xref: /freebsd/sys/fs/nfsserver/nfs_nfsdstate.c (revision b197d4b893974c9eb4d7b38704c6d5c486235d6f)
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 #include <sys/extattr.h>
36 #include <fs/nfs/nfsport.h>
37 
38 struct nfsrv_stablefirst nfsrv_stablefirst;
39 int nfsrv_issuedelegs = 0;
40 int nfsrv_dolocallocks = 0;
41 struct nfsv4lock nfsv4rootfs_lock;
42 time_t nfsdev_time = 0;
43 int nfsrv_layouthashsize;
44 volatile int nfsrv_layoutcnt = 0;
45 extern uint32_t nfs_srvmaxio;
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 
127 static u_int32_t nfsrv_openpluslock = 0, nfsrv_delegatecnt = 0;
128 static time_t nfsrvboottime;
129 static int nfsrv_returnoldstateid = 0, nfsrv_clients = 0;
130 static int nfsrv_clienthighwater = NFSRV_CLIENTHIGHWATER;
131 static int nfsrv_nogsscallback = 0;
132 static volatile int nfsrv_writedelegcnt = 0;
133 static int nfsrv_faildscnt;
134 
135 /* local functions */
136 static void nfsrv_dumpaclient(struct nfsclient *clp,
137     struct nfsd_dumpclients *dumpp);
138 static void nfsrv_freeopenowner(struct nfsstate *stp, int cansleep,
139     NFSPROC_T *p);
140 static int nfsrv_freeopen(struct nfsstate *stp, vnode_t vp, int cansleep,
141     NFSPROC_T *p);
142 static void nfsrv_freelockowner(struct nfsstate *stp, vnode_t vp, int cansleep,
143     NFSPROC_T *p);
144 static void nfsrv_freeallnfslocks(struct nfsstate *stp, vnode_t vp,
145     int cansleep, NFSPROC_T *p);
146 static void nfsrv_freenfslock(struct nfslock *lop);
147 static void nfsrv_freenfslockfile(struct nfslockfile *lfp);
148 static void nfsrv_freedeleg(struct nfsstate *);
149 static int nfsrv_getstate(struct nfsclient *clp, nfsv4stateid_t *stateidp,
150     u_int32_t flags, struct nfsstate **stpp);
151 static void nfsrv_getowner(struct nfsstatehead *hp, struct nfsstate *new_stp,
152     struct nfsstate **stpp);
153 static int nfsrv_getlockfh(vnode_t vp, u_short flags,
154     struct nfslockfile *new_lfp, fhandle_t *nfhp, NFSPROC_T *p);
155 static int nfsrv_getlockfile(u_short flags, struct nfslockfile **new_lfpp,
156     struct nfslockfile **lfpp, fhandle_t *nfhp, int lockit);
157 static void nfsrv_insertlock(struct nfslock *new_lop,
158     struct nfslock *insert_lop, struct nfsstate *stp, struct nfslockfile *lfp);
159 static void nfsrv_updatelock(struct nfsstate *stp, struct nfslock **new_lopp,
160     struct nfslock **other_lopp, struct nfslockfile *lfp);
161 static int nfsrv_getipnumber(u_char *cp);
162 static int nfsrv_checkrestart(nfsquad_t clientid, u_int32_t flags,
163     nfsv4stateid_t *stateidp, int specialid);
164 static int nfsrv_checkgrace(struct nfsrv_descript *nd, struct nfsclient *clp,
165     u_int32_t flags);
166 static int nfsrv_docallback(struct nfsclient *clp, int procnum,
167     nfsv4stateid_t *stateidp, int trunc, fhandle_t *fhp,
168     struct nfsvattr *nap, nfsattrbit_t *attrbitp, int laytype, NFSPROC_T *p);
169 static int nfsrv_cbcallargs(struct nfsrv_descript *nd, struct nfsclient *clp,
170     uint32_t callback, int op, const char *optag, struct nfsdsession **sepp,
171     int *slotposp);
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, int *slotposp);
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 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 			confirmp->lval[0] = ++confirm_index;
341 			new_clp->lc_confirm.lval[0] = confirmp->lval[0] - 1;
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_prevsess = 0;
352 		new_clp->lc_cbref = 0;
353 		new_clp->lc_expiry = nfsrv_leaseexpiry();
354 		LIST_INIT(&new_clp->lc_open);
355 		LIST_INIT(&new_clp->lc_deleg);
356 		LIST_INIT(&new_clp->lc_olddeleg);
357 		LIST_INIT(&new_clp->lc_session);
358 		for (i = 0; i < nfsrv_statehashsize; i++)
359 			LIST_INIT(&new_clp->lc_stateid[i]);
360 		LIST_INSERT_HEAD(NFSCLIENTHASH(new_clp->lc_clientid), new_clp,
361 		    lc_hash);
362 		nfsstatsv1.srvclients++;
363 		nfsrv_openpluslock++;
364 		nfsrv_clients++;
365 		NFSLOCKV4ROOTMUTEX();
366 		nfsv4_unlock(&nfsv4rootfs_lock, 1);
367 		NFSUNLOCKV4ROOTMUTEX();
368 		if (zapit)
369 			nfsrv_zapclient(clp, p);
370 		*new_clpp = NULL;
371 		goto out;
372 	}
373 
374 	/*
375 	 * Now, handle the cases where the id is already issued.
376 	 */
377 	if (nfsrv_notsamecredname(nd, clp)) {
378 	    /*
379 	     * Check to see if there is expired state that should go away.
380 	     */
381 	    if (clp->lc_expiry < NFSD_MONOSEC &&
382 	        (!LIST_EMPTY(&clp->lc_open) || !LIST_EMPTY(&clp->lc_deleg))) {
383 		nfsrv_cleanclient(clp, p);
384 		nfsrv_freedeleglist(&clp->lc_deleg);
385 	    }
386 
387 	    /*
388 	     * If there is outstanding state, then reply NFSERR_CLIDINUSE per
389 	     * RFC3530 Sec. 8.1.2 last para.
390 	     */
391 	    if (!LIST_EMPTY(&clp->lc_deleg)) {
392 		hasstate = 1;
393 	    } else if (LIST_EMPTY(&clp->lc_open)) {
394 		hasstate = 0;
395 	    } else {
396 		hasstate = 0;
397 		/* Look for an Open on the OpenOwner */
398 		LIST_FOREACH(stp, &clp->lc_open, ls_list) {
399 		    if (!LIST_EMPTY(&stp->ls_open)) {
400 			hasstate = 1;
401 			break;
402 		    }
403 		}
404 	    }
405 	    if (hasstate) {
406 		/*
407 		 * If the uid doesn't match, return NFSERR_CLIDINUSE after
408 		 * filling out the correct ipaddr and portnum.
409 		 */
410 		switch (clp->lc_req.nr_nam->sa_family) {
411 #ifdef INET
412 		case AF_INET:
413 			sin = (struct sockaddr_in *)new_clp->lc_req.nr_nam;
414 			rin = (struct sockaddr_in *)clp->lc_req.nr_nam;
415 			sin->sin_addr.s_addr = rin->sin_addr.s_addr;
416 			sin->sin_port = rin->sin_port;
417 			break;
418 #endif
419 #ifdef INET6
420 		case AF_INET6:
421 			sin6 = (struct sockaddr_in6 *)new_clp->lc_req.nr_nam;
422 			rin6 = (struct sockaddr_in6 *)clp->lc_req.nr_nam;
423 			sin6->sin6_addr = rin6->sin6_addr;
424 			sin6->sin6_port = rin6->sin6_port;
425 			break;
426 #endif
427 		}
428 		NFSLOCKV4ROOTMUTEX();
429 		nfsv4_unlock(&nfsv4rootfs_lock, 1);
430 		NFSUNLOCKV4ROOTMUTEX();
431 		error = NFSERR_CLIDINUSE;
432 		goto out;
433 	    }
434 	}
435 
436 	if (NFSBCMP(new_clp->lc_verf, clp->lc_verf, NFSX_VERF)) {
437 		/*
438 		 * If the verifier has changed, the client has rebooted
439 		 * and a new client id is issued. The old state info
440 		 * can be thrown away once the SETCLIENTID_CONFIRM occurs.
441 		 */
442 		LIST_REMOVE(clp, lc_hash);
443 
444 		/* Get rid of all sessions on this clientid. */
445 		LIST_FOREACH_SAFE(sep, &clp->lc_session, sess_list, nsep) {
446 			ret = nfsrv_freesession(sep, NULL);
447 			if (ret != 0)
448 				printf("nfsrv_setclient: verifier changed free"
449 				    " session failed=%d\n", ret);
450 		}
451 
452 		new_clp->lc_flags |= LCL_NEEDSCONFIRM;
453 		if ((nd->nd_flag & ND_NFSV41) != 0) {
454 			confirmp->lval[0] = ++confirm_index;
455 			new_clp->lc_confirm.lval[0] = confirmp->lval[0] - 1;
456 		} else
457 			confirmp->qval = new_clp->lc_confirm.qval =
458 			    ++confirm_index;
459 		clientidp->lval[0] = new_clp->lc_clientid.lval[0] =
460 		    nfsrvboottime;
461 		clientidp->lval[1] = new_clp->lc_clientid.lval[1] =
462 		    nfsrv_nextclientindex();
463 		new_clp->lc_stateindex = 0;
464 		new_clp->lc_statemaxindex = 0;
465 		new_clp->lc_prevsess = 0;
466 		new_clp->lc_cbref = 0;
467 		new_clp->lc_expiry = nfsrv_leaseexpiry();
468 
469 		/*
470 		 * Save the state until confirmed.
471 		 */
472 		LIST_NEWHEAD(&new_clp->lc_open, &clp->lc_open, ls_list);
473 		LIST_FOREACH(tstp, &new_clp->lc_open, ls_list)
474 			tstp->ls_clp = new_clp;
475 		LIST_NEWHEAD(&new_clp->lc_deleg, &clp->lc_deleg, ls_list);
476 		LIST_FOREACH(tstp, &new_clp->lc_deleg, ls_list)
477 			tstp->ls_clp = new_clp;
478 		LIST_NEWHEAD(&new_clp->lc_olddeleg, &clp->lc_olddeleg,
479 		    ls_list);
480 		LIST_FOREACH(tstp, &new_clp->lc_olddeleg, ls_list)
481 			tstp->ls_clp = new_clp;
482 		for (i = 0; i < nfsrv_statehashsize; i++) {
483 			LIST_NEWHEAD(&new_clp->lc_stateid[i],
484 			    &clp->lc_stateid[i], ls_hash);
485 			LIST_FOREACH(tstp, &new_clp->lc_stateid[i], ls_hash)
486 				tstp->ls_clp = new_clp;
487 		}
488 		LIST_INIT(&new_clp->lc_session);
489 		LIST_INSERT_HEAD(NFSCLIENTHASH(new_clp->lc_clientid), new_clp,
490 		    lc_hash);
491 		nfsstatsv1.srvclients++;
492 		nfsrv_openpluslock++;
493 		nfsrv_clients++;
494 		NFSLOCKV4ROOTMUTEX();
495 		nfsv4_unlock(&nfsv4rootfs_lock, 1);
496 		NFSUNLOCKV4ROOTMUTEX();
497 
498 		/*
499 		 * Must wait until any outstanding callback on the old clp
500 		 * completes.
501 		 */
502 		NFSLOCKSTATE();
503 		while (clp->lc_cbref) {
504 			clp->lc_flags |= LCL_WAKEUPWANTED;
505 			(void)mtx_sleep(clp, NFSSTATEMUTEXPTR, PZERO - 1,
506 			    "nfsd clp", 10 * hz);
507 		}
508 		NFSUNLOCKSTATE();
509 		nfsrv_zapclient(clp, p);
510 		*new_clpp = NULL;
511 		goto out;
512 	}
513 
514 	/* For NFSv4.1, mark that we found a confirmed clientid. */
515 	if ((nd->nd_flag & ND_NFSV41) != 0) {
516 		clientidp->lval[0] = clp->lc_clientid.lval[0];
517 		clientidp->lval[1] = clp->lc_clientid.lval[1];
518 		confirmp->lval[0] = 0;	/* Ignored by client */
519 		confirmp->lval[1] = 1;
520 	} else {
521 		/*
522 		 * id and verifier match, so update the net address info
523 		 * and get rid of any existing callback authentication
524 		 * handle, so a new one will be acquired.
525 		 */
526 		LIST_REMOVE(clp, lc_hash);
527 		new_clp->lc_flags |= (LCL_NEEDSCONFIRM | LCL_DONTCLEAN);
528 		new_clp->lc_expiry = nfsrv_leaseexpiry();
529 		confirmp->qval = new_clp->lc_confirm.qval = ++confirm_index;
530 		clientidp->lval[0] = new_clp->lc_clientid.lval[0] =
531 		    clp->lc_clientid.lval[0];
532 		clientidp->lval[1] = new_clp->lc_clientid.lval[1] =
533 		    clp->lc_clientid.lval[1];
534 		new_clp->lc_delegtime = clp->lc_delegtime;
535 		new_clp->lc_stateindex = clp->lc_stateindex;
536 		new_clp->lc_statemaxindex = clp->lc_statemaxindex;
537 		new_clp->lc_cbref = 0;
538 		LIST_NEWHEAD(&new_clp->lc_open, &clp->lc_open, ls_list);
539 		LIST_FOREACH(tstp, &new_clp->lc_open, ls_list)
540 			tstp->ls_clp = new_clp;
541 		LIST_NEWHEAD(&new_clp->lc_deleg, &clp->lc_deleg, ls_list);
542 		LIST_FOREACH(tstp, &new_clp->lc_deleg, ls_list)
543 			tstp->ls_clp = new_clp;
544 		LIST_NEWHEAD(&new_clp->lc_olddeleg, &clp->lc_olddeleg, ls_list);
545 		LIST_FOREACH(tstp, &new_clp->lc_olddeleg, ls_list)
546 			tstp->ls_clp = new_clp;
547 		for (i = 0; i < nfsrv_statehashsize; i++) {
548 			LIST_NEWHEAD(&new_clp->lc_stateid[i],
549 			    &clp->lc_stateid[i], ls_hash);
550 			LIST_FOREACH(tstp, &new_clp->lc_stateid[i], ls_hash)
551 				tstp->ls_clp = new_clp;
552 		}
553 		LIST_INIT(&new_clp->lc_session);
554 		LIST_INSERT_HEAD(NFSCLIENTHASH(new_clp->lc_clientid), new_clp,
555 		    lc_hash);
556 		nfsstatsv1.srvclients++;
557 		nfsrv_openpluslock++;
558 		nfsrv_clients++;
559 	}
560 	NFSLOCKV4ROOTMUTEX();
561 	nfsv4_unlock(&nfsv4rootfs_lock, 1);
562 	NFSUNLOCKV4ROOTMUTEX();
563 
564 	if ((nd->nd_flag & ND_NFSV41) == 0) {
565 		/*
566 		 * Must wait until any outstanding callback on the old clp
567 		 * completes.
568 		 */
569 		NFSLOCKSTATE();
570 		while (clp->lc_cbref) {
571 			clp->lc_flags |= LCL_WAKEUPWANTED;
572 			(void)mtx_sleep(clp, NFSSTATEMUTEXPTR, PZERO - 1,
573 			    "nfsdclp", 10 * hz);
574 		}
575 		NFSUNLOCKSTATE();
576 		nfsrv_zapclient(clp, p);
577 		*new_clpp = NULL;
578 	}
579 
580 out:
581 	NFSEXITCODE2(error, nd);
582 	return (error);
583 }
584 
585 /*
586  * Check to see if the client id exists and optionally confirm it.
587  */
588 int
589 nfsrv_getclient(nfsquad_t clientid, int opflags, struct nfsclient **clpp,
590     struct nfsdsession *nsep, nfsquad_t confirm, uint32_t cbprogram,
591     struct nfsrv_descript *nd, NFSPROC_T *p)
592 {
593 	struct nfsclient *clp;
594 	struct nfsstate *stp;
595 	int i;
596 	struct nfsclienthashhead *hp;
597 	int error = 0, igotlock, doneok;
598 	struct nfssessionhash *shp;
599 	struct nfsdsession *sep;
600 	uint64_t sessid[2];
601 	bool sess_replay;
602 	static uint64_t next_sess = 0;
603 
604 	if (clpp)
605 		*clpp = NULL;
606 	if ((nd == NULL || (nd->nd_flag & ND_NFSV41) == 0 ||
607 	    opflags != CLOPS_RENEW) && nfsrvboottime != clientid.lval[0]) {
608 		error = NFSERR_STALECLIENTID;
609 		goto out;
610 	}
611 
612 	/*
613 	 * If called with opflags == CLOPS_RENEW, the State Lock is
614 	 * already held. Otherwise, we need to get either that or,
615 	 * for the case of Confirm, lock out the nfsd threads.
616 	 */
617 	if (opflags & CLOPS_CONFIRM) {
618 		NFSLOCKV4ROOTMUTEX();
619 		nfsv4_relref(&nfsv4rootfs_lock);
620 		do {
621 			igotlock = nfsv4_lock(&nfsv4rootfs_lock, 1, NULL,
622 			    NFSV4ROOTLOCKMUTEXPTR, NULL);
623 		} while (!igotlock);
624 		/*
625 		 * Create a new sessionid here, since we need to do it where
626 		 * there is a mutex held to serialize update of next_sess.
627 		 */
628 		if ((nd->nd_flag & ND_NFSV41) != 0) {
629 			sessid[0] = ++next_sess;
630 			sessid[1] = clientid.qval;
631 		}
632 		NFSUNLOCKV4ROOTMUTEX();
633 	} else if (opflags != CLOPS_RENEW) {
634 		NFSLOCKSTATE();
635 	}
636 
637 	/* For NFSv4.1, the clp is acquired from the associated session. */
638 	if (nd != NULL && (nd->nd_flag & ND_NFSV41) != 0 &&
639 	    opflags == CLOPS_RENEW) {
640 		clp = NULL;
641 		if ((nd->nd_flag & ND_HASSEQUENCE) != 0) {
642 			shp = NFSSESSIONHASH(nd->nd_sessionid);
643 			NFSLOCKSESSION(shp);
644 			sep = nfsrv_findsession(nd->nd_sessionid);
645 			if (sep != NULL)
646 				clp = sep->sess_clp;
647 			NFSUNLOCKSESSION(shp);
648 		}
649 	} else {
650 		hp = NFSCLIENTHASH(clientid);
651 		LIST_FOREACH(clp, hp, lc_hash) {
652 			if (clp->lc_clientid.lval[1] == clientid.lval[1])
653 				break;
654 		}
655 	}
656 	if (clp == NULL) {
657 		if (opflags & CLOPS_CONFIRM)
658 			error = NFSERR_STALECLIENTID;
659 		else
660 			error = NFSERR_EXPIRED;
661 	} else if (clp->lc_flags & LCL_ADMINREVOKED) {
662 		/*
663 		 * If marked admin revoked, just return the error.
664 		 */
665 		error = NFSERR_ADMINREVOKED;
666 	}
667 	if (error) {
668 		if (opflags & CLOPS_CONFIRM) {
669 			NFSLOCKV4ROOTMUTEX();
670 			nfsv4_unlock(&nfsv4rootfs_lock, 1);
671 			NFSUNLOCKV4ROOTMUTEX();
672 		} else if (opflags != CLOPS_RENEW) {
673 			NFSUNLOCKSTATE();
674 		}
675 		goto out;
676 	}
677 
678 	/*
679 	 * Perform any operations specified by the opflags.
680 	 */
681 	if (opflags & CLOPS_CONFIRM) {
682 		sess_replay = false;
683 		if ((nd->nd_flag & ND_NFSV41) != 0) {
684 		    /*
685 		     * For the case where lc_confirm.lval[0] == confirm.lval[0],
686 		     * use the new session, but with the previous sessionid.
687 		     * This is not exactly what the RFC describes, but should
688 		     * result in the same reply as the previous CreateSession.
689 		     */
690 		    if (clp->lc_confirm.lval[0] + 1 == confirm.lval[0]) {
691 			clp->lc_confirm.lval[0] = confirm.lval[0];
692 			clp->lc_prevsess = sessid[0];
693 		    } else if (clp->lc_confirm.lval[0] == confirm.lval[0]) {
694 			if (clp->lc_prevsess == 0)
695 			    error = NFSERR_SEQMISORDERED;
696 			else
697 			    sessid[0] = clp->lc_prevsess;
698 			sess_replay = true;
699 		    } else
700 			error = NFSERR_SEQMISORDERED;
701 		} else if ((nd->nd_flag & ND_NFSV41) == 0 &&
702 		     clp->lc_confirm.qval != confirm.qval)
703 			error = NFSERR_STALECLIENTID;
704 		if (error == 0 && nfsrv_notsamecredname(nd, clp))
705 			error = NFSERR_CLIDINUSE;
706 
707 		if (!error) {
708 		    if ((clp->lc_flags & (LCL_NEEDSCONFIRM | LCL_DONTCLEAN)) ==
709 			LCL_NEEDSCONFIRM) {
710 			/*
711 			 * Hang onto the delegations (as old delegations)
712 			 * for an Open with CLAIM_DELEGATE_PREV unless in
713 			 * grace, but get rid of the rest of the state.
714 			 */
715 			nfsrv_cleanclient(clp, p);
716 			nfsrv_freedeleglist(&clp->lc_olddeleg);
717 			if (nfsrv_checkgrace(nd, clp, 0)) {
718 			    /* In grace, so just delete delegations */
719 			    nfsrv_freedeleglist(&clp->lc_deleg);
720 			} else {
721 			    LIST_FOREACH(stp, &clp->lc_deleg, ls_list)
722 				stp->ls_flags |= NFSLCK_OLDDELEG;
723 			    clp->lc_delegtime = NFSD_MONOSEC +
724 				nfsrv_lease + NFSRV_LEASEDELTA;
725 			    LIST_NEWHEAD(&clp->lc_olddeleg, &clp->lc_deleg,
726 				ls_list);
727 			}
728 			if ((nd->nd_flag & ND_NFSV41) != 0)
729 			    clp->lc_program = cbprogram;
730 		    }
731 		    clp->lc_flags &= ~(LCL_NEEDSCONFIRM | LCL_DONTCLEAN);
732 		    if (clp->lc_program)
733 			clp->lc_flags |= LCL_NEEDSCBNULL;
734 		    /* For NFSv4.1, link the session onto the client. */
735 		    if (nsep != NULL) {
736 			/* Hold a reference on the xprt for a backchannel. */
737 			if ((nsep->sess_crflags & NFSV4CRSESS_CONNBACKCHAN)
738 			    != 0 && !sess_replay) {
739 			    if (clp->lc_req.nr_client == NULL)
740 				clp->lc_req.nr_client = (struct __rpc_client *)
741 				    clnt_bck_create(nd->nd_xprt->xp_socket,
742 				    cbprogram, NFSV4_CBVERS);
743 			    if (clp->lc_req.nr_client != NULL) {
744 				SVC_ACQUIRE(nd->nd_xprt);
745 				CLNT_ACQUIRE(clp->lc_req.nr_client);
746 				nd->nd_xprt->xp_p2 = clp->lc_req.nr_client;
747 				/* Disable idle timeout. */
748 				nd->nd_xprt->xp_idletimeout = 0;
749 				nsep->sess_cbsess.nfsess_xprt = nd->nd_xprt;
750 			    } else
751 				nsep->sess_crflags &= ~NFSV4CRSESS_CONNBACKCHAN;
752 			}
753 			NFSBCOPY(sessid, nsep->sess_sessionid,
754 			    NFSX_V4SESSIONID);
755 			NFSBCOPY(sessid, nsep->sess_cbsess.nfsess_sessionid,
756 			    NFSX_V4SESSIONID);
757 			if (!sess_replay) {
758 			    shp = NFSSESSIONHASH(nsep->sess_sessionid);
759 			    NFSLOCKSTATE();
760 			    NFSLOCKSESSION(shp);
761 			    LIST_INSERT_HEAD(&shp->list, nsep, sess_hash);
762 			    LIST_INSERT_HEAD(&clp->lc_session, nsep, sess_list);
763 			    nsep->sess_clp = clp;
764 			    NFSUNLOCKSESSION(shp);
765 			    NFSUNLOCKSTATE();
766 			}
767 		    }
768 		}
769 	} else if (clp->lc_flags & LCL_NEEDSCONFIRM) {
770 		error = NFSERR_EXPIRED;
771 	}
772 
773 	/*
774 	 * If called by the Renew Op, we must check the principal.
775 	 */
776 	if (!error && (opflags & CLOPS_RENEWOP)) {
777 	    if (nfsrv_notsamecredname(nd, clp)) {
778 		doneok = 0;
779 		for (i = 0; i < nfsrv_statehashsize && doneok == 0; i++) {
780 		    LIST_FOREACH(stp, &clp->lc_stateid[i], ls_hash) {
781 			if ((stp->ls_flags & NFSLCK_OPEN) &&
782 			    stp->ls_uid == nd->nd_cred->cr_uid) {
783 				doneok = 1;
784 				break;
785 			}
786 		    }
787 		}
788 		if (!doneok)
789 			error = NFSERR_ACCES;
790 	    }
791 	    if (!error && (clp->lc_flags & LCL_CBDOWN))
792 		error = NFSERR_CBPATHDOWN;
793 	}
794 	if ((!error || error == NFSERR_CBPATHDOWN) &&
795 	     (opflags & CLOPS_RENEW)) {
796 		clp->lc_expiry = nfsrv_leaseexpiry();
797 	}
798 	if (opflags & CLOPS_CONFIRM) {
799 		NFSLOCKV4ROOTMUTEX();
800 		nfsv4_unlock(&nfsv4rootfs_lock, 1);
801 		NFSUNLOCKV4ROOTMUTEX();
802 	} else if (opflags != CLOPS_RENEW) {
803 		NFSUNLOCKSTATE();
804 	}
805 	if (clpp)
806 		*clpp = clp;
807 
808 out:
809 	NFSEXITCODE2(error, nd);
810 	return (error);
811 }
812 
813 /*
814  * Perform the NFSv4.1 destroy clientid.
815  */
816 int
817 nfsrv_destroyclient(nfsquad_t clientid, NFSPROC_T *p)
818 {
819 	struct nfsclient *clp;
820 	struct nfsclienthashhead *hp;
821 	int error = 0, i, igotlock;
822 
823 	if (nfsrvboottime != clientid.lval[0]) {
824 		error = NFSERR_STALECLIENTID;
825 		goto out;
826 	}
827 
828 	/* Lock out other nfsd threads */
829 	NFSLOCKV4ROOTMUTEX();
830 	nfsv4_relref(&nfsv4rootfs_lock);
831 	do {
832 		igotlock = nfsv4_lock(&nfsv4rootfs_lock, 1, NULL,
833 		    NFSV4ROOTLOCKMUTEXPTR, NULL);
834 	} while (igotlock == 0);
835 	NFSUNLOCKV4ROOTMUTEX();
836 
837 	hp = NFSCLIENTHASH(clientid);
838 	LIST_FOREACH(clp, hp, lc_hash) {
839 		if (clp->lc_clientid.lval[1] == clientid.lval[1])
840 			break;
841 	}
842 	if (clp == NULL) {
843 		NFSLOCKV4ROOTMUTEX();
844 		nfsv4_unlock(&nfsv4rootfs_lock, 1);
845 		NFSUNLOCKV4ROOTMUTEX();
846 		/* Just return ok, since it is gone. */
847 		goto out;
848 	}
849 
850 	/*
851 	 * Free up all layouts on the clientid.  Should the client return the
852 	 * layouts?
853 	 */
854 	nfsrv_freelayoutlist(clientid);
855 
856 	/* Scan for state on the clientid. */
857 	for (i = 0; i < nfsrv_statehashsize; i++)
858 		if (!LIST_EMPTY(&clp->lc_stateid[i])) {
859 			NFSLOCKV4ROOTMUTEX();
860 			nfsv4_unlock(&nfsv4rootfs_lock, 1);
861 			NFSUNLOCKV4ROOTMUTEX();
862 			error = NFSERR_CLIENTIDBUSY;
863 			goto out;
864 		}
865 	if (!LIST_EMPTY(&clp->lc_session) || !LIST_EMPTY(&clp->lc_deleg)) {
866 		NFSLOCKV4ROOTMUTEX();
867 		nfsv4_unlock(&nfsv4rootfs_lock, 1);
868 		NFSUNLOCKV4ROOTMUTEX();
869 		error = NFSERR_CLIENTIDBUSY;
870 		goto out;
871 	}
872 
873 	/* Destroy the clientid and return ok. */
874 	nfsrv_cleanclient(clp, p);
875 	nfsrv_freedeleglist(&clp->lc_deleg);
876 	nfsrv_freedeleglist(&clp->lc_olddeleg);
877 	LIST_REMOVE(clp, lc_hash);
878 	NFSLOCKV4ROOTMUTEX();
879 	nfsv4_unlock(&nfsv4rootfs_lock, 1);
880 	NFSUNLOCKV4ROOTMUTEX();
881 	nfsrv_zapclient(clp, p);
882 out:
883 	NFSEXITCODE2(error, nd);
884 	return (error);
885 }
886 
887 /*
888  * Called from the new nfssvc syscall to admin revoke a clientid.
889  * Returns 0 for success, error otherwise.
890  */
891 int
892 nfsrv_adminrevoke(struct nfsd_clid *revokep, NFSPROC_T *p)
893 {
894 	struct nfsclient *clp = NULL;
895 	int i, error = 0;
896 	int gotit, igotlock;
897 
898 	/*
899 	 * First, lock out the nfsd so that state won't change while the
900 	 * revocation record is being written to the stable storage restart
901 	 * file.
902 	 */
903 	NFSLOCKV4ROOTMUTEX();
904 	do {
905 		igotlock = nfsv4_lock(&nfsv4rootfs_lock, 1, NULL,
906 		    NFSV4ROOTLOCKMUTEXPTR, NULL);
907 	} while (!igotlock);
908 	NFSUNLOCKV4ROOTMUTEX();
909 
910 	/*
911 	 * Search for a match in the client list.
912 	 */
913 	gotit = i = 0;
914 	while (i < nfsrv_clienthashsize && !gotit) {
915 	    LIST_FOREACH(clp, &nfsclienthash[i], lc_hash) {
916 		if (revokep->nclid_idlen == clp->lc_idlen &&
917 		    !NFSBCMP(revokep->nclid_id, clp->lc_id, clp->lc_idlen)) {
918 			gotit = 1;
919 			break;
920 		}
921 	    }
922 	    i++;
923 	}
924 	if (!gotit) {
925 		NFSLOCKV4ROOTMUTEX();
926 		nfsv4_unlock(&nfsv4rootfs_lock, 0);
927 		NFSUNLOCKV4ROOTMUTEX();
928 		error = EPERM;
929 		goto out;
930 	}
931 
932 	/*
933 	 * Now, write out the revocation record
934 	 */
935 	nfsrv_writestable(clp->lc_id, clp->lc_idlen, NFSNST_REVOKE, p);
936 	nfsrv_backupstable();
937 
938 	/*
939 	 * and clear out the state, marking the clientid revoked.
940 	 */
941 	clp->lc_flags &= ~LCL_CALLBACKSON;
942 	clp->lc_flags |= LCL_ADMINREVOKED;
943 	nfsrv_cleanclient(clp, p);
944 	nfsrv_freedeleglist(&clp->lc_deleg);
945 	nfsrv_freedeleglist(&clp->lc_olddeleg);
946 	NFSLOCKV4ROOTMUTEX();
947 	nfsv4_unlock(&nfsv4rootfs_lock, 0);
948 	NFSUNLOCKV4ROOTMUTEX();
949 
950 out:
951 	NFSEXITCODE(error);
952 	return (error);
953 }
954 
955 /*
956  * Dump out stats for all clients. Called from nfssvc(2), that is used
957  * nfsstatsv1.
958  */
959 void
960 nfsrv_dumpclients(struct nfsd_dumpclients *dumpp, int maxcnt)
961 {
962 	struct nfsclient *clp;
963 	int i = 0, cnt = 0;
964 
965 	/*
966 	 * First, get a reference on the nfsv4rootfs_lock so that an
967 	 * exclusive lock cannot be acquired while dumping the clients.
968 	 */
969 	NFSLOCKV4ROOTMUTEX();
970 	nfsv4_getref(&nfsv4rootfs_lock, NULL, NFSV4ROOTLOCKMUTEXPTR, NULL);
971 	NFSUNLOCKV4ROOTMUTEX();
972 	NFSLOCKSTATE();
973 	/*
974 	 * Rattle through the client lists until done.
975 	 */
976 	while (i < nfsrv_clienthashsize && cnt < maxcnt) {
977 	    clp = LIST_FIRST(&nfsclienthash[i]);
978 	    while (clp != LIST_END(&nfsclienthash[i]) && cnt < maxcnt) {
979 		nfsrv_dumpaclient(clp, &dumpp[cnt]);
980 		cnt++;
981 		clp = LIST_NEXT(clp, lc_hash);
982 	    }
983 	    i++;
984 	}
985 	if (cnt < maxcnt)
986 	    dumpp[cnt].ndcl_clid.nclid_idlen = 0;
987 	NFSUNLOCKSTATE();
988 	NFSLOCKV4ROOTMUTEX();
989 	nfsv4_relref(&nfsv4rootfs_lock);
990 	NFSUNLOCKV4ROOTMUTEX();
991 }
992 
993 /*
994  * Dump stats for a client. Must be called with the NFSSTATELOCK and spl'd.
995  */
996 static void
997 nfsrv_dumpaclient(struct nfsclient *clp, struct nfsd_dumpclients *dumpp)
998 {
999 	struct nfsstate *stp, *openstp, *lckownstp;
1000 	struct nfslock *lop;
1001 	sa_family_t af;
1002 #ifdef INET
1003 	struct sockaddr_in *rin;
1004 #endif
1005 #ifdef INET6
1006 	struct sockaddr_in6 *rin6;
1007 #endif
1008 
1009 	dumpp->ndcl_nopenowners = dumpp->ndcl_nlockowners = 0;
1010 	dumpp->ndcl_nopens = dumpp->ndcl_nlocks = 0;
1011 	dumpp->ndcl_ndelegs = dumpp->ndcl_nolddelegs = 0;
1012 	dumpp->ndcl_flags = clp->lc_flags;
1013 	dumpp->ndcl_clid.nclid_idlen = clp->lc_idlen;
1014 	NFSBCOPY(clp->lc_id, dumpp->ndcl_clid.nclid_id, clp->lc_idlen);
1015 	af = clp->lc_req.nr_nam->sa_family;
1016 	dumpp->ndcl_addrfam = af;
1017 	switch (af) {
1018 #ifdef INET
1019 	case AF_INET:
1020 		rin = (struct sockaddr_in *)clp->lc_req.nr_nam;
1021 		dumpp->ndcl_cbaddr.sin_addr = rin->sin_addr;
1022 		break;
1023 #endif
1024 #ifdef INET6
1025 	case AF_INET6:
1026 		rin6 = (struct sockaddr_in6 *)clp->lc_req.nr_nam;
1027 		dumpp->ndcl_cbaddr.sin6_addr = rin6->sin6_addr;
1028 		break;
1029 #endif
1030 	}
1031 
1032 	/*
1033 	 * Now, scan the state lists and total up the opens and locks.
1034 	 */
1035 	LIST_FOREACH(stp, &clp->lc_open, ls_list) {
1036 	    dumpp->ndcl_nopenowners++;
1037 	    LIST_FOREACH(openstp, &stp->ls_open, ls_list) {
1038 		dumpp->ndcl_nopens++;
1039 		LIST_FOREACH(lckownstp, &openstp->ls_open, ls_list) {
1040 		    dumpp->ndcl_nlockowners++;
1041 		    LIST_FOREACH(lop, &lckownstp->ls_lock, lo_lckowner) {
1042 			dumpp->ndcl_nlocks++;
1043 		    }
1044 		}
1045 	    }
1046 	}
1047 
1048 	/*
1049 	 * and the delegation lists.
1050 	 */
1051 	LIST_FOREACH(stp, &clp->lc_deleg, ls_list) {
1052 	    dumpp->ndcl_ndelegs++;
1053 	}
1054 	LIST_FOREACH(stp, &clp->lc_olddeleg, ls_list) {
1055 	    dumpp->ndcl_nolddelegs++;
1056 	}
1057 }
1058 
1059 /*
1060  * Dump out lock stats for a file.
1061  */
1062 void
1063 nfsrv_dumplocks(vnode_t vp, struct nfsd_dumplocks *ldumpp, int maxcnt,
1064     NFSPROC_T *p)
1065 {
1066 	struct nfsstate *stp;
1067 	struct nfslock *lop;
1068 	int cnt = 0;
1069 	struct nfslockfile *lfp;
1070 	sa_family_t af;
1071 #ifdef INET
1072 	struct sockaddr_in *rin;
1073 #endif
1074 #ifdef INET6
1075 	struct sockaddr_in6 *rin6;
1076 #endif
1077 	int ret;
1078 	fhandle_t nfh;
1079 
1080 	ret = nfsrv_getlockfh(vp, 0, NULL, &nfh, p);
1081 	/*
1082 	 * First, get a reference on the nfsv4rootfs_lock so that an
1083 	 * exclusive lock on it cannot be acquired while dumping the locks.
1084 	 */
1085 	NFSLOCKV4ROOTMUTEX();
1086 	nfsv4_getref(&nfsv4rootfs_lock, NULL, NFSV4ROOTLOCKMUTEXPTR, NULL);
1087 	NFSUNLOCKV4ROOTMUTEX();
1088 	NFSLOCKSTATE();
1089 	if (!ret)
1090 		ret = nfsrv_getlockfile(0, NULL, &lfp, &nfh, 0);
1091 	if (ret) {
1092 		ldumpp[0].ndlck_clid.nclid_idlen = 0;
1093 		NFSUNLOCKSTATE();
1094 		NFSLOCKV4ROOTMUTEX();
1095 		nfsv4_relref(&nfsv4rootfs_lock);
1096 		NFSUNLOCKV4ROOTMUTEX();
1097 		return;
1098 	}
1099 
1100 	/*
1101 	 * For each open share on file, dump it out.
1102 	 */
1103 	stp = LIST_FIRST(&lfp->lf_open);
1104 	while (stp != LIST_END(&lfp->lf_open) && cnt < maxcnt) {
1105 		ldumpp[cnt].ndlck_flags = stp->ls_flags;
1106 		ldumpp[cnt].ndlck_stateid.seqid = stp->ls_stateid.seqid;
1107 		ldumpp[cnt].ndlck_stateid.other[0] = stp->ls_stateid.other[0];
1108 		ldumpp[cnt].ndlck_stateid.other[1] = stp->ls_stateid.other[1];
1109 		ldumpp[cnt].ndlck_stateid.other[2] = stp->ls_stateid.other[2];
1110 		ldumpp[cnt].ndlck_owner.nclid_idlen =
1111 		    stp->ls_openowner->ls_ownerlen;
1112 		NFSBCOPY(stp->ls_openowner->ls_owner,
1113 		    ldumpp[cnt].ndlck_owner.nclid_id,
1114 		    stp->ls_openowner->ls_ownerlen);
1115 		ldumpp[cnt].ndlck_clid.nclid_idlen = stp->ls_clp->lc_idlen;
1116 		NFSBCOPY(stp->ls_clp->lc_id, ldumpp[cnt].ndlck_clid.nclid_id,
1117 		    stp->ls_clp->lc_idlen);
1118 		af = stp->ls_clp->lc_req.nr_nam->sa_family;
1119 		ldumpp[cnt].ndlck_addrfam = af;
1120 		switch (af) {
1121 #ifdef INET
1122 		case AF_INET:
1123 			rin = (struct sockaddr_in *)stp->ls_clp->lc_req.nr_nam;
1124 			ldumpp[cnt].ndlck_cbaddr.sin_addr = rin->sin_addr;
1125 			break;
1126 #endif
1127 #ifdef INET6
1128 		case AF_INET6:
1129 			rin6 = (struct sockaddr_in6 *)
1130 			    stp->ls_clp->lc_req.nr_nam;
1131 			ldumpp[cnt].ndlck_cbaddr.sin6_addr = rin6->sin6_addr;
1132 			break;
1133 #endif
1134 		}
1135 		stp = LIST_NEXT(stp, ls_file);
1136 		cnt++;
1137 	}
1138 
1139 	/*
1140 	 * and all locks.
1141 	 */
1142 	lop = LIST_FIRST(&lfp->lf_lock);
1143 	while (lop != LIST_END(&lfp->lf_lock) && cnt < maxcnt) {
1144 		stp = lop->lo_stp;
1145 		ldumpp[cnt].ndlck_flags = lop->lo_flags;
1146 		ldumpp[cnt].ndlck_first = lop->lo_first;
1147 		ldumpp[cnt].ndlck_end = lop->lo_end;
1148 		ldumpp[cnt].ndlck_stateid.seqid = stp->ls_stateid.seqid;
1149 		ldumpp[cnt].ndlck_stateid.other[0] = stp->ls_stateid.other[0];
1150 		ldumpp[cnt].ndlck_stateid.other[1] = stp->ls_stateid.other[1];
1151 		ldumpp[cnt].ndlck_stateid.other[2] = stp->ls_stateid.other[2];
1152 		ldumpp[cnt].ndlck_owner.nclid_idlen = stp->ls_ownerlen;
1153 		NFSBCOPY(stp->ls_owner, ldumpp[cnt].ndlck_owner.nclid_id,
1154 		    stp->ls_ownerlen);
1155 		ldumpp[cnt].ndlck_clid.nclid_idlen = stp->ls_clp->lc_idlen;
1156 		NFSBCOPY(stp->ls_clp->lc_id, ldumpp[cnt].ndlck_clid.nclid_id,
1157 		    stp->ls_clp->lc_idlen);
1158 		af = stp->ls_clp->lc_req.nr_nam->sa_family;
1159 		ldumpp[cnt].ndlck_addrfam = af;
1160 		switch (af) {
1161 #ifdef INET
1162 		case AF_INET:
1163 			rin = (struct sockaddr_in *)stp->ls_clp->lc_req.nr_nam;
1164 			ldumpp[cnt].ndlck_cbaddr.sin_addr = rin->sin_addr;
1165 			break;
1166 #endif
1167 #ifdef INET6
1168 		case AF_INET6:
1169 			rin6 = (struct sockaddr_in6 *)
1170 			    stp->ls_clp->lc_req.nr_nam;
1171 			ldumpp[cnt].ndlck_cbaddr.sin6_addr = rin6->sin6_addr;
1172 			break;
1173 #endif
1174 		}
1175 		lop = LIST_NEXT(lop, lo_lckfile);
1176 		cnt++;
1177 	}
1178 
1179 	/*
1180 	 * and the delegations.
1181 	 */
1182 	stp = LIST_FIRST(&lfp->lf_deleg);
1183 	while (stp != LIST_END(&lfp->lf_deleg) && cnt < maxcnt) {
1184 		ldumpp[cnt].ndlck_flags = stp->ls_flags;
1185 		ldumpp[cnt].ndlck_stateid.seqid = stp->ls_stateid.seqid;
1186 		ldumpp[cnt].ndlck_stateid.other[0] = stp->ls_stateid.other[0];
1187 		ldumpp[cnt].ndlck_stateid.other[1] = stp->ls_stateid.other[1];
1188 		ldumpp[cnt].ndlck_stateid.other[2] = stp->ls_stateid.other[2];
1189 		ldumpp[cnt].ndlck_owner.nclid_idlen = 0;
1190 		ldumpp[cnt].ndlck_clid.nclid_idlen = stp->ls_clp->lc_idlen;
1191 		NFSBCOPY(stp->ls_clp->lc_id, ldumpp[cnt].ndlck_clid.nclid_id,
1192 		    stp->ls_clp->lc_idlen);
1193 		af = stp->ls_clp->lc_req.nr_nam->sa_family;
1194 		ldumpp[cnt].ndlck_addrfam = af;
1195 		switch (af) {
1196 #ifdef INET
1197 		case AF_INET:
1198 			rin = (struct sockaddr_in *)stp->ls_clp->lc_req.nr_nam;
1199 			ldumpp[cnt].ndlck_cbaddr.sin_addr = rin->sin_addr;
1200 			break;
1201 #endif
1202 #ifdef INET6
1203 		case AF_INET6:
1204 			rin6 = (struct sockaddr_in6 *)
1205 			    stp->ls_clp->lc_req.nr_nam;
1206 			ldumpp[cnt].ndlck_cbaddr.sin6_addr = rin6->sin6_addr;
1207 			break;
1208 #endif
1209 		}
1210 		stp = LIST_NEXT(stp, ls_file);
1211 		cnt++;
1212 	}
1213 
1214 	/*
1215 	 * If list isn't full, mark end of list by setting the client name
1216 	 * to zero length.
1217 	 */
1218 	if (cnt < maxcnt)
1219 		ldumpp[cnt].ndlck_clid.nclid_idlen = 0;
1220 	NFSUNLOCKSTATE();
1221 	NFSLOCKV4ROOTMUTEX();
1222 	nfsv4_relref(&nfsv4rootfs_lock);
1223 	NFSUNLOCKV4ROOTMUTEX();
1224 }
1225 
1226 /*
1227  * Server timer routine. It can scan any linked list, so long
1228  * as it holds the spin/mutex lock and there is no exclusive lock on
1229  * nfsv4rootfs_lock.
1230  * (For OpenBSD, a kthread is ok. For FreeBSD, I think it is ok
1231  *  to do this from a callout, since the spin locks work. For
1232  *  Darwin, I'm not sure what will work correctly yet.)
1233  * Should be called once per second.
1234  */
1235 void
1236 nfsrv_servertimer(void)
1237 {
1238 	struct nfsclient *clp, *nclp;
1239 	struct nfsstate *stp, *nstp;
1240 	int got_ref, i;
1241 
1242 	/*
1243 	 * Make sure nfsboottime is set. This is used by V3 as well
1244 	 * as V4. Note that nfsboottime is not nfsrvboottime, which is
1245 	 * only used by the V4 server for leases.
1246 	 */
1247 	if (nfsboottime.tv_sec == 0)
1248 		NFSSETBOOTTIME(nfsboottime);
1249 
1250 	/*
1251 	 * If server hasn't started yet, just return.
1252 	 */
1253 	NFSLOCKSTATE();
1254 	if (nfsrv_stablefirst.nsf_eograce == 0) {
1255 		NFSUNLOCKSTATE();
1256 		return;
1257 	}
1258 	if (!(nfsrv_stablefirst.nsf_flags & NFSNSF_UPDATEDONE)) {
1259 		if (!(nfsrv_stablefirst.nsf_flags & NFSNSF_GRACEOVER) &&
1260 		    NFSD_MONOSEC > nfsrv_stablefirst.nsf_eograce)
1261 			nfsrv_stablefirst.nsf_flags |=
1262 			    (NFSNSF_GRACEOVER | NFSNSF_NEEDLOCK);
1263 		NFSUNLOCKSTATE();
1264 		return;
1265 	}
1266 
1267 	/*
1268 	 * Try and get a reference count on the nfsv4rootfs_lock so that
1269 	 * no nfsd thread can acquire an exclusive lock on it before this
1270 	 * call is done. If it is already exclusively locked, just return.
1271 	 */
1272 	NFSLOCKV4ROOTMUTEX();
1273 	got_ref = nfsv4_getref_nonblock(&nfsv4rootfs_lock);
1274 	NFSUNLOCKV4ROOTMUTEX();
1275 	if (got_ref == 0) {
1276 		NFSUNLOCKSTATE();
1277 		return;
1278 	}
1279 
1280 	/*
1281 	 * For each client...
1282 	 */
1283 	for (i = 0; i < nfsrv_clienthashsize; i++) {
1284 	    clp = LIST_FIRST(&nfsclienthash[i]);
1285 	    while (clp != LIST_END(&nfsclienthash[i])) {
1286 		nclp = LIST_NEXT(clp, lc_hash);
1287 		if (!(clp->lc_flags & LCL_EXPIREIT)) {
1288 		    if (((clp->lc_expiry + NFSRV_STALELEASE) < NFSD_MONOSEC
1289 			 && ((LIST_EMPTY(&clp->lc_deleg)
1290 			      && LIST_EMPTY(&clp->lc_open)) ||
1291 			     nfsrv_clients > nfsrv_clienthighwater)) ||
1292 			(clp->lc_expiry + NFSRV_MOULDYLEASE) < NFSD_MONOSEC ||
1293 			(clp->lc_expiry < NFSD_MONOSEC &&
1294 			 (nfsrv_openpluslock * 10 / 9) > nfsrv_v4statelimit)) {
1295 			/*
1296 			 * Lease has expired several nfsrv_lease times ago:
1297 			 * PLUS
1298 			 *    - no state is associated with it
1299 			 *    OR
1300 			 *    - above high water mark for number of clients
1301 			 *      (nfsrv_clienthighwater should be large enough
1302 			 *       that this only occurs when clients fail to
1303 			 *       use the same nfs_client_id4.id. Maybe somewhat
1304 			 *       higher that the maximum number of clients that
1305 			 *       will mount this server?)
1306 			 * OR
1307 			 * Lease has expired a very long time ago
1308 			 * OR
1309 			 * Lease has expired PLUS the number of opens + locks
1310 			 * has exceeded 90% of capacity
1311 			 *
1312 			 * --> Mark for expiry. The actual expiry will be done
1313 			 *     by an nfsd sometime soon.
1314 			 */
1315 			clp->lc_flags |= LCL_EXPIREIT;
1316 			nfsrv_stablefirst.nsf_flags |=
1317 			    (NFSNSF_NEEDLOCK | NFSNSF_EXPIREDCLIENT);
1318 		    } else {
1319 			/*
1320 			 * If there are no opens, increment no open tick cnt
1321 			 * If time exceeds NFSNOOPEN, mark it to be thrown away
1322 			 * otherwise, if there is an open, reset no open time
1323 			 * Hopefully, this will avoid excessive re-creation
1324 			 * of open owners and subsequent open confirms.
1325 			 */
1326 			stp = LIST_FIRST(&clp->lc_open);
1327 			while (stp != LIST_END(&clp->lc_open)) {
1328 				nstp = LIST_NEXT(stp, ls_list);
1329 				if (LIST_EMPTY(&stp->ls_open)) {
1330 					stp->ls_noopens++;
1331 					if (stp->ls_noopens > NFSNOOPEN ||
1332 					    (nfsrv_openpluslock * 2) >
1333 					    nfsrv_v4statelimit)
1334 						nfsrv_stablefirst.nsf_flags |=
1335 							NFSNSF_NOOPENS;
1336 				} else {
1337 					stp->ls_noopens = 0;
1338 				}
1339 				stp = nstp;
1340 			}
1341 		    }
1342 		}
1343 		clp = nclp;
1344 	    }
1345 	}
1346 	NFSUNLOCKSTATE();
1347 	NFSLOCKV4ROOTMUTEX();
1348 	nfsv4_relref(&nfsv4rootfs_lock);
1349 	NFSUNLOCKV4ROOTMUTEX();
1350 }
1351 
1352 /*
1353  * The following set of functions free up the various data structures.
1354  */
1355 /*
1356  * Clear out all open/lock state related to this nfsclient.
1357  * Caller must hold an exclusive lock on nfsv4rootfs_lock, so that
1358  * there are no other active nfsd threads.
1359  */
1360 void
1361 nfsrv_cleanclient(struct nfsclient *clp, NFSPROC_T *p)
1362 {
1363 	struct nfsstate *stp, *nstp;
1364 	struct nfsdsession *sep, *nsep;
1365 
1366 	LIST_FOREACH_SAFE(stp, &clp->lc_open, ls_list, nstp)
1367 		nfsrv_freeopenowner(stp, 1, p);
1368 	if ((clp->lc_flags & LCL_ADMINREVOKED) == 0)
1369 		LIST_FOREACH_SAFE(sep, &clp->lc_session, sess_list, nsep)
1370 			(void)nfsrv_freesession(sep, NULL);
1371 }
1372 
1373 /*
1374  * Free a client that has been cleaned. It should also already have been
1375  * removed from the lists.
1376  * (Just to be safe w.r.t. newnfs_disconnect(), call this function when
1377  *  softclock interrupts are enabled.)
1378  */
1379 void
1380 nfsrv_zapclient(struct nfsclient *clp, NFSPROC_T *p)
1381 {
1382 
1383 #ifdef notyet
1384 	if ((clp->lc_flags & (LCL_GSS | LCL_CALLBACKSON)) ==
1385 	     (LCL_GSS | LCL_CALLBACKSON) &&
1386 	    (clp->lc_hand.nfsh_flag & NFSG_COMPLETE) &&
1387 	    clp->lc_handlelen > 0) {
1388 		clp->lc_hand.nfsh_flag &= ~NFSG_COMPLETE;
1389 		clp->lc_hand.nfsh_flag |= NFSG_DESTROYED;
1390 		(void) nfsrv_docallback(clp, NFSV4PROC_CBNULL,
1391 			NULL, 0, NULL, NULL, NULL, 0, p);
1392 	}
1393 #endif
1394 	newnfs_disconnect(NULL, &clp->lc_req);
1395 	free(clp->lc_req.nr_nam, M_SONAME);
1396 	NFSFREEMUTEX(&clp->lc_req.nr_mtx);
1397 	free(clp->lc_stateid, M_NFSDCLIENT);
1398 	free(clp, M_NFSDCLIENT);
1399 	NFSLOCKSTATE();
1400 	nfsstatsv1.srvclients--;
1401 	nfsrv_openpluslock--;
1402 	nfsrv_clients--;
1403 	NFSUNLOCKSTATE();
1404 }
1405 
1406 /*
1407  * Free a list of delegation state structures.
1408  * (This function will also free all nfslockfile structures that no
1409  *  longer have associated state.)
1410  */
1411 void
1412 nfsrv_freedeleglist(struct nfsstatehead *sthp)
1413 {
1414 	struct nfsstate *stp, *nstp;
1415 
1416 	LIST_FOREACH_SAFE(stp, sthp, ls_list, nstp) {
1417 		nfsrv_freedeleg(stp);
1418 	}
1419 	LIST_INIT(sthp);
1420 }
1421 
1422 /*
1423  * Free up a delegation.
1424  */
1425 static void
1426 nfsrv_freedeleg(struct nfsstate *stp)
1427 {
1428 	struct nfslockfile *lfp;
1429 
1430 	LIST_REMOVE(stp, ls_hash);
1431 	LIST_REMOVE(stp, ls_list);
1432 	LIST_REMOVE(stp, ls_file);
1433 	if ((stp->ls_flags & NFSLCK_DELEGWRITE) != 0)
1434 		nfsrv_writedelegcnt--;
1435 	lfp = stp->ls_lfp;
1436 	if (LIST_EMPTY(&lfp->lf_open) &&
1437 	    LIST_EMPTY(&lfp->lf_lock) && LIST_EMPTY(&lfp->lf_deleg) &&
1438 	    LIST_EMPTY(&lfp->lf_locallock) && LIST_EMPTY(&lfp->lf_rollback) &&
1439 	    lfp->lf_usecount == 0 &&
1440 	    nfsv4_testlock(&lfp->lf_locallock_lck) == 0)
1441 		nfsrv_freenfslockfile(lfp);
1442 	free(stp, M_NFSDSTATE);
1443 	nfsstatsv1.srvdelegates--;
1444 	nfsrv_openpluslock--;
1445 	nfsrv_delegatecnt--;
1446 }
1447 
1448 /*
1449  * This function frees an open owner and all associated opens.
1450  */
1451 static void
1452 nfsrv_freeopenowner(struct nfsstate *stp, int cansleep, NFSPROC_T *p)
1453 {
1454 	struct nfsstate *nstp, *tstp;
1455 
1456 	LIST_REMOVE(stp, ls_list);
1457 	/*
1458 	 * Now, free all associated opens.
1459 	 */
1460 	nstp = LIST_FIRST(&stp->ls_open);
1461 	while (nstp != LIST_END(&stp->ls_open)) {
1462 		tstp = nstp;
1463 		nstp = LIST_NEXT(nstp, ls_list);
1464 		(void) nfsrv_freeopen(tstp, NULL, cansleep, p);
1465 	}
1466 	if (stp->ls_op)
1467 		nfsrvd_derefcache(stp->ls_op);
1468 	free(stp, M_NFSDSTATE);
1469 	nfsstatsv1.srvopenowners--;
1470 	nfsrv_openpluslock--;
1471 }
1472 
1473 /*
1474  * This function frees an open (nfsstate open structure) with all associated
1475  * lock_owners and locks. It also frees the nfslockfile structure iff there
1476  * are no other opens on the file.
1477  * Returns 1 if it free'd the nfslockfile, 0 otherwise.
1478  */
1479 static int
1480 nfsrv_freeopen(struct nfsstate *stp, vnode_t vp, int cansleep, NFSPROC_T *p)
1481 {
1482 	struct nfsstate *nstp, *tstp;
1483 	struct nfslockfile *lfp;
1484 	int ret;
1485 
1486 	LIST_REMOVE(stp, ls_hash);
1487 	LIST_REMOVE(stp, ls_list);
1488 	LIST_REMOVE(stp, ls_file);
1489 
1490 	lfp = stp->ls_lfp;
1491 	/*
1492 	 * Now, free all lockowners associated with this open.
1493 	 */
1494 	LIST_FOREACH_SAFE(tstp, &stp->ls_open, ls_list, nstp)
1495 		nfsrv_freelockowner(tstp, vp, cansleep, p);
1496 
1497 	/*
1498 	 * The nfslockfile is freed here if there are no locks
1499 	 * associated with the open.
1500 	 * If there are locks associated with the open, the
1501 	 * nfslockfile structure can be freed via nfsrv_freelockowner().
1502 	 * Acquire the state mutex to avoid races with calls to
1503 	 * nfsrv_getlockfile().
1504 	 */
1505 	if (cansleep != 0)
1506 		NFSLOCKSTATE();
1507 	if (lfp != NULL && LIST_EMPTY(&lfp->lf_open) &&
1508 	    LIST_EMPTY(&lfp->lf_deleg) && LIST_EMPTY(&lfp->lf_lock) &&
1509 	    LIST_EMPTY(&lfp->lf_locallock) && LIST_EMPTY(&lfp->lf_rollback) &&
1510 	    lfp->lf_usecount == 0 &&
1511 	    (cansleep != 0 || nfsv4_testlock(&lfp->lf_locallock_lck) == 0)) {
1512 		nfsrv_freenfslockfile(lfp);
1513 		ret = 1;
1514 	} else
1515 		ret = 0;
1516 	if (cansleep != 0)
1517 		NFSUNLOCKSTATE();
1518 	free(stp, M_NFSDSTATE);
1519 	nfsstatsv1.srvopens--;
1520 	nfsrv_openpluslock--;
1521 	return (ret);
1522 }
1523 
1524 /*
1525  * Frees a lockowner and all associated locks.
1526  */
1527 static void
1528 nfsrv_freelockowner(struct nfsstate *stp, vnode_t vp, int cansleep,
1529     NFSPROC_T *p)
1530 {
1531 
1532 	LIST_REMOVE(stp, ls_hash);
1533 	LIST_REMOVE(stp, ls_list);
1534 	nfsrv_freeallnfslocks(stp, vp, cansleep, p);
1535 	if (stp->ls_op)
1536 		nfsrvd_derefcache(stp->ls_op);
1537 	free(stp, M_NFSDSTATE);
1538 	nfsstatsv1.srvlockowners--;
1539 	nfsrv_openpluslock--;
1540 }
1541 
1542 /*
1543  * Free all the nfs locks on a lockowner.
1544  */
1545 static void
1546 nfsrv_freeallnfslocks(struct nfsstate *stp, vnode_t vp, int cansleep,
1547     NFSPROC_T *p)
1548 {
1549 	struct nfslock *lop, *nlop;
1550 	struct nfsrollback *rlp, *nrlp;
1551 	struct nfslockfile *lfp = NULL;
1552 	int gottvp = 0;
1553 	vnode_t tvp = NULL;
1554 	uint64_t first, end;
1555 
1556 	if (vp != NULL)
1557 		ASSERT_VOP_UNLOCKED(vp, "nfsrv_freeallnfslocks: vnode locked");
1558 	lop = LIST_FIRST(&stp->ls_lock);
1559 	while (lop != LIST_END(&stp->ls_lock)) {
1560 		nlop = LIST_NEXT(lop, lo_lckowner);
1561 		/*
1562 		 * Since all locks should be for the same file, lfp should
1563 		 * not change.
1564 		 */
1565 		if (lfp == NULL)
1566 			lfp = lop->lo_lfp;
1567 		else if (lfp != lop->lo_lfp)
1568 			panic("allnfslocks");
1569 		/*
1570 		 * If vp is NULL and cansleep != 0, a vnode must be acquired
1571 		 * from the file handle. This only occurs when called from
1572 		 * nfsrv_cleanclient().
1573 		 */
1574 		if (gottvp == 0) {
1575 			if (nfsrv_dolocallocks == 0)
1576 				tvp = NULL;
1577 			else if (vp == NULL && cansleep != 0) {
1578 				tvp = nfsvno_getvp(&lfp->lf_fh);
1579 				if (tvp != NULL)
1580 					NFSVOPUNLOCK(tvp);
1581 			} else
1582 				tvp = vp;
1583 			gottvp = 1;
1584 		}
1585 
1586 		if (tvp != NULL) {
1587 			if (cansleep == 0)
1588 				panic("allnfs2");
1589 			first = lop->lo_first;
1590 			end = lop->lo_end;
1591 			nfsrv_freenfslock(lop);
1592 			nfsrv_localunlock(tvp, lfp, first, end, p);
1593 			LIST_FOREACH_SAFE(rlp, &lfp->lf_rollback, rlck_list,
1594 			    nrlp)
1595 				free(rlp, M_NFSDROLLBACK);
1596 			LIST_INIT(&lfp->lf_rollback);
1597 		} else
1598 			nfsrv_freenfslock(lop);
1599 		lop = nlop;
1600 	}
1601 	if (vp == NULL && tvp != NULL)
1602 		vrele(tvp);
1603 }
1604 
1605 /*
1606  * Free an nfslock structure.
1607  */
1608 static void
1609 nfsrv_freenfslock(struct nfslock *lop)
1610 {
1611 
1612 	if (lop->lo_lckfile.le_prev != NULL) {
1613 		LIST_REMOVE(lop, lo_lckfile);
1614 		nfsstatsv1.srvlocks--;
1615 		nfsrv_openpluslock--;
1616 	}
1617 	LIST_REMOVE(lop, lo_lckowner);
1618 	free(lop, M_NFSDLOCK);
1619 }
1620 
1621 /*
1622  * This function frees an nfslockfile structure.
1623  */
1624 static void
1625 nfsrv_freenfslockfile(struct nfslockfile *lfp)
1626 {
1627 
1628 	LIST_REMOVE(lfp, lf_hash);
1629 	free(lfp, M_NFSDLOCKFILE);
1630 }
1631 
1632 /*
1633  * This function looks up an nfsstate structure via stateid.
1634  */
1635 static int
1636 nfsrv_getstate(struct nfsclient *clp, nfsv4stateid_t *stateidp, __unused u_int32_t flags,
1637     struct nfsstate **stpp)
1638 {
1639 	struct nfsstate *stp;
1640 	struct nfsstatehead *hp;
1641 	int error = 0;
1642 
1643 	*stpp = NULL;
1644 	hp = NFSSTATEHASH(clp, *stateidp);
1645 	LIST_FOREACH(stp, hp, ls_hash) {
1646 		if (!NFSBCMP(stp->ls_stateid.other, stateidp->other,
1647 			NFSX_STATEIDOTHER))
1648 			break;
1649 	}
1650 
1651 	/*
1652 	 * If no state id in list, return NFSERR_BADSTATEID.
1653 	 */
1654 	if (stp == LIST_END(hp)) {
1655 		error = NFSERR_BADSTATEID;
1656 		goto out;
1657 	}
1658 	*stpp = stp;
1659 
1660 out:
1661 	NFSEXITCODE(error);
1662 	return (error);
1663 }
1664 
1665 /*
1666  * This function gets an nfsstate structure via owner string.
1667  */
1668 static void
1669 nfsrv_getowner(struct nfsstatehead *hp, struct nfsstate *new_stp,
1670     struct nfsstate **stpp)
1671 {
1672 	struct nfsstate *stp;
1673 
1674 	*stpp = NULL;
1675 	LIST_FOREACH(stp, hp, ls_list) {
1676 		if (new_stp->ls_ownerlen == stp->ls_ownerlen &&
1677 		  !NFSBCMP(new_stp->ls_owner,stp->ls_owner,stp->ls_ownerlen)) {
1678 			*stpp = stp;
1679 			return;
1680 		}
1681 	}
1682 }
1683 
1684 /*
1685  * Lock control function called to update lock status.
1686  * Returns 0 upon success, -1 if there is no lock and the flags indicate
1687  * that one isn't to be created and an NFSERR_xxx for other errors.
1688  * The structures new_stp and new_lop are passed in as pointers that should
1689  * be set to NULL if the structure is used and shouldn't be free'd.
1690  * For the NFSLCK_TEST and NFSLCK_CHECK cases, the structures are
1691  * never used and can safely be allocated on the stack. For all other
1692  * cases, *new_stpp and *new_lopp should be malloc'd before the call,
1693  * in case they are used.
1694  */
1695 int
1696 nfsrv_lockctrl(vnode_t vp, struct nfsstate **new_stpp,
1697     struct nfslock **new_lopp, struct nfslockconflict *cfp,
1698     nfsquad_t clientid, nfsv4stateid_t *stateidp,
1699     __unused struct nfsexstuff *exp,
1700     struct nfsrv_descript *nd, NFSPROC_T *p)
1701 {
1702 	struct nfslock *lop;
1703 	struct nfsstate *new_stp = *new_stpp;
1704 	struct nfslock *new_lop = *new_lopp;
1705 	struct nfsstate *tstp, *mystp, *nstp;
1706 	int specialid = 0;
1707 	struct nfslockfile *lfp;
1708 	struct nfslock *other_lop = NULL;
1709 	struct nfsstate *stp, *lckstp = NULL;
1710 	struct nfsclient *clp = NULL;
1711 	u_int32_t bits;
1712 	int error = 0, haslock = 0, ret, reterr;
1713 	int getlckret, delegation = 0, filestruct_locked, vnode_unlocked = 0;
1714 	fhandle_t nfh;
1715 	uint64_t first, end;
1716 	uint32_t lock_flags;
1717 
1718 	if (new_stp->ls_flags & (NFSLCK_CHECK | NFSLCK_SETATTR)) {
1719 		/*
1720 		 * Note the special cases of "all 1s" or "all 0s" stateids and
1721 		 * let reads with all 1s go ahead.
1722 		 */
1723 		if (new_stp->ls_stateid.seqid == 0x0 &&
1724 		    new_stp->ls_stateid.other[0] == 0x0 &&
1725 		    new_stp->ls_stateid.other[1] == 0x0 &&
1726 		    new_stp->ls_stateid.other[2] == 0x0)
1727 			specialid = 1;
1728 		else if (new_stp->ls_stateid.seqid == 0xffffffff &&
1729 		    new_stp->ls_stateid.other[0] == 0xffffffff &&
1730 		    new_stp->ls_stateid.other[1] == 0xffffffff &&
1731 		    new_stp->ls_stateid.other[2] == 0xffffffff)
1732 			specialid = 2;
1733 	}
1734 
1735 	/*
1736 	 * Check for restart conditions (client and server).
1737 	 */
1738 	error = nfsrv_checkrestart(clientid, new_stp->ls_flags,
1739 	    &new_stp->ls_stateid, specialid);
1740 	if (error)
1741 		goto out;
1742 
1743 	/*
1744 	 * Check for state resource limit exceeded.
1745 	 */
1746 	if ((new_stp->ls_flags & NFSLCK_LOCK) &&
1747 	    nfsrv_openpluslock > nfsrv_v4statelimit) {
1748 		error = NFSERR_RESOURCE;
1749 		goto out;
1750 	}
1751 
1752 	/*
1753 	 * For the lock case, get another nfslock structure,
1754 	 * just in case we need it.
1755 	 * Malloc now, before we start sifting through the linked lists,
1756 	 * in case we have to wait for memory.
1757 	 */
1758 tryagain:
1759 	if (new_stp->ls_flags & NFSLCK_LOCK)
1760 		other_lop = malloc(sizeof (struct nfslock),
1761 		    M_NFSDLOCK, M_WAITOK);
1762 	filestruct_locked = 0;
1763 	reterr = 0;
1764 	lfp = NULL;
1765 
1766 	/*
1767 	 * Get the lockfile structure for CFH now, so we can do a sanity
1768 	 * check against the stateid, before incrementing the seqid#, since
1769 	 * we want to return NFSERR_BADSTATEID on failure and the seqid#
1770 	 * shouldn't be incremented for this case.
1771 	 * If nfsrv_getlockfile() returns -1, it means "not found", which
1772 	 * will be handled later.
1773 	 * If we are doing Lock/LockU and local locking is enabled, sleep
1774 	 * lock the nfslockfile structure.
1775 	 */
1776 	getlckret = nfsrv_getlockfh(vp, new_stp->ls_flags, NULL, &nfh, p);
1777 	NFSLOCKSTATE();
1778 	if (getlckret == 0) {
1779 		if ((new_stp->ls_flags & (NFSLCK_LOCK | NFSLCK_UNLOCK)) != 0 &&
1780 		    nfsrv_dolocallocks != 0 && nd->nd_repstat == 0) {
1781 			getlckret = nfsrv_getlockfile(new_stp->ls_flags, NULL,
1782 			    &lfp, &nfh, 1);
1783 			if (getlckret == 0)
1784 				filestruct_locked = 1;
1785 		} else
1786 			getlckret = nfsrv_getlockfile(new_stp->ls_flags, NULL,
1787 			    &lfp, &nfh, 0);
1788 	}
1789 	if (getlckret != 0 && getlckret != -1)
1790 		reterr = getlckret;
1791 
1792 	if (filestruct_locked != 0) {
1793 		LIST_INIT(&lfp->lf_rollback);
1794 		if ((new_stp->ls_flags & NFSLCK_LOCK)) {
1795 			/*
1796 			 * For local locking, do the advisory locking now, so
1797 			 * that any conflict can be detected. A failure later
1798 			 * can be rolled back locally. If an error is returned,
1799 			 * struct nfslockfile has been unlocked and any local
1800 			 * locking rolled back.
1801 			 */
1802 			NFSUNLOCKSTATE();
1803 			if (vnode_unlocked == 0) {
1804 				ASSERT_VOP_ELOCKED(vp, "nfsrv_lockctrl1");
1805 				vnode_unlocked = 1;
1806 				NFSVOPUNLOCK(vp);
1807 			}
1808 			reterr = nfsrv_locallock(vp, lfp,
1809 			    (new_lop->lo_flags & (NFSLCK_READ | NFSLCK_WRITE)),
1810 			    new_lop->lo_first, new_lop->lo_end, cfp, p);
1811 			NFSLOCKSTATE();
1812 		}
1813 	}
1814 
1815 	if (specialid == 0) {
1816 	    if (new_stp->ls_flags & NFSLCK_TEST) {
1817 		/*
1818 		 * RFC 3530 does not list LockT as an op that renews a
1819 		 * lease, but the consensus seems to be that it is ok
1820 		 * for a server to do so.
1821 		 */
1822 		error = nfsrv_getclient(clientid, CLOPS_RENEW, &clp, NULL,
1823 		    (nfsquad_t)((u_quad_t)0), 0, nd, p);
1824 
1825 		/*
1826 		 * Since NFSERR_EXPIRED, NFSERR_ADMINREVOKED are not valid
1827 		 * error returns for LockT, just go ahead and test for a lock,
1828 		 * since there are no locks for this client, but other locks
1829 		 * can conflict. (ie. same client will always be false)
1830 		 */
1831 		if (error == NFSERR_EXPIRED || error == NFSERR_ADMINREVOKED)
1832 		    error = 0;
1833 		lckstp = new_stp;
1834 	    } else {
1835 	      error = nfsrv_getclient(clientid, CLOPS_RENEW, &clp, NULL,
1836 		(nfsquad_t)((u_quad_t)0), 0, nd, p);
1837 	      if (error == 0)
1838 		/*
1839 		 * Look up the stateid
1840 		 */
1841 		error = nfsrv_getstate(clp, &new_stp->ls_stateid,
1842 		  new_stp->ls_flags, &stp);
1843 	      /*
1844 	       * do some sanity checks for an unconfirmed open or a
1845 	       * stateid that refers to the wrong file, for an open stateid
1846 	       */
1847 	      if (error == 0 && (stp->ls_flags & NFSLCK_OPEN) &&
1848 		  ((stp->ls_openowner->ls_flags & NFSLCK_NEEDSCONFIRM) ||
1849 		   (getlckret == 0 && stp->ls_lfp != lfp))){
1850 		      /*
1851 		       * NFSLCK_SETATTR should return OK rather than NFSERR_BADSTATEID
1852 		       * The only exception is using SETATTR with SIZE.
1853 		       * */
1854                     if ((new_stp->ls_flags &
1855                          (NFSLCK_SETATTR | NFSLCK_CHECK)) != NFSLCK_SETATTR)
1856 			     error = NFSERR_BADSTATEID;
1857 	      }
1858 
1859 		if (error == 0 &&
1860 		  (stp->ls_flags & (NFSLCK_DELEGREAD | NFSLCK_DELEGWRITE)) &&
1861 		  getlckret == 0 && stp->ls_lfp != lfp)
1862 			error = NFSERR_BADSTATEID;
1863 
1864 	      /*
1865 	       * If the lockowner stateid doesn't refer to the same file,
1866 	       * I believe that is considered ok, since some clients will
1867 	       * only create a single lockowner and use that for all locks
1868 	       * on all files.
1869 	       * For now, log it as a diagnostic, instead of considering it
1870 	       * a BadStateid.
1871 	       */
1872 	      if (error == 0 && (stp->ls_flags &
1873 		  (NFSLCK_OPEN | NFSLCK_DELEGREAD | NFSLCK_DELEGWRITE)) == 0 &&
1874 		  getlckret == 0 && stp->ls_lfp != lfp) {
1875 #ifdef DIAGNOSTIC
1876 		  printf("Got a lock statid for different file open\n");
1877 #endif
1878 		  /*
1879 		  error = NFSERR_BADSTATEID;
1880 		  */
1881 	      }
1882 
1883 	      if (error == 0) {
1884 		    if (new_stp->ls_flags & NFSLCK_OPENTOLOCK) {
1885 			/*
1886 			 * If haslock set, we've already checked the seqid.
1887 			 */
1888 			if (!haslock) {
1889 			    if (stp->ls_flags & NFSLCK_OPEN)
1890 				error = nfsrv_checkseqid(nd, new_stp->ls_seq,
1891 				    stp->ls_openowner, new_stp->ls_op);
1892 			    else
1893 				error = NFSERR_BADSTATEID;
1894 			}
1895 			if (!error)
1896 			    nfsrv_getowner(&stp->ls_open, new_stp, &lckstp);
1897 			if (lckstp) {
1898 			    /*
1899 			     * For NFSv4.1 and NFSv4.2 allow an
1900 			     * open_to_lock_owner when the lock_owner already
1901 			     * exists.  Just clear NFSLCK_OPENTOLOCK so that
1902 			     * a new lock_owner will not be created.
1903 			     * RFC7530 states that the error for NFSv4.0
1904 			     * is NFS4ERR_BAD_SEQID.
1905 			     */
1906 			    if ((nd->nd_flag & ND_NFSV41) != 0)
1907 				new_stp->ls_flags &= ~NFSLCK_OPENTOLOCK;
1908 			    else
1909 				error = NFSERR_BADSEQID;
1910 			} else
1911 			    lckstp = new_stp;
1912 		    } else if (new_stp->ls_flags&(NFSLCK_LOCK|NFSLCK_UNLOCK)) {
1913 			/*
1914 			 * If haslock set, ditto above.
1915 			 */
1916 			if (!haslock) {
1917 			    if (stp->ls_flags & NFSLCK_OPEN)
1918 				error = NFSERR_BADSTATEID;
1919 			    else
1920 				error = nfsrv_checkseqid(nd, new_stp->ls_seq,
1921 				    stp, new_stp->ls_op);
1922 			}
1923 			lckstp = stp;
1924 		    } else {
1925 			lckstp = stp;
1926 		    }
1927 	      }
1928 	      /*
1929 	       * If the seqid part of the stateid isn't the same, return
1930 	       * NFSERR_OLDSTATEID for cases other than I/O Ops.
1931 	       * For I/O Ops, only return NFSERR_OLDSTATEID if
1932 	       * nfsrv_returnoldstateid is set. (The consensus on the email
1933 	       * list was that most clients would prefer to not receive
1934 	       * NFSERR_OLDSTATEID for I/O Ops, but the RFC suggests that that
1935 	       * is what will happen, so I use the nfsrv_returnoldstateid to
1936 	       * allow for either server configuration.)
1937 	       */
1938 	      if (!error && stp->ls_stateid.seqid!=new_stp->ls_stateid.seqid &&
1939 		  (((nd->nd_flag & ND_NFSV41) == 0 &&
1940 		   (!(new_stp->ls_flags & NFSLCK_CHECK) ||
1941 		    nfsrv_returnoldstateid)) ||
1942 		   ((nd->nd_flag & ND_NFSV41) != 0 &&
1943 		    new_stp->ls_stateid.seqid != 0)))
1944 		    error = NFSERR_OLDSTATEID;
1945 	    }
1946 	}
1947 
1948 	/*
1949 	 * Now we can check for grace.
1950 	 */
1951 	if (!error)
1952 		error = nfsrv_checkgrace(nd, clp, new_stp->ls_flags);
1953 	if ((new_stp->ls_flags & NFSLCK_RECLAIM) && !error &&
1954 		nfsrv_checkstable(clp))
1955 		error = NFSERR_NOGRACE;
1956 	/*
1957 	 * If we successfully Reclaimed state, note that.
1958 	 */
1959 	if ((new_stp->ls_flags & NFSLCK_RECLAIM) && !error)
1960 		nfsrv_markstable(clp);
1961 
1962 	/*
1963 	 * At this point, either error == NFSERR_BADSTATEID or the
1964 	 * seqid# has been updated, so we can return any error.
1965 	 * If error == 0, there may be an error in:
1966 	 *    nd_repstat - Set by the calling function.
1967 	 *    reterr - Set above, if getting the nfslockfile structure
1968 	 *       or acquiring the local lock failed.
1969 	 *    (If both of these are set, nd_repstat should probably be
1970 	 *     returned, since that error was detected before this
1971 	 *     function call.)
1972 	 */
1973 	if (error != 0 || nd->nd_repstat != 0 || reterr != 0) {
1974 		if (error == 0) {
1975 			if (nd->nd_repstat != 0)
1976 				error = nd->nd_repstat;
1977 			else
1978 				error = reterr;
1979 		}
1980 		if (filestruct_locked != 0) {
1981 			/* Roll back local locks. */
1982 			NFSUNLOCKSTATE();
1983 			if (vnode_unlocked == 0) {
1984 				ASSERT_VOP_ELOCKED(vp, "nfsrv_lockctrl2");
1985 				vnode_unlocked = 1;
1986 				NFSVOPUNLOCK(vp);
1987 			}
1988 			nfsrv_locallock_rollback(vp, lfp, p);
1989 			NFSLOCKSTATE();
1990 			nfsrv_unlocklf(lfp);
1991 		}
1992 		NFSUNLOCKSTATE();
1993 		goto out;
1994 	}
1995 
1996 	/*
1997 	 * Check the nfsrv_getlockfile return.
1998 	 * Returned -1 if no structure found.
1999 	 */
2000 	if (getlckret == -1) {
2001 		error = NFSERR_EXPIRED;
2002 		/*
2003 		 * Called from lockt, so no lock is OK.
2004 		 */
2005 		if (new_stp->ls_flags & NFSLCK_TEST) {
2006 			error = 0;
2007 		} else if (new_stp->ls_flags &
2008 		    (NFSLCK_CHECK | NFSLCK_SETATTR)) {
2009 			/*
2010 			 * Called to check for a lock, OK if the stateid is all
2011 			 * 1s or all 0s, but there should be an nfsstate
2012 			 * otherwise.
2013 			 * (ie. If there is no open, I'll assume no share
2014 			 *  deny bits.)
2015 			 */
2016 			if (specialid)
2017 				error = 0;
2018 			else
2019 				error = NFSERR_BADSTATEID;
2020 		}
2021 		NFSUNLOCKSTATE();
2022 		goto out;
2023 	}
2024 
2025 	/*
2026 	 * For NFSLCK_CHECK and NFSLCK_LOCK, test for a share conflict.
2027 	 * For NFSLCK_CHECK, allow a read if write access is granted,
2028 	 * but check for a deny. For NFSLCK_LOCK, require correct access,
2029 	 * which implies a conflicting deny can't exist.
2030 	 */
2031 	if (new_stp->ls_flags & (NFSLCK_CHECK | NFSLCK_LOCK)) {
2032 	    /*
2033 	     * Four kinds of state id:
2034 	     * - specialid (all 0s or all 1s), only for NFSLCK_CHECK
2035 	     * - stateid for an open
2036 	     * - stateid for a delegation
2037 	     * - stateid for a lock owner
2038 	     */
2039 	    if (!specialid) {
2040 		if (stp->ls_flags & (NFSLCK_DELEGREAD | NFSLCK_DELEGWRITE)) {
2041 		    delegation = 1;
2042 		    mystp = stp;
2043 		    nfsrv_delaydelegtimeout(stp);
2044 	        } else if (stp->ls_flags & NFSLCK_OPEN) {
2045 		    mystp = stp;
2046 		} else {
2047 		    mystp = stp->ls_openstp;
2048 		}
2049 		/*
2050 		 * If locking or checking, require correct access
2051 		 * bit set.
2052 		 */
2053 		if (((new_stp->ls_flags & NFSLCK_LOCK) &&
2054 		     !((new_lop->lo_flags >> NFSLCK_LOCKSHIFT) &
2055 		       mystp->ls_flags & NFSLCK_ACCESSBITS)) ||
2056 		    ((new_stp->ls_flags & (NFSLCK_CHECK|NFSLCK_READACCESS)) ==
2057 		      (NFSLCK_CHECK | NFSLCK_READACCESS) &&
2058 		     !(mystp->ls_flags & NFSLCK_READACCESS) &&
2059 		     nfsrv_allowreadforwriteopen == 0) ||
2060 		    ((new_stp->ls_flags & (NFSLCK_CHECK|NFSLCK_WRITEACCESS)) ==
2061 		      (NFSLCK_CHECK | NFSLCK_WRITEACCESS) &&
2062 		     !(mystp->ls_flags & NFSLCK_WRITEACCESS))) {
2063 			if (filestruct_locked != 0) {
2064 				/* Roll back local locks. */
2065 				NFSUNLOCKSTATE();
2066 				if (vnode_unlocked == 0) {
2067 					ASSERT_VOP_ELOCKED(vp,
2068 					    "nfsrv_lockctrl3");
2069 					vnode_unlocked = 1;
2070 					NFSVOPUNLOCK(vp);
2071 				}
2072 				nfsrv_locallock_rollback(vp, lfp, p);
2073 				NFSLOCKSTATE();
2074 				nfsrv_unlocklf(lfp);
2075 			}
2076 			NFSUNLOCKSTATE();
2077 			error = NFSERR_OPENMODE;
2078 			goto out;
2079 		}
2080 	    } else
2081 		mystp = NULL;
2082 	    if ((new_stp->ls_flags & NFSLCK_CHECK) && !delegation) {
2083 		/*
2084 		 * Check for a conflicting deny bit.
2085 		 */
2086 		LIST_FOREACH(tstp, &lfp->lf_open, ls_file) {
2087 		    if (tstp != mystp) {
2088 			bits = tstp->ls_flags;
2089 			bits >>= NFSLCK_SHIFT;
2090 			if (new_stp->ls_flags & bits & NFSLCK_ACCESSBITS) {
2091 			    KASSERT(vnode_unlocked == 0,
2092 				("nfsrv_lockctrl: vnode unlocked1"));
2093 			    ret = nfsrv_clientconflict(tstp->ls_clp, &haslock,
2094 				vp, p);
2095 			    if (ret == 1) {
2096 				/*
2097 				* nfsrv_clientconflict unlocks state
2098 				 * when it returns non-zero.
2099 				 */
2100 				lckstp = NULL;
2101 				goto tryagain;
2102 			    }
2103 			    if (ret == 0)
2104 				NFSUNLOCKSTATE();
2105 			    if (ret == 2)
2106 				error = NFSERR_PERM;
2107 			    else
2108 				error = NFSERR_OPENMODE;
2109 			    goto out;
2110 			}
2111 		    }
2112 		}
2113 
2114 		/* We're outta here */
2115 		NFSUNLOCKSTATE();
2116 		goto out;
2117 	    }
2118 	}
2119 
2120 	/*
2121 	 * For setattr, just get rid of all the Delegations for other clients.
2122 	 */
2123 	if (new_stp->ls_flags & NFSLCK_SETATTR) {
2124 		KASSERT(vnode_unlocked == 0,
2125 		    ("nfsrv_lockctrl: vnode unlocked2"));
2126 		ret = nfsrv_cleandeleg(vp, lfp, clp, &haslock, p);
2127 		if (ret) {
2128 			/*
2129 			 * nfsrv_cleandeleg() unlocks state when it
2130 			 * returns non-zero.
2131 			 */
2132 			if (ret == -1) {
2133 				lckstp = NULL;
2134 				goto tryagain;
2135 			}
2136 			error = ret;
2137 			goto out;
2138 		}
2139 		if (!(new_stp->ls_flags & NFSLCK_CHECK) ||
2140 		    (LIST_EMPTY(&lfp->lf_open) && LIST_EMPTY(&lfp->lf_lock) &&
2141 		     LIST_EMPTY(&lfp->lf_deleg))) {
2142 			NFSUNLOCKSTATE();
2143 			goto out;
2144 		}
2145 	}
2146 
2147 	/*
2148 	 * Check for a conflicting delegation. If one is found, call
2149 	 * nfsrv_delegconflict() to handle it. If the v4root lock hasn't
2150 	 * been set yet, it will get the lock. Otherwise, it will recall
2151 	 * the delegation. Then, we try try again...
2152 	 * I currently believe the conflict algorithm to be:
2153 	 * For Lock Ops (Lock/LockT/LockU)
2154 	 * - there is a conflict iff a different client has a write delegation
2155 	 * For Reading (Read Op)
2156 	 * - there is a conflict iff a different client has a write delegation
2157 	 *   (the specialids are always a different client)
2158 	 * For Writing (Write/Setattr of size)
2159 	 * - there is a conflict if a different client has any delegation
2160 	 * - there is a conflict if the same client has a read delegation
2161 	 *   (I don't understand why this isn't allowed, but that seems to be
2162 	 *    the current consensus?)
2163 	 */
2164 	tstp = LIST_FIRST(&lfp->lf_deleg);
2165 	while (tstp != LIST_END(&lfp->lf_deleg)) {
2166 	    nstp = LIST_NEXT(tstp, ls_file);
2167 	    if ((((new_stp->ls_flags&(NFSLCK_LOCK|NFSLCK_UNLOCK|NFSLCK_TEST))||
2168 		 ((new_stp->ls_flags & NFSLCK_CHECK) &&
2169 		  (new_lop->lo_flags & NFSLCK_READ))) &&
2170 		  clp != tstp->ls_clp &&
2171 		 (tstp->ls_flags & NFSLCK_DELEGWRITE)) ||
2172 		 ((new_stp->ls_flags & NFSLCK_CHECK) &&
2173 		   (new_lop->lo_flags & NFSLCK_WRITE) &&
2174 		  (clp != tstp->ls_clp ||
2175 		   (tstp->ls_flags & NFSLCK_DELEGREAD)))) {
2176 		ret = 0;
2177 		if (filestruct_locked != 0) {
2178 			/* Roll back local locks. */
2179 			NFSUNLOCKSTATE();
2180 			if (vnode_unlocked == 0) {
2181 				ASSERT_VOP_ELOCKED(vp, "nfsrv_lockctrl4");
2182 				NFSVOPUNLOCK(vp);
2183 			}
2184 			nfsrv_locallock_rollback(vp, lfp, p);
2185 			NFSLOCKSTATE();
2186 			nfsrv_unlocklf(lfp);
2187 			NFSUNLOCKSTATE();
2188 			NFSVOPLOCK(vp, LK_EXCLUSIVE | LK_RETRY);
2189 			vnode_unlocked = 0;
2190 			if (VN_IS_DOOMED(vp))
2191 				ret = NFSERR_SERVERFAULT;
2192 			NFSLOCKSTATE();
2193 		}
2194 		if (ret == 0)
2195 			ret = nfsrv_delegconflict(tstp, &haslock, p, vp);
2196 		if (ret) {
2197 		    /*
2198 		     * nfsrv_delegconflict unlocks state when it
2199 		     * returns non-zero, which it always does.
2200 		     */
2201 		    if (other_lop) {
2202 			free(other_lop, M_NFSDLOCK);
2203 			other_lop = NULL;
2204 		    }
2205 		    if (ret == -1) {
2206 			lckstp = NULL;
2207 			goto tryagain;
2208 		    }
2209 		    error = ret;
2210 		    goto out;
2211 		}
2212 		/* Never gets here. */
2213 	    }
2214 	    tstp = nstp;
2215 	}
2216 
2217 	/*
2218 	 * Handle the unlock case by calling nfsrv_updatelock().
2219 	 * (Should I have done some access checking above for unlock? For now,
2220 	 *  just let it happen.)
2221 	 */
2222 	if (new_stp->ls_flags & NFSLCK_UNLOCK) {
2223 		first = new_lop->lo_first;
2224 		end = new_lop->lo_end;
2225 		nfsrv_updatelock(stp, new_lopp, &other_lop, lfp);
2226 		stateidp->seqid = ++(stp->ls_stateid.seqid);
2227 		if ((nd->nd_flag & ND_NFSV41) != 0 && stateidp->seqid == 0)
2228 			stateidp->seqid = stp->ls_stateid.seqid = 1;
2229 		stateidp->other[0] = stp->ls_stateid.other[0];
2230 		stateidp->other[1] = stp->ls_stateid.other[1];
2231 		stateidp->other[2] = stp->ls_stateid.other[2];
2232 		if (filestruct_locked != 0) {
2233 			NFSUNLOCKSTATE();
2234 			if (vnode_unlocked == 0) {
2235 				ASSERT_VOP_ELOCKED(vp, "nfsrv_lockctrl5");
2236 				vnode_unlocked = 1;
2237 				NFSVOPUNLOCK(vp);
2238 			}
2239 			/* Update the local locks. */
2240 			nfsrv_localunlock(vp, lfp, first, end, p);
2241 			NFSLOCKSTATE();
2242 			nfsrv_unlocklf(lfp);
2243 		}
2244 		NFSUNLOCKSTATE();
2245 		goto out;
2246 	}
2247 
2248 	/*
2249 	 * Search for a conflicting lock. A lock conflicts if:
2250 	 * - the lock range overlaps and
2251 	 * - at least one lock is a write lock and
2252 	 * - it is not owned by the same lock owner
2253 	 */
2254 	if (!delegation) {
2255 	  LIST_FOREACH(lop, &lfp->lf_lock, lo_lckfile) {
2256 	    if (new_lop->lo_end > lop->lo_first &&
2257 		new_lop->lo_first < lop->lo_end &&
2258 		(new_lop->lo_flags == NFSLCK_WRITE ||
2259 		 lop->lo_flags == NFSLCK_WRITE) &&
2260 		lckstp != lop->lo_stp &&
2261 		(clp != lop->lo_stp->ls_clp ||
2262 		 lckstp->ls_ownerlen != lop->lo_stp->ls_ownerlen ||
2263 		 NFSBCMP(lckstp->ls_owner, lop->lo_stp->ls_owner,
2264 		    lckstp->ls_ownerlen))) {
2265 		if (other_lop) {
2266 		    free(other_lop, M_NFSDLOCK);
2267 		    other_lop = NULL;
2268 		}
2269 		if (vnode_unlocked != 0)
2270 		    ret = nfsrv_clientconflict(lop->lo_stp->ls_clp, &haslock,
2271 			NULL, p);
2272 		else
2273 		    ret = nfsrv_clientconflict(lop->lo_stp->ls_clp, &haslock,
2274 			vp, p);
2275 		if (ret == 1) {
2276 		    if (filestruct_locked != 0) {
2277 			if (vnode_unlocked == 0) {
2278 				ASSERT_VOP_ELOCKED(vp, "nfsrv_lockctrl6");
2279 				NFSVOPUNLOCK(vp);
2280 			}
2281 			/* Roll back local locks. */
2282 			nfsrv_locallock_rollback(vp, lfp, p);
2283 			NFSLOCKSTATE();
2284 			nfsrv_unlocklf(lfp);
2285 			NFSUNLOCKSTATE();
2286 			NFSVOPLOCK(vp, LK_EXCLUSIVE | LK_RETRY);
2287 			vnode_unlocked = 0;
2288 			if (VN_IS_DOOMED(vp)) {
2289 				error = NFSERR_SERVERFAULT;
2290 				goto out;
2291 			}
2292 		    }
2293 		    /*
2294 		     * nfsrv_clientconflict() unlocks state when it
2295 		     * returns non-zero.
2296 		     */
2297 		    lckstp = NULL;
2298 		    goto tryagain;
2299 		}
2300 		/*
2301 		 * Found a conflicting lock, so record the conflict and
2302 		 * return the error.
2303 		 */
2304 		if (cfp != NULL && ret == 0) {
2305 		    cfp->cl_clientid.lval[0]=lop->lo_stp->ls_stateid.other[0];
2306 		    cfp->cl_clientid.lval[1]=lop->lo_stp->ls_stateid.other[1];
2307 		    cfp->cl_first = lop->lo_first;
2308 		    cfp->cl_end = lop->lo_end;
2309 		    cfp->cl_flags = lop->lo_flags;
2310 		    cfp->cl_ownerlen = lop->lo_stp->ls_ownerlen;
2311 		    NFSBCOPY(lop->lo_stp->ls_owner, cfp->cl_owner,
2312 			cfp->cl_ownerlen);
2313 		}
2314 		if (ret == 2)
2315 		    error = NFSERR_PERM;
2316 		else if (new_stp->ls_flags & NFSLCK_RECLAIM)
2317 		    error = NFSERR_RECLAIMCONFLICT;
2318 		else if (new_stp->ls_flags & NFSLCK_CHECK)
2319 		    error = NFSERR_LOCKED;
2320 		else
2321 		    error = NFSERR_DENIED;
2322 		if (filestruct_locked != 0 && ret == 0) {
2323 			/* Roll back local locks. */
2324 			NFSUNLOCKSTATE();
2325 			if (vnode_unlocked == 0) {
2326 				ASSERT_VOP_ELOCKED(vp, "nfsrv_lockctrl7");
2327 				vnode_unlocked = 1;
2328 				NFSVOPUNLOCK(vp);
2329 			}
2330 			nfsrv_locallock_rollback(vp, lfp, p);
2331 			NFSLOCKSTATE();
2332 			nfsrv_unlocklf(lfp);
2333 		}
2334 		if (ret == 0)
2335 			NFSUNLOCKSTATE();
2336 		goto out;
2337 	    }
2338 	  }
2339 	}
2340 
2341 	/*
2342 	 * We only get here if there was no lock that conflicted.
2343 	 */
2344 	if (new_stp->ls_flags & (NFSLCK_TEST | NFSLCK_CHECK)) {
2345 		NFSUNLOCKSTATE();
2346 		goto out;
2347 	}
2348 
2349 	/*
2350 	 * We only get here when we are creating or modifying a lock.
2351 	 * There are two variants:
2352 	 * - exist_lock_owner where lock_owner exists
2353 	 * - open_to_lock_owner with new lock_owner
2354 	 */
2355 	first = new_lop->lo_first;
2356 	end = new_lop->lo_end;
2357 	lock_flags = new_lop->lo_flags;
2358 	if (!(new_stp->ls_flags & NFSLCK_OPENTOLOCK)) {
2359 		nfsrv_updatelock(lckstp, new_lopp, &other_lop, lfp);
2360 		stateidp->seqid = ++(lckstp->ls_stateid.seqid);
2361 		if ((nd->nd_flag & ND_NFSV41) != 0 && stateidp->seqid == 0)
2362 			stateidp->seqid = lckstp->ls_stateid.seqid = 1;
2363 		stateidp->other[0] = lckstp->ls_stateid.other[0];
2364 		stateidp->other[1] = lckstp->ls_stateid.other[1];
2365 		stateidp->other[2] = lckstp->ls_stateid.other[2];
2366 	} else {
2367 		/*
2368 		 * The new open_to_lock_owner case.
2369 		 * Link the new nfsstate into the lists.
2370 		 */
2371 		new_stp->ls_seq = new_stp->ls_opentolockseq;
2372 		nfsrvd_refcache(new_stp->ls_op);
2373 		stateidp->seqid = new_stp->ls_stateid.seqid = 1;
2374 		stateidp->other[0] = new_stp->ls_stateid.other[0] =
2375 		    clp->lc_clientid.lval[0];
2376 		stateidp->other[1] = new_stp->ls_stateid.other[1] =
2377 		    clp->lc_clientid.lval[1];
2378 		stateidp->other[2] = new_stp->ls_stateid.other[2] =
2379 		    nfsrv_nextstateindex(clp);
2380 		new_stp->ls_clp = clp;
2381 		LIST_INIT(&new_stp->ls_lock);
2382 		new_stp->ls_openstp = stp;
2383 		new_stp->ls_lfp = lfp;
2384 		nfsrv_insertlock(new_lop, (struct nfslock *)new_stp, new_stp,
2385 		    lfp);
2386 		LIST_INSERT_HEAD(NFSSTATEHASH(clp, new_stp->ls_stateid),
2387 		    new_stp, ls_hash);
2388 		LIST_INSERT_HEAD(&stp->ls_open, new_stp, ls_list);
2389 		*new_lopp = NULL;
2390 		*new_stpp = NULL;
2391 		nfsstatsv1.srvlockowners++;
2392 		nfsrv_openpluslock++;
2393 	}
2394 	if (filestruct_locked != 0) {
2395 		NFSUNLOCKSTATE();
2396 		nfsrv_locallock_commit(lfp, lock_flags, first, end);
2397 		NFSLOCKSTATE();
2398 		nfsrv_unlocklf(lfp);
2399 	}
2400 	NFSUNLOCKSTATE();
2401 
2402 out:
2403 	if (haslock) {
2404 		NFSLOCKV4ROOTMUTEX();
2405 		nfsv4_unlock(&nfsv4rootfs_lock, 1);
2406 		NFSUNLOCKV4ROOTMUTEX();
2407 	}
2408 	if (vnode_unlocked != 0) {
2409 		NFSVOPLOCK(vp, LK_EXCLUSIVE | LK_RETRY);
2410 		if (error == 0 && VN_IS_DOOMED(vp))
2411 			error = NFSERR_SERVERFAULT;
2412 	}
2413 	if (other_lop)
2414 		free(other_lop, M_NFSDLOCK);
2415 	NFSEXITCODE2(error, nd);
2416 	return (error);
2417 }
2418 
2419 /*
2420  * Check for state errors for Open.
2421  * repstat is passed back out as an error if more critical errors
2422  * are not detected.
2423  */
2424 int
2425 nfsrv_opencheck(nfsquad_t clientid, nfsv4stateid_t *stateidp,
2426     struct nfsstate *new_stp, vnode_t vp, struct nfsrv_descript *nd,
2427     NFSPROC_T *p, int repstat)
2428 {
2429 	struct nfsstate *stp, *nstp;
2430 	struct nfsclient *clp;
2431 	struct nfsstate *ownerstp;
2432 	struct nfslockfile *lfp, *new_lfp;
2433 	int error = 0, haslock = 0, ret, readonly = 0, getfhret = 0;
2434 
2435 	if ((new_stp->ls_flags & NFSLCK_SHAREBITS) == NFSLCK_READACCESS)
2436 		readonly = 1;
2437 	/*
2438 	 * Check for restart conditions (client and server).
2439 	 */
2440 	error = nfsrv_checkrestart(clientid, new_stp->ls_flags,
2441 		&new_stp->ls_stateid, 0);
2442 	if (error)
2443 		goto out;
2444 
2445 	/*
2446 	 * Check for state resource limit exceeded.
2447 	 * Technically this should be SMP protected, but the worst
2448 	 * case error is "out by one or two" on the count when it
2449 	 * returns NFSERR_RESOURCE and the limit is just a rather
2450 	 * arbitrary high water mark, so no harm is done.
2451 	 */
2452 	if (nfsrv_openpluslock > nfsrv_v4statelimit) {
2453 		error = NFSERR_RESOURCE;
2454 		goto out;
2455 	}
2456 
2457 tryagain:
2458 	new_lfp = malloc(sizeof (struct nfslockfile),
2459 	    M_NFSDLOCKFILE, M_WAITOK);
2460 	if (vp)
2461 		getfhret = nfsrv_getlockfh(vp, new_stp->ls_flags, new_lfp,
2462 		    NULL, p);
2463 	NFSLOCKSTATE();
2464 	/*
2465 	 * Get the nfsclient structure.
2466 	 */
2467 	error = nfsrv_getclient(clientid, CLOPS_RENEW, &clp, NULL,
2468 	    (nfsquad_t)((u_quad_t)0), 0, nd, p);
2469 
2470 	/*
2471 	 * Look up the open owner. See if it needs confirmation and
2472 	 * check the seq#, as required.
2473 	 */
2474 	if (!error)
2475 		nfsrv_getowner(&clp->lc_open, new_stp, &ownerstp);
2476 
2477 	if (!error && ownerstp) {
2478 		error = nfsrv_checkseqid(nd, new_stp->ls_seq, ownerstp,
2479 		    new_stp->ls_op);
2480 		/*
2481 		 * If the OpenOwner hasn't been confirmed, assume the
2482 		 * old one was a replay and this one is ok.
2483 		 * See: RFC3530 Sec. 14.2.18.
2484 		 */
2485 		if (error == NFSERR_BADSEQID &&
2486 		    (ownerstp->ls_flags & NFSLCK_NEEDSCONFIRM))
2487 			error = 0;
2488 	}
2489 
2490 	/*
2491 	 * Check for grace.
2492 	 */
2493 	if (!error)
2494 		error = nfsrv_checkgrace(nd, clp, new_stp->ls_flags);
2495 	if ((new_stp->ls_flags & NFSLCK_RECLAIM) && !error &&
2496 		nfsrv_checkstable(clp))
2497 		error = NFSERR_NOGRACE;
2498 
2499 	/*
2500 	 * If none of the above errors occurred, let repstat be
2501 	 * returned.
2502 	 */
2503 	if (repstat && !error)
2504 		error = repstat;
2505 	if (error) {
2506 		NFSUNLOCKSTATE();
2507 		if (haslock) {
2508 			NFSLOCKV4ROOTMUTEX();
2509 			nfsv4_unlock(&nfsv4rootfs_lock, 1);
2510 			NFSUNLOCKV4ROOTMUTEX();
2511 		}
2512 		free(new_lfp, M_NFSDLOCKFILE);
2513 		goto out;
2514 	}
2515 
2516 	/*
2517 	 * If vp == NULL, the file doesn't exist yet, so return ok.
2518 	 * (This always happens on the first pass, so haslock must be 0.)
2519 	 */
2520 	if (vp == NULL) {
2521 		NFSUNLOCKSTATE();
2522 		free(new_lfp, M_NFSDLOCKFILE);
2523 		goto out;
2524 	}
2525 
2526 	/*
2527 	 * Get the structure for the underlying file.
2528 	 */
2529 	if (getfhret)
2530 		error = getfhret;
2531 	else
2532 		error = nfsrv_getlockfile(new_stp->ls_flags, &new_lfp, &lfp,
2533 		    NULL, 0);
2534 	if (new_lfp)
2535 		free(new_lfp, M_NFSDLOCKFILE);
2536 	if (error) {
2537 		NFSUNLOCKSTATE();
2538 		if (haslock) {
2539 			NFSLOCKV4ROOTMUTEX();
2540 			nfsv4_unlock(&nfsv4rootfs_lock, 1);
2541 			NFSUNLOCKV4ROOTMUTEX();
2542 		}
2543 		goto out;
2544 	}
2545 
2546 	/*
2547 	 * Search for a conflicting open/share.
2548 	 */
2549 	if (new_stp->ls_flags & NFSLCK_DELEGCUR) {
2550 	    /*
2551 	     * For Delegate_Cur, search for the matching Delegation,
2552 	     * which indicates no conflict.
2553 	     * An old delegation should have been recovered by the
2554 	     * client doing a Claim_DELEGATE_Prev, so I won't let
2555 	     * it match and return NFSERR_EXPIRED. Should I let it
2556 	     * match?
2557 	     */
2558 	    LIST_FOREACH(stp, &lfp->lf_deleg, ls_file) {
2559 		if (!(stp->ls_flags & NFSLCK_OLDDELEG) &&
2560 		    (((nd->nd_flag & ND_NFSV41) != 0 &&
2561 		    stateidp->seqid == 0) ||
2562 		    stateidp->seqid == stp->ls_stateid.seqid) &&
2563 		    !NFSBCMP(stateidp->other, stp->ls_stateid.other,
2564 			  NFSX_STATEIDOTHER))
2565 			break;
2566 	    }
2567 	    if (stp == LIST_END(&lfp->lf_deleg) ||
2568 		((new_stp->ls_flags & NFSLCK_WRITEACCESS) &&
2569 		 (stp->ls_flags & NFSLCK_DELEGREAD))) {
2570 		NFSUNLOCKSTATE();
2571 		if (haslock) {
2572 			NFSLOCKV4ROOTMUTEX();
2573 			nfsv4_unlock(&nfsv4rootfs_lock, 1);
2574 			NFSUNLOCKV4ROOTMUTEX();
2575 		}
2576 		error = NFSERR_EXPIRED;
2577 		goto out;
2578 	    }
2579 	}
2580 
2581 	/*
2582 	 * Check for access/deny bit conflicts. I check for the same
2583 	 * owner as well, in case the client didn't bother.
2584 	 */
2585 	LIST_FOREACH(stp, &lfp->lf_open, ls_file) {
2586 		if (!(new_stp->ls_flags & NFSLCK_DELEGCUR) &&
2587 		    (((new_stp->ls_flags & NFSLCK_ACCESSBITS) &
2588 		      ((stp->ls_flags>>NFSLCK_SHIFT) & NFSLCK_ACCESSBITS))||
2589 		     ((stp->ls_flags & NFSLCK_ACCESSBITS) &
2590 		      ((new_stp->ls_flags>>NFSLCK_SHIFT)&NFSLCK_ACCESSBITS)))){
2591 			ret = nfsrv_clientconflict(stp->ls_clp,&haslock,vp,p);
2592 			if (ret == 1) {
2593 				/*
2594 				 * nfsrv_clientconflict() unlocks
2595 				 * state when it returns non-zero.
2596 				 */
2597 				goto tryagain;
2598 			}
2599 			if (ret == 2)
2600 				error = NFSERR_PERM;
2601 			else if (new_stp->ls_flags & NFSLCK_RECLAIM)
2602 				error = NFSERR_RECLAIMCONFLICT;
2603 			else
2604 				error = NFSERR_SHAREDENIED;
2605 			if (ret == 0)
2606 				NFSUNLOCKSTATE();
2607 			if (haslock) {
2608 				NFSLOCKV4ROOTMUTEX();
2609 				nfsv4_unlock(&nfsv4rootfs_lock, 1);
2610 				NFSUNLOCKV4ROOTMUTEX();
2611 			}
2612 			goto out;
2613 		}
2614 	}
2615 
2616 	/*
2617 	 * Check for a conflicting delegation. If one is found, call
2618 	 * nfsrv_delegconflict() to handle it. If the v4root lock hasn't
2619 	 * been set yet, it will get the lock. Otherwise, it will recall
2620 	 * the delegation. Then, we try try again...
2621 	 * (If NFSLCK_DELEGCUR is set, it has a delegation, so there
2622 	 *  isn't a conflict.)
2623 	 * I currently believe the conflict algorithm to be:
2624 	 * For Open with Read Access and Deny None
2625 	 * - there is a conflict iff a different client has a write delegation
2626 	 * For Open with other Write Access or any Deny except None
2627 	 * - there is a conflict if a different client has any delegation
2628 	 * - there is a conflict if the same client has a read delegation
2629 	 *   (The current consensus is that this last case should be
2630 	 *    considered a conflict since the client with a read delegation
2631 	 *    could have done an Open with ReadAccess and WriteDeny
2632 	 *    locally and then not have checked for the WriteDeny.)
2633 	 * Don't check for a Reclaim, since that will be dealt with
2634 	 * by nfsrv_openctrl().
2635 	 */
2636 	if (!(new_stp->ls_flags &
2637 		(NFSLCK_DELEGPREV | NFSLCK_DELEGCUR | NFSLCK_RECLAIM))) {
2638 	    stp = LIST_FIRST(&lfp->lf_deleg);
2639 	    while (stp != LIST_END(&lfp->lf_deleg)) {
2640 		nstp = LIST_NEXT(stp, ls_file);
2641 		if ((readonly && stp->ls_clp != clp &&
2642 		       (stp->ls_flags & NFSLCK_DELEGWRITE)) ||
2643 		    (!readonly && (stp->ls_clp != clp ||
2644 		         (stp->ls_flags & NFSLCK_DELEGREAD)))) {
2645 			ret = nfsrv_delegconflict(stp, &haslock, p, vp);
2646 			if (ret) {
2647 			    /*
2648 			     * nfsrv_delegconflict() unlocks state
2649 			     * when it returns non-zero.
2650 			     */
2651 			    if (ret == -1)
2652 				goto tryagain;
2653 			    error = ret;
2654 			    goto out;
2655 			}
2656 		}
2657 		stp = nstp;
2658 	    }
2659 	}
2660 	NFSUNLOCKSTATE();
2661 	if (haslock) {
2662 		NFSLOCKV4ROOTMUTEX();
2663 		nfsv4_unlock(&nfsv4rootfs_lock, 1);
2664 		NFSUNLOCKV4ROOTMUTEX();
2665 	}
2666 
2667 out:
2668 	NFSEXITCODE2(error, nd);
2669 	return (error);
2670 }
2671 
2672 /*
2673  * Open control function to create/update open state for an open.
2674  */
2675 int
2676 nfsrv_openctrl(struct nfsrv_descript *nd, vnode_t vp,
2677     struct nfsstate **new_stpp, nfsquad_t clientid, nfsv4stateid_t *stateidp,
2678     nfsv4stateid_t *delegstateidp, u_int32_t *rflagsp, struct nfsexstuff *exp,
2679     NFSPROC_T *p, u_quad_t filerev)
2680 {
2681 	struct nfsstate *new_stp = *new_stpp;
2682 	struct nfsstate *stp, *nstp;
2683 	struct nfsstate *openstp = NULL, *new_open, *ownerstp, *new_deleg;
2684 	struct nfslockfile *lfp, *new_lfp;
2685 	struct nfsclient *clp;
2686 	int error = 0, haslock = 0, ret, delegate = 1, writedeleg = 1;
2687 	int readonly = 0, cbret = 1, getfhret = 0;
2688 	int gotstate = 0, len = 0;
2689 	u_char *clidp = NULL;
2690 
2691 	if ((new_stp->ls_flags & NFSLCK_SHAREBITS) == NFSLCK_READACCESS)
2692 		readonly = 1;
2693 	/*
2694 	 * Check for restart conditions (client and server).
2695 	 * (Paranoia, should have been detected by nfsrv_opencheck().)
2696 	 * If an error does show up, return NFSERR_EXPIRED, since the
2697 	 * the seqid# has already been incremented.
2698 	 */
2699 	error = nfsrv_checkrestart(clientid, new_stp->ls_flags,
2700 	    &new_stp->ls_stateid, 0);
2701 	if (error) {
2702 		printf("Nfsd: openctrl unexpected restart err=%d\n",
2703 		    error);
2704 		error = NFSERR_EXPIRED;
2705 		goto out;
2706 	}
2707 
2708 	clidp = malloc(NFSV4_OPAQUELIMIT, M_TEMP, M_WAITOK);
2709 tryagain:
2710 	new_lfp = malloc(sizeof (struct nfslockfile),
2711 	    M_NFSDLOCKFILE, M_WAITOK);
2712 	new_open = malloc(sizeof (struct nfsstate),
2713 	    M_NFSDSTATE, M_WAITOK);
2714 	new_deleg = malloc(sizeof (struct nfsstate),
2715 	    M_NFSDSTATE, M_WAITOK);
2716 	getfhret = nfsrv_getlockfh(vp, new_stp->ls_flags, new_lfp,
2717 	    NULL, p);
2718 	NFSLOCKSTATE();
2719 	/*
2720 	 * Get the client structure. Since the linked lists could be changed
2721 	 * by other nfsd processes if this process does a tsleep(), one of
2722 	 * two things must be done.
2723 	 * 1 - don't tsleep()
2724 	 * or
2725 	 * 2 - get the nfsv4_lock() { indicated by haslock == 1 }
2726 	 *     before using the lists, since this lock stops the other
2727 	 *     nfsd. This should only be used for rare cases, since it
2728 	 *     essentially single threads the nfsd.
2729 	 *     At this time, it is only done for cases where the stable
2730 	 *     storage file must be written prior to completion of state
2731 	 *     expiration.
2732 	 */
2733 	error = nfsrv_getclient(clientid, CLOPS_RENEW, &clp, NULL,
2734 	    (nfsquad_t)((u_quad_t)0), 0, nd, p);
2735 	if (!error && (clp->lc_flags & LCL_NEEDSCBNULL) &&
2736 	    clp->lc_program) {
2737 		/*
2738 		 * This happens on the first open for a client
2739 		 * that supports callbacks.
2740 		 */
2741 		NFSUNLOCKSTATE();
2742 		/*
2743 		 * Although nfsrv_docallback() will sleep, clp won't
2744 		 * go away, since they are only removed when the
2745 		 * nfsv4_lock() has blocked the nfsd threads. The
2746 		 * fields in clp can change, but having multiple
2747 		 * threads do this Null callback RPC should be
2748 		 * harmless.
2749 		 */
2750 		cbret = nfsrv_docallback(clp, NFSV4PROC_CBNULL,
2751 		    NULL, 0, NULL, NULL, NULL, 0, p);
2752 		NFSLOCKSTATE();
2753 		clp->lc_flags &= ~LCL_NEEDSCBNULL;
2754 		if (!cbret)
2755 			clp->lc_flags |= LCL_CALLBACKSON;
2756 	}
2757 
2758 	/*
2759 	 * Look up the open owner. See if it needs confirmation and
2760 	 * check the seq#, as required.
2761 	 */
2762 	if (!error)
2763 		nfsrv_getowner(&clp->lc_open, new_stp, &ownerstp);
2764 
2765 	if (error) {
2766 		NFSUNLOCKSTATE();
2767 		printf("Nfsd: openctrl unexpected state err=%d\n",
2768 			error);
2769 		free(new_lfp, M_NFSDLOCKFILE);
2770 		free(new_open, M_NFSDSTATE);
2771 		free(new_deleg, M_NFSDSTATE);
2772 		if (haslock) {
2773 			NFSLOCKV4ROOTMUTEX();
2774 			nfsv4_unlock(&nfsv4rootfs_lock, 1);
2775 			NFSUNLOCKV4ROOTMUTEX();
2776 		}
2777 		error = NFSERR_EXPIRED;
2778 		goto out;
2779 	}
2780 
2781 	if (new_stp->ls_flags & NFSLCK_RECLAIM)
2782 		nfsrv_markstable(clp);
2783 
2784 	/*
2785 	 * Get the structure for the underlying file.
2786 	 */
2787 	if (getfhret)
2788 		error = getfhret;
2789 	else
2790 		error = nfsrv_getlockfile(new_stp->ls_flags, &new_lfp, &lfp,
2791 		    NULL, 0);
2792 	if (new_lfp)
2793 		free(new_lfp, M_NFSDLOCKFILE);
2794 	if (error) {
2795 		NFSUNLOCKSTATE();
2796 		printf("Nfsd openctrl unexpected getlockfile err=%d\n",
2797 		    error);
2798 		free(new_open, M_NFSDSTATE);
2799 		free(new_deleg, M_NFSDSTATE);
2800 		if (haslock) {
2801 			NFSLOCKV4ROOTMUTEX();
2802 			nfsv4_unlock(&nfsv4rootfs_lock, 1);
2803 			NFSUNLOCKV4ROOTMUTEX();
2804 		}
2805 		goto out;
2806 	}
2807 
2808 	/*
2809 	 * Search for a conflicting open/share.
2810 	 */
2811 	if (new_stp->ls_flags & NFSLCK_DELEGCUR) {
2812 	    /*
2813 	     * For Delegate_Cur, search for the matching Delegation,
2814 	     * which indicates no conflict.
2815 	     * An old delegation should have been recovered by the
2816 	     * client doing a Claim_DELEGATE_Prev, so I won't let
2817 	     * it match and return NFSERR_EXPIRED. Should I let it
2818 	     * match?
2819 	     */
2820 	    LIST_FOREACH(stp, &lfp->lf_deleg, ls_file) {
2821 		if (!(stp->ls_flags & NFSLCK_OLDDELEG) &&
2822 		    (((nd->nd_flag & ND_NFSV41) != 0 &&
2823 		    stateidp->seqid == 0) ||
2824 		    stateidp->seqid == stp->ls_stateid.seqid) &&
2825 		    !NFSBCMP(stateidp->other, stp->ls_stateid.other,
2826 			NFSX_STATEIDOTHER))
2827 			break;
2828 	    }
2829 	    if (stp == LIST_END(&lfp->lf_deleg) ||
2830 		((new_stp->ls_flags & NFSLCK_WRITEACCESS) &&
2831 		 (stp->ls_flags & NFSLCK_DELEGREAD))) {
2832 		NFSUNLOCKSTATE();
2833 		printf("Nfsd openctrl unexpected expiry\n");
2834 		free(new_open, M_NFSDSTATE);
2835 		free(new_deleg, M_NFSDSTATE);
2836 		if (haslock) {
2837 			NFSLOCKV4ROOTMUTEX();
2838 			nfsv4_unlock(&nfsv4rootfs_lock, 1);
2839 			NFSUNLOCKV4ROOTMUTEX();
2840 		}
2841 		error = NFSERR_EXPIRED;
2842 		goto out;
2843 	    }
2844 
2845 	    /*
2846 	     * Don't issue a Delegation, since one already exists and
2847 	     * delay delegation timeout, as required.
2848 	     */
2849 	    delegate = 0;
2850 	    nfsrv_delaydelegtimeout(stp);
2851 	}
2852 
2853 	/*
2854 	 * Check for access/deny bit conflicts. I also check for the
2855 	 * same owner, since the client might not have bothered to check.
2856 	 * Also, note an open for the same file and owner, if found,
2857 	 * which is all we do here for Delegate_Cur, since conflict
2858 	 * checking is already done.
2859 	 */
2860 	LIST_FOREACH(stp, &lfp->lf_open, ls_file) {
2861 		if (ownerstp && stp->ls_openowner == ownerstp)
2862 			openstp = stp;
2863 		if (!(new_stp->ls_flags & NFSLCK_DELEGCUR)) {
2864 		    /*
2865 		     * If another client has the file open, the only
2866 		     * delegation that can be issued is a Read delegation
2867 		     * and only if it is a Read open with Deny none.
2868 		     */
2869 		    if (clp != stp->ls_clp) {
2870 			if ((stp->ls_flags & NFSLCK_SHAREBITS) ==
2871 			    NFSLCK_READACCESS)
2872 			    writedeleg = 0;
2873 			else
2874 			    delegate = 0;
2875 		    }
2876 		    if(((new_stp->ls_flags & NFSLCK_ACCESSBITS) &
2877 		        ((stp->ls_flags>>NFSLCK_SHIFT) & NFSLCK_ACCESSBITS))||
2878 		       ((stp->ls_flags & NFSLCK_ACCESSBITS) &
2879 		        ((new_stp->ls_flags>>NFSLCK_SHIFT)&NFSLCK_ACCESSBITS))){
2880 			ret = nfsrv_clientconflict(stp->ls_clp,&haslock,vp,p);
2881 			if (ret == 1) {
2882 				/*
2883 				 * nfsrv_clientconflict() unlocks state
2884 				 * when it returns non-zero.
2885 				 */
2886 				free(new_open, M_NFSDSTATE);
2887 				free(new_deleg, M_NFSDSTATE);
2888 				openstp = NULL;
2889 				goto tryagain;
2890 			}
2891 			if (ret == 2)
2892 				error = NFSERR_PERM;
2893 			else if (new_stp->ls_flags & NFSLCK_RECLAIM)
2894 				error = NFSERR_RECLAIMCONFLICT;
2895 			else
2896 				error = NFSERR_SHAREDENIED;
2897 			if (ret == 0)
2898 				NFSUNLOCKSTATE();
2899 			if (haslock) {
2900 				NFSLOCKV4ROOTMUTEX();
2901 				nfsv4_unlock(&nfsv4rootfs_lock, 1);
2902 				NFSUNLOCKV4ROOTMUTEX();
2903 			}
2904 			free(new_open, M_NFSDSTATE);
2905 			free(new_deleg, M_NFSDSTATE);
2906 			printf("nfsd openctrl unexpected client cnfl\n");
2907 			goto out;
2908 		    }
2909 		}
2910 	}
2911 
2912 	/*
2913 	 * Check for a conflicting delegation. If one is found, call
2914 	 * nfsrv_delegconflict() to handle it. If the v4root lock hasn't
2915 	 * been set yet, it will get the lock. Otherwise, it will recall
2916 	 * the delegation. Then, we try try again...
2917 	 * (If NFSLCK_DELEGCUR is set, it has a delegation, so there
2918 	 *  isn't a conflict.)
2919 	 * I currently believe the conflict algorithm to be:
2920 	 * For Open with Read Access and Deny None
2921 	 * - there is a conflict iff a different client has a write delegation
2922 	 * For Open with other Write Access or any Deny except None
2923 	 * - there is a conflict if a different client has any delegation
2924 	 * - there is a conflict if the same client has a read delegation
2925 	 *   (The current consensus is that this last case should be
2926 	 *    considered a conflict since the client with a read delegation
2927 	 *    could have done an Open with ReadAccess and WriteDeny
2928 	 *    locally and then not have checked for the WriteDeny.)
2929 	 */
2930 	if (!(new_stp->ls_flags & (NFSLCK_DELEGPREV | NFSLCK_DELEGCUR))) {
2931 	    stp = LIST_FIRST(&lfp->lf_deleg);
2932 	    while (stp != LIST_END(&lfp->lf_deleg)) {
2933 		nstp = LIST_NEXT(stp, ls_file);
2934 		if (stp->ls_clp != clp && (stp->ls_flags & NFSLCK_DELEGREAD))
2935 			writedeleg = 0;
2936 		else
2937 			delegate = 0;
2938 		if ((readonly && stp->ls_clp != clp &&
2939 		       (stp->ls_flags & NFSLCK_DELEGWRITE)) ||
2940 		    (!readonly && (stp->ls_clp != clp ||
2941 		         (stp->ls_flags & NFSLCK_DELEGREAD)))) {
2942 		    if (new_stp->ls_flags & NFSLCK_RECLAIM) {
2943 			delegate = 2;
2944 		    } else {
2945 			ret = nfsrv_delegconflict(stp, &haslock, p, vp);
2946 			if (ret) {
2947 			    /*
2948 			     * nfsrv_delegconflict() unlocks state
2949 			     * when it returns non-zero.
2950 			     */
2951 			    printf("Nfsd openctrl unexpected deleg cnfl\n");
2952 			    free(new_open, M_NFSDSTATE);
2953 			    free(new_deleg, M_NFSDSTATE);
2954 			    if (ret == -1) {
2955 				openstp = NULL;
2956 				goto tryagain;
2957 			    }
2958 			    error = ret;
2959 			    goto out;
2960 			}
2961 		    }
2962 		}
2963 		stp = nstp;
2964 	    }
2965 	}
2966 
2967 	/*
2968 	 * We only get here if there was no open that conflicted.
2969 	 * If an open for the owner exists, or in the access/deny bits.
2970 	 * Otherwise it is a new open. If the open_owner hasn't been
2971 	 * confirmed, replace the open with the new one needing confirmation,
2972 	 * otherwise add the open.
2973 	 */
2974 	if (new_stp->ls_flags & NFSLCK_DELEGPREV) {
2975 	    /*
2976 	     * Handle NFSLCK_DELEGPREV by searching the old delegations for
2977 	     * a match. If found, just move the old delegation to the current
2978 	     * delegation list and issue open. If not found, return
2979 	     * NFSERR_EXPIRED.
2980 	     */
2981 	    LIST_FOREACH(stp, &clp->lc_olddeleg, ls_list) {
2982 		if (stp->ls_lfp == lfp) {
2983 		    /* Found it */
2984 		    if (stp->ls_clp != clp)
2985 			panic("olddeleg clp");
2986 		    LIST_REMOVE(stp, ls_list);
2987 		    LIST_REMOVE(stp, ls_hash);
2988 		    stp->ls_flags &= ~NFSLCK_OLDDELEG;
2989 		    stp->ls_stateid.seqid = delegstateidp->seqid = 1;
2990 		    stp->ls_stateid.other[0] = delegstateidp->other[0] =
2991 			clp->lc_clientid.lval[0];
2992 		    stp->ls_stateid.other[1] = delegstateidp->other[1] =
2993 			clp->lc_clientid.lval[1];
2994 		    stp->ls_stateid.other[2] = delegstateidp->other[2] =
2995 			nfsrv_nextstateindex(clp);
2996 		    stp->ls_compref = nd->nd_compref;
2997 		    LIST_INSERT_HEAD(&clp->lc_deleg, stp, ls_list);
2998 		    LIST_INSERT_HEAD(NFSSTATEHASH(clp,
2999 			stp->ls_stateid), stp, ls_hash);
3000 		    if (stp->ls_flags & NFSLCK_DELEGWRITE)
3001 			*rflagsp |= NFSV4OPEN_WRITEDELEGATE;
3002 		    else
3003 			*rflagsp |= NFSV4OPEN_READDELEGATE;
3004 		    clp->lc_delegtime = NFSD_MONOSEC +
3005 			nfsrv_lease + NFSRV_LEASEDELTA;
3006 
3007 		    /*
3008 		     * Now, do the associated open.
3009 		     */
3010 		    new_open->ls_stateid.seqid = 1;
3011 		    new_open->ls_stateid.other[0] = clp->lc_clientid.lval[0];
3012 		    new_open->ls_stateid.other[1] = clp->lc_clientid.lval[1];
3013 		    new_open->ls_stateid.other[2] = nfsrv_nextstateindex(clp);
3014 		    new_open->ls_flags = (new_stp->ls_flags&NFSLCK_DENYBITS)|
3015 			NFSLCK_OPEN;
3016 		    if (stp->ls_flags & NFSLCK_DELEGWRITE)
3017 			new_open->ls_flags |= (NFSLCK_READACCESS |
3018 			    NFSLCK_WRITEACCESS);
3019 		    else
3020 			new_open->ls_flags |= NFSLCK_READACCESS;
3021 		    new_open->ls_uid = new_stp->ls_uid;
3022 		    new_open->ls_lfp = lfp;
3023 		    new_open->ls_clp = clp;
3024 		    LIST_INIT(&new_open->ls_open);
3025 		    LIST_INSERT_HEAD(&lfp->lf_open, new_open, ls_file);
3026 		    LIST_INSERT_HEAD(NFSSTATEHASH(clp, new_open->ls_stateid),
3027 			new_open, ls_hash);
3028 		    /*
3029 		     * and handle the open owner
3030 		     */
3031 		    if (ownerstp) {
3032 			new_open->ls_openowner = ownerstp;
3033 			LIST_INSERT_HEAD(&ownerstp->ls_open,new_open,ls_list);
3034 		    } else {
3035 			new_open->ls_openowner = new_stp;
3036 			new_stp->ls_flags = 0;
3037 			nfsrvd_refcache(new_stp->ls_op);
3038 			new_stp->ls_noopens = 0;
3039 			LIST_INIT(&new_stp->ls_open);
3040 			LIST_INSERT_HEAD(&new_stp->ls_open, new_open, ls_list);
3041 			LIST_INSERT_HEAD(&clp->lc_open, new_stp, ls_list);
3042 			*new_stpp = NULL;
3043 			nfsstatsv1.srvopenowners++;
3044 			nfsrv_openpluslock++;
3045 		    }
3046 		    openstp = new_open;
3047 		    new_open = NULL;
3048 		    nfsstatsv1.srvopens++;
3049 		    nfsrv_openpluslock++;
3050 		    break;
3051 		}
3052 	    }
3053 	    if (stp == LIST_END(&clp->lc_olddeleg))
3054 		error = NFSERR_EXPIRED;
3055 	} else if (new_stp->ls_flags & (NFSLCK_DELEGREAD | NFSLCK_DELEGWRITE)) {
3056 	    /*
3057 	     * Scan to see that no delegation for this client and file
3058 	     * doesn't already exist.
3059 	     * There also shouldn't yet be an Open for this file and
3060 	     * openowner.
3061 	     */
3062 	    LIST_FOREACH(stp, &lfp->lf_deleg, ls_file) {
3063 		if (stp->ls_clp == clp)
3064 		    break;
3065 	    }
3066 	    if (stp == LIST_END(&lfp->lf_deleg) && openstp == NULL) {
3067 		/*
3068 		 * This is the Claim_Previous case with a delegation
3069 		 * type != Delegate_None.
3070 		 */
3071 		/*
3072 		 * First, add the delegation. (Although we must issue the
3073 		 * delegation, we can also ask for an immediate return.)
3074 		 */
3075 		new_deleg->ls_stateid.seqid = delegstateidp->seqid = 1;
3076 		new_deleg->ls_stateid.other[0] = delegstateidp->other[0] =
3077 		    clp->lc_clientid.lval[0];
3078 		new_deleg->ls_stateid.other[1] = delegstateidp->other[1] =
3079 		    clp->lc_clientid.lval[1];
3080 		new_deleg->ls_stateid.other[2] = delegstateidp->other[2] =
3081 		    nfsrv_nextstateindex(clp);
3082 		if (new_stp->ls_flags & NFSLCK_DELEGWRITE) {
3083 		    new_deleg->ls_flags = (NFSLCK_DELEGWRITE |
3084 			NFSLCK_READACCESS | NFSLCK_WRITEACCESS);
3085 		    *rflagsp |= NFSV4OPEN_WRITEDELEGATE;
3086 		    nfsrv_writedelegcnt++;
3087 		} else {
3088 		    new_deleg->ls_flags = (NFSLCK_DELEGREAD |
3089 			NFSLCK_READACCESS);
3090 		    *rflagsp |= NFSV4OPEN_READDELEGATE;
3091 		}
3092 		new_deleg->ls_uid = new_stp->ls_uid;
3093 		new_deleg->ls_lfp = lfp;
3094 		new_deleg->ls_clp = clp;
3095 		new_deleg->ls_filerev = filerev;
3096 		new_deleg->ls_compref = nd->nd_compref;
3097 		new_deleg->ls_lastrecall = 0;
3098 		LIST_INSERT_HEAD(&lfp->lf_deleg, new_deleg, ls_file);
3099 		LIST_INSERT_HEAD(NFSSTATEHASH(clp,
3100 		    new_deleg->ls_stateid), new_deleg, ls_hash);
3101 		LIST_INSERT_HEAD(&clp->lc_deleg, new_deleg, ls_list);
3102 		new_deleg = NULL;
3103 		if (delegate == 2 || nfsrv_issuedelegs == 0 ||
3104 		    (clp->lc_flags & (LCL_CALLBACKSON | LCL_CBDOWN)) !=
3105 		     LCL_CALLBACKSON ||
3106 		    NFSRV_V4DELEGLIMIT(nfsrv_delegatecnt) ||
3107 		    !NFSVNO_DELEGOK(vp))
3108 		    *rflagsp |= NFSV4OPEN_RECALL;
3109 		nfsstatsv1.srvdelegates++;
3110 		nfsrv_openpluslock++;
3111 		nfsrv_delegatecnt++;
3112 
3113 		/*
3114 		 * Now, do the associated open.
3115 		 */
3116 		new_open->ls_stateid.seqid = 1;
3117 		new_open->ls_stateid.other[0] = clp->lc_clientid.lval[0];
3118 		new_open->ls_stateid.other[1] = clp->lc_clientid.lval[1];
3119 		new_open->ls_stateid.other[2] = nfsrv_nextstateindex(clp);
3120 		new_open->ls_flags = (new_stp->ls_flags & NFSLCK_DENYBITS) |
3121 		    NFSLCK_OPEN;
3122 		if (new_stp->ls_flags & NFSLCK_DELEGWRITE)
3123 			new_open->ls_flags |= (NFSLCK_READACCESS |
3124 			    NFSLCK_WRITEACCESS);
3125 		else
3126 			new_open->ls_flags |= NFSLCK_READACCESS;
3127 		new_open->ls_uid = new_stp->ls_uid;
3128 		new_open->ls_lfp = lfp;
3129 		new_open->ls_clp = clp;
3130 		LIST_INIT(&new_open->ls_open);
3131 		LIST_INSERT_HEAD(&lfp->lf_open, new_open, ls_file);
3132 		LIST_INSERT_HEAD(NFSSTATEHASH(clp, new_open->ls_stateid),
3133 		   new_open, ls_hash);
3134 		/*
3135 		 * and handle the open owner
3136 		 */
3137 		if (ownerstp) {
3138 		    new_open->ls_openowner = ownerstp;
3139 		    LIST_INSERT_HEAD(&ownerstp->ls_open, new_open, ls_list);
3140 		} else {
3141 		    new_open->ls_openowner = new_stp;
3142 		    new_stp->ls_flags = 0;
3143 		    nfsrvd_refcache(new_stp->ls_op);
3144 		    new_stp->ls_noopens = 0;
3145 		    LIST_INIT(&new_stp->ls_open);
3146 		    LIST_INSERT_HEAD(&new_stp->ls_open, new_open, ls_list);
3147 		    LIST_INSERT_HEAD(&clp->lc_open, new_stp, ls_list);
3148 		    *new_stpp = NULL;
3149 		    nfsstatsv1.srvopenowners++;
3150 		    nfsrv_openpluslock++;
3151 		}
3152 		openstp = new_open;
3153 		new_open = NULL;
3154 		nfsstatsv1.srvopens++;
3155 		nfsrv_openpluslock++;
3156 	    } else {
3157 		error = NFSERR_RECLAIMCONFLICT;
3158 	    }
3159 	} else if (ownerstp) {
3160 		if (ownerstp->ls_flags & NFSLCK_NEEDSCONFIRM) {
3161 		    /* Replace the open */
3162 		    if (ownerstp->ls_op)
3163 			nfsrvd_derefcache(ownerstp->ls_op);
3164 		    ownerstp->ls_op = new_stp->ls_op;
3165 		    nfsrvd_refcache(ownerstp->ls_op);
3166 		    ownerstp->ls_seq = new_stp->ls_seq;
3167 		    *rflagsp |= NFSV4OPEN_RESULTCONFIRM;
3168 		    stp = LIST_FIRST(&ownerstp->ls_open);
3169 		    stp->ls_flags = (new_stp->ls_flags & NFSLCK_SHAREBITS) |
3170 			NFSLCK_OPEN;
3171 		    stp->ls_stateid.seqid = 1;
3172 		    stp->ls_uid = new_stp->ls_uid;
3173 		    if (lfp != stp->ls_lfp) {
3174 			LIST_REMOVE(stp, ls_file);
3175 			LIST_INSERT_HEAD(&lfp->lf_open, stp, ls_file);
3176 			stp->ls_lfp = lfp;
3177 		    }
3178 		    openstp = stp;
3179 		} else if (openstp) {
3180 		    openstp->ls_flags |= (new_stp->ls_flags & NFSLCK_SHAREBITS);
3181 		    openstp->ls_stateid.seqid++;
3182 		    if ((nd->nd_flag & ND_NFSV41) != 0 &&
3183 			openstp->ls_stateid.seqid == 0)
3184 			openstp->ls_stateid.seqid = 1;
3185 
3186 		    /*
3187 		     * This is where we can choose to issue a delegation.
3188 		     */
3189 		    if ((new_stp->ls_flags & NFSLCK_WANTNODELEG) != 0)
3190 			*rflagsp |= NFSV4OPEN_WDNOTWANTED;
3191 		    else if (nfsrv_issuedelegs == 0)
3192 			*rflagsp |= NFSV4OPEN_WDSUPPFTYPE;
3193 		    else if (NFSRV_V4DELEGLIMIT(nfsrv_delegatecnt))
3194 			*rflagsp |= NFSV4OPEN_WDRESOURCE;
3195 		    else if (delegate == 0 || writedeleg == 0 ||
3196 			NFSVNO_EXRDONLY(exp) || (readonly != 0 &&
3197 			nfsrv_writedelegifpos == 0) ||
3198 			!NFSVNO_DELEGOK(vp) ||
3199 			(new_stp->ls_flags & NFSLCK_WANTRDELEG) != 0 ||
3200 			(clp->lc_flags & (LCL_CALLBACKSON | LCL_CBDOWN)) !=
3201 			 LCL_CALLBACKSON)
3202 			*rflagsp |= NFSV4OPEN_WDCONTENTION;
3203 		    else {
3204 			new_deleg->ls_stateid.seqid = delegstateidp->seqid = 1;
3205 			new_deleg->ls_stateid.other[0] = delegstateidp->other[0]
3206 			    = clp->lc_clientid.lval[0];
3207 			new_deleg->ls_stateid.other[1] = delegstateidp->other[1]
3208 			    = clp->lc_clientid.lval[1];
3209 			new_deleg->ls_stateid.other[2] = delegstateidp->other[2]
3210 			    = nfsrv_nextstateindex(clp);
3211 			new_deleg->ls_flags = (NFSLCK_DELEGWRITE |
3212 			    NFSLCK_READACCESS | NFSLCK_WRITEACCESS);
3213 			*rflagsp |= NFSV4OPEN_WRITEDELEGATE;
3214 			new_deleg->ls_uid = new_stp->ls_uid;
3215 			new_deleg->ls_lfp = lfp;
3216 			new_deleg->ls_clp = clp;
3217 			new_deleg->ls_filerev = filerev;
3218 			new_deleg->ls_compref = nd->nd_compref;
3219 			new_deleg->ls_lastrecall = 0;
3220 			nfsrv_writedelegcnt++;
3221 			LIST_INSERT_HEAD(&lfp->lf_deleg, new_deleg, ls_file);
3222 			LIST_INSERT_HEAD(NFSSTATEHASH(clp,
3223 			    new_deleg->ls_stateid), new_deleg, ls_hash);
3224 			LIST_INSERT_HEAD(&clp->lc_deleg, new_deleg, ls_list);
3225 			new_deleg = NULL;
3226 			nfsstatsv1.srvdelegates++;
3227 			nfsrv_openpluslock++;
3228 			nfsrv_delegatecnt++;
3229 		    }
3230 		} else {
3231 		    new_open->ls_stateid.seqid = 1;
3232 		    new_open->ls_stateid.other[0] = clp->lc_clientid.lval[0];
3233 		    new_open->ls_stateid.other[1] = clp->lc_clientid.lval[1];
3234 		    new_open->ls_stateid.other[2] = nfsrv_nextstateindex(clp);
3235 		    new_open->ls_flags = (new_stp->ls_flags & NFSLCK_SHAREBITS)|
3236 			NFSLCK_OPEN;
3237 		    new_open->ls_uid = new_stp->ls_uid;
3238 		    new_open->ls_openowner = ownerstp;
3239 		    new_open->ls_lfp = lfp;
3240 		    new_open->ls_clp = clp;
3241 		    LIST_INIT(&new_open->ls_open);
3242 		    LIST_INSERT_HEAD(&lfp->lf_open, new_open, ls_file);
3243 		    LIST_INSERT_HEAD(&ownerstp->ls_open, new_open, ls_list);
3244 		    LIST_INSERT_HEAD(NFSSTATEHASH(clp, new_open->ls_stateid),
3245 			new_open, ls_hash);
3246 		    openstp = new_open;
3247 		    new_open = NULL;
3248 		    nfsstatsv1.srvopens++;
3249 		    nfsrv_openpluslock++;
3250 
3251 		    /*
3252 		     * This is where we can choose to issue a delegation.
3253 		     */
3254 		    if ((new_stp->ls_flags & NFSLCK_WANTNODELEG) != 0)
3255 			*rflagsp |= NFSV4OPEN_WDNOTWANTED;
3256 		    else if (nfsrv_issuedelegs == 0)
3257 			*rflagsp |= NFSV4OPEN_WDSUPPFTYPE;
3258 		    else if (NFSRV_V4DELEGLIMIT(nfsrv_delegatecnt))
3259 			*rflagsp |= NFSV4OPEN_WDRESOURCE;
3260 		    else if (delegate == 0 || (writedeleg == 0 &&
3261 			readonly == 0) || !NFSVNO_DELEGOK(vp) ||
3262 			(clp->lc_flags & (LCL_CALLBACKSON | LCL_CBDOWN)) !=
3263 			 LCL_CALLBACKSON)
3264 			*rflagsp |= NFSV4OPEN_WDCONTENTION;
3265 		    else {
3266 			new_deleg->ls_stateid.seqid = delegstateidp->seqid = 1;
3267 			new_deleg->ls_stateid.other[0] = delegstateidp->other[0]
3268 			    = clp->lc_clientid.lval[0];
3269 			new_deleg->ls_stateid.other[1] = delegstateidp->other[1]
3270 			    = clp->lc_clientid.lval[1];
3271 			new_deleg->ls_stateid.other[2] = delegstateidp->other[2]
3272 			    = nfsrv_nextstateindex(clp);
3273 			if (writedeleg && !NFSVNO_EXRDONLY(exp) &&
3274 			    (nfsrv_writedelegifpos || !readonly) &&
3275 			    (new_stp->ls_flags & NFSLCK_WANTRDELEG) == 0) {
3276 			    new_deleg->ls_flags = (NFSLCK_DELEGWRITE |
3277 				NFSLCK_READACCESS | NFSLCK_WRITEACCESS);
3278 			    *rflagsp |= NFSV4OPEN_WRITEDELEGATE;
3279 			    nfsrv_writedelegcnt++;
3280 			} else {
3281 			    new_deleg->ls_flags = (NFSLCK_DELEGREAD |
3282 				NFSLCK_READACCESS);
3283 			    *rflagsp |= NFSV4OPEN_READDELEGATE;
3284 			}
3285 			new_deleg->ls_uid = new_stp->ls_uid;
3286 			new_deleg->ls_lfp = lfp;
3287 			new_deleg->ls_clp = clp;
3288 			new_deleg->ls_filerev = filerev;
3289 			new_deleg->ls_compref = nd->nd_compref;
3290 			new_deleg->ls_lastrecall = 0;
3291 			LIST_INSERT_HEAD(&lfp->lf_deleg, new_deleg, ls_file);
3292 			LIST_INSERT_HEAD(NFSSTATEHASH(clp,
3293 			    new_deleg->ls_stateid), new_deleg, ls_hash);
3294 			LIST_INSERT_HEAD(&clp->lc_deleg, new_deleg, ls_list);
3295 			new_deleg = NULL;
3296 			nfsstatsv1.srvdelegates++;
3297 			nfsrv_openpluslock++;
3298 			nfsrv_delegatecnt++;
3299 		    }
3300 		}
3301 	} else {
3302 		/*
3303 		 * New owner case. Start the open_owner sequence with a
3304 		 * Needs confirmation (unless a reclaim) and hang the
3305 		 * new open off it.
3306 		 */
3307 		new_open->ls_stateid.seqid = 1;
3308 		new_open->ls_stateid.other[0] = clp->lc_clientid.lval[0];
3309 		new_open->ls_stateid.other[1] = clp->lc_clientid.lval[1];
3310 		new_open->ls_stateid.other[2] = nfsrv_nextstateindex(clp);
3311 		new_open->ls_flags = (new_stp->ls_flags & NFSLCK_SHAREBITS) |
3312 		    NFSLCK_OPEN;
3313 		new_open->ls_uid = new_stp->ls_uid;
3314 		LIST_INIT(&new_open->ls_open);
3315 		new_open->ls_openowner = new_stp;
3316 		new_open->ls_lfp = lfp;
3317 		new_open->ls_clp = clp;
3318 		LIST_INSERT_HEAD(&lfp->lf_open, new_open, ls_file);
3319 		if (new_stp->ls_flags & NFSLCK_RECLAIM) {
3320 			new_stp->ls_flags = 0;
3321 		} else if ((nd->nd_flag & ND_NFSV41) != 0) {
3322 			/* NFSv4.1 never needs confirmation. */
3323 			new_stp->ls_flags = 0;
3324 
3325 			/*
3326 			 * This is where we can choose to issue a delegation.
3327 			 */
3328 			if (delegate && nfsrv_issuedelegs &&
3329 			    (writedeleg || readonly) &&
3330 			    (clp->lc_flags & (LCL_CALLBACKSON | LCL_CBDOWN)) ==
3331 			     LCL_CALLBACKSON &&
3332 			    !NFSRV_V4DELEGLIMIT(nfsrv_delegatecnt) &&
3333 			    NFSVNO_DELEGOK(vp) &&
3334 			    ((nd->nd_flag & ND_NFSV41) == 0 ||
3335 			     (new_stp->ls_flags & NFSLCK_WANTNODELEG) == 0)) {
3336 				new_deleg->ls_stateid.seqid =
3337 				    delegstateidp->seqid = 1;
3338 				new_deleg->ls_stateid.other[0] =
3339 				    delegstateidp->other[0]
3340 				    = clp->lc_clientid.lval[0];
3341 				new_deleg->ls_stateid.other[1] =
3342 				    delegstateidp->other[1]
3343 				    = clp->lc_clientid.lval[1];
3344 				new_deleg->ls_stateid.other[2] =
3345 				    delegstateidp->other[2]
3346 				    = nfsrv_nextstateindex(clp);
3347 				if (writedeleg && !NFSVNO_EXRDONLY(exp) &&
3348 				    (nfsrv_writedelegifpos || !readonly) &&
3349 				    ((nd->nd_flag & ND_NFSV41) == 0 ||
3350 				     (new_stp->ls_flags & NFSLCK_WANTRDELEG) ==
3351 				     0)) {
3352 					new_deleg->ls_flags =
3353 					    (NFSLCK_DELEGWRITE |
3354 					     NFSLCK_READACCESS |
3355 					     NFSLCK_WRITEACCESS);
3356 					*rflagsp |= NFSV4OPEN_WRITEDELEGATE;
3357 					nfsrv_writedelegcnt++;
3358 				} else {
3359 					new_deleg->ls_flags =
3360 					    (NFSLCK_DELEGREAD |
3361 					     NFSLCK_READACCESS);
3362 					*rflagsp |= NFSV4OPEN_READDELEGATE;
3363 				}
3364 				new_deleg->ls_uid = new_stp->ls_uid;
3365 				new_deleg->ls_lfp = lfp;
3366 				new_deleg->ls_clp = clp;
3367 				new_deleg->ls_filerev = filerev;
3368 				new_deleg->ls_compref = nd->nd_compref;
3369 				new_deleg->ls_lastrecall = 0;
3370 				LIST_INSERT_HEAD(&lfp->lf_deleg, new_deleg,
3371 				    ls_file);
3372 				LIST_INSERT_HEAD(NFSSTATEHASH(clp,
3373 				    new_deleg->ls_stateid), new_deleg, ls_hash);
3374 				LIST_INSERT_HEAD(&clp->lc_deleg, new_deleg,
3375 				    ls_list);
3376 				new_deleg = NULL;
3377 				nfsstatsv1.srvdelegates++;
3378 				nfsrv_openpluslock++;
3379 				nfsrv_delegatecnt++;
3380 			}
3381 			/*
3382 			 * Since NFSv4.1 never does an OpenConfirm, the first
3383 			 * open state will be acquired here.
3384 			 */
3385 			if (!(clp->lc_flags & LCL_STAMPEDSTABLE)) {
3386 				clp->lc_flags |= LCL_STAMPEDSTABLE;
3387 				len = clp->lc_idlen;
3388 				NFSBCOPY(clp->lc_id, clidp, len);
3389 				gotstate = 1;
3390 			}
3391 		} else {
3392 			*rflagsp |= NFSV4OPEN_RESULTCONFIRM;
3393 			new_stp->ls_flags = NFSLCK_NEEDSCONFIRM;
3394 		}
3395 		nfsrvd_refcache(new_stp->ls_op);
3396 		new_stp->ls_noopens = 0;
3397 		LIST_INIT(&new_stp->ls_open);
3398 		LIST_INSERT_HEAD(&new_stp->ls_open, new_open, ls_list);
3399 		LIST_INSERT_HEAD(&clp->lc_open, new_stp, ls_list);
3400 		LIST_INSERT_HEAD(NFSSTATEHASH(clp, new_open->ls_stateid),
3401 		    new_open, ls_hash);
3402 		openstp = new_open;
3403 		new_open = NULL;
3404 		*new_stpp = NULL;
3405 		nfsstatsv1.srvopens++;
3406 		nfsrv_openpluslock++;
3407 		nfsstatsv1.srvopenowners++;
3408 		nfsrv_openpluslock++;
3409 	}
3410 	if (!error) {
3411 		stateidp->seqid = openstp->ls_stateid.seqid;
3412 		stateidp->other[0] = openstp->ls_stateid.other[0];
3413 		stateidp->other[1] = openstp->ls_stateid.other[1];
3414 		stateidp->other[2] = openstp->ls_stateid.other[2];
3415 	}
3416 	NFSUNLOCKSTATE();
3417 	if (haslock) {
3418 		NFSLOCKV4ROOTMUTEX();
3419 		nfsv4_unlock(&nfsv4rootfs_lock, 1);
3420 		NFSUNLOCKV4ROOTMUTEX();
3421 	}
3422 	if (new_open)
3423 		free(new_open, M_NFSDSTATE);
3424 	if (new_deleg)
3425 		free(new_deleg, M_NFSDSTATE);
3426 
3427 	/*
3428 	 * If the NFSv4.1 client just acquired its first open, write a timestamp
3429 	 * to the stable storage file.
3430 	 */
3431 	if (gotstate != 0) {
3432 		nfsrv_writestable(clidp, len, NFSNST_NEWSTATE, p);
3433 		nfsrv_backupstable();
3434 	}
3435 
3436 out:
3437 	free(clidp, M_TEMP);
3438 	NFSEXITCODE2(error, nd);
3439 	return (error);
3440 }
3441 
3442 /*
3443  * Open update. Does the confirm, downgrade and close.
3444  */
3445 int
3446 nfsrv_openupdate(vnode_t vp, struct nfsstate *new_stp, nfsquad_t clientid,
3447     nfsv4stateid_t *stateidp, struct nfsrv_descript *nd, NFSPROC_T *p,
3448     int *retwriteaccessp)
3449 {
3450 	struct nfsstate *stp;
3451 	struct nfsclient *clp;
3452 	struct nfslockfile *lfp;
3453 	u_int32_t bits;
3454 	int error = 0, gotstate = 0, len = 0;
3455 	u_char *clidp = NULL;
3456 
3457 	/*
3458 	 * Check for restart conditions (client and server).
3459 	 */
3460 	error = nfsrv_checkrestart(clientid, new_stp->ls_flags,
3461 	    &new_stp->ls_stateid, 0);
3462 	if (error)
3463 		goto out;
3464 
3465 	clidp = malloc(NFSV4_OPAQUELIMIT, M_TEMP, M_WAITOK);
3466 	NFSLOCKSTATE();
3467 	/*
3468 	 * Get the open structure via clientid and stateid.
3469 	 */
3470 	error = nfsrv_getclient(clientid, CLOPS_RENEW, &clp, NULL,
3471 	    (nfsquad_t)((u_quad_t)0), 0, nd, p);
3472 	if (!error)
3473 		error = nfsrv_getstate(clp, &new_stp->ls_stateid,
3474 		    new_stp->ls_flags, &stp);
3475 
3476 	/*
3477 	 * Sanity check the open.
3478 	 */
3479 	if (!error && (!(stp->ls_flags & NFSLCK_OPEN) ||
3480 		(!(new_stp->ls_flags & NFSLCK_CONFIRM) &&
3481 		 (stp->ls_openowner->ls_flags & NFSLCK_NEEDSCONFIRM)) ||
3482 		((new_stp->ls_flags & NFSLCK_CONFIRM) &&
3483 		 (!(stp->ls_openowner->ls_flags & NFSLCK_NEEDSCONFIRM)))))
3484 		error = NFSERR_BADSTATEID;
3485 
3486 	if (!error)
3487 		error = nfsrv_checkseqid(nd, new_stp->ls_seq,
3488 		    stp->ls_openowner, new_stp->ls_op);
3489 	if (!error && stp->ls_stateid.seqid != new_stp->ls_stateid.seqid &&
3490 	    (((nd->nd_flag & ND_NFSV41) == 0 &&
3491 	      !(new_stp->ls_flags & NFSLCK_CONFIRM)) ||
3492 	     ((nd->nd_flag & ND_NFSV41) != 0 &&
3493 	      new_stp->ls_stateid.seqid != 0)))
3494 		error = NFSERR_OLDSTATEID;
3495 	if (!error && vp->v_type != VREG) {
3496 		if (vp->v_type == VDIR)
3497 			error = NFSERR_ISDIR;
3498 		else
3499 			error = NFSERR_INVAL;
3500 	}
3501 
3502 	if (error) {
3503 		/*
3504 		 * If a client tries to confirm an Open with a bad
3505 		 * seqid# and there are no byte range locks or other Opens
3506 		 * on the openowner, just throw it away, so the next use of the
3507 		 * openowner will start a fresh seq#.
3508 		 */
3509 		if (error == NFSERR_BADSEQID &&
3510 		    (new_stp->ls_flags & NFSLCK_CONFIRM) &&
3511 		    nfsrv_nootherstate(stp))
3512 			nfsrv_freeopenowner(stp->ls_openowner, 0, p);
3513 		NFSUNLOCKSTATE();
3514 		goto out;
3515 	}
3516 
3517 	/*
3518 	 * Set the return stateid.
3519 	 */
3520 	stateidp->seqid = stp->ls_stateid.seqid + 1;
3521 	if ((nd->nd_flag & ND_NFSV41) != 0 && stateidp->seqid == 0)
3522 		stateidp->seqid = 1;
3523 	stateidp->other[0] = stp->ls_stateid.other[0];
3524 	stateidp->other[1] = stp->ls_stateid.other[1];
3525 	stateidp->other[2] = stp->ls_stateid.other[2];
3526 	/*
3527 	 * Now, handle the three cases.
3528 	 */
3529 	if (new_stp->ls_flags & NFSLCK_CONFIRM) {
3530 		/*
3531 		 * If the open doesn't need confirmation, it seems to me that
3532 		 * there is a client error, but I'll just log it and keep going?
3533 		 */
3534 		if (!(stp->ls_openowner->ls_flags & NFSLCK_NEEDSCONFIRM))
3535 			printf("Nfsv4d: stray open confirm\n");
3536 		stp->ls_openowner->ls_flags = 0;
3537 		stp->ls_stateid.seqid++;
3538 		if ((nd->nd_flag & ND_NFSV41) != 0 &&
3539 		    stp->ls_stateid.seqid == 0)
3540 			stp->ls_stateid.seqid = 1;
3541 		if (!(clp->lc_flags & LCL_STAMPEDSTABLE)) {
3542 			clp->lc_flags |= LCL_STAMPEDSTABLE;
3543 			len = clp->lc_idlen;
3544 			NFSBCOPY(clp->lc_id, clidp, len);
3545 			gotstate = 1;
3546 		}
3547 		NFSUNLOCKSTATE();
3548 	} else if (new_stp->ls_flags & NFSLCK_CLOSE) {
3549 		lfp = stp->ls_lfp;
3550 		if (retwriteaccessp != NULL) {
3551 			if ((stp->ls_flags & NFSLCK_WRITEACCESS) != 0)
3552 				*retwriteaccessp = 1;
3553 			else
3554 				*retwriteaccessp = 0;
3555 		}
3556 		if (nfsrv_dolocallocks != 0 && !LIST_EMPTY(&stp->ls_open)) {
3557 			/* Get the lf lock */
3558 			nfsrv_locklf(lfp);
3559 			NFSUNLOCKSTATE();
3560 			ASSERT_VOP_ELOCKED(vp, "nfsrv_openupdate");
3561 			NFSVOPUNLOCK(vp);
3562 			if (nfsrv_freeopen(stp, vp, 1, p) == 0) {
3563 				NFSLOCKSTATE();
3564 				nfsrv_unlocklf(lfp);
3565 				NFSUNLOCKSTATE();
3566 			}
3567 			NFSVOPLOCK(vp, LK_EXCLUSIVE | LK_RETRY);
3568 		} else {
3569 			(void) nfsrv_freeopen(stp, NULL, 0, p);
3570 			NFSUNLOCKSTATE();
3571 		}
3572 	} else {
3573 		/*
3574 		 * Update the share bits, making sure that the new set are a
3575 		 * subset of the old ones.
3576 		 */
3577 		bits = (new_stp->ls_flags & NFSLCK_SHAREBITS);
3578 		if (~(stp->ls_flags) & bits) {
3579 			NFSUNLOCKSTATE();
3580 			error = NFSERR_INVAL;
3581 			goto out;
3582 		}
3583 		stp->ls_flags = (bits | NFSLCK_OPEN);
3584 		stp->ls_stateid.seqid++;
3585 		if ((nd->nd_flag & ND_NFSV41) != 0 &&
3586 		    stp->ls_stateid.seqid == 0)
3587 			stp->ls_stateid.seqid = 1;
3588 		NFSUNLOCKSTATE();
3589 	}
3590 
3591 	/*
3592 	 * If the client just confirmed its first open, write a timestamp
3593 	 * to the stable storage file.
3594 	 */
3595 	if (gotstate != 0) {
3596 		nfsrv_writestable(clidp, len, NFSNST_NEWSTATE, p);
3597 		nfsrv_backupstable();
3598 	}
3599 
3600 out:
3601 	free(clidp, M_TEMP);
3602 	NFSEXITCODE2(error, nd);
3603 	return (error);
3604 }
3605 
3606 /*
3607  * Delegation update. Does the purge and return.
3608  */
3609 int
3610 nfsrv_delegupdate(struct nfsrv_descript *nd, nfsquad_t clientid,
3611     nfsv4stateid_t *stateidp, vnode_t vp, int op, struct ucred *cred,
3612     NFSPROC_T *p, int *retwriteaccessp)
3613 {
3614 	struct nfsstate *stp;
3615 	struct nfsclient *clp;
3616 	int error = 0;
3617 	fhandle_t fh;
3618 
3619 	/*
3620 	 * Do a sanity check against the file handle for DelegReturn.
3621 	 */
3622 	if (vp) {
3623 		error = nfsvno_getfh(vp, &fh, p);
3624 		if (error)
3625 			goto out;
3626 	}
3627 	/*
3628 	 * Check for restart conditions (client and server).
3629 	 */
3630 	if (op == NFSV4OP_DELEGRETURN)
3631 		error = nfsrv_checkrestart(clientid, NFSLCK_DELEGRETURN,
3632 			stateidp, 0);
3633 	else
3634 		error = nfsrv_checkrestart(clientid, NFSLCK_DELEGPURGE,
3635 			stateidp, 0);
3636 
3637 	NFSLOCKSTATE();
3638 	/*
3639 	 * Get the open structure via clientid and stateid.
3640 	 */
3641 	if (!error)
3642 	    error = nfsrv_getclient(clientid, CLOPS_RENEW, &clp, NULL,
3643 		(nfsquad_t)((u_quad_t)0), 0, nd, p);
3644 	if (error) {
3645 		if (error == NFSERR_CBPATHDOWN)
3646 			error = 0;
3647 		if (error == NFSERR_STALECLIENTID && op == NFSV4OP_DELEGRETURN)
3648 			error = NFSERR_STALESTATEID;
3649 	}
3650 	if (!error && op == NFSV4OP_DELEGRETURN) {
3651 	    error = nfsrv_getstate(clp, stateidp, NFSLCK_DELEGRETURN, &stp);
3652 	    if (!error && stp->ls_stateid.seqid != stateidp->seqid &&
3653 		((nd->nd_flag & ND_NFSV41) == 0 || stateidp->seqid != 0))
3654 		error = NFSERR_OLDSTATEID;
3655 	}
3656 	/*
3657 	 * NFSERR_EXPIRED means that the state has gone away,
3658 	 * so Delegations have been purged. Just return ok.
3659 	 */
3660 	if (error == NFSERR_EXPIRED && op == NFSV4OP_DELEGPURGE) {
3661 		NFSUNLOCKSTATE();
3662 		error = 0;
3663 		goto out;
3664 	}
3665 	if (error) {
3666 		NFSUNLOCKSTATE();
3667 		goto out;
3668 	}
3669 
3670 	if (op == NFSV4OP_DELEGRETURN) {
3671 		if (NFSBCMP((caddr_t)&fh, (caddr_t)&stp->ls_lfp->lf_fh,
3672 		    sizeof (fhandle_t))) {
3673 			NFSUNLOCKSTATE();
3674 			error = NFSERR_BADSTATEID;
3675 			goto out;
3676 		}
3677 		if (retwriteaccessp != NULL) {
3678 			if ((stp->ls_flags & NFSLCK_DELEGWRITE) != 0)
3679 				*retwriteaccessp = 1;
3680 			else
3681 				*retwriteaccessp = 0;
3682 		}
3683 		nfsrv_freedeleg(stp);
3684 	} else {
3685 		nfsrv_freedeleglist(&clp->lc_olddeleg);
3686 	}
3687 	NFSUNLOCKSTATE();
3688 	error = 0;
3689 
3690 out:
3691 	NFSEXITCODE(error);
3692 	return (error);
3693 }
3694 
3695 /*
3696  * Release lock owner.
3697  */
3698 int
3699 nfsrv_releaselckown(struct nfsstate *new_stp, nfsquad_t clientid,
3700     NFSPROC_T *p)
3701 {
3702 	struct nfsstate *stp, *nstp, *openstp, *ownstp;
3703 	struct nfsclient *clp;
3704 	int error = 0;
3705 
3706 	/*
3707 	 * Check for restart conditions (client and server).
3708 	 */
3709 	error = nfsrv_checkrestart(clientid, new_stp->ls_flags,
3710 	    &new_stp->ls_stateid, 0);
3711 	if (error)
3712 		goto out;
3713 
3714 	NFSLOCKSTATE();
3715 	/*
3716 	 * Get the lock owner by name.
3717 	 */
3718 	error = nfsrv_getclient(clientid, CLOPS_RENEW, &clp, NULL,
3719 	    (nfsquad_t)((u_quad_t)0), 0, NULL, p);
3720 	if (error) {
3721 		NFSUNLOCKSTATE();
3722 		goto out;
3723 	}
3724 	LIST_FOREACH(ownstp, &clp->lc_open, ls_list) {
3725 	    LIST_FOREACH(openstp, &ownstp->ls_open, ls_list) {
3726 		stp = LIST_FIRST(&openstp->ls_open);
3727 		while (stp != LIST_END(&openstp->ls_open)) {
3728 		    nstp = LIST_NEXT(stp, ls_list);
3729 		    /*
3730 		     * If the owner matches, check for locks and
3731 		     * then free or return an error.
3732 		     */
3733 		    if (stp->ls_ownerlen == new_stp->ls_ownerlen &&
3734 			!NFSBCMP(stp->ls_owner, new_stp->ls_owner,
3735 			 stp->ls_ownerlen)){
3736 			if (LIST_EMPTY(&stp->ls_lock)) {
3737 			    nfsrv_freelockowner(stp, NULL, 0, p);
3738 			} else {
3739 			    NFSUNLOCKSTATE();
3740 			    error = NFSERR_LOCKSHELD;
3741 			    goto out;
3742 			}
3743 		    }
3744 		    stp = nstp;
3745 		}
3746 	    }
3747 	}
3748 	NFSUNLOCKSTATE();
3749 
3750 out:
3751 	NFSEXITCODE(error);
3752 	return (error);
3753 }
3754 
3755 /*
3756  * Get the file handle for a lock structure.
3757  */
3758 static int
3759 nfsrv_getlockfh(vnode_t vp, u_short flags, struct nfslockfile *new_lfp,
3760     fhandle_t *nfhp, NFSPROC_T *p)
3761 {
3762 	fhandle_t *fhp = NULL;
3763 	int error;
3764 
3765 	/*
3766 	 * For lock, use the new nfslock structure, otherwise just
3767 	 * a fhandle_t on the stack.
3768 	 */
3769 	if (flags & NFSLCK_OPEN) {
3770 		KASSERT(new_lfp != NULL, ("nfsrv_getlockfh: new_lfp NULL"));
3771 		fhp = &new_lfp->lf_fh;
3772 	} else if (nfhp) {
3773 		fhp = nfhp;
3774 	} else {
3775 		panic("nfsrv_getlockfh");
3776 	}
3777 	error = nfsvno_getfh(vp, fhp, p);
3778 	NFSEXITCODE(error);
3779 	return (error);
3780 }
3781 
3782 /*
3783  * Get an nfs lock structure. Allocate one, as required, and return a
3784  * pointer to it.
3785  * Returns an NFSERR_xxx upon failure or -1 to indicate no current lock.
3786  */
3787 static int
3788 nfsrv_getlockfile(u_short flags, struct nfslockfile **new_lfpp,
3789     struct nfslockfile **lfpp, fhandle_t *nfhp, int lockit)
3790 {
3791 	struct nfslockfile *lfp;
3792 	fhandle_t *fhp = NULL, *tfhp;
3793 	struct nfslockhashhead *hp;
3794 	struct nfslockfile *new_lfp = NULL;
3795 
3796 	/*
3797 	 * For lock, use the new nfslock structure, otherwise just
3798 	 * a fhandle_t on the stack.
3799 	 */
3800 	if (flags & NFSLCK_OPEN) {
3801 		new_lfp = *new_lfpp;
3802 		fhp = &new_lfp->lf_fh;
3803 	} else if (nfhp) {
3804 		fhp = nfhp;
3805 	} else {
3806 		panic("nfsrv_getlockfile");
3807 	}
3808 
3809 	hp = NFSLOCKHASH(fhp);
3810 	LIST_FOREACH(lfp, hp, lf_hash) {
3811 		tfhp = &lfp->lf_fh;
3812 		if (NFSVNO_CMPFH(fhp, tfhp)) {
3813 			if (lockit)
3814 				nfsrv_locklf(lfp);
3815 			*lfpp = lfp;
3816 			return (0);
3817 		}
3818 	}
3819 	if (!(flags & NFSLCK_OPEN))
3820 		return (-1);
3821 
3822 	/*
3823 	 * No match, so chain the new one into the list.
3824 	 */
3825 	LIST_INIT(&new_lfp->lf_open);
3826 	LIST_INIT(&new_lfp->lf_lock);
3827 	LIST_INIT(&new_lfp->lf_deleg);
3828 	LIST_INIT(&new_lfp->lf_locallock);
3829 	LIST_INIT(&new_lfp->lf_rollback);
3830 	new_lfp->lf_locallock_lck.nfslock_usecnt = 0;
3831 	new_lfp->lf_locallock_lck.nfslock_lock = 0;
3832 	new_lfp->lf_usecount = 0;
3833 	LIST_INSERT_HEAD(hp, new_lfp, lf_hash);
3834 	*lfpp = new_lfp;
3835 	*new_lfpp = NULL;
3836 	return (0);
3837 }
3838 
3839 /*
3840  * This function adds a nfslock lock structure to the list for the associated
3841  * nfsstate and nfslockfile structures. It will be inserted after the
3842  * entry pointed at by insert_lop.
3843  */
3844 static void
3845 nfsrv_insertlock(struct nfslock *new_lop, struct nfslock *insert_lop,
3846     struct nfsstate *stp, struct nfslockfile *lfp)
3847 {
3848 	struct nfslock *lop, *nlop;
3849 
3850 	new_lop->lo_stp = stp;
3851 	new_lop->lo_lfp = lfp;
3852 
3853 	if (stp != NULL) {
3854 		/* Insert in increasing lo_first order */
3855 		lop = LIST_FIRST(&lfp->lf_lock);
3856 		if (lop == LIST_END(&lfp->lf_lock) ||
3857 		    new_lop->lo_first <= lop->lo_first) {
3858 			LIST_INSERT_HEAD(&lfp->lf_lock, new_lop, lo_lckfile);
3859 		} else {
3860 			nlop = LIST_NEXT(lop, lo_lckfile);
3861 			while (nlop != LIST_END(&lfp->lf_lock) &&
3862 			       nlop->lo_first < new_lop->lo_first) {
3863 				lop = nlop;
3864 				nlop = LIST_NEXT(lop, lo_lckfile);
3865 			}
3866 			LIST_INSERT_AFTER(lop, new_lop, lo_lckfile);
3867 		}
3868 	} else {
3869 		new_lop->lo_lckfile.le_prev = NULL;	/* list not used */
3870 	}
3871 
3872 	/*
3873 	 * Insert after insert_lop, which is overloaded as stp or lfp for
3874 	 * an empty list.
3875 	 */
3876 	if (stp == NULL && (struct nfslockfile *)insert_lop == lfp)
3877 		LIST_INSERT_HEAD(&lfp->lf_locallock, new_lop, lo_lckowner);
3878 	else if ((struct nfsstate *)insert_lop == stp)
3879 		LIST_INSERT_HEAD(&stp->ls_lock, new_lop, lo_lckowner);
3880 	else
3881 		LIST_INSERT_AFTER(insert_lop, new_lop, lo_lckowner);
3882 	if (stp != NULL) {
3883 		nfsstatsv1.srvlocks++;
3884 		nfsrv_openpluslock++;
3885 	}
3886 }
3887 
3888 /*
3889  * This function updates the locking for a lock owner and given file. It
3890  * maintains a list of lock ranges ordered on increasing file offset that
3891  * are NFSLCK_READ or NFSLCK_WRITE and non-overlapping (aka POSIX style).
3892  * It always adds new_lop to the list and sometimes uses the one pointed
3893  * at by other_lopp.
3894  */
3895 static void
3896 nfsrv_updatelock(struct nfsstate *stp, struct nfslock **new_lopp,
3897     struct nfslock **other_lopp, struct nfslockfile *lfp)
3898 {
3899 	struct nfslock *new_lop = *new_lopp;
3900 	struct nfslock *lop, *tlop, *ilop;
3901 	struct nfslock *other_lop = *other_lopp;
3902 	int unlock = 0, myfile = 0;
3903 	u_int64_t tmp;
3904 
3905 	/*
3906 	 * Work down the list until the lock is merged.
3907 	 */
3908 	if (new_lop->lo_flags & NFSLCK_UNLOCK)
3909 		unlock = 1;
3910 	if (stp != NULL) {
3911 		ilop = (struct nfslock *)stp;
3912 		lop = LIST_FIRST(&stp->ls_lock);
3913 	} else {
3914 		ilop = (struct nfslock *)lfp;
3915 		lop = LIST_FIRST(&lfp->lf_locallock);
3916 	}
3917 	while (lop != NULL) {
3918 	    /*
3919 	     * Only check locks for this file that aren't before the start of
3920 	     * new lock's range.
3921 	     */
3922 	    if (lop->lo_lfp == lfp) {
3923 	      myfile = 1;
3924 	      if (lop->lo_end >= new_lop->lo_first) {
3925 		if (new_lop->lo_end < lop->lo_first) {
3926 			/*
3927 			 * If the new lock ends before the start of the
3928 			 * current lock's range, no merge, just insert
3929 			 * the new lock.
3930 			 */
3931 			break;
3932 		}
3933 		if (new_lop->lo_flags == lop->lo_flags ||
3934 		    (new_lop->lo_first <= lop->lo_first &&
3935 		     new_lop->lo_end >= lop->lo_end)) {
3936 			/*
3937 			 * This lock can be absorbed by the new lock/unlock.
3938 			 * This happens when it covers the entire range
3939 			 * of the old lock or is contiguous
3940 			 * with the old lock and is of the same type or an
3941 			 * unlock.
3942 			 */
3943 			if (lop->lo_first < new_lop->lo_first)
3944 				new_lop->lo_first = lop->lo_first;
3945 			if (lop->lo_end > new_lop->lo_end)
3946 				new_lop->lo_end = lop->lo_end;
3947 			tlop = lop;
3948 			lop = LIST_NEXT(lop, lo_lckowner);
3949 			nfsrv_freenfslock(tlop);
3950 			continue;
3951 		}
3952 
3953 		/*
3954 		 * All these cases are for contiguous locks that are not the
3955 		 * same type, so they can't be merged.
3956 		 */
3957 		if (new_lop->lo_first <= lop->lo_first) {
3958 			/*
3959 			 * This case is where the new lock overlaps with the
3960 			 * first part of the old lock. Move the start of the
3961 			 * old lock to just past the end of the new lock. The
3962 			 * new lock will be inserted in front of the old, since
3963 			 * ilop hasn't been updated. (We are done now.)
3964 			 */
3965 			lop->lo_first = new_lop->lo_end;
3966 			break;
3967 		}
3968 		if (new_lop->lo_end >= lop->lo_end) {
3969 			/*
3970 			 * This case is where the new lock overlaps with the
3971 			 * end of the old lock's range. Move the old lock's
3972 			 * end to just before the new lock's first and insert
3973 			 * the new lock after the old lock.
3974 			 * Might not be done yet, since the new lock could
3975 			 * overlap further locks with higher ranges.
3976 			 */
3977 			lop->lo_end = new_lop->lo_first;
3978 			ilop = lop;
3979 			lop = LIST_NEXT(lop, lo_lckowner);
3980 			continue;
3981 		}
3982 		/*
3983 		 * The final case is where the new lock's range is in the
3984 		 * middle of the current lock's and splits the current lock
3985 		 * up. Use *other_lopp to handle the second part of the
3986 		 * split old lock range. (We are done now.)
3987 		 * For unlock, we use new_lop as other_lop and tmp, since
3988 		 * other_lop and new_lop are the same for this case.
3989 		 * We noted the unlock case above, so we don't need
3990 		 * new_lop->lo_flags any longer.
3991 		 */
3992 		tmp = new_lop->lo_first;
3993 		if (other_lop == NULL) {
3994 			if (!unlock)
3995 				panic("nfsd srv update unlock");
3996 			other_lop = new_lop;
3997 			*new_lopp = NULL;
3998 		}
3999 		other_lop->lo_first = new_lop->lo_end;
4000 		other_lop->lo_end = lop->lo_end;
4001 		other_lop->lo_flags = lop->lo_flags;
4002 		other_lop->lo_stp = stp;
4003 		other_lop->lo_lfp = lfp;
4004 		lop->lo_end = tmp;
4005 		nfsrv_insertlock(other_lop, lop, stp, lfp);
4006 		*other_lopp = NULL;
4007 		ilop = lop;
4008 		break;
4009 	      }
4010 	    }
4011 	    ilop = lop;
4012 	    lop = LIST_NEXT(lop, lo_lckowner);
4013 	    if (myfile && (lop == NULL || lop->lo_lfp != lfp))
4014 		break;
4015 	}
4016 
4017 	/*
4018 	 * Insert the new lock in the list at the appropriate place.
4019 	 */
4020 	if (!unlock) {
4021 		nfsrv_insertlock(new_lop, ilop, stp, lfp);
4022 		*new_lopp = NULL;
4023 	}
4024 }
4025 
4026 /*
4027  * This function handles sequencing of locks, etc.
4028  * It returns an error that indicates what the caller should do.
4029  */
4030 static int
4031 nfsrv_checkseqid(struct nfsrv_descript *nd, u_int32_t seqid,
4032     struct nfsstate *stp, struct nfsrvcache *op)
4033 {
4034 	int error = 0;
4035 
4036 	if ((nd->nd_flag & ND_NFSV41) != 0)
4037 		/* NFSv4.1 ignores the open_seqid and lock_seqid. */
4038 		goto out;
4039 	if (op != nd->nd_rp)
4040 		panic("nfsrvstate checkseqid");
4041 	if (!(op->rc_flag & RC_INPROG))
4042 		panic("nfsrvstate not inprog");
4043 	if (stp->ls_op && stp->ls_op->rc_refcnt <= 0) {
4044 		printf("refcnt=%d\n", stp->ls_op->rc_refcnt);
4045 		panic("nfsrvstate op refcnt");
4046 	}
4047 
4048 	/* If ND_ERELOOKUP is set, the seqid has already been handled. */
4049 	if ((nd->nd_flag & ND_ERELOOKUP) != 0)
4050 		goto out;
4051 
4052 	if ((stp->ls_seq + 1) == seqid) {
4053 		if (stp->ls_op)
4054 			nfsrvd_derefcache(stp->ls_op);
4055 		stp->ls_op = op;
4056 		nfsrvd_refcache(op);
4057 		stp->ls_seq = seqid;
4058 		goto out;
4059 	} else if (stp->ls_seq == seqid && stp->ls_op &&
4060 		op->rc_xid == stp->ls_op->rc_xid &&
4061 		op->rc_refcnt == 0 &&
4062 		op->rc_reqlen == stp->ls_op->rc_reqlen &&
4063 		op->rc_cksum == stp->ls_op->rc_cksum) {
4064 		if (stp->ls_op->rc_flag & RC_INPROG) {
4065 			error = NFSERR_DONTREPLY;
4066 			goto out;
4067 		}
4068 		nd->nd_rp = stp->ls_op;
4069 		nd->nd_rp->rc_flag |= RC_INPROG;
4070 		nfsrvd_delcache(op);
4071 		error = NFSERR_REPLYFROMCACHE;
4072 		goto out;
4073 	}
4074 	error = NFSERR_BADSEQID;
4075 
4076 out:
4077 	NFSEXITCODE2(error, nd);
4078 	return (error);
4079 }
4080 
4081 /*
4082  * Get the client ip address for callbacks. If the strings can't be parsed,
4083  * just set lc_program to 0 to indicate no callbacks are possible.
4084  * (For cases where the address can't be parsed or is 0.0.0.0.0.0, set
4085  *  the address to the client's transport address. This won't be used
4086  *  for callbacks, but can be printed out by nfsstats for info.)
4087  * Return error if the xdr can't be parsed, 0 otherwise.
4088  */
4089 int
4090 nfsrv_getclientipaddr(struct nfsrv_descript *nd, struct nfsclient *clp)
4091 {
4092 	u_int32_t *tl;
4093 	u_char *cp, *cp2;
4094 	int i, j, maxalen = 0, minalen = 0;
4095 	sa_family_t af;
4096 #ifdef INET
4097 	struct sockaddr_in *rin = NULL, *sin;
4098 #endif
4099 #ifdef INET6
4100 	struct sockaddr_in6 *rin6 = NULL, *sin6;
4101 #endif
4102 	u_char *addr;
4103 	int error = 0, cantparse = 0;
4104 	union {
4105 		in_addr_t ival;
4106 		u_char cval[4];
4107 	} ip;
4108 	union {
4109 		in_port_t sval;
4110 		u_char cval[2];
4111 	} port;
4112 
4113 	/* 8 is the maximum length of the port# string. */
4114 	addr = malloc(INET6_ADDRSTRLEN + 8, M_TEMP, M_WAITOK);
4115 	clp->lc_req.nr_client = NULL;
4116 	clp->lc_req.nr_lock = 0;
4117 	af = AF_UNSPEC;
4118 	NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
4119 	i = fxdr_unsigned(int, *tl);
4120 	if (i >= 3 && i <= 4) {
4121 		error = nfsrv_mtostr(nd, addr, i);
4122 		if (error)
4123 			goto nfsmout;
4124 #ifdef INET
4125 		if (!strcmp(addr, "tcp")) {
4126 			clp->lc_flags |= LCL_TCPCALLBACK;
4127 			clp->lc_req.nr_sotype = SOCK_STREAM;
4128 			clp->lc_req.nr_soproto = IPPROTO_TCP;
4129 			af = AF_INET;
4130 		} else if (!strcmp(addr, "udp")) {
4131 			clp->lc_req.nr_sotype = SOCK_DGRAM;
4132 			clp->lc_req.nr_soproto = IPPROTO_UDP;
4133 			af = AF_INET;
4134 		}
4135 #endif
4136 #ifdef INET6
4137 		if (af == AF_UNSPEC) {
4138 			if (!strcmp(addr, "tcp6")) {
4139 				clp->lc_flags |= LCL_TCPCALLBACK;
4140 				clp->lc_req.nr_sotype = SOCK_STREAM;
4141 				clp->lc_req.nr_soproto = IPPROTO_TCP;
4142 				af = AF_INET6;
4143 			} else if (!strcmp(addr, "udp6")) {
4144 				clp->lc_req.nr_sotype = SOCK_DGRAM;
4145 				clp->lc_req.nr_soproto = IPPROTO_UDP;
4146 				af = AF_INET6;
4147 			}
4148 		}
4149 #endif
4150 		if (af == AF_UNSPEC) {
4151 			cantparse = 1;
4152 		}
4153 	} else {
4154 		cantparse = 1;
4155 		if (i > 0) {
4156 			error = nfsm_advance(nd, NFSM_RNDUP(i), -1);
4157 			if (error)
4158 				goto nfsmout;
4159 		}
4160 	}
4161 	/*
4162 	 * The caller has allocated clp->lc_req.nr_nam to be large enough
4163 	 * for either AF_INET or AF_INET6 and zeroed out the contents.
4164 	 * maxalen is set to the maximum length of the host IP address string
4165 	 * plus 8 for the maximum length of the port#.
4166 	 * minalen is set to the minimum length of the host IP address string
4167 	 * plus 4 for the minimum length of the port#.
4168 	 * These lengths do not include NULL termination,
4169 	 * so INET[6]_ADDRSTRLEN - 1 is used in the calculations.
4170 	 */
4171 	switch (af) {
4172 #ifdef INET
4173 	case AF_INET:
4174 		rin = (struct sockaddr_in *)clp->lc_req.nr_nam;
4175 		rin->sin_family = AF_INET;
4176 		rin->sin_len = sizeof(struct sockaddr_in);
4177 		maxalen = INET_ADDRSTRLEN - 1 + 8;
4178 		minalen = 7 + 4;
4179 		break;
4180 #endif
4181 #ifdef INET6
4182 	case AF_INET6:
4183 		rin6 = (struct sockaddr_in6 *)clp->lc_req.nr_nam;
4184 		rin6->sin6_family = AF_INET6;
4185 		rin6->sin6_len = sizeof(struct sockaddr_in6);
4186 		maxalen = INET6_ADDRSTRLEN - 1 + 8;
4187 		minalen = 3 + 4;
4188 		break;
4189 #endif
4190 	}
4191 	NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
4192 	i = fxdr_unsigned(int, *tl);
4193 	if (i < 0) {
4194 		error = NFSERR_BADXDR;
4195 		goto nfsmout;
4196 	} else if (i == 0) {
4197 		cantparse = 1;
4198 	} else if (!cantparse && i <= maxalen && i >= minalen) {
4199 		error = nfsrv_mtostr(nd, addr, i);
4200 		if (error)
4201 			goto nfsmout;
4202 
4203 		/*
4204 		 * Parse out the address fields. We expect 6 decimal numbers
4205 		 * separated by '.'s for AF_INET and two decimal numbers
4206 		 * preceeded by '.'s for AF_INET6.
4207 		 */
4208 		cp = NULL;
4209 		switch (af) {
4210 #ifdef INET6
4211 		/*
4212 		 * For AF_INET6, first parse the host address.
4213 		 */
4214 		case AF_INET6:
4215 			cp = strchr(addr, '.');
4216 			if (cp != NULL) {
4217 				*cp++ = '\0';
4218 				if (inet_pton(af, addr, &rin6->sin6_addr) == 1)
4219 					i = 4;
4220 				else {
4221 					cp = NULL;
4222 					cantparse = 1;
4223 				}
4224 			}
4225 			break;
4226 #endif
4227 #ifdef INET
4228 		case AF_INET:
4229 			cp = addr;
4230 			i = 0;
4231 			break;
4232 #endif
4233 		}
4234 		while (cp != NULL && *cp && i < 6) {
4235 			cp2 = cp;
4236 			while (*cp2 && *cp2 != '.')
4237 				cp2++;
4238 			if (*cp2)
4239 				*cp2++ = '\0';
4240 			else if (i != 5) {
4241 				cantparse = 1;
4242 				break;
4243 			}
4244 			j = nfsrv_getipnumber(cp);
4245 			if (j >= 0) {
4246 				if (i < 4)
4247 					ip.cval[3 - i] = j;
4248 				else
4249 					port.cval[5 - i] = j;
4250 			} else {
4251 				cantparse = 1;
4252 				break;
4253 			}
4254 			cp = cp2;
4255 			i++;
4256 		}
4257 		if (!cantparse) {
4258 			/*
4259 			 * The host address INADDR_ANY is (mis)used to indicate
4260 			 * "there is no valid callback address".
4261 			 */
4262 			switch (af) {
4263 #ifdef INET6
4264 			case AF_INET6:
4265 				if (!IN6_ARE_ADDR_EQUAL(&rin6->sin6_addr,
4266 				    &in6addr_any))
4267 					rin6->sin6_port = htons(port.sval);
4268 				else
4269 					cantparse = 1;
4270 				break;
4271 #endif
4272 #ifdef INET
4273 			case AF_INET:
4274 				if (ip.ival != INADDR_ANY) {
4275 					rin->sin_addr.s_addr = htonl(ip.ival);
4276 					rin->sin_port = htons(port.sval);
4277 				} else {
4278 					cantparse = 1;
4279 				}
4280 				break;
4281 #endif
4282 			}
4283 		}
4284 	} else {
4285 		cantparse = 1;
4286 		if (i > 0) {
4287 			error = nfsm_advance(nd, NFSM_RNDUP(i), -1);
4288 			if (error)
4289 				goto nfsmout;
4290 		}
4291 	}
4292 	if (cantparse) {
4293 		switch (nd->nd_nam->sa_family) {
4294 #ifdef INET
4295 		case AF_INET:
4296 			sin = (struct sockaddr_in *)nd->nd_nam;
4297 			rin = (struct sockaddr_in *)clp->lc_req.nr_nam;
4298 			rin->sin_family = AF_INET;
4299 			rin->sin_len = sizeof(struct sockaddr_in);
4300 			rin->sin_addr.s_addr = sin->sin_addr.s_addr;
4301 			rin->sin_port = 0x0;
4302 			break;
4303 #endif
4304 #ifdef INET6
4305 		case AF_INET6:
4306 			sin6 = (struct sockaddr_in6 *)nd->nd_nam;
4307 			rin6 = (struct sockaddr_in6 *)clp->lc_req.nr_nam;
4308 			rin6->sin6_family = AF_INET6;
4309 			rin6->sin6_len = sizeof(struct sockaddr_in6);
4310 			rin6->sin6_addr = sin6->sin6_addr;
4311 			rin6->sin6_port = 0x0;
4312 			break;
4313 #endif
4314 		}
4315 		clp->lc_program = 0;
4316 	}
4317 nfsmout:
4318 	free(addr, M_TEMP);
4319 	NFSEXITCODE2(error, nd);
4320 	return (error);
4321 }
4322 
4323 /*
4324  * Turn a string of up to three decimal digits into a number. Return -1 upon
4325  * error.
4326  */
4327 static int
4328 nfsrv_getipnumber(u_char *cp)
4329 {
4330 	int i = 0, j = 0;
4331 
4332 	while (*cp) {
4333 		if (j > 2 || *cp < '0' || *cp > '9')
4334 			return (-1);
4335 		i *= 10;
4336 		i += (*cp - '0');
4337 		cp++;
4338 		j++;
4339 	}
4340 	if (i < 256)
4341 		return (i);
4342 	return (-1);
4343 }
4344 
4345 /*
4346  * This function checks for restart conditions.
4347  */
4348 static int
4349 nfsrv_checkrestart(nfsquad_t clientid, u_int32_t flags,
4350     nfsv4stateid_t *stateidp, int specialid)
4351 {
4352 	int ret = 0;
4353 
4354 	/*
4355 	 * First check for a server restart. Open, LockT, ReleaseLockOwner
4356 	 * and DelegPurge have a clientid, the rest a stateid.
4357 	 */
4358 	if (flags &
4359 	    (NFSLCK_OPEN | NFSLCK_TEST | NFSLCK_RELEASE | NFSLCK_DELEGPURGE)) {
4360 		if (clientid.lval[0] != nfsrvboottime) {
4361 			ret = NFSERR_STALECLIENTID;
4362 			goto out;
4363 		}
4364 	} else if (stateidp->other[0] != nfsrvboottime &&
4365 		specialid == 0) {
4366 		ret = NFSERR_STALESTATEID;
4367 		goto out;
4368 	}
4369 
4370 	/*
4371 	 * Read, Write, Setattr and LockT can return NFSERR_GRACE and do
4372 	 * not use a lock/open owner seqid#, so the check can be done now.
4373 	 * (The others will be checked, as required, later.)
4374 	 */
4375 	if (!(flags & (NFSLCK_CHECK | NFSLCK_TEST)))
4376 		goto out;
4377 
4378 	NFSLOCKSTATE();
4379 	ret = nfsrv_checkgrace(NULL, NULL, flags);
4380 	NFSUNLOCKSTATE();
4381 
4382 out:
4383 	NFSEXITCODE(ret);
4384 	return (ret);
4385 }
4386 
4387 /*
4388  * Check for grace.
4389  */
4390 static int
4391 nfsrv_checkgrace(struct nfsrv_descript *nd, struct nfsclient *clp,
4392     u_int32_t flags)
4393 {
4394 	int error = 0, notreclaimed;
4395 	struct nfsrv_stable *sp;
4396 
4397 	if ((nfsrv_stablefirst.nsf_flags & (NFSNSF_UPDATEDONE |
4398 	     NFSNSF_GRACEOVER)) == 0) {
4399 		/*
4400 		 * First, check to see if all of the clients have done a
4401 		 * ReclaimComplete.  If so, grace can end now.
4402 		 */
4403 		notreclaimed = 0;
4404 		LIST_FOREACH(sp, &nfsrv_stablefirst.nsf_head, nst_list) {
4405 			if ((sp->nst_flag & NFSNST_RECLAIMED) == 0) {
4406 				notreclaimed = 1;
4407 				break;
4408 			}
4409 		}
4410 		if (notreclaimed == 0)
4411 			nfsrv_stablefirst.nsf_flags |= (NFSNSF_GRACEOVER |
4412 			    NFSNSF_NEEDLOCK);
4413 	}
4414 
4415 	if ((nfsrv_stablefirst.nsf_flags & NFSNSF_GRACEOVER) != 0) {
4416 		if (flags & NFSLCK_RECLAIM) {
4417 			error = NFSERR_NOGRACE;
4418 			goto out;
4419 		}
4420 	} else {
4421 		if (!(flags & NFSLCK_RECLAIM)) {
4422 			error = NFSERR_GRACE;
4423 			goto out;
4424 		}
4425 		if (nd != NULL && clp != NULL &&
4426 		    (nd->nd_flag & ND_NFSV41) != 0 &&
4427 		    (clp->lc_flags & LCL_RECLAIMCOMPLETE) != 0) {
4428 			error = NFSERR_NOGRACE;
4429 			goto out;
4430 		}
4431 
4432 		/*
4433 		 * If grace is almost over and we are still getting Reclaims,
4434 		 * extend grace a bit.
4435 		 */
4436 		if ((NFSD_MONOSEC + NFSRV_LEASEDELTA) >
4437 		    nfsrv_stablefirst.nsf_eograce)
4438 			nfsrv_stablefirst.nsf_eograce = NFSD_MONOSEC +
4439 				NFSRV_LEASEDELTA;
4440 	}
4441 
4442 out:
4443 	NFSEXITCODE(error);
4444 	return (error);
4445 }
4446 
4447 /*
4448  * Do a server callback.
4449  * The "trunc" argument is slightly overloaded and refers to different
4450  * boolean arguments for CBRECALL and CBLAYOUTRECALL.
4451  */
4452 static int
4453 nfsrv_docallback(struct nfsclient *clp, int procnum, nfsv4stateid_t *stateidp,
4454     int trunc, fhandle_t *fhp, struct nfsvattr *nap, nfsattrbit_t *attrbitp,
4455     int laytype, NFSPROC_T *p)
4456 {
4457 	struct mbuf *m;
4458 	u_int32_t *tl;
4459 	struct nfsrv_descript *nd;
4460 	struct ucred *cred;
4461 	int error = 0, slotpos;
4462 	u_int32_t callback;
4463 	struct nfsdsession *sep = NULL;
4464 	uint64_t tval;
4465 	bool dotls;
4466 
4467 	nd = malloc(sizeof(*nd), M_TEMP, M_WAITOK | M_ZERO);
4468 	cred = newnfs_getcred();
4469 	NFSLOCKSTATE();	/* mostly for lc_cbref++ */
4470 	if (clp->lc_flags & LCL_NEEDSCONFIRM) {
4471 		NFSUNLOCKSTATE();
4472 		panic("docallb");
4473 	}
4474 	clp->lc_cbref++;
4475 
4476 	/*
4477 	 * Fill the callback program# and version into the request
4478 	 * structure for newnfs_connect() to use.
4479 	 */
4480 	clp->lc_req.nr_prog = clp->lc_program;
4481 #ifdef notnow
4482 	if ((clp->lc_flags & LCL_NFSV41) != 0)
4483 		clp->lc_req.nr_vers = NFSV41_CBVERS;
4484 	else
4485 #endif
4486 		clp->lc_req.nr_vers = NFSV4_CBVERS;
4487 
4488 	/*
4489 	 * First, fill in some of the fields of nd and cr.
4490 	 */
4491 	nd->nd_flag = ND_NFSV4;
4492 	if (clp->lc_flags & LCL_GSS)
4493 		nd->nd_flag |= ND_KERBV;
4494 	if ((clp->lc_flags & LCL_NFSV41) != 0)
4495 		nd->nd_flag |= ND_NFSV41;
4496 	if ((clp->lc_flags & LCL_NFSV42) != 0)
4497 		nd->nd_flag |= ND_NFSV42;
4498 	nd->nd_repstat = 0;
4499 	cred->cr_uid = clp->lc_uid;
4500 	cred->cr_gid = clp->lc_gid;
4501 	callback = clp->lc_callback;
4502 	NFSUNLOCKSTATE();
4503 	cred->cr_ngroups = 1;
4504 
4505 	/*
4506 	 * Get the first mbuf for the request.
4507 	 */
4508 	MGET(m, M_WAITOK, MT_DATA);
4509 	m->m_len = 0;
4510 	nd->nd_mreq = nd->nd_mb = m;
4511 	nd->nd_bpos = mtod(m, caddr_t);
4512 
4513 	/*
4514 	 * and build the callback request.
4515 	 */
4516 	if (procnum == NFSV4OP_CBGETATTR) {
4517 		nd->nd_procnum = NFSV4PROC_CBCOMPOUND;
4518 		error = nfsrv_cbcallargs(nd, clp, callback, NFSV4OP_CBGETATTR,
4519 		    "CB Getattr", &sep, &slotpos);
4520 		if (error != 0) {
4521 			m_freem(nd->nd_mreq);
4522 			goto errout;
4523 		}
4524 		(void)nfsm_fhtom(nd, (u_int8_t *)fhp, NFSX_MYFH, 0);
4525 		(void)nfsrv_putattrbit(nd, attrbitp);
4526 	} else if (procnum == NFSV4OP_CBRECALL) {
4527 		nd->nd_procnum = NFSV4PROC_CBCOMPOUND;
4528 		error = nfsrv_cbcallargs(nd, clp, callback, NFSV4OP_CBRECALL,
4529 		    "CB Recall", &sep, &slotpos);
4530 		if (error != 0) {
4531 			m_freem(nd->nd_mreq);
4532 			goto errout;
4533 		}
4534 		NFSM_BUILD(tl, u_int32_t *, NFSX_UNSIGNED + NFSX_STATEID);
4535 		*tl++ = txdr_unsigned(stateidp->seqid);
4536 		NFSBCOPY((caddr_t)stateidp->other, (caddr_t)tl,
4537 		    NFSX_STATEIDOTHER);
4538 		tl += (NFSX_STATEIDOTHER / NFSX_UNSIGNED);
4539 		if (trunc)
4540 			*tl = newnfs_true;
4541 		else
4542 			*tl = newnfs_false;
4543 		(void)nfsm_fhtom(nd, (u_int8_t *)fhp, NFSX_MYFH, 0);
4544 	} else if (procnum == NFSV4OP_CBLAYOUTRECALL) {
4545 		NFSD_DEBUG(4, "docallback layout recall\n");
4546 		nd->nd_procnum = NFSV4PROC_CBCOMPOUND;
4547 		error = nfsrv_cbcallargs(nd, clp, callback,
4548 		    NFSV4OP_CBLAYOUTRECALL, "CB Reclayout", &sep, &slotpos);
4549 		NFSD_DEBUG(4, "aft cbcallargs=%d\n", error);
4550 		if (error != 0) {
4551 			m_freem(nd->nd_mreq);
4552 			goto errout;
4553 		}
4554 		NFSM_BUILD(tl, u_int32_t *, 4 * NFSX_UNSIGNED);
4555 		*tl++ = txdr_unsigned(laytype);
4556 		*tl++ = txdr_unsigned(NFSLAYOUTIOMODE_ANY);
4557 		if (trunc)
4558 			*tl++ = newnfs_true;
4559 		else
4560 			*tl++ = newnfs_false;
4561 		*tl = txdr_unsigned(NFSV4LAYOUTRET_FILE);
4562 		nfsm_fhtom(nd, (uint8_t *)fhp, NFSX_MYFH, 0);
4563 		NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_HYPER + NFSX_STATEID);
4564 		tval = 0;
4565 		txdr_hyper(tval, tl); tl += 2;
4566 		tval = UINT64_MAX;
4567 		txdr_hyper(tval, tl); tl += 2;
4568 		*tl++ = txdr_unsigned(stateidp->seqid);
4569 		NFSBCOPY(stateidp->other, tl, NFSX_STATEIDOTHER);
4570 		tl += (NFSX_STATEIDOTHER / NFSX_UNSIGNED);
4571 		NFSD_DEBUG(4, "aft args\n");
4572 	} else if (procnum == NFSV4PROC_CBNULL) {
4573 		nd->nd_procnum = NFSV4PROC_CBNULL;
4574 		if ((clp->lc_flags & LCL_NFSV41) != 0) {
4575 			error = nfsv4_getcbsession(clp, &sep);
4576 			if (error != 0) {
4577 				m_freem(nd->nd_mreq);
4578 				goto errout;
4579 			}
4580 		}
4581 	} else {
4582 		error = NFSERR_SERVERFAULT;
4583 		m_freem(nd->nd_mreq);
4584 		goto errout;
4585 	}
4586 
4587 	/*
4588 	 * Call newnfs_connect(), as required, and then newnfs_request().
4589 	 */
4590 	dotls = false;
4591 	if ((clp->lc_flags & LCL_TLSCB) != 0)
4592 		dotls = true;
4593 	(void) newnfs_sndlock(&clp->lc_req.nr_lock);
4594 	if (clp->lc_req.nr_client == NULL) {
4595 		if ((clp->lc_flags & LCL_NFSV41) != 0) {
4596 			error = ECONNREFUSED;
4597 			if (procnum != NFSV4PROC_CBNULL)
4598 				nfsv4_freeslot(&sep->sess_cbsess, slotpos,
4599 				    true);
4600 			nfsrv_freesession(sep, NULL);
4601 		} else if (nd->nd_procnum == NFSV4PROC_CBNULL)
4602 			error = newnfs_connect(NULL, &clp->lc_req, cred,
4603 			    NULL, 1, dotls, &clp->lc_req.nr_client);
4604 		else
4605 			error = newnfs_connect(NULL, &clp->lc_req, cred,
4606 			    NULL, 3, dotls, &clp->lc_req.nr_client);
4607 	}
4608 	newnfs_sndunlock(&clp->lc_req.nr_lock);
4609 	NFSD_DEBUG(4, "aft sndunlock=%d\n", error);
4610 	if (!error) {
4611 		if ((nd->nd_flag & ND_NFSV41) != 0) {
4612 			KASSERT(sep != NULL, ("sep NULL"));
4613 			if (sep->sess_cbsess.nfsess_xprt != NULL)
4614 				error = newnfs_request(nd, NULL, clp,
4615 				    &clp->lc_req, NULL, NULL, cred,
4616 				    clp->lc_program, clp->lc_req.nr_vers, NULL,
4617 				    1, NULL, &sep->sess_cbsess);
4618 			else {
4619 				/*
4620 				 * This should probably never occur, but if a
4621 				 * client somehow does an RPC without a
4622 				 * SequenceID Op that causes a callback just
4623 				 * after the nfsd threads have been terminated
4624 				 * and restared we could conceivably get here
4625 				 * without a backchannel xprt.
4626 				 */
4627 				printf("nfsrv_docallback: no xprt\n");
4628 				error = ECONNREFUSED;
4629 			}
4630 			NFSD_DEBUG(4, "aft newnfs_request=%d\n", error);
4631 			if (error != 0 && procnum != NFSV4PROC_CBNULL) {
4632 				/*
4633 				 * It is likely that the callback was never
4634 				 * processed by the client and, as such,
4635 				 * the sequence# for the session slot needs
4636 				 * to be backed up by one to avoid a
4637 				 * NFSERR_SEQMISORDERED error reply.
4638 				 * For the unlikely case where the callback
4639 				 * was processed by the client, this will
4640 				 * make the next callback on the slot
4641 				 * appear to be a retry.
4642 				 * Since callbacks never specify that the
4643 				 * reply be cached, this "apparent retry"
4644 				 * should not be a problem.
4645 				 */
4646 				nfsv4_freeslot(&sep->sess_cbsess, slotpos,
4647 				    true);
4648 			}
4649 			nfsrv_freesession(sep, NULL);
4650 		} else
4651 			error = newnfs_request(nd, NULL, clp, &clp->lc_req,
4652 			    NULL, NULL, cred, clp->lc_program,
4653 			    clp->lc_req.nr_vers, NULL, 1, NULL, NULL);
4654 	}
4655 errout:
4656 	NFSFREECRED(cred);
4657 
4658 	/*
4659 	 * If error is set here, the Callback path isn't working
4660 	 * properly, so twiddle the appropriate LCL_ flags.
4661 	 * (nd_repstat != 0 indicates the Callback path is working,
4662 	 *  but the callback failed on the client.)
4663 	 */
4664 	if (error) {
4665 		/*
4666 		 * Mark the callback pathway down, which disabled issuing
4667 		 * of delegations and gets Renew to return NFSERR_CBPATHDOWN.
4668 		 */
4669 		NFSLOCKSTATE();
4670 		clp->lc_flags |= LCL_CBDOWN;
4671 		NFSUNLOCKSTATE();
4672 	} else {
4673 		/*
4674 		 * Callback worked. If the callback path was down, disable
4675 		 * callbacks, so no more delegations will be issued. (This
4676 		 * is done on the assumption that the callback pathway is
4677 		 * flakey.)
4678 		 */
4679 		NFSLOCKSTATE();
4680 		if (clp->lc_flags & LCL_CBDOWN)
4681 			clp->lc_flags &= ~(LCL_CBDOWN | LCL_CALLBACKSON);
4682 		NFSUNLOCKSTATE();
4683 		if (nd->nd_repstat) {
4684 			error = nd->nd_repstat;
4685 			NFSD_DEBUG(1, "nfsrv_docallback op=%d err=%d\n",
4686 			    procnum, error);
4687 		} else if (error == 0 && procnum == NFSV4OP_CBGETATTR)
4688 			error = nfsv4_loadattr(nd, NULL, nap, NULL, NULL, 0,
4689 			    NULL, NULL, NULL, NULL, NULL, 0, NULL, NULL, NULL,
4690 			    p, NULL);
4691 		m_freem(nd->nd_mrep);
4692 	}
4693 	NFSLOCKSTATE();
4694 	clp->lc_cbref--;
4695 	if ((clp->lc_flags & LCL_WAKEUPWANTED) && clp->lc_cbref == 0) {
4696 		clp->lc_flags &= ~LCL_WAKEUPWANTED;
4697 		wakeup(clp);
4698 	}
4699 	NFSUNLOCKSTATE();
4700 
4701 	free(nd, M_TEMP);
4702 	NFSEXITCODE(error);
4703 	return (error);
4704 }
4705 
4706 /*
4707  * Set up the compound RPC for the callback.
4708  */
4709 static int
4710 nfsrv_cbcallargs(struct nfsrv_descript *nd, struct nfsclient *clp,
4711     uint32_t callback, int op, const char *optag, struct nfsdsession **sepp,
4712     int *slotposp)
4713 {
4714 	uint32_t *tl;
4715 	int error, len;
4716 
4717 	len = strlen(optag);
4718 	(void)nfsm_strtom(nd, optag, len);
4719 	NFSM_BUILD(tl, uint32_t *, 4 * NFSX_UNSIGNED);
4720 	if ((nd->nd_flag & ND_NFSV41) != 0) {
4721 		if ((nd->nd_flag & ND_NFSV42) != 0)
4722 			*tl++ = txdr_unsigned(NFSV42_MINORVERSION);
4723 		else
4724 			*tl++ = txdr_unsigned(NFSV41_MINORVERSION);
4725 		*tl++ = txdr_unsigned(callback);
4726 		*tl++ = txdr_unsigned(2);
4727 		*tl = txdr_unsigned(NFSV4OP_CBSEQUENCE);
4728 		error = nfsv4_setcbsequence(nd, clp, 1, sepp, slotposp);
4729 		if (error != 0)
4730 			return (error);
4731 		NFSM_BUILD(tl, u_int32_t *, NFSX_UNSIGNED);
4732 		*tl = txdr_unsigned(op);
4733 	} else {
4734 		*tl++ = txdr_unsigned(NFSV4_MINORVERSION);
4735 		*tl++ = txdr_unsigned(callback);
4736 		*tl++ = txdr_unsigned(1);
4737 		*tl = txdr_unsigned(op);
4738 	}
4739 	return (0);
4740 }
4741 
4742 /*
4743  * Return the next index# for a clientid. Mostly just increment and return
4744  * the next one, but... if the 32bit unsigned does actually wrap around,
4745  * it should be rebooted.
4746  * At an average rate of one new client per second, it will wrap around in
4747  * approximately 136 years. (I think the server will have been shut
4748  * down or rebooted before then.)
4749  */
4750 static u_int32_t
4751 nfsrv_nextclientindex(void)
4752 {
4753 	static u_int32_t client_index = 0;
4754 
4755 	client_index++;
4756 	if (client_index != 0)
4757 		return (client_index);
4758 
4759 	printf("%s: out of clientids\n", __func__);
4760 	return (client_index);
4761 }
4762 
4763 /*
4764  * Return the next index# for a stateid. Mostly just increment and return
4765  * the next one, but... if the 32bit unsigned does actually wrap around
4766  * (will a BSD server stay up that long?), find
4767  * new start and end values.
4768  */
4769 static u_int32_t
4770 nfsrv_nextstateindex(struct nfsclient *clp)
4771 {
4772 	struct nfsstate *stp;
4773 	int i;
4774 	u_int32_t canuse, min_index, max_index;
4775 
4776 	if (!(clp->lc_flags & LCL_INDEXNOTOK)) {
4777 		clp->lc_stateindex++;
4778 		if (clp->lc_stateindex != clp->lc_statemaxindex)
4779 			return (clp->lc_stateindex);
4780 	}
4781 
4782 	/*
4783 	 * Yuck, we've hit the end.
4784 	 * Look for a new min and max.
4785 	 */
4786 	min_index = 0;
4787 	max_index = 0xffffffff;
4788 	for (i = 0; i < nfsrv_statehashsize; i++) {
4789 	    LIST_FOREACH(stp, &clp->lc_stateid[i], ls_hash) {
4790 		if (stp->ls_stateid.other[2] > 0x80000000) {
4791 		    if (stp->ls_stateid.other[2] < max_index)
4792 			max_index = stp->ls_stateid.other[2];
4793 		} else {
4794 		    if (stp->ls_stateid.other[2] > min_index)
4795 			min_index = stp->ls_stateid.other[2];
4796 		}
4797 	    }
4798 	}
4799 
4800 	/*
4801 	 * Yikes, highly unlikely, but I'll handle it anyhow.
4802 	 */
4803 	if (min_index == 0x80000000 && max_index == 0x80000001) {
4804 	    canuse = 0;
4805 	    /*
4806 	     * Loop around until we find an unused entry. Return that
4807 	     * and set LCL_INDEXNOTOK, so the search will continue next time.
4808 	     * (This is one of those rare cases where a goto is the
4809 	     *  cleanest way to code the loop.)
4810 	     */
4811 tryagain:
4812 	    for (i = 0; i < nfsrv_statehashsize; i++) {
4813 		LIST_FOREACH(stp, &clp->lc_stateid[i], ls_hash) {
4814 		    if (stp->ls_stateid.other[2] == canuse) {
4815 			canuse++;
4816 			goto tryagain;
4817 		    }
4818 		}
4819 	    }
4820 	    clp->lc_flags |= LCL_INDEXNOTOK;
4821 	    return (canuse);
4822 	}
4823 
4824 	/*
4825 	 * Ok to start again from min + 1.
4826 	 */
4827 	clp->lc_stateindex = min_index + 1;
4828 	clp->lc_statemaxindex = max_index;
4829 	clp->lc_flags &= ~LCL_INDEXNOTOK;
4830 	return (clp->lc_stateindex);
4831 }
4832 
4833 /*
4834  * The following functions handle the stable storage file that deals with
4835  * the edge conditions described in RFC3530 Sec. 8.6.3.
4836  * The file is as follows:
4837  * - a single record at the beginning that has the lease time of the
4838  *   previous server instance (before the last reboot) and the nfsrvboottime
4839  *   values for the previous server boots.
4840  *   These previous boot times are used to ensure that the current
4841  *   nfsrvboottime does not, somehow, get set to a previous one.
4842  *   (This is important so that Stale ClientIDs and StateIDs can
4843  *    be recognized.)
4844  *   The number of previous nfsvrboottime values precedes the list.
4845  * - followed by some number of appended records with:
4846  *   - client id string
4847  *   - flag that indicates it is a record revoking state via lease
4848  *     expiration or similar
4849  *     OR has successfully acquired state.
4850  * These structures vary in length, with the client string at the end, up
4851  * to NFSV4_OPAQUELIMIT in size.
4852  *
4853  * At the end of the grace period, the file is truncated, the first
4854  * record is rewritten with updated information and any acquired state
4855  * records for successful reclaims of state are written.
4856  *
4857  * Subsequent records are appended when the first state is issued to
4858  * a client and when state is revoked for a client.
4859  *
4860  * When reading the file in, state issued records that come later in
4861  * the file override older ones, since the append log is in cronological order.
4862  * If, for some reason, the file can't be read, the grace period is
4863  * immediately terminated and all reclaims get NFSERR_NOGRACE.
4864  */
4865 
4866 /*
4867  * Read in the stable storage file. Called by nfssvc() before the nfsd
4868  * processes start servicing requests.
4869  */
4870 void
4871 nfsrv_setupstable(NFSPROC_T *p)
4872 {
4873 	struct nfsrv_stablefirst *sf = &nfsrv_stablefirst;
4874 	struct nfsrv_stable *sp, *nsp;
4875 	struct nfst_rec *tsp;
4876 	int error, i, tryagain;
4877 	off_t off = 0;
4878 	ssize_t aresid, len;
4879 
4880 	/*
4881 	 * If NFSNSF_UPDATEDONE is set, this is a restart of the nfsds without
4882 	 * a reboot, so state has not been lost.
4883 	 */
4884 	if (sf->nsf_flags & NFSNSF_UPDATEDONE)
4885 		return;
4886 	/*
4887 	 * Set Grace over just until the file reads successfully.
4888 	 */
4889 	nfsrvboottime = time_second;
4890 	LIST_INIT(&sf->nsf_head);
4891 	sf->nsf_flags = (NFSNSF_GRACEOVER | NFSNSF_NEEDLOCK);
4892 	sf->nsf_eograce = NFSD_MONOSEC + NFSRV_LEASEDELTA;
4893 	if (sf->nsf_fp == NULL)
4894 		return;
4895 	error = NFSD_RDWR(UIO_READ, NFSFPVNODE(sf->nsf_fp),
4896 	    (caddr_t)&sf->nsf_rec, sizeof (struct nfsf_rec), off, UIO_SYSSPACE,
4897 	    0, NFSFPCRED(sf->nsf_fp), &aresid, p);
4898 	if (error || aresid || sf->nsf_numboots == 0 ||
4899 		sf->nsf_numboots > NFSNSF_MAXNUMBOOTS)
4900 		return;
4901 
4902 	/*
4903 	 * Now, read in the boottimes.
4904 	 */
4905 	sf->nsf_bootvals = (time_t *)malloc((sf->nsf_numboots + 1) *
4906 		sizeof (time_t), M_TEMP, M_WAITOK);
4907 	off = sizeof (struct nfsf_rec);
4908 	error = NFSD_RDWR(UIO_READ, NFSFPVNODE(sf->nsf_fp),
4909 	    (caddr_t)sf->nsf_bootvals, sf->nsf_numboots * sizeof (time_t), off,
4910 	    UIO_SYSSPACE, 0, NFSFPCRED(sf->nsf_fp), &aresid, p);
4911 	if (error || aresid) {
4912 		free(sf->nsf_bootvals, M_TEMP);
4913 		sf->nsf_bootvals = NULL;
4914 		return;
4915 	}
4916 
4917 	/*
4918 	 * Make sure this nfsrvboottime is different from all recorded
4919 	 * previous ones.
4920 	 */
4921 	do {
4922 		tryagain = 0;
4923 		for (i = 0; i < sf->nsf_numboots; i++) {
4924 			if (nfsrvboottime == sf->nsf_bootvals[i]) {
4925 				nfsrvboottime++;
4926 				tryagain = 1;
4927 				break;
4928 			}
4929 		}
4930 	} while (tryagain);
4931 
4932 	sf->nsf_flags |= NFSNSF_OK;
4933 	off += (sf->nsf_numboots * sizeof (time_t));
4934 
4935 	/*
4936 	 * Read through the file, building a list of records for grace
4937 	 * checking.
4938 	 * Each record is between sizeof (struct nfst_rec) and
4939 	 * sizeof (struct nfst_rec) + NFSV4_OPAQUELIMIT - 1
4940 	 * and is actually sizeof (struct nfst_rec) + nst_len - 1.
4941 	 */
4942 	tsp = (struct nfst_rec *)malloc(sizeof (struct nfst_rec) +
4943 		NFSV4_OPAQUELIMIT - 1, M_TEMP, M_WAITOK);
4944 	do {
4945 	    error = NFSD_RDWR(UIO_READ, NFSFPVNODE(sf->nsf_fp),
4946 	        (caddr_t)tsp, sizeof (struct nfst_rec) + NFSV4_OPAQUELIMIT - 1,
4947 	        off, UIO_SYSSPACE, 0, NFSFPCRED(sf->nsf_fp), &aresid, p);
4948 	    len = (sizeof (struct nfst_rec) + NFSV4_OPAQUELIMIT - 1) - aresid;
4949 	    if (error || (len > 0 && (len < sizeof (struct nfst_rec) ||
4950 		len < (sizeof (struct nfst_rec) + tsp->len - 1)))) {
4951 		/*
4952 		 * Yuck, the file has been corrupted, so just return
4953 		 * after clearing out any restart state, so the grace period
4954 		 * is over.
4955 		 */
4956 		LIST_FOREACH_SAFE(sp, &sf->nsf_head, nst_list, nsp) {
4957 			LIST_REMOVE(sp, nst_list);
4958 			free(sp, M_TEMP);
4959 		}
4960 		free(tsp, M_TEMP);
4961 		sf->nsf_flags &= ~NFSNSF_OK;
4962 		free(sf->nsf_bootvals, M_TEMP);
4963 		sf->nsf_bootvals = NULL;
4964 		return;
4965 	    }
4966 	    if (len > 0) {
4967 		off += sizeof (struct nfst_rec) + tsp->len - 1;
4968 		/*
4969 		 * Search the list for a matching client.
4970 		 */
4971 		LIST_FOREACH(sp, &sf->nsf_head, nst_list) {
4972 			if (tsp->len == sp->nst_len &&
4973 			    !NFSBCMP(tsp->client, sp->nst_client, tsp->len))
4974 				break;
4975 		}
4976 		if (sp == LIST_END(&sf->nsf_head)) {
4977 			sp = (struct nfsrv_stable *)malloc(tsp->len +
4978 				sizeof (struct nfsrv_stable) - 1, M_TEMP,
4979 				M_WAITOK);
4980 			NFSBCOPY((caddr_t)tsp, (caddr_t)&sp->nst_rec,
4981 				sizeof (struct nfst_rec) + tsp->len - 1);
4982 			LIST_INSERT_HEAD(&sf->nsf_head, sp, nst_list);
4983 		} else {
4984 			if (tsp->flag == NFSNST_REVOKE)
4985 				sp->nst_flag |= NFSNST_REVOKE;
4986 			else
4987 				/*
4988 				 * A subsequent timestamp indicates the client
4989 				 * did a setclientid/confirm and any previous
4990 				 * revoke is no longer relevant.
4991 				 */
4992 				sp->nst_flag &= ~NFSNST_REVOKE;
4993 		}
4994 	    }
4995 	} while (len > 0);
4996 	free(tsp, M_TEMP);
4997 	sf->nsf_flags = NFSNSF_OK;
4998 	sf->nsf_eograce = NFSD_MONOSEC + sf->nsf_lease +
4999 		NFSRV_LEASEDELTA;
5000 }
5001 
5002 /*
5003  * Update the stable storage file, now that the grace period is over.
5004  */
5005 void
5006 nfsrv_updatestable(NFSPROC_T *p)
5007 {
5008 	struct nfsrv_stablefirst *sf = &nfsrv_stablefirst;
5009 	struct nfsrv_stable *sp, *nsp;
5010 	int i;
5011 	struct nfsvattr nva;
5012 	vnode_t vp;
5013 #if defined(__FreeBSD_version) && (__FreeBSD_version >= 500000)
5014 	mount_t mp = NULL;
5015 #endif
5016 	int error;
5017 
5018 	if (sf->nsf_fp == NULL || (sf->nsf_flags & NFSNSF_UPDATEDONE))
5019 		return;
5020 	sf->nsf_flags |= NFSNSF_UPDATEDONE;
5021 	/*
5022 	 * Ok, we need to rewrite the stable storage file.
5023 	 * - truncate to 0 length
5024 	 * - write the new first structure
5025 	 * - loop through the data structures, writing out any that
5026 	 *   have timestamps older than the old boot
5027 	 */
5028 	if (sf->nsf_bootvals) {
5029 		sf->nsf_numboots++;
5030 		for (i = sf->nsf_numboots - 2; i >= 0; i--)
5031 			sf->nsf_bootvals[i + 1] = sf->nsf_bootvals[i];
5032 	} else {
5033 		sf->nsf_numboots = 1;
5034 		sf->nsf_bootvals = (time_t *)malloc(sizeof (time_t),
5035 			M_TEMP, M_WAITOK);
5036 	}
5037 	sf->nsf_bootvals[0] = nfsrvboottime;
5038 	sf->nsf_lease = nfsrv_lease;
5039 	NFSVNO_ATTRINIT(&nva);
5040 	NFSVNO_SETATTRVAL(&nva, size, 0);
5041 	vp = NFSFPVNODE(sf->nsf_fp);
5042 	vn_start_write(vp, &mp, V_WAIT);
5043 	if (NFSVOPLOCK(vp, LK_EXCLUSIVE) == 0) {
5044 		error = nfsvno_setattr(vp, &nva, NFSFPCRED(sf->nsf_fp), p,
5045 		    NULL);
5046 		NFSVOPUNLOCK(vp);
5047 	} else
5048 		error = EPERM;
5049 	vn_finished_write(mp);
5050 	if (!error)
5051 	    error = NFSD_RDWR(UIO_WRITE, vp,
5052 		(caddr_t)&sf->nsf_rec, sizeof (struct nfsf_rec), (off_t)0,
5053 		UIO_SYSSPACE, IO_SYNC, NFSFPCRED(sf->nsf_fp), NULL, p);
5054 	if (!error)
5055 	    error = NFSD_RDWR(UIO_WRITE, vp,
5056 		(caddr_t)sf->nsf_bootvals,
5057 		sf->nsf_numboots * sizeof (time_t),
5058 		(off_t)(sizeof (struct nfsf_rec)),
5059 		UIO_SYSSPACE, IO_SYNC, NFSFPCRED(sf->nsf_fp), NULL, p);
5060 	free(sf->nsf_bootvals, M_TEMP);
5061 	sf->nsf_bootvals = NULL;
5062 	if (error) {
5063 		sf->nsf_flags &= ~NFSNSF_OK;
5064 		printf("EEK! Can't write NfsV4 stable storage file\n");
5065 		return;
5066 	}
5067 	sf->nsf_flags |= NFSNSF_OK;
5068 
5069 	/*
5070 	 * Loop through the list and write out timestamp records for
5071 	 * any clients that successfully reclaimed state.
5072 	 */
5073 	LIST_FOREACH_SAFE(sp, &sf->nsf_head, nst_list, nsp) {
5074 		if (sp->nst_flag & NFSNST_GOTSTATE) {
5075 			nfsrv_writestable(sp->nst_client, sp->nst_len,
5076 				NFSNST_NEWSTATE, p);
5077 			sp->nst_clp->lc_flags |= LCL_STAMPEDSTABLE;
5078 		}
5079 		LIST_REMOVE(sp, nst_list);
5080 		free(sp, M_TEMP);
5081 	}
5082 	nfsrv_backupstable();
5083 }
5084 
5085 /*
5086  * Append a record to the stable storage file.
5087  */
5088 void
5089 nfsrv_writestable(u_char *client, int len, int flag, NFSPROC_T *p)
5090 {
5091 	struct nfsrv_stablefirst *sf = &nfsrv_stablefirst;
5092 	struct nfst_rec *sp;
5093 	int error;
5094 
5095 	if (!(sf->nsf_flags & NFSNSF_OK) || sf->nsf_fp == NULL)
5096 		return;
5097 	sp = (struct nfst_rec *)malloc(sizeof (struct nfst_rec) +
5098 		len - 1, M_TEMP, M_WAITOK);
5099 	sp->len = len;
5100 	NFSBCOPY(client, sp->client, len);
5101 	sp->flag = flag;
5102 	error = NFSD_RDWR(UIO_WRITE, NFSFPVNODE(sf->nsf_fp),
5103 	    (caddr_t)sp, sizeof (struct nfst_rec) + len - 1, (off_t)0,
5104 	    UIO_SYSSPACE, (IO_SYNC | IO_APPEND), NFSFPCRED(sf->nsf_fp), NULL, p);
5105 	free(sp, M_TEMP);
5106 	if (error) {
5107 		sf->nsf_flags &= ~NFSNSF_OK;
5108 		printf("EEK! Can't write NfsV4 stable storage file\n");
5109 	}
5110 }
5111 
5112 /*
5113  * This function is called during the grace period to mark a client
5114  * that successfully reclaimed state.
5115  */
5116 static void
5117 nfsrv_markstable(struct nfsclient *clp)
5118 {
5119 	struct nfsrv_stable *sp;
5120 
5121 	/*
5122 	 * First find the client structure.
5123 	 */
5124 	LIST_FOREACH(sp, &nfsrv_stablefirst.nsf_head, nst_list) {
5125 		if (sp->nst_len == clp->lc_idlen &&
5126 		    !NFSBCMP(sp->nst_client, clp->lc_id, sp->nst_len))
5127 			break;
5128 	}
5129 	if (sp == LIST_END(&nfsrv_stablefirst.nsf_head))
5130 		return;
5131 
5132 	/*
5133 	 * Now, just mark it and set the nfsclient back pointer.
5134 	 */
5135 	sp->nst_flag |= NFSNST_GOTSTATE;
5136 	sp->nst_clp = clp;
5137 }
5138 
5139 /*
5140  * This function is called when a NFSv4.1 client does a ReclaimComplete.
5141  * Very similar to nfsrv_markstable(), except for the flag being set.
5142  */
5143 static void
5144 nfsrv_markreclaim(struct nfsclient *clp)
5145 {
5146 	struct nfsrv_stable *sp;
5147 
5148 	/*
5149 	 * First find the client structure.
5150 	 */
5151 	LIST_FOREACH(sp, &nfsrv_stablefirst.nsf_head, nst_list) {
5152 		if (sp->nst_len == clp->lc_idlen &&
5153 		    !NFSBCMP(sp->nst_client, clp->lc_id, sp->nst_len))
5154 			break;
5155 	}
5156 	if (sp == LIST_END(&nfsrv_stablefirst.nsf_head))
5157 		return;
5158 
5159 	/*
5160 	 * Now, just set the flag.
5161 	 */
5162 	sp->nst_flag |= NFSNST_RECLAIMED;
5163 }
5164 
5165 /*
5166  * This function is called for a reclaim, to see if it gets grace.
5167  * It returns 0 if a reclaim is allowed, 1 otherwise.
5168  */
5169 static int
5170 nfsrv_checkstable(struct nfsclient *clp)
5171 {
5172 	struct nfsrv_stable *sp;
5173 
5174 	/*
5175 	 * First, find the entry for the client.
5176 	 */
5177 	LIST_FOREACH(sp, &nfsrv_stablefirst.nsf_head, nst_list) {
5178 		if (sp->nst_len == clp->lc_idlen &&
5179 		    !NFSBCMP(sp->nst_client, clp->lc_id, sp->nst_len))
5180 			break;
5181 	}
5182 
5183 	/*
5184 	 * If not in the list, state was revoked or no state was issued
5185 	 * since the previous reboot, a reclaim is denied.
5186 	 */
5187 	if (sp == LIST_END(&nfsrv_stablefirst.nsf_head) ||
5188 	    (sp->nst_flag & NFSNST_REVOKE) ||
5189 	    !(nfsrv_stablefirst.nsf_flags & NFSNSF_OK))
5190 		return (1);
5191 	return (0);
5192 }
5193 
5194 /*
5195  * Test for and try to clear out a conflicting client. This is called by
5196  * nfsrv_lockctrl() and nfsrv_openctrl() when conflicts with other clients
5197  * a found.
5198  * The trick here is that it can't revoke a conflicting client with an
5199  * expired lease unless it holds the v4root lock, so...
5200  * If no v4root lock, get the lock and return 1 to indicate "try again".
5201  * Return 0 to indicate the conflict can't be revoked and 1 to indicate
5202  * the revocation worked and the conflicting client is "bye, bye", so it
5203  * can be tried again.
5204  * Return 2 to indicate that the vnode is VIRF_DOOMED after NFSVOPLOCK().
5205  * Unlocks State before a non-zero value is returned.
5206  */
5207 static int
5208 nfsrv_clientconflict(struct nfsclient *clp, int *haslockp, vnode_t vp,
5209     NFSPROC_T *p)
5210 {
5211 	int gotlock, lktype = 0;
5212 
5213 	/*
5214 	 * If lease hasn't expired, we can't fix it.
5215 	 */
5216 	if (clp->lc_expiry >= NFSD_MONOSEC ||
5217 	    !(nfsrv_stablefirst.nsf_flags & NFSNSF_UPDATEDONE))
5218 		return (0);
5219 	if (*haslockp == 0) {
5220 		NFSUNLOCKSTATE();
5221 		if (vp != NULL) {
5222 			lktype = NFSVOPISLOCKED(vp);
5223 			NFSVOPUNLOCK(vp);
5224 		}
5225 		NFSLOCKV4ROOTMUTEX();
5226 		nfsv4_relref(&nfsv4rootfs_lock);
5227 		do {
5228 			gotlock = nfsv4_lock(&nfsv4rootfs_lock, 1, NULL,
5229 			    NFSV4ROOTLOCKMUTEXPTR, NULL);
5230 		} while (!gotlock);
5231 		NFSUNLOCKV4ROOTMUTEX();
5232 		*haslockp = 1;
5233 		if (vp != NULL) {
5234 			NFSVOPLOCK(vp, lktype | LK_RETRY);
5235 			if (VN_IS_DOOMED(vp))
5236 				return (2);
5237 		}
5238 		return (1);
5239 	}
5240 	NFSUNLOCKSTATE();
5241 
5242 	/*
5243 	 * Ok, we can expire the conflicting client.
5244 	 */
5245 	nfsrv_writestable(clp->lc_id, clp->lc_idlen, NFSNST_REVOKE, p);
5246 	nfsrv_backupstable();
5247 	nfsrv_cleanclient(clp, p);
5248 	nfsrv_freedeleglist(&clp->lc_deleg);
5249 	nfsrv_freedeleglist(&clp->lc_olddeleg);
5250 	LIST_REMOVE(clp, lc_hash);
5251 	nfsrv_zapclient(clp, p);
5252 	return (1);
5253 }
5254 
5255 /*
5256  * Resolve a delegation conflict.
5257  * Returns 0 to indicate the conflict was resolved without sleeping.
5258  * Return -1 to indicate that the caller should check for conflicts again.
5259  * Return > 0 for an error that should be returned, normally NFSERR_DELAY.
5260  *
5261  * Also, manipulate the nfsv4root_lock, as required. It isn't changed
5262  * for a return of 0, since there was no sleep and it could be required
5263  * later. It is released for a return of NFSERR_DELAY, since the caller
5264  * will return that error. It is released when a sleep was done waiting
5265  * for the delegation to be returned or expire (so that other nfsds can
5266  * handle ops). Then, it must be acquired for the write to stable storage.
5267  * (This function is somewhat similar to nfsrv_clientconflict(), but
5268  *  the semantics differ in a couple of subtle ways. The return of 0
5269  *  indicates the conflict was resolved without sleeping here, not
5270  *  that the conflict can't be resolved and the handling of nfsv4root_lock
5271  *  differs, as noted above.)
5272  * Unlocks State before returning a non-zero value.
5273  */
5274 static int
5275 nfsrv_delegconflict(struct nfsstate *stp, int *haslockp, NFSPROC_T *p,
5276     vnode_t vp)
5277 {
5278 	struct nfsclient *clp = stp->ls_clp;
5279 	int gotlock, error, lktype = 0, retrycnt, zapped_clp;
5280 	nfsv4stateid_t tstateid;
5281 	fhandle_t tfh;
5282 
5283 	/*
5284 	 * If the conflict is with an old delegation...
5285 	 */
5286 	if (stp->ls_flags & NFSLCK_OLDDELEG) {
5287 		/*
5288 		 * You can delete it, if it has expired.
5289 		 */
5290 		if (clp->lc_delegtime < NFSD_MONOSEC) {
5291 			nfsrv_freedeleg(stp);
5292 			NFSUNLOCKSTATE();
5293 			error = -1;
5294 			goto out;
5295 		}
5296 		NFSUNLOCKSTATE();
5297 		/*
5298 		 * During this delay, the old delegation could expire or it
5299 		 * could be recovered by the client via an Open with
5300 		 * CLAIM_DELEGATE_PREV.
5301 		 * Release the nfsv4root_lock, if held.
5302 		 */
5303 		if (*haslockp) {
5304 			*haslockp = 0;
5305 			NFSLOCKV4ROOTMUTEX();
5306 			nfsv4_unlock(&nfsv4rootfs_lock, 1);
5307 			NFSUNLOCKV4ROOTMUTEX();
5308 		}
5309 		error = NFSERR_DELAY;
5310 		goto out;
5311 	}
5312 
5313 	/*
5314 	 * It's a current delegation, so:
5315 	 * - check to see if the delegation has expired
5316 	 *   - if so, get the v4root lock and then expire it
5317 	 */
5318 	if ((stp->ls_flags & NFSLCK_DELEGRECALL) == 0 || (stp->ls_lastrecall <
5319 	    NFSD_MONOSEC && clp->lc_expiry >= NFSD_MONOSEC &&
5320 	    stp->ls_delegtime >= NFSD_MONOSEC)) {
5321 		/*
5322 		 * - do a recall callback, since not yet done
5323 		 * For now, never allow truncate to be set. To use
5324 		 * truncate safely, it must be guaranteed that the
5325 		 * Remove, Rename or Setattr with size of 0 will
5326 		 * succeed and that would require major changes to
5327 		 * the VFS/Vnode OPs.
5328 		 * Set the expiry time large enough so that it won't expire
5329 		 * until after the callback, then set it correctly, once
5330 		 * the callback is done. (The delegation will now time
5331 		 * out whether or not the Recall worked ok. The timeout
5332 		 * will be extended when ops are done on the delegation
5333 		 * stateid, up to the timelimit.)
5334 		 */
5335 		if ((stp->ls_flags & NFSLCK_DELEGRECALL) == 0) {
5336 			stp->ls_delegtime = NFSD_MONOSEC + (2 * nfsrv_lease) +
5337 			    NFSRV_LEASEDELTA;
5338 			stp->ls_delegtimelimit = NFSD_MONOSEC + (6 *
5339 			    nfsrv_lease) + NFSRV_LEASEDELTA;
5340 			stp->ls_flags |= NFSLCK_DELEGRECALL;
5341 		}
5342 		stp->ls_lastrecall = time_uptime + 1;
5343 
5344 		/*
5345 		 * Loop NFSRV_CBRETRYCNT times while the CBRecall replies
5346 		 * NFSERR_BADSTATEID or NFSERR_BADHANDLE. This is done
5347 		 * in order to try and avoid a race that could happen
5348 		 * when a CBRecall request passed the Open reply with
5349 		 * the delegation in it when transitting the network.
5350 		 * Since nfsrv_docallback will sleep, don't use stp after
5351 		 * the call.
5352 		 */
5353 		NFSBCOPY((caddr_t)&stp->ls_stateid, (caddr_t)&tstateid,
5354 		    sizeof (tstateid));
5355 		NFSBCOPY((caddr_t)&stp->ls_lfp->lf_fh, (caddr_t)&tfh,
5356 		    sizeof (tfh));
5357 		NFSUNLOCKSTATE();
5358 		if (*haslockp) {
5359 			*haslockp = 0;
5360 			NFSLOCKV4ROOTMUTEX();
5361 			nfsv4_unlock(&nfsv4rootfs_lock, 1);
5362 			NFSUNLOCKV4ROOTMUTEX();
5363 		}
5364 		retrycnt = 0;
5365 		do {
5366 		    error = nfsrv_docallback(clp, NFSV4OP_CBRECALL,
5367 			&tstateid, 0, &tfh, NULL, NULL, 0, p);
5368 		    retrycnt++;
5369 		} while ((error == NFSERR_BADSTATEID ||
5370 		    error == NFSERR_BADHANDLE) && retrycnt < NFSV4_CBRETRYCNT);
5371 		error = NFSERR_DELAY;
5372 		goto out;
5373 	}
5374 
5375 	if (clp->lc_expiry >= NFSD_MONOSEC &&
5376 	    stp->ls_delegtime >= NFSD_MONOSEC) {
5377 		NFSUNLOCKSTATE();
5378 		/*
5379 		 * A recall has been done, but it has not yet expired.
5380 		 * So, RETURN_DELAY.
5381 		 */
5382 		if (*haslockp) {
5383 			*haslockp = 0;
5384 			NFSLOCKV4ROOTMUTEX();
5385 			nfsv4_unlock(&nfsv4rootfs_lock, 1);
5386 			NFSUNLOCKV4ROOTMUTEX();
5387 		}
5388 		error = NFSERR_DELAY;
5389 		goto out;
5390 	}
5391 
5392 	/*
5393 	 * If we don't yet have the lock, just get it and then return,
5394 	 * since we need that before deleting expired state, such as
5395 	 * this delegation.
5396 	 * When getting the lock, unlock the vnode, so other nfsds that
5397 	 * are in progress, won't get stuck waiting for the vnode lock.
5398 	 */
5399 	if (*haslockp == 0) {
5400 		NFSUNLOCKSTATE();
5401 		if (vp != NULL) {
5402 			lktype = NFSVOPISLOCKED(vp);
5403 			NFSVOPUNLOCK(vp);
5404 		}
5405 		NFSLOCKV4ROOTMUTEX();
5406 		nfsv4_relref(&nfsv4rootfs_lock);
5407 		do {
5408 			gotlock = nfsv4_lock(&nfsv4rootfs_lock, 1, NULL,
5409 			    NFSV4ROOTLOCKMUTEXPTR, NULL);
5410 		} while (!gotlock);
5411 		NFSUNLOCKV4ROOTMUTEX();
5412 		*haslockp = 1;
5413 		if (vp != NULL) {
5414 			NFSVOPLOCK(vp, lktype | LK_RETRY);
5415 			if (VN_IS_DOOMED(vp)) {
5416 				*haslockp = 0;
5417 				NFSLOCKV4ROOTMUTEX();
5418 				nfsv4_unlock(&nfsv4rootfs_lock, 1);
5419 				NFSUNLOCKV4ROOTMUTEX();
5420 				error = NFSERR_PERM;
5421 				goto out;
5422 			}
5423 		}
5424 		error = -1;
5425 		goto out;
5426 	}
5427 
5428 	NFSUNLOCKSTATE();
5429 	/*
5430 	 * Ok, we can delete the expired delegation.
5431 	 * First, write the Revoke record to stable storage and then
5432 	 * clear out the conflict.
5433 	 * Since all other nfsd threads are now blocked, we can safely
5434 	 * sleep without the state changing.
5435 	 */
5436 	nfsrv_writestable(clp->lc_id, clp->lc_idlen, NFSNST_REVOKE, p);
5437 	nfsrv_backupstable();
5438 	if (clp->lc_expiry < NFSD_MONOSEC) {
5439 		nfsrv_cleanclient(clp, p);
5440 		nfsrv_freedeleglist(&clp->lc_deleg);
5441 		nfsrv_freedeleglist(&clp->lc_olddeleg);
5442 		LIST_REMOVE(clp, lc_hash);
5443 		zapped_clp = 1;
5444 	} else {
5445 		nfsrv_freedeleg(stp);
5446 		zapped_clp = 0;
5447 	}
5448 	if (zapped_clp)
5449 		nfsrv_zapclient(clp, p);
5450 	error = -1;
5451 
5452 out:
5453 	NFSEXITCODE(error);
5454 	return (error);
5455 }
5456 
5457 /*
5458  * Check for a remove allowed, if remove is set to 1 and get rid of
5459  * delegations.
5460  */
5461 int
5462 nfsrv_checkremove(vnode_t vp, int remove, struct nfsrv_descript *nd,
5463     nfsquad_t clientid, NFSPROC_T *p)
5464 {
5465 	struct nfsclient *clp;
5466 	struct nfsstate *stp;
5467 	struct nfslockfile *lfp;
5468 	int error, haslock = 0;
5469 	fhandle_t nfh;
5470 
5471 	clp = NULL;
5472 	/*
5473 	 * First, get the lock file structure.
5474 	 * (A return of -1 means no associated state, so remove ok.)
5475 	 */
5476 	error = nfsrv_getlockfh(vp, NFSLCK_CHECK, NULL, &nfh, p);
5477 tryagain:
5478 	NFSLOCKSTATE();
5479 	if (error == 0 && clientid.qval != 0)
5480 		error = nfsrv_getclient(clientid, CLOPS_RENEW, &clp, NULL,
5481 		    (nfsquad_t)((u_quad_t)0), 0, nd, p);
5482 	if (!error)
5483 		error = nfsrv_getlockfile(NFSLCK_CHECK, NULL, &lfp, &nfh, 0);
5484 	if (error) {
5485 		NFSUNLOCKSTATE();
5486 		if (haslock) {
5487 			NFSLOCKV4ROOTMUTEX();
5488 			nfsv4_unlock(&nfsv4rootfs_lock, 1);
5489 			NFSUNLOCKV4ROOTMUTEX();
5490 		}
5491 		if (error == -1)
5492 			error = 0;
5493 		goto out;
5494 	}
5495 
5496 	/*
5497 	 * Now, we must Recall any delegations.
5498 	 */
5499 	error = nfsrv_cleandeleg(vp, lfp, clp, &haslock, p);
5500 	if (error) {
5501 		/*
5502 		 * nfsrv_cleandeleg() unlocks state for non-zero
5503 		 * return.
5504 		 */
5505 		if (error == -1)
5506 			goto tryagain;
5507 		if (haslock) {
5508 			NFSLOCKV4ROOTMUTEX();
5509 			nfsv4_unlock(&nfsv4rootfs_lock, 1);
5510 			NFSUNLOCKV4ROOTMUTEX();
5511 		}
5512 		goto out;
5513 	}
5514 
5515 	/*
5516 	 * Now, look for a conflicting open share.
5517 	 */
5518 	if (remove) {
5519 		/*
5520 		 * If the entry in the directory was the last reference to the
5521 		 * corresponding filesystem object, the object can be destroyed
5522 		 * */
5523 		if(lfp->lf_usecount>1)
5524 			LIST_FOREACH(stp, &lfp->lf_open, ls_file) {
5525 				if (stp->ls_flags & NFSLCK_WRITEDENY) {
5526 					error = NFSERR_FILEOPEN;
5527 					break;
5528 				}
5529 			}
5530 	}
5531 
5532 	NFSUNLOCKSTATE();
5533 	if (haslock) {
5534 		NFSLOCKV4ROOTMUTEX();
5535 		nfsv4_unlock(&nfsv4rootfs_lock, 1);
5536 		NFSUNLOCKV4ROOTMUTEX();
5537 	}
5538 
5539 out:
5540 	NFSEXITCODE(error);
5541 	return (error);
5542 }
5543 
5544 /*
5545  * Clear out all delegations for the file referred to by lfp.
5546  * May return NFSERR_DELAY, if there will be a delay waiting for
5547  * delegations to expire.
5548  * Returns -1 to indicate it slept while recalling a delegation.
5549  * This function has the side effect of deleting the nfslockfile structure,
5550  * if it no longer has associated state and didn't have to sleep.
5551  * Unlocks State before a non-zero value is returned.
5552  */
5553 static int
5554 nfsrv_cleandeleg(vnode_t vp, struct nfslockfile *lfp,
5555     struct nfsclient *clp, int *haslockp, NFSPROC_T *p)
5556 {
5557 	struct nfsstate *stp, *nstp;
5558 	int ret = 0;
5559 
5560 	stp = LIST_FIRST(&lfp->lf_deleg);
5561 	while (stp != LIST_END(&lfp->lf_deleg)) {
5562 		nstp = LIST_NEXT(stp, ls_file);
5563 		if (stp->ls_clp != clp) {
5564 			ret = nfsrv_delegconflict(stp, haslockp, p, vp);
5565 			if (ret) {
5566 				/*
5567 				 * nfsrv_delegconflict() unlocks state
5568 				 * when it returns non-zero.
5569 				 */
5570 				goto out;
5571 			}
5572 		}
5573 		stp = nstp;
5574 	}
5575 out:
5576 	NFSEXITCODE(ret);
5577 	return (ret);
5578 }
5579 
5580 /*
5581  * There are certain operations that, when being done outside of NFSv4,
5582  * require that any NFSv4 delegation for the file be recalled.
5583  * This function is to be called for those cases:
5584  * VOP_RENAME() - When a delegation is being recalled for any reason,
5585  *	the client may have to do Opens against the server, using the file's
5586  *	final component name. If the file has been renamed on the server,
5587  *	that component name will be incorrect and the Open will fail.
5588  * VOP_REMOVE() - Theoretically, a client could Open a file after it has
5589  *	been removed on the server, if there is a delegation issued to
5590  *	that client for the file. I say "theoretically" since clients
5591  *	normally do an Access Op before the Open and that Access Op will
5592  *	fail with ESTALE. Note that NFSv2 and 3 don't even do Opens, so
5593  *	they will detect the file's removal in the same manner. (There is
5594  *	one case where RFC3530 allows a client to do an Open without first
5595  *	doing an Access Op, which is passage of a check against the ACE
5596  *	returned with a Write delegation, but current practice is to ignore
5597  *	the ACE and always do an Access Op.)
5598  *	Since the functions can only be called with an unlocked vnode, this
5599  *	can't be done at this time.
5600  * VOP_ADVLOCK() - When a client holds a delegation, it can issue byte range
5601  *	locks locally in the client, which are not visible to the server. To
5602  *	deal with this, issuing of delegations for a vnode must be disabled
5603  *	and all delegations for the vnode recalled. This is done via the
5604  *	second function, using the VV_DISABLEDELEG vflag on the vnode.
5605  */
5606 void
5607 nfsd_recalldelegation(vnode_t vp, NFSPROC_T *p)
5608 {
5609 	time_t starttime;
5610 	int error;
5611 
5612 	/*
5613 	 * First, check to see if the server is currently running and it has
5614 	 * been called for a regular file when issuing delegations.
5615 	 */
5616 	if (newnfs_numnfsd == 0 || vp->v_type != VREG ||
5617 	    nfsrv_issuedelegs == 0)
5618 		return;
5619 
5620 	KASSERT((NFSVOPISLOCKED(vp) != LK_EXCLUSIVE), ("vp %p is locked", vp));
5621 	/*
5622 	 * First, get a reference on the nfsv4rootfs_lock so that an
5623 	 * exclusive lock cannot be acquired by another thread.
5624 	 */
5625 	NFSLOCKV4ROOTMUTEX();
5626 	nfsv4_getref(&nfsv4rootfs_lock, NULL, NFSV4ROOTLOCKMUTEXPTR, NULL);
5627 	NFSUNLOCKV4ROOTMUTEX();
5628 
5629 	/*
5630 	 * Now, call nfsrv_checkremove() in a loop while it returns
5631 	 * NFSERR_DELAY. Return upon any other error or when timed out.
5632 	 */
5633 	starttime = NFSD_MONOSEC;
5634 	do {
5635 		if (NFSVOPLOCK(vp, LK_EXCLUSIVE) == 0) {
5636 			error = nfsrv_checkremove(vp, 0, NULL,
5637 			    (nfsquad_t)((u_quad_t)0), p);
5638 			NFSVOPUNLOCK(vp);
5639 		} else
5640 			error = EPERM;
5641 		if (error == NFSERR_DELAY) {
5642 			if (NFSD_MONOSEC - starttime > NFS_REMOVETIMEO)
5643 				break;
5644 			/* Sleep for a short period of time */
5645 			(void) nfs_catnap(PZERO, 0, "nfsremove");
5646 		}
5647 	} while (error == NFSERR_DELAY);
5648 	NFSLOCKV4ROOTMUTEX();
5649 	nfsv4_relref(&nfsv4rootfs_lock);
5650 	NFSUNLOCKV4ROOTMUTEX();
5651 }
5652 
5653 void
5654 nfsd_disabledelegation(vnode_t vp, NFSPROC_T *p)
5655 {
5656 
5657 #ifdef VV_DISABLEDELEG
5658 	/*
5659 	 * First, flag issuance of delegations disabled.
5660 	 */
5661 	atomic_set_long(&vp->v_vflag, VV_DISABLEDELEG);
5662 #endif
5663 
5664 	/*
5665 	 * Then call nfsd_recalldelegation() to get rid of all extant
5666 	 * delegations.
5667 	 */
5668 	nfsd_recalldelegation(vp, p);
5669 }
5670 
5671 /*
5672  * Check for conflicting locks, etc. and then get rid of delegations.
5673  * (At one point I thought that I should get rid of delegations for any
5674  *  Setattr, since it could potentially disallow the I/O op (read or write)
5675  *  allowed by the delegation. However, Setattr Ops that aren't changing
5676  *  the size get a stateid of all 0s, so you can't tell if it is a delegation
5677  *  for the same client or a different one, so I decided to only get rid
5678  *  of delegations for other clients when the size is being changed.)
5679  * In general, a Setattr can disable NFS I/O Ops that are outstanding, such
5680  * as Write backs, even if there is no delegation, so it really isn't any
5681  * different?)
5682  */
5683 int
5684 nfsrv_checksetattr(vnode_t vp, struct nfsrv_descript *nd,
5685     nfsv4stateid_t *stateidp, struct nfsvattr *nvap, nfsattrbit_t *attrbitp,
5686     struct nfsexstuff *exp, NFSPROC_T *p)
5687 {
5688 	struct nfsstate st, *stp = &st;
5689 	struct nfslock lo, *lop = &lo;
5690 	int error = 0;
5691 	nfsquad_t clientid;
5692 
5693 	if (NFSISSET_ATTRBIT(attrbitp, NFSATTRBIT_SIZE)) {
5694 		stp->ls_flags = (NFSLCK_CHECK | NFSLCK_WRITEACCESS);
5695 		lop->lo_first = nvap->na_size;
5696 	} else {
5697 		stp->ls_flags = 0;
5698 		lop->lo_first = 0;
5699 	}
5700 	if (NFSISSET_ATTRBIT(attrbitp, NFSATTRBIT_OWNER) ||
5701 	    NFSISSET_ATTRBIT(attrbitp, NFSATTRBIT_OWNERGROUP) ||
5702 	    NFSISSET_ATTRBIT(attrbitp, NFSATTRBIT_MODE) ||
5703 	    NFSISSET_ATTRBIT(attrbitp, NFSATTRBIT_ACL))
5704 		stp->ls_flags |= NFSLCK_SETATTR;
5705 	if (stp->ls_flags == 0)
5706 		goto out;
5707 	lop->lo_end = NFS64BITSSET;
5708 	lop->lo_flags = NFSLCK_WRITE;
5709 	stp->ls_ownerlen = 0;
5710 	stp->ls_op = NULL;
5711 	stp->ls_uid = nd->nd_cred->cr_uid;
5712 	stp->ls_stateid.seqid = stateidp->seqid;
5713 	clientid.lval[0] = stp->ls_stateid.other[0] = stateidp->other[0];
5714 	clientid.lval[1] = stp->ls_stateid.other[1] = stateidp->other[1];
5715 	stp->ls_stateid.other[2] = stateidp->other[2];
5716 	error = nfsrv_lockctrl(vp, &stp, &lop, NULL, clientid,
5717 	    stateidp, exp, nd, p);
5718 
5719 out:
5720 	NFSEXITCODE2(error, nd);
5721 	return (error);
5722 }
5723 
5724 /*
5725  * Check for a write delegation and do a CBGETATTR if there is one, updating
5726  * the attributes, as required.
5727  * Should I return an error if I can't get the attributes? (For now, I'll
5728  * just return ok.
5729  */
5730 int
5731 nfsrv_checkgetattr(struct nfsrv_descript *nd, vnode_t vp,
5732     struct nfsvattr *nvap, nfsattrbit_t *attrbitp, NFSPROC_T *p)
5733 {
5734 	struct nfsstate *stp;
5735 	struct nfslockfile *lfp;
5736 	struct nfsclient *clp;
5737 	struct nfsvattr nva;
5738 	fhandle_t nfh;
5739 	int error = 0;
5740 	nfsattrbit_t cbbits;
5741 	u_quad_t delegfilerev;
5742 
5743 	NFSCBGETATTR_ATTRBIT(attrbitp, &cbbits);
5744 	if (!NFSNONZERO_ATTRBIT(&cbbits))
5745 		goto out;
5746 	if (nfsrv_writedelegcnt == 0)
5747 		goto out;
5748 
5749 	/*
5750 	 * Get the lock file structure.
5751 	 * (A return of -1 means no associated state, so return ok.)
5752 	 */
5753 	error = nfsrv_getlockfh(vp, NFSLCK_CHECK, NULL, &nfh, p);
5754 	NFSLOCKSTATE();
5755 	if (!error)
5756 		error = nfsrv_getlockfile(NFSLCK_CHECK, NULL, &lfp, &nfh, 0);
5757 	if (error) {
5758 		NFSUNLOCKSTATE();
5759 		if (error == -1)
5760 			error = 0;
5761 		goto out;
5762 	}
5763 
5764 	/*
5765 	 * Now, look for a write delegation.
5766 	 */
5767 	LIST_FOREACH(stp, &lfp->lf_deleg, ls_file) {
5768 		if (stp->ls_flags & NFSLCK_DELEGWRITE)
5769 			break;
5770 	}
5771 	if (stp == LIST_END(&lfp->lf_deleg)) {
5772 		NFSUNLOCKSTATE();
5773 		goto out;
5774 	}
5775 	clp = stp->ls_clp;
5776 
5777 	/* If the clientid is not confirmed, ignore the delegation. */
5778 	if (clp->lc_flags & LCL_NEEDSCONFIRM) {
5779 		NFSUNLOCKSTATE();
5780 		goto out;
5781 	}
5782 
5783 	delegfilerev = stp->ls_filerev;
5784 	/*
5785 	 * If the Write delegation was issued as a part of this Compound RPC
5786 	 * or if we have an Implied Clientid (used in a previous Op in this
5787 	 * compound) and it is the client the delegation was issued to,
5788 	 * just return ok.
5789 	 * I also assume that it is from the same client iff the network
5790 	 * host IP address is the same as the callback address. (Not
5791 	 * exactly correct by the RFC, but avoids a lot of Getattr
5792 	 * callbacks.)
5793 	 */
5794 	if (nd->nd_compref == stp->ls_compref ||
5795 	    ((nd->nd_flag & ND_IMPLIEDCLID) &&
5796 	     clp->lc_clientid.qval == nd->nd_clientid.qval) ||
5797 	     nfsaddr2_match(clp->lc_req.nr_nam, nd->nd_nam)) {
5798 		NFSUNLOCKSTATE();
5799 		goto out;
5800 	}
5801 
5802 	/*
5803 	 * We are now done with the delegation state structure,
5804 	 * so the statelock can be released and we can now tsleep().
5805 	 */
5806 
5807 	/*
5808 	 * Now, we must do the CB Getattr callback, to see if Change or Size
5809 	 * has changed.
5810 	 */
5811 	if (clp->lc_expiry >= NFSD_MONOSEC) {
5812 		NFSUNLOCKSTATE();
5813 		NFSVNO_ATTRINIT(&nva);
5814 		nva.na_filerev = NFS64BITSSET;
5815 		error = nfsrv_docallback(clp, NFSV4OP_CBGETATTR, NULL,
5816 		    0, &nfh, &nva, &cbbits, 0, p);
5817 		if (!error) {
5818 			if ((nva.na_filerev != NFS64BITSSET &&
5819 			    nva.na_filerev > delegfilerev) ||
5820 			    (NFSVNO_ISSETSIZE(&nva) &&
5821 			     nva.na_size != nvap->na_size)) {
5822 				error = nfsvno_updfilerev(vp, nvap, nd, p);
5823 				if (NFSVNO_ISSETSIZE(&nva))
5824 					nvap->na_size = nva.na_size;
5825 			}
5826 		} else
5827 			error = 0;	/* Ignore callback errors for now. */
5828 	} else {
5829 		NFSUNLOCKSTATE();
5830 	}
5831 
5832 out:
5833 	NFSEXITCODE2(error, nd);
5834 	return (error);
5835 }
5836 
5837 /*
5838  * This function looks for openowners that haven't had any opens for
5839  * a while and throws them away. Called by an nfsd when NFSNSF_NOOPENS
5840  * is set.
5841  */
5842 void
5843 nfsrv_throwawayopens(NFSPROC_T *p)
5844 {
5845 	struct nfsclient *clp, *nclp;
5846 	struct nfsstate *stp, *nstp;
5847 	int i;
5848 
5849 	NFSLOCKSTATE();
5850 	nfsrv_stablefirst.nsf_flags &= ~NFSNSF_NOOPENS;
5851 	/*
5852 	 * For each client...
5853 	 */
5854 	for (i = 0; i < nfsrv_clienthashsize; i++) {
5855 	    LIST_FOREACH_SAFE(clp, &nfsclienthash[i], lc_hash, nclp) {
5856 		LIST_FOREACH_SAFE(stp, &clp->lc_open, ls_list, nstp) {
5857 			if (LIST_EMPTY(&stp->ls_open) &&
5858 			    (stp->ls_noopens > NFSNOOPEN ||
5859 			     (nfsrv_openpluslock * 2) >
5860 			     nfsrv_v4statelimit))
5861 				nfsrv_freeopenowner(stp, 0, p);
5862 		}
5863 	    }
5864 	}
5865 	NFSUNLOCKSTATE();
5866 }
5867 
5868 /*
5869  * This function checks to see if the credentials are the same.
5870  * Returns 1 for not same, 0 otherwise.
5871  */
5872 static int
5873 nfsrv_notsamecredname(struct nfsrv_descript *nd, struct nfsclient *clp)
5874 {
5875 
5876 	if (nd->nd_flag & ND_GSS) {
5877 		if (!(clp->lc_flags & LCL_GSS))
5878 			return (1);
5879 		if (clp->lc_flags & LCL_NAME) {
5880 			if (nd->nd_princlen != clp->lc_namelen ||
5881 			    NFSBCMP(nd->nd_principal, clp->lc_name,
5882 				clp->lc_namelen))
5883 				return (1);
5884 			else
5885 				return (0);
5886 		}
5887 		if (nd->nd_cred->cr_uid == clp->lc_uid)
5888 			return (0);
5889 		else
5890 			return (1);
5891 	} else if (clp->lc_flags & LCL_GSS)
5892 		return (1);
5893 	/*
5894 	 * For AUTH_SYS, allow the same uid or root. (This is underspecified
5895 	 * in RFC3530, which talks about principals, but doesn't say anything
5896 	 * about uids for AUTH_SYS.)
5897 	 */
5898 	if (nd->nd_cred->cr_uid == clp->lc_uid || nd->nd_cred->cr_uid == 0)
5899 		return (0);
5900 	else
5901 		return (1);
5902 }
5903 
5904 /*
5905  * Calculate the lease expiry time.
5906  */
5907 static time_t
5908 nfsrv_leaseexpiry(void)
5909 {
5910 
5911 	if (nfsrv_stablefirst.nsf_eograce > NFSD_MONOSEC)
5912 		return (NFSD_MONOSEC + 2 * (nfsrv_lease + NFSRV_LEASEDELTA));
5913 	return (NFSD_MONOSEC + nfsrv_lease + NFSRV_LEASEDELTA);
5914 }
5915 
5916 /*
5917  * Delay the delegation timeout as far as ls_delegtimelimit, as required.
5918  */
5919 static void
5920 nfsrv_delaydelegtimeout(struct nfsstate *stp)
5921 {
5922 
5923 	if ((stp->ls_flags & NFSLCK_DELEGRECALL) == 0)
5924 		return;
5925 
5926 	if ((stp->ls_delegtime + 15) > NFSD_MONOSEC &&
5927 	    stp->ls_delegtime < stp->ls_delegtimelimit) {
5928 		stp->ls_delegtime += nfsrv_lease;
5929 		if (stp->ls_delegtime > stp->ls_delegtimelimit)
5930 			stp->ls_delegtime = stp->ls_delegtimelimit;
5931 	}
5932 }
5933 
5934 /*
5935  * This function checks to see if there is any other state associated
5936  * with the openowner for this Open.
5937  * It returns 1 if there is no other state, 0 otherwise.
5938  */
5939 static int
5940 nfsrv_nootherstate(struct nfsstate *stp)
5941 {
5942 	struct nfsstate *tstp;
5943 
5944 	LIST_FOREACH(tstp, &stp->ls_openowner->ls_open, ls_list) {
5945 		if (tstp != stp || !LIST_EMPTY(&tstp->ls_lock))
5946 			return (0);
5947 	}
5948 	return (1);
5949 }
5950 
5951 /*
5952  * Create a list of lock deltas (changes to local byte range locking
5953  * that can be rolled back using the list) and apply the changes via
5954  * nfsvno_advlock(). Optionally, lock the list. It is expected that either
5955  * the rollback or update function will be called after this.
5956  * It returns an error (and rolls back, as required), if any nfsvno_advlock()
5957  * call fails. If it returns an error, it will unlock the list.
5958  */
5959 static int
5960 nfsrv_locallock(vnode_t vp, struct nfslockfile *lfp, int flags,
5961     uint64_t first, uint64_t end, struct nfslockconflict *cfp, NFSPROC_T *p)
5962 {
5963 	struct nfslock *lop, *nlop;
5964 	int error = 0;
5965 
5966 	/* Loop through the list of locks. */
5967 	lop = LIST_FIRST(&lfp->lf_locallock);
5968 	while (first < end && lop != NULL) {
5969 		nlop = LIST_NEXT(lop, lo_lckowner);
5970 		if (first >= lop->lo_end) {
5971 			/* not there yet */
5972 			lop = nlop;
5973 		} else if (first < lop->lo_first) {
5974 			/* new one starts before entry in list */
5975 			if (end <= lop->lo_first) {
5976 				/* no overlap between old and new */
5977 				error = nfsrv_dolocal(vp, lfp, flags,
5978 				    NFSLCK_UNLOCK, first, end, cfp, p);
5979 				if (error != 0)
5980 					break;
5981 				first = end;
5982 			} else {
5983 				/* handle fragment overlapped with new one */
5984 				error = nfsrv_dolocal(vp, lfp, flags,
5985 				    NFSLCK_UNLOCK, first, lop->lo_first, cfp,
5986 				    p);
5987 				if (error != 0)
5988 					break;
5989 				first = lop->lo_first;
5990 			}
5991 		} else {
5992 			/* new one overlaps this entry in list */
5993 			if (end <= lop->lo_end) {
5994 				/* overlaps all of new one */
5995 				error = nfsrv_dolocal(vp, lfp, flags,
5996 				    lop->lo_flags, first, end, cfp, p);
5997 				if (error != 0)
5998 					break;
5999 				first = end;
6000 			} else {
6001 				/* handle fragment overlapped with new one */
6002 				error = nfsrv_dolocal(vp, lfp, flags,
6003 				    lop->lo_flags, first, lop->lo_end, cfp, p);
6004 				if (error != 0)
6005 					break;
6006 				first = lop->lo_end;
6007 				lop = nlop;
6008 			}
6009 		}
6010 	}
6011 	if (first < end && error == 0)
6012 		/* handle fragment past end of list */
6013 		error = nfsrv_dolocal(vp, lfp, flags, NFSLCK_UNLOCK, first,
6014 		    end, cfp, p);
6015 
6016 	NFSEXITCODE(error);
6017 	return (error);
6018 }
6019 
6020 /*
6021  * Local lock unlock. Unlock all byte ranges that are no longer locked
6022  * by NFSv4. To do this, unlock any subranges of first-->end that
6023  * do not overlap with the byte ranges of any lock in the lfp->lf_lock
6024  * list. This list has all locks for the file held by other
6025  * <clientid, lockowner> tuples. The list is ordered by increasing
6026  * lo_first value, but may have entries that overlap each other, for
6027  * the case of read locks.
6028  */
6029 static void
6030 nfsrv_localunlock(vnode_t vp, struct nfslockfile *lfp, uint64_t init_first,
6031     uint64_t init_end, NFSPROC_T *p)
6032 {
6033 	struct nfslock *lop;
6034 	uint64_t first, end, prevfirst __unused;
6035 
6036 	first = init_first;
6037 	end = init_end;
6038 	while (first < init_end) {
6039 		/* Loop through all nfs locks, adjusting first and end */
6040 		prevfirst = 0;
6041 		LIST_FOREACH(lop, &lfp->lf_lock, lo_lckfile) {
6042 			KASSERT(prevfirst <= lop->lo_first,
6043 			    ("nfsv4 locks out of order"));
6044 			KASSERT(lop->lo_first < lop->lo_end,
6045 			    ("nfsv4 bogus lock"));
6046 			prevfirst = lop->lo_first;
6047 			if (first >= lop->lo_first &&
6048 			    first < lop->lo_end)
6049 				/*
6050 				 * Overlaps with initial part, so trim
6051 				 * off that initial part by moving first past
6052 				 * it.
6053 				 */
6054 				first = lop->lo_end;
6055 			else if (end > lop->lo_first &&
6056 			    lop->lo_first > first) {
6057 				/*
6058 				 * This lock defines the end of the
6059 				 * segment to unlock, so set end to the
6060 				 * start of it and break out of the loop.
6061 				 */
6062 				end = lop->lo_first;
6063 				break;
6064 			}
6065 			if (first >= end)
6066 				/*
6067 				 * There is no segment left to do, so
6068 				 * break out of this loop and then exit
6069 				 * the outer while() since first will be set
6070 				 * to end, which must equal init_end here.
6071 				 */
6072 				break;
6073 		}
6074 		if (first < end) {
6075 			/* Unlock this segment */
6076 			(void) nfsrv_dolocal(vp, lfp, NFSLCK_UNLOCK,
6077 			    NFSLCK_READ, first, end, NULL, p);
6078 			nfsrv_locallock_commit(lfp, NFSLCK_UNLOCK,
6079 			    first, end);
6080 		}
6081 		/*
6082 		 * Now move past this segment and look for any further
6083 		 * segment in the range, if there is one.
6084 		 */
6085 		first = end;
6086 		end = init_end;
6087 	}
6088 }
6089 
6090 /*
6091  * Do the local lock operation and update the rollback list, as required.
6092  * Perform the rollback and return the error if nfsvno_advlock() fails.
6093  */
6094 static int
6095 nfsrv_dolocal(vnode_t vp, struct nfslockfile *lfp, int flags, int oldflags,
6096     uint64_t first, uint64_t end, struct nfslockconflict *cfp, NFSPROC_T *p)
6097 {
6098 	struct nfsrollback *rlp;
6099 	int error = 0, ltype, oldltype;
6100 
6101 	if (flags & NFSLCK_WRITE)
6102 		ltype = F_WRLCK;
6103 	else if (flags & NFSLCK_READ)
6104 		ltype = F_RDLCK;
6105 	else
6106 		ltype = F_UNLCK;
6107 	if (oldflags & NFSLCK_WRITE)
6108 		oldltype = F_WRLCK;
6109 	else if (oldflags & NFSLCK_READ)
6110 		oldltype = F_RDLCK;
6111 	else
6112 		oldltype = F_UNLCK;
6113 	if (ltype == oldltype || (oldltype == F_WRLCK && ltype == F_RDLCK))
6114 		/* nothing to do */
6115 		goto out;
6116 	error = nfsvno_advlock(vp, ltype, first, end, p);
6117 	if (error != 0) {
6118 		if (cfp != NULL) {
6119 			cfp->cl_clientid.lval[0] = 0;
6120 			cfp->cl_clientid.lval[1] = 0;
6121 			cfp->cl_first = 0;
6122 			cfp->cl_end = NFS64BITSSET;
6123 			cfp->cl_flags = NFSLCK_WRITE;
6124 			cfp->cl_ownerlen = 5;
6125 			NFSBCOPY("LOCAL", cfp->cl_owner, 5);
6126 		}
6127 		nfsrv_locallock_rollback(vp, lfp, p);
6128 	} else if (ltype != F_UNLCK) {
6129 		rlp = malloc(sizeof (struct nfsrollback), M_NFSDROLLBACK,
6130 		    M_WAITOK);
6131 		rlp->rlck_first = first;
6132 		rlp->rlck_end = end;
6133 		rlp->rlck_type = oldltype;
6134 		LIST_INSERT_HEAD(&lfp->lf_rollback, rlp, rlck_list);
6135 	}
6136 
6137 out:
6138 	NFSEXITCODE(error);
6139 	return (error);
6140 }
6141 
6142 /*
6143  * Roll back local lock changes and free up the rollback list.
6144  */
6145 static void
6146 nfsrv_locallock_rollback(vnode_t vp, struct nfslockfile *lfp, NFSPROC_T *p)
6147 {
6148 	struct nfsrollback *rlp, *nrlp;
6149 
6150 	LIST_FOREACH_SAFE(rlp, &lfp->lf_rollback, rlck_list, nrlp) {
6151 		(void) nfsvno_advlock(vp, rlp->rlck_type, rlp->rlck_first,
6152 		    rlp->rlck_end, p);
6153 		free(rlp, M_NFSDROLLBACK);
6154 	}
6155 	LIST_INIT(&lfp->lf_rollback);
6156 }
6157 
6158 /*
6159  * Update local lock list and delete rollback list (ie now committed to the
6160  * local locks). Most of the work is done by the internal function.
6161  */
6162 static void
6163 nfsrv_locallock_commit(struct nfslockfile *lfp, int flags, uint64_t first,
6164     uint64_t end)
6165 {
6166 	struct nfsrollback *rlp, *nrlp;
6167 	struct nfslock *new_lop, *other_lop;
6168 
6169 	new_lop = malloc(sizeof (struct nfslock), M_NFSDLOCK, M_WAITOK);
6170 	if (flags & (NFSLCK_READ | NFSLCK_WRITE))
6171 		other_lop = malloc(sizeof (struct nfslock), M_NFSDLOCK,
6172 		    M_WAITOK);
6173 	else
6174 		other_lop = NULL;
6175 	new_lop->lo_flags = flags;
6176 	new_lop->lo_first = first;
6177 	new_lop->lo_end = end;
6178 	nfsrv_updatelock(NULL, &new_lop, &other_lop, lfp);
6179 	if (new_lop != NULL)
6180 		free(new_lop, M_NFSDLOCK);
6181 	if (other_lop != NULL)
6182 		free(other_lop, M_NFSDLOCK);
6183 
6184 	/* and get rid of the rollback list */
6185 	LIST_FOREACH_SAFE(rlp, &lfp->lf_rollback, rlck_list, nrlp)
6186 		free(rlp, M_NFSDROLLBACK);
6187 	LIST_INIT(&lfp->lf_rollback);
6188 }
6189 
6190 /*
6191  * Lock the struct nfslockfile for local lock updating.
6192  */
6193 static void
6194 nfsrv_locklf(struct nfslockfile *lfp)
6195 {
6196 	int gotlock;
6197 
6198 	/* lf_usecount ensures *lfp won't be free'd */
6199 	lfp->lf_usecount++;
6200 	do {
6201 		gotlock = nfsv4_lock(&lfp->lf_locallock_lck, 1, NULL,
6202 		    NFSSTATEMUTEXPTR, NULL);
6203 	} while (gotlock == 0);
6204 	lfp->lf_usecount--;
6205 }
6206 
6207 /*
6208  * Unlock the struct nfslockfile after local lock updating.
6209  */
6210 static void
6211 nfsrv_unlocklf(struct nfslockfile *lfp)
6212 {
6213 
6214 	nfsv4_unlock(&lfp->lf_locallock_lck, 0);
6215 }
6216 
6217 /*
6218  * Clear out all state for the NFSv4 server.
6219  * Must be called by a thread that can sleep when no nfsds are running.
6220  */
6221 void
6222 nfsrv_throwawayallstate(NFSPROC_T *p)
6223 {
6224 	struct nfsclient *clp, *nclp;
6225 	struct nfslockfile *lfp, *nlfp;
6226 	int i;
6227 
6228 	/*
6229 	 * For each client, clean out the state and then free the structure.
6230 	 */
6231 	for (i = 0; i < nfsrv_clienthashsize; i++) {
6232 		LIST_FOREACH_SAFE(clp, &nfsclienthash[i], lc_hash, nclp) {
6233 			nfsrv_cleanclient(clp, p);
6234 			nfsrv_freedeleglist(&clp->lc_deleg);
6235 			nfsrv_freedeleglist(&clp->lc_olddeleg);
6236 			free(clp->lc_stateid, M_NFSDCLIENT);
6237 			free(clp, M_NFSDCLIENT);
6238 		}
6239 	}
6240 
6241 	/*
6242 	 * Also, free up any remaining lock file structures.
6243 	 */
6244 	for (i = 0; i < nfsrv_lockhashsize; i++) {
6245 		LIST_FOREACH_SAFE(lfp, &nfslockhash[i], lf_hash, nlfp) {
6246 			printf("nfsd unload: fnd a lock file struct\n");
6247 			nfsrv_freenfslockfile(lfp);
6248 		}
6249 	}
6250 
6251 	/* And get rid of the deviceid structures and layouts. */
6252 	nfsrv_freealllayoutsanddevids();
6253 }
6254 
6255 /*
6256  * Check the sequence# for the session and slot provided as an argument.
6257  * Also, renew the lease if the session will return NFS_OK.
6258  */
6259 int
6260 nfsrv_checksequence(struct nfsrv_descript *nd, uint32_t sequenceid,
6261     uint32_t *highest_slotidp, uint32_t *target_highest_slotidp, int cache_this,
6262     uint32_t *sflagsp, NFSPROC_T *p)
6263 {
6264 	struct nfsdsession *sep;
6265 	struct nfssessionhash *shp;
6266 	int error;
6267 
6268 	shp = NFSSESSIONHASH(nd->nd_sessionid);
6269 	NFSLOCKSESSION(shp);
6270 	sep = nfsrv_findsession(nd->nd_sessionid);
6271 	if (sep == NULL) {
6272 		NFSUNLOCKSESSION(shp);
6273 		return (NFSERR_BADSESSION);
6274 	}
6275 	error = nfsv4_seqsession(sequenceid, nd->nd_slotid, *highest_slotidp,
6276 	    sep->sess_slots, NULL, NFSV4_SLOTS - 1);
6277 	if (error != 0) {
6278 		NFSUNLOCKSESSION(shp);
6279 		return (error);
6280 	}
6281 	if (cache_this != 0)
6282 		nd->nd_flag |= ND_SAVEREPLY;
6283 	/* Renew the lease. */
6284 	sep->sess_clp->lc_expiry = nfsrv_leaseexpiry();
6285 	nd->nd_clientid.qval = sep->sess_clp->lc_clientid.qval;
6286 	nd->nd_flag |= ND_IMPLIEDCLID;
6287 
6288 	/* Save maximum request and reply sizes. */
6289 	nd->nd_maxreq = sep->sess_maxreq;
6290 	nd->nd_maxresp = sep->sess_maxresp;
6291 
6292 	*sflagsp = 0;
6293 	if (sep->sess_clp->lc_req.nr_client == NULL ||
6294 	    (sep->sess_clp->lc_flags & LCL_CBDOWN) != 0)
6295 		*sflagsp |= NFSV4SEQ_CBPATHDOWN;
6296 	NFSUNLOCKSESSION(shp);
6297 	if (error == NFSERR_EXPIRED) {
6298 		*sflagsp |= NFSV4SEQ_EXPIREDALLSTATEREVOKED;
6299 		error = 0;
6300 	} else if (error == NFSERR_ADMINREVOKED) {
6301 		*sflagsp |= NFSV4SEQ_ADMINSTATEREVOKED;
6302 		error = 0;
6303 	}
6304 	*highest_slotidp = *target_highest_slotidp = NFSV4_SLOTS - 1;
6305 	return (0);
6306 }
6307 
6308 /*
6309  * Check/set reclaim complete for this session/clientid.
6310  */
6311 int
6312 nfsrv_checkreclaimcomplete(struct nfsrv_descript *nd, int onefs)
6313 {
6314 	struct nfsdsession *sep;
6315 	struct nfssessionhash *shp;
6316 	int error = 0;
6317 
6318 	shp = NFSSESSIONHASH(nd->nd_sessionid);
6319 	NFSLOCKSTATE();
6320 	NFSLOCKSESSION(shp);
6321 	sep = nfsrv_findsession(nd->nd_sessionid);
6322 	if (sep == NULL) {
6323 		NFSUNLOCKSESSION(shp);
6324 		NFSUNLOCKSTATE();
6325 		return (NFSERR_BADSESSION);
6326 	}
6327 
6328 	if (onefs != 0)
6329 		sep->sess_clp->lc_flags |= LCL_RECLAIMONEFS;
6330 		/* Check to see if reclaim complete has already happened. */
6331 	else if ((sep->sess_clp->lc_flags & LCL_RECLAIMCOMPLETE) != 0)
6332 		error = NFSERR_COMPLETEALREADY;
6333 	else {
6334 		sep->sess_clp->lc_flags |= LCL_RECLAIMCOMPLETE;
6335 		nfsrv_markreclaim(sep->sess_clp);
6336 	}
6337 	NFSUNLOCKSESSION(shp);
6338 	NFSUNLOCKSTATE();
6339 	return (error);
6340 }
6341 
6342 /*
6343  * Cache the reply in a session slot.
6344  */
6345 void
6346 nfsrv_cache_session(struct nfsrv_descript *nd, struct mbuf **m)
6347 {
6348 	struct nfsdsession *sep;
6349 	struct nfssessionhash *shp;
6350 	char *buf, *cp;
6351 #ifdef INET
6352 	struct sockaddr_in *sin;
6353 #endif
6354 #ifdef INET6
6355 	struct sockaddr_in6 *sin6;
6356 #endif
6357 
6358 	shp = NFSSESSIONHASH(nd->nd_sessionid);
6359 	NFSLOCKSESSION(shp);
6360 	sep = nfsrv_findsession(nd->nd_sessionid);
6361 	if (sep == NULL) {
6362 		NFSUNLOCKSESSION(shp);
6363 		if ((nfsrv_stablefirst.nsf_flags & NFSNSF_GRACEOVER) != 0) {
6364 			buf = malloc(INET6_ADDRSTRLEN, M_TEMP, M_WAITOK);
6365 			switch (nd->nd_nam->sa_family) {
6366 #ifdef INET
6367 			case AF_INET:
6368 				sin = (struct sockaddr_in *)nd->nd_nam;
6369 				cp = inet_ntop(sin->sin_family,
6370 				    &sin->sin_addr.s_addr, buf,
6371 				    INET6_ADDRSTRLEN);
6372 				break;
6373 #endif
6374 #ifdef INET6
6375 			case AF_INET6:
6376 				sin6 = (struct sockaddr_in6 *)nd->nd_nam;
6377 				cp = inet_ntop(sin6->sin6_family,
6378 				    &sin6->sin6_addr, buf, INET6_ADDRSTRLEN);
6379 				break;
6380 #endif
6381 			default:
6382 				cp = NULL;
6383 			}
6384 			if (cp != NULL)
6385 				printf("nfsrv_cache_session: no session "
6386 				    "IPaddr=%s, check NFS clients for unique "
6387 				    "/etc/hostid's\n", cp);
6388 			else
6389 				printf("nfsrv_cache_session: no session, "
6390 				    "check NFS clients for unique "
6391 				    "/etc/hostid's\n");
6392 			free(buf, M_TEMP);
6393 		}
6394 		m_freem(*m);
6395 		return;
6396 	}
6397 	nfsv4_seqsess_cacherep(nd->nd_slotid, sep->sess_slots, nd->nd_repstat,
6398 	    m);
6399 	NFSUNLOCKSESSION(shp);
6400 }
6401 
6402 /*
6403  * Search for a session that matches the sessionid.
6404  */
6405 static struct nfsdsession *
6406 nfsrv_findsession(uint8_t *sessionid)
6407 {
6408 	struct nfsdsession *sep;
6409 	struct nfssessionhash *shp;
6410 
6411 	shp = NFSSESSIONHASH(sessionid);
6412 	LIST_FOREACH(sep, &shp->list, sess_hash) {
6413 		if (!NFSBCMP(sessionid, sep->sess_sessionid, NFSX_V4SESSIONID))
6414 			break;
6415 	}
6416 	return (sep);
6417 }
6418 
6419 /*
6420  * Destroy a session.
6421  */
6422 int
6423 nfsrv_destroysession(struct nfsrv_descript *nd, uint8_t *sessionid)
6424 {
6425 	int error, igotlock, samesess;
6426 
6427 	samesess = 0;
6428 	if (!NFSBCMP(sessionid, nd->nd_sessionid, NFSX_V4SESSIONID) &&
6429 	    (nd->nd_flag & ND_HASSEQUENCE) != 0) {
6430 		samesess = 1;
6431 		if ((nd->nd_flag & ND_LASTOP) == 0)
6432 			return (NFSERR_BADSESSION);
6433 	}
6434 
6435 	/* Lock out other nfsd threads */
6436 	NFSLOCKV4ROOTMUTEX();
6437 	nfsv4_relref(&nfsv4rootfs_lock);
6438 	do {
6439 		igotlock = nfsv4_lock(&nfsv4rootfs_lock, 1, NULL,
6440 		    NFSV4ROOTLOCKMUTEXPTR, NULL);
6441 	} while (igotlock == 0);
6442 	NFSUNLOCKV4ROOTMUTEX();
6443 
6444 	error = nfsrv_freesession(NULL, sessionid);
6445 	if (error == 0 && samesess != 0)
6446 		nd->nd_flag &= ~ND_HASSEQUENCE;
6447 
6448 	NFSLOCKV4ROOTMUTEX();
6449 	nfsv4_unlock(&nfsv4rootfs_lock, 1);
6450 	NFSUNLOCKV4ROOTMUTEX();
6451 	return (error);
6452 }
6453 
6454 /*
6455  * Bind a connection to a session.
6456  * For now, only certain variants are supported, since the current session
6457  * structure can only handle a single backchannel entry, which will be
6458  * applied to all connections if it is set.
6459  */
6460 int
6461 nfsrv_bindconnsess(struct nfsrv_descript *nd, uint8_t *sessionid, int *foreaftp)
6462 {
6463 	struct nfssessionhash *shp;
6464 	struct nfsdsession *sep;
6465 	struct nfsclient *clp;
6466 	SVCXPRT *savxprt;
6467 	int error;
6468 
6469 	error = 0;
6470 	savxprt = NULL;
6471 	shp = NFSSESSIONHASH(sessionid);
6472 	NFSLOCKSTATE();
6473 	NFSLOCKSESSION(shp);
6474 	sep = nfsrv_findsession(sessionid);
6475 	if (sep != NULL) {
6476 		clp = sep->sess_clp;
6477 		if (*foreaftp == NFSCDFC4_BACK ||
6478 		    *foreaftp == NFSCDFC4_BACK_OR_BOTH ||
6479 		    *foreaftp == NFSCDFC4_FORE_OR_BOTH) {
6480 			/* Try to set up a backchannel. */
6481 			if (clp->lc_req.nr_client == NULL) {
6482 				NFSD_DEBUG(2, "nfsrv_bindconnsess: acquire "
6483 				    "backchannel\n");
6484 				clp->lc_req.nr_client = (struct __rpc_client *)
6485 				    clnt_bck_create(nd->nd_xprt->xp_socket,
6486 				    sep->sess_cbprogram, NFSV4_CBVERS);
6487 			}
6488 			if (clp->lc_req.nr_client != NULL) {
6489 				NFSD_DEBUG(2, "nfsrv_bindconnsess: set up "
6490 				    "backchannel\n");
6491 				savxprt = sep->sess_cbsess.nfsess_xprt;
6492 				SVC_ACQUIRE(nd->nd_xprt);
6493 				CLNT_ACQUIRE(clp->lc_req.nr_client);
6494 				nd->nd_xprt->xp_p2 = clp->lc_req.nr_client;
6495 				/* Disable idle timeout. */
6496 				nd->nd_xprt->xp_idletimeout = 0;
6497 				sep->sess_cbsess.nfsess_xprt = nd->nd_xprt;
6498 				sep->sess_crflags |= NFSV4CRSESS_CONNBACKCHAN;
6499 				clp->lc_flags |= LCL_DONEBINDCONN |
6500 				    LCL_NEEDSCBNULL;
6501 				clp->lc_flags &= ~LCL_CBDOWN;
6502 				if (*foreaftp == NFSCDFS4_BACK)
6503 					*foreaftp = NFSCDFS4_BACK;
6504 				else
6505 					*foreaftp = NFSCDFS4_BOTH;
6506 			} else if (*foreaftp != NFSCDFC4_BACK) {
6507 				NFSD_DEBUG(2, "nfsrv_bindconnsess: can't set "
6508 				    "up backchannel\n");
6509 				sep->sess_crflags &= ~NFSV4CRSESS_CONNBACKCHAN;
6510 				clp->lc_flags |= LCL_DONEBINDCONN;
6511 				*foreaftp = NFSCDFS4_FORE;
6512 			} else {
6513 				error = NFSERR_NOTSUPP;
6514 				printf("nfsrv_bindconnsess: Can't add "
6515 				    "backchannel\n");
6516 			}
6517 		} else {
6518 			NFSD_DEBUG(2, "nfsrv_bindconnsess: Set forechannel\n");
6519 			clp->lc_flags |= LCL_DONEBINDCONN;
6520 			*foreaftp = NFSCDFS4_FORE;
6521 		}
6522 	} else
6523 		error = NFSERR_BADSESSION;
6524 	NFSUNLOCKSESSION(shp);
6525 	NFSUNLOCKSTATE();
6526 	if (savxprt != NULL)
6527 		SVC_RELEASE(savxprt);
6528 	return (error);
6529 }
6530 
6531 /*
6532  * Free up a session structure.
6533  */
6534 static int
6535 nfsrv_freesession(struct nfsdsession *sep, uint8_t *sessionid)
6536 {
6537 	struct nfssessionhash *shp;
6538 	int i;
6539 
6540 	NFSLOCKSTATE();
6541 	if (sep == NULL) {
6542 		shp = NFSSESSIONHASH(sessionid);
6543 		NFSLOCKSESSION(shp);
6544 		sep = nfsrv_findsession(sessionid);
6545 	} else {
6546 		shp = NFSSESSIONHASH(sep->sess_sessionid);
6547 		NFSLOCKSESSION(shp);
6548 	}
6549 	if (sep != NULL) {
6550 		sep->sess_refcnt--;
6551 		if (sep->sess_refcnt > 0) {
6552 			NFSUNLOCKSESSION(shp);
6553 			NFSUNLOCKSTATE();
6554 			return (NFSERR_BACKCHANBUSY);
6555 		}
6556 		LIST_REMOVE(sep, sess_hash);
6557 		LIST_REMOVE(sep, sess_list);
6558 	}
6559 	NFSUNLOCKSESSION(shp);
6560 	NFSUNLOCKSTATE();
6561 	if (sep == NULL)
6562 		return (NFSERR_BADSESSION);
6563 	for (i = 0; i < NFSV4_SLOTS; i++)
6564 		if (sep->sess_slots[i].nfssl_reply != NULL)
6565 			m_freem(sep->sess_slots[i].nfssl_reply);
6566 	if (sep->sess_cbsess.nfsess_xprt != NULL)
6567 		SVC_RELEASE(sep->sess_cbsess.nfsess_xprt);
6568 	free(sep, M_NFSDSESSION);
6569 	return (0);
6570 }
6571 
6572 /*
6573  * Free a stateid.
6574  * RFC5661 says that it should fail when there are associated opens, locks
6575  * or delegations. Since stateids represent opens, I don't see how you can
6576  * free an open stateid (it will be free'd when closed), so this function
6577  * only works for lock stateids (freeing the lock_owner) or delegations.
6578  */
6579 int
6580 nfsrv_freestateid(struct nfsrv_descript *nd, nfsv4stateid_t *stateidp,
6581     NFSPROC_T *p)
6582 {
6583 	struct nfsclient *clp;
6584 	struct nfsstate *stp;
6585 	int error;
6586 
6587 	NFSLOCKSTATE();
6588 	/*
6589 	 * Look up the stateid
6590 	 */
6591 	error = nfsrv_getclient((nfsquad_t)((u_quad_t)0), CLOPS_RENEW, &clp,
6592 	    NULL, (nfsquad_t)((u_quad_t)0), 0, nd, p);
6593 	if (error == 0) {
6594 		/* First, check for a delegation. */
6595 		LIST_FOREACH(stp, &clp->lc_deleg, ls_list) {
6596 			if (!NFSBCMP(stp->ls_stateid.other, stateidp->other,
6597 			    NFSX_STATEIDOTHER))
6598 				break;
6599 		}
6600 		if (stp != NULL) {
6601 			nfsrv_freedeleg(stp);
6602 			NFSUNLOCKSTATE();
6603 			return (error);
6604 		}
6605 	}
6606 	/* Not a delegation, try for a lock_owner. */
6607 	if (error == 0)
6608 		error = nfsrv_getstate(clp, stateidp, 0, &stp);
6609 	if (error == 0 && ((stp->ls_flags & (NFSLCK_OPEN | NFSLCK_DELEGREAD |
6610 	    NFSLCK_DELEGWRITE)) != 0 || (stp->ls_flags & NFSLCK_LOCK) == 0))
6611 		/* Not a lock_owner stateid. */
6612 		error = NFSERR_LOCKSHELD;
6613 	if (error == 0 && !LIST_EMPTY(&stp->ls_lock))
6614 		error = NFSERR_LOCKSHELD;
6615 	if (error == 0)
6616 		nfsrv_freelockowner(stp, NULL, 0, p);
6617 	NFSUNLOCKSTATE();
6618 	return (error);
6619 }
6620 
6621 /*
6622  * Test a stateid.
6623  */
6624 int
6625 nfsrv_teststateid(struct nfsrv_descript *nd, nfsv4stateid_t *stateidp,
6626     NFSPROC_T *p)
6627 {
6628 	struct nfsclient *clp;
6629 	struct nfsstate *stp;
6630 	int error;
6631 
6632 	NFSLOCKSTATE();
6633 	/*
6634 	 * Look up the stateid
6635 	 */
6636 	error = nfsrv_getclient((nfsquad_t)((u_quad_t)0), CLOPS_RENEW, &clp,
6637 	    NULL, (nfsquad_t)((u_quad_t)0), 0, nd, p);
6638 	if (error == 0)
6639 		error = nfsrv_getstate(clp, stateidp, 0, &stp);
6640 	if (error == 0 && stateidp->seqid != 0 &&
6641 	    SEQ_LT(stateidp->seqid, stp->ls_stateid.seqid))
6642 		error = NFSERR_OLDSTATEID;
6643 	NFSUNLOCKSTATE();
6644 	return (error);
6645 }
6646 
6647 /*
6648  * Generate the xdr for an NFSv4.1 CBSequence Operation.
6649  */
6650 static int
6651 nfsv4_setcbsequence(struct nfsrv_descript *nd, struct nfsclient *clp,
6652     int dont_replycache, struct nfsdsession **sepp, int *slotposp)
6653 {
6654 	struct nfsdsession *sep;
6655 	uint32_t *tl, slotseq = 0;
6656 	int maxslot;
6657 	uint8_t sessionid[NFSX_V4SESSIONID];
6658 	int error;
6659 
6660 	error = nfsv4_getcbsession(clp, sepp);
6661 	if (error != 0)
6662 		return (error);
6663 	sep = *sepp;
6664 	nfsv4_sequencelookup(NULL, &sep->sess_cbsess, slotposp, &maxslot,
6665 	    &slotseq, sessionid, true);
6666 	KASSERT(maxslot >= 0, ("nfsv4_setcbsequence neg maxslot"));
6667 
6668 	/* Build the Sequence arguments. */
6669 	NFSM_BUILD(tl, uint32_t *, NFSX_V4SESSIONID + 5 * NFSX_UNSIGNED);
6670 	bcopy(sessionid, tl, NFSX_V4SESSIONID);
6671 	tl += NFSX_V4SESSIONID / NFSX_UNSIGNED;
6672 	nd->nd_slotseq = tl;
6673 	nd->nd_slotid = *slotposp;
6674 	nd->nd_flag |= ND_HASSLOTID;
6675 	*tl++ = txdr_unsigned(slotseq);
6676 	*tl++ = txdr_unsigned(*slotposp);
6677 	*tl++ = txdr_unsigned(maxslot);
6678 	if (dont_replycache == 0)
6679 		*tl++ = newnfs_true;
6680 	else
6681 		*tl++ = newnfs_false;
6682 	*tl = 0;			/* No referring call list, for now. */
6683 	nd->nd_flag |= ND_HASSEQUENCE;
6684 	return (0);
6685 }
6686 
6687 /*
6688  * Get a session for the callback.
6689  */
6690 static int
6691 nfsv4_getcbsession(struct nfsclient *clp, struct nfsdsession **sepp)
6692 {
6693 	struct nfsdsession *sep;
6694 
6695 	NFSLOCKSTATE();
6696 	LIST_FOREACH(sep, &clp->lc_session, sess_list) {
6697 		if ((sep->sess_crflags & NFSV4CRSESS_CONNBACKCHAN) != 0)
6698 			break;
6699 	}
6700 	if (sep == NULL) {
6701 		NFSUNLOCKSTATE();
6702 		return (NFSERR_BADSESSION);
6703 	}
6704 	sep->sess_refcnt++;
6705 	*sepp = sep;
6706 	NFSUNLOCKSTATE();
6707 	return (0);
6708 }
6709 
6710 /*
6711  * Free up all backchannel xprts.  This needs to be done when the nfsd threads
6712  * exit, since those transports will all be going away.
6713  * This is only called after all the nfsd threads are done performing RPCs,
6714  * so locking shouldn't be an issue.
6715  */
6716 void
6717 nfsrv_freeallbackchannel_xprts(void)
6718 {
6719 	struct nfsdsession *sep;
6720 	struct nfsclient *clp;
6721 	SVCXPRT *xprt;
6722 	int i;
6723 
6724 	for (i = 0; i < nfsrv_clienthashsize; i++) {
6725 		LIST_FOREACH(clp, &nfsclienthash[i], lc_hash) {
6726 			LIST_FOREACH(sep, &clp->lc_session, sess_list) {
6727 				xprt = sep->sess_cbsess.nfsess_xprt;
6728 				sep->sess_cbsess.nfsess_xprt = NULL;
6729 				if (xprt != NULL)
6730 					SVC_RELEASE(xprt);
6731 			}
6732 		}
6733 	}
6734 }
6735 
6736 /*
6737  * Do a layout commit.  Actually just call nfsrv_updatemdsattr().
6738  * I have no idea if the rest of these arguments will ever be useful?
6739  */
6740 int
6741 nfsrv_layoutcommit(struct nfsrv_descript *nd, vnode_t vp, int layouttype,
6742     int hasnewoff, uint64_t newoff, uint64_t offset, uint64_t len,
6743     int hasnewmtime, struct timespec *newmtimep, int reclaim,
6744     nfsv4stateid_t *stateidp, int maxcnt, char *layp, int *hasnewsizep,
6745     uint64_t *newsizep, struct ucred *cred, NFSPROC_T *p)
6746 {
6747 	struct nfsvattr na;
6748 	int error;
6749 
6750 	error = nfsrv_updatemdsattr(vp, &na, p);
6751 	if (error == 0) {
6752 		*hasnewsizep = 1;
6753 		*newsizep = na.na_size;
6754 	}
6755 	return (error);
6756 }
6757 
6758 /*
6759  * Try and get a layout.
6760  */
6761 int
6762 nfsrv_layoutget(struct nfsrv_descript *nd, vnode_t vp, struct nfsexstuff *exp,
6763     int layouttype, int *iomode, uint64_t *offset, uint64_t *len,
6764     uint64_t minlen, nfsv4stateid_t *stateidp, int maxcnt, int *retonclose,
6765     int *layoutlenp, char *layp, struct ucred *cred, NFSPROC_T *p)
6766 {
6767 	struct nfslayouthash *lhyp;
6768 	struct nfslayout *lyp;
6769 	char *devid;
6770 	fhandle_t fh, *dsfhp;
6771 	int error, mirrorcnt;
6772 
6773 	if (nfsrv_devidcnt == 0)
6774 		return (NFSERR_UNKNLAYOUTTYPE);
6775 
6776 	if (*offset != 0)
6777 		printf("nfsrv_layoutget: off=%ju len=%ju\n", (uintmax_t)*offset,
6778 		    (uintmax_t)*len);
6779 	error = nfsvno_getfh(vp, &fh, p);
6780 	NFSD_DEBUG(4, "layoutget getfh=%d\n", error);
6781 	if (error != 0)
6782 		return (error);
6783 
6784 	/*
6785 	 * For now, all layouts are for entire files.
6786 	 * Only issue Read/Write layouts if requested for a non-readonly fs.
6787 	 */
6788 	if (NFSVNO_EXRDONLY(exp)) {
6789 		if (*iomode == NFSLAYOUTIOMODE_RW)
6790 			return (NFSERR_LAYOUTTRYLATER);
6791 		*iomode = NFSLAYOUTIOMODE_READ;
6792 	}
6793 	if (*iomode != NFSLAYOUTIOMODE_RW)
6794 		*iomode = NFSLAYOUTIOMODE_READ;
6795 
6796 	/*
6797 	 * Check to see if a write layout can be issued for this file.
6798 	 * This is used during mirror recovery to avoid RW layouts being
6799 	 * issued for a file while it is being copied to the recovered
6800 	 * mirror.
6801 	 */
6802 	if (*iomode == NFSLAYOUTIOMODE_RW && nfsrv_dontlayout(&fh) != 0)
6803 		return (NFSERR_LAYOUTTRYLATER);
6804 
6805 	*retonclose = 0;
6806 	*offset = 0;
6807 	*len = UINT64_MAX;
6808 
6809 	/* First, see if a layout already exists and return if found. */
6810 	lhyp = NFSLAYOUTHASH(&fh);
6811 	NFSLOCKLAYOUT(lhyp);
6812 	error = nfsrv_findlayout(&nd->nd_clientid, &fh, layouttype, p, &lyp);
6813 	NFSD_DEBUG(4, "layoutget findlay=%d\n", error);
6814 	/*
6815 	 * Not sure if the seqid must be the same, so I won't check it.
6816 	 */
6817 	if (error == 0 && (stateidp->other[0] != lyp->lay_stateid.other[0] ||
6818 	    stateidp->other[1] != lyp->lay_stateid.other[1] ||
6819 	    stateidp->other[2] != lyp->lay_stateid.other[2])) {
6820 		if ((lyp->lay_flags & NFSLAY_CALLB) == 0) {
6821 			NFSUNLOCKLAYOUT(lhyp);
6822 			NFSD_DEBUG(1, "ret bad stateid\n");
6823 			return (NFSERR_BADSTATEID);
6824 		}
6825 		/*
6826 		 * I believe we get here because there is a race between
6827 		 * the client processing the CBLAYOUTRECALL and the layout
6828 		 * being deleted here on the server.
6829 		 * The client has now done a LayoutGet with a non-layout
6830 		 * stateid, as it would when there is no layout.
6831 		 * As such, free this layout and set error == NFSERR_BADSTATEID
6832 		 * so the code below will create a new layout structure as
6833 		 * would happen if no layout was found.
6834 		 * "lyp" will be set before being used below, but set it NULL
6835 		 * as a safety belt.
6836 		 */
6837 		nfsrv_freelayout(&lhyp->list, lyp);
6838 		lyp = NULL;
6839 		error = NFSERR_BADSTATEID;
6840 	}
6841 	if (error == 0) {
6842 		if (lyp->lay_layoutlen > maxcnt) {
6843 			NFSUNLOCKLAYOUT(lhyp);
6844 			NFSD_DEBUG(1, "ret layout too small\n");
6845 			return (NFSERR_TOOSMALL);
6846 		}
6847 		if (*iomode == NFSLAYOUTIOMODE_RW) {
6848 			if ((lyp->lay_flags & NFSLAY_NOSPC) != 0) {
6849 				NFSUNLOCKLAYOUT(lhyp);
6850 				NFSD_DEBUG(1, "ret layout nospace\n");
6851 				return (NFSERR_NOSPC);
6852 			}
6853 			lyp->lay_flags |= NFSLAY_RW;
6854 		} else
6855 			lyp->lay_flags |= NFSLAY_READ;
6856 		NFSBCOPY(lyp->lay_xdr, layp, lyp->lay_layoutlen);
6857 		*layoutlenp = lyp->lay_layoutlen;
6858 		if (++lyp->lay_stateid.seqid == 0)
6859 			lyp->lay_stateid.seqid = 1;
6860 		stateidp->seqid = lyp->lay_stateid.seqid;
6861 		NFSUNLOCKLAYOUT(lhyp);
6862 		NFSD_DEBUG(4, "ret fnd layout\n");
6863 		return (0);
6864 	}
6865 	NFSUNLOCKLAYOUT(lhyp);
6866 
6867 	/* Find the device id and file handle. */
6868 	dsfhp = malloc(sizeof(fhandle_t) * NFSDEV_MAXMIRRORS, M_TEMP, M_WAITOK);
6869 	devid = malloc(NFSX_V4DEVICEID * NFSDEV_MAXMIRRORS, M_TEMP, M_WAITOK);
6870 	error = nfsrv_dsgetdevandfh(vp, p, &mirrorcnt, dsfhp, devid);
6871 	NFSD_DEBUG(4, "layoutget devandfh=%d\n", error);
6872 	if (error == 0) {
6873 		if (layouttype == NFSLAYOUT_NFSV4_1_FILES) {
6874 			if (NFSX_V4FILELAYOUT > maxcnt)
6875 				error = NFSERR_TOOSMALL;
6876 			else
6877 				lyp = nfsrv_filelayout(nd, *iomode, &fh, dsfhp,
6878 				    devid, vp->v_mount->mnt_stat.f_fsid);
6879 		} else {
6880 			if (NFSX_V4FLEXLAYOUT(mirrorcnt) > maxcnt)
6881 				error = NFSERR_TOOSMALL;
6882 			else
6883 				lyp = nfsrv_flexlayout(nd, *iomode, mirrorcnt,
6884 				    &fh, dsfhp, devid,
6885 				    vp->v_mount->mnt_stat.f_fsid);
6886 		}
6887 	}
6888 	free(dsfhp, M_TEMP);
6889 	free(devid, M_TEMP);
6890 	if (error != 0)
6891 		return (error);
6892 
6893 	/*
6894 	 * Now, add this layout to the list.
6895 	 */
6896 	error = nfsrv_addlayout(nd, &lyp, stateidp, layp, layoutlenp, p);
6897 	NFSD_DEBUG(4, "layoutget addl=%d\n", error);
6898 	/*
6899 	 * The lyp will be set to NULL by nfsrv_addlayout() if it
6900 	 * linked the new structure into the lists.
6901 	 */
6902 	free(lyp, M_NFSDSTATE);
6903 	return (error);
6904 }
6905 
6906 /*
6907  * Generate a File Layout.
6908  */
6909 static struct nfslayout *
6910 nfsrv_filelayout(struct nfsrv_descript *nd, int iomode, fhandle_t *fhp,
6911     fhandle_t *dsfhp, char *devid, fsid_t fs)
6912 {
6913 	uint32_t *tl;
6914 	struct nfslayout *lyp;
6915 	uint64_t pattern_offset;
6916 
6917 	lyp = malloc(sizeof(struct nfslayout) + NFSX_V4FILELAYOUT, M_NFSDSTATE,
6918 	    M_WAITOK | M_ZERO);
6919 	lyp->lay_type = NFSLAYOUT_NFSV4_1_FILES;
6920 	if (iomode == NFSLAYOUTIOMODE_RW)
6921 		lyp->lay_flags = NFSLAY_RW;
6922 	else
6923 		lyp->lay_flags = NFSLAY_READ;
6924 	NFSBCOPY(fhp, &lyp->lay_fh, sizeof(*fhp));
6925 	lyp->lay_clientid.qval = nd->nd_clientid.qval;
6926 	lyp->lay_fsid = fs;
6927 	NFSBCOPY(devid, lyp->lay_deviceid, NFSX_V4DEVICEID);
6928 
6929 	/* Fill in the xdr for the files layout. */
6930 	tl = (uint32_t *)lyp->lay_xdr;
6931 	NFSBCOPY(devid, tl, NFSX_V4DEVICEID);		/* Device ID. */
6932 	tl += (NFSX_V4DEVICEID / NFSX_UNSIGNED);
6933 
6934 	/* Set the stripe size to the maximum I/O size. */
6935 	*tl++ = txdr_unsigned(nfs_srvmaxio & NFSFLAYUTIL_STRIPE_MASK);
6936 	*tl++ = 0;					/* 1st stripe index. */
6937 	pattern_offset = 0;
6938 	txdr_hyper(pattern_offset, tl); tl += 2;	/* Pattern offset. */
6939 	*tl++ = txdr_unsigned(1);			/* 1 file handle. */
6940 	*tl++ = txdr_unsigned(NFSX_V4PNFSFH);
6941 	NFSBCOPY(dsfhp, tl, sizeof(*dsfhp));
6942 	lyp->lay_layoutlen = NFSX_V4FILELAYOUT;
6943 	return (lyp);
6944 }
6945 
6946 #define	FLEX_OWNERID	"999"
6947 #define	FLEX_UID0	"0"
6948 /*
6949  * Generate a Flex File Layout.
6950  * The FLEX_OWNERID can be any string of 3 decimal digits. Although this
6951  * string goes on the wire, it isn't supposed to be used by the client,
6952  * since this server uses tight coupling.
6953  * Although not recommended by the spec., if vfs.nfsd.flexlinuxhack=1 use
6954  * a string of "0". This works around the Linux Flex File Layout driver bug
6955  * which uses the synthetic uid/gid strings for the "tightly coupled" case.
6956  */
6957 static struct nfslayout *
6958 nfsrv_flexlayout(struct nfsrv_descript *nd, int iomode, int mirrorcnt,
6959     fhandle_t *fhp, fhandle_t *dsfhp, char *devid, fsid_t fs)
6960 {
6961 	uint32_t *tl;
6962 	struct nfslayout *lyp;
6963 	uint64_t lenval;
6964 	int i;
6965 
6966 	lyp = malloc(sizeof(struct nfslayout) + NFSX_V4FLEXLAYOUT(mirrorcnt),
6967 	    M_NFSDSTATE, M_WAITOK | M_ZERO);
6968 	lyp->lay_type = NFSLAYOUT_FLEXFILE;
6969 	if (iomode == NFSLAYOUTIOMODE_RW)
6970 		lyp->lay_flags = NFSLAY_RW;
6971 	else
6972 		lyp->lay_flags = NFSLAY_READ;
6973 	NFSBCOPY(fhp, &lyp->lay_fh, sizeof(*fhp));
6974 	lyp->lay_clientid.qval = nd->nd_clientid.qval;
6975 	lyp->lay_fsid = fs;
6976 	lyp->lay_mirrorcnt = mirrorcnt;
6977 	NFSBCOPY(devid, lyp->lay_deviceid, NFSX_V4DEVICEID);
6978 
6979 	/* Fill in the xdr for the files layout. */
6980 	tl = (uint32_t *)lyp->lay_xdr;
6981 	lenval = 0;
6982 	txdr_hyper(lenval, tl); tl += 2;		/* Stripe unit. */
6983 	*tl++ = txdr_unsigned(mirrorcnt);		/* # of mirrors. */
6984 	for (i = 0; i < mirrorcnt; i++) {
6985 		*tl++ = txdr_unsigned(1);		/* One stripe. */
6986 		NFSBCOPY(devid, tl, NFSX_V4DEVICEID);	/* Device ID. */
6987 		tl += (NFSX_V4DEVICEID / NFSX_UNSIGNED);
6988 		devid += NFSX_V4DEVICEID;
6989 		*tl++ = txdr_unsigned(1);		/* Efficiency. */
6990 		*tl++ = 0;				/* Proxy Stateid. */
6991 		*tl++ = 0x55555555;
6992 		*tl++ = 0x55555555;
6993 		*tl++ = 0x55555555;
6994 		*tl++ = txdr_unsigned(1);		/* 1 file handle. */
6995 		*tl++ = txdr_unsigned(NFSX_V4PNFSFH);
6996 		NFSBCOPY(dsfhp, tl, sizeof(*dsfhp));
6997 		tl += (NFSM_RNDUP(NFSX_V4PNFSFH) / NFSX_UNSIGNED);
6998 		dsfhp++;
6999 		if (nfsrv_flexlinuxhack != 0) {
7000 			*tl++ = txdr_unsigned(strlen(FLEX_UID0));
7001 			*tl = 0;		/* 0 pad string. */
7002 			NFSBCOPY(FLEX_UID0, tl++, strlen(FLEX_UID0));
7003 			*tl++ = txdr_unsigned(strlen(FLEX_UID0));
7004 			*tl = 0;		/* 0 pad string. */
7005 			NFSBCOPY(FLEX_UID0, tl++, strlen(FLEX_UID0));
7006 		} else {
7007 			*tl++ = txdr_unsigned(strlen(FLEX_OWNERID));
7008 			NFSBCOPY(FLEX_OWNERID, tl++, NFSX_UNSIGNED);
7009 			*tl++ = txdr_unsigned(strlen(FLEX_OWNERID));
7010 			NFSBCOPY(FLEX_OWNERID, tl++, NFSX_UNSIGNED);
7011 		}
7012 	}
7013 	*tl++ = txdr_unsigned(0);		/* ff_flags. */
7014 	*tl = txdr_unsigned(60);		/* Status interval hint. */
7015 	lyp->lay_layoutlen = NFSX_V4FLEXLAYOUT(mirrorcnt);
7016 	return (lyp);
7017 }
7018 
7019 /*
7020  * Parse and process Flex File errors returned via LayoutReturn.
7021  */
7022 static void
7023 nfsrv_flexlayouterr(struct nfsrv_descript *nd, uint32_t *layp, int maxcnt,
7024     NFSPROC_T *p)
7025 {
7026 	uint32_t *tl;
7027 	int cnt, errcnt, i, j, opnum, stat;
7028 	char devid[NFSX_V4DEVICEID];
7029 
7030 	tl = layp;
7031 	maxcnt -= NFSX_UNSIGNED;
7032 	if (maxcnt > 0)
7033 		cnt = fxdr_unsigned(int, *tl++);
7034 	else
7035 		cnt = 0;
7036 	NFSD_DEBUG(4, "flexlayouterr cnt=%d\n", cnt);
7037 	for (i = 0; i < cnt; i++) {
7038 		maxcnt -= NFSX_STATEID + 2 * NFSX_HYPER +
7039 		    NFSX_UNSIGNED;
7040 		if (maxcnt <= 0)
7041 			break;
7042 		/* Skip offset, length and stateid for now. */
7043 		tl += (4 + NFSX_STATEID / NFSX_UNSIGNED);
7044 		errcnt = fxdr_unsigned(int, *tl++);
7045 		NFSD_DEBUG(4, "flexlayouterr errcnt=%d\n", errcnt);
7046 		for (j = 0; j < errcnt; j++) {
7047 			maxcnt -= NFSX_V4DEVICEID + 2 * NFSX_UNSIGNED;
7048 			if (maxcnt < 0)
7049 				break;
7050 			NFSBCOPY(tl, devid, NFSX_V4DEVICEID);
7051 			tl += (NFSX_V4DEVICEID / NFSX_UNSIGNED);
7052 			stat = fxdr_unsigned(int, *tl++);
7053 			opnum = fxdr_unsigned(int, *tl++);
7054 			NFSD_DEBUG(4, "flexlayouterr op=%d stat=%d\n", opnum,
7055 			    stat);
7056 			/*
7057 			 * Except for NFSERR_ACCES, NFSERR_STALE and
7058 			 * NFSERR_NOSPC errors, disable the mirror.
7059 			 */
7060 			if (stat != NFSERR_ACCES && stat != NFSERR_STALE &&
7061 			    stat != NFSERR_NOSPC)
7062 				nfsrv_delds(devid, p);
7063 
7064 			/* For NFSERR_NOSPC, mark all devids and layouts. */
7065 			if (stat == NFSERR_NOSPC)
7066 				nfsrv_marknospc(devid, true);
7067 		}
7068 	}
7069 }
7070 
7071 /*
7072  * This function removes all flex file layouts which has a mirror with
7073  * a device id that matches the argument.
7074  * Called when the DS represented by the device id has failed.
7075  */
7076 void
7077 nfsrv_flexmirrordel(char *devid, NFSPROC_T *p)
7078 {
7079 	uint32_t *tl;
7080 	struct nfslayout *lyp, *nlyp;
7081 	struct nfslayouthash *lhyp;
7082 	struct nfslayouthead loclyp;
7083 	int i, j;
7084 
7085 	NFSD_DEBUG(4, "flexmirrordel\n");
7086 	/* Move all layouts found onto a local list. */
7087 	TAILQ_INIT(&loclyp);
7088 	for (i = 0; i < nfsrv_layouthashsize; i++) {
7089 		lhyp = &nfslayouthash[i];
7090 		NFSLOCKLAYOUT(lhyp);
7091 		TAILQ_FOREACH_SAFE(lyp, &lhyp->list, lay_list, nlyp) {
7092 			if (lyp->lay_type == NFSLAYOUT_FLEXFILE &&
7093 			    lyp->lay_mirrorcnt > 1) {
7094 				NFSD_DEBUG(4, "possible match\n");
7095 				tl = lyp->lay_xdr;
7096 				tl += 3;
7097 				for (j = 0; j < lyp->lay_mirrorcnt; j++) {
7098 					tl++;
7099 					if (NFSBCMP(devid, tl, NFSX_V4DEVICEID)
7100 					    == 0) {
7101 						/* Found one. */
7102 						NFSD_DEBUG(4, "fnd one\n");
7103 						TAILQ_REMOVE(&lhyp->list, lyp,
7104 						    lay_list);
7105 						TAILQ_INSERT_HEAD(&loclyp, lyp,
7106 						    lay_list);
7107 						break;
7108 					}
7109 					tl += (NFSX_V4DEVICEID / NFSX_UNSIGNED +
7110 					    NFSM_RNDUP(NFSX_V4PNFSFH) /
7111 					    NFSX_UNSIGNED + 11 * NFSX_UNSIGNED);
7112 				}
7113 			}
7114 		}
7115 		NFSUNLOCKLAYOUT(lhyp);
7116 	}
7117 
7118 	/* Now, try to do a Layout recall for each one found. */
7119 	TAILQ_FOREACH_SAFE(lyp, &loclyp, lay_list, nlyp) {
7120 		NFSD_DEBUG(4, "do layout recall\n");
7121 		/*
7122 		 * The layout stateid.seqid needs to be incremented
7123 		 * before doing a LAYOUT_RECALL callback.
7124 		 */
7125 		if (++lyp->lay_stateid.seqid == 0)
7126 			lyp->lay_stateid.seqid = 1;
7127 		nfsrv_recalllayout(lyp->lay_clientid, &lyp->lay_stateid,
7128 		    &lyp->lay_fh, lyp, 1, lyp->lay_type, p);
7129 		nfsrv_freelayout(&loclyp, lyp);
7130 	}
7131 }
7132 
7133 /*
7134  * Do a recall callback to the client for this layout.
7135  */
7136 static int
7137 nfsrv_recalllayout(nfsquad_t clid, nfsv4stateid_t *stateidp, fhandle_t *fhp,
7138     struct nfslayout *lyp, int changed, int laytype, NFSPROC_T *p)
7139 {
7140 	struct nfsclient *clp;
7141 	int error;
7142 
7143 	NFSD_DEBUG(4, "nfsrv_recalllayout\n");
7144 	error = nfsrv_getclient(clid, 0, &clp, NULL, (nfsquad_t)((u_quad_t)0),
7145 	    0, NULL, p);
7146 	NFSD_DEBUG(4, "aft nfsrv_getclient=%d\n", error);
7147 	if (error != 0) {
7148 		printf("nfsrv_recalllayout: getclient err=%d\n", error);
7149 		return (error);
7150 	}
7151 	if ((clp->lc_flags & LCL_NFSV41) != 0) {
7152 		error = nfsrv_docallback(clp, NFSV4OP_CBLAYOUTRECALL,
7153 		    stateidp, changed, fhp, NULL, NULL, laytype, p);
7154 		/* If lyp != NULL, handle an error return here. */
7155 		if (error != 0 && lyp != NULL) {
7156 			NFSDRECALLLOCK();
7157 			/*
7158 			 * Mark it returned, since no layout recall
7159 			 * has been done.
7160 			 * All errors seem to be non-recoverable, although
7161 			 * NFSERR_NOMATCHLAYOUT is a normal event.
7162 			 */
7163 			if ((lyp->lay_flags & NFSLAY_RECALL) != 0) {
7164 				lyp->lay_flags |= NFSLAY_RETURNED;
7165 				wakeup(lyp);
7166 			}
7167 			NFSDRECALLUNLOCK();
7168 			if (error != NFSERR_NOMATCHLAYOUT)
7169 				printf("nfsrv_recalllayout: err=%d\n", error);
7170 		}
7171 	} else
7172 		printf("nfsrv_recalllayout: clp not NFSv4.1\n");
7173 	return (error);
7174 }
7175 
7176 /*
7177  * Find a layout to recall when we exceed our high water mark.
7178  */
7179 void
7180 nfsrv_recalloldlayout(NFSPROC_T *p)
7181 {
7182 	struct nfslayouthash *lhyp;
7183 	struct nfslayout *lyp;
7184 	nfsquad_t clientid;
7185 	nfsv4stateid_t stateid;
7186 	fhandle_t fh;
7187 	int error, laytype = 0, ret;
7188 
7189 	lhyp = &nfslayouthash[arc4random() % nfsrv_layouthashsize];
7190 	NFSLOCKLAYOUT(lhyp);
7191 	TAILQ_FOREACH_REVERSE(lyp, &lhyp->list, nfslayouthead, lay_list) {
7192 		if ((lyp->lay_flags & NFSLAY_CALLB) == 0) {
7193 			lyp->lay_flags |= NFSLAY_CALLB;
7194 			/*
7195 			 * The layout stateid.seqid needs to be incremented
7196 			 * before doing a LAYOUT_RECALL callback.
7197 			 */
7198 			if (++lyp->lay_stateid.seqid == 0)
7199 				lyp->lay_stateid.seqid = 1;
7200 			clientid = lyp->lay_clientid;
7201 			stateid = lyp->lay_stateid;
7202 			NFSBCOPY(&lyp->lay_fh, &fh, sizeof(fh));
7203 			laytype = lyp->lay_type;
7204 			break;
7205 		}
7206 	}
7207 	NFSUNLOCKLAYOUT(lhyp);
7208 	if (lyp != NULL) {
7209 		error = nfsrv_recalllayout(clientid, &stateid, &fh, NULL, 0,
7210 		    laytype, p);
7211 		if (error != 0 && error != NFSERR_NOMATCHLAYOUT)
7212 			NFSD_DEBUG(4, "recallold=%d\n", error);
7213 		if (error != 0) {
7214 			NFSLOCKLAYOUT(lhyp);
7215 			/*
7216 			 * Since the hash list was unlocked, we need to
7217 			 * find it again.
7218 			 */
7219 			ret = nfsrv_findlayout(&clientid, &fh, laytype, p,
7220 			    &lyp);
7221 			if (ret == 0 &&
7222 			    (lyp->lay_flags & NFSLAY_CALLB) != 0 &&
7223 			    lyp->lay_stateid.other[0] == stateid.other[0] &&
7224 			    lyp->lay_stateid.other[1] == stateid.other[1] &&
7225 			    lyp->lay_stateid.other[2] == stateid.other[2]) {
7226 				/*
7227 				 * The client no longer knows this layout, so
7228 				 * it can be free'd now.
7229 				 */
7230 				if (error == NFSERR_NOMATCHLAYOUT)
7231 					nfsrv_freelayout(&lhyp->list, lyp);
7232 				else {
7233 					/*
7234 					 * Leave it to be tried later by
7235 					 * clearing NFSLAY_CALLB and moving
7236 					 * it to the head of the list, so it
7237 					 * won't be tried again for a while.
7238 					 */
7239 					lyp->lay_flags &= ~NFSLAY_CALLB;
7240 					TAILQ_REMOVE(&lhyp->list, lyp,
7241 					    lay_list);
7242 					TAILQ_INSERT_HEAD(&lhyp->list, lyp,
7243 					    lay_list);
7244 				}
7245 			}
7246 			NFSUNLOCKLAYOUT(lhyp);
7247 		}
7248 	}
7249 }
7250 
7251 /*
7252  * Try and return layout(s).
7253  */
7254 int
7255 nfsrv_layoutreturn(struct nfsrv_descript *nd, vnode_t vp,
7256     int layouttype, int iomode, uint64_t offset, uint64_t len, int reclaim,
7257     int kind, nfsv4stateid_t *stateidp, int maxcnt, uint32_t *layp, int *fndp,
7258     struct ucred *cred, NFSPROC_T *p)
7259 {
7260 	struct nfsvattr na;
7261 	struct nfslayouthash *lhyp;
7262 	struct nfslayout *lyp;
7263 	fhandle_t fh;
7264 	int error = 0;
7265 
7266 	*fndp = 0;
7267 	if (kind == NFSV4LAYOUTRET_FILE) {
7268 		error = nfsvno_getfh(vp, &fh, p);
7269 		if (error == 0) {
7270 			error = nfsrv_updatemdsattr(vp, &na, p);
7271 			if (error != 0)
7272 				printf("nfsrv_layoutreturn: updatemdsattr"
7273 				    " failed=%d\n", error);
7274 		}
7275 		if (error == 0) {
7276 			if (reclaim == newnfs_true) {
7277 				error = nfsrv_checkgrace(NULL, NULL,
7278 				    NFSLCK_RECLAIM);
7279 				if (error != NFSERR_NOGRACE)
7280 					error = 0;
7281 				return (error);
7282 			}
7283 			lhyp = NFSLAYOUTHASH(&fh);
7284 			NFSDRECALLLOCK();
7285 			NFSLOCKLAYOUT(lhyp);
7286 			error = nfsrv_findlayout(&nd->nd_clientid, &fh,
7287 			    layouttype, p, &lyp);
7288 			NFSD_DEBUG(4, "layoutret findlay=%d\n", error);
7289 			if (error == 0 &&
7290 			    stateidp->other[0] == lyp->lay_stateid.other[0] &&
7291 			    stateidp->other[1] == lyp->lay_stateid.other[1] &&
7292 			    stateidp->other[2] == lyp->lay_stateid.other[2]) {
7293 				NFSD_DEBUG(4, "nfsrv_layoutreturn: stateid %d"
7294 				    " %x %x %x laystateid %d %x %x %x"
7295 				    " off=%ju len=%ju flgs=0x%x\n",
7296 				    stateidp->seqid, stateidp->other[0],
7297 				    stateidp->other[1], stateidp->other[2],
7298 				    lyp->lay_stateid.seqid,
7299 				    lyp->lay_stateid.other[0],
7300 				    lyp->lay_stateid.other[1],
7301 				    lyp->lay_stateid.other[2],
7302 				    (uintmax_t)offset, (uintmax_t)len,
7303 				    lyp->lay_flags);
7304 				if (++lyp->lay_stateid.seqid == 0)
7305 					lyp->lay_stateid.seqid = 1;
7306 				stateidp->seqid = lyp->lay_stateid.seqid;
7307 				if (offset == 0 && len == UINT64_MAX) {
7308 					if ((iomode & NFSLAYOUTIOMODE_READ) !=
7309 					    0)
7310 						lyp->lay_flags &= ~NFSLAY_READ;
7311 					if ((iomode & NFSLAYOUTIOMODE_RW) != 0)
7312 						lyp->lay_flags &= ~NFSLAY_RW;
7313 					if ((lyp->lay_flags & (NFSLAY_READ |
7314 					    NFSLAY_RW)) == 0)
7315 						nfsrv_freelayout(&lhyp->list,
7316 						    lyp);
7317 					else
7318 						*fndp = 1;
7319 				} else
7320 					*fndp = 1;
7321 			}
7322 			NFSUNLOCKLAYOUT(lhyp);
7323 			/* Search the nfsrv_recalllist for a match. */
7324 			TAILQ_FOREACH(lyp, &nfsrv_recalllisthead, lay_list) {
7325 				if (NFSBCMP(&lyp->lay_fh, &fh,
7326 				    sizeof(fh)) == 0 &&
7327 				    lyp->lay_clientid.qval ==
7328 				    nd->nd_clientid.qval &&
7329 				    stateidp->other[0] ==
7330 				    lyp->lay_stateid.other[0] &&
7331 				    stateidp->other[1] ==
7332 				    lyp->lay_stateid.other[1] &&
7333 				    stateidp->other[2] ==
7334 				    lyp->lay_stateid.other[2]) {
7335 					lyp->lay_flags |= NFSLAY_RETURNED;
7336 					wakeup(lyp);
7337 					error = 0;
7338 				}
7339 			}
7340 			NFSDRECALLUNLOCK();
7341 		}
7342 		if (layouttype == NFSLAYOUT_FLEXFILE && layp != NULL)
7343 			nfsrv_flexlayouterr(nd, layp, maxcnt, p);
7344 	} else if (kind == NFSV4LAYOUTRET_FSID)
7345 		nfsrv_freelayouts(&nd->nd_clientid,
7346 		    &vp->v_mount->mnt_stat.f_fsid, layouttype, iomode);
7347 	else if (kind == NFSV4LAYOUTRET_ALL)
7348 		nfsrv_freelayouts(&nd->nd_clientid, NULL, layouttype, iomode);
7349 	else
7350 		error = NFSERR_INVAL;
7351 	if (error == -1)
7352 		error = 0;
7353 	return (error);
7354 }
7355 
7356 /*
7357  * Look for an existing layout.
7358  */
7359 static int
7360 nfsrv_findlayout(nfsquad_t *clientidp, fhandle_t *fhp, int laytype,
7361     NFSPROC_T *p, struct nfslayout **lypp)
7362 {
7363 	struct nfslayouthash *lhyp;
7364 	struct nfslayout *lyp;
7365 	int ret;
7366 
7367 	*lypp = NULL;
7368 	ret = 0;
7369 	lhyp = NFSLAYOUTHASH(fhp);
7370 	TAILQ_FOREACH(lyp, &lhyp->list, lay_list) {
7371 		if (NFSBCMP(&lyp->lay_fh, fhp, sizeof(*fhp)) == 0 &&
7372 		    lyp->lay_clientid.qval == clientidp->qval &&
7373 		    lyp->lay_type == laytype)
7374 			break;
7375 	}
7376 	if (lyp != NULL)
7377 		*lypp = lyp;
7378 	else
7379 		ret = -1;
7380 	return (ret);
7381 }
7382 
7383 /*
7384  * Add the new layout, as required.
7385  */
7386 static int
7387 nfsrv_addlayout(struct nfsrv_descript *nd, struct nfslayout **lypp,
7388     nfsv4stateid_t *stateidp, char *layp, int *layoutlenp, NFSPROC_T *p)
7389 {
7390 	struct nfsclient *clp;
7391 	struct nfslayouthash *lhyp;
7392 	struct nfslayout *lyp, *nlyp;
7393 	fhandle_t *fhp;
7394 	int error;
7395 
7396 	KASSERT((nd->nd_flag & ND_IMPLIEDCLID) != 0,
7397 	    ("nfsrv_layoutget: no nd_clientid\n"));
7398 	lyp = *lypp;
7399 	fhp = &lyp->lay_fh;
7400 	NFSLOCKSTATE();
7401 	error = nfsrv_getclient((nfsquad_t)((u_quad_t)0), CLOPS_RENEW, &clp,
7402 	    NULL, (nfsquad_t)((u_quad_t)0), 0, nd, p);
7403 	if (error != 0) {
7404 		NFSUNLOCKSTATE();
7405 		return (error);
7406 	}
7407 	lyp->lay_stateid.seqid = stateidp->seqid = 1;
7408 	lyp->lay_stateid.other[0] = stateidp->other[0] =
7409 	    clp->lc_clientid.lval[0];
7410 	lyp->lay_stateid.other[1] = stateidp->other[1] =
7411 	    clp->lc_clientid.lval[1];
7412 	lyp->lay_stateid.other[2] = stateidp->other[2] =
7413 	    nfsrv_nextstateindex(clp);
7414 	NFSUNLOCKSTATE();
7415 
7416 	lhyp = NFSLAYOUTHASH(fhp);
7417 	NFSLOCKLAYOUT(lhyp);
7418 	TAILQ_FOREACH(nlyp, &lhyp->list, lay_list) {
7419 		if (NFSBCMP(&nlyp->lay_fh, fhp, sizeof(*fhp)) == 0 &&
7420 		    nlyp->lay_clientid.qval == nd->nd_clientid.qval)
7421 			break;
7422 	}
7423 	if (nlyp != NULL) {
7424 		/* A layout already exists, so use it. */
7425 		nlyp->lay_flags |= (lyp->lay_flags & (NFSLAY_READ | NFSLAY_RW));
7426 		NFSBCOPY(nlyp->lay_xdr, layp, nlyp->lay_layoutlen);
7427 		*layoutlenp = nlyp->lay_layoutlen;
7428 		if (++nlyp->lay_stateid.seqid == 0)
7429 			nlyp->lay_stateid.seqid = 1;
7430 		stateidp->seqid = nlyp->lay_stateid.seqid;
7431 		stateidp->other[0] = nlyp->lay_stateid.other[0];
7432 		stateidp->other[1] = nlyp->lay_stateid.other[1];
7433 		stateidp->other[2] = nlyp->lay_stateid.other[2];
7434 		NFSUNLOCKLAYOUT(lhyp);
7435 		return (0);
7436 	}
7437 
7438 	/* Insert the new layout in the lists. */
7439 	*lypp = NULL;
7440 	atomic_add_int(&nfsrv_layoutcnt, 1);
7441 	nfsstatsv1.srvlayouts++;
7442 	NFSBCOPY(lyp->lay_xdr, layp, lyp->lay_layoutlen);
7443 	*layoutlenp = lyp->lay_layoutlen;
7444 	TAILQ_INSERT_HEAD(&lhyp->list, lyp, lay_list);
7445 	NFSUNLOCKLAYOUT(lhyp);
7446 	return (0);
7447 }
7448 
7449 /*
7450  * Get the devinfo for a deviceid.
7451  */
7452 int
7453 nfsrv_getdevinfo(char *devid, int layouttype, uint32_t *maxcnt,
7454     uint32_t *notify, int *devaddrlen, char **devaddr)
7455 {
7456 	struct nfsdevice *ds;
7457 
7458 	if ((layouttype != NFSLAYOUT_NFSV4_1_FILES && layouttype !=
7459 	     NFSLAYOUT_FLEXFILE) ||
7460 	    (nfsrv_maxpnfsmirror > 1 && layouttype == NFSLAYOUT_NFSV4_1_FILES))
7461 		return (NFSERR_UNKNLAYOUTTYPE);
7462 
7463 	/*
7464 	 * Now, search for the device id.  Note that the structures won't go
7465 	 * away, but the order changes in the list.  As such, the lock only
7466 	 * needs to be held during the search through the list.
7467 	 */
7468 	NFSDDSLOCK();
7469 	TAILQ_FOREACH(ds, &nfsrv_devidhead, nfsdev_list) {
7470 		if (NFSBCMP(devid, ds->nfsdev_deviceid, NFSX_V4DEVICEID) == 0 &&
7471 		    ds->nfsdev_nmp != NULL)
7472 			break;
7473 	}
7474 	NFSDDSUNLOCK();
7475 	if (ds == NULL)
7476 		return (NFSERR_NOENT);
7477 
7478 	/* If the correct nfsdev_XXXXaddrlen is > 0, we have the device info. */
7479 	*devaddrlen = 0;
7480 	if (layouttype == NFSLAYOUT_NFSV4_1_FILES) {
7481 		*devaddrlen = ds->nfsdev_fileaddrlen;
7482 		*devaddr = ds->nfsdev_fileaddr;
7483 	} else if (layouttype == NFSLAYOUT_FLEXFILE) {
7484 		*devaddrlen = ds->nfsdev_flexaddrlen;
7485 		*devaddr = ds->nfsdev_flexaddr;
7486 	}
7487 	if (*devaddrlen == 0)
7488 		return (NFSERR_UNKNLAYOUTTYPE);
7489 
7490 	/*
7491 	 * The XDR overhead is 3 unsigned values: layout_type,
7492 	 * length_of_address and notify bitmap.
7493 	 * If the notify array is changed to not all zeros, the
7494 	 * count of unsigned values must be increased.
7495 	 */
7496 	if (*maxcnt > 0 && *maxcnt < NFSM_RNDUP(*devaddrlen) +
7497 	    3 * NFSX_UNSIGNED) {
7498 		*maxcnt = NFSM_RNDUP(*devaddrlen) + 3 * NFSX_UNSIGNED;
7499 		return (NFSERR_TOOSMALL);
7500 	}
7501 	return (0);
7502 }
7503 
7504 /*
7505  * Free a list of layout state structures.
7506  */
7507 static void
7508 nfsrv_freelayoutlist(nfsquad_t clientid)
7509 {
7510 	struct nfslayouthash *lhyp;
7511 	struct nfslayout *lyp, *nlyp;
7512 	int i;
7513 
7514 	for (i = 0; i < nfsrv_layouthashsize; i++) {
7515 		lhyp = &nfslayouthash[i];
7516 		NFSLOCKLAYOUT(lhyp);
7517 		TAILQ_FOREACH_SAFE(lyp, &lhyp->list, lay_list, nlyp) {
7518 			if (lyp->lay_clientid.qval == clientid.qval)
7519 				nfsrv_freelayout(&lhyp->list, lyp);
7520 		}
7521 		NFSUNLOCKLAYOUT(lhyp);
7522 	}
7523 }
7524 
7525 /*
7526  * Free up a layout.
7527  */
7528 static void
7529 nfsrv_freelayout(struct nfslayouthead *lhp, struct nfslayout *lyp)
7530 {
7531 
7532 	NFSD_DEBUG(4, "Freelayout=%p\n", lyp);
7533 	atomic_add_int(&nfsrv_layoutcnt, -1);
7534 	nfsstatsv1.srvlayouts--;
7535 	TAILQ_REMOVE(lhp, lyp, lay_list);
7536 	free(lyp, M_NFSDSTATE);
7537 }
7538 
7539 /*
7540  * Free up a device id.
7541  */
7542 void
7543 nfsrv_freeonedevid(struct nfsdevice *ds)
7544 {
7545 	int i;
7546 
7547 	atomic_add_int(&nfsrv_devidcnt, -1);
7548 	vrele(ds->nfsdev_dvp);
7549 	for (i = 0; i < nfsrv_dsdirsize; i++)
7550 		if (ds->nfsdev_dsdir[i] != NULL)
7551 			vrele(ds->nfsdev_dsdir[i]);
7552 	free(ds->nfsdev_fileaddr, M_NFSDSTATE);
7553 	free(ds->nfsdev_flexaddr, M_NFSDSTATE);
7554 	free(ds->nfsdev_host, M_NFSDSTATE);
7555 	free(ds, M_NFSDSTATE);
7556 }
7557 
7558 /*
7559  * Free up a device id and its mirrors.
7560  */
7561 static void
7562 nfsrv_freedevid(struct nfsdevice *ds)
7563 {
7564 
7565 	TAILQ_REMOVE(&nfsrv_devidhead, ds, nfsdev_list);
7566 	nfsrv_freeonedevid(ds);
7567 }
7568 
7569 /*
7570  * Free all layouts and device ids.
7571  * Done when the nfsd threads are shut down since there may be a new
7572  * modified device id list created when the nfsd is restarted.
7573  */
7574 void
7575 nfsrv_freealllayoutsanddevids(void)
7576 {
7577 	struct nfsdontlist *mrp, *nmrp;
7578 	struct nfslayout *lyp, *nlyp;
7579 
7580 	/* Get rid of the deviceid structures. */
7581 	nfsrv_freealldevids();
7582 	TAILQ_INIT(&nfsrv_devidhead);
7583 	nfsrv_devidcnt = 0;
7584 
7585 	/* Get rid of all layouts. */
7586 	nfsrv_freealllayouts();
7587 
7588 	/* Get rid of any nfsdontlist entries. */
7589 	LIST_FOREACH_SAFE(mrp, &nfsrv_dontlisthead, nfsmr_list, nmrp)
7590 		free(mrp, M_NFSDSTATE);
7591 	LIST_INIT(&nfsrv_dontlisthead);
7592 	nfsrv_dontlistlen = 0;
7593 
7594 	/* Free layouts in the recall list. */
7595 	TAILQ_FOREACH_SAFE(lyp, &nfsrv_recalllisthead, lay_list, nlyp)
7596 		nfsrv_freelayout(&nfsrv_recalllisthead, lyp);
7597 	TAILQ_INIT(&nfsrv_recalllisthead);
7598 }
7599 
7600 /*
7601  * Free layouts that match the arguments.
7602  */
7603 static void
7604 nfsrv_freelayouts(nfsquad_t *clid, fsid_t *fs, int laytype, int iomode)
7605 {
7606 	struct nfslayouthash *lhyp;
7607 	struct nfslayout *lyp, *nlyp;
7608 	int i;
7609 
7610 	for (i = 0; i < nfsrv_layouthashsize; i++) {
7611 		lhyp = &nfslayouthash[i];
7612 		NFSLOCKLAYOUT(lhyp);
7613 		TAILQ_FOREACH_SAFE(lyp, &lhyp->list, lay_list, nlyp) {
7614 			if (clid->qval != lyp->lay_clientid.qval)
7615 				continue;
7616 			if (fs != NULL && fsidcmp(fs, &lyp->lay_fsid) != 0)
7617 				continue;
7618 			if (laytype != lyp->lay_type)
7619 				continue;
7620 			if ((iomode & NFSLAYOUTIOMODE_READ) != 0)
7621 				lyp->lay_flags &= ~NFSLAY_READ;
7622 			if ((iomode & NFSLAYOUTIOMODE_RW) != 0)
7623 				lyp->lay_flags &= ~NFSLAY_RW;
7624 			if ((lyp->lay_flags & (NFSLAY_READ | NFSLAY_RW)) == 0)
7625 				nfsrv_freelayout(&lhyp->list, lyp);
7626 		}
7627 		NFSUNLOCKLAYOUT(lhyp);
7628 	}
7629 }
7630 
7631 /*
7632  * Free all layouts for the argument file.
7633  */
7634 void
7635 nfsrv_freefilelayouts(fhandle_t *fhp)
7636 {
7637 	struct nfslayouthash *lhyp;
7638 	struct nfslayout *lyp, *nlyp;
7639 
7640 	lhyp = NFSLAYOUTHASH(fhp);
7641 	NFSLOCKLAYOUT(lhyp);
7642 	TAILQ_FOREACH_SAFE(lyp, &lhyp->list, lay_list, nlyp) {
7643 		if (NFSBCMP(&lyp->lay_fh, fhp, sizeof(*fhp)) == 0)
7644 			nfsrv_freelayout(&lhyp->list, lyp);
7645 	}
7646 	NFSUNLOCKLAYOUT(lhyp);
7647 }
7648 
7649 /*
7650  * Free all layouts.
7651  */
7652 static void
7653 nfsrv_freealllayouts(void)
7654 {
7655 	struct nfslayouthash *lhyp;
7656 	struct nfslayout *lyp, *nlyp;
7657 	int i;
7658 
7659 	for (i = 0; i < nfsrv_layouthashsize; i++) {
7660 		lhyp = &nfslayouthash[i];
7661 		NFSLOCKLAYOUT(lhyp);
7662 		TAILQ_FOREACH_SAFE(lyp, &lhyp->list, lay_list, nlyp)
7663 			nfsrv_freelayout(&lhyp->list, lyp);
7664 		NFSUNLOCKLAYOUT(lhyp);
7665 	}
7666 }
7667 
7668 /*
7669  * Look up the mount path for the DS server.
7670  */
7671 static int
7672 nfsrv_setdsserver(char *dspathp, char *mdspathp, NFSPROC_T *p,
7673     struct nfsdevice **dsp)
7674 {
7675 	struct nameidata nd;
7676 	struct nfsdevice *ds;
7677 	struct mount *mp;
7678 	int error, i;
7679 	char *dsdirpath;
7680 	size_t dsdirsize;
7681 
7682 	NFSD_DEBUG(4, "setdssrv path=%s\n", dspathp);
7683 	*dsp = NULL;
7684 	NDINIT(&nd, LOOKUP, FOLLOW | LOCKSHARED | LOCKLEAF, UIO_SYSSPACE,
7685 	    dspathp);
7686 	error = namei(&nd);
7687 	NFSD_DEBUG(4, "lookup=%d\n", error);
7688 	if (error != 0)
7689 		return (error);
7690 	if (nd.ni_vp->v_type != VDIR) {
7691 		vput(nd.ni_vp);
7692 		NFSD_DEBUG(4, "dspath not dir\n");
7693 		return (ENOTDIR);
7694 	}
7695 	if (strcmp(nd.ni_vp->v_mount->mnt_vfc->vfc_name, "nfs") != 0) {
7696 		vput(nd.ni_vp);
7697 		NFSD_DEBUG(4, "dspath not an NFS mount\n");
7698 		return (ENXIO);
7699 	}
7700 
7701 	/*
7702 	 * Allocate a DS server structure with the NFS mounted directory
7703 	 * vnode reference counted, so that a non-forced dismount will
7704 	 * fail with EBUSY.
7705 	 * This structure is always linked into the list, even if an error
7706 	 * is being returned.  The caller will free the entire list upon
7707 	 * an error return.
7708 	 */
7709 	*dsp = ds = malloc(sizeof(*ds) + nfsrv_dsdirsize * sizeof(vnode_t),
7710 	    M_NFSDSTATE, M_WAITOK | M_ZERO);
7711 	ds->nfsdev_dvp = nd.ni_vp;
7712 	ds->nfsdev_nmp = VFSTONFS(nd.ni_vp->v_mount);
7713 	NFSVOPUNLOCK(nd.ni_vp);
7714 
7715 	dsdirsize = strlen(dspathp) + 16;
7716 	dsdirpath = malloc(dsdirsize, M_TEMP, M_WAITOK);
7717 	/* Now, create the DS directory structures. */
7718 	for (i = 0; i < nfsrv_dsdirsize; i++) {
7719 		snprintf(dsdirpath, dsdirsize, "%s/ds%d", dspathp, i);
7720 		NDINIT(&nd, LOOKUP, FOLLOW | LOCKSHARED | LOCKLEAF,
7721 		    UIO_SYSSPACE, dsdirpath);
7722 		error = namei(&nd);
7723 		NFSD_DEBUG(4, "dsdirpath=%s lookup=%d\n", dsdirpath, error);
7724 		if (error != 0)
7725 			break;
7726 		if (nd.ni_vp->v_type != VDIR) {
7727 			vput(nd.ni_vp);
7728 			error = ENOTDIR;
7729 			NFSD_DEBUG(4, "dsdirpath not a VDIR\n");
7730 			break;
7731 		}
7732 		if (strcmp(nd.ni_vp->v_mount->mnt_vfc->vfc_name, "nfs") != 0) {
7733 			vput(nd.ni_vp);
7734 			error = ENXIO;
7735 			NFSD_DEBUG(4, "dsdirpath not an NFS mount\n");
7736 			break;
7737 		}
7738 		ds->nfsdev_dsdir[i] = nd.ni_vp;
7739 		NFSVOPUNLOCK(nd.ni_vp);
7740 	}
7741 	free(dsdirpath, M_TEMP);
7742 
7743 	if (strlen(mdspathp) > 0) {
7744 		/*
7745 		 * This DS stores file for a specific MDS exported file
7746 		 * system.
7747 		 */
7748 		NDINIT(&nd, LOOKUP, FOLLOW | LOCKSHARED | LOCKLEAF,
7749 		    UIO_SYSSPACE, mdspathp);
7750 		error = namei(&nd);
7751 		NFSD_DEBUG(4, "mds lookup=%d\n", error);
7752 		if (error != 0)
7753 			goto out;
7754 		if (nd.ni_vp->v_type != VDIR) {
7755 			vput(nd.ni_vp);
7756 			error = ENOTDIR;
7757 			NFSD_DEBUG(4, "mdspath not dir\n");
7758 			goto out;
7759 		}
7760 		mp = nd.ni_vp->v_mount;
7761 		if ((mp->mnt_flag & MNT_EXPORTED) == 0) {
7762 			vput(nd.ni_vp);
7763 			error = ENXIO;
7764 			NFSD_DEBUG(4, "mdspath not an exported fs\n");
7765 			goto out;
7766 		}
7767 		ds->nfsdev_mdsfsid = mp->mnt_stat.f_fsid;
7768 		ds->nfsdev_mdsisset = 1;
7769 		vput(nd.ni_vp);
7770 	}
7771 
7772 out:
7773 	TAILQ_INSERT_TAIL(&nfsrv_devidhead, ds, nfsdev_list);
7774 	atomic_add_int(&nfsrv_devidcnt, 1);
7775 	return (error);
7776 }
7777 
7778 /*
7779  * Look up the mount path for the DS server and delete it.
7780  */
7781 int
7782 nfsrv_deldsserver(int op, char *dspathp, NFSPROC_T *p)
7783 {
7784 	struct mount *mp;
7785 	struct nfsmount *nmp;
7786 	struct nfsdevice *ds;
7787 	int error;
7788 
7789 	NFSD_DEBUG(4, "deldssrv path=%s\n", dspathp);
7790 	/*
7791 	 * Search for the path in the mount list.  Avoid looking the path
7792 	 * up, since this mount point may be hung, with associated locked
7793 	 * vnodes, etc.
7794 	 * Set NFSMNTP_CANCELRPCS so that any forced dismount will be blocked
7795 	 * until this completes.
7796 	 * As noted in the man page, this should be done before any forced
7797 	 * dismount on the mount point, but at least the handshake on
7798 	 * NFSMNTP_CANCELRPCS should make it safe.
7799 	 */
7800 	error = 0;
7801 	ds = NULL;
7802 	nmp = NULL;
7803 	mtx_lock(&mountlist_mtx);
7804 	TAILQ_FOREACH(mp, &mountlist, mnt_list) {
7805 		if (strcmp(mp->mnt_stat.f_mntonname, dspathp) == 0 &&
7806 		    strcmp(mp->mnt_stat.f_fstypename, "nfs") == 0 &&
7807 		    mp->mnt_data != NULL) {
7808 			nmp = VFSTONFS(mp);
7809 			NFSLOCKMNT(nmp);
7810 			if ((nmp->nm_privflag & (NFSMNTP_FORCEDISM |
7811 			     NFSMNTP_CANCELRPCS)) == 0) {
7812 				nmp->nm_privflag |= NFSMNTP_CANCELRPCS;
7813 				NFSUNLOCKMNT(nmp);
7814 			} else {
7815 				NFSUNLOCKMNT(nmp);
7816 				nmp = NULL;
7817 			}
7818 			break;
7819 		}
7820 	}
7821 	mtx_unlock(&mountlist_mtx);
7822 
7823 	if (nmp != NULL) {
7824 		ds = nfsrv_deldsnmp(op, nmp, p);
7825 		NFSD_DEBUG(4, "deldsnmp=%p\n", ds);
7826 		if (ds != NULL) {
7827 			nfsrv_killrpcs(nmp);
7828 			NFSD_DEBUG(4, "aft killrpcs\n");
7829 		} else
7830 			error = ENXIO;
7831 		NFSLOCKMNT(nmp);
7832 		nmp->nm_privflag &= ~NFSMNTP_CANCELRPCS;
7833 		wakeup(nmp);
7834 		NFSUNLOCKMNT(nmp);
7835 	} else
7836 		error = EINVAL;
7837 	return (error);
7838 }
7839 
7840 /*
7841  * Search for and remove a DS entry which matches the "nmp" argument.
7842  * The nfsdevice structure pointer is returned so that the caller can
7843  * free it via nfsrv_freeonedevid().
7844  * For the forced case, do not try to do LayoutRecalls, since the server
7845  * must be shut down now anyhow.
7846  */
7847 struct nfsdevice *
7848 nfsrv_deldsnmp(int op, struct nfsmount *nmp, NFSPROC_T *p)
7849 {
7850 	struct nfsdevice *fndds;
7851 
7852 	NFSD_DEBUG(4, "deldsdvp\n");
7853 	NFSDDSLOCK();
7854 	if (op == PNFSDOP_FORCEDELDS)
7855 		fndds = nfsv4_findmirror(nmp);
7856 	else
7857 		fndds = nfsrv_findmirroredds(nmp);
7858 	if (fndds != NULL)
7859 		nfsrv_deleteds(fndds);
7860 	NFSDDSUNLOCK();
7861 	if (fndds != NULL) {
7862 		if (op != PNFSDOP_FORCEDELDS)
7863 			nfsrv_flexmirrordel(fndds->nfsdev_deviceid, p);
7864 		printf("pNFS server: mirror %s failed\n", fndds->nfsdev_host);
7865 	}
7866 	return (fndds);
7867 }
7868 
7869 /*
7870  * Similar to nfsrv_deldsnmp(), except that the DS is indicated by deviceid.
7871  * This function also calls nfsrv_killrpcs() to unblock RPCs on the mount
7872  * point.
7873  * Also, returns an error instead of the nfsdevice found.
7874  */
7875 int
7876 nfsrv_delds(char *devid, NFSPROC_T *p)
7877 {
7878 	struct nfsdevice *ds, *fndds;
7879 	struct nfsmount *nmp;
7880 	int fndmirror;
7881 
7882 	NFSD_DEBUG(4, "delds\n");
7883 	/*
7884 	 * Search the DS server list for a match with devid.
7885 	 * Remove the DS entry if found and there is a mirror.
7886 	 */
7887 	fndds = NULL;
7888 	nmp = NULL;
7889 	fndmirror = 0;
7890 	NFSDDSLOCK();
7891 	TAILQ_FOREACH(ds, &nfsrv_devidhead, nfsdev_list) {
7892 		if (NFSBCMP(ds->nfsdev_deviceid, devid, NFSX_V4DEVICEID) == 0 &&
7893 		    ds->nfsdev_nmp != NULL) {
7894 			NFSD_DEBUG(4, "fnd main ds\n");
7895 			fndds = ds;
7896 			break;
7897 		}
7898 	}
7899 	if (fndds == NULL) {
7900 		NFSDDSUNLOCK();
7901 		return (ENXIO);
7902 	}
7903 	if (fndds->nfsdev_mdsisset == 0 && nfsrv_faildscnt > 0)
7904 		fndmirror = 1;
7905 	else if (fndds->nfsdev_mdsisset != 0) {
7906 		/* For the fsid is set case, search for a mirror. */
7907 		TAILQ_FOREACH(ds, &nfsrv_devidhead, nfsdev_list) {
7908 			if (ds != fndds && ds->nfsdev_nmp != NULL &&
7909 			    ds->nfsdev_mdsisset != 0 &&
7910 			    fsidcmp(&ds->nfsdev_mdsfsid,
7911 			    &fndds->nfsdev_mdsfsid) == 0) {
7912 				fndmirror = 1;
7913 				break;
7914 			}
7915 		}
7916 	}
7917 	if (fndmirror != 0) {
7918 		nmp = fndds->nfsdev_nmp;
7919 		NFSLOCKMNT(nmp);
7920 		if ((nmp->nm_privflag & (NFSMNTP_FORCEDISM |
7921 		     NFSMNTP_CANCELRPCS)) == 0) {
7922 			nmp->nm_privflag |= NFSMNTP_CANCELRPCS;
7923 			NFSUNLOCKMNT(nmp);
7924 			nfsrv_deleteds(fndds);
7925 		} else {
7926 			NFSUNLOCKMNT(nmp);
7927 			nmp = NULL;
7928 		}
7929 	}
7930 	NFSDDSUNLOCK();
7931 	if (nmp != NULL) {
7932 		nfsrv_flexmirrordel(fndds->nfsdev_deviceid, p);
7933 		printf("pNFS server: mirror %s failed\n", fndds->nfsdev_host);
7934 		nfsrv_killrpcs(nmp);
7935 		NFSLOCKMNT(nmp);
7936 		nmp->nm_privflag &= ~NFSMNTP_CANCELRPCS;
7937 		wakeup(nmp);
7938 		NFSUNLOCKMNT(nmp);
7939 		return (0);
7940 	}
7941 	return (ENXIO);
7942 }
7943 
7944 /*
7945  * Mark a DS as disabled by setting nfsdev_nmp = NULL.
7946  */
7947 static void
7948 nfsrv_deleteds(struct nfsdevice *fndds)
7949 {
7950 
7951 	NFSD_DEBUG(4, "deleteds: deleting a mirror\n");
7952 	fndds->nfsdev_nmp = NULL;
7953 	if (fndds->nfsdev_mdsisset == 0)
7954 		nfsrv_faildscnt--;
7955 }
7956 
7957 /*
7958  * Fill in the addr structures for the File and Flex File layouts.
7959  */
7960 static void
7961 nfsrv_allocdevid(struct nfsdevice *ds, char *addr, char *dnshost)
7962 {
7963 	uint32_t *tl;
7964 	char *netprot;
7965 	int addrlen;
7966 	static uint64_t new_devid = 0;
7967 
7968 	if (strchr(addr, ':') != NULL)
7969 		netprot = "tcp6";
7970 	else
7971 		netprot = "tcp";
7972 
7973 	/* Fill in the device id. */
7974 	NFSBCOPY(&nfsdev_time, ds->nfsdev_deviceid, sizeof(nfsdev_time));
7975 	new_devid++;
7976 	NFSBCOPY(&new_devid, &ds->nfsdev_deviceid[sizeof(nfsdev_time)],
7977 	    sizeof(new_devid));
7978 
7979 	/*
7980 	 * Fill in the file addr (actually the nfsv4_file_layout_ds_addr4
7981 	 * as defined in RFC5661) in XDR.
7982 	 */
7983 	addrlen = NFSM_RNDUP(strlen(addr)) + NFSM_RNDUP(strlen(netprot)) +
7984 	    6 * NFSX_UNSIGNED;
7985 	NFSD_DEBUG(4, "hn=%s addr=%s netprot=%s\n", dnshost, addr, netprot);
7986 	ds->nfsdev_fileaddrlen = addrlen;
7987 	tl = malloc(addrlen, M_NFSDSTATE, M_WAITOK | M_ZERO);
7988 	ds->nfsdev_fileaddr = (char *)tl;
7989 	*tl++ = txdr_unsigned(1);		/* One stripe with index 0. */
7990 	*tl++ = 0;
7991 	*tl++ = txdr_unsigned(1);		/* One multipath list */
7992 	*tl++ = txdr_unsigned(1);		/* with one entry in it. */
7993 	/* The netaddr for this one entry. */
7994 	*tl++ = txdr_unsigned(strlen(netprot));
7995 	NFSBCOPY(netprot, tl, strlen(netprot));
7996 	tl += (NFSM_RNDUP(strlen(netprot)) / NFSX_UNSIGNED);
7997 	*tl++ = txdr_unsigned(strlen(addr));
7998 	NFSBCOPY(addr, tl, strlen(addr));
7999 
8000 	/*
8001 	 * Fill in the flex file addr (actually the ff_device_addr4
8002 	 * as defined for Flexible File Layout) in XDR.
8003 	 */
8004 	addrlen = NFSM_RNDUP(strlen(addr)) + NFSM_RNDUP(strlen(netprot)) +
8005 	    14 * NFSX_UNSIGNED;
8006 	ds->nfsdev_flexaddrlen = addrlen;
8007 	tl = malloc(addrlen, M_NFSDSTATE, M_WAITOK | M_ZERO);
8008 	ds->nfsdev_flexaddr = (char *)tl;
8009 	*tl++ = txdr_unsigned(1);		/* One multipath entry. */
8010 	/* The netaddr for this one entry. */
8011 	*tl++ = txdr_unsigned(strlen(netprot));
8012 	NFSBCOPY(netprot, tl, strlen(netprot));
8013 	tl += (NFSM_RNDUP(strlen(netprot)) / NFSX_UNSIGNED);
8014 	*tl++ = txdr_unsigned(strlen(addr));
8015 	NFSBCOPY(addr, tl, strlen(addr));
8016 	tl += (NFSM_RNDUP(strlen(addr)) / NFSX_UNSIGNED);
8017 	*tl++ = txdr_unsigned(2);		/* Two NFS Versions. */
8018 	*tl++ = txdr_unsigned(NFS_VER4);	/* NFSv4. */
8019 	*tl++ = txdr_unsigned(NFSV42_MINORVERSION); /* Minor version 2. */
8020 	*tl++ = txdr_unsigned(nfs_srvmaxio);	/* DS max rsize. */
8021 	*tl++ = txdr_unsigned(nfs_srvmaxio);	/* DS max wsize. */
8022 	*tl++ = newnfs_true;			/* Tightly coupled. */
8023 	*tl++ = txdr_unsigned(NFS_VER4);	/* NFSv4. */
8024 	*tl++ = txdr_unsigned(NFSV41_MINORVERSION); /* Minor version 1. */
8025 	*tl++ = txdr_unsigned(nfs_srvmaxio);	/* DS max rsize. */
8026 	*tl++ = txdr_unsigned(nfs_srvmaxio);	/* DS max wsize. */
8027 	*tl = newnfs_true;			/* Tightly coupled. */
8028 
8029 	ds->nfsdev_hostnamelen = strlen(dnshost);
8030 	ds->nfsdev_host = malloc(ds->nfsdev_hostnamelen + 1, M_NFSDSTATE,
8031 	    M_WAITOK);
8032 	NFSBCOPY(dnshost, ds->nfsdev_host, ds->nfsdev_hostnamelen + 1);
8033 }
8034 
8035 /*
8036  * Create the device id list.
8037  * Return 0 if the nfsd threads are to run and ENXIO if the "-p" argument
8038  * is misconfigured.
8039  */
8040 int
8041 nfsrv_createdevids(struct nfsd_nfsd_args *args, NFSPROC_T *p)
8042 {
8043 	struct nfsdevice *ds;
8044 	char *addrp, *dnshostp, *dspathp, *mdspathp;
8045 	int error, i;
8046 
8047 	addrp = args->addr;
8048 	dnshostp = args->dnshost;
8049 	dspathp = args->dspath;
8050 	mdspathp = args->mdspath;
8051 	nfsrv_maxpnfsmirror = args->mirrorcnt;
8052 	if (addrp == NULL || dnshostp == NULL || dspathp == NULL ||
8053 	    mdspathp == NULL)
8054 		return (0);
8055 
8056 	/*
8057 	 * Loop around for each nul-terminated string in args->addr,
8058 	 * args->dnshost, args->dnspath and args->mdspath.
8059 	 */
8060 	while (addrp < (args->addr + args->addrlen) &&
8061 	    dnshostp < (args->dnshost + args->dnshostlen) &&
8062 	    dspathp < (args->dspath + args->dspathlen) &&
8063 	    mdspathp < (args->mdspath + args->mdspathlen)) {
8064 		error = nfsrv_setdsserver(dspathp, mdspathp, p, &ds);
8065 		if (error != 0) {
8066 			/* Free all DS servers. */
8067 			nfsrv_freealldevids();
8068 			nfsrv_devidcnt = 0;
8069 			return (ENXIO);
8070 		}
8071 		nfsrv_allocdevid(ds, addrp, dnshostp);
8072 		addrp += (strlen(addrp) + 1);
8073 		dnshostp += (strlen(dnshostp) + 1);
8074 		dspathp += (strlen(dspathp) + 1);
8075 		mdspathp += (strlen(mdspathp) + 1);
8076 	}
8077 	if (nfsrv_devidcnt < nfsrv_maxpnfsmirror) {
8078 		/* Free all DS servers. */
8079 		nfsrv_freealldevids();
8080 		nfsrv_devidcnt = 0;
8081 		nfsrv_maxpnfsmirror = 1;
8082 		return (ENXIO);
8083 	}
8084 	/* We can fail at most one less DS than the mirror level. */
8085 	nfsrv_faildscnt = nfsrv_maxpnfsmirror - 1;
8086 
8087 	/*
8088 	 * Allocate the nfslayout hash table now, since this is a pNFS server.
8089 	 * Make it 1% of the high water mark and at least 100.
8090 	 */
8091 	if (nfslayouthash == NULL) {
8092 		nfsrv_layouthashsize = nfsrv_layouthighwater / 100;
8093 		if (nfsrv_layouthashsize < 100)
8094 			nfsrv_layouthashsize = 100;
8095 		nfslayouthash = mallocarray(nfsrv_layouthashsize,
8096 		    sizeof(struct nfslayouthash), M_NFSDSESSION, M_WAITOK |
8097 		    M_ZERO);
8098 		for (i = 0; i < nfsrv_layouthashsize; i++) {
8099 			mtx_init(&nfslayouthash[i].mtx, "nfslm", NULL, MTX_DEF);
8100 			TAILQ_INIT(&nfslayouthash[i].list);
8101 		}
8102 	}
8103 	return (0);
8104 }
8105 
8106 /*
8107  * Free all device ids.
8108  */
8109 static void
8110 nfsrv_freealldevids(void)
8111 {
8112 	struct nfsdevice *ds, *nds;
8113 
8114 	TAILQ_FOREACH_SAFE(ds, &nfsrv_devidhead, nfsdev_list, nds)
8115 		nfsrv_freedevid(ds);
8116 }
8117 
8118 /*
8119  * Check to see if there is a Read/Write Layout plus either:
8120  * - A Write Delegation
8121  * or
8122  * - An Open with Write_access.
8123  * Return 1 if this is the case and 0 otherwise.
8124  * This function is used by nfsrv_proxyds() to decide if doing a Proxy
8125  * Getattr RPC to the Data Server (DS) is necessary.
8126  */
8127 #define	NFSCLIDVECSIZE	6
8128 int
8129 nfsrv_checkdsattr(vnode_t vp, NFSPROC_T *p)
8130 {
8131 	fhandle_t fh, *tfhp;
8132 	struct nfsstate *stp;
8133 	struct nfslayout *lyp;
8134 	struct nfslayouthash *lhyp;
8135 	struct nfslockhashhead *hp;
8136 	struct nfslockfile *lfp;
8137 	nfsquad_t clid[NFSCLIDVECSIZE];
8138 	int clidcnt, ret;
8139 
8140 	ret = nfsvno_getfh(vp, &fh, p);
8141 	if (ret != 0)
8142 		return (0);
8143 
8144 	/* First check for a Read/Write Layout. */
8145 	clidcnt = 0;
8146 	lhyp = NFSLAYOUTHASH(&fh);
8147 	NFSLOCKLAYOUT(lhyp);
8148 	TAILQ_FOREACH(lyp, &lhyp->list, lay_list) {
8149 		if (NFSBCMP(&lyp->lay_fh, &fh, sizeof(fh)) == 0 &&
8150 		    ((lyp->lay_flags & NFSLAY_RW) != 0 ||
8151 		     ((lyp->lay_flags & NFSLAY_READ) != 0 &&
8152 		      nfsrv_pnfsatime != 0))) {
8153 			if (clidcnt < NFSCLIDVECSIZE)
8154 				clid[clidcnt].qval = lyp->lay_clientid.qval;
8155 			clidcnt++;
8156 		}
8157 	}
8158 	NFSUNLOCKLAYOUT(lhyp);
8159 	if (clidcnt == 0) {
8160 		/* None found, so return 0. */
8161 		return (0);
8162 	}
8163 
8164 	/* Get the nfslockfile for this fh. */
8165 	NFSLOCKSTATE();
8166 	hp = NFSLOCKHASH(&fh);
8167 	LIST_FOREACH(lfp, hp, lf_hash) {
8168 		tfhp = &lfp->lf_fh;
8169 		if (NFSVNO_CMPFH(&fh, tfhp))
8170 			break;
8171 	}
8172 	if (lfp == NULL) {
8173 		/* None found, so return 0. */
8174 		NFSUNLOCKSTATE();
8175 		return (0);
8176 	}
8177 
8178 	/* Now, look for a Write delegation for this clientid. */
8179 	LIST_FOREACH(stp, &lfp->lf_deleg, ls_file) {
8180 		if ((stp->ls_flags & NFSLCK_DELEGWRITE) != 0 &&
8181 		    nfsrv_fndclid(clid, stp->ls_clp->lc_clientid, clidcnt) != 0)
8182 			break;
8183 	}
8184 	if (stp != NULL) {
8185 		/* Found one, so return 1. */
8186 		NFSUNLOCKSTATE();
8187 		return (1);
8188 	}
8189 
8190 	/* No Write delegation, so look for an Open with Write_access. */
8191 	LIST_FOREACH(stp, &lfp->lf_open, ls_file) {
8192 		KASSERT((stp->ls_flags & NFSLCK_OPEN) != 0,
8193 		    ("nfsrv_checkdsattr: Non-open in Open list\n"));
8194 		if ((stp->ls_flags & NFSLCK_WRITEACCESS) != 0 &&
8195 		    nfsrv_fndclid(clid, stp->ls_clp->lc_clientid, clidcnt) != 0)
8196 			break;
8197 	}
8198 	NFSUNLOCKSTATE();
8199 	if (stp != NULL)
8200 		return (1);
8201 	return (0);
8202 }
8203 
8204 /*
8205  * Look for a matching clientid in the vector. Return 1 if one might match.
8206  */
8207 static int
8208 nfsrv_fndclid(nfsquad_t *clidvec, nfsquad_t clid, int clidcnt)
8209 {
8210 	int i;
8211 
8212 	/* If too many for the vector, return 1 since there might be a match. */
8213 	if (clidcnt > NFSCLIDVECSIZE)
8214 		return (1);
8215 
8216 	for (i = 0; i < clidcnt; i++)
8217 		if (clidvec[i].qval == clid.qval)
8218 			return (1);
8219 	return (0);
8220 }
8221 
8222 /*
8223  * Check the don't list for "vp" and see if issuing an rw layout is allowed.
8224  * Return 1 if issuing an rw layout isn't allowed, 0 otherwise.
8225  */
8226 static int
8227 nfsrv_dontlayout(fhandle_t *fhp)
8228 {
8229 	struct nfsdontlist *mrp;
8230 	int ret;
8231 
8232 	if (nfsrv_dontlistlen == 0)
8233 		return (0);
8234 	ret = 0;
8235 	NFSDDONTLISTLOCK();
8236 	LIST_FOREACH(mrp, &nfsrv_dontlisthead, nfsmr_list) {
8237 		if (NFSBCMP(fhp, &mrp->nfsmr_fh, sizeof(*fhp)) == 0 &&
8238 		    (mrp->nfsmr_flags & NFSMR_DONTLAYOUT) != 0) {
8239 			ret = 1;
8240 			break;
8241 		}
8242 	}
8243 	NFSDDONTLISTUNLOCK();
8244 	return (ret);
8245 }
8246 
8247 #define	PNFSDS_COPYSIZ	65536
8248 /*
8249  * Create a new file on a DS and copy the contents of an extant DS file to it.
8250  * This can be used for recovery of a DS file onto a recovered DS.
8251  * The steps are:
8252  * - When called, the MDS file's vnode is locked, blocking LayoutGet operations.
8253  * - Disable issuing of read/write layouts for the file via the nfsdontlist,
8254  *   so that they will be disabled after the MDS file's vnode is unlocked.
8255  * - Set up the nfsrv_recalllist so that recall of read/write layouts can
8256  *   be done.
8257  * - Unlock the MDS file's vnode, so that the client(s) can perform proxied
8258  *   writes, LayoutCommits and LayoutReturns for the file when completing the
8259  *   LayoutReturn requested by the LayoutRecall callback.
8260  * - Issue a LayoutRecall callback for all read/write layouts and wait for
8261  *   them to be returned. (If the LayoutRecall callback replies
8262  *   NFSERR_NOMATCHLAYOUT, they are gone and no LayoutReturn is needed.)
8263  * - Exclusively lock the MDS file's vnode.  This ensures that no proxied
8264  *   writes are in progress or can occur during the DS file copy.
8265  *   It also blocks Setattr operations.
8266  * - Create the file on the recovered mirror.
8267  * - Copy the file from the operational DS.
8268  * - Copy any ACL from the MDS file to the new DS file.
8269  * - Set the modify time of the new DS file to that of the MDS file.
8270  * - Update the extended attribute for the MDS file.
8271  * - Enable issuing of rw layouts by deleting the nfsdontlist entry.
8272  * - The caller will unlock the MDS file's vnode allowing operations
8273  *   to continue normally, since it is now on the mirror again.
8274  */
8275 int
8276 nfsrv_copymr(vnode_t vp, vnode_t fvp, vnode_t dvp, struct nfsdevice *ds,
8277     struct pnfsdsfile *pf, struct pnfsdsfile *wpf, int mirrorcnt,
8278     struct ucred *cred, NFSPROC_T *p)
8279 {
8280 	struct nfsdontlist *mrp, *nmrp;
8281 	struct nfslayouthash *lhyp;
8282 	struct nfslayout *lyp, *nlyp;
8283 	struct nfslayouthead thl;
8284 	struct mount *mp, *tvmp;
8285 	struct acl *aclp;
8286 	struct vattr va;
8287 	struct timespec mtime;
8288 	fhandle_t fh;
8289 	vnode_t tvp;
8290 	off_t rdpos, wrpos;
8291 	ssize_t aresid;
8292 	char *dat;
8293 	int didprintf, ret, retacl, xfer;
8294 
8295 	ASSERT_VOP_LOCKED(fvp, "nfsrv_copymr fvp");
8296 	ASSERT_VOP_LOCKED(vp, "nfsrv_copymr vp");
8297 	/*
8298 	 * Allocate a nfsdontlist entry and set the NFSMR_DONTLAYOUT flag
8299 	 * so that no more RW layouts will get issued.
8300 	 */
8301 	ret = nfsvno_getfh(vp, &fh, p);
8302 	if (ret != 0) {
8303 		NFSD_DEBUG(4, "nfsrv_copymr: getfh=%d\n", ret);
8304 		return (ret);
8305 	}
8306 	nmrp = malloc(sizeof(*nmrp), M_NFSDSTATE, M_WAITOK);
8307 	nmrp->nfsmr_flags = NFSMR_DONTLAYOUT;
8308 	NFSBCOPY(&fh, &nmrp->nfsmr_fh, sizeof(fh));
8309 	NFSDDONTLISTLOCK();
8310 	LIST_FOREACH(mrp, &nfsrv_dontlisthead, nfsmr_list) {
8311 		if (NFSBCMP(&fh, &mrp->nfsmr_fh, sizeof(fh)) == 0)
8312 			break;
8313 	}
8314 	if (mrp == NULL) {
8315 		LIST_INSERT_HEAD(&nfsrv_dontlisthead, nmrp, nfsmr_list);
8316 		mrp = nmrp;
8317 		nmrp = NULL;
8318 		nfsrv_dontlistlen++;
8319 		NFSD_DEBUG(4, "nfsrv_copymr: in dontlist\n");
8320 	} else {
8321 		NFSDDONTLISTUNLOCK();
8322 		free(nmrp, M_NFSDSTATE);
8323 		NFSD_DEBUG(4, "nfsrv_copymr: dup dontlist\n");
8324 		return (ENXIO);
8325 	}
8326 	NFSDDONTLISTUNLOCK();
8327 
8328 	/*
8329 	 * Search for all RW layouts for this file.  Move them to the
8330 	 * recall list, so they can be recalled and their return noted.
8331 	 */
8332 	lhyp = NFSLAYOUTHASH(&fh);
8333 	NFSDRECALLLOCK();
8334 	NFSLOCKLAYOUT(lhyp);
8335 	TAILQ_FOREACH_SAFE(lyp, &lhyp->list, lay_list, nlyp) {
8336 		if (NFSBCMP(&lyp->lay_fh, &fh, sizeof(fh)) == 0 &&
8337 		    (lyp->lay_flags & NFSLAY_RW) != 0) {
8338 			TAILQ_REMOVE(&lhyp->list, lyp, lay_list);
8339 			TAILQ_INSERT_HEAD(&nfsrv_recalllisthead, lyp, lay_list);
8340 			lyp->lay_trycnt = 0;
8341 		}
8342 	}
8343 	NFSUNLOCKLAYOUT(lhyp);
8344 	NFSDRECALLUNLOCK();
8345 
8346 	ret = 0;
8347 	mp = tvmp = NULL;
8348 	didprintf = 0;
8349 	TAILQ_INIT(&thl);
8350 	/* Unlock the MDS vp, so that a LayoutReturn can be done on it. */
8351 	NFSVOPUNLOCK(vp);
8352 	/* Now, do a recall for all layouts not yet recalled. */
8353 tryagain:
8354 	NFSDRECALLLOCK();
8355 	TAILQ_FOREACH(lyp, &nfsrv_recalllisthead, lay_list) {
8356 		if (NFSBCMP(&lyp->lay_fh, &fh, sizeof(fh)) == 0 &&
8357 		    (lyp->lay_flags & NFSLAY_RECALL) == 0) {
8358 			lyp->lay_flags |= NFSLAY_RECALL;
8359 			/*
8360 			 * The layout stateid.seqid needs to be incremented
8361 			 * before doing a LAYOUT_RECALL callback.
8362 			 */
8363 			if (++lyp->lay_stateid.seqid == 0)
8364 				lyp->lay_stateid.seqid = 1;
8365 			NFSDRECALLUNLOCK();
8366 			nfsrv_recalllayout(lyp->lay_clientid, &lyp->lay_stateid,
8367 			    &lyp->lay_fh, lyp, 0, lyp->lay_type, p);
8368 			NFSD_DEBUG(4, "nfsrv_copymr: recalled layout\n");
8369 			goto tryagain;
8370 		}
8371 	}
8372 
8373 	/* Now wait for them to be returned. */
8374 tryagain2:
8375 	TAILQ_FOREACH(lyp, &nfsrv_recalllisthead, lay_list) {
8376 		if (NFSBCMP(&lyp->lay_fh, &fh, sizeof(fh)) == 0) {
8377 			if ((lyp->lay_flags & NFSLAY_RETURNED) != 0) {
8378 				TAILQ_REMOVE(&nfsrv_recalllisthead, lyp,
8379 				    lay_list);
8380 				TAILQ_INSERT_HEAD(&thl, lyp, lay_list);
8381 				NFSD_DEBUG(4,
8382 				    "nfsrv_copymr: layout returned\n");
8383 			} else {
8384 				lyp->lay_trycnt++;
8385 				ret = mtx_sleep(lyp, NFSDRECALLMUTEXPTR,
8386 				    PVFS | PCATCH, "nfsmrl", hz);
8387 				NFSD_DEBUG(4, "nfsrv_copymr: aft sleep=%d\n",
8388 				    ret);
8389 				if (ret == EINTR || ret == ERESTART)
8390 					break;
8391 				if ((lyp->lay_flags & NFSLAY_RETURNED) == 0) {
8392 					/*
8393 					 * Give up after 60sec and return
8394 					 * ENXIO, failing the copymr.
8395 					 * This layout will remain on the
8396 					 * recalllist.  It can only be cleared
8397 					 * by restarting the nfsd.
8398 					 * This seems the safe way to handle
8399 					 * it, since it cannot be safely copied
8400 					 * with an outstanding RW layout.
8401 					 */
8402 					if (lyp->lay_trycnt >= 60) {
8403 						ret = ENXIO;
8404 						break;
8405 					}
8406 					if (didprintf == 0) {
8407 						printf("nfsrv_copymr: layout "
8408 						    "not returned\n");
8409 						didprintf = 1;
8410 					}
8411 				}
8412 			}
8413 			goto tryagain2;
8414 		}
8415 	}
8416 	NFSDRECALLUNLOCK();
8417 	/* We can now get rid of the layouts that have been returned. */
8418 	TAILQ_FOREACH_SAFE(lyp, &thl, lay_list, nlyp)
8419 		nfsrv_freelayout(&thl, lyp);
8420 
8421 	/*
8422 	 * Do the vn_start_write() calls here, before the MDS vnode is
8423 	 * locked and the tvp is created (locked) in the NFS file system
8424 	 * that dvp is in.
8425 	 * For tvmp, this probably isn't necessary, since it will be an
8426 	 * NFS mount and they are not suspendable at this time.
8427 	 */
8428 	if (ret == 0)
8429 		ret = vn_start_write(vp, &mp, V_WAIT | PCATCH);
8430 	if (ret == 0) {
8431 		tvmp = dvp->v_mount;
8432 		ret = vn_start_write(NULL, &tvmp, V_WAIT | PCATCH);
8433 	}
8434 
8435 	/*
8436 	 * LK_EXCLUSIVE lock the MDS vnode, so that any
8437 	 * proxied writes through the MDS will be blocked until we have
8438 	 * completed the copy and update of the extended attributes.
8439 	 * This will also ensure that any attributes and ACL will not be
8440 	 * changed until the copy is complete.
8441 	 */
8442 	NFSVOPLOCK(vp, LK_EXCLUSIVE | LK_RETRY);
8443 	if (ret == 0 && VN_IS_DOOMED(vp)) {
8444 		NFSD_DEBUG(4, "nfsrv_copymr: lk_exclusive doomed\n");
8445 		ret = ESTALE;
8446 	}
8447 
8448 	/* Create the data file on the recovered DS. */
8449 	if (ret == 0)
8450 		ret = nfsrv_createdsfile(vp, &fh, pf, dvp, ds, cred, p, &tvp);
8451 
8452 	/* Copy the DS file, if created successfully. */
8453 	if (ret == 0) {
8454 		/*
8455 		 * Get any NFSv4 ACL on the MDS file, so that it can be set
8456 		 * on the new DS file.
8457 		 */
8458 		aclp = acl_alloc(M_WAITOK | M_ZERO);
8459 		retacl = VOP_GETACL(vp, ACL_TYPE_NFS4, aclp, cred, p);
8460 		if (retacl != 0 && retacl != ENOATTR)
8461 			NFSD_DEBUG(1, "nfsrv_copymr: vop_getacl=%d\n", retacl);
8462 		dat = malloc(PNFSDS_COPYSIZ, M_TEMP, M_WAITOK);
8463 		/* Malloc a block of 0s used to check for holes. */
8464 		if (nfsrv_zeropnfsdat == NULL)
8465 			nfsrv_zeropnfsdat = malloc(PNFSDS_COPYSIZ, M_TEMP,
8466 			    M_WAITOK | M_ZERO);
8467 		rdpos = wrpos = 0;
8468 		ret = VOP_GETATTR(fvp, &va, cred);
8469 		aresid = 0;
8470 		while (ret == 0 && aresid == 0) {
8471 			ret = vn_rdwr(UIO_READ, fvp, dat, PNFSDS_COPYSIZ,
8472 			    rdpos, UIO_SYSSPACE, IO_NODELOCKED, cred, NULL,
8473 			    &aresid, p);
8474 			xfer = PNFSDS_COPYSIZ - aresid;
8475 			if (ret == 0 && xfer > 0) {
8476 				rdpos += xfer;
8477 				/*
8478 				 * Skip the write for holes, except for the
8479 				 * last block.
8480 				 */
8481 				if (xfer < PNFSDS_COPYSIZ || rdpos ==
8482 				    va.va_size || NFSBCMP(dat,
8483 				    nfsrv_zeropnfsdat, PNFSDS_COPYSIZ) != 0)
8484 					ret = vn_rdwr(UIO_WRITE, tvp, dat, xfer,
8485 					    wrpos, UIO_SYSSPACE, IO_NODELOCKED,
8486 					    cred, NULL, NULL, p);
8487 				if (ret == 0)
8488 					wrpos += xfer;
8489 			}
8490 		}
8491 
8492 		/* If there is an ACL and the copy succeeded, set the ACL. */
8493 		if (ret == 0 && retacl == 0) {
8494 			ret = VOP_SETACL(tvp, ACL_TYPE_NFS4, aclp, cred, p);
8495 			/*
8496 			 * Don't consider these as errors, since VOP_GETACL()
8497 			 * can return an ACL when they are not actually
8498 			 * supported.  For example, for UFS, VOP_GETACL()
8499 			 * will return a trivial ACL based on the uid/gid/mode
8500 			 * when there is no ACL on the file.
8501 			 * This case should be recognized as a trivial ACL
8502 			 * by UFS's VOP_SETACL() and succeed, but...
8503 			 */
8504 			if (ret == ENOATTR || ret == EOPNOTSUPP || ret == EPERM)
8505 				ret = 0;
8506 		}
8507 
8508 		if (ret == 0)
8509 			ret = VOP_FSYNC(tvp, MNT_WAIT, p);
8510 
8511 		/* Set the DS data file's modify time that of the MDS file. */
8512 		if (ret == 0)
8513 			ret = VOP_GETATTR(vp, &va, cred);
8514 		if (ret == 0) {
8515 			mtime = va.va_mtime;
8516 			VATTR_NULL(&va);
8517 			va.va_mtime = mtime;
8518 			ret = VOP_SETATTR(tvp, &va, cred);
8519 		}
8520 
8521 		vput(tvp);
8522 		acl_free(aclp);
8523 		free(dat, M_TEMP);
8524 	}
8525 	if (tvmp != NULL)
8526 		vn_finished_write(tvmp);
8527 
8528 	/* Update the extended attributes for the newly created DS file. */
8529 	if (ret == 0)
8530 		ret = vn_extattr_set(vp, IO_NODELOCKED,
8531 		    EXTATTR_NAMESPACE_SYSTEM, "pnfsd.dsfile",
8532 		    sizeof(*wpf) * mirrorcnt, (char *)wpf, p);
8533 	if (mp != NULL)
8534 		vn_finished_write(mp);
8535 
8536 	/* Get rid of the dontlist entry, so that Layouts can be issued. */
8537 	NFSDDONTLISTLOCK();
8538 	LIST_REMOVE(mrp, nfsmr_list);
8539 	NFSDDONTLISTUNLOCK();
8540 	free(mrp, M_NFSDSTATE);
8541 	return (ret);
8542 }
8543 
8544 /*
8545  * Create a data storage file on the recovered DS.
8546  */
8547 static int
8548 nfsrv_createdsfile(vnode_t vp, fhandle_t *fhp, struct pnfsdsfile *pf,
8549     vnode_t dvp, struct nfsdevice *ds, struct ucred *cred, NFSPROC_T *p,
8550     vnode_t *tvpp)
8551 {
8552 	struct vattr va, nva;
8553 	int error;
8554 
8555 	/* Make data file name based on FH. */
8556 	error = VOP_GETATTR(vp, &va, cred);
8557 	if (error == 0) {
8558 		/* Set the attributes for "vp" to Setattr the DS vp. */
8559 		VATTR_NULL(&nva);
8560 		nva.va_uid = va.va_uid;
8561 		nva.va_gid = va.va_gid;
8562 		nva.va_mode = va.va_mode;
8563 		nva.va_size = 0;
8564 		VATTR_NULL(&va);
8565 		va.va_type = VREG;
8566 		va.va_mode = nva.va_mode;
8567 		NFSD_DEBUG(4, "nfsrv_dscreatefile: dvp=%p pf=%p\n", dvp, pf);
8568 		error = nfsrv_dscreate(dvp, &va, &nva, fhp, pf, NULL,
8569 		    pf->dsf_filename, cred, p, tvpp);
8570 	}
8571 	return (error);
8572 }
8573 
8574 /*
8575  * Look up the MDS file shared locked, and then get the extended attribute
8576  * to find the extant DS file to be copied to the new mirror.
8577  * If successful, *vpp is set to the MDS file's vp and *nvpp is
8578  * set to a DS data file for the MDS file, both exclusively locked.
8579  * The "buf" argument has the pnfsdsfile structure from the MDS file
8580  * in it and buflen is set to its length.
8581  */
8582 int
8583 nfsrv_mdscopymr(char *mdspathp, char *dspathp, char *curdspathp, char *buf,
8584     int *buflenp, char *fname, NFSPROC_T *p, struct vnode **vpp,
8585     struct vnode **nvpp, struct pnfsdsfile **pfp, struct nfsdevice **dsp,
8586     struct nfsdevice **fdsp)
8587 {
8588 	struct nameidata nd;
8589 	struct vnode *vp, *curvp;
8590 	struct pnfsdsfile *pf;
8591 	struct nfsmount *nmp, *curnmp;
8592 	int dsdir, error, mirrorcnt, ippos;
8593 
8594 	vp = NULL;
8595 	curvp = NULL;
8596 	curnmp = NULL;
8597 	*dsp = NULL;
8598 	*fdsp = NULL;
8599 	if (dspathp == NULL && curdspathp != NULL)
8600 		return (EPERM);
8601 
8602 	/*
8603 	 * Look up the MDS file shared locked.  The lock will be upgraded
8604 	 * to an exclusive lock after any rw layouts have been returned.
8605 	 */
8606 	NFSD_DEBUG(4, "mdsopen path=%s\n", mdspathp);
8607 	NDINIT(&nd, LOOKUP, FOLLOW | LOCKSHARED | LOCKLEAF, UIO_SYSSPACE,
8608 	    mdspathp);
8609 	error = namei(&nd);
8610 	NFSD_DEBUG(4, "lookup=%d\n", error);
8611 	if (error != 0)
8612 		return (error);
8613 	if (nd.ni_vp->v_type != VREG) {
8614 		vput(nd.ni_vp);
8615 		NFSD_DEBUG(4, "mdspath not reg\n");
8616 		return (EISDIR);
8617 	}
8618 	vp = nd.ni_vp;
8619 
8620 	if (curdspathp != NULL) {
8621 		/*
8622 		 * Look up the current DS path and find the nfsdev structure for
8623 		 * it.
8624 		 */
8625 		NFSD_DEBUG(4, "curmdsdev path=%s\n", curdspathp);
8626 		NDINIT(&nd, LOOKUP, FOLLOW | LOCKSHARED | LOCKLEAF,
8627 		    UIO_SYSSPACE, curdspathp);
8628 		error = namei(&nd);
8629 		NFSD_DEBUG(4, "ds lookup=%d\n", error);
8630 		if (error != 0) {
8631 			vput(vp);
8632 			return (error);
8633 		}
8634 		if (nd.ni_vp->v_type != VDIR) {
8635 			vput(nd.ni_vp);
8636 			vput(vp);
8637 			NFSD_DEBUG(4, "curdspath not dir\n");
8638 			return (ENOTDIR);
8639 		}
8640 		if (strcmp(nd.ni_vp->v_mount->mnt_vfc->vfc_name, "nfs") != 0) {
8641 			vput(nd.ni_vp);
8642 			vput(vp);
8643 			NFSD_DEBUG(4, "curdspath not an NFS mount\n");
8644 			return (ENXIO);
8645 		}
8646 		curnmp = VFSTONFS(nd.ni_vp->v_mount);
8647 
8648 		/* Search the nfsdev list for a match. */
8649 		NFSDDSLOCK();
8650 		*fdsp = nfsv4_findmirror(curnmp);
8651 		NFSDDSUNLOCK();
8652 		if (*fdsp == NULL)
8653 			curnmp = NULL;
8654 		if (curnmp == NULL) {
8655 			vput(nd.ni_vp);
8656 			vput(vp);
8657 			NFSD_DEBUG(4, "mdscopymr: no current ds\n");
8658 			return (ENXIO);
8659 		}
8660 		curvp = nd.ni_vp;
8661 	}
8662 
8663 	if (dspathp != NULL) {
8664 		/* Look up the nfsdev path and find the nfsdev structure. */
8665 		NFSD_DEBUG(4, "mdsdev path=%s\n", dspathp);
8666 		NDINIT(&nd, LOOKUP, FOLLOW | LOCKSHARED | LOCKLEAF,
8667 		    UIO_SYSSPACE, dspathp);
8668 		error = namei(&nd);
8669 		NFSD_DEBUG(4, "ds lookup=%d\n", error);
8670 		if (error != 0) {
8671 			vput(vp);
8672 			if (curvp != NULL)
8673 				vput(curvp);
8674 			return (error);
8675 		}
8676 		if (nd.ni_vp->v_type != VDIR || nd.ni_vp == curvp) {
8677 			vput(nd.ni_vp);
8678 			vput(vp);
8679 			if (curvp != NULL)
8680 				vput(curvp);
8681 			NFSD_DEBUG(4, "dspath not dir\n");
8682 			if (nd.ni_vp == curvp)
8683 				return (EPERM);
8684 			return (ENOTDIR);
8685 		}
8686 		if (strcmp(nd.ni_vp->v_mount->mnt_vfc->vfc_name, "nfs") != 0) {
8687 			vput(nd.ni_vp);
8688 			vput(vp);
8689 			if (curvp != NULL)
8690 				vput(curvp);
8691 			NFSD_DEBUG(4, "dspath not an NFS mount\n");
8692 			return (ENXIO);
8693 		}
8694 		nmp = VFSTONFS(nd.ni_vp->v_mount);
8695 
8696 		/*
8697 		 * Search the nfsdevice list for a match.  If curnmp == NULL,
8698 		 * this is a recovery and there must be a mirror.
8699 		 */
8700 		NFSDDSLOCK();
8701 		if (curnmp == NULL)
8702 			*dsp = nfsrv_findmirroredds(nmp);
8703 		else
8704 			*dsp = nfsv4_findmirror(nmp);
8705 		NFSDDSUNLOCK();
8706 		if (*dsp == NULL) {
8707 			vput(nd.ni_vp);
8708 			vput(vp);
8709 			if (curvp != NULL)
8710 				vput(curvp);
8711 			NFSD_DEBUG(4, "mdscopymr: no ds\n");
8712 			return (ENXIO);
8713 		}
8714 	} else {
8715 		nd.ni_vp = NULL;
8716 		nmp = NULL;
8717 	}
8718 
8719 	/*
8720 	 * Get a vp for an available DS data file using the extended
8721 	 * attribute on the MDS file.
8722 	 * If there is a valid entry for the new DS in the extended attribute
8723 	 * on the MDS file (as checked via the nmp argument),
8724 	 * nfsrv_dsgetsockmnt() returns EEXIST, so no copying will occur.
8725 	 */
8726 	error = nfsrv_dsgetsockmnt(vp, 0, buf, buflenp, &mirrorcnt, p,
8727 	    NULL, NULL, NULL, fname, nvpp, &nmp, curnmp, &ippos, &dsdir);
8728 	if (curvp != NULL)
8729 		vput(curvp);
8730 	if (nd.ni_vp == NULL) {
8731 		if (error == 0 && nmp != NULL) {
8732 			/* Search the nfsdev list for a match. */
8733 			NFSDDSLOCK();
8734 			*dsp = nfsrv_findmirroredds(nmp);
8735 			NFSDDSUNLOCK();
8736 		}
8737 		if (error == 0 && (nmp == NULL || *dsp == NULL)) {
8738 			if (nvpp != NULL && *nvpp != NULL) {
8739 				vput(*nvpp);
8740 				*nvpp = NULL;
8741 			}
8742 			error = ENXIO;
8743 		}
8744 	} else
8745 		vput(nd.ni_vp);
8746 
8747 	/*
8748 	 * When dspathp != NULL and curdspathp == NULL, this is a recovery
8749 	 * and is only allowed if there is a 0.0.0.0 IP address entry.
8750 	 * When curdspathp != NULL, the ippos will be set to that entry.
8751 	 */
8752 	if (error == 0 && dspathp != NULL && ippos == -1) {
8753 		if (nvpp != NULL && *nvpp != NULL) {
8754 			vput(*nvpp);
8755 			*nvpp = NULL;
8756 		}
8757 		error = ENXIO;
8758 	}
8759 	if (error == 0) {
8760 		*vpp = vp;
8761 
8762 		pf = (struct pnfsdsfile *)buf;
8763 		if (ippos == -1) {
8764 			/* If no zeroip pnfsdsfile, add one. */
8765 			ippos = *buflenp / sizeof(*pf);
8766 			*buflenp += sizeof(*pf);
8767 			pf += ippos;
8768 			pf->dsf_dir = dsdir;
8769 			strlcpy(pf->dsf_filename, fname,
8770 			    sizeof(pf->dsf_filename));
8771 		} else
8772 			pf += ippos;
8773 		*pfp = pf;
8774 	} else
8775 		vput(vp);
8776 	return (error);
8777 }
8778 
8779 /*
8780  * Search for a matching pnfsd mirror device structure, base on the nmp arg.
8781  * Return one if found, NULL otherwise.
8782  */
8783 static struct nfsdevice *
8784 nfsrv_findmirroredds(struct nfsmount *nmp)
8785 {
8786 	struct nfsdevice *ds, *fndds;
8787 	int fndmirror;
8788 
8789 	mtx_assert(NFSDDSMUTEXPTR, MA_OWNED);
8790 	/*
8791 	 * Search the DS server list for a match with nmp.
8792 	 * Remove the DS entry if found and there is a mirror.
8793 	 */
8794 	fndds = NULL;
8795 	fndmirror = 0;
8796 	if (nfsrv_devidcnt == 0)
8797 		return (fndds);
8798 	TAILQ_FOREACH(ds, &nfsrv_devidhead, nfsdev_list) {
8799 		if (ds->nfsdev_nmp == nmp) {
8800 			NFSD_DEBUG(4, "nfsrv_findmirroredds: fnd main ds\n");
8801 			fndds = ds;
8802 			break;
8803 		}
8804 	}
8805 	if (fndds == NULL)
8806 		return (fndds);
8807 	if (fndds->nfsdev_mdsisset == 0 && nfsrv_faildscnt > 0)
8808 		fndmirror = 1;
8809 	else if (fndds->nfsdev_mdsisset != 0) {
8810 		/* For the fsid is set case, search for a mirror. */
8811 		TAILQ_FOREACH(ds, &nfsrv_devidhead, nfsdev_list) {
8812 			if (ds != fndds && ds->nfsdev_nmp != NULL &&
8813 			    ds->nfsdev_mdsisset != 0 &&
8814 			    fsidcmp(&ds->nfsdev_mdsfsid,
8815 			    &fndds->nfsdev_mdsfsid) == 0) {
8816 				fndmirror = 1;
8817 				break;
8818 			}
8819 		}
8820 	}
8821 	if (fndmirror == 0) {
8822 		NFSD_DEBUG(4, "nfsrv_findmirroredds: no mirror for DS\n");
8823 		return (NULL);
8824 	}
8825 	return (fndds);
8826 }
8827 
8828 /*
8829  * Mark the appropriate devid and all associated layout as "out of space".
8830  */
8831 void
8832 nfsrv_marknospc(char *devid, bool setit)
8833 {
8834 	struct nfsdevice *ds;
8835 	struct nfslayout *lyp;
8836 	struct nfslayouthash *lhyp;
8837 	int i;
8838 
8839 	NFSDDSLOCK();
8840 	TAILQ_FOREACH(ds, &nfsrv_devidhead, nfsdev_list) {
8841 		if (NFSBCMP(ds->nfsdev_deviceid, devid, NFSX_V4DEVICEID) == 0) {
8842 			NFSD_DEBUG(1, "nfsrv_marknospc: devid %d\n", setit);
8843 			ds->nfsdev_nospc = setit;
8844 		}
8845 	}
8846 	NFSDDSUNLOCK();
8847 
8848 	for (i = 0; i < nfsrv_layouthashsize; i++) {
8849 		lhyp = &nfslayouthash[i];
8850 		NFSLOCKLAYOUT(lhyp);
8851 		TAILQ_FOREACH(lyp, &lhyp->list, lay_list) {
8852 			if (NFSBCMP(lyp->lay_deviceid, devid,
8853 			    NFSX_V4DEVICEID) == 0) {
8854 				NFSD_DEBUG(1, "nfsrv_marknospc: layout %d\n",
8855 				    setit);
8856 				if (setit)
8857 					lyp->lay_flags |= NFSLAY_NOSPC;
8858 				else
8859 					lyp->lay_flags &= ~NFSLAY_NOSPC;
8860 			}
8861 		}
8862 		NFSUNLOCKLAYOUT(lhyp);
8863 	}
8864 }
8865