xref: /freebsd/sys/fs/nfsserver/nfs_nfsdstate.c (revision d0ba1baed3f6e4936a0c1b89c25f6c59168ef6de)
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 NFSV4ROOTLOCKMUTEX;
47 NFSSTATESPINLOCK;
48 
49 SYSCTL_DECL(_vfs_nfsd);
50 int	nfsrv_statehashsize = NFSSTATEHASHSIZE;
51 SYSCTL_INT(_vfs_nfsd, OID_AUTO, statehashsize, CTLFLAG_RDTUN,
52     &nfsrv_statehashsize, 0,
53     "Size of state hash table set via loader.conf");
54 
55 int	nfsrv_clienthashsize = NFSCLIENTHASHSIZE;
56 SYSCTL_INT(_vfs_nfsd, OID_AUTO, clienthashsize, CTLFLAG_RDTUN,
57     &nfsrv_clienthashsize, 0,
58     "Size of client hash table set via loader.conf");
59 
60 int	nfsrv_lockhashsize = NFSLOCKHASHSIZE;
61 SYSCTL_INT(_vfs_nfsd, OID_AUTO, fhhashsize, CTLFLAG_RDTUN,
62     &nfsrv_lockhashsize, 0,
63     "Size of file handle hash table set via loader.conf");
64 
65 int	nfsrv_sessionhashsize = NFSSESSIONHASHSIZE;
66 SYSCTL_INT(_vfs_nfsd, OID_AUTO, sessionhashsize, CTLFLAG_RDTUN,
67     &nfsrv_sessionhashsize, 0,
68     "Size of session hash table set via loader.conf");
69 
70 static int	nfsrv_v4statelimit = NFSRV_V4STATELIMIT;
71 SYSCTL_INT(_vfs_nfsd, OID_AUTO, v4statelimit, CTLFLAG_RWTUN,
72     &nfsrv_v4statelimit, 0,
73     "High water limit for NFSv4 opens+locks+delegations");
74 
75 static int	nfsrv_writedelegifpos = 0;
76 SYSCTL_INT(_vfs_nfsd, OID_AUTO, writedelegifpos, CTLFLAG_RW,
77     &nfsrv_writedelegifpos, 0,
78     "Issue a write delegation for read opens if possible");
79 
80 static int	nfsrv_allowreadforwriteopen = 1;
81 SYSCTL_INT(_vfs_nfsd, OID_AUTO, allowreadforwriteopen, CTLFLAG_RW,
82     &nfsrv_allowreadforwriteopen, 0,
83     "Allow Reads to be done with Write Access StateIDs");
84 
85 /*
86  * Hash lists for nfs V4.
87  */
88 struct nfsclienthashhead	*nfsclienthash;
89 struct nfslockhashhead		*nfslockhash;
90 struct nfssessionhash		*nfssessionhash;
91 #endif	/* !APPLEKEXT */
92 
93 static u_int32_t nfsrv_openpluslock = 0, nfsrv_delegatecnt = 0;
94 static time_t nfsrvboottime;
95 static int nfsrv_returnoldstateid = 0, nfsrv_clients = 0;
96 static int nfsrv_clienthighwater = NFSRV_CLIENTHIGHWATER;
97 static int nfsrv_nogsscallback = 0;
98 static volatile int nfsrv_writedelegcnt = 0;
99 
100 /* local functions */
101 static void nfsrv_dumpaclient(struct nfsclient *clp,
102     struct nfsd_dumpclients *dumpp);
103 static void nfsrv_freeopenowner(struct nfsstate *stp, int cansleep,
104     NFSPROC_T *p);
105 static int nfsrv_freeopen(struct nfsstate *stp, vnode_t vp, int cansleep,
106     NFSPROC_T *p);
107 static void nfsrv_freelockowner(struct nfsstate *stp, vnode_t vp, int cansleep,
108     NFSPROC_T *p);
109 static void nfsrv_freeallnfslocks(struct nfsstate *stp, vnode_t vp,
110     int cansleep, NFSPROC_T *p);
111 static void nfsrv_freenfslock(struct nfslock *lop);
112 static void nfsrv_freenfslockfile(struct nfslockfile *lfp);
113 static void nfsrv_freedeleg(struct nfsstate *);
114 static int nfsrv_getstate(struct nfsclient *clp, nfsv4stateid_t *stateidp,
115     u_int32_t flags, struct nfsstate **stpp);
116 static void nfsrv_getowner(struct nfsstatehead *hp, struct nfsstate *new_stp,
117     struct nfsstate **stpp);
118 static int nfsrv_getlockfh(vnode_t vp, u_short flags,
119     struct nfslockfile *new_lfp, fhandle_t *nfhp, NFSPROC_T *p);
120 static int nfsrv_getlockfile(u_short flags, struct nfslockfile **new_lfpp,
121     struct nfslockfile **lfpp, fhandle_t *nfhp, int lockit);
122 static void nfsrv_insertlock(struct nfslock *new_lop,
123     struct nfslock *insert_lop, struct nfsstate *stp, struct nfslockfile *lfp);
124 static void nfsrv_updatelock(struct nfsstate *stp, struct nfslock **new_lopp,
125     struct nfslock **other_lopp, struct nfslockfile *lfp);
126 static int nfsrv_getipnumber(u_char *cp);
127 static int nfsrv_checkrestart(nfsquad_t clientid, u_int32_t flags,
128     nfsv4stateid_t *stateidp, int specialid);
129 static int nfsrv_checkgrace(struct nfsrv_descript *nd, struct nfsclient *clp,
130     u_int32_t flags);
131 static int nfsrv_docallback(struct nfsclient *clp, int procnum,
132     nfsv4stateid_t *stateidp, int trunc, fhandle_t *fhp,
133     struct nfsvattr *nap, nfsattrbit_t *attrbitp, NFSPROC_T *p);
134 static int nfsrv_cbcallargs(struct nfsrv_descript *nd, struct nfsclient *clp,
135     uint32_t callback, int op, const char *optag, struct nfsdsession **sepp);
136 static u_int32_t nfsrv_nextclientindex(void);
137 static u_int32_t nfsrv_nextstateindex(struct nfsclient *clp);
138 static void nfsrv_markstable(struct nfsclient *clp);
139 static void nfsrv_markreclaim(struct nfsclient *clp);
140 static int nfsrv_checkstable(struct nfsclient *clp);
141 static int nfsrv_clientconflict(struct nfsclient *clp, int *haslockp, struct
142     vnode *vp, NFSPROC_T *p);
143 static int nfsrv_delegconflict(struct nfsstate *stp, int *haslockp,
144     NFSPROC_T *p, vnode_t vp);
145 static int nfsrv_cleandeleg(vnode_t vp, struct nfslockfile *lfp,
146     struct nfsclient *clp, int *haslockp, NFSPROC_T *p);
147 static int nfsrv_notsamecredname(struct nfsrv_descript *nd,
148     struct nfsclient *clp);
149 static time_t nfsrv_leaseexpiry(void);
150 static void nfsrv_delaydelegtimeout(struct nfsstate *stp);
151 static int nfsrv_checkseqid(struct nfsrv_descript *nd, u_int32_t seqid,
152     struct nfsstate *stp, struct nfsrvcache *op);
153 static int nfsrv_nootherstate(struct nfsstate *stp);
154 static int nfsrv_locallock(vnode_t vp, struct nfslockfile *lfp, int flags,
155     uint64_t first, uint64_t end, struct nfslockconflict *cfp, NFSPROC_T *p);
156 static void nfsrv_localunlock(vnode_t vp, struct nfslockfile *lfp,
157     uint64_t init_first, uint64_t init_end, NFSPROC_T *p);
158 static int nfsrv_dolocal(vnode_t vp, struct nfslockfile *lfp, int flags,
159     int oldflags, uint64_t first, uint64_t end, struct nfslockconflict *cfp,
160     NFSPROC_T *p);
161 static void nfsrv_locallock_rollback(vnode_t vp, struct nfslockfile *lfp,
162     NFSPROC_T *p);
163 static void nfsrv_locallock_commit(struct nfslockfile *lfp, int flags,
164     uint64_t first, uint64_t end);
165 static void nfsrv_locklf(struct nfslockfile *lfp);
166 static void nfsrv_unlocklf(struct nfslockfile *lfp);
167 static struct nfsdsession *nfsrv_findsession(uint8_t *sessionid);
168 static int nfsrv_freesession(struct nfsdsession *sep, uint8_t *sessionid);
169 static int nfsv4_setcbsequence(struct nfsrv_descript *nd, struct nfsclient *clp,
170     int dont_replycache, struct nfsdsession **sepp);
171 static int nfsv4_getcbsession(struct nfsclient *clp, struct nfsdsession **sepp);
172 
173 /*
174  * Scan the client list for a match and either return the current one,
175  * create a new entry or return an error.
176  * If returning a non-error, the clp structure must either be linked into
177  * the client list or free'd.
178  */
179 APPLESTATIC int
180 nfsrv_setclient(struct nfsrv_descript *nd, struct nfsclient **new_clpp,
181     nfsquad_t *clientidp, nfsquad_t *confirmp, NFSPROC_T *p)
182 {
183 	struct nfsclient *clp = NULL, *new_clp = *new_clpp;
184 	int i, error = 0, ret;
185 	struct nfsstate *stp, *tstp;
186 	struct sockaddr_in *sad, *rad;
187 	struct nfsdsession *sep, *nsep;
188 	int zapit = 0, gotit, hasstate = 0, igotlock;
189 	static u_int64_t confirm_index = 0;
190 
191 	/*
192 	 * Check for state resource limit exceeded.
193 	 */
194 	if (nfsrv_openpluslock > nfsrv_v4statelimit) {
195 		error = NFSERR_RESOURCE;
196 		goto out;
197 	}
198 
199 	if (nfsrv_issuedelegs == 0 ||
200 	    ((nd->nd_flag & ND_GSS) != 0 && nfsrv_nogsscallback != 0))
201 		/*
202 		 * Don't do callbacks when delegations are disabled or
203 		 * for AUTH_GSS unless enabled via nfsrv_nogsscallback.
204 		 * If establishing a callback connection is attempted
205 		 * when a firewall is blocking the callback path, the
206 		 * server may wait too long for the connect attempt to
207 		 * succeed during the Open. Some clients, such as Linux,
208 		 * may timeout and give up on the Open before the server
209 		 * replies. Also, since AUTH_GSS callbacks are not
210 		 * yet interoperability tested, they might cause the
211 		 * server to crap out, if they get past the Init call to
212 		 * the client.
213 		 */
214 		new_clp->lc_program = 0;
215 
216 	/* Lock out other nfsd threads */
217 	NFSLOCKV4ROOTMUTEX();
218 	nfsv4_relref(&nfsv4rootfs_lock);
219 	do {
220 		igotlock = nfsv4_lock(&nfsv4rootfs_lock, 1, NULL,
221 		    NFSV4ROOTLOCKMUTEXPTR, NULL);
222 	} while (!igotlock);
223 	NFSUNLOCKV4ROOTMUTEX();
224 
225 	/*
226 	 * Search for a match in the client list.
227 	 */
228 	gotit = i = 0;
229 	while (i < nfsrv_clienthashsize && !gotit) {
230 	    LIST_FOREACH(clp, &nfsclienthash[i], lc_hash) {
231 		if (new_clp->lc_idlen == clp->lc_idlen &&
232 		    !NFSBCMP(new_clp->lc_id, clp->lc_id, clp->lc_idlen)) {
233 			gotit = 1;
234 			break;
235 		}
236 	    }
237 	    if (gotit == 0)
238 		i++;
239 	}
240 	if (!gotit ||
241 	    (clp->lc_flags & (LCL_NEEDSCONFIRM | LCL_ADMINREVOKED))) {
242 		if ((nd->nd_flag & ND_NFSV41) != 0 && confirmp->lval[1] != 0) {
243 			/*
244 			 * For NFSv4.1, if confirmp->lval[1] is non-zero, the
245 			 * client is trying to update a confirmed clientid.
246 			 */
247 			NFSLOCKV4ROOTMUTEX();
248 			nfsv4_unlock(&nfsv4rootfs_lock, 1);
249 			NFSUNLOCKV4ROOTMUTEX();
250 			confirmp->lval[1] = 0;
251 			error = NFSERR_NOENT;
252 			goto out;
253 		}
254 		/*
255 		 * Get rid of the old one.
256 		 */
257 		if (i != nfsrv_clienthashsize) {
258 			LIST_REMOVE(clp, lc_hash);
259 			nfsrv_cleanclient(clp, p);
260 			nfsrv_freedeleglist(&clp->lc_deleg);
261 			nfsrv_freedeleglist(&clp->lc_olddeleg);
262 			zapit = 1;
263 		}
264 		/*
265 		 * Add it after assigning a client id to it.
266 		 */
267 		new_clp->lc_flags |= LCL_NEEDSCONFIRM;
268 		if ((nd->nd_flag & ND_NFSV41) != 0)
269 			new_clp->lc_confirm.lval[0] = confirmp->lval[0] =
270 			    ++confirm_index;
271 		else
272 			confirmp->qval = new_clp->lc_confirm.qval =
273 			    ++confirm_index;
274 		clientidp->lval[0] = new_clp->lc_clientid.lval[0] =
275 		    (u_int32_t)nfsrvboottime;
276 		clientidp->lval[1] = new_clp->lc_clientid.lval[1] =
277 		    nfsrv_nextclientindex();
278 		new_clp->lc_stateindex = 0;
279 		new_clp->lc_statemaxindex = 0;
280 		new_clp->lc_cbref = 0;
281 		new_clp->lc_expiry = nfsrv_leaseexpiry();
282 		LIST_INIT(&new_clp->lc_open);
283 		LIST_INIT(&new_clp->lc_deleg);
284 		LIST_INIT(&new_clp->lc_olddeleg);
285 		LIST_INIT(&new_clp->lc_session);
286 		for (i = 0; i < nfsrv_statehashsize; i++)
287 			LIST_INIT(&new_clp->lc_stateid[i]);
288 		LIST_INSERT_HEAD(NFSCLIENTHASH(new_clp->lc_clientid), new_clp,
289 		    lc_hash);
290 		nfsstatsv1.srvclients++;
291 		nfsrv_openpluslock++;
292 		nfsrv_clients++;
293 		NFSLOCKV4ROOTMUTEX();
294 		nfsv4_unlock(&nfsv4rootfs_lock, 1);
295 		NFSUNLOCKV4ROOTMUTEX();
296 		if (zapit)
297 			nfsrv_zapclient(clp, p);
298 		*new_clpp = NULL;
299 		goto out;
300 	}
301 
302 	/*
303 	 * Now, handle the cases where the id is already issued.
304 	 */
305 	if (nfsrv_notsamecredname(nd, clp)) {
306 	    /*
307 	     * Check to see if there is expired state that should go away.
308 	     */
309 	    if (clp->lc_expiry < NFSD_MONOSEC &&
310 	        (!LIST_EMPTY(&clp->lc_open) || !LIST_EMPTY(&clp->lc_deleg))) {
311 		nfsrv_cleanclient(clp, p);
312 		nfsrv_freedeleglist(&clp->lc_deleg);
313 	    }
314 
315 	    /*
316 	     * If there is outstanding state, then reply NFSERR_CLIDINUSE per
317 	     * RFC3530 Sec. 8.1.2 last para.
318 	     */
319 	    if (!LIST_EMPTY(&clp->lc_deleg)) {
320 		hasstate = 1;
321 	    } else if (LIST_EMPTY(&clp->lc_open)) {
322 		hasstate = 0;
323 	    } else {
324 		hasstate = 0;
325 		/* Look for an Open on the OpenOwner */
326 		LIST_FOREACH(stp, &clp->lc_open, ls_list) {
327 		    if (!LIST_EMPTY(&stp->ls_open)) {
328 			hasstate = 1;
329 			break;
330 		    }
331 		}
332 	    }
333 	    if (hasstate) {
334 		/*
335 		 * If the uid doesn't match, return NFSERR_CLIDINUSE after
336 		 * filling out the correct ipaddr and portnum.
337 		 */
338 		sad = NFSSOCKADDR(new_clp->lc_req.nr_nam, struct sockaddr_in *);
339 		rad = NFSSOCKADDR(clp->lc_req.nr_nam, struct sockaddr_in *);
340 		sad->sin_addr.s_addr = rad->sin_addr.s_addr;
341 		sad->sin_port = rad->sin_port;
342 		NFSLOCKV4ROOTMUTEX();
343 		nfsv4_unlock(&nfsv4rootfs_lock, 1);
344 		NFSUNLOCKV4ROOTMUTEX();
345 		error = NFSERR_CLIDINUSE;
346 		goto out;
347 	    }
348 	}
349 
350 	if (NFSBCMP(new_clp->lc_verf, clp->lc_verf, NFSX_VERF)) {
351 		/*
352 		 * If the verifier has changed, the client has rebooted
353 		 * and a new client id is issued. The old state info
354 		 * can be thrown away once the SETCLIENTID_CONFIRM occurs.
355 		 */
356 		LIST_REMOVE(clp, lc_hash);
357 
358 		/* Get rid of all sessions on this clientid. */
359 		LIST_FOREACH_SAFE(sep, &clp->lc_session, sess_list, nsep) {
360 			ret = nfsrv_freesession(sep, NULL);
361 			if (ret != 0)
362 				printf("nfsrv_setclient: verifier changed free"
363 				    " session failed=%d\n", ret);
364 		}
365 
366 		new_clp->lc_flags |= LCL_NEEDSCONFIRM;
367 		if ((nd->nd_flag & ND_NFSV41) != 0)
368 			new_clp->lc_confirm.lval[0] = confirmp->lval[0] =
369 			    ++confirm_index;
370 		else
371 			confirmp->qval = new_clp->lc_confirm.qval =
372 			    ++confirm_index;
373 		clientidp->lval[0] = new_clp->lc_clientid.lval[0] =
374 		    nfsrvboottime;
375 		clientidp->lval[1] = new_clp->lc_clientid.lval[1] =
376 		    nfsrv_nextclientindex();
377 		new_clp->lc_stateindex = 0;
378 		new_clp->lc_statemaxindex = 0;
379 		new_clp->lc_cbref = 0;
380 		new_clp->lc_expiry = nfsrv_leaseexpiry();
381 
382 		/*
383 		 * Save the state until confirmed.
384 		 */
385 		LIST_NEWHEAD(&new_clp->lc_open, &clp->lc_open, ls_list);
386 		LIST_FOREACH(tstp, &new_clp->lc_open, ls_list)
387 			tstp->ls_clp = new_clp;
388 		LIST_NEWHEAD(&new_clp->lc_deleg, &clp->lc_deleg, ls_list);
389 		LIST_FOREACH(tstp, &new_clp->lc_deleg, ls_list)
390 			tstp->ls_clp = new_clp;
391 		LIST_NEWHEAD(&new_clp->lc_olddeleg, &clp->lc_olddeleg,
392 		    ls_list);
393 		LIST_FOREACH(tstp, &new_clp->lc_olddeleg, ls_list)
394 			tstp->ls_clp = new_clp;
395 		for (i = 0; i < nfsrv_statehashsize; i++) {
396 			LIST_NEWHEAD(&new_clp->lc_stateid[i],
397 			    &clp->lc_stateid[i], ls_hash);
398 			LIST_FOREACH(tstp, &new_clp->lc_stateid[i], ls_hash)
399 				tstp->ls_clp = new_clp;
400 		}
401 		LIST_INIT(&new_clp->lc_session);
402 		LIST_INSERT_HEAD(NFSCLIENTHASH(new_clp->lc_clientid), new_clp,
403 		    lc_hash);
404 		nfsstatsv1.srvclients++;
405 		nfsrv_openpluslock++;
406 		nfsrv_clients++;
407 		NFSLOCKV4ROOTMUTEX();
408 		nfsv4_unlock(&nfsv4rootfs_lock, 1);
409 		NFSUNLOCKV4ROOTMUTEX();
410 
411 		/*
412 		 * Must wait until any outstanding callback on the old clp
413 		 * completes.
414 		 */
415 		NFSLOCKSTATE();
416 		while (clp->lc_cbref) {
417 			clp->lc_flags |= LCL_WAKEUPWANTED;
418 			(void)mtx_sleep(clp, NFSSTATEMUTEXPTR, PZERO - 1,
419 			    "nfsd clp", 10 * hz);
420 		}
421 		NFSUNLOCKSTATE();
422 		nfsrv_zapclient(clp, p);
423 		*new_clpp = NULL;
424 		goto out;
425 	}
426 
427 	/* For NFSv4.1, mark that we found a confirmed clientid. */
428 	if ((nd->nd_flag & ND_NFSV41) != 0) {
429 		clientidp->lval[0] = clp->lc_clientid.lval[0];
430 		clientidp->lval[1] = clp->lc_clientid.lval[1];
431 		confirmp->lval[0] = 0;	/* Ignored by client */
432 		confirmp->lval[1] = 1;
433 	} else {
434 		/*
435 		 * id and verifier match, so update the net address info
436 		 * and get rid of any existing callback authentication
437 		 * handle, so a new one will be acquired.
438 		 */
439 		LIST_REMOVE(clp, lc_hash);
440 		new_clp->lc_flags |= (LCL_NEEDSCONFIRM | LCL_DONTCLEAN);
441 		new_clp->lc_expiry = nfsrv_leaseexpiry();
442 		confirmp->qval = new_clp->lc_confirm.qval = ++confirm_index;
443 		clientidp->lval[0] = new_clp->lc_clientid.lval[0] =
444 		    clp->lc_clientid.lval[0];
445 		clientidp->lval[1] = new_clp->lc_clientid.lval[1] =
446 		    clp->lc_clientid.lval[1];
447 		new_clp->lc_delegtime = clp->lc_delegtime;
448 		new_clp->lc_stateindex = clp->lc_stateindex;
449 		new_clp->lc_statemaxindex = clp->lc_statemaxindex;
450 		new_clp->lc_cbref = 0;
451 		LIST_NEWHEAD(&new_clp->lc_open, &clp->lc_open, ls_list);
452 		LIST_FOREACH(tstp, &new_clp->lc_open, ls_list)
453 			tstp->ls_clp = new_clp;
454 		LIST_NEWHEAD(&new_clp->lc_deleg, &clp->lc_deleg, ls_list);
455 		LIST_FOREACH(tstp, &new_clp->lc_deleg, ls_list)
456 			tstp->ls_clp = new_clp;
457 		LIST_NEWHEAD(&new_clp->lc_olddeleg, &clp->lc_olddeleg, ls_list);
458 		LIST_FOREACH(tstp, &new_clp->lc_olddeleg, ls_list)
459 			tstp->ls_clp = new_clp;
460 		for (i = 0; i < nfsrv_statehashsize; i++) {
461 			LIST_NEWHEAD(&new_clp->lc_stateid[i],
462 			    &clp->lc_stateid[i], ls_hash);
463 			LIST_FOREACH(tstp, &new_clp->lc_stateid[i], ls_hash)
464 				tstp->ls_clp = new_clp;
465 		}
466 		LIST_INIT(&new_clp->lc_session);
467 		LIST_INSERT_HEAD(NFSCLIENTHASH(new_clp->lc_clientid), new_clp,
468 		    lc_hash);
469 		nfsstatsv1.srvclients++;
470 		nfsrv_openpluslock++;
471 		nfsrv_clients++;
472 	}
473 	NFSLOCKV4ROOTMUTEX();
474 	nfsv4_unlock(&nfsv4rootfs_lock, 1);
475 	NFSUNLOCKV4ROOTMUTEX();
476 
477 	if ((nd->nd_flag & ND_NFSV41) == 0) {
478 		/*
479 		 * Must wait until any outstanding callback on the old clp
480 		 * completes.
481 		 */
482 		NFSLOCKSTATE();
483 		while (clp->lc_cbref) {
484 			clp->lc_flags |= LCL_WAKEUPWANTED;
485 			(void)mtx_sleep(clp, NFSSTATEMUTEXPTR, PZERO - 1,
486 			    "nfsdclp", 10 * hz);
487 		}
488 		NFSUNLOCKSTATE();
489 		nfsrv_zapclient(clp, p);
490 		*new_clpp = NULL;
491 	}
492 
493 out:
494 	NFSEXITCODE2(error, nd);
495 	return (error);
496 }
497 
498 /*
499  * Check to see if the client id exists and optionally confirm it.
500  */
501 APPLESTATIC int
502 nfsrv_getclient(nfsquad_t clientid, int opflags, struct nfsclient **clpp,
503     struct nfsdsession *nsep, nfsquad_t confirm, uint32_t cbprogram,
504     struct nfsrv_descript *nd, NFSPROC_T *p)
505 {
506 	struct nfsclient *clp;
507 	struct nfsstate *stp;
508 	int i;
509 	struct nfsclienthashhead *hp;
510 	int error = 0, igotlock, doneok;
511 	struct nfssessionhash *shp;
512 	struct nfsdsession *sep;
513 	uint64_t sessid[2];
514 	static uint64_t next_sess = 0;
515 
516 	if (clpp)
517 		*clpp = NULL;
518 	if ((nd == NULL || (nd->nd_flag & ND_NFSV41) == 0 ||
519 	    opflags != CLOPS_RENEW) && nfsrvboottime != clientid.lval[0]) {
520 		error = NFSERR_STALECLIENTID;
521 		goto out;
522 	}
523 
524 	/*
525 	 * If called with opflags == CLOPS_RENEW, the State Lock is
526 	 * already held. Otherwise, we need to get either that or,
527 	 * for the case of Confirm, lock out the nfsd threads.
528 	 */
529 	if (opflags & CLOPS_CONFIRM) {
530 		NFSLOCKV4ROOTMUTEX();
531 		nfsv4_relref(&nfsv4rootfs_lock);
532 		do {
533 			igotlock = nfsv4_lock(&nfsv4rootfs_lock, 1, NULL,
534 			    NFSV4ROOTLOCKMUTEXPTR, NULL);
535 		} while (!igotlock);
536 		/*
537 		 * Create a new sessionid here, since we need to do it where
538 		 * there is a mutex held to serialize update of next_sess.
539 		 */
540 		if ((nd->nd_flag & ND_NFSV41) != 0) {
541 			sessid[0] = ++next_sess;
542 			sessid[1] = clientid.qval;
543 		}
544 		NFSUNLOCKV4ROOTMUTEX();
545 	} else if (opflags != CLOPS_RENEW) {
546 		NFSLOCKSTATE();
547 	}
548 
549 	/* For NFSv4.1, the clp is acquired from the associated session. */
550 	if (nd != NULL && (nd->nd_flag & ND_NFSV41) != 0 &&
551 	    opflags == CLOPS_RENEW) {
552 		clp = NULL;
553 		if ((nd->nd_flag & ND_HASSEQUENCE) != 0) {
554 			shp = NFSSESSIONHASH(nd->nd_sessionid);
555 			NFSLOCKSESSION(shp);
556 			sep = nfsrv_findsession(nd->nd_sessionid);
557 			if (sep != NULL)
558 				clp = sep->sess_clp;
559 			NFSUNLOCKSESSION(shp);
560 		}
561 	} else {
562 		hp = NFSCLIENTHASH(clientid);
563 		LIST_FOREACH(clp, hp, lc_hash) {
564 			if (clp->lc_clientid.lval[1] == clientid.lval[1])
565 				break;
566 		}
567 	}
568 	if (clp == NULL) {
569 		if (opflags & CLOPS_CONFIRM)
570 			error = NFSERR_STALECLIENTID;
571 		else
572 			error = NFSERR_EXPIRED;
573 	} else if (clp->lc_flags & LCL_ADMINREVOKED) {
574 		/*
575 		 * If marked admin revoked, just return the error.
576 		 */
577 		error = NFSERR_ADMINREVOKED;
578 	}
579 	if (error) {
580 		if (opflags & CLOPS_CONFIRM) {
581 			NFSLOCKV4ROOTMUTEX();
582 			nfsv4_unlock(&nfsv4rootfs_lock, 1);
583 			NFSUNLOCKV4ROOTMUTEX();
584 		} else if (opflags != CLOPS_RENEW) {
585 			NFSUNLOCKSTATE();
586 		}
587 		goto out;
588 	}
589 
590 	/*
591 	 * Perform any operations specified by the opflags.
592 	 */
593 	if (opflags & CLOPS_CONFIRM) {
594 		if (((nd->nd_flag & ND_NFSV41) != 0 &&
595 		     clp->lc_confirm.lval[0] != confirm.lval[0]) ||
596 		    ((nd->nd_flag & ND_NFSV41) == 0 &&
597 		     clp->lc_confirm.qval != confirm.qval))
598 			error = NFSERR_STALECLIENTID;
599 		else if (nfsrv_notsamecredname(nd, clp))
600 			error = NFSERR_CLIDINUSE;
601 
602 		if (!error) {
603 		    if ((clp->lc_flags & (LCL_NEEDSCONFIRM | LCL_DONTCLEAN)) ==
604 			LCL_NEEDSCONFIRM) {
605 			/*
606 			 * Hang onto the delegations (as old delegations)
607 			 * for an Open with CLAIM_DELEGATE_PREV unless in
608 			 * grace, but get rid of the rest of the state.
609 			 */
610 			nfsrv_cleanclient(clp, p);
611 			nfsrv_freedeleglist(&clp->lc_olddeleg);
612 			if (nfsrv_checkgrace(nd, clp, 0)) {
613 			    /* In grace, so just delete delegations */
614 			    nfsrv_freedeleglist(&clp->lc_deleg);
615 			} else {
616 			    LIST_FOREACH(stp, &clp->lc_deleg, ls_list)
617 				stp->ls_flags |= NFSLCK_OLDDELEG;
618 			    clp->lc_delegtime = NFSD_MONOSEC +
619 				nfsrv_lease + NFSRV_LEASEDELTA;
620 			    LIST_NEWHEAD(&clp->lc_olddeleg, &clp->lc_deleg,
621 				ls_list);
622 			}
623 			if ((nd->nd_flag & ND_NFSV41) != 0)
624 			    clp->lc_program = cbprogram;
625 		    }
626 		    clp->lc_flags &= ~(LCL_NEEDSCONFIRM | LCL_DONTCLEAN);
627 		    if (clp->lc_program)
628 			clp->lc_flags |= LCL_NEEDSCBNULL;
629 		    /* For NFSv4.1, link the session onto the client. */
630 		    if (nsep != NULL) {
631 			/* Hold a reference on the xprt for a backchannel. */
632 			if ((nsep->sess_crflags & NFSV4CRSESS_CONNBACKCHAN)
633 			    != 0 && clp->lc_req.nr_client == NULL) {
634 			    clp->lc_req.nr_client = (struct __rpc_client *)
635 				clnt_bck_create(nd->nd_xprt->xp_socket,
636 				cbprogram, NFSV4_CBVERS);
637 			    if (clp->lc_req.nr_client != NULL) {
638 				SVC_ACQUIRE(nd->nd_xprt);
639 				nd->nd_xprt->xp_p2 =
640 				    clp->lc_req.nr_client->cl_private;
641 				/* Disable idle timeout. */
642 				nd->nd_xprt->xp_idletimeout = 0;
643 				nsep->sess_cbsess.nfsess_xprt = nd->nd_xprt;
644 			    } else
645 				nsep->sess_crflags &= ~NFSV4CRSESS_CONNBACKCHAN;
646 			}
647 			NFSBCOPY(sessid, nsep->sess_sessionid,
648 			    NFSX_V4SESSIONID);
649 			NFSBCOPY(sessid, nsep->sess_cbsess.nfsess_sessionid,
650 			    NFSX_V4SESSIONID);
651 			shp = NFSSESSIONHASH(nsep->sess_sessionid);
652 			NFSLOCKSTATE();
653 			NFSLOCKSESSION(shp);
654 			LIST_INSERT_HEAD(&shp->list, nsep, sess_hash);
655 			LIST_INSERT_HEAD(&clp->lc_session, nsep, sess_list);
656 			nsep->sess_clp = clp;
657 			NFSUNLOCKSESSION(shp);
658 			NFSUNLOCKSTATE();
659 		    }
660 		}
661 	} else if (clp->lc_flags & LCL_NEEDSCONFIRM) {
662 		error = NFSERR_EXPIRED;
663 	}
664 
665 	/*
666 	 * If called by the Renew Op, we must check the principal.
667 	 */
668 	if (!error && (opflags & CLOPS_RENEWOP)) {
669 	    if (nfsrv_notsamecredname(nd, clp)) {
670 		doneok = 0;
671 		for (i = 0; i < nfsrv_statehashsize && doneok == 0; i++) {
672 		    LIST_FOREACH(stp, &clp->lc_stateid[i], ls_hash) {
673 			if ((stp->ls_flags & NFSLCK_OPEN) &&
674 			    stp->ls_uid == nd->nd_cred->cr_uid) {
675 				doneok = 1;
676 				break;
677 			}
678 		    }
679 		}
680 		if (!doneok)
681 			error = NFSERR_ACCES;
682 	    }
683 	    if (!error && (clp->lc_flags & LCL_CBDOWN))
684 		error = NFSERR_CBPATHDOWN;
685 	}
686 	if ((!error || error == NFSERR_CBPATHDOWN) &&
687 	     (opflags & CLOPS_RENEW)) {
688 		clp->lc_expiry = nfsrv_leaseexpiry();
689 	}
690 	if (opflags & CLOPS_CONFIRM) {
691 		NFSLOCKV4ROOTMUTEX();
692 		nfsv4_unlock(&nfsv4rootfs_lock, 1);
693 		NFSUNLOCKV4ROOTMUTEX();
694 	} else if (opflags != CLOPS_RENEW) {
695 		NFSUNLOCKSTATE();
696 	}
697 	if (clpp)
698 		*clpp = clp;
699 
700 out:
701 	NFSEXITCODE2(error, nd);
702 	return (error);
703 }
704 
705 /*
706  * Perform the NFSv4.1 destroy clientid.
707  */
708 int
709 nfsrv_destroyclient(nfsquad_t clientid, NFSPROC_T *p)
710 {
711 	struct nfsclient *clp;
712 	struct nfsclienthashhead *hp;
713 	int error = 0, i, igotlock;
714 
715 	if (nfsrvboottime != clientid.lval[0]) {
716 		error = NFSERR_STALECLIENTID;
717 		goto out;
718 	}
719 
720 	/* Lock out other nfsd threads */
721 	NFSLOCKV4ROOTMUTEX();
722 	nfsv4_relref(&nfsv4rootfs_lock);
723 	do {
724 		igotlock = nfsv4_lock(&nfsv4rootfs_lock, 1, NULL,
725 		    NFSV4ROOTLOCKMUTEXPTR, NULL);
726 	} while (igotlock == 0);
727 	NFSUNLOCKV4ROOTMUTEX();
728 
729 	hp = NFSCLIENTHASH(clientid);
730 	LIST_FOREACH(clp, hp, lc_hash) {
731 		if (clp->lc_clientid.lval[1] == clientid.lval[1])
732 			break;
733 	}
734 	if (clp == NULL) {
735 		NFSLOCKV4ROOTMUTEX();
736 		nfsv4_unlock(&nfsv4rootfs_lock, 1);
737 		NFSUNLOCKV4ROOTMUTEX();
738 		/* Just return ok, since it is gone. */
739 		goto out;
740 	}
741 
742 	/* Scan for state on the clientid. */
743 	for (i = 0; i < nfsrv_statehashsize; i++)
744 		if (!LIST_EMPTY(&clp->lc_stateid[i])) {
745 			NFSLOCKV4ROOTMUTEX();
746 			nfsv4_unlock(&nfsv4rootfs_lock, 1);
747 			NFSUNLOCKV4ROOTMUTEX();
748 			error = NFSERR_CLIENTIDBUSY;
749 			goto out;
750 		}
751 	if (!LIST_EMPTY(&clp->lc_session) || !LIST_EMPTY(&clp->lc_deleg)) {
752 		NFSLOCKV4ROOTMUTEX();
753 		nfsv4_unlock(&nfsv4rootfs_lock, 1);
754 		NFSUNLOCKV4ROOTMUTEX();
755 		error = NFSERR_CLIENTIDBUSY;
756 		goto out;
757 	}
758 
759 	/* Destroy the clientid and return ok. */
760 	nfsrv_cleanclient(clp, p);
761 	nfsrv_freedeleglist(&clp->lc_deleg);
762 	nfsrv_freedeleglist(&clp->lc_olddeleg);
763 	LIST_REMOVE(clp, lc_hash);
764 	NFSLOCKV4ROOTMUTEX();
765 	nfsv4_unlock(&nfsv4rootfs_lock, 1);
766 	NFSUNLOCKV4ROOTMUTEX();
767 	nfsrv_zapclient(clp, p);
768 out:
769 	NFSEXITCODE2(error, nd);
770 	return (error);
771 }
772 
773 /*
774  * Called from the new nfssvc syscall to admin revoke a clientid.
775  * Returns 0 for success, error otherwise.
776  */
777 APPLESTATIC int
778 nfsrv_adminrevoke(struct nfsd_clid *revokep, NFSPROC_T *p)
779 {
780 	struct nfsclient *clp = NULL;
781 	int i, error = 0;
782 	int gotit, igotlock;
783 
784 	/*
785 	 * First, lock out the nfsd so that state won't change while the
786 	 * revocation record is being written to the stable storage restart
787 	 * file.
788 	 */
789 	NFSLOCKV4ROOTMUTEX();
790 	do {
791 		igotlock = nfsv4_lock(&nfsv4rootfs_lock, 1, NULL,
792 		    NFSV4ROOTLOCKMUTEXPTR, NULL);
793 	} while (!igotlock);
794 	NFSUNLOCKV4ROOTMUTEX();
795 
796 	/*
797 	 * Search for a match in the client list.
798 	 */
799 	gotit = i = 0;
800 	while (i < nfsrv_clienthashsize && !gotit) {
801 	    LIST_FOREACH(clp, &nfsclienthash[i], lc_hash) {
802 		if (revokep->nclid_idlen == clp->lc_idlen &&
803 		    !NFSBCMP(revokep->nclid_id, clp->lc_id, clp->lc_idlen)) {
804 			gotit = 1;
805 			break;
806 		}
807 	    }
808 	    i++;
809 	}
810 	if (!gotit) {
811 		NFSLOCKV4ROOTMUTEX();
812 		nfsv4_unlock(&nfsv4rootfs_lock, 0);
813 		NFSUNLOCKV4ROOTMUTEX();
814 		error = EPERM;
815 		goto out;
816 	}
817 
818 	/*
819 	 * Now, write out the revocation record
820 	 */
821 	nfsrv_writestable(clp->lc_id, clp->lc_idlen, NFSNST_REVOKE, p);
822 	nfsrv_backupstable();
823 
824 	/*
825 	 * and clear out the state, marking the clientid revoked.
826 	 */
827 	clp->lc_flags &= ~LCL_CALLBACKSON;
828 	clp->lc_flags |= LCL_ADMINREVOKED;
829 	nfsrv_cleanclient(clp, p);
830 	nfsrv_freedeleglist(&clp->lc_deleg);
831 	nfsrv_freedeleglist(&clp->lc_olddeleg);
832 	NFSLOCKV4ROOTMUTEX();
833 	nfsv4_unlock(&nfsv4rootfs_lock, 0);
834 	NFSUNLOCKV4ROOTMUTEX();
835 
836 out:
837 	NFSEXITCODE(error);
838 	return (error);
839 }
840 
841 /*
842  * Dump out stats for all clients. Called from nfssvc(2), that is used
843  * nfsstatsv1.
844  */
845 APPLESTATIC void
846 nfsrv_dumpclients(struct nfsd_dumpclients *dumpp, int maxcnt)
847 {
848 	struct nfsclient *clp;
849 	int i = 0, cnt = 0;
850 
851 	/*
852 	 * First, get a reference on the nfsv4rootfs_lock so that an
853 	 * exclusive lock cannot be acquired while dumping the clients.
854 	 */
855 	NFSLOCKV4ROOTMUTEX();
856 	nfsv4_getref(&nfsv4rootfs_lock, NULL, NFSV4ROOTLOCKMUTEXPTR, NULL);
857 	NFSUNLOCKV4ROOTMUTEX();
858 	NFSLOCKSTATE();
859 	/*
860 	 * Rattle through the client lists until done.
861 	 */
862 	while (i < nfsrv_clienthashsize && cnt < maxcnt) {
863 	    clp = LIST_FIRST(&nfsclienthash[i]);
864 	    while (clp != LIST_END(&nfsclienthash[i]) && cnt < maxcnt) {
865 		nfsrv_dumpaclient(clp, &dumpp[cnt]);
866 		cnt++;
867 		clp = LIST_NEXT(clp, lc_hash);
868 	    }
869 	    i++;
870 	}
871 	if (cnt < maxcnt)
872 	    dumpp[cnt].ndcl_clid.nclid_idlen = 0;
873 	NFSUNLOCKSTATE();
874 	NFSLOCKV4ROOTMUTEX();
875 	nfsv4_relref(&nfsv4rootfs_lock);
876 	NFSUNLOCKV4ROOTMUTEX();
877 }
878 
879 /*
880  * Dump stats for a client. Must be called with the NFSSTATELOCK and spl'd.
881  */
882 static void
883 nfsrv_dumpaclient(struct nfsclient *clp, struct nfsd_dumpclients *dumpp)
884 {
885 	struct nfsstate *stp, *openstp, *lckownstp;
886 	struct nfslock *lop;
887 	struct sockaddr *sad;
888 	struct sockaddr_in *rad;
889 	struct sockaddr_in6 *rad6;
890 
891 	dumpp->ndcl_nopenowners = dumpp->ndcl_nlockowners = 0;
892 	dumpp->ndcl_nopens = dumpp->ndcl_nlocks = 0;
893 	dumpp->ndcl_ndelegs = dumpp->ndcl_nolddelegs = 0;
894 	dumpp->ndcl_flags = clp->lc_flags;
895 	dumpp->ndcl_clid.nclid_idlen = clp->lc_idlen;
896 	NFSBCOPY(clp->lc_id, dumpp->ndcl_clid.nclid_id, clp->lc_idlen);
897 	sad = NFSSOCKADDR(clp->lc_req.nr_nam, struct sockaddr *);
898 	dumpp->ndcl_addrfam = sad->sa_family;
899 	if (sad->sa_family == AF_INET) {
900 		rad = (struct sockaddr_in *)sad;
901 		dumpp->ndcl_cbaddr.sin_addr = rad->sin_addr;
902 	} else {
903 		rad6 = (struct sockaddr_in6 *)sad;
904 		dumpp->ndcl_cbaddr.sin6_addr = rad6->sin6_addr;
905 	}
906 
907 	/*
908 	 * Now, scan the state lists and total up the opens and locks.
909 	 */
910 	LIST_FOREACH(stp, &clp->lc_open, ls_list) {
911 	    dumpp->ndcl_nopenowners++;
912 	    LIST_FOREACH(openstp, &stp->ls_open, ls_list) {
913 		dumpp->ndcl_nopens++;
914 		LIST_FOREACH(lckownstp, &openstp->ls_open, ls_list) {
915 		    dumpp->ndcl_nlockowners++;
916 		    LIST_FOREACH(lop, &lckownstp->ls_lock, lo_lckowner) {
917 			dumpp->ndcl_nlocks++;
918 		    }
919 		}
920 	    }
921 	}
922 
923 	/*
924 	 * and the delegation lists.
925 	 */
926 	LIST_FOREACH(stp, &clp->lc_deleg, ls_list) {
927 	    dumpp->ndcl_ndelegs++;
928 	}
929 	LIST_FOREACH(stp, &clp->lc_olddeleg, ls_list) {
930 	    dumpp->ndcl_nolddelegs++;
931 	}
932 }
933 
934 /*
935  * Dump out lock stats for a file.
936  */
937 APPLESTATIC void
938 nfsrv_dumplocks(vnode_t vp, struct nfsd_dumplocks *ldumpp, int maxcnt,
939     NFSPROC_T *p)
940 {
941 	struct nfsstate *stp;
942 	struct nfslock *lop;
943 	int cnt = 0;
944 	struct nfslockfile *lfp;
945 	struct sockaddr *sad;
946 	struct sockaddr_in *rad;
947 	struct sockaddr_in6 *rad6;
948 	int ret;
949 	fhandle_t nfh;
950 
951 	ret = nfsrv_getlockfh(vp, 0, NULL, &nfh, p);
952 	/*
953 	 * First, get a reference on the nfsv4rootfs_lock so that an
954 	 * exclusive lock on it cannot be acquired while dumping the locks.
955 	 */
956 	NFSLOCKV4ROOTMUTEX();
957 	nfsv4_getref(&nfsv4rootfs_lock, NULL, NFSV4ROOTLOCKMUTEXPTR, NULL);
958 	NFSUNLOCKV4ROOTMUTEX();
959 	NFSLOCKSTATE();
960 	if (!ret)
961 		ret = nfsrv_getlockfile(0, NULL, &lfp, &nfh, 0);
962 	if (ret) {
963 		ldumpp[0].ndlck_clid.nclid_idlen = 0;
964 		NFSUNLOCKSTATE();
965 		NFSLOCKV4ROOTMUTEX();
966 		nfsv4_relref(&nfsv4rootfs_lock);
967 		NFSUNLOCKV4ROOTMUTEX();
968 		return;
969 	}
970 
971 	/*
972 	 * For each open share on file, dump it out.
973 	 */
974 	stp = LIST_FIRST(&lfp->lf_open);
975 	while (stp != LIST_END(&lfp->lf_open) && cnt < maxcnt) {
976 		ldumpp[cnt].ndlck_flags = stp->ls_flags;
977 		ldumpp[cnt].ndlck_stateid.seqid = stp->ls_stateid.seqid;
978 		ldumpp[cnt].ndlck_stateid.other[0] = stp->ls_stateid.other[0];
979 		ldumpp[cnt].ndlck_stateid.other[1] = stp->ls_stateid.other[1];
980 		ldumpp[cnt].ndlck_stateid.other[2] = stp->ls_stateid.other[2];
981 		ldumpp[cnt].ndlck_owner.nclid_idlen =
982 		    stp->ls_openowner->ls_ownerlen;
983 		NFSBCOPY(stp->ls_openowner->ls_owner,
984 		    ldumpp[cnt].ndlck_owner.nclid_id,
985 		    stp->ls_openowner->ls_ownerlen);
986 		ldumpp[cnt].ndlck_clid.nclid_idlen = stp->ls_clp->lc_idlen;
987 		NFSBCOPY(stp->ls_clp->lc_id, ldumpp[cnt].ndlck_clid.nclid_id,
988 		    stp->ls_clp->lc_idlen);
989 		sad=NFSSOCKADDR(stp->ls_clp->lc_req.nr_nam, struct sockaddr *);
990 		ldumpp[cnt].ndlck_addrfam = sad->sa_family;
991 		if (sad->sa_family == AF_INET) {
992 			rad = (struct sockaddr_in *)sad;
993 			ldumpp[cnt].ndlck_cbaddr.sin_addr = rad->sin_addr;
994 		} else {
995 			rad6 = (struct sockaddr_in6 *)sad;
996 			ldumpp[cnt].ndlck_cbaddr.sin6_addr = rad6->sin6_addr;
997 		}
998 		stp = LIST_NEXT(stp, ls_file);
999 		cnt++;
1000 	}
1001 
1002 	/*
1003 	 * and all locks.
1004 	 */
1005 	lop = LIST_FIRST(&lfp->lf_lock);
1006 	while (lop != LIST_END(&lfp->lf_lock) && cnt < maxcnt) {
1007 		stp = lop->lo_stp;
1008 		ldumpp[cnt].ndlck_flags = lop->lo_flags;
1009 		ldumpp[cnt].ndlck_first = lop->lo_first;
1010 		ldumpp[cnt].ndlck_end = lop->lo_end;
1011 		ldumpp[cnt].ndlck_stateid.seqid = stp->ls_stateid.seqid;
1012 		ldumpp[cnt].ndlck_stateid.other[0] = stp->ls_stateid.other[0];
1013 		ldumpp[cnt].ndlck_stateid.other[1] = stp->ls_stateid.other[1];
1014 		ldumpp[cnt].ndlck_stateid.other[2] = stp->ls_stateid.other[2];
1015 		ldumpp[cnt].ndlck_owner.nclid_idlen = stp->ls_ownerlen;
1016 		NFSBCOPY(stp->ls_owner, ldumpp[cnt].ndlck_owner.nclid_id,
1017 		    stp->ls_ownerlen);
1018 		ldumpp[cnt].ndlck_clid.nclid_idlen = stp->ls_clp->lc_idlen;
1019 		NFSBCOPY(stp->ls_clp->lc_id, ldumpp[cnt].ndlck_clid.nclid_id,
1020 		    stp->ls_clp->lc_idlen);
1021 		sad=NFSSOCKADDR(stp->ls_clp->lc_req.nr_nam, struct sockaddr *);
1022 		ldumpp[cnt].ndlck_addrfam = sad->sa_family;
1023 		if (sad->sa_family == AF_INET) {
1024 			rad = (struct sockaddr_in *)sad;
1025 			ldumpp[cnt].ndlck_cbaddr.sin_addr = rad->sin_addr;
1026 		} else {
1027 			rad6 = (struct sockaddr_in6 *)sad;
1028 			ldumpp[cnt].ndlck_cbaddr.sin6_addr = rad6->sin6_addr;
1029 		}
1030 		lop = LIST_NEXT(lop, lo_lckfile);
1031 		cnt++;
1032 	}
1033 
1034 	/*
1035 	 * and the delegations.
1036 	 */
1037 	stp = LIST_FIRST(&lfp->lf_deleg);
1038 	while (stp != LIST_END(&lfp->lf_deleg) && cnt < maxcnt) {
1039 		ldumpp[cnt].ndlck_flags = stp->ls_flags;
1040 		ldumpp[cnt].ndlck_stateid.seqid = stp->ls_stateid.seqid;
1041 		ldumpp[cnt].ndlck_stateid.other[0] = stp->ls_stateid.other[0];
1042 		ldumpp[cnt].ndlck_stateid.other[1] = stp->ls_stateid.other[1];
1043 		ldumpp[cnt].ndlck_stateid.other[2] = stp->ls_stateid.other[2];
1044 		ldumpp[cnt].ndlck_owner.nclid_idlen = 0;
1045 		ldumpp[cnt].ndlck_clid.nclid_idlen = stp->ls_clp->lc_idlen;
1046 		NFSBCOPY(stp->ls_clp->lc_id, ldumpp[cnt].ndlck_clid.nclid_id,
1047 		    stp->ls_clp->lc_idlen);
1048 		sad=NFSSOCKADDR(stp->ls_clp->lc_req.nr_nam, struct sockaddr *);
1049 		ldumpp[cnt].ndlck_addrfam = sad->sa_family;
1050 		if (sad->sa_family == AF_INET) {
1051 			rad = (struct sockaddr_in *)sad;
1052 			ldumpp[cnt].ndlck_cbaddr.sin_addr = rad->sin_addr;
1053 		} else {
1054 			rad6 = (struct sockaddr_in6 *)sad;
1055 			ldumpp[cnt].ndlck_cbaddr.sin6_addr = rad6->sin6_addr;
1056 		}
1057 		stp = LIST_NEXT(stp, ls_file);
1058 		cnt++;
1059 	}
1060 
1061 	/*
1062 	 * If list isn't full, mark end of list by setting the client name
1063 	 * to zero length.
1064 	 */
1065 	if (cnt < maxcnt)
1066 		ldumpp[cnt].ndlck_clid.nclid_idlen = 0;
1067 	NFSUNLOCKSTATE();
1068 	NFSLOCKV4ROOTMUTEX();
1069 	nfsv4_relref(&nfsv4rootfs_lock);
1070 	NFSUNLOCKV4ROOTMUTEX();
1071 }
1072 
1073 /*
1074  * Server timer routine. It can scan any linked list, so long
1075  * as it holds the spin/mutex lock and there is no exclusive lock on
1076  * nfsv4rootfs_lock.
1077  * (For OpenBSD, a kthread is ok. For FreeBSD, I think it is ok
1078  *  to do this from a callout, since the spin locks work. For
1079  *  Darwin, I'm not sure what will work correctly yet.)
1080  * Should be called once per second.
1081  */
1082 APPLESTATIC void
1083 nfsrv_servertimer(void)
1084 {
1085 	struct nfsclient *clp, *nclp;
1086 	struct nfsstate *stp, *nstp;
1087 	int got_ref, i;
1088 
1089 	/*
1090 	 * Make sure nfsboottime is set. This is used by V3 as well
1091 	 * as V4. Note that nfsboottime is not nfsrvboottime, which is
1092 	 * only used by the V4 server for leases.
1093 	 */
1094 	if (nfsboottime.tv_sec == 0)
1095 		NFSSETBOOTTIME(nfsboottime);
1096 
1097 	/*
1098 	 * If server hasn't started yet, just return.
1099 	 */
1100 	NFSLOCKSTATE();
1101 	if (nfsrv_stablefirst.nsf_eograce == 0) {
1102 		NFSUNLOCKSTATE();
1103 		return;
1104 	}
1105 	if (!(nfsrv_stablefirst.nsf_flags & NFSNSF_UPDATEDONE)) {
1106 		if (!(nfsrv_stablefirst.nsf_flags & NFSNSF_GRACEOVER) &&
1107 		    NFSD_MONOSEC > nfsrv_stablefirst.nsf_eograce)
1108 			nfsrv_stablefirst.nsf_flags |=
1109 			    (NFSNSF_GRACEOVER | NFSNSF_NEEDLOCK);
1110 		NFSUNLOCKSTATE();
1111 		return;
1112 	}
1113 
1114 	/*
1115 	 * Try and get a reference count on the nfsv4rootfs_lock so that
1116 	 * no nfsd thread can acquire an exclusive lock on it before this
1117 	 * call is done. If it is already exclusively locked, just return.
1118 	 */
1119 	NFSLOCKV4ROOTMUTEX();
1120 	got_ref = nfsv4_getref_nonblock(&nfsv4rootfs_lock);
1121 	NFSUNLOCKV4ROOTMUTEX();
1122 	if (got_ref == 0) {
1123 		NFSUNLOCKSTATE();
1124 		return;
1125 	}
1126 
1127 	/*
1128 	 * For each client...
1129 	 */
1130 	for (i = 0; i < nfsrv_clienthashsize; i++) {
1131 	    clp = LIST_FIRST(&nfsclienthash[i]);
1132 	    while (clp != LIST_END(&nfsclienthash[i])) {
1133 		nclp = LIST_NEXT(clp, lc_hash);
1134 		if (!(clp->lc_flags & LCL_EXPIREIT)) {
1135 		    if (((clp->lc_expiry + NFSRV_STALELEASE) < NFSD_MONOSEC
1136 			 && ((LIST_EMPTY(&clp->lc_deleg)
1137 			      && LIST_EMPTY(&clp->lc_open)) ||
1138 			     nfsrv_clients > nfsrv_clienthighwater)) ||
1139 			(clp->lc_expiry + NFSRV_MOULDYLEASE) < NFSD_MONOSEC ||
1140 			(clp->lc_expiry < NFSD_MONOSEC &&
1141 			 (nfsrv_openpluslock * 10 / 9) > nfsrv_v4statelimit)) {
1142 			/*
1143 			 * Lease has expired several nfsrv_lease times ago:
1144 			 * PLUS
1145 			 *    - no state is associated with it
1146 			 *    OR
1147 			 *    - above high water mark for number of clients
1148 			 *      (nfsrv_clienthighwater should be large enough
1149 			 *       that this only occurs when clients fail to
1150 			 *       use the same nfs_client_id4.id. Maybe somewhat
1151 			 *       higher that the maximum number of clients that
1152 			 *       will mount this server?)
1153 			 * OR
1154 			 * Lease has expired a very long time ago
1155 			 * OR
1156 			 * Lease has expired PLUS the number of opens + locks
1157 			 * has exceeded 90% of capacity
1158 			 *
1159 			 * --> Mark for expiry. The actual expiry will be done
1160 			 *     by an nfsd sometime soon.
1161 			 */
1162 			clp->lc_flags |= LCL_EXPIREIT;
1163 			nfsrv_stablefirst.nsf_flags |=
1164 			    (NFSNSF_NEEDLOCK | NFSNSF_EXPIREDCLIENT);
1165 		    } else {
1166 			/*
1167 			 * If there are no opens, increment no open tick cnt
1168 			 * If time exceeds NFSNOOPEN, mark it to be thrown away
1169 			 * otherwise, if there is an open, reset no open time
1170 			 * Hopefully, this will avoid excessive re-creation
1171 			 * of open owners and subsequent open confirms.
1172 			 */
1173 			stp = LIST_FIRST(&clp->lc_open);
1174 			while (stp != LIST_END(&clp->lc_open)) {
1175 				nstp = LIST_NEXT(stp, ls_list);
1176 				if (LIST_EMPTY(&stp->ls_open)) {
1177 					stp->ls_noopens++;
1178 					if (stp->ls_noopens > NFSNOOPEN ||
1179 					    (nfsrv_openpluslock * 2) >
1180 					    nfsrv_v4statelimit)
1181 						nfsrv_stablefirst.nsf_flags |=
1182 							NFSNSF_NOOPENS;
1183 				} else {
1184 					stp->ls_noopens = 0;
1185 				}
1186 				stp = nstp;
1187 			}
1188 		    }
1189 		}
1190 		clp = nclp;
1191 	    }
1192 	}
1193 	NFSUNLOCKSTATE();
1194 	NFSLOCKV4ROOTMUTEX();
1195 	nfsv4_relref(&nfsv4rootfs_lock);
1196 	NFSUNLOCKV4ROOTMUTEX();
1197 }
1198 
1199 /*
1200  * The following set of functions free up the various data structures.
1201  */
1202 /*
1203  * Clear out all open/lock state related to this nfsclient.
1204  * Caller must hold an exclusive lock on nfsv4rootfs_lock, so that
1205  * there are no other active nfsd threads.
1206  */
1207 APPLESTATIC void
1208 nfsrv_cleanclient(struct nfsclient *clp, NFSPROC_T *p)
1209 {
1210 	struct nfsstate *stp, *nstp;
1211 	struct nfsdsession *sep, *nsep;
1212 
1213 	LIST_FOREACH_SAFE(stp, &clp->lc_open, ls_list, nstp)
1214 		nfsrv_freeopenowner(stp, 1, p);
1215 	if ((clp->lc_flags & LCL_ADMINREVOKED) == 0)
1216 		LIST_FOREACH_SAFE(sep, &clp->lc_session, sess_list, nsep)
1217 			(void)nfsrv_freesession(sep, NULL);
1218 }
1219 
1220 /*
1221  * Free a client that has been cleaned. It should also already have been
1222  * removed from the lists.
1223  * (Just to be safe w.r.t. newnfs_disconnect(), call this function when
1224  *  softclock interrupts are enabled.)
1225  */
1226 APPLESTATIC void
1227 nfsrv_zapclient(struct nfsclient *clp, NFSPROC_T *p)
1228 {
1229 
1230 #ifdef notyet
1231 	if ((clp->lc_flags & (LCL_GSS | LCL_CALLBACKSON)) ==
1232 	     (LCL_GSS | LCL_CALLBACKSON) &&
1233 	    (clp->lc_hand.nfsh_flag & NFSG_COMPLETE) &&
1234 	    clp->lc_handlelen > 0) {
1235 		clp->lc_hand.nfsh_flag &= ~NFSG_COMPLETE;
1236 		clp->lc_hand.nfsh_flag |= NFSG_DESTROYED;
1237 		(void) nfsrv_docallback(clp, NFSV4PROC_CBNULL,
1238 			NULL, 0, NULL, NULL, NULL, p);
1239 	}
1240 #endif
1241 	newnfs_disconnect(&clp->lc_req);
1242 	free(clp->lc_req.nr_nam, M_SONAME);
1243 	NFSFREEMUTEX(&clp->lc_req.nr_mtx);
1244 	free(clp->lc_stateid, M_NFSDCLIENT);
1245 	free(clp, M_NFSDCLIENT);
1246 	NFSLOCKSTATE();
1247 	nfsstatsv1.srvclients--;
1248 	nfsrv_openpluslock--;
1249 	nfsrv_clients--;
1250 	NFSUNLOCKSTATE();
1251 }
1252 
1253 /*
1254  * Free a list of delegation state structures.
1255  * (This function will also free all nfslockfile structures that no
1256  *  longer have associated state.)
1257  */
1258 APPLESTATIC void
1259 nfsrv_freedeleglist(struct nfsstatehead *sthp)
1260 {
1261 	struct nfsstate *stp, *nstp;
1262 
1263 	LIST_FOREACH_SAFE(stp, sthp, ls_list, nstp) {
1264 		nfsrv_freedeleg(stp);
1265 	}
1266 	LIST_INIT(sthp);
1267 }
1268 
1269 /*
1270  * Free up a delegation.
1271  */
1272 static void
1273 nfsrv_freedeleg(struct nfsstate *stp)
1274 {
1275 	struct nfslockfile *lfp;
1276 
1277 	LIST_REMOVE(stp, ls_hash);
1278 	LIST_REMOVE(stp, ls_list);
1279 	LIST_REMOVE(stp, ls_file);
1280 	if ((stp->ls_flags & NFSLCK_DELEGWRITE) != 0)
1281 		nfsrv_writedelegcnt--;
1282 	lfp = stp->ls_lfp;
1283 	if (LIST_EMPTY(&lfp->lf_open) &&
1284 	    LIST_EMPTY(&lfp->lf_lock) && LIST_EMPTY(&lfp->lf_deleg) &&
1285 	    LIST_EMPTY(&lfp->lf_locallock) && LIST_EMPTY(&lfp->lf_rollback) &&
1286 	    lfp->lf_usecount == 0 &&
1287 	    nfsv4_testlock(&lfp->lf_locallock_lck) == 0)
1288 		nfsrv_freenfslockfile(lfp);
1289 	free(stp, M_NFSDSTATE);
1290 	nfsstatsv1.srvdelegates--;
1291 	nfsrv_openpluslock--;
1292 	nfsrv_delegatecnt--;
1293 }
1294 
1295 /*
1296  * This function frees an open owner and all associated opens.
1297  */
1298 static void
1299 nfsrv_freeopenowner(struct nfsstate *stp, int cansleep, NFSPROC_T *p)
1300 {
1301 	struct nfsstate *nstp, *tstp;
1302 
1303 	LIST_REMOVE(stp, ls_list);
1304 	/*
1305 	 * Now, free all associated opens.
1306 	 */
1307 	nstp = LIST_FIRST(&stp->ls_open);
1308 	while (nstp != LIST_END(&stp->ls_open)) {
1309 		tstp = nstp;
1310 		nstp = LIST_NEXT(nstp, ls_list);
1311 		(void) nfsrv_freeopen(tstp, NULL, cansleep, p);
1312 	}
1313 	if (stp->ls_op)
1314 		nfsrvd_derefcache(stp->ls_op);
1315 	free(stp, M_NFSDSTATE);
1316 	nfsstatsv1.srvopenowners--;
1317 	nfsrv_openpluslock--;
1318 }
1319 
1320 /*
1321  * This function frees an open (nfsstate open structure) with all associated
1322  * lock_owners and locks. It also frees the nfslockfile structure iff there
1323  * are no other opens on the file.
1324  * Returns 1 if it free'd the nfslockfile, 0 otherwise.
1325  */
1326 static int
1327 nfsrv_freeopen(struct nfsstate *stp, vnode_t vp, int cansleep, NFSPROC_T *p)
1328 {
1329 	struct nfsstate *nstp, *tstp;
1330 	struct nfslockfile *lfp;
1331 	int ret;
1332 
1333 	LIST_REMOVE(stp, ls_hash);
1334 	LIST_REMOVE(stp, ls_list);
1335 	LIST_REMOVE(stp, ls_file);
1336 
1337 	lfp = stp->ls_lfp;
1338 	/*
1339 	 * Now, free all lockowners associated with this open.
1340 	 */
1341 	LIST_FOREACH_SAFE(tstp, &stp->ls_open, ls_list, nstp)
1342 		nfsrv_freelockowner(tstp, vp, cansleep, p);
1343 
1344 	/*
1345 	 * The nfslockfile is freed here if there are no locks
1346 	 * associated with the open.
1347 	 * If there are locks associated with the open, the
1348 	 * nfslockfile structure can be freed via nfsrv_freelockowner().
1349 	 * Acquire the state mutex to avoid races with calls to
1350 	 * nfsrv_getlockfile().
1351 	 */
1352 	if (cansleep != 0)
1353 		NFSLOCKSTATE();
1354 	if (lfp != NULL && LIST_EMPTY(&lfp->lf_open) &&
1355 	    LIST_EMPTY(&lfp->lf_deleg) && LIST_EMPTY(&lfp->lf_lock) &&
1356 	    LIST_EMPTY(&lfp->lf_locallock) && LIST_EMPTY(&lfp->lf_rollback) &&
1357 	    lfp->lf_usecount == 0 &&
1358 	    (cansleep != 0 || nfsv4_testlock(&lfp->lf_locallock_lck) == 0)) {
1359 		nfsrv_freenfslockfile(lfp);
1360 		ret = 1;
1361 	} else
1362 		ret = 0;
1363 	if (cansleep != 0)
1364 		NFSUNLOCKSTATE();
1365 	free(stp, M_NFSDSTATE);
1366 	nfsstatsv1.srvopens--;
1367 	nfsrv_openpluslock--;
1368 	return (ret);
1369 }
1370 
1371 /*
1372  * Frees a lockowner and all associated locks.
1373  */
1374 static void
1375 nfsrv_freelockowner(struct nfsstate *stp, vnode_t vp, int cansleep,
1376     NFSPROC_T *p)
1377 {
1378 
1379 	LIST_REMOVE(stp, ls_hash);
1380 	LIST_REMOVE(stp, ls_list);
1381 	nfsrv_freeallnfslocks(stp, vp, cansleep, p);
1382 	if (stp->ls_op)
1383 		nfsrvd_derefcache(stp->ls_op);
1384 	free(stp, M_NFSDSTATE);
1385 	nfsstatsv1.srvlockowners--;
1386 	nfsrv_openpluslock--;
1387 }
1388 
1389 /*
1390  * Free all the nfs locks on a lockowner.
1391  */
1392 static void
1393 nfsrv_freeallnfslocks(struct nfsstate *stp, vnode_t vp, int cansleep,
1394     NFSPROC_T *p)
1395 {
1396 	struct nfslock *lop, *nlop;
1397 	struct nfsrollback *rlp, *nrlp;
1398 	struct nfslockfile *lfp = NULL;
1399 	int gottvp = 0;
1400 	vnode_t tvp = NULL;
1401 	uint64_t first, end;
1402 
1403 	if (vp != NULL)
1404 		ASSERT_VOP_UNLOCKED(vp, "nfsrv_freeallnfslocks: vnode locked");
1405 	lop = LIST_FIRST(&stp->ls_lock);
1406 	while (lop != LIST_END(&stp->ls_lock)) {
1407 		nlop = LIST_NEXT(lop, lo_lckowner);
1408 		/*
1409 		 * Since all locks should be for the same file, lfp should
1410 		 * not change.
1411 		 */
1412 		if (lfp == NULL)
1413 			lfp = lop->lo_lfp;
1414 		else if (lfp != lop->lo_lfp)
1415 			panic("allnfslocks");
1416 		/*
1417 		 * If vp is NULL and cansleep != 0, a vnode must be acquired
1418 		 * from the file handle. This only occurs when called from
1419 		 * nfsrv_cleanclient().
1420 		 */
1421 		if (gottvp == 0) {
1422 			if (nfsrv_dolocallocks == 0)
1423 				tvp = NULL;
1424 			else if (vp == NULL && cansleep != 0) {
1425 				tvp = nfsvno_getvp(&lfp->lf_fh);
1426 				NFSVOPUNLOCK(tvp, 0);
1427 			} else
1428 				tvp = vp;
1429 			gottvp = 1;
1430 		}
1431 
1432 		if (tvp != NULL) {
1433 			if (cansleep == 0)
1434 				panic("allnfs2");
1435 			first = lop->lo_first;
1436 			end = lop->lo_end;
1437 			nfsrv_freenfslock(lop);
1438 			nfsrv_localunlock(tvp, lfp, first, end, p);
1439 			LIST_FOREACH_SAFE(rlp, &lfp->lf_rollback, rlck_list,
1440 			    nrlp)
1441 				free(rlp, M_NFSDROLLBACK);
1442 			LIST_INIT(&lfp->lf_rollback);
1443 		} else
1444 			nfsrv_freenfslock(lop);
1445 		lop = nlop;
1446 	}
1447 	if (vp == NULL && tvp != NULL)
1448 		vrele(tvp);
1449 }
1450 
1451 /*
1452  * Free an nfslock structure.
1453  */
1454 static void
1455 nfsrv_freenfslock(struct nfslock *lop)
1456 {
1457 
1458 	if (lop->lo_lckfile.le_prev != NULL) {
1459 		LIST_REMOVE(lop, lo_lckfile);
1460 		nfsstatsv1.srvlocks--;
1461 		nfsrv_openpluslock--;
1462 	}
1463 	LIST_REMOVE(lop, lo_lckowner);
1464 	free(lop, M_NFSDLOCK);
1465 }
1466 
1467 /*
1468  * This function frees an nfslockfile structure.
1469  */
1470 static void
1471 nfsrv_freenfslockfile(struct nfslockfile *lfp)
1472 {
1473 
1474 	LIST_REMOVE(lfp, lf_hash);
1475 	free(lfp, M_NFSDLOCKFILE);
1476 }
1477 
1478 /*
1479  * This function looks up an nfsstate structure via stateid.
1480  */
1481 static int
1482 nfsrv_getstate(struct nfsclient *clp, nfsv4stateid_t *stateidp, __unused u_int32_t flags,
1483     struct nfsstate **stpp)
1484 {
1485 	struct nfsstate *stp;
1486 	struct nfsstatehead *hp;
1487 	int error = 0;
1488 
1489 	*stpp = NULL;
1490 	hp = NFSSTATEHASH(clp, *stateidp);
1491 	LIST_FOREACH(stp, hp, ls_hash) {
1492 		if (!NFSBCMP(stp->ls_stateid.other, stateidp->other,
1493 			NFSX_STATEIDOTHER))
1494 			break;
1495 	}
1496 
1497 	/*
1498 	 * If no state id in list, return NFSERR_BADSTATEID.
1499 	 */
1500 	if (stp == LIST_END(hp)) {
1501 		error = NFSERR_BADSTATEID;
1502 		goto out;
1503 	}
1504 	*stpp = stp;
1505 
1506 out:
1507 	NFSEXITCODE(error);
1508 	return (error);
1509 }
1510 
1511 /*
1512  * This function gets an nfsstate structure via owner string.
1513  */
1514 static void
1515 nfsrv_getowner(struct nfsstatehead *hp, struct nfsstate *new_stp,
1516     struct nfsstate **stpp)
1517 {
1518 	struct nfsstate *stp;
1519 
1520 	*stpp = NULL;
1521 	LIST_FOREACH(stp, hp, ls_list) {
1522 		if (new_stp->ls_ownerlen == stp->ls_ownerlen &&
1523 		  !NFSBCMP(new_stp->ls_owner,stp->ls_owner,stp->ls_ownerlen)) {
1524 			*stpp = stp;
1525 			return;
1526 		}
1527 	}
1528 }
1529 
1530 /*
1531  * Lock control function called to update lock status.
1532  * Returns 0 upon success, -1 if there is no lock and the flags indicate
1533  * that one isn't to be created and an NFSERR_xxx for other errors.
1534  * The structures new_stp and new_lop are passed in as pointers that should
1535  * be set to NULL if the structure is used and shouldn't be free'd.
1536  * For the NFSLCK_TEST and NFSLCK_CHECK cases, the structures are
1537  * never used and can safely be allocated on the stack. For all other
1538  * cases, *new_stpp and *new_lopp should be malloc'd before the call,
1539  * in case they are used.
1540  */
1541 APPLESTATIC int
1542 nfsrv_lockctrl(vnode_t vp, struct nfsstate **new_stpp,
1543     struct nfslock **new_lopp, struct nfslockconflict *cfp,
1544     nfsquad_t clientid, nfsv4stateid_t *stateidp,
1545     __unused struct nfsexstuff *exp,
1546     struct nfsrv_descript *nd, NFSPROC_T *p)
1547 {
1548 	struct nfslock *lop;
1549 	struct nfsstate *new_stp = *new_stpp;
1550 	struct nfslock *new_lop = *new_lopp;
1551 	struct nfsstate *tstp, *mystp, *nstp;
1552 	int specialid = 0;
1553 	struct nfslockfile *lfp;
1554 	struct nfslock *other_lop = NULL;
1555 	struct nfsstate *stp, *lckstp = NULL;
1556 	struct nfsclient *clp = NULL;
1557 	u_int32_t bits;
1558 	int error = 0, haslock = 0, ret, reterr;
1559 	int getlckret, delegation = 0, filestruct_locked, vnode_unlocked = 0;
1560 	fhandle_t nfh;
1561 	uint64_t first, end;
1562 	uint32_t lock_flags;
1563 
1564 	if (new_stp->ls_flags & (NFSLCK_CHECK | NFSLCK_SETATTR)) {
1565 		/*
1566 		 * Note the special cases of "all 1s" or "all 0s" stateids and
1567 		 * let reads with all 1s go ahead.
1568 		 */
1569 		if (new_stp->ls_stateid.seqid == 0x0 &&
1570 		    new_stp->ls_stateid.other[0] == 0x0 &&
1571 		    new_stp->ls_stateid.other[1] == 0x0 &&
1572 		    new_stp->ls_stateid.other[2] == 0x0)
1573 			specialid = 1;
1574 		else if (new_stp->ls_stateid.seqid == 0xffffffff &&
1575 		    new_stp->ls_stateid.other[0] == 0xffffffff &&
1576 		    new_stp->ls_stateid.other[1] == 0xffffffff &&
1577 		    new_stp->ls_stateid.other[2] == 0xffffffff)
1578 			specialid = 2;
1579 	}
1580 
1581 	/*
1582 	 * Check for restart conditions (client and server).
1583 	 */
1584 	error = nfsrv_checkrestart(clientid, new_stp->ls_flags,
1585 	    &new_stp->ls_stateid, specialid);
1586 	if (error)
1587 		goto out;
1588 
1589 	/*
1590 	 * Check for state resource limit exceeded.
1591 	 */
1592 	if ((new_stp->ls_flags & NFSLCK_LOCK) &&
1593 	    nfsrv_openpluslock > nfsrv_v4statelimit) {
1594 		error = NFSERR_RESOURCE;
1595 		goto out;
1596 	}
1597 
1598 	/*
1599 	 * For the lock case, get another nfslock structure,
1600 	 * just in case we need it.
1601 	 * Malloc now, before we start sifting through the linked lists,
1602 	 * in case we have to wait for memory.
1603 	 */
1604 tryagain:
1605 	if (new_stp->ls_flags & NFSLCK_LOCK)
1606 		other_lop = malloc(sizeof (struct nfslock),
1607 		    M_NFSDLOCK, M_WAITOK);
1608 	filestruct_locked = 0;
1609 	reterr = 0;
1610 	lfp = NULL;
1611 
1612 	/*
1613 	 * Get the lockfile structure for CFH now, so we can do a sanity
1614 	 * check against the stateid, before incrementing the seqid#, since
1615 	 * we want to return NFSERR_BADSTATEID on failure and the seqid#
1616 	 * shouldn't be incremented for this case.
1617 	 * If nfsrv_getlockfile() returns -1, it means "not found", which
1618 	 * will be handled later.
1619 	 * If we are doing Lock/LockU and local locking is enabled, sleep
1620 	 * lock the nfslockfile structure.
1621 	 */
1622 	getlckret = nfsrv_getlockfh(vp, new_stp->ls_flags, NULL, &nfh, p);
1623 	NFSLOCKSTATE();
1624 	if (getlckret == 0) {
1625 		if ((new_stp->ls_flags & (NFSLCK_LOCK | NFSLCK_UNLOCK)) != 0 &&
1626 		    nfsrv_dolocallocks != 0 && nd->nd_repstat == 0) {
1627 			getlckret = nfsrv_getlockfile(new_stp->ls_flags, NULL,
1628 			    &lfp, &nfh, 1);
1629 			if (getlckret == 0)
1630 				filestruct_locked = 1;
1631 		} else
1632 			getlckret = nfsrv_getlockfile(new_stp->ls_flags, NULL,
1633 			    &lfp, &nfh, 0);
1634 	}
1635 	if (getlckret != 0 && getlckret != -1)
1636 		reterr = getlckret;
1637 
1638 	if (filestruct_locked != 0) {
1639 		LIST_INIT(&lfp->lf_rollback);
1640 		if ((new_stp->ls_flags & NFSLCK_LOCK)) {
1641 			/*
1642 			 * For local locking, do the advisory locking now, so
1643 			 * that any conflict can be detected. A failure later
1644 			 * can be rolled back locally. If an error is returned,
1645 			 * struct nfslockfile has been unlocked and any local
1646 			 * locking rolled back.
1647 			 */
1648 			NFSUNLOCKSTATE();
1649 			if (vnode_unlocked == 0) {
1650 				ASSERT_VOP_ELOCKED(vp, "nfsrv_lockctrl1");
1651 				vnode_unlocked = 1;
1652 				NFSVOPUNLOCK(vp, 0);
1653 			}
1654 			reterr = nfsrv_locallock(vp, lfp,
1655 			    (new_lop->lo_flags & (NFSLCK_READ | NFSLCK_WRITE)),
1656 			    new_lop->lo_first, new_lop->lo_end, cfp, p);
1657 			NFSLOCKSTATE();
1658 		}
1659 	}
1660 
1661 	if (specialid == 0) {
1662 	    if (new_stp->ls_flags & NFSLCK_TEST) {
1663 		/*
1664 		 * RFC 3530 does not list LockT as an op that renews a
1665 		 * lease, but the consensus seems to be that it is ok
1666 		 * for a server to do so.
1667 		 */
1668 		error = nfsrv_getclient(clientid, CLOPS_RENEW, &clp, NULL,
1669 		    (nfsquad_t)((u_quad_t)0), 0, nd, p);
1670 
1671 		/*
1672 		 * Since NFSERR_EXPIRED, NFSERR_ADMINREVOKED are not valid
1673 		 * error returns for LockT, just go ahead and test for a lock,
1674 		 * since there are no locks for this client, but other locks
1675 		 * can conflict. (ie. same client will always be false)
1676 		 */
1677 		if (error == NFSERR_EXPIRED || error == NFSERR_ADMINREVOKED)
1678 		    error = 0;
1679 		lckstp = new_stp;
1680 	    } else {
1681 	      error = nfsrv_getclient(clientid, CLOPS_RENEW, &clp, NULL,
1682 		(nfsquad_t)((u_quad_t)0), 0, nd, p);
1683 	      if (error == 0)
1684 		/*
1685 		 * Look up the stateid
1686 		 */
1687 		error = nfsrv_getstate(clp, &new_stp->ls_stateid,
1688 		  new_stp->ls_flags, &stp);
1689 	      /*
1690 	       * do some sanity checks for an unconfirmed open or a
1691 	       * stateid that refers to the wrong file, for an open stateid
1692 	       */
1693 	      if (error == 0 && (stp->ls_flags & NFSLCK_OPEN) &&
1694 		  ((stp->ls_openowner->ls_flags & NFSLCK_NEEDSCONFIRM) ||
1695 		   (getlckret == 0 && stp->ls_lfp != lfp))){
1696 		      /*
1697 		       * NFSLCK_SETATTR should return OK rather than NFSERR_BADSTATEID
1698 		       * The only exception is using SETATTR with SIZE.
1699 		       * */
1700                     if ((new_stp->ls_flags &
1701                          (NFSLCK_SETATTR | NFSLCK_CHECK)) != NFSLCK_SETATTR)
1702 			     error = NFSERR_BADSTATEID;
1703 	      }
1704 
1705 		if (error == 0 &&
1706 		  (stp->ls_flags & (NFSLCK_DELEGREAD | NFSLCK_DELEGWRITE)) &&
1707 		  getlckret == 0 && stp->ls_lfp != lfp)
1708 			error = NFSERR_BADSTATEID;
1709 
1710 	      /*
1711 	       * If the lockowner stateid doesn't refer to the same file,
1712 	       * I believe that is considered ok, since some clients will
1713 	       * only create a single lockowner and use that for all locks
1714 	       * on all files.
1715 	       * For now, log it as a diagnostic, instead of considering it
1716 	       * a BadStateid.
1717 	       */
1718 	      if (error == 0 && (stp->ls_flags &
1719 		  (NFSLCK_OPEN | NFSLCK_DELEGREAD | NFSLCK_DELEGWRITE)) == 0 &&
1720 		  getlckret == 0 && stp->ls_lfp != lfp) {
1721 #ifdef DIAGNOSTIC
1722 		  printf("Got a lock statid for different file open\n");
1723 #endif
1724 		  /*
1725 		  error = NFSERR_BADSTATEID;
1726 		  */
1727 	      }
1728 
1729 	      if (error == 0) {
1730 		    if (new_stp->ls_flags & NFSLCK_OPENTOLOCK) {
1731 			/*
1732 			 * If haslock set, we've already checked the seqid.
1733 			 */
1734 			if (!haslock) {
1735 			    if (stp->ls_flags & NFSLCK_OPEN)
1736 				error = nfsrv_checkseqid(nd, new_stp->ls_seq,
1737 				    stp->ls_openowner, new_stp->ls_op);
1738 			    else
1739 				error = NFSERR_BADSTATEID;
1740 			}
1741 			if (!error)
1742 			    nfsrv_getowner(&stp->ls_open, new_stp, &lckstp);
1743 			if (lckstp)
1744 			    /*
1745 			     * I believe this should be an error, but it
1746 			     * isn't obvious what NFSERR_xxx would be
1747 			     * appropriate, so I'll use NFSERR_INVAL for now.
1748 			     */
1749 			    error = NFSERR_INVAL;
1750 			else
1751 			    lckstp = new_stp;
1752 		    } else if (new_stp->ls_flags&(NFSLCK_LOCK|NFSLCK_UNLOCK)) {
1753 			/*
1754 			 * If haslock set, ditto above.
1755 			 */
1756 			if (!haslock) {
1757 			    if (stp->ls_flags & NFSLCK_OPEN)
1758 				error = NFSERR_BADSTATEID;
1759 			    else
1760 				error = nfsrv_checkseqid(nd, new_stp->ls_seq,
1761 				    stp, new_stp->ls_op);
1762 			}
1763 			lckstp = stp;
1764 		    } else {
1765 			lckstp = stp;
1766 		    }
1767 	      }
1768 	      /*
1769 	       * If the seqid part of the stateid isn't the same, return
1770 	       * NFSERR_OLDSTATEID for cases other than I/O Ops.
1771 	       * For I/O Ops, only return NFSERR_OLDSTATEID if
1772 	       * nfsrv_returnoldstateid is set. (The consensus on the email
1773 	       * list was that most clients would prefer to not receive
1774 	       * NFSERR_OLDSTATEID for I/O Ops, but the RFC suggests that that
1775 	       * is what will happen, so I use the nfsrv_returnoldstateid to
1776 	       * allow for either server configuration.)
1777 	       */
1778 	      if (!error && stp->ls_stateid.seqid!=new_stp->ls_stateid.seqid &&
1779 		  (((nd->nd_flag & ND_NFSV41) == 0 &&
1780 		   (!(new_stp->ls_flags & NFSLCK_CHECK) ||
1781 		    nfsrv_returnoldstateid)) ||
1782 		   ((nd->nd_flag & ND_NFSV41) != 0 &&
1783 		    new_stp->ls_stateid.seqid != 0)))
1784 		    error = NFSERR_OLDSTATEID;
1785 	    }
1786 	}
1787 
1788 	/*
1789 	 * Now we can check for grace.
1790 	 */
1791 	if (!error)
1792 		error = nfsrv_checkgrace(nd, clp, new_stp->ls_flags);
1793 	if ((new_stp->ls_flags & NFSLCK_RECLAIM) && !error &&
1794 		nfsrv_checkstable(clp))
1795 		error = NFSERR_NOGRACE;
1796 	/*
1797 	 * If we successfully Reclaimed state, note that.
1798 	 */
1799 	if ((new_stp->ls_flags & NFSLCK_RECLAIM) && !error)
1800 		nfsrv_markstable(clp);
1801 
1802 	/*
1803 	 * At this point, either error == NFSERR_BADSTATEID or the
1804 	 * seqid# has been updated, so we can return any error.
1805 	 * If error == 0, there may be an error in:
1806 	 *    nd_repstat - Set by the calling function.
1807 	 *    reterr - Set above, if getting the nfslockfile structure
1808 	 *       or acquiring the local lock failed.
1809 	 *    (If both of these are set, nd_repstat should probably be
1810 	 *     returned, since that error was detected before this
1811 	 *     function call.)
1812 	 */
1813 	if (error != 0 || nd->nd_repstat != 0 || reterr != 0) {
1814 		if (error == 0) {
1815 			if (nd->nd_repstat != 0)
1816 				error = nd->nd_repstat;
1817 			else
1818 				error = reterr;
1819 		}
1820 		if (filestruct_locked != 0) {
1821 			/* Roll back local locks. */
1822 			NFSUNLOCKSTATE();
1823 			if (vnode_unlocked == 0) {
1824 				ASSERT_VOP_ELOCKED(vp, "nfsrv_lockctrl2");
1825 				vnode_unlocked = 1;
1826 				NFSVOPUNLOCK(vp, 0);
1827 			}
1828 			nfsrv_locallock_rollback(vp, lfp, p);
1829 			NFSLOCKSTATE();
1830 			nfsrv_unlocklf(lfp);
1831 		}
1832 		NFSUNLOCKSTATE();
1833 		goto out;
1834 	}
1835 
1836 	/*
1837 	 * Check the nfsrv_getlockfile return.
1838 	 * Returned -1 if no structure found.
1839 	 */
1840 	if (getlckret == -1) {
1841 		error = NFSERR_EXPIRED;
1842 		/*
1843 		 * Called from lockt, so no lock is OK.
1844 		 */
1845 		if (new_stp->ls_flags & NFSLCK_TEST) {
1846 			error = 0;
1847 		} else if (new_stp->ls_flags &
1848 		    (NFSLCK_CHECK | NFSLCK_SETATTR)) {
1849 			/*
1850 			 * Called to check for a lock, OK if the stateid is all
1851 			 * 1s or all 0s, but there should be an nfsstate
1852 			 * otherwise.
1853 			 * (ie. If there is no open, I'll assume no share
1854 			 *  deny bits.)
1855 			 */
1856 			if (specialid)
1857 				error = 0;
1858 			else
1859 				error = NFSERR_BADSTATEID;
1860 		}
1861 		NFSUNLOCKSTATE();
1862 		goto out;
1863 	}
1864 
1865 	/*
1866 	 * For NFSLCK_CHECK and NFSLCK_LOCK, test for a share conflict.
1867 	 * For NFSLCK_CHECK, allow a read if write access is granted,
1868 	 * but check for a deny. For NFSLCK_LOCK, require correct access,
1869 	 * which implies a conflicting deny can't exist.
1870 	 */
1871 	if (new_stp->ls_flags & (NFSLCK_CHECK | NFSLCK_LOCK)) {
1872 	    /*
1873 	     * Four kinds of state id:
1874 	     * - specialid (all 0s or all 1s), only for NFSLCK_CHECK
1875 	     * - stateid for an open
1876 	     * - stateid for a delegation
1877 	     * - stateid for a lock owner
1878 	     */
1879 	    if (!specialid) {
1880 		if (stp->ls_flags & (NFSLCK_DELEGREAD | NFSLCK_DELEGWRITE)) {
1881 		    delegation = 1;
1882 		    mystp = stp;
1883 		    nfsrv_delaydelegtimeout(stp);
1884 	        } else if (stp->ls_flags & NFSLCK_OPEN) {
1885 		    mystp = stp;
1886 		} else {
1887 		    mystp = stp->ls_openstp;
1888 		}
1889 		/*
1890 		 * If locking or checking, require correct access
1891 		 * bit set.
1892 		 */
1893 		if (((new_stp->ls_flags & NFSLCK_LOCK) &&
1894 		     !((new_lop->lo_flags >> NFSLCK_LOCKSHIFT) &
1895 		       mystp->ls_flags & NFSLCK_ACCESSBITS)) ||
1896 		    ((new_stp->ls_flags & (NFSLCK_CHECK|NFSLCK_READACCESS)) ==
1897 		      (NFSLCK_CHECK | NFSLCK_READACCESS) &&
1898 		     !(mystp->ls_flags & NFSLCK_READACCESS) &&
1899 		     nfsrv_allowreadforwriteopen == 0) ||
1900 		    ((new_stp->ls_flags & (NFSLCK_CHECK|NFSLCK_WRITEACCESS)) ==
1901 		      (NFSLCK_CHECK | NFSLCK_WRITEACCESS) &&
1902 		     !(mystp->ls_flags & NFSLCK_WRITEACCESS))) {
1903 			if (filestruct_locked != 0) {
1904 				/* Roll back local locks. */
1905 				NFSUNLOCKSTATE();
1906 				if (vnode_unlocked == 0) {
1907 					ASSERT_VOP_ELOCKED(vp,
1908 					    "nfsrv_lockctrl3");
1909 					vnode_unlocked = 1;
1910 					NFSVOPUNLOCK(vp, 0);
1911 				}
1912 				nfsrv_locallock_rollback(vp, lfp, p);
1913 				NFSLOCKSTATE();
1914 				nfsrv_unlocklf(lfp);
1915 			}
1916 			NFSUNLOCKSTATE();
1917 			error = NFSERR_OPENMODE;
1918 			goto out;
1919 		}
1920 	    } else
1921 		mystp = NULL;
1922 	    if ((new_stp->ls_flags & NFSLCK_CHECK) && !delegation) {
1923 		/*
1924 		 * Check for a conflicting deny bit.
1925 		 */
1926 		LIST_FOREACH(tstp, &lfp->lf_open, ls_file) {
1927 		    if (tstp != mystp) {
1928 			bits = tstp->ls_flags;
1929 			bits >>= NFSLCK_SHIFT;
1930 			if (new_stp->ls_flags & bits & NFSLCK_ACCESSBITS) {
1931 			    KASSERT(vnode_unlocked == 0,
1932 				("nfsrv_lockctrl: vnode unlocked1"));
1933 			    ret = nfsrv_clientconflict(tstp->ls_clp, &haslock,
1934 				vp, p);
1935 			    if (ret == 1) {
1936 				/*
1937 				* nfsrv_clientconflict unlocks state
1938 				 * when it returns non-zero.
1939 				 */
1940 				lckstp = NULL;
1941 				goto tryagain;
1942 			    }
1943 			    if (ret == 0)
1944 				NFSUNLOCKSTATE();
1945 			    if (ret == 2)
1946 				error = NFSERR_PERM;
1947 			    else
1948 				error = NFSERR_OPENMODE;
1949 			    goto out;
1950 			}
1951 		    }
1952 		}
1953 
1954 		/* We're outta here */
1955 		NFSUNLOCKSTATE();
1956 		goto out;
1957 	    }
1958 	}
1959 
1960 	/*
1961 	 * For setattr, just get rid of all the Delegations for other clients.
1962 	 */
1963 	if (new_stp->ls_flags & NFSLCK_SETATTR) {
1964 		KASSERT(vnode_unlocked == 0,
1965 		    ("nfsrv_lockctrl: vnode unlocked2"));
1966 		ret = nfsrv_cleandeleg(vp, lfp, clp, &haslock, p);
1967 		if (ret) {
1968 			/*
1969 			 * nfsrv_cleandeleg() unlocks state when it
1970 			 * returns non-zero.
1971 			 */
1972 			if (ret == -1) {
1973 				lckstp = NULL;
1974 				goto tryagain;
1975 			}
1976 			error = ret;
1977 			goto out;
1978 		}
1979 		if (!(new_stp->ls_flags & NFSLCK_CHECK) ||
1980 		    (LIST_EMPTY(&lfp->lf_open) && LIST_EMPTY(&lfp->lf_lock) &&
1981 		     LIST_EMPTY(&lfp->lf_deleg))) {
1982 			NFSUNLOCKSTATE();
1983 			goto out;
1984 		}
1985 	}
1986 
1987 	/*
1988 	 * Check for a conflicting delegation. If one is found, call
1989 	 * nfsrv_delegconflict() to handle it. If the v4root lock hasn't
1990 	 * been set yet, it will get the lock. Otherwise, it will recall
1991 	 * the delegation. Then, we try try again...
1992 	 * I currently believe the conflict algorithm to be:
1993 	 * For Lock Ops (Lock/LockT/LockU)
1994 	 * - there is a conflict iff a different client has a write delegation
1995 	 * For Reading (Read Op)
1996 	 * - there is a conflict iff a different client has a write delegation
1997 	 *   (the specialids are always a different client)
1998 	 * For Writing (Write/Setattr of size)
1999 	 * - there is a conflict if a different client has any delegation
2000 	 * - there is a conflict if the same client has a read delegation
2001 	 *   (I don't understand why this isn't allowed, but that seems to be
2002 	 *    the current consensus?)
2003 	 */
2004 	tstp = LIST_FIRST(&lfp->lf_deleg);
2005 	while (tstp != LIST_END(&lfp->lf_deleg)) {
2006 	    nstp = LIST_NEXT(tstp, ls_file);
2007 	    if ((((new_stp->ls_flags&(NFSLCK_LOCK|NFSLCK_UNLOCK|NFSLCK_TEST))||
2008 		 ((new_stp->ls_flags & NFSLCK_CHECK) &&
2009 		  (new_lop->lo_flags & NFSLCK_READ))) &&
2010 		  clp != tstp->ls_clp &&
2011 		 (tstp->ls_flags & NFSLCK_DELEGWRITE)) ||
2012 		 ((new_stp->ls_flags & NFSLCK_CHECK) &&
2013 		   (new_lop->lo_flags & NFSLCK_WRITE) &&
2014 		  (clp != tstp->ls_clp ||
2015 		   (tstp->ls_flags & NFSLCK_DELEGREAD)))) {
2016 		ret = 0;
2017 		if (filestruct_locked != 0) {
2018 			/* Roll back local locks. */
2019 			NFSUNLOCKSTATE();
2020 			if (vnode_unlocked == 0) {
2021 				ASSERT_VOP_ELOCKED(vp, "nfsrv_lockctrl4");
2022 				NFSVOPUNLOCK(vp, 0);
2023 			}
2024 			nfsrv_locallock_rollback(vp, lfp, p);
2025 			NFSLOCKSTATE();
2026 			nfsrv_unlocklf(lfp);
2027 			NFSUNLOCKSTATE();
2028 			NFSVOPLOCK(vp, LK_EXCLUSIVE | LK_RETRY);
2029 			vnode_unlocked = 0;
2030 			if ((vp->v_iflag & VI_DOOMED) != 0)
2031 				ret = NFSERR_SERVERFAULT;
2032 			NFSLOCKSTATE();
2033 		}
2034 		if (ret == 0)
2035 			ret = nfsrv_delegconflict(tstp, &haslock, p, vp);
2036 		if (ret) {
2037 		    /*
2038 		     * nfsrv_delegconflict unlocks state when it
2039 		     * returns non-zero, which it always does.
2040 		     */
2041 		    if (other_lop) {
2042 			free(other_lop, M_NFSDLOCK);
2043 			other_lop = NULL;
2044 		    }
2045 		    if (ret == -1) {
2046 			lckstp = NULL;
2047 			goto tryagain;
2048 		    }
2049 		    error = ret;
2050 		    goto out;
2051 		}
2052 		/* Never gets here. */
2053 	    }
2054 	    tstp = nstp;
2055 	}
2056 
2057 	/*
2058 	 * Handle the unlock case by calling nfsrv_updatelock().
2059 	 * (Should I have done some access checking above for unlock? For now,
2060 	 *  just let it happen.)
2061 	 */
2062 	if (new_stp->ls_flags & NFSLCK_UNLOCK) {
2063 		first = new_lop->lo_first;
2064 		end = new_lop->lo_end;
2065 		nfsrv_updatelock(stp, new_lopp, &other_lop, lfp);
2066 		stateidp->seqid = ++(stp->ls_stateid.seqid);
2067 		if ((nd->nd_flag & ND_NFSV41) != 0 && stateidp->seqid == 0)
2068 			stateidp->seqid = stp->ls_stateid.seqid = 1;
2069 		stateidp->other[0] = stp->ls_stateid.other[0];
2070 		stateidp->other[1] = stp->ls_stateid.other[1];
2071 		stateidp->other[2] = stp->ls_stateid.other[2];
2072 		if (filestruct_locked != 0) {
2073 			NFSUNLOCKSTATE();
2074 			if (vnode_unlocked == 0) {
2075 				ASSERT_VOP_ELOCKED(vp, "nfsrv_lockctrl5");
2076 				vnode_unlocked = 1;
2077 				NFSVOPUNLOCK(vp, 0);
2078 			}
2079 			/* Update the local locks. */
2080 			nfsrv_localunlock(vp, lfp, first, end, p);
2081 			NFSLOCKSTATE();
2082 			nfsrv_unlocklf(lfp);
2083 		}
2084 		NFSUNLOCKSTATE();
2085 		goto out;
2086 	}
2087 
2088 	/*
2089 	 * Search for a conflicting lock. A lock conflicts if:
2090 	 * - the lock range overlaps and
2091 	 * - at least one lock is a write lock and
2092 	 * - it is not owned by the same lock owner
2093 	 */
2094 	if (!delegation) {
2095 	  LIST_FOREACH(lop, &lfp->lf_lock, lo_lckfile) {
2096 	    if (new_lop->lo_end > lop->lo_first &&
2097 		new_lop->lo_first < lop->lo_end &&
2098 		(new_lop->lo_flags == NFSLCK_WRITE ||
2099 		 lop->lo_flags == NFSLCK_WRITE) &&
2100 		lckstp != lop->lo_stp &&
2101 		(clp != lop->lo_stp->ls_clp ||
2102 		 lckstp->ls_ownerlen != lop->lo_stp->ls_ownerlen ||
2103 		 NFSBCMP(lckstp->ls_owner, lop->lo_stp->ls_owner,
2104 		    lckstp->ls_ownerlen))) {
2105 		if (other_lop) {
2106 		    free(other_lop, M_NFSDLOCK);
2107 		    other_lop = NULL;
2108 		}
2109 		if (vnode_unlocked != 0)
2110 		    ret = nfsrv_clientconflict(lop->lo_stp->ls_clp, &haslock,
2111 			NULL, p);
2112 		else
2113 		    ret = nfsrv_clientconflict(lop->lo_stp->ls_clp, &haslock,
2114 			vp, p);
2115 		if (ret == 1) {
2116 		    if (filestruct_locked != 0) {
2117 			if (vnode_unlocked == 0) {
2118 				ASSERT_VOP_ELOCKED(vp, "nfsrv_lockctrl6");
2119 				NFSVOPUNLOCK(vp, 0);
2120 			}
2121 			/* Roll back local locks. */
2122 			nfsrv_locallock_rollback(vp, lfp, p);
2123 			NFSLOCKSTATE();
2124 			nfsrv_unlocklf(lfp);
2125 			NFSUNLOCKSTATE();
2126 			NFSVOPLOCK(vp, LK_EXCLUSIVE | LK_RETRY);
2127 			vnode_unlocked = 0;
2128 			if ((vp->v_iflag & VI_DOOMED) != 0) {
2129 				error = NFSERR_SERVERFAULT;
2130 				goto out;
2131 			}
2132 		    }
2133 		    /*
2134 		     * nfsrv_clientconflict() unlocks state when it
2135 		     * returns non-zero.
2136 		     */
2137 		    lckstp = NULL;
2138 		    goto tryagain;
2139 		}
2140 		/*
2141 		 * Found a conflicting lock, so record the conflict and
2142 		 * return the error.
2143 		 */
2144 		if (cfp != NULL && ret == 0) {
2145 		    cfp->cl_clientid.lval[0]=lop->lo_stp->ls_stateid.other[0];
2146 		    cfp->cl_clientid.lval[1]=lop->lo_stp->ls_stateid.other[1];
2147 		    cfp->cl_first = lop->lo_first;
2148 		    cfp->cl_end = lop->lo_end;
2149 		    cfp->cl_flags = lop->lo_flags;
2150 		    cfp->cl_ownerlen = lop->lo_stp->ls_ownerlen;
2151 		    NFSBCOPY(lop->lo_stp->ls_owner, cfp->cl_owner,
2152 			cfp->cl_ownerlen);
2153 		}
2154 		if (ret == 2)
2155 		    error = NFSERR_PERM;
2156 		else if (new_stp->ls_flags & NFSLCK_RECLAIM)
2157 		    error = NFSERR_RECLAIMCONFLICT;
2158 		else if (new_stp->ls_flags & NFSLCK_CHECK)
2159 		    error = NFSERR_LOCKED;
2160 		else
2161 		    error = NFSERR_DENIED;
2162 		if (filestruct_locked != 0 && ret == 0) {
2163 			/* Roll back local locks. */
2164 			NFSUNLOCKSTATE();
2165 			if (vnode_unlocked == 0) {
2166 				ASSERT_VOP_ELOCKED(vp, "nfsrv_lockctrl7");
2167 				vnode_unlocked = 1;
2168 				NFSVOPUNLOCK(vp, 0);
2169 			}
2170 			nfsrv_locallock_rollback(vp, lfp, p);
2171 			NFSLOCKSTATE();
2172 			nfsrv_unlocklf(lfp);
2173 		}
2174 		if (ret == 0)
2175 			NFSUNLOCKSTATE();
2176 		goto out;
2177 	    }
2178 	  }
2179 	}
2180 
2181 	/*
2182 	 * We only get here if there was no lock that conflicted.
2183 	 */
2184 	if (new_stp->ls_flags & (NFSLCK_TEST | NFSLCK_CHECK)) {
2185 		NFSUNLOCKSTATE();
2186 		goto out;
2187 	}
2188 
2189 	/*
2190 	 * We only get here when we are creating or modifying a lock.
2191 	 * There are two variants:
2192 	 * - exist_lock_owner where lock_owner exists
2193 	 * - open_to_lock_owner with new lock_owner
2194 	 */
2195 	first = new_lop->lo_first;
2196 	end = new_lop->lo_end;
2197 	lock_flags = new_lop->lo_flags;
2198 	if (!(new_stp->ls_flags & NFSLCK_OPENTOLOCK)) {
2199 		nfsrv_updatelock(lckstp, new_lopp, &other_lop, lfp);
2200 		stateidp->seqid = ++(lckstp->ls_stateid.seqid);
2201 		if ((nd->nd_flag & ND_NFSV41) != 0 && stateidp->seqid == 0)
2202 			stateidp->seqid = lckstp->ls_stateid.seqid = 1;
2203 		stateidp->other[0] = lckstp->ls_stateid.other[0];
2204 		stateidp->other[1] = lckstp->ls_stateid.other[1];
2205 		stateidp->other[2] = lckstp->ls_stateid.other[2];
2206 	} else {
2207 		/*
2208 		 * The new open_to_lock_owner case.
2209 		 * Link the new nfsstate into the lists.
2210 		 */
2211 		new_stp->ls_seq = new_stp->ls_opentolockseq;
2212 		nfsrvd_refcache(new_stp->ls_op);
2213 		stateidp->seqid = new_stp->ls_stateid.seqid = 1;
2214 		stateidp->other[0] = new_stp->ls_stateid.other[0] =
2215 		    clp->lc_clientid.lval[0];
2216 		stateidp->other[1] = new_stp->ls_stateid.other[1] =
2217 		    clp->lc_clientid.lval[1];
2218 		stateidp->other[2] = new_stp->ls_stateid.other[2] =
2219 		    nfsrv_nextstateindex(clp);
2220 		new_stp->ls_clp = clp;
2221 		LIST_INIT(&new_stp->ls_lock);
2222 		new_stp->ls_openstp = stp;
2223 		new_stp->ls_lfp = lfp;
2224 		nfsrv_insertlock(new_lop, (struct nfslock *)new_stp, new_stp,
2225 		    lfp);
2226 		LIST_INSERT_HEAD(NFSSTATEHASH(clp, new_stp->ls_stateid),
2227 		    new_stp, ls_hash);
2228 		LIST_INSERT_HEAD(&stp->ls_open, new_stp, ls_list);
2229 		*new_lopp = NULL;
2230 		*new_stpp = NULL;
2231 		nfsstatsv1.srvlockowners++;
2232 		nfsrv_openpluslock++;
2233 	}
2234 	if (filestruct_locked != 0) {
2235 		NFSUNLOCKSTATE();
2236 		nfsrv_locallock_commit(lfp, lock_flags, first, end);
2237 		NFSLOCKSTATE();
2238 		nfsrv_unlocklf(lfp);
2239 	}
2240 	NFSUNLOCKSTATE();
2241 
2242 out:
2243 	if (haslock) {
2244 		NFSLOCKV4ROOTMUTEX();
2245 		nfsv4_unlock(&nfsv4rootfs_lock, 1);
2246 		NFSUNLOCKV4ROOTMUTEX();
2247 	}
2248 	if (vnode_unlocked != 0) {
2249 		NFSVOPLOCK(vp, LK_EXCLUSIVE | LK_RETRY);
2250 		if (error == 0 && (vp->v_iflag & VI_DOOMED) != 0)
2251 			error = NFSERR_SERVERFAULT;
2252 	}
2253 	if (other_lop)
2254 		free(other_lop, M_NFSDLOCK);
2255 	NFSEXITCODE2(error, nd);
2256 	return (error);
2257 }
2258 
2259 /*
2260  * Check for state errors for Open.
2261  * repstat is passed back out as an error if more critical errors
2262  * are not detected.
2263  */
2264 APPLESTATIC int
2265 nfsrv_opencheck(nfsquad_t clientid, nfsv4stateid_t *stateidp,
2266     struct nfsstate *new_stp, vnode_t vp, struct nfsrv_descript *nd,
2267     NFSPROC_T *p, int repstat)
2268 {
2269 	struct nfsstate *stp, *nstp;
2270 	struct nfsclient *clp;
2271 	struct nfsstate *ownerstp;
2272 	struct nfslockfile *lfp, *new_lfp;
2273 	int error = 0, haslock = 0, ret, readonly = 0, getfhret = 0;
2274 
2275 	if ((new_stp->ls_flags & NFSLCK_SHAREBITS) == NFSLCK_READACCESS)
2276 		readonly = 1;
2277 	/*
2278 	 * Check for restart conditions (client and server).
2279 	 */
2280 	error = nfsrv_checkrestart(clientid, new_stp->ls_flags,
2281 		&new_stp->ls_stateid, 0);
2282 	if (error)
2283 		goto out;
2284 
2285 	/*
2286 	 * Check for state resource limit exceeded.
2287 	 * Technically this should be SMP protected, but the worst
2288 	 * case error is "out by one or two" on the count when it
2289 	 * returns NFSERR_RESOURCE and the limit is just a rather
2290 	 * arbitrary high water mark, so no harm is done.
2291 	 */
2292 	if (nfsrv_openpluslock > nfsrv_v4statelimit) {
2293 		error = NFSERR_RESOURCE;
2294 		goto out;
2295 	}
2296 
2297 tryagain:
2298 	new_lfp = malloc(sizeof (struct nfslockfile),
2299 	    M_NFSDLOCKFILE, M_WAITOK);
2300 	if (vp)
2301 		getfhret = nfsrv_getlockfh(vp, new_stp->ls_flags, new_lfp,
2302 		    NULL, p);
2303 	NFSLOCKSTATE();
2304 	/*
2305 	 * Get the nfsclient structure.
2306 	 */
2307 	error = nfsrv_getclient(clientid, CLOPS_RENEW, &clp, NULL,
2308 	    (nfsquad_t)((u_quad_t)0), 0, nd, p);
2309 
2310 	/*
2311 	 * Look up the open owner. See if it needs confirmation and
2312 	 * check the seq#, as required.
2313 	 */
2314 	if (!error)
2315 		nfsrv_getowner(&clp->lc_open, new_stp, &ownerstp);
2316 
2317 	if (!error && ownerstp) {
2318 		error = nfsrv_checkseqid(nd, new_stp->ls_seq, ownerstp,
2319 		    new_stp->ls_op);
2320 		/*
2321 		 * If the OpenOwner hasn't been confirmed, assume the
2322 		 * old one was a replay and this one is ok.
2323 		 * See: RFC3530 Sec. 14.2.18.
2324 		 */
2325 		if (error == NFSERR_BADSEQID &&
2326 		    (ownerstp->ls_flags & NFSLCK_NEEDSCONFIRM))
2327 			error = 0;
2328 	}
2329 
2330 	/*
2331 	 * Check for grace.
2332 	 */
2333 	if (!error)
2334 		error = nfsrv_checkgrace(nd, clp, new_stp->ls_flags);
2335 	if ((new_stp->ls_flags & NFSLCK_RECLAIM) && !error &&
2336 		nfsrv_checkstable(clp))
2337 		error = NFSERR_NOGRACE;
2338 
2339 	/*
2340 	 * If none of the above errors occurred, let repstat be
2341 	 * returned.
2342 	 */
2343 	if (repstat && !error)
2344 		error = repstat;
2345 	if (error) {
2346 		NFSUNLOCKSTATE();
2347 		if (haslock) {
2348 			NFSLOCKV4ROOTMUTEX();
2349 			nfsv4_unlock(&nfsv4rootfs_lock, 1);
2350 			NFSUNLOCKV4ROOTMUTEX();
2351 		}
2352 		free(new_lfp, M_NFSDLOCKFILE);
2353 		goto out;
2354 	}
2355 
2356 	/*
2357 	 * If vp == NULL, the file doesn't exist yet, so return ok.
2358 	 * (This always happens on the first pass, so haslock must be 0.)
2359 	 */
2360 	if (vp == NULL) {
2361 		NFSUNLOCKSTATE();
2362 		free(new_lfp, M_NFSDLOCKFILE);
2363 		goto out;
2364 	}
2365 
2366 	/*
2367 	 * Get the structure for the underlying file.
2368 	 */
2369 	if (getfhret)
2370 		error = getfhret;
2371 	else
2372 		error = nfsrv_getlockfile(new_stp->ls_flags, &new_lfp, &lfp,
2373 		    NULL, 0);
2374 	if (new_lfp)
2375 		free(new_lfp, M_NFSDLOCKFILE);
2376 	if (error) {
2377 		NFSUNLOCKSTATE();
2378 		if (haslock) {
2379 			NFSLOCKV4ROOTMUTEX();
2380 			nfsv4_unlock(&nfsv4rootfs_lock, 1);
2381 			NFSUNLOCKV4ROOTMUTEX();
2382 		}
2383 		goto out;
2384 	}
2385 
2386 	/*
2387 	 * Search for a conflicting open/share.
2388 	 */
2389 	if (new_stp->ls_flags & NFSLCK_DELEGCUR) {
2390 	    /*
2391 	     * For Delegate_Cur, search for the matching Delegation,
2392 	     * which indicates no conflict.
2393 	     * An old delegation should have been recovered by the
2394 	     * client doing a Claim_DELEGATE_Prev, so I won't let
2395 	     * it match and return NFSERR_EXPIRED. Should I let it
2396 	     * match?
2397 	     */
2398 	    LIST_FOREACH(stp, &lfp->lf_deleg, ls_file) {
2399 		if (!(stp->ls_flags & NFSLCK_OLDDELEG) &&
2400 		    (((nd->nd_flag & ND_NFSV41) != 0 &&
2401 		    stateidp->seqid == 0) ||
2402 		    stateidp->seqid == stp->ls_stateid.seqid) &&
2403 		    !NFSBCMP(stateidp->other, stp->ls_stateid.other,
2404 			  NFSX_STATEIDOTHER))
2405 			break;
2406 	    }
2407 	    if (stp == LIST_END(&lfp->lf_deleg) ||
2408 		((new_stp->ls_flags & NFSLCK_WRITEACCESS) &&
2409 		 (stp->ls_flags & NFSLCK_DELEGREAD))) {
2410 		NFSUNLOCKSTATE();
2411 		if (haslock) {
2412 			NFSLOCKV4ROOTMUTEX();
2413 			nfsv4_unlock(&nfsv4rootfs_lock, 1);
2414 			NFSUNLOCKV4ROOTMUTEX();
2415 		}
2416 		error = NFSERR_EXPIRED;
2417 		goto out;
2418 	    }
2419 	}
2420 
2421 	/*
2422 	 * Check for access/deny bit conflicts. I check for the same
2423 	 * owner as well, in case the client didn't bother.
2424 	 */
2425 	LIST_FOREACH(stp, &lfp->lf_open, ls_file) {
2426 		if (!(new_stp->ls_flags & NFSLCK_DELEGCUR) &&
2427 		    (((new_stp->ls_flags & NFSLCK_ACCESSBITS) &
2428 		      ((stp->ls_flags>>NFSLCK_SHIFT) & NFSLCK_ACCESSBITS))||
2429 		     ((stp->ls_flags & NFSLCK_ACCESSBITS) &
2430 		      ((new_stp->ls_flags>>NFSLCK_SHIFT)&NFSLCK_ACCESSBITS)))){
2431 			ret = nfsrv_clientconflict(stp->ls_clp,&haslock,vp,p);
2432 			if (ret == 1) {
2433 				/*
2434 				 * nfsrv_clientconflict() unlocks
2435 				 * state when it returns non-zero.
2436 				 */
2437 				goto tryagain;
2438 			}
2439 			if (ret == 2)
2440 				error = NFSERR_PERM;
2441 			else if (new_stp->ls_flags & NFSLCK_RECLAIM)
2442 				error = NFSERR_RECLAIMCONFLICT;
2443 			else
2444 				error = NFSERR_SHAREDENIED;
2445 			if (ret == 0)
2446 				NFSUNLOCKSTATE();
2447 			if (haslock) {
2448 				NFSLOCKV4ROOTMUTEX();
2449 				nfsv4_unlock(&nfsv4rootfs_lock, 1);
2450 				NFSUNLOCKV4ROOTMUTEX();
2451 			}
2452 			goto out;
2453 		}
2454 	}
2455 
2456 	/*
2457 	 * Check for a conflicting delegation. If one is found, call
2458 	 * nfsrv_delegconflict() to handle it. If the v4root lock hasn't
2459 	 * been set yet, it will get the lock. Otherwise, it will recall
2460 	 * the delegation. Then, we try try again...
2461 	 * (If NFSLCK_DELEGCUR is set, it has a delegation, so there
2462 	 *  isn't a conflict.)
2463 	 * I currently believe the conflict algorithm to be:
2464 	 * For Open with Read Access and Deny None
2465 	 * - there is a conflict iff a different client has a write delegation
2466 	 * For Open with other Write Access or any Deny except None
2467 	 * - there is a conflict if a different client has any delegation
2468 	 * - there is a conflict if the same client has a read delegation
2469 	 *   (The current consensus is that this last case should be
2470 	 *    considered a conflict since the client with a read delegation
2471 	 *    could have done an Open with ReadAccess and WriteDeny
2472 	 *    locally and then not have checked for the WriteDeny.)
2473 	 * Don't check for a Reclaim, since that will be dealt with
2474 	 * by nfsrv_openctrl().
2475 	 */
2476 	if (!(new_stp->ls_flags &
2477 		(NFSLCK_DELEGPREV | NFSLCK_DELEGCUR | NFSLCK_RECLAIM))) {
2478 	    stp = LIST_FIRST(&lfp->lf_deleg);
2479 	    while (stp != LIST_END(&lfp->lf_deleg)) {
2480 		nstp = LIST_NEXT(stp, ls_file);
2481 		if ((readonly && stp->ls_clp != clp &&
2482 		       (stp->ls_flags & NFSLCK_DELEGWRITE)) ||
2483 		    (!readonly && (stp->ls_clp != clp ||
2484 		         (stp->ls_flags & NFSLCK_DELEGREAD)))) {
2485 			ret = nfsrv_delegconflict(stp, &haslock, p, vp);
2486 			if (ret) {
2487 			    /*
2488 			     * nfsrv_delegconflict() unlocks state
2489 			     * when it returns non-zero.
2490 			     */
2491 			    if (ret == -1)
2492 				goto tryagain;
2493 			    error = ret;
2494 			    goto out;
2495 			}
2496 		}
2497 		stp = nstp;
2498 	    }
2499 	}
2500 	NFSUNLOCKSTATE();
2501 	if (haslock) {
2502 		NFSLOCKV4ROOTMUTEX();
2503 		nfsv4_unlock(&nfsv4rootfs_lock, 1);
2504 		NFSUNLOCKV4ROOTMUTEX();
2505 	}
2506 
2507 out:
2508 	NFSEXITCODE2(error, nd);
2509 	return (error);
2510 }
2511 
2512 /*
2513  * Open control function to create/update open state for an open.
2514  */
2515 APPLESTATIC int
2516 nfsrv_openctrl(struct nfsrv_descript *nd, vnode_t vp,
2517     struct nfsstate **new_stpp, nfsquad_t clientid, nfsv4stateid_t *stateidp,
2518     nfsv4stateid_t *delegstateidp, u_int32_t *rflagsp, struct nfsexstuff *exp,
2519     NFSPROC_T *p, u_quad_t filerev)
2520 {
2521 	struct nfsstate *new_stp = *new_stpp;
2522 	struct nfsstate *stp, *nstp;
2523 	struct nfsstate *openstp = NULL, *new_open, *ownerstp, *new_deleg;
2524 	struct nfslockfile *lfp, *new_lfp;
2525 	struct nfsclient *clp;
2526 	int error = 0, haslock = 0, ret, delegate = 1, writedeleg = 1;
2527 	int readonly = 0, cbret = 1, getfhret = 0;
2528 	int gotstate = 0, len = 0;
2529 	u_char *clidp = NULL;
2530 
2531 	if ((new_stp->ls_flags & NFSLCK_SHAREBITS) == NFSLCK_READACCESS)
2532 		readonly = 1;
2533 	/*
2534 	 * Check for restart conditions (client and server).
2535 	 * (Paranoia, should have been detected by nfsrv_opencheck().)
2536 	 * If an error does show up, return NFSERR_EXPIRED, since the
2537 	 * the seqid# has already been incremented.
2538 	 */
2539 	error = nfsrv_checkrestart(clientid, new_stp->ls_flags,
2540 	    &new_stp->ls_stateid, 0);
2541 	if (error) {
2542 		printf("Nfsd: openctrl unexpected restart err=%d\n",
2543 		    error);
2544 		error = NFSERR_EXPIRED;
2545 		goto out;
2546 	}
2547 
2548 	clidp = malloc(NFSV4_OPAQUELIMIT, M_TEMP, M_WAITOK);
2549 tryagain:
2550 	new_lfp = malloc(sizeof (struct nfslockfile),
2551 	    M_NFSDLOCKFILE, M_WAITOK);
2552 	new_open = malloc(sizeof (struct nfsstate),
2553 	    M_NFSDSTATE, M_WAITOK);
2554 	new_deleg = malloc(sizeof (struct nfsstate),
2555 	    M_NFSDSTATE, M_WAITOK);
2556 	getfhret = nfsrv_getlockfh(vp, new_stp->ls_flags, new_lfp,
2557 	    NULL, p);
2558 	NFSLOCKSTATE();
2559 	/*
2560 	 * Get the client structure. Since the linked lists could be changed
2561 	 * by other nfsd processes if this process does a tsleep(), one of
2562 	 * two things must be done.
2563 	 * 1 - don't tsleep()
2564 	 * or
2565 	 * 2 - get the nfsv4_lock() { indicated by haslock == 1 }
2566 	 *     before using the lists, since this lock stops the other
2567 	 *     nfsd. This should only be used for rare cases, since it
2568 	 *     essentially single threads the nfsd.
2569 	 *     At this time, it is only done for cases where the stable
2570 	 *     storage file must be written prior to completion of state
2571 	 *     expiration.
2572 	 */
2573 	error = nfsrv_getclient(clientid, CLOPS_RENEW, &clp, NULL,
2574 	    (nfsquad_t)((u_quad_t)0), 0, nd, p);
2575 	if (!error && (clp->lc_flags & LCL_NEEDSCBNULL) &&
2576 	    clp->lc_program) {
2577 		/*
2578 		 * This happens on the first open for a client
2579 		 * that supports callbacks.
2580 		 */
2581 		NFSUNLOCKSTATE();
2582 		/*
2583 		 * Although nfsrv_docallback() will sleep, clp won't
2584 		 * go away, since they are only removed when the
2585 		 * nfsv4_lock() has blocked the nfsd threads. The
2586 		 * fields in clp can change, but having multiple
2587 		 * threads do this Null callback RPC should be
2588 		 * harmless.
2589 		 */
2590 		cbret = nfsrv_docallback(clp, NFSV4PROC_CBNULL,
2591 		    NULL, 0, NULL, NULL, NULL, p);
2592 		NFSLOCKSTATE();
2593 		clp->lc_flags &= ~LCL_NEEDSCBNULL;
2594 		if (!cbret)
2595 			clp->lc_flags |= LCL_CALLBACKSON;
2596 	}
2597 
2598 	/*
2599 	 * Look up the open owner. See if it needs confirmation and
2600 	 * check the seq#, as required.
2601 	 */
2602 	if (!error)
2603 		nfsrv_getowner(&clp->lc_open, new_stp, &ownerstp);
2604 
2605 	if (error) {
2606 		NFSUNLOCKSTATE();
2607 		printf("Nfsd: openctrl unexpected state err=%d\n",
2608 			error);
2609 		free(new_lfp, M_NFSDLOCKFILE);
2610 		free(new_open, M_NFSDSTATE);
2611 		free(new_deleg, M_NFSDSTATE);
2612 		if (haslock) {
2613 			NFSLOCKV4ROOTMUTEX();
2614 			nfsv4_unlock(&nfsv4rootfs_lock, 1);
2615 			NFSUNLOCKV4ROOTMUTEX();
2616 		}
2617 		error = NFSERR_EXPIRED;
2618 		goto out;
2619 	}
2620 
2621 	if (new_stp->ls_flags & NFSLCK_RECLAIM)
2622 		nfsrv_markstable(clp);
2623 
2624 	/*
2625 	 * Get the structure for the underlying file.
2626 	 */
2627 	if (getfhret)
2628 		error = getfhret;
2629 	else
2630 		error = nfsrv_getlockfile(new_stp->ls_flags, &new_lfp, &lfp,
2631 		    NULL, 0);
2632 	if (new_lfp)
2633 		free(new_lfp, M_NFSDLOCKFILE);
2634 	if (error) {
2635 		NFSUNLOCKSTATE();
2636 		printf("Nfsd openctrl unexpected getlockfile err=%d\n",
2637 		    error);
2638 		free(new_open, M_NFSDSTATE);
2639 		free(new_deleg, M_NFSDSTATE);
2640 		if (haslock) {
2641 			NFSLOCKV4ROOTMUTEX();
2642 			nfsv4_unlock(&nfsv4rootfs_lock, 1);
2643 			NFSUNLOCKV4ROOTMUTEX();
2644 		}
2645 		goto out;
2646 	}
2647 
2648 	/*
2649 	 * Search for a conflicting open/share.
2650 	 */
2651 	if (new_stp->ls_flags & NFSLCK_DELEGCUR) {
2652 	    /*
2653 	     * For Delegate_Cur, search for the matching Delegation,
2654 	     * which indicates no conflict.
2655 	     * An old delegation should have been recovered by the
2656 	     * client doing a Claim_DELEGATE_Prev, so I won't let
2657 	     * it match and return NFSERR_EXPIRED. Should I let it
2658 	     * match?
2659 	     */
2660 	    LIST_FOREACH(stp, &lfp->lf_deleg, ls_file) {
2661 		if (!(stp->ls_flags & NFSLCK_OLDDELEG) &&
2662 		    (((nd->nd_flag & ND_NFSV41) != 0 &&
2663 		    stateidp->seqid == 0) ||
2664 		    stateidp->seqid == stp->ls_stateid.seqid) &&
2665 		    !NFSBCMP(stateidp->other, stp->ls_stateid.other,
2666 			NFSX_STATEIDOTHER))
2667 			break;
2668 	    }
2669 	    if (stp == LIST_END(&lfp->lf_deleg) ||
2670 		((new_stp->ls_flags & NFSLCK_WRITEACCESS) &&
2671 		 (stp->ls_flags & NFSLCK_DELEGREAD))) {
2672 		NFSUNLOCKSTATE();
2673 		printf("Nfsd openctrl unexpected expiry\n");
2674 		free(new_open, M_NFSDSTATE);
2675 		free(new_deleg, M_NFSDSTATE);
2676 		if (haslock) {
2677 			NFSLOCKV4ROOTMUTEX();
2678 			nfsv4_unlock(&nfsv4rootfs_lock, 1);
2679 			NFSUNLOCKV4ROOTMUTEX();
2680 		}
2681 		error = NFSERR_EXPIRED;
2682 		goto out;
2683 	    }
2684 
2685 	    /*
2686 	     * Don't issue a Delegation, since one already exists and
2687 	     * delay delegation timeout, as required.
2688 	     */
2689 	    delegate = 0;
2690 	    nfsrv_delaydelegtimeout(stp);
2691 	}
2692 
2693 	/*
2694 	 * Check for access/deny bit conflicts. I also check for the
2695 	 * same owner, since the client might not have bothered to check.
2696 	 * Also, note an open for the same file and owner, if found,
2697 	 * which is all we do here for Delegate_Cur, since conflict
2698 	 * checking is already done.
2699 	 */
2700 	LIST_FOREACH(stp, &lfp->lf_open, ls_file) {
2701 		if (ownerstp && stp->ls_openowner == ownerstp)
2702 			openstp = stp;
2703 		if (!(new_stp->ls_flags & NFSLCK_DELEGCUR)) {
2704 		    /*
2705 		     * If another client has the file open, the only
2706 		     * delegation that can be issued is a Read delegation
2707 		     * and only if it is a Read open with Deny none.
2708 		     */
2709 		    if (clp != stp->ls_clp) {
2710 			if ((stp->ls_flags & NFSLCK_SHAREBITS) ==
2711 			    NFSLCK_READACCESS)
2712 			    writedeleg = 0;
2713 			else
2714 			    delegate = 0;
2715 		    }
2716 		    if(((new_stp->ls_flags & NFSLCK_ACCESSBITS) &
2717 		        ((stp->ls_flags>>NFSLCK_SHIFT) & NFSLCK_ACCESSBITS))||
2718 		       ((stp->ls_flags & NFSLCK_ACCESSBITS) &
2719 		        ((new_stp->ls_flags>>NFSLCK_SHIFT)&NFSLCK_ACCESSBITS))){
2720 			ret = nfsrv_clientconflict(stp->ls_clp,&haslock,vp,p);
2721 			if (ret == 1) {
2722 				/*
2723 				 * nfsrv_clientconflict() unlocks state
2724 				 * when it returns non-zero.
2725 				 */
2726 				free(new_open, M_NFSDSTATE);
2727 				free(new_deleg, M_NFSDSTATE);
2728 				openstp = NULL;
2729 				goto tryagain;
2730 			}
2731 			if (ret == 2)
2732 				error = NFSERR_PERM;
2733 			else if (new_stp->ls_flags & NFSLCK_RECLAIM)
2734 				error = NFSERR_RECLAIMCONFLICT;
2735 			else
2736 				error = NFSERR_SHAREDENIED;
2737 			if (ret == 0)
2738 				NFSUNLOCKSTATE();
2739 			if (haslock) {
2740 				NFSLOCKV4ROOTMUTEX();
2741 				nfsv4_unlock(&nfsv4rootfs_lock, 1);
2742 				NFSUNLOCKV4ROOTMUTEX();
2743 			}
2744 			free(new_open, M_NFSDSTATE);
2745 			free(new_deleg, M_NFSDSTATE);
2746 			printf("nfsd openctrl unexpected client cnfl\n");
2747 			goto out;
2748 		    }
2749 		}
2750 	}
2751 
2752 	/*
2753 	 * Check for a conflicting delegation. If one is found, call
2754 	 * nfsrv_delegconflict() to handle it. If the v4root lock hasn't
2755 	 * been set yet, it will get the lock. Otherwise, it will recall
2756 	 * the delegation. Then, we try try again...
2757 	 * (If NFSLCK_DELEGCUR is set, it has a delegation, so there
2758 	 *  isn't a conflict.)
2759 	 * I currently believe the conflict algorithm to be:
2760 	 * For Open with Read Access and Deny None
2761 	 * - there is a conflict iff a different client has a write delegation
2762 	 * For Open with other Write Access or any Deny except None
2763 	 * - there is a conflict if a different client has any delegation
2764 	 * - there is a conflict if the same client has a read delegation
2765 	 *   (The current consensus is that this last case should be
2766 	 *    considered a conflict since the client with a read delegation
2767 	 *    could have done an Open with ReadAccess and WriteDeny
2768 	 *    locally and then not have checked for the WriteDeny.)
2769 	 */
2770 	if (!(new_stp->ls_flags & (NFSLCK_DELEGPREV | NFSLCK_DELEGCUR))) {
2771 	    stp = LIST_FIRST(&lfp->lf_deleg);
2772 	    while (stp != LIST_END(&lfp->lf_deleg)) {
2773 		nstp = LIST_NEXT(stp, ls_file);
2774 		if (stp->ls_clp != clp && (stp->ls_flags & NFSLCK_DELEGREAD))
2775 			writedeleg = 0;
2776 		else
2777 			delegate = 0;
2778 		if ((readonly && stp->ls_clp != clp &&
2779 		       (stp->ls_flags & NFSLCK_DELEGWRITE)) ||
2780 		    (!readonly && (stp->ls_clp != clp ||
2781 		         (stp->ls_flags & NFSLCK_DELEGREAD)))) {
2782 		    if (new_stp->ls_flags & NFSLCK_RECLAIM) {
2783 			delegate = 2;
2784 		    } else {
2785 			ret = nfsrv_delegconflict(stp, &haslock, p, vp);
2786 			if (ret) {
2787 			    /*
2788 			     * nfsrv_delegconflict() unlocks state
2789 			     * when it returns non-zero.
2790 			     */
2791 			    printf("Nfsd openctrl unexpected deleg cnfl\n");
2792 			    free(new_open, M_NFSDSTATE);
2793 			    free(new_deleg, M_NFSDSTATE);
2794 			    if (ret == -1) {
2795 				openstp = NULL;
2796 				goto tryagain;
2797 			    }
2798 			    error = ret;
2799 			    goto out;
2800 			}
2801 		    }
2802 		}
2803 		stp = nstp;
2804 	    }
2805 	}
2806 
2807 	/*
2808 	 * We only get here if there was no open that conflicted.
2809 	 * If an open for the owner exists, or in the access/deny bits.
2810 	 * Otherwise it is a new open. If the open_owner hasn't been
2811 	 * confirmed, replace the open with the new one needing confirmation,
2812 	 * otherwise add the open.
2813 	 */
2814 	if (new_stp->ls_flags & NFSLCK_DELEGPREV) {
2815 	    /*
2816 	     * Handle NFSLCK_DELEGPREV by searching the old delegations for
2817 	     * a match. If found, just move the old delegation to the current
2818 	     * delegation list and issue open. If not found, return
2819 	     * NFSERR_EXPIRED.
2820 	     */
2821 	    LIST_FOREACH(stp, &clp->lc_olddeleg, ls_list) {
2822 		if (stp->ls_lfp == lfp) {
2823 		    /* Found it */
2824 		    if (stp->ls_clp != clp)
2825 			panic("olddeleg clp");
2826 		    LIST_REMOVE(stp, ls_list);
2827 		    LIST_REMOVE(stp, ls_hash);
2828 		    stp->ls_flags &= ~NFSLCK_OLDDELEG;
2829 		    stp->ls_stateid.seqid = delegstateidp->seqid = 1;
2830 		    stp->ls_stateid.other[0] = delegstateidp->other[0] =
2831 			clp->lc_clientid.lval[0];
2832 		    stp->ls_stateid.other[1] = delegstateidp->other[1] =
2833 			clp->lc_clientid.lval[1];
2834 		    stp->ls_stateid.other[2] = delegstateidp->other[2] =
2835 			nfsrv_nextstateindex(clp);
2836 		    stp->ls_compref = nd->nd_compref;
2837 		    LIST_INSERT_HEAD(&clp->lc_deleg, stp, ls_list);
2838 		    LIST_INSERT_HEAD(NFSSTATEHASH(clp,
2839 			stp->ls_stateid), stp, ls_hash);
2840 		    if (stp->ls_flags & NFSLCK_DELEGWRITE)
2841 			*rflagsp |= NFSV4OPEN_WRITEDELEGATE;
2842 		    else
2843 			*rflagsp |= NFSV4OPEN_READDELEGATE;
2844 		    clp->lc_delegtime = NFSD_MONOSEC +
2845 			nfsrv_lease + NFSRV_LEASEDELTA;
2846 
2847 		    /*
2848 		     * Now, do the associated open.
2849 		     */
2850 		    new_open->ls_stateid.seqid = 1;
2851 		    new_open->ls_stateid.other[0] = clp->lc_clientid.lval[0];
2852 		    new_open->ls_stateid.other[1] = clp->lc_clientid.lval[1];
2853 		    new_open->ls_stateid.other[2] = nfsrv_nextstateindex(clp);
2854 		    new_open->ls_flags = (new_stp->ls_flags&NFSLCK_DENYBITS)|
2855 			NFSLCK_OPEN;
2856 		    if (stp->ls_flags & NFSLCK_DELEGWRITE)
2857 			new_open->ls_flags |= (NFSLCK_READACCESS |
2858 			    NFSLCK_WRITEACCESS);
2859 		    else
2860 			new_open->ls_flags |= NFSLCK_READACCESS;
2861 		    new_open->ls_uid = new_stp->ls_uid;
2862 		    new_open->ls_lfp = lfp;
2863 		    new_open->ls_clp = clp;
2864 		    LIST_INIT(&new_open->ls_open);
2865 		    LIST_INSERT_HEAD(&lfp->lf_open, new_open, ls_file);
2866 		    LIST_INSERT_HEAD(NFSSTATEHASH(clp, new_open->ls_stateid),
2867 			new_open, ls_hash);
2868 		    /*
2869 		     * and handle the open owner
2870 		     */
2871 		    if (ownerstp) {
2872 			new_open->ls_openowner = ownerstp;
2873 			LIST_INSERT_HEAD(&ownerstp->ls_open,new_open,ls_list);
2874 		    } else {
2875 			new_open->ls_openowner = new_stp;
2876 			new_stp->ls_flags = 0;
2877 			nfsrvd_refcache(new_stp->ls_op);
2878 			new_stp->ls_noopens = 0;
2879 			LIST_INIT(&new_stp->ls_open);
2880 			LIST_INSERT_HEAD(&new_stp->ls_open, new_open, ls_list);
2881 			LIST_INSERT_HEAD(&clp->lc_open, new_stp, ls_list);
2882 			*new_stpp = NULL;
2883 			nfsstatsv1.srvopenowners++;
2884 			nfsrv_openpluslock++;
2885 		    }
2886 		    openstp = new_open;
2887 		    new_open = NULL;
2888 		    nfsstatsv1.srvopens++;
2889 		    nfsrv_openpluslock++;
2890 		    break;
2891 		}
2892 	    }
2893 	    if (stp == LIST_END(&clp->lc_olddeleg))
2894 		error = NFSERR_EXPIRED;
2895 	} else if (new_stp->ls_flags & (NFSLCK_DELEGREAD | NFSLCK_DELEGWRITE)) {
2896 	    /*
2897 	     * Scan to see that no delegation for this client and file
2898 	     * doesn't already exist.
2899 	     * There also shouldn't yet be an Open for this file and
2900 	     * openowner.
2901 	     */
2902 	    LIST_FOREACH(stp, &lfp->lf_deleg, ls_file) {
2903 		if (stp->ls_clp == clp)
2904 		    break;
2905 	    }
2906 	    if (stp == LIST_END(&lfp->lf_deleg) && openstp == NULL) {
2907 		/*
2908 		 * This is the Claim_Previous case with a delegation
2909 		 * type != Delegate_None.
2910 		 */
2911 		/*
2912 		 * First, add the delegation. (Although we must issue the
2913 		 * delegation, we can also ask for an immediate return.)
2914 		 */
2915 		new_deleg->ls_stateid.seqid = delegstateidp->seqid = 1;
2916 		new_deleg->ls_stateid.other[0] = delegstateidp->other[0] =
2917 		    clp->lc_clientid.lval[0];
2918 		new_deleg->ls_stateid.other[1] = delegstateidp->other[1] =
2919 		    clp->lc_clientid.lval[1];
2920 		new_deleg->ls_stateid.other[2] = delegstateidp->other[2] =
2921 		    nfsrv_nextstateindex(clp);
2922 		if (new_stp->ls_flags & NFSLCK_DELEGWRITE) {
2923 		    new_deleg->ls_flags = (NFSLCK_DELEGWRITE |
2924 			NFSLCK_READACCESS | NFSLCK_WRITEACCESS);
2925 		    *rflagsp |= NFSV4OPEN_WRITEDELEGATE;
2926 		    nfsrv_writedelegcnt++;
2927 		} else {
2928 		    new_deleg->ls_flags = (NFSLCK_DELEGREAD |
2929 			NFSLCK_READACCESS);
2930 		    *rflagsp |= NFSV4OPEN_READDELEGATE;
2931 		}
2932 		new_deleg->ls_uid = new_stp->ls_uid;
2933 		new_deleg->ls_lfp = lfp;
2934 		new_deleg->ls_clp = clp;
2935 		new_deleg->ls_filerev = filerev;
2936 		new_deleg->ls_compref = nd->nd_compref;
2937 		LIST_INSERT_HEAD(&lfp->lf_deleg, new_deleg, ls_file);
2938 		LIST_INSERT_HEAD(NFSSTATEHASH(clp,
2939 		    new_deleg->ls_stateid), new_deleg, ls_hash);
2940 		LIST_INSERT_HEAD(&clp->lc_deleg, new_deleg, ls_list);
2941 		new_deleg = NULL;
2942 		if (delegate == 2 || nfsrv_issuedelegs == 0 ||
2943 		    (clp->lc_flags & (LCL_CALLBACKSON | LCL_CBDOWN)) !=
2944 		     LCL_CALLBACKSON ||
2945 		    NFSRV_V4DELEGLIMIT(nfsrv_delegatecnt) ||
2946 		    !NFSVNO_DELEGOK(vp))
2947 		    *rflagsp |= NFSV4OPEN_RECALL;
2948 		nfsstatsv1.srvdelegates++;
2949 		nfsrv_openpluslock++;
2950 		nfsrv_delegatecnt++;
2951 
2952 		/*
2953 		 * Now, do the associated open.
2954 		 */
2955 		new_open->ls_stateid.seqid = 1;
2956 		new_open->ls_stateid.other[0] = clp->lc_clientid.lval[0];
2957 		new_open->ls_stateid.other[1] = clp->lc_clientid.lval[1];
2958 		new_open->ls_stateid.other[2] = nfsrv_nextstateindex(clp);
2959 		new_open->ls_flags = (new_stp->ls_flags & NFSLCK_DENYBITS) |
2960 		    NFSLCK_OPEN;
2961 		if (new_stp->ls_flags & NFSLCK_DELEGWRITE)
2962 			new_open->ls_flags |= (NFSLCK_READACCESS |
2963 			    NFSLCK_WRITEACCESS);
2964 		else
2965 			new_open->ls_flags |= NFSLCK_READACCESS;
2966 		new_open->ls_uid = new_stp->ls_uid;
2967 		new_open->ls_lfp = lfp;
2968 		new_open->ls_clp = clp;
2969 		LIST_INIT(&new_open->ls_open);
2970 		LIST_INSERT_HEAD(&lfp->lf_open, new_open, ls_file);
2971 		LIST_INSERT_HEAD(NFSSTATEHASH(clp, new_open->ls_stateid),
2972 		   new_open, ls_hash);
2973 		/*
2974 		 * and handle the open owner
2975 		 */
2976 		if (ownerstp) {
2977 		    new_open->ls_openowner = ownerstp;
2978 		    LIST_INSERT_HEAD(&ownerstp->ls_open, new_open, ls_list);
2979 		} else {
2980 		    new_open->ls_openowner = new_stp;
2981 		    new_stp->ls_flags = 0;
2982 		    nfsrvd_refcache(new_stp->ls_op);
2983 		    new_stp->ls_noopens = 0;
2984 		    LIST_INIT(&new_stp->ls_open);
2985 		    LIST_INSERT_HEAD(&new_stp->ls_open, new_open, ls_list);
2986 		    LIST_INSERT_HEAD(&clp->lc_open, new_stp, ls_list);
2987 		    *new_stpp = NULL;
2988 		    nfsstatsv1.srvopenowners++;
2989 		    nfsrv_openpluslock++;
2990 		}
2991 		openstp = new_open;
2992 		new_open = NULL;
2993 		nfsstatsv1.srvopens++;
2994 		nfsrv_openpluslock++;
2995 	    } else {
2996 		error = NFSERR_RECLAIMCONFLICT;
2997 	    }
2998 	} else if (ownerstp) {
2999 		if (ownerstp->ls_flags & NFSLCK_NEEDSCONFIRM) {
3000 		    /* Replace the open */
3001 		    if (ownerstp->ls_op)
3002 			nfsrvd_derefcache(ownerstp->ls_op);
3003 		    ownerstp->ls_op = new_stp->ls_op;
3004 		    nfsrvd_refcache(ownerstp->ls_op);
3005 		    ownerstp->ls_seq = new_stp->ls_seq;
3006 		    *rflagsp |= NFSV4OPEN_RESULTCONFIRM;
3007 		    stp = LIST_FIRST(&ownerstp->ls_open);
3008 		    stp->ls_flags = (new_stp->ls_flags & NFSLCK_SHAREBITS) |
3009 			NFSLCK_OPEN;
3010 		    stp->ls_stateid.seqid = 1;
3011 		    stp->ls_uid = new_stp->ls_uid;
3012 		    if (lfp != stp->ls_lfp) {
3013 			LIST_REMOVE(stp, ls_file);
3014 			LIST_INSERT_HEAD(&lfp->lf_open, stp, ls_file);
3015 			stp->ls_lfp = lfp;
3016 		    }
3017 		    openstp = stp;
3018 		} else if (openstp) {
3019 		    openstp->ls_flags |= (new_stp->ls_flags & NFSLCK_SHAREBITS);
3020 		    openstp->ls_stateid.seqid++;
3021 		    if ((nd->nd_flag & ND_NFSV41) != 0 &&
3022 			openstp->ls_stateid.seqid == 0)
3023 			openstp->ls_stateid.seqid = 1;
3024 
3025 		    /*
3026 		     * This is where we can choose to issue a delegation.
3027 		     */
3028 		    if (delegate == 0 || writedeleg == 0 ||
3029 			NFSVNO_EXRDONLY(exp) || (readonly != 0 &&
3030 			nfsrv_writedelegifpos == 0) ||
3031 			!NFSVNO_DELEGOK(vp) ||
3032 			(new_stp->ls_flags & NFSLCK_WANTRDELEG) != 0 ||
3033 			(clp->lc_flags & (LCL_CALLBACKSON | LCL_CBDOWN)) !=
3034 			 LCL_CALLBACKSON)
3035 			*rflagsp |= NFSV4OPEN_WDCONTENTION;
3036 		    else if (nfsrv_issuedelegs == 0 ||
3037 			NFSRV_V4DELEGLIMIT(nfsrv_delegatecnt))
3038 			*rflagsp |= NFSV4OPEN_WDRESOURCE;
3039 		    else if ((new_stp->ls_flags & NFSLCK_WANTNODELEG) != 0)
3040 			*rflagsp |= NFSV4OPEN_WDNOTWANTED;
3041 		    else {
3042 			new_deleg->ls_stateid.seqid = delegstateidp->seqid = 1;
3043 			new_deleg->ls_stateid.other[0] = delegstateidp->other[0]
3044 			    = clp->lc_clientid.lval[0];
3045 			new_deleg->ls_stateid.other[1] = delegstateidp->other[1]
3046 			    = clp->lc_clientid.lval[1];
3047 			new_deleg->ls_stateid.other[2] = delegstateidp->other[2]
3048 			    = nfsrv_nextstateindex(clp);
3049 			new_deleg->ls_flags = (NFSLCK_DELEGWRITE |
3050 			    NFSLCK_READACCESS | NFSLCK_WRITEACCESS);
3051 			*rflagsp |= NFSV4OPEN_WRITEDELEGATE;
3052 			new_deleg->ls_uid = new_stp->ls_uid;
3053 			new_deleg->ls_lfp = lfp;
3054 			new_deleg->ls_clp = clp;
3055 			new_deleg->ls_filerev = filerev;
3056 			new_deleg->ls_compref = nd->nd_compref;
3057 			nfsrv_writedelegcnt++;
3058 			LIST_INSERT_HEAD(&lfp->lf_deleg, new_deleg, ls_file);
3059 			LIST_INSERT_HEAD(NFSSTATEHASH(clp,
3060 			    new_deleg->ls_stateid), new_deleg, ls_hash);
3061 			LIST_INSERT_HEAD(&clp->lc_deleg, new_deleg, ls_list);
3062 			new_deleg = NULL;
3063 			nfsstatsv1.srvdelegates++;
3064 			nfsrv_openpluslock++;
3065 			nfsrv_delegatecnt++;
3066 		    }
3067 		} else {
3068 		    new_open->ls_stateid.seqid = 1;
3069 		    new_open->ls_stateid.other[0] = clp->lc_clientid.lval[0];
3070 		    new_open->ls_stateid.other[1] = clp->lc_clientid.lval[1];
3071 		    new_open->ls_stateid.other[2] = nfsrv_nextstateindex(clp);
3072 		    new_open->ls_flags = (new_stp->ls_flags & NFSLCK_SHAREBITS)|
3073 			NFSLCK_OPEN;
3074 		    new_open->ls_uid = new_stp->ls_uid;
3075 		    new_open->ls_openowner = ownerstp;
3076 		    new_open->ls_lfp = lfp;
3077 		    new_open->ls_clp = clp;
3078 		    LIST_INIT(&new_open->ls_open);
3079 		    LIST_INSERT_HEAD(&lfp->lf_open, new_open, ls_file);
3080 		    LIST_INSERT_HEAD(&ownerstp->ls_open, new_open, ls_list);
3081 		    LIST_INSERT_HEAD(NFSSTATEHASH(clp, new_open->ls_stateid),
3082 			new_open, ls_hash);
3083 		    openstp = new_open;
3084 		    new_open = NULL;
3085 		    nfsstatsv1.srvopens++;
3086 		    nfsrv_openpluslock++;
3087 
3088 		    /*
3089 		     * This is where we can choose to issue a delegation.
3090 		     */
3091 		    if (delegate == 0 || (writedeleg == 0 && readonly == 0) ||
3092 			!NFSVNO_DELEGOK(vp) ||
3093 			(clp->lc_flags & (LCL_CALLBACKSON | LCL_CBDOWN)) !=
3094 			 LCL_CALLBACKSON)
3095 			*rflagsp |= NFSV4OPEN_WDCONTENTION;
3096 		    else if (nfsrv_issuedelegs == 0 ||
3097 			NFSRV_V4DELEGLIMIT(nfsrv_delegatecnt))
3098 			*rflagsp |= NFSV4OPEN_WDRESOURCE;
3099 		    else if ((new_stp->ls_flags & NFSLCK_WANTNODELEG) != 0)
3100 			*rflagsp |= NFSV4OPEN_WDNOTWANTED;
3101 		    else {
3102 			new_deleg->ls_stateid.seqid = delegstateidp->seqid = 1;
3103 			new_deleg->ls_stateid.other[0] = delegstateidp->other[0]
3104 			    = clp->lc_clientid.lval[0];
3105 			new_deleg->ls_stateid.other[1] = delegstateidp->other[1]
3106 			    = clp->lc_clientid.lval[1];
3107 			new_deleg->ls_stateid.other[2] = delegstateidp->other[2]
3108 			    = nfsrv_nextstateindex(clp);
3109 			if (writedeleg && !NFSVNO_EXRDONLY(exp) &&
3110 			    (nfsrv_writedelegifpos || !readonly) &&
3111 			    (new_stp->ls_flags & NFSLCK_WANTRDELEG) == 0) {
3112 			    new_deleg->ls_flags = (NFSLCK_DELEGWRITE |
3113 				NFSLCK_READACCESS | NFSLCK_WRITEACCESS);
3114 			    *rflagsp |= NFSV4OPEN_WRITEDELEGATE;
3115 			    nfsrv_writedelegcnt++;
3116 			} else {
3117 			    new_deleg->ls_flags = (NFSLCK_DELEGREAD |
3118 				NFSLCK_READACCESS);
3119 			    *rflagsp |= NFSV4OPEN_READDELEGATE;
3120 			}
3121 			new_deleg->ls_uid = new_stp->ls_uid;
3122 			new_deleg->ls_lfp = lfp;
3123 			new_deleg->ls_clp = clp;
3124 			new_deleg->ls_filerev = filerev;
3125 			new_deleg->ls_compref = nd->nd_compref;
3126 			LIST_INSERT_HEAD(&lfp->lf_deleg, new_deleg, ls_file);
3127 			LIST_INSERT_HEAD(NFSSTATEHASH(clp,
3128 			    new_deleg->ls_stateid), new_deleg, ls_hash);
3129 			LIST_INSERT_HEAD(&clp->lc_deleg, new_deleg, ls_list);
3130 			new_deleg = NULL;
3131 			nfsstatsv1.srvdelegates++;
3132 			nfsrv_openpluslock++;
3133 			nfsrv_delegatecnt++;
3134 		    }
3135 		}
3136 	} else {
3137 		/*
3138 		 * New owner case. Start the open_owner sequence with a
3139 		 * Needs confirmation (unless a reclaim) and hang the
3140 		 * new open off it.
3141 		 */
3142 		new_open->ls_stateid.seqid = 1;
3143 		new_open->ls_stateid.other[0] = clp->lc_clientid.lval[0];
3144 		new_open->ls_stateid.other[1] = clp->lc_clientid.lval[1];
3145 		new_open->ls_stateid.other[2] = nfsrv_nextstateindex(clp);
3146 		new_open->ls_flags = (new_stp->ls_flags & NFSLCK_SHAREBITS) |
3147 		    NFSLCK_OPEN;
3148 		new_open->ls_uid = new_stp->ls_uid;
3149 		LIST_INIT(&new_open->ls_open);
3150 		new_open->ls_openowner = new_stp;
3151 		new_open->ls_lfp = lfp;
3152 		new_open->ls_clp = clp;
3153 		LIST_INSERT_HEAD(&lfp->lf_open, new_open, ls_file);
3154 		if (new_stp->ls_flags & NFSLCK_RECLAIM) {
3155 			new_stp->ls_flags = 0;
3156 		} else if ((nd->nd_flag & ND_NFSV41) != 0) {
3157 			/* NFSv4.1 never needs confirmation. */
3158 			new_stp->ls_flags = 0;
3159 
3160 			/*
3161 			 * This is where we can choose to issue a delegation.
3162 			 */
3163 			if (delegate && nfsrv_issuedelegs &&
3164 			    (writedeleg || readonly) &&
3165 			    (clp->lc_flags & (LCL_CALLBACKSON | LCL_CBDOWN)) ==
3166 			     LCL_CALLBACKSON &&
3167 			    !NFSRV_V4DELEGLIMIT(nfsrv_delegatecnt) &&
3168 			    NFSVNO_DELEGOK(vp) &&
3169 			    ((nd->nd_flag & ND_NFSV41) == 0 ||
3170 			     (new_stp->ls_flags & NFSLCK_WANTNODELEG) == 0)) {
3171 				new_deleg->ls_stateid.seqid =
3172 				    delegstateidp->seqid = 1;
3173 				new_deleg->ls_stateid.other[0] =
3174 				    delegstateidp->other[0]
3175 				    = clp->lc_clientid.lval[0];
3176 				new_deleg->ls_stateid.other[1] =
3177 				    delegstateidp->other[1]
3178 				    = clp->lc_clientid.lval[1];
3179 				new_deleg->ls_stateid.other[2] =
3180 				    delegstateidp->other[2]
3181 				    = nfsrv_nextstateindex(clp);
3182 				if (writedeleg && !NFSVNO_EXRDONLY(exp) &&
3183 				    (nfsrv_writedelegifpos || !readonly) &&
3184 				    ((nd->nd_flag & ND_NFSV41) == 0 ||
3185 				     (new_stp->ls_flags & NFSLCK_WANTRDELEG) ==
3186 				     0)) {
3187 					new_deleg->ls_flags =
3188 					    (NFSLCK_DELEGWRITE |
3189 					     NFSLCK_READACCESS |
3190 					     NFSLCK_WRITEACCESS);
3191 					*rflagsp |= NFSV4OPEN_WRITEDELEGATE;
3192 					nfsrv_writedelegcnt++;
3193 				} else {
3194 					new_deleg->ls_flags =
3195 					    (NFSLCK_DELEGREAD |
3196 					     NFSLCK_READACCESS);
3197 					*rflagsp |= NFSV4OPEN_READDELEGATE;
3198 				}
3199 				new_deleg->ls_uid = new_stp->ls_uid;
3200 				new_deleg->ls_lfp = lfp;
3201 				new_deleg->ls_clp = clp;
3202 				new_deleg->ls_filerev = filerev;
3203 				new_deleg->ls_compref = nd->nd_compref;
3204 				LIST_INSERT_HEAD(&lfp->lf_deleg, new_deleg,
3205 				    ls_file);
3206 				LIST_INSERT_HEAD(NFSSTATEHASH(clp,
3207 				    new_deleg->ls_stateid), new_deleg, ls_hash);
3208 				LIST_INSERT_HEAD(&clp->lc_deleg, new_deleg,
3209 				    ls_list);
3210 				new_deleg = NULL;
3211 				nfsstatsv1.srvdelegates++;
3212 				nfsrv_openpluslock++;
3213 				nfsrv_delegatecnt++;
3214 			}
3215 			/*
3216 			 * Since NFSv4.1 never does an OpenConfirm, the first
3217 			 * open state will be acquired here.
3218 			 */
3219 			if (!(clp->lc_flags & LCL_STAMPEDSTABLE)) {
3220 				clp->lc_flags |= LCL_STAMPEDSTABLE;
3221 				len = clp->lc_idlen;
3222 				NFSBCOPY(clp->lc_id, clidp, len);
3223 				gotstate = 1;
3224 			}
3225 		} else {
3226 			*rflagsp |= NFSV4OPEN_RESULTCONFIRM;
3227 			new_stp->ls_flags = NFSLCK_NEEDSCONFIRM;
3228 		}
3229 		nfsrvd_refcache(new_stp->ls_op);
3230 		new_stp->ls_noopens = 0;
3231 		LIST_INIT(&new_stp->ls_open);
3232 		LIST_INSERT_HEAD(&new_stp->ls_open, new_open, ls_list);
3233 		LIST_INSERT_HEAD(&clp->lc_open, new_stp, ls_list);
3234 		LIST_INSERT_HEAD(NFSSTATEHASH(clp, new_open->ls_stateid),
3235 		    new_open, ls_hash);
3236 		openstp = new_open;
3237 		new_open = NULL;
3238 		*new_stpp = NULL;
3239 		nfsstatsv1.srvopens++;
3240 		nfsrv_openpluslock++;
3241 		nfsstatsv1.srvopenowners++;
3242 		nfsrv_openpluslock++;
3243 	}
3244 	if (!error) {
3245 		stateidp->seqid = openstp->ls_stateid.seqid;
3246 		stateidp->other[0] = openstp->ls_stateid.other[0];
3247 		stateidp->other[1] = openstp->ls_stateid.other[1];
3248 		stateidp->other[2] = openstp->ls_stateid.other[2];
3249 	}
3250 	NFSUNLOCKSTATE();
3251 	if (haslock) {
3252 		NFSLOCKV4ROOTMUTEX();
3253 		nfsv4_unlock(&nfsv4rootfs_lock, 1);
3254 		NFSUNLOCKV4ROOTMUTEX();
3255 	}
3256 	if (new_open)
3257 		free(new_open, M_NFSDSTATE);
3258 	if (new_deleg)
3259 		free(new_deleg, M_NFSDSTATE);
3260 
3261 	/*
3262 	 * If the NFSv4.1 client just acquired its first open, write a timestamp
3263 	 * to the stable storage file.
3264 	 */
3265 	if (gotstate != 0) {
3266 		nfsrv_writestable(clidp, len, NFSNST_NEWSTATE, p);
3267 		nfsrv_backupstable();
3268 	}
3269 
3270 out:
3271 	free(clidp, M_TEMP);
3272 	NFSEXITCODE2(error, nd);
3273 	return (error);
3274 }
3275 
3276 /*
3277  * Open update. Does the confirm, downgrade and close.
3278  */
3279 APPLESTATIC int
3280 nfsrv_openupdate(vnode_t vp, struct nfsstate *new_stp, nfsquad_t clientid,
3281     nfsv4stateid_t *stateidp, struct nfsrv_descript *nd, NFSPROC_T *p)
3282 {
3283 	struct nfsstate *stp;
3284 	struct nfsclient *clp;
3285 	struct nfslockfile *lfp;
3286 	u_int32_t bits;
3287 	int error = 0, gotstate = 0, len = 0;
3288 	u_char *clidp = NULL;
3289 
3290 	/*
3291 	 * Check for restart conditions (client and server).
3292 	 */
3293 	error = nfsrv_checkrestart(clientid, new_stp->ls_flags,
3294 	    &new_stp->ls_stateid, 0);
3295 	if (error)
3296 		goto out;
3297 
3298 	clidp = malloc(NFSV4_OPAQUELIMIT, M_TEMP, M_WAITOK);
3299 	NFSLOCKSTATE();
3300 	/*
3301 	 * Get the open structure via clientid and stateid.
3302 	 */
3303 	error = nfsrv_getclient(clientid, CLOPS_RENEW, &clp, NULL,
3304 	    (nfsquad_t)((u_quad_t)0), 0, nd, p);
3305 	if (!error)
3306 		error = nfsrv_getstate(clp, &new_stp->ls_stateid,
3307 		    new_stp->ls_flags, &stp);
3308 
3309 	/*
3310 	 * Sanity check the open.
3311 	 */
3312 	if (!error && (!(stp->ls_flags & NFSLCK_OPEN) ||
3313 		(!(new_stp->ls_flags & NFSLCK_CONFIRM) &&
3314 		 (stp->ls_openowner->ls_flags & NFSLCK_NEEDSCONFIRM)) ||
3315 		((new_stp->ls_flags & NFSLCK_CONFIRM) &&
3316 		 (!(stp->ls_openowner->ls_flags & NFSLCK_NEEDSCONFIRM)))))
3317 		error = NFSERR_BADSTATEID;
3318 
3319 	if (!error)
3320 		error = nfsrv_checkseqid(nd, new_stp->ls_seq,
3321 		    stp->ls_openowner, new_stp->ls_op);
3322 	if (!error && stp->ls_stateid.seqid != new_stp->ls_stateid.seqid &&
3323 	    (((nd->nd_flag & ND_NFSV41) == 0 &&
3324 	      !(new_stp->ls_flags & NFSLCK_CONFIRM)) ||
3325 	     ((nd->nd_flag & ND_NFSV41) != 0 &&
3326 	      new_stp->ls_stateid.seqid != 0)))
3327 		error = NFSERR_OLDSTATEID;
3328 	if (!error && vnode_vtype(vp) != VREG) {
3329 		if (vnode_vtype(vp) == VDIR)
3330 			error = NFSERR_ISDIR;
3331 		else
3332 			error = NFSERR_INVAL;
3333 	}
3334 
3335 	if (error) {
3336 		/*
3337 		 * If a client tries to confirm an Open with a bad
3338 		 * seqid# and there are no byte range locks or other Opens
3339 		 * on the openowner, just throw it away, so the next use of the
3340 		 * openowner will start a fresh seq#.
3341 		 */
3342 		if (error == NFSERR_BADSEQID &&
3343 		    (new_stp->ls_flags & NFSLCK_CONFIRM) &&
3344 		    nfsrv_nootherstate(stp))
3345 			nfsrv_freeopenowner(stp->ls_openowner, 0, p);
3346 		NFSUNLOCKSTATE();
3347 		goto out;
3348 	}
3349 
3350 	/*
3351 	 * Set the return stateid.
3352 	 */
3353 	stateidp->seqid = stp->ls_stateid.seqid + 1;
3354 	if ((nd->nd_flag & ND_NFSV41) != 0 && stateidp->seqid == 0)
3355 		stateidp->seqid = 1;
3356 	stateidp->other[0] = stp->ls_stateid.other[0];
3357 	stateidp->other[1] = stp->ls_stateid.other[1];
3358 	stateidp->other[2] = stp->ls_stateid.other[2];
3359 	/*
3360 	 * Now, handle the three cases.
3361 	 */
3362 	if (new_stp->ls_flags & NFSLCK_CONFIRM) {
3363 		/*
3364 		 * If the open doesn't need confirmation, it seems to me that
3365 		 * there is a client error, but I'll just log it and keep going?
3366 		 */
3367 		if (!(stp->ls_openowner->ls_flags & NFSLCK_NEEDSCONFIRM))
3368 			printf("Nfsv4d: stray open confirm\n");
3369 		stp->ls_openowner->ls_flags = 0;
3370 		stp->ls_stateid.seqid++;
3371 		if ((nd->nd_flag & ND_NFSV41) != 0 &&
3372 		    stp->ls_stateid.seqid == 0)
3373 			stp->ls_stateid.seqid = 1;
3374 		if (!(clp->lc_flags & LCL_STAMPEDSTABLE)) {
3375 			clp->lc_flags |= LCL_STAMPEDSTABLE;
3376 			len = clp->lc_idlen;
3377 			NFSBCOPY(clp->lc_id, clidp, len);
3378 			gotstate = 1;
3379 		}
3380 		NFSUNLOCKSTATE();
3381 	} else if (new_stp->ls_flags & NFSLCK_CLOSE) {
3382 		lfp = stp->ls_lfp;
3383 		if (nfsrv_dolocallocks != 0 && !LIST_EMPTY(&stp->ls_open)) {
3384 			/* Get the lf lock */
3385 			nfsrv_locklf(lfp);
3386 			NFSUNLOCKSTATE();
3387 			ASSERT_VOP_ELOCKED(vp, "nfsrv_openupdate");
3388 			NFSVOPUNLOCK(vp, 0);
3389 			if (nfsrv_freeopen(stp, vp, 1, p) == 0) {
3390 				NFSLOCKSTATE();
3391 				nfsrv_unlocklf(lfp);
3392 				NFSUNLOCKSTATE();
3393 			}
3394 			NFSVOPLOCK(vp, LK_EXCLUSIVE | LK_RETRY);
3395 		} else {
3396 			(void) nfsrv_freeopen(stp, NULL, 0, p);
3397 			NFSUNLOCKSTATE();
3398 		}
3399 	} else {
3400 		/*
3401 		 * Update the share bits, making sure that the new set are a
3402 		 * subset of the old ones.
3403 		 */
3404 		bits = (new_stp->ls_flags & NFSLCK_SHAREBITS);
3405 		if (~(stp->ls_flags) & bits) {
3406 			NFSUNLOCKSTATE();
3407 			error = NFSERR_INVAL;
3408 			goto out;
3409 		}
3410 		stp->ls_flags = (bits | NFSLCK_OPEN);
3411 		stp->ls_stateid.seqid++;
3412 		if ((nd->nd_flag & ND_NFSV41) != 0 &&
3413 		    stp->ls_stateid.seqid == 0)
3414 			stp->ls_stateid.seqid = 1;
3415 		NFSUNLOCKSTATE();
3416 	}
3417 
3418 	/*
3419 	 * If the client just confirmed its first open, write a timestamp
3420 	 * to the stable storage file.
3421 	 */
3422 	if (gotstate != 0) {
3423 		nfsrv_writestable(clidp, len, NFSNST_NEWSTATE, p);
3424 		nfsrv_backupstable();
3425 	}
3426 
3427 out:
3428 	free(clidp, M_TEMP);
3429 	NFSEXITCODE2(error, nd);
3430 	return (error);
3431 }
3432 
3433 /*
3434  * Delegation update. Does the purge and return.
3435  */
3436 APPLESTATIC int
3437 nfsrv_delegupdate(struct nfsrv_descript *nd, nfsquad_t clientid,
3438     nfsv4stateid_t *stateidp, vnode_t vp, int op, struct ucred *cred,
3439     NFSPROC_T *p)
3440 {
3441 	struct nfsstate *stp;
3442 	struct nfsclient *clp;
3443 	int error = 0;
3444 	fhandle_t fh;
3445 
3446 	/*
3447 	 * Do a sanity check against the file handle for DelegReturn.
3448 	 */
3449 	if (vp) {
3450 		error = nfsvno_getfh(vp, &fh, p);
3451 		if (error)
3452 			goto out;
3453 	}
3454 	/*
3455 	 * Check for restart conditions (client and server).
3456 	 */
3457 	if (op == NFSV4OP_DELEGRETURN)
3458 		error = nfsrv_checkrestart(clientid, NFSLCK_DELEGRETURN,
3459 			stateidp, 0);
3460 	else
3461 		error = nfsrv_checkrestart(clientid, NFSLCK_DELEGPURGE,
3462 			stateidp, 0);
3463 
3464 	NFSLOCKSTATE();
3465 	/*
3466 	 * Get the open structure via clientid and stateid.
3467 	 */
3468 	if (!error)
3469 	    error = nfsrv_getclient(clientid, CLOPS_RENEW, &clp, NULL,
3470 		(nfsquad_t)((u_quad_t)0), 0, nd, p);
3471 	if (error) {
3472 		if (error == NFSERR_CBPATHDOWN)
3473 			error = 0;
3474 		if (error == NFSERR_STALECLIENTID && op == NFSV4OP_DELEGRETURN)
3475 			error = NFSERR_STALESTATEID;
3476 	}
3477 	if (!error && op == NFSV4OP_DELEGRETURN) {
3478 	    error = nfsrv_getstate(clp, stateidp, NFSLCK_DELEGRETURN, &stp);
3479 	    if (!error && stp->ls_stateid.seqid != stateidp->seqid &&
3480 		((nd->nd_flag & ND_NFSV41) == 0 || stateidp->seqid != 0))
3481 		error = NFSERR_OLDSTATEID;
3482 	}
3483 	/*
3484 	 * NFSERR_EXPIRED means that the state has gone away,
3485 	 * so Delegations have been purged. Just return ok.
3486 	 */
3487 	if (error == NFSERR_EXPIRED && op == NFSV4OP_DELEGPURGE) {
3488 		NFSUNLOCKSTATE();
3489 		error = 0;
3490 		goto out;
3491 	}
3492 	if (error) {
3493 		NFSUNLOCKSTATE();
3494 		goto out;
3495 	}
3496 
3497 	if (op == NFSV4OP_DELEGRETURN) {
3498 		if (NFSBCMP((caddr_t)&fh, (caddr_t)&stp->ls_lfp->lf_fh,
3499 		    sizeof (fhandle_t))) {
3500 			NFSUNLOCKSTATE();
3501 			error = NFSERR_BADSTATEID;
3502 			goto out;
3503 		}
3504 		nfsrv_freedeleg(stp);
3505 	} else {
3506 		nfsrv_freedeleglist(&clp->lc_olddeleg);
3507 	}
3508 	NFSUNLOCKSTATE();
3509 	error = 0;
3510 
3511 out:
3512 	NFSEXITCODE(error);
3513 	return (error);
3514 }
3515 
3516 /*
3517  * Release lock owner.
3518  */
3519 APPLESTATIC int
3520 nfsrv_releaselckown(struct nfsstate *new_stp, nfsquad_t clientid,
3521     NFSPROC_T *p)
3522 {
3523 	struct nfsstate *stp, *nstp, *openstp, *ownstp;
3524 	struct nfsclient *clp;
3525 	int error = 0;
3526 
3527 	/*
3528 	 * Check for restart conditions (client and server).
3529 	 */
3530 	error = nfsrv_checkrestart(clientid, new_stp->ls_flags,
3531 	    &new_stp->ls_stateid, 0);
3532 	if (error)
3533 		goto out;
3534 
3535 	NFSLOCKSTATE();
3536 	/*
3537 	 * Get the lock owner by name.
3538 	 */
3539 	error = nfsrv_getclient(clientid, CLOPS_RENEW, &clp, NULL,
3540 	    (nfsquad_t)((u_quad_t)0), 0, NULL, p);
3541 	if (error) {
3542 		NFSUNLOCKSTATE();
3543 		goto out;
3544 	}
3545 	LIST_FOREACH(ownstp, &clp->lc_open, ls_list) {
3546 	    LIST_FOREACH(openstp, &ownstp->ls_open, ls_list) {
3547 		stp = LIST_FIRST(&openstp->ls_open);
3548 		while (stp != LIST_END(&openstp->ls_open)) {
3549 		    nstp = LIST_NEXT(stp, ls_list);
3550 		    /*
3551 		     * If the owner matches, check for locks and
3552 		     * then free or return an error.
3553 		     */
3554 		    if (stp->ls_ownerlen == new_stp->ls_ownerlen &&
3555 			!NFSBCMP(stp->ls_owner, new_stp->ls_owner,
3556 			 stp->ls_ownerlen)){
3557 			if (LIST_EMPTY(&stp->ls_lock)) {
3558 			    nfsrv_freelockowner(stp, NULL, 0, p);
3559 			} else {
3560 			    NFSUNLOCKSTATE();
3561 			    error = NFSERR_LOCKSHELD;
3562 			    goto out;
3563 			}
3564 		    }
3565 		    stp = nstp;
3566 		}
3567 	    }
3568 	}
3569 	NFSUNLOCKSTATE();
3570 
3571 out:
3572 	NFSEXITCODE(error);
3573 	return (error);
3574 }
3575 
3576 /*
3577  * Get the file handle for a lock structure.
3578  */
3579 static int
3580 nfsrv_getlockfh(vnode_t vp, u_short flags, struct nfslockfile *new_lfp,
3581     fhandle_t *nfhp, NFSPROC_T *p)
3582 {
3583 	fhandle_t *fhp = NULL;
3584 	int error;
3585 
3586 	/*
3587 	 * For lock, use the new nfslock structure, otherwise just
3588 	 * a fhandle_t on the stack.
3589 	 */
3590 	if (flags & NFSLCK_OPEN) {
3591 		KASSERT(new_lfp != NULL, ("nfsrv_getlockfh: new_lfp NULL"));
3592 		fhp = &new_lfp->lf_fh;
3593 	} else if (nfhp) {
3594 		fhp = nfhp;
3595 	} else {
3596 		panic("nfsrv_getlockfh");
3597 	}
3598 	error = nfsvno_getfh(vp, fhp, p);
3599 	NFSEXITCODE(error);
3600 	return (error);
3601 }
3602 
3603 /*
3604  * Get an nfs lock structure. Allocate one, as required, and return a
3605  * pointer to it.
3606  * Returns an NFSERR_xxx upon failure or -1 to indicate no current lock.
3607  */
3608 static int
3609 nfsrv_getlockfile(u_short flags, struct nfslockfile **new_lfpp,
3610     struct nfslockfile **lfpp, fhandle_t *nfhp, int lockit)
3611 {
3612 	struct nfslockfile *lfp;
3613 	fhandle_t *fhp = NULL, *tfhp;
3614 	struct nfslockhashhead *hp;
3615 	struct nfslockfile *new_lfp = NULL;
3616 
3617 	/*
3618 	 * For lock, use the new nfslock structure, otherwise just
3619 	 * a fhandle_t on the stack.
3620 	 */
3621 	if (flags & NFSLCK_OPEN) {
3622 		new_lfp = *new_lfpp;
3623 		fhp = &new_lfp->lf_fh;
3624 	} else if (nfhp) {
3625 		fhp = nfhp;
3626 	} else {
3627 		panic("nfsrv_getlockfile");
3628 	}
3629 
3630 	hp = NFSLOCKHASH(fhp);
3631 	LIST_FOREACH(lfp, hp, lf_hash) {
3632 		tfhp = &lfp->lf_fh;
3633 		if (NFSVNO_CMPFH(fhp, tfhp)) {
3634 			if (lockit)
3635 				nfsrv_locklf(lfp);
3636 			*lfpp = lfp;
3637 			return (0);
3638 		}
3639 	}
3640 	if (!(flags & NFSLCK_OPEN))
3641 		return (-1);
3642 
3643 	/*
3644 	 * No match, so chain the new one into the list.
3645 	 */
3646 	LIST_INIT(&new_lfp->lf_open);
3647 	LIST_INIT(&new_lfp->lf_lock);
3648 	LIST_INIT(&new_lfp->lf_deleg);
3649 	LIST_INIT(&new_lfp->lf_locallock);
3650 	LIST_INIT(&new_lfp->lf_rollback);
3651 	new_lfp->lf_locallock_lck.nfslock_usecnt = 0;
3652 	new_lfp->lf_locallock_lck.nfslock_lock = 0;
3653 	new_lfp->lf_usecount = 0;
3654 	LIST_INSERT_HEAD(hp, new_lfp, lf_hash);
3655 	*lfpp = new_lfp;
3656 	*new_lfpp = NULL;
3657 	return (0);
3658 }
3659 
3660 /*
3661  * This function adds a nfslock lock structure to the list for the associated
3662  * nfsstate and nfslockfile structures. It will be inserted after the
3663  * entry pointed at by insert_lop.
3664  */
3665 static void
3666 nfsrv_insertlock(struct nfslock *new_lop, struct nfslock *insert_lop,
3667     struct nfsstate *stp, struct nfslockfile *lfp)
3668 {
3669 	struct nfslock *lop, *nlop;
3670 
3671 	new_lop->lo_stp = stp;
3672 	new_lop->lo_lfp = lfp;
3673 
3674 	if (stp != NULL) {
3675 		/* Insert in increasing lo_first order */
3676 		lop = LIST_FIRST(&lfp->lf_lock);
3677 		if (lop == LIST_END(&lfp->lf_lock) ||
3678 		    new_lop->lo_first <= lop->lo_first) {
3679 			LIST_INSERT_HEAD(&lfp->lf_lock, new_lop, lo_lckfile);
3680 		} else {
3681 			nlop = LIST_NEXT(lop, lo_lckfile);
3682 			while (nlop != LIST_END(&lfp->lf_lock) &&
3683 			       nlop->lo_first < new_lop->lo_first) {
3684 				lop = nlop;
3685 				nlop = LIST_NEXT(lop, lo_lckfile);
3686 			}
3687 			LIST_INSERT_AFTER(lop, new_lop, lo_lckfile);
3688 		}
3689 	} else {
3690 		new_lop->lo_lckfile.le_prev = NULL;	/* list not used */
3691 	}
3692 
3693 	/*
3694 	 * Insert after insert_lop, which is overloaded as stp or lfp for
3695 	 * an empty list.
3696 	 */
3697 	if (stp == NULL && (struct nfslockfile *)insert_lop == lfp)
3698 		LIST_INSERT_HEAD(&lfp->lf_locallock, new_lop, lo_lckowner);
3699 	else if ((struct nfsstate *)insert_lop == stp)
3700 		LIST_INSERT_HEAD(&stp->ls_lock, new_lop, lo_lckowner);
3701 	else
3702 		LIST_INSERT_AFTER(insert_lop, new_lop, lo_lckowner);
3703 	if (stp != NULL) {
3704 		nfsstatsv1.srvlocks++;
3705 		nfsrv_openpluslock++;
3706 	}
3707 }
3708 
3709 /*
3710  * This function updates the locking for a lock owner and given file. It
3711  * maintains a list of lock ranges ordered on increasing file offset that
3712  * are NFSLCK_READ or NFSLCK_WRITE and non-overlapping (aka POSIX style).
3713  * It always adds new_lop to the list and sometimes uses the one pointed
3714  * at by other_lopp.
3715  */
3716 static void
3717 nfsrv_updatelock(struct nfsstate *stp, struct nfslock **new_lopp,
3718     struct nfslock **other_lopp, struct nfslockfile *lfp)
3719 {
3720 	struct nfslock *new_lop = *new_lopp;
3721 	struct nfslock *lop, *tlop, *ilop;
3722 	struct nfslock *other_lop = *other_lopp;
3723 	int unlock = 0, myfile = 0;
3724 	u_int64_t tmp;
3725 
3726 	/*
3727 	 * Work down the list until the lock is merged.
3728 	 */
3729 	if (new_lop->lo_flags & NFSLCK_UNLOCK)
3730 		unlock = 1;
3731 	if (stp != NULL) {
3732 		ilop = (struct nfslock *)stp;
3733 		lop = LIST_FIRST(&stp->ls_lock);
3734 	} else {
3735 		ilop = (struct nfslock *)lfp;
3736 		lop = LIST_FIRST(&lfp->lf_locallock);
3737 	}
3738 	while (lop != NULL) {
3739 	    /*
3740 	     * Only check locks for this file that aren't before the start of
3741 	     * new lock's range.
3742 	     */
3743 	    if (lop->lo_lfp == lfp) {
3744 	      myfile = 1;
3745 	      if (lop->lo_end >= new_lop->lo_first) {
3746 		if (new_lop->lo_end < lop->lo_first) {
3747 			/*
3748 			 * If the new lock ends before the start of the
3749 			 * current lock's range, no merge, just insert
3750 			 * the new lock.
3751 			 */
3752 			break;
3753 		}
3754 		if (new_lop->lo_flags == lop->lo_flags ||
3755 		    (new_lop->lo_first <= lop->lo_first &&
3756 		     new_lop->lo_end >= lop->lo_end)) {
3757 			/*
3758 			 * This lock can be absorbed by the new lock/unlock.
3759 			 * This happens when it covers the entire range
3760 			 * of the old lock or is contiguous
3761 			 * with the old lock and is of the same type or an
3762 			 * unlock.
3763 			 */
3764 			if (lop->lo_first < new_lop->lo_first)
3765 				new_lop->lo_first = lop->lo_first;
3766 			if (lop->lo_end > new_lop->lo_end)
3767 				new_lop->lo_end = lop->lo_end;
3768 			tlop = lop;
3769 			lop = LIST_NEXT(lop, lo_lckowner);
3770 			nfsrv_freenfslock(tlop);
3771 			continue;
3772 		}
3773 
3774 		/*
3775 		 * All these cases are for contiguous locks that are not the
3776 		 * same type, so they can't be merged.
3777 		 */
3778 		if (new_lop->lo_first <= lop->lo_first) {
3779 			/*
3780 			 * This case is where the new lock overlaps with the
3781 			 * first part of the old lock. Move the start of the
3782 			 * old lock to just past the end of the new lock. The
3783 			 * new lock will be inserted in front of the old, since
3784 			 * ilop hasn't been updated. (We are done now.)
3785 			 */
3786 			lop->lo_first = new_lop->lo_end;
3787 			break;
3788 		}
3789 		if (new_lop->lo_end >= lop->lo_end) {
3790 			/*
3791 			 * This case is where the new lock overlaps with the
3792 			 * end of the old lock's range. Move the old lock's
3793 			 * end to just before the new lock's first and insert
3794 			 * the new lock after the old lock.
3795 			 * Might not be done yet, since the new lock could
3796 			 * overlap further locks with higher ranges.
3797 			 */
3798 			lop->lo_end = new_lop->lo_first;
3799 			ilop = lop;
3800 			lop = LIST_NEXT(lop, lo_lckowner);
3801 			continue;
3802 		}
3803 		/*
3804 		 * The final case is where the new lock's range is in the
3805 		 * middle of the current lock's and splits the current lock
3806 		 * up. Use *other_lopp to handle the second part of the
3807 		 * split old lock range. (We are done now.)
3808 		 * For unlock, we use new_lop as other_lop and tmp, since
3809 		 * other_lop and new_lop are the same for this case.
3810 		 * We noted the unlock case above, so we don't need
3811 		 * new_lop->lo_flags any longer.
3812 		 */
3813 		tmp = new_lop->lo_first;
3814 		if (other_lop == NULL) {
3815 			if (!unlock)
3816 				panic("nfsd srv update unlock");
3817 			other_lop = new_lop;
3818 			*new_lopp = NULL;
3819 		}
3820 		other_lop->lo_first = new_lop->lo_end;
3821 		other_lop->lo_end = lop->lo_end;
3822 		other_lop->lo_flags = lop->lo_flags;
3823 		other_lop->lo_stp = stp;
3824 		other_lop->lo_lfp = lfp;
3825 		lop->lo_end = tmp;
3826 		nfsrv_insertlock(other_lop, lop, stp, lfp);
3827 		*other_lopp = NULL;
3828 		ilop = lop;
3829 		break;
3830 	      }
3831 	    }
3832 	    ilop = lop;
3833 	    lop = LIST_NEXT(lop, lo_lckowner);
3834 	    if (myfile && (lop == NULL || lop->lo_lfp != lfp))
3835 		break;
3836 	}
3837 
3838 	/*
3839 	 * Insert the new lock in the list at the appropriate place.
3840 	 */
3841 	if (!unlock) {
3842 		nfsrv_insertlock(new_lop, ilop, stp, lfp);
3843 		*new_lopp = NULL;
3844 	}
3845 }
3846 
3847 /*
3848  * This function handles sequencing of locks, etc.
3849  * It returns an error that indicates what the caller should do.
3850  */
3851 static int
3852 nfsrv_checkseqid(struct nfsrv_descript *nd, u_int32_t seqid,
3853     struct nfsstate *stp, struct nfsrvcache *op)
3854 {
3855 	int error = 0;
3856 
3857 	if ((nd->nd_flag & ND_NFSV41) != 0)
3858 		/* NFSv4.1 ignores the open_seqid and lock_seqid. */
3859 		goto out;
3860 	if (op != nd->nd_rp)
3861 		panic("nfsrvstate checkseqid");
3862 	if (!(op->rc_flag & RC_INPROG))
3863 		panic("nfsrvstate not inprog");
3864 	if (stp->ls_op && stp->ls_op->rc_refcnt <= 0) {
3865 		printf("refcnt=%d\n", stp->ls_op->rc_refcnt);
3866 		panic("nfsrvstate op refcnt");
3867 	}
3868 	if ((stp->ls_seq + 1) == seqid) {
3869 		if (stp->ls_op)
3870 			nfsrvd_derefcache(stp->ls_op);
3871 		stp->ls_op = op;
3872 		nfsrvd_refcache(op);
3873 		stp->ls_seq = seqid;
3874 		goto out;
3875 	} else if (stp->ls_seq == seqid && stp->ls_op &&
3876 		op->rc_xid == stp->ls_op->rc_xid &&
3877 		op->rc_refcnt == 0 &&
3878 		op->rc_reqlen == stp->ls_op->rc_reqlen &&
3879 		op->rc_cksum == stp->ls_op->rc_cksum) {
3880 		if (stp->ls_op->rc_flag & RC_INPROG) {
3881 			error = NFSERR_DONTREPLY;
3882 			goto out;
3883 		}
3884 		nd->nd_rp = stp->ls_op;
3885 		nd->nd_rp->rc_flag |= RC_INPROG;
3886 		nfsrvd_delcache(op);
3887 		error = NFSERR_REPLYFROMCACHE;
3888 		goto out;
3889 	}
3890 	error = NFSERR_BADSEQID;
3891 
3892 out:
3893 	NFSEXITCODE2(error, nd);
3894 	return (error);
3895 }
3896 
3897 /*
3898  * Get the client ip address for callbacks. If the strings can't be parsed,
3899  * just set lc_program to 0 to indicate no callbacks are possible.
3900  * (For cases where the address can't be parsed or is 0.0.0.0.0.0, set
3901  *  the address to the client's transport address. This won't be used
3902  *  for callbacks, but can be printed out by nfsstats for info.)
3903  * Return error if the xdr can't be parsed, 0 otherwise.
3904  */
3905 APPLESTATIC int
3906 nfsrv_getclientipaddr(struct nfsrv_descript *nd, struct nfsclient *clp)
3907 {
3908 	u_int32_t *tl;
3909 	u_char *cp, *cp2;
3910 	int i, j;
3911 	struct sockaddr_in *rad, *sad;
3912 	u_char protocol[5], addr[24];
3913 	int error = 0, cantparse = 0;
3914 	union {
3915 		in_addr_t ival;
3916 		u_char cval[4];
3917 	} ip;
3918 	union {
3919 		in_port_t sval;
3920 		u_char cval[2];
3921 	} port;
3922 
3923 	rad = NFSSOCKADDR(clp->lc_req.nr_nam, struct sockaddr_in *);
3924 	rad->sin_family = AF_INET;
3925 	rad->sin_len = sizeof (struct sockaddr_in);
3926 	rad->sin_addr.s_addr = 0;
3927 	rad->sin_port = 0;
3928 	clp->lc_req.nr_client = NULL;
3929 	clp->lc_req.nr_lock = 0;
3930 	NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
3931 	i = fxdr_unsigned(int, *tl);
3932 	if (i >= 3 && i <= 4) {
3933 		error = nfsrv_mtostr(nd, protocol, i);
3934 		if (error)
3935 			goto nfsmout;
3936 		if (!strcmp(protocol, "tcp")) {
3937 			clp->lc_flags |= LCL_TCPCALLBACK;
3938 			clp->lc_req.nr_sotype = SOCK_STREAM;
3939 			clp->lc_req.nr_soproto = IPPROTO_TCP;
3940 		} else if (!strcmp(protocol, "udp")) {
3941 			clp->lc_req.nr_sotype = SOCK_DGRAM;
3942 			clp->lc_req.nr_soproto = IPPROTO_UDP;
3943 		} else {
3944 			cantparse = 1;
3945 		}
3946 	} else {
3947 		cantparse = 1;
3948 		if (i > 0) {
3949 			error = nfsm_advance(nd, NFSM_RNDUP(i), -1);
3950 			if (error)
3951 				goto nfsmout;
3952 		}
3953 	}
3954 	NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
3955 	i = fxdr_unsigned(int, *tl);
3956 	if (i < 0) {
3957 		error = NFSERR_BADXDR;
3958 		goto nfsmout;
3959 	} else if (i == 0) {
3960 		cantparse = 1;
3961 	} else if (!cantparse && i <= 23 && i >= 11) {
3962 		error = nfsrv_mtostr(nd, addr, i);
3963 		if (error)
3964 			goto nfsmout;
3965 
3966 		/*
3967 		 * Parse out the address fields. We expect 6 decimal numbers
3968 		 * separated by '.'s.
3969 		 */
3970 		cp = addr;
3971 		i = 0;
3972 		while (*cp && i < 6) {
3973 			cp2 = cp;
3974 			while (*cp2 && *cp2 != '.')
3975 				cp2++;
3976 			if (*cp2)
3977 				*cp2++ = '\0';
3978 			else if (i != 5) {
3979 				cantparse = 1;
3980 				break;
3981 			}
3982 			j = nfsrv_getipnumber(cp);
3983 			if (j >= 0) {
3984 				if (i < 4)
3985 					ip.cval[3 - i] = j;
3986 				else
3987 					port.cval[5 - i] = j;
3988 			} else {
3989 				cantparse = 1;
3990 				break;
3991 			}
3992 			cp = cp2;
3993 			i++;
3994 		}
3995 		if (!cantparse) {
3996 			if (ip.ival != 0x0) {
3997 				rad->sin_addr.s_addr = htonl(ip.ival);
3998 				rad->sin_port = htons(port.sval);
3999 			} else {
4000 				cantparse = 1;
4001 			}
4002 		}
4003 	} else {
4004 		cantparse = 1;
4005 		if (i > 0) {
4006 			error = nfsm_advance(nd, NFSM_RNDUP(i), -1);
4007 			if (error)
4008 				goto nfsmout;
4009 		}
4010 	}
4011 	if (cantparse) {
4012 		sad = NFSSOCKADDR(nd->nd_nam, struct sockaddr_in *);
4013 		if (sad->sin_family == AF_INET) {
4014 			rad->sin_addr.s_addr = sad->sin_addr.s_addr;
4015 			rad->sin_port = 0x0;
4016 		}
4017 		clp->lc_program = 0;
4018 	}
4019 nfsmout:
4020 	NFSEXITCODE2(error, nd);
4021 	return (error);
4022 }
4023 
4024 /*
4025  * Turn a string of up to three decimal digits into a number. Return -1 upon
4026  * error.
4027  */
4028 static int
4029 nfsrv_getipnumber(u_char *cp)
4030 {
4031 	int i = 0, j = 0;
4032 
4033 	while (*cp) {
4034 		if (j > 2 || *cp < '0' || *cp > '9')
4035 			return (-1);
4036 		i *= 10;
4037 		i += (*cp - '0');
4038 		cp++;
4039 		j++;
4040 	}
4041 	if (i < 256)
4042 		return (i);
4043 	return (-1);
4044 }
4045 
4046 /*
4047  * This function checks for restart conditions.
4048  */
4049 static int
4050 nfsrv_checkrestart(nfsquad_t clientid, u_int32_t flags,
4051     nfsv4stateid_t *stateidp, int specialid)
4052 {
4053 	int ret = 0;
4054 
4055 	/*
4056 	 * First check for a server restart. Open, LockT, ReleaseLockOwner
4057 	 * and DelegPurge have a clientid, the rest a stateid.
4058 	 */
4059 	if (flags &
4060 	    (NFSLCK_OPEN | NFSLCK_TEST | NFSLCK_RELEASE | NFSLCK_DELEGPURGE)) {
4061 		if (clientid.lval[0] != nfsrvboottime) {
4062 			ret = NFSERR_STALECLIENTID;
4063 			goto out;
4064 		}
4065 	} else if (stateidp->other[0] != nfsrvboottime &&
4066 		specialid == 0) {
4067 		ret = NFSERR_STALESTATEID;
4068 		goto out;
4069 	}
4070 
4071 	/*
4072 	 * Read, Write, Setattr and LockT can return NFSERR_GRACE and do
4073 	 * not use a lock/open owner seqid#, so the check can be done now.
4074 	 * (The others will be checked, as required, later.)
4075 	 */
4076 	if (!(flags & (NFSLCK_CHECK | NFSLCK_TEST)))
4077 		goto out;
4078 
4079 	NFSLOCKSTATE();
4080 	ret = nfsrv_checkgrace(NULL, NULL, flags);
4081 	NFSUNLOCKSTATE();
4082 
4083 out:
4084 	NFSEXITCODE(ret);
4085 	return (ret);
4086 }
4087 
4088 /*
4089  * Check for grace.
4090  */
4091 static int
4092 nfsrv_checkgrace(struct nfsrv_descript *nd, struct nfsclient *clp,
4093     u_int32_t flags)
4094 {
4095 	int error = 0, notreclaimed;
4096 	struct nfsrv_stable *sp;
4097 
4098 	if ((nfsrv_stablefirst.nsf_flags & (NFSNSF_UPDATEDONE |
4099 	     NFSNSF_GRACEOVER)) == 0) {
4100 		/*
4101 		 * First, check to see if all of the clients have done a
4102 		 * ReclaimComplete.  If so, grace can end now.
4103 		 */
4104 		notreclaimed = 0;
4105 		LIST_FOREACH(sp, &nfsrv_stablefirst.nsf_head, nst_list) {
4106 			if ((sp->nst_flag & NFSNST_RECLAIMED) == 0) {
4107 				notreclaimed = 1;
4108 				break;
4109 			}
4110 		}
4111 		if (notreclaimed == 0)
4112 			nfsrv_stablefirst.nsf_flags |= (NFSNSF_GRACEOVER |
4113 			    NFSNSF_NEEDLOCK);
4114 	}
4115 
4116 	if ((nfsrv_stablefirst.nsf_flags & NFSNSF_GRACEOVER) != 0) {
4117 		if (flags & NFSLCK_RECLAIM) {
4118 			error = NFSERR_NOGRACE;
4119 			goto out;
4120 		}
4121 	} else {
4122 		if (!(flags & NFSLCK_RECLAIM)) {
4123 			error = NFSERR_GRACE;
4124 			goto out;
4125 		}
4126 		if (nd != NULL && clp != NULL &&
4127 		    (nd->nd_flag & ND_NFSV41) != 0 &&
4128 		    (clp->lc_flags & LCL_RECLAIMCOMPLETE) != 0) {
4129 			error = NFSERR_NOGRACE;
4130 			goto out;
4131 		}
4132 
4133 		/*
4134 		 * If grace is almost over and we are still getting Reclaims,
4135 		 * extend grace a bit.
4136 		 */
4137 		if ((NFSD_MONOSEC + NFSRV_LEASEDELTA) >
4138 		    nfsrv_stablefirst.nsf_eograce)
4139 			nfsrv_stablefirst.nsf_eograce = NFSD_MONOSEC +
4140 				NFSRV_LEASEDELTA;
4141 	}
4142 
4143 out:
4144 	NFSEXITCODE(error);
4145 	return (error);
4146 }
4147 
4148 /*
4149  * Do a server callback.
4150  */
4151 static int
4152 nfsrv_docallback(struct nfsclient *clp, int procnum,
4153     nfsv4stateid_t *stateidp, int trunc, fhandle_t *fhp,
4154     struct nfsvattr *nap, nfsattrbit_t *attrbitp, NFSPROC_T *p)
4155 {
4156 	mbuf_t m;
4157 	u_int32_t *tl;
4158 	struct nfsrv_descript nfsd, *nd = &nfsd;
4159 	struct ucred *cred;
4160 	int error = 0;
4161 	u_int32_t callback;
4162 	struct nfsdsession *sep = NULL;
4163 
4164 	cred = newnfs_getcred();
4165 	NFSLOCKSTATE();	/* mostly for lc_cbref++ */
4166 	if (clp->lc_flags & LCL_NEEDSCONFIRM) {
4167 		NFSUNLOCKSTATE();
4168 		panic("docallb");
4169 	}
4170 	clp->lc_cbref++;
4171 
4172 	/*
4173 	 * Fill the callback program# and version into the request
4174 	 * structure for newnfs_connect() to use.
4175 	 */
4176 	clp->lc_req.nr_prog = clp->lc_program;
4177 #ifdef notnow
4178 	if ((clp->lc_flags & LCL_NFSV41) != 0)
4179 		clp->lc_req.nr_vers = NFSV41_CBVERS;
4180 	else
4181 #endif
4182 		clp->lc_req.nr_vers = NFSV4_CBVERS;
4183 
4184 	/*
4185 	 * First, fill in some of the fields of nd and cr.
4186 	 */
4187 	nd->nd_flag = ND_NFSV4;
4188 	if (clp->lc_flags & LCL_GSS)
4189 		nd->nd_flag |= ND_KERBV;
4190 	if ((clp->lc_flags & LCL_NFSV41) != 0)
4191 		nd->nd_flag |= ND_NFSV41;
4192 	nd->nd_repstat = 0;
4193 	cred->cr_uid = clp->lc_uid;
4194 	cred->cr_gid = clp->lc_gid;
4195 	callback = clp->lc_callback;
4196 	NFSUNLOCKSTATE();
4197 	cred->cr_ngroups = 1;
4198 
4199 	/*
4200 	 * Get the first mbuf for the request.
4201 	 */
4202 	MGET(m, M_WAITOK, MT_DATA);
4203 	mbuf_setlen(m, 0);
4204 	nd->nd_mreq = nd->nd_mb = m;
4205 	nd->nd_bpos = NFSMTOD(m, caddr_t);
4206 
4207 	/*
4208 	 * and build the callback request.
4209 	 */
4210 	if (procnum == NFSV4OP_CBGETATTR) {
4211 		nd->nd_procnum = NFSV4PROC_CBCOMPOUND;
4212 		error = nfsrv_cbcallargs(nd, clp, callback, NFSV4OP_CBGETATTR,
4213 		    "CB Getattr", &sep);
4214 		if (error != 0) {
4215 			mbuf_freem(nd->nd_mreq);
4216 			goto errout;
4217 		}
4218 		(void)nfsm_fhtom(nd, (u_int8_t *)fhp, NFSX_MYFH, 0);
4219 		(void)nfsrv_putattrbit(nd, attrbitp);
4220 	} else if (procnum == NFSV4OP_CBRECALL) {
4221 		nd->nd_procnum = NFSV4PROC_CBCOMPOUND;
4222 		error = nfsrv_cbcallargs(nd, clp, callback, NFSV4OP_CBRECALL,
4223 		    "CB Recall", &sep);
4224 		if (error != 0) {
4225 			mbuf_freem(nd->nd_mreq);
4226 			goto errout;
4227 		}
4228 		NFSM_BUILD(tl, u_int32_t *, NFSX_UNSIGNED + NFSX_STATEID);
4229 		*tl++ = txdr_unsigned(stateidp->seqid);
4230 		NFSBCOPY((caddr_t)stateidp->other, (caddr_t)tl,
4231 		    NFSX_STATEIDOTHER);
4232 		tl += (NFSX_STATEIDOTHER / NFSX_UNSIGNED);
4233 		if (trunc)
4234 			*tl = newnfs_true;
4235 		else
4236 			*tl = newnfs_false;
4237 		(void)nfsm_fhtom(nd, (u_int8_t *)fhp, NFSX_MYFH, 0);
4238 	} else if (procnum == NFSV4PROC_CBNULL) {
4239 		nd->nd_procnum = NFSV4PROC_CBNULL;
4240 		if ((clp->lc_flags & LCL_NFSV41) != 0) {
4241 			error = nfsv4_getcbsession(clp, &sep);
4242 			if (error != 0) {
4243 				mbuf_freem(nd->nd_mreq);
4244 				goto errout;
4245 			}
4246 		}
4247 	} else {
4248 		error = NFSERR_SERVERFAULT;
4249 		mbuf_freem(nd->nd_mreq);
4250 		goto errout;
4251 	}
4252 
4253 	/*
4254 	 * Call newnfs_connect(), as required, and then newnfs_request().
4255 	 */
4256 	(void) newnfs_sndlock(&clp->lc_req.nr_lock);
4257 	if (clp->lc_req.nr_client == NULL) {
4258 		if ((clp->lc_flags & LCL_NFSV41) != 0) {
4259 			error = ECONNREFUSED;
4260 			nfsrv_freesession(sep, NULL);
4261 		} else if (nd->nd_procnum == NFSV4PROC_CBNULL)
4262 			error = newnfs_connect(NULL, &clp->lc_req, cred,
4263 			    NULL, 1);
4264 		else
4265 			error = newnfs_connect(NULL, &clp->lc_req, cred,
4266 			    NULL, 3);
4267 	}
4268 	newnfs_sndunlock(&clp->lc_req.nr_lock);
4269 	if (!error) {
4270 		if ((nd->nd_flag & ND_NFSV41) != 0) {
4271 			KASSERT(sep != NULL, ("sep NULL"));
4272 			if (sep->sess_cbsess.nfsess_xprt != NULL)
4273 				error = newnfs_request(nd, NULL, clp,
4274 				    &clp->lc_req, NULL, NULL, cred,
4275 				    clp->lc_program, clp->lc_req.nr_vers, NULL,
4276 				    1, NULL, &sep->sess_cbsess);
4277 			else {
4278 				/*
4279 				 * This should probably never occur, but if a
4280 				 * client somehow does an RPC without a
4281 				 * SequenceID Op that causes a callback just
4282 				 * after the nfsd threads have been terminated
4283 				 * and restared we could conceivably get here
4284 				 * without a backchannel xprt.
4285 				 */
4286 				printf("nfsrv_docallback: no xprt\n");
4287 				error = ECONNREFUSED;
4288 			}
4289 			nfsrv_freesession(sep, NULL);
4290 		} else
4291 			error = newnfs_request(nd, NULL, clp, &clp->lc_req,
4292 			    NULL, NULL, cred, clp->lc_program,
4293 			    clp->lc_req.nr_vers, NULL, 1, NULL, NULL);
4294 	}
4295 errout:
4296 	NFSFREECRED(cred);
4297 
4298 	/*
4299 	 * If error is set here, the Callback path isn't working
4300 	 * properly, so twiddle the appropriate LCL_ flags.
4301 	 * (nd_repstat != 0 indicates the Callback path is working,
4302 	 *  but the callback failed on the client.)
4303 	 */
4304 	if (error) {
4305 		/*
4306 		 * Mark the callback pathway down, which disabled issuing
4307 		 * of delegations and gets Renew to return NFSERR_CBPATHDOWN.
4308 		 */
4309 		NFSLOCKSTATE();
4310 		clp->lc_flags |= LCL_CBDOWN;
4311 		NFSUNLOCKSTATE();
4312 	} else {
4313 		/*
4314 		 * Callback worked. If the callback path was down, disable
4315 		 * callbacks, so no more delegations will be issued. (This
4316 		 * is done on the assumption that the callback pathway is
4317 		 * flakey.)
4318 		 */
4319 		NFSLOCKSTATE();
4320 		if (clp->lc_flags & LCL_CBDOWN)
4321 			clp->lc_flags &= ~(LCL_CBDOWN | LCL_CALLBACKSON);
4322 		NFSUNLOCKSTATE();
4323 		if (nd->nd_repstat)
4324 			error = nd->nd_repstat;
4325 		else if (error == 0 && procnum == NFSV4OP_CBGETATTR)
4326 			error = nfsv4_loadattr(nd, NULL, nap, NULL, NULL, 0,
4327 			    NULL, NULL, NULL, NULL, NULL, 0, NULL, NULL, NULL,
4328 			    p, NULL);
4329 		mbuf_freem(nd->nd_mrep);
4330 	}
4331 	NFSLOCKSTATE();
4332 	clp->lc_cbref--;
4333 	if ((clp->lc_flags & LCL_WAKEUPWANTED) && clp->lc_cbref == 0) {
4334 		clp->lc_flags &= ~LCL_WAKEUPWANTED;
4335 		wakeup(clp);
4336 	}
4337 	NFSUNLOCKSTATE();
4338 
4339 	NFSEXITCODE(error);
4340 	return (error);
4341 }
4342 
4343 /*
4344  * Set up the compound RPC for the callback.
4345  */
4346 static int
4347 nfsrv_cbcallargs(struct nfsrv_descript *nd, struct nfsclient *clp,
4348     uint32_t callback, int op, const char *optag, struct nfsdsession **sepp)
4349 {
4350 	uint32_t *tl;
4351 	int error, len;
4352 
4353 	len = strlen(optag);
4354 	(void)nfsm_strtom(nd, optag, len);
4355 	NFSM_BUILD(tl, uint32_t *, 4 * NFSX_UNSIGNED);
4356 	if ((nd->nd_flag & ND_NFSV41) != 0) {
4357 		*tl++ = txdr_unsigned(NFSV41_MINORVERSION);
4358 		*tl++ = txdr_unsigned(callback);
4359 		*tl++ = txdr_unsigned(2);
4360 		*tl = txdr_unsigned(NFSV4OP_CBSEQUENCE);
4361 		error = nfsv4_setcbsequence(nd, clp, 1, sepp);
4362 		if (error != 0)
4363 			return (error);
4364 		NFSM_BUILD(tl, u_int32_t *, NFSX_UNSIGNED);
4365 		*tl = txdr_unsigned(op);
4366 	} else {
4367 		*tl++ = txdr_unsigned(NFSV4_MINORVERSION);
4368 		*tl++ = txdr_unsigned(callback);
4369 		*tl++ = txdr_unsigned(1);
4370 		*tl = txdr_unsigned(op);
4371 	}
4372 	return (0);
4373 }
4374 
4375 /*
4376  * Return the next index# for a clientid. Mostly just increment and return
4377  * the next one, but... if the 32bit unsigned does actually wrap around,
4378  * it should be rebooted.
4379  * At an average rate of one new client per second, it will wrap around in
4380  * approximately 136 years. (I think the server will have been shut
4381  * down or rebooted before then.)
4382  */
4383 static u_int32_t
4384 nfsrv_nextclientindex(void)
4385 {
4386 	static u_int32_t client_index = 0;
4387 
4388 	client_index++;
4389 	if (client_index != 0)
4390 		return (client_index);
4391 
4392 	printf("%s: out of clientids\n", __func__);
4393 	return (client_index);
4394 }
4395 
4396 /*
4397  * Return the next index# for a stateid. Mostly just increment and return
4398  * the next one, but... if the 32bit unsigned does actually wrap around
4399  * (will a BSD server stay up that long?), find
4400  * new start and end values.
4401  */
4402 static u_int32_t
4403 nfsrv_nextstateindex(struct nfsclient *clp)
4404 {
4405 	struct nfsstate *stp;
4406 	int i;
4407 	u_int32_t canuse, min_index, max_index;
4408 
4409 	if (!(clp->lc_flags & LCL_INDEXNOTOK)) {
4410 		clp->lc_stateindex++;
4411 		if (clp->lc_stateindex != clp->lc_statemaxindex)
4412 			return (clp->lc_stateindex);
4413 	}
4414 
4415 	/*
4416 	 * Yuck, we've hit the end.
4417 	 * Look for a new min and max.
4418 	 */
4419 	min_index = 0;
4420 	max_index = 0xffffffff;
4421 	for (i = 0; i < nfsrv_statehashsize; i++) {
4422 	    LIST_FOREACH(stp, &clp->lc_stateid[i], ls_hash) {
4423 		if (stp->ls_stateid.other[2] > 0x80000000) {
4424 		    if (stp->ls_stateid.other[2] < max_index)
4425 			max_index = stp->ls_stateid.other[2];
4426 		} else {
4427 		    if (stp->ls_stateid.other[2] > min_index)
4428 			min_index = stp->ls_stateid.other[2];
4429 		}
4430 	    }
4431 	}
4432 
4433 	/*
4434 	 * Yikes, highly unlikely, but I'll handle it anyhow.
4435 	 */
4436 	if (min_index == 0x80000000 && max_index == 0x80000001) {
4437 	    canuse = 0;
4438 	    /*
4439 	     * Loop around until we find an unused entry. Return that
4440 	     * and set LCL_INDEXNOTOK, so the search will continue next time.
4441 	     * (This is one of those rare cases where a goto is the
4442 	     *  cleanest way to code the loop.)
4443 	     */
4444 tryagain:
4445 	    for (i = 0; i < nfsrv_statehashsize; i++) {
4446 		LIST_FOREACH(stp, &clp->lc_stateid[i], ls_hash) {
4447 		    if (stp->ls_stateid.other[2] == canuse) {
4448 			canuse++;
4449 			goto tryagain;
4450 		    }
4451 		}
4452 	    }
4453 	    clp->lc_flags |= LCL_INDEXNOTOK;
4454 	    return (canuse);
4455 	}
4456 
4457 	/*
4458 	 * Ok to start again from min + 1.
4459 	 */
4460 	clp->lc_stateindex = min_index + 1;
4461 	clp->lc_statemaxindex = max_index;
4462 	clp->lc_flags &= ~LCL_INDEXNOTOK;
4463 	return (clp->lc_stateindex);
4464 }
4465 
4466 /*
4467  * The following functions handle the stable storage file that deals with
4468  * the edge conditions described in RFC3530 Sec. 8.6.3.
4469  * The file is as follows:
4470  * - a single record at the beginning that has the lease time of the
4471  *   previous server instance (before the last reboot) and the nfsrvboottime
4472  *   values for the previous server boots.
4473  *   These previous boot times are used to ensure that the current
4474  *   nfsrvboottime does not, somehow, get set to a previous one.
4475  *   (This is important so that Stale ClientIDs and StateIDs can
4476  *    be recognized.)
4477  *   The number of previous nfsvrboottime values precedes the list.
4478  * - followed by some number of appended records with:
4479  *   - client id string
4480  *   - flag that indicates it is a record revoking state via lease
4481  *     expiration or similar
4482  *     OR has successfully acquired state.
4483  * These structures vary in length, with the client string at the end, up
4484  * to NFSV4_OPAQUELIMIT in size.
4485  *
4486  * At the end of the grace period, the file is truncated, the first
4487  * record is rewritten with updated information and any acquired state
4488  * records for successful reclaims of state are written.
4489  *
4490  * Subsequent records are appended when the first state is issued to
4491  * a client and when state is revoked for a client.
4492  *
4493  * When reading the file in, state issued records that come later in
4494  * the file override older ones, since the append log is in cronological order.
4495  * If, for some reason, the file can't be read, the grace period is
4496  * immediately terminated and all reclaims get NFSERR_NOGRACE.
4497  */
4498 
4499 /*
4500  * Read in the stable storage file. Called by nfssvc() before the nfsd
4501  * processes start servicing requests.
4502  */
4503 APPLESTATIC void
4504 nfsrv_setupstable(NFSPROC_T *p)
4505 {
4506 	struct nfsrv_stablefirst *sf = &nfsrv_stablefirst;
4507 	struct nfsrv_stable *sp, *nsp;
4508 	struct nfst_rec *tsp;
4509 	int error, i, tryagain;
4510 	off_t off = 0;
4511 	ssize_t aresid, len;
4512 
4513 	/*
4514 	 * If NFSNSF_UPDATEDONE is set, this is a restart of the nfsds without
4515 	 * a reboot, so state has not been lost.
4516 	 */
4517 	if (sf->nsf_flags & NFSNSF_UPDATEDONE)
4518 		return;
4519 	/*
4520 	 * Set Grace over just until the file reads successfully.
4521 	 */
4522 	nfsrvboottime = time_second;
4523 	LIST_INIT(&sf->nsf_head);
4524 	sf->nsf_flags = (NFSNSF_GRACEOVER | NFSNSF_NEEDLOCK);
4525 	sf->nsf_eograce = NFSD_MONOSEC + NFSRV_LEASEDELTA;
4526 	if (sf->nsf_fp == NULL)
4527 		return;
4528 	error = NFSD_RDWR(UIO_READ, NFSFPVNODE(sf->nsf_fp),
4529 	    (caddr_t)&sf->nsf_rec, sizeof (struct nfsf_rec), off, UIO_SYSSPACE,
4530 	    0, NFSFPCRED(sf->nsf_fp), &aresid, p);
4531 	if (error || aresid || sf->nsf_numboots == 0 ||
4532 		sf->nsf_numboots > NFSNSF_MAXNUMBOOTS)
4533 		return;
4534 
4535 	/*
4536 	 * Now, read in the boottimes.
4537 	 */
4538 	sf->nsf_bootvals = (time_t *)malloc((sf->nsf_numboots + 1) *
4539 		sizeof (time_t), M_TEMP, M_WAITOK);
4540 	off = sizeof (struct nfsf_rec);
4541 	error = NFSD_RDWR(UIO_READ, NFSFPVNODE(sf->nsf_fp),
4542 	    (caddr_t)sf->nsf_bootvals, sf->nsf_numboots * sizeof (time_t), off,
4543 	    UIO_SYSSPACE, 0, NFSFPCRED(sf->nsf_fp), &aresid, p);
4544 	if (error || aresid) {
4545 		free(sf->nsf_bootvals, M_TEMP);
4546 		sf->nsf_bootvals = NULL;
4547 		return;
4548 	}
4549 
4550 	/*
4551 	 * Make sure this nfsrvboottime is different from all recorded
4552 	 * previous ones.
4553 	 */
4554 	do {
4555 		tryagain = 0;
4556 		for (i = 0; i < sf->nsf_numboots; i++) {
4557 			if (nfsrvboottime == sf->nsf_bootvals[i]) {
4558 				nfsrvboottime++;
4559 				tryagain = 1;
4560 				break;
4561 			}
4562 		}
4563 	} while (tryagain);
4564 
4565 	sf->nsf_flags |= NFSNSF_OK;
4566 	off += (sf->nsf_numboots * sizeof (time_t));
4567 
4568 	/*
4569 	 * Read through the file, building a list of records for grace
4570 	 * checking.
4571 	 * Each record is between sizeof (struct nfst_rec) and
4572 	 * sizeof (struct nfst_rec) + NFSV4_OPAQUELIMIT - 1
4573 	 * and is actually sizeof (struct nfst_rec) + nst_len - 1.
4574 	 */
4575 	tsp = (struct nfst_rec *)malloc(sizeof (struct nfst_rec) +
4576 		NFSV4_OPAQUELIMIT - 1, M_TEMP, M_WAITOK);
4577 	do {
4578 	    error = NFSD_RDWR(UIO_READ, NFSFPVNODE(sf->nsf_fp),
4579 	        (caddr_t)tsp, sizeof (struct nfst_rec) + NFSV4_OPAQUELIMIT - 1,
4580 	        off, UIO_SYSSPACE, 0, NFSFPCRED(sf->nsf_fp), &aresid, p);
4581 	    len = (sizeof (struct nfst_rec) + NFSV4_OPAQUELIMIT - 1) - aresid;
4582 	    if (error || (len > 0 && (len < sizeof (struct nfst_rec) ||
4583 		len < (sizeof (struct nfst_rec) + tsp->len - 1)))) {
4584 		/*
4585 		 * Yuck, the file has been corrupted, so just return
4586 		 * after clearing out any restart state, so the grace period
4587 		 * is over.
4588 		 */
4589 		LIST_FOREACH_SAFE(sp, &sf->nsf_head, nst_list, nsp) {
4590 			LIST_REMOVE(sp, nst_list);
4591 			free(sp, M_TEMP);
4592 		}
4593 		free(tsp, M_TEMP);
4594 		sf->nsf_flags &= ~NFSNSF_OK;
4595 		free(sf->nsf_bootvals, M_TEMP);
4596 		sf->nsf_bootvals = NULL;
4597 		return;
4598 	    }
4599 	    if (len > 0) {
4600 		off += sizeof (struct nfst_rec) + tsp->len - 1;
4601 		/*
4602 		 * Search the list for a matching client.
4603 		 */
4604 		LIST_FOREACH(sp, &sf->nsf_head, nst_list) {
4605 			if (tsp->len == sp->nst_len &&
4606 			    !NFSBCMP(tsp->client, sp->nst_client, tsp->len))
4607 				break;
4608 		}
4609 		if (sp == LIST_END(&sf->nsf_head)) {
4610 			sp = (struct nfsrv_stable *)malloc(tsp->len +
4611 				sizeof (struct nfsrv_stable) - 1, M_TEMP,
4612 				M_WAITOK);
4613 			NFSBCOPY((caddr_t)tsp, (caddr_t)&sp->nst_rec,
4614 				sizeof (struct nfst_rec) + tsp->len - 1);
4615 			LIST_INSERT_HEAD(&sf->nsf_head, sp, nst_list);
4616 		} else {
4617 			if (tsp->flag == NFSNST_REVOKE)
4618 				sp->nst_flag |= NFSNST_REVOKE;
4619 			else
4620 				/*
4621 				 * A subsequent timestamp indicates the client
4622 				 * did a setclientid/confirm and any previous
4623 				 * revoke is no longer relevant.
4624 				 */
4625 				sp->nst_flag &= ~NFSNST_REVOKE;
4626 		}
4627 	    }
4628 	} while (len > 0);
4629 	free(tsp, M_TEMP);
4630 	sf->nsf_flags = NFSNSF_OK;
4631 	sf->nsf_eograce = NFSD_MONOSEC + sf->nsf_lease +
4632 		NFSRV_LEASEDELTA;
4633 }
4634 
4635 /*
4636  * Update the stable storage file, now that the grace period is over.
4637  */
4638 APPLESTATIC void
4639 nfsrv_updatestable(NFSPROC_T *p)
4640 {
4641 	struct nfsrv_stablefirst *sf = &nfsrv_stablefirst;
4642 	struct nfsrv_stable *sp, *nsp;
4643 	int i;
4644 	struct nfsvattr nva;
4645 	vnode_t vp;
4646 #if defined(__FreeBSD_version) && (__FreeBSD_version >= 500000)
4647 	mount_t mp = NULL;
4648 #endif
4649 	int error;
4650 
4651 	if (sf->nsf_fp == NULL || (sf->nsf_flags & NFSNSF_UPDATEDONE))
4652 		return;
4653 	sf->nsf_flags |= NFSNSF_UPDATEDONE;
4654 	/*
4655 	 * Ok, we need to rewrite the stable storage file.
4656 	 * - truncate to 0 length
4657 	 * - write the new first structure
4658 	 * - loop through the data structures, writing out any that
4659 	 *   have timestamps older than the old boot
4660 	 */
4661 	if (sf->nsf_bootvals) {
4662 		sf->nsf_numboots++;
4663 		for (i = sf->nsf_numboots - 2; i >= 0; i--)
4664 			sf->nsf_bootvals[i + 1] = sf->nsf_bootvals[i];
4665 	} else {
4666 		sf->nsf_numboots = 1;
4667 		sf->nsf_bootvals = (time_t *)malloc(sizeof (time_t),
4668 			M_TEMP, M_WAITOK);
4669 	}
4670 	sf->nsf_bootvals[0] = nfsrvboottime;
4671 	sf->nsf_lease = nfsrv_lease;
4672 	NFSVNO_ATTRINIT(&nva);
4673 	NFSVNO_SETATTRVAL(&nva, size, 0);
4674 	vp = NFSFPVNODE(sf->nsf_fp);
4675 	vn_start_write(vp, &mp, V_WAIT);
4676 	if (NFSVOPLOCK(vp, LK_EXCLUSIVE) == 0) {
4677 		error = nfsvno_setattr(vp, &nva, NFSFPCRED(sf->nsf_fp), p,
4678 		    NULL);
4679 		NFSVOPUNLOCK(vp, 0);
4680 	} else
4681 		error = EPERM;
4682 	vn_finished_write(mp);
4683 	if (!error)
4684 	    error = NFSD_RDWR(UIO_WRITE, vp,
4685 		(caddr_t)&sf->nsf_rec, sizeof (struct nfsf_rec), (off_t)0,
4686 		UIO_SYSSPACE, IO_SYNC, NFSFPCRED(sf->nsf_fp), NULL, p);
4687 	if (!error)
4688 	    error = NFSD_RDWR(UIO_WRITE, vp,
4689 		(caddr_t)sf->nsf_bootvals,
4690 		sf->nsf_numboots * sizeof (time_t),
4691 		(off_t)(sizeof (struct nfsf_rec)),
4692 		UIO_SYSSPACE, IO_SYNC, NFSFPCRED(sf->nsf_fp), NULL, p);
4693 	free(sf->nsf_bootvals, M_TEMP);
4694 	sf->nsf_bootvals = NULL;
4695 	if (error) {
4696 		sf->nsf_flags &= ~NFSNSF_OK;
4697 		printf("EEK! Can't write NfsV4 stable storage file\n");
4698 		return;
4699 	}
4700 	sf->nsf_flags |= NFSNSF_OK;
4701 
4702 	/*
4703 	 * Loop through the list and write out timestamp records for
4704 	 * any clients that successfully reclaimed state.
4705 	 */
4706 	LIST_FOREACH_SAFE(sp, &sf->nsf_head, nst_list, nsp) {
4707 		if (sp->nst_flag & NFSNST_GOTSTATE) {
4708 			nfsrv_writestable(sp->nst_client, sp->nst_len,
4709 				NFSNST_NEWSTATE, p);
4710 			sp->nst_clp->lc_flags |= LCL_STAMPEDSTABLE;
4711 		}
4712 		LIST_REMOVE(sp, nst_list);
4713 		free(sp, M_TEMP);
4714 	}
4715 	nfsrv_backupstable();
4716 }
4717 
4718 /*
4719  * Append a record to the stable storage file.
4720  */
4721 APPLESTATIC void
4722 nfsrv_writestable(u_char *client, int len, int flag, NFSPROC_T *p)
4723 {
4724 	struct nfsrv_stablefirst *sf = &nfsrv_stablefirst;
4725 	struct nfst_rec *sp;
4726 	int error;
4727 
4728 	if (!(sf->nsf_flags & NFSNSF_OK) || sf->nsf_fp == NULL)
4729 		return;
4730 	sp = (struct nfst_rec *)malloc(sizeof (struct nfst_rec) +
4731 		len - 1, M_TEMP, M_WAITOK);
4732 	sp->len = len;
4733 	NFSBCOPY(client, sp->client, len);
4734 	sp->flag = flag;
4735 	error = NFSD_RDWR(UIO_WRITE, NFSFPVNODE(sf->nsf_fp),
4736 	    (caddr_t)sp, sizeof (struct nfst_rec) + len - 1, (off_t)0,
4737 	    UIO_SYSSPACE, (IO_SYNC | IO_APPEND), NFSFPCRED(sf->nsf_fp), NULL, p);
4738 	free(sp, M_TEMP);
4739 	if (error) {
4740 		sf->nsf_flags &= ~NFSNSF_OK;
4741 		printf("EEK! Can't write NfsV4 stable storage file\n");
4742 	}
4743 }
4744 
4745 /*
4746  * This function is called during the grace period to mark a client
4747  * that successfully reclaimed state.
4748  */
4749 static void
4750 nfsrv_markstable(struct nfsclient *clp)
4751 {
4752 	struct nfsrv_stable *sp;
4753 
4754 	/*
4755 	 * First find the client structure.
4756 	 */
4757 	LIST_FOREACH(sp, &nfsrv_stablefirst.nsf_head, nst_list) {
4758 		if (sp->nst_len == clp->lc_idlen &&
4759 		    !NFSBCMP(sp->nst_client, clp->lc_id, sp->nst_len))
4760 			break;
4761 	}
4762 	if (sp == LIST_END(&nfsrv_stablefirst.nsf_head))
4763 		return;
4764 
4765 	/*
4766 	 * Now, just mark it and set the nfsclient back pointer.
4767 	 */
4768 	sp->nst_flag |= NFSNST_GOTSTATE;
4769 	sp->nst_clp = clp;
4770 }
4771 
4772 /*
4773  * This function is called when a NFSv4.1 client does a ReclaimComplete.
4774  * Very similar to nfsrv_markstable(), except for the flag being set.
4775  */
4776 static void
4777 nfsrv_markreclaim(struct nfsclient *clp)
4778 {
4779 	struct nfsrv_stable *sp;
4780 
4781 	/*
4782 	 * First find the client structure.
4783 	 */
4784 	LIST_FOREACH(sp, &nfsrv_stablefirst.nsf_head, nst_list) {
4785 		if (sp->nst_len == clp->lc_idlen &&
4786 		    !NFSBCMP(sp->nst_client, clp->lc_id, sp->nst_len))
4787 			break;
4788 	}
4789 	if (sp == LIST_END(&nfsrv_stablefirst.nsf_head))
4790 		return;
4791 
4792 	/*
4793 	 * Now, just set the flag.
4794 	 */
4795 	sp->nst_flag |= NFSNST_RECLAIMED;
4796 }
4797 
4798 /*
4799  * This function is called for a reclaim, to see if it gets grace.
4800  * It returns 0 if a reclaim is allowed, 1 otherwise.
4801  */
4802 static int
4803 nfsrv_checkstable(struct nfsclient *clp)
4804 {
4805 	struct nfsrv_stable *sp;
4806 
4807 	/*
4808 	 * First, find the entry for the client.
4809 	 */
4810 	LIST_FOREACH(sp, &nfsrv_stablefirst.nsf_head, nst_list) {
4811 		if (sp->nst_len == clp->lc_idlen &&
4812 		    !NFSBCMP(sp->nst_client, clp->lc_id, sp->nst_len))
4813 			break;
4814 	}
4815 
4816 	/*
4817 	 * If not in the list, state was revoked or no state was issued
4818 	 * since the previous reboot, a reclaim is denied.
4819 	 */
4820 	if (sp == LIST_END(&nfsrv_stablefirst.nsf_head) ||
4821 	    (sp->nst_flag & NFSNST_REVOKE) ||
4822 	    !(nfsrv_stablefirst.nsf_flags & NFSNSF_OK))
4823 		return (1);
4824 	return (0);
4825 }
4826 
4827 /*
4828  * Test for and try to clear out a conflicting client. This is called by
4829  * nfsrv_lockctrl() and nfsrv_openctrl() when conflicts with other clients
4830  * a found.
4831  * The trick here is that it can't revoke a conflicting client with an
4832  * expired lease unless it holds the v4root lock, so...
4833  * If no v4root lock, get the lock and return 1 to indicate "try again".
4834  * Return 0 to indicate the conflict can't be revoked and 1 to indicate
4835  * the revocation worked and the conflicting client is "bye, bye", so it
4836  * can be tried again.
4837  * Return 2 to indicate that the vnode is VI_DOOMED after NFSVOPLOCK().
4838  * Unlocks State before a non-zero value is returned.
4839  */
4840 static int
4841 nfsrv_clientconflict(struct nfsclient *clp, int *haslockp, vnode_t vp,
4842     NFSPROC_T *p)
4843 {
4844 	int gotlock, lktype = 0;
4845 
4846 	/*
4847 	 * If lease hasn't expired, we can't fix it.
4848 	 */
4849 	if (clp->lc_expiry >= NFSD_MONOSEC ||
4850 	    !(nfsrv_stablefirst.nsf_flags & NFSNSF_UPDATEDONE))
4851 		return (0);
4852 	if (*haslockp == 0) {
4853 		NFSUNLOCKSTATE();
4854 		if (vp != NULL) {
4855 			lktype = NFSVOPISLOCKED(vp);
4856 			NFSVOPUNLOCK(vp, 0);
4857 		}
4858 		NFSLOCKV4ROOTMUTEX();
4859 		nfsv4_relref(&nfsv4rootfs_lock);
4860 		do {
4861 			gotlock = nfsv4_lock(&nfsv4rootfs_lock, 1, NULL,
4862 			    NFSV4ROOTLOCKMUTEXPTR, NULL);
4863 		} while (!gotlock);
4864 		NFSUNLOCKV4ROOTMUTEX();
4865 		*haslockp = 1;
4866 		if (vp != NULL) {
4867 			NFSVOPLOCK(vp, lktype | LK_RETRY);
4868 			if ((vp->v_iflag & VI_DOOMED) != 0)
4869 				return (2);
4870 		}
4871 		return (1);
4872 	}
4873 	NFSUNLOCKSTATE();
4874 
4875 	/*
4876 	 * Ok, we can expire the conflicting client.
4877 	 */
4878 	nfsrv_writestable(clp->lc_id, clp->lc_idlen, NFSNST_REVOKE, p);
4879 	nfsrv_backupstable();
4880 	nfsrv_cleanclient(clp, p);
4881 	nfsrv_freedeleglist(&clp->lc_deleg);
4882 	nfsrv_freedeleglist(&clp->lc_olddeleg);
4883 	LIST_REMOVE(clp, lc_hash);
4884 	nfsrv_zapclient(clp, p);
4885 	return (1);
4886 }
4887 
4888 /*
4889  * Resolve a delegation conflict.
4890  * Returns 0 to indicate the conflict was resolved without sleeping.
4891  * Return -1 to indicate that the caller should check for conflicts again.
4892  * Return > 0 for an error that should be returned, normally NFSERR_DELAY.
4893  *
4894  * Also, manipulate the nfsv4root_lock, as required. It isn't changed
4895  * for a return of 0, since there was no sleep and it could be required
4896  * later. It is released for a return of NFSERR_DELAY, since the caller
4897  * will return that error. It is released when a sleep was done waiting
4898  * for the delegation to be returned or expire (so that other nfsds can
4899  * handle ops). Then, it must be acquired for the write to stable storage.
4900  * (This function is somewhat similar to nfsrv_clientconflict(), but
4901  *  the semantics differ in a couple of subtle ways. The return of 0
4902  *  indicates the conflict was resolved without sleeping here, not
4903  *  that the conflict can't be resolved and the handling of nfsv4root_lock
4904  *  differs, as noted above.)
4905  * Unlocks State before returning a non-zero value.
4906  */
4907 static int
4908 nfsrv_delegconflict(struct nfsstate *stp, int *haslockp, NFSPROC_T *p,
4909     vnode_t vp)
4910 {
4911 	struct nfsclient *clp = stp->ls_clp;
4912 	int gotlock, error, lktype = 0, retrycnt, zapped_clp;
4913 	nfsv4stateid_t tstateid;
4914 	fhandle_t tfh;
4915 
4916 	/*
4917 	 * If the conflict is with an old delegation...
4918 	 */
4919 	if (stp->ls_flags & NFSLCK_OLDDELEG) {
4920 		/*
4921 		 * You can delete it, if it has expired.
4922 		 */
4923 		if (clp->lc_delegtime < NFSD_MONOSEC) {
4924 			nfsrv_freedeleg(stp);
4925 			NFSUNLOCKSTATE();
4926 			error = -1;
4927 			goto out;
4928 		}
4929 		NFSUNLOCKSTATE();
4930 		/*
4931 		 * During this delay, the old delegation could expire or it
4932 		 * could be recovered by the client via an Open with
4933 		 * CLAIM_DELEGATE_PREV.
4934 		 * Release the nfsv4root_lock, if held.
4935 		 */
4936 		if (*haslockp) {
4937 			*haslockp = 0;
4938 			NFSLOCKV4ROOTMUTEX();
4939 			nfsv4_unlock(&nfsv4rootfs_lock, 1);
4940 			NFSUNLOCKV4ROOTMUTEX();
4941 		}
4942 		error = NFSERR_DELAY;
4943 		goto out;
4944 	}
4945 
4946 	/*
4947 	 * It's a current delegation, so:
4948 	 * - check to see if the delegation has expired
4949 	 *   - if so, get the v4root lock and then expire it
4950 	 */
4951 	if (!(stp->ls_flags & NFSLCK_DELEGRECALL)) {
4952 		/*
4953 		 * - do a recall callback, since not yet done
4954 		 * For now, never allow truncate to be set. To use
4955 		 * truncate safely, it must be guaranteed that the
4956 		 * Remove, Rename or Setattr with size of 0 will
4957 		 * succeed and that would require major changes to
4958 		 * the VFS/Vnode OPs.
4959 		 * Set the expiry time large enough so that it won't expire
4960 		 * until after the callback, then set it correctly, once
4961 		 * the callback is done. (The delegation will now time
4962 		 * out whether or not the Recall worked ok. The timeout
4963 		 * will be extended when ops are done on the delegation
4964 		 * stateid, up to the timelimit.)
4965 		 */
4966 		stp->ls_delegtime = NFSD_MONOSEC + (2 * nfsrv_lease) +
4967 		    NFSRV_LEASEDELTA;
4968 		stp->ls_delegtimelimit = NFSD_MONOSEC + (6 * nfsrv_lease) +
4969 		    NFSRV_LEASEDELTA;
4970 		stp->ls_flags |= NFSLCK_DELEGRECALL;
4971 
4972 		/*
4973 		 * Loop NFSRV_CBRETRYCNT times while the CBRecall replies
4974 		 * NFSERR_BADSTATEID or NFSERR_BADHANDLE. This is done
4975 		 * in order to try and avoid a race that could happen
4976 		 * when a CBRecall request passed the Open reply with
4977 		 * the delegation in it when transitting the network.
4978 		 * Since nfsrv_docallback will sleep, don't use stp after
4979 		 * the call.
4980 		 */
4981 		NFSBCOPY((caddr_t)&stp->ls_stateid, (caddr_t)&tstateid,
4982 		    sizeof (tstateid));
4983 		NFSBCOPY((caddr_t)&stp->ls_lfp->lf_fh, (caddr_t)&tfh,
4984 		    sizeof (tfh));
4985 		NFSUNLOCKSTATE();
4986 		if (*haslockp) {
4987 			*haslockp = 0;
4988 			NFSLOCKV4ROOTMUTEX();
4989 			nfsv4_unlock(&nfsv4rootfs_lock, 1);
4990 			NFSUNLOCKV4ROOTMUTEX();
4991 		}
4992 		retrycnt = 0;
4993 		do {
4994 		    error = nfsrv_docallback(clp, NFSV4OP_CBRECALL,
4995 			&tstateid, 0, &tfh, NULL, NULL, p);
4996 		    retrycnt++;
4997 		} while ((error == NFSERR_BADSTATEID ||
4998 		    error == NFSERR_BADHANDLE) && retrycnt < NFSV4_CBRETRYCNT);
4999 		error = NFSERR_DELAY;
5000 		goto out;
5001 	}
5002 
5003 	if (clp->lc_expiry >= NFSD_MONOSEC &&
5004 	    stp->ls_delegtime >= NFSD_MONOSEC) {
5005 		NFSUNLOCKSTATE();
5006 		/*
5007 		 * A recall has been done, but it has not yet expired.
5008 		 * So, RETURN_DELAY.
5009 		 */
5010 		if (*haslockp) {
5011 			*haslockp = 0;
5012 			NFSLOCKV4ROOTMUTEX();
5013 			nfsv4_unlock(&nfsv4rootfs_lock, 1);
5014 			NFSUNLOCKV4ROOTMUTEX();
5015 		}
5016 		error = NFSERR_DELAY;
5017 		goto out;
5018 	}
5019 
5020 	/*
5021 	 * If we don't yet have the lock, just get it and then return,
5022 	 * since we need that before deleting expired state, such as
5023 	 * this delegation.
5024 	 * When getting the lock, unlock the vnode, so other nfsds that
5025 	 * are in progress, won't get stuck waiting for the vnode lock.
5026 	 */
5027 	if (*haslockp == 0) {
5028 		NFSUNLOCKSTATE();
5029 		if (vp != NULL) {
5030 			lktype = NFSVOPISLOCKED(vp);
5031 			NFSVOPUNLOCK(vp, 0);
5032 		}
5033 		NFSLOCKV4ROOTMUTEX();
5034 		nfsv4_relref(&nfsv4rootfs_lock);
5035 		do {
5036 			gotlock = nfsv4_lock(&nfsv4rootfs_lock, 1, NULL,
5037 			    NFSV4ROOTLOCKMUTEXPTR, NULL);
5038 		} while (!gotlock);
5039 		NFSUNLOCKV4ROOTMUTEX();
5040 		*haslockp = 1;
5041 		if (vp != NULL) {
5042 			NFSVOPLOCK(vp, lktype | LK_RETRY);
5043 			if ((vp->v_iflag & VI_DOOMED) != 0) {
5044 				*haslockp = 0;
5045 				NFSLOCKV4ROOTMUTEX();
5046 				nfsv4_unlock(&nfsv4rootfs_lock, 1);
5047 				NFSUNLOCKV4ROOTMUTEX();
5048 				error = NFSERR_PERM;
5049 				goto out;
5050 			}
5051 		}
5052 		error = -1;
5053 		goto out;
5054 	}
5055 
5056 	NFSUNLOCKSTATE();
5057 	/*
5058 	 * Ok, we can delete the expired delegation.
5059 	 * First, write the Revoke record to stable storage and then
5060 	 * clear out the conflict.
5061 	 * Since all other nfsd threads are now blocked, we can safely
5062 	 * sleep without the state changing.
5063 	 */
5064 	nfsrv_writestable(clp->lc_id, clp->lc_idlen, NFSNST_REVOKE, p);
5065 	nfsrv_backupstable();
5066 	if (clp->lc_expiry < NFSD_MONOSEC) {
5067 		nfsrv_cleanclient(clp, p);
5068 		nfsrv_freedeleglist(&clp->lc_deleg);
5069 		nfsrv_freedeleglist(&clp->lc_olddeleg);
5070 		LIST_REMOVE(clp, lc_hash);
5071 		zapped_clp = 1;
5072 	} else {
5073 		nfsrv_freedeleg(stp);
5074 		zapped_clp = 0;
5075 	}
5076 	if (zapped_clp)
5077 		nfsrv_zapclient(clp, p);
5078 	error = -1;
5079 
5080 out:
5081 	NFSEXITCODE(error);
5082 	return (error);
5083 }
5084 
5085 /*
5086  * Check for a remove allowed, if remove is set to 1 and get rid of
5087  * delegations.
5088  */
5089 APPLESTATIC int
5090 nfsrv_checkremove(vnode_t vp, int remove, NFSPROC_T *p)
5091 {
5092 	struct nfsstate *stp;
5093 	struct nfslockfile *lfp;
5094 	int error, haslock = 0;
5095 	fhandle_t nfh;
5096 
5097 	/*
5098 	 * First, get the lock file structure.
5099 	 * (A return of -1 means no associated state, so remove ok.)
5100 	 */
5101 	error = nfsrv_getlockfh(vp, NFSLCK_CHECK, NULL, &nfh, p);
5102 tryagain:
5103 	NFSLOCKSTATE();
5104 	if (!error)
5105 		error = nfsrv_getlockfile(NFSLCK_CHECK, NULL, &lfp, &nfh, 0);
5106 	if (error) {
5107 		NFSUNLOCKSTATE();
5108 		if (haslock) {
5109 			NFSLOCKV4ROOTMUTEX();
5110 			nfsv4_unlock(&nfsv4rootfs_lock, 1);
5111 			NFSUNLOCKV4ROOTMUTEX();
5112 		}
5113 		if (error == -1)
5114 			error = 0;
5115 		goto out;
5116 	}
5117 
5118 	/*
5119 	 * Now, we must Recall any delegations.
5120 	 */
5121 	error = nfsrv_cleandeleg(vp, lfp, NULL, &haslock, p);
5122 	if (error) {
5123 		/*
5124 		 * nfsrv_cleandeleg() unlocks state for non-zero
5125 		 * return.
5126 		 */
5127 		if (error == -1)
5128 			goto tryagain;
5129 		if (haslock) {
5130 			NFSLOCKV4ROOTMUTEX();
5131 			nfsv4_unlock(&nfsv4rootfs_lock, 1);
5132 			NFSUNLOCKV4ROOTMUTEX();
5133 		}
5134 		goto out;
5135 	}
5136 
5137 	/*
5138 	 * Now, look for a conflicting open share.
5139 	 */
5140 	if (remove) {
5141 		/*
5142 		 * If the entry in the directory was the last reference to the
5143 		 * corresponding filesystem object, the object can be destroyed
5144 		 * */
5145 		if(lfp->lf_usecount>1)
5146 			LIST_FOREACH(stp, &lfp->lf_open, ls_file) {
5147 				if (stp->ls_flags & NFSLCK_WRITEDENY) {
5148 					error = NFSERR_FILEOPEN;
5149 					break;
5150 				}
5151 			}
5152 	}
5153 
5154 	NFSUNLOCKSTATE();
5155 	if (haslock) {
5156 		NFSLOCKV4ROOTMUTEX();
5157 		nfsv4_unlock(&nfsv4rootfs_lock, 1);
5158 		NFSUNLOCKV4ROOTMUTEX();
5159 	}
5160 
5161 out:
5162 	NFSEXITCODE(error);
5163 	return (error);
5164 }
5165 
5166 /*
5167  * Clear out all delegations for the file referred to by lfp.
5168  * May return NFSERR_DELAY, if there will be a delay waiting for
5169  * delegations to expire.
5170  * Returns -1 to indicate it slept while recalling a delegation.
5171  * This function has the side effect of deleting the nfslockfile structure,
5172  * if it no longer has associated state and didn't have to sleep.
5173  * Unlocks State before a non-zero value is returned.
5174  */
5175 static int
5176 nfsrv_cleandeleg(vnode_t vp, struct nfslockfile *lfp,
5177     struct nfsclient *clp, int *haslockp, NFSPROC_T *p)
5178 {
5179 	struct nfsstate *stp, *nstp;
5180 	int ret = 0;
5181 
5182 	stp = LIST_FIRST(&lfp->lf_deleg);
5183 	while (stp != LIST_END(&lfp->lf_deleg)) {
5184 		nstp = LIST_NEXT(stp, ls_file);
5185 		if (stp->ls_clp != clp) {
5186 			ret = nfsrv_delegconflict(stp, haslockp, p, vp);
5187 			if (ret) {
5188 				/*
5189 				 * nfsrv_delegconflict() unlocks state
5190 				 * when it returns non-zero.
5191 				 */
5192 				goto out;
5193 			}
5194 		}
5195 		stp = nstp;
5196 	}
5197 out:
5198 	NFSEXITCODE(ret);
5199 	return (ret);
5200 }
5201 
5202 /*
5203  * There are certain operations that, when being done outside of NFSv4,
5204  * require that any NFSv4 delegation for the file be recalled.
5205  * This function is to be called for those cases:
5206  * VOP_RENAME() - When a delegation is being recalled for any reason,
5207  *	the client may have to do Opens against the server, using the file's
5208  *	final component name. If the file has been renamed on the server,
5209  *	that component name will be incorrect and the Open will fail.
5210  * VOP_REMOVE() - Theoretically, a client could Open a file after it has
5211  *	been removed on the server, if there is a delegation issued to
5212  *	that client for the file. I say "theoretically" since clients
5213  *	normally do an Access Op before the Open and that Access Op will
5214  *	fail with ESTALE. Note that NFSv2 and 3 don't even do Opens, so
5215  *	they will detect the file's removal in the same manner. (There is
5216  *	one case where RFC3530 allows a client to do an Open without first
5217  *	doing an Access Op, which is passage of a check against the ACE
5218  *	returned with a Write delegation, but current practice is to ignore
5219  *	the ACE and always do an Access Op.)
5220  *	Since the functions can only be called with an unlocked vnode, this
5221  *	can't be done at this time.
5222  * VOP_ADVLOCK() - When a client holds a delegation, it can issue byte range
5223  *	locks locally in the client, which are not visible to the server. To
5224  *	deal with this, issuing of delegations for a vnode must be disabled
5225  *	and all delegations for the vnode recalled. This is done via the
5226  *	second function, using the VV_DISABLEDELEG vflag on the vnode.
5227  */
5228 APPLESTATIC void
5229 nfsd_recalldelegation(vnode_t vp, NFSPROC_T *p)
5230 {
5231 	time_t starttime;
5232 	int error;
5233 
5234 	/*
5235 	 * First, check to see if the server is currently running and it has
5236 	 * been called for a regular file when issuing delegations.
5237 	 */
5238 	if (newnfs_numnfsd == 0 || vp->v_type != VREG ||
5239 	    nfsrv_issuedelegs == 0)
5240 		return;
5241 
5242 	KASSERT((NFSVOPISLOCKED(vp) != LK_EXCLUSIVE), ("vp %p is locked", vp));
5243 	/*
5244 	 * First, get a reference on the nfsv4rootfs_lock so that an
5245 	 * exclusive lock cannot be acquired by another thread.
5246 	 */
5247 	NFSLOCKV4ROOTMUTEX();
5248 	nfsv4_getref(&nfsv4rootfs_lock, NULL, NFSV4ROOTLOCKMUTEXPTR, NULL);
5249 	NFSUNLOCKV4ROOTMUTEX();
5250 
5251 	/*
5252 	 * Now, call nfsrv_checkremove() in a loop while it returns
5253 	 * NFSERR_DELAY. Return upon any other error or when timed out.
5254 	 */
5255 	starttime = NFSD_MONOSEC;
5256 	do {
5257 		if (NFSVOPLOCK(vp, LK_EXCLUSIVE) == 0) {
5258 			error = nfsrv_checkremove(vp, 0, p);
5259 			NFSVOPUNLOCK(vp, 0);
5260 		} else
5261 			error = EPERM;
5262 		if (error == NFSERR_DELAY) {
5263 			if (NFSD_MONOSEC - starttime > NFS_REMOVETIMEO)
5264 				break;
5265 			/* Sleep for a short period of time */
5266 			(void) nfs_catnap(PZERO, 0, "nfsremove");
5267 		}
5268 	} while (error == NFSERR_DELAY);
5269 	NFSLOCKV4ROOTMUTEX();
5270 	nfsv4_relref(&nfsv4rootfs_lock);
5271 	NFSUNLOCKV4ROOTMUTEX();
5272 }
5273 
5274 APPLESTATIC void
5275 nfsd_disabledelegation(vnode_t vp, NFSPROC_T *p)
5276 {
5277 
5278 #ifdef VV_DISABLEDELEG
5279 	/*
5280 	 * First, flag issuance of delegations disabled.
5281 	 */
5282 	atomic_set_long(&vp->v_vflag, VV_DISABLEDELEG);
5283 #endif
5284 
5285 	/*
5286 	 * Then call nfsd_recalldelegation() to get rid of all extant
5287 	 * delegations.
5288 	 */
5289 	nfsd_recalldelegation(vp, p);
5290 }
5291 
5292 /*
5293  * Check for conflicting locks, etc. and then get rid of delegations.
5294  * (At one point I thought that I should get rid of delegations for any
5295  *  Setattr, since it could potentially disallow the I/O op (read or write)
5296  *  allowed by the delegation. However, Setattr Ops that aren't changing
5297  *  the size get a stateid of all 0s, so you can't tell if it is a delegation
5298  *  for the same client or a different one, so I decided to only get rid
5299  *  of delegations for other clients when the size is being changed.)
5300  * In general, a Setattr can disable NFS I/O Ops that are outstanding, such
5301  * as Write backs, even if there is no delegation, so it really isn't any
5302  * different?)
5303  */
5304 APPLESTATIC int
5305 nfsrv_checksetattr(vnode_t vp, struct nfsrv_descript *nd,
5306     nfsv4stateid_t *stateidp, struct nfsvattr *nvap, nfsattrbit_t *attrbitp,
5307     struct nfsexstuff *exp, NFSPROC_T *p)
5308 {
5309 	struct nfsstate st, *stp = &st;
5310 	struct nfslock lo, *lop = &lo;
5311 	int error = 0;
5312 	nfsquad_t clientid;
5313 
5314 	if (NFSISSET_ATTRBIT(attrbitp, NFSATTRBIT_SIZE)) {
5315 		stp->ls_flags = (NFSLCK_CHECK | NFSLCK_WRITEACCESS);
5316 		lop->lo_first = nvap->na_size;
5317 	} else {
5318 		stp->ls_flags = 0;
5319 		lop->lo_first = 0;
5320 	}
5321 	if (NFSISSET_ATTRBIT(attrbitp, NFSATTRBIT_OWNER) ||
5322 	    NFSISSET_ATTRBIT(attrbitp, NFSATTRBIT_OWNERGROUP) ||
5323 	    NFSISSET_ATTRBIT(attrbitp, NFSATTRBIT_MODE) ||
5324 	    NFSISSET_ATTRBIT(attrbitp, NFSATTRBIT_ACL))
5325 		stp->ls_flags |= NFSLCK_SETATTR;
5326 	if (stp->ls_flags == 0)
5327 		goto out;
5328 	lop->lo_end = NFS64BITSSET;
5329 	lop->lo_flags = NFSLCK_WRITE;
5330 	stp->ls_ownerlen = 0;
5331 	stp->ls_op = NULL;
5332 	stp->ls_uid = nd->nd_cred->cr_uid;
5333 	stp->ls_stateid.seqid = stateidp->seqid;
5334 	clientid.lval[0] = stp->ls_stateid.other[0] = stateidp->other[0];
5335 	clientid.lval[1] = stp->ls_stateid.other[1] = stateidp->other[1];
5336 	stp->ls_stateid.other[2] = stateidp->other[2];
5337 	error = nfsrv_lockctrl(vp, &stp, &lop, NULL, clientid,
5338 	    stateidp, exp, nd, p);
5339 
5340 out:
5341 	NFSEXITCODE2(error, nd);
5342 	return (error);
5343 }
5344 
5345 /*
5346  * Check for a write delegation and do a CBGETATTR if there is one, updating
5347  * the attributes, as required.
5348  * Should I return an error if I can't get the attributes? (For now, I'll
5349  * just return ok.
5350  */
5351 APPLESTATIC int
5352 nfsrv_checkgetattr(struct nfsrv_descript *nd, vnode_t vp,
5353     struct nfsvattr *nvap, nfsattrbit_t *attrbitp, struct ucred *cred,
5354     NFSPROC_T *p)
5355 {
5356 	struct nfsstate *stp;
5357 	struct nfslockfile *lfp;
5358 	struct nfsclient *clp;
5359 	struct nfsvattr nva;
5360 	fhandle_t nfh;
5361 	int error = 0;
5362 	nfsattrbit_t cbbits;
5363 	u_quad_t delegfilerev;
5364 
5365 	NFSCBGETATTR_ATTRBIT(attrbitp, &cbbits);
5366 	if (!NFSNONZERO_ATTRBIT(&cbbits))
5367 		goto out;
5368 	if (nfsrv_writedelegcnt == 0)
5369 		goto out;
5370 
5371 	/*
5372 	 * Get the lock file structure.
5373 	 * (A return of -1 means no associated state, so return ok.)
5374 	 */
5375 	error = nfsrv_getlockfh(vp, NFSLCK_CHECK, NULL, &nfh, p);
5376 	NFSLOCKSTATE();
5377 	if (!error)
5378 		error = nfsrv_getlockfile(NFSLCK_CHECK, NULL, &lfp, &nfh, 0);
5379 	if (error) {
5380 		NFSUNLOCKSTATE();
5381 		if (error == -1)
5382 			error = 0;
5383 		goto out;
5384 	}
5385 
5386 	/*
5387 	 * Now, look for a write delegation.
5388 	 */
5389 	LIST_FOREACH(stp, &lfp->lf_deleg, ls_file) {
5390 		if (stp->ls_flags & NFSLCK_DELEGWRITE)
5391 			break;
5392 	}
5393 	if (stp == LIST_END(&lfp->lf_deleg)) {
5394 		NFSUNLOCKSTATE();
5395 		goto out;
5396 	}
5397 	clp = stp->ls_clp;
5398 	delegfilerev = stp->ls_filerev;
5399 
5400 	/*
5401 	 * If the Write delegation was issued as a part of this Compound RPC
5402 	 * or if we have an Implied Clientid (used in a previous Op in this
5403 	 * compound) and it is the client the delegation was issued to,
5404 	 * just return ok.
5405 	 * I also assume that it is from the same client iff the network
5406 	 * host IP address is the same as the callback address. (Not
5407 	 * exactly correct by the RFC, but avoids a lot of Getattr
5408 	 * callbacks.)
5409 	 */
5410 	if (nd->nd_compref == stp->ls_compref ||
5411 	    ((nd->nd_flag & ND_IMPLIEDCLID) &&
5412 	     clp->lc_clientid.qval == nd->nd_clientid.qval) ||
5413 	     nfsaddr2_match(clp->lc_req.nr_nam, nd->nd_nam)) {
5414 		NFSUNLOCKSTATE();
5415 		goto out;
5416 	}
5417 
5418 	/*
5419 	 * We are now done with the delegation state structure,
5420 	 * so the statelock can be released and we can now tsleep().
5421 	 */
5422 
5423 	/*
5424 	 * Now, we must do the CB Getattr callback, to see if Change or Size
5425 	 * has changed.
5426 	 */
5427 	if (clp->lc_expiry >= NFSD_MONOSEC) {
5428 		NFSUNLOCKSTATE();
5429 		NFSVNO_ATTRINIT(&nva);
5430 		nva.na_filerev = NFS64BITSSET;
5431 		error = nfsrv_docallback(clp, NFSV4OP_CBGETATTR, NULL,
5432 		    0, &nfh, &nva, &cbbits, p);
5433 		if (!error) {
5434 			if ((nva.na_filerev != NFS64BITSSET &&
5435 			    nva.na_filerev > delegfilerev) ||
5436 			    (NFSVNO_ISSETSIZE(&nva) &&
5437 			     nva.na_size != nvap->na_size)) {
5438 				error = nfsvno_updfilerev(vp, nvap, cred, p);
5439 				if (NFSVNO_ISSETSIZE(&nva))
5440 					nvap->na_size = nva.na_size;
5441 			}
5442 		} else
5443 			error = 0;	/* Ignore callback errors for now. */
5444 	} else {
5445 		NFSUNLOCKSTATE();
5446 	}
5447 
5448 out:
5449 	NFSEXITCODE2(error, nd);
5450 	return (error);
5451 }
5452 
5453 /*
5454  * This function looks for openowners that haven't had any opens for
5455  * a while and throws them away. Called by an nfsd when NFSNSF_NOOPENS
5456  * is set.
5457  */
5458 APPLESTATIC void
5459 nfsrv_throwawayopens(NFSPROC_T *p)
5460 {
5461 	struct nfsclient *clp, *nclp;
5462 	struct nfsstate *stp, *nstp;
5463 	int i;
5464 
5465 	NFSLOCKSTATE();
5466 	nfsrv_stablefirst.nsf_flags &= ~NFSNSF_NOOPENS;
5467 	/*
5468 	 * For each client...
5469 	 */
5470 	for (i = 0; i < nfsrv_clienthashsize; i++) {
5471 	    LIST_FOREACH_SAFE(clp, &nfsclienthash[i], lc_hash, nclp) {
5472 		LIST_FOREACH_SAFE(stp, &clp->lc_open, ls_list, nstp) {
5473 			if (LIST_EMPTY(&stp->ls_open) &&
5474 			    (stp->ls_noopens > NFSNOOPEN ||
5475 			     (nfsrv_openpluslock * 2) >
5476 			     nfsrv_v4statelimit))
5477 				nfsrv_freeopenowner(stp, 0, p);
5478 		}
5479 	    }
5480 	}
5481 	NFSUNLOCKSTATE();
5482 }
5483 
5484 /*
5485  * This function checks to see if the credentials are the same.
5486  * Returns 1 for not same, 0 otherwise.
5487  */
5488 static int
5489 nfsrv_notsamecredname(struct nfsrv_descript *nd, struct nfsclient *clp)
5490 {
5491 
5492 	if (nd->nd_flag & ND_GSS) {
5493 		if (!(clp->lc_flags & LCL_GSS))
5494 			return (1);
5495 		if (clp->lc_flags & LCL_NAME) {
5496 			if (nd->nd_princlen != clp->lc_namelen ||
5497 			    NFSBCMP(nd->nd_principal, clp->lc_name,
5498 				clp->lc_namelen))
5499 				return (1);
5500 			else
5501 				return (0);
5502 		}
5503 		if (nd->nd_cred->cr_uid == clp->lc_uid)
5504 			return (0);
5505 		else
5506 			return (1);
5507 	} else if (clp->lc_flags & LCL_GSS)
5508 		return (1);
5509 	/*
5510 	 * For AUTH_SYS, allow the same uid or root. (This is underspecified
5511 	 * in RFC3530, which talks about principals, but doesn't say anything
5512 	 * about uids for AUTH_SYS.)
5513 	 */
5514 	if (nd->nd_cred->cr_uid == clp->lc_uid || nd->nd_cred->cr_uid == 0)
5515 		return (0);
5516 	else
5517 		return (1);
5518 }
5519 
5520 /*
5521  * Calculate the lease expiry time.
5522  */
5523 static time_t
5524 nfsrv_leaseexpiry(void)
5525 {
5526 
5527 	if (nfsrv_stablefirst.nsf_eograce > NFSD_MONOSEC)
5528 		return (NFSD_MONOSEC + 2 * (nfsrv_lease + NFSRV_LEASEDELTA));
5529 	return (NFSD_MONOSEC + nfsrv_lease + NFSRV_LEASEDELTA);
5530 }
5531 
5532 /*
5533  * Delay the delegation timeout as far as ls_delegtimelimit, as required.
5534  */
5535 static void
5536 nfsrv_delaydelegtimeout(struct nfsstate *stp)
5537 {
5538 
5539 	if ((stp->ls_flags & NFSLCK_DELEGRECALL) == 0)
5540 		return;
5541 
5542 	if ((stp->ls_delegtime + 15) > NFSD_MONOSEC &&
5543 	    stp->ls_delegtime < stp->ls_delegtimelimit) {
5544 		stp->ls_delegtime += nfsrv_lease;
5545 		if (stp->ls_delegtime > stp->ls_delegtimelimit)
5546 			stp->ls_delegtime = stp->ls_delegtimelimit;
5547 	}
5548 }
5549 
5550 /*
5551  * This function checks to see if there is any other state associated
5552  * with the openowner for this Open.
5553  * It returns 1 if there is no other state, 0 otherwise.
5554  */
5555 static int
5556 nfsrv_nootherstate(struct nfsstate *stp)
5557 {
5558 	struct nfsstate *tstp;
5559 
5560 	LIST_FOREACH(tstp, &stp->ls_openowner->ls_open, ls_list) {
5561 		if (tstp != stp || !LIST_EMPTY(&tstp->ls_lock))
5562 			return (0);
5563 	}
5564 	return (1);
5565 }
5566 
5567 /*
5568  * Create a list of lock deltas (changes to local byte range locking
5569  * that can be rolled back using the list) and apply the changes via
5570  * nfsvno_advlock(). Optionally, lock the list. It is expected that either
5571  * the rollback or update function will be called after this.
5572  * It returns an error (and rolls back, as required), if any nfsvno_advlock()
5573  * call fails. If it returns an error, it will unlock the list.
5574  */
5575 static int
5576 nfsrv_locallock(vnode_t vp, struct nfslockfile *lfp, int flags,
5577     uint64_t first, uint64_t end, struct nfslockconflict *cfp, NFSPROC_T *p)
5578 {
5579 	struct nfslock *lop, *nlop;
5580 	int error = 0;
5581 
5582 	/* Loop through the list of locks. */
5583 	lop = LIST_FIRST(&lfp->lf_locallock);
5584 	while (first < end && lop != NULL) {
5585 		nlop = LIST_NEXT(lop, lo_lckowner);
5586 		if (first >= lop->lo_end) {
5587 			/* not there yet */
5588 			lop = nlop;
5589 		} else if (first < lop->lo_first) {
5590 			/* new one starts before entry in list */
5591 			if (end <= lop->lo_first) {
5592 				/* no overlap between old and new */
5593 				error = nfsrv_dolocal(vp, lfp, flags,
5594 				    NFSLCK_UNLOCK, first, end, cfp, p);
5595 				if (error != 0)
5596 					break;
5597 				first = end;
5598 			} else {
5599 				/* handle fragment overlapped with new one */
5600 				error = nfsrv_dolocal(vp, lfp, flags,
5601 				    NFSLCK_UNLOCK, first, lop->lo_first, cfp,
5602 				    p);
5603 				if (error != 0)
5604 					break;
5605 				first = lop->lo_first;
5606 			}
5607 		} else {
5608 			/* new one overlaps this entry in list */
5609 			if (end <= lop->lo_end) {
5610 				/* overlaps all of new one */
5611 				error = nfsrv_dolocal(vp, lfp, flags,
5612 				    lop->lo_flags, first, end, cfp, p);
5613 				if (error != 0)
5614 					break;
5615 				first = end;
5616 			} else {
5617 				/* handle fragment overlapped with new one */
5618 				error = nfsrv_dolocal(vp, lfp, flags,
5619 				    lop->lo_flags, first, lop->lo_end, cfp, p);
5620 				if (error != 0)
5621 					break;
5622 				first = lop->lo_end;
5623 				lop = nlop;
5624 			}
5625 		}
5626 	}
5627 	if (first < end && error == 0)
5628 		/* handle fragment past end of list */
5629 		error = nfsrv_dolocal(vp, lfp, flags, NFSLCK_UNLOCK, first,
5630 		    end, cfp, p);
5631 
5632 	NFSEXITCODE(error);
5633 	return (error);
5634 }
5635 
5636 /*
5637  * Local lock unlock. Unlock all byte ranges that are no longer locked
5638  * by NFSv4. To do this, unlock any subranges of first-->end that
5639  * do not overlap with the byte ranges of any lock in the lfp->lf_lock
5640  * list. This list has all locks for the file held by other
5641  * <clientid, lockowner> tuples. The list is ordered by increasing
5642  * lo_first value, but may have entries that overlap each other, for
5643  * the case of read locks.
5644  */
5645 static void
5646 nfsrv_localunlock(vnode_t vp, struct nfslockfile *lfp, uint64_t init_first,
5647     uint64_t init_end, NFSPROC_T *p)
5648 {
5649 	struct nfslock *lop;
5650 	uint64_t first, end, prevfirst;
5651 
5652 	first = init_first;
5653 	end = init_end;
5654 	while (first < init_end) {
5655 		/* Loop through all nfs locks, adjusting first and end */
5656 		prevfirst = 0;
5657 		LIST_FOREACH(lop, &lfp->lf_lock, lo_lckfile) {
5658 			KASSERT(prevfirst <= lop->lo_first,
5659 			    ("nfsv4 locks out of order"));
5660 			KASSERT(lop->lo_first < lop->lo_end,
5661 			    ("nfsv4 bogus lock"));
5662 			prevfirst = lop->lo_first;
5663 			if (first >= lop->lo_first &&
5664 			    first < lop->lo_end)
5665 				/*
5666 				 * Overlaps with initial part, so trim
5667 				 * off that initial part by moving first past
5668 				 * it.
5669 				 */
5670 				first = lop->lo_end;
5671 			else if (end > lop->lo_first &&
5672 			    lop->lo_first > first) {
5673 				/*
5674 				 * This lock defines the end of the
5675 				 * segment to unlock, so set end to the
5676 				 * start of it and break out of the loop.
5677 				 */
5678 				end = lop->lo_first;
5679 				break;
5680 			}
5681 			if (first >= end)
5682 				/*
5683 				 * There is no segment left to do, so
5684 				 * break out of this loop and then exit
5685 				 * the outer while() since first will be set
5686 				 * to end, which must equal init_end here.
5687 				 */
5688 				break;
5689 		}
5690 		if (first < end) {
5691 			/* Unlock this segment */
5692 			(void) nfsrv_dolocal(vp, lfp, NFSLCK_UNLOCK,
5693 			    NFSLCK_READ, first, end, NULL, p);
5694 			nfsrv_locallock_commit(lfp, NFSLCK_UNLOCK,
5695 			    first, end);
5696 		}
5697 		/*
5698 		 * Now move past this segment and look for any further
5699 		 * segment in the range, if there is one.
5700 		 */
5701 		first = end;
5702 		end = init_end;
5703 	}
5704 }
5705 
5706 /*
5707  * Do the local lock operation and update the rollback list, as required.
5708  * Perform the rollback and return the error if nfsvno_advlock() fails.
5709  */
5710 static int
5711 nfsrv_dolocal(vnode_t vp, struct nfslockfile *lfp, int flags, int oldflags,
5712     uint64_t first, uint64_t end, struct nfslockconflict *cfp, NFSPROC_T *p)
5713 {
5714 	struct nfsrollback *rlp;
5715 	int error = 0, ltype, oldltype;
5716 
5717 	if (flags & NFSLCK_WRITE)
5718 		ltype = F_WRLCK;
5719 	else if (flags & NFSLCK_READ)
5720 		ltype = F_RDLCK;
5721 	else
5722 		ltype = F_UNLCK;
5723 	if (oldflags & NFSLCK_WRITE)
5724 		oldltype = F_WRLCK;
5725 	else if (oldflags & NFSLCK_READ)
5726 		oldltype = F_RDLCK;
5727 	else
5728 		oldltype = F_UNLCK;
5729 	if (ltype == oldltype || (oldltype == F_WRLCK && ltype == F_RDLCK))
5730 		/* nothing to do */
5731 		goto out;
5732 	error = nfsvno_advlock(vp, ltype, first, end, p);
5733 	if (error != 0) {
5734 		if (cfp != NULL) {
5735 			cfp->cl_clientid.lval[0] = 0;
5736 			cfp->cl_clientid.lval[1] = 0;
5737 			cfp->cl_first = 0;
5738 			cfp->cl_end = NFS64BITSSET;
5739 			cfp->cl_flags = NFSLCK_WRITE;
5740 			cfp->cl_ownerlen = 5;
5741 			NFSBCOPY("LOCAL", cfp->cl_owner, 5);
5742 		}
5743 		nfsrv_locallock_rollback(vp, lfp, p);
5744 	} else if (ltype != F_UNLCK) {
5745 		rlp = malloc(sizeof (struct nfsrollback), M_NFSDROLLBACK,
5746 		    M_WAITOK);
5747 		rlp->rlck_first = first;
5748 		rlp->rlck_end = end;
5749 		rlp->rlck_type = oldltype;
5750 		LIST_INSERT_HEAD(&lfp->lf_rollback, rlp, rlck_list);
5751 	}
5752 
5753 out:
5754 	NFSEXITCODE(error);
5755 	return (error);
5756 }
5757 
5758 /*
5759  * Roll back local lock changes and free up the rollback list.
5760  */
5761 static void
5762 nfsrv_locallock_rollback(vnode_t vp, struct nfslockfile *lfp, NFSPROC_T *p)
5763 {
5764 	struct nfsrollback *rlp, *nrlp;
5765 
5766 	LIST_FOREACH_SAFE(rlp, &lfp->lf_rollback, rlck_list, nrlp) {
5767 		(void) nfsvno_advlock(vp, rlp->rlck_type, rlp->rlck_first,
5768 		    rlp->rlck_end, p);
5769 		free(rlp, M_NFSDROLLBACK);
5770 	}
5771 	LIST_INIT(&lfp->lf_rollback);
5772 }
5773 
5774 /*
5775  * Update local lock list and delete rollback list (ie now committed to the
5776  * local locks). Most of the work is done by the internal function.
5777  */
5778 static void
5779 nfsrv_locallock_commit(struct nfslockfile *lfp, int flags, uint64_t first,
5780     uint64_t end)
5781 {
5782 	struct nfsrollback *rlp, *nrlp;
5783 	struct nfslock *new_lop, *other_lop;
5784 
5785 	new_lop = malloc(sizeof (struct nfslock), M_NFSDLOCK, M_WAITOK);
5786 	if (flags & (NFSLCK_READ | NFSLCK_WRITE))
5787 		other_lop = malloc(sizeof (struct nfslock), M_NFSDLOCK,
5788 		    M_WAITOK);
5789 	else
5790 		other_lop = NULL;
5791 	new_lop->lo_flags = flags;
5792 	new_lop->lo_first = first;
5793 	new_lop->lo_end = end;
5794 	nfsrv_updatelock(NULL, &new_lop, &other_lop, lfp);
5795 	if (new_lop != NULL)
5796 		free(new_lop, M_NFSDLOCK);
5797 	if (other_lop != NULL)
5798 		free(other_lop, M_NFSDLOCK);
5799 
5800 	/* and get rid of the rollback list */
5801 	LIST_FOREACH_SAFE(rlp, &lfp->lf_rollback, rlck_list, nrlp)
5802 		free(rlp, M_NFSDROLLBACK);
5803 	LIST_INIT(&lfp->lf_rollback);
5804 }
5805 
5806 /*
5807  * Lock the struct nfslockfile for local lock updating.
5808  */
5809 static void
5810 nfsrv_locklf(struct nfslockfile *lfp)
5811 {
5812 	int gotlock;
5813 
5814 	/* lf_usecount ensures *lfp won't be free'd */
5815 	lfp->lf_usecount++;
5816 	do {
5817 		gotlock = nfsv4_lock(&lfp->lf_locallock_lck, 1, NULL,
5818 		    NFSSTATEMUTEXPTR, NULL);
5819 	} while (gotlock == 0);
5820 	lfp->lf_usecount--;
5821 }
5822 
5823 /*
5824  * Unlock the struct nfslockfile after local lock updating.
5825  */
5826 static void
5827 nfsrv_unlocklf(struct nfslockfile *lfp)
5828 {
5829 
5830 	nfsv4_unlock(&lfp->lf_locallock_lck, 0);
5831 }
5832 
5833 /*
5834  * Clear out all state for the NFSv4 server.
5835  * Must be called by a thread that can sleep when no nfsds are running.
5836  */
5837 void
5838 nfsrv_throwawayallstate(NFSPROC_T *p)
5839 {
5840 	struct nfsclient *clp, *nclp;
5841 	struct nfslockfile *lfp, *nlfp;
5842 	int i;
5843 
5844 	/*
5845 	 * For each client, clean out the state and then free the structure.
5846 	 */
5847 	for (i = 0; i < nfsrv_clienthashsize; i++) {
5848 		LIST_FOREACH_SAFE(clp, &nfsclienthash[i], lc_hash, nclp) {
5849 			nfsrv_cleanclient(clp, p);
5850 			nfsrv_freedeleglist(&clp->lc_deleg);
5851 			nfsrv_freedeleglist(&clp->lc_olddeleg);
5852 			free(clp->lc_stateid, M_NFSDCLIENT);
5853 			free(clp, M_NFSDCLIENT);
5854 		}
5855 	}
5856 
5857 	/*
5858 	 * Also, free up any remaining lock file structures.
5859 	 */
5860 	for (i = 0; i < nfsrv_lockhashsize; i++) {
5861 		LIST_FOREACH_SAFE(lfp, &nfslockhash[i], lf_hash, nlfp)
5862 			nfsrv_freenfslockfile(lfp);
5863 	}
5864 }
5865 
5866 /*
5867  * Check the sequence# for the session and slot provided as an argument.
5868  * Also, renew the lease if the session will return NFS_OK.
5869  */
5870 int
5871 nfsrv_checksequence(struct nfsrv_descript *nd, uint32_t sequenceid,
5872     uint32_t *highest_slotidp, uint32_t *target_highest_slotidp, int cache_this,
5873     uint32_t *sflagsp, NFSPROC_T *p)
5874 {
5875 	struct nfsdsession *sep;
5876 	struct nfssessionhash *shp;
5877 	int error;
5878 	SVCXPRT *savxprt;
5879 
5880 	shp = NFSSESSIONHASH(nd->nd_sessionid);
5881 	NFSLOCKSESSION(shp);
5882 	sep = nfsrv_findsession(nd->nd_sessionid);
5883 	if (sep == NULL) {
5884 		NFSUNLOCKSESSION(shp);
5885 		return (NFSERR_BADSESSION);
5886 	}
5887 	error = nfsv4_seqsession(sequenceid, nd->nd_slotid, *highest_slotidp,
5888 	    sep->sess_slots, NULL, NFSV4_SLOTS - 1);
5889 	if (error != 0) {
5890 		NFSUNLOCKSESSION(shp);
5891 		return (error);
5892 	}
5893 	if (cache_this != 0)
5894 		nd->nd_flag |= ND_SAVEREPLY;
5895 	/* Renew the lease. */
5896 	sep->sess_clp->lc_expiry = nfsrv_leaseexpiry();
5897 	nd->nd_clientid.qval = sep->sess_clp->lc_clientid.qval;
5898 	nd->nd_flag |= ND_IMPLIEDCLID;
5899 
5900 	/*
5901 	 * If this session handles the backchannel, save the nd_xprt for this
5902 	 * RPC, since this is the one being used.
5903 	 */
5904 	if (sep->sess_clp->lc_req.nr_client != NULL &&
5905 	    (sep->sess_crflags & NFSV4CRSESS_CONNBACKCHAN) != 0) {
5906 		savxprt = sep->sess_cbsess.nfsess_xprt;
5907 		SVC_ACQUIRE(nd->nd_xprt);
5908 		nd->nd_xprt->xp_p2 =
5909 		    sep->sess_clp->lc_req.nr_client->cl_private;
5910 		nd->nd_xprt->xp_idletimeout = 0;	/* Disable timeout. */
5911 		sep->sess_cbsess.nfsess_xprt = nd->nd_xprt;
5912 		if (savxprt != NULL)
5913 			SVC_RELEASE(savxprt);
5914 	}
5915 
5916 	*sflagsp = 0;
5917 	if (sep->sess_clp->lc_req.nr_client == NULL)
5918 		*sflagsp |= NFSV4SEQ_CBPATHDOWN;
5919 	NFSUNLOCKSESSION(shp);
5920 	if (error == NFSERR_EXPIRED) {
5921 		*sflagsp |= NFSV4SEQ_EXPIREDALLSTATEREVOKED;
5922 		error = 0;
5923 	} else if (error == NFSERR_ADMINREVOKED) {
5924 		*sflagsp |= NFSV4SEQ_ADMINSTATEREVOKED;
5925 		error = 0;
5926 	}
5927 	*highest_slotidp = *target_highest_slotidp = NFSV4_SLOTS - 1;
5928 	return (0);
5929 }
5930 
5931 /*
5932  * Check/set reclaim complete for this session/clientid.
5933  */
5934 int
5935 nfsrv_checkreclaimcomplete(struct nfsrv_descript *nd)
5936 {
5937 	struct nfsdsession *sep;
5938 	struct nfssessionhash *shp;
5939 	int error = 0;
5940 
5941 	shp = NFSSESSIONHASH(nd->nd_sessionid);
5942 	NFSLOCKSTATE();
5943 	NFSLOCKSESSION(shp);
5944 	sep = nfsrv_findsession(nd->nd_sessionid);
5945 	if (sep == NULL) {
5946 		NFSUNLOCKSESSION(shp);
5947 		NFSUNLOCKSTATE();
5948 		return (NFSERR_BADSESSION);
5949 	}
5950 
5951 	/* Check to see if reclaim complete has already happened. */
5952 	if ((sep->sess_clp->lc_flags & LCL_RECLAIMCOMPLETE) != 0)
5953 		error = NFSERR_COMPLETEALREADY;
5954 	else {
5955 		sep->sess_clp->lc_flags |= LCL_RECLAIMCOMPLETE;
5956 		nfsrv_markreclaim(sep->sess_clp);
5957 	}
5958 	NFSUNLOCKSESSION(shp);
5959 	NFSUNLOCKSTATE();
5960 	return (error);
5961 }
5962 
5963 /*
5964  * Cache the reply in a session slot.
5965  */
5966 void
5967 nfsrv_cache_session(uint8_t *sessionid, uint32_t slotid, int repstat,
5968    struct mbuf **m)
5969 {
5970 	struct nfsdsession *sep;
5971 	struct nfssessionhash *shp;
5972 
5973 	shp = NFSSESSIONHASH(sessionid);
5974 	NFSLOCKSESSION(shp);
5975 	sep = nfsrv_findsession(sessionid);
5976 	if (sep == NULL) {
5977 		NFSUNLOCKSESSION(shp);
5978 		printf("nfsrv_cache_session: no session\n");
5979 		m_freem(*m);
5980 		return;
5981 	}
5982 	nfsv4_seqsess_cacherep(slotid, sep->sess_slots, repstat, m);
5983 	NFSUNLOCKSESSION(shp);
5984 }
5985 
5986 /*
5987  * Search for a session that matches the sessionid.
5988  */
5989 static struct nfsdsession *
5990 nfsrv_findsession(uint8_t *sessionid)
5991 {
5992 	struct nfsdsession *sep;
5993 	struct nfssessionhash *shp;
5994 
5995 	shp = NFSSESSIONHASH(sessionid);
5996 	LIST_FOREACH(sep, &shp->list, sess_hash) {
5997 		if (!NFSBCMP(sessionid, sep->sess_sessionid, NFSX_V4SESSIONID))
5998 			break;
5999 	}
6000 	return (sep);
6001 }
6002 
6003 /*
6004  * Destroy a session.
6005  */
6006 int
6007 nfsrv_destroysession(struct nfsrv_descript *nd, uint8_t *sessionid)
6008 {
6009 	int error, samesess;
6010 
6011 	samesess = 0;
6012 	if (!NFSBCMP(sessionid, nd->nd_sessionid, NFSX_V4SESSIONID)) {
6013 		samesess = 1;
6014 		if ((nd->nd_flag & ND_LASTOP) == 0)
6015 			return (NFSERR_BADSESSION);
6016 	}
6017 	error = nfsrv_freesession(NULL, sessionid);
6018 	if (error == 0 && samesess != 0)
6019 		nd->nd_flag &= ~ND_HASSEQUENCE;
6020 	return (error);
6021 }
6022 
6023 /*
6024  * Free up a session structure.
6025  */
6026 static int
6027 nfsrv_freesession(struct nfsdsession *sep, uint8_t *sessionid)
6028 {
6029 	struct nfssessionhash *shp;
6030 	int i;
6031 
6032 	NFSLOCKSTATE();
6033 	if (sep == NULL) {
6034 		shp = NFSSESSIONHASH(sessionid);
6035 		NFSLOCKSESSION(shp);
6036 		sep = nfsrv_findsession(sessionid);
6037 	} else {
6038 		shp = NFSSESSIONHASH(sep->sess_sessionid);
6039 		NFSLOCKSESSION(shp);
6040 	}
6041 	if (sep != NULL) {
6042 		sep->sess_refcnt--;
6043 		if (sep->sess_refcnt > 0) {
6044 			NFSUNLOCKSESSION(shp);
6045 			NFSUNLOCKSTATE();
6046 			return (NFSERR_BACKCHANBUSY);
6047 		}
6048 		LIST_REMOVE(sep, sess_hash);
6049 		LIST_REMOVE(sep, sess_list);
6050 	}
6051 	NFSUNLOCKSESSION(shp);
6052 	NFSUNLOCKSTATE();
6053 	if (sep == NULL)
6054 		return (NFSERR_BADSESSION);
6055 	for (i = 0; i < NFSV4_SLOTS; i++)
6056 		if (sep->sess_slots[i].nfssl_reply != NULL)
6057 			m_freem(sep->sess_slots[i].nfssl_reply);
6058 	if (sep->sess_cbsess.nfsess_xprt != NULL)
6059 		SVC_RELEASE(sep->sess_cbsess.nfsess_xprt);
6060 	free(sep, M_NFSDSESSION);
6061 	return (0);
6062 }
6063 
6064 /*
6065  * Free a stateid.
6066  * RFC5661 says that it should fail when there are associated opens, locks
6067  * or delegations. Since stateids represent opens, I don't see how you can
6068  * free an open stateid (it will be free'd when closed), so this function
6069  * only works for lock stateids (freeing the lock_owner) or delegations.
6070  */
6071 int
6072 nfsrv_freestateid(struct nfsrv_descript *nd, nfsv4stateid_t *stateidp,
6073     NFSPROC_T *p)
6074 {
6075 	struct nfsclient *clp;
6076 	struct nfsstate *stp;
6077 	int error;
6078 
6079 	NFSLOCKSTATE();
6080 	/*
6081 	 * Look up the stateid
6082 	 */
6083 	error = nfsrv_getclient((nfsquad_t)((u_quad_t)0), CLOPS_RENEW, &clp,
6084 	    NULL, (nfsquad_t)((u_quad_t)0), 0, nd, p);
6085 	if (error == 0) {
6086 		/* First, check for a delegation. */
6087 		LIST_FOREACH(stp, &clp->lc_deleg, ls_list) {
6088 			if (!NFSBCMP(stp->ls_stateid.other, stateidp->other,
6089 			    NFSX_STATEIDOTHER))
6090 				break;
6091 		}
6092 		if (stp != NULL) {
6093 			nfsrv_freedeleg(stp);
6094 			NFSUNLOCKSTATE();
6095 			return (error);
6096 		}
6097 	}
6098 	/* Not a delegation, try for a lock_owner. */
6099 	if (error == 0)
6100 		error = nfsrv_getstate(clp, stateidp, 0, &stp);
6101 	if (error == 0 && ((stp->ls_flags & (NFSLCK_OPEN | NFSLCK_DELEGREAD |
6102 	    NFSLCK_DELEGWRITE)) != 0 || (stp->ls_flags & NFSLCK_LOCK) == 0))
6103 		/* Not a lock_owner stateid. */
6104 		error = NFSERR_LOCKSHELD;
6105 	if (error == 0 && !LIST_EMPTY(&stp->ls_lock))
6106 		error = NFSERR_LOCKSHELD;
6107 	if (error == 0)
6108 		nfsrv_freelockowner(stp, NULL, 0, p);
6109 	NFSUNLOCKSTATE();
6110 	return (error);
6111 }
6112 
6113 /*
6114  * Test a stateid.
6115  */
6116 int
6117 nfsrv_teststateid(struct nfsrv_descript *nd, nfsv4stateid_t *stateidp,
6118     NFSPROC_T *p)
6119 {
6120 	struct nfsclient *clp;
6121 	struct nfsstate *stp;
6122 	int error;
6123 
6124 	NFSLOCKSTATE();
6125 	/*
6126 	 * Look up the stateid
6127 	 */
6128 	error = nfsrv_getclient((nfsquad_t)((u_quad_t)0), CLOPS_RENEW, &clp,
6129 	    NULL, (nfsquad_t)((u_quad_t)0), 0, nd, p);
6130 	if (error == 0)
6131 		error = nfsrv_getstate(clp, stateidp, 0, &stp);
6132 	if (error == 0 && stateidp->seqid != 0 &&
6133 	    SEQ_LT(stateidp->seqid, stp->ls_stateid.seqid))
6134 		error = NFSERR_OLDSTATEID;
6135 	NFSUNLOCKSTATE();
6136 	return (error);
6137 }
6138 
6139 /*
6140  * Generate the xdr for an NFSv4.1 CBSequence Operation.
6141  */
6142 static int
6143 nfsv4_setcbsequence(struct nfsrv_descript *nd, struct nfsclient *clp,
6144     int dont_replycache, struct nfsdsession **sepp)
6145 {
6146 	struct nfsdsession *sep;
6147 	uint32_t *tl, slotseq = 0;
6148 	int maxslot, slotpos;
6149 	uint8_t sessionid[NFSX_V4SESSIONID];
6150 	int error;
6151 
6152 	error = nfsv4_getcbsession(clp, sepp);
6153 	if (error != 0)
6154 		return (error);
6155 	sep = *sepp;
6156 	(void)nfsv4_sequencelookup(NULL, &sep->sess_cbsess, &slotpos, &maxslot,
6157 	    &slotseq, sessionid);
6158 	KASSERT(maxslot >= 0, ("nfsv4_setcbsequence neg maxslot"));
6159 
6160 	/* Build the Sequence arguments. */
6161 	NFSM_BUILD(tl, uint32_t *, NFSX_V4SESSIONID + 5 * NFSX_UNSIGNED);
6162 	bcopy(sessionid, tl, NFSX_V4SESSIONID);
6163 	tl += NFSX_V4SESSIONID / NFSX_UNSIGNED;
6164 	nd->nd_slotseq = tl;
6165 	*tl++ = txdr_unsigned(slotseq);
6166 	*tl++ = txdr_unsigned(slotpos);
6167 	*tl++ = txdr_unsigned(maxslot);
6168 	if (dont_replycache == 0)
6169 		*tl++ = newnfs_true;
6170 	else
6171 		*tl++ = newnfs_false;
6172 	*tl = 0;			/* No referring call list, for now. */
6173 	nd->nd_flag |= ND_HASSEQUENCE;
6174 	return (0);
6175 }
6176 
6177 /*
6178  * Get a session for the callback.
6179  */
6180 static int
6181 nfsv4_getcbsession(struct nfsclient *clp, struct nfsdsession **sepp)
6182 {
6183 	struct nfsdsession *sep;
6184 
6185 	NFSLOCKSTATE();
6186 	LIST_FOREACH(sep, &clp->lc_session, sess_list) {
6187 		if ((sep->sess_crflags & NFSV4CRSESS_CONNBACKCHAN) != 0)
6188 			break;
6189 	}
6190 	if (sep == NULL) {
6191 		NFSUNLOCKSTATE();
6192 		return (NFSERR_BADSESSION);
6193 	}
6194 	sep->sess_refcnt++;
6195 	*sepp = sep;
6196 	NFSUNLOCKSTATE();
6197 	return (0);
6198 }
6199 
6200 /*
6201  * Free up all backchannel xprts.  This needs to be done when the nfsd threads
6202  * exit, since those transports will all be going away.
6203  * This is only called after all the nfsd threads are done performing RPCs,
6204  * so locking shouldn't be an issue.
6205  */
6206 APPLESTATIC void
6207 nfsrv_freeallbackchannel_xprts(void)
6208 {
6209 	struct nfsdsession *sep;
6210 	struct nfsclient *clp;
6211 	SVCXPRT *xprt;
6212 	int i;
6213 
6214 	for (i = 0; i < nfsrv_clienthashsize; i++) {
6215 		LIST_FOREACH(clp, &nfsclienthash[i], lc_hash) {
6216 			LIST_FOREACH(sep, &clp->lc_session, sess_list) {
6217 				xprt = sep->sess_cbsess.nfsess_xprt;
6218 				sep->sess_cbsess.nfsess_xprt = NULL;
6219 				if (xprt != NULL)
6220 					SVC_RELEASE(xprt);
6221 			}
6222 		}
6223 	}
6224 }
6225 
6226