xref: /freebsd/sys/fs/nfsserver/nfs_nfsdstate.c (revision 8df8b2d3e51d1b816201d8a1fe8bc29fe192e562)
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 <sys/extattr.h>
35 #include <fs/nfs/nfsport.h>
36 
37 struct nfsrv_stablefirst nfsrv_stablefirst;
38 int nfsrv_issuedelegs = 0;
39 int nfsrv_dolocallocks = 0;
40 struct nfsv4lock nfsv4rootfs_lock;
41 time_t nfsdev_time = 0;
42 int nfsrv_layouthashsize;
43 volatile int nfsrv_layoutcnt = 0;
44 
45 extern int newnfs_numnfsd;
46 extern struct nfsstatsv1 nfsstatsv1;
47 extern int nfsrv_lease;
48 extern struct timeval nfsboottime;
49 extern u_int32_t newnfs_true, newnfs_false;
50 extern struct mtx nfsrv_dslock_mtx;
51 extern struct mtx nfsrv_recalllock_mtx;
52 extern struct mtx nfsrv_dontlistlock_mtx;
53 extern int nfsd_debuglevel;
54 extern u_int nfsrv_dsdirsize;
55 extern struct nfsdevicehead nfsrv_devidhead;
56 extern int nfsrv_doflexfile;
57 extern int nfsrv_maxpnfsmirror;
58 NFSV4ROOTLOCKMUTEX;
59 NFSSTATESPINLOCK;
60 extern struct nfsdontlisthead nfsrv_dontlisthead;
61 extern volatile int nfsrv_devidcnt;
62 extern struct nfslayouthead nfsrv_recalllisthead;
63 extern char *nfsrv_zeropnfsdat;
64 
65 SYSCTL_DECL(_vfs_nfsd);
66 int	nfsrv_statehashsize = NFSSTATEHASHSIZE;
67 SYSCTL_INT(_vfs_nfsd, OID_AUTO, statehashsize, CTLFLAG_RDTUN,
68     &nfsrv_statehashsize, 0,
69     "Size of state hash table set via loader.conf");
70 
71 int	nfsrv_clienthashsize = NFSCLIENTHASHSIZE;
72 SYSCTL_INT(_vfs_nfsd, OID_AUTO, clienthashsize, CTLFLAG_RDTUN,
73     &nfsrv_clienthashsize, 0,
74     "Size of client hash table set via loader.conf");
75 
76 int	nfsrv_lockhashsize = NFSLOCKHASHSIZE;
77 SYSCTL_INT(_vfs_nfsd, OID_AUTO, fhhashsize, CTLFLAG_RDTUN,
78     &nfsrv_lockhashsize, 0,
79     "Size of file handle hash table set via loader.conf");
80 
81 int	nfsrv_sessionhashsize = NFSSESSIONHASHSIZE;
82 SYSCTL_INT(_vfs_nfsd, OID_AUTO, sessionhashsize, CTLFLAG_RDTUN,
83     &nfsrv_sessionhashsize, 0,
84     "Size of session hash table set via loader.conf");
85 
86 int	nfsrv_layouthighwater = NFSLAYOUTHIGHWATER;
87 SYSCTL_INT(_vfs_nfsd, OID_AUTO, layouthighwater, CTLFLAG_RDTUN,
88     &nfsrv_layouthighwater, 0,
89     "High water mark for number of layouts set via loader.conf");
90 
91 static int	nfsrv_v4statelimit = NFSRV_V4STATELIMIT;
92 SYSCTL_INT(_vfs_nfsd, OID_AUTO, v4statelimit, CTLFLAG_RWTUN,
93     &nfsrv_v4statelimit, 0,
94     "High water limit for NFSv4 opens+locks+delegations");
95 
96 static int	nfsrv_writedelegifpos = 0;
97 SYSCTL_INT(_vfs_nfsd, OID_AUTO, writedelegifpos, CTLFLAG_RW,
98     &nfsrv_writedelegifpos, 0,
99     "Issue a write delegation for read opens if possible");
100 
101 static int	nfsrv_allowreadforwriteopen = 1;
102 SYSCTL_INT(_vfs_nfsd, OID_AUTO, allowreadforwriteopen, CTLFLAG_RW,
103     &nfsrv_allowreadforwriteopen, 0,
104     "Allow Reads to be done with Write Access StateIDs");
105 
106 int	nfsrv_pnfsatime = 0;
107 SYSCTL_INT(_vfs_nfsd, OID_AUTO, pnfsstrictatime, CTLFLAG_RW,
108     &nfsrv_pnfsatime, 0,
109     "For pNFS service, do Getattr ops to keep atime up-to-date");
110 
111 int	nfsrv_flexlinuxhack = 0;
112 SYSCTL_INT(_vfs_nfsd, OID_AUTO, flexlinuxhack, CTLFLAG_RW,
113     &nfsrv_flexlinuxhack, 0,
114     "For Linux clients, hack around Flex File Layout bug");
115 
116 /*
117  * Hash lists for nfs V4.
118  */
119 struct nfsclienthashhead	*nfsclienthash;
120 struct nfslockhashhead		*nfslockhash;
121 struct nfssessionhash		*nfssessionhash;
122 struct nfslayouthash		*nfslayouthash;
123 volatile int nfsrv_dontlistlen = 0;
124 #endif	/* !APPLEKEXT */
125 
126 static u_int32_t nfsrv_openpluslock = 0, nfsrv_delegatecnt = 0;
127 static time_t nfsrvboottime;
128 static int nfsrv_returnoldstateid = 0, nfsrv_clients = 0;
129 static int nfsrv_clienthighwater = NFSRV_CLIENTHIGHWATER;
130 static int nfsrv_nogsscallback = 0;
131 static volatile int nfsrv_writedelegcnt = 0;
132 static int nfsrv_faildscnt;
133 
134 /* local functions */
135 static void nfsrv_dumpaclient(struct nfsclient *clp,
136     struct nfsd_dumpclients *dumpp);
137 static void nfsrv_freeopenowner(struct nfsstate *stp, int cansleep,
138     NFSPROC_T *p);
139 static int nfsrv_freeopen(struct nfsstate *stp, vnode_t vp, int cansleep,
140     NFSPROC_T *p);
141 static void nfsrv_freelockowner(struct nfsstate *stp, vnode_t vp, int cansleep,
142     NFSPROC_T *p);
143 static void nfsrv_freeallnfslocks(struct nfsstate *stp, vnode_t vp,
144     int cansleep, NFSPROC_T *p);
145 static void nfsrv_freenfslock(struct nfslock *lop);
146 static void nfsrv_freenfslockfile(struct nfslockfile *lfp);
147 static void nfsrv_freedeleg(struct nfsstate *);
148 static int nfsrv_getstate(struct nfsclient *clp, nfsv4stateid_t *stateidp,
149     u_int32_t flags, struct nfsstate **stpp);
150 static void nfsrv_getowner(struct nfsstatehead *hp, struct nfsstate *new_stp,
151     struct nfsstate **stpp);
152 static int nfsrv_getlockfh(vnode_t vp, u_short flags,
153     struct nfslockfile *new_lfp, fhandle_t *nfhp, NFSPROC_T *p);
154 static int nfsrv_getlockfile(u_short flags, struct nfslockfile **new_lfpp,
155     struct nfslockfile **lfpp, fhandle_t *nfhp, int lockit);
156 static void nfsrv_insertlock(struct nfslock *new_lop,
157     struct nfslock *insert_lop, struct nfsstate *stp, struct nfslockfile *lfp);
158 static void nfsrv_updatelock(struct nfsstate *stp, struct nfslock **new_lopp,
159     struct nfslock **other_lopp, struct nfslockfile *lfp);
160 static int nfsrv_getipnumber(u_char *cp);
161 static int nfsrv_checkrestart(nfsquad_t clientid, u_int32_t flags,
162     nfsv4stateid_t *stateidp, int specialid);
163 static int nfsrv_checkgrace(struct nfsrv_descript *nd, struct nfsclient *clp,
164     u_int32_t flags);
165 static int nfsrv_docallback(struct nfsclient *clp, int procnum,
166     nfsv4stateid_t *stateidp, int trunc, fhandle_t *fhp,
167     struct nfsvattr *nap, nfsattrbit_t *attrbitp, int laytype, NFSPROC_T *p);
168 static int nfsrv_cbcallargs(struct nfsrv_descript *nd, struct nfsclient *clp,
169     uint32_t callback, int op, const char *optag, struct nfsdsession **sepp);
170 static u_int32_t nfsrv_nextclientindex(void);
171 static u_int32_t nfsrv_nextstateindex(struct nfsclient *clp);
172 static void nfsrv_markstable(struct nfsclient *clp);
173 static void nfsrv_markreclaim(struct nfsclient *clp);
174 static int nfsrv_checkstable(struct nfsclient *clp);
175 static int nfsrv_clientconflict(struct nfsclient *clp, int *haslockp, struct
176     vnode *vp, NFSPROC_T *p);
177 static int nfsrv_delegconflict(struct nfsstate *stp, int *haslockp,
178     NFSPROC_T *p, vnode_t vp);
179 static int nfsrv_cleandeleg(vnode_t vp, struct nfslockfile *lfp,
180     struct nfsclient *clp, int *haslockp, NFSPROC_T *p);
181 static int nfsrv_notsamecredname(struct nfsrv_descript *nd,
182     struct nfsclient *clp);
183 static time_t nfsrv_leaseexpiry(void);
184 static void nfsrv_delaydelegtimeout(struct nfsstate *stp);
185 static int nfsrv_checkseqid(struct nfsrv_descript *nd, u_int32_t seqid,
186     struct nfsstate *stp, struct nfsrvcache *op);
187 static int nfsrv_nootherstate(struct nfsstate *stp);
188 static int nfsrv_locallock(vnode_t vp, struct nfslockfile *lfp, int flags,
189     uint64_t first, uint64_t end, struct nfslockconflict *cfp, NFSPROC_T *p);
190 static void nfsrv_localunlock(vnode_t vp, struct nfslockfile *lfp,
191     uint64_t init_first, uint64_t init_end, NFSPROC_T *p);
192 static int nfsrv_dolocal(vnode_t vp, struct nfslockfile *lfp, int flags,
193     int oldflags, uint64_t first, uint64_t end, struct nfslockconflict *cfp,
194     NFSPROC_T *p);
195 static void nfsrv_locallock_rollback(vnode_t vp, struct nfslockfile *lfp,
196     NFSPROC_T *p);
197 static void nfsrv_locallock_commit(struct nfslockfile *lfp, int flags,
198     uint64_t first, uint64_t end);
199 static void nfsrv_locklf(struct nfslockfile *lfp);
200 static void nfsrv_unlocklf(struct nfslockfile *lfp);
201 static struct nfsdsession *nfsrv_findsession(uint8_t *sessionid);
202 static int nfsrv_freesession(struct nfsdsession *sep, uint8_t *sessionid);
203 static int nfsv4_setcbsequence(struct nfsrv_descript *nd, struct nfsclient *clp,
204     int dont_replycache, struct nfsdsession **sepp);
205 static int nfsv4_getcbsession(struct nfsclient *clp, struct nfsdsession **sepp);
206 static int nfsrv_addlayout(struct nfsrv_descript *nd, struct nfslayout **lypp,
207     nfsv4stateid_t *stateidp, char *layp, int *layoutlenp, NFSPROC_T *p);
208 static void nfsrv_freelayout(struct nfslayouthead *lhp, struct nfslayout *lyp);
209 static void nfsrv_freelayoutlist(nfsquad_t clientid);
210 static void nfsrv_freelayouts(nfsquad_t *clid, fsid_t *fs, int laytype,
211     int iomode);
212 static void nfsrv_freealllayouts(void);
213 static void nfsrv_freedevid(struct nfsdevice *ds);
214 static int nfsrv_setdsserver(char *dspathp, char *mdspathp, NFSPROC_T *p,
215     struct nfsdevice **dsp);
216 static int nfsrv_delds(char *devid, NFSPROC_T *p);
217 static void nfsrv_deleteds(struct nfsdevice *fndds);
218 static void nfsrv_allocdevid(struct nfsdevice *ds, char *addr, char *dnshost);
219 static void nfsrv_freealldevids(void);
220 static void nfsrv_flexlayouterr(struct nfsrv_descript *nd, uint32_t *layp,
221     int maxcnt, NFSPROC_T *p);
222 static int nfsrv_recalllayout(nfsquad_t clid, nfsv4stateid_t *stateidp,
223     fhandle_t *fhp, struct nfslayout *lyp, int changed, int laytype,
224     NFSPROC_T *p);
225 static int nfsrv_findlayout(nfsquad_t *clientidp, fhandle_t *fhp, int laytype,
226     NFSPROC_T *, struct nfslayout **lypp);
227 static int nfsrv_fndclid(nfsquad_t *clidvec, nfsquad_t clid, int clidcnt);
228 static struct nfslayout *nfsrv_filelayout(struct nfsrv_descript *nd, int iomode,
229     fhandle_t *fhp, fhandle_t *dsfhp, char *devid, fsid_t fs);
230 static struct nfslayout *nfsrv_flexlayout(struct nfsrv_descript *nd, int iomode,
231     int mirrorcnt, fhandle_t *fhp, fhandle_t *dsfhp, char *devid, fsid_t fs);
232 static int nfsrv_dontlayout(fhandle_t *fhp);
233 static int nfsrv_createdsfile(vnode_t vp, fhandle_t *fhp, struct pnfsdsfile *pf,
234     vnode_t dvp, struct nfsdevice *ds, struct ucred *cred, NFSPROC_T *p,
235     vnode_t *tvpp);
236 static struct nfsdevice *nfsrv_findmirroredds(struct nfsmount *nmp);
237 
238 /*
239  * Scan the client list for a match and either return the current one,
240  * create a new entry or return an error.
241  * If returning a non-error, the clp structure must either be linked into
242  * the client list or free'd.
243  */
244 APPLESTATIC int
245 nfsrv_setclient(struct nfsrv_descript *nd, struct nfsclient **new_clpp,
246     nfsquad_t *clientidp, nfsquad_t *confirmp, NFSPROC_T *p)
247 {
248 	struct nfsclient *clp = NULL, *new_clp = *new_clpp;
249 	int i, error = 0, ret;
250 	struct nfsstate *stp, *tstp;
251 	struct sockaddr_in *sad, *rad;
252 	struct nfsdsession *sep, *nsep;
253 	int zapit = 0, gotit, hasstate = 0, igotlock;
254 	static u_int64_t confirm_index = 0;
255 
256 	/*
257 	 * Check for state resource limit exceeded.
258 	 */
259 	if (nfsrv_openpluslock > nfsrv_v4statelimit) {
260 		error = NFSERR_RESOURCE;
261 		goto out;
262 	}
263 
264 	if (nfsrv_issuedelegs == 0 ||
265 	    ((nd->nd_flag & ND_GSS) != 0 && nfsrv_nogsscallback != 0))
266 		/*
267 		 * Don't do callbacks when delegations are disabled or
268 		 * for AUTH_GSS unless enabled via nfsrv_nogsscallback.
269 		 * If establishing a callback connection is attempted
270 		 * when a firewall is blocking the callback path, the
271 		 * server may wait too long for the connect attempt to
272 		 * succeed during the Open. Some clients, such as Linux,
273 		 * may timeout and give up on the Open before the server
274 		 * replies. Also, since AUTH_GSS callbacks are not
275 		 * yet interoperability tested, they might cause the
276 		 * server to crap out, if they get past the Init call to
277 		 * the client.
278 		 */
279 		new_clp->lc_program = 0;
280 
281 	/* Lock out other nfsd threads */
282 	NFSLOCKV4ROOTMUTEX();
283 	nfsv4_relref(&nfsv4rootfs_lock);
284 	do {
285 		igotlock = nfsv4_lock(&nfsv4rootfs_lock, 1, NULL,
286 		    NFSV4ROOTLOCKMUTEXPTR, NULL);
287 	} while (!igotlock);
288 	NFSUNLOCKV4ROOTMUTEX();
289 
290 	/*
291 	 * Search for a match in the client list.
292 	 */
293 	gotit = i = 0;
294 	while (i < nfsrv_clienthashsize && !gotit) {
295 	    LIST_FOREACH(clp, &nfsclienthash[i], lc_hash) {
296 		if (new_clp->lc_idlen == clp->lc_idlen &&
297 		    !NFSBCMP(new_clp->lc_id, clp->lc_id, clp->lc_idlen)) {
298 			gotit = 1;
299 			break;
300 		}
301 	    }
302 	    if (gotit == 0)
303 		i++;
304 	}
305 	if (!gotit ||
306 	    (clp->lc_flags & (LCL_NEEDSCONFIRM | LCL_ADMINREVOKED))) {
307 		if ((nd->nd_flag & ND_NFSV41) != 0 && confirmp->lval[1] != 0) {
308 			/*
309 			 * For NFSv4.1, if confirmp->lval[1] is non-zero, the
310 			 * client is trying to update a confirmed clientid.
311 			 */
312 			NFSLOCKV4ROOTMUTEX();
313 			nfsv4_unlock(&nfsv4rootfs_lock, 1);
314 			NFSUNLOCKV4ROOTMUTEX();
315 			confirmp->lval[1] = 0;
316 			error = NFSERR_NOENT;
317 			goto out;
318 		}
319 		/*
320 		 * Get rid of the old one.
321 		 */
322 		if (i != nfsrv_clienthashsize) {
323 			LIST_REMOVE(clp, lc_hash);
324 			nfsrv_cleanclient(clp, p);
325 			nfsrv_freedeleglist(&clp->lc_deleg);
326 			nfsrv_freedeleglist(&clp->lc_olddeleg);
327 			zapit = 1;
328 		}
329 		/*
330 		 * Add it after assigning a client id to it.
331 		 */
332 		new_clp->lc_flags |= LCL_NEEDSCONFIRM;
333 		if ((nd->nd_flag & ND_NFSV41) != 0)
334 			new_clp->lc_confirm.lval[0] = confirmp->lval[0] =
335 			    ++confirm_index;
336 		else
337 			confirmp->qval = new_clp->lc_confirm.qval =
338 			    ++confirm_index;
339 		clientidp->lval[0] = new_clp->lc_clientid.lval[0] =
340 		    (u_int32_t)nfsrvboottime;
341 		clientidp->lval[1] = new_clp->lc_clientid.lval[1] =
342 		    nfsrv_nextclientindex();
343 		new_clp->lc_stateindex = 0;
344 		new_clp->lc_statemaxindex = 0;
345 		new_clp->lc_cbref = 0;
346 		new_clp->lc_expiry = nfsrv_leaseexpiry();
347 		LIST_INIT(&new_clp->lc_open);
348 		LIST_INIT(&new_clp->lc_deleg);
349 		LIST_INIT(&new_clp->lc_olddeleg);
350 		LIST_INIT(&new_clp->lc_session);
351 		for (i = 0; i < nfsrv_statehashsize; i++)
352 			LIST_INIT(&new_clp->lc_stateid[i]);
353 		LIST_INSERT_HEAD(NFSCLIENTHASH(new_clp->lc_clientid), new_clp,
354 		    lc_hash);
355 		nfsstatsv1.srvclients++;
356 		nfsrv_openpluslock++;
357 		nfsrv_clients++;
358 		NFSLOCKV4ROOTMUTEX();
359 		nfsv4_unlock(&nfsv4rootfs_lock, 1);
360 		NFSUNLOCKV4ROOTMUTEX();
361 		if (zapit)
362 			nfsrv_zapclient(clp, p);
363 		*new_clpp = NULL;
364 		goto out;
365 	}
366 
367 	/*
368 	 * Now, handle the cases where the id is already issued.
369 	 */
370 	if (nfsrv_notsamecredname(nd, clp)) {
371 	    /*
372 	     * Check to see if there is expired state that should go away.
373 	     */
374 	    if (clp->lc_expiry < NFSD_MONOSEC &&
375 	        (!LIST_EMPTY(&clp->lc_open) || !LIST_EMPTY(&clp->lc_deleg))) {
376 		nfsrv_cleanclient(clp, p);
377 		nfsrv_freedeleglist(&clp->lc_deleg);
378 	    }
379 
380 	    /*
381 	     * If there is outstanding state, then reply NFSERR_CLIDINUSE per
382 	     * RFC3530 Sec. 8.1.2 last para.
383 	     */
384 	    if (!LIST_EMPTY(&clp->lc_deleg)) {
385 		hasstate = 1;
386 	    } else if (LIST_EMPTY(&clp->lc_open)) {
387 		hasstate = 0;
388 	    } else {
389 		hasstate = 0;
390 		/* Look for an Open on the OpenOwner */
391 		LIST_FOREACH(stp, &clp->lc_open, ls_list) {
392 		    if (!LIST_EMPTY(&stp->ls_open)) {
393 			hasstate = 1;
394 			break;
395 		    }
396 		}
397 	    }
398 	    if (hasstate) {
399 		/*
400 		 * If the uid doesn't match, return NFSERR_CLIDINUSE after
401 		 * filling out the correct ipaddr and portnum.
402 		 */
403 		sad = NFSSOCKADDR(new_clp->lc_req.nr_nam, struct sockaddr_in *);
404 		rad = NFSSOCKADDR(clp->lc_req.nr_nam, struct sockaddr_in *);
405 		sad->sin_addr.s_addr = rad->sin_addr.s_addr;
406 		sad->sin_port = rad->sin_port;
407 		NFSLOCKV4ROOTMUTEX();
408 		nfsv4_unlock(&nfsv4rootfs_lock, 1);
409 		NFSUNLOCKV4ROOTMUTEX();
410 		error = NFSERR_CLIDINUSE;
411 		goto out;
412 	    }
413 	}
414 
415 	if (NFSBCMP(new_clp->lc_verf, clp->lc_verf, NFSX_VERF)) {
416 		/*
417 		 * If the verifier has changed, the client has rebooted
418 		 * and a new client id is issued. The old state info
419 		 * can be thrown away once the SETCLIENTID_CONFIRM occurs.
420 		 */
421 		LIST_REMOVE(clp, lc_hash);
422 
423 		/* Get rid of all sessions on this clientid. */
424 		LIST_FOREACH_SAFE(sep, &clp->lc_session, sess_list, nsep) {
425 			ret = nfsrv_freesession(sep, NULL);
426 			if (ret != 0)
427 				printf("nfsrv_setclient: verifier changed free"
428 				    " session failed=%d\n", ret);
429 		}
430 
431 		new_clp->lc_flags |= LCL_NEEDSCONFIRM;
432 		if ((nd->nd_flag & ND_NFSV41) != 0)
433 			new_clp->lc_confirm.lval[0] = confirmp->lval[0] =
434 			    ++confirm_index;
435 		else
436 			confirmp->qval = new_clp->lc_confirm.qval =
437 			    ++confirm_index;
438 		clientidp->lval[0] = new_clp->lc_clientid.lval[0] =
439 		    nfsrvboottime;
440 		clientidp->lval[1] = new_clp->lc_clientid.lval[1] =
441 		    nfsrv_nextclientindex();
442 		new_clp->lc_stateindex = 0;
443 		new_clp->lc_statemaxindex = 0;
444 		new_clp->lc_cbref = 0;
445 		new_clp->lc_expiry = nfsrv_leaseexpiry();
446 
447 		/*
448 		 * Save the state until confirmed.
449 		 */
450 		LIST_NEWHEAD(&new_clp->lc_open, &clp->lc_open, ls_list);
451 		LIST_FOREACH(tstp, &new_clp->lc_open, ls_list)
452 			tstp->ls_clp = new_clp;
453 		LIST_NEWHEAD(&new_clp->lc_deleg, &clp->lc_deleg, ls_list);
454 		LIST_FOREACH(tstp, &new_clp->lc_deleg, ls_list)
455 			tstp->ls_clp = new_clp;
456 		LIST_NEWHEAD(&new_clp->lc_olddeleg, &clp->lc_olddeleg,
457 		    ls_list);
458 		LIST_FOREACH(tstp, &new_clp->lc_olddeleg, ls_list)
459 			tstp->ls_clp = new_clp;
460 		for (i = 0; i < nfsrv_statehashsize; i++) {
461 			LIST_NEWHEAD(&new_clp->lc_stateid[i],
462 			    &clp->lc_stateid[i], ls_hash);
463 			LIST_FOREACH(tstp, &new_clp->lc_stateid[i], ls_hash)
464 				tstp->ls_clp = new_clp;
465 		}
466 		LIST_INIT(&new_clp->lc_session);
467 		LIST_INSERT_HEAD(NFSCLIENTHASH(new_clp->lc_clientid), new_clp,
468 		    lc_hash);
469 		nfsstatsv1.srvclients++;
470 		nfsrv_openpluslock++;
471 		nfsrv_clients++;
472 		NFSLOCKV4ROOTMUTEX();
473 		nfsv4_unlock(&nfsv4rootfs_lock, 1);
474 		NFSUNLOCKV4ROOTMUTEX();
475 
476 		/*
477 		 * Must wait until any outstanding callback on the old clp
478 		 * completes.
479 		 */
480 		NFSLOCKSTATE();
481 		while (clp->lc_cbref) {
482 			clp->lc_flags |= LCL_WAKEUPWANTED;
483 			(void)mtx_sleep(clp, NFSSTATEMUTEXPTR, PZERO - 1,
484 			    "nfsd clp", 10 * hz);
485 		}
486 		NFSUNLOCKSTATE();
487 		nfsrv_zapclient(clp, p);
488 		*new_clpp = NULL;
489 		goto out;
490 	}
491 
492 	/* For NFSv4.1, mark that we found a confirmed clientid. */
493 	if ((nd->nd_flag & ND_NFSV41) != 0) {
494 		clientidp->lval[0] = clp->lc_clientid.lval[0];
495 		clientidp->lval[1] = clp->lc_clientid.lval[1];
496 		confirmp->lval[0] = 0;	/* Ignored by client */
497 		confirmp->lval[1] = 1;
498 	} else {
499 		/*
500 		 * id and verifier match, so update the net address info
501 		 * and get rid of any existing callback authentication
502 		 * handle, so a new one will be acquired.
503 		 */
504 		LIST_REMOVE(clp, lc_hash);
505 		new_clp->lc_flags |= (LCL_NEEDSCONFIRM | LCL_DONTCLEAN);
506 		new_clp->lc_expiry = nfsrv_leaseexpiry();
507 		confirmp->qval = new_clp->lc_confirm.qval = ++confirm_index;
508 		clientidp->lval[0] = new_clp->lc_clientid.lval[0] =
509 		    clp->lc_clientid.lval[0];
510 		clientidp->lval[1] = new_clp->lc_clientid.lval[1] =
511 		    clp->lc_clientid.lval[1];
512 		new_clp->lc_delegtime = clp->lc_delegtime;
513 		new_clp->lc_stateindex = clp->lc_stateindex;
514 		new_clp->lc_statemaxindex = clp->lc_statemaxindex;
515 		new_clp->lc_cbref = 0;
516 		LIST_NEWHEAD(&new_clp->lc_open, &clp->lc_open, ls_list);
517 		LIST_FOREACH(tstp, &new_clp->lc_open, ls_list)
518 			tstp->ls_clp = new_clp;
519 		LIST_NEWHEAD(&new_clp->lc_deleg, &clp->lc_deleg, ls_list);
520 		LIST_FOREACH(tstp, &new_clp->lc_deleg, ls_list)
521 			tstp->ls_clp = new_clp;
522 		LIST_NEWHEAD(&new_clp->lc_olddeleg, &clp->lc_olddeleg, ls_list);
523 		LIST_FOREACH(tstp, &new_clp->lc_olddeleg, ls_list)
524 			tstp->ls_clp = new_clp;
525 		for (i = 0; i < nfsrv_statehashsize; i++) {
526 			LIST_NEWHEAD(&new_clp->lc_stateid[i],
527 			    &clp->lc_stateid[i], ls_hash);
528 			LIST_FOREACH(tstp, &new_clp->lc_stateid[i], ls_hash)
529 				tstp->ls_clp = new_clp;
530 		}
531 		LIST_INIT(&new_clp->lc_session);
532 		LIST_INSERT_HEAD(NFSCLIENTHASH(new_clp->lc_clientid), new_clp,
533 		    lc_hash);
534 		nfsstatsv1.srvclients++;
535 		nfsrv_openpluslock++;
536 		nfsrv_clients++;
537 	}
538 	NFSLOCKV4ROOTMUTEX();
539 	nfsv4_unlock(&nfsv4rootfs_lock, 1);
540 	NFSUNLOCKV4ROOTMUTEX();
541 
542 	if ((nd->nd_flag & ND_NFSV41) == 0) {
543 		/*
544 		 * Must wait until any outstanding callback on the old clp
545 		 * completes.
546 		 */
547 		NFSLOCKSTATE();
548 		while (clp->lc_cbref) {
549 			clp->lc_flags |= LCL_WAKEUPWANTED;
550 			(void)mtx_sleep(clp, NFSSTATEMUTEXPTR, PZERO - 1,
551 			    "nfsdclp", 10 * hz);
552 		}
553 		NFSUNLOCKSTATE();
554 		nfsrv_zapclient(clp, p);
555 		*new_clpp = NULL;
556 	}
557 
558 out:
559 	NFSEXITCODE2(error, nd);
560 	return (error);
561 }
562 
563 /*
564  * Check to see if the client id exists and optionally confirm it.
565  */
566 APPLESTATIC int
567 nfsrv_getclient(nfsquad_t clientid, int opflags, struct nfsclient **clpp,
568     struct nfsdsession *nsep, nfsquad_t confirm, uint32_t cbprogram,
569     struct nfsrv_descript *nd, NFSPROC_T *p)
570 {
571 	struct nfsclient *clp;
572 	struct nfsstate *stp;
573 	int i;
574 	struct nfsclienthashhead *hp;
575 	int error = 0, igotlock, doneok;
576 	struct nfssessionhash *shp;
577 	struct nfsdsession *sep;
578 	uint64_t sessid[2];
579 	static uint64_t next_sess = 0;
580 
581 	if (clpp)
582 		*clpp = NULL;
583 	if ((nd == NULL || (nd->nd_flag & ND_NFSV41) == 0 ||
584 	    opflags != CLOPS_RENEW) && nfsrvboottime != clientid.lval[0]) {
585 		error = NFSERR_STALECLIENTID;
586 		goto out;
587 	}
588 
589 	/*
590 	 * If called with opflags == CLOPS_RENEW, the State Lock is
591 	 * already held. Otherwise, we need to get either that or,
592 	 * for the case of Confirm, lock out the nfsd threads.
593 	 */
594 	if (opflags & CLOPS_CONFIRM) {
595 		NFSLOCKV4ROOTMUTEX();
596 		nfsv4_relref(&nfsv4rootfs_lock);
597 		do {
598 			igotlock = nfsv4_lock(&nfsv4rootfs_lock, 1, NULL,
599 			    NFSV4ROOTLOCKMUTEXPTR, NULL);
600 		} while (!igotlock);
601 		/*
602 		 * Create a new sessionid here, since we need to do it where
603 		 * there is a mutex held to serialize update of next_sess.
604 		 */
605 		if ((nd->nd_flag & ND_NFSV41) != 0) {
606 			sessid[0] = ++next_sess;
607 			sessid[1] = clientid.qval;
608 		}
609 		NFSUNLOCKV4ROOTMUTEX();
610 	} else if (opflags != CLOPS_RENEW) {
611 		NFSLOCKSTATE();
612 	}
613 
614 	/* For NFSv4.1, the clp is acquired from the associated session. */
615 	if (nd != NULL && (nd->nd_flag & ND_NFSV41) != 0 &&
616 	    opflags == CLOPS_RENEW) {
617 		clp = NULL;
618 		if ((nd->nd_flag & ND_HASSEQUENCE) != 0) {
619 			shp = NFSSESSIONHASH(nd->nd_sessionid);
620 			NFSLOCKSESSION(shp);
621 			sep = nfsrv_findsession(nd->nd_sessionid);
622 			if (sep != NULL)
623 				clp = sep->sess_clp;
624 			NFSUNLOCKSESSION(shp);
625 		}
626 	} else {
627 		hp = NFSCLIENTHASH(clientid);
628 		LIST_FOREACH(clp, hp, lc_hash) {
629 			if (clp->lc_clientid.lval[1] == clientid.lval[1])
630 				break;
631 		}
632 	}
633 	if (clp == NULL) {
634 		if (opflags & CLOPS_CONFIRM)
635 			error = NFSERR_STALECLIENTID;
636 		else
637 			error = NFSERR_EXPIRED;
638 	} else if (clp->lc_flags & LCL_ADMINREVOKED) {
639 		/*
640 		 * If marked admin revoked, just return the error.
641 		 */
642 		error = NFSERR_ADMINREVOKED;
643 	}
644 	if (error) {
645 		if (opflags & CLOPS_CONFIRM) {
646 			NFSLOCKV4ROOTMUTEX();
647 			nfsv4_unlock(&nfsv4rootfs_lock, 1);
648 			NFSUNLOCKV4ROOTMUTEX();
649 		} else if (opflags != CLOPS_RENEW) {
650 			NFSUNLOCKSTATE();
651 		}
652 		goto out;
653 	}
654 
655 	/*
656 	 * Perform any operations specified by the opflags.
657 	 */
658 	if (opflags & CLOPS_CONFIRM) {
659 		if (((nd->nd_flag & ND_NFSV41) != 0 &&
660 		     clp->lc_confirm.lval[0] != confirm.lval[0]) ||
661 		    ((nd->nd_flag & ND_NFSV41) == 0 &&
662 		     clp->lc_confirm.qval != confirm.qval))
663 			error = NFSERR_STALECLIENTID;
664 		else if (nfsrv_notsamecredname(nd, clp))
665 			error = NFSERR_CLIDINUSE;
666 
667 		if (!error) {
668 		    if ((clp->lc_flags & (LCL_NEEDSCONFIRM | LCL_DONTCLEAN)) ==
669 			LCL_NEEDSCONFIRM) {
670 			/*
671 			 * Hang onto the delegations (as old delegations)
672 			 * for an Open with CLAIM_DELEGATE_PREV unless in
673 			 * grace, but get rid of the rest of the state.
674 			 */
675 			nfsrv_cleanclient(clp, p);
676 			nfsrv_freedeleglist(&clp->lc_olddeleg);
677 			if (nfsrv_checkgrace(nd, clp, 0)) {
678 			    /* In grace, so just delete delegations */
679 			    nfsrv_freedeleglist(&clp->lc_deleg);
680 			} else {
681 			    LIST_FOREACH(stp, &clp->lc_deleg, ls_list)
682 				stp->ls_flags |= NFSLCK_OLDDELEG;
683 			    clp->lc_delegtime = NFSD_MONOSEC +
684 				nfsrv_lease + NFSRV_LEASEDELTA;
685 			    LIST_NEWHEAD(&clp->lc_olddeleg, &clp->lc_deleg,
686 				ls_list);
687 			}
688 			if ((nd->nd_flag & ND_NFSV41) != 0)
689 			    clp->lc_program = cbprogram;
690 		    }
691 		    clp->lc_flags &= ~(LCL_NEEDSCONFIRM | LCL_DONTCLEAN);
692 		    if (clp->lc_program)
693 			clp->lc_flags |= LCL_NEEDSCBNULL;
694 		    /* For NFSv4.1, link the session onto the client. */
695 		    if (nsep != NULL) {
696 			/* Hold a reference on the xprt for a backchannel. */
697 			if ((nsep->sess_crflags & NFSV4CRSESS_CONNBACKCHAN)
698 			    != 0) {
699 			    if (clp->lc_req.nr_client == NULL)
700 				clp->lc_req.nr_client = (struct __rpc_client *)
701 				    clnt_bck_create(nd->nd_xprt->xp_socket,
702 				    cbprogram, NFSV4_CBVERS);
703 			    if (clp->lc_req.nr_client != NULL) {
704 				SVC_ACQUIRE(nd->nd_xprt);
705 				nd->nd_xprt->xp_p2 =
706 				    clp->lc_req.nr_client->cl_private;
707 				/* Disable idle timeout. */
708 				nd->nd_xprt->xp_idletimeout = 0;
709 				nsep->sess_cbsess.nfsess_xprt = nd->nd_xprt;
710 			    } else
711 				nsep->sess_crflags &= ~NFSV4CRSESS_CONNBACKCHAN;
712 			}
713 			NFSBCOPY(sessid, nsep->sess_sessionid,
714 			    NFSX_V4SESSIONID);
715 			NFSBCOPY(sessid, nsep->sess_cbsess.nfsess_sessionid,
716 			    NFSX_V4SESSIONID);
717 			shp = NFSSESSIONHASH(nsep->sess_sessionid);
718 			NFSLOCKSTATE();
719 			NFSLOCKSESSION(shp);
720 			LIST_INSERT_HEAD(&shp->list, nsep, sess_hash);
721 			LIST_INSERT_HEAD(&clp->lc_session, nsep, sess_list);
722 			nsep->sess_clp = clp;
723 			NFSUNLOCKSESSION(shp);
724 			NFSUNLOCKSTATE();
725 		    }
726 		}
727 	} else if (clp->lc_flags & LCL_NEEDSCONFIRM) {
728 		error = NFSERR_EXPIRED;
729 	}
730 
731 	/*
732 	 * If called by the Renew Op, we must check the principal.
733 	 */
734 	if (!error && (opflags & CLOPS_RENEWOP)) {
735 	    if (nfsrv_notsamecredname(nd, clp)) {
736 		doneok = 0;
737 		for (i = 0; i < nfsrv_statehashsize && doneok == 0; i++) {
738 		    LIST_FOREACH(stp, &clp->lc_stateid[i], ls_hash) {
739 			if ((stp->ls_flags & NFSLCK_OPEN) &&
740 			    stp->ls_uid == nd->nd_cred->cr_uid) {
741 				doneok = 1;
742 				break;
743 			}
744 		    }
745 		}
746 		if (!doneok)
747 			error = NFSERR_ACCES;
748 	    }
749 	    if (!error && (clp->lc_flags & LCL_CBDOWN))
750 		error = NFSERR_CBPATHDOWN;
751 	}
752 	if ((!error || error == NFSERR_CBPATHDOWN) &&
753 	     (opflags & CLOPS_RENEW)) {
754 		clp->lc_expiry = nfsrv_leaseexpiry();
755 	}
756 	if (opflags & CLOPS_CONFIRM) {
757 		NFSLOCKV4ROOTMUTEX();
758 		nfsv4_unlock(&nfsv4rootfs_lock, 1);
759 		NFSUNLOCKV4ROOTMUTEX();
760 	} else if (opflags != CLOPS_RENEW) {
761 		NFSUNLOCKSTATE();
762 	}
763 	if (clpp)
764 		*clpp = clp;
765 
766 out:
767 	NFSEXITCODE2(error, nd);
768 	return (error);
769 }
770 
771 /*
772  * Perform the NFSv4.1 destroy clientid.
773  */
774 int
775 nfsrv_destroyclient(nfsquad_t clientid, NFSPROC_T *p)
776 {
777 	struct nfsclient *clp;
778 	struct nfsclienthashhead *hp;
779 	int error = 0, i, igotlock;
780 
781 	if (nfsrvboottime != clientid.lval[0]) {
782 		error = NFSERR_STALECLIENTID;
783 		goto out;
784 	}
785 
786 	/* Lock out other nfsd threads */
787 	NFSLOCKV4ROOTMUTEX();
788 	nfsv4_relref(&nfsv4rootfs_lock);
789 	do {
790 		igotlock = nfsv4_lock(&nfsv4rootfs_lock, 1, NULL,
791 		    NFSV4ROOTLOCKMUTEXPTR, NULL);
792 	} while (igotlock == 0);
793 	NFSUNLOCKV4ROOTMUTEX();
794 
795 	hp = NFSCLIENTHASH(clientid);
796 	LIST_FOREACH(clp, hp, lc_hash) {
797 		if (clp->lc_clientid.lval[1] == clientid.lval[1])
798 			break;
799 	}
800 	if (clp == NULL) {
801 		NFSLOCKV4ROOTMUTEX();
802 		nfsv4_unlock(&nfsv4rootfs_lock, 1);
803 		NFSUNLOCKV4ROOTMUTEX();
804 		/* Just return ok, since it is gone. */
805 		goto out;
806 	}
807 
808 	/*
809 	 * Free up all layouts on the clientid.  Should the client return the
810 	 * layouts?
811 	 */
812 	nfsrv_freelayoutlist(clientid);
813 
814 	/* Scan for state on the clientid. */
815 	for (i = 0; i < nfsrv_statehashsize; i++)
816 		if (!LIST_EMPTY(&clp->lc_stateid[i])) {
817 			NFSLOCKV4ROOTMUTEX();
818 			nfsv4_unlock(&nfsv4rootfs_lock, 1);
819 			NFSUNLOCKV4ROOTMUTEX();
820 			error = NFSERR_CLIENTIDBUSY;
821 			goto out;
822 		}
823 	if (!LIST_EMPTY(&clp->lc_session) || !LIST_EMPTY(&clp->lc_deleg)) {
824 		NFSLOCKV4ROOTMUTEX();
825 		nfsv4_unlock(&nfsv4rootfs_lock, 1);
826 		NFSUNLOCKV4ROOTMUTEX();
827 		error = NFSERR_CLIENTIDBUSY;
828 		goto out;
829 	}
830 
831 	/* Destroy the clientid and return ok. */
832 	nfsrv_cleanclient(clp, p);
833 	nfsrv_freedeleglist(&clp->lc_deleg);
834 	nfsrv_freedeleglist(&clp->lc_olddeleg);
835 	LIST_REMOVE(clp, lc_hash);
836 	NFSLOCKV4ROOTMUTEX();
837 	nfsv4_unlock(&nfsv4rootfs_lock, 1);
838 	NFSUNLOCKV4ROOTMUTEX();
839 	nfsrv_zapclient(clp, p);
840 out:
841 	NFSEXITCODE2(error, nd);
842 	return (error);
843 }
844 
845 /*
846  * Called from the new nfssvc syscall to admin revoke a clientid.
847  * Returns 0 for success, error otherwise.
848  */
849 APPLESTATIC int
850 nfsrv_adminrevoke(struct nfsd_clid *revokep, NFSPROC_T *p)
851 {
852 	struct nfsclient *clp = NULL;
853 	int i, error = 0;
854 	int gotit, igotlock;
855 
856 	/*
857 	 * First, lock out the nfsd so that state won't change while the
858 	 * revocation record is being written to the stable storage restart
859 	 * file.
860 	 */
861 	NFSLOCKV4ROOTMUTEX();
862 	do {
863 		igotlock = nfsv4_lock(&nfsv4rootfs_lock, 1, NULL,
864 		    NFSV4ROOTLOCKMUTEXPTR, NULL);
865 	} while (!igotlock);
866 	NFSUNLOCKV4ROOTMUTEX();
867 
868 	/*
869 	 * Search for a match in the client list.
870 	 */
871 	gotit = i = 0;
872 	while (i < nfsrv_clienthashsize && !gotit) {
873 	    LIST_FOREACH(clp, &nfsclienthash[i], lc_hash) {
874 		if (revokep->nclid_idlen == clp->lc_idlen &&
875 		    !NFSBCMP(revokep->nclid_id, clp->lc_id, clp->lc_idlen)) {
876 			gotit = 1;
877 			break;
878 		}
879 	    }
880 	    i++;
881 	}
882 	if (!gotit) {
883 		NFSLOCKV4ROOTMUTEX();
884 		nfsv4_unlock(&nfsv4rootfs_lock, 0);
885 		NFSUNLOCKV4ROOTMUTEX();
886 		error = EPERM;
887 		goto out;
888 	}
889 
890 	/*
891 	 * Now, write out the revocation record
892 	 */
893 	nfsrv_writestable(clp->lc_id, clp->lc_idlen, NFSNST_REVOKE, p);
894 	nfsrv_backupstable();
895 
896 	/*
897 	 * and clear out the state, marking the clientid revoked.
898 	 */
899 	clp->lc_flags &= ~LCL_CALLBACKSON;
900 	clp->lc_flags |= LCL_ADMINREVOKED;
901 	nfsrv_cleanclient(clp, p);
902 	nfsrv_freedeleglist(&clp->lc_deleg);
903 	nfsrv_freedeleglist(&clp->lc_olddeleg);
904 	NFSLOCKV4ROOTMUTEX();
905 	nfsv4_unlock(&nfsv4rootfs_lock, 0);
906 	NFSUNLOCKV4ROOTMUTEX();
907 
908 out:
909 	NFSEXITCODE(error);
910 	return (error);
911 }
912 
913 /*
914  * Dump out stats for all clients. Called from nfssvc(2), that is used
915  * nfsstatsv1.
916  */
917 APPLESTATIC void
918 nfsrv_dumpclients(struct nfsd_dumpclients *dumpp, int maxcnt)
919 {
920 	struct nfsclient *clp;
921 	int i = 0, cnt = 0;
922 
923 	/*
924 	 * First, get a reference on the nfsv4rootfs_lock so that an
925 	 * exclusive lock cannot be acquired while dumping the clients.
926 	 */
927 	NFSLOCKV4ROOTMUTEX();
928 	nfsv4_getref(&nfsv4rootfs_lock, NULL, NFSV4ROOTLOCKMUTEXPTR, NULL);
929 	NFSUNLOCKV4ROOTMUTEX();
930 	NFSLOCKSTATE();
931 	/*
932 	 * Rattle through the client lists until done.
933 	 */
934 	while (i < nfsrv_clienthashsize && cnt < maxcnt) {
935 	    clp = LIST_FIRST(&nfsclienthash[i]);
936 	    while (clp != LIST_END(&nfsclienthash[i]) && cnt < maxcnt) {
937 		nfsrv_dumpaclient(clp, &dumpp[cnt]);
938 		cnt++;
939 		clp = LIST_NEXT(clp, lc_hash);
940 	    }
941 	    i++;
942 	}
943 	if (cnt < maxcnt)
944 	    dumpp[cnt].ndcl_clid.nclid_idlen = 0;
945 	NFSUNLOCKSTATE();
946 	NFSLOCKV4ROOTMUTEX();
947 	nfsv4_relref(&nfsv4rootfs_lock);
948 	NFSUNLOCKV4ROOTMUTEX();
949 }
950 
951 /*
952  * Dump stats for a client. Must be called with the NFSSTATELOCK and spl'd.
953  */
954 static void
955 nfsrv_dumpaclient(struct nfsclient *clp, struct nfsd_dumpclients *dumpp)
956 {
957 	struct nfsstate *stp, *openstp, *lckownstp;
958 	struct nfslock *lop;
959 	struct sockaddr *sad;
960 	struct sockaddr_in *rad;
961 	struct sockaddr_in6 *rad6;
962 
963 	dumpp->ndcl_nopenowners = dumpp->ndcl_nlockowners = 0;
964 	dumpp->ndcl_nopens = dumpp->ndcl_nlocks = 0;
965 	dumpp->ndcl_ndelegs = dumpp->ndcl_nolddelegs = 0;
966 	dumpp->ndcl_flags = clp->lc_flags;
967 	dumpp->ndcl_clid.nclid_idlen = clp->lc_idlen;
968 	NFSBCOPY(clp->lc_id, dumpp->ndcl_clid.nclid_id, clp->lc_idlen);
969 	sad = NFSSOCKADDR(clp->lc_req.nr_nam, struct sockaddr *);
970 	dumpp->ndcl_addrfam = sad->sa_family;
971 	if (sad->sa_family == AF_INET) {
972 		rad = (struct sockaddr_in *)sad;
973 		dumpp->ndcl_cbaddr.sin_addr = rad->sin_addr;
974 	} else {
975 		rad6 = (struct sockaddr_in6 *)sad;
976 		dumpp->ndcl_cbaddr.sin6_addr = rad6->sin6_addr;
977 	}
978 
979 	/*
980 	 * Now, scan the state lists and total up the opens and locks.
981 	 */
982 	LIST_FOREACH(stp, &clp->lc_open, ls_list) {
983 	    dumpp->ndcl_nopenowners++;
984 	    LIST_FOREACH(openstp, &stp->ls_open, ls_list) {
985 		dumpp->ndcl_nopens++;
986 		LIST_FOREACH(lckownstp, &openstp->ls_open, ls_list) {
987 		    dumpp->ndcl_nlockowners++;
988 		    LIST_FOREACH(lop, &lckownstp->ls_lock, lo_lckowner) {
989 			dumpp->ndcl_nlocks++;
990 		    }
991 		}
992 	    }
993 	}
994 
995 	/*
996 	 * and the delegation lists.
997 	 */
998 	LIST_FOREACH(stp, &clp->lc_deleg, ls_list) {
999 	    dumpp->ndcl_ndelegs++;
1000 	}
1001 	LIST_FOREACH(stp, &clp->lc_olddeleg, ls_list) {
1002 	    dumpp->ndcl_nolddelegs++;
1003 	}
1004 }
1005 
1006 /*
1007  * Dump out lock stats for a file.
1008  */
1009 APPLESTATIC void
1010 nfsrv_dumplocks(vnode_t vp, struct nfsd_dumplocks *ldumpp, int maxcnt,
1011     NFSPROC_T *p)
1012 {
1013 	struct nfsstate *stp;
1014 	struct nfslock *lop;
1015 	int cnt = 0;
1016 	struct nfslockfile *lfp;
1017 	struct sockaddr *sad;
1018 	struct sockaddr_in *rad;
1019 	struct sockaddr_in6 *rad6;
1020 	int ret;
1021 	fhandle_t nfh;
1022 
1023 	ret = nfsrv_getlockfh(vp, 0, NULL, &nfh, p);
1024 	/*
1025 	 * First, get a reference on the nfsv4rootfs_lock so that an
1026 	 * exclusive lock on it cannot be acquired while dumping the locks.
1027 	 */
1028 	NFSLOCKV4ROOTMUTEX();
1029 	nfsv4_getref(&nfsv4rootfs_lock, NULL, NFSV4ROOTLOCKMUTEXPTR, NULL);
1030 	NFSUNLOCKV4ROOTMUTEX();
1031 	NFSLOCKSTATE();
1032 	if (!ret)
1033 		ret = nfsrv_getlockfile(0, NULL, &lfp, &nfh, 0);
1034 	if (ret) {
1035 		ldumpp[0].ndlck_clid.nclid_idlen = 0;
1036 		NFSUNLOCKSTATE();
1037 		NFSLOCKV4ROOTMUTEX();
1038 		nfsv4_relref(&nfsv4rootfs_lock);
1039 		NFSUNLOCKV4ROOTMUTEX();
1040 		return;
1041 	}
1042 
1043 	/*
1044 	 * For each open share on file, dump it out.
1045 	 */
1046 	stp = LIST_FIRST(&lfp->lf_open);
1047 	while (stp != LIST_END(&lfp->lf_open) && cnt < maxcnt) {
1048 		ldumpp[cnt].ndlck_flags = stp->ls_flags;
1049 		ldumpp[cnt].ndlck_stateid.seqid = stp->ls_stateid.seqid;
1050 		ldumpp[cnt].ndlck_stateid.other[0] = stp->ls_stateid.other[0];
1051 		ldumpp[cnt].ndlck_stateid.other[1] = stp->ls_stateid.other[1];
1052 		ldumpp[cnt].ndlck_stateid.other[2] = stp->ls_stateid.other[2];
1053 		ldumpp[cnt].ndlck_owner.nclid_idlen =
1054 		    stp->ls_openowner->ls_ownerlen;
1055 		NFSBCOPY(stp->ls_openowner->ls_owner,
1056 		    ldumpp[cnt].ndlck_owner.nclid_id,
1057 		    stp->ls_openowner->ls_ownerlen);
1058 		ldumpp[cnt].ndlck_clid.nclid_idlen = stp->ls_clp->lc_idlen;
1059 		NFSBCOPY(stp->ls_clp->lc_id, ldumpp[cnt].ndlck_clid.nclid_id,
1060 		    stp->ls_clp->lc_idlen);
1061 		sad=NFSSOCKADDR(stp->ls_clp->lc_req.nr_nam, struct sockaddr *);
1062 		ldumpp[cnt].ndlck_addrfam = sad->sa_family;
1063 		if (sad->sa_family == AF_INET) {
1064 			rad = (struct sockaddr_in *)sad;
1065 			ldumpp[cnt].ndlck_cbaddr.sin_addr = rad->sin_addr;
1066 		} else {
1067 			rad6 = (struct sockaddr_in6 *)sad;
1068 			ldumpp[cnt].ndlck_cbaddr.sin6_addr = rad6->sin6_addr;
1069 		}
1070 		stp = LIST_NEXT(stp, ls_file);
1071 		cnt++;
1072 	}
1073 
1074 	/*
1075 	 * and all locks.
1076 	 */
1077 	lop = LIST_FIRST(&lfp->lf_lock);
1078 	while (lop != LIST_END(&lfp->lf_lock) && cnt < maxcnt) {
1079 		stp = lop->lo_stp;
1080 		ldumpp[cnt].ndlck_flags = lop->lo_flags;
1081 		ldumpp[cnt].ndlck_first = lop->lo_first;
1082 		ldumpp[cnt].ndlck_end = lop->lo_end;
1083 		ldumpp[cnt].ndlck_stateid.seqid = stp->ls_stateid.seqid;
1084 		ldumpp[cnt].ndlck_stateid.other[0] = stp->ls_stateid.other[0];
1085 		ldumpp[cnt].ndlck_stateid.other[1] = stp->ls_stateid.other[1];
1086 		ldumpp[cnt].ndlck_stateid.other[2] = stp->ls_stateid.other[2];
1087 		ldumpp[cnt].ndlck_owner.nclid_idlen = stp->ls_ownerlen;
1088 		NFSBCOPY(stp->ls_owner, ldumpp[cnt].ndlck_owner.nclid_id,
1089 		    stp->ls_ownerlen);
1090 		ldumpp[cnt].ndlck_clid.nclid_idlen = stp->ls_clp->lc_idlen;
1091 		NFSBCOPY(stp->ls_clp->lc_id, ldumpp[cnt].ndlck_clid.nclid_id,
1092 		    stp->ls_clp->lc_idlen);
1093 		sad=NFSSOCKADDR(stp->ls_clp->lc_req.nr_nam, struct sockaddr *);
1094 		ldumpp[cnt].ndlck_addrfam = sad->sa_family;
1095 		if (sad->sa_family == AF_INET) {
1096 			rad = (struct sockaddr_in *)sad;
1097 			ldumpp[cnt].ndlck_cbaddr.sin_addr = rad->sin_addr;
1098 		} else {
1099 			rad6 = (struct sockaddr_in6 *)sad;
1100 			ldumpp[cnt].ndlck_cbaddr.sin6_addr = rad6->sin6_addr;
1101 		}
1102 		lop = LIST_NEXT(lop, lo_lckfile);
1103 		cnt++;
1104 	}
1105 
1106 	/*
1107 	 * and the delegations.
1108 	 */
1109 	stp = LIST_FIRST(&lfp->lf_deleg);
1110 	while (stp != LIST_END(&lfp->lf_deleg) && cnt < maxcnt) {
1111 		ldumpp[cnt].ndlck_flags = stp->ls_flags;
1112 		ldumpp[cnt].ndlck_stateid.seqid = stp->ls_stateid.seqid;
1113 		ldumpp[cnt].ndlck_stateid.other[0] = stp->ls_stateid.other[0];
1114 		ldumpp[cnt].ndlck_stateid.other[1] = stp->ls_stateid.other[1];
1115 		ldumpp[cnt].ndlck_stateid.other[2] = stp->ls_stateid.other[2];
1116 		ldumpp[cnt].ndlck_owner.nclid_idlen = 0;
1117 		ldumpp[cnt].ndlck_clid.nclid_idlen = stp->ls_clp->lc_idlen;
1118 		NFSBCOPY(stp->ls_clp->lc_id, ldumpp[cnt].ndlck_clid.nclid_id,
1119 		    stp->ls_clp->lc_idlen);
1120 		sad=NFSSOCKADDR(stp->ls_clp->lc_req.nr_nam, struct sockaddr *);
1121 		ldumpp[cnt].ndlck_addrfam = sad->sa_family;
1122 		if (sad->sa_family == AF_INET) {
1123 			rad = (struct sockaddr_in *)sad;
1124 			ldumpp[cnt].ndlck_cbaddr.sin_addr = rad->sin_addr;
1125 		} else {
1126 			rad6 = (struct sockaddr_in6 *)sad;
1127 			ldumpp[cnt].ndlck_cbaddr.sin6_addr = rad6->sin6_addr;
1128 		}
1129 		stp = LIST_NEXT(stp, ls_file);
1130 		cnt++;
1131 	}
1132 
1133 	/*
1134 	 * If list isn't full, mark end of list by setting the client name
1135 	 * to zero length.
1136 	 */
1137 	if (cnt < maxcnt)
1138 		ldumpp[cnt].ndlck_clid.nclid_idlen = 0;
1139 	NFSUNLOCKSTATE();
1140 	NFSLOCKV4ROOTMUTEX();
1141 	nfsv4_relref(&nfsv4rootfs_lock);
1142 	NFSUNLOCKV4ROOTMUTEX();
1143 }
1144 
1145 /*
1146  * Server timer routine. It can scan any linked list, so long
1147  * as it holds the spin/mutex lock and there is no exclusive lock on
1148  * nfsv4rootfs_lock.
1149  * (For OpenBSD, a kthread is ok. For FreeBSD, I think it is ok
1150  *  to do this from a callout, since the spin locks work. For
1151  *  Darwin, I'm not sure what will work correctly yet.)
1152  * Should be called once per second.
1153  */
1154 APPLESTATIC void
1155 nfsrv_servertimer(void)
1156 {
1157 	struct nfsclient *clp, *nclp;
1158 	struct nfsstate *stp, *nstp;
1159 	int got_ref, i;
1160 
1161 	/*
1162 	 * Make sure nfsboottime is set. This is used by V3 as well
1163 	 * as V4. Note that nfsboottime is not nfsrvboottime, which is
1164 	 * only used by the V4 server for leases.
1165 	 */
1166 	if (nfsboottime.tv_sec == 0)
1167 		NFSSETBOOTTIME(nfsboottime);
1168 
1169 	/*
1170 	 * If server hasn't started yet, just return.
1171 	 */
1172 	NFSLOCKSTATE();
1173 	if (nfsrv_stablefirst.nsf_eograce == 0) {
1174 		NFSUNLOCKSTATE();
1175 		return;
1176 	}
1177 	if (!(nfsrv_stablefirst.nsf_flags & NFSNSF_UPDATEDONE)) {
1178 		if (!(nfsrv_stablefirst.nsf_flags & NFSNSF_GRACEOVER) &&
1179 		    NFSD_MONOSEC > nfsrv_stablefirst.nsf_eograce)
1180 			nfsrv_stablefirst.nsf_flags |=
1181 			    (NFSNSF_GRACEOVER | NFSNSF_NEEDLOCK);
1182 		NFSUNLOCKSTATE();
1183 		return;
1184 	}
1185 
1186 	/*
1187 	 * Try and get a reference count on the nfsv4rootfs_lock so that
1188 	 * no nfsd thread can acquire an exclusive lock on it before this
1189 	 * call is done. If it is already exclusively locked, just return.
1190 	 */
1191 	NFSLOCKV4ROOTMUTEX();
1192 	got_ref = nfsv4_getref_nonblock(&nfsv4rootfs_lock);
1193 	NFSUNLOCKV4ROOTMUTEX();
1194 	if (got_ref == 0) {
1195 		NFSUNLOCKSTATE();
1196 		return;
1197 	}
1198 
1199 	/*
1200 	 * For each client...
1201 	 */
1202 	for (i = 0; i < nfsrv_clienthashsize; i++) {
1203 	    clp = LIST_FIRST(&nfsclienthash[i]);
1204 	    while (clp != LIST_END(&nfsclienthash[i])) {
1205 		nclp = LIST_NEXT(clp, lc_hash);
1206 		if (!(clp->lc_flags & LCL_EXPIREIT)) {
1207 		    if (((clp->lc_expiry + NFSRV_STALELEASE) < NFSD_MONOSEC
1208 			 && ((LIST_EMPTY(&clp->lc_deleg)
1209 			      && LIST_EMPTY(&clp->lc_open)) ||
1210 			     nfsrv_clients > nfsrv_clienthighwater)) ||
1211 			(clp->lc_expiry + NFSRV_MOULDYLEASE) < NFSD_MONOSEC ||
1212 			(clp->lc_expiry < NFSD_MONOSEC &&
1213 			 (nfsrv_openpluslock * 10 / 9) > nfsrv_v4statelimit)) {
1214 			/*
1215 			 * Lease has expired several nfsrv_lease times ago:
1216 			 * PLUS
1217 			 *    - no state is associated with it
1218 			 *    OR
1219 			 *    - above high water mark for number of clients
1220 			 *      (nfsrv_clienthighwater should be large enough
1221 			 *       that this only occurs when clients fail to
1222 			 *       use the same nfs_client_id4.id. Maybe somewhat
1223 			 *       higher that the maximum number of clients that
1224 			 *       will mount this server?)
1225 			 * OR
1226 			 * Lease has expired a very long time ago
1227 			 * OR
1228 			 * Lease has expired PLUS the number of opens + locks
1229 			 * has exceeded 90% of capacity
1230 			 *
1231 			 * --> Mark for expiry. The actual expiry will be done
1232 			 *     by an nfsd sometime soon.
1233 			 */
1234 			clp->lc_flags |= LCL_EXPIREIT;
1235 			nfsrv_stablefirst.nsf_flags |=
1236 			    (NFSNSF_NEEDLOCK | NFSNSF_EXPIREDCLIENT);
1237 		    } else {
1238 			/*
1239 			 * If there are no opens, increment no open tick cnt
1240 			 * If time exceeds NFSNOOPEN, mark it to be thrown away
1241 			 * otherwise, if there is an open, reset no open time
1242 			 * Hopefully, this will avoid excessive re-creation
1243 			 * of open owners and subsequent open confirms.
1244 			 */
1245 			stp = LIST_FIRST(&clp->lc_open);
1246 			while (stp != LIST_END(&clp->lc_open)) {
1247 				nstp = LIST_NEXT(stp, ls_list);
1248 				if (LIST_EMPTY(&stp->ls_open)) {
1249 					stp->ls_noopens++;
1250 					if (stp->ls_noopens > NFSNOOPEN ||
1251 					    (nfsrv_openpluslock * 2) >
1252 					    nfsrv_v4statelimit)
1253 						nfsrv_stablefirst.nsf_flags |=
1254 							NFSNSF_NOOPENS;
1255 				} else {
1256 					stp->ls_noopens = 0;
1257 				}
1258 				stp = nstp;
1259 			}
1260 		    }
1261 		}
1262 		clp = nclp;
1263 	    }
1264 	}
1265 	NFSUNLOCKSTATE();
1266 	NFSLOCKV4ROOTMUTEX();
1267 	nfsv4_relref(&nfsv4rootfs_lock);
1268 	NFSUNLOCKV4ROOTMUTEX();
1269 }
1270 
1271 /*
1272  * The following set of functions free up the various data structures.
1273  */
1274 /*
1275  * Clear out all open/lock state related to this nfsclient.
1276  * Caller must hold an exclusive lock on nfsv4rootfs_lock, so that
1277  * there are no other active nfsd threads.
1278  */
1279 APPLESTATIC void
1280 nfsrv_cleanclient(struct nfsclient *clp, NFSPROC_T *p)
1281 {
1282 	struct nfsstate *stp, *nstp;
1283 	struct nfsdsession *sep, *nsep;
1284 
1285 	LIST_FOREACH_SAFE(stp, &clp->lc_open, ls_list, nstp)
1286 		nfsrv_freeopenowner(stp, 1, p);
1287 	if ((clp->lc_flags & LCL_ADMINREVOKED) == 0)
1288 		LIST_FOREACH_SAFE(sep, &clp->lc_session, sess_list, nsep)
1289 			(void)nfsrv_freesession(sep, NULL);
1290 }
1291 
1292 /*
1293  * Free a client that has been cleaned. It should also already have been
1294  * removed from the lists.
1295  * (Just to be safe w.r.t. newnfs_disconnect(), call this function when
1296  *  softclock interrupts are enabled.)
1297  */
1298 APPLESTATIC void
1299 nfsrv_zapclient(struct nfsclient *clp, NFSPROC_T *p)
1300 {
1301 
1302 #ifdef notyet
1303 	if ((clp->lc_flags & (LCL_GSS | LCL_CALLBACKSON)) ==
1304 	     (LCL_GSS | LCL_CALLBACKSON) &&
1305 	    (clp->lc_hand.nfsh_flag & NFSG_COMPLETE) &&
1306 	    clp->lc_handlelen > 0) {
1307 		clp->lc_hand.nfsh_flag &= ~NFSG_COMPLETE;
1308 		clp->lc_hand.nfsh_flag |= NFSG_DESTROYED;
1309 		(void) nfsrv_docallback(clp, NFSV4PROC_CBNULL,
1310 			NULL, 0, NULL, NULL, NULL, 0, p);
1311 	}
1312 #endif
1313 	newnfs_disconnect(&clp->lc_req);
1314 	free(clp->lc_req.nr_nam, M_SONAME);
1315 	NFSFREEMUTEX(&clp->lc_req.nr_mtx);
1316 	free(clp->lc_stateid, M_NFSDCLIENT);
1317 	free(clp, M_NFSDCLIENT);
1318 	NFSLOCKSTATE();
1319 	nfsstatsv1.srvclients--;
1320 	nfsrv_openpluslock--;
1321 	nfsrv_clients--;
1322 	NFSUNLOCKSTATE();
1323 }
1324 
1325 /*
1326  * Free a list of delegation state structures.
1327  * (This function will also free all nfslockfile structures that no
1328  *  longer have associated state.)
1329  */
1330 APPLESTATIC void
1331 nfsrv_freedeleglist(struct nfsstatehead *sthp)
1332 {
1333 	struct nfsstate *stp, *nstp;
1334 
1335 	LIST_FOREACH_SAFE(stp, sthp, ls_list, nstp) {
1336 		nfsrv_freedeleg(stp);
1337 	}
1338 	LIST_INIT(sthp);
1339 }
1340 
1341 /*
1342  * Free up a delegation.
1343  */
1344 static void
1345 nfsrv_freedeleg(struct nfsstate *stp)
1346 {
1347 	struct nfslockfile *lfp;
1348 
1349 	LIST_REMOVE(stp, ls_hash);
1350 	LIST_REMOVE(stp, ls_list);
1351 	LIST_REMOVE(stp, ls_file);
1352 	if ((stp->ls_flags & NFSLCK_DELEGWRITE) != 0)
1353 		nfsrv_writedelegcnt--;
1354 	lfp = stp->ls_lfp;
1355 	if (LIST_EMPTY(&lfp->lf_open) &&
1356 	    LIST_EMPTY(&lfp->lf_lock) && LIST_EMPTY(&lfp->lf_deleg) &&
1357 	    LIST_EMPTY(&lfp->lf_locallock) && LIST_EMPTY(&lfp->lf_rollback) &&
1358 	    lfp->lf_usecount == 0 &&
1359 	    nfsv4_testlock(&lfp->lf_locallock_lck) == 0)
1360 		nfsrv_freenfslockfile(lfp);
1361 	free(stp, M_NFSDSTATE);
1362 	nfsstatsv1.srvdelegates--;
1363 	nfsrv_openpluslock--;
1364 	nfsrv_delegatecnt--;
1365 }
1366 
1367 /*
1368  * This function frees an open owner and all associated opens.
1369  */
1370 static void
1371 nfsrv_freeopenowner(struct nfsstate *stp, int cansleep, NFSPROC_T *p)
1372 {
1373 	struct nfsstate *nstp, *tstp;
1374 
1375 	LIST_REMOVE(stp, ls_list);
1376 	/*
1377 	 * Now, free all associated opens.
1378 	 */
1379 	nstp = LIST_FIRST(&stp->ls_open);
1380 	while (nstp != LIST_END(&stp->ls_open)) {
1381 		tstp = nstp;
1382 		nstp = LIST_NEXT(nstp, ls_list);
1383 		(void) nfsrv_freeopen(tstp, NULL, cansleep, p);
1384 	}
1385 	if (stp->ls_op)
1386 		nfsrvd_derefcache(stp->ls_op);
1387 	free(stp, M_NFSDSTATE);
1388 	nfsstatsv1.srvopenowners--;
1389 	nfsrv_openpluslock--;
1390 }
1391 
1392 /*
1393  * This function frees an open (nfsstate open structure) with all associated
1394  * lock_owners and locks. It also frees the nfslockfile structure iff there
1395  * are no other opens on the file.
1396  * Returns 1 if it free'd the nfslockfile, 0 otherwise.
1397  */
1398 static int
1399 nfsrv_freeopen(struct nfsstate *stp, vnode_t vp, int cansleep, NFSPROC_T *p)
1400 {
1401 	struct nfsstate *nstp, *tstp;
1402 	struct nfslockfile *lfp;
1403 	int ret;
1404 
1405 	LIST_REMOVE(stp, ls_hash);
1406 	LIST_REMOVE(stp, ls_list);
1407 	LIST_REMOVE(stp, ls_file);
1408 
1409 	lfp = stp->ls_lfp;
1410 	/*
1411 	 * Now, free all lockowners associated with this open.
1412 	 */
1413 	LIST_FOREACH_SAFE(tstp, &stp->ls_open, ls_list, nstp)
1414 		nfsrv_freelockowner(tstp, vp, cansleep, p);
1415 
1416 	/*
1417 	 * The nfslockfile is freed here if there are no locks
1418 	 * associated with the open.
1419 	 * If there are locks associated with the open, the
1420 	 * nfslockfile structure can be freed via nfsrv_freelockowner().
1421 	 * Acquire the state mutex to avoid races with calls to
1422 	 * nfsrv_getlockfile().
1423 	 */
1424 	if (cansleep != 0)
1425 		NFSLOCKSTATE();
1426 	if (lfp != NULL && LIST_EMPTY(&lfp->lf_open) &&
1427 	    LIST_EMPTY(&lfp->lf_deleg) && LIST_EMPTY(&lfp->lf_lock) &&
1428 	    LIST_EMPTY(&lfp->lf_locallock) && LIST_EMPTY(&lfp->lf_rollback) &&
1429 	    lfp->lf_usecount == 0 &&
1430 	    (cansleep != 0 || nfsv4_testlock(&lfp->lf_locallock_lck) == 0)) {
1431 		nfsrv_freenfslockfile(lfp);
1432 		ret = 1;
1433 	} else
1434 		ret = 0;
1435 	if (cansleep != 0)
1436 		NFSUNLOCKSTATE();
1437 	free(stp, M_NFSDSTATE);
1438 	nfsstatsv1.srvopens--;
1439 	nfsrv_openpluslock--;
1440 	return (ret);
1441 }
1442 
1443 /*
1444  * Frees a lockowner and all associated locks.
1445  */
1446 static void
1447 nfsrv_freelockowner(struct nfsstate *stp, vnode_t vp, int cansleep,
1448     NFSPROC_T *p)
1449 {
1450 
1451 	LIST_REMOVE(stp, ls_hash);
1452 	LIST_REMOVE(stp, ls_list);
1453 	nfsrv_freeallnfslocks(stp, vp, cansleep, p);
1454 	if (stp->ls_op)
1455 		nfsrvd_derefcache(stp->ls_op);
1456 	free(stp, M_NFSDSTATE);
1457 	nfsstatsv1.srvlockowners--;
1458 	nfsrv_openpluslock--;
1459 }
1460 
1461 /*
1462  * Free all the nfs locks on a lockowner.
1463  */
1464 static void
1465 nfsrv_freeallnfslocks(struct nfsstate *stp, vnode_t vp, int cansleep,
1466     NFSPROC_T *p)
1467 {
1468 	struct nfslock *lop, *nlop;
1469 	struct nfsrollback *rlp, *nrlp;
1470 	struct nfslockfile *lfp = NULL;
1471 	int gottvp = 0;
1472 	vnode_t tvp = NULL;
1473 	uint64_t first, end;
1474 
1475 	if (vp != NULL)
1476 		ASSERT_VOP_UNLOCKED(vp, "nfsrv_freeallnfslocks: vnode locked");
1477 	lop = LIST_FIRST(&stp->ls_lock);
1478 	while (lop != LIST_END(&stp->ls_lock)) {
1479 		nlop = LIST_NEXT(lop, lo_lckowner);
1480 		/*
1481 		 * Since all locks should be for the same file, lfp should
1482 		 * not change.
1483 		 */
1484 		if (lfp == NULL)
1485 			lfp = lop->lo_lfp;
1486 		else if (lfp != lop->lo_lfp)
1487 			panic("allnfslocks");
1488 		/*
1489 		 * If vp is NULL and cansleep != 0, a vnode must be acquired
1490 		 * from the file handle. This only occurs when called from
1491 		 * nfsrv_cleanclient().
1492 		 */
1493 		if (gottvp == 0) {
1494 			if (nfsrv_dolocallocks == 0)
1495 				tvp = NULL;
1496 			else if (vp == NULL && cansleep != 0) {
1497 				tvp = nfsvno_getvp(&lfp->lf_fh);
1498 				NFSVOPUNLOCK(tvp, 0);
1499 			} else
1500 				tvp = vp;
1501 			gottvp = 1;
1502 		}
1503 
1504 		if (tvp != NULL) {
1505 			if (cansleep == 0)
1506 				panic("allnfs2");
1507 			first = lop->lo_first;
1508 			end = lop->lo_end;
1509 			nfsrv_freenfslock(lop);
1510 			nfsrv_localunlock(tvp, lfp, first, end, p);
1511 			LIST_FOREACH_SAFE(rlp, &lfp->lf_rollback, rlck_list,
1512 			    nrlp)
1513 				free(rlp, M_NFSDROLLBACK);
1514 			LIST_INIT(&lfp->lf_rollback);
1515 		} else
1516 			nfsrv_freenfslock(lop);
1517 		lop = nlop;
1518 	}
1519 	if (vp == NULL && tvp != NULL)
1520 		vrele(tvp);
1521 }
1522 
1523 /*
1524  * Free an nfslock structure.
1525  */
1526 static void
1527 nfsrv_freenfslock(struct nfslock *lop)
1528 {
1529 
1530 	if (lop->lo_lckfile.le_prev != NULL) {
1531 		LIST_REMOVE(lop, lo_lckfile);
1532 		nfsstatsv1.srvlocks--;
1533 		nfsrv_openpluslock--;
1534 	}
1535 	LIST_REMOVE(lop, lo_lckowner);
1536 	free(lop, M_NFSDLOCK);
1537 }
1538 
1539 /*
1540  * This function frees an nfslockfile structure.
1541  */
1542 static void
1543 nfsrv_freenfslockfile(struct nfslockfile *lfp)
1544 {
1545 
1546 	LIST_REMOVE(lfp, lf_hash);
1547 	free(lfp, M_NFSDLOCKFILE);
1548 }
1549 
1550 /*
1551  * This function looks up an nfsstate structure via stateid.
1552  */
1553 static int
1554 nfsrv_getstate(struct nfsclient *clp, nfsv4stateid_t *stateidp, __unused u_int32_t flags,
1555     struct nfsstate **stpp)
1556 {
1557 	struct nfsstate *stp;
1558 	struct nfsstatehead *hp;
1559 	int error = 0;
1560 
1561 	*stpp = NULL;
1562 	hp = NFSSTATEHASH(clp, *stateidp);
1563 	LIST_FOREACH(stp, hp, ls_hash) {
1564 		if (!NFSBCMP(stp->ls_stateid.other, stateidp->other,
1565 			NFSX_STATEIDOTHER))
1566 			break;
1567 	}
1568 
1569 	/*
1570 	 * If no state id in list, return NFSERR_BADSTATEID.
1571 	 */
1572 	if (stp == LIST_END(hp)) {
1573 		error = NFSERR_BADSTATEID;
1574 		goto out;
1575 	}
1576 	*stpp = stp;
1577 
1578 out:
1579 	NFSEXITCODE(error);
1580 	return (error);
1581 }
1582 
1583 /*
1584  * This function gets an nfsstate structure via owner string.
1585  */
1586 static void
1587 nfsrv_getowner(struct nfsstatehead *hp, struct nfsstate *new_stp,
1588     struct nfsstate **stpp)
1589 {
1590 	struct nfsstate *stp;
1591 
1592 	*stpp = NULL;
1593 	LIST_FOREACH(stp, hp, ls_list) {
1594 		if (new_stp->ls_ownerlen == stp->ls_ownerlen &&
1595 		  !NFSBCMP(new_stp->ls_owner,stp->ls_owner,stp->ls_ownerlen)) {
1596 			*stpp = stp;
1597 			return;
1598 		}
1599 	}
1600 }
1601 
1602 /*
1603  * Lock control function called to update lock status.
1604  * Returns 0 upon success, -1 if there is no lock and the flags indicate
1605  * that one isn't to be created and an NFSERR_xxx for other errors.
1606  * The structures new_stp and new_lop are passed in as pointers that should
1607  * be set to NULL if the structure is used and shouldn't be free'd.
1608  * For the NFSLCK_TEST and NFSLCK_CHECK cases, the structures are
1609  * never used and can safely be allocated on the stack. For all other
1610  * cases, *new_stpp and *new_lopp should be malloc'd before the call,
1611  * in case they are used.
1612  */
1613 APPLESTATIC int
1614 nfsrv_lockctrl(vnode_t vp, struct nfsstate **new_stpp,
1615     struct nfslock **new_lopp, struct nfslockconflict *cfp,
1616     nfsquad_t clientid, nfsv4stateid_t *stateidp,
1617     __unused struct nfsexstuff *exp,
1618     struct nfsrv_descript *nd, NFSPROC_T *p)
1619 {
1620 	struct nfslock *lop;
1621 	struct nfsstate *new_stp = *new_stpp;
1622 	struct nfslock *new_lop = *new_lopp;
1623 	struct nfsstate *tstp, *mystp, *nstp;
1624 	int specialid = 0;
1625 	struct nfslockfile *lfp;
1626 	struct nfslock *other_lop = NULL;
1627 	struct nfsstate *stp, *lckstp = NULL;
1628 	struct nfsclient *clp = NULL;
1629 	u_int32_t bits;
1630 	int error = 0, haslock = 0, ret, reterr;
1631 	int getlckret, delegation = 0, filestruct_locked, vnode_unlocked = 0;
1632 	fhandle_t nfh;
1633 	uint64_t first, end;
1634 	uint32_t lock_flags;
1635 
1636 	if (new_stp->ls_flags & (NFSLCK_CHECK | NFSLCK_SETATTR)) {
1637 		/*
1638 		 * Note the special cases of "all 1s" or "all 0s" stateids and
1639 		 * let reads with all 1s go ahead.
1640 		 */
1641 		if (new_stp->ls_stateid.seqid == 0x0 &&
1642 		    new_stp->ls_stateid.other[0] == 0x0 &&
1643 		    new_stp->ls_stateid.other[1] == 0x0 &&
1644 		    new_stp->ls_stateid.other[2] == 0x0)
1645 			specialid = 1;
1646 		else if (new_stp->ls_stateid.seqid == 0xffffffff &&
1647 		    new_stp->ls_stateid.other[0] == 0xffffffff &&
1648 		    new_stp->ls_stateid.other[1] == 0xffffffff &&
1649 		    new_stp->ls_stateid.other[2] == 0xffffffff)
1650 			specialid = 2;
1651 	}
1652 
1653 	/*
1654 	 * Check for restart conditions (client and server).
1655 	 */
1656 	error = nfsrv_checkrestart(clientid, new_stp->ls_flags,
1657 	    &new_stp->ls_stateid, specialid);
1658 	if (error)
1659 		goto out;
1660 
1661 	/*
1662 	 * Check for state resource limit exceeded.
1663 	 */
1664 	if ((new_stp->ls_flags & NFSLCK_LOCK) &&
1665 	    nfsrv_openpluslock > nfsrv_v4statelimit) {
1666 		error = NFSERR_RESOURCE;
1667 		goto out;
1668 	}
1669 
1670 	/*
1671 	 * For the lock case, get another nfslock structure,
1672 	 * just in case we need it.
1673 	 * Malloc now, before we start sifting through the linked lists,
1674 	 * in case we have to wait for memory.
1675 	 */
1676 tryagain:
1677 	if (new_stp->ls_flags & NFSLCK_LOCK)
1678 		other_lop = malloc(sizeof (struct nfslock),
1679 		    M_NFSDLOCK, M_WAITOK);
1680 	filestruct_locked = 0;
1681 	reterr = 0;
1682 	lfp = NULL;
1683 
1684 	/*
1685 	 * Get the lockfile structure for CFH now, so we can do a sanity
1686 	 * check against the stateid, before incrementing the seqid#, since
1687 	 * we want to return NFSERR_BADSTATEID on failure and the seqid#
1688 	 * shouldn't be incremented for this case.
1689 	 * If nfsrv_getlockfile() returns -1, it means "not found", which
1690 	 * will be handled later.
1691 	 * If we are doing Lock/LockU and local locking is enabled, sleep
1692 	 * lock the nfslockfile structure.
1693 	 */
1694 	getlckret = nfsrv_getlockfh(vp, new_stp->ls_flags, NULL, &nfh, p);
1695 	NFSLOCKSTATE();
1696 	if (getlckret == 0) {
1697 		if ((new_stp->ls_flags & (NFSLCK_LOCK | NFSLCK_UNLOCK)) != 0 &&
1698 		    nfsrv_dolocallocks != 0 && nd->nd_repstat == 0) {
1699 			getlckret = nfsrv_getlockfile(new_stp->ls_flags, NULL,
1700 			    &lfp, &nfh, 1);
1701 			if (getlckret == 0)
1702 				filestruct_locked = 1;
1703 		} else
1704 			getlckret = nfsrv_getlockfile(new_stp->ls_flags, NULL,
1705 			    &lfp, &nfh, 0);
1706 	}
1707 	if (getlckret != 0 && getlckret != -1)
1708 		reterr = getlckret;
1709 
1710 	if (filestruct_locked != 0) {
1711 		LIST_INIT(&lfp->lf_rollback);
1712 		if ((new_stp->ls_flags & NFSLCK_LOCK)) {
1713 			/*
1714 			 * For local locking, do the advisory locking now, so
1715 			 * that any conflict can be detected. A failure later
1716 			 * can be rolled back locally. If an error is returned,
1717 			 * struct nfslockfile has been unlocked and any local
1718 			 * locking rolled back.
1719 			 */
1720 			NFSUNLOCKSTATE();
1721 			if (vnode_unlocked == 0) {
1722 				ASSERT_VOP_ELOCKED(vp, "nfsrv_lockctrl1");
1723 				vnode_unlocked = 1;
1724 				NFSVOPUNLOCK(vp, 0);
1725 			}
1726 			reterr = nfsrv_locallock(vp, lfp,
1727 			    (new_lop->lo_flags & (NFSLCK_READ | NFSLCK_WRITE)),
1728 			    new_lop->lo_first, new_lop->lo_end, cfp, p);
1729 			NFSLOCKSTATE();
1730 		}
1731 	}
1732 
1733 	if (specialid == 0) {
1734 	    if (new_stp->ls_flags & NFSLCK_TEST) {
1735 		/*
1736 		 * RFC 3530 does not list LockT as an op that renews a
1737 		 * lease, but the consensus seems to be that it is ok
1738 		 * for a server to do so.
1739 		 */
1740 		error = nfsrv_getclient(clientid, CLOPS_RENEW, &clp, NULL,
1741 		    (nfsquad_t)((u_quad_t)0), 0, nd, p);
1742 
1743 		/*
1744 		 * Since NFSERR_EXPIRED, NFSERR_ADMINREVOKED are not valid
1745 		 * error returns for LockT, just go ahead and test for a lock,
1746 		 * since there are no locks for this client, but other locks
1747 		 * can conflict. (ie. same client will always be false)
1748 		 */
1749 		if (error == NFSERR_EXPIRED || error == NFSERR_ADMINREVOKED)
1750 		    error = 0;
1751 		lckstp = new_stp;
1752 	    } else {
1753 	      error = nfsrv_getclient(clientid, CLOPS_RENEW, &clp, NULL,
1754 		(nfsquad_t)((u_quad_t)0), 0, nd, p);
1755 	      if (error == 0)
1756 		/*
1757 		 * Look up the stateid
1758 		 */
1759 		error = nfsrv_getstate(clp, &new_stp->ls_stateid,
1760 		  new_stp->ls_flags, &stp);
1761 	      /*
1762 	       * do some sanity checks for an unconfirmed open or a
1763 	       * stateid that refers to the wrong file, for an open stateid
1764 	       */
1765 	      if (error == 0 && (stp->ls_flags & NFSLCK_OPEN) &&
1766 		  ((stp->ls_openowner->ls_flags & NFSLCK_NEEDSCONFIRM) ||
1767 		   (getlckret == 0 && stp->ls_lfp != lfp))){
1768 		      /*
1769 		       * NFSLCK_SETATTR should return OK rather than NFSERR_BADSTATEID
1770 		       * The only exception is using SETATTR with SIZE.
1771 		       * */
1772                     if ((new_stp->ls_flags &
1773                          (NFSLCK_SETATTR | NFSLCK_CHECK)) != NFSLCK_SETATTR)
1774 			     error = NFSERR_BADSTATEID;
1775 	      }
1776 
1777 		if (error == 0 &&
1778 		  (stp->ls_flags & (NFSLCK_DELEGREAD | NFSLCK_DELEGWRITE)) &&
1779 		  getlckret == 0 && stp->ls_lfp != lfp)
1780 			error = NFSERR_BADSTATEID;
1781 
1782 	      /*
1783 	       * If the lockowner stateid doesn't refer to the same file,
1784 	       * I believe that is considered ok, since some clients will
1785 	       * only create a single lockowner and use that for all locks
1786 	       * on all files.
1787 	       * For now, log it as a diagnostic, instead of considering it
1788 	       * a BadStateid.
1789 	       */
1790 	      if (error == 0 && (stp->ls_flags &
1791 		  (NFSLCK_OPEN | NFSLCK_DELEGREAD | NFSLCK_DELEGWRITE)) == 0 &&
1792 		  getlckret == 0 && stp->ls_lfp != lfp) {
1793 #ifdef DIAGNOSTIC
1794 		  printf("Got a lock statid for different file open\n");
1795 #endif
1796 		  /*
1797 		  error = NFSERR_BADSTATEID;
1798 		  */
1799 	      }
1800 
1801 	      if (error == 0) {
1802 		    if (new_stp->ls_flags & NFSLCK_OPENTOLOCK) {
1803 			/*
1804 			 * If haslock set, we've already checked the seqid.
1805 			 */
1806 			if (!haslock) {
1807 			    if (stp->ls_flags & NFSLCK_OPEN)
1808 				error = nfsrv_checkseqid(nd, new_stp->ls_seq,
1809 				    stp->ls_openowner, new_stp->ls_op);
1810 			    else
1811 				error = NFSERR_BADSTATEID;
1812 			}
1813 			if (!error)
1814 			    nfsrv_getowner(&stp->ls_open, new_stp, &lckstp);
1815 			if (lckstp)
1816 			    /*
1817 			     * I believe this should be an error, but it
1818 			     * isn't obvious what NFSERR_xxx would be
1819 			     * appropriate, so I'll use NFSERR_INVAL for now.
1820 			     */
1821 			    error = NFSERR_INVAL;
1822 			else
1823 			    lckstp = new_stp;
1824 		    } else if (new_stp->ls_flags&(NFSLCK_LOCK|NFSLCK_UNLOCK)) {
1825 			/*
1826 			 * If haslock set, ditto above.
1827 			 */
1828 			if (!haslock) {
1829 			    if (stp->ls_flags & NFSLCK_OPEN)
1830 				error = NFSERR_BADSTATEID;
1831 			    else
1832 				error = nfsrv_checkseqid(nd, new_stp->ls_seq,
1833 				    stp, new_stp->ls_op);
1834 			}
1835 			lckstp = stp;
1836 		    } else {
1837 			lckstp = stp;
1838 		    }
1839 	      }
1840 	      /*
1841 	       * If the seqid part of the stateid isn't the same, return
1842 	       * NFSERR_OLDSTATEID for cases other than I/O Ops.
1843 	       * For I/O Ops, only return NFSERR_OLDSTATEID if
1844 	       * nfsrv_returnoldstateid is set. (The consensus on the email
1845 	       * list was that most clients would prefer to not receive
1846 	       * NFSERR_OLDSTATEID for I/O Ops, but the RFC suggests that that
1847 	       * is what will happen, so I use the nfsrv_returnoldstateid to
1848 	       * allow for either server configuration.)
1849 	       */
1850 	      if (!error && stp->ls_stateid.seqid!=new_stp->ls_stateid.seqid &&
1851 		  (((nd->nd_flag & ND_NFSV41) == 0 &&
1852 		   (!(new_stp->ls_flags & NFSLCK_CHECK) ||
1853 		    nfsrv_returnoldstateid)) ||
1854 		   ((nd->nd_flag & ND_NFSV41) != 0 &&
1855 		    new_stp->ls_stateid.seqid != 0)))
1856 		    error = NFSERR_OLDSTATEID;
1857 	    }
1858 	}
1859 
1860 	/*
1861 	 * Now we can check for grace.
1862 	 */
1863 	if (!error)
1864 		error = nfsrv_checkgrace(nd, clp, new_stp->ls_flags);
1865 	if ((new_stp->ls_flags & NFSLCK_RECLAIM) && !error &&
1866 		nfsrv_checkstable(clp))
1867 		error = NFSERR_NOGRACE;
1868 	/*
1869 	 * If we successfully Reclaimed state, note that.
1870 	 */
1871 	if ((new_stp->ls_flags & NFSLCK_RECLAIM) && !error)
1872 		nfsrv_markstable(clp);
1873 
1874 	/*
1875 	 * At this point, either error == NFSERR_BADSTATEID or the
1876 	 * seqid# has been updated, so we can return any error.
1877 	 * If error == 0, there may be an error in:
1878 	 *    nd_repstat - Set by the calling function.
1879 	 *    reterr - Set above, if getting the nfslockfile structure
1880 	 *       or acquiring the local lock failed.
1881 	 *    (If both of these are set, nd_repstat should probably be
1882 	 *     returned, since that error was detected before this
1883 	 *     function call.)
1884 	 */
1885 	if (error != 0 || nd->nd_repstat != 0 || reterr != 0) {
1886 		if (error == 0) {
1887 			if (nd->nd_repstat != 0)
1888 				error = nd->nd_repstat;
1889 			else
1890 				error = reterr;
1891 		}
1892 		if (filestruct_locked != 0) {
1893 			/* Roll back local locks. */
1894 			NFSUNLOCKSTATE();
1895 			if (vnode_unlocked == 0) {
1896 				ASSERT_VOP_ELOCKED(vp, "nfsrv_lockctrl2");
1897 				vnode_unlocked = 1;
1898 				NFSVOPUNLOCK(vp, 0);
1899 			}
1900 			nfsrv_locallock_rollback(vp, lfp, p);
1901 			NFSLOCKSTATE();
1902 			nfsrv_unlocklf(lfp);
1903 		}
1904 		NFSUNLOCKSTATE();
1905 		goto out;
1906 	}
1907 
1908 	/*
1909 	 * Check the nfsrv_getlockfile return.
1910 	 * Returned -1 if no structure found.
1911 	 */
1912 	if (getlckret == -1) {
1913 		error = NFSERR_EXPIRED;
1914 		/*
1915 		 * Called from lockt, so no lock is OK.
1916 		 */
1917 		if (new_stp->ls_flags & NFSLCK_TEST) {
1918 			error = 0;
1919 		} else if (new_stp->ls_flags &
1920 		    (NFSLCK_CHECK | NFSLCK_SETATTR)) {
1921 			/*
1922 			 * Called to check for a lock, OK if the stateid is all
1923 			 * 1s or all 0s, but there should be an nfsstate
1924 			 * otherwise.
1925 			 * (ie. If there is no open, I'll assume no share
1926 			 *  deny bits.)
1927 			 */
1928 			if (specialid)
1929 				error = 0;
1930 			else
1931 				error = NFSERR_BADSTATEID;
1932 		}
1933 		NFSUNLOCKSTATE();
1934 		goto out;
1935 	}
1936 
1937 	/*
1938 	 * For NFSLCK_CHECK and NFSLCK_LOCK, test for a share conflict.
1939 	 * For NFSLCK_CHECK, allow a read if write access is granted,
1940 	 * but check for a deny. For NFSLCK_LOCK, require correct access,
1941 	 * which implies a conflicting deny can't exist.
1942 	 */
1943 	if (new_stp->ls_flags & (NFSLCK_CHECK | NFSLCK_LOCK)) {
1944 	    /*
1945 	     * Four kinds of state id:
1946 	     * - specialid (all 0s or all 1s), only for NFSLCK_CHECK
1947 	     * - stateid for an open
1948 	     * - stateid for a delegation
1949 	     * - stateid for a lock owner
1950 	     */
1951 	    if (!specialid) {
1952 		if (stp->ls_flags & (NFSLCK_DELEGREAD | NFSLCK_DELEGWRITE)) {
1953 		    delegation = 1;
1954 		    mystp = stp;
1955 		    nfsrv_delaydelegtimeout(stp);
1956 	        } else if (stp->ls_flags & NFSLCK_OPEN) {
1957 		    mystp = stp;
1958 		} else {
1959 		    mystp = stp->ls_openstp;
1960 		}
1961 		/*
1962 		 * If locking or checking, require correct access
1963 		 * bit set.
1964 		 */
1965 		if (((new_stp->ls_flags & NFSLCK_LOCK) &&
1966 		     !((new_lop->lo_flags >> NFSLCK_LOCKSHIFT) &
1967 		       mystp->ls_flags & NFSLCK_ACCESSBITS)) ||
1968 		    ((new_stp->ls_flags & (NFSLCK_CHECK|NFSLCK_READACCESS)) ==
1969 		      (NFSLCK_CHECK | NFSLCK_READACCESS) &&
1970 		     !(mystp->ls_flags & NFSLCK_READACCESS) &&
1971 		     nfsrv_allowreadforwriteopen == 0) ||
1972 		    ((new_stp->ls_flags & (NFSLCK_CHECK|NFSLCK_WRITEACCESS)) ==
1973 		      (NFSLCK_CHECK | NFSLCK_WRITEACCESS) &&
1974 		     !(mystp->ls_flags & NFSLCK_WRITEACCESS))) {
1975 			if (filestruct_locked != 0) {
1976 				/* Roll back local locks. */
1977 				NFSUNLOCKSTATE();
1978 				if (vnode_unlocked == 0) {
1979 					ASSERT_VOP_ELOCKED(vp,
1980 					    "nfsrv_lockctrl3");
1981 					vnode_unlocked = 1;
1982 					NFSVOPUNLOCK(vp, 0);
1983 				}
1984 				nfsrv_locallock_rollback(vp, lfp, p);
1985 				NFSLOCKSTATE();
1986 				nfsrv_unlocklf(lfp);
1987 			}
1988 			NFSUNLOCKSTATE();
1989 			error = NFSERR_OPENMODE;
1990 			goto out;
1991 		}
1992 	    } else
1993 		mystp = NULL;
1994 	    if ((new_stp->ls_flags & NFSLCK_CHECK) && !delegation) {
1995 		/*
1996 		 * Check for a conflicting deny bit.
1997 		 */
1998 		LIST_FOREACH(tstp, &lfp->lf_open, ls_file) {
1999 		    if (tstp != mystp) {
2000 			bits = tstp->ls_flags;
2001 			bits >>= NFSLCK_SHIFT;
2002 			if (new_stp->ls_flags & bits & NFSLCK_ACCESSBITS) {
2003 			    KASSERT(vnode_unlocked == 0,
2004 				("nfsrv_lockctrl: vnode unlocked1"));
2005 			    ret = nfsrv_clientconflict(tstp->ls_clp, &haslock,
2006 				vp, p);
2007 			    if (ret == 1) {
2008 				/*
2009 				* nfsrv_clientconflict unlocks state
2010 				 * when it returns non-zero.
2011 				 */
2012 				lckstp = NULL;
2013 				goto tryagain;
2014 			    }
2015 			    if (ret == 0)
2016 				NFSUNLOCKSTATE();
2017 			    if (ret == 2)
2018 				error = NFSERR_PERM;
2019 			    else
2020 				error = NFSERR_OPENMODE;
2021 			    goto out;
2022 			}
2023 		    }
2024 		}
2025 
2026 		/* We're outta here */
2027 		NFSUNLOCKSTATE();
2028 		goto out;
2029 	    }
2030 	}
2031 
2032 	/*
2033 	 * For setattr, just get rid of all the Delegations for other clients.
2034 	 */
2035 	if (new_stp->ls_flags & NFSLCK_SETATTR) {
2036 		KASSERT(vnode_unlocked == 0,
2037 		    ("nfsrv_lockctrl: vnode unlocked2"));
2038 		ret = nfsrv_cleandeleg(vp, lfp, clp, &haslock, p);
2039 		if (ret) {
2040 			/*
2041 			 * nfsrv_cleandeleg() unlocks state when it
2042 			 * returns non-zero.
2043 			 */
2044 			if (ret == -1) {
2045 				lckstp = NULL;
2046 				goto tryagain;
2047 			}
2048 			error = ret;
2049 			goto out;
2050 		}
2051 		if (!(new_stp->ls_flags & NFSLCK_CHECK) ||
2052 		    (LIST_EMPTY(&lfp->lf_open) && LIST_EMPTY(&lfp->lf_lock) &&
2053 		     LIST_EMPTY(&lfp->lf_deleg))) {
2054 			NFSUNLOCKSTATE();
2055 			goto out;
2056 		}
2057 	}
2058 
2059 	/*
2060 	 * Check for a conflicting delegation. If one is found, call
2061 	 * nfsrv_delegconflict() to handle it. If the v4root lock hasn't
2062 	 * been set yet, it will get the lock. Otherwise, it will recall
2063 	 * the delegation. Then, we try try again...
2064 	 * I currently believe the conflict algorithm to be:
2065 	 * For Lock Ops (Lock/LockT/LockU)
2066 	 * - there is a conflict iff a different client has a write delegation
2067 	 * For Reading (Read Op)
2068 	 * - there is a conflict iff a different client has a write delegation
2069 	 *   (the specialids are always a different client)
2070 	 * For Writing (Write/Setattr of size)
2071 	 * - there is a conflict if a different client has any delegation
2072 	 * - there is a conflict if the same client has a read delegation
2073 	 *   (I don't understand why this isn't allowed, but that seems to be
2074 	 *    the current consensus?)
2075 	 */
2076 	tstp = LIST_FIRST(&lfp->lf_deleg);
2077 	while (tstp != LIST_END(&lfp->lf_deleg)) {
2078 	    nstp = LIST_NEXT(tstp, ls_file);
2079 	    if ((((new_stp->ls_flags&(NFSLCK_LOCK|NFSLCK_UNLOCK|NFSLCK_TEST))||
2080 		 ((new_stp->ls_flags & NFSLCK_CHECK) &&
2081 		  (new_lop->lo_flags & NFSLCK_READ))) &&
2082 		  clp != tstp->ls_clp &&
2083 		 (tstp->ls_flags & NFSLCK_DELEGWRITE)) ||
2084 		 ((new_stp->ls_flags & NFSLCK_CHECK) &&
2085 		   (new_lop->lo_flags & NFSLCK_WRITE) &&
2086 		  (clp != tstp->ls_clp ||
2087 		   (tstp->ls_flags & NFSLCK_DELEGREAD)))) {
2088 		ret = 0;
2089 		if (filestruct_locked != 0) {
2090 			/* Roll back local locks. */
2091 			NFSUNLOCKSTATE();
2092 			if (vnode_unlocked == 0) {
2093 				ASSERT_VOP_ELOCKED(vp, "nfsrv_lockctrl4");
2094 				NFSVOPUNLOCK(vp, 0);
2095 			}
2096 			nfsrv_locallock_rollback(vp, lfp, p);
2097 			NFSLOCKSTATE();
2098 			nfsrv_unlocklf(lfp);
2099 			NFSUNLOCKSTATE();
2100 			NFSVOPLOCK(vp, LK_EXCLUSIVE | LK_RETRY);
2101 			vnode_unlocked = 0;
2102 			if ((vp->v_iflag & VI_DOOMED) != 0)
2103 				ret = NFSERR_SERVERFAULT;
2104 			NFSLOCKSTATE();
2105 		}
2106 		if (ret == 0)
2107 			ret = nfsrv_delegconflict(tstp, &haslock, p, vp);
2108 		if (ret) {
2109 		    /*
2110 		     * nfsrv_delegconflict unlocks state when it
2111 		     * returns non-zero, which it always does.
2112 		     */
2113 		    if (other_lop) {
2114 			free(other_lop, M_NFSDLOCK);
2115 			other_lop = NULL;
2116 		    }
2117 		    if (ret == -1) {
2118 			lckstp = NULL;
2119 			goto tryagain;
2120 		    }
2121 		    error = ret;
2122 		    goto out;
2123 		}
2124 		/* Never gets here. */
2125 	    }
2126 	    tstp = nstp;
2127 	}
2128 
2129 	/*
2130 	 * Handle the unlock case by calling nfsrv_updatelock().
2131 	 * (Should I have done some access checking above for unlock? For now,
2132 	 *  just let it happen.)
2133 	 */
2134 	if (new_stp->ls_flags & NFSLCK_UNLOCK) {
2135 		first = new_lop->lo_first;
2136 		end = new_lop->lo_end;
2137 		nfsrv_updatelock(stp, new_lopp, &other_lop, lfp);
2138 		stateidp->seqid = ++(stp->ls_stateid.seqid);
2139 		if ((nd->nd_flag & ND_NFSV41) != 0 && stateidp->seqid == 0)
2140 			stateidp->seqid = stp->ls_stateid.seqid = 1;
2141 		stateidp->other[0] = stp->ls_stateid.other[0];
2142 		stateidp->other[1] = stp->ls_stateid.other[1];
2143 		stateidp->other[2] = stp->ls_stateid.other[2];
2144 		if (filestruct_locked != 0) {
2145 			NFSUNLOCKSTATE();
2146 			if (vnode_unlocked == 0) {
2147 				ASSERT_VOP_ELOCKED(vp, "nfsrv_lockctrl5");
2148 				vnode_unlocked = 1;
2149 				NFSVOPUNLOCK(vp, 0);
2150 			}
2151 			/* Update the local locks. */
2152 			nfsrv_localunlock(vp, lfp, first, end, p);
2153 			NFSLOCKSTATE();
2154 			nfsrv_unlocklf(lfp);
2155 		}
2156 		NFSUNLOCKSTATE();
2157 		goto out;
2158 	}
2159 
2160 	/*
2161 	 * Search for a conflicting lock. A lock conflicts if:
2162 	 * - the lock range overlaps and
2163 	 * - at least one lock is a write lock and
2164 	 * - it is not owned by the same lock owner
2165 	 */
2166 	if (!delegation) {
2167 	  LIST_FOREACH(lop, &lfp->lf_lock, lo_lckfile) {
2168 	    if (new_lop->lo_end > lop->lo_first &&
2169 		new_lop->lo_first < lop->lo_end &&
2170 		(new_lop->lo_flags == NFSLCK_WRITE ||
2171 		 lop->lo_flags == NFSLCK_WRITE) &&
2172 		lckstp != lop->lo_stp &&
2173 		(clp != lop->lo_stp->ls_clp ||
2174 		 lckstp->ls_ownerlen != lop->lo_stp->ls_ownerlen ||
2175 		 NFSBCMP(lckstp->ls_owner, lop->lo_stp->ls_owner,
2176 		    lckstp->ls_ownerlen))) {
2177 		if (other_lop) {
2178 		    free(other_lop, M_NFSDLOCK);
2179 		    other_lop = NULL;
2180 		}
2181 		if (vnode_unlocked != 0)
2182 		    ret = nfsrv_clientconflict(lop->lo_stp->ls_clp, &haslock,
2183 			NULL, p);
2184 		else
2185 		    ret = nfsrv_clientconflict(lop->lo_stp->ls_clp, &haslock,
2186 			vp, p);
2187 		if (ret == 1) {
2188 		    if (filestruct_locked != 0) {
2189 			if (vnode_unlocked == 0) {
2190 				ASSERT_VOP_ELOCKED(vp, "nfsrv_lockctrl6");
2191 				NFSVOPUNLOCK(vp, 0);
2192 			}
2193 			/* Roll back local locks. */
2194 			nfsrv_locallock_rollback(vp, lfp, p);
2195 			NFSLOCKSTATE();
2196 			nfsrv_unlocklf(lfp);
2197 			NFSUNLOCKSTATE();
2198 			NFSVOPLOCK(vp, LK_EXCLUSIVE | LK_RETRY);
2199 			vnode_unlocked = 0;
2200 			if ((vp->v_iflag & VI_DOOMED) != 0) {
2201 				error = NFSERR_SERVERFAULT;
2202 				goto out;
2203 			}
2204 		    }
2205 		    /*
2206 		     * nfsrv_clientconflict() unlocks state when it
2207 		     * returns non-zero.
2208 		     */
2209 		    lckstp = NULL;
2210 		    goto tryagain;
2211 		}
2212 		/*
2213 		 * Found a conflicting lock, so record the conflict and
2214 		 * return the error.
2215 		 */
2216 		if (cfp != NULL && ret == 0) {
2217 		    cfp->cl_clientid.lval[0]=lop->lo_stp->ls_stateid.other[0];
2218 		    cfp->cl_clientid.lval[1]=lop->lo_stp->ls_stateid.other[1];
2219 		    cfp->cl_first = lop->lo_first;
2220 		    cfp->cl_end = lop->lo_end;
2221 		    cfp->cl_flags = lop->lo_flags;
2222 		    cfp->cl_ownerlen = lop->lo_stp->ls_ownerlen;
2223 		    NFSBCOPY(lop->lo_stp->ls_owner, cfp->cl_owner,
2224 			cfp->cl_ownerlen);
2225 		}
2226 		if (ret == 2)
2227 		    error = NFSERR_PERM;
2228 		else if (new_stp->ls_flags & NFSLCK_RECLAIM)
2229 		    error = NFSERR_RECLAIMCONFLICT;
2230 		else if (new_stp->ls_flags & NFSLCK_CHECK)
2231 		    error = NFSERR_LOCKED;
2232 		else
2233 		    error = NFSERR_DENIED;
2234 		if (filestruct_locked != 0 && ret == 0) {
2235 			/* Roll back local locks. */
2236 			NFSUNLOCKSTATE();
2237 			if (vnode_unlocked == 0) {
2238 				ASSERT_VOP_ELOCKED(vp, "nfsrv_lockctrl7");
2239 				vnode_unlocked = 1;
2240 				NFSVOPUNLOCK(vp, 0);
2241 			}
2242 			nfsrv_locallock_rollback(vp, lfp, p);
2243 			NFSLOCKSTATE();
2244 			nfsrv_unlocklf(lfp);
2245 		}
2246 		if (ret == 0)
2247 			NFSUNLOCKSTATE();
2248 		goto out;
2249 	    }
2250 	  }
2251 	}
2252 
2253 	/*
2254 	 * We only get here if there was no lock that conflicted.
2255 	 */
2256 	if (new_stp->ls_flags & (NFSLCK_TEST | NFSLCK_CHECK)) {
2257 		NFSUNLOCKSTATE();
2258 		goto out;
2259 	}
2260 
2261 	/*
2262 	 * We only get here when we are creating or modifying a lock.
2263 	 * There are two variants:
2264 	 * - exist_lock_owner where lock_owner exists
2265 	 * - open_to_lock_owner with new lock_owner
2266 	 */
2267 	first = new_lop->lo_first;
2268 	end = new_lop->lo_end;
2269 	lock_flags = new_lop->lo_flags;
2270 	if (!(new_stp->ls_flags & NFSLCK_OPENTOLOCK)) {
2271 		nfsrv_updatelock(lckstp, new_lopp, &other_lop, lfp);
2272 		stateidp->seqid = ++(lckstp->ls_stateid.seqid);
2273 		if ((nd->nd_flag & ND_NFSV41) != 0 && stateidp->seqid == 0)
2274 			stateidp->seqid = lckstp->ls_stateid.seqid = 1;
2275 		stateidp->other[0] = lckstp->ls_stateid.other[0];
2276 		stateidp->other[1] = lckstp->ls_stateid.other[1];
2277 		stateidp->other[2] = lckstp->ls_stateid.other[2];
2278 	} else {
2279 		/*
2280 		 * The new open_to_lock_owner case.
2281 		 * Link the new nfsstate into the lists.
2282 		 */
2283 		new_stp->ls_seq = new_stp->ls_opentolockseq;
2284 		nfsrvd_refcache(new_stp->ls_op);
2285 		stateidp->seqid = new_stp->ls_stateid.seqid = 1;
2286 		stateidp->other[0] = new_stp->ls_stateid.other[0] =
2287 		    clp->lc_clientid.lval[0];
2288 		stateidp->other[1] = new_stp->ls_stateid.other[1] =
2289 		    clp->lc_clientid.lval[1];
2290 		stateidp->other[2] = new_stp->ls_stateid.other[2] =
2291 		    nfsrv_nextstateindex(clp);
2292 		new_stp->ls_clp = clp;
2293 		LIST_INIT(&new_stp->ls_lock);
2294 		new_stp->ls_openstp = stp;
2295 		new_stp->ls_lfp = lfp;
2296 		nfsrv_insertlock(new_lop, (struct nfslock *)new_stp, new_stp,
2297 		    lfp);
2298 		LIST_INSERT_HEAD(NFSSTATEHASH(clp, new_stp->ls_stateid),
2299 		    new_stp, ls_hash);
2300 		LIST_INSERT_HEAD(&stp->ls_open, new_stp, ls_list);
2301 		*new_lopp = NULL;
2302 		*new_stpp = NULL;
2303 		nfsstatsv1.srvlockowners++;
2304 		nfsrv_openpluslock++;
2305 	}
2306 	if (filestruct_locked != 0) {
2307 		NFSUNLOCKSTATE();
2308 		nfsrv_locallock_commit(lfp, lock_flags, first, end);
2309 		NFSLOCKSTATE();
2310 		nfsrv_unlocklf(lfp);
2311 	}
2312 	NFSUNLOCKSTATE();
2313 
2314 out:
2315 	if (haslock) {
2316 		NFSLOCKV4ROOTMUTEX();
2317 		nfsv4_unlock(&nfsv4rootfs_lock, 1);
2318 		NFSUNLOCKV4ROOTMUTEX();
2319 	}
2320 	if (vnode_unlocked != 0) {
2321 		NFSVOPLOCK(vp, LK_EXCLUSIVE | LK_RETRY);
2322 		if (error == 0 && (vp->v_iflag & VI_DOOMED) != 0)
2323 			error = NFSERR_SERVERFAULT;
2324 	}
2325 	if (other_lop)
2326 		free(other_lop, M_NFSDLOCK);
2327 	NFSEXITCODE2(error, nd);
2328 	return (error);
2329 }
2330 
2331 /*
2332  * Check for state errors for Open.
2333  * repstat is passed back out as an error if more critical errors
2334  * are not detected.
2335  */
2336 APPLESTATIC int
2337 nfsrv_opencheck(nfsquad_t clientid, nfsv4stateid_t *stateidp,
2338     struct nfsstate *new_stp, vnode_t vp, struct nfsrv_descript *nd,
2339     NFSPROC_T *p, int repstat)
2340 {
2341 	struct nfsstate *stp, *nstp;
2342 	struct nfsclient *clp;
2343 	struct nfsstate *ownerstp;
2344 	struct nfslockfile *lfp, *new_lfp;
2345 	int error = 0, haslock = 0, ret, readonly = 0, getfhret = 0;
2346 
2347 	if ((new_stp->ls_flags & NFSLCK_SHAREBITS) == NFSLCK_READACCESS)
2348 		readonly = 1;
2349 	/*
2350 	 * Check for restart conditions (client and server).
2351 	 */
2352 	error = nfsrv_checkrestart(clientid, new_stp->ls_flags,
2353 		&new_stp->ls_stateid, 0);
2354 	if (error)
2355 		goto out;
2356 
2357 	/*
2358 	 * Check for state resource limit exceeded.
2359 	 * Technically this should be SMP protected, but the worst
2360 	 * case error is "out by one or two" on the count when it
2361 	 * returns NFSERR_RESOURCE and the limit is just a rather
2362 	 * arbitrary high water mark, so no harm is done.
2363 	 */
2364 	if (nfsrv_openpluslock > nfsrv_v4statelimit) {
2365 		error = NFSERR_RESOURCE;
2366 		goto out;
2367 	}
2368 
2369 tryagain:
2370 	new_lfp = malloc(sizeof (struct nfslockfile),
2371 	    M_NFSDLOCKFILE, M_WAITOK);
2372 	if (vp)
2373 		getfhret = nfsrv_getlockfh(vp, new_stp->ls_flags, new_lfp,
2374 		    NULL, p);
2375 	NFSLOCKSTATE();
2376 	/*
2377 	 * Get the nfsclient structure.
2378 	 */
2379 	error = nfsrv_getclient(clientid, CLOPS_RENEW, &clp, NULL,
2380 	    (nfsquad_t)((u_quad_t)0), 0, nd, p);
2381 
2382 	/*
2383 	 * Look up the open owner. See if it needs confirmation and
2384 	 * check the seq#, as required.
2385 	 */
2386 	if (!error)
2387 		nfsrv_getowner(&clp->lc_open, new_stp, &ownerstp);
2388 
2389 	if (!error && ownerstp) {
2390 		error = nfsrv_checkseqid(nd, new_stp->ls_seq, ownerstp,
2391 		    new_stp->ls_op);
2392 		/*
2393 		 * If the OpenOwner hasn't been confirmed, assume the
2394 		 * old one was a replay and this one is ok.
2395 		 * See: RFC3530 Sec. 14.2.18.
2396 		 */
2397 		if (error == NFSERR_BADSEQID &&
2398 		    (ownerstp->ls_flags & NFSLCK_NEEDSCONFIRM))
2399 			error = 0;
2400 	}
2401 
2402 	/*
2403 	 * Check for grace.
2404 	 */
2405 	if (!error)
2406 		error = nfsrv_checkgrace(nd, clp, new_stp->ls_flags);
2407 	if ((new_stp->ls_flags & NFSLCK_RECLAIM) && !error &&
2408 		nfsrv_checkstable(clp))
2409 		error = NFSERR_NOGRACE;
2410 
2411 	/*
2412 	 * If none of the above errors occurred, let repstat be
2413 	 * returned.
2414 	 */
2415 	if (repstat && !error)
2416 		error = repstat;
2417 	if (error) {
2418 		NFSUNLOCKSTATE();
2419 		if (haslock) {
2420 			NFSLOCKV4ROOTMUTEX();
2421 			nfsv4_unlock(&nfsv4rootfs_lock, 1);
2422 			NFSUNLOCKV4ROOTMUTEX();
2423 		}
2424 		free(new_lfp, M_NFSDLOCKFILE);
2425 		goto out;
2426 	}
2427 
2428 	/*
2429 	 * If vp == NULL, the file doesn't exist yet, so return ok.
2430 	 * (This always happens on the first pass, so haslock must be 0.)
2431 	 */
2432 	if (vp == NULL) {
2433 		NFSUNLOCKSTATE();
2434 		free(new_lfp, M_NFSDLOCKFILE);
2435 		goto out;
2436 	}
2437 
2438 	/*
2439 	 * Get the structure for the underlying file.
2440 	 */
2441 	if (getfhret)
2442 		error = getfhret;
2443 	else
2444 		error = nfsrv_getlockfile(new_stp->ls_flags, &new_lfp, &lfp,
2445 		    NULL, 0);
2446 	if (new_lfp)
2447 		free(new_lfp, M_NFSDLOCKFILE);
2448 	if (error) {
2449 		NFSUNLOCKSTATE();
2450 		if (haslock) {
2451 			NFSLOCKV4ROOTMUTEX();
2452 			nfsv4_unlock(&nfsv4rootfs_lock, 1);
2453 			NFSUNLOCKV4ROOTMUTEX();
2454 		}
2455 		goto out;
2456 	}
2457 
2458 	/*
2459 	 * Search for a conflicting open/share.
2460 	 */
2461 	if (new_stp->ls_flags & NFSLCK_DELEGCUR) {
2462 	    /*
2463 	     * For Delegate_Cur, search for the matching Delegation,
2464 	     * which indicates no conflict.
2465 	     * An old delegation should have been recovered by the
2466 	     * client doing a Claim_DELEGATE_Prev, so I won't let
2467 	     * it match and return NFSERR_EXPIRED. Should I let it
2468 	     * match?
2469 	     */
2470 	    LIST_FOREACH(stp, &lfp->lf_deleg, ls_file) {
2471 		if (!(stp->ls_flags & NFSLCK_OLDDELEG) &&
2472 		    (((nd->nd_flag & ND_NFSV41) != 0 &&
2473 		    stateidp->seqid == 0) ||
2474 		    stateidp->seqid == stp->ls_stateid.seqid) &&
2475 		    !NFSBCMP(stateidp->other, stp->ls_stateid.other,
2476 			  NFSX_STATEIDOTHER))
2477 			break;
2478 	    }
2479 	    if (stp == LIST_END(&lfp->lf_deleg) ||
2480 		((new_stp->ls_flags & NFSLCK_WRITEACCESS) &&
2481 		 (stp->ls_flags & NFSLCK_DELEGREAD))) {
2482 		NFSUNLOCKSTATE();
2483 		if (haslock) {
2484 			NFSLOCKV4ROOTMUTEX();
2485 			nfsv4_unlock(&nfsv4rootfs_lock, 1);
2486 			NFSUNLOCKV4ROOTMUTEX();
2487 		}
2488 		error = NFSERR_EXPIRED;
2489 		goto out;
2490 	    }
2491 	}
2492 
2493 	/*
2494 	 * Check for access/deny bit conflicts. I check for the same
2495 	 * owner as well, in case the client didn't bother.
2496 	 */
2497 	LIST_FOREACH(stp, &lfp->lf_open, ls_file) {
2498 		if (!(new_stp->ls_flags & NFSLCK_DELEGCUR) &&
2499 		    (((new_stp->ls_flags & NFSLCK_ACCESSBITS) &
2500 		      ((stp->ls_flags>>NFSLCK_SHIFT) & NFSLCK_ACCESSBITS))||
2501 		     ((stp->ls_flags & NFSLCK_ACCESSBITS) &
2502 		      ((new_stp->ls_flags>>NFSLCK_SHIFT)&NFSLCK_ACCESSBITS)))){
2503 			ret = nfsrv_clientconflict(stp->ls_clp,&haslock,vp,p);
2504 			if (ret == 1) {
2505 				/*
2506 				 * nfsrv_clientconflict() unlocks
2507 				 * state when it returns non-zero.
2508 				 */
2509 				goto tryagain;
2510 			}
2511 			if (ret == 2)
2512 				error = NFSERR_PERM;
2513 			else if (new_stp->ls_flags & NFSLCK_RECLAIM)
2514 				error = NFSERR_RECLAIMCONFLICT;
2515 			else
2516 				error = NFSERR_SHAREDENIED;
2517 			if (ret == 0)
2518 				NFSUNLOCKSTATE();
2519 			if (haslock) {
2520 				NFSLOCKV4ROOTMUTEX();
2521 				nfsv4_unlock(&nfsv4rootfs_lock, 1);
2522 				NFSUNLOCKV4ROOTMUTEX();
2523 			}
2524 			goto out;
2525 		}
2526 	}
2527 
2528 	/*
2529 	 * Check for a conflicting delegation. If one is found, call
2530 	 * nfsrv_delegconflict() to handle it. If the v4root lock hasn't
2531 	 * been set yet, it will get the lock. Otherwise, it will recall
2532 	 * the delegation. Then, we try try again...
2533 	 * (If NFSLCK_DELEGCUR is set, it has a delegation, so there
2534 	 *  isn't a conflict.)
2535 	 * I currently believe the conflict algorithm to be:
2536 	 * For Open with Read Access and Deny None
2537 	 * - there is a conflict iff a different client has a write delegation
2538 	 * For Open with other Write Access or any Deny except None
2539 	 * - there is a conflict if a different client has any delegation
2540 	 * - there is a conflict if the same client has a read delegation
2541 	 *   (The current consensus is that this last case should be
2542 	 *    considered a conflict since the client with a read delegation
2543 	 *    could have done an Open with ReadAccess and WriteDeny
2544 	 *    locally and then not have checked for the WriteDeny.)
2545 	 * Don't check for a Reclaim, since that will be dealt with
2546 	 * by nfsrv_openctrl().
2547 	 */
2548 	if (!(new_stp->ls_flags &
2549 		(NFSLCK_DELEGPREV | NFSLCK_DELEGCUR | NFSLCK_RECLAIM))) {
2550 	    stp = LIST_FIRST(&lfp->lf_deleg);
2551 	    while (stp != LIST_END(&lfp->lf_deleg)) {
2552 		nstp = LIST_NEXT(stp, ls_file);
2553 		if ((readonly && stp->ls_clp != clp &&
2554 		       (stp->ls_flags & NFSLCK_DELEGWRITE)) ||
2555 		    (!readonly && (stp->ls_clp != clp ||
2556 		         (stp->ls_flags & NFSLCK_DELEGREAD)))) {
2557 			ret = nfsrv_delegconflict(stp, &haslock, p, vp);
2558 			if (ret) {
2559 			    /*
2560 			     * nfsrv_delegconflict() unlocks state
2561 			     * when it returns non-zero.
2562 			     */
2563 			    if (ret == -1)
2564 				goto tryagain;
2565 			    error = ret;
2566 			    goto out;
2567 			}
2568 		}
2569 		stp = nstp;
2570 	    }
2571 	}
2572 	NFSUNLOCKSTATE();
2573 	if (haslock) {
2574 		NFSLOCKV4ROOTMUTEX();
2575 		nfsv4_unlock(&nfsv4rootfs_lock, 1);
2576 		NFSUNLOCKV4ROOTMUTEX();
2577 	}
2578 
2579 out:
2580 	NFSEXITCODE2(error, nd);
2581 	return (error);
2582 }
2583 
2584 /*
2585  * Open control function to create/update open state for an open.
2586  */
2587 APPLESTATIC int
2588 nfsrv_openctrl(struct nfsrv_descript *nd, vnode_t vp,
2589     struct nfsstate **new_stpp, nfsquad_t clientid, nfsv4stateid_t *stateidp,
2590     nfsv4stateid_t *delegstateidp, u_int32_t *rflagsp, struct nfsexstuff *exp,
2591     NFSPROC_T *p, u_quad_t filerev)
2592 {
2593 	struct nfsstate *new_stp = *new_stpp;
2594 	struct nfsstate *stp, *nstp;
2595 	struct nfsstate *openstp = NULL, *new_open, *ownerstp, *new_deleg;
2596 	struct nfslockfile *lfp, *new_lfp;
2597 	struct nfsclient *clp;
2598 	int error = 0, haslock = 0, ret, delegate = 1, writedeleg = 1;
2599 	int readonly = 0, cbret = 1, getfhret = 0;
2600 	int gotstate = 0, len = 0;
2601 	u_char *clidp = NULL;
2602 
2603 	if ((new_stp->ls_flags & NFSLCK_SHAREBITS) == NFSLCK_READACCESS)
2604 		readonly = 1;
2605 	/*
2606 	 * Check for restart conditions (client and server).
2607 	 * (Paranoia, should have been detected by nfsrv_opencheck().)
2608 	 * If an error does show up, return NFSERR_EXPIRED, since the
2609 	 * the seqid# has already been incremented.
2610 	 */
2611 	error = nfsrv_checkrestart(clientid, new_stp->ls_flags,
2612 	    &new_stp->ls_stateid, 0);
2613 	if (error) {
2614 		printf("Nfsd: openctrl unexpected restart err=%d\n",
2615 		    error);
2616 		error = NFSERR_EXPIRED;
2617 		goto out;
2618 	}
2619 
2620 	clidp = malloc(NFSV4_OPAQUELIMIT, M_TEMP, M_WAITOK);
2621 tryagain:
2622 	new_lfp = malloc(sizeof (struct nfslockfile),
2623 	    M_NFSDLOCKFILE, M_WAITOK);
2624 	new_open = malloc(sizeof (struct nfsstate),
2625 	    M_NFSDSTATE, M_WAITOK);
2626 	new_deleg = malloc(sizeof (struct nfsstate),
2627 	    M_NFSDSTATE, M_WAITOK);
2628 	getfhret = nfsrv_getlockfh(vp, new_stp->ls_flags, new_lfp,
2629 	    NULL, p);
2630 	NFSLOCKSTATE();
2631 	/*
2632 	 * Get the client structure. Since the linked lists could be changed
2633 	 * by other nfsd processes if this process does a tsleep(), one of
2634 	 * two things must be done.
2635 	 * 1 - don't tsleep()
2636 	 * or
2637 	 * 2 - get the nfsv4_lock() { indicated by haslock == 1 }
2638 	 *     before using the lists, since this lock stops the other
2639 	 *     nfsd. This should only be used for rare cases, since it
2640 	 *     essentially single threads the nfsd.
2641 	 *     At this time, it is only done for cases where the stable
2642 	 *     storage file must be written prior to completion of state
2643 	 *     expiration.
2644 	 */
2645 	error = nfsrv_getclient(clientid, CLOPS_RENEW, &clp, NULL,
2646 	    (nfsquad_t)((u_quad_t)0), 0, nd, p);
2647 	if (!error && (clp->lc_flags & LCL_NEEDSCBNULL) &&
2648 	    clp->lc_program) {
2649 		/*
2650 		 * This happens on the first open for a client
2651 		 * that supports callbacks.
2652 		 */
2653 		NFSUNLOCKSTATE();
2654 		/*
2655 		 * Although nfsrv_docallback() will sleep, clp won't
2656 		 * go away, since they are only removed when the
2657 		 * nfsv4_lock() has blocked the nfsd threads. The
2658 		 * fields in clp can change, but having multiple
2659 		 * threads do this Null callback RPC should be
2660 		 * harmless.
2661 		 */
2662 		cbret = nfsrv_docallback(clp, NFSV4PROC_CBNULL,
2663 		    NULL, 0, NULL, NULL, NULL, 0, p);
2664 		NFSLOCKSTATE();
2665 		clp->lc_flags &= ~LCL_NEEDSCBNULL;
2666 		if (!cbret)
2667 			clp->lc_flags |= LCL_CALLBACKSON;
2668 	}
2669 
2670 	/*
2671 	 * Look up the open owner. See if it needs confirmation and
2672 	 * check the seq#, as required.
2673 	 */
2674 	if (!error)
2675 		nfsrv_getowner(&clp->lc_open, new_stp, &ownerstp);
2676 
2677 	if (error) {
2678 		NFSUNLOCKSTATE();
2679 		printf("Nfsd: openctrl unexpected state err=%d\n",
2680 			error);
2681 		free(new_lfp, M_NFSDLOCKFILE);
2682 		free(new_open, M_NFSDSTATE);
2683 		free(new_deleg, M_NFSDSTATE);
2684 		if (haslock) {
2685 			NFSLOCKV4ROOTMUTEX();
2686 			nfsv4_unlock(&nfsv4rootfs_lock, 1);
2687 			NFSUNLOCKV4ROOTMUTEX();
2688 		}
2689 		error = NFSERR_EXPIRED;
2690 		goto out;
2691 	}
2692 
2693 	if (new_stp->ls_flags & NFSLCK_RECLAIM)
2694 		nfsrv_markstable(clp);
2695 
2696 	/*
2697 	 * Get the structure for the underlying file.
2698 	 */
2699 	if (getfhret)
2700 		error = getfhret;
2701 	else
2702 		error = nfsrv_getlockfile(new_stp->ls_flags, &new_lfp, &lfp,
2703 		    NULL, 0);
2704 	if (new_lfp)
2705 		free(new_lfp, M_NFSDLOCKFILE);
2706 	if (error) {
2707 		NFSUNLOCKSTATE();
2708 		printf("Nfsd openctrl unexpected getlockfile err=%d\n",
2709 		    error);
2710 		free(new_open, M_NFSDSTATE);
2711 		free(new_deleg, M_NFSDSTATE);
2712 		if (haslock) {
2713 			NFSLOCKV4ROOTMUTEX();
2714 			nfsv4_unlock(&nfsv4rootfs_lock, 1);
2715 			NFSUNLOCKV4ROOTMUTEX();
2716 		}
2717 		goto out;
2718 	}
2719 
2720 	/*
2721 	 * Search for a conflicting open/share.
2722 	 */
2723 	if (new_stp->ls_flags & NFSLCK_DELEGCUR) {
2724 	    /*
2725 	     * For Delegate_Cur, search for the matching Delegation,
2726 	     * which indicates no conflict.
2727 	     * An old delegation should have been recovered by the
2728 	     * client doing a Claim_DELEGATE_Prev, so I won't let
2729 	     * it match and return NFSERR_EXPIRED. Should I let it
2730 	     * match?
2731 	     */
2732 	    LIST_FOREACH(stp, &lfp->lf_deleg, ls_file) {
2733 		if (!(stp->ls_flags & NFSLCK_OLDDELEG) &&
2734 		    (((nd->nd_flag & ND_NFSV41) != 0 &&
2735 		    stateidp->seqid == 0) ||
2736 		    stateidp->seqid == stp->ls_stateid.seqid) &&
2737 		    !NFSBCMP(stateidp->other, stp->ls_stateid.other,
2738 			NFSX_STATEIDOTHER))
2739 			break;
2740 	    }
2741 	    if (stp == LIST_END(&lfp->lf_deleg) ||
2742 		((new_stp->ls_flags & NFSLCK_WRITEACCESS) &&
2743 		 (stp->ls_flags & NFSLCK_DELEGREAD))) {
2744 		NFSUNLOCKSTATE();
2745 		printf("Nfsd openctrl unexpected expiry\n");
2746 		free(new_open, M_NFSDSTATE);
2747 		free(new_deleg, M_NFSDSTATE);
2748 		if (haslock) {
2749 			NFSLOCKV4ROOTMUTEX();
2750 			nfsv4_unlock(&nfsv4rootfs_lock, 1);
2751 			NFSUNLOCKV4ROOTMUTEX();
2752 		}
2753 		error = NFSERR_EXPIRED;
2754 		goto out;
2755 	    }
2756 
2757 	    /*
2758 	     * Don't issue a Delegation, since one already exists and
2759 	     * delay delegation timeout, as required.
2760 	     */
2761 	    delegate = 0;
2762 	    nfsrv_delaydelegtimeout(stp);
2763 	}
2764 
2765 	/*
2766 	 * Check for access/deny bit conflicts. I also check for the
2767 	 * same owner, since the client might not have bothered to check.
2768 	 * Also, note an open for the same file and owner, if found,
2769 	 * which is all we do here for Delegate_Cur, since conflict
2770 	 * checking is already done.
2771 	 */
2772 	LIST_FOREACH(stp, &lfp->lf_open, ls_file) {
2773 		if (ownerstp && stp->ls_openowner == ownerstp)
2774 			openstp = stp;
2775 		if (!(new_stp->ls_flags & NFSLCK_DELEGCUR)) {
2776 		    /*
2777 		     * If another client has the file open, the only
2778 		     * delegation that can be issued is a Read delegation
2779 		     * and only if it is a Read open with Deny none.
2780 		     */
2781 		    if (clp != stp->ls_clp) {
2782 			if ((stp->ls_flags & NFSLCK_SHAREBITS) ==
2783 			    NFSLCK_READACCESS)
2784 			    writedeleg = 0;
2785 			else
2786 			    delegate = 0;
2787 		    }
2788 		    if(((new_stp->ls_flags & NFSLCK_ACCESSBITS) &
2789 		        ((stp->ls_flags>>NFSLCK_SHIFT) & NFSLCK_ACCESSBITS))||
2790 		       ((stp->ls_flags & NFSLCK_ACCESSBITS) &
2791 		        ((new_stp->ls_flags>>NFSLCK_SHIFT)&NFSLCK_ACCESSBITS))){
2792 			ret = nfsrv_clientconflict(stp->ls_clp,&haslock,vp,p);
2793 			if (ret == 1) {
2794 				/*
2795 				 * nfsrv_clientconflict() unlocks state
2796 				 * when it returns non-zero.
2797 				 */
2798 				free(new_open, M_NFSDSTATE);
2799 				free(new_deleg, M_NFSDSTATE);
2800 				openstp = NULL;
2801 				goto tryagain;
2802 			}
2803 			if (ret == 2)
2804 				error = NFSERR_PERM;
2805 			else if (new_stp->ls_flags & NFSLCK_RECLAIM)
2806 				error = NFSERR_RECLAIMCONFLICT;
2807 			else
2808 				error = NFSERR_SHAREDENIED;
2809 			if (ret == 0)
2810 				NFSUNLOCKSTATE();
2811 			if (haslock) {
2812 				NFSLOCKV4ROOTMUTEX();
2813 				nfsv4_unlock(&nfsv4rootfs_lock, 1);
2814 				NFSUNLOCKV4ROOTMUTEX();
2815 			}
2816 			free(new_open, M_NFSDSTATE);
2817 			free(new_deleg, M_NFSDSTATE);
2818 			printf("nfsd openctrl unexpected client cnfl\n");
2819 			goto out;
2820 		    }
2821 		}
2822 	}
2823 
2824 	/*
2825 	 * Check for a conflicting delegation. If one is found, call
2826 	 * nfsrv_delegconflict() to handle it. If the v4root lock hasn't
2827 	 * been set yet, it will get the lock. Otherwise, it will recall
2828 	 * the delegation. Then, we try try again...
2829 	 * (If NFSLCK_DELEGCUR is set, it has a delegation, so there
2830 	 *  isn't a conflict.)
2831 	 * I currently believe the conflict algorithm to be:
2832 	 * For Open with Read Access and Deny None
2833 	 * - there is a conflict iff a different client has a write delegation
2834 	 * For Open with other Write Access or any Deny except None
2835 	 * - there is a conflict if a different client has any delegation
2836 	 * - there is a conflict if the same client has a read delegation
2837 	 *   (The current consensus is that this last case should be
2838 	 *    considered a conflict since the client with a read delegation
2839 	 *    could have done an Open with ReadAccess and WriteDeny
2840 	 *    locally and then not have checked for the WriteDeny.)
2841 	 */
2842 	if (!(new_stp->ls_flags & (NFSLCK_DELEGPREV | NFSLCK_DELEGCUR))) {
2843 	    stp = LIST_FIRST(&lfp->lf_deleg);
2844 	    while (stp != LIST_END(&lfp->lf_deleg)) {
2845 		nstp = LIST_NEXT(stp, ls_file);
2846 		if (stp->ls_clp != clp && (stp->ls_flags & NFSLCK_DELEGREAD))
2847 			writedeleg = 0;
2848 		else
2849 			delegate = 0;
2850 		if ((readonly && stp->ls_clp != clp &&
2851 		       (stp->ls_flags & NFSLCK_DELEGWRITE)) ||
2852 		    (!readonly && (stp->ls_clp != clp ||
2853 		         (stp->ls_flags & NFSLCK_DELEGREAD)))) {
2854 		    if (new_stp->ls_flags & NFSLCK_RECLAIM) {
2855 			delegate = 2;
2856 		    } else {
2857 			ret = nfsrv_delegconflict(stp, &haslock, p, vp);
2858 			if (ret) {
2859 			    /*
2860 			     * nfsrv_delegconflict() unlocks state
2861 			     * when it returns non-zero.
2862 			     */
2863 			    printf("Nfsd openctrl unexpected deleg cnfl\n");
2864 			    free(new_open, M_NFSDSTATE);
2865 			    free(new_deleg, M_NFSDSTATE);
2866 			    if (ret == -1) {
2867 				openstp = NULL;
2868 				goto tryagain;
2869 			    }
2870 			    error = ret;
2871 			    goto out;
2872 			}
2873 		    }
2874 		}
2875 		stp = nstp;
2876 	    }
2877 	}
2878 
2879 	/*
2880 	 * We only get here if there was no open that conflicted.
2881 	 * If an open for the owner exists, or in the access/deny bits.
2882 	 * Otherwise it is a new open. If the open_owner hasn't been
2883 	 * confirmed, replace the open with the new one needing confirmation,
2884 	 * otherwise add the open.
2885 	 */
2886 	if (new_stp->ls_flags & NFSLCK_DELEGPREV) {
2887 	    /*
2888 	     * Handle NFSLCK_DELEGPREV by searching the old delegations for
2889 	     * a match. If found, just move the old delegation to the current
2890 	     * delegation list and issue open. If not found, return
2891 	     * NFSERR_EXPIRED.
2892 	     */
2893 	    LIST_FOREACH(stp, &clp->lc_olddeleg, ls_list) {
2894 		if (stp->ls_lfp == lfp) {
2895 		    /* Found it */
2896 		    if (stp->ls_clp != clp)
2897 			panic("olddeleg clp");
2898 		    LIST_REMOVE(stp, ls_list);
2899 		    LIST_REMOVE(stp, ls_hash);
2900 		    stp->ls_flags &= ~NFSLCK_OLDDELEG;
2901 		    stp->ls_stateid.seqid = delegstateidp->seqid = 1;
2902 		    stp->ls_stateid.other[0] = delegstateidp->other[0] =
2903 			clp->lc_clientid.lval[0];
2904 		    stp->ls_stateid.other[1] = delegstateidp->other[1] =
2905 			clp->lc_clientid.lval[1];
2906 		    stp->ls_stateid.other[2] = delegstateidp->other[2] =
2907 			nfsrv_nextstateindex(clp);
2908 		    stp->ls_compref = nd->nd_compref;
2909 		    LIST_INSERT_HEAD(&clp->lc_deleg, stp, ls_list);
2910 		    LIST_INSERT_HEAD(NFSSTATEHASH(clp,
2911 			stp->ls_stateid), stp, ls_hash);
2912 		    if (stp->ls_flags & NFSLCK_DELEGWRITE)
2913 			*rflagsp |= NFSV4OPEN_WRITEDELEGATE;
2914 		    else
2915 			*rflagsp |= NFSV4OPEN_READDELEGATE;
2916 		    clp->lc_delegtime = NFSD_MONOSEC +
2917 			nfsrv_lease + NFSRV_LEASEDELTA;
2918 
2919 		    /*
2920 		     * Now, do the associated open.
2921 		     */
2922 		    new_open->ls_stateid.seqid = 1;
2923 		    new_open->ls_stateid.other[0] = clp->lc_clientid.lval[0];
2924 		    new_open->ls_stateid.other[1] = clp->lc_clientid.lval[1];
2925 		    new_open->ls_stateid.other[2] = nfsrv_nextstateindex(clp);
2926 		    new_open->ls_flags = (new_stp->ls_flags&NFSLCK_DENYBITS)|
2927 			NFSLCK_OPEN;
2928 		    if (stp->ls_flags & NFSLCK_DELEGWRITE)
2929 			new_open->ls_flags |= (NFSLCK_READACCESS |
2930 			    NFSLCK_WRITEACCESS);
2931 		    else
2932 			new_open->ls_flags |= NFSLCK_READACCESS;
2933 		    new_open->ls_uid = new_stp->ls_uid;
2934 		    new_open->ls_lfp = lfp;
2935 		    new_open->ls_clp = clp;
2936 		    LIST_INIT(&new_open->ls_open);
2937 		    LIST_INSERT_HEAD(&lfp->lf_open, new_open, ls_file);
2938 		    LIST_INSERT_HEAD(NFSSTATEHASH(clp, new_open->ls_stateid),
2939 			new_open, ls_hash);
2940 		    /*
2941 		     * and handle the open owner
2942 		     */
2943 		    if (ownerstp) {
2944 			new_open->ls_openowner = ownerstp;
2945 			LIST_INSERT_HEAD(&ownerstp->ls_open,new_open,ls_list);
2946 		    } else {
2947 			new_open->ls_openowner = new_stp;
2948 			new_stp->ls_flags = 0;
2949 			nfsrvd_refcache(new_stp->ls_op);
2950 			new_stp->ls_noopens = 0;
2951 			LIST_INIT(&new_stp->ls_open);
2952 			LIST_INSERT_HEAD(&new_stp->ls_open, new_open, ls_list);
2953 			LIST_INSERT_HEAD(&clp->lc_open, new_stp, ls_list);
2954 			*new_stpp = NULL;
2955 			nfsstatsv1.srvopenowners++;
2956 			nfsrv_openpluslock++;
2957 		    }
2958 		    openstp = new_open;
2959 		    new_open = NULL;
2960 		    nfsstatsv1.srvopens++;
2961 		    nfsrv_openpluslock++;
2962 		    break;
2963 		}
2964 	    }
2965 	    if (stp == LIST_END(&clp->lc_olddeleg))
2966 		error = NFSERR_EXPIRED;
2967 	} else if (new_stp->ls_flags & (NFSLCK_DELEGREAD | NFSLCK_DELEGWRITE)) {
2968 	    /*
2969 	     * Scan to see that no delegation for this client and file
2970 	     * doesn't already exist.
2971 	     * There also shouldn't yet be an Open for this file and
2972 	     * openowner.
2973 	     */
2974 	    LIST_FOREACH(stp, &lfp->lf_deleg, ls_file) {
2975 		if (stp->ls_clp == clp)
2976 		    break;
2977 	    }
2978 	    if (stp == LIST_END(&lfp->lf_deleg) && openstp == NULL) {
2979 		/*
2980 		 * This is the Claim_Previous case with a delegation
2981 		 * type != Delegate_None.
2982 		 */
2983 		/*
2984 		 * First, add the delegation. (Although we must issue the
2985 		 * delegation, we can also ask for an immediate return.)
2986 		 */
2987 		new_deleg->ls_stateid.seqid = delegstateidp->seqid = 1;
2988 		new_deleg->ls_stateid.other[0] = delegstateidp->other[0] =
2989 		    clp->lc_clientid.lval[0];
2990 		new_deleg->ls_stateid.other[1] = delegstateidp->other[1] =
2991 		    clp->lc_clientid.lval[1];
2992 		new_deleg->ls_stateid.other[2] = delegstateidp->other[2] =
2993 		    nfsrv_nextstateindex(clp);
2994 		if (new_stp->ls_flags & NFSLCK_DELEGWRITE) {
2995 		    new_deleg->ls_flags = (NFSLCK_DELEGWRITE |
2996 			NFSLCK_READACCESS | NFSLCK_WRITEACCESS);
2997 		    *rflagsp |= NFSV4OPEN_WRITEDELEGATE;
2998 		    nfsrv_writedelegcnt++;
2999 		} else {
3000 		    new_deleg->ls_flags = (NFSLCK_DELEGREAD |
3001 			NFSLCK_READACCESS);
3002 		    *rflagsp |= NFSV4OPEN_READDELEGATE;
3003 		}
3004 		new_deleg->ls_uid = new_stp->ls_uid;
3005 		new_deleg->ls_lfp = lfp;
3006 		new_deleg->ls_clp = clp;
3007 		new_deleg->ls_filerev = filerev;
3008 		new_deleg->ls_compref = nd->nd_compref;
3009 		LIST_INSERT_HEAD(&lfp->lf_deleg, new_deleg, ls_file);
3010 		LIST_INSERT_HEAD(NFSSTATEHASH(clp,
3011 		    new_deleg->ls_stateid), new_deleg, ls_hash);
3012 		LIST_INSERT_HEAD(&clp->lc_deleg, new_deleg, ls_list);
3013 		new_deleg = NULL;
3014 		if (delegate == 2 || nfsrv_issuedelegs == 0 ||
3015 		    (clp->lc_flags & (LCL_CALLBACKSON | LCL_CBDOWN)) !=
3016 		     LCL_CALLBACKSON ||
3017 		    NFSRV_V4DELEGLIMIT(nfsrv_delegatecnt) ||
3018 		    !NFSVNO_DELEGOK(vp))
3019 		    *rflagsp |= NFSV4OPEN_RECALL;
3020 		nfsstatsv1.srvdelegates++;
3021 		nfsrv_openpluslock++;
3022 		nfsrv_delegatecnt++;
3023 
3024 		/*
3025 		 * Now, do the associated open.
3026 		 */
3027 		new_open->ls_stateid.seqid = 1;
3028 		new_open->ls_stateid.other[0] = clp->lc_clientid.lval[0];
3029 		new_open->ls_stateid.other[1] = clp->lc_clientid.lval[1];
3030 		new_open->ls_stateid.other[2] = nfsrv_nextstateindex(clp);
3031 		new_open->ls_flags = (new_stp->ls_flags & NFSLCK_DENYBITS) |
3032 		    NFSLCK_OPEN;
3033 		if (new_stp->ls_flags & NFSLCK_DELEGWRITE)
3034 			new_open->ls_flags |= (NFSLCK_READACCESS |
3035 			    NFSLCK_WRITEACCESS);
3036 		else
3037 			new_open->ls_flags |= NFSLCK_READACCESS;
3038 		new_open->ls_uid = new_stp->ls_uid;
3039 		new_open->ls_lfp = lfp;
3040 		new_open->ls_clp = clp;
3041 		LIST_INIT(&new_open->ls_open);
3042 		LIST_INSERT_HEAD(&lfp->lf_open, new_open, ls_file);
3043 		LIST_INSERT_HEAD(NFSSTATEHASH(clp, new_open->ls_stateid),
3044 		   new_open, ls_hash);
3045 		/*
3046 		 * and handle the open owner
3047 		 */
3048 		if (ownerstp) {
3049 		    new_open->ls_openowner = ownerstp;
3050 		    LIST_INSERT_HEAD(&ownerstp->ls_open, new_open, ls_list);
3051 		} else {
3052 		    new_open->ls_openowner = new_stp;
3053 		    new_stp->ls_flags = 0;
3054 		    nfsrvd_refcache(new_stp->ls_op);
3055 		    new_stp->ls_noopens = 0;
3056 		    LIST_INIT(&new_stp->ls_open);
3057 		    LIST_INSERT_HEAD(&new_stp->ls_open, new_open, ls_list);
3058 		    LIST_INSERT_HEAD(&clp->lc_open, new_stp, ls_list);
3059 		    *new_stpp = NULL;
3060 		    nfsstatsv1.srvopenowners++;
3061 		    nfsrv_openpluslock++;
3062 		}
3063 		openstp = new_open;
3064 		new_open = NULL;
3065 		nfsstatsv1.srvopens++;
3066 		nfsrv_openpluslock++;
3067 	    } else {
3068 		error = NFSERR_RECLAIMCONFLICT;
3069 	    }
3070 	} else if (ownerstp) {
3071 		if (ownerstp->ls_flags & NFSLCK_NEEDSCONFIRM) {
3072 		    /* Replace the open */
3073 		    if (ownerstp->ls_op)
3074 			nfsrvd_derefcache(ownerstp->ls_op);
3075 		    ownerstp->ls_op = new_stp->ls_op;
3076 		    nfsrvd_refcache(ownerstp->ls_op);
3077 		    ownerstp->ls_seq = new_stp->ls_seq;
3078 		    *rflagsp |= NFSV4OPEN_RESULTCONFIRM;
3079 		    stp = LIST_FIRST(&ownerstp->ls_open);
3080 		    stp->ls_flags = (new_stp->ls_flags & NFSLCK_SHAREBITS) |
3081 			NFSLCK_OPEN;
3082 		    stp->ls_stateid.seqid = 1;
3083 		    stp->ls_uid = new_stp->ls_uid;
3084 		    if (lfp != stp->ls_lfp) {
3085 			LIST_REMOVE(stp, ls_file);
3086 			LIST_INSERT_HEAD(&lfp->lf_open, stp, ls_file);
3087 			stp->ls_lfp = lfp;
3088 		    }
3089 		    openstp = stp;
3090 		} else if (openstp) {
3091 		    openstp->ls_flags |= (new_stp->ls_flags & NFSLCK_SHAREBITS);
3092 		    openstp->ls_stateid.seqid++;
3093 		    if ((nd->nd_flag & ND_NFSV41) != 0 &&
3094 			openstp->ls_stateid.seqid == 0)
3095 			openstp->ls_stateid.seqid = 1;
3096 
3097 		    /*
3098 		     * This is where we can choose to issue a delegation.
3099 		     */
3100 		    if ((new_stp->ls_flags & NFSLCK_WANTNODELEG) != 0)
3101 			*rflagsp |= NFSV4OPEN_WDNOTWANTED;
3102 		    else if (nfsrv_issuedelegs == 0)
3103 			*rflagsp |= NFSV4OPEN_WDSUPPFTYPE;
3104 		    else if (NFSRV_V4DELEGLIMIT(nfsrv_delegatecnt))
3105 			*rflagsp |= NFSV4OPEN_WDRESOURCE;
3106 		    else if (delegate == 0 || writedeleg == 0 ||
3107 			NFSVNO_EXRDONLY(exp) || (readonly != 0 &&
3108 			nfsrv_writedelegifpos == 0) ||
3109 			!NFSVNO_DELEGOK(vp) ||
3110 			(new_stp->ls_flags & NFSLCK_WANTRDELEG) != 0 ||
3111 			(clp->lc_flags & (LCL_CALLBACKSON | LCL_CBDOWN)) !=
3112 			 LCL_CALLBACKSON)
3113 			*rflagsp |= NFSV4OPEN_WDCONTENTION;
3114 		    else {
3115 			new_deleg->ls_stateid.seqid = delegstateidp->seqid = 1;
3116 			new_deleg->ls_stateid.other[0] = delegstateidp->other[0]
3117 			    = clp->lc_clientid.lval[0];
3118 			new_deleg->ls_stateid.other[1] = delegstateidp->other[1]
3119 			    = clp->lc_clientid.lval[1];
3120 			new_deleg->ls_stateid.other[2] = delegstateidp->other[2]
3121 			    = nfsrv_nextstateindex(clp);
3122 			new_deleg->ls_flags = (NFSLCK_DELEGWRITE |
3123 			    NFSLCK_READACCESS | NFSLCK_WRITEACCESS);
3124 			*rflagsp |= NFSV4OPEN_WRITEDELEGATE;
3125 			new_deleg->ls_uid = new_stp->ls_uid;
3126 			new_deleg->ls_lfp = lfp;
3127 			new_deleg->ls_clp = clp;
3128 			new_deleg->ls_filerev = filerev;
3129 			new_deleg->ls_compref = nd->nd_compref;
3130 			nfsrv_writedelegcnt++;
3131 			LIST_INSERT_HEAD(&lfp->lf_deleg, new_deleg, ls_file);
3132 			LIST_INSERT_HEAD(NFSSTATEHASH(clp,
3133 			    new_deleg->ls_stateid), new_deleg, ls_hash);
3134 			LIST_INSERT_HEAD(&clp->lc_deleg, new_deleg, ls_list);
3135 			new_deleg = NULL;
3136 			nfsstatsv1.srvdelegates++;
3137 			nfsrv_openpluslock++;
3138 			nfsrv_delegatecnt++;
3139 		    }
3140 		} else {
3141 		    new_open->ls_stateid.seqid = 1;
3142 		    new_open->ls_stateid.other[0] = clp->lc_clientid.lval[0];
3143 		    new_open->ls_stateid.other[1] = clp->lc_clientid.lval[1];
3144 		    new_open->ls_stateid.other[2] = nfsrv_nextstateindex(clp);
3145 		    new_open->ls_flags = (new_stp->ls_flags & NFSLCK_SHAREBITS)|
3146 			NFSLCK_OPEN;
3147 		    new_open->ls_uid = new_stp->ls_uid;
3148 		    new_open->ls_openowner = ownerstp;
3149 		    new_open->ls_lfp = lfp;
3150 		    new_open->ls_clp = clp;
3151 		    LIST_INIT(&new_open->ls_open);
3152 		    LIST_INSERT_HEAD(&lfp->lf_open, new_open, ls_file);
3153 		    LIST_INSERT_HEAD(&ownerstp->ls_open, new_open, ls_list);
3154 		    LIST_INSERT_HEAD(NFSSTATEHASH(clp, new_open->ls_stateid),
3155 			new_open, ls_hash);
3156 		    openstp = new_open;
3157 		    new_open = NULL;
3158 		    nfsstatsv1.srvopens++;
3159 		    nfsrv_openpluslock++;
3160 
3161 		    /*
3162 		     * This is where we can choose to issue a delegation.
3163 		     */
3164 		    if ((new_stp->ls_flags & NFSLCK_WANTNODELEG) != 0)
3165 			*rflagsp |= NFSV4OPEN_WDNOTWANTED;
3166 		    else if (nfsrv_issuedelegs == 0)
3167 			*rflagsp |= NFSV4OPEN_WDSUPPFTYPE;
3168 		    else if (NFSRV_V4DELEGLIMIT(nfsrv_delegatecnt))
3169 			*rflagsp |= NFSV4OPEN_WDRESOURCE;
3170 		    else if (delegate == 0 || (writedeleg == 0 &&
3171 			readonly == 0) || !NFSVNO_DELEGOK(vp) ||
3172 			(clp->lc_flags & (LCL_CALLBACKSON | LCL_CBDOWN)) !=
3173 			 LCL_CALLBACKSON)
3174 			*rflagsp |= NFSV4OPEN_WDCONTENTION;
3175 		    else {
3176 			new_deleg->ls_stateid.seqid = delegstateidp->seqid = 1;
3177 			new_deleg->ls_stateid.other[0] = delegstateidp->other[0]
3178 			    = clp->lc_clientid.lval[0];
3179 			new_deleg->ls_stateid.other[1] = delegstateidp->other[1]
3180 			    = clp->lc_clientid.lval[1];
3181 			new_deleg->ls_stateid.other[2] = delegstateidp->other[2]
3182 			    = nfsrv_nextstateindex(clp);
3183 			if (writedeleg && !NFSVNO_EXRDONLY(exp) &&
3184 			    (nfsrv_writedelegifpos || !readonly) &&
3185 			    (new_stp->ls_flags & NFSLCK_WANTRDELEG) == 0) {
3186 			    new_deleg->ls_flags = (NFSLCK_DELEGWRITE |
3187 				NFSLCK_READACCESS | NFSLCK_WRITEACCESS);
3188 			    *rflagsp |= NFSV4OPEN_WRITEDELEGATE;
3189 			    nfsrv_writedelegcnt++;
3190 			} else {
3191 			    new_deleg->ls_flags = (NFSLCK_DELEGREAD |
3192 				NFSLCK_READACCESS);
3193 			    *rflagsp |= NFSV4OPEN_READDELEGATE;
3194 			}
3195 			new_deleg->ls_uid = new_stp->ls_uid;
3196 			new_deleg->ls_lfp = lfp;
3197 			new_deleg->ls_clp = clp;
3198 			new_deleg->ls_filerev = filerev;
3199 			new_deleg->ls_compref = nd->nd_compref;
3200 			LIST_INSERT_HEAD(&lfp->lf_deleg, new_deleg, ls_file);
3201 			LIST_INSERT_HEAD(NFSSTATEHASH(clp,
3202 			    new_deleg->ls_stateid), new_deleg, ls_hash);
3203 			LIST_INSERT_HEAD(&clp->lc_deleg, new_deleg, ls_list);
3204 			new_deleg = NULL;
3205 			nfsstatsv1.srvdelegates++;
3206 			nfsrv_openpluslock++;
3207 			nfsrv_delegatecnt++;
3208 		    }
3209 		}
3210 	} else {
3211 		/*
3212 		 * New owner case. Start the open_owner sequence with a
3213 		 * Needs confirmation (unless a reclaim) and hang the
3214 		 * new open off it.
3215 		 */
3216 		new_open->ls_stateid.seqid = 1;
3217 		new_open->ls_stateid.other[0] = clp->lc_clientid.lval[0];
3218 		new_open->ls_stateid.other[1] = clp->lc_clientid.lval[1];
3219 		new_open->ls_stateid.other[2] = nfsrv_nextstateindex(clp);
3220 		new_open->ls_flags = (new_stp->ls_flags & NFSLCK_SHAREBITS) |
3221 		    NFSLCK_OPEN;
3222 		new_open->ls_uid = new_stp->ls_uid;
3223 		LIST_INIT(&new_open->ls_open);
3224 		new_open->ls_openowner = new_stp;
3225 		new_open->ls_lfp = lfp;
3226 		new_open->ls_clp = clp;
3227 		LIST_INSERT_HEAD(&lfp->lf_open, new_open, ls_file);
3228 		if (new_stp->ls_flags & NFSLCK_RECLAIM) {
3229 			new_stp->ls_flags = 0;
3230 		} else if ((nd->nd_flag & ND_NFSV41) != 0) {
3231 			/* NFSv4.1 never needs confirmation. */
3232 			new_stp->ls_flags = 0;
3233 
3234 			/*
3235 			 * This is where we can choose to issue a delegation.
3236 			 */
3237 			if (delegate && nfsrv_issuedelegs &&
3238 			    (writedeleg || readonly) &&
3239 			    (clp->lc_flags & (LCL_CALLBACKSON | LCL_CBDOWN)) ==
3240 			     LCL_CALLBACKSON &&
3241 			    !NFSRV_V4DELEGLIMIT(nfsrv_delegatecnt) &&
3242 			    NFSVNO_DELEGOK(vp) &&
3243 			    ((nd->nd_flag & ND_NFSV41) == 0 ||
3244 			     (new_stp->ls_flags & NFSLCK_WANTNODELEG) == 0)) {
3245 				new_deleg->ls_stateid.seqid =
3246 				    delegstateidp->seqid = 1;
3247 				new_deleg->ls_stateid.other[0] =
3248 				    delegstateidp->other[0]
3249 				    = clp->lc_clientid.lval[0];
3250 				new_deleg->ls_stateid.other[1] =
3251 				    delegstateidp->other[1]
3252 				    = clp->lc_clientid.lval[1];
3253 				new_deleg->ls_stateid.other[2] =
3254 				    delegstateidp->other[2]
3255 				    = nfsrv_nextstateindex(clp);
3256 				if (writedeleg && !NFSVNO_EXRDONLY(exp) &&
3257 				    (nfsrv_writedelegifpos || !readonly) &&
3258 				    ((nd->nd_flag & ND_NFSV41) == 0 ||
3259 				     (new_stp->ls_flags & NFSLCK_WANTRDELEG) ==
3260 				     0)) {
3261 					new_deleg->ls_flags =
3262 					    (NFSLCK_DELEGWRITE |
3263 					     NFSLCK_READACCESS |
3264 					     NFSLCK_WRITEACCESS);
3265 					*rflagsp |= NFSV4OPEN_WRITEDELEGATE;
3266 					nfsrv_writedelegcnt++;
3267 				} else {
3268 					new_deleg->ls_flags =
3269 					    (NFSLCK_DELEGREAD |
3270 					     NFSLCK_READACCESS);
3271 					*rflagsp |= NFSV4OPEN_READDELEGATE;
3272 				}
3273 				new_deleg->ls_uid = new_stp->ls_uid;
3274 				new_deleg->ls_lfp = lfp;
3275 				new_deleg->ls_clp = clp;
3276 				new_deleg->ls_filerev = filerev;
3277 				new_deleg->ls_compref = nd->nd_compref;
3278 				LIST_INSERT_HEAD(&lfp->lf_deleg, new_deleg,
3279 				    ls_file);
3280 				LIST_INSERT_HEAD(NFSSTATEHASH(clp,
3281 				    new_deleg->ls_stateid), new_deleg, ls_hash);
3282 				LIST_INSERT_HEAD(&clp->lc_deleg, new_deleg,
3283 				    ls_list);
3284 				new_deleg = NULL;
3285 				nfsstatsv1.srvdelegates++;
3286 				nfsrv_openpluslock++;
3287 				nfsrv_delegatecnt++;
3288 			}
3289 			/*
3290 			 * Since NFSv4.1 never does an OpenConfirm, the first
3291 			 * open state will be acquired here.
3292 			 */
3293 			if (!(clp->lc_flags & LCL_STAMPEDSTABLE)) {
3294 				clp->lc_flags |= LCL_STAMPEDSTABLE;
3295 				len = clp->lc_idlen;
3296 				NFSBCOPY(clp->lc_id, clidp, len);
3297 				gotstate = 1;
3298 			}
3299 		} else {
3300 			*rflagsp |= NFSV4OPEN_RESULTCONFIRM;
3301 			new_stp->ls_flags = NFSLCK_NEEDSCONFIRM;
3302 		}
3303 		nfsrvd_refcache(new_stp->ls_op);
3304 		new_stp->ls_noopens = 0;
3305 		LIST_INIT(&new_stp->ls_open);
3306 		LIST_INSERT_HEAD(&new_stp->ls_open, new_open, ls_list);
3307 		LIST_INSERT_HEAD(&clp->lc_open, new_stp, ls_list);
3308 		LIST_INSERT_HEAD(NFSSTATEHASH(clp, new_open->ls_stateid),
3309 		    new_open, ls_hash);
3310 		openstp = new_open;
3311 		new_open = NULL;
3312 		*new_stpp = NULL;
3313 		nfsstatsv1.srvopens++;
3314 		nfsrv_openpluslock++;
3315 		nfsstatsv1.srvopenowners++;
3316 		nfsrv_openpluslock++;
3317 	}
3318 	if (!error) {
3319 		stateidp->seqid = openstp->ls_stateid.seqid;
3320 		stateidp->other[0] = openstp->ls_stateid.other[0];
3321 		stateidp->other[1] = openstp->ls_stateid.other[1];
3322 		stateidp->other[2] = openstp->ls_stateid.other[2];
3323 	}
3324 	NFSUNLOCKSTATE();
3325 	if (haslock) {
3326 		NFSLOCKV4ROOTMUTEX();
3327 		nfsv4_unlock(&nfsv4rootfs_lock, 1);
3328 		NFSUNLOCKV4ROOTMUTEX();
3329 	}
3330 	if (new_open)
3331 		free(new_open, M_NFSDSTATE);
3332 	if (new_deleg)
3333 		free(new_deleg, M_NFSDSTATE);
3334 
3335 	/*
3336 	 * If the NFSv4.1 client just acquired its first open, write a timestamp
3337 	 * to the stable storage file.
3338 	 */
3339 	if (gotstate != 0) {
3340 		nfsrv_writestable(clidp, len, NFSNST_NEWSTATE, p);
3341 		nfsrv_backupstable();
3342 	}
3343 
3344 out:
3345 	free(clidp, M_TEMP);
3346 	NFSEXITCODE2(error, nd);
3347 	return (error);
3348 }
3349 
3350 /*
3351  * Open update. Does the confirm, downgrade and close.
3352  */
3353 APPLESTATIC int
3354 nfsrv_openupdate(vnode_t vp, struct nfsstate *new_stp, nfsquad_t clientid,
3355     nfsv4stateid_t *stateidp, struct nfsrv_descript *nd, NFSPROC_T *p,
3356     int *retwriteaccessp)
3357 {
3358 	struct nfsstate *stp;
3359 	struct nfsclient *clp;
3360 	struct nfslockfile *lfp;
3361 	u_int32_t bits;
3362 	int error = 0, gotstate = 0, len = 0;
3363 	u_char *clidp = NULL;
3364 
3365 	/*
3366 	 * Check for restart conditions (client and server).
3367 	 */
3368 	error = nfsrv_checkrestart(clientid, new_stp->ls_flags,
3369 	    &new_stp->ls_stateid, 0);
3370 	if (error)
3371 		goto out;
3372 
3373 	clidp = malloc(NFSV4_OPAQUELIMIT, M_TEMP, M_WAITOK);
3374 	NFSLOCKSTATE();
3375 	/*
3376 	 * Get the open structure via clientid and stateid.
3377 	 */
3378 	error = nfsrv_getclient(clientid, CLOPS_RENEW, &clp, NULL,
3379 	    (nfsquad_t)((u_quad_t)0), 0, nd, p);
3380 	if (!error)
3381 		error = nfsrv_getstate(clp, &new_stp->ls_stateid,
3382 		    new_stp->ls_flags, &stp);
3383 
3384 	/*
3385 	 * Sanity check the open.
3386 	 */
3387 	if (!error && (!(stp->ls_flags & NFSLCK_OPEN) ||
3388 		(!(new_stp->ls_flags & NFSLCK_CONFIRM) &&
3389 		 (stp->ls_openowner->ls_flags & NFSLCK_NEEDSCONFIRM)) ||
3390 		((new_stp->ls_flags & NFSLCK_CONFIRM) &&
3391 		 (!(stp->ls_openowner->ls_flags & NFSLCK_NEEDSCONFIRM)))))
3392 		error = NFSERR_BADSTATEID;
3393 
3394 	if (!error)
3395 		error = nfsrv_checkseqid(nd, new_stp->ls_seq,
3396 		    stp->ls_openowner, new_stp->ls_op);
3397 	if (!error && stp->ls_stateid.seqid != new_stp->ls_stateid.seqid &&
3398 	    (((nd->nd_flag & ND_NFSV41) == 0 &&
3399 	      !(new_stp->ls_flags & NFSLCK_CONFIRM)) ||
3400 	     ((nd->nd_flag & ND_NFSV41) != 0 &&
3401 	      new_stp->ls_stateid.seqid != 0)))
3402 		error = NFSERR_OLDSTATEID;
3403 	if (!error && vnode_vtype(vp) != VREG) {
3404 		if (vnode_vtype(vp) == VDIR)
3405 			error = NFSERR_ISDIR;
3406 		else
3407 			error = NFSERR_INVAL;
3408 	}
3409 
3410 	if (error) {
3411 		/*
3412 		 * If a client tries to confirm an Open with a bad
3413 		 * seqid# and there are no byte range locks or other Opens
3414 		 * on the openowner, just throw it away, so the next use of the
3415 		 * openowner will start a fresh seq#.
3416 		 */
3417 		if (error == NFSERR_BADSEQID &&
3418 		    (new_stp->ls_flags & NFSLCK_CONFIRM) &&
3419 		    nfsrv_nootherstate(stp))
3420 			nfsrv_freeopenowner(stp->ls_openowner, 0, p);
3421 		NFSUNLOCKSTATE();
3422 		goto out;
3423 	}
3424 
3425 	/*
3426 	 * Set the return stateid.
3427 	 */
3428 	stateidp->seqid = stp->ls_stateid.seqid + 1;
3429 	if ((nd->nd_flag & ND_NFSV41) != 0 && stateidp->seqid == 0)
3430 		stateidp->seqid = 1;
3431 	stateidp->other[0] = stp->ls_stateid.other[0];
3432 	stateidp->other[1] = stp->ls_stateid.other[1];
3433 	stateidp->other[2] = stp->ls_stateid.other[2];
3434 	/*
3435 	 * Now, handle the three cases.
3436 	 */
3437 	if (new_stp->ls_flags & NFSLCK_CONFIRM) {
3438 		/*
3439 		 * If the open doesn't need confirmation, it seems to me that
3440 		 * there is a client error, but I'll just log it and keep going?
3441 		 */
3442 		if (!(stp->ls_openowner->ls_flags & NFSLCK_NEEDSCONFIRM))
3443 			printf("Nfsv4d: stray open confirm\n");
3444 		stp->ls_openowner->ls_flags = 0;
3445 		stp->ls_stateid.seqid++;
3446 		if ((nd->nd_flag & ND_NFSV41) != 0 &&
3447 		    stp->ls_stateid.seqid == 0)
3448 			stp->ls_stateid.seqid = 1;
3449 		if (!(clp->lc_flags & LCL_STAMPEDSTABLE)) {
3450 			clp->lc_flags |= LCL_STAMPEDSTABLE;
3451 			len = clp->lc_idlen;
3452 			NFSBCOPY(clp->lc_id, clidp, len);
3453 			gotstate = 1;
3454 		}
3455 		NFSUNLOCKSTATE();
3456 	} else if (new_stp->ls_flags & NFSLCK_CLOSE) {
3457 		lfp = stp->ls_lfp;
3458 		if (retwriteaccessp != NULL) {
3459 			if ((stp->ls_flags & NFSLCK_WRITEACCESS) != 0)
3460 				*retwriteaccessp = 1;
3461 			else
3462 				*retwriteaccessp = 0;
3463 		}
3464 		if (nfsrv_dolocallocks != 0 && !LIST_EMPTY(&stp->ls_open)) {
3465 			/* Get the lf lock */
3466 			nfsrv_locklf(lfp);
3467 			NFSUNLOCKSTATE();
3468 			ASSERT_VOP_ELOCKED(vp, "nfsrv_openupdate");
3469 			NFSVOPUNLOCK(vp, 0);
3470 			if (nfsrv_freeopen(stp, vp, 1, p) == 0) {
3471 				NFSLOCKSTATE();
3472 				nfsrv_unlocklf(lfp);
3473 				NFSUNLOCKSTATE();
3474 			}
3475 			NFSVOPLOCK(vp, LK_EXCLUSIVE | LK_RETRY);
3476 		} else {
3477 			(void) nfsrv_freeopen(stp, NULL, 0, p);
3478 			NFSUNLOCKSTATE();
3479 		}
3480 	} else {
3481 		/*
3482 		 * Update the share bits, making sure that the new set are a
3483 		 * subset of the old ones.
3484 		 */
3485 		bits = (new_stp->ls_flags & NFSLCK_SHAREBITS);
3486 		if (~(stp->ls_flags) & bits) {
3487 			NFSUNLOCKSTATE();
3488 			error = NFSERR_INVAL;
3489 			goto out;
3490 		}
3491 		stp->ls_flags = (bits | NFSLCK_OPEN);
3492 		stp->ls_stateid.seqid++;
3493 		if ((nd->nd_flag & ND_NFSV41) != 0 &&
3494 		    stp->ls_stateid.seqid == 0)
3495 			stp->ls_stateid.seqid = 1;
3496 		NFSUNLOCKSTATE();
3497 	}
3498 
3499 	/*
3500 	 * If the client just confirmed its first open, write a timestamp
3501 	 * to the stable storage file.
3502 	 */
3503 	if (gotstate != 0) {
3504 		nfsrv_writestable(clidp, len, NFSNST_NEWSTATE, p);
3505 		nfsrv_backupstable();
3506 	}
3507 
3508 out:
3509 	free(clidp, M_TEMP);
3510 	NFSEXITCODE2(error, nd);
3511 	return (error);
3512 }
3513 
3514 /*
3515  * Delegation update. Does the purge and return.
3516  */
3517 APPLESTATIC int
3518 nfsrv_delegupdate(struct nfsrv_descript *nd, nfsquad_t clientid,
3519     nfsv4stateid_t *stateidp, vnode_t vp, int op, struct ucred *cred,
3520     NFSPROC_T *p, int *retwriteaccessp)
3521 {
3522 	struct nfsstate *stp;
3523 	struct nfsclient *clp;
3524 	int error = 0;
3525 	fhandle_t fh;
3526 
3527 	/*
3528 	 * Do a sanity check against the file handle for DelegReturn.
3529 	 */
3530 	if (vp) {
3531 		error = nfsvno_getfh(vp, &fh, p);
3532 		if (error)
3533 			goto out;
3534 	}
3535 	/*
3536 	 * Check for restart conditions (client and server).
3537 	 */
3538 	if (op == NFSV4OP_DELEGRETURN)
3539 		error = nfsrv_checkrestart(clientid, NFSLCK_DELEGRETURN,
3540 			stateidp, 0);
3541 	else
3542 		error = nfsrv_checkrestart(clientid, NFSLCK_DELEGPURGE,
3543 			stateidp, 0);
3544 
3545 	NFSLOCKSTATE();
3546 	/*
3547 	 * Get the open structure via clientid and stateid.
3548 	 */
3549 	if (!error)
3550 	    error = nfsrv_getclient(clientid, CLOPS_RENEW, &clp, NULL,
3551 		(nfsquad_t)((u_quad_t)0), 0, nd, p);
3552 	if (error) {
3553 		if (error == NFSERR_CBPATHDOWN)
3554 			error = 0;
3555 		if (error == NFSERR_STALECLIENTID && op == NFSV4OP_DELEGRETURN)
3556 			error = NFSERR_STALESTATEID;
3557 	}
3558 	if (!error && op == NFSV4OP_DELEGRETURN) {
3559 	    error = nfsrv_getstate(clp, stateidp, NFSLCK_DELEGRETURN, &stp);
3560 	    if (!error && stp->ls_stateid.seqid != stateidp->seqid &&
3561 		((nd->nd_flag & ND_NFSV41) == 0 || stateidp->seqid != 0))
3562 		error = NFSERR_OLDSTATEID;
3563 	}
3564 	/*
3565 	 * NFSERR_EXPIRED means that the state has gone away,
3566 	 * so Delegations have been purged. Just return ok.
3567 	 */
3568 	if (error == NFSERR_EXPIRED && op == NFSV4OP_DELEGPURGE) {
3569 		NFSUNLOCKSTATE();
3570 		error = 0;
3571 		goto out;
3572 	}
3573 	if (error) {
3574 		NFSUNLOCKSTATE();
3575 		goto out;
3576 	}
3577 
3578 	if (op == NFSV4OP_DELEGRETURN) {
3579 		if (NFSBCMP((caddr_t)&fh, (caddr_t)&stp->ls_lfp->lf_fh,
3580 		    sizeof (fhandle_t))) {
3581 			NFSUNLOCKSTATE();
3582 			error = NFSERR_BADSTATEID;
3583 			goto out;
3584 		}
3585 		if (retwriteaccessp != NULL) {
3586 			if ((stp->ls_flags & NFSLCK_DELEGWRITE) != 0)
3587 				*retwriteaccessp = 1;
3588 			else
3589 				*retwriteaccessp = 0;
3590 		}
3591 		nfsrv_freedeleg(stp);
3592 	} else {
3593 		nfsrv_freedeleglist(&clp->lc_olddeleg);
3594 	}
3595 	NFSUNLOCKSTATE();
3596 	error = 0;
3597 
3598 out:
3599 	NFSEXITCODE(error);
3600 	return (error);
3601 }
3602 
3603 /*
3604  * Release lock owner.
3605  */
3606 APPLESTATIC int
3607 nfsrv_releaselckown(struct nfsstate *new_stp, nfsquad_t clientid,
3608     NFSPROC_T *p)
3609 {
3610 	struct nfsstate *stp, *nstp, *openstp, *ownstp;
3611 	struct nfsclient *clp;
3612 	int error = 0;
3613 
3614 	/*
3615 	 * Check for restart conditions (client and server).
3616 	 */
3617 	error = nfsrv_checkrestart(clientid, new_stp->ls_flags,
3618 	    &new_stp->ls_stateid, 0);
3619 	if (error)
3620 		goto out;
3621 
3622 	NFSLOCKSTATE();
3623 	/*
3624 	 * Get the lock owner by name.
3625 	 */
3626 	error = nfsrv_getclient(clientid, CLOPS_RENEW, &clp, NULL,
3627 	    (nfsquad_t)((u_quad_t)0), 0, NULL, p);
3628 	if (error) {
3629 		NFSUNLOCKSTATE();
3630 		goto out;
3631 	}
3632 	LIST_FOREACH(ownstp, &clp->lc_open, ls_list) {
3633 	    LIST_FOREACH(openstp, &ownstp->ls_open, ls_list) {
3634 		stp = LIST_FIRST(&openstp->ls_open);
3635 		while (stp != LIST_END(&openstp->ls_open)) {
3636 		    nstp = LIST_NEXT(stp, ls_list);
3637 		    /*
3638 		     * If the owner matches, check for locks and
3639 		     * then free or return an error.
3640 		     */
3641 		    if (stp->ls_ownerlen == new_stp->ls_ownerlen &&
3642 			!NFSBCMP(stp->ls_owner, new_stp->ls_owner,
3643 			 stp->ls_ownerlen)){
3644 			if (LIST_EMPTY(&stp->ls_lock)) {
3645 			    nfsrv_freelockowner(stp, NULL, 0, p);
3646 			} else {
3647 			    NFSUNLOCKSTATE();
3648 			    error = NFSERR_LOCKSHELD;
3649 			    goto out;
3650 			}
3651 		    }
3652 		    stp = nstp;
3653 		}
3654 	    }
3655 	}
3656 	NFSUNLOCKSTATE();
3657 
3658 out:
3659 	NFSEXITCODE(error);
3660 	return (error);
3661 }
3662 
3663 /*
3664  * Get the file handle for a lock structure.
3665  */
3666 static int
3667 nfsrv_getlockfh(vnode_t vp, u_short flags, struct nfslockfile *new_lfp,
3668     fhandle_t *nfhp, NFSPROC_T *p)
3669 {
3670 	fhandle_t *fhp = NULL;
3671 	int error;
3672 
3673 	/*
3674 	 * For lock, use the new nfslock structure, otherwise just
3675 	 * a fhandle_t on the stack.
3676 	 */
3677 	if (flags & NFSLCK_OPEN) {
3678 		KASSERT(new_lfp != NULL, ("nfsrv_getlockfh: new_lfp NULL"));
3679 		fhp = &new_lfp->lf_fh;
3680 	} else if (nfhp) {
3681 		fhp = nfhp;
3682 	} else {
3683 		panic("nfsrv_getlockfh");
3684 	}
3685 	error = nfsvno_getfh(vp, fhp, p);
3686 	NFSEXITCODE(error);
3687 	return (error);
3688 }
3689 
3690 /*
3691  * Get an nfs lock structure. Allocate one, as required, and return a
3692  * pointer to it.
3693  * Returns an NFSERR_xxx upon failure or -1 to indicate no current lock.
3694  */
3695 static int
3696 nfsrv_getlockfile(u_short flags, struct nfslockfile **new_lfpp,
3697     struct nfslockfile **lfpp, fhandle_t *nfhp, int lockit)
3698 {
3699 	struct nfslockfile *lfp;
3700 	fhandle_t *fhp = NULL, *tfhp;
3701 	struct nfslockhashhead *hp;
3702 	struct nfslockfile *new_lfp = NULL;
3703 
3704 	/*
3705 	 * For lock, use the new nfslock structure, otherwise just
3706 	 * a fhandle_t on the stack.
3707 	 */
3708 	if (flags & NFSLCK_OPEN) {
3709 		new_lfp = *new_lfpp;
3710 		fhp = &new_lfp->lf_fh;
3711 	} else if (nfhp) {
3712 		fhp = nfhp;
3713 	} else {
3714 		panic("nfsrv_getlockfile");
3715 	}
3716 
3717 	hp = NFSLOCKHASH(fhp);
3718 	LIST_FOREACH(lfp, hp, lf_hash) {
3719 		tfhp = &lfp->lf_fh;
3720 		if (NFSVNO_CMPFH(fhp, tfhp)) {
3721 			if (lockit)
3722 				nfsrv_locklf(lfp);
3723 			*lfpp = lfp;
3724 			return (0);
3725 		}
3726 	}
3727 	if (!(flags & NFSLCK_OPEN))
3728 		return (-1);
3729 
3730 	/*
3731 	 * No match, so chain the new one into the list.
3732 	 */
3733 	LIST_INIT(&new_lfp->lf_open);
3734 	LIST_INIT(&new_lfp->lf_lock);
3735 	LIST_INIT(&new_lfp->lf_deleg);
3736 	LIST_INIT(&new_lfp->lf_locallock);
3737 	LIST_INIT(&new_lfp->lf_rollback);
3738 	new_lfp->lf_locallock_lck.nfslock_usecnt = 0;
3739 	new_lfp->lf_locallock_lck.nfslock_lock = 0;
3740 	new_lfp->lf_usecount = 0;
3741 	LIST_INSERT_HEAD(hp, new_lfp, lf_hash);
3742 	*lfpp = new_lfp;
3743 	*new_lfpp = NULL;
3744 	return (0);
3745 }
3746 
3747 /*
3748  * This function adds a nfslock lock structure to the list for the associated
3749  * nfsstate and nfslockfile structures. It will be inserted after the
3750  * entry pointed at by insert_lop.
3751  */
3752 static void
3753 nfsrv_insertlock(struct nfslock *new_lop, struct nfslock *insert_lop,
3754     struct nfsstate *stp, struct nfslockfile *lfp)
3755 {
3756 	struct nfslock *lop, *nlop;
3757 
3758 	new_lop->lo_stp = stp;
3759 	new_lop->lo_lfp = lfp;
3760 
3761 	if (stp != NULL) {
3762 		/* Insert in increasing lo_first order */
3763 		lop = LIST_FIRST(&lfp->lf_lock);
3764 		if (lop == LIST_END(&lfp->lf_lock) ||
3765 		    new_lop->lo_first <= lop->lo_first) {
3766 			LIST_INSERT_HEAD(&lfp->lf_lock, new_lop, lo_lckfile);
3767 		} else {
3768 			nlop = LIST_NEXT(lop, lo_lckfile);
3769 			while (nlop != LIST_END(&lfp->lf_lock) &&
3770 			       nlop->lo_first < new_lop->lo_first) {
3771 				lop = nlop;
3772 				nlop = LIST_NEXT(lop, lo_lckfile);
3773 			}
3774 			LIST_INSERT_AFTER(lop, new_lop, lo_lckfile);
3775 		}
3776 	} else {
3777 		new_lop->lo_lckfile.le_prev = NULL;	/* list not used */
3778 	}
3779 
3780 	/*
3781 	 * Insert after insert_lop, which is overloaded as stp or lfp for
3782 	 * an empty list.
3783 	 */
3784 	if (stp == NULL && (struct nfslockfile *)insert_lop == lfp)
3785 		LIST_INSERT_HEAD(&lfp->lf_locallock, new_lop, lo_lckowner);
3786 	else if ((struct nfsstate *)insert_lop == stp)
3787 		LIST_INSERT_HEAD(&stp->ls_lock, new_lop, lo_lckowner);
3788 	else
3789 		LIST_INSERT_AFTER(insert_lop, new_lop, lo_lckowner);
3790 	if (stp != NULL) {
3791 		nfsstatsv1.srvlocks++;
3792 		nfsrv_openpluslock++;
3793 	}
3794 }
3795 
3796 /*
3797  * This function updates the locking for a lock owner and given file. It
3798  * maintains a list of lock ranges ordered on increasing file offset that
3799  * are NFSLCK_READ or NFSLCK_WRITE and non-overlapping (aka POSIX style).
3800  * It always adds new_lop to the list and sometimes uses the one pointed
3801  * at by other_lopp.
3802  */
3803 static void
3804 nfsrv_updatelock(struct nfsstate *stp, struct nfslock **new_lopp,
3805     struct nfslock **other_lopp, struct nfslockfile *lfp)
3806 {
3807 	struct nfslock *new_lop = *new_lopp;
3808 	struct nfslock *lop, *tlop, *ilop;
3809 	struct nfslock *other_lop = *other_lopp;
3810 	int unlock = 0, myfile = 0;
3811 	u_int64_t tmp;
3812 
3813 	/*
3814 	 * Work down the list until the lock is merged.
3815 	 */
3816 	if (new_lop->lo_flags & NFSLCK_UNLOCK)
3817 		unlock = 1;
3818 	if (stp != NULL) {
3819 		ilop = (struct nfslock *)stp;
3820 		lop = LIST_FIRST(&stp->ls_lock);
3821 	} else {
3822 		ilop = (struct nfslock *)lfp;
3823 		lop = LIST_FIRST(&lfp->lf_locallock);
3824 	}
3825 	while (lop != NULL) {
3826 	    /*
3827 	     * Only check locks for this file that aren't before the start of
3828 	     * new lock's range.
3829 	     */
3830 	    if (lop->lo_lfp == lfp) {
3831 	      myfile = 1;
3832 	      if (lop->lo_end >= new_lop->lo_first) {
3833 		if (new_lop->lo_end < lop->lo_first) {
3834 			/*
3835 			 * If the new lock ends before the start of the
3836 			 * current lock's range, no merge, just insert
3837 			 * the new lock.
3838 			 */
3839 			break;
3840 		}
3841 		if (new_lop->lo_flags == lop->lo_flags ||
3842 		    (new_lop->lo_first <= lop->lo_first &&
3843 		     new_lop->lo_end >= lop->lo_end)) {
3844 			/*
3845 			 * This lock can be absorbed by the new lock/unlock.
3846 			 * This happens when it covers the entire range
3847 			 * of the old lock or is contiguous
3848 			 * with the old lock and is of the same type or an
3849 			 * unlock.
3850 			 */
3851 			if (lop->lo_first < new_lop->lo_first)
3852 				new_lop->lo_first = lop->lo_first;
3853 			if (lop->lo_end > new_lop->lo_end)
3854 				new_lop->lo_end = lop->lo_end;
3855 			tlop = lop;
3856 			lop = LIST_NEXT(lop, lo_lckowner);
3857 			nfsrv_freenfslock(tlop);
3858 			continue;
3859 		}
3860 
3861 		/*
3862 		 * All these cases are for contiguous locks that are not the
3863 		 * same type, so they can't be merged.
3864 		 */
3865 		if (new_lop->lo_first <= lop->lo_first) {
3866 			/*
3867 			 * This case is where the new lock overlaps with the
3868 			 * first part of the old lock. Move the start of the
3869 			 * old lock to just past the end of the new lock. The
3870 			 * new lock will be inserted in front of the old, since
3871 			 * ilop hasn't been updated. (We are done now.)
3872 			 */
3873 			lop->lo_first = new_lop->lo_end;
3874 			break;
3875 		}
3876 		if (new_lop->lo_end >= lop->lo_end) {
3877 			/*
3878 			 * This case is where the new lock overlaps with the
3879 			 * end of the old lock's range. Move the old lock's
3880 			 * end to just before the new lock's first and insert
3881 			 * the new lock after the old lock.
3882 			 * Might not be done yet, since the new lock could
3883 			 * overlap further locks with higher ranges.
3884 			 */
3885 			lop->lo_end = new_lop->lo_first;
3886 			ilop = lop;
3887 			lop = LIST_NEXT(lop, lo_lckowner);
3888 			continue;
3889 		}
3890 		/*
3891 		 * The final case is where the new lock's range is in the
3892 		 * middle of the current lock's and splits the current lock
3893 		 * up. Use *other_lopp to handle the second part of the
3894 		 * split old lock range. (We are done now.)
3895 		 * For unlock, we use new_lop as other_lop and tmp, since
3896 		 * other_lop and new_lop are the same for this case.
3897 		 * We noted the unlock case above, so we don't need
3898 		 * new_lop->lo_flags any longer.
3899 		 */
3900 		tmp = new_lop->lo_first;
3901 		if (other_lop == NULL) {
3902 			if (!unlock)
3903 				panic("nfsd srv update unlock");
3904 			other_lop = new_lop;
3905 			*new_lopp = NULL;
3906 		}
3907 		other_lop->lo_first = new_lop->lo_end;
3908 		other_lop->lo_end = lop->lo_end;
3909 		other_lop->lo_flags = lop->lo_flags;
3910 		other_lop->lo_stp = stp;
3911 		other_lop->lo_lfp = lfp;
3912 		lop->lo_end = tmp;
3913 		nfsrv_insertlock(other_lop, lop, stp, lfp);
3914 		*other_lopp = NULL;
3915 		ilop = lop;
3916 		break;
3917 	      }
3918 	    }
3919 	    ilop = lop;
3920 	    lop = LIST_NEXT(lop, lo_lckowner);
3921 	    if (myfile && (lop == NULL || lop->lo_lfp != lfp))
3922 		break;
3923 	}
3924 
3925 	/*
3926 	 * Insert the new lock in the list at the appropriate place.
3927 	 */
3928 	if (!unlock) {
3929 		nfsrv_insertlock(new_lop, ilop, stp, lfp);
3930 		*new_lopp = NULL;
3931 	}
3932 }
3933 
3934 /*
3935  * This function handles sequencing of locks, etc.
3936  * It returns an error that indicates what the caller should do.
3937  */
3938 static int
3939 nfsrv_checkseqid(struct nfsrv_descript *nd, u_int32_t seqid,
3940     struct nfsstate *stp, struct nfsrvcache *op)
3941 {
3942 	int error = 0;
3943 
3944 	if ((nd->nd_flag & ND_NFSV41) != 0)
3945 		/* NFSv4.1 ignores the open_seqid and lock_seqid. */
3946 		goto out;
3947 	if (op != nd->nd_rp)
3948 		panic("nfsrvstate checkseqid");
3949 	if (!(op->rc_flag & RC_INPROG))
3950 		panic("nfsrvstate not inprog");
3951 	if (stp->ls_op && stp->ls_op->rc_refcnt <= 0) {
3952 		printf("refcnt=%d\n", stp->ls_op->rc_refcnt);
3953 		panic("nfsrvstate op refcnt");
3954 	}
3955 	if ((stp->ls_seq + 1) == seqid) {
3956 		if (stp->ls_op)
3957 			nfsrvd_derefcache(stp->ls_op);
3958 		stp->ls_op = op;
3959 		nfsrvd_refcache(op);
3960 		stp->ls_seq = seqid;
3961 		goto out;
3962 	} else if (stp->ls_seq == seqid && stp->ls_op &&
3963 		op->rc_xid == stp->ls_op->rc_xid &&
3964 		op->rc_refcnt == 0 &&
3965 		op->rc_reqlen == stp->ls_op->rc_reqlen &&
3966 		op->rc_cksum == stp->ls_op->rc_cksum) {
3967 		if (stp->ls_op->rc_flag & RC_INPROG) {
3968 			error = NFSERR_DONTREPLY;
3969 			goto out;
3970 		}
3971 		nd->nd_rp = stp->ls_op;
3972 		nd->nd_rp->rc_flag |= RC_INPROG;
3973 		nfsrvd_delcache(op);
3974 		error = NFSERR_REPLYFROMCACHE;
3975 		goto out;
3976 	}
3977 	error = NFSERR_BADSEQID;
3978 
3979 out:
3980 	NFSEXITCODE2(error, nd);
3981 	return (error);
3982 }
3983 
3984 /*
3985  * Get the client ip address for callbacks. If the strings can't be parsed,
3986  * just set lc_program to 0 to indicate no callbacks are possible.
3987  * (For cases where the address can't be parsed or is 0.0.0.0.0.0, set
3988  *  the address to the client's transport address. This won't be used
3989  *  for callbacks, but can be printed out by nfsstats for info.)
3990  * Return error if the xdr can't be parsed, 0 otherwise.
3991  */
3992 APPLESTATIC int
3993 nfsrv_getclientipaddr(struct nfsrv_descript *nd, struct nfsclient *clp)
3994 {
3995 	u_int32_t *tl;
3996 	u_char *cp, *cp2;
3997 	int i, j;
3998 	struct sockaddr_in *rad, *sad;
3999 	u_char protocol[5], addr[24];
4000 	int error = 0, cantparse = 0;
4001 	union {
4002 		in_addr_t ival;
4003 		u_char cval[4];
4004 	} ip;
4005 	union {
4006 		in_port_t sval;
4007 		u_char cval[2];
4008 	} port;
4009 
4010 	rad = NFSSOCKADDR(clp->lc_req.nr_nam, struct sockaddr_in *);
4011 	rad->sin_family = AF_INET;
4012 	rad->sin_len = sizeof (struct sockaddr_in);
4013 	rad->sin_addr.s_addr = 0;
4014 	rad->sin_port = 0;
4015 	clp->lc_req.nr_client = NULL;
4016 	clp->lc_req.nr_lock = 0;
4017 	NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
4018 	i = fxdr_unsigned(int, *tl);
4019 	if (i >= 3 && i <= 4) {
4020 		error = nfsrv_mtostr(nd, protocol, i);
4021 		if (error)
4022 			goto nfsmout;
4023 		if (!strcmp(protocol, "tcp")) {
4024 			clp->lc_flags |= LCL_TCPCALLBACK;
4025 			clp->lc_req.nr_sotype = SOCK_STREAM;
4026 			clp->lc_req.nr_soproto = IPPROTO_TCP;
4027 		} else if (!strcmp(protocol, "udp")) {
4028 			clp->lc_req.nr_sotype = SOCK_DGRAM;
4029 			clp->lc_req.nr_soproto = IPPROTO_UDP;
4030 		} else {
4031 			cantparse = 1;
4032 		}
4033 	} else {
4034 		cantparse = 1;
4035 		if (i > 0) {
4036 			error = nfsm_advance(nd, NFSM_RNDUP(i), -1);
4037 			if (error)
4038 				goto nfsmout;
4039 		}
4040 	}
4041 	NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
4042 	i = fxdr_unsigned(int, *tl);
4043 	if (i < 0) {
4044 		error = NFSERR_BADXDR;
4045 		goto nfsmout;
4046 	} else if (i == 0) {
4047 		cantparse = 1;
4048 	} else if (!cantparse && i <= 23 && i >= 11) {
4049 		error = nfsrv_mtostr(nd, addr, i);
4050 		if (error)
4051 			goto nfsmout;
4052 
4053 		/*
4054 		 * Parse out the address fields. We expect 6 decimal numbers
4055 		 * separated by '.'s.
4056 		 */
4057 		cp = addr;
4058 		i = 0;
4059 		while (*cp && i < 6) {
4060 			cp2 = cp;
4061 			while (*cp2 && *cp2 != '.')
4062 				cp2++;
4063 			if (*cp2)
4064 				*cp2++ = '\0';
4065 			else if (i != 5) {
4066 				cantparse = 1;
4067 				break;
4068 			}
4069 			j = nfsrv_getipnumber(cp);
4070 			if (j >= 0) {
4071 				if (i < 4)
4072 					ip.cval[3 - i] = j;
4073 				else
4074 					port.cval[5 - i] = j;
4075 			} else {
4076 				cantparse = 1;
4077 				break;
4078 			}
4079 			cp = cp2;
4080 			i++;
4081 		}
4082 		if (!cantparse) {
4083 			if (ip.ival != 0x0) {
4084 				rad->sin_addr.s_addr = htonl(ip.ival);
4085 				rad->sin_port = htons(port.sval);
4086 			} else {
4087 				cantparse = 1;
4088 			}
4089 		}
4090 	} else {
4091 		cantparse = 1;
4092 		if (i > 0) {
4093 			error = nfsm_advance(nd, NFSM_RNDUP(i), -1);
4094 			if (error)
4095 				goto nfsmout;
4096 		}
4097 	}
4098 	if (cantparse) {
4099 		sad = NFSSOCKADDR(nd->nd_nam, struct sockaddr_in *);
4100 		if (sad->sin_family == AF_INET) {
4101 			rad->sin_addr.s_addr = sad->sin_addr.s_addr;
4102 			rad->sin_port = 0x0;
4103 		}
4104 		clp->lc_program = 0;
4105 	}
4106 nfsmout:
4107 	NFSEXITCODE2(error, nd);
4108 	return (error);
4109 }
4110 
4111 /*
4112  * Turn a string of up to three decimal digits into a number. Return -1 upon
4113  * error.
4114  */
4115 static int
4116 nfsrv_getipnumber(u_char *cp)
4117 {
4118 	int i = 0, j = 0;
4119 
4120 	while (*cp) {
4121 		if (j > 2 || *cp < '0' || *cp > '9')
4122 			return (-1);
4123 		i *= 10;
4124 		i += (*cp - '0');
4125 		cp++;
4126 		j++;
4127 	}
4128 	if (i < 256)
4129 		return (i);
4130 	return (-1);
4131 }
4132 
4133 /*
4134  * This function checks for restart conditions.
4135  */
4136 static int
4137 nfsrv_checkrestart(nfsquad_t clientid, u_int32_t flags,
4138     nfsv4stateid_t *stateidp, int specialid)
4139 {
4140 	int ret = 0;
4141 
4142 	/*
4143 	 * First check for a server restart. Open, LockT, ReleaseLockOwner
4144 	 * and DelegPurge have a clientid, the rest a stateid.
4145 	 */
4146 	if (flags &
4147 	    (NFSLCK_OPEN | NFSLCK_TEST | NFSLCK_RELEASE | NFSLCK_DELEGPURGE)) {
4148 		if (clientid.lval[0] != nfsrvboottime) {
4149 			ret = NFSERR_STALECLIENTID;
4150 			goto out;
4151 		}
4152 	} else if (stateidp->other[0] != nfsrvboottime &&
4153 		specialid == 0) {
4154 		ret = NFSERR_STALESTATEID;
4155 		goto out;
4156 	}
4157 
4158 	/*
4159 	 * Read, Write, Setattr and LockT can return NFSERR_GRACE and do
4160 	 * not use a lock/open owner seqid#, so the check can be done now.
4161 	 * (The others will be checked, as required, later.)
4162 	 */
4163 	if (!(flags & (NFSLCK_CHECK | NFSLCK_TEST)))
4164 		goto out;
4165 
4166 	NFSLOCKSTATE();
4167 	ret = nfsrv_checkgrace(NULL, NULL, flags);
4168 	NFSUNLOCKSTATE();
4169 
4170 out:
4171 	NFSEXITCODE(ret);
4172 	return (ret);
4173 }
4174 
4175 /*
4176  * Check for grace.
4177  */
4178 static int
4179 nfsrv_checkgrace(struct nfsrv_descript *nd, struct nfsclient *clp,
4180     u_int32_t flags)
4181 {
4182 	int error = 0, notreclaimed;
4183 	struct nfsrv_stable *sp;
4184 
4185 	if ((nfsrv_stablefirst.nsf_flags & (NFSNSF_UPDATEDONE |
4186 	     NFSNSF_GRACEOVER)) == 0) {
4187 		/*
4188 		 * First, check to see if all of the clients have done a
4189 		 * ReclaimComplete.  If so, grace can end now.
4190 		 */
4191 		notreclaimed = 0;
4192 		LIST_FOREACH(sp, &nfsrv_stablefirst.nsf_head, nst_list) {
4193 			if ((sp->nst_flag & NFSNST_RECLAIMED) == 0) {
4194 				notreclaimed = 1;
4195 				break;
4196 			}
4197 		}
4198 		if (notreclaimed == 0)
4199 			nfsrv_stablefirst.nsf_flags |= (NFSNSF_GRACEOVER |
4200 			    NFSNSF_NEEDLOCK);
4201 	}
4202 
4203 	if ((nfsrv_stablefirst.nsf_flags & NFSNSF_GRACEOVER) != 0) {
4204 		if (flags & NFSLCK_RECLAIM) {
4205 			error = NFSERR_NOGRACE;
4206 			goto out;
4207 		}
4208 	} else {
4209 		if (!(flags & NFSLCK_RECLAIM)) {
4210 			error = NFSERR_GRACE;
4211 			goto out;
4212 		}
4213 		if (nd != NULL && clp != NULL &&
4214 		    (nd->nd_flag & ND_NFSV41) != 0 &&
4215 		    (clp->lc_flags & LCL_RECLAIMCOMPLETE) != 0) {
4216 			error = NFSERR_NOGRACE;
4217 			goto out;
4218 		}
4219 
4220 		/*
4221 		 * If grace is almost over and we are still getting Reclaims,
4222 		 * extend grace a bit.
4223 		 */
4224 		if ((NFSD_MONOSEC + NFSRV_LEASEDELTA) >
4225 		    nfsrv_stablefirst.nsf_eograce)
4226 			nfsrv_stablefirst.nsf_eograce = NFSD_MONOSEC +
4227 				NFSRV_LEASEDELTA;
4228 	}
4229 
4230 out:
4231 	NFSEXITCODE(error);
4232 	return (error);
4233 }
4234 
4235 /*
4236  * Do a server callback.
4237  * The "trunc" argument is slightly overloaded and refers to different
4238  * boolean arguments for CBRECALL and CBLAYOUTRECALL.
4239  */
4240 static int
4241 nfsrv_docallback(struct nfsclient *clp, int procnum, nfsv4stateid_t *stateidp,
4242     int trunc, fhandle_t *fhp, struct nfsvattr *nap, nfsattrbit_t *attrbitp,
4243     int laytype, NFSPROC_T *p)
4244 {
4245 	mbuf_t m;
4246 	u_int32_t *tl;
4247 	struct nfsrv_descript *nd;
4248 	struct ucred *cred;
4249 	int error = 0;
4250 	u_int32_t callback;
4251 	struct nfsdsession *sep = NULL;
4252 	uint64_t tval;
4253 
4254 	nd = malloc(sizeof(*nd), M_TEMP, M_WAITOK | M_ZERO);
4255 	cred = newnfs_getcred();
4256 	NFSLOCKSTATE();	/* mostly for lc_cbref++ */
4257 	if (clp->lc_flags & LCL_NEEDSCONFIRM) {
4258 		NFSUNLOCKSTATE();
4259 		panic("docallb");
4260 	}
4261 	clp->lc_cbref++;
4262 
4263 	/*
4264 	 * Fill the callback program# and version into the request
4265 	 * structure for newnfs_connect() to use.
4266 	 */
4267 	clp->lc_req.nr_prog = clp->lc_program;
4268 #ifdef notnow
4269 	if ((clp->lc_flags & LCL_NFSV41) != 0)
4270 		clp->lc_req.nr_vers = NFSV41_CBVERS;
4271 	else
4272 #endif
4273 		clp->lc_req.nr_vers = NFSV4_CBVERS;
4274 
4275 	/*
4276 	 * First, fill in some of the fields of nd and cr.
4277 	 */
4278 	nd->nd_flag = ND_NFSV4;
4279 	if (clp->lc_flags & LCL_GSS)
4280 		nd->nd_flag |= ND_KERBV;
4281 	if ((clp->lc_flags & LCL_NFSV41) != 0)
4282 		nd->nd_flag |= ND_NFSV41;
4283 	nd->nd_repstat = 0;
4284 	cred->cr_uid = clp->lc_uid;
4285 	cred->cr_gid = clp->lc_gid;
4286 	callback = clp->lc_callback;
4287 	NFSUNLOCKSTATE();
4288 	cred->cr_ngroups = 1;
4289 
4290 	/*
4291 	 * Get the first mbuf for the request.
4292 	 */
4293 	MGET(m, M_WAITOK, MT_DATA);
4294 	mbuf_setlen(m, 0);
4295 	nd->nd_mreq = nd->nd_mb = m;
4296 	nd->nd_bpos = NFSMTOD(m, caddr_t);
4297 
4298 	/*
4299 	 * and build the callback request.
4300 	 */
4301 	if (procnum == NFSV4OP_CBGETATTR) {
4302 		nd->nd_procnum = NFSV4PROC_CBCOMPOUND;
4303 		error = nfsrv_cbcallargs(nd, clp, callback, NFSV4OP_CBGETATTR,
4304 		    "CB Getattr", &sep);
4305 		if (error != 0) {
4306 			mbuf_freem(nd->nd_mreq);
4307 			goto errout;
4308 		}
4309 		(void)nfsm_fhtom(nd, (u_int8_t *)fhp, NFSX_MYFH, 0);
4310 		(void)nfsrv_putattrbit(nd, attrbitp);
4311 	} else if (procnum == NFSV4OP_CBRECALL) {
4312 		nd->nd_procnum = NFSV4PROC_CBCOMPOUND;
4313 		error = nfsrv_cbcallargs(nd, clp, callback, NFSV4OP_CBRECALL,
4314 		    "CB Recall", &sep);
4315 		if (error != 0) {
4316 			mbuf_freem(nd->nd_mreq);
4317 			goto errout;
4318 		}
4319 		NFSM_BUILD(tl, u_int32_t *, NFSX_UNSIGNED + NFSX_STATEID);
4320 		*tl++ = txdr_unsigned(stateidp->seqid);
4321 		NFSBCOPY((caddr_t)stateidp->other, (caddr_t)tl,
4322 		    NFSX_STATEIDOTHER);
4323 		tl += (NFSX_STATEIDOTHER / NFSX_UNSIGNED);
4324 		if (trunc)
4325 			*tl = newnfs_true;
4326 		else
4327 			*tl = newnfs_false;
4328 		(void)nfsm_fhtom(nd, (u_int8_t *)fhp, NFSX_MYFH, 0);
4329 	} else if (procnum == NFSV4OP_CBLAYOUTRECALL) {
4330 		NFSD_DEBUG(4, "docallback layout recall\n");
4331 		nd->nd_procnum = NFSV4PROC_CBCOMPOUND;
4332 		error = nfsrv_cbcallargs(nd, clp, callback,
4333 		    NFSV4OP_CBLAYOUTRECALL, "CB Reclayout", &sep);
4334 		NFSD_DEBUG(4, "aft cbcallargs=%d\n", error);
4335 		if (error != 0) {
4336 			mbuf_freem(nd->nd_mreq);
4337 			goto errout;
4338 		}
4339 		NFSM_BUILD(tl, u_int32_t *, 4 * NFSX_UNSIGNED);
4340 		*tl++ = txdr_unsigned(laytype);
4341 		*tl++ = txdr_unsigned(NFSLAYOUTIOMODE_ANY);
4342 		if (trunc)
4343 			*tl++ = newnfs_true;
4344 		else
4345 			*tl++ = newnfs_false;
4346 		*tl = txdr_unsigned(NFSV4LAYOUTRET_FILE);
4347 		nfsm_fhtom(nd, (uint8_t *)fhp, NFSX_MYFH, 0);
4348 		NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_HYPER + NFSX_STATEID);
4349 		tval = 0;
4350 		txdr_hyper(tval, tl); tl += 2;
4351 		tval = UINT64_MAX;
4352 		txdr_hyper(tval, tl); tl += 2;
4353 		*tl++ = txdr_unsigned(stateidp->seqid);
4354 		NFSBCOPY(stateidp->other, tl, NFSX_STATEIDOTHER);
4355 		tl += (NFSX_STATEIDOTHER / NFSX_UNSIGNED);
4356 		NFSD_DEBUG(4, "aft args\n");
4357 	} else if (procnum == NFSV4PROC_CBNULL) {
4358 		nd->nd_procnum = NFSV4PROC_CBNULL;
4359 		if ((clp->lc_flags & LCL_NFSV41) != 0) {
4360 			error = nfsv4_getcbsession(clp, &sep);
4361 			if (error != 0) {
4362 				mbuf_freem(nd->nd_mreq);
4363 				goto errout;
4364 			}
4365 		}
4366 	} else {
4367 		error = NFSERR_SERVERFAULT;
4368 		mbuf_freem(nd->nd_mreq);
4369 		goto errout;
4370 	}
4371 
4372 	/*
4373 	 * Call newnfs_connect(), as required, and then newnfs_request().
4374 	 */
4375 	(void) newnfs_sndlock(&clp->lc_req.nr_lock);
4376 	if (clp->lc_req.nr_client == NULL) {
4377 		if ((clp->lc_flags & LCL_NFSV41) != 0) {
4378 			error = ECONNREFUSED;
4379 			nfsrv_freesession(sep, NULL);
4380 		} else if (nd->nd_procnum == NFSV4PROC_CBNULL)
4381 			error = newnfs_connect(NULL, &clp->lc_req, cred,
4382 			    NULL, 1);
4383 		else
4384 			error = newnfs_connect(NULL, &clp->lc_req, cred,
4385 			    NULL, 3);
4386 	}
4387 	newnfs_sndunlock(&clp->lc_req.nr_lock);
4388 	NFSD_DEBUG(4, "aft sndunlock=%d\n", error);
4389 	if (!error) {
4390 		if ((nd->nd_flag & ND_NFSV41) != 0) {
4391 			KASSERT(sep != NULL, ("sep NULL"));
4392 			if (sep->sess_cbsess.nfsess_xprt != NULL)
4393 				error = newnfs_request(nd, NULL, clp,
4394 				    &clp->lc_req, NULL, NULL, cred,
4395 				    clp->lc_program, clp->lc_req.nr_vers, NULL,
4396 				    1, NULL, &sep->sess_cbsess);
4397 			else {
4398 				/*
4399 				 * This should probably never occur, but if a
4400 				 * client somehow does an RPC without a
4401 				 * SequenceID Op that causes a callback just
4402 				 * after the nfsd threads have been terminated
4403 				 * and restared we could conceivably get here
4404 				 * without a backchannel xprt.
4405 				 */
4406 				printf("nfsrv_docallback: no xprt\n");
4407 				error = ECONNREFUSED;
4408 			}
4409 			NFSD_DEBUG(4, "aft newnfs_request=%d\n", error);
4410 			nfsrv_freesession(sep, NULL);
4411 		} else
4412 			error = newnfs_request(nd, NULL, clp, &clp->lc_req,
4413 			    NULL, NULL, cred, clp->lc_program,
4414 			    clp->lc_req.nr_vers, NULL, 1, NULL, NULL);
4415 	}
4416 errout:
4417 	NFSFREECRED(cred);
4418 
4419 	/*
4420 	 * If error is set here, the Callback path isn't working
4421 	 * properly, so twiddle the appropriate LCL_ flags.
4422 	 * (nd_repstat != 0 indicates the Callback path is working,
4423 	 *  but the callback failed on the client.)
4424 	 */
4425 	if (error) {
4426 		/*
4427 		 * Mark the callback pathway down, which disabled issuing
4428 		 * of delegations and gets Renew to return NFSERR_CBPATHDOWN.
4429 		 */
4430 		NFSLOCKSTATE();
4431 		clp->lc_flags |= LCL_CBDOWN;
4432 		NFSUNLOCKSTATE();
4433 	} else {
4434 		/*
4435 		 * Callback worked. If the callback path was down, disable
4436 		 * callbacks, so no more delegations will be issued. (This
4437 		 * is done on the assumption that the callback pathway is
4438 		 * flakey.)
4439 		 */
4440 		NFSLOCKSTATE();
4441 		if (clp->lc_flags & LCL_CBDOWN)
4442 			clp->lc_flags &= ~(LCL_CBDOWN | LCL_CALLBACKSON);
4443 		NFSUNLOCKSTATE();
4444 		if (nd->nd_repstat) {
4445 			error = nd->nd_repstat;
4446 			NFSD_DEBUG(1, "nfsrv_docallback op=%d err=%d\n",
4447 			    procnum, error);
4448 		} else if (error == 0 && procnum == NFSV4OP_CBGETATTR)
4449 			error = nfsv4_loadattr(nd, NULL, nap, NULL, NULL, 0,
4450 			    NULL, NULL, NULL, NULL, NULL, 0, NULL, NULL, NULL,
4451 			    p, NULL);
4452 		mbuf_freem(nd->nd_mrep);
4453 	}
4454 	NFSLOCKSTATE();
4455 	clp->lc_cbref--;
4456 	if ((clp->lc_flags & LCL_WAKEUPWANTED) && clp->lc_cbref == 0) {
4457 		clp->lc_flags &= ~LCL_WAKEUPWANTED;
4458 		wakeup(clp);
4459 	}
4460 	NFSUNLOCKSTATE();
4461 
4462 	free(nd, M_TEMP);
4463 	NFSEXITCODE(error);
4464 	return (error);
4465 }
4466 
4467 /*
4468  * Set up the compound RPC for the callback.
4469  */
4470 static int
4471 nfsrv_cbcallargs(struct nfsrv_descript *nd, struct nfsclient *clp,
4472     uint32_t callback, int op, const char *optag, struct nfsdsession **sepp)
4473 {
4474 	uint32_t *tl;
4475 	int error, len;
4476 
4477 	len = strlen(optag);
4478 	(void)nfsm_strtom(nd, optag, len);
4479 	NFSM_BUILD(tl, uint32_t *, 4 * NFSX_UNSIGNED);
4480 	if ((nd->nd_flag & ND_NFSV41) != 0) {
4481 		*tl++ = txdr_unsigned(NFSV41_MINORVERSION);
4482 		*tl++ = txdr_unsigned(callback);
4483 		*tl++ = txdr_unsigned(2);
4484 		*tl = txdr_unsigned(NFSV4OP_CBSEQUENCE);
4485 		error = nfsv4_setcbsequence(nd, clp, 1, sepp);
4486 		if (error != 0)
4487 			return (error);
4488 		NFSM_BUILD(tl, u_int32_t *, NFSX_UNSIGNED);
4489 		*tl = txdr_unsigned(op);
4490 	} else {
4491 		*tl++ = txdr_unsigned(NFSV4_MINORVERSION);
4492 		*tl++ = txdr_unsigned(callback);
4493 		*tl++ = txdr_unsigned(1);
4494 		*tl = txdr_unsigned(op);
4495 	}
4496 	return (0);
4497 }
4498 
4499 /*
4500  * Return the next index# for a clientid. Mostly just increment and return
4501  * the next one, but... if the 32bit unsigned does actually wrap around,
4502  * it should be rebooted.
4503  * At an average rate of one new client per second, it will wrap around in
4504  * approximately 136 years. (I think the server will have been shut
4505  * down or rebooted before then.)
4506  */
4507 static u_int32_t
4508 nfsrv_nextclientindex(void)
4509 {
4510 	static u_int32_t client_index = 0;
4511 
4512 	client_index++;
4513 	if (client_index != 0)
4514 		return (client_index);
4515 
4516 	printf("%s: out of clientids\n", __func__);
4517 	return (client_index);
4518 }
4519 
4520 /*
4521  * Return the next index# for a stateid. Mostly just increment and return
4522  * the next one, but... if the 32bit unsigned does actually wrap around
4523  * (will a BSD server stay up that long?), find
4524  * new start and end values.
4525  */
4526 static u_int32_t
4527 nfsrv_nextstateindex(struct nfsclient *clp)
4528 {
4529 	struct nfsstate *stp;
4530 	int i;
4531 	u_int32_t canuse, min_index, max_index;
4532 
4533 	if (!(clp->lc_flags & LCL_INDEXNOTOK)) {
4534 		clp->lc_stateindex++;
4535 		if (clp->lc_stateindex != clp->lc_statemaxindex)
4536 			return (clp->lc_stateindex);
4537 	}
4538 
4539 	/*
4540 	 * Yuck, we've hit the end.
4541 	 * Look for a new min and max.
4542 	 */
4543 	min_index = 0;
4544 	max_index = 0xffffffff;
4545 	for (i = 0; i < nfsrv_statehashsize; i++) {
4546 	    LIST_FOREACH(stp, &clp->lc_stateid[i], ls_hash) {
4547 		if (stp->ls_stateid.other[2] > 0x80000000) {
4548 		    if (stp->ls_stateid.other[2] < max_index)
4549 			max_index = stp->ls_stateid.other[2];
4550 		} else {
4551 		    if (stp->ls_stateid.other[2] > min_index)
4552 			min_index = stp->ls_stateid.other[2];
4553 		}
4554 	    }
4555 	}
4556 
4557 	/*
4558 	 * Yikes, highly unlikely, but I'll handle it anyhow.
4559 	 */
4560 	if (min_index == 0x80000000 && max_index == 0x80000001) {
4561 	    canuse = 0;
4562 	    /*
4563 	     * Loop around until we find an unused entry. Return that
4564 	     * and set LCL_INDEXNOTOK, so the search will continue next time.
4565 	     * (This is one of those rare cases where a goto is the
4566 	     *  cleanest way to code the loop.)
4567 	     */
4568 tryagain:
4569 	    for (i = 0; i < nfsrv_statehashsize; i++) {
4570 		LIST_FOREACH(stp, &clp->lc_stateid[i], ls_hash) {
4571 		    if (stp->ls_stateid.other[2] == canuse) {
4572 			canuse++;
4573 			goto tryagain;
4574 		    }
4575 		}
4576 	    }
4577 	    clp->lc_flags |= LCL_INDEXNOTOK;
4578 	    return (canuse);
4579 	}
4580 
4581 	/*
4582 	 * Ok to start again from min + 1.
4583 	 */
4584 	clp->lc_stateindex = min_index + 1;
4585 	clp->lc_statemaxindex = max_index;
4586 	clp->lc_flags &= ~LCL_INDEXNOTOK;
4587 	return (clp->lc_stateindex);
4588 }
4589 
4590 /*
4591  * The following functions handle the stable storage file that deals with
4592  * the edge conditions described in RFC3530 Sec. 8.6.3.
4593  * The file is as follows:
4594  * - a single record at the beginning that has the lease time of the
4595  *   previous server instance (before the last reboot) and the nfsrvboottime
4596  *   values for the previous server boots.
4597  *   These previous boot times are used to ensure that the current
4598  *   nfsrvboottime does not, somehow, get set to a previous one.
4599  *   (This is important so that Stale ClientIDs and StateIDs can
4600  *    be recognized.)
4601  *   The number of previous nfsvrboottime values precedes the list.
4602  * - followed by some number of appended records with:
4603  *   - client id string
4604  *   - flag that indicates it is a record revoking state via lease
4605  *     expiration or similar
4606  *     OR has successfully acquired state.
4607  * These structures vary in length, with the client string at the end, up
4608  * to NFSV4_OPAQUELIMIT in size.
4609  *
4610  * At the end of the grace period, the file is truncated, the first
4611  * record is rewritten with updated information and any acquired state
4612  * records for successful reclaims of state are written.
4613  *
4614  * Subsequent records are appended when the first state is issued to
4615  * a client and when state is revoked for a client.
4616  *
4617  * When reading the file in, state issued records that come later in
4618  * the file override older ones, since the append log is in cronological order.
4619  * If, for some reason, the file can't be read, the grace period is
4620  * immediately terminated and all reclaims get NFSERR_NOGRACE.
4621  */
4622 
4623 /*
4624  * Read in the stable storage file. Called by nfssvc() before the nfsd
4625  * processes start servicing requests.
4626  */
4627 APPLESTATIC void
4628 nfsrv_setupstable(NFSPROC_T *p)
4629 {
4630 	struct nfsrv_stablefirst *sf = &nfsrv_stablefirst;
4631 	struct nfsrv_stable *sp, *nsp;
4632 	struct nfst_rec *tsp;
4633 	int error, i, tryagain;
4634 	off_t off = 0;
4635 	ssize_t aresid, len;
4636 
4637 	/*
4638 	 * If NFSNSF_UPDATEDONE is set, this is a restart of the nfsds without
4639 	 * a reboot, so state has not been lost.
4640 	 */
4641 	if (sf->nsf_flags & NFSNSF_UPDATEDONE)
4642 		return;
4643 	/*
4644 	 * Set Grace over just until the file reads successfully.
4645 	 */
4646 	nfsrvboottime = time_second;
4647 	LIST_INIT(&sf->nsf_head);
4648 	sf->nsf_flags = (NFSNSF_GRACEOVER | NFSNSF_NEEDLOCK);
4649 	sf->nsf_eograce = NFSD_MONOSEC + NFSRV_LEASEDELTA;
4650 	if (sf->nsf_fp == NULL)
4651 		return;
4652 	error = NFSD_RDWR(UIO_READ, NFSFPVNODE(sf->nsf_fp),
4653 	    (caddr_t)&sf->nsf_rec, sizeof (struct nfsf_rec), off, UIO_SYSSPACE,
4654 	    0, NFSFPCRED(sf->nsf_fp), &aresid, p);
4655 	if (error || aresid || sf->nsf_numboots == 0 ||
4656 		sf->nsf_numboots > NFSNSF_MAXNUMBOOTS)
4657 		return;
4658 
4659 	/*
4660 	 * Now, read in the boottimes.
4661 	 */
4662 	sf->nsf_bootvals = (time_t *)malloc((sf->nsf_numboots + 1) *
4663 		sizeof (time_t), M_TEMP, M_WAITOK);
4664 	off = sizeof (struct nfsf_rec);
4665 	error = NFSD_RDWR(UIO_READ, NFSFPVNODE(sf->nsf_fp),
4666 	    (caddr_t)sf->nsf_bootvals, sf->nsf_numboots * sizeof (time_t), off,
4667 	    UIO_SYSSPACE, 0, NFSFPCRED(sf->nsf_fp), &aresid, p);
4668 	if (error || aresid) {
4669 		free(sf->nsf_bootvals, M_TEMP);
4670 		sf->nsf_bootvals = NULL;
4671 		return;
4672 	}
4673 
4674 	/*
4675 	 * Make sure this nfsrvboottime is different from all recorded
4676 	 * previous ones.
4677 	 */
4678 	do {
4679 		tryagain = 0;
4680 		for (i = 0; i < sf->nsf_numboots; i++) {
4681 			if (nfsrvboottime == sf->nsf_bootvals[i]) {
4682 				nfsrvboottime++;
4683 				tryagain = 1;
4684 				break;
4685 			}
4686 		}
4687 	} while (tryagain);
4688 
4689 	sf->nsf_flags |= NFSNSF_OK;
4690 	off += (sf->nsf_numboots * sizeof (time_t));
4691 
4692 	/*
4693 	 * Read through the file, building a list of records for grace
4694 	 * checking.
4695 	 * Each record is between sizeof (struct nfst_rec) and
4696 	 * sizeof (struct nfst_rec) + NFSV4_OPAQUELIMIT - 1
4697 	 * and is actually sizeof (struct nfst_rec) + nst_len - 1.
4698 	 */
4699 	tsp = (struct nfst_rec *)malloc(sizeof (struct nfst_rec) +
4700 		NFSV4_OPAQUELIMIT - 1, M_TEMP, M_WAITOK);
4701 	do {
4702 	    error = NFSD_RDWR(UIO_READ, NFSFPVNODE(sf->nsf_fp),
4703 	        (caddr_t)tsp, sizeof (struct nfst_rec) + NFSV4_OPAQUELIMIT - 1,
4704 	        off, UIO_SYSSPACE, 0, NFSFPCRED(sf->nsf_fp), &aresid, p);
4705 	    len = (sizeof (struct nfst_rec) + NFSV4_OPAQUELIMIT - 1) - aresid;
4706 	    if (error || (len > 0 && (len < sizeof (struct nfst_rec) ||
4707 		len < (sizeof (struct nfst_rec) + tsp->len - 1)))) {
4708 		/*
4709 		 * Yuck, the file has been corrupted, so just return
4710 		 * after clearing out any restart state, so the grace period
4711 		 * is over.
4712 		 */
4713 		LIST_FOREACH_SAFE(sp, &sf->nsf_head, nst_list, nsp) {
4714 			LIST_REMOVE(sp, nst_list);
4715 			free(sp, M_TEMP);
4716 		}
4717 		free(tsp, M_TEMP);
4718 		sf->nsf_flags &= ~NFSNSF_OK;
4719 		free(sf->nsf_bootvals, M_TEMP);
4720 		sf->nsf_bootvals = NULL;
4721 		return;
4722 	    }
4723 	    if (len > 0) {
4724 		off += sizeof (struct nfst_rec) + tsp->len - 1;
4725 		/*
4726 		 * Search the list for a matching client.
4727 		 */
4728 		LIST_FOREACH(sp, &sf->nsf_head, nst_list) {
4729 			if (tsp->len == sp->nst_len &&
4730 			    !NFSBCMP(tsp->client, sp->nst_client, tsp->len))
4731 				break;
4732 		}
4733 		if (sp == LIST_END(&sf->nsf_head)) {
4734 			sp = (struct nfsrv_stable *)malloc(tsp->len +
4735 				sizeof (struct nfsrv_stable) - 1, M_TEMP,
4736 				M_WAITOK);
4737 			NFSBCOPY((caddr_t)tsp, (caddr_t)&sp->nst_rec,
4738 				sizeof (struct nfst_rec) + tsp->len - 1);
4739 			LIST_INSERT_HEAD(&sf->nsf_head, sp, nst_list);
4740 		} else {
4741 			if (tsp->flag == NFSNST_REVOKE)
4742 				sp->nst_flag |= NFSNST_REVOKE;
4743 			else
4744 				/*
4745 				 * A subsequent timestamp indicates the client
4746 				 * did a setclientid/confirm and any previous
4747 				 * revoke is no longer relevant.
4748 				 */
4749 				sp->nst_flag &= ~NFSNST_REVOKE;
4750 		}
4751 	    }
4752 	} while (len > 0);
4753 	free(tsp, M_TEMP);
4754 	sf->nsf_flags = NFSNSF_OK;
4755 	sf->nsf_eograce = NFSD_MONOSEC + sf->nsf_lease +
4756 		NFSRV_LEASEDELTA;
4757 }
4758 
4759 /*
4760  * Update the stable storage file, now that the grace period is over.
4761  */
4762 APPLESTATIC void
4763 nfsrv_updatestable(NFSPROC_T *p)
4764 {
4765 	struct nfsrv_stablefirst *sf = &nfsrv_stablefirst;
4766 	struct nfsrv_stable *sp, *nsp;
4767 	int i;
4768 	struct nfsvattr nva;
4769 	vnode_t vp;
4770 #if defined(__FreeBSD_version) && (__FreeBSD_version >= 500000)
4771 	mount_t mp = NULL;
4772 #endif
4773 	int error;
4774 
4775 	if (sf->nsf_fp == NULL || (sf->nsf_flags & NFSNSF_UPDATEDONE))
4776 		return;
4777 	sf->nsf_flags |= NFSNSF_UPDATEDONE;
4778 	/*
4779 	 * Ok, we need to rewrite the stable storage file.
4780 	 * - truncate to 0 length
4781 	 * - write the new first structure
4782 	 * - loop through the data structures, writing out any that
4783 	 *   have timestamps older than the old boot
4784 	 */
4785 	if (sf->nsf_bootvals) {
4786 		sf->nsf_numboots++;
4787 		for (i = sf->nsf_numboots - 2; i >= 0; i--)
4788 			sf->nsf_bootvals[i + 1] = sf->nsf_bootvals[i];
4789 	} else {
4790 		sf->nsf_numboots = 1;
4791 		sf->nsf_bootvals = (time_t *)malloc(sizeof (time_t),
4792 			M_TEMP, M_WAITOK);
4793 	}
4794 	sf->nsf_bootvals[0] = nfsrvboottime;
4795 	sf->nsf_lease = nfsrv_lease;
4796 	NFSVNO_ATTRINIT(&nva);
4797 	NFSVNO_SETATTRVAL(&nva, size, 0);
4798 	vp = NFSFPVNODE(sf->nsf_fp);
4799 	vn_start_write(vp, &mp, V_WAIT);
4800 	if (NFSVOPLOCK(vp, LK_EXCLUSIVE) == 0) {
4801 		error = nfsvno_setattr(vp, &nva, NFSFPCRED(sf->nsf_fp), p,
4802 		    NULL);
4803 		NFSVOPUNLOCK(vp, 0);
4804 	} else
4805 		error = EPERM;
4806 	vn_finished_write(mp);
4807 	if (!error)
4808 	    error = NFSD_RDWR(UIO_WRITE, vp,
4809 		(caddr_t)&sf->nsf_rec, sizeof (struct nfsf_rec), (off_t)0,
4810 		UIO_SYSSPACE, IO_SYNC, NFSFPCRED(sf->nsf_fp), NULL, p);
4811 	if (!error)
4812 	    error = NFSD_RDWR(UIO_WRITE, vp,
4813 		(caddr_t)sf->nsf_bootvals,
4814 		sf->nsf_numboots * sizeof (time_t),
4815 		(off_t)(sizeof (struct nfsf_rec)),
4816 		UIO_SYSSPACE, IO_SYNC, NFSFPCRED(sf->nsf_fp), NULL, p);
4817 	free(sf->nsf_bootvals, M_TEMP);
4818 	sf->nsf_bootvals = NULL;
4819 	if (error) {
4820 		sf->nsf_flags &= ~NFSNSF_OK;
4821 		printf("EEK! Can't write NfsV4 stable storage file\n");
4822 		return;
4823 	}
4824 	sf->nsf_flags |= NFSNSF_OK;
4825 
4826 	/*
4827 	 * Loop through the list and write out timestamp records for
4828 	 * any clients that successfully reclaimed state.
4829 	 */
4830 	LIST_FOREACH_SAFE(sp, &sf->nsf_head, nst_list, nsp) {
4831 		if (sp->nst_flag & NFSNST_GOTSTATE) {
4832 			nfsrv_writestable(sp->nst_client, sp->nst_len,
4833 				NFSNST_NEWSTATE, p);
4834 			sp->nst_clp->lc_flags |= LCL_STAMPEDSTABLE;
4835 		}
4836 		LIST_REMOVE(sp, nst_list);
4837 		free(sp, M_TEMP);
4838 	}
4839 	nfsrv_backupstable();
4840 }
4841 
4842 /*
4843  * Append a record to the stable storage file.
4844  */
4845 APPLESTATIC void
4846 nfsrv_writestable(u_char *client, int len, int flag, NFSPROC_T *p)
4847 {
4848 	struct nfsrv_stablefirst *sf = &nfsrv_stablefirst;
4849 	struct nfst_rec *sp;
4850 	int error;
4851 
4852 	if (!(sf->nsf_flags & NFSNSF_OK) || sf->nsf_fp == NULL)
4853 		return;
4854 	sp = (struct nfst_rec *)malloc(sizeof (struct nfst_rec) +
4855 		len - 1, M_TEMP, M_WAITOK);
4856 	sp->len = len;
4857 	NFSBCOPY(client, sp->client, len);
4858 	sp->flag = flag;
4859 	error = NFSD_RDWR(UIO_WRITE, NFSFPVNODE(sf->nsf_fp),
4860 	    (caddr_t)sp, sizeof (struct nfst_rec) + len - 1, (off_t)0,
4861 	    UIO_SYSSPACE, (IO_SYNC | IO_APPEND), NFSFPCRED(sf->nsf_fp), NULL, p);
4862 	free(sp, M_TEMP);
4863 	if (error) {
4864 		sf->nsf_flags &= ~NFSNSF_OK;
4865 		printf("EEK! Can't write NfsV4 stable storage file\n");
4866 	}
4867 }
4868 
4869 /*
4870  * This function is called during the grace period to mark a client
4871  * that successfully reclaimed state.
4872  */
4873 static void
4874 nfsrv_markstable(struct nfsclient *clp)
4875 {
4876 	struct nfsrv_stable *sp;
4877 
4878 	/*
4879 	 * First find the client structure.
4880 	 */
4881 	LIST_FOREACH(sp, &nfsrv_stablefirst.nsf_head, nst_list) {
4882 		if (sp->nst_len == clp->lc_idlen &&
4883 		    !NFSBCMP(sp->nst_client, clp->lc_id, sp->nst_len))
4884 			break;
4885 	}
4886 	if (sp == LIST_END(&nfsrv_stablefirst.nsf_head))
4887 		return;
4888 
4889 	/*
4890 	 * Now, just mark it and set the nfsclient back pointer.
4891 	 */
4892 	sp->nst_flag |= NFSNST_GOTSTATE;
4893 	sp->nst_clp = clp;
4894 }
4895 
4896 /*
4897  * This function is called when a NFSv4.1 client does a ReclaimComplete.
4898  * Very similar to nfsrv_markstable(), except for the flag being set.
4899  */
4900 static void
4901 nfsrv_markreclaim(struct nfsclient *clp)
4902 {
4903 	struct nfsrv_stable *sp;
4904 
4905 	/*
4906 	 * First find the client structure.
4907 	 */
4908 	LIST_FOREACH(sp, &nfsrv_stablefirst.nsf_head, nst_list) {
4909 		if (sp->nst_len == clp->lc_idlen &&
4910 		    !NFSBCMP(sp->nst_client, clp->lc_id, sp->nst_len))
4911 			break;
4912 	}
4913 	if (sp == LIST_END(&nfsrv_stablefirst.nsf_head))
4914 		return;
4915 
4916 	/*
4917 	 * Now, just set the flag.
4918 	 */
4919 	sp->nst_flag |= NFSNST_RECLAIMED;
4920 }
4921 
4922 /*
4923  * This function is called for a reclaim, to see if it gets grace.
4924  * It returns 0 if a reclaim is allowed, 1 otherwise.
4925  */
4926 static int
4927 nfsrv_checkstable(struct nfsclient *clp)
4928 {
4929 	struct nfsrv_stable *sp;
4930 
4931 	/*
4932 	 * First, find the entry for the client.
4933 	 */
4934 	LIST_FOREACH(sp, &nfsrv_stablefirst.nsf_head, nst_list) {
4935 		if (sp->nst_len == clp->lc_idlen &&
4936 		    !NFSBCMP(sp->nst_client, clp->lc_id, sp->nst_len))
4937 			break;
4938 	}
4939 
4940 	/*
4941 	 * If not in the list, state was revoked or no state was issued
4942 	 * since the previous reboot, a reclaim is denied.
4943 	 */
4944 	if (sp == LIST_END(&nfsrv_stablefirst.nsf_head) ||
4945 	    (sp->nst_flag & NFSNST_REVOKE) ||
4946 	    !(nfsrv_stablefirst.nsf_flags & NFSNSF_OK))
4947 		return (1);
4948 	return (0);
4949 }
4950 
4951 /*
4952  * Test for and try to clear out a conflicting client. This is called by
4953  * nfsrv_lockctrl() and nfsrv_openctrl() when conflicts with other clients
4954  * a found.
4955  * The trick here is that it can't revoke a conflicting client with an
4956  * expired lease unless it holds the v4root lock, so...
4957  * If no v4root lock, get the lock and return 1 to indicate "try again".
4958  * Return 0 to indicate the conflict can't be revoked and 1 to indicate
4959  * the revocation worked and the conflicting client is "bye, bye", so it
4960  * can be tried again.
4961  * Return 2 to indicate that the vnode is VI_DOOMED after NFSVOPLOCK().
4962  * Unlocks State before a non-zero value is returned.
4963  */
4964 static int
4965 nfsrv_clientconflict(struct nfsclient *clp, int *haslockp, vnode_t vp,
4966     NFSPROC_T *p)
4967 {
4968 	int gotlock, lktype = 0;
4969 
4970 	/*
4971 	 * If lease hasn't expired, we can't fix it.
4972 	 */
4973 	if (clp->lc_expiry >= NFSD_MONOSEC ||
4974 	    !(nfsrv_stablefirst.nsf_flags & NFSNSF_UPDATEDONE))
4975 		return (0);
4976 	if (*haslockp == 0) {
4977 		NFSUNLOCKSTATE();
4978 		if (vp != NULL) {
4979 			lktype = NFSVOPISLOCKED(vp);
4980 			NFSVOPUNLOCK(vp, 0);
4981 		}
4982 		NFSLOCKV4ROOTMUTEX();
4983 		nfsv4_relref(&nfsv4rootfs_lock);
4984 		do {
4985 			gotlock = nfsv4_lock(&nfsv4rootfs_lock, 1, NULL,
4986 			    NFSV4ROOTLOCKMUTEXPTR, NULL);
4987 		} while (!gotlock);
4988 		NFSUNLOCKV4ROOTMUTEX();
4989 		*haslockp = 1;
4990 		if (vp != NULL) {
4991 			NFSVOPLOCK(vp, lktype | LK_RETRY);
4992 			if ((vp->v_iflag & VI_DOOMED) != 0)
4993 				return (2);
4994 		}
4995 		return (1);
4996 	}
4997 	NFSUNLOCKSTATE();
4998 
4999 	/*
5000 	 * Ok, we can expire the conflicting client.
5001 	 */
5002 	nfsrv_writestable(clp->lc_id, clp->lc_idlen, NFSNST_REVOKE, p);
5003 	nfsrv_backupstable();
5004 	nfsrv_cleanclient(clp, p);
5005 	nfsrv_freedeleglist(&clp->lc_deleg);
5006 	nfsrv_freedeleglist(&clp->lc_olddeleg);
5007 	LIST_REMOVE(clp, lc_hash);
5008 	nfsrv_zapclient(clp, p);
5009 	return (1);
5010 }
5011 
5012 /*
5013  * Resolve a delegation conflict.
5014  * Returns 0 to indicate the conflict was resolved without sleeping.
5015  * Return -1 to indicate that the caller should check for conflicts again.
5016  * Return > 0 for an error that should be returned, normally NFSERR_DELAY.
5017  *
5018  * Also, manipulate the nfsv4root_lock, as required. It isn't changed
5019  * for a return of 0, since there was no sleep and it could be required
5020  * later. It is released for a return of NFSERR_DELAY, since the caller
5021  * will return that error. It is released when a sleep was done waiting
5022  * for the delegation to be returned or expire (so that other nfsds can
5023  * handle ops). Then, it must be acquired for the write to stable storage.
5024  * (This function is somewhat similar to nfsrv_clientconflict(), but
5025  *  the semantics differ in a couple of subtle ways. The return of 0
5026  *  indicates the conflict was resolved without sleeping here, not
5027  *  that the conflict can't be resolved and the handling of nfsv4root_lock
5028  *  differs, as noted above.)
5029  * Unlocks State before returning a non-zero value.
5030  */
5031 static int
5032 nfsrv_delegconflict(struct nfsstate *stp, int *haslockp, NFSPROC_T *p,
5033     vnode_t vp)
5034 {
5035 	struct nfsclient *clp = stp->ls_clp;
5036 	int gotlock, error, lktype = 0, retrycnt, zapped_clp;
5037 	nfsv4stateid_t tstateid;
5038 	fhandle_t tfh;
5039 
5040 	/*
5041 	 * If the conflict is with an old delegation...
5042 	 */
5043 	if (stp->ls_flags & NFSLCK_OLDDELEG) {
5044 		/*
5045 		 * You can delete it, if it has expired.
5046 		 */
5047 		if (clp->lc_delegtime < NFSD_MONOSEC) {
5048 			nfsrv_freedeleg(stp);
5049 			NFSUNLOCKSTATE();
5050 			error = -1;
5051 			goto out;
5052 		}
5053 		NFSUNLOCKSTATE();
5054 		/*
5055 		 * During this delay, the old delegation could expire or it
5056 		 * could be recovered by the client via an Open with
5057 		 * CLAIM_DELEGATE_PREV.
5058 		 * Release the nfsv4root_lock, if held.
5059 		 */
5060 		if (*haslockp) {
5061 			*haslockp = 0;
5062 			NFSLOCKV4ROOTMUTEX();
5063 			nfsv4_unlock(&nfsv4rootfs_lock, 1);
5064 			NFSUNLOCKV4ROOTMUTEX();
5065 		}
5066 		error = NFSERR_DELAY;
5067 		goto out;
5068 	}
5069 
5070 	/*
5071 	 * It's a current delegation, so:
5072 	 * - check to see if the delegation has expired
5073 	 *   - if so, get the v4root lock and then expire it
5074 	 */
5075 	if (!(stp->ls_flags & NFSLCK_DELEGRECALL)) {
5076 		/*
5077 		 * - do a recall callback, since not yet done
5078 		 * For now, never allow truncate to be set. To use
5079 		 * truncate safely, it must be guaranteed that the
5080 		 * Remove, Rename or Setattr with size of 0 will
5081 		 * succeed and that would require major changes to
5082 		 * the VFS/Vnode OPs.
5083 		 * Set the expiry time large enough so that it won't expire
5084 		 * until after the callback, then set it correctly, once
5085 		 * the callback is done. (The delegation will now time
5086 		 * out whether or not the Recall worked ok. The timeout
5087 		 * will be extended when ops are done on the delegation
5088 		 * stateid, up to the timelimit.)
5089 		 */
5090 		stp->ls_delegtime = NFSD_MONOSEC + (2 * nfsrv_lease) +
5091 		    NFSRV_LEASEDELTA;
5092 		stp->ls_delegtimelimit = NFSD_MONOSEC + (6 * nfsrv_lease) +
5093 		    NFSRV_LEASEDELTA;
5094 		stp->ls_flags |= NFSLCK_DELEGRECALL;
5095 
5096 		/*
5097 		 * Loop NFSRV_CBRETRYCNT times while the CBRecall replies
5098 		 * NFSERR_BADSTATEID or NFSERR_BADHANDLE. This is done
5099 		 * in order to try and avoid a race that could happen
5100 		 * when a CBRecall request passed the Open reply with
5101 		 * the delegation in it when transitting the network.
5102 		 * Since nfsrv_docallback will sleep, don't use stp after
5103 		 * the call.
5104 		 */
5105 		NFSBCOPY((caddr_t)&stp->ls_stateid, (caddr_t)&tstateid,
5106 		    sizeof (tstateid));
5107 		NFSBCOPY((caddr_t)&stp->ls_lfp->lf_fh, (caddr_t)&tfh,
5108 		    sizeof (tfh));
5109 		NFSUNLOCKSTATE();
5110 		if (*haslockp) {
5111 			*haslockp = 0;
5112 			NFSLOCKV4ROOTMUTEX();
5113 			nfsv4_unlock(&nfsv4rootfs_lock, 1);
5114 			NFSUNLOCKV4ROOTMUTEX();
5115 		}
5116 		retrycnt = 0;
5117 		do {
5118 		    error = nfsrv_docallback(clp, NFSV4OP_CBRECALL,
5119 			&tstateid, 0, &tfh, NULL, NULL, 0, p);
5120 		    retrycnt++;
5121 		} while ((error == NFSERR_BADSTATEID ||
5122 		    error == NFSERR_BADHANDLE) && retrycnt < NFSV4_CBRETRYCNT);
5123 		error = NFSERR_DELAY;
5124 		goto out;
5125 	}
5126 
5127 	if (clp->lc_expiry >= NFSD_MONOSEC &&
5128 	    stp->ls_delegtime >= NFSD_MONOSEC) {
5129 		NFSUNLOCKSTATE();
5130 		/*
5131 		 * A recall has been done, but it has not yet expired.
5132 		 * So, RETURN_DELAY.
5133 		 */
5134 		if (*haslockp) {
5135 			*haslockp = 0;
5136 			NFSLOCKV4ROOTMUTEX();
5137 			nfsv4_unlock(&nfsv4rootfs_lock, 1);
5138 			NFSUNLOCKV4ROOTMUTEX();
5139 		}
5140 		error = NFSERR_DELAY;
5141 		goto out;
5142 	}
5143 
5144 	/*
5145 	 * If we don't yet have the lock, just get it and then return,
5146 	 * since we need that before deleting expired state, such as
5147 	 * this delegation.
5148 	 * When getting the lock, unlock the vnode, so other nfsds that
5149 	 * are in progress, won't get stuck waiting for the vnode lock.
5150 	 */
5151 	if (*haslockp == 0) {
5152 		NFSUNLOCKSTATE();
5153 		if (vp != NULL) {
5154 			lktype = NFSVOPISLOCKED(vp);
5155 			NFSVOPUNLOCK(vp, 0);
5156 		}
5157 		NFSLOCKV4ROOTMUTEX();
5158 		nfsv4_relref(&nfsv4rootfs_lock);
5159 		do {
5160 			gotlock = nfsv4_lock(&nfsv4rootfs_lock, 1, NULL,
5161 			    NFSV4ROOTLOCKMUTEXPTR, NULL);
5162 		} while (!gotlock);
5163 		NFSUNLOCKV4ROOTMUTEX();
5164 		*haslockp = 1;
5165 		if (vp != NULL) {
5166 			NFSVOPLOCK(vp, lktype | LK_RETRY);
5167 			if ((vp->v_iflag & VI_DOOMED) != 0) {
5168 				*haslockp = 0;
5169 				NFSLOCKV4ROOTMUTEX();
5170 				nfsv4_unlock(&nfsv4rootfs_lock, 1);
5171 				NFSUNLOCKV4ROOTMUTEX();
5172 				error = NFSERR_PERM;
5173 				goto out;
5174 			}
5175 		}
5176 		error = -1;
5177 		goto out;
5178 	}
5179 
5180 	NFSUNLOCKSTATE();
5181 	/*
5182 	 * Ok, we can delete the expired delegation.
5183 	 * First, write the Revoke record to stable storage and then
5184 	 * clear out the conflict.
5185 	 * Since all other nfsd threads are now blocked, we can safely
5186 	 * sleep without the state changing.
5187 	 */
5188 	nfsrv_writestable(clp->lc_id, clp->lc_idlen, NFSNST_REVOKE, p);
5189 	nfsrv_backupstable();
5190 	if (clp->lc_expiry < NFSD_MONOSEC) {
5191 		nfsrv_cleanclient(clp, p);
5192 		nfsrv_freedeleglist(&clp->lc_deleg);
5193 		nfsrv_freedeleglist(&clp->lc_olddeleg);
5194 		LIST_REMOVE(clp, lc_hash);
5195 		zapped_clp = 1;
5196 	} else {
5197 		nfsrv_freedeleg(stp);
5198 		zapped_clp = 0;
5199 	}
5200 	if (zapped_clp)
5201 		nfsrv_zapclient(clp, p);
5202 	error = -1;
5203 
5204 out:
5205 	NFSEXITCODE(error);
5206 	return (error);
5207 }
5208 
5209 /*
5210  * Check for a remove allowed, if remove is set to 1 and get rid of
5211  * delegations.
5212  */
5213 APPLESTATIC int
5214 nfsrv_checkremove(vnode_t vp, int remove, NFSPROC_T *p)
5215 {
5216 	struct nfsstate *stp;
5217 	struct nfslockfile *lfp;
5218 	int error, haslock = 0;
5219 	fhandle_t nfh;
5220 
5221 	/*
5222 	 * First, get the lock file structure.
5223 	 * (A return of -1 means no associated state, so remove ok.)
5224 	 */
5225 	error = nfsrv_getlockfh(vp, NFSLCK_CHECK, NULL, &nfh, p);
5226 tryagain:
5227 	NFSLOCKSTATE();
5228 	if (!error)
5229 		error = nfsrv_getlockfile(NFSLCK_CHECK, NULL, &lfp, &nfh, 0);
5230 	if (error) {
5231 		NFSUNLOCKSTATE();
5232 		if (haslock) {
5233 			NFSLOCKV4ROOTMUTEX();
5234 			nfsv4_unlock(&nfsv4rootfs_lock, 1);
5235 			NFSUNLOCKV4ROOTMUTEX();
5236 		}
5237 		if (error == -1)
5238 			error = 0;
5239 		goto out;
5240 	}
5241 
5242 	/*
5243 	 * Now, we must Recall any delegations.
5244 	 */
5245 	error = nfsrv_cleandeleg(vp, lfp, NULL, &haslock, p);
5246 	if (error) {
5247 		/*
5248 		 * nfsrv_cleandeleg() unlocks state for non-zero
5249 		 * return.
5250 		 */
5251 		if (error == -1)
5252 			goto tryagain;
5253 		if (haslock) {
5254 			NFSLOCKV4ROOTMUTEX();
5255 			nfsv4_unlock(&nfsv4rootfs_lock, 1);
5256 			NFSUNLOCKV4ROOTMUTEX();
5257 		}
5258 		goto out;
5259 	}
5260 
5261 	/*
5262 	 * Now, look for a conflicting open share.
5263 	 */
5264 	if (remove) {
5265 		/*
5266 		 * If the entry in the directory was the last reference to the
5267 		 * corresponding filesystem object, the object can be destroyed
5268 		 * */
5269 		if(lfp->lf_usecount>1)
5270 			LIST_FOREACH(stp, &lfp->lf_open, ls_file) {
5271 				if (stp->ls_flags & NFSLCK_WRITEDENY) {
5272 					error = NFSERR_FILEOPEN;
5273 					break;
5274 				}
5275 			}
5276 	}
5277 
5278 	NFSUNLOCKSTATE();
5279 	if (haslock) {
5280 		NFSLOCKV4ROOTMUTEX();
5281 		nfsv4_unlock(&nfsv4rootfs_lock, 1);
5282 		NFSUNLOCKV4ROOTMUTEX();
5283 	}
5284 
5285 out:
5286 	NFSEXITCODE(error);
5287 	return (error);
5288 }
5289 
5290 /*
5291  * Clear out all delegations for the file referred to by lfp.
5292  * May return NFSERR_DELAY, if there will be a delay waiting for
5293  * delegations to expire.
5294  * Returns -1 to indicate it slept while recalling a delegation.
5295  * This function has the side effect of deleting the nfslockfile structure,
5296  * if it no longer has associated state and didn't have to sleep.
5297  * Unlocks State before a non-zero value is returned.
5298  */
5299 static int
5300 nfsrv_cleandeleg(vnode_t vp, struct nfslockfile *lfp,
5301     struct nfsclient *clp, int *haslockp, NFSPROC_T *p)
5302 {
5303 	struct nfsstate *stp, *nstp;
5304 	int ret = 0;
5305 
5306 	stp = LIST_FIRST(&lfp->lf_deleg);
5307 	while (stp != LIST_END(&lfp->lf_deleg)) {
5308 		nstp = LIST_NEXT(stp, ls_file);
5309 		if (stp->ls_clp != clp) {
5310 			ret = nfsrv_delegconflict(stp, haslockp, p, vp);
5311 			if (ret) {
5312 				/*
5313 				 * nfsrv_delegconflict() unlocks state
5314 				 * when it returns non-zero.
5315 				 */
5316 				goto out;
5317 			}
5318 		}
5319 		stp = nstp;
5320 	}
5321 out:
5322 	NFSEXITCODE(ret);
5323 	return (ret);
5324 }
5325 
5326 /*
5327  * There are certain operations that, when being done outside of NFSv4,
5328  * require that any NFSv4 delegation for the file be recalled.
5329  * This function is to be called for those cases:
5330  * VOP_RENAME() - When a delegation is being recalled for any reason,
5331  *	the client may have to do Opens against the server, using the file's
5332  *	final component name. If the file has been renamed on the server,
5333  *	that component name will be incorrect and the Open will fail.
5334  * VOP_REMOVE() - Theoretically, a client could Open a file after it has
5335  *	been removed on the server, if there is a delegation issued to
5336  *	that client for the file. I say "theoretically" since clients
5337  *	normally do an Access Op before the Open and that Access Op will
5338  *	fail with ESTALE. Note that NFSv2 and 3 don't even do Opens, so
5339  *	they will detect the file's removal in the same manner. (There is
5340  *	one case where RFC3530 allows a client to do an Open without first
5341  *	doing an Access Op, which is passage of a check against the ACE
5342  *	returned with a Write delegation, but current practice is to ignore
5343  *	the ACE and always do an Access Op.)
5344  *	Since the functions can only be called with an unlocked vnode, this
5345  *	can't be done at this time.
5346  * VOP_ADVLOCK() - When a client holds a delegation, it can issue byte range
5347  *	locks locally in the client, which are not visible to the server. To
5348  *	deal with this, issuing of delegations for a vnode must be disabled
5349  *	and all delegations for the vnode recalled. This is done via the
5350  *	second function, using the VV_DISABLEDELEG vflag on the vnode.
5351  */
5352 APPLESTATIC void
5353 nfsd_recalldelegation(vnode_t vp, NFSPROC_T *p)
5354 {
5355 	time_t starttime;
5356 	int error;
5357 
5358 	/*
5359 	 * First, check to see if the server is currently running and it has
5360 	 * been called for a regular file when issuing delegations.
5361 	 */
5362 	if (newnfs_numnfsd == 0 || vp->v_type != VREG ||
5363 	    nfsrv_issuedelegs == 0)
5364 		return;
5365 
5366 	KASSERT((NFSVOPISLOCKED(vp) != LK_EXCLUSIVE), ("vp %p is locked", vp));
5367 	/*
5368 	 * First, get a reference on the nfsv4rootfs_lock so that an
5369 	 * exclusive lock cannot be acquired by another thread.
5370 	 */
5371 	NFSLOCKV4ROOTMUTEX();
5372 	nfsv4_getref(&nfsv4rootfs_lock, NULL, NFSV4ROOTLOCKMUTEXPTR, NULL);
5373 	NFSUNLOCKV4ROOTMUTEX();
5374 
5375 	/*
5376 	 * Now, call nfsrv_checkremove() in a loop while it returns
5377 	 * NFSERR_DELAY. Return upon any other error or when timed out.
5378 	 */
5379 	starttime = NFSD_MONOSEC;
5380 	do {
5381 		if (NFSVOPLOCK(vp, LK_EXCLUSIVE) == 0) {
5382 			error = nfsrv_checkremove(vp, 0, p);
5383 			NFSVOPUNLOCK(vp, 0);
5384 		} else
5385 			error = EPERM;
5386 		if (error == NFSERR_DELAY) {
5387 			if (NFSD_MONOSEC - starttime > NFS_REMOVETIMEO)
5388 				break;
5389 			/* Sleep for a short period of time */
5390 			(void) nfs_catnap(PZERO, 0, "nfsremove");
5391 		}
5392 	} while (error == NFSERR_DELAY);
5393 	NFSLOCKV4ROOTMUTEX();
5394 	nfsv4_relref(&nfsv4rootfs_lock);
5395 	NFSUNLOCKV4ROOTMUTEX();
5396 }
5397 
5398 APPLESTATIC void
5399 nfsd_disabledelegation(vnode_t vp, NFSPROC_T *p)
5400 {
5401 
5402 #ifdef VV_DISABLEDELEG
5403 	/*
5404 	 * First, flag issuance of delegations disabled.
5405 	 */
5406 	atomic_set_long(&vp->v_vflag, VV_DISABLEDELEG);
5407 #endif
5408 
5409 	/*
5410 	 * Then call nfsd_recalldelegation() to get rid of all extant
5411 	 * delegations.
5412 	 */
5413 	nfsd_recalldelegation(vp, p);
5414 }
5415 
5416 /*
5417  * Check for conflicting locks, etc. and then get rid of delegations.
5418  * (At one point I thought that I should get rid of delegations for any
5419  *  Setattr, since it could potentially disallow the I/O op (read or write)
5420  *  allowed by the delegation. However, Setattr Ops that aren't changing
5421  *  the size get a stateid of all 0s, so you can't tell if it is a delegation
5422  *  for the same client or a different one, so I decided to only get rid
5423  *  of delegations for other clients when the size is being changed.)
5424  * In general, a Setattr can disable NFS I/O Ops that are outstanding, such
5425  * as Write backs, even if there is no delegation, so it really isn't any
5426  * different?)
5427  */
5428 APPLESTATIC int
5429 nfsrv_checksetattr(vnode_t vp, struct nfsrv_descript *nd,
5430     nfsv4stateid_t *stateidp, struct nfsvattr *nvap, nfsattrbit_t *attrbitp,
5431     struct nfsexstuff *exp, NFSPROC_T *p)
5432 {
5433 	struct nfsstate st, *stp = &st;
5434 	struct nfslock lo, *lop = &lo;
5435 	int error = 0;
5436 	nfsquad_t clientid;
5437 
5438 	if (NFSISSET_ATTRBIT(attrbitp, NFSATTRBIT_SIZE)) {
5439 		stp->ls_flags = (NFSLCK_CHECK | NFSLCK_WRITEACCESS);
5440 		lop->lo_first = nvap->na_size;
5441 	} else {
5442 		stp->ls_flags = 0;
5443 		lop->lo_first = 0;
5444 	}
5445 	if (NFSISSET_ATTRBIT(attrbitp, NFSATTRBIT_OWNER) ||
5446 	    NFSISSET_ATTRBIT(attrbitp, NFSATTRBIT_OWNERGROUP) ||
5447 	    NFSISSET_ATTRBIT(attrbitp, NFSATTRBIT_MODE) ||
5448 	    NFSISSET_ATTRBIT(attrbitp, NFSATTRBIT_ACL))
5449 		stp->ls_flags |= NFSLCK_SETATTR;
5450 	if (stp->ls_flags == 0)
5451 		goto out;
5452 	lop->lo_end = NFS64BITSSET;
5453 	lop->lo_flags = NFSLCK_WRITE;
5454 	stp->ls_ownerlen = 0;
5455 	stp->ls_op = NULL;
5456 	stp->ls_uid = nd->nd_cred->cr_uid;
5457 	stp->ls_stateid.seqid = stateidp->seqid;
5458 	clientid.lval[0] = stp->ls_stateid.other[0] = stateidp->other[0];
5459 	clientid.lval[1] = stp->ls_stateid.other[1] = stateidp->other[1];
5460 	stp->ls_stateid.other[2] = stateidp->other[2];
5461 	error = nfsrv_lockctrl(vp, &stp, &lop, NULL, clientid,
5462 	    stateidp, exp, nd, p);
5463 
5464 out:
5465 	NFSEXITCODE2(error, nd);
5466 	return (error);
5467 }
5468 
5469 /*
5470  * Check for a write delegation and do a CBGETATTR if there is one, updating
5471  * the attributes, as required.
5472  * Should I return an error if I can't get the attributes? (For now, I'll
5473  * just return ok.
5474  */
5475 APPLESTATIC int
5476 nfsrv_checkgetattr(struct nfsrv_descript *nd, vnode_t vp,
5477     struct nfsvattr *nvap, nfsattrbit_t *attrbitp, NFSPROC_T *p)
5478 {
5479 	struct nfsstate *stp;
5480 	struct nfslockfile *lfp;
5481 	struct nfsclient *clp;
5482 	struct nfsvattr nva;
5483 	fhandle_t nfh;
5484 	int error = 0;
5485 	nfsattrbit_t cbbits;
5486 	u_quad_t delegfilerev;
5487 
5488 	NFSCBGETATTR_ATTRBIT(attrbitp, &cbbits);
5489 	if (!NFSNONZERO_ATTRBIT(&cbbits))
5490 		goto out;
5491 	if (nfsrv_writedelegcnt == 0)
5492 		goto out;
5493 
5494 	/*
5495 	 * Get the lock file structure.
5496 	 * (A return of -1 means no associated state, so return ok.)
5497 	 */
5498 	error = nfsrv_getlockfh(vp, NFSLCK_CHECK, NULL, &nfh, p);
5499 	NFSLOCKSTATE();
5500 	if (!error)
5501 		error = nfsrv_getlockfile(NFSLCK_CHECK, NULL, &lfp, &nfh, 0);
5502 	if (error) {
5503 		NFSUNLOCKSTATE();
5504 		if (error == -1)
5505 			error = 0;
5506 		goto out;
5507 	}
5508 
5509 	/*
5510 	 * Now, look for a write delegation.
5511 	 */
5512 	LIST_FOREACH(stp, &lfp->lf_deleg, ls_file) {
5513 		if (stp->ls_flags & NFSLCK_DELEGWRITE)
5514 			break;
5515 	}
5516 	if (stp == LIST_END(&lfp->lf_deleg)) {
5517 		NFSUNLOCKSTATE();
5518 		goto out;
5519 	}
5520 	clp = stp->ls_clp;
5521 	delegfilerev = stp->ls_filerev;
5522 
5523 	/*
5524 	 * If the Write delegation was issued as a part of this Compound RPC
5525 	 * or if we have an Implied Clientid (used in a previous Op in this
5526 	 * compound) and it is the client the delegation was issued to,
5527 	 * just return ok.
5528 	 * I also assume that it is from the same client iff the network
5529 	 * host IP address is the same as the callback address. (Not
5530 	 * exactly correct by the RFC, but avoids a lot of Getattr
5531 	 * callbacks.)
5532 	 */
5533 	if (nd->nd_compref == stp->ls_compref ||
5534 	    ((nd->nd_flag & ND_IMPLIEDCLID) &&
5535 	     clp->lc_clientid.qval == nd->nd_clientid.qval) ||
5536 	     nfsaddr2_match(clp->lc_req.nr_nam, nd->nd_nam)) {
5537 		NFSUNLOCKSTATE();
5538 		goto out;
5539 	}
5540 
5541 	/*
5542 	 * We are now done with the delegation state structure,
5543 	 * so the statelock can be released and we can now tsleep().
5544 	 */
5545 
5546 	/*
5547 	 * Now, we must do the CB Getattr callback, to see if Change or Size
5548 	 * has changed.
5549 	 */
5550 	if (clp->lc_expiry >= NFSD_MONOSEC) {
5551 		NFSUNLOCKSTATE();
5552 		NFSVNO_ATTRINIT(&nva);
5553 		nva.na_filerev = NFS64BITSSET;
5554 		error = nfsrv_docallback(clp, NFSV4OP_CBGETATTR, NULL,
5555 		    0, &nfh, &nva, &cbbits, 0, p);
5556 		if (!error) {
5557 			if ((nva.na_filerev != NFS64BITSSET &&
5558 			    nva.na_filerev > delegfilerev) ||
5559 			    (NFSVNO_ISSETSIZE(&nva) &&
5560 			     nva.na_size != nvap->na_size)) {
5561 				error = nfsvno_updfilerev(vp, nvap, nd, p);
5562 				if (NFSVNO_ISSETSIZE(&nva))
5563 					nvap->na_size = nva.na_size;
5564 			}
5565 		} else
5566 			error = 0;	/* Ignore callback errors for now. */
5567 	} else {
5568 		NFSUNLOCKSTATE();
5569 	}
5570 
5571 out:
5572 	NFSEXITCODE2(error, nd);
5573 	return (error);
5574 }
5575 
5576 /*
5577  * This function looks for openowners that haven't had any opens for
5578  * a while and throws them away. Called by an nfsd when NFSNSF_NOOPENS
5579  * is set.
5580  */
5581 APPLESTATIC void
5582 nfsrv_throwawayopens(NFSPROC_T *p)
5583 {
5584 	struct nfsclient *clp, *nclp;
5585 	struct nfsstate *stp, *nstp;
5586 	int i;
5587 
5588 	NFSLOCKSTATE();
5589 	nfsrv_stablefirst.nsf_flags &= ~NFSNSF_NOOPENS;
5590 	/*
5591 	 * For each client...
5592 	 */
5593 	for (i = 0; i < nfsrv_clienthashsize; i++) {
5594 	    LIST_FOREACH_SAFE(clp, &nfsclienthash[i], lc_hash, nclp) {
5595 		LIST_FOREACH_SAFE(stp, &clp->lc_open, ls_list, nstp) {
5596 			if (LIST_EMPTY(&stp->ls_open) &&
5597 			    (stp->ls_noopens > NFSNOOPEN ||
5598 			     (nfsrv_openpluslock * 2) >
5599 			     nfsrv_v4statelimit))
5600 				nfsrv_freeopenowner(stp, 0, p);
5601 		}
5602 	    }
5603 	}
5604 	NFSUNLOCKSTATE();
5605 }
5606 
5607 /*
5608  * This function checks to see if the credentials are the same.
5609  * Returns 1 for not same, 0 otherwise.
5610  */
5611 static int
5612 nfsrv_notsamecredname(struct nfsrv_descript *nd, struct nfsclient *clp)
5613 {
5614 
5615 	if (nd->nd_flag & ND_GSS) {
5616 		if (!(clp->lc_flags & LCL_GSS))
5617 			return (1);
5618 		if (clp->lc_flags & LCL_NAME) {
5619 			if (nd->nd_princlen != clp->lc_namelen ||
5620 			    NFSBCMP(nd->nd_principal, clp->lc_name,
5621 				clp->lc_namelen))
5622 				return (1);
5623 			else
5624 				return (0);
5625 		}
5626 		if (nd->nd_cred->cr_uid == clp->lc_uid)
5627 			return (0);
5628 		else
5629 			return (1);
5630 	} else if (clp->lc_flags & LCL_GSS)
5631 		return (1);
5632 	/*
5633 	 * For AUTH_SYS, allow the same uid or root. (This is underspecified
5634 	 * in RFC3530, which talks about principals, but doesn't say anything
5635 	 * about uids for AUTH_SYS.)
5636 	 */
5637 	if (nd->nd_cred->cr_uid == clp->lc_uid || nd->nd_cred->cr_uid == 0)
5638 		return (0);
5639 	else
5640 		return (1);
5641 }
5642 
5643 /*
5644  * Calculate the lease expiry time.
5645  */
5646 static time_t
5647 nfsrv_leaseexpiry(void)
5648 {
5649 
5650 	if (nfsrv_stablefirst.nsf_eograce > NFSD_MONOSEC)
5651 		return (NFSD_MONOSEC + 2 * (nfsrv_lease + NFSRV_LEASEDELTA));
5652 	return (NFSD_MONOSEC + nfsrv_lease + NFSRV_LEASEDELTA);
5653 }
5654 
5655 /*
5656  * Delay the delegation timeout as far as ls_delegtimelimit, as required.
5657  */
5658 static void
5659 nfsrv_delaydelegtimeout(struct nfsstate *stp)
5660 {
5661 
5662 	if ((stp->ls_flags & NFSLCK_DELEGRECALL) == 0)
5663 		return;
5664 
5665 	if ((stp->ls_delegtime + 15) > NFSD_MONOSEC &&
5666 	    stp->ls_delegtime < stp->ls_delegtimelimit) {
5667 		stp->ls_delegtime += nfsrv_lease;
5668 		if (stp->ls_delegtime > stp->ls_delegtimelimit)
5669 			stp->ls_delegtime = stp->ls_delegtimelimit;
5670 	}
5671 }
5672 
5673 /*
5674  * This function checks to see if there is any other state associated
5675  * with the openowner for this Open.
5676  * It returns 1 if there is no other state, 0 otherwise.
5677  */
5678 static int
5679 nfsrv_nootherstate(struct nfsstate *stp)
5680 {
5681 	struct nfsstate *tstp;
5682 
5683 	LIST_FOREACH(tstp, &stp->ls_openowner->ls_open, ls_list) {
5684 		if (tstp != stp || !LIST_EMPTY(&tstp->ls_lock))
5685 			return (0);
5686 	}
5687 	return (1);
5688 }
5689 
5690 /*
5691  * Create a list of lock deltas (changes to local byte range locking
5692  * that can be rolled back using the list) and apply the changes via
5693  * nfsvno_advlock(). Optionally, lock the list. It is expected that either
5694  * the rollback or update function will be called after this.
5695  * It returns an error (and rolls back, as required), if any nfsvno_advlock()
5696  * call fails. If it returns an error, it will unlock the list.
5697  */
5698 static int
5699 nfsrv_locallock(vnode_t vp, struct nfslockfile *lfp, int flags,
5700     uint64_t first, uint64_t end, struct nfslockconflict *cfp, NFSPROC_T *p)
5701 {
5702 	struct nfslock *lop, *nlop;
5703 	int error = 0;
5704 
5705 	/* Loop through the list of locks. */
5706 	lop = LIST_FIRST(&lfp->lf_locallock);
5707 	while (first < end && lop != NULL) {
5708 		nlop = LIST_NEXT(lop, lo_lckowner);
5709 		if (first >= lop->lo_end) {
5710 			/* not there yet */
5711 			lop = nlop;
5712 		} else if (first < lop->lo_first) {
5713 			/* new one starts before entry in list */
5714 			if (end <= lop->lo_first) {
5715 				/* no overlap between old and new */
5716 				error = nfsrv_dolocal(vp, lfp, flags,
5717 				    NFSLCK_UNLOCK, first, end, cfp, p);
5718 				if (error != 0)
5719 					break;
5720 				first = end;
5721 			} else {
5722 				/* handle fragment overlapped with new one */
5723 				error = nfsrv_dolocal(vp, lfp, flags,
5724 				    NFSLCK_UNLOCK, first, lop->lo_first, cfp,
5725 				    p);
5726 				if (error != 0)
5727 					break;
5728 				first = lop->lo_first;
5729 			}
5730 		} else {
5731 			/* new one overlaps this entry in list */
5732 			if (end <= lop->lo_end) {
5733 				/* overlaps all of new one */
5734 				error = nfsrv_dolocal(vp, lfp, flags,
5735 				    lop->lo_flags, first, end, cfp, p);
5736 				if (error != 0)
5737 					break;
5738 				first = end;
5739 			} else {
5740 				/* handle fragment overlapped with new one */
5741 				error = nfsrv_dolocal(vp, lfp, flags,
5742 				    lop->lo_flags, first, lop->lo_end, cfp, p);
5743 				if (error != 0)
5744 					break;
5745 				first = lop->lo_end;
5746 				lop = nlop;
5747 			}
5748 		}
5749 	}
5750 	if (first < end && error == 0)
5751 		/* handle fragment past end of list */
5752 		error = nfsrv_dolocal(vp, lfp, flags, NFSLCK_UNLOCK, first,
5753 		    end, cfp, p);
5754 
5755 	NFSEXITCODE(error);
5756 	return (error);
5757 }
5758 
5759 /*
5760  * Local lock unlock. Unlock all byte ranges that are no longer locked
5761  * by NFSv4. To do this, unlock any subranges of first-->end that
5762  * do not overlap with the byte ranges of any lock in the lfp->lf_lock
5763  * list. This list has all locks for the file held by other
5764  * <clientid, lockowner> tuples. The list is ordered by increasing
5765  * lo_first value, but may have entries that overlap each other, for
5766  * the case of read locks.
5767  */
5768 static void
5769 nfsrv_localunlock(vnode_t vp, struct nfslockfile *lfp, uint64_t init_first,
5770     uint64_t init_end, NFSPROC_T *p)
5771 {
5772 	struct nfslock *lop;
5773 	uint64_t first, end, prevfirst __unused;
5774 
5775 	first = init_first;
5776 	end = init_end;
5777 	while (first < init_end) {
5778 		/* Loop through all nfs locks, adjusting first and end */
5779 		prevfirst = 0;
5780 		LIST_FOREACH(lop, &lfp->lf_lock, lo_lckfile) {
5781 			KASSERT(prevfirst <= lop->lo_first,
5782 			    ("nfsv4 locks out of order"));
5783 			KASSERT(lop->lo_first < lop->lo_end,
5784 			    ("nfsv4 bogus lock"));
5785 			prevfirst = lop->lo_first;
5786 			if (first >= lop->lo_first &&
5787 			    first < lop->lo_end)
5788 				/*
5789 				 * Overlaps with initial part, so trim
5790 				 * off that initial part by moving first past
5791 				 * it.
5792 				 */
5793 				first = lop->lo_end;
5794 			else if (end > lop->lo_first &&
5795 			    lop->lo_first > first) {
5796 				/*
5797 				 * This lock defines the end of the
5798 				 * segment to unlock, so set end to the
5799 				 * start of it and break out of the loop.
5800 				 */
5801 				end = lop->lo_first;
5802 				break;
5803 			}
5804 			if (first >= end)
5805 				/*
5806 				 * There is no segment left to do, so
5807 				 * break out of this loop and then exit
5808 				 * the outer while() since first will be set
5809 				 * to end, which must equal init_end here.
5810 				 */
5811 				break;
5812 		}
5813 		if (first < end) {
5814 			/* Unlock this segment */
5815 			(void) nfsrv_dolocal(vp, lfp, NFSLCK_UNLOCK,
5816 			    NFSLCK_READ, first, end, NULL, p);
5817 			nfsrv_locallock_commit(lfp, NFSLCK_UNLOCK,
5818 			    first, end);
5819 		}
5820 		/*
5821 		 * Now move past this segment and look for any further
5822 		 * segment in the range, if there is one.
5823 		 */
5824 		first = end;
5825 		end = init_end;
5826 	}
5827 }
5828 
5829 /*
5830  * Do the local lock operation and update the rollback list, as required.
5831  * Perform the rollback and return the error if nfsvno_advlock() fails.
5832  */
5833 static int
5834 nfsrv_dolocal(vnode_t vp, struct nfslockfile *lfp, int flags, int oldflags,
5835     uint64_t first, uint64_t end, struct nfslockconflict *cfp, NFSPROC_T *p)
5836 {
5837 	struct nfsrollback *rlp;
5838 	int error = 0, ltype, oldltype;
5839 
5840 	if (flags & NFSLCK_WRITE)
5841 		ltype = F_WRLCK;
5842 	else if (flags & NFSLCK_READ)
5843 		ltype = F_RDLCK;
5844 	else
5845 		ltype = F_UNLCK;
5846 	if (oldflags & NFSLCK_WRITE)
5847 		oldltype = F_WRLCK;
5848 	else if (oldflags & NFSLCK_READ)
5849 		oldltype = F_RDLCK;
5850 	else
5851 		oldltype = F_UNLCK;
5852 	if (ltype == oldltype || (oldltype == F_WRLCK && ltype == F_RDLCK))
5853 		/* nothing to do */
5854 		goto out;
5855 	error = nfsvno_advlock(vp, ltype, first, end, p);
5856 	if (error != 0) {
5857 		if (cfp != NULL) {
5858 			cfp->cl_clientid.lval[0] = 0;
5859 			cfp->cl_clientid.lval[1] = 0;
5860 			cfp->cl_first = 0;
5861 			cfp->cl_end = NFS64BITSSET;
5862 			cfp->cl_flags = NFSLCK_WRITE;
5863 			cfp->cl_ownerlen = 5;
5864 			NFSBCOPY("LOCAL", cfp->cl_owner, 5);
5865 		}
5866 		nfsrv_locallock_rollback(vp, lfp, p);
5867 	} else if (ltype != F_UNLCK) {
5868 		rlp = malloc(sizeof (struct nfsrollback), M_NFSDROLLBACK,
5869 		    M_WAITOK);
5870 		rlp->rlck_first = first;
5871 		rlp->rlck_end = end;
5872 		rlp->rlck_type = oldltype;
5873 		LIST_INSERT_HEAD(&lfp->lf_rollback, rlp, rlck_list);
5874 	}
5875 
5876 out:
5877 	NFSEXITCODE(error);
5878 	return (error);
5879 }
5880 
5881 /*
5882  * Roll back local lock changes and free up the rollback list.
5883  */
5884 static void
5885 nfsrv_locallock_rollback(vnode_t vp, struct nfslockfile *lfp, NFSPROC_T *p)
5886 {
5887 	struct nfsrollback *rlp, *nrlp;
5888 
5889 	LIST_FOREACH_SAFE(rlp, &lfp->lf_rollback, rlck_list, nrlp) {
5890 		(void) nfsvno_advlock(vp, rlp->rlck_type, rlp->rlck_first,
5891 		    rlp->rlck_end, p);
5892 		free(rlp, M_NFSDROLLBACK);
5893 	}
5894 	LIST_INIT(&lfp->lf_rollback);
5895 }
5896 
5897 /*
5898  * Update local lock list and delete rollback list (ie now committed to the
5899  * local locks). Most of the work is done by the internal function.
5900  */
5901 static void
5902 nfsrv_locallock_commit(struct nfslockfile *lfp, int flags, uint64_t first,
5903     uint64_t end)
5904 {
5905 	struct nfsrollback *rlp, *nrlp;
5906 	struct nfslock *new_lop, *other_lop;
5907 
5908 	new_lop = malloc(sizeof (struct nfslock), M_NFSDLOCK, M_WAITOK);
5909 	if (flags & (NFSLCK_READ | NFSLCK_WRITE))
5910 		other_lop = malloc(sizeof (struct nfslock), M_NFSDLOCK,
5911 		    M_WAITOK);
5912 	else
5913 		other_lop = NULL;
5914 	new_lop->lo_flags = flags;
5915 	new_lop->lo_first = first;
5916 	new_lop->lo_end = end;
5917 	nfsrv_updatelock(NULL, &new_lop, &other_lop, lfp);
5918 	if (new_lop != NULL)
5919 		free(new_lop, M_NFSDLOCK);
5920 	if (other_lop != NULL)
5921 		free(other_lop, M_NFSDLOCK);
5922 
5923 	/* and get rid of the rollback list */
5924 	LIST_FOREACH_SAFE(rlp, &lfp->lf_rollback, rlck_list, nrlp)
5925 		free(rlp, M_NFSDROLLBACK);
5926 	LIST_INIT(&lfp->lf_rollback);
5927 }
5928 
5929 /*
5930  * Lock the struct nfslockfile for local lock updating.
5931  */
5932 static void
5933 nfsrv_locklf(struct nfslockfile *lfp)
5934 {
5935 	int gotlock;
5936 
5937 	/* lf_usecount ensures *lfp won't be free'd */
5938 	lfp->lf_usecount++;
5939 	do {
5940 		gotlock = nfsv4_lock(&lfp->lf_locallock_lck, 1, NULL,
5941 		    NFSSTATEMUTEXPTR, NULL);
5942 	} while (gotlock == 0);
5943 	lfp->lf_usecount--;
5944 }
5945 
5946 /*
5947  * Unlock the struct nfslockfile after local lock updating.
5948  */
5949 static void
5950 nfsrv_unlocklf(struct nfslockfile *lfp)
5951 {
5952 
5953 	nfsv4_unlock(&lfp->lf_locallock_lck, 0);
5954 }
5955 
5956 /*
5957  * Clear out all state for the NFSv4 server.
5958  * Must be called by a thread that can sleep when no nfsds are running.
5959  */
5960 void
5961 nfsrv_throwawayallstate(NFSPROC_T *p)
5962 {
5963 	struct nfsclient *clp, *nclp;
5964 	struct nfslockfile *lfp, *nlfp;
5965 	int i;
5966 
5967 	/*
5968 	 * For each client, clean out the state and then free the structure.
5969 	 */
5970 	for (i = 0; i < nfsrv_clienthashsize; i++) {
5971 		LIST_FOREACH_SAFE(clp, &nfsclienthash[i], lc_hash, nclp) {
5972 			nfsrv_cleanclient(clp, p);
5973 			nfsrv_freedeleglist(&clp->lc_deleg);
5974 			nfsrv_freedeleglist(&clp->lc_olddeleg);
5975 			free(clp->lc_stateid, M_NFSDCLIENT);
5976 			free(clp, M_NFSDCLIENT);
5977 		}
5978 	}
5979 
5980 	/*
5981 	 * Also, free up any remaining lock file structures.
5982 	 */
5983 	for (i = 0; i < nfsrv_lockhashsize; i++) {
5984 		LIST_FOREACH_SAFE(lfp, &nfslockhash[i], lf_hash, nlfp) {
5985 			printf("nfsd unload: fnd a lock file struct\n");
5986 			nfsrv_freenfslockfile(lfp);
5987 		}
5988 	}
5989 
5990 	/* And get rid of the deviceid structures and layouts. */
5991 	nfsrv_freealllayoutsanddevids();
5992 }
5993 
5994 /*
5995  * Check the sequence# for the session and slot provided as an argument.
5996  * Also, renew the lease if the session will return NFS_OK.
5997  */
5998 int
5999 nfsrv_checksequence(struct nfsrv_descript *nd, uint32_t sequenceid,
6000     uint32_t *highest_slotidp, uint32_t *target_highest_slotidp, int cache_this,
6001     uint32_t *sflagsp, NFSPROC_T *p)
6002 {
6003 	struct nfsdsession *sep;
6004 	struct nfssessionhash *shp;
6005 	int error;
6006 	SVCXPRT *savxprt;
6007 
6008 	shp = NFSSESSIONHASH(nd->nd_sessionid);
6009 	NFSLOCKSESSION(shp);
6010 	sep = nfsrv_findsession(nd->nd_sessionid);
6011 	if (sep == NULL) {
6012 		NFSUNLOCKSESSION(shp);
6013 		return (NFSERR_BADSESSION);
6014 	}
6015 	error = nfsv4_seqsession(sequenceid, nd->nd_slotid, *highest_slotidp,
6016 	    sep->sess_slots, NULL, NFSV4_SLOTS - 1);
6017 	if (error != 0) {
6018 		NFSUNLOCKSESSION(shp);
6019 		return (error);
6020 	}
6021 	if (cache_this != 0)
6022 		nd->nd_flag |= ND_SAVEREPLY;
6023 	/* Renew the lease. */
6024 	sep->sess_clp->lc_expiry = nfsrv_leaseexpiry();
6025 	nd->nd_clientid.qval = sep->sess_clp->lc_clientid.qval;
6026 	nd->nd_flag |= ND_IMPLIEDCLID;
6027 
6028 	/*
6029 	 * If this session handles the backchannel, save the nd_xprt for this
6030 	 * RPC, since this is the one being used.
6031 	 * RFC-5661 specifies that the fore channel will be implicitly
6032 	 * bound by a Sequence operation.  However, since some NFSv4.1 clients
6033 	 * erroneously assumed that the back channel would be implicitly
6034 	 * bound as well, do the implicit binding unless a
6035 	 * BindConnectiontoSession has already been done on the session.
6036 	 */
6037 	if (sep->sess_clp->lc_req.nr_client != NULL &&
6038 	    sep->sess_cbsess.nfsess_xprt != nd->nd_xprt &&
6039 	    (sep->sess_crflags & NFSV4CRSESS_CONNBACKCHAN) != 0 &&
6040 	    (sep->sess_clp->lc_flags & LCL_DONEBINDCONN) == 0) {
6041 		NFSD_DEBUG(2,
6042 		    "nfsrv_checksequence: implicit back channel bind\n");
6043 		savxprt = sep->sess_cbsess.nfsess_xprt;
6044 		SVC_ACQUIRE(nd->nd_xprt);
6045 		nd->nd_xprt->xp_p2 =
6046 		    sep->sess_clp->lc_req.nr_client->cl_private;
6047 		nd->nd_xprt->xp_idletimeout = 0;	/* Disable timeout. */
6048 		sep->sess_cbsess.nfsess_xprt = nd->nd_xprt;
6049 		if (savxprt != NULL)
6050 			SVC_RELEASE(savxprt);
6051 	}
6052 
6053 	*sflagsp = 0;
6054 	if (sep->sess_clp->lc_req.nr_client == NULL)
6055 		*sflagsp |= NFSV4SEQ_CBPATHDOWN;
6056 	NFSUNLOCKSESSION(shp);
6057 	if (error == NFSERR_EXPIRED) {
6058 		*sflagsp |= NFSV4SEQ_EXPIREDALLSTATEREVOKED;
6059 		error = 0;
6060 	} else if (error == NFSERR_ADMINREVOKED) {
6061 		*sflagsp |= NFSV4SEQ_ADMINSTATEREVOKED;
6062 		error = 0;
6063 	}
6064 	*highest_slotidp = *target_highest_slotidp = NFSV4_SLOTS - 1;
6065 	return (0);
6066 }
6067 
6068 /*
6069  * Check/set reclaim complete for this session/clientid.
6070  */
6071 int
6072 nfsrv_checkreclaimcomplete(struct nfsrv_descript *nd, int onefs)
6073 {
6074 	struct nfsdsession *sep;
6075 	struct nfssessionhash *shp;
6076 	int error = 0;
6077 
6078 	shp = NFSSESSIONHASH(nd->nd_sessionid);
6079 	NFSLOCKSTATE();
6080 	NFSLOCKSESSION(shp);
6081 	sep = nfsrv_findsession(nd->nd_sessionid);
6082 	if (sep == NULL) {
6083 		NFSUNLOCKSESSION(shp);
6084 		NFSUNLOCKSTATE();
6085 		return (NFSERR_BADSESSION);
6086 	}
6087 
6088 	if (onefs != 0)
6089 		sep->sess_clp->lc_flags |= LCL_RECLAIMONEFS;
6090 		/* Check to see if reclaim complete has already happened. */
6091 	else if ((sep->sess_clp->lc_flags & LCL_RECLAIMCOMPLETE) != 0)
6092 		error = NFSERR_COMPLETEALREADY;
6093 	else {
6094 		sep->sess_clp->lc_flags |= LCL_RECLAIMCOMPLETE;
6095 		nfsrv_markreclaim(sep->sess_clp);
6096 	}
6097 	NFSUNLOCKSESSION(shp);
6098 	NFSUNLOCKSTATE();
6099 	return (error);
6100 }
6101 
6102 /*
6103  * Cache the reply in a session slot.
6104  */
6105 void
6106 nfsrv_cache_session(uint8_t *sessionid, uint32_t slotid, int repstat,
6107    struct mbuf **m)
6108 {
6109 	struct nfsdsession *sep;
6110 	struct nfssessionhash *shp;
6111 
6112 	shp = NFSSESSIONHASH(sessionid);
6113 	NFSLOCKSESSION(shp);
6114 	sep = nfsrv_findsession(sessionid);
6115 	if (sep == NULL) {
6116 		NFSUNLOCKSESSION(shp);
6117 		printf("nfsrv_cache_session: no session\n");
6118 		m_freem(*m);
6119 		return;
6120 	}
6121 	nfsv4_seqsess_cacherep(slotid, sep->sess_slots, repstat, m);
6122 	NFSUNLOCKSESSION(shp);
6123 }
6124 
6125 /*
6126  * Search for a session that matches the sessionid.
6127  */
6128 static struct nfsdsession *
6129 nfsrv_findsession(uint8_t *sessionid)
6130 {
6131 	struct nfsdsession *sep;
6132 	struct nfssessionhash *shp;
6133 
6134 	shp = NFSSESSIONHASH(sessionid);
6135 	LIST_FOREACH(sep, &shp->list, sess_hash) {
6136 		if (!NFSBCMP(sessionid, sep->sess_sessionid, NFSX_V4SESSIONID))
6137 			break;
6138 	}
6139 	return (sep);
6140 }
6141 
6142 /*
6143  * Destroy a session.
6144  */
6145 int
6146 nfsrv_destroysession(struct nfsrv_descript *nd, uint8_t *sessionid)
6147 {
6148 	int error, igotlock, samesess;
6149 
6150 	samesess = 0;
6151 	if (!NFSBCMP(sessionid, nd->nd_sessionid, NFSX_V4SESSIONID) &&
6152 	    (nd->nd_flag & ND_HASSEQUENCE) != 0) {
6153 		samesess = 1;
6154 		if ((nd->nd_flag & ND_LASTOP) == 0)
6155 			return (NFSERR_BADSESSION);
6156 	}
6157 
6158 	/* Lock out other nfsd threads */
6159 	NFSLOCKV4ROOTMUTEX();
6160 	nfsv4_relref(&nfsv4rootfs_lock);
6161 	do {
6162 		igotlock = nfsv4_lock(&nfsv4rootfs_lock, 1, NULL,
6163 		    NFSV4ROOTLOCKMUTEXPTR, NULL);
6164 	} while (igotlock == 0);
6165 	NFSUNLOCKV4ROOTMUTEX();
6166 
6167 	error = nfsrv_freesession(NULL, sessionid);
6168 	if (error == 0 && samesess != 0)
6169 		nd->nd_flag &= ~ND_HASSEQUENCE;
6170 
6171 	NFSLOCKV4ROOTMUTEX();
6172 	nfsv4_unlock(&nfsv4rootfs_lock, 1);
6173 	NFSUNLOCKV4ROOTMUTEX();
6174 	return (error);
6175 }
6176 
6177 /*
6178  * Bind a connection to a session.
6179  * For now, only certain variants are supported, since the current session
6180  * structure can only handle a single backchannel entry, which will be
6181  * applied to all connections if it is set.
6182  */
6183 int
6184 nfsrv_bindconnsess(struct nfsrv_descript *nd, uint8_t *sessionid, int *foreaftp)
6185 {
6186 	struct nfssessionhash *shp;
6187 	struct nfsdsession *sep;
6188 	struct nfsclient *clp;
6189 	SVCXPRT *savxprt;
6190 	int error;
6191 
6192 	error = 0;
6193 	shp = NFSSESSIONHASH(sessionid);
6194 	NFSLOCKSTATE();
6195 	NFSLOCKSESSION(shp);
6196 	sep = nfsrv_findsession(sessionid);
6197 	if (sep != NULL) {
6198 		clp = sep->sess_clp;
6199 		if (*foreaftp == NFSCDFC4_BACK ||
6200 		    *foreaftp == NFSCDFC4_BACK_OR_BOTH ||
6201 		    *foreaftp == NFSCDFC4_FORE_OR_BOTH) {
6202 			/* Try to set up a backchannel. */
6203 			if (clp->lc_req.nr_client == NULL) {
6204 				NFSD_DEBUG(2, "nfsrv_bindconnsess: acquire "
6205 				    "backchannel\n");
6206 				clp->lc_req.nr_client = (struct __rpc_client *)
6207 				    clnt_bck_create(nd->nd_xprt->xp_socket,
6208 				    sep->sess_cbprogram, NFSV4_CBVERS);
6209 			}
6210 			if (clp->lc_req.nr_client != NULL) {
6211 				NFSD_DEBUG(2, "nfsrv_bindconnsess: set up "
6212 				    "backchannel\n");
6213 				savxprt = sep->sess_cbsess.nfsess_xprt;
6214 				SVC_ACQUIRE(nd->nd_xprt);
6215 				nd->nd_xprt->xp_p2 =
6216 				    clp->lc_req.nr_client->cl_private;
6217 				/* Disable idle timeout. */
6218 				nd->nd_xprt->xp_idletimeout = 0;
6219 				sep->sess_cbsess.nfsess_xprt = nd->nd_xprt;
6220 				if (savxprt != NULL)
6221 					SVC_RELEASE(savxprt);
6222 				sep->sess_crflags |= NFSV4CRSESS_CONNBACKCHAN;
6223 				clp->lc_flags |= LCL_DONEBINDCONN;
6224 				if (*foreaftp == NFSCDFS4_BACK)
6225 					*foreaftp = NFSCDFS4_BACK;
6226 				else
6227 					*foreaftp = NFSCDFS4_BOTH;
6228 			} else if (*foreaftp != NFSCDFC4_BACK) {
6229 				NFSD_DEBUG(2, "nfsrv_bindconnsess: can't set "
6230 				    "up backchannel\n");
6231 				sep->sess_crflags &= ~NFSV4CRSESS_CONNBACKCHAN;
6232 				clp->lc_flags |= LCL_DONEBINDCONN;
6233 				*foreaftp = NFSCDFS4_FORE;
6234 			} else {
6235 				error = NFSERR_NOTSUPP;
6236 				printf("nfsrv_bindconnsess: Can't add "
6237 				    "backchannel\n");
6238 			}
6239 		} else {
6240 			NFSD_DEBUG(2, "nfsrv_bindconnsess: Set forechannel\n");
6241 			clp->lc_flags |= LCL_DONEBINDCONN;
6242 			*foreaftp = NFSCDFS4_FORE;
6243 		}
6244 	} else
6245 		error = NFSERR_BADSESSION;
6246 	NFSUNLOCKSESSION(shp);
6247 	NFSUNLOCKSTATE();
6248 	return (error);
6249 }
6250 
6251 /*
6252  * Free up a session structure.
6253  */
6254 static int
6255 nfsrv_freesession(struct nfsdsession *sep, uint8_t *sessionid)
6256 {
6257 	struct nfssessionhash *shp;
6258 	int i;
6259 
6260 	NFSLOCKSTATE();
6261 	if (sep == NULL) {
6262 		shp = NFSSESSIONHASH(sessionid);
6263 		NFSLOCKSESSION(shp);
6264 		sep = nfsrv_findsession(sessionid);
6265 	} else {
6266 		shp = NFSSESSIONHASH(sep->sess_sessionid);
6267 		NFSLOCKSESSION(shp);
6268 	}
6269 	if (sep != NULL) {
6270 		sep->sess_refcnt--;
6271 		if (sep->sess_refcnt > 0) {
6272 			NFSUNLOCKSESSION(shp);
6273 			NFSUNLOCKSTATE();
6274 			return (NFSERR_BACKCHANBUSY);
6275 		}
6276 		LIST_REMOVE(sep, sess_hash);
6277 		LIST_REMOVE(sep, sess_list);
6278 	}
6279 	NFSUNLOCKSESSION(shp);
6280 	NFSUNLOCKSTATE();
6281 	if (sep == NULL)
6282 		return (NFSERR_BADSESSION);
6283 	for (i = 0; i < NFSV4_SLOTS; i++)
6284 		if (sep->sess_slots[i].nfssl_reply != NULL)
6285 			m_freem(sep->sess_slots[i].nfssl_reply);
6286 	if (sep->sess_cbsess.nfsess_xprt != NULL)
6287 		SVC_RELEASE(sep->sess_cbsess.nfsess_xprt);
6288 	free(sep, M_NFSDSESSION);
6289 	return (0);
6290 }
6291 
6292 /*
6293  * Free a stateid.
6294  * RFC5661 says that it should fail when there are associated opens, locks
6295  * or delegations. Since stateids represent opens, I don't see how you can
6296  * free an open stateid (it will be free'd when closed), so this function
6297  * only works for lock stateids (freeing the lock_owner) or delegations.
6298  */
6299 int
6300 nfsrv_freestateid(struct nfsrv_descript *nd, nfsv4stateid_t *stateidp,
6301     NFSPROC_T *p)
6302 {
6303 	struct nfsclient *clp;
6304 	struct nfsstate *stp;
6305 	int error;
6306 
6307 	NFSLOCKSTATE();
6308 	/*
6309 	 * Look up the stateid
6310 	 */
6311 	error = nfsrv_getclient((nfsquad_t)((u_quad_t)0), CLOPS_RENEW, &clp,
6312 	    NULL, (nfsquad_t)((u_quad_t)0), 0, nd, p);
6313 	if (error == 0) {
6314 		/* First, check for a delegation. */
6315 		LIST_FOREACH(stp, &clp->lc_deleg, ls_list) {
6316 			if (!NFSBCMP(stp->ls_stateid.other, stateidp->other,
6317 			    NFSX_STATEIDOTHER))
6318 				break;
6319 		}
6320 		if (stp != NULL) {
6321 			nfsrv_freedeleg(stp);
6322 			NFSUNLOCKSTATE();
6323 			return (error);
6324 		}
6325 	}
6326 	/* Not a delegation, try for a lock_owner. */
6327 	if (error == 0)
6328 		error = nfsrv_getstate(clp, stateidp, 0, &stp);
6329 	if (error == 0 && ((stp->ls_flags & (NFSLCK_OPEN | NFSLCK_DELEGREAD |
6330 	    NFSLCK_DELEGWRITE)) != 0 || (stp->ls_flags & NFSLCK_LOCK) == 0))
6331 		/* Not a lock_owner stateid. */
6332 		error = NFSERR_LOCKSHELD;
6333 	if (error == 0 && !LIST_EMPTY(&stp->ls_lock))
6334 		error = NFSERR_LOCKSHELD;
6335 	if (error == 0)
6336 		nfsrv_freelockowner(stp, NULL, 0, p);
6337 	NFSUNLOCKSTATE();
6338 	return (error);
6339 }
6340 
6341 /*
6342  * Test a stateid.
6343  */
6344 int
6345 nfsrv_teststateid(struct nfsrv_descript *nd, nfsv4stateid_t *stateidp,
6346     NFSPROC_T *p)
6347 {
6348 	struct nfsclient *clp;
6349 	struct nfsstate *stp;
6350 	int error;
6351 
6352 	NFSLOCKSTATE();
6353 	/*
6354 	 * Look up the stateid
6355 	 */
6356 	error = nfsrv_getclient((nfsquad_t)((u_quad_t)0), CLOPS_RENEW, &clp,
6357 	    NULL, (nfsquad_t)((u_quad_t)0), 0, nd, p);
6358 	if (error == 0)
6359 		error = nfsrv_getstate(clp, stateidp, 0, &stp);
6360 	if (error == 0 && stateidp->seqid != 0 &&
6361 	    SEQ_LT(stateidp->seqid, stp->ls_stateid.seqid))
6362 		error = NFSERR_OLDSTATEID;
6363 	NFSUNLOCKSTATE();
6364 	return (error);
6365 }
6366 
6367 /*
6368  * Generate the xdr for an NFSv4.1 CBSequence Operation.
6369  */
6370 static int
6371 nfsv4_setcbsequence(struct nfsrv_descript *nd, struct nfsclient *clp,
6372     int dont_replycache, struct nfsdsession **sepp)
6373 {
6374 	struct nfsdsession *sep;
6375 	uint32_t *tl, slotseq = 0;
6376 	int maxslot, slotpos;
6377 	uint8_t sessionid[NFSX_V4SESSIONID];
6378 	int error;
6379 
6380 	error = nfsv4_getcbsession(clp, sepp);
6381 	if (error != 0)
6382 		return (error);
6383 	sep = *sepp;
6384 	(void)nfsv4_sequencelookup(NULL, &sep->sess_cbsess, &slotpos, &maxslot,
6385 	    &slotseq, sessionid);
6386 	KASSERT(maxslot >= 0, ("nfsv4_setcbsequence neg maxslot"));
6387 
6388 	/* Build the Sequence arguments. */
6389 	NFSM_BUILD(tl, uint32_t *, NFSX_V4SESSIONID + 5 * NFSX_UNSIGNED);
6390 	bcopy(sessionid, tl, NFSX_V4SESSIONID);
6391 	tl += NFSX_V4SESSIONID / NFSX_UNSIGNED;
6392 	nd->nd_slotseq = tl;
6393 	*tl++ = txdr_unsigned(slotseq);
6394 	*tl++ = txdr_unsigned(slotpos);
6395 	*tl++ = txdr_unsigned(maxslot);
6396 	if (dont_replycache == 0)
6397 		*tl++ = newnfs_true;
6398 	else
6399 		*tl++ = newnfs_false;
6400 	*tl = 0;			/* No referring call list, for now. */
6401 	nd->nd_flag |= ND_HASSEQUENCE;
6402 	return (0);
6403 }
6404 
6405 /*
6406  * Get a session for the callback.
6407  */
6408 static int
6409 nfsv4_getcbsession(struct nfsclient *clp, struct nfsdsession **sepp)
6410 {
6411 	struct nfsdsession *sep;
6412 
6413 	NFSLOCKSTATE();
6414 	LIST_FOREACH(sep, &clp->lc_session, sess_list) {
6415 		if ((sep->sess_crflags & NFSV4CRSESS_CONNBACKCHAN) != 0)
6416 			break;
6417 	}
6418 	if (sep == NULL) {
6419 		NFSUNLOCKSTATE();
6420 		return (NFSERR_BADSESSION);
6421 	}
6422 	sep->sess_refcnt++;
6423 	*sepp = sep;
6424 	NFSUNLOCKSTATE();
6425 	return (0);
6426 }
6427 
6428 /*
6429  * Free up all backchannel xprts.  This needs to be done when the nfsd threads
6430  * exit, since those transports will all be going away.
6431  * This is only called after all the nfsd threads are done performing RPCs,
6432  * so locking shouldn't be an issue.
6433  */
6434 APPLESTATIC void
6435 nfsrv_freeallbackchannel_xprts(void)
6436 {
6437 	struct nfsdsession *sep;
6438 	struct nfsclient *clp;
6439 	SVCXPRT *xprt;
6440 	int i;
6441 
6442 	for (i = 0; i < nfsrv_clienthashsize; i++) {
6443 		LIST_FOREACH(clp, &nfsclienthash[i], lc_hash) {
6444 			LIST_FOREACH(sep, &clp->lc_session, sess_list) {
6445 				xprt = sep->sess_cbsess.nfsess_xprt;
6446 				sep->sess_cbsess.nfsess_xprt = NULL;
6447 				if (xprt != NULL)
6448 					SVC_RELEASE(xprt);
6449 			}
6450 		}
6451 	}
6452 }
6453 
6454 /*
6455  * Do a layout commit.  Actually just call nfsrv_updatemdsattr().
6456  * I have no idea if the rest of these arguments will ever be useful?
6457  */
6458 int
6459 nfsrv_layoutcommit(struct nfsrv_descript *nd, vnode_t vp, int layouttype,
6460     int hasnewoff, uint64_t newoff, uint64_t offset, uint64_t len,
6461     int hasnewmtime, struct timespec *newmtimep, int reclaim,
6462     nfsv4stateid_t *stateidp, int maxcnt, char *layp, int *hasnewsizep,
6463     uint64_t *newsizep, struct ucred *cred, NFSPROC_T *p)
6464 {
6465 	struct nfsvattr na;
6466 	int error;
6467 
6468 	error = nfsrv_updatemdsattr(vp, &na, p);
6469 	if (error == 0) {
6470 		*hasnewsizep = 1;
6471 		*newsizep = na.na_size;
6472 	}
6473 	return (error);
6474 }
6475 
6476 /*
6477  * Try and get a layout.
6478  */
6479 int
6480 nfsrv_layoutget(struct nfsrv_descript *nd, vnode_t vp, struct nfsexstuff *exp,
6481     int layouttype, int *iomode, uint64_t *offset, uint64_t *len,
6482     uint64_t minlen, nfsv4stateid_t *stateidp, int maxcnt, int *retonclose,
6483     int *layoutlenp, char *layp, struct ucred *cred, NFSPROC_T *p)
6484 {
6485 	struct nfslayouthash *lhyp;
6486 	struct nfslayout *lyp;
6487 	char *devid;
6488 	fhandle_t fh, *dsfhp;
6489 	int error, mirrorcnt;
6490 
6491 	if (nfsrv_devidcnt == 0)
6492 		return (NFSERR_UNKNLAYOUTTYPE);
6493 
6494 	if (*offset != 0)
6495 		printf("nfsrv_layoutget: off=%ju len=%ju\n", (uintmax_t)*offset,
6496 		    (uintmax_t)*len);
6497 	error = nfsvno_getfh(vp, &fh, p);
6498 	NFSD_DEBUG(4, "layoutget getfh=%d\n", error);
6499 	if (error != 0)
6500 		return (error);
6501 
6502 	/*
6503 	 * For now, all layouts are for entire files.
6504 	 * Only issue Read/Write layouts if requested for a non-readonly fs.
6505 	 */
6506 	if (NFSVNO_EXRDONLY(exp)) {
6507 		if (*iomode == NFSLAYOUTIOMODE_RW)
6508 			return (NFSERR_LAYOUTTRYLATER);
6509 		*iomode = NFSLAYOUTIOMODE_READ;
6510 	}
6511 	if (*iomode != NFSLAYOUTIOMODE_RW)
6512 		*iomode = NFSLAYOUTIOMODE_READ;
6513 
6514 	/*
6515 	 * Check to see if a write layout can be issued for this file.
6516 	 * This is used during mirror recovery to avoid RW layouts being
6517 	 * issued for a file while it is being copied to the recovered
6518 	 * mirror.
6519 	 */
6520 	if (*iomode == NFSLAYOUTIOMODE_RW && nfsrv_dontlayout(&fh) != 0)
6521 		return (NFSERR_LAYOUTTRYLATER);
6522 
6523 	*retonclose = 0;
6524 	*offset = 0;
6525 	*len = UINT64_MAX;
6526 
6527 	/* First, see if a layout already exists and return if found. */
6528 	lhyp = NFSLAYOUTHASH(&fh);
6529 	NFSLOCKLAYOUT(lhyp);
6530 	error = nfsrv_findlayout(&nd->nd_clientid, &fh, layouttype, p, &lyp);
6531 	NFSD_DEBUG(4, "layoutget findlay=%d\n", error);
6532 	/*
6533 	 * Not sure if the seqid must be the same, so I won't check it.
6534 	 */
6535 	if (error == 0 && (stateidp->other[0] != lyp->lay_stateid.other[0] ||
6536 	    stateidp->other[1] != lyp->lay_stateid.other[1] ||
6537 	    stateidp->other[2] != lyp->lay_stateid.other[2])) {
6538 		if ((lyp->lay_flags & NFSLAY_CALLB) == 0) {
6539 			NFSUNLOCKLAYOUT(lhyp);
6540 			NFSD_DEBUG(1, "ret bad stateid\n");
6541 			return (NFSERR_BADSTATEID);
6542 		}
6543 		/*
6544 		 * I believe we get here because there is a race between
6545 		 * the client processing the CBLAYOUTRECALL and the layout
6546 		 * being deleted here on the server.
6547 		 * The client has now done a LayoutGet with a non-layout
6548 		 * stateid, as it would when there is no layout.
6549 		 * As such, free this layout and set error == NFSERR_BADSTATEID
6550 		 * so the code below will create a new layout structure as
6551 		 * would happen if no layout was found.
6552 		 * "lyp" will be set before being used below, but set it NULL
6553 		 * as a safety belt.
6554 		 */
6555 		nfsrv_freelayout(&lhyp->list, lyp);
6556 		lyp = NULL;
6557 		error = NFSERR_BADSTATEID;
6558 	}
6559 	if (error == 0) {
6560 		if (lyp->lay_layoutlen > maxcnt) {
6561 			NFSUNLOCKLAYOUT(lhyp);
6562 			NFSD_DEBUG(1, "ret layout too small\n");
6563 			return (NFSERR_TOOSMALL);
6564 		}
6565 		if (*iomode == NFSLAYOUTIOMODE_RW)
6566 			lyp->lay_flags |= NFSLAY_RW;
6567 		else
6568 			lyp->lay_flags |= NFSLAY_READ;
6569 		NFSBCOPY(lyp->lay_xdr, layp, lyp->lay_layoutlen);
6570 		*layoutlenp = lyp->lay_layoutlen;
6571 		if (++lyp->lay_stateid.seqid == 0)
6572 			lyp->lay_stateid.seqid = 1;
6573 		stateidp->seqid = lyp->lay_stateid.seqid;
6574 		NFSUNLOCKLAYOUT(lhyp);
6575 		NFSD_DEBUG(4, "ret fnd layout\n");
6576 		return (0);
6577 	}
6578 	NFSUNLOCKLAYOUT(lhyp);
6579 
6580 	/* Find the device id and file handle. */
6581 	dsfhp = malloc(sizeof(fhandle_t) * NFSDEV_MAXMIRRORS, M_TEMP, M_WAITOK);
6582 	devid = malloc(NFSX_V4DEVICEID * NFSDEV_MAXMIRRORS, M_TEMP, M_WAITOK);
6583 	error = nfsrv_dsgetdevandfh(vp, p, &mirrorcnt, dsfhp, devid);
6584 	NFSD_DEBUG(4, "layoutget devandfh=%d\n", error);
6585 	if (error == 0) {
6586 		if (layouttype == NFSLAYOUT_NFSV4_1_FILES) {
6587 			if (NFSX_V4FILELAYOUT > maxcnt)
6588 				error = NFSERR_TOOSMALL;
6589 			else
6590 				lyp = nfsrv_filelayout(nd, *iomode, &fh, dsfhp,
6591 				    devid, vp->v_mount->mnt_stat.f_fsid);
6592 		} else {
6593 			if (NFSX_V4FLEXLAYOUT(mirrorcnt) > maxcnt)
6594 				error = NFSERR_TOOSMALL;
6595 			else
6596 				lyp = nfsrv_flexlayout(nd, *iomode, mirrorcnt,
6597 				    &fh, dsfhp, devid,
6598 				    vp->v_mount->mnt_stat.f_fsid);
6599 		}
6600 	}
6601 	free(dsfhp, M_TEMP);
6602 	free(devid, M_TEMP);
6603 	if (error != 0)
6604 		return (error);
6605 
6606 	/*
6607 	 * Now, add this layout to the list.
6608 	 */
6609 	error = nfsrv_addlayout(nd, &lyp, stateidp, layp, layoutlenp, p);
6610 	NFSD_DEBUG(4, "layoutget addl=%d\n", error);
6611 	/*
6612 	 * The lyp will be set to NULL by nfsrv_addlayout() if it
6613 	 * linked the new structure into the lists.
6614 	 */
6615 	free(lyp, M_NFSDSTATE);
6616 	return (error);
6617 }
6618 
6619 /*
6620  * Generate a File Layout.
6621  */
6622 static struct nfslayout *
6623 nfsrv_filelayout(struct nfsrv_descript *nd, int iomode, fhandle_t *fhp,
6624     fhandle_t *dsfhp, char *devid, fsid_t fs)
6625 {
6626 	uint32_t *tl;
6627 	struct nfslayout *lyp;
6628 	uint64_t pattern_offset;
6629 
6630 	lyp = malloc(sizeof(struct nfslayout) + NFSX_V4FILELAYOUT, M_NFSDSTATE,
6631 	    M_WAITOK | M_ZERO);
6632 	lyp->lay_type = NFSLAYOUT_NFSV4_1_FILES;
6633 	if (iomode == NFSLAYOUTIOMODE_RW)
6634 		lyp->lay_flags = NFSLAY_RW;
6635 	else
6636 		lyp->lay_flags = NFSLAY_READ;
6637 	NFSBCOPY(fhp, &lyp->lay_fh, sizeof(*fhp));
6638 	lyp->lay_clientid.qval = nd->nd_clientid.qval;
6639 	lyp->lay_fsid = fs;
6640 
6641 	/* Fill in the xdr for the files layout. */
6642 	tl = (uint32_t *)lyp->lay_xdr;
6643 	NFSBCOPY(devid, tl, NFSX_V4DEVICEID);		/* Device ID. */
6644 	tl += (NFSX_V4DEVICEID / NFSX_UNSIGNED);
6645 
6646 	/*
6647 	 * Make the stripe size as many 64K blocks as will fit in the stripe
6648 	 * mask. Since there is only one stripe, the stripe size doesn't really
6649 	 * matter, except that the Linux client will only handle an exact
6650 	 * multiple of their PAGE_SIZE (usually 4K).  I chose 64K as a value
6651 	 * that should cover most/all arches w.r.t. PAGE_SIZE.
6652 	 */
6653 	*tl++ = txdr_unsigned(NFSFLAYUTIL_STRIPE_MASK & ~0xffff);
6654 	*tl++ = 0;					/* 1st stripe index. */
6655 	pattern_offset = 0;
6656 	txdr_hyper(pattern_offset, tl); tl += 2;	/* Pattern offset. */
6657 	*tl++ = txdr_unsigned(1);			/* 1 file handle. */
6658 	*tl++ = txdr_unsigned(NFSX_V4PNFSFH);
6659 	NFSBCOPY(dsfhp, tl, sizeof(*dsfhp));
6660 	lyp->lay_layoutlen = NFSX_V4FILELAYOUT;
6661 	return (lyp);
6662 }
6663 
6664 #define	FLEX_OWNERID	"999"
6665 #define	FLEX_UID0	"0"
6666 /*
6667  * Generate a Flex File Layout.
6668  * The FLEX_OWNERID can be any string of 3 decimal digits. Although this
6669  * string goes on the wire, it isn't supposed to be used by the client,
6670  * since this server uses tight coupling.
6671  * Although not recommended by the spec., if vfs.nfsd.flexlinuxhack=1 use
6672  * a string of "0". This works around the Linux Flex File Layout driver bug
6673  * which uses the synthetic uid/gid strings for the "tightly coupled" case.
6674  */
6675 static struct nfslayout *
6676 nfsrv_flexlayout(struct nfsrv_descript *nd, int iomode, int mirrorcnt,
6677     fhandle_t *fhp, fhandle_t *dsfhp, char *devid, fsid_t fs)
6678 {
6679 	uint32_t *tl;
6680 	struct nfslayout *lyp;
6681 	uint64_t lenval;
6682 	int i;
6683 
6684 	lyp = malloc(sizeof(struct nfslayout) + NFSX_V4FLEXLAYOUT(mirrorcnt),
6685 	    M_NFSDSTATE, M_WAITOK | M_ZERO);
6686 	lyp->lay_type = NFSLAYOUT_FLEXFILE;
6687 	if (iomode == NFSLAYOUTIOMODE_RW)
6688 		lyp->lay_flags = NFSLAY_RW;
6689 	else
6690 		lyp->lay_flags = NFSLAY_READ;
6691 	NFSBCOPY(fhp, &lyp->lay_fh, sizeof(*fhp));
6692 	lyp->lay_clientid.qval = nd->nd_clientid.qval;
6693 	lyp->lay_fsid = fs;
6694 	lyp->lay_mirrorcnt = mirrorcnt;
6695 
6696 	/* Fill in the xdr for the files layout. */
6697 	tl = (uint32_t *)lyp->lay_xdr;
6698 	lenval = 0;
6699 	txdr_hyper(lenval, tl); tl += 2;		/* Stripe unit. */
6700 	*tl++ = txdr_unsigned(mirrorcnt);		/* # of mirrors. */
6701 	for (i = 0; i < mirrorcnt; i++) {
6702 		*tl++ = txdr_unsigned(1);		/* One stripe. */
6703 		NFSBCOPY(devid, tl, NFSX_V4DEVICEID);	/* Device ID. */
6704 		tl += (NFSX_V4DEVICEID / NFSX_UNSIGNED);
6705 		devid += NFSX_V4DEVICEID;
6706 		*tl++ = txdr_unsigned(1);		/* Efficiency. */
6707 		*tl++ = 0;				/* Proxy Stateid. */
6708 		*tl++ = 0x55555555;
6709 		*tl++ = 0x55555555;
6710 		*tl++ = 0x55555555;
6711 		*tl++ = txdr_unsigned(1);		/* 1 file handle. */
6712 		*tl++ = txdr_unsigned(NFSX_V4PNFSFH);
6713 		NFSBCOPY(dsfhp, tl, sizeof(*dsfhp));
6714 		tl += (NFSM_RNDUP(NFSX_V4PNFSFH) / NFSX_UNSIGNED);
6715 		dsfhp++;
6716 		if (nfsrv_flexlinuxhack != 0) {
6717 			*tl++ = txdr_unsigned(strlen(FLEX_UID0));
6718 			*tl = 0;		/* 0 pad string. */
6719 			NFSBCOPY(FLEX_UID0, tl++, strlen(FLEX_UID0));
6720 			*tl++ = txdr_unsigned(strlen(FLEX_UID0));
6721 			*tl = 0;		/* 0 pad string. */
6722 			NFSBCOPY(FLEX_UID0, tl++, strlen(FLEX_UID0));
6723 		} else {
6724 			*tl++ = txdr_unsigned(strlen(FLEX_OWNERID));
6725 			NFSBCOPY(FLEX_OWNERID, tl++, NFSX_UNSIGNED);
6726 			*tl++ = txdr_unsigned(strlen(FLEX_OWNERID));
6727 			NFSBCOPY(FLEX_OWNERID, tl++, NFSX_UNSIGNED);
6728 		}
6729 	}
6730 	*tl++ = txdr_unsigned(0);		/* ff_flags. */
6731 	*tl = txdr_unsigned(60);		/* Status interval hint. */
6732 	lyp->lay_layoutlen = NFSX_V4FLEXLAYOUT(mirrorcnt);
6733 	return (lyp);
6734 }
6735 
6736 /*
6737  * Parse and process Flex File errors returned via LayoutReturn.
6738  */
6739 static void
6740 nfsrv_flexlayouterr(struct nfsrv_descript *nd, uint32_t *layp, int maxcnt,
6741     NFSPROC_T *p)
6742 {
6743 	uint32_t *tl;
6744 	int cnt, errcnt, i, j, opnum, stat;
6745 	char devid[NFSX_V4DEVICEID];
6746 
6747 	tl = layp;
6748 	cnt = fxdr_unsigned(int, *tl++);
6749 	NFSD_DEBUG(4, "flexlayouterr cnt=%d\n", cnt);
6750 	for (i = 0; i < cnt; i++) {
6751 		/* Skip offset, length and stateid for now. */
6752 		tl += (4 + NFSX_STATEID / NFSX_UNSIGNED);
6753 		errcnt = fxdr_unsigned(int, *tl++);
6754 		NFSD_DEBUG(4, "flexlayouterr errcnt=%d\n", errcnt);
6755 		for (j = 0; j < errcnt; j++) {
6756 			NFSBCOPY(tl, devid, NFSX_V4DEVICEID);
6757 			tl += (NFSX_V4DEVICEID / NFSX_UNSIGNED);
6758 			stat = fxdr_unsigned(int, *tl++);
6759 			opnum = fxdr_unsigned(int, *tl++);
6760 			NFSD_DEBUG(4, "flexlayouterr op=%d stat=%d\n", opnum,
6761 			    stat);
6762 			/*
6763 			 * Except for NFSERR_ACCES and NFSERR_STALE errors,
6764 			 * disable the mirror.
6765 			 */
6766 			if (stat != NFSERR_ACCES && stat != NFSERR_STALE)
6767 				nfsrv_delds(devid, p);
6768 		}
6769 	}
6770 }
6771 
6772 /*
6773  * This function removes all flex file layouts which has a mirror with
6774  * a device id that matches the argument.
6775  * Called when the DS represented by the device id has failed.
6776  */
6777 void
6778 nfsrv_flexmirrordel(char *devid, NFSPROC_T *p)
6779 {
6780 	uint32_t *tl;
6781 	struct nfslayout *lyp, *nlyp;
6782 	struct nfslayouthash *lhyp;
6783 	struct nfslayouthead loclyp;
6784 	int i, j;
6785 
6786 	NFSD_DEBUG(4, "flexmirrordel\n");
6787 	/* Move all layouts found onto a local list. */
6788 	TAILQ_INIT(&loclyp);
6789 	for (i = 0; i < nfsrv_layouthashsize; i++) {
6790 		lhyp = &nfslayouthash[i];
6791 		NFSLOCKLAYOUT(lhyp);
6792 		TAILQ_FOREACH_SAFE(lyp, &lhyp->list, lay_list, nlyp) {
6793 			if (lyp->lay_type == NFSLAYOUT_FLEXFILE &&
6794 			    lyp->lay_mirrorcnt > 1) {
6795 				NFSD_DEBUG(4, "possible match\n");
6796 				tl = lyp->lay_xdr;
6797 				tl += 3;
6798 				for (j = 0; j < lyp->lay_mirrorcnt; j++) {
6799 					tl++;
6800 					if (NFSBCMP(devid, tl, NFSX_V4DEVICEID)
6801 					    == 0) {
6802 						/* Found one. */
6803 						NFSD_DEBUG(4, "fnd one\n");
6804 						TAILQ_REMOVE(&lhyp->list, lyp,
6805 						    lay_list);
6806 						TAILQ_INSERT_HEAD(&loclyp, lyp,
6807 						    lay_list);
6808 						break;
6809 					}
6810 					tl += (NFSX_V4DEVICEID / NFSX_UNSIGNED +
6811 					    NFSM_RNDUP(NFSX_V4PNFSFH) /
6812 					    NFSX_UNSIGNED + 11 * NFSX_UNSIGNED);
6813 				}
6814 			}
6815 		}
6816 		NFSUNLOCKLAYOUT(lhyp);
6817 	}
6818 
6819 	/* Now, try to do a Layout recall for each one found. */
6820 	TAILQ_FOREACH_SAFE(lyp, &loclyp, lay_list, nlyp) {
6821 		NFSD_DEBUG(4, "do layout recall\n");
6822 		/*
6823 		 * The layout stateid.seqid needs to be incremented
6824 		 * before doing a LAYOUT_RECALL callback.
6825 		 */
6826 		if (++lyp->lay_stateid.seqid == 0)
6827 			lyp->lay_stateid.seqid = 1;
6828 		nfsrv_recalllayout(lyp->lay_clientid, &lyp->lay_stateid,
6829 		    &lyp->lay_fh, lyp, 1, lyp->lay_type, p);
6830 		nfsrv_freelayout(&loclyp, lyp);
6831 	}
6832 }
6833 
6834 /*
6835  * Do a recall callback to the client for this layout.
6836  */
6837 static int
6838 nfsrv_recalllayout(nfsquad_t clid, nfsv4stateid_t *stateidp, fhandle_t *fhp,
6839     struct nfslayout *lyp, int changed, int laytype, NFSPROC_T *p)
6840 {
6841 	struct nfsclient *clp;
6842 	int error;
6843 
6844 	NFSD_DEBUG(4, "nfsrv_recalllayout\n");
6845 	error = nfsrv_getclient(clid, 0, &clp, NULL, (nfsquad_t)((u_quad_t)0),
6846 	    0, NULL, p);
6847 	NFSD_DEBUG(4, "aft nfsrv_getclient=%d\n", error);
6848 	if (error != 0) {
6849 		printf("nfsrv_recalllayout: getclient err=%d\n", error);
6850 		return (error);
6851 	}
6852 	if ((clp->lc_flags & LCL_NFSV41) != 0) {
6853 		error = nfsrv_docallback(clp, NFSV4OP_CBLAYOUTRECALL,
6854 		    stateidp, changed, fhp, NULL, NULL, laytype, p);
6855 		/* If lyp != NULL, handle an error return here. */
6856 		if (error != 0 && lyp != NULL) {
6857 			NFSDRECALLLOCK();
6858 			/*
6859 			 * Mark it returned, since no layout recall
6860 			 * has been done.
6861 			 * All errors seem to be non-recoverable, although
6862 			 * NFSERR_NOMATCHLAYOUT is a normal event.
6863 			 */
6864 			if ((lyp->lay_flags & NFSLAY_RECALL) != 0) {
6865 				lyp->lay_flags |= NFSLAY_RETURNED;
6866 				wakeup(lyp);
6867 			}
6868 			NFSDRECALLUNLOCK();
6869 			if (error != NFSERR_NOMATCHLAYOUT)
6870 				printf("nfsrv_recalllayout: err=%d\n", error);
6871 		}
6872 	} else
6873 		printf("nfsrv_recalllayout: clp not NFSv4.1\n");
6874 	return (error);
6875 }
6876 
6877 /*
6878  * Find a layout to recall when we exceed our high water mark.
6879  */
6880 void
6881 nfsrv_recalloldlayout(NFSPROC_T *p)
6882 {
6883 	struct nfslayouthash *lhyp;
6884 	struct nfslayout *lyp;
6885 	nfsquad_t clientid;
6886 	nfsv4stateid_t stateid;
6887 	fhandle_t fh;
6888 	int error, laytype, ret;
6889 
6890 	lhyp = &nfslayouthash[arc4random() % nfsrv_layouthashsize];
6891 	NFSLOCKLAYOUT(lhyp);
6892 	TAILQ_FOREACH_REVERSE(lyp, &lhyp->list, nfslayouthead, lay_list) {
6893 		if ((lyp->lay_flags & NFSLAY_CALLB) == 0) {
6894 			lyp->lay_flags |= NFSLAY_CALLB;
6895 			/*
6896 			 * The layout stateid.seqid needs to be incremented
6897 			 * before doing a LAYOUT_RECALL callback.
6898 			 */
6899 			if (++lyp->lay_stateid.seqid == 0)
6900 				lyp->lay_stateid.seqid = 1;
6901 			clientid = lyp->lay_clientid;
6902 			stateid = lyp->lay_stateid;
6903 			NFSBCOPY(&lyp->lay_fh, &fh, sizeof(fh));
6904 			laytype = lyp->lay_type;
6905 			break;
6906 		}
6907 	}
6908 	NFSUNLOCKLAYOUT(lhyp);
6909 	if (lyp != NULL) {
6910 		error = nfsrv_recalllayout(clientid, &stateid, &fh, NULL, 0,
6911 		    laytype, p);
6912 		if (error != 0 && error != NFSERR_NOMATCHLAYOUT)
6913 			NFSD_DEBUG(4, "recallold=%d\n", error);
6914 		if (error != 0) {
6915 			NFSLOCKLAYOUT(lhyp);
6916 			/*
6917 			 * Since the hash list was unlocked, we need to
6918 			 * find it again.
6919 			 */
6920 			ret = nfsrv_findlayout(&clientid, &fh, laytype, p,
6921 			    &lyp);
6922 			if (ret == 0 &&
6923 			    (lyp->lay_flags & NFSLAY_CALLB) != 0 &&
6924 			    lyp->lay_stateid.other[0] == stateid.other[0] &&
6925 			    lyp->lay_stateid.other[1] == stateid.other[1] &&
6926 			    lyp->lay_stateid.other[2] == stateid.other[2]) {
6927 				/*
6928 				 * The client no longer knows this layout, so
6929 				 * it can be free'd now.
6930 				 */
6931 				if (error == NFSERR_NOMATCHLAYOUT)
6932 					nfsrv_freelayout(&lhyp->list, lyp);
6933 				else {
6934 					/*
6935 					 * Leave it to be tried later by
6936 					 * clearing NFSLAY_CALLB and moving
6937 					 * it to the head of the list, so it
6938 					 * won't be tried again for a while.
6939 					 */
6940 					lyp->lay_flags &= ~NFSLAY_CALLB;
6941 					TAILQ_REMOVE(&lhyp->list, lyp,
6942 					    lay_list);
6943 					TAILQ_INSERT_HEAD(&lhyp->list, lyp,
6944 					    lay_list);
6945 				}
6946 			}
6947 			NFSUNLOCKLAYOUT(lhyp);
6948 		}
6949 	}
6950 }
6951 
6952 /*
6953  * Try and return layout(s).
6954  */
6955 int
6956 nfsrv_layoutreturn(struct nfsrv_descript *nd, vnode_t vp,
6957     int layouttype, int iomode, uint64_t offset, uint64_t len, int reclaim,
6958     int kind, nfsv4stateid_t *stateidp, int maxcnt, uint32_t *layp, int *fndp,
6959     struct ucred *cred, NFSPROC_T *p)
6960 {
6961 	struct nfsvattr na;
6962 	struct nfslayouthash *lhyp;
6963 	struct nfslayout *lyp;
6964 	fhandle_t fh;
6965 	int error = 0;
6966 
6967 	*fndp = 0;
6968 	if (kind == NFSV4LAYOUTRET_FILE) {
6969 		error = nfsvno_getfh(vp, &fh, p);
6970 		if (error == 0) {
6971 			error = nfsrv_updatemdsattr(vp, &na, p);
6972 			if (error != 0)
6973 				printf("nfsrv_layoutreturn: updatemdsattr"
6974 				    " failed=%d\n", error);
6975 		}
6976 		if (error == 0) {
6977 			if (reclaim == newnfs_true) {
6978 				error = nfsrv_checkgrace(NULL, NULL,
6979 				    NFSLCK_RECLAIM);
6980 				if (error != NFSERR_NOGRACE)
6981 					error = 0;
6982 				return (error);
6983 			}
6984 			lhyp = NFSLAYOUTHASH(&fh);
6985 			NFSDRECALLLOCK();
6986 			NFSLOCKLAYOUT(lhyp);
6987 			error = nfsrv_findlayout(&nd->nd_clientid, &fh,
6988 			    layouttype, p, &lyp);
6989 			NFSD_DEBUG(4, "layoutret findlay=%d\n", error);
6990 			if (error == 0 &&
6991 			    stateidp->other[0] == lyp->lay_stateid.other[0] &&
6992 			    stateidp->other[1] == lyp->lay_stateid.other[1] &&
6993 			    stateidp->other[2] == lyp->lay_stateid.other[2]) {
6994 				NFSD_DEBUG(4, "nfsrv_layoutreturn: stateid %d"
6995 				    " %x %x %x laystateid %d %x %x %x"
6996 				    " off=%ju len=%ju flgs=0x%x\n",
6997 				    stateidp->seqid, stateidp->other[0],
6998 				    stateidp->other[1], stateidp->other[2],
6999 				    lyp->lay_stateid.seqid,
7000 				    lyp->lay_stateid.other[0],
7001 				    lyp->lay_stateid.other[1],
7002 				    lyp->lay_stateid.other[2],
7003 				    (uintmax_t)offset, (uintmax_t)len,
7004 				    lyp->lay_flags);
7005 				if (++lyp->lay_stateid.seqid == 0)
7006 					lyp->lay_stateid.seqid = 1;
7007 				stateidp->seqid = lyp->lay_stateid.seqid;
7008 				if (offset == 0 && len == UINT64_MAX) {
7009 					if ((iomode & NFSLAYOUTIOMODE_READ) !=
7010 					    0)
7011 						lyp->lay_flags &= ~NFSLAY_READ;
7012 					if ((iomode & NFSLAYOUTIOMODE_RW) != 0)
7013 						lyp->lay_flags &= ~NFSLAY_RW;
7014 					if ((lyp->lay_flags & (NFSLAY_READ |
7015 					    NFSLAY_RW)) == 0)
7016 						nfsrv_freelayout(&lhyp->list,
7017 						    lyp);
7018 					else
7019 						*fndp = 1;
7020 				} else
7021 					*fndp = 1;
7022 			}
7023 			NFSUNLOCKLAYOUT(lhyp);
7024 			/* Search the nfsrv_recalllist for a match. */
7025 			TAILQ_FOREACH(lyp, &nfsrv_recalllisthead, lay_list) {
7026 				if (NFSBCMP(&lyp->lay_fh, &fh,
7027 				    sizeof(fh)) == 0 &&
7028 				    lyp->lay_clientid.qval ==
7029 				    nd->nd_clientid.qval &&
7030 				    stateidp->other[0] ==
7031 				    lyp->lay_stateid.other[0] &&
7032 				    stateidp->other[1] ==
7033 				    lyp->lay_stateid.other[1] &&
7034 				    stateidp->other[2] ==
7035 				    lyp->lay_stateid.other[2]) {
7036 					lyp->lay_flags |= NFSLAY_RETURNED;
7037 					wakeup(lyp);
7038 					error = 0;
7039 				}
7040 			}
7041 			NFSDRECALLUNLOCK();
7042 		}
7043 		if (layouttype == NFSLAYOUT_FLEXFILE)
7044 			nfsrv_flexlayouterr(nd, layp, maxcnt, p);
7045 	} else if (kind == NFSV4LAYOUTRET_FSID)
7046 		nfsrv_freelayouts(&nd->nd_clientid,
7047 		    &vp->v_mount->mnt_stat.f_fsid, layouttype, iomode);
7048 	else if (kind == NFSV4LAYOUTRET_ALL)
7049 		nfsrv_freelayouts(&nd->nd_clientid, NULL, layouttype, iomode);
7050 	else
7051 		error = NFSERR_INVAL;
7052 	if (error == -1)
7053 		error = 0;
7054 	return (error);
7055 }
7056 
7057 /*
7058  * Look for an existing layout.
7059  */
7060 static int
7061 nfsrv_findlayout(nfsquad_t *clientidp, fhandle_t *fhp, int laytype,
7062     NFSPROC_T *p, struct nfslayout **lypp)
7063 {
7064 	struct nfslayouthash *lhyp;
7065 	struct nfslayout *lyp;
7066 	int ret;
7067 
7068 	*lypp = NULL;
7069 	ret = 0;
7070 	lhyp = NFSLAYOUTHASH(fhp);
7071 	TAILQ_FOREACH(lyp, &lhyp->list, lay_list) {
7072 		if (NFSBCMP(&lyp->lay_fh, fhp, sizeof(*fhp)) == 0 &&
7073 		    lyp->lay_clientid.qval == clientidp->qval &&
7074 		    lyp->lay_type == laytype)
7075 			break;
7076 	}
7077 	if (lyp != NULL)
7078 		*lypp = lyp;
7079 	else
7080 		ret = -1;
7081 	return (ret);
7082 }
7083 
7084 /*
7085  * Add the new layout, as required.
7086  */
7087 static int
7088 nfsrv_addlayout(struct nfsrv_descript *nd, struct nfslayout **lypp,
7089     nfsv4stateid_t *stateidp, char *layp, int *layoutlenp, NFSPROC_T *p)
7090 {
7091 	struct nfsclient *clp;
7092 	struct nfslayouthash *lhyp;
7093 	struct nfslayout *lyp, *nlyp;
7094 	fhandle_t *fhp;
7095 	int error;
7096 
7097 	KASSERT((nd->nd_flag & ND_IMPLIEDCLID) != 0,
7098 	    ("nfsrv_layoutget: no nd_clientid\n"));
7099 	lyp = *lypp;
7100 	fhp = &lyp->lay_fh;
7101 	NFSLOCKSTATE();
7102 	error = nfsrv_getclient((nfsquad_t)((u_quad_t)0), CLOPS_RENEW, &clp,
7103 	    NULL, (nfsquad_t)((u_quad_t)0), 0, nd, p);
7104 	if (error != 0) {
7105 		NFSUNLOCKSTATE();
7106 		return (error);
7107 	}
7108 	lyp->lay_stateid.seqid = stateidp->seqid = 1;
7109 	lyp->lay_stateid.other[0] = stateidp->other[0] =
7110 	    clp->lc_clientid.lval[0];
7111 	lyp->lay_stateid.other[1] = stateidp->other[1] =
7112 	    clp->lc_clientid.lval[1];
7113 	lyp->lay_stateid.other[2] = stateidp->other[2] =
7114 	    nfsrv_nextstateindex(clp);
7115 	NFSUNLOCKSTATE();
7116 
7117 	lhyp = NFSLAYOUTHASH(fhp);
7118 	NFSLOCKLAYOUT(lhyp);
7119 	TAILQ_FOREACH(nlyp, &lhyp->list, lay_list) {
7120 		if (NFSBCMP(&nlyp->lay_fh, fhp, sizeof(*fhp)) == 0 &&
7121 		    nlyp->lay_clientid.qval == nd->nd_clientid.qval)
7122 			break;
7123 	}
7124 	if (nlyp != NULL) {
7125 		/* A layout already exists, so use it. */
7126 		nlyp->lay_flags |= (lyp->lay_flags & (NFSLAY_READ | NFSLAY_RW));
7127 		NFSBCOPY(nlyp->lay_xdr, layp, nlyp->lay_layoutlen);
7128 		*layoutlenp = nlyp->lay_layoutlen;
7129 		if (++nlyp->lay_stateid.seqid == 0)
7130 			nlyp->lay_stateid.seqid = 1;
7131 		stateidp->seqid = nlyp->lay_stateid.seqid;
7132 		stateidp->other[0] = nlyp->lay_stateid.other[0];
7133 		stateidp->other[1] = nlyp->lay_stateid.other[1];
7134 		stateidp->other[2] = nlyp->lay_stateid.other[2];
7135 		NFSUNLOCKLAYOUT(lhyp);
7136 		return (0);
7137 	}
7138 
7139 	/* Insert the new layout in the lists. */
7140 	*lypp = NULL;
7141 	atomic_add_int(&nfsrv_layoutcnt, 1);
7142 	NFSBCOPY(lyp->lay_xdr, layp, lyp->lay_layoutlen);
7143 	*layoutlenp = lyp->lay_layoutlen;
7144 	TAILQ_INSERT_HEAD(&lhyp->list, lyp, lay_list);
7145 	NFSUNLOCKLAYOUT(lhyp);
7146 	return (0);
7147 }
7148 
7149 /*
7150  * Get the devinfo for a deviceid.
7151  */
7152 int
7153 nfsrv_getdevinfo(char *devid, int layouttype, uint32_t *maxcnt,
7154     uint32_t *notify, int *devaddrlen, char **devaddr)
7155 {
7156 	struct nfsdevice *ds;
7157 
7158 	if ((layouttype != NFSLAYOUT_NFSV4_1_FILES && layouttype !=
7159 	     NFSLAYOUT_FLEXFILE) ||
7160 	    (nfsrv_maxpnfsmirror > 1 && layouttype == NFSLAYOUT_NFSV4_1_FILES))
7161 		return (NFSERR_UNKNLAYOUTTYPE);
7162 
7163 	/*
7164 	 * Now, search for the device id.  Note that the structures won't go
7165 	 * away, but the order changes in the list.  As such, the lock only
7166 	 * needs to be held during the search through the list.
7167 	 */
7168 	NFSDDSLOCK();
7169 	TAILQ_FOREACH(ds, &nfsrv_devidhead, nfsdev_list) {
7170 		if (NFSBCMP(devid, ds->nfsdev_deviceid, NFSX_V4DEVICEID) == 0 &&
7171 		    ds->nfsdev_nmp != NULL)
7172 			break;
7173 	}
7174 	NFSDDSUNLOCK();
7175 	if (ds == NULL)
7176 		return (NFSERR_NOENT);
7177 
7178 	/* If the correct nfsdev_XXXXaddrlen is > 0, we have the device info. */
7179 	*devaddrlen = 0;
7180 	if (layouttype == NFSLAYOUT_NFSV4_1_FILES) {
7181 		*devaddrlen = ds->nfsdev_fileaddrlen;
7182 		*devaddr = ds->nfsdev_fileaddr;
7183 	} else if (layouttype == NFSLAYOUT_FLEXFILE) {
7184 		*devaddrlen = ds->nfsdev_flexaddrlen;
7185 		*devaddr = ds->nfsdev_flexaddr;
7186 	}
7187 	if (*devaddrlen == 0)
7188 		return (NFSERR_UNKNLAYOUTTYPE);
7189 
7190 	/*
7191 	 * The XDR overhead is 3 unsigned values: layout_type,
7192 	 * length_of_address and notify bitmap.
7193 	 * If the notify array is changed to not all zeros, the
7194 	 * count of unsigned values must be increased.
7195 	 */
7196 	if (*maxcnt > 0 && *maxcnt < NFSM_RNDUP(*devaddrlen) +
7197 	    3 * NFSX_UNSIGNED) {
7198 		*maxcnt = NFSM_RNDUP(*devaddrlen) + 3 * NFSX_UNSIGNED;
7199 		return (NFSERR_TOOSMALL);
7200 	}
7201 	return (0);
7202 }
7203 
7204 /*
7205  * Free a list of layout state structures.
7206  */
7207 static void
7208 nfsrv_freelayoutlist(nfsquad_t clientid)
7209 {
7210 	struct nfslayouthash *lhyp;
7211 	struct nfslayout *lyp, *nlyp;
7212 	int i;
7213 
7214 	for (i = 0; i < nfsrv_layouthashsize; i++) {
7215 		lhyp = &nfslayouthash[i];
7216 		NFSLOCKLAYOUT(lhyp);
7217 		TAILQ_FOREACH_SAFE(lyp, &lhyp->list, lay_list, nlyp) {
7218 			if (lyp->lay_clientid.qval == clientid.qval)
7219 				nfsrv_freelayout(&lhyp->list, lyp);
7220 		}
7221 		NFSUNLOCKLAYOUT(lhyp);
7222 	}
7223 }
7224 
7225 /*
7226  * Free up a layout.
7227  */
7228 static void
7229 nfsrv_freelayout(struct nfslayouthead *lhp, struct nfslayout *lyp)
7230 {
7231 
7232 	NFSD_DEBUG(4, "Freelayout=%p\n", lyp);
7233 	atomic_add_int(&nfsrv_layoutcnt, -1);
7234 	TAILQ_REMOVE(lhp, lyp, lay_list);
7235 	free(lyp, M_NFSDSTATE);
7236 }
7237 
7238 /*
7239  * Free up a device id.
7240  */
7241 void
7242 nfsrv_freeonedevid(struct nfsdevice *ds)
7243 {
7244 	int i;
7245 
7246 	atomic_add_int(&nfsrv_devidcnt, -1);
7247 	vrele(ds->nfsdev_dvp);
7248 	for (i = 0; i < nfsrv_dsdirsize; i++)
7249 		if (ds->nfsdev_dsdir[i] != NULL)
7250 			vrele(ds->nfsdev_dsdir[i]);
7251 	free(ds->nfsdev_fileaddr, M_NFSDSTATE);
7252 	free(ds->nfsdev_flexaddr, M_NFSDSTATE);
7253 	free(ds->nfsdev_host, M_NFSDSTATE);
7254 	free(ds, M_NFSDSTATE);
7255 }
7256 
7257 /*
7258  * Free up a device id and its mirrors.
7259  */
7260 static void
7261 nfsrv_freedevid(struct nfsdevice *ds)
7262 {
7263 
7264 	TAILQ_REMOVE(&nfsrv_devidhead, ds, nfsdev_list);
7265 	nfsrv_freeonedevid(ds);
7266 }
7267 
7268 /*
7269  * Free all layouts and device ids.
7270  * Done when the nfsd threads are shut down since there may be a new
7271  * modified device id list created when the nfsd is restarted.
7272  */
7273 void
7274 nfsrv_freealllayoutsanddevids(void)
7275 {
7276 	struct nfsdontlist *mrp, *nmrp;
7277 	struct nfslayout *lyp, *nlyp;
7278 
7279 	/* Get rid of the deviceid structures. */
7280 	nfsrv_freealldevids();
7281 	TAILQ_INIT(&nfsrv_devidhead);
7282 	nfsrv_devidcnt = 0;
7283 
7284 	/* Get rid of all layouts. */
7285 	nfsrv_freealllayouts();
7286 
7287 	/* Get rid of any nfsdontlist entries. */
7288 	LIST_FOREACH_SAFE(mrp, &nfsrv_dontlisthead, nfsmr_list, nmrp)
7289 		free(mrp, M_NFSDSTATE);
7290 	LIST_INIT(&nfsrv_dontlisthead);
7291 	nfsrv_dontlistlen = 0;
7292 
7293 	/* Free layouts in the recall list. */
7294 	TAILQ_FOREACH_SAFE(lyp, &nfsrv_recalllisthead, lay_list, nlyp)
7295 		nfsrv_freelayout(&nfsrv_recalllisthead, lyp);
7296 	TAILQ_INIT(&nfsrv_recalllisthead);
7297 }
7298 
7299 /*
7300  * Free layouts that match the arguments.
7301  */
7302 static void
7303 nfsrv_freelayouts(nfsquad_t *clid, fsid_t *fs, int laytype, int iomode)
7304 {
7305 	struct nfslayouthash *lhyp;
7306 	struct nfslayout *lyp, *nlyp;
7307 	int i;
7308 
7309 	for (i = 0; i < nfsrv_layouthashsize; i++) {
7310 		lhyp = &nfslayouthash[i];
7311 		NFSLOCKLAYOUT(lhyp);
7312 		TAILQ_FOREACH_SAFE(lyp, &lhyp->list, lay_list, nlyp) {
7313 			if (clid->qval != lyp->lay_clientid.qval)
7314 				continue;
7315 			if (fs != NULL && (fs->val[0] != lyp->lay_fsid.val[0] ||
7316 			    fs->val[1] != lyp->lay_fsid.val[1]))
7317 				continue;
7318 			if (laytype != lyp->lay_type)
7319 				continue;
7320 			if ((iomode & NFSLAYOUTIOMODE_READ) != 0)
7321 				lyp->lay_flags &= ~NFSLAY_READ;
7322 			if ((iomode & NFSLAYOUTIOMODE_RW) != 0)
7323 				lyp->lay_flags &= ~NFSLAY_RW;
7324 			if ((lyp->lay_flags & (NFSLAY_READ | NFSLAY_RW)) == 0)
7325 				nfsrv_freelayout(&lhyp->list, lyp);
7326 		}
7327 		NFSUNLOCKLAYOUT(lhyp);
7328 	}
7329 }
7330 
7331 /*
7332  * Free all layouts for the argument file.
7333  */
7334 void
7335 nfsrv_freefilelayouts(fhandle_t *fhp)
7336 {
7337 	struct nfslayouthash *lhyp;
7338 	struct nfslayout *lyp, *nlyp;
7339 
7340 	lhyp = NFSLAYOUTHASH(fhp);
7341 	NFSLOCKLAYOUT(lhyp);
7342 	TAILQ_FOREACH_SAFE(lyp, &lhyp->list, lay_list, nlyp) {
7343 		if (NFSBCMP(&lyp->lay_fh, fhp, sizeof(*fhp)) == 0)
7344 			nfsrv_freelayout(&lhyp->list, lyp);
7345 	}
7346 	NFSUNLOCKLAYOUT(lhyp);
7347 }
7348 
7349 /*
7350  * Free all layouts.
7351  */
7352 static void
7353 nfsrv_freealllayouts(void)
7354 {
7355 	struct nfslayouthash *lhyp;
7356 	struct nfslayout *lyp, *nlyp;
7357 	int i;
7358 
7359 	for (i = 0; i < nfsrv_layouthashsize; i++) {
7360 		lhyp = &nfslayouthash[i];
7361 		NFSLOCKLAYOUT(lhyp);
7362 		TAILQ_FOREACH_SAFE(lyp, &lhyp->list, lay_list, nlyp)
7363 			nfsrv_freelayout(&lhyp->list, lyp);
7364 		NFSUNLOCKLAYOUT(lhyp);
7365 	}
7366 }
7367 
7368 /*
7369  * Look up the mount path for the DS server.
7370  */
7371 static int
7372 nfsrv_setdsserver(char *dspathp, char *mdspathp, NFSPROC_T *p,
7373     struct nfsdevice **dsp)
7374 {
7375 	struct nameidata nd;
7376 	struct nfsdevice *ds;
7377 	struct mount *mp;
7378 	int error, i;
7379 	char *dsdirpath;
7380 	size_t dsdirsize;
7381 
7382 	NFSD_DEBUG(4, "setdssrv path=%s\n", dspathp);
7383 	*dsp = NULL;
7384 	NDINIT(&nd, LOOKUP, FOLLOW | LOCKSHARED | LOCKLEAF, UIO_SYSSPACE,
7385 	    dspathp, p);
7386 	error = namei(&nd);
7387 	NFSD_DEBUG(4, "lookup=%d\n", error);
7388 	if (error != 0)
7389 		return (error);
7390 	if (nd.ni_vp->v_type != VDIR) {
7391 		vput(nd.ni_vp);
7392 		NFSD_DEBUG(4, "dspath not dir\n");
7393 		return (ENOTDIR);
7394 	}
7395 	if (strcmp(nd.ni_vp->v_mount->mnt_vfc->vfc_name, "nfs") != 0) {
7396 		vput(nd.ni_vp);
7397 		NFSD_DEBUG(4, "dspath not an NFS mount\n");
7398 		return (ENXIO);
7399 	}
7400 
7401 	/*
7402 	 * Allocate a DS server structure with the NFS mounted directory
7403 	 * vnode reference counted, so that a non-forced dismount will
7404 	 * fail with EBUSY.
7405 	 * This structure is always linked into the list, even if an error
7406 	 * is being returned.  The caller will free the entire list upon
7407 	 * an error return.
7408 	 */
7409 	*dsp = ds = malloc(sizeof(*ds) + nfsrv_dsdirsize * sizeof(vnode_t),
7410 	    M_NFSDSTATE, M_WAITOK | M_ZERO);
7411 	ds->nfsdev_dvp = nd.ni_vp;
7412 	ds->nfsdev_nmp = VFSTONFS(nd.ni_vp->v_mount);
7413 	NFSVOPUNLOCK(nd.ni_vp, 0);
7414 
7415 	dsdirsize = strlen(dspathp) + 16;
7416 	dsdirpath = malloc(dsdirsize, M_TEMP, M_WAITOK);
7417 	/* Now, create the DS directory structures. */
7418 	for (i = 0; i < nfsrv_dsdirsize; i++) {
7419 		snprintf(dsdirpath, dsdirsize, "%s/ds%d", dspathp, i);
7420 		NDINIT(&nd, LOOKUP, FOLLOW | LOCKSHARED | LOCKLEAF,
7421 		    UIO_SYSSPACE, dsdirpath, p);
7422 		error = namei(&nd);
7423 		NFSD_DEBUG(4, "dsdirpath=%s lookup=%d\n", dsdirpath, error);
7424 		if (error != 0)
7425 			break;
7426 		if (nd.ni_vp->v_type != VDIR) {
7427 			vput(nd.ni_vp);
7428 			error = ENOTDIR;
7429 			NFSD_DEBUG(4, "dsdirpath not a VDIR\n");
7430 			break;
7431 		}
7432 		if (strcmp(nd.ni_vp->v_mount->mnt_vfc->vfc_name, "nfs") != 0) {
7433 			vput(nd.ni_vp);
7434 			error = ENXIO;
7435 			NFSD_DEBUG(4, "dsdirpath not an NFS mount\n");
7436 			break;
7437 		}
7438 		ds->nfsdev_dsdir[i] = nd.ni_vp;
7439 		NFSVOPUNLOCK(nd.ni_vp, 0);
7440 	}
7441 	free(dsdirpath, M_TEMP);
7442 
7443 	if (strlen(mdspathp) > 0) {
7444 		/*
7445 		 * This DS stores file for a specific MDS exported file
7446 		 * system.
7447 		 */
7448 		NDINIT(&nd, LOOKUP, FOLLOW | LOCKSHARED | LOCKLEAF,
7449 		    UIO_SYSSPACE, mdspathp, p);
7450 		error = namei(&nd);
7451 		NFSD_DEBUG(4, "mds lookup=%d\n", error);
7452 		if (error != 0)
7453 			goto out;
7454 		if (nd.ni_vp->v_type != VDIR) {
7455 			vput(nd.ni_vp);
7456 			error = ENOTDIR;
7457 			NFSD_DEBUG(4, "mdspath not dir\n");
7458 			goto out;
7459 		}
7460 		mp = nd.ni_vp->v_mount;
7461 		if ((mp->mnt_flag & MNT_EXPORTED) == 0) {
7462 			vput(nd.ni_vp);
7463 			error = ENXIO;
7464 			NFSD_DEBUG(4, "mdspath not an exported fs\n");
7465 			goto out;
7466 		}
7467 		ds->nfsdev_mdsfsid = mp->mnt_stat.f_fsid;
7468 		ds->nfsdev_mdsisset = 1;
7469 		vput(nd.ni_vp);
7470 	}
7471 
7472 out:
7473 	TAILQ_INSERT_TAIL(&nfsrv_devidhead, ds, nfsdev_list);
7474 	atomic_add_int(&nfsrv_devidcnt, 1);
7475 	return (error);
7476 }
7477 
7478 /*
7479  * Look up the mount path for the DS server and delete it.
7480  */
7481 int
7482 nfsrv_deldsserver(int op, char *dspathp, NFSPROC_T *p)
7483 {
7484 	struct mount *mp;
7485 	struct nfsmount *nmp;
7486 	struct nfsdevice *ds;
7487 	int error;
7488 
7489 	NFSD_DEBUG(4, "deldssrv path=%s\n", dspathp);
7490 	/*
7491 	 * Search for the path in the mount list.  Avoid looking the path
7492 	 * up, since this mount point may be hung, with associated locked
7493 	 * vnodes, etc.
7494 	 * Set NFSMNTP_CANCELRPCS so that any forced dismount will be blocked
7495 	 * until this completes.
7496 	 * As noted in the man page, this should be done before any forced
7497 	 * dismount on the mount point, but at least the handshake on
7498 	 * NFSMNTP_CANCELRPCS should make it safe.
7499 	 */
7500 	error = 0;
7501 	ds = NULL;
7502 	nmp = NULL;
7503 	mtx_lock(&mountlist_mtx);
7504 	TAILQ_FOREACH(mp, &mountlist, mnt_list) {
7505 		if (strcmp(mp->mnt_stat.f_mntonname, dspathp) == 0 &&
7506 		    strcmp(mp->mnt_stat.f_fstypename, "nfs") == 0 &&
7507 		    mp->mnt_data != NULL) {
7508 			nmp = VFSTONFS(mp);
7509 			NFSLOCKMNT(nmp);
7510 			if ((nmp->nm_privflag & (NFSMNTP_FORCEDISM |
7511 			     NFSMNTP_CANCELRPCS)) == 0) {
7512 				nmp->nm_privflag |= NFSMNTP_CANCELRPCS;
7513 				NFSUNLOCKMNT(nmp);
7514 			} else {
7515 				NFSUNLOCKMNT(nmp);
7516 				nmp = NULL;
7517 			}
7518 			break;
7519 		}
7520 	}
7521 	mtx_unlock(&mountlist_mtx);
7522 
7523 	if (nmp != NULL) {
7524 		ds = nfsrv_deldsnmp(op, nmp, p);
7525 		NFSD_DEBUG(4, "deldsnmp=%p\n", ds);
7526 		if (ds != NULL) {
7527 			nfsrv_killrpcs(nmp);
7528 			NFSD_DEBUG(4, "aft killrpcs\n");
7529 		} else
7530 			error = ENXIO;
7531 		NFSLOCKMNT(nmp);
7532 		nmp->nm_privflag &= ~NFSMNTP_CANCELRPCS;
7533 		wakeup(nmp);
7534 		NFSUNLOCKMNT(nmp);
7535 	} else
7536 		error = EINVAL;
7537 	return (error);
7538 }
7539 
7540 /*
7541  * Search for and remove a DS entry which matches the "nmp" argument.
7542  * The nfsdevice structure pointer is returned so that the caller can
7543  * free it via nfsrv_freeonedevid().
7544  * For the forced case, do not try to do LayoutRecalls, since the server
7545  * must be shut down now anyhow.
7546  */
7547 struct nfsdevice *
7548 nfsrv_deldsnmp(int op, struct nfsmount *nmp, NFSPROC_T *p)
7549 {
7550 	struct nfsdevice *fndds;
7551 
7552 	NFSD_DEBUG(4, "deldsdvp\n");
7553 	NFSDDSLOCK();
7554 	if (op == PNFSDOP_FORCEDELDS)
7555 		fndds = nfsv4_findmirror(nmp);
7556 	else
7557 		fndds = nfsrv_findmirroredds(nmp);
7558 	if (fndds != NULL)
7559 		nfsrv_deleteds(fndds);
7560 	NFSDDSUNLOCK();
7561 	if (fndds != NULL) {
7562 		if (op != PNFSDOP_FORCEDELDS)
7563 			nfsrv_flexmirrordel(fndds->nfsdev_deviceid, p);
7564 		printf("pNFS server: mirror %s failed\n", fndds->nfsdev_host);
7565 	}
7566 	return (fndds);
7567 }
7568 
7569 /*
7570  * Similar to nfsrv_deldsnmp(), except that the DS is indicated by deviceid.
7571  * This function also calls nfsrv_killrpcs() to unblock RPCs on the mount
7572  * point.
7573  * Also, returns an error instead of the nfsdevice found.
7574  */
7575 static int
7576 nfsrv_delds(char *devid, NFSPROC_T *p)
7577 {
7578 	struct nfsdevice *ds, *fndds;
7579 	struct nfsmount *nmp;
7580 	int fndmirror;
7581 
7582 	NFSD_DEBUG(4, "delds\n");
7583 	/*
7584 	 * Search the DS server list for a match with devid.
7585 	 * Remove the DS entry if found and there is a mirror.
7586 	 */
7587 	fndds = NULL;
7588 	nmp = NULL;
7589 	fndmirror = 0;
7590 	NFSDDSLOCK();
7591 	TAILQ_FOREACH(ds, &nfsrv_devidhead, nfsdev_list) {
7592 		if (NFSBCMP(ds->nfsdev_deviceid, devid, NFSX_V4DEVICEID) == 0 &&
7593 		    ds->nfsdev_nmp != NULL) {
7594 			NFSD_DEBUG(4, "fnd main ds\n");
7595 			fndds = ds;
7596 			break;
7597 		}
7598 	}
7599 	if (fndds == NULL) {
7600 		NFSDDSUNLOCK();
7601 		return (ENXIO);
7602 	}
7603 	if (fndds->nfsdev_mdsisset == 0 && nfsrv_faildscnt > 0)
7604 		fndmirror = 1;
7605 	else if (fndds->nfsdev_mdsisset != 0) {
7606 		/* For the fsid is set case, search for a mirror. */
7607 		TAILQ_FOREACH(ds, &nfsrv_devidhead, nfsdev_list) {
7608 			if (ds != fndds && ds->nfsdev_nmp != NULL &&
7609 			    ds->nfsdev_mdsisset != 0 &&
7610 			    ds->nfsdev_mdsfsid.val[0] ==
7611 			    fndds->nfsdev_mdsfsid.val[0] &&
7612 			    ds->nfsdev_mdsfsid.val[1] ==
7613 			    fndds->nfsdev_mdsfsid.val[1]) {
7614 				fndmirror = 1;
7615 				break;
7616 			}
7617 		}
7618 	}
7619 	if (fndmirror != 0) {
7620 		nmp = fndds->nfsdev_nmp;
7621 		NFSLOCKMNT(nmp);
7622 		if ((nmp->nm_privflag & (NFSMNTP_FORCEDISM |
7623 		     NFSMNTP_CANCELRPCS)) == 0) {
7624 			nmp->nm_privflag |= NFSMNTP_CANCELRPCS;
7625 			NFSUNLOCKMNT(nmp);
7626 			nfsrv_deleteds(fndds);
7627 		} else {
7628 			NFSUNLOCKMNT(nmp);
7629 			nmp = NULL;
7630 		}
7631 	}
7632 	NFSDDSUNLOCK();
7633 	if (nmp != NULL) {
7634 		nfsrv_flexmirrordel(fndds->nfsdev_deviceid, p);
7635 		printf("pNFS server: mirror %s failed\n", fndds->nfsdev_host);
7636 		nfsrv_killrpcs(nmp);
7637 		NFSLOCKMNT(nmp);
7638 		nmp->nm_privflag &= ~NFSMNTP_CANCELRPCS;
7639 		wakeup(nmp);
7640 		NFSUNLOCKMNT(nmp);
7641 		return (0);
7642 	}
7643 	return (ENXIO);
7644 }
7645 
7646 /*
7647  * Mark a DS as disabled by setting nfsdev_nmp = NULL.
7648  */
7649 static void
7650 nfsrv_deleteds(struct nfsdevice *fndds)
7651 {
7652 
7653 	NFSD_DEBUG(4, "deleteds: deleting a mirror\n");
7654 	fndds->nfsdev_nmp = NULL;
7655 	if (fndds->nfsdev_mdsisset == 0)
7656 		nfsrv_faildscnt--;
7657 }
7658 
7659 /*
7660  * Fill in the addr structures for the File and Flex File layouts.
7661  */
7662 static void
7663 nfsrv_allocdevid(struct nfsdevice *ds, char *addr, char *dnshost)
7664 {
7665 	uint32_t *tl;
7666 	char *netprot;
7667 	int addrlen;
7668 	static uint64_t new_devid = 0;
7669 
7670 	if (strchr(addr, ':') != NULL)
7671 		netprot = "tcp6";
7672 	else
7673 		netprot = "tcp";
7674 
7675 	/* Fill in the device id. */
7676 	NFSBCOPY(&nfsdev_time, ds->nfsdev_deviceid, sizeof(nfsdev_time));
7677 	new_devid++;
7678 	NFSBCOPY(&new_devid, &ds->nfsdev_deviceid[sizeof(nfsdev_time)],
7679 	    sizeof(new_devid));
7680 
7681 	/*
7682 	 * Fill in the file addr (actually the nfsv4_file_layout_ds_addr4
7683 	 * as defined in RFC5661) in XDR.
7684 	 */
7685 	addrlen = NFSM_RNDUP(strlen(addr)) + NFSM_RNDUP(strlen(netprot)) +
7686 	    6 * NFSX_UNSIGNED;
7687 	NFSD_DEBUG(4, "hn=%s addr=%s netprot=%s\n", dnshost, addr, netprot);
7688 	ds->nfsdev_fileaddrlen = addrlen;
7689 	tl = malloc(addrlen, M_NFSDSTATE, M_WAITOK | M_ZERO);
7690 	ds->nfsdev_fileaddr = (char *)tl;
7691 	*tl++ = txdr_unsigned(1);		/* One stripe with index 0. */
7692 	*tl++ = 0;
7693 	*tl++ = txdr_unsigned(1);		/* One multipath list */
7694 	*tl++ = txdr_unsigned(1);		/* with one entry in it. */
7695 	/* The netaddr for this one entry. */
7696 	*tl++ = txdr_unsigned(strlen(netprot));
7697 	NFSBCOPY(netprot, tl, strlen(netprot));
7698 	tl += (NFSM_RNDUP(strlen(netprot)) / NFSX_UNSIGNED);
7699 	*tl++ = txdr_unsigned(strlen(addr));
7700 	NFSBCOPY(addr, tl, strlen(addr));
7701 
7702 	/*
7703 	 * Fill in the flex file addr (actually the ff_device_addr4
7704 	 * as defined for Flexible File Layout) in XDR.
7705 	 */
7706 	addrlen = NFSM_RNDUP(strlen(addr)) + NFSM_RNDUP(strlen(netprot)) +
7707 	    9 * NFSX_UNSIGNED;
7708 	ds->nfsdev_flexaddrlen = addrlen;
7709 	tl = malloc(addrlen, M_NFSDSTATE, M_WAITOK | M_ZERO);
7710 	ds->nfsdev_flexaddr = (char *)tl;
7711 	*tl++ = txdr_unsigned(1);		/* One multipath entry. */
7712 	/* The netaddr for this one entry. */
7713 	*tl++ = txdr_unsigned(strlen(netprot));
7714 	NFSBCOPY(netprot, tl, strlen(netprot));
7715 	tl += (NFSM_RNDUP(strlen(netprot)) / NFSX_UNSIGNED);
7716 	*tl++ = txdr_unsigned(strlen(addr));
7717 	NFSBCOPY(addr, tl, strlen(addr));
7718 	tl += (NFSM_RNDUP(strlen(addr)) / NFSX_UNSIGNED);
7719 	*tl++ = txdr_unsigned(1);		/* One NFS Version. */
7720 	*tl++ = txdr_unsigned(NFS_VER4);	/* NFSv4. */
7721 	*tl++ = txdr_unsigned(NFSV41_MINORVERSION); /* Minor version 1. */
7722 	*tl++ = txdr_unsigned(NFS_SRVMAXIO);	/* DS max rsize. */
7723 	*tl++ = txdr_unsigned(NFS_SRVMAXIO);	/* DS max wsize. */
7724 	*tl = newnfs_true;			/* Tightly coupled. */
7725 
7726 	ds->nfsdev_hostnamelen = strlen(dnshost);
7727 	ds->nfsdev_host = malloc(ds->nfsdev_hostnamelen + 1, M_NFSDSTATE,
7728 	    M_WAITOK);
7729 	NFSBCOPY(dnshost, ds->nfsdev_host, ds->nfsdev_hostnamelen + 1);
7730 }
7731 
7732 
7733 /*
7734  * Create the device id list.
7735  * Return 0 if the nfsd threads are to run and ENXIO if the "-p" argument
7736  * is misconfigured.
7737  */
7738 int
7739 nfsrv_createdevids(struct nfsd_nfsd_args *args, NFSPROC_T *p)
7740 {
7741 	struct nfsdevice *ds;
7742 	char *addrp, *dnshostp, *dspathp, *mdspathp;
7743 	int error, i;
7744 
7745 	addrp = args->addr;
7746 	dnshostp = args->dnshost;
7747 	dspathp = args->dspath;
7748 	mdspathp = args->mdspath;
7749 	nfsrv_maxpnfsmirror = args->mirrorcnt;
7750 	if (addrp == NULL || dnshostp == NULL || dspathp == NULL ||
7751 	    mdspathp == NULL)
7752 		return (0);
7753 
7754 	/*
7755 	 * Loop around for each nul-terminated string in args->addr,
7756 	 * args->dnshost, args->dnspath and args->mdspath.
7757 	 */
7758 	while (addrp < (args->addr + args->addrlen) &&
7759 	    dnshostp < (args->dnshost + args->dnshostlen) &&
7760 	    dspathp < (args->dspath + args->dspathlen) &&
7761 	    mdspathp < (args->mdspath + args->mdspathlen)) {
7762 		error = nfsrv_setdsserver(dspathp, mdspathp, p, &ds);
7763 		if (error != 0) {
7764 			/* Free all DS servers. */
7765 			nfsrv_freealldevids();
7766 			nfsrv_devidcnt = 0;
7767 			return (ENXIO);
7768 		}
7769 		nfsrv_allocdevid(ds, addrp, dnshostp);
7770 		addrp += (strlen(addrp) + 1);
7771 		dnshostp += (strlen(dnshostp) + 1);
7772 		dspathp += (strlen(dspathp) + 1);
7773 		mdspathp += (strlen(mdspathp) + 1);
7774 	}
7775 	if (nfsrv_devidcnt < nfsrv_maxpnfsmirror) {
7776 		/* Free all DS servers. */
7777 		nfsrv_freealldevids();
7778 		nfsrv_devidcnt = 0;
7779 		nfsrv_maxpnfsmirror = 1;
7780 		return (ENXIO);
7781 	}
7782 	/* We can fail at most one less DS than the mirror level. */
7783 	nfsrv_faildscnt = nfsrv_maxpnfsmirror - 1;
7784 
7785 	/*
7786 	 * Allocate the nfslayout hash table now, since this is a pNFS server.
7787 	 * Make it 1% of the high water mark and at least 100.
7788 	 */
7789 	if (nfslayouthash == NULL) {
7790 		nfsrv_layouthashsize = nfsrv_layouthighwater / 100;
7791 		if (nfsrv_layouthashsize < 100)
7792 			nfsrv_layouthashsize = 100;
7793 		nfslayouthash = mallocarray(nfsrv_layouthashsize,
7794 		    sizeof(struct nfslayouthash), M_NFSDSESSION, M_WAITOK |
7795 		    M_ZERO);
7796 		for (i = 0; i < nfsrv_layouthashsize; i++) {
7797 			mtx_init(&nfslayouthash[i].mtx, "nfslm", NULL, MTX_DEF);
7798 			TAILQ_INIT(&nfslayouthash[i].list);
7799 		}
7800 	}
7801 	return (0);
7802 }
7803 
7804 /*
7805  * Free all device ids.
7806  */
7807 static void
7808 nfsrv_freealldevids(void)
7809 {
7810 	struct nfsdevice *ds, *nds;
7811 
7812 	TAILQ_FOREACH_SAFE(ds, &nfsrv_devidhead, nfsdev_list, nds)
7813 		nfsrv_freedevid(ds);
7814 }
7815 
7816 /*
7817  * Check to see if there is a Read/Write Layout plus either:
7818  * - A Write Delegation
7819  * or
7820  * - An Open with Write_access.
7821  * Return 1 if this is the case and 0 otherwise.
7822  * This function is used by nfsrv_proxyds() to decide if doing a Proxy
7823  * Getattr RPC to the Data Server (DS) is necessary.
7824  */
7825 #define	NFSCLIDVECSIZE	6
7826 APPLESTATIC int
7827 nfsrv_checkdsattr(struct nfsrv_descript *nd, vnode_t vp, NFSPROC_T *p)
7828 {
7829 	fhandle_t fh, *tfhp;
7830 	struct nfsstate *stp;
7831 	struct nfslayout *lyp;
7832 	struct nfslayouthash *lhyp;
7833 	struct nfslockhashhead *hp;
7834 	struct nfslockfile *lfp;
7835 	nfsquad_t clid[NFSCLIDVECSIZE];
7836 	int clidcnt, ret;
7837 
7838 	ret = nfsvno_getfh(vp, &fh, p);
7839 	if (ret != 0)
7840 		return (0);
7841 
7842 	/* First check for a Read/Write Layout. */
7843 	clidcnt = 0;
7844 	lhyp = NFSLAYOUTHASH(&fh);
7845 	NFSLOCKLAYOUT(lhyp);
7846 	TAILQ_FOREACH(lyp, &lhyp->list, lay_list) {
7847 		if (NFSBCMP(&lyp->lay_fh, &fh, sizeof(fh)) == 0 &&
7848 		    ((lyp->lay_flags & NFSLAY_RW) != 0 ||
7849 		     ((lyp->lay_flags & NFSLAY_READ) != 0 &&
7850 		      nfsrv_pnfsatime != 0))) {
7851 			if (clidcnt < NFSCLIDVECSIZE)
7852 				clid[clidcnt].qval = lyp->lay_clientid.qval;
7853 			clidcnt++;
7854 		}
7855 	}
7856 	NFSUNLOCKLAYOUT(lhyp);
7857 	if (clidcnt == 0) {
7858 		/* None found, so return 0. */
7859 		return (0);
7860 	}
7861 
7862 	/* Get the nfslockfile for this fh. */
7863 	NFSLOCKSTATE();
7864 	hp = NFSLOCKHASH(&fh);
7865 	LIST_FOREACH(lfp, hp, lf_hash) {
7866 		tfhp = &lfp->lf_fh;
7867 		if (NFSVNO_CMPFH(&fh, tfhp))
7868 			break;
7869 	}
7870 	if (lfp == NULL) {
7871 		/* None found, so return 0. */
7872 		NFSUNLOCKSTATE();
7873 		return (0);
7874 	}
7875 
7876 	/* Now, look for a Write delegation for this clientid. */
7877 	LIST_FOREACH(stp, &lfp->lf_deleg, ls_file) {
7878 		if ((stp->ls_flags & NFSLCK_DELEGWRITE) != 0 &&
7879 		    nfsrv_fndclid(clid, stp->ls_clp->lc_clientid, clidcnt) != 0)
7880 			break;
7881 	}
7882 	if (stp != NULL) {
7883 		/* Found one, so return 1. */
7884 		NFSUNLOCKSTATE();
7885 		return (1);
7886 	}
7887 
7888 	/* No Write delegation, so look for an Open with Write_access. */
7889 	LIST_FOREACH(stp, &lfp->lf_open, ls_file) {
7890 		KASSERT((stp->ls_flags & NFSLCK_OPEN) != 0,
7891 		    ("nfsrv_checkdsattr: Non-open in Open list\n"));
7892 		if ((stp->ls_flags & NFSLCK_WRITEACCESS) != 0 &&
7893 		    nfsrv_fndclid(clid, stp->ls_clp->lc_clientid, clidcnt) != 0)
7894 			break;
7895 	}
7896 	NFSUNLOCKSTATE();
7897 	if (stp != NULL)
7898 		return (1);
7899 	return (0);
7900 }
7901 
7902 /*
7903  * Look for a matching clientid in the vector. Return 1 if one might match.
7904  */
7905 static int
7906 nfsrv_fndclid(nfsquad_t *clidvec, nfsquad_t clid, int clidcnt)
7907 {
7908 	int i;
7909 
7910 	/* If too many for the vector, return 1 since there might be a match. */
7911 	if (clidcnt > NFSCLIDVECSIZE)
7912 		return (1);
7913 
7914 	for (i = 0; i < clidcnt; i++)
7915 		if (clidvec[i].qval == clid.qval)
7916 			return (1);
7917 	return (0);
7918 }
7919 
7920 /*
7921  * Check the don't list for "vp" and see if issuing an rw layout is allowed.
7922  * Return 1 if issuing an rw layout isn't allowed, 0 otherwise.
7923  */
7924 static int
7925 nfsrv_dontlayout(fhandle_t *fhp)
7926 {
7927 	struct nfsdontlist *mrp;
7928 	int ret;
7929 
7930 	if (nfsrv_dontlistlen == 0)
7931 		return (0);
7932 	ret = 0;
7933 	NFSDDONTLISTLOCK();
7934 	LIST_FOREACH(mrp, &nfsrv_dontlisthead, nfsmr_list) {
7935 		if (NFSBCMP(fhp, &mrp->nfsmr_fh, sizeof(*fhp)) == 0 &&
7936 		    (mrp->nfsmr_flags & NFSMR_DONTLAYOUT) != 0) {
7937 			ret = 1;
7938 			break;
7939 		}
7940 	}
7941 	NFSDDONTLISTUNLOCK();
7942 	return (ret);
7943 }
7944 
7945 #define	PNFSDS_COPYSIZ	65536
7946 /*
7947  * Create a new file on a DS and copy the contents of an extant DS file to it.
7948  * This can be used for recovery of a DS file onto a recovered DS.
7949  * The steps are:
7950  * - When called, the MDS file's vnode is locked, blocking LayoutGet operations.
7951  * - Disable issuing of read/write layouts for the file via the nfsdontlist,
7952  *   so that they will be disabled after the MDS file's vnode is unlocked.
7953  * - Set up the nfsrv_recalllist so that recall of read/write layouts can
7954  *   be done.
7955  * - Unlock the MDS file's vnode, so that the client(s) can perform proxied
7956  *   writes, LayoutCommits and LayoutReturns for the file when completing the
7957  *   LayoutReturn requested by the LayoutRecall callback.
7958  * - Issue a LayoutRecall callback for all read/write layouts and wait for
7959  *   them to be returned. (If the LayoutRecall callback replies
7960  *   NFSERR_NOMATCHLAYOUT, they are gone and no LayoutReturn is needed.)
7961  * - Exclusively lock the MDS file's vnode.  This ensures that no proxied
7962  *   writes are in progress or can occur during the DS file copy.
7963  *   It also blocks Setattr operations.
7964  * - Create the file on the recovered mirror.
7965  * - Copy the file from the operational DS.
7966  * - Copy any ACL from the MDS file to the new DS file.
7967  * - Set the modify time of the new DS file to that of the MDS file.
7968  * - Update the extended attribute for the MDS file.
7969  * - Enable issuing of rw layouts by deleting the nfsdontlist entry.
7970  * - The caller will unlock the MDS file's vnode allowing operations
7971  *   to continue normally, since it is now on the mirror again.
7972  */
7973 int
7974 nfsrv_copymr(vnode_t vp, vnode_t fvp, vnode_t dvp, struct nfsdevice *ds,
7975     struct pnfsdsfile *pf, struct pnfsdsfile *wpf, int mirrorcnt,
7976     struct ucred *cred, NFSPROC_T *p)
7977 {
7978 	struct nfsdontlist *mrp, *nmrp;
7979 	struct nfslayouthash *lhyp;
7980 	struct nfslayout *lyp, *nlyp;
7981 	struct nfslayouthead thl;
7982 	struct mount *mp, *tvmp;
7983 	struct acl *aclp;
7984 	struct vattr va;
7985 	struct timespec mtime;
7986 	fhandle_t fh;
7987 	vnode_t tvp;
7988 	off_t rdpos, wrpos;
7989 	ssize_t aresid;
7990 	char *dat;
7991 	int didprintf, ret, retacl, xfer;
7992 
7993 	ASSERT_VOP_LOCKED(fvp, "nfsrv_copymr fvp");
7994 	ASSERT_VOP_LOCKED(vp, "nfsrv_copymr vp");
7995 	/*
7996 	 * Allocate a nfsdontlist entry and set the NFSMR_DONTLAYOUT flag
7997 	 * so that no more RW layouts will get issued.
7998 	 */
7999 	ret = nfsvno_getfh(vp, &fh, p);
8000 	if (ret != 0) {
8001 		NFSD_DEBUG(4, "nfsrv_copymr: getfh=%d\n", ret);
8002 		return (ret);
8003 	}
8004 	nmrp = malloc(sizeof(*nmrp), M_NFSDSTATE, M_WAITOK);
8005 	nmrp->nfsmr_flags = NFSMR_DONTLAYOUT;
8006 	NFSBCOPY(&fh, &nmrp->nfsmr_fh, sizeof(fh));
8007 	NFSDDONTLISTLOCK();
8008 	LIST_FOREACH(mrp, &nfsrv_dontlisthead, nfsmr_list) {
8009 		if (NFSBCMP(&fh, &mrp->nfsmr_fh, sizeof(fh)) == 0)
8010 			break;
8011 	}
8012 	if (mrp == NULL) {
8013 		LIST_INSERT_HEAD(&nfsrv_dontlisthead, nmrp, nfsmr_list);
8014 		mrp = nmrp;
8015 		nmrp = NULL;
8016 		nfsrv_dontlistlen++;
8017 		NFSD_DEBUG(4, "nfsrv_copymr: in dontlist\n");
8018 	} else {
8019 		NFSDDONTLISTUNLOCK();
8020 		free(nmrp, M_NFSDSTATE);
8021 		NFSD_DEBUG(4, "nfsrv_copymr: dup dontlist\n");
8022 		return (ENXIO);
8023 	}
8024 	NFSDDONTLISTUNLOCK();
8025 
8026 	/*
8027 	 * Search for all RW layouts for this file.  Move them to the
8028 	 * recall list, so they can be recalled and their return noted.
8029 	 */
8030 	lhyp = NFSLAYOUTHASH(&fh);
8031 	NFSDRECALLLOCK();
8032 	NFSLOCKLAYOUT(lhyp);
8033 	TAILQ_FOREACH_SAFE(lyp, &lhyp->list, lay_list, nlyp) {
8034 		if (NFSBCMP(&lyp->lay_fh, &fh, sizeof(fh)) == 0 &&
8035 		    (lyp->lay_flags & NFSLAY_RW) != 0) {
8036 			TAILQ_REMOVE(&lhyp->list, lyp, lay_list);
8037 			TAILQ_INSERT_HEAD(&nfsrv_recalllisthead, lyp, lay_list);
8038 			lyp->lay_trycnt = 0;
8039 		}
8040 	}
8041 	NFSUNLOCKLAYOUT(lhyp);
8042 	NFSDRECALLUNLOCK();
8043 
8044 	ret = 0;
8045 	mp = tvmp = NULL;
8046 	didprintf = 0;
8047 	TAILQ_INIT(&thl);
8048 	/* Unlock the MDS vp, so that a LayoutReturn can be done on it. */
8049 	NFSVOPUNLOCK(vp, 0);
8050 	/* Now, do a recall for all layouts not yet recalled. */
8051 tryagain:
8052 	NFSDRECALLLOCK();
8053 	TAILQ_FOREACH(lyp, &nfsrv_recalllisthead, lay_list) {
8054 		if (NFSBCMP(&lyp->lay_fh, &fh, sizeof(fh)) == 0 &&
8055 		    (lyp->lay_flags & NFSLAY_RECALL) == 0) {
8056 			lyp->lay_flags |= NFSLAY_RECALL;
8057 			/*
8058 			 * The layout stateid.seqid needs to be incremented
8059 			 * before doing a LAYOUT_RECALL callback.
8060 			 */
8061 			if (++lyp->lay_stateid.seqid == 0)
8062 				lyp->lay_stateid.seqid = 1;
8063 			NFSDRECALLUNLOCK();
8064 			nfsrv_recalllayout(lyp->lay_clientid, &lyp->lay_stateid,
8065 			    &lyp->lay_fh, lyp, 0, lyp->lay_type, p);
8066 			NFSD_DEBUG(4, "nfsrv_copymr: recalled layout\n");
8067 			goto tryagain;
8068 		}
8069 	}
8070 
8071 	/* Now wait for them to be returned. */
8072 tryagain2:
8073 	TAILQ_FOREACH(lyp, &nfsrv_recalllisthead, lay_list) {
8074 		if (NFSBCMP(&lyp->lay_fh, &fh, sizeof(fh)) == 0) {
8075 			if ((lyp->lay_flags & NFSLAY_RETURNED) != 0) {
8076 				TAILQ_REMOVE(&nfsrv_recalllisthead, lyp,
8077 				    lay_list);
8078 				TAILQ_INSERT_HEAD(&thl, lyp, lay_list);
8079 				NFSD_DEBUG(4,
8080 				    "nfsrv_copymr: layout returned\n");
8081 			} else {
8082 				lyp->lay_trycnt++;
8083 				ret = mtx_sleep(lyp, NFSDRECALLMUTEXPTR,
8084 				    PVFS | PCATCH, "nfsmrl", hz);
8085 				NFSD_DEBUG(4, "nfsrv_copymr: aft sleep=%d\n",
8086 				    ret);
8087 				if (ret == EINTR || ret == ERESTART)
8088 					break;
8089 				if ((lyp->lay_flags & NFSLAY_RETURNED) == 0) {
8090 					/*
8091 					 * Give up after 60sec and return
8092 					 * ENXIO, failing the copymr.
8093 					 * This layout will remain on the
8094 					 * recalllist.  It can only be cleared
8095 					 * by restarting the nfsd.
8096 					 * This seems the safe way to handle
8097 					 * it, since it cannot be safely copied
8098 					 * with an outstanding RW layout.
8099 					 */
8100 					if (lyp->lay_trycnt >= 60) {
8101 						ret = ENXIO;
8102 						break;
8103 					}
8104 					if (didprintf == 0) {
8105 						printf("nfsrv_copymr: layout "
8106 						    "not returned\n");
8107 						didprintf = 1;
8108 					}
8109 				}
8110 			}
8111 			goto tryagain2;
8112 		}
8113 	}
8114 	NFSDRECALLUNLOCK();
8115 	/* We can now get rid of the layouts that have been returned. */
8116 	TAILQ_FOREACH_SAFE(lyp, &thl, lay_list, nlyp)
8117 		nfsrv_freelayout(&thl, lyp);
8118 
8119 	/*
8120 	 * Do the vn_start_write() calls here, before the MDS vnode is
8121 	 * locked and the tvp is created (locked) in the NFS file system
8122 	 * that dvp is in.
8123 	 * For tvmp, this probably isn't necessary, since it will be an
8124 	 * NFS mount and they are not suspendable at this time.
8125 	 */
8126 	if (ret == 0)
8127 		ret = vn_start_write(vp, &mp, V_WAIT | PCATCH);
8128 	if (ret == 0) {
8129 		tvmp = dvp->v_mount;
8130 		ret = vn_start_write(NULL, &tvmp, V_WAIT | PCATCH);
8131 	}
8132 
8133 	/*
8134 	 * LK_EXCLUSIVE lock the MDS vnode, so that any
8135 	 * proxied writes through the MDS will be blocked until we have
8136 	 * completed the copy and update of the extended attributes.
8137 	 * This will also ensure that any attributes and ACL will not be
8138 	 * changed until the copy is complete.
8139 	 */
8140 	NFSVOPLOCK(vp, LK_EXCLUSIVE | LK_RETRY);
8141 	if (ret == 0 && (vp->v_iflag & VI_DOOMED) != 0) {
8142 		NFSD_DEBUG(4, "nfsrv_copymr: lk_exclusive doomed\n");
8143 		ret = ESTALE;
8144 	}
8145 
8146 	/* Create the data file on the recovered DS. */
8147 	if (ret == 0)
8148 		ret = nfsrv_createdsfile(vp, &fh, pf, dvp, ds, cred, p, &tvp);
8149 
8150 	/* Copy the DS file, if created successfully. */
8151 	if (ret == 0) {
8152 		/*
8153 		 * Get any NFSv4 ACL on the MDS file, so that it can be set
8154 		 * on the new DS file.
8155 		 */
8156 		aclp = acl_alloc(M_WAITOK | M_ZERO);
8157 		retacl = VOP_GETACL(vp, ACL_TYPE_NFS4, aclp, cred, p);
8158 		if (retacl != 0 && retacl != ENOATTR)
8159 			NFSD_DEBUG(1, "nfsrv_copymr: vop_getacl=%d\n", retacl);
8160 		dat = malloc(PNFSDS_COPYSIZ, M_TEMP, M_WAITOK);
8161 		/* Malloc a block of 0s used to check for holes. */
8162 		if (nfsrv_zeropnfsdat == NULL)
8163 			nfsrv_zeropnfsdat = malloc(PNFSDS_COPYSIZ, M_TEMP,
8164 			    M_WAITOK | M_ZERO);
8165 		rdpos = wrpos = 0;
8166 		ret = VOP_GETATTR(fvp, &va, cred);
8167 		aresid = 0;
8168 		while (ret == 0 && aresid == 0) {
8169 			ret = vn_rdwr(UIO_READ, fvp, dat, PNFSDS_COPYSIZ,
8170 			    rdpos, UIO_SYSSPACE, IO_NODELOCKED, cred, NULL,
8171 			    &aresid, p);
8172 			xfer = PNFSDS_COPYSIZ - aresid;
8173 			if (ret == 0 && xfer > 0) {
8174 				rdpos += xfer;
8175 				/*
8176 				 * Skip the write for holes, except for the
8177 				 * last block.
8178 				 */
8179 				if (xfer < PNFSDS_COPYSIZ || rdpos ==
8180 				    va.va_size || NFSBCMP(dat,
8181 				    nfsrv_zeropnfsdat, PNFSDS_COPYSIZ) != 0)
8182 					ret = vn_rdwr(UIO_WRITE, tvp, dat, xfer,
8183 					    wrpos, UIO_SYSSPACE, IO_NODELOCKED,
8184 					    cred, NULL, NULL, p);
8185 				if (ret == 0)
8186 					wrpos += xfer;
8187 			}
8188 		}
8189 
8190 		/* If there is an ACL and the copy succeeded, set the ACL. */
8191 		if (ret == 0 && retacl == 0) {
8192 			ret = VOP_SETACL(tvp, ACL_TYPE_NFS4, aclp, cred, p);
8193 			/*
8194 			 * Don't consider these as errors, since VOP_GETACL()
8195 			 * can return an ACL when they are not actually
8196 			 * supported.  For example, for UFS, VOP_GETACL()
8197 			 * will return a trivial ACL based on the uid/gid/mode
8198 			 * when there is no ACL on the file.
8199 			 * This case should be recognized as a trivial ACL
8200 			 * by UFS's VOP_SETACL() and succeed, but...
8201 			 */
8202 			if (ret == ENOATTR || ret == EOPNOTSUPP || ret == EPERM)
8203 				ret = 0;
8204 		}
8205 
8206 		if (ret == 0)
8207 			ret = VOP_FSYNC(tvp, MNT_WAIT, p);
8208 
8209 		/* Set the DS data file's modify time that of the MDS file. */
8210 		if (ret == 0)
8211 			ret = VOP_GETATTR(vp, &va, cred);
8212 		if (ret == 0) {
8213 			mtime = va.va_mtime;
8214 			VATTR_NULL(&va);
8215 			va.va_mtime = mtime;
8216 			ret = VOP_SETATTR(tvp, &va, cred);
8217 		}
8218 
8219 		vput(tvp);
8220 		acl_free(aclp);
8221 		free(dat, M_TEMP);
8222 	}
8223 	if (tvmp != NULL)
8224 		vn_finished_write(tvmp);
8225 
8226 	/* Update the extended attributes for the newly created DS file. */
8227 	if (ret == 0)
8228 		ret = vn_extattr_set(vp, IO_NODELOCKED,
8229 		    EXTATTR_NAMESPACE_SYSTEM, "pnfsd.dsfile",
8230 		    sizeof(*wpf) * mirrorcnt, (char *)wpf, p);
8231 	if (mp != NULL)
8232 		vn_finished_write(mp);
8233 
8234 	/* Get rid of the dontlist entry, so that Layouts can be issued. */
8235 	NFSDDONTLISTLOCK();
8236 	LIST_REMOVE(mrp, nfsmr_list);
8237 	NFSDDONTLISTUNLOCK();
8238 	free(mrp, M_NFSDSTATE);
8239 	return (ret);
8240 }
8241 
8242 /*
8243  * Create a data storage file on the recovered DS.
8244  */
8245 static int
8246 nfsrv_createdsfile(vnode_t vp, fhandle_t *fhp, struct pnfsdsfile *pf,
8247     vnode_t dvp, struct nfsdevice *ds, struct ucred *cred, NFSPROC_T *p,
8248     vnode_t *tvpp)
8249 {
8250 	struct vattr va, nva;
8251 	int error;
8252 
8253 	/* Make data file name based on FH. */
8254 	error = VOP_GETATTR(vp, &va, cred);
8255 	if (error == 0) {
8256 		/* Set the attributes for "vp" to Setattr the DS vp. */
8257 		VATTR_NULL(&nva);
8258 		nva.va_uid = va.va_uid;
8259 		nva.va_gid = va.va_gid;
8260 		nva.va_mode = va.va_mode;
8261 		nva.va_size = 0;
8262 		VATTR_NULL(&va);
8263 		va.va_type = VREG;
8264 		va.va_mode = nva.va_mode;
8265 		NFSD_DEBUG(4, "nfsrv_dscreatefile: dvp=%p pf=%p\n", dvp, pf);
8266 		error = nfsrv_dscreate(dvp, &va, &nva, fhp, pf, NULL,
8267 		    pf->dsf_filename, cred, p, tvpp);
8268 	}
8269 	return (error);
8270 }
8271 
8272 /*
8273  * Look up the MDS file shared locked, and then get the extended attribute
8274  * to find the extant DS file to be copied to the new mirror.
8275  * If successful, *vpp is set to the MDS file's vp and *nvpp is
8276  * set to a DS data file for the MDS file, both exclusively locked.
8277  * The "buf" argument has the pnfsdsfile structure from the MDS file
8278  * in it and buflen is set to its length.
8279  */
8280 int
8281 nfsrv_mdscopymr(char *mdspathp, char *dspathp, char *curdspathp, char *buf,
8282     int *buflenp, char *fname, NFSPROC_T *p, struct vnode **vpp,
8283     struct vnode **nvpp, struct pnfsdsfile **pfp, struct nfsdevice **dsp,
8284     struct nfsdevice **fdsp)
8285 {
8286 	struct nameidata nd;
8287 	struct vnode *vp, *curvp;
8288 	struct pnfsdsfile *pf;
8289 	struct nfsmount *nmp, *curnmp;
8290 	int dsdir, error, mirrorcnt, ippos;
8291 
8292 	vp = NULL;
8293 	curvp = NULL;
8294 	curnmp = NULL;
8295 	*dsp = NULL;
8296 	*fdsp = NULL;
8297 	if (dspathp == NULL && curdspathp != NULL)
8298 		return (EPERM);
8299 
8300 	/*
8301 	 * Look up the MDS file shared locked.  The lock will be upgraded
8302 	 * to an exclusive lock after any rw layouts have been returned.
8303 	 */
8304 	NFSD_DEBUG(4, "mdsopen path=%s\n", mdspathp);
8305 	NDINIT(&nd, LOOKUP, FOLLOW | LOCKSHARED | LOCKLEAF, UIO_SYSSPACE,
8306 	    mdspathp, p);
8307 	error = namei(&nd);
8308 	NFSD_DEBUG(4, "lookup=%d\n", error);
8309 	if (error != 0)
8310 		return (error);
8311 	if (nd.ni_vp->v_type != VREG) {
8312 		vput(nd.ni_vp);
8313 		NFSD_DEBUG(4, "mdspath not reg\n");
8314 		return (EISDIR);
8315 	}
8316 	vp = nd.ni_vp;
8317 
8318 	if (curdspathp != NULL) {
8319 		/*
8320 		 * Look up the current DS path and find the nfsdev structure for
8321 		 * it.
8322 		 */
8323 		NFSD_DEBUG(4, "curmdsdev path=%s\n", curdspathp);
8324 		NDINIT(&nd, LOOKUP, FOLLOW | LOCKSHARED | LOCKLEAF,
8325 		    UIO_SYSSPACE, curdspathp, p);
8326 		error = namei(&nd);
8327 		NFSD_DEBUG(4, "ds lookup=%d\n", error);
8328 		if (error != 0) {
8329 			vput(vp);
8330 			return (error);
8331 		}
8332 		if (nd.ni_vp->v_type != VDIR) {
8333 			vput(nd.ni_vp);
8334 			vput(vp);
8335 			NFSD_DEBUG(4, "curdspath not dir\n");
8336 			return (ENOTDIR);
8337 		}
8338 		if (strcmp(nd.ni_vp->v_mount->mnt_vfc->vfc_name, "nfs") != 0) {
8339 			vput(nd.ni_vp);
8340 			vput(vp);
8341 			NFSD_DEBUG(4, "curdspath not an NFS mount\n");
8342 			return (ENXIO);
8343 		}
8344 		curnmp = VFSTONFS(nd.ni_vp->v_mount);
8345 
8346 		/* Search the nfsdev list for a match. */
8347 		NFSDDSLOCK();
8348 		*fdsp = nfsv4_findmirror(curnmp);
8349 		NFSDDSUNLOCK();
8350 		if (*fdsp == NULL)
8351 			curnmp = NULL;
8352 		if (curnmp == NULL) {
8353 			vput(nd.ni_vp);
8354 			vput(vp);
8355 			NFSD_DEBUG(4, "mdscopymr: no current ds\n");
8356 			return (ENXIO);
8357 		}
8358 		curvp = nd.ni_vp;
8359 	}
8360 
8361 	if (dspathp != NULL) {
8362 		/* Look up the nfsdev path and find the nfsdev structure. */
8363 		NFSD_DEBUG(4, "mdsdev path=%s\n", dspathp);
8364 		NDINIT(&nd, LOOKUP, FOLLOW | LOCKSHARED | LOCKLEAF,
8365 		    UIO_SYSSPACE, dspathp, p);
8366 		error = namei(&nd);
8367 		NFSD_DEBUG(4, "ds lookup=%d\n", error);
8368 		if (error != 0) {
8369 			vput(vp);
8370 			if (curvp != NULL)
8371 				vput(curvp);
8372 			return (error);
8373 		}
8374 		if (nd.ni_vp->v_type != VDIR || nd.ni_vp == curvp) {
8375 			vput(nd.ni_vp);
8376 			vput(vp);
8377 			if (curvp != NULL)
8378 				vput(curvp);
8379 			NFSD_DEBUG(4, "dspath not dir\n");
8380 			if (nd.ni_vp == curvp)
8381 				return (EPERM);
8382 			return (ENOTDIR);
8383 		}
8384 		if (strcmp(nd.ni_vp->v_mount->mnt_vfc->vfc_name, "nfs") != 0) {
8385 			vput(nd.ni_vp);
8386 			vput(vp);
8387 			if (curvp != NULL)
8388 				vput(curvp);
8389 			NFSD_DEBUG(4, "dspath not an NFS mount\n");
8390 			return (ENXIO);
8391 		}
8392 		nmp = VFSTONFS(nd.ni_vp->v_mount);
8393 
8394 		/*
8395 		 * Search the nfsdevice list for a match.  If curnmp == NULL,
8396 		 * this is a recovery and there must be a mirror.
8397 		 */
8398 		NFSDDSLOCK();
8399 		if (curnmp == NULL)
8400 			*dsp = nfsrv_findmirroredds(nmp);
8401 		else
8402 			*dsp = nfsv4_findmirror(nmp);
8403 		NFSDDSUNLOCK();
8404 		if (*dsp == NULL) {
8405 			vput(nd.ni_vp);
8406 			vput(vp);
8407 			if (curvp != NULL)
8408 				vput(curvp);
8409 			NFSD_DEBUG(4, "mdscopymr: no ds\n");
8410 			return (ENXIO);
8411 		}
8412 	} else {
8413 		nd.ni_vp = NULL;
8414 		nmp = NULL;
8415 	}
8416 
8417 	/*
8418 	 * Get a vp for an available DS data file using the extended
8419 	 * attribute on the MDS file.
8420 	 * If there is a valid entry for the new DS in the extended attribute
8421 	 * on the MDS file (as checked via the nmp argument),
8422 	 * nfsrv_dsgetsockmnt() returns EEXIST, so no copying will occur.
8423 	 */
8424 	error = nfsrv_dsgetsockmnt(vp, 0, buf, buflenp, &mirrorcnt, p,
8425 	    NULL, NULL, NULL, fname, nvpp, &nmp, curnmp, &ippos, &dsdir);
8426 	if (curvp != NULL)
8427 		vput(curvp);
8428 	if (nd.ni_vp == NULL) {
8429 		if (error == 0 && nmp != NULL) {
8430 			/* Search the nfsdev list for a match. */
8431 			NFSDDSLOCK();
8432 			*dsp = nfsrv_findmirroredds(nmp);
8433 			NFSDDSUNLOCK();
8434 		}
8435 		if (error == 0 && (nmp == NULL || *dsp == NULL)) {
8436 			if (nvpp != NULL && *nvpp != NULL) {
8437 				vput(*nvpp);
8438 				*nvpp = NULL;
8439 			}
8440 			error = ENXIO;
8441 		}
8442 	} else
8443 		vput(nd.ni_vp);
8444 
8445 	/*
8446 	 * When dspathp != NULL and curdspathp == NULL, this is a recovery
8447 	 * and is only allowed if there is a 0.0.0.0 IP address entry.
8448 	 * When curdspathp != NULL, the ippos will be set to that entry.
8449 	 */
8450 	if (error == 0 && dspathp != NULL && ippos == -1) {
8451 		if (nvpp != NULL && *nvpp != NULL) {
8452 			vput(*nvpp);
8453 			*nvpp = NULL;
8454 		}
8455 		error = ENXIO;
8456 	}
8457 	if (error == 0) {
8458 		*vpp = vp;
8459 
8460 		pf = (struct pnfsdsfile *)buf;
8461 		if (ippos == -1) {
8462 			/* If no zeroip pnfsdsfile, add one. */
8463 			ippos = *buflenp / sizeof(*pf);
8464 			*buflenp += sizeof(*pf);
8465 			pf += ippos;
8466 			pf->dsf_dir = dsdir;
8467 			strlcpy(pf->dsf_filename, fname,
8468 			    sizeof(pf->dsf_filename));
8469 		} else
8470 			pf += ippos;
8471 		*pfp = pf;
8472 	} else
8473 		vput(vp);
8474 	return (error);
8475 }
8476 
8477 /*
8478  * Search for a matching pnfsd mirror device structure, base on the nmp arg.
8479  * Return one if found, NULL otherwise.
8480  */
8481 static struct nfsdevice *
8482 nfsrv_findmirroredds(struct nfsmount *nmp)
8483 {
8484 	struct nfsdevice *ds, *fndds;
8485 	int fndmirror;
8486 
8487 	mtx_assert(NFSDDSMUTEXPTR, MA_OWNED);
8488 	/*
8489 	 * Search the DS server list for a match with nmp.
8490 	 * Remove the DS entry if found and there is a mirror.
8491 	 */
8492 	fndds = NULL;
8493 	fndmirror = 0;
8494 	if (nfsrv_devidcnt == 0)
8495 		return (fndds);
8496 	TAILQ_FOREACH(ds, &nfsrv_devidhead, nfsdev_list) {
8497 		if (ds->nfsdev_nmp == nmp) {
8498 			NFSD_DEBUG(4, "nfsrv_findmirroredds: fnd main ds\n");
8499 			fndds = ds;
8500 			break;
8501 		}
8502 	}
8503 	if (fndds == NULL)
8504 		return (fndds);
8505 	if (fndds->nfsdev_mdsisset == 0 && nfsrv_faildscnt > 0)
8506 		fndmirror = 1;
8507 	else if (fndds->nfsdev_mdsisset != 0) {
8508 		/* For the fsid is set case, search for a mirror. */
8509 		TAILQ_FOREACH(ds, &nfsrv_devidhead, nfsdev_list) {
8510 			if (ds != fndds && ds->nfsdev_nmp != NULL &&
8511 			    ds->nfsdev_mdsisset != 0 &&
8512 			    ds->nfsdev_mdsfsid.val[0] ==
8513 			    fndds->nfsdev_mdsfsid.val[0] &&
8514 			    ds->nfsdev_mdsfsid.val[1] ==
8515 			    fndds->nfsdev_mdsfsid.val[1]) {
8516 				fndmirror = 1;
8517 				break;
8518 			}
8519 		}
8520 	}
8521 	if (fndmirror == 0) {
8522 		NFSD_DEBUG(4, "nfsrv_findmirroredds: no mirror for DS\n");
8523 		return (NULL);
8524 	}
8525 	return (fndds);
8526 }
8527 
8528