclient *clp)
4173 {
4174 	struct nfsrv_stable *sp;
4175 
4176 	/*
4177 	 * First, find the entry for the client.
4178 	 */
4179 	LIST_FOREACH(sp, &nfsrv_stablefirst.nsf_head, nst_list) {
4180 		if (sp->nst_len == clp->lc_idlen &&
4181 		    !NFSBCMP(sp->nst_client, clp->lc_id, sp->nst_len))
4182 			break;
4183 	}
4184 
4185 	/*
4186 	 * If not in the list, state was revoked or no state was issued
4187 	 * since the previous reboot, a reclaim is denied.
4188 	 */
4189 	if (sp == LIST_END(&nfsrv_stablefirst.nsf_head) ||
4190 	    (sp->nst_flag & NFSNST_REVOKE) ||
4191 	    !(nfsrv_stablefirst.nsf_flags & NFSNSF_OK))
4192 		return (1);
4193 	return (0);
4194 }
4195 
4196 /*
4197  * Test for and try to clear out a conflicting client. This is called by
4198  * nfsrv_lockctrl() and nfsrv_openctrl() when conflicts with other clients
4199  * a found.
4200  * The trick here is that it can't revoke a conflicting client with an
4201  * expired lease unless it holds the v4root lock, so...
4202  * If no v4root lock, get the lock and return 1 to indicate "try again".
4203  * Return 0 to indicate the conflict can't be revoked and 1 to indicate
4204  * the revocation worked and the conflicting client is "bye, bye", so it
4205  * can be tried again.
4206  * Unlocks State before a non-zero value is returned.
4207  */
4208 static int
4209 nfsrv_clientconflict(struct nfsclient *clp, int *haslockp, __unused vnode_t vp,
4210     NFSPROC_T *p)
4211 {
4212 	int gotlock;
4213 
4214 	/*
4215 	 * If lease hasn't expired, we can't fix it.
4216 	 */
4217 	if (clp->lc_expiry >= NFSD_MONOSEC ||
4218 	    !(nfsrv_stablefirst.nsf_flags & NFSNSF_UPDATEDONE))
4219 		return (0);
4220 	if (*haslockp == 0) {
4221 		NFSUNLOCKSTATE();
4222 		NFSVOPUNLOCK(vp, 0, p);
4223 		NFSLOCKV4ROOTMUTEX();
4224 		nfsv4_relref(&nfsv4rootfs_lock);
4225 		do {
4226 			gotlock = nfsv4_lock(&nfsv4rootfs_lock, 1, NULL,
4227 			    NFSV4ROOTLOCKMUTEXPTR);
4228 		} while (!gotlock);
4229 		NFSUNLOCKV4ROOTMUTEX();
4230 		*haslockp = 1;
4231 		NFSVOPLOCK(vp, LK_EXCLUSIVE | LK_RETRY, p);
4232 		return (1);
4233 	}
4234 	NFSUNLOCKSTATE();
4235 
4236 	/*
4237 	 * Ok, we can expire the conflicting client.
4238 	 */
4239 	nfsrv_writestable(clp->lc_id, clp->lc_idlen, NFSNST_REVOKE, p);
4240 	nfsrv_cleanclient(clp, p);
4241 	nfsrv_freedeleglist(&clp->lc_deleg);
4242 	nfsrv_freedeleglist(&clp->lc_olddeleg);
4243 	LIST_REMOVE(clp, lc_hash);
4244 	nfsrv_zapclient(clp, p);
4245 	return (1);
4246 }
4247 
4248 
4249 /*
4250  * Resolve a delegation conflict.
4251  * Returns 0 to indicate the conflict was resolved without sleeping.
4252  * Return -1 to indicate that the caller should check for conflicts again.
4253  * Return > 0 for an error that should be returned, normally NFSERR_DELAY.
4254  *
4255  * Also, manipulate the nfsv4root_lock, as required. It isn't changed
4256  * for a return of 0, since there was no sleep and it could be required
4257  * later. It is released for a return of NFSERR_DELAY, since the caller
4258  * will return that error. It is released when a sleep was done waiting
4259  * for the delegation to be returned or expire (so that other nfsds can
4260  * handle ops). Then, it must be acquired for the write to stable storage.
4261  * (This function is somewhat similar to nfsrv_clientconflict(), but
4262  *  the semantics differ in a couple of subtle ways. The return of 0
4263  *  indicates the conflict was resolved without sleeping here, not
4264  *  that the conflict can't be resolved and the handling of nfsv4root_lock
4265  *  differs, as noted above.)
4266  * Unlocks State before returning a non-zero value.
4267  */
4268 static int
4269 nfsrv_delegconflict(struct nfsstate *stp, int *haslockp, NFSPROC_T *p,
4270     __unused vnode_t vp)
4271 {
4272 	struct nfsclient *clp = stp->ls_clp;
4273 	int gotlock, error, retrycnt, zapped_clp;
4274 	nfsv4stateid_t tstateid;
4275 	fhandle_t tfh;
4276 
4277 	/*
4278 	 * If the conflict is with an old delegation...
4279 	 */
4280 	if (stp->ls_flags & NFSLCK_OLDDELEG) {
4281 		/*
4282 		 * You can delete it, if it has expired.
4283 		 */
4284 		if (clp->lc_delegtime < NFSD_MONOSEC) {
4285 			nfsrv_freedeleg(stp);
4286 			NFSUNLOCKSTATE();
4287 			return (-1);
4288 		}
4289 		NFSUNLOCKSTATE();
4290 		/*
4291 		 * During this delay, the old delegation could expire or it
4292 		 * could be recovered by the client via an Open with
4293 		 * CLAIM_DELEGATE_PREV.
4294 		 * Release the nfsv4root_lock, if held.
4295 		 */
4296 		if (*haslockp) {
4297 			*haslockp = 0;
4298 			NFSLOCKV4ROOTMUTEX();
4299 			nfsv4_unlock(&nfsv4rootfs_lock, 1);
4300 			NFSUNLOCKV4ROOTMUTEX();
4301 		}
4302 		return (NFSERR_DELAY);
4303 	}
4304 
4305 	/*
4306 	 * It's a current delegation, so:
4307 	 * - check to see if the delegation has expired
4308 	 *   - if so, get the v4root lock and then expire it
4309 	 */
4310 	if (!(stp->ls_flags & NFSLCK_DELEGRECALL)) {
4311 		/*
4312 		 * - do a recall callback, since not yet done
4313 		 * For now, never allow truncate to be set. To use
4314 		 * truncate safely, it must be guaranteed that the
4315 		 * Remove, Rename or Setattr with size of 0 will
4316 		 * succeed and that would require major changes to
4317 		 * the VFS/Vnode OPs.
4318 		 * Set the expiry time large enough so that it won't expire
4319 		 * until after the callback, then set it correctly, once
4320 		 * the callback is done. (The delegation will now time
4321 		 * out whether or not the Recall worked ok. The timeout
4322 		 * will be extended when ops are done on the delegation
4323 		 * stateid, up to the timelimit.)
4324 		 */
4325 		stp->ls_delegtime = NFSD_MONOSEC + (2 * nfsrv_lease) +
4326 		    NFSRV_LEASEDELTA;
4327 		stp->ls_delegtimelimit = NFSD_MONOSEC + (6 * nfsrv_lease) +
4328 		    NFSRV_LEASEDELTA;
4329 		stp->ls_flags |= NFSLCK_DELEGRECALL;
4330 
4331 		/*
4332 		 * Loop NFSRV_CBRETRYCNT times while the CBRecall replies
4333 		 * NFSERR_BADSTATEID or NFSERR_BADHANDLE. This is done
4334 		 * in order to try and avoid a race that could happen
4335 		 * when a CBRecall request passed the Open reply with
4336 		 * the delegation in it when transitting the network.
4337 		 * Since nfsrv_docallback will sleep, don't use stp after
4338 		 * the call.
4339 		 */
4340 		NFSBCOPY((caddr_t)&stp->ls_stateid, (caddr_t)&tstateid,
4341 		    sizeof (tstateid));
4342 		NFSBCOPY((caddr_t)&stp->ls_lfp->lf_fh, (caddr_t)&tfh,
4343 		    sizeof (tfh));
4344 		NFSUNLOCKSTATE();
4345 		if (*haslockp) {
4346 			*haslockp = 0;
4347 			NFSLOCKV4ROOTMUTEX();
4348 			nfsv4_unlock(&nfsv4rootfs_lock, 1);
4349 			NFSUNLOCKV4ROOTMUTEX();
4350 		}
4351 		retrycnt = 0;
4352 		do {
4353 		    error = nfsrv_docallback(clp, NFSV4OP_CBRECALL,
4354 			&tstateid, 0, &tfh, NULL, NULL, p);
4355 		    retrycnt++;
4356 		} while ((error == NFSERR_BADSTATEID ||
4357 		    error == NFSERR_BADHANDLE) && retrycnt < NFSV4_CBRETRYCNT);
4358 		return (NFSERR_DELAY);
4359 	}
4360 
4361 	if (clp->lc_expiry >= NFSD_MONOSEC &&
4362 	    stp->ls_delegtime >= NFSD_MONOSEC) {
4363 		NFSUNLOCKSTATE();
4364 		/*
4365 		 * A recall has been done, but it has not yet expired.
4366 		 * So, RETURN_DELAY.
4367 		 */
4368 		if (*haslockp) {
4369 			*haslockp = 0;
4370 			NFSLOCKV4ROOTMUTEX();
4371 			nfsv4_unlock(&nfsv4rootfs_lock, 1);
4372 			NFSUNLOCKV4ROOTMUTEX();
4373 		}
4374 		return (NFSERR_DELAY);
4375 	}
4376 
4377 	/*
4378 	 * If we don't yet have the lock, just get it and then return,
4379 	 * since we need that before deleting expired state, such as
4380 	 * this delegation.
4381 	 * When getting the lock, unlock the vnode, so other nfsds that
4382 	 * are in progress, won't get stuck waiting for the vnode lock.
4383 	 */
4384 	if (*haslockp == 0) {
4385 		NFSUNLOCKSTATE();
4386 		NFSVOPUNLOCK(vp, 0, p);
4387 		NFSLOCKV4ROOTMUTEX();
4388 		nfsv4_relref(&nfsv4rootfs_lock);
4389 		do {
4390 			gotlock = nfsv4_lock(&nfsv4rootfs_lock, 1, NULL,
4391 			    NFSV4ROOTLOCKMUTEXPTR);
4392 		} while (!gotlock);
4393 		NFSUNLOCKV4ROOTMUTEX();
4394 		*haslockp = 1;
4395 		NFSVOPLOCK(vp, LK_EXCLUSIVE | LK_RETRY, p);
4396 		return (-1);
4397 	}
4398 
4399 	NFSUNLOCKSTATE();
4400 	/*
4401 	 * Ok, we can delete the expired delegation.
4402 	 * First, write the Revoke record to stable storage and then
4403 	 * clear out the conflict.
4404 	 * Since all other nfsd threads are now blocked, we can safely
4405 	 * sleep without the state changing.
4406 	 */
4407 	nfsrv_writestable(clp->lc_id, clp->lc_idlen, NFSNST_REVOKE, p);
4408 	if (clp->lc_expiry < NFSD_MONOSEC) {
4409 		nfsrv_cleanclient(clp, p);
4410 		nfsrv_freedeleglist(&clp->lc_deleg);
4411 		nfsrv_freedeleglist(&clp->lc_olddeleg);
4412 		LIST_REMOVE(clp, lc_hash);
4413 		zapped_clp = 1;
4414 	} else {
4415 		nfsrv_freedeleg(stp);
4416 		zapped_clp = 0;
4417 	}
4418 	if (zapped_clp)
4419 		nfsrv_zapclient(clp, p);
4420 	return (-1);
4421 }
4422 
4423 /*
4424  * Check for a remove allowed, if remove is set to 1 and get rid of
4425  * delegations.
4426  */
4427 APPLESTATIC int
4428 nfsrv_checkremove(vnode_t vp, int remove, NFSPROC_T *p)
4429 {
4430 	struct nfsstate *stp;
4431 	struct nfslockfile *lfp;
4432 	int error, haslock = 0;
4433 	fhandle_t nfh;
4434 
4435 	/*
4436 	 * First, get the lock file structure.
4437 	 * (A return of -1 means no associated state, so remove ok.)
4438 	 */
4439 	error = nfsrv_getlockfh(vp, NFSLCK_CHECK, NULL, &nfh, p);
4440 tryagain:
4441 	NFSLOCKSTATE();
4442 	if (!error)
4443 		error = nfsrv_getlockfile(NFSLCK_CHECK, NULL, &lfp, &nfh, 0);
4444 	if (error) {
4445 		NFSUNLOCKSTATE();
4446 		if (haslock) {
4447 			NFSLOCKV4ROOTMUTEX();
4448 			nfsv4_unlock(&nfsv4rootfs_lock, 1);
4449 			NFSUNLOCKV4ROOTMUTEX();
4450 		}
4451 		if (error == -1)
4452 			return (0);
4453 		return (error);
4454 	}
4455 
4456 	/*
4457 	 * Now, we must Recall any delegations.
4458 	 */
4459 	error = nfsrv_cleandeleg(vp, lfp, NULL, &haslock, p);
4460 	if (error) {
4461 		/*
4462 		 * nfsrv_cleandeleg() unlocks state for non-zero
4463 		 * return.
4464 		 */
4465 		if (error == -1)
4466 			goto tryagain;
4467 		if (haslock) {
4468 			NFSLOCKV4ROOTMUTEX();
4469 			nfsv4_unlock(&nfsv4rootfs_lock, 1);
4470 			NFSUNLOCKV4ROOTMUTEX();
4471 		}
4472 		return (error);
4473 	}
4474 
4475 	/*
4476 	 * Now, look for a conflicting open share.
4477 	 */
4478 	if (remove) {
4479 		LIST_FOREACH(stp, &lfp->lf_open, ls_file) {
4480 			if (stp->ls_flags & NFSLCK_WRITEDENY) {
4481 				error = NFSERR_FILEOPEN;
4482 				break;
4483 			}
4484 		}
4485 	}
4486 
4487 	NFSUNLOCKSTATE();
4488 	if (haslock) {
4489 		NFSLOCKV4ROOTMUTEX();
4490 		nfsv4_unlock(&nfsv4rootfs_lock, 1);
4491 		NFSUNLOCKV4ROOTMUTEX();
4492 	}
4493 	return (error);
4494 }
4495 
4496 /*
4497  * Clear out all delegations for the file referred to by lfp.
4498  * May return NFSERR_DELAY, if there will be a delay waiting for
4499  * delegations to expire.
4500  * Returns -1 to indicate it slept while recalling a delegation.
4501  * This function has the side effect of deleting the nfslockfile structure,
4502  * if it no longer has associated state and didn't have to sleep.
4503  * Unlocks State before a non-zero value is returned.
4504  */
4505 static int
4506 nfsrv_cleandeleg(vnode_t vp, struct nfslockfile *lfp,
4507     struct nfsclient *clp, int *haslockp, NFSPROC_T *p)
4508 {
4509 	struct nfsstate *stp, *nstp;
4510 	int ret;
4511 
4512 	stp = LIST_FIRST(&lfp->lf_deleg);
4513 	while (stp != LIST_END(&lfp->lf_deleg)) {
4514 		nstp = LIST_NEXT(stp, ls_file);
4515 		if (stp->ls_clp != clp) {
4516 			ret = nfsrv_delegconflict(stp, haslockp, p, vp);
4517 			if (ret) {
4518 				/*
4519 				 * nfsrv_delegconflict() unlocks state
4520 				 * when it returns non-zero.
4521 				 */
4522 				return (ret);
4523 			}
4524 		}
4525 		stp = nstp;
4526 	}
4527 	return (0);
4528 }
4529 
4530 /*
4531  * There are certain operations that, when being done outside of NFSv4,
4532  * require that any NFSv4 delegation for the file be recalled.
4533  * This function is to be called for those cases:
4534  * VOP_RENAME() - When a delegation is being recalled for any reason,
4535  *	the client may have to do Opens against the server, using the file's
4536  *	final component name. If the file has been renamed on the server,
4537  *	that component name will be incorrect and the Open will fail.
4538  * VOP_REMOVE() - Theoretically, a client could Open a file after it has
4539  *	been removed on the server, if there is a delegation issued to
4540  *	that client for the file. I say "theoretically" since clients
4541  *	normally do an Access Op before the Open and that Access Op will
4542  *	fail with ESTALE. Note that NFSv2 and 3 don't even do Opens, so
4543  *	they will detect the file's removal in the same manner. (There is
4544  *	one case where RFC3530 allows a client to do an Open without first
4545  *	doing an Access Op, which is passage of a check against the ACE
4546  *	returned with a Write delegation, but current practice is to ignore
4547  *	the ACE and always do an Access Op.)
4548  *	Since the functions can only be called with an unlocked vnode, this
4549  *	can't be done at this time.
4550  * VOP_ADVLOCK() - When a client holds a delegation, it can issue byte range
4551  *	locks locally in the client, which are not visible to the server. To
4552  *	deal with this, issuing of delegations for a vnode must be disabled
4553  *	and all delegations for the vnode recalled. This is done via the
4554  *	second function, using the VV_DISABLEDELEG vflag on the vnode.
4555  */
4556 APPLESTATIC void
4557 nfsd_recalldelegation(vnode_t vp, NFSPROC_T *p)
4558 {
4559 	struct timespec mytime;
4560 	int32_t starttime;
4561 	int error;
4562 
4563 	KASSERT(!VOP_ISLOCKED(vp), ("vp %p is locked", vp));
4564 
4565 	/*
4566 	 * First, check to see if the server is currently running and it has
4567 	 * been called for a regular file when issuing delegations.
4568 	 */
4569 	if (newnfs_numnfsd == 0 || vp->v_type != VREG ||
4570 	    nfsrv_issuedelegs == 0)
4571 		return;
4572 
4573 	/*
4574 	 * First, get a reference on the nfsv4rootfs_lock so that an
4575 	 * exclusive lock cannot be acquired by another thread.
4576 	 */
4577 	NFSLOCKV4ROOTMUTEX();
4578 	nfsv4_getref(&nfsv4rootfs_lock, NULL, NFSV4ROOTLOCKMUTEXPTR);
4579 	NFSUNLOCKV4ROOTMUTEX();
4580 
4581 	/*
4582 	 * Now, call nfsrv_checkremove() in a loop while it returns
4583 	 * NFSERR_DELAY. Return upon any other error or when timed out.
4584 	 */
4585 	NFSGETNANOTIME(&mytime);
4586 	starttime = (u_int32_t)mytime.tv_sec;
4587 	do {
4588 		error = nfsrv_checkremove(vp, 0, p);
4589 		if (error == NFSERR_DELAY) {
4590 			NFSGETNANOTIME(&mytime);
4591 			if (((u_int32_t)mytime.tv_sec - starttime) >
4592 			    NFS_REMOVETIMEO &&
4593 			    ((u_int32_t)mytime.tv_sec - starttime) <
4594 			    100000)
4595 				break;
4596 			/* Sleep for a short period of time */
4597 			(void) nfs_catnap(PZERO, 0, "nfsremove");
4598 		}
4599 	} while (error == NFSERR_DELAY);
4600 	NFSLOCKV4ROOTMUTEX();
4601 	nfsv4_relref(&nfsv4rootfs_lock);
4602 	NFSUNLOCKV4ROOTMUTEX();
4603 }
4604 
4605 APPLESTATIC void
4606 nfsd_disabledelegation(vnode_t vp, NFSPROC_T *p)
4607 {
4608 
4609 #ifdef VV_DISABLEDELEG
4610 	/*
4611 	 * First, flag issuance of delegations disabled.
4612 	 */
4613 	atomic_set_long(&vp->v_vflag, VV_DISABLEDELEG);
4614 #endif
4615 
4616 	/*
4617 	 * Then call nfsd_recalldelegation() to get rid of all extant
4618 	 * delegations.
4619 	 */
4620 	nfsd_recalldelegation(vp, p);
4621 }
4622 
4623 /*
4624  * Check for conflicting locks, etc. and then get rid of delegations.
4625  * (At one point I thought that I should get rid of delegations for any
4626  *  Setattr, since it could potentially disallow the I/O op (read or write)
4627  *  allowed by the delegation. However, Setattr Ops that aren't changing
4628  *  the size get a stateid of all 0s, so you can't tell if it is a delegation
4629  *  for the same client or a different one, so I decided to only get rid
4630  *  of delegations for other clients when the size is being changed.)
4631  * In general, a Setattr can disable NFS I/O Ops that are outstanding, such
4632  * as Write backs, even if there is no delegation, so it really isn't any
4633  * different?)
4634  */
4635 APPLESTATIC int
4636 nfsrv_checksetattr(vnode_t vp, struct nfsrv_descript *nd,
4637     nfsv4stateid_t *stateidp, struct nfsvattr *nvap, nfsattrbit_t *attrbitp,
4638     struct nfsexstuff *exp, NFSPROC_T *p)
4639 {
4640 	struct nfsstate st, *stp = &st;
4641 	struct nfslock lo, *lop = &lo;
4642 	int error = 0;
4643 	nfsquad_t clientid;
4644 
4645 	if (NFSISSET_ATTRBIT(attrbitp, NFSATTRBIT_SIZE)) {
4646 		stp->ls_flags = (NFSLCK_CHECK | NFSLCK_WRITEACCESS);
4647 		lop->lo_first = nvap->na_size;
4648 	} else {
4649 		stp->ls_flags = 0;
4650 		lop->lo_first = 0;
4651 	}
4652 	if (NFSISSET_ATTRBIT(attrbitp, NFSATTRBIT_OWNER) ||
4653 	    NFSISSET_ATTRBIT(attrbitp, NFSATTRBIT_OWNERGROUP) ||
4654 	    NFSISSET_ATTRBIT(attrbitp, NFSATTRBIT_MODE) ||
4655 	    NFSISSET_ATTRBIT(attrbitp, NFSATTRBIT_ACL))
4656 		stp->ls_flags |= NFSLCK_SETATTR;
4657 	if (stp->ls_flags == 0)
4658 		return (0);
4659 	lop->lo_end = NFS64BITSSET;
4660 	lop->lo_flags = NFSLCK_WRITE;
4661 	stp->ls_ownerlen = 0;
4662 	stp->ls_op = NULL;
4663 	stp->ls_uid = nd->nd_cred->cr_uid;
4664 	stp->ls_stateid.seqid = stateidp->seqid;
4665 	clientid.lval[0] = stp->ls_stateid.other[0] = stateidp->other[0];
4666 	clientid.lval[1] = stp->ls_stateid.other[1] = stateidp->other[1];
4667 	stp->ls_stateid.other[2] = stateidp->other[2];
4668 	error = nfsrv_lockctrl(vp, &stp, &lop, NULL, clientid,
4669 	    stateidp, exp, nd, p);
4670 	return (error);
4671 }
4672 
4673 /*
4674  * Check for a write delegation and do a CBGETATTR if there is one, updating
4675  * the attributes, as required.
4676  * Should I return an error if I can't get the attributes? (For now, I'll
4677  * just return ok.
4678  */
4679 APPLESTATIC int
4680 nfsrv_checkgetattr(struct nfsrv_descript *nd, vnode_t vp,
4681     struct nfsvattr *nvap, nfsattrbit_t *attrbitp, struct ucred *cred,
4682     NFSPROC_T *p)
4683 {
4684 	struct nfsstate *stp;
4685 	struct nfslockfile *lfp;
4686 	struct nfsclient *clp;
4687 	struct nfsvattr nva;
4688 	fhandle_t nfh;
4689 	int error;
4690 	nfsattrbit_t cbbits;
4691 	u_quad_t delegfilerev;
4692 
4693 	NFSCBGETATTR_ATTRBIT(attrbitp, &cbbits);
4694 	if (!NFSNONZERO_ATTRBIT(&cbbits))
4695 		return (0);
4696 
4697 	/*
4698 	 * Get the lock file structure.
4699 	 * (A return of -1 means no associated state, so return ok.)
4700 	 */
4701 	error = nfsrv_getlockfh(vp, NFSLCK_CHECK, NULL, &nfh, p);
4702 	NFSLOCKSTATE();
4703 	if (!error)
4704 		error = nfsrv_getlockfile(NFSLCK_CHECK, NULL, &lfp, &nfh, 0);
4705 	if (error) {
4706 		NFSUNLOCKSTATE();
4707 		if (error == -1)
4708 			return (0);
4709 		return (error);
4710 	}
4711 
4712 	/*
4713 	 * Now, look for a write delegation.
4714 	 */
4715 	LIST_FOREACH(stp, &lfp->lf_deleg, ls_file) {
4716 		if (stp->ls_flags & NFSLCK_DELEGWRITE)
4717 			break;
4718 	}
4719 	if (stp == LIST_END(&lfp->lf_deleg)) {
4720 		NFSUNLOCKSTATE();
4721 		return (0);
4722 	}
4723 	clp = stp->ls_clp;
4724 	delegfilerev = stp->ls_filerev;
4725 
4726 	/*
4727 	 * If the Write delegation was issued as a part of this Compound RPC
4728 	 * or if we have an Implied Clientid (used in a previous Op in this
4729 	 * compound) and it is the client the delegation was issued to,
4730 	 * just return ok.
4731 	 * I also assume that it is from the same client iff the network
4732 	 * host IP address is the same as the callback address. (Not
4733 	 * exactly correct by the RFC, but avoids a lot of Getattr
4734 	 * callbacks.)
4735 	 */
4736 	if (nd->nd_compref == stp->ls_compref ||
4737 	    ((nd->nd_flag & ND_IMPLIEDCLID) &&
4738 	     clp->lc_clientid.qval == nd->nd_clientid.qval) ||
4739 	     nfsaddr2_match(clp->lc_req.nr_nam, nd->nd_nam)) {
4740 		NFSUNLOCKSTATE();
4741 		return (0);
4742 	}
4743 
4744 	/*
4745 	 * We are now done with the delegation state structure,
4746 	 * so the statelock can be released and we can now tsleep().
4747 	 */
4748 
4749 	/*
4750 	 * Now, we must do the CB Getattr callback, to see if Change or Size
4751 	 * has changed.
4752 	 */
4753 	if (clp->lc_expiry >= NFSD_MONOSEC) {
4754 		NFSUNLOCKSTATE();
4755 		NFSVNO_ATTRINIT(&nva);
4756 		nva.na_filerev = NFS64BITSSET;
4757 		error = nfsrv_docallback(clp, NFSV4OP_CBGETATTR, NULL,
4758 		    0, &nfh, &nva, &cbbits, p);
4759 		if (!error) {
4760 			if ((nva.na_filerev != NFS64BITSSET &&
4761 			    nva.na_filerev > delegfilerev) ||
4762 			    (NFSVNO_ISSETSIZE(&nva) &&
4763 			     nva.na_size != nvap->na_size)) {
4764 				nfsvno_updfilerev(vp, nvap, cred, p);
4765 				if (NFSVNO_ISSETSIZE(&nva))
4766 					nvap->na_size = nva.na_size;
4767 			}
4768 		}
4769 	} else {
4770 		NFSUNLOCKSTATE();
4771 	}
4772 	return (0);
4773 }
4774 
4775 /*
4776  * This function looks for openowners that haven't had any opens for
4777  * a while and throws them away. Called by an nfsd when NFSNSF_NOOPENS
4778  * is set.
4779  */
4780 APPLESTATIC void
4781 nfsrv_throwawayopens(NFSPROC_T *p)
4782 {
4783 	struct nfsclient *clp, *nclp;
4784 	struct nfsstate *stp, *nstp;
4785 	int i;
4786 
4787 	NFSLOCKSTATE();
4788 	nfsrv_stablefirst.nsf_flags &= ~NFSNSF_NOOPENS;
4789 	/*
4790 	 * For each client...
4791 	 */
4792 	for (i = 0; i < NFSCLIENTHASHSIZE; i++) {
4793 	    LIST_FOREACH_SAFE(clp, &nfsclienthash[i], lc_hash, nclp) {
4794 		LIST_FOREACH_SAFE(stp, &clp->lc_open, ls_list, nstp) {
4795 			if (LIST_EMPTY(&stp->ls_open) &&
4796 			    (stp->ls_noopens > NFSNOOPEN ||
4797 			     (nfsrv_openpluslock * 2) >
4798 			     NFSRV_V4STATELIMIT))
4799 				nfsrv_freeopenowner(stp, 0, p);
4800 		}
4801 	    }
4802 	}
4803 	NFSUNLOCKSTATE();
4804 }
4805 
4806 /*
4807  * This function checks to see if the credentials are the same.
4808  * Returns 1 for not same, 0 otherwise.
4809  */
4810 static int
4811 nfsrv_notsamecredname(struct nfsrv_descript *nd, struct nfsclient *clp)
4812 {
4813 
4814 	if (nd->nd_flag & ND_GSS) {
4815 		if (!(clp->lc_flags & LCL_GSS))
4816 			return (1);
4817 		if (clp->lc_flags & LCL_NAME) {
4818 			if (nd->nd_princlen != clp->lc_namelen ||
4819 			    NFSBCMP(nd->nd_principal, clp->lc_name,
4820 				clp->lc_namelen))
4821 				return (1);
4822 			else
4823 				return (0);
4824 		}
4825 		if (nd->nd_cred->cr_uid == clp->lc_uid)
4826 			return (0);
4827 		else
4828 			return (1);
4829 	} else if (clp->lc_flags & LCL_GSS)
4830 		return (1);
4831 	/*
4832 	 * For AUTH_SYS, allow the same uid or root. (This is underspecified
4833 	 * in RFC3530, which talks about principals, but doesn't say anything
4834 	 * about uids for AUTH_SYS.)
4835 	 */
4836 	if (nd->nd_cred->cr_uid == clp->lc_uid || nd->nd_cred->cr_uid == 0)
4837 		return (0);
4838 	else
4839 		return (1);
4840 }
4841 
4842 /*
4843  * Calculate the lease expiry time.
4844  */
4845 static time_t
4846 nfsrv_leaseexpiry(void)
4847 {
4848 	struct timeval curtime;
4849 
4850 	NFSGETTIME(&curtime);
4851 	if (nfsrv_stablefirst.nsf_eograce > NFSD_MONOSEC)
4852 		return (NFSD_MONOSEC + 2 * (nfsrv_lease + NFSRV_LEASEDELTA));
4853 	return (NFSD_MONOSEC + nfsrv_lease + NFSRV_LEASEDELTA);
4854 }
4855 
4856 /*
4857  * Delay the delegation timeout as far as ls_delegtimelimit, as required.
4858  */
4859 static void
4860 nfsrv_delaydelegtimeout(struct nfsstate *stp)
4861 {
4862 
4863 	if ((stp->ls_flags & NFSLCK_DELEGRECALL) == 0)
4864 		return;
4865 
4866 	if ((stp->ls_delegtime + 15) > NFSD_MONOSEC &&
4867 	    stp->ls_delegtime < stp->ls_delegtimelimit) {
4868 		stp->ls_delegtime += nfsrv_lease;
4869 		if (stp->ls_delegtime > stp->ls_delegtimelimit)
4870 			stp->ls_delegtime = stp->ls_delegtimelimit;
4871 	}
4872 }
4873 
4874 /*
4875  * This function checks to see if there is any other state associated
4876  * with the openowner for this Open.
4877  * It returns 1 if there is no other state, 0 otherwise.
4878  */
4879 static int
4880 nfsrv_nootherstate(struct nfsstate *stp)
4881 {
4882 	struct nfsstate *tstp;
4883 
4884 	LIST_FOREACH(tstp, &stp->ls_openowner->ls_open, ls_list) {
4885 		if (tstp != stp || !LIST_EMPTY(&tstp->ls_lock))
4886 			return (0);
4887 	}
4888 	return (1);
4889 }
4890 
4891 /*
4892  * Create a list of lock deltas (changes to local byte range locking
4893  * that can be rolled back using the list) and apply the changes via
4894  * nfsvno_advlock(). Optionally, lock the list. It is expected that either
4895  * the rollback or update function will be called after this.
4896  * It returns an error (and rolls back, as required), if any nfsvno_advlock()
4897  * call fails. If it returns an error, it will unlock the list.
4898  */
4899 static int
4900 nfsrv_locallock(vnode_t vp, struct nfslockfile *lfp, int flags,
4901     uint64_t first, uint64_t end, struct nfslockconflict *cfp, NFSPROC_T *p)
4902 {
4903 	struct nfslock *lop, *nlop;
4904 	int error = 0;
4905 
4906 	/* Loop through the list of locks. */
4907 	lop = LIST_FIRST(&lfp->lf_locallock);
4908 	while (first < end && lop != NULL) {
4909 		nlop = LIST_NEXT(lop, lo_lckowner);
4910 		if (first >= lop->lo_end) {
4911 			/* not there yet */
4912 			lop = nlop;
4913 		} else if (first < lop->lo_first) {
4914 			/* new one starts before entry in list */
4915 			if (end <= lop->lo_first) {
4916 				/* no overlap between old and new */
4917 				error = nfsrv_dolocal(vp, lfp, flags,
4918 				    NFSLCK_UNLOCK, first, end, cfp, p);
4919 				if (error != 0)
4920 					break;
4921 				first = end;
4922 			} else {
4923 				/* handle fragment overlapped with new one */
4924 				error = nfsrv_dolocal(vp, lfp, flags,
4925 				    NFSLCK_UNLOCK, first, lop->lo_first, cfp,
4926 				    p);
4927 				if (error != 0)
4928 					break;
4929 				first = lop->lo_first;
4930 			}
4931 		} else {
4932 			/* new one overlaps this entry in list */
4933 			if (end <= lop->lo_end) {
4934 				/* overlaps all of new one */
4935 				error = nfsrv_dolocal(vp, lfp, flags,
4936 				    lop->lo_flags, first, end, cfp, p);
4937 				if (error != 0)
4938 					break;
4939 				first = end;
4940 			} else {
4941 				/* handle fragment overlapped with new one */
4942 				error = nfsrv_dolocal(vp, lfp, flags,
4943 				    lop->lo_flags, first, lop->lo_end, cfp, p);
4944 				if (error != 0)
4945 					break;
4946 				first = lop->lo_end;
4947 				lop = nlop;
4948 			}
4949 		}
4950 	}
4951 	if (first < end && error == 0)
4952 		/* handle fragment past end of list */
4953 		error = nfsrv_dolocal(vp, lfp, flags, NFSLCK_UNLOCK, first,
4954 		    end, cfp, p);
4955 	return (error);
4956 }
4957 
4958 /*
4959  * Local lock unlock. Unlock all byte ranges that are no longer locked
4960  * by NFSv4. To do this, unlock any subranges of first-->end that
4961  * do not overlap with the byte ranges of any lock in the lfp->lf_lock
4962  * list. This list has all locks for the file held by other
4963  * <clientid, lockowner> tuples. The list is ordered by increasing
4964  * lo_first value, but may have entries that overlap each other, for
4965  * the case of read locks.
4966  */
4967 static void
4968 nfsrv_localunlock(vnode_t vp, struct nfslockfile *lfp, uint64_t init_first,
4969     uint64_t init_end, NFSPROC_T *p)
4970 {
4971 	struct nfslock *lop;
4972 	uint64_t first, end, prevfirst;
4973 
4974 	first = init_first;
4975 	end = init_end;
4976 	while (first < init_end) {
4977 		/* Loop through all nfs locks, adjusting first and end */
4978 		prevfirst = 0;
4979 		LIST_FOREACH(lop, &lfp->lf_lock, lo_lckfile) {
4980 			KASSERT(prevfirst <= lop->lo_first,
4981 			    ("nfsv4 locks out of order"));
4982 			KASSERT(lop->lo_first < lop->lo_end,
4983 			    ("nfsv4 bogus lock"));
4984 			prevfirst = lop->lo_first;
4985 			if (first >= lop->lo_first &&
4986 			    first < lop->lo_end)
4987 				/*
4988 				 * Overlaps with initial part, so trim
4989 				 * off that initial part by moving first past
4990 				 * it.
4991 				 */
4992 				first = lop->lo_end;
4993 			else if (end > lop->lo_first &&
4994 			    lop->lo_first > first) {
4995 				/*
4996 				 * This lock defines the end of the
4997 				 * segment to unlock, so set end to the
4998 				 * start of it and break out of the loop.
4999 				 */
5000 				end = lop->lo_first;
5001 				break;
5002 			}
5003 			if (first >= end)
5004 				/*
5005 				 * There is no segment left to do, so
5006 				 * break out of this loop and then exit
5007 				 * the outer while() since first will be set
5008 				 * to end, which must equal init_end here.
5009 				 */
5010 				break;
5011 		}
5012 		if (first < end) {
5013 			/* Unlock this segment */
5014 			(void) nfsrv_dolocal(vp, lfp, NFSLCK_UNLOCK,
5015 			    NFSLCK_READ, first, end, NULL, p);
5016 			nfsrv_locallock_commit(lfp, NFSLCK_UNLOCK,
5017 			    first, end);
5018 		}
5019 		/*
5020 		 * Now move past this segment and look for any further
5021 		 * segment in the range, if there is one.
5022 		 */
5023 		first = end;
5024 		end = init_end;
5025 	}
5026 }
5027 
5028 /*
5029  * Do the local lock operation and update the rollback list, as required.
5030  * Perform the rollback and return the error if nfsvno_advlock() fails.
5031  */
5032 static int
5033 nfsrv_dolocal(vnode_t vp, struct nfslockfile *lfp, int flags, int oldflags,
5034     uint64_t first, uint64_t end, struct nfslockconflict *cfp, NFSPROC_T *p)
5035 {
5036 	struct nfsrollback *rlp;
5037 	int error, ltype, oldltype;
5038 
5039 	if (flags & NFSLCK_WRITE)
5040 		ltype = F_WRLCK;
5041 	else if (flags & NFSLCK_READ)
5042 		ltype = F_RDLCK;
5043 	else
5044 		ltype = F_UNLCK;
5045 	if (oldflags & NFSLCK_WRITE)
5046 		oldltype = F_WRLCK;
5047 	else if (oldflags & NFSLCK_READ)
5048 		oldltype = F_RDLCK;
5049 	else
5050 		oldltype = F_UNLCK;
5051 	if (ltype == oldltype || (oldltype == F_WRLCK && ltype == F_RDLCK))
5052 		/* nothing to do */
5053 		return (0);
5054 	error = nfsvno_advlock(vp, ltype, first, end, p);
5055 	if (error != 0) {
5056 		if (cfp != NULL) {
5057 			cfp->cl_clientid.lval[0] = 0;
5058 			cfp->cl_clientid.lval[1] = 0;
5059 			cfp->cl_first = 0;
5060 			cfp->cl_end = NFS64BITSSET;
5061 			cfp->cl_flags = NFSLCK_WRITE;
5062 			cfp->cl_ownerlen = 5;
5063 			NFSBCOPY("LOCAL", cfp->cl_owner, 5);
5064 		}
5065 		nfsrv_locallock_rollback(vp, lfp, p);
5066 	} else if (ltype != F_UNLCK) {
5067 		rlp = malloc(sizeof (struct nfsrollback), M_NFSDROLLBACK,
5068 		    M_WAITOK);
5069 		rlp->rlck_first = first;
5070 		rlp->rlck_end = end;
5071 		rlp->rlck_type = oldltype;
5072 		LIST_INSERT_HEAD(&lfp->lf_rollback, rlp, rlck_list);
5073 	}
5074 	return (error);
5075 }
5076 
5077 /*
5078  * Roll back local lock changes and free up the rollback list.
5079  */
5080 static void
5081 nfsrv_locallock_rollback(vnode_t vp, struct nfslockfile *lfp, NFSPROC_T *p)
5082 {
5083 	struct nfsrollback *rlp, *nrlp;
5084 
5085 	LIST_FOREACH_SAFE(rlp, &lfp->lf_rollback, rlck_list, nrlp) {
5086 		(void) nfsvno_advlock(vp, rlp->rlck_type, rlp->rlck_first,
5087 		    rlp->rlck_end, p);
5088 		free(rlp, M_NFSDROLLBACK);
5089 	}
5090 	LIST_INIT(&lfp->lf_rollback);
5091 }
5092 
5093 /*
5094  * Update local lock list and delete rollback list (ie now committed to the
5095  * local locks). Most of the work is done by the internal function.
5096  */
5097 static void
5098 nfsrv_locallock_commit(struct nfslockfile *lfp, int flags, uint64_t first,
5099     uint64_t end)
5100 {
5101 	struct nfsrollback *rlp, *nrlp;
5102 	struct nfslock *new_lop, *other_lop;
5103 
5104 	new_lop = malloc(sizeof (struct nfslock), M_NFSDLOCK, M_WAITOK);
5105 	if (flags & (NFSLCK_READ | NFSLCK_WRITE))
5106 		other_lop = malloc(sizeof (struct nfslock), M_NFSDLOCK,
5107 		    M_WAITOK);
5108 	else
5109 		other_lop = NULL;
5110 	new_lop->lo_flags = flags;
5111 	new_lop->lo_first = first;
5112 	new_lop->lo_end = end;
5113 	nfsrv_updatelock(NULL, &new_lop, &other_lop, lfp);
5114 	if (new_lop != NULL)
5115 		free(new_lop, M_NFSDLOCK);
5116 	if (other_lop != NULL)
5117 		free(other_lop, M_NFSDLOCK);
5118 
5119 	/* and get rid of the rollback list */
5120 	LIST_FOREACH_SAFE(rlp, &lfp->lf_rollback, rlck_list, nrlp)
5121 		free(rlp, M_NFSDROLLBACK);
5122 	LIST_INIT(&lfp->lf_rollback);
5123 }
5124 
5125 /*
5126  * Lock the struct nfslockfile for local lock updating.
5127  */
5128 static void
5129 nfsrv_locklf(struct nfslockfile *lfp)
5130 {
5131 	int gotlock;
5132 
5133 	/* lf_usecount ensures *lfp won't be free'd */
5134 	lfp->lf_usecount++;
5135 	do {
5136 		gotlock = nfsv4_lock(&lfp->lf_locallock_lck, 1, NULL,
5137 		    NFSSTATEMUTEXPTR);
5138 	} while (gotlock == 0);
5139 	lfp->lf_usecount--;
5140 }
5141 
5142 /*
5143  * Unlock the struct nfslockfile after local lock updating.
5144  */
5145 static void
5146 nfsrv_unlocklf(struct nfslockfile *lfp)
5147 {
5148 
5149 	nfsv4_unlock(&lfp->lf_locallock_lck, 0);
5150 }
5151 
5152