xref: /freebsd/sys/fs/nfsserver/nfs_nfsdstate.c (revision a0ee8cc6)
1 /*-
2  * Copyright (c) 2009 Rick Macklem, University of Guelph
3  * All rights reserved.
4  *
5  * Redistribution and use in source and binary forms, with or without
6  * modification, are permitted provided that the following conditions
7  * are met:
8  * 1. Redistributions of source code must retain the above copyright
9  *    notice, this list of conditions and the following disclaimer.
10  * 2. Redistributions in binary form must reproduce the above copyright
11  *    notice, this list of conditions and the following disclaimer in the
12  *    documentation and/or other materials provided with the distribution.
13  *
14  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
15  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
16  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
17  * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
18  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
19  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
20  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
21  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
22  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
23  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
24  * SUCH DAMAGE.
25  *
26  */
27 
28 #include <sys/cdefs.h>
29 __FBSDID("$FreeBSD$");
30 
31 #ifndef APPLEKEXT
32 #include <fs/nfs/nfsport.h>
33 
34 struct nfsrv_stablefirst nfsrv_stablefirst;
35 int nfsrv_issuedelegs = 0;
36 int nfsrv_dolocallocks = 0;
37 struct nfsv4lock nfsv4rootfs_lock;
38 
39 extern int newnfs_numnfsd;
40 extern struct nfsstats newnfsstats;
41 extern int nfsrv_lease;
42 extern struct timeval nfsboottime;
43 extern u_int32_t newnfs_true, newnfs_false;
44 NFSV4ROOTLOCKMUTEX;
45 NFSSTATESPINLOCK;
46 
47 SYSCTL_DECL(_vfs_nfsd);
48 int	nfsrv_statehashsize = NFSSTATEHASHSIZE;
49 SYSCTL_INT(_vfs_nfsd, OID_AUTO, statehashsize, CTLFLAG_RDTUN,
50     &nfsrv_statehashsize, 0,
51     "Size of state hash table set via loader.conf");
52 
53 int	nfsrv_clienthashsize = NFSCLIENTHASHSIZE;
54 SYSCTL_INT(_vfs_nfsd, OID_AUTO, clienthashsize, CTLFLAG_RDTUN,
55     &nfsrv_clienthashsize, 0,
56     "Size of client hash table set via loader.conf");
57 
58 int	nfsrv_lockhashsize = NFSLOCKHASHSIZE;
59 SYSCTL_INT(_vfs_nfsd, OID_AUTO, fhhashsize, CTLFLAG_RDTUN,
60     &nfsrv_lockhashsize, 0,
61     "Size of file handle hash table set via loader.conf");
62 
63 int	nfsrv_sessionhashsize = NFSSESSIONHASHSIZE;
64 SYSCTL_INT(_vfs_nfsd, OID_AUTO, sessionhashsize, CTLFLAG_RDTUN,
65     &nfsrv_sessionhashsize, 0,
66     "Size of session hash table set via loader.conf");
67 
68 static int	nfsrv_v4statelimit = NFSRV_V4STATELIMIT;
69 SYSCTL_INT(_vfs_nfsd, OID_AUTO, v4statelimit, CTLFLAG_RWTUN,
70     &nfsrv_v4statelimit, 0,
71     "High water limit for NFSv4 opens+locks+delegations");
72 
73 /*
74  * Hash lists for nfs V4.
75  */
76 struct nfsclienthashhead	*nfsclienthash;
77 struct nfslockhashhead		*nfslockhash;
78 struct nfssessionhash		*nfssessionhash;
79 #endif	/* !APPLEKEXT */
80 
81 static u_int32_t nfsrv_openpluslock = 0, nfsrv_delegatecnt = 0;
82 static time_t nfsrvboottime;
83 static int nfsrv_writedelegifpos = 1;
84 static int nfsrv_returnoldstateid = 0, nfsrv_clients = 0;
85 static int nfsrv_clienthighwater = NFSRV_CLIENTHIGHWATER;
86 static int nfsrv_nogsscallback = 0;
87 
88 /* local functions */
89 static void nfsrv_dumpaclient(struct nfsclient *clp,
90     struct nfsd_dumpclients *dumpp);
91 static void nfsrv_freeopenowner(struct nfsstate *stp, int cansleep,
92     NFSPROC_T *p);
93 static int nfsrv_freeopen(struct nfsstate *stp, vnode_t vp, int cansleep,
94     NFSPROC_T *p);
95 static void nfsrv_freelockowner(struct nfsstate *stp, vnode_t vp, int cansleep,
96     NFSPROC_T *p);
97 static void nfsrv_freeallnfslocks(struct nfsstate *stp, vnode_t vp,
98     int cansleep, NFSPROC_T *p);
99 static void nfsrv_freenfslock(struct nfslock *lop);
100 static void nfsrv_freenfslockfile(struct nfslockfile *lfp);
101 static void nfsrv_freedeleg(struct nfsstate *);
102 static int nfsrv_getstate(struct nfsclient *clp, nfsv4stateid_t *stateidp,
103     u_int32_t flags, struct nfsstate **stpp);
104 static void nfsrv_getowner(struct nfsstatehead *hp, struct nfsstate *new_stp,
105     struct nfsstate **stpp);
106 static int nfsrv_getlockfh(vnode_t vp, u_short flags,
107     struct nfslockfile *new_lfp, fhandle_t *nfhp, NFSPROC_T *p);
108 static int nfsrv_getlockfile(u_short flags, struct nfslockfile **new_lfpp,
109     struct nfslockfile **lfpp, fhandle_t *nfhp, int lockit);
110 static void nfsrv_insertlock(struct nfslock *new_lop,
111     struct nfslock *insert_lop, struct nfsstate *stp, struct nfslockfile *lfp);
112 static void nfsrv_updatelock(struct nfsstate *stp, struct nfslock **new_lopp,
113     struct nfslock **other_lopp, struct nfslockfile *lfp);
114 static int nfsrv_getipnumber(u_char *cp);
115 static int nfsrv_checkrestart(nfsquad_t clientid, u_int32_t flags,
116     nfsv4stateid_t *stateidp, int specialid);
117 static int nfsrv_checkgrace(struct nfsrv_descript *nd, struct nfsclient *clp,
118     u_int32_t flags);
119 static int nfsrv_docallback(struct nfsclient *clp, int procnum,
120     nfsv4stateid_t *stateidp, int trunc, fhandle_t *fhp,
121     struct nfsvattr *nap, nfsattrbit_t *attrbitp, NFSPROC_T *p);
122 static int nfsrv_cbcallargs(struct nfsrv_descript *nd, struct nfsclient *clp,
123     uint32_t callback, int op, const char *optag, struct nfsdsession **sepp);
124 static u_int32_t nfsrv_nextclientindex(void);
125 static u_int32_t nfsrv_nextstateindex(struct nfsclient *clp);
126 static void nfsrv_markstable(struct nfsclient *clp);
127 static int nfsrv_checkstable(struct nfsclient *clp);
128 static int nfsrv_clientconflict(struct nfsclient *clp, int *haslockp, struct
129     vnode *vp, NFSPROC_T *p);
130 static int nfsrv_delegconflict(struct nfsstate *stp, int *haslockp,
131     NFSPROC_T *p, vnode_t vp);
132 static int nfsrv_cleandeleg(vnode_t vp, struct nfslockfile *lfp,
133     struct nfsclient *clp, int *haslockp, NFSPROC_T *p);
134 static int nfsrv_notsamecredname(struct nfsrv_descript *nd,
135     struct nfsclient *clp);
136 static time_t nfsrv_leaseexpiry(void);
137 static void nfsrv_delaydelegtimeout(struct nfsstate *stp);
138 static int nfsrv_checkseqid(struct nfsrv_descript *nd, u_int32_t seqid,
139     struct nfsstate *stp, struct nfsrvcache *op);
140 static int nfsrv_nootherstate(struct nfsstate *stp);
141 static int nfsrv_locallock(vnode_t vp, struct nfslockfile *lfp, int flags,
142     uint64_t first, uint64_t end, struct nfslockconflict *cfp, NFSPROC_T *p);
143 static void nfsrv_localunlock(vnode_t vp, struct nfslockfile *lfp,
144     uint64_t init_first, uint64_t init_end, NFSPROC_T *p);
145 static int nfsrv_dolocal(vnode_t vp, struct nfslockfile *lfp, int flags,
146     int oldflags, uint64_t first, uint64_t end, struct nfslockconflict *cfp,
147     NFSPROC_T *p);
148 static void nfsrv_locallock_rollback(vnode_t vp, struct nfslockfile *lfp,
149     NFSPROC_T *p);
150 static void nfsrv_locallock_commit(struct nfslockfile *lfp, int flags,
151     uint64_t first, uint64_t end);
152 static void nfsrv_locklf(struct nfslockfile *lfp);
153 static void nfsrv_unlocklf(struct nfslockfile *lfp);
154 static struct nfsdsession *nfsrv_findsession(uint8_t *sessionid);
155 static int nfsrv_freesession(struct nfsdsession *sep, uint8_t *sessionid);
156 static int nfsv4_setcbsequence(struct nfsrv_descript *nd, struct nfsclient *clp,
157     int dont_replycache, struct nfsdsession **sepp);
158 static int nfsv4_getcbsession(struct nfsclient *clp, struct nfsdsession **sepp);
159 
160 /*
161  * Scan the client list for a match and either return the current one,
162  * create a new entry or return an error.
163  * If returning a non-error, the clp structure must either be linked into
164  * the client list or free'd.
165  */
166 APPLESTATIC int
167 nfsrv_setclient(struct nfsrv_descript *nd, struct nfsclient **new_clpp,
168     nfsquad_t *clientidp, nfsquad_t *confirmp, NFSPROC_T *p)
169 {
170 	struct nfsclient *clp = NULL, *new_clp = *new_clpp;
171 	int i, error = 0;
172 	struct nfsstate *stp, *tstp;
173 	struct sockaddr_in *sad, *rad;
174 	int zapit = 0, gotit, hasstate = 0, igotlock;
175 	static u_int64_t confirm_index = 0;
176 
177 	/*
178 	 * Check for state resource limit exceeded.
179 	 */
180 	if (nfsrv_openpluslock > nfsrv_v4statelimit) {
181 		error = NFSERR_RESOURCE;
182 		goto out;
183 	}
184 
185 	if (nfsrv_issuedelegs == 0 ||
186 	    ((nd->nd_flag & ND_GSS) != 0 && nfsrv_nogsscallback != 0))
187 		/*
188 		 * Don't do callbacks when delegations are disabled or
189 		 * for AUTH_GSS unless enabled via nfsrv_nogsscallback.
190 		 * If establishing a callback connection is attempted
191 		 * when a firewall is blocking the callback path, the
192 		 * server may wait too long for the connect attempt to
193 		 * succeed during the Open. Some clients, such as Linux,
194 		 * may timeout and give up on the Open before the server
195 		 * replies. Also, since AUTH_GSS callbacks are not
196 		 * yet interoperability tested, they might cause the
197 		 * server to crap out, if they get past the Init call to
198 		 * the client.
199 		 */
200 		new_clp->lc_program = 0;
201 
202 	/* Lock out other nfsd threads */
203 	NFSLOCKV4ROOTMUTEX();
204 	nfsv4_relref(&nfsv4rootfs_lock);
205 	do {
206 		igotlock = nfsv4_lock(&nfsv4rootfs_lock, 1, NULL,
207 		    NFSV4ROOTLOCKMUTEXPTR, NULL);
208 	} while (!igotlock);
209 	NFSUNLOCKV4ROOTMUTEX();
210 
211 	/*
212 	 * Search for a match in the client list.
213 	 */
214 	gotit = i = 0;
215 	while (i < nfsrv_clienthashsize && !gotit) {
216 	    LIST_FOREACH(clp, &nfsclienthash[i], lc_hash) {
217 		if (new_clp->lc_idlen == clp->lc_idlen &&
218 		    !NFSBCMP(new_clp->lc_id, clp->lc_id, clp->lc_idlen)) {
219 			gotit = 1;
220 			break;
221 		}
222 	    }
223 	    if (gotit == 0)
224 		i++;
225 	}
226 	if (!gotit ||
227 	    (clp->lc_flags & (LCL_NEEDSCONFIRM | LCL_ADMINREVOKED))) {
228 		if ((nd->nd_flag & ND_NFSV41) != 0 && confirmp->lval[1] != 0) {
229 			/*
230 			 * For NFSv4.1, if confirmp->lval[1] is non-zero, the
231 			 * client is trying to update a confirmed clientid.
232 			 */
233 			NFSLOCKV4ROOTMUTEX();
234 			nfsv4_unlock(&nfsv4rootfs_lock, 1);
235 			NFSUNLOCKV4ROOTMUTEX();
236 			confirmp->lval[1] = 0;
237 			error = NFSERR_NOENT;
238 			goto out;
239 		}
240 		/*
241 		 * Get rid of the old one.
242 		 */
243 		if (i != nfsrv_clienthashsize) {
244 			LIST_REMOVE(clp, lc_hash);
245 			nfsrv_cleanclient(clp, p);
246 			nfsrv_freedeleglist(&clp->lc_deleg);
247 			nfsrv_freedeleglist(&clp->lc_olddeleg);
248 			zapit = 1;
249 		}
250 		/*
251 		 * Add it after assigning a client id to it.
252 		 */
253 		new_clp->lc_flags |= LCL_NEEDSCONFIRM;
254 		if ((nd->nd_flag & ND_NFSV41) != 0)
255 			new_clp->lc_confirm.lval[0] = confirmp->lval[0] =
256 			    ++confirm_index;
257 		else
258 			confirmp->qval = new_clp->lc_confirm.qval =
259 			    ++confirm_index;
260 		clientidp->lval[0] = new_clp->lc_clientid.lval[0] =
261 		    (u_int32_t)nfsrvboottime;
262 		clientidp->lval[1] = new_clp->lc_clientid.lval[1] =
263 		    nfsrv_nextclientindex();
264 		new_clp->lc_stateindex = 0;
265 		new_clp->lc_statemaxindex = 0;
266 		new_clp->lc_cbref = 0;
267 		new_clp->lc_expiry = nfsrv_leaseexpiry();
268 		LIST_INIT(&new_clp->lc_open);
269 		LIST_INIT(&new_clp->lc_deleg);
270 		LIST_INIT(&new_clp->lc_olddeleg);
271 		LIST_INIT(&new_clp->lc_session);
272 		for (i = 0; i < nfsrv_statehashsize; i++)
273 			LIST_INIT(&new_clp->lc_stateid[i]);
274 		LIST_INSERT_HEAD(NFSCLIENTHASH(new_clp->lc_clientid), new_clp,
275 		    lc_hash);
276 		newnfsstats.srvclients++;
277 		nfsrv_openpluslock++;
278 		nfsrv_clients++;
279 		NFSLOCKV4ROOTMUTEX();
280 		nfsv4_unlock(&nfsv4rootfs_lock, 1);
281 		NFSUNLOCKV4ROOTMUTEX();
282 		if (zapit)
283 			nfsrv_zapclient(clp, p);
284 		*new_clpp = NULL;
285 		goto out;
286 	}
287 
288 	/*
289 	 * Now, handle the cases where the id is already issued.
290 	 */
291 	if (nfsrv_notsamecredname(nd, clp)) {
292 	    /*
293 	     * Check to see if there is expired state that should go away.
294 	     */
295 	    if (clp->lc_expiry < NFSD_MONOSEC &&
296 	        (!LIST_EMPTY(&clp->lc_open) || !LIST_EMPTY(&clp->lc_deleg))) {
297 		nfsrv_cleanclient(clp, p);
298 		nfsrv_freedeleglist(&clp->lc_deleg);
299 	    }
300 
301 	    /*
302 	     * If there is outstanding state, then reply NFSERR_CLIDINUSE per
303 	     * RFC3530 Sec. 8.1.2 last para.
304 	     */
305 	    if (!LIST_EMPTY(&clp->lc_deleg)) {
306 		hasstate = 1;
307 	    } else if (LIST_EMPTY(&clp->lc_open)) {
308 		hasstate = 0;
309 	    } else {
310 		hasstate = 0;
311 		/* Look for an Open on the OpenOwner */
312 		LIST_FOREACH(stp, &clp->lc_open, ls_list) {
313 		    if (!LIST_EMPTY(&stp->ls_open)) {
314 			hasstate = 1;
315 			break;
316 		    }
317 		}
318 	    }
319 	    if (hasstate) {
320 		/*
321 		 * If the uid doesn't match, return NFSERR_CLIDINUSE after
322 		 * filling out the correct ipaddr and portnum.
323 		 */
324 		sad = NFSSOCKADDR(new_clp->lc_req.nr_nam, struct sockaddr_in *);
325 		rad = NFSSOCKADDR(clp->lc_req.nr_nam, struct sockaddr_in *);
326 		sad->sin_addr.s_addr = rad->sin_addr.s_addr;
327 		sad->sin_port = rad->sin_port;
328 		NFSLOCKV4ROOTMUTEX();
329 		nfsv4_unlock(&nfsv4rootfs_lock, 1);
330 		NFSUNLOCKV4ROOTMUTEX();
331 		error = NFSERR_CLIDINUSE;
332 		goto out;
333 	    }
334 	}
335 
336 	if (NFSBCMP(new_clp->lc_verf, clp->lc_verf, NFSX_VERF)) {
337 		/*
338 		 * If the verifier has changed, the client has rebooted
339 		 * and a new client id is issued. The old state info
340 		 * can be thrown away once the SETCLIENTID_CONFIRM occurs.
341 		 */
342 		LIST_REMOVE(clp, lc_hash);
343 		new_clp->lc_flags |= LCL_NEEDSCONFIRM;
344 		if ((nd->nd_flag & ND_NFSV41) != 0)
345 			new_clp->lc_confirm.lval[0] = confirmp->lval[0] =
346 			    ++confirm_index;
347 		else
348 			confirmp->qval = new_clp->lc_confirm.qval =
349 			    ++confirm_index;
350 		clientidp->lval[0] = new_clp->lc_clientid.lval[0] =
351 		    nfsrvboottime;
352 		clientidp->lval[1] = new_clp->lc_clientid.lval[1] =
353 		    nfsrv_nextclientindex();
354 		new_clp->lc_stateindex = 0;
355 		new_clp->lc_statemaxindex = 0;
356 		new_clp->lc_cbref = 0;
357 		new_clp->lc_expiry = nfsrv_leaseexpiry();
358 
359 		/*
360 		 * Save the state until confirmed.
361 		 */
362 		LIST_NEWHEAD(&new_clp->lc_open, &clp->lc_open, ls_list);
363 		LIST_FOREACH(tstp, &new_clp->lc_open, ls_list)
364 			tstp->ls_clp = new_clp;
365 		LIST_NEWHEAD(&new_clp->lc_deleg, &clp->lc_deleg, ls_list);
366 		LIST_FOREACH(tstp, &new_clp->lc_deleg, ls_list)
367 			tstp->ls_clp = new_clp;
368 		LIST_NEWHEAD(&new_clp->lc_olddeleg, &clp->lc_olddeleg,
369 		    ls_list);
370 		LIST_FOREACH(tstp, &new_clp->lc_olddeleg, ls_list)
371 			tstp->ls_clp = new_clp;
372 		for (i = 0; i < nfsrv_statehashsize; i++) {
373 			LIST_NEWHEAD(&new_clp->lc_stateid[i],
374 			    &clp->lc_stateid[i], ls_hash);
375 			LIST_FOREACH(tstp, &new_clp->lc_stateid[i], ls_hash)
376 				tstp->ls_clp = new_clp;
377 		}
378 		LIST_INSERT_HEAD(NFSCLIENTHASH(new_clp->lc_clientid), new_clp,
379 		    lc_hash);
380 		newnfsstats.srvclients++;
381 		nfsrv_openpluslock++;
382 		nfsrv_clients++;
383 		NFSLOCKV4ROOTMUTEX();
384 		nfsv4_unlock(&nfsv4rootfs_lock, 1);
385 		NFSUNLOCKV4ROOTMUTEX();
386 
387 		/*
388 		 * Must wait until any outstanding callback on the old clp
389 		 * completes.
390 		 */
391 		NFSLOCKSTATE();
392 		while (clp->lc_cbref) {
393 			clp->lc_flags |= LCL_WAKEUPWANTED;
394 			(void)mtx_sleep(clp, NFSSTATEMUTEXPTR, PZERO - 1,
395 			    "nfsd clp", 10 * hz);
396 		}
397 		NFSUNLOCKSTATE();
398 		nfsrv_zapclient(clp, p);
399 		*new_clpp = NULL;
400 		goto out;
401 	}
402 
403 	/* For NFSv4.1, mark that we found a confirmed clientid. */
404 	if ((nd->nd_flag & ND_NFSV41) != 0) {
405 		clientidp->lval[0] = clp->lc_clientid.lval[0];
406 		clientidp->lval[1] = clp->lc_clientid.lval[1];
407 		confirmp->lval[0] = 0;	/* Ignored by client */
408 		confirmp->lval[1] = 1;
409 	} else {
410 		/*
411 		 * id and verifier match, so update the net address info
412 		 * and get rid of any existing callback authentication
413 		 * handle, so a new one will be acquired.
414 		 */
415 		LIST_REMOVE(clp, lc_hash);
416 		new_clp->lc_flags |= (LCL_NEEDSCONFIRM | LCL_DONTCLEAN);
417 		new_clp->lc_expiry = nfsrv_leaseexpiry();
418 		confirmp->qval = new_clp->lc_confirm.qval = ++confirm_index;
419 		clientidp->lval[0] = new_clp->lc_clientid.lval[0] =
420 		    clp->lc_clientid.lval[0];
421 		clientidp->lval[1] = new_clp->lc_clientid.lval[1] =
422 		    clp->lc_clientid.lval[1];
423 		new_clp->lc_delegtime = clp->lc_delegtime;
424 		new_clp->lc_stateindex = clp->lc_stateindex;
425 		new_clp->lc_statemaxindex = clp->lc_statemaxindex;
426 		new_clp->lc_cbref = 0;
427 		LIST_NEWHEAD(&new_clp->lc_open, &clp->lc_open, ls_list);
428 		LIST_FOREACH(tstp, &new_clp->lc_open, ls_list)
429 			tstp->ls_clp = new_clp;
430 		LIST_NEWHEAD(&new_clp->lc_deleg, &clp->lc_deleg, ls_list);
431 		LIST_FOREACH(tstp, &new_clp->lc_deleg, ls_list)
432 			tstp->ls_clp = new_clp;
433 		LIST_NEWHEAD(&new_clp->lc_olddeleg, &clp->lc_olddeleg, ls_list);
434 		LIST_FOREACH(tstp, &new_clp->lc_olddeleg, ls_list)
435 			tstp->ls_clp = new_clp;
436 		for (i = 0; i < nfsrv_statehashsize; i++) {
437 			LIST_NEWHEAD(&new_clp->lc_stateid[i],
438 			    &clp->lc_stateid[i], ls_hash);
439 			LIST_FOREACH(tstp, &new_clp->lc_stateid[i], ls_hash)
440 				tstp->ls_clp = new_clp;
441 		}
442 		LIST_INSERT_HEAD(NFSCLIENTHASH(new_clp->lc_clientid), new_clp,
443 		    lc_hash);
444 		newnfsstats.srvclients++;
445 		nfsrv_openpluslock++;
446 		nfsrv_clients++;
447 	}
448 	NFSLOCKV4ROOTMUTEX();
449 	nfsv4_unlock(&nfsv4rootfs_lock, 1);
450 	NFSUNLOCKV4ROOTMUTEX();
451 
452 	if ((nd->nd_flag & ND_NFSV41) == 0) {
453 		/*
454 		 * Must wait until any outstanding callback on the old clp
455 		 * completes.
456 		 */
457 		NFSLOCKSTATE();
458 		while (clp->lc_cbref) {
459 			clp->lc_flags |= LCL_WAKEUPWANTED;
460 			(void)mtx_sleep(clp, NFSSTATEMUTEXPTR, PZERO - 1,
461 			    "nfsdclp", 10 * hz);
462 		}
463 		NFSUNLOCKSTATE();
464 		nfsrv_zapclient(clp, p);
465 		*new_clpp = NULL;
466 	}
467 
468 out:
469 	NFSEXITCODE2(error, nd);
470 	return (error);
471 }
472 
473 /*
474  * Check to see if the client id exists and optionally confirm it.
475  */
476 APPLESTATIC int
477 nfsrv_getclient(nfsquad_t clientid, int opflags, struct nfsclient **clpp,
478     struct nfsdsession *nsep, nfsquad_t confirm, uint32_t cbprogram,
479     struct nfsrv_descript *nd, NFSPROC_T *p)
480 {
481 	struct nfsclient *clp;
482 	struct nfsstate *stp;
483 	int i;
484 	struct nfsclienthashhead *hp;
485 	int error = 0, igotlock, doneok;
486 	struct nfssessionhash *shp;
487 	struct nfsdsession *sep;
488 	uint64_t sessid[2];
489 	static uint64_t next_sess = 0;
490 
491 	if (clpp)
492 		*clpp = NULL;
493 	if ((nd == NULL || (nd->nd_flag & ND_NFSV41) == 0 ||
494 	    opflags != CLOPS_RENEW) && nfsrvboottime != clientid.lval[0]) {
495 		error = NFSERR_STALECLIENTID;
496 		goto out;
497 	}
498 
499 	/*
500 	 * If called with opflags == CLOPS_RENEW, the State Lock is
501 	 * already held. Otherwise, we need to get either that or,
502 	 * for the case of Confirm, lock out the nfsd threads.
503 	 */
504 	if (opflags & CLOPS_CONFIRM) {
505 		NFSLOCKV4ROOTMUTEX();
506 		nfsv4_relref(&nfsv4rootfs_lock);
507 		do {
508 			igotlock = nfsv4_lock(&nfsv4rootfs_lock, 1, NULL,
509 			    NFSV4ROOTLOCKMUTEXPTR, NULL);
510 		} while (!igotlock);
511 		/*
512 		 * Create a new sessionid here, since we need to do it where
513 		 * there is a mutex held to serialize update of next_sess.
514 		 */
515 		if ((nd->nd_flag & ND_NFSV41) != 0) {
516 			sessid[0] = ++next_sess;
517 			sessid[1] = clientid.qval;
518 		}
519 		NFSUNLOCKV4ROOTMUTEX();
520 	} else if (opflags != CLOPS_RENEW) {
521 		NFSLOCKSTATE();
522 	}
523 
524 	/* For NFSv4.1, the clp is acquired from the associated session. */
525 	if (nd != NULL && (nd->nd_flag & ND_NFSV41) != 0 &&
526 	    opflags == CLOPS_RENEW) {
527 		clp = NULL;
528 		if ((nd->nd_flag & ND_HASSEQUENCE) != 0) {
529 			shp = NFSSESSIONHASH(nd->nd_sessionid);
530 			NFSLOCKSESSION(shp);
531 			sep = nfsrv_findsession(nd->nd_sessionid);
532 			if (sep != NULL)
533 				clp = sep->sess_clp;
534 			NFSUNLOCKSESSION(shp);
535 		}
536 	} else {
537 		hp = NFSCLIENTHASH(clientid);
538 		LIST_FOREACH(clp, hp, lc_hash) {
539 			if (clp->lc_clientid.lval[1] == clientid.lval[1])
540 				break;
541 		}
542 	}
543 	if (clp == NULL) {
544 		if (opflags & CLOPS_CONFIRM)
545 			error = NFSERR_STALECLIENTID;
546 		else
547 			error = NFSERR_EXPIRED;
548 	} else if (clp->lc_flags & LCL_ADMINREVOKED) {
549 		/*
550 		 * If marked admin revoked, just return the error.
551 		 */
552 		error = NFSERR_ADMINREVOKED;
553 	}
554 	if (error) {
555 		if (opflags & CLOPS_CONFIRM) {
556 			NFSLOCKV4ROOTMUTEX();
557 			nfsv4_unlock(&nfsv4rootfs_lock, 1);
558 			NFSUNLOCKV4ROOTMUTEX();
559 		} else if (opflags != CLOPS_RENEW) {
560 			NFSUNLOCKSTATE();
561 		}
562 		goto out;
563 	}
564 
565 	/*
566 	 * Perform any operations specified by the opflags.
567 	 */
568 	if (opflags & CLOPS_CONFIRM) {
569 		if (((nd->nd_flag & ND_NFSV41) != 0 &&
570 		     clp->lc_confirm.lval[0] != confirm.lval[0]) ||
571 		    ((nd->nd_flag & ND_NFSV41) == 0 &&
572 		     clp->lc_confirm.qval != confirm.qval))
573 			error = NFSERR_STALECLIENTID;
574 		else if (nfsrv_notsamecredname(nd, clp))
575 			error = NFSERR_CLIDINUSE;
576 
577 		if (!error) {
578 		    if ((clp->lc_flags & (LCL_NEEDSCONFIRM | LCL_DONTCLEAN)) ==
579 			LCL_NEEDSCONFIRM) {
580 			/*
581 			 * Hang onto the delegations (as old delegations)
582 			 * for an Open with CLAIM_DELEGATE_PREV unless in
583 			 * grace, but get rid of the rest of the state.
584 			 */
585 			nfsrv_cleanclient(clp, p);
586 			nfsrv_freedeleglist(&clp->lc_olddeleg);
587 			if (nfsrv_checkgrace(nd, clp, 0)) {
588 			    /* In grace, so just delete delegations */
589 			    nfsrv_freedeleglist(&clp->lc_deleg);
590 			} else {
591 			    LIST_FOREACH(stp, &clp->lc_deleg, ls_list)
592 				stp->ls_flags |= NFSLCK_OLDDELEG;
593 			    clp->lc_delegtime = NFSD_MONOSEC +
594 				nfsrv_lease + NFSRV_LEASEDELTA;
595 			    LIST_NEWHEAD(&clp->lc_olddeleg, &clp->lc_deleg,
596 				ls_list);
597 			}
598 			if ((nd->nd_flag & ND_NFSV41) != 0)
599 			    clp->lc_program = cbprogram;
600 		    }
601 		    clp->lc_flags &= ~(LCL_NEEDSCONFIRM | LCL_DONTCLEAN);
602 		    if (clp->lc_program)
603 			clp->lc_flags |= LCL_NEEDSCBNULL;
604 		    /* For NFSv4.1, link the session onto the client. */
605 		    if (nsep != NULL) {
606 			/* Hold a reference on the xprt for a backchannel. */
607 			if ((nsep->sess_crflags & NFSV4CRSESS_CONNBACKCHAN)
608 			    != 0 && clp->lc_req.nr_client == NULL) {
609 			    clp->lc_req.nr_client = (struct __rpc_client *)
610 				clnt_bck_create(nd->nd_xprt->xp_socket,
611 				cbprogram, NFSV4_CBVERS);
612 			    if (clp->lc_req.nr_client != NULL) {
613 				SVC_ACQUIRE(nd->nd_xprt);
614 				nd->nd_xprt->xp_p2 =
615 				    clp->lc_req.nr_client->cl_private;
616 				/* Disable idle timeout. */
617 				nd->nd_xprt->xp_idletimeout = 0;
618 				nsep->sess_cbsess.nfsess_xprt = nd->nd_xprt;
619 			    } else
620 				nsep->sess_crflags &= ~NFSV4CRSESS_CONNBACKCHAN;
621 			}
622 			NFSBCOPY(sessid, nsep->sess_sessionid,
623 			    NFSX_V4SESSIONID);
624 			NFSBCOPY(sessid, nsep->sess_cbsess.nfsess_sessionid,
625 			    NFSX_V4SESSIONID);
626 			shp = NFSSESSIONHASH(nsep->sess_sessionid);
627 			NFSLOCKSESSION(shp);
628 			LIST_INSERT_HEAD(&shp->list, nsep, sess_hash);
629 			NFSLOCKSTATE();
630 			LIST_INSERT_HEAD(&clp->lc_session, nsep, sess_list);
631 			nsep->sess_clp = clp;
632 			NFSUNLOCKSTATE();
633 			NFSUNLOCKSESSION(shp);
634 		    }
635 		}
636 	} else if (clp->lc_flags & LCL_NEEDSCONFIRM) {
637 		error = NFSERR_EXPIRED;
638 	}
639 
640 	/*
641 	 * If called by the Renew Op, we must check the principal.
642 	 */
643 	if (!error && (opflags & CLOPS_RENEWOP)) {
644 	    if (nfsrv_notsamecredname(nd, clp)) {
645 		doneok = 0;
646 		for (i = 0; i < nfsrv_statehashsize && doneok == 0; i++) {
647 		    LIST_FOREACH(stp, &clp->lc_stateid[i], ls_hash) {
648 			if ((stp->ls_flags & NFSLCK_OPEN) &&
649 			    stp->ls_uid == nd->nd_cred->cr_uid) {
650 				doneok = 1;
651 				break;
652 			}
653 		    }
654 		}
655 		if (!doneok)
656 			error = NFSERR_ACCES;
657 	    }
658 	    if (!error && (clp->lc_flags & LCL_CBDOWN))
659 		error = NFSERR_CBPATHDOWN;
660 	}
661 	if ((!error || error == NFSERR_CBPATHDOWN) &&
662 	     (opflags & CLOPS_RENEW)) {
663 		clp->lc_expiry = nfsrv_leaseexpiry();
664 	}
665 	if (opflags & CLOPS_CONFIRM) {
666 		NFSLOCKV4ROOTMUTEX();
667 		nfsv4_unlock(&nfsv4rootfs_lock, 1);
668 		NFSUNLOCKV4ROOTMUTEX();
669 	} else if (opflags != CLOPS_RENEW) {
670 		NFSUNLOCKSTATE();
671 	}
672 	if (clpp)
673 		*clpp = clp;
674 
675 out:
676 	NFSEXITCODE2(error, nd);
677 	return (error);
678 }
679 
680 /*
681  * Perform the NFSv4.1 destroy clientid.
682  */
683 int
684 nfsrv_destroyclient(nfsquad_t clientid, NFSPROC_T *p)
685 {
686 	struct nfsclient *clp;
687 	struct nfsclienthashhead *hp;
688 	int error = 0, i, igotlock;
689 
690 	if (nfsrvboottime != clientid.lval[0]) {
691 		error = NFSERR_STALECLIENTID;
692 		goto out;
693 	}
694 
695 	/* Lock out other nfsd threads */
696 	NFSLOCKV4ROOTMUTEX();
697 	nfsv4_relref(&nfsv4rootfs_lock);
698 	do {
699 		igotlock = nfsv4_lock(&nfsv4rootfs_lock, 1, NULL,
700 		    NFSV4ROOTLOCKMUTEXPTR, NULL);
701 	} while (igotlock == 0);
702 	NFSUNLOCKV4ROOTMUTEX();
703 
704 	hp = NFSCLIENTHASH(clientid);
705 	LIST_FOREACH(clp, hp, lc_hash) {
706 		if (clp->lc_clientid.lval[1] == clientid.lval[1])
707 			break;
708 	}
709 	if (clp == NULL) {
710 		NFSLOCKV4ROOTMUTEX();
711 		nfsv4_unlock(&nfsv4rootfs_lock, 1);
712 		NFSUNLOCKV4ROOTMUTEX();
713 		/* Just return ok, since it is gone. */
714 		goto out;
715 	}
716 
717 	/* Scan for state on the clientid. */
718 	for (i = 0; i < nfsrv_statehashsize; i++)
719 		if (!LIST_EMPTY(&clp->lc_stateid[i])) {
720 			NFSLOCKV4ROOTMUTEX();
721 			nfsv4_unlock(&nfsv4rootfs_lock, 1);
722 			NFSUNLOCKV4ROOTMUTEX();
723 			error = NFSERR_CLIENTIDBUSY;
724 			goto out;
725 		}
726 	if (!LIST_EMPTY(&clp->lc_session) || !LIST_EMPTY(&clp->lc_deleg)) {
727 		NFSLOCKV4ROOTMUTEX();
728 		nfsv4_unlock(&nfsv4rootfs_lock, 1);
729 		NFSUNLOCKV4ROOTMUTEX();
730 		error = NFSERR_CLIENTIDBUSY;
731 		goto out;
732 	}
733 
734 	/* Destroy the clientid and return ok. */
735 	nfsrv_cleanclient(clp, p);
736 	nfsrv_freedeleglist(&clp->lc_deleg);
737 	nfsrv_freedeleglist(&clp->lc_olddeleg);
738 	LIST_REMOVE(clp, lc_hash);
739 	NFSLOCKV4ROOTMUTEX();
740 	nfsv4_unlock(&nfsv4rootfs_lock, 1);
741 	NFSUNLOCKV4ROOTMUTEX();
742 	nfsrv_zapclient(clp, p);
743 out:
744 	NFSEXITCODE2(error, nd);
745 	return (error);
746 }
747 
748 /*
749  * Called from the new nfssvc syscall to admin revoke a clientid.
750  * Returns 0 for success, error otherwise.
751  */
752 APPLESTATIC int
753 nfsrv_adminrevoke(struct nfsd_clid *revokep, NFSPROC_T *p)
754 {
755 	struct nfsclient *clp = NULL;
756 	int i, error = 0;
757 	int gotit, igotlock;
758 
759 	/*
760 	 * First, lock out the nfsd so that state won't change while the
761 	 * revocation record is being written to the stable storage restart
762 	 * file.
763 	 */
764 	NFSLOCKV4ROOTMUTEX();
765 	do {
766 		igotlock = nfsv4_lock(&nfsv4rootfs_lock, 1, NULL,
767 		    NFSV4ROOTLOCKMUTEXPTR, NULL);
768 	} while (!igotlock);
769 	NFSUNLOCKV4ROOTMUTEX();
770 
771 	/*
772 	 * Search for a match in the client list.
773 	 */
774 	gotit = i = 0;
775 	while (i < nfsrv_clienthashsize && !gotit) {
776 	    LIST_FOREACH(clp, &nfsclienthash[i], lc_hash) {
777 		if (revokep->nclid_idlen == clp->lc_idlen &&
778 		    !NFSBCMP(revokep->nclid_id, clp->lc_id, clp->lc_idlen)) {
779 			gotit = 1;
780 			break;
781 		}
782 	    }
783 	    i++;
784 	}
785 	if (!gotit) {
786 		NFSLOCKV4ROOTMUTEX();
787 		nfsv4_unlock(&nfsv4rootfs_lock, 0);
788 		NFSUNLOCKV4ROOTMUTEX();
789 		error = EPERM;
790 		goto out;
791 	}
792 
793 	/*
794 	 * Now, write out the revocation record
795 	 */
796 	nfsrv_writestable(clp->lc_id, clp->lc_idlen, NFSNST_REVOKE, p);
797 	nfsrv_backupstable();
798 
799 	/*
800 	 * and clear out the state, marking the clientid revoked.
801 	 */
802 	clp->lc_flags &= ~LCL_CALLBACKSON;
803 	clp->lc_flags |= LCL_ADMINREVOKED;
804 	nfsrv_cleanclient(clp, p);
805 	nfsrv_freedeleglist(&clp->lc_deleg);
806 	nfsrv_freedeleglist(&clp->lc_olddeleg);
807 	NFSLOCKV4ROOTMUTEX();
808 	nfsv4_unlock(&nfsv4rootfs_lock, 0);
809 	NFSUNLOCKV4ROOTMUTEX();
810 
811 out:
812 	NFSEXITCODE(error);
813 	return (error);
814 }
815 
816 /*
817  * Dump out stats for all clients. Called from nfssvc(2), that is used
818  * newnfsstats.
819  */
820 APPLESTATIC void
821 nfsrv_dumpclients(struct nfsd_dumpclients *dumpp, int maxcnt)
822 {
823 	struct nfsclient *clp;
824 	int i = 0, cnt = 0;
825 
826 	/*
827 	 * First, get a reference on the nfsv4rootfs_lock so that an
828 	 * exclusive lock cannot be acquired while dumping the clients.
829 	 */
830 	NFSLOCKV4ROOTMUTEX();
831 	nfsv4_getref(&nfsv4rootfs_lock, NULL, NFSV4ROOTLOCKMUTEXPTR, NULL);
832 	NFSUNLOCKV4ROOTMUTEX();
833 	NFSLOCKSTATE();
834 	/*
835 	 * Rattle through the client lists until done.
836 	 */
837 	while (i < nfsrv_clienthashsize && cnt < maxcnt) {
838 	    clp = LIST_FIRST(&nfsclienthash[i]);
839 	    while (clp != LIST_END(&nfsclienthash[i]) && cnt < maxcnt) {
840 		nfsrv_dumpaclient(clp, &dumpp[cnt]);
841 		cnt++;
842 		clp = LIST_NEXT(clp, lc_hash);
843 	    }
844 	    i++;
845 	}
846 	if (cnt < maxcnt)
847 	    dumpp[cnt].ndcl_clid.nclid_idlen = 0;
848 	NFSUNLOCKSTATE();
849 	NFSLOCKV4ROOTMUTEX();
850 	nfsv4_relref(&nfsv4rootfs_lock);
851 	NFSUNLOCKV4ROOTMUTEX();
852 }
853 
854 /*
855  * Dump stats for a client. Must be called with the NFSSTATELOCK and spl'd.
856  */
857 static void
858 nfsrv_dumpaclient(struct nfsclient *clp, struct nfsd_dumpclients *dumpp)
859 {
860 	struct nfsstate *stp, *openstp, *lckownstp;
861 	struct nfslock *lop;
862 	struct sockaddr *sad;
863 	struct sockaddr_in *rad;
864 	struct sockaddr_in6 *rad6;
865 
866 	dumpp->ndcl_nopenowners = dumpp->ndcl_nlockowners = 0;
867 	dumpp->ndcl_nopens = dumpp->ndcl_nlocks = 0;
868 	dumpp->ndcl_ndelegs = dumpp->ndcl_nolddelegs = 0;
869 	dumpp->ndcl_flags = clp->lc_flags;
870 	dumpp->ndcl_clid.nclid_idlen = clp->lc_idlen;
871 	NFSBCOPY(clp->lc_id, dumpp->ndcl_clid.nclid_id, clp->lc_idlen);
872 	sad = NFSSOCKADDR(clp->lc_req.nr_nam, struct sockaddr *);
873 	dumpp->ndcl_addrfam = sad->sa_family;
874 	if (sad->sa_family == AF_INET) {
875 		rad = (struct sockaddr_in *)sad;
876 		dumpp->ndcl_cbaddr.sin_addr = rad->sin_addr;
877 	} else {
878 		rad6 = (struct sockaddr_in6 *)sad;
879 		dumpp->ndcl_cbaddr.sin6_addr = rad6->sin6_addr;
880 	}
881 
882 	/*
883 	 * Now, scan the state lists and total up the opens and locks.
884 	 */
885 	LIST_FOREACH(stp, &clp->lc_open, ls_list) {
886 	    dumpp->ndcl_nopenowners++;
887 	    LIST_FOREACH(openstp, &stp->ls_open, ls_list) {
888 		dumpp->ndcl_nopens++;
889 		LIST_FOREACH(lckownstp, &openstp->ls_open, ls_list) {
890 		    dumpp->ndcl_nlockowners++;
891 		    LIST_FOREACH(lop, &lckownstp->ls_lock, lo_lckowner) {
892 			dumpp->ndcl_nlocks++;
893 		    }
894 		}
895 	    }
896 	}
897 
898 	/*
899 	 * and the delegation lists.
900 	 */
901 	LIST_FOREACH(stp, &clp->lc_deleg, ls_list) {
902 	    dumpp->ndcl_ndelegs++;
903 	}
904 	LIST_FOREACH(stp, &clp->lc_olddeleg, ls_list) {
905 	    dumpp->ndcl_nolddelegs++;
906 	}
907 }
908 
909 /*
910  * Dump out lock stats for a file.
911  */
912 APPLESTATIC void
913 nfsrv_dumplocks(vnode_t vp, struct nfsd_dumplocks *ldumpp, int maxcnt,
914     NFSPROC_T *p)
915 {
916 	struct nfsstate *stp;
917 	struct nfslock *lop;
918 	int cnt = 0;
919 	struct nfslockfile *lfp;
920 	struct sockaddr *sad;
921 	struct sockaddr_in *rad;
922 	struct sockaddr_in6 *rad6;
923 	int ret;
924 	fhandle_t nfh;
925 
926 	ret = nfsrv_getlockfh(vp, 0, NULL, &nfh, p);
927 	/*
928 	 * First, get a reference on the nfsv4rootfs_lock so that an
929 	 * exclusive lock on it cannot be acquired while dumping the locks.
930 	 */
931 	NFSLOCKV4ROOTMUTEX();
932 	nfsv4_getref(&nfsv4rootfs_lock, NULL, NFSV4ROOTLOCKMUTEXPTR, NULL);
933 	NFSUNLOCKV4ROOTMUTEX();
934 	NFSLOCKSTATE();
935 	if (!ret)
936 		ret = nfsrv_getlockfile(0, NULL, &lfp, &nfh, 0);
937 	if (ret) {
938 		ldumpp[0].ndlck_clid.nclid_idlen = 0;
939 		NFSUNLOCKSTATE();
940 		NFSLOCKV4ROOTMUTEX();
941 		nfsv4_relref(&nfsv4rootfs_lock);
942 		NFSUNLOCKV4ROOTMUTEX();
943 		return;
944 	}
945 
946 	/*
947 	 * For each open share on file, dump it out.
948 	 */
949 	stp = LIST_FIRST(&lfp->lf_open);
950 	while (stp != LIST_END(&lfp->lf_open) && cnt < maxcnt) {
951 		ldumpp[cnt].ndlck_flags = stp->ls_flags;
952 		ldumpp[cnt].ndlck_stateid.seqid = stp->ls_stateid.seqid;
953 		ldumpp[cnt].ndlck_stateid.other[0] = stp->ls_stateid.other[0];
954 		ldumpp[cnt].ndlck_stateid.other[1] = stp->ls_stateid.other[1];
955 		ldumpp[cnt].ndlck_stateid.other[2] = stp->ls_stateid.other[2];
956 		ldumpp[cnt].ndlck_owner.nclid_idlen =
957 		    stp->ls_openowner->ls_ownerlen;
958 		NFSBCOPY(stp->ls_openowner->ls_owner,
959 		    ldumpp[cnt].ndlck_owner.nclid_id,
960 		    stp->ls_openowner->ls_ownerlen);
961 		ldumpp[cnt].ndlck_clid.nclid_idlen = stp->ls_clp->lc_idlen;
962 		NFSBCOPY(stp->ls_clp->lc_id, ldumpp[cnt].ndlck_clid.nclid_id,
963 		    stp->ls_clp->lc_idlen);
964 		sad=NFSSOCKADDR(stp->ls_clp->lc_req.nr_nam, struct sockaddr *);
965 		ldumpp[cnt].ndlck_addrfam = sad->sa_family;
966 		if (sad->sa_family == AF_INET) {
967 			rad = (struct sockaddr_in *)sad;
968 			ldumpp[cnt].ndlck_cbaddr.sin_addr = rad->sin_addr;
969 		} else {
970 			rad6 = (struct sockaddr_in6 *)sad;
971 			ldumpp[cnt].ndlck_cbaddr.sin6_addr = rad6->sin6_addr;
972 		}
973 		stp = LIST_NEXT(stp, ls_file);
974 		cnt++;
975 	}
976 
977 	/*
978 	 * and all locks.
979 	 */
980 	lop = LIST_FIRST(&lfp->lf_lock);
981 	while (lop != LIST_END(&lfp->lf_lock) && cnt < maxcnt) {
982 		stp = lop->lo_stp;
983 		ldumpp[cnt].ndlck_flags = lop->lo_flags;
984 		ldumpp[cnt].ndlck_first = lop->lo_first;
985 		ldumpp[cnt].ndlck_end = lop->lo_end;
986 		ldumpp[cnt].ndlck_stateid.seqid = stp->ls_stateid.seqid;
987 		ldumpp[cnt].ndlck_stateid.other[0] = stp->ls_stateid.other[0];
988 		ldumpp[cnt].ndlck_stateid.other[1] = stp->ls_stateid.other[1];
989 		ldumpp[cnt].ndlck_stateid.other[2] = stp->ls_stateid.other[2];
990 		ldumpp[cnt].ndlck_owner.nclid_idlen = stp->ls_ownerlen;
991 		NFSBCOPY(stp->ls_owner, ldumpp[cnt].ndlck_owner.nclid_id,
992 		    stp->ls_ownerlen);
993 		ldumpp[cnt].ndlck_clid.nclid_idlen = stp->ls_clp->lc_idlen;
994 		NFSBCOPY(stp->ls_clp->lc_id, ldumpp[cnt].ndlck_clid.nclid_id,
995 		    stp->ls_clp->lc_idlen);
996 		sad=NFSSOCKADDR(stp->ls_clp->lc_req.nr_nam, struct sockaddr *);
997 		ldumpp[cnt].ndlck_addrfam = sad->sa_family;
998 		if (sad->sa_family == AF_INET) {
999 			rad = (struct sockaddr_in *)sad;
1000 			ldumpp[cnt].ndlck_cbaddr.sin_addr = rad->sin_addr;
1001 		} else {
1002 			rad6 = (struct sockaddr_in6 *)sad;
1003 			ldumpp[cnt].ndlck_cbaddr.sin6_addr = rad6->sin6_addr;
1004 		}
1005 		lop = LIST_NEXT(lop, lo_lckfile);
1006 		cnt++;
1007 	}
1008 
1009 	/*
1010 	 * and the delegations.
1011 	 */
1012 	stp = LIST_FIRST(&lfp->lf_deleg);
1013 	while (stp != LIST_END(&lfp->lf_deleg) && cnt < maxcnt) {
1014 		ldumpp[cnt].ndlck_flags = stp->ls_flags;
1015 		ldumpp[cnt].ndlck_stateid.seqid = stp->ls_stateid.seqid;
1016 		ldumpp[cnt].ndlck_stateid.other[0] = stp->ls_stateid.other[0];
1017 		ldumpp[cnt].ndlck_stateid.other[1] = stp->ls_stateid.other[1];
1018 		ldumpp[cnt].ndlck_stateid.other[2] = stp->ls_stateid.other[2];
1019 		ldumpp[cnt].ndlck_owner.nclid_idlen = 0;
1020 		ldumpp[cnt].ndlck_clid.nclid_idlen = stp->ls_clp->lc_idlen;
1021 		NFSBCOPY(stp->ls_clp->lc_id, ldumpp[cnt].ndlck_clid.nclid_id,
1022 		    stp->ls_clp->lc_idlen);
1023 		sad=NFSSOCKADDR(stp->ls_clp->lc_req.nr_nam, struct sockaddr *);
1024 		ldumpp[cnt].ndlck_addrfam = sad->sa_family;
1025 		if (sad->sa_family == AF_INET) {
1026 			rad = (struct sockaddr_in *)sad;
1027 			ldumpp[cnt].ndlck_cbaddr.sin_addr = rad->sin_addr;
1028 		} else {
1029 			rad6 = (struct sockaddr_in6 *)sad;
1030 			ldumpp[cnt].ndlck_cbaddr.sin6_addr = rad6->sin6_addr;
1031 		}
1032 		stp = LIST_NEXT(stp, ls_file);
1033 		cnt++;
1034 	}
1035 
1036 	/*
1037 	 * If list isn't full, mark end of list by setting the client name
1038 	 * to zero length.
1039 	 */
1040 	if (cnt < maxcnt)
1041 		ldumpp[cnt].ndlck_clid.nclid_idlen = 0;
1042 	NFSUNLOCKSTATE();
1043 	NFSLOCKV4ROOTMUTEX();
1044 	nfsv4_relref(&nfsv4rootfs_lock);
1045 	NFSUNLOCKV4ROOTMUTEX();
1046 }
1047 
1048 /*
1049  * Server timer routine. It can scan any linked list, so long
1050  * as it holds the spin/mutex lock and there is no exclusive lock on
1051  * nfsv4rootfs_lock.
1052  * (For OpenBSD, a kthread is ok. For FreeBSD, I think it is ok
1053  *  to do this from a callout, since the spin locks work. For
1054  *  Darwin, I'm not sure what will work correctly yet.)
1055  * Should be called once per second.
1056  */
1057 APPLESTATIC void
1058 nfsrv_servertimer(void)
1059 {
1060 	struct nfsclient *clp, *nclp;
1061 	struct nfsstate *stp, *nstp;
1062 	int got_ref, i;
1063 
1064 	/*
1065 	 * Make sure nfsboottime is set. This is used by V3 as well
1066 	 * as V4. Note that nfsboottime is not nfsrvboottime, which is
1067 	 * only used by the V4 server for leases.
1068 	 */
1069 	if (nfsboottime.tv_sec == 0)
1070 		NFSSETBOOTTIME(nfsboottime);
1071 
1072 	/*
1073 	 * If server hasn't started yet, just return.
1074 	 */
1075 	NFSLOCKSTATE();
1076 	if (nfsrv_stablefirst.nsf_eograce == 0) {
1077 		NFSUNLOCKSTATE();
1078 		return;
1079 	}
1080 	if (!(nfsrv_stablefirst.nsf_flags & NFSNSF_UPDATEDONE)) {
1081 		if (!(nfsrv_stablefirst.nsf_flags & NFSNSF_GRACEOVER) &&
1082 		    NFSD_MONOSEC > nfsrv_stablefirst.nsf_eograce)
1083 			nfsrv_stablefirst.nsf_flags |=
1084 			    (NFSNSF_GRACEOVER | NFSNSF_NEEDLOCK);
1085 		NFSUNLOCKSTATE();
1086 		return;
1087 	}
1088 
1089 	/*
1090 	 * Try and get a reference count on the nfsv4rootfs_lock so that
1091 	 * no nfsd thread can acquire an exclusive lock on it before this
1092 	 * call is done. If it is already exclusively locked, just return.
1093 	 */
1094 	NFSLOCKV4ROOTMUTEX();
1095 	got_ref = nfsv4_getref_nonblock(&nfsv4rootfs_lock);
1096 	NFSUNLOCKV4ROOTMUTEX();
1097 	if (got_ref == 0) {
1098 		NFSUNLOCKSTATE();
1099 		return;
1100 	}
1101 
1102 	/*
1103 	 * For each client...
1104 	 */
1105 	for (i = 0; i < nfsrv_clienthashsize; i++) {
1106 	    clp = LIST_FIRST(&nfsclienthash[i]);
1107 	    while (clp != LIST_END(&nfsclienthash[i])) {
1108 		nclp = LIST_NEXT(clp, lc_hash);
1109 		if (!(clp->lc_flags & LCL_EXPIREIT)) {
1110 		    if (((clp->lc_expiry + NFSRV_STALELEASE) < NFSD_MONOSEC
1111 			 && ((LIST_EMPTY(&clp->lc_deleg)
1112 			      && LIST_EMPTY(&clp->lc_open)) ||
1113 			     nfsrv_clients > nfsrv_clienthighwater)) ||
1114 			(clp->lc_expiry + NFSRV_MOULDYLEASE) < NFSD_MONOSEC ||
1115 			(clp->lc_expiry < NFSD_MONOSEC &&
1116 			 (nfsrv_openpluslock * 10 / 9) > nfsrv_v4statelimit)) {
1117 			/*
1118 			 * Lease has expired several nfsrv_lease times ago:
1119 			 * PLUS
1120 			 *    - no state is associated with it
1121 			 *    OR
1122 			 *    - above high water mark for number of clients
1123 			 *      (nfsrv_clienthighwater should be large enough
1124 			 *       that this only occurs when clients fail to
1125 			 *       use the same nfs_client_id4.id. Maybe somewhat
1126 			 *       higher that the maximum number of clients that
1127 			 *       will mount this server?)
1128 			 * OR
1129 			 * Lease has expired a very long time ago
1130 			 * OR
1131 			 * Lease has expired PLUS the number of opens + locks
1132 			 * has exceeded 90% of capacity
1133 			 *
1134 			 * --> Mark for expiry. The actual expiry will be done
1135 			 *     by an nfsd sometime soon.
1136 			 */
1137 			clp->lc_flags |= LCL_EXPIREIT;
1138 			nfsrv_stablefirst.nsf_flags |=
1139 			    (NFSNSF_NEEDLOCK | NFSNSF_EXPIREDCLIENT);
1140 		    } else {
1141 			/*
1142 			 * If there are no opens, increment no open tick cnt
1143 			 * If time exceeds NFSNOOPEN, mark it to be thrown away
1144 			 * otherwise, if there is an open, reset no open time
1145 			 * Hopefully, this will avoid excessive re-creation
1146 			 * of open owners and subsequent open confirms.
1147 			 */
1148 			stp = LIST_FIRST(&clp->lc_open);
1149 			while (stp != LIST_END(&clp->lc_open)) {
1150 				nstp = LIST_NEXT(stp, ls_list);
1151 				if (LIST_EMPTY(&stp->ls_open)) {
1152 					stp->ls_noopens++;
1153 					if (stp->ls_noopens > NFSNOOPEN ||
1154 					    (nfsrv_openpluslock * 2) >
1155 					    nfsrv_v4statelimit)
1156 						nfsrv_stablefirst.nsf_flags |=
1157 							NFSNSF_NOOPENS;
1158 				} else {
1159 					stp->ls_noopens = 0;
1160 				}
1161 				stp = nstp;
1162 			}
1163 		    }
1164 		}
1165 		clp = nclp;
1166 	    }
1167 	}
1168 	NFSUNLOCKSTATE();
1169 	NFSLOCKV4ROOTMUTEX();
1170 	nfsv4_relref(&nfsv4rootfs_lock);
1171 	NFSUNLOCKV4ROOTMUTEX();
1172 }
1173 
1174 /*
1175  * The following set of functions free up the various data structures.
1176  */
1177 /*
1178  * Clear out all open/lock state related to this nfsclient.
1179  * Caller must hold an exclusive lock on nfsv4rootfs_lock, so that
1180  * there are no other active nfsd threads.
1181  */
1182 APPLESTATIC void
1183 nfsrv_cleanclient(struct nfsclient *clp, NFSPROC_T *p)
1184 {
1185 	struct nfsstate *stp, *nstp;
1186 	struct nfsdsession *sep, *nsep;
1187 
1188 	LIST_FOREACH_SAFE(stp, &clp->lc_open, ls_list, nstp)
1189 		nfsrv_freeopenowner(stp, 1, p);
1190 	if ((clp->lc_flags & LCL_ADMINREVOKED) == 0)
1191 		LIST_FOREACH_SAFE(sep, &clp->lc_session, sess_list, nsep)
1192 			(void)nfsrv_freesession(sep, NULL);
1193 }
1194 
1195 /*
1196  * Free a client that has been cleaned. It should also already have been
1197  * removed from the lists.
1198  * (Just to be safe w.r.t. newnfs_disconnect(), call this function when
1199  *  softclock interrupts are enabled.)
1200  */
1201 APPLESTATIC void
1202 nfsrv_zapclient(struct nfsclient *clp, NFSPROC_T *p)
1203 {
1204 
1205 #ifdef notyet
1206 	if ((clp->lc_flags & (LCL_GSS | LCL_CALLBACKSON)) ==
1207 	     (LCL_GSS | LCL_CALLBACKSON) &&
1208 	    (clp->lc_hand.nfsh_flag & NFSG_COMPLETE) &&
1209 	    clp->lc_handlelen > 0) {
1210 		clp->lc_hand.nfsh_flag &= ~NFSG_COMPLETE;
1211 		clp->lc_hand.nfsh_flag |= NFSG_DESTROYED;
1212 		(void) nfsrv_docallback(clp, NFSV4PROC_CBNULL,
1213 			NULL, 0, NULL, NULL, NULL, p);
1214 	}
1215 #endif
1216 	newnfs_disconnect(&clp->lc_req);
1217 	NFSSOCKADDRFREE(clp->lc_req.nr_nam);
1218 	NFSFREEMUTEX(&clp->lc_req.nr_mtx);
1219 	free(clp->lc_stateid, M_NFSDCLIENT);
1220 	free(clp, M_NFSDCLIENT);
1221 	NFSLOCKSTATE();
1222 	newnfsstats.srvclients--;
1223 	nfsrv_openpluslock--;
1224 	nfsrv_clients--;
1225 	NFSUNLOCKSTATE();
1226 }
1227 
1228 /*
1229  * Free a list of delegation state structures.
1230  * (This function will also free all nfslockfile structures that no
1231  *  longer have associated state.)
1232  */
1233 APPLESTATIC void
1234 nfsrv_freedeleglist(struct nfsstatehead *sthp)
1235 {
1236 	struct nfsstate *stp, *nstp;
1237 
1238 	LIST_FOREACH_SAFE(stp, sthp, ls_list, nstp) {
1239 		nfsrv_freedeleg(stp);
1240 	}
1241 	LIST_INIT(sthp);
1242 }
1243 
1244 /*
1245  * Free up a delegation.
1246  */
1247 static void
1248 nfsrv_freedeleg(struct nfsstate *stp)
1249 {
1250 	struct nfslockfile *lfp;
1251 
1252 	LIST_REMOVE(stp, ls_hash);
1253 	LIST_REMOVE(stp, ls_list);
1254 	LIST_REMOVE(stp, ls_file);
1255 	lfp = stp->ls_lfp;
1256 	if (LIST_EMPTY(&lfp->lf_open) &&
1257 	    LIST_EMPTY(&lfp->lf_lock) && LIST_EMPTY(&lfp->lf_deleg) &&
1258 	    LIST_EMPTY(&lfp->lf_locallock) && LIST_EMPTY(&lfp->lf_rollback) &&
1259 	    lfp->lf_usecount == 0 &&
1260 	    nfsv4_testlock(&lfp->lf_locallock_lck) == 0)
1261 		nfsrv_freenfslockfile(lfp);
1262 	FREE((caddr_t)stp, M_NFSDSTATE);
1263 	newnfsstats.srvdelegates--;
1264 	nfsrv_openpluslock--;
1265 	nfsrv_delegatecnt--;
1266 }
1267 
1268 /*
1269  * This function frees an open owner and all associated opens.
1270  */
1271 static void
1272 nfsrv_freeopenowner(struct nfsstate *stp, int cansleep, NFSPROC_T *p)
1273 {
1274 	struct nfsstate *nstp, *tstp;
1275 
1276 	LIST_REMOVE(stp, ls_list);
1277 	/*
1278 	 * Now, free all associated opens.
1279 	 */
1280 	nstp = LIST_FIRST(&stp->ls_open);
1281 	while (nstp != LIST_END(&stp->ls_open)) {
1282 		tstp = nstp;
1283 		nstp = LIST_NEXT(nstp, ls_list);
1284 		(void) nfsrv_freeopen(tstp, NULL, cansleep, p);
1285 	}
1286 	if (stp->ls_op)
1287 		nfsrvd_derefcache(stp->ls_op);
1288 	FREE((caddr_t)stp, M_NFSDSTATE);
1289 	newnfsstats.srvopenowners--;
1290 	nfsrv_openpluslock--;
1291 }
1292 
1293 /*
1294  * This function frees an open (nfsstate open structure) with all associated
1295  * lock_owners and locks. It also frees the nfslockfile structure iff there
1296  * are no other opens on the file.
1297  * Returns 1 if it free'd the nfslockfile, 0 otherwise.
1298  */
1299 static int
1300 nfsrv_freeopen(struct nfsstate *stp, vnode_t vp, int cansleep, NFSPROC_T *p)
1301 {
1302 	struct nfsstate *nstp, *tstp;
1303 	struct nfslockfile *lfp;
1304 	int ret;
1305 
1306 	LIST_REMOVE(stp, ls_hash);
1307 	LIST_REMOVE(stp, ls_list);
1308 	LIST_REMOVE(stp, ls_file);
1309 
1310 	lfp = stp->ls_lfp;
1311 	/*
1312 	 * Now, free all lockowners associated with this open.
1313 	 */
1314 	LIST_FOREACH_SAFE(tstp, &stp->ls_open, ls_list, nstp)
1315 		nfsrv_freelockowner(tstp, vp, cansleep, p);
1316 
1317 	/*
1318 	 * The nfslockfile is freed here if there are no locks
1319 	 * associated with the open.
1320 	 * If there are locks associated with the open, the
1321 	 * nfslockfile structure can be freed via nfsrv_freelockowner().
1322 	 * Acquire the state mutex to avoid races with calls to
1323 	 * nfsrv_getlockfile().
1324 	 */
1325 	if (cansleep != 0)
1326 		NFSLOCKSTATE();
1327 	if (lfp != NULL && LIST_EMPTY(&lfp->lf_open) &&
1328 	    LIST_EMPTY(&lfp->lf_deleg) && LIST_EMPTY(&lfp->lf_lock) &&
1329 	    LIST_EMPTY(&lfp->lf_locallock) && LIST_EMPTY(&lfp->lf_rollback) &&
1330 	    lfp->lf_usecount == 0 &&
1331 	    (cansleep != 0 || nfsv4_testlock(&lfp->lf_locallock_lck) == 0)) {
1332 		nfsrv_freenfslockfile(lfp);
1333 		ret = 1;
1334 	} else
1335 		ret = 0;
1336 	if (cansleep != 0)
1337 		NFSUNLOCKSTATE();
1338 	FREE((caddr_t)stp, M_NFSDSTATE);
1339 	newnfsstats.srvopens--;
1340 	nfsrv_openpluslock--;
1341 	return (ret);
1342 }
1343 
1344 /*
1345  * Frees a lockowner and all associated locks.
1346  */
1347 static void
1348 nfsrv_freelockowner(struct nfsstate *stp, vnode_t vp, int cansleep,
1349     NFSPROC_T *p)
1350 {
1351 
1352 	LIST_REMOVE(stp, ls_hash);
1353 	LIST_REMOVE(stp, ls_list);
1354 	nfsrv_freeallnfslocks(stp, vp, cansleep, p);
1355 	if (stp->ls_op)
1356 		nfsrvd_derefcache(stp->ls_op);
1357 	FREE((caddr_t)stp, M_NFSDSTATE);
1358 	newnfsstats.srvlockowners--;
1359 	nfsrv_openpluslock--;
1360 }
1361 
1362 /*
1363  * Free all the nfs locks on a lockowner.
1364  */
1365 static void
1366 nfsrv_freeallnfslocks(struct nfsstate *stp, vnode_t vp, int cansleep,
1367     NFSPROC_T *p)
1368 {
1369 	struct nfslock *lop, *nlop;
1370 	struct nfsrollback *rlp, *nrlp;
1371 	struct nfslockfile *lfp = NULL;
1372 	int gottvp = 0;
1373 	vnode_t tvp = NULL;
1374 	uint64_t first, end;
1375 
1376 	if (vp != NULL)
1377 		ASSERT_VOP_UNLOCKED(vp, "nfsrv_freeallnfslocks: vnode locked");
1378 	lop = LIST_FIRST(&stp->ls_lock);
1379 	while (lop != LIST_END(&stp->ls_lock)) {
1380 		nlop = LIST_NEXT(lop, lo_lckowner);
1381 		/*
1382 		 * Since all locks should be for the same file, lfp should
1383 		 * not change.
1384 		 */
1385 		if (lfp == NULL)
1386 			lfp = lop->lo_lfp;
1387 		else if (lfp != lop->lo_lfp)
1388 			panic("allnfslocks");
1389 		/*
1390 		 * If vp is NULL and cansleep != 0, a vnode must be acquired
1391 		 * from the file handle. This only occurs when called from
1392 		 * nfsrv_cleanclient().
1393 		 */
1394 		if (gottvp == 0) {
1395 			if (nfsrv_dolocallocks == 0)
1396 				tvp = NULL;
1397 			else if (vp == NULL && cansleep != 0) {
1398 				tvp = nfsvno_getvp(&lfp->lf_fh);
1399 				NFSVOPUNLOCK(tvp, 0);
1400 			} else
1401 				tvp = vp;
1402 			gottvp = 1;
1403 		}
1404 
1405 		if (tvp != NULL) {
1406 			if (cansleep == 0)
1407 				panic("allnfs2");
1408 			first = lop->lo_first;
1409 			end = lop->lo_end;
1410 			nfsrv_freenfslock(lop);
1411 			nfsrv_localunlock(tvp, lfp, first, end, p);
1412 			LIST_FOREACH_SAFE(rlp, &lfp->lf_rollback, rlck_list,
1413 			    nrlp)
1414 				free(rlp, M_NFSDROLLBACK);
1415 			LIST_INIT(&lfp->lf_rollback);
1416 		} else
1417 			nfsrv_freenfslock(lop);
1418 		lop = nlop;
1419 	}
1420 	if (vp == NULL && tvp != NULL)
1421 		vrele(tvp);
1422 }
1423 
1424 /*
1425  * Free an nfslock structure.
1426  */
1427 static void
1428 nfsrv_freenfslock(struct nfslock *lop)
1429 {
1430 
1431 	if (lop->lo_lckfile.le_prev != NULL) {
1432 		LIST_REMOVE(lop, lo_lckfile);
1433 		newnfsstats.srvlocks--;
1434 		nfsrv_openpluslock--;
1435 	}
1436 	LIST_REMOVE(lop, lo_lckowner);
1437 	FREE((caddr_t)lop, M_NFSDLOCK);
1438 }
1439 
1440 /*
1441  * This function frees an nfslockfile structure.
1442  */
1443 static void
1444 nfsrv_freenfslockfile(struct nfslockfile *lfp)
1445 {
1446 
1447 	LIST_REMOVE(lfp, lf_hash);
1448 	FREE((caddr_t)lfp, M_NFSDLOCKFILE);
1449 }
1450 
1451 /*
1452  * This function looks up an nfsstate structure via stateid.
1453  */
1454 static int
1455 nfsrv_getstate(struct nfsclient *clp, nfsv4stateid_t *stateidp, __unused u_int32_t flags,
1456     struct nfsstate **stpp)
1457 {
1458 	struct nfsstate *stp;
1459 	struct nfsstatehead *hp;
1460 	int error = 0;
1461 
1462 	*stpp = NULL;
1463 	hp = NFSSTATEHASH(clp, *stateidp);
1464 	LIST_FOREACH(stp, hp, ls_hash) {
1465 		if (!NFSBCMP(stp->ls_stateid.other, stateidp->other,
1466 			NFSX_STATEIDOTHER))
1467 			break;
1468 	}
1469 
1470 	/*
1471 	 * If no state id in list, return NFSERR_BADSTATEID.
1472 	 */
1473 	if (stp == LIST_END(hp)) {
1474 		error = NFSERR_BADSTATEID;
1475 		goto out;
1476 	}
1477 	*stpp = stp;
1478 
1479 out:
1480 	NFSEXITCODE(error);
1481 	return (error);
1482 }
1483 
1484 /*
1485  * This function gets an nfsstate structure via owner string.
1486  */
1487 static void
1488 nfsrv_getowner(struct nfsstatehead *hp, struct nfsstate *new_stp,
1489     struct nfsstate **stpp)
1490 {
1491 	struct nfsstate *stp;
1492 
1493 	*stpp = NULL;
1494 	LIST_FOREACH(stp, hp, ls_list) {
1495 		if (new_stp->ls_ownerlen == stp->ls_ownerlen &&
1496 		  !NFSBCMP(new_stp->ls_owner,stp->ls_owner,stp->ls_ownerlen)) {
1497 			*stpp = stp;
1498 			return;
1499 		}
1500 	}
1501 }
1502 
1503 /*
1504  * Lock control function called to update lock status.
1505  * Returns 0 upon success, -1 if there is no lock and the flags indicate
1506  * that one isn't to be created and an NFSERR_xxx for other errors.
1507  * The structures new_stp and new_lop are passed in as pointers that should
1508  * be set to NULL if the structure is used and shouldn't be free'd.
1509  * For the NFSLCK_TEST and NFSLCK_CHECK cases, the structures are
1510  * never used and can safely be allocated on the stack. For all other
1511  * cases, *new_stpp and *new_lopp should be malloc'd before the call,
1512  * in case they are used.
1513  */
1514 APPLESTATIC int
1515 nfsrv_lockctrl(vnode_t vp, struct nfsstate **new_stpp,
1516     struct nfslock **new_lopp, struct nfslockconflict *cfp,
1517     nfsquad_t clientid, nfsv4stateid_t *stateidp,
1518     __unused struct nfsexstuff *exp,
1519     struct nfsrv_descript *nd, NFSPROC_T *p)
1520 {
1521 	struct nfslock *lop;
1522 	struct nfsstate *new_stp = *new_stpp;
1523 	struct nfslock *new_lop = *new_lopp;
1524 	struct nfsstate *tstp, *mystp, *nstp;
1525 	int specialid = 0;
1526 	struct nfslockfile *lfp;
1527 	struct nfslock *other_lop = NULL;
1528 	struct nfsstate *stp, *lckstp = NULL;
1529 	struct nfsclient *clp = NULL;
1530 	u_int32_t bits;
1531 	int error = 0, haslock = 0, ret, reterr;
1532 	int getlckret, delegation = 0, filestruct_locked, vnode_unlocked = 0;
1533 	fhandle_t nfh;
1534 	uint64_t first, end;
1535 	uint32_t lock_flags;
1536 
1537 	if (new_stp->ls_flags & (NFSLCK_CHECK | NFSLCK_SETATTR)) {
1538 		/*
1539 		 * Note the special cases of "all 1s" or "all 0s" stateids and
1540 		 * let reads with all 1s go ahead.
1541 		 */
1542 		if (new_stp->ls_stateid.seqid == 0x0 &&
1543 		    new_stp->ls_stateid.other[0] == 0x0 &&
1544 		    new_stp->ls_stateid.other[1] == 0x0 &&
1545 		    new_stp->ls_stateid.other[2] == 0x0)
1546 			specialid = 1;
1547 		else if (new_stp->ls_stateid.seqid == 0xffffffff &&
1548 		    new_stp->ls_stateid.other[0] == 0xffffffff &&
1549 		    new_stp->ls_stateid.other[1] == 0xffffffff &&
1550 		    new_stp->ls_stateid.other[2] == 0xffffffff)
1551 			specialid = 2;
1552 	}
1553 
1554 	/*
1555 	 * Check for restart conditions (client and server).
1556 	 */
1557 	error = nfsrv_checkrestart(clientid, new_stp->ls_flags,
1558 	    &new_stp->ls_stateid, specialid);
1559 	if (error)
1560 		goto out;
1561 
1562 	/*
1563 	 * Check for state resource limit exceeded.
1564 	 */
1565 	if ((new_stp->ls_flags & NFSLCK_LOCK) &&
1566 	    nfsrv_openpluslock > nfsrv_v4statelimit) {
1567 		error = NFSERR_RESOURCE;
1568 		goto out;
1569 	}
1570 
1571 	/*
1572 	 * For the lock case, get another nfslock structure,
1573 	 * just in case we need it.
1574 	 * Malloc now, before we start sifting through the linked lists,
1575 	 * in case we have to wait for memory.
1576 	 */
1577 tryagain:
1578 	if (new_stp->ls_flags & NFSLCK_LOCK)
1579 		MALLOC(other_lop, struct nfslock *, sizeof (struct nfslock),
1580 		    M_NFSDLOCK, M_WAITOK);
1581 	filestruct_locked = 0;
1582 	reterr = 0;
1583 	lfp = NULL;
1584 
1585 	/*
1586 	 * Get the lockfile structure for CFH now, so we can do a sanity
1587 	 * check against the stateid, before incrementing the seqid#, since
1588 	 * we want to return NFSERR_BADSTATEID on failure and the seqid#
1589 	 * shouldn't be incremented for this case.
1590 	 * If nfsrv_getlockfile() returns -1, it means "not found", which
1591 	 * will be handled later.
1592 	 * If we are doing Lock/LockU and local locking is enabled, sleep
1593 	 * lock the nfslockfile structure.
1594 	 */
1595 	getlckret = nfsrv_getlockfh(vp, new_stp->ls_flags, NULL, &nfh, p);
1596 	NFSLOCKSTATE();
1597 	if (getlckret == 0) {
1598 		if ((new_stp->ls_flags & (NFSLCK_LOCK | NFSLCK_UNLOCK)) != 0 &&
1599 		    nfsrv_dolocallocks != 0 && nd->nd_repstat == 0) {
1600 			getlckret = nfsrv_getlockfile(new_stp->ls_flags, NULL,
1601 			    &lfp, &nfh, 1);
1602 			if (getlckret == 0)
1603 				filestruct_locked = 1;
1604 		} else
1605 			getlckret = nfsrv_getlockfile(new_stp->ls_flags, NULL,
1606 			    &lfp, &nfh, 0);
1607 	}
1608 	if (getlckret != 0 && getlckret != -1)
1609 		reterr = getlckret;
1610 
1611 	if (filestruct_locked != 0) {
1612 		LIST_INIT(&lfp->lf_rollback);
1613 		if ((new_stp->ls_flags & NFSLCK_LOCK)) {
1614 			/*
1615 			 * For local locking, do the advisory locking now, so
1616 			 * that any conflict can be detected. A failure later
1617 			 * can be rolled back locally. If an error is returned,
1618 			 * struct nfslockfile has been unlocked and any local
1619 			 * locking rolled back.
1620 			 */
1621 			NFSUNLOCKSTATE();
1622 			if (vnode_unlocked == 0) {
1623 				ASSERT_VOP_ELOCKED(vp, "nfsrv_lockctrl1");
1624 				vnode_unlocked = 1;
1625 				NFSVOPUNLOCK(vp, 0);
1626 			}
1627 			reterr = nfsrv_locallock(vp, lfp,
1628 			    (new_lop->lo_flags & (NFSLCK_READ | NFSLCK_WRITE)),
1629 			    new_lop->lo_first, new_lop->lo_end, cfp, p);
1630 			NFSLOCKSTATE();
1631 		}
1632 	}
1633 
1634 	if (specialid == 0) {
1635 	    if (new_stp->ls_flags & NFSLCK_TEST) {
1636 		/*
1637 		 * RFC 3530 does not list LockT as an op that renews a
1638 		 * lease, but the concensus seems to be that it is ok
1639 		 * for a server to do so.
1640 		 */
1641 		error = nfsrv_getclient(clientid, CLOPS_RENEW, &clp, NULL,
1642 		    (nfsquad_t)((u_quad_t)0), 0, nd, p);
1643 
1644 		/*
1645 		 * Since NFSERR_EXPIRED, NFSERR_ADMINREVOKED are not valid
1646 		 * error returns for LockT, just go ahead and test for a lock,
1647 		 * since there are no locks for this client, but other locks
1648 		 * can conflict. (ie. same client will always be false)
1649 		 */
1650 		if (error == NFSERR_EXPIRED || error == NFSERR_ADMINREVOKED)
1651 		    error = 0;
1652 		lckstp = new_stp;
1653 	    } else {
1654 	      error = nfsrv_getclient(clientid, CLOPS_RENEW, &clp, NULL,
1655 		(nfsquad_t)((u_quad_t)0), 0, nd, p);
1656 	      if (error == 0)
1657 		/*
1658 		 * Look up the stateid
1659 		 */
1660 		error = nfsrv_getstate(clp, &new_stp->ls_stateid,
1661 		  new_stp->ls_flags, &stp);
1662 	      /*
1663 	       * do some sanity checks for an unconfirmed open or a
1664 	       * stateid that refers to the wrong file, for an open stateid
1665 	       */
1666 	      if (error == 0 && (stp->ls_flags & NFSLCK_OPEN) &&
1667 		  ((stp->ls_openowner->ls_flags & NFSLCK_NEEDSCONFIRM) ||
1668 		   (getlckret == 0 && stp->ls_lfp != lfp))){
1669 		      /*
1670 		       * NFSLCK_SETATTR should return OK rather than NFSERR_BADSTATEID
1671 		       * The only exception is using SETATTR with SIZE.
1672 		       * */
1673                     if ((new_stp->ls_flags &
1674                          (NFSLCK_SETATTR | NFSLCK_CHECK)) != NFSLCK_SETATTR)
1675 			     error = NFSERR_BADSTATEID;
1676 	      }
1677 
1678 		if (error == 0 &&
1679 		  (stp->ls_flags & (NFSLCK_DELEGREAD | NFSLCK_DELEGWRITE)) &&
1680 		  getlckret == 0 && stp->ls_lfp != lfp)
1681 			error = NFSERR_BADSTATEID;
1682 
1683 	      /*
1684 	       * If the lockowner stateid doesn't refer to the same file,
1685 	       * I believe that is considered ok, since some clients will
1686 	       * only create a single lockowner and use that for all locks
1687 	       * on all files.
1688 	       * For now, log it as a diagnostic, instead of considering it
1689 	       * a BadStateid.
1690 	       */
1691 	      if (error == 0 && (stp->ls_flags &
1692 		  (NFSLCK_OPEN | NFSLCK_DELEGREAD | NFSLCK_DELEGWRITE)) == 0 &&
1693 		  getlckret == 0 && stp->ls_lfp != lfp) {
1694 #ifdef DIAGNOSTIC
1695 		  printf("Got a lock statid for different file open\n");
1696 #endif
1697 		  /*
1698 		  error = NFSERR_BADSTATEID;
1699 		  */
1700 	      }
1701 
1702 	      if (error == 0) {
1703 		    if (new_stp->ls_flags & NFSLCK_OPENTOLOCK) {
1704 			/*
1705 			 * If haslock set, we've already checked the seqid.
1706 			 */
1707 			if (!haslock) {
1708 			    if (stp->ls_flags & NFSLCK_OPEN)
1709 				error = nfsrv_checkseqid(nd, new_stp->ls_seq,
1710 				    stp->ls_openowner, new_stp->ls_op);
1711 			    else
1712 				error = NFSERR_BADSTATEID;
1713 			}
1714 			if (!error)
1715 			    nfsrv_getowner(&stp->ls_open, new_stp, &lckstp);
1716 			if (lckstp)
1717 			    /*
1718 			     * I believe this should be an error, but it
1719 			     * isn't obvious what NFSERR_xxx would be
1720 			     * appropriate, so I'll use NFSERR_INVAL for now.
1721 			     */
1722 			    error = NFSERR_INVAL;
1723 			else
1724 			    lckstp = new_stp;
1725 		    } else if (new_stp->ls_flags&(NFSLCK_LOCK|NFSLCK_UNLOCK)) {
1726 			/*
1727 			 * If haslock set, ditto above.
1728 			 */
1729 			if (!haslock) {
1730 			    if (stp->ls_flags & NFSLCK_OPEN)
1731 				error = NFSERR_BADSTATEID;
1732 			    else
1733 				error = nfsrv_checkseqid(nd, new_stp->ls_seq,
1734 				    stp, new_stp->ls_op);
1735 			}
1736 			lckstp = stp;
1737 		    } else {
1738 			lckstp = stp;
1739 		    }
1740 	      }
1741 	      /*
1742 	       * If the seqid part of the stateid isn't the same, return
1743 	       * NFSERR_OLDSTATEID for cases other than I/O Ops.
1744 	       * For I/O Ops, only return NFSERR_OLDSTATEID if
1745 	       * nfsrv_returnoldstateid is set. (The concensus on the email
1746 	       * list was that most clients would prefer to not receive
1747 	       * NFSERR_OLDSTATEID for I/O Ops, but the RFC suggests that that
1748 	       * is what will happen, so I use the nfsrv_returnoldstateid to
1749 	       * allow for either server configuration.)
1750 	       */
1751 	      if (!error && stp->ls_stateid.seqid!=new_stp->ls_stateid.seqid &&
1752 		  (((nd->nd_flag & ND_NFSV41) == 0 &&
1753 		   (!(new_stp->ls_flags & NFSLCK_CHECK) ||
1754 		    nfsrv_returnoldstateid)) ||
1755 		   ((nd->nd_flag & ND_NFSV41) != 0 &&
1756 		    new_stp->ls_stateid.seqid != 0)))
1757 		    error = NFSERR_OLDSTATEID;
1758 	    }
1759 	}
1760 
1761 	/*
1762 	 * Now we can check for grace.
1763 	 */
1764 	if (!error)
1765 		error = nfsrv_checkgrace(nd, clp, new_stp->ls_flags);
1766 	if ((new_stp->ls_flags & NFSLCK_RECLAIM) && !error &&
1767 		nfsrv_checkstable(clp))
1768 		error = NFSERR_NOGRACE;
1769 	/*
1770 	 * If we successfully Reclaimed state, note that.
1771 	 */
1772 	if ((new_stp->ls_flags & NFSLCK_RECLAIM) && !error)
1773 		nfsrv_markstable(clp);
1774 
1775 	/*
1776 	 * At this point, either error == NFSERR_BADSTATEID or the
1777 	 * seqid# has been updated, so we can return any error.
1778 	 * If error == 0, there may be an error in:
1779 	 *    nd_repstat - Set by the calling function.
1780 	 *    reterr - Set above, if getting the nfslockfile structure
1781 	 *       or acquiring the local lock failed.
1782 	 *    (If both of these are set, nd_repstat should probably be
1783 	 *     returned, since that error was detected before this
1784 	 *     function call.)
1785 	 */
1786 	if (error != 0 || nd->nd_repstat != 0 || reterr != 0) {
1787 		if (error == 0) {
1788 			if (nd->nd_repstat != 0)
1789 				error = nd->nd_repstat;
1790 			else
1791 				error = reterr;
1792 		}
1793 		if (filestruct_locked != 0) {
1794 			/* Roll back local locks. */
1795 			NFSUNLOCKSTATE();
1796 			if (vnode_unlocked == 0) {
1797 				ASSERT_VOP_ELOCKED(vp, "nfsrv_lockctrl2");
1798 				vnode_unlocked = 1;
1799 				NFSVOPUNLOCK(vp, 0);
1800 			}
1801 			nfsrv_locallock_rollback(vp, lfp, p);
1802 			NFSLOCKSTATE();
1803 			nfsrv_unlocklf(lfp);
1804 		}
1805 		NFSUNLOCKSTATE();
1806 		goto out;
1807 	}
1808 
1809 	/*
1810 	 * Check the nfsrv_getlockfile return.
1811 	 * Returned -1 if no structure found.
1812 	 */
1813 	if (getlckret == -1) {
1814 		error = NFSERR_EXPIRED;
1815 		/*
1816 		 * Called from lockt, so no lock is OK.
1817 		 */
1818 		if (new_stp->ls_flags & NFSLCK_TEST) {
1819 			error = 0;
1820 		} else if (new_stp->ls_flags &
1821 		    (NFSLCK_CHECK | NFSLCK_SETATTR)) {
1822 			/*
1823 			 * Called to check for a lock, OK if the stateid is all
1824 			 * 1s or all 0s, but there should be an nfsstate
1825 			 * otherwise.
1826 			 * (ie. If there is no open, I'll assume no share
1827 			 *  deny bits.)
1828 			 */
1829 			if (specialid)
1830 				error = 0;
1831 			else
1832 				error = NFSERR_BADSTATEID;
1833 		}
1834 		NFSUNLOCKSTATE();
1835 		goto out;
1836 	}
1837 
1838 	/*
1839 	 * For NFSLCK_CHECK and NFSLCK_LOCK, test for a share conflict.
1840 	 * For NFSLCK_CHECK, allow a read if write access is granted,
1841 	 * but check for a deny. For NFSLCK_LOCK, require correct access,
1842 	 * which implies a conflicting deny can't exist.
1843 	 */
1844 	if (new_stp->ls_flags & (NFSLCK_CHECK | NFSLCK_LOCK)) {
1845 	    /*
1846 	     * Four kinds of state id:
1847 	     * - specialid (all 0s or all 1s), only for NFSLCK_CHECK
1848 	     * - stateid for an open
1849 	     * - stateid for a delegation
1850 	     * - stateid for a lock owner
1851 	     */
1852 	    if (!specialid) {
1853 		if (stp->ls_flags & (NFSLCK_DELEGREAD | NFSLCK_DELEGWRITE)) {
1854 		    delegation = 1;
1855 		    mystp = stp;
1856 		    nfsrv_delaydelegtimeout(stp);
1857 	        } else if (stp->ls_flags & NFSLCK_OPEN) {
1858 		    mystp = stp;
1859 		} else {
1860 		    mystp = stp->ls_openstp;
1861 		}
1862 		/*
1863 		 * If locking or checking, require correct access
1864 		 * bit set.
1865 		 */
1866 		if (((new_stp->ls_flags & NFSLCK_LOCK) &&
1867 		     !((new_lop->lo_flags >> NFSLCK_LOCKSHIFT) &
1868 		       mystp->ls_flags & NFSLCK_ACCESSBITS)) ||
1869 		    ((new_stp->ls_flags & (NFSLCK_CHECK|NFSLCK_READACCESS)) ==
1870 		      (NFSLCK_CHECK | NFSLCK_READACCESS) &&
1871 		     !(mystp->ls_flags & NFSLCK_READACCESS)) ||
1872 		    ((new_stp->ls_flags & (NFSLCK_CHECK|NFSLCK_WRITEACCESS)) ==
1873 		      (NFSLCK_CHECK | NFSLCK_WRITEACCESS) &&
1874 		     !(mystp->ls_flags & NFSLCK_WRITEACCESS))) {
1875 			if (filestruct_locked != 0) {
1876 				/* Roll back local locks. */
1877 				NFSUNLOCKSTATE();
1878 				if (vnode_unlocked == 0) {
1879 					ASSERT_VOP_ELOCKED(vp,
1880 					    "nfsrv_lockctrl3");
1881 					vnode_unlocked = 1;
1882 					NFSVOPUNLOCK(vp, 0);
1883 				}
1884 				nfsrv_locallock_rollback(vp, lfp, p);
1885 				NFSLOCKSTATE();
1886 				nfsrv_unlocklf(lfp);
1887 			}
1888 			NFSUNLOCKSTATE();
1889 			error = NFSERR_OPENMODE;
1890 			goto out;
1891 		}
1892 	    } else
1893 		mystp = NULL;
1894 	    if ((new_stp->ls_flags & NFSLCK_CHECK) && !delegation) {
1895 		/*
1896 		 * Check for a conflicting deny bit.
1897 		 */
1898 		LIST_FOREACH(tstp, &lfp->lf_open, ls_file) {
1899 		    if (tstp != mystp) {
1900 			bits = tstp->ls_flags;
1901 			bits >>= NFSLCK_SHIFT;
1902 			if (new_stp->ls_flags & bits & NFSLCK_ACCESSBITS) {
1903 			    KASSERT(vnode_unlocked == 0,
1904 				("nfsrv_lockctrl: vnode unlocked1"));
1905 			    ret = nfsrv_clientconflict(tstp->ls_clp, &haslock,
1906 				vp, p);
1907 			    if (ret == 1) {
1908 				/*
1909 				* nfsrv_clientconflict unlocks state
1910 				 * when it returns non-zero.
1911 				 */
1912 				lckstp = NULL;
1913 				goto tryagain;
1914 			    }
1915 			    if (ret == 0)
1916 				NFSUNLOCKSTATE();
1917 			    if (ret == 2)
1918 				error = NFSERR_PERM;
1919 			    else
1920 				error = NFSERR_OPENMODE;
1921 			    goto out;
1922 			}
1923 		    }
1924 		}
1925 
1926 		/* We're outta here */
1927 		NFSUNLOCKSTATE();
1928 		goto out;
1929 	    }
1930 	}
1931 
1932 	/*
1933 	 * For setattr, just get rid of all the Delegations for other clients.
1934 	 */
1935 	if (new_stp->ls_flags & NFSLCK_SETATTR) {
1936 		KASSERT(vnode_unlocked == 0,
1937 		    ("nfsrv_lockctrl: vnode unlocked2"));
1938 		ret = nfsrv_cleandeleg(vp, lfp, clp, &haslock, p);
1939 		if (ret) {
1940 			/*
1941 			 * nfsrv_cleandeleg() unlocks state when it
1942 			 * returns non-zero.
1943 			 */
1944 			if (ret == -1) {
1945 				lckstp = NULL;
1946 				goto tryagain;
1947 			}
1948 			error = ret;
1949 			goto out;
1950 		}
1951 		if (!(new_stp->ls_flags & NFSLCK_CHECK) ||
1952 		    (LIST_EMPTY(&lfp->lf_open) && LIST_EMPTY(&lfp->lf_lock) &&
1953 		     LIST_EMPTY(&lfp->lf_deleg))) {
1954 			NFSUNLOCKSTATE();
1955 			goto out;
1956 		}
1957 	}
1958 
1959 	/*
1960 	 * Check for a conflicting delegation. If one is found, call
1961 	 * nfsrv_delegconflict() to handle it. If the v4root lock hasn't
1962 	 * been set yet, it will get the lock. Otherwise, it will recall
1963 	 * the delegation. Then, we try try again...
1964 	 * I currently believe the conflict algorithm to be:
1965 	 * For Lock Ops (Lock/LockT/LockU)
1966 	 * - there is a conflict iff a different client has a write delegation
1967 	 * For Reading (Read Op)
1968 	 * - there is a conflict iff a different client has a write delegation
1969 	 *   (the specialids are always a different client)
1970 	 * For Writing (Write/Setattr of size)
1971 	 * - there is a conflict if a different client has any delegation
1972 	 * - there is a conflict if the same client has a read delegation
1973 	 *   (I don't understand why this isn't allowed, but that seems to be
1974 	 *    the current concensus?)
1975 	 */
1976 	tstp = LIST_FIRST(&lfp->lf_deleg);
1977 	while (tstp != LIST_END(&lfp->lf_deleg)) {
1978 	    nstp = LIST_NEXT(tstp, ls_file);
1979 	    if ((((new_stp->ls_flags&(NFSLCK_LOCK|NFSLCK_UNLOCK|NFSLCK_TEST))||
1980 		 ((new_stp->ls_flags & NFSLCK_CHECK) &&
1981 		  (new_lop->lo_flags & NFSLCK_READ))) &&
1982 		  clp != tstp->ls_clp &&
1983 		 (tstp->ls_flags & NFSLCK_DELEGWRITE)) ||
1984 		 ((new_stp->ls_flags & NFSLCK_CHECK) &&
1985 		   (new_lop->lo_flags & NFSLCK_WRITE) &&
1986 		  (clp != tstp->ls_clp ||
1987 		   (tstp->ls_flags & NFSLCK_DELEGREAD)))) {
1988 		ret = 0;
1989 		if (filestruct_locked != 0) {
1990 			/* Roll back local locks. */
1991 			NFSUNLOCKSTATE();
1992 			if (vnode_unlocked == 0) {
1993 				ASSERT_VOP_ELOCKED(vp, "nfsrv_lockctrl4");
1994 				NFSVOPUNLOCK(vp, 0);
1995 			}
1996 			nfsrv_locallock_rollback(vp, lfp, p);
1997 			NFSLOCKSTATE();
1998 			nfsrv_unlocklf(lfp);
1999 			NFSUNLOCKSTATE();
2000 			NFSVOPLOCK(vp, LK_EXCLUSIVE | LK_RETRY);
2001 			vnode_unlocked = 0;
2002 			if ((vp->v_iflag & VI_DOOMED) != 0)
2003 				ret = NFSERR_SERVERFAULT;
2004 			NFSLOCKSTATE();
2005 		}
2006 		if (ret == 0)
2007 			ret = nfsrv_delegconflict(tstp, &haslock, p, vp);
2008 		if (ret) {
2009 		    /*
2010 		     * nfsrv_delegconflict unlocks state when it
2011 		     * returns non-zero, which it always does.
2012 		     */
2013 		    if (other_lop) {
2014 			FREE((caddr_t)other_lop, M_NFSDLOCK);
2015 			other_lop = NULL;
2016 		    }
2017 		    if (ret == -1) {
2018 			lckstp = NULL;
2019 			goto tryagain;
2020 		    }
2021 		    error = ret;
2022 		    goto out;
2023 		}
2024 		/* Never gets here. */
2025 	    }
2026 	    tstp = nstp;
2027 	}
2028 
2029 	/*
2030 	 * Handle the unlock case by calling nfsrv_updatelock().
2031 	 * (Should I have done some access checking above for unlock? For now,
2032 	 *  just let it happen.)
2033 	 */
2034 	if (new_stp->ls_flags & NFSLCK_UNLOCK) {
2035 		first = new_lop->lo_first;
2036 		end = new_lop->lo_end;
2037 		nfsrv_updatelock(stp, new_lopp, &other_lop, lfp);
2038 		stateidp->seqid = ++(stp->ls_stateid.seqid);
2039 		if ((nd->nd_flag & ND_NFSV41) != 0 && stateidp->seqid == 0)
2040 			stateidp->seqid = stp->ls_stateid.seqid = 1;
2041 		stateidp->other[0] = stp->ls_stateid.other[0];
2042 		stateidp->other[1] = stp->ls_stateid.other[1];
2043 		stateidp->other[2] = stp->ls_stateid.other[2];
2044 		if (filestruct_locked != 0) {
2045 			NFSUNLOCKSTATE();
2046 			if (vnode_unlocked == 0) {
2047 				ASSERT_VOP_ELOCKED(vp, "nfsrv_lockctrl5");
2048 				vnode_unlocked = 1;
2049 				NFSVOPUNLOCK(vp, 0);
2050 			}
2051 			/* Update the local locks. */
2052 			nfsrv_localunlock(vp, lfp, first, end, p);
2053 			NFSLOCKSTATE();
2054 			nfsrv_unlocklf(lfp);
2055 		}
2056 		NFSUNLOCKSTATE();
2057 		goto out;
2058 	}
2059 
2060 	/*
2061 	 * Search for a conflicting lock. A lock conflicts if:
2062 	 * - the lock range overlaps and
2063 	 * - at least one lock is a write lock and
2064 	 * - it is not owned by the same lock owner
2065 	 */
2066 	if (!delegation) {
2067 	  LIST_FOREACH(lop, &lfp->lf_lock, lo_lckfile) {
2068 	    if (new_lop->lo_end > lop->lo_first &&
2069 		new_lop->lo_first < lop->lo_end &&
2070 		(new_lop->lo_flags == NFSLCK_WRITE ||
2071 		 lop->lo_flags == NFSLCK_WRITE) &&
2072 		lckstp != lop->lo_stp &&
2073 		(clp != lop->lo_stp->ls_clp ||
2074 		 lckstp->ls_ownerlen != lop->lo_stp->ls_ownerlen ||
2075 		 NFSBCMP(lckstp->ls_owner, lop->lo_stp->ls_owner,
2076 		    lckstp->ls_ownerlen))) {
2077 		if (other_lop) {
2078 		    FREE((caddr_t)other_lop, M_NFSDLOCK);
2079 		    other_lop = NULL;
2080 		}
2081 		if (vnode_unlocked != 0)
2082 		    ret = nfsrv_clientconflict(lop->lo_stp->ls_clp, &haslock,
2083 			NULL, p);
2084 		else
2085 		    ret = nfsrv_clientconflict(lop->lo_stp->ls_clp, &haslock,
2086 			vp, p);
2087 		if (ret == 1) {
2088 		    if (filestruct_locked != 0) {
2089 			if (vnode_unlocked == 0) {
2090 				ASSERT_VOP_ELOCKED(vp, "nfsrv_lockctrl6");
2091 				NFSVOPUNLOCK(vp, 0);
2092 			}
2093 			/* Roll back local locks. */
2094 			nfsrv_locallock_rollback(vp, lfp, p);
2095 			NFSLOCKSTATE();
2096 			nfsrv_unlocklf(lfp);
2097 			NFSUNLOCKSTATE();
2098 			NFSVOPLOCK(vp, LK_EXCLUSIVE | LK_RETRY);
2099 			vnode_unlocked = 0;
2100 			if ((vp->v_iflag & VI_DOOMED) != 0) {
2101 				error = NFSERR_SERVERFAULT;
2102 				goto out;
2103 			}
2104 		    }
2105 		    /*
2106 		     * nfsrv_clientconflict() unlocks state when it
2107 		     * returns non-zero.
2108 		     */
2109 		    lckstp = NULL;
2110 		    goto tryagain;
2111 		}
2112 		/*
2113 		 * Found a conflicting lock, so record the conflict and
2114 		 * return the error.
2115 		 */
2116 		if (cfp != NULL && ret == 0) {
2117 		    cfp->cl_clientid.lval[0]=lop->lo_stp->ls_stateid.other[0];
2118 		    cfp->cl_clientid.lval[1]=lop->lo_stp->ls_stateid.other[1];
2119 		    cfp->cl_first = lop->lo_first;
2120 		    cfp->cl_end = lop->lo_end;
2121 		    cfp->cl_flags = lop->lo_flags;
2122 		    cfp->cl_ownerlen = lop->lo_stp->ls_ownerlen;
2123 		    NFSBCOPY(lop->lo_stp->ls_owner, cfp->cl_owner,
2124 			cfp->cl_ownerlen);
2125 		}
2126 		if (ret == 2)
2127 		    error = NFSERR_PERM;
2128 		else if (new_stp->ls_flags & NFSLCK_RECLAIM)
2129 		    error = NFSERR_RECLAIMCONFLICT;
2130 		else if (new_stp->ls_flags & NFSLCK_CHECK)
2131 		    error = NFSERR_LOCKED;
2132 		else
2133 		    error = NFSERR_DENIED;
2134 		if (filestruct_locked != 0 && ret == 0) {
2135 			/* Roll back local locks. */
2136 			NFSUNLOCKSTATE();
2137 			if (vnode_unlocked == 0) {
2138 				ASSERT_VOP_ELOCKED(vp, "nfsrv_lockctrl7");
2139 				vnode_unlocked = 1;
2140 				NFSVOPUNLOCK(vp, 0);
2141 			}
2142 			nfsrv_locallock_rollback(vp, lfp, p);
2143 			NFSLOCKSTATE();
2144 			nfsrv_unlocklf(lfp);
2145 		}
2146 		if (ret == 0)
2147 			NFSUNLOCKSTATE();
2148 		goto out;
2149 	    }
2150 	  }
2151 	}
2152 
2153 	/*
2154 	 * We only get here if there was no lock that conflicted.
2155 	 */
2156 	if (new_stp->ls_flags & (NFSLCK_TEST | NFSLCK_CHECK)) {
2157 		NFSUNLOCKSTATE();
2158 		goto out;
2159 	}
2160 
2161 	/*
2162 	 * We only get here when we are creating or modifying a lock.
2163 	 * There are two variants:
2164 	 * - exist_lock_owner where lock_owner exists
2165 	 * - open_to_lock_owner with new lock_owner
2166 	 */
2167 	first = new_lop->lo_first;
2168 	end = new_lop->lo_end;
2169 	lock_flags = new_lop->lo_flags;
2170 	if (!(new_stp->ls_flags & NFSLCK_OPENTOLOCK)) {
2171 		nfsrv_updatelock(lckstp, new_lopp, &other_lop, lfp);
2172 		stateidp->seqid = ++(lckstp->ls_stateid.seqid);
2173 		if ((nd->nd_flag & ND_NFSV41) != 0 && stateidp->seqid == 0)
2174 			stateidp->seqid = lckstp->ls_stateid.seqid = 1;
2175 		stateidp->other[0] = lckstp->ls_stateid.other[0];
2176 		stateidp->other[1] = lckstp->ls_stateid.other[1];
2177 		stateidp->other[2] = lckstp->ls_stateid.other[2];
2178 	} else {
2179 		/*
2180 		 * The new open_to_lock_owner case.
2181 		 * Link the new nfsstate into the lists.
2182 		 */
2183 		new_stp->ls_seq = new_stp->ls_opentolockseq;
2184 		nfsrvd_refcache(new_stp->ls_op);
2185 		stateidp->seqid = new_stp->ls_stateid.seqid = 1;
2186 		stateidp->other[0] = new_stp->ls_stateid.other[0] =
2187 		    clp->lc_clientid.lval[0];
2188 		stateidp->other[1] = new_stp->ls_stateid.other[1] =
2189 		    clp->lc_clientid.lval[1];
2190 		stateidp->other[2] = new_stp->ls_stateid.other[2] =
2191 		    nfsrv_nextstateindex(clp);
2192 		new_stp->ls_clp = clp;
2193 		LIST_INIT(&new_stp->ls_lock);
2194 		new_stp->ls_openstp = stp;
2195 		new_stp->ls_lfp = lfp;
2196 		nfsrv_insertlock(new_lop, (struct nfslock *)new_stp, new_stp,
2197 		    lfp);
2198 		LIST_INSERT_HEAD(NFSSTATEHASH(clp, new_stp->ls_stateid),
2199 		    new_stp, ls_hash);
2200 		LIST_INSERT_HEAD(&stp->ls_open, new_stp, ls_list);
2201 		*new_lopp = NULL;
2202 		*new_stpp = NULL;
2203 		newnfsstats.srvlockowners++;
2204 		nfsrv_openpluslock++;
2205 	}
2206 	if (filestruct_locked != 0) {
2207 		NFSUNLOCKSTATE();
2208 		nfsrv_locallock_commit(lfp, lock_flags, first, end);
2209 		NFSLOCKSTATE();
2210 		nfsrv_unlocklf(lfp);
2211 	}
2212 	NFSUNLOCKSTATE();
2213 
2214 out:
2215 	if (haslock) {
2216 		NFSLOCKV4ROOTMUTEX();
2217 		nfsv4_unlock(&nfsv4rootfs_lock, 1);
2218 		NFSUNLOCKV4ROOTMUTEX();
2219 	}
2220 	if (vnode_unlocked != 0) {
2221 		NFSVOPLOCK(vp, LK_EXCLUSIVE | LK_RETRY);
2222 		if (error == 0 && (vp->v_iflag & VI_DOOMED) != 0)
2223 			error = NFSERR_SERVERFAULT;
2224 	}
2225 	if (other_lop)
2226 		FREE((caddr_t)other_lop, M_NFSDLOCK);
2227 	NFSEXITCODE2(error, nd);
2228 	return (error);
2229 }
2230 
2231 /*
2232  * Check for state errors for Open.
2233  * repstat is passed back out as an error if more critical errors
2234  * are not detected.
2235  */
2236 APPLESTATIC int
2237 nfsrv_opencheck(nfsquad_t clientid, nfsv4stateid_t *stateidp,
2238     struct nfsstate *new_stp, vnode_t vp, struct nfsrv_descript *nd,
2239     NFSPROC_T *p, int repstat)
2240 {
2241 	struct nfsstate *stp, *nstp;
2242 	struct nfsclient *clp;
2243 	struct nfsstate *ownerstp;
2244 	struct nfslockfile *lfp, *new_lfp;
2245 	int error = 0, haslock = 0, ret, readonly = 0, getfhret = 0;
2246 
2247 	if ((new_stp->ls_flags & NFSLCK_SHAREBITS) == NFSLCK_READACCESS)
2248 		readonly = 1;
2249 	/*
2250 	 * Check for restart conditions (client and server).
2251 	 */
2252 	error = nfsrv_checkrestart(clientid, new_stp->ls_flags,
2253 		&new_stp->ls_stateid, 0);
2254 	if (error)
2255 		goto out;
2256 
2257 	/*
2258 	 * Check for state resource limit exceeded.
2259 	 * Technically this should be SMP protected, but the worst
2260 	 * case error is "out by one or two" on the count when it
2261 	 * returns NFSERR_RESOURCE and the limit is just a rather
2262 	 * arbitrary high water mark, so no harm is done.
2263 	 */
2264 	if (nfsrv_openpluslock > nfsrv_v4statelimit) {
2265 		error = NFSERR_RESOURCE;
2266 		goto out;
2267 	}
2268 
2269 tryagain:
2270 	MALLOC(new_lfp, struct nfslockfile *, sizeof (struct nfslockfile),
2271 	    M_NFSDLOCKFILE, M_WAITOK);
2272 	if (vp)
2273 		getfhret = nfsrv_getlockfh(vp, new_stp->ls_flags, new_lfp,
2274 		    NULL, p);
2275 	NFSLOCKSTATE();
2276 	/*
2277 	 * Get the nfsclient structure.
2278 	 */
2279 	error = nfsrv_getclient(clientid, CLOPS_RENEW, &clp, NULL,
2280 	    (nfsquad_t)((u_quad_t)0), 0, nd, p);
2281 
2282 	/*
2283 	 * Look up the open owner. See if it needs confirmation and
2284 	 * check the seq#, as required.
2285 	 */
2286 	if (!error)
2287 		nfsrv_getowner(&clp->lc_open, new_stp, &ownerstp);
2288 
2289 	if (!error && ownerstp) {
2290 		error = nfsrv_checkseqid(nd, new_stp->ls_seq, ownerstp,
2291 		    new_stp->ls_op);
2292 		/*
2293 		 * If the OpenOwner hasn't been confirmed, assume the
2294 		 * old one was a replay and this one is ok.
2295 		 * See: RFC3530 Sec. 14.2.18.
2296 		 */
2297 		if (error == NFSERR_BADSEQID &&
2298 		    (ownerstp->ls_flags & NFSLCK_NEEDSCONFIRM))
2299 			error = 0;
2300 	}
2301 
2302 	/*
2303 	 * Check for grace.
2304 	 */
2305 	if (!error)
2306 		error = nfsrv_checkgrace(nd, clp, new_stp->ls_flags);
2307 	if ((new_stp->ls_flags & NFSLCK_RECLAIM) && !error &&
2308 		nfsrv_checkstable(clp))
2309 		error = NFSERR_NOGRACE;
2310 
2311 	/*
2312 	 * If none of the above errors occurred, let repstat be
2313 	 * returned.
2314 	 */
2315 	if (repstat && !error)
2316 		error = repstat;
2317 	if (error) {
2318 		NFSUNLOCKSTATE();
2319 		if (haslock) {
2320 			NFSLOCKV4ROOTMUTEX();
2321 			nfsv4_unlock(&nfsv4rootfs_lock, 1);
2322 			NFSUNLOCKV4ROOTMUTEX();
2323 		}
2324 		free((caddr_t)new_lfp, M_NFSDLOCKFILE);
2325 		goto out;
2326 	}
2327 
2328 	/*
2329 	 * If vp == NULL, the file doesn't exist yet, so return ok.
2330 	 * (This always happens on the first pass, so haslock must be 0.)
2331 	 */
2332 	if (vp == NULL) {
2333 		NFSUNLOCKSTATE();
2334 		FREE((caddr_t)new_lfp, M_NFSDLOCKFILE);
2335 		goto out;
2336 	}
2337 
2338 	/*
2339 	 * Get the structure for the underlying file.
2340 	 */
2341 	if (getfhret)
2342 		error = getfhret;
2343 	else
2344 		error = nfsrv_getlockfile(new_stp->ls_flags, &new_lfp, &lfp,
2345 		    NULL, 0);
2346 	if (new_lfp)
2347 		FREE((caddr_t)new_lfp, M_NFSDLOCKFILE);
2348 	if (error) {
2349 		NFSUNLOCKSTATE();
2350 		if (haslock) {
2351 			NFSLOCKV4ROOTMUTEX();
2352 			nfsv4_unlock(&nfsv4rootfs_lock, 1);
2353 			NFSUNLOCKV4ROOTMUTEX();
2354 		}
2355 		goto out;
2356 	}
2357 
2358 	/*
2359 	 * Search for a conflicting open/share.
2360 	 */
2361 	if (new_stp->ls_flags & NFSLCK_DELEGCUR) {
2362 	    /*
2363 	     * For Delegate_Cur, search for the matching Delegation,
2364 	     * which indicates no conflict.
2365 	     * An old delegation should have been recovered by the
2366 	     * client doing a Claim_DELEGATE_Prev, so I won't let
2367 	     * it match and return NFSERR_EXPIRED. Should I let it
2368 	     * match?
2369 	     */
2370 	    LIST_FOREACH(stp, &lfp->lf_deleg, ls_file) {
2371 		if (!(stp->ls_flags & NFSLCK_OLDDELEG) &&
2372 		    (((nd->nd_flag & ND_NFSV41) != 0 &&
2373 		    stateidp->seqid == 0) ||
2374 		    stateidp->seqid == stp->ls_stateid.seqid) &&
2375 		    !NFSBCMP(stateidp->other, stp->ls_stateid.other,
2376 			  NFSX_STATEIDOTHER))
2377 			break;
2378 	    }
2379 	    if (stp == LIST_END(&lfp->lf_deleg) ||
2380 		((new_stp->ls_flags & NFSLCK_WRITEACCESS) &&
2381 		 (stp->ls_flags & NFSLCK_DELEGREAD))) {
2382 		NFSUNLOCKSTATE();
2383 		if (haslock) {
2384 			NFSLOCKV4ROOTMUTEX();
2385 			nfsv4_unlock(&nfsv4rootfs_lock, 1);
2386 			NFSUNLOCKV4ROOTMUTEX();
2387 		}
2388 		error = NFSERR_EXPIRED;
2389 		goto out;
2390 	    }
2391 	}
2392 
2393 	/*
2394 	 * Check for access/deny bit conflicts. I check for the same
2395 	 * owner as well, in case the client didn't bother.
2396 	 */
2397 	LIST_FOREACH(stp, &lfp->lf_open, ls_file) {
2398 		if (!(new_stp->ls_flags & NFSLCK_DELEGCUR) &&
2399 		    (((new_stp->ls_flags & NFSLCK_ACCESSBITS) &
2400 		      ((stp->ls_flags>>NFSLCK_SHIFT) & NFSLCK_ACCESSBITS))||
2401 		     ((stp->ls_flags & NFSLCK_ACCESSBITS) &
2402 		      ((new_stp->ls_flags>>NFSLCK_SHIFT)&NFSLCK_ACCESSBITS)))){
2403 			ret = nfsrv_clientconflict(stp->ls_clp,&haslock,vp,p);
2404 			if (ret == 1) {
2405 				/*
2406 				 * nfsrv_clientconflict() unlocks
2407 				 * state when it returns non-zero.
2408 				 */
2409 				goto tryagain;
2410 			}
2411 			if (ret == 2)
2412 				error = NFSERR_PERM;
2413 			else if (new_stp->ls_flags & NFSLCK_RECLAIM)
2414 				error = NFSERR_RECLAIMCONFLICT;
2415 			else
2416 				error = NFSERR_SHAREDENIED;
2417 			if (ret == 0)
2418 				NFSUNLOCKSTATE();
2419 			if (haslock) {
2420 				NFSLOCKV4ROOTMUTEX();
2421 				nfsv4_unlock(&nfsv4rootfs_lock, 1);
2422 				NFSUNLOCKV4ROOTMUTEX();
2423 			}
2424 			goto out;
2425 		}
2426 	}
2427 
2428 	/*
2429 	 * Check for a conflicting delegation. If one is found, call
2430 	 * nfsrv_delegconflict() to handle it. If the v4root lock hasn't
2431 	 * been set yet, it will get the lock. Otherwise, it will recall
2432 	 * the delegation. Then, we try try again...
2433 	 * (If NFSLCK_DELEGCUR is set, it has a delegation, so there
2434 	 *  isn't a conflict.)
2435 	 * I currently believe the conflict algorithm to be:
2436 	 * For Open with Read Access and Deny None
2437 	 * - there is a conflict iff a different client has a write delegation
2438 	 * For Open with other Write Access or any Deny except None
2439 	 * - there is a conflict if a different client has any delegation
2440 	 * - there is a conflict if the same client has a read delegation
2441 	 *   (The current concensus is that this last case should be
2442 	 *    considered a conflict since the client with a read delegation
2443 	 *    could have done an Open with ReadAccess and WriteDeny
2444 	 *    locally and then not have checked for the WriteDeny.)
2445 	 * Don't check for a Reclaim, since that will be dealt with
2446 	 * by nfsrv_openctrl().
2447 	 */
2448 	if (!(new_stp->ls_flags &
2449 		(NFSLCK_DELEGPREV | NFSLCK_DELEGCUR | NFSLCK_RECLAIM))) {
2450 	    stp = LIST_FIRST(&lfp->lf_deleg);
2451 	    while (stp != LIST_END(&lfp->lf_deleg)) {
2452 		nstp = LIST_NEXT(stp, ls_file);
2453 		if ((readonly && stp->ls_clp != clp &&
2454 		       (stp->ls_flags & NFSLCK_DELEGWRITE)) ||
2455 		    (!readonly && (stp->ls_clp != clp ||
2456 		         (stp->ls_flags & NFSLCK_DELEGREAD)))) {
2457 			ret = nfsrv_delegconflict(stp, &haslock, p, vp);
2458 			if (ret) {
2459 			    /*
2460 			     * nfsrv_delegconflict() unlocks state
2461 			     * when it returns non-zero.
2462 			     */
2463 			    if (ret == -1)
2464 				goto tryagain;
2465 			    error = ret;
2466 			    goto out;
2467 			}
2468 		}
2469 		stp = nstp;
2470 	    }
2471 	}
2472 	NFSUNLOCKSTATE();
2473 	if (haslock) {
2474 		NFSLOCKV4ROOTMUTEX();
2475 		nfsv4_unlock(&nfsv4rootfs_lock, 1);
2476 		NFSUNLOCKV4ROOTMUTEX();
2477 	}
2478 
2479 out:
2480 	NFSEXITCODE2(error, nd);
2481 	return (error);
2482 }
2483 
2484 /*
2485  * Open control function to create/update open state for an open.
2486  */
2487 APPLESTATIC int
2488 nfsrv_openctrl(struct nfsrv_descript *nd, vnode_t vp,
2489     struct nfsstate **new_stpp, nfsquad_t clientid, nfsv4stateid_t *stateidp,
2490     nfsv4stateid_t *delegstateidp, u_int32_t *rflagsp, struct nfsexstuff *exp,
2491     NFSPROC_T *p, u_quad_t filerev)
2492 {
2493 	struct nfsstate *new_stp = *new_stpp;
2494 	struct nfsstate *stp, *nstp;
2495 	struct nfsstate *openstp = NULL, *new_open, *ownerstp, *new_deleg;
2496 	struct nfslockfile *lfp, *new_lfp;
2497 	struct nfsclient *clp;
2498 	int error = 0, haslock = 0, ret, delegate = 1, writedeleg = 1;
2499 	int readonly = 0, cbret = 1, getfhret = 0;
2500 
2501 	if ((new_stp->ls_flags & NFSLCK_SHAREBITS) == NFSLCK_READACCESS)
2502 		readonly = 1;
2503 	/*
2504 	 * Check for restart conditions (client and server).
2505 	 * (Paranoia, should have been detected by nfsrv_opencheck().)
2506 	 * If an error does show up, return NFSERR_EXPIRED, since the
2507 	 * the seqid# has already been incremented.
2508 	 */
2509 	error = nfsrv_checkrestart(clientid, new_stp->ls_flags,
2510 	    &new_stp->ls_stateid, 0);
2511 	if (error) {
2512 		printf("Nfsd: openctrl unexpected restart err=%d\n",
2513 		    error);
2514 		error = NFSERR_EXPIRED;
2515 		goto out;
2516 	}
2517 
2518 tryagain:
2519 	MALLOC(new_lfp, struct nfslockfile *, sizeof (struct nfslockfile),
2520 	    M_NFSDLOCKFILE, M_WAITOK);
2521 	MALLOC(new_open, struct nfsstate *, sizeof (struct nfsstate),
2522 	    M_NFSDSTATE, M_WAITOK);
2523 	MALLOC(new_deleg, struct nfsstate *, sizeof (struct nfsstate),
2524 	    M_NFSDSTATE, M_WAITOK);
2525 	getfhret = nfsrv_getlockfh(vp, new_stp->ls_flags, new_lfp,
2526 	    NULL, p);
2527 	NFSLOCKSTATE();
2528 	/*
2529 	 * Get the client structure. Since the linked lists could be changed
2530 	 * by other nfsd processes if this process does a tsleep(), one of
2531 	 * two things must be done.
2532 	 * 1 - don't tsleep()
2533 	 * or
2534 	 * 2 - get the nfsv4_lock() { indicated by haslock == 1 }
2535 	 *     before using the lists, since this lock stops the other
2536 	 *     nfsd. This should only be used for rare cases, since it
2537 	 *     essentially single threads the nfsd.
2538 	 *     At this time, it is only done for cases where the stable
2539 	 *     storage file must be written prior to completion of state
2540 	 *     expiration.
2541 	 */
2542 	error = nfsrv_getclient(clientid, CLOPS_RENEW, &clp, NULL,
2543 	    (nfsquad_t)((u_quad_t)0), 0, nd, p);
2544 	if (!error && (clp->lc_flags & LCL_NEEDSCBNULL) &&
2545 	    clp->lc_program) {
2546 		/*
2547 		 * This happens on the first open for a client
2548 		 * that supports callbacks.
2549 		 */
2550 		NFSUNLOCKSTATE();
2551 		/*
2552 		 * Although nfsrv_docallback() will sleep, clp won't
2553 		 * go away, since they are only removed when the
2554 		 * nfsv4_lock() has blocked the nfsd threads. The
2555 		 * fields in clp can change, but having multiple
2556 		 * threads do this Null callback RPC should be
2557 		 * harmless.
2558 		 */
2559 		cbret = nfsrv_docallback(clp, NFSV4PROC_CBNULL,
2560 		    NULL, 0, NULL, NULL, NULL, p);
2561 		NFSLOCKSTATE();
2562 		clp->lc_flags &= ~LCL_NEEDSCBNULL;
2563 		if (!cbret)
2564 			clp->lc_flags |= LCL_CALLBACKSON;
2565 	}
2566 
2567 	/*
2568 	 * Look up the open owner. See if it needs confirmation and
2569 	 * check the seq#, as required.
2570 	 */
2571 	if (!error)
2572 		nfsrv_getowner(&clp->lc_open, new_stp, &ownerstp);
2573 
2574 	if (error) {
2575 		NFSUNLOCKSTATE();
2576 		printf("Nfsd: openctrl unexpected state err=%d\n",
2577 			error);
2578 		free((caddr_t)new_lfp, M_NFSDLOCKFILE);
2579 		free((caddr_t)new_open, M_NFSDSTATE);
2580 		free((caddr_t)new_deleg, M_NFSDSTATE);
2581 		if (haslock) {
2582 			NFSLOCKV4ROOTMUTEX();
2583 			nfsv4_unlock(&nfsv4rootfs_lock, 1);
2584 			NFSUNLOCKV4ROOTMUTEX();
2585 		}
2586 		error = NFSERR_EXPIRED;
2587 		goto out;
2588 	}
2589 
2590 	if (new_stp->ls_flags & NFSLCK_RECLAIM)
2591 		nfsrv_markstable(clp);
2592 
2593 	/*
2594 	 * Get the structure for the underlying file.
2595 	 */
2596 	if (getfhret)
2597 		error = getfhret;
2598 	else
2599 		error = nfsrv_getlockfile(new_stp->ls_flags, &new_lfp, &lfp,
2600 		    NULL, 0);
2601 	if (new_lfp)
2602 		FREE((caddr_t)new_lfp, M_NFSDLOCKFILE);
2603 	if (error) {
2604 		NFSUNLOCKSTATE();
2605 		printf("Nfsd openctrl unexpected getlockfile err=%d\n",
2606 		    error);
2607 		free((caddr_t)new_open, M_NFSDSTATE);
2608 		free((caddr_t)new_deleg, M_NFSDSTATE);
2609 		if (haslock) {
2610 			NFSLOCKV4ROOTMUTEX();
2611 			nfsv4_unlock(&nfsv4rootfs_lock, 1);
2612 			NFSUNLOCKV4ROOTMUTEX();
2613 		}
2614 		goto out;
2615 	}
2616 
2617 	/*
2618 	 * Search for a conflicting open/share.
2619 	 */
2620 	if (new_stp->ls_flags & NFSLCK_DELEGCUR) {
2621 	    /*
2622 	     * For Delegate_Cur, search for the matching Delegation,
2623 	     * which indicates no conflict.
2624 	     * An old delegation should have been recovered by the
2625 	     * client doing a Claim_DELEGATE_Prev, so I won't let
2626 	     * it match and return NFSERR_EXPIRED. Should I let it
2627 	     * match?
2628 	     */
2629 	    LIST_FOREACH(stp, &lfp->lf_deleg, ls_file) {
2630 		if (!(stp->ls_flags & NFSLCK_OLDDELEG) &&
2631 		    (((nd->nd_flag & ND_NFSV41) != 0 &&
2632 		    stateidp->seqid == 0) ||
2633 		    stateidp->seqid == stp->ls_stateid.seqid) &&
2634 		    !NFSBCMP(stateidp->other, stp->ls_stateid.other,
2635 			NFSX_STATEIDOTHER))
2636 			break;
2637 	    }
2638 	    if (stp == LIST_END(&lfp->lf_deleg) ||
2639 		((new_stp->ls_flags & NFSLCK_WRITEACCESS) &&
2640 		 (stp->ls_flags & NFSLCK_DELEGREAD))) {
2641 		NFSUNLOCKSTATE();
2642 		printf("Nfsd openctrl unexpected expiry\n");
2643 		free((caddr_t)new_open, M_NFSDSTATE);
2644 		free((caddr_t)new_deleg, M_NFSDSTATE);
2645 		if (haslock) {
2646 			NFSLOCKV4ROOTMUTEX();
2647 			nfsv4_unlock(&nfsv4rootfs_lock, 1);
2648 			NFSUNLOCKV4ROOTMUTEX();
2649 		}
2650 		error = NFSERR_EXPIRED;
2651 		goto out;
2652 	    }
2653 
2654 	    /*
2655 	     * Don't issue a Delegation, since one already exists and
2656 	     * delay delegation timeout, as required.
2657 	     */
2658 	    delegate = 0;
2659 	    nfsrv_delaydelegtimeout(stp);
2660 	}
2661 
2662 	/*
2663 	 * Check for access/deny bit conflicts. I also check for the
2664 	 * same owner, since the client might not have bothered to check.
2665 	 * Also, note an open for the same file and owner, if found,
2666 	 * which is all we do here for Delegate_Cur, since conflict
2667 	 * checking is already done.
2668 	 */
2669 	LIST_FOREACH(stp, &lfp->lf_open, ls_file) {
2670 		if (ownerstp && stp->ls_openowner == ownerstp)
2671 			openstp = stp;
2672 		if (!(new_stp->ls_flags & NFSLCK_DELEGCUR)) {
2673 		    /*
2674 		     * If another client has the file open, the only
2675 		     * delegation that can be issued is a Read delegation
2676 		     * and only if it is a Read open with Deny none.
2677 		     */
2678 		    if (clp != stp->ls_clp) {
2679 			if ((stp->ls_flags & NFSLCK_SHAREBITS) ==
2680 			    NFSLCK_READACCESS)
2681 			    writedeleg = 0;
2682 			else
2683 			    delegate = 0;
2684 		    }
2685 		    if(((new_stp->ls_flags & NFSLCK_ACCESSBITS) &
2686 		        ((stp->ls_flags>>NFSLCK_SHIFT) & NFSLCK_ACCESSBITS))||
2687 		       ((stp->ls_flags & NFSLCK_ACCESSBITS) &
2688 		        ((new_stp->ls_flags>>NFSLCK_SHIFT)&NFSLCK_ACCESSBITS))){
2689 			ret = nfsrv_clientconflict(stp->ls_clp,&haslock,vp,p);
2690 			if (ret == 1) {
2691 				/*
2692 				 * nfsrv_clientconflict() unlocks state
2693 				 * when it returns non-zero.
2694 				 */
2695 				free((caddr_t)new_open, M_NFSDSTATE);
2696 				free((caddr_t)new_deleg, M_NFSDSTATE);
2697 				openstp = NULL;
2698 				goto tryagain;
2699 			}
2700 			if (ret == 2)
2701 				error = NFSERR_PERM;
2702 			else if (new_stp->ls_flags & NFSLCK_RECLAIM)
2703 				error = NFSERR_RECLAIMCONFLICT;
2704 			else
2705 				error = NFSERR_SHAREDENIED;
2706 			if (ret == 0)
2707 				NFSUNLOCKSTATE();
2708 			if (haslock) {
2709 				NFSLOCKV4ROOTMUTEX();
2710 				nfsv4_unlock(&nfsv4rootfs_lock, 1);
2711 				NFSUNLOCKV4ROOTMUTEX();
2712 			}
2713 			free((caddr_t)new_open, M_NFSDSTATE);
2714 			free((caddr_t)new_deleg, M_NFSDSTATE);
2715 			printf("nfsd openctrl unexpected client cnfl\n");
2716 			goto out;
2717 		    }
2718 		}
2719 	}
2720 
2721 	/*
2722 	 * Check for a conflicting delegation. If one is found, call
2723 	 * nfsrv_delegconflict() to handle it. If the v4root lock hasn't
2724 	 * been set yet, it will get the lock. Otherwise, it will recall
2725 	 * the delegation. Then, we try try again...
2726 	 * (If NFSLCK_DELEGCUR is set, it has a delegation, so there
2727 	 *  isn't a conflict.)
2728 	 * I currently believe the conflict algorithm to be:
2729 	 * For Open with Read Access and Deny None
2730 	 * - there is a conflict iff a different client has a write delegation
2731 	 * For Open with other Write Access or any Deny except None
2732 	 * - there is a conflict if a different client has any delegation
2733 	 * - there is a conflict if the same client has a read delegation
2734 	 *   (The current concensus is that this last case should be
2735 	 *    considered a conflict since the client with a read delegation
2736 	 *    could have done an Open with ReadAccess and WriteDeny
2737 	 *    locally and then not have checked for the WriteDeny.)
2738 	 */
2739 	if (!(new_stp->ls_flags & (NFSLCK_DELEGPREV | NFSLCK_DELEGCUR))) {
2740 	    stp = LIST_FIRST(&lfp->lf_deleg);
2741 	    while (stp != LIST_END(&lfp->lf_deleg)) {
2742 		nstp = LIST_NEXT(stp, ls_file);
2743 		if (stp->ls_clp != clp && (stp->ls_flags & NFSLCK_DELEGREAD))
2744 			writedeleg = 0;
2745 		else
2746 			delegate = 0;
2747 		if ((readonly && stp->ls_clp != clp &&
2748 		       (stp->ls_flags & NFSLCK_DELEGWRITE)) ||
2749 		    (!readonly && (stp->ls_clp != clp ||
2750 		         (stp->ls_flags & NFSLCK_DELEGREAD)))) {
2751 		    if (new_stp->ls_flags & NFSLCK_RECLAIM) {
2752 			delegate = 2;
2753 		    } else {
2754 			ret = nfsrv_delegconflict(stp, &haslock, p, vp);
2755 			if (ret) {
2756 			    /*
2757 			     * nfsrv_delegconflict() unlocks state
2758 			     * when it returns non-zero.
2759 			     */
2760 			    printf("Nfsd openctrl unexpected deleg cnfl\n");
2761 			    free((caddr_t)new_open, M_NFSDSTATE);
2762 			    free((caddr_t)new_deleg, M_NFSDSTATE);
2763 			    if (ret == -1) {
2764 				openstp = NULL;
2765 				goto tryagain;
2766 			    }
2767 			    error = ret;
2768 			    goto out;
2769 			}
2770 		    }
2771 		}
2772 		stp = nstp;
2773 	    }
2774 	}
2775 
2776 	/*
2777 	 * We only get here if there was no open that conflicted.
2778 	 * If an open for the owner exists, or in the access/deny bits.
2779 	 * Otherwise it is a new open. If the open_owner hasn't been
2780 	 * confirmed, replace the open with the new one needing confirmation,
2781 	 * otherwise add the open.
2782 	 */
2783 	if (new_stp->ls_flags & NFSLCK_DELEGPREV) {
2784 	    /*
2785 	     * Handle NFSLCK_DELEGPREV by searching the old delegations for
2786 	     * a match. If found, just move the old delegation to the current
2787 	     * delegation list and issue open. If not found, return
2788 	     * NFSERR_EXPIRED.
2789 	     */
2790 	    LIST_FOREACH(stp, &clp->lc_olddeleg, ls_list) {
2791 		if (stp->ls_lfp == lfp) {
2792 		    /* Found it */
2793 		    if (stp->ls_clp != clp)
2794 			panic("olddeleg clp");
2795 		    LIST_REMOVE(stp, ls_list);
2796 		    LIST_REMOVE(stp, ls_hash);
2797 		    stp->ls_flags &= ~NFSLCK_OLDDELEG;
2798 		    stp->ls_stateid.seqid = delegstateidp->seqid = 1;
2799 		    stp->ls_stateid.other[0] = delegstateidp->other[0] =
2800 			clp->lc_clientid.lval[0];
2801 		    stp->ls_stateid.other[1] = delegstateidp->other[1] =
2802 			clp->lc_clientid.lval[1];
2803 		    stp->ls_stateid.other[2] = delegstateidp->other[2] =
2804 			nfsrv_nextstateindex(clp);
2805 		    stp->ls_compref = nd->nd_compref;
2806 		    LIST_INSERT_HEAD(&clp->lc_deleg, stp, ls_list);
2807 		    LIST_INSERT_HEAD(NFSSTATEHASH(clp,
2808 			stp->ls_stateid), stp, ls_hash);
2809 		    if (stp->ls_flags & NFSLCK_DELEGWRITE)
2810 			*rflagsp |= NFSV4OPEN_WRITEDELEGATE;
2811 		    else
2812 			*rflagsp |= NFSV4OPEN_READDELEGATE;
2813 		    clp->lc_delegtime = NFSD_MONOSEC +
2814 			nfsrv_lease + NFSRV_LEASEDELTA;
2815 
2816 		    /*
2817 		     * Now, do the associated open.
2818 		     */
2819 		    new_open->ls_stateid.seqid = 1;
2820 		    new_open->ls_stateid.other[0] = clp->lc_clientid.lval[0];
2821 		    new_open->ls_stateid.other[1] = clp->lc_clientid.lval[1];
2822 		    new_open->ls_stateid.other[2] = nfsrv_nextstateindex(clp);
2823 		    new_open->ls_flags = (new_stp->ls_flags&NFSLCK_DENYBITS)|
2824 			NFSLCK_OPEN;
2825 		    if (stp->ls_flags & NFSLCK_DELEGWRITE)
2826 			new_open->ls_flags |= (NFSLCK_READACCESS |
2827 			    NFSLCK_WRITEACCESS);
2828 		    else
2829 			new_open->ls_flags |= NFSLCK_READACCESS;
2830 		    new_open->ls_uid = new_stp->ls_uid;
2831 		    new_open->ls_lfp = lfp;
2832 		    new_open->ls_clp = clp;
2833 		    LIST_INIT(&new_open->ls_open);
2834 		    LIST_INSERT_HEAD(&lfp->lf_open, new_open, ls_file);
2835 		    LIST_INSERT_HEAD(NFSSTATEHASH(clp, new_open->ls_stateid),
2836 			new_open, ls_hash);
2837 		    /*
2838 		     * and handle the open owner
2839 		     */
2840 		    if (ownerstp) {
2841 			new_open->ls_openowner = ownerstp;
2842 			LIST_INSERT_HEAD(&ownerstp->ls_open,new_open,ls_list);
2843 		    } else {
2844 			new_open->ls_openowner = new_stp;
2845 			new_stp->ls_flags = 0;
2846 			nfsrvd_refcache(new_stp->ls_op);
2847 			new_stp->ls_noopens = 0;
2848 			LIST_INIT(&new_stp->ls_open);
2849 			LIST_INSERT_HEAD(&new_stp->ls_open, new_open, ls_list);
2850 			LIST_INSERT_HEAD(&clp->lc_open, new_stp, ls_list);
2851 			*new_stpp = NULL;
2852 			newnfsstats.srvopenowners++;
2853 			nfsrv_openpluslock++;
2854 		    }
2855 		    openstp = new_open;
2856 		    new_open = NULL;
2857 		    newnfsstats.srvopens++;
2858 		    nfsrv_openpluslock++;
2859 		    break;
2860 		}
2861 	    }
2862 	    if (stp == LIST_END(&clp->lc_olddeleg))
2863 		error = NFSERR_EXPIRED;
2864 	} else if (new_stp->ls_flags & (NFSLCK_DELEGREAD | NFSLCK_DELEGWRITE)) {
2865 	    /*
2866 	     * Scan to see that no delegation for this client and file
2867 	     * doesn't already exist.
2868 	     * There also shouldn't yet be an Open for this file and
2869 	     * openowner.
2870 	     */
2871 	    LIST_FOREACH(stp, &lfp->lf_deleg, ls_file) {
2872 		if (stp->ls_clp == clp)
2873 		    break;
2874 	    }
2875 	    if (stp == LIST_END(&lfp->lf_deleg) && openstp == NULL) {
2876 		/*
2877 		 * This is the Claim_Previous case with a delegation
2878 		 * type != Delegate_None.
2879 		 */
2880 		/*
2881 		 * First, add the delegation. (Although we must issue the
2882 		 * delegation, we can also ask for an immediate return.)
2883 		 */
2884 		new_deleg->ls_stateid.seqid = delegstateidp->seqid = 1;
2885 		new_deleg->ls_stateid.other[0] = delegstateidp->other[0] =
2886 		    clp->lc_clientid.lval[0];
2887 		new_deleg->ls_stateid.other[1] = delegstateidp->other[1] =
2888 		    clp->lc_clientid.lval[1];
2889 		new_deleg->ls_stateid.other[2] = delegstateidp->other[2] =
2890 		    nfsrv_nextstateindex(clp);
2891 		if (new_stp->ls_flags & NFSLCK_DELEGWRITE) {
2892 		    new_deleg->ls_flags = (NFSLCK_DELEGWRITE |
2893 			NFSLCK_READACCESS | NFSLCK_WRITEACCESS);
2894 		    *rflagsp |= NFSV4OPEN_WRITEDELEGATE;
2895 		} else {
2896 		    new_deleg->ls_flags = (NFSLCK_DELEGREAD |
2897 			NFSLCK_READACCESS);
2898 		    *rflagsp |= NFSV4OPEN_READDELEGATE;
2899 		}
2900 		new_deleg->ls_uid = new_stp->ls_uid;
2901 		new_deleg->ls_lfp = lfp;
2902 		new_deleg->ls_clp = clp;
2903 		new_deleg->ls_filerev = filerev;
2904 		new_deleg->ls_compref = nd->nd_compref;
2905 		LIST_INSERT_HEAD(&lfp->lf_deleg, new_deleg, ls_file);
2906 		LIST_INSERT_HEAD(NFSSTATEHASH(clp,
2907 		    new_deleg->ls_stateid), new_deleg, ls_hash);
2908 		LIST_INSERT_HEAD(&clp->lc_deleg, new_deleg, ls_list);
2909 		new_deleg = NULL;
2910 		if (delegate == 2 || nfsrv_issuedelegs == 0 ||
2911 		    (clp->lc_flags & (LCL_CALLBACKSON | LCL_CBDOWN)) !=
2912 		     LCL_CALLBACKSON ||
2913 		    NFSRV_V4DELEGLIMIT(nfsrv_delegatecnt) ||
2914 		    !NFSVNO_DELEGOK(vp))
2915 		    *rflagsp |= NFSV4OPEN_RECALL;
2916 		newnfsstats.srvdelegates++;
2917 		nfsrv_openpluslock++;
2918 		nfsrv_delegatecnt++;
2919 
2920 		/*
2921 		 * Now, do the associated open.
2922 		 */
2923 		new_open->ls_stateid.seqid = 1;
2924 		new_open->ls_stateid.other[0] = clp->lc_clientid.lval[0];
2925 		new_open->ls_stateid.other[1] = clp->lc_clientid.lval[1];
2926 		new_open->ls_stateid.other[2] = nfsrv_nextstateindex(clp);
2927 		new_open->ls_flags = (new_stp->ls_flags & NFSLCK_DENYBITS) |
2928 		    NFSLCK_OPEN;
2929 		if (new_stp->ls_flags & NFSLCK_DELEGWRITE)
2930 			new_open->ls_flags |= (NFSLCK_READACCESS |
2931 			    NFSLCK_WRITEACCESS);
2932 		else
2933 			new_open->ls_flags |= NFSLCK_READACCESS;
2934 		new_open->ls_uid = new_stp->ls_uid;
2935 		new_open->ls_lfp = lfp;
2936 		new_open->ls_clp = clp;
2937 		LIST_INIT(&new_open->ls_open);
2938 		LIST_INSERT_HEAD(&lfp->lf_open, new_open, ls_file);
2939 		LIST_INSERT_HEAD(NFSSTATEHASH(clp, new_open->ls_stateid),
2940 		   new_open, ls_hash);
2941 		/*
2942 		 * and handle the open owner
2943 		 */
2944 		if (ownerstp) {
2945 		    new_open->ls_openowner = ownerstp;
2946 		    LIST_INSERT_HEAD(&ownerstp->ls_open, new_open, ls_list);
2947 		} else {
2948 		    new_open->ls_openowner = new_stp;
2949 		    new_stp->ls_flags = 0;
2950 		    nfsrvd_refcache(new_stp->ls_op);
2951 		    new_stp->ls_noopens = 0;
2952 		    LIST_INIT(&new_stp->ls_open);
2953 		    LIST_INSERT_HEAD(&new_stp->ls_open, new_open, ls_list);
2954 		    LIST_INSERT_HEAD(&clp->lc_open, new_stp, ls_list);
2955 		    *new_stpp = NULL;
2956 		    newnfsstats.srvopenowners++;
2957 		    nfsrv_openpluslock++;
2958 		}
2959 		openstp = new_open;
2960 		new_open = NULL;
2961 		newnfsstats.srvopens++;
2962 		nfsrv_openpluslock++;
2963 	    } else {
2964 		error = NFSERR_RECLAIMCONFLICT;
2965 	    }
2966 	} else if (ownerstp) {
2967 		if (ownerstp->ls_flags & NFSLCK_NEEDSCONFIRM) {
2968 		    /* Replace the open */
2969 		    if (ownerstp->ls_op)
2970 			nfsrvd_derefcache(ownerstp->ls_op);
2971 		    ownerstp->ls_op = new_stp->ls_op;
2972 		    nfsrvd_refcache(ownerstp->ls_op);
2973 		    ownerstp->ls_seq = new_stp->ls_seq;
2974 		    *rflagsp |= NFSV4OPEN_RESULTCONFIRM;
2975 		    stp = LIST_FIRST(&ownerstp->ls_open);
2976 		    stp->ls_flags = (new_stp->ls_flags & NFSLCK_SHAREBITS) |
2977 			NFSLCK_OPEN;
2978 		    stp->ls_stateid.seqid = 1;
2979 		    stp->ls_uid = new_stp->ls_uid;
2980 		    if (lfp != stp->ls_lfp) {
2981 			LIST_REMOVE(stp, ls_file);
2982 			LIST_INSERT_HEAD(&lfp->lf_open, stp, ls_file);
2983 			stp->ls_lfp = lfp;
2984 		    }
2985 		    openstp = stp;
2986 		} else if (openstp) {
2987 		    openstp->ls_flags |= (new_stp->ls_flags & NFSLCK_SHAREBITS);
2988 		    openstp->ls_stateid.seqid++;
2989 		    if ((nd->nd_flag & ND_NFSV41) != 0 &&
2990 			openstp->ls_stateid.seqid == 0)
2991 			openstp->ls_stateid.seqid = 1;
2992 
2993 		    /*
2994 		     * This is where we can choose to issue a delegation.
2995 		     */
2996 		    if (delegate == 0 || writedeleg == 0 ||
2997 			NFSVNO_EXRDONLY(exp) || (readonly != 0 &&
2998 			nfsrv_writedelegifpos == 0) ||
2999 			!NFSVNO_DELEGOK(vp) ||
3000 			(new_stp->ls_flags & NFSLCK_WANTRDELEG) != 0 ||
3001 			(clp->lc_flags & (LCL_CALLBACKSON | LCL_CBDOWN)) !=
3002 			 LCL_CALLBACKSON)
3003 			*rflagsp |= NFSV4OPEN_WDCONTENTION;
3004 		    else if (nfsrv_issuedelegs == 0 ||
3005 			NFSRV_V4DELEGLIMIT(nfsrv_delegatecnt))
3006 			*rflagsp |= NFSV4OPEN_WDRESOURCE;
3007 		    else if ((new_stp->ls_flags & NFSLCK_WANTNODELEG) != 0)
3008 			*rflagsp |= NFSV4OPEN_WDNOTWANTED;
3009 		    else {
3010 			new_deleg->ls_stateid.seqid = delegstateidp->seqid = 1;
3011 			new_deleg->ls_stateid.other[0] = delegstateidp->other[0]
3012 			    = clp->lc_clientid.lval[0];
3013 			new_deleg->ls_stateid.other[1] = delegstateidp->other[1]
3014 			    = clp->lc_clientid.lval[1];
3015 			new_deleg->ls_stateid.other[2] = delegstateidp->other[2]
3016 			    = nfsrv_nextstateindex(clp);
3017 			new_deleg->ls_flags = (NFSLCK_DELEGWRITE |
3018 			    NFSLCK_READACCESS | NFSLCK_WRITEACCESS);
3019 			*rflagsp |= NFSV4OPEN_WRITEDELEGATE;
3020 			new_deleg->ls_uid = new_stp->ls_uid;
3021 			new_deleg->ls_lfp = lfp;
3022 			new_deleg->ls_clp = clp;
3023 			new_deleg->ls_filerev = filerev;
3024 			new_deleg->ls_compref = nd->nd_compref;
3025 			LIST_INSERT_HEAD(&lfp->lf_deleg, new_deleg, ls_file);
3026 			LIST_INSERT_HEAD(NFSSTATEHASH(clp,
3027 			    new_deleg->ls_stateid), new_deleg, ls_hash);
3028 			LIST_INSERT_HEAD(&clp->lc_deleg, new_deleg, ls_list);
3029 			new_deleg = NULL;
3030 			newnfsstats.srvdelegates++;
3031 			nfsrv_openpluslock++;
3032 			nfsrv_delegatecnt++;
3033 		    }
3034 		} else {
3035 		    new_open->ls_stateid.seqid = 1;
3036 		    new_open->ls_stateid.other[0] = clp->lc_clientid.lval[0];
3037 		    new_open->ls_stateid.other[1] = clp->lc_clientid.lval[1];
3038 		    new_open->ls_stateid.other[2] = nfsrv_nextstateindex(clp);
3039 		    new_open->ls_flags = (new_stp->ls_flags & NFSLCK_SHAREBITS)|
3040 			NFSLCK_OPEN;
3041 		    new_open->ls_uid = new_stp->ls_uid;
3042 		    new_open->ls_openowner = ownerstp;
3043 		    new_open->ls_lfp = lfp;
3044 		    new_open->ls_clp = clp;
3045 		    LIST_INIT(&new_open->ls_open);
3046 		    LIST_INSERT_HEAD(&lfp->lf_open, new_open, ls_file);
3047 		    LIST_INSERT_HEAD(&ownerstp->ls_open, new_open, ls_list);
3048 		    LIST_INSERT_HEAD(NFSSTATEHASH(clp, new_open->ls_stateid),
3049 			new_open, ls_hash);
3050 		    openstp = new_open;
3051 		    new_open = NULL;
3052 		    newnfsstats.srvopens++;
3053 		    nfsrv_openpluslock++;
3054 
3055 		    /*
3056 		     * This is where we can choose to issue a delegation.
3057 		     */
3058 		    if (delegate == 0 || (writedeleg == 0 && readonly == 0) ||
3059 			!NFSVNO_DELEGOK(vp) ||
3060 			(clp->lc_flags & (LCL_CALLBACKSON | LCL_CBDOWN)) !=
3061 			 LCL_CALLBACKSON)
3062 			*rflagsp |= NFSV4OPEN_WDCONTENTION;
3063 		    else if (nfsrv_issuedelegs == 0 ||
3064 			NFSRV_V4DELEGLIMIT(nfsrv_delegatecnt))
3065 			*rflagsp |= NFSV4OPEN_WDRESOURCE;
3066 		    else if ((new_stp->ls_flags & NFSLCK_WANTNODELEG) != 0)
3067 			*rflagsp |= NFSV4OPEN_WDNOTWANTED;
3068 		    else {
3069 			new_deleg->ls_stateid.seqid = delegstateidp->seqid = 1;
3070 			new_deleg->ls_stateid.other[0] = delegstateidp->other[0]
3071 			    = clp->lc_clientid.lval[0];
3072 			new_deleg->ls_stateid.other[1] = delegstateidp->other[1]
3073 			    = clp->lc_clientid.lval[1];
3074 			new_deleg->ls_stateid.other[2] = delegstateidp->other[2]
3075 			    = nfsrv_nextstateindex(clp);
3076 			if (writedeleg && !NFSVNO_EXRDONLY(exp) &&
3077 			    (nfsrv_writedelegifpos || !readonly) &&
3078 			    (new_stp->ls_flags & NFSLCK_WANTRDELEG) == 0) {
3079 			    new_deleg->ls_flags = (NFSLCK_DELEGWRITE |
3080 				NFSLCK_READACCESS | NFSLCK_WRITEACCESS);
3081 			    *rflagsp |= NFSV4OPEN_WRITEDELEGATE;
3082 			} else {
3083 			    new_deleg->ls_flags = (NFSLCK_DELEGREAD |
3084 				NFSLCK_READACCESS);
3085 			    *rflagsp |= NFSV4OPEN_READDELEGATE;
3086 			}
3087 			new_deleg->ls_uid = new_stp->ls_uid;
3088 			new_deleg->ls_lfp = lfp;
3089 			new_deleg->ls_clp = clp;
3090 			new_deleg->ls_filerev = filerev;
3091 			new_deleg->ls_compref = nd->nd_compref;
3092 			LIST_INSERT_HEAD(&lfp->lf_deleg, new_deleg, ls_file);
3093 			LIST_INSERT_HEAD(NFSSTATEHASH(clp,
3094 			    new_deleg->ls_stateid), new_deleg, ls_hash);
3095 			LIST_INSERT_HEAD(&clp->lc_deleg, new_deleg, ls_list);
3096 			new_deleg = NULL;
3097 			newnfsstats.srvdelegates++;
3098 			nfsrv_openpluslock++;
3099 			nfsrv_delegatecnt++;
3100 		    }
3101 		}
3102 	} else {
3103 		/*
3104 		 * New owner case. Start the open_owner sequence with a
3105 		 * Needs confirmation (unless a reclaim) and hang the
3106 		 * new open off it.
3107 		 */
3108 		new_open->ls_stateid.seqid = 1;
3109 		new_open->ls_stateid.other[0] = clp->lc_clientid.lval[0];
3110 		new_open->ls_stateid.other[1] = clp->lc_clientid.lval[1];
3111 		new_open->ls_stateid.other[2] = nfsrv_nextstateindex(clp);
3112 		new_open->ls_flags = (new_stp->ls_flags & NFSLCK_SHAREBITS) |
3113 		    NFSLCK_OPEN;
3114 		new_open->ls_uid = new_stp->ls_uid;
3115 		LIST_INIT(&new_open->ls_open);
3116 		new_open->ls_openowner = new_stp;
3117 		new_open->ls_lfp = lfp;
3118 		new_open->ls_clp = clp;
3119 		LIST_INSERT_HEAD(&lfp->lf_open, new_open, ls_file);
3120 		if (new_stp->ls_flags & NFSLCK_RECLAIM) {
3121 			new_stp->ls_flags = 0;
3122 		} else if ((nd->nd_flag & ND_NFSV41) != 0) {
3123 			/* NFSv4.1 never needs confirmation. */
3124 			new_stp->ls_flags = 0;
3125 
3126 			/*
3127 			 * This is where we can choose to issue a delegation.
3128 			 */
3129 			if (delegate && nfsrv_issuedelegs &&
3130 			    (writedeleg || readonly) &&
3131 			    (clp->lc_flags & (LCL_CALLBACKSON | LCL_CBDOWN)) ==
3132 			     LCL_CALLBACKSON &&
3133 			    !NFSRV_V4DELEGLIMIT(nfsrv_delegatecnt) &&
3134 			    NFSVNO_DELEGOK(vp) &&
3135 			    ((nd->nd_flag & ND_NFSV41) == 0 ||
3136 			     (new_stp->ls_flags & NFSLCK_WANTNODELEG) == 0)) {
3137 				new_deleg->ls_stateid.seqid =
3138 				    delegstateidp->seqid = 1;
3139 				new_deleg->ls_stateid.other[0] =
3140 				    delegstateidp->other[0]
3141 				    = clp->lc_clientid.lval[0];
3142 				new_deleg->ls_stateid.other[1] =
3143 				    delegstateidp->other[1]
3144 				    = clp->lc_clientid.lval[1];
3145 				new_deleg->ls_stateid.other[2] =
3146 				    delegstateidp->other[2]
3147 				    = nfsrv_nextstateindex(clp);
3148 				if (writedeleg && !NFSVNO_EXRDONLY(exp) &&
3149 				    (nfsrv_writedelegifpos || !readonly) &&
3150 				    ((nd->nd_flag & ND_NFSV41) == 0 ||
3151 				     (new_stp->ls_flags & NFSLCK_WANTRDELEG) ==
3152 				     0)) {
3153 					new_deleg->ls_flags =
3154 					    (NFSLCK_DELEGWRITE |
3155 					     NFSLCK_READACCESS |
3156 					     NFSLCK_WRITEACCESS);
3157 					*rflagsp |= NFSV4OPEN_WRITEDELEGATE;
3158 				} else {
3159 					new_deleg->ls_flags =
3160 					    (NFSLCK_DELEGREAD |
3161 					     NFSLCK_READACCESS);
3162 					*rflagsp |= NFSV4OPEN_READDELEGATE;
3163 				}
3164 				new_deleg->ls_uid = new_stp->ls_uid;
3165 				new_deleg->ls_lfp = lfp;
3166 				new_deleg->ls_clp = clp;
3167 				new_deleg->ls_filerev = filerev;
3168 				new_deleg->ls_compref = nd->nd_compref;
3169 				LIST_INSERT_HEAD(&lfp->lf_deleg, new_deleg,
3170 				    ls_file);
3171 				LIST_INSERT_HEAD(NFSSTATEHASH(clp,
3172 				    new_deleg->ls_stateid), new_deleg, ls_hash);
3173 				LIST_INSERT_HEAD(&clp->lc_deleg, new_deleg,
3174 				    ls_list);
3175 				new_deleg = NULL;
3176 				newnfsstats.srvdelegates++;
3177 				nfsrv_openpluslock++;
3178 				nfsrv_delegatecnt++;
3179 			}
3180 		} else {
3181 			*rflagsp |= NFSV4OPEN_RESULTCONFIRM;
3182 			new_stp->ls_flags = NFSLCK_NEEDSCONFIRM;
3183 		}
3184 		nfsrvd_refcache(new_stp->ls_op);
3185 		new_stp->ls_noopens = 0;
3186 		LIST_INIT(&new_stp->ls_open);
3187 		LIST_INSERT_HEAD(&new_stp->ls_open, new_open, ls_list);
3188 		LIST_INSERT_HEAD(&clp->lc_open, new_stp, ls_list);
3189 		LIST_INSERT_HEAD(NFSSTATEHASH(clp, new_open->ls_stateid),
3190 		    new_open, ls_hash);
3191 		openstp = new_open;
3192 		new_open = NULL;
3193 		*new_stpp = NULL;
3194 		newnfsstats.srvopens++;
3195 		nfsrv_openpluslock++;
3196 		newnfsstats.srvopenowners++;
3197 		nfsrv_openpluslock++;
3198 	}
3199 	if (!error) {
3200 		stateidp->seqid = openstp->ls_stateid.seqid;
3201 		stateidp->other[0] = openstp->ls_stateid.other[0];
3202 		stateidp->other[1] = openstp->ls_stateid.other[1];
3203 		stateidp->other[2] = openstp->ls_stateid.other[2];
3204 	}
3205 	NFSUNLOCKSTATE();
3206 	if (haslock) {
3207 		NFSLOCKV4ROOTMUTEX();
3208 		nfsv4_unlock(&nfsv4rootfs_lock, 1);
3209 		NFSUNLOCKV4ROOTMUTEX();
3210 	}
3211 	if (new_open)
3212 		FREE((caddr_t)new_open, M_NFSDSTATE);
3213 	if (new_deleg)
3214 		FREE((caddr_t)new_deleg, M_NFSDSTATE);
3215 
3216 out:
3217 	NFSEXITCODE2(error, nd);
3218 	return (error);
3219 }
3220 
3221 /*
3222  * Open update. Does the confirm, downgrade and close.
3223  */
3224 APPLESTATIC int
3225 nfsrv_openupdate(vnode_t vp, struct nfsstate *new_stp, nfsquad_t clientid,
3226     nfsv4stateid_t *stateidp, struct nfsrv_descript *nd, NFSPROC_T *p)
3227 {
3228 	struct nfsstate *stp, *ownerstp;
3229 	struct nfsclient *clp;
3230 	struct nfslockfile *lfp;
3231 	u_int32_t bits;
3232 	int error = 0, gotstate = 0, len = 0;
3233 	u_char client[NFSV4_OPAQUELIMIT];
3234 
3235 	/*
3236 	 * Check for restart conditions (client and server).
3237 	 */
3238 	error = nfsrv_checkrestart(clientid, new_stp->ls_flags,
3239 	    &new_stp->ls_stateid, 0);
3240 	if (error)
3241 		goto out;
3242 
3243 	NFSLOCKSTATE();
3244 	/*
3245 	 * Get the open structure via clientid and stateid.
3246 	 */
3247 	error = nfsrv_getclient(clientid, CLOPS_RENEW, &clp, NULL,
3248 	    (nfsquad_t)((u_quad_t)0), 0, nd, p);
3249 	if (!error)
3250 		error = nfsrv_getstate(clp, &new_stp->ls_stateid,
3251 		    new_stp->ls_flags, &stp);
3252 
3253 	/*
3254 	 * Sanity check the open.
3255 	 */
3256 	if (!error && (!(stp->ls_flags & NFSLCK_OPEN) ||
3257 		(!(new_stp->ls_flags & NFSLCK_CONFIRM) &&
3258 		 (stp->ls_openowner->ls_flags & NFSLCK_NEEDSCONFIRM)) ||
3259 		((new_stp->ls_flags & NFSLCK_CONFIRM) &&
3260 		 (!(stp->ls_openowner->ls_flags & NFSLCK_NEEDSCONFIRM)))))
3261 		error = NFSERR_BADSTATEID;
3262 
3263 	if (!error)
3264 		error = nfsrv_checkseqid(nd, new_stp->ls_seq,
3265 		    stp->ls_openowner, new_stp->ls_op);
3266 	if (!error && stp->ls_stateid.seqid != new_stp->ls_stateid.seqid &&
3267 	    (((nd->nd_flag & ND_NFSV41) == 0 &&
3268 	      !(new_stp->ls_flags & NFSLCK_CONFIRM)) ||
3269 	     ((nd->nd_flag & ND_NFSV41) != 0 &&
3270 	      new_stp->ls_stateid.seqid != 0)))
3271 		error = NFSERR_OLDSTATEID;
3272 	if (!error && vnode_vtype(vp) != VREG) {
3273 		if (vnode_vtype(vp) == VDIR)
3274 			error = NFSERR_ISDIR;
3275 		else
3276 			error = NFSERR_INVAL;
3277 	}
3278 
3279 	if (error) {
3280 		/*
3281 		 * If a client tries to confirm an Open with a bad
3282 		 * seqid# and there are no byte range locks or other Opens
3283 		 * on the openowner, just throw it away, so the next use of the
3284 		 * openowner will start a fresh seq#.
3285 		 */
3286 		if (error == NFSERR_BADSEQID &&
3287 		    (new_stp->ls_flags & NFSLCK_CONFIRM) &&
3288 		    nfsrv_nootherstate(stp))
3289 			nfsrv_freeopenowner(stp->ls_openowner, 0, p);
3290 		NFSUNLOCKSTATE();
3291 		goto out;
3292 	}
3293 
3294 	/*
3295 	 * Set the return stateid.
3296 	 */
3297 	stateidp->seqid = stp->ls_stateid.seqid + 1;
3298 	if ((nd->nd_flag & ND_NFSV41) != 0 && stateidp->seqid == 0)
3299 		stateidp->seqid = 1;
3300 	stateidp->other[0] = stp->ls_stateid.other[0];
3301 	stateidp->other[1] = stp->ls_stateid.other[1];
3302 	stateidp->other[2] = stp->ls_stateid.other[2];
3303 	/*
3304 	 * Now, handle the three cases.
3305 	 */
3306 	if (new_stp->ls_flags & NFSLCK_CONFIRM) {
3307 		/*
3308 		 * If the open doesn't need confirmation, it seems to me that
3309 		 * there is a client error, but I'll just log it and keep going?
3310 		 */
3311 		if (!(stp->ls_openowner->ls_flags & NFSLCK_NEEDSCONFIRM))
3312 			printf("Nfsv4d: stray open confirm\n");
3313 		stp->ls_openowner->ls_flags = 0;
3314 		stp->ls_stateid.seqid++;
3315 		if ((nd->nd_flag & ND_NFSV41) != 0 &&
3316 		    stp->ls_stateid.seqid == 0)
3317 			stp->ls_stateid.seqid = 1;
3318 		if (!(clp->lc_flags & LCL_STAMPEDSTABLE)) {
3319 			clp->lc_flags |= LCL_STAMPEDSTABLE;
3320 			len = clp->lc_idlen;
3321 			NFSBCOPY(clp->lc_id, client, len);
3322 			gotstate = 1;
3323 		}
3324 		NFSUNLOCKSTATE();
3325 	} else if (new_stp->ls_flags & NFSLCK_CLOSE) {
3326 		ownerstp = stp->ls_openowner;
3327 		lfp = stp->ls_lfp;
3328 		if (nfsrv_dolocallocks != 0 && !LIST_EMPTY(&stp->ls_open)) {
3329 			/* Get the lf lock */
3330 			nfsrv_locklf(lfp);
3331 			NFSUNLOCKSTATE();
3332 			ASSERT_VOP_ELOCKED(vp, "nfsrv_openupdate");
3333 			NFSVOPUNLOCK(vp, 0);
3334 			if (nfsrv_freeopen(stp, vp, 1, p) == 0) {
3335 				NFSLOCKSTATE();
3336 				nfsrv_unlocklf(lfp);
3337 				NFSUNLOCKSTATE();
3338 			}
3339 			NFSVOPLOCK(vp, LK_EXCLUSIVE | LK_RETRY);
3340 		} else {
3341 			(void) nfsrv_freeopen(stp, NULL, 0, p);
3342 			NFSUNLOCKSTATE();
3343 		}
3344 	} else {
3345 		/*
3346 		 * Update the share bits, making sure that the new set are a
3347 		 * subset of the old ones.
3348 		 */
3349 		bits = (new_stp->ls_flags & NFSLCK_SHAREBITS);
3350 		if (~(stp->ls_flags) & bits) {
3351 			NFSUNLOCKSTATE();
3352 			error = NFSERR_INVAL;
3353 			goto out;
3354 		}
3355 		stp->ls_flags = (bits | NFSLCK_OPEN);
3356 		stp->ls_stateid.seqid++;
3357 		if ((nd->nd_flag & ND_NFSV41) != 0 &&
3358 		    stp->ls_stateid.seqid == 0)
3359 			stp->ls_stateid.seqid = 1;
3360 		NFSUNLOCKSTATE();
3361 	}
3362 
3363 	/*
3364 	 * If the client just confirmed its first open, write a timestamp
3365 	 * to the stable storage file.
3366 	 */
3367 	if (gotstate != 0) {
3368 		nfsrv_writestable(client, len, NFSNST_NEWSTATE, p);
3369 		nfsrv_backupstable();
3370 	}
3371 
3372 out:
3373 	NFSEXITCODE2(error, nd);
3374 	return (error);
3375 }
3376 
3377 /*
3378  * Delegation update. Does the purge and return.
3379  */
3380 APPLESTATIC int
3381 nfsrv_delegupdate(struct nfsrv_descript *nd, nfsquad_t clientid,
3382     nfsv4stateid_t *stateidp, vnode_t vp, int op, struct ucred *cred,
3383     NFSPROC_T *p)
3384 {
3385 	struct nfsstate *stp;
3386 	struct nfsclient *clp;
3387 	int error = 0;
3388 	fhandle_t fh;
3389 
3390 	/*
3391 	 * Do a sanity check against the file handle for DelegReturn.
3392 	 */
3393 	if (vp) {
3394 		error = nfsvno_getfh(vp, &fh, p);
3395 		if (error)
3396 			goto out;
3397 	}
3398 	/*
3399 	 * Check for restart conditions (client and server).
3400 	 */
3401 	if (op == NFSV4OP_DELEGRETURN)
3402 		error = nfsrv_checkrestart(clientid, NFSLCK_DELEGRETURN,
3403 			stateidp, 0);
3404 	else
3405 		error = nfsrv_checkrestart(clientid, NFSLCK_DELEGPURGE,
3406 			stateidp, 0);
3407 
3408 	NFSLOCKSTATE();
3409 	/*
3410 	 * Get the open structure via clientid and stateid.
3411 	 */
3412 	if (!error)
3413 	    error = nfsrv_getclient(clientid, CLOPS_RENEW, &clp, NULL,
3414 		(nfsquad_t)((u_quad_t)0), 0, nd, p);
3415 	if (error) {
3416 		if (error == NFSERR_CBPATHDOWN)
3417 			error = 0;
3418 		if (error == NFSERR_STALECLIENTID && op == NFSV4OP_DELEGRETURN)
3419 			error = NFSERR_STALESTATEID;
3420 	}
3421 	if (!error && op == NFSV4OP_DELEGRETURN) {
3422 	    error = nfsrv_getstate(clp, stateidp, NFSLCK_DELEGRETURN, &stp);
3423 	    if (!error && stp->ls_stateid.seqid != stateidp->seqid &&
3424 		((nd->nd_flag & ND_NFSV41) == 0 || stateidp->seqid != 0))
3425 		error = NFSERR_OLDSTATEID;
3426 	}
3427 	/*
3428 	 * NFSERR_EXPIRED means that the state has gone away,
3429 	 * so Delegations have been purged. Just return ok.
3430 	 */
3431 	if (error == NFSERR_EXPIRED && op == NFSV4OP_DELEGPURGE) {
3432 		NFSUNLOCKSTATE();
3433 		error = 0;
3434 		goto out;
3435 	}
3436 	if (error) {
3437 		NFSUNLOCKSTATE();
3438 		goto out;
3439 	}
3440 
3441 	if (op == NFSV4OP_DELEGRETURN) {
3442 		if (NFSBCMP((caddr_t)&fh, (caddr_t)&stp->ls_lfp->lf_fh,
3443 		    sizeof (fhandle_t))) {
3444 			NFSUNLOCKSTATE();
3445 			error = NFSERR_BADSTATEID;
3446 			goto out;
3447 		}
3448 		nfsrv_freedeleg(stp);
3449 	} else {
3450 		nfsrv_freedeleglist(&clp->lc_olddeleg);
3451 	}
3452 	NFSUNLOCKSTATE();
3453 	error = 0;
3454 
3455 out:
3456 	NFSEXITCODE(error);
3457 	return (error);
3458 }
3459 
3460 /*
3461  * Release lock owner.
3462  */
3463 APPLESTATIC int
3464 nfsrv_releaselckown(struct nfsstate *new_stp, nfsquad_t clientid,
3465     NFSPROC_T *p)
3466 {
3467 	struct nfsstate *stp, *nstp, *openstp, *ownstp;
3468 	struct nfsclient *clp;
3469 	int error = 0;
3470 
3471 	/*
3472 	 * Check for restart conditions (client and server).
3473 	 */
3474 	error = nfsrv_checkrestart(clientid, new_stp->ls_flags,
3475 	    &new_stp->ls_stateid, 0);
3476 	if (error)
3477 		goto out;
3478 
3479 	NFSLOCKSTATE();
3480 	/*
3481 	 * Get the lock owner by name.
3482 	 */
3483 	error = nfsrv_getclient(clientid, CLOPS_RENEW, &clp, NULL,
3484 	    (nfsquad_t)((u_quad_t)0), 0, NULL, p);
3485 	if (error) {
3486 		NFSUNLOCKSTATE();
3487 		goto out;
3488 	}
3489 	LIST_FOREACH(ownstp, &clp->lc_open, ls_list) {
3490 	    LIST_FOREACH(openstp, &ownstp->ls_open, ls_list) {
3491 		stp = LIST_FIRST(&openstp->ls_open);
3492 		while (stp != LIST_END(&openstp->ls_open)) {
3493 		    nstp = LIST_NEXT(stp, ls_list);
3494 		    /*
3495 		     * If the owner matches, check for locks and
3496 		     * then free or return an error.
3497 		     */
3498 		    if (stp->ls_ownerlen == new_stp->ls_ownerlen &&
3499 			!NFSBCMP(stp->ls_owner, new_stp->ls_owner,
3500 			 stp->ls_ownerlen)){
3501 			if (LIST_EMPTY(&stp->ls_lock)) {
3502 			    nfsrv_freelockowner(stp, NULL, 0, p);
3503 			} else {
3504 			    NFSUNLOCKSTATE();
3505 			    error = NFSERR_LOCKSHELD;
3506 			    goto out;
3507 			}
3508 		    }
3509 		    stp = nstp;
3510 		}
3511 	    }
3512 	}
3513 	NFSUNLOCKSTATE();
3514 
3515 out:
3516 	NFSEXITCODE(error);
3517 	return (error);
3518 }
3519 
3520 /*
3521  * Get the file handle for a lock structure.
3522  */
3523 static int
3524 nfsrv_getlockfh(vnode_t vp, u_short flags, struct nfslockfile *new_lfp,
3525     fhandle_t *nfhp, NFSPROC_T *p)
3526 {
3527 	fhandle_t *fhp = NULL;
3528 	int error;
3529 
3530 	/*
3531 	 * For lock, use the new nfslock structure, otherwise just
3532 	 * a fhandle_t on the stack.
3533 	 */
3534 	if (flags & NFSLCK_OPEN) {
3535 		KASSERT(new_lfp != NULL, ("nfsrv_getlockfh: new_lfp NULL"));
3536 		fhp = &new_lfp->lf_fh;
3537 	} else if (nfhp) {
3538 		fhp = nfhp;
3539 	} else {
3540 		panic("nfsrv_getlockfh");
3541 	}
3542 	error = nfsvno_getfh(vp, fhp, p);
3543 	NFSEXITCODE(error);
3544 	return (error);
3545 }
3546 
3547 /*
3548  * Get an nfs lock structure. Allocate one, as required, and return a
3549  * pointer to it.
3550  * Returns an NFSERR_xxx upon failure or -1 to indicate no current lock.
3551  */
3552 static int
3553 nfsrv_getlockfile(u_short flags, struct nfslockfile **new_lfpp,
3554     struct nfslockfile **lfpp, fhandle_t *nfhp, int lockit)
3555 {
3556 	struct nfslockfile *lfp;
3557 	fhandle_t *fhp = NULL, *tfhp;
3558 	struct nfslockhashhead *hp;
3559 	struct nfslockfile *new_lfp = NULL;
3560 
3561 	/*
3562 	 * For lock, use the new nfslock structure, otherwise just
3563 	 * a fhandle_t on the stack.
3564 	 */
3565 	if (flags & NFSLCK_OPEN) {
3566 		new_lfp = *new_lfpp;
3567 		fhp = &new_lfp->lf_fh;
3568 	} else if (nfhp) {
3569 		fhp = nfhp;
3570 	} else {
3571 		panic("nfsrv_getlockfile");
3572 	}
3573 
3574 	hp = NFSLOCKHASH(fhp);
3575 	LIST_FOREACH(lfp, hp, lf_hash) {
3576 		tfhp = &lfp->lf_fh;
3577 		if (NFSVNO_CMPFH(fhp, tfhp)) {
3578 			if (lockit)
3579 				nfsrv_locklf(lfp);
3580 			*lfpp = lfp;
3581 			return (0);
3582 		}
3583 	}
3584 	if (!(flags & NFSLCK_OPEN))
3585 		return (-1);
3586 
3587 	/*
3588 	 * No match, so chain the new one into the list.
3589 	 */
3590 	LIST_INIT(&new_lfp->lf_open);
3591 	LIST_INIT(&new_lfp->lf_lock);
3592 	LIST_INIT(&new_lfp->lf_deleg);
3593 	LIST_INIT(&new_lfp->lf_locallock);
3594 	LIST_INIT(&new_lfp->lf_rollback);
3595 	new_lfp->lf_locallock_lck.nfslock_usecnt = 0;
3596 	new_lfp->lf_locallock_lck.nfslock_lock = 0;
3597 	new_lfp->lf_usecount = 0;
3598 	LIST_INSERT_HEAD(hp, new_lfp, lf_hash);
3599 	*lfpp = new_lfp;
3600 	*new_lfpp = NULL;
3601 	return (0);
3602 }
3603 
3604 /*
3605  * This function adds a nfslock lock structure to the list for the associated
3606  * nfsstate and nfslockfile structures. It will be inserted after the
3607  * entry pointed at by insert_lop.
3608  */
3609 static void
3610 nfsrv_insertlock(struct nfslock *new_lop, struct nfslock *insert_lop,
3611     struct nfsstate *stp, struct nfslockfile *lfp)
3612 {
3613 	struct nfslock *lop, *nlop;
3614 
3615 	new_lop->lo_stp = stp;
3616 	new_lop->lo_lfp = lfp;
3617 
3618 	if (stp != NULL) {
3619 		/* Insert in increasing lo_first order */
3620 		lop = LIST_FIRST(&lfp->lf_lock);
3621 		if (lop == LIST_END(&lfp->lf_lock) ||
3622 		    new_lop->lo_first <= lop->lo_first) {
3623 			LIST_INSERT_HEAD(&lfp->lf_lock, new_lop, lo_lckfile);
3624 		} else {
3625 			nlop = LIST_NEXT(lop, lo_lckfile);
3626 			while (nlop != LIST_END(&lfp->lf_lock) &&
3627 			       nlop->lo_first < new_lop->lo_first) {
3628 				lop = nlop;
3629 				nlop = LIST_NEXT(lop, lo_lckfile);
3630 			}
3631 			LIST_INSERT_AFTER(lop, new_lop, lo_lckfile);
3632 		}
3633 	} else {
3634 		new_lop->lo_lckfile.le_prev = NULL;	/* list not used */
3635 	}
3636 
3637 	/*
3638 	 * Insert after insert_lop, which is overloaded as stp or lfp for
3639 	 * an empty list.
3640 	 */
3641 	if (stp == NULL && (struct nfslockfile *)insert_lop == lfp)
3642 		LIST_INSERT_HEAD(&lfp->lf_locallock, new_lop, lo_lckowner);
3643 	else if ((struct nfsstate *)insert_lop == stp)
3644 		LIST_INSERT_HEAD(&stp->ls_lock, new_lop, lo_lckowner);
3645 	else
3646 		LIST_INSERT_AFTER(insert_lop, new_lop, lo_lckowner);
3647 	if (stp != NULL) {
3648 		newnfsstats.srvlocks++;
3649 		nfsrv_openpluslock++;
3650 	}
3651 }
3652 
3653 /*
3654  * This function updates the locking for a lock owner and given file. It
3655  * maintains a list of lock ranges ordered on increasing file offset that
3656  * are NFSLCK_READ or NFSLCK_WRITE and non-overlapping (aka POSIX style).
3657  * It always adds new_lop to the list and sometimes uses the one pointed
3658  * at by other_lopp.
3659  */
3660 static void
3661 nfsrv_updatelock(struct nfsstate *stp, struct nfslock **new_lopp,
3662     struct nfslock **other_lopp, struct nfslockfile *lfp)
3663 {
3664 	struct nfslock *new_lop = *new_lopp;
3665 	struct nfslock *lop, *tlop, *ilop;
3666 	struct nfslock *other_lop = *other_lopp;
3667 	int unlock = 0, myfile = 0;
3668 	u_int64_t tmp;
3669 
3670 	/*
3671 	 * Work down the list until the lock is merged.
3672 	 */
3673 	if (new_lop->lo_flags & NFSLCK_UNLOCK)
3674 		unlock = 1;
3675 	if (stp != NULL) {
3676 		ilop = (struct nfslock *)stp;
3677 		lop = LIST_FIRST(&stp->ls_lock);
3678 	} else {
3679 		ilop = (struct nfslock *)lfp;
3680 		lop = LIST_FIRST(&lfp->lf_locallock);
3681 	}
3682 	while (lop != NULL) {
3683 	    /*
3684 	     * Only check locks for this file that aren't before the start of
3685 	     * new lock's range.
3686 	     */
3687 	    if (lop->lo_lfp == lfp) {
3688 	      myfile = 1;
3689 	      if (lop->lo_end >= new_lop->lo_first) {
3690 		if (new_lop->lo_end < lop->lo_first) {
3691 			/*
3692 			 * If the new lock ends before the start of the
3693 			 * current lock's range, no merge, just insert
3694 			 * the new lock.
3695 			 */
3696 			break;
3697 		}
3698 		if (new_lop->lo_flags == lop->lo_flags ||
3699 		    (new_lop->lo_first <= lop->lo_first &&
3700 		     new_lop->lo_end >= lop->lo_end)) {
3701 			/*
3702 			 * This lock can be absorbed by the new lock/unlock.
3703 			 * This happens when it covers the entire range
3704 			 * of the old lock or is contiguous
3705 			 * with the old lock and is of the same type or an
3706 			 * unlock.
3707 			 */
3708 			if (lop->lo_first < new_lop->lo_first)
3709 				new_lop->lo_first = lop->lo_first;
3710 			if (lop->lo_end > new_lop->lo_end)
3711 				new_lop->lo_end = lop->lo_end;
3712 			tlop = lop;
3713 			lop = LIST_NEXT(lop, lo_lckowner);
3714 			nfsrv_freenfslock(tlop);
3715 			continue;
3716 		}
3717 
3718 		/*
3719 		 * All these cases are for contiguous locks that are not the
3720 		 * same type, so they can't be merged.
3721 		 */
3722 		if (new_lop->lo_first <= lop->lo_first) {
3723 			/*
3724 			 * This case is where the new lock overlaps with the
3725 			 * first part of the old lock. Move the start of the
3726 			 * old lock to just past the end of the new lock. The
3727 			 * new lock will be inserted in front of the old, since
3728 			 * ilop hasn't been updated. (We are done now.)
3729 			 */
3730 			lop->lo_first = new_lop->lo_end;
3731 			break;
3732 		}
3733 		if (new_lop->lo_end >= lop->lo_end) {
3734 			/*
3735 			 * This case is where the new lock overlaps with the
3736 			 * end of the old lock's range. Move the old lock's
3737 			 * end to just before the new lock's first and insert
3738 			 * the new lock after the old lock.
3739 			 * Might not be done yet, since the new lock could
3740 			 * overlap further locks with higher ranges.
3741 			 */
3742 			lop->lo_end = new_lop->lo_first;
3743 			ilop = lop;
3744 			lop = LIST_NEXT(lop, lo_lckowner);
3745 			continue;
3746 		}
3747 		/*
3748 		 * The final case is where the new lock's range is in the
3749 		 * middle of the current lock's and splits the current lock
3750 		 * up. Use *other_lopp to handle the second part of the
3751 		 * split old lock range. (We are done now.)
3752 		 * For unlock, we use new_lop as other_lop and tmp, since
3753 		 * other_lop and new_lop are the same for this case.
3754 		 * We noted the unlock case above, so we don't need
3755 		 * new_lop->lo_flags any longer.
3756 		 */
3757 		tmp = new_lop->lo_first;
3758 		if (other_lop == NULL) {
3759 			if (!unlock)
3760 				panic("nfsd srv update unlock");
3761 			other_lop = new_lop;
3762 			*new_lopp = NULL;
3763 		}
3764 		other_lop->lo_first = new_lop->lo_end;
3765 		other_lop->lo_end = lop->lo_end;
3766 		other_lop->lo_flags = lop->lo_flags;
3767 		other_lop->lo_stp = stp;
3768 		other_lop->lo_lfp = lfp;
3769 		lop->lo_end = tmp;
3770 		nfsrv_insertlock(other_lop, lop, stp, lfp);
3771 		*other_lopp = NULL;
3772 		ilop = lop;
3773 		break;
3774 	      }
3775 	    }
3776 	    ilop = lop;
3777 	    lop = LIST_NEXT(lop, lo_lckowner);
3778 	    if (myfile && (lop == NULL || lop->lo_lfp != lfp))
3779 		break;
3780 	}
3781 
3782 	/*
3783 	 * Insert the new lock in the list at the appropriate place.
3784 	 */
3785 	if (!unlock) {
3786 		nfsrv_insertlock(new_lop, ilop, stp, lfp);
3787 		*new_lopp = NULL;
3788 	}
3789 }
3790 
3791 /*
3792  * This function handles sequencing of locks, etc.
3793  * It returns an error that indicates what the caller should do.
3794  */
3795 static int
3796 nfsrv_checkseqid(struct nfsrv_descript *nd, u_int32_t seqid,
3797     struct nfsstate *stp, struct nfsrvcache *op)
3798 {
3799 	int error = 0;
3800 
3801 	if ((nd->nd_flag & ND_NFSV41) != 0)
3802 		/* NFSv4.1 ignores the open_seqid and lock_seqid. */
3803 		goto out;
3804 	if (op != nd->nd_rp)
3805 		panic("nfsrvstate checkseqid");
3806 	if (!(op->rc_flag & RC_INPROG))
3807 		panic("nfsrvstate not inprog");
3808 	if (stp->ls_op && stp->ls_op->rc_refcnt <= 0) {
3809 		printf("refcnt=%d\n", stp->ls_op->rc_refcnt);
3810 		panic("nfsrvstate op refcnt");
3811 	}
3812 	if ((stp->ls_seq + 1) == seqid) {
3813 		if (stp->ls_op)
3814 			nfsrvd_derefcache(stp->ls_op);
3815 		stp->ls_op = op;
3816 		nfsrvd_refcache(op);
3817 		stp->ls_seq = seqid;
3818 		goto out;
3819 	} else if (stp->ls_seq == seqid && stp->ls_op &&
3820 		op->rc_xid == stp->ls_op->rc_xid &&
3821 		op->rc_refcnt == 0 &&
3822 		op->rc_reqlen == stp->ls_op->rc_reqlen &&
3823 		op->rc_cksum == stp->ls_op->rc_cksum) {
3824 		if (stp->ls_op->rc_flag & RC_INPROG) {
3825 			error = NFSERR_DONTREPLY;
3826 			goto out;
3827 		}
3828 		nd->nd_rp = stp->ls_op;
3829 		nd->nd_rp->rc_flag |= RC_INPROG;
3830 		nfsrvd_delcache(op);
3831 		error = NFSERR_REPLYFROMCACHE;
3832 		goto out;
3833 	}
3834 	error = NFSERR_BADSEQID;
3835 
3836 out:
3837 	NFSEXITCODE2(error, nd);
3838 	return (error);
3839 }
3840 
3841 /*
3842  * Get the client ip address for callbacks. If the strings can't be parsed,
3843  * just set lc_program to 0 to indicate no callbacks are possible.
3844  * (For cases where the address can't be parsed or is 0.0.0.0.0.0, set
3845  *  the address to the client's transport address. This won't be used
3846  *  for callbacks, but can be printed out by newnfsstats for info.)
3847  * Return error if the xdr can't be parsed, 0 otherwise.
3848  */
3849 APPLESTATIC int
3850 nfsrv_getclientipaddr(struct nfsrv_descript *nd, struct nfsclient *clp)
3851 {
3852 	u_int32_t *tl;
3853 	u_char *cp, *cp2;
3854 	int i, j;
3855 	struct sockaddr_in *rad, *sad;
3856 	u_char protocol[5], addr[24];
3857 	int error = 0, cantparse = 0;
3858 	union {
3859 		u_long ival;
3860 		u_char cval[4];
3861 	} ip;
3862 	union {
3863 		u_short sval;
3864 		u_char cval[2];
3865 	} port;
3866 
3867 	rad = NFSSOCKADDR(clp->lc_req.nr_nam, struct sockaddr_in *);
3868 	rad->sin_family = AF_INET;
3869 	rad->sin_len = sizeof (struct sockaddr_in);
3870 	rad->sin_addr.s_addr = 0;
3871 	rad->sin_port = 0;
3872 	clp->lc_req.nr_client = NULL;
3873 	clp->lc_req.nr_lock = 0;
3874 	NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
3875 	i = fxdr_unsigned(int, *tl);
3876 	if (i >= 3 && i <= 4) {
3877 		error = nfsrv_mtostr(nd, protocol, i);
3878 		if (error)
3879 			goto nfsmout;
3880 		if (!strcmp(protocol, "tcp")) {
3881 			clp->lc_flags |= LCL_TCPCALLBACK;
3882 			clp->lc_req.nr_sotype = SOCK_STREAM;
3883 			clp->lc_req.nr_soproto = IPPROTO_TCP;
3884 		} else if (!strcmp(protocol, "udp")) {
3885 			clp->lc_req.nr_sotype = SOCK_DGRAM;
3886 			clp->lc_req.nr_soproto = IPPROTO_UDP;
3887 		} else {
3888 			cantparse = 1;
3889 		}
3890 	} else {
3891 		cantparse = 1;
3892 		if (i > 0) {
3893 			error = nfsm_advance(nd, NFSM_RNDUP(i), -1);
3894 			if (error)
3895 				goto nfsmout;
3896 		}
3897 	}
3898 	NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
3899 	i = fxdr_unsigned(int, *tl);
3900 	if (i < 0) {
3901 		error = NFSERR_BADXDR;
3902 		goto nfsmout;
3903 	} else if (i == 0) {
3904 		cantparse = 1;
3905 	} else if (!cantparse && i <= 23 && i >= 11) {
3906 		error = nfsrv_mtostr(nd, addr, i);
3907 		if (error)
3908 			goto nfsmout;
3909 
3910 		/*
3911 		 * Parse out the address fields. We expect 6 decimal numbers
3912 		 * separated by '.'s.
3913 		 */
3914 		cp = addr;
3915 		i = 0;
3916 		while (*cp && i < 6) {
3917 			cp2 = cp;
3918 			while (*cp2 && *cp2 != '.')
3919 				cp2++;
3920 			if (*cp2)
3921 				*cp2++ = '\0';
3922 			else if (i != 5) {
3923 				cantparse = 1;
3924 				break;
3925 			}
3926 			j = nfsrv_getipnumber(cp);
3927 			if (j >= 0) {
3928 				if (i < 4)
3929 					ip.cval[3 - i] = j;
3930 				else
3931 					port.cval[5 - i] = j;
3932 			} else {
3933 				cantparse = 1;
3934 				break;
3935 			}
3936 			cp = cp2;
3937 			i++;
3938 		}
3939 		if (!cantparse) {
3940 			if (ip.ival != 0x0) {
3941 				rad->sin_addr.s_addr = htonl(ip.ival);
3942 				rad->sin_port = htons(port.sval);
3943 			} else {
3944 				cantparse = 1;
3945 			}
3946 		}
3947 	} else {
3948 		cantparse = 1;
3949 		if (i > 0) {
3950 			error = nfsm_advance(nd, NFSM_RNDUP(i), -1);
3951 			if (error)
3952 				goto nfsmout;
3953 		}
3954 	}
3955 	if (cantparse) {
3956 		sad = NFSSOCKADDR(nd->nd_nam, struct sockaddr_in *);
3957 		rad->sin_addr.s_addr = sad->sin_addr.s_addr;
3958 		rad->sin_port = 0x0;
3959 		clp->lc_program = 0;
3960 	}
3961 nfsmout:
3962 	NFSEXITCODE2(error, nd);
3963 	return (error);
3964 }
3965 
3966 /*
3967  * Turn a string of up to three decimal digits into a number. Return -1 upon
3968  * error.
3969  */
3970 static int
3971 nfsrv_getipnumber(u_char *cp)
3972 {
3973 	int i = 0, j = 0;
3974 
3975 	while (*cp) {
3976 		if (j > 2 || *cp < '0' || *cp > '9')
3977 			return (-1);
3978 		i *= 10;
3979 		i += (*cp - '0');
3980 		cp++;
3981 		j++;
3982 	}
3983 	if (i < 256)
3984 		return (i);
3985 	return (-1);
3986 }
3987 
3988 /*
3989  * This function checks for restart conditions.
3990  */
3991 static int
3992 nfsrv_checkrestart(nfsquad_t clientid, u_int32_t flags,
3993     nfsv4stateid_t *stateidp, int specialid)
3994 {
3995 	int ret = 0;
3996 
3997 	/*
3998 	 * First check for a server restart. Open, LockT, ReleaseLockOwner
3999 	 * and DelegPurge have a clientid, the rest a stateid.
4000 	 */
4001 	if (flags &
4002 	    (NFSLCK_OPEN | NFSLCK_TEST | NFSLCK_RELEASE | NFSLCK_DELEGPURGE)) {
4003 		if (clientid.lval[0] != nfsrvboottime) {
4004 			ret = NFSERR_STALECLIENTID;
4005 			goto out;
4006 		}
4007 	} else if (stateidp->other[0] != nfsrvboottime &&
4008 		specialid == 0) {
4009 		ret = NFSERR_STALESTATEID;
4010 		goto out;
4011 	}
4012 
4013 	/*
4014 	 * Read, Write, Setattr and LockT can return NFSERR_GRACE and do
4015 	 * not use a lock/open owner seqid#, so the check can be done now.
4016 	 * (The others will be checked, as required, later.)
4017 	 */
4018 	if (!(flags & (NFSLCK_CHECK | NFSLCK_TEST)))
4019 		goto out;
4020 
4021 	NFSLOCKSTATE();
4022 	ret = nfsrv_checkgrace(NULL, NULL, flags);
4023 	NFSUNLOCKSTATE();
4024 
4025 out:
4026 	NFSEXITCODE(ret);
4027 	return (ret);
4028 }
4029 
4030 /*
4031  * Check for grace.
4032  */
4033 static int
4034 nfsrv_checkgrace(struct nfsrv_descript *nd, struct nfsclient *clp,
4035     u_int32_t flags)
4036 {
4037 	int error = 0;
4038 
4039 	if ((nfsrv_stablefirst.nsf_flags & NFSNSF_GRACEOVER) != 0) {
4040 		if (flags & NFSLCK_RECLAIM) {
4041 			error = NFSERR_NOGRACE;
4042 			goto out;
4043 		}
4044 	} else {
4045 		if (!(flags & NFSLCK_RECLAIM)) {
4046 			error = NFSERR_GRACE;
4047 			goto out;
4048 		}
4049 		if (nd != NULL && clp != NULL &&
4050 		    (nd->nd_flag & ND_NFSV41) != 0 &&
4051 		    (clp->lc_flags & LCL_RECLAIMCOMPLETE) != 0) {
4052 			error = NFSERR_NOGRACE;
4053 			goto out;
4054 		}
4055 
4056 		/*
4057 		 * If grace is almost over and we are still getting Reclaims,
4058 		 * extend grace a bit.
4059 		 */
4060 		if ((NFSD_MONOSEC + NFSRV_LEASEDELTA) >
4061 		    nfsrv_stablefirst.nsf_eograce)
4062 			nfsrv_stablefirst.nsf_eograce = NFSD_MONOSEC +
4063 				NFSRV_LEASEDELTA;
4064 	}
4065 
4066 out:
4067 	NFSEXITCODE(error);
4068 	return (error);
4069 }
4070 
4071 /*
4072  * Do a server callback.
4073  */
4074 static int
4075 nfsrv_docallback(struct nfsclient *clp, int procnum,
4076     nfsv4stateid_t *stateidp, int trunc, fhandle_t *fhp,
4077     struct nfsvattr *nap, nfsattrbit_t *attrbitp, NFSPROC_T *p)
4078 {
4079 	mbuf_t m;
4080 	u_int32_t *tl;
4081 	struct nfsrv_descript nfsd, *nd = &nfsd;
4082 	struct ucred *cred;
4083 	int error = 0;
4084 	u_int32_t callback;
4085 	struct nfsdsession *sep = NULL;
4086 
4087 	cred = newnfs_getcred();
4088 	NFSLOCKSTATE();	/* mostly for lc_cbref++ */
4089 	if (clp->lc_flags & LCL_NEEDSCONFIRM) {
4090 		NFSUNLOCKSTATE();
4091 		panic("docallb");
4092 	}
4093 	clp->lc_cbref++;
4094 
4095 	/*
4096 	 * Fill the callback program# and version into the request
4097 	 * structure for newnfs_connect() to use.
4098 	 */
4099 	clp->lc_req.nr_prog = clp->lc_program;
4100 #ifdef notnow
4101 	if ((clp->lc_flags & LCL_NFSV41) != 0)
4102 		clp->lc_req.nr_vers = NFSV41_CBVERS;
4103 	else
4104 #endif
4105 		clp->lc_req.nr_vers = NFSV4_CBVERS;
4106 
4107 	/*
4108 	 * First, fill in some of the fields of nd and cr.
4109 	 */
4110 	nd->nd_flag = ND_NFSV4;
4111 	if (clp->lc_flags & LCL_GSS)
4112 		nd->nd_flag |= ND_KERBV;
4113 	if ((clp->lc_flags & LCL_NFSV41) != 0)
4114 		nd->nd_flag |= ND_NFSV41;
4115 	nd->nd_repstat = 0;
4116 	cred->cr_uid = clp->lc_uid;
4117 	cred->cr_gid = clp->lc_gid;
4118 	callback = clp->lc_callback;
4119 	NFSUNLOCKSTATE();
4120 	cred->cr_ngroups = 1;
4121 
4122 	/*
4123 	 * Get the first mbuf for the request.
4124 	 */
4125 	MGET(m, M_WAITOK, MT_DATA);
4126 	mbuf_setlen(m, 0);
4127 	nd->nd_mreq = nd->nd_mb = m;
4128 	nd->nd_bpos = NFSMTOD(m, caddr_t);
4129 
4130 	/*
4131 	 * and build the callback request.
4132 	 */
4133 	if (procnum == NFSV4OP_CBGETATTR) {
4134 		nd->nd_procnum = NFSV4PROC_CBCOMPOUND;
4135 		error = nfsrv_cbcallargs(nd, clp, callback, NFSV4OP_CBGETATTR,
4136 		    "CB Getattr", &sep);
4137 		if (error != 0) {
4138 			mbuf_freem(nd->nd_mreq);
4139 			goto errout;
4140 		}
4141 		(void)nfsm_fhtom(nd, (u_int8_t *)fhp, NFSX_MYFH, 0);
4142 		(void)nfsrv_putattrbit(nd, attrbitp);
4143 	} else if (procnum == NFSV4OP_CBRECALL) {
4144 		nd->nd_procnum = NFSV4PROC_CBCOMPOUND;
4145 		error = nfsrv_cbcallargs(nd, clp, callback, NFSV4OP_CBRECALL,
4146 		    "CB Recall", &sep);
4147 		if (error != 0) {
4148 			mbuf_freem(nd->nd_mreq);
4149 			goto errout;
4150 		}
4151 		NFSM_BUILD(tl, u_int32_t *, NFSX_UNSIGNED + NFSX_STATEID);
4152 		*tl++ = txdr_unsigned(stateidp->seqid);
4153 		NFSBCOPY((caddr_t)stateidp->other, (caddr_t)tl,
4154 		    NFSX_STATEIDOTHER);
4155 		tl += (NFSX_STATEIDOTHER / NFSX_UNSIGNED);
4156 		if (trunc)
4157 			*tl = newnfs_true;
4158 		else
4159 			*tl = newnfs_false;
4160 		(void)nfsm_fhtom(nd, (u_int8_t *)fhp, NFSX_MYFH, 0);
4161 	} else if (procnum == NFSV4PROC_CBNULL) {
4162 		nd->nd_procnum = NFSV4PROC_CBNULL;
4163 		if ((clp->lc_flags & LCL_NFSV41) != 0) {
4164 			error = nfsv4_getcbsession(clp, &sep);
4165 			if (error != 0) {
4166 				mbuf_freem(nd->nd_mreq);
4167 				goto errout;
4168 			}
4169 		}
4170 	} else {
4171 		error = NFSERR_SERVERFAULT;
4172 		mbuf_freem(nd->nd_mreq);
4173 		goto errout;
4174 	}
4175 
4176 	/*
4177 	 * Call newnfs_connect(), as required, and then newnfs_request().
4178 	 */
4179 	(void) newnfs_sndlock(&clp->lc_req.nr_lock);
4180 	if (clp->lc_req.nr_client == NULL) {
4181 		if ((clp->lc_flags & LCL_NFSV41) != 0)
4182 			error = ECONNREFUSED;
4183 		else if (nd->nd_procnum == NFSV4PROC_CBNULL)
4184 			error = newnfs_connect(NULL, &clp->lc_req, cred,
4185 			    NULL, 1);
4186 		else
4187 			error = newnfs_connect(NULL, &clp->lc_req, cred,
4188 			    NULL, 3);
4189 	}
4190 	newnfs_sndunlock(&clp->lc_req.nr_lock);
4191 	if (!error) {
4192 		if ((nd->nd_flag & ND_NFSV41) != 0) {
4193 			KASSERT(sep != NULL, ("sep NULL"));
4194 			if (sep->sess_cbsess.nfsess_xprt != NULL)
4195 				error = newnfs_request(nd, NULL, clp,
4196 				    &clp->lc_req, NULL, NULL, cred,
4197 				    clp->lc_program, clp->lc_req.nr_vers, NULL,
4198 				    1, NULL, &sep->sess_cbsess);
4199 			else {
4200 				/*
4201 				 * This should probably never occur, but if a
4202 				 * client somehow does an RPC without a
4203 				 * SequenceID Op that causes a callback just
4204 				 * after the nfsd threads have been terminated
4205 				 * and restared we could conceivably get here
4206 				 * without a backchannel xprt.
4207 				 */
4208 				printf("nfsrv_docallback: no xprt\n");
4209 				error = ECONNREFUSED;
4210 			}
4211 			nfsrv_freesession(sep, NULL);
4212 		} else
4213 			error = newnfs_request(nd, NULL, clp, &clp->lc_req,
4214 			    NULL, NULL, cred, clp->lc_program,
4215 			    clp->lc_req.nr_vers, NULL, 1, NULL, NULL);
4216 	}
4217 errout:
4218 	NFSFREECRED(cred);
4219 
4220 	/*
4221 	 * If error is set here, the Callback path isn't working
4222 	 * properly, so twiddle the appropriate LCL_ flags.
4223 	 * (nd_repstat != 0 indicates the Callback path is working,
4224 	 *  but the callback failed on the client.)
4225 	 */
4226 	if (error) {
4227 		/*
4228 		 * Mark the callback pathway down, which disabled issuing
4229 		 * of delegations and gets Renew to return NFSERR_CBPATHDOWN.
4230 		 */
4231 		NFSLOCKSTATE();
4232 		clp->lc_flags |= LCL_CBDOWN;
4233 		NFSUNLOCKSTATE();
4234 	} else {
4235 		/*
4236 		 * Callback worked. If the callback path was down, disable
4237 		 * callbacks, so no more delegations will be issued. (This
4238 		 * is done on the assumption that the callback pathway is
4239 		 * flakey.)
4240 		 */
4241 		NFSLOCKSTATE();
4242 		if (clp->lc_flags & LCL_CBDOWN)
4243 			clp->lc_flags &= ~(LCL_CBDOWN | LCL_CALLBACKSON);
4244 		NFSUNLOCKSTATE();
4245 		if (nd->nd_repstat)
4246 			error = nd->nd_repstat;
4247 		else if (error == 0 && procnum == NFSV4OP_CBGETATTR)
4248 			error = nfsv4_loadattr(nd, NULL, nap, NULL, NULL, 0,
4249 			    NULL, NULL, NULL, NULL, NULL, 0, NULL, NULL, NULL,
4250 			    p, NULL);
4251 		mbuf_freem(nd->nd_mrep);
4252 	}
4253 	NFSLOCKSTATE();
4254 	clp->lc_cbref--;
4255 	if ((clp->lc_flags & LCL_WAKEUPWANTED) && clp->lc_cbref == 0) {
4256 		clp->lc_flags &= ~LCL_WAKEUPWANTED;
4257 		wakeup(clp);
4258 	}
4259 	NFSUNLOCKSTATE();
4260 
4261 	NFSEXITCODE(error);
4262 	return (error);
4263 }
4264 
4265 /*
4266  * Set up the compound RPC for the callback.
4267  */
4268 static int
4269 nfsrv_cbcallargs(struct nfsrv_descript *nd, struct nfsclient *clp,
4270     uint32_t callback, int op, const char *optag, struct nfsdsession **sepp)
4271 {
4272 	uint32_t *tl;
4273 	int error, len;
4274 
4275 	len = strlen(optag);
4276 	(void)nfsm_strtom(nd, optag, len);
4277 	NFSM_BUILD(tl, uint32_t *, 4 * NFSX_UNSIGNED);
4278 	if ((nd->nd_flag & ND_NFSV41) != 0) {
4279 		*tl++ = txdr_unsigned(NFSV41_MINORVERSION);
4280 		*tl++ = txdr_unsigned(callback);
4281 		*tl++ = txdr_unsigned(2);
4282 		*tl = txdr_unsigned(NFSV4OP_CBSEQUENCE);
4283 		error = nfsv4_setcbsequence(nd, clp, 1, sepp);
4284 		if (error != 0)
4285 			return (error);
4286 		NFSM_BUILD(tl, u_int32_t *, NFSX_UNSIGNED);
4287 		*tl = txdr_unsigned(op);
4288 	} else {
4289 		*tl++ = txdr_unsigned(NFSV4_MINORVERSION);
4290 		*tl++ = txdr_unsigned(callback);
4291 		*tl++ = txdr_unsigned(1);
4292 		*tl = txdr_unsigned(op);
4293 	}
4294 	return (0);
4295 }
4296 
4297 /*
4298  * Return the next index# for a clientid. Mostly just increment and return
4299  * the next one, but... if the 32bit unsigned does actually wrap around,
4300  * it should be rebooted.
4301  * At an average rate of one new client per second, it will wrap around in
4302  * approximately 136 years. (I think the server will have been shut
4303  * down or rebooted before then.)
4304  */
4305 static u_int32_t
4306 nfsrv_nextclientindex(void)
4307 {
4308 	static u_int32_t client_index = 0;
4309 
4310 	client_index++;
4311 	if (client_index != 0)
4312 		return (client_index);
4313 
4314 	printf("%s: out of clientids\n", __func__);
4315 	return (client_index);
4316 }
4317 
4318 /*
4319  * Return the next index# for a stateid. Mostly just increment and return
4320  * the next one, but... if the 32bit unsigned does actually wrap around
4321  * (will a BSD server stay up that long?), find
4322  * new start and end values.
4323  */
4324 static u_int32_t
4325 nfsrv_nextstateindex(struct nfsclient *clp)
4326 {
4327 	struct nfsstate *stp;
4328 	int i;
4329 	u_int32_t canuse, min_index, max_index;
4330 
4331 	if (!(clp->lc_flags & LCL_INDEXNOTOK)) {
4332 		clp->lc_stateindex++;
4333 		if (clp->lc_stateindex != clp->lc_statemaxindex)
4334 			return (clp->lc_stateindex);
4335 	}
4336 
4337 	/*
4338 	 * Yuck, we've hit the end.
4339 	 * Look for a new min and max.
4340 	 */
4341 	min_index = 0;
4342 	max_index = 0xffffffff;
4343 	for (i = 0; i < nfsrv_statehashsize; i++) {
4344 	    LIST_FOREACH(stp, &clp->lc_stateid[i], ls_hash) {
4345 		if (stp->ls_stateid.other[2] > 0x80000000) {
4346 		    if (stp->ls_stateid.other[2] < max_index)
4347 			max_index = stp->ls_stateid.other[2];
4348 		} else {
4349 		    if (stp->ls_stateid.other[2] > min_index)
4350 			min_index = stp->ls_stateid.other[2];
4351 		}
4352 	    }
4353 	}
4354 
4355 	/*
4356 	 * Yikes, highly unlikely, but I'll handle it anyhow.
4357 	 */
4358 	if (min_index == 0x80000000 && max_index == 0x80000001) {
4359 	    canuse = 0;
4360 	    /*
4361 	     * Loop around until we find an unused entry. Return that
4362 	     * and set LCL_INDEXNOTOK, so the search will continue next time.
4363 	     * (This is one of those rare cases where a goto is the
4364 	     *  cleanest way to code the loop.)
4365 	     */
4366 tryagain:
4367 	    for (i = 0; i < nfsrv_statehashsize; i++) {
4368 		LIST_FOREACH(stp, &clp->lc_stateid[i], ls_hash) {
4369 		    if (stp->ls_stateid.other[2] == canuse) {
4370 			canuse++;
4371 			goto tryagain;
4372 		    }
4373 		}
4374 	    }
4375 	    clp->lc_flags |= LCL_INDEXNOTOK;
4376 	    return (canuse);
4377 	}
4378 
4379 	/*
4380 	 * Ok to start again from min + 1.
4381 	 */
4382 	clp->lc_stateindex = min_index + 1;
4383 	clp->lc_statemaxindex = max_index;
4384 	clp->lc_flags &= ~LCL_INDEXNOTOK;
4385 	return (clp->lc_stateindex);
4386 }
4387 
4388 /*
4389  * The following functions handle the stable storage file that deals with
4390  * the edge conditions described in RFC3530 Sec. 8.6.3.
4391  * The file is as follows:
4392  * - a single record at the beginning that has the lease time of the
4393  *   previous server instance (before the last reboot) and the nfsrvboottime
4394  *   values for the previous server boots.
4395  *   These previous boot times are used to ensure that the current
4396  *   nfsrvboottime does not, somehow, get set to a previous one.
4397  *   (This is important so that Stale ClientIDs and StateIDs can
4398  *    be recognized.)
4399  *   The number of previous nfsvrboottime values preceeds the list.
4400  * - followed by some number of appended records with:
4401  *   - client id string
4402  *   - flag that indicates it is a record revoking state via lease
4403  *     expiration or similar
4404  *     OR has successfully acquired state.
4405  * These structures vary in length, with the client string at the end, up
4406  * to NFSV4_OPAQUELIMIT in size.
4407  *
4408  * At the end of the grace period, the file is truncated, the first
4409  * record is rewritten with updated information and any acquired state
4410  * records for successful reclaims of state are written.
4411  *
4412  * Subsequent records are appended when the first state is issued to
4413  * a client and when state is revoked for a client.
4414  *
4415  * When reading the file in, state issued records that come later in
4416  * the file override older ones, since the append log is in cronological order.
4417  * If, for some reason, the file can't be read, the grace period is
4418  * immediately terminated and all reclaims get NFSERR_NOGRACE.
4419  */
4420 
4421 /*
4422  * Read in the stable storage file. Called by nfssvc() before the nfsd
4423  * processes start servicing requests.
4424  */
4425 APPLESTATIC void
4426 nfsrv_setupstable(NFSPROC_T *p)
4427 {
4428 	struct nfsrv_stablefirst *sf = &nfsrv_stablefirst;
4429 	struct nfsrv_stable *sp, *nsp;
4430 	struct nfst_rec *tsp;
4431 	int error, i, tryagain;
4432 	off_t off = 0;
4433 	ssize_t aresid, len;
4434 
4435 	/*
4436 	 * If NFSNSF_UPDATEDONE is set, this is a restart of the nfsds without
4437 	 * a reboot, so state has not been lost.
4438 	 */
4439 	if (sf->nsf_flags & NFSNSF_UPDATEDONE)
4440 		return;
4441 	/*
4442 	 * Set Grace over just until the file reads successfully.
4443 	 */
4444 	nfsrvboottime = time_second;
4445 	LIST_INIT(&sf->nsf_head);
4446 	sf->nsf_flags = (NFSNSF_GRACEOVER | NFSNSF_NEEDLOCK);
4447 	sf->nsf_eograce = NFSD_MONOSEC + NFSRV_LEASEDELTA;
4448 	if (sf->nsf_fp == NULL)
4449 		return;
4450 	error = NFSD_RDWR(UIO_READ, NFSFPVNODE(sf->nsf_fp),
4451 	    (caddr_t)&sf->nsf_rec, sizeof (struct nfsf_rec), off, UIO_SYSSPACE,
4452 	    0, NFSFPCRED(sf->nsf_fp), &aresid, p);
4453 	if (error || aresid || sf->nsf_numboots == 0 ||
4454 		sf->nsf_numboots > NFSNSF_MAXNUMBOOTS)
4455 		return;
4456 
4457 	/*
4458 	 * Now, read in the boottimes.
4459 	 */
4460 	sf->nsf_bootvals = (time_t *)malloc((sf->nsf_numboots + 1) *
4461 		sizeof (time_t), M_TEMP, M_WAITOK);
4462 	off = sizeof (struct nfsf_rec);
4463 	error = NFSD_RDWR(UIO_READ, NFSFPVNODE(sf->nsf_fp),
4464 	    (caddr_t)sf->nsf_bootvals, sf->nsf_numboots * sizeof (time_t), off,
4465 	    UIO_SYSSPACE, 0, NFSFPCRED(sf->nsf_fp), &aresid, p);
4466 	if (error || aresid) {
4467 		free((caddr_t)sf->nsf_bootvals, M_TEMP);
4468 		sf->nsf_bootvals = NULL;
4469 		return;
4470 	}
4471 
4472 	/*
4473 	 * Make sure this nfsrvboottime is different from all recorded
4474 	 * previous ones.
4475 	 */
4476 	do {
4477 		tryagain = 0;
4478 		for (i = 0; i < sf->nsf_numboots; i++) {
4479 			if (nfsrvboottime == sf->nsf_bootvals[i]) {
4480 				nfsrvboottime++;
4481 				tryagain = 1;
4482 				break;
4483 			}
4484 		}
4485 	} while (tryagain);
4486 
4487 	sf->nsf_flags |= NFSNSF_OK;
4488 	off += (sf->nsf_numboots * sizeof (time_t));
4489 
4490 	/*
4491 	 * Read through the file, building a list of records for grace
4492 	 * checking.
4493 	 * Each record is between sizeof (struct nfst_rec) and
4494 	 * sizeof (struct nfst_rec) + NFSV4_OPAQUELIMIT - 1
4495 	 * and is actually sizeof (struct nfst_rec) + nst_len - 1.
4496 	 */
4497 	tsp = (struct nfst_rec *)malloc(sizeof (struct nfst_rec) +
4498 		NFSV4_OPAQUELIMIT - 1, M_TEMP, M_WAITOK);
4499 	do {
4500 	    error = NFSD_RDWR(UIO_READ, NFSFPVNODE(sf->nsf_fp),
4501 	        (caddr_t)tsp, sizeof (struct nfst_rec) + NFSV4_OPAQUELIMIT - 1,
4502 	        off, UIO_SYSSPACE, 0, NFSFPCRED(sf->nsf_fp), &aresid, p);
4503 	    len = (sizeof (struct nfst_rec) + NFSV4_OPAQUELIMIT - 1) - aresid;
4504 	    if (error || (len > 0 && (len < sizeof (struct nfst_rec) ||
4505 		len < (sizeof (struct nfst_rec) + tsp->len - 1)))) {
4506 		/*
4507 		 * Yuck, the file has been corrupted, so just return
4508 		 * after clearing out any restart state, so the grace period
4509 		 * is over.
4510 		 */
4511 		LIST_FOREACH_SAFE(sp, &sf->nsf_head, nst_list, nsp) {
4512 			LIST_REMOVE(sp, nst_list);
4513 			free((caddr_t)sp, M_TEMP);
4514 		}
4515 		free((caddr_t)tsp, M_TEMP);
4516 		sf->nsf_flags &= ~NFSNSF_OK;
4517 		free((caddr_t)sf->nsf_bootvals, M_TEMP);
4518 		sf->nsf_bootvals = NULL;
4519 		return;
4520 	    }
4521 	    if (len > 0) {
4522 		off += sizeof (struct nfst_rec) + tsp->len - 1;
4523 		/*
4524 		 * Search the list for a matching client.
4525 		 */
4526 		LIST_FOREACH(sp, &sf->nsf_head, nst_list) {
4527 			if (tsp->len == sp->nst_len &&
4528 			    !NFSBCMP(tsp->client, sp->nst_client, tsp->len))
4529 				break;
4530 		}
4531 		if (sp == LIST_END(&sf->nsf_head)) {
4532 			sp = (struct nfsrv_stable *)malloc(tsp->len +
4533 				sizeof (struct nfsrv_stable) - 1, M_TEMP,
4534 				M_WAITOK);
4535 			NFSBCOPY((caddr_t)tsp, (caddr_t)&sp->nst_rec,
4536 				sizeof (struct nfst_rec) + tsp->len - 1);
4537 			LIST_INSERT_HEAD(&sf->nsf_head, sp, nst_list);
4538 		} else {
4539 			if (tsp->flag == NFSNST_REVOKE)
4540 				sp->nst_flag |= NFSNST_REVOKE;
4541 			else
4542 				/*
4543 				 * A subsequent timestamp indicates the client
4544 				 * did a setclientid/confirm and any previous
4545 				 * revoke is no longer relevant.
4546 				 */
4547 				sp->nst_flag &= ~NFSNST_REVOKE;
4548 		}
4549 	    }
4550 	} while (len > 0);
4551 	free((caddr_t)tsp, M_TEMP);
4552 	sf->nsf_flags = NFSNSF_OK;
4553 	sf->nsf_eograce = NFSD_MONOSEC + sf->nsf_lease +
4554 		NFSRV_LEASEDELTA;
4555 }
4556 
4557 /*
4558  * Update the stable storage file, now that the grace period is over.
4559  */
4560 APPLESTATIC void
4561 nfsrv_updatestable(NFSPROC_T *p)
4562 {
4563 	struct nfsrv_stablefirst *sf = &nfsrv_stablefirst;
4564 	struct nfsrv_stable *sp, *nsp;
4565 	int i;
4566 	struct nfsvattr nva;
4567 	vnode_t vp;
4568 #if defined(__FreeBSD_version) && (__FreeBSD_version >= 500000)
4569 	mount_t mp = NULL;
4570 #endif
4571 	int error;
4572 
4573 	if (sf->nsf_fp == NULL || (sf->nsf_flags & NFSNSF_UPDATEDONE))
4574 		return;
4575 	sf->nsf_flags |= NFSNSF_UPDATEDONE;
4576 	/*
4577 	 * Ok, we need to rewrite the stable storage file.
4578 	 * - truncate to 0 length
4579 	 * - write the new first structure
4580 	 * - loop through the data structures, writing out any that
4581 	 *   have timestamps older than the old boot
4582 	 */
4583 	if (sf->nsf_bootvals) {
4584 		sf->nsf_numboots++;
4585 		for (i = sf->nsf_numboots - 2; i >= 0; i--)
4586 			sf->nsf_bootvals[i + 1] = sf->nsf_bootvals[i];
4587 	} else {
4588 		sf->nsf_numboots = 1;
4589 		sf->nsf_bootvals = (time_t *)malloc(sizeof (time_t),
4590 			M_TEMP, M_WAITOK);
4591 	}
4592 	sf->nsf_bootvals[0] = nfsrvboottime;
4593 	sf->nsf_lease = nfsrv_lease;
4594 	NFSVNO_ATTRINIT(&nva);
4595 	NFSVNO_SETATTRVAL(&nva, size, 0);
4596 	vp = NFSFPVNODE(sf->nsf_fp);
4597 	vn_start_write(vp, &mp, V_WAIT);
4598 	if (NFSVOPLOCK(vp, LK_EXCLUSIVE) == 0) {
4599 		error = nfsvno_setattr(vp, &nva, NFSFPCRED(sf->nsf_fp), p,
4600 		    NULL);
4601 		NFSVOPUNLOCK(vp, 0);
4602 	} else
4603 		error = EPERM;
4604 	vn_finished_write(mp);
4605 	if (!error)
4606 	    error = NFSD_RDWR(UIO_WRITE, vp,
4607 		(caddr_t)&sf->nsf_rec, sizeof (struct nfsf_rec), (off_t)0,
4608 		UIO_SYSSPACE, IO_SYNC, NFSFPCRED(sf->nsf_fp), NULL, p);
4609 	if (!error)
4610 	    error = NFSD_RDWR(UIO_WRITE, vp,
4611 		(caddr_t)sf->nsf_bootvals,
4612 		sf->nsf_numboots * sizeof (time_t),
4613 		(off_t)(sizeof (struct nfsf_rec)),
4614 		UIO_SYSSPACE, IO_SYNC, NFSFPCRED(sf->nsf_fp), NULL, p);
4615 	free((caddr_t)sf->nsf_bootvals, M_TEMP);
4616 	sf->nsf_bootvals = NULL;
4617 	if (error) {
4618 		sf->nsf_flags &= ~NFSNSF_OK;
4619 		printf("EEK! Can't write NfsV4 stable storage file\n");
4620 		return;
4621 	}
4622 	sf->nsf_flags |= NFSNSF_OK;
4623 
4624 	/*
4625 	 * Loop through the list and write out timestamp records for
4626 	 * any clients that successfully reclaimed state.
4627 	 */
4628 	LIST_FOREACH_SAFE(sp, &sf->nsf_head, nst_list, nsp) {
4629 		if (sp->nst_flag & NFSNST_GOTSTATE) {
4630 			nfsrv_writestable(sp->nst_client, sp->nst_len,
4631 				NFSNST_NEWSTATE, p);
4632 			sp->nst_clp->lc_flags |= LCL_STAMPEDSTABLE;
4633 		}
4634 		LIST_REMOVE(sp, nst_list);
4635 		free((caddr_t)sp, M_TEMP);
4636 	}
4637 	nfsrv_backupstable();
4638 }
4639 
4640 /*
4641  * Append a record to the stable storage file.
4642  */
4643 APPLESTATIC void
4644 nfsrv_writestable(u_char *client, int len, int flag, NFSPROC_T *p)
4645 {
4646 	struct nfsrv_stablefirst *sf = &nfsrv_stablefirst;
4647 	struct nfst_rec *sp;
4648 	int error;
4649 
4650 	if (!(sf->nsf_flags & NFSNSF_OK) || sf->nsf_fp == NULL)
4651 		return;
4652 	sp = (struct nfst_rec *)malloc(sizeof (struct nfst_rec) +
4653 		len - 1, M_TEMP, M_WAITOK);
4654 	sp->len = len;
4655 	NFSBCOPY(client, sp->client, len);
4656 	sp->flag = flag;
4657 	error = NFSD_RDWR(UIO_WRITE, NFSFPVNODE(sf->nsf_fp),
4658 	    (caddr_t)sp, sizeof (struct nfst_rec) + len - 1, (off_t)0,
4659 	    UIO_SYSSPACE, (IO_SYNC | IO_APPEND), NFSFPCRED(sf->nsf_fp), NULL, p);
4660 	free((caddr_t)sp, M_TEMP);
4661 	if (error) {
4662 		sf->nsf_flags &= ~NFSNSF_OK;
4663 		printf("EEK! Can't write NfsV4 stable storage file\n");
4664 	}
4665 }
4666 
4667 /*
4668  * This function is called during the grace period to mark a client
4669  * that successfully reclaimed state.
4670  */
4671 static void
4672 nfsrv_markstable(struct nfsclient *clp)
4673 {
4674 	struct nfsrv_stable *sp;
4675 
4676 	/*
4677 	 * First find the client structure.
4678 	 */
4679 	LIST_FOREACH(sp, &nfsrv_stablefirst.nsf_head, nst_list) {
4680 		if (sp->nst_len == clp->lc_idlen &&
4681 		    !NFSBCMP(sp->nst_client, clp->lc_id, sp->nst_len))
4682 			break;
4683 	}
4684 	if (sp == LIST_END(&nfsrv_stablefirst.nsf_head))
4685 		return;
4686 
4687 	/*
4688 	 * Now, just mark it and set the nfsclient back pointer.
4689 	 */
4690 	sp->nst_flag |= NFSNST_GOTSTATE;
4691 	sp->nst_clp = clp;
4692 }
4693 
4694 /*
4695  * This function is called for a reclaim, to see if it gets grace.
4696  * It returns 0 if a reclaim is allowed, 1 otherwise.
4697  */
4698 static int
4699 nfsrv_checkstable(struct nfsclient *clp)
4700 {
4701 	struct nfsrv_stable *sp;
4702 
4703 	/*
4704 	 * First, find the entry for the client.
4705 	 */
4706 	LIST_FOREACH(sp, &nfsrv_stablefirst.nsf_head, nst_list) {
4707 		if (sp->nst_len == clp->lc_idlen &&
4708 		    !NFSBCMP(sp->nst_client, clp->lc_id, sp->nst_len))
4709 			break;
4710 	}
4711 
4712 	/*
4713 	 * If not in the list, state was revoked or no state was issued
4714 	 * since the previous reboot, a reclaim is denied.
4715 	 */
4716 	if (sp == LIST_END(&nfsrv_stablefirst.nsf_head) ||
4717 	    (sp->nst_flag & NFSNST_REVOKE) ||
4718 	    !(nfsrv_stablefirst.nsf_flags & NFSNSF_OK))
4719 		return (1);
4720 	return (0);
4721 }
4722 
4723 /*
4724  * Test for and try to clear out a conflicting client. This is called by
4725  * nfsrv_lockctrl() and nfsrv_openctrl() when conflicts with other clients
4726  * a found.
4727  * The trick here is that it can't revoke a conflicting client with an
4728  * expired lease unless it holds the v4root lock, so...
4729  * If no v4root lock, get the lock and return 1 to indicate "try again".
4730  * Return 0 to indicate the conflict can't be revoked and 1 to indicate
4731  * the revocation worked and the conflicting client is "bye, bye", so it
4732  * can be tried again.
4733  * Return 2 to indicate that the vnode is VI_DOOMED after NFSVOPLOCK().
4734  * Unlocks State before a non-zero value is returned.
4735  */
4736 static int
4737 nfsrv_clientconflict(struct nfsclient *clp, int *haslockp, vnode_t vp,
4738     NFSPROC_T *p)
4739 {
4740 	int gotlock, lktype = 0;
4741 
4742 	/*
4743 	 * If lease hasn't expired, we can't fix it.
4744 	 */
4745 	if (clp->lc_expiry >= NFSD_MONOSEC ||
4746 	    !(nfsrv_stablefirst.nsf_flags & NFSNSF_UPDATEDONE))
4747 		return (0);
4748 	if (*haslockp == 0) {
4749 		NFSUNLOCKSTATE();
4750 		if (vp != NULL) {
4751 			lktype = NFSVOPISLOCKED(vp);
4752 			NFSVOPUNLOCK(vp, 0);
4753 		}
4754 		NFSLOCKV4ROOTMUTEX();
4755 		nfsv4_relref(&nfsv4rootfs_lock);
4756 		do {
4757 			gotlock = nfsv4_lock(&nfsv4rootfs_lock, 1, NULL,
4758 			    NFSV4ROOTLOCKMUTEXPTR, NULL);
4759 		} while (!gotlock);
4760 		NFSUNLOCKV4ROOTMUTEX();
4761 		*haslockp = 1;
4762 		if (vp != NULL) {
4763 			NFSVOPLOCK(vp, lktype | LK_RETRY);
4764 			if ((vp->v_iflag & VI_DOOMED) != 0)
4765 				return (2);
4766 		}
4767 		return (1);
4768 	}
4769 	NFSUNLOCKSTATE();
4770 
4771 	/*
4772 	 * Ok, we can expire the conflicting client.
4773 	 */
4774 	nfsrv_writestable(clp->lc_id, clp->lc_idlen, NFSNST_REVOKE, p);
4775 	nfsrv_backupstable();
4776 	nfsrv_cleanclient(clp, p);
4777 	nfsrv_freedeleglist(&clp->lc_deleg);
4778 	nfsrv_freedeleglist(&clp->lc_olddeleg);
4779 	LIST_REMOVE(clp, lc_hash);
4780 	nfsrv_zapclient(clp, p);
4781 	return (1);
4782 }
4783 
4784 /*
4785  * Resolve a delegation conflict.
4786  * Returns 0 to indicate the conflict was resolved without sleeping.
4787  * Return -1 to indicate that the caller should check for conflicts again.
4788  * Return > 0 for an error that should be returned, normally NFSERR_DELAY.
4789  *
4790  * Also, manipulate the nfsv4root_lock, as required. It isn't changed
4791  * for a return of 0, since there was no sleep and it could be required
4792  * later. It is released for a return of NFSERR_DELAY, since the caller
4793  * will return that error. It is released when a sleep was done waiting
4794  * for the delegation to be returned or expire (so that other nfsds can
4795  * handle ops). Then, it must be acquired for the write to stable storage.
4796  * (This function is somewhat similar to nfsrv_clientconflict(), but
4797  *  the semantics differ in a couple of subtle ways. The return of 0
4798  *  indicates the conflict was resolved without sleeping here, not
4799  *  that the conflict can't be resolved and the handling of nfsv4root_lock
4800  *  differs, as noted above.)
4801  * Unlocks State before returning a non-zero value.
4802  */
4803 static int
4804 nfsrv_delegconflict(struct nfsstate *stp, int *haslockp, NFSPROC_T *p,
4805     vnode_t vp)
4806 {
4807 	struct nfsclient *clp = stp->ls_clp;
4808 	int gotlock, error, lktype = 0, retrycnt, zapped_clp;
4809 	nfsv4stateid_t tstateid;
4810 	fhandle_t tfh;
4811 
4812 	/*
4813 	 * If the conflict is with an old delegation...
4814 	 */
4815 	if (stp->ls_flags & NFSLCK_OLDDELEG) {
4816 		/*
4817 		 * You can delete it, if it has expired.
4818 		 */
4819 		if (clp->lc_delegtime < NFSD_MONOSEC) {
4820 			nfsrv_freedeleg(stp);
4821 			NFSUNLOCKSTATE();
4822 			error = -1;
4823 			goto out;
4824 		}
4825 		NFSUNLOCKSTATE();
4826 		/*
4827 		 * During this delay, the old delegation could expire or it
4828 		 * could be recovered by the client via an Open with
4829 		 * CLAIM_DELEGATE_PREV.
4830 		 * Release the nfsv4root_lock, if held.
4831 		 */
4832 		if (*haslockp) {
4833 			*haslockp = 0;
4834 			NFSLOCKV4ROOTMUTEX();
4835 			nfsv4_unlock(&nfsv4rootfs_lock, 1);
4836 			NFSUNLOCKV4ROOTMUTEX();
4837 		}
4838 		error = NFSERR_DELAY;
4839 		goto out;
4840 	}
4841 
4842 	/*
4843 	 * It's a current delegation, so:
4844 	 * - check to see if the delegation has expired
4845 	 *   - if so, get the v4root lock and then expire it
4846 	 */
4847 	if (!(stp->ls_flags & NFSLCK_DELEGRECALL)) {
4848 		/*
4849 		 * - do a recall callback, since not yet done
4850 		 * For now, never allow truncate to be set. To use
4851 		 * truncate safely, it must be guaranteed that the
4852 		 * Remove, Rename or Setattr with size of 0 will
4853 		 * succeed and that would require major changes to
4854 		 * the VFS/Vnode OPs.
4855 		 * Set the expiry time large enough so that it won't expire
4856 		 * until after the callback, then set it correctly, once
4857 		 * the callback is done. (The delegation will now time
4858 		 * out whether or not the Recall worked ok. The timeout
4859 		 * will be extended when ops are done on the delegation
4860 		 * stateid, up to the timelimit.)
4861 		 */
4862 		stp->ls_delegtime = NFSD_MONOSEC + (2 * nfsrv_lease) +
4863 		    NFSRV_LEASEDELTA;
4864 		stp->ls_delegtimelimit = NFSD_MONOSEC + (6 * nfsrv_lease) +
4865 		    NFSRV_LEASEDELTA;
4866 		stp->ls_flags |= NFSLCK_DELEGRECALL;
4867 
4868 		/*
4869 		 * Loop NFSRV_CBRETRYCNT times while the CBRecall replies
4870 		 * NFSERR_BADSTATEID or NFSERR_BADHANDLE. This is done
4871 		 * in order to try and avoid a race that could happen
4872 		 * when a CBRecall request passed the Open reply with
4873 		 * the delegation in it when transitting the network.
4874 		 * Since nfsrv_docallback will sleep, don't use stp after
4875 		 * the call.
4876 		 */
4877 		NFSBCOPY((caddr_t)&stp->ls_stateid, (caddr_t)&tstateid,
4878 		    sizeof (tstateid));
4879 		NFSBCOPY((caddr_t)&stp->ls_lfp->lf_fh, (caddr_t)&tfh,
4880 		    sizeof (tfh));
4881 		NFSUNLOCKSTATE();
4882 		if (*haslockp) {
4883 			*haslockp = 0;
4884 			NFSLOCKV4ROOTMUTEX();
4885 			nfsv4_unlock(&nfsv4rootfs_lock, 1);
4886 			NFSUNLOCKV4ROOTMUTEX();
4887 		}
4888 		retrycnt = 0;
4889 		do {
4890 		    error = nfsrv_docallback(clp, NFSV4OP_CBRECALL,
4891 			&tstateid, 0, &tfh, NULL, NULL, p);
4892 		    retrycnt++;
4893 		} while ((error == NFSERR_BADSTATEID ||
4894 		    error == NFSERR_BADHANDLE) && retrycnt < NFSV4_CBRETRYCNT);
4895 		error = NFSERR_DELAY;
4896 		goto out;
4897 	}
4898 
4899 	if (clp->lc_expiry >= NFSD_MONOSEC &&
4900 	    stp->ls_delegtime >= NFSD_MONOSEC) {
4901 		NFSUNLOCKSTATE();
4902 		/*
4903 		 * A recall has been done, but it has not yet expired.
4904 		 * So, RETURN_DELAY.
4905 		 */
4906 		if (*haslockp) {
4907 			*haslockp = 0;
4908 			NFSLOCKV4ROOTMUTEX();
4909 			nfsv4_unlock(&nfsv4rootfs_lock, 1);
4910 			NFSUNLOCKV4ROOTMUTEX();
4911 		}
4912 		error = NFSERR_DELAY;
4913 		goto out;
4914 	}
4915 
4916 	/*
4917 	 * If we don't yet have the lock, just get it and then return,
4918 	 * since we need that before deleting expired state, such as
4919 	 * this delegation.
4920 	 * When getting the lock, unlock the vnode, so other nfsds that
4921 	 * are in progress, won't get stuck waiting for the vnode lock.
4922 	 */
4923 	if (*haslockp == 0) {
4924 		NFSUNLOCKSTATE();
4925 		if (vp != NULL) {
4926 			lktype = NFSVOPISLOCKED(vp);
4927 			NFSVOPUNLOCK(vp, 0);
4928 		}
4929 		NFSLOCKV4ROOTMUTEX();
4930 		nfsv4_relref(&nfsv4rootfs_lock);
4931 		do {
4932 			gotlock = nfsv4_lock(&nfsv4rootfs_lock, 1, NULL,
4933 			    NFSV4ROOTLOCKMUTEXPTR, NULL);
4934 		} while (!gotlock);
4935 		NFSUNLOCKV4ROOTMUTEX();
4936 		*haslockp = 1;
4937 		if (vp != NULL) {
4938 			NFSVOPLOCK(vp, lktype | LK_RETRY);
4939 			if ((vp->v_iflag & VI_DOOMED) != 0) {
4940 				*haslockp = 0;
4941 				NFSLOCKV4ROOTMUTEX();
4942 				nfsv4_unlock(&nfsv4rootfs_lock, 1);
4943 				NFSUNLOCKV4ROOTMUTEX();
4944 				error = NFSERR_PERM;
4945 				goto out;
4946 			}
4947 		}
4948 		error = -1;
4949 		goto out;
4950 	}
4951 
4952 	NFSUNLOCKSTATE();
4953 	/*
4954 	 * Ok, we can delete the expired delegation.
4955 	 * First, write the Revoke record to stable storage and then
4956 	 * clear out the conflict.
4957 	 * Since all other nfsd threads are now blocked, we can safely
4958 	 * sleep without the state changing.
4959 	 */
4960 	nfsrv_writestable(clp->lc_id, clp->lc_idlen, NFSNST_REVOKE, p);
4961 	nfsrv_backupstable();
4962 	if (clp->lc_expiry < NFSD_MONOSEC) {
4963 		nfsrv_cleanclient(clp, p);
4964 		nfsrv_freedeleglist(&clp->lc_deleg);
4965 		nfsrv_freedeleglist(&clp->lc_olddeleg);
4966 		LIST_REMOVE(clp, lc_hash);
4967 		zapped_clp = 1;
4968 	} else {
4969 		nfsrv_freedeleg(stp);
4970 		zapped_clp = 0;
4971 	}
4972 	if (zapped_clp)
4973 		nfsrv_zapclient(clp, p);
4974 	error = -1;
4975 
4976 out:
4977 	NFSEXITCODE(error);
4978 	return (error);
4979 }
4980 
4981 /*
4982  * Check for a remove allowed, if remove is set to 1 and get rid of
4983  * delegations.
4984  */
4985 APPLESTATIC int
4986 nfsrv_checkremove(vnode_t vp, int remove, NFSPROC_T *p)
4987 {
4988 	struct nfsstate *stp;
4989 	struct nfslockfile *lfp;
4990 	int error, haslock = 0;
4991 	fhandle_t nfh;
4992 
4993 	/*
4994 	 * First, get the lock file structure.
4995 	 * (A return of -1 means no associated state, so remove ok.)
4996 	 */
4997 	error = nfsrv_getlockfh(vp, NFSLCK_CHECK, NULL, &nfh, p);
4998 tryagain:
4999 	NFSLOCKSTATE();
5000 	if (!error)
5001 		error = nfsrv_getlockfile(NFSLCK_CHECK, NULL, &lfp, &nfh, 0);
5002 	if (error) {
5003 		NFSUNLOCKSTATE();
5004 		if (haslock) {
5005 			NFSLOCKV4ROOTMUTEX();
5006 			nfsv4_unlock(&nfsv4rootfs_lock, 1);
5007 			NFSUNLOCKV4ROOTMUTEX();
5008 		}
5009 		if (error == -1)
5010 			error = 0;
5011 		goto out;
5012 	}
5013 
5014 	/*
5015 	 * Now, we must Recall any delegations.
5016 	 */
5017 	error = nfsrv_cleandeleg(vp, lfp, NULL, &haslock, p);
5018 	if (error) {
5019 		/*
5020 		 * nfsrv_cleandeleg() unlocks state for non-zero
5021 		 * return.
5022 		 */
5023 		if (error == -1)
5024 			goto tryagain;
5025 		if (haslock) {
5026 			NFSLOCKV4ROOTMUTEX();
5027 			nfsv4_unlock(&nfsv4rootfs_lock, 1);
5028 			NFSUNLOCKV4ROOTMUTEX();
5029 		}
5030 		goto out;
5031 	}
5032 
5033 	/*
5034 	 * Now, look for a conflicting open share.
5035 	 */
5036 	if (remove) {
5037 		/*
5038 		 * If the entry in the directory was the last reference to the
5039 		 * corresponding filesystem object, the object can be destroyed
5040 		 * */
5041 		if(lfp->lf_usecount>1)
5042 			LIST_FOREACH(stp, &lfp->lf_open, ls_file) {
5043 				if (stp->ls_flags & NFSLCK_WRITEDENY) {
5044 					error = NFSERR_FILEOPEN;
5045 					break;
5046 				}
5047 			}
5048 	}
5049 
5050 	NFSUNLOCKSTATE();
5051 	if (haslock) {
5052 		NFSLOCKV4ROOTMUTEX();
5053 		nfsv4_unlock(&nfsv4rootfs_lock, 1);
5054 		NFSUNLOCKV4ROOTMUTEX();
5055 	}
5056 
5057 out:
5058 	NFSEXITCODE(error);
5059 	return (error);
5060 }
5061 
5062 /*
5063  * Clear out all delegations for the file referred to by lfp.
5064  * May return NFSERR_DELAY, if there will be a delay waiting for
5065  * delegations to expire.
5066  * Returns -1 to indicate it slept while recalling a delegation.
5067  * This function has the side effect of deleting the nfslockfile structure,
5068  * if it no longer has associated state and didn't have to sleep.
5069  * Unlocks State before a non-zero value is returned.
5070  */
5071 static int
5072 nfsrv_cleandeleg(vnode_t vp, struct nfslockfile *lfp,
5073     struct nfsclient *clp, int *haslockp, NFSPROC_T *p)
5074 {
5075 	struct nfsstate *stp, *nstp;
5076 	int ret = 0;
5077 
5078 	stp = LIST_FIRST(&lfp->lf_deleg);
5079 	while (stp != LIST_END(&lfp->lf_deleg)) {
5080 		nstp = LIST_NEXT(stp, ls_file);
5081 		if (stp->ls_clp != clp) {
5082 			ret = nfsrv_delegconflict(stp, haslockp, p, vp);
5083 			if (ret) {
5084 				/*
5085 				 * nfsrv_delegconflict() unlocks state
5086 				 * when it returns non-zero.
5087 				 */
5088 				goto out;
5089 			}
5090 		}
5091 		stp = nstp;
5092 	}
5093 out:
5094 	NFSEXITCODE(ret);
5095 	return (ret);
5096 }
5097 
5098 /*
5099  * There are certain operations that, when being done outside of NFSv4,
5100  * require that any NFSv4 delegation for the file be recalled.
5101  * This function is to be called for those cases:
5102  * VOP_RENAME() - When a delegation is being recalled for any reason,
5103  *	the client may have to do Opens against the server, using the file's
5104  *	final component name. If the file has been renamed on the server,
5105  *	that component name will be incorrect and the Open will fail.
5106  * VOP_REMOVE() - Theoretically, a client could Open a file after it has
5107  *	been removed on the server, if there is a delegation issued to
5108  *	that client for the file. I say "theoretically" since clients
5109  *	normally do an Access Op before the Open and that Access Op will
5110  *	fail with ESTALE. Note that NFSv2 and 3 don't even do Opens, so
5111  *	they will detect the file's removal in the same manner. (There is
5112  *	one case where RFC3530 allows a client to do an Open without first
5113  *	doing an Access Op, which is passage of a check against the ACE
5114  *	returned with a Write delegation, but current practice is to ignore
5115  *	the ACE and always do an Access Op.)
5116  *	Since the functions can only be called with an unlocked vnode, this
5117  *	can't be done at this time.
5118  * VOP_ADVLOCK() - When a client holds a delegation, it can issue byte range
5119  *	locks locally in the client, which are not visible to the server. To
5120  *	deal with this, issuing of delegations for a vnode must be disabled
5121  *	and all delegations for the vnode recalled. This is done via the
5122  *	second function, using the VV_DISABLEDELEG vflag on the vnode.
5123  */
5124 APPLESTATIC void
5125 nfsd_recalldelegation(vnode_t vp, NFSPROC_T *p)
5126 {
5127 	time_t starttime;
5128 	int error;
5129 
5130 	/*
5131 	 * First, check to see if the server is currently running and it has
5132 	 * been called for a regular file when issuing delegations.
5133 	 */
5134 	if (newnfs_numnfsd == 0 || vp->v_type != VREG ||
5135 	    nfsrv_issuedelegs == 0)
5136 		return;
5137 
5138 	KASSERT((NFSVOPISLOCKED(vp) != LK_EXCLUSIVE), ("vp %p is locked", vp));
5139 	/*
5140 	 * First, get a reference on the nfsv4rootfs_lock so that an
5141 	 * exclusive lock cannot be acquired by another thread.
5142 	 */
5143 	NFSLOCKV4ROOTMUTEX();
5144 	nfsv4_getref(&nfsv4rootfs_lock, NULL, NFSV4ROOTLOCKMUTEXPTR, NULL);
5145 	NFSUNLOCKV4ROOTMUTEX();
5146 
5147 	/*
5148 	 * Now, call nfsrv_checkremove() in a loop while it returns
5149 	 * NFSERR_DELAY. Return upon any other error or when timed out.
5150 	 */
5151 	starttime = NFSD_MONOSEC;
5152 	do {
5153 		if (NFSVOPLOCK(vp, LK_EXCLUSIVE) == 0) {
5154 			error = nfsrv_checkremove(vp, 0, p);
5155 			NFSVOPUNLOCK(vp, 0);
5156 		} else
5157 			error = EPERM;
5158 		if (error == NFSERR_DELAY) {
5159 			if (NFSD_MONOSEC - starttime > NFS_REMOVETIMEO)
5160 				break;
5161 			/* Sleep for a short period of time */
5162 			(void) nfs_catnap(PZERO, 0, "nfsremove");
5163 		}
5164 	} while (error == NFSERR_DELAY);
5165 	NFSLOCKV4ROOTMUTEX();
5166 	nfsv4_relref(&nfsv4rootfs_lock);
5167 	NFSUNLOCKV4ROOTMUTEX();
5168 }
5169 
5170 APPLESTATIC void
5171 nfsd_disabledelegation(vnode_t vp, NFSPROC_T *p)
5172 {
5173 
5174 #ifdef VV_DISABLEDELEG
5175 	/*
5176 	 * First, flag issuance of delegations disabled.
5177 	 */
5178 	atomic_set_long(&vp->v_vflag, VV_DISABLEDELEG);
5179 #endif
5180 
5181 	/*
5182 	 * Then call nfsd_recalldelegation() to get rid of all extant
5183 	 * delegations.
5184 	 */
5185 	nfsd_recalldelegation(vp, p);
5186 }
5187 
5188 /*
5189  * Check for conflicting locks, etc. and then get rid of delegations.
5190  * (At one point I thought that I should get rid of delegations for any
5191  *  Setattr, since it could potentially disallow the I/O op (read or write)
5192  *  allowed by the delegation. However, Setattr Ops that aren't changing
5193  *  the size get a stateid of all 0s, so you can't tell if it is a delegation
5194  *  for the same client or a different one, so I decided to only get rid
5195  *  of delegations for other clients when the size is being changed.)
5196  * In general, a Setattr can disable NFS I/O Ops that are outstanding, such
5197  * as Write backs, even if there is no delegation, so it really isn't any
5198  * different?)
5199  */
5200 APPLESTATIC int
5201 nfsrv_checksetattr(vnode_t vp, struct nfsrv_descript *nd,
5202     nfsv4stateid_t *stateidp, struct nfsvattr *nvap, nfsattrbit_t *attrbitp,
5203     struct nfsexstuff *exp, NFSPROC_T *p)
5204 {
5205 	struct nfsstate st, *stp = &st;
5206 	struct nfslock lo, *lop = &lo;
5207 	int error = 0;
5208 	nfsquad_t clientid;
5209 
5210 	if (NFSISSET_ATTRBIT(attrbitp, NFSATTRBIT_SIZE)) {
5211 		stp->ls_flags = (NFSLCK_CHECK | NFSLCK_WRITEACCESS);
5212 		lop->lo_first = nvap->na_size;
5213 	} else {
5214 		stp->ls_flags = 0;
5215 		lop->lo_first = 0;
5216 	}
5217 	if (NFSISSET_ATTRBIT(attrbitp, NFSATTRBIT_OWNER) ||
5218 	    NFSISSET_ATTRBIT(attrbitp, NFSATTRBIT_OWNERGROUP) ||
5219 	    NFSISSET_ATTRBIT(attrbitp, NFSATTRBIT_MODE) ||
5220 	    NFSISSET_ATTRBIT(attrbitp, NFSATTRBIT_ACL))
5221 		stp->ls_flags |= NFSLCK_SETATTR;
5222 	if (stp->ls_flags == 0)
5223 		goto out;
5224 	lop->lo_end = NFS64BITSSET;
5225 	lop->lo_flags = NFSLCK_WRITE;
5226 	stp->ls_ownerlen = 0;
5227 	stp->ls_op = NULL;
5228 	stp->ls_uid = nd->nd_cred->cr_uid;
5229 	stp->ls_stateid.seqid = stateidp->seqid;
5230 	clientid.lval[0] = stp->ls_stateid.other[0] = stateidp->other[0];
5231 	clientid.lval[1] = stp->ls_stateid.other[1] = stateidp->other[1];
5232 	stp->ls_stateid.other[2] = stateidp->other[2];
5233 	error = nfsrv_lockctrl(vp, &stp, &lop, NULL, clientid,
5234 	    stateidp, exp, nd, p);
5235 
5236 out:
5237 	NFSEXITCODE2(error, nd);
5238 	return (error);
5239 }
5240 
5241 /*
5242  * Check for a write delegation and do a CBGETATTR if there is one, updating
5243  * the attributes, as required.
5244  * Should I return an error if I can't get the attributes? (For now, I'll
5245  * just return ok.
5246  */
5247 APPLESTATIC int
5248 nfsrv_checkgetattr(struct nfsrv_descript *nd, vnode_t vp,
5249     struct nfsvattr *nvap, nfsattrbit_t *attrbitp, struct ucred *cred,
5250     NFSPROC_T *p)
5251 {
5252 	struct nfsstate *stp;
5253 	struct nfslockfile *lfp;
5254 	struct nfsclient *clp;
5255 	struct nfsvattr nva;
5256 	fhandle_t nfh;
5257 	int error = 0;
5258 	nfsattrbit_t cbbits;
5259 	u_quad_t delegfilerev;
5260 
5261 	NFSCBGETATTR_ATTRBIT(attrbitp, &cbbits);
5262 	if (!NFSNONZERO_ATTRBIT(&cbbits))
5263 		goto out;
5264 
5265 	/*
5266 	 * Get the lock file structure.
5267 	 * (A return of -1 means no associated state, so return ok.)
5268 	 */
5269 	error = nfsrv_getlockfh(vp, NFSLCK_CHECK, NULL, &nfh, p);
5270 	NFSLOCKSTATE();
5271 	if (!error)
5272 		error = nfsrv_getlockfile(NFSLCK_CHECK, NULL, &lfp, &nfh, 0);
5273 	if (error) {
5274 		NFSUNLOCKSTATE();
5275 		if (error == -1)
5276 			error = 0;
5277 		goto out;
5278 	}
5279 
5280 	/*
5281 	 * Now, look for a write delegation.
5282 	 */
5283 	LIST_FOREACH(stp, &lfp->lf_deleg, ls_file) {
5284 		if (stp->ls_flags & NFSLCK_DELEGWRITE)
5285 			break;
5286 	}
5287 	if (stp == LIST_END(&lfp->lf_deleg)) {
5288 		NFSUNLOCKSTATE();
5289 		goto out;
5290 	}
5291 	clp = stp->ls_clp;
5292 	delegfilerev = stp->ls_filerev;
5293 
5294 	/*
5295 	 * If the Write delegation was issued as a part of this Compound RPC
5296 	 * or if we have an Implied Clientid (used in a previous Op in this
5297 	 * compound) and it is the client the delegation was issued to,
5298 	 * just return ok.
5299 	 * I also assume that it is from the same client iff the network
5300 	 * host IP address is the same as the callback address. (Not
5301 	 * exactly correct by the RFC, but avoids a lot of Getattr
5302 	 * callbacks.)
5303 	 */
5304 	if (nd->nd_compref == stp->ls_compref ||
5305 	    ((nd->nd_flag & ND_IMPLIEDCLID) &&
5306 	     clp->lc_clientid.qval == nd->nd_clientid.qval) ||
5307 	     nfsaddr2_match(clp->lc_req.nr_nam, nd->nd_nam)) {
5308 		NFSUNLOCKSTATE();
5309 		goto out;
5310 	}
5311 
5312 	/*
5313 	 * We are now done with the delegation state structure,
5314 	 * so the statelock can be released and we can now tsleep().
5315 	 */
5316 
5317 	/*
5318 	 * Now, we must do the CB Getattr callback, to see if Change or Size
5319 	 * has changed.
5320 	 */
5321 	if (clp->lc_expiry >= NFSD_MONOSEC) {
5322 		NFSUNLOCKSTATE();
5323 		NFSVNO_ATTRINIT(&nva);
5324 		nva.na_filerev = NFS64BITSSET;
5325 		error = nfsrv_docallback(clp, NFSV4OP_CBGETATTR, NULL,
5326 		    0, &nfh, &nva, &cbbits, p);
5327 		if (!error) {
5328 			if ((nva.na_filerev != NFS64BITSSET &&
5329 			    nva.na_filerev > delegfilerev) ||
5330 			    (NFSVNO_ISSETSIZE(&nva) &&
5331 			     nva.na_size != nvap->na_size)) {
5332 				error = nfsvno_updfilerev(vp, nvap, cred, p);
5333 				if (NFSVNO_ISSETSIZE(&nva))
5334 					nvap->na_size = nva.na_size;
5335 			}
5336 		} else
5337 			error = 0;	/* Ignore callback errors for now. */
5338 	} else {
5339 		NFSUNLOCKSTATE();
5340 	}
5341 
5342 out:
5343 	NFSEXITCODE2(error, nd);
5344 	return (error);
5345 }
5346 
5347 /*
5348  * This function looks for openowners that haven't had any opens for
5349  * a while and throws them away. Called by an nfsd when NFSNSF_NOOPENS
5350  * is set.
5351  */
5352 APPLESTATIC void
5353 nfsrv_throwawayopens(NFSPROC_T *p)
5354 {
5355 	struct nfsclient *clp, *nclp;
5356 	struct nfsstate *stp, *nstp;
5357 	int i;
5358 
5359 	NFSLOCKSTATE();
5360 	nfsrv_stablefirst.nsf_flags &= ~NFSNSF_NOOPENS;
5361 	/*
5362 	 * For each client...
5363 	 */
5364 	for (i = 0; i < nfsrv_clienthashsize; i++) {
5365 	    LIST_FOREACH_SAFE(clp, &nfsclienthash[i], lc_hash, nclp) {
5366 		LIST_FOREACH_SAFE(stp, &clp->lc_open, ls_list, nstp) {
5367 			if (LIST_EMPTY(&stp->ls_open) &&
5368 			    (stp->ls_noopens > NFSNOOPEN ||
5369 			     (nfsrv_openpluslock * 2) >
5370 			     nfsrv_v4statelimit))
5371 				nfsrv_freeopenowner(stp, 0, p);
5372 		}
5373 	    }
5374 	}
5375 	NFSUNLOCKSTATE();
5376 }
5377 
5378 /*
5379  * This function checks to see if the credentials are the same.
5380  * Returns 1 for not same, 0 otherwise.
5381  */
5382 static int
5383 nfsrv_notsamecredname(struct nfsrv_descript *nd, struct nfsclient *clp)
5384 {
5385 
5386 	if (nd->nd_flag & ND_GSS) {
5387 		if (!(clp->lc_flags & LCL_GSS))
5388 			return (1);
5389 		if (clp->lc_flags & LCL_NAME) {
5390 			if (nd->nd_princlen != clp->lc_namelen ||
5391 			    NFSBCMP(nd->nd_principal, clp->lc_name,
5392 				clp->lc_namelen))
5393 				return (1);
5394 			else
5395 				return (0);
5396 		}
5397 		if (nd->nd_cred->cr_uid == clp->lc_uid)
5398 			return (0);
5399 		else
5400 			return (1);
5401 	} else if (clp->lc_flags & LCL_GSS)
5402 		return (1);
5403 	/*
5404 	 * For AUTH_SYS, allow the same uid or root. (This is underspecified
5405 	 * in RFC3530, which talks about principals, but doesn't say anything
5406 	 * about uids for AUTH_SYS.)
5407 	 */
5408 	if (nd->nd_cred->cr_uid == clp->lc_uid || nd->nd_cred->cr_uid == 0)
5409 		return (0);
5410 	else
5411 		return (1);
5412 }
5413 
5414 /*
5415  * Calculate the lease expiry time.
5416  */
5417 static time_t
5418 nfsrv_leaseexpiry(void)
5419 {
5420 
5421 	if (nfsrv_stablefirst.nsf_eograce > NFSD_MONOSEC)
5422 		return (NFSD_MONOSEC + 2 * (nfsrv_lease + NFSRV_LEASEDELTA));
5423 	return (NFSD_MONOSEC + nfsrv_lease + NFSRV_LEASEDELTA);
5424 }
5425 
5426 /*
5427  * Delay the delegation timeout as far as ls_delegtimelimit, as required.
5428  */
5429 static void
5430 nfsrv_delaydelegtimeout(struct nfsstate *stp)
5431 {
5432 
5433 	if ((stp->ls_flags & NFSLCK_DELEGRECALL) == 0)
5434 		return;
5435 
5436 	if ((stp->ls_delegtime + 15) > NFSD_MONOSEC &&
5437 	    stp->ls_delegtime < stp->ls_delegtimelimit) {
5438 		stp->ls_delegtime += nfsrv_lease;
5439 		if (stp->ls_delegtime > stp->ls_delegtimelimit)
5440 			stp->ls_delegtime = stp->ls_delegtimelimit;
5441 	}
5442 }
5443 
5444 /*
5445  * This function checks to see if there is any other state associated
5446  * with the openowner for this Open.
5447  * It returns 1 if there is no other state, 0 otherwise.
5448  */
5449 static int
5450 nfsrv_nootherstate(struct nfsstate *stp)
5451 {
5452 	struct nfsstate *tstp;
5453 
5454 	LIST_FOREACH(tstp, &stp->ls_openowner->ls_open, ls_list) {
5455 		if (tstp != stp || !LIST_EMPTY(&tstp->ls_lock))
5456 			return (0);
5457 	}
5458 	return (1);
5459 }
5460 
5461 /*
5462  * Create a list of lock deltas (changes to local byte range locking
5463  * that can be rolled back using the list) and apply the changes via
5464  * nfsvno_advlock(). Optionally, lock the list. It is expected that either
5465  * the rollback or update function will be called after this.
5466  * It returns an error (and rolls back, as required), if any nfsvno_advlock()
5467  * call fails. If it returns an error, it will unlock the list.
5468  */
5469 static int
5470 nfsrv_locallock(vnode_t vp, struct nfslockfile *lfp, int flags,
5471     uint64_t first, uint64_t end, struct nfslockconflict *cfp, NFSPROC_T *p)
5472 {
5473 	struct nfslock *lop, *nlop;
5474 	int error = 0;
5475 
5476 	/* Loop through the list of locks. */
5477 	lop = LIST_FIRST(&lfp->lf_locallock);
5478 	while (first < end && lop != NULL) {
5479 		nlop = LIST_NEXT(lop, lo_lckowner);
5480 		if (first >= lop->lo_end) {
5481 			/* not there yet */
5482 			lop = nlop;
5483 		} else if (first < lop->lo_first) {
5484 			/* new one starts before entry in list */
5485 			if (end <= lop->lo_first) {
5486 				/* no overlap between old and new */
5487 				error = nfsrv_dolocal(vp, lfp, flags,
5488 				    NFSLCK_UNLOCK, first, end, cfp, p);
5489 				if (error != 0)
5490 					break;
5491 				first = end;
5492 			} else {
5493 				/* handle fragment overlapped with new one */
5494 				error = nfsrv_dolocal(vp, lfp, flags,
5495 				    NFSLCK_UNLOCK, first, lop->lo_first, cfp,
5496 				    p);
5497 				if (error != 0)
5498 					break;
5499 				first = lop->lo_first;
5500 			}
5501 		} else {
5502 			/* new one overlaps this entry in list */
5503 			if (end <= lop->lo_end) {
5504 				/* overlaps all of new one */
5505 				error = nfsrv_dolocal(vp, lfp, flags,
5506 				    lop->lo_flags, first, end, cfp, p);
5507 				if (error != 0)
5508 					break;
5509 				first = end;
5510 			} else {
5511 				/* handle fragment overlapped with new one */
5512 				error = nfsrv_dolocal(vp, lfp, flags,
5513 				    lop->lo_flags, first, lop->lo_end, cfp, p);
5514 				if (error != 0)
5515 					break;
5516 				first = lop->lo_end;
5517 				lop = nlop;
5518 			}
5519 		}
5520 	}
5521 	if (first < end && error == 0)
5522 		/* handle fragment past end of list */
5523 		error = nfsrv_dolocal(vp, lfp, flags, NFSLCK_UNLOCK, first,
5524 		    end, cfp, p);
5525 
5526 	NFSEXITCODE(error);
5527 	return (error);
5528 }
5529 
5530 /*
5531  * Local lock unlock. Unlock all byte ranges that are no longer locked
5532  * by NFSv4. To do this, unlock any subranges of first-->end that
5533  * do not overlap with the byte ranges of any lock in the lfp->lf_lock
5534  * list. This list has all locks for the file held by other
5535  * <clientid, lockowner> tuples. The list is ordered by increasing
5536  * lo_first value, but may have entries that overlap each other, for
5537  * the case of read locks.
5538  */
5539 static void
5540 nfsrv_localunlock(vnode_t vp, struct nfslockfile *lfp, uint64_t init_first,
5541     uint64_t init_end, NFSPROC_T *p)
5542 {
5543 	struct nfslock *lop;
5544 	uint64_t first, end, prevfirst;
5545 
5546 	first = init_first;
5547 	end = init_end;
5548 	while (first < init_end) {
5549 		/* Loop through all nfs locks, adjusting first and end */
5550 		prevfirst = 0;
5551 		LIST_FOREACH(lop, &lfp->lf_lock, lo_lckfile) {
5552 			KASSERT(prevfirst <= lop->lo_first,
5553 			    ("nfsv4 locks out of order"));
5554 			KASSERT(lop->lo_first < lop->lo_end,
5555 			    ("nfsv4 bogus lock"));
5556 			prevfirst = lop->lo_first;
5557 			if (first >= lop->lo_first &&
5558 			    first < lop->lo_end)
5559 				/*
5560 				 * Overlaps with initial part, so trim
5561 				 * off that initial part by moving first past
5562 				 * it.
5563 				 */
5564 				first = lop->lo_end;
5565 			else if (end > lop->lo_first &&
5566 			    lop->lo_first > first) {
5567 				/*
5568 				 * This lock defines the end of the
5569 				 * segment to unlock, so set end to the
5570 				 * start of it and break out of the loop.
5571 				 */
5572 				end = lop->lo_first;
5573 				break;
5574 			}
5575 			if (first >= end)
5576 				/*
5577 				 * There is no segment left to do, so
5578 				 * break out of this loop and then exit
5579 				 * the outer while() since first will be set
5580 				 * to end, which must equal init_end here.
5581 				 */
5582 				break;
5583 		}
5584 		if (first < end) {
5585 			/* Unlock this segment */
5586 			(void) nfsrv_dolocal(vp, lfp, NFSLCK_UNLOCK,
5587 			    NFSLCK_READ, first, end, NULL, p);
5588 			nfsrv_locallock_commit(lfp, NFSLCK_UNLOCK,
5589 			    first, end);
5590 		}
5591 		/*
5592 		 * Now move past this segment and look for any further
5593 		 * segment in the range, if there is one.
5594 		 */
5595 		first = end;
5596 		end = init_end;
5597 	}
5598 }
5599 
5600 /*
5601  * Do the local lock operation and update the rollback list, as required.
5602  * Perform the rollback and return the error if nfsvno_advlock() fails.
5603  */
5604 static int
5605 nfsrv_dolocal(vnode_t vp, struct nfslockfile *lfp, int flags, int oldflags,
5606     uint64_t first, uint64_t end, struct nfslockconflict *cfp, NFSPROC_T *p)
5607 {
5608 	struct nfsrollback *rlp;
5609 	int error = 0, ltype, oldltype;
5610 
5611 	if (flags & NFSLCK_WRITE)
5612 		ltype = F_WRLCK;
5613 	else if (flags & NFSLCK_READ)
5614 		ltype = F_RDLCK;
5615 	else
5616 		ltype = F_UNLCK;
5617 	if (oldflags & NFSLCK_WRITE)
5618 		oldltype = F_WRLCK;
5619 	else if (oldflags & NFSLCK_READ)
5620 		oldltype = F_RDLCK;
5621 	else
5622 		oldltype = F_UNLCK;
5623 	if (ltype == oldltype || (oldltype == F_WRLCK && ltype == F_RDLCK))
5624 		/* nothing to do */
5625 		goto out;
5626 	error = nfsvno_advlock(vp, ltype, first, end, p);
5627 	if (error != 0) {
5628 		if (cfp != NULL) {
5629 			cfp->cl_clientid.lval[0] = 0;
5630 			cfp->cl_clientid.lval[1] = 0;
5631 			cfp->cl_first = 0;
5632 			cfp->cl_end = NFS64BITSSET;
5633 			cfp->cl_flags = NFSLCK_WRITE;
5634 			cfp->cl_ownerlen = 5;
5635 			NFSBCOPY("LOCAL", cfp->cl_owner, 5);
5636 		}
5637 		nfsrv_locallock_rollback(vp, lfp, p);
5638 	} else if (ltype != F_UNLCK) {
5639 		rlp = malloc(sizeof (struct nfsrollback), M_NFSDROLLBACK,
5640 		    M_WAITOK);
5641 		rlp->rlck_first = first;
5642 		rlp->rlck_end = end;
5643 		rlp->rlck_type = oldltype;
5644 		LIST_INSERT_HEAD(&lfp->lf_rollback, rlp, rlck_list);
5645 	}
5646 
5647 out:
5648 	NFSEXITCODE(error);
5649 	return (error);
5650 }
5651 
5652 /*
5653  * Roll back local lock changes and free up the rollback list.
5654  */
5655 static void
5656 nfsrv_locallock_rollback(vnode_t vp, struct nfslockfile *lfp, NFSPROC_T *p)
5657 {
5658 	struct nfsrollback *rlp, *nrlp;
5659 
5660 	LIST_FOREACH_SAFE(rlp, &lfp->lf_rollback, rlck_list, nrlp) {
5661 		(void) nfsvno_advlock(vp, rlp->rlck_type, rlp->rlck_first,
5662 		    rlp->rlck_end, p);
5663 		free(rlp, M_NFSDROLLBACK);
5664 	}
5665 	LIST_INIT(&lfp->lf_rollback);
5666 }
5667 
5668 /*
5669  * Update local lock list and delete rollback list (ie now committed to the
5670  * local locks). Most of the work is done by the internal function.
5671  */
5672 static void
5673 nfsrv_locallock_commit(struct nfslockfile *lfp, int flags, uint64_t first,
5674     uint64_t end)
5675 {
5676 	struct nfsrollback *rlp, *nrlp;
5677 	struct nfslock *new_lop, *other_lop;
5678 
5679 	new_lop = malloc(sizeof (struct nfslock), M_NFSDLOCK, M_WAITOK);
5680 	if (flags & (NFSLCK_READ | NFSLCK_WRITE))
5681 		other_lop = malloc(sizeof (struct nfslock), M_NFSDLOCK,
5682 		    M_WAITOK);
5683 	else
5684 		other_lop = NULL;
5685 	new_lop->lo_flags = flags;
5686 	new_lop->lo_first = first;
5687 	new_lop->lo_end = end;
5688 	nfsrv_updatelock(NULL, &new_lop, &other_lop, lfp);
5689 	if (new_lop != NULL)
5690 		free(new_lop, M_NFSDLOCK);
5691 	if (other_lop != NULL)
5692 		free(other_lop, M_NFSDLOCK);
5693 
5694 	/* and get rid of the rollback list */
5695 	LIST_FOREACH_SAFE(rlp, &lfp->lf_rollback, rlck_list, nrlp)
5696 		free(rlp, M_NFSDROLLBACK);
5697 	LIST_INIT(&lfp->lf_rollback);
5698 }
5699 
5700 /*
5701  * Lock the struct nfslockfile for local lock updating.
5702  */
5703 static void
5704 nfsrv_locklf(struct nfslockfile *lfp)
5705 {
5706 	int gotlock;
5707 
5708 	/* lf_usecount ensures *lfp won't be free'd */
5709 	lfp->lf_usecount++;
5710 	do {
5711 		gotlock = nfsv4_lock(&lfp->lf_locallock_lck, 1, NULL,
5712 		    NFSSTATEMUTEXPTR, NULL);
5713 	} while (gotlock == 0);
5714 	lfp->lf_usecount--;
5715 }
5716 
5717 /*
5718  * Unlock the struct nfslockfile after local lock updating.
5719  */
5720 static void
5721 nfsrv_unlocklf(struct nfslockfile *lfp)
5722 {
5723 
5724 	nfsv4_unlock(&lfp->lf_locallock_lck, 0);
5725 }
5726 
5727 /*
5728  * Clear out all state for the NFSv4 server.
5729  * Must be called by a thread that can sleep when no nfsds are running.
5730  */
5731 void
5732 nfsrv_throwawayallstate(NFSPROC_T *p)
5733 {
5734 	struct nfsclient *clp, *nclp;
5735 	struct nfslockfile *lfp, *nlfp;
5736 	int i;
5737 
5738 	/*
5739 	 * For each client, clean out the state and then free the structure.
5740 	 */
5741 	for (i = 0; i < nfsrv_clienthashsize; i++) {
5742 		LIST_FOREACH_SAFE(clp, &nfsclienthash[i], lc_hash, nclp) {
5743 			nfsrv_cleanclient(clp, p);
5744 			nfsrv_freedeleglist(&clp->lc_deleg);
5745 			nfsrv_freedeleglist(&clp->lc_olddeleg);
5746 			free(clp->lc_stateid, M_NFSDCLIENT);
5747 			free(clp, M_NFSDCLIENT);
5748 		}
5749 	}
5750 
5751 	/*
5752 	 * Also, free up any remaining lock file structures.
5753 	 */
5754 	for (i = 0; i < nfsrv_lockhashsize; i++) {
5755 		LIST_FOREACH_SAFE(lfp, &nfslockhash[i], lf_hash, nlfp) {
5756 			printf("nfsd unload: fnd a lock file struct\n");
5757 			nfsrv_freenfslockfile(lfp);
5758 		}
5759 	}
5760 }
5761 
5762 /*
5763  * Check the sequence# for the session and slot provided as an argument.
5764  * Also, renew the lease if the session will return NFS_OK.
5765  */
5766 int
5767 nfsrv_checksequence(struct nfsrv_descript *nd, uint32_t sequenceid,
5768     uint32_t *highest_slotidp, uint32_t *target_highest_slotidp, int cache_this,
5769     uint32_t *sflagsp, NFSPROC_T *p)
5770 {
5771 	struct nfsdsession *sep;
5772 	struct nfssessionhash *shp;
5773 	int error;
5774 	SVCXPRT *savxprt;
5775 
5776 	shp = NFSSESSIONHASH(nd->nd_sessionid);
5777 	NFSLOCKSESSION(shp);
5778 	sep = nfsrv_findsession(nd->nd_sessionid);
5779 	if (sep == NULL) {
5780 		NFSUNLOCKSESSION(shp);
5781 		return (NFSERR_BADSESSION);
5782 	}
5783 	error = nfsv4_seqsession(sequenceid, nd->nd_slotid, *highest_slotidp,
5784 	    sep->sess_slots, NULL, NFSV4_SLOTS - 1);
5785 	if (error != 0) {
5786 		NFSUNLOCKSESSION(shp);
5787 		return (error);
5788 	}
5789 	if (cache_this != 0)
5790 		nd->nd_flag |= ND_SAVEREPLY;
5791 	/* Renew the lease. */
5792 	sep->sess_clp->lc_expiry = nfsrv_leaseexpiry();
5793 	nd->nd_clientid.qval = sep->sess_clp->lc_clientid.qval;
5794 	nd->nd_flag |= ND_IMPLIEDCLID;
5795 
5796 	/*
5797 	 * If this session handles the backchannel, save the nd_xprt for this
5798 	 * RPC, since this is the one being used.
5799 	 */
5800 	if (sep->sess_clp->lc_req.nr_client != NULL &&
5801 	    (sep->sess_crflags & NFSV4CRSESS_CONNBACKCHAN) != 0) {
5802 		savxprt = sep->sess_cbsess.nfsess_xprt;
5803 		SVC_ACQUIRE(nd->nd_xprt);
5804 		nd->nd_xprt->xp_p2 =
5805 		    sep->sess_clp->lc_req.nr_client->cl_private;
5806 		nd->nd_xprt->xp_idletimeout = 0;	/* Disable timeout. */
5807 		sep->sess_cbsess.nfsess_xprt = nd->nd_xprt;
5808 		if (savxprt != NULL)
5809 			SVC_RELEASE(savxprt);
5810 	}
5811 
5812 	*sflagsp = 0;
5813 	if (sep->sess_clp->lc_req.nr_client == NULL)
5814 		*sflagsp |= NFSV4SEQ_CBPATHDOWN;
5815 	NFSUNLOCKSESSION(shp);
5816 	if (error == NFSERR_EXPIRED) {
5817 		*sflagsp |= NFSV4SEQ_EXPIREDALLSTATEREVOKED;
5818 		error = 0;
5819 	} else if (error == NFSERR_ADMINREVOKED) {
5820 		*sflagsp |= NFSV4SEQ_ADMINSTATEREVOKED;
5821 		error = 0;
5822 	}
5823 	*highest_slotidp = *target_highest_slotidp = NFSV4_SLOTS - 1;
5824 	return (0);
5825 }
5826 
5827 /*
5828  * Check/set reclaim complete for this session/clientid.
5829  */
5830 int
5831 nfsrv_checkreclaimcomplete(struct nfsrv_descript *nd)
5832 {
5833 	struct nfsdsession *sep;
5834 	struct nfssessionhash *shp;
5835 	int error = 0;
5836 
5837 	shp = NFSSESSIONHASH(nd->nd_sessionid);
5838 	NFSLOCKSTATE();
5839 	NFSLOCKSESSION(shp);
5840 	sep = nfsrv_findsession(nd->nd_sessionid);
5841 	if (sep == NULL) {
5842 		NFSUNLOCKSESSION(shp);
5843 		NFSUNLOCKSTATE();
5844 		return (NFSERR_BADSESSION);
5845 	}
5846 
5847 	/* Check to see if reclaim complete has already happened. */
5848 	if ((sep->sess_clp->lc_flags & LCL_RECLAIMCOMPLETE) != 0)
5849 		error = NFSERR_COMPLETEALREADY;
5850 	else
5851 		sep->sess_clp->lc_flags |= LCL_RECLAIMCOMPLETE;
5852 	NFSUNLOCKSESSION(shp);
5853 	NFSUNLOCKSTATE();
5854 	return (error);
5855 }
5856 
5857 /*
5858  * Cache the reply in a session slot.
5859  */
5860 void
5861 nfsrv_cache_session(uint8_t *sessionid, uint32_t slotid, int repstat,
5862    struct mbuf **m)
5863 {
5864 	struct nfsdsession *sep;
5865 	struct nfssessionhash *shp;
5866 
5867 	shp = NFSSESSIONHASH(sessionid);
5868 	NFSLOCKSESSION(shp);
5869 	sep = nfsrv_findsession(sessionid);
5870 	if (sep == NULL) {
5871 		NFSUNLOCKSESSION(shp);
5872 		printf("nfsrv_cache_session: no session\n");
5873 		m_freem(*m);
5874 		return;
5875 	}
5876 	nfsv4_seqsess_cacherep(slotid, sep->sess_slots, repstat, m);
5877 	NFSUNLOCKSESSION(shp);
5878 }
5879 
5880 /*
5881  * Search for a session that matches the sessionid.
5882  */
5883 static struct nfsdsession *
5884 nfsrv_findsession(uint8_t *sessionid)
5885 {
5886 	struct nfsdsession *sep;
5887 	struct nfssessionhash *shp;
5888 
5889 	shp = NFSSESSIONHASH(sessionid);
5890 	LIST_FOREACH(sep, &shp->list, sess_hash) {
5891 		if (!NFSBCMP(sessionid, sep->sess_sessionid, NFSX_V4SESSIONID))
5892 			break;
5893 	}
5894 	return (sep);
5895 }
5896 
5897 /*
5898  * Destroy a session.
5899  */
5900 int
5901 nfsrv_destroysession(struct nfsrv_descript *nd, uint8_t *sessionid)
5902 {
5903 	int error, samesess;
5904 
5905 	samesess = 0;
5906 	if (!NFSBCMP(sessionid, nd->nd_sessionid, NFSX_V4SESSIONID)) {
5907 		samesess = 1;
5908 		if ((nd->nd_flag & ND_LASTOP) == 0)
5909 			return (NFSERR_BADSESSION);
5910 	}
5911 	error = nfsrv_freesession(NULL, sessionid);
5912 	if (error == 0 && samesess != 0)
5913 		nd->nd_flag &= ~ND_HASSEQUENCE;
5914 	return (error);
5915 }
5916 
5917 /*
5918  * Free up a session structure.
5919  */
5920 static int
5921 nfsrv_freesession(struct nfsdsession *sep, uint8_t *sessionid)
5922 {
5923 	struct nfssessionhash *shp;
5924 	int i;
5925 
5926 	if (sep == NULL) {
5927 		shp = NFSSESSIONHASH(sessionid);
5928 		NFSLOCKSESSION(shp);
5929 		sep = nfsrv_findsession(sessionid);
5930 	} else {
5931 		shp = NFSSESSIONHASH(sep->sess_sessionid);
5932 		NFSLOCKSESSION(shp);
5933 	}
5934 	if (sep != NULL) {
5935 		NFSLOCKSTATE();
5936 		sep->sess_refcnt--;
5937 		if (sep->sess_refcnt > 0) {
5938 			NFSUNLOCKSTATE();
5939 			NFSUNLOCKSESSION(shp);
5940 			return (0);
5941 		}
5942 		LIST_REMOVE(sep, sess_hash);
5943 		LIST_REMOVE(sep, sess_list);
5944 		NFSUNLOCKSTATE();
5945 	}
5946 	NFSUNLOCKSESSION(shp);
5947 	if (sep == NULL)
5948 		return (NFSERR_BADSESSION);
5949 	for (i = 0; i < NFSV4_SLOTS; i++)
5950 		if (sep->sess_slots[i].nfssl_reply != NULL)
5951 			m_freem(sep->sess_slots[i].nfssl_reply);
5952 	if (sep->sess_cbsess.nfsess_xprt != NULL)
5953 		SVC_RELEASE(sep->sess_cbsess.nfsess_xprt);
5954 	free(sep, M_NFSDSESSION);
5955 	return (0);
5956 }
5957 
5958 /*
5959  * Free a stateid.
5960  * RFC5661 says that it should fail when there are associated opens, locks
5961  * or delegations. Since stateids represent opens, I don't see how you can
5962  * free an open stateid (it will be free'd when closed), so this function
5963  * only works for lock stateids (freeing the lock_owner) or delegations.
5964  */
5965 int
5966 nfsrv_freestateid(struct nfsrv_descript *nd, nfsv4stateid_t *stateidp,
5967     NFSPROC_T *p)
5968 {
5969 	struct nfsclient *clp;
5970 	struct nfsstate *stp;
5971 	int error;
5972 
5973 	NFSLOCKSTATE();
5974 	/*
5975 	 * Look up the stateid
5976 	 */
5977 	error = nfsrv_getclient((nfsquad_t)((u_quad_t)0), CLOPS_RENEW, &clp,
5978 	    NULL, (nfsquad_t)((u_quad_t)0), 0, nd, p);
5979 	if (error == 0) {
5980 		/* First, check for a delegation. */
5981 		LIST_FOREACH(stp, &clp->lc_deleg, ls_list) {
5982 			if (!NFSBCMP(stp->ls_stateid.other, stateidp->other,
5983 			    NFSX_STATEIDOTHER))
5984 				break;
5985 		}
5986 		if (stp != NULL) {
5987 			nfsrv_freedeleg(stp);
5988 			NFSUNLOCKSTATE();
5989 			return (error);
5990 		}
5991 	}
5992 	/* Not a delegation, try for a lock_owner. */
5993 	if (error == 0)
5994 		error = nfsrv_getstate(clp, stateidp, 0, &stp);
5995 	if (error == 0 && ((stp->ls_flags & (NFSLCK_OPEN | NFSLCK_DELEGREAD |
5996 	    NFSLCK_DELEGWRITE)) != 0 || (stp->ls_flags & NFSLCK_LOCK) == 0))
5997 		/* Not a lock_owner stateid. */
5998 		error = NFSERR_LOCKSHELD;
5999 	if (error == 0 && !LIST_EMPTY(&stp->ls_lock))
6000 		error = NFSERR_LOCKSHELD;
6001 	if (error == 0)
6002 		nfsrv_freelockowner(stp, NULL, 0, p);
6003 	NFSUNLOCKSTATE();
6004 	return (error);
6005 }
6006 
6007 /*
6008  * Generate the xdr for an NFSv4.1 CBSequence Operation.
6009  */
6010 static int
6011 nfsv4_setcbsequence(struct nfsrv_descript *nd, struct nfsclient *clp,
6012     int dont_replycache, struct nfsdsession **sepp)
6013 {
6014 	struct nfsdsession *sep;
6015 	uint32_t *tl, slotseq = 0;
6016 	int maxslot, slotpos;
6017 	uint8_t sessionid[NFSX_V4SESSIONID];
6018 	int error;
6019 
6020 	error = nfsv4_getcbsession(clp, sepp);
6021 	if (error != 0)
6022 		return (error);
6023 	sep = *sepp;
6024 	(void)nfsv4_sequencelookup(NULL, &sep->sess_cbsess, &slotpos, &maxslot,
6025 	    &slotseq, sessionid);
6026 	KASSERT(maxslot >= 0, ("nfsv4_setcbsequence neg maxslot"));
6027 
6028 	/* Build the Sequence arguments. */
6029 	NFSM_BUILD(tl, uint32_t *, NFSX_V4SESSIONID + 5 * NFSX_UNSIGNED);
6030 	bcopy(sessionid, tl, NFSX_V4SESSIONID);
6031 	tl += NFSX_V4SESSIONID / NFSX_UNSIGNED;
6032 	nd->nd_slotseq = tl;
6033 	*tl++ = txdr_unsigned(slotseq);
6034 	*tl++ = txdr_unsigned(slotpos);
6035 	*tl++ = txdr_unsigned(maxslot);
6036 	if (dont_replycache == 0)
6037 		*tl++ = newnfs_true;
6038 	else
6039 		*tl++ = newnfs_false;
6040 	*tl = 0;			/* No referring call list, for now. */
6041 	nd->nd_flag |= ND_HASSEQUENCE;
6042 	return (0);
6043 }
6044 
6045 /*
6046  * Get a session for the callback.
6047  */
6048 static int
6049 nfsv4_getcbsession(struct nfsclient *clp, struct nfsdsession **sepp)
6050 {
6051 	struct nfsdsession *sep;
6052 
6053 	NFSLOCKSTATE();
6054 	LIST_FOREACH(sep, &clp->lc_session, sess_list) {
6055 		if ((sep->sess_crflags & NFSV4CRSESS_CONNBACKCHAN) != 0)
6056 			break;
6057 	}
6058 	if (sep == NULL) {
6059 		NFSUNLOCKSTATE();
6060 		return (NFSERR_BADSESSION);
6061 	}
6062 	sep->sess_refcnt++;
6063 	*sepp = sep;
6064 	NFSUNLOCKSTATE();
6065 	return (0);
6066 }
6067 
6068 /*
6069  * Free up all backchannel xprts.  This needs to be done when the nfsd threads
6070  * exit, since those transports will all be going away.
6071  * This is only called after all the nfsd threads are done performing RPCs,
6072  * so locking shouldn't be an issue.
6073  */
6074 APPLESTATIC void
6075 nfsrv_freeallbackchannel_xprts(void)
6076 {
6077 	struct nfsdsession *sep;
6078 	struct nfsclient *clp;
6079 	SVCXPRT *xprt;
6080 	int i;
6081 
6082 	for (i = 0; i < nfsrv_clienthashsize; i++) {
6083 		LIST_FOREACH(clp, &nfsclienthash[i], lc_hash) {
6084 			LIST_FOREACH(sep, &clp->lc_session, sess_list) {
6085 				xprt = sep->sess_cbsess.nfsess_xprt;
6086 				sep->sess_cbsess.nfsess_xprt = NULL;
6087 				if (xprt != NULL)
6088 					SVC_RELEASE(xprt);
6089 			}
6090 		}
6091 	}
6092 }
6093 
6094