xref: /freebsd/sys/fs/nfsserver/nfs_nfsdstate.c (revision def7fe87e9b28032572ca6f820a260677fd0c2d5)
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 nfsstatsv1 nfsstatsv1;
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 static int	nfsrv_writedelegifpos = 0;
74 SYSCTL_INT(_vfs_nfsd, OID_AUTO, writedelegifpos, CTLFLAG_RW,
75     &nfsrv_writedelegifpos, 0,
76     "Issue a write delegation for read opens if possible");
77 
78 /*
79  * Hash lists for nfs V4.
80  */
81 struct nfsclienthashhead	*nfsclienthash;
82 struct nfslockhashhead		*nfslockhash;
83 struct nfssessionhash		*nfssessionhash;
84 #endif	/* !APPLEKEXT */
85 
86 static u_int32_t nfsrv_openpluslock = 0, nfsrv_delegatecnt = 0;
87 static time_t nfsrvboottime;
88 static int nfsrv_returnoldstateid = 0, nfsrv_clients = 0;
89 static int nfsrv_clienthighwater = NFSRV_CLIENTHIGHWATER;
90 static int nfsrv_nogsscallback = 0;
91 
92 /* local functions */
93 static void nfsrv_dumpaclient(struct nfsclient *clp,
94     struct nfsd_dumpclients *dumpp);
95 static void nfsrv_freeopenowner(struct nfsstate *stp, int cansleep,
96     NFSPROC_T *p);
97 static int nfsrv_freeopen(struct nfsstate *stp, vnode_t vp, int cansleep,
98     NFSPROC_T *p);
99 static void nfsrv_freelockowner(struct nfsstate *stp, vnode_t vp, int cansleep,
100     NFSPROC_T *p);
101 static void nfsrv_freeallnfslocks(struct nfsstate *stp, vnode_t vp,
102     int cansleep, NFSPROC_T *p);
103 static void nfsrv_freenfslock(struct nfslock *lop);
104 static void nfsrv_freenfslockfile(struct nfslockfile *lfp);
105 static void nfsrv_freedeleg(struct nfsstate *);
106 static int nfsrv_getstate(struct nfsclient *clp, nfsv4stateid_t *stateidp,
107     u_int32_t flags, struct nfsstate **stpp);
108 static void nfsrv_getowner(struct nfsstatehead *hp, struct nfsstate *new_stp,
109     struct nfsstate **stpp);
110 static int nfsrv_getlockfh(vnode_t vp, u_short flags,
111     struct nfslockfile *new_lfp, fhandle_t *nfhp, NFSPROC_T *p);
112 static int nfsrv_getlockfile(u_short flags, struct nfslockfile **new_lfpp,
113     struct nfslockfile **lfpp, fhandle_t *nfhp, int lockit);
114 static void nfsrv_insertlock(struct nfslock *new_lop,
115     struct nfslock *insert_lop, struct nfsstate *stp, struct nfslockfile *lfp);
116 static void nfsrv_updatelock(struct nfsstate *stp, struct nfslock **new_lopp,
117     struct nfslock **other_lopp, struct nfslockfile *lfp);
118 static int nfsrv_getipnumber(u_char *cp);
119 static int nfsrv_checkrestart(nfsquad_t clientid, u_int32_t flags,
120     nfsv4stateid_t *stateidp, int specialid);
121 static int nfsrv_checkgrace(struct nfsrv_descript *nd, struct nfsclient *clp,
122     u_int32_t flags);
123 static int nfsrv_docallback(struct nfsclient *clp, int procnum,
124     nfsv4stateid_t *stateidp, int trunc, fhandle_t *fhp,
125     struct nfsvattr *nap, nfsattrbit_t *attrbitp, NFSPROC_T *p);
126 static int nfsrv_cbcallargs(struct nfsrv_descript *nd, struct nfsclient *clp,
127     uint32_t callback, int op, const char *optag, struct nfsdsession **sepp);
128 static u_int32_t nfsrv_nextclientindex(void);
129 static u_int32_t nfsrv_nextstateindex(struct nfsclient *clp);
130 static void nfsrv_markstable(struct nfsclient *clp);
131 static int nfsrv_checkstable(struct nfsclient *clp);
132 static int nfsrv_clientconflict(struct nfsclient *clp, int *haslockp, struct
133     vnode *vp, NFSPROC_T *p);
134 static int nfsrv_delegconflict(struct nfsstate *stp, int *haslockp,
135     NFSPROC_T *p, vnode_t vp);
136 static int nfsrv_cleandeleg(vnode_t vp, struct nfslockfile *lfp,
137     struct nfsclient *clp, int *haslockp, NFSPROC_T *p);
138 static int nfsrv_notsamecredname(struct nfsrv_descript *nd,
139     struct nfsclient *clp);
140 static time_t nfsrv_leaseexpiry(void);
141 static void nfsrv_delaydelegtimeout(struct nfsstate *stp);
142 static int nfsrv_checkseqid(struct nfsrv_descript *nd, u_int32_t seqid,
143     struct nfsstate *stp, struct nfsrvcache *op);
144 static int nfsrv_nootherstate(struct nfsstate *stp);
145 static int nfsrv_locallock(vnode_t vp, struct nfslockfile *lfp, int flags,
146     uint64_t first, uint64_t end, struct nfslockconflict *cfp, NFSPROC_T *p);
147 static void nfsrv_localunlock(vnode_t vp, struct nfslockfile *lfp,
148     uint64_t init_first, uint64_t init_end, NFSPROC_T *p);
149 static int nfsrv_dolocal(vnode_t vp, struct nfslockfile *lfp, int flags,
150     int oldflags, uint64_t first, uint64_t end, struct nfslockconflict *cfp,
151     NFSPROC_T *p);
152 static void nfsrv_locallock_rollback(vnode_t vp, struct nfslockfile *lfp,
153     NFSPROC_T *p);
154 static void nfsrv_locallock_commit(struct nfslockfile *lfp, int flags,
155     uint64_t first, uint64_t end);
156 static void nfsrv_locklf(struct nfslockfile *lfp);
157 static void nfsrv_unlocklf(struct nfslockfile *lfp);
158 static struct nfsdsession *nfsrv_findsession(uint8_t *sessionid);
159 static int nfsrv_freesession(struct nfsdsession *sep, uint8_t *sessionid);
160 static int nfsv4_setcbsequence(struct nfsrv_descript *nd, struct nfsclient *clp,
161     int dont_replycache, struct nfsdsession **sepp);
162 static int nfsv4_getcbsession(struct nfsclient *clp, struct nfsdsession **sepp);
163 
164 /*
165  * Scan the client list for a match and either return the current one,
166  * create a new entry or return an error.
167  * If returning a non-error, the clp structure must either be linked into
168  * the client list or free'd.
169  */
170 APPLESTATIC int
171 nfsrv_setclient(struct nfsrv_descript *nd, struct nfsclient **new_clpp,
172     nfsquad_t *clientidp, nfsquad_t *confirmp, NFSPROC_T *p)
173 {
174 	struct nfsclient *clp = NULL, *new_clp = *new_clpp;
175 	int i, error = 0;
176 	struct nfsstate *stp, *tstp;
177 	struct sockaddr_in *sad, *rad;
178 	int zapit = 0, gotit, hasstate = 0, igotlock;
179 	static u_int64_t confirm_index = 0;
180 
181 	/*
182 	 * Check for state resource limit exceeded.
183 	 */
184 	if (nfsrv_openpluslock > nfsrv_v4statelimit) {
185 		error = NFSERR_RESOURCE;
186 		goto out;
187 	}
188 
189 	if (nfsrv_issuedelegs == 0 ||
190 	    ((nd->nd_flag & ND_GSS) != 0 && nfsrv_nogsscallback != 0))
191 		/*
192 		 * Don't do callbacks when delegations are disabled or
193 		 * for AUTH_GSS unless enabled via nfsrv_nogsscallback.
194 		 * If establishing a callback connection is attempted
195 		 * when a firewall is blocking the callback path, the
196 		 * server may wait too long for the connect attempt to
197 		 * succeed during the Open. Some clients, such as Linux,
198 		 * may timeout and give up on the Open before the server
199 		 * replies. Also, since AUTH_GSS callbacks are not
200 		 * yet interoperability tested, they might cause the
201 		 * server to crap out, if they get past the Init call to
202 		 * the client.
203 		 */
204 		new_clp->lc_program = 0;
205 
206 	/* Lock out other nfsd threads */
207 	NFSLOCKV4ROOTMUTEX();
208 	nfsv4_relref(&nfsv4rootfs_lock);
209 	do {
210 		igotlock = nfsv4_lock(&nfsv4rootfs_lock, 1, NULL,
211 		    NFSV4ROOTLOCKMUTEXPTR, NULL);
212 	} while (!igotlock);
213 	NFSUNLOCKV4ROOTMUTEX();
214 
215 	/*
216 	 * Search for a match in the client list.
217 	 */
218 	gotit = i = 0;
219 	while (i < nfsrv_clienthashsize && !gotit) {
220 	    LIST_FOREACH(clp, &nfsclienthash[i], lc_hash) {
221 		if (new_clp->lc_idlen == clp->lc_idlen &&
222 		    !NFSBCMP(new_clp->lc_id, clp->lc_id, clp->lc_idlen)) {
223 			gotit = 1;
224 			break;
225 		}
226 	    }
227 	    if (gotit == 0)
228 		i++;
229 	}
230 	if (!gotit ||
231 	    (clp->lc_flags & (LCL_NEEDSCONFIRM | LCL_ADMINREVOKED))) {
232 		if ((nd->nd_flag & ND_NFSV41) != 0 && confirmp->lval[1] != 0) {
233 			/*
234 			 * For NFSv4.1, if confirmp->lval[1] is non-zero, the
235 			 * client is trying to update a confirmed clientid.
236 			 */
237 			NFSLOCKV4ROOTMUTEX();
238 			nfsv4_unlock(&nfsv4rootfs_lock, 1);
239 			NFSUNLOCKV4ROOTMUTEX();
240 			confirmp->lval[1] = 0;
241 			error = NFSERR_NOENT;
242 			goto out;
243 		}
244 		/*
245 		 * Get rid of the old one.
246 		 */
247 		if (i != nfsrv_clienthashsize) {
248 			LIST_REMOVE(clp, lc_hash);
249 			nfsrv_cleanclient(clp, p);
250 			nfsrv_freedeleglist(&clp->lc_deleg);
251 			nfsrv_freedeleglist(&clp->lc_olddeleg);
252 			zapit = 1;
253 		}
254 		/*
255 		 * Add it after assigning a client id to it.
256 		 */
257 		new_clp->lc_flags |= LCL_NEEDSCONFIRM;
258 		if ((nd->nd_flag & ND_NFSV41) != 0)
259 			new_clp->lc_confirm.lval[0] = confirmp->lval[0] =
260 			    ++confirm_index;
261 		else
262 			confirmp->qval = new_clp->lc_confirm.qval =
263 			    ++confirm_index;
264 		clientidp->lval[0] = new_clp->lc_clientid.lval[0] =
265 		    (u_int32_t)nfsrvboottime;
266 		clientidp->lval[1] = new_clp->lc_clientid.lval[1] =
267 		    nfsrv_nextclientindex();
268 		new_clp->lc_stateindex = 0;
269 		new_clp->lc_statemaxindex = 0;
270 		new_clp->lc_cbref = 0;
271 		new_clp->lc_expiry = nfsrv_leaseexpiry();
272 		LIST_INIT(&new_clp->lc_open);
273 		LIST_INIT(&new_clp->lc_deleg);
274 		LIST_INIT(&new_clp->lc_olddeleg);
275 		LIST_INIT(&new_clp->lc_session);
276 		for (i = 0; i < nfsrv_statehashsize; i++)
277 			LIST_INIT(&new_clp->lc_stateid[i]);
278 		LIST_INSERT_HEAD(NFSCLIENTHASH(new_clp->lc_clientid), new_clp,
279 		    lc_hash);
280 		nfsstatsv1.srvclients++;
281 		nfsrv_openpluslock++;
282 		nfsrv_clients++;
283 		NFSLOCKV4ROOTMUTEX();
284 		nfsv4_unlock(&nfsv4rootfs_lock, 1);
285 		NFSUNLOCKV4ROOTMUTEX();
286 		if (zapit)
287 			nfsrv_zapclient(clp, p);
288 		*new_clpp = NULL;
289 		goto out;
290 	}
291 
292 	/*
293 	 * Now, handle the cases where the id is already issued.
294 	 */
295 	if (nfsrv_notsamecredname(nd, clp)) {
296 	    /*
297 	     * Check to see if there is expired state that should go away.
298 	     */
299 	    if (clp->lc_expiry < NFSD_MONOSEC &&
300 	        (!LIST_EMPTY(&clp->lc_open) || !LIST_EMPTY(&clp->lc_deleg))) {
301 		nfsrv_cleanclient(clp, p);
302 		nfsrv_freedeleglist(&clp->lc_deleg);
303 	    }
304 
305 	    /*
306 	     * If there is outstanding state, then reply NFSERR_CLIDINUSE per
307 	     * RFC3530 Sec. 8.1.2 last para.
308 	     */
309 	    if (!LIST_EMPTY(&clp->lc_deleg)) {
310 		hasstate = 1;
311 	    } else if (LIST_EMPTY(&clp->lc_open)) {
312 		hasstate = 0;
313 	    } else {
314 		hasstate = 0;
315 		/* Look for an Open on the OpenOwner */
316 		LIST_FOREACH(stp, &clp->lc_open, ls_list) {
317 		    if (!LIST_EMPTY(&stp->ls_open)) {
318 			hasstate = 1;
319 			break;
320 		    }
321 		}
322 	    }
323 	    if (hasstate) {
324 		/*
325 		 * If the uid doesn't match, return NFSERR_CLIDINUSE after
326 		 * filling out the correct ipaddr and portnum.
327 		 */
328 		sad = NFSSOCKADDR(new_clp->lc_req.nr_nam, struct sockaddr_in *);
329 		rad = NFSSOCKADDR(clp->lc_req.nr_nam, struct sockaddr_in *);
330 		sad->sin_addr.s_addr = rad->sin_addr.s_addr;
331 		sad->sin_port = rad->sin_port;
332 		NFSLOCKV4ROOTMUTEX();
333 		nfsv4_unlock(&nfsv4rootfs_lock, 1);
334 		NFSUNLOCKV4ROOTMUTEX();
335 		error = NFSERR_CLIDINUSE;
336 		goto out;
337 	    }
338 	}
339 
340 	if (NFSBCMP(new_clp->lc_verf, clp->lc_verf, NFSX_VERF)) {
341 		/*
342 		 * If the verifier has changed, the client has rebooted
343 		 * and a new client id is issued. The old state info
344 		 * can be thrown away once the SETCLIENTID_CONFIRM occurs.
345 		 */
346 		LIST_REMOVE(clp, lc_hash);
347 		new_clp->lc_flags |= LCL_NEEDSCONFIRM;
348 		if ((nd->nd_flag & ND_NFSV41) != 0)
349 			new_clp->lc_confirm.lval[0] = confirmp->lval[0] =
350 			    ++confirm_index;
351 		else
352 			confirmp->qval = new_clp->lc_confirm.qval =
353 			    ++confirm_index;
354 		clientidp->lval[0] = new_clp->lc_clientid.lval[0] =
355 		    nfsrvboottime;
356 		clientidp->lval[1] = new_clp->lc_clientid.lval[1] =
357 		    nfsrv_nextclientindex();
358 		new_clp->lc_stateindex = 0;
359 		new_clp->lc_statemaxindex = 0;
360 		new_clp->lc_cbref = 0;
361 		new_clp->lc_expiry = nfsrv_leaseexpiry();
362 
363 		/*
364 		 * Save the state until confirmed.
365 		 */
366 		LIST_NEWHEAD(&new_clp->lc_open, &clp->lc_open, ls_list);
367 		LIST_FOREACH(tstp, &new_clp->lc_open, ls_list)
368 			tstp->ls_clp = new_clp;
369 		LIST_NEWHEAD(&new_clp->lc_deleg, &clp->lc_deleg, ls_list);
370 		LIST_FOREACH(tstp, &new_clp->lc_deleg, ls_list)
371 			tstp->ls_clp = new_clp;
372 		LIST_NEWHEAD(&new_clp->lc_olddeleg, &clp->lc_olddeleg,
373 		    ls_list);
374 		LIST_FOREACH(tstp, &new_clp->lc_olddeleg, ls_list)
375 			tstp->ls_clp = new_clp;
376 		for (i = 0; i < nfsrv_statehashsize; i++) {
377 			LIST_NEWHEAD(&new_clp->lc_stateid[i],
378 			    &clp->lc_stateid[i], ls_hash);
379 			LIST_FOREACH(tstp, &new_clp->lc_stateid[i], ls_hash)
380 				tstp->ls_clp = new_clp;
381 		}
382 		LIST_INSERT_HEAD(NFSCLIENTHASH(new_clp->lc_clientid), new_clp,
383 		    lc_hash);
384 		nfsstatsv1.srvclients++;
385 		nfsrv_openpluslock++;
386 		nfsrv_clients++;
387 		NFSLOCKV4ROOTMUTEX();
388 		nfsv4_unlock(&nfsv4rootfs_lock, 1);
389 		NFSUNLOCKV4ROOTMUTEX();
390 
391 		/*
392 		 * Must wait until any outstanding callback on the old clp
393 		 * completes.
394 		 */
395 		NFSLOCKSTATE();
396 		while (clp->lc_cbref) {
397 			clp->lc_flags |= LCL_WAKEUPWANTED;
398 			(void)mtx_sleep(clp, NFSSTATEMUTEXPTR, PZERO - 1,
399 			    "nfsd clp", 10 * hz);
400 		}
401 		NFSUNLOCKSTATE();
402 		nfsrv_zapclient(clp, p);
403 		*new_clpp = NULL;
404 		goto out;
405 	}
406 
407 	/* For NFSv4.1, mark that we found a confirmed clientid. */
408 	if ((nd->nd_flag & ND_NFSV41) != 0) {
409 		clientidp->lval[0] = clp->lc_clientid.lval[0];
410 		clientidp->lval[1] = clp->lc_clientid.lval[1];
411 		confirmp->lval[0] = 0;	/* Ignored by client */
412 		confirmp->lval[1] = 1;
413 	} else {
414 		/*
415 		 * id and verifier match, so update the net address info
416 		 * and get rid of any existing callback authentication
417 		 * handle, so a new one will be acquired.
418 		 */
419 		LIST_REMOVE(clp, lc_hash);
420 		new_clp->lc_flags |= (LCL_NEEDSCONFIRM | LCL_DONTCLEAN);
421 		new_clp->lc_expiry = nfsrv_leaseexpiry();
422 		confirmp->qval = new_clp->lc_confirm.qval = ++confirm_index;
423 		clientidp->lval[0] = new_clp->lc_clientid.lval[0] =
424 		    clp->lc_clientid.lval[0];
425 		clientidp->lval[1] = new_clp->lc_clientid.lval[1] =
426 		    clp->lc_clientid.lval[1];
427 		new_clp->lc_delegtime = clp->lc_delegtime;
428 		new_clp->lc_stateindex = clp->lc_stateindex;
429 		new_clp->lc_statemaxindex = clp->lc_statemaxindex;
430 		new_clp->lc_cbref = 0;
431 		LIST_NEWHEAD(&new_clp->lc_open, &clp->lc_open, ls_list);
432 		LIST_FOREACH(tstp, &new_clp->lc_open, ls_list)
433 			tstp->ls_clp = new_clp;
434 		LIST_NEWHEAD(&new_clp->lc_deleg, &clp->lc_deleg, ls_list);
435 		LIST_FOREACH(tstp, &new_clp->lc_deleg, ls_list)
436 			tstp->ls_clp = new_clp;
437 		LIST_NEWHEAD(&new_clp->lc_olddeleg, &clp->lc_olddeleg, ls_list);
438 		LIST_FOREACH(tstp, &new_clp->lc_olddeleg, ls_list)
439 			tstp->ls_clp = new_clp;
440 		for (i = 0; i < nfsrv_statehashsize; i++) {
441 			LIST_NEWHEAD(&new_clp->lc_stateid[i],
442 			    &clp->lc_stateid[i], ls_hash);
443 			LIST_FOREACH(tstp, &new_clp->lc_stateid[i], ls_hash)
444 				tstp->ls_clp = new_clp;
445 		}
446 		LIST_INSERT_HEAD(NFSCLIENTHASH(new_clp->lc_clientid), new_clp,
447 		    lc_hash);
448 		nfsstatsv1.srvclients++;
449 		nfsrv_openpluslock++;
450 		nfsrv_clients++;
451 	}
452 	NFSLOCKV4ROOTMUTEX();
453 	nfsv4_unlock(&nfsv4rootfs_lock, 1);
454 	NFSUNLOCKV4ROOTMUTEX();
455 
456 	if ((nd->nd_flag & ND_NFSV41) == 0) {
457 		/*
458 		 * Must wait until any outstanding callback on the old clp
459 		 * completes.
460 		 */
461 		NFSLOCKSTATE();
462 		while (clp->lc_cbref) {
463 			clp->lc_flags |= LCL_WAKEUPWANTED;
464 			(void)mtx_sleep(clp, NFSSTATEMUTEXPTR, PZERO - 1,
465 			    "nfsdclp", 10 * hz);
466 		}
467 		NFSUNLOCKSTATE();
468 		nfsrv_zapclient(clp, p);
469 		*new_clpp = NULL;
470 	}
471 
472 out:
473 	NFSEXITCODE2(error, nd);
474 	return (error);
475 }
476 
477 /*
478  * Check to see if the client id exists and optionally confirm it.
479  */
480 APPLESTATIC int
481 nfsrv_getclient(nfsquad_t clientid, int opflags, struct nfsclient **clpp,
482     struct nfsdsession *nsep, nfsquad_t confirm, uint32_t cbprogram,
483     struct nfsrv_descript *nd, NFSPROC_T *p)
484 {
485 	struct nfsclient *clp;
486 	struct nfsstate *stp;
487 	int i;
488 	struct nfsclienthashhead *hp;
489 	int error = 0, igotlock, doneok;
490 	struct nfssessionhash *shp;
491 	struct nfsdsession *sep;
492 	uint64_t sessid[2];
493 	static uint64_t next_sess = 0;
494 
495 	if (clpp)
496 		*clpp = NULL;
497 	if ((nd == NULL || (nd->nd_flag & ND_NFSV41) == 0 ||
498 	    opflags != CLOPS_RENEW) && nfsrvboottime != clientid.lval[0]) {
499 		error = NFSERR_STALECLIENTID;
500 		goto out;
501 	}
502 
503 	/*
504 	 * If called with opflags == CLOPS_RENEW, the State Lock is
505 	 * already held. Otherwise, we need to get either that or,
506 	 * for the case of Confirm, lock out the nfsd threads.
507 	 */
508 	if (opflags & CLOPS_CONFIRM) {
509 		NFSLOCKV4ROOTMUTEX();
510 		nfsv4_relref(&nfsv4rootfs_lock);
511 		do {
512 			igotlock = nfsv4_lock(&nfsv4rootfs_lock, 1, NULL,
513 			    NFSV4ROOTLOCKMUTEXPTR, NULL);
514 		} while (!igotlock);
515 		/*
516 		 * Create a new sessionid here, since we need to do it where
517 		 * there is a mutex held to serialize update of next_sess.
518 		 */
519 		if ((nd->nd_flag & ND_NFSV41) != 0) {
520 			sessid[0] = ++next_sess;
521 			sessid[1] = clientid.qval;
522 		}
523 		NFSUNLOCKV4ROOTMUTEX();
524 	} else if (opflags != CLOPS_RENEW) {
525 		NFSLOCKSTATE();
526 	}
527 
528 	/* For NFSv4.1, the clp is acquired from the associated session. */
529 	if (nd != NULL && (nd->nd_flag & ND_NFSV41) != 0 &&
530 	    opflags == CLOPS_RENEW) {
531 		clp = NULL;
532 		if ((nd->nd_flag & ND_HASSEQUENCE) != 0) {
533 			shp = NFSSESSIONHASH(nd->nd_sessionid);
534 			NFSLOCKSESSION(shp);
535 			sep = nfsrv_findsession(nd->nd_sessionid);
536 			if (sep != NULL)
537 				clp = sep->sess_clp;
538 			NFSUNLOCKSESSION(shp);
539 		}
540 	} else {
541 		hp = NFSCLIENTHASH(clientid);
542 		LIST_FOREACH(clp, hp, lc_hash) {
543 			if (clp->lc_clientid.lval[1] == clientid.lval[1])
544 				break;
545 		}
546 	}
547 	if (clp == NULL) {
548 		if (opflags & CLOPS_CONFIRM)
549 			error = NFSERR_STALECLIENTID;
550 		else
551 			error = NFSERR_EXPIRED;
552 	} else if (clp->lc_flags & LCL_ADMINREVOKED) {
553 		/*
554 		 * If marked admin revoked, just return the error.
555 		 */
556 		error = NFSERR_ADMINREVOKED;
557 	}
558 	if (error) {
559 		if (opflags & CLOPS_CONFIRM) {
560 			NFSLOCKV4ROOTMUTEX();
561 			nfsv4_unlock(&nfsv4rootfs_lock, 1);
562 			NFSUNLOCKV4ROOTMUTEX();
563 		} else if (opflags != CLOPS_RENEW) {
564 			NFSUNLOCKSTATE();
565 		}
566 		goto out;
567 	}
568 
569 	/*
570 	 * Perform any operations specified by the opflags.
571 	 */
572 	if (opflags & CLOPS_CONFIRM) {
573 		if (((nd->nd_flag & ND_NFSV41) != 0 &&
574 		     clp->lc_confirm.lval[0] != confirm.lval[0]) ||
575 		    ((nd->nd_flag & ND_NFSV41) == 0 &&
576 		     clp->lc_confirm.qval != confirm.qval))
577 			error = NFSERR_STALECLIENTID;
578 		else if (nfsrv_notsamecredname(nd, clp))
579 			error = NFSERR_CLIDINUSE;
580 
581 		if (!error) {
582 		    if ((clp->lc_flags & (LCL_NEEDSCONFIRM | LCL_DONTCLEAN)) ==
583 			LCL_NEEDSCONFIRM) {
584 			/*
585 			 * Hang onto the delegations (as old delegations)
586 			 * for an Open with CLAIM_DELEGATE_PREV unless in
587 			 * grace, but get rid of the rest of the state.
588 			 */
589 			nfsrv_cleanclient(clp, p);
590 			nfsrv_freedeleglist(&clp->lc_olddeleg);
591 			if (nfsrv_checkgrace(nd, clp, 0)) {
592 			    /* In grace, so just delete delegations */
593 			    nfsrv_freedeleglist(&clp->lc_deleg);
594 			} else {
595 			    LIST_FOREACH(stp, &clp->lc_deleg, ls_list)
596 				stp->ls_flags |= NFSLCK_OLDDELEG;
597 			    clp->lc_delegtime = NFSD_MONOSEC +
598 				nfsrv_lease + NFSRV_LEASEDELTA;
599 			    LIST_NEWHEAD(&clp->lc_olddeleg, &clp->lc_deleg,
600 				ls_list);
601 			}
602 			if ((nd->nd_flag & ND_NFSV41) != 0)
603 			    clp->lc_program = cbprogram;
604 		    }
605 		    clp->lc_flags &= ~(LCL_NEEDSCONFIRM | LCL_DONTCLEAN);
606 		    if (clp->lc_program)
607 			clp->lc_flags |= LCL_NEEDSCBNULL;
608 		    /* For NFSv4.1, link the session onto the client. */
609 		    if (nsep != NULL) {
610 			/* Hold a reference on the xprt for a backchannel. */
611 			if ((nsep->sess_crflags & NFSV4CRSESS_CONNBACKCHAN)
612 			    != 0 && clp->lc_req.nr_client == NULL) {
613 			    clp->lc_req.nr_client = (struct __rpc_client *)
614 				clnt_bck_create(nd->nd_xprt->xp_socket,
615 				cbprogram, NFSV4_CBVERS);
616 			    if (clp->lc_req.nr_client != NULL) {
617 				SVC_ACQUIRE(nd->nd_xprt);
618 				nd->nd_xprt->xp_p2 =
619 				    clp->lc_req.nr_client->cl_private;
620 				/* Disable idle timeout. */
621 				nd->nd_xprt->xp_idletimeout = 0;
622 				nsep->sess_cbsess.nfsess_xprt = nd->nd_xprt;
623 			    } else
624 				nsep->sess_crflags &= ~NFSV4CRSESS_CONNBACKCHAN;
625 			}
626 			NFSBCOPY(sessid, nsep->sess_sessionid,
627 			    NFSX_V4SESSIONID);
628 			NFSBCOPY(sessid, nsep->sess_cbsess.nfsess_sessionid,
629 			    NFSX_V4SESSIONID);
630 			shp = NFSSESSIONHASH(nsep->sess_sessionid);
631 			NFSLOCKSTATE();
632 			NFSLOCKSESSION(shp);
633 			LIST_INSERT_HEAD(&shp->list, nsep, sess_hash);
634 			LIST_INSERT_HEAD(&clp->lc_session, nsep, sess_list);
635 			nsep->sess_clp = clp;
636 			NFSUNLOCKSESSION(shp);
637 			NFSUNLOCKSTATE();
638 		    }
639 		}
640 	} else if (clp->lc_flags & LCL_NEEDSCONFIRM) {
641 		error = NFSERR_EXPIRED;
642 	}
643 
644 	/*
645 	 * If called by the Renew Op, we must check the principal.
646 	 */
647 	if (!error && (opflags & CLOPS_RENEWOP)) {
648 	    if (nfsrv_notsamecredname(nd, clp)) {
649 		doneok = 0;
650 		for (i = 0; i < nfsrv_statehashsize && doneok == 0; i++) {
651 		    LIST_FOREACH(stp, &clp->lc_stateid[i], ls_hash) {
652 			if ((stp->ls_flags & NFSLCK_OPEN) &&
653 			    stp->ls_uid == nd->nd_cred->cr_uid) {
654 				doneok = 1;
655 				break;
656 			}
657 		    }
658 		}
659 		if (!doneok)
660 			error = NFSERR_ACCES;
661 	    }
662 	    if (!error && (clp->lc_flags & LCL_CBDOWN))
663 		error = NFSERR_CBPATHDOWN;
664 	}
665 	if ((!error || error == NFSERR_CBPATHDOWN) &&
666 	     (opflags & CLOPS_RENEW)) {
667 		clp->lc_expiry = nfsrv_leaseexpiry();
668 	}
669 	if (opflags & CLOPS_CONFIRM) {
670 		NFSLOCKV4ROOTMUTEX();
671 		nfsv4_unlock(&nfsv4rootfs_lock, 1);
672 		NFSUNLOCKV4ROOTMUTEX();
673 	} else if (opflags != CLOPS_RENEW) {
674 		NFSUNLOCKSTATE();
675 	}
676 	if (clpp)
677 		*clpp = clp;
678 
679 out:
680 	NFSEXITCODE2(error, nd);
681 	return (error);
682 }
683 
684 /*
685  * Perform the NFSv4.1 destroy clientid.
686  */
687 int
688 nfsrv_destroyclient(nfsquad_t clientid, NFSPROC_T *p)
689 {
690 	struct nfsclient *clp;
691 	struct nfsclienthashhead *hp;
692 	int error = 0, i, igotlock;
693 
694 	if (nfsrvboottime != clientid.lval[0]) {
695 		error = NFSERR_STALECLIENTID;
696 		goto out;
697 	}
698 
699 	/* Lock out other nfsd threads */
700 	NFSLOCKV4ROOTMUTEX();
701 	nfsv4_relref(&nfsv4rootfs_lock);
702 	do {
703 		igotlock = nfsv4_lock(&nfsv4rootfs_lock, 1, NULL,
704 		    NFSV4ROOTLOCKMUTEXPTR, NULL);
705 	} while (igotlock == 0);
706 	NFSUNLOCKV4ROOTMUTEX();
707 
708 	hp = NFSCLIENTHASH(clientid);
709 	LIST_FOREACH(clp, hp, lc_hash) {
710 		if (clp->lc_clientid.lval[1] == clientid.lval[1])
711 			break;
712 	}
713 	if (clp == NULL) {
714 		NFSLOCKV4ROOTMUTEX();
715 		nfsv4_unlock(&nfsv4rootfs_lock, 1);
716 		NFSUNLOCKV4ROOTMUTEX();
717 		/* Just return ok, since it is gone. */
718 		goto out;
719 	}
720 
721 	/* Scan for state on the clientid. */
722 	for (i = 0; i < nfsrv_statehashsize; i++)
723 		if (!LIST_EMPTY(&clp->lc_stateid[i])) {
724 			NFSLOCKV4ROOTMUTEX();
725 			nfsv4_unlock(&nfsv4rootfs_lock, 1);
726 			NFSUNLOCKV4ROOTMUTEX();
727 			error = NFSERR_CLIENTIDBUSY;
728 			goto out;
729 		}
730 	if (!LIST_EMPTY(&clp->lc_session) || !LIST_EMPTY(&clp->lc_deleg)) {
731 		NFSLOCKV4ROOTMUTEX();
732 		nfsv4_unlock(&nfsv4rootfs_lock, 1);
733 		NFSUNLOCKV4ROOTMUTEX();
734 		error = NFSERR_CLIENTIDBUSY;
735 		goto out;
736 	}
737 
738 	/* Destroy the clientid and return ok. */
739 	nfsrv_cleanclient(clp, p);
740 	nfsrv_freedeleglist(&clp->lc_deleg);
741 	nfsrv_freedeleglist(&clp->lc_olddeleg);
742 	LIST_REMOVE(clp, lc_hash);
743 	NFSLOCKV4ROOTMUTEX();
744 	nfsv4_unlock(&nfsv4rootfs_lock, 1);
745 	NFSUNLOCKV4ROOTMUTEX();
746 	nfsrv_zapclient(clp, p);
747 out:
748 	NFSEXITCODE2(error, nd);
749 	return (error);
750 }
751 
752 /*
753  * Called from the new nfssvc syscall to admin revoke a clientid.
754  * Returns 0 for success, error otherwise.
755  */
756 APPLESTATIC int
757 nfsrv_adminrevoke(struct nfsd_clid *revokep, NFSPROC_T *p)
758 {
759 	struct nfsclient *clp = NULL;
760 	int i, error = 0;
761 	int gotit, igotlock;
762 
763 	/*
764 	 * First, lock out the nfsd so that state won't change while the
765 	 * revocation record is being written to the stable storage restart
766 	 * file.
767 	 */
768 	NFSLOCKV4ROOTMUTEX();
769 	do {
770 		igotlock = nfsv4_lock(&nfsv4rootfs_lock, 1, NULL,
771 		    NFSV4ROOTLOCKMUTEXPTR, NULL);
772 	} while (!igotlock);
773 	NFSUNLOCKV4ROOTMUTEX();
774 
775 	/*
776 	 * Search for a match in the client list.
777 	 */
778 	gotit = i = 0;
779 	while (i < nfsrv_clienthashsize && !gotit) {
780 	    LIST_FOREACH(clp, &nfsclienthash[i], lc_hash) {
781 		if (revokep->nclid_idlen == clp->lc_idlen &&
782 		    !NFSBCMP(revokep->nclid_id, clp->lc_id, clp->lc_idlen)) {
783 			gotit = 1;
784 			break;
785 		}
786 	    }
787 	    i++;
788 	}
789 	if (!gotit) {
790 		NFSLOCKV4ROOTMUTEX();
791 		nfsv4_unlock(&nfsv4rootfs_lock, 0);
792 		NFSUNLOCKV4ROOTMUTEX();
793 		error = EPERM;
794 		goto out;
795 	}
796 
797 	/*
798 	 * Now, write out the revocation record
799 	 */
800 	nfsrv_writestable(clp->lc_id, clp->lc_idlen, NFSNST_REVOKE, p);
801 	nfsrv_backupstable();
802 
803 	/*
804 	 * and clear out the state, marking the clientid revoked.
805 	 */
806 	clp->lc_flags &= ~LCL_CALLBACKSON;
807 	clp->lc_flags |= LCL_ADMINREVOKED;
808 	nfsrv_cleanclient(clp, p);
809 	nfsrv_freedeleglist(&clp->lc_deleg);
810 	nfsrv_freedeleglist(&clp->lc_olddeleg);
811 	NFSLOCKV4ROOTMUTEX();
812 	nfsv4_unlock(&nfsv4rootfs_lock, 0);
813 	NFSUNLOCKV4ROOTMUTEX();
814 
815 out:
816 	NFSEXITCODE(error);
817 	return (error);
818 }
819 
820 /*
821  * Dump out stats for all clients. Called from nfssvc(2), that is used
822  * nfsstatsv1.
823  */
824 APPLESTATIC void
825 nfsrv_dumpclients(struct nfsd_dumpclients *dumpp, int maxcnt)
826 {
827 	struct nfsclient *clp;
828 	int i = 0, cnt = 0;
829 
830 	/*
831 	 * First, get a reference on the nfsv4rootfs_lock so that an
832 	 * exclusive lock cannot be acquired while dumping the clients.
833 	 */
834 	NFSLOCKV4ROOTMUTEX();
835 	nfsv4_getref(&nfsv4rootfs_lock, NULL, NFSV4ROOTLOCKMUTEXPTR, NULL);
836 	NFSUNLOCKV4ROOTMUTEX();
837 	NFSLOCKSTATE();
838 	/*
839 	 * Rattle through the client lists until done.
840 	 */
841 	while (i < nfsrv_clienthashsize && cnt < maxcnt) {
842 	    clp = LIST_FIRST(&nfsclienthash[i]);
843 	    while (clp != LIST_END(&nfsclienthash[i]) && cnt < maxcnt) {
844 		nfsrv_dumpaclient(clp, &dumpp[cnt]);
845 		cnt++;
846 		clp = LIST_NEXT(clp, lc_hash);
847 	    }
848 	    i++;
849 	}
850 	if (cnt < maxcnt)
851 	    dumpp[cnt].ndcl_clid.nclid_idlen = 0;
852 	NFSUNLOCKSTATE();
853 	NFSLOCKV4ROOTMUTEX();
854 	nfsv4_relref(&nfsv4rootfs_lock);
855 	NFSUNLOCKV4ROOTMUTEX();
856 }
857 
858 /*
859  * Dump stats for a client. Must be called with the NFSSTATELOCK and spl'd.
860  */
861 static void
862 nfsrv_dumpaclient(struct nfsclient *clp, struct nfsd_dumpclients *dumpp)
863 {
864 	struct nfsstate *stp, *openstp, *lckownstp;
865 	struct nfslock *lop;
866 	struct sockaddr *sad;
867 	struct sockaddr_in *rad;
868 	struct sockaddr_in6 *rad6;
869 
870 	dumpp->ndcl_nopenowners = dumpp->ndcl_nlockowners = 0;
871 	dumpp->ndcl_nopens = dumpp->ndcl_nlocks = 0;
872 	dumpp->ndcl_ndelegs = dumpp->ndcl_nolddelegs = 0;
873 	dumpp->ndcl_flags = clp->lc_flags;
874 	dumpp->ndcl_clid.nclid_idlen = clp->lc_idlen;
875 	NFSBCOPY(clp->lc_id, dumpp->ndcl_clid.nclid_id, clp->lc_idlen);
876 	sad = NFSSOCKADDR(clp->lc_req.nr_nam, struct sockaddr *);
877 	dumpp->ndcl_addrfam = sad->sa_family;
878 	if (sad->sa_family == AF_INET) {
879 		rad = (struct sockaddr_in *)sad;
880 		dumpp->ndcl_cbaddr.sin_addr = rad->sin_addr;
881 	} else {
882 		rad6 = (struct sockaddr_in6 *)sad;
883 		dumpp->ndcl_cbaddr.sin6_addr = rad6->sin6_addr;
884 	}
885 
886 	/*
887 	 * Now, scan the state lists and total up the opens and locks.
888 	 */
889 	LIST_FOREACH(stp, &clp->lc_open, ls_list) {
890 	    dumpp->ndcl_nopenowners++;
891 	    LIST_FOREACH(openstp, &stp->ls_open, ls_list) {
892 		dumpp->ndcl_nopens++;
893 		LIST_FOREACH(lckownstp, &openstp->ls_open, ls_list) {
894 		    dumpp->ndcl_nlockowners++;
895 		    LIST_FOREACH(lop, &lckownstp->ls_lock, lo_lckowner) {
896 			dumpp->ndcl_nlocks++;
897 		    }
898 		}
899 	    }
900 	}
901 
902 	/*
903 	 * and the delegation lists.
904 	 */
905 	LIST_FOREACH(stp, &clp->lc_deleg, ls_list) {
906 	    dumpp->ndcl_ndelegs++;
907 	}
908 	LIST_FOREACH(stp, &clp->lc_olddeleg, ls_list) {
909 	    dumpp->ndcl_nolddelegs++;
910 	}
911 }
912 
913 /*
914  * Dump out lock stats for a file.
915  */
916 APPLESTATIC void
917 nfsrv_dumplocks(vnode_t vp, struct nfsd_dumplocks *ldumpp, int maxcnt,
918     NFSPROC_T *p)
919 {
920 	struct nfsstate *stp;
921 	struct nfslock *lop;
922 	int cnt = 0;
923 	struct nfslockfile *lfp;
924 	struct sockaddr *sad;
925 	struct sockaddr_in *rad;
926 	struct sockaddr_in6 *rad6;
927 	int ret;
928 	fhandle_t nfh;
929 
930 	ret = nfsrv_getlockfh(vp, 0, NULL, &nfh, p);
931 	/*
932 	 * First, get a reference on the nfsv4rootfs_lock so that an
933 	 * exclusive lock on it cannot be acquired while dumping the locks.
934 	 */
935 	NFSLOCKV4ROOTMUTEX();
936 	nfsv4_getref(&nfsv4rootfs_lock, NULL, NFSV4ROOTLOCKMUTEXPTR, NULL);
937 	NFSUNLOCKV4ROOTMUTEX();
938 	NFSLOCKSTATE();
939 	if (!ret)
940 		ret = nfsrv_getlockfile(0, NULL, &lfp, &nfh, 0);
941 	if (ret) {
942 		ldumpp[0].ndlck_clid.nclid_idlen = 0;
943 		NFSUNLOCKSTATE();
944 		NFSLOCKV4ROOTMUTEX();
945 		nfsv4_relref(&nfsv4rootfs_lock);
946 		NFSUNLOCKV4ROOTMUTEX();
947 		return;
948 	}
949 
950 	/*
951 	 * For each open share on file, dump it out.
952 	 */
953 	stp = LIST_FIRST(&lfp->lf_open);
954 	while (stp != LIST_END(&lfp->lf_open) && cnt < maxcnt) {
955 		ldumpp[cnt].ndlck_flags = stp->ls_flags;
956 		ldumpp[cnt].ndlck_stateid.seqid = stp->ls_stateid.seqid;
957 		ldumpp[cnt].ndlck_stateid.other[0] = stp->ls_stateid.other[0];
958 		ldumpp[cnt].ndlck_stateid.other[1] = stp->ls_stateid.other[1];
959 		ldumpp[cnt].ndlck_stateid.other[2] = stp->ls_stateid.other[2];
960 		ldumpp[cnt].ndlck_owner.nclid_idlen =
961 		    stp->ls_openowner->ls_ownerlen;
962 		NFSBCOPY(stp->ls_openowner->ls_owner,
963 		    ldumpp[cnt].ndlck_owner.nclid_id,
964 		    stp->ls_openowner->ls_ownerlen);
965 		ldumpp[cnt].ndlck_clid.nclid_idlen = stp->ls_clp->lc_idlen;
966 		NFSBCOPY(stp->ls_clp->lc_id, ldumpp[cnt].ndlck_clid.nclid_id,
967 		    stp->ls_clp->lc_idlen);
968 		sad=NFSSOCKADDR(stp->ls_clp->lc_req.nr_nam, struct sockaddr *);
969 		ldumpp[cnt].ndlck_addrfam = sad->sa_family;
970 		if (sad->sa_family == AF_INET) {
971 			rad = (struct sockaddr_in *)sad;
972 			ldumpp[cnt].ndlck_cbaddr.sin_addr = rad->sin_addr;
973 		} else {
974 			rad6 = (struct sockaddr_in6 *)sad;
975 			ldumpp[cnt].ndlck_cbaddr.sin6_addr = rad6->sin6_addr;
976 		}
977 		stp = LIST_NEXT(stp, ls_file);
978 		cnt++;
979 	}
980 
981 	/*
982 	 * and all locks.
983 	 */
984 	lop = LIST_FIRST(&lfp->lf_lock);
985 	while (lop != LIST_END(&lfp->lf_lock) && cnt < maxcnt) {
986 		stp = lop->lo_stp;
987 		ldumpp[cnt].ndlck_flags = lop->lo_flags;
988 		ldumpp[cnt].ndlck_first = lop->lo_first;
989 		ldumpp[cnt].ndlck_end = lop->lo_end;
990 		ldumpp[cnt].ndlck_stateid.seqid = stp->ls_stateid.seqid;
991 		ldumpp[cnt].ndlck_stateid.other[0] = stp->ls_stateid.other[0];
992 		ldumpp[cnt].ndlck_stateid.other[1] = stp->ls_stateid.other[1];
993 		ldumpp[cnt].ndlck_stateid.other[2] = stp->ls_stateid.other[2];
994 		ldumpp[cnt].ndlck_owner.nclid_idlen = stp->ls_ownerlen;
995 		NFSBCOPY(stp->ls_owner, ldumpp[cnt].ndlck_owner.nclid_id,
996 		    stp->ls_ownerlen);
997 		ldumpp[cnt].ndlck_clid.nclid_idlen = stp->ls_clp->lc_idlen;
998 		NFSBCOPY(stp->ls_clp->lc_id, ldumpp[cnt].ndlck_clid.nclid_id,
999 		    stp->ls_clp->lc_idlen);
1000 		sad=NFSSOCKADDR(stp->ls_clp->lc_req.nr_nam, struct sockaddr *);
1001 		ldumpp[cnt].ndlck_addrfam = sad->sa_family;
1002 		if (sad->sa_family == AF_INET) {
1003 			rad = (struct sockaddr_in *)sad;
1004 			ldumpp[cnt].ndlck_cbaddr.sin_addr = rad->sin_addr;
1005 		} else {
1006 			rad6 = (struct sockaddr_in6 *)sad;
1007 			ldumpp[cnt].ndlck_cbaddr.sin6_addr = rad6->sin6_addr;
1008 		}
1009 		lop = LIST_NEXT(lop, lo_lckfile);
1010 		cnt++;
1011 	}
1012 
1013 	/*
1014 	 * and the delegations.
1015 	 */
1016 	stp = LIST_FIRST(&lfp->lf_deleg);
1017 	while (stp != LIST_END(&lfp->lf_deleg) && cnt < maxcnt) {
1018 		ldumpp[cnt].ndlck_flags = stp->ls_flags;
1019 		ldumpp[cnt].ndlck_stateid.seqid = stp->ls_stateid.seqid;
1020 		ldumpp[cnt].ndlck_stateid.other[0] = stp->ls_stateid.other[0];
1021 		ldumpp[cnt].ndlck_stateid.other[1] = stp->ls_stateid.other[1];
1022 		ldumpp[cnt].ndlck_stateid.other[2] = stp->ls_stateid.other[2];
1023 		ldumpp[cnt].ndlck_owner.nclid_idlen = 0;
1024 		ldumpp[cnt].ndlck_clid.nclid_idlen = stp->ls_clp->lc_idlen;
1025 		NFSBCOPY(stp->ls_clp->lc_id, ldumpp[cnt].ndlck_clid.nclid_id,
1026 		    stp->ls_clp->lc_idlen);
1027 		sad=NFSSOCKADDR(stp->ls_clp->lc_req.nr_nam, struct sockaddr *);
1028 		ldumpp[cnt].ndlck_addrfam = sad->sa_family;
1029 		if (sad->sa_family == AF_INET) {
1030 			rad = (struct sockaddr_in *)sad;
1031 			ldumpp[cnt].ndlck_cbaddr.sin_addr = rad->sin_addr;
1032 		} else {
1033 			rad6 = (struct sockaddr_in6 *)sad;
1034 			ldumpp[cnt].ndlck_cbaddr.sin6_addr = rad6->sin6_addr;
1035 		}
1036 		stp = LIST_NEXT(stp, ls_file);
1037 		cnt++;
1038 	}
1039 
1040 	/*
1041 	 * If list isn't full, mark end of list by setting the client name
1042 	 * to zero length.
1043 	 */
1044 	if (cnt < maxcnt)
1045 		ldumpp[cnt].ndlck_clid.nclid_idlen = 0;
1046 	NFSUNLOCKSTATE();
1047 	NFSLOCKV4ROOTMUTEX();
1048 	nfsv4_relref(&nfsv4rootfs_lock);
1049 	NFSUNLOCKV4ROOTMUTEX();
1050 }
1051 
1052 /*
1053  * Server timer routine. It can scan any linked list, so long
1054  * as it holds the spin/mutex lock and there is no exclusive lock on
1055  * nfsv4rootfs_lock.
1056  * (For OpenBSD, a kthread is ok. For FreeBSD, I think it is ok
1057  *  to do this from a callout, since the spin locks work. For
1058  *  Darwin, I'm not sure what will work correctly yet.)
1059  * Should be called once per second.
1060  */
1061 APPLESTATIC void
1062 nfsrv_servertimer(void)
1063 {
1064 	struct nfsclient *clp, *nclp;
1065 	struct nfsstate *stp, *nstp;
1066 	int got_ref, i;
1067 
1068 	/*
1069 	 * Make sure nfsboottime is set. This is used by V3 as well
1070 	 * as V4. Note that nfsboottime is not nfsrvboottime, which is
1071 	 * only used by the V4 server for leases.
1072 	 */
1073 	if (nfsboottime.tv_sec == 0)
1074 		NFSSETBOOTTIME(nfsboottime);
1075 
1076 	/*
1077 	 * If server hasn't started yet, just return.
1078 	 */
1079 	NFSLOCKSTATE();
1080 	if (nfsrv_stablefirst.nsf_eograce == 0) {
1081 		NFSUNLOCKSTATE();
1082 		return;
1083 	}
1084 	if (!(nfsrv_stablefirst.nsf_flags & NFSNSF_UPDATEDONE)) {
1085 		if (!(nfsrv_stablefirst.nsf_flags & NFSNSF_GRACEOVER) &&
1086 		    NFSD_MONOSEC > nfsrv_stablefirst.nsf_eograce)
1087 			nfsrv_stablefirst.nsf_flags |=
1088 			    (NFSNSF_GRACEOVER | NFSNSF_NEEDLOCK);
1089 		NFSUNLOCKSTATE();
1090 		return;
1091 	}
1092 
1093 	/*
1094 	 * Try and get a reference count on the nfsv4rootfs_lock so that
1095 	 * no nfsd thread can acquire an exclusive lock on it before this
1096 	 * call is done. If it is already exclusively locked, just return.
1097 	 */
1098 	NFSLOCKV4ROOTMUTEX();
1099 	got_ref = nfsv4_getref_nonblock(&nfsv4rootfs_lock);
1100 	NFSUNLOCKV4ROOTMUTEX();
1101 	if (got_ref == 0) {
1102 		NFSUNLOCKSTATE();
1103 		return;
1104 	}
1105 
1106 	/*
1107 	 * For each client...
1108 	 */
1109 	for (i = 0; i < nfsrv_clienthashsize; i++) {
1110 	    clp = LIST_FIRST(&nfsclienthash[i]);
1111 	    while (clp != LIST_END(&nfsclienthash[i])) {
1112 		nclp = LIST_NEXT(clp, lc_hash);
1113 		if (!(clp->lc_flags & LCL_EXPIREIT)) {
1114 		    if (((clp->lc_expiry + NFSRV_STALELEASE) < NFSD_MONOSEC
1115 			 && ((LIST_EMPTY(&clp->lc_deleg)
1116 			      && LIST_EMPTY(&clp->lc_open)) ||
1117 			     nfsrv_clients > nfsrv_clienthighwater)) ||
1118 			(clp->lc_expiry + NFSRV_MOULDYLEASE) < NFSD_MONOSEC ||
1119 			(clp->lc_expiry < NFSD_MONOSEC &&
1120 			 (nfsrv_openpluslock * 10 / 9) > nfsrv_v4statelimit)) {
1121 			/*
1122 			 * Lease has expired several nfsrv_lease times ago:
1123 			 * PLUS
1124 			 *    - no state is associated with it
1125 			 *    OR
1126 			 *    - above high water mark for number of clients
1127 			 *      (nfsrv_clienthighwater should be large enough
1128 			 *       that this only occurs when clients fail to
1129 			 *       use the same nfs_client_id4.id. Maybe somewhat
1130 			 *       higher that the maximum number of clients that
1131 			 *       will mount this server?)
1132 			 * OR
1133 			 * Lease has expired a very long time ago
1134 			 * OR
1135 			 * Lease has expired PLUS the number of opens + locks
1136 			 * has exceeded 90% of capacity
1137 			 *
1138 			 * --> Mark for expiry. The actual expiry will be done
1139 			 *     by an nfsd sometime soon.
1140 			 */
1141 			clp->lc_flags |= LCL_EXPIREIT;
1142 			nfsrv_stablefirst.nsf_flags |=
1143 			    (NFSNSF_NEEDLOCK | NFSNSF_EXPIREDCLIENT);
1144 		    } else {
1145 			/*
1146 			 * If there are no opens, increment no open tick cnt
1147 			 * If time exceeds NFSNOOPEN, mark it to be thrown away
1148 			 * otherwise, if there is an open, reset no open time
1149 			 * Hopefully, this will avoid excessive re-creation
1150 			 * of open owners and subsequent open confirms.
1151 			 */
1152 			stp = LIST_FIRST(&clp->lc_open);
1153 			while (stp != LIST_END(&clp->lc_open)) {
1154 				nstp = LIST_NEXT(stp, ls_list);
1155 				if (LIST_EMPTY(&stp->ls_open)) {
1156 					stp->ls_noopens++;
1157 					if (stp->ls_noopens > NFSNOOPEN ||
1158 					    (nfsrv_openpluslock * 2) >
1159 					    nfsrv_v4statelimit)
1160 						nfsrv_stablefirst.nsf_flags |=
1161 							NFSNSF_NOOPENS;
1162 				} else {
1163 					stp->ls_noopens = 0;
1164 				}
1165 				stp = nstp;
1166 			}
1167 		    }
1168 		}
1169 		clp = nclp;
1170 	    }
1171 	}
1172 	NFSUNLOCKSTATE();
1173 	NFSLOCKV4ROOTMUTEX();
1174 	nfsv4_relref(&nfsv4rootfs_lock);
1175 	NFSUNLOCKV4ROOTMUTEX();
1176 }
1177 
1178 /*
1179  * The following set of functions free up the various data structures.
1180  */
1181 /*
1182  * Clear out all open/lock state related to this nfsclient.
1183  * Caller must hold an exclusive lock on nfsv4rootfs_lock, so that
1184  * there are no other active nfsd threads.
1185  */
1186 APPLESTATIC void
1187 nfsrv_cleanclient(struct nfsclient *clp, NFSPROC_T *p)
1188 {
1189 	struct nfsstate *stp, *nstp;
1190 	struct nfsdsession *sep, *nsep;
1191 
1192 	LIST_FOREACH_SAFE(stp, &clp->lc_open, ls_list, nstp)
1193 		nfsrv_freeopenowner(stp, 1, p);
1194 	if ((clp->lc_flags & LCL_ADMINREVOKED) == 0)
1195 		LIST_FOREACH_SAFE(sep, &clp->lc_session, sess_list, nsep)
1196 			(void)nfsrv_freesession(sep, NULL);
1197 }
1198 
1199 /*
1200  * Free a client that has been cleaned. It should also already have been
1201  * removed from the lists.
1202  * (Just to be safe w.r.t. newnfs_disconnect(), call this function when
1203  *  softclock interrupts are enabled.)
1204  */
1205 APPLESTATIC void
1206 nfsrv_zapclient(struct nfsclient *clp, NFSPROC_T *p)
1207 {
1208 
1209 #ifdef notyet
1210 	if ((clp->lc_flags & (LCL_GSS | LCL_CALLBACKSON)) ==
1211 	     (LCL_GSS | LCL_CALLBACKSON) &&
1212 	    (clp->lc_hand.nfsh_flag & NFSG_COMPLETE) &&
1213 	    clp->lc_handlelen > 0) {
1214 		clp->lc_hand.nfsh_flag &= ~NFSG_COMPLETE;
1215 		clp->lc_hand.nfsh_flag |= NFSG_DESTROYED;
1216 		(void) nfsrv_docallback(clp, NFSV4PROC_CBNULL,
1217 			NULL, 0, NULL, NULL, NULL, p);
1218 	}
1219 #endif
1220 	newnfs_disconnect(&clp->lc_req);
1221 	NFSSOCKADDRFREE(clp->lc_req.nr_nam);
1222 	NFSFREEMUTEX(&clp->lc_req.nr_mtx);
1223 	free(clp->lc_stateid, M_NFSDCLIENT);
1224 	free(clp, M_NFSDCLIENT);
1225 	NFSLOCKSTATE();
1226 	nfsstatsv1.srvclients--;
1227 	nfsrv_openpluslock--;
1228 	nfsrv_clients--;
1229 	NFSUNLOCKSTATE();
1230 }
1231 
1232 /*
1233  * Free a list of delegation state structures.
1234  * (This function will also free all nfslockfile structures that no
1235  *  longer have associated state.)
1236  */
1237 APPLESTATIC void
1238 nfsrv_freedeleglist(struct nfsstatehead *sthp)
1239 {
1240 	struct nfsstate *stp, *nstp;
1241 
1242 	LIST_FOREACH_SAFE(stp, sthp, ls_list, nstp) {
1243 		nfsrv_freedeleg(stp);
1244 	}
1245 	LIST_INIT(sthp);
1246 }
1247 
1248 /*
1249  * Free up a delegation.
1250  */
1251 static void
1252 nfsrv_freedeleg(struct nfsstate *stp)
1253 {
1254 	struct nfslockfile *lfp;
1255 
1256 	LIST_REMOVE(stp, ls_hash);
1257 	LIST_REMOVE(stp, ls_list);
1258 	LIST_REMOVE(stp, ls_file);
1259 	lfp = stp->ls_lfp;
1260 	if (LIST_EMPTY(&lfp->lf_open) &&
1261 	    LIST_EMPTY(&lfp->lf_lock) && LIST_EMPTY(&lfp->lf_deleg) &&
1262 	    LIST_EMPTY(&lfp->lf_locallock) && LIST_EMPTY(&lfp->lf_rollback) &&
1263 	    lfp->lf_usecount == 0 &&
1264 	    nfsv4_testlock(&lfp->lf_locallock_lck) == 0)
1265 		nfsrv_freenfslockfile(lfp);
1266 	FREE((caddr_t)stp, M_NFSDSTATE);
1267 	nfsstatsv1.srvdelegates--;
1268 	nfsrv_openpluslock--;
1269 	nfsrv_delegatecnt--;
1270 }
1271 
1272 /*
1273  * This function frees an open owner and all associated opens.
1274  */
1275 static void
1276 nfsrv_freeopenowner(struct nfsstate *stp, int cansleep, NFSPROC_T *p)
1277 {
1278 	struct nfsstate *nstp, *tstp;
1279 
1280 	LIST_REMOVE(stp, ls_list);
1281 	/*
1282 	 * Now, free all associated opens.
1283 	 */
1284 	nstp = LIST_FIRST(&stp->ls_open);
1285 	while (nstp != LIST_END(&stp->ls_open)) {
1286 		tstp = nstp;
1287 		nstp = LIST_NEXT(nstp, ls_list);
1288 		(void) nfsrv_freeopen(tstp, NULL, cansleep, p);
1289 	}
1290 	if (stp->ls_op)
1291 		nfsrvd_derefcache(stp->ls_op);
1292 	FREE((caddr_t)stp, M_NFSDSTATE);
1293 	nfsstatsv1.srvopenowners--;
1294 	nfsrv_openpluslock--;
1295 }
1296 
1297 /*
1298  * This function frees an open (nfsstate open structure) with all associated
1299  * lock_owners and locks. It also frees the nfslockfile structure iff there
1300  * are no other opens on the file.
1301  * Returns 1 if it free'd the nfslockfile, 0 otherwise.
1302  */
1303 static int
1304 nfsrv_freeopen(struct nfsstate *stp, vnode_t vp, int cansleep, NFSPROC_T *p)
1305 {
1306 	struct nfsstate *nstp, *tstp;
1307 	struct nfslockfile *lfp;
1308 	int ret;
1309 
1310 	LIST_REMOVE(stp, ls_hash);
1311 	LIST_REMOVE(stp, ls_list);
1312 	LIST_REMOVE(stp, ls_file);
1313 
1314 	lfp = stp->ls_lfp;
1315 	/*
1316 	 * Now, free all lockowners associated with this open.
1317 	 */
1318 	LIST_FOREACH_SAFE(tstp, &stp->ls_open, ls_list, nstp)
1319 		nfsrv_freelockowner(tstp, vp, cansleep, p);
1320 
1321 	/*
1322 	 * The nfslockfile is freed here if there are no locks
1323 	 * associated with the open.
1324 	 * If there are locks associated with the open, the
1325 	 * nfslockfile structure can be freed via nfsrv_freelockowner().
1326 	 * Acquire the state mutex to avoid races with calls to
1327 	 * nfsrv_getlockfile().
1328 	 */
1329 	if (cansleep != 0)
1330 		NFSLOCKSTATE();
1331 	if (lfp != NULL && LIST_EMPTY(&lfp->lf_open) &&
1332 	    LIST_EMPTY(&lfp->lf_deleg) && LIST_EMPTY(&lfp->lf_lock) &&
1333 	    LIST_EMPTY(&lfp->lf_locallock) && LIST_EMPTY(&lfp->lf_rollback) &&
1334 	    lfp->lf_usecount == 0 &&
1335 	    (cansleep != 0 || nfsv4_testlock(&lfp->lf_locallock_lck) == 0)) {
1336 		nfsrv_freenfslockfile(lfp);
1337 		ret = 1;
1338 	} else
1339 		ret = 0;
1340 	if (cansleep != 0)
1341 		NFSUNLOCKSTATE();
1342 	FREE((caddr_t)stp, M_NFSDSTATE);
1343 	nfsstatsv1.srvopens--;
1344 	nfsrv_openpluslock--;
1345 	return (ret);
1346 }
1347 
1348 /*
1349  * Frees a lockowner and all associated locks.
1350  */
1351 static void
1352 nfsrv_freelockowner(struct nfsstate *stp, vnode_t vp, int cansleep,
1353     NFSPROC_T *p)
1354 {
1355 
1356 	LIST_REMOVE(stp, ls_hash);
1357 	LIST_REMOVE(stp, ls_list);
1358 	nfsrv_freeallnfslocks(stp, vp, cansleep, p);
1359 	if (stp->ls_op)
1360 		nfsrvd_derefcache(stp->ls_op);
1361 	FREE((caddr_t)stp, M_NFSDSTATE);
1362 	nfsstatsv1.srvlockowners--;
1363 	nfsrv_openpluslock--;
1364 }
1365 
1366 /*
1367  * Free all the nfs locks on a lockowner.
1368  */
1369 static void
1370 nfsrv_freeallnfslocks(struct nfsstate *stp, vnode_t vp, int cansleep,
1371     NFSPROC_T *p)
1372 {
1373 	struct nfslock *lop, *nlop;
1374 	struct nfsrollback *rlp, *nrlp;
1375 	struct nfslockfile *lfp = NULL;
1376 	int gottvp = 0;
1377 	vnode_t tvp = NULL;
1378 	uint64_t first, end;
1379 
1380 	if (vp != NULL)
1381 		ASSERT_VOP_UNLOCKED(vp, "nfsrv_freeallnfslocks: vnode locked");
1382 	lop = LIST_FIRST(&stp->ls_lock);
1383 	while (lop != LIST_END(&stp->ls_lock)) {
1384 		nlop = LIST_NEXT(lop, lo_lckowner);
1385 		/*
1386 		 * Since all locks should be for the same file, lfp should
1387 		 * not change.
1388 		 */
1389 		if (lfp == NULL)
1390 			lfp = lop->lo_lfp;
1391 		else if (lfp != lop->lo_lfp)
1392 			panic("allnfslocks");
1393 		/*
1394 		 * If vp is NULL and cansleep != 0, a vnode must be acquired
1395 		 * from the file handle. This only occurs when called from
1396 		 * nfsrv_cleanclient().
1397 		 */
1398 		if (gottvp == 0) {
1399 			if (nfsrv_dolocallocks == 0)
1400 				tvp = NULL;
1401 			else if (vp == NULL && cansleep != 0) {
1402 				tvp = nfsvno_getvp(&lfp->lf_fh);
1403 				NFSVOPUNLOCK(tvp, 0);
1404 			} else
1405 				tvp = vp;
1406 			gottvp = 1;
1407 		}
1408 
1409 		if (tvp != NULL) {
1410 			if (cansleep == 0)
1411 				panic("allnfs2");
1412 			first = lop->lo_first;
1413 			end = lop->lo_end;
1414 			nfsrv_freenfslock(lop);
1415 			nfsrv_localunlock(tvp, lfp, first, end, p);
1416 			LIST_FOREACH_SAFE(rlp, &lfp->lf_rollback, rlck_list,
1417 			    nrlp)
1418 				free(rlp, M_NFSDROLLBACK);
1419 			LIST_INIT(&lfp->lf_rollback);
1420 		} else
1421 			nfsrv_freenfslock(lop);
1422 		lop = nlop;
1423 	}
1424 	if (vp == NULL && tvp != NULL)
1425 		vrele(tvp);
1426 }
1427 
1428 /*
1429  * Free an nfslock structure.
1430  */
1431 static void
1432 nfsrv_freenfslock(struct nfslock *lop)
1433 {
1434 
1435 	if (lop->lo_lckfile.le_prev != NULL) {
1436 		LIST_REMOVE(lop, lo_lckfile);
1437 		nfsstatsv1.srvlocks--;
1438 		nfsrv_openpluslock--;
1439 	}
1440 	LIST_REMOVE(lop, lo_lckowner);
1441 	FREE((caddr_t)lop, M_NFSDLOCK);
1442 }
1443 
1444 /*
1445  * This function frees an nfslockfile structure.
1446  */
1447 static void
1448 nfsrv_freenfslockfile(struct nfslockfile *lfp)
1449 {
1450 
1451 	LIST_REMOVE(lfp, lf_hash);
1452 	FREE((caddr_t)lfp, M_NFSDLOCKFILE);
1453 }
1454 
1455 /*
1456  * This function looks up an nfsstate structure via stateid.
1457  */
1458 static int
1459 nfsrv_getstate(struct nfsclient *clp, nfsv4stateid_t *stateidp, __unused u_int32_t flags,
1460     struct nfsstate **stpp)
1461 {
1462 	struct nfsstate *stp;
1463 	struct nfsstatehead *hp;
1464 	int error = 0;
1465 
1466 	*stpp = NULL;
1467 	hp = NFSSTATEHASH(clp, *stateidp);
1468 	LIST_FOREACH(stp, hp, ls_hash) {
1469 		if (!NFSBCMP(stp->ls_stateid.other, stateidp->other,
1470 			NFSX_STATEIDOTHER))
1471 			break;
1472 	}
1473 
1474 	/*
1475 	 * If no state id in list, return NFSERR_BADSTATEID.
1476 	 */
1477 	if (stp == LIST_END(hp)) {
1478 		error = NFSERR_BADSTATEID;
1479 		goto out;
1480 	}
1481 	*stpp = stp;
1482 
1483 out:
1484 	NFSEXITCODE(error);
1485 	return (error);
1486 }
1487 
1488 /*
1489  * This function gets an nfsstate structure via owner string.
1490  */
1491 static void
1492 nfsrv_getowner(struct nfsstatehead *hp, struct nfsstate *new_stp,
1493     struct nfsstate **stpp)
1494 {
1495 	struct nfsstate *stp;
1496 
1497 	*stpp = NULL;
1498 	LIST_FOREACH(stp, hp, ls_list) {
1499 		if (new_stp->ls_ownerlen == stp->ls_ownerlen &&
1500 		  !NFSBCMP(new_stp->ls_owner,stp->ls_owner,stp->ls_ownerlen)) {
1501 			*stpp = stp;
1502 			return;
1503 		}
1504 	}
1505 }
1506 
1507 /*
1508  * Lock control function called to update lock status.
1509  * Returns 0 upon success, -1 if there is no lock and the flags indicate
1510  * that one isn't to be created and an NFSERR_xxx for other errors.
1511  * The structures new_stp and new_lop are passed in as pointers that should
1512  * be set to NULL if the structure is used and shouldn't be free'd.
1513  * For the NFSLCK_TEST and NFSLCK_CHECK cases, the structures are
1514  * never used and can safely be allocated on the stack. For all other
1515  * cases, *new_stpp and *new_lopp should be malloc'd before the call,
1516  * in case they are used.
1517  */
1518 APPLESTATIC int
1519 nfsrv_lockctrl(vnode_t vp, struct nfsstate **new_stpp,
1520     struct nfslock **new_lopp, struct nfslockconflict *cfp,
1521     nfsquad_t clientid, nfsv4stateid_t *stateidp,
1522     __unused struct nfsexstuff *exp,
1523     struct nfsrv_descript *nd, NFSPROC_T *p)
1524 {
1525 	struct nfslock *lop;
1526 	struct nfsstate *new_stp = *new_stpp;
1527 	struct nfslock *new_lop = *new_lopp;
1528 	struct nfsstate *tstp, *mystp, *nstp;
1529 	int specialid = 0;
1530 	struct nfslockfile *lfp;
1531 	struct nfslock *other_lop = NULL;
1532 	struct nfsstate *stp, *lckstp = NULL;
1533 	struct nfsclient *clp = NULL;
1534 	u_int32_t bits;
1535 	int error = 0, haslock = 0, ret, reterr;
1536 	int getlckret, delegation = 0, filestruct_locked, vnode_unlocked = 0;
1537 	fhandle_t nfh;
1538 	uint64_t first, end;
1539 	uint32_t lock_flags;
1540 
1541 	if (new_stp->ls_flags & (NFSLCK_CHECK | NFSLCK_SETATTR)) {
1542 		/*
1543 		 * Note the special cases of "all 1s" or "all 0s" stateids and
1544 		 * let reads with all 1s go ahead.
1545 		 */
1546 		if (new_stp->ls_stateid.seqid == 0x0 &&
1547 		    new_stp->ls_stateid.other[0] == 0x0 &&
1548 		    new_stp->ls_stateid.other[1] == 0x0 &&
1549 		    new_stp->ls_stateid.other[2] == 0x0)
1550 			specialid = 1;
1551 		else if (new_stp->ls_stateid.seqid == 0xffffffff &&
1552 		    new_stp->ls_stateid.other[0] == 0xffffffff &&
1553 		    new_stp->ls_stateid.other[1] == 0xffffffff &&
1554 		    new_stp->ls_stateid.other[2] == 0xffffffff)
1555 			specialid = 2;
1556 	}
1557 
1558 	/*
1559 	 * Check for restart conditions (client and server).
1560 	 */
1561 	error = nfsrv_checkrestart(clientid, new_stp->ls_flags,
1562 	    &new_stp->ls_stateid, specialid);
1563 	if (error)
1564 		goto out;
1565 
1566 	/*
1567 	 * Check for state resource limit exceeded.
1568 	 */
1569 	if ((new_stp->ls_flags & NFSLCK_LOCK) &&
1570 	    nfsrv_openpluslock > nfsrv_v4statelimit) {
1571 		error = NFSERR_RESOURCE;
1572 		goto out;
1573 	}
1574 
1575 	/*
1576 	 * For the lock case, get another nfslock structure,
1577 	 * just in case we need it.
1578 	 * Malloc now, before we start sifting through the linked lists,
1579 	 * in case we have to wait for memory.
1580 	 */
1581 tryagain:
1582 	if (new_stp->ls_flags & NFSLCK_LOCK)
1583 		MALLOC(other_lop, struct nfslock *, sizeof (struct nfslock),
1584 		    M_NFSDLOCK, M_WAITOK);
1585 	filestruct_locked = 0;
1586 	reterr = 0;
1587 	lfp = NULL;
1588 
1589 	/*
1590 	 * Get the lockfile structure for CFH now, so we can do a sanity
1591 	 * check against the stateid, before incrementing the seqid#, since
1592 	 * we want to return NFSERR_BADSTATEID on failure and the seqid#
1593 	 * shouldn't be incremented for this case.
1594 	 * If nfsrv_getlockfile() returns -1, it means "not found", which
1595 	 * will be handled later.
1596 	 * If we are doing Lock/LockU and local locking is enabled, sleep
1597 	 * lock the nfslockfile structure.
1598 	 */
1599 	getlckret = nfsrv_getlockfh(vp, new_stp->ls_flags, NULL, &nfh, p);
1600 	NFSLOCKSTATE();
1601 	if (getlckret == 0) {
1602 		if ((new_stp->ls_flags & (NFSLCK_LOCK | NFSLCK_UNLOCK)) != 0 &&
1603 		    nfsrv_dolocallocks != 0 && nd->nd_repstat == 0) {
1604 			getlckret = nfsrv_getlockfile(new_stp->ls_flags, NULL,
1605 			    &lfp, &nfh, 1);
1606 			if (getlckret == 0)
1607 				filestruct_locked = 1;
1608 		} else
1609 			getlckret = nfsrv_getlockfile(new_stp->ls_flags, NULL,
1610 			    &lfp, &nfh, 0);
1611 	}
1612 	if (getlckret != 0 && getlckret != -1)
1613 		reterr = getlckret;
1614 
1615 	if (filestruct_locked != 0) {
1616 		LIST_INIT(&lfp->lf_rollback);
1617 		if ((new_stp->ls_flags & NFSLCK_LOCK)) {
1618 			/*
1619 			 * For local locking, do the advisory locking now, so
1620 			 * that any conflict can be detected. A failure later
1621 			 * can be rolled back locally. If an error is returned,
1622 			 * struct nfslockfile has been unlocked and any local
1623 			 * locking rolled back.
1624 			 */
1625 			NFSUNLOCKSTATE();
1626 			if (vnode_unlocked == 0) {
1627 				ASSERT_VOP_ELOCKED(vp, "nfsrv_lockctrl1");
1628 				vnode_unlocked = 1;
1629 				NFSVOPUNLOCK(vp, 0);
1630 			}
1631 			reterr = nfsrv_locallock(vp, lfp,
1632 			    (new_lop->lo_flags & (NFSLCK_READ | NFSLCK_WRITE)),
1633 			    new_lop->lo_first, new_lop->lo_end, cfp, p);
1634 			NFSLOCKSTATE();
1635 		}
1636 	}
1637 
1638 	if (specialid == 0) {
1639 	    if (new_stp->ls_flags & NFSLCK_TEST) {
1640 		/*
1641 		 * RFC 3530 does not list LockT as an op that renews a
1642 		 * lease, but the consensus seems to be that it is ok
1643 		 * for a server to do so.
1644 		 */
1645 		error = nfsrv_getclient(clientid, CLOPS_RENEW, &clp, NULL,
1646 		    (nfsquad_t)((u_quad_t)0), 0, nd, p);
1647 
1648 		/*
1649 		 * Since NFSERR_EXPIRED, NFSERR_ADMINREVOKED are not valid
1650 		 * error returns for LockT, just go ahead and test for a lock,
1651 		 * since there are no locks for this client, but other locks
1652 		 * can conflict. (ie. same client will always be false)
1653 		 */
1654 		if (error == NFSERR_EXPIRED || error == NFSERR_ADMINREVOKED)
1655 		    error = 0;
1656 		lckstp = new_stp;
1657 	    } else {
1658 	      error = nfsrv_getclient(clientid, CLOPS_RENEW, &clp, NULL,
1659 		(nfsquad_t)((u_quad_t)0), 0, nd, p);
1660 	      if (error == 0)
1661 		/*
1662 		 * Look up the stateid
1663 		 */
1664 		error = nfsrv_getstate(clp, &new_stp->ls_stateid,
1665 		  new_stp->ls_flags, &stp);
1666 	      /*
1667 	       * do some sanity checks for an unconfirmed open or a
1668 	       * stateid that refers to the wrong file, for an open stateid
1669 	       */
1670 	      if (error == 0 && (stp->ls_flags & NFSLCK_OPEN) &&
1671 		  ((stp->ls_openowner->ls_flags & NFSLCK_NEEDSCONFIRM) ||
1672 		   (getlckret == 0 && stp->ls_lfp != lfp))){
1673 		      /*
1674 		       * NFSLCK_SETATTR should return OK rather than NFSERR_BADSTATEID
1675 		       * The only exception is using SETATTR with SIZE.
1676 		       * */
1677                     if ((new_stp->ls_flags &
1678                          (NFSLCK_SETATTR | NFSLCK_CHECK)) != NFSLCK_SETATTR)
1679 			     error = NFSERR_BADSTATEID;
1680 	      }
1681 
1682 		if (error == 0 &&
1683 		  (stp->ls_flags & (NFSLCK_DELEGREAD | NFSLCK_DELEGWRITE)) &&
1684 		  getlckret == 0 && stp->ls_lfp != lfp)
1685 			error = NFSERR_BADSTATEID;
1686 
1687 	      /*
1688 	       * If the lockowner stateid doesn't refer to the same file,
1689 	       * I believe that is considered ok, since some clients will
1690 	       * only create a single lockowner and use that for all locks
1691 	       * on all files.
1692 	       * For now, log it as a diagnostic, instead of considering it
1693 	       * a BadStateid.
1694 	       */
1695 	      if (error == 0 && (stp->ls_flags &
1696 		  (NFSLCK_OPEN | NFSLCK_DELEGREAD | NFSLCK_DELEGWRITE)) == 0 &&
1697 		  getlckret == 0 && stp->ls_lfp != lfp) {
1698 #ifdef DIAGNOSTIC
1699 		  printf("Got a lock statid for different file open\n");
1700 #endif
1701 		  /*
1702 		  error = NFSERR_BADSTATEID;
1703 		  */
1704 	      }
1705 
1706 	      if (error == 0) {
1707 		    if (new_stp->ls_flags & NFSLCK_OPENTOLOCK) {
1708 			/*
1709 			 * If haslock set, we've already checked the seqid.
1710 			 */
1711 			if (!haslock) {
1712 			    if (stp->ls_flags & NFSLCK_OPEN)
1713 				error = nfsrv_checkseqid(nd, new_stp->ls_seq,
1714 				    stp->ls_openowner, new_stp->ls_op);
1715 			    else
1716 				error = NFSERR_BADSTATEID;
1717 			}
1718 			if (!error)
1719 			    nfsrv_getowner(&stp->ls_open, new_stp, &lckstp);
1720 			if (lckstp)
1721 			    /*
1722 			     * I believe this should be an error, but it
1723 			     * isn't obvious what NFSERR_xxx would be
1724 			     * appropriate, so I'll use NFSERR_INVAL for now.
1725 			     */
1726 			    error = NFSERR_INVAL;
1727 			else
1728 			    lckstp = new_stp;
1729 		    } else if (new_stp->ls_flags&(NFSLCK_LOCK|NFSLCK_UNLOCK)) {
1730 			/*
1731 			 * If haslock set, ditto above.
1732 			 */
1733 			if (!haslock) {
1734 			    if (stp->ls_flags & NFSLCK_OPEN)
1735 				error = NFSERR_BADSTATEID;
1736 			    else
1737 				error = nfsrv_checkseqid(nd, new_stp->ls_seq,
1738 				    stp, new_stp->ls_op);
1739 			}
1740 			lckstp = stp;
1741 		    } else {
1742 			lckstp = stp;
1743 		    }
1744 	      }
1745 	      /*
1746 	       * If the seqid part of the stateid isn't the same, return
1747 	       * NFSERR_OLDSTATEID for cases other than I/O Ops.
1748 	       * For I/O Ops, only return NFSERR_OLDSTATEID if
1749 	       * nfsrv_returnoldstateid is set. (The consensus on the email
1750 	       * list was that most clients would prefer to not receive
1751 	       * NFSERR_OLDSTATEID for I/O Ops, but the RFC suggests that that
1752 	       * is what will happen, so I use the nfsrv_returnoldstateid to
1753 	       * allow for either server configuration.)
1754 	       */
1755 	      if (!error && stp->ls_stateid.seqid!=new_stp->ls_stateid.seqid &&
1756 		  (((nd->nd_flag & ND_NFSV41) == 0 &&
1757 		   (!(new_stp->ls_flags & NFSLCK_CHECK) ||
1758 		    nfsrv_returnoldstateid)) ||
1759 		   ((nd->nd_flag & ND_NFSV41) != 0 &&
1760 		    new_stp->ls_stateid.seqid != 0)))
1761 		    error = NFSERR_OLDSTATEID;
1762 	    }
1763 	}
1764 
1765 	/*
1766 	 * Now we can check for grace.
1767 	 */
1768 	if (!error)
1769 		error = nfsrv_checkgrace(nd, clp, new_stp->ls_flags);
1770 	if ((new_stp->ls_flags & NFSLCK_RECLAIM) && !error &&
1771 		nfsrv_checkstable(clp))
1772 		error = NFSERR_NOGRACE;
1773 	/*
1774 	 * If we successfully Reclaimed state, note that.
1775 	 */
1776 	if ((new_stp->ls_flags & NFSLCK_RECLAIM) && !error)
1777 		nfsrv_markstable(clp);
1778 
1779 	/*
1780 	 * At this point, either error == NFSERR_BADSTATEID or the
1781 	 * seqid# has been updated, so we can return any error.
1782 	 * If error == 0, there may be an error in:
1783 	 *    nd_repstat - Set by the calling function.
1784 	 *    reterr - Set above, if getting the nfslockfile structure
1785 	 *       or acquiring the local lock failed.
1786 	 *    (If both of these are set, nd_repstat should probably be
1787 	 *     returned, since that error was detected before this
1788 	 *     function call.)
1789 	 */
1790 	if (error != 0 || nd->nd_repstat != 0 || reterr != 0) {
1791 		if (error == 0) {
1792 			if (nd->nd_repstat != 0)
1793 				error = nd->nd_repstat;
1794 			else
1795 				error = reterr;
1796 		}
1797 		if (filestruct_locked != 0) {
1798 			/* Roll back local locks. */
1799 			NFSUNLOCKSTATE();
1800 			if (vnode_unlocked == 0) {
1801 				ASSERT_VOP_ELOCKED(vp, "nfsrv_lockctrl2");
1802 				vnode_unlocked = 1;
1803 				NFSVOPUNLOCK(vp, 0);
1804 			}
1805 			nfsrv_locallock_rollback(vp, lfp, p);
1806 			NFSLOCKSTATE();
1807 			nfsrv_unlocklf(lfp);
1808 		}
1809 		NFSUNLOCKSTATE();
1810 		goto out;
1811 	}
1812 
1813 	/*
1814 	 * Check the nfsrv_getlockfile return.
1815 	 * Returned -1 if no structure found.
1816 	 */
1817 	if (getlckret == -1) {
1818 		error = NFSERR_EXPIRED;
1819 		/*
1820 		 * Called from lockt, so no lock is OK.
1821 		 */
1822 		if (new_stp->ls_flags & NFSLCK_TEST) {
1823 			error = 0;
1824 		} else if (new_stp->ls_flags &
1825 		    (NFSLCK_CHECK | NFSLCK_SETATTR)) {
1826 			/*
1827 			 * Called to check for a lock, OK if the stateid is all
1828 			 * 1s or all 0s, but there should be an nfsstate
1829 			 * otherwise.
1830 			 * (ie. If there is no open, I'll assume no share
1831 			 *  deny bits.)
1832 			 */
1833 			if (specialid)
1834 				error = 0;
1835 			else
1836 				error = NFSERR_BADSTATEID;
1837 		}
1838 		NFSUNLOCKSTATE();
1839 		goto out;
1840 	}
1841 
1842 	/*
1843 	 * For NFSLCK_CHECK and NFSLCK_LOCK, test for a share conflict.
1844 	 * For NFSLCK_CHECK, allow a read if write access is granted,
1845 	 * but check for a deny. For NFSLCK_LOCK, require correct access,
1846 	 * which implies a conflicting deny can't exist.
1847 	 */
1848 	if (new_stp->ls_flags & (NFSLCK_CHECK | NFSLCK_LOCK)) {
1849 	    /*
1850 	     * Four kinds of state id:
1851 	     * - specialid (all 0s or all 1s), only for NFSLCK_CHECK
1852 	     * - stateid for an open
1853 	     * - stateid for a delegation
1854 	     * - stateid for a lock owner
1855 	     */
1856 	    if (!specialid) {
1857 		if (stp->ls_flags & (NFSLCK_DELEGREAD | NFSLCK_DELEGWRITE)) {
1858 		    delegation = 1;
1859 		    mystp = stp;
1860 		    nfsrv_delaydelegtimeout(stp);
1861 	        } else if (stp->ls_flags & NFSLCK_OPEN) {
1862 		    mystp = stp;
1863 		} else {
1864 		    mystp = stp->ls_openstp;
1865 		}
1866 		/*
1867 		 * If locking or checking, require correct access
1868 		 * bit set.
1869 		 */
1870 		if (((new_stp->ls_flags & NFSLCK_LOCK) &&
1871 		     !((new_lop->lo_flags >> NFSLCK_LOCKSHIFT) &
1872 		       mystp->ls_flags & NFSLCK_ACCESSBITS)) ||
1873 		    ((new_stp->ls_flags & (NFSLCK_CHECK|NFSLCK_READACCESS)) ==
1874 		      (NFSLCK_CHECK | NFSLCK_READACCESS) &&
1875 		     !(mystp->ls_flags & NFSLCK_READACCESS)) ||
1876 		    ((new_stp->ls_flags & (NFSLCK_CHECK|NFSLCK_WRITEACCESS)) ==
1877 		      (NFSLCK_CHECK | NFSLCK_WRITEACCESS) &&
1878 		     !(mystp->ls_flags & NFSLCK_WRITEACCESS))) {
1879 			if (filestruct_locked != 0) {
1880 				/* Roll back local locks. */
1881 				NFSUNLOCKSTATE();
1882 				if (vnode_unlocked == 0) {
1883 					ASSERT_VOP_ELOCKED(vp,
1884 					    "nfsrv_lockctrl3");
1885 					vnode_unlocked = 1;
1886 					NFSVOPUNLOCK(vp, 0);
1887 				}
1888 				nfsrv_locallock_rollback(vp, lfp, p);
1889 				NFSLOCKSTATE();
1890 				nfsrv_unlocklf(lfp);
1891 			}
1892 			NFSUNLOCKSTATE();
1893 			error = NFSERR_OPENMODE;
1894 			goto out;
1895 		}
1896 	    } else
1897 		mystp = NULL;
1898 	    if ((new_stp->ls_flags & NFSLCK_CHECK) && !delegation) {
1899 		/*
1900 		 * Check for a conflicting deny bit.
1901 		 */
1902 		LIST_FOREACH(tstp, &lfp->lf_open, ls_file) {
1903 		    if (tstp != mystp) {
1904 			bits = tstp->ls_flags;
1905 			bits >>= NFSLCK_SHIFT;
1906 			if (new_stp->ls_flags & bits & NFSLCK_ACCESSBITS) {
1907 			    KASSERT(vnode_unlocked == 0,
1908 				("nfsrv_lockctrl: vnode unlocked1"));
1909 			    ret = nfsrv_clientconflict(tstp->ls_clp, &haslock,
1910 				vp, p);
1911 			    if (ret == 1) {
1912 				/*
1913 				* nfsrv_clientconflict unlocks state
1914 				 * when it returns non-zero.
1915 				 */
1916 				lckstp = NULL;
1917 				goto tryagain;
1918 			    }
1919 			    if (ret == 0)
1920 				NFSUNLOCKSTATE();
1921 			    if (ret == 2)
1922 				error = NFSERR_PERM;
1923 			    else
1924 				error = NFSERR_OPENMODE;
1925 			    goto out;
1926 			}
1927 		    }
1928 		}
1929 
1930 		/* We're outta here */
1931 		NFSUNLOCKSTATE();
1932 		goto out;
1933 	    }
1934 	}
1935 
1936 	/*
1937 	 * For setattr, just get rid of all the Delegations for other clients.
1938 	 */
1939 	if (new_stp->ls_flags & NFSLCK_SETATTR) {
1940 		KASSERT(vnode_unlocked == 0,
1941 		    ("nfsrv_lockctrl: vnode unlocked2"));
1942 		ret = nfsrv_cleandeleg(vp, lfp, clp, &haslock, p);
1943 		if (ret) {
1944 			/*
1945 			 * nfsrv_cleandeleg() unlocks state when it
1946 			 * returns non-zero.
1947 			 */
1948 			if (ret == -1) {
1949 				lckstp = NULL;
1950 				goto tryagain;
1951 			}
1952 			error = ret;
1953 			goto out;
1954 		}
1955 		if (!(new_stp->ls_flags & NFSLCK_CHECK) ||
1956 		    (LIST_EMPTY(&lfp->lf_open) && LIST_EMPTY(&lfp->lf_lock) &&
1957 		     LIST_EMPTY(&lfp->lf_deleg))) {
1958 			NFSUNLOCKSTATE();
1959 			goto out;
1960 		}
1961 	}
1962 
1963 	/*
1964 	 * Check for a conflicting delegation. If one is found, call
1965 	 * nfsrv_delegconflict() to handle it. If the v4root lock hasn't
1966 	 * been set yet, it will get the lock. Otherwise, it will recall
1967 	 * the delegation. Then, we try try again...
1968 	 * I currently believe the conflict algorithm to be:
1969 	 * For Lock Ops (Lock/LockT/LockU)
1970 	 * - there is a conflict iff a different client has a write delegation
1971 	 * For Reading (Read Op)
1972 	 * - there is a conflict iff a different client has a write delegation
1973 	 *   (the specialids are always a different client)
1974 	 * For Writing (Write/Setattr of size)
1975 	 * - there is a conflict if a different client has any delegation
1976 	 * - there is a conflict if the same client has a read delegation
1977 	 *   (I don't understand why this isn't allowed, but that seems to be
1978 	 *    the current consensus?)
1979 	 */
1980 	tstp = LIST_FIRST(&lfp->lf_deleg);
1981 	while (tstp != LIST_END(&lfp->lf_deleg)) {
1982 	    nstp = LIST_NEXT(tstp, ls_file);
1983 	    if ((((new_stp->ls_flags&(NFSLCK_LOCK|NFSLCK_UNLOCK|NFSLCK_TEST))||
1984 		 ((new_stp->ls_flags & NFSLCK_CHECK) &&
1985 		  (new_lop->lo_flags & NFSLCK_READ))) &&
1986 		  clp != tstp->ls_clp &&
1987 		 (tstp->ls_flags & NFSLCK_DELEGWRITE)) ||
1988 		 ((new_stp->ls_flags & NFSLCK_CHECK) &&
1989 		   (new_lop->lo_flags & NFSLCK_WRITE) &&
1990 		  (clp != tstp->ls_clp ||
1991 		   (tstp->ls_flags & NFSLCK_DELEGREAD)))) {
1992 		ret = 0;
1993 		if (filestruct_locked != 0) {
1994 			/* Roll back local locks. */
1995 			NFSUNLOCKSTATE();
1996 			if (vnode_unlocked == 0) {
1997 				ASSERT_VOP_ELOCKED(vp, "nfsrv_lockctrl4");
1998 				NFSVOPUNLOCK(vp, 0);
1999 			}
2000 			nfsrv_locallock_rollback(vp, lfp, p);
2001 			NFSLOCKSTATE();
2002 			nfsrv_unlocklf(lfp);
2003 			NFSUNLOCKSTATE();
2004 			NFSVOPLOCK(vp, LK_EXCLUSIVE | LK_RETRY);
2005 			vnode_unlocked = 0;
2006 			if ((vp->v_iflag & VI_DOOMED) != 0)
2007 				ret = NFSERR_SERVERFAULT;
2008 			NFSLOCKSTATE();
2009 		}
2010 		if (ret == 0)
2011 			ret = nfsrv_delegconflict(tstp, &haslock, p, vp);
2012 		if (ret) {
2013 		    /*
2014 		     * nfsrv_delegconflict unlocks state when it
2015 		     * returns non-zero, which it always does.
2016 		     */
2017 		    if (other_lop) {
2018 			FREE((caddr_t)other_lop, M_NFSDLOCK);
2019 			other_lop = NULL;
2020 		    }
2021 		    if (ret == -1) {
2022 			lckstp = NULL;
2023 			goto tryagain;
2024 		    }
2025 		    error = ret;
2026 		    goto out;
2027 		}
2028 		/* Never gets here. */
2029 	    }
2030 	    tstp = nstp;
2031 	}
2032 
2033 	/*
2034 	 * Handle the unlock case by calling nfsrv_updatelock().
2035 	 * (Should I have done some access checking above for unlock? For now,
2036 	 *  just let it happen.)
2037 	 */
2038 	if (new_stp->ls_flags & NFSLCK_UNLOCK) {
2039 		first = new_lop->lo_first;
2040 		end = new_lop->lo_end;
2041 		nfsrv_updatelock(stp, new_lopp, &other_lop, lfp);
2042 		stateidp->seqid = ++(stp->ls_stateid.seqid);
2043 		if ((nd->nd_flag & ND_NFSV41) != 0 && stateidp->seqid == 0)
2044 			stateidp->seqid = stp->ls_stateid.seqid = 1;
2045 		stateidp->other[0] = stp->ls_stateid.other[0];
2046 		stateidp->other[1] = stp->ls_stateid.other[1];
2047 		stateidp->other[2] = stp->ls_stateid.other[2];
2048 		if (filestruct_locked != 0) {
2049 			NFSUNLOCKSTATE();
2050 			if (vnode_unlocked == 0) {
2051 				ASSERT_VOP_ELOCKED(vp, "nfsrv_lockctrl5");
2052 				vnode_unlocked = 1;
2053 				NFSVOPUNLOCK(vp, 0);
2054 			}
2055 			/* Update the local locks. */
2056 			nfsrv_localunlock(vp, lfp, first, end, p);
2057 			NFSLOCKSTATE();
2058 			nfsrv_unlocklf(lfp);
2059 		}
2060 		NFSUNLOCKSTATE();
2061 		goto out;
2062 	}
2063 
2064 	/*
2065 	 * Search for a conflicting lock. A lock conflicts if:
2066 	 * - the lock range overlaps and
2067 	 * - at least one lock is a write lock and
2068 	 * - it is not owned by the same lock owner
2069 	 */
2070 	if (!delegation) {
2071 	  LIST_FOREACH(lop, &lfp->lf_lock, lo_lckfile) {
2072 	    if (new_lop->lo_end > lop->lo_first &&
2073 		new_lop->lo_first < lop->lo_end &&
2074 		(new_lop->lo_flags == NFSLCK_WRITE ||
2075 		 lop->lo_flags == NFSLCK_WRITE) &&
2076 		lckstp != lop->lo_stp &&
2077 		(clp != lop->lo_stp->ls_clp ||
2078 		 lckstp->ls_ownerlen != lop->lo_stp->ls_ownerlen ||
2079 		 NFSBCMP(lckstp->ls_owner, lop->lo_stp->ls_owner,
2080 		    lckstp->ls_ownerlen))) {
2081 		if (other_lop) {
2082 		    FREE((caddr_t)other_lop, M_NFSDLOCK);
2083 		    other_lop = NULL;
2084 		}
2085 		if (vnode_unlocked != 0)
2086 		    ret = nfsrv_clientconflict(lop->lo_stp->ls_clp, &haslock,
2087 			NULL, p);
2088 		else
2089 		    ret = nfsrv_clientconflict(lop->lo_stp->ls_clp, &haslock,
2090 			vp, p);
2091 		if (ret == 1) {
2092 		    if (filestruct_locked != 0) {
2093 			if (vnode_unlocked == 0) {
2094 				ASSERT_VOP_ELOCKED(vp, "nfsrv_lockctrl6");
2095 				NFSVOPUNLOCK(vp, 0);
2096 			}
2097 			/* Roll back local locks. */
2098 			nfsrv_locallock_rollback(vp, lfp, p);
2099 			NFSLOCKSTATE();
2100 			nfsrv_unlocklf(lfp);
2101 			NFSUNLOCKSTATE();
2102 			NFSVOPLOCK(vp, LK_EXCLUSIVE | LK_RETRY);
2103 			vnode_unlocked = 0;
2104 			if ((vp->v_iflag & VI_DOOMED) != 0) {
2105 				error = NFSERR_SERVERFAULT;
2106 				goto out;
2107 			}
2108 		    }
2109 		    /*
2110 		     * nfsrv_clientconflict() unlocks state when it
2111 		     * returns non-zero.
2112 		     */
2113 		    lckstp = NULL;
2114 		    goto tryagain;
2115 		}
2116 		/*
2117 		 * Found a conflicting lock, so record the conflict and
2118 		 * return the error.
2119 		 */
2120 		if (cfp != NULL && ret == 0) {
2121 		    cfp->cl_clientid.lval[0]=lop->lo_stp->ls_stateid.other[0];
2122 		    cfp->cl_clientid.lval[1]=lop->lo_stp->ls_stateid.other[1];
2123 		    cfp->cl_first = lop->lo_first;
2124 		    cfp->cl_end = lop->lo_end;
2125 		    cfp->cl_flags = lop->lo_flags;
2126 		    cfp->cl_ownerlen = lop->lo_stp->ls_ownerlen;
2127 		    NFSBCOPY(lop->lo_stp->ls_owner, cfp->cl_owner,
2128 			cfp->cl_ownerlen);
2129 		}
2130 		if (ret == 2)
2131 		    error = NFSERR_PERM;
2132 		else if (new_stp->ls_flags & NFSLCK_RECLAIM)
2133 		    error = NFSERR_RECLAIMCONFLICT;
2134 		else if (new_stp->ls_flags & NFSLCK_CHECK)
2135 		    error = NFSERR_LOCKED;
2136 		else
2137 		    error = NFSERR_DENIED;
2138 		if (filestruct_locked != 0 && ret == 0) {
2139 			/* Roll back local locks. */
2140 			NFSUNLOCKSTATE();
2141 			if (vnode_unlocked == 0) {
2142 				ASSERT_VOP_ELOCKED(vp, "nfsrv_lockctrl7");
2143 				vnode_unlocked = 1;
2144 				NFSVOPUNLOCK(vp, 0);
2145 			}
2146 			nfsrv_locallock_rollback(vp, lfp, p);
2147 			NFSLOCKSTATE();
2148 			nfsrv_unlocklf(lfp);
2149 		}
2150 		if (ret == 0)
2151 			NFSUNLOCKSTATE();
2152 		goto out;
2153 	    }
2154 	  }
2155 	}
2156 
2157 	/*
2158 	 * We only get here if there was no lock that conflicted.
2159 	 */
2160 	if (new_stp->ls_flags & (NFSLCK_TEST | NFSLCK_CHECK)) {
2161 		NFSUNLOCKSTATE();
2162 		goto out;
2163 	}
2164 
2165 	/*
2166 	 * We only get here when we are creating or modifying a lock.
2167 	 * There are two variants:
2168 	 * - exist_lock_owner where lock_owner exists
2169 	 * - open_to_lock_owner with new lock_owner
2170 	 */
2171 	first = new_lop->lo_first;
2172 	end = new_lop->lo_end;
2173 	lock_flags = new_lop->lo_flags;
2174 	if (!(new_stp->ls_flags & NFSLCK_OPENTOLOCK)) {
2175 		nfsrv_updatelock(lckstp, new_lopp, &other_lop, lfp);
2176 		stateidp->seqid = ++(lckstp->ls_stateid.seqid);
2177 		if ((nd->nd_flag & ND_NFSV41) != 0 && stateidp->seqid == 0)
2178 			stateidp->seqid = lckstp->ls_stateid.seqid = 1;
2179 		stateidp->other[0] = lckstp->ls_stateid.other[0];
2180 		stateidp->other[1] = lckstp->ls_stateid.other[1];
2181 		stateidp->other[2] = lckstp->ls_stateid.other[2];
2182 	} else {
2183 		/*
2184 		 * The new open_to_lock_owner case.
2185 		 * Link the new nfsstate into the lists.
2186 		 */
2187 		new_stp->ls_seq = new_stp->ls_opentolockseq;
2188 		nfsrvd_refcache(new_stp->ls_op);
2189 		stateidp->seqid = new_stp->ls_stateid.seqid = 1;
2190 		stateidp->other[0] = new_stp->ls_stateid.other[0] =
2191 		    clp->lc_clientid.lval[0];
2192 		stateidp->other[1] = new_stp->ls_stateid.other[1] =
2193 		    clp->lc_clientid.lval[1];
2194 		stateidp->other[2] = new_stp->ls_stateid.other[2] =
2195 		    nfsrv_nextstateindex(clp);
2196 		new_stp->ls_clp = clp;
2197 		LIST_INIT(&new_stp->ls_lock);
2198 		new_stp->ls_openstp = stp;
2199 		new_stp->ls_lfp = lfp;
2200 		nfsrv_insertlock(new_lop, (struct nfslock *)new_stp, new_stp,
2201 		    lfp);
2202 		LIST_INSERT_HEAD(NFSSTATEHASH(clp, new_stp->ls_stateid),
2203 		    new_stp, ls_hash);
2204 		LIST_INSERT_HEAD(&stp->ls_open, new_stp, ls_list);
2205 		*new_lopp = NULL;
2206 		*new_stpp = NULL;
2207 		nfsstatsv1.srvlockowners++;
2208 		nfsrv_openpluslock++;
2209 	}
2210 	if (filestruct_locked != 0) {
2211 		NFSUNLOCKSTATE();
2212 		nfsrv_locallock_commit(lfp, lock_flags, first, end);
2213 		NFSLOCKSTATE();
2214 		nfsrv_unlocklf(lfp);
2215 	}
2216 	NFSUNLOCKSTATE();
2217 
2218 out:
2219 	if (haslock) {
2220 		NFSLOCKV4ROOTMUTEX();
2221 		nfsv4_unlock(&nfsv4rootfs_lock, 1);
2222 		NFSUNLOCKV4ROOTMUTEX();
2223 	}
2224 	if (vnode_unlocked != 0) {
2225 		NFSVOPLOCK(vp, LK_EXCLUSIVE | LK_RETRY);
2226 		if (error == 0 && (vp->v_iflag & VI_DOOMED) != 0)
2227 			error = NFSERR_SERVERFAULT;
2228 	}
2229 	if (other_lop)
2230 		FREE((caddr_t)other_lop, M_NFSDLOCK);
2231 	NFSEXITCODE2(error, nd);
2232 	return (error);
2233 }
2234 
2235 /*
2236  * Check for state errors for Open.
2237  * repstat is passed back out as an error if more critical errors
2238  * are not detected.
2239  */
2240 APPLESTATIC int
2241 nfsrv_opencheck(nfsquad_t clientid, nfsv4stateid_t *stateidp,
2242     struct nfsstate *new_stp, vnode_t vp, struct nfsrv_descript *nd,
2243     NFSPROC_T *p, int repstat)
2244 {
2245 	struct nfsstate *stp, *nstp;
2246 	struct nfsclient *clp;
2247 	struct nfsstate *ownerstp;
2248 	struct nfslockfile *lfp, *new_lfp;
2249 	int error = 0, haslock = 0, ret, readonly = 0, getfhret = 0;
2250 
2251 	if ((new_stp->ls_flags & NFSLCK_SHAREBITS) == NFSLCK_READACCESS)
2252 		readonly = 1;
2253 	/*
2254 	 * Check for restart conditions (client and server).
2255 	 */
2256 	error = nfsrv_checkrestart(clientid, new_stp->ls_flags,
2257 		&new_stp->ls_stateid, 0);
2258 	if (error)
2259 		goto out;
2260 
2261 	/*
2262 	 * Check for state resource limit exceeded.
2263 	 * Technically this should be SMP protected, but the worst
2264 	 * case error is "out by one or two" on the count when it
2265 	 * returns NFSERR_RESOURCE and the limit is just a rather
2266 	 * arbitrary high water mark, so no harm is done.
2267 	 */
2268 	if (nfsrv_openpluslock > nfsrv_v4statelimit) {
2269 		error = NFSERR_RESOURCE;
2270 		goto out;
2271 	}
2272 
2273 tryagain:
2274 	MALLOC(new_lfp, struct nfslockfile *, sizeof (struct nfslockfile),
2275 	    M_NFSDLOCKFILE, M_WAITOK);
2276 	if (vp)
2277 		getfhret = nfsrv_getlockfh(vp, new_stp->ls_flags, new_lfp,
2278 		    NULL, p);
2279 	NFSLOCKSTATE();
2280 	/*
2281 	 * Get the nfsclient structure.
2282 	 */
2283 	error = nfsrv_getclient(clientid, CLOPS_RENEW, &clp, NULL,
2284 	    (nfsquad_t)((u_quad_t)0), 0, nd, p);
2285 
2286 	/*
2287 	 * Look up the open owner. See if it needs confirmation and
2288 	 * check the seq#, as required.
2289 	 */
2290 	if (!error)
2291 		nfsrv_getowner(&clp->lc_open, new_stp, &ownerstp);
2292 
2293 	if (!error && ownerstp) {
2294 		error = nfsrv_checkseqid(nd, new_stp->ls_seq, ownerstp,
2295 		    new_stp->ls_op);
2296 		/*
2297 		 * If the OpenOwner hasn't been confirmed, assume the
2298 		 * old one was a replay and this one is ok.
2299 		 * See: RFC3530 Sec. 14.2.18.
2300 		 */
2301 		if (error == NFSERR_BADSEQID &&
2302 		    (ownerstp->ls_flags & NFSLCK_NEEDSCONFIRM))
2303 			error = 0;
2304 	}
2305 
2306 	/*
2307 	 * Check for grace.
2308 	 */
2309 	if (!error)
2310 		error = nfsrv_checkgrace(nd, clp, new_stp->ls_flags);
2311 	if ((new_stp->ls_flags & NFSLCK_RECLAIM) && !error &&
2312 		nfsrv_checkstable(clp))
2313 		error = NFSERR_NOGRACE;
2314 
2315 	/*
2316 	 * If none of the above errors occurred, let repstat be
2317 	 * returned.
2318 	 */
2319 	if (repstat && !error)
2320 		error = repstat;
2321 	if (error) {
2322 		NFSUNLOCKSTATE();
2323 		if (haslock) {
2324 			NFSLOCKV4ROOTMUTEX();
2325 			nfsv4_unlock(&nfsv4rootfs_lock, 1);
2326 			NFSUNLOCKV4ROOTMUTEX();
2327 		}
2328 		free((caddr_t)new_lfp, M_NFSDLOCKFILE);
2329 		goto out;
2330 	}
2331 
2332 	/*
2333 	 * If vp == NULL, the file doesn't exist yet, so return ok.
2334 	 * (This always happens on the first pass, so haslock must be 0.)
2335 	 */
2336 	if (vp == NULL) {
2337 		NFSUNLOCKSTATE();
2338 		FREE((caddr_t)new_lfp, M_NFSDLOCKFILE);
2339 		goto out;
2340 	}
2341 
2342 	/*
2343 	 * Get the structure for the underlying file.
2344 	 */
2345 	if (getfhret)
2346 		error = getfhret;
2347 	else
2348 		error = nfsrv_getlockfile(new_stp->ls_flags, &new_lfp, &lfp,
2349 		    NULL, 0);
2350 	if (new_lfp)
2351 		FREE((caddr_t)new_lfp, M_NFSDLOCKFILE);
2352 	if (error) {
2353 		NFSUNLOCKSTATE();
2354 		if (haslock) {
2355 			NFSLOCKV4ROOTMUTEX();
2356 			nfsv4_unlock(&nfsv4rootfs_lock, 1);
2357 			NFSUNLOCKV4ROOTMUTEX();
2358 		}
2359 		goto out;
2360 	}
2361 
2362 	/*
2363 	 * Search for a conflicting open/share.
2364 	 */
2365 	if (new_stp->ls_flags & NFSLCK_DELEGCUR) {
2366 	    /*
2367 	     * For Delegate_Cur, search for the matching Delegation,
2368 	     * which indicates no conflict.
2369 	     * An old delegation should have been recovered by the
2370 	     * client doing a Claim_DELEGATE_Prev, so I won't let
2371 	     * it match and return NFSERR_EXPIRED. Should I let it
2372 	     * match?
2373 	     */
2374 	    LIST_FOREACH(stp, &lfp->lf_deleg, ls_file) {
2375 		if (!(stp->ls_flags & NFSLCK_OLDDELEG) &&
2376 		    (((nd->nd_flag & ND_NFSV41) != 0 &&
2377 		    stateidp->seqid == 0) ||
2378 		    stateidp->seqid == stp->ls_stateid.seqid) &&
2379 		    !NFSBCMP(stateidp->other, stp->ls_stateid.other,
2380 			  NFSX_STATEIDOTHER))
2381 			break;
2382 	    }
2383 	    if (stp == LIST_END(&lfp->lf_deleg) ||
2384 		((new_stp->ls_flags & NFSLCK_WRITEACCESS) &&
2385 		 (stp->ls_flags & NFSLCK_DELEGREAD))) {
2386 		NFSUNLOCKSTATE();
2387 		if (haslock) {
2388 			NFSLOCKV4ROOTMUTEX();
2389 			nfsv4_unlock(&nfsv4rootfs_lock, 1);
2390 			NFSUNLOCKV4ROOTMUTEX();
2391 		}
2392 		error = NFSERR_EXPIRED;
2393 		goto out;
2394 	    }
2395 	}
2396 
2397 	/*
2398 	 * Check for access/deny bit conflicts. I check for the same
2399 	 * owner as well, in case the client didn't bother.
2400 	 */
2401 	LIST_FOREACH(stp, &lfp->lf_open, ls_file) {
2402 		if (!(new_stp->ls_flags & NFSLCK_DELEGCUR) &&
2403 		    (((new_stp->ls_flags & NFSLCK_ACCESSBITS) &
2404 		      ((stp->ls_flags>>NFSLCK_SHIFT) & NFSLCK_ACCESSBITS))||
2405 		     ((stp->ls_flags & NFSLCK_ACCESSBITS) &
2406 		      ((new_stp->ls_flags>>NFSLCK_SHIFT)&NFSLCK_ACCESSBITS)))){
2407 			ret = nfsrv_clientconflict(stp->ls_clp,&haslock,vp,p);
2408 			if (ret == 1) {
2409 				/*
2410 				 * nfsrv_clientconflict() unlocks
2411 				 * state when it returns non-zero.
2412 				 */
2413 				goto tryagain;
2414 			}
2415 			if (ret == 2)
2416 				error = NFSERR_PERM;
2417 			else if (new_stp->ls_flags & NFSLCK_RECLAIM)
2418 				error = NFSERR_RECLAIMCONFLICT;
2419 			else
2420 				error = NFSERR_SHAREDENIED;
2421 			if (ret == 0)
2422 				NFSUNLOCKSTATE();
2423 			if (haslock) {
2424 				NFSLOCKV4ROOTMUTEX();
2425 				nfsv4_unlock(&nfsv4rootfs_lock, 1);
2426 				NFSUNLOCKV4ROOTMUTEX();
2427 			}
2428 			goto out;
2429 		}
2430 	}
2431 
2432 	/*
2433 	 * Check for a conflicting delegation. If one is found, call
2434 	 * nfsrv_delegconflict() to handle it. If the v4root lock hasn't
2435 	 * been set yet, it will get the lock. Otherwise, it will recall
2436 	 * the delegation. Then, we try try again...
2437 	 * (If NFSLCK_DELEGCUR is set, it has a delegation, so there
2438 	 *  isn't a conflict.)
2439 	 * I currently believe the conflict algorithm to be:
2440 	 * For Open with Read Access and Deny None
2441 	 * - there is a conflict iff a different client has a write delegation
2442 	 * For Open with other Write Access or any Deny except None
2443 	 * - there is a conflict if a different client has any delegation
2444 	 * - there is a conflict if the same client has a read delegation
2445 	 *   (The current consensus is that this last case should be
2446 	 *    considered a conflict since the client with a read delegation
2447 	 *    could have done an Open with ReadAccess and WriteDeny
2448 	 *    locally and then not have checked for the WriteDeny.)
2449 	 * Don't check for a Reclaim, since that will be dealt with
2450 	 * by nfsrv_openctrl().
2451 	 */
2452 	if (!(new_stp->ls_flags &
2453 		(NFSLCK_DELEGPREV | NFSLCK_DELEGCUR | NFSLCK_RECLAIM))) {
2454 	    stp = LIST_FIRST(&lfp->lf_deleg);
2455 	    while (stp != LIST_END(&lfp->lf_deleg)) {
2456 		nstp = LIST_NEXT(stp, ls_file);
2457 		if ((readonly && stp->ls_clp != clp &&
2458 		       (stp->ls_flags & NFSLCK_DELEGWRITE)) ||
2459 		    (!readonly && (stp->ls_clp != clp ||
2460 		         (stp->ls_flags & NFSLCK_DELEGREAD)))) {
2461 			ret = nfsrv_delegconflict(stp, &haslock, p, vp);
2462 			if (ret) {
2463 			    /*
2464 			     * nfsrv_delegconflict() unlocks state
2465 			     * when it returns non-zero.
2466 			     */
2467 			    if (ret == -1)
2468 				goto tryagain;
2469 			    error = ret;
2470 			    goto out;
2471 			}
2472 		}
2473 		stp = nstp;
2474 	    }
2475 	}
2476 	NFSUNLOCKSTATE();
2477 	if (haslock) {
2478 		NFSLOCKV4ROOTMUTEX();
2479 		nfsv4_unlock(&nfsv4rootfs_lock, 1);
2480 		NFSUNLOCKV4ROOTMUTEX();
2481 	}
2482 
2483 out:
2484 	NFSEXITCODE2(error, nd);
2485 	return (error);
2486 }
2487 
2488 /*
2489  * Open control function to create/update open state for an open.
2490  */
2491 APPLESTATIC int
2492 nfsrv_openctrl(struct nfsrv_descript *nd, vnode_t vp,
2493     struct nfsstate **new_stpp, nfsquad_t clientid, nfsv4stateid_t *stateidp,
2494     nfsv4stateid_t *delegstateidp, u_int32_t *rflagsp, struct nfsexstuff *exp,
2495     NFSPROC_T *p, u_quad_t filerev)
2496 {
2497 	struct nfsstate *new_stp = *new_stpp;
2498 	struct nfsstate *stp, *nstp;
2499 	struct nfsstate *openstp = NULL, *new_open, *ownerstp, *new_deleg;
2500 	struct nfslockfile *lfp, *new_lfp;
2501 	struct nfsclient *clp;
2502 	int error = 0, haslock = 0, ret, delegate = 1, writedeleg = 1;
2503 	int readonly = 0, cbret = 1, getfhret = 0;
2504 
2505 	if ((new_stp->ls_flags & NFSLCK_SHAREBITS) == NFSLCK_READACCESS)
2506 		readonly = 1;
2507 	/*
2508 	 * Check for restart conditions (client and server).
2509 	 * (Paranoia, should have been detected by nfsrv_opencheck().)
2510 	 * If an error does show up, return NFSERR_EXPIRED, since the
2511 	 * the seqid# has already been incremented.
2512 	 */
2513 	error = nfsrv_checkrestart(clientid, new_stp->ls_flags,
2514 	    &new_stp->ls_stateid, 0);
2515 	if (error) {
2516 		printf("Nfsd: openctrl unexpected restart err=%d\n",
2517 		    error);
2518 		error = NFSERR_EXPIRED;
2519 		goto out;
2520 	}
2521 
2522 tryagain:
2523 	MALLOC(new_lfp, struct nfslockfile *, sizeof (struct nfslockfile),
2524 	    M_NFSDLOCKFILE, M_WAITOK);
2525 	MALLOC(new_open, struct nfsstate *, sizeof (struct nfsstate),
2526 	    M_NFSDSTATE, M_WAITOK);
2527 	MALLOC(new_deleg, struct nfsstate *, sizeof (struct nfsstate),
2528 	    M_NFSDSTATE, M_WAITOK);
2529 	getfhret = nfsrv_getlockfh(vp, new_stp->ls_flags, new_lfp,
2530 	    NULL, p);
2531 	NFSLOCKSTATE();
2532 	/*
2533 	 * Get the client structure. Since the linked lists could be changed
2534 	 * by other nfsd processes if this process does a tsleep(), one of
2535 	 * two things must be done.
2536 	 * 1 - don't tsleep()
2537 	 * or
2538 	 * 2 - get the nfsv4_lock() { indicated by haslock == 1 }
2539 	 *     before using the lists, since this lock stops the other
2540 	 *     nfsd. This should only be used for rare cases, since it
2541 	 *     essentially single threads the nfsd.
2542 	 *     At this time, it is only done for cases where the stable
2543 	 *     storage file must be written prior to completion of state
2544 	 *     expiration.
2545 	 */
2546 	error = nfsrv_getclient(clientid, CLOPS_RENEW, &clp, NULL,
2547 	    (nfsquad_t)((u_quad_t)0), 0, nd, p);
2548 	if (!error && (clp->lc_flags & LCL_NEEDSCBNULL) &&
2549 	    clp->lc_program) {
2550 		/*
2551 		 * This happens on the first open for a client
2552 		 * that supports callbacks.
2553 		 */
2554 		NFSUNLOCKSTATE();
2555 		/*
2556 		 * Although nfsrv_docallback() will sleep, clp won't
2557 		 * go away, since they are only removed when the
2558 		 * nfsv4_lock() has blocked the nfsd threads. The
2559 		 * fields in clp can change, but having multiple
2560 		 * threads do this Null callback RPC should be
2561 		 * harmless.
2562 		 */
2563 		cbret = nfsrv_docallback(clp, NFSV4PROC_CBNULL,
2564 		    NULL, 0, NULL, NULL, NULL, p);
2565 		NFSLOCKSTATE();
2566 		clp->lc_flags &= ~LCL_NEEDSCBNULL;
2567 		if (!cbret)
2568 			clp->lc_flags |= LCL_CALLBACKSON;
2569 	}
2570 
2571 	/*
2572 	 * Look up the open owner. See if it needs confirmation and
2573 	 * check the seq#, as required.
2574 	 */
2575 	if (!error)
2576 		nfsrv_getowner(&clp->lc_open, new_stp, &ownerstp);
2577 
2578 	if (error) {
2579 		NFSUNLOCKSTATE();
2580 		printf("Nfsd: openctrl unexpected state err=%d\n",
2581 			error);
2582 		free((caddr_t)new_lfp, M_NFSDLOCKFILE);
2583 		free((caddr_t)new_open, M_NFSDSTATE);
2584 		free((caddr_t)new_deleg, M_NFSDSTATE);
2585 		if (haslock) {
2586 			NFSLOCKV4ROOTMUTEX();
2587 			nfsv4_unlock(&nfsv4rootfs_lock, 1);
2588 			NFSUNLOCKV4ROOTMUTEX();
2589 		}
2590 		error = NFSERR_EXPIRED;
2591 		goto out;
2592 	}
2593 
2594 	if (new_stp->ls_flags & NFSLCK_RECLAIM)
2595 		nfsrv_markstable(clp);
2596 
2597 	/*
2598 	 * Get the structure for the underlying file.
2599 	 */
2600 	if (getfhret)
2601 		error = getfhret;
2602 	else
2603 		error = nfsrv_getlockfile(new_stp->ls_flags, &new_lfp, &lfp,
2604 		    NULL, 0);
2605 	if (new_lfp)
2606 		FREE((caddr_t)new_lfp, M_NFSDLOCKFILE);
2607 	if (error) {
2608 		NFSUNLOCKSTATE();
2609 		printf("Nfsd openctrl unexpected getlockfile err=%d\n",
2610 		    error);
2611 		free((caddr_t)new_open, M_NFSDSTATE);
2612 		free((caddr_t)new_deleg, M_NFSDSTATE);
2613 		if (haslock) {
2614 			NFSLOCKV4ROOTMUTEX();
2615 			nfsv4_unlock(&nfsv4rootfs_lock, 1);
2616 			NFSUNLOCKV4ROOTMUTEX();
2617 		}
2618 		goto out;
2619 	}
2620 
2621 	/*
2622 	 * Search for a conflicting open/share.
2623 	 */
2624 	if (new_stp->ls_flags & NFSLCK_DELEGCUR) {
2625 	    /*
2626 	     * For Delegate_Cur, search for the matching Delegation,
2627 	     * which indicates no conflict.
2628 	     * An old delegation should have been recovered by the
2629 	     * client doing a Claim_DELEGATE_Prev, so I won't let
2630 	     * it match and return NFSERR_EXPIRED. Should I let it
2631 	     * match?
2632 	     */
2633 	    LIST_FOREACH(stp, &lfp->lf_deleg, ls_file) {
2634 		if (!(stp->ls_flags & NFSLCK_OLDDELEG) &&
2635 		    (((nd->nd_flag & ND_NFSV41) != 0 &&
2636 		    stateidp->seqid == 0) ||
2637 		    stateidp->seqid == stp->ls_stateid.seqid) &&
2638 		    !NFSBCMP(stateidp->other, stp->ls_stateid.other,
2639 			NFSX_STATEIDOTHER))
2640 			break;
2641 	    }
2642 	    if (stp == LIST_END(&lfp->lf_deleg) ||
2643 		((new_stp->ls_flags & NFSLCK_WRITEACCESS) &&
2644 		 (stp->ls_flags & NFSLCK_DELEGREAD))) {
2645 		NFSUNLOCKSTATE();
2646 		printf("Nfsd openctrl unexpected expiry\n");
2647 		free((caddr_t)new_open, M_NFSDSTATE);
2648 		free((caddr_t)new_deleg, M_NFSDSTATE);
2649 		if (haslock) {
2650 			NFSLOCKV4ROOTMUTEX();
2651 			nfsv4_unlock(&nfsv4rootfs_lock, 1);
2652 			NFSUNLOCKV4ROOTMUTEX();
2653 		}
2654 		error = NFSERR_EXPIRED;
2655 		goto out;
2656 	    }
2657 
2658 	    /*
2659 	     * Don't issue a Delegation, since one already exists and
2660 	     * delay delegation timeout, as required.
2661 	     */
2662 	    delegate = 0;
2663 	    nfsrv_delaydelegtimeout(stp);
2664 	}
2665 
2666 	/*
2667 	 * Check for access/deny bit conflicts. I also check for the
2668 	 * same owner, since the client might not have bothered to check.
2669 	 * Also, note an open for the same file and owner, if found,
2670 	 * which is all we do here for Delegate_Cur, since conflict
2671 	 * checking is already done.
2672 	 */
2673 	LIST_FOREACH(stp, &lfp->lf_open, ls_file) {
2674 		if (ownerstp && stp->ls_openowner == ownerstp)
2675 			openstp = stp;
2676 		if (!(new_stp->ls_flags & NFSLCK_DELEGCUR)) {
2677 		    /*
2678 		     * If another client has the file open, the only
2679 		     * delegation that can be issued is a Read delegation
2680 		     * and only if it is a Read open with Deny none.
2681 		     */
2682 		    if (clp != stp->ls_clp) {
2683 			if ((stp->ls_flags & NFSLCK_SHAREBITS) ==
2684 			    NFSLCK_READACCESS)
2685 			    writedeleg = 0;
2686 			else
2687 			    delegate = 0;
2688 		    }
2689 		    if(((new_stp->ls_flags & NFSLCK_ACCESSBITS) &
2690 		        ((stp->ls_flags>>NFSLCK_SHIFT) & NFSLCK_ACCESSBITS))||
2691 		       ((stp->ls_flags & NFSLCK_ACCESSBITS) &
2692 		        ((new_stp->ls_flags>>NFSLCK_SHIFT)&NFSLCK_ACCESSBITS))){
2693 			ret = nfsrv_clientconflict(stp->ls_clp,&haslock,vp,p);
2694 			if (ret == 1) {
2695 				/*
2696 				 * nfsrv_clientconflict() unlocks state
2697 				 * when it returns non-zero.
2698 				 */
2699 				free((caddr_t)new_open, M_NFSDSTATE);
2700 				free((caddr_t)new_deleg, M_NFSDSTATE);
2701 				openstp = NULL;
2702 				goto tryagain;
2703 			}
2704 			if (ret == 2)
2705 				error = NFSERR_PERM;
2706 			else if (new_stp->ls_flags & NFSLCK_RECLAIM)
2707 				error = NFSERR_RECLAIMCONFLICT;
2708 			else
2709 				error = NFSERR_SHAREDENIED;
2710 			if (ret == 0)
2711 				NFSUNLOCKSTATE();
2712 			if (haslock) {
2713 				NFSLOCKV4ROOTMUTEX();
2714 				nfsv4_unlock(&nfsv4rootfs_lock, 1);
2715 				NFSUNLOCKV4ROOTMUTEX();
2716 			}
2717 			free((caddr_t)new_open, M_NFSDSTATE);
2718 			free((caddr_t)new_deleg, M_NFSDSTATE);
2719 			printf("nfsd openctrl unexpected client cnfl\n");
2720 			goto out;
2721 		    }
2722 		}
2723 	}
2724 
2725 	/*
2726 	 * Check for a conflicting delegation. If one is found, call
2727 	 * nfsrv_delegconflict() to handle it. If the v4root lock hasn't
2728 	 * been set yet, it will get the lock. Otherwise, it will recall
2729 	 * the delegation. Then, we try try again...
2730 	 * (If NFSLCK_DELEGCUR is set, it has a delegation, so there
2731 	 *  isn't a conflict.)
2732 	 * I currently believe the conflict algorithm to be:
2733 	 * For Open with Read Access and Deny None
2734 	 * - there is a conflict iff a different client has a write delegation
2735 	 * For Open with other Write Access or any Deny except None
2736 	 * - there is a conflict if a different client has any delegation
2737 	 * - there is a conflict if the same client has a read delegation
2738 	 *   (The current consensus is that this last case should be
2739 	 *    considered a conflict since the client with a read delegation
2740 	 *    could have done an Open with ReadAccess and WriteDeny
2741 	 *    locally and then not have checked for the WriteDeny.)
2742 	 */
2743 	if (!(new_stp->ls_flags & (NFSLCK_DELEGPREV | NFSLCK_DELEGCUR))) {
2744 	    stp = LIST_FIRST(&lfp->lf_deleg);
2745 	    while (stp != LIST_END(&lfp->lf_deleg)) {
2746 		nstp = LIST_NEXT(stp, ls_file);
2747 		if (stp->ls_clp != clp && (stp->ls_flags & NFSLCK_DELEGREAD))
2748 			writedeleg = 0;
2749 		else
2750 			delegate = 0;
2751 		if ((readonly && stp->ls_clp != clp &&
2752 		       (stp->ls_flags & NFSLCK_DELEGWRITE)) ||
2753 		    (!readonly && (stp->ls_clp != clp ||
2754 		         (stp->ls_flags & NFSLCK_DELEGREAD)))) {
2755 		    if (new_stp->ls_flags & NFSLCK_RECLAIM) {
2756 			delegate = 2;
2757 		    } else {
2758 			ret = nfsrv_delegconflict(stp, &haslock, p, vp);
2759 			if (ret) {
2760 			    /*
2761 			     * nfsrv_delegconflict() unlocks state
2762 			     * when it returns non-zero.
2763 			     */
2764 			    printf("Nfsd openctrl unexpected deleg cnfl\n");
2765 			    free((caddr_t)new_open, M_NFSDSTATE);
2766 			    free((caddr_t)new_deleg, M_NFSDSTATE);
2767 			    if (ret == -1) {
2768 				openstp = NULL;
2769 				goto tryagain;
2770 			    }
2771 			    error = ret;
2772 			    goto out;
2773 			}
2774 		    }
2775 		}
2776 		stp = nstp;
2777 	    }
2778 	}
2779 
2780 	/*
2781 	 * We only get here if there was no open that conflicted.
2782 	 * If an open for the owner exists, or in the access/deny bits.
2783 	 * Otherwise it is a new open. If the open_owner hasn't been
2784 	 * confirmed, replace the open with the new one needing confirmation,
2785 	 * otherwise add the open.
2786 	 */
2787 	if (new_stp->ls_flags & NFSLCK_DELEGPREV) {
2788 	    /*
2789 	     * Handle NFSLCK_DELEGPREV by searching the old delegations for
2790 	     * a match. If found, just move the old delegation to the current
2791 	     * delegation list and issue open. If not found, return
2792 	     * NFSERR_EXPIRED.
2793 	     */
2794 	    LIST_FOREACH(stp, &clp->lc_olddeleg, ls_list) {
2795 		if (stp->ls_lfp == lfp) {
2796 		    /* Found it */
2797 		    if (stp->ls_clp != clp)
2798 			panic("olddeleg clp");
2799 		    LIST_REMOVE(stp, ls_list);
2800 		    LIST_REMOVE(stp, ls_hash);
2801 		    stp->ls_flags &= ~NFSLCK_OLDDELEG;
2802 		    stp->ls_stateid.seqid = delegstateidp->seqid = 1;
2803 		    stp->ls_stateid.other[0] = delegstateidp->other[0] =
2804 			clp->lc_clientid.lval[0];
2805 		    stp->ls_stateid.other[1] = delegstateidp->other[1] =
2806 			clp->lc_clientid.lval[1];
2807 		    stp->ls_stateid.other[2] = delegstateidp->other[2] =
2808 			nfsrv_nextstateindex(clp);
2809 		    stp->ls_compref = nd->nd_compref;
2810 		    LIST_INSERT_HEAD(&clp->lc_deleg, stp, ls_list);
2811 		    LIST_INSERT_HEAD(NFSSTATEHASH(clp,
2812 			stp->ls_stateid), stp, ls_hash);
2813 		    if (stp->ls_flags & NFSLCK_DELEGWRITE)
2814 			*rflagsp |= NFSV4OPEN_WRITEDELEGATE;
2815 		    else
2816 			*rflagsp |= NFSV4OPEN_READDELEGATE;
2817 		    clp->lc_delegtime = NFSD_MONOSEC +
2818 			nfsrv_lease + NFSRV_LEASEDELTA;
2819 
2820 		    /*
2821 		     * Now, do the associated open.
2822 		     */
2823 		    new_open->ls_stateid.seqid = 1;
2824 		    new_open->ls_stateid.other[0] = clp->lc_clientid.lval[0];
2825 		    new_open->ls_stateid.other[1] = clp->lc_clientid.lval[1];
2826 		    new_open->ls_stateid.other[2] = nfsrv_nextstateindex(clp);
2827 		    new_open->ls_flags = (new_stp->ls_flags&NFSLCK_DENYBITS)|
2828 			NFSLCK_OPEN;
2829 		    if (stp->ls_flags & NFSLCK_DELEGWRITE)
2830 			new_open->ls_flags |= (NFSLCK_READACCESS |
2831 			    NFSLCK_WRITEACCESS);
2832 		    else
2833 			new_open->ls_flags |= NFSLCK_READACCESS;
2834 		    new_open->ls_uid = new_stp->ls_uid;
2835 		    new_open->ls_lfp = lfp;
2836 		    new_open->ls_clp = clp;
2837 		    LIST_INIT(&new_open->ls_open);
2838 		    LIST_INSERT_HEAD(&lfp->lf_open, new_open, ls_file);
2839 		    LIST_INSERT_HEAD(NFSSTATEHASH(clp, new_open->ls_stateid),
2840 			new_open, ls_hash);
2841 		    /*
2842 		     * and handle the open owner
2843 		     */
2844 		    if (ownerstp) {
2845 			new_open->ls_openowner = ownerstp;
2846 			LIST_INSERT_HEAD(&ownerstp->ls_open,new_open,ls_list);
2847 		    } else {
2848 			new_open->ls_openowner = new_stp;
2849 			new_stp->ls_flags = 0;
2850 			nfsrvd_refcache(new_stp->ls_op);
2851 			new_stp->ls_noopens = 0;
2852 			LIST_INIT(&new_stp->ls_open);
2853 			LIST_INSERT_HEAD(&new_stp->ls_open, new_open, ls_list);
2854 			LIST_INSERT_HEAD(&clp->lc_open, new_stp, ls_list);
2855 			*new_stpp = NULL;
2856 			nfsstatsv1.srvopenowners++;
2857 			nfsrv_openpluslock++;
2858 		    }
2859 		    openstp = new_open;
2860 		    new_open = NULL;
2861 		    nfsstatsv1.srvopens++;
2862 		    nfsrv_openpluslock++;
2863 		    break;
2864 		}
2865 	    }
2866 	    if (stp == LIST_END(&clp->lc_olddeleg))
2867 		error = NFSERR_EXPIRED;
2868 	} else if (new_stp->ls_flags & (NFSLCK_DELEGREAD | NFSLCK_DELEGWRITE)) {
2869 	    /*
2870 	     * Scan to see that no delegation for this client and file
2871 	     * doesn't already exist.
2872 	     * There also shouldn't yet be an Open for this file and
2873 	     * openowner.
2874 	     */
2875 	    LIST_FOREACH(stp, &lfp->lf_deleg, ls_file) {
2876 		if (stp->ls_clp == clp)
2877 		    break;
2878 	    }
2879 	    if (stp == LIST_END(&lfp->lf_deleg) && openstp == NULL) {
2880 		/*
2881 		 * This is the Claim_Previous case with a delegation
2882 		 * type != Delegate_None.
2883 		 */
2884 		/*
2885 		 * First, add the delegation. (Although we must issue the
2886 		 * delegation, we can also ask for an immediate return.)
2887 		 */
2888 		new_deleg->ls_stateid.seqid = delegstateidp->seqid = 1;
2889 		new_deleg->ls_stateid.other[0] = delegstateidp->other[0] =
2890 		    clp->lc_clientid.lval[0];
2891 		new_deleg->ls_stateid.other[1] = delegstateidp->other[1] =
2892 		    clp->lc_clientid.lval[1];
2893 		new_deleg->ls_stateid.other[2] = delegstateidp->other[2] =
2894 		    nfsrv_nextstateindex(clp);
2895 		if (new_stp->ls_flags & NFSLCK_DELEGWRITE) {
2896 		    new_deleg->ls_flags = (NFSLCK_DELEGWRITE |
2897 			NFSLCK_READACCESS | NFSLCK_WRITEACCESS);
2898 		    *rflagsp |= NFSV4OPEN_WRITEDELEGATE;
2899 		} else {
2900 		    new_deleg->ls_flags = (NFSLCK_DELEGREAD |
2901 			NFSLCK_READACCESS);
2902 		    *rflagsp |= NFSV4OPEN_READDELEGATE;
2903 		}
2904 		new_deleg->ls_uid = new_stp->ls_uid;
2905 		new_deleg->ls_lfp = lfp;
2906 		new_deleg->ls_clp = clp;
2907 		new_deleg->ls_filerev = filerev;
2908 		new_deleg->ls_compref = nd->nd_compref;
2909 		LIST_INSERT_HEAD(&lfp->lf_deleg, new_deleg, ls_file);
2910 		LIST_INSERT_HEAD(NFSSTATEHASH(clp,
2911 		    new_deleg->ls_stateid), new_deleg, ls_hash);
2912 		LIST_INSERT_HEAD(&clp->lc_deleg, new_deleg, ls_list);
2913 		new_deleg = NULL;
2914 		if (delegate == 2 || nfsrv_issuedelegs == 0 ||
2915 		    (clp->lc_flags & (LCL_CALLBACKSON | LCL_CBDOWN)) !=
2916 		     LCL_CALLBACKSON ||
2917 		    NFSRV_V4DELEGLIMIT(nfsrv_delegatecnt) ||
2918 		    !NFSVNO_DELEGOK(vp))
2919 		    *rflagsp |= NFSV4OPEN_RECALL;
2920 		nfsstatsv1.srvdelegates++;
2921 		nfsrv_openpluslock++;
2922 		nfsrv_delegatecnt++;
2923 
2924 		/*
2925 		 * Now, do the associated open.
2926 		 */
2927 		new_open->ls_stateid.seqid = 1;
2928 		new_open->ls_stateid.other[0] = clp->lc_clientid.lval[0];
2929 		new_open->ls_stateid.other[1] = clp->lc_clientid.lval[1];
2930 		new_open->ls_stateid.other[2] = nfsrv_nextstateindex(clp);
2931 		new_open->ls_flags = (new_stp->ls_flags & NFSLCK_DENYBITS) |
2932 		    NFSLCK_OPEN;
2933 		if (new_stp->ls_flags & NFSLCK_DELEGWRITE)
2934 			new_open->ls_flags |= (NFSLCK_READACCESS |
2935 			    NFSLCK_WRITEACCESS);
2936 		else
2937 			new_open->ls_flags |= NFSLCK_READACCESS;
2938 		new_open->ls_uid = new_stp->ls_uid;
2939 		new_open->ls_lfp = lfp;
2940 		new_open->ls_clp = clp;
2941 		LIST_INIT(&new_open->ls_open);
2942 		LIST_INSERT_HEAD(&lfp->lf_open, new_open, ls_file);
2943 		LIST_INSERT_HEAD(NFSSTATEHASH(clp, new_open->ls_stateid),
2944 		   new_open, ls_hash);
2945 		/*
2946 		 * and handle the open owner
2947 		 */
2948 		if (ownerstp) {
2949 		    new_open->ls_openowner = ownerstp;
2950 		    LIST_INSERT_HEAD(&ownerstp->ls_open, new_open, ls_list);
2951 		} else {
2952 		    new_open->ls_openowner = new_stp;
2953 		    new_stp->ls_flags = 0;
2954 		    nfsrvd_refcache(new_stp->ls_op);
2955 		    new_stp->ls_noopens = 0;
2956 		    LIST_INIT(&new_stp->ls_open);
2957 		    LIST_INSERT_HEAD(&new_stp->ls_open, new_open, ls_list);
2958 		    LIST_INSERT_HEAD(&clp->lc_open, new_stp, ls_list);
2959 		    *new_stpp = NULL;
2960 		    nfsstatsv1.srvopenowners++;
2961 		    nfsrv_openpluslock++;
2962 		}
2963 		openstp = new_open;
2964 		new_open = NULL;
2965 		nfsstatsv1.srvopens++;
2966 		nfsrv_openpluslock++;
2967 	    } else {
2968 		error = NFSERR_RECLAIMCONFLICT;
2969 	    }
2970 	} else if (ownerstp) {
2971 		if (ownerstp->ls_flags & NFSLCK_NEEDSCONFIRM) {
2972 		    /* Replace the open */
2973 		    if (ownerstp->ls_op)
2974 			nfsrvd_derefcache(ownerstp->ls_op);
2975 		    ownerstp->ls_op = new_stp->ls_op;
2976 		    nfsrvd_refcache(ownerstp->ls_op);
2977 		    ownerstp->ls_seq = new_stp->ls_seq;
2978 		    *rflagsp |= NFSV4OPEN_RESULTCONFIRM;
2979 		    stp = LIST_FIRST(&ownerstp->ls_open);
2980 		    stp->ls_flags = (new_stp->ls_flags & NFSLCK_SHAREBITS) |
2981 			NFSLCK_OPEN;
2982 		    stp->ls_stateid.seqid = 1;
2983 		    stp->ls_uid = new_stp->ls_uid;
2984 		    if (lfp != stp->ls_lfp) {
2985 			LIST_REMOVE(stp, ls_file);
2986 			LIST_INSERT_HEAD(&lfp->lf_open, stp, ls_file);
2987 			stp->ls_lfp = lfp;
2988 		    }
2989 		    openstp = stp;
2990 		} else if (openstp) {
2991 		    openstp->ls_flags |= (new_stp->ls_flags & NFSLCK_SHAREBITS);
2992 		    openstp->ls_stateid.seqid++;
2993 		    if ((nd->nd_flag & ND_NFSV41) != 0 &&
2994 			openstp->ls_stateid.seqid == 0)
2995 			openstp->ls_stateid.seqid = 1;
2996 
2997 		    /*
2998 		     * This is where we can choose to issue a delegation.
2999 		     */
3000 		    if (delegate == 0 || writedeleg == 0 ||
3001 			NFSVNO_EXRDONLY(exp) || (readonly != 0 &&
3002 			nfsrv_writedelegifpos == 0) ||
3003 			!NFSVNO_DELEGOK(vp) ||
3004 			(new_stp->ls_flags & NFSLCK_WANTRDELEG) != 0 ||
3005 			(clp->lc_flags & (LCL_CALLBACKSON | LCL_CBDOWN)) !=
3006 			 LCL_CALLBACKSON)
3007 			*rflagsp |= NFSV4OPEN_WDCONTENTION;
3008 		    else if (nfsrv_issuedelegs == 0 ||
3009 			NFSRV_V4DELEGLIMIT(nfsrv_delegatecnt))
3010 			*rflagsp |= NFSV4OPEN_WDRESOURCE;
3011 		    else if ((new_stp->ls_flags & NFSLCK_WANTNODELEG) != 0)
3012 			*rflagsp |= NFSV4OPEN_WDNOTWANTED;
3013 		    else {
3014 			new_deleg->ls_stateid.seqid = delegstateidp->seqid = 1;
3015 			new_deleg->ls_stateid.other[0] = delegstateidp->other[0]
3016 			    = clp->lc_clientid.lval[0];
3017 			new_deleg->ls_stateid.other[1] = delegstateidp->other[1]
3018 			    = clp->lc_clientid.lval[1];
3019 			new_deleg->ls_stateid.other[2] = delegstateidp->other[2]
3020 			    = nfsrv_nextstateindex(clp);
3021 			new_deleg->ls_flags = (NFSLCK_DELEGWRITE |
3022 			    NFSLCK_READACCESS | NFSLCK_WRITEACCESS);
3023 			*rflagsp |= NFSV4OPEN_WRITEDELEGATE;
3024 			new_deleg->ls_uid = new_stp->ls_uid;
3025 			new_deleg->ls_lfp = lfp;
3026 			new_deleg->ls_clp = clp;
3027 			new_deleg->ls_filerev = filerev;
3028 			new_deleg->ls_compref = nd->nd_compref;
3029 			LIST_INSERT_HEAD(&lfp->lf_deleg, new_deleg, ls_file);
3030 			LIST_INSERT_HEAD(NFSSTATEHASH(clp,
3031 			    new_deleg->ls_stateid), new_deleg, ls_hash);
3032 			LIST_INSERT_HEAD(&clp->lc_deleg, new_deleg, ls_list);
3033 			new_deleg = NULL;
3034 			nfsstatsv1.srvdelegates++;
3035 			nfsrv_openpluslock++;
3036 			nfsrv_delegatecnt++;
3037 		    }
3038 		} else {
3039 		    new_open->ls_stateid.seqid = 1;
3040 		    new_open->ls_stateid.other[0] = clp->lc_clientid.lval[0];
3041 		    new_open->ls_stateid.other[1] = clp->lc_clientid.lval[1];
3042 		    new_open->ls_stateid.other[2] = nfsrv_nextstateindex(clp);
3043 		    new_open->ls_flags = (new_stp->ls_flags & NFSLCK_SHAREBITS)|
3044 			NFSLCK_OPEN;
3045 		    new_open->ls_uid = new_stp->ls_uid;
3046 		    new_open->ls_openowner = ownerstp;
3047 		    new_open->ls_lfp = lfp;
3048 		    new_open->ls_clp = clp;
3049 		    LIST_INIT(&new_open->ls_open);
3050 		    LIST_INSERT_HEAD(&lfp->lf_open, new_open, ls_file);
3051 		    LIST_INSERT_HEAD(&ownerstp->ls_open, new_open, ls_list);
3052 		    LIST_INSERT_HEAD(NFSSTATEHASH(clp, new_open->ls_stateid),
3053 			new_open, ls_hash);
3054 		    openstp = new_open;
3055 		    new_open = NULL;
3056 		    nfsstatsv1.srvopens++;
3057 		    nfsrv_openpluslock++;
3058 
3059 		    /*
3060 		     * This is where we can choose to issue a delegation.
3061 		     */
3062 		    if (delegate == 0 || (writedeleg == 0 && readonly == 0) ||
3063 			!NFSVNO_DELEGOK(vp) ||
3064 			(clp->lc_flags & (LCL_CALLBACKSON | LCL_CBDOWN)) !=
3065 			 LCL_CALLBACKSON)
3066 			*rflagsp |= NFSV4OPEN_WDCONTENTION;
3067 		    else if (nfsrv_issuedelegs == 0 ||
3068 			NFSRV_V4DELEGLIMIT(nfsrv_delegatecnt))
3069 			*rflagsp |= NFSV4OPEN_WDRESOURCE;
3070 		    else if ((new_stp->ls_flags & NFSLCK_WANTNODELEG) != 0)
3071 			*rflagsp |= NFSV4OPEN_WDNOTWANTED;
3072 		    else {
3073 			new_deleg->ls_stateid.seqid = delegstateidp->seqid = 1;
3074 			new_deleg->ls_stateid.other[0] = delegstateidp->other[0]
3075 			    = clp->lc_clientid.lval[0];
3076 			new_deleg->ls_stateid.other[1] = delegstateidp->other[1]
3077 			    = clp->lc_clientid.lval[1];
3078 			new_deleg->ls_stateid.other[2] = delegstateidp->other[2]
3079 			    = nfsrv_nextstateindex(clp);
3080 			if (writedeleg && !NFSVNO_EXRDONLY(exp) &&
3081 			    (nfsrv_writedelegifpos || !readonly) &&
3082 			    (new_stp->ls_flags & NFSLCK_WANTRDELEG) == 0) {
3083 			    new_deleg->ls_flags = (NFSLCK_DELEGWRITE |
3084 				NFSLCK_READACCESS | NFSLCK_WRITEACCESS);
3085 			    *rflagsp |= NFSV4OPEN_WRITEDELEGATE;
3086 			} else {
3087 			    new_deleg->ls_flags = (NFSLCK_DELEGREAD |
3088 				NFSLCK_READACCESS);
3089 			    *rflagsp |= NFSV4OPEN_READDELEGATE;
3090 			}
3091 			new_deleg->ls_uid = new_stp->ls_uid;
3092 			new_deleg->ls_lfp = lfp;
3093 			new_deleg->ls_clp = clp;
3094 			new_deleg->ls_filerev = filerev;
3095 			new_deleg->ls_compref = nd->nd_compref;
3096 			LIST_INSERT_HEAD(&lfp->lf_deleg, new_deleg, ls_file);
3097 			LIST_INSERT_HEAD(NFSSTATEHASH(clp,
3098 			    new_deleg->ls_stateid), new_deleg, ls_hash);
3099 			LIST_INSERT_HEAD(&clp->lc_deleg, new_deleg, ls_list);
3100 			new_deleg = NULL;
3101 			nfsstatsv1.srvdelegates++;
3102 			nfsrv_openpluslock++;
3103 			nfsrv_delegatecnt++;
3104 		    }
3105 		}
3106 	} else {
3107 		/*
3108 		 * New owner case. Start the open_owner sequence with a
3109 		 * Needs confirmation (unless a reclaim) and hang the
3110 		 * new open off it.
3111 		 */
3112 		new_open->ls_stateid.seqid = 1;
3113 		new_open->ls_stateid.other[0] = clp->lc_clientid.lval[0];
3114 		new_open->ls_stateid.other[1] = clp->lc_clientid.lval[1];
3115 		new_open->ls_stateid.other[2] = nfsrv_nextstateindex(clp);
3116 		new_open->ls_flags = (new_stp->ls_flags & NFSLCK_SHAREBITS) |
3117 		    NFSLCK_OPEN;
3118 		new_open->ls_uid = new_stp->ls_uid;
3119 		LIST_INIT(&new_open->ls_open);
3120 		new_open->ls_openowner = new_stp;
3121 		new_open->ls_lfp = lfp;
3122 		new_open->ls_clp = clp;
3123 		LIST_INSERT_HEAD(&lfp->lf_open, new_open, ls_file);
3124 		if (new_stp->ls_flags & NFSLCK_RECLAIM) {
3125 			new_stp->ls_flags = 0;
3126 		} else if ((nd->nd_flag & ND_NFSV41) != 0) {
3127 			/* NFSv4.1 never needs confirmation. */
3128 			new_stp->ls_flags = 0;
3129 
3130 			/*
3131 			 * This is where we can choose to issue a delegation.
3132 			 */
3133 			if (delegate && nfsrv_issuedelegs &&
3134 			    (writedeleg || readonly) &&
3135 			    (clp->lc_flags & (LCL_CALLBACKSON | LCL_CBDOWN)) ==
3136 			     LCL_CALLBACKSON &&
3137 			    !NFSRV_V4DELEGLIMIT(nfsrv_delegatecnt) &&
3138 			    NFSVNO_DELEGOK(vp) &&
3139 			    ((nd->nd_flag & ND_NFSV41) == 0 ||
3140 			     (new_stp->ls_flags & NFSLCK_WANTNODELEG) == 0)) {
3141 				new_deleg->ls_stateid.seqid =
3142 				    delegstateidp->seqid = 1;
3143 				new_deleg->ls_stateid.other[0] =
3144 				    delegstateidp->other[0]
3145 				    = clp->lc_clientid.lval[0];
3146 				new_deleg->ls_stateid.other[1] =
3147 				    delegstateidp->other[1]
3148 				    = clp->lc_clientid.lval[1];
3149 				new_deleg->ls_stateid.other[2] =
3150 				    delegstateidp->other[2]
3151 				    = nfsrv_nextstateindex(clp);
3152 				if (writedeleg && !NFSVNO_EXRDONLY(exp) &&
3153 				    (nfsrv_writedelegifpos || !readonly) &&
3154 				    ((nd->nd_flag & ND_NFSV41) == 0 ||
3155 				     (new_stp->ls_flags & NFSLCK_WANTRDELEG) ==
3156 				     0)) {
3157 					new_deleg->ls_flags =
3158 					    (NFSLCK_DELEGWRITE |
3159 					     NFSLCK_READACCESS |
3160 					     NFSLCK_WRITEACCESS);
3161 					*rflagsp |= NFSV4OPEN_WRITEDELEGATE;
3162 				} else {
3163 					new_deleg->ls_flags =
3164 					    (NFSLCK_DELEGREAD |
3165 					     NFSLCK_READACCESS);
3166 					*rflagsp |= NFSV4OPEN_READDELEGATE;
3167 				}
3168 				new_deleg->ls_uid = new_stp->ls_uid;
3169 				new_deleg->ls_lfp = lfp;
3170 				new_deleg->ls_clp = clp;
3171 				new_deleg->ls_filerev = filerev;
3172 				new_deleg->ls_compref = nd->nd_compref;
3173 				LIST_INSERT_HEAD(&lfp->lf_deleg, new_deleg,
3174 				    ls_file);
3175 				LIST_INSERT_HEAD(NFSSTATEHASH(clp,
3176 				    new_deleg->ls_stateid), new_deleg, ls_hash);
3177 				LIST_INSERT_HEAD(&clp->lc_deleg, new_deleg,
3178 				    ls_list);
3179 				new_deleg = NULL;
3180 				nfsstatsv1.srvdelegates++;
3181 				nfsrv_openpluslock++;
3182 				nfsrv_delegatecnt++;
3183 			}
3184 		} else {
3185 			*rflagsp |= NFSV4OPEN_RESULTCONFIRM;
3186 			new_stp->ls_flags = NFSLCK_NEEDSCONFIRM;
3187 		}
3188 		nfsrvd_refcache(new_stp->ls_op);
3189 		new_stp->ls_noopens = 0;
3190 		LIST_INIT(&new_stp->ls_open);
3191 		LIST_INSERT_HEAD(&new_stp->ls_open, new_open, ls_list);
3192 		LIST_INSERT_HEAD(&clp->lc_open, new_stp, ls_list);
3193 		LIST_INSERT_HEAD(NFSSTATEHASH(clp, new_open->ls_stateid),
3194 		    new_open, ls_hash);
3195 		openstp = new_open;
3196 		new_open = NULL;
3197 		*new_stpp = NULL;
3198 		nfsstatsv1.srvopens++;
3199 		nfsrv_openpluslock++;
3200 		nfsstatsv1.srvopenowners++;
3201 		nfsrv_openpluslock++;
3202 	}
3203 	if (!error) {
3204 		stateidp->seqid = openstp->ls_stateid.seqid;
3205 		stateidp->other[0] = openstp->ls_stateid.other[0];
3206 		stateidp->other[1] = openstp->ls_stateid.other[1];
3207 		stateidp->other[2] = openstp->ls_stateid.other[2];
3208 	}
3209 	NFSUNLOCKSTATE();
3210 	if (haslock) {
3211 		NFSLOCKV4ROOTMUTEX();
3212 		nfsv4_unlock(&nfsv4rootfs_lock, 1);
3213 		NFSUNLOCKV4ROOTMUTEX();
3214 	}
3215 	if (new_open)
3216 		FREE((caddr_t)new_open, M_NFSDSTATE);
3217 	if (new_deleg)
3218 		FREE((caddr_t)new_deleg, M_NFSDSTATE);
3219 
3220 out:
3221 	NFSEXITCODE2(error, nd);
3222 	return (error);
3223 }
3224 
3225 /*
3226  * Open update. Does the confirm, downgrade and close.
3227  */
3228 APPLESTATIC int
3229 nfsrv_openupdate(vnode_t vp, struct nfsstate *new_stp, nfsquad_t clientid,
3230     nfsv4stateid_t *stateidp, struct nfsrv_descript *nd, NFSPROC_T *p)
3231 {
3232 	struct nfsstate *stp, *ownerstp;
3233 	struct nfsclient *clp;
3234 	struct nfslockfile *lfp;
3235 	u_int32_t bits;
3236 	int error = 0, gotstate = 0, len = 0;
3237 	u_char client[NFSV4_OPAQUELIMIT];
3238 
3239 	/*
3240 	 * Check for restart conditions (client and server).
3241 	 */
3242 	error = nfsrv_checkrestart(clientid, new_stp->ls_flags,
3243 	    &new_stp->ls_stateid, 0);
3244 	if (error)
3245 		goto out;
3246 
3247 	NFSLOCKSTATE();
3248 	/*
3249 	 * Get the open structure via clientid and stateid.
3250 	 */
3251 	error = nfsrv_getclient(clientid, CLOPS_RENEW, &clp, NULL,
3252 	    (nfsquad_t)((u_quad_t)0), 0, nd, p);
3253 	if (!error)
3254 		error = nfsrv_getstate(clp, &new_stp->ls_stateid,
3255 		    new_stp->ls_flags, &stp);
3256 
3257 	/*
3258 	 * Sanity check the open.
3259 	 */
3260 	if (!error && (!(stp->ls_flags & NFSLCK_OPEN) ||
3261 		(!(new_stp->ls_flags & NFSLCK_CONFIRM) &&
3262 		 (stp->ls_openowner->ls_flags & NFSLCK_NEEDSCONFIRM)) ||
3263 		((new_stp->ls_flags & NFSLCK_CONFIRM) &&
3264 		 (!(stp->ls_openowner->ls_flags & NFSLCK_NEEDSCONFIRM)))))
3265 		error = NFSERR_BADSTATEID;
3266 
3267 	if (!error)
3268 		error = nfsrv_checkseqid(nd, new_stp->ls_seq,
3269 		    stp->ls_openowner, new_stp->ls_op);
3270 	if (!error && stp->ls_stateid.seqid != new_stp->ls_stateid.seqid &&
3271 	    (((nd->nd_flag & ND_NFSV41) == 0 &&
3272 	      !(new_stp->ls_flags & NFSLCK_CONFIRM)) ||
3273 	     ((nd->nd_flag & ND_NFSV41) != 0 &&
3274 	      new_stp->ls_stateid.seqid != 0)))
3275 		error = NFSERR_OLDSTATEID;
3276 	if (!error && vnode_vtype(vp) != VREG) {
3277 		if (vnode_vtype(vp) == VDIR)
3278 			error = NFSERR_ISDIR;
3279 		else
3280 			error = NFSERR_INVAL;
3281 	}
3282 
3283 	if (error) {
3284 		/*
3285 		 * If a client tries to confirm an Open with a bad
3286 		 * seqid# and there are no byte range locks or other Opens
3287 		 * on the openowner, just throw it away, so the next use of the
3288 		 * openowner will start a fresh seq#.
3289 		 */
3290 		if (error == NFSERR_BADSEQID &&
3291 		    (new_stp->ls_flags & NFSLCK_CONFIRM) &&
3292 		    nfsrv_nootherstate(stp))
3293 			nfsrv_freeopenowner(stp->ls_openowner, 0, p);
3294 		NFSUNLOCKSTATE();
3295 		goto out;
3296 	}
3297 
3298 	/*
3299 	 * Set the return stateid.
3300 	 */
3301 	stateidp->seqid = stp->ls_stateid.seqid + 1;
3302 	if ((nd->nd_flag & ND_NFSV41) != 0 && stateidp->seqid == 0)
3303 		stateidp->seqid = 1;
3304 	stateidp->other[0] = stp->ls_stateid.other[0];
3305 	stateidp->other[1] = stp->ls_stateid.other[1];
3306 	stateidp->other[2] = stp->ls_stateid.other[2];
3307 	/*
3308 	 * Now, handle the three cases.
3309 	 */
3310 	if (new_stp->ls_flags & NFSLCK_CONFIRM) {
3311 		/*
3312 		 * If the open doesn't need confirmation, it seems to me that
3313 		 * there is a client error, but I'll just log it and keep going?
3314 		 */
3315 		if (!(stp->ls_openowner->ls_flags & NFSLCK_NEEDSCONFIRM))
3316 			printf("Nfsv4d: stray open confirm\n");
3317 		stp->ls_openowner->ls_flags = 0;
3318 		stp->ls_stateid.seqid++;
3319 		if ((nd->nd_flag & ND_NFSV41) != 0 &&
3320 		    stp->ls_stateid.seqid == 0)
3321 			stp->ls_stateid.seqid = 1;
3322 		if (!(clp->lc_flags & LCL_STAMPEDSTABLE)) {
3323 			clp->lc_flags |= LCL_STAMPEDSTABLE;
3324 			len = clp->lc_idlen;
3325 			NFSBCOPY(clp->lc_id, client, len);
3326 			gotstate = 1;
3327 		}
3328 		NFSUNLOCKSTATE();
3329 	} else if (new_stp->ls_flags & NFSLCK_CLOSE) {
3330 		ownerstp = stp->ls_openowner;
3331 		lfp = stp->ls_lfp;
3332 		if (nfsrv_dolocallocks != 0 && !LIST_EMPTY(&stp->ls_open)) {
3333 			/* Get the lf lock */
3334 			nfsrv_locklf(lfp);
3335 			NFSUNLOCKSTATE();
3336 			ASSERT_VOP_ELOCKED(vp, "nfsrv_openupdate");
3337 			NFSVOPUNLOCK(vp, 0);
3338 			if (nfsrv_freeopen(stp, vp, 1, p) == 0) {
3339 				NFSLOCKSTATE();
3340 				nfsrv_unlocklf(lfp);
3341 				NFSUNLOCKSTATE();
3342 			}
3343 			NFSVOPLOCK(vp, LK_EXCLUSIVE | LK_RETRY);
3344 		} else {
3345 			(void) nfsrv_freeopen(stp, NULL, 0, p);
3346 			NFSUNLOCKSTATE();
3347 		}
3348 	} else {
3349 		/*
3350 		 * Update the share bits, making sure that the new set are a
3351 		 * subset of the old ones.
3352 		 */
3353 		bits = (new_stp->ls_flags & NFSLCK_SHAREBITS);
3354 		if (~(stp->ls_flags) & bits) {
3355 			NFSUNLOCKSTATE();
3356 			error = NFSERR_INVAL;
3357 			goto out;
3358 		}
3359 		stp->ls_flags = (bits | NFSLCK_OPEN);
3360 		stp->ls_stateid.seqid++;
3361 		if ((nd->nd_flag & ND_NFSV41) != 0 &&
3362 		    stp->ls_stateid.seqid == 0)
3363 			stp->ls_stateid.seqid = 1;
3364 		NFSUNLOCKSTATE();
3365 	}
3366 
3367 	/*
3368 	 * If the client just confirmed its first open, write a timestamp
3369 	 * to the stable storage file.
3370 	 */
3371 	if (gotstate != 0) {
3372 		nfsrv_writestable(client, len, NFSNST_NEWSTATE, p);
3373 		nfsrv_backupstable();
3374 	}
3375 
3376 out:
3377 	NFSEXITCODE2(error, nd);
3378 	return (error);
3379 }
3380 
3381 /*
3382  * Delegation update. Does the purge and return.
3383  */
3384 APPLESTATIC int
3385 nfsrv_delegupdate(struct nfsrv_descript *nd, nfsquad_t clientid,
3386     nfsv4stateid_t *stateidp, vnode_t vp, int op, struct ucred *cred,
3387     NFSPROC_T *p)
3388 {
3389 	struct nfsstate *stp;
3390 	struct nfsclient *clp;
3391 	int error = 0;
3392 	fhandle_t fh;
3393 
3394 	/*
3395 	 * Do a sanity check against the file handle for DelegReturn.
3396 	 */
3397 	if (vp) {
3398 		error = nfsvno_getfh(vp, &fh, p);
3399 		if (error)
3400 			goto out;
3401 	}
3402 	/*
3403 	 * Check for restart conditions (client and server).
3404 	 */
3405 	if (op == NFSV4OP_DELEGRETURN)
3406 		error = nfsrv_checkrestart(clientid, NFSLCK_DELEGRETURN,
3407 			stateidp, 0);
3408 	else
3409 		error = nfsrv_checkrestart(clientid, NFSLCK_DELEGPURGE,
3410 			stateidp, 0);
3411 
3412 	NFSLOCKSTATE();
3413 	/*
3414 	 * Get the open structure via clientid and stateid.
3415 	 */
3416 	if (!error)
3417 	    error = nfsrv_getclient(clientid, CLOPS_RENEW, &clp, NULL,
3418 		(nfsquad_t)((u_quad_t)0), 0, nd, p);
3419 	if (error) {
3420 		if (error == NFSERR_CBPATHDOWN)
3421 			error = 0;
3422 		if (error == NFSERR_STALECLIENTID && op == NFSV4OP_DELEGRETURN)
3423 			error = NFSERR_STALESTATEID;
3424 	}
3425 	if (!error && op == NFSV4OP_DELEGRETURN) {
3426 	    error = nfsrv_getstate(clp, stateidp, NFSLCK_DELEGRETURN, &stp);
3427 	    if (!error && stp->ls_stateid.seqid != stateidp->seqid &&
3428 		((nd->nd_flag & ND_NFSV41) == 0 || stateidp->seqid != 0))
3429 		error = NFSERR_OLDSTATEID;
3430 	}
3431 	/*
3432 	 * NFSERR_EXPIRED means that the state has gone away,
3433 	 * so Delegations have been purged. Just return ok.
3434 	 */
3435 	if (error == NFSERR_EXPIRED && op == NFSV4OP_DELEGPURGE) {
3436 		NFSUNLOCKSTATE();
3437 		error = 0;
3438 		goto out;
3439 	}
3440 	if (error) {
3441 		NFSUNLOCKSTATE();
3442 		goto out;
3443 	}
3444 
3445 	if (op == NFSV4OP_DELEGRETURN) {
3446 		if (NFSBCMP((caddr_t)&fh, (caddr_t)&stp->ls_lfp->lf_fh,
3447 		    sizeof (fhandle_t))) {
3448 			NFSUNLOCKSTATE();
3449 			error = NFSERR_BADSTATEID;
3450 			goto out;
3451 		}
3452 		nfsrv_freedeleg(stp);
3453 	} else {
3454 		nfsrv_freedeleglist(&clp->lc_olddeleg);
3455 	}
3456 	NFSUNLOCKSTATE();
3457 	error = 0;
3458 
3459 out:
3460 	NFSEXITCODE(error);
3461 	return (error);
3462 }
3463 
3464 /*
3465  * Release lock owner.
3466  */
3467 APPLESTATIC int
3468 nfsrv_releaselckown(struct nfsstate *new_stp, nfsquad_t clientid,
3469     NFSPROC_T *p)
3470 {
3471 	struct nfsstate *stp, *nstp, *openstp, *ownstp;
3472 	struct nfsclient *clp;
3473 	int error = 0;
3474 
3475 	/*
3476 	 * Check for restart conditions (client and server).
3477 	 */
3478 	error = nfsrv_checkrestart(clientid, new_stp->ls_flags,
3479 	    &new_stp->ls_stateid, 0);
3480 	if (error)
3481 		goto out;
3482 
3483 	NFSLOCKSTATE();
3484 	/*
3485 	 * Get the lock owner by name.
3486 	 */
3487 	error = nfsrv_getclient(clientid, CLOPS_RENEW, &clp, NULL,
3488 	    (nfsquad_t)((u_quad_t)0), 0, NULL, p);
3489 	if (error) {
3490 		NFSUNLOCKSTATE();
3491 		goto out;
3492 	}
3493 	LIST_FOREACH(ownstp, &clp->lc_open, ls_list) {
3494 	    LIST_FOREACH(openstp, &ownstp->ls_open, ls_list) {
3495 		stp = LIST_FIRST(&openstp->ls_open);
3496 		while (stp != LIST_END(&openstp->ls_open)) {
3497 		    nstp = LIST_NEXT(stp, ls_list);
3498 		    /*
3499 		     * If the owner matches, check for locks and
3500 		     * then free or return an error.
3501 		     */
3502 		    if (stp->ls_ownerlen == new_stp->ls_ownerlen &&
3503 			!NFSBCMP(stp->ls_owner, new_stp->ls_owner,
3504 			 stp->ls_ownerlen)){
3505 			if (LIST_EMPTY(&stp->ls_lock)) {
3506 			    nfsrv_freelockowner(stp, NULL, 0, p);
3507 			} else {
3508 			    NFSUNLOCKSTATE();
3509 			    error = NFSERR_LOCKSHELD;
3510 			    goto out;
3511 			}
3512 		    }
3513 		    stp = nstp;
3514 		}
3515 	    }
3516 	}
3517 	NFSUNLOCKSTATE();
3518 
3519 out:
3520 	NFSEXITCODE(error);
3521 	return (error);
3522 }
3523 
3524 /*
3525  * Get the file handle for a lock structure.
3526  */
3527 static int
3528 nfsrv_getlockfh(vnode_t vp, u_short flags, struct nfslockfile *new_lfp,
3529     fhandle_t *nfhp, NFSPROC_T *p)
3530 {
3531 	fhandle_t *fhp = NULL;
3532 	int error;
3533 
3534 	/*
3535 	 * For lock, use the new nfslock structure, otherwise just
3536 	 * a fhandle_t on the stack.
3537 	 */
3538 	if (flags & NFSLCK_OPEN) {
3539 		KASSERT(new_lfp != NULL, ("nfsrv_getlockfh: new_lfp NULL"));
3540 		fhp = &new_lfp->lf_fh;
3541 	} else if (nfhp) {
3542 		fhp = nfhp;
3543 	} else {
3544 		panic("nfsrv_getlockfh");
3545 	}
3546 	error = nfsvno_getfh(vp, fhp, p);
3547 	NFSEXITCODE(error);
3548 	return (error);
3549 }
3550 
3551 /*
3552  * Get an nfs lock structure. Allocate one, as required, and return a
3553  * pointer to it.
3554  * Returns an NFSERR_xxx upon failure or -1 to indicate no current lock.
3555  */
3556 static int
3557 nfsrv_getlockfile(u_short flags, struct nfslockfile **new_lfpp,
3558     struct nfslockfile **lfpp, fhandle_t *nfhp, int lockit)
3559 {
3560 	struct nfslockfile *lfp;
3561 	fhandle_t *fhp = NULL, *tfhp;
3562 	struct nfslockhashhead *hp;
3563 	struct nfslockfile *new_lfp = NULL;
3564 
3565 	/*
3566 	 * For lock, use the new nfslock structure, otherwise just
3567 	 * a fhandle_t on the stack.
3568 	 */
3569 	if (flags & NFSLCK_OPEN) {
3570 		new_lfp = *new_lfpp;
3571 		fhp = &new_lfp->lf_fh;
3572 	} else if (nfhp) {
3573 		fhp = nfhp;
3574 	} else {
3575 		panic("nfsrv_getlockfile");
3576 	}
3577 
3578 	hp = NFSLOCKHASH(fhp);
3579 	LIST_FOREACH(lfp, hp, lf_hash) {
3580 		tfhp = &lfp->lf_fh;
3581 		if (NFSVNO_CMPFH(fhp, tfhp)) {
3582 			if (lockit)
3583 				nfsrv_locklf(lfp);
3584 			*lfpp = lfp;
3585 			return (0);
3586 		}
3587 	}
3588 	if (!(flags & NFSLCK_OPEN))
3589 		return (-1);
3590 
3591 	/*
3592 	 * No match, so chain the new one into the list.
3593 	 */
3594 	LIST_INIT(&new_lfp->lf_open);
3595 	LIST_INIT(&new_lfp->lf_lock);
3596 	LIST_INIT(&new_lfp->lf_deleg);
3597 	LIST_INIT(&new_lfp->lf_locallock);
3598 	LIST_INIT(&new_lfp->lf_rollback);
3599 	new_lfp->lf_locallock_lck.nfslock_usecnt = 0;
3600 	new_lfp->lf_locallock_lck.nfslock_lock = 0;
3601 	new_lfp->lf_usecount = 0;
3602 	LIST_INSERT_HEAD(hp, new_lfp, lf_hash);
3603 	*lfpp = new_lfp;
3604 	*new_lfpp = NULL;
3605 	return (0);
3606 }
3607 
3608 /*
3609  * This function adds a nfslock lock structure to the list for the associated
3610  * nfsstate and nfslockfile structures. It will be inserted after the
3611  * entry pointed at by insert_lop.
3612  */
3613 static void
3614 nfsrv_insertlock(struct nfslock *new_lop, struct nfslock *insert_lop,
3615     struct nfsstate *stp, struct nfslockfile *lfp)
3616 {
3617 	struct nfslock *lop, *nlop;
3618 
3619 	new_lop->lo_stp = stp;
3620 	new_lop->lo_lfp = lfp;
3621 
3622 	if (stp != NULL) {
3623 		/* Insert in increasing lo_first order */
3624 		lop = LIST_FIRST(&lfp->lf_lock);
3625 		if (lop == LIST_END(&lfp->lf_lock) ||
3626 		    new_lop->lo_first <= lop->lo_first) {
3627 			LIST_INSERT_HEAD(&lfp->lf_lock, new_lop, lo_lckfile);
3628 		} else {
3629 			nlop = LIST_NEXT(lop, lo_lckfile);
3630 			while (nlop != LIST_END(&lfp->lf_lock) &&
3631 			       nlop->lo_first < new_lop->lo_first) {
3632 				lop = nlop;
3633 				nlop = LIST_NEXT(lop, lo_lckfile);
3634 			}
3635 			LIST_INSERT_AFTER(lop, new_lop, lo_lckfile);
3636 		}
3637 	} else {
3638 		new_lop->lo_lckfile.le_prev = NULL;	/* list not used */
3639 	}
3640 
3641 	/*
3642 	 * Insert after insert_lop, which is overloaded as stp or lfp for
3643 	 * an empty list.
3644 	 */
3645 	if (stp == NULL && (struct nfslockfile *)insert_lop == lfp)
3646 		LIST_INSERT_HEAD(&lfp->lf_locallock, new_lop, lo_lckowner);
3647 	else if ((struct nfsstate *)insert_lop == stp)
3648 		LIST_INSERT_HEAD(&stp->ls_lock, new_lop, lo_lckowner);
3649 	else
3650 		LIST_INSERT_AFTER(insert_lop, new_lop, lo_lckowner);
3651 	if (stp != NULL) {
3652 		nfsstatsv1.srvlocks++;
3653 		nfsrv_openpluslock++;
3654 	}
3655 }
3656 
3657 /*
3658  * This function updates the locking for a lock owner and given file. It
3659  * maintains a list of lock ranges ordered on increasing file offset that
3660  * are NFSLCK_READ or NFSLCK_WRITE and non-overlapping (aka POSIX style).
3661  * It always adds new_lop to the list and sometimes uses the one pointed
3662  * at by other_lopp.
3663  */
3664 static void
3665 nfsrv_updatelock(struct nfsstate *stp, struct nfslock **new_lopp,
3666     struct nfslock **other_lopp, struct nfslockfile *lfp)
3667 {
3668 	struct nfslock *new_lop = *new_lopp;
3669 	struct nfslock *lop, *tlop, *ilop;
3670 	struct nfslock *other_lop = *other_lopp;
3671 	int unlock = 0, myfile = 0;
3672 	u_int64_t tmp;
3673 
3674 	/*
3675 	 * Work down the list until the lock is merged.
3676 	 */
3677 	if (new_lop->lo_flags & NFSLCK_UNLOCK)
3678 		unlock = 1;
3679 	if (stp != NULL) {
3680 		ilop = (struct nfslock *)stp;
3681 		lop = LIST_FIRST(&stp->ls_lock);
3682 	} else {
3683 		ilop = (struct nfslock *)lfp;
3684 		lop = LIST_FIRST(&lfp->lf_locallock);
3685 	}
3686 	while (lop != NULL) {
3687 	    /*
3688 	     * Only check locks for this file that aren't before the start of
3689 	     * new lock's range.
3690 	     */
3691 	    if (lop->lo_lfp == lfp) {
3692 	      myfile = 1;
3693 	      if (lop->lo_end >= new_lop->lo_first) {
3694 		if (new_lop->lo_end < lop->lo_first) {
3695 			/*
3696 			 * If the new lock ends before the start of the
3697 			 * current lock's range, no merge, just insert
3698 			 * the new lock.
3699 			 */
3700 			break;
3701 		}
3702 		if (new_lop->lo_flags == lop->lo_flags ||
3703 		    (new_lop->lo_first <= lop->lo_first &&
3704 		     new_lop->lo_end >= lop->lo_end)) {
3705 			/*
3706 			 * This lock can be absorbed by the new lock/unlock.
3707 			 * This happens when it covers the entire range
3708 			 * of the old lock or is contiguous
3709 			 * with the old lock and is of the same type or an
3710 			 * unlock.
3711 			 */
3712 			if (lop->lo_first < new_lop->lo_first)
3713 				new_lop->lo_first = lop->lo_first;
3714 			if (lop->lo_end > new_lop->lo_end)
3715 				new_lop->lo_end = lop->lo_end;
3716 			tlop = lop;
3717 			lop = LIST_NEXT(lop, lo_lckowner);
3718 			nfsrv_freenfslock(tlop);
3719 			continue;
3720 		}
3721 
3722 		/*
3723 		 * All these cases are for contiguous locks that are not the
3724 		 * same type, so they can't be merged.
3725 		 */
3726 		if (new_lop->lo_first <= lop->lo_first) {
3727 			/*
3728 			 * This case is where the new lock overlaps with the
3729 			 * first part of the old lock. Move the start of the
3730 			 * old lock to just past the end of the new lock. The
3731 			 * new lock will be inserted in front of the old, since
3732 			 * ilop hasn't been updated. (We are done now.)
3733 			 */
3734 			lop->lo_first = new_lop->lo_end;
3735 			break;
3736 		}
3737 		if (new_lop->lo_end >= lop->lo_end) {
3738 			/*
3739 			 * This case is where the new lock overlaps with the
3740 			 * end of the old lock's range. Move the old lock's
3741 			 * end to just before the new lock's first and insert
3742 			 * the new lock after the old lock.
3743 			 * Might not be done yet, since the new lock could
3744 			 * overlap further locks with higher ranges.
3745 			 */
3746 			lop->lo_end = new_lop->lo_first;
3747 			ilop = lop;
3748 			lop = LIST_NEXT(lop, lo_lckowner);
3749 			continue;
3750 		}
3751 		/*
3752 		 * The final case is where the new lock's range is in the
3753 		 * middle of the current lock's and splits the current lock
3754 		 * up. Use *other_lopp to handle the second part of the
3755 		 * split old lock range. (We are done now.)
3756 		 * For unlock, we use new_lop as other_lop and tmp, since
3757 		 * other_lop and new_lop are the same for this case.
3758 		 * We noted the unlock case above, so we don't need
3759 		 * new_lop->lo_flags any longer.
3760 		 */
3761 		tmp = new_lop->lo_first;
3762 		if (other_lop == NULL) {
3763 			if (!unlock)
3764 				panic("nfsd srv update unlock");
3765 			other_lop = new_lop;
3766 			*new_lopp = NULL;
3767 		}
3768 		other_lop->lo_first = new_lop->lo_end;
3769 		other_lop->lo_end = lop->lo_end;
3770 		other_lop->lo_flags = lop->lo_flags;
3771 		other_lop->lo_stp = stp;
3772 		other_lop->lo_lfp = lfp;
3773 		lop->lo_end = tmp;
3774 		nfsrv_insertlock(other_lop, lop, stp, lfp);
3775 		*other_lopp = NULL;
3776 		ilop = lop;
3777 		break;
3778 	      }
3779 	    }
3780 	    ilop = lop;
3781 	    lop = LIST_NEXT(lop, lo_lckowner);
3782 	    if (myfile && (lop == NULL || lop->lo_lfp != lfp))
3783 		break;
3784 	}
3785 
3786 	/*
3787 	 * Insert the new lock in the list at the appropriate place.
3788 	 */
3789 	if (!unlock) {
3790 		nfsrv_insertlock(new_lop, ilop, stp, lfp);
3791 		*new_lopp = NULL;
3792 	}
3793 }
3794 
3795 /*
3796  * This function handles sequencing of locks, etc.
3797  * It returns an error that indicates what the caller should do.
3798  */
3799 static int
3800 nfsrv_checkseqid(struct nfsrv_descript *nd, u_int32_t seqid,
3801     struct nfsstate *stp, struct nfsrvcache *op)
3802 {
3803 	int error = 0;
3804 
3805 	if ((nd->nd_flag & ND_NFSV41) != 0)
3806 		/* NFSv4.1 ignores the open_seqid and lock_seqid. */
3807 		goto out;
3808 	if (op != nd->nd_rp)
3809 		panic("nfsrvstate checkseqid");
3810 	if (!(op->rc_flag & RC_INPROG))
3811 		panic("nfsrvstate not inprog");
3812 	if (stp->ls_op && stp->ls_op->rc_refcnt <= 0) {
3813 		printf("refcnt=%d\n", stp->ls_op->rc_refcnt);
3814 		panic("nfsrvstate op refcnt");
3815 	}
3816 	if ((stp->ls_seq + 1) == seqid) {
3817 		if (stp->ls_op)
3818 			nfsrvd_derefcache(stp->ls_op);
3819 		stp->ls_op = op;
3820 		nfsrvd_refcache(op);
3821 		stp->ls_seq = seqid;
3822 		goto out;
3823 	} else if (stp->ls_seq == seqid && stp->ls_op &&
3824 		op->rc_xid == stp->ls_op->rc_xid &&
3825 		op->rc_refcnt == 0 &&
3826 		op->rc_reqlen == stp->ls_op->rc_reqlen &&
3827 		op->rc_cksum == stp->ls_op->rc_cksum) {
3828 		if (stp->ls_op->rc_flag & RC_INPROG) {
3829 			error = NFSERR_DONTREPLY;
3830 			goto out;
3831 		}
3832 		nd->nd_rp = stp->ls_op;
3833 		nd->nd_rp->rc_flag |= RC_INPROG;
3834 		nfsrvd_delcache(op);
3835 		error = NFSERR_REPLYFROMCACHE;
3836 		goto out;
3837 	}
3838 	error = NFSERR_BADSEQID;
3839 
3840 out:
3841 	NFSEXITCODE2(error, nd);
3842 	return (error);
3843 }
3844 
3845 /*
3846  * Get the client ip address for callbacks. If the strings can't be parsed,
3847  * just set lc_program to 0 to indicate no callbacks are possible.
3848  * (For cases where the address can't be parsed or is 0.0.0.0.0.0, set
3849  *  the address to the client's transport address. This won't be used
3850  *  for callbacks, but can be printed out by nfsstats for info.)
3851  * Return error if the xdr can't be parsed, 0 otherwise.
3852  */
3853 APPLESTATIC int
3854 nfsrv_getclientipaddr(struct nfsrv_descript *nd, struct nfsclient *clp)
3855 {
3856 	u_int32_t *tl;
3857 	u_char *cp, *cp2;
3858 	int i, j;
3859 	struct sockaddr_in *rad, *sad;
3860 	u_char protocol[5], addr[24];
3861 	int error = 0, cantparse = 0;
3862 	union {
3863 		u_long ival;
3864 		u_char cval[4];
3865 	} ip;
3866 	union {
3867 		u_short sval;
3868 		u_char cval[2];
3869 	} port;
3870 
3871 	rad = NFSSOCKADDR(clp->lc_req.nr_nam, struct sockaddr_in *);
3872 	rad->sin_family = AF_INET;
3873 	rad->sin_len = sizeof (struct sockaddr_in);
3874 	rad->sin_addr.s_addr = 0;
3875 	rad->sin_port = 0;
3876 	clp->lc_req.nr_client = NULL;
3877 	clp->lc_req.nr_lock = 0;
3878 	NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
3879 	i = fxdr_unsigned(int, *tl);
3880 	if (i >= 3 && i <= 4) {
3881 		error = nfsrv_mtostr(nd, protocol, i);
3882 		if (error)
3883 			goto nfsmout;
3884 		if (!strcmp(protocol, "tcp")) {
3885 			clp->lc_flags |= LCL_TCPCALLBACK;
3886 			clp->lc_req.nr_sotype = SOCK_STREAM;
3887 			clp->lc_req.nr_soproto = IPPROTO_TCP;
3888 		} else if (!strcmp(protocol, "udp")) {
3889 			clp->lc_req.nr_sotype = SOCK_DGRAM;
3890 			clp->lc_req.nr_soproto = IPPROTO_UDP;
3891 		} else {
3892 			cantparse = 1;
3893 		}
3894 	} else {
3895 		cantparse = 1;
3896 		if (i > 0) {
3897 			error = nfsm_advance(nd, NFSM_RNDUP(i), -1);
3898 			if (error)
3899 				goto nfsmout;
3900 		}
3901 	}
3902 	NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
3903 	i = fxdr_unsigned(int, *tl);
3904 	if (i < 0) {
3905 		error = NFSERR_BADXDR;
3906 		goto nfsmout;
3907 	} else if (i == 0) {
3908 		cantparse = 1;
3909 	} else if (!cantparse && i <= 23 && i >= 11) {
3910 		error = nfsrv_mtostr(nd, addr, i);
3911 		if (error)
3912 			goto nfsmout;
3913 
3914 		/*
3915 		 * Parse out the address fields. We expect 6 decimal numbers
3916 		 * separated by '.'s.
3917 		 */
3918 		cp = addr;
3919 		i = 0;
3920 		while (*cp && i < 6) {
3921 			cp2 = cp;
3922 			while (*cp2 && *cp2 != '.')
3923 				cp2++;
3924 			if (*cp2)
3925 				*cp2++ = '\0';
3926 			else if (i != 5) {
3927 				cantparse = 1;
3928 				break;
3929 			}
3930 			j = nfsrv_getipnumber(cp);
3931 			if (j >= 0) {
3932 				if (i < 4)
3933 					ip.cval[3 - i] = j;
3934 				else
3935 					port.cval[5 - i] = j;
3936 			} else {
3937 				cantparse = 1;
3938 				break;
3939 			}
3940 			cp = cp2;
3941 			i++;
3942 		}
3943 		if (!cantparse) {
3944 			if (ip.ival != 0x0) {
3945 				rad->sin_addr.s_addr = htonl(ip.ival);
3946 				rad->sin_port = htons(port.sval);
3947 			} else {
3948 				cantparse = 1;
3949 			}
3950 		}
3951 	} else {
3952 		cantparse = 1;
3953 		if (i > 0) {
3954 			error = nfsm_advance(nd, NFSM_RNDUP(i), -1);
3955 			if (error)
3956 				goto nfsmout;
3957 		}
3958 	}
3959 	if (cantparse) {
3960 		sad = NFSSOCKADDR(nd->nd_nam, struct sockaddr_in *);
3961 		rad->sin_addr.s_addr = sad->sin_addr.s_addr;
3962 		rad->sin_port = 0x0;
3963 		clp->lc_program = 0;
3964 	}
3965 nfsmout:
3966 	NFSEXITCODE2(error, nd);
3967 	return (error);
3968 }
3969 
3970 /*
3971  * Turn a string of up to three decimal digits into a number. Return -1 upon
3972  * error.
3973  */
3974 static int
3975 nfsrv_getipnumber(u_char *cp)
3976 {
3977 	int i = 0, j = 0;
3978 
3979 	while (*cp) {
3980 		if (j > 2 || *cp < '0' || *cp > '9')
3981 			return (-1);
3982 		i *= 10;
3983 		i += (*cp - '0');
3984 		cp++;
3985 		j++;
3986 	}
3987 	if (i < 256)
3988 		return (i);
3989 	return (-1);
3990 }
3991 
3992 /*
3993  * This function checks for restart conditions.
3994  */
3995 static int
3996 nfsrv_checkrestart(nfsquad_t clientid, u_int32_t flags,
3997     nfsv4stateid_t *stateidp, int specialid)
3998 {
3999 	int ret = 0;
4000 
4001 	/*
4002 	 * First check for a server restart. Open, LockT, ReleaseLockOwner
4003 	 * and DelegPurge have a clientid, the rest a stateid.
4004 	 */
4005 	if (flags &
4006 	    (NFSLCK_OPEN | NFSLCK_TEST | NFSLCK_RELEASE | NFSLCK_DELEGPURGE)) {
4007 		if (clientid.lval[0] != nfsrvboottime) {
4008 			ret = NFSERR_STALECLIENTID;
4009 			goto out;
4010 		}
4011 	} else if (stateidp->other[0] != nfsrvboottime &&
4012 		specialid == 0) {
4013 		ret = NFSERR_STALESTATEID;
4014 		goto out;
4015 	}
4016 
4017 	/*
4018 	 * Read, Write, Setattr and LockT can return NFSERR_GRACE and do
4019 	 * not use a lock/open owner seqid#, so the check can be done now.
4020 	 * (The others will be checked, as required, later.)
4021 	 */
4022 	if (!(flags & (NFSLCK_CHECK | NFSLCK_TEST)))
4023 		goto out;
4024 
4025 	NFSLOCKSTATE();
4026 	ret = nfsrv_checkgrace(NULL, NULL, flags);
4027 	NFSUNLOCKSTATE();
4028 
4029 out:
4030 	NFSEXITCODE(ret);
4031 	return (ret);
4032 }
4033 
4034 /*
4035  * Check for grace.
4036  */
4037 static int
4038 nfsrv_checkgrace(struct nfsrv_descript *nd, struct nfsclient *clp,
4039     u_int32_t flags)
4040 {
4041 	int error = 0;
4042 
4043 	if ((nfsrv_stablefirst.nsf_flags & NFSNSF_GRACEOVER) != 0) {
4044 		if (flags & NFSLCK_RECLAIM) {
4045 			error = NFSERR_NOGRACE;
4046 			goto out;
4047 		}
4048 	} else {
4049 		if (!(flags & NFSLCK_RECLAIM)) {
4050 			error = NFSERR_GRACE;
4051 			goto out;
4052 		}
4053 		if (nd != NULL && clp != NULL &&
4054 		    (nd->nd_flag & ND_NFSV41) != 0 &&
4055 		    (clp->lc_flags & LCL_RECLAIMCOMPLETE) != 0) {
4056 			error = NFSERR_NOGRACE;
4057 			goto out;
4058 		}
4059 
4060 		/*
4061 		 * If grace is almost over and we are still getting Reclaims,
4062 		 * extend grace a bit.
4063 		 */
4064 		if ((NFSD_MONOSEC + NFSRV_LEASEDELTA) >
4065 		    nfsrv_stablefirst.nsf_eograce)
4066 			nfsrv_stablefirst.nsf_eograce = NFSD_MONOSEC +
4067 				NFSRV_LEASEDELTA;
4068 	}
4069 
4070 out:
4071 	NFSEXITCODE(error);
4072 	return (error);
4073 }
4074 
4075 /*
4076  * Do a server callback.
4077  */
4078 static int
4079 nfsrv_docallback(struct nfsclient *clp, int procnum,
4080     nfsv4stateid_t *stateidp, int trunc, fhandle_t *fhp,
4081     struct nfsvattr *nap, nfsattrbit_t *attrbitp, NFSPROC_T *p)
4082 {
4083 	mbuf_t m;
4084 	u_int32_t *tl;
4085 	struct nfsrv_descript nfsd, *nd = &nfsd;
4086 	struct ucred *cred;
4087 	int error = 0;
4088 	u_int32_t callback;
4089 	struct nfsdsession *sep = NULL;
4090 
4091 	cred = newnfs_getcred();
4092 	NFSLOCKSTATE();	/* mostly for lc_cbref++ */
4093 	if (clp->lc_flags & LCL_NEEDSCONFIRM) {
4094 		NFSUNLOCKSTATE();
4095 		panic("docallb");
4096 	}
4097 	clp->lc_cbref++;
4098 
4099 	/*
4100 	 * Fill the callback program# and version into the request
4101 	 * structure for newnfs_connect() to use.
4102 	 */
4103 	clp->lc_req.nr_prog = clp->lc_program;
4104 #ifdef notnow
4105 	if ((clp->lc_flags & LCL_NFSV41) != 0)
4106 		clp->lc_req.nr_vers = NFSV41_CBVERS;
4107 	else
4108 #endif
4109 		clp->lc_req.nr_vers = NFSV4_CBVERS;
4110 
4111 	/*
4112 	 * First, fill in some of the fields of nd and cr.
4113 	 */
4114 	nd->nd_flag = ND_NFSV4;
4115 	if (clp->lc_flags & LCL_GSS)
4116 		nd->nd_flag |= ND_KERBV;
4117 	if ((clp->lc_flags & LCL_NFSV41) != 0)
4118 		nd->nd_flag |= ND_NFSV41;
4119 	nd->nd_repstat = 0;
4120 	cred->cr_uid = clp->lc_uid;
4121 	cred->cr_gid = clp->lc_gid;
4122 	callback = clp->lc_callback;
4123 	NFSUNLOCKSTATE();
4124 	cred->cr_ngroups = 1;
4125 
4126 	/*
4127 	 * Get the first mbuf for the request.
4128 	 */
4129 	MGET(m, M_WAITOK, MT_DATA);
4130 	mbuf_setlen(m, 0);
4131 	nd->nd_mreq = nd->nd_mb = m;
4132 	nd->nd_bpos = NFSMTOD(m, caddr_t);
4133 
4134 	/*
4135 	 * and build the callback request.
4136 	 */
4137 	if (procnum == NFSV4OP_CBGETATTR) {
4138 		nd->nd_procnum = NFSV4PROC_CBCOMPOUND;
4139 		error = nfsrv_cbcallargs(nd, clp, callback, NFSV4OP_CBGETATTR,
4140 		    "CB Getattr", &sep);
4141 		if (error != 0) {
4142 			mbuf_freem(nd->nd_mreq);
4143 			goto errout;
4144 		}
4145 		(void)nfsm_fhtom(nd, (u_int8_t *)fhp, NFSX_MYFH, 0);
4146 		(void)nfsrv_putattrbit(nd, attrbitp);
4147 	} else if (procnum == NFSV4OP_CBRECALL) {
4148 		nd->nd_procnum = NFSV4PROC_CBCOMPOUND;
4149 		error = nfsrv_cbcallargs(nd, clp, callback, NFSV4OP_CBRECALL,
4150 		    "CB Recall", &sep);
4151 		if (error != 0) {
4152 			mbuf_freem(nd->nd_mreq);
4153 			goto errout;
4154 		}
4155 		NFSM_BUILD(tl, u_int32_t *, NFSX_UNSIGNED + NFSX_STATEID);
4156 		*tl++ = txdr_unsigned(stateidp->seqid);
4157 		NFSBCOPY((caddr_t)stateidp->other, (caddr_t)tl,
4158 		    NFSX_STATEIDOTHER);
4159 		tl += (NFSX_STATEIDOTHER / NFSX_UNSIGNED);
4160 		if (trunc)
4161 			*tl = newnfs_true;
4162 		else
4163 			*tl = newnfs_false;
4164 		(void)nfsm_fhtom(nd, (u_int8_t *)fhp, NFSX_MYFH, 0);
4165 	} else if (procnum == NFSV4PROC_CBNULL) {
4166 		nd->nd_procnum = NFSV4PROC_CBNULL;
4167 		if ((clp->lc_flags & LCL_NFSV41) != 0) {
4168 			error = nfsv4_getcbsession(clp, &sep);
4169 			if (error != 0) {
4170 				mbuf_freem(nd->nd_mreq);
4171 				goto errout;
4172 			}
4173 		}
4174 	} else {
4175 		error = NFSERR_SERVERFAULT;
4176 		mbuf_freem(nd->nd_mreq);
4177 		goto errout;
4178 	}
4179 
4180 	/*
4181 	 * Call newnfs_connect(), as required, and then newnfs_request().
4182 	 */
4183 	(void) newnfs_sndlock(&clp->lc_req.nr_lock);
4184 	if (clp->lc_req.nr_client == NULL) {
4185 		if ((clp->lc_flags & LCL_NFSV41) != 0)
4186 			error = ECONNREFUSED;
4187 		else if (nd->nd_procnum == NFSV4PROC_CBNULL)
4188 			error = newnfs_connect(NULL, &clp->lc_req, cred,
4189 			    NULL, 1);
4190 		else
4191 			error = newnfs_connect(NULL, &clp->lc_req, cred,
4192 			    NULL, 3);
4193 	}
4194 	newnfs_sndunlock(&clp->lc_req.nr_lock);
4195 	if (!error) {
4196 		if ((nd->nd_flag & ND_NFSV41) != 0) {
4197 			KASSERT(sep != NULL, ("sep NULL"));
4198 			if (sep->sess_cbsess.nfsess_xprt != NULL)
4199 				error = newnfs_request(nd, NULL, clp,
4200 				    &clp->lc_req, NULL, NULL, cred,
4201 				    clp->lc_program, clp->lc_req.nr_vers, NULL,
4202 				    1, NULL, &sep->sess_cbsess);
4203 			else {
4204 				/*
4205 				 * This should probably never occur, but if a
4206 				 * client somehow does an RPC without a
4207 				 * SequenceID Op that causes a callback just
4208 				 * after the nfsd threads have been terminated
4209 				 * and restared we could conceivably get here
4210 				 * without a backchannel xprt.
4211 				 */
4212 				printf("nfsrv_docallback: no xprt\n");
4213 				error = ECONNREFUSED;
4214 			}
4215 			nfsrv_freesession(sep, NULL);
4216 		} else
4217 			error = newnfs_request(nd, NULL, clp, &clp->lc_req,
4218 			    NULL, NULL, cred, clp->lc_program,
4219 			    clp->lc_req.nr_vers, NULL, 1, NULL, NULL);
4220 	}
4221 errout:
4222 	NFSFREECRED(cred);
4223 
4224 	/*
4225 	 * If error is set here, the Callback path isn't working
4226 	 * properly, so twiddle the appropriate LCL_ flags.
4227 	 * (nd_repstat != 0 indicates the Callback path is working,
4228 	 *  but the callback failed on the client.)
4229 	 */
4230 	if (error) {
4231 		/*
4232 		 * Mark the callback pathway down, which disabled issuing
4233 		 * of delegations and gets Renew to return NFSERR_CBPATHDOWN.
4234 		 */
4235 		NFSLOCKSTATE();
4236 		clp->lc_flags |= LCL_CBDOWN;
4237 		NFSUNLOCKSTATE();
4238 	} else {
4239 		/*
4240 		 * Callback worked. If the callback path was down, disable
4241 		 * callbacks, so no more delegations will be issued. (This
4242 		 * is done on the assumption that the callback pathway is
4243 		 * flakey.)
4244 		 */
4245 		NFSLOCKSTATE();
4246 		if (clp->lc_flags & LCL_CBDOWN)
4247 			clp->lc_flags &= ~(LCL_CBDOWN | LCL_CALLBACKSON);
4248 		NFSUNLOCKSTATE();
4249 		if (nd->nd_repstat)
4250 			error = nd->nd_repstat;
4251 		else if (error == 0 && procnum == NFSV4OP_CBGETATTR)
4252 			error = nfsv4_loadattr(nd, NULL, nap, NULL, NULL, 0,
4253 			    NULL, NULL, NULL, NULL, NULL, 0, NULL, NULL, NULL,
4254 			    p, NULL);
4255 		mbuf_freem(nd->nd_mrep);
4256 	}
4257 	NFSLOCKSTATE();
4258 	clp->lc_cbref--;
4259 	if ((clp->lc_flags & LCL_WAKEUPWANTED) && clp->lc_cbref == 0) {
4260 		clp->lc_flags &= ~LCL_WAKEUPWANTED;
4261 		wakeup(clp);
4262 	}
4263 	NFSUNLOCKSTATE();
4264 
4265 	NFSEXITCODE(error);
4266 	return (error);
4267 }
4268 
4269 /*
4270  * Set up the compound RPC for the callback.
4271  */
4272 static int
4273 nfsrv_cbcallargs(struct nfsrv_descript *nd, struct nfsclient *clp,
4274     uint32_t callback, int op, const char *optag, struct nfsdsession **sepp)
4275 {
4276 	uint32_t *tl;
4277 	int error, len;
4278 
4279 	len = strlen(optag);
4280 	(void)nfsm_strtom(nd, optag, len);
4281 	NFSM_BUILD(tl, uint32_t *, 4 * NFSX_UNSIGNED);
4282 	if ((nd->nd_flag & ND_NFSV41) != 0) {
4283 		*tl++ = txdr_unsigned(NFSV41_MINORVERSION);
4284 		*tl++ = txdr_unsigned(callback);
4285 		*tl++ = txdr_unsigned(2);
4286 		*tl = txdr_unsigned(NFSV4OP_CBSEQUENCE);
4287 		error = nfsv4_setcbsequence(nd, clp, 1, sepp);
4288 		if (error != 0)
4289 			return (error);
4290 		NFSM_BUILD(tl, u_int32_t *, NFSX_UNSIGNED);
4291 		*tl = txdr_unsigned(op);
4292 	} else {
4293 		*tl++ = txdr_unsigned(NFSV4_MINORVERSION);
4294 		*tl++ = txdr_unsigned(callback);
4295 		*tl++ = txdr_unsigned(1);
4296 		*tl = txdr_unsigned(op);
4297 	}
4298 	return (0);
4299 }
4300 
4301 /*
4302  * Return the next index# for a clientid. Mostly just increment and return
4303  * the next one, but... if the 32bit unsigned does actually wrap around,
4304  * it should be rebooted.
4305  * At an average rate of one new client per second, it will wrap around in
4306  * approximately 136 years. (I think the server will have been shut
4307  * down or rebooted before then.)
4308  */
4309 static u_int32_t
4310 nfsrv_nextclientindex(void)
4311 {
4312 	static u_int32_t client_index = 0;
4313 
4314 	client_index++;
4315 	if (client_index != 0)
4316 		return (client_index);
4317 
4318 	printf("%s: out of clientids\n", __func__);
4319 	return (client_index);
4320 }
4321 
4322 /*
4323  * Return the next index# for a stateid. Mostly just increment and return
4324  * the next one, but... if the 32bit unsigned does actually wrap around
4325  * (will a BSD server stay up that long?), find
4326  * new start and end values.
4327  */
4328 static u_int32_t
4329 nfsrv_nextstateindex(struct nfsclient *clp)
4330 {
4331 	struct nfsstate *stp;
4332 	int i;
4333 	u_int32_t canuse, min_index, max_index;
4334 
4335 	if (!(clp->lc_flags & LCL_INDEXNOTOK)) {
4336 		clp->lc_stateindex++;
4337 		if (clp->lc_stateindex != clp->lc_statemaxindex)
4338 			return (clp->lc_stateindex);
4339 	}
4340 
4341 	/*
4342 	 * Yuck, we've hit the end.
4343 	 * Look for a new min and max.
4344 	 */
4345 	min_index = 0;
4346 	max_index = 0xffffffff;
4347 	for (i = 0; i < nfsrv_statehashsize; i++) {
4348 	    LIST_FOREACH(stp, &clp->lc_stateid[i], ls_hash) {
4349 		if (stp->ls_stateid.other[2] > 0x80000000) {
4350 		    if (stp->ls_stateid.other[2] < max_index)
4351 			max_index = stp->ls_stateid.other[2];
4352 		} else {
4353 		    if (stp->ls_stateid.other[2] > min_index)
4354 			min_index = stp->ls_stateid.other[2];
4355 		}
4356 	    }
4357 	}
4358 
4359 	/*
4360 	 * Yikes, highly unlikely, but I'll handle it anyhow.
4361 	 */
4362 	if (min_index == 0x80000000 && max_index == 0x80000001) {
4363 	    canuse = 0;
4364 	    /*
4365 	     * Loop around until we find an unused entry. Return that
4366 	     * and set LCL_INDEXNOTOK, so the search will continue next time.
4367 	     * (This is one of those rare cases where a goto is the
4368 	     *  cleanest way to code the loop.)
4369 	     */
4370 tryagain:
4371 	    for (i = 0; i < nfsrv_statehashsize; i++) {
4372 		LIST_FOREACH(stp, &clp->lc_stateid[i], ls_hash) {
4373 		    if (stp->ls_stateid.other[2] == canuse) {
4374 			canuse++;
4375 			goto tryagain;
4376 		    }
4377 		}
4378 	    }
4379 	    clp->lc_flags |= LCL_INDEXNOTOK;
4380 	    return (canuse);
4381 	}
4382 
4383 	/*
4384 	 * Ok to start again from min + 1.
4385 	 */
4386 	clp->lc_stateindex = min_index + 1;
4387 	clp->lc_statemaxindex = max_index;
4388 	clp->lc_flags &= ~LCL_INDEXNOTOK;
4389 	return (clp->lc_stateindex);
4390 }
4391 
4392 /*
4393  * The following functions handle the stable storage file that deals with
4394  * the edge conditions described in RFC3530 Sec. 8.6.3.
4395  * The file is as follows:
4396  * - a single record at the beginning that has the lease time of the
4397  *   previous server instance (before the last reboot) and the nfsrvboottime
4398  *   values for the previous server boots.
4399  *   These previous boot times are used to ensure that the current
4400  *   nfsrvboottime does not, somehow, get set to a previous one.
4401  *   (This is important so that Stale ClientIDs and StateIDs can
4402  *    be recognized.)
4403  *   The number of previous nfsvrboottime values precedes the list.
4404  * - followed by some number of appended records with:
4405  *   - client id string
4406  *   - flag that indicates it is a record revoking state via lease
4407  *     expiration or similar
4408  *     OR has successfully acquired state.
4409  * These structures vary in length, with the client string at the end, up
4410  * to NFSV4_OPAQUELIMIT in size.
4411  *
4412  * At the end of the grace period, the file is truncated, the first
4413  * record is rewritten with updated information and any acquired state
4414  * records for successful reclaims of state are written.
4415  *
4416  * Subsequent records are appended when the first state is issued to
4417  * a client and when state is revoked for a client.
4418  *
4419  * When reading the file in, state issued records that come later in
4420  * the file override older ones, since the append log is in cronological order.
4421  * If, for some reason, the file can't be read, the grace period is
4422  * immediately terminated and all reclaims get NFSERR_NOGRACE.
4423  */
4424 
4425 /*
4426  * Read in the stable storage file. Called by nfssvc() before the nfsd
4427  * processes start servicing requests.
4428  */
4429 APPLESTATIC void
4430 nfsrv_setupstable(NFSPROC_T *p)
4431 {
4432 	struct nfsrv_stablefirst *sf = &nfsrv_stablefirst;
4433 	struct nfsrv_stable *sp, *nsp;
4434 	struct nfst_rec *tsp;
4435 	int error, i, tryagain;
4436 	off_t off = 0;
4437 	ssize_t aresid, len;
4438 
4439 	/*
4440 	 * If NFSNSF_UPDATEDONE is set, this is a restart of the nfsds without
4441 	 * a reboot, so state has not been lost.
4442 	 */
4443 	if (sf->nsf_flags & NFSNSF_UPDATEDONE)
4444 		return;
4445 	/*
4446 	 * Set Grace over just until the file reads successfully.
4447 	 */
4448 	nfsrvboottime = time_second;
4449 	LIST_INIT(&sf->nsf_head);
4450 	sf->nsf_flags = (NFSNSF_GRACEOVER | NFSNSF_NEEDLOCK);
4451 	sf->nsf_eograce = NFSD_MONOSEC + NFSRV_LEASEDELTA;
4452 	if (sf->nsf_fp == NULL)
4453 		return;
4454 	error = NFSD_RDWR(UIO_READ, NFSFPVNODE(sf->nsf_fp),
4455 	    (caddr_t)&sf->nsf_rec, sizeof (struct nfsf_rec), off, UIO_SYSSPACE,
4456 	    0, NFSFPCRED(sf->nsf_fp), &aresid, p);
4457 	if (error || aresid || sf->nsf_numboots == 0 ||
4458 		sf->nsf_numboots > NFSNSF_MAXNUMBOOTS)
4459 		return;
4460 
4461 	/*
4462 	 * Now, read in the boottimes.
4463 	 */
4464 	sf->nsf_bootvals = (time_t *)malloc((sf->nsf_numboots + 1) *
4465 		sizeof (time_t), M_TEMP, M_WAITOK);
4466 	off = sizeof (struct nfsf_rec);
4467 	error = NFSD_RDWR(UIO_READ, NFSFPVNODE(sf->nsf_fp),
4468 	    (caddr_t)sf->nsf_bootvals, sf->nsf_numboots * sizeof (time_t), off,
4469 	    UIO_SYSSPACE, 0, NFSFPCRED(sf->nsf_fp), &aresid, p);
4470 	if (error || aresid) {
4471 		free((caddr_t)sf->nsf_bootvals, M_TEMP);
4472 		sf->nsf_bootvals = NULL;
4473 		return;
4474 	}
4475 
4476 	/*
4477 	 * Make sure this nfsrvboottime is different from all recorded
4478 	 * previous ones.
4479 	 */
4480 	do {
4481 		tryagain = 0;
4482 		for (i = 0; i < sf->nsf_numboots; i++) {
4483 			if (nfsrvboottime == sf->nsf_bootvals[i]) {
4484 				nfsrvboottime++;
4485 				tryagain = 1;
4486 				break;
4487 			}
4488 		}
4489 	} while (tryagain);
4490 
4491 	sf->nsf_flags |= NFSNSF_OK;
4492 	off += (sf->nsf_numboots * sizeof (time_t));
4493 
4494 	/*
4495 	 * Read through the file, building a list of records for grace
4496 	 * checking.
4497 	 * Each record is between sizeof (struct nfst_rec) and
4498 	 * sizeof (struct nfst_rec) + NFSV4_OPAQUELIMIT - 1
4499 	 * and is actually sizeof (struct nfst_rec) + nst_len - 1.
4500 	 */
4501 	tsp = (struct nfst_rec *)malloc(sizeof (struct nfst_rec) +
4502 		NFSV4_OPAQUELIMIT - 1, M_TEMP, M_WAITOK);
4503 	do {
4504 	    error = NFSD_RDWR(UIO_READ, NFSFPVNODE(sf->nsf_fp),
4505 	        (caddr_t)tsp, sizeof (struct nfst_rec) + NFSV4_OPAQUELIMIT - 1,
4506 	        off, UIO_SYSSPACE, 0, NFSFPCRED(sf->nsf_fp), &aresid, p);
4507 	    len = (sizeof (struct nfst_rec) + NFSV4_OPAQUELIMIT - 1) - aresid;
4508 	    if (error || (len > 0 && (len < sizeof (struct nfst_rec) ||
4509 		len < (sizeof (struct nfst_rec) + tsp->len - 1)))) {
4510 		/*
4511 		 * Yuck, the file has been corrupted, so just return
4512 		 * after clearing out any restart state, so the grace period
4513 		 * is over.
4514 		 */
4515 		LIST_FOREACH_SAFE(sp, &sf->nsf_head, nst_list, nsp) {
4516 			LIST_REMOVE(sp, nst_list);
4517 			free((caddr_t)sp, M_TEMP);
4518 		}
4519 		free((caddr_t)tsp, M_TEMP);
4520 		sf->nsf_flags &= ~NFSNSF_OK;
4521 		free((caddr_t)sf->nsf_bootvals, M_TEMP);
4522 		sf->nsf_bootvals = NULL;
4523 		return;
4524 	    }
4525 	    if (len > 0) {
4526 		off += sizeof (struct nfst_rec) + tsp->len - 1;
4527 		/*
4528 		 * Search the list for a matching client.
4529 		 */
4530 		LIST_FOREACH(sp, &sf->nsf_head, nst_list) {
4531 			if (tsp->len == sp->nst_len &&
4532 			    !NFSBCMP(tsp->client, sp->nst_client, tsp->len))
4533 				break;
4534 		}
4535 		if (sp == LIST_END(&sf->nsf_head)) {
4536 			sp = (struct nfsrv_stable *)malloc(tsp->len +
4537 				sizeof (struct nfsrv_stable) - 1, M_TEMP,
4538 				M_WAITOK);
4539 			NFSBCOPY((caddr_t)tsp, (caddr_t)&sp->nst_rec,
4540 				sizeof (struct nfst_rec) + tsp->len - 1);
4541 			LIST_INSERT_HEAD(&sf->nsf_head, sp, nst_list);
4542 		} else {
4543 			if (tsp->flag == NFSNST_REVOKE)
4544 				sp->nst_flag |= NFSNST_REVOKE;
4545 			else
4546 				/*
4547 				 * A subsequent timestamp indicates the client
4548 				 * did a setclientid/confirm and any previous
4549 				 * revoke is no longer relevant.
4550 				 */
4551 				sp->nst_flag &= ~NFSNST_REVOKE;
4552 		}
4553 	    }
4554 	} while (len > 0);
4555 	free((caddr_t)tsp, M_TEMP);
4556 	sf->nsf_flags = NFSNSF_OK;
4557 	sf->nsf_eograce = NFSD_MONOSEC + sf->nsf_lease +
4558 		NFSRV_LEASEDELTA;
4559 }
4560 
4561 /*
4562  * Update the stable storage file, now that the grace period is over.
4563  */
4564 APPLESTATIC void
4565 nfsrv_updatestable(NFSPROC_T *p)
4566 {
4567 	struct nfsrv_stablefirst *sf = &nfsrv_stablefirst;
4568 	struct nfsrv_stable *sp, *nsp;
4569 	int i;
4570 	struct nfsvattr nva;
4571 	vnode_t vp;
4572 #if defined(__FreeBSD_version) && (__FreeBSD_version >= 500000)
4573 	mount_t mp = NULL;
4574 #endif
4575 	int error;
4576 
4577 	if (sf->nsf_fp == NULL || (sf->nsf_flags & NFSNSF_UPDATEDONE))
4578 		return;
4579 	sf->nsf_flags |= NFSNSF_UPDATEDONE;
4580 	/*
4581 	 * Ok, we need to rewrite the stable storage file.
4582 	 * - truncate to 0 length
4583 	 * - write the new first structure
4584 	 * - loop through the data structures, writing out any that
4585 	 *   have timestamps older than the old boot
4586 	 */
4587 	if (sf->nsf_bootvals) {
4588 		sf->nsf_numboots++;
4589 		for (i = sf->nsf_numboots - 2; i >= 0; i--)
4590 			sf->nsf_bootvals[i + 1] = sf->nsf_bootvals[i];
4591 	} else {
4592 		sf->nsf_numboots = 1;
4593 		sf->nsf_bootvals = (time_t *)malloc(sizeof (time_t),
4594 			M_TEMP, M_WAITOK);
4595 	}
4596 	sf->nsf_bootvals[0] = nfsrvboottime;
4597 	sf->nsf_lease = nfsrv_lease;
4598 	NFSVNO_ATTRINIT(&nva);
4599 	NFSVNO_SETATTRVAL(&nva, size, 0);
4600 	vp = NFSFPVNODE(sf->nsf_fp);
4601 	vn_start_write(vp, &mp, V_WAIT);
4602 	if (NFSVOPLOCK(vp, LK_EXCLUSIVE) == 0) {
4603 		error = nfsvno_setattr(vp, &nva, NFSFPCRED(sf->nsf_fp), p,
4604 		    NULL);
4605 		NFSVOPUNLOCK(vp, 0);
4606 	} else
4607 		error = EPERM;
4608 	vn_finished_write(mp);
4609 	if (!error)
4610 	    error = NFSD_RDWR(UIO_WRITE, vp,
4611 		(caddr_t)&sf->nsf_rec, sizeof (struct nfsf_rec), (off_t)0,
4612 		UIO_SYSSPACE, IO_SYNC, NFSFPCRED(sf->nsf_fp), NULL, p);
4613 	if (!error)
4614 	    error = NFSD_RDWR(UIO_WRITE, vp,
4615 		(caddr_t)sf->nsf_bootvals,
4616 		sf->nsf_numboots * sizeof (time_t),
4617 		(off_t)(sizeof (struct nfsf_rec)),
4618 		UIO_SYSSPACE, IO_SYNC, NFSFPCRED(sf->nsf_fp), NULL, p);
4619 	free((caddr_t)sf->nsf_bootvals, M_TEMP);
4620 	sf->nsf_bootvals = NULL;
4621 	if (error) {
4622 		sf->nsf_flags &= ~NFSNSF_OK;
4623 		printf("EEK! Can't write NfsV4 stable storage file\n");
4624 		return;
4625 	}
4626 	sf->nsf_flags |= NFSNSF_OK;
4627 
4628 	/*
4629 	 * Loop through the list and write out timestamp records for
4630 	 * any clients that successfully reclaimed state.
4631 	 */
4632 	LIST_FOREACH_SAFE(sp, &sf->nsf_head, nst_list, nsp) {
4633 		if (sp->nst_flag & NFSNST_GOTSTATE) {
4634 			nfsrv_writestable(sp->nst_client, sp->nst_len,
4635 				NFSNST_NEWSTATE, p);
4636 			sp->nst_clp->lc_flags |= LCL_STAMPEDSTABLE;
4637 		}
4638 		LIST_REMOVE(sp, nst_list);
4639 		free((caddr_t)sp, M_TEMP);
4640 	}
4641 	nfsrv_backupstable();
4642 }
4643 
4644 /*
4645  * Append a record to the stable storage file.
4646  */
4647 APPLESTATIC void
4648 nfsrv_writestable(u_char *client, int len, int flag, NFSPROC_T *p)
4649 {
4650 	struct nfsrv_stablefirst *sf = &nfsrv_stablefirst;
4651 	struct nfst_rec *sp;
4652 	int error;
4653 
4654 	if (!(sf->nsf_flags & NFSNSF_OK) || sf->nsf_fp == NULL)
4655 		return;
4656 	sp = (struct nfst_rec *)malloc(sizeof (struct nfst_rec) +
4657 		len - 1, M_TEMP, M_WAITOK);
4658 	sp->len = len;
4659 	NFSBCOPY(client, sp->client, len);
4660 	sp->flag = flag;
4661 	error = NFSD_RDWR(UIO_WRITE, NFSFPVNODE(sf->nsf_fp),
4662 	    (caddr_t)sp, sizeof (struct nfst_rec) + len - 1, (off_t)0,
4663 	    UIO_SYSSPACE, (IO_SYNC | IO_APPEND), NFSFPCRED(sf->nsf_fp), NULL, p);
4664 	free((caddr_t)sp, M_TEMP);
4665 	if (error) {
4666 		sf->nsf_flags &= ~NFSNSF_OK;
4667 		printf("EEK! Can't write NfsV4 stable storage file\n");
4668 	}
4669 }
4670 
4671 /*
4672  * This function is called during the grace period to mark a client
4673  * that successfully reclaimed state.
4674  */
4675 static void
4676 nfsrv_markstable(struct nfsclient *clp)
4677 {
4678 	struct nfsrv_stable *sp;
4679 
4680 	/*
4681 	 * First find the client structure.
4682 	 */
4683 	LIST_FOREACH(sp, &nfsrv_stablefirst.nsf_head, nst_list) {
4684 		if (sp->nst_len == clp->lc_idlen &&
4685 		    !NFSBCMP(sp->nst_client, clp->lc_id, sp->nst_len))
4686 			break;
4687 	}
4688 	if (sp == LIST_END(&nfsrv_stablefirst.nsf_head))
4689 		return;
4690 
4691 	/*
4692 	 * Now, just mark it and set the nfsclient back pointer.
4693 	 */
4694 	sp->nst_flag |= NFSNST_GOTSTATE;
4695 	sp->nst_clp = clp;
4696 }
4697 
4698 /*
4699  * This function is called for a reclaim, to see if it gets grace.
4700  * It returns 0 if a reclaim is allowed, 1 otherwise.
4701  */
4702 static int
4703 nfsrv_checkstable(struct nfsclient *clp)
4704 {
4705 	struct nfsrv_stable *sp;
4706 
4707 	/*
4708 	 * First, find the entry for the client.
4709 	 */
4710 	LIST_FOREACH(sp, &nfsrv_stablefirst.nsf_head, nst_list) {
4711 		if (sp->nst_len == clp->lc_idlen &&
4712 		    !NFSBCMP(sp->nst_client, clp->lc_id, sp->nst_len))
4713 			break;
4714 	}
4715 
4716 	/*
4717 	 * If not in the list, state was revoked or no state was issued
4718 	 * since the previous reboot, a reclaim is denied.
4719 	 */
4720 	if (sp == LIST_END(&nfsrv_stablefirst.nsf_head) ||
4721 	    (sp->nst_flag & NFSNST_REVOKE) ||
4722 	    !(nfsrv_stablefirst.nsf_flags & NFSNSF_OK))
4723 		return (1);
4724 	return (0);
4725 }
4726 
4727 /*
4728  * Test for and try to clear out a conflicting client. This is called by
4729  * nfsrv_lockctrl() and nfsrv_openctrl() when conflicts with other clients
4730  * a found.
4731  * The trick here is that it can't revoke a conflicting client with an
4732  * expired lease unless it holds the v4root lock, so...
4733  * If no v4root lock, get the lock and return 1 to indicate "try again".
4734  * Return 0 to indicate the conflict can't be revoked and 1 to indicate
4735  * the revocation worked and the conflicting client is "bye, bye", so it
4736  * can be tried again.
4737  * Return 2 to indicate that the vnode is VI_DOOMED after NFSVOPLOCK().
4738  * Unlocks State before a non-zero value is returned.
4739  */
4740 static int
4741 nfsrv_clientconflict(struct nfsclient *clp, int *haslockp, vnode_t vp,
4742     NFSPROC_T *p)
4743 {
4744 	int gotlock, lktype = 0;
4745 
4746 	/*
4747 	 * If lease hasn't expired, we can't fix it.
4748 	 */
4749 	if (clp->lc_expiry >= NFSD_MONOSEC ||
4750 	    !(nfsrv_stablefirst.nsf_flags & NFSNSF_UPDATEDONE))
4751 		return (0);
4752 	if (*haslockp == 0) {
4753 		NFSUNLOCKSTATE();
4754 		if (vp != NULL) {
4755 			lktype = NFSVOPISLOCKED(vp);
4756 			NFSVOPUNLOCK(vp, 0);
4757 		}
4758 		NFSLOCKV4ROOTMUTEX();
4759 		nfsv4_relref(&nfsv4rootfs_lock);
4760 		do {
4761 			gotlock = nfsv4_lock(&nfsv4rootfs_lock, 1, NULL,
4762 			    NFSV4ROOTLOCKMUTEXPTR, NULL);
4763 		} while (!gotlock);
4764 		NFSUNLOCKV4ROOTMUTEX();
4765 		*haslockp = 1;
4766 		if (vp != NULL) {
4767 			NFSVOPLOCK(vp, lktype | LK_RETRY);
4768 			if ((vp->v_iflag & VI_DOOMED) != 0)
4769 				return (2);
4770 		}
4771 		return (1);
4772 	}
4773 	NFSUNLOCKSTATE();
4774 
4775 	/*
4776 	 * Ok, we can expire the conflicting client.
4777 	 */
4778 	nfsrv_writestable(clp->lc_id, clp->lc_idlen, NFSNST_REVOKE, p);
4779 	nfsrv_backupstable();
4780 	nfsrv_cleanclient(clp, p);
4781 	nfsrv_freedeleglist(&clp->lc_deleg);
4782 	nfsrv_freedeleglist(&clp->lc_olddeleg);
4783 	LIST_REMOVE(clp, lc_hash);
4784 	nfsrv_zapclient(clp, p);
4785 	return (1);
4786 }
4787 
4788 /*
4789  * Resolve a delegation conflict.
4790  * Returns 0 to indicate the conflict was resolved without sleeping.
4791  * Return -1 to indicate that the caller should check for conflicts again.
4792  * Return > 0 for an error that should be returned, normally NFSERR_DELAY.
4793  *
4794  * Also, manipulate the nfsv4root_lock, as required. It isn't changed
4795  * for a return of 0, since there was no sleep and it could be required
4796  * later. It is released for a return of NFSERR_DELAY, since the caller
4797  * will return that error. It is released when a sleep was done waiting
4798  * for the delegation to be returned or expire (so that other nfsds can
4799  * handle ops). Then, it must be acquired for the write to stable storage.
4800  * (This function is somewhat similar to nfsrv_clientconflict(), but
4801  *  the semantics differ in a couple of subtle ways. The return of 0
4802  *  indicates the conflict was resolved without sleeping here, not
4803  *  that the conflict can't be resolved and the handling of nfsv4root_lock
4804  *  differs, as noted above.)
4805  * Unlocks State before returning a non-zero value.
4806  */
4807 static int
4808 nfsrv_delegconflict(struct nfsstate *stp, int *haslockp, NFSPROC_T *p,
4809     vnode_t vp)
4810 {
4811 	struct nfsclient *clp = stp->ls_clp;
4812 	int gotlock, error, lktype = 0, retrycnt, zapped_clp;
4813 	nfsv4stateid_t tstateid;
4814 	fhandle_t tfh;
4815 
4816 	/*
4817 	 * If the conflict is with an old delegation...
4818 	 */
4819 	if (stp->ls_flags & NFSLCK_OLDDELEG) {
4820 		/*
4821 		 * You can delete it, if it has expired.
4822 		 */
4823 		if (clp->lc_delegtime < NFSD_MONOSEC) {
4824 			nfsrv_freedeleg(stp);
4825 			NFSUNLOCKSTATE();
4826 			error = -1;
4827 			goto out;
4828 		}
4829 		NFSUNLOCKSTATE();
4830 		/*
4831 		 * During this delay, the old delegation could expire or it
4832 		 * could be recovered by the client via an Open with
4833 		 * CLAIM_DELEGATE_PREV.
4834 		 * Release the nfsv4root_lock, if held.
4835 		 */
4836 		if (*haslockp) {
4837 			*haslockp = 0;
4838 			NFSLOCKV4ROOTMUTEX();
4839 			nfsv4_unlock(&nfsv4rootfs_lock, 1);
4840 			NFSUNLOCKV4ROOTMUTEX();
4841 		}
4842 		error = NFSERR_DELAY;
4843 		goto out;
4844 	}
4845 
4846 	/*
4847 	 * It's a current delegation, so:
4848 	 * - check to see if the delegation has expired
4849 	 *   - if so, get the v4root lock and then expire it
4850 	 */
4851 	if (!(stp->ls_flags & NFSLCK_DELEGRECALL)) {
4852 		/*
4853 		 * - do a recall callback, since not yet done
4854 		 * For now, never allow truncate to be set. To use
4855 		 * truncate safely, it must be guaranteed that the
4856 		 * Remove, Rename or Setattr with size of 0 will
4857 		 * succeed and that would require major changes to
4858 		 * the VFS/Vnode OPs.
4859 		 * Set the expiry time large enough so that it won't expire
4860 		 * until after the callback, then set it correctly, once
4861 		 * the callback is done. (The delegation will now time
4862 		 * out whether or not the Recall worked ok. The timeout
4863 		 * will be extended when ops are done on the delegation
4864 		 * stateid, up to the timelimit.)
4865 		 */
4866 		stp->ls_delegtime = NFSD_MONOSEC + (2 * nfsrv_lease) +
4867 		    NFSRV_LEASEDELTA;
4868 		stp->ls_delegtimelimit = NFSD_MONOSEC + (6 * nfsrv_lease) +
4869 		    NFSRV_LEASEDELTA;
4870 		stp->ls_flags |= NFSLCK_DELEGRECALL;
4871 
4872 		/*
4873 		 * Loop NFSRV_CBRETRYCNT times while the CBRecall replies
4874 		 * NFSERR_BADSTATEID or NFSERR_BADHANDLE. This is done
4875 		 * in order to try and avoid a race that could happen
4876 		 * when a CBRecall request passed the Open reply with
4877 		 * the delegation in it when transitting the network.
4878 		 * Since nfsrv_docallback will sleep, don't use stp after
4879 		 * the call.
4880 		 */
4881 		NFSBCOPY((caddr_t)&stp->ls_stateid, (caddr_t)&tstateid,
4882 		    sizeof (tstateid));
4883 		NFSBCOPY((caddr_t)&stp->ls_lfp->lf_fh, (caddr_t)&tfh,
4884 		    sizeof (tfh));
4885 		NFSUNLOCKSTATE();
4886 		if (*haslockp) {
4887 			*haslockp = 0;
4888 			NFSLOCKV4ROOTMUTEX();
4889 			nfsv4_unlock(&nfsv4rootfs_lock, 1);
4890 			NFSUNLOCKV4ROOTMUTEX();
4891 		}
4892 		retrycnt = 0;
4893 		do {
4894 		    error = nfsrv_docallback(clp, NFSV4OP_CBRECALL,
4895 			&tstateid, 0, &tfh, NULL, NULL, p);
4896 		    retrycnt++;
4897 		} while ((error == NFSERR_BADSTATEID ||
4898 		    error == NFSERR_BADHANDLE) && retrycnt < NFSV4_CBRETRYCNT);
4899 		error = NFSERR_DELAY;
4900 		goto out;
4901 	}
4902 
4903 	if (clp->lc_expiry >= NFSD_MONOSEC &&
4904 	    stp->ls_delegtime >= NFSD_MONOSEC) {
4905 		NFSUNLOCKSTATE();
4906 		/*
4907 		 * A recall has been done, but it has not yet expired.
4908 		 * So, RETURN_DELAY.
4909 		 */
4910 		if (*haslockp) {
4911 			*haslockp = 0;
4912 			NFSLOCKV4ROOTMUTEX();
4913 			nfsv4_unlock(&nfsv4rootfs_lock, 1);
4914 			NFSUNLOCKV4ROOTMUTEX();
4915 		}
4916 		error = NFSERR_DELAY;
4917 		goto out;
4918 	}
4919 
4920 	/*
4921 	 * If we don't yet have the lock, just get it and then return,
4922 	 * since we need that before deleting expired state, such as
4923 	 * this delegation.
4924 	 * When getting the lock, unlock the vnode, so other nfsds that
4925 	 * are in progress, won't get stuck waiting for the vnode lock.
4926 	 */
4927 	if (*haslockp == 0) {
4928 		NFSUNLOCKSTATE();
4929 		if (vp != NULL) {
4930 			lktype = NFSVOPISLOCKED(vp);
4931 			NFSVOPUNLOCK(vp, 0);
4932 		}
4933 		NFSLOCKV4ROOTMUTEX();
4934 		nfsv4_relref(&nfsv4rootfs_lock);
4935 		do {
4936 			gotlock = nfsv4_lock(&nfsv4rootfs_lock, 1, NULL,
4937 			    NFSV4ROOTLOCKMUTEXPTR, NULL);
4938 		} while (!gotlock);
4939 		NFSUNLOCKV4ROOTMUTEX();
4940 		*haslockp = 1;
4941 		if (vp != NULL) {
4942 			NFSVOPLOCK(vp, lktype | LK_RETRY);
4943 			if ((vp->v_iflag & VI_DOOMED) != 0) {
4944 				*haslockp = 0;
4945 				NFSLOCKV4ROOTMUTEX();
4946 				nfsv4_unlock(&nfsv4rootfs_lock, 1);
4947 				NFSUNLOCKV4ROOTMUTEX();
4948 				error = NFSERR_PERM;
4949 				goto out;
4950 			}
4951 		}
4952 		error = -1;
4953 		goto out;
4954 	}
4955 
4956 	NFSUNLOCKSTATE();
4957 	/*
4958 	 * Ok, we can delete the expired delegation.
4959 	 * First, write the Revoke record to stable storage and then
4960 	 * clear out the conflict.
4961 	 * Since all other nfsd threads are now blocked, we can safely
4962 	 * sleep without the state changing.
4963 	 */
4964 	nfsrv_writestable(clp->lc_id, clp->lc_idlen, NFSNST_REVOKE, p);
4965 	nfsrv_backupstable();
4966 	if (clp->lc_expiry < NFSD_MONOSEC) {
4967 		nfsrv_cleanclient(clp, p);
4968 		nfsrv_freedeleglist(&clp->lc_deleg);
4969 		nfsrv_freedeleglist(&clp->lc_olddeleg);
4970 		LIST_REMOVE(clp, lc_hash);
4971 		zapped_clp = 1;
4972 	} else {
4973 		nfsrv_freedeleg(stp);
4974 		zapped_clp = 0;
4975 	}
4976 	if (zapped_clp)
4977 		nfsrv_zapclient(clp, p);
4978 	error = -1;
4979 
4980 out:
4981 	NFSEXITCODE(error);
4982 	return (error);
4983 }
4984 
4985 /*
4986  * Check for a remove allowed, if remove is set to 1 and get rid of
4987  * delegations.
4988  */
4989 APPLESTATIC int
4990 nfsrv_checkremove(vnode_t vp, int remove, NFSPROC_T *p)
4991 {
4992 	struct nfsstate *stp;
4993 	struct nfslockfile *lfp;
4994 	int error, haslock = 0;
4995 	fhandle_t nfh;
4996 
4997 	/*
4998 	 * First, get the lock file structure.
4999 	 * (A return of -1 means no associated state, so remove ok.)
5000 	 */
5001 	error = nfsrv_getlockfh(vp, NFSLCK_CHECK, NULL, &nfh, p);
5002 tryagain:
5003 	NFSLOCKSTATE();
5004 	if (!error)
5005 		error = nfsrv_getlockfile(NFSLCK_CHECK, NULL, &lfp, &nfh, 0);
5006 	if (error) {
5007 		NFSUNLOCKSTATE();
5008 		if (haslock) {
5009 			NFSLOCKV4ROOTMUTEX();
5010 			nfsv4_unlock(&nfsv4rootfs_lock, 1);
5011 			NFSUNLOCKV4ROOTMUTEX();
5012 		}
5013 		if (error == -1)
5014 			error = 0;
5015 		goto out;
5016 	}
5017 
5018 	/*
5019 	 * Now, we must Recall any delegations.
5020 	 */
5021 	error = nfsrv_cleandeleg(vp, lfp, NULL, &haslock, p);
5022 	if (error) {
5023 		/*
5024 		 * nfsrv_cleandeleg() unlocks state for non-zero
5025 		 * return.
5026 		 */
5027 		if (error == -1)
5028 			goto tryagain;
5029 		if (haslock) {
5030 			NFSLOCKV4ROOTMUTEX();
5031 			nfsv4_unlock(&nfsv4rootfs_lock, 1);
5032 			NFSUNLOCKV4ROOTMUTEX();
5033 		}
5034 		goto out;
5035 	}
5036 
5037 	/*
5038 	 * Now, look for a conflicting open share.
5039 	 */
5040 	if (remove) {
5041 		/*
5042 		 * If the entry in the directory was the last reference to the
5043 		 * corresponding filesystem object, the object can be destroyed
5044 		 * */
5045 		if(lfp->lf_usecount>1)
5046 			LIST_FOREACH(stp, &lfp->lf_open, ls_file) {
5047 				if (stp->ls_flags & NFSLCK_WRITEDENY) {
5048 					error = NFSERR_FILEOPEN;
5049 					break;
5050 				}
5051 			}
5052 	}
5053 
5054 	NFSUNLOCKSTATE();
5055 	if (haslock) {
5056 		NFSLOCKV4ROOTMUTEX();
5057 		nfsv4_unlock(&nfsv4rootfs_lock, 1);
5058 		NFSUNLOCKV4ROOTMUTEX();
5059 	}
5060 
5061 out:
5062 	NFSEXITCODE(error);
5063 	return (error);
5064 }
5065 
5066 /*
5067  * Clear out all delegations for the file referred to by lfp.
5068  * May return NFSERR_DELAY, if there will be a delay waiting for
5069  * delegations to expire.
5070  * Returns -1 to indicate it slept while recalling a delegation.
5071  * This function has the side effect of deleting the nfslockfile structure,
5072  * if it no longer has associated state and didn't have to sleep.
5073  * Unlocks State before a non-zero value is returned.
5074  */
5075 static int
5076 nfsrv_cleandeleg(vnode_t vp, struct nfslockfile *lfp,
5077     struct nfsclient *clp, int *haslockp, NFSPROC_T *p)
5078 {
5079 	struct nfsstate *stp, *nstp;
5080 	int ret = 0;
5081 
5082 	stp = LIST_FIRST(&lfp->lf_deleg);
5083 	while (stp != LIST_END(&lfp->lf_deleg)) {
5084 		nstp = LIST_NEXT(stp, ls_file);
5085 		if (stp->ls_clp != clp) {
5086 			ret = nfsrv_delegconflict(stp, haslockp, p, vp);
5087 			if (ret) {
5088 				/*
5089 				 * nfsrv_delegconflict() unlocks state
5090 				 * when it returns non-zero.
5091 				 */
5092 				goto out;
5093 			}
5094 		}
5095 		stp = nstp;
5096 	}
5097 out:
5098 	NFSEXITCODE(ret);
5099 	return (ret);
5100 }
5101 
5102 /*
5103  * There are certain operations that, when being done outside of NFSv4,
5104  * require that any NFSv4 delegation for the file be recalled.
5105  * This function is to be called for those cases:
5106  * VOP_RENAME() - When a delegation is being recalled for any reason,
5107  *	the client may have to do Opens against the server, using the file's
5108  *	final component name. If the file has been renamed on the server,
5109  *	that component name will be incorrect and the Open will fail.
5110  * VOP_REMOVE() - Theoretically, a client could Open a file after it has
5111  *	been removed on the server, if there is a delegation issued to
5112  *	that client for the file. I say "theoretically" since clients
5113  *	normally do an Access Op before the Open and that Access Op will
5114  *	fail with ESTALE. Note that NFSv2 and 3 don't even do Opens, so
5115  *	they will detect the file's removal in the same manner. (There is
5116  *	one case where RFC3530 allows a client to do an Open without first
5117  *	doing an Access Op, which is passage of a check against the ACE
5118  *	returned with a Write delegation, but current practice is to ignore
5119  *	the ACE and always do an Access Op.)
5120  *	Since the functions can only be called with an unlocked vnode, this
5121  *	can't be done at this time.
5122  * VOP_ADVLOCK() - When a client holds a delegation, it can issue byte range
5123  *	locks locally in the client, which are not visible to the server. To
5124  *	deal with this, issuing of delegations for a vnode must be disabled
5125  *	and all delegations for the vnode recalled. This is done via the
5126  *	second function, using the VV_DISABLEDELEG vflag on the vnode.
5127  */
5128 APPLESTATIC void
5129 nfsd_recalldelegation(vnode_t vp, NFSPROC_T *p)
5130 {
5131 	time_t starttime;
5132 	int error;
5133 
5134 	/*
5135 	 * First, check to see if the server is currently running and it has
5136 	 * been called for a regular file when issuing delegations.
5137 	 */
5138 	if (newnfs_numnfsd == 0 || vp->v_type != VREG ||
5139 	    nfsrv_issuedelegs == 0)
5140 		return;
5141 
5142 	KASSERT((NFSVOPISLOCKED(vp) != LK_EXCLUSIVE), ("vp %p is locked", vp));
5143 	/*
5144 	 * First, get a reference on the nfsv4rootfs_lock so that an
5145 	 * exclusive lock cannot be acquired by another thread.
5146 	 */
5147 	NFSLOCKV4ROOTMUTEX();
5148 	nfsv4_getref(&nfsv4rootfs_lock, NULL, NFSV4ROOTLOCKMUTEXPTR, NULL);
5149 	NFSUNLOCKV4ROOTMUTEX();
5150 
5151 	/*
5152 	 * Now, call nfsrv_checkremove() in a loop while it returns
5153 	 * NFSERR_DELAY. Return upon any other error or when timed out.
5154 	 */
5155 	starttime = NFSD_MONOSEC;
5156 	do {
5157 		if (NFSVOPLOCK(vp, LK_EXCLUSIVE) == 0) {
5158 			error = nfsrv_checkremove(vp, 0, p);
5159 			NFSVOPUNLOCK(vp, 0);
5160 		} else
5161 			error = EPERM;
5162 		if (error == NFSERR_DELAY) {
5163 			if (NFSD_MONOSEC - starttime > NFS_REMOVETIMEO)
5164 				break;
5165 			/* Sleep for a short period of time */
5166 			(void) nfs_catnap(PZERO, 0, "nfsremove");
5167 		}
5168 	} while (error == NFSERR_DELAY);
5169 	NFSLOCKV4ROOTMUTEX();
5170 	nfsv4_relref(&nfsv4rootfs_lock);
5171 	NFSUNLOCKV4ROOTMUTEX();
5172 }
5173 
5174 APPLESTATIC void
5175 nfsd_disabledelegation(vnode_t vp, NFSPROC_T *p)
5176 {
5177 
5178 #ifdef VV_DISABLEDELEG
5179 	/*
5180 	 * First, flag issuance of delegations disabled.
5181 	 */
5182 	atomic_set_long(&vp->v_vflag, VV_DISABLEDELEG);
5183 #endif
5184 
5185 	/*
5186 	 * Then call nfsd_recalldelegation() to get rid of all extant
5187 	 * delegations.
5188 	 */
5189 	nfsd_recalldelegation(vp, p);
5190 }
5191 
5192 /*
5193  * Check for conflicting locks, etc. and then get rid of delegations.
5194  * (At one point I thought that I should get rid of delegations for any
5195  *  Setattr, since it could potentially disallow the I/O op (read or write)
5196  *  allowed by the delegation. However, Setattr Ops that aren't changing
5197  *  the size get a stateid of all 0s, so you can't tell if it is a delegation
5198  *  for the same client or a different one, so I decided to only get rid
5199  *  of delegations for other clients when the size is being changed.)
5200  * In general, a Setattr can disable NFS I/O Ops that are outstanding, such
5201  * as Write backs, even if there is no delegation, so it really isn't any
5202  * different?)
5203  */
5204 APPLESTATIC int
5205 nfsrv_checksetattr(vnode_t vp, struct nfsrv_descript *nd,
5206     nfsv4stateid_t *stateidp, struct nfsvattr *nvap, nfsattrbit_t *attrbitp,
5207     struct nfsexstuff *exp, NFSPROC_T *p)
5208 {
5209 	struct nfsstate st, *stp = &st;
5210 	struct nfslock lo, *lop = &lo;
5211 	int error = 0;
5212 	nfsquad_t clientid;
5213 
5214 	if (NFSISSET_ATTRBIT(attrbitp, NFSATTRBIT_SIZE)) {
5215 		stp->ls_flags = (NFSLCK_CHECK | NFSLCK_WRITEACCESS);
5216 		lop->lo_first = nvap->na_size;
5217 	} else {
5218 		stp->ls_flags = 0;
5219 		lop->lo_first = 0;
5220 	}
5221 	if (NFSISSET_ATTRBIT(attrbitp, NFSATTRBIT_OWNER) ||
5222 	    NFSISSET_ATTRBIT(attrbitp, NFSATTRBIT_OWNERGROUP) ||
5223 	    NFSISSET_ATTRBIT(attrbitp, NFSATTRBIT_MODE) ||
5224 	    NFSISSET_ATTRBIT(attrbitp, NFSATTRBIT_ACL))
5225 		stp->ls_flags |= NFSLCK_SETATTR;
5226 	if (stp->ls_flags == 0)
5227 		goto out;
5228 	lop->lo_end = NFS64BITSSET;
5229 	lop->lo_flags = NFSLCK_WRITE;
5230 	stp->ls_ownerlen = 0;
5231 	stp->ls_op = NULL;
5232 	stp->ls_uid = nd->nd_cred->cr_uid;
5233 	stp->ls_stateid.seqid = stateidp->seqid;
5234 	clientid.lval[0] = stp->ls_stateid.other[0] = stateidp->other[0];
5235 	clientid.lval[1] = stp->ls_stateid.other[1] = stateidp->other[1];
5236 	stp->ls_stateid.other[2] = stateidp->other[2];
5237 	error = nfsrv_lockctrl(vp, &stp, &lop, NULL, clientid,
5238 	    stateidp, exp, nd, p);
5239 
5240 out:
5241 	NFSEXITCODE2(error, nd);
5242 	return (error);
5243 }
5244 
5245 /*
5246  * Check for a write delegation and do a CBGETATTR if there is one, updating
5247  * the attributes, as required.
5248  * Should I return an error if I can't get the attributes? (For now, I'll
5249  * just return ok.
5250  */
5251 APPLESTATIC int
5252 nfsrv_checkgetattr(struct nfsrv_descript *nd, vnode_t vp,
5253     struct nfsvattr *nvap, nfsattrbit_t *attrbitp, struct ucred *cred,
5254     NFSPROC_T *p)
5255 {
5256 	struct nfsstate *stp;
5257 	struct nfslockfile *lfp;
5258 	struct nfsclient *clp;
5259 	struct nfsvattr nva;
5260 	fhandle_t nfh;
5261 	int error = 0;
5262 	nfsattrbit_t cbbits;
5263 	u_quad_t delegfilerev;
5264 
5265 	NFSCBGETATTR_ATTRBIT(attrbitp, &cbbits);
5266 	if (!NFSNONZERO_ATTRBIT(&cbbits))
5267 		goto out;
5268 
5269 	/*
5270 	 * Get the lock file structure.
5271 	 * (A return of -1 means no associated state, so return ok.)
5272 	 */
5273 	error = nfsrv_getlockfh(vp, NFSLCK_CHECK, NULL, &nfh, p);
5274 	NFSLOCKSTATE();
5275 	if (!error)
5276 		error = nfsrv_getlockfile(NFSLCK_CHECK, NULL, &lfp, &nfh, 0);
5277 	if (error) {
5278 		NFSUNLOCKSTATE();
5279 		if (error == -1)
5280 			error = 0;
5281 		goto out;
5282 	}
5283 
5284 	/*
5285 	 * Now, look for a write delegation.
5286 	 */
5287 	LIST_FOREACH(stp, &lfp->lf_deleg, ls_file) {
5288 		if (stp->ls_flags & NFSLCK_DELEGWRITE)
5289 			break;
5290 	}
5291 	if (stp == LIST_END(&lfp->lf_deleg)) {
5292 		NFSUNLOCKSTATE();
5293 		goto out;
5294 	}
5295 	clp = stp->ls_clp;
5296 	delegfilerev = stp->ls_filerev;
5297 
5298 	/*
5299 	 * If the Write delegation was issued as a part of this Compound RPC
5300 	 * or if we have an Implied Clientid (used in a previous Op in this
5301 	 * compound) and it is the client the delegation was issued to,
5302 	 * just return ok.
5303 	 * I also assume that it is from the same client iff the network
5304 	 * host IP address is the same as the callback address. (Not
5305 	 * exactly correct by the RFC, but avoids a lot of Getattr
5306 	 * callbacks.)
5307 	 */
5308 	if (nd->nd_compref == stp->ls_compref ||
5309 	    ((nd->nd_flag & ND_IMPLIEDCLID) &&
5310 	     clp->lc_clientid.qval == nd->nd_clientid.qval) ||
5311 	     nfsaddr2_match(clp->lc_req.nr_nam, nd->nd_nam)) {
5312 		NFSUNLOCKSTATE();
5313 		goto out;
5314 	}
5315 
5316 	/*
5317 	 * We are now done with the delegation state structure,
5318 	 * so the statelock can be released and we can now tsleep().
5319 	 */
5320 
5321 	/*
5322 	 * Now, we must do the CB Getattr callback, to see if Change or Size
5323 	 * has changed.
5324 	 */
5325 	if (clp->lc_expiry >= NFSD_MONOSEC) {
5326 		NFSUNLOCKSTATE();
5327 		NFSVNO_ATTRINIT(&nva);
5328 		nva.na_filerev = NFS64BITSSET;
5329 		error = nfsrv_docallback(clp, NFSV4OP_CBGETATTR, NULL,
5330 		    0, &nfh, &nva, &cbbits, p);
5331 		if (!error) {
5332 			if ((nva.na_filerev != NFS64BITSSET &&
5333 			    nva.na_filerev > delegfilerev) ||
5334 			    (NFSVNO_ISSETSIZE(&nva) &&
5335 			     nva.na_size != nvap->na_size)) {
5336 				error = nfsvno_updfilerev(vp, nvap, cred, p);
5337 				if (NFSVNO_ISSETSIZE(&nva))
5338 					nvap->na_size = nva.na_size;
5339 			}
5340 		} else
5341 			error = 0;	/* Ignore callback errors for now. */
5342 	} else {
5343 		NFSUNLOCKSTATE();
5344 	}
5345 
5346 out:
5347 	NFSEXITCODE2(error, nd);
5348 	return (error);
5349 }
5350 
5351 /*
5352  * This function looks for openowners that haven't had any opens for
5353  * a while and throws them away. Called by an nfsd when NFSNSF_NOOPENS
5354  * is set.
5355  */
5356 APPLESTATIC void
5357 nfsrv_throwawayopens(NFSPROC_T *p)
5358 {
5359 	struct nfsclient *clp, *nclp;
5360 	struct nfsstate *stp, *nstp;
5361 	int i;
5362 
5363 	NFSLOCKSTATE();
5364 	nfsrv_stablefirst.nsf_flags &= ~NFSNSF_NOOPENS;
5365 	/*
5366 	 * For each client...
5367 	 */
5368 	for (i = 0; i < nfsrv_clienthashsize; i++) {
5369 	    LIST_FOREACH_SAFE(clp, &nfsclienthash[i], lc_hash, nclp) {
5370 		LIST_FOREACH_SAFE(stp, &clp->lc_open, ls_list, nstp) {
5371 			if (LIST_EMPTY(&stp->ls_open) &&
5372 			    (stp->ls_noopens > NFSNOOPEN ||
5373 			     (nfsrv_openpluslock * 2) >
5374 			     nfsrv_v4statelimit))
5375 				nfsrv_freeopenowner(stp, 0, p);
5376 		}
5377 	    }
5378 	}
5379 	NFSUNLOCKSTATE();
5380 }
5381 
5382 /*
5383  * This function checks to see if the credentials are the same.
5384  * Returns 1 for not same, 0 otherwise.
5385  */
5386 static int
5387 nfsrv_notsamecredname(struct nfsrv_descript *nd, struct nfsclient *clp)
5388 {
5389 
5390 	if (nd->nd_flag & ND_GSS) {
5391 		if (!(clp->lc_flags & LCL_GSS))
5392 			return (1);
5393 		if (clp->lc_flags & LCL_NAME) {
5394 			if (nd->nd_princlen != clp->lc_namelen ||
5395 			    NFSBCMP(nd->nd_principal, clp->lc_name,
5396 				clp->lc_namelen))
5397 				return (1);
5398 			else
5399 				return (0);
5400 		}
5401 		if (nd->nd_cred->cr_uid == clp->lc_uid)
5402 			return (0);
5403 		else
5404 			return (1);
5405 	} else if (clp->lc_flags & LCL_GSS)
5406 		return (1);
5407 	/*
5408 	 * For AUTH_SYS, allow the same uid or root. (This is underspecified
5409 	 * in RFC3530, which talks about principals, but doesn't say anything
5410 	 * about uids for AUTH_SYS.)
5411 	 */
5412 	if (nd->nd_cred->cr_uid == clp->lc_uid || nd->nd_cred->cr_uid == 0)
5413 		return (0);
5414 	else
5415 		return (1);
5416 }
5417 
5418 /*
5419  * Calculate the lease expiry time.
5420  */
5421 static time_t
5422 nfsrv_leaseexpiry(void)
5423 {
5424 
5425 	if (nfsrv_stablefirst.nsf_eograce > NFSD_MONOSEC)
5426 		return (NFSD_MONOSEC + 2 * (nfsrv_lease + NFSRV_LEASEDELTA));
5427 	return (NFSD_MONOSEC + nfsrv_lease + NFSRV_LEASEDELTA);
5428 }
5429 
5430 /*
5431  * Delay the delegation timeout as far as ls_delegtimelimit, as required.
5432  */
5433 static void
5434 nfsrv_delaydelegtimeout(struct nfsstate *stp)
5435 {
5436 
5437 	if ((stp->ls_flags & NFSLCK_DELEGRECALL) == 0)
5438 		return;
5439 
5440 	if ((stp->ls_delegtime + 15) > NFSD_MONOSEC &&
5441 	    stp->ls_delegtime < stp->ls_delegtimelimit) {
5442 		stp->ls_delegtime += nfsrv_lease;
5443 		if (stp->ls_delegtime > stp->ls_delegtimelimit)
5444 			stp->ls_delegtime = stp->ls_delegtimelimit;
5445 	}
5446 }
5447 
5448 /*
5449  * This function checks to see if there is any other state associated
5450  * with the openowner for this Open.
5451  * It returns 1 if there is no other state, 0 otherwise.
5452  */
5453 static int
5454 nfsrv_nootherstate(struct nfsstate *stp)
5455 {
5456 	struct nfsstate *tstp;
5457 
5458 	LIST_FOREACH(tstp, &stp->ls_openowner->ls_open, ls_list) {
5459 		if (tstp != stp || !LIST_EMPTY(&tstp->ls_lock))
5460 			return (0);
5461 	}
5462 	return (1);
5463 }
5464 
5465 /*
5466  * Create a list of lock deltas (changes to local byte range locking
5467  * that can be rolled back using the list) and apply the changes via
5468  * nfsvno_advlock(). Optionally, lock the list. It is expected that either
5469  * the rollback or update function will be called after this.
5470  * It returns an error (and rolls back, as required), if any nfsvno_advlock()
5471  * call fails. If it returns an error, it will unlock the list.
5472  */
5473 static int
5474 nfsrv_locallock(vnode_t vp, struct nfslockfile *lfp, int flags,
5475     uint64_t first, uint64_t end, struct nfslockconflict *cfp, NFSPROC_T *p)
5476 {
5477 	struct nfslock *lop, *nlop;
5478 	int error = 0;
5479 
5480 	/* Loop through the list of locks. */
5481 	lop = LIST_FIRST(&lfp->lf_locallock);
5482 	while (first < end && lop != NULL) {
5483 		nlop = LIST_NEXT(lop, lo_lckowner);
5484 		if (first >= lop->lo_end) {
5485 			/* not there yet */
5486 			lop = nlop;
5487 		} else if (first < lop->lo_first) {
5488 			/* new one starts before entry in list */
5489 			if (end <= lop->lo_first) {
5490 				/* no overlap between old and new */
5491 				error = nfsrv_dolocal(vp, lfp, flags,
5492 				    NFSLCK_UNLOCK, first, end, cfp, p);
5493 				if (error != 0)
5494 					break;
5495 				first = end;
5496 			} else {
5497 				/* handle fragment overlapped with new one */
5498 				error = nfsrv_dolocal(vp, lfp, flags,
5499 				    NFSLCK_UNLOCK, first, lop->lo_first, cfp,
5500 				    p);
5501 				if (error != 0)
5502 					break;
5503 				first = lop->lo_first;
5504 			}
5505 		} else {
5506 			/* new one overlaps this entry in list */
5507 			if (end <= lop->lo_end) {
5508 				/* overlaps all of new one */
5509 				error = nfsrv_dolocal(vp, lfp, flags,
5510 				    lop->lo_flags, first, end, cfp, p);
5511 				if (error != 0)
5512 					break;
5513 				first = end;
5514 			} else {
5515 				/* handle fragment overlapped with new one */
5516 				error = nfsrv_dolocal(vp, lfp, flags,
5517 				    lop->lo_flags, first, lop->lo_end, cfp, p);
5518 				if (error != 0)
5519 					break;
5520 				first = lop->lo_end;
5521 				lop = nlop;
5522 			}
5523 		}
5524 	}
5525 	if (first < end && error == 0)
5526 		/* handle fragment past end of list */
5527 		error = nfsrv_dolocal(vp, lfp, flags, NFSLCK_UNLOCK, first,
5528 		    end, cfp, p);
5529 
5530 	NFSEXITCODE(error);
5531 	return (error);
5532 }
5533 
5534 /*
5535  * Local lock unlock. Unlock all byte ranges that are no longer locked
5536  * by NFSv4. To do this, unlock any subranges of first-->end that
5537  * do not overlap with the byte ranges of any lock in the lfp->lf_lock
5538  * list. This list has all locks for the file held by other
5539  * <clientid, lockowner> tuples. The list is ordered by increasing
5540  * lo_first value, but may have entries that overlap each other, for
5541  * the case of read locks.
5542  */
5543 static void
5544 nfsrv_localunlock(vnode_t vp, struct nfslockfile *lfp, uint64_t init_first,
5545     uint64_t init_end, NFSPROC_T *p)
5546 {
5547 	struct nfslock *lop;
5548 	uint64_t first, end, prevfirst;
5549 
5550 	first = init_first;
5551 	end = init_end;
5552 	while (first < init_end) {
5553 		/* Loop through all nfs locks, adjusting first and end */
5554 		prevfirst = 0;
5555 		LIST_FOREACH(lop, &lfp->lf_lock, lo_lckfile) {
5556 			KASSERT(prevfirst <= lop->lo_first,
5557 			    ("nfsv4 locks out of order"));
5558 			KASSERT(lop->lo_first < lop->lo_end,
5559 			    ("nfsv4 bogus lock"));
5560 			prevfirst = lop->lo_first;
5561 			if (first >= lop->lo_first &&
5562 			    first < lop->lo_end)
5563 				/*
5564 				 * Overlaps with initial part, so trim
5565 				 * off that initial part by moving first past
5566 				 * it.
5567 				 */
5568 				first = lop->lo_end;
5569 			else if (end > lop->lo_first &&
5570 			    lop->lo_first > first) {
5571 				/*
5572 				 * This lock defines the end of the
5573 				 * segment to unlock, so set end to the
5574 				 * start of it and break out of the loop.
5575 				 */
5576 				end = lop->lo_first;
5577 				break;
5578 			}
5579 			if (first >= end)
5580 				/*
5581 				 * There is no segment left to do, so
5582 				 * break out of this loop and then exit
5583 				 * the outer while() since first will be set
5584 				 * to end, which must equal init_end here.
5585 				 */
5586 				break;
5587 		}
5588 		if (first < end) {
5589 			/* Unlock this segment */
5590 			(void) nfsrv_dolocal(vp, lfp, NFSLCK_UNLOCK,
5591 			    NFSLCK_READ, first, end, NULL, p);
5592 			nfsrv_locallock_commit(lfp, NFSLCK_UNLOCK,
5593 			    first, end);
5594 		}
5595 		/*
5596 		 * Now move past this segment and look for any further
5597 		 * segment in the range, if there is one.
5598 		 */
5599 		first = end;
5600 		end = init_end;
5601 	}
5602 }
5603 
5604 /*
5605  * Do the local lock operation and update the rollback list, as required.
5606  * Perform the rollback and return the error if nfsvno_advlock() fails.
5607  */
5608 static int
5609 nfsrv_dolocal(vnode_t vp, struct nfslockfile *lfp, int flags, int oldflags,
5610     uint64_t first, uint64_t end, struct nfslockconflict *cfp, NFSPROC_T *p)
5611 {
5612 	struct nfsrollback *rlp;
5613 	int error = 0, ltype, oldltype;
5614 
5615 	if (flags & NFSLCK_WRITE)
5616 		ltype = F_WRLCK;
5617 	else if (flags & NFSLCK_READ)
5618 		ltype = F_RDLCK;
5619 	else
5620 		ltype = F_UNLCK;
5621 	if (oldflags & NFSLCK_WRITE)
5622 		oldltype = F_WRLCK;
5623 	else if (oldflags & NFSLCK_READ)
5624 		oldltype = F_RDLCK;
5625 	else
5626 		oldltype = F_UNLCK;
5627 	if (ltype == oldltype || (oldltype == F_WRLCK && ltype == F_RDLCK))
5628 		/* nothing to do */
5629 		goto out;
5630 	error = nfsvno_advlock(vp, ltype, first, end, p);
5631 	if (error != 0) {
5632 		if (cfp != NULL) {
5633 			cfp->cl_clientid.lval[0] = 0;
5634 			cfp->cl_clientid.lval[1] = 0;
5635 			cfp->cl_first = 0;
5636 			cfp->cl_end = NFS64BITSSET;
5637 			cfp->cl_flags = NFSLCK_WRITE;
5638 			cfp->cl_ownerlen = 5;
5639 			NFSBCOPY("LOCAL", cfp->cl_owner, 5);
5640 		}
5641 		nfsrv_locallock_rollback(vp, lfp, p);
5642 	} else if (ltype != F_UNLCK) {
5643 		rlp = malloc(sizeof (struct nfsrollback), M_NFSDROLLBACK,
5644 		    M_WAITOK);
5645 		rlp->rlck_first = first;
5646 		rlp->rlck_end = end;
5647 		rlp->rlck_type = oldltype;
5648 		LIST_INSERT_HEAD(&lfp->lf_rollback, rlp, rlck_list);
5649 	}
5650 
5651 out:
5652 	NFSEXITCODE(error);
5653 	return (error);
5654 }
5655 
5656 /*
5657  * Roll back local lock changes and free up the rollback list.
5658  */
5659 static void
5660 nfsrv_locallock_rollback(vnode_t vp, struct nfslockfile *lfp, NFSPROC_T *p)
5661 {
5662 	struct nfsrollback *rlp, *nrlp;
5663 
5664 	LIST_FOREACH_SAFE(rlp, &lfp->lf_rollback, rlck_list, nrlp) {
5665 		(void) nfsvno_advlock(vp, rlp->rlck_type, rlp->rlck_first,
5666 		    rlp->rlck_end, p);
5667 		free(rlp, M_NFSDROLLBACK);
5668 	}
5669 	LIST_INIT(&lfp->lf_rollback);
5670 }
5671 
5672 /*
5673  * Update local lock list and delete rollback list (ie now committed to the
5674  * local locks). Most of the work is done by the internal function.
5675  */
5676 static void
5677 nfsrv_locallock_commit(struct nfslockfile *lfp, int flags, uint64_t first,
5678     uint64_t end)
5679 {
5680 	struct nfsrollback *rlp, *nrlp;
5681 	struct nfslock *new_lop, *other_lop;
5682 
5683 	new_lop = malloc(sizeof (struct nfslock), M_NFSDLOCK, M_WAITOK);
5684 	if (flags & (NFSLCK_READ | NFSLCK_WRITE))
5685 		other_lop = malloc(sizeof (struct nfslock), M_NFSDLOCK,
5686 		    M_WAITOK);
5687 	else
5688 		other_lop = NULL;
5689 	new_lop->lo_flags = flags;
5690 	new_lop->lo_first = first;
5691 	new_lop->lo_end = end;
5692 	nfsrv_updatelock(NULL, &new_lop, &other_lop, lfp);
5693 	if (new_lop != NULL)
5694 		free(new_lop, M_NFSDLOCK);
5695 	if (other_lop != NULL)
5696 		free(other_lop, M_NFSDLOCK);
5697 
5698 	/* and get rid of the rollback list */
5699 	LIST_FOREACH_SAFE(rlp, &lfp->lf_rollback, rlck_list, nrlp)
5700 		free(rlp, M_NFSDROLLBACK);
5701 	LIST_INIT(&lfp->lf_rollback);
5702 }
5703 
5704 /*
5705  * Lock the struct nfslockfile for local lock updating.
5706  */
5707 static void
5708 nfsrv_locklf(struct nfslockfile *lfp)
5709 {
5710 	int gotlock;
5711 
5712 	/* lf_usecount ensures *lfp won't be free'd */
5713 	lfp->lf_usecount++;
5714 	do {
5715 		gotlock = nfsv4_lock(&lfp->lf_locallock_lck, 1, NULL,
5716 		    NFSSTATEMUTEXPTR, NULL);
5717 	} while (gotlock == 0);
5718 	lfp->lf_usecount--;
5719 }
5720 
5721 /*
5722  * Unlock the struct nfslockfile after local lock updating.
5723  */
5724 static void
5725 nfsrv_unlocklf(struct nfslockfile *lfp)
5726 {
5727 
5728 	nfsv4_unlock(&lfp->lf_locallock_lck, 0);
5729 }
5730 
5731 /*
5732  * Clear out all state for the NFSv4 server.
5733  * Must be called by a thread that can sleep when no nfsds are running.
5734  */
5735 void
5736 nfsrv_throwawayallstate(NFSPROC_T *p)
5737 {
5738 	struct nfsclient *clp, *nclp;
5739 	struct nfslockfile *lfp, *nlfp;
5740 	int i;
5741 
5742 	/*
5743 	 * For each client, clean out the state and then free the structure.
5744 	 */
5745 	for (i = 0; i < nfsrv_clienthashsize; i++) {
5746 		LIST_FOREACH_SAFE(clp, &nfsclienthash[i], lc_hash, nclp) {
5747 			nfsrv_cleanclient(clp, p);
5748 			nfsrv_freedeleglist(&clp->lc_deleg);
5749 			nfsrv_freedeleglist(&clp->lc_olddeleg);
5750 			free(clp->lc_stateid, M_NFSDCLIENT);
5751 			free(clp, M_NFSDCLIENT);
5752 		}
5753 	}
5754 
5755 	/*
5756 	 * Also, free up any remaining lock file structures.
5757 	 */
5758 	for (i = 0; i < nfsrv_lockhashsize; i++) {
5759 		LIST_FOREACH_SAFE(lfp, &nfslockhash[i], lf_hash, nlfp) {
5760 			printf("nfsd unload: fnd a lock file struct\n");
5761 			nfsrv_freenfslockfile(lfp);
5762 		}
5763 	}
5764 }
5765 
5766 /*
5767  * Check the sequence# for the session and slot provided as an argument.
5768  * Also, renew the lease if the session will return NFS_OK.
5769  */
5770 int
5771 nfsrv_checksequence(struct nfsrv_descript *nd, uint32_t sequenceid,
5772     uint32_t *highest_slotidp, uint32_t *target_highest_slotidp, int cache_this,
5773     uint32_t *sflagsp, NFSPROC_T *p)
5774 {
5775 	struct nfsdsession *sep;
5776 	struct nfssessionhash *shp;
5777 	int error;
5778 	SVCXPRT *savxprt;
5779 
5780 	shp = NFSSESSIONHASH(nd->nd_sessionid);
5781 	NFSLOCKSESSION(shp);
5782 	sep = nfsrv_findsession(nd->nd_sessionid);
5783 	if (sep == NULL) {
5784 		NFSUNLOCKSESSION(shp);
5785 		return (NFSERR_BADSESSION);
5786 	}
5787 	error = nfsv4_seqsession(sequenceid, nd->nd_slotid, *highest_slotidp,
5788 	    sep->sess_slots, NULL, NFSV4_SLOTS - 1);
5789 	if (error != 0) {
5790 		NFSUNLOCKSESSION(shp);
5791 		return (error);
5792 	}
5793 	if (cache_this != 0)
5794 		nd->nd_flag |= ND_SAVEREPLY;
5795 	/* Renew the lease. */
5796 	sep->sess_clp->lc_expiry = nfsrv_leaseexpiry();
5797 	nd->nd_clientid.qval = sep->sess_clp->lc_clientid.qval;
5798 	nd->nd_flag |= ND_IMPLIEDCLID;
5799 
5800 	/*
5801 	 * If this session handles the backchannel, save the nd_xprt for this
5802 	 * RPC, since this is the one being used.
5803 	 */
5804 	if (sep->sess_clp->lc_req.nr_client != NULL &&
5805 	    (sep->sess_crflags & NFSV4CRSESS_CONNBACKCHAN) != 0) {
5806 		savxprt = sep->sess_cbsess.nfsess_xprt;
5807 		SVC_ACQUIRE(nd->nd_xprt);
5808 		nd->nd_xprt->xp_p2 =
5809 		    sep->sess_clp->lc_req.nr_client->cl_private;
5810 		nd->nd_xprt->xp_idletimeout = 0;	/* Disable timeout. */
5811 		sep->sess_cbsess.nfsess_xprt = nd->nd_xprt;
5812 		if (savxprt != NULL)
5813 			SVC_RELEASE(savxprt);
5814 	}
5815 
5816 	*sflagsp = 0;
5817 	if (sep->sess_clp->lc_req.nr_client == NULL)
5818 		*sflagsp |= NFSV4SEQ_CBPATHDOWN;
5819 	NFSUNLOCKSESSION(shp);
5820 	if (error == NFSERR_EXPIRED) {
5821 		*sflagsp |= NFSV4SEQ_EXPIREDALLSTATEREVOKED;
5822 		error = 0;
5823 	} else if (error == NFSERR_ADMINREVOKED) {
5824 		*sflagsp |= NFSV4SEQ_ADMINSTATEREVOKED;
5825 		error = 0;
5826 	}
5827 	*highest_slotidp = *target_highest_slotidp = NFSV4_SLOTS - 1;
5828 	return (0);
5829 }
5830 
5831 /*
5832  * Check/set reclaim complete for this session/clientid.
5833  */
5834 int
5835 nfsrv_checkreclaimcomplete(struct nfsrv_descript *nd)
5836 {
5837 	struct nfsdsession *sep;
5838 	struct nfssessionhash *shp;
5839 	int error = 0;
5840 
5841 	shp = NFSSESSIONHASH(nd->nd_sessionid);
5842 	NFSLOCKSTATE();
5843 	NFSLOCKSESSION(shp);
5844 	sep = nfsrv_findsession(nd->nd_sessionid);
5845 	if (sep == NULL) {
5846 		NFSUNLOCKSESSION(shp);
5847 		NFSUNLOCKSTATE();
5848 		return (NFSERR_BADSESSION);
5849 	}
5850 
5851 	/* Check to see if reclaim complete has already happened. */
5852 	if ((sep->sess_clp->lc_flags & LCL_RECLAIMCOMPLETE) != 0)
5853 		error = NFSERR_COMPLETEALREADY;
5854 	else
5855 		sep->sess_clp->lc_flags |= LCL_RECLAIMCOMPLETE;
5856 	NFSUNLOCKSESSION(shp);
5857 	NFSUNLOCKSTATE();
5858 	return (error);
5859 }
5860 
5861 /*
5862  * Cache the reply in a session slot.
5863  */
5864 void
5865 nfsrv_cache_session(uint8_t *sessionid, uint32_t slotid, int repstat,
5866    struct mbuf **m)
5867 {
5868 	struct nfsdsession *sep;
5869 	struct nfssessionhash *shp;
5870 
5871 	shp = NFSSESSIONHASH(sessionid);
5872 	NFSLOCKSESSION(shp);
5873 	sep = nfsrv_findsession(sessionid);
5874 	if (sep == NULL) {
5875 		NFSUNLOCKSESSION(shp);
5876 		printf("nfsrv_cache_session: no session\n");
5877 		m_freem(*m);
5878 		return;
5879 	}
5880 	nfsv4_seqsess_cacherep(slotid, sep->sess_slots, repstat, m);
5881 	NFSUNLOCKSESSION(shp);
5882 }
5883 
5884 /*
5885  * Search for a session that matches the sessionid.
5886  */
5887 static struct nfsdsession *
5888 nfsrv_findsession(uint8_t *sessionid)
5889 {
5890 	struct nfsdsession *sep;
5891 	struct nfssessionhash *shp;
5892 
5893 	shp = NFSSESSIONHASH(sessionid);
5894 	LIST_FOREACH(sep, &shp->list, sess_hash) {
5895 		if (!NFSBCMP(sessionid, sep->sess_sessionid, NFSX_V4SESSIONID))
5896 			break;
5897 	}
5898 	return (sep);
5899 }
5900 
5901 /*
5902  * Destroy a session.
5903  */
5904 int
5905 nfsrv_destroysession(struct nfsrv_descript *nd, uint8_t *sessionid)
5906 {
5907 	int error, samesess;
5908 
5909 	samesess = 0;
5910 	if (!NFSBCMP(sessionid, nd->nd_sessionid, NFSX_V4SESSIONID)) {
5911 		samesess = 1;
5912 		if ((nd->nd_flag & ND_LASTOP) == 0)
5913 			return (NFSERR_BADSESSION);
5914 	}
5915 	error = nfsrv_freesession(NULL, sessionid);
5916 	if (error == 0 && samesess != 0)
5917 		nd->nd_flag &= ~ND_HASSEQUENCE;
5918 	return (error);
5919 }
5920 
5921 /*
5922  * Free up a session structure.
5923  */
5924 static int
5925 nfsrv_freesession(struct nfsdsession *sep, uint8_t *sessionid)
5926 {
5927 	struct nfssessionhash *shp;
5928 	int i;
5929 
5930 	NFSLOCKSTATE();
5931 	if (sep == NULL) {
5932 		shp = NFSSESSIONHASH(sessionid);
5933 		NFSLOCKSESSION(shp);
5934 		sep = nfsrv_findsession(sessionid);
5935 	} else {
5936 		shp = NFSSESSIONHASH(sep->sess_sessionid);
5937 		NFSLOCKSESSION(shp);
5938 	}
5939 	if (sep != NULL) {
5940 		sep->sess_refcnt--;
5941 		if (sep->sess_refcnt > 0) {
5942 			NFSUNLOCKSESSION(shp);
5943 			NFSUNLOCKSTATE();
5944 			return (0);
5945 		}
5946 		LIST_REMOVE(sep, sess_hash);
5947 		LIST_REMOVE(sep, sess_list);
5948 	}
5949 	NFSUNLOCKSESSION(shp);
5950 	NFSUNLOCKSTATE();
5951 	if (sep == NULL)
5952 		return (NFSERR_BADSESSION);
5953 	for (i = 0; i < NFSV4_SLOTS; i++)
5954 		if (sep->sess_slots[i].nfssl_reply != NULL)
5955 			m_freem(sep->sess_slots[i].nfssl_reply);
5956 	if (sep->sess_cbsess.nfsess_xprt != NULL)
5957 		SVC_RELEASE(sep->sess_cbsess.nfsess_xprt);
5958 	free(sep, M_NFSDSESSION);
5959 	return (0);
5960 }
5961 
5962 /*
5963  * Free a stateid.
5964  * RFC5661 says that it should fail when there are associated opens, locks
5965  * or delegations. Since stateids represent opens, I don't see how you can
5966  * free an open stateid (it will be free'd when closed), so this function
5967  * only works for lock stateids (freeing the lock_owner) or delegations.
5968  */
5969 int
5970 nfsrv_freestateid(struct nfsrv_descript *nd, nfsv4stateid_t *stateidp,
5971     NFSPROC_T *p)
5972 {
5973 	struct nfsclient *clp;
5974 	struct nfsstate *stp;
5975 	int error;
5976 
5977 	NFSLOCKSTATE();
5978 	/*
5979 	 * Look up the stateid
5980 	 */
5981 	error = nfsrv_getclient((nfsquad_t)((u_quad_t)0), CLOPS_RENEW, &clp,
5982 	    NULL, (nfsquad_t)((u_quad_t)0), 0, nd, p);
5983 	if (error == 0) {
5984 		/* First, check for a delegation. */
5985 		LIST_FOREACH(stp, &clp->lc_deleg, ls_list) {
5986 			if (!NFSBCMP(stp->ls_stateid.other, stateidp->other,
5987 			    NFSX_STATEIDOTHER))
5988 				break;
5989 		}
5990 		if (stp != NULL) {
5991 			nfsrv_freedeleg(stp);
5992 			NFSUNLOCKSTATE();
5993 			return (error);
5994 		}
5995 	}
5996 	/* Not a delegation, try for a lock_owner. */
5997 	if (error == 0)
5998 		error = nfsrv_getstate(clp, stateidp, 0, &stp);
5999 	if (error == 0 && ((stp->ls_flags & (NFSLCK_OPEN | NFSLCK_DELEGREAD |
6000 	    NFSLCK_DELEGWRITE)) != 0 || (stp->ls_flags & NFSLCK_LOCK) == 0))
6001 		/* Not a lock_owner stateid. */
6002 		error = NFSERR_LOCKSHELD;
6003 	if (error == 0 && !LIST_EMPTY(&stp->ls_lock))
6004 		error = NFSERR_LOCKSHELD;
6005 	if (error == 0)
6006 		nfsrv_freelockowner(stp, NULL, 0, p);
6007 	NFSUNLOCKSTATE();
6008 	return (error);
6009 }
6010 
6011 /*
6012  * Generate the xdr for an NFSv4.1 CBSequence Operation.
6013  */
6014 static int
6015 nfsv4_setcbsequence(struct nfsrv_descript *nd, struct nfsclient *clp,
6016     int dont_replycache, struct nfsdsession **sepp)
6017 {
6018 	struct nfsdsession *sep;
6019 	uint32_t *tl, slotseq = 0;
6020 	int maxslot, slotpos;
6021 	uint8_t sessionid[NFSX_V4SESSIONID];
6022 	int error;
6023 
6024 	error = nfsv4_getcbsession(clp, sepp);
6025 	if (error != 0)
6026 		return (error);
6027 	sep = *sepp;
6028 	(void)nfsv4_sequencelookup(NULL, &sep->sess_cbsess, &slotpos, &maxslot,
6029 	    &slotseq, sessionid);
6030 	KASSERT(maxslot >= 0, ("nfsv4_setcbsequence neg maxslot"));
6031 
6032 	/* Build the Sequence arguments. */
6033 	NFSM_BUILD(tl, uint32_t *, NFSX_V4SESSIONID + 5 * NFSX_UNSIGNED);
6034 	bcopy(sessionid, tl, NFSX_V4SESSIONID);
6035 	tl += NFSX_V4SESSIONID / NFSX_UNSIGNED;
6036 	nd->nd_slotseq = tl;
6037 	*tl++ = txdr_unsigned(slotseq);
6038 	*tl++ = txdr_unsigned(slotpos);
6039 	*tl++ = txdr_unsigned(maxslot);
6040 	if (dont_replycache == 0)
6041 		*tl++ = newnfs_true;
6042 	else
6043 		*tl++ = newnfs_false;
6044 	*tl = 0;			/* No referring call list, for now. */
6045 	nd->nd_flag |= ND_HASSEQUENCE;
6046 	return (0);
6047 }
6048 
6049 /*
6050  * Get a session for the callback.
6051  */
6052 static int
6053 nfsv4_getcbsession(struct nfsclient *clp, struct nfsdsession **sepp)
6054 {
6055 	struct nfsdsession *sep;
6056 
6057 	NFSLOCKSTATE();
6058 	LIST_FOREACH(sep, &clp->lc_session, sess_list) {
6059 		if ((sep->sess_crflags & NFSV4CRSESS_CONNBACKCHAN) != 0)
6060 			break;
6061 	}
6062 	if (sep == NULL) {
6063 		NFSUNLOCKSTATE();
6064 		return (NFSERR_BADSESSION);
6065 	}
6066 	sep->sess_refcnt++;
6067 	*sepp = sep;
6068 	NFSUNLOCKSTATE();
6069 	return (0);
6070 }
6071 
6072 /*
6073  * Free up all backchannel xprts.  This needs to be done when the nfsd threads
6074  * exit, since those transports will all be going away.
6075  * This is only called after all the nfsd threads are done performing RPCs,
6076  * so locking shouldn't be an issue.
6077  */
6078 APPLESTATIC void
6079 nfsrv_freeallbackchannel_xprts(void)
6080 {
6081 	struct nfsdsession *sep;
6082 	struct nfsclient *clp;
6083 	SVCXPRT *xprt;
6084 	int i;
6085 
6086 	for (i = 0; i < nfsrv_clienthashsize; i++) {
6087 		LIST_FOREACH(clp, &nfsclienthash[i], lc_hash) {
6088 			LIST_FOREACH(sep, &clp->lc_session, sess_list) {
6089 				xprt = sep->sess_cbsess.nfsess_xprt;
6090 				sep->sess_cbsess.nfsess_xprt = NULL;
6091 				if (xprt != NULL)
6092 					SVC_RELEASE(xprt);
6093 			}
6094 		}
6095 	}
6096 }
6097 
6098