xref: /freebsd/sys/fs/nfsserver/nfs_nfsdstate.c (revision 52f72944b8f5abb2386eae924357dee8aea17d5b)
1 /*-
2  * SPDX-License-Identifier: BSD-2-Clause-FreeBSD
3  *
4  * Copyright (c) 2009 Rick Macklem, University of Guelph
5  * All rights reserved.
6  *
7  * Redistribution and use in source and binary forms, with or without
8  * modification, are permitted provided that the following conditions
9  * are met:
10  * 1. Redistributions of source code must retain the above copyright
11  *    notice, this list of conditions and the following disclaimer.
12  * 2. Redistributions in binary form must reproduce the above copyright
13  *    notice, this list of conditions and the following disclaimer in the
14  *    documentation and/or other materials provided with the distribution.
15  *
16  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
17  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
18  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
19  * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
20  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
21  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
22  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
23  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
24  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
25  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
26  * SUCH DAMAGE.
27  *
28  */
29 
30 #include <sys/cdefs.h>
31 __FBSDID("$FreeBSD$");
32 
33 #ifndef APPLEKEXT
34 #include <fs/nfs/nfsport.h>
35 
36 struct nfsrv_stablefirst nfsrv_stablefirst;
37 int nfsrv_issuedelegs = 0;
38 int nfsrv_dolocallocks = 0;
39 struct nfsv4lock nfsv4rootfs_lock;
40 
41 extern int newnfs_numnfsd;
42 extern struct nfsstatsv1 nfsstatsv1;
43 extern int nfsrv_lease;
44 extern struct timeval nfsboottime;
45 extern u_int32_t newnfs_true, newnfs_false;
46 NFSV4ROOTLOCKMUTEX;
47 NFSSTATESPINLOCK;
48 
49 SYSCTL_DECL(_vfs_nfsd);
50 int	nfsrv_statehashsize = NFSSTATEHASHSIZE;
51 SYSCTL_INT(_vfs_nfsd, OID_AUTO, statehashsize, CTLFLAG_RDTUN,
52     &nfsrv_statehashsize, 0,
53     "Size of state hash table set via loader.conf");
54 
55 int	nfsrv_clienthashsize = NFSCLIENTHASHSIZE;
56 SYSCTL_INT(_vfs_nfsd, OID_AUTO, clienthashsize, CTLFLAG_RDTUN,
57     &nfsrv_clienthashsize, 0,
58     "Size of client hash table set via loader.conf");
59 
60 int	nfsrv_lockhashsize = NFSLOCKHASHSIZE;
61 SYSCTL_INT(_vfs_nfsd, OID_AUTO, fhhashsize, CTLFLAG_RDTUN,
62     &nfsrv_lockhashsize, 0,
63     "Size of file handle hash table set via loader.conf");
64 
65 int	nfsrv_sessionhashsize = NFSSESSIONHASHSIZE;
66 SYSCTL_INT(_vfs_nfsd, OID_AUTO, sessionhashsize, CTLFLAG_RDTUN,
67     &nfsrv_sessionhashsize, 0,
68     "Size of session hash table set via loader.conf");
69 
70 static int	nfsrv_v4statelimit = NFSRV_V4STATELIMIT;
71 SYSCTL_INT(_vfs_nfsd, OID_AUTO, v4statelimit, CTLFLAG_RWTUN,
72     &nfsrv_v4statelimit, 0,
73     "High water limit for NFSv4 opens+locks+delegations");
74 
75 static int	nfsrv_writedelegifpos = 0;
76 SYSCTL_INT(_vfs_nfsd, OID_AUTO, writedelegifpos, CTLFLAG_RW,
77     &nfsrv_writedelegifpos, 0,
78     "Issue a write delegation for read opens if possible");
79 
80 static int	nfsrv_allowreadforwriteopen = 1;
81 SYSCTL_INT(_vfs_nfsd, OID_AUTO, allowreadforwriteopen, CTLFLAG_RW,
82     &nfsrv_allowreadforwriteopen, 0,
83     "Allow Reads to be done with Write Access StateIDs");
84 
85 /*
86  * Hash lists for nfs V4.
87  */
88 struct nfsclienthashhead	*nfsclienthash;
89 struct nfslockhashhead		*nfslockhash;
90 struct nfssessionhash		*nfssessionhash;
91 #endif	/* !APPLEKEXT */
92 
93 static u_int32_t nfsrv_openpluslock = 0, nfsrv_delegatecnt = 0;
94 static time_t nfsrvboottime;
95 static int nfsrv_returnoldstateid = 0, nfsrv_clients = 0;
96 static int nfsrv_clienthighwater = NFSRV_CLIENTHIGHWATER;
97 static int nfsrv_nogsscallback = 0;
98 static volatile int nfsrv_writedelegcnt = 0;
99 
100 /* local functions */
101 static void nfsrv_dumpaclient(struct nfsclient *clp,
102     struct nfsd_dumpclients *dumpp);
103 static void nfsrv_freeopenowner(struct nfsstate *stp, int cansleep,
104     NFSPROC_T *p);
105 static int nfsrv_freeopen(struct nfsstate *stp, vnode_t vp, int cansleep,
106     NFSPROC_T *p);
107 static void nfsrv_freelockowner(struct nfsstate *stp, vnode_t vp, int cansleep,
108     NFSPROC_T *p);
109 static void nfsrv_freeallnfslocks(struct nfsstate *stp, vnode_t vp,
110     int cansleep, NFSPROC_T *p);
111 static void nfsrv_freenfslock(struct nfslock *lop);
112 static void nfsrv_freenfslockfile(struct nfslockfile *lfp);
113 static void nfsrv_freedeleg(struct nfsstate *);
114 static int nfsrv_getstate(struct nfsclient *clp, nfsv4stateid_t *stateidp,
115     u_int32_t flags, struct nfsstate **stpp);
116 static void nfsrv_getowner(struct nfsstatehead *hp, struct nfsstate *new_stp,
117     struct nfsstate **stpp);
118 static int nfsrv_getlockfh(vnode_t vp, u_short flags,
119     struct nfslockfile *new_lfp, fhandle_t *nfhp, NFSPROC_T *p);
120 static int nfsrv_getlockfile(u_short flags, struct nfslockfile **new_lfpp,
121     struct nfslockfile **lfpp, fhandle_t *nfhp, int lockit);
122 static void nfsrv_insertlock(struct nfslock *new_lop,
123     struct nfslock *insert_lop, struct nfsstate *stp, struct nfslockfile *lfp);
124 static void nfsrv_updatelock(struct nfsstate *stp, struct nfslock **new_lopp,
125     struct nfslock **other_lopp, struct nfslockfile *lfp);
126 static int nfsrv_getipnumber(u_char *cp);
127 static int nfsrv_checkrestart(nfsquad_t clientid, u_int32_t flags,
128     nfsv4stateid_t *stateidp, int specialid);
129 static int nfsrv_checkgrace(struct nfsrv_descript *nd, struct nfsclient *clp,
130     u_int32_t flags);
131 static int nfsrv_docallback(struct nfsclient *clp, int procnum,
132     nfsv4stateid_t *stateidp, int trunc, fhandle_t *fhp,
133     struct nfsvattr *nap, nfsattrbit_t *attrbitp, NFSPROC_T *p);
134 static int nfsrv_cbcallargs(struct nfsrv_descript *nd, struct nfsclient *clp,
135     uint32_t callback, int op, const char *optag, struct nfsdsession **sepp);
136 static u_int32_t nfsrv_nextclientindex(void);
137 static u_int32_t nfsrv_nextstateindex(struct nfsclient *clp);
138 static void nfsrv_markstable(struct nfsclient *clp);
139 static int nfsrv_checkstable(struct nfsclient *clp);
140 static int nfsrv_clientconflict(struct nfsclient *clp, int *haslockp, struct
141     vnode *vp, NFSPROC_T *p);
142 static int nfsrv_delegconflict(struct nfsstate *stp, int *haslockp,
143     NFSPROC_T *p, vnode_t vp);
144 static int nfsrv_cleandeleg(vnode_t vp, struct nfslockfile *lfp,
145     struct nfsclient *clp, int *haslockp, NFSPROC_T *p);
146 static int nfsrv_notsamecredname(struct nfsrv_descript *nd,
147     struct nfsclient *clp);
148 static time_t nfsrv_leaseexpiry(void);
149 static void nfsrv_delaydelegtimeout(struct nfsstate *stp);
150 static int nfsrv_checkseqid(struct nfsrv_descript *nd, u_int32_t seqid,
151     struct nfsstate *stp, struct nfsrvcache *op);
152 static int nfsrv_nootherstate(struct nfsstate *stp);
153 static int nfsrv_locallock(vnode_t vp, struct nfslockfile *lfp, int flags,
154     uint64_t first, uint64_t end, struct nfslockconflict *cfp, NFSPROC_T *p);
155 static void nfsrv_localunlock(vnode_t vp, struct nfslockfile *lfp,
156     uint64_t init_first, uint64_t init_end, NFSPROC_T *p);
157 static int nfsrv_dolocal(vnode_t vp, struct nfslockfile *lfp, int flags,
158     int oldflags, uint64_t first, uint64_t end, struct nfslockconflict *cfp,
159     NFSPROC_T *p);
160 static void nfsrv_locallock_rollback(vnode_t vp, struct nfslockfile *lfp,
161     NFSPROC_T *p);
162 static void nfsrv_locallock_commit(struct nfslockfile *lfp, int flags,
163     uint64_t first, uint64_t end);
164 static void nfsrv_locklf(struct nfslockfile *lfp);
165 static void nfsrv_unlocklf(struct nfslockfile *lfp);
166 static struct nfsdsession *nfsrv_findsession(uint8_t *sessionid);
167 static int nfsrv_freesession(struct nfsdsession *sep, uint8_t *sessionid);
168 static int nfsv4_setcbsequence(struct nfsrv_descript *nd, struct nfsclient *clp,
169     int dont_replycache, struct nfsdsession **sepp);
170 static int nfsv4_getcbsession(struct nfsclient *clp, struct nfsdsession **sepp);
171 
172 /*
173  * Scan the client list for a match and either return the current one,
174  * create a new entry or return an error.
175  * If returning a non-error, the clp structure must either be linked into
176  * the client list or free'd.
177  */
178 APPLESTATIC int
179 nfsrv_setclient(struct nfsrv_descript *nd, struct nfsclient **new_clpp,
180     nfsquad_t *clientidp, nfsquad_t *confirmp, NFSPROC_T *p)
181 {
182 	struct nfsclient *clp = NULL, *new_clp = *new_clpp;
183 	int i, error = 0;
184 	struct nfsstate *stp, *tstp;
185 	struct sockaddr_in *sad, *rad;
186 	int zapit = 0, gotit, hasstate = 0, igotlock;
187 	static u_int64_t confirm_index = 0;
188 
189 	/*
190 	 * Check for state resource limit exceeded.
191 	 */
192 	if (nfsrv_openpluslock > nfsrv_v4statelimit) {
193 		error = NFSERR_RESOURCE;
194 		goto out;
195 	}
196 
197 	if (nfsrv_issuedelegs == 0 ||
198 	    ((nd->nd_flag & ND_GSS) != 0 && nfsrv_nogsscallback != 0))
199 		/*
200 		 * Don't do callbacks when delegations are disabled or
201 		 * for AUTH_GSS unless enabled via nfsrv_nogsscallback.
202 		 * If establishing a callback connection is attempted
203 		 * when a firewall is blocking the callback path, the
204 		 * server may wait too long for the connect attempt to
205 		 * succeed during the Open. Some clients, such as Linux,
206 		 * may timeout and give up on the Open before the server
207 		 * replies. Also, since AUTH_GSS callbacks are not
208 		 * yet interoperability tested, they might cause the
209 		 * server to crap out, if they get past the Init call to
210 		 * the client.
211 		 */
212 		new_clp->lc_program = 0;
213 
214 	/* Lock out other nfsd threads */
215 	NFSLOCKV4ROOTMUTEX();
216 	nfsv4_relref(&nfsv4rootfs_lock);
217 	do {
218 		igotlock = nfsv4_lock(&nfsv4rootfs_lock, 1, NULL,
219 		    NFSV4ROOTLOCKMUTEXPTR, NULL);
220 	} while (!igotlock);
221 	NFSUNLOCKV4ROOTMUTEX();
222 
223 	/*
224 	 * Search for a match in the client list.
225 	 */
226 	gotit = i = 0;
227 	while (i < nfsrv_clienthashsize && !gotit) {
228 	    LIST_FOREACH(clp, &nfsclienthash[i], lc_hash) {
229 		if (new_clp->lc_idlen == clp->lc_idlen &&
230 		    !NFSBCMP(new_clp->lc_id, clp->lc_id, clp->lc_idlen)) {
231 			gotit = 1;
232 			break;
233 		}
234 	    }
235 	    if (gotit == 0)
236 		i++;
237 	}
238 	if (!gotit ||
239 	    (clp->lc_flags & (LCL_NEEDSCONFIRM | LCL_ADMINREVOKED))) {
240 		if ((nd->nd_flag & ND_NFSV41) != 0 && confirmp->lval[1] != 0) {
241 			/*
242 			 * For NFSv4.1, if confirmp->lval[1] is non-zero, the
243 			 * client is trying to update a confirmed clientid.
244 			 */
245 			NFSLOCKV4ROOTMUTEX();
246 			nfsv4_unlock(&nfsv4rootfs_lock, 1);
247 			NFSUNLOCKV4ROOTMUTEX();
248 			confirmp->lval[1] = 0;
249 			error = NFSERR_NOENT;
250 			goto out;
251 		}
252 		/*
253 		 * Get rid of the old one.
254 		 */
255 		if (i != nfsrv_clienthashsize) {
256 			LIST_REMOVE(clp, lc_hash);
257 			nfsrv_cleanclient(clp, p);
258 			nfsrv_freedeleglist(&clp->lc_deleg);
259 			nfsrv_freedeleglist(&clp->lc_olddeleg);
260 			zapit = 1;
261 		}
262 		/*
263 		 * Add it after assigning a client id to it.
264 		 */
265 		new_clp->lc_flags |= LCL_NEEDSCONFIRM;
266 		if ((nd->nd_flag & ND_NFSV41) != 0)
267 			new_clp->lc_confirm.lval[0] = confirmp->lval[0] =
268 			    ++confirm_index;
269 		else
270 			confirmp->qval = new_clp->lc_confirm.qval =
271 			    ++confirm_index;
272 		clientidp->lval[0] = new_clp->lc_clientid.lval[0] =
273 		    (u_int32_t)nfsrvboottime;
274 		clientidp->lval[1] = new_clp->lc_clientid.lval[1] =
275 		    nfsrv_nextclientindex();
276 		new_clp->lc_stateindex = 0;
277 		new_clp->lc_statemaxindex = 0;
278 		new_clp->lc_cbref = 0;
279 		new_clp->lc_expiry = nfsrv_leaseexpiry();
280 		LIST_INIT(&new_clp->lc_open);
281 		LIST_INIT(&new_clp->lc_deleg);
282 		LIST_INIT(&new_clp->lc_olddeleg);
283 		LIST_INIT(&new_clp->lc_session);
284 		for (i = 0; i < nfsrv_statehashsize; i++)
285 			LIST_INIT(&new_clp->lc_stateid[i]);
286 		LIST_INSERT_HEAD(NFSCLIENTHASH(new_clp->lc_clientid), new_clp,
287 		    lc_hash);
288 		nfsstatsv1.srvclients++;
289 		nfsrv_openpluslock++;
290 		nfsrv_clients++;
291 		NFSLOCKV4ROOTMUTEX();
292 		nfsv4_unlock(&nfsv4rootfs_lock, 1);
293 		NFSUNLOCKV4ROOTMUTEX();
294 		if (zapit)
295 			nfsrv_zapclient(clp, p);
296 		*new_clpp = NULL;
297 		goto out;
298 	}
299 
300 	/*
301 	 * Now, handle the cases where the id is already issued.
302 	 */
303 	if (nfsrv_notsamecredname(nd, clp)) {
304 	    /*
305 	     * Check to see if there is expired state that should go away.
306 	     */
307 	    if (clp->lc_expiry < NFSD_MONOSEC &&
308 	        (!LIST_EMPTY(&clp->lc_open) || !LIST_EMPTY(&clp->lc_deleg))) {
309 		nfsrv_cleanclient(clp, p);
310 		nfsrv_freedeleglist(&clp->lc_deleg);
311 	    }
312 
313 	    /*
314 	     * If there is outstanding state, then reply NFSERR_CLIDINUSE per
315 	     * RFC3530 Sec. 8.1.2 last para.
316 	     */
317 	    if (!LIST_EMPTY(&clp->lc_deleg)) {
318 		hasstate = 1;
319 	    } else if (LIST_EMPTY(&clp->lc_open)) {
320 		hasstate = 0;
321 	    } else {
322 		hasstate = 0;
323 		/* Look for an Open on the OpenOwner */
324 		LIST_FOREACH(stp, &clp->lc_open, ls_list) {
325 		    if (!LIST_EMPTY(&stp->ls_open)) {
326 			hasstate = 1;
327 			break;
328 		    }
329 		}
330 	    }
331 	    if (hasstate) {
332 		/*
333 		 * If the uid doesn't match, return NFSERR_CLIDINUSE after
334 		 * filling out the correct ipaddr and portnum.
335 		 */
336 		sad = NFSSOCKADDR(new_clp->lc_req.nr_nam, struct sockaddr_in *);
337 		rad = NFSSOCKADDR(clp->lc_req.nr_nam, struct sockaddr_in *);
338 		sad->sin_addr.s_addr = rad->sin_addr.s_addr;
339 		sad->sin_port = rad->sin_port;
340 		NFSLOCKV4ROOTMUTEX();
341 		nfsv4_unlock(&nfsv4rootfs_lock, 1);
342 		NFSUNLOCKV4ROOTMUTEX();
343 		error = NFSERR_CLIDINUSE;
344 		goto out;
345 	    }
346 	}
347 
348 	if (NFSBCMP(new_clp->lc_verf, clp->lc_verf, NFSX_VERF)) {
349 		/*
350 		 * If the verifier has changed, the client has rebooted
351 		 * and a new client id is issued. The old state info
352 		 * can be thrown away once the SETCLIENTID_CONFIRM occurs.
353 		 */
354 		LIST_REMOVE(clp, lc_hash);
355 		new_clp->lc_flags |= LCL_NEEDSCONFIRM;
356 		if ((nd->nd_flag & ND_NFSV41) != 0)
357 			new_clp->lc_confirm.lval[0] = confirmp->lval[0] =
358 			    ++confirm_index;
359 		else
360 			confirmp->qval = new_clp->lc_confirm.qval =
361 			    ++confirm_index;
362 		clientidp->lval[0] = new_clp->lc_clientid.lval[0] =
363 		    nfsrvboottime;
364 		clientidp->lval[1] = new_clp->lc_clientid.lval[1] =
365 		    nfsrv_nextclientindex();
366 		new_clp->lc_stateindex = 0;
367 		new_clp->lc_statemaxindex = 0;
368 		new_clp->lc_cbref = 0;
369 		new_clp->lc_expiry = nfsrv_leaseexpiry();
370 
371 		/*
372 		 * Save the state until confirmed.
373 		 */
374 		LIST_NEWHEAD(&new_clp->lc_open, &clp->lc_open, ls_list);
375 		LIST_FOREACH(tstp, &new_clp->lc_open, ls_list)
376 			tstp->ls_clp = new_clp;
377 		LIST_NEWHEAD(&new_clp->lc_deleg, &clp->lc_deleg, ls_list);
378 		LIST_FOREACH(tstp, &new_clp->lc_deleg, ls_list)
379 			tstp->ls_clp = new_clp;
380 		LIST_NEWHEAD(&new_clp->lc_olddeleg, &clp->lc_olddeleg,
381 		    ls_list);
382 		LIST_FOREACH(tstp, &new_clp->lc_olddeleg, ls_list)
383 			tstp->ls_clp = new_clp;
384 		for (i = 0; i < nfsrv_statehashsize; i++) {
385 			LIST_NEWHEAD(&new_clp->lc_stateid[i],
386 			    &clp->lc_stateid[i], ls_hash);
387 			LIST_FOREACH(tstp, &new_clp->lc_stateid[i], ls_hash)
388 				tstp->ls_clp = new_clp;
389 		}
390 		LIST_INSERT_HEAD(NFSCLIENTHASH(new_clp->lc_clientid), new_clp,
391 		    lc_hash);
392 		nfsstatsv1.srvclients++;
393 		nfsrv_openpluslock++;
394 		nfsrv_clients++;
395 		NFSLOCKV4ROOTMUTEX();
396 		nfsv4_unlock(&nfsv4rootfs_lock, 1);
397 		NFSUNLOCKV4ROOTMUTEX();
398 
399 		/*
400 		 * Must wait until any outstanding callback on the old clp
401 		 * completes.
402 		 */
403 		NFSLOCKSTATE();
404 		while (clp->lc_cbref) {
405 			clp->lc_flags |= LCL_WAKEUPWANTED;
406 			(void)mtx_sleep(clp, NFSSTATEMUTEXPTR, PZERO - 1,
407 			    "nfsd clp", 10 * hz);
408 		}
409 		NFSUNLOCKSTATE();
410 		nfsrv_zapclient(clp, p);
411 		*new_clpp = NULL;
412 		goto out;
413 	}
414 
415 	/* For NFSv4.1, mark that we found a confirmed clientid. */
416 	if ((nd->nd_flag & ND_NFSV41) != 0) {
417 		clientidp->lval[0] = clp->lc_clientid.lval[0];
418 		clientidp->lval[1] = clp->lc_clientid.lval[1];
419 		confirmp->lval[0] = 0;	/* Ignored by client */
420 		confirmp->lval[1] = 1;
421 	} else {
422 		/*
423 		 * id and verifier match, so update the net address info
424 		 * and get rid of any existing callback authentication
425 		 * handle, so a new one will be acquired.
426 		 */
427 		LIST_REMOVE(clp, lc_hash);
428 		new_clp->lc_flags |= (LCL_NEEDSCONFIRM | LCL_DONTCLEAN);
429 		new_clp->lc_expiry = nfsrv_leaseexpiry();
430 		confirmp->qval = new_clp->lc_confirm.qval = ++confirm_index;
431 		clientidp->lval[0] = new_clp->lc_clientid.lval[0] =
432 		    clp->lc_clientid.lval[0];
433 		clientidp->lval[1] = new_clp->lc_clientid.lval[1] =
434 		    clp->lc_clientid.lval[1];
435 		new_clp->lc_delegtime = clp->lc_delegtime;
436 		new_clp->lc_stateindex = clp->lc_stateindex;
437 		new_clp->lc_statemaxindex = clp->lc_statemaxindex;
438 		new_clp->lc_cbref = 0;
439 		LIST_NEWHEAD(&new_clp->lc_open, &clp->lc_open, ls_list);
440 		LIST_FOREACH(tstp, &new_clp->lc_open, ls_list)
441 			tstp->ls_clp = new_clp;
442 		LIST_NEWHEAD(&new_clp->lc_deleg, &clp->lc_deleg, ls_list);
443 		LIST_FOREACH(tstp, &new_clp->lc_deleg, ls_list)
444 			tstp->ls_clp = new_clp;
445 		LIST_NEWHEAD(&new_clp->lc_olddeleg, &clp->lc_olddeleg, ls_list);
446 		LIST_FOREACH(tstp, &new_clp->lc_olddeleg, ls_list)
447 			tstp->ls_clp = new_clp;
448 		for (i = 0; i < nfsrv_statehashsize; i++) {
449 			LIST_NEWHEAD(&new_clp->lc_stateid[i],
450 			    &clp->lc_stateid[i], ls_hash);
451 			LIST_FOREACH(tstp, &new_clp->lc_stateid[i], ls_hash)
452 				tstp->ls_clp = new_clp;
453 		}
454 		LIST_INSERT_HEAD(NFSCLIENTHASH(new_clp->lc_clientid), new_clp,
455 		    lc_hash);
456 		nfsstatsv1.srvclients++;
457 		nfsrv_openpluslock++;
458 		nfsrv_clients++;
459 	}
460 	NFSLOCKV4ROOTMUTEX();
461 	nfsv4_unlock(&nfsv4rootfs_lock, 1);
462 	NFSUNLOCKV4ROOTMUTEX();
463 
464 	if ((nd->nd_flag & ND_NFSV41) == 0) {
465 		/*
466 		 * Must wait until any outstanding callback on the old clp
467 		 * completes.
468 		 */
469 		NFSLOCKSTATE();
470 		while (clp->lc_cbref) {
471 			clp->lc_flags |= LCL_WAKEUPWANTED;
472 			(void)mtx_sleep(clp, NFSSTATEMUTEXPTR, PZERO - 1,
473 			    "nfsdclp", 10 * hz);
474 		}
475 		NFSUNLOCKSTATE();
476 		nfsrv_zapclient(clp, p);
477 		*new_clpp = NULL;
478 	}
479 
480 out:
481 	NFSEXITCODE2(error, nd);
482 	return (error);
483 }
484 
485 /*
486  * Check to see if the client id exists and optionally confirm it.
487  */
488 APPLESTATIC int
489 nfsrv_getclient(nfsquad_t clientid, int opflags, struct nfsclient **clpp,
490     struct nfsdsession *nsep, nfsquad_t confirm, uint32_t cbprogram,
491     struct nfsrv_descript *nd, NFSPROC_T *p)
492 {
493 	struct nfsclient *clp;
494 	struct nfsstate *stp;
495 	int i;
496 	struct nfsclienthashhead *hp;
497 	int error = 0, igotlock, doneok;
498 	struct nfssessionhash *shp;
499 	struct nfsdsession *sep;
500 	uint64_t sessid[2];
501 	static uint64_t next_sess = 0;
502 
503 	if (clpp)
504 		*clpp = NULL;
505 	if ((nd == NULL || (nd->nd_flag & ND_NFSV41) == 0 ||
506 	    opflags != CLOPS_RENEW) && nfsrvboottime != clientid.lval[0]) {
507 		error = NFSERR_STALECLIENTID;
508 		goto out;
509 	}
510 
511 	/*
512 	 * If called with opflags == CLOPS_RENEW, the State Lock is
513 	 * already held. Otherwise, we need to get either that or,
514 	 * for the case of Confirm, lock out the nfsd threads.
515 	 */
516 	if (opflags & CLOPS_CONFIRM) {
517 		NFSLOCKV4ROOTMUTEX();
518 		nfsv4_relref(&nfsv4rootfs_lock);
519 		do {
520 			igotlock = nfsv4_lock(&nfsv4rootfs_lock, 1, NULL,
521 			    NFSV4ROOTLOCKMUTEXPTR, NULL);
522 		} while (!igotlock);
523 		/*
524 		 * Create a new sessionid here, since we need to do it where
525 		 * there is a mutex held to serialize update of next_sess.
526 		 */
527 		if ((nd->nd_flag & ND_NFSV41) != 0) {
528 			sessid[0] = ++next_sess;
529 			sessid[1] = clientid.qval;
530 		}
531 		NFSUNLOCKV4ROOTMUTEX();
532 	} else if (opflags != CLOPS_RENEW) {
533 		NFSLOCKSTATE();
534 	}
535 
536 	/* For NFSv4.1, the clp is acquired from the associated session. */
537 	if (nd != NULL && (nd->nd_flag & ND_NFSV41) != 0 &&
538 	    opflags == CLOPS_RENEW) {
539 		clp = NULL;
540 		if ((nd->nd_flag & ND_HASSEQUENCE) != 0) {
541 			shp = NFSSESSIONHASH(nd->nd_sessionid);
542 			NFSLOCKSESSION(shp);
543 			sep = nfsrv_findsession(nd->nd_sessionid);
544 			if (sep != NULL)
545 				clp = sep->sess_clp;
546 			NFSUNLOCKSESSION(shp);
547 		}
548 	} else {
549 		hp = NFSCLIENTHASH(clientid);
550 		LIST_FOREACH(clp, hp, lc_hash) {
551 			if (clp->lc_clientid.lval[1] == clientid.lval[1])
552 				break;
553 		}
554 	}
555 	if (clp == NULL) {
556 		if (opflags & CLOPS_CONFIRM)
557 			error = NFSERR_STALECLIENTID;
558 		else
559 			error = NFSERR_EXPIRED;
560 	} else if (clp->lc_flags & LCL_ADMINREVOKED) {
561 		/*
562 		 * If marked admin revoked, just return the error.
563 		 */
564 		error = NFSERR_ADMINREVOKED;
565 	}
566 	if (error) {
567 		if (opflags & CLOPS_CONFIRM) {
568 			NFSLOCKV4ROOTMUTEX();
569 			nfsv4_unlock(&nfsv4rootfs_lock, 1);
570 			NFSUNLOCKV4ROOTMUTEX();
571 		} else if (opflags != CLOPS_RENEW) {
572 			NFSUNLOCKSTATE();
573 		}
574 		goto out;
575 	}
576 
577 	/*
578 	 * Perform any operations specified by the opflags.
579 	 */
580 	if (opflags & CLOPS_CONFIRM) {
581 		if (((nd->nd_flag & ND_NFSV41) != 0 &&
582 		     clp->lc_confirm.lval[0] != confirm.lval[0]) ||
583 		    ((nd->nd_flag & ND_NFSV41) == 0 &&
584 		     clp->lc_confirm.qval != confirm.qval))
585 			error = NFSERR_STALECLIENTID;
586 		else if (nfsrv_notsamecredname(nd, clp))
587 			error = NFSERR_CLIDINUSE;
588 
589 		if (!error) {
590 		    if ((clp->lc_flags & (LCL_NEEDSCONFIRM | LCL_DONTCLEAN)) ==
591 			LCL_NEEDSCONFIRM) {
592 			/*
593 			 * Hang onto the delegations (as old delegations)
594 			 * for an Open with CLAIM_DELEGATE_PREV unless in
595 			 * grace, but get rid of the rest of the state.
596 			 */
597 			nfsrv_cleanclient(clp, p);
598 			nfsrv_freedeleglist(&clp->lc_olddeleg);
599 			if (nfsrv_checkgrace(nd, clp, 0)) {
600 			    /* In grace, so just delete delegations */
601 			    nfsrv_freedeleglist(&clp->lc_deleg);
602 			} else {
603 			    LIST_FOREACH(stp, &clp->lc_deleg, ls_list)
604 				stp->ls_flags |= NFSLCK_OLDDELEG;
605 			    clp->lc_delegtime = NFSD_MONOSEC +
606 				nfsrv_lease + NFSRV_LEASEDELTA;
607 			    LIST_NEWHEAD(&clp->lc_olddeleg, &clp->lc_deleg,
608 				ls_list);
609 			}
610 			if ((nd->nd_flag & ND_NFSV41) != 0)
611 			    clp->lc_program = cbprogram;
612 		    }
613 		    clp->lc_flags &= ~(LCL_NEEDSCONFIRM | LCL_DONTCLEAN);
614 		    if (clp->lc_program)
615 			clp->lc_flags |= LCL_NEEDSCBNULL;
616 		    /* For NFSv4.1, link the session onto the client. */
617 		    if (nsep != NULL) {
618 			/* Hold a reference on the xprt for a backchannel. */
619 			if ((nsep->sess_crflags & NFSV4CRSESS_CONNBACKCHAN)
620 			    != 0 && clp->lc_req.nr_client == NULL) {
621 			    clp->lc_req.nr_client = (struct __rpc_client *)
622 				clnt_bck_create(nd->nd_xprt->xp_socket,
623 				cbprogram, NFSV4_CBVERS);
624 			    if (clp->lc_req.nr_client != NULL) {
625 				SVC_ACQUIRE(nd->nd_xprt);
626 				nd->nd_xprt->xp_p2 =
627 				    clp->lc_req.nr_client->cl_private;
628 				/* Disable idle timeout. */
629 				nd->nd_xprt->xp_idletimeout = 0;
630 				nsep->sess_cbsess.nfsess_xprt = nd->nd_xprt;
631 			    } else
632 				nsep->sess_crflags &= ~NFSV4CRSESS_CONNBACKCHAN;
633 			}
634 			NFSBCOPY(sessid, nsep->sess_sessionid,
635 			    NFSX_V4SESSIONID);
636 			NFSBCOPY(sessid, nsep->sess_cbsess.nfsess_sessionid,
637 			    NFSX_V4SESSIONID);
638 			shp = NFSSESSIONHASH(nsep->sess_sessionid);
639 			NFSLOCKSTATE();
640 			NFSLOCKSESSION(shp);
641 			LIST_INSERT_HEAD(&shp->list, nsep, sess_hash);
642 			LIST_INSERT_HEAD(&clp->lc_session, nsep, sess_list);
643 			nsep->sess_clp = clp;
644 			NFSUNLOCKSESSION(shp);
645 			NFSUNLOCKSTATE();
646 		    }
647 		}
648 	} else if (clp->lc_flags & LCL_NEEDSCONFIRM) {
649 		error = NFSERR_EXPIRED;
650 	}
651 
652 	/*
653 	 * If called by the Renew Op, we must check the principal.
654 	 */
655 	if (!error && (opflags & CLOPS_RENEWOP)) {
656 	    if (nfsrv_notsamecredname(nd, clp)) {
657 		doneok = 0;
658 		for (i = 0; i < nfsrv_statehashsize && doneok == 0; i++) {
659 		    LIST_FOREACH(stp, &clp->lc_stateid[i], ls_hash) {
660 			if ((stp->ls_flags & NFSLCK_OPEN) &&
661 			    stp->ls_uid == nd->nd_cred->cr_uid) {
662 				doneok = 1;
663 				break;
664 			}
665 		    }
666 		}
667 		if (!doneok)
668 			error = NFSERR_ACCES;
669 	    }
670 	    if (!error && (clp->lc_flags & LCL_CBDOWN))
671 		error = NFSERR_CBPATHDOWN;
672 	}
673 	if ((!error || error == NFSERR_CBPATHDOWN) &&
674 	     (opflags & CLOPS_RENEW)) {
675 		clp->lc_expiry = nfsrv_leaseexpiry();
676 	}
677 	if (opflags & CLOPS_CONFIRM) {
678 		NFSLOCKV4ROOTMUTEX();
679 		nfsv4_unlock(&nfsv4rootfs_lock, 1);
680 		NFSUNLOCKV4ROOTMUTEX();
681 	} else if (opflags != CLOPS_RENEW) {
682 		NFSUNLOCKSTATE();
683 	}
684 	if (clpp)
685 		*clpp = clp;
686 
687 out:
688 	NFSEXITCODE2(error, nd);
689 	return (error);
690 }
691 
692 /*
693  * Perform the NFSv4.1 destroy clientid.
694  */
695 int
696 nfsrv_destroyclient(nfsquad_t clientid, NFSPROC_T *p)
697 {
698 	struct nfsclient *clp;
699 	struct nfsclienthashhead *hp;
700 	int error = 0, i, igotlock;
701 
702 	if (nfsrvboottime != clientid.lval[0]) {
703 		error = NFSERR_STALECLIENTID;
704 		goto out;
705 	}
706 
707 	/* Lock out other nfsd threads */
708 	NFSLOCKV4ROOTMUTEX();
709 	nfsv4_relref(&nfsv4rootfs_lock);
710 	do {
711 		igotlock = nfsv4_lock(&nfsv4rootfs_lock, 1, NULL,
712 		    NFSV4ROOTLOCKMUTEXPTR, NULL);
713 	} while (igotlock == 0);
714 	NFSUNLOCKV4ROOTMUTEX();
715 
716 	hp = NFSCLIENTHASH(clientid);
717 	LIST_FOREACH(clp, hp, lc_hash) {
718 		if (clp->lc_clientid.lval[1] == clientid.lval[1])
719 			break;
720 	}
721 	if (clp == NULL) {
722 		NFSLOCKV4ROOTMUTEX();
723 		nfsv4_unlock(&nfsv4rootfs_lock, 1);
724 		NFSUNLOCKV4ROOTMUTEX();
725 		/* Just return ok, since it is gone. */
726 		goto out;
727 	}
728 
729 	/* Scan for state on the clientid. */
730 	for (i = 0; i < nfsrv_statehashsize; i++)
731 		if (!LIST_EMPTY(&clp->lc_stateid[i])) {
732 			NFSLOCKV4ROOTMUTEX();
733 			nfsv4_unlock(&nfsv4rootfs_lock, 1);
734 			NFSUNLOCKV4ROOTMUTEX();
735 			error = NFSERR_CLIENTIDBUSY;
736 			goto out;
737 		}
738 	if (!LIST_EMPTY(&clp->lc_session) || !LIST_EMPTY(&clp->lc_deleg)) {
739 		NFSLOCKV4ROOTMUTEX();
740 		nfsv4_unlock(&nfsv4rootfs_lock, 1);
741 		NFSUNLOCKV4ROOTMUTEX();
742 		error = NFSERR_CLIENTIDBUSY;
743 		goto out;
744 	}
745 
746 	/* Destroy the clientid and return ok. */
747 	nfsrv_cleanclient(clp, p);
748 	nfsrv_freedeleglist(&clp->lc_deleg);
749 	nfsrv_freedeleglist(&clp->lc_olddeleg);
750 	LIST_REMOVE(clp, lc_hash);
751 	NFSLOCKV4ROOTMUTEX();
752 	nfsv4_unlock(&nfsv4rootfs_lock, 1);
753 	NFSUNLOCKV4ROOTMUTEX();
754 	nfsrv_zapclient(clp, p);
755 out:
756 	NFSEXITCODE2(error, nd);
757 	return (error);
758 }
759 
760 /*
761  * Called from the new nfssvc syscall to admin revoke a clientid.
762  * Returns 0 for success, error otherwise.
763  */
764 APPLESTATIC int
765 nfsrv_adminrevoke(struct nfsd_clid *revokep, NFSPROC_T *p)
766 {
767 	struct nfsclient *clp = NULL;
768 	int i, error = 0;
769 	int gotit, igotlock;
770 
771 	/*
772 	 * First, lock out the nfsd so that state won't change while the
773 	 * revocation record is being written to the stable storage restart
774 	 * file.
775 	 */
776 	NFSLOCKV4ROOTMUTEX();
777 	do {
778 		igotlock = nfsv4_lock(&nfsv4rootfs_lock, 1, NULL,
779 		    NFSV4ROOTLOCKMUTEXPTR, NULL);
780 	} while (!igotlock);
781 	NFSUNLOCKV4ROOTMUTEX();
782 
783 	/*
784 	 * Search for a match in the client list.
785 	 */
786 	gotit = i = 0;
787 	while (i < nfsrv_clienthashsize && !gotit) {
788 	    LIST_FOREACH(clp, &nfsclienthash[i], lc_hash) {
789 		if (revokep->nclid_idlen == clp->lc_idlen &&
790 		    !NFSBCMP(revokep->nclid_id, clp->lc_id, clp->lc_idlen)) {
791 			gotit = 1;
792 			break;
793 		}
794 	    }
795 	    i++;
796 	}
797 	if (!gotit) {
798 		NFSLOCKV4ROOTMUTEX();
799 		nfsv4_unlock(&nfsv4rootfs_lock, 0);
800 		NFSUNLOCKV4ROOTMUTEX();
801 		error = EPERM;
802 		goto out;
803 	}
804 
805 	/*
806 	 * Now, write out the revocation record
807 	 */
808 	nfsrv_writestable(clp->lc_id, clp->lc_idlen, NFSNST_REVOKE, p);
809 	nfsrv_backupstable();
810 
811 	/*
812 	 * and clear out the state, marking the clientid revoked.
813 	 */
814 	clp->lc_flags &= ~LCL_CALLBACKSON;
815 	clp->lc_flags |= LCL_ADMINREVOKED;
816 	nfsrv_cleanclient(clp, p);
817 	nfsrv_freedeleglist(&clp->lc_deleg);
818 	nfsrv_freedeleglist(&clp->lc_olddeleg);
819 	NFSLOCKV4ROOTMUTEX();
820 	nfsv4_unlock(&nfsv4rootfs_lock, 0);
821 	NFSUNLOCKV4ROOTMUTEX();
822 
823 out:
824 	NFSEXITCODE(error);
825 	return (error);
826 }
827 
828 /*
829  * Dump out stats for all clients. Called from nfssvc(2), that is used
830  * nfsstatsv1.
831  */
832 APPLESTATIC void
833 nfsrv_dumpclients(struct nfsd_dumpclients *dumpp, int maxcnt)
834 {
835 	struct nfsclient *clp;
836 	int i = 0, cnt = 0;
837 
838 	/*
839 	 * First, get a reference on the nfsv4rootfs_lock so that an
840 	 * exclusive lock cannot be acquired while dumping the clients.
841 	 */
842 	NFSLOCKV4ROOTMUTEX();
843 	nfsv4_getref(&nfsv4rootfs_lock, NULL, NFSV4ROOTLOCKMUTEXPTR, NULL);
844 	NFSUNLOCKV4ROOTMUTEX();
845 	NFSLOCKSTATE();
846 	/*
847 	 * Rattle through the client lists until done.
848 	 */
849 	while (i < nfsrv_clienthashsize && cnt < maxcnt) {
850 	    clp = LIST_FIRST(&nfsclienthash[i]);
851 	    while (clp != LIST_END(&nfsclienthash[i]) && cnt < maxcnt) {
852 		nfsrv_dumpaclient(clp, &dumpp[cnt]);
853 		cnt++;
854 		clp = LIST_NEXT(clp, lc_hash);
855 	    }
856 	    i++;
857 	}
858 	if (cnt < maxcnt)
859 	    dumpp[cnt].ndcl_clid.nclid_idlen = 0;
860 	NFSUNLOCKSTATE();
861 	NFSLOCKV4ROOTMUTEX();
862 	nfsv4_relref(&nfsv4rootfs_lock);
863 	NFSUNLOCKV4ROOTMUTEX();
864 }
865 
866 /*
867  * Dump stats for a client. Must be called with the NFSSTATELOCK and spl'd.
868  */
869 static void
870 nfsrv_dumpaclient(struct nfsclient *clp, struct nfsd_dumpclients *dumpp)
871 {
872 	struct nfsstate *stp, *openstp, *lckownstp;
873 	struct nfslock *lop;
874 	struct sockaddr *sad;
875 	struct sockaddr_in *rad;
876 	struct sockaddr_in6 *rad6;
877 
878 	dumpp->ndcl_nopenowners = dumpp->ndcl_nlockowners = 0;
879 	dumpp->ndcl_nopens = dumpp->ndcl_nlocks = 0;
880 	dumpp->ndcl_ndelegs = dumpp->ndcl_nolddelegs = 0;
881 	dumpp->ndcl_flags = clp->lc_flags;
882 	dumpp->ndcl_clid.nclid_idlen = clp->lc_idlen;
883 	NFSBCOPY(clp->lc_id, dumpp->ndcl_clid.nclid_id, clp->lc_idlen);
884 	sad = NFSSOCKADDR(clp->lc_req.nr_nam, struct sockaddr *);
885 	dumpp->ndcl_addrfam = sad->sa_family;
886 	if (sad->sa_family == AF_INET) {
887 		rad = (struct sockaddr_in *)sad;
888 		dumpp->ndcl_cbaddr.sin_addr = rad->sin_addr;
889 	} else {
890 		rad6 = (struct sockaddr_in6 *)sad;
891 		dumpp->ndcl_cbaddr.sin6_addr = rad6->sin6_addr;
892 	}
893 
894 	/*
895 	 * Now, scan the state lists and total up the opens and locks.
896 	 */
897 	LIST_FOREACH(stp, &clp->lc_open, ls_list) {
898 	    dumpp->ndcl_nopenowners++;
899 	    LIST_FOREACH(openstp, &stp->ls_open, ls_list) {
900 		dumpp->ndcl_nopens++;
901 		LIST_FOREACH(lckownstp, &openstp->ls_open, ls_list) {
902 		    dumpp->ndcl_nlockowners++;
903 		    LIST_FOREACH(lop, &lckownstp->ls_lock, lo_lckowner) {
904 			dumpp->ndcl_nlocks++;
905 		    }
906 		}
907 	    }
908 	}
909 
910 	/*
911 	 * and the delegation lists.
912 	 */
913 	LIST_FOREACH(stp, &clp->lc_deleg, ls_list) {
914 	    dumpp->ndcl_ndelegs++;
915 	}
916 	LIST_FOREACH(stp, &clp->lc_olddeleg, ls_list) {
917 	    dumpp->ndcl_nolddelegs++;
918 	}
919 }
920 
921 /*
922  * Dump out lock stats for a file.
923  */
924 APPLESTATIC void
925 nfsrv_dumplocks(vnode_t vp, struct nfsd_dumplocks *ldumpp, int maxcnt,
926     NFSPROC_T *p)
927 {
928 	struct nfsstate *stp;
929 	struct nfslock *lop;
930 	int cnt = 0;
931 	struct nfslockfile *lfp;
932 	struct sockaddr *sad;
933 	struct sockaddr_in *rad;
934 	struct sockaddr_in6 *rad6;
935 	int ret;
936 	fhandle_t nfh;
937 
938 	ret = nfsrv_getlockfh(vp, 0, NULL, &nfh, p);
939 	/*
940 	 * First, get a reference on the nfsv4rootfs_lock so that an
941 	 * exclusive lock on it cannot be acquired while dumping the locks.
942 	 */
943 	NFSLOCKV4ROOTMUTEX();
944 	nfsv4_getref(&nfsv4rootfs_lock, NULL, NFSV4ROOTLOCKMUTEXPTR, NULL);
945 	NFSUNLOCKV4ROOTMUTEX();
946 	NFSLOCKSTATE();
947 	if (!ret)
948 		ret = nfsrv_getlockfile(0, NULL, &lfp, &nfh, 0);
949 	if (ret) {
950 		ldumpp[0].ndlck_clid.nclid_idlen = 0;
951 		NFSUNLOCKSTATE();
952 		NFSLOCKV4ROOTMUTEX();
953 		nfsv4_relref(&nfsv4rootfs_lock);
954 		NFSUNLOCKV4ROOTMUTEX();
955 		return;
956 	}
957 
958 	/*
959 	 * For each open share on file, dump it out.
960 	 */
961 	stp = LIST_FIRST(&lfp->lf_open);
962 	while (stp != LIST_END(&lfp->lf_open) && cnt < maxcnt) {
963 		ldumpp[cnt].ndlck_flags = stp->ls_flags;
964 		ldumpp[cnt].ndlck_stateid.seqid = stp->ls_stateid.seqid;
965 		ldumpp[cnt].ndlck_stateid.other[0] = stp->ls_stateid.other[0];
966 		ldumpp[cnt].ndlck_stateid.other[1] = stp->ls_stateid.other[1];
967 		ldumpp[cnt].ndlck_stateid.other[2] = stp->ls_stateid.other[2];
968 		ldumpp[cnt].ndlck_owner.nclid_idlen =
969 		    stp->ls_openowner->ls_ownerlen;
970 		NFSBCOPY(stp->ls_openowner->ls_owner,
971 		    ldumpp[cnt].ndlck_owner.nclid_id,
972 		    stp->ls_openowner->ls_ownerlen);
973 		ldumpp[cnt].ndlck_clid.nclid_idlen = stp->ls_clp->lc_idlen;
974 		NFSBCOPY(stp->ls_clp->lc_id, ldumpp[cnt].ndlck_clid.nclid_id,
975 		    stp->ls_clp->lc_idlen);
976 		sad=NFSSOCKADDR(stp->ls_clp->lc_req.nr_nam, struct sockaddr *);
977 		ldumpp[cnt].ndlck_addrfam = sad->sa_family;
978 		if (sad->sa_family == AF_INET) {
979 			rad = (struct sockaddr_in *)sad;
980 			ldumpp[cnt].ndlck_cbaddr.sin_addr = rad->sin_addr;
981 		} else {
982 			rad6 = (struct sockaddr_in6 *)sad;
983 			ldumpp[cnt].ndlck_cbaddr.sin6_addr = rad6->sin6_addr;
984 		}
985 		stp = LIST_NEXT(stp, ls_file);
986 		cnt++;
987 	}
988 
989 	/*
990 	 * and all locks.
991 	 */
992 	lop = LIST_FIRST(&lfp->lf_lock);
993 	while (lop != LIST_END(&lfp->lf_lock) && cnt < maxcnt) {
994 		stp = lop->lo_stp;
995 		ldumpp[cnt].ndlck_flags = lop->lo_flags;
996 		ldumpp[cnt].ndlck_first = lop->lo_first;
997 		ldumpp[cnt].ndlck_end = lop->lo_end;
998 		ldumpp[cnt].ndlck_stateid.seqid = stp->ls_stateid.seqid;
999 		ldumpp[cnt].ndlck_stateid.other[0] = stp->ls_stateid.other[0];
1000 		ldumpp[cnt].ndlck_stateid.other[1] = stp->ls_stateid.other[1];
1001 		ldumpp[cnt].ndlck_stateid.other[2] = stp->ls_stateid.other[2];
1002 		ldumpp[cnt].ndlck_owner.nclid_idlen = stp->ls_ownerlen;
1003 		NFSBCOPY(stp->ls_owner, ldumpp[cnt].ndlck_owner.nclid_id,
1004 		    stp->ls_ownerlen);
1005 		ldumpp[cnt].ndlck_clid.nclid_idlen = stp->ls_clp->lc_idlen;
1006 		NFSBCOPY(stp->ls_clp->lc_id, ldumpp[cnt].ndlck_clid.nclid_id,
1007 		    stp->ls_clp->lc_idlen);
1008 		sad=NFSSOCKADDR(stp->ls_clp->lc_req.nr_nam, struct sockaddr *);
1009 		ldumpp[cnt].ndlck_addrfam = sad->sa_family;
1010 		if (sad->sa_family == AF_INET) {
1011 			rad = (struct sockaddr_in *)sad;
1012 			ldumpp[cnt].ndlck_cbaddr.sin_addr = rad->sin_addr;
1013 		} else {
1014 			rad6 = (struct sockaddr_in6 *)sad;
1015 			ldumpp[cnt].ndlck_cbaddr.sin6_addr = rad6->sin6_addr;
1016 		}
1017 		lop = LIST_NEXT(lop, lo_lckfile);
1018 		cnt++;
1019 	}
1020 
1021 	/*
1022 	 * and the delegations.
1023 	 */
1024 	stp = LIST_FIRST(&lfp->lf_deleg);
1025 	while (stp != LIST_END(&lfp->lf_deleg) && cnt < maxcnt) {
1026 		ldumpp[cnt].ndlck_flags = stp->ls_flags;
1027 		ldumpp[cnt].ndlck_stateid.seqid = stp->ls_stateid.seqid;
1028 		ldumpp[cnt].ndlck_stateid.other[0] = stp->ls_stateid.other[0];
1029 		ldumpp[cnt].ndlck_stateid.other[1] = stp->ls_stateid.other[1];
1030 		ldumpp[cnt].ndlck_stateid.other[2] = stp->ls_stateid.other[2];
1031 		ldumpp[cnt].ndlck_owner.nclid_idlen = 0;
1032 		ldumpp[cnt].ndlck_clid.nclid_idlen = stp->ls_clp->lc_idlen;
1033 		NFSBCOPY(stp->ls_clp->lc_id, ldumpp[cnt].ndlck_clid.nclid_id,
1034 		    stp->ls_clp->lc_idlen);
1035 		sad=NFSSOCKADDR(stp->ls_clp->lc_req.nr_nam, struct sockaddr *);
1036 		ldumpp[cnt].ndlck_addrfam = sad->sa_family;
1037 		if (sad->sa_family == AF_INET) {
1038 			rad = (struct sockaddr_in *)sad;
1039 			ldumpp[cnt].ndlck_cbaddr.sin_addr = rad->sin_addr;
1040 		} else {
1041 			rad6 = (struct sockaddr_in6 *)sad;
1042 			ldumpp[cnt].ndlck_cbaddr.sin6_addr = rad6->sin6_addr;
1043 		}
1044 		stp = LIST_NEXT(stp, ls_file);
1045 		cnt++;
1046 	}
1047 
1048 	/*
1049 	 * If list isn't full, mark end of list by setting the client name
1050 	 * to zero length.
1051 	 */
1052 	if (cnt < maxcnt)
1053 		ldumpp[cnt].ndlck_clid.nclid_idlen = 0;
1054 	NFSUNLOCKSTATE();
1055 	NFSLOCKV4ROOTMUTEX();
1056 	nfsv4_relref(&nfsv4rootfs_lock);
1057 	NFSUNLOCKV4ROOTMUTEX();
1058 }
1059 
1060 /*
1061  * Server timer routine. It can scan any linked list, so long
1062  * as it holds the spin/mutex lock and there is no exclusive lock on
1063  * nfsv4rootfs_lock.
1064  * (For OpenBSD, a kthread is ok. For FreeBSD, I think it is ok
1065  *  to do this from a callout, since the spin locks work. For
1066  *  Darwin, I'm not sure what will work correctly yet.)
1067  * Should be called once per second.
1068  */
1069 APPLESTATIC void
1070 nfsrv_servertimer(void)
1071 {
1072 	struct nfsclient *clp, *nclp;
1073 	struct nfsstate *stp, *nstp;
1074 	int got_ref, i;
1075 
1076 	/*
1077 	 * Make sure nfsboottime is set. This is used by V3 as well
1078 	 * as V4. Note that nfsboottime is not nfsrvboottime, which is
1079 	 * only used by the V4 server for leases.
1080 	 */
1081 	if (nfsboottime.tv_sec == 0)
1082 		NFSSETBOOTTIME(nfsboottime);
1083 
1084 	/*
1085 	 * If server hasn't started yet, just return.
1086 	 */
1087 	NFSLOCKSTATE();
1088 	if (nfsrv_stablefirst.nsf_eograce == 0) {
1089 		NFSUNLOCKSTATE();
1090 		return;
1091 	}
1092 	if (!(nfsrv_stablefirst.nsf_flags & NFSNSF_UPDATEDONE)) {
1093 		if (!(nfsrv_stablefirst.nsf_flags & NFSNSF_GRACEOVER) &&
1094 		    NFSD_MONOSEC > nfsrv_stablefirst.nsf_eograce)
1095 			nfsrv_stablefirst.nsf_flags |=
1096 			    (NFSNSF_GRACEOVER | NFSNSF_NEEDLOCK);
1097 		NFSUNLOCKSTATE();
1098 		return;
1099 	}
1100 
1101 	/*
1102 	 * Try and get a reference count on the nfsv4rootfs_lock so that
1103 	 * no nfsd thread can acquire an exclusive lock on it before this
1104 	 * call is done. If it is already exclusively locked, just return.
1105 	 */
1106 	NFSLOCKV4ROOTMUTEX();
1107 	got_ref = nfsv4_getref_nonblock(&nfsv4rootfs_lock);
1108 	NFSUNLOCKV4ROOTMUTEX();
1109 	if (got_ref == 0) {
1110 		NFSUNLOCKSTATE();
1111 		return;
1112 	}
1113 
1114 	/*
1115 	 * For each client...
1116 	 */
1117 	for (i = 0; i < nfsrv_clienthashsize; i++) {
1118 	    clp = LIST_FIRST(&nfsclienthash[i]);
1119 	    while (clp != LIST_END(&nfsclienthash[i])) {
1120 		nclp = LIST_NEXT(clp, lc_hash);
1121 		if (!(clp->lc_flags & LCL_EXPIREIT)) {
1122 		    if (((clp->lc_expiry + NFSRV_STALELEASE) < NFSD_MONOSEC
1123 			 && ((LIST_EMPTY(&clp->lc_deleg)
1124 			      && LIST_EMPTY(&clp->lc_open)) ||
1125 			     nfsrv_clients > nfsrv_clienthighwater)) ||
1126 			(clp->lc_expiry + NFSRV_MOULDYLEASE) < NFSD_MONOSEC ||
1127 			(clp->lc_expiry < NFSD_MONOSEC &&
1128 			 (nfsrv_openpluslock * 10 / 9) > nfsrv_v4statelimit)) {
1129 			/*
1130 			 * Lease has expired several nfsrv_lease times ago:
1131 			 * PLUS
1132 			 *    - no state is associated with it
1133 			 *    OR
1134 			 *    - above high water mark for number of clients
1135 			 *      (nfsrv_clienthighwater should be large enough
1136 			 *       that this only occurs when clients fail to
1137 			 *       use the same nfs_client_id4.id. Maybe somewhat
1138 			 *       higher that the maximum number of clients that
1139 			 *       will mount this server?)
1140 			 * OR
1141 			 * Lease has expired a very long time ago
1142 			 * OR
1143 			 * Lease has expired PLUS the number of opens + locks
1144 			 * has exceeded 90% of capacity
1145 			 *
1146 			 * --> Mark for expiry. The actual expiry will be done
1147 			 *     by an nfsd sometime soon.
1148 			 */
1149 			clp->lc_flags |= LCL_EXPIREIT;
1150 			nfsrv_stablefirst.nsf_flags |=
1151 			    (NFSNSF_NEEDLOCK | NFSNSF_EXPIREDCLIENT);
1152 		    } else {
1153 			/*
1154 			 * If there are no opens, increment no open tick cnt
1155 			 * If time exceeds NFSNOOPEN, mark it to be thrown away
1156 			 * otherwise, if there is an open, reset no open time
1157 			 * Hopefully, this will avoid excessive re-creation
1158 			 * of open owners and subsequent open confirms.
1159 			 */
1160 			stp = LIST_FIRST(&clp->lc_open);
1161 			while (stp != LIST_END(&clp->lc_open)) {
1162 				nstp = LIST_NEXT(stp, ls_list);
1163 				if (LIST_EMPTY(&stp->ls_open)) {
1164 					stp->ls_noopens++;
1165 					if (stp->ls_noopens > NFSNOOPEN ||
1166 					    (nfsrv_openpluslock * 2) >
1167 					    nfsrv_v4statelimit)
1168 						nfsrv_stablefirst.nsf_flags |=
1169 							NFSNSF_NOOPENS;
1170 				} else {
1171 					stp->ls_noopens = 0;
1172 				}
1173 				stp = nstp;
1174 			}
1175 		    }
1176 		}
1177 		clp = nclp;
1178 	    }
1179 	}
1180 	NFSUNLOCKSTATE();
1181 	NFSLOCKV4ROOTMUTEX();
1182 	nfsv4_relref(&nfsv4rootfs_lock);
1183 	NFSUNLOCKV4ROOTMUTEX();
1184 }
1185 
1186 /*
1187  * The following set of functions free up the various data structures.
1188  */
1189 /*
1190  * Clear out all open/lock state related to this nfsclient.
1191  * Caller must hold an exclusive lock on nfsv4rootfs_lock, so that
1192  * there are no other active nfsd threads.
1193  */
1194 APPLESTATIC void
1195 nfsrv_cleanclient(struct nfsclient *clp, NFSPROC_T *p)
1196 {
1197 	struct nfsstate *stp, *nstp;
1198 	struct nfsdsession *sep, *nsep;
1199 
1200 	LIST_FOREACH_SAFE(stp, &clp->lc_open, ls_list, nstp)
1201 		nfsrv_freeopenowner(stp, 1, p);
1202 	if ((clp->lc_flags & LCL_ADMINREVOKED) == 0)
1203 		LIST_FOREACH_SAFE(sep, &clp->lc_session, sess_list, nsep)
1204 			(void)nfsrv_freesession(sep, NULL);
1205 }
1206 
1207 /*
1208  * Free a client that has been cleaned. It should also already have been
1209  * removed from the lists.
1210  * (Just to be safe w.r.t. newnfs_disconnect(), call this function when
1211  *  softclock interrupts are enabled.)
1212  */
1213 APPLESTATIC void
1214 nfsrv_zapclient(struct nfsclient *clp, NFSPROC_T *p)
1215 {
1216 
1217 #ifdef notyet
1218 	if ((clp->lc_flags & (LCL_GSS | LCL_CALLBACKSON)) ==
1219 	     (LCL_GSS | LCL_CALLBACKSON) &&
1220 	    (clp->lc_hand.nfsh_flag & NFSG_COMPLETE) &&
1221 	    clp->lc_handlelen > 0) {
1222 		clp->lc_hand.nfsh_flag &= ~NFSG_COMPLETE;
1223 		clp->lc_hand.nfsh_flag |= NFSG_DESTROYED;
1224 		(void) nfsrv_docallback(clp, NFSV4PROC_CBNULL,
1225 			NULL, 0, NULL, NULL, NULL, p);
1226 	}
1227 #endif
1228 	newnfs_disconnect(&clp->lc_req);
1229 	free(clp->lc_req.nr_nam, M_SONAME);
1230 	NFSFREEMUTEX(&clp->lc_req.nr_mtx);
1231 	free(clp->lc_stateid, M_NFSDCLIENT);
1232 	free(clp, M_NFSDCLIENT);
1233 	NFSLOCKSTATE();
1234 	nfsstatsv1.srvclients--;
1235 	nfsrv_openpluslock--;
1236 	nfsrv_clients--;
1237 	NFSUNLOCKSTATE();
1238 }
1239 
1240 /*
1241  * Free a list of delegation state structures.
1242  * (This function will also free all nfslockfile structures that no
1243  *  longer have associated state.)
1244  */
1245 APPLESTATIC void
1246 nfsrv_freedeleglist(struct nfsstatehead *sthp)
1247 {
1248 	struct nfsstate *stp, *nstp;
1249 
1250 	LIST_FOREACH_SAFE(stp, sthp, ls_list, nstp) {
1251 		nfsrv_freedeleg(stp);
1252 	}
1253 	LIST_INIT(sthp);
1254 }
1255 
1256 /*
1257  * Free up a delegation.
1258  */
1259 static void
1260 nfsrv_freedeleg(struct nfsstate *stp)
1261 {
1262 	struct nfslockfile *lfp;
1263 
1264 	LIST_REMOVE(stp, ls_hash);
1265 	LIST_REMOVE(stp, ls_list);
1266 	LIST_REMOVE(stp, ls_file);
1267 	if ((stp->ls_flags & NFSLCK_DELEGWRITE) != 0)
1268 		nfsrv_writedelegcnt--;
1269 	lfp = stp->ls_lfp;
1270 	if (LIST_EMPTY(&lfp->lf_open) &&
1271 	    LIST_EMPTY(&lfp->lf_lock) && LIST_EMPTY(&lfp->lf_deleg) &&
1272 	    LIST_EMPTY(&lfp->lf_locallock) && LIST_EMPTY(&lfp->lf_rollback) &&
1273 	    lfp->lf_usecount == 0 &&
1274 	    nfsv4_testlock(&lfp->lf_locallock_lck) == 0)
1275 		nfsrv_freenfslockfile(lfp);
1276 	free(stp, M_NFSDSTATE);
1277 	nfsstatsv1.srvdelegates--;
1278 	nfsrv_openpluslock--;
1279 	nfsrv_delegatecnt--;
1280 }
1281 
1282 /*
1283  * This function frees an open owner and all associated opens.
1284  */
1285 static void
1286 nfsrv_freeopenowner(struct nfsstate *stp, int cansleep, NFSPROC_T *p)
1287 {
1288 	struct nfsstate *nstp, *tstp;
1289 
1290 	LIST_REMOVE(stp, ls_list);
1291 	/*
1292 	 * Now, free all associated opens.
1293 	 */
1294 	nstp = LIST_FIRST(&stp->ls_open);
1295 	while (nstp != LIST_END(&stp->ls_open)) {
1296 		tstp = nstp;
1297 		nstp = LIST_NEXT(nstp, ls_list);
1298 		(void) nfsrv_freeopen(tstp, NULL, cansleep, p);
1299 	}
1300 	if (stp->ls_op)
1301 		nfsrvd_derefcache(stp->ls_op);
1302 	free(stp, M_NFSDSTATE);
1303 	nfsstatsv1.srvopenowners--;
1304 	nfsrv_openpluslock--;
1305 }
1306 
1307 /*
1308  * This function frees an open (nfsstate open structure) with all associated
1309  * lock_owners and locks. It also frees the nfslockfile structure iff there
1310  * are no other opens on the file.
1311  * Returns 1 if it free'd the nfslockfile, 0 otherwise.
1312  */
1313 static int
1314 nfsrv_freeopen(struct nfsstate *stp, vnode_t vp, int cansleep, NFSPROC_T *p)
1315 {
1316 	struct nfsstate *nstp, *tstp;
1317 	struct nfslockfile *lfp;
1318 	int ret;
1319 
1320 	LIST_REMOVE(stp, ls_hash);
1321 	LIST_REMOVE(stp, ls_list);
1322 	LIST_REMOVE(stp, ls_file);
1323 
1324 	lfp = stp->ls_lfp;
1325 	/*
1326 	 * Now, free all lockowners associated with this open.
1327 	 */
1328 	LIST_FOREACH_SAFE(tstp, &stp->ls_open, ls_list, nstp)
1329 		nfsrv_freelockowner(tstp, vp, cansleep, p);
1330 
1331 	/*
1332 	 * The nfslockfile is freed here if there are no locks
1333 	 * associated with the open.
1334 	 * If there are locks associated with the open, the
1335 	 * nfslockfile structure can be freed via nfsrv_freelockowner().
1336 	 * Acquire the state mutex to avoid races with calls to
1337 	 * nfsrv_getlockfile().
1338 	 */
1339 	if (cansleep != 0)
1340 		NFSLOCKSTATE();
1341 	if (lfp != NULL && LIST_EMPTY(&lfp->lf_open) &&
1342 	    LIST_EMPTY(&lfp->lf_deleg) && LIST_EMPTY(&lfp->lf_lock) &&
1343 	    LIST_EMPTY(&lfp->lf_locallock) && LIST_EMPTY(&lfp->lf_rollback) &&
1344 	    lfp->lf_usecount == 0 &&
1345 	    (cansleep != 0 || nfsv4_testlock(&lfp->lf_locallock_lck) == 0)) {
1346 		nfsrv_freenfslockfile(lfp);
1347 		ret = 1;
1348 	} else
1349 		ret = 0;
1350 	if (cansleep != 0)
1351 		NFSUNLOCKSTATE();
1352 	free(stp, M_NFSDSTATE);
1353 	nfsstatsv1.srvopens--;
1354 	nfsrv_openpluslock--;
1355 	return (ret);
1356 }
1357 
1358 /*
1359  * Frees a lockowner and all associated locks.
1360  */
1361 static void
1362 nfsrv_freelockowner(struct nfsstate *stp, vnode_t vp, int cansleep,
1363     NFSPROC_T *p)
1364 {
1365 
1366 	LIST_REMOVE(stp, ls_hash);
1367 	LIST_REMOVE(stp, ls_list);
1368 	nfsrv_freeallnfslocks(stp, vp, cansleep, p);
1369 	if (stp->ls_op)
1370 		nfsrvd_derefcache(stp->ls_op);
1371 	free(stp, M_NFSDSTATE);
1372 	nfsstatsv1.srvlockowners--;
1373 	nfsrv_openpluslock--;
1374 }
1375 
1376 /*
1377  * Free all the nfs locks on a lockowner.
1378  */
1379 static void
1380 nfsrv_freeallnfslocks(struct nfsstate *stp, vnode_t vp, int cansleep,
1381     NFSPROC_T *p)
1382 {
1383 	struct nfslock *lop, *nlop;
1384 	struct nfsrollback *rlp, *nrlp;
1385 	struct nfslockfile *lfp = NULL;
1386 	int gottvp = 0;
1387 	vnode_t tvp = NULL;
1388 	uint64_t first, end;
1389 
1390 	if (vp != NULL)
1391 		ASSERT_VOP_UNLOCKED(vp, "nfsrv_freeallnfslocks: vnode locked");
1392 	lop = LIST_FIRST(&stp->ls_lock);
1393 	while (lop != LIST_END(&stp->ls_lock)) {
1394 		nlop = LIST_NEXT(lop, lo_lckowner);
1395 		/*
1396 		 * Since all locks should be for the same file, lfp should
1397 		 * not change.
1398 		 */
1399 		if (lfp == NULL)
1400 			lfp = lop->lo_lfp;
1401 		else if (lfp != lop->lo_lfp)
1402 			panic("allnfslocks");
1403 		/*
1404 		 * If vp is NULL and cansleep != 0, a vnode must be acquired
1405 		 * from the file handle. This only occurs when called from
1406 		 * nfsrv_cleanclient().
1407 		 */
1408 		if (gottvp == 0) {
1409 			if (nfsrv_dolocallocks == 0)
1410 				tvp = NULL;
1411 			else if (vp == NULL && cansleep != 0) {
1412 				tvp = nfsvno_getvp(&lfp->lf_fh);
1413 				NFSVOPUNLOCK(tvp, 0);
1414 			} else
1415 				tvp = vp;
1416 			gottvp = 1;
1417 		}
1418 
1419 		if (tvp != NULL) {
1420 			if (cansleep == 0)
1421 				panic("allnfs2");
1422 			first = lop->lo_first;
1423 			end = lop->lo_end;
1424 			nfsrv_freenfslock(lop);
1425 			nfsrv_localunlock(tvp, lfp, first, end, p);
1426 			LIST_FOREACH_SAFE(rlp, &lfp->lf_rollback, rlck_list,
1427 			    nrlp)
1428 				free(rlp, M_NFSDROLLBACK);
1429 			LIST_INIT(&lfp->lf_rollback);
1430 		} else
1431 			nfsrv_freenfslock(lop);
1432 		lop = nlop;
1433 	}
1434 	if (vp == NULL && tvp != NULL)
1435 		vrele(tvp);
1436 }
1437 
1438 /*
1439  * Free an nfslock structure.
1440  */
1441 static void
1442 nfsrv_freenfslock(struct nfslock *lop)
1443 {
1444 
1445 	if (lop->lo_lckfile.le_prev != NULL) {
1446 		LIST_REMOVE(lop, lo_lckfile);
1447 		nfsstatsv1.srvlocks--;
1448 		nfsrv_openpluslock--;
1449 	}
1450 	LIST_REMOVE(lop, lo_lckowner);
1451 	free(lop, M_NFSDLOCK);
1452 }
1453 
1454 /*
1455  * This function frees an nfslockfile structure.
1456  */
1457 static void
1458 nfsrv_freenfslockfile(struct nfslockfile *lfp)
1459 {
1460 
1461 	LIST_REMOVE(lfp, lf_hash);
1462 	free(lfp, M_NFSDLOCKFILE);
1463 }
1464 
1465 /*
1466  * This function looks up an nfsstate structure via stateid.
1467  */
1468 static int
1469 nfsrv_getstate(struct nfsclient *clp, nfsv4stateid_t *stateidp, __unused u_int32_t flags,
1470     struct nfsstate **stpp)
1471 {
1472 	struct nfsstate *stp;
1473 	struct nfsstatehead *hp;
1474 	int error = 0;
1475 
1476 	*stpp = NULL;
1477 	hp = NFSSTATEHASH(clp, *stateidp);
1478 	LIST_FOREACH(stp, hp, ls_hash) {
1479 		if (!NFSBCMP(stp->ls_stateid.other, stateidp->other,
1480 			NFSX_STATEIDOTHER))
1481 			break;
1482 	}
1483 
1484 	/*
1485 	 * If no state id in list, return NFSERR_BADSTATEID.
1486 	 */
1487 	if (stp == LIST_END(hp)) {
1488 		error = NFSERR_BADSTATEID;
1489 		goto out;
1490 	}
1491 	*stpp = stp;
1492 
1493 out:
1494 	NFSEXITCODE(error);
1495 	return (error);
1496 }
1497 
1498 /*
1499  * This function gets an nfsstate structure via owner string.
1500  */
1501 static void
1502 nfsrv_getowner(struct nfsstatehead *hp, struct nfsstate *new_stp,
1503     struct nfsstate **stpp)
1504 {
1505 	struct nfsstate *stp;
1506 
1507 	*stpp = NULL;
1508 	LIST_FOREACH(stp, hp, ls_list) {
1509 		if (new_stp->ls_ownerlen == stp->ls_ownerlen &&
1510 		  !NFSBCMP(new_stp->ls_owner,stp->ls_owner,stp->ls_ownerlen)) {
1511 			*stpp = stp;
1512 			return;
1513 		}
1514 	}
1515 }
1516 
1517 /*
1518  * Lock control function called to update lock status.
1519  * Returns 0 upon success, -1 if there is no lock and the flags indicate
1520  * that one isn't to be created and an NFSERR_xxx for other errors.
1521  * The structures new_stp and new_lop are passed in as pointers that should
1522  * be set to NULL if the structure is used and shouldn't be free'd.
1523  * For the NFSLCK_TEST and NFSLCK_CHECK cases, the structures are
1524  * never used and can safely be allocated on the stack. For all other
1525  * cases, *new_stpp and *new_lopp should be malloc'd before the call,
1526  * in case they are used.
1527  */
1528 APPLESTATIC int
1529 nfsrv_lockctrl(vnode_t vp, struct nfsstate **new_stpp,
1530     struct nfslock **new_lopp, struct nfslockconflict *cfp,
1531     nfsquad_t clientid, nfsv4stateid_t *stateidp,
1532     __unused struct nfsexstuff *exp,
1533     struct nfsrv_descript *nd, NFSPROC_T *p)
1534 {
1535 	struct nfslock *lop;
1536 	struct nfsstate *new_stp = *new_stpp;
1537 	struct nfslock *new_lop = *new_lopp;
1538 	struct nfsstate *tstp, *mystp, *nstp;
1539 	int specialid = 0;
1540 	struct nfslockfile *lfp;
1541 	struct nfslock *other_lop = NULL;
1542 	struct nfsstate *stp, *lckstp = NULL;
1543 	struct nfsclient *clp = NULL;
1544 	u_int32_t bits;
1545 	int error = 0, haslock = 0, ret, reterr;
1546 	int getlckret, delegation = 0, filestruct_locked, vnode_unlocked = 0;
1547 	fhandle_t nfh;
1548 	uint64_t first, end;
1549 	uint32_t lock_flags;
1550 
1551 	if (new_stp->ls_flags & (NFSLCK_CHECK | NFSLCK_SETATTR)) {
1552 		/*
1553 		 * Note the special cases of "all 1s" or "all 0s" stateids and
1554 		 * let reads with all 1s go ahead.
1555 		 */
1556 		if (new_stp->ls_stateid.seqid == 0x0 &&
1557 		    new_stp->ls_stateid.other[0] == 0x0 &&
1558 		    new_stp->ls_stateid.other[1] == 0x0 &&
1559 		    new_stp->ls_stateid.other[2] == 0x0)
1560 			specialid = 1;
1561 		else if (new_stp->ls_stateid.seqid == 0xffffffff &&
1562 		    new_stp->ls_stateid.other[0] == 0xffffffff &&
1563 		    new_stp->ls_stateid.other[1] == 0xffffffff &&
1564 		    new_stp->ls_stateid.other[2] == 0xffffffff)
1565 			specialid = 2;
1566 	}
1567 
1568 	/*
1569 	 * Check for restart conditions (client and server).
1570 	 */
1571 	error = nfsrv_checkrestart(clientid, new_stp->ls_flags,
1572 	    &new_stp->ls_stateid, specialid);
1573 	if (error)
1574 		goto out;
1575 
1576 	/*
1577 	 * Check for state resource limit exceeded.
1578 	 */
1579 	if ((new_stp->ls_flags & NFSLCK_LOCK) &&
1580 	    nfsrv_openpluslock > nfsrv_v4statelimit) {
1581 		error = NFSERR_RESOURCE;
1582 		goto out;
1583 	}
1584 
1585 	/*
1586 	 * For the lock case, get another nfslock structure,
1587 	 * just in case we need it.
1588 	 * Malloc now, before we start sifting through the linked lists,
1589 	 * in case we have to wait for memory.
1590 	 */
1591 tryagain:
1592 	if (new_stp->ls_flags & NFSLCK_LOCK)
1593 		other_lop = malloc(sizeof (struct nfslock),
1594 		    M_NFSDLOCK, M_WAITOK);
1595 	filestruct_locked = 0;
1596 	reterr = 0;
1597 	lfp = NULL;
1598 
1599 	/*
1600 	 * Get the lockfile structure for CFH now, so we can do a sanity
1601 	 * check against the stateid, before incrementing the seqid#, since
1602 	 * we want to return NFSERR_BADSTATEID on failure and the seqid#
1603 	 * shouldn't be incremented for this case.
1604 	 * If nfsrv_getlockfile() returns -1, it means "not found", which
1605 	 * will be handled later.
1606 	 * If we are doing Lock/LockU and local locking is enabled, sleep
1607 	 * lock the nfslockfile structure.
1608 	 */
1609 	getlckret = nfsrv_getlockfh(vp, new_stp->ls_flags, NULL, &nfh, p);
1610 	NFSLOCKSTATE();
1611 	if (getlckret == 0) {
1612 		if ((new_stp->ls_flags & (NFSLCK_LOCK | NFSLCK_UNLOCK)) != 0 &&
1613 		    nfsrv_dolocallocks != 0 && nd->nd_repstat == 0) {
1614 			getlckret = nfsrv_getlockfile(new_stp->ls_flags, NULL,
1615 			    &lfp, &nfh, 1);
1616 			if (getlckret == 0)
1617 				filestruct_locked = 1;
1618 		} else
1619 			getlckret = nfsrv_getlockfile(new_stp->ls_flags, NULL,
1620 			    &lfp, &nfh, 0);
1621 	}
1622 	if (getlckret != 0 && getlckret != -1)
1623 		reterr = getlckret;
1624 
1625 	if (filestruct_locked != 0) {
1626 		LIST_INIT(&lfp->lf_rollback);
1627 		if ((new_stp->ls_flags & NFSLCK_LOCK)) {
1628 			/*
1629 			 * For local locking, do the advisory locking now, so
1630 			 * that any conflict can be detected. A failure later
1631 			 * can be rolled back locally. If an error is returned,
1632 			 * struct nfslockfile has been unlocked and any local
1633 			 * locking rolled back.
1634 			 */
1635 			NFSUNLOCKSTATE();
1636 			if (vnode_unlocked == 0) {
1637 				ASSERT_VOP_ELOCKED(vp, "nfsrv_lockctrl1");
1638 				vnode_unlocked = 1;
1639 				NFSVOPUNLOCK(vp, 0);
1640 			}
1641 			reterr = nfsrv_locallock(vp, lfp,
1642 			    (new_lop->lo_flags & (NFSLCK_READ | NFSLCK_WRITE)),
1643 			    new_lop->lo_first, new_lop->lo_end, cfp, p);
1644 			NFSLOCKSTATE();
1645 		}
1646 	}
1647 
1648 	if (specialid == 0) {
1649 	    if (new_stp->ls_flags & NFSLCK_TEST) {
1650 		/*
1651 		 * RFC 3530 does not list LockT as an op that renews a
1652 		 * lease, but the consensus seems to be that it is ok
1653 		 * for a server to do so.
1654 		 */
1655 		error = nfsrv_getclient(clientid, CLOPS_RENEW, &clp, NULL,
1656 		    (nfsquad_t)((u_quad_t)0), 0, nd, p);
1657 
1658 		/*
1659 		 * Since NFSERR_EXPIRED, NFSERR_ADMINREVOKED are not valid
1660 		 * error returns for LockT, just go ahead and test for a lock,
1661 		 * since there are no locks for this client, but other locks
1662 		 * can conflict. (ie. same client will always be false)
1663 		 */
1664 		if (error == NFSERR_EXPIRED || error == NFSERR_ADMINREVOKED)
1665 		    error = 0;
1666 		lckstp = new_stp;
1667 	    } else {
1668 	      error = nfsrv_getclient(clientid, CLOPS_RENEW, &clp, NULL,
1669 		(nfsquad_t)((u_quad_t)0), 0, nd, p);
1670 	      if (error == 0)
1671 		/*
1672 		 * Look up the stateid
1673 		 */
1674 		error = nfsrv_getstate(clp, &new_stp->ls_stateid,
1675 		  new_stp->ls_flags, &stp);
1676 	      /*
1677 	       * do some sanity checks for an unconfirmed open or a
1678 	       * stateid that refers to the wrong file, for an open stateid
1679 	       */
1680 	      if (error == 0 && (stp->ls_flags & NFSLCK_OPEN) &&
1681 		  ((stp->ls_openowner->ls_flags & NFSLCK_NEEDSCONFIRM) ||
1682 		   (getlckret == 0 && stp->ls_lfp != lfp))){
1683 		      /*
1684 		       * NFSLCK_SETATTR should return OK rather than NFSERR_BADSTATEID
1685 		       * The only exception is using SETATTR with SIZE.
1686 		       * */
1687                     if ((new_stp->ls_flags &
1688                          (NFSLCK_SETATTR | NFSLCK_CHECK)) != NFSLCK_SETATTR)
1689 			     error = NFSERR_BADSTATEID;
1690 	      }
1691 
1692 		if (error == 0 &&
1693 		  (stp->ls_flags & (NFSLCK_DELEGREAD | NFSLCK_DELEGWRITE)) &&
1694 		  getlckret == 0 && stp->ls_lfp != lfp)
1695 			error = NFSERR_BADSTATEID;
1696 
1697 	      /*
1698 	       * If the lockowner stateid doesn't refer to the same file,
1699 	       * I believe that is considered ok, since some clients will
1700 	       * only create a single lockowner and use that for all locks
1701 	       * on all files.
1702 	       * For now, log it as a diagnostic, instead of considering it
1703 	       * a BadStateid.
1704 	       */
1705 	      if (error == 0 && (stp->ls_flags &
1706 		  (NFSLCK_OPEN | NFSLCK_DELEGREAD | NFSLCK_DELEGWRITE)) == 0 &&
1707 		  getlckret == 0 && stp->ls_lfp != lfp) {
1708 #ifdef DIAGNOSTIC
1709 		  printf("Got a lock statid for different file open\n");
1710 #endif
1711 		  /*
1712 		  error = NFSERR_BADSTATEID;
1713 		  */
1714 	      }
1715 
1716 	      if (error == 0) {
1717 		    if (new_stp->ls_flags & NFSLCK_OPENTOLOCK) {
1718 			/*
1719 			 * If haslock set, we've already checked the seqid.
1720 			 */
1721 			if (!haslock) {
1722 			    if (stp->ls_flags & NFSLCK_OPEN)
1723 				error = nfsrv_checkseqid(nd, new_stp->ls_seq,
1724 				    stp->ls_openowner, new_stp->ls_op);
1725 			    else
1726 				error = NFSERR_BADSTATEID;
1727 			}
1728 			if (!error)
1729 			    nfsrv_getowner(&stp->ls_open, new_stp, &lckstp);
1730 			if (lckstp)
1731 			    /*
1732 			     * I believe this should be an error, but it
1733 			     * isn't obvious what NFSERR_xxx would be
1734 			     * appropriate, so I'll use NFSERR_INVAL for now.
1735 			     */
1736 			    error = NFSERR_INVAL;
1737 			else
1738 			    lckstp = new_stp;
1739 		    } else if (new_stp->ls_flags&(NFSLCK_LOCK|NFSLCK_UNLOCK)) {
1740 			/*
1741 			 * If haslock set, ditto above.
1742 			 */
1743 			if (!haslock) {
1744 			    if (stp->ls_flags & NFSLCK_OPEN)
1745 				error = NFSERR_BADSTATEID;
1746 			    else
1747 				error = nfsrv_checkseqid(nd, new_stp->ls_seq,
1748 				    stp, new_stp->ls_op);
1749 			}
1750 			lckstp = stp;
1751 		    } else {
1752 			lckstp = stp;
1753 		    }
1754 	      }
1755 	      /*
1756 	       * If the seqid part of the stateid isn't the same, return
1757 	       * NFSERR_OLDSTATEID for cases other than I/O Ops.
1758 	       * For I/O Ops, only return NFSERR_OLDSTATEID if
1759 	       * nfsrv_returnoldstateid is set. (The consensus on the email
1760 	       * list was that most clients would prefer to not receive
1761 	       * NFSERR_OLDSTATEID for I/O Ops, but the RFC suggests that that
1762 	       * is what will happen, so I use the nfsrv_returnoldstateid to
1763 	       * allow for either server configuration.)
1764 	       */
1765 	      if (!error && stp->ls_stateid.seqid!=new_stp->ls_stateid.seqid &&
1766 		  (((nd->nd_flag & ND_NFSV41) == 0 &&
1767 		   (!(new_stp->ls_flags & NFSLCK_CHECK) ||
1768 		    nfsrv_returnoldstateid)) ||
1769 		   ((nd->nd_flag & ND_NFSV41) != 0 &&
1770 		    new_stp->ls_stateid.seqid != 0)))
1771 		    error = NFSERR_OLDSTATEID;
1772 	    }
1773 	}
1774 
1775 	/*
1776 	 * Now we can check for grace.
1777 	 */
1778 	if (!error)
1779 		error = nfsrv_checkgrace(nd, clp, new_stp->ls_flags);
1780 	if ((new_stp->ls_flags & NFSLCK_RECLAIM) && !error &&
1781 		nfsrv_checkstable(clp))
1782 		error = NFSERR_NOGRACE;
1783 	/*
1784 	 * If we successfully Reclaimed state, note that.
1785 	 */
1786 	if ((new_stp->ls_flags & NFSLCK_RECLAIM) && !error)
1787 		nfsrv_markstable(clp);
1788 
1789 	/*
1790 	 * At this point, either error == NFSERR_BADSTATEID or the
1791 	 * seqid# has been updated, so we can return any error.
1792 	 * If error == 0, there may be an error in:
1793 	 *    nd_repstat - Set by the calling function.
1794 	 *    reterr - Set above, if getting the nfslockfile structure
1795 	 *       or acquiring the local lock failed.
1796 	 *    (If both of these are set, nd_repstat should probably be
1797 	 *     returned, since that error was detected before this
1798 	 *     function call.)
1799 	 */
1800 	if (error != 0 || nd->nd_repstat != 0 || reterr != 0) {
1801 		if (error == 0) {
1802 			if (nd->nd_repstat != 0)
1803 				error = nd->nd_repstat;
1804 			else
1805 				error = reterr;
1806 		}
1807 		if (filestruct_locked != 0) {
1808 			/* Roll back local locks. */
1809 			NFSUNLOCKSTATE();
1810 			if (vnode_unlocked == 0) {
1811 				ASSERT_VOP_ELOCKED(vp, "nfsrv_lockctrl2");
1812 				vnode_unlocked = 1;
1813 				NFSVOPUNLOCK(vp, 0);
1814 			}
1815 			nfsrv_locallock_rollback(vp, lfp, p);
1816 			NFSLOCKSTATE();
1817 			nfsrv_unlocklf(lfp);
1818 		}
1819 		NFSUNLOCKSTATE();
1820 		goto out;
1821 	}
1822 
1823 	/*
1824 	 * Check the nfsrv_getlockfile return.
1825 	 * Returned -1 if no structure found.
1826 	 */
1827 	if (getlckret == -1) {
1828 		error = NFSERR_EXPIRED;
1829 		/*
1830 		 * Called from lockt, so no lock is OK.
1831 		 */
1832 		if (new_stp->ls_flags & NFSLCK_TEST) {
1833 			error = 0;
1834 		} else if (new_stp->ls_flags &
1835 		    (NFSLCK_CHECK | NFSLCK_SETATTR)) {
1836 			/*
1837 			 * Called to check for a lock, OK if the stateid is all
1838 			 * 1s or all 0s, but there should be an nfsstate
1839 			 * otherwise.
1840 			 * (ie. If there is no open, I'll assume no share
1841 			 *  deny bits.)
1842 			 */
1843 			if (specialid)
1844 				error = 0;
1845 			else
1846 				error = NFSERR_BADSTATEID;
1847 		}
1848 		NFSUNLOCKSTATE();
1849 		goto out;
1850 	}
1851 
1852 	/*
1853 	 * For NFSLCK_CHECK and NFSLCK_LOCK, test for a share conflict.
1854 	 * For NFSLCK_CHECK, allow a read if write access is granted,
1855 	 * but check for a deny. For NFSLCK_LOCK, require correct access,
1856 	 * which implies a conflicting deny can't exist.
1857 	 */
1858 	if (new_stp->ls_flags & (NFSLCK_CHECK | NFSLCK_LOCK)) {
1859 	    /*
1860 	     * Four kinds of state id:
1861 	     * - specialid (all 0s or all 1s), only for NFSLCK_CHECK
1862 	     * - stateid for an open
1863 	     * - stateid for a delegation
1864 	     * - stateid for a lock owner
1865 	     */
1866 	    if (!specialid) {
1867 		if (stp->ls_flags & (NFSLCK_DELEGREAD | NFSLCK_DELEGWRITE)) {
1868 		    delegation = 1;
1869 		    mystp = stp;
1870 		    nfsrv_delaydelegtimeout(stp);
1871 	        } else if (stp->ls_flags & NFSLCK_OPEN) {
1872 		    mystp = stp;
1873 		} else {
1874 		    mystp = stp->ls_openstp;
1875 		}
1876 		/*
1877 		 * If locking or checking, require correct access
1878 		 * bit set.
1879 		 */
1880 		if (((new_stp->ls_flags & NFSLCK_LOCK) &&
1881 		     !((new_lop->lo_flags >> NFSLCK_LOCKSHIFT) &
1882 		       mystp->ls_flags & NFSLCK_ACCESSBITS)) ||
1883 		    ((new_stp->ls_flags & (NFSLCK_CHECK|NFSLCK_READACCESS)) ==
1884 		      (NFSLCK_CHECK | NFSLCK_READACCESS) &&
1885 		     !(mystp->ls_flags & NFSLCK_READACCESS) &&
1886 		     nfsrv_allowreadforwriteopen == 0) ||
1887 		    ((new_stp->ls_flags & (NFSLCK_CHECK|NFSLCK_WRITEACCESS)) ==
1888 		      (NFSLCK_CHECK | NFSLCK_WRITEACCESS) &&
1889 		     !(mystp->ls_flags & NFSLCK_WRITEACCESS))) {
1890 			if (filestruct_locked != 0) {
1891 				/* Roll back local locks. */
1892 				NFSUNLOCKSTATE();
1893 				if (vnode_unlocked == 0) {
1894 					ASSERT_VOP_ELOCKED(vp,
1895 					    "nfsrv_lockctrl3");
1896 					vnode_unlocked = 1;
1897 					NFSVOPUNLOCK(vp, 0);
1898 				}
1899 				nfsrv_locallock_rollback(vp, lfp, p);
1900 				NFSLOCKSTATE();
1901 				nfsrv_unlocklf(lfp);
1902 			}
1903 			NFSUNLOCKSTATE();
1904 			error = NFSERR_OPENMODE;
1905 			goto out;
1906 		}
1907 	    } else
1908 		mystp = NULL;
1909 	    if ((new_stp->ls_flags & NFSLCK_CHECK) && !delegation) {
1910 		/*
1911 		 * Check for a conflicting deny bit.
1912 		 */
1913 		LIST_FOREACH(tstp, &lfp->lf_open, ls_file) {
1914 		    if (tstp != mystp) {
1915 			bits = tstp->ls_flags;
1916 			bits >>= NFSLCK_SHIFT;
1917 			if (new_stp->ls_flags & bits & NFSLCK_ACCESSBITS) {
1918 			    KASSERT(vnode_unlocked == 0,
1919 				("nfsrv_lockctrl: vnode unlocked1"));
1920 			    ret = nfsrv_clientconflict(tstp->ls_clp, &haslock,
1921 				vp, p);
1922 			    if (ret == 1) {
1923 				/*
1924 				* nfsrv_clientconflict unlocks state
1925 				 * when it returns non-zero.
1926 				 */
1927 				lckstp = NULL;
1928 				goto tryagain;
1929 			    }
1930 			    if (ret == 0)
1931 				NFSUNLOCKSTATE();
1932 			    if (ret == 2)
1933 				error = NFSERR_PERM;
1934 			    else
1935 				error = NFSERR_OPENMODE;
1936 			    goto out;
1937 			}
1938 		    }
1939 		}
1940 
1941 		/* We're outta here */
1942 		NFSUNLOCKSTATE();
1943 		goto out;
1944 	    }
1945 	}
1946 
1947 	/*
1948 	 * For setattr, just get rid of all the Delegations for other clients.
1949 	 */
1950 	if (new_stp->ls_flags & NFSLCK_SETATTR) {
1951 		KASSERT(vnode_unlocked == 0,
1952 		    ("nfsrv_lockctrl: vnode unlocked2"));
1953 		ret = nfsrv_cleandeleg(vp, lfp, clp, &haslock, p);
1954 		if (ret) {
1955 			/*
1956 			 * nfsrv_cleandeleg() unlocks state when it
1957 			 * returns non-zero.
1958 			 */
1959 			if (ret == -1) {
1960 				lckstp = NULL;
1961 				goto tryagain;
1962 			}
1963 			error = ret;
1964 			goto out;
1965 		}
1966 		if (!(new_stp->ls_flags & NFSLCK_CHECK) ||
1967 		    (LIST_EMPTY(&lfp->lf_open) && LIST_EMPTY(&lfp->lf_lock) &&
1968 		     LIST_EMPTY(&lfp->lf_deleg))) {
1969 			NFSUNLOCKSTATE();
1970 			goto out;
1971 		}
1972 	}
1973 
1974 	/*
1975 	 * Check for a conflicting delegation. If one is found, call
1976 	 * nfsrv_delegconflict() to handle it. If the v4root lock hasn't
1977 	 * been set yet, it will get the lock. Otherwise, it will recall
1978 	 * the delegation. Then, we try try again...
1979 	 * I currently believe the conflict algorithm to be:
1980 	 * For Lock Ops (Lock/LockT/LockU)
1981 	 * - there is a conflict iff a different client has a write delegation
1982 	 * For Reading (Read Op)
1983 	 * - there is a conflict iff a different client has a write delegation
1984 	 *   (the specialids are always a different client)
1985 	 * For Writing (Write/Setattr of size)
1986 	 * - there is a conflict if a different client has any delegation
1987 	 * - there is a conflict if the same client has a read delegation
1988 	 *   (I don't understand why this isn't allowed, but that seems to be
1989 	 *    the current consensus?)
1990 	 */
1991 	tstp = LIST_FIRST(&lfp->lf_deleg);
1992 	while (tstp != LIST_END(&lfp->lf_deleg)) {
1993 	    nstp = LIST_NEXT(tstp, ls_file);
1994 	    if ((((new_stp->ls_flags&(NFSLCK_LOCK|NFSLCK_UNLOCK|NFSLCK_TEST))||
1995 		 ((new_stp->ls_flags & NFSLCK_CHECK) &&
1996 		  (new_lop->lo_flags & NFSLCK_READ))) &&
1997 		  clp != tstp->ls_clp &&
1998 		 (tstp->ls_flags & NFSLCK_DELEGWRITE)) ||
1999 		 ((new_stp->ls_flags & NFSLCK_CHECK) &&
2000 		   (new_lop->lo_flags & NFSLCK_WRITE) &&
2001 		  (clp != tstp->ls_clp ||
2002 		   (tstp->ls_flags & NFSLCK_DELEGREAD)))) {
2003 		ret = 0;
2004 		if (filestruct_locked != 0) {
2005 			/* Roll back local locks. */
2006 			NFSUNLOCKSTATE();
2007 			if (vnode_unlocked == 0) {
2008 				ASSERT_VOP_ELOCKED(vp, "nfsrv_lockctrl4");
2009 				NFSVOPUNLOCK(vp, 0);
2010 			}
2011 			nfsrv_locallock_rollback(vp, lfp, p);
2012 			NFSLOCKSTATE();
2013 			nfsrv_unlocklf(lfp);
2014 			NFSUNLOCKSTATE();
2015 			NFSVOPLOCK(vp, LK_EXCLUSIVE | LK_RETRY);
2016 			vnode_unlocked = 0;
2017 			if ((vp->v_iflag & VI_DOOMED) != 0)
2018 				ret = NFSERR_SERVERFAULT;
2019 			NFSLOCKSTATE();
2020 		}
2021 		if (ret == 0)
2022 			ret = nfsrv_delegconflict(tstp, &haslock, p, vp);
2023 		if (ret) {
2024 		    /*
2025 		     * nfsrv_delegconflict unlocks state when it
2026 		     * returns non-zero, which it always does.
2027 		     */
2028 		    if (other_lop) {
2029 			free(other_lop, M_NFSDLOCK);
2030 			other_lop = NULL;
2031 		    }
2032 		    if (ret == -1) {
2033 			lckstp = NULL;
2034 			goto tryagain;
2035 		    }
2036 		    error = ret;
2037 		    goto out;
2038 		}
2039 		/* Never gets here. */
2040 	    }
2041 	    tstp = nstp;
2042 	}
2043 
2044 	/*
2045 	 * Handle the unlock case by calling nfsrv_updatelock().
2046 	 * (Should I have done some access checking above for unlock? For now,
2047 	 *  just let it happen.)
2048 	 */
2049 	if (new_stp->ls_flags & NFSLCK_UNLOCK) {
2050 		first = new_lop->lo_first;
2051 		end = new_lop->lo_end;
2052 		nfsrv_updatelock(stp, new_lopp, &other_lop, lfp);
2053 		stateidp->seqid = ++(stp->ls_stateid.seqid);
2054 		if ((nd->nd_flag & ND_NFSV41) != 0 && stateidp->seqid == 0)
2055 			stateidp->seqid = stp->ls_stateid.seqid = 1;
2056 		stateidp->other[0] = stp->ls_stateid.other[0];
2057 		stateidp->other[1] = stp->ls_stateid.other[1];
2058 		stateidp->other[2] = stp->ls_stateid.other[2];
2059 		if (filestruct_locked != 0) {
2060 			NFSUNLOCKSTATE();
2061 			if (vnode_unlocked == 0) {
2062 				ASSERT_VOP_ELOCKED(vp, "nfsrv_lockctrl5");
2063 				vnode_unlocked = 1;
2064 				NFSVOPUNLOCK(vp, 0);
2065 			}
2066 			/* Update the local locks. */
2067 			nfsrv_localunlock(vp, lfp, first, end, p);
2068 			NFSLOCKSTATE();
2069 			nfsrv_unlocklf(lfp);
2070 		}
2071 		NFSUNLOCKSTATE();
2072 		goto out;
2073 	}
2074 
2075 	/*
2076 	 * Search for a conflicting lock. A lock conflicts if:
2077 	 * - the lock range overlaps and
2078 	 * - at least one lock is a write lock and
2079 	 * - it is not owned by the same lock owner
2080 	 */
2081 	if (!delegation) {
2082 	  LIST_FOREACH(lop, &lfp->lf_lock, lo_lckfile) {
2083 	    if (new_lop->lo_end > lop->lo_first &&
2084 		new_lop->lo_first < lop->lo_end &&
2085 		(new_lop->lo_flags == NFSLCK_WRITE ||
2086 		 lop->lo_flags == NFSLCK_WRITE) &&
2087 		lckstp != lop->lo_stp &&
2088 		(clp != lop->lo_stp->ls_clp ||
2089 		 lckstp->ls_ownerlen != lop->lo_stp->ls_ownerlen ||
2090 		 NFSBCMP(lckstp->ls_owner, lop->lo_stp->ls_owner,
2091 		    lckstp->ls_ownerlen))) {
2092 		if (other_lop) {
2093 		    free(other_lop, M_NFSDLOCK);
2094 		    other_lop = NULL;
2095 		}
2096 		if (vnode_unlocked != 0)
2097 		    ret = nfsrv_clientconflict(lop->lo_stp->ls_clp, &haslock,
2098 			NULL, p);
2099 		else
2100 		    ret = nfsrv_clientconflict(lop->lo_stp->ls_clp, &haslock,
2101 			vp, p);
2102 		if (ret == 1) {
2103 		    if (filestruct_locked != 0) {
2104 			if (vnode_unlocked == 0) {
2105 				ASSERT_VOP_ELOCKED(vp, "nfsrv_lockctrl6");
2106 				NFSVOPUNLOCK(vp, 0);
2107 			}
2108 			/* Roll back local locks. */
2109 			nfsrv_locallock_rollback(vp, lfp, p);
2110 			NFSLOCKSTATE();
2111 			nfsrv_unlocklf(lfp);
2112 			NFSUNLOCKSTATE();
2113 			NFSVOPLOCK(vp, LK_EXCLUSIVE | LK_RETRY);
2114 			vnode_unlocked = 0;
2115 			if ((vp->v_iflag & VI_DOOMED) != 0) {
2116 				error = NFSERR_SERVERFAULT;
2117 				goto out;
2118 			}
2119 		    }
2120 		    /*
2121 		     * nfsrv_clientconflict() unlocks state when it
2122 		     * returns non-zero.
2123 		     */
2124 		    lckstp = NULL;
2125 		    goto tryagain;
2126 		}
2127 		/*
2128 		 * Found a conflicting lock, so record the conflict and
2129 		 * return the error.
2130 		 */
2131 		if (cfp != NULL && ret == 0) {
2132 		    cfp->cl_clientid.lval[0]=lop->lo_stp->ls_stateid.other[0];
2133 		    cfp->cl_clientid.lval[1]=lop->lo_stp->ls_stateid.other[1];
2134 		    cfp->cl_first = lop->lo_first;
2135 		    cfp->cl_end = lop->lo_end;
2136 		    cfp->cl_flags = lop->lo_flags;
2137 		    cfp->cl_ownerlen = lop->lo_stp->ls_ownerlen;
2138 		    NFSBCOPY(lop->lo_stp->ls_owner, cfp->cl_owner,
2139 			cfp->cl_ownerlen);
2140 		}
2141 		if (ret == 2)
2142 		    error = NFSERR_PERM;
2143 		else if (new_stp->ls_flags & NFSLCK_RECLAIM)
2144 		    error = NFSERR_RECLAIMCONFLICT;
2145 		else if (new_stp->ls_flags & NFSLCK_CHECK)
2146 		    error = NFSERR_LOCKED;
2147 		else
2148 		    error = NFSERR_DENIED;
2149 		if (filestruct_locked != 0 && ret == 0) {
2150 			/* Roll back local locks. */
2151 			NFSUNLOCKSTATE();
2152 			if (vnode_unlocked == 0) {
2153 				ASSERT_VOP_ELOCKED(vp, "nfsrv_lockctrl7");
2154 				vnode_unlocked = 1;
2155 				NFSVOPUNLOCK(vp, 0);
2156 			}
2157 			nfsrv_locallock_rollback(vp, lfp, p);
2158 			NFSLOCKSTATE();
2159 			nfsrv_unlocklf(lfp);
2160 		}
2161 		if (ret == 0)
2162 			NFSUNLOCKSTATE();
2163 		goto out;
2164 	    }
2165 	  }
2166 	}
2167 
2168 	/*
2169 	 * We only get here if there was no lock that conflicted.
2170 	 */
2171 	if (new_stp->ls_flags & (NFSLCK_TEST | NFSLCK_CHECK)) {
2172 		NFSUNLOCKSTATE();
2173 		goto out;
2174 	}
2175 
2176 	/*
2177 	 * We only get here when we are creating or modifying a lock.
2178 	 * There are two variants:
2179 	 * - exist_lock_owner where lock_owner exists
2180 	 * - open_to_lock_owner with new lock_owner
2181 	 */
2182 	first = new_lop->lo_first;
2183 	end = new_lop->lo_end;
2184 	lock_flags = new_lop->lo_flags;
2185 	if (!(new_stp->ls_flags & NFSLCK_OPENTOLOCK)) {
2186 		nfsrv_updatelock(lckstp, new_lopp, &other_lop, lfp);
2187 		stateidp->seqid = ++(lckstp->ls_stateid.seqid);
2188 		if ((nd->nd_flag & ND_NFSV41) != 0 && stateidp->seqid == 0)
2189 			stateidp->seqid = lckstp->ls_stateid.seqid = 1;
2190 		stateidp->other[0] = lckstp->ls_stateid.other[0];
2191 		stateidp->other[1] = lckstp->ls_stateid.other[1];
2192 		stateidp->other[2] = lckstp->ls_stateid.other[2];
2193 	} else {
2194 		/*
2195 		 * The new open_to_lock_owner case.
2196 		 * Link the new nfsstate into the lists.
2197 		 */
2198 		new_stp->ls_seq = new_stp->ls_opentolockseq;
2199 		nfsrvd_refcache(new_stp->ls_op);
2200 		stateidp->seqid = new_stp->ls_stateid.seqid = 1;
2201 		stateidp->other[0] = new_stp->ls_stateid.other[0] =
2202 		    clp->lc_clientid.lval[0];
2203 		stateidp->other[1] = new_stp->ls_stateid.other[1] =
2204 		    clp->lc_clientid.lval[1];
2205 		stateidp->other[2] = new_stp->ls_stateid.other[2] =
2206 		    nfsrv_nextstateindex(clp);
2207 		new_stp->ls_clp = clp;
2208 		LIST_INIT(&new_stp->ls_lock);
2209 		new_stp->ls_openstp = stp;
2210 		new_stp->ls_lfp = lfp;
2211 		nfsrv_insertlock(new_lop, (struct nfslock *)new_stp, new_stp,
2212 		    lfp);
2213 		LIST_INSERT_HEAD(NFSSTATEHASH(clp, new_stp->ls_stateid),
2214 		    new_stp, ls_hash);
2215 		LIST_INSERT_HEAD(&stp->ls_open, new_stp, ls_list);
2216 		*new_lopp = NULL;
2217 		*new_stpp = NULL;
2218 		nfsstatsv1.srvlockowners++;
2219 		nfsrv_openpluslock++;
2220 	}
2221 	if (filestruct_locked != 0) {
2222 		NFSUNLOCKSTATE();
2223 		nfsrv_locallock_commit(lfp, lock_flags, first, end);
2224 		NFSLOCKSTATE();
2225 		nfsrv_unlocklf(lfp);
2226 	}
2227 	NFSUNLOCKSTATE();
2228 
2229 out:
2230 	if (haslock) {
2231 		NFSLOCKV4ROOTMUTEX();
2232 		nfsv4_unlock(&nfsv4rootfs_lock, 1);
2233 		NFSUNLOCKV4ROOTMUTEX();
2234 	}
2235 	if (vnode_unlocked != 0) {
2236 		NFSVOPLOCK(vp, LK_EXCLUSIVE | LK_RETRY);
2237 		if (error == 0 && (vp->v_iflag & VI_DOOMED) != 0)
2238 			error = NFSERR_SERVERFAULT;
2239 	}
2240 	if (other_lop)
2241 		free(other_lop, M_NFSDLOCK);
2242 	NFSEXITCODE2(error, nd);
2243 	return (error);
2244 }
2245 
2246 /*
2247  * Check for state errors for Open.
2248  * repstat is passed back out as an error if more critical errors
2249  * are not detected.
2250  */
2251 APPLESTATIC int
2252 nfsrv_opencheck(nfsquad_t clientid, nfsv4stateid_t *stateidp,
2253     struct nfsstate *new_stp, vnode_t vp, struct nfsrv_descript *nd,
2254     NFSPROC_T *p, int repstat)
2255 {
2256 	struct nfsstate *stp, *nstp;
2257 	struct nfsclient *clp;
2258 	struct nfsstate *ownerstp;
2259 	struct nfslockfile *lfp, *new_lfp;
2260 	int error = 0, haslock = 0, ret, readonly = 0, getfhret = 0;
2261 
2262 	if ((new_stp->ls_flags & NFSLCK_SHAREBITS) == NFSLCK_READACCESS)
2263 		readonly = 1;
2264 	/*
2265 	 * Check for restart conditions (client and server).
2266 	 */
2267 	error = nfsrv_checkrestart(clientid, new_stp->ls_flags,
2268 		&new_stp->ls_stateid, 0);
2269 	if (error)
2270 		goto out;
2271 
2272 	/*
2273 	 * Check for state resource limit exceeded.
2274 	 * Technically this should be SMP protected, but the worst
2275 	 * case error is "out by one or two" on the count when it
2276 	 * returns NFSERR_RESOURCE and the limit is just a rather
2277 	 * arbitrary high water mark, so no harm is done.
2278 	 */
2279 	if (nfsrv_openpluslock > nfsrv_v4statelimit) {
2280 		error = NFSERR_RESOURCE;
2281 		goto out;
2282 	}
2283 
2284 tryagain:
2285 	new_lfp = malloc(sizeof (struct nfslockfile),
2286 	    M_NFSDLOCKFILE, M_WAITOK);
2287 	if (vp)
2288 		getfhret = nfsrv_getlockfh(vp, new_stp->ls_flags, new_lfp,
2289 		    NULL, p);
2290 	NFSLOCKSTATE();
2291 	/*
2292 	 * Get the nfsclient structure.
2293 	 */
2294 	error = nfsrv_getclient(clientid, CLOPS_RENEW, &clp, NULL,
2295 	    (nfsquad_t)((u_quad_t)0), 0, nd, p);
2296 
2297 	/*
2298 	 * Look up the open owner. See if it needs confirmation and
2299 	 * check the seq#, as required.
2300 	 */
2301 	if (!error)
2302 		nfsrv_getowner(&clp->lc_open, new_stp, &ownerstp);
2303 
2304 	if (!error && ownerstp) {
2305 		error = nfsrv_checkseqid(nd, new_stp->ls_seq, ownerstp,
2306 		    new_stp->ls_op);
2307 		/*
2308 		 * If the OpenOwner hasn't been confirmed, assume the
2309 		 * old one was a replay and this one is ok.
2310 		 * See: RFC3530 Sec. 14.2.18.
2311 		 */
2312 		if (error == NFSERR_BADSEQID &&
2313 		    (ownerstp->ls_flags & NFSLCK_NEEDSCONFIRM))
2314 			error = 0;
2315 	}
2316 
2317 	/*
2318 	 * Check for grace.
2319 	 */
2320 	if (!error)
2321 		error = nfsrv_checkgrace(nd, clp, new_stp->ls_flags);
2322 	if ((new_stp->ls_flags & NFSLCK_RECLAIM) && !error &&
2323 		nfsrv_checkstable(clp))
2324 		error = NFSERR_NOGRACE;
2325 
2326 	/*
2327 	 * If none of the above errors occurred, let repstat be
2328 	 * returned.
2329 	 */
2330 	if (repstat && !error)
2331 		error = repstat;
2332 	if (error) {
2333 		NFSUNLOCKSTATE();
2334 		if (haslock) {
2335 			NFSLOCKV4ROOTMUTEX();
2336 			nfsv4_unlock(&nfsv4rootfs_lock, 1);
2337 			NFSUNLOCKV4ROOTMUTEX();
2338 		}
2339 		free(new_lfp, M_NFSDLOCKFILE);
2340 		goto out;
2341 	}
2342 
2343 	/*
2344 	 * If vp == NULL, the file doesn't exist yet, so return ok.
2345 	 * (This always happens on the first pass, so haslock must be 0.)
2346 	 */
2347 	if (vp == NULL) {
2348 		NFSUNLOCKSTATE();
2349 		free(new_lfp, M_NFSDLOCKFILE);
2350 		goto out;
2351 	}
2352 
2353 	/*
2354 	 * Get the structure for the underlying file.
2355 	 */
2356 	if (getfhret)
2357 		error = getfhret;
2358 	else
2359 		error = nfsrv_getlockfile(new_stp->ls_flags, &new_lfp, &lfp,
2360 		    NULL, 0);
2361 	if (new_lfp)
2362 		free(new_lfp, M_NFSDLOCKFILE);
2363 	if (error) {
2364 		NFSUNLOCKSTATE();
2365 		if (haslock) {
2366 			NFSLOCKV4ROOTMUTEX();
2367 			nfsv4_unlock(&nfsv4rootfs_lock, 1);
2368 			NFSUNLOCKV4ROOTMUTEX();
2369 		}
2370 		goto out;
2371 	}
2372 
2373 	/*
2374 	 * Search for a conflicting open/share.
2375 	 */
2376 	if (new_stp->ls_flags & NFSLCK_DELEGCUR) {
2377 	    /*
2378 	     * For Delegate_Cur, search for the matching Delegation,
2379 	     * which indicates no conflict.
2380 	     * An old delegation should have been recovered by the
2381 	     * client doing a Claim_DELEGATE_Prev, so I won't let
2382 	     * it match and return NFSERR_EXPIRED. Should I let it
2383 	     * match?
2384 	     */
2385 	    LIST_FOREACH(stp, &lfp->lf_deleg, ls_file) {
2386 		if (!(stp->ls_flags & NFSLCK_OLDDELEG) &&
2387 		    (((nd->nd_flag & ND_NFSV41) != 0 &&
2388 		    stateidp->seqid == 0) ||
2389 		    stateidp->seqid == stp->ls_stateid.seqid) &&
2390 		    !NFSBCMP(stateidp->other, stp->ls_stateid.other,
2391 			  NFSX_STATEIDOTHER))
2392 			break;
2393 	    }
2394 	    if (stp == LIST_END(&lfp->lf_deleg) ||
2395 		((new_stp->ls_flags & NFSLCK_WRITEACCESS) &&
2396 		 (stp->ls_flags & NFSLCK_DELEGREAD))) {
2397 		NFSUNLOCKSTATE();
2398 		if (haslock) {
2399 			NFSLOCKV4ROOTMUTEX();
2400 			nfsv4_unlock(&nfsv4rootfs_lock, 1);
2401 			NFSUNLOCKV4ROOTMUTEX();
2402 		}
2403 		error = NFSERR_EXPIRED;
2404 		goto out;
2405 	    }
2406 	}
2407 
2408 	/*
2409 	 * Check for access/deny bit conflicts. I check for the same
2410 	 * owner as well, in case the client didn't bother.
2411 	 */
2412 	LIST_FOREACH(stp, &lfp->lf_open, ls_file) {
2413 		if (!(new_stp->ls_flags & NFSLCK_DELEGCUR) &&
2414 		    (((new_stp->ls_flags & NFSLCK_ACCESSBITS) &
2415 		      ((stp->ls_flags>>NFSLCK_SHIFT) & NFSLCK_ACCESSBITS))||
2416 		     ((stp->ls_flags & NFSLCK_ACCESSBITS) &
2417 		      ((new_stp->ls_flags>>NFSLCK_SHIFT)&NFSLCK_ACCESSBITS)))){
2418 			ret = nfsrv_clientconflict(stp->ls_clp,&haslock,vp,p);
2419 			if (ret == 1) {
2420 				/*
2421 				 * nfsrv_clientconflict() unlocks
2422 				 * state when it returns non-zero.
2423 				 */
2424 				goto tryagain;
2425 			}
2426 			if (ret == 2)
2427 				error = NFSERR_PERM;
2428 			else if (new_stp->ls_flags & NFSLCK_RECLAIM)
2429 				error = NFSERR_RECLAIMCONFLICT;
2430 			else
2431 				error = NFSERR_SHAREDENIED;
2432 			if (ret == 0)
2433 				NFSUNLOCKSTATE();
2434 			if (haslock) {
2435 				NFSLOCKV4ROOTMUTEX();
2436 				nfsv4_unlock(&nfsv4rootfs_lock, 1);
2437 				NFSUNLOCKV4ROOTMUTEX();
2438 			}
2439 			goto out;
2440 		}
2441 	}
2442 
2443 	/*
2444 	 * Check for a conflicting delegation. If one is found, call
2445 	 * nfsrv_delegconflict() to handle it. If the v4root lock hasn't
2446 	 * been set yet, it will get the lock. Otherwise, it will recall
2447 	 * the delegation. Then, we try try again...
2448 	 * (If NFSLCK_DELEGCUR is set, it has a delegation, so there
2449 	 *  isn't a conflict.)
2450 	 * I currently believe the conflict algorithm to be:
2451 	 * For Open with Read Access and Deny None
2452 	 * - there is a conflict iff a different client has a write delegation
2453 	 * For Open with other Write Access or any Deny except None
2454 	 * - there is a conflict if a different client has any delegation
2455 	 * - there is a conflict if the same client has a read delegation
2456 	 *   (The current consensus is that this last case should be
2457 	 *    considered a conflict since the client with a read delegation
2458 	 *    could have done an Open with ReadAccess and WriteDeny
2459 	 *    locally and then not have checked for the WriteDeny.)
2460 	 * Don't check for a Reclaim, since that will be dealt with
2461 	 * by nfsrv_openctrl().
2462 	 */
2463 	if (!(new_stp->ls_flags &
2464 		(NFSLCK_DELEGPREV | NFSLCK_DELEGCUR | NFSLCK_RECLAIM))) {
2465 	    stp = LIST_FIRST(&lfp->lf_deleg);
2466 	    while (stp != LIST_END(&lfp->lf_deleg)) {
2467 		nstp = LIST_NEXT(stp, ls_file);
2468 		if ((readonly && stp->ls_clp != clp &&
2469 		       (stp->ls_flags & NFSLCK_DELEGWRITE)) ||
2470 		    (!readonly && (stp->ls_clp != clp ||
2471 		         (stp->ls_flags & NFSLCK_DELEGREAD)))) {
2472 			ret = nfsrv_delegconflict(stp, &haslock, p, vp);
2473 			if (ret) {
2474 			    /*
2475 			     * nfsrv_delegconflict() unlocks state
2476 			     * when it returns non-zero.
2477 			     */
2478 			    if (ret == -1)
2479 				goto tryagain;
2480 			    error = ret;
2481 			    goto out;
2482 			}
2483 		}
2484 		stp = nstp;
2485 	    }
2486 	}
2487 	NFSUNLOCKSTATE();
2488 	if (haslock) {
2489 		NFSLOCKV4ROOTMUTEX();
2490 		nfsv4_unlock(&nfsv4rootfs_lock, 1);
2491 		NFSUNLOCKV4ROOTMUTEX();
2492 	}
2493 
2494 out:
2495 	NFSEXITCODE2(error, nd);
2496 	return (error);
2497 }
2498 
2499 /*
2500  * Open control function to create/update open state for an open.
2501  */
2502 APPLESTATIC int
2503 nfsrv_openctrl(struct nfsrv_descript *nd, vnode_t vp,
2504     struct nfsstate **new_stpp, nfsquad_t clientid, nfsv4stateid_t *stateidp,
2505     nfsv4stateid_t *delegstateidp, u_int32_t *rflagsp, struct nfsexstuff *exp,
2506     NFSPROC_T *p, u_quad_t filerev)
2507 {
2508 	struct nfsstate *new_stp = *new_stpp;
2509 	struct nfsstate *stp, *nstp;
2510 	struct nfsstate *openstp = NULL, *new_open, *ownerstp, *new_deleg;
2511 	struct nfslockfile *lfp, *new_lfp;
2512 	struct nfsclient *clp;
2513 	int error = 0, haslock = 0, ret, delegate = 1, writedeleg = 1;
2514 	int readonly = 0, cbret = 1, getfhret = 0;
2515 	int gotstate = 0, len = 0;
2516 	u_char *clidp = NULL;
2517 
2518 	if ((new_stp->ls_flags & NFSLCK_SHAREBITS) == NFSLCK_READACCESS)
2519 		readonly = 1;
2520 	/*
2521 	 * Check for restart conditions (client and server).
2522 	 * (Paranoia, should have been detected by nfsrv_opencheck().)
2523 	 * If an error does show up, return NFSERR_EXPIRED, since the
2524 	 * the seqid# has already been incremented.
2525 	 */
2526 	error = nfsrv_checkrestart(clientid, new_stp->ls_flags,
2527 	    &new_stp->ls_stateid, 0);
2528 	if (error) {
2529 		printf("Nfsd: openctrl unexpected restart err=%d\n",
2530 		    error);
2531 		error = NFSERR_EXPIRED;
2532 		goto out;
2533 	}
2534 
2535 	clidp = malloc(NFSV4_OPAQUELIMIT, M_TEMP, M_WAITOK);
2536 tryagain:
2537 	new_lfp = malloc(sizeof (struct nfslockfile),
2538 	    M_NFSDLOCKFILE, M_WAITOK);
2539 	new_open = malloc(sizeof (struct nfsstate),
2540 	    M_NFSDSTATE, M_WAITOK);
2541 	new_deleg = malloc(sizeof (struct nfsstate),
2542 	    M_NFSDSTATE, M_WAITOK);
2543 	getfhret = nfsrv_getlockfh(vp, new_stp->ls_flags, new_lfp,
2544 	    NULL, p);
2545 	NFSLOCKSTATE();
2546 	/*
2547 	 * Get the client structure. Since the linked lists could be changed
2548 	 * by other nfsd processes if this process does a tsleep(), one of
2549 	 * two things must be done.
2550 	 * 1 - don't tsleep()
2551 	 * or
2552 	 * 2 - get the nfsv4_lock() { indicated by haslock == 1 }
2553 	 *     before using the lists, since this lock stops the other
2554 	 *     nfsd. This should only be used for rare cases, since it
2555 	 *     essentially single threads the nfsd.
2556 	 *     At this time, it is only done for cases where the stable
2557 	 *     storage file must be written prior to completion of state
2558 	 *     expiration.
2559 	 */
2560 	error = nfsrv_getclient(clientid, CLOPS_RENEW, &clp, NULL,
2561 	    (nfsquad_t)((u_quad_t)0), 0, nd, p);
2562 	if (!error && (clp->lc_flags & LCL_NEEDSCBNULL) &&
2563 	    clp->lc_program) {
2564 		/*
2565 		 * This happens on the first open for a client
2566 		 * that supports callbacks.
2567 		 */
2568 		NFSUNLOCKSTATE();
2569 		/*
2570 		 * Although nfsrv_docallback() will sleep, clp won't
2571 		 * go away, since they are only removed when the
2572 		 * nfsv4_lock() has blocked the nfsd threads. The
2573 		 * fields in clp can change, but having multiple
2574 		 * threads do this Null callback RPC should be
2575 		 * harmless.
2576 		 */
2577 		cbret = nfsrv_docallback(clp, NFSV4PROC_CBNULL,
2578 		    NULL, 0, NULL, NULL, NULL, p);
2579 		NFSLOCKSTATE();
2580 		clp->lc_flags &= ~LCL_NEEDSCBNULL;
2581 		if (!cbret)
2582 			clp->lc_flags |= LCL_CALLBACKSON;
2583 	}
2584 
2585 	/*
2586 	 * Look up the open owner. See if it needs confirmation and
2587 	 * check the seq#, as required.
2588 	 */
2589 	if (!error)
2590 		nfsrv_getowner(&clp->lc_open, new_stp, &ownerstp);
2591 
2592 	if (error) {
2593 		NFSUNLOCKSTATE();
2594 		printf("Nfsd: openctrl unexpected state err=%d\n",
2595 			error);
2596 		free(new_lfp, M_NFSDLOCKFILE);
2597 		free(new_open, M_NFSDSTATE);
2598 		free(new_deleg, M_NFSDSTATE);
2599 		if (haslock) {
2600 			NFSLOCKV4ROOTMUTEX();
2601 			nfsv4_unlock(&nfsv4rootfs_lock, 1);
2602 			NFSUNLOCKV4ROOTMUTEX();
2603 		}
2604 		error = NFSERR_EXPIRED;
2605 		goto out;
2606 	}
2607 
2608 	if (new_stp->ls_flags & NFSLCK_RECLAIM)
2609 		nfsrv_markstable(clp);
2610 
2611 	/*
2612 	 * Get the structure for the underlying file.
2613 	 */
2614 	if (getfhret)
2615 		error = getfhret;
2616 	else
2617 		error = nfsrv_getlockfile(new_stp->ls_flags, &new_lfp, &lfp,
2618 		    NULL, 0);
2619 	if (new_lfp)
2620 		free(new_lfp, M_NFSDLOCKFILE);
2621 	if (error) {
2622 		NFSUNLOCKSTATE();
2623 		printf("Nfsd openctrl unexpected getlockfile err=%d\n",
2624 		    error);
2625 		free(new_open, M_NFSDSTATE);
2626 		free(new_deleg, M_NFSDSTATE);
2627 		if (haslock) {
2628 			NFSLOCKV4ROOTMUTEX();
2629 			nfsv4_unlock(&nfsv4rootfs_lock, 1);
2630 			NFSUNLOCKV4ROOTMUTEX();
2631 		}
2632 		goto out;
2633 	}
2634 
2635 	/*
2636 	 * Search for a conflicting open/share.
2637 	 */
2638 	if (new_stp->ls_flags & NFSLCK_DELEGCUR) {
2639 	    /*
2640 	     * For Delegate_Cur, search for the matching Delegation,
2641 	     * which indicates no conflict.
2642 	     * An old delegation should have been recovered by the
2643 	     * client doing a Claim_DELEGATE_Prev, so I won't let
2644 	     * it match and return NFSERR_EXPIRED. Should I let it
2645 	     * match?
2646 	     */
2647 	    LIST_FOREACH(stp, &lfp->lf_deleg, ls_file) {
2648 		if (!(stp->ls_flags & NFSLCK_OLDDELEG) &&
2649 		    (((nd->nd_flag & ND_NFSV41) != 0 &&
2650 		    stateidp->seqid == 0) ||
2651 		    stateidp->seqid == stp->ls_stateid.seqid) &&
2652 		    !NFSBCMP(stateidp->other, stp->ls_stateid.other,
2653 			NFSX_STATEIDOTHER))
2654 			break;
2655 	    }
2656 	    if (stp == LIST_END(&lfp->lf_deleg) ||
2657 		((new_stp->ls_flags & NFSLCK_WRITEACCESS) &&
2658 		 (stp->ls_flags & NFSLCK_DELEGREAD))) {
2659 		NFSUNLOCKSTATE();
2660 		printf("Nfsd openctrl unexpected expiry\n");
2661 		free(new_open, M_NFSDSTATE);
2662 		free(new_deleg, M_NFSDSTATE);
2663 		if (haslock) {
2664 			NFSLOCKV4ROOTMUTEX();
2665 			nfsv4_unlock(&nfsv4rootfs_lock, 1);
2666 			NFSUNLOCKV4ROOTMUTEX();
2667 		}
2668 		error = NFSERR_EXPIRED;
2669 		goto out;
2670 	    }
2671 
2672 	    /*
2673 	     * Don't issue a Delegation, since one already exists and
2674 	     * delay delegation timeout, as required.
2675 	     */
2676 	    delegate = 0;
2677 	    nfsrv_delaydelegtimeout(stp);
2678 	}
2679 
2680 	/*
2681 	 * Check for access/deny bit conflicts. I also check for the
2682 	 * same owner, since the client might not have bothered to check.
2683 	 * Also, note an open for the same file and owner, if found,
2684 	 * which is all we do here for Delegate_Cur, since conflict
2685 	 * checking is already done.
2686 	 */
2687 	LIST_FOREACH(stp, &lfp->lf_open, ls_file) {
2688 		if (ownerstp && stp->ls_openowner == ownerstp)
2689 			openstp = stp;
2690 		if (!(new_stp->ls_flags & NFSLCK_DELEGCUR)) {
2691 		    /*
2692 		     * If another client has the file open, the only
2693 		     * delegation that can be issued is a Read delegation
2694 		     * and only if it is a Read open with Deny none.
2695 		     */
2696 		    if (clp != stp->ls_clp) {
2697 			if ((stp->ls_flags & NFSLCK_SHAREBITS) ==
2698 			    NFSLCK_READACCESS)
2699 			    writedeleg = 0;
2700 			else
2701 			    delegate = 0;
2702 		    }
2703 		    if(((new_stp->ls_flags & NFSLCK_ACCESSBITS) &
2704 		        ((stp->ls_flags>>NFSLCK_SHIFT) & NFSLCK_ACCESSBITS))||
2705 		       ((stp->ls_flags & NFSLCK_ACCESSBITS) &
2706 		        ((new_stp->ls_flags>>NFSLCK_SHIFT)&NFSLCK_ACCESSBITS))){
2707 			ret = nfsrv_clientconflict(stp->ls_clp,&haslock,vp,p);
2708 			if (ret == 1) {
2709 				/*
2710 				 * nfsrv_clientconflict() unlocks state
2711 				 * when it returns non-zero.
2712 				 */
2713 				free(new_open, M_NFSDSTATE);
2714 				free(new_deleg, M_NFSDSTATE);
2715 				openstp = NULL;
2716 				goto tryagain;
2717 			}
2718 			if (ret == 2)
2719 				error = NFSERR_PERM;
2720 			else if (new_stp->ls_flags & NFSLCK_RECLAIM)
2721 				error = NFSERR_RECLAIMCONFLICT;
2722 			else
2723 				error = NFSERR_SHAREDENIED;
2724 			if (ret == 0)
2725 				NFSUNLOCKSTATE();
2726 			if (haslock) {
2727 				NFSLOCKV4ROOTMUTEX();
2728 				nfsv4_unlock(&nfsv4rootfs_lock, 1);
2729 				NFSUNLOCKV4ROOTMUTEX();
2730 			}
2731 			free(new_open, M_NFSDSTATE);
2732 			free(new_deleg, M_NFSDSTATE);
2733 			printf("nfsd openctrl unexpected client cnfl\n");
2734 			goto out;
2735 		    }
2736 		}
2737 	}
2738 
2739 	/*
2740 	 * Check for a conflicting delegation. If one is found, call
2741 	 * nfsrv_delegconflict() to handle it. If the v4root lock hasn't
2742 	 * been set yet, it will get the lock. Otherwise, it will recall
2743 	 * the delegation. Then, we try try again...
2744 	 * (If NFSLCK_DELEGCUR is set, it has a delegation, so there
2745 	 *  isn't a conflict.)
2746 	 * I currently believe the conflict algorithm to be:
2747 	 * For Open with Read Access and Deny None
2748 	 * - there is a conflict iff a different client has a write delegation
2749 	 * For Open with other Write Access or any Deny except None
2750 	 * - there is a conflict if a different client has any delegation
2751 	 * - there is a conflict if the same client has a read delegation
2752 	 *   (The current consensus is that this last case should be
2753 	 *    considered a conflict since the client with a read delegation
2754 	 *    could have done an Open with ReadAccess and WriteDeny
2755 	 *    locally and then not have checked for the WriteDeny.)
2756 	 */
2757 	if (!(new_stp->ls_flags & (NFSLCK_DELEGPREV | NFSLCK_DELEGCUR))) {
2758 	    stp = LIST_FIRST(&lfp->lf_deleg);
2759 	    while (stp != LIST_END(&lfp->lf_deleg)) {
2760 		nstp = LIST_NEXT(stp, ls_file);
2761 		if (stp->ls_clp != clp && (stp->ls_flags & NFSLCK_DELEGREAD))
2762 			writedeleg = 0;
2763 		else
2764 			delegate = 0;
2765 		if ((readonly && stp->ls_clp != clp &&
2766 		       (stp->ls_flags & NFSLCK_DELEGWRITE)) ||
2767 		    (!readonly && (stp->ls_clp != clp ||
2768 		         (stp->ls_flags & NFSLCK_DELEGREAD)))) {
2769 		    if (new_stp->ls_flags & NFSLCK_RECLAIM) {
2770 			delegate = 2;
2771 		    } else {
2772 			ret = nfsrv_delegconflict(stp, &haslock, p, vp);
2773 			if (ret) {
2774 			    /*
2775 			     * nfsrv_delegconflict() unlocks state
2776 			     * when it returns non-zero.
2777 			     */
2778 			    printf("Nfsd openctrl unexpected deleg cnfl\n");
2779 			    free(new_open, M_NFSDSTATE);
2780 			    free(new_deleg, M_NFSDSTATE);
2781 			    if (ret == -1) {
2782 				openstp = NULL;
2783 				goto tryagain;
2784 			    }
2785 			    error = ret;
2786 			    goto out;
2787 			}
2788 		    }
2789 		}
2790 		stp = nstp;
2791 	    }
2792 	}
2793 
2794 	/*
2795 	 * We only get here if there was no open that conflicted.
2796 	 * If an open for the owner exists, or in the access/deny bits.
2797 	 * Otherwise it is a new open. If the open_owner hasn't been
2798 	 * confirmed, replace the open with the new one needing confirmation,
2799 	 * otherwise add the open.
2800 	 */
2801 	if (new_stp->ls_flags & NFSLCK_DELEGPREV) {
2802 	    /*
2803 	     * Handle NFSLCK_DELEGPREV by searching the old delegations for
2804 	     * a match. If found, just move the old delegation to the current
2805 	     * delegation list and issue open. If not found, return
2806 	     * NFSERR_EXPIRED.
2807 	     */
2808 	    LIST_FOREACH(stp, &clp->lc_olddeleg, ls_list) {
2809 		if (stp->ls_lfp == lfp) {
2810 		    /* Found it */
2811 		    if (stp->ls_clp != clp)
2812 			panic("olddeleg clp");
2813 		    LIST_REMOVE(stp, ls_list);
2814 		    LIST_REMOVE(stp, ls_hash);
2815 		    stp->ls_flags &= ~NFSLCK_OLDDELEG;
2816 		    stp->ls_stateid.seqid = delegstateidp->seqid = 1;
2817 		    stp->ls_stateid.other[0] = delegstateidp->other[0] =
2818 			clp->lc_clientid.lval[0];
2819 		    stp->ls_stateid.other[1] = delegstateidp->other[1] =
2820 			clp->lc_clientid.lval[1];
2821 		    stp->ls_stateid.other[2] = delegstateidp->other[2] =
2822 			nfsrv_nextstateindex(clp);
2823 		    stp->ls_compref = nd->nd_compref;
2824 		    LIST_INSERT_HEAD(&clp->lc_deleg, stp, ls_list);
2825 		    LIST_INSERT_HEAD(NFSSTATEHASH(clp,
2826 			stp->ls_stateid), stp, ls_hash);
2827 		    if (stp->ls_flags & NFSLCK_DELEGWRITE)
2828 			*rflagsp |= NFSV4OPEN_WRITEDELEGATE;
2829 		    else
2830 			*rflagsp |= NFSV4OPEN_READDELEGATE;
2831 		    clp->lc_delegtime = NFSD_MONOSEC +
2832 			nfsrv_lease + NFSRV_LEASEDELTA;
2833 
2834 		    /*
2835 		     * Now, do the associated open.
2836 		     */
2837 		    new_open->ls_stateid.seqid = 1;
2838 		    new_open->ls_stateid.other[0] = clp->lc_clientid.lval[0];
2839 		    new_open->ls_stateid.other[1] = clp->lc_clientid.lval[1];
2840 		    new_open->ls_stateid.other[2] = nfsrv_nextstateindex(clp);
2841 		    new_open->ls_flags = (new_stp->ls_flags&NFSLCK_DENYBITS)|
2842 			NFSLCK_OPEN;
2843 		    if (stp->ls_flags & NFSLCK_DELEGWRITE)
2844 			new_open->ls_flags |= (NFSLCK_READACCESS |
2845 			    NFSLCK_WRITEACCESS);
2846 		    else
2847 			new_open->ls_flags |= NFSLCK_READACCESS;
2848 		    new_open->ls_uid = new_stp->ls_uid;
2849 		    new_open->ls_lfp = lfp;
2850 		    new_open->ls_clp = clp;
2851 		    LIST_INIT(&new_open->ls_open);
2852 		    LIST_INSERT_HEAD(&lfp->lf_open, new_open, ls_file);
2853 		    LIST_INSERT_HEAD(NFSSTATEHASH(clp, new_open->ls_stateid),
2854 			new_open, ls_hash);
2855 		    /*
2856 		     * and handle the open owner
2857 		     */
2858 		    if (ownerstp) {
2859 			new_open->ls_openowner = ownerstp;
2860 			LIST_INSERT_HEAD(&ownerstp->ls_open,new_open,ls_list);
2861 		    } else {
2862 			new_open->ls_openowner = new_stp;
2863 			new_stp->ls_flags = 0;
2864 			nfsrvd_refcache(new_stp->ls_op);
2865 			new_stp->ls_noopens = 0;
2866 			LIST_INIT(&new_stp->ls_open);
2867 			LIST_INSERT_HEAD(&new_stp->ls_open, new_open, ls_list);
2868 			LIST_INSERT_HEAD(&clp->lc_open, new_stp, ls_list);
2869 			*new_stpp = NULL;
2870 			nfsstatsv1.srvopenowners++;
2871 			nfsrv_openpluslock++;
2872 		    }
2873 		    openstp = new_open;
2874 		    new_open = NULL;
2875 		    nfsstatsv1.srvopens++;
2876 		    nfsrv_openpluslock++;
2877 		    break;
2878 		}
2879 	    }
2880 	    if (stp == LIST_END(&clp->lc_olddeleg))
2881 		error = NFSERR_EXPIRED;
2882 	} else if (new_stp->ls_flags & (NFSLCK_DELEGREAD | NFSLCK_DELEGWRITE)) {
2883 	    /*
2884 	     * Scan to see that no delegation for this client and file
2885 	     * doesn't already exist.
2886 	     * There also shouldn't yet be an Open for this file and
2887 	     * openowner.
2888 	     */
2889 	    LIST_FOREACH(stp, &lfp->lf_deleg, ls_file) {
2890 		if (stp->ls_clp == clp)
2891 		    break;
2892 	    }
2893 	    if (stp == LIST_END(&lfp->lf_deleg) && openstp == NULL) {
2894 		/*
2895 		 * This is the Claim_Previous case with a delegation
2896 		 * type != Delegate_None.
2897 		 */
2898 		/*
2899 		 * First, add the delegation. (Although we must issue the
2900 		 * delegation, we can also ask for an immediate return.)
2901 		 */
2902 		new_deleg->ls_stateid.seqid = delegstateidp->seqid = 1;
2903 		new_deleg->ls_stateid.other[0] = delegstateidp->other[0] =
2904 		    clp->lc_clientid.lval[0];
2905 		new_deleg->ls_stateid.other[1] = delegstateidp->other[1] =
2906 		    clp->lc_clientid.lval[1];
2907 		new_deleg->ls_stateid.other[2] = delegstateidp->other[2] =
2908 		    nfsrv_nextstateindex(clp);
2909 		if (new_stp->ls_flags & NFSLCK_DELEGWRITE) {
2910 		    new_deleg->ls_flags = (NFSLCK_DELEGWRITE |
2911 			NFSLCK_READACCESS | NFSLCK_WRITEACCESS);
2912 		    *rflagsp |= NFSV4OPEN_WRITEDELEGATE;
2913 		    nfsrv_writedelegcnt++;
2914 		} else {
2915 		    new_deleg->ls_flags = (NFSLCK_DELEGREAD |
2916 			NFSLCK_READACCESS);
2917 		    *rflagsp |= NFSV4OPEN_READDELEGATE;
2918 		}
2919 		new_deleg->ls_uid = new_stp->ls_uid;
2920 		new_deleg->ls_lfp = lfp;
2921 		new_deleg->ls_clp = clp;
2922 		new_deleg->ls_filerev = filerev;
2923 		new_deleg->ls_compref = nd->nd_compref;
2924 		LIST_INSERT_HEAD(&lfp->lf_deleg, new_deleg, ls_file);
2925 		LIST_INSERT_HEAD(NFSSTATEHASH(clp,
2926 		    new_deleg->ls_stateid), new_deleg, ls_hash);
2927 		LIST_INSERT_HEAD(&clp->lc_deleg, new_deleg, ls_list);
2928 		new_deleg = NULL;
2929 		if (delegate == 2 || nfsrv_issuedelegs == 0 ||
2930 		    (clp->lc_flags & (LCL_CALLBACKSON | LCL_CBDOWN)) !=
2931 		     LCL_CALLBACKSON ||
2932 		    NFSRV_V4DELEGLIMIT(nfsrv_delegatecnt) ||
2933 		    !NFSVNO_DELEGOK(vp))
2934 		    *rflagsp |= NFSV4OPEN_RECALL;
2935 		nfsstatsv1.srvdelegates++;
2936 		nfsrv_openpluslock++;
2937 		nfsrv_delegatecnt++;
2938 
2939 		/*
2940 		 * Now, do the associated open.
2941 		 */
2942 		new_open->ls_stateid.seqid = 1;
2943 		new_open->ls_stateid.other[0] = clp->lc_clientid.lval[0];
2944 		new_open->ls_stateid.other[1] = clp->lc_clientid.lval[1];
2945 		new_open->ls_stateid.other[2] = nfsrv_nextstateindex(clp);
2946 		new_open->ls_flags = (new_stp->ls_flags & NFSLCK_DENYBITS) |
2947 		    NFSLCK_OPEN;
2948 		if (new_stp->ls_flags & NFSLCK_DELEGWRITE)
2949 			new_open->ls_flags |= (NFSLCK_READACCESS |
2950 			    NFSLCK_WRITEACCESS);
2951 		else
2952 			new_open->ls_flags |= NFSLCK_READACCESS;
2953 		new_open->ls_uid = new_stp->ls_uid;
2954 		new_open->ls_lfp = lfp;
2955 		new_open->ls_clp = clp;
2956 		LIST_INIT(&new_open->ls_open);
2957 		LIST_INSERT_HEAD(&lfp->lf_open, new_open, ls_file);
2958 		LIST_INSERT_HEAD(NFSSTATEHASH(clp, new_open->ls_stateid),
2959 		   new_open, ls_hash);
2960 		/*
2961 		 * and handle the open owner
2962 		 */
2963 		if (ownerstp) {
2964 		    new_open->ls_openowner = ownerstp;
2965 		    LIST_INSERT_HEAD(&ownerstp->ls_open, new_open, ls_list);
2966 		} else {
2967 		    new_open->ls_openowner = new_stp;
2968 		    new_stp->ls_flags = 0;
2969 		    nfsrvd_refcache(new_stp->ls_op);
2970 		    new_stp->ls_noopens = 0;
2971 		    LIST_INIT(&new_stp->ls_open);
2972 		    LIST_INSERT_HEAD(&new_stp->ls_open, new_open, ls_list);
2973 		    LIST_INSERT_HEAD(&clp->lc_open, new_stp, ls_list);
2974 		    *new_stpp = NULL;
2975 		    nfsstatsv1.srvopenowners++;
2976 		    nfsrv_openpluslock++;
2977 		}
2978 		openstp = new_open;
2979 		new_open = NULL;
2980 		nfsstatsv1.srvopens++;
2981 		nfsrv_openpluslock++;
2982 	    } else {
2983 		error = NFSERR_RECLAIMCONFLICT;
2984 	    }
2985 	} else if (ownerstp) {
2986 		if (ownerstp->ls_flags & NFSLCK_NEEDSCONFIRM) {
2987 		    /* Replace the open */
2988 		    if (ownerstp->ls_op)
2989 			nfsrvd_derefcache(ownerstp->ls_op);
2990 		    ownerstp->ls_op = new_stp->ls_op;
2991 		    nfsrvd_refcache(ownerstp->ls_op);
2992 		    ownerstp->ls_seq = new_stp->ls_seq;
2993 		    *rflagsp |= NFSV4OPEN_RESULTCONFIRM;
2994 		    stp = LIST_FIRST(&ownerstp->ls_open);
2995 		    stp->ls_flags = (new_stp->ls_flags & NFSLCK_SHAREBITS) |
2996 			NFSLCK_OPEN;
2997 		    stp->ls_stateid.seqid = 1;
2998 		    stp->ls_uid = new_stp->ls_uid;
2999 		    if (lfp != stp->ls_lfp) {
3000 			LIST_REMOVE(stp, ls_file);
3001 			LIST_INSERT_HEAD(&lfp->lf_open, stp, ls_file);
3002 			stp->ls_lfp = lfp;
3003 		    }
3004 		    openstp = stp;
3005 		} else if (openstp) {
3006 		    openstp->ls_flags |= (new_stp->ls_flags & NFSLCK_SHAREBITS);
3007 		    openstp->ls_stateid.seqid++;
3008 		    if ((nd->nd_flag & ND_NFSV41) != 0 &&
3009 			openstp->ls_stateid.seqid == 0)
3010 			openstp->ls_stateid.seqid = 1;
3011 
3012 		    /*
3013 		     * This is where we can choose to issue a delegation.
3014 		     */
3015 		    if (delegate == 0 || writedeleg == 0 ||
3016 			NFSVNO_EXRDONLY(exp) || (readonly != 0 &&
3017 			nfsrv_writedelegifpos == 0) ||
3018 			!NFSVNO_DELEGOK(vp) ||
3019 			(new_stp->ls_flags & NFSLCK_WANTRDELEG) != 0 ||
3020 			(clp->lc_flags & (LCL_CALLBACKSON | LCL_CBDOWN)) !=
3021 			 LCL_CALLBACKSON)
3022 			*rflagsp |= NFSV4OPEN_WDCONTENTION;
3023 		    else if (nfsrv_issuedelegs == 0 ||
3024 			NFSRV_V4DELEGLIMIT(nfsrv_delegatecnt))
3025 			*rflagsp |= NFSV4OPEN_WDRESOURCE;
3026 		    else if ((new_stp->ls_flags & NFSLCK_WANTNODELEG) != 0)
3027 			*rflagsp |= NFSV4OPEN_WDNOTWANTED;
3028 		    else {
3029 			new_deleg->ls_stateid.seqid = delegstateidp->seqid = 1;
3030 			new_deleg->ls_stateid.other[0] = delegstateidp->other[0]
3031 			    = clp->lc_clientid.lval[0];
3032 			new_deleg->ls_stateid.other[1] = delegstateidp->other[1]
3033 			    = clp->lc_clientid.lval[1];
3034 			new_deleg->ls_stateid.other[2] = delegstateidp->other[2]
3035 			    = nfsrv_nextstateindex(clp);
3036 			new_deleg->ls_flags = (NFSLCK_DELEGWRITE |
3037 			    NFSLCK_READACCESS | NFSLCK_WRITEACCESS);
3038 			*rflagsp |= NFSV4OPEN_WRITEDELEGATE;
3039 			new_deleg->ls_uid = new_stp->ls_uid;
3040 			new_deleg->ls_lfp = lfp;
3041 			new_deleg->ls_clp = clp;
3042 			new_deleg->ls_filerev = filerev;
3043 			new_deleg->ls_compref = nd->nd_compref;
3044 			nfsrv_writedelegcnt++;
3045 			LIST_INSERT_HEAD(&lfp->lf_deleg, new_deleg, ls_file);
3046 			LIST_INSERT_HEAD(NFSSTATEHASH(clp,
3047 			    new_deleg->ls_stateid), new_deleg, ls_hash);
3048 			LIST_INSERT_HEAD(&clp->lc_deleg, new_deleg, ls_list);
3049 			new_deleg = NULL;
3050 			nfsstatsv1.srvdelegates++;
3051 			nfsrv_openpluslock++;
3052 			nfsrv_delegatecnt++;
3053 		    }
3054 		} else {
3055 		    new_open->ls_stateid.seqid = 1;
3056 		    new_open->ls_stateid.other[0] = clp->lc_clientid.lval[0];
3057 		    new_open->ls_stateid.other[1] = clp->lc_clientid.lval[1];
3058 		    new_open->ls_stateid.other[2] = nfsrv_nextstateindex(clp);
3059 		    new_open->ls_flags = (new_stp->ls_flags & NFSLCK_SHAREBITS)|
3060 			NFSLCK_OPEN;
3061 		    new_open->ls_uid = new_stp->ls_uid;
3062 		    new_open->ls_openowner = ownerstp;
3063 		    new_open->ls_lfp = lfp;
3064 		    new_open->ls_clp = clp;
3065 		    LIST_INIT(&new_open->ls_open);
3066 		    LIST_INSERT_HEAD(&lfp->lf_open, new_open, ls_file);
3067 		    LIST_INSERT_HEAD(&ownerstp->ls_open, new_open, ls_list);
3068 		    LIST_INSERT_HEAD(NFSSTATEHASH(clp, new_open->ls_stateid),
3069 			new_open, ls_hash);
3070 		    openstp = new_open;
3071 		    new_open = NULL;
3072 		    nfsstatsv1.srvopens++;
3073 		    nfsrv_openpluslock++;
3074 
3075 		    /*
3076 		     * This is where we can choose to issue a delegation.
3077 		     */
3078 		    if (delegate == 0 || (writedeleg == 0 && readonly == 0) ||
3079 			!NFSVNO_DELEGOK(vp) ||
3080 			(clp->lc_flags & (LCL_CALLBACKSON | LCL_CBDOWN)) !=
3081 			 LCL_CALLBACKSON)
3082 			*rflagsp |= NFSV4OPEN_WDCONTENTION;
3083 		    else if (nfsrv_issuedelegs == 0 ||
3084 			NFSRV_V4DELEGLIMIT(nfsrv_delegatecnt))
3085 			*rflagsp |= NFSV4OPEN_WDRESOURCE;
3086 		    else if ((new_stp->ls_flags & NFSLCK_WANTNODELEG) != 0)
3087 			*rflagsp |= NFSV4OPEN_WDNOTWANTED;
3088 		    else {
3089 			new_deleg->ls_stateid.seqid = delegstateidp->seqid = 1;
3090 			new_deleg->ls_stateid.other[0] = delegstateidp->other[0]
3091 			    = clp->lc_clientid.lval[0];
3092 			new_deleg->ls_stateid.other[1] = delegstateidp->other[1]
3093 			    = clp->lc_clientid.lval[1];
3094 			new_deleg->ls_stateid.other[2] = delegstateidp->other[2]
3095 			    = nfsrv_nextstateindex(clp);
3096 			if (writedeleg && !NFSVNO_EXRDONLY(exp) &&
3097 			    (nfsrv_writedelegifpos || !readonly) &&
3098 			    (new_stp->ls_flags & NFSLCK_WANTRDELEG) == 0) {
3099 			    new_deleg->ls_flags = (NFSLCK_DELEGWRITE |
3100 				NFSLCK_READACCESS | NFSLCK_WRITEACCESS);
3101 			    *rflagsp |= NFSV4OPEN_WRITEDELEGATE;
3102 			    nfsrv_writedelegcnt++;
3103 			} else {
3104 			    new_deleg->ls_flags = (NFSLCK_DELEGREAD |
3105 				NFSLCK_READACCESS);
3106 			    *rflagsp |= NFSV4OPEN_READDELEGATE;
3107 			}
3108 			new_deleg->ls_uid = new_stp->ls_uid;
3109 			new_deleg->ls_lfp = lfp;
3110 			new_deleg->ls_clp = clp;
3111 			new_deleg->ls_filerev = filerev;
3112 			new_deleg->ls_compref = nd->nd_compref;
3113 			LIST_INSERT_HEAD(&lfp->lf_deleg, new_deleg, ls_file);
3114 			LIST_INSERT_HEAD(NFSSTATEHASH(clp,
3115 			    new_deleg->ls_stateid), new_deleg, ls_hash);
3116 			LIST_INSERT_HEAD(&clp->lc_deleg, new_deleg, ls_list);
3117 			new_deleg = NULL;
3118 			nfsstatsv1.srvdelegates++;
3119 			nfsrv_openpluslock++;
3120 			nfsrv_delegatecnt++;
3121 		    }
3122 		}
3123 	} else {
3124 		/*
3125 		 * New owner case. Start the open_owner sequence with a
3126 		 * Needs confirmation (unless a reclaim) and hang the
3127 		 * new open off it.
3128 		 */
3129 		new_open->ls_stateid.seqid = 1;
3130 		new_open->ls_stateid.other[0] = clp->lc_clientid.lval[0];
3131 		new_open->ls_stateid.other[1] = clp->lc_clientid.lval[1];
3132 		new_open->ls_stateid.other[2] = nfsrv_nextstateindex(clp);
3133 		new_open->ls_flags = (new_stp->ls_flags & NFSLCK_SHAREBITS) |
3134 		    NFSLCK_OPEN;
3135 		new_open->ls_uid = new_stp->ls_uid;
3136 		LIST_INIT(&new_open->ls_open);
3137 		new_open->ls_openowner = new_stp;
3138 		new_open->ls_lfp = lfp;
3139 		new_open->ls_clp = clp;
3140 		LIST_INSERT_HEAD(&lfp->lf_open, new_open, ls_file);
3141 		if (new_stp->ls_flags & NFSLCK_RECLAIM) {
3142 			new_stp->ls_flags = 0;
3143 		} else if ((nd->nd_flag & ND_NFSV41) != 0) {
3144 			/* NFSv4.1 never needs confirmation. */
3145 			new_stp->ls_flags = 0;
3146 
3147 			/*
3148 			 * This is where we can choose to issue a delegation.
3149 			 */
3150 			if (delegate && nfsrv_issuedelegs &&
3151 			    (writedeleg || readonly) &&
3152 			    (clp->lc_flags & (LCL_CALLBACKSON | LCL_CBDOWN)) ==
3153 			     LCL_CALLBACKSON &&
3154 			    !NFSRV_V4DELEGLIMIT(nfsrv_delegatecnt) &&
3155 			    NFSVNO_DELEGOK(vp) &&
3156 			    ((nd->nd_flag & ND_NFSV41) == 0 ||
3157 			     (new_stp->ls_flags & NFSLCK_WANTNODELEG) == 0)) {
3158 				new_deleg->ls_stateid.seqid =
3159 				    delegstateidp->seqid = 1;
3160 				new_deleg->ls_stateid.other[0] =
3161 				    delegstateidp->other[0]
3162 				    = clp->lc_clientid.lval[0];
3163 				new_deleg->ls_stateid.other[1] =
3164 				    delegstateidp->other[1]
3165 				    = clp->lc_clientid.lval[1];
3166 				new_deleg->ls_stateid.other[2] =
3167 				    delegstateidp->other[2]
3168 				    = nfsrv_nextstateindex(clp);
3169 				if (writedeleg && !NFSVNO_EXRDONLY(exp) &&
3170 				    (nfsrv_writedelegifpos || !readonly) &&
3171 				    ((nd->nd_flag & ND_NFSV41) == 0 ||
3172 				     (new_stp->ls_flags & NFSLCK_WANTRDELEG) ==
3173 				     0)) {
3174 					new_deleg->ls_flags =
3175 					    (NFSLCK_DELEGWRITE |
3176 					     NFSLCK_READACCESS |
3177 					     NFSLCK_WRITEACCESS);
3178 					*rflagsp |= NFSV4OPEN_WRITEDELEGATE;
3179 					nfsrv_writedelegcnt++;
3180 				} else {
3181 					new_deleg->ls_flags =
3182 					    (NFSLCK_DELEGREAD |
3183 					     NFSLCK_READACCESS);
3184 					*rflagsp |= NFSV4OPEN_READDELEGATE;
3185 				}
3186 				new_deleg->ls_uid = new_stp->ls_uid;
3187 				new_deleg->ls_lfp = lfp;
3188 				new_deleg->ls_clp = clp;
3189 				new_deleg->ls_filerev = filerev;
3190 				new_deleg->ls_compref = nd->nd_compref;
3191 				LIST_INSERT_HEAD(&lfp->lf_deleg, new_deleg,
3192 				    ls_file);
3193 				LIST_INSERT_HEAD(NFSSTATEHASH(clp,
3194 				    new_deleg->ls_stateid), new_deleg, ls_hash);
3195 				LIST_INSERT_HEAD(&clp->lc_deleg, new_deleg,
3196 				    ls_list);
3197 				new_deleg = NULL;
3198 				nfsstatsv1.srvdelegates++;
3199 				nfsrv_openpluslock++;
3200 				nfsrv_delegatecnt++;
3201 			}
3202 			/*
3203 			 * Since NFSv4.1 never does an OpenConfirm, the first
3204 			 * open state will be acquired here.
3205 			 */
3206 			if (!(clp->lc_flags & LCL_STAMPEDSTABLE)) {
3207 				clp->lc_flags |= LCL_STAMPEDSTABLE;
3208 				len = clp->lc_idlen;
3209 				NFSBCOPY(clp->lc_id, clidp, len);
3210 				gotstate = 1;
3211 			}
3212 		} else {
3213 			*rflagsp |= NFSV4OPEN_RESULTCONFIRM;
3214 			new_stp->ls_flags = NFSLCK_NEEDSCONFIRM;
3215 		}
3216 		nfsrvd_refcache(new_stp->ls_op);
3217 		new_stp->ls_noopens = 0;
3218 		LIST_INIT(&new_stp->ls_open);
3219 		LIST_INSERT_HEAD(&new_stp->ls_open, new_open, ls_list);
3220 		LIST_INSERT_HEAD(&clp->lc_open, new_stp, ls_list);
3221 		LIST_INSERT_HEAD(NFSSTATEHASH(clp, new_open->ls_stateid),
3222 		    new_open, ls_hash);
3223 		openstp = new_open;
3224 		new_open = NULL;
3225 		*new_stpp = NULL;
3226 		nfsstatsv1.srvopens++;
3227 		nfsrv_openpluslock++;
3228 		nfsstatsv1.srvopenowners++;
3229 		nfsrv_openpluslock++;
3230 	}
3231 	if (!error) {
3232 		stateidp->seqid = openstp->ls_stateid.seqid;
3233 		stateidp->other[0] = openstp->ls_stateid.other[0];
3234 		stateidp->other[1] = openstp->ls_stateid.other[1];
3235 		stateidp->other[2] = openstp->ls_stateid.other[2];
3236 	}
3237 	NFSUNLOCKSTATE();
3238 	if (haslock) {
3239 		NFSLOCKV4ROOTMUTEX();
3240 		nfsv4_unlock(&nfsv4rootfs_lock, 1);
3241 		NFSUNLOCKV4ROOTMUTEX();
3242 	}
3243 	if (new_open)
3244 		free(new_open, M_NFSDSTATE);
3245 	if (new_deleg)
3246 		free(new_deleg, M_NFSDSTATE);
3247 
3248 	/*
3249 	 * If the NFSv4.1 client just acquired its first open, write a timestamp
3250 	 * to the stable storage file.
3251 	 */
3252 	if (gotstate != 0) {
3253 		nfsrv_writestable(clidp, len, NFSNST_NEWSTATE, p);
3254 		nfsrv_backupstable();
3255 	}
3256 
3257 out:
3258 	free(clidp, M_TEMP);
3259 	NFSEXITCODE2(error, nd);
3260 	return (error);
3261 }
3262 
3263 /*
3264  * Open update. Does the confirm, downgrade and close.
3265  */
3266 APPLESTATIC int
3267 nfsrv_openupdate(vnode_t vp, struct nfsstate *new_stp, nfsquad_t clientid,
3268     nfsv4stateid_t *stateidp, struct nfsrv_descript *nd, NFSPROC_T *p)
3269 {
3270 	struct nfsstate *stp;
3271 	struct nfsclient *clp;
3272 	struct nfslockfile *lfp;
3273 	u_int32_t bits;
3274 	int error = 0, gotstate = 0, len = 0;
3275 	u_char *clidp = NULL;
3276 
3277 	/*
3278 	 * Check for restart conditions (client and server).
3279 	 */
3280 	error = nfsrv_checkrestart(clientid, new_stp->ls_flags,
3281 	    &new_stp->ls_stateid, 0);
3282 	if (error)
3283 		goto out;
3284 
3285 	clidp = malloc(NFSV4_OPAQUELIMIT, M_TEMP, M_WAITOK);
3286 	NFSLOCKSTATE();
3287 	/*
3288 	 * Get the open structure via clientid and stateid.
3289 	 */
3290 	error = nfsrv_getclient(clientid, CLOPS_RENEW, &clp, NULL,
3291 	    (nfsquad_t)((u_quad_t)0), 0, nd, p);
3292 	if (!error)
3293 		error = nfsrv_getstate(clp, &new_stp->ls_stateid,
3294 		    new_stp->ls_flags, &stp);
3295 
3296 	/*
3297 	 * Sanity check the open.
3298 	 */
3299 	if (!error && (!(stp->ls_flags & NFSLCK_OPEN) ||
3300 		(!(new_stp->ls_flags & NFSLCK_CONFIRM) &&
3301 		 (stp->ls_openowner->ls_flags & NFSLCK_NEEDSCONFIRM)) ||
3302 		((new_stp->ls_flags & NFSLCK_CONFIRM) &&
3303 		 (!(stp->ls_openowner->ls_flags & NFSLCK_NEEDSCONFIRM)))))
3304 		error = NFSERR_BADSTATEID;
3305 
3306 	if (!error)
3307 		error = nfsrv_checkseqid(nd, new_stp->ls_seq,
3308 		    stp->ls_openowner, new_stp->ls_op);
3309 	if (!error && stp->ls_stateid.seqid != new_stp->ls_stateid.seqid &&
3310 	    (((nd->nd_flag & ND_NFSV41) == 0 &&
3311 	      !(new_stp->ls_flags & NFSLCK_CONFIRM)) ||
3312 	     ((nd->nd_flag & ND_NFSV41) != 0 &&
3313 	      new_stp->ls_stateid.seqid != 0)))
3314 		error = NFSERR_OLDSTATEID;
3315 	if (!error && vnode_vtype(vp) != VREG) {
3316 		if (vnode_vtype(vp) == VDIR)
3317 			error = NFSERR_ISDIR;
3318 		else
3319 			error = NFSERR_INVAL;
3320 	}
3321 
3322 	if (error) {
3323 		/*
3324 		 * If a client tries to confirm an Open with a bad
3325 		 * seqid# and there are no byte range locks or other Opens
3326 		 * on the openowner, just throw it away, so the next use of the
3327 		 * openowner will start a fresh seq#.
3328 		 */
3329 		if (error == NFSERR_BADSEQID &&
3330 		    (new_stp->ls_flags & NFSLCK_CONFIRM) &&
3331 		    nfsrv_nootherstate(stp))
3332 			nfsrv_freeopenowner(stp->ls_openowner, 0, p);
3333 		NFSUNLOCKSTATE();
3334 		goto out;
3335 	}
3336 
3337 	/*
3338 	 * Set the return stateid.
3339 	 */
3340 	stateidp->seqid = stp->ls_stateid.seqid + 1;
3341 	if ((nd->nd_flag & ND_NFSV41) != 0 && stateidp->seqid == 0)
3342 		stateidp->seqid = 1;
3343 	stateidp->other[0] = stp->ls_stateid.other[0];
3344 	stateidp->other[1] = stp->ls_stateid.other[1];
3345 	stateidp->other[2] = stp->ls_stateid.other[2];
3346 	/*
3347 	 * Now, handle the three cases.
3348 	 */
3349 	if (new_stp->ls_flags & NFSLCK_CONFIRM) {
3350 		/*
3351 		 * If the open doesn't need confirmation, it seems to me that
3352 		 * there is a client error, but I'll just log it and keep going?
3353 		 */
3354 		if (!(stp->ls_openowner->ls_flags & NFSLCK_NEEDSCONFIRM))
3355 			printf("Nfsv4d: stray open confirm\n");
3356 		stp->ls_openowner->ls_flags = 0;
3357 		stp->ls_stateid.seqid++;
3358 		if ((nd->nd_flag & ND_NFSV41) != 0 &&
3359 		    stp->ls_stateid.seqid == 0)
3360 			stp->ls_stateid.seqid = 1;
3361 		if (!(clp->lc_flags & LCL_STAMPEDSTABLE)) {
3362 			clp->lc_flags |= LCL_STAMPEDSTABLE;
3363 			len = clp->lc_idlen;
3364 			NFSBCOPY(clp->lc_id, clidp, len);
3365 			gotstate = 1;
3366 		}
3367 		NFSUNLOCKSTATE();
3368 	} else if (new_stp->ls_flags & NFSLCK_CLOSE) {
3369 		lfp = stp->ls_lfp;
3370 		if (nfsrv_dolocallocks != 0 && !LIST_EMPTY(&stp->ls_open)) {
3371 			/* Get the lf lock */
3372 			nfsrv_locklf(lfp);
3373 			NFSUNLOCKSTATE();
3374 			ASSERT_VOP_ELOCKED(vp, "nfsrv_openupdate");
3375 			NFSVOPUNLOCK(vp, 0);
3376 			if (nfsrv_freeopen(stp, vp, 1, p) == 0) {
3377 				NFSLOCKSTATE();
3378 				nfsrv_unlocklf(lfp);
3379 				NFSUNLOCKSTATE();
3380 			}
3381 			NFSVOPLOCK(vp, LK_EXCLUSIVE | LK_RETRY);
3382 		} else {
3383 			(void) nfsrv_freeopen(stp, NULL, 0, p);
3384 			NFSUNLOCKSTATE();
3385 		}
3386 	} else {
3387 		/*
3388 		 * Update the share bits, making sure that the new set are a
3389 		 * subset of the old ones.
3390 		 */
3391 		bits = (new_stp->ls_flags & NFSLCK_SHAREBITS);
3392 		if (~(stp->ls_flags) & bits) {
3393 			NFSUNLOCKSTATE();
3394 			error = NFSERR_INVAL;
3395 			goto out;
3396 		}
3397 		stp->ls_flags = (bits | NFSLCK_OPEN);
3398 		stp->ls_stateid.seqid++;
3399 		if ((nd->nd_flag & ND_NFSV41) != 0 &&
3400 		    stp->ls_stateid.seqid == 0)
3401 			stp->ls_stateid.seqid = 1;
3402 		NFSUNLOCKSTATE();
3403 	}
3404 
3405 	/*
3406 	 * If the client just confirmed its first open, write a timestamp
3407 	 * to the stable storage file.
3408 	 */
3409 	if (gotstate != 0) {
3410 		nfsrv_writestable(clidp, len, NFSNST_NEWSTATE, p);
3411 		nfsrv_backupstable();
3412 	}
3413 
3414 out:
3415 	free(clidp, M_TEMP);
3416 	NFSEXITCODE2(error, nd);
3417 	return (error);
3418 }
3419 
3420 /*
3421  * Delegation update. Does the purge and return.
3422  */
3423 APPLESTATIC int
3424 nfsrv_delegupdate(struct nfsrv_descript *nd, nfsquad_t clientid,
3425     nfsv4stateid_t *stateidp, vnode_t vp, int op, struct ucred *cred,
3426     NFSPROC_T *p)
3427 {
3428 	struct nfsstate *stp;
3429 	struct nfsclient *clp;
3430 	int error = 0;
3431 	fhandle_t fh;
3432 
3433 	/*
3434 	 * Do a sanity check against the file handle for DelegReturn.
3435 	 */
3436 	if (vp) {
3437 		error = nfsvno_getfh(vp, &fh, p);
3438 		if (error)
3439 			goto out;
3440 	}
3441 	/*
3442 	 * Check for restart conditions (client and server).
3443 	 */
3444 	if (op == NFSV4OP_DELEGRETURN)
3445 		error = nfsrv_checkrestart(clientid, NFSLCK_DELEGRETURN,
3446 			stateidp, 0);
3447 	else
3448 		error = nfsrv_checkrestart(clientid, NFSLCK_DELEGPURGE,
3449 			stateidp, 0);
3450 
3451 	NFSLOCKSTATE();
3452 	/*
3453 	 * Get the open structure via clientid and stateid.
3454 	 */
3455 	if (!error)
3456 	    error = nfsrv_getclient(clientid, CLOPS_RENEW, &clp, NULL,
3457 		(nfsquad_t)((u_quad_t)0), 0, nd, p);
3458 	if (error) {
3459 		if (error == NFSERR_CBPATHDOWN)
3460 			error = 0;
3461 		if (error == NFSERR_STALECLIENTID && op == NFSV4OP_DELEGRETURN)
3462 			error = NFSERR_STALESTATEID;
3463 	}
3464 	if (!error && op == NFSV4OP_DELEGRETURN) {
3465 	    error = nfsrv_getstate(clp, stateidp, NFSLCK_DELEGRETURN, &stp);
3466 	    if (!error && stp->ls_stateid.seqid != stateidp->seqid &&
3467 		((nd->nd_flag & ND_NFSV41) == 0 || stateidp->seqid != 0))
3468 		error = NFSERR_OLDSTATEID;
3469 	}
3470 	/*
3471 	 * NFSERR_EXPIRED means that the state has gone away,
3472 	 * so Delegations have been purged. Just return ok.
3473 	 */
3474 	if (error == NFSERR_EXPIRED && op == NFSV4OP_DELEGPURGE) {
3475 		NFSUNLOCKSTATE();
3476 		error = 0;
3477 		goto out;
3478 	}
3479 	if (error) {
3480 		NFSUNLOCKSTATE();
3481 		goto out;
3482 	}
3483 
3484 	if (op == NFSV4OP_DELEGRETURN) {
3485 		if (NFSBCMP((caddr_t)&fh, (caddr_t)&stp->ls_lfp->lf_fh,
3486 		    sizeof (fhandle_t))) {
3487 			NFSUNLOCKSTATE();
3488 			error = NFSERR_BADSTATEID;
3489 			goto out;
3490 		}
3491 		nfsrv_freedeleg(stp);
3492 	} else {
3493 		nfsrv_freedeleglist(&clp->lc_olddeleg);
3494 	}
3495 	NFSUNLOCKSTATE();
3496 	error = 0;
3497 
3498 out:
3499 	NFSEXITCODE(error);
3500 	return (error);
3501 }
3502 
3503 /*
3504  * Release lock owner.
3505  */
3506 APPLESTATIC int
3507 nfsrv_releaselckown(struct nfsstate *new_stp, nfsquad_t clientid,
3508     NFSPROC_T *p)
3509 {
3510 	struct nfsstate *stp, *nstp, *openstp, *ownstp;
3511 	struct nfsclient *clp;
3512 	int error = 0;
3513 
3514 	/*
3515 	 * Check for restart conditions (client and server).
3516 	 */
3517 	error = nfsrv_checkrestart(clientid, new_stp->ls_flags,
3518 	    &new_stp->ls_stateid, 0);
3519 	if (error)
3520 		goto out;
3521 
3522 	NFSLOCKSTATE();
3523 	/*
3524 	 * Get the lock owner by name.
3525 	 */
3526 	error = nfsrv_getclient(clientid, CLOPS_RENEW, &clp, NULL,
3527 	    (nfsquad_t)((u_quad_t)0), 0, NULL, p);
3528 	if (error) {
3529 		NFSUNLOCKSTATE();
3530 		goto out;
3531 	}
3532 	LIST_FOREACH(ownstp, &clp->lc_open, ls_list) {
3533 	    LIST_FOREACH(openstp, &ownstp->ls_open, ls_list) {
3534 		stp = LIST_FIRST(&openstp->ls_open);
3535 		while (stp != LIST_END(&openstp->ls_open)) {
3536 		    nstp = LIST_NEXT(stp, ls_list);
3537 		    /*
3538 		     * If the owner matches, check for locks and
3539 		     * then free or return an error.
3540 		     */
3541 		    if (stp->ls_ownerlen == new_stp->ls_ownerlen &&
3542 			!NFSBCMP(stp->ls_owner, new_stp->ls_owner,
3543 			 stp->ls_ownerlen)){
3544 			if (LIST_EMPTY(&stp->ls_lock)) {
3545 			    nfsrv_freelockowner(stp, NULL, 0, p);
3546 			} else {
3547 			    NFSUNLOCKSTATE();
3548 			    error = NFSERR_LOCKSHELD;
3549 			    goto out;
3550 			}
3551 		    }
3552 		    stp = nstp;
3553 		}
3554 	    }
3555 	}
3556 	NFSUNLOCKSTATE();
3557 
3558 out:
3559 	NFSEXITCODE(error);
3560 	return (error);
3561 }
3562 
3563 /*
3564  * Get the file handle for a lock structure.
3565  */
3566 static int
3567 nfsrv_getlockfh(vnode_t vp, u_short flags, struct nfslockfile *new_lfp,
3568     fhandle_t *nfhp, NFSPROC_T *p)
3569 {
3570 	fhandle_t *fhp = NULL;
3571 	int error;
3572 
3573 	/*
3574 	 * For lock, use the new nfslock structure, otherwise just
3575 	 * a fhandle_t on the stack.
3576 	 */
3577 	if (flags & NFSLCK_OPEN) {
3578 		KASSERT(new_lfp != NULL, ("nfsrv_getlockfh: new_lfp NULL"));
3579 		fhp = &new_lfp->lf_fh;
3580 	} else if (nfhp) {
3581 		fhp = nfhp;
3582 	} else {
3583 		panic("nfsrv_getlockfh");
3584 	}
3585 	error = nfsvno_getfh(vp, fhp, p);
3586 	NFSEXITCODE(error);
3587 	return (error);
3588 }
3589 
3590 /*
3591  * Get an nfs lock structure. Allocate one, as required, and return a
3592  * pointer to it.
3593  * Returns an NFSERR_xxx upon failure or -1 to indicate no current lock.
3594  */
3595 static int
3596 nfsrv_getlockfile(u_short flags, struct nfslockfile **new_lfpp,
3597     struct nfslockfile **lfpp, fhandle_t *nfhp, int lockit)
3598 {
3599 	struct nfslockfile *lfp;
3600 	fhandle_t *fhp = NULL, *tfhp;
3601 	struct nfslockhashhead *hp;
3602 	struct nfslockfile *new_lfp = NULL;
3603 
3604 	/*
3605 	 * For lock, use the new nfslock structure, otherwise just
3606 	 * a fhandle_t on the stack.
3607 	 */
3608 	if (flags & NFSLCK_OPEN) {
3609 		new_lfp = *new_lfpp;
3610 		fhp = &new_lfp->lf_fh;
3611 	} else if (nfhp) {
3612 		fhp = nfhp;
3613 	} else {
3614 		panic("nfsrv_getlockfile");
3615 	}
3616 
3617 	hp = NFSLOCKHASH(fhp);
3618 	LIST_FOREACH(lfp, hp, lf_hash) {
3619 		tfhp = &lfp->lf_fh;
3620 		if (NFSVNO_CMPFH(fhp, tfhp)) {
3621 			if (lockit)
3622 				nfsrv_locklf(lfp);
3623 			*lfpp = lfp;
3624 			return (0);
3625 		}
3626 	}
3627 	if (!(flags & NFSLCK_OPEN))
3628 		return (-1);
3629 
3630 	/*
3631 	 * No match, so chain the new one into the list.
3632 	 */
3633 	LIST_INIT(&new_lfp->lf_open);
3634 	LIST_INIT(&new_lfp->lf_lock);
3635 	LIST_INIT(&new_lfp->lf_deleg);
3636 	LIST_INIT(&new_lfp->lf_locallock);
3637 	LIST_INIT(&new_lfp->lf_rollback);
3638 	new_lfp->lf_locallock_lck.nfslock_usecnt = 0;
3639 	new_lfp->lf_locallock_lck.nfslock_lock = 0;
3640 	new_lfp->lf_usecount = 0;
3641 	LIST_INSERT_HEAD(hp, new_lfp, lf_hash);
3642 	*lfpp = new_lfp;
3643 	*new_lfpp = NULL;
3644 	return (0);
3645 }
3646 
3647 /*
3648  * This function adds a nfslock lock structure to the list for the associated
3649  * nfsstate and nfslockfile structures. It will be inserted after the
3650  * entry pointed at by insert_lop.
3651  */
3652 static void
3653 nfsrv_insertlock(struct nfslock *new_lop, struct nfslock *insert_lop,
3654     struct nfsstate *stp, struct nfslockfile *lfp)
3655 {
3656 	struct nfslock *lop, *nlop;
3657 
3658 	new_lop->lo_stp = stp;
3659 	new_lop->lo_lfp = lfp;
3660 
3661 	if (stp != NULL) {
3662 		/* Insert in increasing lo_first order */
3663 		lop = LIST_FIRST(&lfp->lf_lock);
3664 		if (lop == LIST_END(&lfp->lf_lock) ||
3665 		    new_lop->lo_first <= lop->lo_first) {
3666 			LIST_INSERT_HEAD(&lfp->lf_lock, new_lop, lo_lckfile);
3667 		} else {
3668 			nlop = LIST_NEXT(lop, lo_lckfile);
3669 			while (nlop != LIST_END(&lfp->lf_lock) &&
3670 			       nlop->lo_first < new_lop->lo_first) {
3671 				lop = nlop;
3672 				nlop = LIST_NEXT(lop, lo_lckfile);
3673 			}
3674 			LIST_INSERT_AFTER(lop, new_lop, lo_lckfile);
3675 		}
3676 	} else {
3677 		new_lop->lo_lckfile.le_prev = NULL;	/* list not used */
3678 	}
3679 
3680 	/*
3681 	 * Insert after insert_lop, which is overloaded as stp or lfp for
3682 	 * an empty list.
3683 	 */
3684 	if (stp == NULL && (struct nfslockfile *)insert_lop == lfp)
3685 		LIST_INSERT_HEAD(&lfp->lf_locallock, new_lop, lo_lckowner);
3686 	else if ((struct nfsstate *)insert_lop == stp)
3687 		LIST_INSERT_HEAD(&stp->ls_lock, new_lop, lo_lckowner);
3688 	else
3689 		LIST_INSERT_AFTER(insert_lop, new_lop, lo_lckowner);
3690 	if (stp != NULL) {
3691 		nfsstatsv1.srvlocks++;
3692 		nfsrv_openpluslock++;
3693 	}
3694 }
3695 
3696 /*
3697  * This function updates the locking for a lock owner and given file. It
3698  * maintains a list of lock ranges ordered on increasing file offset that
3699  * are NFSLCK_READ or NFSLCK_WRITE and non-overlapping (aka POSIX style).
3700  * It always adds new_lop to the list and sometimes uses the one pointed
3701  * at by other_lopp.
3702  */
3703 static void
3704 nfsrv_updatelock(struct nfsstate *stp, struct nfslock **new_lopp,
3705     struct nfslock **other_lopp, struct nfslockfile *lfp)
3706 {
3707 	struct nfslock *new_lop = *new_lopp;
3708 	struct nfslock *lop, *tlop, *ilop;
3709 	struct nfslock *other_lop = *other_lopp;
3710 	int unlock = 0, myfile = 0;
3711 	u_int64_t tmp;
3712 
3713 	/*
3714 	 * Work down the list until the lock is merged.
3715 	 */
3716 	if (new_lop->lo_flags & NFSLCK_UNLOCK)
3717 		unlock = 1;
3718 	if (stp != NULL) {
3719 		ilop = (struct nfslock *)stp;
3720 		lop = LIST_FIRST(&stp->ls_lock);
3721 	} else {
3722 		ilop = (struct nfslock *)lfp;
3723 		lop = LIST_FIRST(&lfp->lf_locallock);
3724 	}
3725 	while (lop != NULL) {
3726 	    /*
3727 	     * Only check locks for this file that aren't before the start of
3728 	     * new lock's range.
3729 	     */
3730 	    if (lop->lo_lfp == lfp) {
3731 	      myfile = 1;
3732 	      if (lop->lo_end >= new_lop->lo_first) {
3733 		if (new_lop->lo_end < lop->lo_first) {
3734 			/*
3735 			 * If the new lock ends before the start of the
3736 			 * current lock's range, no merge, just insert
3737 			 * the new lock.
3738 			 */
3739 			break;
3740 		}
3741 		if (new_lop->lo_flags == lop->lo_flags ||
3742 		    (new_lop->lo_first <= lop->lo_first &&
3743 		     new_lop->lo_end >= lop->lo_end)) {
3744 			/*
3745 			 * This lock can be absorbed by the new lock/unlock.
3746 			 * This happens when it covers the entire range
3747 			 * of the old lock or is contiguous
3748 			 * with the old lock and is of the same type or an
3749 			 * unlock.
3750 			 */
3751 			if (lop->lo_first < new_lop->lo_first)
3752 				new_lop->lo_first = lop->lo_first;
3753 			if (lop->lo_end > new_lop->lo_end)
3754 				new_lop->lo_end = lop->lo_end;
3755 			tlop = lop;
3756 			lop = LIST_NEXT(lop, lo_lckowner);
3757 			nfsrv_freenfslock(tlop);
3758 			continue;
3759 		}
3760 
3761 		/*
3762 		 * All these cases are for contiguous locks that are not the
3763 		 * same type, so they can't be merged.
3764 		 */
3765 		if (new_lop->lo_first <= lop->lo_first) {
3766 			/*
3767 			 * This case is where the new lock overlaps with the
3768 			 * first part of the old lock. Move the start of the
3769 			 * old lock to just past the end of the new lock. The
3770 			 * new lock will be inserted in front of the old, since
3771 			 * ilop hasn't been updated. (We are done now.)
3772 			 */
3773 			lop->lo_first = new_lop->lo_end;
3774 			break;
3775 		}
3776 		if (new_lop->lo_end >= lop->lo_end) {
3777 			/*
3778 			 * This case is where the new lock overlaps with the
3779 			 * end of the old lock's range. Move the old lock's
3780 			 * end to just before the new lock's first and insert
3781 			 * the new lock after the old lock.
3782 			 * Might not be done yet, since the new lock could
3783 			 * overlap further locks with higher ranges.
3784 			 */
3785 			lop->lo_end = new_lop->lo_first;
3786 			ilop = lop;
3787 			lop = LIST_NEXT(lop, lo_lckowner);
3788 			continue;
3789 		}
3790 		/*
3791 		 * The final case is where the new lock's range is in the
3792 		 * middle of the current lock's and splits the current lock
3793 		 * up. Use *other_lopp to handle the second part of the
3794 		 * split old lock range. (We are done now.)
3795 		 * For unlock, we use new_lop as other_lop and tmp, since
3796 		 * other_lop and new_lop are the same for this case.
3797 		 * We noted the unlock case above, so we don't need
3798 		 * new_lop->lo_flags any longer.
3799 		 */
3800 		tmp = new_lop->lo_first;
3801 		if (other_lop == NULL) {
3802 			if (!unlock)
3803 				panic("nfsd srv update unlock");
3804 			other_lop = new_lop;
3805 			*new_lopp = NULL;
3806 		}
3807 		other_lop->lo_first = new_lop->lo_end;
3808 		other_lop->lo_end = lop->lo_end;
3809 		other_lop->lo_flags = lop->lo_flags;
3810 		other_lop->lo_stp = stp;
3811 		other_lop->lo_lfp = lfp;
3812 		lop->lo_end = tmp;
3813 		nfsrv_insertlock(other_lop, lop, stp, lfp);
3814 		*other_lopp = NULL;
3815 		ilop = lop;
3816 		break;
3817 	      }
3818 	    }
3819 	    ilop = lop;
3820 	    lop = LIST_NEXT(lop, lo_lckowner);
3821 	    if (myfile && (lop == NULL || lop->lo_lfp != lfp))
3822 		break;
3823 	}
3824 
3825 	/*
3826 	 * Insert the new lock in the list at the appropriate place.
3827 	 */
3828 	if (!unlock) {
3829 		nfsrv_insertlock(new_lop, ilop, stp, lfp);
3830 		*new_lopp = NULL;
3831 	}
3832 }
3833 
3834 /*
3835  * This function handles sequencing of locks, etc.
3836  * It returns an error that indicates what the caller should do.
3837  */
3838 static int
3839 nfsrv_checkseqid(struct nfsrv_descript *nd, u_int32_t seqid,
3840     struct nfsstate *stp, struct nfsrvcache *op)
3841 {
3842 	int error = 0;
3843 
3844 	if ((nd->nd_flag & ND_NFSV41) != 0)
3845 		/* NFSv4.1 ignores the open_seqid and lock_seqid. */
3846 		goto out;
3847 	if (op != nd->nd_rp)
3848 		panic("nfsrvstate checkseqid");
3849 	if (!(op->rc_flag & RC_INPROG))
3850 		panic("nfsrvstate not inprog");
3851 	if (stp->ls_op && stp->ls_op->rc_refcnt <= 0) {
3852 		printf("refcnt=%d\n", stp->ls_op->rc_refcnt);
3853 		panic("nfsrvstate op refcnt");
3854 	}
3855 	if ((stp->ls_seq + 1) == seqid) {
3856 		if (stp->ls_op)
3857 			nfsrvd_derefcache(stp->ls_op);
3858 		stp->ls_op = op;
3859 		nfsrvd_refcache(op);
3860 		stp->ls_seq = seqid;
3861 		goto out;
3862 	} else if (stp->ls_seq == seqid && stp->ls_op &&
3863 		op->rc_xid == stp->ls_op->rc_xid &&
3864 		op->rc_refcnt == 0 &&
3865 		op->rc_reqlen == stp->ls_op->rc_reqlen &&
3866 		op->rc_cksum == stp->ls_op->rc_cksum) {
3867 		if (stp->ls_op->rc_flag & RC_INPROG) {
3868 			error = NFSERR_DONTREPLY;
3869 			goto out;
3870 		}
3871 		nd->nd_rp = stp->ls_op;
3872 		nd->nd_rp->rc_flag |= RC_INPROG;
3873 		nfsrvd_delcache(op);
3874 		error = NFSERR_REPLYFROMCACHE;
3875 		goto out;
3876 	}
3877 	error = NFSERR_BADSEQID;
3878 
3879 out:
3880 	NFSEXITCODE2(error, nd);
3881 	return (error);
3882 }
3883 
3884 /*
3885  * Get the client ip address for callbacks. If the strings can't be parsed,
3886  * just set lc_program to 0 to indicate no callbacks are possible.
3887  * (For cases where the address can't be parsed or is 0.0.0.0.0.0, set
3888  *  the address to the client's transport address. This won't be used
3889  *  for callbacks, but can be printed out by nfsstats for info.)
3890  * Return error if the xdr can't be parsed, 0 otherwise.
3891  */
3892 APPLESTATIC int
3893 nfsrv_getclientipaddr(struct nfsrv_descript *nd, struct nfsclient *clp)
3894 {
3895 	u_int32_t *tl;
3896 	u_char *cp, *cp2;
3897 	int i, j;
3898 	struct sockaddr_in *rad, *sad;
3899 	u_char protocol[5], addr[24];
3900 	int error = 0, cantparse = 0;
3901 	union {
3902 		in_addr_t ival;
3903 		u_char cval[4];
3904 	} ip;
3905 	union {
3906 		in_port_t sval;
3907 		u_char cval[2];
3908 	} port;
3909 
3910 	rad = NFSSOCKADDR(clp->lc_req.nr_nam, struct sockaddr_in *);
3911 	rad->sin_family = AF_INET;
3912 	rad->sin_len = sizeof (struct sockaddr_in);
3913 	rad->sin_addr.s_addr = 0;
3914 	rad->sin_port = 0;
3915 	clp->lc_req.nr_client = NULL;
3916 	clp->lc_req.nr_lock = 0;
3917 	NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
3918 	i = fxdr_unsigned(int, *tl);
3919 	if (i >= 3 && i <= 4) {
3920 		error = nfsrv_mtostr(nd, protocol, i);
3921 		if (error)
3922 			goto nfsmout;
3923 		if (!strcmp(protocol, "tcp")) {
3924 			clp->lc_flags |= LCL_TCPCALLBACK;
3925 			clp->lc_req.nr_sotype = SOCK_STREAM;
3926 			clp->lc_req.nr_soproto = IPPROTO_TCP;
3927 		} else if (!strcmp(protocol, "udp")) {
3928 			clp->lc_req.nr_sotype = SOCK_DGRAM;
3929 			clp->lc_req.nr_soproto = IPPROTO_UDP;
3930 		} else {
3931 			cantparse = 1;
3932 		}
3933 	} else {
3934 		cantparse = 1;
3935 		if (i > 0) {
3936 			error = nfsm_advance(nd, NFSM_RNDUP(i), -1);
3937 			if (error)
3938 				goto nfsmout;
3939 		}
3940 	}
3941 	NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
3942 	i = fxdr_unsigned(int, *tl);
3943 	if (i < 0) {
3944 		error = NFSERR_BADXDR;
3945 		goto nfsmout;
3946 	} else if (i == 0) {
3947 		cantparse = 1;
3948 	} else if (!cantparse && i <= 23 && i >= 11) {
3949 		error = nfsrv_mtostr(nd, addr, i);
3950 		if (error)
3951 			goto nfsmout;
3952 
3953 		/*
3954 		 * Parse out the address fields. We expect 6 decimal numbers
3955 		 * separated by '.'s.
3956 		 */
3957 		cp = addr;
3958 		i = 0;
3959 		while (*cp && i < 6) {
3960 			cp2 = cp;
3961 			while (*cp2 && *cp2 != '.')
3962 				cp2++;
3963 			if (*cp2)
3964 				*cp2++ = '\0';
3965 			else if (i != 5) {
3966 				cantparse = 1;
3967 				break;
3968 			}
3969 			j = nfsrv_getipnumber(cp);
3970 			if (j >= 0) {
3971 				if (i < 4)
3972 					ip.cval[3 - i] = j;
3973 				else
3974 					port.cval[5 - i] = j;
3975 			} else {
3976 				cantparse = 1;
3977 				break;
3978 			}
3979 			cp = cp2;
3980 			i++;
3981 		}
3982 		if (!cantparse) {
3983 			if (ip.ival != 0x0) {
3984 				rad->sin_addr.s_addr = htonl(ip.ival);
3985 				rad->sin_port = htons(port.sval);
3986 			} else {
3987 				cantparse = 1;
3988 			}
3989 		}
3990 	} else {
3991 		cantparse = 1;
3992 		if (i > 0) {
3993 			error = nfsm_advance(nd, NFSM_RNDUP(i), -1);
3994 			if (error)
3995 				goto nfsmout;
3996 		}
3997 	}
3998 	if (cantparse) {
3999 		sad = NFSSOCKADDR(nd->nd_nam, struct sockaddr_in *);
4000 		if (sad->sin_family == AF_INET) {
4001 			rad->sin_addr.s_addr = sad->sin_addr.s_addr;
4002 			rad->sin_port = 0x0;
4003 		}
4004 		clp->lc_program = 0;
4005 	}
4006 nfsmout:
4007 	NFSEXITCODE2(error, nd);
4008 	return (error);
4009 }
4010 
4011 /*
4012  * Turn a string of up to three decimal digits into a number. Return -1 upon
4013  * error.
4014  */
4015 static int
4016 nfsrv_getipnumber(u_char *cp)
4017 {
4018 	int i = 0, j = 0;
4019 
4020 	while (*cp) {
4021 		if (j > 2 || *cp < '0' || *cp > '9')
4022 			return (-1);
4023 		i *= 10;
4024 		i += (*cp - '0');
4025 		cp++;
4026 		j++;
4027 	}
4028 	if (i < 256)
4029 		return (i);
4030 	return (-1);
4031 }
4032 
4033 /*
4034  * This function checks for restart conditions.
4035  */
4036 static int
4037 nfsrv_checkrestart(nfsquad_t clientid, u_int32_t flags,
4038     nfsv4stateid_t *stateidp, int specialid)
4039 {
4040 	int ret = 0;
4041 
4042 	/*
4043 	 * First check for a server restart. Open, LockT, ReleaseLockOwner
4044 	 * and DelegPurge have a clientid, the rest a stateid.
4045 	 */
4046 	if (flags &
4047 	    (NFSLCK_OPEN | NFSLCK_TEST | NFSLCK_RELEASE | NFSLCK_DELEGPURGE)) {
4048 		if (clientid.lval[0] != nfsrvboottime) {
4049 			ret = NFSERR_STALECLIENTID;
4050 			goto out;
4051 		}
4052 	} else if (stateidp->other[0] != nfsrvboottime &&
4053 		specialid == 0) {
4054 		ret = NFSERR_STALESTATEID;
4055 		goto out;
4056 	}
4057 
4058 	/*
4059 	 * Read, Write, Setattr and LockT can return NFSERR_GRACE and do
4060 	 * not use a lock/open owner seqid#, so the check can be done now.
4061 	 * (The others will be checked, as required, later.)
4062 	 */
4063 	if (!(flags & (NFSLCK_CHECK | NFSLCK_TEST)))
4064 		goto out;
4065 
4066 	NFSLOCKSTATE();
4067 	ret = nfsrv_checkgrace(NULL, NULL, flags);
4068 	NFSUNLOCKSTATE();
4069 
4070 out:
4071 	NFSEXITCODE(ret);
4072 	return (ret);
4073 }
4074 
4075 /*
4076  * Check for grace.
4077  */
4078 static int
4079 nfsrv_checkgrace(struct nfsrv_descript *nd, struct nfsclient *clp,
4080     u_int32_t flags)
4081 {
4082 	int error = 0;
4083 
4084 	if ((nfsrv_stablefirst.nsf_flags & NFSNSF_GRACEOVER) != 0) {
4085 		if (flags & NFSLCK_RECLAIM) {
4086 			error = NFSERR_NOGRACE;
4087 			goto out;
4088 		}
4089 	} else {
4090 		if (!(flags & NFSLCK_RECLAIM)) {
4091 			error = NFSERR_GRACE;
4092 			goto out;
4093 		}
4094 		if (nd != NULL && clp != NULL &&
4095 		    (nd->nd_flag & ND_NFSV41) != 0 &&
4096 		    (clp->lc_flags & LCL_RECLAIMCOMPLETE) != 0) {
4097 			error = NFSERR_NOGRACE;
4098 			goto out;
4099 		}
4100 
4101 		/*
4102 		 * If grace is almost over and we are still getting Reclaims,
4103 		 * extend grace a bit.
4104 		 */
4105 		if ((NFSD_MONOSEC + NFSRV_LEASEDELTA) >
4106 		    nfsrv_stablefirst.nsf_eograce)
4107 			nfsrv_stablefirst.nsf_eograce = NFSD_MONOSEC +
4108 				NFSRV_LEASEDELTA;
4109 	}
4110 
4111 out:
4112 	NFSEXITCODE(error);
4113 	return (error);
4114 }
4115 
4116 /*
4117  * Do a server callback.
4118  */
4119 static int
4120 nfsrv_docallback(struct nfsclient *clp, int procnum,
4121     nfsv4stateid_t *stateidp, int trunc, fhandle_t *fhp,
4122     struct nfsvattr *nap, nfsattrbit_t *attrbitp, NFSPROC_T *p)
4123 {
4124 	mbuf_t m;
4125 	u_int32_t *tl;
4126 	struct nfsrv_descript nfsd, *nd = &nfsd;
4127 	struct ucred *cred;
4128 	int error = 0;
4129 	u_int32_t callback;
4130 	struct nfsdsession *sep = NULL;
4131 
4132 	cred = newnfs_getcred();
4133 	NFSLOCKSTATE();	/* mostly for lc_cbref++ */
4134 	if (clp->lc_flags & LCL_NEEDSCONFIRM) {
4135 		NFSUNLOCKSTATE();
4136 		panic("docallb");
4137 	}
4138 	clp->lc_cbref++;
4139 
4140 	/*
4141 	 * Fill the callback program# and version into the request
4142 	 * structure for newnfs_connect() to use.
4143 	 */
4144 	clp->lc_req.nr_prog = clp->lc_program;
4145 #ifdef notnow
4146 	if ((clp->lc_flags & LCL_NFSV41) != 0)
4147 		clp->lc_req.nr_vers = NFSV41_CBVERS;
4148 	else
4149 #endif
4150 		clp->lc_req.nr_vers = NFSV4_CBVERS;
4151 
4152 	/*
4153 	 * First, fill in some of the fields of nd and cr.
4154 	 */
4155 	nd->nd_flag = ND_NFSV4;
4156 	if (clp->lc_flags & LCL_GSS)
4157 		nd->nd_flag |= ND_KERBV;
4158 	if ((clp->lc_flags & LCL_NFSV41) != 0)
4159 		nd->nd_flag |= ND_NFSV41;
4160 	nd->nd_repstat = 0;
4161 	cred->cr_uid = clp->lc_uid;
4162 	cred->cr_gid = clp->lc_gid;
4163 	callback = clp->lc_callback;
4164 	NFSUNLOCKSTATE();
4165 	cred->cr_ngroups = 1;
4166 
4167 	/*
4168 	 * Get the first mbuf for the request.
4169 	 */
4170 	MGET(m, M_WAITOK, MT_DATA);
4171 	mbuf_setlen(m, 0);
4172 	nd->nd_mreq = nd->nd_mb = m;
4173 	nd->nd_bpos = NFSMTOD(m, caddr_t);
4174 
4175 	/*
4176 	 * and build the callback request.
4177 	 */
4178 	if (procnum == NFSV4OP_CBGETATTR) {
4179 		nd->nd_procnum = NFSV4PROC_CBCOMPOUND;
4180 		error = nfsrv_cbcallargs(nd, clp, callback, NFSV4OP_CBGETATTR,
4181 		    "CB Getattr", &sep);
4182 		if (error != 0) {
4183 			mbuf_freem(nd->nd_mreq);
4184 			goto errout;
4185 		}
4186 		(void)nfsm_fhtom(nd, (u_int8_t *)fhp, NFSX_MYFH, 0);
4187 		(void)nfsrv_putattrbit(nd, attrbitp);
4188 	} else if (procnum == NFSV4OP_CBRECALL) {
4189 		nd->nd_procnum = NFSV4PROC_CBCOMPOUND;
4190 		error = nfsrv_cbcallargs(nd, clp, callback, NFSV4OP_CBRECALL,
4191 		    "CB Recall", &sep);
4192 		if (error != 0) {
4193 			mbuf_freem(nd->nd_mreq);
4194 			goto errout;
4195 		}
4196 		NFSM_BUILD(tl, u_int32_t *, NFSX_UNSIGNED + NFSX_STATEID);
4197 		*tl++ = txdr_unsigned(stateidp->seqid);
4198 		NFSBCOPY((caddr_t)stateidp->other, (caddr_t)tl,
4199 		    NFSX_STATEIDOTHER);
4200 		tl += (NFSX_STATEIDOTHER / NFSX_UNSIGNED);
4201 		if (trunc)
4202 			*tl = newnfs_true;
4203 		else
4204 			*tl = newnfs_false;
4205 		(void)nfsm_fhtom(nd, (u_int8_t *)fhp, NFSX_MYFH, 0);
4206 	} else if (procnum == NFSV4PROC_CBNULL) {
4207 		nd->nd_procnum = NFSV4PROC_CBNULL;
4208 		if ((clp->lc_flags & LCL_NFSV41) != 0) {
4209 			error = nfsv4_getcbsession(clp, &sep);
4210 			if (error != 0) {
4211 				mbuf_freem(nd->nd_mreq);
4212 				goto errout;
4213 			}
4214 		}
4215 	} else {
4216 		error = NFSERR_SERVERFAULT;
4217 		mbuf_freem(nd->nd_mreq);
4218 		goto errout;
4219 	}
4220 
4221 	/*
4222 	 * Call newnfs_connect(), as required, and then newnfs_request().
4223 	 */
4224 	(void) newnfs_sndlock(&clp->lc_req.nr_lock);
4225 	if (clp->lc_req.nr_client == NULL) {
4226 		if ((clp->lc_flags & LCL_NFSV41) != 0)
4227 			error = ECONNREFUSED;
4228 		else if (nd->nd_procnum == NFSV4PROC_CBNULL)
4229 			error = newnfs_connect(NULL, &clp->lc_req, cred,
4230 			    NULL, 1);
4231 		else
4232 			error = newnfs_connect(NULL, &clp->lc_req, cred,
4233 			    NULL, 3);
4234 	}
4235 	newnfs_sndunlock(&clp->lc_req.nr_lock);
4236 	if (!error) {
4237 		if ((nd->nd_flag & ND_NFSV41) != 0) {
4238 			KASSERT(sep != NULL, ("sep NULL"));
4239 			if (sep->sess_cbsess.nfsess_xprt != NULL)
4240 				error = newnfs_request(nd, NULL, clp,
4241 				    &clp->lc_req, NULL, NULL, cred,
4242 				    clp->lc_program, clp->lc_req.nr_vers, NULL,
4243 				    1, NULL, &sep->sess_cbsess);
4244 			else {
4245 				/*
4246 				 * This should probably never occur, but if a
4247 				 * client somehow does an RPC without a
4248 				 * SequenceID Op that causes a callback just
4249 				 * after the nfsd threads have been terminated
4250 				 * and restared we could conceivably get here
4251 				 * without a backchannel xprt.
4252 				 */
4253 				printf("nfsrv_docallback: no xprt\n");
4254 				error = ECONNREFUSED;
4255 			}
4256 			nfsrv_freesession(sep, NULL);
4257 		} else
4258 			error = newnfs_request(nd, NULL, clp, &clp->lc_req,
4259 			    NULL, NULL, cred, clp->lc_program,
4260 			    clp->lc_req.nr_vers, NULL, 1, NULL, NULL);
4261 	}
4262 errout:
4263 	NFSFREECRED(cred);
4264 
4265 	/*
4266 	 * If error is set here, the Callback path isn't working
4267 	 * properly, so twiddle the appropriate LCL_ flags.
4268 	 * (nd_repstat != 0 indicates the Callback path is working,
4269 	 *  but the callback failed on the client.)
4270 	 */
4271 	if (error) {
4272 		/*
4273 		 * Mark the callback pathway down, which disabled issuing
4274 		 * of delegations and gets Renew to return NFSERR_CBPATHDOWN.
4275 		 */
4276 		NFSLOCKSTATE();
4277 		clp->lc_flags |= LCL_CBDOWN;
4278 		NFSUNLOCKSTATE();
4279 	} else {
4280 		/*
4281 		 * Callback worked. If the callback path was down, disable
4282 		 * callbacks, so no more delegations will be issued. (This
4283 		 * is done on the assumption that the callback pathway is
4284 		 * flakey.)
4285 		 */
4286 		NFSLOCKSTATE();
4287 		if (clp->lc_flags & LCL_CBDOWN)
4288 			clp->lc_flags &= ~(LCL_CBDOWN | LCL_CALLBACKSON);
4289 		NFSUNLOCKSTATE();
4290 		if (nd->nd_repstat)
4291 			error = nd->nd_repstat;
4292 		else if (error == 0 && procnum == NFSV4OP_CBGETATTR)
4293 			error = nfsv4_loadattr(nd, NULL, nap, NULL, NULL, 0,
4294 			    NULL, NULL, NULL, NULL, NULL, 0, NULL, NULL, NULL,
4295 			    p, NULL);
4296 		mbuf_freem(nd->nd_mrep);
4297 	}
4298 	NFSLOCKSTATE();
4299 	clp->lc_cbref--;
4300 	if ((clp->lc_flags & LCL_WAKEUPWANTED) && clp->lc_cbref == 0) {
4301 		clp->lc_flags &= ~LCL_WAKEUPWANTED;
4302 		wakeup(clp);
4303 	}
4304 	NFSUNLOCKSTATE();
4305 
4306 	NFSEXITCODE(error);
4307 	return (error);
4308 }
4309 
4310 /*
4311  * Set up the compound RPC for the callback.
4312  */
4313 static int
4314 nfsrv_cbcallargs(struct nfsrv_descript *nd, struct nfsclient *clp,
4315     uint32_t callback, int op, const char *optag, struct nfsdsession **sepp)
4316 {
4317 	uint32_t *tl;
4318 	int error, len;
4319 
4320 	len = strlen(optag);
4321 	(void)nfsm_strtom(nd, optag, len);
4322 	NFSM_BUILD(tl, uint32_t *, 4 * NFSX_UNSIGNED);
4323 	if ((nd->nd_flag & ND_NFSV41) != 0) {
4324 		*tl++ = txdr_unsigned(NFSV41_MINORVERSION);
4325 		*tl++ = txdr_unsigned(callback);
4326 		*tl++ = txdr_unsigned(2);
4327 		*tl = txdr_unsigned(NFSV4OP_CBSEQUENCE);
4328 		error = nfsv4_setcbsequence(nd, clp, 1, sepp);
4329 		if (error != 0)
4330 			return (error);
4331 		NFSM_BUILD(tl, u_int32_t *, NFSX_UNSIGNED);
4332 		*tl = txdr_unsigned(op);
4333 	} else {
4334 		*tl++ = txdr_unsigned(NFSV4_MINORVERSION);
4335 		*tl++ = txdr_unsigned(callback);
4336 		*tl++ = txdr_unsigned(1);
4337 		*tl = txdr_unsigned(op);
4338 	}
4339 	return (0);
4340 }
4341 
4342 /*
4343  * Return the next index# for a clientid. Mostly just increment and return
4344  * the next one, but... if the 32bit unsigned does actually wrap around,
4345  * it should be rebooted.
4346  * At an average rate of one new client per second, it will wrap around in
4347  * approximately 136 years. (I think the server will have been shut
4348  * down or rebooted before then.)
4349  */
4350 static u_int32_t
4351 nfsrv_nextclientindex(void)
4352 {
4353 	static u_int32_t client_index = 0;
4354 
4355 	client_index++;
4356 	if (client_index != 0)
4357 		return (client_index);
4358 
4359 	printf("%s: out of clientids\n", __func__);
4360 	return (client_index);
4361 }
4362 
4363 /*
4364  * Return the next index# for a stateid. Mostly just increment and return
4365  * the next one, but... if the 32bit unsigned does actually wrap around
4366  * (will a BSD server stay up that long?), find
4367  * new start and end values.
4368  */
4369 static u_int32_t
4370 nfsrv_nextstateindex(struct nfsclient *clp)
4371 {
4372 	struct nfsstate *stp;
4373 	int i;
4374 	u_int32_t canuse, min_index, max_index;
4375 
4376 	if (!(clp->lc_flags & LCL_INDEXNOTOK)) {
4377 		clp->lc_stateindex++;
4378 		if (clp->lc_stateindex != clp->lc_statemaxindex)
4379 			return (clp->lc_stateindex);
4380 	}
4381 
4382 	/*
4383 	 * Yuck, we've hit the end.
4384 	 * Look for a new min and max.
4385 	 */
4386 	min_index = 0;
4387 	max_index = 0xffffffff;
4388 	for (i = 0; i < nfsrv_statehashsize; i++) {
4389 	    LIST_FOREACH(stp, &clp->lc_stateid[i], ls_hash) {
4390 		if (stp->ls_stateid.other[2] > 0x80000000) {
4391 		    if (stp->ls_stateid.other[2] < max_index)
4392 			max_index = stp->ls_stateid.other[2];
4393 		} else {
4394 		    if (stp->ls_stateid.other[2] > min_index)
4395 			min_index = stp->ls_stateid.other[2];
4396 		}
4397 	    }
4398 	}
4399 
4400 	/*
4401 	 * Yikes, highly unlikely, but I'll handle it anyhow.
4402 	 */
4403 	if (min_index == 0x80000000 && max_index == 0x80000001) {
4404 	    canuse = 0;
4405 	    /*
4406 	     * Loop around until we find an unused entry. Return that
4407 	     * and set LCL_INDEXNOTOK, so the search will continue next time.
4408 	     * (This is one of those rare cases where a goto is the
4409 	     *  cleanest way to code the loop.)
4410 	     */
4411 tryagain:
4412 	    for (i = 0; i < nfsrv_statehashsize; i++) {
4413 		LIST_FOREACH(stp, &clp->lc_stateid[i], ls_hash) {
4414 		    if (stp->ls_stateid.other[2] == canuse) {
4415 			canuse++;
4416 			goto tryagain;
4417 		    }
4418 		}
4419 	    }
4420 	    clp->lc_flags |= LCL_INDEXNOTOK;
4421 	    return (canuse);
4422 	}
4423 
4424 	/*
4425 	 * Ok to start again from min + 1.
4426 	 */
4427 	clp->lc_stateindex = min_index + 1;
4428 	clp->lc_statemaxindex = max_index;
4429 	clp->lc_flags &= ~LCL_INDEXNOTOK;
4430 	return (clp->lc_stateindex);
4431 }
4432 
4433 /*
4434  * The following functions handle the stable storage file that deals with
4435  * the edge conditions described in RFC3530 Sec. 8.6.3.
4436  * The file is as follows:
4437  * - a single record at the beginning that has the lease time of the
4438  *   previous server instance (before the last reboot) and the nfsrvboottime
4439  *   values for the previous server boots.
4440  *   These previous boot times are used to ensure that the current
4441  *   nfsrvboottime does not, somehow, get set to a previous one.
4442  *   (This is important so that Stale ClientIDs and StateIDs can
4443  *    be recognized.)
4444  *   The number of previous nfsvrboottime values precedes the list.
4445  * - followed by some number of appended records with:
4446  *   - client id string
4447  *   - flag that indicates it is a record revoking state via lease
4448  *     expiration or similar
4449  *     OR has successfully acquired state.
4450  * These structures vary in length, with the client string at the end, up
4451  * to NFSV4_OPAQUELIMIT in size.
4452  *
4453  * At the end of the grace period, the file is truncated, the first
4454  * record is rewritten with updated information and any acquired state
4455  * records for successful reclaims of state are written.
4456  *
4457  * Subsequent records are appended when the first state is issued to
4458  * a client and when state is revoked for a client.
4459  *
4460  * When reading the file in, state issued records that come later in
4461  * the file override older ones, since the append log is in cronological order.
4462  * If, for some reason, the file can't be read, the grace period is
4463  * immediately terminated and all reclaims get NFSERR_NOGRACE.
4464  */
4465 
4466 /*
4467  * Read in the stable storage file. Called by nfssvc() before the nfsd
4468  * processes start servicing requests.
4469  */
4470 APPLESTATIC void
4471 nfsrv_setupstable(NFSPROC_T *p)
4472 {
4473 	struct nfsrv_stablefirst *sf = &nfsrv_stablefirst;
4474 	struct nfsrv_stable *sp, *nsp;
4475 	struct nfst_rec *tsp;
4476 	int error, i, tryagain;
4477 	off_t off = 0;
4478 	ssize_t aresid, len;
4479 
4480 	/*
4481 	 * If NFSNSF_UPDATEDONE is set, this is a restart of the nfsds without
4482 	 * a reboot, so state has not been lost.
4483 	 */
4484 	if (sf->nsf_flags & NFSNSF_UPDATEDONE)
4485 		return;
4486 	/*
4487 	 * Set Grace over just until the file reads successfully.
4488 	 */
4489 	nfsrvboottime = time_second;
4490 	LIST_INIT(&sf->nsf_head);
4491 	sf->nsf_flags = (NFSNSF_GRACEOVER | NFSNSF_NEEDLOCK);
4492 	sf->nsf_eograce = NFSD_MONOSEC + NFSRV_LEASEDELTA;
4493 	if (sf->nsf_fp == NULL)
4494 		return;
4495 	error = NFSD_RDWR(UIO_READ, NFSFPVNODE(sf->nsf_fp),
4496 	    (caddr_t)&sf->nsf_rec, sizeof (struct nfsf_rec), off, UIO_SYSSPACE,
4497 	    0, NFSFPCRED(sf->nsf_fp), &aresid, p);
4498 	if (error || aresid || sf->nsf_numboots == 0 ||
4499 		sf->nsf_numboots > NFSNSF_MAXNUMBOOTS)
4500 		return;
4501 
4502 	/*
4503 	 * Now, read in the boottimes.
4504 	 */
4505 	sf->nsf_bootvals = (time_t *)malloc((sf->nsf_numboots + 1) *
4506 		sizeof (time_t), M_TEMP, M_WAITOK);
4507 	off = sizeof (struct nfsf_rec);
4508 	error = NFSD_RDWR(UIO_READ, NFSFPVNODE(sf->nsf_fp),
4509 	    (caddr_t)sf->nsf_bootvals, sf->nsf_numboots * sizeof (time_t), off,
4510 	    UIO_SYSSPACE, 0, NFSFPCRED(sf->nsf_fp), &aresid, p);
4511 	if (error || aresid) {
4512 		free(sf->nsf_bootvals, M_TEMP);
4513 		sf->nsf_bootvals = NULL;
4514 		return;
4515 	}
4516 
4517 	/*
4518 	 * Make sure this nfsrvboottime is different from all recorded
4519 	 * previous ones.
4520 	 */
4521 	do {
4522 		tryagain = 0;
4523 		for (i = 0; i < sf->nsf_numboots; i++) {
4524 			if (nfsrvboottime == sf->nsf_bootvals[i]) {
4525 				nfsrvboottime++;
4526 				tryagain = 1;
4527 				break;
4528 			}
4529 		}
4530 	} while (tryagain);
4531 
4532 	sf->nsf_flags |= NFSNSF_OK;
4533 	off += (sf->nsf_numboots * sizeof (time_t));
4534 
4535 	/*
4536 	 * Read through the file, building a list of records for grace
4537 	 * checking.
4538 	 * Each record is between sizeof (struct nfst_rec) and
4539 	 * sizeof (struct nfst_rec) + NFSV4_OPAQUELIMIT - 1
4540 	 * and is actually sizeof (struct nfst_rec) + nst_len - 1.
4541 	 */
4542 	tsp = (struct nfst_rec *)malloc(sizeof (struct nfst_rec) +
4543 		NFSV4_OPAQUELIMIT - 1, M_TEMP, M_WAITOK);
4544 	do {
4545 	    error = NFSD_RDWR(UIO_READ, NFSFPVNODE(sf->nsf_fp),
4546 	        (caddr_t)tsp, sizeof (struct nfst_rec) + NFSV4_OPAQUELIMIT - 1,
4547 	        off, UIO_SYSSPACE, 0, NFSFPCRED(sf->nsf_fp), &aresid, p);
4548 	    len = (sizeof (struct nfst_rec) + NFSV4_OPAQUELIMIT - 1) - aresid;
4549 	    if (error || (len > 0 && (len < sizeof (struct nfst_rec) ||
4550 		len < (sizeof (struct nfst_rec) + tsp->len - 1)))) {
4551 		/*
4552 		 * Yuck, the file has been corrupted, so just return
4553 		 * after clearing out any restart state, so the grace period
4554 		 * is over.
4555 		 */
4556 		LIST_FOREACH_SAFE(sp, &sf->nsf_head, nst_list, nsp) {
4557 			LIST_REMOVE(sp, nst_list);
4558 			free(sp, M_TEMP);
4559 		}
4560 		free(tsp, M_TEMP);
4561 		sf->nsf_flags &= ~NFSNSF_OK;
4562 		free(sf->nsf_bootvals, M_TEMP);
4563 		sf->nsf_bootvals = NULL;
4564 		return;
4565 	    }
4566 	    if (len > 0) {
4567 		off += sizeof (struct nfst_rec) + tsp->len - 1;
4568 		/*
4569 		 * Search the list for a matching client.
4570 		 */
4571 		LIST_FOREACH(sp, &sf->nsf_head, nst_list) {
4572 			if (tsp->len == sp->nst_len &&
4573 			    !NFSBCMP(tsp->client, sp->nst_client, tsp->len))
4574 				break;
4575 		}
4576 		if (sp == LIST_END(&sf->nsf_head)) {
4577 			sp = (struct nfsrv_stable *)malloc(tsp->len +
4578 				sizeof (struct nfsrv_stable) - 1, M_TEMP,
4579 				M_WAITOK);
4580 			NFSBCOPY((caddr_t)tsp, (caddr_t)&sp->nst_rec,
4581 				sizeof (struct nfst_rec) + tsp->len - 1);
4582 			LIST_INSERT_HEAD(&sf->nsf_head, sp, nst_list);
4583 		} else {
4584 			if (tsp->flag == NFSNST_REVOKE)
4585 				sp->nst_flag |= NFSNST_REVOKE;
4586 			else
4587 				/*
4588 				 * A subsequent timestamp indicates the client
4589 				 * did a setclientid/confirm and any previous
4590 				 * revoke is no longer relevant.
4591 				 */
4592 				sp->nst_flag &= ~NFSNST_REVOKE;
4593 		}
4594 	    }
4595 	} while (len > 0);
4596 	free(tsp, M_TEMP);
4597 	sf->nsf_flags = NFSNSF_OK;
4598 	sf->nsf_eograce = NFSD_MONOSEC + sf->nsf_lease +
4599 		NFSRV_LEASEDELTA;
4600 }
4601 
4602 /*
4603  * Update the stable storage file, now that the grace period is over.
4604  */
4605 APPLESTATIC void
4606 nfsrv_updatestable(NFSPROC_T *p)
4607 {
4608 	struct nfsrv_stablefirst *sf = &nfsrv_stablefirst;
4609 	struct nfsrv_stable *sp, *nsp;
4610 	int i;
4611 	struct nfsvattr nva;
4612 	vnode_t vp;
4613 #if defined(__FreeBSD_version) && (__FreeBSD_version >= 500000)
4614 	mount_t mp = NULL;
4615 #endif
4616 	int error;
4617 
4618 	if (sf->nsf_fp == NULL || (sf->nsf_flags & NFSNSF_UPDATEDONE))
4619 		return;
4620 	sf->nsf_flags |= NFSNSF_UPDATEDONE;
4621 	/*
4622 	 * Ok, we need to rewrite the stable storage file.
4623 	 * - truncate to 0 length
4624 	 * - write the new first structure
4625 	 * - loop through the data structures, writing out any that
4626 	 *   have timestamps older than the old boot
4627 	 */
4628 	if (sf->nsf_bootvals) {
4629 		sf->nsf_numboots++;
4630 		for (i = sf->nsf_numboots - 2; i >= 0; i--)
4631 			sf->nsf_bootvals[i + 1] = sf->nsf_bootvals[i];
4632 	} else {
4633 		sf->nsf_numboots = 1;
4634 		sf->nsf_bootvals = (time_t *)malloc(sizeof (time_t),
4635 			M_TEMP, M_WAITOK);
4636 	}
4637 	sf->nsf_bootvals[0] = nfsrvboottime;
4638 	sf->nsf_lease = nfsrv_lease;
4639 	NFSVNO_ATTRINIT(&nva);
4640 	NFSVNO_SETATTRVAL(&nva, size, 0);
4641 	vp = NFSFPVNODE(sf->nsf_fp);
4642 	vn_start_write(vp, &mp, V_WAIT);
4643 	if (NFSVOPLOCK(vp, LK_EXCLUSIVE) == 0) {
4644 		error = nfsvno_setattr(vp, &nva, NFSFPCRED(sf->nsf_fp), p,
4645 		    NULL);
4646 		NFSVOPUNLOCK(vp, 0);
4647 	} else
4648 		error = EPERM;
4649 	vn_finished_write(mp);
4650 	if (!error)
4651 	    error = NFSD_RDWR(UIO_WRITE, vp,
4652 		(caddr_t)&sf->nsf_rec, sizeof (struct nfsf_rec), (off_t)0,
4653 		UIO_SYSSPACE, IO_SYNC, NFSFPCRED(sf->nsf_fp), NULL, p);
4654 	if (!error)
4655 	    error = NFSD_RDWR(UIO_WRITE, vp,
4656 		(caddr_t)sf->nsf_bootvals,
4657 		sf->nsf_numboots * sizeof (time_t),
4658 		(off_t)(sizeof (struct nfsf_rec)),
4659 		UIO_SYSSPACE, IO_SYNC, NFSFPCRED(sf->nsf_fp), NULL, p);
4660 	free(sf->nsf_bootvals, M_TEMP);
4661 	sf->nsf_bootvals = NULL;
4662 	if (error) {
4663 		sf->nsf_flags &= ~NFSNSF_OK;
4664 		printf("EEK! Can't write NfsV4 stable storage file\n");
4665 		return;
4666 	}
4667 	sf->nsf_flags |= NFSNSF_OK;
4668 
4669 	/*
4670 	 * Loop through the list and write out timestamp records for
4671 	 * any clients that successfully reclaimed state.
4672 	 */
4673 	LIST_FOREACH_SAFE(sp, &sf->nsf_head, nst_list, nsp) {
4674 		if (sp->nst_flag & NFSNST_GOTSTATE) {
4675 			nfsrv_writestable(sp->nst_client, sp->nst_len,
4676 				NFSNST_NEWSTATE, p);
4677 			sp->nst_clp->lc_flags |= LCL_STAMPEDSTABLE;
4678 		}
4679 		LIST_REMOVE(sp, nst_list);
4680 		free(sp, M_TEMP);
4681 	}
4682 	nfsrv_backupstable();
4683 }
4684 
4685 /*
4686  * Append a record to the stable storage file.
4687  */
4688 APPLESTATIC void
4689 nfsrv_writestable(u_char *client, int len, int flag, NFSPROC_T *p)
4690 {
4691 	struct nfsrv_stablefirst *sf = &nfsrv_stablefirst;
4692 	struct nfst_rec *sp;
4693 	int error;
4694 
4695 	if (!(sf->nsf_flags & NFSNSF_OK) || sf->nsf_fp == NULL)
4696 		return;
4697 	sp = (struct nfst_rec *)malloc(sizeof (struct nfst_rec) +
4698 		len - 1, M_TEMP, M_WAITOK);
4699 	sp->len = len;
4700 	NFSBCOPY(client, sp->client, len);
4701 	sp->flag = flag;
4702 	error = NFSD_RDWR(UIO_WRITE, NFSFPVNODE(sf->nsf_fp),
4703 	    (caddr_t)sp, sizeof (struct nfst_rec) + len - 1, (off_t)0,
4704 	    UIO_SYSSPACE, (IO_SYNC | IO_APPEND), NFSFPCRED(sf->nsf_fp), NULL, p);
4705 	free(sp, M_TEMP);
4706 	if (error) {
4707 		sf->nsf_flags &= ~NFSNSF_OK;
4708 		printf("EEK! Can't write NfsV4 stable storage file\n");
4709 	}
4710 }
4711 
4712 /*
4713  * This function is called during the grace period to mark a client
4714  * that successfully reclaimed state.
4715  */
4716 static void
4717 nfsrv_markstable(struct nfsclient *clp)
4718 {
4719 	struct nfsrv_stable *sp;
4720 
4721 	/*
4722 	 * First find the client structure.
4723 	 */
4724 	LIST_FOREACH(sp, &nfsrv_stablefirst.nsf_head, nst_list) {
4725 		if (sp->nst_len == clp->lc_idlen &&
4726 		    !NFSBCMP(sp->nst_client, clp->lc_id, sp->nst_len))
4727 			break;
4728 	}
4729 	if (sp == LIST_END(&nfsrv_stablefirst.nsf_head))
4730 		return;
4731 
4732 	/*
4733 	 * Now, just mark it and set the nfsclient back pointer.
4734 	 */
4735 	sp->nst_flag |= NFSNST_GOTSTATE;
4736 	sp->nst_clp = clp;
4737 }
4738 
4739 /*
4740  * This function is called for a reclaim, to see if it gets grace.
4741  * It returns 0 if a reclaim is allowed, 1 otherwise.
4742  */
4743 static int
4744 nfsrv_checkstable(struct nfsclient *clp)
4745 {
4746 	struct nfsrv_stable *sp;
4747 
4748 	/*
4749 	 * First, find the entry for the client.
4750 	 */
4751 	LIST_FOREACH(sp, &nfsrv_stablefirst.nsf_head, nst_list) {
4752 		if (sp->nst_len == clp->lc_idlen &&
4753 		    !NFSBCMP(sp->nst_client, clp->lc_id, sp->nst_len))
4754 			break;
4755 	}
4756 
4757 	/*
4758 	 * If not in the list, state was revoked or no state was issued
4759 	 * since the previous reboot, a reclaim is denied.
4760 	 */
4761 	if (sp == LIST_END(&nfsrv_stablefirst.nsf_head) ||
4762 	    (sp->nst_flag & NFSNST_REVOKE) ||
4763 	    !(nfsrv_stablefirst.nsf_flags & NFSNSF_OK))
4764 		return (1);
4765 	return (0);
4766 }
4767 
4768 /*
4769  * Test for and try to clear out a conflicting client. This is called by
4770  * nfsrv_lockctrl() and nfsrv_openctrl() when conflicts with other clients
4771  * a found.
4772  * The trick here is that it can't revoke a conflicting client with an
4773  * expired lease unless it holds the v4root lock, so...
4774  * If no v4root lock, get the lock and return 1 to indicate "try again".
4775  * Return 0 to indicate the conflict can't be revoked and 1 to indicate
4776  * the revocation worked and the conflicting client is "bye, bye", so it
4777  * can be tried again.
4778  * Return 2 to indicate that the vnode is VI_DOOMED after NFSVOPLOCK().
4779  * Unlocks State before a non-zero value is returned.
4780  */
4781 static int
4782 nfsrv_clientconflict(struct nfsclient *clp, int *haslockp, vnode_t vp,
4783     NFSPROC_T *p)
4784 {
4785 	int gotlock, lktype = 0;
4786 
4787 	/*
4788 	 * If lease hasn't expired, we can't fix it.
4789 	 */
4790 	if (clp->lc_expiry >= NFSD_MONOSEC ||
4791 	    !(nfsrv_stablefirst.nsf_flags & NFSNSF_UPDATEDONE))
4792 		return (0);
4793 	if (*haslockp == 0) {
4794 		NFSUNLOCKSTATE();
4795 		if (vp != NULL) {
4796 			lktype = NFSVOPISLOCKED(vp);
4797 			NFSVOPUNLOCK(vp, 0);
4798 		}
4799 		NFSLOCKV4ROOTMUTEX();
4800 		nfsv4_relref(&nfsv4rootfs_lock);
4801 		do {
4802 			gotlock = nfsv4_lock(&nfsv4rootfs_lock, 1, NULL,
4803 			    NFSV4ROOTLOCKMUTEXPTR, NULL);
4804 		} while (!gotlock);
4805 		NFSUNLOCKV4ROOTMUTEX();
4806 		*haslockp = 1;
4807 		if (vp != NULL) {
4808 			NFSVOPLOCK(vp, lktype | LK_RETRY);
4809 			if ((vp->v_iflag & VI_DOOMED) != 0)
4810 				return (2);
4811 		}
4812 		return (1);
4813 	}
4814 	NFSUNLOCKSTATE();
4815 
4816 	/*
4817 	 * Ok, we can expire the conflicting client.
4818 	 */
4819 	nfsrv_writestable(clp->lc_id, clp->lc_idlen, NFSNST_REVOKE, p);
4820 	nfsrv_backupstable();
4821 	nfsrv_cleanclient(clp, p);
4822 	nfsrv_freedeleglist(&clp->lc_deleg);
4823 	nfsrv_freedeleglist(&clp->lc_olddeleg);
4824 	LIST_REMOVE(clp, lc_hash);
4825 	nfsrv_zapclient(clp, p);
4826 	return (1);
4827 }
4828 
4829 /*
4830  * Resolve a delegation conflict.
4831  * Returns 0 to indicate the conflict was resolved without sleeping.
4832  * Return -1 to indicate that the caller should check for conflicts again.
4833  * Return > 0 for an error that should be returned, normally NFSERR_DELAY.
4834  *
4835  * Also, manipulate the nfsv4root_lock, as required. It isn't changed
4836  * for a return of 0, since there was no sleep and it could be required
4837  * later. It is released for a return of NFSERR_DELAY, since the caller
4838  * will return that error. It is released when a sleep was done waiting
4839  * for the delegation to be returned or expire (so that other nfsds can
4840  * handle ops). Then, it must be acquired for the write to stable storage.
4841  * (This function is somewhat similar to nfsrv_clientconflict(), but
4842  *  the semantics differ in a couple of subtle ways. The return of 0
4843  *  indicates the conflict was resolved without sleeping here, not
4844  *  that the conflict can't be resolved and the handling of nfsv4root_lock
4845  *  differs, as noted above.)
4846  * Unlocks State before returning a non-zero value.
4847  */
4848 static int
4849 nfsrv_delegconflict(struct nfsstate *stp, int *haslockp, NFSPROC_T *p,
4850     vnode_t vp)
4851 {
4852 	struct nfsclient *clp = stp->ls_clp;
4853 	int gotlock, error, lktype = 0, retrycnt, zapped_clp;
4854 	nfsv4stateid_t tstateid;
4855 	fhandle_t tfh;
4856 
4857 	/*
4858 	 * If the conflict is with an old delegation...
4859 	 */
4860 	if (stp->ls_flags & NFSLCK_OLDDELEG) {
4861 		/*
4862 		 * You can delete it, if it has expired.
4863 		 */
4864 		if (clp->lc_delegtime < NFSD_MONOSEC) {
4865 			nfsrv_freedeleg(stp);
4866 			NFSUNLOCKSTATE();
4867 			error = -1;
4868 			goto out;
4869 		}
4870 		NFSUNLOCKSTATE();
4871 		/*
4872 		 * During this delay, the old delegation could expire or it
4873 		 * could be recovered by the client via an Open with
4874 		 * CLAIM_DELEGATE_PREV.
4875 		 * Release the nfsv4root_lock, if held.
4876 		 */
4877 		if (*haslockp) {
4878 			*haslockp = 0;
4879 			NFSLOCKV4ROOTMUTEX();
4880 			nfsv4_unlock(&nfsv4rootfs_lock, 1);
4881 			NFSUNLOCKV4ROOTMUTEX();
4882 		}
4883 		error = NFSERR_DELAY;
4884 		goto out;
4885 	}
4886 
4887 	/*
4888 	 * It's a current delegation, so:
4889 	 * - check to see if the delegation has expired
4890 	 *   - if so, get the v4root lock and then expire it
4891 	 */
4892 	if (!(stp->ls_flags & NFSLCK_DELEGRECALL)) {
4893 		/*
4894 		 * - do a recall callback, since not yet done
4895 		 * For now, never allow truncate to be set. To use
4896 		 * truncate safely, it must be guaranteed that the
4897 		 * Remove, Rename or Setattr with size of 0 will
4898 		 * succeed and that would require major changes to
4899 		 * the VFS/Vnode OPs.
4900 		 * Set the expiry time large enough so that it won't expire
4901 		 * until after the callback, then set it correctly, once
4902 		 * the callback is done. (The delegation will now time
4903 		 * out whether or not the Recall worked ok. The timeout
4904 		 * will be extended when ops are done on the delegation
4905 		 * stateid, up to the timelimit.)
4906 		 */
4907 		stp->ls_delegtime = NFSD_MONOSEC + (2 * nfsrv_lease) +
4908 		    NFSRV_LEASEDELTA;
4909 		stp->ls_delegtimelimit = NFSD_MONOSEC + (6 * nfsrv_lease) +
4910 		    NFSRV_LEASEDELTA;
4911 		stp->ls_flags |= NFSLCK_DELEGRECALL;
4912 
4913 		/*
4914 		 * Loop NFSRV_CBRETRYCNT times while the CBRecall replies
4915 		 * NFSERR_BADSTATEID or NFSERR_BADHANDLE. This is done
4916 		 * in order to try and avoid a race that could happen
4917 		 * when a CBRecall request passed the Open reply with
4918 		 * the delegation in it when transitting the network.
4919 		 * Since nfsrv_docallback will sleep, don't use stp after
4920 		 * the call.
4921 		 */
4922 		NFSBCOPY((caddr_t)&stp->ls_stateid, (caddr_t)&tstateid,
4923 		    sizeof (tstateid));
4924 		NFSBCOPY((caddr_t)&stp->ls_lfp->lf_fh, (caddr_t)&tfh,
4925 		    sizeof (tfh));
4926 		NFSUNLOCKSTATE();
4927 		if (*haslockp) {
4928 			*haslockp = 0;
4929 			NFSLOCKV4ROOTMUTEX();
4930 			nfsv4_unlock(&nfsv4rootfs_lock, 1);
4931 			NFSUNLOCKV4ROOTMUTEX();
4932 		}
4933 		retrycnt = 0;
4934 		do {
4935 		    error = nfsrv_docallback(clp, NFSV4OP_CBRECALL,
4936 			&tstateid, 0, &tfh, NULL, NULL, p);
4937 		    retrycnt++;
4938 		} while ((error == NFSERR_BADSTATEID ||
4939 		    error == NFSERR_BADHANDLE) && retrycnt < NFSV4_CBRETRYCNT);
4940 		error = NFSERR_DELAY;
4941 		goto out;
4942 	}
4943 
4944 	if (clp->lc_expiry >= NFSD_MONOSEC &&
4945 	    stp->ls_delegtime >= NFSD_MONOSEC) {
4946 		NFSUNLOCKSTATE();
4947 		/*
4948 		 * A recall has been done, but it has not yet expired.
4949 		 * So, RETURN_DELAY.
4950 		 */
4951 		if (*haslockp) {
4952 			*haslockp = 0;
4953 			NFSLOCKV4ROOTMUTEX();
4954 			nfsv4_unlock(&nfsv4rootfs_lock, 1);
4955 			NFSUNLOCKV4ROOTMUTEX();
4956 		}
4957 		error = NFSERR_DELAY;
4958 		goto out;
4959 	}
4960 
4961 	/*
4962 	 * If we don't yet have the lock, just get it and then return,
4963 	 * since we need that before deleting expired state, such as
4964 	 * this delegation.
4965 	 * When getting the lock, unlock the vnode, so other nfsds that
4966 	 * are in progress, won't get stuck waiting for the vnode lock.
4967 	 */
4968 	if (*haslockp == 0) {
4969 		NFSUNLOCKSTATE();
4970 		if (vp != NULL) {
4971 			lktype = NFSVOPISLOCKED(vp);
4972 			NFSVOPUNLOCK(vp, 0);
4973 		}
4974 		NFSLOCKV4ROOTMUTEX();
4975 		nfsv4_relref(&nfsv4rootfs_lock);
4976 		do {
4977 			gotlock = nfsv4_lock(&nfsv4rootfs_lock, 1, NULL,
4978 			    NFSV4ROOTLOCKMUTEXPTR, NULL);
4979 		} while (!gotlock);
4980 		NFSUNLOCKV4ROOTMUTEX();
4981 		*haslockp = 1;
4982 		if (vp != NULL) {
4983 			NFSVOPLOCK(vp, lktype | LK_RETRY);
4984 			if ((vp->v_iflag & VI_DOOMED) != 0) {
4985 				*haslockp = 0;
4986 				NFSLOCKV4ROOTMUTEX();
4987 				nfsv4_unlock(&nfsv4rootfs_lock, 1);
4988 				NFSUNLOCKV4ROOTMUTEX();
4989 				error = NFSERR_PERM;
4990 				goto out;
4991 			}
4992 		}
4993 		error = -1;
4994 		goto out;
4995 	}
4996 
4997 	NFSUNLOCKSTATE();
4998 	/*
4999 	 * Ok, we can delete the expired delegation.
5000 	 * First, write the Revoke record to stable storage and then
5001 	 * clear out the conflict.
5002 	 * Since all other nfsd threads are now blocked, we can safely
5003 	 * sleep without the state changing.
5004 	 */
5005 	nfsrv_writestable(clp->lc_id, clp->lc_idlen, NFSNST_REVOKE, p);
5006 	nfsrv_backupstable();
5007 	if (clp->lc_expiry < NFSD_MONOSEC) {
5008 		nfsrv_cleanclient(clp, p);
5009 		nfsrv_freedeleglist(&clp->lc_deleg);
5010 		nfsrv_freedeleglist(&clp->lc_olddeleg);
5011 		LIST_REMOVE(clp, lc_hash);
5012 		zapped_clp = 1;
5013 	} else {
5014 		nfsrv_freedeleg(stp);
5015 		zapped_clp = 0;
5016 	}
5017 	if (zapped_clp)
5018 		nfsrv_zapclient(clp, p);
5019 	error = -1;
5020 
5021 out:
5022 	NFSEXITCODE(error);
5023 	return (error);
5024 }
5025 
5026 /*
5027  * Check for a remove allowed, if remove is set to 1 and get rid of
5028  * delegations.
5029  */
5030 APPLESTATIC int
5031 nfsrv_checkremove(vnode_t vp, int remove, NFSPROC_T *p)
5032 {
5033 	struct nfsstate *stp;
5034 	struct nfslockfile *lfp;
5035 	int error, haslock = 0;
5036 	fhandle_t nfh;
5037 
5038 	/*
5039 	 * First, get the lock file structure.
5040 	 * (A return of -1 means no associated state, so remove ok.)
5041 	 */
5042 	error = nfsrv_getlockfh(vp, NFSLCK_CHECK, NULL, &nfh, p);
5043 tryagain:
5044 	NFSLOCKSTATE();
5045 	if (!error)
5046 		error = nfsrv_getlockfile(NFSLCK_CHECK, NULL, &lfp, &nfh, 0);
5047 	if (error) {
5048 		NFSUNLOCKSTATE();
5049 		if (haslock) {
5050 			NFSLOCKV4ROOTMUTEX();
5051 			nfsv4_unlock(&nfsv4rootfs_lock, 1);
5052 			NFSUNLOCKV4ROOTMUTEX();
5053 		}
5054 		if (error == -1)
5055 			error = 0;
5056 		goto out;
5057 	}
5058 
5059 	/*
5060 	 * Now, we must Recall any delegations.
5061 	 */
5062 	error = nfsrv_cleandeleg(vp, lfp, NULL, &haslock, p);
5063 	if (error) {
5064 		/*
5065 		 * nfsrv_cleandeleg() unlocks state for non-zero
5066 		 * return.
5067 		 */
5068 		if (error == -1)
5069 			goto tryagain;
5070 		if (haslock) {
5071 			NFSLOCKV4ROOTMUTEX();
5072 			nfsv4_unlock(&nfsv4rootfs_lock, 1);
5073 			NFSUNLOCKV4ROOTMUTEX();
5074 		}
5075 		goto out;
5076 	}
5077 
5078 	/*
5079 	 * Now, look for a conflicting open share.
5080 	 */
5081 	if (remove) {
5082 		/*
5083 		 * If the entry in the directory was the last reference to the
5084 		 * corresponding filesystem object, the object can be destroyed
5085 		 * */
5086 		if(lfp->lf_usecount>1)
5087 			LIST_FOREACH(stp, &lfp->lf_open, ls_file) {
5088 				if (stp->ls_flags & NFSLCK_WRITEDENY) {
5089 					error = NFSERR_FILEOPEN;
5090 					break;
5091 				}
5092 			}
5093 	}
5094 
5095 	NFSUNLOCKSTATE();
5096 	if (haslock) {
5097 		NFSLOCKV4ROOTMUTEX();
5098 		nfsv4_unlock(&nfsv4rootfs_lock, 1);
5099 		NFSUNLOCKV4ROOTMUTEX();
5100 	}
5101 
5102 out:
5103 	NFSEXITCODE(error);
5104 	return (error);
5105 }
5106 
5107 /*
5108  * Clear out all delegations for the file referred to by lfp.
5109  * May return NFSERR_DELAY, if there will be a delay waiting for
5110  * delegations to expire.
5111  * Returns -1 to indicate it slept while recalling a delegation.
5112  * This function has the side effect of deleting the nfslockfile structure,
5113  * if it no longer has associated state and didn't have to sleep.
5114  * Unlocks State before a non-zero value is returned.
5115  */
5116 static int
5117 nfsrv_cleandeleg(vnode_t vp, struct nfslockfile *lfp,
5118     struct nfsclient *clp, int *haslockp, NFSPROC_T *p)
5119 {
5120 	struct nfsstate *stp, *nstp;
5121 	int ret = 0;
5122 
5123 	stp = LIST_FIRST(&lfp->lf_deleg);
5124 	while (stp != LIST_END(&lfp->lf_deleg)) {
5125 		nstp = LIST_NEXT(stp, ls_file);
5126 		if (stp->ls_clp != clp) {
5127 			ret = nfsrv_delegconflict(stp, haslockp, p, vp);
5128 			if (ret) {
5129 				/*
5130 				 * nfsrv_delegconflict() unlocks state
5131 				 * when it returns non-zero.
5132 				 */
5133 				goto out;
5134 			}
5135 		}
5136 		stp = nstp;
5137 	}
5138 out:
5139 	NFSEXITCODE(ret);
5140 	return (ret);
5141 }
5142 
5143 /*
5144  * There are certain operations that, when being done outside of NFSv4,
5145  * require that any NFSv4 delegation for the file be recalled.
5146  * This function is to be called for those cases:
5147  * VOP_RENAME() - When a delegation is being recalled for any reason,
5148  *	the client may have to do Opens against the server, using the file's
5149  *	final component name. If the file has been renamed on the server,
5150  *	that component name will be incorrect and the Open will fail.
5151  * VOP_REMOVE() - Theoretically, a client could Open a file after it has
5152  *	been removed on the server, if there is a delegation issued to
5153  *	that client for the file. I say "theoretically" since clients
5154  *	normally do an Access Op before the Open and that Access Op will
5155  *	fail with ESTALE. Note that NFSv2 and 3 don't even do Opens, so
5156  *	they will detect the file's removal in the same manner. (There is
5157  *	one case where RFC3530 allows a client to do an Open without first
5158  *	doing an Access Op, which is passage of a check against the ACE
5159  *	returned with a Write delegation, but current practice is to ignore
5160  *	the ACE and always do an Access Op.)
5161  *	Since the functions can only be called with an unlocked vnode, this
5162  *	can't be done at this time.
5163  * VOP_ADVLOCK() - When a client holds a delegation, it can issue byte range
5164  *	locks locally in the client, which are not visible to the server. To
5165  *	deal with this, issuing of delegations for a vnode must be disabled
5166  *	and all delegations for the vnode recalled. This is done via the
5167  *	second function, using the VV_DISABLEDELEG vflag on the vnode.
5168  */
5169 APPLESTATIC void
5170 nfsd_recalldelegation(vnode_t vp, NFSPROC_T *p)
5171 {
5172 	time_t starttime;
5173 	int error;
5174 
5175 	/*
5176 	 * First, check to see if the server is currently running and it has
5177 	 * been called for a regular file when issuing delegations.
5178 	 */
5179 	if (newnfs_numnfsd == 0 || vp->v_type != VREG ||
5180 	    nfsrv_issuedelegs == 0)
5181 		return;
5182 
5183 	KASSERT((NFSVOPISLOCKED(vp) != LK_EXCLUSIVE), ("vp %p is locked", vp));
5184 	/*
5185 	 * First, get a reference on the nfsv4rootfs_lock so that an
5186 	 * exclusive lock cannot be acquired by another thread.
5187 	 */
5188 	NFSLOCKV4ROOTMUTEX();
5189 	nfsv4_getref(&nfsv4rootfs_lock, NULL, NFSV4ROOTLOCKMUTEXPTR, NULL);
5190 	NFSUNLOCKV4ROOTMUTEX();
5191 
5192 	/*
5193 	 * Now, call nfsrv_checkremove() in a loop while it returns
5194 	 * NFSERR_DELAY. Return upon any other error or when timed out.
5195 	 */
5196 	starttime = NFSD_MONOSEC;
5197 	do {
5198 		if (NFSVOPLOCK(vp, LK_EXCLUSIVE) == 0) {
5199 			error = nfsrv_checkremove(vp, 0, p);
5200 			NFSVOPUNLOCK(vp, 0);
5201 		} else
5202 			error = EPERM;
5203 		if (error == NFSERR_DELAY) {
5204 			if (NFSD_MONOSEC - starttime > NFS_REMOVETIMEO)
5205 				break;
5206 			/* Sleep for a short period of time */
5207 			(void) nfs_catnap(PZERO, 0, "nfsremove");
5208 		}
5209 	} while (error == NFSERR_DELAY);
5210 	NFSLOCKV4ROOTMUTEX();
5211 	nfsv4_relref(&nfsv4rootfs_lock);
5212 	NFSUNLOCKV4ROOTMUTEX();
5213 }
5214 
5215 APPLESTATIC void
5216 nfsd_disabledelegation(vnode_t vp, NFSPROC_T *p)
5217 {
5218 
5219 #ifdef VV_DISABLEDELEG
5220 	/*
5221 	 * First, flag issuance of delegations disabled.
5222 	 */
5223 	atomic_set_long(&vp->v_vflag, VV_DISABLEDELEG);
5224 #endif
5225 
5226 	/*
5227 	 * Then call nfsd_recalldelegation() to get rid of all extant
5228 	 * delegations.
5229 	 */
5230 	nfsd_recalldelegation(vp, p);
5231 }
5232 
5233 /*
5234  * Check for conflicting locks, etc. and then get rid of delegations.
5235  * (At one point I thought that I should get rid of delegations for any
5236  *  Setattr, since it could potentially disallow the I/O op (read or write)
5237  *  allowed by the delegation. However, Setattr Ops that aren't changing
5238  *  the size get a stateid of all 0s, so you can't tell if it is a delegation
5239  *  for the same client or a different one, so I decided to only get rid
5240  *  of delegations for other clients when the size is being changed.)
5241  * In general, a Setattr can disable NFS I/O Ops that are outstanding, such
5242  * as Write backs, even if there is no delegation, so it really isn't any
5243  * different?)
5244  */
5245 APPLESTATIC int
5246 nfsrv_checksetattr(vnode_t vp, struct nfsrv_descript *nd,
5247     nfsv4stateid_t *stateidp, struct nfsvattr *nvap, nfsattrbit_t *attrbitp,
5248     struct nfsexstuff *exp, NFSPROC_T *p)
5249 {
5250 	struct nfsstate st, *stp = &st;
5251 	struct nfslock lo, *lop = &lo;
5252 	int error = 0;
5253 	nfsquad_t clientid;
5254 
5255 	if (NFSISSET_ATTRBIT(attrbitp, NFSATTRBIT_SIZE)) {
5256 		stp->ls_flags = (NFSLCK_CHECK | NFSLCK_WRITEACCESS);
5257 		lop->lo_first = nvap->na_size;
5258 	} else {
5259 		stp->ls_flags = 0;
5260 		lop->lo_first = 0;
5261 	}
5262 	if (NFSISSET_ATTRBIT(attrbitp, NFSATTRBIT_OWNER) ||
5263 	    NFSISSET_ATTRBIT(attrbitp, NFSATTRBIT_OWNERGROUP) ||
5264 	    NFSISSET_ATTRBIT(attrbitp, NFSATTRBIT_MODE) ||
5265 	    NFSISSET_ATTRBIT(attrbitp, NFSATTRBIT_ACL))
5266 		stp->ls_flags |= NFSLCK_SETATTR;
5267 	if (stp->ls_flags == 0)
5268 		goto out;
5269 	lop->lo_end = NFS64BITSSET;
5270 	lop->lo_flags = NFSLCK_WRITE;
5271 	stp->ls_ownerlen = 0;
5272 	stp->ls_op = NULL;
5273 	stp->ls_uid = nd->nd_cred->cr_uid;
5274 	stp->ls_stateid.seqid = stateidp->seqid;
5275 	clientid.lval[0] = stp->ls_stateid.other[0] = stateidp->other[0];
5276 	clientid.lval[1] = stp->ls_stateid.other[1] = stateidp->other[1];
5277 	stp->ls_stateid.other[2] = stateidp->other[2];
5278 	error = nfsrv_lockctrl(vp, &stp, &lop, NULL, clientid,
5279 	    stateidp, exp, nd, p);
5280 
5281 out:
5282 	NFSEXITCODE2(error, nd);
5283 	return (error);
5284 }
5285 
5286 /*
5287  * Check for a write delegation and do a CBGETATTR if there is one, updating
5288  * the attributes, as required.
5289  * Should I return an error if I can't get the attributes? (For now, I'll
5290  * just return ok.
5291  */
5292 APPLESTATIC int
5293 nfsrv_checkgetattr(struct nfsrv_descript *nd, vnode_t vp,
5294     struct nfsvattr *nvap, nfsattrbit_t *attrbitp, struct ucred *cred,
5295     NFSPROC_T *p)
5296 {
5297 	struct nfsstate *stp;
5298 	struct nfslockfile *lfp;
5299 	struct nfsclient *clp;
5300 	struct nfsvattr nva;
5301 	fhandle_t nfh;
5302 	int error = 0;
5303 	nfsattrbit_t cbbits;
5304 	u_quad_t delegfilerev;
5305 
5306 	NFSCBGETATTR_ATTRBIT(attrbitp, &cbbits);
5307 	if (!NFSNONZERO_ATTRBIT(&cbbits))
5308 		goto out;
5309 	if (nfsrv_writedelegcnt == 0)
5310 		goto out;
5311 
5312 	/*
5313 	 * Get the lock file structure.
5314 	 * (A return of -1 means no associated state, so return ok.)
5315 	 */
5316 	error = nfsrv_getlockfh(vp, NFSLCK_CHECK, NULL, &nfh, p);
5317 	NFSLOCKSTATE();
5318 	if (!error)
5319 		error = nfsrv_getlockfile(NFSLCK_CHECK, NULL, &lfp, &nfh, 0);
5320 	if (error) {
5321 		NFSUNLOCKSTATE();
5322 		if (error == -1)
5323 			error = 0;
5324 		goto out;
5325 	}
5326 
5327 	/*
5328 	 * Now, look for a write delegation.
5329 	 */
5330 	LIST_FOREACH(stp, &lfp->lf_deleg, ls_file) {
5331 		if (stp->ls_flags & NFSLCK_DELEGWRITE)
5332 			break;
5333 	}
5334 	if (stp == LIST_END(&lfp->lf_deleg)) {
5335 		NFSUNLOCKSTATE();
5336 		goto out;
5337 	}
5338 	clp = stp->ls_clp;
5339 	delegfilerev = stp->ls_filerev;
5340 
5341 	/*
5342 	 * If the Write delegation was issued as a part of this Compound RPC
5343 	 * or if we have an Implied Clientid (used in a previous Op in this
5344 	 * compound) and it is the client the delegation was issued to,
5345 	 * just return ok.
5346 	 * I also assume that it is from the same client iff the network
5347 	 * host IP address is the same as the callback address. (Not
5348 	 * exactly correct by the RFC, but avoids a lot of Getattr
5349 	 * callbacks.)
5350 	 */
5351 	if (nd->nd_compref == stp->ls_compref ||
5352 	    ((nd->nd_flag & ND_IMPLIEDCLID) &&
5353 	     clp->lc_clientid.qval == nd->nd_clientid.qval) ||
5354 	     nfsaddr2_match(clp->lc_req.nr_nam, nd->nd_nam)) {
5355 		NFSUNLOCKSTATE();
5356 		goto out;
5357 	}
5358 
5359 	/*
5360 	 * We are now done with the delegation state structure,
5361 	 * so the statelock can be released and we can now tsleep().
5362 	 */
5363 
5364 	/*
5365 	 * Now, we must do the CB Getattr callback, to see if Change or Size
5366 	 * has changed.
5367 	 */
5368 	if (clp->lc_expiry >= NFSD_MONOSEC) {
5369 		NFSUNLOCKSTATE();
5370 		NFSVNO_ATTRINIT(&nva);
5371 		nva.na_filerev = NFS64BITSSET;
5372 		error = nfsrv_docallback(clp, NFSV4OP_CBGETATTR, NULL,
5373 		    0, &nfh, &nva, &cbbits, p);
5374 		if (!error) {
5375 			if ((nva.na_filerev != NFS64BITSSET &&
5376 			    nva.na_filerev > delegfilerev) ||
5377 			    (NFSVNO_ISSETSIZE(&nva) &&
5378 			     nva.na_size != nvap->na_size)) {
5379 				error = nfsvno_updfilerev(vp, nvap, cred, p);
5380 				if (NFSVNO_ISSETSIZE(&nva))
5381 					nvap->na_size = nva.na_size;
5382 			}
5383 		} else
5384 			error = 0;	/* Ignore callback errors for now. */
5385 	} else {
5386 		NFSUNLOCKSTATE();
5387 	}
5388 
5389 out:
5390 	NFSEXITCODE2(error, nd);
5391 	return (error);
5392 }
5393 
5394 /*
5395  * This function looks for openowners that haven't had any opens for
5396  * a while and throws them away. Called by an nfsd when NFSNSF_NOOPENS
5397  * is set.
5398  */
5399 APPLESTATIC void
5400 nfsrv_throwawayopens(NFSPROC_T *p)
5401 {
5402 	struct nfsclient *clp, *nclp;
5403 	struct nfsstate *stp, *nstp;
5404 	int i;
5405 
5406 	NFSLOCKSTATE();
5407 	nfsrv_stablefirst.nsf_flags &= ~NFSNSF_NOOPENS;
5408 	/*
5409 	 * For each client...
5410 	 */
5411 	for (i = 0; i < nfsrv_clienthashsize; i++) {
5412 	    LIST_FOREACH_SAFE(clp, &nfsclienthash[i], lc_hash, nclp) {
5413 		LIST_FOREACH_SAFE(stp, &clp->lc_open, ls_list, nstp) {
5414 			if (LIST_EMPTY(&stp->ls_open) &&
5415 			    (stp->ls_noopens > NFSNOOPEN ||
5416 			     (nfsrv_openpluslock * 2) >
5417 			     nfsrv_v4statelimit))
5418 				nfsrv_freeopenowner(stp, 0, p);
5419 		}
5420 	    }
5421 	}
5422 	NFSUNLOCKSTATE();
5423 }
5424 
5425 /*
5426  * This function checks to see if the credentials are the same.
5427  * Returns 1 for not same, 0 otherwise.
5428  */
5429 static int
5430 nfsrv_notsamecredname(struct nfsrv_descript *nd, struct nfsclient *clp)
5431 {
5432 
5433 	if (nd->nd_flag & ND_GSS) {
5434 		if (!(clp->lc_flags & LCL_GSS))
5435 			return (1);
5436 		if (clp->lc_flags & LCL_NAME) {
5437 			if (nd->nd_princlen != clp->lc_namelen ||
5438 			    NFSBCMP(nd->nd_principal, clp->lc_name,
5439 				clp->lc_namelen))
5440 				return (1);
5441 			else
5442 				return (0);
5443 		}
5444 		if (nd->nd_cred->cr_uid == clp->lc_uid)
5445 			return (0);
5446 		else
5447 			return (1);
5448 	} else if (clp->lc_flags & LCL_GSS)
5449 		return (1);
5450 	/*
5451 	 * For AUTH_SYS, allow the same uid or root. (This is underspecified
5452 	 * in RFC3530, which talks about principals, but doesn't say anything
5453 	 * about uids for AUTH_SYS.)
5454 	 */
5455 	if (nd->nd_cred->cr_uid == clp->lc_uid || nd->nd_cred->cr_uid == 0)
5456 		return (0);
5457 	else
5458 		return (1);
5459 }
5460 
5461 /*
5462  * Calculate the lease expiry time.
5463  */
5464 static time_t
5465 nfsrv_leaseexpiry(void)
5466 {
5467 
5468 	if (nfsrv_stablefirst.nsf_eograce > NFSD_MONOSEC)
5469 		return (NFSD_MONOSEC + 2 * (nfsrv_lease + NFSRV_LEASEDELTA));
5470 	return (NFSD_MONOSEC + nfsrv_lease + NFSRV_LEASEDELTA);
5471 }
5472 
5473 /*
5474  * Delay the delegation timeout as far as ls_delegtimelimit, as required.
5475  */
5476 static void
5477 nfsrv_delaydelegtimeout(struct nfsstate *stp)
5478 {
5479 
5480 	if ((stp->ls_flags & NFSLCK_DELEGRECALL) == 0)
5481 		return;
5482 
5483 	if ((stp->ls_delegtime + 15) > NFSD_MONOSEC &&
5484 	    stp->ls_delegtime < stp->ls_delegtimelimit) {
5485 		stp->ls_delegtime += nfsrv_lease;
5486 		if (stp->ls_delegtime > stp->ls_delegtimelimit)
5487 			stp->ls_delegtime = stp->ls_delegtimelimit;
5488 	}
5489 }
5490 
5491 /*
5492  * This function checks to see if there is any other state associated
5493  * with the openowner for this Open.
5494  * It returns 1 if there is no other state, 0 otherwise.
5495  */
5496 static int
5497 nfsrv_nootherstate(struct nfsstate *stp)
5498 {
5499 	struct nfsstate *tstp;
5500 
5501 	LIST_FOREACH(tstp, &stp->ls_openowner->ls_open, ls_list) {
5502 		if (tstp != stp || !LIST_EMPTY(&tstp->ls_lock))
5503 			return (0);
5504 	}
5505 	return (1);
5506 }
5507 
5508 /*
5509  * Create a list of lock deltas (changes to local byte range locking
5510  * that can be rolled back using the list) and apply the changes via
5511  * nfsvno_advlock(). Optionally, lock the list. It is expected that either
5512  * the rollback or update function will be called after this.
5513  * It returns an error (and rolls back, as required), if any nfsvno_advlock()
5514  * call fails. If it returns an error, it will unlock the list.
5515  */
5516 static int
5517 nfsrv_locallock(vnode_t vp, struct nfslockfile *lfp, int flags,
5518     uint64_t first, uint64_t end, struct nfslockconflict *cfp, NFSPROC_T *p)
5519 {
5520 	struct nfslock *lop, *nlop;
5521 	int error = 0;
5522 
5523 	/* Loop through the list of locks. */
5524 	lop = LIST_FIRST(&lfp->lf_locallock);
5525 	while (first < end && lop != NULL) {
5526 		nlop = LIST_NEXT(lop, lo_lckowner);
5527 		if (first >= lop->lo_end) {
5528 			/* not there yet */
5529 			lop = nlop;
5530 		} else if (first < lop->lo_first) {
5531 			/* new one starts before entry in list */
5532 			if (end <= lop->lo_first) {
5533 				/* no overlap between old and new */
5534 				error = nfsrv_dolocal(vp, lfp, flags,
5535 				    NFSLCK_UNLOCK, 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 				    NFSLCK_UNLOCK, first, lop->lo_first, cfp,
5543 				    p);
5544 				if (error != 0)
5545 					break;
5546 				first = lop->lo_first;
5547 			}
5548 		} else {
5549 			/* new one overlaps this entry in list */
5550 			if (end <= lop->lo_end) {
5551 				/* overlaps all of new one */
5552 				error = nfsrv_dolocal(vp, lfp, flags,
5553 				    lop->lo_flags, first, end, cfp, p);
5554 				if (error != 0)
5555 					break;
5556 				first = end;
5557 			} else {
5558 				/* handle fragment overlapped with new one */
5559 				error = nfsrv_dolocal(vp, lfp, flags,
5560 				    lop->lo_flags, first, lop->lo_end, cfp, p);
5561 				if (error != 0)
5562 					break;
5563 				first = lop->lo_end;
5564 				lop = nlop;
5565 			}
5566 		}
5567 	}
5568 	if (first < end && error == 0)
5569 		/* handle fragment past end of list */
5570 		error = nfsrv_dolocal(vp, lfp, flags, NFSLCK_UNLOCK, first,
5571 		    end, cfp, p);
5572 
5573 	NFSEXITCODE(error);
5574 	return (error);
5575 }
5576 
5577 /*
5578  * Local lock unlock. Unlock all byte ranges that are no longer locked
5579  * by NFSv4. To do this, unlock any subranges of first-->end that
5580  * do not overlap with the byte ranges of any lock in the lfp->lf_lock
5581  * list. This list has all locks for the file held by other
5582  * <clientid, lockowner> tuples. The list is ordered by increasing
5583  * lo_first value, but may have entries that overlap each other, for
5584  * the case of read locks.
5585  */
5586 static void
5587 nfsrv_localunlock(vnode_t vp, struct nfslockfile *lfp, uint64_t init_first,
5588     uint64_t init_end, NFSPROC_T *p)
5589 {
5590 	struct nfslock *lop;
5591 	uint64_t first, end, prevfirst;
5592 
5593 	first = init_first;
5594 	end = init_end;
5595 	while (first < init_end) {
5596 		/* Loop through all nfs locks, adjusting first and end */
5597 		prevfirst = 0;
5598 		LIST_FOREACH(lop, &lfp->lf_lock, lo_lckfile) {
5599 			KASSERT(prevfirst <= lop->lo_first,
5600 			    ("nfsv4 locks out of order"));
5601 			KASSERT(lop->lo_first < lop->lo_end,
5602 			    ("nfsv4 bogus lock"));
5603 			prevfirst = lop->lo_first;
5604 			if (first >= lop->lo_first &&
5605 			    first < lop->lo_end)
5606 				/*
5607 				 * Overlaps with initial part, so trim
5608 				 * off that initial part by moving first past
5609 				 * it.
5610 				 */
5611 				first = lop->lo_end;
5612 			else if (end > lop->lo_first &&
5613 			    lop->lo_first > first) {
5614 				/*
5615 				 * This lock defines the end of the
5616 				 * segment to unlock, so set end to the
5617 				 * start of it and break out of the loop.
5618 				 */
5619 				end = lop->lo_first;
5620 				break;
5621 			}
5622 			if (first >= end)
5623 				/*
5624 				 * There is no segment left to do, so
5625 				 * break out of this loop and then exit
5626 				 * the outer while() since first will be set
5627 				 * to end, which must equal init_end here.
5628 				 */
5629 				break;
5630 		}
5631 		if (first < end) {
5632 			/* Unlock this segment */
5633 			(void) nfsrv_dolocal(vp, lfp, NFSLCK_UNLOCK,
5634 			    NFSLCK_READ, first, end, NULL, p);
5635 			nfsrv_locallock_commit(lfp, NFSLCK_UNLOCK,
5636 			    first, end);
5637 		}
5638 		/*
5639 		 * Now move past this segment and look for any further
5640 		 * segment in the range, if there is one.
5641 		 */
5642 		first = end;
5643 		end = init_end;
5644 	}
5645 }
5646 
5647 /*
5648  * Do the local lock operation and update the rollback list, as required.
5649  * Perform the rollback and return the error if nfsvno_advlock() fails.
5650  */
5651 static int
5652 nfsrv_dolocal(vnode_t vp, struct nfslockfile *lfp, int flags, int oldflags,
5653     uint64_t first, uint64_t end, struct nfslockconflict *cfp, NFSPROC_T *p)
5654 {
5655 	struct nfsrollback *rlp;
5656 	int error = 0, ltype, oldltype;
5657 
5658 	if (flags & NFSLCK_WRITE)
5659 		ltype = F_WRLCK;
5660 	else if (flags & NFSLCK_READ)
5661 		ltype = F_RDLCK;
5662 	else
5663 		ltype = F_UNLCK;
5664 	if (oldflags & NFSLCK_WRITE)
5665 		oldltype = F_WRLCK;
5666 	else if (oldflags & NFSLCK_READ)
5667 		oldltype = F_RDLCK;
5668 	else
5669 		oldltype = F_UNLCK;
5670 	if (ltype == oldltype || (oldltype == F_WRLCK && ltype == F_RDLCK))
5671 		/* nothing to do */
5672 		goto out;
5673 	error = nfsvno_advlock(vp, ltype, first, end, p);
5674 	if (error != 0) {
5675 		if (cfp != NULL) {
5676 			cfp->cl_clientid.lval[0] = 0;
5677 			cfp->cl_clientid.lval[1] = 0;
5678 			cfp->cl_first = 0;
5679 			cfp->cl_end = NFS64BITSSET;
5680 			cfp->cl_flags = NFSLCK_WRITE;
5681 			cfp->cl_ownerlen = 5;
5682 			NFSBCOPY("LOCAL", cfp->cl_owner, 5);
5683 		}
5684 		nfsrv_locallock_rollback(vp, lfp, p);
5685 	} else if (ltype != F_UNLCK) {
5686 		rlp = malloc(sizeof (struct nfsrollback), M_NFSDROLLBACK,
5687 		    M_WAITOK);
5688 		rlp->rlck_first = first;
5689 		rlp->rlck_end = end;
5690 		rlp->rlck_type = oldltype;
5691 		LIST_INSERT_HEAD(&lfp->lf_rollback, rlp, rlck_list);
5692 	}
5693 
5694 out:
5695 	NFSEXITCODE(error);
5696 	return (error);
5697 }
5698 
5699 /*
5700  * Roll back local lock changes and free up the rollback list.
5701  */
5702 static void
5703 nfsrv_locallock_rollback(vnode_t vp, struct nfslockfile *lfp, NFSPROC_T *p)
5704 {
5705 	struct nfsrollback *rlp, *nrlp;
5706 
5707 	LIST_FOREACH_SAFE(rlp, &lfp->lf_rollback, rlck_list, nrlp) {
5708 		(void) nfsvno_advlock(vp, rlp->rlck_type, rlp->rlck_first,
5709 		    rlp->rlck_end, p);
5710 		free(rlp, M_NFSDROLLBACK);
5711 	}
5712 	LIST_INIT(&lfp->lf_rollback);
5713 }
5714 
5715 /*
5716  * Update local lock list and delete rollback list (ie now committed to the
5717  * local locks). Most of the work is done by the internal function.
5718  */
5719 static void
5720 nfsrv_locallock_commit(struct nfslockfile *lfp, int flags, uint64_t first,
5721     uint64_t end)
5722 {
5723 	struct nfsrollback *rlp, *nrlp;
5724 	struct nfslock *new_lop, *other_lop;
5725 
5726 	new_lop = malloc(sizeof (struct nfslock), M_NFSDLOCK, M_WAITOK);
5727 	if (flags & (NFSLCK_READ | NFSLCK_WRITE))
5728 		other_lop = malloc(sizeof (struct nfslock), M_NFSDLOCK,
5729 		    M_WAITOK);
5730 	else
5731 		other_lop = NULL;
5732 	new_lop->lo_flags = flags;
5733 	new_lop->lo_first = first;
5734 	new_lop->lo_end = end;
5735 	nfsrv_updatelock(NULL, &new_lop, &other_lop, lfp);
5736 	if (new_lop != NULL)
5737 		free(new_lop, M_NFSDLOCK);
5738 	if (other_lop != NULL)
5739 		free(other_lop, M_NFSDLOCK);
5740 
5741 	/* and get rid of the rollback list */
5742 	LIST_FOREACH_SAFE(rlp, &lfp->lf_rollback, rlck_list, nrlp)
5743 		free(rlp, M_NFSDROLLBACK);
5744 	LIST_INIT(&lfp->lf_rollback);
5745 }
5746 
5747 /*
5748  * Lock the struct nfslockfile for local lock updating.
5749  */
5750 static void
5751 nfsrv_locklf(struct nfslockfile *lfp)
5752 {
5753 	int gotlock;
5754 
5755 	/* lf_usecount ensures *lfp won't be free'd */
5756 	lfp->lf_usecount++;
5757 	do {
5758 		gotlock = nfsv4_lock(&lfp->lf_locallock_lck, 1, NULL,
5759 		    NFSSTATEMUTEXPTR, NULL);
5760 	} while (gotlock == 0);
5761 	lfp->lf_usecount--;
5762 }
5763 
5764 /*
5765  * Unlock the struct nfslockfile after local lock updating.
5766  */
5767 static void
5768 nfsrv_unlocklf(struct nfslockfile *lfp)
5769 {
5770 
5771 	nfsv4_unlock(&lfp->lf_locallock_lck, 0);
5772 }
5773 
5774 /*
5775  * Clear out all state for the NFSv4 server.
5776  * Must be called by a thread that can sleep when no nfsds are running.
5777  */
5778 void
5779 nfsrv_throwawayallstate(NFSPROC_T *p)
5780 {
5781 	struct nfsclient *clp, *nclp;
5782 	struct nfslockfile *lfp, *nlfp;
5783 	int i;
5784 
5785 	/*
5786 	 * For each client, clean out the state and then free the structure.
5787 	 */
5788 	for (i = 0; i < nfsrv_clienthashsize; i++) {
5789 		LIST_FOREACH_SAFE(clp, &nfsclienthash[i], lc_hash, nclp) {
5790 			nfsrv_cleanclient(clp, p);
5791 			nfsrv_freedeleglist(&clp->lc_deleg);
5792 			nfsrv_freedeleglist(&clp->lc_olddeleg);
5793 			free(clp->lc_stateid, M_NFSDCLIENT);
5794 			free(clp, M_NFSDCLIENT);
5795 		}
5796 	}
5797 
5798 	/*
5799 	 * Also, free up any remaining lock file structures.
5800 	 */
5801 	for (i = 0; i < nfsrv_lockhashsize; i++) {
5802 		LIST_FOREACH_SAFE(lfp, &nfslockhash[i], lf_hash, nlfp)
5803 			nfsrv_freenfslockfile(lfp);
5804 	}
5805 }
5806 
5807 /*
5808  * Check the sequence# for the session and slot provided as an argument.
5809  * Also, renew the lease if the session will return NFS_OK.
5810  */
5811 int
5812 nfsrv_checksequence(struct nfsrv_descript *nd, uint32_t sequenceid,
5813     uint32_t *highest_slotidp, uint32_t *target_highest_slotidp, int cache_this,
5814     uint32_t *sflagsp, NFSPROC_T *p)
5815 {
5816 	struct nfsdsession *sep;
5817 	struct nfssessionhash *shp;
5818 	int error;
5819 	SVCXPRT *savxprt;
5820 
5821 	shp = NFSSESSIONHASH(nd->nd_sessionid);
5822 	NFSLOCKSESSION(shp);
5823 	sep = nfsrv_findsession(nd->nd_sessionid);
5824 	if (sep == NULL) {
5825 		NFSUNLOCKSESSION(shp);
5826 		return (NFSERR_BADSESSION);
5827 	}
5828 	error = nfsv4_seqsession(sequenceid, nd->nd_slotid, *highest_slotidp,
5829 	    sep->sess_slots, NULL, NFSV4_SLOTS - 1);
5830 	if (error != 0) {
5831 		NFSUNLOCKSESSION(shp);
5832 		return (error);
5833 	}
5834 	if (cache_this != 0)
5835 		nd->nd_flag |= ND_SAVEREPLY;
5836 	/* Renew the lease. */
5837 	sep->sess_clp->lc_expiry = nfsrv_leaseexpiry();
5838 	nd->nd_clientid.qval = sep->sess_clp->lc_clientid.qval;
5839 	nd->nd_flag |= ND_IMPLIEDCLID;
5840 
5841 	/*
5842 	 * If this session handles the backchannel, save the nd_xprt for this
5843 	 * RPC, since this is the one being used.
5844 	 */
5845 	if (sep->sess_clp->lc_req.nr_client != NULL &&
5846 	    (sep->sess_crflags & NFSV4CRSESS_CONNBACKCHAN) != 0) {
5847 		savxprt = sep->sess_cbsess.nfsess_xprt;
5848 		SVC_ACQUIRE(nd->nd_xprt);
5849 		nd->nd_xprt->xp_p2 =
5850 		    sep->sess_clp->lc_req.nr_client->cl_private;
5851 		nd->nd_xprt->xp_idletimeout = 0;	/* Disable timeout. */
5852 		sep->sess_cbsess.nfsess_xprt = nd->nd_xprt;
5853 		if (savxprt != NULL)
5854 			SVC_RELEASE(savxprt);
5855 	}
5856 
5857 	*sflagsp = 0;
5858 	if (sep->sess_clp->lc_req.nr_client == NULL)
5859 		*sflagsp |= NFSV4SEQ_CBPATHDOWN;
5860 	NFSUNLOCKSESSION(shp);
5861 	if (error == NFSERR_EXPIRED) {
5862 		*sflagsp |= NFSV4SEQ_EXPIREDALLSTATEREVOKED;
5863 		error = 0;
5864 	} else if (error == NFSERR_ADMINREVOKED) {
5865 		*sflagsp |= NFSV4SEQ_ADMINSTATEREVOKED;
5866 		error = 0;
5867 	}
5868 	*highest_slotidp = *target_highest_slotidp = NFSV4_SLOTS - 1;
5869 	return (0);
5870 }
5871 
5872 /*
5873  * Check/set reclaim complete for this session/clientid.
5874  */
5875 int
5876 nfsrv_checkreclaimcomplete(struct nfsrv_descript *nd)
5877 {
5878 	struct nfsdsession *sep;
5879 	struct nfssessionhash *shp;
5880 	int error = 0;
5881 
5882 	shp = NFSSESSIONHASH(nd->nd_sessionid);
5883 	NFSLOCKSTATE();
5884 	NFSLOCKSESSION(shp);
5885 	sep = nfsrv_findsession(nd->nd_sessionid);
5886 	if (sep == NULL) {
5887 		NFSUNLOCKSESSION(shp);
5888 		NFSUNLOCKSTATE();
5889 		return (NFSERR_BADSESSION);
5890 	}
5891 
5892 	/* Check to see if reclaim complete has already happened. */
5893 	if ((sep->sess_clp->lc_flags & LCL_RECLAIMCOMPLETE) != 0)
5894 		error = NFSERR_COMPLETEALREADY;
5895 	else
5896 		sep->sess_clp->lc_flags |= LCL_RECLAIMCOMPLETE;
5897 	NFSUNLOCKSESSION(shp);
5898 	NFSUNLOCKSTATE();
5899 	return (error);
5900 }
5901 
5902 /*
5903  * Cache the reply in a session slot.
5904  */
5905 void
5906 nfsrv_cache_session(uint8_t *sessionid, uint32_t slotid, int repstat,
5907    struct mbuf **m)
5908 {
5909 	struct nfsdsession *sep;
5910 	struct nfssessionhash *shp;
5911 
5912 	shp = NFSSESSIONHASH(sessionid);
5913 	NFSLOCKSESSION(shp);
5914 	sep = nfsrv_findsession(sessionid);
5915 	if (sep == NULL) {
5916 		NFSUNLOCKSESSION(shp);
5917 		printf("nfsrv_cache_session: no session\n");
5918 		m_freem(*m);
5919 		return;
5920 	}
5921 	nfsv4_seqsess_cacherep(slotid, sep->sess_slots, repstat, m);
5922 	NFSUNLOCKSESSION(shp);
5923 }
5924 
5925 /*
5926  * Search for a session that matches the sessionid.
5927  */
5928 static struct nfsdsession *
5929 nfsrv_findsession(uint8_t *sessionid)
5930 {
5931 	struct nfsdsession *sep;
5932 	struct nfssessionhash *shp;
5933 
5934 	shp = NFSSESSIONHASH(sessionid);
5935 	LIST_FOREACH(sep, &shp->list, sess_hash) {
5936 		if (!NFSBCMP(sessionid, sep->sess_sessionid, NFSX_V4SESSIONID))
5937 			break;
5938 	}
5939 	return (sep);
5940 }
5941 
5942 /*
5943  * Destroy a session.
5944  */
5945 int
5946 nfsrv_destroysession(struct nfsrv_descript *nd, uint8_t *sessionid)
5947 {
5948 	int error, samesess;
5949 
5950 	samesess = 0;
5951 	if (!NFSBCMP(sessionid, nd->nd_sessionid, NFSX_V4SESSIONID)) {
5952 		samesess = 1;
5953 		if ((nd->nd_flag & ND_LASTOP) == 0)
5954 			return (NFSERR_BADSESSION);
5955 	}
5956 	error = nfsrv_freesession(NULL, sessionid);
5957 	if (error == 0 && samesess != 0)
5958 		nd->nd_flag &= ~ND_HASSEQUENCE;
5959 	return (error);
5960 }
5961 
5962 /*
5963  * Free up a session structure.
5964  */
5965 static int
5966 nfsrv_freesession(struct nfsdsession *sep, uint8_t *sessionid)
5967 {
5968 	struct nfssessionhash *shp;
5969 	int i;
5970 
5971 	NFSLOCKSTATE();
5972 	if (sep == NULL) {
5973 		shp = NFSSESSIONHASH(sessionid);
5974 		NFSLOCKSESSION(shp);
5975 		sep = nfsrv_findsession(sessionid);
5976 	} else {
5977 		shp = NFSSESSIONHASH(sep->sess_sessionid);
5978 		NFSLOCKSESSION(shp);
5979 	}
5980 	if (sep != NULL) {
5981 		sep->sess_refcnt--;
5982 		if (sep->sess_refcnt > 0) {
5983 			NFSUNLOCKSESSION(shp);
5984 			NFSUNLOCKSTATE();
5985 			return (0);
5986 		}
5987 		LIST_REMOVE(sep, sess_hash);
5988 		LIST_REMOVE(sep, sess_list);
5989 	}
5990 	NFSUNLOCKSESSION(shp);
5991 	NFSUNLOCKSTATE();
5992 	if (sep == NULL)
5993 		return (NFSERR_BADSESSION);
5994 	for (i = 0; i < NFSV4_SLOTS; i++)
5995 		if (sep->sess_slots[i].nfssl_reply != NULL)
5996 			m_freem(sep->sess_slots[i].nfssl_reply);
5997 	if (sep->sess_cbsess.nfsess_xprt != NULL)
5998 		SVC_RELEASE(sep->sess_cbsess.nfsess_xprt);
5999 	free(sep, M_NFSDSESSION);
6000 	return (0);
6001 }
6002 
6003 /*
6004  * Free a stateid.
6005  * RFC5661 says that it should fail when there are associated opens, locks
6006  * or delegations. Since stateids represent opens, I don't see how you can
6007  * free an open stateid (it will be free'd when closed), so this function
6008  * only works for lock stateids (freeing the lock_owner) or delegations.
6009  */
6010 int
6011 nfsrv_freestateid(struct nfsrv_descript *nd, nfsv4stateid_t *stateidp,
6012     NFSPROC_T *p)
6013 {
6014 	struct nfsclient *clp;
6015 	struct nfsstate *stp;
6016 	int error;
6017 
6018 	NFSLOCKSTATE();
6019 	/*
6020 	 * Look up the stateid
6021 	 */
6022 	error = nfsrv_getclient((nfsquad_t)((u_quad_t)0), CLOPS_RENEW, &clp,
6023 	    NULL, (nfsquad_t)((u_quad_t)0), 0, nd, p);
6024 	if (error == 0) {
6025 		/* First, check for a delegation. */
6026 		LIST_FOREACH(stp, &clp->lc_deleg, ls_list) {
6027 			if (!NFSBCMP(stp->ls_stateid.other, stateidp->other,
6028 			    NFSX_STATEIDOTHER))
6029 				break;
6030 		}
6031 		if (stp != NULL) {
6032 			nfsrv_freedeleg(stp);
6033 			NFSUNLOCKSTATE();
6034 			return (error);
6035 		}
6036 	}
6037 	/* Not a delegation, try for a lock_owner. */
6038 	if (error == 0)
6039 		error = nfsrv_getstate(clp, stateidp, 0, &stp);
6040 	if (error == 0 && ((stp->ls_flags & (NFSLCK_OPEN | NFSLCK_DELEGREAD |
6041 	    NFSLCK_DELEGWRITE)) != 0 || (stp->ls_flags & NFSLCK_LOCK) == 0))
6042 		/* Not a lock_owner stateid. */
6043 		error = NFSERR_LOCKSHELD;
6044 	if (error == 0 && !LIST_EMPTY(&stp->ls_lock))
6045 		error = NFSERR_LOCKSHELD;
6046 	if (error == 0)
6047 		nfsrv_freelockowner(stp, NULL, 0, p);
6048 	NFSUNLOCKSTATE();
6049 	return (error);
6050 }
6051 
6052 /*
6053  * Generate the xdr for an NFSv4.1 CBSequence Operation.
6054  */
6055 static int
6056 nfsv4_setcbsequence(struct nfsrv_descript *nd, struct nfsclient *clp,
6057     int dont_replycache, struct nfsdsession **sepp)
6058 {
6059 	struct nfsdsession *sep;
6060 	uint32_t *tl, slotseq = 0;
6061 	int maxslot, slotpos;
6062 	uint8_t sessionid[NFSX_V4SESSIONID];
6063 	int error;
6064 
6065 	error = nfsv4_getcbsession(clp, sepp);
6066 	if (error != 0)
6067 		return (error);
6068 	sep = *sepp;
6069 	(void)nfsv4_sequencelookup(NULL, &sep->sess_cbsess, &slotpos, &maxslot,
6070 	    &slotseq, sessionid);
6071 	KASSERT(maxslot >= 0, ("nfsv4_setcbsequence neg maxslot"));
6072 
6073 	/* Build the Sequence arguments. */
6074 	NFSM_BUILD(tl, uint32_t *, NFSX_V4SESSIONID + 5 * NFSX_UNSIGNED);
6075 	bcopy(sessionid, tl, NFSX_V4SESSIONID);
6076 	tl += NFSX_V4SESSIONID / NFSX_UNSIGNED;
6077 	nd->nd_slotseq = tl;
6078 	*tl++ = txdr_unsigned(slotseq);
6079 	*tl++ = txdr_unsigned(slotpos);
6080 	*tl++ = txdr_unsigned(maxslot);
6081 	if (dont_replycache == 0)
6082 		*tl++ = newnfs_true;
6083 	else
6084 		*tl++ = newnfs_false;
6085 	*tl = 0;			/* No referring call list, for now. */
6086 	nd->nd_flag |= ND_HASSEQUENCE;
6087 	return (0);
6088 }
6089 
6090 /*
6091  * Get a session for the callback.
6092  */
6093 static int
6094 nfsv4_getcbsession(struct nfsclient *clp, struct nfsdsession **sepp)
6095 {
6096 	struct nfsdsession *sep;
6097 
6098 	NFSLOCKSTATE();
6099 	LIST_FOREACH(sep, &clp->lc_session, sess_list) {
6100 		if ((sep->sess_crflags & NFSV4CRSESS_CONNBACKCHAN) != 0)
6101 			break;
6102 	}
6103 	if (sep == NULL) {
6104 		NFSUNLOCKSTATE();
6105 		return (NFSERR_BADSESSION);
6106 	}
6107 	sep->sess_refcnt++;
6108 	*sepp = sep;
6109 	NFSUNLOCKSTATE();
6110 	return (0);
6111 }
6112 
6113 /*
6114  * Free up all backchannel xprts.  This needs to be done when the nfsd threads
6115  * exit, since those transports will all be going away.
6116  * This is only called after all the nfsd threads are done performing RPCs,
6117  * so locking shouldn't be an issue.
6118  */
6119 APPLESTATIC void
6120 nfsrv_freeallbackchannel_xprts(void)
6121 {
6122 	struct nfsdsession *sep;
6123 	struct nfsclient *clp;
6124 	SVCXPRT *xprt;
6125 	int i;
6126 
6127 	for (i = 0; i < nfsrv_clienthashsize; i++) {
6128 		LIST_FOREACH(clp, &nfsclienthash[i], lc_hash) {
6129 			LIST_FOREACH(sep, &clp->lc_session, sess_list) {
6130 				xprt = sep->sess_cbsess.nfsess_xprt;
6131 				sep->sess_cbsess.nfsess_xprt = NULL;
6132 				if (xprt != NULL)
6133 					SVC_RELEASE(xprt);
6134 			}
6135 		}
6136 	}
6137 }
6138 
6139