xref: /freebsd/sys/fs/nfsserver/nfs_nfsdstate.c (revision e52d92164754cbfff84767d4c6eb3cc93e8c21ae)
1 /*-
2  * SPDX-License-Identifier: BSD-2-Clause-FreeBSD
3  *
4  * Copyright (c) 2009 Rick Macklem, University of Guelph
5  * All rights reserved.
6  *
7  * Redistribution and use in source and binary forms, with or without
8  * modification, are permitted provided that the following conditions
9  * are met:
10  * 1. Redistributions of source code must retain the above copyright
11  *    notice, this list of conditions and the following disclaimer.
12  * 2. Redistributions in binary form must reproduce the above copyright
13  *    notice, this list of conditions and the following disclaimer in the
14  *    documentation and/or other materials provided with the distribution.
15  *
16  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
17  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
18  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
19  * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
20  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
21  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
22  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
23  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
24  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
25  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
26  * SUCH DAMAGE.
27  *
28  */
29 
30 #include <sys/cdefs.h>
31 __FBSDID("$FreeBSD$");
32 
33 #ifndef APPLEKEXT
34 #include <fs/nfs/nfsport.h>
35 
36 struct nfsrv_stablefirst nfsrv_stablefirst;
37 int nfsrv_issuedelegs = 0;
38 int nfsrv_dolocallocks = 0;
39 struct nfsv4lock nfsv4rootfs_lock;
40 
41 extern int newnfs_numnfsd;
42 extern struct nfsstatsv1 nfsstatsv1;
43 extern int nfsrv_lease;
44 extern struct timeval nfsboottime;
45 extern u_int32_t newnfs_true, newnfs_false;
46 extern int nfsd_debuglevel;
47 NFSV4ROOTLOCKMUTEX;
48 NFSSTATESPINLOCK;
49 
50 SYSCTL_DECL(_vfs_nfsd);
51 int	nfsrv_statehashsize = NFSSTATEHASHSIZE;
52 SYSCTL_INT(_vfs_nfsd, OID_AUTO, statehashsize, CTLFLAG_RDTUN,
53     &nfsrv_statehashsize, 0,
54     "Size of state hash table set via loader.conf");
55 
56 int	nfsrv_clienthashsize = NFSCLIENTHASHSIZE;
57 SYSCTL_INT(_vfs_nfsd, OID_AUTO, clienthashsize, CTLFLAG_RDTUN,
58     &nfsrv_clienthashsize, 0,
59     "Size of client hash table set via loader.conf");
60 
61 int	nfsrv_lockhashsize = NFSLOCKHASHSIZE;
62 SYSCTL_INT(_vfs_nfsd, OID_AUTO, fhhashsize, CTLFLAG_RDTUN,
63     &nfsrv_lockhashsize, 0,
64     "Size of file handle hash table set via loader.conf");
65 
66 int	nfsrv_sessionhashsize = NFSSESSIONHASHSIZE;
67 SYSCTL_INT(_vfs_nfsd, OID_AUTO, sessionhashsize, CTLFLAG_RDTUN,
68     &nfsrv_sessionhashsize, 0,
69     "Size of session hash table set via loader.conf");
70 
71 static int	nfsrv_v4statelimit = NFSRV_V4STATELIMIT;
72 SYSCTL_INT(_vfs_nfsd, OID_AUTO, v4statelimit, CTLFLAG_RWTUN,
73     &nfsrv_v4statelimit, 0,
74     "High water limit for NFSv4 opens+locks+delegations");
75 
76 static int	nfsrv_writedelegifpos = 0;
77 SYSCTL_INT(_vfs_nfsd, OID_AUTO, writedelegifpos, CTLFLAG_RW,
78     &nfsrv_writedelegifpos, 0,
79     "Issue a write delegation for read opens if possible");
80 
81 static int	nfsrv_allowreadforwriteopen = 1;
82 SYSCTL_INT(_vfs_nfsd, OID_AUTO, allowreadforwriteopen, CTLFLAG_RW,
83     &nfsrv_allowreadforwriteopen, 0,
84     "Allow Reads to be done with Write Access StateIDs");
85 
86 /*
87  * Hash lists for nfs V4.
88  */
89 struct nfsclienthashhead	*nfsclienthash;
90 struct nfslockhashhead		*nfslockhash;
91 struct nfssessionhash		*nfssessionhash;
92 #endif	/* !APPLEKEXT */
93 
94 static u_int32_t nfsrv_openpluslock = 0, nfsrv_delegatecnt = 0;
95 static time_t nfsrvboottime;
96 static int nfsrv_returnoldstateid = 0, nfsrv_clients = 0;
97 static int nfsrv_clienthighwater = NFSRV_CLIENTHIGHWATER;
98 static int nfsrv_nogsscallback = 0;
99 static volatile int nfsrv_writedelegcnt = 0;
100 
101 /* local functions */
102 static void nfsrv_dumpaclient(struct nfsclient *clp,
103     struct nfsd_dumpclients *dumpp);
104 static void nfsrv_freeopenowner(struct nfsstate *stp, int cansleep,
105     NFSPROC_T *p);
106 static int nfsrv_freeopen(struct nfsstate *stp, vnode_t vp, int cansleep,
107     NFSPROC_T *p);
108 static void nfsrv_freelockowner(struct nfsstate *stp, vnode_t vp, int cansleep,
109     NFSPROC_T *p);
110 static void nfsrv_freeallnfslocks(struct nfsstate *stp, vnode_t vp,
111     int cansleep, NFSPROC_T *p);
112 static void nfsrv_freenfslock(struct nfslock *lop);
113 static void nfsrv_freenfslockfile(struct nfslockfile *lfp);
114 static void nfsrv_freedeleg(struct nfsstate *);
115 static int nfsrv_getstate(struct nfsclient *clp, nfsv4stateid_t *stateidp,
116     u_int32_t flags, struct nfsstate **stpp);
117 static void nfsrv_getowner(struct nfsstatehead *hp, struct nfsstate *new_stp,
118     struct nfsstate **stpp);
119 static int nfsrv_getlockfh(vnode_t vp, u_short flags,
120     struct nfslockfile *new_lfp, fhandle_t *nfhp, NFSPROC_T *p);
121 static int nfsrv_getlockfile(u_short flags, struct nfslockfile **new_lfpp,
122     struct nfslockfile **lfpp, fhandle_t *nfhp, int lockit);
123 static void nfsrv_insertlock(struct nfslock *new_lop,
124     struct nfslock *insert_lop, struct nfsstate *stp, struct nfslockfile *lfp);
125 static void nfsrv_updatelock(struct nfsstate *stp, struct nfslock **new_lopp,
126     struct nfslock **other_lopp, struct nfslockfile *lfp);
127 static int nfsrv_getipnumber(u_char *cp);
128 static int nfsrv_checkrestart(nfsquad_t clientid, u_int32_t flags,
129     nfsv4stateid_t *stateidp, int specialid);
130 static int nfsrv_checkgrace(struct nfsrv_descript *nd, struct nfsclient *clp,
131     u_int32_t flags);
132 static int nfsrv_docallback(struct nfsclient *clp, int procnum,
133     nfsv4stateid_t *stateidp, int trunc, fhandle_t *fhp,
134     struct nfsvattr *nap, nfsattrbit_t *attrbitp, NFSPROC_T *p);
135 static int nfsrv_cbcallargs(struct nfsrv_descript *nd, struct nfsclient *clp,
136     uint32_t callback, int op, const char *optag, struct nfsdsession **sepp);
137 static u_int32_t nfsrv_nextclientindex(void);
138 static u_int32_t nfsrv_nextstateindex(struct nfsclient *clp);
139 static void nfsrv_markstable(struct nfsclient *clp);
140 static void nfsrv_markreclaim(struct nfsclient *clp);
141 static int nfsrv_checkstable(struct nfsclient *clp);
142 static int nfsrv_clientconflict(struct nfsclient *clp, int *haslockp, struct
143     vnode *vp, NFSPROC_T *p);
144 static int nfsrv_delegconflict(struct nfsstate *stp, int *haslockp,
145     NFSPROC_T *p, vnode_t vp);
146 static int nfsrv_cleandeleg(vnode_t vp, struct nfslockfile *lfp,
147     struct nfsclient *clp, int *haslockp, NFSPROC_T *p);
148 static int nfsrv_notsamecredname(struct nfsrv_descript *nd,
149     struct nfsclient *clp);
150 static time_t nfsrv_leaseexpiry(void);
151 static void nfsrv_delaydelegtimeout(struct nfsstate *stp);
152 static int nfsrv_checkseqid(struct nfsrv_descript *nd, u_int32_t seqid,
153     struct nfsstate *stp, struct nfsrvcache *op);
154 static int nfsrv_nootherstate(struct nfsstate *stp);
155 static int nfsrv_locallock(vnode_t vp, struct nfslockfile *lfp, int flags,
156     uint64_t first, uint64_t end, struct nfslockconflict *cfp, NFSPROC_T *p);
157 static void nfsrv_localunlock(vnode_t vp, struct nfslockfile *lfp,
158     uint64_t init_first, uint64_t init_end, NFSPROC_T *p);
159 static int nfsrv_dolocal(vnode_t vp, struct nfslockfile *lfp, int flags,
160     int oldflags, uint64_t first, uint64_t end, struct nfslockconflict *cfp,
161     NFSPROC_T *p);
162 static void nfsrv_locallock_rollback(vnode_t vp, struct nfslockfile *lfp,
163     NFSPROC_T *p);
164 static void nfsrv_locallock_commit(struct nfslockfile *lfp, int flags,
165     uint64_t first, uint64_t end);
166 static void nfsrv_locklf(struct nfslockfile *lfp);
167 static void nfsrv_unlocklf(struct nfslockfile *lfp);
168 static struct nfsdsession *nfsrv_findsession(uint8_t *sessionid);
169 static int nfsrv_freesession(struct nfsdsession *sep, uint8_t *sessionid);
170 static int nfsv4_setcbsequence(struct nfsrv_descript *nd, struct nfsclient *clp,
171     int dont_replycache, struct nfsdsession **sepp);
172 static int nfsv4_getcbsession(struct nfsclient *clp, struct nfsdsession **sepp);
173 
174 /*
175  * Scan the client list for a match and either return the current one,
176  * create a new entry or return an error.
177  * If returning a non-error, the clp structure must either be linked into
178  * the client list or free'd.
179  */
180 APPLESTATIC int
181 nfsrv_setclient(struct nfsrv_descript *nd, struct nfsclient **new_clpp,
182     nfsquad_t *clientidp, nfsquad_t *confirmp, NFSPROC_T *p)
183 {
184 	struct nfsclient *clp = NULL, *new_clp = *new_clpp;
185 	int i, error = 0, ret;
186 	struct nfsstate *stp, *tstp;
187 	struct sockaddr_in *sad, *rad;
188 	struct nfsdsession *sep, *nsep;
189 	int zapit = 0, gotit, hasstate = 0, igotlock;
190 	static u_int64_t confirm_index = 0;
191 
192 	/*
193 	 * Check for state resource limit exceeded.
194 	 */
195 	if (nfsrv_openpluslock > nfsrv_v4statelimit) {
196 		error = NFSERR_RESOURCE;
197 		goto out;
198 	}
199 
200 	if (nfsrv_issuedelegs == 0 ||
201 	    ((nd->nd_flag & ND_GSS) != 0 && nfsrv_nogsscallback != 0))
202 		/*
203 		 * Don't do callbacks when delegations are disabled or
204 		 * for AUTH_GSS unless enabled via nfsrv_nogsscallback.
205 		 * If establishing a callback connection is attempted
206 		 * when a firewall is blocking the callback path, the
207 		 * server may wait too long for the connect attempt to
208 		 * succeed during the Open. Some clients, such as Linux,
209 		 * may timeout and give up on the Open before the server
210 		 * replies. Also, since AUTH_GSS callbacks are not
211 		 * yet interoperability tested, they might cause the
212 		 * server to crap out, if they get past the Init call to
213 		 * the client.
214 		 */
215 		new_clp->lc_program = 0;
216 
217 	/* Lock out other nfsd threads */
218 	NFSLOCKV4ROOTMUTEX();
219 	nfsv4_relref(&nfsv4rootfs_lock);
220 	do {
221 		igotlock = nfsv4_lock(&nfsv4rootfs_lock, 1, NULL,
222 		    NFSV4ROOTLOCKMUTEXPTR, NULL);
223 	} while (!igotlock);
224 	NFSUNLOCKV4ROOTMUTEX();
225 
226 	/*
227 	 * Search for a match in the client list.
228 	 */
229 	gotit = i = 0;
230 	while (i < nfsrv_clienthashsize && !gotit) {
231 	    LIST_FOREACH(clp, &nfsclienthash[i], lc_hash) {
232 		if (new_clp->lc_idlen == clp->lc_idlen &&
233 		    !NFSBCMP(new_clp->lc_id, clp->lc_id, clp->lc_idlen)) {
234 			gotit = 1;
235 			break;
236 		}
237 	    }
238 	    if (gotit == 0)
239 		i++;
240 	}
241 	if (!gotit ||
242 	    (clp->lc_flags & (LCL_NEEDSCONFIRM | LCL_ADMINREVOKED))) {
243 		if ((nd->nd_flag & ND_NFSV41) != 0 && confirmp->lval[1] != 0) {
244 			/*
245 			 * For NFSv4.1, if confirmp->lval[1] is non-zero, the
246 			 * client is trying to update a confirmed clientid.
247 			 */
248 			NFSLOCKV4ROOTMUTEX();
249 			nfsv4_unlock(&nfsv4rootfs_lock, 1);
250 			NFSUNLOCKV4ROOTMUTEX();
251 			confirmp->lval[1] = 0;
252 			error = NFSERR_NOENT;
253 			goto out;
254 		}
255 		/*
256 		 * Get rid of the old one.
257 		 */
258 		if (i != nfsrv_clienthashsize) {
259 			LIST_REMOVE(clp, lc_hash);
260 			nfsrv_cleanclient(clp, p);
261 			nfsrv_freedeleglist(&clp->lc_deleg);
262 			nfsrv_freedeleglist(&clp->lc_olddeleg);
263 			zapit = 1;
264 		}
265 		/*
266 		 * Add it after assigning a client id to it.
267 		 */
268 		new_clp->lc_flags |= LCL_NEEDSCONFIRM;
269 		if ((nd->nd_flag & ND_NFSV41) != 0)
270 			new_clp->lc_confirm.lval[0] = confirmp->lval[0] =
271 			    ++confirm_index;
272 		else
273 			confirmp->qval = new_clp->lc_confirm.qval =
274 			    ++confirm_index;
275 		clientidp->lval[0] = new_clp->lc_clientid.lval[0] =
276 		    (u_int32_t)nfsrvboottime;
277 		clientidp->lval[1] = new_clp->lc_clientid.lval[1] =
278 		    nfsrv_nextclientindex();
279 		new_clp->lc_stateindex = 0;
280 		new_clp->lc_statemaxindex = 0;
281 		new_clp->lc_cbref = 0;
282 		new_clp->lc_expiry = nfsrv_leaseexpiry();
283 		LIST_INIT(&new_clp->lc_open);
284 		LIST_INIT(&new_clp->lc_deleg);
285 		LIST_INIT(&new_clp->lc_olddeleg);
286 		LIST_INIT(&new_clp->lc_session);
287 		for (i = 0; i < nfsrv_statehashsize; i++)
288 			LIST_INIT(&new_clp->lc_stateid[i]);
289 		LIST_INSERT_HEAD(NFSCLIENTHASH(new_clp->lc_clientid), new_clp,
290 		    lc_hash);
291 		nfsstatsv1.srvclients++;
292 		nfsrv_openpluslock++;
293 		nfsrv_clients++;
294 		NFSLOCKV4ROOTMUTEX();
295 		nfsv4_unlock(&nfsv4rootfs_lock, 1);
296 		NFSUNLOCKV4ROOTMUTEX();
297 		if (zapit)
298 			nfsrv_zapclient(clp, p);
299 		*new_clpp = NULL;
300 		goto out;
301 	}
302 
303 	/*
304 	 * Now, handle the cases where the id is already issued.
305 	 */
306 	if (nfsrv_notsamecredname(nd, clp)) {
307 	    /*
308 	     * Check to see if there is expired state that should go away.
309 	     */
310 	    if (clp->lc_expiry < NFSD_MONOSEC &&
311 	        (!LIST_EMPTY(&clp->lc_open) || !LIST_EMPTY(&clp->lc_deleg))) {
312 		nfsrv_cleanclient(clp, p);
313 		nfsrv_freedeleglist(&clp->lc_deleg);
314 	    }
315 
316 	    /*
317 	     * If there is outstanding state, then reply NFSERR_CLIDINUSE per
318 	     * RFC3530 Sec. 8.1.2 last para.
319 	     */
320 	    if (!LIST_EMPTY(&clp->lc_deleg)) {
321 		hasstate = 1;
322 	    } else if (LIST_EMPTY(&clp->lc_open)) {
323 		hasstate = 0;
324 	    } else {
325 		hasstate = 0;
326 		/* Look for an Open on the OpenOwner */
327 		LIST_FOREACH(stp, &clp->lc_open, ls_list) {
328 		    if (!LIST_EMPTY(&stp->ls_open)) {
329 			hasstate = 1;
330 			break;
331 		    }
332 		}
333 	    }
334 	    if (hasstate) {
335 		/*
336 		 * If the uid doesn't match, return NFSERR_CLIDINUSE after
337 		 * filling out the correct ipaddr and portnum.
338 		 */
339 		sad = NFSSOCKADDR(new_clp->lc_req.nr_nam, struct sockaddr_in *);
340 		rad = NFSSOCKADDR(clp->lc_req.nr_nam, struct sockaddr_in *);
341 		sad->sin_addr.s_addr = rad->sin_addr.s_addr;
342 		sad->sin_port = rad->sin_port;
343 		NFSLOCKV4ROOTMUTEX();
344 		nfsv4_unlock(&nfsv4rootfs_lock, 1);
345 		NFSUNLOCKV4ROOTMUTEX();
346 		error = NFSERR_CLIDINUSE;
347 		goto out;
348 	    }
349 	}
350 
351 	if (NFSBCMP(new_clp->lc_verf, clp->lc_verf, NFSX_VERF)) {
352 		/*
353 		 * If the verifier has changed, the client has rebooted
354 		 * and a new client id is issued. The old state info
355 		 * can be thrown away once the SETCLIENTID_CONFIRM occurs.
356 		 */
357 		LIST_REMOVE(clp, lc_hash);
358 
359 		/* Get rid of all sessions on this clientid. */
360 		LIST_FOREACH_SAFE(sep, &clp->lc_session, sess_list, nsep) {
361 			ret = nfsrv_freesession(sep, NULL);
362 			if (ret != 0)
363 				printf("nfsrv_setclient: verifier changed free"
364 				    " session failed=%d\n", ret);
365 		}
366 
367 		new_clp->lc_flags |= LCL_NEEDSCONFIRM;
368 		if ((nd->nd_flag & ND_NFSV41) != 0)
369 			new_clp->lc_confirm.lval[0] = confirmp->lval[0] =
370 			    ++confirm_index;
371 		else
372 			confirmp->qval = new_clp->lc_confirm.qval =
373 			    ++confirm_index;
374 		clientidp->lval[0] = new_clp->lc_clientid.lval[0] =
375 		    nfsrvboottime;
376 		clientidp->lval[1] = new_clp->lc_clientid.lval[1] =
377 		    nfsrv_nextclientindex();
378 		new_clp->lc_stateindex = 0;
379 		new_clp->lc_statemaxindex = 0;
380 		new_clp->lc_cbref = 0;
381 		new_clp->lc_expiry = nfsrv_leaseexpiry();
382 
383 		/*
384 		 * Save the state until confirmed.
385 		 */
386 		LIST_NEWHEAD(&new_clp->lc_open, &clp->lc_open, ls_list);
387 		LIST_FOREACH(tstp, &new_clp->lc_open, ls_list)
388 			tstp->ls_clp = new_clp;
389 		LIST_NEWHEAD(&new_clp->lc_deleg, &clp->lc_deleg, ls_list);
390 		LIST_FOREACH(tstp, &new_clp->lc_deleg, ls_list)
391 			tstp->ls_clp = new_clp;
392 		LIST_NEWHEAD(&new_clp->lc_olddeleg, &clp->lc_olddeleg,
393 		    ls_list);
394 		LIST_FOREACH(tstp, &new_clp->lc_olddeleg, ls_list)
395 			tstp->ls_clp = new_clp;
396 		for (i = 0; i < nfsrv_statehashsize; i++) {
397 			LIST_NEWHEAD(&new_clp->lc_stateid[i],
398 			    &clp->lc_stateid[i], ls_hash);
399 			LIST_FOREACH(tstp, &new_clp->lc_stateid[i], ls_hash)
400 				tstp->ls_clp = new_clp;
401 		}
402 		LIST_INIT(&new_clp->lc_session);
403 		LIST_INSERT_HEAD(NFSCLIENTHASH(new_clp->lc_clientid), new_clp,
404 		    lc_hash);
405 		nfsstatsv1.srvclients++;
406 		nfsrv_openpluslock++;
407 		nfsrv_clients++;
408 		NFSLOCKV4ROOTMUTEX();
409 		nfsv4_unlock(&nfsv4rootfs_lock, 1);
410 		NFSUNLOCKV4ROOTMUTEX();
411 
412 		/*
413 		 * Must wait until any outstanding callback on the old clp
414 		 * completes.
415 		 */
416 		NFSLOCKSTATE();
417 		while (clp->lc_cbref) {
418 			clp->lc_flags |= LCL_WAKEUPWANTED;
419 			(void)mtx_sleep(clp, NFSSTATEMUTEXPTR, PZERO - 1,
420 			    "nfsd clp", 10 * hz);
421 		}
422 		NFSUNLOCKSTATE();
423 		nfsrv_zapclient(clp, p);
424 		*new_clpp = NULL;
425 		goto out;
426 	}
427 
428 	/* For NFSv4.1, mark that we found a confirmed clientid. */
429 	if ((nd->nd_flag & ND_NFSV41) != 0) {
430 		clientidp->lval[0] = clp->lc_clientid.lval[0];
431 		clientidp->lval[1] = clp->lc_clientid.lval[1];
432 		confirmp->lval[0] = 0;	/* Ignored by client */
433 		confirmp->lval[1] = 1;
434 	} else {
435 		/*
436 		 * id and verifier match, so update the net address info
437 		 * and get rid of any existing callback authentication
438 		 * handle, so a new one will be acquired.
439 		 */
440 		LIST_REMOVE(clp, lc_hash);
441 		new_clp->lc_flags |= (LCL_NEEDSCONFIRM | LCL_DONTCLEAN);
442 		new_clp->lc_expiry = nfsrv_leaseexpiry();
443 		confirmp->qval = new_clp->lc_confirm.qval = ++confirm_index;
444 		clientidp->lval[0] = new_clp->lc_clientid.lval[0] =
445 		    clp->lc_clientid.lval[0];
446 		clientidp->lval[1] = new_clp->lc_clientid.lval[1] =
447 		    clp->lc_clientid.lval[1];
448 		new_clp->lc_delegtime = clp->lc_delegtime;
449 		new_clp->lc_stateindex = clp->lc_stateindex;
450 		new_clp->lc_statemaxindex = clp->lc_statemaxindex;
451 		new_clp->lc_cbref = 0;
452 		LIST_NEWHEAD(&new_clp->lc_open, &clp->lc_open, ls_list);
453 		LIST_FOREACH(tstp, &new_clp->lc_open, ls_list)
454 			tstp->ls_clp = new_clp;
455 		LIST_NEWHEAD(&new_clp->lc_deleg, &clp->lc_deleg, ls_list);
456 		LIST_FOREACH(tstp, &new_clp->lc_deleg, ls_list)
457 			tstp->ls_clp = new_clp;
458 		LIST_NEWHEAD(&new_clp->lc_olddeleg, &clp->lc_olddeleg, ls_list);
459 		LIST_FOREACH(tstp, &new_clp->lc_olddeleg, ls_list)
460 			tstp->ls_clp = new_clp;
461 		for (i = 0; i < nfsrv_statehashsize; i++) {
462 			LIST_NEWHEAD(&new_clp->lc_stateid[i],
463 			    &clp->lc_stateid[i], ls_hash);
464 			LIST_FOREACH(tstp, &new_clp->lc_stateid[i], ls_hash)
465 				tstp->ls_clp = new_clp;
466 		}
467 		LIST_INIT(&new_clp->lc_session);
468 		LIST_INSERT_HEAD(NFSCLIENTHASH(new_clp->lc_clientid), new_clp,
469 		    lc_hash);
470 		nfsstatsv1.srvclients++;
471 		nfsrv_openpluslock++;
472 		nfsrv_clients++;
473 	}
474 	NFSLOCKV4ROOTMUTEX();
475 	nfsv4_unlock(&nfsv4rootfs_lock, 1);
476 	NFSUNLOCKV4ROOTMUTEX();
477 
478 	if ((nd->nd_flag & ND_NFSV41) == 0) {
479 		/*
480 		 * Must wait until any outstanding callback on the old clp
481 		 * completes.
482 		 */
483 		NFSLOCKSTATE();
484 		while (clp->lc_cbref) {
485 			clp->lc_flags |= LCL_WAKEUPWANTED;
486 			(void)mtx_sleep(clp, NFSSTATEMUTEXPTR, PZERO - 1,
487 			    "nfsdclp", 10 * hz);
488 		}
489 		NFSUNLOCKSTATE();
490 		nfsrv_zapclient(clp, p);
491 		*new_clpp = NULL;
492 	}
493 
494 out:
495 	NFSEXITCODE2(error, nd);
496 	return (error);
497 }
498 
499 /*
500  * Check to see if the client id exists and optionally confirm it.
501  */
502 APPLESTATIC int
503 nfsrv_getclient(nfsquad_t clientid, int opflags, struct nfsclient **clpp,
504     struct nfsdsession *nsep, nfsquad_t confirm, uint32_t cbprogram,
505     struct nfsrv_descript *nd, NFSPROC_T *p)
506 {
507 	struct nfsclient *clp;
508 	struct nfsstate *stp;
509 	int i;
510 	struct nfsclienthashhead *hp;
511 	int error = 0, igotlock, doneok;
512 	struct nfssessionhash *shp;
513 	struct nfsdsession *sep;
514 	uint64_t sessid[2];
515 	static uint64_t next_sess = 0;
516 
517 	if (clpp)
518 		*clpp = NULL;
519 	if ((nd == NULL || (nd->nd_flag & ND_NFSV41) == 0 ||
520 	    opflags != CLOPS_RENEW) && nfsrvboottime != clientid.lval[0]) {
521 		error = NFSERR_STALECLIENTID;
522 		goto out;
523 	}
524 
525 	/*
526 	 * If called with opflags == CLOPS_RENEW, the State Lock is
527 	 * already held. Otherwise, we need to get either that or,
528 	 * for the case of Confirm, lock out the nfsd threads.
529 	 */
530 	if (opflags & CLOPS_CONFIRM) {
531 		NFSLOCKV4ROOTMUTEX();
532 		nfsv4_relref(&nfsv4rootfs_lock);
533 		do {
534 			igotlock = nfsv4_lock(&nfsv4rootfs_lock, 1, NULL,
535 			    NFSV4ROOTLOCKMUTEXPTR, NULL);
536 		} while (!igotlock);
537 		/*
538 		 * Create a new sessionid here, since we need to do it where
539 		 * there is a mutex held to serialize update of next_sess.
540 		 */
541 		if ((nd->nd_flag & ND_NFSV41) != 0) {
542 			sessid[0] = ++next_sess;
543 			sessid[1] = clientid.qval;
544 		}
545 		NFSUNLOCKV4ROOTMUTEX();
546 	} else if (opflags != CLOPS_RENEW) {
547 		NFSLOCKSTATE();
548 	}
549 
550 	/* For NFSv4.1, the clp is acquired from the associated session. */
551 	if (nd != NULL && (nd->nd_flag & ND_NFSV41) != 0 &&
552 	    opflags == CLOPS_RENEW) {
553 		clp = NULL;
554 		if ((nd->nd_flag & ND_HASSEQUENCE) != 0) {
555 			shp = NFSSESSIONHASH(nd->nd_sessionid);
556 			NFSLOCKSESSION(shp);
557 			sep = nfsrv_findsession(nd->nd_sessionid);
558 			if (sep != NULL)
559 				clp = sep->sess_clp;
560 			NFSUNLOCKSESSION(shp);
561 		}
562 	} else {
563 		hp = NFSCLIENTHASH(clientid);
564 		LIST_FOREACH(clp, hp, lc_hash) {
565 			if (clp->lc_clientid.lval[1] == clientid.lval[1])
566 				break;
567 		}
568 	}
569 	if (clp == NULL) {
570 		if (opflags & CLOPS_CONFIRM)
571 			error = NFSERR_STALECLIENTID;
572 		else
573 			error = NFSERR_EXPIRED;
574 	} else if (clp->lc_flags & LCL_ADMINREVOKED) {
575 		/*
576 		 * If marked admin revoked, just return the error.
577 		 */
578 		error = NFSERR_ADMINREVOKED;
579 	}
580 	if (error) {
581 		if (opflags & CLOPS_CONFIRM) {
582 			NFSLOCKV4ROOTMUTEX();
583 			nfsv4_unlock(&nfsv4rootfs_lock, 1);
584 			NFSUNLOCKV4ROOTMUTEX();
585 		} else if (opflags != CLOPS_RENEW) {
586 			NFSUNLOCKSTATE();
587 		}
588 		goto out;
589 	}
590 
591 	/*
592 	 * Perform any operations specified by the opflags.
593 	 */
594 	if (opflags & CLOPS_CONFIRM) {
595 		if (((nd->nd_flag & ND_NFSV41) != 0 &&
596 		     clp->lc_confirm.lval[0] != confirm.lval[0]) ||
597 		    ((nd->nd_flag & ND_NFSV41) == 0 &&
598 		     clp->lc_confirm.qval != confirm.qval))
599 			error = NFSERR_STALECLIENTID;
600 		else if (nfsrv_notsamecredname(nd, clp))
601 			error = NFSERR_CLIDINUSE;
602 
603 		if (!error) {
604 		    if ((clp->lc_flags & (LCL_NEEDSCONFIRM | LCL_DONTCLEAN)) ==
605 			LCL_NEEDSCONFIRM) {
606 			/*
607 			 * Hang onto the delegations (as old delegations)
608 			 * for an Open with CLAIM_DELEGATE_PREV unless in
609 			 * grace, but get rid of the rest of the state.
610 			 */
611 			nfsrv_cleanclient(clp, p);
612 			nfsrv_freedeleglist(&clp->lc_olddeleg);
613 			if (nfsrv_checkgrace(nd, clp, 0)) {
614 			    /* In grace, so just delete delegations */
615 			    nfsrv_freedeleglist(&clp->lc_deleg);
616 			} else {
617 			    LIST_FOREACH(stp, &clp->lc_deleg, ls_list)
618 				stp->ls_flags |= NFSLCK_OLDDELEG;
619 			    clp->lc_delegtime = NFSD_MONOSEC +
620 				nfsrv_lease + NFSRV_LEASEDELTA;
621 			    LIST_NEWHEAD(&clp->lc_olddeleg, &clp->lc_deleg,
622 				ls_list);
623 			}
624 			if ((nd->nd_flag & ND_NFSV41) != 0)
625 			    clp->lc_program = cbprogram;
626 		    }
627 		    clp->lc_flags &= ~(LCL_NEEDSCONFIRM | LCL_DONTCLEAN);
628 		    if (clp->lc_program)
629 			clp->lc_flags |= LCL_NEEDSCBNULL;
630 		    /* For NFSv4.1, link the session onto the client. */
631 		    if (nsep != NULL) {
632 			/* Hold a reference on the xprt for a backchannel. */
633 			if ((nsep->sess_crflags & NFSV4CRSESS_CONNBACKCHAN)
634 			    != 0) {
635 			    if (clp->lc_req.nr_client == NULL)
636 				clp->lc_req.nr_client = (struct __rpc_client *)
637 				    clnt_bck_create(nd->nd_xprt->xp_socket,
638 				    cbprogram, NFSV4_CBVERS);
639 			    if (clp->lc_req.nr_client != NULL) {
640 				SVC_ACQUIRE(nd->nd_xprt);
641 				nd->nd_xprt->xp_p2 =
642 				    clp->lc_req.nr_client->cl_private;
643 				/* Disable idle timeout. */
644 				nd->nd_xprt->xp_idletimeout = 0;
645 				nsep->sess_cbsess.nfsess_xprt = nd->nd_xprt;
646 			    } else
647 				nsep->sess_crflags &= ~NFSV4CRSESS_CONNBACKCHAN;
648 			}
649 			NFSBCOPY(sessid, nsep->sess_sessionid,
650 			    NFSX_V4SESSIONID);
651 			NFSBCOPY(sessid, nsep->sess_cbsess.nfsess_sessionid,
652 			    NFSX_V4SESSIONID);
653 			shp = NFSSESSIONHASH(nsep->sess_sessionid);
654 			NFSLOCKSTATE();
655 			NFSLOCKSESSION(shp);
656 			LIST_INSERT_HEAD(&shp->list, nsep, sess_hash);
657 			LIST_INSERT_HEAD(&clp->lc_session, nsep, sess_list);
658 			nsep->sess_clp = clp;
659 			NFSUNLOCKSESSION(shp);
660 			NFSUNLOCKSTATE();
661 		    }
662 		}
663 	} else if (clp->lc_flags & LCL_NEEDSCONFIRM) {
664 		error = NFSERR_EXPIRED;
665 	}
666 
667 	/*
668 	 * If called by the Renew Op, we must check the principal.
669 	 */
670 	if (!error && (opflags & CLOPS_RENEWOP)) {
671 	    if (nfsrv_notsamecredname(nd, clp)) {
672 		doneok = 0;
673 		for (i = 0; i < nfsrv_statehashsize && doneok == 0; i++) {
674 		    LIST_FOREACH(stp, &clp->lc_stateid[i], ls_hash) {
675 			if ((stp->ls_flags & NFSLCK_OPEN) &&
676 			    stp->ls_uid == nd->nd_cred->cr_uid) {
677 				doneok = 1;
678 				break;
679 			}
680 		    }
681 		}
682 		if (!doneok)
683 			error = NFSERR_ACCES;
684 	    }
685 	    if (!error && (clp->lc_flags & LCL_CBDOWN))
686 		error = NFSERR_CBPATHDOWN;
687 	}
688 	if ((!error || error == NFSERR_CBPATHDOWN) &&
689 	     (opflags & CLOPS_RENEW)) {
690 		clp->lc_expiry = nfsrv_leaseexpiry();
691 	}
692 	if (opflags & CLOPS_CONFIRM) {
693 		NFSLOCKV4ROOTMUTEX();
694 		nfsv4_unlock(&nfsv4rootfs_lock, 1);
695 		NFSUNLOCKV4ROOTMUTEX();
696 	} else if (opflags != CLOPS_RENEW) {
697 		NFSUNLOCKSTATE();
698 	}
699 	if (clpp)
700 		*clpp = clp;
701 
702 out:
703 	NFSEXITCODE2(error, nd);
704 	return (error);
705 }
706 
707 /*
708  * Perform the NFSv4.1 destroy clientid.
709  */
710 int
711 nfsrv_destroyclient(nfsquad_t clientid, NFSPROC_T *p)
712 {
713 	struct nfsclient *clp;
714 	struct nfsclienthashhead *hp;
715 	int error = 0, i, igotlock;
716 
717 	if (nfsrvboottime != clientid.lval[0]) {
718 		error = NFSERR_STALECLIENTID;
719 		goto out;
720 	}
721 
722 	/* Lock out other nfsd threads */
723 	NFSLOCKV4ROOTMUTEX();
724 	nfsv4_relref(&nfsv4rootfs_lock);
725 	do {
726 		igotlock = nfsv4_lock(&nfsv4rootfs_lock, 1, NULL,
727 		    NFSV4ROOTLOCKMUTEXPTR, NULL);
728 	} while (igotlock == 0);
729 	NFSUNLOCKV4ROOTMUTEX();
730 
731 	hp = NFSCLIENTHASH(clientid);
732 	LIST_FOREACH(clp, hp, lc_hash) {
733 		if (clp->lc_clientid.lval[1] == clientid.lval[1])
734 			break;
735 	}
736 	if (clp == NULL) {
737 		NFSLOCKV4ROOTMUTEX();
738 		nfsv4_unlock(&nfsv4rootfs_lock, 1);
739 		NFSUNLOCKV4ROOTMUTEX();
740 		/* Just return ok, since it is gone. */
741 		goto out;
742 	}
743 
744 	/* Scan for state on the clientid. */
745 	for (i = 0; i < nfsrv_statehashsize; i++)
746 		if (!LIST_EMPTY(&clp->lc_stateid[i])) {
747 			NFSLOCKV4ROOTMUTEX();
748 			nfsv4_unlock(&nfsv4rootfs_lock, 1);
749 			NFSUNLOCKV4ROOTMUTEX();
750 			error = NFSERR_CLIENTIDBUSY;
751 			goto out;
752 		}
753 	if (!LIST_EMPTY(&clp->lc_session) || !LIST_EMPTY(&clp->lc_deleg)) {
754 		NFSLOCKV4ROOTMUTEX();
755 		nfsv4_unlock(&nfsv4rootfs_lock, 1);
756 		NFSUNLOCKV4ROOTMUTEX();
757 		error = NFSERR_CLIENTIDBUSY;
758 		goto out;
759 	}
760 
761 	/* Destroy the clientid and return ok. */
762 	nfsrv_cleanclient(clp, p);
763 	nfsrv_freedeleglist(&clp->lc_deleg);
764 	nfsrv_freedeleglist(&clp->lc_olddeleg);
765 	LIST_REMOVE(clp, lc_hash);
766 	NFSLOCKV4ROOTMUTEX();
767 	nfsv4_unlock(&nfsv4rootfs_lock, 1);
768 	NFSUNLOCKV4ROOTMUTEX();
769 	nfsrv_zapclient(clp, p);
770 out:
771 	NFSEXITCODE2(error, nd);
772 	return (error);
773 }
774 
775 /*
776  * Called from the new nfssvc syscall to admin revoke a clientid.
777  * Returns 0 for success, error otherwise.
778  */
779 APPLESTATIC int
780 nfsrv_adminrevoke(struct nfsd_clid *revokep, NFSPROC_T *p)
781 {
782 	struct nfsclient *clp = NULL;
783 	int i, error = 0;
784 	int gotit, igotlock;
785 
786 	/*
787 	 * First, lock out the nfsd so that state won't change while the
788 	 * revocation record is being written to the stable storage restart
789 	 * file.
790 	 */
791 	NFSLOCKV4ROOTMUTEX();
792 	do {
793 		igotlock = nfsv4_lock(&nfsv4rootfs_lock, 1, NULL,
794 		    NFSV4ROOTLOCKMUTEXPTR, NULL);
795 	} while (!igotlock);
796 	NFSUNLOCKV4ROOTMUTEX();
797 
798 	/*
799 	 * Search for a match in the client list.
800 	 */
801 	gotit = i = 0;
802 	while (i < nfsrv_clienthashsize && !gotit) {
803 	    LIST_FOREACH(clp, &nfsclienthash[i], lc_hash) {
804 		if (revokep->nclid_idlen == clp->lc_idlen &&
805 		    !NFSBCMP(revokep->nclid_id, clp->lc_id, clp->lc_idlen)) {
806 			gotit = 1;
807 			break;
808 		}
809 	    }
810 	    i++;
811 	}
812 	if (!gotit) {
813 		NFSLOCKV4ROOTMUTEX();
814 		nfsv4_unlock(&nfsv4rootfs_lock, 0);
815 		NFSUNLOCKV4ROOTMUTEX();
816 		error = EPERM;
817 		goto out;
818 	}
819 
820 	/*
821 	 * Now, write out the revocation record
822 	 */
823 	nfsrv_writestable(clp->lc_id, clp->lc_idlen, NFSNST_REVOKE, p);
824 	nfsrv_backupstable();
825 
826 	/*
827 	 * and clear out the state, marking the clientid revoked.
828 	 */
829 	clp->lc_flags &= ~LCL_CALLBACKSON;
830 	clp->lc_flags |= LCL_ADMINREVOKED;
831 	nfsrv_cleanclient(clp, p);
832 	nfsrv_freedeleglist(&clp->lc_deleg);
833 	nfsrv_freedeleglist(&clp->lc_olddeleg);
834 	NFSLOCKV4ROOTMUTEX();
835 	nfsv4_unlock(&nfsv4rootfs_lock, 0);
836 	NFSUNLOCKV4ROOTMUTEX();
837 
838 out:
839 	NFSEXITCODE(error);
840 	return (error);
841 }
842 
843 /*
844  * Dump out stats for all clients. Called from nfssvc(2), that is used
845  * nfsstatsv1.
846  */
847 APPLESTATIC void
848 nfsrv_dumpclients(struct nfsd_dumpclients *dumpp, int maxcnt)
849 {
850 	struct nfsclient *clp;
851 	int i = 0, cnt = 0;
852 
853 	/*
854 	 * First, get a reference on the nfsv4rootfs_lock so that an
855 	 * exclusive lock cannot be acquired while dumping the clients.
856 	 */
857 	NFSLOCKV4ROOTMUTEX();
858 	nfsv4_getref(&nfsv4rootfs_lock, NULL, NFSV4ROOTLOCKMUTEXPTR, NULL);
859 	NFSUNLOCKV4ROOTMUTEX();
860 	NFSLOCKSTATE();
861 	/*
862 	 * Rattle through the client lists until done.
863 	 */
864 	while (i < nfsrv_clienthashsize && cnt < maxcnt) {
865 	    clp = LIST_FIRST(&nfsclienthash[i]);
866 	    while (clp != LIST_END(&nfsclienthash[i]) && cnt < maxcnt) {
867 		nfsrv_dumpaclient(clp, &dumpp[cnt]);
868 		cnt++;
869 		clp = LIST_NEXT(clp, lc_hash);
870 	    }
871 	    i++;
872 	}
873 	if (cnt < maxcnt)
874 	    dumpp[cnt].ndcl_clid.nclid_idlen = 0;
875 	NFSUNLOCKSTATE();
876 	NFSLOCKV4ROOTMUTEX();
877 	nfsv4_relref(&nfsv4rootfs_lock);
878 	NFSUNLOCKV4ROOTMUTEX();
879 }
880 
881 /*
882  * Dump stats for a client. Must be called with the NFSSTATELOCK and spl'd.
883  */
884 static void
885 nfsrv_dumpaclient(struct nfsclient *clp, struct nfsd_dumpclients *dumpp)
886 {
887 	struct nfsstate *stp, *openstp, *lckownstp;
888 	struct nfslock *lop;
889 	struct sockaddr *sad;
890 	struct sockaddr_in *rad;
891 	struct sockaddr_in6 *rad6;
892 
893 	dumpp->ndcl_nopenowners = dumpp->ndcl_nlockowners = 0;
894 	dumpp->ndcl_nopens = dumpp->ndcl_nlocks = 0;
895 	dumpp->ndcl_ndelegs = dumpp->ndcl_nolddelegs = 0;
896 	dumpp->ndcl_flags = clp->lc_flags;
897 	dumpp->ndcl_clid.nclid_idlen = clp->lc_idlen;
898 	NFSBCOPY(clp->lc_id, dumpp->ndcl_clid.nclid_id, clp->lc_idlen);
899 	sad = NFSSOCKADDR(clp->lc_req.nr_nam, struct sockaddr *);
900 	dumpp->ndcl_addrfam = sad->sa_family;
901 	if (sad->sa_family == AF_INET) {
902 		rad = (struct sockaddr_in *)sad;
903 		dumpp->ndcl_cbaddr.sin_addr = rad->sin_addr;
904 	} else {
905 		rad6 = (struct sockaddr_in6 *)sad;
906 		dumpp->ndcl_cbaddr.sin6_addr = rad6->sin6_addr;
907 	}
908 
909 	/*
910 	 * Now, scan the state lists and total up the opens and locks.
911 	 */
912 	LIST_FOREACH(stp, &clp->lc_open, ls_list) {
913 	    dumpp->ndcl_nopenowners++;
914 	    LIST_FOREACH(openstp, &stp->ls_open, ls_list) {
915 		dumpp->ndcl_nopens++;
916 		LIST_FOREACH(lckownstp, &openstp->ls_open, ls_list) {
917 		    dumpp->ndcl_nlockowners++;
918 		    LIST_FOREACH(lop, &lckownstp->ls_lock, lo_lckowner) {
919 			dumpp->ndcl_nlocks++;
920 		    }
921 		}
922 	    }
923 	}
924 
925 	/*
926 	 * and the delegation lists.
927 	 */
928 	LIST_FOREACH(stp, &clp->lc_deleg, ls_list) {
929 	    dumpp->ndcl_ndelegs++;
930 	}
931 	LIST_FOREACH(stp, &clp->lc_olddeleg, ls_list) {
932 	    dumpp->ndcl_nolddelegs++;
933 	}
934 }
935 
936 /*
937  * Dump out lock stats for a file.
938  */
939 APPLESTATIC void
940 nfsrv_dumplocks(vnode_t vp, struct nfsd_dumplocks *ldumpp, int maxcnt,
941     NFSPROC_T *p)
942 {
943 	struct nfsstate *stp;
944 	struct nfslock *lop;
945 	int cnt = 0;
946 	struct nfslockfile *lfp;
947 	struct sockaddr *sad;
948 	struct sockaddr_in *rad;
949 	struct sockaddr_in6 *rad6;
950 	int ret;
951 	fhandle_t nfh;
952 
953 	ret = nfsrv_getlockfh(vp, 0, NULL, &nfh, p);
954 	/*
955 	 * First, get a reference on the nfsv4rootfs_lock so that an
956 	 * exclusive lock on it cannot be acquired while dumping the locks.
957 	 */
958 	NFSLOCKV4ROOTMUTEX();
959 	nfsv4_getref(&nfsv4rootfs_lock, NULL, NFSV4ROOTLOCKMUTEXPTR, NULL);
960 	NFSUNLOCKV4ROOTMUTEX();
961 	NFSLOCKSTATE();
962 	if (!ret)
963 		ret = nfsrv_getlockfile(0, NULL, &lfp, &nfh, 0);
964 	if (ret) {
965 		ldumpp[0].ndlck_clid.nclid_idlen = 0;
966 		NFSUNLOCKSTATE();
967 		NFSLOCKV4ROOTMUTEX();
968 		nfsv4_relref(&nfsv4rootfs_lock);
969 		NFSUNLOCKV4ROOTMUTEX();
970 		return;
971 	}
972 
973 	/*
974 	 * For each open share on file, dump it out.
975 	 */
976 	stp = LIST_FIRST(&lfp->lf_open);
977 	while (stp != LIST_END(&lfp->lf_open) && cnt < maxcnt) {
978 		ldumpp[cnt].ndlck_flags = stp->ls_flags;
979 		ldumpp[cnt].ndlck_stateid.seqid = stp->ls_stateid.seqid;
980 		ldumpp[cnt].ndlck_stateid.other[0] = stp->ls_stateid.other[0];
981 		ldumpp[cnt].ndlck_stateid.other[1] = stp->ls_stateid.other[1];
982 		ldumpp[cnt].ndlck_stateid.other[2] = stp->ls_stateid.other[2];
983 		ldumpp[cnt].ndlck_owner.nclid_idlen =
984 		    stp->ls_openowner->ls_ownerlen;
985 		NFSBCOPY(stp->ls_openowner->ls_owner,
986 		    ldumpp[cnt].ndlck_owner.nclid_id,
987 		    stp->ls_openowner->ls_ownerlen);
988 		ldumpp[cnt].ndlck_clid.nclid_idlen = stp->ls_clp->lc_idlen;
989 		NFSBCOPY(stp->ls_clp->lc_id, ldumpp[cnt].ndlck_clid.nclid_id,
990 		    stp->ls_clp->lc_idlen);
991 		sad=NFSSOCKADDR(stp->ls_clp->lc_req.nr_nam, struct sockaddr *);
992 		ldumpp[cnt].ndlck_addrfam = sad->sa_family;
993 		if (sad->sa_family == AF_INET) {
994 			rad = (struct sockaddr_in *)sad;
995 			ldumpp[cnt].ndlck_cbaddr.sin_addr = rad->sin_addr;
996 		} else {
997 			rad6 = (struct sockaddr_in6 *)sad;
998 			ldumpp[cnt].ndlck_cbaddr.sin6_addr = rad6->sin6_addr;
999 		}
1000 		stp = LIST_NEXT(stp, ls_file);
1001 		cnt++;
1002 	}
1003 
1004 	/*
1005 	 * and all locks.
1006 	 */
1007 	lop = LIST_FIRST(&lfp->lf_lock);
1008 	while (lop != LIST_END(&lfp->lf_lock) && cnt < maxcnt) {
1009 		stp = lop->lo_stp;
1010 		ldumpp[cnt].ndlck_flags = lop->lo_flags;
1011 		ldumpp[cnt].ndlck_first = lop->lo_first;
1012 		ldumpp[cnt].ndlck_end = lop->lo_end;
1013 		ldumpp[cnt].ndlck_stateid.seqid = stp->ls_stateid.seqid;
1014 		ldumpp[cnt].ndlck_stateid.other[0] = stp->ls_stateid.other[0];
1015 		ldumpp[cnt].ndlck_stateid.other[1] = stp->ls_stateid.other[1];
1016 		ldumpp[cnt].ndlck_stateid.other[2] = stp->ls_stateid.other[2];
1017 		ldumpp[cnt].ndlck_owner.nclid_idlen = stp->ls_ownerlen;
1018 		NFSBCOPY(stp->ls_owner, ldumpp[cnt].ndlck_owner.nclid_id,
1019 		    stp->ls_ownerlen);
1020 		ldumpp[cnt].ndlck_clid.nclid_idlen = stp->ls_clp->lc_idlen;
1021 		NFSBCOPY(stp->ls_clp->lc_id, ldumpp[cnt].ndlck_clid.nclid_id,
1022 		    stp->ls_clp->lc_idlen);
1023 		sad=NFSSOCKADDR(stp->ls_clp->lc_req.nr_nam, struct sockaddr *);
1024 		ldumpp[cnt].ndlck_addrfam = sad->sa_family;
1025 		if (sad->sa_family == AF_INET) {
1026 			rad = (struct sockaddr_in *)sad;
1027 			ldumpp[cnt].ndlck_cbaddr.sin_addr = rad->sin_addr;
1028 		} else {
1029 			rad6 = (struct sockaddr_in6 *)sad;
1030 			ldumpp[cnt].ndlck_cbaddr.sin6_addr = rad6->sin6_addr;
1031 		}
1032 		lop = LIST_NEXT(lop, lo_lckfile);
1033 		cnt++;
1034 	}
1035 
1036 	/*
1037 	 * and the delegations.
1038 	 */
1039 	stp = LIST_FIRST(&lfp->lf_deleg);
1040 	while (stp != LIST_END(&lfp->lf_deleg) && cnt < maxcnt) {
1041 		ldumpp[cnt].ndlck_flags = stp->ls_flags;
1042 		ldumpp[cnt].ndlck_stateid.seqid = stp->ls_stateid.seqid;
1043 		ldumpp[cnt].ndlck_stateid.other[0] = stp->ls_stateid.other[0];
1044 		ldumpp[cnt].ndlck_stateid.other[1] = stp->ls_stateid.other[1];
1045 		ldumpp[cnt].ndlck_stateid.other[2] = stp->ls_stateid.other[2];
1046 		ldumpp[cnt].ndlck_owner.nclid_idlen = 0;
1047 		ldumpp[cnt].ndlck_clid.nclid_idlen = stp->ls_clp->lc_idlen;
1048 		NFSBCOPY(stp->ls_clp->lc_id, ldumpp[cnt].ndlck_clid.nclid_id,
1049 		    stp->ls_clp->lc_idlen);
1050 		sad=NFSSOCKADDR(stp->ls_clp->lc_req.nr_nam, struct sockaddr *);
1051 		ldumpp[cnt].ndlck_addrfam = sad->sa_family;
1052 		if (sad->sa_family == AF_INET) {
1053 			rad = (struct sockaddr_in *)sad;
1054 			ldumpp[cnt].ndlck_cbaddr.sin_addr = rad->sin_addr;
1055 		} else {
1056 			rad6 = (struct sockaddr_in6 *)sad;
1057 			ldumpp[cnt].ndlck_cbaddr.sin6_addr = rad6->sin6_addr;
1058 		}
1059 		stp = LIST_NEXT(stp, ls_file);
1060 		cnt++;
1061 	}
1062 
1063 	/*
1064 	 * If list isn't full, mark end of list by setting the client name
1065 	 * to zero length.
1066 	 */
1067 	if (cnt < maxcnt)
1068 		ldumpp[cnt].ndlck_clid.nclid_idlen = 0;
1069 	NFSUNLOCKSTATE();
1070 	NFSLOCKV4ROOTMUTEX();
1071 	nfsv4_relref(&nfsv4rootfs_lock);
1072 	NFSUNLOCKV4ROOTMUTEX();
1073 }
1074 
1075 /*
1076  * Server timer routine. It can scan any linked list, so long
1077  * as it holds the spin/mutex lock and there is no exclusive lock on
1078  * nfsv4rootfs_lock.
1079  * (For OpenBSD, a kthread is ok. For FreeBSD, I think it is ok
1080  *  to do this from a callout, since the spin locks work. For
1081  *  Darwin, I'm not sure what will work correctly yet.)
1082  * Should be called once per second.
1083  */
1084 APPLESTATIC void
1085 nfsrv_servertimer(void)
1086 {
1087 	struct nfsclient *clp, *nclp;
1088 	struct nfsstate *stp, *nstp;
1089 	int got_ref, i;
1090 
1091 	/*
1092 	 * Make sure nfsboottime is set. This is used by V3 as well
1093 	 * as V4. Note that nfsboottime is not nfsrvboottime, which is
1094 	 * only used by the V4 server for leases.
1095 	 */
1096 	if (nfsboottime.tv_sec == 0)
1097 		NFSSETBOOTTIME(nfsboottime);
1098 
1099 	/*
1100 	 * If server hasn't started yet, just return.
1101 	 */
1102 	NFSLOCKSTATE();
1103 	if (nfsrv_stablefirst.nsf_eograce == 0) {
1104 		NFSUNLOCKSTATE();
1105 		return;
1106 	}
1107 	if (!(nfsrv_stablefirst.nsf_flags & NFSNSF_UPDATEDONE)) {
1108 		if (!(nfsrv_stablefirst.nsf_flags & NFSNSF_GRACEOVER) &&
1109 		    NFSD_MONOSEC > nfsrv_stablefirst.nsf_eograce)
1110 			nfsrv_stablefirst.nsf_flags |=
1111 			    (NFSNSF_GRACEOVER | NFSNSF_NEEDLOCK);
1112 		NFSUNLOCKSTATE();
1113 		return;
1114 	}
1115 
1116 	/*
1117 	 * Try and get a reference count on the nfsv4rootfs_lock so that
1118 	 * no nfsd thread can acquire an exclusive lock on it before this
1119 	 * call is done. If it is already exclusively locked, just return.
1120 	 */
1121 	NFSLOCKV4ROOTMUTEX();
1122 	got_ref = nfsv4_getref_nonblock(&nfsv4rootfs_lock);
1123 	NFSUNLOCKV4ROOTMUTEX();
1124 	if (got_ref == 0) {
1125 		NFSUNLOCKSTATE();
1126 		return;
1127 	}
1128 
1129 	/*
1130 	 * For each client...
1131 	 */
1132 	for (i = 0; i < nfsrv_clienthashsize; i++) {
1133 	    clp = LIST_FIRST(&nfsclienthash[i]);
1134 	    while (clp != LIST_END(&nfsclienthash[i])) {
1135 		nclp = LIST_NEXT(clp, lc_hash);
1136 		if (!(clp->lc_flags & LCL_EXPIREIT)) {
1137 		    if (((clp->lc_expiry + NFSRV_STALELEASE) < NFSD_MONOSEC
1138 			 && ((LIST_EMPTY(&clp->lc_deleg)
1139 			      && LIST_EMPTY(&clp->lc_open)) ||
1140 			     nfsrv_clients > nfsrv_clienthighwater)) ||
1141 			(clp->lc_expiry + NFSRV_MOULDYLEASE) < NFSD_MONOSEC ||
1142 			(clp->lc_expiry < NFSD_MONOSEC &&
1143 			 (nfsrv_openpluslock * 10 / 9) > nfsrv_v4statelimit)) {
1144 			/*
1145 			 * Lease has expired several nfsrv_lease times ago:
1146 			 * PLUS
1147 			 *    - no state is associated with it
1148 			 *    OR
1149 			 *    - above high water mark for number of clients
1150 			 *      (nfsrv_clienthighwater should be large enough
1151 			 *       that this only occurs when clients fail to
1152 			 *       use the same nfs_client_id4.id. Maybe somewhat
1153 			 *       higher that the maximum number of clients that
1154 			 *       will mount this server?)
1155 			 * OR
1156 			 * Lease has expired a very long time ago
1157 			 * OR
1158 			 * Lease has expired PLUS the number of opens + locks
1159 			 * has exceeded 90% of capacity
1160 			 *
1161 			 * --> Mark for expiry. The actual expiry will be done
1162 			 *     by an nfsd sometime soon.
1163 			 */
1164 			clp->lc_flags |= LCL_EXPIREIT;
1165 			nfsrv_stablefirst.nsf_flags |=
1166 			    (NFSNSF_NEEDLOCK | NFSNSF_EXPIREDCLIENT);
1167 		    } else {
1168 			/*
1169 			 * If there are no opens, increment no open tick cnt
1170 			 * If time exceeds NFSNOOPEN, mark it to be thrown away
1171 			 * otherwise, if there is an open, reset no open time
1172 			 * Hopefully, this will avoid excessive re-creation
1173 			 * of open owners and subsequent open confirms.
1174 			 */
1175 			stp = LIST_FIRST(&clp->lc_open);
1176 			while (stp != LIST_END(&clp->lc_open)) {
1177 				nstp = LIST_NEXT(stp, ls_list);
1178 				if (LIST_EMPTY(&stp->ls_open)) {
1179 					stp->ls_noopens++;
1180 					if (stp->ls_noopens > NFSNOOPEN ||
1181 					    (nfsrv_openpluslock * 2) >
1182 					    nfsrv_v4statelimit)
1183 						nfsrv_stablefirst.nsf_flags |=
1184 							NFSNSF_NOOPENS;
1185 				} else {
1186 					stp->ls_noopens = 0;
1187 				}
1188 				stp = nstp;
1189 			}
1190 		    }
1191 		}
1192 		clp = nclp;
1193 	    }
1194 	}
1195 	NFSUNLOCKSTATE();
1196 	NFSLOCKV4ROOTMUTEX();
1197 	nfsv4_relref(&nfsv4rootfs_lock);
1198 	NFSUNLOCKV4ROOTMUTEX();
1199 }
1200 
1201 /*
1202  * The following set of functions free up the various data structures.
1203  */
1204 /*
1205  * Clear out all open/lock state related to this nfsclient.
1206  * Caller must hold an exclusive lock on nfsv4rootfs_lock, so that
1207  * there are no other active nfsd threads.
1208  */
1209 APPLESTATIC void
1210 nfsrv_cleanclient(struct nfsclient *clp, NFSPROC_T *p)
1211 {
1212 	struct nfsstate *stp, *nstp;
1213 	struct nfsdsession *sep, *nsep;
1214 
1215 	LIST_FOREACH_SAFE(stp, &clp->lc_open, ls_list, nstp)
1216 		nfsrv_freeopenowner(stp, 1, p);
1217 	if ((clp->lc_flags & LCL_ADMINREVOKED) == 0)
1218 		LIST_FOREACH_SAFE(sep, &clp->lc_session, sess_list, nsep)
1219 			(void)nfsrv_freesession(sep, NULL);
1220 }
1221 
1222 /*
1223  * Free a client that has been cleaned. It should also already have been
1224  * removed from the lists.
1225  * (Just to be safe w.r.t. newnfs_disconnect(), call this function when
1226  *  softclock interrupts are enabled.)
1227  */
1228 APPLESTATIC void
1229 nfsrv_zapclient(struct nfsclient *clp, NFSPROC_T *p)
1230 {
1231 
1232 #ifdef notyet
1233 	if ((clp->lc_flags & (LCL_GSS | LCL_CALLBACKSON)) ==
1234 	     (LCL_GSS | LCL_CALLBACKSON) &&
1235 	    (clp->lc_hand.nfsh_flag & NFSG_COMPLETE) &&
1236 	    clp->lc_handlelen > 0) {
1237 		clp->lc_hand.nfsh_flag &= ~NFSG_COMPLETE;
1238 		clp->lc_hand.nfsh_flag |= NFSG_DESTROYED;
1239 		(void) nfsrv_docallback(clp, NFSV4PROC_CBNULL,
1240 			NULL, 0, NULL, NULL, NULL, p);
1241 	}
1242 #endif
1243 	newnfs_disconnect(&clp->lc_req);
1244 	free(clp->lc_req.nr_nam, M_SONAME);
1245 	NFSFREEMUTEX(&clp->lc_req.nr_mtx);
1246 	free(clp->lc_stateid, M_NFSDCLIENT);
1247 	free(clp, M_NFSDCLIENT);
1248 	NFSLOCKSTATE();
1249 	nfsstatsv1.srvclients--;
1250 	nfsrv_openpluslock--;
1251 	nfsrv_clients--;
1252 	NFSUNLOCKSTATE();
1253 }
1254 
1255 /*
1256  * Free a list of delegation state structures.
1257  * (This function will also free all nfslockfile structures that no
1258  *  longer have associated state.)
1259  */
1260 APPLESTATIC void
1261 nfsrv_freedeleglist(struct nfsstatehead *sthp)
1262 {
1263 	struct nfsstate *stp, *nstp;
1264 
1265 	LIST_FOREACH_SAFE(stp, sthp, ls_list, nstp) {
1266 		nfsrv_freedeleg(stp);
1267 	}
1268 	LIST_INIT(sthp);
1269 }
1270 
1271 /*
1272  * Free up a delegation.
1273  */
1274 static void
1275 nfsrv_freedeleg(struct nfsstate *stp)
1276 {
1277 	struct nfslockfile *lfp;
1278 
1279 	LIST_REMOVE(stp, ls_hash);
1280 	LIST_REMOVE(stp, ls_list);
1281 	LIST_REMOVE(stp, ls_file);
1282 	if ((stp->ls_flags & NFSLCK_DELEGWRITE) != 0)
1283 		nfsrv_writedelegcnt--;
1284 	lfp = stp->ls_lfp;
1285 	if (LIST_EMPTY(&lfp->lf_open) &&
1286 	    LIST_EMPTY(&lfp->lf_lock) && LIST_EMPTY(&lfp->lf_deleg) &&
1287 	    LIST_EMPTY(&lfp->lf_locallock) && LIST_EMPTY(&lfp->lf_rollback) &&
1288 	    lfp->lf_usecount == 0 &&
1289 	    nfsv4_testlock(&lfp->lf_locallock_lck) == 0)
1290 		nfsrv_freenfslockfile(lfp);
1291 	free(stp, M_NFSDSTATE);
1292 	nfsstatsv1.srvdelegates--;
1293 	nfsrv_openpluslock--;
1294 	nfsrv_delegatecnt--;
1295 }
1296 
1297 /*
1298  * This function frees an open owner and all associated opens.
1299  */
1300 static void
1301 nfsrv_freeopenowner(struct nfsstate *stp, int cansleep, NFSPROC_T *p)
1302 {
1303 	struct nfsstate *nstp, *tstp;
1304 
1305 	LIST_REMOVE(stp, ls_list);
1306 	/*
1307 	 * Now, free all associated opens.
1308 	 */
1309 	nstp = LIST_FIRST(&stp->ls_open);
1310 	while (nstp != LIST_END(&stp->ls_open)) {
1311 		tstp = nstp;
1312 		nstp = LIST_NEXT(nstp, ls_list);
1313 		(void) nfsrv_freeopen(tstp, NULL, cansleep, p);
1314 	}
1315 	if (stp->ls_op)
1316 		nfsrvd_derefcache(stp->ls_op);
1317 	free(stp, M_NFSDSTATE);
1318 	nfsstatsv1.srvopenowners--;
1319 	nfsrv_openpluslock--;
1320 }
1321 
1322 /*
1323  * This function frees an open (nfsstate open structure) with all associated
1324  * lock_owners and locks. It also frees the nfslockfile structure iff there
1325  * are no other opens on the file.
1326  * Returns 1 if it free'd the nfslockfile, 0 otherwise.
1327  */
1328 static int
1329 nfsrv_freeopen(struct nfsstate *stp, vnode_t vp, int cansleep, NFSPROC_T *p)
1330 {
1331 	struct nfsstate *nstp, *tstp;
1332 	struct nfslockfile *lfp;
1333 	int ret;
1334 
1335 	LIST_REMOVE(stp, ls_hash);
1336 	LIST_REMOVE(stp, ls_list);
1337 	LIST_REMOVE(stp, ls_file);
1338 
1339 	lfp = stp->ls_lfp;
1340 	/*
1341 	 * Now, free all lockowners associated with this open.
1342 	 */
1343 	LIST_FOREACH_SAFE(tstp, &stp->ls_open, ls_list, nstp)
1344 		nfsrv_freelockowner(tstp, vp, cansleep, p);
1345 
1346 	/*
1347 	 * The nfslockfile is freed here if there are no locks
1348 	 * associated with the open.
1349 	 * If there are locks associated with the open, the
1350 	 * nfslockfile structure can be freed via nfsrv_freelockowner().
1351 	 * Acquire the state mutex to avoid races with calls to
1352 	 * nfsrv_getlockfile().
1353 	 */
1354 	if (cansleep != 0)
1355 		NFSLOCKSTATE();
1356 	if (lfp != NULL && LIST_EMPTY(&lfp->lf_open) &&
1357 	    LIST_EMPTY(&lfp->lf_deleg) && LIST_EMPTY(&lfp->lf_lock) &&
1358 	    LIST_EMPTY(&lfp->lf_locallock) && LIST_EMPTY(&lfp->lf_rollback) &&
1359 	    lfp->lf_usecount == 0 &&
1360 	    (cansleep != 0 || nfsv4_testlock(&lfp->lf_locallock_lck) == 0)) {
1361 		nfsrv_freenfslockfile(lfp);
1362 		ret = 1;
1363 	} else
1364 		ret = 0;
1365 	if (cansleep != 0)
1366 		NFSUNLOCKSTATE();
1367 	free(stp, M_NFSDSTATE);
1368 	nfsstatsv1.srvopens--;
1369 	nfsrv_openpluslock--;
1370 	return (ret);
1371 }
1372 
1373 /*
1374  * Frees a lockowner and all associated locks.
1375  */
1376 static void
1377 nfsrv_freelockowner(struct nfsstate *stp, vnode_t vp, int cansleep,
1378     NFSPROC_T *p)
1379 {
1380 
1381 	LIST_REMOVE(stp, ls_hash);
1382 	LIST_REMOVE(stp, ls_list);
1383 	nfsrv_freeallnfslocks(stp, vp, cansleep, p);
1384 	if (stp->ls_op)
1385 		nfsrvd_derefcache(stp->ls_op);
1386 	free(stp, M_NFSDSTATE);
1387 	nfsstatsv1.srvlockowners--;
1388 	nfsrv_openpluslock--;
1389 }
1390 
1391 /*
1392  * Free all the nfs locks on a lockowner.
1393  */
1394 static void
1395 nfsrv_freeallnfslocks(struct nfsstate *stp, vnode_t vp, int cansleep,
1396     NFSPROC_T *p)
1397 {
1398 	struct nfslock *lop, *nlop;
1399 	struct nfsrollback *rlp, *nrlp;
1400 	struct nfslockfile *lfp = NULL;
1401 	int gottvp = 0;
1402 	vnode_t tvp = NULL;
1403 	uint64_t first, end;
1404 
1405 	if (vp != NULL)
1406 		ASSERT_VOP_UNLOCKED(vp, "nfsrv_freeallnfslocks: vnode locked");
1407 	lop = LIST_FIRST(&stp->ls_lock);
1408 	while (lop != LIST_END(&stp->ls_lock)) {
1409 		nlop = LIST_NEXT(lop, lo_lckowner);
1410 		/*
1411 		 * Since all locks should be for the same file, lfp should
1412 		 * not change.
1413 		 */
1414 		if (lfp == NULL)
1415 			lfp = lop->lo_lfp;
1416 		else if (lfp != lop->lo_lfp)
1417 			panic("allnfslocks");
1418 		/*
1419 		 * If vp is NULL and cansleep != 0, a vnode must be acquired
1420 		 * from the file handle. This only occurs when called from
1421 		 * nfsrv_cleanclient().
1422 		 */
1423 		if (gottvp == 0) {
1424 			if (nfsrv_dolocallocks == 0)
1425 				tvp = NULL;
1426 			else if (vp == NULL && cansleep != 0) {
1427 				tvp = nfsvno_getvp(&lfp->lf_fh);
1428 				NFSVOPUNLOCK(tvp, 0);
1429 			} else
1430 				tvp = vp;
1431 			gottvp = 1;
1432 		}
1433 
1434 		if (tvp != NULL) {
1435 			if (cansleep == 0)
1436 				panic("allnfs2");
1437 			first = lop->lo_first;
1438 			end = lop->lo_end;
1439 			nfsrv_freenfslock(lop);
1440 			nfsrv_localunlock(tvp, lfp, first, end, p);
1441 			LIST_FOREACH_SAFE(rlp, &lfp->lf_rollback, rlck_list,
1442 			    nrlp)
1443 				free(rlp, M_NFSDROLLBACK);
1444 			LIST_INIT(&lfp->lf_rollback);
1445 		} else
1446 			nfsrv_freenfslock(lop);
1447 		lop = nlop;
1448 	}
1449 	if (vp == NULL && tvp != NULL)
1450 		vrele(tvp);
1451 }
1452 
1453 /*
1454  * Free an nfslock structure.
1455  */
1456 static void
1457 nfsrv_freenfslock(struct nfslock *lop)
1458 {
1459 
1460 	if (lop->lo_lckfile.le_prev != NULL) {
1461 		LIST_REMOVE(lop, lo_lckfile);
1462 		nfsstatsv1.srvlocks--;
1463 		nfsrv_openpluslock--;
1464 	}
1465 	LIST_REMOVE(lop, lo_lckowner);
1466 	free(lop, M_NFSDLOCK);
1467 }
1468 
1469 /*
1470  * This function frees an nfslockfile structure.
1471  */
1472 static void
1473 nfsrv_freenfslockfile(struct nfslockfile *lfp)
1474 {
1475 
1476 	LIST_REMOVE(lfp, lf_hash);
1477 	free(lfp, M_NFSDLOCKFILE);
1478 }
1479 
1480 /*
1481  * This function looks up an nfsstate structure via stateid.
1482  */
1483 static int
1484 nfsrv_getstate(struct nfsclient *clp, nfsv4stateid_t *stateidp, __unused u_int32_t flags,
1485     struct nfsstate **stpp)
1486 {
1487 	struct nfsstate *stp;
1488 	struct nfsstatehead *hp;
1489 	int error = 0;
1490 
1491 	*stpp = NULL;
1492 	hp = NFSSTATEHASH(clp, *stateidp);
1493 	LIST_FOREACH(stp, hp, ls_hash) {
1494 		if (!NFSBCMP(stp->ls_stateid.other, stateidp->other,
1495 			NFSX_STATEIDOTHER))
1496 			break;
1497 	}
1498 
1499 	/*
1500 	 * If no state id in list, return NFSERR_BADSTATEID.
1501 	 */
1502 	if (stp == LIST_END(hp)) {
1503 		error = NFSERR_BADSTATEID;
1504 		goto out;
1505 	}
1506 	*stpp = stp;
1507 
1508 out:
1509 	NFSEXITCODE(error);
1510 	return (error);
1511 }
1512 
1513 /*
1514  * This function gets an nfsstate structure via owner string.
1515  */
1516 static void
1517 nfsrv_getowner(struct nfsstatehead *hp, struct nfsstate *new_stp,
1518     struct nfsstate **stpp)
1519 {
1520 	struct nfsstate *stp;
1521 
1522 	*stpp = NULL;
1523 	LIST_FOREACH(stp, hp, ls_list) {
1524 		if (new_stp->ls_ownerlen == stp->ls_ownerlen &&
1525 		  !NFSBCMP(new_stp->ls_owner,stp->ls_owner,stp->ls_ownerlen)) {
1526 			*stpp = stp;
1527 			return;
1528 		}
1529 	}
1530 }
1531 
1532 /*
1533  * Lock control function called to update lock status.
1534  * Returns 0 upon success, -1 if there is no lock and the flags indicate
1535  * that one isn't to be created and an NFSERR_xxx for other errors.
1536  * The structures new_stp and new_lop are passed in as pointers that should
1537  * be set to NULL if the structure is used and shouldn't be free'd.
1538  * For the NFSLCK_TEST and NFSLCK_CHECK cases, the structures are
1539  * never used and can safely be allocated on the stack. For all other
1540  * cases, *new_stpp and *new_lopp should be malloc'd before the call,
1541  * in case they are used.
1542  */
1543 APPLESTATIC int
1544 nfsrv_lockctrl(vnode_t vp, struct nfsstate **new_stpp,
1545     struct nfslock **new_lopp, struct nfslockconflict *cfp,
1546     nfsquad_t clientid, nfsv4stateid_t *stateidp,
1547     __unused struct nfsexstuff *exp,
1548     struct nfsrv_descript *nd, NFSPROC_T *p)
1549 {
1550 	struct nfslock *lop;
1551 	struct nfsstate *new_stp = *new_stpp;
1552 	struct nfslock *new_lop = *new_lopp;
1553 	struct nfsstate *tstp, *mystp, *nstp;
1554 	int specialid = 0;
1555 	struct nfslockfile *lfp;
1556 	struct nfslock *other_lop = NULL;
1557 	struct nfsstate *stp, *lckstp = NULL;
1558 	struct nfsclient *clp = NULL;
1559 	u_int32_t bits;
1560 	int error = 0, haslock = 0, ret, reterr;
1561 	int getlckret, delegation = 0, filestruct_locked, vnode_unlocked = 0;
1562 	fhandle_t nfh;
1563 	uint64_t first, end;
1564 	uint32_t lock_flags;
1565 
1566 	if (new_stp->ls_flags & (NFSLCK_CHECK | NFSLCK_SETATTR)) {
1567 		/*
1568 		 * Note the special cases of "all 1s" or "all 0s" stateids and
1569 		 * let reads with all 1s go ahead.
1570 		 */
1571 		if (new_stp->ls_stateid.seqid == 0x0 &&
1572 		    new_stp->ls_stateid.other[0] == 0x0 &&
1573 		    new_stp->ls_stateid.other[1] == 0x0 &&
1574 		    new_stp->ls_stateid.other[2] == 0x0)
1575 			specialid = 1;
1576 		else if (new_stp->ls_stateid.seqid == 0xffffffff &&
1577 		    new_stp->ls_stateid.other[0] == 0xffffffff &&
1578 		    new_stp->ls_stateid.other[1] == 0xffffffff &&
1579 		    new_stp->ls_stateid.other[2] == 0xffffffff)
1580 			specialid = 2;
1581 	}
1582 
1583 	/*
1584 	 * Check for restart conditions (client and server).
1585 	 */
1586 	error = nfsrv_checkrestart(clientid, new_stp->ls_flags,
1587 	    &new_stp->ls_stateid, specialid);
1588 	if (error)
1589 		goto out;
1590 
1591 	/*
1592 	 * Check for state resource limit exceeded.
1593 	 */
1594 	if ((new_stp->ls_flags & NFSLCK_LOCK) &&
1595 	    nfsrv_openpluslock > nfsrv_v4statelimit) {
1596 		error = NFSERR_RESOURCE;
1597 		goto out;
1598 	}
1599 
1600 	/*
1601 	 * For the lock case, get another nfslock structure,
1602 	 * just in case we need it.
1603 	 * Malloc now, before we start sifting through the linked lists,
1604 	 * in case we have to wait for memory.
1605 	 */
1606 tryagain:
1607 	if (new_stp->ls_flags & NFSLCK_LOCK)
1608 		other_lop = malloc(sizeof (struct nfslock),
1609 		    M_NFSDLOCK, M_WAITOK);
1610 	filestruct_locked = 0;
1611 	reterr = 0;
1612 	lfp = NULL;
1613 
1614 	/*
1615 	 * Get the lockfile structure for CFH now, so we can do a sanity
1616 	 * check against the stateid, before incrementing the seqid#, since
1617 	 * we want to return NFSERR_BADSTATEID on failure and the seqid#
1618 	 * shouldn't be incremented for this case.
1619 	 * If nfsrv_getlockfile() returns -1, it means "not found", which
1620 	 * will be handled later.
1621 	 * If we are doing Lock/LockU and local locking is enabled, sleep
1622 	 * lock the nfslockfile structure.
1623 	 */
1624 	getlckret = nfsrv_getlockfh(vp, new_stp->ls_flags, NULL, &nfh, p);
1625 	NFSLOCKSTATE();
1626 	if (getlckret == 0) {
1627 		if ((new_stp->ls_flags & (NFSLCK_LOCK | NFSLCK_UNLOCK)) != 0 &&
1628 		    nfsrv_dolocallocks != 0 && nd->nd_repstat == 0) {
1629 			getlckret = nfsrv_getlockfile(new_stp->ls_flags, NULL,
1630 			    &lfp, &nfh, 1);
1631 			if (getlckret == 0)
1632 				filestruct_locked = 1;
1633 		} else
1634 			getlckret = nfsrv_getlockfile(new_stp->ls_flags, NULL,
1635 			    &lfp, &nfh, 0);
1636 	}
1637 	if (getlckret != 0 && getlckret != -1)
1638 		reterr = getlckret;
1639 
1640 	if (filestruct_locked != 0) {
1641 		LIST_INIT(&lfp->lf_rollback);
1642 		if ((new_stp->ls_flags & NFSLCK_LOCK)) {
1643 			/*
1644 			 * For local locking, do the advisory locking now, so
1645 			 * that any conflict can be detected. A failure later
1646 			 * can be rolled back locally. If an error is returned,
1647 			 * struct nfslockfile has been unlocked and any local
1648 			 * locking rolled back.
1649 			 */
1650 			NFSUNLOCKSTATE();
1651 			if (vnode_unlocked == 0) {
1652 				ASSERT_VOP_ELOCKED(vp, "nfsrv_lockctrl1");
1653 				vnode_unlocked = 1;
1654 				NFSVOPUNLOCK(vp, 0);
1655 			}
1656 			reterr = nfsrv_locallock(vp, lfp,
1657 			    (new_lop->lo_flags & (NFSLCK_READ | NFSLCK_WRITE)),
1658 			    new_lop->lo_first, new_lop->lo_end, cfp, p);
1659 			NFSLOCKSTATE();
1660 		}
1661 	}
1662 
1663 	if (specialid == 0) {
1664 	    if (new_stp->ls_flags & NFSLCK_TEST) {
1665 		/*
1666 		 * RFC 3530 does not list LockT as an op that renews a
1667 		 * lease, but the consensus seems to be that it is ok
1668 		 * for a server to do so.
1669 		 */
1670 		error = nfsrv_getclient(clientid, CLOPS_RENEW, &clp, NULL,
1671 		    (nfsquad_t)((u_quad_t)0), 0, nd, p);
1672 
1673 		/*
1674 		 * Since NFSERR_EXPIRED, NFSERR_ADMINREVOKED are not valid
1675 		 * error returns for LockT, just go ahead and test for a lock,
1676 		 * since there are no locks for this client, but other locks
1677 		 * can conflict. (ie. same client will always be false)
1678 		 */
1679 		if (error == NFSERR_EXPIRED || error == NFSERR_ADMINREVOKED)
1680 		    error = 0;
1681 		lckstp = new_stp;
1682 	    } else {
1683 	      error = nfsrv_getclient(clientid, CLOPS_RENEW, &clp, NULL,
1684 		(nfsquad_t)((u_quad_t)0), 0, nd, p);
1685 	      if (error == 0)
1686 		/*
1687 		 * Look up the stateid
1688 		 */
1689 		error = nfsrv_getstate(clp, &new_stp->ls_stateid,
1690 		  new_stp->ls_flags, &stp);
1691 	      /*
1692 	       * do some sanity checks for an unconfirmed open or a
1693 	       * stateid that refers to the wrong file, for an open stateid
1694 	       */
1695 	      if (error == 0 && (stp->ls_flags & NFSLCK_OPEN) &&
1696 		  ((stp->ls_openowner->ls_flags & NFSLCK_NEEDSCONFIRM) ||
1697 		   (getlckret == 0 && stp->ls_lfp != lfp))){
1698 		      /*
1699 		       * NFSLCK_SETATTR should return OK rather than NFSERR_BADSTATEID
1700 		       * The only exception is using SETATTR with SIZE.
1701 		       * */
1702                     if ((new_stp->ls_flags &
1703                          (NFSLCK_SETATTR | NFSLCK_CHECK)) != NFSLCK_SETATTR)
1704 			     error = NFSERR_BADSTATEID;
1705 	      }
1706 
1707 		if (error == 0 &&
1708 		  (stp->ls_flags & (NFSLCK_DELEGREAD | NFSLCK_DELEGWRITE)) &&
1709 		  getlckret == 0 && stp->ls_lfp != lfp)
1710 			error = NFSERR_BADSTATEID;
1711 
1712 	      /*
1713 	       * If the lockowner stateid doesn't refer to the same file,
1714 	       * I believe that is considered ok, since some clients will
1715 	       * only create a single lockowner and use that for all locks
1716 	       * on all files.
1717 	       * For now, log it as a diagnostic, instead of considering it
1718 	       * a BadStateid.
1719 	       */
1720 	      if (error == 0 && (stp->ls_flags &
1721 		  (NFSLCK_OPEN | NFSLCK_DELEGREAD | NFSLCK_DELEGWRITE)) == 0 &&
1722 		  getlckret == 0 && stp->ls_lfp != lfp) {
1723 #ifdef DIAGNOSTIC
1724 		  printf("Got a lock statid for different file open\n");
1725 #endif
1726 		  /*
1727 		  error = NFSERR_BADSTATEID;
1728 		  */
1729 	      }
1730 
1731 	      if (error == 0) {
1732 		    if (new_stp->ls_flags & NFSLCK_OPENTOLOCK) {
1733 			/*
1734 			 * If haslock set, we've already checked the seqid.
1735 			 */
1736 			if (!haslock) {
1737 			    if (stp->ls_flags & NFSLCK_OPEN)
1738 				error = nfsrv_checkseqid(nd, new_stp->ls_seq,
1739 				    stp->ls_openowner, new_stp->ls_op);
1740 			    else
1741 				error = NFSERR_BADSTATEID;
1742 			}
1743 			if (!error)
1744 			    nfsrv_getowner(&stp->ls_open, new_stp, &lckstp);
1745 			if (lckstp)
1746 			    /*
1747 			     * I believe this should be an error, but it
1748 			     * isn't obvious what NFSERR_xxx would be
1749 			     * appropriate, so I'll use NFSERR_INVAL for now.
1750 			     */
1751 			    error = NFSERR_INVAL;
1752 			else
1753 			    lckstp = new_stp;
1754 		    } else if (new_stp->ls_flags&(NFSLCK_LOCK|NFSLCK_UNLOCK)) {
1755 			/*
1756 			 * If haslock set, ditto above.
1757 			 */
1758 			if (!haslock) {
1759 			    if (stp->ls_flags & NFSLCK_OPEN)
1760 				error = NFSERR_BADSTATEID;
1761 			    else
1762 				error = nfsrv_checkseqid(nd, new_stp->ls_seq,
1763 				    stp, new_stp->ls_op);
1764 			}
1765 			lckstp = stp;
1766 		    } else {
1767 			lckstp = stp;
1768 		    }
1769 	      }
1770 	      /*
1771 	       * If the seqid part of the stateid isn't the same, return
1772 	       * NFSERR_OLDSTATEID for cases other than I/O Ops.
1773 	       * For I/O Ops, only return NFSERR_OLDSTATEID if
1774 	       * nfsrv_returnoldstateid is set. (The consensus on the email
1775 	       * list was that most clients would prefer to not receive
1776 	       * NFSERR_OLDSTATEID for I/O Ops, but the RFC suggests that that
1777 	       * is what will happen, so I use the nfsrv_returnoldstateid to
1778 	       * allow for either server configuration.)
1779 	       */
1780 	      if (!error && stp->ls_stateid.seqid!=new_stp->ls_stateid.seqid &&
1781 		  (((nd->nd_flag & ND_NFSV41) == 0 &&
1782 		   (!(new_stp->ls_flags & NFSLCK_CHECK) ||
1783 		    nfsrv_returnoldstateid)) ||
1784 		   ((nd->nd_flag & ND_NFSV41) != 0 &&
1785 		    new_stp->ls_stateid.seqid != 0)))
1786 		    error = NFSERR_OLDSTATEID;
1787 	    }
1788 	}
1789 
1790 	/*
1791 	 * Now we can check for grace.
1792 	 */
1793 	if (!error)
1794 		error = nfsrv_checkgrace(nd, clp, new_stp->ls_flags);
1795 	if ((new_stp->ls_flags & NFSLCK_RECLAIM) && !error &&
1796 		nfsrv_checkstable(clp))
1797 		error = NFSERR_NOGRACE;
1798 	/*
1799 	 * If we successfully Reclaimed state, note that.
1800 	 */
1801 	if ((new_stp->ls_flags & NFSLCK_RECLAIM) && !error)
1802 		nfsrv_markstable(clp);
1803 
1804 	/*
1805 	 * At this point, either error == NFSERR_BADSTATEID or the
1806 	 * seqid# has been updated, so we can return any error.
1807 	 * If error == 0, there may be an error in:
1808 	 *    nd_repstat - Set by the calling function.
1809 	 *    reterr - Set above, if getting the nfslockfile structure
1810 	 *       or acquiring the local lock failed.
1811 	 *    (If both of these are set, nd_repstat should probably be
1812 	 *     returned, since that error was detected before this
1813 	 *     function call.)
1814 	 */
1815 	if (error != 0 || nd->nd_repstat != 0 || reterr != 0) {
1816 		if (error == 0) {
1817 			if (nd->nd_repstat != 0)
1818 				error = nd->nd_repstat;
1819 			else
1820 				error = reterr;
1821 		}
1822 		if (filestruct_locked != 0) {
1823 			/* Roll back local locks. */
1824 			NFSUNLOCKSTATE();
1825 			if (vnode_unlocked == 0) {
1826 				ASSERT_VOP_ELOCKED(vp, "nfsrv_lockctrl2");
1827 				vnode_unlocked = 1;
1828 				NFSVOPUNLOCK(vp, 0);
1829 			}
1830 			nfsrv_locallock_rollback(vp, lfp, p);
1831 			NFSLOCKSTATE();
1832 			nfsrv_unlocklf(lfp);
1833 		}
1834 		NFSUNLOCKSTATE();
1835 		goto out;
1836 	}
1837 
1838 	/*
1839 	 * Check the nfsrv_getlockfile return.
1840 	 * Returned -1 if no structure found.
1841 	 */
1842 	if (getlckret == -1) {
1843 		error = NFSERR_EXPIRED;
1844 		/*
1845 		 * Called from lockt, so no lock is OK.
1846 		 */
1847 		if (new_stp->ls_flags & NFSLCK_TEST) {
1848 			error = 0;
1849 		} else if (new_stp->ls_flags &
1850 		    (NFSLCK_CHECK | NFSLCK_SETATTR)) {
1851 			/*
1852 			 * Called to check for a lock, OK if the stateid is all
1853 			 * 1s or all 0s, but there should be an nfsstate
1854 			 * otherwise.
1855 			 * (ie. If there is no open, I'll assume no share
1856 			 *  deny bits.)
1857 			 */
1858 			if (specialid)
1859 				error = 0;
1860 			else
1861 				error = NFSERR_BADSTATEID;
1862 		}
1863 		NFSUNLOCKSTATE();
1864 		goto out;
1865 	}
1866 
1867 	/*
1868 	 * For NFSLCK_CHECK and NFSLCK_LOCK, test for a share conflict.
1869 	 * For NFSLCK_CHECK, allow a read if write access is granted,
1870 	 * but check for a deny. For NFSLCK_LOCK, require correct access,
1871 	 * which implies a conflicting deny can't exist.
1872 	 */
1873 	if (new_stp->ls_flags & (NFSLCK_CHECK | NFSLCK_LOCK)) {
1874 	    /*
1875 	     * Four kinds of state id:
1876 	     * - specialid (all 0s or all 1s), only for NFSLCK_CHECK
1877 	     * - stateid for an open
1878 	     * - stateid for a delegation
1879 	     * - stateid for a lock owner
1880 	     */
1881 	    if (!specialid) {
1882 		if (stp->ls_flags & (NFSLCK_DELEGREAD | NFSLCK_DELEGWRITE)) {
1883 		    delegation = 1;
1884 		    mystp = stp;
1885 		    nfsrv_delaydelegtimeout(stp);
1886 	        } else if (stp->ls_flags & NFSLCK_OPEN) {
1887 		    mystp = stp;
1888 		} else {
1889 		    mystp = stp->ls_openstp;
1890 		}
1891 		/*
1892 		 * If locking or checking, require correct access
1893 		 * bit set.
1894 		 */
1895 		if (((new_stp->ls_flags & NFSLCK_LOCK) &&
1896 		     !((new_lop->lo_flags >> NFSLCK_LOCKSHIFT) &
1897 		       mystp->ls_flags & NFSLCK_ACCESSBITS)) ||
1898 		    ((new_stp->ls_flags & (NFSLCK_CHECK|NFSLCK_READACCESS)) ==
1899 		      (NFSLCK_CHECK | NFSLCK_READACCESS) &&
1900 		     !(mystp->ls_flags & NFSLCK_READACCESS) &&
1901 		     nfsrv_allowreadforwriteopen == 0) ||
1902 		    ((new_stp->ls_flags & (NFSLCK_CHECK|NFSLCK_WRITEACCESS)) ==
1903 		      (NFSLCK_CHECK | NFSLCK_WRITEACCESS) &&
1904 		     !(mystp->ls_flags & NFSLCK_WRITEACCESS))) {
1905 			if (filestruct_locked != 0) {
1906 				/* Roll back local locks. */
1907 				NFSUNLOCKSTATE();
1908 				if (vnode_unlocked == 0) {
1909 					ASSERT_VOP_ELOCKED(vp,
1910 					    "nfsrv_lockctrl3");
1911 					vnode_unlocked = 1;
1912 					NFSVOPUNLOCK(vp, 0);
1913 				}
1914 				nfsrv_locallock_rollback(vp, lfp, p);
1915 				NFSLOCKSTATE();
1916 				nfsrv_unlocklf(lfp);
1917 			}
1918 			NFSUNLOCKSTATE();
1919 			error = NFSERR_OPENMODE;
1920 			goto out;
1921 		}
1922 	    } else
1923 		mystp = NULL;
1924 	    if ((new_stp->ls_flags & NFSLCK_CHECK) && !delegation) {
1925 		/*
1926 		 * Check for a conflicting deny bit.
1927 		 */
1928 		LIST_FOREACH(tstp, &lfp->lf_open, ls_file) {
1929 		    if (tstp != mystp) {
1930 			bits = tstp->ls_flags;
1931 			bits >>= NFSLCK_SHIFT;
1932 			if (new_stp->ls_flags & bits & NFSLCK_ACCESSBITS) {
1933 			    KASSERT(vnode_unlocked == 0,
1934 				("nfsrv_lockctrl: vnode unlocked1"));
1935 			    ret = nfsrv_clientconflict(tstp->ls_clp, &haslock,
1936 				vp, p);
1937 			    if (ret == 1) {
1938 				/*
1939 				* nfsrv_clientconflict unlocks state
1940 				 * when it returns non-zero.
1941 				 */
1942 				lckstp = NULL;
1943 				goto tryagain;
1944 			    }
1945 			    if (ret == 0)
1946 				NFSUNLOCKSTATE();
1947 			    if (ret == 2)
1948 				error = NFSERR_PERM;
1949 			    else
1950 				error = NFSERR_OPENMODE;
1951 			    goto out;
1952 			}
1953 		    }
1954 		}
1955 
1956 		/* We're outta here */
1957 		NFSUNLOCKSTATE();
1958 		goto out;
1959 	    }
1960 	}
1961 
1962 	/*
1963 	 * For setattr, just get rid of all the Delegations for other clients.
1964 	 */
1965 	if (new_stp->ls_flags & NFSLCK_SETATTR) {
1966 		KASSERT(vnode_unlocked == 0,
1967 		    ("nfsrv_lockctrl: vnode unlocked2"));
1968 		ret = nfsrv_cleandeleg(vp, lfp, clp, &haslock, p);
1969 		if (ret) {
1970 			/*
1971 			 * nfsrv_cleandeleg() unlocks state when it
1972 			 * returns non-zero.
1973 			 */
1974 			if (ret == -1) {
1975 				lckstp = NULL;
1976 				goto tryagain;
1977 			}
1978 			error = ret;
1979 			goto out;
1980 		}
1981 		if (!(new_stp->ls_flags & NFSLCK_CHECK) ||
1982 		    (LIST_EMPTY(&lfp->lf_open) && LIST_EMPTY(&lfp->lf_lock) &&
1983 		     LIST_EMPTY(&lfp->lf_deleg))) {
1984 			NFSUNLOCKSTATE();
1985 			goto out;
1986 		}
1987 	}
1988 
1989 	/*
1990 	 * Check for a conflicting delegation. If one is found, call
1991 	 * nfsrv_delegconflict() to handle it. If the v4root lock hasn't
1992 	 * been set yet, it will get the lock. Otherwise, it will recall
1993 	 * the delegation. Then, we try try again...
1994 	 * I currently believe the conflict algorithm to be:
1995 	 * For Lock Ops (Lock/LockT/LockU)
1996 	 * - there is a conflict iff a different client has a write delegation
1997 	 * For Reading (Read Op)
1998 	 * - there is a conflict iff a different client has a write delegation
1999 	 *   (the specialids are always a different client)
2000 	 * For Writing (Write/Setattr of size)
2001 	 * - there is a conflict if a different client has any delegation
2002 	 * - there is a conflict if the same client has a read delegation
2003 	 *   (I don't understand why this isn't allowed, but that seems to be
2004 	 *    the current consensus?)
2005 	 */
2006 	tstp = LIST_FIRST(&lfp->lf_deleg);
2007 	while (tstp != LIST_END(&lfp->lf_deleg)) {
2008 	    nstp = LIST_NEXT(tstp, ls_file);
2009 	    if ((((new_stp->ls_flags&(NFSLCK_LOCK|NFSLCK_UNLOCK|NFSLCK_TEST))||
2010 		 ((new_stp->ls_flags & NFSLCK_CHECK) &&
2011 		  (new_lop->lo_flags & NFSLCK_READ))) &&
2012 		  clp != tstp->ls_clp &&
2013 		 (tstp->ls_flags & NFSLCK_DELEGWRITE)) ||
2014 		 ((new_stp->ls_flags & NFSLCK_CHECK) &&
2015 		   (new_lop->lo_flags & NFSLCK_WRITE) &&
2016 		  (clp != tstp->ls_clp ||
2017 		   (tstp->ls_flags & NFSLCK_DELEGREAD)))) {
2018 		ret = 0;
2019 		if (filestruct_locked != 0) {
2020 			/* Roll back local locks. */
2021 			NFSUNLOCKSTATE();
2022 			if (vnode_unlocked == 0) {
2023 				ASSERT_VOP_ELOCKED(vp, "nfsrv_lockctrl4");
2024 				NFSVOPUNLOCK(vp, 0);
2025 			}
2026 			nfsrv_locallock_rollback(vp, lfp, p);
2027 			NFSLOCKSTATE();
2028 			nfsrv_unlocklf(lfp);
2029 			NFSUNLOCKSTATE();
2030 			NFSVOPLOCK(vp, LK_EXCLUSIVE | LK_RETRY);
2031 			vnode_unlocked = 0;
2032 			if ((vp->v_iflag & VI_DOOMED) != 0)
2033 				ret = NFSERR_SERVERFAULT;
2034 			NFSLOCKSTATE();
2035 		}
2036 		if (ret == 0)
2037 			ret = nfsrv_delegconflict(tstp, &haslock, p, vp);
2038 		if (ret) {
2039 		    /*
2040 		     * nfsrv_delegconflict unlocks state when it
2041 		     * returns non-zero, which it always does.
2042 		     */
2043 		    if (other_lop) {
2044 			free(other_lop, M_NFSDLOCK);
2045 			other_lop = NULL;
2046 		    }
2047 		    if (ret == -1) {
2048 			lckstp = NULL;
2049 			goto tryagain;
2050 		    }
2051 		    error = ret;
2052 		    goto out;
2053 		}
2054 		/* Never gets here. */
2055 	    }
2056 	    tstp = nstp;
2057 	}
2058 
2059 	/*
2060 	 * Handle the unlock case by calling nfsrv_updatelock().
2061 	 * (Should I have done some access checking above for unlock? For now,
2062 	 *  just let it happen.)
2063 	 */
2064 	if (new_stp->ls_flags & NFSLCK_UNLOCK) {
2065 		first = new_lop->lo_first;
2066 		end = new_lop->lo_end;
2067 		nfsrv_updatelock(stp, new_lopp, &other_lop, lfp);
2068 		stateidp->seqid = ++(stp->ls_stateid.seqid);
2069 		if ((nd->nd_flag & ND_NFSV41) != 0 && stateidp->seqid == 0)
2070 			stateidp->seqid = stp->ls_stateid.seqid = 1;
2071 		stateidp->other[0] = stp->ls_stateid.other[0];
2072 		stateidp->other[1] = stp->ls_stateid.other[1];
2073 		stateidp->other[2] = stp->ls_stateid.other[2];
2074 		if (filestruct_locked != 0) {
2075 			NFSUNLOCKSTATE();
2076 			if (vnode_unlocked == 0) {
2077 				ASSERT_VOP_ELOCKED(vp, "nfsrv_lockctrl5");
2078 				vnode_unlocked = 1;
2079 				NFSVOPUNLOCK(vp, 0);
2080 			}
2081 			/* Update the local locks. */
2082 			nfsrv_localunlock(vp, lfp, first, end, p);
2083 			NFSLOCKSTATE();
2084 			nfsrv_unlocklf(lfp);
2085 		}
2086 		NFSUNLOCKSTATE();
2087 		goto out;
2088 	}
2089 
2090 	/*
2091 	 * Search for a conflicting lock. A lock conflicts if:
2092 	 * - the lock range overlaps and
2093 	 * - at least one lock is a write lock and
2094 	 * - it is not owned by the same lock owner
2095 	 */
2096 	if (!delegation) {
2097 	  LIST_FOREACH(lop, &lfp->lf_lock, lo_lckfile) {
2098 	    if (new_lop->lo_end > lop->lo_first &&
2099 		new_lop->lo_first < lop->lo_end &&
2100 		(new_lop->lo_flags == NFSLCK_WRITE ||
2101 		 lop->lo_flags == NFSLCK_WRITE) &&
2102 		lckstp != lop->lo_stp &&
2103 		(clp != lop->lo_stp->ls_clp ||
2104 		 lckstp->ls_ownerlen != lop->lo_stp->ls_ownerlen ||
2105 		 NFSBCMP(lckstp->ls_owner, lop->lo_stp->ls_owner,
2106 		    lckstp->ls_ownerlen))) {
2107 		if (other_lop) {
2108 		    free(other_lop, M_NFSDLOCK);
2109 		    other_lop = NULL;
2110 		}
2111 		if (vnode_unlocked != 0)
2112 		    ret = nfsrv_clientconflict(lop->lo_stp->ls_clp, &haslock,
2113 			NULL, p);
2114 		else
2115 		    ret = nfsrv_clientconflict(lop->lo_stp->ls_clp, &haslock,
2116 			vp, p);
2117 		if (ret == 1) {
2118 		    if (filestruct_locked != 0) {
2119 			if (vnode_unlocked == 0) {
2120 				ASSERT_VOP_ELOCKED(vp, "nfsrv_lockctrl6");
2121 				NFSVOPUNLOCK(vp, 0);
2122 			}
2123 			/* Roll back local locks. */
2124 			nfsrv_locallock_rollback(vp, lfp, p);
2125 			NFSLOCKSTATE();
2126 			nfsrv_unlocklf(lfp);
2127 			NFSUNLOCKSTATE();
2128 			NFSVOPLOCK(vp, LK_EXCLUSIVE | LK_RETRY);
2129 			vnode_unlocked = 0;
2130 			if ((vp->v_iflag & VI_DOOMED) != 0) {
2131 				error = NFSERR_SERVERFAULT;
2132 				goto out;
2133 			}
2134 		    }
2135 		    /*
2136 		     * nfsrv_clientconflict() unlocks state when it
2137 		     * returns non-zero.
2138 		     */
2139 		    lckstp = NULL;
2140 		    goto tryagain;
2141 		}
2142 		/*
2143 		 * Found a conflicting lock, so record the conflict and
2144 		 * return the error.
2145 		 */
2146 		if (cfp != NULL && ret == 0) {
2147 		    cfp->cl_clientid.lval[0]=lop->lo_stp->ls_stateid.other[0];
2148 		    cfp->cl_clientid.lval[1]=lop->lo_stp->ls_stateid.other[1];
2149 		    cfp->cl_first = lop->lo_first;
2150 		    cfp->cl_end = lop->lo_end;
2151 		    cfp->cl_flags = lop->lo_flags;
2152 		    cfp->cl_ownerlen = lop->lo_stp->ls_ownerlen;
2153 		    NFSBCOPY(lop->lo_stp->ls_owner, cfp->cl_owner,
2154 			cfp->cl_ownerlen);
2155 		}
2156 		if (ret == 2)
2157 		    error = NFSERR_PERM;
2158 		else if (new_stp->ls_flags & NFSLCK_RECLAIM)
2159 		    error = NFSERR_RECLAIMCONFLICT;
2160 		else if (new_stp->ls_flags & NFSLCK_CHECK)
2161 		    error = NFSERR_LOCKED;
2162 		else
2163 		    error = NFSERR_DENIED;
2164 		if (filestruct_locked != 0 && ret == 0) {
2165 			/* Roll back local locks. */
2166 			NFSUNLOCKSTATE();
2167 			if (vnode_unlocked == 0) {
2168 				ASSERT_VOP_ELOCKED(vp, "nfsrv_lockctrl7");
2169 				vnode_unlocked = 1;
2170 				NFSVOPUNLOCK(vp, 0);
2171 			}
2172 			nfsrv_locallock_rollback(vp, lfp, p);
2173 			NFSLOCKSTATE();
2174 			nfsrv_unlocklf(lfp);
2175 		}
2176 		if (ret == 0)
2177 			NFSUNLOCKSTATE();
2178 		goto out;
2179 	    }
2180 	  }
2181 	}
2182 
2183 	/*
2184 	 * We only get here if there was no lock that conflicted.
2185 	 */
2186 	if (new_stp->ls_flags & (NFSLCK_TEST | NFSLCK_CHECK)) {
2187 		NFSUNLOCKSTATE();
2188 		goto out;
2189 	}
2190 
2191 	/*
2192 	 * We only get here when we are creating or modifying a lock.
2193 	 * There are two variants:
2194 	 * - exist_lock_owner where lock_owner exists
2195 	 * - open_to_lock_owner with new lock_owner
2196 	 */
2197 	first = new_lop->lo_first;
2198 	end = new_lop->lo_end;
2199 	lock_flags = new_lop->lo_flags;
2200 	if (!(new_stp->ls_flags & NFSLCK_OPENTOLOCK)) {
2201 		nfsrv_updatelock(lckstp, new_lopp, &other_lop, lfp);
2202 		stateidp->seqid = ++(lckstp->ls_stateid.seqid);
2203 		if ((nd->nd_flag & ND_NFSV41) != 0 && stateidp->seqid == 0)
2204 			stateidp->seqid = lckstp->ls_stateid.seqid = 1;
2205 		stateidp->other[0] = lckstp->ls_stateid.other[0];
2206 		stateidp->other[1] = lckstp->ls_stateid.other[1];
2207 		stateidp->other[2] = lckstp->ls_stateid.other[2];
2208 	} else {
2209 		/*
2210 		 * The new open_to_lock_owner case.
2211 		 * Link the new nfsstate into the lists.
2212 		 */
2213 		new_stp->ls_seq = new_stp->ls_opentolockseq;
2214 		nfsrvd_refcache(new_stp->ls_op);
2215 		stateidp->seqid = new_stp->ls_stateid.seqid = 1;
2216 		stateidp->other[0] = new_stp->ls_stateid.other[0] =
2217 		    clp->lc_clientid.lval[0];
2218 		stateidp->other[1] = new_stp->ls_stateid.other[1] =
2219 		    clp->lc_clientid.lval[1];
2220 		stateidp->other[2] = new_stp->ls_stateid.other[2] =
2221 		    nfsrv_nextstateindex(clp);
2222 		new_stp->ls_clp = clp;
2223 		LIST_INIT(&new_stp->ls_lock);
2224 		new_stp->ls_openstp = stp;
2225 		new_stp->ls_lfp = lfp;
2226 		nfsrv_insertlock(new_lop, (struct nfslock *)new_stp, new_stp,
2227 		    lfp);
2228 		LIST_INSERT_HEAD(NFSSTATEHASH(clp, new_stp->ls_stateid),
2229 		    new_stp, ls_hash);
2230 		LIST_INSERT_HEAD(&stp->ls_open, new_stp, ls_list);
2231 		*new_lopp = NULL;
2232 		*new_stpp = NULL;
2233 		nfsstatsv1.srvlockowners++;
2234 		nfsrv_openpluslock++;
2235 	}
2236 	if (filestruct_locked != 0) {
2237 		NFSUNLOCKSTATE();
2238 		nfsrv_locallock_commit(lfp, lock_flags, first, end);
2239 		NFSLOCKSTATE();
2240 		nfsrv_unlocklf(lfp);
2241 	}
2242 	NFSUNLOCKSTATE();
2243 
2244 out:
2245 	if (haslock) {
2246 		NFSLOCKV4ROOTMUTEX();
2247 		nfsv4_unlock(&nfsv4rootfs_lock, 1);
2248 		NFSUNLOCKV4ROOTMUTEX();
2249 	}
2250 	if (vnode_unlocked != 0) {
2251 		NFSVOPLOCK(vp, LK_EXCLUSIVE | LK_RETRY);
2252 		if (error == 0 && (vp->v_iflag & VI_DOOMED) != 0)
2253 			error = NFSERR_SERVERFAULT;
2254 	}
2255 	if (other_lop)
2256 		free(other_lop, M_NFSDLOCK);
2257 	NFSEXITCODE2(error, nd);
2258 	return (error);
2259 }
2260 
2261 /*
2262  * Check for state errors for Open.
2263  * repstat is passed back out as an error if more critical errors
2264  * are not detected.
2265  */
2266 APPLESTATIC int
2267 nfsrv_opencheck(nfsquad_t clientid, nfsv4stateid_t *stateidp,
2268     struct nfsstate *new_stp, vnode_t vp, struct nfsrv_descript *nd,
2269     NFSPROC_T *p, int repstat)
2270 {
2271 	struct nfsstate *stp, *nstp;
2272 	struct nfsclient *clp;
2273 	struct nfsstate *ownerstp;
2274 	struct nfslockfile *lfp, *new_lfp;
2275 	int error = 0, haslock = 0, ret, readonly = 0, getfhret = 0;
2276 
2277 	if ((new_stp->ls_flags & NFSLCK_SHAREBITS) == NFSLCK_READACCESS)
2278 		readonly = 1;
2279 	/*
2280 	 * Check for restart conditions (client and server).
2281 	 */
2282 	error = nfsrv_checkrestart(clientid, new_stp->ls_flags,
2283 		&new_stp->ls_stateid, 0);
2284 	if (error)
2285 		goto out;
2286 
2287 	/*
2288 	 * Check for state resource limit exceeded.
2289 	 * Technically this should be SMP protected, but the worst
2290 	 * case error is "out by one or two" on the count when it
2291 	 * returns NFSERR_RESOURCE and the limit is just a rather
2292 	 * arbitrary high water mark, so no harm is done.
2293 	 */
2294 	if (nfsrv_openpluslock > nfsrv_v4statelimit) {
2295 		error = NFSERR_RESOURCE;
2296 		goto out;
2297 	}
2298 
2299 tryagain:
2300 	new_lfp = malloc(sizeof (struct nfslockfile),
2301 	    M_NFSDLOCKFILE, M_WAITOK);
2302 	if (vp)
2303 		getfhret = nfsrv_getlockfh(vp, new_stp->ls_flags, new_lfp,
2304 		    NULL, p);
2305 	NFSLOCKSTATE();
2306 	/*
2307 	 * Get the nfsclient structure.
2308 	 */
2309 	error = nfsrv_getclient(clientid, CLOPS_RENEW, &clp, NULL,
2310 	    (nfsquad_t)((u_quad_t)0), 0, nd, p);
2311 
2312 	/*
2313 	 * Look up the open owner. See if it needs confirmation and
2314 	 * check the seq#, as required.
2315 	 */
2316 	if (!error)
2317 		nfsrv_getowner(&clp->lc_open, new_stp, &ownerstp);
2318 
2319 	if (!error && ownerstp) {
2320 		error = nfsrv_checkseqid(nd, new_stp->ls_seq, ownerstp,
2321 		    new_stp->ls_op);
2322 		/*
2323 		 * If the OpenOwner hasn't been confirmed, assume the
2324 		 * old one was a replay and this one is ok.
2325 		 * See: RFC3530 Sec. 14.2.18.
2326 		 */
2327 		if (error == NFSERR_BADSEQID &&
2328 		    (ownerstp->ls_flags & NFSLCK_NEEDSCONFIRM))
2329 			error = 0;
2330 	}
2331 
2332 	/*
2333 	 * Check for grace.
2334 	 */
2335 	if (!error)
2336 		error = nfsrv_checkgrace(nd, clp, new_stp->ls_flags);
2337 	if ((new_stp->ls_flags & NFSLCK_RECLAIM) && !error &&
2338 		nfsrv_checkstable(clp))
2339 		error = NFSERR_NOGRACE;
2340 
2341 	/*
2342 	 * If none of the above errors occurred, let repstat be
2343 	 * returned.
2344 	 */
2345 	if (repstat && !error)
2346 		error = repstat;
2347 	if (error) {
2348 		NFSUNLOCKSTATE();
2349 		if (haslock) {
2350 			NFSLOCKV4ROOTMUTEX();
2351 			nfsv4_unlock(&nfsv4rootfs_lock, 1);
2352 			NFSUNLOCKV4ROOTMUTEX();
2353 		}
2354 		free(new_lfp, M_NFSDLOCKFILE);
2355 		goto out;
2356 	}
2357 
2358 	/*
2359 	 * If vp == NULL, the file doesn't exist yet, so return ok.
2360 	 * (This always happens on the first pass, so haslock must be 0.)
2361 	 */
2362 	if (vp == NULL) {
2363 		NFSUNLOCKSTATE();
2364 		free(new_lfp, M_NFSDLOCKFILE);
2365 		goto out;
2366 	}
2367 
2368 	/*
2369 	 * Get the structure for the underlying file.
2370 	 */
2371 	if (getfhret)
2372 		error = getfhret;
2373 	else
2374 		error = nfsrv_getlockfile(new_stp->ls_flags, &new_lfp, &lfp,
2375 		    NULL, 0);
2376 	if (new_lfp)
2377 		free(new_lfp, M_NFSDLOCKFILE);
2378 	if (error) {
2379 		NFSUNLOCKSTATE();
2380 		if (haslock) {
2381 			NFSLOCKV4ROOTMUTEX();
2382 			nfsv4_unlock(&nfsv4rootfs_lock, 1);
2383 			NFSUNLOCKV4ROOTMUTEX();
2384 		}
2385 		goto out;
2386 	}
2387 
2388 	/*
2389 	 * Search for a conflicting open/share.
2390 	 */
2391 	if (new_stp->ls_flags & NFSLCK_DELEGCUR) {
2392 	    /*
2393 	     * For Delegate_Cur, search for the matching Delegation,
2394 	     * which indicates no conflict.
2395 	     * An old delegation should have been recovered by the
2396 	     * client doing a Claim_DELEGATE_Prev, so I won't let
2397 	     * it match and return NFSERR_EXPIRED. Should I let it
2398 	     * match?
2399 	     */
2400 	    LIST_FOREACH(stp, &lfp->lf_deleg, ls_file) {
2401 		if (!(stp->ls_flags & NFSLCK_OLDDELEG) &&
2402 		    (((nd->nd_flag & ND_NFSV41) != 0 &&
2403 		    stateidp->seqid == 0) ||
2404 		    stateidp->seqid == stp->ls_stateid.seqid) &&
2405 		    !NFSBCMP(stateidp->other, stp->ls_stateid.other,
2406 			  NFSX_STATEIDOTHER))
2407 			break;
2408 	    }
2409 	    if (stp == LIST_END(&lfp->lf_deleg) ||
2410 		((new_stp->ls_flags & NFSLCK_WRITEACCESS) &&
2411 		 (stp->ls_flags & NFSLCK_DELEGREAD))) {
2412 		NFSUNLOCKSTATE();
2413 		if (haslock) {
2414 			NFSLOCKV4ROOTMUTEX();
2415 			nfsv4_unlock(&nfsv4rootfs_lock, 1);
2416 			NFSUNLOCKV4ROOTMUTEX();
2417 		}
2418 		error = NFSERR_EXPIRED;
2419 		goto out;
2420 	    }
2421 	}
2422 
2423 	/*
2424 	 * Check for access/deny bit conflicts. I check for the same
2425 	 * owner as well, in case the client didn't bother.
2426 	 */
2427 	LIST_FOREACH(stp, &lfp->lf_open, ls_file) {
2428 		if (!(new_stp->ls_flags & NFSLCK_DELEGCUR) &&
2429 		    (((new_stp->ls_flags & NFSLCK_ACCESSBITS) &
2430 		      ((stp->ls_flags>>NFSLCK_SHIFT) & NFSLCK_ACCESSBITS))||
2431 		     ((stp->ls_flags & NFSLCK_ACCESSBITS) &
2432 		      ((new_stp->ls_flags>>NFSLCK_SHIFT)&NFSLCK_ACCESSBITS)))){
2433 			ret = nfsrv_clientconflict(stp->ls_clp,&haslock,vp,p);
2434 			if (ret == 1) {
2435 				/*
2436 				 * nfsrv_clientconflict() unlocks
2437 				 * state when it returns non-zero.
2438 				 */
2439 				goto tryagain;
2440 			}
2441 			if (ret == 2)
2442 				error = NFSERR_PERM;
2443 			else if (new_stp->ls_flags & NFSLCK_RECLAIM)
2444 				error = NFSERR_RECLAIMCONFLICT;
2445 			else
2446 				error = NFSERR_SHAREDENIED;
2447 			if (ret == 0)
2448 				NFSUNLOCKSTATE();
2449 			if (haslock) {
2450 				NFSLOCKV4ROOTMUTEX();
2451 				nfsv4_unlock(&nfsv4rootfs_lock, 1);
2452 				NFSUNLOCKV4ROOTMUTEX();
2453 			}
2454 			goto out;
2455 		}
2456 	}
2457 
2458 	/*
2459 	 * Check for a conflicting delegation. If one is found, call
2460 	 * nfsrv_delegconflict() to handle it. If the v4root lock hasn't
2461 	 * been set yet, it will get the lock. Otherwise, it will recall
2462 	 * the delegation. Then, we try try again...
2463 	 * (If NFSLCK_DELEGCUR is set, it has a delegation, so there
2464 	 *  isn't a conflict.)
2465 	 * I currently believe the conflict algorithm to be:
2466 	 * For Open with Read Access and Deny None
2467 	 * - there is a conflict iff a different client has a write delegation
2468 	 * For Open with other Write Access or any Deny except None
2469 	 * - there is a conflict if a different client has any delegation
2470 	 * - there is a conflict if the same client has a read delegation
2471 	 *   (The current consensus is that this last case should be
2472 	 *    considered a conflict since the client with a read delegation
2473 	 *    could have done an Open with ReadAccess and WriteDeny
2474 	 *    locally and then not have checked for the WriteDeny.)
2475 	 * Don't check for a Reclaim, since that will be dealt with
2476 	 * by nfsrv_openctrl().
2477 	 */
2478 	if (!(new_stp->ls_flags &
2479 		(NFSLCK_DELEGPREV | NFSLCK_DELEGCUR | NFSLCK_RECLAIM))) {
2480 	    stp = LIST_FIRST(&lfp->lf_deleg);
2481 	    while (stp != LIST_END(&lfp->lf_deleg)) {
2482 		nstp = LIST_NEXT(stp, ls_file);
2483 		if ((readonly && stp->ls_clp != clp &&
2484 		       (stp->ls_flags & NFSLCK_DELEGWRITE)) ||
2485 		    (!readonly && (stp->ls_clp != clp ||
2486 		         (stp->ls_flags & NFSLCK_DELEGREAD)))) {
2487 			ret = nfsrv_delegconflict(stp, &haslock, p, vp);
2488 			if (ret) {
2489 			    /*
2490 			     * nfsrv_delegconflict() unlocks state
2491 			     * when it returns non-zero.
2492 			     */
2493 			    if (ret == -1)
2494 				goto tryagain;
2495 			    error = ret;
2496 			    goto out;
2497 			}
2498 		}
2499 		stp = nstp;
2500 	    }
2501 	}
2502 	NFSUNLOCKSTATE();
2503 	if (haslock) {
2504 		NFSLOCKV4ROOTMUTEX();
2505 		nfsv4_unlock(&nfsv4rootfs_lock, 1);
2506 		NFSUNLOCKV4ROOTMUTEX();
2507 	}
2508 
2509 out:
2510 	NFSEXITCODE2(error, nd);
2511 	return (error);
2512 }
2513 
2514 /*
2515  * Open control function to create/update open state for an open.
2516  */
2517 APPLESTATIC int
2518 nfsrv_openctrl(struct nfsrv_descript *nd, vnode_t vp,
2519     struct nfsstate **new_stpp, nfsquad_t clientid, nfsv4stateid_t *stateidp,
2520     nfsv4stateid_t *delegstateidp, u_int32_t *rflagsp, struct nfsexstuff *exp,
2521     NFSPROC_T *p, u_quad_t filerev)
2522 {
2523 	struct nfsstate *new_stp = *new_stpp;
2524 	struct nfsstate *stp, *nstp;
2525 	struct nfsstate *openstp = NULL, *new_open, *ownerstp, *new_deleg;
2526 	struct nfslockfile *lfp, *new_lfp;
2527 	struct nfsclient *clp;
2528 	int error = 0, haslock = 0, ret, delegate = 1, writedeleg = 1;
2529 	int readonly = 0, cbret = 1, getfhret = 0;
2530 	int gotstate = 0, len = 0;
2531 	u_char *clidp = NULL;
2532 
2533 	if ((new_stp->ls_flags & NFSLCK_SHAREBITS) == NFSLCK_READACCESS)
2534 		readonly = 1;
2535 	/*
2536 	 * Check for restart conditions (client and server).
2537 	 * (Paranoia, should have been detected by nfsrv_opencheck().)
2538 	 * If an error does show up, return NFSERR_EXPIRED, since the
2539 	 * the seqid# has already been incremented.
2540 	 */
2541 	error = nfsrv_checkrestart(clientid, new_stp->ls_flags,
2542 	    &new_stp->ls_stateid, 0);
2543 	if (error) {
2544 		printf("Nfsd: openctrl unexpected restart err=%d\n",
2545 		    error);
2546 		error = NFSERR_EXPIRED;
2547 		goto out;
2548 	}
2549 
2550 	clidp = malloc(NFSV4_OPAQUELIMIT, M_TEMP, M_WAITOK);
2551 tryagain:
2552 	new_lfp = malloc(sizeof (struct nfslockfile),
2553 	    M_NFSDLOCKFILE, M_WAITOK);
2554 	new_open = malloc(sizeof (struct nfsstate),
2555 	    M_NFSDSTATE, M_WAITOK);
2556 	new_deleg = malloc(sizeof (struct nfsstate),
2557 	    M_NFSDSTATE, M_WAITOK);
2558 	getfhret = nfsrv_getlockfh(vp, new_stp->ls_flags, new_lfp,
2559 	    NULL, p);
2560 	NFSLOCKSTATE();
2561 	/*
2562 	 * Get the client structure. Since the linked lists could be changed
2563 	 * by other nfsd processes if this process does a tsleep(), one of
2564 	 * two things must be done.
2565 	 * 1 - don't tsleep()
2566 	 * or
2567 	 * 2 - get the nfsv4_lock() { indicated by haslock == 1 }
2568 	 *     before using the lists, since this lock stops the other
2569 	 *     nfsd. This should only be used for rare cases, since it
2570 	 *     essentially single threads the nfsd.
2571 	 *     At this time, it is only done for cases where the stable
2572 	 *     storage file must be written prior to completion of state
2573 	 *     expiration.
2574 	 */
2575 	error = nfsrv_getclient(clientid, CLOPS_RENEW, &clp, NULL,
2576 	    (nfsquad_t)((u_quad_t)0), 0, nd, p);
2577 	if (!error && (clp->lc_flags & LCL_NEEDSCBNULL) &&
2578 	    clp->lc_program) {
2579 		/*
2580 		 * This happens on the first open for a client
2581 		 * that supports callbacks.
2582 		 */
2583 		NFSUNLOCKSTATE();
2584 		/*
2585 		 * Although nfsrv_docallback() will sleep, clp won't
2586 		 * go away, since they are only removed when the
2587 		 * nfsv4_lock() has blocked the nfsd threads. The
2588 		 * fields in clp can change, but having multiple
2589 		 * threads do this Null callback RPC should be
2590 		 * harmless.
2591 		 */
2592 		cbret = nfsrv_docallback(clp, NFSV4PROC_CBNULL,
2593 		    NULL, 0, NULL, NULL, NULL, p);
2594 		NFSLOCKSTATE();
2595 		clp->lc_flags &= ~LCL_NEEDSCBNULL;
2596 		if (!cbret)
2597 			clp->lc_flags |= LCL_CALLBACKSON;
2598 	}
2599 
2600 	/*
2601 	 * Look up the open owner. See if it needs confirmation and
2602 	 * check the seq#, as required.
2603 	 */
2604 	if (!error)
2605 		nfsrv_getowner(&clp->lc_open, new_stp, &ownerstp);
2606 
2607 	if (error) {
2608 		NFSUNLOCKSTATE();
2609 		printf("Nfsd: openctrl unexpected state err=%d\n",
2610 			error);
2611 		free(new_lfp, M_NFSDLOCKFILE);
2612 		free(new_open, M_NFSDSTATE);
2613 		free(new_deleg, M_NFSDSTATE);
2614 		if (haslock) {
2615 			NFSLOCKV4ROOTMUTEX();
2616 			nfsv4_unlock(&nfsv4rootfs_lock, 1);
2617 			NFSUNLOCKV4ROOTMUTEX();
2618 		}
2619 		error = NFSERR_EXPIRED;
2620 		goto out;
2621 	}
2622 
2623 	if (new_stp->ls_flags & NFSLCK_RECLAIM)
2624 		nfsrv_markstable(clp);
2625 
2626 	/*
2627 	 * Get the structure for the underlying file.
2628 	 */
2629 	if (getfhret)
2630 		error = getfhret;
2631 	else
2632 		error = nfsrv_getlockfile(new_stp->ls_flags, &new_lfp, &lfp,
2633 		    NULL, 0);
2634 	if (new_lfp)
2635 		free(new_lfp, M_NFSDLOCKFILE);
2636 	if (error) {
2637 		NFSUNLOCKSTATE();
2638 		printf("Nfsd openctrl unexpected getlockfile err=%d\n",
2639 		    error);
2640 		free(new_open, M_NFSDSTATE);
2641 		free(new_deleg, M_NFSDSTATE);
2642 		if (haslock) {
2643 			NFSLOCKV4ROOTMUTEX();
2644 			nfsv4_unlock(&nfsv4rootfs_lock, 1);
2645 			NFSUNLOCKV4ROOTMUTEX();
2646 		}
2647 		goto out;
2648 	}
2649 
2650 	/*
2651 	 * Search for a conflicting open/share.
2652 	 */
2653 	if (new_stp->ls_flags & NFSLCK_DELEGCUR) {
2654 	    /*
2655 	     * For Delegate_Cur, search for the matching Delegation,
2656 	     * which indicates no conflict.
2657 	     * An old delegation should have been recovered by the
2658 	     * client doing a Claim_DELEGATE_Prev, so I won't let
2659 	     * it match and return NFSERR_EXPIRED. Should I let it
2660 	     * match?
2661 	     */
2662 	    LIST_FOREACH(stp, &lfp->lf_deleg, ls_file) {
2663 		if (!(stp->ls_flags & NFSLCK_OLDDELEG) &&
2664 		    (((nd->nd_flag & ND_NFSV41) != 0 &&
2665 		    stateidp->seqid == 0) ||
2666 		    stateidp->seqid == stp->ls_stateid.seqid) &&
2667 		    !NFSBCMP(stateidp->other, stp->ls_stateid.other,
2668 			NFSX_STATEIDOTHER))
2669 			break;
2670 	    }
2671 	    if (stp == LIST_END(&lfp->lf_deleg) ||
2672 		((new_stp->ls_flags & NFSLCK_WRITEACCESS) &&
2673 		 (stp->ls_flags & NFSLCK_DELEGREAD))) {
2674 		NFSUNLOCKSTATE();
2675 		printf("Nfsd openctrl unexpected expiry\n");
2676 		free(new_open, M_NFSDSTATE);
2677 		free(new_deleg, M_NFSDSTATE);
2678 		if (haslock) {
2679 			NFSLOCKV4ROOTMUTEX();
2680 			nfsv4_unlock(&nfsv4rootfs_lock, 1);
2681 			NFSUNLOCKV4ROOTMUTEX();
2682 		}
2683 		error = NFSERR_EXPIRED;
2684 		goto out;
2685 	    }
2686 
2687 	    /*
2688 	     * Don't issue a Delegation, since one already exists and
2689 	     * delay delegation timeout, as required.
2690 	     */
2691 	    delegate = 0;
2692 	    nfsrv_delaydelegtimeout(stp);
2693 	}
2694 
2695 	/*
2696 	 * Check for access/deny bit conflicts. I also check for the
2697 	 * same owner, since the client might not have bothered to check.
2698 	 * Also, note an open for the same file and owner, if found,
2699 	 * which is all we do here for Delegate_Cur, since conflict
2700 	 * checking is already done.
2701 	 */
2702 	LIST_FOREACH(stp, &lfp->lf_open, ls_file) {
2703 		if (ownerstp && stp->ls_openowner == ownerstp)
2704 			openstp = stp;
2705 		if (!(new_stp->ls_flags & NFSLCK_DELEGCUR)) {
2706 		    /*
2707 		     * If another client has the file open, the only
2708 		     * delegation that can be issued is a Read delegation
2709 		     * and only if it is a Read open with Deny none.
2710 		     */
2711 		    if (clp != stp->ls_clp) {
2712 			if ((stp->ls_flags & NFSLCK_SHAREBITS) ==
2713 			    NFSLCK_READACCESS)
2714 			    writedeleg = 0;
2715 			else
2716 			    delegate = 0;
2717 		    }
2718 		    if(((new_stp->ls_flags & NFSLCK_ACCESSBITS) &
2719 		        ((stp->ls_flags>>NFSLCK_SHIFT) & NFSLCK_ACCESSBITS))||
2720 		       ((stp->ls_flags & NFSLCK_ACCESSBITS) &
2721 		        ((new_stp->ls_flags>>NFSLCK_SHIFT)&NFSLCK_ACCESSBITS))){
2722 			ret = nfsrv_clientconflict(stp->ls_clp,&haslock,vp,p);
2723 			if (ret == 1) {
2724 				/*
2725 				 * nfsrv_clientconflict() unlocks state
2726 				 * when it returns non-zero.
2727 				 */
2728 				free(new_open, M_NFSDSTATE);
2729 				free(new_deleg, M_NFSDSTATE);
2730 				openstp = NULL;
2731 				goto tryagain;
2732 			}
2733 			if (ret == 2)
2734 				error = NFSERR_PERM;
2735 			else if (new_stp->ls_flags & NFSLCK_RECLAIM)
2736 				error = NFSERR_RECLAIMCONFLICT;
2737 			else
2738 				error = NFSERR_SHAREDENIED;
2739 			if (ret == 0)
2740 				NFSUNLOCKSTATE();
2741 			if (haslock) {
2742 				NFSLOCKV4ROOTMUTEX();
2743 				nfsv4_unlock(&nfsv4rootfs_lock, 1);
2744 				NFSUNLOCKV4ROOTMUTEX();
2745 			}
2746 			free(new_open, M_NFSDSTATE);
2747 			free(new_deleg, M_NFSDSTATE);
2748 			printf("nfsd openctrl unexpected client cnfl\n");
2749 			goto out;
2750 		    }
2751 		}
2752 	}
2753 
2754 	/*
2755 	 * Check for a conflicting delegation. If one is found, call
2756 	 * nfsrv_delegconflict() to handle it. If the v4root lock hasn't
2757 	 * been set yet, it will get the lock. Otherwise, it will recall
2758 	 * the delegation. Then, we try try again...
2759 	 * (If NFSLCK_DELEGCUR is set, it has a delegation, so there
2760 	 *  isn't a conflict.)
2761 	 * I currently believe the conflict algorithm to be:
2762 	 * For Open with Read Access and Deny None
2763 	 * - there is a conflict iff a different client has a write delegation
2764 	 * For Open with other Write Access or any Deny except None
2765 	 * - there is a conflict if a different client has any delegation
2766 	 * - there is a conflict if the same client has a read delegation
2767 	 *   (The current consensus is that this last case should be
2768 	 *    considered a conflict since the client with a read delegation
2769 	 *    could have done an Open with ReadAccess and WriteDeny
2770 	 *    locally and then not have checked for the WriteDeny.)
2771 	 */
2772 	if (!(new_stp->ls_flags & (NFSLCK_DELEGPREV | NFSLCK_DELEGCUR))) {
2773 	    stp = LIST_FIRST(&lfp->lf_deleg);
2774 	    while (stp != LIST_END(&lfp->lf_deleg)) {
2775 		nstp = LIST_NEXT(stp, ls_file);
2776 		if (stp->ls_clp != clp && (stp->ls_flags & NFSLCK_DELEGREAD))
2777 			writedeleg = 0;
2778 		else
2779 			delegate = 0;
2780 		if ((readonly && stp->ls_clp != clp &&
2781 		       (stp->ls_flags & NFSLCK_DELEGWRITE)) ||
2782 		    (!readonly && (stp->ls_clp != clp ||
2783 		         (stp->ls_flags & NFSLCK_DELEGREAD)))) {
2784 		    if (new_stp->ls_flags & NFSLCK_RECLAIM) {
2785 			delegate = 2;
2786 		    } else {
2787 			ret = nfsrv_delegconflict(stp, &haslock, p, vp);
2788 			if (ret) {
2789 			    /*
2790 			     * nfsrv_delegconflict() unlocks state
2791 			     * when it returns non-zero.
2792 			     */
2793 			    printf("Nfsd openctrl unexpected deleg cnfl\n");
2794 			    free(new_open, M_NFSDSTATE);
2795 			    free(new_deleg, M_NFSDSTATE);
2796 			    if (ret == -1) {
2797 				openstp = NULL;
2798 				goto tryagain;
2799 			    }
2800 			    error = ret;
2801 			    goto out;
2802 			}
2803 		    }
2804 		}
2805 		stp = nstp;
2806 	    }
2807 	}
2808 
2809 	/*
2810 	 * We only get here if there was no open that conflicted.
2811 	 * If an open for the owner exists, or in the access/deny bits.
2812 	 * Otherwise it is a new open. If the open_owner hasn't been
2813 	 * confirmed, replace the open with the new one needing confirmation,
2814 	 * otherwise add the open.
2815 	 */
2816 	if (new_stp->ls_flags & NFSLCK_DELEGPREV) {
2817 	    /*
2818 	     * Handle NFSLCK_DELEGPREV by searching the old delegations for
2819 	     * a match. If found, just move the old delegation to the current
2820 	     * delegation list and issue open. If not found, return
2821 	     * NFSERR_EXPIRED.
2822 	     */
2823 	    LIST_FOREACH(stp, &clp->lc_olddeleg, ls_list) {
2824 		if (stp->ls_lfp == lfp) {
2825 		    /* Found it */
2826 		    if (stp->ls_clp != clp)
2827 			panic("olddeleg clp");
2828 		    LIST_REMOVE(stp, ls_list);
2829 		    LIST_REMOVE(stp, ls_hash);
2830 		    stp->ls_flags &= ~NFSLCK_OLDDELEG;
2831 		    stp->ls_stateid.seqid = delegstateidp->seqid = 1;
2832 		    stp->ls_stateid.other[0] = delegstateidp->other[0] =
2833 			clp->lc_clientid.lval[0];
2834 		    stp->ls_stateid.other[1] = delegstateidp->other[1] =
2835 			clp->lc_clientid.lval[1];
2836 		    stp->ls_stateid.other[2] = delegstateidp->other[2] =
2837 			nfsrv_nextstateindex(clp);
2838 		    stp->ls_compref = nd->nd_compref;
2839 		    LIST_INSERT_HEAD(&clp->lc_deleg, stp, ls_list);
2840 		    LIST_INSERT_HEAD(NFSSTATEHASH(clp,
2841 			stp->ls_stateid), stp, ls_hash);
2842 		    if (stp->ls_flags & NFSLCK_DELEGWRITE)
2843 			*rflagsp |= NFSV4OPEN_WRITEDELEGATE;
2844 		    else
2845 			*rflagsp |= NFSV4OPEN_READDELEGATE;
2846 		    clp->lc_delegtime = NFSD_MONOSEC +
2847 			nfsrv_lease + NFSRV_LEASEDELTA;
2848 
2849 		    /*
2850 		     * Now, do the associated open.
2851 		     */
2852 		    new_open->ls_stateid.seqid = 1;
2853 		    new_open->ls_stateid.other[0] = clp->lc_clientid.lval[0];
2854 		    new_open->ls_stateid.other[1] = clp->lc_clientid.lval[1];
2855 		    new_open->ls_stateid.other[2] = nfsrv_nextstateindex(clp);
2856 		    new_open->ls_flags = (new_stp->ls_flags&NFSLCK_DENYBITS)|
2857 			NFSLCK_OPEN;
2858 		    if (stp->ls_flags & NFSLCK_DELEGWRITE)
2859 			new_open->ls_flags |= (NFSLCK_READACCESS |
2860 			    NFSLCK_WRITEACCESS);
2861 		    else
2862 			new_open->ls_flags |= NFSLCK_READACCESS;
2863 		    new_open->ls_uid = new_stp->ls_uid;
2864 		    new_open->ls_lfp = lfp;
2865 		    new_open->ls_clp = clp;
2866 		    LIST_INIT(&new_open->ls_open);
2867 		    LIST_INSERT_HEAD(&lfp->lf_open, new_open, ls_file);
2868 		    LIST_INSERT_HEAD(NFSSTATEHASH(clp, new_open->ls_stateid),
2869 			new_open, ls_hash);
2870 		    /*
2871 		     * and handle the open owner
2872 		     */
2873 		    if (ownerstp) {
2874 			new_open->ls_openowner = ownerstp;
2875 			LIST_INSERT_HEAD(&ownerstp->ls_open,new_open,ls_list);
2876 		    } else {
2877 			new_open->ls_openowner = new_stp;
2878 			new_stp->ls_flags = 0;
2879 			nfsrvd_refcache(new_stp->ls_op);
2880 			new_stp->ls_noopens = 0;
2881 			LIST_INIT(&new_stp->ls_open);
2882 			LIST_INSERT_HEAD(&new_stp->ls_open, new_open, ls_list);
2883 			LIST_INSERT_HEAD(&clp->lc_open, new_stp, ls_list);
2884 			*new_stpp = NULL;
2885 			nfsstatsv1.srvopenowners++;
2886 			nfsrv_openpluslock++;
2887 		    }
2888 		    openstp = new_open;
2889 		    new_open = NULL;
2890 		    nfsstatsv1.srvopens++;
2891 		    nfsrv_openpluslock++;
2892 		    break;
2893 		}
2894 	    }
2895 	    if (stp == LIST_END(&clp->lc_olddeleg))
2896 		error = NFSERR_EXPIRED;
2897 	} else if (new_stp->ls_flags & (NFSLCK_DELEGREAD | NFSLCK_DELEGWRITE)) {
2898 	    /*
2899 	     * Scan to see that no delegation for this client and file
2900 	     * doesn't already exist.
2901 	     * There also shouldn't yet be an Open for this file and
2902 	     * openowner.
2903 	     */
2904 	    LIST_FOREACH(stp, &lfp->lf_deleg, ls_file) {
2905 		if (stp->ls_clp == clp)
2906 		    break;
2907 	    }
2908 	    if (stp == LIST_END(&lfp->lf_deleg) && openstp == NULL) {
2909 		/*
2910 		 * This is the Claim_Previous case with a delegation
2911 		 * type != Delegate_None.
2912 		 */
2913 		/*
2914 		 * First, add the delegation. (Although we must issue the
2915 		 * delegation, we can also ask for an immediate return.)
2916 		 */
2917 		new_deleg->ls_stateid.seqid = delegstateidp->seqid = 1;
2918 		new_deleg->ls_stateid.other[0] = delegstateidp->other[0] =
2919 		    clp->lc_clientid.lval[0];
2920 		new_deleg->ls_stateid.other[1] = delegstateidp->other[1] =
2921 		    clp->lc_clientid.lval[1];
2922 		new_deleg->ls_stateid.other[2] = delegstateidp->other[2] =
2923 		    nfsrv_nextstateindex(clp);
2924 		if (new_stp->ls_flags & NFSLCK_DELEGWRITE) {
2925 		    new_deleg->ls_flags = (NFSLCK_DELEGWRITE |
2926 			NFSLCK_READACCESS | NFSLCK_WRITEACCESS);
2927 		    *rflagsp |= NFSV4OPEN_WRITEDELEGATE;
2928 		    nfsrv_writedelegcnt++;
2929 		} else {
2930 		    new_deleg->ls_flags = (NFSLCK_DELEGREAD |
2931 			NFSLCK_READACCESS);
2932 		    *rflagsp |= NFSV4OPEN_READDELEGATE;
2933 		}
2934 		new_deleg->ls_uid = new_stp->ls_uid;
2935 		new_deleg->ls_lfp = lfp;
2936 		new_deleg->ls_clp = clp;
2937 		new_deleg->ls_filerev = filerev;
2938 		new_deleg->ls_compref = nd->nd_compref;
2939 		LIST_INSERT_HEAD(&lfp->lf_deleg, new_deleg, ls_file);
2940 		LIST_INSERT_HEAD(NFSSTATEHASH(clp,
2941 		    new_deleg->ls_stateid), new_deleg, ls_hash);
2942 		LIST_INSERT_HEAD(&clp->lc_deleg, new_deleg, ls_list);
2943 		new_deleg = NULL;
2944 		if (delegate == 2 || nfsrv_issuedelegs == 0 ||
2945 		    (clp->lc_flags & (LCL_CALLBACKSON | LCL_CBDOWN)) !=
2946 		     LCL_CALLBACKSON ||
2947 		    NFSRV_V4DELEGLIMIT(nfsrv_delegatecnt) ||
2948 		    !NFSVNO_DELEGOK(vp))
2949 		    *rflagsp |= NFSV4OPEN_RECALL;
2950 		nfsstatsv1.srvdelegates++;
2951 		nfsrv_openpluslock++;
2952 		nfsrv_delegatecnt++;
2953 
2954 		/*
2955 		 * Now, do the associated open.
2956 		 */
2957 		new_open->ls_stateid.seqid = 1;
2958 		new_open->ls_stateid.other[0] = clp->lc_clientid.lval[0];
2959 		new_open->ls_stateid.other[1] = clp->lc_clientid.lval[1];
2960 		new_open->ls_stateid.other[2] = nfsrv_nextstateindex(clp);
2961 		new_open->ls_flags = (new_stp->ls_flags & NFSLCK_DENYBITS) |
2962 		    NFSLCK_OPEN;
2963 		if (new_stp->ls_flags & NFSLCK_DELEGWRITE)
2964 			new_open->ls_flags |= (NFSLCK_READACCESS |
2965 			    NFSLCK_WRITEACCESS);
2966 		else
2967 			new_open->ls_flags |= NFSLCK_READACCESS;
2968 		new_open->ls_uid = new_stp->ls_uid;
2969 		new_open->ls_lfp = lfp;
2970 		new_open->ls_clp = clp;
2971 		LIST_INIT(&new_open->ls_open);
2972 		LIST_INSERT_HEAD(&lfp->lf_open, new_open, ls_file);
2973 		LIST_INSERT_HEAD(NFSSTATEHASH(clp, new_open->ls_stateid),
2974 		   new_open, ls_hash);
2975 		/*
2976 		 * and handle the open owner
2977 		 */
2978 		if (ownerstp) {
2979 		    new_open->ls_openowner = ownerstp;
2980 		    LIST_INSERT_HEAD(&ownerstp->ls_open, new_open, ls_list);
2981 		} else {
2982 		    new_open->ls_openowner = new_stp;
2983 		    new_stp->ls_flags = 0;
2984 		    nfsrvd_refcache(new_stp->ls_op);
2985 		    new_stp->ls_noopens = 0;
2986 		    LIST_INIT(&new_stp->ls_open);
2987 		    LIST_INSERT_HEAD(&new_stp->ls_open, new_open, ls_list);
2988 		    LIST_INSERT_HEAD(&clp->lc_open, new_stp, ls_list);
2989 		    *new_stpp = NULL;
2990 		    nfsstatsv1.srvopenowners++;
2991 		    nfsrv_openpluslock++;
2992 		}
2993 		openstp = new_open;
2994 		new_open = NULL;
2995 		nfsstatsv1.srvopens++;
2996 		nfsrv_openpluslock++;
2997 	    } else {
2998 		error = NFSERR_RECLAIMCONFLICT;
2999 	    }
3000 	} else if (ownerstp) {
3001 		if (ownerstp->ls_flags & NFSLCK_NEEDSCONFIRM) {
3002 		    /* Replace the open */
3003 		    if (ownerstp->ls_op)
3004 			nfsrvd_derefcache(ownerstp->ls_op);
3005 		    ownerstp->ls_op = new_stp->ls_op;
3006 		    nfsrvd_refcache(ownerstp->ls_op);
3007 		    ownerstp->ls_seq = new_stp->ls_seq;
3008 		    *rflagsp |= NFSV4OPEN_RESULTCONFIRM;
3009 		    stp = LIST_FIRST(&ownerstp->ls_open);
3010 		    stp->ls_flags = (new_stp->ls_flags & NFSLCK_SHAREBITS) |
3011 			NFSLCK_OPEN;
3012 		    stp->ls_stateid.seqid = 1;
3013 		    stp->ls_uid = new_stp->ls_uid;
3014 		    if (lfp != stp->ls_lfp) {
3015 			LIST_REMOVE(stp, ls_file);
3016 			LIST_INSERT_HEAD(&lfp->lf_open, stp, ls_file);
3017 			stp->ls_lfp = lfp;
3018 		    }
3019 		    openstp = stp;
3020 		} else if (openstp) {
3021 		    openstp->ls_flags |= (new_stp->ls_flags & NFSLCK_SHAREBITS);
3022 		    openstp->ls_stateid.seqid++;
3023 		    if ((nd->nd_flag & ND_NFSV41) != 0 &&
3024 			openstp->ls_stateid.seqid == 0)
3025 			openstp->ls_stateid.seqid = 1;
3026 
3027 		    /*
3028 		     * This is where we can choose to issue a delegation.
3029 		     */
3030 		    if (delegate == 0 || writedeleg == 0 ||
3031 			NFSVNO_EXRDONLY(exp) || (readonly != 0 &&
3032 			nfsrv_writedelegifpos == 0) ||
3033 			!NFSVNO_DELEGOK(vp) ||
3034 			(new_stp->ls_flags & NFSLCK_WANTRDELEG) != 0 ||
3035 			(clp->lc_flags & (LCL_CALLBACKSON | LCL_CBDOWN)) !=
3036 			 LCL_CALLBACKSON)
3037 			*rflagsp |= NFSV4OPEN_WDCONTENTION;
3038 		    else if (nfsrv_issuedelegs == 0 ||
3039 			NFSRV_V4DELEGLIMIT(nfsrv_delegatecnt))
3040 			*rflagsp |= NFSV4OPEN_WDRESOURCE;
3041 		    else if ((new_stp->ls_flags & NFSLCK_WANTNODELEG) != 0)
3042 			*rflagsp |= NFSV4OPEN_WDNOTWANTED;
3043 		    else {
3044 			new_deleg->ls_stateid.seqid = delegstateidp->seqid = 1;
3045 			new_deleg->ls_stateid.other[0] = delegstateidp->other[0]
3046 			    = clp->lc_clientid.lval[0];
3047 			new_deleg->ls_stateid.other[1] = delegstateidp->other[1]
3048 			    = clp->lc_clientid.lval[1];
3049 			new_deleg->ls_stateid.other[2] = delegstateidp->other[2]
3050 			    = nfsrv_nextstateindex(clp);
3051 			new_deleg->ls_flags = (NFSLCK_DELEGWRITE |
3052 			    NFSLCK_READACCESS | NFSLCK_WRITEACCESS);
3053 			*rflagsp |= NFSV4OPEN_WRITEDELEGATE;
3054 			new_deleg->ls_uid = new_stp->ls_uid;
3055 			new_deleg->ls_lfp = lfp;
3056 			new_deleg->ls_clp = clp;
3057 			new_deleg->ls_filerev = filerev;
3058 			new_deleg->ls_compref = nd->nd_compref;
3059 			nfsrv_writedelegcnt++;
3060 			LIST_INSERT_HEAD(&lfp->lf_deleg, new_deleg, ls_file);
3061 			LIST_INSERT_HEAD(NFSSTATEHASH(clp,
3062 			    new_deleg->ls_stateid), new_deleg, ls_hash);
3063 			LIST_INSERT_HEAD(&clp->lc_deleg, new_deleg, ls_list);
3064 			new_deleg = NULL;
3065 			nfsstatsv1.srvdelegates++;
3066 			nfsrv_openpluslock++;
3067 			nfsrv_delegatecnt++;
3068 		    }
3069 		} else {
3070 		    new_open->ls_stateid.seqid = 1;
3071 		    new_open->ls_stateid.other[0] = clp->lc_clientid.lval[0];
3072 		    new_open->ls_stateid.other[1] = clp->lc_clientid.lval[1];
3073 		    new_open->ls_stateid.other[2] = nfsrv_nextstateindex(clp);
3074 		    new_open->ls_flags = (new_stp->ls_flags & NFSLCK_SHAREBITS)|
3075 			NFSLCK_OPEN;
3076 		    new_open->ls_uid = new_stp->ls_uid;
3077 		    new_open->ls_openowner = ownerstp;
3078 		    new_open->ls_lfp = lfp;
3079 		    new_open->ls_clp = clp;
3080 		    LIST_INIT(&new_open->ls_open);
3081 		    LIST_INSERT_HEAD(&lfp->lf_open, new_open, ls_file);
3082 		    LIST_INSERT_HEAD(&ownerstp->ls_open, new_open, ls_list);
3083 		    LIST_INSERT_HEAD(NFSSTATEHASH(clp, new_open->ls_stateid),
3084 			new_open, ls_hash);
3085 		    openstp = new_open;
3086 		    new_open = NULL;
3087 		    nfsstatsv1.srvopens++;
3088 		    nfsrv_openpluslock++;
3089 
3090 		    /*
3091 		     * This is where we can choose to issue a delegation.
3092 		     */
3093 		    if (delegate == 0 || (writedeleg == 0 && readonly == 0) ||
3094 			!NFSVNO_DELEGOK(vp) ||
3095 			(clp->lc_flags & (LCL_CALLBACKSON | LCL_CBDOWN)) !=
3096 			 LCL_CALLBACKSON)
3097 			*rflagsp |= NFSV4OPEN_WDCONTENTION;
3098 		    else if (nfsrv_issuedelegs == 0 ||
3099 			NFSRV_V4DELEGLIMIT(nfsrv_delegatecnt))
3100 			*rflagsp |= NFSV4OPEN_WDRESOURCE;
3101 		    else if ((new_stp->ls_flags & NFSLCK_WANTNODELEG) != 0)
3102 			*rflagsp |= NFSV4OPEN_WDNOTWANTED;
3103 		    else {
3104 			new_deleg->ls_stateid.seqid = delegstateidp->seqid = 1;
3105 			new_deleg->ls_stateid.other[0] = delegstateidp->other[0]
3106 			    = clp->lc_clientid.lval[0];
3107 			new_deleg->ls_stateid.other[1] = delegstateidp->other[1]
3108 			    = clp->lc_clientid.lval[1];
3109 			new_deleg->ls_stateid.other[2] = delegstateidp->other[2]
3110 			    = nfsrv_nextstateindex(clp);
3111 			if (writedeleg && !NFSVNO_EXRDONLY(exp) &&
3112 			    (nfsrv_writedelegifpos || !readonly) &&
3113 			    (new_stp->ls_flags & NFSLCK_WANTRDELEG) == 0) {
3114 			    new_deleg->ls_flags = (NFSLCK_DELEGWRITE |
3115 				NFSLCK_READACCESS | NFSLCK_WRITEACCESS);
3116 			    *rflagsp |= NFSV4OPEN_WRITEDELEGATE;
3117 			    nfsrv_writedelegcnt++;
3118 			} else {
3119 			    new_deleg->ls_flags = (NFSLCK_DELEGREAD |
3120 				NFSLCK_READACCESS);
3121 			    *rflagsp |= NFSV4OPEN_READDELEGATE;
3122 			}
3123 			new_deleg->ls_uid = new_stp->ls_uid;
3124 			new_deleg->ls_lfp = lfp;
3125 			new_deleg->ls_clp = clp;
3126 			new_deleg->ls_filerev = filerev;
3127 			new_deleg->ls_compref = nd->nd_compref;
3128 			LIST_INSERT_HEAD(&lfp->lf_deleg, new_deleg, ls_file);
3129 			LIST_INSERT_HEAD(NFSSTATEHASH(clp,
3130 			    new_deleg->ls_stateid), new_deleg, ls_hash);
3131 			LIST_INSERT_HEAD(&clp->lc_deleg, new_deleg, ls_list);
3132 			new_deleg = NULL;
3133 			nfsstatsv1.srvdelegates++;
3134 			nfsrv_openpluslock++;
3135 			nfsrv_delegatecnt++;
3136 		    }
3137 		}
3138 	} else {
3139 		/*
3140 		 * New owner case. Start the open_owner sequence with a
3141 		 * Needs confirmation (unless a reclaim) and hang the
3142 		 * new open off it.
3143 		 */
3144 		new_open->ls_stateid.seqid = 1;
3145 		new_open->ls_stateid.other[0] = clp->lc_clientid.lval[0];
3146 		new_open->ls_stateid.other[1] = clp->lc_clientid.lval[1];
3147 		new_open->ls_stateid.other[2] = nfsrv_nextstateindex(clp);
3148 		new_open->ls_flags = (new_stp->ls_flags & NFSLCK_SHAREBITS) |
3149 		    NFSLCK_OPEN;
3150 		new_open->ls_uid = new_stp->ls_uid;
3151 		LIST_INIT(&new_open->ls_open);
3152 		new_open->ls_openowner = new_stp;
3153 		new_open->ls_lfp = lfp;
3154 		new_open->ls_clp = clp;
3155 		LIST_INSERT_HEAD(&lfp->lf_open, new_open, ls_file);
3156 		if (new_stp->ls_flags & NFSLCK_RECLAIM) {
3157 			new_stp->ls_flags = 0;
3158 		} else if ((nd->nd_flag & ND_NFSV41) != 0) {
3159 			/* NFSv4.1 never needs confirmation. */
3160 			new_stp->ls_flags = 0;
3161 
3162 			/*
3163 			 * This is where we can choose to issue a delegation.
3164 			 */
3165 			if (delegate && nfsrv_issuedelegs &&
3166 			    (writedeleg || readonly) &&
3167 			    (clp->lc_flags & (LCL_CALLBACKSON | LCL_CBDOWN)) ==
3168 			     LCL_CALLBACKSON &&
3169 			    !NFSRV_V4DELEGLIMIT(nfsrv_delegatecnt) &&
3170 			    NFSVNO_DELEGOK(vp) &&
3171 			    ((nd->nd_flag & ND_NFSV41) == 0 ||
3172 			     (new_stp->ls_flags & NFSLCK_WANTNODELEG) == 0)) {
3173 				new_deleg->ls_stateid.seqid =
3174 				    delegstateidp->seqid = 1;
3175 				new_deleg->ls_stateid.other[0] =
3176 				    delegstateidp->other[0]
3177 				    = clp->lc_clientid.lval[0];
3178 				new_deleg->ls_stateid.other[1] =
3179 				    delegstateidp->other[1]
3180 				    = clp->lc_clientid.lval[1];
3181 				new_deleg->ls_stateid.other[2] =
3182 				    delegstateidp->other[2]
3183 				    = nfsrv_nextstateindex(clp);
3184 				if (writedeleg && !NFSVNO_EXRDONLY(exp) &&
3185 				    (nfsrv_writedelegifpos || !readonly) &&
3186 				    ((nd->nd_flag & ND_NFSV41) == 0 ||
3187 				     (new_stp->ls_flags & NFSLCK_WANTRDELEG) ==
3188 				     0)) {
3189 					new_deleg->ls_flags =
3190 					    (NFSLCK_DELEGWRITE |
3191 					     NFSLCK_READACCESS |
3192 					     NFSLCK_WRITEACCESS);
3193 					*rflagsp |= NFSV4OPEN_WRITEDELEGATE;
3194 					nfsrv_writedelegcnt++;
3195 				} else {
3196 					new_deleg->ls_flags =
3197 					    (NFSLCK_DELEGREAD |
3198 					     NFSLCK_READACCESS);
3199 					*rflagsp |= NFSV4OPEN_READDELEGATE;
3200 				}
3201 				new_deleg->ls_uid = new_stp->ls_uid;
3202 				new_deleg->ls_lfp = lfp;
3203 				new_deleg->ls_clp = clp;
3204 				new_deleg->ls_filerev = filerev;
3205 				new_deleg->ls_compref = nd->nd_compref;
3206 				LIST_INSERT_HEAD(&lfp->lf_deleg, new_deleg,
3207 				    ls_file);
3208 				LIST_INSERT_HEAD(NFSSTATEHASH(clp,
3209 				    new_deleg->ls_stateid), new_deleg, ls_hash);
3210 				LIST_INSERT_HEAD(&clp->lc_deleg, new_deleg,
3211 				    ls_list);
3212 				new_deleg = NULL;
3213 				nfsstatsv1.srvdelegates++;
3214 				nfsrv_openpluslock++;
3215 				nfsrv_delegatecnt++;
3216 			}
3217 			/*
3218 			 * Since NFSv4.1 never does an OpenConfirm, the first
3219 			 * open state will be acquired here.
3220 			 */
3221 			if (!(clp->lc_flags & LCL_STAMPEDSTABLE)) {
3222 				clp->lc_flags |= LCL_STAMPEDSTABLE;
3223 				len = clp->lc_idlen;
3224 				NFSBCOPY(clp->lc_id, clidp, len);
3225 				gotstate = 1;
3226 			}
3227 		} else {
3228 			*rflagsp |= NFSV4OPEN_RESULTCONFIRM;
3229 			new_stp->ls_flags = NFSLCK_NEEDSCONFIRM;
3230 		}
3231 		nfsrvd_refcache(new_stp->ls_op);
3232 		new_stp->ls_noopens = 0;
3233 		LIST_INIT(&new_stp->ls_open);
3234 		LIST_INSERT_HEAD(&new_stp->ls_open, new_open, ls_list);
3235 		LIST_INSERT_HEAD(&clp->lc_open, new_stp, ls_list);
3236 		LIST_INSERT_HEAD(NFSSTATEHASH(clp, new_open->ls_stateid),
3237 		    new_open, ls_hash);
3238 		openstp = new_open;
3239 		new_open = NULL;
3240 		*new_stpp = NULL;
3241 		nfsstatsv1.srvopens++;
3242 		nfsrv_openpluslock++;
3243 		nfsstatsv1.srvopenowners++;
3244 		nfsrv_openpluslock++;
3245 	}
3246 	if (!error) {
3247 		stateidp->seqid = openstp->ls_stateid.seqid;
3248 		stateidp->other[0] = openstp->ls_stateid.other[0];
3249 		stateidp->other[1] = openstp->ls_stateid.other[1];
3250 		stateidp->other[2] = openstp->ls_stateid.other[2];
3251 	}
3252 	NFSUNLOCKSTATE();
3253 	if (haslock) {
3254 		NFSLOCKV4ROOTMUTEX();
3255 		nfsv4_unlock(&nfsv4rootfs_lock, 1);
3256 		NFSUNLOCKV4ROOTMUTEX();
3257 	}
3258 	if (new_open)
3259 		free(new_open, M_NFSDSTATE);
3260 	if (new_deleg)
3261 		free(new_deleg, M_NFSDSTATE);
3262 
3263 	/*
3264 	 * If the NFSv4.1 client just acquired its first open, write a timestamp
3265 	 * to the stable storage file.
3266 	 */
3267 	if (gotstate != 0) {
3268 		nfsrv_writestable(clidp, len, NFSNST_NEWSTATE, p);
3269 		nfsrv_backupstable();
3270 	}
3271 
3272 out:
3273 	free(clidp, M_TEMP);
3274 	NFSEXITCODE2(error, nd);
3275 	return (error);
3276 }
3277 
3278 /*
3279  * Open update. Does the confirm, downgrade and close.
3280  */
3281 APPLESTATIC int
3282 nfsrv_openupdate(vnode_t vp, struct nfsstate *new_stp, nfsquad_t clientid,
3283     nfsv4stateid_t *stateidp, struct nfsrv_descript *nd, NFSPROC_T *p)
3284 {
3285 	struct nfsstate *stp;
3286 	struct nfsclient *clp;
3287 	struct nfslockfile *lfp;
3288 	u_int32_t bits;
3289 	int error = 0, gotstate = 0, len = 0;
3290 	u_char *clidp = NULL;
3291 
3292 	/*
3293 	 * Check for restart conditions (client and server).
3294 	 */
3295 	error = nfsrv_checkrestart(clientid, new_stp->ls_flags,
3296 	    &new_stp->ls_stateid, 0);
3297 	if (error)
3298 		goto out;
3299 
3300 	clidp = malloc(NFSV4_OPAQUELIMIT, M_TEMP, M_WAITOK);
3301 	NFSLOCKSTATE();
3302 	/*
3303 	 * Get the open structure via clientid and stateid.
3304 	 */
3305 	error = nfsrv_getclient(clientid, CLOPS_RENEW, &clp, NULL,
3306 	    (nfsquad_t)((u_quad_t)0), 0, nd, p);
3307 	if (!error)
3308 		error = nfsrv_getstate(clp, &new_stp->ls_stateid,
3309 		    new_stp->ls_flags, &stp);
3310 
3311 	/*
3312 	 * Sanity check the open.
3313 	 */
3314 	if (!error && (!(stp->ls_flags & NFSLCK_OPEN) ||
3315 		(!(new_stp->ls_flags & NFSLCK_CONFIRM) &&
3316 		 (stp->ls_openowner->ls_flags & NFSLCK_NEEDSCONFIRM)) ||
3317 		((new_stp->ls_flags & NFSLCK_CONFIRM) &&
3318 		 (!(stp->ls_openowner->ls_flags & NFSLCK_NEEDSCONFIRM)))))
3319 		error = NFSERR_BADSTATEID;
3320 
3321 	if (!error)
3322 		error = nfsrv_checkseqid(nd, new_stp->ls_seq,
3323 		    stp->ls_openowner, new_stp->ls_op);
3324 	if (!error && stp->ls_stateid.seqid != new_stp->ls_stateid.seqid &&
3325 	    (((nd->nd_flag & ND_NFSV41) == 0 &&
3326 	      !(new_stp->ls_flags & NFSLCK_CONFIRM)) ||
3327 	     ((nd->nd_flag & ND_NFSV41) != 0 &&
3328 	      new_stp->ls_stateid.seqid != 0)))
3329 		error = NFSERR_OLDSTATEID;
3330 	if (!error && vnode_vtype(vp) != VREG) {
3331 		if (vnode_vtype(vp) == VDIR)
3332 			error = NFSERR_ISDIR;
3333 		else
3334 			error = NFSERR_INVAL;
3335 	}
3336 
3337 	if (error) {
3338 		/*
3339 		 * If a client tries to confirm an Open with a bad
3340 		 * seqid# and there are no byte range locks or other Opens
3341 		 * on the openowner, just throw it away, so the next use of the
3342 		 * openowner will start a fresh seq#.
3343 		 */
3344 		if (error == NFSERR_BADSEQID &&
3345 		    (new_stp->ls_flags & NFSLCK_CONFIRM) &&
3346 		    nfsrv_nootherstate(stp))
3347 			nfsrv_freeopenowner(stp->ls_openowner, 0, p);
3348 		NFSUNLOCKSTATE();
3349 		goto out;
3350 	}
3351 
3352 	/*
3353 	 * Set the return stateid.
3354 	 */
3355 	stateidp->seqid = stp->ls_stateid.seqid + 1;
3356 	if ((nd->nd_flag & ND_NFSV41) != 0 && stateidp->seqid == 0)
3357 		stateidp->seqid = 1;
3358 	stateidp->other[0] = stp->ls_stateid.other[0];
3359 	stateidp->other[1] = stp->ls_stateid.other[1];
3360 	stateidp->other[2] = stp->ls_stateid.other[2];
3361 	/*
3362 	 * Now, handle the three cases.
3363 	 */
3364 	if (new_stp->ls_flags & NFSLCK_CONFIRM) {
3365 		/*
3366 		 * If the open doesn't need confirmation, it seems to me that
3367 		 * there is a client error, but I'll just log it and keep going?
3368 		 */
3369 		if (!(stp->ls_openowner->ls_flags & NFSLCK_NEEDSCONFIRM))
3370 			printf("Nfsv4d: stray open confirm\n");
3371 		stp->ls_openowner->ls_flags = 0;
3372 		stp->ls_stateid.seqid++;
3373 		if ((nd->nd_flag & ND_NFSV41) != 0 &&
3374 		    stp->ls_stateid.seqid == 0)
3375 			stp->ls_stateid.seqid = 1;
3376 		if (!(clp->lc_flags & LCL_STAMPEDSTABLE)) {
3377 			clp->lc_flags |= LCL_STAMPEDSTABLE;
3378 			len = clp->lc_idlen;
3379 			NFSBCOPY(clp->lc_id, clidp, len);
3380 			gotstate = 1;
3381 		}
3382 		NFSUNLOCKSTATE();
3383 	} else if (new_stp->ls_flags & NFSLCK_CLOSE) {
3384 		lfp = stp->ls_lfp;
3385 		if (nfsrv_dolocallocks != 0 && !LIST_EMPTY(&stp->ls_open)) {
3386 			/* Get the lf lock */
3387 			nfsrv_locklf(lfp);
3388 			NFSUNLOCKSTATE();
3389 			ASSERT_VOP_ELOCKED(vp, "nfsrv_openupdate");
3390 			NFSVOPUNLOCK(vp, 0);
3391 			if (nfsrv_freeopen(stp, vp, 1, p) == 0) {
3392 				NFSLOCKSTATE();
3393 				nfsrv_unlocklf(lfp);
3394 				NFSUNLOCKSTATE();
3395 			}
3396 			NFSVOPLOCK(vp, LK_EXCLUSIVE | LK_RETRY);
3397 		} else {
3398 			(void) nfsrv_freeopen(stp, NULL, 0, p);
3399 			NFSUNLOCKSTATE();
3400 		}
3401 	} else {
3402 		/*
3403 		 * Update the share bits, making sure that the new set are a
3404 		 * subset of the old ones.
3405 		 */
3406 		bits = (new_stp->ls_flags & NFSLCK_SHAREBITS);
3407 		if (~(stp->ls_flags) & bits) {
3408 			NFSUNLOCKSTATE();
3409 			error = NFSERR_INVAL;
3410 			goto out;
3411 		}
3412 		stp->ls_flags = (bits | NFSLCK_OPEN);
3413 		stp->ls_stateid.seqid++;
3414 		if ((nd->nd_flag & ND_NFSV41) != 0 &&
3415 		    stp->ls_stateid.seqid == 0)
3416 			stp->ls_stateid.seqid = 1;
3417 		NFSUNLOCKSTATE();
3418 	}
3419 
3420 	/*
3421 	 * If the client just confirmed its first open, write a timestamp
3422 	 * to the stable storage file.
3423 	 */
3424 	if (gotstate != 0) {
3425 		nfsrv_writestable(clidp, len, NFSNST_NEWSTATE, p);
3426 		nfsrv_backupstable();
3427 	}
3428 
3429 out:
3430 	free(clidp, M_TEMP);
3431 	NFSEXITCODE2(error, nd);
3432 	return (error);
3433 }
3434 
3435 /*
3436  * Delegation update. Does the purge and return.
3437  */
3438 APPLESTATIC int
3439 nfsrv_delegupdate(struct nfsrv_descript *nd, nfsquad_t clientid,
3440     nfsv4stateid_t *stateidp, vnode_t vp, int op, struct ucred *cred,
3441     NFSPROC_T *p)
3442 {
3443 	struct nfsstate *stp;
3444 	struct nfsclient *clp;
3445 	int error = 0;
3446 	fhandle_t fh;
3447 
3448 	/*
3449 	 * Do a sanity check against the file handle for DelegReturn.
3450 	 */
3451 	if (vp) {
3452 		error = nfsvno_getfh(vp, &fh, p);
3453 		if (error)
3454 			goto out;
3455 	}
3456 	/*
3457 	 * Check for restart conditions (client and server).
3458 	 */
3459 	if (op == NFSV4OP_DELEGRETURN)
3460 		error = nfsrv_checkrestart(clientid, NFSLCK_DELEGRETURN,
3461 			stateidp, 0);
3462 	else
3463 		error = nfsrv_checkrestart(clientid, NFSLCK_DELEGPURGE,
3464 			stateidp, 0);
3465 
3466 	NFSLOCKSTATE();
3467 	/*
3468 	 * Get the open structure via clientid and stateid.
3469 	 */
3470 	if (!error)
3471 	    error = nfsrv_getclient(clientid, CLOPS_RENEW, &clp, NULL,
3472 		(nfsquad_t)((u_quad_t)0), 0, nd, p);
3473 	if (error) {
3474 		if (error == NFSERR_CBPATHDOWN)
3475 			error = 0;
3476 		if (error == NFSERR_STALECLIENTID && op == NFSV4OP_DELEGRETURN)
3477 			error = NFSERR_STALESTATEID;
3478 	}
3479 	if (!error && op == NFSV4OP_DELEGRETURN) {
3480 	    error = nfsrv_getstate(clp, stateidp, NFSLCK_DELEGRETURN, &stp);
3481 	    if (!error && stp->ls_stateid.seqid != stateidp->seqid &&
3482 		((nd->nd_flag & ND_NFSV41) == 0 || stateidp->seqid != 0))
3483 		error = NFSERR_OLDSTATEID;
3484 	}
3485 	/*
3486 	 * NFSERR_EXPIRED means that the state has gone away,
3487 	 * so Delegations have been purged. Just return ok.
3488 	 */
3489 	if (error == NFSERR_EXPIRED && op == NFSV4OP_DELEGPURGE) {
3490 		NFSUNLOCKSTATE();
3491 		error = 0;
3492 		goto out;
3493 	}
3494 	if (error) {
3495 		NFSUNLOCKSTATE();
3496 		goto out;
3497 	}
3498 
3499 	if (op == NFSV4OP_DELEGRETURN) {
3500 		if (NFSBCMP((caddr_t)&fh, (caddr_t)&stp->ls_lfp->lf_fh,
3501 		    sizeof (fhandle_t))) {
3502 			NFSUNLOCKSTATE();
3503 			error = NFSERR_BADSTATEID;
3504 			goto out;
3505 		}
3506 		nfsrv_freedeleg(stp);
3507 	} else {
3508 		nfsrv_freedeleglist(&clp->lc_olddeleg);
3509 	}
3510 	NFSUNLOCKSTATE();
3511 	error = 0;
3512 
3513 out:
3514 	NFSEXITCODE(error);
3515 	return (error);
3516 }
3517 
3518 /*
3519  * Release lock owner.
3520  */
3521 APPLESTATIC int
3522 nfsrv_releaselckown(struct nfsstate *new_stp, nfsquad_t clientid,
3523     NFSPROC_T *p)
3524 {
3525 	struct nfsstate *stp, *nstp, *openstp, *ownstp;
3526 	struct nfsclient *clp;
3527 	int error = 0;
3528 
3529 	/*
3530 	 * Check for restart conditions (client and server).
3531 	 */
3532 	error = nfsrv_checkrestart(clientid, new_stp->ls_flags,
3533 	    &new_stp->ls_stateid, 0);
3534 	if (error)
3535 		goto out;
3536 
3537 	NFSLOCKSTATE();
3538 	/*
3539 	 * Get the lock owner by name.
3540 	 */
3541 	error = nfsrv_getclient(clientid, CLOPS_RENEW, &clp, NULL,
3542 	    (nfsquad_t)((u_quad_t)0), 0, NULL, p);
3543 	if (error) {
3544 		NFSUNLOCKSTATE();
3545 		goto out;
3546 	}
3547 	LIST_FOREACH(ownstp, &clp->lc_open, ls_list) {
3548 	    LIST_FOREACH(openstp, &ownstp->ls_open, ls_list) {
3549 		stp = LIST_FIRST(&openstp->ls_open);
3550 		while (stp != LIST_END(&openstp->ls_open)) {
3551 		    nstp = LIST_NEXT(stp, ls_list);
3552 		    /*
3553 		     * If the owner matches, check for locks and
3554 		     * then free or return an error.
3555 		     */
3556 		    if (stp->ls_ownerlen == new_stp->ls_ownerlen &&
3557 			!NFSBCMP(stp->ls_owner, new_stp->ls_owner,
3558 			 stp->ls_ownerlen)){
3559 			if (LIST_EMPTY(&stp->ls_lock)) {
3560 			    nfsrv_freelockowner(stp, NULL, 0, p);
3561 			} else {
3562 			    NFSUNLOCKSTATE();
3563 			    error = NFSERR_LOCKSHELD;
3564 			    goto out;
3565 			}
3566 		    }
3567 		    stp = nstp;
3568 		}
3569 	    }
3570 	}
3571 	NFSUNLOCKSTATE();
3572 
3573 out:
3574 	NFSEXITCODE(error);
3575 	return (error);
3576 }
3577 
3578 /*
3579  * Get the file handle for a lock structure.
3580  */
3581 static int
3582 nfsrv_getlockfh(vnode_t vp, u_short flags, struct nfslockfile *new_lfp,
3583     fhandle_t *nfhp, NFSPROC_T *p)
3584 {
3585 	fhandle_t *fhp = NULL;
3586 	int error;
3587 
3588 	/*
3589 	 * For lock, use the new nfslock structure, otherwise just
3590 	 * a fhandle_t on the stack.
3591 	 */
3592 	if (flags & NFSLCK_OPEN) {
3593 		KASSERT(new_lfp != NULL, ("nfsrv_getlockfh: new_lfp NULL"));
3594 		fhp = &new_lfp->lf_fh;
3595 	} else if (nfhp) {
3596 		fhp = nfhp;
3597 	} else {
3598 		panic("nfsrv_getlockfh");
3599 	}
3600 	error = nfsvno_getfh(vp, fhp, p);
3601 	NFSEXITCODE(error);
3602 	return (error);
3603 }
3604 
3605 /*
3606  * Get an nfs lock structure. Allocate one, as required, and return a
3607  * pointer to it.
3608  * Returns an NFSERR_xxx upon failure or -1 to indicate no current lock.
3609  */
3610 static int
3611 nfsrv_getlockfile(u_short flags, struct nfslockfile **new_lfpp,
3612     struct nfslockfile **lfpp, fhandle_t *nfhp, int lockit)
3613 {
3614 	struct nfslockfile *lfp;
3615 	fhandle_t *fhp = NULL, *tfhp;
3616 	struct nfslockhashhead *hp;
3617 	struct nfslockfile *new_lfp = NULL;
3618 
3619 	/*
3620 	 * For lock, use the new nfslock structure, otherwise just
3621 	 * a fhandle_t on the stack.
3622 	 */
3623 	if (flags & NFSLCK_OPEN) {
3624 		new_lfp = *new_lfpp;
3625 		fhp = &new_lfp->lf_fh;
3626 	} else if (nfhp) {
3627 		fhp = nfhp;
3628 	} else {
3629 		panic("nfsrv_getlockfile");
3630 	}
3631 
3632 	hp = NFSLOCKHASH(fhp);
3633 	LIST_FOREACH(lfp, hp, lf_hash) {
3634 		tfhp = &lfp->lf_fh;
3635 		if (NFSVNO_CMPFH(fhp, tfhp)) {
3636 			if (lockit)
3637 				nfsrv_locklf(lfp);
3638 			*lfpp = lfp;
3639 			return (0);
3640 		}
3641 	}
3642 	if (!(flags & NFSLCK_OPEN))
3643 		return (-1);
3644 
3645 	/*
3646 	 * No match, so chain the new one into the list.
3647 	 */
3648 	LIST_INIT(&new_lfp->lf_open);
3649 	LIST_INIT(&new_lfp->lf_lock);
3650 	LIST_INIT(&new_lfp->lf_deleg);
3651 	LIST_INIT(&new_lfp->lf_locallock);
3652 	LIST_INIT(&new_lfp->lf_rollback);
3653 	new_lfp->lf_locallock_lck.nfslock_usecnt = 0;
3654 	new_lfp->lf_locallock_lck.nfslock_lock = 0;
3655 	new_lfp->lf_usecount = 0;
3656 	LIST_INSERT_HEAD(hp, new_lfp, lf_hash);
3657 	*lfpp = new_lfp;
3658 	*new_lfpp = NULL;
3659 	return (0);
3660 }
3661 
3662 /*
3663  * This function adds a nfslock lock structure to the list for the associated
3664  * nfsstate and nfslockfile structures. It will be inserted after the
3665  * entry pointed at by insert_lop.
3666  */
3667 static void
3668 nfsrv_insertlock(struct nfslock *new_lop, struct nfslock *insert_lop,
3669     struct nfsstate *stp, struct nfslockfile *lfp)
3670 {
3671 	struct nfslock *lop, *nlop;
3672 
3673 	new_lop->lo_stp = stp;
3674 	new_lop->lo_lfp = lfp;
3675 
3676 	if (stp != NULL) {
3677 		/* Insert in increasing lo_first order */
3678 		lop = LIST_FIRST(&lfp->lf_lock);
3679 		if (lop == LIST_END(&lfp->lf_lock) ||
3680 		    new_lop->lo_first <= lop->lo_first) {
3681 			LIST_INSERT_HEAD(&lfp->lf_lock, new_lop, lo_lckfile);
3682 		} else {
3683 			nlop = LIST_NEXT(lop, lo_lckfile);
3684 			while (nlop != LIST_END(&lfp->lf_lock) &&
3685 			       nlop->lo_first < new_lop->lo_first) {
3686 				lop = nlop;
3687 				nlop = LIST_NEXT(lop, lo_lckfile);
3688 			}
3689 			LIST_INSERT_AFTER(lop, new_lop, lo_lckfile);
3690 		}
3691 	} else {
3692 		new_lop->lo_lckfile.le_prev = NULL;	/* list not used */
3693 	}
3694 
3695 	/*
3696 	 * Insert after insert_lop, which is overloaded as stp or lfp for
3697 	 * an empty list.
3698 	 */
3699 	if (stp == NULL && (struct nfslockfile *)insert_lop == lfp)
3700 		LIST_INSERT_HEAD(&lfp->lf_locallock, new_lop, lo_lckowner);
3701 	else if ((struct nfsstate *)insert_lop == stp)
3702 		LIST_INSERT_HEAD(&stp->ls_lock, new_lop, lo_lckowner);
3703 	else
3704 		LIST_INSERT_AFTER(insert_lop, new_lop, lo_lckowner);
3705 	if (stp != NULL) {
3706 		nfsstatsv1.srvlocks++;
3707 		nfsrv_openpluslock++;
3708 	}
3709 }
3710 
3711 /*
3712  * This function updates the locking for a lock owner and given file. It
3713  * maintains a list of lock ranges ordered on increasing file offset that
3714  * are NFSLCK_READ or NFSLCK_WRITE and non-overlapping (aka POSIX style).
3715  * It always adds new_lop to the list and sometimes uses the one pointed
3716  * at by other_lopp.
3717  */
3718 static void
3719 nfsrv_updatelock(struct nfsstate *stp, struct nfslock **new_lopp,
3720     struct nfslock **other_lopp, struct nfslockfile *lfp)
3721 {
3722 	struct nfslock *new_lop = *new_lopp;
3723 	struct nfslock *lop, *tlop, *ilop;
3724 	struct nfslock *other_lop = *other_lopp;
3725 	int unlock = 0, myfile = 0;
3726 	u_int64_t tmp;
3727 
3728 	/*
3729 	 * Work down the list until the lock is merged.
3730 	 */
3731 	if (new_lop->lo_flags & NFSLCK_UNLOCK)
3732 		unlock = 1;
3733 	if (stp != NULL) {
3734 		ilop = (struct nfslock *)stp;
3735 		lop = LIST_FIRST(&stp->ls_lock);
3736 	} else {
3737 		ilop = (struct nfslock *)lfp;
3738 		lop = LIST_FIRST(&lfp->lf_locallock);
3739 	}
3740 	while (lop != NULL) {
3741 	    /*
3742 	     * Only check locks for this file that aren't before the start of
3743 	     * new lock's range.
3744 	     */
3745 	    if (lop->lo_lfp == lfp) {
3746 	      myfile = 1;
3747 	      if (lop->lo_end >= new_lop->lo_first) {
3748 		if (new_lop->lo_end < lop->lo_first) {
3749 			/*
3750 			 * If the new lock ends before the start of the
3751 			 * current lock's range, no merge, just insert
3752 			 * the new lock.
3753 			 */
3754 			break;
3755 		}
3756 		if (new_lop->lo_flags == lop->lo_flags ||
3757 		    (new_lop->lo_first <= lop->lo_first &&
3758 		     new_lop->lo_end >= lop->lo_end)) {
3759 			/*
3760 			 * This lock can be absorbed by the new lock/unlock.
3761 			 * This happens when it covers the entire range
3762 			 * of the old lock or is contiguous
3763 			 * with the old lock and is of the same type or an
3764 			 * unlock.
3765 			 */
3766 			if (lop->lo_first < new_lop->lo_first)
3767 				new_lop->lo_first = lop->lo_first;
3768 			if (lop->lo_end > new_lop->lo_end)
3769 				new_lop->lo_end = lop->lo_end;
3770 			tlop = lop;
3771 			lop = LIST_NEXT(lop, lo_lckowner);
3772 			nfsrv_freenfslock(tlop);
3773 			continue;
3774 		}
3775 
3776 		/*
3777 		 * All these cases are for contiguous locks that are not the
3778 		 * same type, so they can't be merged.
3779 		 */
3780 		if (new_lop->lo_first <= lop->lo_first) {
3781 			/*
3782 			 * This case is where the new lock overlaps with the
3783 			 * first part of the old lock. Move the start of the
3784 			 * old lock to just past the end of the new lock. The
3785 			 * new lock will be inserted in front of the old, since
3786 			 * ilop hasn't been updated. (We are done now.)
3787 			 */
3788 			lop->lo_first = new_lop->lo_end;
3789 			break;
3790 		}
3791 		if (new_lop->lo_end >= lop->lo_end) {
3792 			/*
3793 			 * This case is where the new lock overlaps with the
3794 			 * end of the old lock's range. Move the old lock's
3795 			 * end to just before the new lock's first and insert
3796 			 * the new lock after the old lock.
3797 			 * Might not be done yet, since the new lock could
3798 			 * overlap further locks with higher ranges.
3799 			 */
3800 			lop->lo_end = new_lop->lo_first;
3801 			ilop = lop;
3802 			lop = LIST_NEXT(lop, lo_lckowner);
3803 			continue;
3804 		}
3805 		/*
3806 		 * The final case is where the new lock's range is in the
3807 		 * middle of the current lock's and splits the current lock
3808 		 * up. Use *other_lopp to handle the second part of the
3809 		 * split old lock range. (We are done now.)
3810 		 * For unlock, we use new_lop as other_lop and tmp, since
3811 		 * other_lop and new_lop are the same for this case.
3812 		 * We noted the unlock case above, so we don't need
3813 		 * new_lop->lo_flags any longer.
3814 		 */
3815 		tmp = new_lop->lo_first;
3816 		if (other_lop == NULL) {
3817 			if (!unlock)
3818 				panic("nfsd srv update unlock");
3819 			other_lop = new_lop;
3820 			*new_lopp = NULL;
3821 		}
3822 		other_lop->lo_first = new_lop->lo_end;
3823 		other_lop->lo_end = lop->lo_end;
3824 		other_lop->lo_flags = lop->lo_flags;
3825 		other_lop->lo_stp = stp;
3826 		other_lop->lo_lfp = lfp;
3827 		lop->lo_end = tmp;
3828 		nfsrv_insertlock(other_lop, lop, stp, lfp);
3829 		*other_lopp = NULL;
3830 		ilop = lop;
3831 		break;
3832 	      }
3833 	    }
3834 	    ilop = lop;
3835 	    lop = LIST_NEXT(lop, lo_lckowner);
3836 	    if (myfile && (lop == NULL || lop->lo_lfp != lfp))
3837 		break;
3838 	}
3839 
3840 	/*
3841 	 * Insert the new lock in the list at the appropriate place.
3842 	 */
3843 	if (!unlock) {
3844 		nfsrv_insertlock(new_lop, ilop, stp, lfp);
3845 		*new_lopp = NULL;
3846 	}
3847 }
3848 
3849 /*
3850  * This function handles sequencing of locks, etc.
3851  * It returns an error that indicates what the caller should do.
3852  */
3853 static int
3854 nfsrv_checkseqid(struct nfsrv_descript *nd, u_int32_t seqid,
3855     struct nfsstate *stp, struct nfsrvcache *op)
3856 {
3857 	int error = 0;
3858 
3859 	if ((nd->nd_flag & ND_NFSV41) != 0)
3860 		/* NFSv4.1 ignores the open_seqid and lock_seqid. */
3861 		goto out;
3862 	if (op != nd->nd_rp)
3863 		panic("nfsrvstate checkseqid");
3864 	if (!(op->rc_flag & RC_INPROG))
3865 		panic("nfsrvstate not inprog");
3866 	if (stp->ls_op && stp->ls_op->rc_refcnt <= 0) {
3867 		printf("refcnt=%d\n", stp->ls_op->rc_refcnt);
3868 		panic("nfsrvstate op refcnt");
3869 	}
3870 	if ((stp->ls_seq + 1) == seqid) {
3871 		if (stp->ls_op)
3872 			nfsrvd_derefcache(stp->ls_op);
3873 		stp->ls_op = op;
3874 		nfsrvd_refcache(op);
3875 		stp->ls_seq = seqid;
3876 		goto out;
3877 	} else if (stp->ls_seq == seqid && stp->ls_op &&
3878 		op->rc_xid == stp->ls_op->rc_xid &&
3879 		op->rc_refcnt == 0 &&
3880 		op->rc_reqlen == stp->ls_op->rc_reqlen &&
3881 		op->rc_cksum == stp->ls_op->rc_cksum) {
3882 		if (stp->ls_op->rc_flag & RC_INPROG) {
3883 			error = NFSERR_DONTREPLY;
3884 			goto out;
3885 		}
3886 		nd->nd_rp = stp->ls_op;
3887 		nd->nd_rp->rc_flag |= RC_INPROG;
3888 		nfsrvd_delcache(op);
3889 		error = NFSERR_REPLYFROMCACHE;
3890 		goto out;
3891 	}
3892 	error = NFSERR_BADSEQID;
3893 
3894 out:
3895 	NFSEXITCODE2(error, nd);
3896 	return (error);
3897 }
3898 
3899 /*
3900  * Get the client ip address for callbacks. If the strings can't be parsed,
3901  * just set lc_program to 0 to indicate no callbacks are possible.
3902  * (For cases where the address can't be parsed or is 0.0.0.0.0.0, set
3903  *  the address to the client's transport address. This won't be used
3904  *  for callbacks, but can be printed out by nfsstats for info.)
3905  * Return error if the xdr can't be parsed, 0 otherwise.
3906  */
3907 APPLESTATIC int
3908 nfsrv_getclientipaddr(struct nfsrv_descript *nd, struct nfsclient *clp)
3909 {
3910 	u_int32_t *tl;
3911 	u_char *cp, *cp2;
3912 	int i, j;
3913 	struct sockaddr_in *rad, *sad;
3914 	u_char protocol[5], addr[24];
3915 	int error = 0, cantparse = 0;
3916 	union {
3917 		in_addr_t ival;
3918 		u_char cval[4];
3919 	} ip;
3920 	union {
3921 		in_port_t sval;
3922 		u_char cval[2];
3923 	} port;
3924 
3925 	rad = NFSSOCKADDR(clp->lc_req.nr_nam, struct sockaddr_in *);
3926 	rad->sin_family = AF_INET;
3927 	rad->sin_len = sizeof (struct sockaddr_in);
3928 	rad->sin_addr.s_addr = 0;
3929 	rad->sin_port = 0;
3930 	clp->lc_req.nr_client = NULL;
3931 	clp->lc_req.nr_lock = 0;
3932 	NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
3933 	i = fxdr_unsigned(int, *tl);
3934 	if (i >= 3 && i <= 4) {
3935 		error = nfsrv_mtostr(nd, protocol, i);
3936 		if (error)
3937 			goto nfsmout;
3938 		if (!strcmp(protocol, "tcp")) {
3939 			clp->lc_flags |= LCL_TCPCALLBACK;
3940 			clp->lc_req.nr_sotype = SOCK_STREAM;
3941 			clp->lc_req.nr_soproto = IPPROTO_TCP;
3942 		} else if (!strcmp(protocol, "udp")) {
3943 			clp->lc_req.nr_sotype = SOCK_DGRAM;
3944 			clp->lc_req.nr_soproto = IPPROTO_UDP;
3945 		} else {
3946 			cantparse = 1;
3947 		}
3948 	} else {
3949 		cantparse = 1;
3950 		if (i > 0) {
3951 			error = nfsm_advance(nd, NFSM_RNDUP(i), -1);
3952 			if (error)
3953 				goto nfsmout;
3954 		}
3955 	}
3956 	NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
3957 	i = fxdr_unsigned(int, *tl);
3958 	if (i < 0) {
3959 		error = NFSERR_BADXDR;
3960 		goto nfsmout;
3961 	} else if (i == 0) {
3962 		cantparse = 1;
3963 	} else if (!cantparse && i <= 23 && i >= 11) {
3964 		error = nfsrv_mtostr(nd, addr, i);
3965 		if (error)
3966 			goto nfsmout;
3967 
3968 		/*
3969 		 * Parse out the address fields. We expect 6 decimal numbers
3970 		 * separated by '.'s.
3971 		 */
3972 		cp = addr;
3973 		i = 0;
3974 		while (*cp && i < 6) {
3975 			cp2 = cp;
3976 			while (*cp2 && *cp2 != '.')
3977 				cp2++;
3978 			if (*cp2)
3979 				*cp2++ = '\0';
3980 			else if (i != 5) {
3981 				cantparse = 1;
3982 				break;
3983 			}
3984 			j = nfsrv_getipnumber(cp);
3985 			if (j >= 0) {
3986 				if (i < 4)
3987 					ip.cval[3 - i] = j;
3988 				else
3989 					port.cval[5 - i] = j;
3990 			} else {
3991 				cantparse = 1;
3992 				break;
3993 			}
3994 			cp = cp2;
3995 			i++;
3996 		}
3997 		if (!cantparse) {
3998 			if (ip.ival != 0x0) {
3999 				rad->sin_addr.s_addr = htonl(ip.ival);
4000 				rad->sin_port = htons(port.sval);
4001 			} else {
4002 				cantparse = 1;
4003 			}
4004 		}
4005 	} else {
4006 		cantparse = 1;
4007 		if (i > 0) {
4008 			error = nfsm_advance(nd, NFSM_RNDUP(i), -1);
4009 			if (error)
4010 				goto nfsmout;
4011 		}
4012 	}
4013 	if (cantparse) {
4014 		sad = NFSSOCKADDR(nd->nd_nam, struct sockaddr_in *);
4015 		if (sad->sin_family == AF_INET) {
4016 			rad->sin_addr.s_addr = sad->sin_addr.s_addr;
4017 			rad->sin_port = 0x0;
4018 		}
4019 		clp->lc_program = 0;
4020 	}
4021 nfsmout:
4022 	NFSEXITCODE2(error, nd);
4023 	return (error);
4024 }
4025 
4026 /*
4027  * Turn a string of up to three decimal digits into a number. Return -1 upon
4028  * error.
4029  */
4030 static int
4031 nfsrv_getipnumber(u_char *cp)
4032 {
4033 	int i = 0, j = 0;
4034 
4035 	while (*cp) {
4036 		if (j > 2 || *cp < '0' || *cp > '9')
4037 			return (-1);
4038 		i *= 10;
4039 		i += (*cp - '0');
4040 		cp++;
4041 		j++;
4042 	}
4043 	if (i < 256)
4044 		return (i);
4045 	return (-1);
4046 }
4047 
4048 /*
4049  * This function checks for restart conditions.
4050  */
4051 static int
4052 nfsrv_checkrestart(nfsquad_t clientid, u_int32_t flags,
4053     nfsv4stateid_t *stateidp, int specialid)
4054 {
4055 	int ret = 0;
4056 
4057 	/*
4058 	 * First check for a server restart. Open, LockT, ReleaseLockOwner
4059 	 * and DelegPurge have a clientid, the rest a stateid.
4060 	 */
4061 	if (flags &
4062 	    (NFSLCK_OPEN | NFSLCK_TEST | NFSLCK_RELEASE | NFSLCK_DELEGPURGE)) {
4063 		if (clientid.lval[0] != nfsrvboottime) {
4064 			ret = NFSERR_STALECLIENTID;
4065 			goto out;
4066 		}
4067 	} else if (stateidp->other[0] != nfsrvboottime &&
4068 		specialid == 0) {
4069 		ret = NFSERR_STALESTATEID;
4070 		goto out;
4071 	}
4072 
4073 	/*
4074 	 * Read, Write, Setattr and LockT can return NFSERR_GRACE and do
4075 	 * not use a lock/open owner seqid#, so the check can be done now.
4076 	 * (The others will be checked, as required, later.)
4077 	 */
4078 	if (!(flags & (NFSLCK_CHECK | NFSLCK_TEST)))
4079 		goto out;
4080 
4081 	NFSLOCKSTATE();
4082 	ret = nfsrv_checkgrace(NULL, NULL, flags);
4083 	NFSUNLOCKSTATE();
4084 
4085 out:
4086 	NFSEXITCODE(ret);
4087 	return (ret);
4088 }
4089 
4090 /*
4091  * Check for grace.
4092  */
4093 static int
4094 nfsrv_checkgrace(struct nfsrv_descript *nd, struct nfsclient *clp,
4095     u_int32_t flags)
4096 {
4097 	int error = 0, notreclaimed;
4098 	struct nfsrv_stable *sp;
4099 
4100 	if ((nfsrv_stablefirst.nsf_flags & (NFSNSF_UPDATEDONE |
4101 	     NFSNSF_GRACEOVER)) == 0) {
4102 		/*
4103 		 * First, check to see if all of the clients have done a
4104 		 * ReclaimComplete.  If so, grace can end now.
4105 		 */
4106 		notreclaimed = 0;
4107 		LIST_FOREACH(sp, &nfsrv_stablefirst.nsf_head, nst_list) {
4108 			if ((sp->nst_flag & NFSNST_RECLAIMED) == 0) {
4109 				notreclaimed = 1;
4110 				break;
4111 			}
4112 		}
4113 		if (notreclaimed == 0)
4114 			nfsrv_stablefirst.nsf_flags |= (NFSNSF_GRACEOVER |
4115 			    NFSNSF_NEEDLOCK);
4116 	}
4117 
4118 	if ((nfsrv_stablefirst.nsf_flags & NFSNSF_GRACEOVER) != 0) {
4119 		if (flags & NFSLCK_RECLAIM) {
4120 			error = NFSERR_NOGRACE;
4121 			goto out;
4122 		}
4123 	} else {
4124 		if (!(flags & NFSLCK_RECLAIM)) {
4125 			error = NFSERR_GRACE;
4126 			goto out;
4127 		}
4128 		if (nd != NULL && clp != NULL &&
4129 		    (nd->nd_flag & ND_NFSV41) != 0 &&
4130 		    (clp->lc_flags & LCL_RECLAIMCOMPLETE) != 0) {
4131 			error = NFSERR_NOGRACE;
4132 			goto out;
4133 		}
4134 
4135 		/*
4136 		 * If grace is almost over and we are still getting Reclaims,
4137 		 * extend grace a bit.
4138 		 */
4139 		if ((NFSD_MONOSEC + NFSRV_LEASEDELTA) >
4140 		    nfsrv_stablefirst.nsf_eograce)
4141 			nfsrv_stablefirst.nsf_eograce = NFSD_MONOSEC +
4142 				NFSRV_LEASEDELTA;
4143 	}
4144 
4145 out:
4146 	NFSEXITCODE(error);
4147 	return (error);
4148 }
4149 
4150 /*
4151  * Do a server callback.
4152  */
4153 static int
4154 nfsrv_docallback(struct nfsclient *clp, int procnum,
4155     nfsv4stateid_t *stateidp, int trunc, fhandle_t *fhp,
4156     struct nfsvattr *nap, nfsattrbit_t *attrbitp, NFSPROC_T *p)
4157 {
4158 	mbuf_t m;
4159 	u_int32_t *tl;
4160 	struct nfsrv_descript nfsd, *nd = &nfsd;
4161 	struct ucred *cred;
4162 	int error = 0;
4163 	u_int32_t callback;
4164 	struct nfsdsession *sep = NULL;
4165 
4166 	cred = newnfs_getcred();
4167 	NFSLOCKSTATE();	/* mostly for lc_cbref++ */
4168 	if (clp->lc_flags & LCL_NEEDSCONFIRM) {
4169 		NFSUNLOCKSTATE();
4170 		panic("docallb");
4171 	}
4172 	clp->lc_cbref++;
4173 
4174 	/*
4175 	 * Fill the callback program# and version into the request
4176 	 * structure for newnfs_connect() to use.
4177 	 */
4178 	clp->lc_req.nr_prog = clp->lc_program;
4179 #ifdef notnow
4180 	if ((clp->lc_flags & LCL_NFSV41) != 0)
4181 		clp->lc_req.nr_vers = NFSV41_CBVERS;
4182 	else
4183 #endif
4184 		clp->lc_req.nr_vers = NFSV4_CBVERS;
4185 
4186 	/*
4187 	 * First, fill in some of the fields of nd and cr.
4188 	 */
4189 	nd->nd_flag = ND_NFSV4;
4190 	if (clp->lc_flags & LCL_GSS)
4191 		nd->nd_flag |= ND_KERBV;
4192 	if ((clp->lc_flags & LCL_NFSV41) != 0)
4193 		nd->nd_flag |= ND_NFSV41;
4194 	nd->nd_repstat = 0;
4195 	cred->cr_uid = clp->lc_uid;
4196 	cred->cr_gid = clp->lc_gid;
4197 	callback = clp->lc_callback;
4198 	NFSUNLOCKSTATE();
4199 	cred->cr_ngroups = 1;
4200 
4201 	/*
4202 	 * Get the first mbuf for the request.
4203 	 */
4204 	MGET(m, M_WAITOK, MT_DATA);
4205 	mbuf_setlen(m, 0);
4206 	nd->nd_mreq = nd->nd_mb = m;
4207 	nd->nd_bpos = NFSMTOD(m, caddr_t);
4208 
4209 	/*
4210 	 * and build the callback request.
4211 	 */
4212 	if (procnum == NFSV4OP_CBGETATTR) {
4213 		nd->nd_procnum = NFSV4PROC_CBCOMPOUND;
4214 		error = nfsrv_cbcallargs(nd, clp, callback, NFSV4OP_CBGETATTR,
4215 		    "CB Getattr", &sep);
4216 		if (error != 0) {
4217 			mbuf_freem(nd->nd_mreq);
4218 			goto errout;
4219 		}
4220 		(void)nfsm_fhtom(nd, (u_int8_t *)fhp, NFSX_MYFH, 0);
4221 		(void)nfsrv_putattrbit(nd, attrbitp);
4222 	} else if (procnum == NFSV4OP_CBRECALL) {
4223 		nd->nd_procnum = NFSV4PROC_CBCOMPOUND;
4224 		error = nfsrv_cbcallargs(nd, clp, callback, NFSV4OP_CBRECALL,
4225 		    "CB Recall", &sep);
4226 		if (error != 0) {
4227 			mbuf_freem(nd->nd_mreq);
4228 			goto errout;
4229 		}
4230 		NFSM_BUILD(tl, u_int32_t *, NFSX_UNSIGNED + NFSX_STATEID);
4231 		*tl++ = txdr_unsigned(stateidp->seqid);
4232 		NFSBCOPY((caddr_t)stateidp->other, (caddr_t)tl,
4233 		    NFSX_STATEIDOTHER);
4234 		tl += (NFSX_STATEIDOTHER / NFSX_UNSIGNED);
4235 		if (trunc)
4236 			*tl = newnfs_true;
4237 		else
4238 			*tl = newnfs_false;
4239 		(void)nfsm_fhtom(nd, (u_int8_t *)fhp, NFSX_MYFH, 0);
4240 	} else if (procnum == NFSV4PROC_CBNULL) {
4241 		nd->nd_procnum = NFSV4PROC_CBNULL;
4242 		if ((clp->lc_flags & LCL_NFSV41) != 0) {
4243 			error = nfsv4_getcbsession(clp, &sep);
4244 			if (error != 0) {
4245 				mbuf_freem(nd->nd_mreq);
4246 				goto errout;
4247 			}
4248 		}
4249 	} else {
4250 		error = NFSERR_SERVERFAULT;
4251 		mbuf_freem(nd->nd_mreq);
4252 		goto errout;
4253 	}
4254 
4255 	/*
4256 	 * Call newnfs_connect(), as required, and then newnfs_request().
4257 	 */
4258 	(void) newnfs_sndlock(&clp->lc_req.nr_lock);
4259 	if (clp->lc_req.nr_client == NULL) {
4260 		if ((clp->lc_flags & LCL_NFSV41) != 0) {
4261 			error = ECONNREFUSED;
4262 			nfsrv_freesession(sep, NULL);
4263 		} else if (nd->nd_procnum == NFSV4PROC_CBNULL)
4264 			error = newnfs_connect(NULL, &clp->lc_req, cred,
4265 			    NULL, 1);
4266 		else
4267 			error = newnfs_connect(NULL, &clp->lc_req, cred,
4268 			    NULL, 3);
4269 	}
4270 	newnfs_sndunlock(&clp->lc_req.nr_lock);
4271 	if (!error) {
4272 		if ((nd->nd_flag & ND_NFSV41) != 0) {
4273 			KASSERT(sep != NULL, ("sep NULL"));
4274 			if (sep->sess_cbsess.nfsess_xprt != NULL)
4275 				error = newnfs_request(nd, NULL, clp,
4276 				    &clp->lc_req, NULL, NULL, cred,
4277 				    clp->lc_program, clp->lc_req.nr_vers, NULL,
4278 				    1, NULL, &sep->sess_cbsess);
4279 			else {
4280 				/*
4281 				 * This should probably never occur, but if a
4282 				 * client somehow does an RPC without a
4283 				 * SequenceID Op that causes a callback just
4284 				 * after the nfsd threads have been terminated
4285 				 * and restared we could conceivably get here
4286 				 * without a backchannel xprt.
4287 				 */
4288 				printf("nfsrv_docallback: no xprt\n");
4289 				error = ECONNREFUSED;
4290 			}
4291 			nfsrv_freesession(sep, NULL);
4292 		} else
4293 			error = newnfs_request(nd, NULL, clp, &clp->lc_req,
4294 			    NULL, NULL, cred, clp->lc_program,
4295 			    clp->lc_req.nr_vers, NULL, 1, NULL, NULL);
4296 	}
4297 errout:
4298 	NFSFREECRED(cred);
4299 
4300 	/*
4301 	 * If error is set here, the Callback path isn't working
4302 	 * properly, so twiddle the appropriate LCL_ flags.
4303 	 * (nd_repstat != 0 indicates the Callback path is working,
4304 	 *  but the callback failed on the client.)
4305 	 */
4306 	if (error) {
4307 		/*
4308 		 * Mark the callback pathway down, which disabled issuing
4309 		 * of delegations and gets Renew to return NFSERR_CBPATHDOWN.
4310 		 */
4311 		NFSLOCKSTATE();
4312 		clp->lc_flags |= LCL_CBDOWN;
4313 		NFSUNLOCKSTATE();
4314 	} else {
4315 		/*
4316 		 * Callback worked. If the callback path was down, disable
4317 		 * callbacks, so no more delegations will be issued. (This
4318 		 * is done on the assumption that the callback pathway is
4319 		 * flakey.)
4320 		 */
4321 		NFSLOCKSTATE();
4322 		if (clp->lc_flags & LCL_CBDOWN)
4323 			clp->lc_flags &= ~(LCL_CBDOWN | LCL_CALLBACKSON);
4324 		NFSUNLOCKSTATE();
4325 		if (nd->nd_repstat)
4326 			error = nd->nd_repstat;
4327 		else if (error == 0 && procnum == NFSV4OP_CBGETATTR)
4328 			error = nfsv4_loadattr(nd, NULL, nap, NULL, NULL, 0,
4329 			    NULL, NULL, NULL, NULL, NULL, 0, NULL, NULL, NULL,
4330 			    p, NULL);
4331 		mbuf_freem(nd->nd_mrep);
4332 	}
4333 	NFSLOCKSTATE();
4334 	clp->lc_cbref--;
4335 	if ((clp->lc_flags & LCL_WAKEUPWANTED) && clp->lc_cbref == 0) {
4336 		clp->lc_flags &= ~LCL_WAKEUPWANTED;
4337 		wakeup(clp);
4338 	}
4339 	NFSUNLOCKSTATE();
4340 
4341 	NFSEXITCODE(error);
4342 	return (error);
4343 }
4344 
4345 /*
4346  * Set up the compound RPC for the callback.
4347  */
4348 static int
4349 nfsrv_cbcallargs(struct nfsrv_descript *nd, struct nfsclient *clp,
4350     uint32_t callback, int op, const char *optag, struct nfsdsession **sepp)
4351 {
4352 	uint32_t *tl;
4353 	int error, len;
4354 
4355 	len = strlen(optag);
4356 	(void)nfsm_strtom(nd, optag, len);
4357 	NFSM_BUILD(tl, uint32_t *, 4 * NFSX_UNSIGNED);
4358 	if ((nd->nd_flag & ND_NFSV41) != 0) {
4359 		*tl++ = txdr_unsigned(NFSV41_MINORVERSION);
4360 		*tl++ = txdr_unsigned(callback);
4361 		*tl++ = txdr_unsigned(2);
4362 		*tl = txdr_unsigned(NFSV4OP_CBSEQUENCE);
4363 		error = nfsv4_setcbsequence(nd, clp, 1, sepp);
4364 		if (error != 0)
4365 			return (error);
4366 		NFSM_BUILD(tl, u_int32_t *, NFSX_UNSIGNED);
4367 		*tl = txdr_unsigned(op);
4368 	} else {
4369 		*tl++ = txdr_unsigned(NFSV4_MINORVERSION);
4370 		*tl++ = txdr_unsigned(callback);
4371 		*tl++ = txdr_unsigned(1);
4372 		*tl = txdr_unsigned(op);
4373 	}
4374 	return (0);
4375 }
4376 
4377 /*
4378  * Return the next index# for a clientid. Mostly just increment and return
4379  * the next one, but... if the 32bit unsigned does actually wrap around,
4380  * it should be rebooted.
4381  * At an average rate of one new client per second, it will wrap around in
4382  * approximately 136 years. (I think the server will have been shut
4383  * down or rebooted before then.)
4384  */
4385 static u_int32_t
4386 nfsrv_nextclientindex(void)
4387 {
4388 	static u_int32_t client_index = 0;
4389 
4390 	client_index++;
4391 	if (client_index != 0)
4392 		return (client_index);
4393 
4394 	printf("%s: out of clientids\n", __func__);
4395 	return (client_index);
4396 }
4397 
4398 /*
4399  * Return the next index# for a stateid. Mostly just increment and return
4400  * the next one, but... if the 32bit unsigned does actually wrap around
4401  * (will a BSD server stay up that long?), find
4402  * new start and end values.
4403  */
4404 static u_int32_t
4405 nfsrv_nextstateindex(struct nfsclient *clp)
4406 {
4407 	struct nfsstate *stp;
4408 	int i;
4409 	u_int32_t canuse, min_index, max_index;
4410 
4411 	if (!(clp->lc_flags & LCL_INDEXNOTOK)) {
4412 		clp->lc_stateindex++;
4413 		if (clp->lc_stateindex != clp->lc_statemaxindex)
4414 			return (clp->lc_stateindex);
4415 	}
4416 
4417 	/*
4418 	 * Yuck, we've hit the end.
4419 	 * Look for a new min and max.
4420 	 */
4421 	min_index = 0;
4422 	max_index = 0xffffffff;
4423 	for (i = 0; i < nfsrv_statehashsize; i++) {
4424 	    LIST_FOREACH(stp, &clp->lc_stateid[i], ls_hash) {
4425 		if (stp->ls_stateid.other[2] > 0x80000000) {
4426 		    if (stp->ls_stateid.other[2] < max_index)
4427 			max_index = stp->ls_stateid.other[2];
4428 		} else {
4429 		    if (stp->ls_stateid.other[2] > min_index)
4430 			min_index = stp->ls_stateid.other[2];
4431 		}
4432 	    }
4433 	}
4434 
4435 	/*
4436 	 * Yikes, highly unlikely, but I'll handle it anyhow.
4437 	 */
4438 	if (min_index == 0x80000000 && max_index == 0x80000001) {
4439 	    canuse = 0;
4440 	    /*
4441 	     * Loop around until we find an unused entry. Return that
4442 	     * and set LCL_INDEXNOTOK, so the search will continue next time.
4443 	     * (This is one of those rare cases where a goto is the
4444 	     *  cleanest way to code the loop.)
4445 	     */
4446 tryagain:
4447 	    for (i = 0; i < nfsrv_statehashsize; i++) {
4448 		LIST_FOREACH(stp, &clp->lc_stateid[i], ls_hash) {
4449 		    if (stp->ls_stateid.other[2] == canuse) {
4450 			canuse++;
4451 			goto tryagain;
4452 		    }
4453 		}
4454 	    }
4455 	    clp->lc_flags |= LCL_INDEXNOTOK;
4456 	    return (canuse);
4457 	}
4458 
4459 	/*
4460 	 * Ok to start again from min + 1.
4461 	 */
4462 	clp->lc_stateindex = min_index + 1;
4463 	clp->lc_statemaxindex = max_index;
4464 	clp->lc_flags &= ~LCL_INDEXNOTOK;
4465 	return (clp->lc_stateindex);
4466 }
4467 
4468 /*
4469  * The following functions handle the stable storage file that deals with
4470  * the edge conditions described in RFC3530 Sec. 8.6.3.
4471  * The file is as follows:
4472  * - a single record at the beginning that has the lease time of the
4473  *   previous server instance (before the last reboot) and the nfsrvboottime
4474  *   values for the previous server boots.
4475  *   These previous boot times are used to ensure that the current
4476  *   nfsrvboottime does not, somehow, get set to a previous one.
4477  *   (This is important so that Stale ClientIDs and StateIDs can
4478  *    be recognized.)
4479  *   The number of previous nfsvrboottime values precedes the list.
4480  * - followed by some number of appended records with:
4481  *   - client id string
4482  *   - flag that indicates it is a record revoking state via lease
4483  *     expiration or similar
4484  *     OR has successfully acquired state.
4485  * These structures vary in length, with the client string at the end, up
4486  * to NFSV4_OPAQUELIMIT in size.
4487  *
4488  * At the end of the grace period, the file is truncated, the first
4489  * record is rewritten with updated information and any acquired state
4490  * records for successful reclaims of state are written.
4491  *
4492  * Subsequent records are appended when the first state is issued to
4493  * a client and when state is revoked for a client.
4494  *
4495  * When reading the file in, state issued records that come later in
4496  * the file override older ones, since the append log is in cronological order.
4497  * If, for some reason, the file can't be read, the grace period is
4498  * immediately terminated and all reclaims get NFSERR_NOGRACE.
4499  */
4500 
4501 /*
4502  * Read in the stable storage file. Called by nfssvc() before the nfsd
4503  * processes start servicing requests.
4504  */
4505 APPLESTATIC void
4506 nfsrv_setupstable(NFSPROC_T *p)
4507 {
4508 	struct nfsrv_stablefirst *sf = &nfsrv_stablefirst;
4509 	struct nfsrv_stable *sp, *nsp;
4510 	struct nfst_rec *tsp;
4511 	int error, i, tryagain;
4512 	off_t off = 0;
4513 	ssize_t aresid, len;
4514 
4515 	/*
4516 	 * If NFSNSF_UPDATEDONE is set, this is a restart of the nfsds without
4517 	 * a reboot, so state has not been lost.
4518 	 */
4519 	if (sf->nsf_flags & NFSNSF_UPDATEDONE)
4520 		return;
4521 	/*
4522 	 * Set Grace over just until the file reads successfully.
4523 	 */
4524 	nfsrvboottime = time_second;
4525 	LIST_INIT(&sf->nsf_head);
4526 	sf->nsf_flags = (NFSNSF_GRACEOVER | NFSNSF_NEEDLOCK);
4527 	sf->nsf_eograce = NFSD_MONOSEC + NFSRV_LEASEDELTA;
4528 	if (sf->nsf_fp == NULL)
4529 		return;
4530 	error = NFSD_RDWR(UIO_READ, NFSFPVNODE(sf->nsf_fp),
4531 	    (caddr_t)&sf->nsf_rec, sizeof (struct nfsf_rec), off, UIO_SYSSPACE,
4532 	    0, NFSFPCRED(sf->nsf_fp), &aresid, p);
4533 	if (error || aresid || sf->nsf_numboots == 0 ||
4534 		sf->nsf_numboots > NFSNSF_MAXNUMBOOTS)
4535 		return;
4536 
4537 	/*
4538 	 * Now, read in the boottimes.
4539 	 */
4540 	sf->nsf_bootvals = (time_t *)malloc((sf->nsf_numboots + 1) *
4541 		sizeof (time_t), M_TEMP, M_WAITOK);
4542 	off = sizeof (struct nfsf_rec);
4543 	error = NFSD_RDWR(UIO_READ, NFSFPVNODE(sf->nsf_fp),
4544 	    (caddr_t)sf->nsf_bootvals, sf->nsf_numboots * sizeof (time_t), off,
4545 	    UIO_SYSSPACE, 0, NFSFPCRED(sf->nsf_fp), &aresid, p);
4546 	if (error || aresid) {
4547 		free(sf->nsf_bootvals, M_TEMP);
4548 		sf->nsf_bootvals = NULL;
4549 		return;
4550 	}
4551 
4552 	/*
4553 	 * Make sure this nfsrvboottime is different from all recorded
4554 	 * previous ones.
4555 	 */
4556 	do {
4557 		tryagain = 0;
4558 		for (i = 0; i < sf->nsf_numboots; i++) {
4559 			if (nfsrvboottime == sf->nsf_bootvals[i]) {
4560 				nfsrvboottime++;
4561 				tryagain = 1;
4562 				break;
4563 			}
4564 		}
4565 	} while (tryagain);
4566 
4567 	sf->nsf_flags |= NFSNSF_OK;
4568 	off += (sf->nsf_numboots * sizeof (time_t));
4569 
4570 	/*
4571 	 * Read through the file, building a list of records for grace
4572 	 * checking.
4573 	 * Each record is between sizeof (struct nfst_rec) and
4574 	 * sizeof (struct nfst_rec) + NFSV4_OPAQUELIMIT - 1
4575 	 * and is actually sizeof (struct nfst_rec) + nst_len - 1.
4576 	 */
4577 	tsp = (struct nfst_rec *)malloc(sizeof (struct nfst_rec) +
4578 		NFSV4_OPAQUELIMIT - 1, M_TEMP, M_WAITOK);
4579 	do {
4580 	    error = NFSD_RDWR(UIO_READ, NFSFPVNODE(sf->nsf_fp),
4581 	        (caddr_t)tsp, sizeof (struct nfst_rec) + NFSV4_OPAQUELIMIT - 1,
4582 	        off, UIO_SYSSPACE, 0, NFSFPCRED(sf->nsf_fp), &aresid, p);
4583 	    len = (sizeof (struct nfst_rec) + NFSV4_OPAQUELIMIT - 1) - aresid;
4584 	    if (error || (len > 0 && (len < sizeof (struct nfst_rec) ||
4585 		len < (sizeof (struct nfst_rec) + tsp->len - 1)))) {
4586 		/*
4587 		 * Yuck, the file has been corrupted, so just return
4588 		 * after clearing out any restart state, so the grace period
4589 		 * is over.
4590 		 */
4591 		LIST_FOREACH_SAFE(sp, &sf->nsf_head, nst_list, nsp) {
4592 			LIST_REMOVE(sp, nst_list);
4593 			free(sp, M_TEMP);
4594 		}
4595 		free(tsp, M_TEMP);
4596 		sf->nsf_flags &= ~NFSNSF_OK;
4597 		free(sf->nsf_bootvals, M_TEMP);
4598 		sf->nsf_bootvals = NULL;
4599 		return;
4600 	    }
4601 	    if (len > 0) {
4602 		off += sizeof (struct nfst_rec) + tsp->len - 1;
4603 		/*
4604 		 * Search the list for a matching client.
4605 		 */
4606 		LIST_FOREACH(sp, &sf->nsf_head, nst_list) {
4607 			if (tsp->len == sp->nst_len &&
4608 			    !NFSBCMP(tsp->client, sp->nst_client, tsp->len))
4609 				break;
4610 		}
4611 		if (sp == LIST_END(&sf->nsf_head)) {
4612 			sp = (struct nfsrv_stable *)malloc(tsp->len +
4613 				sizeof (struct nfsrv_stable) - 1, M_TEMP,
4614 				M_WAITOK);
4615 			NFSBCOPY((caddr_t)tsp, (caddr_t)&sp->nst_rec,
4616 				sizeof (struct nfst_rec) + tsp->len - 1);
4617 			LIST_INSERT_HEAD(&sf->nsf_head, sp, nst_list);
4618 		} else {
4619 			if (tsp->flag == NFSNST_REVOKE)
4620 				sp->nst_flag |= NFSNST_REVOKE;
4621 			else
4622 				/*
4623 				 * A subsequent timestamp indicates the client
4624 				 * did a setclientid/confirm and any previous
4625 				 * revoke is no longer relevant.
4626 				 */
4627 				sp->nst_flag &= ~NFSNST_REVOKE;
4628 		}
4629 	    }
4630 	} while (len > 0);
4631 	free(tsp, M_TEMP);
4632 	sf->nsf_flags = NFSNSF_OK;
4633 	sf->nsf_eograce = NFSD_MONOSEC + sf->nsf_lease +
4634 		NFSRV_LEASEDELTA;
4635 }
4636 
4637 /*
4638  * Update the stable storage file, now that the grace period is over.
4639  */
4640 APPLESTATIC void
4641 nfsrv_updatestable(NFSPROC_T *p)
4642 {
4643 	struct nfsrv_stablefirst *sf = &nfsrv_stablefirst;
4644 	struct nfsrv_stable *sp, *nsp;
4645 	int i;
4646 	struct nfsvattr nva;
4647 	vnode_t vp;
4648 #if defined(__FreeBSD_version) && (__FreeBSD_version >= 500000)
4649 	mount_t mp = NULL;
4650 #endif
4651 	int error;
4652 
4653 	if (sf->nsf_fp == NULL || (sf->nsf_flags & NFSNSF_UPDATEDONE))
4654 		return;
4655 	sf->nsf_flags |= NFSNSF_UPDATEDONE;
4656 	/*
4657 	 * Ok, we need to rewrite the stable storage file.
4658 	 * - truncate to 0 length
4659 	 * - write the new first structure
4660 	 * - loop through the data structures, writing out any that
4661 	 *   have timestamps older than the old boot
4662 	 */
4663 	if (sf->nsf_bootvals) {
4664 		sf->nsf_numboots++;
4665 		for (i = sf->nsf_numboots - 2; i >= 0; i--)
4666 			sf->nsf_bootvals[i + 1] = sf->nsf_bootvals[i];
4667 	} else {
4668 		sf->nsf_numboots = 1;
4669 		sf->nsf_bootvals = (time_t *)malloc(sizeof (time_t),
4670 			M_TEMP, M_WAITOK);
4671 	}
4672 	sf->nsf_bootvals[0] = nfsrvboottime;
4673 	sf->nsf_lease = nfsrv_lease;
4674 	NFSVNO_ATTRINIT(&nva);
4675 	NFSVNO_SETATTRVAL(&nva, size, 0);
4676 	vp = NFSFPVNODE(sf->nsf_fp);
4677 	vn_start_write(vp, &mp, V_WAIT);
4678 	if (NFSVOPLOCK(vp, LK_EXCLUSIVE) == 0) {
4679 		error = nfsvno_setattr(vp, &nva, NFSFPCRED(sf->nsf_fp), p,
4680 		    NULL);
4681 		NFSVOPUNLOCK(vp, 0);
4682 	} else
4683 		error = EPERM;
4684 	vn_finished_write(mp);
4685 	if (!error)
4686 	    error = NFSD_RDWR(UIO_WRITE, vp,
4687 		(caddr_t)&sf->nsf_rec, sizeof (struct nfsf_rec), (off_t)0,
4688 		UIO_SYSSPACE, IO_SYNC, NFSFPCRED(sf->nsf_fp), NULL, p);
4689 	if (!error)
4690 	    error = NFSD_RDWR(UIO_WRITE, vp,
4691 		(caddr_t)sf->nsf_bootvals,
4692 		sf->nsf_numboots * sizeof (time_t),
4693 		(off_t)(sizeof (struct nfsf_rec)),
4694 		UIO_SYSSPACE, IO_SYNC, NFSFPCRED(sf->nsf_fp), NULL, p);
4695 	free(sf->nsf_bootvals, M_TEMP);
4696 	sf->nsf_bootvals = NULL;
4697 	if (error) {
4698 		sf->nsf_flags &= ~NFSNSF_OK;
4699 		printf("EEK! Can't write NfsV4 stable storage file\n");
4700 		return;
4701 	}
4702 	sf->nsf_flags |= NFSNSF_OK;
4703 
4704 	/*
4705 	 * Loop through the list and write out timestamp records for
4706 	 * any clients that successfully reclaimed state.
4707 	 */
4708 	LIST_FOREACH_SAFE(sp, &sf->nsf_head, nst_list, nsp) {
4709 		if (sp->nst_flag & NFSNST_GOTSTATE) {
4710 			nfsrv_writestable(sp->nst_client, sp->nst_len,
4711 				NFSNST_NEWSTATE, p);
4712 			sp->nst_clp->lc_flags |= LCL_STAMPEDSTABLE;
4713 		}
4714 		LIST_REMOVE(sp, nst_list);
4715 		free(sp, M_TEMP);
4716 	}
4717 	nfsrv_backupstable();
4718 }
4719 
4720 /*
4721  * Append a record to the stable storage file.
4722  */
4723 APPLESTATIC void
4724 nfsrv_writestable(u_char *client, int len, int flag, NFSPROC_T *p)
4725 {
4726 	struct nfsrv_stablefirst *sf = &nfsrv_stablefirst;
4727 	struct nfst_rec *sp;
4728 	int error;
4729 
4730 	if (!(sf->nsf_flags & NFSNSF_OK) || sf->nsf_fp == NULL)
4731 		return;
4732 	sp = (struct nfst_rec *)malloc(sizeof (struct nfst_rec) +
4733 		len - 1, M_TEMP, M_WAITOK);
4734 	sp->len = len;
4735 	NFSBCOPY(client, sp->client, len);
4736 	sp->flag = flag;
4737 	error = NFSD_RDWR(UIO_WRITE, NFSFPVNODE(sf->nsf_fp),
4738 	    (caddr_t)sp, sizeof (struct nfst_rec) + len - 1, (off_t)0,
4739 	    UIO_SYSSPACE, (IO_SYNC | IO_APPEND), NFSFPCRED(sf->nsf_fp), NULL, p);
4740 	free(sp, M_TEMP);
4741 	if (error) {
4742 		sf->nsf_flags &= ~NFSNSF_OK;
4743 		printf("EEK! Can't write NfsV4 stable storage file\n");
4744 	}
4745 }
4746 
4747 /*
4748  * This function is called during the grace period to mark a client
4749  * that successfully reclaimed state.
4750  */
4751 static void
4752 nfsrv_markstable(struct nfsclient *clp)
4753 {
4754 	struct nfsrv_stable *sp;
4755 
4756 	/*
4757 	 * First find the client structure.
4758 	 */
4759 	LIST_FOREACH(sp, &nfsrv_stablefirst.nsf_head, nst_list) {
4760 		if (sp->nst_len == clp->lc_idlen &&
4761 		    !NFSBCMP(sp->nst_client, clp->lc_id, sp->nst_len))
4762 			break;
4763 	}
4764 	if (sp == LIST_END(&nfsrv_stablefirst.nsf_head))
4765 		return;
4766 
4767 	/*
4768 	 * Now, just mark it and set the nfsclient back pointer.
4769 	 */
4770 	sp->nst_flag |= NFSNST_GOTSTATE;
4771 	sp->nst_clp = clp;
4772 }
4773 
4774 /*
4775  * This function is called when a NFSv4.1 client does a ReclaimComplete.
4776  * Very similar to nfsrv_markstable(), except for the flag being set.
4777  */
4778 static void
4779 nfsrv_markreclaim(struct nfsclient *clp)
4780 {
4781 	struct nfsrv_stable *sp;
4782 
4783 	/*
4784 	 * First find the client structure.
4785 	 */
4786 	LIST_FOREACH(sp, &nfsrv_stablefirst.nsf_head, nst_list) {
4787 		if (sp->nst_len == clp->lc_idlen &&
4788 		    !NFSBCMP(sp->nst_client, clp->lc_id, sp->nst_len))
4789 			break;
4790 	}
4791 	if (sp == LIST_END(&nfsrv_stablefirst.nsf_head))
4792 		return;
4793 
4794 	/*
4795 	 * Now, just set the flag.
4796 	 */
4797 	sp->nst_flag |= NFSNST_RECLAIMED;
4798 }
4799 
4800 /*
4801  * This function is called for a reclaim, to see if it gets grace.
4802  * It returns 0 if a reclaim is allowed, 1 otherwise.
4803  */
4804 static int
4805 nfsrv_checkstable(struct nfsclient *clp)
4806 {
4807 	struct nfsrv_stable *sp;
4808 
4809 	/*
4810 	 * First, find the entry for the client.
4811 	 */
4812 	LIST_FOREACH(sp, &nfsrv_stablefirst.nsf_head, nst_list) {
4813 		if (sp->nst_len == clp->lc_idlen &&
4814 		    !NFSBCMP(sp->nst_client, clp->lc_id, sp->nst_len))
4815 			break;
4816 	}
4817 
4818 	/*
4819 	 * If not in the list, state was revoked or no state was issued
4820 	 * since the previous reboot, a reclaim is denied.
4821 	 */
4822 	if (sp == LIST_END(&nfsrv_stablefirst.nsf_head) ||
4823 	    (sp->nst_flag & NFSNST_REVOKE) ||
4824 	    !(nfsrv_stablefirst.nsf_flags & NFSNSF_OK))
4825 		return (1);
4826 	return (0);
4827 }
4828 
4829 /*
4830  * Test for and try to clear out a conflicting client. This is called by
4831  * nfsrv_lockctrl() and nfsrv_openctrl() when conflicts with other clients
4832  * a found.
4833  * The trick here is that it can't revoke a conflicting client with an
4834  * expired lease unless it holds the v4root lock, so...
4835  * If no v4root lock, get the lock and return 1 to indicate "try again".
4836  * Return 0 to indicate the conflict can't be revoked and 1 to indicate
4837  * the revocation worked and the conflicting client is "bye, bye", so it
4838  * can be tried again.
4839  * Return 2 to indicate that the vnode is VI_DOOMED after NFSVOPLOCK().
4840  * Unlocks State before a non-zero value is returned.
4841  */
4842 static int
4843 nfsrv_clientconflict(struct nfsclient *clp, int *haslockp, vnode_t vp,
4844     NFSPROC_T *p)
4845 {
4846 	int gotlock, lktype = 0;
4847 
4848 	/*
4849 	 * If lease hasn't expired, we can't fix it.
4850 	 */
4851 	if (clp->lc_expiry >= NFSD_MONOSEC ||
4852 	    !(nfsrv_stablefirst.nsf_flags & NFSNSF_UPDATEDONE))
4853 		return (0);
4854 	if (*haslockp == 0) {
4855 		NFSUNLOCKSTATE();
4856 		if (vp != NULL) {
4857 			lktype = NFSVOPISLOCKED(vp);
4858 			NFSVOPUNLOCK(vp, 0);
4859 		}
4860 		NFSLOCKV4ROOTMUTEX();
4861 		nfsv4_relref(&nfsv4rootfs_lock);
4862 		do {
4863 			gotlock = nfsv4_lock(&nfsv4rootfs_lock, 1, NULL,
4864 			    NFSV4ROOTLOCKMUTEXPTR, NULL);
4865 		} while (!gotlock);
4866 		NFSUNLOCKV4ROOTMUTEX();
4867 		*haslockp = 1;
4868 		if (vp != NULL) {
4869 			NFSVOPLOCK(vp, lktype | LK_RETRY);
4870 			if ((vp->v_iflag & VI_DOOMED) != 0)
4871 				return (2);
4872 		}
4873 		return (1);
4874 	}
4875 	NFSUNLOCKSTATE();
4876 
4877 	/*
4878 	 * Ok, we can expire the conflicting client.
4879 	 */
4880 	nfsrv_writestable(clp->lc_id, clp->lc_idlen, NFSNST_REVOKE, p);
4881 	nfsrv_backupstable();
4882 	nfsrv_cleanclient(clp, p);
4883 	nfsrv_freedeleglist(&clp->lc_deleg);
4884 	nfsrv_freedeleglist(&clp->lc_olddeleg);
4885 	LIST_REMOVE(clp, lc_hash);
4886 	nfsrv_zapclient(clp, p);
4887 	return (1);
4888 }
4889 
4890 /*
4891  * Resolve a delegation conflict.
4892  * Returns 0 to indicate the conflict was resolved without sleeping.
4893  * Return -1 to indicate that the caller should check for conflicts again.
4894  * Return > 0 for an error that should be returned, normally NFSERR_DELAY.
4895  *
4896  * Also, manipulate the nfsv4root_lock, as required. It isn't changed
4897  * for a return of 0, since there was no sleep and it could be required
4898  * later. It is released for a return of NFSERR_DELAY, since the caller
4899  * will return that error. It is released when a sleep was done waiting
4900  * for the delegation to be returned or expire (so that other nfsds can
4901  * handle ops). Then, it must be acquired for the write to stable storage.
4902  * (This function is somewhat similar to nfsrv_clientconflict(), but
4903  *  the semantics differ in a couple of subtle ways. The return of 0
4904  *  indicates the conflict was resolved without sleeping here, not
4905  *  that the conflict can't be resolved and the handling of nfsv4root_lock
4906  *  differs, as noted above.)
4907  * Unlocks State before returning a non-zero value.
4908  */
4909 static int
4910 nfsrv_delegconflict(struct nfsstate *stp, int *haslockp, NFSPROC_T *p,
4911     vnode_t vp)
4912 {
4913 	struct nfsclient *clp = stp->ls_clp;
4914 	int gotlock, error, lktype = 0, retrycnt, zapped_clp;
4915 	nfsv4stateid_t tstateid;
4916 	fhandle_t tfh;
4917 
4918 	/*
4919 	 * If the conflict is with an old delegation...
4920 	 */
4921 	if (stp->ls_flags & NFSLCK_OLDDELEG) {
4922 		/*
4923 		 * You can delete it, if it has expired.
4924 		 */
4925 		if (clp->lc_delegtime < NFSD_MONOSEC) {
4926 			nfsrv_freedeleg(stp);
4927 			NFSUNLOCKSTATE();
4928 			error = -1;
4929 			goto out;
4930 		}
4931 		NFSUNLOCKSTATE();
4932 		/*
4933 		 * During this delay, the old delegation could expire or it
4934 		 * could be recovered by the client via an Open with
4935 		 * CLAIM_DELEGATE_PREV.
4936 		 * Release the nfsv4root_lock, if held.
4937 		 */
4938 		if (*haslockp) {
4939 			*haslockp = 0;
4940 			NFSLOCKV4ROOTMUTEX();
4941 			nfsv4_unlock(&nfsv4rootfs_lock, 1);
4942 			NFSUNLOCKV4ROOTMUTEX();
4943 		}
4944 		error = NFSERR_DELAY;
4945 		goto out;
4946 	}
4947 
4948 	/*
4949 	 * It's a current delegation, so:
4950 	 * - check to see if the delegation has expired
4951 	 *   - if so, get the v4root lock and then expire it
4952 	 */
4953 	if (!(stp->ls_flags & NFSLCK_DELEGRECALL)) {
4954 		/*
4955 		 * - do a recall callback, since not yet done
4956 		 * For now, never allow truncate to be set. To use
4957 		 * truncate safely, it must be guaranteed that the
4958 		 * Remove, Rename or Setattr with size of 0 will
4959 		 * succeed and that would require major changes to
4960 		 * the VFS/Vnode OPs.
4961 		 * Set the expiry time large enough so that it won't expire
4962 		 * until after the callback, then set it correctly, once
4963 		 * the callback is done. (The delegation will now time
4964 		 * out whether or not the Recall worked ok. The timeout
4965 		 * will be extended when ops are done on the delegation
4966 		 * stateid, up to the timelimit.)
4967 		 */
4968 		stp->ls_delegtime = NFSD_MONOSEC + (2 * nfsrv_lease) +
4969 		    NFSRV_LEASEDELTA;
4970 		stp->ls_delegtimelimit = NFSD_MONOSEC + (6 * nfsrv_lease) +
4971 		    NFSRV_LEASEDELTA;
4972 		stp->ls_flags |= NFSLCK_DELEGRECALL;
4973 
4974 		/*
4975 		 * Loop NFSRV_CBRETRYCNT times while the CBRecall replies
4976 		 * NFSERR_BADSTATEID or NFSERR_BADHANDLE. This is done
4977 		 * in order to try and avoid a race that could happen
4978 		 * when a CBRecall request passed the Open reply with
4979 		 * the delegation in it when transitting the network.
4980 		 * Since nfsrv_docallback will sleep, don't use stp after
4981 		 * the call.
4982 		 */
4983 		NFSBCOPY((caddr_t)&stp->ls_stateid, (caddr_t)&tstateid,
4984 		    sizeof (tstateid));
4985 		NFSBCOPY((caddr_t)&stp->ls_lfp->lf_fh, (caddr_t)&tfh,
4986 		    sizeof (tfh));
4987 		NFSUNLOCKSTATE();
4988 		if (*haslockp) {
4989 			*haslockp = 0;
4990 			NFSLOCKV4ROOTMUTEX();
4991 			nfsv4_unlock(&nfsv4rootfs_lock, 1);
4992 			NFSUNLOCKV4ROOTMUTEX();
4993 		}
4994 		retrycnt = 0;
4995 		do {
4996 		    error = nfsrv_docallback(clp, NFSV4OP_CBRECALL,
4997 			&tstateid, 0, &tfh, NULL, NULL, p);
4998 		    retrycnt++;
4999 		} while ((error == NFSERR_BADSTATEID ||
5000 		    error == NFSERR_BADHANDLE) && retrycnt < NFSV4_CBRETRYCNT);
5001 		error = NFSERR_DELAY;
5002 		goto out;
5003 	}
5004 
5005 	if (clp->lc_expiry >= NFSD_MONOSEC &&
5006 	    stp->ls_delegtime >= NFSD_MONOSEC) {
5007 		NFSUNLOCKSTATE();
5008 		/*
5009 		 * A recall has been done, but it has not yet expired.
5010 		 * So, RETURN_DELAY.
5011 		 */
5012 		if (*haslockp) {
5013 			*haslockp = 0;
5014 			NFSLOCKV4ROOTMUTEX();
5015 			nfsv4_unlock(&nfsv4rootfs_lock, 1);
5016 			NFSUNLOCKV4ROOTMUTEX();
5017 		}
5018 		error = NFSERR_DELAY;
5019 		goto out;
5020 	}
5021 
5022 	/*
5023 	 * If we don't yet have the lock, just get it and then return,
5024 	 * since we need that before deleting expired state, such as
5025 	 * this delegation.
5026 	 * When getting the lock, unlock the vnode, so other nfsds that
5027 	 * are in progress, won't get stuck waiting for the vnode lock.
5028 	 */
5029 	if (*haslockp == 0) {
5030 		NFSUNLOCKSTATE();
5031 		if (vp != NULL) {
5032 			lktype = NFSVOPISLOCKED(vp);
5033 			NFSVOPUNLOCK(vp, 0);
5034 		}
5035 		NFSLOCKV4ROOTMUTEX();
5036 		nfsv4_relref(&nfsv4rootfs_lock);
5037 		do {
5038 			gotlock = nfsv4_lock(&nfsv4rootfs_lock, 1, NULL,
5039 			    NFSV4ROOTLOCKMUTEXPTR, NULL);
5040 		} while (!gotlock);
5041 		NFSUNLOCKV4ROOTMUTEX();
5042 		*haslockp = 1;
5043 		if (vp != NULL) {
5044 			NFSVOPLOCK(vp, lktype | LK_RETRY);
5045 			if ((vp->v_iflag & VI_DOOMED) != 0) {
5046 				*haslockp = 0;
5047 				NFSLOCKV4ROOTMUTEX();
5048 				nfsv4_unlock(&nfsv4rootfs_lock, 1);
5049 				NFSUNLOCKV4ROOTMUTEX();
5050 				error = NFSERR_PERM;
5051 				goto out;
5052 			}
5053 		}
5054 		error = -1;
5055 		goto out;
5056 	}
5057 
5058 	NFSUNLOCKSTATE();
5059 	/*
5060 	 * Ok, we can delete the expired delegation.
5061 	 * First, write the Revoke record to stable storage and then
5062 	 * clear out the conflict.
5063 	 * Since all other nfsd threads are now blocked, we can safely
5064 	 * sleep without the state changing.
5065 	 */
5066 	nfsrv_writestable(clp->lc_id, clp->lc_idlen, NFSNST_REVOKE, p);
5067 	nfsrv_backupstable();
5068 	if (clp->lc_expiry < NFSD_MONOSEC) {
5069 		nfsrv_cleanclient(clp, p);
5070 		nfsrv_freedeleglist(&clp->lc_deleg);
5071 		nfsrv_freedeleglist(&clp->lc_olddeleg);
5072 		LIST_REMOVE(clp, lc_hash);
5073 		zapped_clp = 1;
5074 	} else {
5075 		nfsrv_freedeleg(stp);
5076 		zapped_clp = 0;
5077 	}
5078 	if (zapped_clp)
5079 		nfsrv_zapclient(clp, p);
5080 	error = -1;
5081 
5082 out:
5083 	NFSEXITCODE(error);
5084 	return (error);
5085 }
5086 
5087 /*
5088  * Check for a remove allowed, if remove is set to 1 and get rid of
5089  * delegations.
5090  */
5091 APPLESTATIC int
5092 nfsrv_checkremove(vnode_t vp, int remove, NFSPROC_T *p)
5093 {
5094 	struct nfsstate *stp;
5095 	struct nfslockfile *lfp;
5096 	int error, haslock = 0;
5097 	fhandle_t nfh;
5098 
5099 	/*
5100 	 * First, get the lock file structure.
5101 	 * (A return of -1 means no associated state, so remove ok.)
5102 	 */
5103 	error = nfsrv_getlockfh(vp, NFSLCK_CHECK, NULL, &nfh, p);
5104 tryagain:
5105 	NFSLOCKSTATE();
5106 	if (!error)
5107 		error = nfsrv_getlockfile(NFSLCK_CHECK, NULL, &lfp, &nfh, 0);
5108 	if (error) {
5109 		NFSUNLOCKSTATE();
5110 		if (haslock) {
5111 			NFSLOCKV4ROOTMUTEX();
5112 			nfsv4_unlock(&nfsv4rootfs_lock, 1);
5113 			NFSUNLOCKV4ROOTMUTEX();
5114 		}
5115 		if (error == -1)
5116 			error = 0;
5117 		goto out;
5118 	}
5119 
5120 	/*
5121 	 * Now, we must Recall any delegations.
5122 	 */
5123 	error = nfsrv_cleandeleg(vp, lfp, NULL, &haslock, p);
5124 	if (error) {
5125 		/*
5126 		 * nfsrv_cleandeleg() unlocks state for non-zero
5127 		 * return.
5128 		 */
5129 		if (error == -1)
5130 			goto tryagain;
5131 		if (haslock) {
5132 			NFSLOCKV4ROOTMUTEX();
5133 			nfsv4_unlock(&nfsv4rootfs_lock, 1);
5134 			NFSUNLOCKV4ROOTMUTEX();
5135 		}
5136 		goto out;
5137 	}
5138 
5139 	/*
5140 	 * Now, look for a conflicting open share.
5141 	 */
5142 	if (remove) {
5143 		/*
5144 		 * If the entry in the directory was the last reference to the
5145 		 * corresponding filesystem object, the object can be destroyed
5146 		 * */
5147 		if(lfp->lf_usecount>1)
5148 			LIST_FOREACH(stp, &lfp->lf_open, ls_file) {
5149 				if (stp->ls_flags & NFSLCK_WRITEDENY) {
5150 					error = NFSERR_FILEOPEN;
5151 					break;
5152 				}
5153 			}
5154 	}
5155 
5156 	NFSUNLOCKSTATE();
5157 	if (haslock) {
5158 		NFSLOCKV4ROOTMUTEX();
5159 		nfsv4_unlock(&nfsv4rootfs_lock, 1);
5160 		NFSUNLOCKV4ROOTMUTEX();
5161 	}
5162 
5163 out:
5164 	NFSEXITCODE(error);
5165 	return (error);
5166 }
5167 
5168 /*
5169  * Clear out all delegations for the file referred to by lfp.
5170  * May return NFSERR_DELAY, if there will be a delay waiting for
5171  * delegations to expire.
5172  * Returns -1 to indicate it slept while recalling a delegation.
5173  * This function has the side effect of deleting the nfslockfile structure,
5174  * if it no longer has associated state and didn't have to sleep.
5175  * Unlocks State before a non-zero value is returned.
5176  */
5177 static int
5178 nfsrv_cleandeleg(vnode_t vp, struct nfslockfile *lfp,
5179     struct nfsclient *clp, int *haslockp, NFSPROC_T *p)
5180 {
5181 	struct nfsstate *stp, *nstp;
5182 	int ret = 0;
5183 
5184 	stp = LIST_FIRST(&lfp->lf_deleg);
5185 	while (stp != LIST_END(&lfp->lf_deleg)) {
5186 		nstp = LIST_NEXT(stp, ls_file);
5187 		if (stp->ls_clp != clp) {
5188 			ret = nfsrv_delegconflict(stp, haslockp, p, vp);
5189 			if (ret) {
5190 				/*
5191 				 * nfsrv_delegconflict() unlocks state
5192 				 * when it returns non-zero.
5193 				 */
5194 				goto out;
5195 			}
5196 		}
5197 		stp = nstp;
5198 	}
5199 out:
5200 	NFSEXITCODE(ret);
5201 	return (ret);
5202 }
5203 
5204 /*
5205  * There are certain operations that, when being done outside of NFSv4,
5206  * require that any NFSv4 delegation for the file be recalled.
5207  * This function is to be called for those cases:
5208  * VOP_RENAME() - When a delegation is being recalled for any reason,
5209  *	the client may have to do Opens against the server, using the file's
5210  *	final component name. If the file has been renamed on the server,
5211  *	that component name will be incorrect and the Open will fail.
5212  * VOP_REMOVE() - Theoretically, a client could Open a file after it has
5213  *	been removed on the server, if there is a delegation issued to
5214  *	that client for the file. I say "theoretically" since clients
5215  *	normally do an Access Op before the Open and that Access Op will
5216  *	fail with ESTALE. Note that NFSv2 and 3 don't even do Opens, so
5217  *	they will detect the file's removal in the same manner. (There is
5218  *	one case where RFC3530 allows a client to do an Open without first
5219  *	doing an Access Op, which is passage of a check against the ACE
5220  *	returned with a Write delegation, but current practice is to ignore
5221  *	the ACE and always do an Access Op.)
5222  *	Since the functions can only be called with an unlocked vnode, this
5223  *	can't be done at this time.
5224  * VOP_ADVLOCK() - When a client holds a delegation, it can issue byte range
5225  *	locks locally in the client, which are not visible to the server. To
5226  *	deal with this, issuing of delegations for a vnode must be disabled
5227  *	and all delegations for the vnode recalled. This is done via the
5228  *	second function, using the VV_DISABLEDELEG vflag on the vnode.
5229  */
5230 APPLESTATIC void
5231 nfsd_recalldelegation(vnode_t vp, NFSPROC_T *p)
5232 {
5233 	time_t starttime;
5234 	int error;
5235 
5236 	/*
5237 	 * First, check to see if the server is currently running and it has
5238 	 * been called for a regular file when issuing delegations.
5239 	 */
5240 	if (newnfs_numnfsd == 0 || vp->v_type != VREG ||
5241 	    nfsrv_issuedelegs == 0)
5242 		return;
5243 
5244 	KASSERT((NFSVOPISLOCKED(vp) != LK_EXCLUSIVE), ("vp %p is locked", vp));
5245 	/*
5246 	 * First, get a reference on the nfsv4rootfs_lock so that an
5247 	 * exclusive lock cannot be acquired by another thread.
5248 	 */
5249 	NFSLOCKV4ROOTMUTEX();
5250 	nfsv4_getref(&nfsv4rootfs_lock, NULL, NFSV4ROOTLOCKMUTEXPTR, NULL);
5251 	NFSUNLOCKV4ROOTMUTEX();
5252 
5253 	/*
5254 	 * Now, call nfsrv_checkremove() in a loop while it returns
5255 	 * NFSERR_DELAY. Return upon any other error or when timed out.
5256 	 */
5257 	starttime = NFSD_MONOSEC;
5258 	do {
5259 		if (NFSVOPLOCK(vp, LK_EXCLUSIVE) == 0) {
5260 			error = nfsrv_checkremove(vp, 0, p);
5261 			NFSVOPUNLOCK(vp, 0);
5262 		} else
5263 			error = EPERM;
5264 		if (error == NFSERR_DELAY) {
5265 			if (NFSD_MONOSEC - starttime > NFS_REMOVETIMEO)
5266 				break;
5267 			/* Sleep for a short period of time */
5268 			(void) nfs_catnap(PZERO, 0, "nfsremove");
5269 		}
5270 	} while (error == NFSERR_DELAY);
5271 	NFSLOCKV4ROOTMUTEX();
5272 	nfsv4_relref(&nfsv4rootfs_lock);
5273 	NFSUNLOCKV4ROOTMUTEX();
5274 }
5275 
5276 APPLESTATIC void
5277 nfsd_disabledelegation(vnode_t vp, NFSPROC_T *p)
5278 {
5279 
5280 #ifdef VV_DISABLEDELEG
5281 	/*
5282 	 * First, flag issuance of delegations disabled.
5283 	 */
5284 	atomic_set_long(&vp->v_vflag, VV_DISABLEDELEG);
5285 #endif
5286 
5287 	/*
5288 	 * Then call nfsd_recalldelegation() to get rid of all extant
5289 	 * delegations.
5290 	 */
5291 	nfsd_recalldelegation(vp, p);
5292 }
5293 
5294 /*
5295  * Check for conflicting locks, etc. and then get rid of delegations.
5296  * (At one point I thought that I should get rid of delegations for any
5297  *  Setattr, since it could potentially disallow the I/O op (read or write)
5298  *  allowed by the delegation. However, Setattr Ops that aren't changing
5299  *  the size get a stateid of all 0s, so you can't tell if it is a delegation
5300  *  for the same client or a different one, so I decided to only get rid
5301  *  of delegations for other clients when the size is being changed.)
5302  * In general, a Setattr can disable NFS I/O Ops that are outstanding, such
5303  * as Write backs, even if there is no delegation, so it really isn't any
5304  * different?)
5305  */
5306 APPLESTATIC int
5307 nfsrv_checksetattr(vnode_t vp, struct nfsrv_descript *nd,
5308     nfsv4stateid_t *stateidp, struct nfsvattr *nvap, nfsattrbit_t *attrbitp,
5309     struct nfsexstuff *exp, NFSPROC_T *p)
5310 {
5311 	struct nfsstate st, *stp = &st;
5312 	struct nfslock lo, *lop = &lo;
5313 	int error = 0;
5314 	nfsquad_t clientid;
5315 
5316 	if (NFSISSET_ATTRBIT(attrbitp, NFSATTRBIT_SIZE)) {
5317 		stp->ls_flags = (NFSLCK_CHECK | NFSLCK_WRITEACCESS);
5318 		lop->lo_first = nvap->na_size;
5319 	} else {
5320 		stp->ls_flags = 0;
5321 		lop->lo_first = 0;
5322 	}
5323 	if (NFSISSET_ATTRBIT(attrbitp, NFSATTRBIT_OWNER) ||
5324 	    NFSISSET_ATTRBIT(attrbitp, NFSATTRBIT_OWNERGROUP) ||
5325 	    NFSISSET_ATTRBIT(attrbitp, NFSATTRBIT_MODE) ||
5326 	    NFSISSET_ATTRBIT(attrbitp, NFSATTRBIT_ACL))
5327 		stp->ls_flags |= NFSLCK_SETATTR;
5328 	if (stp->ls_flags == 0)
5329 		goto out;
5330 	lop->lo_end = NFS64BITSSET;
5331 	lop->lo_flags = NFSLCK_WRITE;
5332 	stp->ls_ownerlen = 0;
5333 	stp->ls_op = NULL;
5334 	stp->ls_uid = nd->nd_cred->cr_uid;
5335 	stp->ls_stateid.seqid = stateidp->seqid;
5336 	clientid.lval[0] = stp->ls_stateid.other[0] = stateidp->other[0];
5337 	clientid.lval[1] = stp->ls_stateid.other[1] = stateidp->other[1];
5338 	stp->ls_stateid.other[2] = stateidp->other[2];
5339 	error = nfsrv_lockctrl(vp, &stp, &lop, NULL, clientid,
5340 	    stateidp, exp, nd, p);
5341 
5342 out:
5343 	NFSEXITCODE2(error, nd);
5344 	return (error);
5345 }
5346 
5347 /*
5348  * Check for a write delegation and do a CBGETATTR if there is one, updating
5349  * the attributes, as required.
5350  * Should I return an error if I can't get the attributes? (For now, I'll
5351  * just return ok.
5352  */
5353 APPLESTATIC int
5354 nfsrv_checkgetattr(struct nfsrv_descript *nd, vnode_t vp,
5355     struct nfsvattr *nvap, nfsattrbit_t *attrbitp, struct ucred *cred,
5356     NFSPROC_T *p)
5357 {
5358 	struct nfsstate *stp;
5359 	struct nfslockfile *lfp;
5360 	struct nfsclient *clp;
5361 	struct nfsvattr nva;
5362 	fhandle_t nfh;
5363 	int error = 0;
5364 	nfsattrbit_t cbbits;
5365 	u_quad_t delegfilerev;
5366 
5367 	NFSCBGETATTR_ATTRBIT(attrbitp, &cbbits);
5368 	if (!NFSNONZERO_ATTRBIT(&cbbits))
5369 		goto out;
5370 	if (nfsrv_writedelegcnt == 0)
5371 		goto out;
5372 
5373 	/*
5374 	 * Get the lock file structure.
5375 	 * (A return of -1 means no associated state, so return ok.)
5376 	 */
5377 	error = nfsrv_getlockfh(vp, NFSLCK_CHECK, NULL, &nfh, p);
5378 	NFSLOCKSTATE();
5379 	if (!error)
5380 		error = nfsrv_getlockfile(NFSLCK_CHECK, NULL, &lfp, &nfh, 0);
5381 	if (error) {
5382 		NFSUNLOCKSTATE();
5383 		if (error == -1)
5384 			error = 0;
5385 		goto out;
5386 	}
5387 
5388 	/*
5389 	 * Now, look for a write delegation.
5390 	 */
5391 	LIST_FOREACH(stp, &lfp->lf_deleg, ls_file) {
5392 		if (stp->ls_flags & NFSLCK_DELEGWRITE)
5393 			break;
5394 	}
5395 	if (stp == LIST_END(&lfp->lf_deleg)) {
5396 		NFSUNLOCKSTATE();
5397 		goto out;
5398 	}
5399 	clp = stp->ls_clp;
5400 	delegfilerev = stp->ls_filerev;
5401 
5402 	/*
5403 	 * If the Write delegation was issued as a part of this Compound RPC
5404 	 * or if we have an Implied Clientid (used in a previous Op in this
5405 	 * compound) and it is the client the delegation was issued to,
5406 	 * just return ok.
5407 	 * I also assume that it is from the same client iff the network
5408 	 * host IP address is the same as the callback address. (Not
5409 	 * exactly correct by the RFC, but avoids a lot of Getattr
5410 	 * callbacks.)
5411 	 */
5412 	if (nd->nd_compref == stp->ls_compref ||
5413 	    ((nd->nd_flag & ND_IMPLIEDCLID) &&
5414 	     clp->lc_clientid.qval == nd->nd_clientid.qval) ||
5415 	     nfsaddr2_match(clp->lc_req.nr_nam, nd->nd_nam)) {
5416 		NFSUNLOCKSTATE();
5417 		goto out;
5418 	}
5419 
5420 	/*
5421 	 * We are now done with the delegation state structure,
5422 	 * so the statelock can be released and we can now tsleep().
5423 	 */
5424 
5425 	/*
5426 	 * Now, we must do the CB Getattr callback, to see if Change or Size
5427 	 * has changed.
5428 	 */
5429 	if (clp->lc_expiry >= NFSD_MONOSEC) {
5430 		NFSUNLOCKSTATE();
5431 		NFSVNO_ATTRINIT(&nva);
5432 		nva.na_filerev = NFS64BITSSET;
5433 		error = nfsrv_docallback(clp, NFSV4OP_CBGETATTR, NULL,
5434 		    0, &nfh, &nva, &cbbits, p);
5435 		if (!error) {
5436 			if ((nva.na_filerev != NFS64BITSSET &&
5437 			    nva.na_filerev > delegfilerev) ||
5438 			    (NFSVNO_ISSETSIZE(&nva) &&
5439 			     nva.na_size != nvap->na_size)) {
5440 				error = nfsvno_updfilerev(vp, nvap, cred, p);
5441 				if (NFSVNO_ISSETSIZE(&nva))
5442 					nvap->na_size = nva.na_size;
5443 			}
5444 		} else
5445 			error = 0;	/* Ignore callback errors for now. */
5446 	} else {
5447 		NFSUNLOCKSTATE();
5448 	}
5449 
5450 out:
5451 	NFSEXITCODE2(error, nd);
5452 	return (error);
5453 }
5454 
5455 /*
5456  * This function looks for openowners that haven't had any opens for
5457  * a while and throws them away. Called by an nfsd when NFSNSF_NOOPENS
5458  * is set.
5459  */
5460 APPLESTATIC void
5461 nfsrv_throwawayopens(NFSPROC_T *p)
5462 {
5463 	struct nfsclient *clp, *nclp;
5464 	struct nfsstate *stp, *nstp;
5465 	int i;
5466 
5467 	NFSLOCKSTATE();
5468 	nfsrv_stablefirst.nsf_flags &= ~NFSNSF_NOOPENS;
5469 	/*
5470 	 * For each client...
5471 	 */
5472 	for (i = 0; i < nfsrv_clienthashsize; i++) {
5473 	    LIST_FOREACH_SAFE(clp, &nfsclienthash[i], lc_hash, nclp) {
5474 		LIST_FOREACH_SAFE(stp, &clp->lc_open, ls_list, nstp) {
5475 			if (LIST_EMPTY(&stp->ls_open) &&
5476 			    (stp->ls_noopens > NFSNOOPEN ||
5477 			     (nfsrv_openpluslock * 2) >
5478 			     nfsrv_v4statelimit))
5479 				nfsrv_freeopenowner(stp, 0, p);
5480 		}
5481 	    }
5482 	}
5483 	NFSUNLOCKSTATE();
5484 }
5485 
5486 /*
5487  * This function checks to see if the credentials are the same.
5488  * Returns 1 for not same, 0 otherwise.
5489  */
5490 static int
5491 nfsrv_notsamecredname(struct nfsrv_descript *nd, struct nfsclient *clp)
5492 {
5493 
5494 	if (nd->nd_flag & ND_GSS) {
5495 		if (!(clp->lc_flags & LCL_GSS))
5496 			return (1);
5497 		if (clp->lc_flags & LCL_NAME) {
5498 			if (nd->nd_princlen != clp->lc_namelen ||
5499 			    NFSBCMP(nd->nd_principal, clp->lc_name,
5500 				clp->lc_namelen))
5501 				return (1);
5502 			else
5503 				return (0);
5504 		}
5505 		if (nd->nd_cred->cr_uid == clp->lc_uid)
5506 			return (0);
5507 		else
5508 			return (1);
5509 	} else if (clp->lc_flags & LCL_GSS)
5510 		return (1);
5511 	/*
5512 	 * For AUTH_SYS, allow the same uid or root. (This is underspecified
5513 	 * in RFC3530, which talks about principals, but doesn't say anything
5514 	 * about uids for AUTH_SYS.)
5515 	 */
5516 	if (nd->nd_cred->cr_uid == clp->lc_uid || nd->nd_cred->cr_uid == 0)
5517 		return (0);
5518 	else
5519 		return (1);
5520 }
5521 
5522 /*
5523  * Calculate the lease expiry time.
5524  */
5525 static time_t
5526 nfsrv_leaseexpiry(void)
5527 {
5528 
5529 	if (nfsrv_stablefirst.nsf_eograce > NFSD_MONOSEC)
5530 		return (NFSD_MONOSEC + 2 * (nfsrv_lease + NFSRV_LEASEDELTA));
5531 	return (NFSD_MONOSEC + nfsrv_lease + NFSRV_LEASEDELTA);
5532 }
5533 
5534 /*
5535  * Delay the delegation timeout as far as ls_delegtimelimit, as required.
5536  */
5537 static void
5538 nfsrv_delaydelegtimeout(struct nfsstate *stp)
5539 {
5540 
5541 	if ((stp->ls_flags & NFSLCK_DELEGRECALL) == 0)
5542 		return;
5543 
5544 	if ((stp->ls_delegtime + 15) > NFSD_MONOSEC &&
5545 	    stp->ls_delegtime < stp->ls_delegtimelimit) {
5546 		stp->ls_delegtime += nfsrv_lease;
5547 		if (stp->ls_delegtime > stp->ls_delegtimelimit)
5548 			stp->ls_delegtime = stp->ls_delegtimelimit;
5549 	}
5550 }
5551 
5552 /*
5553  * This function checks to see if there is any other state associated
5554  * with the openowner for this Open.
5555  * It returns 1 if there is no other state, 0 otherwise.
5556  */
5557 static int
5558 nfsrv_nootherstate(struct nfsstate *stp)
5559 {
5560 	struct nfsstate *tstp;
5561 
5562 	LIST_FOREACH(tstp, &stp->ls_openowner->ls_open, ls_list) {
5563 		if (tstp != stp || !LIST_EMPTY(&tstp->ls_lock))
5564 			return (0);
5565 	}
5566 	return (1);
5567 }
5568 
5569 /*
5570  * Create a list of lock deltas (changes to local byte range locking
5571  * that can be rolled back using the list) and apply the changes via
5572  * nfsvno_advlock(). Optionally, lock the list. It is expected that either
5573  * the rollback or update function will be called after this.
5574  * It returns an error (and rolls back, as required), if any nfsvno_advlock()
5575  * call fails. If it returns an error, it will unlock the list.
5576  */
5577 static int
5578 nfsrv_locallock(vnode_t vp, struct nfslockfile *lfp, int flags,
5579     uint64_t first, uint64_t end, struct nfslockconflict *cfp, NFSPROC_T *p)
5580 {
5581 	struct nfslock *lop, *nlop;
5582 	int error = 0;
5583 
5584 	/* Loop through the list of locks. */
5585 	lop = LIST_FIRST(&lfp->lf_locallock);
5586 	while (first < end && lop != NULL) {
5587 		nlop = LIST_NEXT(lop, lo_lckowner);
5588 		if (first >= lop->lo_end) {
5589 			/* not there yet */
5590 			lop = nlop;
5591 		} else if (first < lop->lo_first) {
5592 			/* new one starts before entry in list */
5593 			if (end <= lop->lo_first) {
5594 				/* no overlap between old and new */
5595 				error = nfsrv_dolocal(vp, lfp, flags,
5596 				    NFSLCK_UNLOCK, first, end, cfp, p);
5597 				if (error != 0)
5598 					break;
5599 				first = end;
5600 			} else {
5601 				/* handle fragment overlapped with new one */
5602 				error = nfsrv_dolocal(vp, lfp, flags,
5603 				    NFSLCK_UNLOCK, first, lop->lo_first, cfp,
5604 				    p);
5605 				if (error != 0)
5606 					break;
5607 				first = lop->lo_first;
5608 			}
5609 		} else {
5610 			/* new one overlaps this entry in list */
5611 			if (end <= lop->lo_end) {
5612 				/* overlaps all of new one */
5613 				error = nfsrv_dolocal(vp, lfp, flags,
5614 				    lop->lo_flags, first, end, cfp, p);
5615 				if (error != 0)
5616 					break;
5617 				first = end;
5618 			} else {
5619 				/* handle fragment overlapped with new one */
5620 				error = nfsrv_dolocal(vp, lfp, flags,
5621 				    lop->lo_flags, first, lop->lo_end, cfp, p);
5622 				if (error != 0)
5623 					break;
5624 				first = lop->lo_end;
5625 				lop = nlop;
5626 			}
5627 		}
5628 	}
5629 	if (first < end && error == 0)
5630 		/* handle fragment past end of list */
5631 		error = nfsrv_dolocal(vp, lfp, flags, NFSLCK_UNLOCK, first,
5632 		    end, cfp, p);
5633 
5634 	NFSEXITCODE(error);
5635 	return (error);
5636 }
5637 
5638 /*
5639  * Local lock unlock. Unlock all byte ranges that are no longer locked
5640  * by NFSv4. To do this, unlock any subranges of first-->end that
5641  * do not overlap with the byte ranges of any lock in the lfp->lf_lock
5642  * list. This list has all locks for the file held by other
5643  * <clientid, lockowner> tuples. The list is ordered by increasing
5644  * lo_first value, but may have entries that overlap each other, for
5645  * the case of read locks.
5646  */
5647 static void
5648 nfsrv_localunlock(vnode_t vp, struct nfslockfile *lfp, uint64_t init_first,
5649     uint64_t init_end, NFSPROC_T *p)
5650 {
5651 	struct nfslock *lop;
5652 	uint64_t first, end, prevfirst;
5653 
5654 	first = init_first;
5655 	end = init_end;
5656 	while (first < init_end) {
5657 		/* Loop through all nfs locks, adjusting first and end */
5658 		prevfirst = 0;
5659 		LIST_FOREACH(lop, &lfp->lf_lock, lo_lckfile) {
5660 			KASSERT(prevfirst <= lop->lo_first,
5661 			    ("nfsv4 locks out of order"));
5662 			KASSERT(lop->lo_first < lop->lo_end,
5663 			    ("nfsv4 bogus lock"));
5664 			prevfirst = lop->lo_first;
5665 			if (first >= lop->lo_first &&
5666 			    first < lop->lo_end)
5667 				/*
5668 				 * Overlaps with initial part, so trim
5669 				 * off that initial part by moving first past
5670 				 * it.
5671 				 */
5672 				first = lop->lo_end;
5673 			else if (end > lop->lo_first &&
5674 			    lop->lo_first > first) {
5675 				/*
5676 				 * This lock defines the end of the
5677 				 * segment to unlock, so set end to the
5678 				 * start of it and break out of the loop.
5679 				 */
5680 				end = lop->lo_first;
5681 				break;
5682 			}
5683 			if (first >= end)
5684 				/*
5685 				 * There is no segment left to do, so
5686 				 * break out of this loop and then exit
5687 				 * the outer while() since first will be set
5688 				 * to end, which must equal init_end here.
5689 				 */
5690 				break;
5691 		}
5692 		if (first < end) {
5693 			/* Unlock this segment */
5694 			(void) nfsrv_dolocal(vp, lfp, NFSLCK_UNLOCK,
5695 			    NFSLCK_READ, first, end, NULL, p);
5696 			nfsrv_locallock_commit(lfp, NFSLCK_UNLOCK,
5697 			    first, end);
5698 		}
5699 		/*
5700 		 * Now move past this segment and look for any further
5701 		 * segment in the range, if there is one.
5702 		 */
5703 		first = end;
5704 		end = init_end;
5705 	}
5706 }
5707 
5708 /*
5709  * Do the local lock operation and update the rollback list, as required.
5710  * Perform the rollback and return the error if nfsvno_advlock() fails.
5711  */
5712 static int
5713 nfsrv_dolocal(vnode_t vp, struct nfslockfile *lfp, int flags, int oldflags,
5714     uint64_t first, uint64_t end, struct nfslockconflict *cfp, NFSPROC_T *p)
5715 {
5716 	struct nfsrollback *rlp;
5717 	int error = 0, ltype, oldltype;
5718 
5719 	if (flags & NFSLCK_WRITE)
5720 		ltype = F_WRLCK;
5721 	else if (flags & NFSLCK_READ)
5722 		ltype = F_RDLCK;
5723 	else
5724 		ltype = F_UNLCK;
5725 	if (oldflags & NFSLCK_WRITE)
5726 		oldltype = F_WRLCK;
5727 	else if (oldflags & NFSLCK_READ)
5728 		oldltype = F_RDLCK;
5729 	else
5730 		oldltype = F_UNLCK;
5731 	if (ltype == oldltype || (oldltype == F_WRLCK && ltype == F_RDLCK))
5732 		/* nothing to do */
5733 		goto out;
5734 	error = nfsvno_advlock(vp, ltype, first, end, p);
5735 	if (error != 0) {
5736 		if (cfp != NULL) {
5737 			cfp->cl_clientid.lval[0] = 0;
5738 			cfp->cl_clientid.lval[1] = 0;
5739 			cfp->cl_first = 0;
5740 			cfp->cl_end = NFS64BITSSET;
5741 			cfp->cl_flags = NFSLCK_WRITE;
5742 			cfp->cl_ownerlen = 5;
5743 			NFSBCOPY("LOCAL", cfp->cl_owner, 5);
5744 		}
5745 		nfsrv_locallock_rollback(vp, lfp, p);
5746 	} else if (ltype != F_UNLCK) {
5747 		rlp = malloc(sizeof (struct nfsrollback), M_NFSDROLLBACK,
5748 		    M_WAITOK);
5749 		rlp->rlck_first = first;
5750 		rlp->rlck_end = end;
5751 		rlp->rlck_type = oldltype;
5752 		LIST_INSERT_HEAD(&lfp->lf_rollback, rlp, rlck_list);
5753 	}
5754 
5755 out:
5756 	NFSEXITCODE(error);
5757 	return (error);
5758 }
5759 
5760 /*
5761  * Roll back local lock changes and free up the rollback list.
5762  */
5763 static void
5764 nfsrv_locallock_rollback(vnode_t vp, struct nfslockfile *lfp, NFSPROC_T *p)
5765 {
5766 	struct nfsrollback *rlp, *nrlp;
5767 
5768 	LIST_FOREACH_SAFE(rlp, &lfp->lf_rollback, rlck_list, nrlp) {
5769 		(void) nfsvno_advlock(vp, rlp->rlck_type, rlp->rlck_first,
5770 		    rlp->rlck_end, p);
5771 		free(rlp, M_NFSDROLLBACK);
5772 	}
5773 	LIST_INIT(&lfp->lf_rollback);
5774 }
5775 
5776 /*
5777  * Update local lock list and delete rollback list (ie now committed to the
5778  * local locks). Most of the work is done by the internal function.
5779  */
5780 static void
5781 nfsrv_locallock_commit(struct nfslockfile *lfp, int flags, uint64_t first,
5782     uint64_t end)
5783 {
5784 	struct nfsrollback *rlp, *nrlp;
5785 	struct nfslock *new_lop, *other_lop;
5786 
5787 	new_lop = malloc(sizeof (struct nfslock), M_NFSDLOCK, M_WAITOK);
5788 	if (flags & (NFSLCK_READ | NFSLCK_WRITE))
5789 		other_lop = malloc(sizeof (struct nfslock), M_NFSDLOCK,
5790 		    M_WAITOK);
5791 	else
5792 		other_lop = NULL;
5793 	new_lop->lo_flags = flags;
5794 	new_lop->lo_first = first;
5795 	new_lop->lo_end = end;
5796 	nfsrv_updatelock(NULL, &new_lop, &other_lop, lfp);
5797 	if (new_lop != NULL)
5798 		free(new_lop, M_NFSDLOCK);
5799 	if (other_lop != NULL)
5800 		free(other_lop, M_NFSDLOCK);
5801 
5802 	/* and get rid of the rollback list */
5803 	LIST_FOREACH_SAFE(rlp, &lfp->lf_rollback, rlck_list, nrlp)
5804 		free(rlp, M_NFSDROLLBACK);
5805 	LIST_INIT(&lfp->lf_rollback);
5806 }
5807 
5808 /*
5809  * Lock the struct nfslockfile for local lock updating.
5810  */
5811 static void
5812 nfsrv_locklf(struct nfslockfile *lfp)
5813 {
5814 	int gotlock;
5815 
5816 	/* lf_usecount ensures *lfp won't be free'd */
5817 	lfp->lf_usecount++;
5818 	do {
5819 		gotlock = nfsv4_lock(&lfp->lf_locallock_lck, 1, NULL,
5820 		    NFSSTATEMUTEXPTR, NULL);
5821 	} while (gotlock == 0);
5822 	lfp->lf_usecount--;
5823 }
5824 
5825 /*
5826  * Unlock the struct nfslockfile after local lock updating.
5827  */
5828 static void
5829 nfsrv_unlocklf(struct nfslockfile *lfp)
5830 {
5831 
5832 	nfsv4_unlock(&lfp->lf_locallock_lck, 0);
5833 }
5834 
5835 /*
5836  * Clear out all state for the NFSv4 server.
5837  * Must be called by a thread that can sleep when no nfsds are running.
5838  */
5839 void
5840 nfsrv_throwawayallstate(NFSPROC_T *p)
5841 {
5842 	struct nfsclient *clp, *nclp;
5843 	struct nfslockfile *lfp, *nlfp;
5844 	int i;
5845 
5846 	/*
5847 	 * For each client, clean out the state and then free the structure.
5848 	 */
5849 	for (i = 0; i < nfsrv_clienthashsize; i++) {
5850 		LIST_FOREACH_SAFE(clp, &nfsclienthash[i], lc_hash, nclp) {
5851 			nfsrv_cleanclient(clp, p);
5852 			nfsrv_freedeleglist(&clp->lc_deleg);
5853 			nfsrv_freedeleglist(&clp->lc_olddeleg);
5854 			free(clp->lc_stateid, M_NFSDCLIENT);
5855 			free(clp, M_NFSDCLIENT);
5856 		}
5857 	}
5858 
5859 	/*
5860 	 * Also, free up any remaining lock file structures.
5861 	 */
5862 	for (i = 0; i < nfsrv_lockhashsize; i++) {
5863 		LIST_FOREACH_SAFE(lfp, &nfslockhash[i], lf_hash, nlfp)
5864 			nfsrv_freenfslockfile(lfp);
5865 	}
5866 }
5867 
5868 /*
5869  * Check the sequence# for the session and slot provided as an argument.
5870  * Also, renew the lease if the session will return NFS_OK.
5871  */
5872 int
5873 nfsrv_checksequence(struct nfsrv_descript *nd, uint32_t sequenceid,
5874     uint32_t *highest_slotidp, uint32_t *target_highest_slotidp, int cache_this,
5875     uint32_t *sflagsp, NFSPROC_T *p)
5876 {
5877 	struct nfsdsession *sep;
5878 	struct nfssessionhash *shp;
5879 	int error;
5880 	SVCXPRT *savxprt;
5881 
5882 	shp = NFSSESSIONHASH(nd->nd_sessionid);
5883 	NFSLOCKSESSION(shp);
5884 	sep = nfsrv_findsession(nd->nd_sessionid);
5885 	if (sep == NULL) {
5886 		NFSUNLOCKSESSION(shp);
5887 		return (NFSERR_BADSESSION);
5888 	}
5889 	error = nfsv4_seqsession(sequenceid, nd->nd_slotid, *highest_slotidp,
5890 	    sep->sess_slots, NULL, NFSV4_SLOTS - 1);
5891 	if (error != 0) {
5892 		NFSUNLOCKSESSION(shp);
5893 		return (error);
5894 	}
5895 	if (cache_this != 0)
5896 		nd->nd_flag |= ND_SAVEREPLY;
5897 	/* Renew the lease. */
5898 	sep->sess_clp->lc_expiry = nfsrv_leaseexpiry();
5899 	nd->nd_clientid.qval = sep->sess_clp->lc_clientid.qval;
5900 	nd->nd_flag |= ND_IMPLIEDCLID;
5901 
5902 	/*
5903 	 * If this session handles the backchannel, save the nd_xprt for this
5904 	 * RPC, since this is the one being used.
5905 	 * RFC-5661 specifies that the fore channel will be implicitly
5906 	 * bound by a Sequence operation.  However, since some NFSv4.1 clients
5907 	 * erroneously assumed that the back channel would be implicitly
5908 	 * bound as well, do the implicit binding unless a
5909 	 * BindConnectiontoSession has already been done on the session.
5910 	 */
5911 	if (sep->sess_clp->lc_req.nr_client != NULL &&
5912 	    sep->sess_cbsess.nfsess_xprt != nd->nd_xprt &&
5913 	    (sep->sess_crflags & NFSV4CRSESS_CONNBACKCHAN) != 0 &&
5914 	    (sep->sess_clp->lc_flags & LCL_DONEBINDCONN) == 0) {
5915 		NFSD_DEBUG(2,
5916 		    "nfsrv_checksequence: implicit back channel bind\n");
5917 		savxprt = sep->sess_cbsess.nfsess_xprt;
5918 		SVC_ACQUIRE(nd->nd_xprt);
5919 		nd->nd_xprt->xp_p2 =
5920 		    sep->sess_clp->lc_req.nr_client->cl_private;
5921 		nd->nd_xprt->xp_idletimeout = 0;	/* Disable timeout. */
5922 		sep->sess_cbsess.nfsess_xprt = nd->nd_xprt;
5923 		if (savxprt != NULL)
5924 			SVC_RELEASE(savxprt);
5925 	}
5926 
5927 	*sflagsp = 0;
5928 	if (sep->sess_clp->lc_req.nr_client == NULL)
5929 		*sflagsp |= NFSV4SEQ_CBPATHDOWN;
5930 	NFSUNLOCKSESSION(shp);
5931 	if (error == NFSERR_EXPIRED) {
5932 		*sflagsp |= NFSV4SEQ_EXPIREDALLSTATEREVOKED;
5933 		error = 0;
5934 	} else if (error == NFSERR_ADMINREVOKED) {
5935 		*sflagsp |= NFSV4SEQ_ADMINSTATEREVOKED;
5936 		error = 0;
5937 	}
5938 	*highest_slotidp = *target_highest_slotidp = NFSV4_SLOTS - 1;
5939 	return (0);
5940 }
5941 
5942 /*
5943  * Check/set reclaim complete for this session/clientid.
5944  */
5945 int
5946 nfsrv_checkreclaimcomplete(struct nfsrv_descript *nd)
5947 {
5948 	struct nfsdsession *sep;
5949 	struct nfssessionhash *shp;
5950 	int error = 0;
5951 
5952 	shp = NFSSESSIONHASH(nd->nd_sessionid);
5953 	NFSLOCKSTATE();
5954 	NFSLOCKSESSION(shp);
5955 	sep = nfsrv_findsession(nd->nd_sessionid);
5956 	if (sep == NULL) {
5957 		NFSUNLOCKSESSION(shp);
5958 		NFSUNLOCKSTATE();
5959 		return (NFSERR_BADSESSION);
5960 	}
5961 
5962 	/* Check to see if reclaim complete has already happened. */
5963 	if ((sep->sess_clp->lc_flags & LCL_RECLAIMCOMPLETE) != 0)
5964 		error = NFSERR_COMPLETEALREADY;
5965 	else {
5966 		sep->sess_clp->lc_flags |= LCL_RECLAIMCOMPLETE;
5967 		nfsrv_markreclaim(sep->sess_clp);
5968 	}
5969 	NFSUNLOCKSESSION(shp);
5970 	NFSUNLOCKSTATE();
5971 	return (error);
5972 }
5973 
5974 /*
5975  * Cache the reply in a session slot.
5976  */
5977 void
5978 nfsrv_cache_session(uint8_t *sessionid, uint32_t slotid, int repstat,
5979    struct mbuf **m)
5980 {
5981 	struct nfsdsession *sep;
5982 	struct nfssessionhash *shp;
5983 
5984 	shp = NFSSESSIONHASH(sessionid);
5985 	NFSLOCKSESSION(shp);
5986 	sep = nfsrv_findsession(sessionid);
5987 	if (sep == NULL) {
5988 		NFSUNLOCKSESSION(shp);
5989 		printf("nfsrv_cache_session: no session\n");
5990 		m_freem(*m);
5991 		return;
5992 	}
5993 	nfsv4_seqsess_cacherep(slotid, sep->sess_slots, repstat, m);
5994 	NFSUNLOCKSESSION(shp);
5995 }
5996 
5997 /*
5998  * Search for a session that matches the sessionid.
5999  */
6000 static struct nfsdsession *
6001 nfsrv_findsession(uint8_t *sessionid)
6002 {
6003 	struct nfsdsession *sep;
6004 	struct nfssessionhash *shp;
6005 
6006 	shp = NFSSESSIONHASH(sessionid);
6007 	LIST_FOREACH(sep, &shp->list, sess_hash) {
6008 		if (!NFSBCMP(sessionid, sep->sess_sessionid, NFSX_V4SESSIONID))
6009 			break;
6010 	}
6011 	return (sep);
6012 }
6013 
6014 /*
6015  * Destroy a session.
6016  */
6017 int
6018 nfsrv_destroysession(struct nfsrv_descript *nd, uint8_t *sessionid)
6019 {
6020 	int error, igotlock, samesess;
6021 
6022 	samesess = 0;
6023 	if (!NFSBCMP(sessionid, nd->nd_sessionid, NFSX_V4SESSIONID) &&
6024 	    (nd->nd_flag & ND_HASSEQUENCE) != 0) {
6025 		samesess = 1;
6026 		if ((nd->nd_flag & ND_LASTOP) == 0)
6027 			return (NFSERR_BADSESSION);
6028 	}
6029 
6030 	/* Lock out other nfsd threads */
6031 	NFSLOCKV4ROOTMUTEX();
6032 	nfsv4_relref(&nfsv4rootfs_lock);
6033 	do {
6034 		igotlock = nfsv4_lock(&nfsv4rootfs_lock, 1, NULL,
6035 		    NFSV4ROOTLOCKMUTEXPTR, NULL);
6036 	} while (igotlock == 0);
6037 	NFSUNLOCKV4ROOTMUTEX();
6038 
6039 	error = nfsrv_freesession(NULL, sessionid);
6040 	if (error == 0 && samesess != 0)
6041 		nd->nd_flag &= ~ND_HASSEQUENCE;
6042 
6043 	NFSLOCKV4ROOTMUTEX();
6044 	nfsv4_unlock(&nfsv4rootfs_lock, 1);
6045 	NFSUNLOCKV4ROOTMUTEX();
6046 	return (error);
6047 }
6048 
6049 /*
6050  * Bind a connection to a session.
6051  * For now, only certain variants are supported, since the current session
6052  * structure can only handle a single backchannel entry, which will be
6053  * applied to all connections if it is set.
6054  */
6055 int
6056 nfsrv_bindconnsess(struct nfsrv_descript *nd, uint8_t *sessionid, int *foreaftp)
6057 {
6058 	struct nfssessionhash *shp;
6059 	struct nfsdsession *sep;
6060 	struct nfsclient *clp;
6061 	SVCXPRT *savxprt;
6062 	int error;
6063 
6064 	error = 0;
6065 	shp = NFSSESSIONHASH(sessionid);
6066 	NFSLOCKSTATE();
6067 	NFSLOCKSESSION(shp);
6068 	sep = nfsrv_findsession(sessionid);
6069 	if (sep != NULL) {
6070 		clp = sep->sess_clp;
6071 		if (*foreaftp == NFSCDFC4_BACK ||
6072 		    *foreaftp == NFSCDFC4_BACK_OR_BOTH ||
6073 		    *foreaftp == NFSCDFC4_FORE_OR_BOTH) {
6074 			/* Try to set up a backchannel. */
6075 			if (clp->lc_req.nr_client == NULL) {
6076 				NFSD_DEBUG(2, "nfsrv_bindconnsess: acquire "
6077 				    "backchannel\n");
6078 				clp->lc_req.nr_client = (struct __rpc_client *)
6079 				    clnt_bck_create(nd->nd_xprt->xp_socket,
6080 				    sep->sess_cbprogram, NFSV4_CBVERS);
6081 			}
6082 			if (clp->lc_req.nr_client != NULL) {
6083 				NFSD_DEBUG(2, "nfsrv_bindconnsess: set up "
6084 				    "backchannel\n");
6085 				savxprt = sep->sess_cbsess.nfsess_xprt;
6086 				SVC_ACQUIRE(nd->nd_xprt);
6087 				nd->nd_xprt->xp_p2 =
6088 				    clp->lc_req.nr_client->cl_private;
6089 				/* Disable idle timeout. */
6090 				nd->nd_xprt->xp_idletimeout = 0;
6091 				sep->sess_cbsess.nfsess_xprt = nd->nd_xprt;
6092 				if (savxprt != NULL)
6093 					SVC_RELEASE(savxprt);
6094 				sep->sess_crflags |= NFSV4CRSESS_CONNBACKCHAN;
6095 				clp->lc_flags |= LCL_DONEBINDCONN;
6096 				if (*foreaftp == NFSCDFS4_BACK)
6097 					*foreaftp = NFSCDFS4_BACK;
6098 				else
6099 					*foreaftp = NFSCDFS4_BOTH;
6100 			} else if (*foreaftp != NFSCDFC4_BACK) {
6101 				NFSD_DEBUG(2, "nfsrv_bindconnsess: can't set "
6102 				    "up backchannel\n");
6103 				sep->sess_crflags &= ~NFSV4CRSESS_CONNBACKCHAN;
6104 				clp->lc_flags |= LCL_DONEBINDCONN;
6105 				*foreaftp = NFSCDFS4_FORE;
6106 			} else {
6107 				error = NFSERR_NOTSUPP;
6108 				printf("nfsrv_bindconnsess: Can't add "
6109 				    "backchannel\n");
6110 			}
6111 		} else {
6112 			NFSD_DEBUG(2, "nfsrv_bindconnsess: Set forechannel\n");
6113 			clp->lc_flags |= LCL_DONEBINDCONN;
6114 			*foreaftp = NFSCDFS4_FORE;
6115 		}
6116 	} else
6117 		error = NFSERR_BADSESSION;
6118 	NFSUNLOCKSESSION(shp);
6119 	NFSUNLOCKSTATE();
6120 	return (error);
6121 }
6122 
6123 /*
6124  * Free up a session structure.
6125  */
6126 static int
6127 nfsrv_freesession(struct nfsdsession *sep, uint8_t *sessionid)
6128 {
6129 	struct nfssessionhash *shp;
6130 	int i;
6131 
6132 	NFSLOCKSTATE();
6133 	if (sep == NULL) {
6134 		shp = NFSSESSIONHASH(sessionid);
6135 		NFSLOCKSESSION(shp);
6136 		sep = nfsrv_findsession(sessionid);
6137 	} else {
6138 		shp = NFSSESSIONHASH(sep->sess_sessionid);
6139 		NFSLOCKSESSION(shp);
6140 	}
6141 	if (sep != NULL) {
6142 		sep->sess_refcnt--;
6143 		if (sep->sess_refcnt > 0) {
6144 			NFSUNLOCKSESSION(shp);
6145 			NFSUNLOCKSTATE();
6146 			return (NFSERR_BACKCHANBUSY);
6147 		}
6148 		LIST_REMOVE(sep, sess_hash);
6149 		LIST_REMOVE(sep, sess_list);
6150 	}
6151 	NFSUNLOCKSESSION(shp);
6152 	NFSUNLOCKSTATE();
6153 	if (sep == NULL)
6154 		return (NFSERR_BADSESSION);
6155 	for (i = 0; i < NFSV4_SLOTS; i++)
6156 		if (sep->sess_slots[i].nfssl_reply != NULL)
6157 			m_freem(sep->sess_slots[i].nfssl_reply);
6158 	if (sep->sess_cbsess.nfsess_xprt != NULL)
6159 		SVC_RELEASE(sep->sess_cbsess.nfsess_xprt);
6160 	free(sep, M_NFSDSESSION);
6161 	return (0);
6162 }
6163 
6164 /*
6165  * Free a stateid.
6166  * RFC5661 says that it should fail when there are associated opens, locks
6167  * or delegations. Since stateids represent opens, I don't see how you can
6168  * free an open stateid (it will be free'd when closed), so this function
6169  * only works for lock stateids (freeing the lock_owner) or delegations.
6170  */
6171 int
6172 nfsrv_freestateid(struct nfsrv_descript *nd, nfsv4stateid_t *stateidp,
6173     NFSPROC_T *p)
6174 {
6175 	struct nfsclient *clp;
6176 	struct nfsstate *stp;
6177 	int error;
6178 
6179 	NFSLOCKSTATE();
6180 	/*
6181 	 * Look up the stateid
6182 	 */
6183 	error = nfsrv_getclient((nfsquad_t)((u_quad_t)0), CLOPS_RENEW, &clp,
6184 	    NULL, (nfsquad_t)((u_quad_t)0), 0, nd, p);
6185 	if (error == 0) {
6186 		/* First, check for a delegation. */
6187 		LIST_FOREACH(stp, &clp->lc_deleg, ls_list) {
6188 			if (!NFSBCMP(stp->ls_stateid.other, stateidp->other,
6189 			    NFSX_STATEIDOTHER))
6190 				break;
6191 		}
6192 		if (stp != NULL) {
6193 			nfsrv_freedeleg(stp);
6194 			NFSUNLOCKSTATE();
6195 			return (error);
6196 		}
6197 	}
6198 	/* Not a delegation, try for a lock_owner. */
6199 	if (error == 0)
6200 		error = nfsrv_getstate(clp, stateidp, 0, &stp);
6201 	if (error == 0 && ((stp->ls_flags & (NFSLCK_OPEN | NFSLCK_DELEGREAD |
6202 	    NFSLCK_DELEGWRITE)) != 0 || (stp->ls_flags & NFSLCK_LOCK) == 0))
6203 		/* Not a lock_owner stateid. */
6204 		error = NFSERR_LOCKSHELD;
6205 	if (error == 0 && !LIST_EMPTY(&stp->ls_lock))
6206 		error = NFSERR_LOCKSHELD;
6207 	if (error == 0)
6208 		nfsrv_freelockowner(stp, NULL, 0, p);
6209 	NFSUNLOCKSTATE();
6210 	return (error);
6211 }
6212 
6213 /*
6214  * Test a stateid.
6215  */
6216 int
6217 nfsrv_teststateid(struct nfsrv_descript *nd, nfsv4stateid_t *stateidp,
6218     NFSPROC_T *p)
6219 {
6220 	struct nfsclient *clp;
6221 	struct nfsstate *stp;
6222 	int error;
6223 
6224 	NFSLOCKSTATE();
6225 	/*
6226 	 * Look up the stateid
6227 	 */
6228 	error = nfsrv_getclient((nfsquad_t)((u_quad_t)0), CLOPS_RENEW, &clp,
6229 	    NULL, (nfsquad_t)((u_quad_t)0), 0, nd, p);
6230 	if (error == 0)
6231 		error = nfsrv_getstate(clp, stateidp, 0, &stp);
6232 	if (error == 0 && stateidp->seqid != 0 &&
6233 	    SEQ_LT(stateidp->seqid, stp->ls_stateid.seqid))
6234 		error = NFSERR_OLDSTATEID;
6235 	NFSUNLOCKSTATE();
6236 	return (error);
6237 }
6238 
6239 /*
6240  * Generate the xdr for an NFSv4.1 CBSequence Operation.
6241  */
6242 static int
6243 nfsv4_setcbsequence(struct nfsrv_descript *nd, struct nfsclient *clp,
6244     int dont_replycache, struct nfsdsession **sepp)
6245 {
6246 	struct nfsdsession *sep;
6247 	uint32_t *tl, slotseq = 0;
6248 	int maxslot, slotpos;
6249 	uint8_t sessionid[NFSX_V4SESSIONID];
6250 	int error;
6251 
6252 	error = nfsv4_getcbsession(clp, sepp);
6253 	if (error != 0)
6254 		return (error);
6255 	sep = *sepp;
6256 	(void)nfsv4_sequencelookup(NULL, &sep->sess_cbsess, &slotpos, &maxslot,
6257 	    &slotseq, sessionid);
6258 	KASSERT(maxslot >= 0, ("nfsv4_setcbsequence neg maxslot"));
6259 
6260 	/* Build the Sequence arguments. */
6261 	NFSM_BUILD(tl, uint32_t *, NFSX_V4SESSIONID + 5 * NFSX_UNSIGNED);
6262 	bcopy(sessionid, tl, NFSX_V4SESSIONID);
6263 	tl += NFSX_V4SESSIONID / NFSX_UNSIGNED;
6264 	nd->nd_slotseq = tl;
6265 	*tl++ = txdr_unsigned(slotseq);
6266 	*tl++ = txdr_unsigned(slotpos);
6267 	*tl++ = txdr_unsigned(maxslot);
6268 	if (dont_replycache == 0)
6269 		*tl++ = newnfs_true;
6270 	else
6271 		*tl++ = newnfs_false;
6272 	*tl = 0;			/* No referring call list, for now. */
6273 	nd->nd_flag |= ND_HASSEQUENCE;
6274 	return (0);
6275 }
6276 
6277 /*
6278  * Get a session for the callback.
6279  */
6280 static int
6281 nfsv4_getcbsession(struct nfsclient *clp, struct nfsdsession **sepp)
6282 {
6283 	struct nfsdsession *sep;
6284 
6285 	NFSLOCKSTATE();
6286 	LIST_FOREACH(sep, &clp->lc_session, sess_list) {
6287 		if ((sep->sess_crflags & NFSV4CRSESS_CONNBACKCHAN) != 0)
6288 			break;
6289 	}
6290 	if (sep == NULL) {
6291 		NFSUNLOCKSTATE();
6292 		return (NFSERR_BADSESSION);
6293 	}
6294 	sep->sess_refcnt++;
6295 	*sepp = sep;
6296 	NFSUNLOCKSTATE();
6297 	return (0);
6298 }
6299 
6300 /*
6301  * Free up all backchannel xprts.  This needs to be done when the nfsd threads
6302  * exit, since those transports will all be going away.
6303  * This is only called after all the nfsd threads are done performing RPCs,
6304  * so locking shouldn't be an issue.
6305  */
6306 APPLESTATIC void
6307 nfsrv_freeallbackchannel_xprts(void)
6308 {
6309 	struct nfsdsession *sep;
6310 	struct nfsclient *clp;
6311 	SVCXPRT *xprt;
6312 	int i;
6313 
6314 	for (i = 0; i < nfsrv_clienthashsize; i++) {
6315 		LIST_FOREACH(clp, &nfsclienthash[i], lc_hash) {
6316 			LIST_FOREACH(sep, &clp->lc_session, sess_list) {
6317 				xprt = sep->sess_cbsess.nfsess_xprt;
6318 				sep->sess_cbsess.nfsess_xprt = NULL;
6319 				if (xprt != NULL)
6320 					SVC_RELEASE(xprt);
6321 			}
6322 		}
6323 	}
6324 }
6325 
6326