xref: /freebsd/sys/fs/nfsserver/nfs_nfsdstate.c (revision 076ad2f8)
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 	int gotstate = 0, len = 0;
2505 	u_char *clidp = NULL;
2506 
2507 	if ((new_stp->ls_flags & NFSLCK_SHAREBITS) == NFSLCK_READACCESS)
2508 		readonly = 1;
2509 	/*
2510 	 * Check for restart conditions (client and server).
2511 	 * (Paranoia, should have been detected by nfsrv_opencheck().)
2512 	 * If an error does show up, return NFSERR_EXPIRED, since the
2513 	 * the seqid# has already been incremented.
2514 	 */
2515 	error = nfsrv_checkrestart(clientid, new_stp->ls_flags,
2516 	    &new_stp->ls_stateid, 0);
2517 	if (error) {
2518 		printf("Nfsd: openctrl unexpected restart err=%d\n",
2519 		    error);
2520 		error = NFSERR_EXPIRED;
2521 		goto out;
2522 	}
2523 
2524 	clidp = malloc(NFSV4_OPAQUELIMIT, M_TEMP, M_WAITOK);
2525 tryagain:
2526 	MALLOC(new_lfp, struct nfslockfile *, sizeof (struct nfslockfile),
2527 	    M_NFSDLOCKFILE, M_WAITOK);
2528 	MALLOC(new_open, struct nfsstate *, sizeof (struct nfsstate),
2529 	    M_NFSDSTATE, M_WAITOK);
2530 	MALLOC(new_deleg, struct nfsstate *, sizeof (struct nfsstate),
2531 	    M_NFSDSTATE, M_WAITOK);
2532 	getfhret = nfsrv_getlockfh(vp, new_stp->ls_flags, new_lfp,
2533 	    NULL, p);
2534 	NFSLOCKSTATE();
2535 	/*
2536 	 * Get the client structure. Since the linked lists could be changed
2537 	 * by other nfsd processes if this process does a tsleep(), one of
2538 	 * two things must be done.
2539 	 * 1 - don't tsleep()
2540 	 * or
2541 	 * 2 - get the nfsv4_lock() { indicated by haslock == 1 }
2542 	 *     before using the lists, since this lock stops the other
2543 	 *     nfsd. This should only be used for rare cases, since it
2544 	 *     essentially single threads the nfsd.
2545 	 *     At this time, it is only done for cases where the stable
2546 	 *     storage file must be written prior to completion of state
2547 	 *     expiration.
2548 	 */
2549 	error = nfsrv_getclient(clientid, CLOPS_RENEW, &clp, NULL,
2550 	    (nfsquad_t)((u_quad_t)0), 0, nd, p);
2551 	if (!error && (clp->lc_flags & LCL_NEEDSCBNULL) &&
2552 	    clp->lc_program) {
2553 		/*
2554 		 * This happens on the first open for a client
2555 		 * that supports callbacks.
2556 		 */
2557 		NFSUNLOCKSTATE();
2558 		/*
2559 		 * Although nfsrv_docallback() will sleep, clp won't
2560 		 * go away, since they are only removed when the
2561 		 * nfsv4_lock() has blocked the nfsd threads. The
2562 		 * fields in clp can change, but having multiple
2563 		 * threads do this Null callback RPC should be
2564 		 * harmless.
2565 		 */
2566 		cbret = nfsrv_docallback(clp, NFSV4PROC_CBNULL,
2567 		    NULL, 0, NULL, NULL, NULL, p);
2568 		NFSLOCKSTATE();
2569 		clp->lc_flags &= ~LCL_NEEDSCBNULL;
2570 		if (!cbret)
2571 			clp->lc_flags |= LCL_CALLBACKSON;
2572 	}
2573 
2574 	/*
2575 	 * Look up the open owner. See if it needs confirmation and
2576 	 * check the seq#, as required.
2577 	 */
2578 	if (!error)
2579 		nfsrv_getowner(&clp->lc_open, new_stp, &ownerstp);
2580 
2581 	if (error) {
2582 		NFSUNLOCKSTATE();
2583 		printf("Nfsd: openctrl unexpected state err=%d\n",
2584 			error);
2585 		free((caddr_t)new_lfp, M_NFSDLOCKFILE);
2586 		free((caddr_t)new_open, M_NFSDSTATE);
2587 		free((caddr_t)new_deleg, M_NFSDSTATE);
2588 		if (haslock) {
2589 			NFSLOCKV4ROOTMUTEX();
2590 			nfsv4_unlock(&nfsv4rootfs_lock, 1);
2591 			NFSUNLOCKV4ROOTMUTEX();
2592 		}
2593 		error = NFSERR_EXPIRED;
2594 		goto out;
2595 	}
2596 
2597 	if (new_stp->ls_flags & NFSLCK_RECLAIM)
2598 		nfsrv_markstable(clp);
2599 
2600 	/*
2601 	 * Get the structure for the underlying file.
2602 	 */
2603 	if (getfhret)
2604 		error = getfhret;
2605 	else
2606 		error = nfsrv_getlockfile(new_stp->ls_flags, &new_lfp, &lfp,
2607 		    NULL, 0);
2608 	if (new_lfp)
2609 		FREE((caddr_t)new_lfp, M_NFSDLOCKFILE);
2610 	if (error) {
2611 		NFSUNLOCKSTATE();
2612 		printf("Nfsd openctrl unexpected getlockfile err=%d\n",
2613 		    error);
2614 		free((caddr_t)new_open, M_NFSDSTATE);
2615 		free((caddr_t)new_deleg, M_NFSDSTATE);
2616 		if (haslock) {
2617 			NFSLOCKV4ROOTMUTEX();
2618 			nfsv4_unlock(&nfsv4rootfs_lock, 1);
2619 			NFSUNLOCKV4ROOTMUTEX();
2620 		}
2621 		goto out;
2622 	}
2623 
2624 	/*
2625 	 * Search for a conflicting open/share.
2626 	 */
2627 	if (new_stp->ls_flags & NFSLCK_DELEGCUR) {
2628 	    /*
2629 	     * For Delegate_Cur, search for the matching Delegation,
2630 	     * which indicates no conflict.
2631 	     * An old delegation should have been recovered by the
2632 	     * client doing a Claim_DELEGATE_Prev, so I won't let
2633 	     * it match and return NFSERR_EXPIRED. Should I let it
2634 	     * match?
2635 	     */
2636 	    LIST_FOREACH(stp, &lfp->lf_deleg, ls_file) {
2637 		if (!(stp->ls_flags & NFSLCK_OLDDELEG) &&
2638 		    (((nd->nd_flag & ND_NFSV41) != 0 &&
2639 		    stateidp->seqid == 0) ||
2640 		    stateidp->seqid == stp->ls_stateid.seqid) &&
2641 		    !NFSBCMP(stateidp->other, stp->ls_stateid.other,
2642 			NFSX_STATEIDOTHER))
2643 			break;
2644 	    }
2645 	    if (stp == LIST_END(&lfp->lf_deleg) ||
2646 		((new_stp->ls_flags & NFSLCK_WRITEACCESS) &&
2647 		 (stp->ls_flags & NFSLCK_DELEGREAD))) {
2648 		NFSUNLOCKSTATE();
2649 		printf("Nfsd openctrl unexpected expiry\n");
2650 		free((caddr_t)new_open, M_NFSDSTATE);
2651 		free((caddr_t)new_deleg, M_NFSDSTATE);
2652 		if (haslock) {
2653 			NFSLOCKV4ROOTMUTEX();
2654 			nfsv4_unlock(&nfsv4rootfs_lock, 1);
2655 			NFSUNLOCKV4ROOTMUTEX();
2656 		}
2657 		error = NFSERR_EXPIRED;
2658 		goto out;
2659 	    }
2660 
2661 	    /*
2662 	     * Don't issue a Delegation, since one already exists and
2663 	     * delay delegation timeout, as required.
2664 	     */
2665 	    delegate = 0;
2666 	    nfsrv_delaydelegtimeout(stp);
2667 	}
2668 
2669 	/*
2670 	 * Check for access/deny bit conflicts. I also check for the
2671 	 * same owner, since the client might not have bothered to check.
2672 	 * Also, note an open for the same file and owner, if found,
2673 	 * which is all we do here for Delegate_Cur, since conflict
2674 	 * checking is already done.
2675 	 */
2676 	LIST_FOREACH(stp, &lfp->lf_open, ls_file) {
2677 		if (ownerstp && stp->ls_openowner == ownerstp)
2678 			openstp = stp;
2679 		if (!(new_stp->ls_flags & NFSLCK_DELEGCUR)) {
2680 		    /*
2681 		     * If another client has the file open, the only
2682 		     * delegation that can be issued is a Read delegation
2683 		     * and only if it is a Read open with Deny none.
2684 		     */
2685 		    if (clp != stp->ls_clp) {
2686 			if ((stp->ls_flags & NFSLCK_SHAREBITS) ==
2687 			    NFSLCK_READACCESS)
2688 			    writedeleg = 0;
2689 			else
2690 			    delegate = 0;
2691 		    }
2692 		    if(((new_stp->ls_flags & NFSLCK_ACCESSBITS) &
2693 		        ((stp->ls_flags>>NFSLCK_SHIFT) & NFSLCK_ACCESSBITS))||
2694 		       ((stp->ls_flags & NFSLCK_ACCESSBITS) &
2695 		        ((new_stp->ls_flags>>NFSLCK_SHIFT)&NFSLCK_ACCESSBITS))){
2696 			ret = nfsrv_clientconflict(stp->ls_clp,&haslock,vp,p);
2697 			if (ret == 1) {
2698 				/*
2699 				 * nfsrv_clientconflict() unlocks state
2700 				 * when it returns non-zero.
2701 				 */
2702 				free((caddr_t)new_open, M_NFSDSTATE);
2703 				free((caddr_t)new_deleg, M_NFSDSTATE);
2704 				openstp = NULL;
2705 				goto tryagain;
2706 			}
2707 			if (ret == 2)
2708 				error = NFSERR_PERM;
2709 			else if (new_stp->ls_flags & NFSLCK_RECLAIM)
2710 				error = NFSERR_RECLAIMCONFLICT;
2711 			else
2712 				error = NFSERR_SHAREDENIED;
2713 			if (ret == 0)
2714 				NFSUNLOCKSTATE();
2715 			if (haslock) {
2716 				NFSLOCKV4ROOTMUTEX();
2717 				nfsv4_unlock(&nfsv4rootfs_lock, 1);
2718 				NFSUNLOCKV4ROOTMUTEX();
2719 			}
2720 			free((caddr_t)new_open, M_NFSDSTATE);
2721 			free((caddr_t)new_deleg, M_NFSDSTATE);
2722 			printf("nfsd openctrl unexpected client cnfl\n");
2723 			goto out;
2724 		    }
2725 		}
2726 	}
2727 
2728 	/*
2729 	 * Check for a conflicting delegation. If one is found, call
2730 	 * nfsrv_delegconflict() to handle it. If the v4root lock hasn't
2731 	 * been set yet, it will get the lock. Otherwise, it will recall
2732 	 * the delegation. Then, we try try again...
2733 	 * (If NFSLCK_DELEGCUR is set, it has a delegation, so there
2734 	 *  isn't a conflict.)
2735 	 * I currently believe the conflict algorithm to be:
2736 	 * For Open with Read Access and Deny None
2737 	 * - there is a conflict iff a different client has a write delegation
2738 	 * For Open with other Write Access or any Deny except None
2739 	 * - there is a conflict if a different client has any delegation
2740 	 * - there is a conflict if the same client has a read delegation
2741 	 *   (The current consensus is that this last case should be
2742 	 *    considered a conflict since the client with a read delegation
2743 	 *    could have done an Open with ReadAccess and WriteDeny
2744 	 *    locally and then not have checked for the WriteDeny.)
2745 	 */
2746 	if (!(new_stp->ls_flags & (NFSLCK_DELEGPREV | NFSLCK_DELEGCUR))) {
2747 	    stp = LIST_FIRST(&lfp->lf_deleg);
2748 	    while (stp != LIST_END(&lfp->lf_deleg)) {
2749 		nstp = LIST_NEXT(stp, ls_file);
2750 		if (stp->ls_clp != clp && (stp->ls_flags & NFSLCK_DELEGREAD))
2751 			writedeleg = 0;
2752 		else
2753 			delegate = 0;
2754 		if ((readonly && stp->ls_clp != clp &&
2755 		       (stp->ls_flags & NFSLCK_DELEGWRITE)) ||
2756 		    (!readonly && (stp->ls_clp != clp ||
2757 		         (stp->ls_flags & NFSLCK_DELEGREAD)))) {
2758 		    if (new_stp->ls_flags & NFSLCK_RECLAIM) {
2759 			delegate = 2;
2760 		    } else {
2761 			ret = nfsrv_delegconflict(stp, &haslock, p, vp);
2762 			if (ret) {
2763 			    /*
2764 			     * nfsrv_delegconflict() unlocks state
2765 			     * when it returns non-zero.
2766 			     */
2767 			    printf("Nfsd openctrl unexpected deleg cnfl\n");
2768 			    free((caddr_t)new_open, M_NFSDSTATE);
2769 			    free((caddr_t)new_deleg, M_NFSDSTATE);
2770 			    if (ret == -1) {
2771 				openstp = NULL;
2772 				goto tryagain;
2773 			    }
2774 			    error = ret;
2775 			    goto out;
2776 			}
2777 		    }
2778 		}
2779 		stp = nstp;
2780 	    }
2781 	}
2782 
2783 	/*
2784 	 * We only get here if there was no open that conflicted.
2785 	 * If an open for the owner exists, or in the access/deny bits.
2786 	 * Otherwise it is a new open. If the open_owner hasn't been
2787 	 * confirmed, replace the open with the new one needing confirmation,
2788 	 * otherwise add the open.
2789 	 */
2790 	if (new_stp->ls_flags & NFSLCK_DELEGPREV) {
2791 	    /*
2792 	     * Handle NFSLCK_DELEGPREV by searching the old delegations for
2793 	     * a match. If found, just move the old delegation to the current
2794 	     * delegation list and issue open. If not found, return
2795 	     * NFSERR_EXPIRED.
2796 	     */
2797 	    LIST_FOREACH(stp, &clp->lc_olddeleg, ls_list) {
2798 		if (stp->ls_lfp == lfp) {
2799 		    /* Found it */
2800 		    if (stp->ls_clp != clp)
2801 			panic("olddeleg clp");
2802 		    LIST_REMOVE(stp, ls_list);
2803 		    LIST_REMOVE(stp, ls_hash);
2804 		    stp->ls_flags &= ~NFSLCK_OLDDELEG;
2805 		    stp->ls_stateid.seqid = delegstateidp->seqid = 1;
2806 		    stp->ls_stateid.other[0] = delegstateidp->other[0] =
2807 			clp->lc_clientid.lval[0];
2808 		    stp->ls_stateid.other[1] = delegstateidp->other[1] =
2809 			clp->lc_clientid.lval[1];
2810 		    stp->ls_stateid.other[2] = delegstateidp->other[2] =
2811 			nfsrv_nextstateindex(clp);
2812 		    stp->ls_compref = nd->nd_compref;
2813 		    LIST_INSERT_HEAD(&clp->lc_deleg, stp, ls_list);
2814 		    LIST_INSERT_HEAD(NFSSTATEHASH(clp,
2815 			stp->ls_stateid), stp, ls_hash);
2816 		    if (stp->ls_flags & NFSLCK_DELEGWRITE)
2817 			*rflagsp |= NFSV4OPEN_WRITEDELEGATE;
2818 		    else
2819 			*rflagsp |= NFSV4OPEN_READDELEGATE;
2820 		    clp->lc_delegtime = NFSD_MONOSEC +
2821 			nfsrv_lease + NFSRV_LEASEDELTA;
2822 
2823 		    /*
2824 		     * Now, do the associated open.
2825 		     */
2826 		    new_open->ls_stateid.seqid = 1;
2827 		    new_open->ls_stateid.other[0] = clp->lc_clientid.lval[0];
2828 		    new_open->ls_stateid.other[1] = clp->lc_clientid.lval[1];
2829 		    new_open->ls_stateid.other[2] = nfsrv_nextstateindex(clp);
2830 		    new_open->ls_flags = (new_stp->ls_flags&NFSLCK_DENYBITS)|
2831 			NFSLCK_OPEN;
2832 		    if (stp->ls_flags & NFSLCK_DELEGWRITE)
2833 			new_open->ls_flags |= (NFSLCK_READACCESS |
2834 			    NFSLCK_WRITEACCESS);
2835 		    else
2836 			new_open->ls_flags |= NFSLCK_READACCESS;
2837 		    new_open->ls_uid = new_stp->ls_uid;
2838 		    new_open->ls_lfp = lfp;
2839 		    new_open->ls_clp = clp;
2840 		    LIST_INIT(&new_open->ls_open);
2841 		    LIST_INSERT_HEAD(&lfp->lf_open, new_open, ls_file);
2842 		    LIST_INSERT_HEAD(NFSSTATEHASH(clp, new_open->ls_stateid),
2843 			new_open, ls_hash);
2844 		    /*
2845 		     * and handle the open owner
2846 		     */
2847 		    if (ownerstp) {
2848 			new_open->ls_openowner = ownerstp;
2849 			LIST_INSERT_HEAD(&ownerstp->ls_open,new_open,ls_list);
2850 		    } else {
2851 			new_open->ls_openowner = new_stp;
2852 			new_stp->ls_flags = 0;
2853 			nfsrvd_refcache(new_stp->ls_op);
2854 			new_stp->ls_noopens = 0;
2855 			LIST_INIT(&new_stp->ls_open);
2856 			LIST_INSERT_HEAD(&new_stp->ls_open, new_open, ls_list);
2857 			LIST_INSERT_HEAD(&clp->lc_open, new_stp, ls_list);
2858 			*new_stpp = NULL;
2859 			nfsstatsv1.srvopenowners++;
2860 			nfsrv_openpluslock++;
2861 		    }
2862 		    openstp = new_open;
2863 		    new_open = NULL;
2864 		    nfsstatsv1.srvopens++;
2865 		    nfsrv_openpluslock++;
2866 		    break;
2867 		}
2868 	    }
2869 	    if (stp == LIST_END(&clp->lc_olddeleg))
2870 		error = NFSERR_EXPIRED;
2871 	} else if (new_stp->ls_flags & (NFSLCK_DELEGREAD | NFSLCK_DELEGWRITE)) {
2872 	    /*
2873 	     * Scan to see that no delegation for this client and file
2874 	     * doesn't already exist.
2875 	     * There also shouldn't yet be an Open for this file and
2876 	     * openowner.
2877 	     */
2878 	    LIST_FOREACH(stp, &lfp->lf_deleg, ls_file) {
2879 		if (stp->ls_clp == clp)
2880 		    break;
2881 	    }
2882 	    if (stp == LIST_END(&lfp->lf_deleg) && openstp == NULL) {
2883 		/*
2884 		 * This is the Claim_Previous case with a delegation
2885 		 * type != Delegate_None.
2886 		 */
2887 		/*
2888 		 * First, add the delegation. (Although we must issue the
2889 		 * delegation, we can also ask for an immediate return.)
2890 		 */
2891 		new_deleg->ls_stateid.seqid = delegstateidp->seqid = 1;
2892 		new_deleg->ls_stateid.other[0] = delegstateidp->other[0] =
2893 		    clp->lc_clientid.lval[0];
2894 		new_deleg->ls_stateid.other[1] = delegstateidp->other[1] =
2895 		    clp->lc_clientid.lval[1];
2896 		new_deleg->ls_stateid.other[2] = delegstateidp->other[2] =
2897 		    nfsrv_nextstateindex(clp);
2898 		if (new_stp->ls_flags & NFSLCK_DELEGWRITE) {
2899 		    new_deleg->ls_flags = (NFSLCK_DELEGWRITE |
2900 			NFSLCK_READACCESS | NFSLCK_WRITEACCESS);
2901 		    *rflagsp |= NFSV4OPEN_WRITEDELEGATE;
2902 		} else {
2903 		    new_deleg->ls_flags = (NFSLCK_DELEGREAD |
2904 			NFSLCK_READACCESS);
2905 		    *rflagsp |= NFSV4OPEN_READDELEGATE;
2906 		}
2907 		new_deleg->ls_uid = new_stp->ls_uid;
2908 		new_deleg->ls_lfp = lfp;
2909 		new_deleg->ls_clp = clp;
2910 		new_deleg->ls_filerev = filerev;
2911 		new_deleg->ls_compref = nd->nd_compref;
2912 		LIST_INSERT_HEAD(&lfp->lf_deleg, new_deleg, ls_file);
2913 		LIST_INSERT_HEAD(NFSSTATEHASH(clp,
2914 		    new_deleg->ls_stateid), new_deleg, ls_hash);
2915 		LIST_INSERT_HEAD(&clp->lc_deleg, new_deleg, ls_list);
2916 		new_deleg = NULL;
2917 		if (delegate == 2 || nfsrv_issuedelegs == 0 ||
2918 		    (clp->lc_flags & (LCL_CALLBACKSON | LCL_CBDOWN)) !=
2919 		     LCL_CALLBACKSON ||
2920 		    NFSRV_V4DELEGLIMIT(nfsrv_delegatecnt) ||
2921 		    !NFSVNO_DELEGOK(vp))
2922 		    *rflagsp |= NFSV4OPEN_RECALL;
2923 		nfsstatsv1.srvdelegates++;
2924 		nfsrv_openpluslock++;
2925 		nfsrv_delegatecnt++;
2926 
2927 		/*
2928 		 * Now, do the associated open.
2929 		 */
2930 		new_open->ls_stateid.seqid = 1;
2931 		new_open->ls_stateid.other[0] = clp->lc_clientid.lval[0];
2932 		new_open->ls_stateid.other[1] = clp->lc_clientid.lval[1];
2933 		new_open->ls_stateid.other[2] = nfsrv_nextstateindex(clp);
2934 		new_open->ls_flags = (new_stp->ls_flags & NFSLCK_DENYBITS) |
2935 		    NFSLCK_OPEN;
2936 		if (new_stp->ls_flags & NFSLCK_DELEGWRITE)
2937 			new_open->ls_flags |= (NFSLCK_READACCESS |
2938 			    NFSLCK_WRITEACCESS);
2939 		else
2940 			new_open->ls_flags |= NFSLCK_READACCESS;
2941 		new_open->ls_uid = new_stp->ls_uid;
2942 		new_open->ls_lfp = lfp;
2943 		new_open->ls_clp = clp;
2944 		LIST_INIT(&new_open->ls_open);
2945 		LIST_INSERT_HEAD(&lfp->lf_open, new_open, ls_file);
2946 		LIST_INSERT_HEAD(NFSSTATEHASH(clp, new_open->ls_stateid),
2947 		   new_open, ls_hash);
2948 		/*
2949 		 * and handle the open owner
2950 		 */
2951 		if (ownerstp) {
2952 		    new_open->ls_openowner = ownerstp;
2953 		    LIST_INSERT_HEAD(&ownerstp->ls_open, new_open, ls_list);
2954 		} else {
2955 		    new_open->ls_openowner = new_stp;
2956 		    new_stp->ls_flags = 0;
2957 		    nfsrvd_refcache(new_stp->ls_op);
2958 		    new_stp->ls_noopens = 0;
2959 		    LIST_INIT(&new_stp->ls_open);
2960 		    LIST_INSERT_HEAD(&new_stp->ls_open, new_open, ls_list);
2961 		    LIST_INSERT_HEAD(&clp->lc_open, new_stp, ls_list);
2962 		    *new_stpp = NULL;
2963 		    nfsstatsv1.srvopenowners++;
2964 		    nfsrv_openpluslock++;
2965 		}
2966 		openstp = new_open;
2967 		new_open = NULL;
2968 		nfsstatsv1.srvopens++;
2969 		nfsrv_openpluslock++;
2970 	    } else {
2971 		error = NFSERR_RECLAIMCONFLICT;
2972 	    }
2973 	} else if (ownerstp) {
2974 		if (ownerstp->ls_flags & NFSLCK_NEEDSCONFIRM) {
2975 		    /* Replace the open */
2976 		    if (ownerstp->ls_op)
2977 			nfsrvd_derefcache(ownerstp->ls_op);
2978 		    ownerstp->ls_op = new_stp->ls_op;
2979 		    nfsrvd_refcache(ownerstp->ls_op);
2980 		    ownerstp->ls_seq = new_stp->ls_seq;
2981 		    *rflagsp |= NFSV4OPEN_RESULTCONFIRM;
2982 		    stp = LIST_FIRST(&ownerstp->ls_open);
2983 		    stp->ls_flags = (new_stp->ls_flags & NFSLCK_SHAREBITS) |
2984 			NFSLCK_OPEN;
2985 		    stp->ls_stateid.seqid = 1;
2986 		    stp->ls_uid = new_stp->ls_uid;
2987 		    if (lfp != stp->ls_lfp) {
2988 			LIST_REMOVE(stp, ls_file);
2989 			LIST_INSERT_HEAD(&lfp->lf_open, stp, ls_file);
2990 			stp->ls_lfp = lfp;
2991 		    }
2992 		    openstp = stp;
2993 		} else if (openstp) {
2994 		    openstp->ls_flags |= (new_stp->ls_flags & NFSLCK_SHAREBITS);
2995 		    openstp->ls_stateid.seqid++;
2996 		    if ((nd->nd_flag & ND_NFSV41) != 0 &&
2997 			openstp->ls_stateid.seqid == 0)
2998 			openstp->ls_stateid.seqid = 1;
2999 
3000 		    /*
3001 		     * This is where we can choose to issue a delegation.
3002 		     */
3003 		    if (delegate == 0 || writedeleg == 0 ||
3004 			NFSVNO_EXRDONLY(exp) || (readonly != 0 &&
3005 			nfsrv_writedelegifpos == 0) ||
3006 			!NFSVNO_DELEGOK(vp) ||
3007 			(new_stp->ls_flags & NFSLCK_WANTRDELEG) != 0 ||
3008 			(clp->lc_flags & (LCL_CALLBACKSON | LCL_CBDOWN)) !=
3009 			 LCL_CALLBACKSON)
3010 			*rflagsp |= NFSV4OPEN_WDCONTENTION;
3011 		    else if (nfsrv_issuedelegs == 0 ||
3012 			NFSRV_V4DELEGLIMIT(nfsrv_delegatecnt))
3013 			*rflagsp |= NFSV4OPEN_WDRESOURCE;
3014 		    else if ((new_stp->ls_flags & NFSLCK_WANTNODELEG) != 0)
3015 			*rflagsp |= NFSV4OPEN_WDNOTWANTED;
3016 		    else {
3017 			new_deleg->ls_stateid.seqid = delegstateidp->seqid = 1;
3018 			new_deleg->ls_stateid.other[0] = delegstateidp->other[0]
3019 			    = clp->lc_clientid.lval[0];
3020 			new_deleg->ls_stateid.other[1] = delegstateidp->other[1]
3021 			    = clp->lc_clientid.lval[1];
3022 			new_deleg->ls_stateid.other[2] = delegstateidp->other[2]
3023 			    = nfsrv_nextstateindex(clp);
3024 			new_deleg->ls_flags = (NFSLCK_DELEGWRITE |
3025 			    NFSLCK_READACCESS | NFSLCK_WRITEACCESS);
3026 			*rflagsp |= NFSV4OPEN_WRITEDELEGATE;
3027 			new_deleg->ls_uid = new_stp->ls_uid;
3028 			new_deleg->ls_lfp = lfp;
3029 			new_deleg->ls_clp = clp;
3030 			new_deleg->ls_filerev = filerev;
3031 			new_deleg->ls_compref = nd->nd_compref;
3032 			LIST_INSERT_HEAD(&lfp->lf_deleg, new_deleg, ls_file);
3033 			LIST_INSERT_HEAD(NFSSTATEHASH(clp,
3034 			    new_deleg->ls_stateid), new_deleg, ls_hash);
3035 			LIST_INSERT_HEAD(&clp->lc_deleg, new_deleg, ls_list);
3036 			new_deleg = NULL;
3037 			nfsstatsv1.srvdelegates++;
3038 			nfsrv_openpluslock++;
3039 			nfsrv_delegatecnt++;
3040 		    }
3041 		} else {
3042 		    new_open->ls_stateid.seqid = 1;
3043 		    new_open->ls_stateid.other[0] = clp->lc_clientid.lval[0];
3044 		    new_open->ls_stateid.other[1] = clp->lc_clientid.lval[1];
3045 		    new_open->ls_stateid.other[2] = nfsrv_nextstateindex(clp);
3046 		    new_open->ls_flags = (new_stp->ls_flags & NFSLCK_SHAREBITS)|
3047 			NFSLCK_OPEN;
3048 		    new_open->ls_uid = new_stp->ls_uid;
3049 		    new_open->ls_openowner = ownerstp;
3050 		    new_open->ls_lfp = lfp;
3051 		    new_open->ls_clp = clp;
3052 		    LIST_INIT(&new_open->ls_open);
3053 		    LIST_INSERT_HEAD(&lfp->lf_open, new_open, ls_file);
3054 		    LIST_INSERT_HEAD(&ownerstp->ls_open, new_open, ls_list);
3055 		    LIST_INSERT_HEAD(NFSSTATEHASH(clp, new_open->ls_stateid),
3056 			new_open, ls_hash);
3057 		    openstp = new_open;
3058 		    new_open = NULL;
3059 		    nfsstatsv1.srvopens++;
3060 		    nfsrv_openpluslock++;
3061 
3062 		    /*
3063 		     * This is where we can choose to issue a delegation.
3064 		     */
3065 		    if (delegate == 0 || (writedeleg == 0 && readonly == 0) ||
3066 			!NFSVNO_DELEGOK(vp) ||
3067 			(clp->lc_flags & (LCL_CALLBACKSON | LCL_CBDOWN)) !=
3068 			 LCL_CALLBACKSON)
3069 			*rflagsp |= NFSV4OPEN_WDCONTENTION;
3070 		    else if (nfsrv_issuedelegs == 0 ||
3071 			NFSRV_V4DELEGLIMIT(nfsrv_delegatecnt))
3072 			*rflagsp |= NFSV4OPEN_WDRESOURCE;
3073 		    else if ((new_stp->ls_flags & NFSLCK_WANTNODELEG) != 0)
3074 			*rflagsp |= NFSV4OPEN_WDNOTWANTED;
3075 		    else {
3076 			new_deleg->ls_stateid.seqid = delegstateidp->seqid = 1;
3077 			new_deleg->ls_stateid.other[0] = delegstateidp->other[0]
3078 			    = clp->lc_clientid.lval[0];
3079 			new_deleg->ls_stateid.other[1] = delegstateidp->other[1]
3080 			    = clp->lc_clientid.lval[1];
3081 			new_deleg->ls_stateid.other[2] = delegstateidp->other[2]
3082 			    = nfsrv_nextstateindex(clp);
3083 			if (writedeleg && !NFSVNO_EXRDONLY(exp) &&
3084 			    (nfsrv_writedelegifpos || !readonly) &&
3085 			    (new_stp->ls_flags & NFSLCK_WANTRDELEG) == 0) {
3086 			    new_deleg->ls_flags = (NFSLCK_DELEGWRITE |
3087 				NFSLCK_READACCESS | NFSLCK_WRITEACCESS);
3088 			    *rflagsp |= NFSV4OPEN_WRITEDELEGATE;
3089 			} else {
3090 			    new_deleg->ls_flags = (NFSLCK_DELEGREAD |
3091 				NFSLCK_READACCESS);
3092 			    *rflagsp |= NFSV4OPEN_READDELEGATE;
3093 			}
3094 			new_deleg->ls_uid = new_stp->ls_uid;
3095 			new_deleg->ls_lfp = lfp;
3096 			new_deleg->ls_clp = clp;
3097 			new_deleg->ls_filerev = filerev;
3098 			new_deleg->ls_compref = nd->nd_compref;
3099 			LIST_INSERT_HEAD(&lfp->lf_deleg, new_deleg, ls_file);
3100 			LIST_INSERT_HEAD(NFSSTATEHASH(clp,
3101 			    new_deleg->ls_stateid), new_deleg, ls_hash);
3102 			LIST_INSERT_HEAD(&clp->lc_deleg, new_deleg, ls_list);
3103 			new_deleg = NULL;
3104 			nfsstatsv1.srvdelegates++;
3105 			nfsrv_openpluslock++;
3106 			nfsrv_delegatecnt++;
3107 		    }
3108 		}
3109 	} else {
3110 		/*
3111 		 * New owner case. Start the open_owner sequence with a
3112 		 * Needs confirmation (unless a reclaim) and hang the
3113 		 * new open off it.
3114 		 */
3115 		new_open->ls_stateid.seqid = 1;
3116 		new_open->ls_stateid.other[0] = clp->lc_clientid.lval[0];
3117 		new_open->ls_stateid.other[1] = clp->lc_clientid.lval[1];
3118 		new_open->ls_stateid.other[2] = nfsrv_nextstateindex(clp);
3119 		new_open->ls_flags = (new_stp->ls_flags & NFSLCK_SHAREBITS) |
3120 		    NFSLCK_OPEN;
3121 		new_open->ls_uid = new_stp->ls_uid;
3122 		LIST_INIT(&new_open->ls_open);
3123 		new_open->ls_openowner = new_stp;
3124 		new_open->ls_lfp = lfp;
3125 		new_open->ls_clp = clp;
3126 		LIST_INSERT_HEAD(&lfp->lf_open, new_open, ls_file);
3127 		if (new_stp->ls_flags & NFSLCK_RECLAIM) {
3128 			new_stp->ls_flags = 0;
3129 		} else if ((nd->nd_flag & ND_NFSV41) != 0) {
3130 			/* NFSv4.1 never needs confirmation. */
3131 			new_stp->ls_flags = 0;
3132 
3133 			/*
3134 			 * This is where we can choose to issue a delegation.
3135 			 */
3136 			if (delegate && nfsrv_issuedelegs &&
3137 			    (writedeleg || readonly) &&
3138 			    (clp->lc_flags & (LCL_CALLBACKSON | LCL_CBDOWN)) ==
3139 			     LCL_CALLBACKSON &&
3140 			    !NFSRV_V4DELEGLIMIT(nfsrv_delegatecnt) &&
3141 			    NFSVNO_DELEGOK(vp) &&
3142 			    ((nd->nd_flag & ND_NFSV41) == 0 ||
3143 			     (new_stp->ls_flags & NFSLCK_WANTNODELEG) == 0)) {
3144 				new_deleg->ls_stateid.seqid =
3145 				    delegstateidp->seqid = 1;
3146 				new_deleg->ls_stateid.other[0] =
3147 				    delegstateidp->other[0]
3148 				    = clp->lc_clientid.lval[0];
3149 				new_deleg->ls_stateid.other[1] =
3150 				    delegstateidp->other[1]
3151 				    = clp->lc_clientid.lval[1];
3152 				new_deleg->ls_stateid.other[2] =
3153 				    delegstateidp->other[2]
3154 				    = nfsrv_nextstateindex(clp);
3155 				if (writedeleg && !NFSVNO_EXRDONLY(exp) &&
3156 				    (nfsrv_writedelegifpos || !readonly) &&
3157 				    ((nd->nd_flag & ND_NFSV41) == 0 ||
3158 				     (new_stp->ls_flags & NFSLCK_WANTRDELEG) ==
3159 				     0)) {
3160 					new_deleg->ls_flags =
3161 					    (NFSLCK_DELEGWRITE |
3162 					     NFSLCK_READACCESS |
3163 					     NFSLCK_WRITEACCESS);
3164 					*rflagsp |= NFSV4OPEN_WRITEDELEGATE;
3165 				} else {
3166 					new_deleg->ls_flags =
3167 					    (NFSLCK_DELEGREAD |
3168 					     NFSLCK_READACCESS);
3169 					*rflagsp |= NFSV4OPEN_READDELEGATE;
3170 				}
3171 				new_deleg->ls_uid = new_stp->ls_uid;
3172 				new_deleg->ls_lfp = lfp;
3173 				new_deleg->ls_clp = clp;
3174 				new_deleg->ls_filerev = filerev;
3175 				new_deleg->ls_compref = nd->nd_compref;
3176 				LIST_INSERT_HEAD(&lfp->lf_deleg, new_deleg,
3177 				    ls_file);
3178 				LIST_INSERT_HEAD(NFSSTATEHASH(clp,
3179 				    new_deleg->ls_stateid), new_deleg, ls_hash);
3180 				LIST_INSERT_HEAD(&clp->lc_deleg, new_deleg,
3181 				    ls_list);
3182 				new_deleg = NULL;
3183 				nfsstatsv1.srvdelegates++;
3184 				nfsrv_openpluslock++;
3185 				nfsrv_delegatecnt++;
3186 			}
3187 			/*
3188 			 * Since NFSv4.1 never does an OpenConfirm, the first
3189 			 * open state will be acquired here.
3190 			 */
3191 			if (!(clp->lc_flags & LCL_STAMPEDSTABLE)) {
3192 				clp->lc_flags |= LCL_STAMPEDSTABLE;
3193 				len = clp->lc_idlen;
3194 				NFSBCOPY(clp->lc_id, clidp, len);
3195 				gotstate = 1;
3196 			}
3197 		} else {
3198 			*rflagsp |= NFSV4OPEN_RESULTCONFIRM;
3199 			new_stp->ls_flags = NFSLCK_NEEDSCONFIRM;
3200 		}
3201 		nfsrvd_refcache(new_stp->ls_op);
3202 		new_stp->ls_noopens = 0;
3203 		LIST_INIT(&new_stp->ls_open);
3204 		LIST_INSERT_HEAD(&new_stp->ls_open, new_open, ls_list);
3205 		LIST_INSERT_HEAD(&clp->lc_open, new_stp, ls_list);
3206 		LIST_INSERT_HEAD(NFSSTATEHASH(clp, new_open->ls_stateid),
3207 		    new_open, ls_hash);
3208 		openstp = new_open;
3209 		new_open = NULL;
3210 		*new_stpp = NULL;
3211 		nfsstatsv1.srvopens++;
3212 		nfsrv_openpluslock++;
3213 		nfsstatsv1.srvopenowners++;
3214 		nfsrv_openpluslock++;
3215 	}
3216 	if (!error) {
3217 		stateidp->seqid = openstp->ls_stateid.seqid;
3218 		stateidp->other[0] = openstp->ls_stateid.other[0];
3219 		stateidp->other[1] = openstp->ls_stateid.other[1];
3220 		stateidp->other[2] = openstp->ls_stateid.other[2];
3221 	}
3222 	NFSUNLOCKSTATE();
3223 	if (haslock) {
3224 		NFSLOCKV4ROOTMUTEX();
3225 		nfsv4_unlock(&nfsv4rootfs_lock, 1);
3226 		NFSUNLOCKV4ROOTMUTEX();
3227 	}
3228 	if (new_open)
3229 		FREE((caddr_t)new_open, M_NFSDSTATE);
3230 	if (new_deleg)
3231 		FREE((caddr_t)new_deleg, M_NFSDSTATE);
3232 
3233 	/*
3234 	 * If the NFSv4.1 client just acquired its first open, write a timestamp
3235 	 * to the stable storage file.
3236 	 */
3237 	if (gotstate != 0) {
3238 		nfsrv_writestable(clidp, len, NFSNST_NEWSTATE, p);
3239 		nfsrv_backupstable();
3240 	}
3241 
3242 out:
3243 	free(clidp, M_TEMP);
3244 	NFSEXITCODE2(error, nd);
3245 	return (error);
3246 }
3247 
3248 /*
3249  * Open update. Does the confirm, downgrade and close.
3250  */
3251 APPLESTATIC int
3252 nfsrv_openupdate(vnode_t vp, struct nfsstate *new_stp, nfsquad_t clientid,
3253     nfsv4stateid_t *stateidp, struct nfsrv_descript *nd, NFSPROC_T *p)
3254 {
3255 	struct nfsstate *stp, *ownerstp;
3256 	struct nfsclient *clp;
3257 	struct nfslockfile *lfp;
3258 	u_int32_t bits;
3259 	int error = 0, gotstate = 0, len = 0;
3260 	u_char *clidp = NULL;
3261 
3262 	/*
3263 	 * Check for restart conditions (client and server).
3264 	 */
3265 	error = nfsrv_checkrestart(clientid, new_stp->ls_flags,
3266 	    &new_stp->ls_stateid, 0);
3267 	if (error)
3268 		goto out;
3269 
3270 	clidp = malloc(NFSV4_OPAQUELIMIT, M_TEMP, M_WAITOK);
3271 	NFSLOCKSTATE();
3272 	/*
3273 	 * Get the open structure via clientid and stateid.
3274 	 */
3275 	error = nfsrv_getclient(clientid, CLOPS_RENEW, &clp, NULL,
3276 	    (nfsquad_t)((u_quad_t)0), 0, nd, p);
3277 	if (!error)
3278 		error = nfsrv_getstate(clp, &new_stp->ls_stateid,
3279 		    new_stp->ls_flags, &stp);
3280 
3281 	/*
3282 	 * Sanity check the open.
3283 	 */
3284 	if (!error && (!(stp->ls_flags & NFSLCK_OPEN) ||
3285 		(!(new_stp->ls_flags & NFSLCK_CONFIRM) &&
3286 		 (stp->ls_openowner->ls_flags & NFSLCK_NEEDSCONFIRM)) ||
3287 		((new_stp->ls_flags & NFSLCK_CONFIRM) &&
3288 		 (!(stp->ls_openowner->ls_flags & NFSLCK_NEEDSCONFIRM)))))
3289 		error = NFSERR_BADSTATEID;
3290 
3291 	if (!error)
3292 		error = nfsrv_checkseqid(nd, new_stp->ls_seq,
3293 		    stp->ls_openowner, new_stp->ls_op);
3294 	if (!error && stp->ls_stateid.seqid != new_stp->ls_stateid.seqid &&
3295 	    (((nd->nd_flag & ND_NFSV41) == 0 &&
3296 	      !(new_stp->ls_flags & NFSLCK_CONFIRM)) ||
3297 	     ((nd->nd_flag & ND_NFSV41) != 0 &&
3298 	      new_stp->ls_stateid.seqid != 0)))
3299 		error = NFSERR_OLDSTATEID;
3300 	if (!error && vnode_vtype(vp) != VREG) {
3301 		if (vnode_vtype(vp) == VDIR)
3302 			error = NFSERR_ISDIR;
3303 		else
3304 			error = NFSERR_INVAL;
3305 	}
3306 
3307 	if (error) {
3308 		/*
3309 		 * If a client tries to confirm an Open with a bad
3310 		 * seqid# and there are no byte range locks or other Opens
3311 		 * on the openowner, just throw it away, so the next use of the
3312 		 * openowner will start a fresh seq#.
3313 		 */
3314 		if (error == NFSERR_BADSEQID &&
3315 		    (new_stp->ls_flags & NFSLCK_CONFIRM) &&
3316 		    nfsrv_nootherstate(stp))
3317 			nfsrv_freeopenowner(stp->ls_openowner, 0, p);
3318 		NFSUNLOCKSTATE();
3319 		goto out;
3320 	}
3321 
3322 	/*
3323 	 * Set the return stateid.
3324 	 */
3325 	stateidp->seqid = stp->ls_stateid.seqid + 1;
3326 	if ((nd->nd_flag & ND_NFSV41) != 0 && stateidp->seqid == 0)
3327 		stateidp->seqid = 1;
3328 	stateidp->other[0] = stp->ls_stateid.other[0];
3329 	stateidp->other[1] = stp->ls_stateid.other[1];
3330 	stateidp->other[2] = stp->ls_stateid.other[2];
3331 	/*
3332 	 * Now, handle the three cases.
3333 	 */
3334 	if (new_stp->ls_flags & NFSLCK_CONFIRM) {
3335 		/*
3336 		 * If the open doesn't need confirmation, it seems to me that
3337 		 * there is a client error, but I'll just log it and keep going?
3338 		 */
3339 		if (!(stp->ls_openowner->ls_flags & NFSLCK_NEEDSCONFIRM))
3340 			printf("Nfsv4d: stray open confirm\n");
3341 		stp->ls_openowner->ls_flags = 0;
3342 		stp->ls_stateid.seqid++;
3343 		if ((nd->nd_flag & ND_NFSV41) != 0 &&
3344 		    stp->ls_stateid.seqid == 0)
3345 			stp->ls_stateid.seqid = 1;
3346 		if (!(clp->lc_flags & LCL_STAMPEDSTABLE)) {
3347 			clp->lc_flags |= LCL_STAMPEDSTABLE;
3348 			len = clp->lc_idlen;
3349 			NFSBCOPY(clp->lc_id, clidp, len);
3350 			gotstate = 1;
3351 		}
3352 		NFSUNLOCKSTATE();
3353 	} else if (new_stp->ls_flags & NFSLCK_CLOSE) {
3354 		ownerstp = stp->ls_openowner;
3355 		lfp = stp->ls_lfp;
3356 		if (nfsrv_dolocallocks != 0 && !LIST_EMPTY(&stp->ls_open)) {
3357 			/* Get the lf lock */
3358 			nfsrv_locklf(lfp);
3359 			NFSUNLOCKSTATE();
3360 			ASSERT_VOP_ELOCKED(vp, "nfsrv_openupdate");
3361 			NFSVOPUNLOCK(vp, 0);
3362 			if (nfsrv_freeopen(stp, vp, 1, p) == 0) {
3363 				NFSLOCKSTATE();
3364 				nfsrv_unlocklf(lfp);
3365 				NFSUNLOCKSTATE();
3366 			}
3367 			NFSVOPLOCK(vp, LK_EXCLUSIVE | LK_RETRY);
3368 		} else {
3369 			(void) nfsrv_freeopen(stp, NULL, 0, p);
3370 			NFSUNLOCKSTATE();
3371 		}
3372 	} else {
3373 		/*
3374 		 * Update the share bits, making sure that the new set are a
3375 		 * subset of the old ones.
3376 		 */
3377 		bits = (new_stp->ls_flags & NFSLCK_SHAREBITS);
3378 		if (~(stp->ls_flags) & bits) {
3379 			NFSUNLOCKSTATE();
3380 			error = NFSERR_INVAL;
3381 			goto out;
3382 		}
3383 		stp->ls_flags = (bits | NFSLCK_OPEN);
3384 		stp->ls_stateid.seqid++;
3385 		if ((nd->nd_flag & ND_NFSV41) != 0 &&
3386 		    stp->ls_stateid.seqid == 0)
3387 			stp->ls_stateid.seqid = 1;
3388 		NFSUNLOCKSTATE();
3389 	}
3390 
3391 	/*
3392 	 * If the client just confirmed its first open, write a timestamp
3393 	 * to the stable storage file.
3394 	 */
3395 	if (gotstate != 0) {
3396 		nfsrv_writestable(clidp, len, NFSNST_NEWSTATE, p);
3397 		nfsrv_backupstable();
3398 	}
3399 
3400 out:
3401 	free(clidp, M_TEMP);
3402 	NFSEXITCODE2(error, nd);
3403 	return (error);
3404 }
3405 
3406 /*
3407  * Delegation update. Does the purge and return.
3408  */
3409 APPLESTATIC int
3410 nfsrv_delegupdate(struct nfsrv_descript *nd, nfsquad_t clientid,
3411     nfsv4stateid_t *stateidp, vnode_t vp, int op, struct ucred *cred,
3412     NFSPROC_T *p)
3413 {
3414 	struct nfsstate *stp;
3415 	struct nfsclient *clp;
3416 	int error = 0;
3417 	fhandle_t fh;
3418 
3419 	/*
3420 	 * Do a sanity check against the file handle for DelegReturn.
3421 	 */
3422 	if (vp) {
3423 		error = nfsvno_getfh(vp, &fh, p);
3424 		if (error)
3425 			goto out;
3426 	}
3427 	/*
3428 	 * Check for restart conditions (client and server).
3429 	 */
3430 	if (op == NFSV4OP_DELEGRETURN)
3431 		error = nfsrv_checkrestart(clientid, NFSLCK_DELEGRETURN,
3432 			stateidp, 0);
3433 	else
3434 		error = nfsrv_checkrestart(clientid, NFSLCK_DELEGPURGE,
3435 			stateidp, 0);
3436 
3437 	NFSLOCKSTATE();
3438 	/*
3439 	 * Get the open structure via clientid and stateid.
3440 	 */
3441 	if (!error)
3442 	    error = nfsrv_getclient(clientid, CLOPS_RENEW, &clp, NULL,
3443 		(nfsquad_t)((u_quad_t)0), 0, nd, p);
3444 	if (error) {
3445 		if (error == NFSERR_CBPATHDOWN)
3446 			error = 0;
3447 		if (error == NFSERR_STALECLIENTID && op == NFSV4OP_DELEGRETURN)
3448 			error = NFSERR_STALESTATEID;
3449 	}
3450 	if (!error && op == NFSV4OP_DELEGRETURN) {
3451 	    error = nfsrv_getstate(clp, stateidp, NFSLCK_DELEGRETURN, &stp);
3452 	    if (!error && stp->ls_stateid.seqid != stateidp->seqid &&
3453 		((nd->nd_flag & ND_NFSV41) == 0 || stateidp->seqid != 0))
3454 		error = NFSERR_OLDSTATEID;
3455 	}
3456 	/*
3457 	 * NFSERR_EXPIRED means that the state has gone away,
3458 	 * so Delegations have been purged. Just return ok.
3459 	 */
3460 	if (error == NFSERR_EXPIRED && op == NFSV4OP_DELEGPURGE) {
3461 		NFSUNLOCKSTATE();
3462 		error = 0;
3463 		goto out;
3464 	}
3465 	if (error) {
3466 		NFSUNLOCKSTATE();
3467 		goto out;
3468 	}
3469 
3470 	if (op == NFSV4OP_DELEGRETURN) {
3471 		if (NFSBCMP((caddr_t)&fh, (caddr_t)&stp->ls_lfp->lf_fh,
3472 		    sizeof (fhandle_t))) {
3473 			NFSUNLOCKSTATE();
3474 			error = NFSERR_BADSTATEID;
3475 			goto out;
3476 		}
3477 		nfsrv_freedeleg(stp);
3478 	} else {
3479 		nfsrv_freedeleglist(&clp->lc_olddeleg);
3480 	}
3481 	NFSUNLOCKSTATE();
3482 	error = 0;
3483 
3484 out:
3485 	NFSEXITCODE(error);
3486 	return (error);
3487 }
3488 
3489 /*
3490  * Release lock owner.
3491  */
3492 APPLESTATIC int
3493 nfsrv_releaselckown(struct nfsstate *new_stp, nfsquad_t clientid,
3494     NFSPROC_T *p)
3495 {
3496 	struct nfsstate *stp, *nstp, *openstp, *ownstp;
3497 	struct nfsclient *clp;
3498 	int error = 0;
3499 
3500 	/*
3501 	 * Check for restart conditions (client and server).
3502 	 */
3503 	error = nfsrv_checkrestart(clientid, new_stp->ls_flags,
3504 	    &new_stp->ls_stateid, 0);
3505 	if (error)
3506 		goto out;
3507 
3508 	NFSLOCKSTATE();
3509 	/*
3510 	 * Get the lock owner by name.
3511 	 */
3512 	error = nfsrv_getclient(clientid, CLOPS_RENEW, &clp, NULL,
3513 	    (nfsquad_t)((u_quad_t)0), 0, NULL, p);
3514 	if (error) {
3515 		NFSUNLOCKSTATE();
3516 		goto out;
3517 	}
3518 	LIST_FOREACH(ownstp, &clp->lc_open, ls_list) {
3519 	    LIST_FOREACH(openstp, &ownstp->ls_open, ls_list) {
3520 		stp = LIST_FIRST(&openstp->ls_open);
3521 		while (stp != LIST_END(&openstp->ls_open)) {
3522 		    nstp = LIST_NEXT(stp, ls_list);
3523 		    /*
3524 		     * If the owner matches, check for locks and
3525 		     * then free or return an error.
3526 		     */
3527 		    if (stp->ls_ownerlen == new_stp->ls_ownerlen &&
3528 			!NFSBCMP(stp->ls_owner, new_stp->ls_owner,
3529 			 stp->ls_ownerlen)){
3530 			if (LIST_EMPTY(&stp->ls_lock)) {
3531 			    nfsrv_freelockowner(stp, NULL, 0, p);
3532 			} else {
3533 			    NFSUNLOCKSTATE();
3534 			    error = NFSERR_LOCKSHELD;
3535 			    goto out;
3536 			}
3537 		    }
3538 		    stp = nstp;
3539 		}
3540 	    }
3541 	}
3542 	NFSUNLOCKSTATE();
3543 
3544 out:
3545 	NFSEXITCODE(error);
3546 	return (error);
3547 }
3548 
3549 /*
3550  * Get the file handle for a lock structure.
3551  */
3552 static int
3553 nfsrv_getlockfh(vnode_t vp, u_short flags, struct nfslockfile *new_lfp,
3554     fhandle_t *nfhp, NFSPROC_T *p)
3555 {
3556 	fhandle_t *fhp = NULL;
3557 	int error;
3558 
3559 	/*
3560 	 * For lock, use the new nfslock structure, otherwise just
3561 	 * a fhandle_t on the stack.
3562 	 */
3563 	if (flags & NFSLCK_OPEN) {
3564 		KASSERT(new_lfp != NULL, ("nfsrv_getlockfh: new_lfp NULL"));
3565 		fhp = &new_lfp->lf_fh;
3566 	} else if (nfhp) {
3567 		fhp = nfhp;
3568 	} else {
3569 		panic("nfsrv_getlockfh");
3570 	}
3571 	error = nfsvno_getfh(vp, fhp, p);
3572 	NFSEXITCODE(error);
3573 	return (error);
3574 }
3575 
3576 /*
3577  * Get an nfs lock structure. Allocate one, as required, and return a
3578  * pointer to it.
3579  * Returns an NFSERR_xxx upon failure or -1 to indicate no current lock.
3580  */
3581 static int
3582 nfsrv_getlockfile(u_short flags, struct nfslockfile **new_lfpp,
3583     struct nfslockfile **lfpp, fhandle_t *nfhp, int lockit)
3584 {
3585 	struct nfslockfile *lfp;
3586 	fhandle_t *fhp = NULL, *tfhp;
3587 	struct nfslockhashhead *hp;
3588 	struct nfslockfile *new_lfp = NULL;
3589 
3590 	/*
3591 	 * For lock, use the new nfslock structure, otherwise just
3592 	 * a fhandle_t on the stack.
3593 	 */
3594 	if (flags & NFSLCK_OPEN) {
3595 		new_lfp = *new_lfpp;
3596 		fhp = &new_lfp->lf_fh;
3597 	} else if (nfhp) {
3598 		fhp = nfhp;
3599 	} else {
3600 		panic("nfsrv_getlockfile");
3601 	}
3602 
3603 	hp = NFSLOCKHASH(fhp);
3604 	LIST_FOREACH(lfp, hp, lf_hash) {
3605 		tfhp = &lfp->lf_fh;
3606 		if (NFSVNO_CMPFH(fhp, tfhp)) {
3607 			if (lockit)
3608 				nfsrv_locklf(lfp);
3609 			*lfpp = lfp;
3610 			return (0);
3611 		}
3612 	}
3613 	if (!(flags & NFSLCK_OPEN))
3614 		return (-1);
3615 
3616 	/*
3617 	 * No match, so chain the new one into the list.
3618 	 */
3619 	LIST_INIT(&new_lfp->lf_open);
3620 	LIST_INIT(&new_lfp->lf_lock);
3621 	LIST_INIT(&new_lfp->lf_deleg);
3622 	LIST_INIT(&new_lfp->lf_locallock);
3623 	LIST_INIT(&new_lfp->lf_rollback);
3624 	new_lfp->lf_locallock_lck.nfslock_usecnt = 0;
3625 	new_lfp->lf_locallock_lck.nfslock_lock = 0;
3626 	new_lfp->lf_usecount = 0;
3627 	LIST_INSERT_HEAD(hp, new_lfp, lf_hash);
3628 	*lfpp = new_lfp;
3629 	*new_lfpp = NULL;
3630 	return (0);
3631 }
3632 
3633 /*
3634  * This function adds a nfslock lock structure to the list for the associated
3635  * nfsstate and nfslockfile structures. It will be inserted after the
3636  * entry pointed at by insert_lop.
3637  */
3638 static void
3639 nfsrv_insertlock(struct nfslock *new_lop, struct nfslock *insert_lop,
3640     struct nfsstate *stp, struct nfslockfile *lfp)
3641 {
3642 	struct nfslock *lop, *nlop;
3643 
3644 	new_lop->lo_stp = stp;
3645 	new_lop->lo_lfp = lfp;
3646 
3647 	if (stp != NULL) {
3648 		/* Insert in increasing lo_first order */
3649 		lop = LIST_FIRST(&lfp->lf_lock);
3650 		if (lop == LIST_END(&lfp->lf_lock) ||
3651 		    new_lop->lo_first <= lop->lo_first) {
3652 			LIST_INSERT_HEAD(&lfp->lf_lock, new_lop, lo_lckfile);
3653 		} else {
3654 			nlop = LIST_NEXT(lop, lo_lckfile);
3655 			while (nlop != LIST_END(&lfp->lf_lock) &&
3656 			       nlop->lo_first < new_lop->lo_first) {
3657 				lop = nlop;
3658 				nlop = LIST_NEXT(lop, lo_lckfile);
3659 			}
3660 			LIST_INSERT_AFTER(lop, new_lop, lo_lckfile);
3661 		}
3662 	} else {
3663 		new_lop->lo_lckfile.le_prev = NULL;	/* list not used */
3664 	}
3665 
3666 	/*
3667 	 * Insert after insert_lop, which is overloaded as stp or lfp for
3668 	 * an empty list.
3669 	 */
3670 	if (stp == NULL && (struct nfslockfile *)insert_lop == lfp)
3671 		LIST_INSERT_HEAD(&lfp->lf_locallock, new_lop, lo_lckowner);
3672 	else if ((struct nfsstate *)insert_lop == stp)
3673 		LIST_INSERT_HEAD(&stp->ls_lock, new_lop, lo_lckowner);
3674 	else
3675 		LIST_INSERT_AFTER(insert_lop, new_lop, lo_lckowner);
3676 	if (stp != NULL) {
3677 		nfsstatsv1.srvlocks++;
3678 		nfsrv_openpluslock++;
3679 	}
3680 }
3681 
3682 /*
3683  * This function updates the locking for a lock owner and given file. It
3684  * maintains a list of lock ranges ordered on increasing file offset that
3685  * are NFSLCK_READ or NFSLCK_WRITE and non-overlapping (aka POSIX style).
3686  * It always adds new_lop to the list and sometimes uses the one pointed
3687  * at by other_lopp.
3688  */
3689 static void
3690 nfsrv_updatelock(struct nfsstate *stp, struct nfslock **new_lopp,
3691     struct nfslock **other_lopp, struct nfslockfile *lfp)
3692 {
3693 	struct nfslock *new_lop = *new_lopp;
3694 	struct nfslock *lop, *tlop, *ilop;
3695 	struct nfslock *other_lop = *other_lopp;
3696 	int unlock = 0, myfile = 0;
3697 	u_int64_t tmp;
3698 
3699 	/*
3700 	 * Work down the list until the lock is merged.
3701 	 */
3702 	if (new_lop->lo_flags & NFSLCK_UNLOCK)
3703 		unlock = 1;
3704 	if (stp != NULL) {
3705 		ilop = (struct nfslock *)stp;
3706 		lop = LIST_FIRST(&stp->ls_lock);
3707 	} else {
3708 		ilop = (struct nfslock *)lfp;
3709 		lop = LIST_FIRST(&lfp->lf_locallock);
3710 	}
3711 	while (lop != NULL) {
3712 	    /*
3713 	     * Only check locks for this file that aren't before the start of
3714 	     * new lock's range.
3715 	     */
3716 	    if (lop->lo_lfp == lfp) {
3717 	      myfile = 1;
3718 	      if (lop->lo_end >= new_lop->lo_first) {
3719 		if (new_lop->lo_end < lop->lo_first) {
3720 			/*
3721 			 * If the new lock ends before the start of the
3722 			 * current lock's range, no merge, just insert
3723 			 * the new lock.
3724 			 */
3725 			break;
3726 		}
3727 		if (new_lop->lo_flags == lop->lo_flags ||
3728 		    (new_lop->lo_first <= lop->lo_first &&
3729 		     new_lop->lo_end >= lop->lo_end)) {
3730 			/*
3731 			 * This lock can be absorbed by the new lock/unlock.
3732 			 * This happens when it covers the entire range
3733 			 * of the old lock or is contiguous
3734 			 * with the old lock and is of the same type or an
3735 			 * unlock.
3736 			 */
3737 			if (lop->lo_first < new_lop->lo_first)
3738 				new_lop->lo_first = lop->lo_first;
3739 			if (lop->lo_end > new_lop->lo_end)
3740 				new_lop->lo_end = lop->lo_end;
3741 			tlop = lop;
3742 			lop = LIST_NEXT(lop, lo_lckowner);
3743 			nfsrv_freenfslock(tlop);
3744 			continue;
3745 		}
3746 
3747 		/*
3748 		 * All these cases are for contiguous locks that are not the
3749 		 * same type, so they can't be merged.
3750 		 */
3751 		if (new_lop->lo_first <= lop->lo_first) {
3752 			/*
3753 			 * This case is where the new lock overlaps with the
3754 			 * first part of the old lock. Move the start of the
3755 			 * old lock to just past the end of the new lock. The
3756 			 * new lock will be inserted in front of the old, since
3757 			 * ilop hasn't been updated. (We are done now.)
3758 			 */
3759 			lop->lo_first = new_lop->lo_end;
3760 			break;
3761 		}
3762 		if (new_lop->lo_end >= lop->lo_end) {
3763 			/*
3764 			 * This case is where the new lock overlaps with the
3765 			 * end of the old lock's range. Move the old lock's
3766 			 * end to just before the new lock's first and insert
3767 			 * the new lock after the old lock.
3768 			 * Might not be done yet, since the new lock could
3769 			 * overlap further locks with higher ranges.
3770 			 */
3771 			lop->lo_end = new_lop->lo_first;
3772 			ilop = lop;
3773 			lop = LIST_NEXT(lop, lo_lckowner);
3774 			continue;
3775 		}
3776 		/*
3777 		 * The final case is where the new lock's range is in the
3778 		 * middle of the current lock's and splits the current lock
3779 		 * up. Use *other_lopp to handle the second part of the
3780 		 * split old lock range. (We are done now.)
3781 		 * For unlock, we use new_lop as other_lop and tmp, since
3782 		 * other_lop and new_lop are the same for this case.
3783 		 * We noted the unlock case above, so we don't need
3784 		 * new_lop->lo_flags any longer.
3785 		 */
3786 		tmp = new_lop->lo_first;
3787 		if (other_lop == NULL) {
3788 			if (!unlock)
3789 				panic("nfsd srv update unlock");
3790 			other_lop = new_lop;
3791 			*new_lopp = NULL;
3792 		}
3793 		other_lop->lo_first = new_lop->lo_end;
3794 		other_lop->lo_end = lop->lo_end;
3795 		other_lop->lo_flags = lop->lo_flags;
3796 		other_lop->lo_stp = stp;
3797 		other_lop->lo_lfp = lfp;
3798 		lop->lo_end = tmp;
3799 		nfsrv_insertlock(other_lop, lop, stp, lfp);
3800 		*other_lopp = NULL;
3801 		ilop = lop;
3802 		break;
3803 	      }
3804 	    }
3805 	    ilop = lop;
3806 	    lop = LIST_NEXT(lop, lo_lckowner);
3807 	    if (myfile && (lop == NULL || lop->lo_lfp != lfp))
3808 		break;
3809 	}
3810 
3811 	/*
3812 	 * Insert the new lock in the list at the appropriate place.
3813 	 */
3814 	if (!unlock) {
3815 		nfsrv_insertlock(new_lop, ilop, stp, lfp);
3816 		*new_lopp = NULL;
3817 	}
3818 }
3819 
3820 /*
3821  * This function handles sequencing of locks, etc.
3822  * It returns an error that indicates what the caller should do.
3823  */
3824 static int
3825 nfsrv_checkseqid(struct nfsrv_descript *nd, u_int32_t seqid,
3826     struct nfsstate *stp, struct nfsrvcache *op)
3827 {
3828 	int error = 0;
3829 
3830 	if ((nd->nd_flag & ND_NFSV41) != 0)
3831 		/* NFSv4.1 ignores the open_seqid and lock_seqid. */
3832 		goto out;
3833 	if (op != nd->nd_rp)
3834 		panic("nfsrvstate checkseqid");
3835 	if (!(op->rc_flag & RC_INPROG))
3836 		panic("nfsrvstate not inprog");
3837 	if (stp->ls_op && stp->ls_op->rc_refcnt <= 0) {
3838 		printf("refcnt=%d\n", stp->ls_op->rc_refcnt);
3839 		panic("nfsrvstate op refcnt");
3840 	}
3841 	if ((stp->ls_seq + 1) == seqid) {
3842 		if (stp->ls_op)
3843 			nfsrvd_derefcache(stp->ls_op);
3844 		stp->ls_op = op;
3845 		nfsrvd_refcache(op);
3846 		stp->ls_seq = seqid;
3847 		goto out;
3848 	} else if (stp->ls_seq == seqid && stp->ls_op &&
3849 		op->rc_xid == stp->ls_op->rc_xid &&
3850 		op->rc_refcnt == 0 &&
3851 		op->rc_reqlen == stp->ls_op->rc_reqlen &&
3852 		op->rc_cksum == stp->ls_op->rc_cksum) {
3853 		if (stp->ls_op->rc_flag & RC_INPROG) {
3854 			error = NFSERR_DONTREPLY;
3855 			goto out;
3856 		}
3857 		nd->nd_rp = stp->ls_op;
3858 		nd->nd_rp->rc_flag |= RC_INPROG;
3859 		nfsrvd_delcache(op);
3860 		error = NFSERR_REPLYFROMCACHE;
3861 		goto out;
3862 	}
3863 	error = NFSERR_BADSEQID;
3864 
3865 out:
3866 	NFSEXITCODE2(error, nd);
3867 	return (error);
3868 }
3869 
3870 /*
3871  * Get the client ip address for callbacks. If the strings can't be parsed,
3872  * just set lc_program to 0 to indicate no callbacks are possible.
3873  * (For cases where the address can't be parsed or is 0.0.0.0.0.0, set
3874  *  the address to the client's transport address. This won't be used
3875  *  for callbacks, but can be printed out by nfsstats for info.)
3876  * Return error if the xdr can't be parsed, 0 otherwise.
3877  */
3878 APPLESTATIC int
3879 nfsrv_getclientipaddr(struct nfsrv_descript *nd, struct nfsclient *clp)
3880 {
3881 	u_int32_t *tl;
3882 	u_char *cp, *cp2;
3883 	int i, j;
3884 	struct sockaddr_in *rad, *sad;
3885 	u_char protocol[5], addr[24];
3886 	int error = 0, cantparse = 0;
3887 	union {
3888 		u_long ival;
3889 		u_char cval[4];
3890 	} ip;
3891 	union {
3892 		u_short sval;
3893 		u_char cval[2];
3894 	} port;
3895 
3896 	rad = NFSSOCKADDR(clp->lc_req.nr_nam, struct sockaddr_in *);
3897 	rad->sin_family = AF_INET;
3898 	rad->sin_len = sizeof (struct sockaddr_in);
3899 	rad->sin_addr.s_addr = 0;
3900 	rad->sin_port = 0;
3901 	clp->lc_req.nr_client = NULL;
3902 	clp->lc_req.nr_lock = 0;
3903 	NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
3904 	i = fxdr_unsigned(int, *tl);
3905 	if (i >= 3 && i <= 4) {
3906 		error = nfsrv_mtostr(nd, protocol, i);
3907 		if (error)
3908 			goto nfsmout;
3909 		if (!strcmp(protocol, "tcp")) {
3910 			clp->lc_flags |= LCL_TCPCALLBACK;
3911 			clp->lc_req.nr_sotype = SOCK_STREAM;
3912 			clp->lc_req.nr_soproto = IPPROTO_TCP;
3913 		} else if (!strcmp(protocol, "udp")) {
3914 			clp->lc_req.nr_sotype = SOCK_DGRAM;
3915 			clp->lc_req.nr_soproto = IPPROTO_UDP;
3916 		} else {
3917 			cantparse = 1;
3918 		}
3919 	} else {
3920 		cantparse = 1;
3921 		if (i > 0) {
3922 			error = nfsm_advance(nd, NFSM_RNDUP(i), -1);
3923 			if (error)
3924 				goto nfsmout;
3925 		}
3926 	}
3927 	NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
3928 	i = fxdr_unsigned(int, *tl);
3929 	if (i < 0) {
3930 		error = NFSERR_BADXDR;
3931 		goto nfsmout;
3932 	} else if (i == 0) {
3933 		cantparse = 1;
3934 	} else if (!cantparse && i <= 23 && i >= 11) {
3935 		error = nfsrv_mtostr(nd, addr, i);
3936 		if (error)
3937 			goto nfsmout;
3938 
3939 		/*
3940 		 * Parse out the address fields. We expect 6 decimal numbers
3941 		 * separated by '.'s.
3942 		 */
3943 		cp = addr;
3944 		i = 0;
3945 		while (*cp && i < 6) {
3946 			cp2 = cp;
3947 			while (*cp2 && *cp2 != '.')
3948 				cp2++;
3949 			if (*cp2)
3950 				*cp2++ = '\0';
3951 			else if (i != 5) {
3952 				cantparse = 1;
3953 				break;
3954 			}
3955 			j = nfsrv_getipnumber(cp);
3956 			if (j >= 0) {
3957 				if (i < 4)
3958 					ip.cval[3 - i] = j;
3959 				else
3960 					port.cval[5 - i] = j;
3961 			} else {
3962 				cantparse = 1;
3963 				break;
3964 			}
3965 			cp = cp2;
3966 			i++;
3967 		}
3968 		if (!cantparse) {
3969 			if (ip.ival != 0x0) {
3970 				rad->sin_addr.s_addr = htonl(ip.ival);
3971 				rad->sin_port = htons(port.sval);
3972 			} else {
3973 				cantparse = 1;
3974 			}
3975 		}
3976 	} else {
3977 		cantparse = 1;
3978 		if (i > 0) {
3979 			error = nfsm_advance(nd, NFSM_RNDUP(i), -1);
3980 			if (error)
3981 				goto nfsmout;
3982 		}
3983 	}
3984 	if (cantparse) {
3985 		sad = NFSSOCKADDR(nd->nd_nam, struct sockaddr_in *);
3986 		rad->sin_addr.s_addr = sad->sin_addr.s_addr;
3987 		rad->sin_port = 0x0;
3988 		clp->lc_program = 0;
3989 	}
3990 nfsmout:
3991 	NFSEXITCODE2(error, nd);
3992 	return (error);
3993 }
3994 
3995 /*
3996  * Turn a string of up to three decimal digits into a number. Return -1 upon
3997  * error.
3998  */
3999 static int
4000 nfsrv_getipnumber(u_char *cp)
4001 {
4002 	int i = 0, j = 0;
4003 
4004 	while (*cp) {
4005 		if (j > 2 || *cp < '0' || *cp > '9')
4006 			return (-1);
4007 		i *= 10;
4008 		i += (*cp - '0');
4009 		cp++;
4010 		j++;
4011 	}
4012 	if (i < 256)
4013 		return (i);
4014 	return (-1);
4015 }
4016 
4017 /*
4018  * This function checks for restart conditions.
4019  */
4020 static int
4021 nfsrv_checkrestart(nfsquad_t clientid, u_int32_t flags,
4022     nfsv4stateid_t *stateidp, int specialid)
4023 {
4024 	int ret = 0;
4025 
4026 	/*
4027 	 * First check for a server restart. Open, LockT, ReleaseLockOwner
4028 	 * and DelegPurge have a clientid, the rest a stateid.
4029 	 */
4030 	if (flags &
4031 	    (NFSLCK_OPEN | NFSLCK_TEST | NFSLCK_RELEASE | NFSLCK_DELEGPURGE)) {
4032 		if (clientid.lval[0] != nfsrvboottime) {
4033 			ret = NFSERR_STALECLIENTID;
4034 			goto out;
4035 		}
4036 	} else if (stateidp->other[0] != nfsrvboottime &&
4037 		specialid == 0) {
4038 		ret = NFSERR_STALESTATEID;
4039 		goto out;
4040 	}
4041 
4042 	/*
4043 	 * Read, Write, Setattr and LockT can return NFSERR_GRACE and do
4044 	 * not use a lock/open owner seqid#, so the check can be done now.
4045 	 * (The others will be checked, as required, later.)
4046 	 */
4047 	if (!(flags & (NFSLCK_CHECK | NFSLCK_TEST)))
4048 		goto out;
4049 
4050 	NFSLOCKSTATE();
4051 	ret = nfsrv_checkgrace(NULL, NULL, flags);
4052 	NFSUNLOCKSTATE();
4053 
4054 out:
4055 	NFSEXITCODE(ret);
4056 	return (ret);
4057 }
4058 
4059 /*
4060  * Check for grace.
4061  */
4062 static int
4063 nfsrv_checkgrace(struct nfsrv_descript *nd, struct nfsclient *clp,
4064     u_int32_t flags)
4065 {
4066 	int error = 0;
4067 
4068 	if ((nfsrv_stablefirst.nsf_flags & NFSNSF_GRACEOVER) != 0) {
4069 		if (flags & NFSLCK_RECLAIM) {
4070 			error = NFSERR_NOGRACE;
4071 			goto out;
4072 		}
4073 	} else {
4074 		if (!(flags & NFSLCK_RECLAIM)) {
4075 			error = NFSERR_GRACE;
4076 			goto out;
4077 		}
4078 		if (nd != NULL && clp != NULL &&
4079 		    (nd->nd_flag & ND_NFSV41) != 0 &&
4080 		    (clp->lc_flags & LCL_RECLAIMCOMPLETE) != 0) {
4081 			error = NFSERR_NOGRACE;
4082 			goto out;
4083 		}
4084 
4085 		/*
4086 		 * If grace is almost over and we are still getting Reclaims,
4087 		 * extend grace a bit.
4088 		 */
4089 		if ((NFSD_MONOSEC + NFSRV_LEASEDELTA) >
4090 		    nfsrv_stablefirst.nsf_eograce)
4091 			nfsrv_stablefirst.nsf_eograce = NFSD_MONOSEC +
4092 				NFSRV_LEASEDELTA;
4093 	}
4094 
4095 out:
4096 	NFSEXITCODE(error);
4097 	return (error);
4098 }
4099 
4100 /*
4101  * Do a server callback.
4102  */
4103 static int
4104 nfsrv_docallback(struct nfsclient *clp, int procnum,
4105     nfsv4stateid_t *stateidp, int trunc, fhandle_t *fhp,
4106     struct nfsvattr *nap, nfsattrbit_t *attrbitp, NFSPROC_T *p)
4107 {
4108 	mbuf_t m;
4109 	u_int32_t *tl;
4110 	struct nfsrv_descript nfsd, *nd = &nfsd;
4111 	struct ucred *cred;
4112 	int error = 0;
4113 	u_int32_t callback;
4114 	struct nfsdsession *sep = NULL;
4115 
4116 	cred = newnfs_getcred();
4117 	NFSLOCKSTATE();	/* mostly for lc_cbref++ */
4118 	if (clp->lc_flags & LCL_NEEDSCONFIRM) {
4119 		NFSUNLOCKSTATE();
4120 		panic("docallb");
4121 	}
4122 	clp->lc_cbref++;
4123 
4124 	/*
4125 	 * Fill the callback program# and version into the request
4126 	 * structure for newnfs_connect() to use.
4127 	 */
4128 	clp->lc_req.nr_prog = clp->lc_program;
4129 #ifdef notnow
4130 	if ((clp->lc_flags & LCL_NFSV41) != 0)
4131 		clp->lc_req.nr_vers = NFSV41_CBVERS;
4132 	else
4133 #endif
4134 		clp->lc_req.nr_vers = NFSV4_CBVERS;
4135 
4136 	/*
4137 	 * First, fill in some of the fields of nd and cr.
4138 	 */
4139 	nd->nd_flag = ND_NFSV4;
4140 	if (clp->lc_flags & LCL_GSS)
4141 		nd->nd_flag |= ND_KERBV;
4142 	if ((clp->lc_flags & LCL_NFSV41) != 0)
4143 		nd->nd_flag |= ND_NFSV41;
4144 	nd->nd_repstat = 0;
4145 	cred->cr_uid = clp->lc_uid;
4146 	cred->cr_gid = clp->lc_gid;
4147 	callback = clp->lc_callback;
4148 	NFSUNLOCKSTATE();
4149 	cred->cr_ngroups = 1;
4150 
4151 	/*
4152 	 * Get the first mbuf for the request.
4153 	 */
4154 	MGET(m, M_WAITOK, MT_DATA);
4155 	mbuf_setlen(m, 0);
4156 	nd->nd_mreq = nd->nd_mb = m;
4157 	nd->nd_bpos = NFSMTOD(m, caddr_t);
4158 
4159 	/*
4160 	 * and build the callback request.
4161 	 */
4162 	if (procnum == NFSV4OP_CBGETATTR) {
4163 		nd->nd_procnum = NFSV4PROC_CBCOMPOUND;
4164 		error = nfsrv_cbcallargs(nd, clp, callback, NFSV4OP_CBGETATTR,
4165 		    "CB Getattr", &sep);
4166 		if (error != 0) {
4167 			mbuf_freem(nd->nd_mreq);
4168 			goto errout;
4169 		}
4170 		(void)nfsm_fhtom(nd, (u_int8_t *)fhp, NFSX_MYFH, 0);
4171 		(void)nfsrv_putattrbit(nd, attrbitp);
4172 	} else if (procnum == NFSV4OP_CBRECALL) {
4173 		nd->nd_procnum = NFSV4PROC_CBCOMPOUND;
4174 		error = nfsrv_cbcallargs(nd, clp, callback, NFSV4OP_CBRECALL,
4175 		    "CB Recall", &sep);
4176 		if (error != 0) {
4177 			mbuf_freem(nd->nd_mreq);
4178 			goto errout;
4179 		}
4180 		NFSM_BUILD(tl, u_int32_t *, NFSX_UNSIGNED + NFSX_STATEID);
4181 		*tl++ = txdr_unsigned(stateidp->seqid);
4182 		NFSBCOPY((caddr_t)stateidp->other, (caddr_t)tl,
4183 		    NFSX_STATEIDOTHER);
4184 		tl += (NFSX_STATEIDOTHER / NFSX_UNSIGNED);
4185 		if (trunc)
4186 			*tl = newnfs_true;
4187 		else
4188 			*tl = newnfs_false;
4189 		(void)nfsm_fhtom(nd, (u_int8_t *)fhp, NFSX_MYFH, 0);
4190 	} else if (procnum == NFSV4PROC_CBNULL) {
4191 		nd->nd_procnum = NFSV4PROC_CBNULL;
4192 		if ((clp->lc_flags & LCL_NFSV41) != 0) {
4193 			error = nfsv4_getcbsession(clp, &sep);
4194 			if (error != 0) {
4195 				mbuf_freem(nd->nd_mreq);
4196 				goto errout;
4197 			}
4198 		}
4199 	} else {
4200 		error = NFSERR_SERVERFAULT;
4201 		mbuf_freem(nd->nd_mreq);
4202 		goto errout;
4203 	}
4204 
4205 	/*
4206 	 * Call newnfs_connect(), as required, and then newnfs_request().
4207 	 */
4208 	(void) newnfs_sndlock(&clp->lc_req.nr_lock);
4209 	if (clp->lc_req.nr_client == NULL) {
4210 		if ((clp->lc_flags & LCL_NFSV41) != 0)
4211 			error = ECONNREFUSED;
4212 		else if (nd->nd_procnum == NFSV4PROC_CBNULL)
4213 			error = newnfs_connect(NULL, &clp->lc_req, cred,
4214 			    NULL, 1);
4215 		else
4216 			error = newnfs_connect(NULL, &clp->lc_req, cred,
4217 			    NULL, 3);
4218 	}
4219 	newnfs_sndunlock(&clp->lc_req.nr_lock);
4220 	if (!error) {
4221 		if ((nd->nd_flag & ND_NFSV41) != 0) {
4222 			KASSERT(sep != NULL, ("sep NULL"));
4223 			if (sep->sess_cbsess.nfsess_xprt != NULL)
4224 				error = newnfs_request(nd, NULL, clp,
4225 				    &clp->lc_req, NULL, NULL, cred,
4226 				    clp->lc_program, clp->lc_req.nr_vers, NULL,
4227 				    1, NULL, &sep->sess_cbsess);
4228 			else {
4229 				/*
4230 				 * This should probably never occur, but if a
4231 				 * client somehow does an RPC without a
4232 				 * SequenceID Op that causes a callback just
4233 				 * after the nfsd threads have been terminated
4234 				 * and restared we could conceivably get here
4235 				 * without a backchannel xprt.
4236 				 */
4237 				printf("nfsrv_docallback: no xprt\n");
4238 				error = ECONNREFUSED;
4239 			}
4240 			nfsrv_freesession(sep, NULL);
4241 		} else
4242 			error = newnfs_request(nd, NULL, clp, &clp->lc_req,
4243 			    NULL, NULL, cred, clp->lc_program,
4244 			    clp->lc_req.nr_vers, NULL, 1, NULL, NULL);
4245 	}
4246 errout:
4247 	NFSFREECRED(cred);
4248 
4249 	/*
4250 	 * If error is set here, the Callback path isn't working
4251 	 * properly, so twiddle the appropriate LCL_ flags.
4252 	 * (nd_repstat != 0 indicates the Callback path is working,
4253 	 *  but the callback failed on the client.)
4254 	 */
4255 	if (error) {
4256 		/*
4257 		 * Mark the callback pathway down, which disabled issuing
4258 		 * of delegations and gets Renew to return NFSERR_CBPATHDOWN.
4259 		 */
4260 		NFSLOCKSTATE();
4261 		clp->lc_flags |= LCL_CBDOWN;
4262 		NFSUNLOCKSTATE();
4263 	} else {
4264 		/*
4265 		 * Callback worked. If the callback path was down, disable
4266 		 * callbacks, so no more delegations will be issued. (This
4267 		 * is done on the assumption that the callback pathway is
4268 		 * flakey.)
4269 		 */
4270 		NFSLOCKSTATE();
4271 		if (clp->lc_flags & LCL_CBDOWN)
4272 			clp->lc_flags &= ~(LCL_CBDOWN | LCL_CALLBACKSON);
4273 		NFSUNLOCKSTATE();
4274 		if (nd->nd_repstat)
4275 			error = nd->nd_repstat;
4276 		else if (error == 0 && procnum == NFSV4OP_CBGETATTR)
4277 			error = nfsv4_loadattr(nd, NULL, nap, NULL, NULL, 0,
4278 			    NULL, NULL, NULL, NULL, NULL, 0, NULL, NULL, NULL,
4279 			    p, NULL);
4280 		mbuf_freem(nd->nd_mrep);
4281 	}
4282 	NFSLOCKSTATE();
4283 	clp->lc_cbref--;
4284 	if ((clp->lc_flags & LCL_WAKEUPWANTED) && clp->lc_cbref == 0) {
4285 		clp->lc_flags &= ~LCL_WAKEUPWANTED;
4286 		wakeup(clp);
4287 	}
4288 	NFSUNLOCKSTATE();
4289 
4290 	NFSEXITCODE(error);
4291 	return (error);
4292 }
4293 
4294 /*
4295  * Set up the compound RPC for the callback.
4296  */
4297 static int
4298 nfsrv_cbcallargs(struct nfsrv_descript *nd, struct nfsclient *clp,
4299     uint32_t callback, int op, const char *optag, struct nfsdsession **sepp)
4300 {
4301 	uint32_t *tl;
4302 	int error, len;
4303 
4304 	len = strlen(optag);
4305 	(void)nfsm_strtom(nd, optag, len);
4306 	NFSM_BUILD(tl, uint32_t *, 4 * NFSX_UNSIGNED);
4307 	if ((nd->nd_flag & ND_NFSV41) != 0) {
4308 		*tl++ = txdr_unsigned(NFSV41_MINORVERSION);
4309 		*tl++ = txdr_unsigned(callback);
4310 		*tl++ = txdr_unsigned(2);
4311 		*tl = txdr_unsigned(NFSV4OP_CBSEQUENCE);
4312 		error = nfsv4_setcbsequence(nd, clp, 1, sepp);
4313 		if (error != 0)
4314 			return (error);
4315 		NFSM_BUILD(tl, u_int32_t *, NFSX_UNSIGNED);
4316 		*tl = txdr_unsigned(op);
4317 	} else {
4318 		*tl++ = txdr_unsigned(NFSV4_MINORVERSION);
4319 		*tl++ = txdr_unsigned(callback);
4320 		*tl++ = txdr_unsigned(1);
4321 		*tl = txdr_unsigned(op);
4322 	}
4323 	return (0);
4324 }
4325 
4326 /*
4327  * Return the next index# for a clientid. Mostly just increment and return
4328  * the next one, but... if the 32bit unsigned does actually wrap around,
4329  * it should be rebooted.
4330  * At an average rate of one new client per second, it will wrap around in
4331  * approximately 136 years. (I think the server will have been shut
4332  * down or rebooted before then.)
4333  */
4334 static u_int32_t
4335 nfsrv_nextclientindex(void)
4336 {
4337 	static u_int32_t client_index = 0;
4338 
4339 	client_index++;
4340 	if (client_index != 0)
4341 		return (client_index);
4342 
4343 	printf("%s: out of clientids\n", __func__);
4344 	return (client_index);
4345 }
4346 
4347 /*
4348  * Return the next index# for a stateid. Mostly just increment and return
4349  * the next one, but... if the 32bit unsigned does actually wrap around
4350  * (will a BSD server stay up that long?), find
4351  * new start and end values.
4352  */
4353 static u_int32_t
4354 nfsrv_nextstateindex(struct nfsclient *clp)
4355 {
4356 	struct nfsstate *stp;
4357 	int i;
4358 	u_int32_t canuse, min_index, max_index;
4359 
4360 	if (!(clp->lc_flags & LCL_INDEXNOTOK)) {
4361 		clp->lc_stateindex++;
4362 		if (clp->lc_stateindex != clp->lc_statemaxindex)
4363 			return (clp->lc_stateindex);
4364 	}
4365 
4366 	/*
4367 	 * Yuck, we've hit the end.
4368 	 * Look for a new min and max.
4369 	 */
4370 	min_index = 0;
4371 	max_index = 0xffffffff;
4372 	for (i = 0; i < nfsrv_statehashsize; i++) {
4373 	    LIST_FOREACH(stp, &clp->lc_stateid[i], ls_hash) {
4374 		if (stp->ls_stateid.other[2] > 0x80000000) {
4375 		    if (stp->ls_stateid.other[2] < max_index)
4376 			max_index = stp->ls_stateid.other[2];
4377 		} else {
4378 		    if (stp->ls_stateid.other[2] > min_index)
4379 			min_index = stp->ls_stateid.other[2];
4380 		}
4381 	    }
4382 	}
4383 
4384 	/*
4385 	 * Yikes, highly unlikely, but I'll handle it anyhow.
4386 	 */
4387 	if (min_index == 0x80000000 && max_index == 0x80000001) {
4388 	    canuse = 0;
4389 	    /*
4390 	     * Loop around until we find an unused entry. Return that
4391 	     * and set LCL_INDEXNOTOK, so the search will continue next time.
4392 	     * (This is one of those rare cases where a goto is the
4393 	     *  cleanest way to code the loop.)
4394 	     */
4395 tryagain:
4396 	    for (i = 0; i < nfsrv_statehashsize; i++) {
4397 		LIST_FOREACH(stp, &clp->lc_stateid[i], ls_hash) {
4398 		    if (stp->ls_stateid.other[2] == canuse) {
4399 			canuse++;
4400 			goto tryagain;
4401 		    }
4402 		}
4403 	    }
4404 	    clp->lc_flags |= LCL_INDEXNOTOK;
4405 	    return (canuse);
4406 	}
4407 
4408 	/*
4409 	 * Ok to start again from min + 1.
4410 	 */
4411 	clp->lc_stateindex = min_index + 1;
4412 	clp->lc_statemaxindex = max_index;
4413 	clp->lc_flags &= ~LCL_INDEXNOTOK;
4414 	return (clp->lc_stateindex);
4415 }
4416 
4417 /*
4418  * The following functions handle the stable storage file that deals with
4419  * the edge conditions described in RFC3530 Sec. 8.6.3.
4420  * The file is as follows:
4421  * - a single record at the beginning that has the lease time of the
4422  *   previous server instance (before the last reboot) and the nfsrvboottime
4423  *   values for the previous server boots.
4424  *   These previous boot times are used to ensure that the current
4425  *   nfsrvboottime does not, somehow, get set to a previous one.
4426  *   (This is important so that Stale ClientIDs and StateIDs can
4427  *    be recognized.)
4428  *   The number of previous nfsvrboottime values precedes the list.
4429  * - followed by some number of appended records with:
4430  *   - client id string
4431  *   - flag that indicates it is a record revoking state via lease
4432  *     expiration or similar
4433  *     OR has successfully acquired state.
4434  * These structures vary in length, with the client string at the end, up
4435  * to NFSV4_OPAQUELIMIT in size.
4436  *
4437  * At the end of the grace period, the file is truncated, the first
4438  * record is rewritten with updated information and any acquired state
4439  * records for successful reclaims of state are written.
4440  *
4441  * Subsequent records are appended when the first state is issued to
4442  * a client and when state is revoked for a client.
4443  *
4444  * When reading the file in, state issued records that come later in
4445  * the file override older ones, since the append log is in cronological order.
4446  * If, for some reason, the file can't be read, the grace period is
4447  * immediately terminated and all reclaims get NFSERR_NOGRACE.
4448  */
4449 
4450 /*
4451  * Read in the stable storage file. Called by nfssvc() before the nfsd
4452  * processes start servicing requests.
4453  */
4454 APPLESTATIC void
4455 nfsrv_setupstable(NFSPROC_T *p)
4456 {
4457 	struct nfsrv_stablefirst *sf = &nfsrv_stablefirst;
4458 	struct nfsrv_stable *sp, *nsp;
4459 	struct nfst_rec *tsp;
4460 	int error, i, tryagain;
4461 	off_t off = 0;
4462 	ssize_t aresid, len;
4463 
4464 	/*
4465 	 * If NFSNSF_UPDATEDONE is set, this is a restart of the nfsds without
4466 	 * a reboot, so state has not been lost.
4467 	 */
4468 	if (sf->nsf_flags & NFSNSF_UPDATEDONE)
4469 		return;
4470 	/*
4471 	 * Set Grace over just until the file reads successfully.
4472 	 */
4473 	nfsrvboottime = time_second;
4474 	LIST_INIT(&sf->nsf_head);
4475 	sf->nsf_flags = (NFSNSF_GRACEOVER | NFSNSF_NEEDLOCK);
4476 	sf->nsf_eograce = NFSD_MONOSEC + NFSRV_LEASEDELTA;
4477 	if (sf->nsf_fp == NULL)
4478 		return;
4479 	error = NFSD_RDWR(UIO_READ, NFSFPVNODE(sf->nsf_fp),
4480 	    (caddr_t)&sf->nsf_rec, sizeof (struct nfsf_rec), off, UIO_SYSSPACE,
4481 	    0, NFSFPCRED(sf->nsf_fp), &aresid, p);
4482 	if (error || aresid || sf->nsf_numboots == 0 ||
4483 		sf->nsf_numboots > NFSNSF_MAXNUMBOOTS)
4484 		return;
4485 
4486 	/*
4487 	 * Now, read in the boottimes.
4488 	 */
4489 	sf->nsf_bootvals = (time_t *)malloc((sf->nsf_numboots + 1) *
4490 		sizeof (time_t), M_TEMP, M_WAITOK);
4491 	off = sizeof (struct nfsf_rec);
4492 	error = NFSD_RDWR(UIO_READ, NFSFPVNODE(sf->nsf_fp),
4493 	    (caddr_t)sf->nsf_bootvals, sf->nsf_numboots * sizeof (time_t), off,
4494 	    UIO_SYSSPACE, 0, NFSFPCRED(sf->nsf_fp), &aresid, p);
4495 	if (error || aresid) {
4496 		free((caddr_t)sf->nsf_bootvals, M_TEMP);
4497 		sf->nsf_bootvals = NULL;
4498 		return;
4499 	}
4500 
4501 	/*
4502 	 * Make sure this nfsrvboottime is different from all recorded
4503 	 * previous ones.
4504 	 */
4505 	do {
4506 		tryagain = 0;
4507 		for (i = 0; i < sf->nsf_numboots; i++) {
4508 			if (nfsrvboottime == sf->nsf_bootvals[i]) {
4509 				nfsrvboottime++;
4510 				tryagain = 1;
4511 				break;
4512 			}
4513 		}
4514 	} while (tryagain);
4515 
4516 	sf->nsf_flags |= NFSNSF_OK;
4517 	off += (sf->nsf_numboots * sizeof (time_t));
4518 
4519 	/*
4520 	 * Read through the file, building a list of records for grace
4521 	 * checking.
4522 	 * Each record is between sizeof (struct nfst_rec) and
4523 	 * sizeof (struct nfst_rec) + NFSV4_OPAQUELIMIT - 1
4524 	 * and is actually sizeof (struct nfst_rec) + nst_len - 1.
4525 	 */
4526 	tsp = (struct nfst_rec *)malloc(sizeof (struct nfst_rec) +
4527 		NFSV4_OPAQUELIMIT - 1, M_TEMP, M_WAITOK);
4528 	do {
4529 	    error = NFSD_RDWR(UIO_READ, NFSFPVNODE(sf->nsf_fp),
4530 	        (caddr_t)tsp, sizeof (struct nfst_rec) + NFSV4_OPAQUELIMIT - 1,
4531 	        off, UIO_SYSSPACE, 0, NFSFPCRED(sf->nsf_fp), &aresid, p);
4532 	    len = (sizeof (struct nfst_rec) + NFSV4_OPAQUELIMIT - 1) - aresid;
4533 	    if (error || (len > 0 && (len < sizeof (struct nfst_rec) ||
4534 		len < (sizeof (struct nfst_rec) + tsp->len - 1)))) {
4535 		/*
4536 		 * Yuck, the file has been corrupted, so just return
4537 		 * after clearing out any restart state, so the grace period
4538 		 * is over.
4539 		 */
4540 		LIST_FOREACH_SAFE(sp, &sf->nsf_head, nst_list, nsp) {
4541 			LIST_REMOVE(sp, nst_list);
4542 			free((caddr_t)sp, M_TEMP);
4543 		}
4544 		free((caddr_t)tsp, M_TEMP);
4545 		sf->nsf_flags &= ~NFSNSF_OK;
4546 		free((caddr_t)sf->nsf_bootvals, M_TEMP);
4547 		sf->nsf_bootvals = NULL;
4548 		return;
4549 	    }
4550 	    if (len > 0) {
4551 		off += sizeof (struct nfst_rec) + tsp->len - 1;
4552 		/*
4553 		 * Search the list for a matching client.
4554 		 */
4555 		LIST_FOREACH(sp, &sf->nsf_head, nst_list) {
4556 			if (tsp->len == sp->nst_len &&
4557 			    !NFSBCMP(tsp->client, sp->nst_client, tsp->len))
4558 				break;
4559 		}
4560 		if (sp == LIST_END(&sf->nsf_head)) {
4561 			sp = (struct nfsrv_stable *)malloc(tsp->len +
4562 				sizeof (struct nfsrv_stable) - 1, M_TEMP,
4563 				M_WAITOK);
4564 			NFSBCOPY((caddr_t)tsp, (caddr_t)&sp->nst_rec,
4565 				sizeof (struct nfst_rec) + tsp->len - 1);
4566 			LIST_INSERT_HEAD(&sf->nsf_head, sp, nst_list);
4567 		} else {
4568 			if (tsp->flag == NFSNST_REVOKE)
4569 				sp->nst_flag |= NFSNST_REVOKE;
4570 			else
4571 				/*
4572 				 * A subsequent timestamp indicates the client
4573 				 * did a setclientid/confirm and any previous
4574 				 * revoke is no longer relevant.
4575 				 */
4576 				sp->nst_flag &= ~NFSNST_REVOKE;
4577 		}
4578 	    }
4579 	} while (len > 0);
4580 	free((caddr_t)tsp, M_TEMP);
4581 	sf->nsf_flags = NFSNSF_OK;
4582 	sf->nsf_eograce = NFSD_MONOSEC + sf->nsf_lease +
4583 		NFSRV_LEASEDELTA;
4584 }
4585 
4586 /*
4587  * Update the stable storage file, now that the grace period is over.
4588  */
4589 APPLESTATIC void
4590 nfsrv_updatestable(NFSPROC_T *p)
4591 {
4592 	struct nfsrv_stablefirst *sf = &nfsrv_stablefirst;
4593 	struct nfsrv_stable *sp, *nsp;
4594 	int i;
4595 	struct nfsvattr nva;
4596 	vnode_t vp;
4597 #if defined(__FreeBSD_version) && (__FreeBSD_version >= 500000)
4598 	mount_t mp = NULL;
4599 #endif
4600 	int error;
4601 
4602 	if (sf->nsf_fp == NULL || (sf->nsf_flags & NFSNSF_UPDATEDONE))
4603 		return;
4604 	sf->nsf_flags |= NFSNSF_UPDATEDONE;
4605 	/*
4606 	 * Ok, we need to rewrite the stable storage file.
4607 	 * - truncate to 0 length
4608 	 * - write the new first structure
4609 	 * - loop through the data structures, writing out any that
4610 	 *   have timestamps older than the old boot
4611 	 */
4612 	if (sf->nsf_bootvals) {
4613 		sf->nsf_numboots++;
4614 		for (i = sf->nsf_numboots - 2; i >= 0; i--)
4615 			sf->nsf_bootvals[i + 1] = sf->nsf_bootvals[i];
4616 	} else {
4617 		sf->nsf_numboots = 1;
4618 		sf->nsf_bootvals = (time_t *)malloc(sizeof (time_t),
4619 			M_TEMP, M_WAITOK);
4620 	}
4621 	sf->nsf_bootvals[0] = nfsrvboottime;
4622 	sf->nsf_lease = nfsrv_lease;
4623 	NFSVNO_ATTRINIT(&nva);
4624 	NFSVNO_SETATTRVAL(&nva, size, 0);
4625 	vp = NFSFPVNODE(sf->nsf_fp);
4626 	vn_start_write(vp, &mp, V_WAIT);
4627 	if (NFSVOPLOCK(vp, LK_EXCLUSIVE) == 0) {
4628 		error = nfsvno_setattr(vp, &nva, NFSFPCRED(sf->nsf_fp), p,
4629 		    NULL);
4630 		NFSVOPUNLOCK(vp, 0);
4631 	} else
4632 		error = EPERM;
4633 	vn_finished_write(mp);
4634 	if (!error)
4635 	    error = NFSD_RDWR(UIO_WRITE, vp,
4636 		(caddr_t)&sf->nsf_rec, sizeof (struct nfsf_rec), (off_t)0,
4637 		UIO_SYSSPACE, IO_SYNC, NFSFPCRED(sf->nsf_fp), NULL, p);
4638 	if (!error)
4639 	    error = NFSD_RDWR(UIO_WRITE, vp,
4640 		(caddr_t)sf->nsf_bootvals,
4641 		sf->nsf_numboots * sizeof (time_t),
4642 		(off_t)(sizeof (struct nfsf_rec)),
4643 		UIO_SYSSPACE, IO_SYNC, NFSFPCRED(sf->nsf_fp), NULL, p);
4644 	free((caddr_t)sf->nsf_bootvals, M_TEMP);
4645 	sf->nsf_bootvals = NULL;
4646 	if (error) {
4647 		sf->nsf_flags &= ~NFSNSF_OK;
4648 		printf("EEK! Can't write NfsV4 stable storage file\n");
4649 		return;
4650 	}
4651 	sf->nsf_flags |= NFSNSF_OK;
4652 
4653 	/*
4654 	 * Loop through the list and write out timestamp records for
4655 	 * any clients that successfully reclaimed state.
4656 	 */
4657 	LIST_FOREACH_SAFE(sp, &sf->nsf_head, nst_list, nsp) {
4658 		if (sp->nst_flag & NFSNST_GOTSTATE) {
4659 			nfsrv_writestable(sp->nst_client, sp->nst_len,
4660 				NFSNST_NEWSTATE, p);
4661 			sp->nst_clp->lc_flags |= LCL_STAMPEDSTABLE;
4662 		}
4663 		LIST_REMOVE(sp, nst_list);
4664 		free((caddr_t)sp, M_TEMP);
4665 	}
4666 	nfsrv_backupstable();
4667 }
4668 
4669 /*
4670  * Append a record to the stable storage file.
4671  */
4672 APPLESTATIC void
4673 nfsrv_writestable(u_char *client, int len, int flag, NFSPROC_T *p)
4674 {
4675 	struct nfsrv_stablefirst *sf = &nfsrv_stablefirst;
4676 	struct nfst_rec *sp;
4677 	int error;
4678 
4679 	if (!(sf->nsf_flags & NFSNSF_OK) || sf->nsf_fp == NULL)
4680 		return;
4681 	sp = (struct nfst_rec *)malloc(sizeof (struct nfst_rec) +
4682 		len - 1, M_TEMP, M_WAITOK);
4683 	sp->len = len;
4684 	NFSBCOPY(client, sp->client, len);
4685 	sp->flag = flag;
4686 	error = NFSD_RDWR(UIO_WRITE, NFSFPVNODE(sf->nsf_fp),
4687 	    (caddr_t)sp, sizeof (struct nfst_rec) + len - 1, (off_t)0,
4688 	    UIO_SYSSPACE, (IO_SYNC | IO_APPEND), NFSFPCRED(sf->nsf_fp), NULL, p);
4689 	free((caddr_t)sp, M_TEMP);
4690 	if (error) {
4691 		sf->nsf_flags &= ~NFSNSF_OK;
4692 		printf("EEK! Can't write NfsV4 stable storage file\n");
4693 	}
4694 }
4695 
4696 /*
4697  * This function is called during the grace period to mark a client
4698  * that successfully reclaimed state.
4699  */
4700 static void
4701 nfsrv_markstable(struct nfsclient *clp)
4702 {
4703 	struct nfsrv_stable *sp;
4704 
4705 	/*
4706 	 * First find the client structure.
4707 	 */
4708 	LIST_FOREACH(sp, &nfsrv_stablefirst.nsf_head, nst_list) {
4709 		if (sp->nst_len == clp->lc_idlen &&
4710 		    !NFSBCMP(sp->nst_client, clp->lc_id, sp->nst_len))
4711 			break;
4712 	}
4713 	if (sp == LIST_END(&nfsrv_stablefirst.nsf_head))
4714 		return;
4715 
4716 	/*
4717 	 * Now, just mark it and set the nfsclient back pointer.
4718 	 */
4719 	sp->nst_flag |= NFSNST_GOTSTATE;
4720 	sp->nst_clp = clp;
4721 }
4722 
4723 /*
4724  * This function is called for a reclaim, to see if it gets grace.
4725  * It returns 0 if a reclaim is allowed, 1 otherwise.
4726  */
4727 static int
4728 nfsrv_checkstable(struct nfsclient *clp)
4729 {
4730 	struct nfsrv_stable *sp;
4731 
4732 	/*
4733 	 * First, find the entry for the client.
4734 	 */
4735 	LIST_FOREACH(sp, &nfsrv_stablefirst.nsf_head, nst_list) {
4736 		if (sp->nst_len == clp->lc_idlen &&
4737 		    !NFSBCMP(sp->nst_client, clp->lc_id, sp->nst_len))
4738 			break;
4739 	}
4740 
4741 	/*
4742 	 * If not in the list, state was revoked or no state was issued
4743 	 * since the previous reboot, a reclaim is denied.
4744 	 */
4745 	if (sp == LIST_END(&nfsrv_stablefirst.nsf_head) ||
4746 	    (sp->nst_flag & NFSNST_REVOKE) ||
4747 	    !(nfsrv_stablefirst.nsf_flags & NFSNSF_OK))
4748 		return (1);
4749 	return (0);
4750 }
4751 
4752 /*
4753  * Test for and try to clear out a conflicting client. This is called by
4754  * nfsrv_lockctrl() and nfsrv_openctrl() when conflicts with other clients
4755  * a found.
4756  * The trick here is that it can't revoke a conflicting client with an
4757  * expired lease unless it holds the v4root lock, so...
4758  * If no v4root lock, get the lock and return 1 to indicate "try again".
4759  * Return 0 to indicate the conflict can't be revoked and 1 to indicate
4760  * the revocation worked and the conflicting client is "bye, bye", so it
4761  * can be tried again.
4762  * Return 2 to indicate that the vnode is VI_DOOMED after NFSVOPLOCK().
4763  * Unlocks State before a non-zero value is returned.
4764  */
4765 static int
4766 nfsrv_clientconflict(struct nfsclient *clp, int *haslockp, vnode_t vp,
4767     NFSPROC_T *p)
4768 {
4769 	int gotlock, lktype = 0;
4770 
4771 	/*
4772 	 * If lease hasn't expired, we can't fix it.
4773 	 */
4774 	if (clp->lc_expiry >= NFSD_MONOSEC ||
4775 	    !(nfsrv_stablefirst.nsf_flags & NFSNSF_UPDATEDONE))
4776 		return (0);
4777 	if (*haslockp == 0) {
4778 		NFSUNLOCKSTATE();
4779 		if (vp != NULL) {
4780 			lktype = NFSVOPISLOCKED(vp);
4781 			NFSVOPUNLOCK(vp, 0);
4782 		}
4783 		NFSLOCKV4ROOTMUTEX();
4784 		nfsv4_relref(&nfsv4rootfs_lock);
4785 		do {
4786 			gotlock = nfsv4_lock(&nfsv4rootfs_lock, 1, NULL,
4787 			    NFSV4ROOTLOCKMUTEXPTR, NULL);
4788 		} while (!gotlock);
4789 		NFSUNLOCKV4ROOTMUTEX();
4790 		*haslockp = 1;
4791 		if (vp != NULL) {
4792 			NFSVOPLOCK(vp, lktype | LK_RETRY);
4793 			if ((vp->v_iflag & VI_DOOMED) != 0)
4794 				return (2);
4795 		}
4796 		return (1);
4797 	}
4798 	NFSUNLOCKSTATE();
4799 
4800 	/*
4801 	 * Ok, we can expire the conflicting client.
4802 	 */
4803 	nfsrv_writestable(clp->lc_id, clp->lc_idlen, NFSNST_REVOKE, p);
4804 	nfsrv_backupstable();
4805 	nfsrv_cleanclient(clp, p);
4806 	nfsrv_freedeleglist(&clp->lc_deleg);
4807 	nfsrv_freedeleglist(&clp->lc_olddeleg);
4808 	LIST_REMOVE(clp, lc_hash);
4809 	nfsrv_zapclient(clp, p);
4810 	return (1);
4811 }
4812 
4813 /*
4814  * Resolve a delegation conflict.
4815  * Returns 0 to indicate the conflict was resolved without sleeping.
4816  * Return -1 to indicate that the caller should check for conflicts again.
4817  * Return > 0 for an error that should be returned, normally NFSERR_DELAY.
4818  *
4819  * Also, manipulate the nfsv4root_lock, as required. It isn't changed
4820  * for a return of 0, since there was no sleep and it could be required
4821  * later. It is released for a return of NFSERR_DELAY, since the caller
4822  * will return that error. It is released when a sleep was done waiting
4823  * for the delegation to be returned or expire (so that other nfsds can
4824  * handle ops). Then, it must be acquired for the write to stable storage.
4825  * (This function is somewhat similar to nfsrv_clientconflict(), but
4826  *  the semantics differ in a couple of subtle ways. The return of 0
4827  *  indicates the conflict was resolved without sleeping here, not
4828  *  that the conflict can't be resolved and the handling of nfsv4root_lock
4829  *  differs, as noted above.)
4830  * Unlocks State before returning a non-zero value.
4831  */
4832 static int
4833 nfsrv_delegconflict(struct nfsstate *stp, int *haslockp, NFSPROC_T *p,
4834     vnode_t vp)
4835 {
4836 	struct nfsclient *clp = stp->ls_clp;
4837 	int gotlock, error, lktype = 0, retrycnt, zapped_clp;
4838 	nfsv4stateid_t tstateid;
4839 	fhandle_t tfh;
4840 
4841 	/*
4842 	 * If the conflict is with an old delegation...
4843 	 */
4844 	if (stp->ls_flags & NFSLCK_OLDDELEG) {
4845 		/*
4846 		 * You can delete it, if it has expired.
4847 		 */
4848 		if (clp->lc_delegtime < NFSD_MONOSEC) {
4849 			nfsrv_freedeleg(stp);
4850 			NFSUNLOCKSTATE();
4851 			error = -1;
4852 			goto out;
4853 		}
4854 		NFSUNLOCKSTATE();
4855 		/*
4856 		 * During this delay, the old delegation could expire or it
4857 		 * could be recovered by the client via an Open with
4858 		 * CLAIM_DELEGATE_PREV.
4859 		 * Release the nfsv4root_lock, if held.
4860 		 */
4861 		if (*haslockp) {
4862 			*haslockp = 0;
4863 			NFSLOCKV4ROOTMUTEX();
4864 			nfsv4_unlock(&nfsv4rootfs_lock, 1);
4865 			NFSUNLOCKV4ROOTMUTEX();
4866 		}
4867 		error = NFSERR_DELAY;
4868 		goto out;
4869 	}
4870 
4871 	/*
4872 	 * It's a current delegation, so:
4873 	 * - check to see if the delegation has expired
4874 	 *   - if so, get the v4root lock and then expire it
4875 	 */
4876 	if (!(stp->ls_flags & NFSLCK_DELEGRECALL)) {
4877 		/*
4878 		 * - do a recall callback, since not yet done
4879 		 * For now, never allow truncate to be set. To use
4880 		 * truncate safely, it must be guaranteed that the
4881 		 * Remove, Rename or Setattr with size of 0 will
4882 		 * succeed and that would require major changes to
4883 		 * the VFS/Vnode OPs.
4884 		 * Set the expiry time large enough so that it won't expire
4885 		 * until after the callback, then set it correctly, once
4886 		 * the callback is done. (The delegation will now time
4887 		 * out whether or not the Recall worked ok. The timeout
4888 		 * will be extended when ops are done on the delegation
4889 		 * stateid, up to the timelimit.)
4890 		 */
4891 		stp->ls_delegtime = NFSD_MONOSEC + (2 * nfsrv_lease) +
4892 		    NFSRV_LEASEDELTA;
4893 		stp->ls_delegtimelimit = NFSD_MONOSEC + (6 * nfsrv_lease) +
4894 		    NFSRV_LEASEDELTA;
4895 		stp->ls_flags |= NFSLCK_DELEGRECALL;
4896 
4897 		/*
4898 		 * Loop NFSRV_CBRETRYCNT times while the CBRecall replies
4899 		 * NFSERR_BADSTATEID or NFSERR_BADHANDLE. This is done
4900 		 * in order to try and avoid a race that could happen
4901 		 * when a CBRecall request passed the Open reply with
4902 		 * the delegation in it when transitting the network.
4903 		 * Since nfsrv_docallback will sleep, don't use stp after
4904 		 * the call.
4905 		 */
4906 		NFSBCOPY((caddr_t)&stp->ls_stateid, (caddr_t)&tstateid,
4907 		    sizeof (tstateid));
4908 		NFSBCOPY((caddr_t)&stp->ls_lfp->lf_fh, (caddr_t)&tfh,
4909 		    sizeof (tfh));
4910 		NFSUNLOCKSTATE();
4911 		if (*haslockp) {
4912 			*haslockp = 0;
4913 			NFSLOCKV4ROOTMUTEX();
4914 			nfsv4_unlock(&nfsv4rootfs_lock, 1);
4915 			NFSUNLOCKV4ROOTMUTEX();
4916 		}
4917 		retrycnt = 0;
4918 		do {
4919 		    error = nfsrv_docallback(clp, NFSV4OP_CBRECALL,
4920 			&tstateid, 0, &tfh, NULL, NULL, p);
4921 		    retrycnt++;
4922 		} while ((error == NFSERR_BADSTATEID ||
4923 		    error == NFSERR_BADHANDLE) && retrycnt < NFSV4_CBRETRYCNT);
4924 		error = NFSERR_DELAY;
4925 		goto out;
4926 	}
4927 
4928 	if (clp->lc_expiry >= NFSD_MONOSEC &&
4929 	    stp->ls_delegtime >= NFSD_MONOSEC) {
4930 		NFSUNLOCKSTATE();
4931 		/*
4932 		 * A recall has been done, but it has not yet expired.
4933 		 * So, RETURN_DELAY.
4934 		 */
4935 		if (*haslockp) {
4936 			*haslockp = 0;
4937 			NFSLOCKV4ROOTMUTEX();
4938 			nfsv4_unlock(&nfsv4rootfs_lock, 1);
4939 			NFSUNLOCKV4ROOTMUTEX();
4940 		}
4941 		error = NFSERR_DELAY;
4942 		goto out;
4943 	}
4944 
4945 	/*
4946 	 * If we don't yet have the lock, just get it and then return,
4947 	 * since we need that before deleting expired state, such as
4948 	 * this delegation.
4949 	 * When getting the lock, unlock the vnode, so other nfsds that
4950 	 * are in progress, won't get stuck waiting for the vnode lock.
4951 	 */
4952 	if (*haslockp == 0) {
4953 		NFSUNLOCKSTATE();
4954 		if (vp != NULL) {
4955 			lktype = NFSVOPISLOCKED(vp);
4956 			NFSVOPUNLOCK(vp, 0);
4957 		}
4958 		NFSLOCKV4ROOTMUTEX();
4959 		nfsv4_relref(&nfsv4rootfs_lock);
4960 		do {
4961 			gotlock = nfsv4_lock(&nfsv4rootfs_lock, 1, NULL,
4962 			    NFSV4ROOTLOCKMUTEXPTR, NULL);
4963 		} while (!gotlock);
4964 		NFSUNLOCKV4ROOTMUTEX();
4965 		*haslockp = 1;
4966 		if (vp != NULL) {
4967 			NFSVOPLOCK(vp, lktype | LK_RETRY);
4968 			if ((vp->v_iflag & VI_DOOMED) != 0) {
4969 				*haslockp = 0;
4970 				NFSLOCKV4ROOTMUTEX();
4971 				nfsv4_unlock(&nfsv4rootfs_lock, 1);
4972 				NFSUNLOCKV4ROOTMUTEX();
4973 				error = NFSERR_PERM;
4974 				goto out;
4975 			}
4976 		}
4977 		error = -1;
4978 		goto out;
4979 	}
4980 
4981 	NFSUNLOCKSTATE();
4982 	/*
4983 	 * Ok, we can delete the expired delegation.
4984 	 * First, write the Revoke record to stable storage and then
4985 	 * clear out the conflict.
4986 	 * Since all other nfsd threads are now blocked, we can safely
4987 	 * sleep without the state changing.
4988 	 */
4989 	nfsrv_writestable(clp->lc_id, clp->lc_idlen, NFSNST_REVOKE, p);
4990 	nfsrv_backupstable();
4991 	if (clp->lc_expiry < NFSD_MONOSEC) {
4992 		nfsrv_cleanclient(clp, p);
4993 		nfsrv_freedeleglist(&clp->lc_deleg);
4994 		nfsrv_freedeleglist(&clp->lc_olddeleg);
4995 		LIST_REMOVE(clp, lc_hash);
4996 		zapped_clp = 1;
4997 	} else {
4998 		nfsrv_freedeleg(stp);
4999 		zapped_clp = 0;
5000 	}
5001 	if (zapped_clp)
5002 		nfsrv_zapclient(clp, p);
5003 	error = -1;
5004 
5005 out:
5006 	NFSEXITCODE(error);
5007 	return (error);
5008 }
5009 
5010 /*
5011  * Check for a remove allowed, if remove is set to 1 and get rid of
5012  * delegations.
5013  */
5014 APPLESTATIC int
5015 nfsrv_checkremove(vnode_t vp, int remove, NFSPROC_T *p)
5016 {
5017 	struct nfsstate *stp;
5018 	struct nfslockfile *lfp;
5019 	int error, haslock = 0;
5020 	fhandle_t nfh;
5021 
5022 	/*
5023 	 * First, get the lock file structure.
5024 	 * (A return of -1 means no associated state, so remove ok.)
5025 	 */
5026 	error = nfsrv_getlockfh(vp, NFSLCK_CHECK, NULL, &nfh, p);
5027 tryagain:
5028 	NFSLOCKSTATE();
5029 	if (!error)
5030 		error = nfsrv_getlockfile(NFSLCK_CHECK, NULL, &lfp, &nfh, 0);
5031 	if (error) {
5032 		NFSUNLOCKSTATE();
5033 		if (haslock) {
5034 			NFSLOCKV4ROOTMUTEX();
5035 			nfsv4_unlock(&nfsv4rootfs_lock, 1);
5036 			NFSUNLOCKV4ROOTMUTEX();
5037 		}
5038 		if (error == -1)
5039 			error = 0;
5040 		goto out;
5041 	}
5042 
5043 	/*
5044 	 * Now, we must Recall any delegations.
5045 	 */
5046 	error = nfsrv_cleandeleg(vp, lfp, NULL, &haslock, p);
5047 	if (error) {
5048 		/*
5049 		 * nfsrv_cleandeleg() unlocks state for non-zero
5050 		 * return.
5051 		 */
5052 		if (error == -1)
5053 			goto tryagain;
5054 		if (haslock) {
5055 			NFSLOCKV4ROOTMUTEX();
5056 			nfsv4_unlock(&nfsv4rootfs_lock, 1);
5057 			NFSUNLOCKV4ROOTMUTEX();
5058 		}
5059 		goto out;
5060 	}
5061 
5062 	/*
5063 	 * Now, look for a conflicting open share.
5064 	 */
5065 	if (remove) {
5066 		/*
5067 		 * If the entry in the directory was the last reference to the
5068 		 * corresponding filesystem object, the object can be destroyed
5069 		 * */
5070 		if(lfp->lf_usecount>1)
5071 			LIST_FOREACH(stp, &lfp->lf_open, ls_file) {
5072 				if (stp->ls_flags & NFSLCK_WRITEDENY) {
5073 					error = NFSERR_FILEOPEN;
5074 					break;
5075 				}
5076 			}
5077 	}
5078 
5079 	NFSUNLOCKSTATE();
5080 	if (haslock) {
5081 		NFSLOCKV4ROOTMUTEX();
5082 		nfsv4_unlock(&nfsv4rootfs_lock, 1);
5083 		NFSUNLOCKV4ROOTMUTEX();
5084 	}
5085 
5086 out:
5087 	NFSEXITCODE(error);
5088 	return (error);
5089 }
5090 
5091 /*
5092  * Clear out all delegations for the file referred to by lfp.
5093  * May return NFSERR_DELAY, if there will be a delay waiting for
5094  * delegations to expire.
5095  * Returns -1 to indicate it slept while recalling a delegation.
5096  * This function has the side effect of deleting the nfslockfile structure,
5097  * if it no longer has associated state and didn't have to sleep.
5098  * Unlocks State before a non-zero value is returned.
5099  */
5100 static int
5101 nfsrv_cleandeleg(vnode_t vp, struct nfslockfile *lfp,
5102     struct nfsclient *clp, int *haslockp, NFSPROC_T *p)
5103 {
5104 	struct nfsstate *stp, *nstp;
5105 	int ret = 0;
5106 
5107 	stp = LIST_FIRST(&lfp->lf_deleg);
5108 	while (stp != LIST_END(&lfp->lf_deleg)) {
5109 		nstp = LIST_NEXT(stp, ls_file);
5110 		if (stp->ls_clp != clp) {
5111 			ret = nfsrv_delegconflict(stp, haslockp, p, vp);
5112 			if (ret) {
5113 				/*
5114 				 * nfsrv_delegconflict() unlocks state
5115 				 * when it returns non-zero.
5116 				 */
5117 				goto out;
5118 			}
5119 		}
5120 		stp = nstp;
5121 	}
5122 out:
5123 	NFSEXITCODE(ret);
5124 	return (ret);
5125 }
5126 
5127 /*
5128  * There are certain operations that, when being done outside of NFSv4,
5129  * require that any NFSv4 delegation for the file be recalled.
5130  * This function is to be called for those cases:
5131  * VOP_RENAME() - When a delegation is being recalled for any reason,
5132  *	the client may have to do Opens against the server, using the file's
5133  *	final component name. If the file has been renamed on the server,
5134  *	that component name will be incorrect and the Open will fail.
5135  * VOP_REMOVE() - Theoretically, a client could Open a file after it has
5136  *	been removed on the server, if there is a delegation issued to
5137  *	that client for the file. I say "theoretically" since clients
5138  *	normally do an Access Op before the Open and that Access Op will
5139  *	fail with ESTALE. Note that NFSv2 and 3 don't even do Opens, so
5140  *	they will detect the file's removal in the same manner. (There is
5141  *	one case where RFC3530 allows a client to do an Open without first
5142  *	doing an Access Op, which is passage of a check against the ACE
5143  *	returned with a Write delegation, but current practice is to ignore
5144  *	the ACE and always do an Access Op.)
5145  *	Since the functions can only be called with an unlocked vnode, this
5146  *	can't be done at this time.
5147  * VOP_ADVLOCK() - When a client holds a delegation, it can issue byte range
5148  *	locks locally in the client, which are not visible to the server. To
5149  *	deal with this, issuing of delegations for a vnode must be disabled
5150  *	and all delegations for the vnode recalled. This is done via the
5151  *	second function, using the VV_DISABLEDELEG vflag on the vnode.
5152  */
5153 APPLESTATIC void
5154 nfsd_recalldelegation(vnode_t vp, NFSPROC_T *p)
5155 {
5156 	time_t starttime;
5157 	int error;
5158 
5159 	/*
5160 	 * First, check to see if the server is currently running and it has
5161 	 * been called for a regular file when issuing delegations.
5162 	 */
5163 	if (newnfs_numnfsd == 0 || vp->v_type != VREG ||
5164 	    nfsrv_issuedelegs == 0)
5165 		return;
5166 
5167 	KASSERT((NFSVOPISLOCKED(vp) != LK_EXCLUSIVE), ("vp %p is locked", vp));
5168 	/*
5169 	 * First, get a reference on the nfsv4rootfs_lock so that an
5170 	 * exclusive lock cannot be acquired by another thread.
5171 	 */
5172 	NFSLOCKV4ROOTMUTEX();
5173 	nfsv4_getref(&nfsv4rootfs_lock, NULL, NFSV4ROOTLOCKMUTEXPTR, NULL);
5174 	NFSUNLOCKV4ROOTMUTEX();
5175 
5176 	/*
5177 	 * Now, call nfsrv_checkremove() in a loop while it returns
5178 	 * NFSERR_DELAY. Return upon any other error or when timed out.
5179 	 */
5180 	starttime = NFSD_MONOSEC;
5181 	do {
5182 		if (NFSVOPLOCK(vp, LK_EXCLUSIVE) == 0) {
5183 			error = nfsrv_checkremove(vp, 0, p);
5184 			NFSVOPUNLOCK(vp, 0);
5185 		} else
5186 			error = EPERM;
5187 		if (error == NFSERR_DELAY) {
5188 			if (NFSD_MONOSEC - starttime > NFS_REMOVETIMEO)
5189 				break;
5190 			/* Sleep for a short period of time */
5191 			(void) nfs_catnap(PZERO, 0, "nfsremove");
5192 		}
5193 	} while (error == NFSERR_DELAY);
5194 	NFSLOCKV4ROOTMUTEX();
5195 	nfsv4_relref(&nfsv4rootfs_lock);
5196 	NFSUNLOCKV4ROOTMUTEX();
5197 }
5198 
5199 APPLESTATIC void
5200 nfsd_disabledelegation(vnode_t vp, NFSPROC_T *p)
5201 {
5202 
5203 #ifdef VV_DISABLEDELEG
5204 	/*
5205 	 * First, flag issuance of delegations disabled.
5206 	 */
5207 	atomic_set_long(&vp->v_vflag, VV_DISABLEDELEG);
5208 #endif
5209 
5210 	/*
5211 	 * Then call nfsd_recalldelegation() to get rid of all extant
5212 	 * delegations.
5213 	 */
5214 	nfsd_recalldelegation(vp, p);
5215 }
5216 
5217 /*
5218  * Check for conflicting locks, etc. and then get rid of delegations.
5219  * (At one point I thought that I should get rid of delegations for any
5220  *  Setattr, since it could potentially disallow the I/O op (read or write)
5221  *  allowed by the delegation. However, Setattr Ops that aren't changing
5222  *  the size get a stateid of all 0s, so you can't tell if it is a delegation
5223  *  for the same client or a different one, so I decided to only get rid
5224  *  of delegations for other clients when the size is being changed.)
5225  * In general, a Setattr can disable NFS I/O Ops that are outstanding, such
5226  * as Write backs, even if there is no delegation, so it really isn't any
5227  * different?)
5228  */
5229 APPLESTATIC int
5230 nfsrv_checksetattr(vnode_t vp, struct nfsrv_descript *nd,
5231     nfsv4stateid_t *stateidp, struct nfsvattr *nvap, nfsattrbit_t *attrbitp,
5232     struct nfsexstuff *exp, NFSPROC_T *p)
5233 {
5234 	struct nfsstate st, *stp = &st;
5235 	struct nfslock lo, *lop = &lo;
5236 	int error = 0;
5237 	nfsquad_t clientid;
5238 
5239 	if (NFSISSET_ATTRBIT(attrbitp, NFSATTRBIT_SIZE)) {
5240 		stp->ls_flags = (NFSLCK_CHECK | NFSLCK_WRITEACCESS);
5241 		lop->lo_first = nvap->na_size;
5242 	} else {
5243 		stp->ls_flags = 0;
5244 		lop->lo_first = 0;
5245 	}
5246 	if (NFSISSET_ATTRBIT(attrbitp, NFSATTRBIT_OWNER) ||
5247 	    NFSISSET_ATTRBIT(attrbitp, NFSATTRBIT_OWNERGROUP) ||
5248 	    NFSISSET_ATTRBIT(attrbitp, NFSATTRBIT_MODE) ||
5249 	    NFSISSET_ATTRBIT(attrbitp, NFSATTRBIT_ACL))
5250 		stp->ls_flags |= NFSLCK_SETATTR;
5251 	if (stp->ls_flags == 0)
5252 		goto out;
5253 	lop->lo_end = NFS64BITSSET;
5254 	lop->lo_flags = NFSLCK_WRITE;
5255 	stp->ls_ownerlen = 0;
5256 	stp->ls_op = NULL;
5257 	stp->ls_uid = nd->nd_cred->cr_uid;
5258 	stp->ls_stateid.seqid = stateidp->seqid;
5259 	clientid.lval[0] = stp->ls_stateid.other[0] = stateidp->other[0];
5260 	clientid.lval[1] = stp->ls_stateid.other[1] = stateidp->other[1];
5261 	stp->ls_stateid.other[2] = stateidp->other[2];
5262 	error = nfsrv_lockctrl(vp, &stp, &lop, NULL, clientid,
5263 	    stateidp, exp, nd, p);
5264 
5265 out:
5266 	NFSEXITCODE2(error, nd);
5267 	return (error);
5268 }
5269 
5270 /*
5271  * Check for a write delegation and do a CBGETATTR if there is one, updating
5272  * the attributes, as required.
5273  * Should I return an error if I can't get the attributes? (For now, I'll
5274  * just return ok.
5275  */
5276 APPLESTATIC int
5277 nfsrv_checkgetattr(struct nfsrv_descript *nd, vnode_t vp,
5278     struct nfsvattr *nvap, nfsattrbit_t *attrbitp, struct ucred *cred,
5279     NFSPROC_T *p)
5280 {
5281 	struct nfsstate *stp;
5282 	struct nfslockfile *lfp;
5283 	struct nfsclient *clp;
5284 	struct nfsvattr nva;
5285 	fhandle_t nfh;
5286 	int error = 0;
5287 	nfsattrbit_t cbbits;
5288 	u_quad_t delegfilerev;
5289 
5290 	NFSCBGETATTR_ATTRBIT(attrbitp, &cbbits);
5291 	if (!NFSNONZERO_ATTRBIT(&cbbits))
5292 		goto out;
5293 
5294 	/*
5295 	 * Get the lock file structure.
5296 	 * (A return of -1 means no associated state, so return ok.)
5297 	 */
5298 	error = nfsrv_getlockfh(vp, NFSLCK_CHECK, NULL, &nfh, p);
5299 	NFSLOCKSTATE();
5300 	if (!error)
5301 		error = nfsrv_getlockfile(NFSLCK_CHECK, NULL, &lfp, &nfh, 0);
5302 	if (error) {
5303 		NFSUNLOCKSTATE();
5304 		if (error == -1)
5305 			error = 0;
5306 		goto out;
5307 	}
5308 
5309 	/*
5310 	 * Now, look for a write delegation.
5311 	 */
5312 	LIST_FOREACH(stp, &lfp->lf_deleg, ls_file) {
5313 		if (stp->ls_flags & NFSLCK_DELEGWRITE)
5314 			break;
5315 	}
5316 	if (stp == LIST_END(&lfp->lf_deleg)) {
5317 		NFSUNLOCKSTATE();
5318 		goto out;
5319 	}
5320 	clp = stp->ls_clp;
5321 	delegfilerev = stp->ls_filerev;
5322 
5323 	/*
5324 	 * If the Write delegation was issued as a part of this Compound RPC
5325 	 * or if we have an Implied Clientid (used in a previous Op in this
5326 	 * compound) and it is the client the delegation was issued to,
5327 	 * just return ok.
5328 	 * I also assume that it is from the same client iff the network
5329 	 * host IP address is the same as the callback address. (Not
5330 	 * exactly correct by the RFC, but avoids a lot of Getattr
5331 	 * callbacks.)
5332 	 */
5333 	if (nd->nd_compref == stp->ls_compref ||
5334 	    ((nd->nd_flag & ND_IMPLIEDCLID) &&
5335 	     clp->lc_clientid.qval == nd->nd_clientid.qval) ||
5336 	     nfsaddr2_match(clp->lc_req.nr_nam, nd->nd_nam)) {
5337 		NFSUNLOCKSTATE();
5338 		goto out;
5339 	}
5340 
5341 	/*
5342 	 * We are now done with the delegation state structure,
5343 	 * so the statelock can be released and we can now tsleep().
5344 	 */
5345 
5346 	/*
5347 	 * Now, we must do the CB Getattr callback, to see if Change or Size
5348 	 * has changed.
5349 	 */
5350 	if (clp->lc_expiry >= NFSD_MONOSEC) {
5351 		NFSUNLOCKSTATE();
5352 		NFSVNO_ATTRINIT(&nva);
5353 		nva.na_filerev = NFS64BITSSET;
5354 		error = nfsrv_docallback(clp, NFSV4OP_CBGETATTR, NULL,
5355 		    0, &nfh, &nva, &cbbits, p);
5356 		if (!error) {
5357 			if ((nva.na_filerev != NFS64BITSSET &&
5358 			    nva.na_filerev > delegfilerev) ||
5359 			    (NFSVNO_ISSETSIZE(&nva) &&
5360 			     nva.na_size != nvap->na_size)) {
5361 				error = nfsvno_updfilerev(vp, nvap, cred, p);
5362 				if (NFSVNO_ISSETSIZE(&nva))
5363 					nvap->na_size = nva.na_size;
5364 			}
5365 		} else
5366 			error = 0;	/* Ignore callback errors for now. */
5367 	} else {
5368 		NFSUNLOCKSTATE();
5369 	}
5370 
5371 out:
5372 	NFSEXITCODE2(error, nd);
5373 	return (error);
5374 }
5375 
5376 /*
5377  * This function looks for openowners that haven't had any opens for
5378  * a while and throws them away. Called by an nfsd when NFSNSF_NOOPENS
5379  * is set.
5380  */
5381 APPLESTATIC void
5382 nfsrv_throwawayopens(NFSPROC_T *p)
5383 {
5384 	struct nfsclient *clp, *nclp;
5385 	struct nfsstate *stp, *nstp;
5386 	int i;
5387 
5388 	NFSLOCKSTATE();
5389 	nfsrv_stablefirst.nsf_flags &= ~NFSNSF_NOOPENS;
5390 	/*
5391 	 * For each client...
5392 	 */
5393 	for (i = 0; i < nfsrv_clienthashsize; i++) {
5394 	    LIST_FOREACH_SAFE(clp, &nfsclienthash[i], lc_hash, nclp) {
5395 		LIST_FOREACH_SAFE(stp, &clp->lc_open, ls_list, nstp) {
5396 			if (LIST_EMPTY(&stp->ls_open) &&
5397 			    (stp->ls_noopens > NFSNOOPEN ||
5398 			     (nfsrv_openpluslock * 2) >
5399 			     nfsrv_v4statelimit))
5400 				nfsrv_freeopenowner(stp, 0, p);
5401 		}
5402 	    }
5403 	}
5404 	NFSUNLOCKSTATE();
5405 }
5406 
5407 /*
5408  * This function checks to see if the credentials are the same.
5409  * Returns 1 for not same, 0 otherwise.
5410  */
5411 static int
5412 nfsrv_notsamecredname(struct nfsrv_descript *nd, struct nfsclient *clp)
5413 {
5414 
5415 	if (nd->nd_flag & ND_GSS) {
5416 		if (!(clp->lc_flags & LCL_GSS))
5417 			return (1);
5418 		if (clp->lc_flags & LCL_NAME) {
5419 			if (nd->nd_princlen != clp->lc_namelen ||
5420 			    NFSBCMP(nd->nd_principal, clp->lc_name,
5421 				clp->lc_namelen))
5422 				return (1);
5423 			else
5424 				return (0);
5425 		}
5426 		if (nd->nd_cred->cr_uid == clp->lc_uid)
5427 			return (0);
5428 		else
5429 			return (1);
5430 	} else if (clp->lc_flags & LCL_GSS)
5431 		return (1);
5432 	/*
5433 	 * For AUTH_SYS, allow the same uid or root. (This is underspecified
5434 	 * in RFC3530, which talks about principals, but doesn't say anything
5435 	 * about uids for AUTH_SYS.)
5436 	 */
5437 	if (nd->nd_cred->cr_uid == clp->lc_uid || nd->nd_cred->cr_uid == 0)
5438 		return (0);
5439 	else
5440 		return (1);
5441 }
5442 
5443 /*
5444  * Calculate the lease expiry time.
5445  */
5446 static time_t
5447 nfsrv_leaseexpiry(void)
5448 {
5449 
5450 	if (nfsrv_stablefirst.nsf_eograce > NFSD_MONOSEC)
5451 		return (NFSD_MONOSEC + 2 * (nfsrv_lease + NFSRV_LEASEDELTA));
5452 	return (NFSD_MONOSEC + nfsrv_lease + NFSRV_LEASEDELTA);
5453 }
5454 
5455 /*
5456  * Delay the delegation timeout as far as ls_delegtimelimit, as required.
5457  */
5458 static void
5459 nfsrv_delaydelegtimeout(struct nfsstate *stp)
5460 {
5461 
5462 	if ((stp->ls_flags & NFSLCK_DELEGRECALL) == 0)
5463 		return;
5464 
5465 	if ((stp->ls_delegtime + 15) > NFSD_MONOSEC &&
5466 	    stp->ls_delegtime < stp->ls_delegtimelimit) {
5467 		stp->ls_delegtime += nfsrv_lease;
5468 		if (stp->ls_delegtime > stp->ls_delegtimelimit)
5469 			stp->ls_delegtime = stp->ls_delegtimelimit;
5470 	}
5471 }
5472 
5473 /*
5474  * This function checks to see if there is any other state associated
5475  * with the openowner for this Open.
5476  * It returns 1 if there is no other state, 0 otherwise.
5477  */
5478 static int
5479 nfsrv_nootherstate(struct nfsstate *stp)
5480 {
5481 	struct nfsstate *tstp;
5482 
5483 	LIST_FOREACH(tstp, &stp->ls_openowner->ls_open, ls_list) {
5484 		if (tstp != stp || !LIST_EMPTY(&tstp->ls_lock))
5485 			return (0);
5486 	}
5487 	return (1);
5488 }
5489 
5490 /*
5491  * Create a list of lock deltas (changes to local byte range locking
5492  * that can be rolled back using the list) and apply the changes via
5493  * nfsvno_advlock(). Optionally, lock the list. It is expected that either
5494  * the rollback or update function will be called after this.
5495  * It returns an error (and rolls back, as required), if any nfsvno_advlock()
5496  * call fails. If it returns an error, it will unlock the list.
5497  */
5498 static int
5499 nfsrv_locallock(vnode_t vp, struct nfslockfile *lfp, int flags,
5500     uint64_t first, uint64_t end, struct nfslockconflict *cfp, NFSPROC_T *p)
5501 {
5502 	struct nfslock *lop, *nlop;
5503 	int error = 0;
5504 
5505 	/* Loop through the list of locks. */
5506 	lop = LIST_FIRST(&lfp->lf_locallock);
5507 	while (first < end && lop != NULL) {
5508 		nlop = LIST_NEXT(lop, lo_lckowner);
5509 		if (first >= lop->lo_end) {
5510 			/* not there yet */
5511 			lop = nlop;
5512 		} else if (first < lop->lo_first) {
5513 			/* new one starts before entry in list */
5514 			if (end <= lop->lo_first) {
5515 				/* no overlap between old and new */
5516 				error = nfsrv_dolocal(vp, lfp, flags,
5517 				    NFSLCK_UNLOCK, first, end, cfp, p);
5518 				if (error != 0)
5519 					break;
5520 				first = end;
5521 			} else {
5522 				/* handle fragment overlapped with new one */
5523 				error = nfsrv_dolocal(vp, lfp, flags,
5524 				    NFSLCK_UNLOCK, first, lop->lo_first, cfp,
5525 				    p);
5526 				if (error != 0)
5527 					break;
5528 				first = lop->lo_first;
5529 			}
5530 		} else {
5531 			/* new one overlaps this entry in list */
5532 			if (end <= lop->lo_end) {
5533 				/* overlaps all of new one */
5534 				error = nfsrv_dolocal(vp, lfp, flags,
5535 				    lop->lo_flags, first, end, cfp, p);
5536 				if (error != 0)
5537 					break;
5538 				first = end;
5539 			} else {
5540 				/* handle fragment overlapped with new one */
5541 				error = nfsrv_dolocal(vp, lfp, flags,
5542 				    lop->lo_flags, first, lop->lo_end, cfp, p);
5543 				if (error != 0)
5544 					break;
5545 				first = lop->lo_end;
5546 				lop = nlop;
5547 			}
5548 		}
5549 	}
5550 	if (first < end && error == 0)
5551 		/* handle fragment past end of list */
5552 		error = nfsrv_dolocal(vp, lfp, flags, NFSLCK_UNLOCK, first,
5553 		    end, cfp, p);
5554 
5555 	NFSEXITCODE(error);
5556 	return (error);
5557 }
5558 
5559 /*
5560  * Local lock unlock. Unlock all byte ranges that are no longer locked
5561  * by NFSv4. To do this, unlock any subranges of first-->end that
5562  * do not overlap with the byte ranges of any lock in the lfp->lf_lock
5563  * list. This list has all locks for the file held by other
5564  * <clientid, lockowner> tuples. The list is ordered by increasing
5565  * lo_first value, but may have entries that overlap each other, for
5566  * the case of read locks.
5567  */
5568 static void
5569 nfsrv_localunlock(vnode_t vp, struct nfslockfile *lfp, uint64_t init_first,
5570     uint64_t init_end, NFSPROC_T *p)
5571 {
5572 	struct nfslock *lop;
5573 	uint64_t first, end, prevfirst;
5574 
5575 	first = init_first;
5576 	end = init_end;
5577 	while (first < init_end) {
5578 		/* Loop through all nfs locks, adjusting first and end */
5579 		prevfirst = 0;
5580 		LIST_FOREACH(lop, &lfp->lf_lock, lo_lckfile) {
5581 			KASSERT(prevfirst <= lop->lo_first,
5582 			    ("nfsv4 locks out of order"));
5583 			KASSERT(lop->lo_first < lop->lo_end,
5584 			    ("nfsv4 bogus lock"));
5585 			prevfirst = lop->lo_first;
5586 			if (first >= lop->lo_first &&
5587 			    first < lop->lo_end)
5588 				/*
5589 				 * Overlaps with initial part, so trim
5590 				 * off that initial part by moving first past
5591 				 * it.
5592 				 */
5593 				first = lop->lo_end;
5594 			else if (end > lop->lo_first &&
5595 			    lop->lo_first > first) {
5596 				/*
5597 				 * This lock defines the end of the
5598 				 * segment to unlock, so set end to the
5599 				 * start of it and break out of the loop.
5600 				 */
5601 				end = lop->lo_first;
5602 				break;
5603 			}
5604 			if (first >= end)
5605 				/*
5606 				 * There is no segment left to do, so
5607 				 * break out of this loop and then exit
5608 				 * the outer while() since first will be set
5609 				 * to end, which must equal init_end here.
5610 				 */
5611 				break;
5612 		}
5613 		if (first < end) {
5614 			/* Unlock this segment */
5615 			(void) nfsrv_dolocal(vp, lfp, NFSLCK_UNLOCK,
5616 			    NFSLCK_READ, first, end, NULL, p);
5617 			nfsrv_locallock_commit(lfp, NFSLCK_UNLOCK,
5618 			    first, end);
5619 		}
5620 		/*
5621 		 * Now move past this segment and look for any further
5622 		 * segment in the range, if there is one.
5623 		 */
5624 		first = end;
5625 		end = init_end;
5626 	}
5627 }
5628 
5629 /*
5630  * Do the local lock operation and update the rollback list, as required.
5631  * Perform the rollback and return the error if nfsvno_advlock() fails.
5632  */
5633 static int
5634 nfsrv_dolocal(vnode_t vp, struct nfslockfile *lfp, int flags, int oldflags,
5635     uint64_t first, uint64_t end, struct nfslockconflict *cfp, NFSPROC_T *p)
5636 {
5637 	struct nfsrollback *rlp;
5638 	int error = 0, ltype, oldltype;
5639 
5640 	if (flags & NFSLCK_WRITE)
5641 		ltype = F_WRLCK;
5642 	else if (flags & NFSLCK_READ)
5643 		ltype = F_RDLCK;
5644 	else
5645 		ltype = F_UNLCK;
5646 	if (oldflags & NFSLCK_WRITE)
5647 		oldltype = F_WRLCK;
5648 	else if (oldflags & NFSLCK_READ)
5649 		oldltype = F_RDLCK;
5650 	else
5651 		oldltype = F_UNLCK;
5652 	if (ltype == oldltype || (oldltype == F_WRLCK && ltype == F_RDLCK))
5653 		/* nothing to do */
5654 		goto out;
5655 	error = nfsvno_advlock(vp, ltype, first, end, p);
5656 	if (error != 0) {
5657 		if (cfp != NULL) {
5658 			cfp->cl_clientid.lval[0] = 0;
5659 			cfp->cl_clientid.lval[1] = 0;
5660 			cfp->cl_first = 0;
5661 			cfp->cl_end = NFS64BITSSET;
5662 			cfp->cl_flags = NFSLCK_WRITE;
5663 			cfp->cl_ownerlen = 5;
5664 			NFSBCOPY("LOCAL", cfp->cl_owner, 5);
5665 		}
5666 		nfsrv_locallock_rollback(vp, lfp, p);
5667 	} else if (ltype != F_UNLCK) {
5668 		rlp = malloc(sizeof (struct nfsrollback), M_NFSDROLLBACK,
5669 		    M_WAITOK);
5670 		rlp->rlck_first = first;
5671 		rlp->rlck_end = end;
5672 		rlp->rlck_type = oldltype;
5673 		LIST_INSERT_HEAD(&lfp->lf_rollback, rlp, rlck_list);
5674 	}
5675 
5676 out:
5677 	NFSEXITCODE(error);
5678 	return (error);
5679 }
5680 
5681 /*
5682  * Roll back local lock changes and free up the rollback list.
5683  */
5684 static void
5685 nfsrv_locallock_rollback(vnode_t vp, struct nfslockfile *lfp, NFSPROC_T *p)
5686 {
5687 	struct nfsrollback *rlp, *nrlp;
5688 
5689 	LIST_FOREACH_SAFE(rlp, &lfp->lf_rollback, rlck_list, nrlp) {
5690 		(void) nfsvno_advlock(vp, rlp->rlck_type, rlp->rlck_first,
5691 		    rlp->rlck_end, p);
5692 		free(rlp, M_NFSDROLLBACK);
5693 	}
5694 	LIST_INIT(&lfp->lf_rollback);
5695 }
5696 
5697 /*
5698  * Update local lock list and delete rollback list (ie now committed to the
5699  * local locks). Most of the work is done by the internal function.
5700  */
5701 static void
5702 nfsrv_locallock_commit(struct nfslockfile *lfp, int flags, uint64_t first,
5703     uint64_t end)
5704 {
5705 	struct nfsrollback *rlp, *nrlp;
5706 	struct nfslock *new_lop, *other_lop;
5707 
5708 	new_lop = malloc(sizeof (struct nfslock), M_NFSDLOCK, M_WAITOK);
5709 	if (flags & (NFSLCK_READ | NFSLCK_WRITE))
5710 		other_lop = malloc(sizeof (struct nfslock), M_NFSDLOCK,
5711 		    M_WAITOK);
5712 	else
5713 		other_lop = NULL;
5714 	new_lop->lo_flags = flags;
5715 	new_lop->lo_first = first;
5716 	new_lop->lo_end = end;
5717 	nfsrv_updatelock(NULL, &new_lop, &other_lop, lfp);
5718 	if (new_lop != NULL)
5719 		free(new_lop, M_NFSDLOCK);
5720 	if (other_lop != NULL)
5721 		free(other_lop, M_NFSDLOCK);
5722 
5723 	/* and get rid of the rollback list */
5724 	LIST_FOREACH_SAFE(rlp, &lfp->lf_rollback, rlck_list, nrlp)
5725 		free(rlp, M_NFSDROLLBACK);
5726 	LIST_INIT(&lfp->lf_rollback);
5727 }
5728 
5729 /*
5730  * Lock the struct nfslockfile for local lock updating.
5731  */
5732 static void
5733 nfsrv_locklf(struct nfslockfile *lfp)
5734 {
5735 	int gotlock;
5736 
5737 	/* lf_usecount ensures *lfp won't be free'd */
5738 	lfp->lf_usecount++;
5739 	do {
5740 		gotlock = nfsv4_lock(&lfp->lf_locallock_lck, 1, NULL,
5741 		    NFSSTATEMUTEXPTR, NULL);
5742 	} while (gotlock == 0);
5743 	lfp->lf_usecount--;
5744 }
5745 
5746 /*
5747  * Unlock the struct nfslockfile after local lock updating.
5748  */
5749 static void
5750 nfsrv_unlocklf(struct nfslockfile *lfp)
5751 {
5752 
5753 	nfsv4_unlock(&lfp->lf_locallock_lck, 0);
5754 }
5755 
5756 /*
5757  * Clear out all state for the NFSv4 server.
5758  * Must be called by a thread that can sleep when no nfsds are running.
5759  */
5760 void
5761 nfsrv_throwawayallstate(NFSPROC_T *p)
5762 {
5763 	struct nfsclient *clp, *nclp;
5764 	struct nfslockfile *lfp, *nlfp;
5765 	int i;
5766 
5767 	/*
5768 	 * For each client, clean out the state and then free the structure.
5769 	 */
5770 	for (i = 0; i < nfsrv_clienthashsize; i++) {
5771 		LIST_FOREACH_SAFE(clp, &nfsclienthash[i], lc_hash, nclp) {
5772 			nfsrv_cleanclient(clp, p);
5773 			nfsrv_freedeleglist(&clp->lc_deleg);
5774 			nfsrv_freedeleglist(&clp->lc_olddeleg);
5775 			free(clp->lc_stateid, M_NFSDCLIENT);
5776 			free(clp, M_NFSDCLIENT);
5777 		}
5778 	}
5779 
5780 	/*
5781 	 * Also, free up any remaining lock file structures.
5782 	 */
5783 	for (i = 0; i < nfsrv_lockhashsize; i++) {
5784 		LIST_FOREACH_SAFE(lfp, &nfslockhash[i], lf_hash, nlfp) {
5785 			printf("nfsd unload: fnd a lock file struct\n");
5786 			nfsrv_freenfslockfile(lfp);
5787 		}
5788 	}
5789 }
5790 
5791 /*
5792  * Check the sequence# for the session and slot provided as an argument.
5793  * Also, renew the lease if the session will return NFS_OK.
5794  */
5795 int
5796 nfsrv_checksequence(struct nfsrv_descript *nd, uint32_t sequenceid,
5797     uint32_t *highest_slotidp, uint32_t *target_highest_slotidp, int cache_this,
5798     uint32_t *sflagsp, NFSPROC_T *p)
5799 {
5800 	struct nfsdsession *sep;
5801 	struct nfssessionhash *shp;
5802 	int error;
5803 	SVCXPRT *savxprt;
5804 
5805 	shp = NFSSESSIONHASH(nd->nd_sessionid);
5806 	NFSLOCKSESSION(shp);
5807 	sep = nfsrv_findsession(nd->nd_sessionid);
5808 	if (sep == NULL) {
5809 		NFSUNLOCKSESSION(shp);
5810 		return (NFSERR_BADSESSION);
5811 	}
5812 	error = nfsv4_seqsession(sequenceid, nd->nd_slotid, *highest_slotidp,
5813 	    sep->sess_slots, NULL, NFSV4_SLOTS - 1);
5814 	if (error != 0) {
5815 		NFSUNLOCKSESSION(shp);
5816 		return (error);
5817 	}
5818 	if (cache_this != 0)
5819 		nd->nd_flag |= ND_SAVEREPLY;
5820 	/* Renew the lease. */
5821 	sep->sess_clp->lc_expiry = nfsrv_leaseexpiry();
5822 	nd->nd_clientid.qval = sep->sess_clp->lc_clientid.qval;
5823 	nd->nd_flag |= ND_IMPLIEDCLID;
5824 
5825 	/*
5826 	 * If this session handles the backchannel, save the nd_xprt for this
5827 	 * RPC, since this is the one being used.
5828 	 */
5829 	if (sep->sess_clp->lc_req.nr_client != NULL &&
5830 	    (sep->sess_crflags & NFSV4CRSESS_CONNBACKCHAN) != 0) {
5831 		savxprt = sep->sess_cbsess.nfsess_xprt;
5832 		SVC_ACQUIRE(nd->nd_xprt);
5833 		nd->nd_xprt->xp_p2 =
5834 		    sep->sess_clp->lc_req.nr_client->cl_private;
5835 		nd->nd_xprt->xp_idletimeout = 0;	/* Disable timeout. */
5836 		sep->sess_cbsess.nfsess_xprt = nd->nd_xprt;
5837 		if (savxprt != NULL)
5838 			SVC_RELEASE(savxprt);
5839 	}
5840 
5841 	*sflagsp = 0;
5842 	if (sep->sess_clp->lc_req.nr_client == NULL)
5843 		*sflagsp |= NFSV4SEQ_CBPATHDOWN;
5844 	NFSUNLOCKSESSION(shp);
5845 	if (error == NFSERR_EXPIRED) {
5846 		*sflagsp |= NFSV4SEQ_EXPIREDALLSTATEREVOKED;
5847 		error = 0;
5848 	} else if (error == NFSERR_ADMINREVOKED) {
5849 		*sflagsp |= NFSV4SEQ_ADMINSTATEREVOKED;
5850 		error = 0;
5851 	}
5852 	*highest_slotidp = *target_highest_slotidp = NFSV4_SLOTS - 1;
5853 	return (0);
5854 }
5855 
5856 /*
5857  * Check/set reclaim complete for this session/clientid.
5858  */
5859 int
5860 nfsrv_checkreclaimcomplete(struct nfsrv_descript *nd)
5861 {
5862 	struct nfsdsession *sep;
5863 	struct nfssessionhash *shp;
5864 	int error = 0;
5865 
5866 	shp = NFSSESSIONHASH(nd->nd_sessionid);
5867 	NFSLOCKSTATE();
5868 	NFSLOCKSESSION(shp);
5869 	sep = nfsrv_findsession(nd->nd_sessionid);
5870 	if (sep == NULL) {
5871 		NFSUNLOCKSESSION(shp);
5872 		NFSUNLOCKSTATE();
5873 		return (NFSERR_BADSESSION);
5874 	}
5875 
5876 	/* Check to see if reclaim complete has already happened. */
5877 	if ((sep->sess_clp->lc_flags & LCL_RECLAIMCOMPLETE) != 0)
5878 		error = NFSERR_COMPLETEALREADY;
5879 	else
5880 		sep->sess_clp->lc_flags |= LCL_RECLAIMCOMPLETE;
5881 	NFSUNLOCKSESSION(shp);
5882 	NFSUNLOCKSTATE();
5883 	return (error);
5884 }
5885 
5886 /*
5887  * Cache the reply in a session slot.
5888  */
5889 void
5890 nfsrv_cache_session(uint8_t *sessionid, uint32_t slotid, int repstat,
5891    struct mbuf **m)
5892 {
5893 	struct nfsdsession *sep;
5894 	struct nfssessionhash *shp;
5895 
5896 	shp = NFSSESSIONHASH(sessionid);
5897 	NFSLOCKSESSION(shp);
5898 	sep = nfsrv_findsession(sessionid);
5899 	if (sep == NULL) {
5900 		NFSUNLOCKSESSION(shp);
5901 		printf("nfsrv_cache_session: no session\n");
5902 		m_freem(*m);
5903 		return;
5904 	}
5905 	nfsv4_seqsess_cacherep(slotid, sep->sess_slots, repstat, m);
5906 	NFSUNLOCKSESSION(shp);
5907 }
5908 
5909 /*
5910  * Search for a session that matches the sessionid.
5911  */
5912 static struct nfsdsession *
5913 nfsrv_findsession(uint8_t *sessionid)
5914 {
5915 	struct nfsdsession *sep;
5916 	struct nfssessionhash *shp;
5917 
5918 	shp = NFSSESSIONHASH(sessionid);
5919 	LIST_FOREACH(sep, &shp->list, sess_hash) {
5920 		if (!NFSBCMP(sessionid, sep->sess_sessionid, NFSX_V4SESSIONID))
5921 			break;
5922 	}
5923 	return (sep);
5924 }
5925 
5926 /*
5927  * Destroy a session.
5928  */
5929 int
5930 nfsrv_destroysession(struct nfsrv_descript *nd, uint8_t *sessionid)
5931 {
5932 	int error, samesess;
5933 
5934 	samesess = 0;
5935 	if (!NFSBCMP(sessionid, nd->nd_sessionid, NFSX_V4SESSIONID)) {
5936 		samesess = 1;
5937 		if ((nd->nd_flag & ND_LASTOP) == 0)
5938 			return (NFSERR_BADSESSION);
5939 	}
5940 	error = nfsrv_freesession(NULL, sessionid);
5941 	if (error == 0 && samesess != 0)
5942 		nd->nd_flag &= ~ND_HASSEQUENCE;
5943 	return (error);
5944 }
5945 
5946 /*
5947  * Free up a session structure.
5948  */
5949 static int
5950 nfsrv_freesession(struct nfsdsession *sep, uint8_t *sessionid)
5951 {
5952 	struct nfssessionhash *shp;
5953 	int i;
5954 
5955 	NFSLOCKSTATE();
5956 	if (sep == NULL) {
5957 		shp = NFSSESSIONHASH(sessionid);
5958 		NFSLOCKSESSION(shp);
5959 		sep = nfsrv_findsession(sessionid);
5960 	} else {
5961 		shp = NFSSESSIONHASH(sep->sess_sessionid);
5962 		NFSLOCKSESSION(shp);
5963 	}
5964 	if (sep != NULL) {
5965 		sep->sess_refcnt--;
5966 		if (sep->sess_refcnt > 0) {
5967 			NFSUNLOCKSESSION(shp);
5968 			NFSUNLOCKSTATE();
5969 			return (0);
5970 		}
5971 		LIST_REMOVE(sep, sess_hash);
5972 		LIST_REMOVE(sep, sess_list);
5973 	}
5974 	NFSUNLOCKSESSION(shp);
5975 	NFSUNLOCKSTATE();
5976 	if (sep == NULL)
5977 		return (NFSERR_BADSESSION);
5978 	for (i = 0; i < NFSV4_SLOTS; i++)
5979 		if (sep->sess_slots[i].nfssl_reply != NULL)
5980 			m_freem(sep->sess_slots[i].nfssl_reply);
5981 	if (sep->sess_cbsess.nfsess_xprt != NULL)
5982 		SVC_RELEASE(sep->sess_cbsess.nfsess_xprt);
5983 	free(sep, M_NFSDSESSION);
5984 	return (0);
5985 }
5986 
5987 /*
5988  * Free a stateid.
5989  * RFC5661 says that it should fail when there are associated opens, locks
5990  * or delegations. Since stateids represent opens, I don't see how you can
5991  * free an open stateid (it will be free'd when closed), so this function
5992  * only works for lock stateids (freeing the lock_owner) or delegations.
5993  */
5994 int
5995 nfsrv_freestateid(struct nfsrv_descript *nd, nfsv4stateid_t *stateidp,
5996     NFSPROC_T *p)
5997 {
5998 	struct nfsclient *clp;
5999 	struct nfsstate *stp;
6000 	int error;
6001 
6002 	NFSLOCKSTATE();
6003 	/*
6004 	 * Look up the stateid
6005 	 */
6006 	error = nfsrv_getclient((nfsquad_t)((u_quad_t)0), CLOPS_RENEW, &clp,
6007 	    NULL, (nfsquad_t)((u_quad_t)0), 0, nd, p);
6008 	if (error == 0) {
6009 		/* First, check for a delegation. */
6010 		LIST_FOREACH(stp, &clp->lc_deleg, ls_list) {
6011 			if (!NFSBCMP(stp->ls_stateid.other, stateidp->other,
6012 			    NFSX_STATEIDOTHER))
6013 				break;
6014 		}
6015 		if (stp != NULL) {
6016 			nfsrv_freedeleg(stp);
6017 			NFSUNLOCKSTATE();
6018 			return (error);
6019 		}
6020 	}
6021 	/* Not a delegation, try for a lock_owner. */
6022 	if (error == 0)
6023 		error = nfsrv_getstate(clp, stateidp, 0, &stp);
6024 	if (error == 0 && ((stp->ls_flags & (NFSLCK_OPEN | NFSLCK_DELEGREAD |
6025 	    NFSLCK_DELEGWRITE)) != 0 || (stp->ls_flags & NFSLCK_LOCK) == 0))
6026 		/* Not a lock_owner stateid. */
6027 		error = NFSERR_LOCKSHELD;
6028 	if (error == 0 && !LIST_EMPTY(&stp->ls_lock))
6029 		error = NFSERR_LOCKSHELD;
6030 	if (error == 0)
6031 		nfsrv_freelockowner(stp, NULL, 0, p);
6032 	NFSUNLOCKSTATE();
6033 	return (error);
6034 }
6035 
6036 /*
6037  * Generate the xdr for an NFSv4.1 CBSequence Operation.
6038  */
6039 static int
6040 nfsv4_setcbsequence(struct nfsrv_descript *nd, struct nfsclient *clp,
6041     int dont_replycache, struct nfsdsession **sepp)
6042 {
6043 	struct nfsdsession *sep;
6044 	uint32_t *tl, slotseq = 0;
6045 	int maxslot, slotpos;
6046 	uint8_t sessionid[NFSX_V4SESSIONID];
6047 	int error;
6048 
6049 	error = nfsv4_getcbsession(clp, sepp);
6050 	if (error != 0)
6051 		return (error);
6052 	sep = *sepp;
6053 	(void)nfsv4_sequencelookup(NULL, &sep->sess_cbsess, &slotpos, &maxslot,
6054 	    &slotseq, sessionid);
6055 	KASSERT(maxslot >= 0, ("nfsv4_setcbsequence neg maxslot"));
6056 
6057 	/* Build the Sequence arguments. */
6058 	NFSM_BUILD(tl, uint32_t *, NFSX_V4SESSIONID + 5 * NFSX_UNSIGNED);
6059 	bcopy(sessionid, tl, NFSX_V4SESSIONID);
6060 	tl += NFSX_V4SESSIONID / NFSX_UNSIGNED;
6061 	nd->nd_slotseq = tl;
6062 	*tl++ = txdr_unsigned(slotseq);
6063 	*tl++ = txdr_unsigned(slotpos);
6064 	*tl++ = txdr_unsigned(maxslot);
6065 	if (dont_replycache == 0)
6066 		*tl++ = newnfs_true;
6067 	else
6068 		*tl++ = newnfs_false;
6069 	*tl = 0;			/* No referring call list, for now. */
6070 	nd->nd_flag |= ND_HASSEQUENCE;
6071 	return (0);
6072 }
6073 
6074 /*
6075  * Get a session for the callback.
6076  */
6077 static int
6078 nfsv4_getcbsession(struct nfsclient *clp, struct nfsdsession **sepp)
6079 {
6080 	struct nfsdsession *sep;
6081 
6082 	NFSLOCKSTATE();
6083 	LIST_FOREACH(sep, &clp->lc_session, sess_list) {
6084 		if ((sep->sess_crflags & NFSV4CRSESS_CONNBACKCHAN) != 0)
6085 			break;
6086 	}
6087 	if (sep == NULL) {
6088 		NFSUNLOCKSTATE();
6089 		return (NFSERR_BADSESSION);
6090 	}
6091 	sep->sess_refcnt++;
6092 	*sepp = sep;
6093 	NFSUNLOCKSTATE();
6094 	return (0);
6095 }
6096 
6097 /*
6098  * Free up all backchannel xprts.  This needs to be done when the nfsd threads
6099  * exit, since those transports will all be going away.
6100  * This is only called after all the nfsd threads are done performing RPCs,
6101  * so locking shouldn't be an issue.
6102  */
6103 APPLESTATIC void
6104 nfsrv_freeallbackchannel_xprts(void)
6105 {
6106 	struct nfsdsession *sep;
6107 	struct nfsclient *clp;
6108 	SVCXPRT *xprt;
6109 	int i;
6110 
6111 	for (i = 0; i < nfsrv_clienthashsize; i++) {
6112 		LIST_FOREACH(clp, &nfsclienthash[i], lc_hash) {
6113 			LIST_FOREACH(sep, &clp->lc_session, sess_list) {
6114 				xprt = sep->sess_cbsess.nfsess_xprt;
6115 				sep->sess_cbsess.nfsess_xprt = NULL;
6116 				if (xprt != NULL)
6117 					SVC_RELEASE(xprt);
6118 			}
6119 		}
6120 	}
6121 }
6122 
6123