xref: /freebsd/sys/fs/nfsserver/nfs_nfsdstate.c (revision 27c43fe1f3795622c5bd4bbfc465a29a800c0799)
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, error = 0;
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 		error = NFSERR_RESOURCE;
149 		goto out;
150 	}
151 
152 	if (nfsrv_issuedelegs == 0 ||
153 	    ((nd->nd_flag & ND_GSS) != 0 && nfsrv_nogsscallback != 0))
154 		/*
155 		 * Don't do callbacks when delegations are disabled or
156 		 * for AUTH_GSS unless enabled via nfsrv_nogsscallback.
157 		 * If establishing a callback connection is attempted
158 		 * when a firewall is blocking the callback path, the
159 		 * server may wait too long for the connect attempt to
160 		 * succeed during the Open. Some clients, such as Linux,
161 		 * may timeout and give up on the Open before the server
162 		 * replies. Also, since AUTH_GSS callbacks are not
163 		 * yet interoperability tested, they might cause the
164 		 * server to crap out, if they get past the Init call to
165 		 * the client.
166 		 */
167 		new_clp->lc_program = 0;
168 
169 	/* Lock out other nfsd threads */
170 	NFSLOCKV4ROOTMUTEX();
171 	nfsv4_relref(&nfsv4rootfs_lock);
172 	do {
173 		igotlock = nfsv4_lock(&nfsv4rootfs_lock, 1, NULL,
174 		    NFSV4ROOTLOCKMUTEXPTR, NULL);
175 	} while (!igotlock);
176 	NFSUNLOCKV4ROOTMUTEX();
177 
178 	/*
179 	 * Search for a match in the client list.
180 	 */
181 	gotit = i = 0;
182 	while (i < NFSCLIENTHASHSIZE && !gotit) {
183 	    LIST_FOREACH(clp, &nfsclienthash[i], lc_hash) {
184 		if (new_clp->lc_idlen == clp->lc_idlen &&
185 		    !NFSBCMP(new_clp->lc_id, clp->lc_id, clp->lc_idlen)) {
186 			gotit = 1;
187 			break;
188 		}
189 	    }
190 	    i++;
191 	}
192 	if (!gotit ||
193 	    (clp->lc_flags & (LCL_NEEDSCONFIRM | LCL_ADMINREVOKED))) {
194 		/*
195 		 * Get rid of the old one.
196 		 */
197 		if (i != NFSCLIENTHASHSIZE) {
198 			LIST_REMOVE(clp, lc_hash);
199 			nfsrv_cleanclient(clp, p);
200 			nfsrv_freedeleglist(&clp->lc_deleg);
201 			nfsrv_freedeleglist(&clp->lc_olddeleg);
202 			zapit = 1;
203 		}
204 		/*
205 		 * Add it after assigning a client id to it.
206 		 */
207 		new_clp->lc_flags |= LCL_NEEDSCONFIRM;
208 		confirmp->qval = new_clp->lc_confirm.qval = ++confirm_index;
209 		clientidp->lval[0] = new_clp->lc_clientid.lval[0] =
210 		    (u_int32_t)nfsrvboottime;
211 		clientidp->lval[1] = new_clp->lc_clientid.lval[1] =
212 		    nfsrv_nextclientindex();
213 		new_clp->lc_stateindex = 0;
214 		new_clp->lc_statemaxindex = 0;
215 		new_clp->lc_cbref = 0;
216 		new_clp->lc_expiry = nfsrv_leaseexpiry();
217 		LIST_INIT(&new_clp->lc_open);
218 		LIST_INIT(&new_clp->lc_deleg);
219 		LIST_INIT(&new_clp->lc_olddeleg);
220 		for (i = 0; i < NFSSTATEHASHSIZE; i++)
221 			LIST_INIT(&new_clp->lc_stateid[i]);
222 		LIST_INSERT_HEAD(NFSCLIENTHASH(new_clp->lc_clientid), new_clp,
223 		    lc_hash);
224 		newnfsstats.srvclients++;
225 		nfsrv_openpluslock++;
226 		nfsrv_clients++;
227 		NFSLOCKV4ROOTMUTEX();
228 		nfsv4_unlock(&nfsv4rootfs_lock, 1);
229 		NFSUNLOCKV4ROOTMUTEX();
230 		if (zapit)
231 			nfsrv_zapclient(clp, p);
232 		*new_clpp = NULL;
233 		goto out;
234 	}
235 
236 	/*
237 	 * Now, handle the cases where the id is already issued.
238 	 */
239 	if (nfsrv_notsamecredname(nd, clp)) {
240 	    /*
241 	     * Check to see if there is expired state that should go away.
242 	     */
243 	    if (clp->lc_expiry < NFSD_MONOSEC &&
244 	        (!LIST_EMPTY(&clp->lc_open) || !LIST_EMPTY(&clp->lc_deleg))) {
245 		nfsrv_cleanclient(clp, p);
246 		nfsrv_freedeleglist(&clp->lc_deleg);
247 	    }
248 
249 	    /*
250 	     * If there is outstanding state, then reply NFSERR_CLIDINUSE per
251 	     * RFC3530 Sec. 8.1.2 last para.
252 	     */
253 	    if (!LIST_EMPTY(&clp->lc_deleg)) {
254 		hasstate = 1;
255 	    } else if (LIST_EMPTY(&clp->lc_open)) {
256 		hasstate = 0;
257 	    } else {
258 		hasstate = 0;
259 		/* Look for an Open on the OpenOwner */
260 		LIST_FOREACH(stp, &clp->lc_open, ls_list) {
261 		    if (!LIST_EMPTY(&stp->ls_open)) {
262 			hasstate = 1;
263 			break;
264 		    }
265 		}
266 	    }
267 	    if (hasstate) {
268 		/*
269 		 * If the uid doesn't match, return NFSERR_CLIDINUSE after
270 		 * filling out the correct ipaddr and portnum.
271 		 */
272 		sad = NFSSOCKADDR(new_clp->lc_req.nr_nam, struct sockaddr_in *);
273 		rad = NFSSOCKADDR(clp->lc_req.nr_nam, struct sockaddr_in *);
274 		sad->sin_addr.s_addr = rad->sin_addr.s_addr;
275 		sad->sin_port = rad->sin_port;
276 		NFSLOCKV4ROOTMUTEX();
277 		nfsv4_unlock(&nfsv4rootfs_lock, 1);
278 		NFSUNLOCKV4ROOTMUTEX();
279 		error = NFSERR_CLIDINUSE;
280 		goto out;
281 	    }
282 	}
283 
284 	if (NFSBCMP(new_clp->lc_verf, clp->lc_verf, NFSX_VERF)) {
285 		/*
286 		 * If the verifier has changed, the client has rebooted
287 		 * and a new client id is issued. The old state info
288 		 * can be thrown away once the SETCLIENTID_CONFIRM occurs.
289 		 */
290 		LIST_REMOVE(clp, lc_hash);
291 		new_clp->lc_flags |= LCL_NEEDSCONFIRM;
292 		confirmp->qval = new_clp->lc_confirm.qval = ++confirm_index;
293 		clientidp->lval[0] = new_clp->lc_clientid.lval[0] =
294 		    nfsrvboottime;
295 		clientidp->lval[1] = new_clp->lc_clientid.lval[1] =
296 		    nfsrv_nextclientindex();
297 		new_clp->lc_stateindex = 0;
298 		new_clp->lc_statemaxindex = 0;
299 		new_clp->lc_cbref = 0;
300 		new_clp->lc_expiry = nfsrv_leaseexpiry();
301 
302 		/*
303 		 * Save the state until confirmed.
304 		 */
305 		LIST_NEWHEAD(&new_clp->lc_open, &clp->lc_open, ls_list);
306 		LIST_FOREACH(tstp, &new_clp->lc_open, ls_list)
307 			tstp->ls_clp = new_clp;
308 		LIST_NEWHEAD(&new_clp->lc_deleg, &clp->lc_deleg, ls_list);
309 		LIST_FOREACH(tstp, &new_clp->lc_deleg, ls_list)
310 			tstp->ls_clp = new_clp;
311 		LIST_NEWHEAD(&new_clp->lc_olddeleg, &clp->lc_olddeleg,
312 		    ls_list);
313 		LIST_FOREACH(tstp, &new_clp->lc_olddeleg, ls_list)
314 			tstp->ls_clp = new_clp;
315 		for (i = 0; i < NFSSTATEHASHSIZE; i++) {
316 			LIST_NEWHEAD(&new_clp->lc_stateid[i],
317 			    &clp->lc_stateid[i], ls_hash);
318 			LIST_FOREACH(tstp, &new_clp->lc_stateid[i], ls_hash)
319 				tstp->ls_clp = new_clp;
320 		}
321 		LIST_INSERT_HEAD(NFSCLIENTHASH(new_clp->lc_clientid), new_clp,
322 		    lc_hash);
323 		newnfsstats.srvclients++;
324 		nfsrv_openpluslock++;
325 		nfsrv_clients++;
326 		NFSLOCKV4ROOTMUTEX();
327 		nfsv4_unlock(&nfsv4rootfs_lock, 1);
328 		NFSUNLOCKV4ROOTMUTEX();
329 
330 		/*
331 		 * Must wait until any outstanding callback on the old clp
332 		 * completes.
333 		 */
334 		NFSLOCKSTATE();
335 		while (clp->lc_cbref) {
336 			clp->lc_flags |= LCL_WAKEUPWANTED;
337 			(void)mtx_sleep(clp, NFSSTATEMUTEXPTR, PZERO - 1,
338 			    "nfsd clp", 10 * hz);
339 		}
340 		NFSUNLOCKSTATE();
341 		nfsrv_zapclient(clp, p);
342 		*new_clpp = NULL;
343 		goto out;
344 	}
345 	/*
346 	 * id and verifier match, so update the net address info
347 	 * and get rid of any existing callback authentication
348 	 * handle, so a new one will be acquired.
349 	 */
350 	LIST_REMOVE(clp, lc_hash);
351 	new_clp->lc_flags |= (LCL_NEEDSCONFIRM | LCL_DONTCLEAN);
352 	new_clp->lc_expiry = nfsrv_leaseexpiry();
353 	confirmp->qval = new_clp->lc_confirm.qval = ++confirm_index;
354 	clientidp->lval[0] = new_clp->lc_clientid.lval[0] =
355 	    clp->lc_clientid.lval[0];
356 	clientidp->lval[1] = new_clp->lc_clientid.lval[1] =
357 	    clp->lc_clientid.lval[1];
358 	new_clp->lc_delegtime = clp->lc_delegtime;
359 	new_clp->lc_stateindex = clp->lc_stateindex;
360 	new_clp->lc_statemaxindex = clp->lc_statemaxindex;
361 	new_clp->lc_cbref = 0;
362 	LIST_NEWHEAD(&new_clp->lc_open, &clp->lc_open, ls_list);
363 	LIST_FOREACH(tstp, &new_clp->lc_open, ls_list)
364 		tstp->ls_clp = new_clp;
365 	LIST_NEWHEAD(&new_clp->lc_deleg, &clp->lc_deleg, ls_list);
366 	LIST_FOREACH(tstp, &new_clp->lc_deleg, ls_list)
367 		tstp->ls_clp = new_clp;
368 	LIST_NEWHEAD(&new_clp->lc_olddeleg, &clp->lc_olddeleg, ls_list);
369 	LIST_FOREACH(tstp, &new_clp->lc_olddeleg, ls_list)
370 		tstp->ls_clp = new_clp;
371 	for (i = 0; i < NFSSTATEHASHSIZE; i++) {
372 		LIST_NEWHEAD(&new_clp->lc_stateid[i], &clp->lc_stateid[i],
373 		    ls_hash);
374 		LIST_FOREACH(tstp, &new_clp->lc_stateid[i], ls_hash)
375 			tstp->ls_clp = new_clp;
376 	}
377 	LIST_INSERT_HEAD(NFSCLIENTHASH(new_clp->lc_clientid), new_clp,
378 	    lc_hash);
379 	newnfsstats.srvclients++;
380 	nfsrv_openpluslock++;
381 	nfsrv_clients++;
382 	NFSLOCKV4ROOTMUTEX();
383 	nfsv4_unlock(&nfsv4rootfs_lock, 1);
384 	NFSUNLOCKV4ROOTMUTEX();
385 
386 	/*
387 	 * Must wait until any outstanding callback on the old clp
388 	 * completes.
389 	 */
390 	NFSLOCKSTATE();
391 	while (clp->lc_cbref) {
392 		clp->lc_flags |= LCL_WAKEUPWANTED;
393 		(void)mtx_sleep(clp, NFSSTATEMUTEXPTR, PZERO - 1, "nfsd clp",
394 		    10 * hz);
395 	}
396 	NFSUNLOCKSTATE();
397 	nfsrv_zapclient(clp, p);
398 	*new_clpp = NULL;
399 
400 out:
401 	NFSEXITCODE2(error, nd);
402 	return (error);
403 }
404 
405 /*
406  * Check to see if the client id exists and optionally confirm it.
407  */
408 APPLESTATIC int
409 nfsrv_getclient(nfsquad_t clientid, int opflags, struct nfsclient **clpp,
410     nfsquad_t confirm, struct nfsrv_descript *nd, NFSPROC_T *p)
411 {
412 	struct nfsclient *clp;
413 	struct nfsstate *stp;
414 	int i;
415 	struct nfsclienthashhead *hp;
416 	int error = 0, igotlock, doneok;
417 
418 	if (clpp)
419 		*clpp = NULL;
420 	if (nfsrvboottime != clientid.lval[0]) {
421 		error = NFSERR_STALECLIENTID;
422 		goto out;
423 	}
424 
425 	/*
426 	 * If called with opflags == CLOPS_RENEW, the State Lock is
427 	 * already held. Otherwise, we need to get either that or,
428 	 * for the case of Confirm, lock out the nfsd threads.
429 	 */
430 	if (opflags & CLOPS_CONFIRM) {
431 		NFSLOCKV4ROOTMUTEX();
432 		nfsv4_relref(&nfsv4rootfs_lock);
433 		do {
434 			igotlock = nfsv4_lock(&nfsv4rootfs_lock, 1, NULL,
435 			    NFSV4ROOTLOCKMUTEXPTR, NULL);
436 		} while (!igotlock);
437 		NFSUNLOCKV4ROOTMUTEX();
438 	} else if (opflags != CLOPS_RENEW) {
439 		NFSLOCKSTATE();
440 	}
441 
442 	hp = NFSCLIENTHASH(clientid);
443 	LIST_FOREACH(clp, hp, lc_hash) {
444 		if (clp->lc_clientid.lval[1] == clientid.lval[1])
445 			break;
446 	}
447 	if (clp == LIST_END(hp)) {
448 		if (opflags & CLOPS_CONFIRM)
449 			error = NFSERR_STALECLIENTID;
450 		else
451 			error = NFSERR_EXPIRED;
452 	} else if (clp->lc_flags & LCL_ADMINREVOKED) {
453 		/*
454 		 * If marked admin revoked, just return the error.
455 		 */
456 		error = NFSERR_ADMINREVOKED;
457 	}
458 	if (error) {
459 		if (opflags & CLOPS_CONFIRM) {
460 			NFSLOCKV4ROOTMUTEX();
461 			nfsv4_unlock(&nfsv4rootfs_lock, 1);
462 			NFSUNLOCKV4ROOTMUTEX();
463 		} else if (opflags != CLOPS_RENEW) {
464 			NFSUNLOCKSTATE();
465 		}
466 		goto out;
467 	}
468 
469 	/*
470 	 * Perform any operations specified by the opflags.
471 	 */
472 	if (opflags & CLOPS_CONFIRM) {
473 		if (clp->lc_confirm.qval != confirm.qval)
474 			error = NFSERR_STALECLIENTID;
475 		else if (nfsrv_notsamecredname(nd, clp))
476 			error = NFSERR_CLIDINUSE;
477 
478 		if (!error) {
479 		    if ((clp->lc_flags & (LCL_NEEDSCONFIRM | LCL_DONTCLEAN)) ==
480 			LCL_NEEDSCONFIRM) {
481 			/*
482 			 * Hang onto the delegations (as old delegations)
483 			 * for an Open with CLAIM_DELEGATE_PREV unless in
484 			 * grace, but get rid of the rest of the state.
485 			 */
486 			nfsrv_cleanclient(clp, p);
487 			nfsrv_freedeleglist(&clp->lc_olddeleg);
488 			if (nfsrv_checkgrace(0)) {
489 			    /* In grace, so just delete delegations */
490 			    nfsrv_freedeleglist(&clp->lc_deleg);
491 			} else {
492 			    LIST_FOREACH(stp, &clp->lc_deleg, ls_list)
493 				stp->ls_flags |= NFSLCK_OLDDELEG;
494 			    clp->lc_delegtime = NFSD_MONOSEC +
495 				nfsrv_lease + NFSRV_LEASEDELTA;
496 			    LIST_NEWHEAD(&clp->lc_olddeleg, &clp->lc_deleg,
497 				ls_list);
498 			}
499 		    }
500 		    clp->lc_flags &= ~(LCL_NEEDSCONFIRM | LCL_DONTCLEAN);
501 		    if (clp->lc_program)
502 			clp->lc_flags |= LCL_NEEDSCBNULL;
503 		}
504 	} else if (clp->lc_flags & LCL_NEEDSCONFIRM) {
505 		error = NFSERR_EXPIRED;
506 	}
507 
508 	/*
509 	 * If called by the Renew Op, we must check the principal.
510 	 */
511 	if (!error && (opflags & CLOPS_RENEWOP)) {
512 	    if (nfsrv_notsamecredname(nd, clp)) {
513 		doneok = 0;
514 		for (i = 0; i < NFSSTATEHASHSIZE && doneok == 0; i++) {
515 		    LIST_FOREACH(stp, &clp->lc_stateid[i], ls_hash) {
516 			if ((stp->ls_flags & NFSLCK_OPEN) &&
517 			    stp->ls_uid == nd->nd_cred->cr_uid) {
518 				doneok = 1;
519 				break;
520 			}
521 		    }
522 		}
523 		if (!doneok)
524 			error = NFSERR_ACCES;
525 	    }
526 	    if (!error && (clp->lc_flags & LCL_CBDOWN))
527 		error = NFSERR_CBPATHDOWN;
528 	}
529 	if ((!error || error == NFSERR_CBPATHDOWN) &&
530 	     (opflags & CLOPS_RENEW)) {
531 		clp->lc_expiry = nfsrv_leaseexpiry();
532 	}
533 	if (opflags & CLOPS_CONFIRM) {
534 		NFSLOCKV4ROOTMUTEX();
535 		nfsv4_unlock(&nfsv4rootfs_lock, 1);
536 		NFSUNLOCKV4ROOTMUTEX();
537 	} else if (opflags != CLOPS_RENEW) {
538 		NFSUNLOCKSTATE();
539 	}
540 	if (clpp)
541 		*clpp = clp;
542 
543 out:
544 	NFSEXITCODE2(error, nd);
545 	return (error);
546 }
547 
548 /*
549  * Called from the new nfssvc syscall to admin revoke a clientid.
550  * Returns 0 for success, error otherwise.
551  */
552 APPLESTATIC int
553 nfsrv_adminrevoke(struct nfsd_clid *revokep, NFSPROC_T *p)
554 {
555 	struct nfsclient *clp = NULL;
556 	int i, error = 0;
557 	int gotit, igotlock;
558 
559 	/*
560 	 * First, lock out the nfsd so that state won't change while the
561 	 * revocation record is being written to the stable storage restart
562 	 * file.
563 	 */
564 	NFSLOCKV4ROOTMUTEX();
565 	do {
566 		igotlock = nfsv4_lock(&nfsv4rootfs_lock, 1, NULL,
567 		    NFSV4ROOTLOCKMUTEXPTR, NULL);
568 	} while (!igotlock);
569 	NFSUNLOCKV4ROOTMUTEX();
570 
571 	/*
572 	 * Search for a match in the client list.
573 	 */
574 	gotit = i = 0;
575 	while (i < NFSCLIENTHASHSIZE && !gotit) {
576 	    LIST_FOREACH(clp, &nfsclienthash[i], lc_hash) {
577 		if (revokep->nclid_idlen == clp->lc_idlen &&
578 		    !NFSBCMP(revokep->nclid_id, clp->lc_id, clp->lc_idlen)) {
579 			gotit = 1;
580 			break;
581 		}
582 	    }
583 	    i++;
584 	}
585 	if (!gotit) {
586 		NFSLOCKV4ROOTMUTEX();
587 		nfsv4_unlock(&nfsv4rootfs_lock, 0);
588 		NFSUNLOCKV4ROOTMUTEX();
589 		error = EPERM;
590 		goto out;
591 	}
592 
593 	/*
594 	 * Now, write out the revocation record
595 	 */
596 	nfsrv_writestable(clp->lc_id, clp->lc_idlen, NFSNST_REVOKE, p);
597 	nfsrv_backupstable();
598 
599 	/*
600 	 * and clear out the state, marking the clientid revoked.
601 	 */
602 	clp->lc_flags &= ~LCL_CALLBACKSON;
603 	clp->lc_flags |= LCL_ADMINREVOKED;
604 	nfsrv_cleanclient(clp, p);
605 	nfsrv_freedeleglist(&clp->lc_deleg);
606 	nfsrv_freedeleglist(&clp->lc_olddeleg);
607 	NFSLOCKV4ROOTMUTEX();
608 	nfsv4_unlock(&nfsv4rootfs_lock, 0);
609 	NFSUNLOCKV4ROOTMUTEX();
610 
611 out:
612 	NFSEXITCODE(error);
613 	return (error);
614 }
615 
616 /*
617  * Dump out stats for all clients. Called from nfssvc(2), that is used
618  * newnfsstats.
619  */
620 APPLESTATIC void
621 nfsrv_dumpclients(struct nfsd_dumpclients *dumpp, int maxcnt)
622 {
623 	struct nfsclient *clp;
624 	int i = 0, cnt = 0;
625 
626 	/*
627 	 * First, get a reference on the nfsv4rootfs_lock so that an
628 	 * exclusive lock cannot be acquired while dumping the clients.
629 	 */
630 	NFSLOCKV4ROOTMUTEX();
631 	nfsv4_getref(&nfsv4rootfs_lock, NULL, NFSV4ROOTLOCKMUTEXPTR, NULL);
632 	NFSUNLOCKV4ROOTMUTEX();
633 	NFSLOCKSTATE();
634 	/*
635 	 * Rattle through the client lists until done.
636 	 */
637 	while (i < NFSCLIENTHASHSIZE && cnt < maxcnt) {
638 	    clp = LIST_FIRST(&nfsclienthash[i]);
639 	    while (clp != LIST_END(&nfsclienthash[i]) && cnt < maxcnt) {
640 		nfsrv_dumpaclient(clp, &dumpp[cnt]);
641 		cnt++;
642 		clp = LIST_NEXT(clp, lc_hash);
643 	    }
644 	    i++;
645 	}
646 	if (cnt < maxcnt)
647 	    dumpp[cnt].ndcl_clid.nclid_idlen = 0;
648 	NFSUNLOCKSTATE();
649 	NFSLOCKV4ROOTMUTEX();
650 	nfsv4_relref(&nfsv4rootfs_lock);
651 	NFSUNLOCKV4ROOTMUTEX();
652 }
653 
654 /*
655  * Dump stats for a client. Must be called with the NFSSTATELOCK and spl'd.
656  */
657 static void
658 nfsrv_dumpaclient(struct nfsclient *clp, struct nfsd_dumpclients *dumpp)
659 {
660 	struct nfsstate *stp, *openstp, *lckownstp;
661 	struct nfslock *lop;
662 	struct sockaddr *sad;
663 	struct sockaddr_in *rad;
664 	struct sockaddr_in6 *rad6;
665 
666 	dumpp->ndcl_nopenowners = dumpp->ndcl_nlockowners = 0;
667 	dumpp->ndcl_nopens = dumpp->ndcl_nlocks = 0;
668 	dumpp->ndcl_ndelegs = dumpp->ndcl_nolddelegs = 0;
669 	dumpp->ndcl_flags = clp->lc_flags;
670 	dumpp->ndcl_clid.nclid_idlen = clp->lc_idlen;
671 	NFSBCOPY(clp->lc_id, dumpp->ndcl_clid.nclid_id, clp->lc_idlen);
672 	sad = NFSSOCKADDR(clp->lc_req.nr_nam, struct sockaddr *);
673 	dumpp->ndcl_addrfam = sad->sa_family;
674 	if (sad->sa_family == AF_INET) {
675 		rad = (struct sockaddr_in *)sad;
676 		dumpp->ndcl_cbaddr.sin_addr = rad->sin_addr;
677 	} else {
678 		rad6 = (struct sockaddr_in6 *)sad;
679 		dumpp->ndcl_cbaddr.sin6_addr = rad6->sin6_addr;
680 	}
681 
682 	/*
683 	 * Now, scan the state lists and total up the opens and locks.
684 	 */
685 	LIST_FOREACH(stp, &clp->lc_open, ls_list) {
686 	    dumpp->ndcl_nopenowners++;
687 	    LIST_FOREACH(openstp, &stp->ls_open, ls_list) {
688 		dumpp->ndcl_nopens++;
689 		LIST_FOREACH(lckownstp, &openstp->ls_open, ls_list) {
690 		    dumpp->ndcl_nlockowners++;
691 		    LIST_FOREACH(lop, &lckownstp->ls_lock, lo_lckowner) {
692 			dumpp->ndcl_nlocks++;
693 		    }
694 		}
695 	    }
696 	}
697 
698 	/*
699 	 * and the delegation lists.
700 	 */
701 	LIST_FOREACH(stp, &clp->lc_deleg, ls_list) {
702 	    dumpp->ndcl_ndelegs++;
703 	}
704 	LIST_FOREACH(stp, &clp->lc_olddeleg, ls_list) {
705 	    dumpp->ndcl_nolddelegs++;
706 	}
707 }
708 
709 /*
710  * Dump out lock stats for a file.
711  */
712 APPLESTATIC void
713 nfsrv_dumplocks(vnode_t vp, struct nfsd_dumplocks *ldumpp, int maxcnt,
714     NFSPROC_T *p)
715 {
716 	struct nfsstate *stp;
717 	struct nfslock *lop;
718 	int cnt = 0;
719 	struct nfslockfile *lfp;
720 	struct sockaddr *sad;
721 	struct sockaddr_in *rad;
722 	struct sockaddr_in6 *rad6;
723 	int ret;
724 	fhandle_t nfh;
725 
726 	ret = nfsrv_getlockfh(vp, 0, NULL, &nfh, p);
727 	/*
728 	 * First, get a reference on the nfsv4rootfs_lock so that an
729 	 * exclusive lock on it cannot be acquired while dumping the locks.
730 	 */
731 	NFSLOCKV4ROOTMUTEX();
732 	nfsv4_getref(&nfsv4rootfs_lock, NULL, NFSV4ROOTLOCKMUTEXPTR, NULL);
733 	NFSUNLOCKV4ROOTMUTEX();
734 	NFSLOCKSTATE();
735 	if (!ret)
736 		ret = nfsrv_getlockfile(0, NULL, &lfp, &nfh, 0);
737 	if (ret) {
738 		ldumpp[0].ndlck_clid.nclid_idlen = 0;
739 		NFSUNLOCKSTATE();
740 		NFSLOCKV4ROOTMUTEX();
741 		nfsv4_relref(&nfsv4rootfs_lock);
742 		NFSUNLOCKV4ROOTMUTEX();
743 		return;
744 	}
745 
746 	/*
747 	 * For each open share on file, dump it out.
748 	 */
749 	stp = LIST_FIRST(&lfp->lf_open);
750 	while (stp != LIST_END(&lfp->lf_open) && cnt < maxcnt) {
751 		ldumpp[cnt].ndlck_flags = stp->ls_flags;
752 		ldumpp[cnt].ndlck_stateid.seqid = stp->ls_stateid.seqid;
753 		ldumpp[cnt].ndlck_stateid.other[0] = stp->ls_stateid.other[0];
754 		ldumpp[cnt].ndlck_stateid.other[1] = stp->ls_stateid.other[1];
755 		ldumpp[cnt].ndlck_stateid.other[2] = stp->ls_stateid.other[2];
756 		ldumpp[cnt].ndlck_owner.nclid_idlen =
757 		    stp->ls_openowner->ls_ownerlen;
758 		NFSBCOPY(stp->ls_openowner->ls_owner,
759 		    ldumpp[cnt].ndlck_owner.nclid_id,
760 		    stp->ls_openowner->ls_ownerlen);
761 		ldumpp[cnt].ndlck_clid.nclid_idlen = stp->ls_clp->lc_idlen;
762 		NFSBCOPY(stp->ls_clp->lc_id, ldumpp[cnt].ndlck_clid.nclid_id,
763 		    stp->ls_clp->lc_idlen);
764 		sad=NFSSOCKADDR(stp->ls_clp->lc_req.nr_nam, struct sockaddr *);
765 		ldumpp[cnt].ndlck_addrfam = sad->sa_family;
766 		if (sad->sa_family == AF_INET) {
767 			rad = (struct sockaddr_in *)sad;
768 			ldumpp[cnt].ndlck_cbaddr.sin_addr = rad->sin_addr;
769 		} else {
770 			rad6 = (struct sockaddr_in6 *)sad;
771 			ldumpp[cnt].ndlck_cbaddr.sin6_addr = rad6->sin6_addr;
772 		}
773 		stp = LIST_NEXT(stp, ls_file);
774 		cnt++;
775 	}
776 
777 	/*
778 	 * and all locks.
779 	 */
780 	lop = LIST_FIRST(&lfp->lf_lock);
781 	while (lop != LIST_END(&lfp->lf_lock) && cnt < maxcnt) {
782 		stp = lop->lo_stp;
783 		ldumpp[cnt].ndlck_flags = lop->lo_flags;
784 		ldumpp[cnt].ndlck_first = lop->lo_first;
785 		ldumpp[cnt].ndlck_end = lop->lo_end;
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 = stp->ls_ownerlen;
791 		NFSBCOPY(stp->ls_owner, ldumpp[cnt].ndlck_owner.nclid_id,
792 		    stp->ls_ownerlen);
793 		ldumpp[cnt].ndlck_clid.nclid_idlen = stp->ls_clp->lc_idlen;
794 		NFSBCOPY(stp->ls_clp->lc_id, ldumpp[cnt].ndlck_clid.nclid_id,
795 		    stp->ls_clp->lc_idlen);
796 		sad=NFSSOCKADDR(stp->ls_clp->lc_req.nr_nam, struct sockaddr *);
797 		ldumpp[cnt].ndlck_addrfam = sad->sa_family;
798 		if (sad->sa_family == AF_INET) {
799 			rad = (struct sockaddr_in *)sad;
800 			ldumpp[cnt].ndlck_cbaddr.sin_addr = rad->sin_addr;
801 		} else {
802 			rad6 = (struct sockaddr_in6 *)sad;
803 			ldumpp[cnt].ndlck_cbaddr.sin6_addr = rad6->sin6_addr;
804 		}
805 		lop = LIST_NEXT(lop, lo_lckfile);
806 		cnt++;
807 	}
808 
809 	/*
810 	 * and the delegations.
811 	 */
812 	stp = LIST_FIRST(&lfp->lf_deleg);
813 	while (stp != LIST_END(&lfp->lf_deleg) && cnt < maxcnt) {
814 		ldumpp[cnt].ndlck_flags = stp->ls_flags;
815 		ldumpp[cnt].ndlck_stateid.seqid = stp->ls_stateid.seqid;
816 		ldumpp[cnt].ndlck_stateid.other[0] = stp->ls_stateid.other[0];
817 		ldumpp[cnt].ndlck_stateid.other[1] = stp->ls_stateid.other[1];
818 		ldumpp[cnt].ndlck_stateid.other[2] = stp->ls_stateid.other[2];
819 		ldumpp[cnt].ndlck_owner.nclid_idlen = 0;
820 		ldumpp[cnt].ndlck_clid.nclid_idlen = stp->ls_clp->lc_idlen;
821 		NFSBCOPY(stp->ls_clp->lc_id, ldumpp[cnt].ndlck_clid.nclid_id,
822 		    stp->ls_clp->lc_idlen);
823 		sad=NFSSOCKADDR(stp->ls_clp->lc_req.nr_nam, struct sockaddr *);
824 		ldumpp[cnt].ndlck_addrfam = sad->sa_family;
825 		if (sad->sa_family == AF_INET) {
826 			rad = (struct sockaddr_in *)sad;
827 			ldumpp[cnt].ndlck_cbaddr.sin_addr = rad->sin_addr;
828 		} else {
829 			rad6 = (struct sockaddr_in6 *)sad;
830 			ldumpp[cnt].ndlck_cbaddr.sin6_addr = rad6->sin6_addr;
831 		}
832 		stp = LIST_NEXT(stp, ls_file);
833 		cnt++;
834 	}
835 
836 	/*
837 	 * If list isn't full, mark end of list by setting the client name
838 	 * to zero length.
839 	 */
840 	if (cnt < maxcnt)
841 		ldumpp[cnt].ndlck_clid.nclid_idlen = 0;
842 	NFSUNLOCKSTATE();
843 	NFSLOCKV4ROOTMUTEX();
844 	nfsv4_relref(&nfsv4rootfs_lock);
845 	NFSUNLOCKV4ROOTMUTEX();
846 }
847 
848 /*
849  * Server timer routine. It can scan any linked list, so long
850  * as it holds the spin/mutex lock and there is no exclusive lock on
851  * nfsv4rootfs_lock.
852  * (For OpenBSD, a kthread is ok. For FreeBSD, I think it is ok
853  *  to do this from a callout, since the spin locks work. For
854  *  Darwin, I'm not sure what will work correctly yet.)
855  * Should be called once per second.
856  */
857 APPLESTATIC void
858 nfsrv_servertimer(void)
859 {
860 	struct nfsclient *clp, *nclp;
861 	struct nfsstate *stp, *nstp;
862 	int got_ref, i;
863 
864 	/*
865 	 * Make sure nfsboottime is set. This is used by V3 as well
866 	 * as V4. Note that nfsboottime is not nfsrvboottime, which is
867 	 * only used by the V4 server for leases.
868 	 */
869 	if (nfsboottime.tv_sec == 0)
870 		NFSSETBOOTTIME(nfsboottime);
871 
872 	/*
873 	 * If server hasn't started yet, just return.
874 	 */
875 	NFSLOCKSTATE();
876 	if (nfsrv_stablefirst.nsf_eograce == 0) {
877 		NFSUNLOCKSTATE();
878 		return;
879 	}
880 	if (!(nfsrv_stablefirst.nsf_flags & NFSNSF_UPDATEDONE)) {
881 		if (!(nfsrv_stablefirst.nsf_flags & NFSNSF_GRACEOVER) &&
882 		    NFSD_MONOSEC > nfsrv_stablefirst.nsf_eograce)
883 			nfsrv_stablefirst.nsf_flags |=
884 			    (NFSNSF_GRACEOVER | NFSNSF_NEEDLOCK);
885 		NFSUNLOCKSTATE();
886 		return;
887 	}
888 
889 	/*
890 	 * Try and get a reference count on the nfsv4rootfs_lock so that
891 	 * no nfsd thread can acquire an exclusive lock on it before this
892 	 * call is done. If it is already exclusively locked, just return.
893 	 */
894 	NFSLOCKV4ROOTMUTEX();
895 	got_ref = nfsv4_getref_nonblock(&nfsv4rootfs_lock);
896 	NFSUNLOCKV4ROOTMUTEX();
897 	if (got_ref == 0) {
898 		NFSUNLOCKSTATE();
899 		return;
900 	}
901 
902 	/*
903 	 * For each client...
904 	 */
905 	for (i = 0; i < NFSCLIENTHASHSIZE; i++) {
906 	    clp = LIST_FIRST(&nfsclienthash[i]);
907 	    while (clp != LIST_END(&nfsclienthash[i])) {
908 		nclp = LIST_NEXT(clp, lc_hash);
909 		if (!(clp->lc_flags & LCL_EXPIREIT)) {
910 		    if (((clp->lc_expiry + NFSRV_STALELEASE) < NFSD_MONOSEC
911 			 && ((LIST_EMPTY(&clp->lc_deleg)
912 			      && LIST_EMPTY(&clp->lc_open)) ||
913 			     nfsrv_clients > nfsrv_clienthighwater)) ||
914 			(clp->lc_expiry + NFSRV_MOULDYLEASE) < NFSD_MONOSEC ||
915 			(clp->lc_expiry < NFSD_MONOSEC &&
916 			 (nfsrv_openpluslock * 10 / 9) > NFSRV_V4STATELIMIT)) {
917 			/*
918 			 * Lease has expired several nfsrv_lease times ago:
919 			 * PLUS
920 			 *    - no state is associated with it
921 			 *    OR
922 			 *    - above high water mark for number of clients
923 			 *      (nfsrv_clienthighwater should be large enough
924 			 *       that this only occurs when clients fail to
925 			 *       use the same nfs_client_id4.id. Maybe somewhat
926 			 *       higher that the maximum number of clients that
927 			 *       will mount this server?)
928 			 * OR
929 			 * Lease has expired a very long time ago
930 			 * OR
931 			 * Lease has expired PLUS the number of opens + locks
932 			 * has exceeded 90% of capacity
933 			 *
934 			 * --> Mark for expiry. The actual expiry will be done
935 			 *     by an nfsd sometime soon.
936 			 */
937 			clp->lc_flags |= LCL_EXPIREIT;
938 			nfsrv_stablefirst.nsf_flags |=
939 			    (NFSNSF_NEEDLOCK | NFSNSF_EXPIREDCLIENT);
940 		    } else {
941 			/*
942 			 * If there are no opens, increment no open tick cnt
943 			 * If time exceeds NFSNOOPEN, mark it to be thrown away
944 			 * otherwise, if there is an open, reset no open time
945 			 * Hopefully, this will avoid excessive re-creation
946 			 * of open owners and subsequent open confirms.
947 			 */
948 			stp = LIST_FIRST(&clp->lc_open);
949 			while (stp != LIST_END(&clp->lc_open)) {
950 				nstp = LIST_NEXT(stp, ls_list);
951 				if (LIST_EMPTY(&stp->ls_open)) {
952 					stp->ls_noopens++;
953 					if (stp->ls_noopens > NFSNOOPEN ||
954 					    (nfsrv_openpluslock * 2) >
955 					    NFSRV_V4STATELIMIT)
956 						nfsrv_stablefirst.nsf_flags |=
957 							NFSNSF_NOOPENS;
958 				} else {
959 					stp->ls_noopens = 0;
960 				}
961 				stp = nstp;
962 			}
963 		    }
964 		}
965 		clp = nclp;
966 	    }
967 	}
968 	NFSUNLOCKSTATE();
969 	NFSLOCKV4ROOTMUTEX();
970 	nfsv4_relref(&nfsv4rootfs_lock);
971 	NFSUNLOCKV4ROOTMUTEX();
972 }
973 
974 /*
975  * The following set of functions free up the various data structures.
976  */
977 /*
978  * Clear out all open/lock state related to this nfsclient.
979  * Caller must hold an exclusive lock on nfsv4rootfs_lock, so that
980  * there are no other active nfsd threads.
981  */
982 APPLESTATIC void
983 nfsrv_cleanclient(struct nfsclient *clp, NFSPROC_T *p)
984 {
985 	struct nfsstate *stp, *nstp;
986 
987 	LIST_FOREACH_SAFE(stp, &clp->lc_open, ls_list, nstp)
988 		nfsrv_freeopenowner(stp, 1, p);
989 }
990 
991 /*
992  * Free a client that has been cleaned. It should also already have been
993  * removed from the lists.
994  * (Just to be safe w.r.t. newnfs_disconnect(), call this function when
995  *  softclock interrupts are enabled.)
996  */
997 APPLESTATIC void
998 nfsrv_zapclient(struct nfsclient *clp, NFSPROC_T *p)
999 {
1000 
1001 #ifdef notyet
1002 	if ((clp->lc_flags & (LCL_GSS | LCL_CALLBACKSON)) ==
1003 	     (LCL_GSS | LCL_CALLBACKSON) &&
1004 	    (clp->lc_hand.nfsh_flag & NFSG_COMPLETE) &&
1005 	    clp->lc_handlelen > 0) {
1006 		clp->lc_hand.nfsh_flag &= ~NFSG_COMPLETE;
1007 		clp->lc_hand.nfsh_flag |= NFSG_DESTROYED;
1008 		(void) nfsrv_docallback(clp, NFSV4PROC_CBNULL,
1009 			NULL, 0, NULL, NULL, NULL, p);
1010 	}
1011 #endif
1012 	newnfs_disconnect(&clp->lc_req);
1013 	NFSSOCKADDRFREE(clp->lc_req.nr_nam);
1014 	NFSFREEMUTEX(&clp->lc_req.nr_mtx);
1015 	free((caddr_t)clp, M_NFSDCLIENT);
1016 	NFSLOCKSTATE();
1017 	newnfsstats.srvclients--;
1018 	nfsrv_openpluslock--;
1019 	nfsrv_clients--;
1020 	NFSUNLOCKSTATE();
1021 }
1022 
1023 /*
1024  * Free a list of delegation state structures.
1025  * (This function will also free all nfslockfile structures that no
1026  *  longer have associated state.)
1027  */
1028 APPLESTATIC void
1029 nfsrv_freedeleglist(struct nfsstatehead *sthp)
1030 {
1031 	struct nfsstate *stp, *nstp;
1032 
1033 	LIST_FOREACH_SAFE(stp, sthp, ls_list, nstp) {
1034 		nfsrv_freedeleg(stp);
1035 	}
1036 	LIST_INIT(sthp);
1037 }
1038 
1039 /*
1040  * Free up a delegation.
1041  */
1042 static void
1043 nfsrv_freedeleg(struct nfsstate *stp)
1044 {
1045 	struct nfslockfile *lfp;
1046 
1047 	LIST_REMOVE(stp, ls_hash);
1048 	LIST_REMOVE(stp, ls_list);
1049 	LIST_REMOVE(stp, ls_file);
1050 	lfp = stp->ls_lfp;
1051 	if (LIST_EMPTY(&lfp->lf_open) &&
1052 	    LIST_EMPTY(&lfp->lf_lock) && LIST_EMPTY(&lfp->lf_deleg) &&
1053 	    LIST_EMPTY(&lfp->lf_locallock) && LIST_EMPTY(&lfp->lf_rollback) &&
1054 	    lfp->lf_usecount == 0 &&
1055 	    nfsv4_testlock(&lfp->lf_locallock_lck) == 0)
1056 		nfsrv_freenfslockfile(lfp);
1057 	FREE((caddr_t)stp, M_NFSDSTATE);
1058 	newnfsstats.srvdelegates--;
1059 	nfsrv_openpluslock--;
1060 	nfsrv_delegatecnt--;
1061 }
1062 
1063 /*
1064  * This function frees an open owner and all associated opens.
1065  */
1066 static void
1067 nfsrv_freeopenowner(struct nfsstate *stp, int cansleep, NFSPROC_T *p)
1068 {
1069 	struct nfsstate *nstp, *tstp;
1070 
1071 	LIST_REMOVE(stp, ls_list);
1072 	/*
1073 	 * Now, free all associated opens.
1074 	 */
1075 	nstp = LIST_FIRST(&stp->ls_open);
1076 	while (nstp != LIST_END(&stp->ls_open)) {
1077 		tstp = nstp;
1078 		nstp = LIST_NEXT(nstp, ls_list);
1079 		(void) nfsrv_freeopen(tstp, NULL, cansleep, p);
1080 	}
1081 	if (stp->ls_op)
1082 		nfsrvd_derefcache(stp->ls_op);
1083 	FREE((caddr_t)stp, M_NFSDSTATE);
1084 	newnfsstats.srvopenowners--;
1085 	nfsrv_openpluslock--;
1086 }
1087 
1088 /*
1089  * This function frees an open (nfsstate open structure) with all associated
1090  * lock_owners and locks. It also frees the nfslockfile structure iff there
1091  * are no other opens on the file.
1092  * Returns 1 if it free'd the nfslockfile, 0 otherwise.
1093  */
1094 static int
1095 nfsrv_freeopen(struct nfsstate *stp, vnode_t vp, int cansleep, NFSPROC_T *p)
1096 {
1097 	struct nfsstate *nstp, *tstp;
1098 	struct nfslockfile *lfp;
1099 	int ret;
1100 
1101 	LIST_REMOVE(stp, ls_hash);
1102 	LIST_REMOVE(stp, ls_list);
1103 	LIST_REMOVE(stp, ls_file);
1104 
1105 	lfp = stp->ls_lfp;
1106 	/*
1107 	 * Now, free all lockowners associated with this open.
1108 	 */
1109 	LIST_FOREACH_SAFE(tstp, &stp->ls_open, ls_list, nstp)
1110 		nfsrv_freelockowner(tstp, vp, cansleep, p);
1111 
1112 	/*
1113 	 * The nfslockfile is freed here if there are no locks
1114 	 * associated with the open.
1115 	 * If there are locks associated with the open, the
1116 	 * nfslockfile structure can be freed via nfsrv_freelockowner().
1117 	 * Acquire the state mutex to avoid races with calls to
1118 	 * nfsrv_getlockfile().
1119 	 */
1120 	if (cansleep != 0)
1121 		NFSLOCKSTATE();
1122 	if (lfp != NULL && LIST_EMPTY(&lfp->lf_open) &&
1123 	    LIST_EMPTY(&lfp->lf_deleg) && LIST_EMPTY(&lfp->lf_lock) &&
1124 	    LIST_EMPTY(&lfp->lf_locallock) && LIST_EMPTY(&lfp->lf_rollback) &&
1125 	    lfp->lf_usecount == 0 &&
1126 	    (cansleep != 0 || nfsv4_testlock(&lfp->lf_locallock_lck) == 0)) {
1127 		nfsrv_freenfslockfile(lfp);
1128 		ret = 1;
1129 	} else
1130 		ret = 0;
1131 	if (cansleep != 0)
1132 		NFSUNLOCKSTATE();
1133 	FREE((caddr_t)stp, M_NFSDSTATE);
1134 	newnfsstats.srvopens--;
1135 	nfsrv_openpluslock--;
1136 	return (ret);
1137 }
1138 
1139 /*
1140  * Frees a lockowner and all associated locks.
1141  */
1142 static void
1143 nfsrv_freelockowner(struct nfsstate *stp, vnode_t vp, int cansleep,
1144     NFSPROC_T *p)
1145 {
1146 
1147 	LIST_REMOVE(stp, ls_hash);
1148 	LIST_REMOVE(stp, ls_list);
1149 	nfsrv_freeallnfslocks(stp, vp, cansleep, p);
1150 	if (stp->ls_op)
1151 		nfsrvd_derefcache(stp->ls_op);
1152 	FREE((caddr_t)stp, M_NFSDSTATE);
1153 	newnfsstats.srvlockowners--;
1154 	nfsrv_openpluslock--;
1155 }
1156 
1157 /*
1158  * Free all the nfs locks on a lockowner.
1159  */
1160 static void
1161 nfsrv_freeallnfslocks(struct nfsstate *stp, vnode_t vp, int cansleep,
1162     NFSPROC_T *p)
1163 {
1164 	struct nfslock *lop, *nlop;
1165 	struct nfsrollback *rlp, *nrlp;
1166 	struct nfslockfile *lfp = NULL;
1167 	int gottvp = 0;
1168 	vnode_t tvp = NULL;
1169 	uint64_t first, end;
1170 
1171 	lop = LIST_FIRST(&stp->ls_lock);
1172 	while (lop != LIST_END(&stp->ls_lock)) {
1173 		nlop = LIST_NEXT(lop, lo_lckowner);
1174 		/*
1175 		 * Since all locks should be for the same file, lfp should
1176 		 * not change.
1177 		 */
1178 		if (lfp == NULL)
1179 			lfp = lop->lo_lfp;
1180 		else if (lfp != lop->lo_lfp)
1181 			panic("allnfslocks");
1182 		/*
1183 		 * If vp is NULL and cansleep != 0, a vnode must be acquired
1184 		 * from the file handle. This only occurs when called from
1185 		 * nfsrv_cleanclient().
1186 		 */
1187 		if (gottvp == 0) {
1188 			if (nfsrv_dolocallocks == 0)
1189 				tvp = NULL;
1190 			else if (vp == NULL && cansleep != 0)
1191 				tvp = nfsvno_getvp(&lfp->lf_fh);
1192 			else
1193 				tvp = vp;
1194 			gottvp = 1;
1195 		}
1196 
1197 		if (tvp != NULL) {
1198 			if (cansleep == 0)
1199 				panic("allnfs2");
1200 			first = lop->lo_first;
1201 			end = lop->lo_end;
1202 			nfsrv_freenfslock(lop);
1203 			nfsrv_localunlock(tvp, lfp, first, end, p);
1204 			LIST_FOREACH_SAFE(rlp, &lfp->lf_rollback, rlck_list,
1205 			    nrlp)
1206 				free(rlp, M_NFSDROLLBACK);
1207 			LIST_INIT(&lfp->lf_rollback);
1208 		} else
1209 			nfsrv_freenfslock(lop);
1210 		lop = nlop;
1211 	}
1212 	if (vp == NULL && tvp != NULL)
1213 		vput(tvp);
1214 }
1215 
1216 /*
1217  * Free an nfslock structure.
1218  */
1219 static void
1220 nfsrv_freenfslock(struct nfslock *lop)
1221 {
1222 
1223 	if (lop->lo_lckfile.le_prev != NULL) {
1224 		LIST_REMOVE(lop, lo_lckfile);
1225 		newnfsstats.srvlocks--;
1226 		nfsrv_openpluslock--;
1227 	}
1228 	LIST_REMOVE(lop, lo_lckowner);
1229 	FREE((caddr_t)lop, M_NFSDLOCK);
1230 }
1231 
1232 /*
1233  * This function frees an nfslockfile structure.
1234  */
1235 static void
1236 nfsrv_freenfslockfile(struct nfslockfile *lfp)
1237 {
1238 
1239 	LIST_REMOVE(lfp, lf_hash);
1240 	FREE((caddr_t)lfp, M_NFSDLOCKFILE);
1241 }
1242 
1243 /*
1244  * This function looks up an nfsstate structure via stateid.
1245  */
1246 static int
1247 nfsrv_getstate(struct nfsclient *clp, nfsv4stateid_t *stateidp, __unused u_int32_t flags,
1248     struct nfsstate **stpp)
1249 {
1250 	struct nfsstate *stp;
1251 	struct nfsstatehead *hp;
1252 	int error = 0;
1253 
1254 	*stpp = NULL;
1255 	hp = NFSSTATEHASH(clp, *stateidp);
1256 	LIST_FOREACH(stp, hp, ls_hash) {
1257 		if (!NFSBCMP(stp->ls_stateid.other, stateidp->other,
1258 			NFSX_STATEIDOTHER))
1259 			break;
1260 	}
1261 
1262 	/*
1263 	 * If no state id in list, return NFSERR_BADSTATEID.
1264 	 */
1265 	if (stp == LIST_END(hp)) {
1266 		error = NFSERR_BADSTATEID;
1267 		goto out;
1268 	}
1269 	*stpp = stp;
1270 
1271 out:
1272 	NFSEXITCODE(error);
1273 	return (error);
1274 }
1275 
1276 /*
1277  * This function gets an nfsstate structure via owner string.
1278  */
1279 static void
1280 nfsrv_getowner(struct nfsstatehead *hp, struct nfsstate *new_stp,
1281     struct nfsstate **stpp)
1282 {
1283 	struct nfsstate *stp;
1284 
1285 	*stpp = NULL;
1286 	LIST_FOREACH(stp, hp, ls_list) {
1287 		if (new_stp->ls_ownerlen == stp->ls_ownerlen &&
1288 		  !NFSBCMP(new_stp->ls_owner,stp->ls_owner,stp->ls_ownerlen)) {
1289 			*stpp = stp;
1290 			return;
1291 		}
1292 	}
1293 }
1294 
1295 /*
1296  * Lock control function called to update lock status.
1297  * Returns 0 upon success, -1 if there is no lock and the flags indicate
1298  * that one isn't to be created and an NFSERR_xxx for other errors.
1299  * The structures new_stp and new_lop are passed in as pointers that should
1300  * be set to NULL if the structure is used and shouldn't be free'd.
1301  * For the NFSLCK_TEST and NFSLCK_CHECK cases, the structures are
1302  * never used and can safely be allocated on the stack. For all other
1303  * cases, *new_stpp and *new_lopp should be malloc'd before the call,
1304  * in case they are used.
1305  */
1306 APPLESTATIC int
1307 nfsrv_lockctrl(vnode_t vp, struct nfsstate **new_stpp,
1308     struct nfslock **new_lopp, struct nfslockconflict *cfp,
1309     nfsquad_t clientid, nfsv4stateid_t *stateidp,
1310     __unused struct nfsexstuff *exp,
1311     struct nfsrv_descript *nd, NFSPROC_T *p)
1312 {
1313 	struct nfslock *lop;
1314 	struct nfsstate *new_stp = *new_stpp;
1315 	struct nfslock *new_lop = *new_lopp;
1316 	struct nfsstate *tstp, *mystp, *nstp;
1317 	int specialid = 0;
1318 	struct nfslockfile *lfp;
1319 	struct nfslock *other_lop = NULL;
1320 	struct nfsstate *stp, *lckstp = NULL;
1321 	struct nfsclient *clp = NULL;
1322 	u_int32_t bits;
1323 	int error = 0, haslock = 0, ret, reterr;
1324 	int getlckret, delegation = 0, filestruct_locked;
1325 	fhandle_t nfh;
1326 	uint64_t first, end;
1327 	uint32_t lock_flags;
1328 
1329 	if (new_stp->ls_flags & (NFSLCK_CHECK | NFSLCK_SETATTR)) {
1330 		/*
1331 		 * Note the special cases of "all 1s" or "all 0s" stateids and
1332 		 * let reads with all 1s go ahead.
1333 		 */
1334 		if (new_stp->ls_stateid.seqid == 0x0 &&
1335 		    new_stp->ls_stateid.other[0] == 0x0 &&
1336 		    new_stp->ls_stateid.other[1] == 0x0 &&
1337 		    new_stp->ls_stateid.other[2] == 0x0)
1338 			specialid = 1;
1339 		else if (new_stp->ls_stateid.seqid == 0xffffffff &&
1340 		    new_stp->ls_stateid.other[0] == 0xffffffff &&
1341 		    new_stp->ls_stateid.other[1] == 0xffffffff &&
1342 		    new_stp->ls_stateid.other[2] == 0xffffffff)
1343 			specialid = 2;
1344 	}
1345 
1346 	/*
1347 	 * Check for restart conditions (client and server).
1348 	 */
1349 	error = nfsrv_checkrestart(clientid, new_stp->ls_flags,
1350 	    &new_stp->ls_stateid, specialid);
1351 	if (error)
1352 		goto out;
1353 
1354 	/*
1355 	 * Check for state resource limit exceeded.
1356 	 */
1357 	if ((new_stp->ls_flags & NFSLCK_LOCK) &&
1358 	    nfsrv_openpluslock > NFSRV_V4STATELIMIT) {
1359 		error = NFSERR_RESOURCE;
1360 		goto out;
1361 	}
1362 
1363 	/*
1364 	 * For the lock case, get another nfslock structure,
1365 	 * just in case we need it.
1366 	 * Malloc now, before we start sifting through the linked lists,
1367 	 * in case we have to wait for memory.
1368 	 */
1369 tryagain:
1370 	if (new_stp->ls_flags & NFSLCK_LOCK)
1371 		MALLOC(other_lop, struct nfslock *, sizeof (struct nfslock),
1372 		    M_NFSDLOCK, M_WAITOK);
1373 	filestruct_locked = 0;
1374 	reterr = 0;
1375 	lfp = NULL;
1376 
1377 	/*
1378 	 * Get the lockfile structure for CFH now, so we can do a sanity
1379 	 * check against the stateid, before incrementing the seqid#, since
1380 	 * we want to return NFSERR_BADSTATEID on failure and the seqid#
1381 	 * shouldn't be incremented for this case.
1382 	 * If nfsrv_getlockfile() returns -1, it means "not found", which
1383 	 * will be handled later.
1384 	 * If we are doing Lock/LockU and local locking is enabled, sleep
1385 	 * lock the nfslockfile structure.
1386 	 */
1387 	getlckret = nfsrv_getlockfh(vp, new_stp->ls_flags, NULL, &nfh, p);
1388 	NFSLOCKSTATE();
1389 	if (getlckret == 0) {
1390 		if ((new_stp->ls_flags & (NFSLCK_LOCK | NFSLCK_UNLOCK)) != 0 &&
1391 		    nfsrv_dolocallocks != 0 && nd->nd_repstat == 0) {
1392 			getlckret = nfsrv_getlockfile(new_stp->ls_flags, NULL,
1393 			    &lfp, &nfh, 1);
1394 			if (getlckret == 0)
1395 				filestruct_locked = 1;
1396 		} else
1397 			getlckret = nfsrv_getlockfile(new_stp->ls_flags, NULL,
1398 			    &lfp, &nfh, 0);
1399 	}
1400 	if (getlckret != 0 && getlckret != -1)
1401 		reterr = getlckret;
1402 
1403 	if (filestruct_locked != 0) {
1404 		LIST_INIT(&lfp->lf_rollback);
1405 		if ((new_stp->ls_flags & NFSLCK_LOCK)) {
1406 			/*
1407 			 * For local locking, do the advisory locking now, so
1408 			 * that any conflict can be detected. A failure later
1409 			 * can be rolled back locally. If an error is returned,
1410 			 * struct nfslockfile has been unlocked and any local
1411 			 * locking rolled back.
1412 			 */
1413 			NFSUNLOCKSTATE();
1414 			reterr = nfsrv_locallock(vp, lfp,
1415 			    (new_lop->lo_flags & (NFSLCK_READ | NFSLCK_WRITE)),
1416 			    new_lop->lo_first, new_lop->lo_end, cfp, p);
1417 			NFSLOCKSTATE();
1418 		}
1419 	}
1420 
1421 	if (specialid == 0) {
1422 	    if (new_stp->ls_flags & NFSLCK_TEST) {
1423 		/*
1424 		 * RFC 3530 does not list LockT as an op that renews a
1425 		 * lease, but the concensus seems to be that it is ok
1426 		 * for a server to do so.
1427 		 */
1428 		error = nfsrv_getclient(clientid, CLOPS_RENEW, &clp,
1429 		    (nfsquad_t)((u_quad_t)0), NULL, p);
1430 
1431 		/*
1432 		 * Since NFSERR_EXPIRED, NFSERR_ADMINREVOKED are not valid
1433 		 * error returns for LockT, just go ahead and test for a lock,
1434 		 * since there are no locks for this client, but other locks
1435 		 * can conflict. (ie. same client will always be false)
1436 		 */
1437 		if (error == NFSERR_EXPIRED || error == NFSERR_ADMINREVOKED)
1438 		    error = 0;
1439 		lckstp = new_stp;
1440 	    } else {
1441 	      error = nfsrv_getclient(clientid, CLOPS_RENEW, &clp,
1442 		(nfsquad_t)((u_quad_t)0), NULL, p);
1443 	      if (error == 0)
1444 		/*
1445 		 * Look up the stateid
1446 		 */
1447 		error = nfsrv_getstate(clp, &new_stp->ls_stateid,
1448 		  new_stp->ls_flags, &stp);
1449 	      /*
1450 	       * do some sanity checks for an unconfirmed open or a
1451 	       * stateid that refers to the wrong file, for an open stateid
1452 	       */
1453 	      if (error == 0 && (stp->ls_flags & NFSLCK_OPEN) &&
1454 		  ((stp->ls_openowner->ls_flags & NFSLCK_NEEDSCONFIRM) ||
1455 		   (getlckret == 0 && stp->ls_lfp != lfp)))
1456 			error = NFSERR_BADSTATEID;
1457 	      if (error == 0 &&
1458 		  (stp->ls_flags & (NFSLCK_DELEGREAD | NFSLCK_DELEGWRITE)) &&
1459 		  getlckret == 0 && stp->ls_lfp != lfp)
1460 			error = NFSERR_BADSTATEID;
1461 
1462 	      /*
1463 	       * If the lockowner stateid doesn't refer to the same file,
1464 	       * I believe that is considered ok, since some clients will
1465 	       * only create a single lockowner and use that for all locks
1466 	       * on all files.
1467 	       * For now, log it as a diagnostic, instead of considering it
1468 	       * a BadStateid.
1469 	       */
1470 	      if (error == 0 && (stp->ls_flags &
1471 		  (NFSLCK_OPEN | NFSLCK_DELEGREAD | NFSLCK_DELEGWRITE)) == 0 &&
1472 		  getlckret == 0 && stp->ls_lfp != lfp) {
1473 #ifdef DIAGNOSTIC
1474 		  printf("Got a lock statid for different file open\n");
1475 #endif
1476 		  /*
1477 		  error = NFSERR_BADSTATEID;
1478 		  */
1479 	      }
1480 
1481 	      if (error == 0) {
1482 		    if (new_stp->ls_flags & NFSLCK_OPENTOLOCK) {
1483 			/*
1484 			 * If haslock set, we've already checked the seqid.
1485 			 */
1486 			if (!haslock) {
1487 			    if (stp->ls_flags & NFSLCK_OPEN)
1488 				error = nfsrv_checkseqid(nd, new_stp->ls_seq,
1489 				    stp->ls_openowner, new_stp->ls_op);
1490 			    else
1491 				error = NFSERR_BADSTATEID;
1492 			}
1493 			if (!error)
1494 			    nfsrv_getowner(&stp->ls_open, new_stp, &lckstp);
1495 			if (lckstp)
1496 			    /*
1497 			     * I believe this should be an error, but it
1498 			     * isn't obvious what NFSERR_xxx would be
1499 			     * appropriate, so I'll use NFSERR_INVAL for now.
1500 			     */
1501 			    error = NFSERR_INVAL;
1502 			else
1503 			    lckstp = new_stp;
1504 		    } else if (new_stp->ls_flags&(NFSLCK_LOCK|NFSLCK_UNLOCK)) {
1505 			/*
1506 			 * If haslock set, ditto above.
1507 			 */
1508 			if (!haslock) {
1509 			    if (stp->ls_flags & NFSLCK_OPEN)
1510 				error = NFSERR_BADSTATEID;
1511 			    else
1512 				error = nfsrv_checkseqid(nd, new_stp->ls_seq,
1513 				    stp, new_stp->ls_op);
1514 			}
1515 			lckstp = stp;
1516 		    } else {
1517 			lckstp = stp;
1518 		    }
1519 	      }
1520 	      /*
1521 	       * If the seqid part of the stateid isn't the same, return
1522 	       * NFSERR_OLDSTATEID for cases other than I/O Ops.
1523 	       * For I/O Ops, only return NFSERR_OLDSTATEID if
1524 	       * nfsrv_returnoldstateid is set. (The concensus on the email
1525 	       * list was that most clients would prefer to not receive
1526 	       * NFSERR_OLDSTATEID for I/O Ops, but the RFC suggests that that
1527 	       * is what will happen, so I use the nfsrv_returnoldstateid to
1528 	       * allow for either server configuration.)
1529 	       */
1530 	      if (!error && stp->ls_stateid.seqid!=new_stp->ls_stateid.seqid &&
1531 		  (!(new_stp->ls_flags & NFSLCK_CHECK) ||
1532 		   nfsrv_returnoldstateid))
1533 		    error = NFSERR_OLDSTATEID;
1534 	    }
1535 	}
1536 
1537 	/*
1538 	 * Now we can check for grace.
1539 	 */
1540 	if (!error)
1541 		error = nfsrv_checkgrace(new_stp->ls_flags);
1542 	if ((new_stp->ls_flags & NFSLCK_RECLAIM) && !error &&
1543 		nfsrv_checkstable(clp))
1544 		error = NFSERR_NOGRACE;
1545 	/*
1546 	 * If we successfully Reclaimed state, note that.
1547 	 */
1548 	if ((new_stp->ls_flags & NFSLCK_RECLAIM) && !error)
1549 		nfsrv_markstable(clp);
1550 
1551 	/*
1552 	 * At this point, either error == NFSERR_BADSTATEID or the
1553 	 * seqid# has been updated, so we can return any error.
1554 	 * If error == 0, there may be an error in:
1555 	 *    nd_repstat - Set by the calling function.
1556 	 *    reterr - Set above, if getting the nfslockfile structure
1557 	 *       or acquiring the local lock failed.
1558 	 *    (If both of these are set, nd_repstat should probably be
1559 	 *     returned, since that error was detected before this
1560 	 *     function call.)
1561 	 */
1562 	if (error != 0 || nd->nd_repstat != 0 || reterr != 0) {
1563 		if (error == 0) {
1564 			if (nd->nd_repstat != 0)
1565 				error = nd->nd_repstat;
1566 			else
1567 				error = reterr;
1568 		}
1569 		if (filestruct_locked != 0) {
1570 			/* Roll back local locks. */
1571 			NFSUNLOCKSTATE();
1572 			nfsrv_locallock_rollback(vp, lfp, p);
1573 			NFSLOCKSTATE();
1574 			nfsrv_unlocklf(lfp);
1575 		}
1576 		NFSUNLOCKSTATE();
1577 		goto out;
1578 	}
1579 
1580 	/*
1581 	 * Check the nfsrv_getlockfile return.
1582 	 * Returned -1 if no structure found.
1583 	 */
1584 	if (getlckret == -1) {
1585 		error = NFSERR_EXPIRED;
1586 		/*
1587 		 * Called from lockt, so no lock is OK.
1588 		 */
1589 		if (new_stp->ls_flags & NFSLCK_TEST) {
1590 			error = 0;
1591 		} else if (new_stp->ls_flags &
1592 		    (NFSLCK_CHECK | NFSLCK_SETATTR)) {
1593 			/*
1594 			 * Called to check for a lock, OK if the stateid is all
1595 			 * 1s or all 0s, but there should be an nfsstate
1596 			 * otherwise.
1597 			 * (ie. If there is no open, I'll assume no share
1598 			 *  deny bits.)
1599 			 */
1600 			if (specialid)
1601 				error = 0;
1602 			else
1603 				error = NFSERR_BADSTATEID;
1604 		}
1605 		NFSUNLOCKSTATE();
1606 		goto out;
1607 	}
1608 
1609 	/*
1610 	 * For NFSLCK_CHECK and NFSLCK_LOCK, test for a share conflict.
1611 	 * For NFSLCK_CHECK, allow a read if write access is granted,
1612 	 * but check for a deny. For NFSLCK_LOCK, require correct access,
1613 	 * which implies a conflicting deny can't exist.
1614 	 */
1615 	if (new_stp->ls_flags & (NFSLCK_CHECK | NFSLCK_LOCK)) {
1616 	    /*
1617 	     * Four kinds of state id:
1618 	     * - specialid (all 0s or all 1s), only for NFSLCK_CHECK
1619 	     * - stateid for an open
1620 	     * - stateid for a delegation
1621 	     * - stateid for a lock owner
1622 	     */
1623 	    if (!specialid) {
1624 		if (stp->ls_flags & (NFSLCK_DELEGREAD | NFSLCK_DELEGWRITE)) {
1625 		    delegation = 1;
1626 		    mystp = stp;
1627 		    nfsrv_delaydelegtimeout(stp);
1628 	        } else if (stp->ls_flags & NFSLCK_OPEN) {
1629 		    mystp = stp;
1630 		} else {
1631 		    mystp = stp->ls_openstp;
1632 		}
1633 		/*
1634 		 * If locking or checking, require correct access
1635 		 * bit set.
1636 		 */
1637 		if (((new_stp->ls_flags & NFSLCK_LOCK) &&
1638 		     !((new_lop->lo_flags >> NFSLCK_LOCKSHIFT) &
1639 		       mystp->ls_flags & NFSLCK_ACCESSBITS)) ||
1640 		    ((new_stp->ls_flags & (NFSLCK_CHECK|NFSLCK_READACCESS)) ==
1641 		      (NFSLCK_CHECK | NFSLCK_READACCESS) &&
1642 		     !(mystp->ls_flags & NFSLCK_READACCESS)) ||
1643 		    ((new_stp->ls_flags & (NFSLCK_CHECK|NFSLCK_WRITEACCESS)) ==
1644 		      (NFSLCK_CHECK | NFSLCK_WRITEACCESS) &&
1645 		     !(mystp->ls_flags & NFSLCK_WRITEACCESS))) {
1646 			if (filestruct_locked != 0) {
1647 				/* Roll back local locks. */
1648 				NFSUNLOCKSTATE();
1649 				nfsrv_locallock_rollback(vp, lfp, p);
1650 				NFSLOCKSTATE();
1651 				nfsrv_unlocklf(lfp);
1652 			}
1653 			NFSUNLOCKSTATE();
1654 			error = NFSERR_OPENMODE;
1655 			goto out;
1656 		}
1657 	    } else
1658 		mystp = NULL;
1659 	    if ((new_stp->ls_flags & NFSLCK_CHECK) && !delegation) {
1660 		/*
1661 		 * Check for a conflicting deny bit.
1662 		 */
1663 		LIST_FOREACH(tstp, &lfp->lf_open, ls_file) {
1664 		    if (tstp != mystp) {
1665 			bits = tstp->ls_flags;
1666 			bits >>= NFSLCK_SHIFT;
1667 			if (new_stp->ls_flags & bits & NFSLCK_ACCESSBITS) {
1668 			    ret = nfsrv_clientconflict(tstp->ls_clp, &haslock,
1669 				vp, p);
1670 			    if (ret == 1) {
1671 				/*
1672 				* nfsrv_clientconflict unlocks state
1673 				 * when it returns non-zero.
1674 				 */
1675 				lckstp = NULL;
1676 				goto tryagain;
1677 			    }
1678 			    if (ret == 0)
1679 				NFSUNLOCKSTATE();
1680 			    if (ret == 2)
1681 				error = NFSERR_PERM;
1682 			    else
1683 				error = NFSERR_OPENMODE;
1684 			    goto out;
1685 			}
1686 		    }
1687 		}
1688 
1689 		/* We're outta here */
1690 		NFSUNLOCKSTATE();
1691 		goto out;
1692 	    }
1693 	}
1694 
1695 	/*
1696 	 * For setattr, just get rid of all the Delegations for other clients.
1697 	 */
1698 	if (new_stp->ls_flags & NFSLCK_SETATTR) {
1699 		ret = nfsrv_cleandeleg(vp, lfp, clp, &haslock, p);
1700 		if (ret) {
1701 			/*
1702 			 * nfsrv_cleandeleg() unlocks state when it
1703 			 * returns non-zero.
1704 			 */
1705 			if (ret == -1) {
1706 				lckstp = NULL;
1707 				goto tryagain;
1708 			}
1709 			error = ret;
1710 			goto out;
1711 		}
1712 		if (!(new_stp->ls_flags & NFSLCK_CHECK) ||
1713 		    (LIST_EMPTY(&lfp->lf_open) && LIST_EMPTY(&lfp->lf_lock) &&
1714 		     LIST_EMPTY(&lfp->lf_deleg))) {
1715 			NFSUNLOCKSTATE();
1716 			goto out;
1717 		}
1718 	}
1719 
1720 	/*
1721 	 * Check for a conflicting delegation. If one is found, call
1722 	 * nfsrv_delegconflict() to handle it. If the v4root lock hasn't
1723 	 * been set yet, it will get the lock. Otherwise, it will recall
1724 	 * the delegation. Then, we try try again...
1725 	 * I currently believe the conflict algorithm to be:
1726 	 * For Lock Ops (Lock/LockT/LockU)
1727 	 * - there is a conflict iff a different client has a write delegation
1728 	 * For Reading (Read Op)
1729 	 * - there is a conflict iff a different client has a write delegation
1730 	 *   (the specialids are always a different client)
1731 	 * For Writing (Write/Setattr of size)
1732 	 * - there is a conflict if a different client has any delegation
1733 	 * - there is a conflict if the same client has a read delegation
1734 	 *   (I don't understand why this isn't allowed, but that seems to be
1735 	 *    the current concensus?)
1736 	 */
1737 	tstp = LIST_FIRST(&lfp->lf_deleg);
1738 	while (tstp != LIST_END(&lfp->lf_deleg)) {
1739 	    nstp = LIST_NEXT(tstp, ls_file);
1740 	    if ((((new_stp->ls_flags&(NFSLCK_LOCK|NFSLCK_UNLOCK|NFSLCK_TEST))||
1741 		 ((new_stp->ls_flags & NFSLCK_CHECK) &&
1742 		  (new_lop->lo_flags & NFSLCK_READ))) &&
1743 		  clp != tstp->ls_clp &&
1744 		 (tstp->ls_flags & NFSLCK_DELEGWRITE)) ||
1745 		 ((new_stp->ls_flags & NFSLCK_CHECK) &&
1746 		   (new_lop->lo_flags & NFSLCK_WRITE) &&
1747 		  (clp != tstp->ls_clp ||
1748 		   (tstp->ls_flags & NFSLCK_DELEGREAD)))) {
1749 		if (filestruct_locked != 0) {
1750 			/* Roll back local locks. */
1751 			NFSUNLOCKSTATE();
1752 			nfsrv_locallock_rollback(vp, lfp, p);
1753 			NFSLOCKSTATE();
1754 			nfsrv_unlocklf(lfp);
1755 		}
1756 		ret = nfsrv_delegconflict(tstp, &haslock, p, vp);
1757 		if (ret) {
1758 		    /*
1759 		     * nfsrv_delegconflict unlocks state when it
1760 		     * returns non-zero, which it always does.
1761 		     */
1762 		    if (other_lop) {
1763 			FREE((caddr_t)other_lop, M_NFSDLOCK);
1764 			other_lop = NULL;
1765 		    }
1766 		    if (ret == -1) {
1767 			lckstp = NULL;
1768 			goto tryagain;
1769 		    }
1770 		    error = ret;
1771 		    goto out;
1772 		}
1773 		/* Never gets here. */
1774 	    }
1775 	    tstp = nstp;
1776 	}
1777 
1778 	/*
1779 	 * Handle the unlock case by calling nfsrv_updatelock().
1780 	 * (Should I have done some access checking above for unlock? For now,
1781 	 *  just let it happen.)
1782 	 */
1783 	if (new_stp->ls_flags & NFSLCK_UNLOCK) {
1784 		first = new_lop->lo_first;
1785 		end = new_lop->lo_end;
1786 		nfsrv_updatelock(stp, new_lopp, &other_lop, lfp);
1787 		stateidp->seqid = ++(stp->ls_stateid.seqid);
1788 		stateidp->other[0] = stp->ls_stateid.other[0];
1789 		stateidp->other[1] = stp->ls_stateid.other[1];
1790 		stateidp->other[2] = stp->ls_stateid.other[2];
1791 		if (filestruct_locked != 0) {
1792 			NFSUNLOCKSTATE();
1793 			/* Update the local locks. */
1794 			nfsrv_localunlock(vp, lfp, first, end, p);
1795 			NFSLOCKSTATE();
1796 			nfsrv_unlocklf(lfp);
1797 		}
1798 		NFSUNLOCKSTATE();
1799 		goto out;
1800 	}
1801 
1802 	/*
1803 	 * Search for a conflicting lock. A lock conflicts if:
1804 	 * - the lock range overlaps and
1805 	 * - at least one lock is a write lock and
1806 	 * - it is not owned by the same lock owner
1807 	 */
1808 	if (!delegation) {
1809 	  LIST_FOREACH(lop, &lfp->lf_lock, lo_lckfile) {
1810 	    if (new_lop->lo_end > lop->lo_first &&
1811 		new_lop->lo_first < lop->lo_end &&
1812 		(new_lop->lo_flags == NFSLCK_WRITE ||
1813 		 lop->lo_flags == NFSLCK_WRITE) &&
1814 		lckstp != lop->lo_stp &&
1815 		(clp != lop->lo_stp->ls_clp ||
1816 		 lckstp->ls_ownerlen != lop->lo_stp->ls_ownerlen ||
1817 		 NFSBCMP(lckstp->ls_owner, lop->lo_stp->ls_owner,
1818 		    lckstp->ls_ownerlen))) {
1819 		if (other_lop) {
1820 		    FREE((caddr_t)other_lop, M_NFSDLOCK);
1821 		    other_lop = NULL;
1822 		}
1823 		ret = nfsrv_clientconflict(lop->lo_stp->ls_clp,&haslock,vp,p);
1824 		if (ret == 1) {
1825 		    if (filestruct_locked != 0) {
1826 			/* Roll back local locks. */
1827 			nfsrv_locallock_rollback(vp, lfp, p);
1828 			NFSLOCKSTATE();
1829 			nfsrv_unlocklf(lfp);
1830 			NFSUNLOCKSTATE();
1831 		    }
1832 		    /*
1833 		     * nfsrv_clientconflict() unlocks state when it
1834 		     * returns non-zero.
1835 		     */
1836 		    lckstp = NULL;
1837 		    goto tryagain;
1838 		}
1839 		/*
1840 		 * Found a conflicting lock, so record the conflict and
1841 		 * return the error.
1842 		 */
1843 		if (cfp != NULL && ret == 0) {
1844 		    cfp->cl_clientid.lval[0]=lop->lo_stp->ls_stateid.other[0];
1845 		    cfp->cl_clientid.lval[1]=lop->lo_stp->ls_stateid.other[1];
1846 		    cfp->cl_first = lop->lo_first;
1847 		    cfp->cl_end = lop->lo_end;
1848 		    cfp->cl_flags = lop->lo_flags;
1849 		    cfp->cl_ownerlen = lop->lo_stp->ls_ownerlen;
1850 		    NFSBCOPY(lop->lo_stp->ls_owner, cfp->cl_owner,
1851 			cfp->cl_ownerlen);
1852 		}
1853 		if (ret == 2)
1854 		    error = NFSERR_PERM;
1855 		else if (new_stp->ls_flags & NFSLCK_RECLAIM)
1856 		    error = NFSERR_RECLAIMCONFLICT;
1857 		else if (new_stp->ls_flags & NFSLCK_CHECK)
1858 		    error = NFSERR_LOCKED;
1859 		else
1860 		    error = NFSERR_DENIED;
1861 		if (filestruct_locked != 0 && ret == 0) {
1862 			/* Roll back local locks. */
1863 			NFSUNLOCKSTATE();
1864 			nfsrv_locallock_rollback(vp, lfp, p);
1865 			NFSLOCKSTATE();
1866 			nfsrv_unlocklf(lfp);
1867 		}
1868 		if (ret == 0)
1869 			NFSUNLOCKSTATE();
1870 		goto out;
1871 	    }
1872 	  }
1873 	}
1874 
1875 	/*
1876 	 * We only get here if there was no lock that conflicted.
1877 	 */
1878 	if (new_stp->ls_flags & (NFSLCK_TEST | NFSLCK_CHECK)) {
1879 		NFSUNLOCKSTATE();
1880 		goto out;
1881 	}
1882 
1883 	/*
1884 	 * We only get here when we are creating or modifying a lock.
1885 	 * There are two variants:
1886 	 * - exist_lock_owner where lock_owner exists
1887 	 * - open_to_lock_owner with new lock_owner
1888 	 */
1889 	first = new_lop->lo_first;
1890 	end = new_lop->lo_end;
1891 	lock_flags = new_lop->lo_flags;
1892 	if (!(new_stp->ls_flags & NFSLCK_OPENTOLOCK)) {
1893 		nfsrv_updatelock(lckstp, new_lopp, &other_lop, lfp);
1894 		stateidp->seqid = ++(lckstp->ls_stateid.seqid);
1895 		stateidp->other[0] = lckstp->ls_stateid.other[0];
1896 		stateidp->other[1] = lckstp->ls_stateid.other[1];
1897 		stateidp->other[2] = lckstp->ls_stateid.other[2];
1898 	} else {
1899 		/*
1900 		 * The new open_to_lock_owner case.
1901 		 * Link the new nfsstate into the lists.
1902 		 */
1903 		new_stp->ls_seq = new_stp->ls_opentolockseq;
1904 		nfsrvd_refcache(new_stp->ls_op);
1905 		stateidp->seqid = new_stp->ls_stateid.seqid = 1;
1906 		stateidp->other[0] = new_stp->ls_stateid.other[0] =
1907 		    clp->lc_clientid.lval[0];
1908 		stateidp->other[1] = new_stp->ls_stateid.other[1] =
1909 		    clp->lc_clientid.lval[1];
1910 		stateidp->other[2] = new_stp->ls_stateid.other[2] =
1911 		    nfsrv_nextstateindex(clp);
1912 		new_stp->ls_clp = clp;
1913 		LIST_INIT(&new_stp->ls_lock);
1914 		new_stp->ls_openstp = stp;
1915 		new_stp->ls_lfp = lfp;
1916 		nfsrv_insertlock(new_lop, (struct nfslock *)new_stp, new_stp,
1917 		    lfp);
1918 		LIST_INSERT_HEAD(NFSSTATEHASH(clp, new_stp->ls_stateid),
1919 		    new_stp, ls_hash);
1920 		LIST_INSERT_HEAD(&stp->ls_open, new_stp, ls_list);
1921 		*new_lopp = NULL;
1922 		*new_stpp = NULL;
1923 		newnfsstats.srvlockowners++;
1924 		nfsrv_openpluslock++;
1925 	}
1926 	if (filestruct_locked != 0) {
1927 		NFSUNLOCKSTATE();
1928 		nfsrv_locallock_commit(lfp, lock_flags, first, end);
1929 		NFSLOCKSTATE();
1930 		nfsrv_unlocklf(lfp);
1931 	}
1932 	NFSUNLOCKSTATE();
1933 
1934 out:
1935 	if (haslock) {
1936 		NFSLOCKV4ROOTMUTEX();
1937 		nfsv4_unlock(&nfsv4rootfs_lock, 1);
1938 		NFSUNLOCKV4ROOTMUTEX();
1939 	}
1940 	if (other_lop)
1941 		FREE((caddr_t)other_lop, M_NFSDLOCK);
1942 	NFSEXITCODE2(error, nd);
1943 	return (error);
1944 }
1945 
1946 /*
1947  * Check for state errors for Open.
1948  * repstat is passed back out as an error if more critical errors
1949  * are not detected.
1950  */
1951 APPLESTATIC int
1952 nfsrv_opencheck(nfsquad_t clientid, nfsv4stateid_t *stateidp,
1953     struct nfsstate *new_stp, vnode_t vp, struct nfsrv_descript *nd,
1954     NFSPROC_T *p, int repstat)
1955 {
1956 	struct nfsstate *stp, *nstp;
1957 	struct nfsclient *clp;
1958 	struct nfsstate *ownerstp;
1959 	struct nfslockfile *lfp, *new_lfp;
1960 	int error = 0, haslock = 0, ret, readonly = 0, getfhret = 0;
1961 
1962 	if ((new_stp->ls_flags & NFSLCK_SHAREBITS) == NFSLCK_READACCESS)
1963 		readonly = 1;
1964 	/*
1965 	 * Check for restart conditions (client and server).
1966 	 */
1967 	error = nfsrv_checkrestart(clientid, new_stp->ls_flags,
1968 		&new_stp->ls_stateid, 0);
1969 	if (error)
1970 		goto out;
1971 
1972 	/*
1973 	 * Check for state resource limit exceeded.
1974 	 * Technically this should be SMP protected, but the worst
1975 	 * case error is "out by one or two" on the count when it
1976 	 * returns NFSERR_RESOURCE and the limit is just a rather
1977 	 * arbitrary high water mark, so no harm is done.
1978 	 */
1979 	if (nfsrv_openpluslock > NFSRV_V4STATELIMIT) {
1980 		error = NFSERR_RESOURCE;
1981 		goto out;
1982 	}
1983 
1984 tryagain:
1985 	MALLOC(new_lfp, struct nfslockfile *, sizeof (struct nfslockfile),
1986 	    M_NFSDLOCKFILE, M_WAITOK);
1987 	if (vp)
1988 		getfhret = nfsrv_getlockfh(vp, new_stp->ls_flags, &new_lfp,
1989 		    NULL, p);
1990 	NFSLOCKSTATE();
1991 	/*
1992 	 * Get the nfsclient structure.
1993 	 */
1994 	error = nfsrv_getclient(clientid, CLOPS_RENEW, &clp,
1995 	    (nfsquad_t)((u_quad_t)0), NULL, p);
1996 
1997 	/*
1998 	 * Look up the open owner. See if it needs confirmation and
1999 	 * check the seq#, as required.
2000 	 */
2001 	if (!error)
2002 		nfsrv_getowner(&clp->lc_open, new_stp, &ownerstp);
2003 
2004 	if (!error && ownerstp) {
2005 		error = nfsrv_checkseqid(nd, new_stp->ls_seq, ownerstp,
2006 		    new_stp->ls_op);
2007 		/*
2008 		 * If the OpenOwner hasn't been confirmed, assume the
2009 		 * old one was a replay and this one is ok.
2010 		 * See: RFC3530 Sec. 14.2.18.
2011 		 */
2012 		if (error == NFSERR_BADSEQID &&
2013 		    (ownerstp->ls_flags & NFSLCK_NEEDSCONFIRM))
2014 			error = 0;
2015 	}
2016 
2017 	/*
2018 	 * Check for grace.
2019 	 */
2020 	if (!error)
2021 		error = nfsrv_checkgrace(new_stp->ls_flags);
2022 	if ((new_stp->ls_flags & NFSLCK_RECLAIM) && !error &&
2023 		nfsrv_checkstable(clp))
2024 		error = NFSERR_NOGRACE;
2025 
2026 	/*
2027 	 * If none of the above errors occurred, let repstat be
2028 	 * returned.
2029 	 */
2030 	if (repstat && !error)
2031 		error = repstat;
2032 	if (error) {
2033 		NFSUNLOCKSTATE();
2034 		if (haslock) {
2035 			NFSLOCKV4ROOTMUTEX();
2036 			nfsv4_unlock(&nfsv4rootfs_lock, 1);
2037 			NFSUNLOCKV4ROOTMUTEX();
2038 		}
2039 		free((caddr_t)new_lfp, M_NFSDLOCKFILE);
2040 		goto out;
2041 	}
2042 
2043 	/*
2044 	 * If vp == NULL, the file doesn't exist yet, so return ok.
2045 	 * (This always happens on the first pass, so haslock must be 0.)
2046 	 */
2047 	if (vp == NULL) {
2048 		NFSUNLOCKSTATE();
2049 		FREE((caddr_t)new_lfp, M_NFSDLOCKFILE);
2050 		goto out;
2051 	}
2052 
2053 	/*
2054 	 * Get the structure for the underlying file.
2055 	 */
2056 	if (getfhret)
2057 		error = getfhret;
2058 	else
2059 		error = nfsrv_getlockfile(new_stp->ls_flags, &new_lfp, &lfp,
2060 		    NULL, 0);
2061 	if (new_lfp)
2062 		FREE((caddr_t)new_lfp, M_NFSDLOCKFILE);
2063 	if (error) {
2064 		NFSUNLOCKSTATE();
2065 		if (haslock) {
2066 			NFSLOCKV4ROOTMUTEX();
2067 			nfsv4_unlock(&nfsv4rootfs_lock, 1);
2068 			NFSUNLOCKV4ROOTMUTEX();
2069 		}
2070 		goto out;
2071 	}
2072 
2073 	/*
2074 	 * Search for a conflicting open/share.
2075 	 */
2076 	if (new_stp->ls_flags & NFSLCK_DELEGCUR) {
2077 	    /*
2078 	     * For Delegate_Cur, search for the matching Delegation,
2079 	     * which indicates no conflict.
2080 	     * An old delegation should have been recovered by the
2081 	     * client doing a Claim_DELEGATE_Prev, so I won't let
2082 	     * it match and return NFSERR_EXPIRED. Should I let it
2083 	     * match?
2084 	     */
2085 	    LIST_FOREACH(stp, &lfp->lf_deleg, ls_file) {
2086 		if (!(stp->ls_flags & NFSLCK_OLDDELEG) &&
2087 		    stateidp->seqid == stp->ls_stateid.seqid &&
2088 		    !NFSBCMP(stateidp->other, stp->ls_stateid.other,
2089 			  NFSX_STATEIDOTHER))
2090 			break;
2091 	    }
2092 	    if (stp == LIST_END(&lfp->lf_deleg) ||
2093 		((new_stp->ls_flags & NFSLCK_WRITEACCESS) &&
2094 		 (stp->ls_flags & NFSLCK_DELEGREAD))) {
2095 		NFSUNLOCKSTATE();
2096 		if (haslock) {
2097 			NFSLOCKV4ROOTMUTEX();
2098 			nfsv4_unlock(&nfsv4rootfs_lock, 1);
2099 			NFSUNLOCKV4ROOTMUTEX();
2100 		}
2101 		error = NFSERR_EXPIRED;
2102 		goto out;
2103 	    }
2104 	}
2105 
2106 	/*
2107 	 * Check for access/deny bit conflicts. I check for the same
2108 	 * owner as well, in case the client didn't bother.
2109 	 */
2110 	LIST_FOREACH(stp, &lfp->lf_open, ls_file) {
2111 		if (!(new_stp->ls_flags & NFSLCK_DELEGCUR) &&
2112 		    (((new_stp->ls_flags & NFSLCK_ACCESSBITS) &
2113 		      ((stp->ls_flags>>NFSLCK_SHIFT) & NFSLCK_ACCESSBITS))||
2114 		     ((stp->ls_flags & NFSLCK_ACCESSBITS) &
2115 		      ((new_stp->ls_flags>>NFSLCK_SHIFT)&NFSLCK_ACCESSBITS)))){
2116 			ret = nfsrv_clientconflict(stp->ls_clp,&haslock,vp,p);
2117 			if (ret == 1) {
2118 				/*
2119 				 * nfsrv_clientconflict() unlocks
2120 				 * state when it returns non-zero.
2121 				 */
2122 				goto tryagain;
2123 			}
2124 			if (ret == 2)
2125 				error = NFSERR_PERM;
2126 			else if (new_stp->ls_flags & NFSLCK_RECLAIM)
2127 				error = NFSERR_RECLAIMCONFLICT;
2128 			else
2129 				error = NFSERR_SHAREDENIED;
2130 			if (ret == 0)
2131 				NFSUNLOCKSTATE();
2132 			if (haslock) {
2133 				NFSLOCKV4ROOTMUTEX();
2134 				nfsv4_unlock(&nfsv4rootfs_lock, 1);
2135 				NFSUNLOCKV4ROOTMUTEX();
2136 			}
2137 			goto out;
2138 		}
2139 	}
2140 
2141 	/*
2142 	 * Check for a conflicting delegation. If one is found, call
2143 	 * nfsrv_delegconflict() to handle it. If the v4root lock hasn't
2144 	 * been set yet, it will get the lock. Otherwise, it will recall
2145 	 * the delegation. Then, we try try again...
2146 	 * (If NFSLCK_DELEGCUR is set, it has a delegation, so there
2147 	 *  isn't a conflict.)
2148 	 * I currently believe the conflict algorithm to be:
2149 	 * For Open with Read Access and Deny None
2150 	 * - there is a conflict iff a different client has a write delegation
2151 	 * For Open with other Write Access or any Deny except None
2152 	 * - there is a conflict if a different client has any delegation
2153 	 * - there is a conflict if the same client has a read delegation
2154 	 *   (The current concensus is that this last case should be
2155 	 *    considered a conflict since the client with a read delegation
2156 	 *    could have done an Open with ReadAccess and WriteDeny
2157 	 *    locally and then not have checked for the WriteDeny.)
2158 	 * Don't check for a Reclaim, since that will be dealt with
2159 	 * by nfsrv_openctrl().
2160 	 */
2161 	if (!(new_stp->ls_flags &
2162 		(NFSLCK_DELEGPREV | NFSLCK_DELEGCUR | NFSLCK_RECLAIM))) {
2163 	    stp = LIST_FIRST(&lfp->lf_deleg);
2164 	    while (stp != LIST_END(&lfp->lf_deleg)) {
2165 		nstp = LIST_NEXT(stp, ls_file);
2166 		if ((readonly && stp->ls_clp != clp &&
2167 		       (stp->ls_flags & NFSLCK_DELEGWRITE)) ||
2168 		    (!readonly && (stp->ls_clp != clp ||
2169 		         (stp->ls_flags & NFSLCK_DELEGREAD)))) {
2170 			ret = nfsrv_delegconflict(stp, &haslock, p, vp);
2171 			if (ret) {
2172 			    /*
2173 			     * nfsrv_delegconflict() unlocks state
2174 			     * when it returns non-zero.
2175 			     */
2176 			    if (ret == -1)
2177 				goto tryagain;
2178 			    error = ret;
2179 			    goto out;
2180 			}
2181 		}
2182 		stp = nstp;
2183 	    }
2184 	}
2185 	NFSUNLOCKSTATE();
2186 	if (haslock) {
2187 		NFSLOCKV4ROOTMUTEX();
2188 		nfsv4_unlock(&nfsv4rootfs_lock, 1);
2189 		NFSUNLOCKV4ROOTMUTEX();
2190 	}
2191 
2192 out:
2193 	NFSEXITCODE2(error, nd);
2194 	return (error);
2195 }
2196 
2197 /*
2198  * Open control function to create/update open state for an open.
2199  */
2200 APPLESTATIC int
2201 nfsrv_openctrl(struct nfsrv_descript *nd, vnode_t vp,
2202     struct nfsstate **new_stpp, nfsquad_t clientid, nfsv4stateid_t *stateidp,
2203     nfsv4stateid_t *delegstateidp, u_int32_t *rflagsp, struct nfsexstuff *exp,
2204     NFSPROC_T *p, u_quad_t filerev)
2205 {
2206 	struct nfsstate *new_stp = *new_stpp;
2207 	struct nfsstate *stp, *nstp;
2208 	struct nfsstate *openstp = NULL, *new_open, *ownerstp, *new_deleg;
2209 	struct nfslockfile *lfp, *new_lfp;
2210 	struct nfsclient *clp;
2211 	int error = 0, haslock = 0, ret, delegate = 1, writedeleg = 1;
2212 	int readonly = 0, cbret = 1, getfhret = 0;
2213 
2214 	if ((new_stp->ls_flags & NFSLCK_SHAREBITS) == NFSLCK_READACCESS)
2215 		readonly = 1;
2216 	/*
2217 	 * Check for restart conditions (client and server).
2218 	 * (Paranoia, should have been detected by nfsrv_opencheck().)
2219 	 * If an error does show up, return NFSERR_EXPIRED, since the
2220 	 * the seqid# has already been incremented.
2221 	 */
2222 	error = nfsrv_checkrestart(clientid, new_stp->ls_flags,
2223 	    &new_stp->ls_stateid, 0);
2224 	if (error) {
2225 		printf("Nfsd: openctrl unexpected restart err=%d\n",
2226 		    error);
2227 		error = NFSERR_EXPIRED;
2228 		goto out;
2229 	}
2230 
2231 tryagain:
2232 	MALLOC(new_lfp, struct nfslockfile *, sizeof (struct nfslockfile),
2233 	    M_NFSDLOCKFILE, M_WAITOK);
2234 	MALLOC(new_open, struct nfsstate *, sizeof (struct nfsstate),
2235 	    M_NFSDSTATE, M_WAITOK);
2236 	MALLOC(new_deleg, struct nfsstate *, sizeof (struct nfsstate),
2237 	    M_NFSDSTATE, M_WAITOK);
2238 	getfhret = nfsrv_getlockfh(vp, new_stp->ls_flags, &new_lfp,
2239 	    NULL, p);
2240 	NFSLOCKSTATE();
2241 	/*
2242 	 * Get the client structure. Since the linked lists could be changed
2243 	 * by other nfsd processes if this process does a tsleep(), one of
2244 	 * two things must be done.
2245 	 * 1 - don't tsleep()
2246 	 * or
2247 	 * 2 - get the nfsv4_lock() { indicated by haslock == 1 }
2248 	 *     before using the lists, since this lock stops the other
2249 	 *     nfsd. This should only be used for rare cases, since it
2250 	 *     essentially single threads the nfsd.
2251 	 *     At this time, it is only done for cases where the stable
2252 	 *     storage file must be written prior to completion of state
2253 	 *     expiration.
2254 	 */
2255 	error = nfsrv_getclient(clientid, CLOPS_RENEW, &clp,
2256 	    (nfsquad_t)((u_quad_t)0), NULL, p);
2257 	if (!error && (clp->lc_flags & LCL_NEEDSCBNULL) &&
2258 	    clp->lc_program) {
2259 		/*
2260 		 * This happens on the first open for a client
2261 		 * that supports callbacks.
2262 		 */
2263 		NFSUNLOCKSTATE();
2264 		/*
2265 		 * Although nfsrv_docallback() will sleep, clp won't
2266 		 * go away, since they are only removed when the
2267 		 * nfsv4_lock() has blocked the nfsd threads. The
2268 		 * fields in clp can change, but having multiple
2269 		 * threads do this Null callback RPC should be
2270 		 * harmless.
2271 		 */
2272 		cbret = nfsrv_docallback(clp, NFSV4PROC_CBNULL,
2273 		    NULL, 0, NULL, NULL, NULL, p);
2274 		NFSLOCKSTATE();
2275 		clp->lc_flags &= ~LCL_NEEDSCBNULL;
2276 		if (!cbret)
2277 			clp->lc_flags |= LCL_CALLBACKSON;
2278 	}
2279 
2280 	/*
2281 	 * Look up the open owner. See if it needs confirmation and
2282 	 * check the seq#, as required.
2283 	 */
2284 	if (!error)
2285 		nfsrv_getowner(&clp->lc_open, new_stp, &ownerstp);
2286 
2287 	if (error) {
2288 		NFSUNLOCKSTATE();
2289 		printf("Nfsd: openctrl unexpected state err=%d\n",
2290 			error);
2291 		free((caddr_t)new_lfp, M_NFSDLOCKFILE);
2292 		free((caddr_t)new_open, M_NFSDSTATE);
2293 		free((caddr_t)new_deleg, M_NFSDSTATE);
2294 		if (haslock) {
2295 			NFSLOCKV4ROOTMUTEX();
2296 			nfsv4_unlock(&nfsv4rootfs_lock, 1);
2297 			NFSUNLOCKV4ROOTMUTEX();
2298 		}
2299 		error = NFSERR_EXPIRED;
2300 		goto out;
2301 	}
2302 
2303 	if (new_stp->ls_flags & NFSLCK_RECLAIM)
2304 		nfsrv_markstable(clp);
2305 
2306 	/*
2307 	 * Get the structure for the underlying file.
2308 	 */
2309 	if (getfhret)
2310 		error = getfhret;
2311 	else
2312 		error = nfsrv_getlockfile(new_stp->ls_flags, &new_lfp, &lfp,
2313 		    NULL, 0);
2314 	if (new_lfp)
2315 		FREE((caddr_t)new_lfp, M_NFSDLOCKFILE);
2316 	if (error) {
2317 		NFSUNLOCKSTATE();
2318 		printf("Nfsd openctrl unexpected getlockfile err=%d\n",
2319 		    error);
2320 		free((caddr_t)new_open, M_NFSDSTATE);
2321 		free((caddr_t)new_deleg, M_NFSDSTATE);
2322 		if (haslock) {
2323 			NFSLOCKV4ROOTMUTEX();
2324 			nfsv4_unlock(&nfsv4rootfs_lock, 1);
2325 			NFSUNLOCKV4ROOTMUTEX();
2326 		}
2327 		goto out;
2328 	}
2329 
2330 	/*
2331 	 * Search for a conflicting open/share.
2332 	 */
2333 	if (new_stp->ls_flags & NFSLCK_DELEGCUR) {
2334 	    /*
2335 	     * For Delegate_Cur, search for the matching Delegation,
2336 	     * which indicates no conflict.
2337 	     * An old delegation should have been recovered by the
2338 	     * client doing a Claim_DELEGATE_Prev, so I won't let
2339 	     * it match and return NFSERR_EXPIRED. Should I let it
2340 	     * match?
2341 	     */
2342 	    LIST_FOREACH(stp, &lfp->lf_deleg, ls_file) {
2343 		if (!(stp->ls_flags & NFSLCK_OLDDELEG) &&
2344 		    stateidp->seqid == stp->ls_stateid.seqid &&
2345 		    !NFSBCMP(stateidp->other, stp->ls_stateid.other,
2346 			NFSX_STATEIDOTHER))
2347 			break;
2348 	    }
2349 	    if (stp == LIST_END(&lfp->lf_deleg) ||
2350 		((new_stp->ls_flags & NFSLCK_WRITEACCESS) &&
2351 		 (stp->ls_flags & NFSLCK_DELEGREAD))) {
2352 		NFSUNLOCKSTATE();
2353 		printf("Nfsd openctrl unexpected expiry\n");
2354 		free((caddr_t)new_open, M_NFSDSTATE);
2355 		free((caddr_t)new_deleg, M_NFSDSTATE);
2356 		if (haslock) {
2357 			NFSLOCKV4ROOTMUTEX();
2358 			nfsv4_unlock(&nfsv4rootfs_lock, 1);
2359 			NFSUNLOCKV4ROOTMUTEX();
2360 		}
2361 		error = NFSERR_EXPIRED;
2362 		goto out;
2363 	    }
2364 
2365 	    /*
2366 	     * Don't issue a Delegation, since one already exists and
2367 	     * delay delegation timeout, as required.
2368 	     */
2369 	    delegate = 0;
2370 	    nfsrv_delaydelegtimeout(stp);
2371 	}
2372 
2373 	/*
2374 	 * Check for access/deny bit conflicts. I also check for the
2375 	 * same owner, since the client might not have bothered to check.
2376 	 * Also, note an open for the same file and owner, if found,
2377 	 * which is all we do here for Delegate_Cur, since conflict
2378 	 * checking is already done.
2379 	 */
2380 	LIST_FOREACH(stp, &lfp->lf_open, ls_file) {
2381 		if (ownerstp && stp->ls_openowner == ownerstp)
2382 			openstp = stp;
2383 		if (!(new_stp->ls_flags & NFSLCK_DELEGCUR)) {
2384 		    /*
2385 		     * If another client has the file open, the only
2386 		     * delegation that can be issued is a Read delegation
2387 		     * and only if it is a Read open with Deny none.
2388 		     */
2389 		    if (clp != stp->ls_clp) {
2390 			if ((stp->ls_flags & NFSLCK_SHAREBITS) ==
2391 			    NFSLCK_READACCESS)
2392 			    writedeleg = 0;
2393 			else
2394 			    delegate = 0;
2395 		    }
2396 		    if(((new_stp->ls_flags & NFSLCK_ACCESSBITS) &
2397 		        ((stp->ls_flags>>NFSLCK_SHIFT) & NFSLCK_ACCESSBITS))||
2398 		       ((stp->ls_flags & NFSLCK_ACCESSBITS) &
2399 		        ((new_stp->ls_flags>>NFSLCK_SHIFT)&NFSLCK_ACCESSBITS))){
2400 			ret = nfsrv_clientconflict(stp->ls_clp,&haslock,vp,p);
2401 			if (ret == 1) {
2402 				/*
2403 				 * nfsrv_clientconflict() unlocks state
2404 				 * when it returns non-zero.
2405 				 */
2406 				free((caddr_t)new_open, M_NFSDSTATE);
2407 				free((caddr_t)new_deleg, M_NFSDSTATE);
2408 				openstp = NULL;
2409 				goto tryagain;
2410 			}
2411 			if (ret == 2)
2412 				error = NFSERR_PERM;
2413 			else if (new_stp->ls_flags & NFSLCK_RECLAIM)
2414 				error = NFSERR_RECLAIMCONFLICT;
2415 			else
2416 				error = NFSERR_SHAREDENIED;
2417 			if (ret == 0)
2418 				NFSUNLOCKSTATE();
2419 			if (haslock) {
2420 				NFSLOCKV4ROOTMUTEX();
2421 				nfsv4_unlock(&nfsv4rootfs_lock, 1);
2422 				NFSUNLOCKV4ROOTMUTEX();
2423 			}
2424 			free((caddr_t)new_open, M_NFSDSTATE);
2425 			free((caddr_t)new_deleg, M_NFSDSTATE);
2426 			printf("nfsd openctrl unexpected client cnfl\n");
2427 			goto out;
2428 		    }
2429 		}
2430 	}
2431 
2432 	/*
2433 	 * Check for a conflicting delegation. If one is found, call
2434 	 * nfsrv_delegconflict() to handle it. If the v4root lock hasn't
2435 	 * been set yet, it will get the lock. Otherwise, it will recall
2436 	 * the delegation. Then, we try try again...
2437 	 * (If NFSLCK_DELEGCUR is set, it has a delegation, so there
2438 	 *  isn't a conflict.)
2439 	 * I currently believe the conflict algorithm to be:
2440 	 * For Open with Read Access and Deny None
2441 	 * - there is a conflict iff a different client has a write delegation
2442 	 * For Open with other Write Access or any Deny except None
2443 	 * - there is a conflict if a different client has any delegation
2444 	 * - there is a conflict if the same client has a read delegation
2445 	 *   (The current concensus is that this last case should be
2446 	 *    considered a conflict since the client with a read delegation
2447 	 *    could have done an Open with ReadAccess and WriteDeny
2448 	 *    locally and then not have checked for the WriteDeny.)
2449 	 */
2450 	if (!(new_stp->ls_flags & (NFSLCK_DELEGPREV | NFSLCK_DELEGCUR))) {
2451 	    stp = LIST_FIRST(&lfp->lf_deleg);
2452 	    while (stp != LIST_END(&lfp->lf_deleg)) {
2453 		nstp = LIST_NEXT(stp, ls_file);
2454 		if (stp->ls_clp != clp && (stp->ls_flags & NFSLCK_DELEGREAD))
2455 			writedeleg = 0;
2456 		else
2457 			delegate = 0;
2458 		if ((readonly && stp->ls_clp != clp &&
2459 		       (stp->ls_flags & NFSLCK_DELEGWRITE)) ||
2460 		    (!readonly && (stp->ls_clp != clp ||
2461 		         (stp->ls_flags & NFSLCK_DELEGREAD)))) {
2462 		    if (new_stp->ls_flags & NFSLCK_RECLAIM) {
2463 			delegate = 2;
2464 		    } else {
2465 			ret = nfsrv_delegconflict(stp, &haslock, p, vp);
2466 			if (ret) {
2467 			    /*
2468 			     * nfsrv_delegconflict() unlocks state
2469 			     * when it returns non-zero.
2470 			     */
2471 			    printf("Nfsd openctrl unexpected deleg cnfl\n");
2472 			    free((caddr_t)new_open, M_NFSDSTATE);
2473 			    free((caddr_t)new_deleg, M_NFSDSTATE);
2474 			    if (ret == -1) {
2475 				openstp = NULL;
2476 				goto tryagain;
2477 			    }
2478 			    error = ret;
2479 			    goto out;
2480 			}
2481 		    }
2482 		}
2483 		stp = nstp;
2484 	    }
2485 	}
2486 
2487 	/*
2488 	 * We only get here if there was no open that conflicted.
2489 	 * If an open for the owner exists, or in the access/deny bits.
2490 	 * Otherwise it is a new open. If the open_owner hasn't been
2491 	 * confirmed, replace the open with the new one needing confirmation,
2492 	 * otherwise add the open.
2493 	 */
2494 	if (new_stp->ls_flags & NFSLCK_DELEGPREV) {
2495 	    /*
2496 	     * Handle NFSLCK_DELEGPREV by searching the old delegations for
2497 	     * a match. If found, just move the old delegation to the current
2498 	     * delegation list and issue open. If not found, return
2499 	     * NFSERR_EXPIRED.
2500 	     */
2501 	    LIST_FOREACH(stp, &clp->lc_olddeleg, ls_list) {
2502 		if (stp->ls_lfp == lfp) {
2503 		    /* Found it */
2504 		    if (stp->ls_clp != clp)
2505 			panic("olddeleg clp");
2506 		    LIST_REMOVE(stp, ls_list);
2507 		    LIST_REMOVE(stp, ls_hash);
2508 		    stp->ls_flags &= ~NFSLCK_OLDDELEG;
2509 		    stp->ls_stateid.seqid = delegstateidp->seqid = 0;
2510 		    stp->ls_stateid.other[0] = delegstateidp->other[0] =
2511 			clp->lc_clientid.lval[0];
2512 		    stp->ls_stateid.other[1] = delegstateidp->other[1] =
2513 			clp->lc_clientid.lval[1];
2514 		    stp->ls_stateid.other[2] = delegstateidp->other[2] =
2515 			nfsrv_nextstateindex(clp);
2516 		    stp->ls_compref = nd->nd_compref;
2517 		    LIST_INSERT_HEAD(&clp->lc_deleg, stp, ls_list);
2518 		    LIST_INSERT_HEAD(NFSSTATEHASH(clp,
2519 			stp->ls_stateid), stp, ls_hash);
2520 		    if (stp->ls_flags & NFSLCK_DELEGWRITE)
2521 			*rflagsp |= NFSV4OPEN_WRITEDELEGATE;
2522 		    else
2523 			*rflagsp |= NFSV4OPEN_READDELEGATE;
2524 		    clp->lc_delegtime = NFSD_MONOSEC +
2525 			nfsrv_lease + NFSRV_LEASEDELTA;
2526 
2527 		    /*
2528 		     * Now, do the associated open.
2529 		     */
2530 		    new_open->ls_stateid.seqid = 0;
2531 		    new_open->ls_stateid.other[0] = clp->lc_clientid.lval[0];
2532 		    new_open->ls_stateid.other[1] = clp->lc_clientid.lval[1];
2533 		    new_open->ls_stateid.other[2] = nfsrv_nextstateindex(clp);
2534 		    new_open->ls_flags = (new_stp->ls_flags&NFSLCK_DENYBITS)|
2535 			NFSLCK_OPEN;
2536 		    if (stp->ls_flags & NFSLCK_DELEGWRITE)
2537 			new_open->ls_flags |= (NFSLCK_READACCESS |
2538 			    NFSLCK_WRITEACCESS);
2539 		    else
2540 			new_open->ls_flags |= NFSLCK_READACCESS;
2541 		    new_open->ls_uid = new_stp->ls_uid;
2542 		    new_open->ls_lfp = lfp;
2543 		    new_open->ls_clp = clp;
2544 		    LIST_INIT(&new_open->ls_open);
2545 		    LIST_INSERT_HEAD(&lfp->lf_open, new_open, ls_file);
2546 		    LIST_INSERT_HEAD(NFSSTATEHASH(clp, new_open->ls_stateid),
2547 			new_open, ls_hash);
2548 		    /*
2549 		     * and handle the open owner
2550 		     */
2551 		    if (ownerstp) {
2552 			new_open->ls_openowner = ownerstp;
2553 			LIST_INSERT_HEAD(&ownerstp->ls_open,new_open,ls_list);
2554 		    } else {
2555 			new_open->ls_openowner = new_stp;
2556 			new_stp->ls_flags = 0;
2557 			nfsrvd_refcache(new_stp->ls_op);
2558 			new_stp->ls_noopens = 0;
2559 			LIST_INIT(&new_stp->ls_open);
2560 			LIST_INSERT_HEAD(&new_stp->ls_open, new_open, ls_list);
2561 			LIST_INSERT_HEAD(&clp->lc_open, new_stp, ls_list);
2562 			*new_stpp = NULL;
2563 			newnfsstats.srvopenowners++;
2564 			nfsrv_openpluslock++;
2565 		    }
2566 		    openstp = new_open;
2567 		    new_open = NULL;
2568 		    newnfsstats.srvopens++;
2569 		    nfsrv_openpluslock++;
2570 		    break;
2571 		}
2572 	    }
2573 	    if (stp == LIST_END(&clp->lc_olddeleg))
2574 		error = NFSERR_EXPIRED;
2575 	} else if (new_stp->ls_flags & (NFSLCK_DELEGREAD | NFSLCK_DELEGWRITE)) {
2576 	    /*
2577 	     * Scan to see that no delegation for this client and file
2578 	     * doesn't already exist.
2579 	     * There also shouldn't yet be an Open for this file and
2580 	     * openowner.
2581 	     */
2582 	    LIST_FOREACH(stp, &lfp->lf_deleg, ls_file) {
2583 		if (stp->ls_clp == clp)
2584 		    break;
2585 	    }
2586 	    if (stp == LIST_END(&lfp->lf_deleg) && openstp == NULL) {
2587 		/*
2588 		 * This is the Claim_Previous case with a delegation
2589 		 * type != Delegate_None.
2590 		 */
2591 		/*
2592 		 * First, add the delegation. (Although we must issue the
2593 		 * delegation, we can also ask for an immediate return.)
2594 		 */
2595 		new_deleg->ls_stateid.seqid = delegstateidp->seqid = 0;
2596 		new_deleg->ls_stateid.other[0] = delegstateidp->other[0] =
2597 		    clp->lc_clientid.lval[0];
2598 		new_deleg->ls_stateid.other[1] = delegstateidp->other[1] =
2599 		    clp->lc_clientid.lval[1];
2600 		new_deleg->ls_stateid.other[2] = delegstateidp->other[2] =
2601 		    nfsrv_nextstateindex(clp);
2602 		if (new_stp->ls_flags & NFSLCK_DELEGWRITE) {
2603 		    new_deleg->ls_flags = (NFSLCK_DELEGWRITE |
2604 			NFSLCK_READACCESS | NFSLCK_WRITEACCESS);
2605 		    *rflagsp |= NFSV4OPEN_WRITEDELEGATE;
2606 		} else {
2607 		    new_deleg->ls_flags = (NFSLCK_DELEGREAD |
2608 			NFSLCK_READACCESS);
2609 		    *rflagsp |= NFSV4OPEN_READDELEGATE;
2610 		}
2611 		new_deleg->ls_uid = new_stp->ls_uid;
2612 		new_deleg->ls_lfp = lfp;
2613 		new_deleg->ls_clp = clp;
2614 		new_deleg->ls_filerev = filerev;
2615 		new_deleg->ls_compref = nd->nd_compref;
2616 		LIST_INSERT_HEAD(&lfp->lf_deleg, new_deleg, ls_file);
2617 		LIST_INSERT_HEAD(NFSSTATEHASH(clp,
2618 		    new_deleg->ls_stateid), new_deleg, ls_hash);
2619 		LIST_INSERT_HEAD(&clp->lc_deleg, new_deleg, ls_list);
2620 		new_deleg = NULL;
2621 		if (delegate == 2 || nfsrv_issuedelegs == 0 ||
2622 		    (clp->lc_flags & (LCL_CALLBACKSON | LCL_CBDOWN)) !=
2623 		     LCL_CALLBACKSON ||
2624 		    NFSRV_V4DELEGLIMIT(nfsrv_delegatecnt) ||
2625 		    !NFSVNO_DELEGOK(vp))
2626 		    *rflagsp |= NFSV4OPEN_RECALL;
2627 		newnfsstats.srvdelegates++;
2628 		nfsrv_openpluslock++;
2629 		nfsrv_delegatecnt++;
2630 
2631 		/*
2632 		 * Now, do the associated open.
2633 		 */
2634 		new_open->ls_stateid.seqid = 0;
2635 		new_open->ls_stateid.other[0] = clp->lc_clientid.lval[0];
2636 		new_open->ls_stateid.other[1] = clp->lc_clientid.lval[1];
2637 		new_open->ls_stateid.other[2] = nfsrv_nextstateindex(clp);
2638 		new_open->ls_flags = (new_stp->ls_flags & NFSLCK_DENYBITS) |
2639 		    NFSLCK_OPEN;
2640 		if (new_stp->ls_flags & NFSLCK_DELEGWRITE)
2641 			new_open->ls_flags |= (NFSLCK_READACCESS |
2642 			    NFSLCK_WRITEACCESS);
2643 		else
2644 			new_open->ls_flags |= NFSLCK_READACCESS;
2645 		new_open->ls_uid = new_stp->ls_uid;
2646 		new_open->ls_lfp = lfp;
2647 		new_open->ls_clp = clp;
2648 		LIST_INIT(&new_open->ls_open);
2649 		LIST_INSERT_HEAD(&lfp->lf_open, new_open, ls_file);
2650 		LIST_INSERT_HEAD(NFSSTATEHASH(clp, new_open->ls_stateid),
2651 		   new_open, ls_hash);
2652 		/*
2653 		 * and handle the open owner
2654 		 */
2655 		if (ownerstp) {
2656 		    new_open->ls_openowner = ownerstp;
2657 		    LIST_INSERT_HEAD(&ownerstp->ls_open, new_open, ls_list);
2658 		} else {
2659 		    new_open->ls_openowner = new_stp;
2660 		    new_stp->ls_flags = 0;
2661 		    nfsrvd_refcache(new_stp->ls_op);
2662 		    new_stp->ls_noopens = 0;
2663 		    LIST_INIT(&new_stp->ls_open);
2664 		    LIST_INSERT_HEAD(&new_stp->ls_open, new_open, ls_list);
2665 		    LIST_INSERT_HEAD(&clp->lc_open, new_stp, ls_list);
2666 		    *new_stpp = NULL;
2667 		    newnfsstats.srvopenowners++;
2668 		    nfsrv_openpluslock++;
2669 		}
2670 		openstp = new_open;
2671 		new_open = NULL;
2672 		newnfsstats.srvopens++;
2673 		nfsrv_openpluslock++;
2674 	    } else {
2675 		error = NFSERR_RECLAIMCONFLICT;
2676 	    }
2677 	} else if (ownerstp) {
2678 		if (ownerstp->ls_flags & NFSLCK_NEEDSCONFIRM) {
2679 		    /* Replace the open */
2680 		    if (ownerstp->ls_op)
2681 			nfsrvd_derefcache(ownerstp->ls_op);
2682 		    ownerstp->ls_op = new_stp->ls_op;
2683 		    nfsrvd_refcache(ownerstp->ls_op);
2684 		    ownerstp->ls_seq = new_stp->ls_seq;
2685 		    *rflagsp |= NFSV4OPEN_RESULTCONFIRM;
2686 		    stp = LIST_FIRST(&ownerstp->ls_open);
2687 		    stp->ls_flags = (new_stp->ls_flags & NFSLCK_SHAREBITS) |
2688 			NFSLCK_OPEN;
2689 		    stp->ls_stateid.seqid = 0;
2690 		    stp->ls_uid = new_stp->ls_uid;
2691 		    if (lfp != stp->ls_lfp) {
2692 			LIST_REMOVE(stp, ls_file);
2693 			LIST_INSERT_HEAD(&lfp->lf_open, stp, ls_file);
2694 			stp->ls_lfp = lfp;
2695 		    }
2696 		    openstp = stp;
2697 		} else if (openstp) {
2698 		    openstp->ls_flags |= (new_stp->ls_flags & NFSLCK_SHAREBITS);
2699 		    openstp->ls_stateid.seqid++;
2700 
2701 		    /*
2702 		     * This is where we can choose to issue a delegation.
2703 		     */
2704 		    if (delegate && nfsrv_issuedelegs &&
2705 			writedeleg && !NFSVNO_EXRDONLY(exp) &&
2706 			(nfsrv_writedelegifpos || !readonly) &&
2707 			(clp->lc_flags & (LCL_CALLBACKSON | LCL_CBDOWN)) ==
2708 			 LCL_CALLBACKSON &&
2709 			!NFSRV_V4DELEGLIMIT(nfsrv_delegatecnt) &&
2710 			NFSVNO_DELEGOK(vp)) {
2711 			new_deleg->ls_stateid.seqid = delegstateidp->seqid = 0;
2712 			new_deleg->ls_stateid.other[0] = delegstateidp->other[0]
2713 			    = clp->lc_clientid.lval[0];
2714 			new_deleg->ls_stateid.other[1] = delegstateidp->other[1]
2715 			    = clp->lc_clientid.lval[1];
2716 			new_deleg->ls_stateid.other[2] = delegstateidp->other[2]
2717 			    = nfsrv_nextstateindex(clp);
2718 			new_deleg->ls_flags = (NFSLCK_DELEGWRITE |
2719 			    NFSLCK_READACCESS | NFSLCK_WRITEACCESS);
2720 			*rflagsp |= NFSV4OPEN_WRITEDELEGATE;
2721 			new_deleg->ls_uid = new_stp->ls_uid;
2722 			new_deleg->ls_lfp = lfp;
2723 			new_deleg->ls_clp = clp;
2724 			new_deleg->ls_filerev = filerev;
2725 			new_deleg->ls_compref = nd->nd_compref;
2726 			LIST_INSERT_HEAD(&lfp->lf_deleg, new_deleg, ls_file);
2727 			LIST_INSERT_HEAD(NFSSTATEHASH(clp,
2728 			    new_deleg->ls_stateid), new_deleg, ls_hash);
2729 			LIST_INSERT_HEAD(&clp->lc_deleg, new_deleg, ls_list);
2730 			new_deleg = NULL;
2731 			newnfsstats.srvdelegates++;
2732 			nfsrv_openpluslock++;
2733 			nfsrv_delegatecnt++;
2734 		    }
2735 		} else {
2736 		    new_open->ls_stateid.seqid = 0;
2737 		    new_open->ls_stateid.other[0] = clp->lc_clientid.lval[0];
2738 		    new_open->ls_stateid.other[1] = clp->lc_clientid.lval[1];
2739 		    new_open->ls_stateid.other[2] = nfsrv_nextstateindex(clp);
2740 		    new_open->ls_flags = (new_stp->ls_flags & NFSLCK_SHAREBITS)|
2741 			NFSLCK_OPEN;
2742 		    new_open->ls_uid = new_stp->ls_uid;
2743 		    new_open->ls_openowner = ownerstp;
2744 		    new_open->ls_lfp = lfp;
2745 		    new_open->ls_clp = clp;
2746 		    LIST_INIT(&new_open->ls_open);
2747 		    LIST_INSERT_HEAD(&lfp->lf_open, new_open, ls_file);
2748 		    LIST_INSERT_HEAD(&ownerstp->ls_open, new_open, ls_list);
2749 		    LIST_INSERT_HEAD(NFSSTATEHASH(clp, new_open->ls_stateid),
2750 			new_open, ls_hash);
2751 		    openstp = new_open;
2752 		    new_open = NULL;
2753 		    newnfsstats.srvopens++;
2754 		    nfsrv_openpluslock++;
2755 
2756 		    /*
2757 		     * This is where we can choose to issue a delegation.
2758 		     */
2759 		    if (delegate && nfsrv_issuedelegs &&
2760 			(writedeleg || readonly) &&
2761 			(clp->lc_flags & (LCL_CALLBACKSON | LCL_CBDOWN)) ==
2762 			 LCL_CALLBACKSON &&
2763 			!NFSRV_V4DELEGLIMIT(nfsrv_delegatecnt) &&
2764 			NFSVNO_DELEGOK(vp)) {
2765 			new_deleg->ls_stateid.seqid = delegstateidp->seqid = 0;
2766 			new_deleg->ls_stateid.other[0] = delegstateidp->other[0]
2767 			    = clp->lc_clientid.lval[0];
2768 			new_deleg->ls_stateid.other[1] = delegstateidp->other[1]
2769 			    = clp->lc_clientid.lval[1];
2770 			new_deleg->ls_stateid.other[2] = delegstateidp->other[2]
2771 			    = nfsrv_nextstateindex(clp);
2772 			if (writedeleg && !NFSVNO_EXRDONLY(exp) &&
2773 			    (nfsrv_writedelegifpos || !readonly)) {
2774 			    new_deleg->ls_flags = (NFSLCK_DELEGWRITE |
2775 				NFSLCK_READACCESS | NFSLCK_WRITEACCESS);
2776 			    *rflagsp |= NFSV4OPEN_WRITEDELEGATE;
2777 			} else {
2778 			    new_deleg->ls_flags = (NFSLCK_DELEGREAD |
2779 				NFSLCK_READACCESS);
2780 			    *rflagsp |= NFSV4OPEN_READDELEGATE;
2781 			}
2782 			new_deleg->ls_uid = new_stp->ls_uid;
2783 			new_deleg->ls_lfp = lfp;
2784 			new_deleg->ls_clp = clp;
2785 			new_deleg->ls_filerev = filerev;
2786 			new_deleg->ls_compref = nd->nd_compref;
2787 			LIST_INSERT_HEAD(&lfp->lf_deleg, new_deleg, ls_file);
2788 			LIST_INSERT_HEAD(NFSSTATEHASH(clp,
2789 			    new_deleg->ls_stateid), new_deleg, ls_hash);
2790 			LIST_INSERT_HEAD(&clp->lc_deleg, new_deleg, ls_list);
2791 			new_deleg = NULL;
2792 			newnfsstats.srvdelegates++;
2793 			nfsrv_openpluslock++;
2794 			nfsrv_delegatecnt++;
2795 		    }
2796 		}
2797 	} else {
2798 		/*
2799 		 * New owner case. Start the open_owner sequence with a
2800 		 * Needs confirmation (unless a reclaim) and hang the
2801 		 * new open off it.
2802 		 */
2803 		new_open->ls_stateid.seqid = 0;
2804 		new_open->ls_stateid.other[0] = clp->lc_clientid.lval[0];
2805 		new_open->ls_stateid.other[1] = clp->lc_clientid.lval[1];
2806 		new_open->ls_stateid.other[2] = nfsrv_nextstateindex(clp);
2807 		new_open->ls_flags = (new_stp->ls_flags & NFSLCK_SHAREBITS) |
2808 		    NFSLCK_OPEN;
2809 		new_open->ls_uid = new_stp->ls_uid;
2810 		LIST_INIT(&new_open->ls_open);
2811 		new_open->ls_openowner = new_stp;
2812 		new_open->ls_lfp = lfp;
2813 		new_open->ls_clp = clp;
2814 		LIST_INSERT_HEAD(&lfp->lf_open, new_open, ls_file);
2815 		if (new_stp->ls_flags & NFSLCK_RECLAIM) {
2816 			new_stp->ls_flags = 0;
2817 		} else {
2818 			*rflagsp |= NFSV4OPEN_RESULTCONFIRM;
2819 			new_stp->ls_flags = NFSLCK_NEEDSCONFIRM;
2820 		}
2821 		nfsrvd_refcache(new_stp->ls_op);
2822 		new_stp->ls_noopens = 0;
2823 		LIST_INIT(&new_stp->ls_open);
2824 		LIST_INSERT_HEAD(&new_stp->ls_open, new_open, ls_list);
2825 		LIST_INSERT_HEAD(&clp->lc_open, new_stp, ls_list);
2826 		LIST_INSERT_HEAD(NFSSTATEHASH(clp, new_open->ls_stateid),
2827 		    new_open, ls_hash);
2828 		openstp = new_open;
2829 		new_open = NULL;
2830 		*new_stpp = NULL;
2831 		newnfsstats.srvopens++;
2832 		nfsrv_openpluslock++;
2833 		newnfsstats.srvopenowners++;
2834 		nfsrv_openpluslock++;
2835 	}
2836 	if (!error) {
2837 		stateidp->seqid = openstp->ls_stateid.seqid;
2838 		stateidp->other[0] = openstp->ls_stateid.other[0];
2839 		stateidp->other[1] = openstp->ls_stateid.other[1];
2840 		stateidp->other[2] = openstp->ls_stateid.other[2];
2841 	}
2842 	NFSUNLOCKSTATE();
2843 	if (haslock) {
2844 		NFSLOCKV4ROOTMUTEX();
2845 		nfsv4_unlock(&nfsv4rootfs_lock, 1);
2846 		NFSUNLOCKV4ROOTMUTEX();
2847 	}
2848 	if (new_open)
2849 		FREE((caddr_t)new_open, M_NFSDSTATE);
2850 	if (new_deleg)
2851 		FREE((caddr_t)new_deleg, M_NFSDSTATE);
2852 
2853 out:
2854 	NFSEXITCODE2(error, nd);
2855 	return (error);
2856 }
2857 
2858 /*
2859  * Open update. Does the confirm, downgrade and close.
2860  */
2861 APPLESTATIC int
2862 nfsrv_openupdate(vnode_t vp, struct nfsstate *new_stp, nfsquad_t clientid,
2863     nfsv4stateid_t *stateidp, struct nfsrv_descript *nd, NFSPROC_T *p)
2864 {
2865 	struct nfsstate *stp, *ownerstp;
2866 	struct nfsclient *clp;
2867 	struct nfslockfile *lfp;
2868 	u_int32_t bits;
2869 	int error = 0, gotstate = 0, len = 0;
2870 	u_char client[NFSV4_OPAQUELIMIT];
2871 
2872 	/*
2873 	 * Check for restart conditions (client and server).
2874 	 */
2875 	error = nfsrv_checkrestart(clientid, new_stp->ls_flags,
2876 	    &new_stp->ls_stateid, 0);
2877 	if (error)
2878 		goto out;
2879 
2880 	NFSLOCKSTATE();
2881 	/*
2882 	 * Get the open structure via clientid and stateid.
2883 	 */
2884 	error = nfsrv_getclient(clientid, CLOPS_RENEW, &clp,
2885 	    (nfsquad_t)((u_quad_t)0), NULL, p);
2886 	if (!error)
2887 		error = nfsrv_getstate(clp, &new_stp->ls_stateid,
2888 		    new_stp->ls_flags, &stp);
2889 
2890 	/*
2891 	 * Sanity check the open.
2892 	 */
2893 	if (!error && (!(stp->ls_flags & NFSLCK_OPEN) ||
2894 		(!(new_stp->ls_flags & NFSLCK_CONFIRM) &&
2895 		 (stp->ls_openowner->ls_flags & NFSLCK_NEEDSCONFIRM)) ||
2896 		((new_stp->ls_flags & NFSLCK_CONFIRM) &&
2897 		 (!(stp->ls_openowner->ls_flags & NFSLCK_NEEDSCONFIRM)))))
2898 		error = NFSERR_BADSTATEID;
2899 
2900 	if (!error)
2901 		error = nfsrv_checkseqid(nd, new_stp->ls_seq,
2902 		    stp->ls_openowner, new_stp->ls_op);
2903 	if (!error && stp->ls_stateid.seqid != new_stp->ls_stateid.seqid &&
2904 	    !(new_stp->ls_flags & NFSLCK_CONFIRM))
2905 		error = NFSERR_OLDSTATEID;
2906 	if (!error && vnode_vtype(vp) != VREG) {
2907 		if (vnode_vtype(vp) == VDIR)
2908 			error = NFSERR_ISDIR;
2909 		else
2910 			error = NFSERR_INVAL;
2911 	}
2912 
2913 	if (error) {
2914 		/*
2915 		 * If a client tries to confirm an Open with a bad
2916 		 * seqid# and there are no byte range locks or other Opens
2917 		 * on the openowner, just throw it away, so the next use of the
2918 		 * openowner will start a fresh seq#.
2919 		 */
2920 		if (error == NFSERR_BADSEQID &&
2921 		    (new_stp->ls_flags & NFSLCK_CONFIRM) &&
2922 		    nfsrv_nootherstate(stp))
2923 			nfsrv_freeopenowner(stp->ls_openowner, 0, p);
2924 		NFSUNLOCKSTATE();
2925 		goto out;
2926 	}
2927 
2928 	/*
2929 	 * Set the return stateid.
2930 	 */
2931 	stateidp->seqid = stp->ls_stateid.seqid + 1;
2932 	stateidp->other[0] = stp->ls_stateid.other[0];
2933 	stateidp->other[1] = stp->ls_stateid.other[1];
2934 	stateidp->other[2] = stp->ls_stateid.other[2];
2935 	/*
2936 	 * Now, handle the three cases.
2937 	 */
2938 	if (new_stp->ls_flags & NFSLCK_CONFIRM) {
2939 		/*
2940 		 * If the open doesn't need confirmation, it seems to me that
2941 		 * there is a client error, but I'll just log it and keep going?
2942 		 */
2943 		if (!(stp->ls_openowner->ls_flags & NFSLCK_NEEDSCONFIRM))
2944 			printf("Nfsv4d: stray open confirm\n");
2945 		stp->ls_openowner->ls_flags = 0;
2946 		stp->ls_stateid.seqid++;
2947 		if (!(clp->lc_flags & LCL_STAMPEDSTABLE)) {
2948 			clp->lc_flags |= LCL_STAMPEDSTABLE;
2949 			len = clp->lc_idlen;
2950 			NFSBCOPY(clp->lc_id, client, len);
2951 			gotstate = 1;
2952 		}
2953 		NFSUNLOCKSTATE();
2954 	} else if (new_stp->ls_flags & NFSLCK_CLOSE) {
2955 		ownerstp = stp->ls_openowner;
2956 		lfp = stp->ls_lfp;
2957 		if (nfsrv_dolocallocks != 0 && !LIST_EMPTY(&stp->ls_open)) {
2958 			/* Get the lf lock */
2959 			nfsrv_locklf(lfp);
2960 			NFSUNLOCKSTATE();
2961 			if (nfsrv_freeopen(stp, vp, 1, p) == 0) {
2962 				NFSLOCKSTATE();
2963 				nfsrv_unlocklf(lfp);
2964 				NFSUNLOCKSTATE();
2965 			}
2966 		} else {
2967 			(void) nfsrv_freeopen(stp, NULL, 0, p);
2968 			NFSUNLOCKSTATE();
2969 		}
2970 	} else {
2971 		/*
2972 		 * Update the share bits, making sure that the new set are a
2973 		 * subset of the old ones.
2974 		 */
2975 		bits = (new_stp->ls_flags & NFSLCK_SHAREBITS);
2976 		if (~(stp->ls_flags) & bits) {
2977 			NFSUNLOCKSTATE();
2978 			error = NFSERR_INVAL;
2979 			goto out;
2980 		}
2981 		stp->ls_flags = (bits | NFSLCK_OPEN);
2982 		stp->ls_stateid.seqid++;
2983 		NFSUNLOCKSTATE();
2984 	}
2985 
2986 	/*
2987 	 * If the client just confirmed its first open, write a timestamp
2988 	 * to the stable storage file.
2989 	 */
2990 	if (gotstate != 0) {
2991 		nfsrv_writestable(client, len, NFSNST_NEWSTATE, p);
2992 		nfsrv_backupstable();
2993 	}
2994 
2995 out:
2996 	NFSEXITCODE2(error, nd);
2997 	return (error);
2998 }
2999 
3000 /*
3001  * Delegation update. Does the purge and return.
3002  */
3003 APPLESTATIC int
3004 nfsrv_delegupdate(nfsquad_t clientid, nfsv4stateid_t *stateidp,
3005     vnode_t vp, int op, struct ucred *cred, NFSPROC_T *p)
3006 {
3007 	struct nfsstate *stp;
3008 	struct nfsclient *clp;
3009 	int error = 0;
3010 	fhandle_t fh;
3011 
3012 	/*
3013 	 * Do a sanity check against the file handle for DelegReturn.
3014 	 */
3015 	if (vp) {
3016 		error = nfsvno_getfh(vp, &fh, p);
3017 		if (error)
3018 			goto out;
3019 	}
3020 	/*
3021 	 * Check for restart conditions (client and server).
3022 	 */
3023 	if (op == NFSV4OP_DELEGRETURN)
3024 		error = nfsrv_checkrestart(clientid, NFSLCK_DELEGRETURN,
3025 			stateidp, 0);
3026 	else
3027 		error = nfsrv_checkrestart(clientid, NFSLCK_DELEGPURGE,
3028 			stateidp, 0);
3029 
3030 	NFSLOCKSTATE();
3031 	/*
3032 	 * Get the open structure via clientid and stateid.
3033 	 */
3034 	if (!error)
3035 	    error = nfsrv_getclient(clientid, CLOPS_RENEW, &clp,
3036 		(nfsquad_t)((u_quad_t)0), NULL, p);
3037 	if (error) {
3038 		if (error == NFSERR_CBPATHDOWN)
3039 			error = 0;
3040 		if (error == NFSERR_STALECLIENTID && op == NFSV4OP_DELEGRETURN)
3041 			error = NFSERR_STALESTATEID;
3042 	}
3043 	if (!error && op == NFSV4OP_DELEGRETURN) {
3044 	    error = nfsrv_getstate(clp, stateidp, NFSLCK_DELEGRETURN, &stp);
3045 	    if (!error && stp->ls_stateid.seqid != stateidp->seqid)
3046 		error = NFSERR_OLDSTATEID;
3047 	}
3048 	/*
3049 	 * NFSERR_EXPIRED means that the state has gone away,
3050 	 * so Delegations have been purged. Just return ok.
3051 	 */
3052 	if (error == NFSERR_EXPIRED && op == NFSV4OP_DELEGPURGE) {
3053 		NFSUNLOCKSTATE();
3054 		error = 0;
3055 		goto out;
3056 	}
3057 	if (error) {
3058 		NFSUNLOCKSTATE();
3059 		goto out;
3060 	}
3061 
3062 	if (op == NFSV4OP_DELEGRETURN) {
3063 		if (NFSBCMP((caddr_t)&fh, (caddr_t)&stp->ls_lfp->lf_fh,
3064 		    sizeof (fhandle_t))) {
3065 			NFSUNLOCKSTATE();
3066 			error = NFSERR_BADSTATEID;
3067 			goto out;
3068 		}
3069 		nfsrv_freedeleg(stp);
3070 	} else {
3071 		nfsrv_freedeleglist(&clp->lc_olddeleg);
3072 	}
3073 	NFSUNLOCKSTATE();
3074 	error = 0;
3075 
3076 out:
3077 	NFSEXITCODE(error);
3078 	return (error);
3079 }
3080 
3081 /*
3082  * Release lock owner.
3083  */
3084 APPLESTATIC int
3085 nfsrv_releaselckown(struct nfsstate *new_stp, nfsquad_t clientid,
3086     NFSPROC_T *p)
3087 {
3088 	struct nfsstate *stp, *nstp, *openstp, *ownstp;
3089 	struct nfsclient *clp;
3090 	int error = 0;
3091 
3092 	/*
3093 	 * Check for restart conditions (client and server).
3094 	 */
3095 	error = nfsrv_checkrestart(clientid, new_stp->ls_flags,
3096 	    &new_stp->ls_stateid, 0);
3097 	if (error)
3098 		goto out;
3099 
3100 	NFSLOCKSTATE();
3101 	/*
3102 	 * Get the lock owner by name.
3103 	 */
3104 	error = nfsrv_getclient(clientid, CLOPS_RENEW, &clp,
3105 	    (nfsquad_t)((u_quad_t)0), NULL, p);
3106 	if (error) {
3107 		NFSUNLOCKSTATE();
3108 		goto out;
3109 	}
3110 	LIST_FOREACH(ownstp, &clp->lc_open, ls_list) {
3111 	    LIST_FOREACH(openstp, &ownstp->ls_open, ls_list) {
3112 		stp = LIST_FIRST(&openstp->ls_open);
3113 		while (stp != LIST_END(&openstp->ls_open)) {
3114 		    nstp = LIST_NEXT(stp, ls_list);
3115 		    /*
3116 		     * If the owner matches, check for locks and
3117 		     * then free or return an error.
3118 		     */
3119 		    if (stp->ls_ownerlen == new_stp->ls_ownerlen &&
3120 			!NFSBCMP(stp->ls_owner, new_stp->ls_owner,
3121 			 stp->ls_ownerlen)){
3122 			if (LIST_EMPTY(&stp->ls_lock)) {
3123 			    nfsrv_freelockowner(stp, NULL, 0, p);
3124 			} else {
3125 			    NFSUNLOCKSTATE();
3126 			    error = NFSERR_LOCKSHELD;
3127 			    goto out;
3128 			}
3129 		    }
3130 		    stp = nstp;
3131 		}
3132 	    }
3133 	}
3134 	NFSUNLOCKSTATE();
3135 
3136 out:
3137 	NFSEXITCODE(error);
3138 	return (error);
3139 }
3140 
3141 /*
3142  * Get the file handle for a lock structure.
3143  */
3144 static int
3145 nfsrv_getlockfh(vnode_t vp, u_short flags,
3146     struct nfslockfile **new_lfpp, fhandle_t *nfhp, NFSPROC_T *p)
3147 {
3148 	fhandle_t *fhp = NULL;
3149 	struct nfslockfile *new_lfp;
3150 	int error;
3151 
3152 	/*
3153 	 * For lock, use the new nfslock structure, otherwise just
3154 	 * a fhandle_t on the stack.
3155 	 */
3156 	if (flags & NFSLCK_OPEN) {
3157 		new_lfp = *new_lfpp;
3158 		fhp = &new_lfp->lf_fh;
3159 	} else if (nfhp) {
3160 		fhp = nfhp;
3161 	} else {
3162 		panic("nfsrv_getlockfh");
3163 	}
3164 	error = nfsvno_getfh(vp, fhp, p);
3165 	NFSEXITCODE(error);
3166 	return (error);
3167 }
3168 
3169 /*
3170  * Get an nfs lock structure. Allocate one, as required, and return a
3171  * pointer to it.
3172  * Returns an NFSERR_xxx upon failure or -1 to indicate no current lock.
3173  */
3174 static int
3175 nfsrv_getlockfile(u_short flags, struct nfslockfile **new_lfpp,
3176     struct nfslockfile **lfpp, fhandle_t *nfhp, int lockit)
3177 {
3178 	struct nfslockfile *lfp;
3179 	fhandle_t *fhp = NULL, *tfhp;
3180 	struct nfslockhashhead *hp;
3181 	struct nfslockfile *new_lfp = NULL;
3182 
3183 	/*
3184 	 * For lock, use the new nfslock structure, otherwise just
3185 	 * a fhandle_t on the stack.
3186 	 */
3187 	if (flags & NFSLCK_OPEN) {
3188 		new_lfp = *new_lfpp;
3189 		fhp = &new_lfp->lf_fh;
3190 	} else if (nfhp) {
3191 		fhp = nfhp;
3192 	} else {
3193 		panic("nfsrv_getlockfile");
3194 	}
3195 
3196 	hp = NFSLOCKHASH(fhp);
3197 	LIST_FOREACH(lfp, hp, lf_hash) {
3198 		tfhp = &lfp->lf_fh;
3199 		if (NFSVNO_CMPFH(fhp, tfhp)) {
3200 			if (lockit)
3201 				nfsrv_locklf(lfp);
3202 			*lfpp = lfp;
3203 			return (0);
3204 		}
3205 	}
3206 	if (!(flags & NFSLCK_OPEN))
3207 		return (-1);
3208 
3209 	/*
3210 	 * No match, so chain the new one into the list.
3211 	 */
3212 	LIST_INIT(&new_lfp->lf_open);
3213 	LIST_INIT(&new_lfp->lf_lock);
3214 	LIST_INIT(&new_lfp->lf_deleg);
3215 	LIST_INIT(&new_lfp->lf_locallock);
3216 	LIST_INIT(&new_lfp->lf_rollback);
3217 	new_lfp->lf_locallock_lck.nfslock_usecnt = 0;
3218 	new_lfp->lf_locallock_lck.nfslock_lock = 0;
3219 	new_lfp->lf_usecount = 0;
3220 	LIST_INSERT_HEAD(hp, new_lfp, lf_hash);
3221 	*lfpp = new_lfp;
3222 	*new_lfpp = NULL;
3223 	return (0);
3224 }
3225 
3226 /*
3227  * This function adds a nfslock lock structure to the list for the associated
3228  * nfsstate and nfslockfile structures. It will be inserted after the
3229  * entry pointed at by insert_lop.
3230  */
3231 static void
3232 nfsrv_insertlock(struct nfslock *new_lop, struct nfslock *insert_lop,
3233     struct nfsstate *stp, struct nfslockfile *lfp)
3234 {
3235 	struct nfslock *lop, *nlop;
3236 
3237 	new_lop->lo_stp = stp;
3238 	new_lop->lo_lfp = lfp;
3239 
3240 	if (stp != NULL) {
3241 		/* Insert in increasing lo_first order */
3242 		lop = LIST_FIRST(&lfp->lf_lock);
3243 		if (lop == LIST_END(&lfp->lf_lock) ||
3244 		    new_lop->lo_first <= lop->lo_first) {
3245 			LIST_INSERT_HEAD(&lfp->lf_lock, new_lop, lo_lckfile);
3246 		} else {
3247 			nlop = LIST_NEXT(lop, lo_lckfile);
3248 			while (nlop != LIST_END(&lfp->lf_lock) &&
3249 			       nlop->lo_first < new_lop->lo_first) {
3250 				lop = nlop;
3251 				nlop = LIST_NEXT(lop, lo_lckfile);
3252 			}
3253 			LIST_INSERT_AFTER(lop, new_lop, lo_lckfile);
3254 		}
3255 	} else {
3256 		new_lop->lo_lckfile.le_prev = NULL;	/* list not used */
3257 	}
3258 
3259 	/*
3260 	 * Insert after insert_lop, which is overloaded as stp or lfp for
3261 	 * an empty list.
3262 	 */
3263 	if (stp == NULL && (struct nfslockfile *)insert_lop == lfp)
3264 		LIST_INSERT_HEAD(&lfp->lf_locallock, new_lop, lo_lckowner);
3265 	else if ((struct nfsstate *)insert_lop == stp)
3266 		LIST_INSERT_HEAD(&stp->ls_lock, new_lop, lo_lckowner);
3267 	else
3268 		LIST_INSERT_AFTER(insert_lop, new_lop, lo_lckowner);
3269 	if (stp != NULL) {
3270 		newnfsstats.srvlocks++;
3271 		nfsrv_openpluslock++;
3272 	}
3273 }
3274 
3275 /*
3276  * This function updates the locking for a lock owner and given file. It
3277  * maintains a list of lock ranges ordered on increasing file offset that
3278  * are NFSLCK_READ or NFSLCK_WRITE and non-overlapping (aka POSIX style).
3279  * It always adds new_lop to the list and sometimes uses the one pointed
3280  * at by other_lopp.
3281  */
3282 static void
3283 nfsrv_updatelock(struct nfsstate *stp, struct nfslock **new_lopp,
3284     struct nfslock **other_lopp, struct nfslockfile *lfp)
3285 {
3286 	struct nfslock *new_lop = *new_lopp;
3287 	struct nfslock *lop, *tlop, *ilop;
3288 	struct nfslock *other_lop = *other_lopp;
3289 	int unlock = 0, myfile = 0;
3290 	u_int64_t tmp;
3291 
3292 	/*
3293 	 * Work down the list until the lock is merged.
3294 	 */
3295 	if (new_lop->lo_flags & NFSLCK_UNLOCK)
3296 		unlock = 1;
3297 	if (stp != NULL) {
3298 		ilop = (struct nfslock *)stp;
3299 		lop = LIST_FIRST(&stp->ls_lock);
3300 	} else {
3301 		ilop = (struct nfslock *)lfp;
3302 		lop = LIST_FIRST(&lfp->lf_locallock);
3303 	}
3304 	while (lop != NULL) {
3305 	    /*
3306 	     * Only check locks for this file that aren't before the start of
3307 	     * new lock's range.
3308 	     */
3309 	    if (lop->lo_lfp == lfp) {
3310 	      myfile = 1;
3311 	      if (lop->lo_end >= new_lop->lo_first) {
3312 		if (new_lop->lo_end < lop->lo_first) {
3313 			/*
3314 			 * If the new lock ends before the start of the
3315 			 * current lock's range, no merge, just insert
3316 			 * the new lock.
3317 			 */
3318 			break;
3319 		}
3320 		if (new_lop->lo_flags == lop->lo_flags ||
3321 		    (new_lop->lo_first <= lop->lo_first &&
3322 		     new_lop->lo_end >= lop->lo_end)) {
3323 			/*
3324 			 * This lock can be absorbed by the new lock/unlock.
3325 			 * This happens when it covers the entire range
3326 			 * of the old lock or is contiguous
3327 			 * with the old lock and is of the same type or an
3328 			 * unlock.
3329 			 */
3330 			if (lop->lo_first < new_lop->lo_first)
3331 				new_lop->lo_first = lop->lo_first;
3332 			if (lop->lo_end > new_lop->lo_end)
3333 				new_lop->lo_end = lop->lo_end;
3334 			tlop = lop;
3335 			lop = LIST_NEXT(lop, lo_lckowner);
3336 			nfsrv_freenfslock(tlop);
3337 			continue;
3338 		}
3339 
3340 		/*
3341 		 * All these cases are for contiguous locks that are not the
3342 		 * same type, so they can't be merged.
3343 		 */
3344 		if (new_lop->lo_first <= lop->lo_first) {
3345 			/*
3346 			 * This case is where the new lock overlaps with the
3347 			 * first part of the old lock. Move the start of the
3348 			 * old lock to just past the end of the new lock. The
3349 			 * new lock will be inserted in front of the old, since
3350 			 * ilop hasn't been updated. (We are done now.)
3351 			 */
3352 			lop->lo_first = new_lop->lo_end;
3353 			break;
3354 		}
3355 		if (new_lop->lo_end >= lop->lo_end) {
3356 			/*
3357 			 * This case is where the new lock overlaps with the
3358 			 * end of the old lock's range. Move the old lock's
3359 			 * end to just before the new lock's first and insert
3360 			 * the new lock after the old lock.
3361 			 * Might not be done yet, since the new lock could
3362 			 * overlap further locks with higher ranges.
3363 			 */
3364 			lop->lo_end = new_lop->lo_first;
3365 			ilop = lop;
3366 			lop = LIST_NEXT(lop, lo_lckowner);
3367 			continue;
3368 		}
3369 		/*
3370 		 * The final case is where the new lock's range is in the
3371 		 * middle of the current lock's and splits the current lock
3372 		 * up. Use *other_lopp to handle the second part of the
3373 		 * split old lock range. (We are done now.)
3374 		 * For unlock, we use new_lop as other_lop and tmp, since
3375 		 * other_lop and new_lop are the same for this case.
3376 		 * We noted the unlock case above, so we don't need
3377 		 * new_lop->lo_flags any longer.
3378 		 */
3379 		tmp = new_lop->lo_first;
3380 		if (other_lop == NULL) {
3381 			if (!unlock)
3382 				panic("nfsd srv update unlock");
3383 			other_lop = new_lop;
3384 			*new_lopp = NULL;
3385 		}
3386 		other_lop->lo_first = new_lop->lo_end;
3387 		other_lop->lo_end = lop->lo_end;
3388 		other_lop->lo_flags = lop->lo_flags;
3389 		other_lop->lo_stp = stp;
3390 		other_lop->lo_lfp = lfp;
3391 		lop->lo_end = tmp;
3392 		nfsrv_insertlock(other_lop, lop, stp, lfp);
3393 		*other_lopp = NULL;
3394 		ilop = lop;
3395 		break;
3396 	      }
3397 	    }
3398 	    ilop = lop;
3399 	    lop = LIST_NEXT(lop, lo_lckowner);
3400 	    if (myfile && (lop == NULL || lop->lo_lfp != lfp))
3401 		break;
3402 	}
3403 
3404 	/*
3405 	 * Insert the new lock in the list at the appropriate place.
3406 	 */
3407 	if (!unlock) {
3408 		nfsrv_insertlock(new_lop, ilop, stp, lfp);
3409 		*new_lopp = NULL;
3410 	}
3411 }
3412 
3413 /*
3414  * This function handles sequencing of locks, etc.
3415  * It returns an error that indicates what the caller should do.
3416  */
3417 static int
3418 nfsrv_checkseqid(struct nfsrv_descript *nd, u_int32_t seqid,
3419     struct nfsstate *stp, struct nfsrvcache *op)
3420 {
3421 	int error = 0;
3422 
3423 	if (op != nd->nd_rp)
3424 		panic("nfsrvstate checkseqid");
3425 	if (!(op->rc_flag & RC_INPROG))
3426 		panic("nfsrvstate not inprog");
3427 	if (stp->ls_op && stp->ls_op->rc_refcnt <= 0) {
3428 		printf("refcnt=%d\n", stp->ls_op->rc_refcnt);
3429 		panic("nfsrvstate op refcnt");
3430 	}
3431 	if ((stp->ls_seq + 1) == seqid) {
3432 		if (stp->ls_op)
3433 			nfsrvd_derefcache(stp->ls_op);
3434 		stp->ls_op = op;
3435 		nfsrvd_refcache(op);
3436 		stp->ls_seq = seqid;
3437 		goto out;
3438 	} else if (stp->ls_seq == seqid && stp->ls_op &&
3439 		op->rc_xid == stp->ls_op->rc_xid &&
3440 		op->rc_refcnt == 0 &&
3441 		op->rc_reqlen == stp->ls_op->rc_reqlen &&
3442 		op->rc_cksum == stp->ls_op->rc_cksum) {
3443 		if (stp->ls_op->rc_flag & RC_INPROG) {
3444 			error = NFSERR_DONTREPLY;
3445 			goto out;
3446 		}
3447 		nd->nd_rp = stp->ls_op;
3448 		nd->nd_rp->rc_flag |= RC_INPROG;
3449 		nfsrvd_delcache(op);
3450 		error = NFSERR_REPLYFROMCACHE;
3451 		goto out;
3452 	}
3453 	error = NFSERR_BADSEQID;
3454 
3455 out:
3456 	NFSEXITCODE2(error, nd);
3457 	return (error);
3458 }
3459 
3460 /*
3461  * Get the client ip address for callbacks. If the strings can't be parsed,
3462  * just set lc_program to 0 to indicate no callbacks are possible.
3463  * (For cases where the address can't be parsed or is 0.0.0.0.0.0, set
3464  *  the address to the client's transport address. This won't be used
3465  *  for callbacks, but can be printed out by newnfsstats for info.)
3466  * Return error if the xdr can't be parsed, 0 otherwise.
3467  */
3468 APPLESTATIC int
3469 nfsrv_getclientipaddr(struct nfsrv_descript *nd, struct nfsclient *clp)
3470 {
3471 	u_int32_t *tl;
3472 	u_char *cp, *cp2;
3473 	int i, j;
3474 	struct sockaddr_in *rad, *sad;
3475 	u_char protocol[5], addr[24];
3476 	int error = 0, cantparse = 0;
3477 	union {
3478 		u_long ival;
3479 		u_char cval[4];
3480 	} ip;
3481 	union {
3482 		u_short sval;
3483 		u_char cval[2];
3484 	} port;
3485 
3486 	rad = NFSSOCKADDR(clp->lc_req.nr_nam, struct sockaddr_in *);
3487 	rad->sin_family = AF_INET;
3488 	rad->sin_len = sizeof (struct sockaddr_in);
3489 	rad->sin_addr.s_addr = 0;
3490 	rad->sin_port = 0;
3491 	clp->lc_req.nr_client = NULL;
3492 	clp->lc_req.nr_lock = 0;
3493 	NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
3494 	i = fxdr_unsigned(int, *tl);
3495 	if (i >= 3 && i <= 4) {
3496 		error = nfsrv_mtostr(nd, protocol, i);
3497 		if (error)
3498 			goto nfsmout;
3499 		if (!strcmp(protocol, "tcp")) {
3500 			clp->lc_flags |= LCL_TCPCALLBACK;
3501 			clp->lc_req.nr_sotype = SOCK_STREAM;
3502 			clp->lc_req.nr_soproto = IPPROTO_TCP;
3503 		} else if (!strcmp(protocol, "udp")) {
3504 			clp->lc_req.nr_sotype = SOCK_DGRAM;
3505 			clp->lc_req.nr_soproto = IPPROTO_UDP;
3506 		} else {
3507 			cantparse = 1;
3508 		}
3509 	} else {
3510 		cantparse = 1;
3511 		if (i > 0) {
3512 			error = nfsm_advance(nd, NFSM_RNDUP(i), -1);
3513 			if (error)
3514 				goto nfsmout;
3515 		}
3516 	}
3517 	NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
3518 	i = fxdr_unsigned(int, *tl);
3519 	if (i < 0) {
3520 		error = NFSERR_BADXDR;
3521 		goto nfsmout;
3522 	} else if (i == 0) {
3523 		cantparse = 1;
3524 	} else if (!cantparse && i <= 23 && i >= 11) {
3525 		error = nfsrv_mtostr(nd, addr, i);
3526 		if (error)
3527 			goto nfsmout;
3528 
3529 		/*
3530 		 * Parse out the address fields. We expect 6 decimal numbers
3531 		 * separated by '.'s.
3532 		 */
3533 		cp = addr;
3534 		i = 0;
3535 		while (*cp && i < 6) {
3536 			cp2 = cp;
3537 			while (*cp2 && *cp2 != '.')
3538 				cp2++;
3539 			if (*cp2)
3540 				*cp2++ = '\0';
3541 			else if (i != 5) {
3542 				cantparse = 1;
3543 				break;
3544 			}
3545 			j = nfsrv_getipnumber(cp);
3546 			if (j >= 0) {
3547 				if (i < 4)
3548 					ip.cval[3 - i] = j;
3549 				else
3550 					port.cval[5 - i] = j;
3551 			} else {
3552 				cantparse = 1;
3553 				break;
3554 			}
3555 			cp = cp2;
3556 			i++;
3557 		}
3558 		if (!cantparse) {
3559 			if (ip.ival != 0x0) {
3560 				rad->sin_addr.s_addr = htonl(ip.ival);
3561 				rad->sin_port = htons(port.sval);
3562 			} else {
3563 				cantparse = 1;
3564 			}
3565 		}
3566 	} else {
3567 		cantparse = 1;
3568 		if (i > 0) {
3569 			error = nfsm_advance(nd, NFSM_RNDUP(i), -1);
3570 			if (error)
3571 				goto nfsmout;
3572 		}
3573 	}
3574 	if (cantparse) {
3575 		sad = NFSSOCKADDR(nd->nd_nam, struct sockaddr_in *);
3576 		rad->sin_addr.s_addr = sad->sin_addr.s_addr;
3577 		rad->sin_port = 0x0;
3578 		clp->lc_program = 0;
3579 	}
3580 nfsmout:
3581 	NFSEXITCODE2(error, nd);
3582 	return (error);
3583 }
3584 
3585 /*
3586  * Turn a string of up to three decimal digits into a number. Return -1 upon
3587  * error.
3588  */
3589 static int
3590 nfsrv_getipnumber(u_char *cp)
3591 {
3592 	int i = 0, j = 0;
3593 
3594 	while (*cp) {
3595 		if (j > 2 || *cp < '0' || *cp > '9')
3596 			return (-1);
3597 		i *= 10;
3598 		i += (*cp - '0');
3599 		cp++;
3600 		j++;
3601 	}
3602 	if (i < 256)
3603 		return (i);
3604 	return (-1);
3605 }
3606 
3607 /*
3608  * This function checks for restart conditions.
3609  */
3610 static int
3611 nfsrv_checkrestart(nfsquad_t clientid, u_int32_t flags,
3612     nfsv4stateid_t *stateidp, int specialid)
3613 {
3614 	int ret = 0;
3615 
3616 	/*
3617 	 * First check for a server restart. Open, LockT, ReleaseLockOwner
3618 	 * and DelegPurge have a clientid, the rest a stateid.
3619 	 */
3620 	if (flags &
3621 	    (NFSLCK_OPEN | NFSLCK_TEST | NFSLCK_RELEASE | NFSLCK_DELEGPURGE)) {
3622 		if (clientid.lval[0] != nfsrvboottime) {
3623 			ret = NFSERR_STALECLIENTID;
3624 			goto out;
3625 		}
3626 	} else if (stateidp->other[0] != nfsrvboottime &&
3627 		specialid == 0) {
3628 		ret = NFSERR_STALESTATEID;
3629 		goto out;
3630 	}
3631 
3632 	/*
3633 	 * Read, Write, Setattr and LockT can return NFSERR_GRACE and do
3634 	 * not use a lock/open owner seqid#, so the check can be done now.
3635 	 * (The others will be checked, as required, later.)
3636 	 */
3637 	if (!(flags & (NFSLCK_CHECK | NFSLCK_TEST)))
3638 		goto out;
3639 
3640 	NFSLOCKSTATE();
3641 	ret = nfsrv_checkgrace(flags);
3642 	NFSUNLOCKSTATE();
3643 
3644 out:
3645 	NFSEXITCODE(ret);
3646 	return (ret);
3647 }
3648 
3649 /*
3650  * Check for grace.
3651  */
3652 static int
3653 nfsrv_checkgrace(u_int32_t flags)
3654 {
3655 	int error = 0;
3656 
3657 	if (nfsrv_stablefirst.nsf_flags & NFSNSF_GRACEOVER) {
3658 		if (flags & NFSLCK_RECLAIM) {
3659 			error = NFSERR_NOGRACE;
3660 			goto out;
3661 		}
3662 	} else {
3663 		if (!(flags & NFSLCK_RECLAIM)) {
3664 			error = NFSERR_GRACE;
3665 			goto out;
3666 		}
3667 
3668 		/*
3669 		 * If grace is almost over and we are still getting Reclaims,
3670 		 * extend grace a bit.
3671 		 */
3672 		if ((NFSD_MONOSEC + NFSRV_LEASEDELTA) >
3673 		    nfsrv_stablefirst.nsf_eograce)
3674 			nfsrv_stablefirst.nsf_eograce = NFSD_MONOSEC +
3675 				NFSRV_LEASEDELTA;
3676 	}
3677 
3678 out:
3679 	NFSEXITCODE(error);
3680 	return (error);
3681 }
3682 
3683 /*
3684  * Do a server callback.
3685  */
3686 static int
3687 nfsrv_docallback(struct nfsclient *clp, int procnum,
3688     nfsv4stateid_t *stateidp, int trunc, fhandle_t *fhp,
3689     struct nfsvattr *nap, nfsattrbit_t *attrbitp, NFSPROC_T *p)
3690 {
3691 	mbuf_t m;
3692 	u_int32_t *tl;
3693 	struct nfsrv_descript nfsd, *nd = &nfsd;
3694 	struct ucred *cred;
3695 	int error = 0;
3696 	u_int32_t callback;
3697 
3698 	cred = newnfs_getcred();
3699 	NFSLOCKSTATE();	/* mostly for lc_cbref++ */
3700 	if (clp->lc_flags & LCL_NEEDSCONFIRM) {
3701 		NFSUNLOCKSTATE();
3702 		panic("docallb");
3703 	}
3704 	clp->lc_cbref++;
3705 
3706 	/*
3707 	 * Fill the callback program# and version into the request
3708 	 * structure for newnfs_connect() to use.
3709 	 */
3710 	clp->lc_req.nr_prog = clp->lc_program;
3711 	clp->lc_req.nr_vers = NFSV4_CBVERS;
3712 
3713 	/*
3714 	 * First, fill in some of the fields of nd and cr.
3715 	 */
3716 	nd->nd_flag = ND_NFSV4;
3717 	if (clp->lc_flags & LCL_GSS)
3718 		nd->nd_flag |= ND_KERBV;
3719 	nd->nd_repstat = 0;
3720 	cred->cr_uid = clp->lc_uid;
3721 	cred->cr_gid = clp->lc_gid;
3722 	callback = clp->lc_callback;
3723 	NFSUNLOCKSTATE();
3724 	cred->cr_ngroups = 1;
3725 
3726 	/*
3727 	 * Get the first mbuf for the request.
3728 	 */
3729 	MGET(m, M_WAITOK, MT_DATA);
3730 	mbuf_setlen(m, 0);
3731 	nd->nd_mreq = nd->nd_mb = m;
3732 	nd->nd_bpos = NFSMTOD(m, caddr_t);
3733 
3734 	/*
3735 	 * and build the callback request.
3736 	 */
3737 	if (procnum == NFSV4OP_CBGETATTR) {
3738 		nd->nd_procnum = NFSV4PROC_CBCOMPOUND;
3739 		(void) nfsm_strtom(nd, "CB Getattr", 10);
3740 		NFSM_BUILD(tl, u_int32_t *, 4 * NFSX_UNSIGNED);
3741 		*tl++ = txdr_unsigned(NFSV4_MINORVERSION);
3742 		*tl++ = txdr_unsigned(callback);
3743 		*tl++ = txdr_unsigned(1);
3744 		*tl = txdr_unsigned(NFSV4OP_CBGETATTR);
3745 		(void) nfsm_fhtom(nd, (u_int8_t *)fhp, NFSX_MYFH, 0);
3746 		(void) nfsrv_putattrbit(nd, attrbitp);
3747 	} else if (procnum == NFSV4OP_CBRECALL) {
3748 		nd->nd_procnum = NFSV4PROC_CBCOMPOUND;
3749 		(void) nfsm_strtom(nd, "CB Recall", 9);
3750 		NFSM_BUILD(tl, u_int32_t *, 5 * NFSX_UNSIGNED + NFSX_STATEID);
3751 		*tl++ = txdr_unsigned(NFSV4_MINORVERSION);
3752 		*tl++ = txdr_unsigned(callback);
3753 		*tl++ = txdr_unsigned(1);
3754 		*tl++ = txdr_unsigned(NFSV4OP_CBRECALL);
3755 		*tl++ = txdr_unsigned(stateidp->seqid);
3756 		NFSBCOPY((caddr_t)stateidp->other, (caddr_t)tl,
3757 		    NFSX_STATEIDOTHER);
3758 		tl += (NFSX_STATEIDOTHER / NFSX_UNSIGNED);
3759 		if (trunc)
3760 			*tl = newnfs_true;
3761 		else
3762 			*tl = newnfs_false;
3763 		(void) nfsm_fhtom(nd, (u_int8_t *)fhp, NFSX_MYFH, 0);
3764 	} else {
3765 		nd->nd_procnum = NFSV4PROC_CBNULL;
3766 	}
3767 
3768 	/*
3769 	 * Call newnfs_connect(), as required, and then newnfs_request().
3770 	 */
3771 	(void) newnfs_sndlock(&clp->lc_req.nr_lock);
3772 	if (clp->lc_req.nr_client == NULL) {
3773 		if (nd->nd_procnum == NFSV4PROC_CBNULL)
3774 			error = newnfs_connect(NULL, &clp->lc_req, cred,
3775 			    NULL, 1);
3776 		else
3777 			error = newnfs_connect(NULL, &clp->lc_req, cred,
3778 			    NULL, 3);
3779 	}
3780 	newnfs_sndunlock(&clp->lc_req.nr_lock);
3781 	if (!error) {
3782 		error = newnfs_request(nd, NULL, clp, &clp->lc_req, NULL,
3783 		    NULL, cred, clp->lc_program, NFSV4_CBVERS, NULL, 1, NULL,
3784 		    NULL);
3785 	}
3786 	NFSFREECRED(cred);
3787 
3788 	/*
3789 	 * If error is set here, the Callback path isn't working
3790 	 * properly, so twiddle the appropriate LCL_ flags.
3791 	 * (nd_repstat != 0 indicates the Callback path is working,
3792 	 *  but the callback failed on the client.)
3793 	 */
3794 	if (error) {
3795 		/*
3796 		 * Mark the callback pathway down, which disabled issuing
3797 		 * of delegations and gets Renew to return NFSERR_CBPATHDOWN.
3798 		 */
3799 		NFSLOCKSTATE();
3800 		clp->lc_flags |= LCL_CBDOWN;
3801 		NFSUNLOCKSTATE();
3802 	} else {
3803 		/*
3804 		 * Callback worked. If the callback path was down, disable
3805 		 * callbacks, so no more delegations will be issued. (This
3806 		 * is done on the assumption that the callback pathway is
3807 		 * flakey.)
3808 		 */
3809 		NFSLOCKSTATE();
3810 		if (clp->lc_flags & LCL_CBDOWN)
3811 			clp->lc_flags &= ~(LCL_CBDOWN | LCL_CALLBACKSON);
3812 		NFSUNLOCKSTATE();
3813 		if (nd->nd_repstat)
3814 			error = nd->nd_repstat;
3815 		else if (procnum == NFSV4OP_CBGETATTR)
3816 			error = nfsv4_loadattr(nd, NULL, nap, NULL, NULL, 0,
3817 			    NULL, NULL, NULL, NULL, NULL, 0, NULL, NULL, NULL,
3818 			    p, NULL);
3819 		mbuf_freem(nd->nd_mrep);
3820 	}
3821 	NFSLOCKSTATE();
3822 	clp->lc_cbref--;
3823 	if ((clp->lc_flags & LCL_WAKEUPWANTED) && clp->lc_cbref == 0) {
3824 		clp->lc_flags &= ~LCL_WAKEUPWANTED;
3825 		wakeup(clp);
3826 	}
3827 	NFSUNLOCKSTATE();
3828 
3829 	NFSEXITCODE(error);
3830 	return (error);
3831 }
3832 
3833 /*
3834  * Return the next index# for a clientid. Mostly just increment and return
3835  * the next one, but... if the 32bit unsigned does actually wrap around,
3836  * it should be rebooted.
3837  * At an average rate of one new client per second, it will wrap around in
3838  * approximately 136 years. (I think the server will have been shut
3839  * down or rebooted before then.)
3840  */
3841 static u_int32_t
3842 nfsrv_nextclientindex(void)
3843 {
3844 	static u_int32_t client_index = 0;
3845 
3846 	client_index++;
3847 	if (client_index != 0)
3848 		return (client_index);
3849 
3850 	printf("%s: out of clientids\n", __func__);
3851 	return (client_index);
3852 }
3853 
3854 /*
3855  * Return the next index# for a stateid. Mostly just increment and return
3856  * the next one, but... if the 32bit unsigned does actually wrap around
3857  * (will a BSD server stay up that long?), find
3858  * new start and end values.
3859  */
3860 static u_int32_t
3861 nfsrv_nextstateindex(struct nfsclient *clp)
3862 {
3863 	struct nfsstate *stp;
3864 	int i;
3865 	u_int32_t canuse, min_index, max_index;
3866 
3867 	if (!(clp->lc_flags & LCL_INDEXNOTOK)) {
3868 		clp->lc_stateindex++;
3869 		if (clp->lc_stateindex != clp->lc_statemaxindex)
3870 			return (clp->lc_stateindex);
3871 	}
3872 
3873 	/*
3874 	 * Yuck, we've hit the end.
3875 	 * Look for a new min and max.
3876 	 */
3877 	min_index = 0;
3878 	max_index = 0xffffffff;
3879 	for (i = 0; i < NFSSTATEHASHSIZE; i++) {
3880 	    LIST_FOREACH(stp, &clp->lc_stateid[i], ls_hash) {
3881 		if (stp->ls_stateid.other[2] > 0x80000000) {
3882 		    if (stp->ls_stateid.other[2] < max_index)
3883 			max_index = stp->ls_stateid.other[2];
3884 		} else {
3885 		    if (stp->ls_stateid.other[2] > min_index)
3886 			min_index = stp->ls_stateid.other[2];
3887 		}
3888 	    }
3889 	}
3890 
3891 	/*
3892 	 * Yikes, highly unlikely, but I'll handle it anyhow.
3893 	 */
3894 	if (min_index == 0x80000000 && max_index == 0x80000001) {
3895 	    canuse = 0;
3896 	    /*
3897 	     * Loop around until we find an unused entry. Return that
3898 	     * and set LCL_INDEXNOTOK, so the search will continue next time.
3899 	     * (This is one of those rare cases where a goto is the
3900 	     *  cleanest way to code the loop.)
3901 	     */
3902 tryagain:
3903 	    for (i = 0; i < NFSSTATEHASHSIZE; i++) {
3904 		LIST_FOREACH(stp, &clp->lc_stateid[i], ls_hash) {
3905 		    if (stp->ls_stateid.other[2] == canuse) {
3906 			canuse++;
3907 			goto tryagain;
3908 		    }
3909 		}
3910 	    }
3911 	    clp->lc_flags |= LCL_INDEXNOTOK;
3912 	    return (canuse);
3913 	}
3914 
3915 	/*
3916 	 * Ok to start again from min + 1.
3917 	 */
3918 	clp->lc_stateindex = min_index + 1;
3919 	clp->lc_statemaxindex = max_index;
3920 	clp->lc_flags &= ~LCL_INDEXNOTOK;
3921 	return (clp->lc_stateindex);
3922 }
3923 
3924 /*
3925  * The following functions handle the stable storage file that deals with
3926  * the edge conditions described in RFC3530 Sec. 8.6.3.
3927  * The file is as follows:
3928  * - a single record at the beginning that has the lease time of the
3929  *   previous server instance (before the last reboot) and the nfsrvboottime
3930  *   values for the previous server boots.
3931  *   These previous boot times are used to ensure that the current
3932  *   nfsrvboottime does not, somehow, get set to a previous one.
3933  *   (This is important so that Stale ClientIDs and StateIDs can
3934  *    be recognized.)
3935  *   The number of previous nfsvrboottime values preceeds the list.
3936  * - followed by some number of appended records with:
3937  *   - client id string
3938  *   - flag that indicates it is a record revoking state via lease
3939  *     expiration or similar
3940  *     OR has successfully acquired state.
3941  * These structures vary in length, with the client string at the end, up
3942  * to NFSV4_OPAQUELIMIT in size.
3943  *
3944  * At the end of the grace period, the file is truncated, the first
3945  * record is rewritten with updated information and any acquired state
3946  * records for successful reclaims of state are written.
3947  *
3948  * Subsequent records are appended when the first state is issued to
3949  * a client and when state is revoked for a client.
3950  *
3951  * When reading the file in, state issued records that come later in
3952  * the file override older ones, since the append log is in cronological order.
3953  * If, for some reason, the file can't be read, the grace period is
3954  * immediately terminated and all reclaims get NFSERR_NOGRACE.
3955  */
3956 
3957 /*
3958  * Read in the stable storage file. Called by nfssvc() before the nfsd
3959  * processes start servicing requests.
3960  */
3961 APPLESTATIC void
3962 nfsrv_setupstable(NFSPROC_T *p)
3963 {
3964 	struct nfsrv_stablefirst *sf = &nfsrv_stablefirst;
3965 	struct nfsrv_stable *sp, *nsp;
3966 	struct nfst_rec *tsp;
3967 	int error, i, tryagain;
3968 	off_t off = 0;
3969 	ssize_t aresid, len;
3970 
3971 	/*
3972 	 * If NFSNSF_UPDATEDONE is set, this is a restart of the nfsds without
3973 	 * a reboot, so state has not been lost.
3974 	 */
3975 	if (sf->nsf_flags & NFSNSF_UPDATEDONE)
3976 		return;
3977 	/*
3978 	 * Set Grace over just until the file reads successfully.
3979 	 */
3980 	nfsrvboottime = time_second;
3981 	LIST_INIT(&sf->nsf_head);
3982 	sf->nsf_flags = (NFSNSF_GRACEOVER | NFSNSF_NEEDLOCK);
3983 	sf->nsf_eograce = NFSD_MONOSEC + NFSRV_LEASEDELTA;
3984 	if (sf->nsf_fp == NULL)
3985 		return;
3986 	error = NFSD_RDWR(UIO_READ, NFSFPVNODE(sf->nsf_fp),
3987 	    (caddr_t)&sf->nsf_rec, sizeof (struct nfsf_rec), off, UIO_SYSSPACE,
3988 	    0, NFSFPCRED(sf->nsf_fp), &aresid, p);
3989 	if (error || aresid || sf->nsf_numboots == 0 ||
3990 		sf->nsf_numboots > NFSNSF_MAXNUMBOOTS)
3991 		return;
3992 
3993 	/*
3994 	 * Now, read in the boottimes.
3995 	 */
3996 	sf->nsf_bootvals = (time_t *)malloc((sf->nsf_numboots + 1) *
3997 		sizeof (time_t), M_TEMP, M_WAITOK);
3998 	off = sizeof (struct nfsf_rec);
3999 	error = NFSD_RDWR(UIO_READ, NFSFPVNODE(sf->nsf_fp),
4000 	    (caddr_t)sf->nsf_bootvals, sf->nsf_numboots * sizeof (time_t), off,
4001 	    UIO_SYSSPACE, 0, NFSFPCRED(sf->nsf_fp), &aresid, p);
4002 	if (error || aresid) {
4003 		free((caddr_t)sf->nsf_bootvals, M_TEMP);
4004 		sf->nsf_bootvals = NULL;
4005 		return;
4006 	}
4007 
4008 	/*
4009 	 * Make sure this nfsrvboottime is different from all recorded
4010 	 * previous ones.
4011 	 */
4012 	do {
4013 		tryagain = 0;
4014 		for (i = 0; i < sf->nsf_numboots; i++) {
4015 			if (nfsrvboottime == sf->nsf_bootvals[i]) {
4016 				nfsrvboottime++;
4017 				tryagain = 1;
4018 				break;
4019 			}
4020 		}
4021 	} while (tryagain);
4022 
4023 	sf->nsf_flags |= NFSNSF_OK;
4024 	off += (sf->nsf_numboots * sizeof (time_t));
4025 
4026 	/*
4027 	 * Read through the file, building a list of records for grace
4028 	 * checking.
4029 	 * Each record is between sizeof (struct nfst_rec) and
4030 	 * sizeof (struct nfst_rec) + NFSV4_OPAQUELIMIT - 1
4031 	 * and is actually sizeof (struct nfst_rec) + nst_len - 1.
4032 	 */
4033 	tsp = (struct nfst_rec *)malloc(sizeof (struct nfst_rec) +
4034 		NFSV4_OPAQUELIMIT - 1, M_TEMP, M_WAITOK);
4035 	do {
4036 	    error = NFSD_RDWR(UIO_READ, NFSFPVNODE(sf->nsf_fp),
4037 	        (caddr_t)tsp, sizeof (struct nfst_rec) + NFSV4_OPAQUELIMIT - 1,
4038 	        off, UIO_SYSSPACE, 0, NFSFPCRED(sf->nsf_fp), &aresid, p);
4039 	    len = (sizeof (struct nfst_rec) + NFSV4_OPAQUELIMIT - 1) - aresid;
4040 	    if (error || (len > 0 && (len < sizeof (struct nfst_rec) ||
4041 		len < (sizeof (struct nfst_rec) + tsp->len - 1)))) {
4042 		/*
4043 		 * Yuck, the file has been corrupted, so just return
4044 		 * after clearing out any restart state, so the grace period
4045 		 * is over.
4046 		 */
4047 		LIST_FOREACH_SAFE(sp, &sf->nsf_head, nst_list, nsp) {
4048 			LIST_REMOVE(sp, nst_list);
4049 			free((caddr_t)sp, M_TEMP);
4050 		}
4051 		free((caddr_t)tsp, M_TEMP);
4052 		sf->nsf_flags &= ~NFSNSF_OK;
4053 		free((caddr_t)sf->nsf_bootvals, M_TEMP);
4054 		sf->nsf_bootvals = NULL;
4055 		return;
4056 	    }
4057 	    if (len > 0) {
4058 		off += sizeof (struct nfst_rec) + tsp->len - 1;
4059 		/*
4060 		 * Search the list for a matching client.
4061 		 */
4062 		LIST_FOREACH(sp, &sf->nsf_head, nst_list) {
4063 			if (tsp->len == sp->nst_len &&
4064 			    !NFSBCMP(tsp->client, sp->nst_client, tsp->len))
4065 				break;
4066 		}
4067 		if (sp == LIST_END(&sf->nsf_head)) {
4068 			sp = (struct nfsrv_stable *)malloc(tsp->len +
4069 				sizeof (struct nfsrv_stable) - 1, M_TEMP,
4070 				M_WAITOK);
4071 			NFSBCOPY((caddr_t)tsp, (caddr_t)&sp->nst_rec,
4072 				sizeof (struct nfst_rec) + tsp->len - 1);
4073 			LIST_INSERT_HEAD(&sf->nsf_head, sp, nst_list);
4074 		} else {
4075 			if (tsp->flag == NFSNST_REVOKE)
4076 				sp->nst_flag |= NFSNST_REVOKE;
4077 			else
4078 				/*
4079 				 * A subsequent timestamp indicates the client
4080 				 * did a setclientid/confirm and any previous
4081 				 * revoke is no longer relevant.
4082 				 */
4083 				sp->nst_flag &= ~NFSNST_REVOKE;
4084 		}
4085 	    }
4086 	} while (len > 0);
4087 	free((caddr_t)tsp, M_TEMP);
4088 	sf->nsf_flags = NFSNSF_OK;
4089 	sf->nsf_eograce = NFSD_MONOSEC + sf->nsf_lease +
4090 		NFSRV_LEASEDELTA;
4091 }
4092 
4093 /*
4094  * Update the stable storage file, now that the grace period is over.
4095  */
4096 APPLESTATIC void
4097 nfsrv_updatestable(NFSPROC_T *p)
4098 {
4099 	struct nfsrv_stablefirst *sf = &nfsrv_stablefirst;
4100 	struct nfsrv_stable *sp, *nsp;
4101 	int i;
4102 	struct nfsvattr nva;
4103 	vnode_t vp;
4104 #if defined(__FreeBSD_version) && (__FreeBSD_version >= 500000)
4105 	mount_t mp = NULL;
4106 #endif
4107 	int error;
4108 
4109 	if (sf->nsf_fp == NULL || (sf->nsf_flags & NFSNSF_UPDATEDONE))
4110 		return;
4111 	sf->nsf_flags |= NFSNSF_UPDATEDONE;
4112 	/*
4113 	 * Ok, we need to rewrite the stable storage file.
4114 	 * - truncate to 0 length
4115 	 * - write the new first structure
4116 	 * - loop through the data structures, writing out any that
4117 	 *   have timestamps older than the old boot
4118 	 */
4119 	if (sf->nsf_bootvals) {
4120 		sf->nsf_numboots++;
4121 		for (i = sf->nsf_numboots - 2; i >= 0; i--)
4122 			sf->nsf_bootvals[i + 1] = sf->nsf_bootvals[i];
4123 	} else {
4124 		sf->nsf_numboots = 1;
4125 		sf->nsf_bootvals = (time_t *)malloc(sizeof (time_t),
4126 			M_TEMP, M_WAITOK);
4127 	}
4128 	sf->nsf_bootvals[0] = nfsrvboottime;
4129 	sf->nsf_lease = nfsrv_lease;
4130 	NFSVNO_ATTRINIT(&nva);
4131 	NFSVNO_SETATTRVAL(&nva, size, 0);
4132 	vp = NFSFPVNODE(sf->nsf_fp);
4133 	vn_start_write(vp, &mp, V_WAIT);
4134 	if (NFSVOPLOCK(vp, LK_EXCLUSIVE) == 0) {
4135 		error = nfsvno_setattr(vp, &nva, NFSFPCRED(sf->nsf_fp), p,
4136 		    NULL);
4137 		NFSVOPUNLOCK(vp, 0);
4138 	} else
4139 		error = EPERM;
4140 	vn_finished_write(mp);
4141 	if (!error)
4142 	    error = NFSD_RDWR(UIO_WRITE, vp,
4143 		(caddr_t)&sf->nsf_rec, sizeof (struct nfsf_rec), (off_t)0,
4144 		UIO_SYSSPACE, IO_SYNC, NFSFPCRED(sf->nsf_fp), NULL, p);
4145 	if (!error)
4146 	    error = NFSD_RDWR(UIO_WRITE, vp,
4147 		(caddr_t)sf->nsf_bootvals,
4148 		sf->nsf_numboots * sizeof (time_t),
4149 		(off_t)(sizeof (struct nfsf_rec)),
4150 		UIO_SYSSPACE, IO_SYNC, NFSFPCRED(sf->nsf_fp), NULL, p);
4151 	free((caddr_t)sf->nsf_bootvals, M_TEMP);
4152 	sf->nsf_bootvals = NULL;
4153 	if (error) {
4154 		sf->nsf_flags &= ~NFSNSF_OK;
4155 		printf("EEK! Can't write NfsV4 stable storage file\n");
4156 		return;
4157 	}
4158 	sf->nsf_flags |= NFSNSF_OK;
4159 
4160 	/*
4161 	 * Loop through the list and write out timestamp records for
4162 	 * any clients that successfully reclaimed state.
4163 	 */
4164 	LIST_FOREACH_SAFE(sp, &sf->nsf_head, nst_list, nsp) {
4165 		if (sp->nst_flag & NFSNST_GOTSTATE) {
4166 			nfsrv_writestable(sp->nst_client, sp->nst_len,
4167 				NFSNST_NEWSTATE, p);
4168 			sp->nst_clp->lc_flags |= LCL_STAMPEDSTABLE;
4169 		}
4170 		LIST_REMOVE(sp, nst_list);
4171 		free((caddr_t)sp, M_TEMP);
4172 	}
4173 	nfsrv_backupstable();
4174 }
4175 
4176 /*
4177  * Append a record to the stable storage file.
4178  */
4179 APPLESTATIC void
4180 nfsrv_writestable(u_char *client, int len, int flag, NFSPROC_T *p)
4181 {
4182 	struct nfsrv_stablefirst *sf = &nfsrv_stablefirst;
4183 	struct nfst_rec *sp;
4184 	int error;
4185 
4186 	if (!(sf->nsf_flags & NFSNSF_OK) || sf->nsf_fp == NULL)
4187 		return;
4188 	sp = (struct nfst_rec *)malloc(sizeof (struct nfst_rec) +
4189 		len - 1, M_TEMP, M_WAITOK);
4190 	sp->len = len;
4191 	NFSBCOPY(client, sp->client, len);
4192 	sp->flag = flag;
4193 	error = NFSD_RDWR(UIO_WRITE, NFSFPVNODE(sf->nsf_fp),
4194 	    (caddr_t)sp, sizeof (struct nfst_rec) + len - 1, (off_t)0,
4195 	    UIO_SYSSPACE, (IO_SYNC | IO_APPEND), NFSFPCRED(sf->nsf_fp), NULL, p);
4196 	free((caddr_t)sp, M_TEMP);
4197 	if (error) {
4198 		sf->nsf_flags &= ~NFSNSF_OK;
4199 		printf("EEK! Can't write NfsV4 stable storage file\n");
4200 	}
4201 }
4202 
4203 /*
4204  * This function is called during the grace period to mark a client
4205  * that successfully reclaimed state.
4206  */
4207 static void
4208 nfsrv_markstable(struct nfsclient *clp)
4209 {
4210 	struct nfsrv_stable *sp;
4211 
4212 	/*
4213 	 * First find the client structure.
4214 	 */
4215 	LIST_FOREACH(sp, &nfsrv_stablefirst.nsf_head, nst_list) {
4216 		if (sp->nst_len == clp->lc_idlen &&
4217 		    !NFSBCMP(sp->nst_client, clp->lc_id, sp->nst_len))
4218 			break;
4219 	}
4220 	if (sp == LIST_END(&nfsrv_stablefirst.nsf_head))
4221 		return;
4222 
4223 	/*
4224 	 * Now, just mark it and set the nfsclient back pointer.
4225 	 */
4226 	sp->nst_flag |= NFSNST_GOTSTATE;
4227 	sp->nst_clp = clp;
4228 }
4229 
4230 /*
4231  * This function is called for a reclaim, to see if it gets grace.
4232  * It returns 0 if a reclaim is allowed, 1 otherwise.
4233  */
4234 static int
4235 nfsrv_checkstable(struct nfsclient *clp)
4236 {
4237 	struct nfsrv_stable *sp;
4238 
4239 	/*
4240 	 * First, find the entry for the client.
4241 	 */
4242 	LIST_FOREACH(sp, &nfsrv_stablefirst.nsf_head, nst_list) {
4243 		if (sp->nst_len == clp->lc_idlen &&
4244 		    !NFSBCMP(sp->nst_client, clp->lc_id, sp->nst_len))
4245 			break;
4246 	}
4247 
4248 	/*
4249 	 * If not in the list, state was revoked or no state was issued
4250 	 * since the previous reboot, a reclaim is denied.
4251 	 */
4252 	if (sp == LIST_END(&nfsrv_stablefirst.nsf_head) ||
4253 	    (sp->nst_flag & NFSNST_REVOKE) ||
4254 	    !(nfsrv_stablefirst.nsf_flags & NFSNSF_OK))
4255 		return (1);
4256 	return (0);
4257 }
4258 
4259 /*
4260  * Test for and try to clear out a conflicting client. This is called by
4261  * nfsrv_lockctrl() and nfsrv_openctrl() when conflicts with other clients
4262  * a found.
4263  * The trick here is that it can't revoke a conflicting client with an
4264  * expired lease unless it holds the v4root lock, so...
4265  * If no v4root lock, get the lock and return 1 to indicate "try again".
4266  * Return 0 to indicate the conflict can't be revoked and 1 to indicate
4267  * the revocation worked and the conflicting client is "bye, bye", so it
4268  * can be tried again.
4269  * Return 2 to indicate that the vnode is VI_DOOMED after NFSVOPLOCK().
4270  * Unlocks State before a non-zero value is returned.
4271  */
4272 static int
4273 nfsrv_clientconflict(struct nfsclient *clp, int *haslockp, vnode_t vp,
4274     NFSPROC_T *p)
4275 {
4276 	int gotlock, lktype;
4277 
4278 	/*
4279 	 * If lease hasn't expired, we can't fix it.
4280 	 */
4281 	if (clp->lc_expiry >= NFSD_MONOSEC ||
4282 	    !(nfsrv_stablefirst.nsf_flags & NFSNSF_UPDATEDONE))
4283 		return (0);
4284 	if (*haslockp == 0) {
4285 		NFSUNLOCKSTATE();
4286 		lktype = NFSVOPISLOCKED(vp);
4287 		NFSVOPUNLOCK(vp, 0);
4288 		NFSLOCKV4ROOTMUTEX();
4289 		nfsv4_relref(&nfsv4rootfs_lock);
4290 		do {
4291 			gotlock = nfsv4_lock(&nfsv4rootfs_lock, 1, NULL,
4292 			    NFSV4ROOTLOCKMUTEXPTR, NULL);
4293 		} while (!gotlock);
4294 		NFSUNLOCKV4ROOTMUTEX();
4295 		*haslockp = 1;
4296 		NFSVOPLOCK(vp, lktype | LK_RETRY);
4297 		if ((vp->v_iflag & VI_DOOMED) != 0)
4298 			return (2);
4299 		else
4300 			return (1);
4301 	}
4302 	NFSUNLOCKSTATE();
4303 
4304 	/*
4305 	 * Ok, we can expire the conflicting client.
4306 	 */
4307 	nfsrv_writestable(clp->lc_id, clp->lc_idlen, NFSNST_REVOKE, p);
4308 	nfsrv_backupstable();
4309 	nfsrv_cleanclient(clp, p);
4310 	nfsrv_freedeleglist(&clp->lc_deleg);
4311 	nfsrv_freedeleglist(&clp->lc_olddeleg);
4312 	LIST_REMOVE(clp, lc_hash);
4313 	nfsrv_zapclient(clp, p);
4314 	return (1);
4315 }
4316 
4317 /*
4318  * Resolve a delegation conflict.
4319  * Returns 0 to indicate the conflict was resolved without sleeping.
4320  * Return -1 to indicate that the caller should check for conflicts again.
4321  * Return > 0 for an error that should be returned, normally NFSERR_DELAY.
4322  *
4323  * Also, manipulate the nfsv4root_lock, as required. It isn't changed
4324  * for a return of 0, since there was no sleep and it could be required
4325  * later. It is released for a return of NFSERR_DELAY, since the caller
4326  * will return that error. It is released when a sleep was done waiting
4327  * for the delegation to be returned or expire (so that other nfsds can
4328  * handle ops). Then, it must be acquired for the write to stable storage.
4329  * (This function is somewhat similar to nfsrv_clientconflict(), but
4330  *  the semantics differ in a couple of subtle ways. The return of 0
4331  *  indicates the conflict was resolved without sleeping here, not
4332  *  that the conflict can't be resolved and the handling of nfsv4root_lock
4333  *  differs, as noted above.)
4334  * Unlocks State before returning a non-zero value.
4335  */
4336 static int
4337 nfsrv_delegconflict(struct nfsstate *stp, int *haslockp, NFSPROC_T *p,
4338     vnode_t vp)
4339 {
4340 	struct nfsclient *clp = stp->ls_clp;
4341 	int gotlock, error, lktype, retrycnt, zapped_clp;
4342 	nfsv4stateid_t tstateid;
4343 	fhandle_t tfh;
4344 
4345 	/*
4346 	 * If the conflict is with an old delegation...
4347 	 */
4348 	if (stp->ls_flags & NFSLCK_OLDDELEG) {
4349 		/*
4350 		 * You can delete it, if it has expired.
4351 		 */
4352 		if (clp->lc_delegtime < NFSD_MONOSEC) {
4353 			nfsrv_freedeleg(stp);
4354 			NFSUNLOCKSTATE();
4355 			error = -1;
4356 			goto out;
4357 		}
4358 		NFSUNLOCKSTATE();
4359 		/*
4360 		 * During this delay, the old delegation could expire or it
4361 		 * could be recovered by the client via an Open with
4362 		 * CLAIM_DELEGATE_PREV.
4363 		 * Release the nfsv4root_lock, if held.
4364 		 */
4365 		if (*haslockp) {
4366 			*haslockp = 0;
4367 			NFSLOCKV4ROOTMUTEX();
4368 			nfsv4_unlock(&nfsv4rootfs_lock, 1);
4369 			NFSUNLOCKV4ROOTMUTEX();
4370 		}
4371 		error = NFSERR_DELAY;
4372 		goto out;
4373 	}
4374 
4375 	/*
4376 	 * It's a current delegation, so:
4377 	 * - check to see if the delegation has expired
4378 	 *   - if so, get the v4root lock and then expire it
4379 	 */
4380 	if (!(stp->ls_flags & NFSLCK_DELEGRECALL)) {
4381 		/*
4382 		 * - do a recall callback, since not yet done
4383 		 * For now, never allow truncate to be set. To use
4384 		 * truncate safely, it must be guaranteed that the
4385 		 * Remove, Rename or Setattr with size of 0 will
4386 		 * succeed and that would require major changes to
4387 		 * the VFS/Vnode OPs.
4388 		 * Set the expiry time large enough so that it won't expire
4389 		 * until after the callback, then set it correctly, once
4390 		 * the callback is done. (The delegation will now time
4391 		 * out whether or not the Recall worked ok. The timeout
4392 		 * will be extended when ops are done on the delegation
4393 		 * stateid, up to the timelimit.)
4394 		 */
4395 		stp->ls_delegtime = NFSD_MONOSEC + (2 * nfsrv_lease) +
4396 		    NFSRV_LEASEDELTA;
4397 		stp->ls_delegtimelimit = NFSD_MONOSEC + (6 * nfsrv_lease) +
4398 		    NFSRV_LEASEDELTA;
4399 		stp->ls_flags |= NFSLCK_DELEGRECALL;
4400 
4401 		/*
4402 		 * Loop NFSRV_CBRETRYCNT times while the CBRecall replies
4403 		 * NFSERR_BADSTATEID or NFSERR_BADHANDLE. This is done
4404 		 * in order to try and avoid a race that could happen
4405 		 * when a CBRecall request passed the Open reply with
4406 		 * the delegation in it when transitting the network.
4407 		 * Since nfsrv_docallback will sleep, don't use stp after
4408 		 * the call.
4409 		 */
4410 		NFSBCOPY((caddr_t)&stp->ls_stateid, (caddr_t)&tstateid,
4411 		    sizeof (tstateid));
4412 		NFSBCOPY((caddr_t)&stp->ls_lfp->lf_fh, (caddr_t)&tfh,
4413 		    sizeof (tfh));
4414 		NFSUNLOCKSTATE();
4415 		if (*haslockp) {
4416 			*haslockp = 0;
4417 			NFSLOCKV4ROOTMUTEX();
4418 			nfsv4_unlock(&nfsv4rootfs_lock, 1);
4419 			NFSUNLOCKV4ROOTMUTEX();
4420 		}
4421 		retrycnt = 0;
4422 		do {
4423 		    error = nfsrv_docallback(clp, NFSV4OP_CBRECALL,
4424 			&tstateid, 0, &tfh, NULL, NULL, p);
4425 		    retrycnt++;
4426 		} while ((error == NFSERR_BADSTATEID ||
4427 		    error == NFSERR_BADHANDLE) && retrycnt < NFSV4_CBRETRYCNT);
4428 		error = NFSERR_DELAY;
4429 		goto out;
4430 	}
4431 
4432 	if (clp->lc_expiry >= NFSD_MONOSEC &&
4433 	    stp->ls_delegtime >= NFSD_MONOSEC) {
4434 		NFSUNLOCKSTATE();
4435 		/*
4436 		 * A recall has been done, but it has not yet expired.
4437 		 * So, RETURN_DELAY.
4438 		 */
4439 		if (*haslockp) {
4440 			*haslockp = 0;
4441 			NFSLOCKV4ROOTMUTEX();
4442 			nfsv4_unlock(&nfsv4rootfs_lock, 1);
4443 			NFSUNLOCKV4ROOTMUTEX();
4444 		}
4445 		error = NFSERR_DELAY;
4446 		goto out;
4447 	}
4448 
4449 	/*
4450 	 * If we don't yet have the lock, just get it and then return,
4451 	 * since we need that before deleting expired state, such as
4452 	 * this delegation.
4453 	 * When getting the lock, unlock the vnode, so other nfsds that
4454 	 * are in progress, won't get stuck waiting for the vnode lock.
4455 	 */
4456 	if (*haslockp == 0) {
4457 		NFSUNLOCKSTATE();
4458 		lktype = NFSVOPISLOCKED(vp);
4459 		NFSVOPUNLOCK(vp, 0);
4460 		NFSLOCKV4ROOTMUTEX();
4461 		nfsv4_relref(&nfsv4rootfs_lock);
4462 		do {
4463 			gotlock = nfsv4_lock(&nfsv4rootfs_lock, 1, NULL,
4464 			    NFSV4ROOTLOCKMUTEXPTR, NULL);
4465 		} while (!gotlock);
4466 		NFSUNLOCKV4ROOTMUTEX();
4467 		*haslockp = 1;
4468 		NFSVOPLOCK(vp, lktype | LK_RETRY);
4469 		if ((vp->v_iflag & VI_DOOMED) != 0) {
4470 			*haslockp = 0;
4471 			NFSLOCKV4ROOTMUTEX();
4472 			nfsv4_unlock(&nfsv4rootfs_lock, 1);
4473 			NFSUNLOCKV4ROOTMUTEX();
4474 			error = NFSERR_PERM;
4475 			goto out;
4476 		}
4477 		error = -1;
4478 		goto out;
4479 	}
4480 
4481 	NFSUNLOCKSTATE();
4482 	/*
4483 	 * Ok, we can delete the expired delegation.
4484 	 * First, write the Revoke record to stable storage and then
4485 	 * clear out the conflict.
4486 	 * Since all other nfsd threads are now blocked, we can safely
4487 	 * sleep without the state changing.
4488 	 */
4489 	nfsrv_writestable(clp->lc_id, clp->lc_idlen, NFSNST_REVOKE, p);
4490 	nfsrv_backupstable();
4491 	if (clp->lc_expiry < NFSD_MONOSEC) {
4492 		nfsrv_cleanclient(clp, p);
4493 		nfsrv_freedeleglist(&clp->lc_deleg);
4494 		nfsrv_freedeleglist(&clp->lc_olddeleg);
4495 		LIST_REMOVE(clp, lc_hash);
4496 		zapped_clp = 1;
4497 	} else {
4498 		nfsrv_freedeleg(stp);
4499 		zapped_clp = 0;
4500 	}
4501 	if (zapped_clp)
4502 		nfsrv_zapclient(clp, p);
4503 	error = -1;
4504 
4505 out:
4506 	NFSEXITCODE(error);
4507 	return (error);
4508 }
4509 
4510 /*
4511  * Check for a remove allowed, if remove is set to 1 and get rid of
4512  * delegations.
4513  */
4514 APPLESTATIC int
4515 nfsrv_checkremove(vnode_t vp, int remove, NFSPROC_T *p)
4516 {
4517 	struct nfsstate *stp;
4518 	struct nfslockfile *lfp;
4519 	int error, haslock = 0;
4520 	fhandle_t nfh;
4521 
4522 	/*
4523 	 * First, get the lock file structure.
4524 	 * (A return of -1 means no associated state, so remove ok.)
4525 	 */
4526 	error = nfsrv_getlockfh(vp, NFSLCK_CHECK, NULL, &nfh, p);
4527 tryagain:
4528 	NFSLOCKSTATE();
4529 	if (!error)
4530 		error = nfsrv_getlockfile(NFSLCK_CHECK, NULL, &lfp, &nfh, 0);
4531 	if (error) {
4532 		NFSUNLOCKSTATE();
4533 		if (haslock) {
4534 			NFSLOCKV4ROOTMUTEX();
4535 			nfsv4_unlock(&nfsv4rootfs_lock, 1);
4536 			NFSUNLOCKV4ROOTMUTEX();
4537 		}
4538 		if (error == -1)
4539 			error = 0;
4540 		goto out;
4541 	}
4542 
4543 	/*
4544 	 * Now, we must Recall any delegations.
4545 	 */
4546 	error = nfsrv_cleandeleg(vp, lfp, NULL, &haslock, p);
4547 	if (error) {
4548 		/*
4549 		 * nfsrv_cleandeleg() unlocks state for non-zero
4550 		 * return.
4551 		 */
4552 		if (error == -1)
4553 			goto tryagain;
4554 		if (haslock) {
4555 			NFSLOCKV4ROOTMUTEX();
4556 			nfsv4_unlock(&nfsv4rootfs_lock, 1);
4557 			NFSUNLOCKV4ROOTMUTEX();
4558 		}
4559 		goto out;
4560 	}
4561 
4562 	/*
4563 	 * Now, look for a conflicting open share.
4564 	 */
4565 	if (remove) {
4566 		LIST_FOREACH(stp, &lfp->lf_open, ls_file) {
4567 			if (stp->ls_flags & NFSLCK_WRITEDENY) {
4568 				error = NFSERR_FILEOPEN;
4569 				break;
4570 			}
4571 		}
4572 	}
4573 
4574 	NFSUNLOCKSTATE();
4575 	if (haslock) {
4576 		NFSLOCKV4ROOTMUTEX();
4577 		nfsv4_unlock(&nfsv4rootfs_lock, 1);
4578 		NFSUNLOCKV4ROOTMUTEX();
4579 	}
4580 
4581 out:
4582 	NFSEXITCODE(error);
4583 	return (error);
4584 }
4585 
4586 /*
4587  * Clear out all delegations for the file referred to by lfp.
4588  * May return NFSERR_DELAY, if there will be a delay waiting for
4589  * delegations to expire.
4590  * Returns -1 to indicate it slept while recalling a delegation.
4591  * This function has the side effect of deleting the nfslockfile structure,
4592  * if it no longer has associated state and didn't have to sleep.
4593  * Unlocks State before a non-zero value is returned.
4594  */
4595 static int
4596 nfsrv_cleandeleg(vnode_t vp, struct nfslockfile *lfp,
4597     struct nfsclient *clp, int *haslockp, NFSPROC_T *p)
4598 {
4599 	struct nfsstate *stp, *nstp;
4600 	int ret = 0;
4601 
4602 	stp = LIST_FIRST(&lfp->lf_deleg);
4603 	while (stp != LIST_END(&lfp->lf_deleg)) {
4604 		nstp = LIST_NEXT(stp, ls_file);
4605 		if (stp->ls_clp != clp) {
4606 			ret = nfsrv_delegconflict(stp, haslockp, p, vp);
4607 			if (ret) {
4608 				/*
4609 				 * nfsrv_delegconflict() unlocks state
4610 				 * when it returns non-zero.
4611 				 */
4612 				goto out;
4613 			}
4614 		}
4615 		stp = nstp;
4616 	}
4617 out:
4618 	NFSEXITCODE(ret);
4619 	return (ret);
4620 }
4621 
4622 /*
4623  * There are certain operations that, when being done outside of NFSv4,
4624  * require that any NFSv4 delegation for the file be recalled.
4625  * This function is to be called for those cases:
4626  * VOP_RENAME() - When a delegation is being recalled for any reason,
4627  *	the client may have to do Opens against the server, using the file's
4628  *	final component name. If the file has been renamed on the server,
4629  *	that component name will be incorrect and the Open will fail.
4630  * VOP_REMOVE() - Theoretically, a client could Open a file after it has
4631  *	been removed on the server, if there is a delegation issued to
4632  *	that client for the file. I say "theoretically" since clients
4633  *	normally do an Access Op before the Open and that Access Op will
4634  *	fail with ESTALE. Note that NFSv2 and 3 don't even do Opens, so
4635  *	they will detect the file's removal in the same manner. (There is
4636  *	one case where RFC3530 allows a client to do an Open without first
4637  *	doing an Access Op, which is passage of a check against the ACE
4638  *	returned with a Write delegation, but current practice is to ignore
4639  *	the ACE and always do an Access Op.)
4640  *	Since the functions can only be called with an unlocked vnode, this
4641  *	can't be done at this time.
4642  * VOP_ADVLOCK() - When a client holds a delegation, it can issue byte range
4643  *	locks locally in the client, which are not visible to the server. To
4644  *	deal with this, issuing of delegations for a vnode must be disabled
4645  *	and all delegations for the vnode recalled. This is done via the
4646  *	second function, using the VV_DISABLEDELEG vflag on the vnode.
4647  */
4648 APPLESTATIC void
4649 nfsd_recalldelegation(vnode_t vp, NFSPROC_T *p)
4650 {
4651 	time_t starttime;
4652 	int error;
4653 
4654 	/*
4655 	 * First, check to see if the server is currently running and it has
4656 	 * been called for a regular file when issuing delegations.
4657 	 */
4658 	if (newnfs_numnfsd == 0 || vp->v_type != VREG ||
4659 	    nfsrv_issuedelegs == 0)
4660 		return;
4661 
4662 	KASSERT((NFSVOPISLOCKED(vp) != LK_EXCLUSIVE), ("vp %p is locked", vp));
4663 	/*
4664 	 * First, get a reference on the nfsv4rootfs_lock so that an
4665 	 * exclusive lock cannot be acquired by another thread.
4666 	 */
4667 	NFSLOCKV4ROOTMUTEX();
4668 	nfsv4_getref(&nfsv4rootfs_lock, NULL, NFSV4ROOTLOCKMUTEXPTR, NULL);
4669 	NFSUNLOCKV4ROOTMUTEX();
4670 
4671 	/*
4672 	 * Now, call nfsrv_checkremove() in a loop while it returns
4673 	 * NFSERR_DELAY. Return upon any other error or when timed out.
4674 	 */
4675 	starttime = NFSD_MONOSEC;
4676 	do {
4677 		if (NFSVOPLOCK(vp, LK_EXCLUSIVE) == 0) {
4678 			error = nfsrv_checkremove(vp, 0, p);
4679 			NFSVOPUNLOCK(vp, 0);
4680 		} else
4681 			error = EPERM;
4682 		if (error == NFSERR_DELAY) {
4683 			if (NFSD_MONOSEC - starttime > NFS_REMOVETIMEO)
4684 				break;
4685 			/* Sleep for a short period of time */
4686 			(void) nfs_catnap(PZERO, 0, "nfsremove");
4687 		}
4688 	} while (error == NFSERR_DELAY);
4689 	NFSLOCKV4ROOTMUTEX();
4690 	nfsv4_relref(&nfsv4rootfs_lock);
4691 	NFSUNLOCKV4ROOTMUTEX();
4692 }
4693 
4694 APPLESTATIC void
4695 nfsd_disabledelegation(vnode_t vp, NFSPROC_T *p)
4696 {
4697 
4698 #ifdef VV_DISABLEDELEG
4699 	/*
4700 	 * First, flag issuance of delegations disabled.
4701 	 */
4702 	atomic_set_long(&vp->v_vflag, VV_DISABLEDELEG);
4703 #endif
4704 
4705 	/*
4706 	 * Then call nfsd_recalldelegation() to get rid of all extant
4707 	 * delegations.
4708 	 */
4709 	nfsd_recalldelegation(vp, p);
4710 }
4711 
4712 /*
4713  * Check for conflicting locks, etc. and then get rid of delegations.
4714  * (At one point I thought that I should get rid of delegations for any
4715  *  Setattr, since it could potentially disallow the I/O op (read or write)
4716  *  allowed by the delegation. However, Setattr Ops that aren't changing
4717  *  the size get a stateid of all 0s, so you can't tell if it is a delegation
4718  *  for the same client or a different one, so I decided to only get rid
4719  *  of delegations for other clients when the size is being changed.)
4720  * In general, a Setattr can disable NFS I/O Ops that are outstanding, such
4721  * as Write backs, even if there is no delegation, so it really isn't any
4722  * different?)
4723  */
4724 APPLESTATIC int
4725 nfsrv_checksetattr(vnode_t vp, struct nfsrv_descript *nd,
4726     nfsv4stateid_t *stateidp, struct nfsvattr *nvap, nfsattrbit_t *attrbitp,
4727     struct nfsexstuff *exp, NFSPROC_T *p)
4728 {
4729 	struct nfsstate st, *stp = &st;
4730 	struct nfslock lo, *lop = &lo;
4731 	int error = 0;
4732 	nfsquad_t clientid;
4733 
4734 	if (NFSISSET_ATTRBIT(attrbitp, NFSATTRBIT_SIZE)) {
4735 		stp->ls_flags = (NFSLCK_CHECK | NFSLCK_WRITEACCESS);
4736 		lop->lo_first = nvap->na_size;
4737 	} else {
4738 		stp->ls_flags = 0;
4739 		lop->lo_first = 0;
4740 	}
4741 	if (NFSISSET_ATTRBIT(attrbitp, NFSATTRBIT_OWNER) ||
4742 	    NFSISSET_ATTRBIT(attrbitp, NFSATTRBIT_OWNERGROUP) ||
4743 	    NFSISSET_ATTRBIT(attrbitp, NFSATTRBIT_MODE) ||
4744 	    NFSISSET_ATTRBIT(attrbitp, NFSATTRBIT_ACL))
4745 		stp->ls_flags |= NFSLCK_SETATTR;
4746 	if (stp->ls_flags == 0)
4747 		goto out;
4748 	lop->lo_end = NFS64BITSSET;
4749 	lop->lo_flags = NFSLCK_WRITE;
4750 	stp->ls_ownerlen = 0;
4751 	stp->ls_op = NULL;
4752 	stp->ls_uid = nd->nd_cred->cr_uid;
4753 	stp->ls_stateid.seqid = stateidp->seqid;
4754 	clientid.lval[0] = stp->ls_stateid.other[0] = stateidp->other[0];
4755 	clientid.lval[1] = stp->ls_stateid.other[1] = stateidp->other[1];
4756 	stp->ls_stateid.other[2] = stateidp->other[2];
4757 	error = nfsrv_lockctrl(vp, &stp, &lop, NULL, clientid,
4758 	    stateidp, exp, nd, p);
4759 
4760 out:
4761 	NFSEXITCODE2(error, nd);
4762 	return (error);
4763 }
4764 
4765 /*
4766  * Check for a write delegation and do a CBGETATTR if there is one, updating
4767  * the attributes, as required.
4768  * Should I return an error if I can't get the attributes? (For now, I'll
4769  * just return ok.
4770  */
4771 APPLESTATIC int
4772 nfsrv_checkgetattr(struct nfsrv_descript *nd, vnode_t vp,
4773     struct nfsvattr *nvap, nfsattrbit_t *attrbitp, struct ucred *cred,
4774     NFSPROC_T *p)
4775 {
4776 	struct nfsstate *stp;
4777 	struct nfslockfile *lfp;
4778 	struct nfsclient *clp;
4779 	struct nfsvattr nva;
4780 	fhandle_t nfh;
4781 	int error = 0;
4782 	nfsattrbit_t cbbits;
4783 	u_quad_t delegfilerev;
4784 
4785 	NFSCBGETATTR_ATTRBIT(attrbitp, &cbbits);
4786 	if (!NFSNONZERO_ATTRBIT(&cbbits))
4787 		goto out;
4788 
4789 	/*
4790 	 * Get the lock file structure.
4791 	 * (A return of -1 means no associated state, so return ok.)
4792 	 */
4793 	error = nfsrv_getlockfh(vp, NFSLCK_CHECK, NULL, &nfh, p);
4794 	NFSLOCKSTATE();
4795 	if (!error)
4796 		error = nfsrv_getlockfile(NFSLCK_CHECK, NULL, &lfp, &nfh, 0);
4797 	if (error) {
4798 		NFSUNLOCKSTATE();
4799 		if (error == -1)
4800 			error = 0;
4801 		goto out;
4802 	}
4803 
4804 	/*
4805 	 * Now, look for a write delegation.
4806 	 */
4807 	LIST_FOREACH(stp, &lfp->lf_deleg, ls_file) {
4808 		if (stp->ls_flags & NFSLCK_DELEGWRITE)
4809 			break;
4810 	}
4811 	if (stp == LIST_END(&lfp->lf_deleg)) {
4812 		NFSUNLOCKSTATE();
4813 		goto out;
4814 	}
4815 	clp = stp->ls_clp;
4816 	delegfilerev = stp->ls_filerev;
4817 
4818 	/*
4819 	 * If the Write delegation was issued as a part of this Compound RPC
4820 	 * or if we have an Implied Clientid (used in a previous Op in this
4821 	 * compound) and it is the client the delegation was issued to,
4822 	 * just return ok.
4823 	 * I also assume that it is from the same client iff the network
4824 	 * host IP address is the same as the callback address. (Not
4825 	 * exactly correct by the RFC, but avoids a lot of Getattr
4826 	 * callbacks.)
4827 	 */
4828 	if (nd->nd_compref == stp->ls_compref ||
4829 	    ((nd->nd_flag & ND_IMPLIEDCLID) &&
4830 	     clp->lc_clientid.qval == nd->nd_clientid.qval) ||
4831 	     nfsaddr2_match(clp->lc_req.nr_nam, nd->nd_nam)) {
4832 		NFSUNLOCKSTATE();
4833 		goto out;
4834 	}
4835 
4836 	/*
4837 	 * We are now done with the delegation state structure,
4838 	 * so the statelock can be released and we can now tsleep().
4839 	 */
4840 
4841 	/*
4842 	 * Now, we must do the CB Getattr callback, to see if Change or Size
4843 	 * has changed.
4844 	 */
4845 	if (clp->lc_expiry >= NFSD_MONOSEC) {
4846 		NFSUNLOCKSTATE();
4847 		NFSVNO_ATTRINIT(&nva);
4848 		nva.na_filerev = NFS64BITSSET;
4849 		error = nfsrv_docallback(clp, NFSV4OP_CBGETATTR, NULL,
4850 		    0, &nfh, &nva, &cbbits, p);
4851 		if (!error) {
4852 			if ((nva.na_filerev != NFS64BITSSET &&
4853 			    nva.na_filerev > delegfilerev) ||
4854 			    (NFSVNO_ISSETSIZE(&nva) &&
4855 			     nva.na_size != nvap->na_size)) {
4856 				error = nfsvno_updfilerev(vp, nvap, cred, p);
4857 				if (NFSVNO_ISSETSIZE(&nva))
4858 					nvap->na_size = nva.na_size;
4859 			}
4860 		} else
4861 			error = 0;	/* Ignore callback errors for now. */
4862 	} else {
4863 		NFSUNLOCKSTATE();
4864 	}
4865 
4866 out:
4867 	NFSEXITCODE2(error, nd);
4868 	return (error);
4869 }
4870 
4871 /*
4872  * This function looks for openowners that haven't had any opens for
4873  * a while and throws them away. Called by an nfsd when NFSNSF_NOOPENS
4874  * is set.
4875  */
4876 APPLESTATIC void
4877 nfsrv_throwawayopens(NFSPROC_T *p)
4878 {
4879 	struct nfsclient *clp, *nclp;
4880 	struct nfsstate *stp, *nstp;
4881 	int i;
4882 
4883 	NFSLOCKSTATE();
4884 	nfsrv_stablefirst.nsf_flags &= ~NFSNSF_NOOPENS;
4885 	/*
4886 	 * For each client...
4887 	 */
4888 	for (i = 0; i < NFSCLIENTHASHSIZE; i++) {
4889 	    LIST_FOREACH_SAFE(clp, &nfsclienthash[i], lc_hash, nclp) {
4890 		LIST_FOREACH_SAFE(stp, &clp->lc_open, ls_list, nstp) {
4891 			if (LIST_EMPTY(&stp->ls_open) &&
4892 			    (stp->ls_noopens > NFSNOOPEN ||
4893 			     (nfsrv_openpluslock * 2) >
4894 			     NFSRV_V4STATELIMIT))
4895 				nfsrv_freeopenowner(stp, 0, p);
4896 		}
4897 	    }
4898 	}
4899 	NFSUNLOCKSTATE();
4900 }
4901 
4902 /*
4903  * This function checks to see if the credentials are the same.
4904  * Returns 1 for not same, 0 otherwise.
4905  */
4906 static int
4907 nfsrv_notsamecredname(struct nfsrv_descript *nd, struct nfsclient *clp)
4908 {
4909 
4910 	if (nd->nd_flag & ND_GSS) {
4911 		if (!(clp->lc_flags & LCL_GSS))
4912 			return (1);
4913 		if (clp->lc_flags & LCL_NAME) {
4914 			if (nd->nd_princlen != clp->lc_namelen ||
4915 			    NFSBCMP(nd->nd_principal, clp->lc_name,
4916 				clp->lc_namelen))
4917 				return (1);
4918 			else
4919 				return (0);
4920 		}
4921 		if (nd->nd_cred->cr_uid == clp->lc_uid)
4922 			return (0);
4923 		else
4924 			return (1);
4925 	} else if (clp->lc_flags & LCL_GSS)
4926 		return (1);
4927 	/*
4928 	 * For AUTH_SYS, allow the same uid or root. (This is underspecified
4929 	 * in RFC3530, which talks about principals, but doesn't say anything
4930 	 * about uids for AUTH_SYS.)
4931 	 */
4932 	if (nd->nd_cred->cr_uid == clp->lc_uid || nd->nd_cred->cr_uid == 0)
4933 		return (0);
4934 	else
4935 		return (1);
4936 }
4937 
4938 /*
4939  * Calculate the lease expiry time.
4940  */
4941 static time_t
4942 nfsrv_leaseexpiry(void)
4943 {
4944 
4945 	if (nfsrv_stablefirst.nsf_eograce > NFSD_MONOSEC)
4946 		return (NFSD_MONOSEC + 2 * (nfsrv_lease + NFSRV_LEASEDELTA));
4947 	return (NFSD_MONOSEC + nfsrv_lease + NFSRV_LEASEDELTA);
4948 }
4949 
4950 /*
4951  * Delay the delegation timeout as far as ls_delegtimelimit, as required.
4952  */
4953 static void
4954 nfsrv_delaydelegtimeout(struct nfsstate *stp)
4955 {
4956 
4957 	if ((stp->ls_flags & NFSLCK_DELEGRECALL) == 0)
4958 		return;
4959 
4960 	if ((stp->ls_delegtime + 15) > NFSD_MONOSEC &&
4961 	    stp->ls_delegtime < stp->ls_delegtimelimit) {
4962 		stp->ls_delegtime += nfsrv_lease;
4963 		if (stp->ls_delegtime > stp->ls_delegtimelimit)
4964 			stp->ls_delegtime = stp->ls_delegtimelimit;
4965 	}
4966 }
4967 
4968 /*
4969  * This function checks to see if there is any other state associated
4970  * with the openowner for this Open.
4971  * It returns 1 if there is no other state, 0 otherwise.
4972  */
4973 static int
4974 nfsrv_nootherstate(struct nfsstate *stp)
4975 {
4976 	struct nfsstate *tstp;
4977 
4978 	LIST_FOREACH(tstp, &stp->ls_openowner->ls_open, ls_list) {
4979 		if (tstp != stp || !LIST_EMPTY(&tstp->ls_lock))
4980 			return (0);
4981 	}
4982 	return (1);
4983 }
4984 
4985 /*
4986  * Create a list of lock deltas (changes to local byte range locking
4987  * that can be rolled back using the list) and apply the changes via
4988  * nfsvno_advlock(). Optionally, lock the list. It is expected that either
4989  * the rollback or update function will be called after this.
4990  * It returns an error (and rolls back, as required), if any nfsvno_advlock()
4991  * call fails. If it returns an error, it will unlock the list.
4992  */
4993 static int
4994 nfsrv_locallock(vnode_t vp, struct nfslockfile *lfp, int flags,
4995     uint64_t first, uint64_t end, struct nfslockconflict *cfp, NFSPROC_T *p)
4996 {
4997 	struct nfslock *lop, *nlop;
4998 	int error = 0;
4999 
5000 	/* Loop through the list of locks. */
5001 	lop = LIST_FIRST(&lfp->lf_locallock);
5002 	while (first < end && lop != NULL) {
5003 		nlop = LIST_NEXT(lop, lo_lckowner);
5004 		if (first >= lop->lo_end) {
5005 			/* not there yet */
5006 			lop = nlop;
5007 		} else if (first < lop->lo_first) {
5008 			/* new one starts before entry in list */
5009 			if (end <= lop->lo_first) {
5010 				/* no overlap between old and new */
5011 				error = nfsrv_dolocal(vp, lfp, flags,
5012 				    NFSLCK_UNLOCK, first, end, cfp, p);
5013 				if (error != 0)
5014 					break;
5015 				first = end;
5016 			} else {
5017 				/* handle fragment overlapped with new one */
5018 				error = nfsrv_dolocal(vp, lfp, flags,
5019 				    NFSLCK_UNLOCK, first, lop->lo_first, cfp,
5020 				    p);
5021 				if (error != 0)
5022 					break;
5023 				first = lop->lo_first;
5024 			}
5025 		} else {
5026 			/* new one overlaps this entry in list */
5027 			if (end <= lop->lo_end) {
5028 				/* overlaps all of new one */
5029 				error = nfsrv_dolocal(vp, lfp, flags,
5030 				    lop->lo_flags, first, end, cfp, p);
5031 				if (error != 0)
5032 					break;
5033 				first = end;
5034 			} else {
5035 				/* handle fragment overlapped with new one */
5036 				error = nfsrv_dolocal(vp, lfp, flags,
5037 				    lop->lo_flags, first, lop->lo_end, cfp, p);
5038 				if (error != 0)
5039 					break;
5040 				first = lop->lo_end;
5041 				lop = nlop;
5042 			}
5043 		}
5044 	}
5045 	if (first < end && error == 0)
5046 		/* handle fragment past end of list */
5047 		error = nfsrv_dolocal(vp, lfp, flags, NFSLCK_UNLOCK, first,
5048 		    end, cfp, p);
5049 
5050 	NFSEXITCODE(error);
5051 	return (error);
5052 }
5053 
5054 /*
5055  * Local lock unlock. Unlock all byte ranges that are no longer locked
5056  * by NFSv4. To do this, unlock any subranges of first-->end that
5057  * do not overlap with the byte ranges of any lock in the lfp->lf_lock
5058  * list. This list has all locks for the file held by other
5059  * <clientid, lockowner> tuples. The list is ordered by increasing
5060  * lo_first value, but may have entries that overlap each other, for
5061  * the case of read locks.
5062  */
5063 static void
5064 nfsrv_localunlock(vnode_t vp, struct nfslockfile *lfp, uint64_t init_first,
5065     uint64_t init_end, NFSPROC_T *p)
5066 {
5067 	struct nfslock *lop;
5068 	uint64_t first, end, prevfirst;
5069 
5070 	first = init_first;
5071 	end = init_end;
5072 	while (first < init_end) {
5073 		/* Loop through all nfs locks, adjusting first and end */
5074 		prevfirst = 0;
5075 		LIST_FOREACH(lop, &lfp->lf_lock, lo_lckfile) {
5076 			KASSERT(prevfirst <= lop->lo_first,
5077 			    ("nfsv4 locks out of order"));
5078 			KASSERT(lop->lo_first < lop->lo_end,
5079 			    ("nfsv4 bogus lock"));
5080 			prevfirst = lop->lo_first;
5081 			if (first >= lop->lo_first &&
5082 			    first < lop->lo_end)
5083 				/*
5084 				 * Overlaps with initial part, so trim
5085 				 * off that initial part by moving first past
5086 				 * it.
5087 				 */
5088 				first = lop->lo_end;
5089 			else if (end > lop->lo_first &&
5090 			    lop->lo_first > first) {
5091 				/*
5092 				 * This lock defines the end of the
5093 				 * segment to unlock, so set end to the
5094 				 * start of it and break out of the loop.
5095 				 */
5096 				end = lop->lo_first;
5097 				break;
5098 			}
5099 			if (first >= end)
5100 				/*
5101 				 * There is no segment left to do, so
5102 				 * break out of this loop and then exit
5103 				 * the outer while() since first will be set
5104 				 * to end, which must equal init_end here.
5105 				 */
5106 				break;
5107 		}
5108 		if (first < end) {
5109 			/* Unlock this segment */
5110 			(void) nfsrv_dolocal(vp, lfp, NFSLCK_UNLOCK,
5111 			    NFSLCK_READ, first, end, NULL, p);
5112 			nfsrv_locallock_commit(lfp, NFSLCK_UNLOCK,
5113 			    first, end);
5114 		}
5115 		/*
5116 		 * Now move past this segment and look for any further
5117 		 * segment in the range, if there is one.
5118 		 */
5119 		first = end;
5120 		end = init_end;
5121 	}
5122 }
5123 
5124 /*
5125  * Do the local lock operation and update the rollback list, as required.
5126  * Perform the rollback and return the error if nfsvno_advlock() fails.
5127  */
5128 static int
5129 nfsrv_dolocal(vnode_t vp, struct nfslockfile *lfp, int flags, int oldflags,
5130     uint64_t first, uint64_t end, struct nfslockconflict *cfp, NFSPROC_T *p)
5131 {
5132 	struct nfsrollback *rlp;
5133 	int error = 0, ltype, oldltype;
5134 
5135 	if (flags & NFSLCK_WRITE)
5136 		ltype = F_WRLCK;
5137 	else if (flags & NFSLCK_READ)
5138 		ltype = F_RDLCK;
5139 	else
5140 		ltype = F_UNLCK;
5141 	if (oldflags & NFSLCK_WRITE)
5142 		oldltype = F_WRLCK;
5143 	else if (oldflags & NFSLCK_READ)
5144 		oldltype = F_RDLCK;
5145 	else
5146 		oldltype = F_UNLCK;
5147 	if (ltype == oldltype || (oldltype == F_WRLCK && ltype == F_RDLCK))
5148 		/* nothing to do */
5149 		goto out;
5150 	error = nfsvno_advlock(vp, ltype, first, end, p);
5151 	if (error != 0) {
5152 		if (cfp != NULL) {
5153 			cfp->cl_clientid.lval[0] = 0;
5154 			cfp->cl_clientid.lval[1] = 0;
5155 			cfp->cl_first = 0;
5156 			cfp->cl_end = NFS64BITSSET;
5157 			cfp->cl_flags = NFSLCK_WRITE;
5158 			cfp->cl_ownerlen = 5;
5159 			NFSBCOPY("LOCAL", cfp->cl_owner, 5);
5160 		}
5161 		nfsrv_locallock_rollback(vp, lfp, p);
5162 	} else if (ltype != F_UNLCK) {
5163 		rlp = malloc(sizeof (struct nfsrollback), M_NFSDROLLBACK,
5164 		    M_WAITOK);
5165 		rlp->rlck_first = first;
5166 		rlp->rlck_end = end;
5167 		rlp->rlck_type = oldltype;
5168 		LIST_INSERT_HEAD(&lfp->lf_rollback, rlp, rlck_list);
5169 	}
5170 
5171 out:
5172 	NFSEXITCODE(error);
5173 	return (error);
5174 }
5175 
5176 /*
5177  * Roll back local lock changes and free up the rollback list.
5178  */
5179 static void
5180 nfsrv_locallock_rollback(vnode_t vp, struct nfslockfile *lfp, NFSPROC_T *p)
5181 {
5182 	struct nfsrollback *rlp, *nrlp;
5183 
5184 	LIST_FOREACH_SAFE(rlp, &lfp->lf_rollback, rlck_list, nrlp) {
5185 		(void) nfsvno_advlock(vp, rlp->rlck_type, rlp->rlck_first,
5186 		    rlp->rlck_end, p);
5187 		free(rlp, M_NFSDROLLBACK);
5188 	}
5189 	LIST_INIT(&lfp->lf_rollback);
5190 }
5191 
5192 /*
5193  * Update local lock list and delete rollback list (ie now committed to the
5194  * local locks). Most of the work is done by the internal function.
5195  */
5196 static void
5197 nfsrv_locallock_commit(struct nfslockfile *lfp, int flags, uint64_t first,
5198     uint64_t end)
5199 {
5200 	struct nfsrollback *rlp, *nrlp;
5201 	struct nfslock *new_lop, *other_lop;
5202 
5203 	new_lop = malloc(sizeof (struct nfslock), M_NFSDLOCK, M_WAITOK);
5204 	if (flags & (NFSLCK_READ | NFSLCK_WRITE))
5205 		other_lop = malloc(sizeof (struct nfslock), M_NFSDLOCK,
5206 		    M_WAITOK);
5207 	else
5208 		other_lop = NULL;
5209 	new_lop->lo_flags = flags;
5210 	new_lop->lo_first = first;
5211 	new_lop->lo_end = end;
5212 	nfsrv_updatelock(NULL, &new_lop, &other_lop, lfp);
5213 	if (new_lop != NULL)
5214 		free(new_lop, M_NFSDLOCK);
5215 	if (other_lop != NULL)
5216 		free(other_lop, M_NFSDLOCK);
5217 
5218 	/* and get rid of the rollback list */
5219 	LIST_FOREACH_SAFE(rlp, &lfp->lf_rollback, rlck_list, nrlp)
5220 		free(rlp, M_NFSDROLLBACK);
5221 	LIST_INIT(&lfp->lf_rollback);
5222 }
5223 
5224 /*
5225  * Lock the struct nfslockfile for local lock updating.
5226  */
5227 static void
5228 nfsrv_locklf(struct nfslockfile *lfp)
5229 {
5230 	int gotlock;
5231 
5232 	/* lf_usecount ensures *lfp won't be free'd */
5233 	lfp->lf_usecount++;
5234 	do {
5235 		gotlock = nfsv4_lock(&lfp->lf_locallock_lck, 1, NULL,
5236 		    NFSSTATEMUTEXPTR, NULL);
5237 	} while (gotlock == 0);
5238 	lfp->lf_usecount--;
5239 }
5240 
5241 /*
5242  * Unlock the struct nfslockfile after local lock updating.
5243  */
5244 static void
5245 nfsrv_unlocklf(struct nfslockfile *lfp)
5246 {
5247 
5248 	nfsv4_unlock(&lfp->lf_locallock_lck, 0);
5249 }
5250 
5251 /*
5252  * Clear out all state for the NFSv4 server.
5253  * Must be called by a thread that can sleep when no nfsds are running.
5254  */
5255 void
5256 nfsrv_throwawayallstate(NFSPROC_T *p)
5257 {
5258 	struct nfsclient *clp, *nclp;
5259 	struct nfslockfile *lfp, *nlfp;
5260 	int i;
5261 
5262 	/*
5263 	 * For each client, clean out the state and then free the structure.
5264 	 */
5265 	for (i = 0; i < NFSCLIENTHASHSIZE; i++) {
5266 		LIST_FOREACH_SAFE(clp, &nfsclienthash[i], lc_hash, nclp) {
5267 			nfsrv_cleanclient(clp, p);
5268 			nfsrv_freedeleglist(&clp->lc_deleg);
5269 			nfsrv_freedeleglist(&clp->lc_olddeleg);
5270 			free(clp, M_NFSDCLIENT);
5271 		}
5272 	}
5273 
5274 	/*
5275 	 * Also, free up any remaining lock file structures.
5276 	 */
5277 	for (i = 0; i < NFSLOCKHASHSIZE; i++) {
5278 		LIST_FOREACH_SAFE(lfp, &nfslockhash[i], lf_hash, nlfp) {
5279 			printf("nfsd unload: fnd a lock file struct\n");
5280 			nfsrv_freenfslockfile(lfp);
5281 		}
5282 	}
5283 }
5284 
5285