xref: /freebsd/sys/fs/nfsserver/nfs_nfsdstate.c (revision 8bcb0991864975618c09697b1aca10683346d9f0)
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 #include "opt_inet.h"
34 #include "opt_inet6.h"
35 #ifndef APPLEKEXT
36 #include <sys/extattr.h>
37 #include <fs/nfs/nfsport.h>
38 
39 struct nfsrv_stablefirst nfsrv_stablefirst;
40 int nfsrv_issuedelegs = 0;
41 int nfsrv_dolocallocks = 0;
42 struct nfsv4lock nfsv4rootfs_lock;
43 time_t nfsdev_time = 0;
44 int nfsrv_layouthashsize;
45 volatile int nfsrv_layoutcnt = 0;
46 
47 extern int newnfs_numnfsd;
48 extern struct nfsstatsv1 nfsstatsv1;
49 extern int nfsrv_lease;
50 extern struct timeval nfsboottime;
51 extern u_int32_t newnfs_true, newnfs_false;
52 extern struct mtx nfsrv_dslock_mtx;
53 extern struct mtx nfsrv_recalllock_mtx;
54 extern struct mtx nfsrv_dontlistlock_mtx;
55 extern int nfsd_debuglevel;
56 extern u_int nfsrv_dsdirsize;
57 extern struct nfsdevicehead nfsrv_devidhead;
58 extern int nfsrv_doflexfile;
59 extern int nfsrv_maxpnfsmirror;
60 NFSV4ROOTLOCKMUTEX;
61 NFSSTATESPINLOCK;
62 extern struct nfsdontlisthead nfsrv_dontlisthead;
63 extern volatile int nfsrv_devidcnt;
64 extern struct nfslayouthead nfsrv_recalllisthead;
65 extern char *nfsrv_zeropnfsdat;
66 
67 SYSCTL_DECL(_vfs_nfsd);
68 int	nfsrv_statehashsize = NFSSTATEHASHSIZE;
69 SYSCTL_INT(_vfs_nfsd, OID_AUTO, statehashsize, CTLFLAG_RDTUN,
70     &nfsrv_statehashsize, 0,
71     "Size of state hash table set via loader.conf");
72 
73 int	nfsrv_clienthashsize = NFSCLIENTHASHSIZE;
74 SYSCTL_INT(_vfs_nfsd, OID_AUTO, clienthashsize, CTLFLAG_RDTUN,
75     &nfsrv_clienthashsize, 0,
76     "Size of client hash table set via loader.conf");
77 
78 int	nfsrv_lockhashsize = NFSLOCKHASHSIZE;
79 SYSCTL_INT(_vfs_nfsd, OID_AUTO, fhhashsize, CTLFLAG_RDTUN,
80     &nfsrv_lockhashsize, 0,
81     "Size of file handle hash table set via loader.conf");
82 
83 int	nfsrv_sessionhashsize = NFSSESSIONHASHSIZE;
84 SYSCTL_INT(_vfs_nfsd, OID_AUTO, sessionhashsize, CTLFLAG_RDTUN,
85     &nfsrv_sessionhashsize, 0,
86     "Size of session hash table set via loader.conf");
87 
88 int	nfsrv_layouthighwater = NFSLAYOUTHIGHWATER;
89 SYSCTL_INT(_vfs_nfsd, OID_AUTO, layouthighwater, CTLFLAG_RDTUN,
90     &nfsrv_layouthighwater, 0,
91     "High water mark for number of layouts set via loader.conf");
92 
93 static int	nfsrv_v4statelimit = NFSRV_V4STATELIMIT;
94 SYSCTL_INT(_vfs_nfsd, OID_AUTO, v4statelimit, CTLFLAG_RWTUN,
95     &nfsrv_v4statelimit, 0,
96     "High water limit for NFSv4 opens+locks+delegations");
97 
98 static int	nfsrv_writedelegifpos = 0;
99 SYSCTL_INT(_vfs_nfsd, OID_AUTO, writedelegifpos, CTLFLAG_RW,
100     &nfsrv_writedelegifpos, 0,
101     "Issue a write delegation for read opens if possible");
102 
103 static int	nfsrv_allowreadforwriteopen = 1;
104 SYSCTL_INT(_vfs_nfsd, OID_AUTO, allowreadforwriteopen, CTLFLAG_RW,
105     &nfsrv_allowreadforwriteopen, 0,
106     "Allow Reads to be done with Write Access StateIDs");
107 
108 int	nfsrv_pnfsatime = 0;
109 SYSCTL_INT(_vfs_nfsd, OID_AUTO, pnfsstrictatime, CTLFLAG_RW,
110     &nfsrv_pnfsatime, 0,
111     "For pNFS service, do Getattr ops to keep atime up-to-date");
112 
113 int	nfsrv_flexlinuxhack = 0;
114 SYSCTL_INT(_vfs_nfsd, OID_AUTO, flexlinuxhack, CTLFLAG_RW,
115     &nfsrv_flexlinuxhack, 0,
116     "For Linux clients, hack around Flex File Layout bug");
117 
118 /*
119  * Hash lists for nfs V4.
120  */
121 struct nfsclienthashhead	*nfsclienthash;
122 struct nfslockhashhead		*nfslockhash;
123 struct nfssessionhash		*nfssessionhash;
124 struct nfslayouthash		*nfslayouthash;
125 volatile int nfsrv_dontlistlen = 0;
126 #endif	/* !APPLEKEXT */
127 
128 static u_int32_t nfsrv_openpluslock = 0, nfsrv_delegatecnt = 0;
129 static time_t nfsrvboottime;
130 static int nfsrv_returnoldstateid = 0, nfsrv_clients = 0;
131 static int nfsrv_clienthighwater = NFSRV_CLIENTHIGHWATER;
132 static int nfsrv_nogsscallback = 0;
133 static volatile int nfsrv_writedelegcnt = 0;
134 static int nfsrv_faildscnt;
135 
136 /* local functions */
137 static void nfsrv_dumpaclient(struct nfsclient *clp,
138     struct nfsd_dumpclients *dumpp);
139 static void nfsrv_freeopenowner(struct nfsstate *stp, int cansleep,
140     NFSPROC_T *p);
141 static int nfsrv_freeopen(struct nfsstate *stp, vnode_t vp, int cansleep,
142     NFSPROC_T *p);
143 static void nfsrv_freelockowner(struct nfsstate *stp, vnode_t vp, int cansleep,
144     NFSPROC_T *p);
145 static void nfsrv_freeallnfslocks(struct nfsstate *stp, vnode_t vp,
146     int cansleep, NFSPROC_T *p);
147 static void nfsrv_freenfslock(struct nfslock *lop);
148 static void nfsrv_freenfslockfile(struct nfslockfile *lfp);
149 static void nfsrv_freedeleg(struct nfsstate *);
150 static int nfsrv_getstate(struct nfsclient *clp, nfsv4stateid_t *stateidp,
151     u_int32_t flags, struct nfsstate **stpp);
152 static void nfsrv_getowner(struct nfsstatehead *hp, struct nfsstate *new_stp,
153     struct nfsstate **stpp);
154 static int nfsrv_getlockfh(vnode_t vp, u_short flags,
155     struct nfslockfile *new_lfp, fhandle_t *nfhp, NFSPROC_T *p);
156 static int nfsrv_getlockfile(u_short flags, struct nfslockfile **new_lfpp,
157     struct nfslockfile **lfpp, fhandle_t *nfhp, int lockit);
158 static void nfsrv_insertlock(struct nfslock *new_lop,
159     struct nfslock *insert_lop, struct nfsstate *stp, struct nfslockfile *lfp);
160 static void nfsrv_updatelock(struct nfsstate *stp, struct nfslock **new_lopp,
161     struct nfslock **other_lopp, struct nfslockfile *lfp);
162 static int nfsrv_getipnumber(u_char *cp);
163 static int nfsrv_checkrestart(nfsquad_t clientid, u_int32_t flags,
164     nfsv4stateid_t *stateidp, int specialid);
165 static int nfsrv_checkgrace(struct nfsrv_descript *nd, struct nfsclient *clp,
166     u_int32_t flags);
167 static int nfsrv_docallback(struct nfsclient *clp, int procnum,
168     nfsv4stateid_t *stateidp, int trunc, fhandle_t *fhp,
169     struct nfsvattr *nap, nfsattrbit_t *attrbitp, int laytype, NFSPROC_T *p);
170 static int nfsrv_cbcallargs(struct nfsrv_descript *nd, struct nfsclient *clp,
171     uint32_t callback, int op, const char *optag, struct nfsdsession **sepp);
172 static u_int32_t nfsrv_nextclientindex(void);
173 static u_int32_t nfsrv_nextstateindex(struct nfsclient *clp);
174 static void nfsrv_markstable(struct nfsclient *clp);
175 static void nfsrv_markreclaim(struct nfsclient *clp);
176 static int nfsrv_checkstable(struct nfsclient *clp);
177 static int nfsrv_clientconflict(struct nfsclient *clp, int *haslockp, struct
178     vnode *vp, NFSPROC_T *p);
179 static int nfsrv_delegconflict(struct nfsstate *stp, int *haslockp,
180     NFSPROC_T *p, vnode_t vp);
181 static int nfsrv_cleandeleg(vnode_t vp, struct nfslockfile *lfp,
182     struct nfsclient *clp, int *haslockp, NFSPROC_T *p);
183 static int nfsrv_notsamecredname(struct nfsrv_descript *nd,
184     struct nfsclient *clp);
185 static time_t nfsrv_leaseexpiry(void);
186 static void nfsrv_delaydelegtimeout(struct nfsstate *stp);
187 static int nfsrv_checkseqid(struct nfsrv_descript *nd, u_int32_t seqid,
188     struct nfsstate *stp, struct nfsrvcache *op);
189 static int nfsrv_nootherstate(struct nfsstate *stp);
190 static int nfsrv_locallock(vnode_t vp, struct nfslockfile *lfp, int flags,
191     uint64_t first, uint64_t end, struct nfslockconflict *cfp, NFSPROC_T *p);
192 static void nfsrv_localunlock(vnode_t vp, struct nfslockfile *lfp,
193     uint64_t init_first, uint64_t init_end, NFSPROC_T *p);
194 static int nfsrv_dolocal(vnode_t vp, struct nfslockfile *lfp, int flags,
195     int oldflags, uint64_t first, uint64_t end, struct nfslockconflict *cfp,
196     NFSPROC_T *p);
197 static void nfsrv_locallock_rollback(vnode_t vp, struct nfslockfile *lfp,
198     NFSPROC_T *p);
199 static void nfsrv_locallock_commit(struct nfslockfile *lfp, int flags,
200     uint64_t first, uint64_t end);
201 static void nfsrv_locklf(struct nfslockfile *lfp);
202 static void nfsrv_unlocklf(struct nfslockfile *lfp);
203 static struct nfsdsession *nfsrv_findsession(uint8_t *sessionid);
204 static int nfsrv_freesession(struct nfsdsession *sep, uint8_t *sessionid);
205 static int nfsv4_setcbsequence(struct nfsrv_descript *nd, struct nfsclient *clp,
206     int dont_replycache, struct nfsdsession **sepp);
207 static int nfsv4_getcbsession(struct nfsclient *clp, struct nfsdsession **sepp);
208 static int nfsrv_addlayout(struct nfsrv_descript *nd, struct nfslayout **lypp,
209     nfsv4stateid_t *stateidp, char *layp, int *layoutlenp, NFSPROC_T *p);
210 static void nfsrv_freelayout(struct nfslayouthead *lhp, struct nfslayout *lyp);
211 static void nfsrv_freelayoutlist(nfsquad_t clientid);
212 static void nfsrv_freelayouts(nfsquad_t *clid, fsid_t *fs, int laytype,
213     int iomode);
214 static void nfsrv_freealllayouts(void);
215 static void nfsrv_freedevid(struct nfsdevice *ds);
216 static int nfsrv_setdsserver(char *dspathp, char *mdspathp, NFSPROC_T *p,
217     struct nfsdevice **dsp);
218 static void nfsrv_deleteds(struct nfsdevice *fndds);
219 static void nfsrv_allocdevid(struct nfsdevice *ds, char *addr, char *dnshost);
220 static void nfsrv_freealldevids(void);
221 static void nfsrv_flexlayouterr(struct nfsrv_descript *nd, uint32_t *layp,
222     int maxcnt, NFSPROC_T *p);
223 static int nfsrv_recalllayout(nfsquad_t clid, nfsv4stateid_t *stateidp,
224     fhandle_t *fhp, struct nfslayout *lyp, int changed, int laytype,
225     NFSPROC_T *p);
226 static int nfsrv_findlayout(nfsquad_t *clientidp, fhandle_t *fhp, int laytype,
227     NFSPROC_T *, struct nfslayout **lypp);
228 static int nfsrv_fndclid(nfsquad_t *clidvec, nfsquad_t clid, int clidcnt);
229 static struct nfslayout *nfsrv_filelayout(struct nfsrv_descript *nd, int iomode,
230     fhandle_t *fhp, fhandle_t *dsfhp, char *devid, fsid_t fs);
231 static struct nfslayout *nfsrv_flexlayout(struct nfsrv_descript *nd, int iomode,
232     int mirrorcnt, fhandle_t *fhp, fhandle_t *dsfhp, char *devid, fsid_t fs);
233 static int nfsrv_dontlayout(fhandle_t *fhp);
234 static int nfsrv_createdsfile(vnode_t vp, fhandle_t *fhp, struct pnfsdsfile *pf,
235     vnode_t dvp, struct nfsdevice *ds, struct ucred *cred, NFSPROC_T *p,
236     vnode_t *tvpp);
237 static struct nfsdevice *nfsrv_findmirroredds(struct nfsmount *nmp);
238 
239 /*
240  * Scan the client list for a match and either return the current one,
241  * create a new entry or return an error.
242  * If returning a non-error, the clp structure must either be linked into
243  * the client list or free'd.
244  */
245 APPLESTATIC int
246 nfsrv_setclient(struct nfsrv_descript *nd, struct nfsclient **new_clpp,
247     nfsquad_t *clientidp, nfsquad_t *confirmp, NFSPROC_T *p)
248 {
249 	struct nfsclient *clp = NULL, *new_clp = *new_clpp;
250 	int i, error = 0, ret;
251 	struct nfsstate *stp, *tstp;
252 #ifdef INET
253 	struct sockaddr_in *sin, *rin;
254 #endif
255 #ifdef INET6
256 	struct sockaddr_in6 *sin6, *rin6;
257 #endif
258 	struct nfsdsession *sep, *nsep;
259 	int zapit = 0, gotit, hasstate = 0, igotlock;
260 	static u_int64_t confirm_index = 0;
261 
262 	/*
263 	 * Check for state resource limit exceeded.
264 	 */
265 	if (nfsrv_openpluslock > nfsrv_v4statelimit) {
266 		error = NFSERR_RESOURCE;
267 		goto out;
268 	}
269 
270 	if (nfsrv_issuedelegs == 0 ||
271 	    ((nd->nd_flag & ND_GSS) != 0 && nfsrv_nogsscallback != 0))
272 		/*
273 		 * Don't do callbacks when delegations are disabled or
274 		 * for AUTH_GSS unless enabled via nfsrv_nogsscallback.
275 		 * If establishing a callback connection is attempted
276 		 * when a firewall is blocking the callback path, the
277 		 * server may wait too long for the connect attempt to
278 		 * succeed during the Open. Some clients, such as Linux,
279 		 * may timeout and give up on the Open before the server
280 		 * replies. Also, since AUTH_GSS callbacks are not
281 		 * yet interoperability tested, they might cause the
282 		 * server to crap out, if they get past the Init call to
283 		 * the client.
284 		 */
285 		new_clp->lc_program = 0;
286 
287 	/* Lock out other nfsd threads */
288 	NFSLOCKV4ROOTMUTEX();
289 	nfsv4_relref(&nfsv4rootfs_lock);
290 	do {
291 		igotlock = nfsv4_lock(&nfsv4rootfs_lock, 1, NULL,
292 		    NFSV4ROOTLOCKMUTEXPTR, NULL);
293 	} while (!igotlock);
294 	NFSUNLOCKV4ROOTMUTEX();
295 
296 	/*
297 	 * Search for a match in the client list.
298 	 */
299 	gotit = i = 0;
300 	while (i < nfsrv_clienthashsize && !gotit) {
301 	    LIST_FOREACH(clp, &nfsclienthash[i], lc_hash) {
302 		if (new_clp->lc_idlen == clp->lc_idlen &&
303 		    !NFSBCMP(new_clp->lc_id, clp->lc_id, clp->lc_idlen)) {
304 			gotit = 1;
305 			break;
306 		}
307 	    }
308 	    if (gotit == 0)
309 		i++;
310 	}
311 	if (!gotit ||
312 	    (clp->lc_flags & (LCL_NEEDSCONFIRM | LCL_ADMINREVOKED))) {
313 		if ((nd->nd_flag & ND_NFSV41) != 0 && confirmp->lval[1] != 0) {
314 			/*
315 			 * For NFSv4.1, if confirmp->lval[1] is non-zero, the
316 			 * client is trying to update a confirmed clientid.
317 			 */
318 			NFSLOCKV4ROOTMUTEX();
319 			nfsv4_unlock(&nfsv4rootfs_lock, 1);
320 			NFSUNLOCKV4ROOTMUTEX();
321 			confirmp->lval[1] = 0;
322 			error = NFSERR_NOENT;
323 			goto out;
324 		}
325 		/*
326 		 * Get rid of the old one.
327 		 */
328 		if (i != nfsrv_clienthashsize) {
329 			LIST_REMOVE(clp, lc_hash);
330 			nfsrv_cleanclient(clp, p);
331 			nfsrv_freedeleglist(&clp->lc_deleg);
332 			nfsrv_freedeleglist(&clp->lc_olddeleg);
333 			zapit = 1;
334 		}
335 		/*
336 		 * Add it after assigning a client id to it.
337 		 */
338 		new_clp->lc_flags |= LCL_NEEDSCONFIRM;
339 		if ((nd->nd_flag & ND_NFSV41) != 0)
340 			new_clp->lc_confirm.lval[0] = confirmp->lval[0] =
341 			    ++confirm_index;
342 		else
343 			confirmp->qval = new_clp->lc_confirm.qval =
344 			    ++confirm_index;
345 		clientidp->lval[0] = new_clp->lc_clientid.lval[0] =
346 		    (u_int32_t)nfsrvboottime;
347 		clientidp->lval[1] = new_clp->lc_clientid.lval[1] =
348 		    nfsrv_nextclientindex();
349 		new_clp->lc_stateindex = 0;
350 		new_clp->lc_statemaxindex = 0;
351 		new_clp->lc_cbref = 0;
352 		new_clp->lc_expiry = nfsrv_leaseexpiry();
353 		LIST_INIT(&new_clp->lc_open);
354 		LIST_INIT(&new_clp->lc_deleg);
355 		LIST_INIT(&new_clp->lc_olddeleg);
356 		LIST_INIT(&new_clp->lc_session);
357 		for (i = 0; i < nfsrv_statehashsize; i++)
358 			LIST_INIT(&new_clp->lc_stateid[i]);
359 		LIST_INSERT_HEAD(NFSCLIENTHASH(new_clp->lc_clientid), new_clp,
360 		    lc_hash);
361 		nfsstatsv1.srvclients++;
362 		nfsrv_openpluslock++;
363 		nfsrv_clients++;
364 		NFSLOCKV4ROOTMUTEX();
365 		nfsv4_unlock(&nfsv4rootfs_lock, 1);
366 		NFSUNLOCKV4ROOTMUTEX();
367 		if (zapit)
368 			nfsrv_zapclient(clp, p);
369 		*new_clpp = NULL;
370 		goto out;
371 	}
372 
373 	/*
374 	 * Now, handle the cases where the id is already issued.
375 	 */
376 	if (nfsrv_notsamecredname(nd, clp)) {
377 	    /*
378 	     * Check to see if there is expired state that should go away.
379 	     */
380 	    if (clp->lc_expiry < NFSD_MONOSEC &&
381 	        (!LIST_EMPTY(&clp->lc_open) || !LIST_EMPTY(&clp->lc_deleg))) {
382 		nfsrv_cleanclient(clp, p);
383 		nfsrv_freedeleglist(&clp->lc_deleg);
384 	    }
385 
386 	    /*
387 	     * If there is outstanding state, then reply NFSERR_CLIDINUSE per
388 	     * RFC3530 Sec. 8.1.2 last para.
389 	     */
390 	    if (!LIST_EMPTY(&clp->lc_deleg)) {
391 		hasstate = 1;
392 	    } else if (LIST_EMPTY(&clp->lc_open)) {
393 		hasstate = 0;
394 	    } else {
395 		hasstate = 0;
396 		/* Look for an Open on the OpenOwner */
397 		LIST_FOREACH(stp, &clp->lc_open, ls_list) {
398 		    if (!LIST_EMPTY(&stp->ls_open)) {
399 			hasstate = 1;
400 			break;
401 		    }
402 		}
403 	    }
404 	    if (hasstate) {
405 		/*
406 		 * If the uid doesn't match, return NFSERR_CLIDINUSE after
407 		 * filling out the correct ipaddr and portnum.
408 		 */
409 		switch (clp->lc_req.nr_nam->sa_family) {
410 #ifdef INET
411 		case AF_INET:
412 			sin = (struct sockaddr_in *)new_clp->lc_req.nr_nam;
413 			rin = (struct sockaddr_in *)clp->lc_req.nr_nam;
414 			sin->sin_addr.s_addr = rin->sin_addr.s_addr;
415 			sin->sin_port = rin->sin_port;
416 			break;
417 #endif
418 #ifdef INET6
419 		case AF_INET6:
420 			sin6 = (struct sockaddr_in6 *)new_clp->lc_req.nr_nam;
421 			rin6 = (struct sockaddr_in6 *)clp->lc_req.nr_nam;
422 			sin6->sin6_addr = rin6->sin6_addr;
423 			sin6->sin6_port = rin6->sin6_port;
424 			break;
425 #endif
426 		}
427 		NFSLOCKV4ROOTMUTEX();
428 		nfsv4_unlock(&nfsv4rootfs_lock, 1);
429 		NFSUNLOCKV4ROOTMUTEX();
430 		error = NFSERR_CLIDINUSE;
431 		goto out;
432 	    }
433 	}
434 
435 	if (NFSBCMP(new_clp->lc_verf, clp->lc_verf, NFSX_VERF)) {
436 		/*
437 		 * If the verifier has changed, the client has rebooted
438 		 * and a new client id is issued. The old state info
439 		 * can be thrown away once the SETCLIENTID_CONFIRM occurs.
440 		 */
441 		LIST_REMOVE(clp, lc_hash);
442 
443 		/* Get rid of all sessions on this clientid. */
444 		LIST_FOREACH_SAFE(sep, &clp->lc_session, sess_list, nsep) {
445 			ret = nfsrv_freesession(sep, NULL);
446 			if (ret != 0)
447 				printf("nfsrv_setclient: verifier changed free"
448 				    " session failed=%d\n", ret);
449 		}
450 
451 		new_clp->lc_flags |= LCL_NEEDSCONFIRM;
452 		if ((nd->nd_flag & ND_NFSV41) != 0)
453 			new_clp->lc_confirm.lval[0] = confirmp->lval[0] =
454 			    ++confirm_index;
455 		else
456 			confirmp->qval = new_clp->lc_confirm.qval =
457 			    ++confirm_index;
458 		clientidp->lval[0] = new_clp->lc_clientid.lval[0] =
459 		    nfsrvboottime;
460 		clientidp->lval[1] = new_clp->lc_clientid.lval[1] =
461 		    nfsrv_nextclientindex();
462 		new_clp->lc_stateindex = 0;
463 		new_clp->lc_statemaxindex = 0;
464 		new_clp->lc_cbref = 0;
465 		new_clp->lc_expiry = nfsrv_leaseexpiry();
466 
467 		/*
468 		 * Save the state until confirmed.
469 		 */
470 		LIST_NEWHEAD(&new_clp->lc_open, &clp->lc_open, ls_list);
471 		LIST_FOREACH(tstp, &new_clp->lc_open, ls_list)
472 			tstp->ls_clp = new_clp;
473 		LIST_NEWHEAD(&new_clp->lc_deleg, &clp->lc_deleg, ls_list);
474 		LIST_FOREACH(tstp, &new_clp->lc_deleg, ls_list)
475 			tstp->ls_clp = new_clp;
476 		LIST_NEWHEAD(&new_clp->lc_olddeleg, &clp->lc_olddeleg,
477 		    ls_list);
478 		LIST_FOREACH(tstp, &new_clp->lc_olddeleg, ls_list)
479 			tstp->ls_clp = new_clp;
480 		for (i = 0; i < nfsrv_statehashsize; i++) {
481 			LIST_NEWHEAD(&new_clp->lc_stateid[i],
482 			    &clp->lc_stateid[i], ls_hash);
483 			LIST_FOREACH(tstp, &new_clp->lc_stateid[i], ls_hash)
484 				tstp->ls_clp = new_clp;
485 		}
486 		LIST_INIT(&new_clp->lc_session);
487 		LIST_INSERT_HEAD(NFSCLIENTHASH(new_clp->lc_clientid), new_clp,
488 		    lc_hash);
489 		nfsstatsv1.srvclients++;
490 		nfsrv_openpluslock++;
491 		nfsrv_clients++;
492 		NFSLOCKV4ROOTMUTEX();
493 		nfsv4_unlock(&nfsv4rootfs_lock, 1);
494 		NFSUNLOCKV4ROOTMUTEX();
495 
496 		/*
497 		 * Must wait until any outstanding callback on the old clp
498 		 * completes.
499 		 */
500 		NFSLOCKSTATE();
501 		while (clp->lc_cbref) {
502 			clp->lc_flags |= LCL_WAKEUPWANTED;
503 			(void)mtx_sleep(clp, NFSSTATEMUTEXPTR, PZERO - 1,
504 			    "nfsd clp", 10 * hz);
505 		}
506 		NFSUNLOCKSTATE();
507 		nfsrv_zapclient(clp, p);
508 		*new_clpp = NULL;
509 		goto out;
510 	}
511 
512 	/* For NFSv4.1, mark that we found a confirmed clientid. */
513 	if ((nd->nd_flag & ND_NFSV41) != 0) {
514 		clientidp->lval[0] = clp->lc_clientid.lval[0];
515 		clientidp->lval[1] = clp->lc_clientid.lval[1];
516 		confirmp->lval[0] = 0;	/* Ignored by client */
517 		confirmp->lval[1] = 1;
518 	} else {
519 		/*
520 		 * id and verifier match, so update the net address info
521 		 * and get rid of any existing callback authentication
522 		 * handle, so a new one will be acquired.
523 		 */
524 		LIST_REMOVE(clp, lc_hash);
525 		new_clp->lc_flags |= (LCL_NEEDSCONFIRM | LCL_DONTCLEAN);
526 		new_clp->lc_expiry = nfsrv_leaseexpiry();
527 		confirmp->qval = new_clp->lc_confirm.qval = ++confirm_index;
528 		clientidp->lval[0] = new_clp->lc_clientid.lval[0] =
529 		    clp->lc_clientid.lval[0];
530 		clientidp->lval[1] = new_clp->lc_clientid.lval[1] =
531 		    clp->lc_clientid.lval[1];
532 		new_clp->lc_delegtime = clp->lc_delegtime;
533 		new_clp->lc_stateindex = clp->lc_stateindex;
534 		new_clp->lc_statemaxindex = clp->lc_statemaxindex;
535 		new_clp->lc_cbref = 0;
536 		LIST_NEWHEAD(&new_clp->lc_open, &clp->lc_open, ls_list);
537 		LIST_FOREACH(tstp, &new_clp->lc_open, ls_list)
538 			tstp->ls_clp = new_clp;
539 		LIST_NEWHEAD(&new_clp->lc_deleg, &clp->lc_deleg, ls_list);
540 		LIST_FOREACH(tstp, &new_clp->lc_deleg, ls_list)
541 			tstp->ls_clp = new_clp;
542 		LIST_NEWHEAD(&new_clp->lc_olddeleg, &clp->lc_olddeleg, ls_list);
543 		LIST_FOREACH(tstp, &new_clp->lc_olddeleg, ls_list)
544 			tstp->ls_clp = new_clp;
545 		for (i = 0; i < nfsrv_statehashsize; i++) {
546 			LIST_NEWHEAD(&new_clp->lc_stateid[i],
547 			    &clp->lc_stateid[i], ls_hash);
548 			LIST_FOREACH(tstp, &new_clp->lc_stateid[i], ls_hash)
549 				tstp->ls_clp = new_clp;
550 		}
551 		LIST_INIT(&new_clp->lc_session);
552 		LIST_INSERT_HEAD(NFSCLIENTHASH(new_clp->lc_clientid), new_clp,
553 		    lc_hash);
554 		nfsstatsv1.srvclients++;
555 		nfsrv_openpluslock++;
556 		nfsrv_clients++;
557 	}
558 	NFSLOCKV4ROOTMUTEX();
559 	nfsv4_unlock(&nfsv4rootfs_lock, 1);
560 	NFSUNLOCKV4ROOTMUTEX();
561 
562 	if ((nd->nd_flag & ND_NFSV41) == 0) {
563 		/*
564 		 * Must wait until any outstanding callback on the old clp
565 		 * completes.
566 		 */
567 		NFSLOCKSTATE();
568 		while (clp->lc_cbref) {
569 			clp->lc_flags |= LCL_WAKEUPWANTED;
570 			(void)mtx_sleep(clp, NFSSTATEMUTEXPTR, PZERO - 1,
571 			    "nfsdclp", 10 * hz);
572 		}
573 		NFSUNLOCKSTATE();
574 		nfsrv_zapclient(clp, p);
575 		*new_clpp = NULL;
576 	}
577 
578 out:
579 	NFSEXITCODE2(error, nd);
580 	return (error);
581 }
582 
583 /*
584  * Check to see if the client id exists and optionally confirm it.
585  */
586 APPLESTATIC int
587 nfsrv_getclient(nfsquad_t clientid, int opflags, struct nfsclient **clpp,
588     struct nfsdsession *nsep, nfsquad_t confirm, uint32_t cbprogram,
589     struct nfsrv_descript *nd, NFSPROC_T *p)
590 {
591 	struct nfsclient *clp;
592 	struct nfsstate *stp;
593 	int i;
594 	struct nfsclienthashhead *hp;
595 	int error = 0, igotlock, doneok;
596 	struct nfssessionhash *shp;
597 	struct nfsdsession *sep;
598 	uint64_t sessid[2];
599 	static uint64_t next_sess = 0;
600 
601 	if (clpp)
602 		*clpp = NULL;
603 	if ((nd == NULL || (nd->nd_flag & ND_NFSV41) == 0 ||
604 	    opflags != CLOPS_RENEW) && nfsrvboottime != clientid.lval[0]) {
605 		error = NFSERR_STALECLIENTID;
606 		goto out;
607 	}
608 
609 	/*
610 	 * If called with opflags == CLOPS_RENEW, the State Lock is
611 	 * already held. Otherwise, we need to get either that or,
612 	 * for the case of Confirm, lock out the nfsd threads.
613 	 */
614 	if (opflags & CLOPS_CONFIRM) {
615 		NFSLOCKV4ROOTMUTEX();
616 		nfsv4_relref(&nfsv4rootfs_lock);
617 		do {
618 			igotlock = nfsv4_lock(&nfsv4rootfs_lock, 1, NULL,
619 			    NFSV4ROOTLOCKMUTEXPTR, NULL);
620 		} while (!igotlock);
621 		/*
622 		 * Create a new sessionid here, since we need to do it where
623 		 * there is a mutex held to serialize update of next_sess.
624 		 */
625 		if ((nd->nd_flag & ND_NFSV41) != 0) {
626 			sessid[0] = ++next_sess;
627 			sessid[1] = clientid.qval;
628 		}
629 		NFSUNLOCKV4ROOTMUTEX();
630 	} else if (opflags != CLOPS_RENEW) {
631 		NFSLOCKSTATE();
632 	}
633 
634 	/* For NFSv4.1, the clp is acquired from the associated session. */
635 	if (nd != NULL && (nd->nd_flag & ND_NFSV41) != 0 &&
636 	    opflags == CLOPS_RENEW) {
637 		clp = NULL;
638 		if ((nd->nd_flag & ND_HASSEQUENCE) != 0) {
639 			shp = NFSSESSIONHASH(nd->nd_sessionid);
640 			NFSLOCKSESSION(shp);
641 			sep = nfsrv_findsession(nd->nd_sessionid);
642 			if (sep != NULL)
643 				clp = sep->sess_clp;
644 			NFSUNLOCKSESSION(shp);
645 		}
646 	} else {
647 		hp = NFSCLIENTHASH(clientid);
648 		LIST_FOREACH(clp, hp, lc_hash) {
649 			if (clp->lc_clientid.lval[1] == clientid.lval[1])
650 				break;
651 		}
652 	}
653 	if (clp == NULL) {
654 		if (opflags & CLOPS_CONFIRM)
655 			error = NFSERR_STALECLIENTID;
656 		else
657 			error = NFSERR_EXPIRED;
658 	} else if (clp->lc_flags & LCL_ADMINREVOKED) {
659 		/*
660 		 * If marked admin revoked, just return the error.
661 		 */
662 		error = NFSERR_ADMINREVOKED;
663 	}
664 	if (error) {
665 		if (opflags & CLOPS_CONFIRM) {
666 			NFSLOCKV4ROOTMUTEX();
667 			nfsv4_unlock(&nfsv4rootfs_lock, 1);
668 			NFSUNLOCKV4ROOTMUTEX();
669 		} else if (opflags != CLOPS_RENEW) {
670 			NFSUNLOCKSTATE();
671 		}
672 		goto out;
673 	}
674 
675 	/*
676 	 * Perform any operations specified by the opflags.
677 	 */
678 	if (opflags & CLOPS_CONFIRM) {
679 		if (((nd->nd_flag & ND_NFSV41) != 0 &&
680 		     clp->lc_confirm.lval[0] != confirm.lval[0]) ||
681 		    ((nd->nd_flag & ND_NFSV41) == 0 &&
682 		     clp->lc_confirm.qval != confirm.qval))
683 			error = NFSERR_STALECLIENTID;
684 		else if (nfsrv_notsamecredname(nd, clp))
685 			error = NFSERR_CLIDINUSE;
686 
687 		if (!error) {
688 		    if ((clp->lc_flags & (LCL_NEEDSCONFIRM | LCL_DONTCLEAN)) ==
689 			LCL_NEEDSCONFIRM) {
690 			/*
691 			 * Hang onto the delegations (as old delegations)
692 			 * for an Open with CLAIM_DELEGATE_PREV unless in
693 			 * grace, but get rid of the rest of the state.
694 			 */
695 			nfsrv_cleanclient(clp, p);
696 			nfsrv_freedeleglist(&clp->lc_olddeleg);
697 			if (nfsrv_checkgrace(nd, clp, 0)) {
698 			    /* In grace, so just delete delegations */
699 			    nfsrv_freedeleglist(&clp->lc_deleg);
700 			} else {
701 			    LIST_FOREACH(stp, &clp->lc_deleg, ls_list)
702 				stp->ls_flags |= NFSLCK_OLDDELEG;
703 			    clp->lc_delegtime = NFSD_MONOSEC +
704 				nfsrv_lease + NFSRV_LEASEDELTA;
705 			    LIST_NEWHEAD(&clp->lc_olddeleg, &clp->lc_deleg,
706 				ls_list);
707 			}
708 			if ((nd->nd_flag & ND_NFSV41) != 0)
709 			    clp->lc_program = cbprogram;
710 		    }
711 		    clp->lc_flags &= ~(LCL_NEEDSCONFIRM | LCL_DONTCLEAN);
712 		    if (clp->lc_program)
713 			clp->lc_flags |= LCL_NEEDSCBNULL;
714 		    /* For NFSv4.1, link the session onto the client. */
715 		    if (nsep != NULL) {
716 			/* Hold a reference on the xprt for a backchannel. */
717 			if ((nsep->sess_crflags & NFSV4CRSESS_CONNBACKCHAN)
718 			    != 0) {
719 			    if (clp->lc_req.nr_client == NULL)
720 				clp->lc_req.nr_client = (struct __rpc_client *)
721 				    clnt_bck_create(nd->nd_xprt->xp_socket,
722 				    cbprogram, NFSV4_CBVERS);
723 			    if (clp->lc_req.nr_client != NULL) {
724 				SVC_ACQUIRE(nd->nd_xprt);
725 				nd->nd_xprt->xp_p2 =
726 				    clp->lc_req.nr_client->cl_private;
727 				/* Disable idle timeout. */
728 				nd->nd_xprt->xp_idletimeout = 0;
729 				nsep->sess_cbsess.nfsess_xprt = nd->nd_xprt;
730 			    } else
731 				nsep->sess_crflags &= ~NFSV4CRSESS_CONNBACKCHAN;
732 			}
733 			NFSBCOPY(sessid, nsep->sess_sessionid,
734 			    NFSX_V4SESSIONID);
735 			NFSBCOPY(sessid, nsep->sess_cbsess.nfsess_sessionid,
736 			    NFSX_V4SESSIONID);
737 			shp = NFSSESSIONHASH(nsep->sess_sessionid);
738 			NFSLOCKSTATE();
739 			NFSLOCKSESSION(shp);
740 			LIST_INSERT_HEAD(&shp->list, nsep, sess_hash);
741 			LIST_INSERT_HEAD(&clp->lc_session, nsep, sess_list);
742 			nsep->sess_clp = clp;
743 			NFSUNLOCKSESSION(shp);
744 			NFSUNLOCKSTATE();
745 		    }
746 		}
747 	} else if (clp->lc_flags & LCL_NEEDSCONFIRM) {
748 		error = NFSERR_EXPIRED;
749 	}
750 
751 	/*
752 	 * If called by the Renew Op, we must check the principal.
753 	 */
754 	if (!error && (opflags & CLOPS_RENEWOP)) {
755 	    if (nfsrv_notsamecredname(nd, clp)) {
756 		doneok = 0;
757 		for (i = 0; i < nfsrv_statehashsize && doneok == 0; i++) {
758 		    LIST_FOREACH(stp, &clp->lc_stateid[i], ls_hash) {
759 			if ((stp->ls_flags & NFSLCK_OPEN) &&
760 			    stp->ls_uid == nd->nd_cred->cr_uid) {
761 				doneok = 1;
762 				break;
763 			}
764 		    }
765 		}
766 		if (!doneok)
767 			error = NFSERR_ACCES;
768 	    }
769 	    if (!error && (clp->lc_flags & LCL_CBDOWN))
770 		error = NFSERR_CBPATHDOWN;
771 	}
772 	if ((!error || error == NFSERR_CBPATHDOWN) &&
773 	     (opflags & CLOPS_RENEW)) {
774 		clp->lc_expiry = nfsrv_leaseexpiry();
775 	}
776 	if (opflags & CLOPS_CONFIRM) {
777 		NFSLOCKV4ROOTMUTEX();
778 		nfsv4_unlock(&nfsv4rootfs_lock, 1);
779 		NFSUNLOCKV4ROOTMUTEX();
780 	} else if (opflags != CLOPS_RENEW) {
781 		NFSUNLOCKSTATE();
782 	}
783 	if (clpp)
784 		*clpp = clp;
785 
786 out:
787 	NFSEXITCODE2(error, nd);
788 	return (error);
789 }
790 
791 /*
792  * Perform the NFSv4.1 destroy clientid.
793  */
794 int
795 nfsrv_destroyclient(nfsquad_t clientid, NFSPROC_T *p)
796 {
797 	struct nfsclient *clp;
798 	struct nfsclienthashhead *hp;
799 	int error = 0, i, igotlock;
800 
801 	if (nfsrvboottime != clientid.lval[0]) {
802 		error = NFSERR_STALECLIENTID;
803 		goto out;
804 	}
805 
806 	/* Lock out other nfsd threads */
807 	NFSLOCKV4ROOTMUTEX();
808 	nfsv4_relref(&nfsv4rootfs_lock);
809 	do {
810 		igotlock = nfsv4_lock(&nfsv4rootfs_lock, 1, NULL,
811 		    NFSV4ROOTLOCKMUTEXPTR, NULL);
812 	} while (igotlock == 0);
813 	NFSUNLOCKV4ROOTMUTEX();
814 
815 	hp = NFSCLIENTHASH(clientid);
816 	LIST_FOREACH(clp, hp, lc_hash) {
817 		if (clp->lc_clientid.lval[1] == clientid.lval[1])
818 			break;
819 	}
820 	if (clp == NULL) {
821 		NFSLOCKV4ROOTMUTEX();
822 		nfsv4_unlock(&nfsv4rootfs_lock, 1);
823 		NFSUNLOCKV4ROOTMUTEX();
824 		/* Just return ok, since it is gone. */
825 		goto out;
826 	}
827 
828 	/*
829 	 * Free up all layouts on the clientid.  Should the client return the
830 	 * layouts?
831 	 */
832 	nfsrv_freelayoutlist(clientid);
833 
834 	/* Scan for state on the clientid. */
835 	for (i = 0; i < nfsrv_statehashsize; i++)
836 		if (!LIST_EMPTY(&clp->lc_stateid[i])) {
837 			NFSLOCKV4ROOTMUTEX();
838 			nfsv4_unlock(&nfsv4rootfs_lock, 1);
839 			NFSUNLOCKV4ROOTMUTEX();
840 			error = NFSERR_CLIENTIDBUSY;
841 			goto out;
842 		}
843 	if (!LIST_EMPTY(&clp->lc_session) || !LIST_EMPTY(&clp->lc_deleg)) {
844 		NFSLOCKV4ROOTMUTEX();
845 		nfsv4_unlock(&nfsv4rootfs_lock, 1);
846 		NFSUNLOCKV4ROOTMUTEX();
847 		error = NFSERR_CLIENTIDBUSY;
848 		goto out;
849 	}
850 
851 	/* Destroy the clientid and return ok. */
852 	nfsrv_cleanclient(clp, p);
853 	nfsrv_freedeleglist(&clp->lc_deleg);
854 	nfsrv_freedeleglist(&clp->lc_olddeleg);
855 	LIST_REMOVE(clp, lc_hash);
856 	NFSLOCKV4ROOTMUTEX();
857 	nfsv4_unlock(&nfsv4rootfs_lock, 1);
858 	NFSUNLOCKV4ROOTMUTEX();
859 	nfsrv_zapclient(clp, p);
860 out:
861 	NFSEXITCODE2(error, nd);
862 	return (error);
863 }
864 
865 /*
866  * Called from the new nfssvc syscall to admin revoke a clientid.
867  * Returns 0 for success, error otherwise.
868  */
869 APPLESTATIC int
870 nfsrv_adminrevoke(struct nfsd_clid *revokep, NFSPROC_T *p)
871 {
872 	struct nfsclient *clp = NULL;
873 	int i, error = 0;
874 	int gotit, igotlock;
875 
876 	/*
877 	 * First, lock out the nfsd so that state won't change while the
878 	 * revocation record is being written to the stable storage restart
879 	 * file.
880 	 */
881 	NFSLOCKV4ROOTMUTEX();
882 	do {
883 		igotlock = nfsv4_lock(&nfsv4rootfs_lock, 1, NULL,
884 		    NFSV4ROOTLOCKMUTEXPTR, NULL);
885 	} while (!igotlock);
886 	NFSUNLOCKV4ROOTMUTEX();
887 
888 	/*
889 	 * Search for a match in the client list.
890 	 */
891 	gotit = i = 0;
892 	while (i < nfsrv_clienthashsize && !gotit) {
893 	    LIST_FOREACH(clp, &nfsclienthash[i], lc_hash) {
894 		if (revokep->nclid_idlen == clp->lc_idlen &&
895 		    !NFSBCMP(revokep->nclid_id, clp->lc_id, clp->lc_idlen)) {
896 			gotit = 1;
897 			break;
898 		}
899 	    }
900 	    i++;
901 	}
902 	if (!gotit) {
903 		NFSLOCKV4ROOTMUTEX();
904 		nfsv4_unlock(&nfsv4rootfs_lock, 0);
905 		NFSUNLOCKV4ROOTMUTEX();
906 		error = EPERM;
907 		goto out;
908 	}
909 
910 	/*
911 	 * Now, write out the revocation record
912 	 */
913 	nfsrv_writestable(clp->lc_id, clp->lc_idlen, NFSNST_REVOKE, p);
914 	nfsrv_backupstable();
915 
916 	/*
917 	 * and clear out the state, marking the clientid revoked.
918 	 */
919 	clp->lc_flags &= ~LCL_CALLBACKSON;
920 	clp->lc_flags |= LCL_ADMINREVOKED;
921 	nfsrv_cleanclient(clp, p);
922 	nfsrv_freedeleglist(&clp->lc_deleg);
923 	nfsrv_freedeleglist(&clp->lc_olddeleg);
924 	NFSLOCKV4ROOTMUTEX();
925 	nfsv4_unlock(&nfsv4rootfs_lock, 0);
926 	NFSUNLOCKV4ROOTMUTEX();
927 
928 out:
929 	NFSEXITCODE(error);
930 	return (error);
931 }
932 
933 /*
934  * Dump out stats for all clients. Called from nfssvc(2), that is used
935  * nfsstatsv1.
936  */
937 APPLESTATIC void
938 nfsrv_dumpclients(struct nfsd_dumpclients *dumpp, int maxcnt)
939 {
940 	struct nfsclient *clp;
941 	int i = 0, cnt = 0;
942 
943 	/*
944 	 * First, get a reference on the nfsv4rootfs_lock so that an
945 	 * exclusive lock cannot be acquired while dumping the clients.
946 	 */
947 	NFSLOCKV4ROOTMUTEX();
948 	nfsv4_getref(&nfsv4rootfs_lock, NULL, NFSV4ROOTLOCKMUTEXPTR, NULL);
949 	NFSUNLOCKV4ROOTMUTEX();
950 	NFSLOCKSTATE();
951 	/*
952 	 * Rattle through the client lists until done.
953 	 */
954 	while (i < nfsrv_clienthashsize && cnt < maxcnt) {
955 	    clp = LIST_FIRST(&nfsclienthash[i]);
956 	    while (clp != LIST_END(&nfsclienthash[i]) && cnt < maxcnt) {
957 		nfsrv_dumpaclient(clp, &dumpp[cnt]);
958 		cnt++;
959 		clp = LIST_NEXT(clp, lc_hash);
960 	    }
961 	    i++;
962 	}
963 	if (cnt < maxcnt)
964 	    dumpp[cnt].ndcl_clid.nclid_idlen = 0;
965 	NFSUNLOCKSTATE();
966 	NFSLOCKV4ROOTMUTEX();
967 	nfsv4_relref(&nfsv4rootfs_lock);
968 	NFSUNLOCKV4ROOTMUTEX();
969 }
970 
971 /*
972  * Dump stats for a client. Must be called with the NFSSTATELOCK and spl'd.
973  */
974 static void
975 nfsrv_dumpaclient(struct nfsclient *clp, struct nfsd_dumpclients *dumpp)
976 {
977 	struct nfsstate *stp, *openstp, *lckownstp;
978 	struct nfslock *lop;
979 	sa_family_t af;
980 #ifdef INET
981 	struct sockaddr_in *rin;
982 #endif
983 #ifdef INET6
984 	struct sockaddr_in6 *rin6;
985 #endif
986 
987 	dumpp->ndcl_nopenowners = dumpp->ndcl_nlockowners = 0;
988 	dumpp->ndcl_nopens = dumpp->ndcl_nlocks = 0;
989 	dumpp->ndcl_ndelegs = dumpp->ndcl_nolddelegs = 0;
990 	dumpp->ndcl_flags = clp->lc_flags;
991 	dumpp->ndcl_clid.nclid_idlen = clp->lc_idlen;
992 	NFSBCOPY(clp->lc_id, dumpp->ndcl_clid.nclid_id, clp->lc_idlen);
993 	af = clp->lc_req.nr_nam->sa_family;
994 	dumpp->ndcl_addrfam = af;
995 	switch (af) {
996 #ifdef INET
997 	case AF_INET:
998 		rin = (struct sockaddr_in *)clp->lc_req.nr_nam;
999 		dumpp->ndcl_cbaddr.sin_addr = rin->sin_addr;
1000 		break;
1001 #endif
1002 #ifdef INET6
1003 	case AF_INET6:
1004 		rin6 = (struct sockaddr_in6 *)clp->lc_req.nr_nam;
1005 		dumpp->ndcl_cbaddr.sin6_addr = rin6->sin6_addr;
1006 		break;
1007 #endif
1008 	}
1009 
1010 	/*
1011 	 * Now, scan the state lists and total up the opens and locks.
1012 	 */
1013 	LIST_FOREACH(stp, &clp->lc_open, ls_list) {
1014 	    dumpp->ndcl_nopenowners++;
1015 	    LIST_FOREACH(openstp, &stp->ls_open, ls_list) {
1016 		dumpp->ndcl_nopens++;
1017 		LIST_FOREACH(lckownstp, &openstp->ls_open, ls_list) {
1018 		    dumpp->ndcl_nlockowners++;
1019 		    LIST_FOREACH(lop, &lckownstp->ls_lock, lo_lckowner) {
1020 			dumpp->ndcl_nlocks++;
1021 		    }
1022 		}
1023 	    }
1024 	}
1025 
1026 	/*
1027 	 * and the delegation lists.
1028 	 */
1029 	LIST_FOREACH(stp, &clp->lc_deleg, ls_list) {
1030 	    dumpp->ndcl_ndelegs++;
1031 	}
1032 	LIST_FOREACH(stp, &clp->lc_olddeleg, ls_list) {
1033 	    dumpp->ndcl_nolddelegs++;
1034 	}
1035 }
1036 
1037 /*
1038  * Dump out lock stats for a file.
1039  */
1040 APPLESTATIC void
1041 nfsrv_dumplocks(vnode_t vp, struct nfsd_dumplocks *ldumpp, int maxcnt,
1042     NFSPROC_T *p)
1043 {
1044 	struct nfsstate *stp;
1045 	struct nfslock *lop;
1046 	int cnt = 0;
1047 	struct nfslockfile *lfp;
1048 	sa_family_t af;
1049 #ifdef INET
1050 	struct sockaddr_in *rin;
1051 #endif
1052 #ifdef INET6
1053 	struct sockaddr_in6 *rin6;
1054 #endif
1055 	int ret;
1056 	fhandle_t nfh;
1057 
1058 	ret = nfsrv_getlockfh(vp, 0, NULL, &nfh, p);
1059 	/*
1060 	 * First, get a reference on the nfsv4rootfs_lock so that an
1061 	 * exclusive lock on it cannot be acquired while dumping the locks.
1062 	 */
1063 	NFSLOCKV4ROOTMUTEX();
1064 	nfsv4_getref(&nfsv4rootfs_lock, NULL, NFSV4ROOTLOCKMUTEXPTR, NULL);
1065 	NFSUNLOCKV4ROOTMUTEX();
1066 	NFSLOCKSTATE();
1067 	if (!ret)
1068 		ret = nfsrv_getlockfile(0, NULL, &lfp, &nfh, 0);
1069 	if (ret) {
1070 		ldumpp[0].ndlck_clid.nclid_idlen = 0;
1071 		NFSUNLOCKSTATE();
1072 		NFSLOCKV4ROOTMUTEX();
1073 		nfsv4_relref(&nfsv4rootfs_lock);
1074 		NFSUNLOCKV4ROOTMUTEX();
1075 		return;
1076 	}
1077 
1078 	/*
1079 	 * For each open share on file, dump it out.
1080 	 */
1081 	stp = LIST_FIRST(&lfp->lf_open);
1082 	while (stp != LIST_END(&lfp->lf_open) && cnt < maxcnt) {
1083 		ldumpp[cnt].ndlck_flags = stp->ls_flags;
1084 		ldumpp[cnt].ndlck_stateid.seqid = stp->ls_stateid.seqid;
1085 		ldumpp[cnt].ndlck_stateid.other[0] = stp->ls_stateid.other[0];
1086 		ldumpp[cnt].ndlck_stateid.other[1] = stp->ls_stateid.other[1];
1087 		ldumpp[cnt].ndlck_stateid.other[2] = stp->ls_stateid.other[2];
1088 		ldumpp[cnt].ndlck_owner.nclid_idlen =
1089 		    stp->ls_openowner->ls_ownerlen;
1090 		NFSBCOPY(stp->ls_openowner->ls_owner,
1091 		    ldumpp[cnt].ndlck_owner.nclid_id,
1092 		    stp->ls_openowner->ls_ownerlen);
1093 		ldumpp[cnt].ndlck_clid.nclid_idlen = stp->ls_clp->lc_idlen;
1094 		NFSBCOPY(stp->ls_clp->lc_id, ldumpp[cnt].ndlck_clid.nclid_id,
1095 		    stp->ls_clp->lc_idlen);
1096 		af = stp->ls_clp->lc_req.nr_nam->sa_family;
1097 		ldumpp[cnt].ndlck_addrfam = af;
1098 		switch (af) {
1099 #ifdef INET
1100 		case AF_INET:
1101 			rin = (struct sockaddr_in *)stp->ls_clp->lc_req.nr_nam;
1102 			ldumpp[cnt].ndlck_cbaddr.sin_addr = rin->sin_addr;
1103 			break;
1104 #endif
1105 #ifdef INET6
1106 		case AF_INET6:
1107 			rin6 = (struct sockaddr_in6 *)
1108 			    stp->ls_clp->lc_req.nr_nam;
1109 			ldumpp[cnt].ndlck_cbaddr.sin6_addr = rin6->sin6_addr;
1110 			break;
1111 #endif
1112 		}
1113 		stp = LIST_NEXT(stp, ls_file);
1114 		cnt++;
1115 	}
1116 
1117 	/*
1118 	 * and all locks.
1119 	 */
1120 	lop = LIST_FIRST(&lfp->lf_lock);
1121 	while (lop != LIST_END(&lfp->lf_lock) && cnt < maxcnt) {
1122 		stp = lop->lo_stp;
1123 		ldumpp[cnt].ndlck_flags = lop->lo_flags;
1124 		ldumpp[cnt].ndlck_first = lop->lo_first;
1125 		ldumpp[cnt].ndlck_end = lop->lo_end;
1126 		ldumpp[cnt].ndlck_stateid.seqid = stp->ls_stateid.seqid;
1127 		ldumpp[cnt].ndlck_stateid.other[0] = stp->ls_stateid.other[0];
1128 		ldumpp[cnt].ndlck_stateid.other[1] = stp->ls_stateid.other[1];
1129 		ldumpp[cnt].ndlck_stateid.other[2] = stp->ls_stateid.other[2];
1130 		ldumpp[cnt].ndlck_owner.nclid_idlen = stp->ls_ownerlen;
1131 		NFSBCOPY(stp->ls_owner, ldumpp[cnt].ndlck_owner.nclid_id,
1132 		    stp->ls_ownerlen);
1133 		ldumpp[cnt].ndlck_clid.nclid_idlen = stp->ls_clp->lc_idlen;
1134 		NFSBCOPY(stp->ls_clp->lc_id, ldumpp[cnt].ndlck_clid.nclid_id,
1135 		    stp->ls_clp->lc_idlen);
1136 		af = stp->ls_clp->lc_req.nr_nam->sa_family;
1137 		ldumpp[cnt].ndlck_addrfam = af;
1138 		switch (af) {
1139 #ifdef INET
1140 		case AF_INET:
1141 			rin = (struct sockaddr_in *)stp->ls_clp->lc_req.nr_nam;
1142 			ldumpp[cnt].ndlck_cbaddr.sin_addr = rin->sin_addr;
1143 			break;
1144 #endif
1145 #ifdef INET6
1146 		case AF_INET6:
1147 			rin6 = (struct sockaddr_in6 *)
1148 			    stp->ls_clp->lc_req.nr_nam;
1149 			ldumpp[cnt].ndlck_cbaddr.sin6_addr = rin6->sin6_addr;
1150 			break;
1151 #endif
1152 		}
1153 		lop = LIST_NEXT(lop, lo_lckfile);
1154 		cnt++;
1155 	}
1156 
1157 	/*
1158 	 * and the delegations.
1159 	 */
1160 	stp = LIST_FIRST(&lfp->lf_deleg);
1161 	while (stp != LIST_END(&lfp->lf_deleg) && cnt < maxcnt) {
1162 		ldumpp[cnt].ndlck_flags = stp->ls_flags;
1163 		ldumpp[cnt].ndlck_stateid.seqid = stp->ls_stateid.seqid;
1164 		ldumpp[cnt].ndlck_stateid.other[0] = stp->ls_stateid.other[0];
1165 		ldumpp[cnt].ndlck_stateid.other[1] = stp->ls_stateid.other[1];
1166 		ldumpp[cnt].ndlck_stateid.other[2] = stp->ls_stateid.other[2];
1167 		ldumpp[cnt].ndlck_owner.nclid_idlen = 0;
1168 		ldumpp[cnt].ndlck_clid.nclid_idlen = stp->ls_clp->lc_idlen;
1169 		NFSBCOPY(stp->ls_clp->lc_id, ldumpp[cnt].ndlck_clid.nclid_id,
1170 		    stp->ls_clp->lc_idlen);
1171 		af = stp->ls_clp->lc_req.nr_nam->sa_family;
1172 		ldumpp[cnt].ndlck_addrfam = af;
1173 		switch (af) {
1174 #ifdef INET
1175 		case AF_INET:
1176 			rin = (struct sockaddr_in *)stp->ls_clp->lc_req.nr_nam;
1177 			ldumpp[cnt].ndlck_cbaddr.sin_addr = rin->sin_addr;
1178 			break;
1179 #endif
1180 #ifdef INET6
1181 		case AF_INET6:
1182 			rin6 = (struct sockaddr_in6 *)
1183 			    stp->ls_clp->lc_req.nr_nam;
1184 			ldumpp[cnt].ndlck_cbaddr.sin6_addr = rin6->sin6_addr;
1185 			break;
1186 #endif
1187 		}
1188 		stp = LIST_NEXT(stp, ls_file);
1189 		cnt++;
1190 	}
1191 
1192 	/*
1193 	 * If list isn't full, mark end of list by setting the client name
1194 	 * to zero length.
1195 	 */
1196 	if (cnt < maxcnt)
1197 		ldumpp[cnt].ndlck_clid.nclid_idlen = 0;
1198 	NFSUNLOCKSTATE();
1199 	NFSLOCKV4ROOTMUTEX();
1200 	nfsv4_relref(&nfsv4rootfs_lock);
1201 	NFSUNLOCKV4ROOTMUTEX();
1202 }
1203 
1204 /*
1205  * Server timer routine. It can scan any linked list, so long
1206  * as it holds the spin/mutex lock and there is no exclusive lock on
1207  * nfsv4rootfs_lock.
1208  * (For OpenBSD, a kthread is ok. For FreeBSD, I think it is ok
1209  *  to do this from a callout, since the spin locks work. For
1210  *  Darwin, I'm not sure what will work correctly yet.)
1211  * Should be called once per second.
1212  */
1213 APPLESTATIC void
1214 nfsrv_servertimer(void)
1215 {
1216 	struct nfsclient *clp, *nclp;
1217 	struct nfsstate *stp, *nstp;
1218 	int got_ref, i;
1219 
1220 	/*
1221 	 * Make sure nfsboottime is set. This is used by V3 as well
1222 	 * as V4. Note that nfsboottime is not nfsrvboottime, which is
1223 	 * only used by the V4 server for leases.
1224 	 */
1225 	if (nfsboottime.tv_sec == 0)
1226 		NFSSETBOOTTIME(nfsboottime);
1227 
1228 	/*
1229 	 * If server hasn't started yet, just return.
1230 	 */
1231 	NFSLOCKSTATE();
1232 	if (nfsrv_stablefirst.nsf_eograce == 0) {
1233 		NFSUNLOCKSTATE();
1234 		return;
1235 	}
1236 	if (!(nfsrv_stablefirst.nsf_flags & NFSNSF_UPDATEDONE)) {
1237 		if (!(nfsrv_stablefirst.nsf_flags & NFSNSF_GRACEOVER) &&
1238 		    NFSD_MONOSEC > nfsrv_stablefirst.nsf_eograce)
1239 			nfsrv_stablefirst.nsf_flags |=
1240 			    (NFSNSF_GRACEOVER | NFSNSF_NEEDLOCK);
1241 		NFSUNLOCKSTATE();
1242 		return;
1243 	}
1244 
1245 	/*
1246 	 * Try and get a reference count on the nfsv4rootfs_lock so that
1247 	 * no nfsd thread can acquire an exclusive lock on it before this
1248 	 * call is done. If it is already exclusively locked, just return.
1249 	 */
1250 	NFSLOCKV4ROOTMUTEX();
1251 	got_ref = nfsv4_getref_nonblock(&nfsv4rootfs_lock);
1252 	NFSUNLOCKV4ROOTMUTEX();
1253 	if (got_ref == 0) {
1254 		NFSUNLOCKSTATE();
1255 		return;
1256 	}
1257 
1258 	/*
1259 	 * For each client...
1260 	 */
1261 	for (i = 0; i < nfsrv_clienthashsize; i++) {
1262 	    clp = LIST_FIRST(&nfsclienthash[i]);
1263 	    while (clp != LIST_END(&nfsclienthash[i])) {
1264 		nclp = LIST_NEXT(clp, lc_hash);
1265 		if (!(clp->lc_flags & LCL_EXPIREIT)) {
1266 		    if (((clp->lc_expiry + NFSRV_STALELEASE) < NFSD_MONOSEC
1267 			 && ((LIST_EMPTY(&clp->lc_deleg)
1268 			      && LIST_EMPTY(&clp->lc_open)) ||
1269 			     nfsrv_clients > nfsrv_clienthighwater)) ||
1270 			(clp->lc_expiry + NFSRV_MOULDYLEASE) < NFSD_MONOSEC ||
1271 			(clp->lc_expiry < NFSD_MONOSEC &&
1272 			 (nfsrv_openpluslock * 10 / 9) > nfsrv_v4statelimit)) {
1273 			/*
1274 			 * Lease has expired several nfsrv_lease times ago:
1275 			 * PLUS
1276 			 *    - no state is associated with it
1277 			 *    OR
1278 			 *    - above high water mark for number of clients
1279 			 *      (nfsrv_clienthighwater should be large enough
1280 			 *       that this only occurs when clients fail to
1281 			 *       use the same nfs_client_id4.id. Maybe somewhat
1282 			 *       higher that the maximum number of clients that
1283 			 *       will mount this server?)
1284 			 * OR
1285 			 * Lease has expired a very long time ago
1286 			 * OR
1287 			 * Lease has expired PLUS the number of opens + locks
1288 			 * has exceeded 90% of capacity
1289 			 *
1290 			 * --> Mark for expiry. The actual expiry will be done
1291 			 *     by an nfsd sometime soon.
1292 			 */
1293 			clp->lc_flags |= LCL_EXPIREIT;
1294 			nfsrv_stablefirst.nsf_flags |=
1295 			    (NFSNSF_NEEDLOCK | NFSNSF_EXPIREDCLIENT);
1296 		    } else {
1297 			/*
1298 			 * If there are no opens, increment no open tick cnt
1299 			 * If time exceeds NFSNOOPEN, mark it to be thrown away
1300 			 * otherwise, if there is an open, reset no open time
1301 			 * Hopefully, this will avoid excessive re-creation
1302 			 * of open owners and subsequent open confirms.
1303 			 */
1304 			stp = LIST_FIRST(&clp->lc_open);
1305 			while (stp != LIST_END(&clp->lc_open)) {
1306 				nstp = LIST_NEXT(stp, ls_list);
1307 				if (LIST_EMPTY(&stp->ls_open)) {
1308 					stp->ls_noopens++;
1309 					if (stp->ls_noopens > NFSNOOPEN ||
1310 					    (nfsrv_openpluslock * 2) >
1311 					    nfsrv_v4statelimit)
1312 						nfsrv_stablefirst.nsf_flags |=
1313 							NFSNSF_NOOPENS;
1314 				} else {
1315 					stp->ls_noopens = 0;
1316 				}
1317 				stp = nstp;
1318 			}
1319 		    }
1320 		}
1321 		clp = nclp;
1322 	    }
1323 	}
1324 	NFSUNLOCKSTATE();
1325 	NFSLOCKV4ROOTMUTEX();
1326 	nfsv4_relref(&nfsv4rootfs_lock);
1327 	NFSUNLOCKV4ROOTMUTEX();
1328 }
1329 
1330 /*
1331  * The following set of functions free up the various data structures.
1332  */
1333 /*
1334  * Clear out all open/lock state related to this nfsclient.
1335  * Caller must hold an exclusive lock on nfsv4rootfs_lock, so that
1336  * there are no other active nfsd threads.
1337  */
1338 APPLESTATIC void
1339 nfsrv_cleanclient(struct nfsclient *clp, NFSPROC_T *p)
1340 {
1341 	struct nfsstate *stp, *nstp;
1342 	struct nfsdsession *sep, *nsep;
1343 
1344 	LIST_FOREACH_SAFE(stp, &clp->lc_open, ls_list, nstp)
1345 		nfsrv_freeopenowner(stp, 1, p);
1346 	if ((clp->lc_flags & LCL_ADMINREVOKED) == 0)
1347 		LIST_FOREACH_SAFE(sep, &clp->lc_session, sess_list, nsep)
1348 			(void)nfsrv_freesession(sep, NULL);
1349 }
1350 
1351 /*
1352  * Free a client that has been cleaned. It should also already have been
1353  * removed from the lists.
1354  * (Just to be safe w.r.t. newnfs_disconnect(), call this function when
1355  *  softclock interrupts are enabled.)
1356  */
1357 APPLESTATIC void
1358 nfsrv_zapclient(struct nfsclient *clp, NFSPROC_T *p)
1359 {
1360 
1361 #ifdef notyet
1362 	if ((clp->lc_flags & (LCL_GSS | LCL_CALLBACKSON)) ==
1363 	     (LCL_GSS | LCL_CALLBACKSON) &&
1364 	    (clp->lc_hand.nfsh_flag & NFSG_COMPLETE) &&
1365 	    clp->lc_handlelen > 0) {
1366 		clp->lc_hand.nfsh_flag &= ~NFSG_COMPLETE;
1367 		clp->lc_hand.nfsh_flag |= NFSG_DESTROYED;
1368 		(void) nfsrv_docallback(clp, NFSV4PROC_CBNULL,
1369 			NULL, 0, NULL, NULL, NULL, 0, p);
1370 	}
1371 #endif
1372 	newnfs_disconnect(&clp->lc_req);
1373 	free(clp->lc_req.nr_nam, M_SONAME);
1374 	NFSFREEMUTEX(&clp->lc_req.nr_mtx);
1375 	free(clp->lc_stateid, M_NFSDCLIENT);
1376 	free(clp, M_NFSDCLIENT);
1377 	NFSLOCKSTATE();
1378 	nfsstatsv1.srvclients--;
1379 	nfsrv_openpluslock--;
1380 	nfsrv_clients--;
1381 	NFSUNLOCKSTATE();
1382 }
1383 
1384 /*
1385  * Free a list of delegation state structures.
1386  * (This function will also free all nfslockfile structures that no
1387  *  longer have associated state.)
1388  */
1389 APPLESTATIC void
1390 nfsrv_freedeleglist(struct nfsstatehead *sthp)
1391 {
1392 	struct nfsstate *stp, *nstp;
1393 
1394 	LIST_FOREACH_SAFE(stp, sthp, ls_list, nstp) {
1395 		nfsrv_freedeleg(stp);
1396 	}
1397 	LIST_INIT(sthp);
1398 }
1399 
1400 /*
1401  * Free up a delegation.
1402  */
1403 static void
1404 nfsrv_freedeleg(struct nfsstate *stp)
1405 {
1406 	struct nfslockfile *lfp;
1407 
1408 	LIST_REMOVE(stp, ls_hash);
1409 	LIST_REMOVE(stp, ls_list);
1410 	LIST_REMOVE(stp, ls_file);
1411 	if ((stp->ls_flags & NFSLCK_DELEGWRITE) != 0)
1412 		nfsrv_writedelegcnt--;
1413 	lfp = stp->ls_lfp;
1414 	if (LIST_EMPTY(&lfp->lf_open) &&
1415 	    LIST_EMPTY(&lfp->lf_lock) && LIST_EMPTY(&lfp->lf_deleg) &&
1416 	    LIST_EMPTY(&lfp->lf_locallock) && LIST_EMPTY(&lfp->lf_rollback) &&
1417 	    lfp->lf_usecount == 0 &&
1418 	    nfsv4_testlock(&lfp->lf_locallock_lck) == 0)
1419 		nfsrv_freenfslockfile(lfp);
1420 	free(stp, M_NFSDSTATE);
1421 	nfsstatsv1.srvdelegates--;
1422 	nfsrv_openpluslock--;
1423 	nfsrv_delegatecnt--;
1424 }
1425 
1426 /*
1427  * This function frees an open owner and all associated opens.
1428  */
1429 static void
1430 nfsrv_freeopenowner(struct nfsstate *stp, int cansleep, NFSPROC_T *p)
1431 {
1432 	struct nfsstate *nstp, *tstp;
1433 
1434 	LIST_REMOVE(stp, ls_list);
1435 	/*
1436 	 * Now, free all associated opens.
1437 	 */
1438 	nstp = LIST_FIRST(&stp->ls_open);
1439 	while (nstp != LIST_END(&stp->ls_open)) {
1440 		tstp = nstp;
1441 		nstp = LIST_NEXT(nstp, ls_list);
1442 		(void) nfsrv_freeopen(tstp, NULL, cansleep, p);
1443 	}
1444 	if (stp->ls_op)
1445 		nfsrvd_derefcache(stp->ls_op);
1446 	free(stp, M_NFSDSTATE);
1447 	nfsstatsv1.srvopenowners--;
1448 	nfsrv_openpluslock--;
1449 }
1450 
1451 /*
1452  * This function frees an open (nfsstate open structure) with all associated
1453  * lock_owners and locks. It also frees the nfslockfile structure iff there
1454  * are no other opens on the file.
1455  * Returns 1 if it free'd the nfslockfile, 0 otherwise.
1456  */
1457 static int
1458 nfsrv_freeopen(struct nfsstate *stp, vnode_t vp, int cansleep, NFSPROC_T *p)
1459 {
1460 	struct nfsstate *nstp, *tstp;
1461 	struct nfslockfile *lfp;
1462 	int ret;
1463 
1464 	LIST_REMOVE(stp, ls_hash);
1465 	LIST_REMOVE(stp, ls_list);
1466 	LIST_REMOVE(stp, ls_file);
1467 
1468 	lfp = stp->ls_lfp;
1469 	/*
1470 	 * Now, free all lockowners associated with this open.
1471 	 */
1472 	LIST_FOREACH_SAFE(tstp, &stp->ls_open, ls_list, nstp)
1473 		nfsrv_freelockowner(tstp, vp, cansleep, p);
1474 
1475 	/*
1476 	 * The nfslockfile is freed here if there are no locks
1477 	 * associated with the open.
1478 	 * If there are locks associated with the open, the
1479 	 * nfslockfile structure can be freed via nfsrv_freelockowner().
1480 	 * Acquire the state mutex to avoid races with calls to
1481 	 * nfsrv_getlockfile().
1482 	 */
1483 	if (cansleep != 0)
1484 		NFSLOCKSTATE();
1485 	if (lfp != NULL && LIST_EMPTY(&lfp->lf_open) &&
1486 	    LIST_EMPTY(&lfp->lf_deleg) && LIST_EMPTY(&lfp->lf_lock) &&
1487 	    LIST_EMPTY(&lfp->lf_locallock) && LIST_EMPTY(&lfp->lf_rollback) &&
1488 	    lfp->lf_usecount == 0 &&
1489 	    (cansleep != 0 || nfsv4_testlock(&lfp->lf_locallock_lck) == 0)) {
1490 		nfsrv_freenfslockfile(lfp);
1491 		ret = 1;
1492 	} else
1493 		ret = 0;
1494 	if (cansleep != 0)
1495 		NFSUNLOCKSTATE();
1496 	free(stp, M_NFSDSTATE);
1497 	nfsstatsv1.srvopens--;
1498 	nfsrv_openpluslock--;
1499 	return (ret);
1500 }
1501 
1502 /*
1503  * Frees a lockowner and all associated locks.
1504  */
1505 static void
1506 nfsrv_freelockowner(struct nfsstate *stp, vnode_t vp, int cansleep,
1507     NFSPROC_T *p)
1508 {
1509 
1510 	LIST_REMOVE(stp, ls_hash);
1511 	LIST_REMOVE(stp, ls_list);
1512 	nfsrv_freeallnfslocks(stp, vp, cansleep, p);
1513 	if (stp->ls_op)
1514 		nfsrvd_derefcache(stp->ls_op);
1515 	free(stp, M_NFSDSTATE);
1516 	nfsstatsv1.srvlockowners--;
1517 	nfsrv_openpluslock--;
1518 }
1519 
1520 /*
1521  * Free all the nfs locks on a lockowner.
1522  */
1523 static void
1524 nfsrv_freeallnfslocks(struct nfsstate *stp, vnode_t vp, int cansleep,
1525     NFSPROC_T *p)
1526 {
1527 	struct nfslock *lop, *nlop;
1528 	struct nfsrollback *rlp, *nrlp;
1529 	struct nfslockfile *lfp = NULL;
1530 	int gottvp = 0;
1531 	vnode_t tvp = NULL;
1532 	uint64_t first, end;
1533 
1534 	if (vp != NULL)
1535 		ASSERT_VOP_UNLOCKED(vp, "nfsrv_freeallnfslocks: vnode locked");
1536 	lop = LIST_FIRST(&stp->ls_lock);
1537 	while (lop != LIST_END(&stp->ls_lock)) {
1538 		nlop = LIST_NEXT(lop, lo_lckowner);
1539 		/*
1540 		 * Since all locks should be for the same file, lfp should
1541 		 * not change.
1542 		 */
1543 		if (lfp == NULL)
1544 			lfp = lop->lo_lfp;
1545 		else if (lfp != lop->lo_lfp)
1546 			panic("allnfslocks");
1547 		/*
1548 		 * If vp is NULL and cansleep != 0, a vnode must be acquired
1549 		 * from the file handle. This only occurs when called from
1550 		 * nfsrv_cleanclient().
1551 		 */
1552 		if (gottvp == 0) {
1553 			if (nfsrv_dolocallocks == 0)
1554 				tvp = NULL;
1555 			else if (vp == NULL && cansleep != 0) {
1556 				tvp = nfsvno_getvp(&lfp->lf_fh);
1557 				NFSVOPUNLOCK(tvp);
1558 			} else
1559 				tvp = vp;
1560 			gottvp = 1;
1561 		}
1562 
1563 		if (tvp != NULL) {
1564 			if (cansleep == 0)
1565 				panic("allnfs2");
1566 			first = lop->lo_first;
1567 			end = lop->lo_end;
1568 			nfsrv_freenfslock(lop);
1569 			nfsrv_localunlock(tvp, lfp, first, end, p);
1570 			LIST_FOREACH_SAFE(rlp, &lfp->lf_rollback, rlck_list,
1571 			    nrlp)
1572 				free(rlp, M_NFSDROLLBACK);
1573 			LIST_INIT(&lfp->lf_rollback);
1574 		} else
1575 			nfsrv_freenfslock(lop);
1576 		lop = nlop;
1577 	}
1578 	if (vp == NULL && tvp != NULL)
1579 		vrele(tvp);
1580 }
1581 
1582 /*
1583  * Free an nfslock structure.
1584  */
1585 static void
1586 nfsrv_freenfslock(struct nfslock *lop)
1587 {
1588 
1589 	if (lop->lo_lckfile.le_prev != NULL) {
1590 		LIST_REMOVE(lop, lo_lckfile);
1591 		nfsstatsv1.srvlocks--;
1592 		nfsrv_openpluslock--;
1593 	}
1594 	LIST_REMOVE(lop, lo_lckowner);
1595 	free(lop, M_NFSDLOCK);
1596 }
1597 
1598 /*
1599  * This function frees an nfslockfile structure.
1600  */
1601 static void
1602 nfsrv_freenfslockfile(struct nfslockfile *lfp)
1603 {
1604 
1605 	LIST_REMOVE(lfp, lf_hash);
1606 	free(lfp, M_NFSDLOCKFILE);
1607 }
1608 
1609 /*
1610  * This function looks up an nfsstate structure via stateid.
1611  */
1612 static int
1613 nfsrv_getstate(struct nfsclient *clp, nfsv4stateid_t *stateidp, __unused u_int32_t flags,
1614     struct nfsstate **stpp)
1615 {
1616 	struct nfsstate *stp;
1617 	struct nfsstatehead *hp;
1618 	int error = 0;
1619 
1620 	*stpp = NULL;
1621 	hp = NFSSTATEHASH(clp, *stateidp);
1622 	LIST_FOREACH(stp, hp, ls_hash) {
1623 		if (!NFSBCMP(stp->ls_stateid.other, stateidp->other,
1624 			NFSX_STATEIDOTHER))
1625 			break;
1626 	}
1627 
1628 	/*
1629 	 * If no state id in list, return NFSERR_BADSTATEID.
1630 	 */
1631 	if (stp == LIST_END(hp)) {
1632 		error = NFSERR_BADSTATEID;
1633 		goto out;
1634 	}
1635 	*stpp = stp;
1636 
1637 out:
1638 	NFSEXITCODE(error);
1639 	return (error);
1640 }
1641 
1642 /*
1643  * This function gets an nfsstate structure via owner string.
1644  */
1645 static void
1646 nfsrv_getowner(struct nfsstatehead *hp, struct nfsstate *new_stp,
1647     struct nfsstate **stpp)
1648 {
1649 	struct nfsstate *stp;
1650 
1651 	*stpp = NULL;
1652 	LIST_FOREACH(stp, hp, ls_list) {
1653 		if (new_stp->ls_ownerlen == stp->ls_ownerlen &&
1654 		  !NFSBCMP(new_stp->ls_owner,stp->ls_owner,stp->ls_ownerlen)) {
1655 			*stpp = stp;
1656 			return;
1657 		}
1658 	}
1659 }
1660 
1661 /*
1662  * Lock control function called to update lock status.
1663  * Returns 0 upon success, -1 if there is no lock and the flags indicate
1664  * that one isn't to be created and an NFSERR_xxx for other errors.
1665  * The structures new_stp and new_lop are passed in as pointers that should
1666  * be set to NULL if the structure is used and shouldn't be free'd.
1667  * For the NFSLCK_TEST and NFSLCK_CHECK cases, the structures are
1668  * never used and can safely be allocated on the stack. For all other
1669  * cases, *new_stpp and *new_lopp should be malloc'd before the call,
1670  * in case they are used.
1671  */
1672 APPLESTATIC int
1673 nfsrv_lockctrl(vnode_t vp, struct nfsstate **new_stpp,
1674     struct nfslock **new_lopp, struct nfslockconflict *cfp,
1675     nfsquad_t clientid, nfsv4stateid_t *stateidp,
1676     __unused struct nfsexstuff *exp,
1677     struct nfsrv_descript *nd, NFSPROC_T *p)
1678 {
1679 	struct nfslock *lop;
1680 	struct nfsstate *new_stp = *new_stpp;
1681 	struct nfslock *new_lop = *new_lopp;
1682 	struct nfsstate *tstp, *mystp, *nstp;
1683 	int specialid = 0;
1684 	struct nfslockfile *lfp;
1685 	struct nfslock *other_lop = NULL;
1686 	struct nfsstate *stp, *lckstp = NULL;
1687 	struct nfsclient *clp = NULL;
1688 	u_int32_t bits;
1689 	int error = 0, haslock = 0, ret, reterr;
1690 	int getlckret, delegation = 0, filestruct_locked, vnode_unlocked = 0;
1691 	fhandle_t nfh;
1692 	uint64_t first, end;
1693 	uint32_t lock_flags;
1694 
1695 	if (new_stp->ls_flags & (NFSLCK_CHECK | NFSLCK_SETATTR)) {
1696 		/*
1697 		 * Note the special cases of "all 1s" or "all 0s" stateids and
1698 		 * let reads with all 1s go ahead.
1699 		 */
1700 		if (new_stp->ls_stateid.seqid == 0x0 &&
1701 		    new_stp->ls_stateid.other[0] == 0x0 &&
1702 		    new_stp->ls_stateid.other[1] == 0x0 &&
1703 		    new_stp->ls_stateid.other[2] == 0x0)
1704 			specialid = 1;
1705 		else if (new_stp->ls_stateid.seqid == 0xffffffff &&
1706 		    new_stp->ls_stateid.other[0] == 0xffffffff &&
1707 		    new_stp->ls_stateid.other[1] == 0xffffffff &&
1708 		    new_stp->ls_stateid.other[2] == 0xffffffff)
1709 			specialid = 2;
1710 	}
1711 
1712 	/*
1713 	 * Check for restart conditions (client and server).
1714 	 */
1715 	error = nfsrv_checkrestart(clientid, new_stp->ls_flags,
1716 	    &new_stp->ls_stateid, specialid);
1717 	if (error)
1718 		goto out;
1719 
1720 	/*
1721 	 * Check for state resource limit exceeded.
1722 	 */
1723 	if ((new_stp->ls_flags & NFSLCK_LOCK) &&
1724 	    nfsrv_openpluslock > nfsrv_v4statelimit) {
1725 		error = NFSERR_RESOURCE;
1726 		goto out;
1727 	}
1728 
1729 	/*
1730 	 * For the lock case, get another nfslock structure,
1731 	 * just in case we need it.
1732 	 * Malloc now, before we start sifting through the linked lists,
1733 	 * in case we have to wait for memory.
1734 	 */
1735 tryagain:
1736 	if (new_stp->ls_flags & NFSLCK_LOCK)
1737 		other_lop = malloc(sizeof (struct nfslock),
1738 		    M_NFSDLOCK, M_WAITOK);
1739 	filestruct_locked = 0;
1740 	reterr = 0;
1741 	lfp = NULL;
1742 
1743 	/*
1744 	 * Get the lockfile structure for CFH now, so we can do a sanity
1745 	 * check against the stateid, before incrementing the seqid#, since
1746 	 * we want to return NFSERR_BADSTATEID on failure and the seqid#
1747 	 * shouldn't be incremented for this case.
1748 	 * If nfsrv_getlockfile() returns -1, it means "not found", which
1749 	 * will be handled later.
1750 	 * If we are doing Lock/LockU and local locking is enabled, sleep
1751 	 * lock the nfslockfile structure.
1752 	 */
1753 	getlckret = nfsrv_getlockfh(vp, new_stp->ls_flags, NULL, &nfh, p);
1754 	NFSLOCKSTATE();
1755 	if (getlckret == 0) {
1756 		if ((new_stp->ls_flags & (NFSLCK_LOCK | NFSLCK_UNLOCK)) != 0 &&
1757 		    nfsrv_dolocallocks != 0 && nd->nd_repstat == 0) {
1758 			getlckret = nfsrv_getlockfile(new_stp->ls_flags, NULL,
1759 			    &lfp, &nfh, 1);
1760 			if (getlckret == 0)
1761 				filestruct_locked = 1;
1762 		} else
1763 			getlckret = nfsrv_getlockfile(new_stp->ls_flags, NULL,
1764 			    &lfp, &nfh, 0);
1765 	}
1766 	if (getlckret != 0 && getlckret != -1)
1767 		reterr = getlckret;
1768 
1769 	if (filestruct_locked != 0) {
1770 		LIST_INIT(&lfp->lf_rollback);
1771 		if ((new_stp->ls_flags & NFSLCK_LOCK)) {
1772 			/*
1773 			 * For local locking, do the advisory locking now, so
1774 			 * that any conflict can be detected. A failure later
1775 			 * can be rolled back locally. If an error is returned,
1776 			 * struct nfslockfile has been unlocked and any local
1777 			 * locking rolled back.
1778 			 */
1779 			NFSUNLOCKSTATE();
1780 			if (vnode_unlocked == 0) {
1781 				ASSERT_VOP_ELOCKED(vp, "nfsrv_lockctrl1");
1782 				vnode_unlocked = 1;
1783 				NFSVOPUNLOCK(vp);
1784 			}
1785 			reterr = nfsrv_locallock(vp, lfp,
1786 			    (new_lop->lo_flags & (NFSLCK_READ | NFSLCK_WRITE)),
1787 			    new_lop->lo_first, new_lop->lo_end, cfp, p);
1788 			NFSLOCKSTATE();
1789 		}
1790 	}
1791 
1792 	if (specialid == 0) {
1793 	    if (new_stp->ls_flags & NFSLCK_TEST) {
1794 		/*
1795 		 * RFC 3530 does not list LockT as an op that renews a
1796 		 * lease, but the consensus seems to be that it is ok
1797 		 * for a server to do so.
1798 		 */
1799 		error = nfsrv_getclient(clientid, CLOPS_RENEW, &clp, NULL,
1800 		    (nfsquad_t)((u_quad_t)0), 0, nd, p);
1801 
1802 		/*
1803 		 * Since NFSERR_EXPIRED, NFSERR_ADMINREVOKED are not valid
1804 		 * error returns for LockT, just go ahead and test for a lock,
1805 		 * since there are no locks for this client, but other locks
1806 		 * can conflict. (ie. same client will always be false)
1807 		 */
1808 		if (error == NFSERR_EXPIRED || error == NFSERR_ADMINREVOKED)
1809 		    error = 0;
1810 		lckstp = new_stp;
1811 	    } else {
1812 	      error = nfsrv_getclient(clientid, CLOPS_RENEW, &clp, NULL,
1813 		(nfsquad_t)((u_quad_t)0), 0, nd, p);
1814 	      if (error == 0)
1815 		/*
1816 		 * Look up the stateid
1817 		 */
1818 		error = nfsrv_getstate(clp, &new_stp->ls_stateid,
1819 		  new_stp->ls_flags, &stp);
1820 	      /*
1821 	       * do some sanity checks for an unconfirmed open or a
1822 	       * stateid that refers to the wrong file, for an open stateid
1823 	       */
1824 	      if (error == 0 && (stp->ls_flags & NFSLCK_OPEN) &&
1825 		  ((stp->ls_openowner->ls_flags & NFSLCK_NEEDSCONFIRM) ||
1826 		   (getlckret == 0 && stp->ls_lfp != lfp))){
1827 		      /*
1828 		       * NFSLCK_SETATTR should return OK rather than NFSERR_BADSTATEID
1829 		       * The only exception is using SETATTR with SIZE.
1830 		       * */
1831                     if ((new_stp->ls_flags &
1832                          (NFSLCK_SETATTR | NFSLCK_CHECK)) != NFSLCK_SETATTR)
1833 			     error = NFSERR_BADSTATEID;
1834 	      }
1835 
1836 		if (error == 0 &&
1837 		  (stp->ls_flags & (NFSLCK_DELEGREAD | NFSLCK_DELEGWRITE)) &&
1838 		  getlckret == 0 && stp->ls_lfp != lfp)
1839 			error = NFSERR_BADSTATEID;
1840 
1841 	      /*
1842 	       * If the lockowner stateid doesn't refer to the same file,
1843 	       * I believe that is considered ok, since some clients will
1844 	       * only create a single lockowner and use that for all locks
1845 	       * on all files.
1846 	       * For now, log it as a diagnostic, instead of considering it
1847 	       * a BadStateid.
1848 	       */
1849 	      if (error == 0 && (stp->ls_flags &
1850 		  (NFSLCK_OPEN | NFSLCK_DELEGREAD | NFSLCK_DELEGWRITE)) == 0 &&
1851 		  getlckret == 0 && stp->ls_lfp != lfp) {
1852 #ifdef DIAGNOSTIC
1853 		  printf("Got a lock statid for different file open\n");
1854 #endif
1855 		  /*
1856 		  error = NFSERR_BADSTATEID;
1857 		  */
1858 	      }
1859 
1860 	      if (error == 0) {
1861 		    if (new_stp->ls_flags & NFSLCK_OPENTOLOCK) {
1862 			/*
1863 			 * If haslock set, we've already checked the seqid.
1864 			 */
1865 			if (!haslock) {
1866 			    if (stp->ls_flags & NFSLCK_OPEN)
1867 				error = nfsrv_checkseqid(nd, new_stp->ls_seq,
1868 				    stp->ls_openowner, new_stp->ls_op);
1869 			    else
1870 				error = NFSERR_BADSTATEID;
1871 			}
1872 			if (!error)
1873 			    nfsrv_getowner(&stp->ls_open, new_stp, &lckstp);
1874 			if (lckstp)
1875 			    /*
1876 			     * I believe this should be an error, but it
1877 			     * isn't obvious what NFSERR_xxx would be
1878 			     * appropriate, so I'll use NFSERR_INVAL for now.
1879 			     */
1880 			    error = NFSERR_INVAL;
1881 			else
1882 			    lckstp = new_stp;
1883 		    } else if (new_stp->ls_flags&(NFSLCK_LOCK|NFSLCK_UNLOCK)) {
1884 			/*
1885 			 * If haslock set, ditto above.
1886 			 */
1887 			if (!haslock) {
1888 			    if (stp->ls_flags & NFSLCK_OPEN)
1889 				error = NFSERR_BADSTATEID;
1890 			    else
1891 				error = nfsrv_checkseqid(nd, new_stp->ls_seq,
1892 				    stp, new_stp->ls_op);
1893 			}
1894 			lckstp = stp;
1895 		    } else {
1896 			lckstp = stp;
1897 		    }
1898 	      }
1899 	      /*
1900 	       * If the seqid part of the stateid isn't the same, return
1901 	       * NFSERR_OLDSTATEID for cases other than I/O Ops.
1902 	       * For I/O Ops, only return NFSERR_OLDSTATEID if
1903 	       * nfsrv_returnoldstateid is set. (The consensus on the email
1904 	       * list was that most clients would prefer to not receive
1905 	       * NFSERR_OLDSTATEID for I/O Ops, but the RFC suggests that that
1906 	       * is what will happen, so I use the nfsrv_returnoldstateid to
1907 	       * allow for either server configuration.)
1908 	       */
1909 	      if (!error && stp->ls_stateid.seqid!=new_stp->ls_stateid.seqid &&
1910 		  (((nd->nd_flag & ND_NFSV41) == 0 &&
1911 		   (!(new_stp->ls_flags & NFSLCK_CHECK) ||
1912 		    nfsrv_returnoldstateid)) ||
1913 		   ((nd->nd_flag & ND_NFSV41) != 0 &&
1914 		    new_stp->ls_stateid.seqid != 0)))
1915 		    error = NFSERR_OLDSTATEID;
1916 	    }
1917 	}
1918 
1919 	/*
1920 	 * Now we can check for grace.
1921 	 */
1922 	if (!error)
1923 		error = nfsrv_checkgrace(nd, clp, new_stp->ls_flags);
1924 	if ((new_stp->ls_flags & NFSLCK_RECLAIM) && !error &&
1925 		nfsrv_checkstable(clp))
1926 		error = NFSERR_NOGRACE;
1927 	/*
1928 	 * If we successfully Reclaimed state, note that.
1929 	 */
1930 	if ((new_stp->ls_flags & NFSLCK_RECLAIM) && !error)
1931 		nfsrv_markstable(clp);
1932 
1933 	/*
1934 	 * At this point, either error == NFSERR_BADSTATEID or the
1935 	 * seqid# has been updated, so we can return any error.
1936 	 * If error == 0, there may be an error in:
1937 	 *    nd_repstat - Set by the calling function.
1938 	 *    reterr - Set above, if getting the nfslockfile structure
1939 	 *       or acquiring the local lock failed.
1940 	 *    (If both of these are set, nd_repstat should probably be
1941 	 *     returned, since that error was detected before this
1942 	 *     function call.)
1943 	 */
1944 	if (error != 0 || nd->nd_repstat != 0 || reterr != 0) {
1945 		if (error == 0) {
1946 			if (nd->nd_repstat != 0)
1947 				error = nd->nd_repstat;
1948 			else
1949 				error = reterr;
1950 		}
1951 		if (filestruct_locked != 0) {
1952 			/* Roll back local locks. */
1953 			NFSUNLOCKSTATE();
1954 			if (vnode_unlocked == 0) {
1955 				ASSERT_VOP_ELOCKED(vp, "nfsrv_lockctrl2");
1956 				vnode_unlocked = 1;
1957 				NFSVOPUNLOCK(vp);
1958 			}
1959 			nfsrv_locallock_rollback(vp, lfp, p);
1960 			NFSLOCKSTATE();
1961 			nfsrv_unlocklf(lfp);
1962 		}
1963 		NFSUNLOCKSTATE();
1964 		goto out;
1965 	}
1966 
1967 	/*
1968 	 * Check the nfsrv_getlockfile return.
1969 	 * Returned -1 if no structure found.
1970 	 */
1971 	if (getlckret == -1) {
1972 		error = NFSERR_EXPIRED;
1973 		/*
1974 		 * Called from lockt, so no lock is OK.
1975 		 */
1976 		if (new_stp->ls_flags & NFSLCK_TEST) {
1977 			error = 0;
1978 		} else if (new_stp->ls_flags &
1979 		    (NFSLCK_CHECK | NFSLCK_SETATTR)) {
1980 			/*
1981 			 * Called to check for a lock, OK if the stateid is all
1982 			 * 1s or all 0s, but there should be an nfsstate
1983 			 * otherwise.
1984 			 * (ie. If there is no open, I'll assume no share
1985 			 *  deny bits.)
1986 			 */
1987 			if (specialid)
1988 				error = 0;
1989 			else
1990 				error = NFSERR_BADSTATEID;
1991 		}
1992 		NFSUNLOCKSTATE();
1993 		goto out;
1994 	}
1995 
1996 	/*
1997 	 * For NFSLCK_CHECK and NFSLCK_LOCK, test for a share conflict.
1998 	 * For NFSLCK_CHECK, allow a read if write access is granted,
1999 	 * but check for a deny. For NFSLCK_LOCK, require correct access,
2000 	 * which implies a conflicting deny can't exist.
2001 	 */
2002 	if (new_stp->ls_flags & (NFSLCK_CHECK | NFSLCK_LOCK)) {
2003 	    /*
2004 	     * Four kinds of state id:
2005 	     * - specialid (all 0s or all 1s), only for NFSLCK_CHECK
2006 	     * - stateid for an open
2007 	     * - stateid for a delegation
2008 	     * - stateid for a lock owner
2009 	     */
2010 	    if (!specialid) {
2011 		if (stp->ls_flags & (NFSLCK_DELEGREAD | NFSLCK_DELEGWRITE)) {
2012 		    delegation = 1;
2013 		    mystp = stp;
2014 		    nfsrv_delaydelegtimeout(stp);
2015 	        } else if (stp->ls_flags & NFSLCK_OPEN) {
2016 		    mystp = stp;
2017 		} else {
2018 		    mystp = stp->ls_openstp;
2019 		}
2020 		/*
2021 		 * If locking or checking, require correct access
2022 		 * bit set.
2023 		 */
2024 		if (((new_stp->ls_flags & NFSLCK_LOCK) &&
2025 		     !((new_lop->lo_flags >> NFSLCK_LOCKSHIFT) &
2026 		       mystp->ls_flags & NFSLCK_ACCESSBITS)) ||
2027 		    ((new_stp->ls_flags & (NFSLCK_CHECK|NFSLCK_READACCESS)) ==
2028 		      (NFSLCK_CHECK | NFSLCK_READACCESS) &&
2029 		     !(mystp->ls_flags & NFSLCK_READACCESS) &&
2030 		     nfsrv_allowreadforwriteopen == 0) ||
2031 		    ((new_stp->ls_flags & (NFSLCK_CHECK|NFSLCK_WRITEACCESS)) ==
2032 		      (NFSLCK_CHECK | NFSLCK_WRITEACCESS) &&
2033 		     !(mystp->ls_flags & NFSLCK_WRITEACCESS))) {
2034 			if (filestruct_locked != 0) {
2035 				/* Roll back local locks. */
2036 				NFSUNLOCKSTATE();
2037 				if (vnode_unlocked == 0) {
2038 					ASSERT_VOP_ELOCKED(vp,
2039 					    "nfsrv_lockctrl3");
2040 					vnode_unlocked = 1;
2041 					NFSVOPUNLOCK(vp);
2042 				}
2043 				nfsrv_locallock_rollback(vp, lfp, p);
2044 				NFSLOCKSTATE();
2045 				nfsrv_unlocklf(lfp);
2046 			}
2047 			NFSUNLOCKSTATE();
2048 			error = NFSERR_OPENMODE;
2049 			goto out;
2050 		}
2051 	    } else
2052 		mystp = NULL;
2053 	    if ((new_stp->ls_flags & NFSLCK_CHECK) && !delegation) {
2054 		/*
2055 		 * Check for a conflicting deny bit.
2056 		 */
2057 		LIST_FOREACH(tstp, &lfp->lf_open, ls_file) {
2058 		    if (tstp != mystp) {
2059 			bits = tstp->ls_flags;
2060 			bits >>= NFSLCK_SHIFT;
2061 			if (new_stp->ls_flags & bits & NFSLCK_ACCESSBITS) {
2062 			    KASSERT(vnode_unlocked == 0,
2063 				("nfsrv_lockctrl: vnode unlocked1"));
2064 			    ret = nfsrv_clientconflict(tstp->ls_clp, &haslock,
2065 				vp, p);
2066 			    if (ret == 1) {
2067 				/*
2068 				* nfsrv_clientconflict unlocks state
2069 				 * when it returns non-zero.
2070 				 */
2071 				lckstp = NULL;
2072 				goto tryagain;
2073 			    }
2074 			    if (ret == 0)
2075 				NFSUNLOCKSTATE();
2076 			    if (ret == 2)
2077 				error = NFSERR_PERM;
2078 			    else
2079 				error = NFSERR_OPENMODE;
2080 			    goto out;
2081 			}
2082 		    }
2083 		}
2084 
2085 		/* We're outta here */
2086 		NFSUNLOCKSTATE();
2087 		goto out;
2088 	    }
2089 	}
2090 
2091 	/*
2092 	 * For setattr, just get rid of all the Delegations for other clients.
2093 	 */
2094 	if (new_stp->ls_flags & NFSLCK_SETATTR) {
2095 		KASSERT(vnode_unlocked == 0,
2096 		    ("nfsrv_lockctrl: vnode unlocked2"));
2097 		ret = nfsrv_cleandeleg(vp, lfp, clp, &haslock, p);
2098 		if (ret) {
2099 			/*
2100 			 * nfsrv_cleandeleg() unlocks state when it
2101 			 * returns non-zero.
2102 			 */
2103 			if (ret == -1) {
2104 				lckstp = NULL;
2105 				goto tryagain;
2106 			}
2107 			error = ret;
2108 			goto out;
2109 		}
2110 		if (!(new_stp->ls_flags & NFSLCK_CHECK) ||
2111 		    (LIST_EMPTY(&lfp->lf_open) && LIST_EMPTY(&lfp->lf_lock) &&
2112 		     LIST_EMPTY(&lfp->lf_deleg))) {
2113 			NFSUNLOCKSTATE();
2114 			goto out;
2115 		}
2116 	}
2117 
2118 	/*
2119 	 * Check for a conflicting delegation. If one is found, call
2120 	 * nfsrv_delegconflict() to handle it. If the v4root lock hasn't
2121 	 * been set yet, it will get the lock. Otherwise, it will recall
2122 	 * the delegation. Then, we try try again...
2123 	 * I currently believe the conflict algorithm to be:
2124 	 * For Lock Ops (Lock/LockT/LockU)
2125 	 * - there is a conflict iff a different client has a write delegation
2126 	 * For Reading (Read Op)
2127 	 * - there is a conflict iff a different client has a write delegation
2128 	 *   (the specialids are always a different client)
2129 	 * For Writing (Write/Setattr of size)
2130 	 * - there is a conflict if a different client has any delegation
2131 	 * - there is a conflict if the same client has a read delegation
2132 	 *   (I don't understand why this isn't allowed, but that seems to be
2133 	 *    the current consensus?)
2134 	 */
2135 	tstp = LIST_FIRST(&lfp->lf_deleg);
2136 	while (tstp != LIST_END(&lfp->lf_deleg)) {
2137 	    nstp = LIST_NEXT(tstp, ls_file);
2138 	    if ((((new_stp->ls_flags&(NFSLCK_LOCK|NFSLCK_UNLOCK|NFSLCK_TEST))||
2139 		 ((new_stp->ls_flags & NFSLCK_CHECK) &&
2140 		  (new_lop->lo_flags & NFSLCK_READ))) &&
2141 		  clp != tstp->ls_clp &&
2142 		 (tstp->ls_flags & NFSLCK_DELEGWRITE)) ||
2143 		 ((new_stp->ls_flags & NFSLCK_CHECK) &&
2144 		   (new_lop->lo_flags & NFSLCK_WRITE) &&
2145 		  (clp != tstp->ls_clp ||
2146 		   (tstp->ls_flags & NFSLCK_DELEGREAD)))) {
2147 		ret = 0;
2148 		if (filestruct_locked != 0) {
2149 			/* Roll back local locks. */
2150 			NFSUNLOCKSTATE();
2151 			if (vnode_unlocked == 0) {
2152 				ASSERT_VOP_ELOCKED(vp, "nfsrv_lockctrl4");
2153 				NFSVOPUNLOCK(vp);
2154 			}
2155 			nfsrv_locallock_rollback(vp, lfp, p);
2156 			NFSLOCKSTATE();
2157 			nfsrv_unlocklf(lfp);
2158 			NFSUNLOCKSTATE();
2159 			NFSVOPLOCK(vp, LK_EXCLUSIVE | LK_RETRY);
2160 			vnode_unlocked = 0;
2161 			if (VN_IS_DOOMED(vp))
2162 				ret = NFSERR_SERVERFAULT;
2163 			NFSLOCKSTATE();
2164 		}
2165 		if (ret == 0)
2166 			ret = nfsrv_delegconflict(tstp, &haslock, p, vp);
2167 		if (ret) {
2168 		    /*
2169 		     * nfsrv_delegconflict unlocks state when it
2170 		     * returns non-zero, which it always does.
2171 		     */
2172 		    if (other_lop) {
2173 			free(other_lop, M_NFSDLOCK);
2174 			other_lop = NULL;
2175 		    }
2176 		    if (ret == -1) {
2177 			lckstp = NULL;
2178 			goto tryagain;
2179 		    }
2180 		    error = ret;
2181 		    goto out;
2182 		}
2183 		/* Never gets here. */
2184 	    }
2185 	    tstp = nstp;
2186 	}
2187 
2188 	/*
2189 	 * Handle the unlock case by calling nfsrv_updatelock().
2190 	 * (Should I have done some access checking above for unlock? For now,
2191 	 *  just let it happen.)
2192 	 */
2193 	if (new_stp->ls_flags & NFSLCK_UNLOCK) {
2194 		first = new_lop->lo_first;
2195 		end = new_lop->lo_end;
2196 		nfsrv_updatelock(stp, new_lopp, &other_lop, lfp);
2197 		stateidp->seqid = ++(stp->ls_stateid.seqid);
2198 		if ((nd->nd_flag & ND_NFSV41) != 0 && stateidp->seqid == 0)
2199 			stateidp->seqid = stp->ls_stateid.seqid = 1;
2200 		stateidp->other[0] = stp->ls_stateid.other[0];
2201 		stateidp->other[1] = stp->ls_stateid.other[1];
2202 		stateidp->other[2] = stp->ls_stateid.other[2];
2203 		if (filestruct_locked != 0) {
2204 			NFSUNLOCKSTATE();
2205 			if (vnode_unlocked == 0) {
2206 				ASSERT_VOP_ELOCKED(vp, "nfsrv_lockctrl5");
2207 				vnode_unlocked = 1;
2208 				NFSVOPUNLOCK(vp);
2209 			}
2210 			/* Update the local locks. */
2211 			nfsrv_localunlock(vp, lfp, first, end, p);
2212 			NFSLOCKSTATE();
2213 			nfsrv_unlocklf(lfp);
2214 		}
2215 		NFSUNLOCKSTATE();
2216 		goto out;
2217 	}
2218 
2219 	/*
2220 	 * Search for a conflicting lock. A lock conflicts if:
2221 	 * - the lock range overlaps and
2222 	 * - at least one lock is a write lock and
2223 	 * - it is not owned by the same lock owner
2224 	 */
2225 	if (!delegation) {
2226 	  LIST_FOREACH(lop, &lfp->lf_lock, lo_lckfile) {
2227 	    if (new_lop->lo_end > lop->lo_first &&
2228 		new_lop->lo_first < lop->lo_end &&
2229 		(new_lop->lo_flags == NFSLCK_WRITE ||
2230 		 lop->lo_flags == NFSLCK_WRITE) &&
2231 		lckstp != lop->lo_stp &&
2232 		(clp != lop->lo_stp->ls_clp ||
2233 		 lckstp->ls_ownerlen != lop->lo_stp->ls_ownerlen ||
2234 		 NFSBCMP(lckstp->ls_owner, lop->lo_stp->ls_owner,
2235 		    lckstp->ls_ownerlen))) {
2236 		if (other_lop) {
2237 		    free(other_lop, M_NFSDLOCK);
2238 		    other_lop = NULL;
2239 		}
2240 		if (vnode_unlocked != 0)
2241 		    ret = nfsrv_clientconflict(lop->lo_stp->ls_clp, &haslock,
2242 			NULL, p);
2243 		else
2244 		    ret = nfsrv_clientconflict(lop->lo_stp->ls_clp, &haslock,
2245 			vp, p);
2246 		if (ret == 1) {
2247 		    if (filestruct_locked != 0) {
2248 			if (vnode_unlocked == 0) {
2249 				ASSERT_VOP_ELOCKED(vp, "nfsrv_lockctrl6");
2250 				NFSVOPUNLOCK(vp);
2251 			}
2252 			/* Roll back local locks. */
2253 			nfsrv_locallock_rollback(vp, lfp, p);
2254 			NFSLOCKSTATE();
2255 			nfsrv_unlocklf(lfp);
2256 			NFSUNLOCKSTATE();
2257 			NFSVOPLOCK(vp, LK_EXCLUSIVE | LK_RETRY);
2258 			vnode_unlocked = 0;
2259 			if (VN_IS_DOOMED(vp)) {
2260 				error = NFSERR_SERVERFAULT;
2261 				goto out;
2262 			}
2263 		    }
2264 		    /*
2265 		     * nfsrv_clientconflict() unlocks state when it
2266 		     * returns non-zero.
2267 		     */
2268 		    lckstp = NULL;
2269 		    goto tryagain;
2270 		}
2271 		/*
2272 		 * Found a conflicting lock, so record the conflict and
2273 		 * return the error.
2274 		 */
2275 		if (cfp != NULL && ret == 0) {
2276 		    cfp->cl_clientid.lval[0]=lop->lo_stp->ls_stateid.other[0];
2277 		    cfp->cl_clientid.lval[1]=lop->lo_stp->ls_stateid.other[1];
2278 		    cfp->cl_first = lop->lo_first;
2279 		    cfp->cl_end = lop->lo_end;
2280 		    cfp->cl_flags = lop->lo_flags;
2281 		    cfp->cl_ownerlen = lop->lo_stp->ls_ownerlen;
2282 		    NFSBCOPY(lop->lo_stp->ls_owner, cfp->cl_owner,
2283 			cfp->cl_ownerlen);
2284 		}
2285 		if (ret == 2)
2286 		    error = NFSERR_PERM;
2287 		else if (new_stp->ls_flags & NFSLCK_RECLAIM)
2288 		    error = NFSERR_RECLAIMCONFLICT;
2289 		else if (new_stp->ls_flags & NFSLCK_CHECK)
2290 		    error = NFSERR_LOCKED;
2291 		else
2292 		    error = NFSERR_DENIED;
2293 		if (filestruct_locked != 0 && ret == 0) {
2294 			/* Roll back local locks. */
2295 			NFSUNLOCKSTATE();
2296 			if (vnode_unlocked == 0) {
2297 				ASSERT_VOP_ELOCKED(vp, "nfsrv_lockctrl7");
2298 				vnode_unlocked = 1;
2299 				NFSVOPUNLOCK(vp);
2300 			}
2301 			nfsrv_locallock_rollback(vp, lfp, p);
2302 			NFSLOCKSTATE();
2303 			nfsrv_unlocklf(lfp);
2304 		}
2305 		if (ret == 0)
2306 			NFSUNLOCKSTATE();
2307 		goto out;
2308 	    }
2309 	  }
2310 	}
2311 
2312 	/*
2313 	 * We only get here if there was no lock that conflicted.
2314 	 */
2315 	if (new_stp->ls_flags & (NFSLCK_TEST | NFSLCK_CHECK)) {
2316 		NFSUNLOCKSTATE();
2317 		goto out;
2318 	}
2319 
2320 	/*
2321 	 * We only get here when we are creating or modifying a lock.
2322 	 * There are two variants:
2323 	 * - exist_lock_owner where lock_owner exists
2324 	 * - open_to_lock_owner with new lock_owner
2325 	 */
2326 	first = new_lop->lo_first;
2327 	end = new_lop->lo_end;
2328 	lock_flags = new_lop->lo_flags;
2329 	if (!(new_stp->ls_flags & NFSLCK_OPENTOLOCK)) {
2330 		nfsrv_updatelock(lckstp, new_lopp, &other_lop, lfp);
2331 		stateidp->seqid = ++(lckstp->ls_stateid.seqid);
2332 		if ((nd->nd_flag & ND_NFSV41) != 0 && stateidp->seqid == 0)
2333 			stateidp->seqid = lckstp->ls_stateid.seqid = 1;
2334 		stateidp->other[0] = lckstp->ls_stateid.other[0];
2335 		stateidp->other[1] = lckstp->ls_stateid.other[1];
2336 		stateidp->other[2] = lckstp->ls_stateid.other[2];
2337 	} else {
2338 		/*
2339 		 * The new open_to_lock_owner case.
2340 		 * Link the new nfsstate into the lists.
2341 		 */
2342 		new_stp->ls_seq = new_stp->ls_opentolockseq;
2343 		nfsrvd_refcache(new_stp->ls_op);
2344 		stateidp->seqid = new_stp->ls_stateid.seqid = 1;
2345 		stateidp->other[0] = new_stp->ls_stateid.other[0] =
2346 		    clp->lc_clientid.lval[0];
2347 		stateidp->other[1] = new_stp->ls_stateid.other[1] =
2348 		    clp->lc_clientid.lval[1];
2349 		stateidp->other[2] = new_stp->ls_stateid.other[2] =
2350 		    nfsrv_nextstateindex(clp);
2351 		new_stp->ls_clp = clp;
2352 		LIST_INIT(&new_stp->ls_lock);
2353 		new_stp->ls_openstp = stp;
2354 		new_stp->ls_lfp = lfp;
2355 		nfsrv_insertlock(new_lop, (struct nfslock *)new_stp, new_stp,
2356 		    lfp);
2357 		LIST_INSERT_HEAD(NFSSTATEHASH(clp, new_stp->ls_stateid),
2358 		    new_stp, ls_hash);
2359 		LIST_INSERT_HEAD(&stp->ls_open, new_stp, ls_list);
2360 		*new_lopp = NULL;
2361 		*new_stpp = NULL;
2362 		nfsstatsv1.srvlockowners++;
2363 		nfsrv_openpluslock++;
2364 	}
2365 	if (filestruct_locked != 0) {
2366 		NFSUNLOCKSTATE();
2367 		nfsrv_locallock_commit(lfp, lock_flags, first, end);
2368 		NFSLOCKSTATE();
2369 		nfsrv_unlocklf(lfp);
2370 	}
2371 	NFSUNLOCKSTATE();
2372 
2373 out:
2374 	if (haslock) {
2375 		NFSLOCKV4ROOTMUTEX();
2376 		nfsv4_unlock(&nfsv4rootfs_lock, 1);
2377 		NFSUNLOCKV4ROOTMUTEX();
2378 	}
2379 	if (vnode_unlocked != 0) {
2380 		NFSVOPLOCK(vp, LK_EXCLUSIVE | LK_RETRY);
2381 		if (error == 0 && VN_IS_DOOMED(vp))
2382 			error = NFSERR_SERVERFAULT;
2383 	}
2384 	if (other_lop)
2385 		free(other_lop, M_NFSDLOCK);
2386 	NFSEXITCODE2(error, nd);
2387 	return (error);
2388 }
2389 
2390 /*
2391  * Check for state errors for Open.
2392  * repstat is passed back out as an error if more critical errors
2393  * are not detected.
2394  */
2395 APPLESTATIC int
2396 nfsrv_opencheck(nfsquad_t clientid, nfsv4stateid_t *stateidp,
2397     struct nfsstate *new_stp, vnode_t vp, struct nfsrv_descript *nd,
2398     NFSPROC_T *p, int repstat)
2399 {
2400 	struct nfsstate *stp, *nstp;
2401 	struct nfsclient *clp;
2402 	struct nfsstate *ownerstp;
2403 	struct nfslockfile *lfp, *new_lfp;
2404 	int error = 0, haslock = 0, ret, readonly = 0, getfhret = 0;
2405 
2406 	if ((new_stp->ls_flags & NFSLCK_SHAREBITS) == NFSLCK_READACCESS)
2407 		readonly = 1;
2408 	/*
2409 	 * Check for restart conditions (client and server).
2410 	 */
2411 	error = nfsrv_checkrestart(clientid, new_stp->ls_flags,
2412 		&new_stp->ls_stateid, 0);
2413 	if (error)
2414 		goto out;
2415 
2416 	/*
2417 	 * Check for state resource limit exceeded.
2418 	 * Technically this should be SMP protected, but the worst
2419 	 * case error is "out by one or two" on the count when it
2420 	 * returns NFSERR_RESOURCE and the limit is just a rather
2421 	 * arbitrary high water mark, so no harm is done.
2422 	 */
2423 	if (nfsrv_openpluslock > nfsrv_v4statelimit) {
2424 		error = NFSERR_RESOURCE;
2425 		goto out;
2426 	}
2427 
2428 tryagain:
2429 	new_lfp = malloc(sizeof (struct nfslockfile),
2430 	    M_NFSDLOCKFILE, M_WAITOK);
2431 	if (vp)
2432 		getfhret = nfsrv_getlockfh(vp, new_stp->ls_flags, new_lfp,
2433 		    NULL, p);
2434 	NFSLOCKSTATE();
2435 	/*
2436 	 * Get the nfsclient structure.
2437 	 */
2438 	error = nfsrv_getclient(clientid, CLOPS_RENEW, &clp, NULL,
2439 	    (nfsquad_t)((u_quad_t)0), 0, nd, p);
2440 
2441 	/*
2442 	 * Look up the open owner. See if it needs confirmation and
2443 	 * check the seq#, as required.
2444 	 */
2445 	if (!error)
2446 		nfsrv_getowner(&clp->lc_open, new_stp, &ownerstp);
2447 
2448 	if (!error && ownerstp) {
2449 		error = nfsrv_checkseqid(nd, new_stp->ls_seq, ownerstp,
2450 		    new_stp->ls_op);
2451 		/*
2452 		 * If the OpenOwner hasn't been confirmed, assume the
2453 		 * old one was a replay and this one is ok.
2454 		 * See: RFC3530 Sec. 14.2.18.
2455 		 */
2456 		if (error == NFSERR_BADSEQID &&
2457 		    (ownerstp->ls_flags & NFSLCK_NEEDSCONFIRM))
2458 			error = 0;
2459 	}
2460 
2461 	/*
2462 	 * Check for grace.
2463 	 */
2464 	if (!error)
2465 		error = nfsrv_checkgrace(nd, clp, new_stp->ls_flags);
2466 	if ((new_stp->ls_flags & NFSLCK_RECLAIM) && !error &&
2467 		nfsrv_checkstable(clp))
2468 		error = NFSERR_NOGRACE;
2469 
2470 	/*
2471 	 * If none of the above errors occurred, let repstat be
2472 	 * returned.
2473 	 */
2474 	if (repstat && !error)
2475 		error = repstat;
2476 	if (error) {
2477 		NFSUNLOCKSTATE();
2478 		if (haslock) {
2479 			NFSLOCKV4ROOTMUTEX();
2480 			nfsv4_unlock(&nfsv4rootfs_lock, 1);
2481 			NFSUNLOCKV4ROOTMUTEX();
2482 		}
2483 		free(new_lfp, M_NFSDLOCKFILE);
2484 		goto out;
2485 	}
2486 
2487 	/*
2488 	 * If vp == NULL, the file doesn't exist yet, so return ok.
2489 	 * (This always happens on the first pass, so haslock must be 0.)
2490 	 */
2491 	if (vp == NULL) {
2492 		NFSUNLOCKSTATE();
2493 		free(new_lfp, M_NFSDLOCKFILE);
2494 		goto out;
2495 	}
2496 
2497 	/*
2498 	 * Get the structure for the underlying file.
2499 	 */
2500 	if (getfhret)
2501 		error = getfhret;
2502 	else
2503 		error = nfsrv_getlockfile(new_stp->ls_flags, &new_lfp, &lfp,
2504 		    NULL, 0);
2505 	if (new_lfp)
2506 		free(new_lfp, M_NFSDLOCKFILE);
2507 	if (error) {
2508 		NFSUNLOCKSTATE();
2509 		if (haslock) {
2510 			NFSLOCKV4ROOTMUTEX();
2511 			nfsv4_unlock(&nfsv4rootfs_lock, 1);
2512 			NFSUNLOCKV4ROOTMUTEX();
2513 		}
2514 		goto out;
2515 	}
2516 
2517 	/*
2518 	 * Search for a conflicting open/share.
2519 	 */
2520 	if (new_stp->ls_flags & NFSLCK_DELEGCUR) {
2521 	    /*
2522 	     * For Delegate_Cur, search for the matching Delegation,
2523 	     * which indicates no conflict.
2524 	     * An old delegation should have been recovered by the
2525 	     * client doing a Claim_DELEGATE_Prev, so I won't let
2526 	     * it match and return NFSERR_EXPIRED. Should I let it
2527 	     * match?
2528 	     */
2529 	    LIST_FOREACH(stp, &lfp->lf_deleg, ls_file) {
2530 		if (!(stp->ls_flags & NFSLCK_OLDDELEG) &&
2531 		    (((nd->nd_flag & ND_NFSV41) != 0 &&
2532 		    stateidp->seqid == 0) ||
2533 		    stateidp->seqid == stp->ls_stateid.seqid) &&
2534 		    !NFSBCMP(stateidp->other, stp->ls_stateid.other,
2535 			  NFSX_STATEIDOTHER))
2536 			break;
2537 	    }
2538 	    if (stp == LIST_END(&lfp->lf_deleg) ||
2539 		((new_stp->ls_flags & NFSLCK_WRITEACCESS) &&
2540 		 (stp->ls_flags & NFSLCK_DELEGREAD))) {
2541 		NFSUNLOCKSTATE();
2542 		if (haslock) {
2543 			NFSLOCKV4ROOTMUTEX();
2544 			nfsv4_unlock(&nfsv4rootfs_lock, 1);
2545 			NFSUNLOCKV4ROOTMUTEX();
2546 		}
2547 		error = NFSERR_EXPIRED;
2548 		goto out;
2549 	    }
2550 	}
2551 
2552 	/*
2553 	 * Check for access/deny bit conflicts. I check for the same
2554 	 * owner as well, in case the client didn't bother.
2555 	 */
2556 	LIST_FOREACH(stp, &lfp->lf_open, ls_file) {
2557 		if (!(new_stp->ls_flags & NFSLCK_DELEGCUR) &&
2558 		    (((new_stp->ls_flags & NFSLCK_ACCESSBITS) &
2559 		      ((stp->ls_flags>>NFSLCK_SHIFT) & NFSLCK_ACCESSBITS))||
2560 		     ((stp->ls_flags & NFSLCK_ACCESSBITS) &
2561 		      ((new_stp->ls_flags>>NFSLCK_SHIFT)&NFSLCK_ACCESSBITS)))){
2562 			ret = nfsrv_clientconflict(stp->ls_clp,&haslock,vp,p);
2563 			if (ret == 1) {
2564 				/*
2565 				 * nfsrv_clientconflict() unlocks
2566 				 * state when it returns non-zero.
2567 				 */
2568 				goto tryagain;
2569 			}
2570 			if (ret == 2)
2571 				error = NFSERR_PERM;
2572 			else if (new_stp->ls_flags & NFSLCK_RECLAIM)
2573 				error = NFSERR_RECLAIMCONFLICT;
2574 			else
2575 				error = NFSERR_SHAREDENIED;
2576 			if (ret == 0)
2577 				NFSUNLOCKSTATE();
2578 			if (haslock) {
2579 				NFSLOCKV4ROOTMUTEX();
2580 				nfsv4_unlock(&nfsv4rootfs_lock, 1);
2581 				NFSUNLOCKV4ROOTMUTEX();
2582 			}
2583 			goto out;
2584 		}
2585 	}
2586 
2587 	/*
2588 	 * Check for a conflicting delegation. If one is found, call
2589 	 * nfsrv_delegconflict() to handle it. If the v4root lock hasn't
2590 	 * been set yet, it will get the lock. Otherwise, it will recall
2591 	 * the delegation. Then, we try try again...
2592 	 * (If NFSLCK_DELEGCUR is set, it has a delegation, so there
2593 	 *  isn't a conflict.)
2594 	 * I currently believe the conflict algorithm to be:
2595 	 * For Open with Read Access and Deny None
2596 	 * - there is a conflict iff a different client has a write delegation
2597 	 * For Open with other Write Access or any Deny except None
2598 	 * - there is a conflict if a different client has any delegation
2599 	 * - there is a conflict if the same client has a read delegation
2600 	 *   (The current consensus is that this last case should be
2601 	 *    considered a conflict since the client with a read delegation
2602 	 *    could have done an Open with ReadAccess and WriteDeny
2603 	 *    locally and then not have checked for the WriteDeny.)
2604 	 * Don't check for a Reclaim, since that will be dealt with
2605 	 * by nfsrv_openctrl().
2606 	 */
2607 	if (!(new_stp->ls_flags &
2608 		(NFSLCK_DELEGPREV | NFSLCK_DELEGCUR | NFSLCK_RECLAIM))) {
2609 	    stp = LIST_FIRST(&lfp->lf_deleg);
2610 	    while (stp != LIST_END(&lfp->lf_deleg)) {
2611 		nstp = LIST_NEXT(stp, ls_file);
2612 		if ((readonly && stp->ls_clp != clp &&
2613 		       (stp->ls_flags & NFSLCK_DELEGWRITE)) ||
2614 		    (!readonly && (stp->ls_clp != clp ||
2615 		         (stp->ls_flags & NFSLCK_DELEGREAD)))) {
2616 			ret = nfsrv_delegconflict(stp, &haslock, p, vp);
2617 			if (ret) {
2618 			    /*
2619 			     * nfsrv_delegconflict() unlocks state
2620 			     * when it returns non-zero.
2621 			     */
2622 			    if (ret == -1)
2623 				goto tryagain;
2624 			    error = ret;
2625 			    goto out;
2626 			}
2627 		}
2628 		stp = nstp;
2629 	    }
2630 	}
2631 	NFSUNLOCKSTATE();
2632 	if (haslock) {
2633 		NFSLOCKV4ROOTMUTEX();
2634 		nfsv4_unlock(&nfsv4rootfs_lock, 1);
2635 		NFSUNLOCKV4ROOTMUTEX();
2636 	}
2637 
2638 out:
2639 	NFSEXITCODE2(error, nd);
2640 	return (error);
2641 }
2642 
2643 /*
2644  * Open control function to create/update open state for an open.
2645  */
2646 APPLESTATIC int
2647 nfsrv_openctrl(struct nfsrv_descript *nd, vnode_t vp,
2648     struct nfsstate **new_stpp, nfsquad_t clientid, nfsv4stateid_t *stateidp,
2649     nfsv4stateid_t *delegstateidp, u_int32_t *rflagsp, struct nfsexstuff *exp,
2650     NFSPROC_T *p, u_quad_t filerev)
2651 {
2652 	struct nfsstate *new_stp = *new_stpp;
2653 	struct nfsstate *stp, *nstp;
2654 	struct nfsstate *openstp = NULL, *new_open, *ownerstp, *new_deleg;
2655 	struct nfslockfile *lfp, *new_lfp;
2656 	struct nfsclient *clp;
2657 	int error = 0, haslock = 0, ret, delegate = 1, writedeleg = 1;
2658 	int readonly = 0, cbret = 1, getfhret = 0;
2659 	int gotstate = 0, len = 0;
2660 	u_char *clidp = NULL;
2661 
2662 	if ((new_stp->ls_flags & NFSLCK_SHAREBITS) == NFSLCK_READACCESS)
2663 		readonly = 1;
2664 	/*
2665 	 * Check for restart conditions (client and server).
2666 	 * (Paranoia, should have been detected by nfsrv_opencheck().)
2667 	 * If an error does show up, return NFSERR_EXPIRED, since the
2668 	 * the seqid# has already been incremented.
2669 	 */
2670 	error = nfsrv_checkrestart(clientid, new_stp->ls_flags,
2671 	    &new_stp->ls_stateid, 0);
2672 	if (error) {
2673 		printf("Nfsd: openctrl unexpected restart err=%d\n",
2674 		    error);
2675 		error = NFSERR_EXPIRED;
2676 		goto out;
2677 	}
2678 
2679 	clidp = malloc(NFSV4_OPAQUELIMIT, M_TEMP, M_WAITOK);
2680 tryagain:
2681 	new_lfp = malloc(sizeof (struct nfslockfile),
2682 	    M_NFSDLOCKFILE, M_WAITOK);
2683 	new_open = malloc(sizeof (struct nfsstate),
2684 	    M_NFSDSTATE, M_WAITOK);
2685 	new_deleg = malloc(sizeof (struct nfsstate),
2686 	    M_NFSDSTATE, M_WAITOK);
2687 	getfhret = nfsrv_getlockfh(vp, new_stp->ls_flags, new_lfp,
2688 	    NULL, p);
2689 	NFSLOCKSTATE();
2690 	/*
2691 	 * Get the client structure. Since the linked lists could be changed
2692 	 * by other nfsd processes if this process does a tsleep(), one of
2693 	 * two things must be done.
2694 	 * 1 - don't tsleep()
2695 	 * or
2696 	 * 2 - get the nfsv4_lock() { indicated by haslock == 1 }
2697 	 *     before using the lists, since this lock stops the other
2698 	 *     nfsd. This should only be used for rare cases, since it
2699 	 *     essentially single threads the nfsd.
2700 	 *     At this time, it is only done for cases where the stable
2701 	 *     storage file must be written prior to completion of state
2702 	 *     expiration.
2703 	 */
2704 	error = nfsrv_getclient(clientid, CLOPS_RENEW, &clp, NULL,
2705 	    (nfsquad_t)((u_quad_t)0), 0, nd, p);
2706 	if (!error && (clp->lc_flags & LCL_NEEDSCBNULL) &&
2707 	    clp->lc_program) {
2708 		/*
2709 		 * This happens on the first open for a client
2710 		 * that supports callbacks.
2711 		 */
2712 		NFSUNLOCKSTATE();
2713 		/*
2714 		 * Although nfsrv_docallback() will sleep, clp won't
2715 		 * go away, since they are only removed when the
2716 		 * nfsv4_lock() has blocked the nfsd threads. The
2717 		 * fields in clp can change, but having multiple
2718 		 * threads do this Null callback RPC should be
2719 		 * harmless.
2720 		 */
2721 		cbret = nfsrv_docallback(clp, NFSV4PROC_CBNULL,
2722 		    NULL, 0, NULL, NULL, NULL, 0, p);
2723 		NFSLOCKSTATE();
2724 		clp->lc_flags &= ~LCL_NEEDSCBNULL;
2725 		if (!cbret)
2726 			clp->lc_flags |= LCL_CALLBACKSON;
2727 	}
2728 
2729 	/*
2730 	 * Look up the open owner. See if it needs confirmation and
2731 	 * check the seq#, as required.
2732 	 */
2733 	if (!error)
2734 		nfsrv_getowner(&clp->lc_open, new_stp, &ownerstp);
2735 
2736 	if (error) {
2737 		NFSUNLOCKSTATE();
2738 		printf("Nfsd: openctrl unexpected state err=%d\n",
2739 			error);
2740 		free(new_lfp, M_NFSDLOCKFILE);
2741 		free(new_open, M_NFSDSTATE);
2742 		free(new_deleg, M_NFSDSTATE);
2743 		if (haslock) {
2744 			NFSLOCKV4ROOTMUTEX();
2745 			nfsv4_unlock(&nfsv4rootfs_lock, 1);
2746 			NFSUNLOCKV4ROOTMUTEX();
2747 		}
2748 		error = NFSERR_EXPIRED;
2749 		goto out;
2750 	}
2751 
2752 	if (new_stp->ls_flags & NFSLCK_RECLAIM)
2753 		nfsrv_markstable(clp);
2754 
2755 	/*
2756 	 * Get the structure for the underlying file.
2757 	 */
2758 	if (getfhret)
2759 		error = getfhret;
2760 	else
2761 		error = nfsrv_getlockfile(new_stp->ls_flags, &new_lfp, &lfp,
2762 		    NULL, 0);
2763 	if (new_lfp)
2764 		free(new_lfp, M_NFSDLOCKFILE);
2765 	if (error) {
2766 		NFSUNLOCKSTATE();
2767 		printf("Nfsd openctrl unexpected getlockfile err=%d\n",
2768 		    error);
2769 		free(new_open, M_NFSDSTATE);
2770 		free(new_deleg, M_NFSDSTATE);
2771 		if (haslock) {
2772 			NFSLOCKV4ROOTMUTEX();
2773 			nfsv4_unlock(&nfsv4rootfs_lock, 1);
2774 			NFSUNLOCKV4ROOTMUTEX();
2775 		}
2776 		goto out;
2777 	}
2778 
2779 	/*
2780 	 * Search for a conflicting open/share.
2781 	 */
2782 	if (new_stp->ls_flags & NFSLCK_DELEGCUR) {
2783 	    /*
2784 	     * For Delegate_Cur, search for the matching Delegation,
2785 	     * which indicates no conflict.
2786 	     * An old delegation should have been recovered by the
2787 	     * client doing a Claim_DELEGATE_Prev, so I won't let
2788 	     * it match and return NFSERR_EXPIRED. Should I let it
2789 	     * match?
2790 	     */
2791 	    LIST_FOREACH(stp, &lfp->lf_deleg, ls_file) {
2792 		if (!(stp->ls_flags & NFSLCK_OLDDELEG) &&
2793 		    (((nd->nd_flag & ND_NFSV41) != 0 &&
2794 		    stateidp->seqid == 0) ||
2795 		    stateidp->seqid == stp->ls_stateid.seqid) &&
2796 		    !NFSBCMP(stateidp->other, stp->ls_stateid.other,
2797 			NFSX_STATEIDOTHER))
2798 			break;
2799 	    }
2800 	    if (stp == LIST_END(&lfp->lf_deleg) ||
2801 		((new_stp->ls_flags & NFSLCK_WRITEACCESS) &&
2802 		 (stp->ls_flags & NFSLCK_DELEGREAD))) {
2803 		NFSUNLOCKSTATE();
2804 		printf("Nfsd openctrl unexpected expiry\n");
2805 		free(new_open, M_NFSDSTATE);
2806 		free(new_deleg, M_NFSDSTATE);
2807 		if (haslock) {
2808 			NFSLOCKV4ROOTMUTEX();
2809 			nfsv4_unlock(&nfsv4rootfs_lock, 1);
2810 			NFSUNLOCKV4ROOTMUTEX();
2811 		}
2812 		error = NFSERR_EXPIRED;
2813 		goto out;
2814 	    }
2815 
2816 	    /*
2817 	     * Don't issue a Delegation, since one already exists and
2818 	     * delay delegation timeout, as required.
2819 	     */
2820 	    delegate = 0;
2821 	    nfsrv_delaydelegtimeout(stp);
2822 	}
2823 
2824 	/*
2825 	 * Check for access/deny bit conflicts. I also check for the
2826 	 * same owner, since the client might not have bothered to check.
2827 	 * Also, note an open for the same file and owner, if found,
2828 	 * which is all we do here for Delegate_Cur, since conflict
2829 	 * checking is already done.
2830 	 */
2831 	LIST_FOREACH(stp, &lfp->lf_open, ls_file) {
2832 		if (ownerstp && stp->ls_openowner == ownerstp)
2833 			openstp = stp;
2834 		if (!(new_stp->ls_flags & NFSLCK_DELEGCUR)) {
2835 		    /*
2836 		     * If another client has the file open, the only
2837 		     * delegation that can be issued is a Read delegation
2838 		     * and only if it is a Read open with Deny none.
2839 		     */
2840 		    if (clp != stp->ls_clp) {
2841 			if ((stp->ls_flags & NFSLCK_SHAREBITS) ==
2842 			    NFSLCK_READACCESS)
2843 			    writedeleg = 0;
2844 			else
2845 			    delegate = 0;
2846 		    }
2847 		    if(((new_stp->ls_flags & NFSLCK_ACCESSBITS) &
2848 		        ((stp->ls_flags>>NFSLCK_SHIFT) & NFSLCK_ACCESSBITS))||
2849 		       ((stp->ls_flags & NFSLCK_ACCESSBITS) &
2850 		        ((new_stp->ls_flags>>NFSLCK_SHIFT)&NFSLCK_ACCESSBITS))){
2851 			ret = nfsrv_clientconflict(stp->ls_clp,&haslock,vp,p);
2852 			if (ret == 1) {
2853 				/*
2854 				 * nfsrv_clientconflict() unlocks state
2855 				 * when it returns non-zero.
2856 				 */
2857 				free(new_open, M_NFSDSTATE);
2858 				free(new_deleg, M_NFSDSTATE);
2859 				openstp = NULL;
2860 				goto tryagain;
2861 			}
2862 			if (ret == 2)
2863 				error = NFSERR_PERM;
2864 			else if (new_stp->ls_flags & NFSLCK_RECLAIM)
2865 				error = NFSERR_RECLAIMCONFLICT;
2866 			else
2867 				error = NFSERR_SHAREDENIED;
2868 			if (ret == 0)
2869 				NFSUNLOCKSTATE();
2870 			if (haslock) {
2871 				NFSLOCKV4ROOTMUTEX();
2872 				nfsv4_unlock(&nfsv4rootfs_lock, 1);
2873 				NFSUNLOCKV4ROOTMUTEX();
2874 			}
2875 			free(new_open, M_NFSDSTATE);
2876 			free(new_deleg, M_NFSDSTATE);
2877 			printf("nfsd openctrl unexpected client cnfl\n");
2878 			goto out;
2879 		    }
2880 		}
2881 	}
2882 
2883 	/*
2884 	 * Check for a conflicting delegation. If one is found, call
2885 	 * nfsrv_delegconflict() to handle it. If the v4root lock hasn't
2886 	 * been set yet, it will get the lock. Otherwise, it will recall
2887 	 * the delegation. Then, we try try again...
2888 	 * (If NFSLCK_DELEGCUR is set, it has a delegation, so there
2889 	 *  isn't a conflict.)
2890 	 * I currently believe the conflict algorithm to be:
2891 	 * For Open with Read Access and Deny None
2892 	 * - there is a conflict iff a different client has a write delegation
2893 	 * For Open with other Write Access or any Deny except None
2894 	 * - there is a conflict if a different client has any delegation
2895 	 * - there is a conflict if the same client has a read delegation
2896 	 *   (The current consensus is that this last case should be
2897 	 *    considered a conflict since the client with a read delegation
2898 	 *    could have done an Open with ReadAccess and WriteDeny
2899 	 *    locally and then not have checked for the WriteDeny.)
2900 	 */
2901 	if (!(new_stp->ls_flags & (NFSLCK_DELEGPREV | NFSLCK_DELEGCUR))) {
2902 	    stp = LIST_FIRST(&lfp->lf_deleg);
2903 	    while (stp != LIST_END(&lfp->lf_deleg)) {
2904 		nstp = LIST_NEXT(stp, ls_file);
2905 		if (stp->ls_clp != clp && (stp->ls_flags & NFSLCK_DELEGREAD))
2906 			writedeleg = 0;
2907 		else
2908 			delegate = 0;
2909 		if ((readonly && stp->ls_clp != clp &&
2910 		       (stp->ls_flags & NFSLCK_DELEGWRITE)) ||
2911 		    (!readonly && (stp->ls_clp != clp ||
2912 		         (stp->ls_flags & NFSLCK_DELEGREAD)))) {
2913 		    if (new_stp->ls_flags & NFSLCK_RECLAIM) {
2914 			delegate = 2;
2915 		    } else {
2916 			ret = nfsrv_delegconflict(stp, &haslock, p, vp);
2917 			if (ret) {
2918 			    /*
2919 			     * nfsrv_delegconflict() unlocks state
2920 			     * when it returns non-zero.
2921 			     */
2922 			    printf("Nfsd openctrl unexpected deleg cnfl\n");
2923 			    free(new_open, M_NFSDSTATE);
2924 			    free(new_deleg, M_NFSDSTATE);
2925 			    if (ret == -1) {
2926 				openstp = NULL;
2927 				goto tryagain;
2928 			    }
2929 			    error = ret;
2930 			    goto out;
2931 			}
2932 		    }
2933 		}
2934 		stp = nstp;
2935 	    }
2936 	}
2937 
2938 	/*
2939 	 * We only get here if there was no open that conflicted.
2940 	 * If an open for the owner exists, or in the access/deny bits.
2941 	 * Otherwise it is a new open. If the open_owner hasn't been
2942 	 * confirmed, replace the open with the new one needing confirmation,
2943 	 * otherwise add the open.
2944 	 */
2945 	if (new_stp->ls_flags & NFSLCK_DELEGPREV) {
2946 	    /*
2947 	     * Handle NFSLCK_DELEGPREV by searching the old delegations for
2948 	     * a match. If found, just move the old delegation to the current
2949 	     * delegation list and issue open. If not found, return
2950 	     * NFSERR_EXPIRED.
2951 	     */
2952 	    LIST_FOREACH(stp, &clp->lc_olddeleg, ls_list) {
2953 		if (stp->ls_lfp == lfp) {
2954 		    /* Found it */
2955 		    if (stp->ls_clp != clp)
2956 			panic("olddeleg clp");
2957 		    LIST_REMOVE(stp, ls_list);
2958 		    LIST_REMOVE(stp, ls_hash);
2959 		    stp->ls_flags &= ~NFSLCK_OLDDELEG;
2960 		    stp->ls_stateid.seqid = delegstateidp->seqid = 1;
2961 		    stp->ls_stateid.other[0] = delegstateidp->other[0] =
2962 			clp->lc_clientid.lval[0];
2963 		    stp->ls_stateid.other[1] = delegstateidp->other[1] =
2964 			clp->lc_clientid.lval[1];
2965 		    stp->ls_stateid.other[2] = delegstateidp->other[2] =
2966 			nfsrv_nextstateindex(clp);
2967 		    stp->ls_compref = nd->nd_compref;
2968 		    LIST_INSERT_HEAD(&clp->lc_deleg, stp, ls_list);
2969 		    LIST_INSERT_HEAD(NFSSTATEHASH(clp,
2970 			stp->ls_stateid), stp, ls_hash);
2971 		    if (stp->ls_flags & NFSLCK_DELEGWRITE)
2972 			*rflagsp |= NFSV4OPEN_WRITEDELEGATE;
2973 		    else
2974 			*rflagsp |= NFSV4OPEN_READDELEGATE;
2975 		    clp->lc_delegtime = NFSD_MONOSEC +
2976 			nfsrv_lease + NFSRV_LEASEDELTA;
2977 
2978 		    /*
2979 		     * Now, do the associated open.
2980 		     */
2981 		    new_open->ls_stateid.seqid = 1;
2982 		    new_open->ls_stateid.other[0] = clp->lc_clientid.lval[0];
2983 		    new_open->ls_stateid.other[1] = clp->lc_clientid.lval[1];
2984 		    new_open->ls_stateid.other[2] = nfsrv_nextstateindex(clp);
2985 		    new_open->ls_flags = (new_stp->ls_flags&NFSLCK_DENYBITS)|
2986 			NFSLCK_OPEN;
2987 		    if (stp->ls_flags & NFSLCK_DELEGWRITE)
2988 			new_open->ls_flags |= (NFSLCK_READACCESS |
2989 			    NFSLCK_WRITEACCESS);
2990 		    else
2991 			new_open->ls_flags |= NFSLCK_READACCESS;
2992 		    new_open->ls_uid = new_stp->ls_uid;
2993 		    new_open->ls_lfp = lfp;
2994 		    new_open->ls_clp = clp;
2995 		    LIST_INIT(&new_open->ls_open);
2996 		    LIST_INSERT_HEAD(&lfp->lf_open, new_open, ls_file);
2997 		    LIST_INSERT_HEAD(NFSSTATEHASH(clp, new_open->ls_stateid),
2998 			new_open, ls_hash);
2999 		    /*
3000 		     * and handle the open owner
3001 		     */
3002 		    if (ownerstp) {
3003 			new_open->ls_openowner = ownerstp;
3004 			LIST_INSERT_HEAD(&ownerstp->ls_open,new_open,ls_list);
3005 		    } else {
3006 			new_open->ls_openowner = new_stp;
3007 			new_stp->ls_flags = 0;
3008 			nfsrvd_refcache(new_stp->ls_op);
3009 			new_stp->ls_noopens = 0;
3010 			LIST_INIT(&new_stp->ls_open);
3011 			LIST_INSERT_HEAD(&new_stp->ls_open, new_open, ls_list);
3012 			LIST_INSERT_HEAD(&clp->lc_open, new_stp, ls_list);
3013 			*new_stpp = NULL;
3014 			nfsstatsv1.srvopenowners++;
3015 			nfsrv_openpluslock++;
3016 		    }
3017 		    openstp = new_open;
3018 		    new_open = NULL;
3019 		    nfsstatsv1.srvopens++;
3020 		    nfsrv_openpluslock++;
3021 		    break;
3022 		}
3023 	    }
3024 	    if (stp == LIST_END(&clp->lc_olddeleg))
3025 		error = NFSERR_EXPIRED;
3026 	} else if (new_stp->ls_flags & (NFSLCK_DELEGREAD | NFSLCK_DELEGWRITE)) {
3027 	    /*
3028 	     * Scan to see that no delegation for this client and file
3029 	     * doesn't already exist.
3030 	     * There also shouldn't yet be an Open for this file and
3031 	     * openowner.
3032 	     */
3033 	    LIST_FOREACH(stp, &lfp->lf_deleg, ls_file) {
3034 		if (stp->ls_clp == clp)
3035 		    break;
3036 	    }
3037 	    if (stp == LIST_END(&lfp->lf_deleg) && openstp == NULL) {
3038 		/*
3039 		 * This is the Claim_Previous case with a delegation
3040 		 * type != Delegate_None.
3041 		 */
3042 		/*
3043 		 * First, add the delegation. (Although we must issue the
3044 		 * delegation, we can also ask for an immediate return.)
3045 		 */
3046 		new_deleg->ls_stateid.seqid = delegstateidp->seqid = 1;
3047 		new_deleg->ls_stateid.other[0] = delegstateidp->other[0] =
3048 		    clp->lc_clientid.lval[0];
3049 		new_deleg->ls_stateid.other[1] = delegstateidp->other[1] =
3050 		    clp->lc_clientid.lval[1];
3051 		new_deleg->ls_stateid.other[2] = delegstateidp->other[2] =
3052 		    nfsrv_nextstateindex(clp);
3053 		if (new_stp->ls_flags & NFSLCK_DELEGWRITE) {
3054 		    new_deleg->ls_flags = (NFSLCK_DELEGWRITE |
3055 			NFSLCK_READACCESS | NFSLCK_WRITEACCESS);
3056 		    *rflagsp |= NFSV4OPEN_WRITEDELEGATE;
3057 		    nfsrv_writedelegcnt++;
3058 		} else {
3059 		    new_deleg->ls_flags = (NFSLCK_DELEGREAD |
3060 			NFSLCK_READACCESS);
3061 		    *rflagsp |= NFSV4OPEN_READDELEGATE;
3062 		}
3063 		new_deleg->ls_uid = new_stp->ls_uid;
3064 		new_deleg->ls_lfp = lfp;
3065 		new_deleg->ls_clp = clp;
3066 		new_deleg->ls_filerev = filerev;
3067 		new_deleg->ls_compref = nd->nd_compref;
3068 		LIST_INSERT_HEAD(&lfp->lf_deleg, new_deleg, ls_file);
3069 		LIST_INSERT_HEAD(NFSSTATEHASH(clp,
3070 		    new_deleg->ls_stateid), new_deleg, ls_hash);
3071 		LIST_INSERT_HEAD(&clp->lc_deleg, new_deleg, ls_list);
3072 		new_deleg = NULL;
3073 		if (delegate == 2 || nfsrv_issuedelegs == 0 ||
3074 		    (clp->lc_flags & (LCL_CALLBACKSON | LCL_CBDOWN)) !=
3075 		     LCL_CALLBACKSON ||
3076 		    NFSRV_V4DELEGLIMIT(nfsrv_delegatecnt) ||
3077 		    !NFSVNO_DELEGOK(vp))
3078 		    *rflagsp |= NFSV4OPEN_RECALL;
3079 		nfsstatsv1.srvdelegates++;
3080 		nfsrv_openpluslock++;
3081 		nfsrv_delegatecnt++;
3082 
3083 		/*
3084 		 * Now, do the associated open.
3085 		 */
3086 		new_open->ls_stateid.seqid = 1;
3087 		new_open->ls_stateid.other[0] = clp->lc_clientid.lval[0];
3088 		new_open->ls_stateid.other[1] = clp->lc_clientid.lval[1];
3089 		new_open->ls_stateid.other[2] = nfsrv_nextstateindex(clp);
3090 		new_open->ls_flags = (new_stp->ls_flags & NFSLCK_DENYBITS) |
3091 		    NFSLCK_OPEN;
3092 		if (new_stp->ls_flags & NFSLCK_DELEGWRITE)
3093 			new_open->ls_flags |= (NFSLCK_READACCESS |
3094 			    NFSLCK_WRITEACCESS);
3095 		else
3096 			new_open->ls_flags |= NFSLCK_READACCESS;
3097 		new_open->ls_uid = new_stp->ls_uid;
3098 		new_open->ls_lfp = lfp;
3099 		new_open->ls_clp = clp;
3100 		LIST_INIT(&new_open->ls_open);
3101 		LIST_INSERT_HEAD(&lfp->lf_open, new_open, ls_file);
3102 		LIST_INSERT_HEAD(NFSSTATEHASH(clp, new_open->ls_stateid),
3103 		   new_open, ls_hash);
3104 		/*
3105 		 * and handle the open owner
3106 		 */
3107 		if (ownerstp) {
3108 		    new_open->ls_openowner = ownerstp;
3109 		    LIST_INSERT_HEAD(&ownerstp->ls_open, new_open, ls_list);
3110 		} else {
3111 		    new_open->ls_openowner = new_stp;
3112 		    new_stp->ls_flags = 0;
3113 		    nfsrvd_refcache(new_stp->ls_op);
3114 		    new_stp->ls_noopens = 0;
3115 		    LIST_INIT(&new_stp->ls_open);
3116 		    LIST_INSERT_HEAD(&new_stp->ls_open, new_open, ls_list);
3117 		    LIST_INSERT_HEAD(&clp->lc_open, new_stp, ls_list);
3118 		    *new_stpp = NULL;
3119 		    nfsstatsv1.srvopenowners++;
3120 		    nfsrv_openpluslock++;
3121 		}
3122 		openstp = new_open;
3123 		new_open = NULL;
3124 		nfsstatsv1.srvopens++;
3125 		nfsrv_openpluslock++;
3126 	    } else {
3127 		error = NFSERR_RECLAIMCONFLICT;
3128 	    }
3129 	} else if (ownerstp) {
3130 		if (ownerstp->ls_flags & NFSLCK_NEEDSCONFIRM) {
3131 		    /* Replace the open */
3132 		    if (ownerstp->ls_op)
3133 			nfsrvd_derefcache(ownerstp->ls_op);
3134 		    ownerstp->ls_op = new_stp->ls_op;
3135 		    nfsrvd_refcache(ownerstp->ls_op);
3136 		    ownerstp->ls_seq = new_stp->ls_seq;
3137 		    *rflagsp |= NFSV4OPEN_RESULTCONFIRM;
3138 		    stp = LIST_FIRST(&ownerstp->ls_open);
3139 		    stp->ls_flags = (new_stp->ls_flags & NFSLCK_SHAREBITS) |
3140 			NFSLCK_OPEN;
3141 		    stp->ls_stateid.seqid = 1;
3142 		    stp->ls_uid = new_stp->ls_uid;
3143 		    if (lfp != stp->ls_lfp) {
3144 			LIST_REMOVE(stp, ls_file);
3145 			LIST_INSERT_HEAD(&lfp->lf_open, stp, ls_file);
3146 			stp->ls_lfp = lfp;
3147 		    }
3148 		    openstp = stp;
3149 		} else if (openstp) {
3150 		    openstp->ls_flags |= (new_stp->ls_flags & NFSLCK_SHAREBITS);
3151 		    openstp->ls_stateid.seqid++;
3152 		    if ((nd->nd_flag & ND_NFSV41) != 0 &&
3153 			openstp->ls_stateid.seqid == 0)
3154 			openstp->ls_stateid.seqid = 1;
3155 
3156 		    /*
3157 		     * This is where we can choose to issue a delegation.
3158 		     */
3159 		    if ((new_stp->ls_flags & NFSLCK_WANTNODELEG) != 0)
3160 			*rflagsp |= NFSV4OPEN_WDNOTWANTED;
3161 		    else if (nfsrv_issuedelegs == 0)
3162 			*rflagsp |= NFSV4OPEN_WDSUPPFTYPE;
3163 		    else if (NFSRV_V4DELEGLIMIT(nfsrv_delegatecnt))
3164 			*rflagsp |= NFSV4OPEN_WDRESOURCE;
3165 		    else if (delegate == 0 || writedeleg == 0 ||
3166 			NFSVNO_EXRDONLY(exp) || (readonly != 0 &&
3167 			nfsrv_writedelegifpos == 0) ||
3168 			!NFSVNO_DELEGOK(vp) ||
3169 			(new_stp->ls_flags & NFSLCK_WANTRDELEG) != 0 ||
3170 			(clp->lc_flags & (LCL_CALLBACKSON | LCL_CBDOWN)) !=
3171 			 LCL_CALLBACKSON)
3172 			*rflagsp |= NFSV4OPEN_WDCONTENTION;
3173 		    else {
3174 			new_deleg->ls_stateid.seqid = delegstateidp->seqid = 1;
3175 			new_deleg->ls_stateid.other[0] = delegstateidp->other[0]
3176 			    = clp->lc_clientid.lval[0];
3177 			new_deleg->ls_stateid.other[1] = delegstateidp->other[1]
3178 			    = clp->lc_clientid.lval[1];
3179 			new_deleg->ls_stateid.other[2] = delegstateidp->other[2]
3180 			    = nfsrv_nextstateindex(clp);
3181 			new_deleg->ls_flags = (NFSLCK_DELEGWRITE |
3182 			    NFSLCK_READACCESS | NFSLCK_WRITEACCESS);
3183 			*rflagsp |= NFSV4OPEN_WRITEDELEGATE;
3184 			new_deleg->ls_uid = new_stp->ls_uid;
3185 			new_deleg->ls_lfp = lfp;
3186 			new_deleg->ls_clp = clp;
3187 			new_deleg->ls_filerev = filerev;
3188 			new_deleg->ls_compref = nd->nd_compref;
3189 			nfsrv_writedelegcnt++;
3190 			LIST_INSERT_HEAD(&lfp->lf_deleg, new_deleg, ls_file);
3191 			LIST_INSERT_HEAD(NFSSTATEHASH(clp,
3192 			    new_deleg->ls_stateid), new_deleg, ls_hash);
3193 			LIST_INSERT_HEAD(&clp->lc_deleg, new_deleg, ls_list);
3194 			new_deleg = NULL;
3195 			nfsstatsv1.srvdelegates++;
3196 			nfsrv_openpluslock++;
3197 			nfsrv_delegatecnt++;
3198 		    }
3199 		} else {
3200 		    new_open->ls_stateid.seqid = 1;
3201 		    new_open->ls_stateid.other[0] = clp->lc_clientid.lval[0];
3202 		    new_open->ls_stateid.other[1] = clp->lc_clientid.lval[1];
3203 		    new_open->ls_stateid.other[2] = nfsrv_nextstateindex(clp);
3204 		    new_open->ls_flags = (new_stp->ls_flags & NFSLCK_SHAREBITS)|
3205 			NFSLCK_OPEN;
3206 		    new_open->ls_uid = new_stp->ls_uid;
3207 		    new_open->ls_openowner = ownerstp;
3208 		    new_open->ls_lfp = lfp;
3209 		    new_open->ls_clp = clp;
3210 		    LIST_INIT(&new_open->ls_open);
3211 		    LIST_INSERT_HEAD(&lfp->lf_open, new_open, ls_file);
3212 		    LIST_INSERT_HEAD(&ownerstp->ls_open, new_open, ls_list);
3213 		    LIST_INSERT_HEAD(NFSSTATEHASH(clp, new_open->ls_stateid),
3214 			new_open, ls_hash);
3215 		    openstp = new_open;
3216 		    new_open = NULL;
3217 		    nfsstatsv1.srvopens++;
3218 		    nfsrv_openpluslock++;
3219 
3220 		    /*
3221 		     * This is where we can choose to issue a delegation.
3222 		     */
3223 		    if ((new_stp->ls_flags & NFSLCK_WANTNODELEG) != 0)
3224 			*rflagsp |= NFSV4OPEN_WDNOTWANTED;
3225 		    else if (nfsrv_issuedelegs == 0)
3226 			*rflagsp |= NFSV4OPEN_WDSUPPFTYPE;
3227 		    else if (NFSRV_V4DELEGLIMIT(nfsrv_delegatecnt))
3228 			*rflagsp |= NFSV4OPEN_WDRESOURCE;
3229 		    else if (delegate == 0 || (writedeleg == 0 &&
3230 			readonly == 0) || !NFSVNO_DELEGOK(vp) ||
3231 			(clp->lc_flags & (LCL_CALLBACKSON | LCL_CBDOWN)) !=
3232 			 LCL_CALLBACKSON)
3233 			*rflagsp |= NFSV4OPEN_WDCONTENTION;
3234 		    else {
3235 			new_deleg->ls_stateid.seqid = delegstateidp->seqid = 1;
3236 			new_deleg->ls_stateid.other[0] = delegstateidp->other[0]
3237 			    = clp->lc_clientid.lval[0];
3238 			new_deleg->ls_stateid.other[1] = delegstateidp->other[1]
3239 			    = clp->lc_clientid.lval[1];
3240 			new_deleg->ls_stateid.other[2] = delegstateidp->other[2]
3241 			    = nfsrv_nextstateindex(clp);
3242 			if (writedeleg && !NFSVNO_EXRDONLY(exp) &&
3243 			    (nfsrv_writedelegifpos || !readonly) &&
3244 			    (new_stp->ls_flags & NFSLCK_WANTRDELEG) == 0) {
3245 			    new_deleg->ls_flags = (NFSLCK_DELEGWRITE |
3246 				NFSLCK_READACCESS | NFSLCK_WRITEACCESS);
3247 			    *rflagsp |= NFSV4OPEN_WRITEDELEGATE;
3248 			    nfsrv_writedelegcnt++;
3249 			} else {
3250 			    new_deleg->ls_flags = (NFSLCK_DELEGREAD |
3251 				NFSLCK_READACCESS);
3252 			    *rflagsp |= NFSV4OPEN_READDELEGATE;
3253 			}
3254 			new_deleg->ls_uid = new_stp->ls_uid;
3255 			new_deleg->ls_lfp = lfp;
3256 			new_deleg->ls_clp = clp;
3257 			new_deleg->ls_filerev = filerev;
3258 			new_deleg->ls_compref = nd->nd_compref;
3259 			LIST_INSERT_HEAD(&lfp->lf_deleg, new_deleg, ls_file);
3260 			LIST_INSERT_HEAD(NFSSTATEHASH(clp,
3261 			    new_deleg->ls_stateid), new_deleg, ls_hash);
3262 			LIST_INSERT_HEAD(&clp->lc_deleg, new_deleg, ls_list);
3263 			new_deleg = NULL;
3264 			nfsstatsv1.srvdelegates++;
3265 			nfsrv_openpluslock++;
3266 			nfsrv_delegatecnt++;
3267 		    }
3268 		}
3269 	} else {
3270 		/*
3271 		 * New owner case. Start the open_owner sequence with a
3272 		 * Needs confirmation (unless a reclaim) and hang the
3273 		 * new open off it.
3274 		 */
3275 		new_open->ls_stateid.seqid = 1;
3276 		new_open->ls_stateid.other[0] = clp->lc_clientid.lval[0];
3277 		new_open->ls_stateid.other[1] = clp->lc_clientid.lval[1];
3278 		new_open->ls_stateid.other[2] = nfsrv_nextstateindex(clp);
3279 		new_open->ls_flags = (new_stp->ls_flags & NFSLCK_SHAREBITS) |
3280 		    NFSLCK_OPEN;
3281 		new_open->ls_uid = new_stp->ls_uid;
3282 		LIST_INIT(&new_open->ls_open);
3283 		new_open->ls_openowner = new_stp;
3284 		new_open->ls_lfp = lfp;
3285 		new_open->ls_clp = clp;
3286 		LIST_INSERT_HEAD(&lfp->lf_open, new_open, ls_file);
3287 		if (new_stp->ls_flags & NFSLCK_RECLAIM) {
3288 			new_stp->ls_flags = 0;
3289 		} else if ((nd->nd_flag & ND_NFSV41) != 0) {
3290 			/* NFSv4.1 never needs confirmation. */
3291 			new_stp->ls_flags = 0;
3292 
3293 			/*
3294 			 * This is where we can choose to issue a delegation.
3295 			 */
3296 			if (delegate && nfsrv_issuedelegs &&
3297 			    (writedeleg || readonly) &&
3298 			    (clp->lc_flags & (LCL_CALLBACKSON | LCL_CBDOWN)) ==
3299 			     LCL_CALLBACKSON &&
3300 			    !NFSRV_V4DELEGLIMIT(nfsrv_delegatecnt) &&
3301 			    NFSVNO_DELEGOK(vp) &&
3302 			    ((nd->nd_flag & ND_NFSV41) == 0 ||
3303 			     (new_stp->ls_flags & NFSLCK_WANTNODELEG) == 0)) {
3304 				new_deleg->ls_stateid.seqid =
3305 				    delegstateidp->seqid = 1;
3306 				new_deleg->ls_stateid.other[0] =
3307 				    delegstateidp->other[0]
3308 				    = clp->lc_clientid.lval[0];
3309 				new_deleg->ls_stateid.other[1] =
3310 				    delegstateidp->other[1]
3311 				    = clp->lc_clientid.lval[1];
3312 				new_deleg->ls_stateid.other[2] =
3313 				    delegstateidp->other[2]
3314 				    = nfsrv_nextstateindex(clp);
3315 				if (writedeleg && !NFSVNO_EXRDONLY(exp) &&
3316 				    (nfsrv_writedelegifpos || !readonly) &&
3317 				    ((nd->nd_flag & ND_NFSV41) == 0 ||
3318 				     (new_stp->ls_flags & NFSLCK_WANTRDELEG) ==
3319 				     0)) {
3320 					new_deleg->ls_flags =
3321 					    (NFSLCK_DELEGWRITE |
3322 					     NFSLCK_READACCESS |
3323 					     NFSLCK_WRITEACCESS);
3324 					*rflagsp |= NFSV4OPEN_WRITEDELEGATE;
3325 					nfsrv_writedelegcnt++;
3326 				} else {
3327 					new_deleg->ls_flags =
3328 					    (NFSLCK_DELEGREAD |
3329 					     NFSLCK_READACCESS);
3330 					*rflagsp |= NFSV4OPEN_READDELEGATE;
3331 				}
3332 				new_deleg->ls_uid = new_stp->ls_uid;
3333 				new_deleg->ls_lfp = lfp;
3334 				new_deleg->ls_clp = clp;
3335 				new_deleg->ls_filerev = filerev;
3336 				new_deleg->ls_compref = nd->nd_compref;
3337 				LIST_INSERT_HEAD(&lfp->lf_deleg, new_deleg,
3338 				    ls_file);
3339 				LIST_INSERT_HEAD(NFSSTATEHASH(clp,
3340 				    new_deleg->ls_stateid), new_deleg, ls_hash);
3341 				LIST_INSERT_HEAD(&clp->lc_deleg, new_deleg,
3342 				    ls_list);
3343 				new_deleg = NULL;
3344 				nfsstatsv1.srvdelegates++;
3345 				nfsrv_openpluslock++;
3346 				nfsrv_delegatecnt++;
3347 			}
3348 			/*
3349 			 * Since NFSv4.1 never does an OpenConfirm, the first
3350 			 * open state will be acquired here.
3351 			 */
3352 			if (!(clp->lc_flags & LCL_STAMPEDSTABLE)) {
3353 				clp->lc_flags |= LCL_STAMPEDSTABLE;
3354 				len = clp->lc_idlen;
3355 				NFSBCOPY(clp->lc_id, clidp, len);
3356 				gotstate = 1;
3357 			}
3358 		} else {
3359 			*rflagsp |= NFSV4OPEN_RESULTCONFIRM;
3360 			new_stp->ls_flags = NFSLCK_NEEDSCONFIRM;
3361 		}
3362 		nfsrvd_refcache(new_stp->ls_op);
3363 		new_stp->ls_noopens = 0;
3364 		LIST_INIT(&new_stp->ls_open);
3365 		LIST_INSERT_HEAD(&new_stp->ls_open, new_open, ls_list);
3366 		LIST_INSERT_HEAD(&clp->lc_open, new_stp, ls_list);
3367 		LIST_INSERT_HEAD(NFSSTATEHASH(clp, new_open->ls_stateid),
3368 		    new_open, ls_hash);
3369 		openstp = new_open;
3370 		new_open = NULL;
3371 		*new_stpp = NULL;
3372 		nfsstatsv1.srvopens++;
3373 		nfsrv_openpluslock++;
3374 		nfsstatsv1.srvopenowners++;
3375 		nfsrv_openpluslock++;
3376 	}
3377 	if (!error) {
3378 		stateidp->seqid = openstp->ls_stateid.seqid;
3379 		stateidp->other[0] = openstp->ls_stateid.other[0];
3380 		stateidp->other[1] = openstp->ls_stateid.other[1];
3381 		stateidp->other[2] = openstp->ls_stateid.other[2];
3382 	}
3383 	NFSUNLOCKSTATE();
3384 	if (haslock) {
3385 		NFSLOCKV4ROOTMUTEX();
3386 		nfsv4_unlock(&nfsv4rootfs_lock, 1);
3387 		NFSUNLOCKV4ROOTMUTEX();
3388 	}
3389 	if (new_open)
3390 		free(new_open, M_NFSDSTATE);
3391 	if (new_deleg)
3392 		free(new_deleg, M_NFSDSTATE);
3393 
3394 	/*
3395 	 * If the NFSv4.1 client just acquired its first open, write a timestamp
3396 	 * to the stable storage file.
3397 	 */
3398 	if (gotstate != 0) {
3399 		nfsrv_writestable(clidp, len, NFSNST_NEWSTATE, p);
3400 		nfsrv_backupstable();
3401 	}
3402 
3403 out:
3404 	free(clidp, M_TEMP);
3405 	NFSEXITCODE2(error, nd);
3406 	return (error);
3407 }
3408 
3409 /*
3410  * Open update. Does the confirm, downgrade and close.
3411  */
3412 APPLESTATIC int
3413 nfsrv_openupdate(vnode_t vp, struct nfsstate *new_stp, nfsquad_t clientid,
3414     nfsv4stateid_t *stateidp, struct nfsrv_descript *nd, NFSPROC_T *p,
3415     int *retwriteaccessp)
3416 {
3417 	struct nfsstate *stp;
3418 	struct nfsclient *clp;
3419 	struct nfslockfile *lfp;
3420 	u_int32_t bits;
3421 	int error = 0, gotstate = 0, len = 0;
3422 	u_char *clidp = NULL;
3423 
3424 	/*
3425 	 * Check for restart conditions (client and server).
3426 	 */
3427 	error = nfsrv_checkrestart(clientid, new_stp->ls_flags,
3428 	    &new_stp->ls_stateid, 0);
3429 	if (error)
3430 		goto out;
3431 
3432 	clidp = malloc(NFSV4_OPAQUELIMIT, M_TEMP, M_WAITOK);
3433 	NFSLOCKSTATE();
3434 	/*
3435 	 * Get the open structure via clientid and stateid.
3436 	 */
3437 	error = nfsrv_getclient(clientid, CLOPS_RENEW, &clp, NULL,
3438 	    (nfsquad_t)((u_quad_t)0), 0, nd, p);
3439 	if (!error)
3440 		error = nfsrv_getstate(clp, &new_stp->ls_stateid,
3441 		    new_stp->ls_flags, &stp);
3442 
3443 	/*
3444 	 * Sanity check the open.
3445 	 */
3446 	if (!error && (!(stp->ls_flags & NFSLCK_OPEN) ||
3447 		(!(new_stp->ls_flags & NFSLCK_CONFIRM) &&
3448 		 (stp->ls_openowner->ls_flags & NFSLCK_NEEDSCONFIRM)) ||
3449 		((new_stp->ls_flags & NFSLCK_CONFIRM) &&
3450 		 (!(stp->ls_openowner->ls_flags & NFSLCK_NEEDSCONFIRM)))))
3451 		error = NFSERR_BADSTATEID;
3452 
3453 	if (!error)
3454 		error = nfsrv_checkseqid(nd, new_stp->ls_seq,
3455 		    stp->ls_openowner, new_stp->ls_op);
3456 	if (!error && stp->ls_stateid.seqid != new_stp->ls_stateid.seqid &&
3457 	    (((nd->nd_flag & ND_NFSV41) == 0 &&
3458 	      !(new_stp->ls_flags & NFSLCK_CONFIRM)) ||
3459 	     ((nd->nd_flag & ND_NFSV41) != 0 &&
3460 	      new_stp->ls_stateid.seqid != 0)))
3461 		error = NFSERR_OLDSTATEID;
3462 	if (!error && vnode_vtype(vp) != VREG) {
3463 		if (vnode_vtype(vp) == VDIR)
3464 			error = NFSERR_ISDIR;
3465 		else
3466 			error = NFSERR_INVAL;
3467 	}
3468 
3469 	if (error) {
3470 		/*
3471 		 * If a client tries to confirm an Open with a bad
3472 		 * seqid# and there are no byte range locks or other Opens
3473 		 * on the openowner, just throw it away, so the next use of the
3474 		 * openowner will start a fresh seq#.
3475 		 */
3476 		if (error == NFSERR_BADSEQID &&
3477 		    (new_stp->ls_flags & NFSLCK_CONFIRM) &&
3478 		    nfsrv_nootherstate(stp))
3479 			nfsrv_freeopenowner(stp->ls_openowner, 0, p);
3480 		NFSUNLOCKSTATE();
3481 		goto out;
3482 	}
3483 
3484 	/*
3485 	 * Set the return stateid.
3486 	 */
3487 	stateidp->seqid = stp->ls_stateid.seqid + 1;
3488 	if ((nd->nd_flag & ND_NFSV41) != 0 && stateidp->seqid == 0)
3489 		stateidp->seqid = 1;
3490 	stateidp->other[0] = stp->ls_stateid.other[0];
3491 	stateidp->other[1] = stp->ls_stateid.other[1];
3492 	stateidp->other[2] = stp->ls_stateid.other[2];
3493 	/*
3494 	 * Now, handle the three cases.
3495 	 */
3496 	if (new_stp->ls_flags & NFSLCK_CONFIRM) {
3497 		/*
3498 		 * If the open doesn't need confirmation, it seems to me that
3499 		 * there is a client error, but I'll just log it and keep going?
3500 		 */
3501 		if (!(stp->ls_openowner->ls_flags & NFSLCK_NEEDSCONFIRM))
3502 			printf("Nfsv4d: stray open confirm\n");
3503 		stp->ls_openowner->ls_flags = 0;
3504 		stp->ls_stateid.seqid++;
3505 		if ((nd->nd_flag & ND_NFSV41) != 0 &&
3506 		    stp->ls_stateid.seqid == 0)
3507 			stp->ls_stateid.seqid = 1;
3508 		if (!(clp->lc_flags & LCL_STAMPEDSTABLE)) {
3509 			clp->lc_flags |= LCL_STAMPEDSTABLE;
3510 			len = clp->lc_idlen;
3511 			NFSBCOPY(clp->lc_id, clidp, len);
3512 			gotstate = 1;
3513 		}
3514 		NFSUNLOCKSTATE();
3515 	} else if (new_stp->ls_flags & NFSLCK_CLOSE) {
3516 		lfp = stp->ls_lfp;
3517 		if (retwriteaccessp != NULL) {
3518 			if ((stp->ls_flags & NFSLCK_WRITEACCESS) != 0)
3519 				*retwriteaccessp = 1;
3520 			else
3521 				*retwriteaccessp = 0;
3522 		}
3523 		if (nfsrv_dolocallocks != 0 && !LIST_EMPTY(&stp->ls_open)) {
3524 			/* Get the lf lock */
3525 			nfsrv_locklf(lfp);
3526 			NFSUNLOCKSTATE();
3527 			ASSERT_VOP_ELOCKED(vp, "nfsrv_openupdate");
3528 			NFSVOPUNLOCK(vp);
3529 			if (nfsrv_freeopen(stp, vp, 1, p) == 0) {
3530 				NFSLOCKSTATE();
3531 				nfsrv_unlocklf(lfp);
3532 				NFSUNLOCKSTATE();
3533 			}
3534 			NFSVOPLOCK(vp, LK_EXCLUSIVE | LK_RETRY);
3535 		} else {
3536 			(void) nfsrv_freeopen(stp, NULL, 0, p);
3537 			NFSUNLOCKSTATE();
3538 		}
3539 	} else {
3540 		/*
3541 		 * Update the share bits, making sure that the new set are a
3542 		 * subset of the old ones.
3543 		 */
3544 		bits = (new_stp->ls_flags & NFSLCK_SHAREBITS);
3545 		if (~(stp->ls_flags) & bits) {
3546 			NFSUNLOCKSTATE();
3547 			error = NFSERR_INVAL;
3548 			goto out;
3549 		}
3550 		stp->ls_flags = (bits | NFSLCK_OPEN);
3551 		stp->ls_stateid.seqid++;
3552 		if ((nd->nd_flag & ND_NFSV41) != 0 &&
3553 		    stp->ls_stateid.seqid == 0)
3554 			stp->ls_stateid.seqid = 1;
3555 		NFSUNLOCKSTATE();
3556 	}
3557 
3558 	/*
3559 	 * If the client just confirmed its first open, write a timestamp
3560 	 * to the stable storage file.
3561 	 */
3562 	if (gotstate != 0) {
3563 		nfsrv_writestable(clidp, len, NFSNST_NEWSTATE, p);
3564 		nfsrv_backupstable();
3565 	}
3566 
3567 out:
3568 	free(clidp, M_TEMP);
3569 	NFSEXITCODE2(error, nd);
3570 	return (error);
3571 }
3572 
3573 /*
3574  * Delegation update. Does the purge and return.
3575  */
3576 APPLESTATIC int
3577 nfsrv_delegupdate(struct nfsrv_descript *nd, nfsquad_t clientid,
3578     nfsv4stateid_t *stateidp, vnode_t vp, int op, struct ucred *cred,
3579     NFSPROC_T *p, int *retwriteaccessp)
3580 {
3581 	struct nfsstate *stp;
3582 	struct nfsclient *clp;
3583 	int error = 0;
3584 	fhandle_t fh;
3585 
3586 	/*
3587 	 * Do a sanity check against the file handle for DelegReturn.
3588 	 */
3589 	if (vp) {
3590 		error = nfsvno_getfh(vp, &fh, p);
3591 		if (error)
3592 			goto out;
3593 	}
3594 	/*
3595 	 * Check for restart conditions (client and server).
3596 	 */
3597 	if (op == NFSV4OP_DELEGRETURN)
3598 		error = nfsrv_checkrestart(clientid, NFSLCK_DELEGRETURN,
3599 			stateidp, 0);
3600 	else
3601 		error = nfsrv_checkrestart(clientid, NFSLCK_DELEGPURGE,
3602 			stateidp, 0);
3603 
3604 	NFSLOCKSTATE();
3605 	/*
3606 	 * Get the open structure via clientid and stateid.
3607 	 */
3608 	if (!error)
3609 	    error = nfsrv_getclient(clientid, CLOPS_RENEW, &clp, NULL,
3610 		(nfsquad_t)((u_quad_t)0), 0, nd, p);
3611 	if (error) {
3612 		if (error == NFSERR_CBPATHDOWN)
3613 			error = 0;
3614 		if (error == NFSERR_STALECLIENTID && op == NFSV4OP_DELEGRETURN)
3615 			error = NFSERR_STALESTATEID;
3616 	}
3617 	if (!error && op == NFSV4OP_DELEGRETURN) {
3618 	    error = nfsrv_getstate(clp, stateidp, NFSLCK_DELEGRETURN, &stp);
3619 	    if (!error && stp->ls_stateid.seqid != stateidp->seqid &&
3620 		((nd->nd_flag & ND_NFSV41) == 0 || stateidp->seqid != 0))
3621 		error = NFSERR_OLDSTATEID;
3622 	}
3623 	/*
3624 	 * NFSERR_EXPIRED means that the state has gone away,
3625 	 * so Delegations have been purged. Just return ok.
3626 	 */
3627 	if (error == NFSERR_EXPIRED && op == NFSV4OP_DELEGPURGE) {
3628 		NFSUNLOCKSTATE();
3629 		error = 0;
3630 		goto out;
3631 	}
3632 	if (error) {
3633 		NFSUNLOCKSTATE();
3634 		goto out;
3635 	}
3636 
3637 	if (op == NFSV4OP_DELEGRETURN) {
3638 		if (NFSBCMP((caddr_t)&fh, (caddr_t)&stp->ls_lfp->lf_fh,
3639 		    sizeof (fhandle_t))) {
3640 			NFSUNLOCKSTATE();
3641 			error = NFSERR_BADSTATEID;
3642 			goto out;
3643 		}
3644 		if (retwriteaccessp != NULL) {
3645 			if ((stp->ls_flags & NFSLCK_DELEGWRITE) != 0)
3646 				*retwriteaccessp = 1;
3647 			else
3648 				*retwriteaccessp = 0;
3649 		}
3650 		nfsrv_freedeleg(stp);
3651 	} else {
3652 		nfsrv_freedeleglist(&clp->lc_olddeleg);
3653 	}
3654 	NFSUNLOCKSTATE();
3655 	error = 0;
3656 
3657 out:
3658 	NFSEXITCODE(error);
3659 	return (error);
3660 }
3661 
3662 /*
3663  * Release lock owner.
3664  */
3665 APPLESTATIC int
3666 nfsrv_releaselckown(struct nfsstate *new_stp, nfsquad_t clientid,
3667     NFSPROC_T *p)
3668 {
3669 	struct nfsstate *stp, *nstp, *openstp, *ownstp;
3670 	struct nfsclient *clp;
3671 	int error = 0;
3672 
3673 	/*
3674 	 * Check for restart conditions (client and server).
3675 	 */
3676 	error = nfsrv_checkrestart(clientid, new_stp->ls_flags,
3677 	    &new_stp->ls_stateid, 0);
3678 	if (error)
3679 		goto out;
3680 
3681 	NFSLOCKSTATE();
3682 	/*
3683 	 * Get the lock owner by name.
3684 	 */
3685 	error = nfsrv_getclient(clientid, CLOPS_RENEW, &clp, NULL,
3686 	    (nfsquad_t)((u_quad_t)0), 0, NULL, p);
3687 	if (error) {
3688 		NFSUNLOCKSTATE();
3689 		goto out;
3690 	}
3691 	LIST_FOREACH(ownstp, &clp->lc_open, ls_list) {
3692 	    LIST_FOREACH(openstp, &ownstp->ls_open, ls_list) {
3693 		stp = LIST_FIRST(&openstp->ls_open);
3694 		while (stp != LIST_END(&openstp->ls_open)) {
3695 		    nstp = LIST_NEXT(stp, ls_list);
3696 		    /*
3697 		     * If the owner matches, check for locks and
3698 		     * then free or return an error.
3699 		     */
3700 		    if (stp->ls_ownerlen == new_stp->ls_ownerlen &&
3701 			!NFSBCMP(stp->ls_owner, new_stp->ls_owner,
3702 			 stp->ls_ownerlen)){
3703 			if (LIST_EMPTY(&stp->ls_lock)) {
3704 			    nfsrv_freelockowner(stp, NULL, 0, p);
3705 			} else {
3706 			    NFSUNLOCKSTATE();
3707 			    error = NFSERR_LOCKSHELD;
3708 			    goto out;
3709 			}
3710 		    }
3711 		    stp = nstp;
3712 		}
3713 	    }
3714 	}
3715 	NFSUNLOCKSTATE();
3716 
3717 out:
3718 	NFSEXITCODE(error);
3719 	return (error);
3720 }
3721 
3722 /*
3723  * Get the file handle for a lock structure.
3724  */
3725 static int
3726 nfsrv_getlockfh(vnode_t vp, u_short flags, struct nfslockfile *new_lfp,
3727     fhandle_t *nfhp, NFSPROC_T *p)
3728 {
3729 	fhandle_t *fhp = NULL;
3730 	int error;
3731 
3732 	/*
3733 	 * For lock, use the new nfslock structure, otherwise just
3734 	 * a fhandle_t on the stack.
3735 	 */
3736 	if (flags & NFSLCK_OPEN) {
3737 		KASSERT(new_lfp != NULL, ("nfsrv_getlockfh: new_lfp NULL"));
3738 		fhp = &new_lfp->lf_fh;
3739 	} else if (nfhp) {
3740 		fhp = nfhp;
3741 	} else {
3742 		panic("nfsrv_getlockfh");
3743 	}
3744 	error = nfsvno_getfh(vp, fhp, p);
3745 	NFSEXITCODE(error);
3746 	return (error);
3747 }
3748 
3749 /*
3750  * Get an nfs lock structure. Allocate one, as required, and return a
3751  * pointer to it.
3752  * Returns an NFSERR_xxx upon failure or -1 to indicate no current lock.
3753  */
3754 static int
3755 nfsrv_getlockfile(u_short flags, struct nfslockfile **new_lfpp,
3756     struct nfslockfile **lfpp, fhandle_t *nfhp, int lockit)
3757 {
3758 	struct nfslockfile *lfp;
3759 	fhandle_t *fhp = NULL, *tfhp;
3760 	struct nfslockhashhead *hp;
3761 	struct nfslockfile *new_lfp = NULL;
3762 
3763 	/*
3764 	 * For lock, use the new nfslock structure, otherwise just
3765 	 * a fhandle_t on the stack.
3766 	 */
3767 	if (flags & NFSLCK_OPEN) {
3768 		new_lfp = *new_lfpp;
3769 		fhp = &new_lfp->lf_fh;
3770 	} else if (nfhp) {
3771 		fhp = nfhp;
3772 	} else {
3773 		panic("nfsrv_getlockfile");
3774 	}
3775 
3776 	hp = NFSLOCKHASH(fhp);
3777 	LIST_FOREACH(lfp, hp, lf_hash) {
3778 		tfhp = &lfp->lf_fh;
3779 		if (NFSVNO_CMPFH(fhp, tfhp)) {
3780 			if (lockit)
3781 				nfsrv_locklf(lfp);
3782 			*lfpp = lfp;
3783 			return (0);
3784 		}
3785 	}
3786 	if (!(flags & NFSLCK_OPEN))
3787 		return (-1);
3788 
3789 	/*
3790 	 * No match, so chain the new one into the list.
3791 	 */
3792 	LIST_INIT(&new_lfp->lf_open);
3793 	LIST_INIT(&new_lfp->lf_lock);
3794 	LIST_INIT(&new_lfp->lf_deleg);
3795 	LIST_INIT(&new_lfp->lf_locallock);
3796 	LIST_INIT(&new_lfp->lf_rollback);
3797 	new_lfp->lf_locallock_lck.nfslock_usecnt = 0;
3798 	new_lfp->lf_locallock_lck.nfslock_lock = 0;
3799 	new_lfp->lf_usecount = 0;
3800 	LIST_INSERT_HEAD(hp, new_lfp, lf_hash);
3801 	*lfpp = new_lfp;
3802 	*new_lfpp = NULL;
3803 	return (0);
3804 }
3805 
3806 /*
3807  * This function adds a nfslock lock structure to the list for the associated
3808  * nfsstate and nfslockfile structures. It will be inserted after the
3809  * entry pointed at by insert_lop.
3810  */
3811 static void
3812 nfsrv_insertlock(struct nfslock *new_lop, struct nfslock *insert_lop,
3813     struct nfsstate *stp, struct nfslockfile *lfp)
3814 {
3815 	struct nfslock *lop, *nlop;
3816 
3817 	new_lop->lo_stp = stp;
3818 	new_lop->lo_lfp = lfp;
3819 
3820 	if (stp != NULL) {
3821 		/* Insert in increasing lo_first order */
3822 		lop = LIST_FIRST(&lfp->lf_lock);
3823 		if (lop == LIST_END(&lfp->lf_lock) ||
3824 		    new_lop->lo_first <= lop->lo_first) {
3825 			LIST_INSERT_HEAD(&lfp->lf_lock, new_lop, lo_lckfile);
3826 		} else {
3827 			nlop = LIST_NEXT(lop, lo_lckfile);
3828 			while (nlop != LIST_END(&lfp->lf_lock) &&
3829 			       nlop->lo_first < new_lop->lo_first) {
3830 				lop = nlop;
3831 				nlop = LIST_NEXT(lop, lo_lckfile);
3832 			}
3833 			LIST_INSERT_AFTER(lop, new_lop, lo_lckfile);
3834 		}
3835 	} else {
3836 		new_lop->lo_lckfile.le_prev = NULL;	/* list not used */
3837 	}
3838 
3839 	/*
3840 	 * Insert after insert_lop, which is overloaded as stp or lfp for
3841 	 * an empty list.
3842 	 */
3843 	if (stp == NULL && (struct nfslockfile *)insert_lop == lfp)
3844 		LIST_INSERT_HEAD(&lfp->lf_locallock, new_lop, lo_lckowner);
3845 	else if ((struct nfsstate *)insert_lop == stp)
3846 		LIST_INSERT_HEAD(&stp->ls_lock, new_lop, lo_lckowner);
3847 	else
3848 		LIST_INSERT_AFTER(insert_lop, new_lop, lo_lckowner);
3849 	if (stp != NULL) {
3850 		nfsstatsv1.srvlocks++;
3851 		nfsrv_openpluslock++;
3852 	}
3853 }
3854 
3855 /*
3856  * This function updates the locking for a lock owner and given file. It
3857  * maintains a list of lock ranges ordered on increasing file offset that
3858  * are NFSLCK_READ or NFSLCK_WRITE and non-overlapping (aka POSIX style).
3859  * It always adds new_lop to the list and sometimes uses the one pointed
3860  * at by other_lopp.
3861  */
3862 static void
3863 nfsrv_updatelock(struct nfsstate *stp, struct nfslock **new_lopp,
3864     struct nfslock **other_lopp, struct nfslockfile *lfp)
3865 {
3866 	struct nfslock *new_lop = *new_lopp;
3867 	struct nfslock *lop, *tlop, *ilop;
3868 	struct nfslock *other_lop = *other_lopp;
3869 	int unlock = 0, myfile = 0;
3870 	u_int64_t tmp;
3871 
3872 	/*
3873 	 * Work down the list until the lock is merged.
3874 	 */
3875 	if (new_lop->lo_flags & NFSLCK_UNLOCK)
3876 		unlock = 1;
3877 	if (stp != NULL) {
3878 		ilop = (struct nfslock *)stp;
3879 		lop = LIST_FIRST(&stp->ls_lock);
3880 	} else {
3881 		ilop = (struct nfslock *)lfp;
3882 		lop = LIST_FIRST(&lfp->lf_locallock);
3883 	}
3884 	while (lop != NULL) {
3885 	    /*
3886 	     * Only check locks for this file that aren't before the start of
3887 	     * new lock's range.
3888 	     */
3889 	    if (lop->lo_lfp == lfp) {
3890 	      myfile = 1;
3891 	      if (lop->lo_end >= new_lop->lo_first) {
3892 		if (new_lop->lo_end < lop->lo_first) {
3893 			/*
3894 			 * If the new lock ends before the start of the
3895 			 * current lock's range, no merge, just insert
3896 			 * the new lock.
3897 			 */
3898 			break;
3899 		}
3900 		if (new_lop->lo_flags == lop->lo_flags ||
3901 		    (new_lop->lo_first <= lop->lo_first &&
3902 		     new_lop->lo_end >= lop->lo_end)) {
3903 			/*
3904 			 * This lock can be absorbed by the new lock/unlock.
3905 			 * This happens when it covers the entire range
3906 			 * of the old lock or is contiguous
3907 			 * with the old lock and is of the same type or an
3908 			 * unlock.
3909 			 */
3910 			if (lop->lo_first < new_lop->lo_first)
3911 				new_lop->lo_first = lop->lo_first;
3912 			if (lop->lo_end > new_lop->lo_end)
3913 				new_lop->lo_end = lop->lo_end;
3914 			tlop = lop;
3915 			lop = LIST_NEXT(lop, lo_lckowner);
3916 			nfsrv_freenfslock(tlop);
3917 			continue;
3918 		}
3919 
3920 		/*
3921 		 * All these cases are for contiguous locks that are not the
3922 		 * same type, so they can't be merged.
3923 		 */
3924 		if (new_lop->lo_first <= lop->lo_first) {
3925 			/*
3926 			 * This case is where the new lock overlaps with the
3927 			 * first part of the old lock. Move the start of the
3928 			 * old lock to just past the end of the new lock. The
3929 			 * new lock will be inserted in front of the old, since
3930 			 * ilop hasn't been updated. (We are done now.)
3931 			 */
3932 			lop->lo_first = new_lop->lo_end;
3933 			break;
3934 		}
3935 		if (new_lop->lo_end >= lop->lo_end) {
3936 			/*
3937 			 * This case is where the new lock overlaps with the
3938 			 * end of the old lock's range. Move the old lock's
3939 			 * end to just before the new lock's first and insert
3940 			 * the new lock after the old lock.
3941 			 * Might not be done yet, since the new lock could
3942 			 * overlap further locks with higher ranges.
3943 			 */
3944 			lop->lo_end = new_lop->lo_first;
3945 			ilop = lop;
3946 			lop = LIST_NEXT(lop, lo_lckowner);
3947 			continue;
3948 		}
3949 		/*
3950 		 * The final case is where the new lock's range is in the
3951 		 * middle of the current lock's and splits the current lock
3952 		 * up. Use *other_lopp to handle the second part of the
3953 		 * split old lock range. (We are done now.)
3954 		 * For unlock, we use new_lop as other_lop and tmp, since
3955 		 * other_lop and new_lop are the same for this case.
3956 		 * We noted the unlock case above, so we don't need
3957 		 * new_lop->lo_flags any longer.
3958 		 */
3959 		tmp = new_lop->lo_first;
3960 		if (other_lop == NULL) {
3961 			if (!unlock)
3962 				panic("nfsd srv update unlock");
3963 			other_lop = new_lop;
3964 			*new_lopp = NULL;
3965 		}
3966 		other_lop->lo_first = new_lop->lo_end;
3967 		other_lop->lo_end = lop->lo_end;
3968 		other_lop->lo_flags = lop->lo_flags;
3969 		other_lop->lo_stp = stp;
3970 		other_lop->lo_lfp = lfp;
3971 		lop->lo_end = tmp;
3972 		nfsrv_insertlock(other_lop, lop, stp, lfp);
3973 		*other_lopp = NULL;
3974 		ilop = lop;
3975 		break;
3976 	      }
3977 	    }
3978 	    ilop = lop;
3979 	    lop = LIST_NEXT(lop, lo_lckowner);
3980 	    if (myfile && (lop == NULL || lop->lo_lfp != lfp))
3981 		break;
3982 	}
3983 
3984 	/*
3985 	 * Insert the new lock in the list at the appropriate place.
3986 	 */
3987 	if (!unlock) {
3988 		nfsrv_insertlock(new_lop, ilop, stp, lfp);
3989 		*new_lopp = NULL;
3990 	}
3991 }
3992 
3993 /*
3994  * This function handles sequencing of locks, etc.
3995  * It returns an error that indicates what the caller should do.
3996  */
3997 static int
3998 nfsrv_checkseqid(struct nfsrv_descript *nd, u_int32_t seqid,
3999     struct nfsstate *stp, struct nfsrvcache *op)
4000 {
4001 	int error = 0;
4002 
4003 	if ((nd->nd_flag & ND_NFSV41) != 0)
4004 		/* NFSv4.1 ignores the open_seqid and lock_seqid. */
4005 		goto out;
4006 	if (op != nd->nd_rp)
4007 		panic("nfsrvstate checkseqid");
4008 	if (!(op->rc_flag & RC_INPROG))
4009 		panic("nfsrvstate not inprog");
4010 	if (stp->ls_op && stp->ls_op->rc_refcnt <= 0) {
4011 		printf("refcnt=%d\n", stp->ls_op->rc_refcnt);
4012 		panic("nfsrvstate op refcnt");
4013 	}
4014 	if ((stp->ls_seq + 1) == seqid) {
4015 		if (stp->ls_op)
4016 			nfsrvd_derefcache(stp->ls_op);
4017 		stp->ls_op = op;
4018 		nfsrvd_refcache(op);
4019 		stp->ls_seq = seqid;
4020 		goto out;
4021 	} else if (stp->ls_seq == seqid && stp->ls_op &&
4022 		op->rc_xid == stp->ls_op->rc_xid &&
4023 		op->rc_refcnt == 0 &&
4024 		op->rc_reqlen == stp->ls_op->rc_reqlen &&
4025 		op->rc_cksum == stp->ls_op->rc_cksum) {
4026 		if (stp->ls_op->rc_flag & RC_INPROG) {
4027 			error = NFSERR_DONTREPLY;
4028 			goto out;
4029 		}
4030 		nd->nd_rp = stp->ls_op;
4031 		nd->nd_rp->rc_flag |= RC_INPROG;
4032 		nfsrvd_delcache(op);
4033 		error = NFSERR_REPLYFROMCACHE;
4034 		goto out;
4035 	}
4036 	error = NFSERR_BADSEQID;
4037 
4038 out:
4039 	NFSEXITCODE2(error, nd);
4040 	return (error);
4041 }
4042 
4043 /*
4044  * Get the client ip address for callbacks. If the strings can't be parsed,
4045  * just set lc_program to 0 to indicate no callbacks are possible.
4046  * (For cases where the address can't be parsed or is 0.0.0.0.0.0, set
4047  *  the address to the client's transport address. This won't be used
4048  *  for callbacks, but can be printed out by nfsstats for info.)
4049  * Return error if the xdr can't be parsed, 0 otherwise.
4050  */
4051 APPLESTATIC int
4052 nfsrv_getclientipaddr(struct nfsrv_descript *nd, struct nfsclient *clp)
4053 {
4054 	u_int32_t *tl;
4055 	u_char *cp, *cp2;
4056 	int i, j, maxalen = 0, minalen = 0;
4057 	sa_family_t af;
4058 #ifdef INET
4059 	struct sockaddr_in *rin = NULL, *sin;
4060 #endif
4061 #ifdef INET6
4062 	struct sockaddr_in6 *rin6 = NULL, *sin6;
4063 #endif
4064 	u_char *addr;
4065 	int error = 0, cantparse = 0;
4066 	union {
4067 		in_addr_t ival;
4068 		u_char cval[4];
4069 	} ip;
4070 	union {
4071 		in_port_t sval;
4072 		u_char cval[2];
4073 	} port;
4074 
4075 	/* 8 is the maximum length of the port# string. */
4076 	addr = malloc(INET6_ADDRSTRLEN + 8, M_TEMP, M_WAITOK);
4077 	clp->lc_req.nr_client = NULL;
4078 	clp->lc_req.nr_lock = 0;
4079 	af = AF_UNSPEC;
4080 	NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
4081 	i = fxdr_unsigned(int, *tl);
4082 	if (i >= 3 && i <= 4) {
4083 		error = nfsrv_mtostr(nd, addr, i);
4084 		if (error)
4085 			goto nfsmout;
4086 #ifdef INET
4087 		if (!strcmp(addr, "tcp")) {
4088 			clp->lc_flags |= LCL_TCPCALLBACK;
4089 			clp->lc_req.nr_sotype = SOCK_STREAM;
4090 			clp->lc_req.nr_soproto = IPPROTO_TCP;
4091 			af = AF_INET;
4092 		} else if (!strcmp(addr, "udp")) {
4093 			clp->lc_req.nr_sotype = SOCK_DGRAM;
4094 			clp->lc_req.nr_soproto = IPPROTO_UDP;
4095 			af = AF_INET;
4096 		}
4097 #endif
4098 #ifdef INET6
4099 		if (af == AF_UNSPEC) {
4100 			if (!strcmp(addr, "tcp6")) {
4101 				clp->lc_flags |= LCL_TCPCALLBACK;
4102 				clp->lc_req.nr_sotype = SOCK_STREAM;
4103 				clp->lc_req.nr_soproto = IPPROTO_TCP;
4104 				af = AF_INET6;
4105 			} else if (!strcmp(addr, "udp6")) {
4106 				clp->lc_req.nr_sotype = SOCK_DGRAM;
4107 				clp->lc_req.nr_soproto = IPPROTO_UDP;
4108 				af = AF_INET6;
4109 			}
4110 		}
4111 #endif
4112 		if (af == AF_UNSPEC) {
4113 			cantparse = 1;
4114 		}
4115 	} else {
4116 		cantparse = 1;
4117 		if (i > 0) {
4118 			error = nfsm_advance(nd, NFSM_RNDUP(i), -1);
4119 			if (error)
4120 				goto nfsmout;
4121 		}
4122 	}
4123 	/*
4124 	 * The caller has allocated clp->lc_req.nr_nam to be large enough
4125 	 * for either AF_INET or AF_INET6 and zeroed out the contents.
4126 	 * maxalen is set to the maximum length of the host IP address string
4127 	 * plus 8 for the maximum length of the port#.
4128 	 * minalen is set to the minimum length of the host IP address string
4129 	 * plus 4 for the minimum length of the port#.
4130 	 * These lengths do not include NULL termination,
4131 	 * so INET[6]_ADDRSTRLEN - 1 is used in the calculations.
4132 	 */
4133 	switch (af) {
4134 #ifdef INET
4135 	case AF_INET:
4136 		rin = (struct sockaddr_in *)clp->lc_req.nr_nam;
4137 		rin->sin_family = AF_INET;
4138 		rin->sin_len = sizeof(struct sockaddr_in);
4139 		maxalen = INET_ADDRSTRLEN - 1 + 8;
4140 		minalen = 7 + 4;
4141 		break;
4142 #endif
4143 #ifdef INET6
4144 	case AF_INET6:
4145 		rin6 = (struct sockaddr_in6 *)clp->lc_req.nr_nam;
4146 		rin6->sin6_family = AF_INET6;
4147 		rin6->sin6_len = sizeof(struct sockaddr_in6);
4148 		maxalen = INET6_ADDRSTRLEN - 1 + 8;
4149 		minalen = 3 + 4;
4150 		break;
4151 #endif
4152 	}
4153 	NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
4154 	i = fxdr_unsigned(int, *tl);
4155 	if (i < 0) {
4156 		error = NFSERR_BADXDR;
4157 		goto nfsmout;
4158 	} else if (i == 0) {
4159 		cantparse = 1;
4160 	} else if (!cantparse && i <= maxalen && i >= minalen) {
4161 		error = nfsrv_mtostr(nd, addr, i);
4162 		if (error)
4163 			goto nfsmout;
4164 
4165 		/*
4166 		 * Parse out the address fields. We expect 6 decimal numbers
4167 		 * separated by '.'s for AF_INET and two decimal numbers
4168 		 * preceeded by '.'s for AF_INET6.
4169 		 */
4170 		cp = NULL;
4171 		switch (af) {
4172 #ifdef INET6
4173 		/*
4174 		 * For AF_INET6, first parse the host address.
4175 		 */
4176 		case AF_INET6:
4177 			cp = strchr(addr, '.');
4178 			if (cp != NULL) {
4179 				*cp++ = '\0';
4180 				if (inet_pton(af, addr, &rin6->sin6_addr) == 1)
4181 					i = 4;
4182 				else {
4183 					cp = NULL;
4184 					cantparse = 1;
4185 				}
4186 			}
4187 			break;
4188 #endif
4189 #ifdef INET
4190 		case AF_INET:
4191 			cp = addr;
4192 			i = 0;
4193 			break;
4194 #endif
4195 		}
4196 		while (cp != NULL && *cp && i < 6) {
4197 			cp2 = cp;
4198 			while (*cp2 && *cp2 != '.')
4199 				cp2++;
4200 			if (*cp2)
4201 				*cp2++ = '\0';
4202 			else if (i != 5) {
4203 				cantparse = 1;
4204 				break;
4205 			}
4206 			j = nfsrv_getipnumber(cp);
4207 			if (j >= 0) {
4208 				if (i < 4)
4209 					ip.cval[3 - i] = j;
4210 				else
4211 					port.cval[5 - i] = j;
4212 			} else {
4213 				cantparse = 1;
4214 				break;
4215 			}
4216 			cp = cp2;
4217 			i++;
4218 		}
4219 		if (!cantparse) {
4220 			/*
4221 			 * The host address INADDR_ANY is (mis)used to indicate
4222 			 * "there is no valid callback address".
4223 			 */
4224 			switch (af) {
4225 #ifdef INET6
4226 			case AF_INET6:
4227 				if (!IN6_ARE_ADDR_EQUAL(&rin6->sin6_addr,
4228 				    &in6addr_any))
4229 					rin6->sin6_port = htons(port.sval);
4230 				else
4231 					cantparse = 1;
4232 				break;
4233 #endif
4234 #ifdef INET
4235 			case AF_INET:
4236 				if (ip.ival != INADDR_ANY) {
4237 					rin->sin_addr.s_addr = htonl(ip.ival);
4238 					rin->sin_port = htons(port.sval);
4239 				} else {
4240 					cantparse = 1;
4241 				}
4242 				break;
4243 #endif
4244 			}
4245 		}
4246 	} else {
4247 		cantparse = 1;
4248 		if (i > 0) {
4249 			error = nfsm_advance(nd, NFSM_RNDUP(i), -1);
4250 			if (error)
4251 				goto nfsmout;
4252 		}
4253 	}
4254 	if (cantparse) {
4255 		switch (nd->nd_nam->sa_family) {
4256 #ifdef INET
4257 		case AF_INET:
4258 			sin = (struct sockaddr_in *)nd->nd_nam;
4259 			rin = (struct sockaddr_in *)clp->lc_req.nr_nam;
4260 			rin->sin_family = AF_INET;
4261 			rin->sin_len = sizeof(struct sockaddr_in);
4262 			rin->sin_addr.s_addr = sin->sin_addr.s_addr;
4263 			rin->sin_port = 0x0;
4264 			break;
4265 #endif
4266 #ifdef INET6
4267 		case AF_INET6:
4268 			sin6 = (struct sockaddr_in6 *)nd->nd_nam;
4269 			rin6 = (struct sockaddr_in6 *)clp->lc_req.nr_nam;
4270 			rin6->sin6_family = AF_INET6;
4271 			rin6->sin6_len = sizeof(struct sockaddr_in6);
4272 			rin6->sin6_addr = sin6->sin6_addr;
4273 			rin6->sin6_port = 0x0;
4274 			break;
4275 #endif
4276 		}
4277 		clp->lc_program = 0;
4278 	}
4279 nfsmout:
4280 	free(addr, M_TEMP);
4281 	NFSEXITCODE2(error, nd);
4282 	return (error);
4283 }
4284 
4285 /*
4286  * Turn a string of up to three decimal digits into a number. Return -1 upon
4287  * error.
4288  */
4289 static int
4290 nfsrv_getipnumber(u_char *cp)
4291 {
4292 	int i = 0, j = 0;
4293 
4294 	while (*cp) {
4295 		if (j > 2 || *cp < '0' || *cp > '9')
4296 			return (-1);
4297 		i *= 10;
4298 		i += (*cp - '0');
4299 		cp++;
4300 		j++;
4301 	}
4302 	if (i < 256)
4303 		return (i);
4304 	return (-1);
4305 }
4306 
4307 /*
4308  * This function checks for restart conditions.
4309  */
4310 static int
4311 nfsrv_checkrestart(nfsquad_t clientid, u_int32_t flags,
4312     nfsv4stateid_t *stateidp, int specialid)
4313 {
4314 	int ret = 0;
4315 
4316 	/*
4317 	 * First check for a server restart. Open, LockT, ReleaseLockOwner
4318 	 * and DelegPurge have a clientid, the rest a stateid.
4319 	 */
4320 	if (flags &
4321 	    (NFSLCK_OPEN | NFSLCK_TEST | NFSLCK_RELEASE | NFSLCK_DELEGPURGE)) {
4322 		if (clientid.lval[0] != nfsrvboottime) {
4323 			ret = NFSERR_STALECLIENTID;
4324 			goto out;
4325 		}
4326 	} else if (stateidp->other[0] != nfsrvboottime &&
4327 		specialid == 0) {
4328 		ret = NFSERR_STALESTATEID;
4329 		goto out;
4330 	}
4331 
4332 	/*
4333 	 * Read, Write, Setattr and LockT can return NFSERR_GRACE and do
4334 	 * not use a lock/open owner seqid#, so the check can be done now.
4335 	 * (The others will be checked, as required, later.)
4336 	 */
4337 	if (!(flags & (NFSLCK_CHECK | NFSLCK_TEST)))
4338 		goto out;
4339 
4340 	NFSLOCKSTATE();
4341 	ret = nfsrv_checkgrace(NULL, NULL, flags);
4342 	NFSUNLOCKSTATE();
4343 
4344 out:
4345 	NFSEXITCODE(ret);
4346 	return (ret);
4347 }
4348 
4349 /*
4350  * Check for grace.
4351  */
4352 static int
4353 nfsrv_checkgrace(struct nfsrv_descript *nd, struct nfsclient *clp,
4354     u_int32_t flags)
4355 {
4356 	int error = 0, notreclaimed;
4357 	struct nfsrv_stable *sp;
4358 
4359 	if ((nfsrv_stablefirst.nsf_flags & (NFSNSF_UPDATEDONE |
4360 	     NFSNSF_GRACEOVER)) == 0) {
4361 		/*
4362 		 * First, check to see if all of the clients have done a
4363 		 * ReclaimComplete.  If so, grace can end now.
4364 		 */
4365 		notreclaimed = 0;
4366 		LIST_FOREACH(sp, &nfsrv_stablefirst.nsf_head, nst_list) {
4367 			if ((sp->nst_flag & NFSNST_RECLAIMED) == 0) {
4368 				notreclaimed = 1;
4369 				break;
4370 			}
4371 		}
4372 		if (notreclaimed == 0)
4373 			nfsrv_stablefirst.nsf_flags |= (NFSNSF_GRACEOVER |
4374 			    NFSNSF_NEEDLOCK);
4375 	}
4376 
4377 	if ((nfsrv_stablefirst.nsf_flags & NFSNSF_GRACEOVER) != 0) {
4378 		if (flags & NFSLCK_RECLAIM) {
4379 			error = NFSERR_NOGRACE;
4380 			goto out;
4381 		}
4382 	} else {
4383 		if (!(flags & NFSLCK_RECLAIM)) {
4384 			error = NFSERR_GRACE;
4385 			goto out;
4386 		}
4387 		if (nd != NULL && clp != NULL &&
4388 		    (nd->nd_flag & ND_NFSV41) != 0 &&
4389 		    (clp->lc_flags & LCL_RECLAIMCOMPLETE) != 0) {
4390 			error = NFSERR_NOGRACE;
4391 			goto out;
4392 		}
4393 
4394 		/*
4395 		 * If grace is almost over and we are still getting Reclaims,
4396 		 * extend grace a bit.
4397 		 */
4398 		if ((NFSD_MONOSEC + NFSRV_LEASEDELTA) >
4399 		    nfsrv_stablefirst.nsf_eograce)
4400 			nfsrv_stablefirst.nsf_eograce = NFSD_MONOSEC +
4401 				NFSRV_LEASEDELTA;
4402 	}
4403 
4404 out:
4405 	NFSEXITCODE(error);
4406 	return (error);
4407 }
4408 
4409 /*
4410  * Do a server callback.
4411  * The "trunc" argument is slightly overloaded and refers to different
4412  * boolean arguments for CBRECALL and CBLAYOUTRECALL.
4413  */
4414 static int
4415 nfsrv_docallback(struct nfsclient *clp, int procnum, nfsv4stateid_t *stateidp,
4416     int trunc, fhandle_t *fhp, struct nfsvattr *nap, nfsattrbit_t *attrbitp,
4417     int laytype, NFSPROC_T *p)
4418 {
4419 	mbuf_t m;
4420 	u_int32_t *tl;
4421 	struct nfsrv_descript *nd;
4422 	struct ucred *cred;
4423 	int error = 0;
4424 	u_int32_t callback;
4425 	struct nfsdsession *sep = NULL;
4426 	uint64_t tval;
4427 
4428 	nd = malloc(sizeof(*nd), M_TEMP, M_WAITOK | M_ZERO);
4429 	cred = newnfs_getcred();
4430 	NFSLOCKSTATE();	/* mostly for lc_cbref++ */
4431 	if (clp->lc_flags & LCL_NEEDSCONFIRM) {
4432 		NFSUNLOCKSTATE();
4433 		panic("docallb");
4434 	}
4435 	clp->lc_cbref++;
4436 
4437 	/*
4438 	 * Fill the callback program# and version into the request
4439 	 * structure for newnfs_connect() to use.
4440 	 */
4441 	clp->lc_req.nr_prog = clp->lc_program;
4442 #ifdef notnow
4443 	if ((clp->lc_flags & LCL_NFSV41) != 0)
4444 		clp->lc_req.nr_vers = NFSV41_CBVERS;
4445 	else
4446 #endif
4447 		clp->lc_req.nr_vers = NFSV4_CBVERS;
4448 
4449 	/*
4450 	 * First, fill in some of the fields of nd and cr.
4451 	 */
4452 	nd->nd_flag = ND_NFSV4;
4453 	if (clp->lc_flags & LCL_GSS)
4454 		nd->nd_flag |= ND_KERBV;
4455 	if ((clp->lc_flags & LCL_NFSV41) != 0)
4456 		nd->nd_flag |= ND_NFSV41;
4457 	if ((clp->lc_flags & LCL_NFSV42) != 0)
4458 		nd->nd_flag |= ND_NFSV42;
4459 	nd->nd_repstat = 0;
4460 	cred->cr_uid = clp->lc_uid;
4461 	cred->cr_gid = clp->lc_gid;
4462 	callback = clp->lc_callback;
4463 	NFSUNLOCKSTATE();
4464 	cred->cr_ngroups = 1;
4465 
4466 	/*
4467 	 * Get the first mbuf for the request.
4468 	 */
4469 	MGET(m, M_WAITOK, MT_DATA);
4470 	mbuf_setlen(m, 0);
4471 	nd->nd_mreq = nd->nd_mb = m;
4472 	nd->nd_bpos = NFSMTOD(m, caddr_t);
4473 
4474 	/*
4475 	 * and build the callback request.
4476 	 */
4477 	if (procnum == NFSV4OP_CBGETATTR) {
4478 		nd->nd_procnum = NFSV4PROC_CBCOMPOUND;
4479 		error = nfsrv_cbcallargs(nd, clp, callback, NFSV4OP_CBGETATTR,
4480 		    "CB Getattr", &sep);
4481 		if (error != 0) {
4482 			mbuf_freem(nd->nd_mreq);
4483 			goto errout;
4484 		}
4485 		(void)nfsm_fhtom(nd, (u_int8_t *)fhp, NFSX_MYFH, 0);
4486 		(void)nfsrv_putattrbit(nd, attrbitp);
4487 	} else if (procnum == NFSV4OP_CBRECALL) {
4488 		nd->nd_procnum = NFSV4PROC_CBCOMPOUND;
4489 		error = nfsrv_cbcallargs(nd, clp, callback, NFSV4OP_CBRECALL,
4490 		    "CB Recall", &sep);
4491 		if (error != 0) {
4492 			mbuf_freem(nd->nd_mreq);
4493 			goto errout;
4494 		}
4495 		NFSM_BUILD(tl, u_int32_t *, NFSX_UNSIGNED + NFSX_STATEID);
4496 		*tl++ = txdr_unsigned(stateidp->seqid);
4497 		NFSBCOPY((caddr_t)stateidp->other, (caddr_t)tl,
4498 		    NFSX_STATEIDOTHER);
4499 		tl += (NFSX_STATEIDOTHER / NFSX_UNSIGNED);
4500 		if (trunc)
4501 			*tl = newnfs_true;
4502 		else
4503 			*tl = newnfs_false;
4504 		(void)nfsm_fhtom(nd, (u_int8_t *)fhp, NFSX_MYFH, 0);
4505 	} else if (procnum == NFSV4OP_CBLAYOUTRECALL) {
4506 		NFSD_DEBUG(4, "docallback layout recall\n");
4507 		nd->nd_procnum = NFSV4PROC_CBCOMPOUND;
4508 		error = nfsrv_cbcallargs(nd, clp, callback,
4509 		    NFSV4OP_CBLAYOUTRECALL, "CB Reclayout", &sep);
4510 		NFSD_DEBUG(4, "aft cbcallargs=%d\n", error);
4511 		if (error != 0) {
4512 			mbuf_freem(nd->nd_mreq);
4513 			goto errout;
4514 		}
4515 		NFSM_BUILD(tl, u_int32_t *, 4 * NFSX_UNSIGNED);
4516 		*tl++ = txdr_unsigned(laytype);
4517 		*tl++ = txdr_unsigned(NFSLAYOUTIOMODE_ANY);
4518 		if (trunc)
4519 			*tl++ = newnfs_true;
4520 		else
4521 			*tl++ = newnfs_false;
4522 		*tl = txdr_unsigned(NFSV4LAYOUTRET_FILE);
4523 		nfsm_fhtom(nd, (uint8_t *)fhp, NFSX_MYFH, 0);
4524 		NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_HYPER + NFSX_STATEID);
4525 		tval = 0;
4526 		txdr_hyper(tval, tl); tl += 2;
4527 		tval = UINT64_MAX;
4528 		txdr_hyper(tval, tl); tl += 2;
4529 		*tl++ = txdr_unsigned(stateidp->seqid);
4530 		NFSBCOPY(stateidp->other, tl, NFSX_STATEIDOTHER);
4531 		tl += (NFSX_STATEIDOTHER / NFSX_UNSIGNED);
4532 		NFSD_DEBUG(4, "aft args\n");
4533 	} else if (procnum == NFSV4PROC_CBNULL) {
4534 		nd->nd_procnum = NFSV4PROC_CBNULL;
4535 		if ((clp->lc_flags & LCL_NFSV41) != 0) {
4536 			error = nfsv4_getcbsession(clp, &sep);
4537 			if (error != 0) {
4538 				mbuf_freem(nd->nd_mreq);
4539 				goto errout;
4540 			}
4541 		}
4542 	} else {
4543 		error = NFSERR_SERVERFAULT;
4544 		mbuf_freem(nd->nd_mreq);
4545 		goto errout;
4546 	}
4547 
4548 	/*
4549 	 * Call newnfs_connect(), as required, and then newnfs_request().
4550 	 */
4551 	(void) newnfs_sndlock(&clp->lc_req.nr_lock);
4552 	if (clp->lc_req.nr_client == NULL) {
4553 		if ((clp->lc_flags & LCL_NFSV41) != 0) {
4554 			error = ECONNREFUSED;
4555 			nfsrv_freesession(sep, NULL);
4556 		} else if (nd->nd_procnum == NFSV4PROC_CBNULL)
4557 			error = newnfs_connect(NULL, &clp->lc_req, cred,
4558 			    NULL, 1);
4559 		else
4560 			error = newnfs_connect(NULL, &clp->lc_req, cred,
4561 			    NULL, 3);
4562 	}
4563 	newnfs_sndunlock(&clp->lc_req.nr_lock);
4564 	NFSD_DEBUG(4, "aft sndunlock=%d\n", error);
4565 	if (!error) {
4566 		if ((nd->nd_flag & ND_NFSV41) != 0) {
4567 			KASSERT(sep != NULL, ("sep NULL"));
4568 			if (sep->sess_cbsess.nfsess_xprt != NULL)
4569 				error = newnfs_request(nd, NULL, clp,
4570 				    &clp->lc_req, NULL, NULL, cred,
4571 				    clp->lc_program, clp->lc_req.nr_vers, NULL,
4572 				    1, NULL, &sep->sess_cbsess);
4573 			else {
4574 				/*
4575 				 * This should probably never occur, but if a
4576 				 * client somehow does an RPC without a
4577 				 * SequenceID Op that causes a callback just
4578 				 * after the nfsd threads have been terminated
4579 				 * and restared we could conceivably get here
4580 				 * without a backchannel xprt.
4581 				 */
4582 				printf("nfsrv_docallback: no xprt\n");
4583 				error = ECONNREFUSED;
4584 			}
4585 			NFSD_DEBUG(4, "aft newnfs_request=%d\n", error);
4586 			nfsrv_freesession(sep, NULL);
4587 		} else
4588 			error = newnfs_request(nd, NULL, clp, &clp->lc_req,
4589 			    NULL, NULL, cred, clp->lc_program,
4590 			    clp->lc_req.nr_vers, NULL, 1, NULL, NULL);
4591 	}
4592 errout:
4593 	NFSFREECRED(cred);
4594 
4595 	/*
4596 	 * If error is set here, the Callback path isn't working
4597 	 * properly, so twiddle the appropriate LCL_ flags.
4598 	 * (nd_repstat != 0 indicates the Callback path is working,
4599 	 *  but the callback failed on the client.)
4600 	 */
4601 	if (error) {
4602 		/*
4603 		 * Mark the callback pathway down, which disabled issuing
4604 		 * of delegations and gets Renew to return NFSERR_CBPATHDOWN.
4605 		 */
4606 		NFSLOCKSTATE();
4607 		clp->lc_flags |= LCL_CBDOWN;
4608 		NFSUNLOCKSTATE();
4609 	} else {
4610 		/*
4611 		 * Callback worked. If the callback path was down, disable
4612 		 * callbacks, so no more delegations will be issued. (This
4613 		 * is done on the assumption that the callback pathway is
4614 		 * flakey.)
4615 		 */
4616 		NFSLOCKSTATE();
4617 		if (clp->lc_flags & LCL_CBDOWN)
4618 			clp->lc_flags &= ~(LCL_CBDOWN | LCL_CALLBACKSON);
4619 		NFSUNLOCKSTATE();
4620 		if (nd->nd_repstat) {
4621 			error = nd->nd_repstat;
4622 			NFSD_DEBUG(1, "nfsrv_docallback op=%d err=%d\n",
4623 			    procnum, error);
4624 		} else if (error == 0 && procnum == NFSV4OP_CBGETATTR)
4625 			error = nfsv4_loadattr(nd, NULL, nap, NULL, NULL, 0,
4626 			    NULL, NULL, NULL, NULL, NULL, 0, NULL, NULL, NULL,
4627 			    p, NULL);
4628 		mbuf_freem(nd->nd_mrep);
4629 	}
4630 	NFSLOCKSTATE();
4631 	clp->lc_cbref--;
4632 	if ((clp->lc_flags & LCL_WAKEUPWANTED) && clp->lc_cbref == 0) {
4633 		clp->lc_flags &= ~LCL_WAKEUPWANTED;
4634 		wakeup(clp);
4635 	}
4636 	NFSUNLOCKSTATE();
4637 
4638 	free(nd, M_TEMP);
4639 	NFSEXITCODE(error);
4640 	return (error);
4641 }
4642 
4643 /*
4644  * Set up the compound RPC for the callback.
4645  */
4646 static int
4647 nfsrv_cbcallargs(struct nfsrv_descript *nd, struct nfsclient *clp,
4648     uint32_t callback, int op, const char *optag, struct nfsdsession **sepp)
4649 {
4650 	uint32_t *tl;
4651 	int error, len;
4652 
4653 	len = strlen(optag);
4654 	(void)nfsm_strtom(nd, optag, len);
4655 	NFSM_BUILD(tl, uint32_t *, 4 * NFSX_UNSIGNED);
4656 	if ((nd->nd_flag & ND_NFSV41) != 0) {
4657 		if ((nd->nd_flag & ND_NFSV42) != 0)
4658 			*tl++ = txdr_unsigned(NFSV42_MINORVERSION);
4659 		else
4660 			*tl++ = txdr_unsigned(NFSV41_MINORVERSION);
4661 		*tl++ = txdr_unsigned(callback);
4662 		*tl++ = txdr_unsigned(2);
4663 		*tl = txdr_unsigned(NFSV4OP_CBSEQUENCE);
4664 		error = nfsv4_setcbsequence(nd, clp, 1, sepp);
4665 		if (error != 0)
4666 			return (error);
4667 		NFSM_BUILD(tl, u_int32_t *, NFSX_UNSIGNED);
4668 		*tl = txdr_unsigned(op);
4669 	} else {
4670 		*tl++ = txdr_unsigned(NFSV4_MINORVERSION);
4671 		*tl++ = txdr_unsigned(callback);
4672 		*tl++ = txdr_unsigned(1);
4673 		*tl = txdr_unsigned(op);
4674 	}
4675 	return (0);
4676 }
4677 
4678 /*
4679  * Return the next index# for a clientid. Mostly just increment and return
4680  * the next one, but... if the 32bit unsigned does actually wrap around,
4681  * it should be rebooted.
4682  * At an average rate of one new client per second, it will wrap around in
4683  * approximately 136 years. (I think the server will have been shut
4684  * down or rebooted before then.)
4685  */
4686 static u_int32_t
4687 nfsrv_nextclientindex(void)
4688 {
4689 	static u_int32_t client_index = 0;
4690 
4691 	client_index++;
4692 	if (client_index != 0)
4693 		return (client_index);
4694 
4695 	printf("%s: out of clientids\n", __func__);
4696 	return (client_index);
4697 }
4698 
4699 /*
4700  * Return the next index# for a stateid. Mostly just increment and return
4701  * the next one, but... if the 32bit unsigned does actually wrap around
4702  * (will a BSD server stay up that long?), find
4703  * new start and end values.
4704  */
4705 static u_int32_t
4706 nfsrv_nextstateindex(struct nfsclient *clp)
4707 {
4708 	struct nfsstate *stp;
4709 	int i;
4710 	u_int32_t canuse, min_index, max_index;
4711 
4712 	if (!(clp->lc_flags & LCL_INDEXNOTOK)) {
4713 		clp->lc_stateindex++;
4714 		if (clp->lc_stateindex != clp->lc_statemaxindex)
4715 			return (clp->lc_stateindex);
4716 	}
4717 
4718 	/*
4719 	 * Yuck, we've hit the end.
4720 	 * Look for a new min and max.
4721 	 */
4722 	min_index = 0;
4723 	max_index = 0xffffffff;
4724 	for (i = 0; i < nfsrv_statehashsize; i++) {
4725 	    LIST_FOREACH(stp, &clp->lc_stateid[i], ls_hash) {
4726 		if (stp->ls_stateid.other[2] > 0x80000000) {
4727 		    if (stp->ls_stateid.other[2] < max_index)
4728 			max_index = stp->ls_stateid.other[2];
4729 		} else {
4730 		    if (stp->ls_stateid.other[2] > min_index)
4731 			min_index = stp->ls_stateid.other[2];
4732 		}
4733 	    }
4734 	}
4735 
4736 	/*
4737 	 * Yikes, highly unlikely, but I'll handle it anyhow.
4738 	 */
4739 	if (min_index == 0x80000000 && max_index == 0x80000001) {
4740 	    canuse = 0;
4741 	    /*
4742 	     * Loop around until we find an unused entry. Return that
4743 	     * and set LCL_INDEXNOTOK, so the search will continue next time.
4744 	     * (This is one of those rare cases where a goto is the
4745 	     *  cleanest way to code the loop.)
4746 	     */
4747 tryagain:
4748 	    for (i = 0; i < nfsrv_statehashsize; i++) {
4749 		LIST_FOREACH(stp, &clp->lc_stateid[i], ls_hash) {
4750 		    if (stp->ls_stateid.other[2] == canuse) {
4751 			canuse++;
4752 			goto tryagain;
4753 		    }
4754 		}
4755 	    }
4756 	    clp->lc_flags |= LCL_INDEXNOTOK;
4757 	    return (canuse);
4758 	}
4759 
4760 	/*
4761 	 * Ok to start again from min + 1.
4762 	 */
4763 	clp->lc_stateindex = min_index + 1;
4764 	clp->lc_statemaxindex = max_index;
4765 	clp->lc_flags &= ~LCL_INDEXNOTOK;
4766 	return (clp->lc_stateindex);
4767 }
4768 
4769 /*
4770  * The following functions handle the stable storage file that deals with
4771  * the edge conditions described in RFC3530 Sec. 8.6.3.
4772  * The file is as follows:
4773  * - a single record at the beginning that has the lease time of the
4774  *   previous server instance (before the last reboot) and the nfsrvboottime
4775  *   values for the previous server boots.
4776  *   These previous boot times are used to ensure that the current
4777  *   nfsrvboottime does not, somehow, get set to a previous one.
4778  *   (This is important so that Stale ClientIDs and StateIDs can
4779  *    be recognized.)
4780  *   The number of previous nfsvrboottime values precedes the list.
4781  * - followed by some number of appended records with:
4782  *   - client id string
4783  *   - flag that indicates it is a record revoking state via lease
4784  *     expiration or similar
4785  *     OR has successfully acquired state.
4786  * These structures vary in length, with the client string at the end, up
4787  * to NFSV4_OPAQUELIMIT in size.
4788  *
4789  * At the end of the grace period, the file is truncated, the first
4790  * record is rewritten with updated information and any acquired state
4791  * records for successful reclaims of state are written.
4792  *
4793  * Subsequent records are appended when the first state is issued to
4794  * a client and when state is revoked for a client.
4795  *
4796  * When reading the file in, state issued records that come later in
4797  * the file override older ones, since the append log is in cronological order.
4798  * If, for some reason, the file can't be read, the grace period is
4799  * immediately terminated and all reclaims get NFSERR_NOGRACE.
4800  */
4801 
4802 /*
4803  * Read in the stable storage file. Called by nfssvc() before the nfsd
4804  * processes start servicing requests.
4805  */
4806 APPLESTATIC void
4807 nfsrv_setupstable(NFSPROC_T *p)
4808 {
4809 	struct nfsrv_stablefirst *sf = &nfsrv_stablefirst;
4810 	struct nfsrv_stable *sp, *nsp;
4811 	struct nfst_rec *tsp;
4812 	int error, i, tryagain;
4813 	off_t off = 0;
4814 	ssize_t aresid, len;
4815 
4816 	/*
4817 	 * If NFSNSF_UPDATEDONE is set, this is a restart of the nfsds without
4818 	 * a reboot, so state has not been lost.
4819 	 */
4820 	if (sf->nsf_flags & NFSNSF_UPDATEDONE)
4821 		return;
4822 	/*
4823 	 * Set Grace over just until the file reads successfully.
4824 	 */
4825 	nfsrvboottime = time_second;
4826 	LIST_INIT(&sf->nsf_head);
4827 	sf->nsf_flags = (NFSNSF_GRACEOVER | NFSNSF_NEEDLOCK);
4828 	sf->nsf_eograce = NFSD_MONOSEC + NFSRV_LEASEDELTA;
4829 	if (sf->nsf_fp == NULL)
4830 		return;
4831 	error = NFSD_RDWR(UIO_READ, NFSFPVNODE(sf->nsf_fp),
4832 	    (caddr_t)&sf->nsf_rec, sizeof (struct nfsf_rec), off, UIO_SYSSPACE,
4833 	    0, NFSFPCRED(sf->nsf_fp), &aresid, p);
4834 	if (error || aresid || sf->nsf_numboots == 0 ||
4835 		sf->nsf_numboots > NFSNSF_MAXNUMBOOTS)
4836 		return;
4837 
4838 	/*
4839 	 * Now, read in the boottimes.
4840 	 */
4841 	sf->nsf_bootvals = (time_t *)malloc((sf->nsf_numboots + 1) *
4842 		sizeof (time_t), M_TEMP, M_WAITOK);
4843 	off = sizeof (struct nfsf_rec);
4844 	error = NFSD_RDWR(UIO_READ, NFSFPVNODE(sf->nsf_fp),
4845 	    (caddr_t)sf->nsf_bootvals, sf->nsf_numboots * sizeof (time_t), off,
4846 	    UIO_SYSSPACE, 0, NFSFPCRED(sf->nsf_fp), &aresid, p);
4847 	if (error || aresid) {
4848 		free(sf->nsf_bootvals, M_TEMP);
4849 		sf->nsf_bootvals = NULL;
4850 		return;
4851 	}
4852 
4853 	/*
4854 	 * Make sure this nfsrvboottime is different from all recorded
4855 	 * previous ones.
4856 	 */
4857 	do {
4858 		tryagain = 0;
4859 		for (i = 0; i < sf->nsf_numboots; i++) {
4860 			if (nfsrvboottime == sf->nsf_bootvals[i]) {
4861 				nfsrvboottime++;
4862 				tryagain = 1;
4863 				break;
4864 			}
4865 		}
4866 	} while (tryagain);
4867 
4868 	sf->nsf_flags |= NFSNSF_OK;
4869 	off += (sf->nsf_numboots * sizeof (time_t));
4870 
4871 	/*
4872 	 * Read through the file, building a list of records for grace
4873 	 * checking.
4874 	 * Each record is between sizeof (struct nfst_rec) and
4875 	 * sizeof (struct nfst_rec) + NFSV4_OPAQUELIMIT - 1
4876 	 * and is actually sizeof (struct nfst_rec) + nst_len - 1.
4877 	 */
4878 	tsp = (struct nfst_rec *)malloc(sizeof (struct nfst_rec) +
4879 		NFSV4_OPAQUELIMIT - 1, M_TEMP, M_WAITOK);
4880 	do {
4881 	    error = NFSD_RDWR(UIO_READ, NFSFPVNODE(sf->nsf_fp),
4882 	        (caddr_t)tsp, sizeof (struct nfst_rec) + NFSV4_OPAQUELIMIT - 1,
4883 	        off, UIO_SYSSPACE, 0, NFSFPCRED(sf->nsf_fp), &aresid, p);
4884 	    len = (sizeof (struct nfst_rec) + NFSV4_OPAQUELIMIT - 1) - aresid;
4885 	    if (error || (len > 0 && (len < sizeof (struct nfst_rec) ||
4886 		len < (sizeof (struct nfst_rec) + tsp->len - 1)))) {
4887 		/*
4888 		 * Yuck, the file has been corrupted, so just return
4889 		 * after clearing out any restart state, so the grace period
4890 		 * is over.
4891 		 */
4892 		LIST_FOREACH_SAFE(sp, &sf->nsf_head, nst_list, nsp) {
4893 			LIST_REMOVE(sp, nst_list);
4894 			free(sp, M_TEMP);
4895 		}
4896 		free(tsp, M_TEMP);
4897 		sf->nsf_flags &= ~NFSNSF_OK;
4898 		free(sf->nsf_bootvals, M_TEMP);
4899 		sf->nsf_bootvals = NULL;
4900 		return;
4901 	    }
4902 	    if (len > 0) {
4903 		off += sizeof (struct nfst_rec) + tsp->len - 1;
4904 		/*
4905 		 * Search the list for a matching client.
4906 		 */
4907 		LIST_FOREACH(sp, &sf->nsf_head, nst_list) {
4908 			if (tsp->len == sp->nst_len &&
4909 			    !NFSBCMP(tsp->client, sp->nst_client, tsp->len))
4910 				break;
4911 		}
4912 		if (sp == LIST_END(&sf->nsf_head)) {
4913 			sp = (struct nfsrv_stable *)malloc(tsp->len +
4914 				sizeof (struct nfsrv_stable) - 1, M_TEMP,
4915 				M_WAITOK);
4916 			NFSBCOPY((caddr_t)tsp, (caddr_t)&sp->nst_rec,
4917 				sizeof (struct nfst_rec) + tsp->len - 1);
4918 			LIST_INSERT_HEAD(&sf->nsf_head, sp, nst_list);
4919 		} else {
4920 			if (tsp->flag == NFSNST_REVOKE)
4921 				sp->nst_flag |= NFSNST_REVOKE;
4922 			else
4923 				/*
4924 				 * A subsequent timestamp indicates the client
4925 				 * did a setclientid/confirm and any previous
4926 				 * revoke is no longer relevant.
4927 				 */
4928 				sp->nst_flag &= ~NFSNST_REVOKE;
4929 		}
4930 	    }
4931 	} while (len > 0);
4932 	free(tsp, M_TEMP);
4933 	sf->nsf_flags = NFSNSF_OK;
4934 	sf->nsf_eograce = NFSD_MONOSEC + sf->nsf_lease +
4935 		NFSRV_LEASEDELTA;
4936 }
4937 
4938 /*
4939  * Update the stable storage file, now that the grace period is over.
4940  */
4941 APPLESTATIC void
4942 nfsrv_updatestable(NFSPROC_T *p)
4943 {
4944 	struct nfsrv_stablefirst *sf = &nfsrv_stablefirst;
4945 	struct nfsrv_stable *sp, *nsp;
4946 	int i;
4947 	struct nfsvattr nva;
4948 	vnode_t vp;
4949 #if defined(__FreeBSD_version) && (__FreeBSD_version >= 500000)
4950 	mount_t mp = NULL;
4951 #endif
4952 	int error;
4953 
4954 	if (sf->nsf_fp == NULL || (sf->nsf_flags & NFSNSF_UPDATEDONE))
4955 		return;
4956 	sf->nsf_flags |= NFSNSF_UPDATEDONE;
4957 	/*
4958 	 * Ok, we need to rewrite the stable storage file.
4959 	 * - truncate to 0 length
4960 	 * - write the new first structure
4961 	 * - loop through the data structures, writing out any that
4962 	 *   have timestamps older than the old boot
4963 	 */
4964 	if (sf->nsf_bootvals) {
4965 		sf->nsf_numboots++;
4966 		for (i = sf->nsf_numboots - 2; i >= 0; i--)
4967 			sf->nsf_bootvals[i + 1] = sf->nsf_bootvals[i];
4968 	} else {
4969 		sf->nsf_numboots = 1;
4970 		sf->nsf_bootvals = (time_t *)malloc(sizeof (time_t),
4971 			M_TEMP, M_WAITOK);
4972 	}
4973 	sf->nsf_bootvals[0] = nfsrvboottime;
4974 	sf->nsf_lease = nfsrv_lease;
4975 	NFSVNO_ATTRINIT(&nva);
4976 	NFSVNO_SETATTRVAL(&nva, size, 0);
4977 	vp = NFSFPVNODE(sf->nsf_fp);
4978 	vn_start_write(vp, &mp, V_WAIT);
4979 	if (NFSVOPLOCK(vp, LK_EXCLUSIVE) == 0) {
4980 		error = nfsvno_setattr(vp, &nva, NFSFPCRED(sf->nsf_fp), p,
4981 		    NULL);
4982 		NFSVOPUNLOCK(vp);
4983 	} else
4984 		error = EPERM;
4985 	vn_finished_write(mp);
4986 	if (!error)
4987 	    error = NFSD_RDWR(UIO_WRITE, vp,
4988 		(caddr_t)&sf->nsf_rec, sizeof (struct nfsf_rec), (off_t)0,
4989 		UIO_SYSSPACE, IO_SYNC, NFSFPCRED(sf->nsf_fp), NULL, p);
4990 	if (!error)
4991 	    error = NFSD_RDWR(UIO_WRITE, vp,
4992 		(caddr_t)sf->nsf_bootvals,
4993 		sf->nsf_numboots * sizeof (time_t),
4994 		(off_t)(sizeof (struct nfsf_rec)),
4995 		UIO_SYSSPACE, IO_SYNC, NFSFPCRED(sf->nsf_fp), NULL, p);
4996 	free(sf->nsf_bootvals, M_TEMP);
4997 	sf->nsf_bootvals = NULL;
4998 	if (error) {
4999 		sf->nsf_flags &= ~NFSNSF_OK;
5000 		printf("EEK! Can't write NfsV4 stable storage file\n");
5001 		return;
5002 	}
5003 	sf->nsf_flags |= NFSNSF_OK;
5004 
5005 	/*
5006 	 * Loop through the list and write out timestamp records for
5007 	 * any clients that successfully reclaimed state.
5008 	 */
5009 	LIST_FOREACH_SAFE(sp, &sf->nsf_head, nst_list, nsp) {
5010 		if (sp->nst_flag & NFSNST_GOTSTATE) {
5011 			nfsrv_writestable(sp->nst_client, sp->nst_len,
5012 				NFSNST_NEWSTATE, p);
5013 			sp->nst_clp->lc_flags |= LCL_STAMPEDSTABLE;
5014 		}
5015 		LIST_REMOVE(sp, nst_list);
5016 		free(sp, M_TEMP);
5017 	}
5018 	nfsrv_backupstable();
5019 }
5020 
5021 /*
5022  * Append a record to the stable storage file.
5023  */
5024 APPLESTATIC void
5025 nfsrv_writestable(u_char *client, int len, int flag, NFSPROC_T *p)
5026 {
5027 	struct nfsrv_stablefirst *sf = &nfsrv_stablefirst;
5028 	struct nfst_rec *sp;
5029 	int error;
5030 
5031 	if (!(sf->nsf_flags & NFSNSF_OK) || sf->nsf_fp == NULL)
5032 		return;
5033 	sp = (struct nfst_rec *)malloc(sizeof (struct nfst_rec) +
5034 		len - 1, M_TEMP, M_WAITOK);
5035 	sp->len = len;
5036 	NFSBCOPY(client, sp->client, len);
5037 	sp->flag = flag;
5038 	error = NFSD_RDWR(UIO_WRITE, NFSFPVNODE(sf->nsf_fp),
5039 	    (caddr_t)sp, sizeof (struct nfst_rec) + len - 1, (off_t)0,
5040 	    UIO_SYSSPACE, (IO_SYNC | IO_APPEND), NFSFPCRED(sf->nsf_fp), NULL, p);
5041 	free(sp, M_TEMP);
5042 	if (error) {
5043 		sf->nsf_flags &= ~NFSNSF_OK;
5044 		printf("EEK! Can't write NfsV4 stable storage file\n");
5045 	}
5046 }
5047 
5048 /*
5049  * This function is called during the grace period to mark a client
5050  * that successfully reclaimed state.
5051  */
5052 static void
5053 nfsrv_markstable(struct nfsclient *clp)
5054 {
5055 	struct nfsrv_stable *sp;
5056 
5057 	/*
5058 	 * First find the client structure.
5059 	 */
5060 	LIST_FOREACH(sp, &nfsrv_stablefirst.nsf_head, nst_list) {
5061 		if (sp->nst_len == clp->lc_idlen &&
5062 		    !NFSBCMP(sp->nst_client, clp->lc_id, sp->nst_len))
5063 			break;
5064 	}
5065 	if (sp == LIST_END(&nfsrv_stablefirst.nsf_head))
5066 		return;
5067 
5068 	/*
5069 	 * Now, just mark it and set the nfsclient back pointer.
5070 	 */
5071 	sp->nst_flag |= NFSNST_GOTSTATE;
5072 	sp->nst_clp = clp;
5073 }
5074 
5075 /*
5076  * This function is called when a NFSv4.1 client does a ReclaimComplete.
5077  * Very similar to nfsrv_markstable(), except for the flag being set.
5078  */
5079 static void
5080 nfsrv_markreclaim(struct nfsclient *clp)
5081 {
5082 	struct nfsrv_stable *sp;
5083 
5084 	/*
5085 	 * First find the client structure.
5086 	 */
5087 	LIST_FOREACH(sp, &nfsrv_stablefirst.nsf_head, nst_list) {
5088 		if (sp->nst_len == clp->lc_idlen &&
5089 		    !NFSBCMP(sp->nst_client, clp->lc_id, sp->nst_len))
5090 			break;
5091 	}
5092 	if (sp == LIST_END(&nfsrv_stablefirst.nsf_head))
5093 		return;
5094 
5095 	/*
5096 	 * Now, just set the flag.
5097 	 */
5098 	sp->nst_flag |= NFSNST_RECLAIMED;
5099 }
5100 
5101 /*
5102  * This function is called for a reclaim, to see if it gets grace.
5103  * It returns 0 if a reclaim is allowed, 1 otherwise.
5104  */
5105 static int
5106 nfsrv_checkstable(struct nfsclient *clp)
5107 {
5108 	struct nfsrv_stable *sp;
5109 
5110 	/*
5111 	 * First, find the entry for the client.
5112 	 */
5113 	LIST_FOREACH(sp, &nfsrv_stablefirst.nsf_head, nst_list) {
5114 		if (sp->nst_len == clp->lc_idlen &&
5115 		    !NFSBCMP(sp->nst_client, clp->lc_id, sp->nst_len))
5116 			break;
5117 	}
5118 
5119 	/*
5120 	 * If not in the list, state was revoked or no state was issued
5121 	 * since the previous reboot, a reclaim is denied.
5122 	 */
5123 	if (sp == LIST_END(&nfsrv_stablefirst.nsf_head) ||
5124 	    (sp->nst_flag & NFSNST_REVOKE) ||
5125 	    !(nfsrv_stablefirst.nsf_flags & NFSNSF_OK))
5126 		return (1);
5127 	return (0);
5128 }
5129 
5130 /*
5131  * Test for and try to clear out a conflicting client. This is called by
5132  * nfsrv_lockctrl() and nfsrv_openctrl() when conflicts with other clients
5133  * a found.
5134  * The trick here is that it can't revoke a conflicting client with an
5135  * expired lease unless it holds the v4root lock, so...
5136  * If no v4root lock, get the lock and return 1 to indicate "try again".
5137  * Return 0 to indicate the conflict can't be revoked and 1 to indicate
5138  * the revocation worked and the conflicting client is "bye, bye", so it
5139  * can be tried again.
5140  * Return 2 to indicate that the vnode is VIRF_DOOMED after NFSVOPLOCK().
5141  * Unlocks State before a non-zero value is returned.
5142  */
5143 static int
5144 nfsrv_clientconflict(struct nfsclient *clp, int *haslockp, vnode_t vp,
5145     NFSPROC_T *p)
5146 {
5147 	int gotlock, lktype = 0;
5148 
5149 	/*
5150 	 * If lease hasn't expired, we can't fix it.
5151 	 */
5152 	if (clp->lc_expiry >= NFSD_MONOSEC ||
5153 	    !(nfsrv_stablefirst.nsf_flags & NFSNSF_UPDATEDONE))
5154 		return (0);
5155 	if (*haslockp == 0) {
5156 		NFSUNLOCKSTATE();
5157 		if (vp != NULL) {
5158 			lktype = NFSVOPISLOCKED(vp);
5159 			NFSVOPUNLOCK(vp);
5160 		}
5161 		NFSLOCKV4ROOTMUTEX();
5162 		nfsv4_relref(&nfsv4rootfs_lock);
5163 		do {
5164 			gotlock = nfsv4_lock(&nfsv4rootfs_lock, 1, NULL,
5165 			    NFSV4ROOTLOCKMUTEXPTR, NULL);
5166 		} while (!gotlock);
5167 		NFSUNLOCKV4ROOTMUTEX();
5168 		*haslockp = 1;
5169 		if (vp != NULL) {
5170 			NFSVOPLOCK(vp, lktype | LK_RETRY);
5171 			if (VN_IS_DOOMED(vp))
5172 				return (2);
5173 		}
5174 		return (1);
5175 	}
5176 	NFSUNLOCKSTATE();
5177 
5178 	/*
5179 	 * Ok, we can expire the conflicting client.
5180 	 */
5181 	nfsrv_writestable(clp->lc_id, clp->lc_idlen, NFSNST_REVOKE, p);
5182 	nfsrv_backupstable();
5183 	nfsrv_cleanclient(clp, p);
5184 	nfsrv_freedeleglist(&clp->lc_deleg);
5185 	nfsrv_freedeleglist(&clp->lc_olddeleg);
5186 	LIST_REMOVE(clp, lc_hash);
5187 	nfsrv_zapclient(clp, p);
5188 	return (1);
5189 }
5190 
5191 /*
5192  * Resolve a delegation conflict.
5193  * Returns 0 to indicate the conflict was resolved without sleeping.
5194  * Return -1 to indicate that the caller should check for conflicts again.
5195  * Return > 0 for an error that should be returned, normally NFSERR_DELAY.
5196  *
5197  * Also, manipulate the nfsv4root_lock, as required. It isn't changed
5198  * for a return of 0, since there was no sleep and it could be required
5199  * later. It is released for a return of NFSERR_DELAY, since the caller
5200  * will return that error. It is released when a sleep was done waiting
5201  * for the delegation to be returned or expire (so that other nfsds can
5202  * handle ops). Then, it must be acquired for the write to stable storage.
5203  * (This function is somewhat similar to nfsrv_clientconflict(), but
5204  *  the semantics differ in a couple of subtle ways. The return of 0
5205  *  indicates the conflict was resolved without sleeping here, not
5206  *  that the conflict can't be resolved and the handling of nfsv4root_lock
5207  *  differs, as noted above.)
5208  * Unlocks State before returning a non-zero value.
5209  */
5210 static int
5211 nfsrv_delegconflict(struct nfsstate *stp, int *haslockp, NFSPROC_T *p,
5212     vnode_t vp)
5213 {
5214 	struct nfsclient *clp = stp->ls_clp;
5215 	int gotlock, error, lktype = 0, retrycnt, zapped_clp;
5216 	nfsv4stateid_t tstateid;
5217 	fhandle_t tfh;
5218 
5219 	/*
5220 	 * If the conflict is with an old delegation...
5221 	 */
5222 	if (stp->ls_flags & NFSLCK_OLDDELEG) {
5223 		/*
5224 		 * You can delete it, if it has expired.
5225 		 */
5226 		if (clp->lc_delegtime < NFSD_MONOSEC) {
5227 			nfsrv_freedeleg(stp);
5228 			NFSUNLOCKSTATE();
5229 			error = -1;
5230 			goto out;
5231 		}
5232 		NFSUNLOCKSTATE();
5233 		/*
5234 		 * During this delay, the old delegation could expire or it
5235 		 * could be recovered by the client via an Open with
5236 		 * CLAIM_DELEGATE_PREV.
5237 		 * Release the nfsv4root_lock, if held.
5238 		 */
5239 		if (*haslockp) {
5240 			*haslockp = 0;
5241 			NFSLOCKV4ROOTMUTEX();
5242 			nfsv4_unlock(&nfsv4rootfs_lock, 1);
5243 			NFSUNLOCKV4ROOTMUTEX();
5244 		}
5245 		error = NFSERR_DELAY;
5246 		goto out;
5247 	}
5248 
5249 	/*
5250 	 * It's a current delegation, so:
5251 	 * - check to see if the delegation has expired
5252 	 *   - if so, get the v4root lock and then expire it
5253 	 */
5254 	if (!(stp->ls_flags & NFSLCK_DELEGRECALL)) {
5255 		/*
5256 		 * - do a recall callback, since not yet done
5257 		 * For now, never allow truncate to be set. To use
5258 		 * truncate safely, it must be guaranteed that the
5259 		 * Remove, Rename or Setattr with size of 0 will
5260 		 * succeed and that would require major changes to
5261 		 * the VFS/Vnode OPs.
5262 		 * Set the expiry time large enough so that it won't expire
5263 		 * until after the callback, then set it correctly, once
5264 		 * the callback is done. (The delegation will now time
5265 		 * out whether or not the Recall worked ok. The timeout
5266 		 * will be extended when ops are done on the delegation
5267 		 * stateid, up to the timelimit.)
5268 		 */
5269 		stp->ls_delegtime = NFSD_MONOSEC + (2 * nfsrv_lease) +
5270 		    NFSRV_LEASEDELTA;
5271 		stp->ls_delegtimelimit = NFSD_MONOSEC + (6 * nfsrv_lease) +
5272 		    NFSRV_LEASEDELTA;
5273 		stp->ls_flags |= NFSLCK_DELEGRECALL;
5274 
5275 		/*
5276 		 * Loop NFSRV_CBRETRYCNT times while the CBRecall replies
5277 		 * NFSERR_BADSTATEID or NFSERR_BADHANDLE. This is done
5278 		 * in order to try and avoid a race that could happen
5279 		 * when a CBRecall request passed the Open reply with
5280 		 * the delegation in it when transitting the network.
5281 		 * Since nfsrv_docallback will sleep, don't use stp after
5282 		 * the call.
5283 		 */
5284 		NFSBCOPY((caddr_t)&stp->ls_stateid, (caddr_t)&tstateid,
5285 		    sizeof (tstateid));
5286 		NFSBCOPY((caddr_t)&stp->ls_lfp->lf_fh, (caddr_t)&tfh,
5287 		    sizeof (tfh));
5288 		NFSUNLOCKSTATE();
5289 		if (*haslockp) {
5290 			*haslockp = 0;
5291 			NFSLOCKV4ROOTMUTEX();
5292 			nfsv4_unlock(&nfsv4rootfs_lock, 1);
5293 			NFSUNLOCKV4ROOTMUTEX();
5294 		}
5295 		retrycnt = 0;
5296 		do {
5297 		    error = nfsrv_docallback(clp, NFSV4OP_CBRECALL,
5298 			&tstateid, 0, &tfh, NULL, NULL, 0, p);
5299 		    retrycnt++;
5300 		} while ((error == NFSERR_BADSTATEID ||
5301 		    error == NFSERR_BADHANDLE) && retrycnt < NFSV4_CBRETRYCNT);
5302 		error = NFSERR_DELAY;
5303 		goto out;
5304 	}
5305 
5306 	if (clp->lc_expiry >= NFSD_MONOSEC &&
5307 	    stp->ls_delegtime >= NFSD_MONOSEC) {
5308 		NFSUNLOCKSTATE();
5309 		/*
5310 		 * A recall has been done, but it has not yet expired.
5311 		 * So, RETURN_DELAY.
5312 		 */
5313 		if (*haslockp) {
5314 			*haslockp = 0;
5315 			NFSLOCKV4ROOTMUTEX();
5316 			nfsv4_unlock(&nfsv4rootfs_lock, 1);
5317 			NFSUNLOCKV4ROOTMUTEX();
5318 		}
5319 		error = NFSERR_DELAY;
5320 		goto out;
5321 	}
5322 
5323 	/*
5324 	 * If we don't yet have the lock, just get it and then return,
5325 	 * since we need that before deleting expired state, such as
5326 	 * this delegation.
5327 	 * When getting the lock, unlock the vnode, so other nfsds that
5328 	 * are in progress, won't get stuck waiting for the vnode lock.
5329 	 */
5330 	if (*haslockp == 0) {
5331 		NFSUNLOCKSTATE();
5332 		if (vp != NULL) {
5333 			lktype = NFSVOPISLOCKED(vp);
5334 			NFSVOPUNLOCK(vp);
5335 		}
5336 		NFSLOCKV4ROOTMUTEX();
5337 		nfsv4_relref(&nfsv4rootfs_lock);
5338 		do {
5339 			gotlock = nfsv4_lock(&nfsv4rootfs_lock, 1, NULL,
5340 			    NFSV4ROOTLOCKMUTEXPTR, NULL);
5341 		} while (!gotlock);
5342 		NFSUNLOCKV4ROOTMUTEX();
5343 		*haslockp = 1;
5344 		if (vp != NULL) {
5345 			NFSVOPLOCK(vp, lktype | LK_RETRY);
5346 			if (VN_IS_DOOMED(vp)) {
5347 				*haslockp = 0;
5348 				NFSLOCKV4ROOTMUTEX();
5349 				nfsv4_unlock(&nfsv4rootfs_lock, 1);
5350 				NFSUNLOCKV4ROOTMUTEX();
5351 				error = NFSERR_PERM;
5352 				goto out;
5353 			}
5354 		}
5355 		error = -1;
5356 		goto out;
5357 	}
5358 
5359 	NFSUNLOCKSTATE();
5360 	/*
5361 	 * Ok, we can delete the expired delegation.
5362 	 * First, write the Revoke record to stable storage and then
5363 	 * clear out the conflict.
5364 	 * Since all other nfsd threads are now blocked, we can safely
5365 	 * sleep without the state changing.
5366 	 */
5367 	nfsrv_writestable(clp->lc_id, clp->lc_idlen, NFSNST_REVOKE, p);
5368 	nfsrv_backupstable();
5369 	if (clp->lc_expiry < NFSD_MONOSEC) {
5370 		nfsrv_cleanclient(clp, p);
5371 		nfsrv_freedeleglist(&clp->lc_deleg);
5372 		nfsrv_freedeleglist(&clp->lc_olddeleg);
5373 		LIST_REMOVE(clp, lc_hash);
5374 		zapped_clp = 1;
5375 	} else {
5376 		nfsrv_freedeleg(stp);
5377 		zapped_clp = 0;
5378 	}
5379 	if (zapped_clp)
5380 		nfsrv_zapclient(clp, p);
5381 	error = -1;
5382 
5383 out:
5384 	NFSEXITCODE(error);
5385 	return (error);
5386 }
5387 
5388 /*
5389  * Check for a remove allowed, if remove is set to 1 and get rid of
5390  * delegations.
5391  */
5392 APPLESTATIC int
5393 nfsrv_checkremove(vnode_t vp, int remove, struct nfsrv_descript *nd,
5394     nfsquad_t clientid, NFSPROC_T *p)
5395 {
5396 	struct nfsclient *clp;
5397 	struct nfsstate *stp;
5398 	struct nfslockfile *lfp;
5399 	int error, haslock = 0;
5400 	fhandle_t nfh;
5401 
5402 	clp = NULL;
5403 	/*
5404 	 * First, get the lock file structure.
5405 	 * (A return of -1 means no associated state, so remove ok.)
5406 	 */
5407 	error = nfsrv_getlockfh(vp, NFSLCK_CHECK, NULL, &nfh, p);
5408 tryagain:
5409 	NFSLOCKSTATE();
5410 	if (error == 0 && clientid.qval != 0)
5411 		error = nfsrv_getclient(clientid, CLOPS_RENEW, &clp, NULL,
5412 		    (nfsquad_t)((u_quad_t)0), 0, nd, p);
5413 	if (!error)
5414 		error = nfsrv_getlockfile(NFSLCK_CHECK, NULL, &lfp, &nfh, 0);
5415 	if (error) {
5416 		NFSUNLOCKSTATE();
5417 		if (haslock) {
5418 			NFSLOCKV4ROOTMUTEX();
5419 			nfsv4_unlock(&nfsv4rootfs_lock, 1);
5420 			NFSUNLOCKV4ROOTMUTEX();
5421 		}
5422 		if (error == -1)
5423 			error = 0;
5424 		goto out;
5425 	}
5426 
5427 	/*
5428 	 * Now, we must Recall any delegations.
5429 	 */
5430 	error = nfsrv_cleandeleg(vp, lfp, clp, &haslock, p);
5431 	if (error) {
5432 		/*
5433 		 * nfsrv_cleandeleg() unlocks state for non-zero
5434 		 * return.
5435 		 */
5436 		if (error == -1)
5437 			goto tryagain;
5438 		if (haslock) {
5439 			NFSLOCKV4ROOTMUTEX();
5440 			nfsv4_unlock(&nfsv4rootfs_lock, 1);
5441 			NFSUNLOCKV4ROOTMUTEX();
5442 		}
5443 		goto out;
5444 	}
5445 
5446 	/*
5447 	 * Now, look for a conflicting open share.
5448 	 */
5449 	if (remove) {
5450 		/*
5451 		 * If the entry in the directory was the last reference to the
5452 		 * corresponding filesystem object, the object can be destroyed
5453 		 * */
5454 		if(lfp->lf_usecount>1)
5455 			LIST_FOREACH(stp, &lfp->lf_open, ls_file) {
5456 				if (stp->ls_flags & NFSLCK_WRITEDENY) {
5457 					error = NFSERR_FILEOPEN;
5458 					break;
5459 				}
5460 			}
5461 	}
5462 
5463 	NFSUNLOCKSTATE();
5464 	if (haslock) {
5465 		NFSLOCKV4ROOTMUTEX();
5466 		nfsv4_unlock(&nfsv4rootfs_lock, 1);
5467 		NFSUNLOCKV4ROOTMUTEX();
5468 	}
5469 
5470 out:
5471 	NFSEXITCODE(error);
5472 	return (error);
5473 }
5474 
5475 /*
5476  * Clear out all delegations for the file referred to by lfp.
5477  * May return NFSERR_DELAY, if there will be a delay waiting for
5478  * delegations to expire.
5479  * Returns -1 to indicate it slept while recalling a delegation.
5480  * This function has the side effect of deleting the nfslockfile structure,
5481  * if it no longer has associated state and didn't have to sleep.
5482  * Unlocks State before a non-zero value is returned.
5483  */
5484 static int
5485 nfsrv_cleandeleg(vnode_t vp, struct nfslockfile *lfp,
5486     struct nfsclient *clp, int *haslockp, NFSPROC_T *p)
5487 {
5488 	struct nfsstate *stp, *nstp;
5489 	int ret = 0;
5490 
5491 	stp = LIST_FIRST(&lfp->lf_deleg);
5492 	while (stp != LIST_END(&lfp->lf_deleg)) {
5493 		nstp = LIST_NEXT(stp, ls_file);
5494 		if (stp->ls_clp != clp) {
5495 			ret = nfsrv_delegconflict(stp, haslockp, p, vp);
5496 			if (ret) {
5497 				/*
5498 				 * nfsrv_delegconflict() unlocks state
5499 				 * when it returns non-zero.
5500 				 */
5501 				goto out;
5502 			}
5503 		}
5504 		stp = nstp;
5505 	}
5506 out:
5507 	NFSEXITCODE(ret);
5508 	return (ret);
5509 }
5510 
5511 /*
5512  * There are certain operations that, when being done outside of NFSv4,
5513  * require that any NFSv4 delegation for the file be recalled.
5514  * This function is to be called for those cases:
5515  * VOP_RENAME() - When a delegation is being recalled for any reason,
5516  *	the client may have to do Opens against the server, using the file's
5517  *	final component name. If the file has been renamed on the server,
5518  *	that component name will be incorrect and the Open will fail.
5519  * VOP_REMOVE() - Theoretically, a client could Open a file after it has
5520  *	been removed on the server, if there is a delegation issued to
5521  *	that client for the file. I say "theoretically" since clients
5522  *	normally do an Access Op before the Open and that Access Op will
5523  *	fail with ESTALE. Note that NFSv2 and 3 don't even do Opens, so
5524  *	they will detect the file's removal in the same manner. (There is
5525  *	one case where RFC3530 allows a client to do an Open without first
5526  *	doing an Access Op, which is passage of a check against the ACE
5527  *	returned with a Write delegation, but current practice is to ignore
5528  *	the ACE and always do an Access Op.)
5529  *	Since the functions can only be called with an unlocked vnode, this
5530  *	can't be done at this time.
5531  * VOP_ADVLOCK() - When a client holds a delegation, it can issue byte range
5532  *	locks locally in the client, which are not visible to the server. To
5533  *	deal with this, issuing of delegations for a vnode must be disabled
5534  *	and all delegations for the vnode recalled. This is done via the
5535  *	second function, using the VV_DISABLEDELEG vflag on the vnode.
5536  */
5537 APPLESTATIC void
5538 nfsd_recalldelegation(vnode_t vp, NFSPROC_T *p)
5539 {
5540 	time_t starttime;
5541 	int error;
5542 
5543 	/*
5544 	 * First, check to see if the server is currently running and it has
5545 	 * been called for a regular file when issuing delegations.
5546 	 */
5547 	if (newnfs_numnfsd == 0 || vp->v_type != VREG ||
5548 	    nfsrv_issuedelegs == 0)
5549 		return;
5550 
5551 	KASSERT((NFSVOPISLOCKED(vp) != LK_EXCLUSIVE), ("vp %p is locked", vp));
5552 	/*
5553 	 * First, get a reference on the nfsv4rootfs_lock so that an
5554 	 * exclusive lock cannot be acquired by another thread.
5555 	 */
5556 	NFSLOCKV4ROOTMUTEX();
5557 	nfsv4_getref(&nfsv4rootfs_lock, NULL, NFSV4ROOTLOCKMUTEXPTR, NULL);
5558 	NFSUNLOCKV4ROOTMUTEX();
5559 
5560 	/*
5561 	 * Now, call nfsrv_checkremove() in a loop while it returns
5562 	 * NFSERR_DELAY. Return upon any other error or when timed out.
5563 	 */
5564 	starttime = NFSD_MONOSEC;
5565 	do {
5566 		if (NFSVOPLOCK(vp, LK_EXCLUSIVE) == 0) {
5567 			error = nfsrv_checkremove(vp, 0, NULL,
5568 			    (nfsquad_t)((u_quad_t)0), p);
5569 			NFSVOPUNLOCK(vp);
5570 		} else
5571 			error = EPERM;
5572 		if (error == NFSERR_DELAY) {
5573 			if (NFSD_MONOSEC - starttime > NFS_REMOVETIMEO)
5574 				break;
5575 			/* Sleep for a short period of time */
5576 			(void) nfs_catnap(PZERO, 0, "nfsremove");
5577 		}
5578 	} while (error == NFSERR_DELAY);
5579 	NFSLOCKV4ROOTMUTEX();
5580 	nfsv4_relref(&nfsv4rootfs_lock);
5581 	NFSUNLOCKV4ROOTMUTEX();
5582 }
5583 
5584 APPLESTATIC void
5585 nfsd_disabledelegation(vnode_t vp, NFSPROC_T *p)
5586 {
5587 
5588 #ifdef VV_DISABLEDELEG
5589 	/*
5590 	 * First, flag issuance of delegations disabled.
5591 	 */
5592 	atomic_set_long(&vp->v_vflag, VV_DISABLEDELEG);
5593 #endif
5594 
5595 	/*
5596 	 * Then call nfsd_recalldelegation() to get rid of all extant
5597 	 * delegations.
5598 	 */
5599 	nfsd_recalldelegation(vp, p);
5600 }
5601 
5602 /*
5603  * Check for conflicting locks, etc. and then get rid of delegations.
5604  * (At one point I thought that I should get rid of delegations for any
5605  *  Setattr, since it could potentially disallow the I/O op (read or write)
5606  *  allowed by the delegation. However, Setattr Ops that aren't changing
5607  *  the size get a stateid of all 0s, so you can't tell if it is a delegation
5608  *  for the same client or a different one, so I decided to only get rid
5609  *  of delegations for other clients when the size is being changed.)
5610  * In general, a Setattr can disable NFS I/O Ops that are outstanding, such
5611  * as Write backs, even if there is no delegation, so it really isn't any
5612  * different?)
5613  */
5614 APPLESTATIC int
5615 nfsrv_checksetattr(vnode_t vp, struct nfsrv_descript *nd,
5616     nfsv4stateid_t *stateidp, struct nfsvattr *nvap, nfsattrbit_t *attrbitp,
5617     struct nfsexstuff *exp, NFSPROC_T *p)
5618 {
5619 	struct nfsstate st, *stp = &st;
5620 	struct nfslock lo, *lop = &lo;
5621 	int error = 0;
5622 	nfsquad_t clientid;
5623 
5624 	if (NFSISSET_ATTRBIT(attrbitp, NFSATTRBIT_SIZE)) {
5625 		stp->ls_flags = (NFSLCK_CHECK | NFSLCK_WRITEACCESS);
5626 		lop->lo_first = nvap->na_size;
5627 	} else {
5628 		stp->ls_flags = 0;
5629 		lop->lo_first = 0;
5630 	}
5631 	if (NFSISSET_ATTRBIT(attrbitp, NFSATTRBIT_OWNER) ||
5632 	    NFSISSET_ATTRBIT(attrbitp, NFSATTRBIT_OWNERGROUP) ||
5633 	    NFSISSET_ATTRBIT(attrbitp, NFSATTRBIT_MODE) ||
5634 	    NFSISSET_ATTRBIT(attrbitp, NFSATTRBIT_ACL))
5635 		stp->ls_flags |= NFSLCK_SETATTR;
5636 	if (stp->ls_flags == 0)
5637 		goto out;
5638 	lop->lo_end = NFS64BITSSET;
5639 	lop->lo_flags = NFSLCK_WRITE;
5640 	stp->ls_ownerlen = 0;
5641 	stp->ls_op = NULL;
5642 	stp->ls_uid = nd->nd_cred->cr_uid;
5643 	stp->ls_stateid.seqid = stateidp->seqid;
5644 	clientid.lval[0] = stp->ls_stateid.other[0] = stateidp->other[0];
5645 	clientid.lval[1] = stp->ls_stateid.other[1] = stateidp->other[1];
5646 	stp->ls_stateid.other[2] = stateidp->other[2];
5647 	error = nfsrv_lockctrl(vp, &stp, &lop, NULL, clientid,
5648 	    stateidp, exp, nd, p);
5649 
5650 out:
5651 	NFSEXITCODE2(error, nd);
5652 	return (error);
5653 }
5654 
5655 /*
5656  * Check for a write delegation and do a CBGETATTR if there is one, updating
5657  * the attributes, as required.
5658  * Should I return an error if I can't get the attributes? (For now, I'll
5659  * just return ok.
5660  */
5661 APPLESTATIC int
5662 nfsrv_checkgetattr(struct nfsrv_descript *nd, vnode_t vp,
5663     struct nfsvattr *nvap, nfsattrbit_t *attrbitp, NFSPROC_T *p)
5664 {
5665 	struct nfsstate *stp;
5666 	struct nfslockfile *lfp;
5667 	struct nfsclient *clp;
5668 	struct nfsvattr nva;
5669 	fhandle_t nfh;
5670 	int error = 0;
5671 	nfsattrbit_t cbbits;
5672 	u_quad_t delegfilerev;
5673 
5674 	NFSCBGETATTR_ATTRBIT(attrbitp, &cbbits);
5675 	if (!NFSNONZERO_ATTRBIT(&cbbits))
5676 		goto out;
5677 	if (nfsrv_writedelegcnt == 0)
5678 		goto out;
5679 
5680 	/*
5681 	 * Get the lock file structure.
5682 	 * (A return of -1 means no associated state, so return ok.)
5683 	 */
5684 	error = nfsrv_getlockfh(vp, NFSLCK_CHECK, NULL, &nfh, p);
5685 	NFSLOCKSTATE();
5686 	if (!error)
5687 		error = nfsrv_getlockfile(NFSLCK_CHECK, NULL, &lfp, &nfh, 0);
5688 	if (error) {
5689 		NFSUNLOCKSTATE();
5690 		if (error == -1)
5691 			error = 0;
5692 		goto out;
5693 	}
5694 
5695 	/*
5696 	 * Now, look for a write delegation.
5697 	 */
5698 	LIST_FOREACH(stp, &lfp->lf_deleg, ls_file) {
5699 		if (stp->ls_flags & NFSLCK_DELEGWRITE)
5700 			break;
5701 	}
5702 	if (stp == LIST_END(&lfp->lf_deleg)) {
5703 		NFSUNLOCKSTATE();
5704 		goto out;
5705 	}
5706 	clp = stp->ls_clp;
5707 	delegfilerev = stp->ls_filerev;
5708 
5709 	/*
5710 	 * If the Write delegation was issued as a part of this Compound RPC
5711 	 * or if we have an Implied Clientid (used in a previous Op in this
5712 	 * compound) and it is the client the delegation was issued to,
5713 	 * just return ok.
5714 	 * I also assume that it is from the same client iff the network
5715 	 * host IP address is the same as the callback address. (Not
5716 	 * exactly correct by the RFC, but avoids a lot of Getattr
5717 	 * callbacks.)
5718 	 */
5719 	if (nd->nd_compref == stp->ls_compref ||
5720 	    ((nd->nd_flag & ND_IMPLIEDCLID) &&
5721 	     clp->lc_clientid.qval == nd->nd_clientid.qval) ||
5722 	     nfsaddr2_match(clp->lc_req.nr_nam, nd->nd_nam)) {
5723 		NFSUNLOCKSTATE();
5724 		goto out;
5725 	}
5726 
5727 	/*
5728 	 * We are now done with the delegation state structure,
5729 	 * so the statelock can be released and we can now tsleep().
5730 	 */
5731 
5732 	/*
5733 	 * Now, we must do the CB Getattr callback, to see if Change or Size
5734 	 * has changed.
5735 	 */
5736 	if (clp->lc_expiry >= NFSD_MONOSEC) {
5737 		NFSUNLOCKSTATE();
5738 		NFSVNO_ATTRINIT(&nva);
5739 		nva.na_filerev = NFS64BITSSET;
5740 		error = nfsrv_docallback(clp, NFSV4OP_CBGETATTR, NULL,
5741 		    0, &nfh, &nva, &cbbits, 0, p);
5742 		if (!error) {
5743 			if ((nva.na_filerev != NFS64BITSSET &&
5744 			    nva.na_filerev > delegfilerev) ||
5745 			    (NFSVNO_ISSETSIZE(&nva) &&
5746 			     nva.na_size != nvap->na_size)) {
5747 				error = nfsvno_updfilerev(vp, nvap, nd, p);
5748 				if (NFSVNO_ISSETSIZE(&nva))
5749 					nvap->na_size = nva.na_size;
5750 			}
5751 		} else
5752 			error = 0;	/* Ignore callback errors for now. */
5753 	} else {
5754 		NFSUNLOCKSTATE();
5755 	}
5756 
5757 out:
5758 	NFSEXITCODE2(error, nd);
5759 	return (error);
5760 }
5761 
5762 /*
5763  * This function looks for openowners that haven't had any opens for
5764  * a while and throws them away. Called by an nfsd when NFSNSF_NOOPENS
5765  * is set.
5766  */
5767 APPLESTATIC void
5768 nfsrv_throwawayopens(NFSPROC_T *p)
5769 {
5770 	struct nfsclient *clp, *nclp;
5771 	struct nfsstate *stp, *nstp;
5772 	int i;
5773 
5774 	NFSLOCKSTATE();
5775 	nfsrv_stablefirst.nsf_flags &= ~NFSNSF_NOOPENS;
5776 	/*
5777 	 * For each client...
5778 	 */
5779 	for (i = 0; i < nfsrv_clienthashsize; i++) {
5780 	    LIST_FOREACH_SAFE(clp, &nfsclienthash[i], lc_hash, nclp) {
5781 		LIST_FOREACH_SAFE(stp, &clp->lc_open, ls_list, nstp) {
5782 			if (LIST_EMPTY(&stp->ls_open) &&
5783 			    (stp->ls_noopens > NFSNOOPEN ||
5784 			     (nfsrv_openpluslock * 2) >
5785 			     nfsrv_v4statelimit))
5786 				nfsrv_freeopenowner(stp, 0, p);
5787 		}
5788 	    }
5789 	}
5790 	NFSUNLOCKSTATE();
5791 }
5792 
5793 /*
5794  * This function checks to see if the credentials are the same.
5795  * Returns 1 for not same, 0 otherwise.
5796  */
5797 static int
5798 nfsrv_notsamecredname(struct nfsrv_descript *nd, struct nfsclient *clp)
5799 {
5800 
5801 	if (nd->nd_flag & ND_GSS) {
5802 		if (!(clp->lc_flags & LCL_GSS))
5803 			return (1);
5804 		if (clp->lc_flags & LCL_NAME) {
5805 			if (nd->nd_princlen != clp->lc_namelen ||
5806 			    NFSBCMP(nd->nd_principal, clp->lc_name,
5807 				clp->lc_namelen))
5808 				return (1);
5809 			else
5810 				return (0);
5811 		}
5812 		if (nd->nd_cred->cr_uid == clp->lc_uid)
5813 			return (0);
5814 		else
5815 			return (1);
5816 	} else if (clp->lc_flags & LCL_GSS)
5817 		return (1);
5818 	/*
5819 	 * For AUTH_SYS, allow the same uid or root. (This is underspecified
5820 	 * in RFC3530, which talks about principals, but doesn't say anything
5821 	 * about uids for AUTH_SYS.)
5822 	 */
5823 	if (nd->nd_cred->cr_uid == clp->lc_uid || nd->nd_cred->cr_uid == 0)
5824 		return (0);
5825 	else
5826 		return (1);
5827 }
5828 
5829 /*
5830  * Calculate the lease expiry time.
5831  */
5832 static time_t
5833 nfsrv_leaseexpiry(void)
5834 {
5835 
5836 	if (nfsrv_stablefirst.nsf_eograce > NFSD_MONOSEC)
5837 		return (NFSD_MONOSEC + 2 * (nfsrv_lease + NFSRV_LEASEDELTA));
5838 	return (NFSD_MONOSEC + nfsrv_lease + NFSRV_LEASEDELTA);
5839 }
5840 
5841 /*
5842  * Delay the delegation timeout as far as ls_delegtimelimit, as required.
5843  */
5844 static void
5845 nfsrv_delaydelegtimeout(struct nfsstate *stp)
5846 {
5847 
5848 	if ((stp->ls_flags & NFSLCK_DELEGRECALL) == 0)
5849 		return;
5850 
5851 	if ((stp->ls_delegtime + 15) > NFSD_MONOSEC &&
5852 	    stp->ls_delegtime < stp->ls_delegtimelimit) {
5853 		stp->ls_delegtime += nfsrv_lease;
5854 		if (stp->ls_delegtime > stp->ls_delegtimelimit)
5855 			stp->ls_delegtime = stp->ls_delegtimelimit;
5856 	}
5857 }
5858 
5859 /*
5860  * This function checks to see if there is any other state associated
5861  * with the openowner for this Open.
5862  * It returns 1 if there is no other state, 0 otherwise.
5863  */
5864 static int
5865 nfsrv_nootherstate(struct nfsstate *stp)
5866 {
5867 	struct nfsstate *tstp;
5868 
5869 	LIST_FOREACH(tstp, &stp->ls_openowner->ls_open, ls_list) {
5870 		if (tstp != stp || !LIST_EMPTY(&tstp->ls_lock))
5871 			return (0);
5872 	}
5873 	return (1);
5874 }
5875 
5876 /*
5877  * Create a list of lock deltas (changes to local byte range locking
5878  * that can be rolled back using the list) and apply the changes via
5879  * nfsvno_advlock(). Optionally, lock the list. It is expected that either
5880  * the rollback or update function will be called after this.
5881  * It returns an error (and rolls back, as required), if any nfsvno_advlock()
5882  * call fails. If it returns an error, it will unlock the list.
5883  */
5884 static int
5885 nfsrv_locallock(vnode_t vp, struct nfslockfile *lfp, int flags,
5886     uint64_t first, uint64_t end, struct nfslockconflict *cfp, NFSPROC_T *p)
5887 {
5888 	struct nfslock *lop, *nlop;
5889 	int error = 0;
5890 
5891 	/* Loop through the list of locks. */
5892 	lop = LIST_FIRST(&lfp->lf_locallock);
5893 	while (first < end && lop != NULL) {
5894 		nlop = LIST_NEXT(lop, lo_lckowner);
5895 		if (first >= lop->lo_end) {
5896 			/* not there yet */
5897 			lop = nlop;
5898 		} else if (first < lop->lo_first) {
5899 			/* new one starts before entry in list */
5900 			if (end <= lop->lo_first) {
5901 				/* no overlap between old and new */
5902 				error = nfsrv_dolocal(vp, lfp, flags,
5903 				    NFSLCK_UNLOCK, first, end, cfp, p);
5904 				if (error != 0)
5905 					break;
5906 				first = end;
5907 			} else {
5908 				/* handle fragment overlapped with new one */
5909 				error = nfsrv_dolocal(vp, lfp, flags,
5910 				    NFSLCK_UNLOCK, first, lop->lo_first, cfp,
5911 				    p);
5912 				if (error != 0)
5913 					break;
5914 				first = lop->lo_first;
5915 			}
5916 		} else {
5917 			/* new one overlaps this entry in list */
5918 			if (end <= lop->lo_end) {
5919 				/* overlaps all of new one */
5920 				error = nfsrv_dolocal(vp, lfp, flags,
5921 				    lop->lo_flags, first, end, cfp, p);
5922 				if (error != 0)
5923 					break;
5924 				first = end;
5925 			} else {
5926 				/* handle fragment overlapped with new one */
5927 				error = nfsrv_dolocal(vp, lfp, flags,
5928 				    lop->lo_flags, first, lop->lo_end, cfp, p);
5929 				if (error != 0)
5930 					break;
5931 				first = lop->lo_end;
5932 				lop = nlop;
5933 			}
5934 		}
5935 	}
5936 	if (first < end && error == 0)
5937 		/* handle fragment past end of list */
5938 		error = nfsrv_dolocal(vp, lfp, flags, NFSLCK_UNLOCK, first,
5939 		    end, cfp, p);
5940 
5941 	NFSEXITCODE(error);
5942 	return (error);
5943 }
5944 
5945 /*
5946  * Local lock unlock. Unlock all byte ranges that are no longer locked
5947  * by NFSv4. To do this, unlock any subranges of first-->end that
5948  * do not overlap with the byte ranges of any lock in the lfp->lf_lock
5949  * list. This list has all locks for the file held by other
5950  * <clientid, lockowner> tuples. The list is ordered by increasing
5951  * lo_first value, but may have entries that overlap each other, for
5952  * the case of read locks.
5953  */
5954 static void
5955 nfsrv_localunlock(vnode_t vp, struct nfslockfile *lfp, uint64_t init_first,
5956     uint64_t init_end, NFSPROC_T *p)
5957 {
5958 	struct nfslock *lop;
5959 	uint64_t first, end, prevfirst __unused;
5960 
5961 	first = init_first;
5962 	end = init_end;
5963 	while (first < init_end) {
5964 		/* Loop through all nfs locks, adjusting first and end */
5965 		prevfirst = 0;
5966 		LIST_FOREACH(lop, &lfp->lf_lock, lo_lckfile) {
5967 			KASSERT(prevfirst <= lop->lo_first,
5968 			    ("nfsv4 locks out of order"));
5969 			KASSERT(lop->lo_first < lop->lo_end,
5970 			    ("nfsv4 bogus lock"));
5971 			prevfirst = lop->lo_first;
5972 			if (first >= lop->lo_first &&
5973 			    first < lop->lo_end)
5974 				/*
5975 				 * Overlaps with initial part, so trim
5976 				 * off that initial part by moving first past
5977 				 * it.
5978 				 */
5979 				first = lop->lo_end;
5980 			else if (end > lop->lo_first &&
5981 			    lop->lo_first > first) {
5982 				/*
5983 				 * This lock defines the end of the
5984 				 * segment to unlock, so set end to the
5985 				 * start of it and break out of the loop.
5986 				 */
5987 				end = lop->lo_first;
5988 				break;
5989 			}
5990 			if (first >= end)
5991 				/*
5992 				 * There is no segment left to do, so
5993 				 * break out of this loop and then exit
5994 				 * the outer while() since first will be set
5995 				 * to end, which must equal init_end here.
5996 				 */
5997 				break;
5998 		}
5999 		if (first < end) {
6000 			/* Unlock this segment */
6001 			(void) nfsrv_dolocal(vp, lfp, NFSLCK_UNLOCK,
6002 			    NFSLCK_READ, first, end, NULL, p);
6003 			nfsrv_locallock_commit(lfp, NFSLCK_UNLOCK,
6004 			    first, end);
6005 		}
6006 		/*
6007 		 * Now move past this segment and look for any further
6008 		 * segment in the range, if there is one.
6009 		 */
6010 		first = end;
6011 		end = init_end;
6012 	}
6013 }
6014 
6015 /*
6016  * Do the local lock operation and update the rollback list, as required.
6017  * Perform the rollback and return the error if nfsvno_advlock() fails.
6018  */
6019 static int
6020 nfsrv_dolocal(vnode_t vp, struct nfslockfile *lfp, int flags, int oldflags,
6021     uint64_t first, uint64_t end, struct nfslockconflict *cfp, NFSPROC_T *p)
6022 {
6023 	struct nfsrollback *rlp;
6024 	int error = 0, ltype, oldltype;
6025 
6026 	if (flags & NFSLCK_WRITE)
6027 		ltype = F_WRLCK;
6028 	else if (flags & NFSLCK_READ)
6029 		ltype = F_RDLCK;
6030 	else
6031 		ltype = F_UNLCK;
6032 	if (oldflags & NFSLCK_WRITE)
6033 		oldltype = F_WRLCK;
6034 	else if (oldflags & NFSLCK_READ)
6035 		oldltype = F_RDLCK;
6036 	else
6037 		oldltype = F_UNLCK;
6038 	if (ltype == oldltype || (oldltype == F_WRLCK && ltype == F_RDLCK))
6039 		/* nothing to do */
6040 		goto out;
6041 	error = nfsvno_advlock(vp, ltype, first, end, p);
6042 	if (error != 0) {
6043 		if (cfp != NULL) {
6044 			cfp->cl_clientid.lval[0] = 0;
6045 			cfp->cl_clientid.lval[1] = 0;
6046 			cfp->cl_first = 0;
6047 			cfp->cl_end = NFS64BITSSET;
6048 			cfp->cl_flags = NFSLCK_WRITE;
6049 			cfp->cl_ownerlen = 5;
6050 			NFSBCOPY("LOCAL", cfp->cl_owner, 5);
6051 		}
6052 		nfsrv_locallock_rollback(vp, lfp, p);
6053 	} else if (ltype != F_UNLCK) {
6054 		rlp = malloc(sizeof (struct nfsrollback), M_NFSDROLLBACK,
6055 		    M_WAITOK);
6056 		rlp->rlck_first = first;
6057 		rlp->rlck_end = end;
6058 		rlp->rlck_type = oldltype;
6059 		LIST_INSERT_HEAD(&lfp->lf_rollback, rlp, rlck_list);
6060 	}
6061 
6062 out:
6063 	NFSEXITCODE(error);
6064 	return (error);
6065 }
6066 
6067 /*
6068  * Roll back local lock changes and free up the rollback list.
6069  */
6070 static void
6071 nfsrv_locallock_rollback(vnode_t vp, struct nfslockfile *lfp, NFSPROC_T *p)
6072 {
6073 	struct nfsrollback *rlp, *nrlp;
6074 
6075 	LIST_FOREACH_SAFE(rlp, &lfp->lf_rollback, rlck_list, nrlp) {
6076 		(void) nfsvno_advlock(vp, rlp->rlck_type, rlp->rlck_first,
6077 		    rlp->rlck_end, p);
6078 		free(rlp, M_NFSDROLLBACK);
6079 	}
6080 	LIST_INIT(&lfp->lf_rollback);
6081 }
6082 
6083 /*
6084  * Update local lock list and delete rollback list (ie now committed to the
6085  * local locks). Most of the work is done by the internal function.
6086  */
6087 static void
6088 nfsrv_locallock_commit(struct nfslockfile *lfp, int flags, uint64_t first,
6089     uint64_t end)
6090 {
6091 	struct nfsrollback *rlp, *nrlp;
6092 	struct nfslock *new_lop, *other_lop;
6093 
6094 	new_lop = malloc(sizeof (struct nfslock), M_NFSDLOCK, M_WAITOK);
6095 	if (flags & (NFSLCK_READ | NFSLCK_WRITE))
6096 		other_lop = malloc(sizeof (struct nfslock), M_NFSDLOCK,
6097 		    M_WAITOK);
6098 	else
6099 		other_lop = NULL;
6100 	new_lop->lo_flags = flags;
6101 	new_lop->lo_first = first;
6102 	new_lop->lo_end = end;
6103 	nfsrv_updatelock(NULL, &new_lop, &other_lop, lfp);
6104 	if (new_lop != NULL)
6105 		free(new_lop, M_NFSDLOCK);
6106 	if (other_lop != NULL)
6107 		free(other_lop, M_NFSDLOCK);
6108 
6109 	/* and get rid of the rollback list */
6110 	LIST_FOREACH_SAFE(rlp, &lfp->lf_rollback, rlck_list, nrlp)
6111 		free(rlp, M_NFSDROLLBACK);
6112 	LIST_INIT(&lfp->lf_rollback);
6113 }
6114 
6115 /*
6116  * Lock the struct nfslockfile for local lock updating.
6117  */
6118 static void
6119 nfsrv_locklf(struct nfslockfile *lfp)
6120 {
6121 	int gotlock;
6122 
6123 	/* lf_usecount ensures *lfp won't be free'd */
6124 	lfp->lf_usecount++;
6125 	do {
6126 		gotlock = nfsv4_lock(&lfp->lf_locallock_lck, 1, NULL,
6127 		    NFSSTATEMUTEXPTR, NULL);
6128 	} while (gotlock == 0);
6129 	lfp->lf_usecount--;
6130 }
6131 
6132 /*
6133  * Unlock the struct nfslockfile after local lock updating.
6134  */
6135 static void
6136 nfsrv_unlocklf(struct nfslockfile *lfp)
6137 {
6138 
6139 	nfsv4_unlock(&lfp->lf_locallock_lck, 0);
6140 }
6141 
6142 /*
6143  * Clear out all state for the NFSv4 server.
6144  * Must be called by a thread that can sleep when no nfsds are running.
6145  */
6146 void
6147 nfsrv_throwawayallstate(NFSPROC_T *p)
6148 {
6149 	struct nfsclient *clp, *nclp;
6150 	struct nfslockfile *lfp, *nlfp;
6151 	int i;
6152 
6153 	/*
6154 	 * For each client, clean out the state and then free the structure.
6155 	 */
6156 	for (i = 0; i < nfsrv_clienthashsize; i++) {
6157 		LIST_FOREACH_SAFE(clp, &nfsclienthash[i], lc_hash, nclp) {
6158 			nfsrv_cleanclient(clp, p);
6159 			nfsrv_freedeleglist(&clp->lc_deleg);
6160 			nfsrv_freedeleglist(&clp->lc_olddeleg);
6161 			free(clp->lc_stateid, M_NFSDCLIENT);
6162 			free(clp, M_NFSDCLIENT);
6163 		}
6164 	}
6165 
6166 	/*
6167 	 * Also, free up any remaining lock file structures.
6168 	 */
6169 	for (i = 0; i < nfsrv_lockhashsize; i++) {
6170 		LIST_FOREACH_SAFE(lfp, &nfslockhash[i], lf_hash, nlfp) {
6171 			printf("nfsd unload: fnd a lock file struct\n");
6172 			nfsrv_freenfslockfile(lfp);
6173 		}
6174 	}
6175 
6176 	/* And get rid of the deviceid structures and layouts. */
6177 	nfsrv_freealllayoutsanddevids();
6178 }
6179 
6180 /*
6181  * Check the sequence# for the session and slot provided as an argument.
6182  * Also, renew the lease if the session will return NFS_OK.
6183  */
6184 int
6185 nfsrv_checksequence(struct nfsrv_descript *nd, uint32_t sequenceid,
6186     uint32_t *highest_slotidp, uint32_t *target_highest_slotidp, int cache_this,
6187     uint32_t *sflagsp, NFSPROC_T *p)
6188 {
6189 	struct nfsdsession *sep;
6190 	struct nfssessionhash *shp;
6191 	int error;
6192 	SVCXPRT *savxprt;
6193 
6194 	shp = NFSSESSIONHASH(nd->nd_sessionid);
6195 	NFSLOCKSESSION(shp);
6196 	sep = nfsrv_findsession(nd->nd_sessionid);
6197 	if (sep == NULL) {
6198 		NFSUNLOCKSESSION(shp);
6199 		return (NFSERR_BADSESSION);
6200 	}
6201 	error = nfsv4_seqsession(sequenceid, nd->nd_slotid, *highest_slotidp,
6202 	    sep->sess_slots, NULL, NFSV4_SLOTS - 1);
6203 	if (error != 0) {
6204 		NFSUNLOCKSESSION(shp);
6205 		return (error);
6206 	}
6207 	if (cache_this != 0)
6208 		nd->nd_flag |= ND_SAVEREPLY;
6209 	/* Renew the lease. */
6210 	sep->sess_clp->lc_expiry = nfsrv_leaseexpiry();
6211 	nd->nd_clientid.qval = sep->sess_clp->lc_clientid.qval;
6212 	nd->nd_flag |= ND_IMPLIEDCLID;
6213 
6214 	/* Save maximum request and reply sizes. */
6215 	nd->nd_maxreq = sep->sess_maxreq;
6216 	nd->nd_maxresp = sep->sess_maxresp;
6217 
6218 	/*
6219 	 * If this session handles the backchannel, save the nd_xprt for this
6220 	 * RPC, since this is the one being used.
6221 	 * RFC-5661 specifies that the fore channel will be implicitly
6222 	 * bound by a Sequence operation.  However, since some NFSv4.1 clients
6223 	 * erroneously assumed that the back channel would be implicitly
6224 	 * bound as well, do the implicit binding unless a
6225 	 * BindConnectiontoSession has already been done on the session.
6226 	 */
6227 	if (sep->sess_clp->lc_req.nr_client != NULL &&
6228 	    sep->sess_cbsess.nfsess_xprt != nd->nd_xprt &&
6229 	    (sep->sess_crflags & NFSV4CRSESS_CONNBACKCHAN) != 0 &&
6230 	    (sep->sess_clp->lc_flags & LCL_DONEBINDCONN) == 0) {
6231 		NFSD_DEBUG(2,
6232 		    "nfsrv_checksequence: implicit back channel bind\n");
6233 		savxprt = sep->sess_cbsess.nfsess_xprt;
6234 		SVC_ACQUIRE(nd->nd_xprt);
6235 		nd->nd_xprt->xp_p2 =
6236 		    sep->sess_clp->lc_req.nr_client->cl_private;
6237 		nd->nd_xprt->xp_idletimeout = 0;	/* Disable timeout. */
6238 		sep->sess_cbsess.nfsess_xprt = nd->nd_xprt;
6239 		if (savxprt != NULL)
6240 			SVC_RELEASE(savxprt);
6241 	}
6242 
6243 	*sflagsp = 0;
6244 	if (sep->sess_clp->lc_req.nr_client == NULL)
6245 		*sflagsp |= NFSV4SEQ_CBPATHDOWN;
6246 	NFSUNLOCKSESSION(shp);
6247 	if (error == NFSERR_EXPIRED) {
6248 		*sflagsp |= NFSV4SEQ_EXPIREDALLSTATEREVOKED;
6249 		error = 0;
6250 	} else if (error == NFSERR_ADMINREVOKED) {
6251 		*sflagsp |= NFSV4SEQ_ADMINSTATEREVOKED;
6252 		error = 0;
6253 	}
6254 	*highest_slotidp = *target_highest_slotidp = NFSV4_SLOTS - 1;
6255 	return (0);
6256 }
6257 
6258 /*
6259  * Check/set reclaim complete for this session/clientid.
6260  */
6261 int
6262 nfsrv_checkreclaimcomplete(struct nfsrv_descript *nd, int onefs)
6263 {
6264 	struct nfsdsession *sep;
6265 	struct nfssessionhash *shp;
6266 	int error = 0;
6267 
6268 	shp = NFSSESSIONHASH(nd->nd_sessionid);
6269 	NFSLOCKSTATE();
6270 	NFSLOCKSESSION(shp);
6271 	sep = nfsrv_findsession(nd->nd_sessionid);
6272 	if (sep == NULL) {
6273 		NFSUNLOCKSESSION(shp);
6274 		NFSUNLOCKSTATE();
6275 		return (NFSERR_BADSESSION);
6276 	}
6277 
6278 	if (onefs != 0)
6279 		sep->sess_clp->lc_flags |= LCL_RECLAIMONEFS;
6280 		/* Check to see if reclaim complete has already happened. */
6281 	else if ((sep->sess_clp->lc_flags & LCL_RECLAIMCOMPLETE) != 0)
6282 		error = NFSERR_COMPLETEALREADY;
6283 	else {
6284 		sep->sess_clp->lc_flags |= LCL_RECLAIMCOMPLETE;
6285 		nfsrv_markreclaim(sep->sess_clp);
6286 	}
6287 	NFSUNLOCKSESSION(shp);
6288 	NFSUNLOCKSTATE();
6289 	return (error);
6290 }
6291 
6292 /*
6293  * Cache the reply in a session slot.
6294  */
6295 void
6296 nfsrv_cache_session(uint8_t *sessionid, uint32_t slotid, int repstat,
6297    struct mbuf **m)
6298 {
6299 	struct nfsdsession *sep;
6300 	struct nfssessionhash *shp;
6301 
6302 	shp = NFSSESSIONHASH(sessionid);
6303 	NFSLOCKSESSION(shp);
6304 	sep = nfsrv_findsession(sessionid);
6305 	if (sep == NULL) {
6306 		NFSUNLOCKSESSION(shp);
6307 		printf("nfsrv_cache_session: no session\n");
6308 		m_freem(*m);
6309 		return;
6310 	}
6311 	nfsv4_seqsess_cacherep(slotid, sep->sess_slots, repstat, m);
6312 	NFSUNLOCKSESSION(shp);
6313 }
6314 
6315 /*
6316  * Search for a session that matches the sessionid.
6317  */
6318 static struct nfsdsession *
6319 nfsrv_findsession(uint8_t *sessionid)
6320 {
6321 	struct nfsdsession *sep;
6322 	struct nfssessionhash *shp;
6323 
6324 	shp = NFSSESSIONHASH(sessionid);
6325 	LIST_FOREACH(sep, &shp->list, sess_hash) {
6326 		if (!NFSBCMP(sessionid, sep->sess_sessionid, NFSX_V4SESSIONID))
6327 			break;
6328 	}
6329 	return (sep);
6330 }
6331 
6332 /*
6333  * Destroy a session.
6334  */
6335 int
6336 nfsrv_destroysession(struct nfsrv_descript *nd, uint8_t *sessionid)
6337 {
6338 	int error, igotlock, samesess;
6339 
6340 	samesess = 0;
6341 	if (!NFSBCMP(sessionid, nd->nd_sessionid, NFSX_V4SESSIONID) &&
6342 	    (nd->nd_flag & ND_HASSEQUENCE) != 0) {
6343 		samesess = 1;
6344 		if ((nd->nd_flag & ND_LASTOP) == 0)
6345 			return (NFSERR_BADSESSION);
6346 	}
6347 
6348 	/* Lock out other nfsd threads */
6349 	NFSLOCKV4ROOTMUTEX();
6350 	nfsv4_relref(&nfsv4rootfs_lock);
6351 	do {
6352 		igotlock = nfsv4_lock(&nfsv4rootfs_lock, 1, NULL,
6353 		    NFSV4ROOTLOCKMUTEXPTR, NULL);
6354 	} while (igotlock == 0);
6355 	NFSUNLOCKV4ROOTMUTEX();
6356 
6357 	error = nfsrv_freesession(NULL, sessionid);
6358 	if (error == 0 && samesess != 0)
6359 		nd->nd_flag &= ~ND_HASSEQUENCE;
6360 
6361 	NFSLOCKV4ROOTMUTEX();
6362 	nfsv4_unlock(&nfsv4rootfs_lock, 1);
6363 	NFSUNLOCKV4ROOTMUTEX();
6364 	return (error);
6365 }
6366 
6367 /*
6368  * Bind a connection to a session.
6369  * For now, only certain variants are supported, since the current session
6370  * structure can only handle a single backchannel entry, which will be
6371  * applied to all connections if it is set.
6372  */
6373 int
6374 nfsrv_bindconnsess(struct nfsrv_descript *nd, uint8_t *sessionid, int *foreaftp)
6375 {
6376 	struct nfssessionhash *shp;
6377 	struct nfsdsession *sep;
6378 	struct nfsclient *clp;
6379 	SVCXPRT *savxprt;
6380 	int error;
6381 
6382 	error = 0;
6383 	shp = NFSSESSIONHASH(sessionid);
6384 	NFSLOCKSTATE();
6385 	NFSLOCKSESSION(shp);
6386 	sep = nfsrv_findsession(sessionid);
6387 	if (sep != NULL) {
6388 		clp = sep->sess_clp;
6389 		if (*foreaftp == NFSCDFC4_BACK ||
6390 		    *foreaftp == NFSCDFC4_BACK_OR_BOTH ||
6391 		    *foreaftp == NFSCDFC4_FORE_OR_BOTH) {
6392 			/* Try to set up a backchannel. */
6393 			if (clp->lc_req.nr_client == NULL) {
6394 				NFSD_DEBUG(2, "nfsrv_bindconnsess: acquire "
6395 				    "backchannel\n");
6396 				clp->lc_req.nr_client = (struct __rpc_client *)
6397 				    clnt_bck_create(nd->nd_xprt->xp_socket,
6398 				    sep->sess_cbprogram, NFSV4_CBVERS);
6399 			}
6400 			if (clp->lc_req.nr_client != NULL) {
6401 				NFSD_DEBUG(2, "nfsrv_bindconnsess: set up "
6402 				    "backchannel\n");
6403 				savxprt = sep->sess_cbsess.nfsess_xprt;
6404 				SVC_ACQUIRE(nd->nd_xprt);
6405 				nd->nd_xprt->xp_p2 =
6406 				    clp->lc_req.nr_client->cl_private;
6407 				/* Disable idle timeout. */
6408 				nd->nd_xprt->xp_idletimeout = 0;
6409 				sep->sess_cbsess.nfsess_xprt = nd->nd_xprt;
6410 				if (savxprt != NULL)
6411 					SVC_RELEASE(savxprt);
6412 				sep->sess_crflags |= NFSV4CRSESS_CONNBACKCHAN;
6413 				clp->lc_flags |= LCL_DONEBINDCONN;
6414 				if (*foreaftp == NFSCDFS4_BACK)
6415 					*foreaftp = NFSCDFS4_BACK;
6416 				else
6417 					*foreaftp = NFSCDFS4_BOTH;
6418 			} else if (*foreaftp != NFSCDFC4_BACK) {
6419 				NFSD_DEBUG(2, "nfsrv_bindconnsess: can't set "
6420 				    "up backchannel\n");
6421 				sep->sess_crflags &= ~NFSV4CRSESS_CONNBACKCHAN;
6422 				clp->lc_flags |= LCL_DONEBINDCONN;
6423 				*foreaftp = NFSCDFS4_FORE;
6424 			} else {
6425 				error = NFSERR_NOTSUPP;
6426 				printf("nfsrv_bindconnsess: Can't add "
6427 				    "backchannel\n");
6428 			}
6429 		} else {
6430 			NFSD_DEBUG(2, "nfsrv_bindconnsess: Set forechannel\n");
6431 			clp->lc_flags |= LCL_DONEBINDCONN;
6432 			*foreaftp = NFSCDFS4_FORE;
6433 		}
6434 	} else
6435 		error = NFSERR_BADSESSION;
6436 	NFSUNLOCKSESSION(shp);
6437 	NFSUNLOCKSTATE();
6438 	return (error);
6439 }
6440 
6441 /*
6442  * Free up a session structure.
6443  */
6444 static int
6445 nfsrv_freesession(struct nfsdsession *sep, uint8_t *sessionid)
6446 {
6447 	struct nfssessionhash *shp;
6448 	int i;
6449 
6450 	NFSLOCKSTATE();
6451 	if (sep == NULL) {
6452 		shp = NFSSESSIONHASH(sessionid);
6453 		NFSLOCKSESSION(shp);
6454 		sep = nfsrv_findsession(sessionid);
6455 	} else {
6456 		shp = NFSSESSIONHASH(sep->sess_sessionid);
6457 		NFSLOCKSESSION(shp);
6458 	}
6459 	if (sep != NULL) {
6460 		sep->sess_refcnt--;
6461 		if (sep->sess_refcnt > 0) {
6462 			NFSUNLOCKSESSION(shp);
6463 			NFSUNLOCKSTATE();
6464 			return (NFSERR_BACKCHANBUSY);
6465 		}
6466 		LIST_REMOVE(sep, sess_hash);
6467 		LIST_REMOVE(sep, sess_list);
6468 	}
6469 	NFSUNLOCKSESSION(shp);
6470 	NFSUNLOCKSTATE();
6471 	if (sep == NULL)
6472 		return (NFSERR_BADSESSION);
6473 	for (i = 0; i < NFSV4_SLOTS; i++)
6474 		if (sep->sess_slots[i].nfssl_reply != NULL)
6475 			m_freem(sep->sess_slots[i].nfssl_reply);
6476 	if (sep->sess_cbsess.nfsess_xprt != NULL)
6477 		SVC_RELEASE(sep->sess_cbsess.nfsess_xprt);
6478 	free(sep, M_NFSDSESSION);
6479 	return (0);
6480 }
6481 
6482 /*
6483  * Free a stateid.
6484  * RFC5661 says that it should fail when there are associated opens, locks
6485  * or delegations. Since stateids represent opens, I don't see how you can
6486  * free an open stateid (it will be free'd when closed), so this function
6487  * only works for lock stateids (freeing the lock_owner) or delegations.
6488  */
6489 int
6490 nfsrv_freestateid(struct nfsrv_descript *nd, nfsv4stateid_t *stateidp,
6491     NFSPROC_T *p)
6492 {
6493 	struct nfsclient *clp;
6494 	struct nfsstate *stp;
6495 	int error;
6496 
6497 	NFSLOCKSTATE();
6498 	/*
6499 	 * Look up the stateid
6500 	 */
6501 	error = nfsrv_getclient((nfsquad_t)((u_quad_t)0), CLOPS_RENEW, &clp,
6502 	    NULL, (nfsquad_t)((u_quad_t)0), 0, nd, p);
6503 	if (error == 0) {
6504 		/* First, check for a delegation. */
6505 		LIST_FOREACH(stp, &clp->lc_deleg, ls_list) {
6506 			if (!NFSBCMP(stp->ls_stateid.other, stateidp->other,
6507 			    NFSX_STATEIDOTHER))
6508 				break;
6509 		}
6510 		if (stp != NULL) {
6511 			nfsrv_freedeleg(stp);
6512 			NFSUNLOCKSTATE();
6513 			return (error);
6514 		}
6515 	}
6516 	/* Not a delegation, try for a lock_owner. */
6517 	if (error == 0)
6518 		error = nfsrv_getstate(clp, stateidp, 0, &stp);
6519 	if (error == 0 && ((stp->ls_flags & (NFSLCK_OPEN | NFSLCK_DELEGREAD |
6520 	    NFSLCK_DELEGWRITE)) != 0 || (stp->ls_flags & NFSLCK_LOCK) == 0))
6521 		/* Not a lock_owner stateid. */
6522 		error = NFSERR_LOCKSHELD;
6523 	if (error == 0 && !LIST_EMPTY(&stp->ls_lock))
6524 		error = NFSERR_LOCKSHELD;
6525 	if (error == 0)
6526 		nfsrv_freelockowner(stp, NULL, 0, p);
6527 	NFSUNLOCKSTATE();
6528 	return (error);
6529 }
6530 
6531 /*
6532  * Test a stateid.
6533  */
6534 int
6535 nfsrv_teststateid(struct nfsrv_descript *nd, nfsv4stateid_t *stateidp,
6536     NFSPROC_T *p)
6537 {
6538 	struct nfsclient *clp;
6539 	struct nfsstate *stp;
6540 	int error;
6541 
6542 	NFSLOCKSTATE();
6543 	/*
6544 	 * Look up the stateid
6545 	 */
6546 	error = nfsrv_getclient((nfsquad_t)((u_quad_t)0), CLOPS_RENEW, &clp,
6547 	    NULL, (nfsquad_t)((u_quad_t)0), 0, nd, p);
6548 	if (error == 0)
6549 		error = nfsrv_getstate(clp, stateidp, 0, &stp);
6550 	if (error == 0 && stateidp->seqid != 0 &&
6551 	    SEQ_LT(stateidp->seqid, stp->ls_stateid.seqid))
6552 		error = NFSERR_OLDSTATEID;
6553 	NFSUNLOCKSTATE();
6554 	return (error);
6555 }
6556 
6557 /*
6558  * Generate the xdr for an NFSv4.1 CBSequence Operation.
6559  */
6560 static int
6561 nfsv4_setcbsequence(struct nfsrv_descript *nd, struct nfsclient *clp,
6562     int dont_replycache, struct nfsdsession **sepp)
6563 {
6564 	struct nfsdsession *sep;
6565 	uint32_t *tl, slotseq = 0;
6566 	int maxslot, slotpos;
6567 	uint8_t sessionid[NFSX_V4SESSIONID];
6568 	int error;
6569 
6570 	error = nfsv4_getcbsession(clp, sepp);
6571 	if (error != 0)
6572 		return (error);
6573 	sep = *sepp;
6574 	(void)nfsv4_sequencelookup(NULL, &sep->sess_cbsess, &slotpos, &maxslot,
6575 	    &slotseq, sessionid);
6576 	KASSERT(maxslot >= 0, ("nfsv4_setcbsequence neg maxslot"));
6577 
6578 	/* Build the Sequence arguments. */
6579 	NFSM_BUILD(tl, uint32_t *, NFSX_V4SESSIONID + 5 * NFSX_UNSIGNED);
6580 	bcopy(sessionid, tl, NFSX_V4SESSIONID);
6581 	tl += NFSX_V4SESSIONID / NFSX_UNSIGNED;
6582 	nd->nd_slotseq = tl;
6583 	*tl++ = txdr_unsigned(slotseq);
6584 	*tl++ = txdr_unsigned(slotpos);
6585 	*tl++ = txdr_unsigned(maxslot);
6586 	if (dont_replycache == 0)
6587 		*tl++ = newnfs_true;
6588 	else
6589 		*tl++ = newnfs_false;
6590 	*tl = 0;			/* No referring call list, for now. */
6591 	nd->nd_flag |= ND_HASSEQUENCE;
6592 	return (0);
6593 }
6594 
6595 /*
6596  * Get a session for the callback.
6597  */
6598 static int
6599 nfsv4_getcbsession(struct nfsclient *clp, struct nfsdsession **sepp)
6600 {
6601 	struct nfsdsession *sep;
6602 
6603 	NFSLOCKSTATE();
6604 	LIST_FOREACH(sep, &clp->lc_session, sess_list) {
6605 		if ((sep->sess_crflags & NFSV4CRSESS_CONNBACKCHAN) != 0)
6606 			break;
6607 	}
6608 	if (sep == NULL) {
6609 		NFSUNLOCKSTATE();
6610 		return (NFSERR_BADSESSION);
6611 	}
6612 	sep->sess_refcnt++;
6613 	*sepp = sep;
6614 	NFSUNLOCKSTATE();
6615 	return (0);
6616 }
6617 
6618 /*
6619  * Free up all backchannel xprts.  This needs to be done when the nfsd threads
6620  * exit, since those transports will all be going away.
6621  * This is only called after all the nfsd threads are done performing RPCs,
6622  * so locking shouldn't be an issue.
6623  */
6624 APPLESTATIC void
6625 nfsrv_freeallbackchannel_xprts(void)
6626 {
6627 	struct nfsdsession *sep;
6628 	struct nfsclient *clp;
6629 	SVCXPRT *xprt;
6630 	int i;
6631 
6632 	for (i = 0; i < nfsrv_clienthashsize; i++) {
6633 		LIST_FOREACH(clp, &nfsclienthash[i], lc_hash) {
6634 			LIST_FOREACH(sep, &clp->lc_session, sess_list) {
6635 				xprt = sep->sess_cbsess.nfsess_xprt;
6636 				sep->sess_cbsess.nfsess_xprt = NULL;
6637 				if (xprt != NULL)
6638 					SVC_RELEASE(xprt);
6639 			}
6640 		}
6641 	}
6642 }
6643 
6644 /*
6645  * Do a layout commit.  Actually just call nfsrv_updatemdsattr().
6646  * I have no idea if the rest of these arguments will ever be useful?
6647  */
6648 int
6649 nfsrv_layoutcommit(struct nfsrv_descript *nd, vnode_t vp, int layouttype,
6650     int hasnewoff, uint64_t newoff, uint64_t offset, uint64_t len,
6651     int hasnewmtime, struct timespec *newmtimep, int reclaim,
6652     nfsv4stateid_t *stateidp, int maxcnt, char *layp, int *hasnewsizep,
6653     uint64_t *newsizep, struct ucred *cred, NFSPROC_T *p)
6654 {
6655 	struct nfsvattr na;
6656 	int error;
6657 
6658 	error = nfsrv_updatemdsattr(vp, &na, p);
6659 	if (error == 0) {
6660 		*hasnewsizep = 1;
6661 		*newsizep = na.na_size;
6662 	}
6663 	return (error);
6664 }
6665 
6666 /*
6667  * Try and get a layout.
6668  */
6669 int
6670 nfsrv_layoutget(struct nfsrv_descript *nd, vnode_t vp, struct nfsexstuff *exp,
6671     int layouttype, int *iomode, uint64_t *offset, uint64_t *len,
6672     uint64_t minlen, nfsv4stateid_t *stateidp, int maxcnt, int *retonclose,
6673     int *layoutlenp, char *layp, struct ucred *cred, NFSPROC_T *p)
6674 {
6675 	struct nfslayouthash *lhyp;
6676 	struct nfslayout *lyp;
6677 	char *devid;
6678 	fhandle_t fh, *dsfhp;
6679 	int error, mirrorcnt;
6680 
6681 	if (nfsrv_devidcnt == 0)
6682 		return (NFSERR_UNKNLAYOUTTYPE);
6683 
6684 	if (*offset != 0)
6685 		printf("nfsrv_layoutget: off=%ju len=%ju\n", (uintmax_t)*offset,
6686 		    (uintmax_t)*len);
6687 	error = nfsvno_getfh(vp, &fh, p);
6688 	NFSD_DEBUG(4, "layoutget getfh=%d\n", error);
6689 	if (error != 0)
6690 		return (error);
6691 
6692 	/*
6693 	 * For now, all layouts are for entire files.
6694 	 * Only issue Read/Write layouts if requested for a non-readonly fs.
6695 	 */
6696 	if (NFSVNO_EXRDONLY(exp)) {
6697 		if (*iomode == NFSLAYOUTIOMODE_RW)
6698 			return (NFSERR_LAYOUTTRYLATER);
6699 		*iomode = NFSLAYOUTIOMODE_READ;
6700 	}
6701 	if (*iomode != NFSLAYOUTIOMODE_RW)
6702 		*iomode = NFSLAYOUTIOMODE_READ;
6703 
6704 	/*
6705 	 * Check to see if a write layout can be issued for this file.
6706 	 * This is used during mirror recovery to avoid RW layouts being
6707 	 * issued for a file while it is being copied to the recovered
6708 	 * mirror.
6709 	 */
6710 	if (*iomode == NFSLAYOUTIOMODE_RW && nfsrv_dontlayout(&fh) != 0)
6711 		return (NFSERR_LAYOUTTRYLATER);
6712 
6713 	*retonclose = 0;
6714 	*offset = 0;
6715 	*len = UINT64_MAX;
6716 
6717 	/* First, see if a layout already exists and return if found. */
6718 	lhyp = NFSLAYOUTHASH(&fh);
6719 	NFSLOCKLAYOUT(lhyp);
6720 	error = nfsrv_findlayout(&nd->nd_clientid, &fh, layouttype, p, &lyp);
6721 	NFSD_DEBUG(4, "layoutget findlay=%d\n", error);
6722 	/*
6723 	 * Not sure if the seqid must be the same, so I won't check it.
6724 	 */
6725 	if (error == 0 && (stateidp->other[0] != lyp->lay_stateid.other[0] ||
6726 	    stateidp->other[1] != lyp->lay_stateid.other[1] ||
6727 	    stateidp->other[2] != lyp->lay_stateid.other[2])) {
6728 		if ((lyp->lay_flags & NFSLAY_CALLB) == 0) {
6729 			NFSUNLOCKLAYOUT(lhyp);
6730 			NFSD_DEBUG(1, "ret bad stateid\n");
6731 			return (NFSERR_BADSTATEID);
6732 		}
6733 		/*
6734 		 * I believe we get here because there is a race between
6735 		 * the client processing the CBLAYOUTRECALL and the layout
6736 		 * being deleted here on the server.
6737 		 * The client has now done a LayoutGet with a non-layout
6738 		 * stateid, as it would when there is no layout.
6739 		 * As such, free this layout and set error == NFSERR_BADSTATEID
6740 		 * so the code below will create a new layout structure as
6741 		 * would happen if no layout was found.
6742 		 * "lyp" will be set before being used below, but set it NULL
6743 		 * as a safety belt.
6744 		 */
6745 		nfsrv_freelayout(&lhyp->list, lyp);
6746 		lyp = NULL;
6747 		error = NFSERR_BADSTATEID;
6748 	}
6749 	if (error == 0) {
6750 		if (lyp->lay_layoutlen > maxcnt) {
6751 			NFSUNLOCKLAYOUT(lhyp);
6752 			NFSD_DEBUG(1, "ret layout too small\n");
6753 			return (NFSERR_TOOSMALL);
6754 		}
6755 		if (*iomode == NFSLAYOUTIOMODE_RW)
6756 			lyp->lay_flags |= NFSLAY_RW;
6757 		else
6758 			lyp->lay_flags |= NFSLAY_READ;
6759 		NFSBCOPY(lyp->lay_xdr, layp, lyp->lay_layoutlen);
6760 		*layoutlenp = lyp->lay_layoutlen;
6761 		if (++lyp->lay_stateid.seqid == 0)
6762 			lyp->lay_stateid.seqid = 1;
6763 		stateidp->seqid = lyp->lay_stateid.seqid;
6764 		NFSUNLOCKLAYOUT(lhyp);
6765 		NFSD_DEBUG(4, "ret fnd layout\n");
6766 		return (0);
6767 	}
6768 	NFSUNLOCKLAYOUT(lhyp);
6769 
6770 	/* Find the device id and file handle. */
6771 	dsfhp = malloc(sizeof(fhandle_t) * NFSDEV_MAXMIRRORS, M_TEMP, M_WAITOK);
6772 	devid = malloc(NFSX_V4DEVICEID * NFSDEV_MAXMIRRORS, M_TEMP, M_WAITOK);
6773 	error = nfsrv_dsgetdevandfh(vp, p, &mirrorcnt, dsfhp, devid);
6774 	NFSD_DEBUG(4, "layoutget devandfh=%d\n", error);
6775 	if (error == 0) {
6776 		if (layouttype == NFSLAYOUT_NFSV4_1_FILES) {
6777 			if (NFSX_V4FILELAYOUT > maxcnt)
6778 				error = NFSERR_TOOSMALL;
6779 			else
6780 				lyp = nfsrv_filelayout(nd, *iomode, &fh, dsfhp,
6781 				    devid, vp->v_mount->mnt_stat.f_fsid);
6782 		} else {
6783 			if (NFSX_V4FLEXLAYOUT(mirrorcnt) > maxcnt)
6784 				error = NFSERR_TOOSMALL;
6785 			else
6786 				lyp = nfsrv_flexlayout(nd, *iomode, mirrorcnt,
6787 				    &fh, dsfhp, devid,
6788 				    vp->v_mount->mnt_stat.f_fsid);
6789 		}
6790 	}
6791 	free(dsfhp, M_TEMP);
6792 	free(devid, M_TEMP);
6793 	if (error != 0)
6794 		return (error);
6795 
6796 	/*
6797 	 * Now, add this layout to the list.
6798 	 */
6799 	error = nfsrv_addlayout(nd, &lyp, stateidp, layp, layoutlenp, p);
6800 	NFSD_DEBUG(4, "layoutget addl=%d\n", error);
6801 	/*
6802 	 * The lyp will be set to NULL by nfsrv_addlayout() if it
6803 	 * linked the new structure into the lists.
6804 	 */
6805 	free(lyp, M_NFSDSTATE);
6806 	return (error);
6807 }
6808 
6809 /*
6810  * Generate a File Layout.
6811  */
6812 static struct nfslayout *
6813 nfsrv_filelayout(struct nfsrv_descript *nd, int iomode, fhandle_t *fhp,
6814     fhandle_t *dsfhp, char *devid, fsid_t fs)
6815 {
6816 	uint32_t *tl;
6817 	struct nfslayout *lyp;
6818 	uint64_t pattern_offset;
6819 
6820 	lyp = malloc(sizeof(struct nfslayout) + NFSX_V4FILELAYOUT, M_NFSDSTATE,
6821 	    M_WAITOK | M_ZERO);
6822 	lyp->lay_type = NFSLAYOUT_NFSV4_1_FILES;
6823 	if (iomode == NFSLAYOUTIOMODE_RW)
6824 		lyp->lay_flags = NFSLAY_RW;
6825 	else
6826 		lyp->lay_flags = NFSLAY_READ;
6827 	NFSBCOPY(fhp, &lyp->lay_fh, sizeof(*fhp));
6828 	lyp->lay_clientid.qval = nd->nd_clientid.qval;
6829 	lyp->lay_fsid = fs;
6830 
6831 	/* Fill in the xdr for the files layout. */
6832 	tl = (uint32_t *)lyp->lay_xdr;
6833 	NFSBCOPY(devid, tl, NFSX_V4DEVICEID);		/* Device ID. */
6834 	tl += (NFSX_V4DEVICEID / NFSX_UNSIGNED);
6835 
6836 	/*
6837 	 * Make the stripe size as many 64K blocks as will fit in the stripe
6838 	 * mask. Since there is only one stripe, the stripe size doesn't really
6839 	 * matter, except that the Linux client will only handle an exact
6840 	 * multiple of their PAGE_SIZE (usually 4K).  I chose 64K as a value
6841 	 * that should cover most/all arches w.r.t. PAGE_SIZE.
6842 	 */
6843 	*tl++ = txdr_unsigned(NFSFLAYUTIL_STRIPE_MASK & ~0xffff);
6844 	*tl++ = 0;					/* 1st stripe index. */
6845 	pattern_offset = 0;
6846 	txdr_hyper(pattern_offset, tl); tl += 2;	/* Pattern offset. */
6847 	*tl++ = txdr_unsigned(1);			/* 1 file handle. */
6848 	*tl++ = txdr_unsigned(NFSX_V4PNFSFH);
6849 	NFSBCOPY(dsfhp, tl, sizeof(*dsfhp));
6850 	lyp->lay_layoutlen = NFSX_V4FILELAYOUT;
6851 	return (lyp);
6852 }
6853 
6854 #define	FLEX_OWNERID	"999"
6855 #define	FLEX_UID0	"0"
6856 /*
6857  * Generate a Flex File Layout.
6858  * The FLEX_OWNERID can be any string of 3 decimal digits. Although this
6859  * string goes on the wire, it isn't supposed to be used by the client,
6860  * since this server uses tight coupling.
6861  * Although not recommended by the spec., if vfs.nfsd.flexlinuxhack=1 use
6862  * a string of "0". This works around the Linux Flex File Layout driver bug
6863  * which uses the synthetic uid/gid strings for the "tightly coupled" case.
6864  */
6865 static struct nfslayout *
6866 nfsrv_flexlayout(struct nfsrv_descript *nd, int iomode, int mirrorcnt,
6867     fhandle_t *fhp, fhandle_t *dsfhp, char *devid, fsid_t fs)
6868 {
6869 	uint32_t *tl;
6870 	struct nfslayout *lyp;
6871 	uint64_t lenval;
6872 	int i;
6873 
6874 	lyp = malloc(sizeof(struct nfslayout) + NFSX_V4FLEXLAYOUT(mirrorcnt),
6875 	    M_NFSDSTATE, M_WAITOK | M_ZERO);
6876 	lyp->lay_type = NFSLAYOUT_FLEXFILE;
6877 	if (iomode == NFSLAYOUTIOMODE_RW)
6878 		lyp->lay_flags = NFSLAY_RW;
6879 	else
6880 		lyp->lay_flags = NFSLAY_READ;
6881 	NFSBCOPY(fhp, &lyp->lay_fh, sizeof(*fhp));
6882 	lyp->lay_clientid.qval = nd->nd_clientid.qval;
6883 	lyp->lay_fsid = fs;
6884 	lyp->lay_mirrorcnt = mirrorcnt;
6885 
6886 	/* Fill in the xdr for the files layout. */
6887 	tl = (uint32_t *)lyp->lay_xdr;
6888 	lenval = 0;
6889 	txdr_hyper(lenval, tl); tl += 2;		/* Stripe unit. */
6890 	*tl++ = txdr_unsigned(mirrorcnt);		/* # of mirrors. */
6891 	for (i = 0; i < mirrorcnt; i++) {
6892 		*tl++ = txdr_unsigned(1);		/* One stripe. */
6893 		NFSBCOPY(devid, tl, NFSX_V4DEVICEID);	/* Device ID. */
6894 		tl += (NFSX_V4DEVICEID / NFSX_UNSIGNED);
6895 		devid += NFSX_V4DEVICEID;
6896 		*tl++ = txdr_unsigned(1);		/* Efficiency. */
6897 		*tl++ = 0;				/* Proxy Stateid. */
6898 		*tl++ = 0x55555555;
6899 		*tl++ = 0x55555555;
6900 		*tl++ = 0x55555555;
6901 		*tl++ = txdr_unsigned(1);		/* 1 file handle. */
6902 		*tl++ = txdr_unsigned(NFSX_V4PNFSFH);
6903 		NFSBCOPY(dsfhp, tl, sizeof(*dsfhp));
6904 		tl += (NFSM_RNDUP(NFSX_V4PNFSFH) / NFSX_UNSIGNED);
6905 		dsfhp++;
6906 		if (nfsrv_flexlinuxhack != 0) {
6907 			*tl++ = txdr_unsigned(strlen(FLEX_UID0));
6908 			*tl = 0;		/* 0 pad string. */
6909 			NFSBCOPY(FLEX_UID0, tl++, strlen(FLEX_UID0));
6910 			*tl++ = txdr_unsigned(strlen(FLEX_UID0));
6911 			*tl = 0;		/* 0 pad string. */
6912 			NFSBCOPY(FLEX_UID0, tl++, strlen(FLEX_UID0));
6913 		} else {
6914 			*tl++ = txdr_unsigned(strlen(FLEX_OWNERID));
6915 			NFSBCOPY(FLEX_OWNERID, tl++, NFSX_UNSIGNED);
6916 			*tl++ = txdr_unsigned(strlen(FLEX_OWNERID));
6917 			NFSBCOPY(FLEX_OWNERID, tl++, NFSX_UNSIGNED);
6918 		}
6919 	}
6920 	*tl++ = txdr_unsigned(0);		/* ff_flags. */
6921 	*tl = txdr_unsigned(60);		/* Status interval hint. */
6922 	lyp->lay_layoutlen = NFSX_V4FLEXLAYOUT(mirrorcnt);
6923 	return (lyp);
6924 }
6925 
6926 /*
6927  * Parse and process Flex File errors returned via LayoutReturn.
6928  */
6929 static void
6930 nfsrv_flexlayouterr(struct nfsrv_descript *nd, uint32_t *layp, int maxcnt,
6931     NFSPROC_T *p)
6932 {
6933 	uint32_t *tl;
6934 	int cnt, errcnt, i, j, opnum, stat;
6935 	char devid[NFSX_V4DEVICEID];
6936 
6937 	tl = layp;
6938 	cnt = fxdr_unsigned(int, *tl++);
6939 	NFSD_DEBUG(4, "flexlayouterr cnt=%d\n", cnt);
6940 	for (i = 0; i < cnt; i++) {
6941 		/* Skip offset, length and stateid for now. */
6942 		tl += (4 + NFSX_STATEID / NFSX_UNSIGNED);
6943 		errcnt = fxdr_unsigned(int, *tl++);
6944 		NFSD_DEBUG(4, "flexlayouterr errcnt=%d\n", errcnt);
6945 		for (j = 0; j < errcnt; j++) {
6946 			NFSBCOPY(tl, devid, NFSX_V4DEVICEID);
6947 			tl += (NFSX_V4DEVICEID / NFSX_UNSIGNED);
6948 			stat = fxdr_unsigned(int, *tl++);
6949 			opnum = fxdr_unsigned(int, *tl++);
6950 			NFSD_DEBUG(4, "flexlayouterr op=%d stat=%d\n", opnum,
6951 			    stat);
6952 			/*
6953 			 * Except for NFSERR_ACCES and NFSERR_STALE errors,
6954 			 * disable the mirror.
6955 			 */
6956 			if (stat != NFSERR_ACCES && stat != NFSERR_STALE)
6957 				nfsrv_delds(devid, p);
6958 		}
6959 	}
6960 }
6961 
6962 /*
6963  * This function removes all flex file layouts which has a mirror with
6964  * a device id that matches the argument.
6965  * Called when the DS represented by the device id has failed.
6966  */
6967 void
6968 nfsrv_flexmirrordel(char *devid, NFSPROC_T *p)
6969 {
6970 	uint32_t *tl;
6971 	struct nfslayout *lyp, *nlyp;
6972 	struct nfslayouthash *lhyp;
6973 	struct nfslayouthead loclyp;
6974 	int i, j;
6975 
6976 	NFSD_DEBUG(4, "flexmirrordel\n");
6977 	/* Move all layouts found onto a local list. */
6978 	TAILQ_INIT(&loclyp);
6979 	for (i = 0; i < nfsrv_layouthashsize; i++) {
6980 		lhyp = &nfslayouthash[i];
6981 		NFSLOCKLAYOUT(lhyp);
6982 		TAILQ_FOREACH_SAFE(lyp, &lhyp->list, lay_list, nlyp) {
6983 			if (lyp->lay_type == NFSLAYOUT_FLEXFILE &&
6984 			    lyp->lay_mirrorcnt > 1) {
6985 				NFSD_DEBUG(4, "possible match\n");
6986 				tl = lyp->lay_xdr;
6987 				tl += 3;
6988 				for (j = 0; j < lyp->lay_mirrorcnt; j++) {
6989 					tl++;
6990 					if (NFSBCMP(devid, tl, NFSX_V4DEVICEID)
6991 					    == 0) {
6992 						/* Found one. */
6993 						NFSD_DEBUG(4, "fnd one\n");
6994 						TAILQ_REMOVE(&lhyp->list, lyp,
6995 						    lay_list);
6996 						TAILQ_INSERT_HEAD(&loclyp, lyp,
6997 						    lay_list);
6998 						break;
6999 					}
7000 					tl += (NFSX_V4DEVICEID / NFSX_UNSIGNED +
7001 					    NFSM_RNDUP(NFSX_V4PNFSFH) /
7002 					    NFSX_UNSIGNED + 11 * NFSX_UNSIGNED);
7003 				}
7004 			}
7005 		}
7006 		NFSUNLOCKLAYOUT(lhyp);
7007 	}
7008 
7009 	/* Now, try to do a Layout recall for each one found. */
7010 	TAILQ_FOREACH_SAFE(lyp, &loclyp, lay_list, nlyp) {
7011 		NFSD_DEBUG(4, "do layout recall\n");
7012 		/*
7013 		 * The layout stateid.seqid needs to be incremented
7014 		 * before doing a LAYOUT_RECALL callback.
7015 		 */
7016 		if (++lyp->lay_stateid.seqid == 0)
7017 			lyp->lay_stateid.seqid = 1;
7018 		nfsrv_recalllayout(lyp->lay_clientid, &lyp->lay_stateid,
7019 		    &lyp->lay_fh, lyp, 1, lyp->lay_type, p);
7020 		nfsrv_freelayout(&loclyp, lyp);
7021 	}
7022 }
7023 
7024 /*
7025  * Do a recall callback to the client for this layout.
7026  */
7027 static int
7028 nfsrv_recalllayout(nfsquad_t clid, nfsv4stateid_t *stateidp, fhandle_t *fhp,
7029     struct nfslayout *lyp, int changed, int laytype, NFSPROC_T *p)
7030 {
7031 	struct nfsclient *clp;
7032 	int error;
7033 
7034 	NFSD_DEBUG(4, "nfsrv_recalllayout\n");
7035 	error = nfsrv_getclient(clid, 0, &clp, NULL, (nfsquad_t)((u_quad_t)0),
7036 	    0, NULL, p);
7037 	NFSD_DEBUG(4, "aft nfsrv_getclient=%d\n", error);
7038 	if (error != 0) {
7039 		printf("nfsrv_recalllayout: getclient err=%d\n", error);
7040 		return (error);
7041 	}
7042 	if ((clp->lc_flags & LCL_NFSV41) != 0) {
7043 		error = nfsrv_docallback(clp, NFSV4OP_CBLAYOUTRECALL,
7044 		    stateidp, changed, fhp, NULL, NULL, laytype, p);
7045 		/* If lyp != NULL, handle an error return here. */
7046 		if (error != 0 && lyp != NULL) {
7047 			NFSDRECALLLOCK();
7048 			/*
7049 			 * Mark it returned, since no layout recall
7050 			 * has been done.
7051 			 * All errors seem to be non-recoverable, although
7052 			 * NFSERR_NOMATCHLAYOUT is a normal event.
7053 			 */
7054 			if ((lyp->lay_flags & NFSLAY_RECALL) != 0) {
7055 				lyp->lay_flags |= NFSLAY_RETURNED;
7056 				wakeup(lyp);
7057 			}
7058 			NFSDRECALLUNLOCK();
7059 			if (error != NFSERR_NOMATCHLAYOUT)
7060 				printf("nfsrv_recalllayout: err=%d\n", error);
7061 		}
7062 	} else
7063 		printf("nfsrv_recalllayout: clp not NFSv4.1\n");
7064 	return (error);
7065 }
7066 
7067 /*
7068  * Find a layout to recall when we exceed our high water mark.
7069  */
7070 void
7071 nfsrv_recalloldlayout(NFSPROC_T *p)
7072 {
7073 	struct nfslayouthash *lhyp;
7074 	struct nfslayout *lyp;
7075 	nfsquad_t clientid;
7076 	nfsv4stateid_t stateid;
7077 	fhandle_t fh;
7078 	int error, laytype = 0, ret;
7079 
7080 	lhyp = &nfslayouthash[arc4random() % nfsrv_layouthashsize];
7081 	NFSLOCKLAYOUT(lhyp);
7082 	TAILQ_FOREACH_REVERSE(lyp, &lhyp->list, nfslayouthead, lay_list) {
7083 		if ((lyp->lay_flags & NFSLAY_CALLB) == 0) {
7084 			lyp->lay_flags |= NFSLAY_CALLB;
7085 			/*
7086 			 * The layout stateid.seqid needs to be incremented
7087 			 * before doing a LAYOUT_RECALL callback.
7088 			 */
7089 			if (++lyp->lay_stateid.seqid == 0)
7090 				lyp->lay_stateid.seqid = 1;
7091 			clientid = lyp->lay_clientid;
7092 			stateid = lyp->lay_stateid;
7093 			NFSBCOPY(&lyp->lay_fh, &fh, sizeof(fh));
7094 			laytype = lyp->lay_type;
7095 			break;
7096 		}
7097 	}
7098 	NFSUNLOCKLAYOUT(lhyp);
7099 	if (lyp != NULL) {
7100 		error = nfsrv_recalllayout(clientid, &stateid, &fh, NULL, 0,
7101 		    laytype, p);
7102 		if (error != 0 && error != NFSERR_NOMATCHLAYOUT)
7103 			NFSD_DEBUG(4, "recallold=%d\n", error);
7104 		if (error != 0) {
7105 			NFSLOCKLAYOUT(lhyp);
7106 			/*
7107 			 * Since the hash list was unlocked, we need to
7108 			 * find it again.
7109 			 */
7110 			ret = nfsrv_findlayout(&clientid, &fh, laytype, p,
7111 			    &lyp);
7112 			if (ret == 0 &&
7113 			    (lyp->lay_flags & NFSLAY_CALLB) != 0 &&
7114 			    lyp->lay_stateid.other[0] == stateid.other[0] &&
7115 			    lyp->lay_stateid.other[1] == stateid.other[1] &&
7116 			    lyp->lay_stateid.other[2] == stateid.other[2]) {
7117 				/*
7118 				 * The client no longer knows this layout, so
7119 				 * it can be free'd now.
7120 				 */
7121 				if (error == NFSERR_NOMATCHLAYOUT)
7122 					nfsrv_freelayout(&lhyp->list, lyp);
7123 				else {
7124 					/*
7125 					 * Leave it to be tried later by
7126 					 * clearing NFSLAY_CALLB and moving
7127 					 * it to the head of the list, so it
7128 					 * won't be tried again for a while.
7129 					 */
7130 					lyp->lay_flags &= ~NFSLAY_CALLB;
7131 					TAILQ_REMOVE(&lhyp->list, lyp,
7132 					    lay_list);
7133 					TAILQ_INSERT_HEAD(&lhyp->list, lyp,
7134 					    lay_list);
7135 				}
7136 			}
7137 			NFSUNLOCKLAYOUT(lhyp);
7138 		}
7139 	}
7140 }
7141 
7142 /*
7143  * Try and return layout(s).
7144  */
7145 int
7146 nfsrv_layoutreturn(struct nfsrv_descript *nd, vnode_t vp,
7147     int layouttype, int iomode, uint64_t offset, uint64_t len, int reclaim,
7148     int kind, nfsv4stateid_t *stateidp, int maxcnt, uint32_t *layp, int *fndp,
7149     struct ucred *cred, NFSPROC_T *p)
7150 {
7151 	struct nfsvattr na;
7152 	struct nfslayouthash *lhyp;
7153 	struct nfslayout *lyp;
7154 	fhandle_t fh;
7155 	int error = 0;
7156 
7157 	*fndp = 0;
7158 	if (kind == NFSV4LAYOUTRET_FILE) {
7159 		error = nfsvno_getfh(vp, &fh, p);
7160 		if (error == 0) {
7161 			error = nfsrv_updatemdsattr(vp, &na, p);
7162 			if (error != 0)
7163 				printf("nfsrv_layoutreturn: updatemdsattr"
7164 				    " failed=%d\n", error);
7165 		}
7166 		if (error == 0) {
7167 			if (reclaim == newnfs_true) {
7168 				error = nfsrv_checkgrace(NULL, NULL,
7169 				    NFSLCK_RECLAIM);
7170 				if (error != NFSERR_NOGRACE)
7171 					error = 0;
7172 				return (error);
7173 			}
7174 			lhyp = NFSLAYOUTHASH(&fh);
7175 			NFSDRECALLLOCK();
7176 			NFSLOCKLAYOUT(lhyp);
7177 			error = nfsrv_findlayout(&nd->nd_clientid, &fh,
7178 			    layouttype, p, &lyp);
7179 			NFSD_DEBUG(4, "layoutret findlay=%d\n", error);
7180 			if (error == 0 &&
7181 			    stateidp->other[0] == lyp->lay_stateid.other[0] &&
7182 			    stateidp->other[1] == lyp->lay_stateid.other[1] &&
7183 			    stateidp->other[2] == lyp->lay_stateid.other[2]) {
7184 				NFSD_DEBUG(4, "nfsrv_layoutreturn: stateid %d"
7185 				    " %x %x %x laystateid %d %x %x %x"
7186 				    " off=%ju len=%ju flgs=0x%x\n",
7187 				    stateidp->seqid, stateidp->other[0],
7188 				    stateidp->other[1], stateidp->other[2],
7189 				    lyp->lay_stateid.seqid,
7190 				    lyp->lay_stateid.other[0],
7191 				    lyp->lay_stateid.other[1],
7192 				    lyp->lay_stateid.other[2],
7193 				    (uintmax_t)offset, (uintmax_t)len,
7194 				    lyp->lay_flags);
7195 				if (++lyp->lay_stateid.seqid == 0)
7196 					lyp->lay_stateid.seqid = 1;
7197 				stateidp->seqid = lyp->lay_stateid.seqid;
7198 				if (offset == 0 && len == UINT64_MAX) {
7199 					if ((iomode & NFSLAYOUTIOMODE_READ) !=
7200 					    0)
7201 						lyp->lay_flags &= ~NFSLAY_READ;
7202 					if ((iomode & NFSLAYOUTIOMODE_RW) != 0)
7203 						lyp->lay_flags &= ~NFSLAY_RW;
7204 					if ((lyp->lay_flags & (NFSLAY_READ |
7205 					    NFSLAY_RW)) == 0)
7206 						nfsrv_freelayout(&lhyp->list,
7207 						    lyp);
7208 					else
7209 						*fndp = 1;
7210 				} else
7211 					*fndp = 1;
7212 			}
7213 			NFSUNLOCKLAYOUT(lhyp);
7214 			/* Search the nfsrv_recalllist for a match. */
7215 			TAILQ_FOREACH(lyp, &nfsrv_recalllisthead, lay_list) {
7216 				if (NFSBCMP(&lyp->lay_fh, &fh,
7217 				    sizeof(fh)) == 0 &&
7218 				    lyp->lay_clientid.qval ==
7219 				    nd->nd_clientid.qval &&
7220 				    stateidp->other[0] ==
7221 				    lyp->lay_stateid.other[0] &&
7222 				    stateidp->other[1] ==
7223 				    lyp->lay_stateid.other[1] &&
7224 				    stateidp->other[2] ==
7225 				    lyp->lay_stateid.other[2]) {
7226 					lyp->lay_flags |= NFSLAY_RETURNED;
7227 					wakeup(lyp);
7228 					error = 0;
7229 				}
7230 			}
7231 			NFSDRECALLUNLOCK();
7232 		}
7233 		if (layouttype == NFSLAYOUT_FLEXFILE)
7234 			nfsrv_flexlayouterr(nd, layp, maxcnt, p);
7235 	} else if (kind == NFSV4LAYOUTRET_FSID)
7236 		nfsrv_freelayouts(&nd->nd_clientid,
7237 		    &vp->v_mount->mnt_stat.f_fsid, layouttype, iomode);
7238 	else if (kind == NFSV4LAYOUTRET_ALL)
7239 		nfsrv_freelayouts(&nd->nd_clientid, NULL, layouttype, iomode);
7240 	else
7241 		error = NFSERR_INVAL;
7242 	if (error == -1)
7243 		error = 0;
7244 	return (error);
7245 }
7246 
7247 /*
7248  * Look for an existing layout.
7249  */
7250 static int
7251 nfsrv_findlayout(nfsquad_t *clientidp, fhandle_t *fhp, int laytype,
7252     NFSPROC_T *p, struct nfslayout **lypp)
7253 {
7254 	struct nfslayouthash *lhyp;
7255 	struct nfslayout *lyp;
7256 	int ret;
7257 
7258 	*lypp = NULL;
7259 	ret = 0;
7260 	lhyp = NFSLAYOUTHASH(fhp);
7261 	TAILQ_FOREACH(lyp, &lhyp->list, lay_list) {
7262 		if (NFSBCMP(&lyp->lay_fh, fhp, sizeof(*fhp)) == 0 &&
7263 		    lyp->lay_clientid.qval == clientidp->qval &&
7264 		    lyp->lay_type == laytype)
7265 			break;
7266 	}
7267 	if (lyp != NULL)
7268 		*lypp = lyp;
7269 	else
7270 		ret = -1;
7271 	return (ret);
7272 }
7273 
7274 /*
7275  * Add the new layout, as required.
7276  */
7277 static int
7278 nfsrv_addlayout(struct nfsrv_descript *nd, struct nfslayout **lypp,
7279     nfsv4stateid_t *stateidp, char *layp, int *layoutlenp, NFSPROC_T *p)
7280 {
7281 	struct nfsclient *clp;
7282 	struct nfslayouthash *lhyp;
7283 	struct nfslayout *lyp, *nlyp;
7284 	fhandle_t *fhp;
7285 	int error;
7286 
7287 	KASSERT((nd->nd_flag & ND_IMPLIEDCLID) != 0,
7288 	    ("nfsrv_layoutget: no nd_clientid\n"));
7289 	lyp = *lypp;
7290 	fhp = &lyp->lay_fh;
7291 	NFSLOCKSTATE();
7292 	error = nfsrv_getclient((nfsquad_t)((u_quad_t)0), CLOPS_RENEW, &clp,
7293 	    NULL, (nfsquad_t)((u_quad_t)0), 0, nd, p);
7294 	if (error != 0) {
7295 		NFSUNLOCKSTATE();
7296 		return (error);
7297 	}
7298 	lyp->lay_stateid.seqid = stateidp->seqid = 1;
7299 	lyp->lay_stateid.other[0] = stateidp->other[0] =
7300 	    clp->lc_clientid.lval[0];
7301 	lyp->lay_stateid.other[1] = stateidp->other[1] =
7302 	    clp->lc_clientid.lval[1];
7303 	lyp->lay_stateid.other[2] = stateidp->other[2] =
7304 	    nfsrv_nextstateindex(clp);
7305 	NFSUNLOCKSTATE();
7306 
7307 	lhyp = NFSLAYOUTHASH(fhp);
7308 	NFSLOCKLAYOUT(lhyp);
7309 	TAILQ_FOREACH(nlyp, &lhyp->list, lay_list) {
7310 		if (NFSBCMP(&nlyp->lay_fh, fhp, sizeof(*fhp)) == 0 &&
7311 		    nlyp->lay_clientid.qval == nd->nd_clientid.qval)
7312 			break;
7313 	}
7314 	if (nlyp != NULL) {
7315 		/* A layout already exists, so use it. */
7316 		nlyp->lay_flags |= (lyp->lay_flags & (NFSLAY_READ | NFSLAY_RW));
7317 		NFSBCOPY(nlyp->lay_xdr, layp, nlyp->lay_layoutlen);
7318 		*layoutlenp = nlyp->lay_layoutlen;
7319 		if (++nlyp->lay_stateid.seqid == 0)
7320 			nlyp->lay_stateid.seqid = 1;
7321 		stateidp->seqid = nlyp->lay_stateid.seqid;
7322 		stateidp->other[0] = nlyp->lay_stateid.other[0];
7323 		stateidp->other[1] = nlyp->lay_stateid.other[1];
7324 		stateidp->other[2] = nlyp->lay_stateid.other[2];
7325 		NFSUNLOCKLAYOUT(lhyp);
7326 		return (0);
7327 	}
7328 
7329 	/* Insert the new layout in the lists. */
7330 	*lypp = NULL;
7331 	atomic_add_int(&nfsrv_layoutcnt, 1);
7332 	NFSBCOPY(lyp->lay_xdr, layp, lyp->lay_layoutlen);
7333 	*layoutlenp = lyp->lay_layoutlen;
7334 	TAILQ_INSERT_HEAD(&lhyp->list, lyp, lay_list);
7335 	NFSUNLOCKLAYOUT(lhyp);
7336 	return (0);
7337 }
7338 
7339 /*
7340  * Get the devinfo for a deviceid.
7341  */
7342 int
7343 nfsrv_getdevinfo(char *devid, int layouttype, uint32_t *maxcnt,
7344     uint32_t *notify, int *devaddrlen, char **devaddr)
7345 {
7346 	struct nfsdevice *ds;
7347 
7348 	if ((layouttype != NFSLAYOUT_NFSV4_1_FILES && layouttype !=
7349 	     NFSLAYOUT_FLEXFILE) ||
7350 	    (nfsrv_maxpnfsmirror > 1 && layouttype == NFSLAYOUT_NFSV4_1_FILES))
7351 		return (NFSERR_UNKNLAYOUTTYPE);
7352 
7353 	/*
7354 	 * Now, search for the device id.  Note that the structures won't go
7355 	 * away, but the order changes in the list.  As such, the lock only
7356 	 * needs to be held during the search through the list.
7357 	 */
7358 	NFSDDSLOCK();
7359 	TAILQ_FOREACH(ds, &nfsrv_devidhead, nfsdev_list) {
7360 		if (NFSBCMP(devid, ds->nfsdev_deviceid, NFSX_V4DEVICEID) == 0 &&
7361 		    ds->nfsdev_nmp != NULL)
7362 			break;
7363 	}
7364 	NFSDDSUNLOCK();
7365 	if (ds == NULL)
7366 		return (NFSERR_NOENT);
7367 
7368 	/* If the correct nfsdev_XXXXaddrlen is > 0, we have the device info. */
7369 	*devaddrlen = 0;
7370 	if (layouttype == NFSLAYOUT_NFSV4_1_FILES) {
7371 		*devaddrlen = ds->nfsdev_fileaddrlen;
7372 		*devaddr = ds->nfsdev_fileaddr;
7373 	} else if (layouttype == NFSLAYOUT_FLEXFILE) {
7374 		*devaddrlen = ds->nfsdev_flexaddrlen;
7375 		*devaddr = ds->nfsdev_flexaddr;
7376 	}
7377 	if (*devaddrlen == 0)
7378 		return (NFSERR_UNKNLAYOUTTYPE);
7379 
7380 	/*
7381 	 * The XDR overhead is 3 unsigned values: layout_type,
7382 	 * length_of_address and notify bitmap.
7383 	 * If the notify array is changed to not all zeros, the
7384 	 * count of unsigned values must be increased.
7385 	 */
7386 	if (*maxcnt > 0 && *maxcnt < NFSM_RNDUP(*devaddrlen) +
7387 	    3 * NFSX_UNSIGNED) {
7388 		*maxcnt = NFSM_RNDUP(*devaddrlen) + 3 * NFSX_UNSIGNED;
7389 		return (NFSERR_TOOSMALL);
7390 	}
7391 	return (0);
7392 }
7393 
7394 /*
7395  * Free a list of layout state structures.
7396  */
7397 static void
7398 nfsrv_freelayoutlist(nfsquad_t clientid)
7399 {
7400 	struct nfslayouthash *lhyp;
7401 	struct nfslayout *lyp, *nlyp;
7402 	int i;
7403 
7404 	for (i = 0; i < nfsrv_layouthashsize; i++) {
7405 		lhyp = &nfslayouthash[i];
7406 		NFSLOCKLAYOUT(lhyp);
7407 		TAILQ_FOREACH_SAFE(lyp, &lhyp->list, lay_list, nlyp) {
7408 			if (lyp->lay_clientid.qval == clientid.qval)
7409 				nfsrv_freelayout(&lhyp->list, lyp);
7410 		}
7411 		NFSUNLOCKLAYOUT(lhyp);
7412 	}
7413 }
7414 
7415 /*
7416  * Free up a layout.
7417  */
7418 static void
7419 nfsrv_freelayout(struct nfslayouthead *lhp, struct nfslayout *lyp)
7420 {
7421 
7422 	NFSD_DEBUG(4, "Freelayout=%p\n", lyp);
7423 	atomic_add_int(&nfsrv_layoutcnt, -1);
7424 	TAILQ_REMOVE(lhp, lyp, lay_list);
7425 	free(lyp, M_NFSDSTATE);
7426 }
7427 
7428 /*
7429  * Free up a device id.
7430  */
7431 void
7432 nfsrv_freeonedevid(struct nfsdevice *ds)
7433 {
7434 	int i;
7435 
7436 	atomic_add_int(&nfsrv_devidcnt, -1);
7437 	vrele(ds->nfsdev_dvp);
7438 	for (i = 0; i < nfsrv_dsdirsize; i++)
7439 		if (ds->nfsdev_dsdir[i] != NULL)
7440 			vrele(ds->nfsdev_dsdir[i]);
7441 	free(ds->nfsdev_fileaddr, M_NFSDSTATE);
7442 	free(ds->nfsdev_flexaddr, M_NFSDSTATE);
7443 	free(ds->nfsdev_host, M_NFSDSTATE);
7444 	free(ds, M_NFSDSTATE);
7445 }
7446 
7447 /*
7448  * Free up a device id and its mirrors.
7449  */
7450 static void
7451 nfsrv_freedevid(struct nfsdevice *ds)
7452 {
7453 
7454 	TAILQ_REMOVE(&nfsrv_devidhead, ds, nfsdev_list);
7455 	nfsrv_freeonedevid(ds);
7456 }
7457 
7458 /*
7459  * Free all layouts and device ids.
7460  * Done when the nfsd threads are shut down since there may be a new
7461  * modified device id list created when the nfsd is restarted.
7462  */
7463 void
7464 nfsrv_freealllayoutsanddevids(void)
7465 {
7466 	struct nfsdontlist *mrp, *nmrp;
7467 	struct nfslayout *lyp, *nlyp;
7468 
7469 	/* Get rid of the deviceid structures. */
7470 	nfsrv_freealldevids();
7471 	TAILQ_INIT(&nfsrv_devidhead);
7472 	nfsrv_devidcnt = 0;
7473 
7474 	/* Get rid of all layouts. */
7475 	nfsrv_freealllayouts();
7476 
7477 	/* Get rid of any nfsdontlist entries. */
7478 	LIST_FOREACH_SAFE(mrp, &nfsrv_dontlisthead, nfsmr_list, nmrp)
7479 		free(mrp, M_NFSDSTATE);
7480 	LIST_INIT(&nfsrv_dontlisthead);
7481 	nfsrv_dontlistlen = 0;
7482 
7483 	/* Free layouts in the recall list. */
7484 	TAILQ_FOREACH_SAFE(lyp, &nfsrv_recalllisthead, lay_list, nlyp)
7485 		nfsrv_freelayout(&nfsrv_recalllisthead, lyp);
7486 	TAILQ_INIT(&nfsrv_recalllisthead);
7487 }
7488 
7489 /*
7490  * Free layouts that match the arguments.
7491  */
7492 static void
7493 nfsrv_freelayouts(nfsquad_t *clid, fsid_t *fs, int laytype, int iomode)
7494 {
7495 	struct nfslayouthash *lhyp;
7496 	struct nfslayout *lyp, *nlyp;
7497 	int i;
7498 
7499 	for (i = 0; i < nfsrv_layouthashsize; i++) {
7500 		lhyp = &nfslayouthash[i];
7501 		NFSLOCKLAYOUT(lhyp);
7502 		TAILQ_FOREACH_SAFE(lyp, &lhyp->list, lay_list, nlyp) {
7503 			if (clid->qval != lyp->lay_clientid.qval)
7504 				continue;
7505 			if (fs != NULL && (fs->val[0] != lyp->lay_fsid.val[0] ||
7506 			    fs->val[1] != lyp->lay_fsid.val[1]))
7507 				continue;
7508 			if (laytype != lyp->lay_type)
7509 				continue;
7510 			if ((iomode & NFSLAYOUTIOMODE_READ) != 0)
7511 				lyp->lay_flags &= ~NFSLAY_READ;
7512 			if ((iomode & NFSLAYOUTIOMODE_RW) != 0)
7513 				lyp->lay_flags &= ~NFSLAY_RW;
7514 			if ((lyp->lay_flags & (NFSLAY_READ | NFSLAY_RW)) == 0)
7515 				nfsrv_freelayout(&lhyp->list, lyp);
7516 		}
7517 		NFSUNLOCKLAYOUT(lhyp);
7518 	}
7519 }
7520 
7521 /*
7522  * Free all layouts for the argument file.
7523  */
7524 void
7525 nfsrv_freefilelayouts(fhandle_t *fhp)
7526 {
7527 	struct nfslayouthash *lhyp;
7528 	struct nfslayout *lyp, *nlyp;
7529 
7530 	lhyp = NFSLAYOUTHASH(fhp);
7531 	NFSLOCKLAYOUT(lhyp);
7532 	TAILQ_FOREACH_SAFE(lyp, &lhyp->list, lay_list, nlyp) {
7533 		if (NFSBCMP(&lyp->lay_fh, fhp, sizeof(*fhp)) == 0)
7534 			nfsrv_freelayout(&lhyp->list, lyp);
7535 	}
7536 	NFSUNLOCKLAYOUT(lhyp);
7537 }
7538 
7539 /*
7540  * Free all layouts.
7541  */
7542 static void
7543 nfsrv_freealllayouts(void)
7544 {
7545 	struct nfslayouthash *lhyp;
7546 	struct nfslayout *lyp, *nlyp;
7547 	int i;
7548 
7549 	for (i = 0; i < nfsrv_layouthashsize; i++) {
7550 		lhyp = &nfslayouthash[i];
7551 		NFSLOCKLAYOUT(lhyp);
7552 		TAILQ_FOREACH_SAFE(lyp, &lhyp->list, lay_list, nlyp)
7553 			nfsrv_freelayout(&lhyp->list, lyp);
7554 		NFSUNLOCKLAYOUT(lhyp);
7555 	}
7556 }
7557 
7558 /*
7559  * Look up the mount path for the DS server.
7560  */
7561 static int
7562 nfsrv_setdsserver(char *dspathp, char *mdspathp, NFSPROC_T *p,
7563     struct nfsdevice **dsp)
7564 {
7565 	struct nameidata nd;
7566 	struct nfsdevice *ds;
7567 	struct mount *mp;
7568 	int error, i;
7569 	char *dsdirpath;
7570 	size_t dsdirsize;
7571 
7572 	NFSD_DEBUG(4, "setdssrv path=%s\n", dspathp);
7573 	*dsp = NULL;
7574 	NDINIT(&nd, LOOKUP, FOLLOW | LOCKSHARED | LOCKLEAF, UIO_SYSSPACE,
7575 	    dspathp, p);
7576 	error = namei(&nd);
7577 	NFSD_DEBUG(4, "lookup=%d\n", error);
7578 	if (error != 0)
7579 		return (error);
7580 	if (nd.ni_vp->v_type != VDIR) {
7581 		vput(nd.ni_vp);
7582 		NFSD_DEBUG(4, "dspath not dir\n");
7583 		return (ENOTDIR);
7584 	}
7585 	if (strcmp(nd.ni_vp->v_mount->mnt_vfc->vfc_name, "nfs") != 0) {
7586 		vput(nd.ni_vp);
7587 		NFSD_DEBUG(4, "dspath not an NFS mount\n");
7588 		return (ENXIO);
7589 	}
7590 
7591 	/*
7592 	 * Allocate a DS server structure with the NFS mounted directory
7593 	 * vnode reference counted, so that a non-forced dismount will
7594 	 * fail with EBUSY.
7595 	 * This structure is always linked into the list, even if an error
7596 	 * is being returned.  The caller will free the entire list upon
7597 	 * an error return.
7598 	 */
7599 	*dsp = ds = malloc(sizeof(*ds) + nfsrv_dsdirsize * sizeof(vnode_t),
7600 	    M_NFSDSTATE, M_WAITOK | M_ZERO);
7601 	ds->nfsdev_dvp = nd.ni_vp;
7602 	ds->nfsdev_nmp = VFSTONFS(nd.ni_vp->v_mount);
7603 	NFSVOPUNLOCK(nd.ni_vp);
7604 
7605 	dsdirsize = strlen(dspathp) + 16;
7606 	dsdirpath = malloc(dsdirsize, M_TEMP, M_WAITOK);
7607 	/* Now, create the DS directory structures. */
7608 	for (i = 0; i < nfsrv_dsdirsize; i++) {
7609 		snprintf(dsdirpath, dsdirsize, "%s/ds%d", dspathp, i);
7610 		NDINIT(&nd, LOOKUP, FOLLOW | LOCKSHARED | LOCKLEAF,
7611 		    UIO_SYSSPACE, dsdirpath, p);
7612 		error = namei(&nd);
7613 		NFSD_DEBUG(4, "dsdirpath=%s lookup=%d\n", dsdirpath, error);
7614 		if (error != 0)
7615 			break;
7616 		if (nd.ni_vp->v_type != VDIR) {
7617 			vput(nd.ni_vp);
7618 			error = ENOTDIR;
7619 			NFSD_DEBUG(4, "dsdirpath not a VDIR\n");
7620 			break;
7621 		}
7622 		if (strcmp(nd.ni_vp->v_mount->mnt_vfc->vfc_name, "nfs") != 0) {
7623 			vput(nd.ni_vp);
7624 			error = ENXIO;
7625 			NFSD_DEBUG(4, "dsdirpath not an NFS mount\n");
7626 			break;
7627 		}
7628 		ds->nfsdev_dsdir[i] = nd.ni_vp;
7629 		NFSVOPUNLOCK(nd.ni_vp);
7630 	}
7631 	free(dsdirpath, M_TEMP);
7632 
7633 	if (strlen(mdspathp) > 0) {
7634 		/*
7635 		 * This DS stores file for a specific MDS exported file
7636 		 * system.
7637 		 */
7638 		NDINIT(&nd, LOOKUP, FOLLOW | LOCKSHARED | LOCKLEAF,
7639 		    UIO_SYSSPACE, mdspathp, p);
7640 		error = namei(&nd);
7641 		NFSD_DEBUG(4, "mds lookup=%d\n", error);
7642 		if (error != 0)
7643 			goto out;
7644 		if (nd.ni_vp->v_type != VDIR) {
7645 			vput(nd.ni_vp);
7646 			error = ENOTDIR;
7647 			NFSD_DEBUG(4, "mdspath not dir\n");
7648 			goto out;
7649 		}
7650 		mp = nd.ni_vp->v_mount;
7651 		if ((mp->mnt_flag & MNT_EXPORTED) == 0) {
7652 			vput(nd.ni_vp);
7653 			error = ENXIO;
7654 			NFSD_DEBUG(4, "mdspath not an exported fs\n");
7655 			goto out;
7656 		}
7657 		ds->nfsdev_mdsfsid = mp->mnt_stat.f_fsid;
7658 		ds->nfsdev_mdsisset = 1;
7659 		vput(nd.ni_vp);
7660 	}
7661 
7662 out:
7663 	TAILQ_INSERT_TAIL(&nfsrv_devidhead, ds, nfsdev_list);
7664 	atomic_add_int(&nfsrv_devidcnt, 1);
7665 	return (error);
7666 }
7667 
7668 /*
7669  * Look up the mount path for the DS server and delete it.
7670  */
7671 int
7672 nfsrv_deldsserver(int op, char *dspathp, NFSPROC_T *p)
7673 {
7674 	struct mount *mp;
7675 	struct nfsmount *nmp;
7676 	struct nfsdevice *ds;
7677 	int error;
7678 
7679 	NFSD_DEBUG(4, "deldssrv path=%s\n", dspathp);
7680 	/*
7681 	 * Search for the path in the mount list.  Avoid looking the path
7682 	 * up, since this mount point may be hung, with associated locked
7683 	 * vnodes, etc.
7684 	 * Set NFSMNTP_CANCELRPCS so that any forced dismount will be blocked
7685 	 * until this completes.
7686 	 * As noted in the man page, this should be done before any forced
7687 	 * dismount on the mount point, but at least the handshake on
7688 	 * NFSMNTP_CANCELRPCS should make it safe.
7689 	 */
7690 	error = 0;
7691 	ds = NULL;
7692 	nmp = NULL;
7693 	mtx_lock(&mountlist_mtx);
7694 	TAILQ_FOREACH(mp, &mountlist, mnt_list) {
7695 		if (strcmp(mp->mnt_stat.f_mntonname, dspathp) == 0 &&
7696 		    strcmp(mp->mnt_stat.f_fstypename, "nfs") == 0 &&
7697 		    mp->mnt_data != NULL) {
7698 			nmp = VFSTONFS(mp);
7699 			NFSLOCKMNT(nmp);
7700 			if ((nmp->nm_privflag & (NFSMNTP_FORCEDISM |
7701 			     NFSMNTP_CANCELRPCS)) == 0) {
7702 				nmp->nm_privflag |= NFSMNTP_CANCELRPCS;
7703 				NFSUNLOCKMNT(nmp);
7704 			} else {
7705 				NFSUNLOCKMNT(nmp);
7706 				nmp = NULL;
7707 			}
7708 			break;
7709 		}
7710 	}
7711 	mtx_unlock(&mountlist_mtx);
7712 
7713 	if (nmp != NULL) {
7714 		ds = nfsrv_deldsnmp(op, nmp, p);
7715 		NFSD_DEBUG(4, "deldsnmp=%p\n", ds);
7716 		if (ds != NULL) {
7717 			nfsrv_killrpcs(nmp);
7718 			NFSD_DEBUG(4, "aft killrpcs\n");
7719 		} else
7720 			error = ENXIO;
7721 		NFSLOCKMNT(nmp);
7722 		nmp->nm_privflag &= ~NFSMNTP_CANCELRPCS;
7723 		wakeup(nmp);
7724 		NFSUNLOCKMNT(nmp);
7725 	} else
7726 		error = EINVAL;
7727 	return (error);
7728 }
7729 
7730 /*
7731  * Search for and remove a DS entry which matches the "nmp" argument.
7732  * The nfsdevice structure pointer is returned so that the caller can
7733  * free it via nfsrv_freeonedevid().
7734  * For the forced case, do not try to do LayoutRecalls, since the server
7735  * must be shut down now anyhow.
7736  */
7737 struct nfsdevice *
7738 nfsrv_deldsnmp(int op, struct nfsmount *nmp, NFSPROC_T *p)
7739 {
7740 	struct nfsdevice *fndds;
7741 
7742 	NFSD_DEBUG(4, "deldsdvp\n");
7743 	NFSDDSLOCK();
7744 	if (op == PNFSDOP_FORCEDELDS)
7745 		fndds = nfsv4_findmirror(nmp);
7746 	else
7747 		fndds = nfsrv_findmirroredds(nmp);
7748 	if (fndds != NULL)
7749 		nfsrv_deleteds(fndds);
7750 	NFSDDSUNLOCK();
7751 	if (fndds != NULL) {
7752 		if (op != PNFSDOP_FORCEDELDS)
7753 			nfsrv_flexmirrordel(fndds->nfsdev_deviceid, p);
7754 		printf("pNFS server: mirror %s failed\n", fndds->nfsdev_host);
7755 	}
7756 	return (fndds);
7757 }
7758 
7759 /*
7760  * Similar to nfsrv_deldsnmp(), except that the DS is indicated by deviceid.
7761  * This function also calls nfsrv_killrpcs() to unblock RPCs on the mount
7762  * point.
7763  * Also, returns an error instead of the nfsdevice found.
7764  */
7765 APPLESTATIC int
7766 nfsrv_delds(char *devid, NFSPROC_T *p)
7767 {
7768 	struct nfsdevice *ds, *fndds;
7769 	struct nfsmount *nmp;
7770 	int fndmirror;
7771 
7772 	NFSD_DEBUG(4, "delds\n");
7773 	/*
7774 	 * Search the DS server list for a match with devid.
7775 	 * Remove the DS entry if found and there is a mirror.
7776 	 */
7777 	fndds = NULL;
7778 	nmp = NULL;
7779 	fndmirror = 0;
7780 	NFSDDSLOCK();
7781 	TAILQ_FOREACH(ds, &nfsrv_devidhead, nfsdev_list) {
7782 		if (NFSBCMP(ds->nfsdev_deviceid, devid, NFSX_V4DEVICEID) == 0 &&
7783 		    ds->nfsdev_nmp != NULL) {
7784 			NFSD_DEBUG(4, "fnd main ds\n");
7785 			fndds = ds;
7786 			break;
7787 		}
7788 	}
7789 	if (fndds == NULL) {
7790 		NFSDDSUNLOCK();
7791 		return (ENXIO);
7792 	}
7793 	if (fndds->nfsdev_mdsisset == 0 && nfsrv_faildscnt > 0)
7794 		fndmirror = 1;
7795 	else if (fndds->nfsdev_mdsisset != 0) {
7796 		/* For the fsid is set case, search for a mirror. */
7797 		TAILQ_FOREACH(ds, &nfsrv_devidhead, nfsdev_list) {
7798 			if (ds != fndds && ds->nfsdev_nmp != NULL &&
7799 			    ds->nfsdev_mdsisset != 0 &&
7800 			    ds->nfsdev_mdsfsid.val[0] ==
7801 			    fndds->nfsdev_mdsfsid.val[0] &&
7802 			    ds->nfsdev_mdsfsid.val[1] ==
7803 			    fndds->nfsdev_mdsfsid.val[1]) {
7804 				fndmirror = 1;
7805 				break;
7806 			}
7807 		}
7808 	}
7809 	if (fndmirror != 0) {
7810 		nmp = fndds->nfsdev_nmp;
7811 		NFSLOCKMNT(nmp);
7812 		if ((nmp->nm_privflag & (NFSMNTP_FORCEDISM |
7813 		     NFSMNTP_CANCELRPCS)) == 0) {
7814 			nmp->nm_privflag |= NFSMNTP_CANCELRPCS;
7815 			NFSUNLOCKMNT(nmp);
7816 			nfsrv_deleteds(fndds);
7817 		} else {
7818 			NFSUNLOCKMNT(nmp);
7819 			nmp = NULL;
7820 		}
7821 	}
7822 	NFSDDSUNLOCK();
7823 	if (nmp != NULL) {
7824 		nfsrv_flexmirrordel(fndds->nfsdev_deviceid, p);
7825 		printf("pNFS server: mirror %s failed\n", fndds->nfsdev_host);
7826 		nfsrv_killrpcs(nmp);
7827 		NFSLOCKMNT(nmp);
7828 		nmp->nm_privflag &= ~NFSMNTP_CANCELRPCS;
7829 		wakeup(nmp);
7830 		NFSUNLOCKMNT(nmp);
7831 		return (0);
7832 	}
7833 	return (ENXIO);
7834 }
7835 
7836 /*
7837  * Mark a DS as disabled by setting nfsdev_nmp = NULL.
7838  */
7839 static void
7840 nfsrv_deleteds(struct nfsdevice *fndds)
7841 {
7842 
7843 	NFSD_DEBUG(4, "deleteds: deleting a mirror\n");
7844 	fndds->nfsdev_nmp = NULL;
7845 	if (fndds->nfsdev_mdsisset == 0)
7846 		nfsrv_faildscnt--;
7847 }
7848 
7849 /*
7850  * Fill in the addr structures for the File and Flex File layouts.
7851  */
7852 static void
7853 nfsrv_allocdevid(struct nfsdevice *ds, char *addr, char *dnshost)
7854 {
7855 	uint32_t *tl;
7856 	char *netprot;
7857 	int addrlen;
7858 	static uint64_t new_devid = 0;
7859 
7860 	if (strchr(addr, ':') != NULL)
7861 		netprot = "tcp6";
7862 	else
7863 		netprot = "tcp";
7864 
7865 	/* Fill in the device id. */
7866 	NFSBCOPY(&nfsdev_time, ds->nfsdev_deviceid, sizeof(nfsdev_time));
7867 	new_devid++;
7868 	NFSBCOPY(&new_devid, &ds->nfsdev_deviceid[sizeof(nfsdev_time)],
7869 	    sizeof(new_devid));
7870 
7871 	/*
7872 	 * Fill in the file addr (actually the nfsv4_file_layout_ds_addr4
7873 	 * as defined in RFC5661) in XDR.
7874 	 */
7875 	addrlen = NFSM_RNDUP(strlen(addr)) + NFSM_RNDUP(strlen(netprot)) +
7876 	    6 * NFSX_UNSIGNED;
7877 	NFSD_DEBUG(4, "hn=%s addr=%s netprot=%s\n", dnshost, addr, netprot);
7878 	ds->nfsdev_fileaddrlen = addrlen;
7879 	tl = malloc(addrlen, M_NFSDSTATE, M_WAITOK | M_ZERO);
7880 	ds->nfsdev_fileaddr = (char *)tl;
7881 	*tl++ = txdr_unsigned(1);		/* One stripe with index 0. */
7882 	*tl++ = 0;
7883 	*tl++ = txdr_unsigned(1);		/* One multipath list */
7884 	*tl++ = txdr_unsigned(1);		/* with one entry in it. */
7885 	/* The netaddr for this one entry. */
7886 	*tl++ = txdr_unsigned(strlen(netprot));
7887 	NFSBCOPY(netprot, tl, strlen(netprot));
7888 	tl += (NFSM_RNDUP(strlen(netprot)) / NFSX_UNSIGNED);
7889 	*tl++ = txdr_unsigned(strlen(addr));
7890 	NFSBCOPY(addr, tl, strlen(addr));
7891 
7892 	/*
7893 	 * Fill in the flex file addr (actually the ff_device_addr4
7894 	 * as defined for Flexible File Layout) in XDR.
7895 	 */
7896 	addrlen = NFSM_RNDUP(strlen(addr)) + NFSM_RNDUP(strlen(netprot)) +
7897 	    14 * NFSX_UNSIGNED;
7898 	ds->nfsdev_flexaddrlen = addrlen;
7899 	tl = malloc(addrlen, M_NFSDSTATE, M_WAITOK | M_ZERO);
7900 	ds->nfsdev_flexaddr = (char *)tl;
7901 	*tl++ = txdr_unsigned(1);		/* One multipath entry. */
7902 	/* The netaddr for this one entry. */
7903 	*tl++ = txdr_unsigned(strlen(netprot));
7904 	NFSBCOPY(netprot, tl, strlen(netprot));
7905 	tl += (NFSM_RNDUP(strlen(netprot)) / NFSX_UNSIGNED);
7906 	*tl++ = txdr_unsigned(strlen(addr));
7907 	NFSBCOPY(addr, tl, strlen(addr));
7908 	tl += (NFSM_RNDUP(strlen(addr)) / NFSX_UNSIGNED);
7909 	*tl++ = txdr_unsigned(2);		/* Two NFS Versions. */
7910 	*tl++ = txdr_unsigned(NFS_VER4);	/* NFSv4. */
7911 	*tl++ = txdr_unsigned(NFSV42_MINORVERSION); /* Minor version 2. */
7912 	*tl++ = txdr_unsigned(NFS_SRVMAXIO);	/* DS max rsize. */
7913 	*tl++ = txdr_unsigned(NFS_SRVMAXIO);	/* DS max wsize. */
7914 	*tl++ = newnfs_true;			/* Tightly coupled. */
7915 	*tl++ = txdr_unsigned(NFS_VER4);	/* NFSv4. */
7916 	*tl++ = txdr_unsigned(NFSV41_MINORVERSION); /* Minor version 1. */
7917 	*tl++ = txdr_unsigned(NFS_SRVMAXIO);	/* DS max rsize. */
7918 	*tl++ = txdr_unsigned(NFS_SRVMAXIO);	/* DS max wsize. */
7919 	*tl = newnfs_true;			/* Tightly coupled. */
7920 
7921 	ds->nfsdev_hostnamelen = strlen(dnshost);
7922 	ds->nfsdev_host = malloc(ds->nfsdev_hostnamelen + 1, M_NFSDSTATE,
7923 	    M_WAITOK);
7924 	NFSBCOPY(dnshost, ds->nfsdev_host, ds->nfsdev_hostnamelen + 1);
7925 }
7926 
7927 
7928 /*
7929  * Create the device id list.
7930  * Return 0 if the nfsd threads are to run and ENXIO if the "-p" argument
7931  * is misconfigured.
7932  */
7933 int
7934 nfsrv_createdevids(struct nfsd_nfsd_args *args, NFSPROC_T *p)
7935 {
7936 	struct nfsdevice *ds;
7937 	char *addrp, *dnshostp, *dspathp, *mdspathp;
7938 	int error, i;
7939 
7940 	addrp = args->addr;
7941 	dnshostp = args->dnshost;
7942 	dspathp = args->dspath;
7943 	mdspathp = args->mdspath;
7944 	nfsrv_maxpnfsmirror = args->mirrorcnt;
7945 	if (addrp == NULL || dnshostp == NULL || dspathp == NULL ||
7946 	    mdspathp == NULL)
7947 		return (0);
7948 
7949 	/*
7950 	 * Loop around for each nul-terminated string in args->addr,
7951 	 * args->dnshost, args->dnspath and args->mdspath.
7952 	 */
7953 	while (addrp < (args->addr + args->addrlen) &&
7954 	    dnshostp < (args->dnshost + args->dnshostlen) &&
7955 	    dspathp < (args->dspath + args->dspathlen) &&
7956 	    mdspathp < (args->mdspath + args->mdspathlen)) {
7957 		error = nfsrv_setdsserver(dspathp, mdspathp, p, &ds);
7958 		if (error != 0) {
7959 			/* Free all DS servers. */
7960 			nfsrv_freealldevids();
7961 			nfsrv_devidcnt = 0;
7962 			return (ENXIO);
7963 		}
7964 		nfsrv_allocdevid(ds, addrp, dnshostp);
7965 		addrp += (strlen(addrp) + 1);
7966 		dnshostp += (strlen(dnshostp) + 1);
7967 		dspathp += (strlen(dspathp) + 1);
7968 		mdspathp += (strlen(mdspathp) + 1);
7969 	}
7970 	if (nfsrv_devidcnt < nfsrv_maxpnfsmirror) {
7971 		/* Free all DS servers. */
7972 		nfsrv_freealldevids();
7973 		nfsrv_devidcnt = 0;
7974 		nfsrv_maxpnfsmirror = 1;
7975 		return (ENXIO);
7976 	}
7977 	/* We can fail at most one less DS than the mirror level. */
7978 	nfsrv_faildscnt = nfsrv_maxpnfsmirror - 1;
7979 
7980 	/*
7981 	 * Allocate the nfslayout hash table now, since this is a pNFS server.
7982 	 * Make it 1% of the high water mark and at least 100.
7983 	 */
7984 	if (nfslayouthash == NULL) {
7985 		nfsrv_layouthashsize = nfsrv_layouthighwater / 100;
7986 		if (nfsrv_layouthashsize < 100)
7987 			nfsrv_layouthashsize = 100;
7988 		nfslayouthash = mallocarray(nfsrv_layouthashsize,
7989 		    sizeof(struct nfslayouthash), M_NFSDSESSION, M_WAITOK |
7990 		    M_ZERO);
7991 		for (i = 0; i < nfsrv_layouthashsize; i++) {
7992 			mtx_init(&nfslayouthash[i].mtx, "nfslm", NULL, MTX_DEF);
7993 			TAILQ_INIT(&nfslayouthash[i].list);
7994 		}
7995 	}
7996 	return (0);
7997 }
7998 
7999 /*
8000  * Free all device ids.
8001  */
8002 static void
8003 nfsrv_freealldevids(void)
8004 {
8005 	struct nfsdevice *ds, *nds;
8006 
8007 	TAILQ_FOREACH_SAFE(ds, &nfsrv_devidhead, nfsdev_list, nds)
8008 		nfsrv_freedevid(ds);
8009 }
8010 
8011 /*
8012  * Check to see if there is a Read/Write Layout plus either:
8013  * - A Write Delegation
8014  * or
8015  * - An Open with Write_access.
8016  * Return 1 if this is the case and 0 otherwise.
8017  * This function is used by nfsrv_proxyds() to decide if doing a Proxy
8018  * Getattr RPC to the Data Server (DS) is necessary.
8019  */
8020 #define	NFSCLIDVECSIZE	6
8021 APPLESTATIC int
8022 nfsrv_checkdsattr(vnode_t vp, NFSPROC_T *p)
8023 {
8024 	fhandle_t fh, *tfhp;
8025 	struct nfsstate *stp;
8026 	struct nfslayout *lyp;
8027 	struct nfslayouthash *lhyp;
8028 	struct nfslockhashhead *hp;
8029 	struct nfslockfile *lfp;
8030 	nfsquad_t clid[NFSCLIDVECSIZE];
8031 	int clidcnt, ret;
8032 
8033 	ret = nfsvno_getfh(vp, &fh, p);
8034 	if (ret != 0)
8035 		return (0);
8036 
8037 	/* First check for a Read/Write Layout. */
8038 	clidcnt = 0;
8039 	lhyp = NFSLAYOUTHASH(&fh);
8040 	NFSLOCKLAYOUT(lhyp);
8041 	TAILQ_FOREACH(lyp, &lhyp->list, lay_list) {
8042 		if (NFSBCMP(&lyp->lay_fh, &fh, sizeof(fh)) == 0 &&
8043 		    ((lyp->lay_flags & NFSLAY_RW) != 0 ||
8044 		     ((lyp->lay_flags & NFSLAY_READ) != 0 &&
8045 		      nfsrv_pnfsatime != 0))) {
8046 			if (clidcnt < NFSCLIDVECSIZE)
8047 				clid[clidcnt].qval = lyp->lay_clientid.qval;
8048 			clidcnt++;
8049 		}
8050 	}
8051 	NFSUNLOCKLAYOUT(lhyp);
8052 	if (clidcnt == 0) {
8053 		/* None found, so return 0. */
8054 		return (0);
8055 	}
8056 
8057 	/* Get the nfslockfile for this fh. */
8058 	NFSLOCKSTATE();
8059 	hp = NFSLOCKHASH(&fh);
8060 	LIST_FOREACH(lfp, hp, lf_hash) {
8061 		tfhp = &lfp->lf_fh;
8062 		if (NFSVNO_CMPFH(&fh, tfhp))
8063 			break;
8064 	}
8065 	if (lfp == NULL) {
8066 		/* None found, so return 0. */
8067 		NFSUNLOCKSTATE();
8068 		return (0);
8069 	}
8070 
8071 	/* Now, look for a Write delegation for this clientid. */
8072 	LIST_FOREACH(stp, &lfp->lf_deleg, ls_file) {
8073 		if ((stp->ls_flags & NFSLCK_DELEGWRITE) != 0 &&
8074 		    nfsrv_fndclid(clid, stp->ls_clp->lc_clientid, clidcnt) != 0)
8075 			break;
8076 	}
8077 	if (stp != NULL) {
8078 		/* Found one, so return 1. */
8079 		NFSUNLOCKSTATE();
8080 		return (1);
8081 	}
8082 
8083 	/* No Write delegation, so look for an Open with Write_access. */
8084 	LIST_FOREACH(stp, &lfp->lf_open, ls_file) {
8085 		KASSERT((stp->ls_flags & NFSLCK_OPEN) != 0,
8086 		    ("nfsrv_checkdsattr: Non-open in Open list\n"));
8087 		if ((stp->ls_flags & NFSLCK_WRITEACCESS) != 0 &&
8088 		    nfsrv_fndclid(clid, stp->ls_clp->lc_clientid, clidcnt) != 0)
8089 			break;
8090 	}
8091 	NFSUNLOCKSTATE();
8092 	if (stp != NULL)
8093 		return (1);
8094 	return (0);
8095 }
8096 
8097 /*
8098  * Look for a matching clientid in the vector. Return 1 if one might match.
8099  */
8100 static int
8101 nfsrv_fndclid(nfsquad_t *clidvec, nfsquad_t clid, int clidcnt)
8102 {
8103 	int i;
8104 
8105 	/* If too many for the vector, return 1 since there might be a match. */
8106 	if (clidcnt > NFSCLIDVECSIZE)
8107 		return (1);
8108 
8109 	for (i = 0; i < clidcnt; i++)
8110 		if (clidvec[i].qval == clid.qval)
8111 			return (1);
8112 	return (0);
8113 }
8114 
8115 /*
8116  * Check the don't list for "vp" and see if issuing an rw layout is allowed.
8117  * Return 1 if issuing an rw layout isn't allowed, 0 otherwise.
8118  */
8119 static int
8120 nfsrv_dontlayout(fhandle_t *fhp)
8121 {
8122 	struct nfsdontlist *mrp;
8123 	int ret;
8124 
8125 	if (nfsrv_dontlistlen == 0)
8126 		return (0);
8127 	ret = 0;
8128 	NFSDDONTLISTLOCK();
8129 	LIST_FOREACH(mrp, &nfsrv_dontlisthead, nfsmr_list) {
8130 		if (NFSBCMP(fhp, &mrp->nfsmr_fh, sizeof(*fhp)) == 0 &&
8131 		    (mrp->nfsmr_flags & NFSMR_DONTLAYOUT) != 0) {
8132 			ret = 1;
8133 			break;
8134 		}
8135 	}
8136 	NFSDDONTLISTUNLOCK();
8137 	return (ret);
8138 }
8139 
8140 #define	PNFSDS_COPYSIZ	65536
8141 /*
8142  * Create a new file on a DS and copy the contents of an extant DS file to it.
8143  * This can be used for recovery of a DS file onto a recovered DS.
8144  * The steps are:
8145  * - When called, the MDS file's vnode is locked, blocking LayoutGet operations.
8146  * - Disable issuing of read/write layouts for the file via the nfsdontlist,
8147  *   so that they will be disabled after the MDS file's vnode is unlocked.
8148  * - Set up the nfsrv_recalllist so that recall of read/write layouts can
8149  *   be done.
8150  * - Unlock the MDS file's vnode, so that the client(s) can perform proxied
8151  *   writes, LayoutCommits and LayoutReturns for the file when completing the
8152  *   LayoutReturn requested by the LayoutRecall callback.
8153  * - Issue a LayoutRecall callback for all read/write layouts and wait for
8154  *   them to be returned. (If the LayoutRecall callback replies
8155  *   NFSERR_NOMATCHLAYOUT, they are gone and no LayoutReturn is needed.)
8156  * - Exclusively lock the MDS file's vnode.  This ensures that no proxied
8157  *   writes are in progress or can occur during the DS file copy.
8158  *   It also blocks Setattr operations.
8159  * - Create the file on the recovered mirror.
8160  * - Copy the file from the operational DS.
8161  * - Copy any ACL from the MDS file to the new DS file.
8162  * - Set the modify time of the new DS file to that of the MDS file.
8163  * - Update the extended attribute for the MDS file.
8164  * - Enable issuing of rw layouts by deleting the nfsdontlist entry.
8165  * - The caller will unlock the MDS file's vnode allowing operations
8166  *   to continue normally, since it is now on the mirror again.
8167  */
8168 int
8169 nfsrv_copymr(vnode_t vp, vnode_t fvp, vnode_t dvp, struct nfsdevice *ds,
8170     struct pnfsdsfile *pf, struct pnfsdsfile *wpf, int mirrorcnt,
8171     struct ucred *cred, NFSPROC_T *p)
8172 {
8173 	struct nfsdontlist *mrp, *nmrp;
8174 	struct nfslayouthash *lhyp;
8175 	struct nfslayout *lyp, *nlyp;
8176 	struct nfslayouthead thl;
8177 	struct mount *mp, *tvmp;
8178 	struct acl *aclp;
8179 	struct vattr va;
8180 	struct timespec mtime;
8181 	fhandle_t fh;
8182 	vnode_t tvp;
8183 	off_t rdpos, wrpos;
8184 	ssize_t aresid;
8185 	char *dat;
8186 	int didprintf, ret, retacl, xfer;
8187 
8188 	ASSERT_VOP_LOCKED(fvp, "nfsrv_copymr fvp");
8189 	ASSERT_VOP_LOCKED(vp, "nfsrv_copymr vp");
8190 	/*
8191 	 * Allocate a nfsdontlist entry and set the NFSMR_DONTLAYOUT flag
8192 	 * so that no more RW layouts will get issued.
8193 	 */
8194 	ret = nfsvno_getfh(vp, &fh, p);
8195 	if (ret != 0) {
8196 		NFSD_DEBUG(4, "nfsrv_copymr: getfh=%d\n", ret);
8197 		return (ret);
8198 	}
8199 	nmrp = malloc(sizeof(*nmrp), M_NFSDSTATE, M_WAITOK);
8200 	nmrp->nfsmr_flags = NFSMR_DONTLAYOUT;
8201 	NFSBCOPY(&fh, &nmrp->nfsmr_fh, sizeof(fh));
8202 	NFSDDONTLISTLOCK();
8203 	LIST_FOREACH(mrp, &nfsrv_dontlisthead, nfsmr_list) {
8204 		if (NFSBCMP(&fh, &mrp->nfsmr_fh, sizeof(fh)) == 0)
8205 			break;
8206 	}
8207 	if (mrp == NULL) {
8208 		LIST_INSERT_HEAD(&nfsrv_dontlisthead, nmrp, nfsmr_list);
8209 		mrp = nmrp;
8210 		nmrp = NULL;
8211 		nfsrv_dontlistlen++;
8212 		NFSD_DEBUG(4, "nfsrv_copymr: in dontlist\n");
8213 	} else {
8214 		NFSDDONTLISTUNLOCK();
8215 		free(nmrp, M_NFSDSTATE);
8216 		NFSD_DEBUG(4, "nfsrv_copymr: dup dontlist\n");
8217 		return (ENXIO);
8218 	}
8219 	NFSDDONTLISTUNLOCK();
8220 
8221 	/*
8222 	 * Search for all RW layouts for this file.  Move them to the
8223 	 * recall list, so they can be recalled and their return noted.
8224 	 */
8225 	lhyp = NFSLAYOUTHASH(&fh);
8226 	NFSDRECALLLOCK();
8227 	NFSLOCKLAYOUT(lhyp);
8228 	TAILQ_FOREACH_SAFE(lyp, &lhyp->list, lay_list, nlyp) {
8229 		if (NFSBCMP(&lyp->lay_fh, &fh, sizeof(fh)) == 0 &&
8230 		    (lyp->lay_flags & NFSLAY_RW) != 0) {
8231 			TAILQ_REMOVE(&lhyp->list, lyp, lay_list);
8232 			TAILQ_INSERT_HEAD(&nfsrv_recalllisthead, lyp, lay_list);
8233 			lyp->lay_trycnt = 0;
8234 		}
8235 	}
8236 	NFSUNLOCKLAYOUT(lhyp);
8237 	NFSDRECALLUNLOCK();
8238 
8239 	ret = 0;
8240 	mp = tvmp = NULL;
8241 	didprintf = 0;
8242 	TAILQ_INIT(&thl);
8243 	/* Unlock the MDS vp, so that a LayoutReturn can be done on it. */
8244 	NFSVOPUNLOCK(vp);
8245 	/* Now, do a recall for all layouts not yet recalled. */
8246 tryagain:
8247 	NFSDRECALLLOCK();
8248 	TAILQ_FOREACH(lyp, &nfsrv_recalllisthead, lay_list) {
8249 		if (NFSBCMP(&lyp->lay_fh, &fh, sizeof(fh)) == 0 &&
8250 		    (lyp->lay_flags & NFSLAY_RECALL) == 0) {
8251 			lyp->lay_flags |= NFSLAY_RECALL;
8252 			/*
8253 			 * The layout stateid.seqid needs to be incremented
8254 			 * before doing a LAYOUT_RECALL callback.
8255 			 */
8256 			if (++lyp->lay_stateid.seqid == 0)
8257 				lyp->lay_stateid.seqid = 1;
8258 			NFSDRECALLUNLOCK();
8259 			nfsrv_recalllayout(lyp->lay_clientid, &lyp->lay_stateid,
8260 			    &lyp->lay_fh, lyp, 0, lyp->lay_type, p);
8261 			NFSD_DEBUG(4, "nfsrv_copymr: recalled layout\n");
8262 			goto tryagain;
8263 		}
8264 	}
8265 
8266 	/* Now wait for them to be returned. */
8267 tryagain2:
8268 	TAILQ_FOREACH(lyp, &nfsrv_recalllisthead, lay_list) {
8269 		if (NFSBCMP(&lyp->lay_fh, &fh, sizeof(fh)) == 0) {
8270 			if ((lyp->lay_flags & NFSLAY_RETURNED) != 0) {
8271 				TAILQ_REMOVE(&nfsrv_recalllisthead, lyp,
8272 				    lay_list);
8273 				TAILQ_INSERT_HEAD(&thl, lyp, lay_list);
8274 				NFSD_DEBUG(4,
8275 				    "nfsrv_copymr: layout returned\n");
8276 			} else {
8277 				lyp->lay_trycnt++;
8278 				ret = mtx_sleep(lyp, NFSDRECALLMUTEXPTR,
8279 				    PVFS | PCATCH, "nfsmrl", hz);
8280 				NFSD_DEBUG(4, "nfsrv_copymr: aft sleep=%d\n",
8281 				    ret);
8282 				if (ret == EINTR || ret == ERESTART)
8283 					break;
8284 				if ((lyp->lay_flags & NFSLAY_RETURNED) == 0) {
8285 					/*
8286 					 * Give up after 60sec and return
8287 					 * ENXIO, failing the copymr.
8288 					 * This layout will remain on the
8289 					 * recalllist.  It can only be cleared
8290 					 * by restarting the nfsd.
8291 					 * This seems the safe way to handle
8292 					 * it, since it cannot be safely copied
8293 					 * with an outstanding RW layout.
8294 					 */
8295 					if (lyp->lay_trycnt >= 60) {
8296 						ret = ENXIO;
8297 						break;
8298 					}
8299 					if (didprintf == 0) {
8300 						printf("nfsrv_copymr: layout "
8301 						    "not returned\n");
8302 						didprintf = 1;
8303 					}
8304 				}
8305 			}
8306 			goto tryagain2;
8307 		}
8308 	}
8309 	NFSDRECALLUNLOCK();
8310 	/* We can now get rid of the layouts that have been returned. */
8311 	TAILQ_FOREACH_SAFE(lyp, &thl, lay_list, nlyp)
8312 		nfsrv_freelayout(&thl, lyp);
8313 
8314 	/*
8315 	 * Do the vn_start_write() calls here, before the MDS vnode is
8316 	 * locked and the tvp is created (locked) in the NFS file system
8317 	 * that dvp is in.
8318 	 * For tvmp, this probably isn't necessary, since it will be an
8319 	 * NFS mount and they are not suspendable at this time.
8320 	 */
8321 	if (ret == 0)
8322 		ret = vn_start_write(vp, &mp, V_WAIT | PCATCH);
8323 	if (ret == 0) {
8324 		tvmp = dvp->v_mount;
8325 		ret = vn_start_write(NULL, &tvmp, V_WAIT | PCATCH);
8326 	}
8327 
8328 	/*
8329 	 * LK_EXCLUSIVE lock the MDS vnode, so that any
8330 	 * proxied writes through the MDS will be blocked until we have
8331 	 * completed the copy and update of the extended attributes.
8332 	 * This will also ensure that any attributes and ACL will not be
8333 	 * changed until the copy is complete.
8334 	 */
8335 	NFSVOPLOCK(vp, LK_EXCLUSIVE | LK_RETRY);
8336 	if (ret == 0 && VN_IS_DOOMED(vp)) {
8337 		NFSD_DEBUG(4, "nfsrv_copymr: lk_exclusive doomed\n");
8338 		ret = ESTALE;
8339 	}
8340 
8341 	/* Create the data file on the recovered DS. */
8342 	if (ret == 0)
8343 		ret = nfsrv_createdsfile(vp, &fh, pf, dvp, ds, cred, p, &tvp);
8344 
8345 	/* Copy the DS file, if created successfully. */
8346 	if (ret == 0) {
8347 		/*
8348 		 * Get any NFSv4 ACL on the MDS file, so that it can be set
8349 		 * on the new DS file.
8350 		 */
8351 		aclp = acl_alloc(M_WAITOK | M_ZERO);
8352 		retacl = VOP_GETACL(vp, ACL_TYPE_NFS4, aclp, cred, p);
8353 		if (retacl != 0 && retacl != ENOATTR)
8354 			NFSD_DEBUG(1, "nfsrv_copymr: vop_getacl=%d\n", retacl);
8355 		dat = malloc(PNFSDS_COPYSIZ, M_TEMP, M_WAITOK);
8356 		/* Malloc a block of 0s used to check for holes. */
8357 		if (nfsrv_zeropnfsdat == NULL)
8358 			nfsrv_zeropnfsdat = malloc(PNFSDS_COPYSIZ, M_TEMP,
8359 			    M_WAITOK | M_ZERO);
8360 		rdpos = wrpos = 0;
8361 		ret = VOP_GETATTR(fvp, &va, cred);
8362 		aresid = 0;
8363 		while (ret == 0 && aresid == 0) {
8364 			ret = vn_rdwr(UIO_READ, fvp, dat, PNFSDS_COPYSIZ,
8365 			    rdpos, UIO_SYSSPACE, IO_NODELOCKED, cred, NULL,
8366 			    &aresid, p);
8367 			xfer = PNFSDS_COPYSIZ - aresid;
8368 			if (ret == 0 && xfer > 0) {
8369 				rdpos += xfer;
8370 				/*
8371 				 * Skip the write for holes, except for the
8372 				 * last block.
8373 				 */
8374 				if (xfer < PNFSDS_COPYSIZ || rdpos ==
8375 				    va.va_size || NFSBCMP(dat,
8376 				    nfsrv_zeropnfsdat, PNFSDS_COPYSIZ) != 0)
8377 					ret = vn_rdwr(UIO_WRITE, tvp, dat, xfer,
8378 					    wrpos, UIO_SYSSPACE, IO_NODELOCKED,
8379 					    cred, NULL, NULL, p);
8380 				if (ret == 0)
8381 					wrpos += xfer;
8382 			}
8383 		}
8384 
8385 		/* If there is an ACL and the copy succeeded, set the ACL. */
8386 		if (ret == 0 && retacl == 0) {
8387 			ret = VOP_SETACL(tvp, ACL_TYPE_NFS4, aclp, cred, p);
8388 			/*
8389 			 * Don't consider these as errors, since VOP_GETACL()
8390 			 * can return an ACL when they are not actually
8391 			 * supported.  For example, for UFS, VOP_GETACL()
8392 			 * will return a trivial ACL based on the uid/gid/mode
8393 			 * when there is no ACL on the file.
8394 			 * This case should be recognized as a trivial ACL
8395 			 * by UFS's VOP_SETACL() and succeed, but...
8396 			 */
8397 			if (ret == ENOATTR || ret == EOPNOTSUPP || ret == EPERM)
8398 				ret = 0;
8399 		}
8400 
8401 		if (ret == 0)
8402 			ret = VOP_FSYNC(tvp, MNT_WAIT, p);
8403 
8404 		/* Set the DS data file's modify time that of the MDS file. */
8405 		if (ret == 0)
8406 			ret = VOP_GETATTR(vp, &va, cred);
8407 		if (ret == 0) {
8408 			mtime = va.va_mtime;
8409 			VATTR_NULL(&va);
8410 			va.va_mtime = mtime;
8411 			ret = VOP_SETATTR(tvp, &va, cred);
8412 		}
8413 
8414 		vput(tvp);
8415 		acl_free(aclp);
8416 		free(dat, M_TEMP);
8417 	}
8418 	if (tvmp != NULL)
8419 		vn_finished_write(tvmp);
8420 
8421 	/* Update the extended attributes for the newly created DS file. */
8422 	if (ret == 0)
8423 		ret = vn_extattr_set(vp, IO_NODELOCKED,
8424 		    EXTATTR_NAMESPACE_SYSTEM, "pnfsd.dsfile",
8425 		    sizeof(*wpf) * mirrorcnt, (char *)wpf, p);
8426 	if (mp != NULL)
8427 		vn_finished_write(mp);
8428 
8429 	/* Get rid of the dontlist entry, so that Layouts can be issued. */
8430 	NFSDDONTLISTLOCK();
8431 	LIST_REMOVE(mrp, nfsmr_list);
8432 	NFSDDONTLISTUNLOCK();
8433 	free(mrp, M_NFSDSTATE);
8434 	return (ret);
8435 }
8436 
8437 /*
8438  * Create a data storage file on the recovered DS.
8439  */
8440 static int
8441 nfsrv_createdsfile(vnode_t vp, fhandle_t *fhp, struct pnfsdsfile *pf,
8442     vnode_t dvp, struct nfsdevice *ds, struct ucred *cred, NFSPROC_T *p,
8443     vnode_t *tvpp)
8444 {
8445 	struct vattr va, nva;
8446 	int error;
8447 
8448 	/* Make data file name based on FH. */
8449 	error = VOP_GETATTR(vp, &va, cred);
8450 	if (error == 0) {
8451 		/* Set the attributes for "vp" to Setattr the DS vp. */
8452 		VATTR_NULL(&nva);
8453 		nva.va_uid = va.va_uid;
8454 		nva.va_gid = va.va_gid;
8455 		nva.va_mode = va.va_mode;
8456 		nva.va_size = 0;
8457 		VATTR_NULL(&va);
8458 		va.va_type = VREG;
8459 		va.va_mode = nva.va_mode;
8460 		NFSD_DEBUG(4, "nfsrv_dscreatefile: dvp=%p pf=%p\n", dvp, pf);
8461 		error = nfsrv_dscreate(dvp, &va, &nva, fhp, pf, NULL,
8462 		    pf->dsf_filename, cred, p, tvpp);
8463 	}
8464 	return (error);
8465 }
8466 
8467 /*
8468  * Look up the MDS file shared locked, and then get the extended attribute
8469  * to find the extant DS file to be copied to the new mirror.
8470  * If successful, *vpp is set to the MDS file's vp and *nvpp is
8471  * set to a DS data file for the MDS file, both exclusively locked.
8472  * The "buf" argument has the pnfsdsfile structure from the MDS file
8473  * in it and buflen is set to its length.
8474  */
8475 int
8476 nfsrv_mdscopymr(char *mdspathp, char *dspathp, char *curdspathp, char *buf,
8477     int *buflenp, char *fname, NFSPROC_T *p, struct vnode **vpp,
8478     struct vnode **nvpp, struct pnfsdsfile **pfp, struct nfsdevice **dsp,
8479     struct nfsdevice **fdsp)
8480 {
8481 	struct nameidata nd;
8482 	struct vnode *vp, *curvp;
8483 	struct pnfsdsfile *pf;
8484 	struct nfsmount *nmp, *curnmp;
8485 	int dsdir, error, mirrorcnt, ippos;
8486 
8487 	vp = NULL;
8488 	curvp = NULL;
8489 	curnmp = NULL;
8490 	*dsp = NULL;
8491 	*fdsp = NULL;
8492 	if (dspathp == NULL && curdspathp != NULL)
8493 		return (EPERM);
8494 
8495 	/*
8496 	 * Look up the MDS file shared locked.  The lock will be upgraded
8497 	 * to an exclusive lock after any rw layouts have been returned.
8498 	 */
8499 	NFSD_DEBUG(4, "mdsopen path=%s\n", mdspathp);
8500 	NDINIT(&nd, LOOKUP, FOLLOW | LOCKSHARED | LOCKLEAF, UIO_SYSSPACE,
8501 	    mdspathp, p);
8502 	error = namei(&nd);
8503 	NFSD_DEBUG(4, "lookup=%d\n", error);
8504 	if (error != 0)
8505 		return (error);
8506 	if (nd.ni_vp->v_type != VREG) {
8507 		vput(nd.ni_vp);
8508 		NFSD_DEBUG(4, "mdspath not reg\n");
8509 		return (EISDIR);
8510 	}
8511 	vp = nd.ni_vp;
8512 
8513 	if (curdspathp != NULL) {
8514 		/*
8515 		 * Look up the current DS path and find the nfsdev structure for
8516 		 * it.
8517 		 */
8518 		NFSD_DEBUG(4, "curmdsdev path=%s\n", curdspathp);
8519 		NDINIT(&nd, LOOKUP, FOLLOW | LOCKSHARED | LOCKLEAF,
8520 		    UIO_SYSSPACE, curdspathp, p);
8521 		error = namei(&nd);
8522 		NFSD_DEBUG(4, "ds lookup=%d\n", error);
8523 		if (error != 0) {
8524 			vput(vp);
8525 			return (error);
8526 		}
8527 		if (nd.ni_vp->v_type != VDIR) {
8528 			vput(nd.ni_vp);
8529 			vput(vp);
8530 			NFSD_DEBUG(4, "curdspath not dir\n");
8531 			return (ENOTDIR);
8532 		}
8533 		if (strcmp(nd.ni_vp->v_mount->mnt_vfc->vfc_name, "nfs") != 0) {
8534 			vput(nd.ni_vp);
8535 			vput(vp);
8536 			NFSD_DEBUG(4, "curdspath not an NFS mount\n");
8537 			return (ENXIO);
8538 		}
8539 		curnmp = VFSTONFS(nd.ni_vp->v_mount);
8540 
8541 		/* Search the nfsdev list for a match. */
8542 		NFSDDSLOCK();
8543 		*fdsp = nfsv4_findmirror(curnmp);
8544 		NFSDDSUNLOCK();
8545 		if (*fdsp == NULL)
8546 			curnmp = NULL;
8547 		if (curnmp == NULL) {
8548 			vput(nd.ni_vp);
8549 			vput(vp);
8550 			NFSD_DEBUG(4, "mdscopymr: no current ds\n");
8551 			return (ENXIO);
8552 		}
8553 		curvp = nd.ni_vp;
8554 	}
8555 
8556 	if (dspathp != NULL) {
8557 		/* Look up the nfsdev path and find the nfsdev structure. */
8558 		NFSD_DEBUG(4, "mdsdev path=%s\n", dspathp);
8559 		NDINIT(&nd, LOOKUP, FOLLOW | LOCKSHARED | LOCKLEAF,
8560 		    UIO_SYSSPACE, dspathp, p);
8561 		error = namei(&nd);
8562 		NFSD_DEBUG(4, "ds lookup=%d\n", error);
8563 		if (error != 0) {
8564 			vput(vp);
8565 			if (curvp != NULL)
8566 				vput(curvp);
8567 			return (error);
8568 		}
8569 		if (nd.ni_vp->v_type != VDIR || nd.ni_vp == curvp) {
8570 			vput(nd.ni_vp);
8571 			vput(vp);
8572 			if (curvp != NULL)
8573 				vput(curvp);
8574 			NFSD_DEBUG(4, "dspath not dir\n");
8575 			if (nd.ni_vp == curvp)
8576 				return (EPERM);
8577 			return (ENOTDIR);
8578 		}
8579 		if (strcmp(nd.ni_vp->v_mount->mnt_vfc->vfc_name, "nfs") != 0) {
8580 			vput(nd.ni_vp);
8581 			vput(vp);
8582 			if (curvp != NULL)
8583 				vput(curvp);
8584 			NFSD_DEBUG(4, "dspath not an NFS mount\n");
8585 			return (ENXIO);
8586 		}
8587 		nmp = VFSTONFS(nd.ni_vp->v_mount);
8588 
8589 		/*
8590 		 * Search the nfsdevice list for a match.  If curnmp == NULL,
8591 		 * this is a recovery and there must be a mirror.
8592 		 */
8593 		NFSDDSLOCK();
8594 		if (curnmp == NULL)
8595 			*dsp = nfsrv_findmirroredds(nmp);
8596 		else
8597 			*dsp = nfsv4_findmirror(nmp);
8598 		NFSDDSUNLOCK();
8599 		if (*dsp == NULL) {
8600 			vput(nd.ni_vp);
8601 			vput(vp);
8602 			if (curvp != NULL)
8603 				vput(curvp);
8604 			NFSD_DEBUG(4, "mdscopymr: no ds\n");
8605 			return (ENXIO);
8606 		}
8607 	} else {
8608 		nd.ni_vp = NULL;
8609 		nmp = NULL;
8610 	}
8611 
8612 	/*
8613 	 * Get a vp for an available DS data file using the extended
8614 	 * attribute on the MDS file.
8615 	 * If there is a valid entry for the new DS in the extended attribute
8616 	 * on the MDS file (as checked via the nmp argument),
8617 	 * nfsrv_dsgetsockmnt() returns EEXIST, so no copying will occur.
8618 	 */
8619 	error = nfsrv_dsgetsockmnt(vp, 0, buf, buflenp, &mirrorcnt, p,
8620 	    NULL, NULL, NULL, fname, nvpp, &nmp, curnmp, &ippos, &dsdir);
8621 	if (curvp != NULL)
8622 		vput(curvp);
8623 	if (nd.ni_vp == NULL) {
8624 		if (error == 0 && nmp != NULL) {
8625 			/* Search the nfsdev list for a match. */
8626 			NFSDDSLOCK();
8627 			*dsp = nfsrv_findmirroredds(nmp);
8628 			NFSDDSUNLOCK();
8629 		}
8630 		if (error == 0 && (nmp == NULL || *dsp == NULL)) {
8631 			if (nvpp != NULL && *nvpp != NULL) {
8632 				vput(*nvpp);
8633 				*nvpp = NULL;
8634 			}
8635 			error = ENXIO;
8636 		}
8637 	} else
8638 		vput(nd.ni_vp);
8639 
8640 	/*
8641 	 * When dspathp != NULL and curdspathp == NULL, this is a recovery
8642 	 * and is only allowed if there is a 0.0.0.0 IP address entry.
8643 	 * When curdspathp != NULL, the ippos will be set to that entry.
8644 	 */
8645 	if (error == 0 && dspathp != NULL && ippos == -1) {
8646 		if (nvpp != NULL && *nvpp != NULL) {
8647 			vput(*nvpp);
8648 			*nvpp = NULL;
8649 		}
8650 		error = ENXIO;
8651 	}
8652 	if (error == 0) {
8653 		*vpp = vp;
8654 
8655 		pf = (struct pnfsdsfile *)buf;
8656 		if (ippos == -1) {
8657 			/* If no zeroip pnfsdsfile, add one. */
8658 			ippos = *buflenp / sizeof(*pf);
8659 			*buflenp += sizeof(*pf);
8660 			pf += ippos;
8661 			pf->dsf_dir = dsdir;
8662 			strlcpy(pf->dsf_filename, fname,
8663 			    sizeof(pf->dsf_filename));
8664 		} else
8665 			pf += ippos;
8666 		*pfp = pf;
8667 	} else
8668 		vput(vp);
8669 	return (error);
8670 }
8671 
8672 /*
8673  * Search for a matching pnfsd mirror device structure, base on the nmp arg.
8674  * Return one if found, NULL otherwise.
8675  */
8676 static struct nfsdevice *
8677 nfsrv_findmirroredds(struct nfsmount *nmp)
8678 {
8679 	struct nfsdevice *ds, *fndds;
8680 	int fndmirror;
8681 
8682 	mtx_assert(NFSDDSMUTEXPTR, MA_OWNED);
8683 	/*
8684 	 * Search the DS server list for a match with nmp.
8685 	 * Remove the DS entry if found and there is a mirror.
8686 	 */
8687 	fndds = NULL;
8688 	fndmirror = 0;
8689 	if (nfsrv_devidcnt == 0)
8690 		return (fndds);
8691 	TAILQ_FOREACH(ds, &nfsrv_devidhead, nfsdev_list) {
8692 		if (ds->nfsdev_nmp == nmp) {
8693 			NFSD_DEBUG(4, "nfsrv_findmirroredds: fnd main ds\n");
8694 			fndds = ds;
8695 			break;
8696 		}
8697 	}
8698 	if (fndds == NULL)
8699 		return (fndds);
8700 	if (fndds->nfsdev_mdsisset == 0 && nfsrv_faildscnt > 0)
8701 		fndmirror = 1;
8702 	else if (fndds->nfsdev_mdsisset != 0) {
8703 		/* For the fsid is set case, search for a mirror. */
8704 		TAILQ_FOREACH(ds, &nfsrv_devidhead, nfsdev_list) {
8705 			if (ds != fndds && ds->nfsdev_nmp != NULL &&
8706 			    ds->nfsdev_mdsisset != 0 &&
8707 			    ds->nfsdev_mdsfsid.val[0] ==
8708 			    fndds->nfsdev_mdsfsid.val[0] &&
8709 			    ds->nfsdev_mdsfsid.val[1] ==
8710 			    fndds->nfsdev_mdsfsid.val[1]) {
8711 				fndmirror = 1;
8712 				break;
8713 			}
8714 		}
8715 	}
8716 	if (fndmirror == 0) {
8717 		NFSD_DEBUG(4, "nfsrv_findmirroredds: no mirror for DS\n");
8718 		return (NULL);
8719 	}
8720 	return (fndds);
8721 }
8722 
8723