xref: /freebsd/sys/fs/nfsserver/nfs_nfsdstate.c (revision 7937bfbc0ca53fe7cdd0d54414f9296e273a518e)
1 /*-
2  * SPDX-License-Identifier: BSD-2-Clause
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 #include "opt_inet.h"
32 #include "opt_inet6.h"
33 #include <sys/extattr.h>
34 #include <fs/nfs/nfsport.h>
35 
36 int nfsrv_issuedelegs = 0;
37 int nfsrv_dolocallocks = 0;
38 struct nfsv4lock nfsv4rootfs_lock;
39 time_t nfsdev_time = 0;
40 int nfsrv_layouthashsize;
41 volatile int nfsrv_layoutcnt = 0;
42 
43 NFSD_VNET_DEFINE(struct nfsrv_stablefirst, nfsrv_stablefirst);
44 
45 NFSD_VNET_DECLARE(int, nfsrv_numnfsd);
46 NFSD_VNET_DECLARE(struct nfsstatsv1 *, nfsstatsv1_p);
47 
48 extern uint32_t nfs_srvmaxio;
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 NFSD_VNET_DEFINE(struct nfsclienthashhead *, nfsclienthash);
122 NFSD_VNET_DEFINE(struct nfslockhashhead *, nfslockhash);
123 NFSD_VNET_DEFINE(struct nfssessionhash *, nfssessionhash);
124 
125 struct nfslayouthash		*nfslayouthash;
126 volatile int nfsrv_dontlistlen = 0;
127 
128 static u_int32_t nfsrv_openpluslock = 0, nfsrv_delegatecnt = 0;
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 NFSD_VNET_DEFINE_STATIC(time_t, nfsrvboottime);
136 
137 /* local functions */
138 static void nfsrv_dumpaclient(struct nfsclient *clp,
139     struct nfsd_dumpclients *dumpp);
140 static void nfsrv_freeopenowner(struct nfsstate *stp, int cansleep,
141     NFSPROC_T *p);
142 static void nfsrv_freeopen(struct nfsstate *stp, vnode_t vp, int cansleep,
143     NFSPROC_T *p);
144 static void nfsrv_freelockowner(struct nfsstate *stp, vnode_t vp, int cansleep,
145     NFSPROC_T *p);
146 static void nfsrv_freeallnfslocks(struct nfsstate *stp, vnode_t vp,
147     int cansleep, NFSPROC_T *p);
148 static void nfsrv_freenfslock(struct nfslock *lop);
149 static void nfsrv_freenfslockfile(struct nfslockfile *lfp);
150 static void nfsrv_freedeleg(struct nfsstate *);
151 static int nfsrv_getstate(struct nfsclient *clp, nfsv4stateid_t *stateidp,
152     u_int32_t flags, struct nfsstate **stpp);
153 static void nfsrv_getowner(struct nfsstatehead *hp, struct nfsstate *new_stp,
154     struct nfsstate **stpp);
155 static int nfsrv_getlockfh(vnode_t vp, u_short flags,
156     struct nfslockfile *new_lfp, fhandle_t *nfhp, NFSPROC_T *p);
157 static int nfsrv_getlockfile(u_short flags, struct nfslockfile **new_lfpp,
158     struct nfslockfile **lfpp, fhandle_t *nfhp, int lockit);
159 static void nfsrv_insertlock(struct nfslock *new_lop,
160     struct nfslock *insert_lop, struct nfsstate *stp, struct nfslockfile *lfp);
161 static void nfsrv_updatelock(struct nfsstate *stp, struct nfslock **new_lopp,
162     struct nfslock **other_lopp, struct nfslockfile *lfp);
163 static int nfsrv_getipnumber(u_char *cp);
164 static int nfsrv_checkrestart(nfsquad_t clientid, u_int32_t flags,
165     nfsv4stateid_t *stateidp, int specialid);
166 static int nfsrv_checkgrace(struct nfsrv_descript *nd, struct nfsclient *clp,
167     u_int32_t flags);
168 static int nfsrv_docallback(struct nfsclient *clp, int procnum,
169     nfsv4stateid_t *stateidp, int trunc, fhandle_t *fhp,
170     struct nfsvattr *nap, nfsattrbit_t *attrbitp, int laytype, NFSPROC_T *p);
171 static int nfsrv_cbcallargs(struct nfsrv_descript *nd, struct nfsclient *clp,
172     uint32_t callback, int op, const char *optag, struct nfsdsession **sepp,
173     int *slotposp);
174 static u_int32_t nfsrv_nextclientindex(void);
175 static u_int32_t nfsrv_nextstateindex(struct nfsclient *clp);
176 static void nfsrv_markstable(struct nfsclient *clp);
177 static void nfsrv_markreclaim(struct nfsclient *clp);
178 static int nfsrv_checkstable(struct nfsclient *clp);
179 static int nfsrv_clientconflict(struct nfsclient *clp, int *haslockp, struct
180     vnode *vp, NFSPROC_T *p);
181 static int nfsrv_delegconflict(struct nfsstate *stp, int *haslockp,
182     NFSPROC_T *p, vnode_t vp);
183 static int nfsrv_cleandeleg(vnode_t vp, struct nfslockfile *lfp,
184     struct nfsclient *clp, int *haslockp, NFSPROC_T *p);
185 static int nfsrv_notsamecredname(int op, struct nfsrv_descript *nd,
186     struct nfsclient *clp);
187 static time_t nfsrv_leaseexpiry(void);
188 static void nfsrv_delaydelegtimeout(struct nfsstate *stp);
189 static int nfsrv_checkseqid(struct nfsrv_descript *nd, u_int32_t seqid,
190     struct nfsstate *stp, struct nfsrvcache *op);
191 static int nfsrv_nootherstate(struct nfsstate *stp);
192 static int nfsrv_locallock(vnode_t vp, struct nfslockfile *lfp, int flags,
193     uint64_t first, uint64_t end, struct nfslockconflict *cfp, NFSPROC_T *p);
194 static void nfsrv_localunlock(vnode_t vp, struct nfslockfile *lfp,
195     uint64_t init_first, uint64_t init_end, NFSPROC_T *p);
196 static int nfsrv_dolocal(vnode_t vp, struct nfslockfile *lfp, int flags,
197     int oldflags, uint64_t first, uint64_t end, struct nfslockconflict *cfp,
198     NFSPROC_T *p);
199 static void nfsrv_locallock_rollback(vnode_t vp, struct nfslockfile *lfp,
200     NFSPROC_T *p);
201 static void nfsrv_locallock_commit(struct nfslockfile *lfp, int flags,
202     uint64_t first, uint64_t end);
203 static void nfsrv_locklf(struct nfslockfile *lfp);
204 static void nfsrv_unlocklf(struct nfslockfile *lfp);
205 static struct nfsdsession *nfsrv_findsession(uint8_t *sessionid);
206 static int nfsrv_freesession(struct nfsrv_descript *nd, struct nfsdsession *sep,
207     uint8_t *sessionid, bool locked, SVCXPRT **old_xprtp);
208 static int nfsv4_setcbsequence(struct nfsrv_descript *nd, struct nfsclient *clp,
209     int dont_replycache, struct nfsdsession **sepp, int *slotposp);
210 static int nfsv4_getcbsession(struct nfsclient *clp, struct nfsdsession **sepp);
211 static int nfsrv_addlayout(struct nfsrv_descript *nd, struct nfslayout **lypp,
212     nfsv4stateid_t *stateidp, char *layp, int *layoutlenp, NFSPROC_T *p);
213 static void nfsrv_freelayout(struct nfslayouthead *lhp, struct nfslayout *lyp);
214 static void nfsrv_freelayoutlist(nfsquad_t clientid);
215 static void nfsrv_freelayouts(nfsquad_t *clid, fsid_t *fs, int laytype,
216     int iomode);
217 static void nfsrv_freealllayouts(void);
218 static void nfsrv_freedevid(struct nfsdevice *ds);
219 static int nfsrv_setdsserver(char *dspathp, char *mdspathp, NFSPROC_T *p,
220     struct nfsdevice **dsp);
221 static void nfsrv_deleteds(struct nfsdevice *fndds);
222 static void nfsrv_allocdevid(struct nfsdevice *ds, char *addr, char *dnshost);
223 static void nfsrv_freealldevids(void);
224 static void nfsrv_flexlayouterr(struct nfsrv_descript *nd, uint32_t *layp,
225     int maxcnt, NFSPROC_T *p);
226 static int nfsrv_recalllayout(nfsquad_t clid, nfsv4stateid_t *stateidp,
227     fhandle_t *fhp, struct nfslayout *lyp, int changed, int laytype,
228     NFSPROC_T *p);
229 static int nfsrv_findlayout(nfsquad_t *clientidp, fhandle_t *fhp, int laytype,
230     NFSPROC_T *, struct nfslayout **lypp);
231 static int nfsrv_fndclid(nfsquad_t *clidvec, nfsquad_t clid, int clidcnt);
232 static struct nfslayout *nfsrv_filelayout(struct nfsrv_descript *nd, int iomode,
233     fhandle_t *fhp, fhandle_t *dsfhp, char *devid, fsid_t fs);
234 static struct nfslayout *nfsrv_flexlayout(struct nfsrv_descript *nd, int iomode,
235     int mirrorcnt, fhandle_t *fhp, fhandle_t *dsfhp, char *devid, fsid_t fs);
236 static int nfsrv_dontlayout(fhandle_t *fhp);
237 static int nfsrv_createdsfile(vnode_t vp, fhandle_t *fhp, struct pnfsdsfile *pf,
238     vnode_t dvp, struct nfsdevice *ds, struct ucred *cred, NFSPROC_T *p,
239     vnode_t *tvpp);
240 static struct nfsdevice *nfsrv_findmirroredds(struct nfsmount *nmp);
241 static int nfsrv_checkmachcred(int op, struct nfsrv_descript *nd,
242     struct nfsclient *clp);
243 static void nfsrv_issuedelegation(struct vnode *vp, struct nfsclient *clp,
244     struct nfsrv_descript *nd, int delegate, int writedeleg, int readonly,
245     u_quad_t filerev, uint64_t rdonly, struct nfsstate **new_delegp,
246     struct nfsstate *new_stp, struct nfslockfile *lfp, uint32_t *rflagsp,
247     nfsv4stateid_t *delegstateidp);
248 static void nfsrv_clientlock(bool mlocked);
249 static void nfsrv_clientunlock(bool mlocked);
250 
251 /*
252  * Lock the client structure, either with the mutex or the exclusive nfsd lock.
253  */
254 static void
255 nfsrv_clientlock(bool mlocked)
256 {
257 	int igotlock;
258 
259 	if (mlocked) {
260 		NFSLOCKSTATE();
261 	} else {
262 		NFSLOCKV4ROOTMUTEX();
263 		nfsv4_relref(&nfsv4rootfs_lock);
264 		do {
265 			igotlock = nfsv4_lock(&nfsv4rootfs_lock, 1, NULL,
266 			    NFSV4ROOTLOCKMUTEXPTR, NULL);
267 		} while (!igotlock);
268 		NFSUNLOCKV4ROOTMUTEX();
269 	}
270 }
271 
272 /*
273  * Unlock the client structure.
274  */
275 static void
276 nfsrv_clientunlock(bool mlocked)
277 {
278 
279 	if (mlocked) {
280 		NFSUNLOCKSTATE();
281 	} else {
282 		NFSLOCKV4ROOTMUTEX();
283 		nfsv4_unlock(&nfsv4rootfs_lock, 1);
284 		NFSUNLOCKV4ROOTMUTEX();
285 	}
286 }
287 
288 /*
289  * Scan the client list for a match and either return the current one,
290  * create a new entry or return an error.
291  * If returning a non-error, the clp structure must either be linked into
292  * the client list or free'd.
293  */
294 int
295 nfsrv_setclient(struct nfsrv_descript *nd, struct nfsclient **new_clpp,
296     nfsquad_t *clientidp, nfsquad_t *confirmp, NFSPROC_T *p)
297 {
298 	struct nfsclient *clp = NULL, *new_clp = *new_clpp;
299 	int i, error = 0, ret;
300 	struct nfsstate *stp, *tstp;
301 #ifdef INET
302 	struct sockaddr_in *sin, *rin;
303 #endif
304 #ifdef INET6
305 	struct sockaddr_in6 *sin6, *rin6;
306 #endif
307 	struct nfsdsession *sep, *nsep;
308 	SVCXPRT *old_xprt;
309 	struct nfssessionhead old_sess;
310 	int zapit = 0, gotit, hasstate = 0;
311 	bool mlocked;
312 	static u_int64_t confirm_index = 0;
313 
314 	/*
315 	 * Check for state resource limit exceeded.
316 	 */
317 	if (nfsrv_openpluslock > nfsrv_v4statelimit) {
318 		error = NFSERR_RESOURCE;
319 		goto out;
320 	}
321 
322 	if (nfsrv_issuedelegs == 0 ||
323 	    ((nd->nd_flag & ND_GSS) != 0 && nfsrv_nogsscallback != 0))
324 		/*
325 		 * Don't do callbacks when delegations are disabled or
326 		 * for AUTH_GSS unless enabled via nfsrv_nogsscallback.
327 		 * If establishing a callback connection is attempted
328 		 * when a firewall is blocking the callback path, the
329 		 * server may wait too long for the connect attempt to
330 		 * succeed during the Open. Some clients, such as Linux,
331 		 * may timeout and give up on the Open before the server
332 		 * replies. Also, since AUTH_GSS callbacks are not
333 		 * yet interoperability tested, they might cause the
334 		 * server to crap out, if they get past the Init call to
335 		 * the client.
336 		 */
337 		new_clp->lc_program = 0;
338 
339 	mlocked = true;
340 	if (nfsrv_dolocallocks != 0)
341 		mlocked = false;
342 	/* Lock out other nfsd threads */
343 	nfsrv_clientlock(mlocked);
344 
345 	/*
346 	 * Search for a match in the client list.
347 	 */
348 	gotit = i = 0;
349 	while (i < nfsrv_clienthashsize && !gotit) {
350 	    LIST_FOREACH(clp, &NFSD_VNET(nfsclienthash)[i], lc_hash) {
351 		if (new_clp->lc_idlen == clp->lc_idlen &&
352 		    !NFSBCMP(new_clp->lc_id, clp->lc_id, clp->lc_idlen)) {
353 			gotit = 1;
354 			break;
355 		}
356 	    }
357 	    if (gotit == 0)
358 		i++;
359 	}
360 	old_xprt = NULL;
361 	if (!gotit ||
362 	    (clp->lc_flags & (LCL_NEEDSCONFIRM | LCL_ADMINREVOKED))) {
363 		if ((nd->nd_flag & ND_NFSV41) != 0 && confirmp->lval[1] != 0) {
364 			/*
365 			 * For NFSv4.1, if confirmp->lval[1] is non-zero, the
366 			 * client is trying to update a confirmed clientid.
367 			 */
368 			nfsrv_clientunlock(mlocked);
369 			confirmp->lval[1] = 0;
370 			error = NFSERR_NOENT;
371 			goto out;
372 		}
373 		/*
374 		 * Get rid of the old one.
375 		 */
376 		if (i != nfsrv_clienthashsize) {
377 			LIST_REMOVE(clp, lc_hash);
378 			if (mlocked)
379 				nfsrv_cleanclient(clp, p, true, &old_xprt);
380 			else
381 				nfsrv_cleanclient(clp, p, false, NULL);
382 			nfsrv_freedeleglist(&clp->lc_deleg);
383 			nfsrv_freedeleglist(&clp->lc_olddeleg);
384 			zapit = 1;
385 		}
386 		/*
387 		 * Add it after assigning a client id to it.
388 		 */
389 		new_clp->lc_flags |= LCL_NEEDSCONFIRM;
390 		if ((nd->nd_flag & ND_NFSV41) != 0) {
391 			confirmp->lval[0] = ++confirm_index;
392 			new_clp->lc_confirm.lval[0] = confirmp->lval[0] - 1;
393 		} else
394 			confirmp->qval = new_clp->lc_confirm.qval =
395 			    ++confirm_index;
396 		clientidp->lval[0] = new_clp->lc_clientid.lval[0] =
397 		    NFSD_VNET(nfsrvboottime);
398 		clientidp->lval[1] = new_clp->lc_clientid.lval[1] =
399 		    nfsrv_nextclientindex();
400 		new_clp->lc_stateindex = 0;
401 		new_clp->lc_statemaxindex = 0;
402 		new_clp->lc_prevsess = 0;
403 		new_clp->lc_cbref = 0;
404 		new_clp->lc_expiry = nfsrv_leaseexpiry();
405 		LIST_INIT(&new_clp->lc_open);
406 		LIST_INIT(&new_clp->lc_deleg);
407 		LIST_INIT(&new_clp->lc_olddeleg);
408 		LIST_INIT(&new_clp->lc_session);
409 		for (i = 0; i < nfsrv_statehashsize; i++)
410 			LIST_INIT(&new_clp->lc_stateid[i]);
411 		LIST_INSERT_HEAD(NFSCLIENTHASH(new_clp->lc_clientid), new_clp,
412 		    lc_hash);
413 		NFSD_VNET(nfsstatsv1_p)->srvclients++;
414 		nfsrv_openpluslock++;
415 		nfsrv_clients++;
416 		nfsrv_clientunlock(mlocked);
417 		if (zapit != 0) {
418 			if (old_xprt != NULL)
419 				SVC_RELEASE(old_xprt);
420 			nfsrv_zapclient(clp, p);
421 		}
422 		*new_clpp = NULL;
423 		goto out;
424 	}
425 
426 	/*
427 	 * Now, handle the cases where the id is already issued.
428 	 */
429 	if (nfsrv_notsamecredname(NFSV4OP_EXCHANGEID, nd, clp)) {
430 	    /*
431 	     * Check to see if there is expired state that should go away.
432 	     */
433 	    if (clp->lc_expiry < NFSD_MONOSEC &&
434 	        (!LIST_EMPTY(&clp->lc_open) || !LIST_EMPTY(&clp->lc_deleg))) {
435 		if (mlocked)
436 		    nfsrv_cleanclient(clp, p, true, &old_xprt);
437 		else
438 		    nfsrv_cleanclient(clp, p, false, NULL);
439 		nfsrv_freedeleglist(&clp->lc_deleg);
440 	    }
441 
442 	    /*
443 	     * If there is outstanding state, then reply NFSERR_CLIDINUSE per
444 	     * RFC3530 Sec. 8.1.2 last para.
445 	     */
446 	    if (!LIST_EMPTY(&clp->lc_deleg)) {
447 		hasstate = 1;
448 	    } else if (LIST_EMPTY(&clp->lc_open)) {
449 		hasstate = 0;
450 	    } else {
451 		hasstate = 0;
452 		/* Look for an Open on the OpenOwner */
453 		LIST_FOREACH(stp, &clp->lc_open, ls_list) {
454 		    if (!LIST_EMPTY(&stp->ls_open)) {
455 			hasstate = 1;
456 			break;
457 		    }
458 		}
459 	    }
460 	    if (hasstate) {
461 		/*
462 		 * If the uid doesn't match, return NFSERR_CLIDINUSE after
463 		 * filling out the correct ipaddr and portnum.
464 		 */
465 		switch (clp->lc_req.nr_nam->sa_family) {
466 #ifdef INET
467 		case AF_INET:
468 			sin = (struct sockaddr_in *)new_clp->lc_req.nr_nam;
469 			rin = (struct sockaddr_in *)clp->lc_req.nr_nam;
470 			sin->sin_addr.s_addr = rin->sin_addr.s_addr;
471 			sin->sin_port = rin->sin_port;
472 			break;
473 #endif
474 #ifdef INET6
475 		case AF_INET6:
476 			sin6 = (struct sockaddr_in6 *)new_clp->lc_req.nr_nam;
477 			rin6 = (struct sockaddr_in6 *)clp->lc_req.nr_nam;
478 			sin6->sin6_addr = rin6->sin6_addr;
479 			sin6->sin6_port = rin6->sin6_port;
480 			break;
481 #endif
482 		}
483 		nfsrv_clientunlock(mlocked);
484 		if (old_xprt != NULL)
485 			SVC_RELEASE(old_xprt);
486 		error = NFSERR_CLIDINUSE;
487 		goto out;
488 	    }
489 	}
490 
491 	if (NFSBCMP(new_clp->lc_verf, clp->lc_verf, NFSX_VERF)) {
492 		/*
493 		 * If the verifier has changed, the client has rebooted
494 		 * and a new client id is issued. The old state info
495 		 * can be thrown away once the SetClientID_Confirm or
496 		 * Create_Session that confirms the clientid occurs.
497 		 */
498 		LIST_REMOVE(clp, lc_hash);
499 
500 		LIST_NEWHEAD(&old_sess, &clp->lc_session, sess_list);
501 
502 		new_clp->lc_flags |= LCL_NEEDSCONFIRM;
503 		if ((nd->nd_flag & ND_NFSV41) != 0) {
504 			confirmp->lval[0] = ++confirm_index;
505 			new_clp->lc_confirm.lval[0] = confirmp->lval[0] - 1;
506 		} else
507 			confirmp->qval = new_clp->lc_confirm.qval =
508 			    ++confirm_index;
509 		clientidp->lval[0] = new_clp->lc_clientid.lval[0] =
510 		    NFSD_VNET(nfsrvboottime);
511 		clientidp->lval[1] = new_clp->lc_clientid.lval[1] =
512 		    nfsrv_nextclientindex();
513 		new_clp->lc_stateindex = 0;
514 		new_clp->lc_statemaxindex = 0;
515 		new_clp->lc_prevsess = 0;
516 		new_clp->lc_cbref = 0;
517 		new_clp->lc_expiry = nfsrv_leaseexpiry();
518 
519 		/*
520 		 * Save the state until confirmed.
521 		 */
522 		LIST_NEWHEAD(&new_clp->lc_open, &clp->lc_open, ls_list);
523 		LIST_FOREACH(tstp, &new_clp->lc_open, ls_list)
524 			tstp->ls_clp = new_clp;
525 		LIST_NEWHEAD(&new_clp->lc_deleg, &clp->lc_deleg, ls_list);
526 		LIST_FOREACH(tstp, &new_clp->lc_deleg, ls_list)
527 			tstp->ls_clp = new_clp;
528 		LIST_NEWHEAD(&new_clp->lc_olddeleg, &clp->lc_olddeleg,
529 		    ls_list);
530 		LIST_FOREACH(tstp, &new_clp->lc_olddeleg, ls_list)
531 			tstp->ls_clp = new_clp;
532 		for (i = 0; i < nfsrv_statehashsize; i++) {
533 			LIST_NEWHEAD(&new_clp->lc_stateid[i],
534 			    &clp->lc_stateid[i], ls_hash);
535 			LIST_FOREACH(tstp, &new_clp->lc_stateid[i], ls_hash)
536 				tstp->ls_clp = new_clp;
537 		}
538 		LIST_INIT(&new_clp->lc_session);
539 		LIST_INSERT_HEAD(NFSCLIENTHASH(new_clp->lc_clientid), new_clp,
540 		    lc_hash);
541 		NFSD_VNET(nfsstatsv1_p)->srvclients++;
542 		nfsrv_openpluslock++;
543 		nfsrv_clients++;
544 		if (!mlocked) {
545 			nfsrv_clientunlock(mlocked);
546 			NFSLOCKSTATE();
547 		}
548 
549 		/*
550 		 * Must wait until any outstanding callback on the old clp
551 		 * completes.
552 		 */
553 		while (clp->lc_cbref) {
554 			clp->lc_flags |= LCL_WAKEUPWANTED;
555 			(void)mtx_sleep(clp, NFSSTATEMUTEXPTR, PZERO - 1,
556 			    "nfsd clp", 10 * hz);
557 		}
558 		NFSUNLOCKSTATE();
559 		if (old_xprt != NULL)
560 			SVC_RELEASE(old_xprt);
561 		/* Get rid of all sessions on this clientid. */
562 		LIST_FOREACH_SAFE(sep, &old_sess, sess_list, nsep) {
563 			ret = nfsrv_freesession(NULL, sep, NULL, false, NULL);
564 			if (ret != 0)
565 				printf("nfsrv_setclient: verifier changed free"
566 				    " session failed=%d\n", ret);
567 		}
568 
569 		nfsrv_zapclient(clp, p);
570 		*new_clpp = NULL;
571 		goto out;
572 	}
573 
574 	/* For NFSv4.1, mark that we found a confirmed clientid. */
575 	if ((nd->nd_flag & ND_NFSV41) != 0) {
576 		clientidp->lval[0] = clp->lc_clientid.lval[0];
577 		clientidp->lval[1] = clp->lc_clientid.lval[1];
578 		confirmp->lval[0] = 0;	/* Ignored by client */
579 		confirmp->lval[1] = 1;
580 	} else {
581 		/*
582 		 * id and verifier match, so update the net address info
583 		 * and get rid of any existing callback authentication
584 		 * handle, so a new one will be acquired.
585 		 */
586 		LIST_REMOVE(clp, lc_hash);
587 		new_clp->lc_flags |= (LCL_NEEDSCONFIRM | LCL_DONTCLEAN);
588 		new_clp->lc_expiry = nfsrv_leaseexpiry();
589 		confirmp->qval = new_clp->lc_confirm.qval = ++confirm_index;
590 		clientidp->lval[0] = new_clp->lc_clientid.lval[0] =
591 		    clp->lc_clientid.lval[0];
592 		clientidp->lval[1] = new_clp->lc_clientid.lval[1] =
593 		    clp->lc_clientid.lval[1];
594 		new_clp->lc_delegtime = clp->lc_delegtime;
595 		new_clp->lc_stateindex = clp->lc_stateindex;
596 		new_clp->lc_statemaxindex = clp->lc_statemaxindex;
597 		new_clp->lc_cbref = 0;
598 		LIST_NEWHEAD(&new_clp->lc_open, &clp->lc_open, ls_list);
599 		LIST_FOREACH(tstp, &new_clp->lc_open, ls_list)
600 			tstp->ls_clp = new_clp;
601 		LIST_NEWHEAD(&new_clp->lc_deleg, &clp->lc_deleg, ls_list);
602 		LIST_FOREACH(tstp, &new_clp->lc_deleg, ls_list)
603 			tstp->ls_clp = new_clp;
604 		LIST_NEWHEAD(&new_clp->lc_olddeleg, &clp->lc_olddeleg, ls_list);
605 		LIST_FOREACH(tstp, &new_clp->lc_olddeleg, ls_list)
606 			tstp->ls_clp = new_clp;
607 		for (i = 0; i < nfsrv_statehashsize; i++) {
608 			LIST_NEWHEAD(&new_clp->lc_stateid[i],
609 			    &clp->lc_stateid[i], ls_hash);
610 			LIST_FOREACH(tstp, &new_clp->lc_stateid[i], ls_hash)
611 				tstp->ls_clp = new_clp;
612 		}
613 		LIST_INIT(&new_clp->lc_session);
614 		LIST_INSERT_HEAD(NFSCLIENTHASH(new_clp->lc_clientid), new_clp,
615 		    lc_hash);
616 		NFSD_VNET(nfsstatsv1_p)->srvclients++;
617 		nfsrv_openpluslock++;
618 		nfsrv_clients++;
619 	}
620 	if (!mlocked)
621 		nfsrv_clientunlock(mlocked);
622 
623 	if ((nd->nd_flag & ND_NFSV41) == 0) {
624 		/*
625 		 * Must wait until any outstanding callback on the old clp
626 		 * completes.
627 		 */
628 		if (!mlocked)
629 			NFSLOCKSTATE();
630 		while (clp->lc_cbref) {
631 			clp->lc_flags |= LCL_WAKEUPWANTED;
632 			(void)mtx_sleep(clp, NFSSTATEMUTEXPTR, PZERO - 1,
633 			    "nfsdclp", 10 * hz);
634 		}
635 		NFSUNLOCKSTATE();
636 		if (old_xprt != NULL)
637 			SVC_RELEASE(old_xprt);
638 		nfsrv_zapclient(clp, p);
639 		*new_clpp = NULL;
640 	} else {
641 		if (mlocked)
642 			NFSUNLOCKSTATE();
643 		if (old_xprt != NULL)
644 			SVC_RELEASE(old_xprt);
645 	}
646 
647 out:
648 	NFSEXITCODE2(error, nd);
649 	return (error);
650 }
651 
652 /*
653  * Check to see if the client id exists and optionally confirm it.
654  */
655 int
656 nfsrv_getclient(nfsquad_t clientid, int opflags, struct nfsclient **clpp,
657     struct nfsdsession *nsep, nfsquad_t confirm, uint32_t cbprogram,
658     struct nfsrv_descript *nd, NFSPROC_T *p)
659 {
660 	struct nfsclient *clp;
661 	struct nfsstate *stp;
662 	int i;
663 	struct nfsclienthashhead *hp;
664 	int error = 0, doneok, igotlock;
665 	struct nfssessionhash *shp;
666 	struct nfsdsession *sep;
667 	uint64_t sessid[2];
668 	CLIENT *client;
669 	SVCXPRT *old_xprt;
670 	bool mlocked, sess_replay;
671 	static uint64_t next_sess = 0;
672 
673 	if (clpp)
674 		*clpp = NULL;
675 	if ((nd == NULL || (nd->nd_flag & ND_NFSV41) == 0 ||
676 	    opflags != CLOPS_RENEW) && NFSD_VNET(nfsrvboottime) !=
677 	    clientid.lval[0]) {
678 		error = NFSERR_STALECLIENTID;
679 		goto out;
680 	}
681 
682 	/*
683 	 * If called with opflags == CLOPS_RENEW, the State Lock is
684 	 * already held. Otherwise, we need to get either that or,
685 	 * for the case of Confirm, lock out the nfsd threads.
686 	 */
687 	client = NULL;
688 	old_xprt = NULL;
689 	mlocked = true;
690 	if (nfsrv_dolocallocks != 0)
691 		mlocked = false;
692 	if (opflags & CLOPS_CONFIRM) {
693 		if (nsep != NULL &&
694 		    (nsep->sess_crflags & NFSV4CRSESS_CONNBACKCHAN) != 0)
695 			client = (struct __rpc_client *)
696 			    clnt_bck_create(nd->nd_xprt->xp_socket,
697 			    cbprogram, NFSV4_CBVERS);
698 		if (mlocked) {
699 			nfsrv_clientlock(mlocked);
700 		} else {
701 			NFSLOCKV4ROOTMUTEX();
702 			nfsv4_relref(&nfsv4rootfs_lock);
703 			do {
704 				igotlock = nfsv4_lock(&nfsv4rootfs_lock, 1,
705 				    NULL, NFSV4ROOTLOCKMUTEXPTR, NULL);
706 			} while (!igotlock);
707 		}
708 		/*
709 		 * Create a new sessionid here, since we need to do it where
710 		 * there is a mutex held to serialize update of next_sess.
711 		 */
712 		if ((nd->nd_flag & ND_NFSV41) != 0) {
713 			sessid[0] = ++next_sess;
714 			sessid[1] = clientid.qval;
715 		}
716 		if (!mlocked)
717 			NFSUNLOCKV4ROOTMUTEX();
718 	} else if (opflags != CLOPS_RENEW) {
719 		NFSLOCKSTATE();
720 	}
721 
722 	/* For NFSv4.1, the clp is acquired from the associated session. */
723 	if (nd != NULL && (nd->nd_flag & ND_NFSV41) != 0 &&
724 	    opflags == CLOPS_RENEW) {
725 		clp = NULL;
726 		if ((nd->nd_flag & ND_HASSEQUENCE) != 0) {
727 			shp = NFSSESSIONHASH(nd->nd_sessionid);
728 			NFSLOCKSESSION(shp);
729 			sep = nfsrv_findsession(nd->nd_sessionid);
730 			if (sep != NULL)
731 				clp = sep->sess_clp;
732 			NFSUNLOCKSESSION(shp);
733 		}
734 	} else {
735 		hp = NFSCLIENTHASH(clientid);
736 		LIST_FOREACH(clp, hp, lc_hash) {
737 			if (clp->lc_clientid.lval[1] == clientid.lval[1])
738 				break;
739 		}
740 	}
741 	if (clp == NULL) {
742 		if (opflags & CLOPS_CONFIRM)
743 			error = NFSERR_STALECLIENTID;
744 		else
745 			error = NFSERR_EXPIRED;
746 	} else if (clp->lc_flags & LCL_ADMINREVOKED) {
747 		/*
748 		 * If marked admin revoked, just return the error.
749 		 */
750 		error = NFSERR_ADMINREVOKED;
751 	}
752 	if (error) {
753 		if (opflags & CLOPS_CONFIRM) {
754 			nfsrv_clientunlock(mlocked);
755 			if (client != NULL)
756 				CLNT_RELEASE(client);
757 		} else if (opflags != CLOPS_RENEW) {
758 			NFSUNLOCKSTATE();
759 		}
760 		goto out;
761 	}
762 
763 	/*
764 	 * Perform any operations specified by the opflags.
765 	 */
766 	if (opflags & CLOPS_CONFIRM) {
767 		sess_replay = false;
768 		if ((nd->nd_flag & ND_NFSV41) != 0) {
769 		    /*
770 		     * For the case where lc_confirm.lval[0] == confirm.lval[0],
771 		     * use the new session, but with the previous sessionid.
772 		     * This is not exactly what the RFC describes, but should
773 		     * result in the same reply as the previous CreateSession.
774 		     */
775 		    if (clp->lc_confirm.lval[0] + 1 == confirm.lval[0]) {
776 			clp->lc_confirm.lval[0] = confirm.lval[0];
777 			clp->lc_prevsess = sessid[0];
778 		    } else if (clp->lc_confirm.lval[0] == confirm.lval[0]) {
779 			if (clp->lc_prevsess == 0)
780 			    error = NFSERR_SEQMISORDERED;
781 			else
782 			    sessid[0] = clp->lc_prevsess;
783 			sess_replay = true;
784 		    } else
785 			error = NFSERR_SEQMISORDERED;
786 		} else if ((nd->nd_flag & ND_NFSV41) == 0 &&
787 		     clp->lc_confirm.qval != confirm.qval)
788 			error = NFSERR_STALECLIENTID;
789 		if (error == 0 && nfsrv_notsamecredname(NFSV4OP_CREATESESSION,
790 		    nd, clp))
791 			error = NFSERR_CLIDINUSE;
792 
793 		if (!error) {
794 		    if ((clp->lc_flags & (LCL_NEEDSCONFIRM | LCL_DONTCLEAN)) ==
795 			LCL_NEEDSCONFIRM) {
796 			/*
797 			 * Hang onto the delegations (as old delegations)
798 			 * for an Open with CLAIM_DELEGATE_PREV unless in
799 			 * grace, but get rid of the rest of the state.
800 			 */
801 			if (mlocked)
802 				nfsrv_cleanclient(clp, p, true, &old_xprt);
803 			else
804 				nfsrv_cleanclient(clp, p, false, NULL);
805 			nfsrv_freedeleglist(&clp->lc_olddeleg);
806 			if (nfsrv_checkgrace(nd, clp, 0)) {
807 			    /* In grace, so just delete delegations */
808 			    nfsrv_freedeleglist(&clp->lc_deleg);
809 			} else {
810 			    LIST_FOREACH(stp, &clp->lc_deleg, ls_list)
811 				stp->ls_flags |= NFSLCK_OLDDELEG;
812 			    clp->lc_delegtime = NFSD_MONOSEC +
813 				nfsrv_lease + NFSRV_LEASEDELTA;
814 			    LIST_NEWHEAD(&clp->lc_olddeleg, &clp->lc_deleg,
815 				ls_list);
816 			}
817 			if ((nd->nd_flag & ND_NFSV41) != 0)
818 			    clp->lc_program = cbprogram;
819 		    }
820 		    clp->lc_flags &= ~(LCL_NEEDSCONFIRM | LCL_DONTCLEAN);
821 		    if (clp->lc_program)
822 			clp->lc_flags |= LCL_NEEDSCBNULL;
823 		    /* For NFSv4.1, link the session onto the client. */
824 		    if (nsep != NULL) {
825 			/* Hold a reference on the xprt for a backchannel. */
826 			if ((nsep->sess_crflags & NFSV4CRSESS_CONNBACKCHAN)
827 			    != 0 && !sess_replay) {
828 			    if (clp->lc_req.nr_client == NULL) {
829 				clp->lc_req.nr_client = client;
830 				client = NULL;
831 			    }
832 			    if (clp->lc_req.nr_client != NULL) {
833 				SVC_ACQUIRE(nd->nd_xprt);
834 				CLNT_ACQUIRE(clp->lc_req.nr_client);
835 				nd->nd_xprt->xp_p2 = clp->lc_req.nr_client;
836 				/* Disable idle timeout. */
837 				nd->nd_xprt->xp_idletimeout = 0;
838 				nsep->sess_cbsess.nfsess_xprt = nd->nd_xprt;
839 			    } else
840 				nsep->sess_crflags &= ~NFSV4CRSESS_CONNBACKCHAN;
841 			}
842 			NFSBCOPY(sessid, nsep->sess_sessionid,
843 			    NFSX_V4SESSIONID);
844 			NFSBCOPY(sessid, nsep->sess_cbsess.nfsess_sessionid,
845 			    NFSX_V4SESSIONID);
846 			if (!sess_replay) {
847 			    shp = NFSSESSIONHASH(nsep->sess_sessionid);
848 			    if (!mlocked)
849 				NFSLOCKSTATE();
850 			    NFSLOCKSESSION(shp);
851 			    LIST_INSERT_HEAD(&shp->list, nsep, sess_hash);
852 			    LIST_INSERT_HEAD(&clp->lc_session, nsep, sess_list);
853 			    nsep->sess_clp = clp;
854 			    NFSUNLOCKSESSION(shp);
855 			    if (!mlocked)
856 				NFSUNLOCKSTATE();
857 			}
858 		    }
859 		}
860 	} else if (clp->lc_flags & LCL_NEEDSCONFIRM) {
861 		error = NFSERR_EXPIRED;
862 	}
863 
864 	/*
865 	 * If called by the Renew Op, we must check the principal.
866 	 */
867 	if (!error && (opflags & CLOPS_RENEWOP)) {
868 	    if (nfsrv_notsamecredname(0, nd, clp)) {
869 		doneok = 0;
870 		for (i = 0; i < nfsrv_statehashsize && doneok == 0; i++) {
871 		    LIST_FOREACH(stp, &clp->lc_stateid[i], ls_hash) {
872 			if ((stp->ls_flags & NFSLCK_OPEN) &&
873 			    stp->ls_uid == nd->nd_cred->cr_uid) {
874 				doneok = 1;
875 				break;
876 			}
877 		    }
878 		}
879 		if (!doneok)
880 			error = NFSERR_ACCES;
881 	    }
882 	    if (!error && (clp->lc_flags & LCL_CBDOWN))
883 		error = NFSERR_CBPATHDOWN;
884 	}
885 	if ((!error || error == NFSERR_CBPATHDOWN) &&
886 	     (opflags & CLOPS_RENEW)) {
887 		clp->lc_expiry = nfsrv_leaseexpiry();
888 	}
889 	if (opflags & CLOPS_CONFIRM) {
890 		nfsrv_clientunlock(mlocked);
891 		if (client != NULL)
892 			CLNT_RELEASE(client);
893 		if (old_xprt != NULL)
894 			SVC_RELEASE(old_xprt);
895 	} else if (opflags != CLOPS_RENEW) {
896 		NFSUNLOCKSTATE();
897 	}
898 	if (clpp)
899 		*clpp = clp;
900 
901 out:
902 	NFSEXITCODE2(error, nd);
903 	return (error);
904 }
905 
906 /*
907  * Perform the NFSv4.1 destroy clientid.
908  */
909 int
910 nfsrv_destroyclient(struct nfsrv_descript *nd, nfsquad_t clientid, NFSPROC_T *p)
911 {
912 	struct nfsclient *clp;
913 	struct nfsclienthashhead *hp;
914 	SVCXPRT *old_xprt;
915 	int error = 0, i;
916 	bool mlocked;
917 
918 	if (NFSD_VNET(nfsrvboottime) != clientid.lval[0]) {
919 		error = NFSERR_STALECLIENTID;
920 		goto out;
921 	}
922 
923 	mlocked = true;
924 	if (nfsrv_dolocallocks != 0)
925 		mlocked = false;
926 	/* Lock out other nfsd threads */
927 	nfsrv_clientlock(mlocked);
928 
929 	hp = NFSCLIENTHASH(clientid);
930 	LIST_FOREACH(clp, hp, lc_hash) {
931 		if (clp->lc_clientid.lval[1] == clientid.lval[1])
932 			break;
933 	}
934 	if (clp == NULL) {
935 		nfsrv_clientunlock(mlocked);
936 		/* Just return ok, since it is gone. */
937 		goto out;
938 	}
939 
940 	/* Check for the SP4_MACH_CRED case. */
941 	error = nfsrv_checkmachcred(NFSV4OP_DESTROYCLIENTID, nd, clp);
942 	if (error != 0) {
943 		nfsrv_clientunlock(mlocked);
944 		goto out;
945 	}
946 
947 	/*
948 	 * Free up all layouts on the clientid.  Should the client return the
949 	 * layouts?
950 	 */
951 	nfsrv_freelayoutlist(clientid);
952 
953 	/* Scan for state on the clientid. */
954 	for (i = 0; i < nfsrv_statehashsize; i++)
955 		if (!LIST_EMPTY(&clp->lc_stateid[i])) {
956 			nfsrv_clientunlock(mlocked);
957 			error = NFSERR_CLIENTIDBUSY;
958 			goto out;
959 		}
960 	if (!LIST_EMPTY(&clp->lc_session) || !LIST_EMPTY(&clp->lc_deleg)) {
961 		nfsrv_clientunlock(mlocked);
962 		error = NFSERR_CLIENTIDBUSY;
963 		goto out;
964 	}
965 
966 	/* Destroy the clientid and return ok. */
967 	old_xprt = NULL;
968 	if (mlocked)
969 		nfsrv_cleanclient(clp, p, true, &old_xprt);
970 	else
971 		nfsrv_cleanclient(clp, p, false, NULL);
972 	nfsrv_freedeleglist(&clp->lc_deleg);
973 	nfsrv_freedeleglist(&clp->lc_olddeleg);
974 	LIST_REMOVE(clp, lc_hash);
975 	nfsrv_clientunlock(mlocked);
976 	if (old_xprt != NULL)
977 		SVC_RELEASE(old_xprt);
978 	nfsrv_zapclient(clp, p);
979 out:
980 	NFSEXITCODE2(error, nd);
981 	return (error);
982 }
983 
984 /*
985  * Called from the new nfssvc syscall to admin revoke a clientid.
986  * Returns 0 for success, error otherwise.
987  */
988 int
989 nfsrv_adminrevoke(struct nfsd_clid *revokep, NFSPROC_T *p)
990 {
991 	struct nfsclient *clp = NULL;
992 	int i, error = 0;
993 	int gotit, igotlock;
994 
995 	/*
996 	 * First, lock out the nfsd so that state won't change while the
997 	 * revocation record is being written to the stable storage restart
998 	 * file.
999 	 */
1000 	NFSLOCKV4ROOTMUTEX();
1001 	do {
1002 		igotlock = nfsv4_lock(&nfsv4rootfs_lock, 1, NULL,
1003 		    NFSV4ROOTLOCKMUTEXPTR, NULL);
1004 	} while (!igotlock);
1005 	NFSUNLOCKV4ROOTMUTEX();
1006 
1007 	/*
1008 	 * Search for a match in the client list.
1009 	 */
1010 	gotit = i = 0;
1011 	while (i < nfsrv_clienthashsize && !gotit) {
1012 	    LIST_FOREACH(clp, &NFSD_VNET(nfsclienthash)[i], lc_hash) {
1013 		if (revokep->nclid_idlen == clp->lc_idlen &&
1014 		    !NFSBCMP(revokep->nclid_id, clp->lc_id, clp->lc_idlen)) {
1015 			gotit = 1;
1016 			break;
1017 		}
1018 	    }
1019 	    i++;
1020 	}
1021 	if (!gotit) {
1022 		NFSLOCKV4ROOTMUTEX();
1023 		nfsv4_unlock(&nfsv4rootfs_lock, 0);
1024 		NFSUNLOCKV4ROOTMUTEX();
1025 		error = EPERM;
1026 		goto out;
1027 	}
1028 
1029 	/*
1030 	 * Now, write out the revocation record
1031 	 */
1032 	nfsrv_writestable(clp->lc_id, clp->lc_idlen, NFSNST_REVOKE, p);
1033 	nfsrv_backupstable();
1034 
1035 	/*
1036 	 * and clear out the state, marking the clientid revoked.
1037 	 */
1038 	clp->lc_flags &= ~LCL_CALLBACKSON;
1039 	clp->lc_flags |= LCL_ADMINREVOKED;
1040 	nfsrv_cleanclient(clp, p, false, NULL);
1041 	nfsrv_freedeleglist(&clp->lc_deleg);
1042 	nfsrv_freedeleglist(&clp->lc_olddeleg);
1043 	NFSLOCKV4ROOTMUTEX();
1044 	nfsv4_unlock(&nfsv4rootfs_lock, 0);
1045 	NFSUNLOCKV4ROOTMUTEX();
1046 
1047 out:
1048 	NFSEXITCODE(error);
1049 	return (error);
1050 }
1051 
1052 /*
1053  * Dump out stats for all clients. Called from nfssvc(2), that is used
1054  * nfsstatsv1.
1055  */
1056 void
1057 nfsrv_dumpclients(struct nfsd_dumpclients *dumpp, int maxcnt)
1058 {
1059 	struct nfsclient *clp;
1060 	int i = 0, cnt = 0;
1061 
1062 	/*
1063 	 * First, get a reference on the nfsv4rootfs_lock so that an
1064 	 * exclusive lock cannot be acquired while dumping the clients.
1065 	 */
1066 	NFSLOCKV4ROOTMUTEX();
1067 	nfsv4_getref(&nfsv4rootfs_lock, NULL, NFSV4ROOTLOCKMUTEXPTR, NULL);
1068 	NFSUNLOCKV4ROOTMUTEX();
1069 	NFSLOCKSTATE();
1070 	/*
1071 	 * Rattle through the client lists until done.
1072 	 */
1073 	while (i < nfsrv_clienthashsize && cnt < maxcnt) {
1074 	    clp = LIST_FIRST(&NFSD_VNET(nfsclienthash)[i]);
1075 	    while (clp != LIST_END(&NFSD_VNET(nfsclienthash)[i]) && cnt <
1076 		maxcnt) {
1077 		nfsrv_dumpaclient(clp, &dumpp[cnt]);
1078 		cnt++;
1079 		clp = LIST_NEXT(clp, lc_hash);
1080 	    }
1081 	    i++;
1082 	}
1083 	if (cnt < maxcnt)
1084 	    dumpp[cnt].ndcl_clid.nclid_idlen = 0;
1085 	NFSUNLOCKSTATE();
1086 	NFSLOCKV4ROOTMUTEX();
1087 	nfsv4_relref(&nfsv4rootfs_lock);
1088 	NFSUNLOCKV4ROOTMUTEX();
1089 }
1090 
1091 /*
1092  * Dump stats for a client. Must be called with the NFSSTATELOCK and spl'd.
1093  */
1094 static void
1095 nfsrv_dumpaclient(struct nfsclient *clp, struct nfsd_dumpclients *dumpp)
1096 {
1097 	struct nfsstate *stp, *openstp, *lckownstp;
1098 	struct nfslock *lop;
1099 	sa_family_t af;
1100 #ifdef INET
1101 	struct sockaddr_in *rin;
1102 #endif
1103 #ifdef INET6
1104 	struct sockaddr_in6 *rin6;
1105 #endif
1106 
1107 	dumpp->ndcl_nopenowners = dumpp->ndcl_nlockowners = 0;
1108 	dumpp->ndcl_nopens = dumpp->ndcl_nlocks = 0;
1109 	dumpp->ndcl_ndelegs = dumpp->ndcl_nolddelegs = 0;
1110 	dumpp->ndcl_flags = clp->lc_flags;
1111 	dumpp->ndcl_clid.nclid_idlen = clp->lc_idlen;
1112 	NFSBCOPY(clp->lc_id, dumpp->ndcl_clid.nclid_id, clp->lc_idlen);
1113 	af = clp->lc_req.nr_nam->sa_family;
1114 	dumpp->ndcl_addrfam = af;
1115 	switch (af) {
1116 #ifdef INET
1117 	case AF_INET:
1118 		rin = (struct sockaddr_in *)clp->lc_req.nr_nam;
1119 		dumpp->ndcl_cbaddr.sin_addr = rin->sin_addr;
1120 		break;
1121 #endif
1122 #ifdef INET6
1123 	case AF_INET6:
1124 		rin6 = (struct sockaddr_in6 *)clp->lc_req.nr_nam;
1125 		dumpp->ndcl_cbaddr.sin6_addr = rin6->sin6_addr;
1126 		break;
1127 #endif
1128 	}
1129 
1130 	/*
1131 	 * Now, scan the state lists and total up the opens and locks.
1132 	 */
1133 	LIST_FOREACH(stp, &clp->lc_open, ls_list) {
1134 	    dumpp->ndcl_nopenowners++;
1135 	    LIST_FOREACH(openstp, &stp->ls_open, ls_list) {
1136 		dumpp->ndcl_nopens++;
1137 		LIST_FOREACH(lckownstp, &openstp->ls_open, ls_list) {
1138 		    dumpp->ndcl_nlockowners++;
1139 		    LIST_FOREACH(lop, &lckownstp->ls_lock, lo_lckowner) {
1140 			dumpp->ndcl_nlocks++;
1141 		    }
1142 		}
1143 	    }
1144 	}
1145 
1146 	/*
1147 	 * and the delegation lists.
1148 	 */
1149 	LIST_FOREACH(stp, &clp->lc_deleg, ls_list) {
1150 	    dumpp->ndcl_ndelegs++;
1151 	}
1152 	LIST_FOREACH(stp, &clp->lc_olddeleg, ls_list) {
1153 	    dumpp->ndcl_nolddelegs++;
1154 	}
1155 }
1156 
1157 /*
1158  * Dump out lock stats for a file.
1159  */
1160 void
1161 nfsrv_dumplocks(vnode_t vp, struct nfsd_dumplocks *ldumpp, int maxcnt,
1162     NFSPROC_T *p)
1163 {
1164 	struct nfsstate *stp;
1165 	struct nfslock *lop;
1166 	int cnt = 0;
1167 	struct nfslockfile *lfp;
1168 	sa_family_t af;
1169 #ifdef INET
1170 	struct sockaddr_in *rin;
1171 #endif
1172 #ifdef INET6
1173 	struct sockaddr_in6 *rin6;
1174 #endif
1175 	int ret;
1176 	fhandle_t nfh;
1177 
1178 	ret = nfsrv_getlockfh(vp, 0, NULL, &nfh, p);
1179 	/*
1180 	 * First, get a reference on the nfsv4rootfs_lock so that an
1181 	 * exclusive lock on it cannot be acquired while dumping the locks.
1182 	 */
1183 	NFSLOCKV4ROOTMUTEX();
1184 	nfsv4_getref(&nfsv4rootfs_lock, NULL, NFSV4ROOTLOCKMUTEXPTR, NULL);
1185 	NFSUNLOCKV4ROOTMUTEX();
1186 	NFSLOCKSTATE();
1187 	if (!ret)
1188 		ret = nfsrv_getlockfile(0, NULL, &lfp, &nfh, 0);
1189 	if (ret) {
1190 		ldumpp[0].ndlck_clid.nclid_idlen = 0;
1191 		NFSUNLOCKSTATE();
1192 		NFSLOCKV4ROOTMUTEX();
1193 		nfsv4_relref(&nfsv4rootfs_lock);
1194 		NFSUNLOCKV4ROOTMUTEX();
1195 		return;
1196 	}
1197 
1198 	/*
1199 	 * For each open share on file, dump it out.
1200 	 */
1201 	stp = LIST_FIRST(&lfp->lf_open);
1202 	while (stp != LIST_END(&lfp->lf_open) && cnt < maxcnt) {
1203 		ldumpp[cnt].ndlck_flags = stp->ls_flags;
1204 		ldumpp[cnt].ndlck_stateid.seqid = stp->ls_stateid.seqid;
1205 		ldumpp[cnt].ndlck_stateid.other[0] = stp->ls_stateid.other[0];
1206 		ldumpp[cnt].ndlck_stateid.other[1] = stp->ls_stateid.other[1];
1207 		ldumpp[cnt].ndlck_stateid.other[2] = stp->ls_stateid.other[2];
1208 		ldumpp[cnt].ndlck_owner.nclid_idlen =
1209 		    stp->ls_openowner->ls_ownerlen;
1210 		NFSBCOPY(stp->ls_openowner->ls_owner,
1211 		    ldumpp[cnt].ndlck_owner.nclid_id,
1212 		    stp->ls_openowner->ls_ownerlen);
1213 		ldumpp[cnt].ndlck_clid.nclid_idlen = stp->ls_clp->lc_idlen;
1214 		NFSBCOPY(stp->ls_clp->lc_id, ldumpp[cnt].ndlck_clid.nclid_id,
1215 		    stp->ls_clp->lc_idlen);
1216 		af = stp->ls_clp->lc_req.nr_nam->sa_family;
1217 		ldumpp[cnt].ndlck_addrfam = af;
1218 		switch (af) {
1219 #ifdef INET
1220 		case AF_INET:
1221 			rin = (struct sockaddr_in *)stp->ls_clp->lc_req.nr_nam;
1222 			ldumpp[cnt].ndlck_cbaddr.sin_addr = rin->sin_addr;
1223 			break;
1224 #endif
1225 #ifdef INET6
1226 		case AF_INET6:
1227 			rin6 = (struct sockaddr_in6 *)
1228 			    stp->ls_clp->lc_req.nr_nam;
1229 			ldumpp[cnt].ndlck_cbaddr.sin6_addr = rin6->sin6_addr;
1230 			break;
1231 #endif
1232 		}
1233 		stp = LIST_NEXT(stp, ls_file);
1234 		cnt++;
1235 	}
1236 
1237 	/*
1238 	 * and all locks.
1239 	 */
1240 	lop = LIST_FIRST(&lfp->lf_lock);
1241 	while (lop != LIST_END(&lfp->lf_lock) && cnt < maxcnt) {
1242 		stp = lop->lo_stp;
1243 		ldumpp[cnt].ndlck_flags = lop->lo_flags;
1244 		ldumpp[cnt].ndlck_first = lop->lo_first;
1245 		ldumpp[cnt].ndlck_end = lop->lo_end;
1246 		ldumpp[cnt].ndlck_stateid.seqid = stp->ls_stateid.seqid;
1247 		ldumpp[cnt].ndlck_stateid.other[0] = stp->ls_stateid.other[0];
1248 		ldumpp[cnt].ndlck_stateid.other[1] = stp->ls_stateid.other[1];
1249 		ldumpp[cnt].ndlck_stateid.other[2] = stp->ls_stateid.other[2];
1250 		ldumpp[cnt].ndlck_owner.nclid_idlen = stp->ls_ownerlen;
1251 		NFSBCOPY(stp->ls_owner, ldumpp[cnt].ndlck_owner.nclid_id,
1252 		    stp->ls_ownerlen);
1253 		ldumpp[cnt].ndlck_clid.nclid_idlen = stp->ls_clp->lc_idlen;
1254 		NFSBCOPY(stp->ls_clp->lc_id, ldumpp[cnt].ndlck_clid.nclid_id,
1255 		    stp->ls_clp->lc_idlen);
1256 		af = stp->ls_clp->lc_req.nr_nam->sa_family;
1257 		ldumpp[cnt].ndlck_addrfam = af;
1258 		switch (af) {
1259 #ifdef INET
1260 		case AF_INET:
1261 			rin = (struct sockaddr_in *)stp->ls_clp->lc_req.nr_nam;
1262 			ldumpp[cnt].ndlck_cbaddr.sin_addr = rin->sin_addr;
1263 			break;
1264 #endif
1265 #ifdef INET6
1266 		case AF_INET6:
1267 			rin6 = (struct sockaddr_in6 *)
1268 			    stp->ls_clp->lc_req.nr_nam;
1269 			ldumpp[cnt].ndlck_cbaddr.sin6_addr = rin6->sin6_addr;
1270 			break;
1271 #endif
1272 		}
1273 		lop = LIST_NEXT(lop, lo_lckfile);
1274 		cnt++;
1275 	}
1276 
1277 	/*
1278 	 * and the delegations.
1279 	 */
1280 	stp = LIST_FIRST(&lfp->lf_deleg);
1281 	while (stp != LIST_END(&lfp->lf_deleg) && cnt < maxcnt) {
1282 		ldumpp[cnt].ndlck_flags = stp->ls_flags;
1283 		ldumpp[cnt].ndlck_stateid.seqid = stp->ls_stateid.seqid;
1284 		ldumpp[cnt].ndlck_stateid.other[0] = stp->ls_stateid.other[0];
1285 		ldumpp[cnt].ndlck_stateid.other[1] = stp->ls_stateid.other[1];
1286 		ldumpp[cnt].ndlck_stateid.other[2] = stp->ls_stateid.other[2];
1287 		ldumpp[cnt].ndlck_owner.nclid_idlen = 0;
1288 		ldumpp[cnt].ndlck_clid.nclid_idlen = stp->ls_clp->lc_idlen;
1289 		NFSBCOPY(stp->ls_clp->lc_id, ldumpp[cnt].ndlck_clid.nclid_id,
1290 		    stp->ls_clp->lc_idlen);
1291 		af = stp->ls_clp->lc_req.nr_nam->sa_family;
1292 		ldumpp[cnt].ndlck_addrfam = af;
1293 		switch (af) {
1294 #ifdef INET
1295 		case AF_INET:
1296 			rin = (struct sockaddr_in *)stp->ls_clp->lc_req.nr_nam;
1297 			ldumpp[cnt].ndlck_cbaddr.sin_addr = rin->sin_addr;
1298 			break;
1299 #endif
1300 #ifdef INET6
1301 		case AF_INET6:
1302 			rin6 = (struct sockaddr_in6 *)
1303 			    stp->ls_clp->lc_req.nr_nam;
1304 			ldumpp[cnt].ndlck_cbaddr.sin6_addr = rin6->sin6_addr;
1305 			break;
1306 #endif
1307 		}
1308 		stp = LIST_NEXT(stp, ls_file);
1309 		cnt++;
1310 	}
1311 
1312 	/*
1313 	 * If list isn't full, mark end of list by setting the client name
1314 	 * to zero length.
1315 	 */
1316 	if (cnt < maxcnt)
1317 		ldumpp[cnt].ndlck_clid.nclid_idlen = 0;
1318 	NFSUNLOCKSTATE();
1319 	NFSLOCKV4ROOTMUTEX();
1320 	nfsv4_relref(&nfsv4rootfs_lock);
1321 	NFSUNLOCKV4ROOTMUTEX();
1322 }
1323 
1324 /*
1325  * Server timer routine. It can scan any linked list, so long
1326  * as it holds the spin/mutex lock and there is no exclusive lock on
1327  * nfsv4rootfs_lock.
1328  * (For OpenBSD, a kthread is ok. For FreeBSD, I think it is ok
1329  *  to do this from a callout, since the spin locks work. For
1330  *  Darwin, I'm not sure what will work correctly yet.)
1331  * Should be called once per second.
1332  */
1333 void
1334 nfsrv_servertimer(void *arg __unused)
1335 {
1336 	struct nfsclient *clp, *nclp;
1337 	struct nfsstate *stp, *nstp;
1338 	int got_ref, i;
1339 
1340 	/*
1341 	 * Make sure nfsboottime is set. This is used by V3 as well
1342 	 * as V4. Note that nfsboottime is not nfsrvboottime, which is
1343 	 * only used by the V4 server for leases.
1344 	 */
1345 	if (nfsboottime.tv_sec == 0)
1346 		NFSSETBOOTTIME(nfsboottime);
1347 
1348 	/*
1349 	 * If server hasn't started yet, just return.
1350 	 */
1351 	NFSLOCKSTATE();
1352 	if (NFSD_VNET(nfsrv_stablefirst).nsf_eograce == 0) {
1353 		NFSUNLOCKSTATE();
1354 		return;
1355 	}
1356 	if (!(NFSD_VNET(nfsrv_stablefirst).nsf_flags & NFSNSF_UPDATEDONE)) {
1357 		if (!(NFSD_VNET(nfsrv_stablefirst).nsf_flags &
1358 		      NFSNSF_GRACEOVER) &&
1359 		    NFSD_MONOSEC > NFSD_VNET(nfsrv_stablefirst).nsf_eograce)
1360 			NFSD_VNET(nfsrv_stablefirst).nsf_flags |=
1361 			    (NFSNSF_GRACEOVER | NFSNSF_NEEDLOCK);
1362 		NFSUNLOCKSTATE();
1363 		return;
1364 	}
1365 
1366 	/*
1367 	 * Try and get a reference count on the nfsv4rootfs_lock so that
1368 	 * no nfsd thread can acquire an exclusive lock on it before this
1369 	 * call is done. If it is already exclusively locked, just return.
1370 	 */
1371 	NFSLOCKV4ROOTMUTEX();
1372 	got_ref = nfsv4_getref_nonblock(&nfsv4rootfs_lock);
1373 	NFSUNLOCKV4ROOTMUTEX();
1374 	if (got_ref == 0) {
1375 		NFSUNLOCKSTATE();
1376 		return;
1377 	}
1378 
1379 	/*
1380 	 * For each client...
1381 	 */
1382 	for (i = 0; i < nfsrv_clienthashsize; i++) {
1383 	    clp = LIST_FIRST(&NFSD_VNET(nfsclienthash)[i]);
1384 	    while (clp != LIST_END(&NFSD_VNET(nfsclienthash)[i])) {
1385 		nclp = LIST_NEXT(clp, lc_hash);
1386 		if (!(clp->lc_flags & LCL_EXPIREIT)) {
1387 		    if (((clp->lc_expiry + NFSRV_STALELEASE) < NFSD_MONOSEC
1388 			 && ((LIST_EMPTY(&clp->lc_deleg)
1389 			      && LIST_EMPTY(&clp->lc_open)) ||
1390 			     nfsrv_clients > nfsrv_clienthighwater)) ||
1391 			(clp->lc_expiry + NFSRV_MOULDYLEASE) < NFSD_MONOSEC ||
1392 			(clp->lc_expiry < NFSD_MONOSEC &&
1393 			 (nfsrv_openpluslock * 10 / 9) > nfsrv_v4statelimit)) {
1394 			/*
1395 			 * Lease has expired several nfsrv_lease times ago:
1396 			 * PLUS
1397 			 *    - no state is associated with it
1398 			 *    OR
1399 			 *    - above high water mark for number of clients
1400 			 *      (nfsrv_clienthighwater should be large enough
1401 			 *       that this only occurs when clients fail to
1402 			 *       use the same nfs_client_id4.id. Maybe somewhat
1403 			 *       higher that the maximum number of clients that
1404 			 *       will mount this server?)
1405 			 * OR
1406 			 * Lease has expired a very long time ago
1407 			 * OR
1408 			 * Lease has expired PLUS the number of opens + locks
1409 			 * has exceeded 90% of capacity
1410 			 *
1411 			 * --> Mark for expiry. The actual expiry will be done
1412 			 *     by an nfsd sometime soon.
1413 			 */
1414 			clp->lc_flags |= LCL_EXPIREIT;
1415 			NFSD_VNET(nfsrv_stablefirst).nsf_flags |=
1416 			    (NFSNSF_NEEDLOCK | NFSNSF_EXPIREDCLIENT);
1417 		    } else {
1418 			/*
1419 			 * If there are no opens, increment no open tick cnt
1420 			 * If time exceeds NFSNOOPEN, mark it to be thrown away
1421 			 * otherwise, if there is an open, reset no open time
1422 			 * Hopefully, this will avoid excessive re-creation
1423 			 * of open owners and subsequent open confirms.
1424 			 */
1425 			stp = LIST_FIRST(&clp->lc_open);
1426 			while (stp != LIST_END(&clp->lc_open)) {
1427 				nstp = LIST_NEXT(stp, ls_list);
1428 				if (LIST_EMPTY(&stp->ls_open)) {
1429 					stp->ls_noopens++;
1430 					if (stp->ls_noopens > NFSNOOPEN ||
1431 					    (nfsrv_openpluslock * 2) >
1432 					    nfsrv_v4statelimit)
1433 						NFSD_VNET(nfsrv_stablefirst).nsf_flags |=
1434 							NFSNSF_NOOPENS;
1435 				} else {
1436 					stp->ls_noopens = 0;
1437 				}
1438 				stp = nstp;
1439 			}
1440 		    }
1441 		}
1442 		clp = nclp;
1443 	    }
1444 	}
1445 	NFSUNLOCKSTATE();
1446 	NFSLOCKV4ROOTMUTEX();
1447 	nfsv4_relref(&nfsv4rootfs_lock);
1448 	NFSUNLOCKV4ROOTMUTEX();
1449 }
1450 
1451 /*
1452  * The following set of functions free up the various data structures.
1453  */
1454 /*
1455  * Clear out all open/lock state related to this nfsclient.
1456  * Caller must hold an exclusive lock on nfsv4rootfs_lock, so that
1457  * there are no other active nfsd threads.
1458  */
1459 void
1460 nfsrv_cleanclient(struct nfsclient *clp, NFSPROC_T *p, bool locked,
1461     SVCXPRT **old_xprtp)
1462 {
1463 	struct nfsstate *stp, *nstp;
1464 	struct nfsdsession *sep, *nsep;
1465 
1466 	LIST_FOREACH_SAFE(stp, &clp->lc_open, ls_list, nstp) {
1467 		if (locked)
1468 			nfsrv_freeopenowner(stp, 0, p);
1469 		else
1470 			nfsrv_freeopenowner(stp, 1, p);
1471 	}
1472 	if ((clp->lc_flags & LCL_ADMINREVOKED) == 0)
1473 		LIST_FOREACH_SAFE(sep, &clp->lc_session, sess_list, nsep)
1474 			(void)nfsrv_freesession(NULL, sep, NULL, locked,
1475 			    old_xprtp);
1476 }
1477 
1478 /*
1479  * Free a client that has been cleaned. It should also already have been
1480  * removed from the lists.
1481  * (Just to be safe w.r.t. newnfs_disconnect(), call this function when
1482  *  softclock interrupts are enabled.)
1483  */
1484 void
1485 nfsrv_zapclient(struct nfsclient *clp, NFSPROC_T *p)
1486 {
1487 
1488 #ifdef notyet
1489 	if ((clp->lc_flags & (LCL_GSS | LCL_CALLBACKSON)) ==
1490 	     (LCL_GSS | LCL_CALLBACKSON) &&
1491 	    (clp->lc_hand.nfsh_flag & NFSG_COMPLETE) &&
1492 	    clp->lc_handlelen > 0) {
1493 		clp->lc_hand.nfsh_flag &= ~NFSG_COMPLETE;
1494 		clp->lc_hand.nfsh_flag |= NFSG_DESTROYED;
1495 		(void) nfsrv_docallback(clp, NFSV4PROC_CBNULL,
1496 			NULL, 0, NULL, NULL, NULL, 0, p);
1497 	}
1498 #endif
1499 	newnfs_disconnect(NULL, &clp->lc_req);
1500 	free(clp->lc_req.nr_nam, M_SONAME);
1501 	NFSFREEMUTEX(&clp->lc_req.nr_mtx);
1502 	free(clp->lc_stateid, M_NFSDCLIENT);
1503 	free(clp, M_NFSDCLIENT);
1504 	NFSLOCKSTATE();
1505 	NFSD_VNET(nfsstatsv1_p)->srvclients--;
1506 	nfsrv_openpluslock--;
1507 	nfsrv_clients--;
1508 	NFSUNLOCKSTATE();
1509 }
1510 
1511 /*
1512  * Free a list of delegation state structures.
1513  * (This function will also free all nfslockfile structures that no
1514  *  longer have associated state.)
1515  */
1516 void
1517 nfsrv_freedeleglist(struct nfsstatehead *sthp)
1518 {
1519 	struct nfsstate *stp, *nstp;
1520 
1521 	LIST_FOREACH_SAFE(stp, sthp, ls_list, nstp) {
1522 		nfsrv_freedeleg(stp);
1523 	}
1524 	LIST_INIT(sthp);
1525 }
1526 
1527 /*
1528  * Free up a delegation.
1529  */
1530 static void
1531 nfsrv_freedeleg(struct nfsstate *stp)
1532 {
1533 	struct nfslockfile *lfp;
1534 
1535 	LIST_REMOVE(stp, ls_hash);
1536 	LIST_REMOVE(stp, ls_list);
1537 	LIST_REMOVE(stp, ls_file);
1538 	if ((stp->ls_flags & NFSLCK_DELEGWRITE) != 0)
1539 		nfsrv_writedelegcnt--;
1540 	lfp = stp->ls_lfp;
1541 	if (LIST_EMPTY(&lfp->lf_open) &&
1542 	    LIST_EMPTY(&lfp->lf_lock) && LIST_EMPTY(&lfp->lf_deleg) &&
1543 	    LIST_EMPTY(&lfp->lf_locallock) && LIST_EMPTY(&lfp->lf_rollback) &&
1544 	    lfp->lf_usecount == 0 &&
1545 	    nfsv4_testlock(&lfp->lf_locallock_lck) == 0)
1546 		nfsrv_freenfslockfile(lfp);
1547 	free(stp, M_NFSDSTATE);
1548 	NFSD_VNET(nfsstatsv1_p)->srvdelegates--;
1549 	nfsrv_openpluslock--;
1550 	nfsrv_delegatecnt--;
1551 }
1552 
1553 /*
1554  * This function frees an open owner and all associated opens.
1555  */
1556 static void
1557 nfsrv_freeopenowner(struct nfsstate *stp, int cansleep, NFSPROC_T *p)
1558 {
1559 	struct nfsstate *nstp, *tstp;
1560 
1561 	LIST_REMOVE(stp, ls_list);
1562 	/*
1563 	 * Now, free all associated opens.
1564 	 */
1565 	nstp = LIST_FIRST(&stp->ls_open);
1566 	while (nstp != LIST_END(&stp->ls_open)) {
1567 		tstp = nstp;
1568 		nstp = LIST_NEXT(nstp, ls_list);
1569 		nfsrv_freeopen(tstp, NULL, cansleep, p);
1570 	}
1571 	if (stp->ls_op)
1572 		nfsrvd_derefcache(stp->ls_op);
1573 	free(stp, M_NFSDSTATE);
1574 	NFSD_VNET(nfsstatsv1_p)->srvopenowners--;
1575 	nfsrv_openpluslock--;
1576 }
1577 
1578 /*
1579  * This function frees an open (nfsstate open structure) with all associated
1580  * lock_owners and locks. It also frees the nfslockfile structure iff there
1581  * are no other opens on the file.
1582  * Returns 1 if it free'd the nfslockfile, 0 otherwise.
1583  */
1584 static void
1585 nfsrv_freeopen(struct nfsstate *stp, vnode_t vp, int cansleep, NFSPROC_T *p)
1586 {
1587 	struct nfsstate *nstp, *tstp;
1588 	struct nfslockfile *lfp;
1589 
1590 	LIST_REMOVE(stp, ls_hash);
1591 	LIST_REMOVE(stp, ls_list);
1592 	LIST_REMOVE(stp, ls_file);
1593 
1594 	lfp = stp->ls_lfp;
1595 	/*
1596 	 * Now, free all lockowners associated with this open.
1597 	 * Note that, if vp != NULL, nfsrv_freelockowner() will
1598 	 * not call nfsrv_freeallnfslocks(), so it needs to be called, below.
1599 	 */
1600 	LIST_FOREACH_SAFE(tstp, &stp->ls_open, ls_list, nstp)
1601 		nfsrv_freelockowner(tstp, vp, cansleep, p);
1602 
1603 	if (vp != NULL) {
1604 		KASSERT(cansleep != 0, ("nfsrv_freeopen: cansleep == 0"));
1605 		mtx_assert(NFSSTATEMUTEXPTR, MA_OWNED);
1606 		/*
1607 		 * Only called with vp != NULL for Close when
1608 		 * vfs.nfsd.enable_locallocks != 0.
1609 		 * Lock the lfp so that it will not go away and do the
1610 		 * nfsrv_freeallnfslocks() call that was not done by
1611 		 * nfsrv_freelockowner().
1612 		 */
1613 		nfsrv_locklf(lfp);
1614 		NFSUNLOCKSTATE();
1615 		NFSVOPUNLOCK(vp);
1616 		nfsrv_freeallnfslocks(stp, vp, cansleep, p);
1617 		NFSVOPLOCK(vp, LK_EXCLUSIVE | LK_RETRY);
1618 		NFSLOCKSTATE();
1619 		nfsrv_unlocklf(lfp);
1620 	}
1621 
1622 	/*
1623 	 * The nfslockfile is freed here if there are no locks
1624 	 * associated with the open.
1625 	 * If there are locks associated with the open, the
1626 	 * nfslockfile structure can be freed via nfsrv_freelockowner().
1627 	 */
1628 	if (lfp != NULL && LIST_EMPTY(&lfp->lf_open) &&
1629 	    LIST_EMPTY(&lfp->lf_deleg) && LIST_EMPTY(&lfp->lf_lock) &&
1630 	    LIST_EMPTY(&lfp->lf_locallock) && LIST_EMPTY(&lfp->lf_rollback) &&
1631 	    lfp->lf_usecount == 0 &&
1632 	    nfsv4_testlock(&lfp->lf_locallock_lck) == 0)
1633 		nfsrv_freenfslockfile(lfp);
1634 	free(stp, M_NFSDSTATE);
1635 	NFSD_VNET(nfsstatsv1_p)->srvopens--;
1636 	nfsrv_openpluslock--;
1637 }
1638 
1639 /*
1640  * Frees a lockowner and all associated locks.
1641  */
1642 static void
1643 nfsrv_freelockowner(struct nfsstate *stp, vnode_t vp, int cansleep,
1644     NFSPROC_T *p)
1645 {
1646 
1647 	LIST_REMOVE(stp, ls_hash);
1648 	LIST_REMOVE(stp, ls_list);
1649 	if (vp == NULL)
1650 		nfsrv_freeallnfslocks(stp, vp, cansleep, p);
1651 	if (stp->ls_op)
1652 		nfsrvd_derefcache(stp->ls_op);
1653 	free(stp, M_NFSDSTATE);
1654 	NFSD_VNET(nfsstatsv1_p)->srvlockowners--;
1655 	nfsrv_openpluslock--;
1656 }
1657 
1658 /*
1659  * Free all the nfs locks on a lockowner.
1660  */
1661 static void
1662 nfsrv_freeallnfslocks(struct nfsstate *stp, vnode_t vp, int cansleep,
1663     NFSPROC_T *p)
1664 {
1665 	struct nfslock *lop, *nlop;
1666 	struct nfsrollback *rlp, *nrlp;
1667 	struct nfslockfile *lfp = NULL;
1668 	int gottvp = 0;
1669 	vnode_t tvp = NULL;
1670 	uint64_t first, end;
1671 
1672 	if (vp != NULL)
1673 		ASSERT_VOP_UNLOCKED(vp, "nfsrv_freeallnfslocks: vnode locked");
1674 	lop = LIST_FIRST(&stp->ls_lock);
1675 	while (lop != LIST_END(&stp->ls_lock)) {
1676 		nlop = LIST_NEXT(lop, lo_lckowner);
1677 		/*
1678 		 * Since all locks should be for the same file, lfp should
1679 		 * not change.
1680 		 */
1681 		if (lfp == NULL)
1682 			lfp = lop->lo_lfp;
1683 		else if (lfp != lop->lo_lfp)
1684 			panic("allnfslocks");
1685 		/*
1686 		 * If vp is NULL and cansleep != 0, a vnode must be acquired
1687 		 * from the file handle. This only occurs when called from
1688 		 * nfsrv_cleanclient().
1689 		 */
1690 		if (gottvp == 0) {
1691 			if (nfsrv_dolocallocks == 0)
1692 				tvp = NULL;
1693 			else if (vp == NULL && cansleep != 0) {
1694 				tvp = nfsvno_getvp(&lfp->lf_fh);
1695 				if (tvp != NULL)
1696 					NFSVOPUNLOCK(tvp);
1697 			} else
1698 				tvp = vp;
1699 			gottvp = 1;
1700 		}
1701 
1702 		if (tvp != NULL) {
1703 			if (cansleep == 0)
1704 				panic("allnfs2");
1705 			first = lop->lo_first;
1706 			end = lop->lo_end;
1707 			nfsrv_freenfslock(lop);
1708 			nfsrv_localunlock(tvp, lfp, first, end, p);
1709 			LIST_FOREACH_SAFE(rlp, &lfp->lf_rollback, rlck_list,
1710 			    nrlp)
1711 				free(rlp, M_NFSDROLLBACK);
1712 			LIST_INIT(&lfp->lf_rollback);
1713 		} else
1714 			nfsrv_freenfslock(lop);
1715 		lop = nlop;
1716 	}
1717 	if (vp == NULL && tvp != NULL)
1718 		vrele(tvp);
1719 }
1720 
1721 /*
1722  * Free an nfslock structure.
1723  */
1724 static void
1725 nfsrv_freenfslock(struct nfslock *lop)
1726 {
1727 
1728 	if (lop->lo_lckfile.le_prev != NULL) {
1729 		LIST_REMOVE(lop, lo_lckfile);
1730 		NFSD_VNET(nfsstatsv1_p)->srvlocks--;
1731 		nfsrv_openpluslock--;
1732 	}
1733 	LIST_REMOVE(lop, lo_lckowner);
1734 	free(lop, M_NFSDLOCK);
1735 }
1736 
1737 /*
1738  * This function frees an nfslockfile structure.
1739  */
1740 static void
1741 nfsrv_freenfslockfile(struct nfslockfile *lfp)
1742 {
1743 
1744 	LIST_REMOVE(lfp, lf_hash);
1745 	free(lfp, M_NFSDLOCKFILE);
1746 }
1747 
1748 /*
1749  * This function looks up an nfsstate structure via stateid.
1750  */
1751 static int
1752 nfsrv_getstate(struct nfsclient *clp, nfsv4stateid_t *stateidp, __unused u_int32_t flags,
1753     struct nfsstate **stpp)
1754 {
1755 	struct nfsstate *stp;
1756 	struct nfsstatehead *hp;
1757 	int error = 0;
1758 
1759 	*stpp = NULL;
1760 	hp = NFSSTATEHASH(clp, *stateidp);
1761 	LIST_FOREACH(stp, hp, ls_hash) {
1762 		if (!NFSBCMP(stp->ls_stateid.other, stateidp->other,
1763 			NFSX_STATEIDOTHER))
1764 			break;
1765 	}
1766 
1767 	/*
1768 	 * If no state id in list, return NFSERR_BADSTATEID.
1769 	 */
1770 	if (stp == LIST_END(hp)) {
1771 		error = NFSERR_BADSTATEID;
1772 		goto out;
1773 	}
1774 	*stpp = stp;
1775 
1776 out:
1777 	NFSEXITCODE(error);
1778 	return (error);
1779 }
1780 
1781 /*
1782  * This function gets an nfsstate structure via owner string.
1783  */
1784 static void
1785 nfsrv_getowner(struct nfsstatehead *hp, struct nfsstate *new_stp,
1786     struct nfsstate **stpp)
1787 {
1788 	struct nfsstate *stp;
1789 
1790 	*stpp = NULL;
1791 	LIST_FOREACH(stp, hp, ls_list) {
1792 		if (new_stp->ls_ownerlen == stp->ls_ownerlen &&
1793 		  !NFSBCMP(new_stp->ls_owner,stp->ls_owner,stp->ls_ownerlen)) {
1794 			*stpp = stp;
1795 			return;
1796 		}
1797 	}
1798 }
1799 
1800 /*
1801  * Lock control function called to update lock status.
1802  * Returns 0 upon success, -1 if there is no lock and the flags indicate
1803  * that one isn't to be created and an NFSERR_xxx for other errors.
1804  * The structures new_stp and new_lop are passed in as pointers that should
1805  * be set to NULL if the structure is used and shouldn't be free'd.
1806  * For the NFSLCK_TEST and NFSLCK_CHECK cases, the structures are
1807  * never used and can safely be allocated on the stack. For all other
1808  * cases, *new_stpp and *new_lopp should be malloc'd before the call,
1809  * in case they are used.
1810  */
1811 int
1812 nfsrv_lockctrl(vnode_t vp, struct nfsstate **new_stpp,
1813     struct nfslock **new_lopp, struct nfslockconflict *cfp,
1814     nfsquad_t clientid, nfsv4stateid_t *stateidp,
1815     __unused struct nfsexstuff *exp,
1816     struct nfsrv_descript *nd, NFSPROC_T *p)
1817 {
1818 	struct nfslock *lop;
1819 	struct nfsstate *new_stp = *new_stpp;
1820 	struct nfslock *new_lop = *new_lopp;
1821 	struct nfsstate *tstp, *mystp, *nstp;
1822 	int specialid = 0;
1823 	struct nfslockfile *lfp;
1824 	struct nfslock *other_lop = NULL;
1825 	struct nfsstate *stp, *lckstp = NULL;
1826 	struct nfsclient *clp = NULL;
1827 	u_int32_t bits;
1828 	int error = 0, haslock = 0, ret, reterr;
1829 	int getlckret, delegation = 0, filestruct_locked, vnode_unlocked = 0;
1830 	fhandle_t nfh;
1831 	uint64_t first, end;
1832 	uint32_t lock_flags;
1833 
1834 	if (new_stp->ls_flags & (NFSLCK_CHECK | NFSLCK_SETATTR)) {
1835 		/*
1836 		 * Note the special cases of "all 1s" or "all 0s" stateids and
1837 		 * let reads with all 1s go ahead.
1838 		 */
1839 		if (new_stp->ls_stateid.seqid == 0x0 &&
1840 		    new_stp->ls_stateid.other[0] == 0x0 &&
1841 		    new_stp->ls_stateid.other[1] == 0x0 &&
1842 		    new_stp->ls_stateid.other[2] == 0x0)
1843 			specialid = 1;
1844 		else if (new_stp->ls_stateid.seqid == 0xffffffff &&
1845 		    new_stp->ls_stateid.other[0] == 0xffffffff &&
1846 		    new_stp->ls_stateid.other[1] == 0xffffffff &&
1847 		    new_stp->ls_stateid.other[2] == 0xffffffff)
1848 			specialid = 2;
1849 	}
1850 
1851 	/*
1852 	 * Check for restart conditions (client and server).
1853 	 */
1854 	error = nfsrv_checkrestart(clientid, new_stp->ls_flags,
1855 	    &new_stp->ls_stateid, specialid);
1856 	if (error)
1857 		goto out;
1858 
1859 	/*
1860 	 * Check for state resource limit exceeded.
1861 	 */
1862 	if ((new_stp->ls_flags & NFSLCK_LOCK) &&
1863 	    nfsrv_openpluslock > nfsrv_v4statelimit) {
1864 		error = NFSERR_RESOURCE;
1865 		goto out;
1866 	}
1867 
1868 	/*
1869 	 * For the lock case, get another nfslock structure,
1870 	 * just in case we need it.
1871 	 * Malloc now, before we start sifting through the linked lists,
1872 	 * in case we have to wait for memory.
1873 	 */
1874 tryagain:
1875 	if (new_stp->ls_flags & NFSLCK_LOCK)
1876 		other_lop = malloc(sizeof (struct nfslock),
1877 		    M_NFSDLOCK, M_WAITOK);
1878 	filestruct_locked = 0;
1879 	reterr = 0;
1880 	lfp = NULL;
1881 
1882 	/*
1883 	 * Get the lockfile structure for CFH now, so we can do a sanity
1884 	 * check against the stateid, before incrementing the seqid#, since
1885 	 * we want to return NFSERR_BADSTATEID on failure and the seqid#
1886 	 * shouldn't be incremented for this case.
1887 	 * If nfsrv_getlockfile() returns -1, it means "not found", which
1888 	 * will be handled later.
1889 	 * If we are doing Lock/LockU and local locking is enabled, sleep
1890 	 * lock the nfslockfile structure.
1891 	 */
1892 	getlckret = nfsrv_getlockfh(vp, new_stp->ls_flags, NULL, &nfh, p);
1893 	NFSLOCKSTATE();
1894 	if (getlckret == 0) {
1895 		if ((new_stp->ls_flags & (NFSLCK_LOCK | NFSLCK_UNLOCK)) != 0 &&
1896 		    nfsrv_dolocallocks != 0 && nd->nd_repstat == 0) {
1897 			getlckret = nfsrv_getlockfile(new_stp->ls_flags, NULL,
1898 			    &lfp, &nfh, 1);
1899 			if (getlckret == 0)
1900 				filestruct_locked = 1;
1901 		} else
1902 			getlckret = nfsrv_getlockfile(new_stp->ls_flags, NULL,
1903 			    &lfp, &nfh, 0);
1904 	}
1905 	if (getlckret != 0 && getlckret != -1)
1906 		reterr = getlckret;
1907 
1908 	if (filestruct_locked != 0) {
1909 		LIST_INIT(&lfp->lf_rollback);
1910 		if ((new_stp->ls_flags & NFSLCK_LOCK)) {
1911 			/*
1912 			 * For local locking, do the advisory locking now, so
1913 			 * that any conflict can be detected. A failure later
1914 			 * can be rolled back locally. If an error is returned,
1915 			 * struct nfslockfile has been unlocked and any local
1916 			 * locking rolled back.
1917 			 */
1918 			NFSUNLOCKSTATE();
1919 			if (vnode_unlocked == 0) {
1920 				ASSERT_VOP_ELOCKED(vp, "nfsrv_lockctrl1");
1921 				vnode_unlocked = 1;
1922 				NFSVOPUNLOCK(vp);
1923 			}
1924 			reterr = nfsrv_locallock(vp, lfp,
1925 			    (new_lop->lo_flags & (NFSLCK_READ | NFSLCK_WRITE)),
1926 			    new_lop->lo_first, new_lop->lo_end, cfp, p);
1927 			NFSLOCKSTATE();
1928 		}
1929 	}
1930 
1931 	if (specialid == 0) {
1932 	    if (new_stp->ls_flags & NFSLCK_TEST) {
1933 		/*
1934 		 * RFC 3530 does not list LockT as an op that renews a
1935 		 * lease, but the consensus seems to be that it is ok
1936 		 * for a server to do so.
1937 		 */
1938 		error = nfsrv_getclient(clientid, CLOPS_RENEW, &clp, NULL,
1939 		    (nfsquad_t)((u_quad_t)0), 0, nd, p);
1940 
1941 		/*
1942 		 * Since NFSERR_EXPIRED, NFSERR_ADMINREVOKED are not valid
1943 		 * error returns for LockT, just go ahead and test for a lock,
1944 		 * since there are no locks for this client, but other locks
1945 		 * can conflict. (ie. same client will always be false)
1946 		 */
1947 		if (error == NFSERR_EXPIRED || error == NFSERR_ADMINREVOKED)
1948 		    error = 0;
1949 		lckstp = new_stp;
1950 	    } else {
1951 	      error = nfsrv_getclient(clientid, CLOPS_RENEW, &clp, NULL,
1952 		(nfsquad_t)((u_quad_t)0), 0, nd, p);
1953 	      if (error == 0)
1954 		/*
1955 		 * Look up the stateid
1956 		 */
1957 		error = nfsrv_getstate(clp, &new_stp->ls_stateid,
1958 		  new_stp->ls_flags, &stp);
1959 	      /*
1960 	       * do some sanity checks for an unconfirmed open or a
1961 	       * stateid that refers to the wrong file, for an open stateid
1962 	       */
1963 	      if (error == 0 && (stp->ls_flags & NFSLCK_OPEN) &&
1964 		  ((stp->ls_openowner->ls_flags & NFSLCK_NEEDSCONFIRM) ||
1965 		   (getlckret == 0 && stp->ls_lfp != lfp))){
1966 		      /*
1967 		       * NFSLCK_SETATTR should return OK rather than NFSERR_BADSTATEID
1968 		       * The only exception is using SETATTR with SIZE.
1969 		       * */
1970                     if ((new_stp->ls_flags &
1971                          (NFSLCK_SETATTR | NFSLCK_CHECK)) != NFSLCK_SETATTR)
1972 			     error = NFSERR_BADSTATEID;
1973 	      }
1974 
1975 		if (error == 0 &&
1976 		  (stp->ls_flags & (NFSLCK_DELEGREAD | NFSLCK_DELEGWRITE)) &&
1977 		  getlckret == 0 && stp->ls_lfp != lfp)
1978 			error = NFSERR_BADSTATEID;
1979 
1980 	      /*
1981 	       * If the lockowner stateid doesn't refer to the same file,
1982 	       * I believe that is considered ok, since some clients will
1983 	       * only create a single lockowner and use that for all locks
1984 	       * on all files.
1985 	       * For now, log it as a diagnostic, instead of considering it
1986 	       * a BadStateid.
1987 	       */
1988 	      if (error == 0 && (stp->ls_flags &
1989 		  (NFSLCK_OPEN | NFSLCK_DELEGREAD | NFSLCK_DELEGWRITE)) == 0 &&
1990 		  getlckret == 0 && stp->ls_lfp != lfp) {
1991 #ifdef DIAGNOSTIC
1992 		  printf("Got a lock statid for different file open\n");
1993 #endif
1994 		  /*
1995 		  error = NFSERR_BADSTATEID;
1996 		  */
1997 	      }
1998 
1999 	      if (error == 0) {
2000 		    if (new_stp->ls_flags & NFSLCK_OPENTOLOCK) {
2001 			/*
2002 			 * If haslock set, we've already checked the seqid.
2003 			 */
2004 			if (!haslock) {
2005 			    if (stp->ls_flags & NFSLCK_OPEN)
2006 				error = nfsrv_checkseqid(nd, new_stp->ls_seq,
2007 				    stp->ls_openowner, new_stp->ls_op);
2008 			    else
2009 				error = NFSERR_BADSTATEID;
2010 			}
2011 			if (!error)
2012 			    nfsrv_getowner(&stp->ls_open, new_stp, &lckstp);
2013 			if (lckstp) {
2014 			    /*
2015 			     * For NFSv4.1 and NFSv4.2 allow an
2016 			     * open_to_lock_owner when the lock_owner already
2017 			     * exists.  Just clear NFSLCK_OPENTOLOCK so that
2018 			     * a new lock_owner will not be created.
2019 			     * RFC7530 states that the error for NFSv4.0
2020 			     * is NFS4ERR_BAD_SEQID.
2021 			     */
2022 			    if ((nd->nd_flag & ND_NFSV41) != 0)
2023 				new_stp->ls_flags &= ~NFSLCK_OPENTOLOCK;
2024 			    else
2025 				error = NFSERR_BADSEQID;
2026 			} else
2027 			    lckstp = new_stp;
2028 		    } else if (new_stp->ls_flags&(NFSLCK_LOCK|NFSLCK_UNLOCK)) {
2029 			/*
2030 			 * If haslock set, ditto above.
2031 			 */
2032 			if (!haslock) {
2033 			    if (stp->ls_flags & NFSLCK_OPEN)
2034 				error = NFSERR_BADSTATEID;
2035 			    else
2036 				error = nfsrv_checkseqid(nd, new_stp->ls_seq,
2037 				    stp, new_stp->ls_op);
2038 			}
2039 			lckstp = stp;
2040 		    } else {
2041 			lckstp = stp;
2042 		    }
2043 	      }
2044 	      /*
2045 	       * If the seqid part of the stateid isn't the same, return
2046 	       * NFSERR_OLDSTATEID for cases other than I/O Ops.
2047 	       * For I/O Ops, only return NFSERR_OLDSTATEID if
2048 	       * nfsrv_returnoldstateid is set. (The consensus on the email
2049 	       * list was that most clients would prefer to not receive
2050 	       * NFSERR_OLDSTATEID for I/O Ops, but the RFC suggests that that
2051 	       * is what will happen, so I use the nfsrv_returnoldstateid to
2052 	       * allow for either server configuration.)
2053 	       */
2054 	      if (!error && stp->ls_stateid.seqid!=new_stp->ls_stateid.seqid &&
2055 		  (((nd->nd_flag & ND_NFSV41) == 0 &&
2056 		   (!(new_stp->ls_flags & NFSLCK_CHECK) ||
2057 		    nfsrv_returnoldstateid)) ||
2058 		   ((nd->nd_flag & ND_NFSV41) != 0 &&
2059 		    new_stp->ls_stateid.seqid != 0)))
2060 		    error = NFSERR_OLDSTATEID;
2061 	    }
2062 	}
2063 
2064 	/*
2065 	 * Now we can check for grace.
2066 	 */
2067 	if (!error)
2068 		error = nfsrv_checkgrace(nd, clp, new_stp->ls_flags);
2069 	if ((new_stp->ls_flags & NFSLCK_RECLAIM) && !error &&
2070 		nfsrv_checkstable(clp))
2071 		error = NFSERR_NOGRACE;
2072 	/*
2073 	 * If we successfully Reclaimed state, note that.
2074 	 */
2075 	if ((new_stp->ls_flags & NFSLCK_RECLAIM) && !error)
2076 		nfsrv_markstable(clp);
2077 
2078 	/*
2079 	 * At this point, either error == NFSERR_BADSTATEID or the
2080 	 * seqid# has been updated, so we can return any error.
2081 	 * If error == 0, there may be an error in:
2082 	 *    nd_repstat - Set by the calling function.
2083 	 *    reterr - Set above, if getting the nfslockfile structure
2084 	 *       or acquiring the local lock failed.
2085 	 *    (If both of these are set, nd_repstat should probably be
2086 	 *     returned, since that error was detected before this
2087 	 *     function call.)
2088 	 */
2089 	if (error != 0 || nd->nd_repstat != 0 || reterr != 0) {
2090 		if (error == 0) {
2091 			if (nd->nd_repstat != 0)
2092 				error = nd->nd_repstat;
2093 			else
2094 				error = reterr;
2095 		}
2096 		if (filestruct_locked != 0) {
2097 			/* Roll back local locks. */
2098 			NFSUNLOCKSTATE();
2099 			if (vnode_unlocked == 0) {
2100 				ASSERT_VOP_ELOCKED(vp, "nfsrv_lockctrl2");
2101 				vnode_unlocked = 1;
2102 				NFSVOPUNLOCK(vp);
2103 			}
2104 			nfsrv_locallock_rollback(vp, lfp, p);
2105 			NFSLOCKSTATE();
2106 			nfsrv_unlocklf(lfp);
2107 		}
2108 		NFSUNLOCKSTATE();
2109 		goto out;
2110 	}
2111 
2112 	/*
2113 	 * Check the nfsrv_getlockfile return.
2114 	 * Returned -1 if no structure found.
2115 	 */
2116 	if (getlckret == -1) {
2117 		error = NFSERR_EXPIRED;
2118 		/*
2119 		 * Called from lockt, so no lock is OK.
2120 		 */
2121 		if (new_stp->ls_flags & NFSLCK_TEST) {
2122 			error = 0;
2123 		} else if (new_stp->ls_flags &
2124 		    (NFSLCK_CHECK | NFSLCK_SETATTR)) {
2125 			/*
2126 			 * Called to check for a lock, OK if the stateid is all
2127 			 * 1s or all 0s, but there should be an nfsstate
2128 			 * otherwise.
2129 			 * (ie. If there is no open, I'll assume no share
2130 			 *  deny bits.)
2131 			 */
2132 			if (specialid)
2133 				error = 0;
2134 			else
2135 				error = NFSERR_BADSTATEID;
2136 		}
2137 		NFSUNLOCKSTATE();
2138 		goto out;
2139 	}
2140 
2141 	/*
2142 	 * For NFSLCK_CHECK and NFSLCK_LOCK, test for a share conflict.
2143 	 * For NFSLCK_CHECK, allow a read if write access is granted,
2144 	 * but check for a deny. For NFSLCK_LOCK, require correct access,
2145 	 * which implies a conflicting deny can't exist.
2146 	 */
2147 	if (new_stp->ls_flags & (NFSLCK_CHECK | NFSLCK_LOCK)) {
2148 	    /*
2149 	     * Four kinds of state id:
2150 	     * - specialid (all 0s or all 1s), only for NFSLCK_CHECK
2151 	     * - stateid for an open
2152 	     * - stateid for a delegation
2153 	     * - stateid for a lock owner
2154 	     */
2155 	    if (!specialid) {
2156 		if (stp->ls_flags & (NFSLCK_DELEGREAD | NFSLCK_DELEGWRITE)) {
2157 		    delegation = 1;
2158 		    mystp = stp;
2159 		    nfsrv_delaydelegtimeout(stp);
2160 	        } else if (stp->ls_flags & NFSLCK_OPEN) {
2161 		    mystp = stp;
2162 		} else {
2163 		    mystp = stp->ls_openstp;
2164 		}
2165 		/*
2166 		 * If locking or checking, require correct access
2167 		 * bit set.
2168 		 */
2169 		if (((new_stp->ls_flags & NFSLCK_LOCK) &&
2170 		     !((new_lop->lo_flags >> NFSLCK_LOCKSHIFT) &
2171 		       mystp->ls_flags & NFSLCK_ACCESSBITS)) ||
2172 		    ((new_stp->ls_flags & (NFSLCK_CHECK|NFSLCK_READACCESS)) ==
2173 		      (NFSLCK_CHECK | NFSLCK_READACCESS) &&
2174 		     !(mystp->ls_flags & NFSLCK_READACCESS) &&
2175 		     nfsrv_allowreadforwriteopen == 0) ||
2176 		    ((new_stp->ls_flags & (NFSLCK_CHECK|NFSLCK_WRITEACCESS)) ==
2177 		      (NFSLCK_CHECK | NFSLCK_WRITEACCESS) &&
2178 		     !(mystp->ls_flags & NFSLCK_WRITEACCESS))) {
2179 			if (filestruct_locked != 0) {
2180 				/* Roll back local locks. */
2181 				NFSUNLOCKSTATE();
2182 				if (vnode_unlocked == 0) {
2183 					ASSERT_VOP_ELOCKED(vp,
2184 					    "nfsrv_lockctrl3");
2185 					vnode_unlocked = 1;
2186 					NFSVOPUNLOCK(vp);
2187 				}
2188 				nfsrv_locallock_rollback(vp, lfp, p);
2189 				NFSLOCKSTATE();
2190 				nfsrv_unlocklf(lfp);
2191 			}
2192 			NFSUNLOCKSTATE();
2193 			error = NFSERR_OPENMODE;
2194 			goto out;
2195 		}
2196 	    } else
2197 		mystp = NULL;
2198 	    if ((new_stp->ls_flags & NFSLCK_CHECK) && !delegation) {
2199 		/*
2200 		 * Check for a conflicting deny bit.
2201 		 */
2202 		LIST_FOREACH(tstp, &lfp->lf_open, ls_file) {
2203 		    if (tstp != mystp) {
2204 			bits = tstp->ls_flags;
2205 			bits >>= NFSLCK_SHIFT;
2206 			if (new_stp->ls_flags & bits & NFSLCK_ACCESSBITS) {
2207 			    KASSERT(vnode_unlocked == 0,
2208 				("nfsrv_lockctrl: vnode unlocked1"));
2209 			    ret = nfsrv_clientconflict(tstp->ls_clp, &haslock,
2210 				vp, p);
2211 			    if (ret == 1) {
2212 				/*
2213 				* nfsrv_clientconflict unlocks state
2214 				 * when it returns non-zero.
2215 				 */
2216 				lckstp = NULL;
2217 				goto tryagain;
2218 			    }
2219 			    if (ret == 0)
2220 				NFSUNLOCKSTATE();
2221 			    if (ret == 2)
2222 				error = NFSERR_PERM;
2223 			    else
2224 				error = NFSERR_OPENMODE;
2225 			    goto out;
2226 			}
2227 		    }
2228 		}
2229 
2230 		/* We're outta here */
2231 		NFSUNLOCKSTATE();
2232 		goto out;
2233 	    }
2234 	}
2235 
2236 	/*
2237 	 * For setattr, just get rid of all the Delegations for other clients.
2238 	 */
2239 	if (new_stp->ls_flags & NFSLCK_SETATTR) {
2240 		KASSERT(vnode_unlocked == 0,
2241 		    ("nfsrv_lockctrl: vnode unlocked2"));
2242 		ret = nfsrv_cleandeleg(vp, lfp, clp, &haslock, p);
2243 		if (ret) {
2244 			/*
2245 			 * nfsrv_cleandeleg() unlocks state when it
2246 			 * returns non-zero.
2247 			 */
2248 			if (ret == -1) {
2249 				lckstp = NULL;
2250 				goto tryagain;
2251 			}
2252 			error = ret;
2253 			goto out;
2254 		}
2255 		if (!(new_stp->ls_flags & NFSLCK_CHECK) ||
2256 		    (LIST_EMPTY(&lfp->lf_open) && LIST_EMPTY(&lfp->lf_lock) &&
2257 		     LIST_EMPTY(&lfp->lf_deleg))) {
2258 			NFSUNLOCKSTATE();
2259 			goto out;
2260 		}
2261 	}
2262 
2263 	/*
2264 	 * Check for a conflicting delegation. If one is found, call
2265 	 * nfsrv_delegconflict() to handle it. If the v4root lock hasn't
2266 	 * been set yet, it will get the lock. Otherwise, it will recall
2267 	 * the delegation. Then, we try try again...
2268 	 * I currently believe the conflict algorithm to be:
2269 	 * For Lock Ops (Lock/LockT/LockU)
2270 	 * - there is a conflict iff a different client has a write delegation
2271 	 * For Reading (Read Op)
2272 	 * - there is a conflict iff a different client has a write delegation
2273 	 *   (the specialids are always a different client)
2274 	 * For Writing (Write/Setattr of size)
2275 	 * - there is a conflict if a different client has any delegation
2276 	 * - there is a conflict if the same client has a read delegation
2277 	 *   (I don't understand why this isn't allowed, but that seems to be
2278 	 *    the current consensus?)
2279 	 */
2280 	tstp = LIST_FIRST(&lfp->lf_deleg);
2281 	while (tstp != LIST_END(&lfp->lf_deleg)) {
2282 	    nstp = LIST_NEXT(tstp, ls_file);
2283 	    if ((((new_stp->ls_flags&(NFSLCK_LOCK|NFSLCK_UNLOCK|NFSLCK_TEST))||
2284 		 ((new_stp->ls_flags & NFSLCK_CHECK) &&
2285 		  (new_lop->lo_flags & NFSLCK_READ))) &&
2286 		  clp != tstp->ls_clp &&
2287 		 (tstp->ls_flags & NFSLCK_DELEGWRITE)) ||
2288 		 ((new_stp->ls_flags & NFSLCK_CHECK) &&
2289 		   (new_lop->lo_flags & NFSLCK_WRITE) &&
2290 		  (clp != tstp->ls_clp ||
2291 		   (tstp->ls_flags & NFSLCK_DELEGREAD)))) {
2292 		ret = 0;
2293 		if (filestruct_locked != 0) {
2294 			/* Roll back local locks. */
2295 			NFSUNLOCKSTATE();
2296 			if (vnode_unlocked == 0) {
2297 				ASSERT_VOP_ELOCKED(vp, "nfsrv_lockctrl4");
2298 				NFSVOPUNLOCK(vp);
2299 			}
2300 			nfsrv_locallock_rollback(vp, lfp, p);
2301 			NFSLOCKSTATE();
2302 			nfsrv_unlocklf(lfp);
2303 			NFSUNLOCKSTATE();
2304 			NFSVOPLOCK(vp, LK_EXCLUSIVE | LK_RETRY);
2305 			vnode_unlocked = 0;
2306 			if (VN_IS_DOOMED(vp))
2307 				ret = NFSERR_SERVERFAULT;
2308 			NFSLOCKSTATE();
2309 		}
2310 		if (ret == 0)
2311 			ret = nfsrv_delegconflict(tstp, &haslock, p, vp);
2312 		if (ret) {
2313 		    /*
2314 		     * nfsrv_delegconflict unlocks state when it
2315 		     * returns non-zero, which it always does.
2316 		     */
2317 		    if (other_lop) {
2318 			free(other_lop, M_NFSDLOCK);
2319 			other_lop = NULL;
2320 		    }
2321 		    if (ret == -1) {
2322 			lckstp = NULL;
2323 			goto tryagain;
2324 		    }
2325 		    error = ret;
2326 		    goto out;
2327 		}
2328 		/* Never gets here. */
2329 	    }
2330 	    tstp = nstp;
2331 	}
2332 
2333 	/*
2334 	 * Handle the unlock case by calling nfsrv_updatelock().
2335 	 * (Should I have done some access checking above for unlock? For now,
2336 	 *  just let it happen.)
2337 	 */
2338 	if (new_stp->ls_flags & NFSLCK_UNLOCK) {
2339 		first = new_lop->lo_first;
2340 		end = new_lop->lo_end;
2341 		nfsrv_updatelock(stp, new_lopp, &other_lop, lfp);
2342 		stateidp->seqid = ++(stp->ls_stateid.seqid);
2343 		if ((nd->nd_flag & ND_NFSV41) != 0 && stateidp->seqid == 0)
2344 			stateidp->seqid = stp->ls_stateid.seqid = 1;
2345 		stateidp->other[0] = stp->ls_stateid.other[0];
2346 		stateidp->other[1] = stp->ls_stateid.other[1];
2347 		stateidp->other[2] = stp->ls_stateid.other[2];
2348 		if (filestruct_locked != 0) {
2349 			NFSUNLOCKSTATE();
2350 			if (vnode_unlocked == 0) {
2351 				ASSERT_VOP_ELOCKED(vp, "nfsrv_lockctrl5");
2352 				vnode_unlocked = 1;
2353 				NFSVOPUNLOCK(vp);
2354 			}
2355 			/* Update the local locks. */
2356 			nfsrv_localunlock(vp, lfp, first, end, p);
2357 			NFSLOCKSTATE();
2358 			nfsrv_unlocklf(lfp);
2359 		}
2360 		NFSUNLOCKSTATE();
2361 		goto out;
2362 	}
2363 
2364 	/*
2365 	 * Search for a conflicting lock. A lock conflicts if:
2366 	 * - the lock range overlaps and
2367 	 * - at least one lock is a write lock and
2368 	 * - it is not owned by the same lock owner
2369 	 */
2370 	if (!delegation) {
2371 	  LIST_FOREACH(lop, &lfp->lf_lock, lo_lckfile) {
2372 	    if (new_lop->lo_end > lop->lo_first &&
2373 		new_lop->lo_first < lop->lo_end &&
2374 		(new_lop->lo_flags == NFSLCK_WRITE ||
2375 		 lop->lo_flags == NFSLCK_WRITE) &&
2376 		lckstp != lop->lo_stp &&
2377 		(clp != lop->lo_stp->ls_clp ||
2378 		 lckstp->ls_ownerlen != lop->lo_stp->ls_ownerlen ||
2379 		 NFSBCMP(lckstp->ls_owner, lop->lo_stp->ls_owner,
2380 		    lckstp->ls_ownerlen))) {
2381 		if (other_lop) {
2382 		    free(other_lop, M_NFSDLOCK);
2383 		    other_lop = NULL;
2384 		}
2385 		if (vnode_unlocked != 0)
2386 		    ret = nfsrv_clientconflict(lop->lo_stp->ls_clp, &haslock,
2387 			NULL, p);
2388 		else
2389 		    ret = nfsrv_clientconflict(lop->lo_stp->ls_clp, &haslock,
2390 			vp, p);
2391 		if (ret == 1) {
2392 		    if (filestruct_locked != 0) {
2393 			if (vnode_unlocked == 0) {
2394 				ASSERT_VOP_ELOCKED(vp, "nfsrv_lockctrl6");
2395 				NFSVOPUNLOCK(vp);
2396 			}
2397 			/* Roll back local locks. */
2398 			nfsrv_locallock_rollback(vp, lfp, p);
2399 			NFSLOCKSTATE();
2400 			nfsrv_unlocklf(lfp);
2401 			NFSUNLOCKSTATE();
2402 			NFSVOPLOCK(vp, LK_EXCLUSIVE | LK_RETRY);
2403 			vnode_unlocked = 0;
2404 			if (VN_IS_DOOMED(vp)) {
2405 				error = NFSERR_SERVERFAULT;
2406 				goto out;
2407 			}
2408 		    }
2409 		    /*
2410 		     * nfsrv_clientconflict() unlocks state when it
2411 		     * returns non-zero.
2412 		     */
2413 		    lckstp = NULL;
2414 		    goto tryagain;
2415 		}
2416 		/*
2417 		 * Found a conflicting lock, so record the conflict and
2418 		 * return the error.
2419 		 */
2420 		if (cfp != NULL && ret == 0) {
2421 		    cfp->cl_clientid.lval[0]=lop->lo_stp->ls_stateid.other[0];
2422 		    cfp->cl_clientid.lval[1]=lop->lo_stp->ls_stateid.other[1];
2423 		    cfp->cl_first = lop->lo_first;
2424 		    cfp->cl_end = lop->lo_end;
2425 		    cfp->cl_flags = lop->lo_flags;
2426 		    cfp->cl_ownerlen = lop->lo_stp->ls_ownerlen;
2427 		    NFSBCOPY(lop->lo_stp->ls_owner, cfp->cl_owner,
2428 			cfp->cl_ownerlen);
2429 		}
2430 		if (ret == 2)
2431 		    error = NFSERR_PERM;
2432 		else if (new_stp->ls_flags & NFSLCK_RECLAIM)
2433 		    error = NFSERR_RECLAIMCONFLICT;
2434 		else if (new_stp->ls_flags & NFSLCK_CHECK)
2435 		    error = NFSERR_LOCKED;
2436 		else
2437 		    error = NFSERR_DENIED;
2438 		if (filestruct_locked != 0 && ret == 0) {
2439 			/* Roll back local locks. */
2440 			NFSUNLOCKSTATE();
2441 			if (vnode_unlocked == 0) {
2442 				ASSERT_VOP_ELOCKED(vp, "nfsrv_lockctrl7");
2443 				vnode_unlocked = 1;
2444 				NFSVOPUNLOCK(vp);
2445 			}
2446 			nfsrv_locallock_rollback(vp, lfp, p);
2447 			NFSLOCKSTATE();
2448 			nfsrv_unlocklf(lfp);
2449 		}
2450 		if (ret == 0)
2451 			NFSUNLOCKSTATE();
2452 		goto out;
2453 	    }
2454 	  }
2455 	}
2456 
2457 	/*
2458 	 * We only get here if there was no lock that conflicted.
2459 	 */
2460 	if (new_stp->ls_flags & (NFSLCK_TEST | NFSLCK_CHECK)) {
2461 		NFSUNLOCKSTATE();
2462 		goto out;
2463 	}
2464 
2465 	/*
2466 	 * We only get here when we are creating or modifying a lock.
2467 	 * There are two variants:
2468 	 * - exist_lock_owner where lock_owner exists
2469 	 * - open_to_lock_owner with new lock_owner
2470 	 */
2471 	first = new_lop->lo_first;
2472 	end = new_lop->lo_end;
2473 	lock_flags = new_lop->lo_flags;
2474 	if (!(new_stp->ls_flags & NFSLCK_OPENTOLOCK)) {
2475 		nfsrv_updatelock(lckstp, new_lopp, &other_lop, lfp);
2476 		stateidp->seqid = ++(lckstp->ls_stateid.seqid);
2477 		if ((nd->nd_flag & ND_NFSV41) != 0 && stateidp->seqid == 0)
2478 			stateidp->seqid = lckstp->ls_stateid.seqid = 1;
2479 		stateidp->other[0] = lckstp->ls_stateid.other[0];
2480 		stateidp->other[1] = lckstp->ls_stateid.other[1];
2481 		stateidp->other[2] = lckstp->ls_stateid.other[2];
2482 	} else {
2483 		/*
2484 		 * The new open_to_lock_owner case.
2485 		 * Link the new nfsstate into the lists.
2486 		 */
2487 		new_stp->ls_seq = new_stp->ls_opentolockseq;
2488 		nfsrvd_refcache(new_stp->ls_op);
2489 		stateidp->seqid = new_stp->ls_stateid.seqid = 1;
2490 		stateidp->other[0] = new_stp->ls_stateid.other[0] =
2491 		    clp->lc_clientid.lval[0];
2492 		stateidp->other[1] = new_stp->ls_stateid.other[1] =
2493 		    clp->lc_clientid.lval[1];
2494 		stateidp->other[2] = new_stp->ls_stateid.other[2] =
2495 		    nfsrv_nextstateindex(clp);
2496 		new_stp->ls_clp = clp;
2497 		LIST_INIT(&new_stp->ls_lock);
2498 		new_stp->ls_openstp = stp;
2499 		new_stp->ls_lfp = lfp;
2500 		nfsrv_insertlock(new_lop, (struct nfslock *)new_stp, new_stp,
2501 		    lfp);
2502 		LIST_INSERT_HEAD(NFSSTATEHASH(clp, new_stp->ls_stateid),
2503 		    new_stp, ls_hash);
2504 		LIST_INSERT_HEAD(&stp->ls_open, new_stp, ls_list);
2505 		*new_lopp = NULL;
2506 		*new_stpp = NULL;
2507 		NFSD_VNET(nfsstatsv1_p)->srvlockowners++;
2508 		nfsrv_openpluslock++;
2509 	}
2510 	if (filestruct_locked != 0) {
2511 		NFSUNLOCKSTATE();
2512 		nfsrv_locallock_commit(lfp, lock_flags, first, end);
2513 		NFSLOCKSTATE();
2514 		nfsrv_unlocklf(lfp);
2515 	}
2516 	NFSUNLOCKSTATE();
2517 
2518 out:
2519 	if (haslock) {
2520 		NFSLOCKV4ROOTMUTEX();
2521 		nfsv4_unlock(&nfsv4rootfs_lock, 1);
2522 		NFSUNLOCKV4ROOTMUTEX();
2523 	}
2524 	if (vnode_unlocked != 0) {
2525 		NFSVOPLOCK(vp, LK_EXCLUSIVE | LK_RETRY);
2526 		if (error == 0 && VN_IS_DOOMED(vp))
2527 			error = NFSERR_SERVERFAULT;
2528 	}
2529 	if (other_lop)
2530 		free(other_lop, M_NFSDLOCK);
2531 	NFSEXITCODE2(error, nd);
2532 	return (error);
2533 }
2534 
2535 /*
2536  * Check for state errors for Open.
2537  * repstat is passed back out as an error if more critical errors
2538  * are not detected.
2539  */
2540 int
2541 nfsrv_opencheck(nfsquad_t clientid, nfsv4stateid_t *stateidp,
2542     struct nfsstate *new_stp, vnode_t vp, struct nfsrv_descript *nd,
2543     NFSPROC_T *p, int repstat)
2544 {
2545 	struct nfsstate *stp, *nstp;
2546 	struct nfsclient *clp;
2547 	struct nfsstate *ownerstp;
2548 	struct nfslockfile *lfp, *new_lfp;
2549 	int error = 0, haslock = 0, ret, readonly = 0, getfhret = 0;
2550 
2551 	if ((new_stp->ls_flags & NFSLCK_SHAREBITS) == NFSLCK_READACCESS)
2552 		readonly = 1;
2553 	/*
2554 	 * Check for restart conditions (client and server).
2555 	 */
2556 	error = nfsrv_checkrestart(clientid, new_stp->ls_flags,
2557 		&new_stp->ls_stateid, 0);
2558 	if (error)
2559 		goto out;
2560 
2561 	/*
2562 	 * Check for state resource limit exceeded.
2563 	 * Technically this should be SMP protected, but the worst
2564 	 * case error is "out by one or two" on the count when it
2565 	 * returns NFSERR_RESOURCE and the limit is just a rather
2566 	 * arbitrary high water mark, so no harm is done.
2567 	 */
2568 	if (nfsrv_openpluslock > nfsrv_v4statelimit) {
2569 		error = NFSERR_RESOURCE;
2570 		goto out;
2571 	}
2572 
2573 tryagain:
2574 	new_lfp = malloc(sizeof (struct nfslockfile),
2575 	    M_NFSDLOCKFILE, M_WAITOK);
2576 	if (vp)
2577 		getfhret = nfsrv_getlockfh(vp, new_stp->ls_flags, new_lfp,
2578 		    NULL, p);
2579 	NFSLOCKSTATE();
2580 	/*
2581 	 * Get the nfsclient structure.
2582 	 */
2583 	error = nfsrv_getclient(clientid, CLOPS_RENEW, &clp, NULL,
2584 	    (nfsquad_t)((u_quad_t)0), 0, nd, p);
2585 
2586 	/*
2587 	 * Look up the open owner. See if it needs confirmation and
2588 	 * check the seq#, as required.
2589 	 */
2590 	if (!error)
2591 		nfsrv_getowner(&clp->lc_open, new_stp, &ownerstp);
2592 
2593 	if (!error && ownerstp) {
2594 		error = nfsrv_checkseqid(nd, new_stp->ls_seq, ownerstp,
2595 		    new_stp->ls_op);
2596 		/*
2597 		 * If the OpenOwner hasn't been confirmed, assume the
2598 		 * old one was a replay and this one is ok.
2599 		 * See: RFC3530 Sec. 14.2.18.
2600 		 */
2601 		if (error == NFSERR_BADSEQID &&
2602 		    (ownerstp->ls_flags & NFSLCK_NEEDSCONFIRM))
2603 			error = 0;
2604 	}
2605 
2606 	/*
2607 	 * Check for grace.
2608 	 */
2609 	if (!error)
2610 		error = nfsrv_checkgrace(nd, clp, new_stp->ls_flags);
2611 	if ((new_stp->ls_flags & NFSLCK_RECLAIM) && !error &&
2612 		nfsrv_checkstable(clp))
2613 		error = NFSERR_NOGRACE;
2614 
2615 	/*
2616 	 * If none of the above errors occurred, let repstat be
2617 	 * returned.
2618 	 */
2619 	if (repstat && !error)
2620 		error = repstat;
2621 	if (error) {
2622 		NFSUNLOCKSTATE();
2623 		if (haslock) {
2624 			NFSLOCKV4ROOTMUTEX();
2625 			nfsv4_unlock(&nfsv4rootfs_lock, 1);
2626 			NFSUNLOCKV4ROOTMUTEX();
2627 		}
2628 		free(new_lfp, M_NFSDLOCKFILE);
2629 		goto out;
2630 	}
2631 
2632 	/*
2633 	 * If vp == NULL, the file doesn't exist yet, so return ok.
2634 	 * (This always happens on the first pass, so haslock must be 0.)
2635 	 */
2636 	if (vp == NULL) {
2637 		NFSUNLOCKSTATE();
2638 		free(new_lfp, M_NFSDLOCKFILE);
2639 		goto out;
2640 	}
2641 
2642 	/*
2643 	 * Get the structure for the underlying file.
2644 	 */
2645 	if (getfhret)
2646 		error = getfhret;
2647 	else
2648 		error = nfsrv_getlockfile(new_stp->ls_flags, &new_lfp, &lfp,
2649 		    NULL, 0);
2650 	if (new_lfp)
2651 		free(new_lfp, M_NFSDLOCKFILE);
2652 	if (error) {
2653 		NFSUNLOCKSTATE();
2654 		if (haslock) {
2655 			NFSLOCKV4ROOTMUTEX();
2656 			nfsv4_unlock(&nfsv4rootfs_lock, 1);
2657 			NFSUNLOCKV4ROOTMUTEX();
2658 		}
2659 		goto out;
2660 	}
2661 
2662 	/*
2663 	 * Search for a conflicting open/share.
2664 	 */
2665 	if (new_stp->ls_flags & NFSLCK_DELEGCUR) {
2666 	    /*
2667 	     * For Delegate_Cur, search for the matching Delegation,
2668 	     * which indicates no conflict.
2669 	     * An old delegation should have been recovered by the
2670 	     * client doing a Claim_DELEGATE_Prev, so I won't let
2671 	     * it match and return NFSERR_EXPIRED. Should I let it
2672 	     * match?
2673 	     */
2674 	    LIST_FOREACH(stp, &lfp->lf_deleg, ls_file) {
2675 		if (!(stp->ls_flags & NFSLCK_OLDDELEG) &&
2676 		    (((nd->nd_flag & ND_NFSV41) != 0 &&
2677 		    stateidp->seqid == 0) ||
2678 		    stateidp->seqid == stp->ls_stateid.seqid) &&
2679 		    !NFSBCMP(stateidp->other, stp->ls_stateid.other,
2680 			  NFSX_STATEIDOTHER))
2681 			break;
2682 	    }
2683 	    if (stp == LIST_END(&lfp->lf_deleg) ||
2684 		((new_stp->ls_flags & NFSLCK_WRITEACCESS) &&
2685 		 (stp->ls_flags & NFSLCK_DELEGREAD))) {
2686 		NFSUNLOCKSTATE();
2687 		if (haslock) {
2688 			NFSLOCKV4ROOTMUTEX();
2689 			nfsv4_unlock(&nfsv4rootfs_lock, 1);
2690 			NFSUNLOCKV4ROOTMUTEX();
2691 		}
2692 		error = NFSERR_EXPIRED;
2693 		goto out;
2694 	    }
2695 	}
2696 
2697 	/*
2698 	 * Check for access/deny bit conflicts. I check for the same
2699 	 * owner as well, in case the client didn't bother.
2700 	 */
2701 	LIST_FOREACH(stp, &lfp->lf_open, ls_file) {
2702 		if (!(new_stp->ls_flags & NFSLCK_DELEGCUR) &&
2703 		    (((new_stp->ls_flags & NFSLCK_ACCESSBITS) &
2704 		      ((stp->ls_flags>>NFSLCK_SHIFT) & NFSLCK_ACCESSBITS))||
2705 		     ((stp->ls_flags & NFSLCK_ACCESSBITS) &
2706 		      ((new_stp->ls_flags>>NFSLCK_SHIFT)&NFSLCK_ACCESSBITS)))){
2707 			ret = nfsrv_clientconflict(stp->ls_clp,&haslock,vp,p);
2708 			if (ret == 1) {
2709 				/*
2710 				 * nfsrv_clientconflict() unlocks
2711 				 * state when it returns non-zero.
2712 				 */
2713 				goto tryagain;
2714 			}
2715 			if (ret == 2)
2716 				error = NFSERR_PERM;
2717 			else if (new_stp->ls_flags & NFSLCK_RECLAIM)
2718 				error = NFSERR_RECLAIMCONFLICT;
2719 			else
2720 				error = NFSERR_SHAREDENIED;
2721 			if (ret == 0)
2722 				NFSUNLOCKSTATE();
2723 			if (haslock) {
2724 				NFSLOCKV4ROOTMUTEX();
2725 				nfsv4_unlock(&nfsv4rootfs_lock, 1);
2726 				NFSUNLOCKV4ROOTMUTEX();
2727 			}
2728 			goto out;
2729 		}
2730 	}
2731 
2732 	/*
2733 	 * Check for a conflicting delegation. If one is found, call
2734 	 * nfsrv_delegconflict() to handle it. If the v4root lock hasn't
2735 	 * been set yet, it will get the lock. Otherwise, it will recall
2736 	 * the delegation. Then, we try try again...
2737 	 * (If NFSLCK_DELEGCUR is set, it has a delegation, so there
2738 	 *  isn't a conflict.)
2739 	 * I currently believe the conflict algorithm to be:
2740 	 * For Open with Read Access and Deny None
2741 	 * - there is a conflict iff a different client has a write delegation
2742 	 * For Open with other Write Access or any Deny except None
2743 	 * - there is a conflict if a different client has any delegation
2744 	 * - there is a conflict if the same client has a read delegation
2745 	 *   (The current consensus is that this last case should be
2746 	 *    considered a conflict since the client with a read delegation
2747 	 *    could have done an Open with ReadAccess and WriteDeny
2748 	 *    locally and then not have checked for the WriteDeny.)
2749 	 *    The exception is a NFSv4.1/4.2 client that has requested
2750 	 *    an atomic upgrade to a write delegation.
2751 	 * Don't check for a Reclaim, since that will be dealt with
2752 	 * by nfsrv_openctrl().
2753 	 */
2754 	if (!(new_stp->ls_flags &
2755 		(NFSLCK_DELEGPREV | NFSLCK_DELEGCUR | NFSLCK_RECLAIM))) {
2756 	    stp = LIST_FIRST(&lfp->lf_deleg);
2757 	    while (stp != LIST_END(&lfp->lf_deleg)) {
2758 		nstp = LIST_NEXT(stp, ls_file);
2759 		if ((readonly && stp->ls_clp != clp &&
2760 		     (stp->ls_flags & NFSLCK_DELEGWRITE) != 0) ||
2761 		    (!readonly && (stp->ls_clp != clp ||
2762 		     ((stp->ls_flags & NFSLCK_DELEGREAD) != 0 &&
2763 		      (new_stp->ls_flags & NFSLCK_WANTWDELEG) == 0)))) {
2764 			ret = nfsrv_delegconflict(stp, &haslock, p, vp);
2765 			if (ret) {
2766 			    /*
2767 			     * nfsrv_delegconflict() unlocks state
2768 			     * when it returns non-zero.
2769 			     */
2770 			    if (ret == -1)
2771 				goto tryagain;
2772 			    error = ret;
2773 			    goto out;
2774 			}
2775 		}
2776 		stp = nstp;
2777 	    }
2778 	}
2779 	NFSUNLOCKSTATE();
2780 	if (haslock) {
2781 		NFSLOCKV4ROOTMUTEX();
2782 		nfsv4_unlock(&nfsv4rootfs_lock, 1);
2783 		NFSUNLOCKV4ROOTMUTEX();
2784 	}
2785 
2786 out:
2787 	NFSEXITCODE2(error, nd);
2788 	return (error);
2789 }
2790 
2791 /*
2792  * Open control function to create/update open state for an open.
2793  */
2794 int
2795 nfsrv_openctrl(struct nfsrv_descript *nd, vnode_t vp,
2796     struct nfsstate **new_stpp, nfsquad_t clientid, nfsv4stateid_t *stateidp,
2797     nfsv4stateid_t *delegstateidp, u_int32_t *rflagsp, struct nfsexstuff *exp,
2798     NFSPROC_T *p, u_quad_t filerev)
2799 {
2800 	struct nfsstate *new_stp = *new_stpp;
2801 	struct nfsstate *stp, *nstp;
2802 	struct nfsstate *openstp = NULL, *new_open, *ownerstp, *new_deleg;
2803 	struct nfslockfile *lfp, *new_lfp;
2804 	struct nfsclient *clp;
2805 	int error = 0, haslock = 0, ret, delegate = 1, writedeleg = 1;
2806 	int readonly = 0, cbret = 1, getfhret = 0;
2807 	int gotstate = 0, len = 0;
2808 	u_char *clidp = NULL;
2809 
2810 	if ((new_stp->ls_flags & NFSLCK_SHAREBITS) == NFSLCK_READACCESS)
2811 		readonly = 1;
2812 	/*
2813 	 * Check for restart conditions (client and server).
2814 	 * (Paranoia, should have been detected by nfsrv_opencheck().)
2815 	 * If an error does show up, return NFSERR_EXPIRED, since the
2816 	 * the seqid# has already been incremented.
2817 	 */
2818 	error = nfsrv_checkrestart(clientid, new_stp->ls_flags,
2819 	    &new_stp->ls_stateid, 0);
2820 	if (error) {
2821 		printf("Nfsd: openctrl unexpected restart err=%d\n",
2822 		    error);
2823 		error = NFSERR_EXPIRED;
2824 		goto out;
2825 	}
2826 
2827 	clidp = malloc(NFSV4_OPAQUELIMIT, M_TEMP, M_WAITOK);
2828 tryagain:
2829 	new_lfp = malloc(sizeof (struct nfslockfile),
2830 	    M_NFSDLOCKFILE, M_WAITOK);
2831 	new_open = malloc(sizeof (struct nfsstate),
2832 	    M_NFSDSTATE, M_WAITOK);
2833 	new_deleg = malloc(sizeof (struct nfsstate),
2834 	    M_NFSDSTATE, M_WAITOK);
2835 	getfhret = nfsrv_getlockfh(vp, new_stp->ls_flags, new_lfp,
2836 	    NULL, p);
2837 	NFSLOCKSTATE();
2838 	/*
2839 	 * Get the client structure. Since the linked lists could be changed
2840 	 * by other nfsd processes if this process does a tsleep(), one of
2841 	 * two things must be done.
2842 	 * 1 - don't tsleep()
2843 	 * or
2844 	 * 2 - get the nfsv4_lock() { indicated by haslock == 1 }
2845 	 *     before using the lists, since this lock stops the other
2846 	 *     nfsd. This should only be used for rare cases, since it
2847 	 *     essentially single threads the nfsd.
2848 	 *     At this time, it is only done for cases where the stable
2849 	 *     storage file must be written prior to completion of state
2850 	 *     expiration.
2851 	 */
2852 	error = nfsrv_getclient(clientid, CLOPS_RENEW, &clp, NULL,
2853 	    (nfsquad_t)((u_quad_t)0), 0, nd, p);
2854 	if (!error && (clp->lc_flags & LCL_NEEDSCBNULL) &&
2855 	    clp->lc_program) {
2856 		/*
2857 		 * This happens on the first open for a client
2858 		 * that supports callbacks.
2859 		 */
2860 		NFSUNLOCKSTATE();
2861 		/*
2862 		 * Although nfsrv_docallback() will sleep, clp won't
2863 		 * go away, since they are only removed when the
2864 		 * nfsv4_lock() has blocked the nfsd threads. The
2865 		 * fields in clp can change, but having multiple
2866 		 * threads do this Null callback RPC should be
2867 		 * harmless.
2868 		 */
2869 		cbret = nfsrv_docallback(clp, NFSV4PROC_CBNULL,
2870 		    NULL, 0, NULL, NULL, NULL, 0, p);
2871 		NFSLOCKSTATE();
2872 		clp->lc_flags &= ~LCL_NEEDSCBNULL;
2873 		if (!cbret)
2874 			clp->lc_flags |= LCL_CALLBACKSON;
2875 	}
2876 
2877 	/*
2878 	 * Look up the open owner. See if it needs confirmation and
2879 	 * check the seq#, as required.
2880 	 */
2881 	if (!error)
2882 		nfsrv_getowner(&clp->lc_open, new_stp, &ownerstp);
2883 
2884 	if (error) {
2885 		NFSUNLOCKSTATE();
2886 		printf("Nfsd: openctrl unexpected state err=%d\n",
2887 			error);
2888 		free(new_lfp, M_NFSDLOCKFILE);
2889 		free(new_open, M_NFSDSTATE);
2890 		free(new_deleg, M_NFSDSTATE);
2891 		if (haslock) {
2892 			NFSLOCKV4ROOTMUTEX();
2893 			nfsv4_unlock(&nfsv4rootfs_lock, 1);
2894 			NFSUNLOCKV4ROOTMUTEX();
2895 		}
2896 		error = NFSERR_EXPIRED;
2897 		goto out;
2898 	}
2899 
2900 	if (new_stp->ls_flags & NFSLCK_RECLAIM)
2901 		nfsrv_markstable(clp);
2902 
2903 	/*
2904 	 * Get the structure for the underlying file.
2905 	 */
2906 	if (getfhret)
2907 		error = getfhret;
2908 	else
2909 		error = nfsrv_getlockfile(new_stp->ls_flags, &new_lfp, &lfp,
2910 		    NULL, 0);
2911 	if (new_lfp)
2912 		free(new_lfp, M_NFSDLOCKFILE);
2913 	if (error) {
2914 		NFSUNLOCKSTATE();
2915 		printf("Nfsd openctrl unexpected getlockfile err=%d\n",
2916 		    error);
2917 		free(new_open, M_NFSDSTATE);
2918 		free(new_deleg, M_NFSDSTATE);
2919 		if (haslock) {
2920 			NFSLOCKV4ROOTMUTEX();
2921 			nfsv4_unlock(&nfsv4rootfs_lock, 1);
2922 			NFSUNLOCKV4ROOTMUTEX();
2923 		}
2924 		goto out;
2925 	}
2926 
2927 	/*
2928 	 * Search for a conflicting open/share.
2929 	 */
2930 	if (new_stp->ls_flags & NFSLCK_DELEGCUR) {
2931 	    /*
2932 	     * For Delegate_Cur, search for the matching Delegation,
2933 	     * which indicates no conflict.
2934 	     * An old delegation should have been recovered by the
2935 	     * client doing a Claim_DELEGATE_Prev, so I won't let
2936 	     * it match and return NFSERR_EXPIRED. Should I let it
2937 	     * match?
2938 	     */
2939 	    LIST_FOREACH(stp, &lfp->lf_deleg, ls_file) {
2940 		if (!(stp->ls_flags & NFSLCK_OLDDELEG) &&
2941 		    (((nd->nd_flag & ND_NFSV41) != 0 &&
2942 		    stateidp->seqid == 0) ||
2943 		    stateidp->seqid == stp->ls_stateid.seqid) &&
2944 		    !NFSBCMP(stateidp->other, stp->ls_stateid.other,
2945 			NFSX_STATEIDOTHER))
2946 			break;
2947 	    }
2948 	    if (stp == LIST_END(&lfp->lf_deleg) ||
2949 		((new_stp->ls_flags & NFSLCK_WRITEACCESS) &&
2950 		 (stp->ls_flags & NFSLCK_DELEGREAD))) {
2951 		NFSUNLOCKSTATE();
2952 		printf("Nfsd openctrl unexpected expiry\n");
2953 		free(new_open, M_NFSDSTATE);
2954 		free(new_deleg, M_NFSDSTATE);
2955 		if (haslock) {
2956 			NFSLOCKV4ROOTMUTEX();
2957 			nfsv4_unlock(&nfsv4rootfs_lock, 1);
2958 			NFSUNLOCKV4ROOTMUTEX();
2959 		}
2960 		error = NFSERR_EXPIRED;
2961 		goto out;
2962 	    }
2963 
2964 	    /*
2965 	     * Don't issue a Delegation, since one already exists and
2966 	     * delay delegation timeout, as required.
2967 	     */
2968 	    delegate = 0;
2969 	    nfsrv_delaydelegtimeout(stp);
2970 	}
2971 
2972 	/*
2973 	 * Check for access/deny bit conflicts. I also check for the
2974 	 * same owner, since the client might not have bothered to check.
2975 	 * Also, note an open for the same file and owner, if found,
2976 	 * which is all we do here for Delegate_Cur, since conflict
2977 	 * checking is already done.
2978 	 */
2979 	LIST_FOREACH(stp, &lfp->lf_open, ls_file) {
2980 		if (ownerstp && stp->ls_openowner == ownerstp)
2981 			openstp = stp;
2982 		if (!(new_stp->ls_flags & NFSLCK_DELEGCUR)) {
2983 		    /*
2984 		     * If another client has the file open, the only
2985 		     * delegation that can be issued is a Read delegation
2986 		     * and only if it is a Read open with Deny none.
2987 		     */
2988 		    if (clp != stp->ls_clp) {
2989 			if ((stp->ls_flags & NFSLCK_SHAREBITS) ==
2990 			    NFSLCK_READACCESS)
2991 			    writedeleg = 0;
2992 			else
2993 			    delegate = 0;
2994 		    }
2995 		    if(((new_stp->ls_flags & NFSLCK_ACCESSBITS) &
2996 		        ((stp->ls_flags>>NFSLCK_SHIFT) & NFSLCK_ACCESSBITS))||
2997 		       ((stp->ls_flags & NFSLCK_ACCESSBITS) &
2998 		        ((new_stp->ls_flags>>NFSLCK_SHIFT)&NFSLCK_ACCESSBITS))){
2999 			ret = nfsrv_clientconflict(stp->ls_clp,&haslock,vp,p);
3000 			if (ret == 1) {
3001 				/*
3002 				 * nfsrv_clientconflict() unlocks state
3003 				 * when it returns non-zero.
3004 				 */
3005 				free(new_open, M_NFSDSTATE);
3006 				free(new_deleg, M_NFSDSTATE);
3007 				openstp = NULL;
3008 				goto tryagain;
3009 			}
3010 			if (ret == 2)
3011 				error = NFSERR_PERM;
3012 			else if (new_stp->ls_flags & NFSLCK_RECLAIM)
3013 				error = NFSERR_RECLAIMCONFLICT;
3014 			else
3015 				error = NFSERR_SHAREDENIED;
3016 			if (ret == 0)
3017 				NFSUNLOCKSTATE();
3018 			if (haslock) {
3019 				NFSLOCKV4ROOTMUTEX();
3020 				nfsv4_unlock(&nfsv4rootfs_lock, 1);
3021 				NFSUNLOCKV4ROOTMUTEX();
3022 			}
3023 			free(new_open, M_NFSDSTATE);
3024 			free(new_deleg, M_NFSDSTATE);
3025 			printf("nfsd openctrl unexpected client cnfl\n");
3026 			goto out;
3027 		    }
3028 		}
3029 	}
3030 
3031 	/*
3032 	 * Check for a conflicting delegation. If one is found, call
3033 	 * nfsrv_delegconflict() to handle it. If the v4root lock hasn't
3034 	 * been set yet, it will get the lock. Otherwise, it will recall
3035 	 * the delegation. Then, we try try again...
3036 	 * (If NFSLCK_DELEGCUR is set, it has a delegation, so there
3037 	 *  isn't a conflict.)
3038 	 * I currently believe the conflict algorithm to be:
3039 	 * For Open with Read Access and Deny None
3040 	 * - there is a conflict iff a different client has a write delegation
3041 	 * For Open with other Write Access or any Deny except None
3042 	 * - there is a conflict if a different client has any delegation
3043 	 * - there is a conflict if the same client has a read delegation
3044 	 *   (The current consensus is that this last case should be
3045 	 *    considered a conflict since the client with a read delegation
3046 	 *    could have done an Open with ReadAccess and WriteDeny
3047 	 *    locally and then not have checked for the WriteDeny.)
3048 	 *    The exception is a NFSv4.1/4.2 client that has requested
3049 	 *    an atomic upgrade to a write delegation.
3050 	 */
3051 	if (!(new_stp->ls_flags & (NFSLCK_DELEGPREV | NFSLCK_DELEGCUR))) {
3052 	    stp = LIST_FIRST(&lfp->lf_deleg);
3053 	    while (stp != LIST_END(&lfp->lf_deleg)) {
3054 		nstp = LIST_NEXT(stp, ls_file);
3055 		if (stp->ls_clp != clp && (stp->ls_flags & NFSLCK_DELEGREAD))
3056 			writedeleg = 0;
3057 		else if (stp->ls_clp != clp ||
3058 		    (stp->ls_flags & NFSLCK_DELEGWRITE) != 0 ||
3059 		    (new_stp->ls_flags & NFSLCK_WANTWDELEG) == 0)
3060 			delegate = 0;
3061 		if ((readonly && stp->ls_clp != clp &&
3062 		     (stp->ls_flags & NFSLCK_DELEGWRITE) != 0) ||
3063 		    (!readonly && (stp->ls_clp != clp ||
3064 		     ((stp->ls_flags & NFSLCK_DELEGREAD) != 0 &&
3065 		      (new_stp->ls_flags & NFSLCK_WANTWDELEG) == 0)))) {
3066 		    if (new_stp->ls_flags & NFSLCK_RECLAIM) {
3067 			delegate = 2;
3068 		    } else {
3069 			ret = nfsrv_delegconflict(stp, &haslock, p, vp);
3070 			if (ret) {
3071 			    /*
3072 			     * nfsrv_delegconflict() unlocks state
3073 			     * when it returns non-zero.
3074 			     */
3075 			    printf("Nfsd openctrl unexpected deleg cnfl\n");
3076 			    free(new_open, M_NFSDSTATE);
3077 			    free(new_deleg, M_NFSDSTATE);
3078 			    if (ret == -1) {
3079 				openstp = NULL;
3080 				goto tryagain;
3081 			    }
3082 			    error = ret;
3083 			    goto out;
3084 			}
3085 		    }
3086 		}
3087 		stp = nstp;
3088 	    }
3089 	}
3090 
3091 	/*
3092 	 * We only get here if there was no open that conflicted.
3093 	 * If an open for the owner exists, or in the access/deny bits.
3094 	 * Otherwise it is a new open. If the open_owner hasn't been
3095 	 * confirmed, replace the open with the new one needing confirmation,
3096 	 * otherwise add the open.
3097 	 */
3098 	if (new_stp->ls_flags & NFSLCK_DELEGPREV) {
3099 	    /*
3100 	     * Handle NFSLCK_DELEGPREV by searching the old delegations for
3101 	     * a match. If found, just move the old delegation to the current
3102 	     * delegation list and issue open. If not found, return
3103 	     * NFSERR_EXPIRED.
3104 	     */
3105 	    LIST_FOREACH(stp, &clp->lc_olddeleg, ls_list) {
3106 		if (stp->ls_lfp == lfp) {
3107 		    /* Found it */
3108 		    if (stp->ls_clp != clp)
3109 			panic("olddeleg clp");
3110 		    LIST_REMOVE(stp, ls_list);
3111 		    LIST_REMOVE(stp, ls_hash);
3112 		    stp->ls_flags &= ~NFSLCK_OLDDELEG;
3113 		    stp->ls_stateid.seqid = delegstateidp->seqid = 1;
3114 		    stp->ls_stateid.other[0] = delegstateidp->other[0] =
3115 			clp->lc_clientid.lval[0];
3116 		    stp->ls_stateid.other[1] = delegstateidp->other[1] =
3117 			clp->lc_clientid.lval[1];
3118 		    stp->ls_stateid.other[2] = delegstateidp->other[2] =
3119 			nfsrv_nextstateindex(clp);
3120 		    stp->ls_compref = nd->nd_compref;
3121 		    LIST_INSERT_HEAD(&clp->lc_deleg, stp, ls_list);
3122 		    LIST_INSERT_HEAD(NFSSTATEHASH(clp,
3123 			stp->ls_stateid), stp, ls_hash);
3124 		    if (stp->ls_flags & NFSLCK_DELEGWRITE)
3125 			*rflagsp |= NFSV4OPEN_WRITEDELEGATE;
3126 		    else
3127 			*rflagsp |= NFSV4OPEN_READDELEGATE;
3128 		    clp->lc_delegtime = NFSD_MONOSEC +
3129 			nfsrv_lease + NFSRV_LEASEDELTA;
3130 
3131 		    /*
3132 		     * Now, do the associated open.
3133 		     */
3134 		    new_open->ls_stateid.seqid = 1;
3135 		    new_open->ls_stateid.other[0] = clp->lc_clientid.lval[0];
3136 		    new_open->ls_stateid.other[1] = clp->lc_clientid.lval[1];
3137 		    new_open->ls_stateid.other[2] = nfsrv_nextstateindex(clp);
3138 		    new_open->ls_flags = (new_stp->ls_flags&NFSLCK_DENYBITS)|
3139 			NFSLCK_OPEN;
3140 		    if (stp->ls_flags & NFSLCK_DELEGWRITE)
3141 			new_open->ls_flags |= (NFSLCK_READACCESS |
3142 			    NFSLCK_WRITEACCESS);
3143 		    else
3144 			new_open->ls_flags |= NFSLCK_READACCESS;
3145 		    new_open->ls_uid = new_stp->ls_uid;
3146 		    new_open->ls_lfp = lfp;
3147 		    new_open->ls_clp = clp;
3148 		    LIST_INIT(&new_open->ls_open);
3149 		    LIST_INSERT_HEAD(&lfp->lf_open, new_open, ls_file);
3150 		    LIST_INSERT_HEAD(NFSSTATEHASH(clp, new_open->ls_stateid),
3151 			new_open, ls_hash);
3152 		    /*
3153 		     * and handle the open owner
3154 		     */
3155 		    if (ownerstp) {
3156 			new_open->ls_openowner = ownerstp;
3157 			LIST_INSERT_HEAD(&ownerstp->ls_open,new_open,ls_list);
3158 		    } else {
3159 			new_open->ls_openowner = new_stp;
3160 			new_stp->ls_flags = 0;
3161 			nfsrvd_refcache(new_stp->ls_op);
3162 			new_stp->ls_noopens = 0;
3163 			LIST_INIT(&new_stp->ls_open);
3164 			LIST_INSERT_HEAD(&new_stp->ls_open, new_open, ls_list);
3165 			LIST_INSERT_HEAD(&clp->lc_open, new_stp, ls_list);
3166 			*new_stpp = NULL;
3167 			NFSD_VNET(nfsstatsv1_p)->srvopenowners++;
3168 			nfsrv_openpluslock++;
3169 		    }
3170 		    openstp = new_open;
3171 		    new_open = NULL;
3172 		    NFSD_VNET(nfsstatsv1_p)->srvopens++;
3173 		    nfsrv_openpluslock++;
3174 		    break;
3175 		}
3176 	    }
3177 	    if (stp == LIST_END(&clp->lc_olddeleg))
3178 		error = NFSERR_EXPIRED;
3179 	} else if (new_stp->ls_flags & (NFSLCK_DELEGREAD | NFSLCK_DELEGWRITE)) {
3180 	    /*
3181 	     * Scan to see that no delegation for this client and file
3182 	     * doesn't already exist.
3183 	     * There also shouldn't yet be an Open for this file and
3184 	     * openowner.
3185 	     */
3186 	    LIST_FOREACH(stp, &lfp->lf_deleg, ls_file) {
3187 		if (stp->ls_clp == clp)
3188 		    break;
3189 	    }
3190 	    if (stp == LIST_END(&lfp->lf_deleg) && openstp == NULL) {
3191 		/*
3192 		 * This is the Claim_Previous case with a delegation
3193 		 * type != Delegate_None.
3194 		 */
3195 		/*
3196 		 * First, add the delegation. (Although we must issue the
3197 		 * delegation, we can also ask for an immediate return.)
3198 		 */
3199 		new_deleg->ls_stateid.seqid = delegstateidp->seqid = 1;
3200 		new_deleg->ls_stateid.other[0] = delegstateidp->other[0] =
3201 		    clp->lc_clientid.lval[0];
3202 		new_deleg->ls_stateid.other[1] = delegstateidp->other[1] =
3203 		    clp->lc_clientid.lval[1];
3204 		new_deleg->ls_stateid.other[2] = delegstateidp->other[2] =
3205 		    nfsrv_nextstateindex(clp);
3206 		if (new_stp->ls_flags & NFSLCK_DELEGWRITE) {
3207 		    new_deleg->ls_flags = (NFSLCK_DELEGWRITE |
3208 			NFSLCK_READACCESS | NFSLCK_WRITEACCESS);
3209 		    *rflagsp |= NFSV4OPEN_WRITEDELEGATE;
3210 		    nfsrv_writedelegcnt++;
3211 		} else {
3212 		    new_deleg->ls_flags = (NFSLCK_DELEGREAD |
3213 			NFSLCK_READACCESS);
3214 		    *rflagsp |= NFSV4OPEN_READDELEGATE;
3215 		}
3216 		new_deleg->ls_uid = new_stp->ls_uid;
3217 		new_deleg->ls_lfp = lfp;
3218 		new_deleg->ls_clp = clp;
3219 		new_deleg->ls_filerev = filerev;
3220 		new_deleg->ls_compref = nd->nd_compref;
3221 		new_deleg->ls_lastrecall = 0;
3222 		LIST_INSERT_HEAD(&lfp->lf_deleg, new_deleg, ls_file);
3223 		LIST_INSERT_HEAD(NFSSTATEHASH(clp,
3224 		    new_deleg->ls_stateid), new_deleg, ls_hash);
3225 		LIST_INSERT_HEAD(&clp->lc_deleg, new_deleg, ls_list);
3226 		new_deleg = NULL;
3227 		if (delegate == 2 || nfsrv_issuedelegs == 0 ||
3228 		    (clp->lc_flags & (LCL_CALLBACKSON | LCL_CBDOWN)) !=
3229 		     LCL_CALLBACKSON ||
3230 		    NFSRV_V4DELEGLIMIT(nfsrv_delegatecnt) ||
3231 		    !NFSVNO_DELEGOK(vp))
3232 		    *rflagsp |= NFSV4OPEN_RECALL;
3233 		NFSD_VNET(nfsstatsv1_p)->srvdelegates++;
3234 		nfsrv_openpluslock++;
3235 		nfsrv_delegatecnt++;
3236 
3237 		/*
3238 		 * Now, do the associated open.
3239 		 */
3240 		new_open->ls_stateid.seqid = 1;
3241 		new_open->ls_stateid.other[0] = clp->lc_clientid.lval[0];
3242 		new_open->ls_stateid.other[1] = clp->lc_clientid.lval[1];
3243 		new_open->ls_stateid.other[2] = nfsrv_nextstateindex(clp);
3244 		new_open->ls_flags = (new_stp->ls_flags & NFSLCK_DENYBITS) |
3245 		    NFSLCK_OPEN;
3246 		if (new_stp->ls_flags & NFSLCK_DELEGWRITE)
3247 			new_open->ls_flags |= (NFSLCK_READACCESS |
3248 			    NFSLCK_WRITEACCESS);
3249 		else
3250 			new_open->ls_flags |= NFSLCK_READACCESS;
3251 		new_open->ls_uid = new_stp->ls_uid;
3252 		new_open->ls_lfp = lfp;
3253 		new_open->ls_clp = clp;
3254 		LIST_INIT(&new_open->ls_open);
3255 		LIST_INSERT_HEAD(&lfp->lf_open, new_open, ls_file);
3256 		LIST_INSERT_HEAD(NFSSTATEHASH(clp, new_open->ls_stateid),
3257 		   new_open, ls_hash);
3258 		/*
3259 		 * and handle the open owner
3260 		 */
3261 		if (ownerstp) {
3262 		    new_open->ls_openowner = ownerstp;
3263 		    LIST_INSERT_HEAD(&ownerstp->ls_open, new_open, ls_list);
3264 		} else {
3265 		    new_open->ls_openowner = new_stp;
3266 		    new_stp->ls_flags = 0;
3267 		    nfsrvd_refcache(new_stp->ls_op);
3268 		    new_stp->ls_noopens = 0;
3269 		    LIST_INIT(&new_stp->ls_open);
3270 		    LIST_INSERT_HEAD(&new_stp->ls_open, new_open, ls_list);
3271 		    LIST_INSERT_HEAD(&clp->lc_open, new_stp, ls_list);
3272 		    *new_stpp = NULL;
3273 		    NFSD_VNET(nfsstatsv1_p)->srvopenowners++;
3274 		    nfsrv_openpluslock++;
3275 		}
3276 		openstp = new_open;
3277 		new_open = NULL;
3278 		NFSD_VNET(nfsstatsv1_p)->srvopens++;
3279 		nfsrv_openpluslock++;
3280 	    } else {
3281 		error = NFSERR_RECLAIMCONFLICT;
3282 	    }
3283 	} else if (ownerstp) {
3284 		if (ownerstp->ls_flags & NFSLCK_NEEDSCONFIRM) {
3285 		    /* Replace the open */
3286 		    if (ownerstp->ls_op)
3287 			nfsrvd_derefcache(ownerstp->ls_op);
3288 		    ownerstp->ls_op = new_stp->ls_op;
3289 		    nfsrvd_refcache(ownerstp->ls_op);
3290 		    ownerstp->ls_seq = new_stp->ls_seq;
3291 		    *rflagsp |= NFSV4OPEN_RESULTCONFIRM;
3292 		    stp = LIST_FIRST(&ownerstp->ls_open);
3293 		    stp->ls_flags = (new_stp->ls_flags & NFSLCK_SHAREBITS) |
3294 			NFSLCK_OPEN;
3295 		    stp->ls_stateid.seqid = 1;
3296 		    stp->ls_uid = new_stp->ls_uid;
3297 		    if (lfp != stp->ls_lfp) {
3298 			LIST_REMOVE(stp, ls_file);
3299 			LIST_INSERT_HEAD(&lfp->lf_open, stp, ls_file);
3300 			stp->ls_lfp = lfp;
3301 		    }
3302 		    openstp = stp;
3303 		} else if (openstp) {
3304 		    openstp->ls_flags |= (new_stp->ls_flags & NFSLCK_SHAREBITS);
3305 		    openstp->ls_stateid.seqid++;
3306 		    if ((nd->nd_flag & ND_NFSV41) != 0 &&
3307 			openstp->ls_stateid.seqid == 0)
3308 			openstp->ls_stateid.seqid = 1;
3309 
3310 		    /*
3311 		     * This is where we can choose to issue a delegation.
3312 		     */
3313 		    nfsrv_issuedelegation(vp, clp, nd, delegate, writedeleg,
3314 			readonly, filerev, NFSVNO_EXRDONLY(exp), &new_deleg,
3315 			new_stp, lfp, rflagsp, delegstateidp);
3316 		} else {
3317 		    new_open->ls_stateid.seqid = 1;
3318 		    new_open->ls_stateid.other[0] = clp->lc_clientid.lval[0];
3319 		    new_open->ls_stateid.other[1] = clp->lc_clientid.lval[1];
3320 		    new_open->ls_stateid.other[2] = nfsrv_nextstateindex(clp);
3321 		    new_open->ls_flags = (new_stp->ls_flags & NFSLCK_SHAREBITS)|
3322 			NFSLCK_OPEN;
3323 		    new_open->ls_uid = new_stp->ls_uid;
3324 		    new_open->ls_openowner = ownerstp;
3325 		    new_open->ls_lfp = lfp;
3326 		    new_open->ls_clp = clp;
3327 		    LIST_INIT(&new_open->ls_open);
3328 		    LIST_INSERT_HEAD(&lfp->lf_open, new_open, ls_file);
3329 		    LIST_INSERT_HEAD(&ownerstp->ls_open, new_open, ls_list);
3330 		    LIST_INSERT_HEAD(NFSSTATEHASH(clp, new_open->ls_stateid),
3331 			new_open, ls_hash);
3332 		    openstp = new_open;
3333 		    new_open = NULL;
3334 		    NFSD_VNET(nfsstatsv1_p)->srvopens++;
3335 		    nfsrv_openpluslock++;
3336 
3337 		    /*
3338 		     * This is where we can choose to issue a delegation.
3339 		     */
3340 		    nfsrv_issuedelegation(vp, clp, nd, delegate, writedeleg,
3341 			readonly, filerev, NFSVNO_EXRDONLY(exp), &new_deleg,
3342 			new_stp, lfp, rflagsp, delegstateidp);
3343 		}
3344 	} else {
3345 		/*
3346 		 * New owner case. Start the open_owner sequence with a
3347 		 * Needs confirmation (unless a reclaim) and hang the
3348 		 * new open off it.
3349 		 */
3350 		new_open->ls_stateid.seqid = 1;
3351 		new_open->ls_stateid.other[0] = clp->lc_clientid.lval[0];
3352 		new_open->ls_stateid.other[1] = clp->lc_clientid.lval[1];
3353 		new_open->ls_stateid.other[2] = nfsrv_nextstateindex(clp);
3354 		new_open->ls_flags = (new_stp->ls_flags & NFSLCK_SHAREBITS) |
3355 		    NFSLCK_OPEN;
3356 		new_open->ls_uid = new_stp->ls_uid;
3357 		LIST_INIT(&new_open->ls_open);
3358 		new_open->ls_openowner = new_stp;
3359 		new_open->ls_lfp = lfp;
3360 		new_open->ls_clp = clp;
3361 		LIST_INSERT_HEAD(&lfp->lf_open, new_open, ls_file);
3362 		if (new_stp->ls_flags & NFSLCK_RECLAIM) {
3363 			new_stp->ls_flags = 0;
3364 		} else if ((nd->nd_flag & ND_NFSV41) != 0) {
3365 		    /*
3366 		     * This is where we can choose to issue a delegation.
3367 		     */
3368 		    nfsrv_issuedelegation(vp, clp, nd, delegate, writedeleg,
3369 			readonly, filerev, NFSVNO_EXRDONLY(exp), &new_deleg,
3370 			new_stp, lfp, rflagsp, delegstateidp);
3371 		    /* NFSv4.1 never needs confirmation. */
3372 		    new_stp->ls_flags = 0;
3373 
3374 		    /*
3375 		     * Since NFSv4.1 never does an OpenConfirm, the first
3376 		     * open state will be acquired here.
3377 		     */
3378 		    if (!(clp->lc_flags & LCL_STAMPEDSTABLE)) {
3379 			clp->lc_flags |= LCL_STAMPEDSTABLE;
3380 			len = clp->lc_idlen;
3381 			NFSBCOPY(clp->lc_id, clidp, len);
3382 			gotstate = 1;
3383 		    }
3384 		} else {
3385 		    *rflagsp |= NFSV4OPEN_RESULTCONFIRM;
3386 		    new_stp->ls_flags = NFSLCK_NEEDSCONFIRM;
3387 		}
3388 		nfsrvd_refcache(new_stp->ls_op);
3389 		new_stp->ls_noopens = 0;
3390 		LIST_INIT(&new_stp->ls_open);
3391 		LIST_INSERT_HEAD(&new_stp->ls_open, new_open, ls_list);
3392 		LIST_INSERT_HEAD(&clp->lc_open, new_stp, ls_list);
3393 		LIST_INSERT_HEAD(NFSSTATEHASH(clp, new_open->ls_stateid),
3394 		    new_open, ls_hash);
3395 		openstp = new_open;
3396 		new_open = NULL;
3397 		*new_stpp = NULL;
3398 		NFSD_VNET(nfsstatsv1_p)->srvopens++;
3399 		nfsrv_openpluslock++;
3400 		NFSD_VNET(nfsstatsv1_p)->srvopenowners++;
3401 		nfsrv_openpluslock++;
3402 	}
3403 	if (!error) {
3404 		stateidp->seqid = openstp->ls_stateid.seqid;
3405 		stateidp->other[0] = openstp->ls_stateid.other[0];
3406 		stateidp->other[1] = openstp->ls_stateid.other[1];
3407 		stateidp->other[2] = openstp->ls_stateid.other[2];
3408 	}
3409 	NFSUNLOCKSTATE();
3410 	if (haslock) {
3411 		NFSLOCKV4ROOTMUTEX();
3412 		nfsv4_unlock(&nfsv4rootfs_lock, 1);
3413 		NFSUNLOCKV4ROOTMUTEX();
3414 	}
3415 	if (new_open)
3416 		free(new_open, M_NFSDSTATE);
3417 	if (new_deleg)
3418 		free(new_deleg, M_NFSDSTATE);
3419 
3420 	/*
3421 	 * If the NFSv4.1 client just acquired its first open, write a timestamp
3422 	 * to the stable storage file.
3423 	 */
3424 	if (gotstate != 0) {
3425 		nfsrv_writestable(clidp, len, NFSNST_NEWSTATE, p);
3426 		nfsrv_backupstable();
3427 	}
3428 
3429 out:
3430 	free(clidp, M_TEMP);
3431 	NFSEXITCODE2(error, nd);
3432 	return (error);
3433 }
3434 
3435 /*
3436  * Open update. Does the confirm, downgrade and close.
3437  */
3438 int
3439 nfsrv_openupdate(vnode_t vp, struct nfsstate *new_stp, nfsquad_t clientid,
3440     nfsv4stateid_t *stateidp, struct nfsrv_descript *nd, NFSPROC_T *p,
3441     int *retwriteaccessp)
3442 {
3443 	struct nfsstate *stp;
3444 	struct nfsclient *clp;
3445 	u_int32_t bits;
3446 	int error = 0, gotstate = 0, len = 0;
3447 	u_char *clidp = NULL;
3448 
3449 	/*
3450 	 * Check for restart conditions (client and server).
3451 	 */
3452 	error = nfsrv_checkrestart(clientid, new_stp->ls_flags,
3453 	    &new_stp->ls_stateid, 0);
3454 	if (error)
3455 		goto out;
3456 
3457 	clidp = malloc(NFSV4_OPAQUELIMIT, M_TEMP, M_WAITOK);
3458 	NFSLOCKSTATE();
3459 	/*
3460 	 * Get the open structure via clientid and stateid.
3461 	 */
3462 	error = nfsrv_getclient(clientid, CLOPS_RENEW, &clp, NULL,
3463 	    (nfsquad_t)((u_quad_t)0), 0, nd, p);
3464 	if (!error)
3465 		error = nfsrv_getstate(clp, &new_stp->ls_stateid,
3466 		    new_stp->ls_flags, &stp);
3467 
3468 	/*
3469 	 * Sanity check the open.
3470 	 */
3471 	if (!error && (!(stp->ls_flags & NFSLCK_OPEN) ||
3472 		(!(new_stp->ls_flags & NFSLCK_CONFIRM) &&
3473 		 (stp->ls_openowner->ls_flags & NFSLCK_NEEDSCONFIRM)) ||
3474 		((new_stp->ls_flags & NFSLCK_CONFIRM) &&
3475 		 (!(stp->ls_openowner->ls_flags & NFSLCK_NEEDSCONFIRM)))))
3476 		error = NFSERR_BADSTATEID;
3477 
3478 	if (!error)
3479 		error = nfsrv_checkseqid(nd, new_stp->ls_seq,
3480 		    stp->ls_openowner, new_stp->ls_op);
3481 	if (!error && stp->ls_stateid.seqid != new_stp->ls_stateid.seqid &&
3482 	    (((nd->nd_flag & ND_NFSV41) == 0 &&
3483 	      !(new_stp->ls_flags & NFSLCK_CONFIRM)) ||
3484 	     ((nd->nd_flag & ND_NFSV41) != 0 &&
3485 	      new_stp->ls_stateid.seqid != 0)))
3486 		error = NFSERR_OLDSTATEID;
3487 	if (!error && vp->v_type != VREG) {
3488 		if (vp->v_type == VDIR)
3489 			error = NFSERR_ISDIR;
3490 		else
3491 			error = NFSERR_INVAL;
3492 	}
3493 
3494 	if (error) {
3495 		/*
3496 		 * If a client tries to confirm an Open with a bad
3497 		 * seqid# and there are no byte range locks or other Opens
3498 		 * on the openowner, just throw it away, so the next use of the
3499 		 * openowner will start a fresh seq#.
3500 		 */
3501 		if (error == NFSERR_BADSEQID &&
3502 		    (new_stp->ls_flags & NFSLCK_CONFIRM) &&
3503 		    nfsrv_nootherstate(stp))
3504 			nfsrv_freeopenowner(stp->ls_openowner, 0, p);
3505 		NFSUNLOCKSTATE();
3506 		goto out;
3507 	}
3508 
3509 	/*
3510 	 * Set the return stateid.
3511 	 */
3512 	stateidp->seqid = stp->ls_stateid.seqid + 1;
3513 	if ((nd->nd_flag & ND_NFSV41) != 0 && stateidp->seqid == 0)
3514 		stateidp->seqid = 1;
3515 	stateidp->other[0] = stp->ls_stateid.other[0];
3516 	stateidp->other[1] = stp->ls_stateid.other[1];
3517 	stateidp->other[2] = stp->ls_stateid.other[2];
3518 	/*
3519 	 * Now, handle the three cases.
3520 	 */
3521 	if (new_stp->ls_flags & NFSLCK_CONFIRM) {
3522 		/*
3523 		 * If the open doesn't need confirmation, it seems to me that
3524 		 * there is a client error, but I'll just log it and keep going?
3525 		 */
3526 		if (!(stp->ls_openowner->ls_flags & NFSLCK_NEEDSCONFIRM))
3527 			printf("Nfsv4d: stray open confirm\n");
3528 		stp->ls_openowner->ls_flags = 0;
3529 		stp->ls_stateid.seqid++;
3530 		if ((nd->nd_flag & ND_NFSV41) != 0 &&
3531 		    stp->ls_stateid.seqid == 0)
3532 			stp->ls_stateid.seqid = 1;
3533 		if (!(clp->lc_flags & LCL_STAMPEDSTABLE)) {
3534 			clp->lc_flags |= LCL_STAMPEDSTABLE;
3535 			len = clp->lc_idlen;
3536 			NFSBCOPY(clp->lc_id, clidp, len);
3537 			gotstate = 1;
3538 		}
3539 	} else if (new_stp->ls_flags & NFSLCK_CLOSE) {
3540 		if (retwriteaccessp != NULL) {
3541 			if ((stp->ls_flags & NFSLCK_WRITEACCESS) != 0)
3542 				*retwriteaccessp = 1;
3543 			else
3544 				*retwriteaccessp = 0;
3545 		}
3546 		if (nfsrv_dolocallocks != 0 && !LIST_EMPTY(&stp->ls_open)) {
3547 			ASSERT_VOP_ELOCKED(vp, "nfsrv_openupdate");
3548 			nfsrv_freeopen(stp, vp, 1, p);
3549 		} else {
3550 			nfsrv_freeopen(stp, NULL, 0, p);
3551 		}
3552 	} else {
3553 		/*
3554 		 * Update the share bits, making sure that the new set are a
3555 		 * subset of the old ones.
3556 		 */
3557 		bits = (new_stp->ls_flags & NFSLCK_SHAREBITS);
3558 		if (~(stp->ls_flags) & bits) {
3559 			NFSUNLOCKSTATE();
3560 			error = NFSERR_INVAL;
3561 			goto out;
3562 		}
3563 		stp->ls_flags = (bits | NFSLCK_OPEN);
3564 		stp->ls_stateid.seqid++;
3565 		if ((nd->nd_flag & ND_NFSV41) != 0 &&
3566 		    stp->ls_stateid.seqid == 0)
3567 			stp->ls_stateid.seqid = 1;
3568 	}
3569 	NFSUNLOCKSTATE();
3570 
3571 	/*
3572 	 * If the client just confirmed its first open, write a timestamp
3573 	 * to the stable storage file.
3574 	 */
3575 	if (gotstate != 0) {
3576 		nfsrv_writestable(clidp, len, NFSNST_NEWSTATE, p);
3577 		nfsrv_backupstable();
3578 	}
3579 
3580 out:
3581 	free(clidp, M_TEMP);
3582 	NFSEXITCODE2(error, nd);
3583 	return (error);
3584 }
3585 
3586 /*
3587  * Delegation update. Does the purge and return.
3588  */
3589 int
3590 nfsrv_delegupdate(struct nfsrv_descript *nd, nfsquad_t clientid,
3591     nfsv4stateid_t *stateidp, vnode_t vp, int op, struct ucred *cred,
3592     NFSPROC_T *p, int *retwriteaccessp)
3593 {
3594 	struct nfsstate *stp;
3595 	struct nfsclient *clp;
3596 	int error = 0;
3597 	fhandle_t fh;
3598 
3599 	/*
3600 	 * Do a sanity check against the file handle for DelegReturn.
3601 	 */
3602 	if (vp) {
3603 		error = nfsvno_getfh(vp, &fh, p);
3604 		if (error)
3605 			goto out;
3606 	}
3607 	/*
3608 	 * Check for restart conditions (client and server).
3609 	 */
3610 	if (op == NFSV4OP_DELEGRETURN)
3611 		error = nfsrv_checkrestart(clientid, NFSLCK_DELEGRETURN,
3612 			stateidp, 0);
3613 	else
3614 		error = nfsrv_checkrestart(clientid, NFSLCK_DELEGPURGE,
3615 			stateidp, 0);
3616 
3617 	NFSLOCKSTATE();
3618 	/*
3619 	 * Get the open structure via clientid and stateid.
3620 	 */
3621 	if (!error)
3622 	    error = nfsrv_getclient(clientid, CLOPS_RENEW, &clp, NULL,
3623 		(nfsquad_t)((u_quad_t)0), 0, nd, p);
3624 	if (error) {
3625 		if (error == NFSERR_CBPATHDOWN)
3626 			error = 0;
3627 		if (error == NFSERR_STALECLIENTID && op == NFSV4OP_DELEGRETURN)
3628 			error = NFSERR_STALESTATEID;
3629 	}
3630 	if (!error && op == NFSV4OP_DELEGRETURN) {
3631 	    error = nfsrv_getstate(clp, stateidp, NFSLCK_DELEGRETURN, &stp);
3632 	    if (!error && stp->ls_stateid.seqid != stateidp->seqid &&
3633 		((nd->nd_flag & ND_NFSV41) == 0 || stateidp->seqid != 0))
3634 		error = NFSERR_OLDSTATEID;
3635 	}
3636 	/*
3637 	 * NFSERR_EXPIRED means that the state has gone away,
3638 	 * so Delegations have been purged. Just return ok.
3639 	 */
3640 	if (error == NFSERR_EXPIRED && op == NFSV4OP_DELEGPURGE) {
3641 		NFSUNLOCKSTATE();
3642 		error = 0;
3643 		goto out;
3644 	}
3645 	if (error) {
3646 		NFSUNLOCKSTATE();
3647 		goto out;
3648 	}
3649 
3650 	if (op == NFSV4OP_DELEGRETURN) {
3651 		if (NFSBCMP((caddr_t)&fh, (caddr_t)&stp->ls_lfp->lf_fh,
3652 		    sizeof (fhandle_t))) {
3653 			NFSUNLOCKSTATE();
3654 			error = NFSERR_BADSTATEID;
3655 			goto out;
3656 		}
3657 		if (retwriteaccessp != NULL) {
3658 			if ((stp->ls_flags & NFSLCK_DELEGWRITE) != 0)
3659 				*retwriteaccessp = 1;
3660 			else
3661 				*retwriteaccessp = 0;
3662 		}
3663 		nfsrv_freedeleg(stp);
3664 	} else {
3665 		nfsrv_freedeleglist(&clp->lc_olddeleg);
3666 	}
3667 	NFSUNLOCKSTATE();
3668 	error = 0;
3669 
3670 out:
3671 	NFSEXITCODE(error);
3672 	return (error);
3673 }
3674 
3675 /*
3676  * Release lock owner.
3677  */
3678 int
3679 nfsrv_releaselckown(struct nfsstate *new_stp, nfsquad_t clientid,
3680     NFSPROC_T *p)
3681 {
3682 	struct nfsstate *stp, *nstp, *openstp, *ownstp;
3683 	struct nfsclient *clp;
3684 	int error = 0;
3685 
3686 	/*
3687 	 * Check for restart conditions (client and server).
3688 	 */
3689 	error = nfsrv_checkrestart(clientid, new_stp->ls_flags,
3690 	    &new_stp->ls_stateid, 0);
3691 	if (error)
3692 		goto out;
3693 
3694 	NFSLOCKSTATE();
3695 	/*
3696 	 * Get the lock owner by name.
3697 	 */
3698 	error = nfsrv_getclient(clientid, CLOPS_RENEW, &clp, NULL,
3699 	    (nfsquad_t)((u_quad_t)0), 0, NULL, p);
3700 	if (error) {
3701 		NFSUNLOCKSTATE();
3702 		goto out;
3703 	}
3704 	LIST_FOREACH(ownstp, &clp->lc_open, ls_list) {
3705 	    LIST_FOREACH(openstp, &ownstp->ls_open, ls_list) {
3706 		stp = LIST_FIRST(&openstp->ls_open);
3707 		while (stp != LIST_END(&openstp->ls_open)) {
3708 		    nstp = LIST_NEXT(stp, ls_list);
3709 		    /*
3710 		     * If the owner matches, check for locks and
3711 		     * then free or return an error.
3712 		     */
3713 		    if (stp->ls_ownerlen == new_stp->ls_ownerlen &&
3714 			!NFSBCMP(stp->ls_owner, new_stp->ls_owner,
3715 			 stp->ls_ownerlen)){
3716 			if (LIST_EMPTY(&stp->ls_lock)) {
3717 			    nfsrv_freelockowner(stp, NULL, 0, p);
3718 			} else {
3719 			    NFSUNLOCKSTATE();
3720 			    error = NFSERR_LOCKSHELD;
3721 			    goto out;
3722 			}
3723 		    }
3724 		    stp = nstp;
3725 		}
3726 	    }
3727 	}
3728 	NFSUNLOCKSTATE();
3729 
3730 out:
3731 	NFSEXITCODE(error);
3732 	return (error);
3733 }
3734 
3735 /*
3736  * Get the file handle for a lock structure.
3737  */
3738 static int
3739 nfsrv_getlockfh(vnode_t vp, u_short flags, struct nfslockfile *new_lfp,
3740     fhandle_t *nfhp, NFSPROC_T *p)
3741 {
3742 	fhandle_t *fhp = NULL;
3743 	int error;
3744 
3745 	/*
3746 	 * For lock, use the new nfslock structure, otherwise just
3747 	 * a fhandle_t on the stack.
3748 	 */
3749 	if (flags & NFSLCK_OPEN) {
3750 		KASSERT(new_lfp != NULL, ("nfsrv_getlockfh: new_lfp NULL"));
3751 		fhp = &new_lfp->lf_fh;
3752 	} else if (nfhp) {
3753 		fhp = nfhp;
3754 	} else {
3755 		panic("nfsrv_getlockfh");
3756 	}
3757 	error = nfsvno_getfh(vp, fhp, p);
3758 	NFSEXITCODE(error);
3759 	return (error);
3760 }
3761 
3762 /*
3763  * Get an nfs lock structure. Allocate one, as required, and return a
3764  * pointer to it.
3765  * Returns an NFSERR_xxx upon failure or -1 to indicate no current lock.
3766  */
3767 static int
3768 nfsrv_getlockfile(u_short flags, struct nfslockfile **new_lfpp,
3769     struct nfslockfile **lfpp, fhandle_t *nfhp, int lockit)
3770 {
3771 	struct nfslockfile *lfp;
3772 	fhandle_t *fhp = NULL, *tfhp;
3773 	struct nfslockhashhead *hp;
3774 	struct nfslockfile *new_lfp = NULL;
3775 
3776 	/*
3777 	 * For lock, use the new nfslock structure, otherwise just
3778 	 * a fhandle_t on the stack.
3779 	 */
3780 	if (flags & NFSLCK_OPEN) {
3781 		new_lfp = *new_lfpp;
3782 		fhp = &new_lfp->lf_fh;
3783 	} else if (nfhp) {
3784 		fhp = nfhp;
3785 	} else {
3786 		panic("nfsrv_getlockfile");
3787 	}
3788 
3789 	hp = NFSLOCKHASH(fhp);
3790 	LIST_FOREACH(lfp, hp, lf_hash) {
3791 		tfhp = &lfp->lf_fh;
3792 		if (NFSVNO_CMPFH(fhp, tfhp)) {
3793 			if (lockit)
3794 				nfsrv_locklf(lfp);
3795 			*lfpp = lfp;
3796 			return (0);
3797 		}
3798 	}
3799 	if (!(flags & NFSLCK_OPEN))
3800 		return (-1);
3801 
3802 	/*
3803 	 * No match, so chain the new one into the list.
3804 	 */
3805 	LIST_INIT(&new_lfp->lf_open);
3806 	LIST_INIT(&new_lfp->lf_lock);
3807 	LIST_INIT(&new_lfp->lf_deleg);
3808 	LIST_INIT(&new_lfp->lf_locallock);
3809 	LIST_INIT(&new_lfp->lf_rollback);
3810 	new_lfp->lf_locallock_lck.nfslock_usecnt = 0;
3811 	new_lfp->lf_locallock_lck.nfslock_lock = 0;
3812 	new_lfp->lf_usecount = 0;
3813 	LIST_INSERT_HEAD(hp, new_lfp, lf_hash);
3814 	*lfpp = new_lfp;
3815 	*new_lfpp = NULL;
3816 	return (0);
3817 }
3818 
3819 /*
3820  * This function adds a nfslock lock structure to the list for the associated
3821  * nfsstate and nfslockfile structures. It will be inserted after the
3822  * entry pointed at by insert_lop.
3823  */
3824 static void
3825 nfsrv_insertlock(struct nfslock *new_lop, struct nfslock *insert_lop,
3826     struct nfsstate *stp, struct nfslockfile *lfp)
3827 {
3828 	struct nfslock *lop, *nlop;
3829 
3830 	new_lop->lo_stp = stp;
3831 	new_lop->lo_lfp = lfp;
3832 
3833 	if (stp != NULL) {
3834 		/* Insert in increasing lo_first order */
3835 		lop = LIST_FIRST(&lfp->lf_lock);
3836 		if (lop == LIST_END(&lfp->lf_lock) ||
3837 		    new_lop->lo_first <= lop->lo_first) {
3838 			LIST_INSERT_HEAD(&lfp->lf_lock, new_lop, lo_lckfile);
3839 		} else {
3840 			nlop = LIST_NEXT(lop, lo_lckfile);
3841 			while (nlop != LIST_END(&lfp->lf_lock) &&
3842 			       nlop->lo_first < new_lop->lo_first) {
3843 				lop = nlop;
3844 				nlop = LIST_NEXT(lop, lo_lckfile);
3845 			}
3846 			LIST_INSERT_AFTER(lop, new_lop, lo_lckfile);
3847 		}
3848 	} else {
3849 		new_lop->lo_lckfile.le_prev = NULL;	/* list not used */
3850 	}
3851 
3852 	/*
3853 	 * Insert after insert_lop, which is overloaded as stp or lfp for
3854 	 * an empty list.
3855 	 */
3856 	if (stp == NULL && (struct nfslockfile *)insert_lop == lfp)
3857 		LIST_INSERT_HEAD(&lfp->lf_locallock, new_lop, lo_lckowner);
3858 	else if ((struct nfsstate *)insert_lop == stp)
3859 		LIST_INSERT_HEAD(&stp->ls_lock, new_lop, lo_lckowner);
3860 	else
3861 		LIST_INSERT_AFTER(insert_lop, new_lop, lo_lckowner);
3862 	if (stp != NULL) {
3863 		NFSD_VNET(nfsstatsv1_p)->srvlocks++;
3864 		nfsrv_openpluslock++;
3865 	}
3866 }
3867 
3868 /*
3869  * This function updates the locking for a lock owner and given file. It
3870  * maintains a list of lock ranges ordered on increasing file offset that
3871  * are NFSLCK_READ or NFSLCK_WRITE and non-overlapping (aka POSIX style).
3872  * It always adds new_lop to the list and sometimes uses the one pointed
3873  * at by other_lopp.
3874  */
3875 static void
3876 nfsrv_updatelock(struct nfsstate *stp, struct nfslock **new_lopp,
3877     struct nfslock **other_lopp, struct nfslockfile *lfp)
3878 {
3879 	struct nfslock *new_lop = *new_lopp;
3880 	struct nfslock *lop, *tlop, *ilop;
3881 	struct nfslock *other_lop = *other_lopp;
3882 	int unlock = 0, myfile = 0;
3883 	u_int64_t tmp;
3884 
3885 	/*
3886 	 * Work down the list until the lock is merged.
3887 	 */
3888 	if (new_lop->lo_flags & NFSLCK_UNLOCK)
3889 		unlock = 1;
3890 	if (stp != NULL) {
3891 		ilop = (struct nfslock *)stp;
3892 		lop = LIST_FIRST(&stp->ls_lock);
3893 	} else {
3894 		ilop = (struct nfslock *)lfp;
3895 		lop = LIST_FIRST(&lfp->lf_locallock);
3896 	}
3897 	while (lop != NULL) {
3898 	    /*
3899 	     * Only check locks for this file that aren't before the start of
3900 	     * new lock's range.
3901 	     */
3902 	    if (lop->lo_lfp == lfp) {
3903 	      myfile = 1;
3904 	      if (lop->lo_end >= new_lop->lo_first) {
3905 		if (new_lop->lo_end < lop->lo_first) {
3906 			/*
3907 			 * If the new lock ends before the start of the
3908 			 * current lock's range, no merge, just insert
3909 			 * the new lock.
3910 			 */
3911 			break;
3912 		}
3913 		if (new_lop->lo_flags == lop->lo_flags ||
3914 		    (new_lop->lo_first <= lop->lo_first &&
3915 		     new_lop->lo_end >= lop->lo_end)) {
3916 			/*
3917 			 * This lock can be absorbed by the new lock/unlock.
3918 			 * This happens when it covers the entire range
3919 			 * of the old lock or is contiguous
3920 			 * with the old lock and is of the same type or an
3921 			 * unlock.
3922 			 */
3923 			if (lop->lo_first < new_lop->lo_first)
3924 				new_lop->lo_first = lop->lo_first;
3925 			if (lop->lo_end > new_lop->lo_end)
3926 				new_lop->lo_end = lop->lo_end;
3927 			tlop = lop;
3928 			lop = LIST_NEXT(lop, lo_lckowner);
3929 			nfsrv_freenfslock(tlop);
3930 			continue;
3931 		}
3932 
3933 		/*
3934 		 * All these cases are for contiguous locks that are not the
3935 		 * same type, so they can't be merged.
3936 		 */
3937 		if (new_lop->lo_first <= lop->lo_first) {
3938 			/*
3939 			 * This case is where the new lock overlaps with the
3940 			 * first part of the old lock. Move the start of the
3941 			 * old lock to just past the end of the new lock. The
3942 			 * new lock will be inserted in front of the old, since
3943 			 * ilop hasn't been updated. (We are done now.)
3944 			 */
3945 			lop->lo_first = new_lop->lo_end;
3946 			break;
3947 		}
3948 		if (new_lop->lo_end >= lop->lo_end) {
3949 			/*
3950 			 * This case is where the new lock overlaps with the
3951 			 * end of the old lock's range. Move the old lock's
3952 			 * end to just before the new lock's first and insert
3953 			 * the new lock after the old lock.
3954 			 * Might not be done yet, since the new lock could
3955 			 * overlap further locks with higher ranges.
3956 			 */
3957 			lop->lo_end = new_lop->lo_first;
3958 			ilop = lop;
3959 			lop = LIST_NEXT(lop, lo_lckowner);
3960 			continue;
3961 		}
3962 		/*
3963 		 * The final case is where the new lock's range is in the
3964 		 * middle of the current lock's and splits the current lock
3965 		 * up. Use *other_lopp to handle the second part of the
3966 		 * split old lock range. (We are done now.)
3967 		 * For unlock, we use new_lop as other_lop and tmp, since
3968 		 * other_lop and new_lop are the same for this case.
3969 		 * We noted the unlock case above, so we don't need
3970 		 * new_lop->lo_flags any longer.
3971 		 */
3972 		tmp = new_lop->lo_first;
3973 		if (other_lop == NULL) {
3974 			if (!unlock)
3975 				panic("nfsd srv update unlock");
3976 			other_lop = new_lop;
3977 			*new_lopp = NULL;
3978 		}
3979 		other_lop->lo_first = new_lop->lo_end;
3980 		other_lop->lo_end = lop->lo_end;
3981 		other_lop->lo_flags = lop->lo_flags;
3982 		other_lop->lo_stp = stp;
3983 		other_lop->lo_lfp = lfp;
3984 		lop->lo_end = tmp;
3985 		nfsrv_insertlock(other_lop, lop, stp, lfp);
3986 		*other_lopp = NULL;
3987 		ilop = lop;
3988 		break;
3989 	      }
3990 	    }
3991 	    ilop = lop;
3992 	    lop = LIST_NEXT(lop, lo_lckowner);
3993 	    if (myfile && (lop == NULL || lop->lo_lfp != lfp))
3994 		break;
3995 	}
3996 
3997 	/*
3998 	 * Insert the new lock in the list at the appropriate place.
3999 	 */
4000 	if (!unlock) {
4001 		nfsrv_insertlock(new_lop, ilop, stp, lfp);
4002 		*new_lopp = NULL;
4003 	}
4004 }
4005 
4006 /*
4007  * This function handles sequencing of locks, etc.
4008  * It returns an error that indicates what the caller should do.
4009  */
4010 static int
4011 nfsrv_checkseqid(struct nfsrv_descript *nd, u_int32_t seqid,
4012     struct nfsstate *stp, struct nfsrvcache *op)
4013 {
4014 	int error = 0;
4015 
4016 	if ((nd->nd_flag & ND_NFSV41) != 0)
4017 		/* NFSv4.1 ignores the open_seqid and lock_seqid. */
4018 		goto out;
4019 	if (op != nd->nd_rp)
4020 		panic("nfsrvstate checkseqid");
4021 	if (!(op->rc_flag & RC_INPROG))
4022 		panic("nfsrvstate not inprog");
4023 	if (stp->ls_op && stp->ls_op->rc_refcnt <= 0) {
4024 		printf("refcnt=%d\n", stp->ls_op->rc_refcnt);
4025 		panic("nfsrvstate op refcnt");
4026 	}
4027 
4028 	/* If ND_ERELOOKUP is set, the seqid has already been handled. */
4029 	if ((nd->nd_flag & ND_ERELOOKUP) != 0)
4030 		goto out;
4031 
4032 	if ((stp->ls_seq + 1) == seqid) {
4033 		if (stp->ls_op)
4034 			nfsrvd_derefcache(stp->ls_op);
4035 		stp->ls_op = op;
4036 		nfsrvd_refcache(op);
4037 		stp->ls_seq = seqid;
4038 		goto out;
4039 	} else if (stp->ls_seq == seqid && stp->ls_op &&
4040 		op->rc_xid == stp->ls_op->rc_xid &&
4041 		op->rc_refcnt == 0 &&
4042 		op->rc_reqlen == stp->ls_op->rc_reqlen &&
4043 		op->rc_cksum == stp->ls_op->rc_cksum) {
4044 		if (stp->ls_op->rc_flag & RC_INPROG) {
4045 			error = NFSERR_DONTREPLY;
4046 			goto out;
4047 		}
4048 		nd->nd_rp = stp->ls_op;
4049 		nd->nd_rp->rc_flag |= RC_INPROG;
4050 		nfsrvd_delcache(op);
4051 		error = NFSERR_REPLYFROMCACHE;
4052 		goto out;
4053 	}
4054 	error = NFSERR_BADSEQID;
4055 
4056 out:
4057 	NFSEXITCODE2(error, nd);
4058 	return (error);
4059 }
4060 
4061 /*
4062  * Get the client ip address for callbacks. If the strings can't be parsed,
4063  * just set lc_program to 0 to indicate no callbacks are possible.
4064  * (For cases where the address can't be parsed or is 0.0.0.0.0.0, set
4065  *  the address to the client's transport address. This won't be used
4066  *  for callbacks, but can be printed out by nfsstats for info.)
4067  * Return error if the xdr can't be parsed, 0 otherwise.
4068  */
4069 int
4070 nfsrv_getclientipaddr(struct nfsrv_descript *nd, struct nfsclient *clp)
4071 {
4072 	u_int32_t *tl;
4073 	u_char *cp, *cp2;
4074 	int i, j, maxalen = 0, minalen = 0;
4075 	sa_family_t af;
4076 #ifdef INET
4077 	struct sockaddr_in *rin = NULL, *sin;
4078 #endif
4079 #ifdef INET6
4080 	struct sockaddr_in6 *rin6 = NULL, *sin6;
4081 #endif
4082 	u_char *addr;
4083 	int error = 0, cantparse = 0;
4084 	union {
4085 		in_addr_t ival;
4086 		u_char cval[4];
4087 	} ip;
4088 	union {
4089 		in_port_t sval;
4090 		u_char cval[2];
4091 	} port;
4092 
4093 	/* 8 is the maximum length of the port# string. */
4094 	addr = malloc(INET6_ADDRSTRLEN + 8, M_TEMP, M_WAITOK);
4095 	clp->lc_req.nr_client = NULL;
4096 	clp->lc_req.nr_lock = 0;
4097 	af = AF_UNSPEC;
4098 	NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
4099 	i = fxdr_unsigned(int, *tl);
4100 	if (i >= 3 && i <= 4) {
4101 		error = nfsrv_mtostr(nd, addr, i);
4102 		if (error)
4103 			goto nfsmout;
4104 #ifdef INET
4105 		if (!strcmp(addr, "tcp")) {
4106 			clp->lc_flags |= LCL_TCPCALLBACK;
4107 			clp->lc_req.nr_sotype = SOCK_STREAM;
4108 			clp->lc_req.nr_soproto = IPPROTO_TCP;
4109 			af = AF_INET;
4110 		} else if (!strcmp(addr, "udp")) {
4111 			clp->lc_req.nr_sotype = SOCK_DGRAM;
4112 			clp->lc_req.nr_soproto = IPPROTO_UDP;
4113 			af = AF_INET;
4114 		}
4115 #endif
4116 #ifdef INET6
4117 		if (af == AF_UNSPEC) {
4118 			if (!strcmp(addr, "tcp6")) {
4119 				clp->lc_flags |= LCL_TCPCALLBACK;
4120 				clp->lc_req.nr_sotype = SOCK_STREAM;
4121 				clp->lc_req.nr_soproto = IPPROTO_TCP;
4122 				af = AF_INET6;
4123 			} else if (!strcmp(addr, "udp6")) {
4124 				clp->lc_req.nr_sotype = SOCK_DGRAM;
4125 				clp->lc_req.nr_soproto = IPPROTO_UDP;
4126 				af = AF_INET6;
4127 			}
4128 		}
4129 #endif
4130 		if (af == AF_UNSPEC) {
4131 			cantparse = 1;
4132 		}
4133 	} else {
4134 		cantparse = 1;
4135 		if (i > 0) {
4136 			error = nfsm_advance(nd, NFSM_RNDUP(i), -1);
4137 			if (error)
4138 				goto nfsmout;
4139 		}
4140 	}
4141 	/*
4142 	 * The caller has allocated clp->lc_req.nr_nam to be large enough
4143 	 * for either AF_INET or AF_INET6 and zeroed out the contents.
4144 	 * maxalen is set to the maximum length of the host IP address string
4145 	 * plus 8 for the maximum length of the port#.
4146 	 * minalen is set to the minimum length of the host IP address string
4147 	 * plus 4 for the minimum length of the port#.
4148 	 * These lengths do not include NULL termination,
4149 	 * so INET[6]_ADDRSTRLEN - 1 is used in the calculations.
4150 	 */
4151 	switch (af) {
4152 #ifdef INET
4153 	case AF_INET:
4154 		rin = (struct sockaddr_in *)clp->lc_req.nr_nam;
4155 		rin->sin_family = AF_INET;
4156 		rin->sin_len = sizeof(struct sockaddr_in);
4157 		maxalen = INET_ADDRSTRLEN - 1 + 8;
4158 		minalen = 7 + 4;
4159 		break;
4160 #endif
4161 #ifdef INET6
4162 	case AF_INET6:
4163 		rin6 = (struct sockaddr_in6 *)clp->lc_req.nr_nam;
4164 		rin6->sin6_family = AF_INET6;
4165 		rin6->sin6_len = sizeof(struct sockaddr_in6);
4166 		maxalen = INET6_ADDRSTRLEN - 1 + 8;
4167 		minalen = 3 + 4;
4168 		break;
4169 #endif
4170 	}
4171 	NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
4172 	i = fxdr_unsigned(int, *tl);
4173 	if (i < 0) {
4174 		error = NFSERR_BADXDR;
4175 		goto nfsmout;
4176 	} else if (i == 0) {
4177 		cantparse = 1;
4178 	} else if (!cantparse && i <= maxalen && i >= minalen) {
4179 		error = nfsrv_mtostr(nd, addr, i);
4180 		if (error)
4181 			goto nfsmout;
4182 
4183 		/*
4184 		 * Parse out the address fields. We expect 6 decimal numbers
4185 		 * separated by '.'s for AF_INET and two decimal numbers
4186 		 * preceeded by '.'s for AF_INET6.
4187 		 */
4188 		cp = NULL;
4189 		switch (af) {
4190 #ifdef INET6
4191 		/*
4192 		 * For AF_INET6, first parse the host address.
4193 		 */
4194 		case AF_INET6:
4195 			cp = strchr(addr, '.');
4196 			if (cp != NULL) {
4197 				*cp++ = '\0';
4198 				if (inet_pton(af, addr, &rin6->sin6_addr) == 1)
4199 					i = 4;
4200 				else {
4201 					cp = NULL;
4202 					cantparse = 1;
4203 				}
4204 			}
4205 			break;
4206 #endif
4207 #ifdef INET
4208 		case AF_INET:
4209 			cp = addr;
4210 			i = 0;
4211 			break;
4212 #endif
4213 		}
4214 		while (cp != NULL && *cp && i < 6) {
4215 			cp2 = cp;
4216 			while (*cp2 && *cp2 != '.')
4217 				cp2++;
4218 			if (*cp2)
4219 				*cp2++ = '\0';
4220 			else if (i != 5) {
4221 				cantparse = 1;
4222 				break;
4223 			}
4224 			j = nfsrv_getipnumber(cp);
4225 			if (j >= 0) {
4226 				if (i < 4)
4227 					ip.cval[3 - i] = j;
4228 				else
4229 					port.cval[5 - i] = j;
4230 			} else {
4231 				cantparse = 1;
4232 				break;
4233 			}
4234 			cp = cp2;
4235 			i++;
4236 		}
4237 		if (!cantparse) {
4238 			/*
4239 			 * The host address INADDR_ANY is (mis)used to indicate
4240 			 * "there is no valid callback address".
4241 			 */
4242 			switch (af) {
4243 #ifdef INET6
4244 			case AF_INET6:
4245 				if (!IN6_ARE_ADDR_EQUAL(&rin6->sin6_addr,
4246 				    &in6addr_any))
4247 					rin6->sin6_port = htons(port.sval);
4248 				else
4249 					cantparse = 1;
4250 				break;
4251 #endif
4252 #ifdef INET
4253 			case AF_INET:
4254 				if (ip.ival != INADDR_ANY) {
4255 					rin->sin_addr.s_addr = htonl(ip.ival);
4256 					rin->sin_port = htons(port.sval);
4257 				} else {
4258 					cantparse = 1;
4259 				}
4260 				break;
4261 #endif
4262 			}
4263 		}
4264 	} else {
4265 		cantparse = 1;
4266 		if (i > 0) {
4267 			error = nfsm_advance(nd, NFSM_RNDUP(i), -1);
4268 			if (error)
4269 				goto nfsmout;
4270 		}
4271 	}
4272 	if (cantparse) {
4273 		switch (nd->nd_nam->sa_family) {
4274 #ifdef INET
4275 		case AF_INET:
4276 			sin = (struct sockaddr_in *)nd->nd_nam;
4277 			rin = (struct sockaddr_in *)clp->lc_req.nr_nam;
4278 			rin->sin_family = AF_INET;
4279 			rin->sin_len = sizeof(struct sockaddr_in);
4280 			rin->sin_addr.s_addr = sin->sin_addr.s_addr;
4281 			rin->sin_port = 0x0;
4282 			break;
4283 #endif
4284 #ifdef INET6
4285 		case AF_INET6:
4286 			sin6 = (struct sockaddr_in6 *)nd->nd_nam;
4287 			rin6 = (struct sockaddr_in6 *)clp->lc_req.nr_nam;
4288 			rin6->sin6_family = AF_INET6;
4289 			rin6->sin6_len = sizeof(struct sockaddr_in6);
4290 			rin6->sin6_addr = sin6->sin6_addr;
4291 			rin6->sin6_port = 0x0;
4292 			break;
4293 #endif
4294 		}
4295 		clp->lc_program = 0;
4296 	}
4297 nfsmout:
4298 	free(addr, M_TEMP);
4299 	NFSEXITCODE2(error, nd);
4300 	return (error);
4301 }
4302 
4303 /*
4304  * Turn a string of up to three decimal digits into a number. Return -1 upon
4305  * error.
4306  */
4307 static int
4308 nfsrv_getipnumber(u_char *cp)
4309 {
4310 	int i = 0, j = 0;
4311 
4312 	while (*cp) {
4313 		if (j > 2 || *cp < '0' || *cp > '9')
4314 			return (-1);
4315 		i *= 10;
4316 		i += (*cp - '0');
4317 		cp++;
4318 		j++;
4319 	}
4320 	if (i < 256)
4321 		return (i);
4322 	return (-1);
4323 }
4324 
4325 /*
4326  * This function checks for restart conditions.
4327  */
4328 static int
4329 nfsrv_checkrestart(nfsquad_t clientid, u_int32_t flags,
4330     nfsv4stateid_t *stateidp, int specialid)
4331 {
4332 	int ret = 0;
4333 
4334 	/*
4335 	 * First check for a server restart. Open, LockT, ReleaseLockOwner
4336 	 * and DelegPurge have a clientid, the rest a stateid.
4337 	 */
4338 	if (flags &
4339 	    (NFSLCK_OPEN | NFSLCK_TEST | NFSLCK_RELEASE | NFSLCK_DELEGPURGE)) {
4340 		if (clientid.lval[0] != NFSD_VNET(nfsrvboottime)) {
4341 			ret = NFSERR_STALECLIENTID;
4342 			goto out;
4343 		}
4344 	} else if (stateidp->other[0] != NFSD_VNET(nfsrvboottime) &&
4345 		specialid == 0) {
4346 		ret = NFSERR_STALESTATEID;
4347 		goto out;
4348 	}
4349 
4350 	/*
4351 	 * Read, Write, Setattr and LockT can return NFSERR_GRACE and do
4352 	 * not use a lock/open owner seqid#, so the check can be done now.
4353 	 * (The others will be checked, as required, later.)
4354 	 */
4355 	if (!(flags & (NFSLCK_CHECK | NFSLCK_TEST)))
4356 		goto out;
4357 
4358 	NFSLOCKSTATE();
4359 	ret = nfsrv_checkgrace(NULL, NULL, flags);
4360 	NFSUNLOCKSTATE();
4361 
4362 out:
4363 	NFSEXITCODE(ret);
4364 	return (ret);
4365 }
4366 
4367 /*
4368  * Check for grace.
4369  */
4370 static int
4371 nfsrv_checkgrace(struct nfsrv_descript *nd, struct nfsclient *clp,
4372     u_int32_t flags)
4373 {
4374 	int error = 0, notreclaimed;
4375 	struct nfsrv_stable *sp;
4376 
4377 	if ((NFSD_VNET(nfsrv_stablefirst).nsf_flags & (NFSNSF_UPDATEDONE |
4378 	     NFSNSF_GRACEOVER)) == 0) {
4379 		/*
4380 		 * First, check to see if all of the clients have done a
4381 		 * ReclaimComplete.  If so, grace can end now.
4382 		 */
4383 		notreclaimed = 0;
4384 		LIST_FOREACH(sp, &NFSD_VNET(nfsrv_stablefirst).nsf_head,
4385 		    nst_list) {
4386 			if ((sp->nst_flag & NFSNST_RECLAIMED) == 0) {
4387 				notreclaimed = 1;
4388 				break;
4389 			}
4390 		}
4391 		if (notreclaimed == 0)
4392 			NFSD_VNET(nfsrv_stablefirst).nsf_flags |=
4393 			    (NFSNSF_GRACEOVER | NFSNSF_NEEDLOCK);
4394 	}
4395 
4396 	if ((NFSD_VNET(nfsrv_stablefirst).nsf_flags & NFSNSF_GRACEOVER) != 0) {
4397 		if (flags & NFSLCK_RECLAIM) {
4398 			error = NFSERR_NOGRACE;
4399 			goto out;
4400 		}
4401 	} else {
4402 		if (!(flags & NFSLCK_RECLAIM)) {
4403 			error = NFSERR_GRACE;
4404 			goto out;
4405 		}
4406 		if (nd != NULL && clp != NULL &&
4407 		    (nd->nd_flag & ND_NFSV41) != 0 &&
4408 		    (clp->lc_flags & LCL_RECLAIMCOMPLETE) != 0) {
4409 			error = NFSERR_NOGRACE;
4410 			goto out;
4411 		}
4412 
4413 		/*
4414 		 * If grace is almost over and we are still getting Reclaims,
4415 		 * extend grace a bit.
4416 		 */
4417 		if ((NFSD_MONOSEC + NFSRV_LEASEDELTA) >
4418 		    NFSD_VNET(nfsrv_stablefirst).nsf_eograce)
4419 			NFSD_VNET(nfsrv_stablefirst).nsf_eograce =
4420 				NFSD_MONOSEC + NFSRV_LEASEDELTA;
4421 	}
4422 
4423 out:
4424 	NFSEXITCODE(error);
4425 	return (error);
4426 }
4427 
4428 /*
4429  * Do a server callback.
4430  * The "trunc" argument is slightly overloaded and refers to different
4431  * boolean arguments for CBRECALL and CBLAYOUTRECALL.
4432  */
4433 static int
4434 nfsrv_docallback(struct nfsclient *clp, int procnum, nfsv4stateid_t *stateidp,
4435     int trunc, fhandle_t *fhp, struct nfsvattr *nap, nfsattrbit_t *attrbitp,
4436     int laytype, NFSPROC_T *p)
4437 {
4438 	struct mbuf *m;
4439 	u_int32_t *tl;
4440 	struct nfsrv_descript *nd;
4441 	struct ucred *cred;
4442 	int error = 0, slotpos;
4443 	u_int32_t callback;
4444 	struct nfsdsession *sep = NULL;
4445 	uint64_t tval;
4446 	bool dotls;
4447 
4448 	nd = malloc(sizeof(*nd), M_TEMP, M_WAITOK | M_ZERO);
4449 	cred = newnfs_getcred();
4450 	NFSLOCKSTATE();	/* mostly for lc_cbref++ */
4451 	if (clp->lc_flags & LCL_NEEDSCONFIRM) {
4452 		NFSUNLOCKSTATE();
4453 		panic("docallb");
4454 	}
4455 	clp->lc_cbref++;
4456 
4457 	/*
4458 	 * Fill the callback program# and version into the request
4459 	 * structure for newnfs_connect() to use.
4460 	 */
4461 	clp->lc_req.nr_prog = clp->lc_program;
4462 #ifdef notnow
4463 	if ((clp->lc_flags & LCL_NFSV41) != 0)
4464 		clp->lc_req.nr_vers = NFSV41_CBVERS;
4465 	else
4466 #endif
4467 		clp->lc_req.nr_vers = NFSV4_CBVERS;
4468 
4469 	/*
4470 	 * First, fill in some of the fields of nd and cr.
4471 	 */
4472 	nd->nd_flag = ND_NFSV4;
4473 	if (clp->lc_flags & LCL_GSS)
4474 		nd->nd_flag |= ND_KERBV;
4475 	if ((clp->lc_flags & LCL_NFSV41) != 0)
4476 		nd->nd_flag |= ND_NFSV41;
4477 	if ((clp->lc_flags & LCL_NFSV42) != 0)
4478 		nd->nd_flag |= ND_NFSV42;
4479 	nd->nd_repstat = 0;
4480 	cred->cr_uid = clp->lc_uid;
4481 	cred->cr_gid = clp->lc_gid;
4482 	callback = clp->lc_callback;
4483 	NFSUNLOCKSTATE();
4484 	cred->cr_ngroups = 1;
4485 
4486 	/*
4487 	 * Get the first mbuf for the request.
4488 	 */
4489 	MGET(m, M_WAITOK, MT_DATA);
4490 	m->m_len = 0;
4491 	nd->nd_mreq = nd->nd_mb = m;
4492 	nd->nd_bpos = mtod(m, caddr_t);
4493 
4494 	/*
4495 	 * and build the callback request.
4496 	 */
4497 	if (procnum == NFSV4OP_CBGETATTR) {
4498 		nd->nd_procnum = NFSV4PROC_CBCOMPOUND;
4499 		error = nfsrv_cbcallargs(nd, clp, callback, NFSV4OP_CBGETATTR,
4500 		    "CB Getattr", &sep, &slotpos);
4501 		if (error != 0) {
4502 			m_freem(nd->nd_mreq);
4503 			goto errout;
4504 		}
4505 		(void)nfsm_fhtom(NULL, nd, (u_int8_t *)fhp, NFSX_MYFH, 0);
4506 		(void)nfsrv_putattrbit(nd, attrbitp);
4507 	} else if (procnum == NFSV4OP_CBRECALL) {
4508 		nd->nd_procnum = NFSV4PROC_CBCOMPOUND;
4509 		error = nfsrv_cbcallargs(nd, clp, callback, NFSV4OP_CBRECALL,
4510 		    "CB Recall", &sep, &slotpos);
4511 		if (error != 0) {
4512 			m_freem(nd->nd_mreq);
4513 			goto errout;
4514 		}
4515 		NFSM_BUILD(tl, u_int32_t *, NFSX_UNSIGNED + NFSX_STATEID);
4516 		*tl++ = txdr_unsigned(stateidp->seqid);
4517 		NFSBCOPY((caddr_t)stateidp->other, (caddr_t)tl,
4518 		    NFSX_STATEIDOTHER);
4519 		tl += (NFSX_STATEIDOTHER / NFSX_UNSIGNED);
4520 		if (trunc)
4521 			*tl = newnfs_true;
4522 		else
4523 			*tl = newnfs_false;
4524 		(void)nfsm_fhtom(NULL, nd, (u_int8_t *)fhp, NFSX_MYFH, 0);
4525 	} else if (procnum == NFSV4OP_CBLAYOUTRECALL) {
4526 		NFSD_DEBUG(4, "docallback layout recall\n");
4527 		nd->nd_procnum = NFSV4PROC_CBCOMPOUND;
4528 		error = nfsrv_cbcallargs(nd, clp, callback,
4529 		    NFSV4OP_CBLAYOUTRECALL, "CB Reclayout", &sep, &slotpos);
4530 		NFSD_DEBUG(4, "aft cbcallargs=%d\n", error);
4531 		if (error != 0) {
4532 			m_freem(nd->nd_mreq);
4533 			goto errout;
4534 		}
4535 		NFSM_BUILD(tl, u_int32_t *, 4 * NFSX_UNSIGNED);
4536 		*tl++ = txdr_unsigned(laytype);
4537 		*tl++ = txdr_unsigned(NFSLAYOUTIOMODE_ANY);
4538 		if (trunc)
4539 			*tl++ = newnfs_true;
4540 		else
4541 			*tl++ = newnfs_false;
4542 		*tl = txdr_unsigned(NFSV4LAYOUTRET_FILE);
4543 		(void)nfsm_fhtom(NULL, nd, (uint8_t *)fhp, NFSX_MYFH, 0);
4544 		NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_HYPER + NFSX_STATEID);
4545 		tval = 0;
4546 		txdr_hyper(tval, tl); tl += 2;
4547 		tval = UINT64_MAX;
4548 		txdr_hyper(tval, tl); tl += 2;
4549 		*tl++ = txdr_unsigned(stateidp->seqid);
4550 		NFSBCOPY(stateidp->other, tl, NFSX_STATEIDOTHER);
4551 		tl += (NFSX_STATEIDOTHER / NFSX_UNSIGNED);
4552 		NFSD_DEBUG(4, "aft args\n");
4553 	} else if (procnum == NFSV4PROC_CBNULL) {
4554 		nd->nd_procnum = NFSV4PROC_CBNULL;
4555 		if ((clp->lc_flags & LCL_NFSV41) != 0) {
4556 			error = nfsv4_getcbsession(clp, &sep);
4557 			if (error != 0) {
4558 				m_freem(nd->nd_mreq);
4559 				goto errout;
4560 			}
4561 		}
4562 	} else {
4563 		error = NFSERR_SERVERFAULT;
4564 		m_freem(nd->nd_mreq);
4565 		goto errout;
4566 	}
4567 
4568 	/*
4569 	 * Call newnfs_connect(), as required, and then newnfs_request().
4570 	 */
4571 	dotls = false;
4572 	if ((clp->lc_flags & LCL_TLSCB) != 0)
4573 		dotls = true;
4574 	(void) newnfs_sndlock(&clp->lc_req.nr_lock);
4575 	if (clp->lc_req.nr_client == NULL) {
4576 		if ((clp->lc_flags & LCL_NFSV41) != 0) {
4577 			error = ECONNREFUSED;
4578 			if (procnum != NFSV4PROC_CBNULL)
4579 				nfsv4_freeslot(&sep->sess_cbsess, slotpos,
4580 				    true);
4581 			nfsrv_freesession(NULL, sep, NULL, false, NULL);
4582 		} else if (nd->nd_procnum == NFSV4PROC_CBNULL)
4583 			error = newnfs_connect(NULL, &clp->lc_req, cred,
4584 			    NULL, 1, dotls, &clp->lc_req.nr_client);
4585 		else
4586 			error = newnfs_connect(NULL, &clp->lc_req, cred,
4587 			    NULL, 3, dotls, &clp->lc_req.nr_client);
4588 	}
4589 	newnfs_sndunlock(&clp->lc_req.nr_lock);
4590 	NFSD_DEBUG(4, "aft sndunlock=%d\n", error);
4591 	if (!error) {
4592 		if ((nd->nd_flag & ND_NFSV41) != 0) {
4593 			KASSERT(sep != NULL, ("sep NULL"));
4594 			if (sep->sess_cbsess.nfsess_xprt != NULL)
4595 				error = newnfs_request(nd, NULL, clp,
4596 				    &clp->lc_req, NULL, NULL, cred,
4597 				    clp->lc_program, clp->lc_req.nr_vers, NULL,
4598 				    1, NULL, &sep->sess_cbsess);
4599 			else {
4600 				/*
4601 				 * This should probably never occur, but if a
4602 				 * client somehow does an RPC without a
4603 				 * SequenceID Op that causes a callback just
4604 				 * after the nfsd threads have been terminated
4605 				 * and restarted we could conceivably get here
4606 				 * without a backchannel xprt.
4607 				 */
4608 				printf("nfsrv_docallback: no xprt\n");
4609 				error = ECONNREFUSED;
4610 			}
4611 			NFSD_DEBUG(4, "aft newnfs_request=%d\n", error);
4612 			if (error != 0 && procnum != NFSV4PROC_CBNULL) {
4613 				/*
4614 				 * It is likely that the callback was never
4615 				 * processed by the client and, as such,
4616 				 * the sequence# for the session slot needs
4617 				 * to be backed up by one to avoid a
4618 				 * NFSERR_SEQMISORDERED error reply.
4619 				 * For the unlikely case where the callback
4620 				 * was processed by the client, this will
4621 				 * make the next callback on the slot
4622 				 * appear to be a retry.
4623 				 * Since callbacks never specify that the
4624 				 * reply be cached, this "apparent retry"
4625 				 * should not be a problem.
4626 				 */
4627 				nfsv4_freeslot(&sep->sess_cbsess, slotpos,
4628 				    true);
4629 			}
4630 			nfsrv_freesession(NULL, sep, NULL, false, NULL);
4631 		} else
4632 			error = newnfs_request(nd, NULL, clp, &clp->lc_req,
4633 			    NULL, NULL, cred, clp->lc_program,
4634 			    clp->lc_req.nr_vers, NULL, 1, NULL, NULL);
4635 	}
4636 errout:
4637 	NFSFREECRED(cred);
4638 
4639 	/*
4640 	 * If error is set here, the Callback path isn't working
4641 	 * properly, so twiddle the appropriate LCL_ flags.
4642 	 * (nd_repstat != 0 indicates the Callback path is working,
4643 	 *  but the callback failed on the client.)
4644 	 */
4645 	if (error) {
4646 		/*
4647 		 * Mark the callback pathway down, which disabled issuing
4648 		 * of delegations and gets Renew to return NFSERR_CBPATHDOWN.
4649 		 */
4650 		NFSLOCKSTATE();
4651 		clp->lc_flags |= LCL_CBDOWN;
4652 		NFSUNLOCKSTATE();
4653 	} else {
4654 		/*
4655 		 * Callback worked. If the callback path was down, disable
4656 		 * callbacks, so no more delegations will be issued. (This
4657 		 * is done on the assumption that the callback pathway is
4658 		 * flakey.)
4659 		 */
4660 		NFSLOCKSTATE();
4661 		if (clp->lc_flags & LCL_CBDOWN)
4662 			clp->lc_flags &= ~(LCL_CBDOWN | LCL_CALLBACKSON);
4663 		NFSUNLOCKSTATE();
4664 		if (nd->nd_repstat) {
4665 			error = nd->nd_repstat;
4666 			NFSD_DEBUG(1, "nfsrv_docallback op=%d err=%d\n",
4667 			    procnum, error);
4668 		} else if (error == 0 && procnum == NFSV4OP_CBGETATTR)
4669 			error = nfsv4_loadattr(nd, NULL, nap, NULL, NULL, 0,
4670 			    NULL, NULL, NULL, NULL, NULL, 0, NULL, NULL, NULL,
4671 			    p, NULL);
4672 		m_freem(nd->nd_mrep);
4673 	}
4674 	NFSLOCKSTATE();
4675 	clp->lc_cbref--;
4676 	if ((clp->lc_flags & LCL_WAKEUPWANTED) && clp->lc_cbref == 0) {
4677 		clp->lc_flags &= ~LCL_WAKEUPWANTED;
4678 		wakeup(clp);
4679 	}
4680 	NFSUNLOCKSTATE();
4681 
4682 	free(nd, M_TEMP);
4683 	NFSEXITCODE(error);
4684 	return (error);
4685 }
4686 
4687 /*
4688  * Set up the compound RPC for the callback.
4689  */
4690 static int
4691 nfsrv_cbcallargs(struct nfsrv_descript *nd, struct nfsclient *clp,
4692     uint32_t callback, int op, const char *optag, struct nfsdsession **sepp,
4693     int *slotposp)
4694 {
4695 	uint32_t *tl;
4696 	int error, len;
4697 
4698 	len = strlen(optag);
4699 	(void)nfsm_strtom(nd, optag, len);
4700 	NFSM_BUILD(tl, uint32_t *, 4 * NFSX_UNSIGNED);
4701 	if ((nd->nd_flag & ND_NFSV41) != 0) {
4702 		if ((nd->nd_flag & ND_NFSV42) != 0)
4703 			*tl++ = txdr_unsigned(NFSV42_MINORVERSION);
4704 		else
4705 			*tl++ = txdr_unsigned(NFSV41_MINORVERSION);
4706 		*tl++ = txdr_unsigned(callback);
4707 		*tl++ = txdr_unsigned(2);
4708 		*tl = txdr_unsigned(NFSV4OP_CBSEQUENCE);
4709 		error = nfsv4_setcbsequence(nd, clp, 1, sepp, slotposp);
4710 		if (error != 0)
4711 			return (error);
4712 		NFSM_BUILD(tl, u_int32_t *, NFSX_UNSIGNED);
4713 		*tl = txdr_unsigned(op);
4714 	} else {
4715 		*tl++ = txdr_unsigned(NFSV4_MINORVERSION);
4716 		*tl++ = txdr_unsigned(callback);
4717 		*tl++ = txdr_unsigned(1);
4718 		*tl = txdr_unsigned(op);
4719 	}
4720 	return (0);
4721 }
4722 
4723 /*
4724  * Return the next index# for a clientid. Mostly just increment and return
4725  * the next one, but... if the 32bit unsigned does actually wrap around,
4726  * it should be rebooted.
4727  * At an average rate of one new client per second, it will wrap around in
4728  * approximately 136 years. (I think the server will have been shut
4729  * down or rebooted before then.)
4730  */
4731 static u_int32_t
4732 nfsrv_nextclientindex(void)
4733 {
4734 	static u_int32_t client_index = 0;
4735 
4736 	client_index++;
4737 	if (client_index != 0)
4738 		return (client_index);
4739 
4740 	printf("%s: out of clientids\n", __func__);
4741 	return (client_index);
4742 }
4743 
4744 /*
4745  * Return the next index# for a stateid. Mostly just increment and return
4746  * the next one, but... if the 32bit unsigned does actually wrap around
4747  * (will a BSD server stay up that long?), find
4748  * new start and end values.
4749  */
4750 static u_int32_t
4751 nfsrv_nextstateindex(struct nfsclient *clp)
4752 {
4753 	struct nfsstate *stp;
4754 	int i;
4755 	u_int32_t canuse, min_index, max_index;
4756 
4757 	if (!(clp->lc_flags & LCL_INDEXNOTOK)) {
4758 		clp->lc_stateindex++;
4759 		if (clp->lc_stateindex != clp->lc_statemaxindex)
4760 			return (clp->lc_stateindex);
4761 	}
4762 
4763 	/*
4764 	 * Yuck, we've hit the end.
4765 	 * Look for a new min and max.
4766 	 */
4767 	min_index = 0;
4768 	max_index = 0xffffffff;
4769 	for (i = 0; i < nfsrv_statehashsize; i++) {
4770 	    LIST_FOREACH(stp, &clp->lc_stateid[i], ls_hash) {
4771 		if (stp->ls_stateid.other[2] > 0x80000000) {
4772 		    if (stp->ls_stateid.other[2] < max_index)
4773 			max_index = stp->ls_stateid.other[2];
4774 		} else {
4775 		    if (stp->ls_stateid.other[2] > min_index)
4776 			min_index = stp->ls_stateid.other[2];
4777 		}
4778 	    }
4779 	}
4780 
4781 	/*
4782 	 * Yikes, highly unlikely, but I'll handle it anyhow.
4783 	 */
4784 	if (min_index == 0x80000000 && max_index == 0x80000001) {
4785 	    canuse = 0;
4786 	    /*
4787 	     * Loop around until we find an unused entry. Return that
4788 	     * and set LCL_INDEXNOTOK, so the search will continue next time.
4789 	     * (This is one of those rare cases where a goto is the
4790 	     *  cleanest way to code the loop.)
4791 	     */
4792 tryagain:
4793 	    for (i = 0; i < nfsrv_statehashsize; i++) {
4794 		LIST_FOREACH(stp, &clp->lc_stateid[i], ls_hash) {
4795 		    if (stp->ls_stateid.other[2] == canuse) {
4796 			canuse++;
4797 			goto tryagain;
4798 		    }
4799 		}
4800 	    }
4801 	    clp->lc_flags |= LCL_INDEXNOTOK;
4802 	    return (canuse);
4803 	}
4804 
4805 	/*
4806 	 * Ok to start again from min + 1.
4807 	 */
4808 	clp->lc_stateindex = min_index + 1;
4809 	clp->lc_statemaxindex = max_index;
4810 	clp->lc_flags &= ~LCL_INDEXNOTOK;
4811 	return (clp->lc_stateindex);
4812 }
4813 
4814 /*
4815  * The following functions handle the stable storage file that deals with
4816  * the edge conditions described in RFC3530 Sec. 8.6.3.
4817  * The file is as follows:
4818  * - a single record at the beginning that has the lease time of the
4819  *   previous server instance (before the last reboot) and the nfsrvboottime
4820  *   values for the previous server boots.
4821  *   These previous boot times are used to ensure that the current
4822  *   nfsrvboottime does not, somehow, get set to a previous one.
4823  *   (This is important so that Stale ClientIDs and StateIDs can
4824  *    be recognized.)
4825  *   The number of previous nfsvrboottime values precedes the list.
4826  * - followed by some number of appended records with:
4827  *   - client id string
4828  *   - flag that indicates it is a record revoking state via lease
4829  *     expiration or similar
4830  *     OR has successfully acquired state.
4831  * These structures vary in length, with the client string at the end, up
4832  * to NFSV4_OPAQUELIMIT in size.
4833  *
4834  * At the end of the grace period, the file is truncated, the first
4835  * record is rewritten with updated information and any acquired state
4836  * records for successful reclaims of state are written.
4837  *
4838  * Subsequent records are appended when the first state is issued to
4839  * a client and when state is revoked for a client.
4840  *
4841  * When reading the file in, state issued records that come later in
4842  * the file override older ones, since the append log is in cronological order.
4843  * If, for some reason, the file can't be read, the grace period is
4844  * immediately terminated and all reclaims get NFSERR_NOGRACE.
4845  */
4846 
4847 /*
4848  * Read in the stable storage file. Called by nfssvc() before the nfsd
4849  * processes start servicing requests.
4850  */
4851 void
4852 nfsrv_setupstable(NFSPROC_T *p)
4853 {
4854 	struct nfsrv_stablefirst *sf = &NFSD_VNET(nfsrv_stablefirst);
4855 	struct nfsrv_stable *sp, *nsp;
4856 	struct nfst_rec *tsp;
4857 	int error, i, tryagain;
4858 	off_t off = 0;
4859 	ssize_t aresid, len;
4860 
4861 	/*
4862 	 * If NFSNSF_UPDATEDONE is set, this is a restart of the nfsds without
4863 	 * a reboot, so state has not been lost.
4864 	 */
4865 	if (sf->nsf_flags & NFSNSF_UPDATEDONE)
4866 		return;
4867 	/*
4868 	 * Set Grace over just until the file reads successfully.
4869 	 */
4870 	NFSD_VNET(nfsrvboottime) = time_second;
4871 	LIST_INIT(&sf->nsf_head);
4872 	sf->nsf_flags = (NFSNSF_GRACEOVER | NFSNSF_NEEDLOCK);
4873 	sf->nsf_eograce = NFSD_MONOSEC + NFSRV_LEASEDELTA;
4874 	if (sf->nsf_fp == NULL)
4875 		return;
4876 	error = NFSD_RDWR(UIO_READ, NFSFPVNODE(sf->nsf_fp),
4877 	    (caddr_t)&sf->nsf_rec, sizeof (struct nfsf_rec), off, UIO_SYSSPACE,
4878 	    0, NFSFPCRED(sf->nsf_fp), &aresid, p);
4879 	if (error || aresid || sf->nsf_numboots == 0 ||
4880 		sf->nsf_numboots > NFSNSF_MAXNUMBOOTS)
4881 		return;
4882 
4883 	/*
4884 	 * Now, read in the boottimes.
4885 	 */
4886 	sf->nsf_bootvals = (time_t *)malloc((sf->nsf_numboots + 1) *
4887 		sizeof(time_t), M_TEMP, M_WAITOK);
4888 	off = sizeof (struct nfsf_rec);
4889 	error = NFSD_RDWR(UIO_READ, NFSFPVNODE(sf->nsf_fp),
4890 	    (caddr_t)sf->nsf_bootvals, sf->nsf_numboots * sizeof (time_t), off,
4891 	    UIO_SYSSPACE, 0, NFSFPCRED(sf->nsf_fp), &aresid, p);
4892 	if (error || aresid) {
4893 		free(sf->nsf_bootvals, M_TEMP);
4894 		sf->nsf_bootvals = NULL;
4895 		return;
4896 	}
4897 
4898 	/*
4899 	 * Make sure this nfsrvboottime is different from all recorded
4900 	 * previous ones.
4901 	 */
4902 	do {
4903 		tryagain = 0;
4904 		for (i = 0; i < sf->nsf_numboots; i++) {
4905 			if (NFSD_VNET(nfsrvboottime) == sf->nsf_bootvals[i]) {
4906 				NFSD_VNET(nfsrvboottime)++;
4907 				tryagain = 1;
4908 				break;
4909 			}
4910 		}
4911 	} while (tryagain);
4912 
4913 	sf->nsf_flags |= NFSNSF_OK;
4914 	off += (sf->nsf_numboots * sizeof (time_t));
4915 
4916 	/*
4917 	 * Read through the file, building a list of records for grace
4918 	 * checking.
4919 	 * Each record is between sizeof (struct nfst_rec) and
4920 	 * sizeof (struct nfst_rec) + NFSV4_OPAQUELIMIT - 1
4921 	 * and is actually sizeof (struct nfst_rec) + nst_len - 1.
4922 	 */
4923 	tsp = (struct nfst_rec *)malloc(sizeof (struct nfst_rec) +
4924 		NFSV4_OPAQUELIMIT - 1, M_TEMP, M_WAITOK);
4925 	do {
4926 	    error = NFSD_RDWR(UIO_READ, NFSFPVNODE(sf->nsf_fp),
4927 	        (caddr_t)tsp, sizeof (struct nfst_rec) + NFSV4_OPAQUELIMIT - 1,
4928 	        off, UIO_SYSSPACE, 0, NFSFPCRED(sf->nsf_fp), &aresid, p);
4929 	    len = (sizeof (struct nfst_rec) + NFSV4_OPAQUELIMIT - 1) - aresid;
4930 	    if (error || (len > 0 && (len < sizeof (struct nfst_rec) ||
4931 		len < (sizeof (struct nfst_rec) + tsp->len - 1)))) {
4932 		/*
4933 		 * Yuck, the file has been corrupted, so just return
4934 		 * after clearing out any restart state, so the grace period
4935 		 * is over.
4936 		 */
4937 		LIST_FOREACH_SAFE(sp, &sf->nsf_head, nst_list, nsp) {
4938 			LIST_REMOVE(sp, nst_list);
4939 			free(sp, M_TEMP);
4940 		}
4941 		free(tsp, M_TEMP);
4942 		sf->nsf_flags &= ~NFSNSF_OK;
4943 		free(sf->nsf_bootvals, M_TEMP);
4944 		sf->nsf_bootvals = NULL;
4945 		return;
4946 	    }
4947 	    if (len > 0) {
4948 		off += sizeof (struct nfst_rec) + tsp->len - 1;
4949 		/*
4950 		 * Search the list for a matching client.
4951 		 */
4952 		LIST_FOREACH(sp, &sf->nsf_head, nst_list) {
4953 			if (tsp->len == sp->nst_len &&
4954 			    !NFSBCMP(tsp->client, sp->nst_client, tsp->len))
4955 				break;
4956 		}
4957 		if (sp == LIST_END(&sf->nsf_head)) {
4958 			sp = (struct nfsrv_stable *)malloc(tsp->len +
4959 				sizeof (struct nfsrv_stable) - 1, M_TEMP,
4960 				M_WAITOK);
4961 			NFSBCOPY((caddr_t)tsp, (caddr_t)&sp->nst_rec,
4962 				sizeof (struct nfst_rec) + tsp->len - 1);
4963 			LIST_INSERT_HEAD(&sf->nsf_head, sp, nst_list);
4964 		} else {
4965 			if (tsp->flag == NFSNST_REVOKE)
4966 				sp->nst_flag |= NFSNST_REVOKE;
4967 			else
4968 				/*
4969 				 * A subsequent timestamp indicates the client
4970 				 * did a setclientid/confirm and any previous
4971 				 * revoke is no longer relevant.
4972 				 */
4973 				sp->nst_flag &= ~NFSNST_REVOKE;
4974 		}
4975 	    }
4976 	} while (len > 0);
4977 	free(tsp, M_TEMP);
4978 	sf->nsf_flags = NFSNSF_OK;
4979 	sf->nsf_eograce = NFSD_MONOSEC + sf->nsf_lease +
4980 		NFSRV_LEASEDELTA;
4981 }
4982 
4983 /*
4984  * Update the stable storage file, now that the grace period is over.
4985  */
4986 void
4987 nfsrv_updatestable(NFSPROC_T *p)
4988 {
4989 	struct nfsrv_stablefirst *sf = &NFSD_VNET(nfsrv_stablefirst);
4990 	struct nfsrv_stable *sp, *nsp;
4991 	int i;
4992 	struct nfsvattr nva;
4993 	vnode_t vp;
4994 #if defined(__FreeBSD_version) && (__FreeBSD_version >= 500000)
4995 	mount_t mp = NULL;
4996 #endif
4997 	int error;
4998 
4999 	if (sf->nsf_fp == NULL || (sf->nsf_flags & NFSNSF_UPDATEDONE))
5000 		return;
5001 	sf->nsf_flags |= NFSNSF_UPDATEDONE;
5002 	/*
5003 	 * Ok, we need to rewrite the stable storage file.
5004 	 * - truncate to 0 length
5005 	 * - write the new first structure
5006 	 * - loop through the data structures, writing out any that
5007 	 *   have timestamps older than the old boot
5008 	 */
5009 	if (sf->nsf_bootvals) {
5010 		sf->nsf_numboots++;
5011 		for (i = sf->nsf_numboots - 2; i >= 0; i--)
5012 			sf->nsf_bootvals[i + 1] = sf->nsf_bootvals[i];
5013 	} else {
5014 		sf->nsf_numboots = 1;
5015 		sf->nsf_bootvals = (time_t *)malloc(sizeof(time_t),
5016 			M_TEMP, M_WAITOK);
5017 	}
5018 	sf->nsf_bootvals[0] = NFSD_VNET(nfsrvboottime);
5019 	sf->nsf_lease = nfsrv_lease;
5020 	NFSVNO_ATTRINIT(&nva);
5021 	NFSVNO_SETATTRVAL(&nva, size, 0);
5022 	vp = NFSFPVNODE(sf->nsf_fp);
5023 	vn_start_write(vp, &mp, V_WAIT);
5024 	if (NFSVOPLOCK(vp, LK_EXCLUSIVE) == 0) {
5025 		error = nfsvno_setattr(vp, &nva, NFSFPCRED(sf->nsf_fp), p,
5026 		    NULL);
5027 		NFSVOPUNLOCK(vp);
5028 	} else
5029 		error = EPERM;
5030 	vn_finished_write(mp);
5031 	if (!error)
5032 	    error = NFSD_RDWR(UIO_WRITE, vp,
5033 		(caddr_t)&sf->nsf_rec, sizeof (struct nfsf_rec), (off_t)0,
5034 		UIO_SYSSPACE, IO_SYNC, NFSFPCRED(sf->nsf_fp), NULL, p);
5035 	if (!error)
5036 	    error = NFSD_RDWR(UIO_WRITE, vp,
5037 		(caddr_t)sf->nsf_bootvals,
5038 		sf->nsf_numboots * sizeof (time_t),
5039 		(off_t)(sizeof (struct nfsf_rec)),
5040 		UIO_SYSSPACE, IO_SYNC, NFSFPCRED(sf->nsf_fp), NULL, p);
5041 	free(sf->nsf_bootvals, M_TEMP);
5042 	sf->nsf_bootvals = NULL;
5043 	if (error) {
5044 		sf->nsf_flags &= ~NFSNSF_OK;
5045 		printf("EEK! Can't write NfsV4 stable storage file\n");
5046 		return;
5047 	}
5048 	sf->nsf_flags |= NFSNSF_OK;
5049 
5050 	/*
5051 	 * Loop through the list and write out timestamp records for
5052 	 * any clients that successfully reclaimed state.
5053 	 */
5054 	LIST_FOREACH_SAFE(sp, &sf->nsf_head, nst_list, nsp) {
5055 		if (sp->nst_flag & NFSNST_GOTSTATE) {
5056 			nfsrv_writestable(sp->nst_client, sp->nst_len,
5057 				NFSNST_NEWSTATE, p);
5058 			sp->nst_clp->lc_flags |= LCL_STAMPEDSTABLE;
5059 		}
5060 		LIST_REMOVE(sp, nst_list);
5061 		free(sp, M_TEMP);
5062 	}
5063 	nfsrv_backupstable();
5064 }
5065 
5066 /*
5067  * Append a record to the stable storage file.
5068  */
5069 void
5070 nfsrv_writestable(u_char *client, int len, int flag, NFSPROC_T *p)
5071 {
5072 	struct nfsrv_stablefirst *sf = &NFSD_VNET(nfsrv_stablefirst);
5073 	struct nfst_rec *sp;
5074 	int error;
5075 
5076 	if (!(sf->nsf_flags & NFSNSF_OK) || sf->nsf_fp == NULL)
5077 		return;
5078 	sp = (struct nfst_rec *)malloc(sizeof (struct nfst_rec) +
5079 		len - 1, M_TEMP, M_WAITOK);
5080 	sp->len = len;
5081 	NFSBCOPY(client, sp->client, len);
5082 	sp->flag = flag;
5083 	error = NFSD_RDWR(UIO_WRITE, NFSFPVNODE(sf->nsf_fp),
5084 	    (caddr_t)sp, sizeof (struct nfst_rec) + len - 1, (off_t)0,
5085 	    UIO_SYSSPACE, (IO_SYNC | IO_APPEND), NFSFPCRED(sf->nsf_fp), NULL, p);
5086 	free(sp, M_TEMP);
5087 	if (error) {
5088 		sf->nsf_flags &= ~NFSNSF_OK;
5089 		printf("EEK! Can't write NfsV4 stable storage file\n");
5090 	}
5091 }
5092 
5093 /*
5094  * This function is called during the grace period to mark a client
5095  * that successfully reclaimed state.
5096  */
5097 static void
5098 nfsrv_markstable(struct nfsclient *clp)
5099 {
5100 	struct nfsrv_stable *sp;
5101 
5102 	/*
5103 	 * First find the client structure.
5104 	 */
5105 	LIST_FOREACH(sp, &NFSD_VNET(nfsrv_stablefirst).nsf_head, nst_list) {
5106 		if (sp->nst_len == clp->lc_idlen &&
5107 		    !NFSBCMP(sp->nst_client, clp->lc_id, sp->nst_len))
5108 			break;
5109 	}
5110 	if (sp == LIST_END(&NFSD_VNET(nfsrv_stablefirst).nsf_head))
5111 		return;
5112 
5113 	/*
5114 	 * Now, just mark it and set the nfsclient back pointer.
5115 	 */
5116 	sp->nst_flag |= NFSNST_GOTSTATE;
5117 	sp->nst_clp = clp;
5118 }
5119 
5120 /*
5121  * This function is called when a NFSv4.1 client does a ReclaimComplete.
5122  * Very similar to nfsrv_markstable(), except for the flag being set.
5123  */
5124 static void
5125 nfsrv_markreclaim(struct nfsclient *clp)
5126 {
5127 	struct nfsrv_stable *sp;
5128 
5129 	/*
5130 	 * First find the client structure.
5131 	 */
5132 	LIST_FOREACH(sp, &NFSD_VNET(nfsrv_stablefirst).nsf_head, nst_list) {
5133 		if (sp->nst_len == clp->lc_idlen &&
5134 		    !NFSBCMP(sp->nst_client, clp->lc_id, sp->nst_len))
5135 			break;
5136 	}
5137 	if (sp == LIST_END(&NFSD_VNET(nfsrv_stablefirst).nsf_head))
5138 		return;
5139 
5140 	/*
5141 	 * Now, just set the flag.
5142 	 */
5143 	sp->nst_flag |= NFSNST_RECLAIMED;
5144 
5145 	/*
5146 	 * Free up any old delegations.
5147 	 */
5148 	nfsrv_freedeleglist(&clp->lc_olddeleg);
5149 }
5150 
5151 /*
5152  * This function is called for a reclaim, to see if it gets grace.
5153  * It returns 0 if a reclaim is allowed, 1 otherwise.
5154  */
5155 static int
5156 nfsrv_checkstable(struct nfsclient *clp)
5157 {
5158 	struct nfsrv_stable *sp;
5159 
5160 	/*
5161 	 * First, find the entry for the client.
5162 	 */
5163 	LIST_FOREACH(sp, &NFSD_VNET(nfsrv_stablefirst).nsf_head, nst_list) {
5164 		if (sp->nst_len == clp->lc_idlen &&
5165 		    !NFSBCMP(sp->nst_client, clp->lc_id, sp->nst_len))
5166 			break;
5167 	}
5168 
5169 	/*
5170 	 * If not in the list, state was revoked or no state was issued
5171 	 * since the previous reboot, a reclaim is denied.
5172 	 */
5173 	if (sp == LIST_END(&NFSD_VNET(nfsrv_stablefirst).nsf_head) ||
5174 	    (sp->nst_flag & NFSNST_REVOKE) ||
5175 	    !(NFSD_VNET(nfsrv_stablefirst).nsf_flags & NFSNSF_OK))
5176 		return (1);
5177 	return (0);
5178 }
5179 
5180 /*
5181  * Test for and try to clear out a conflicting client. This is called by
5182  * nfsrv_lockctrl() and nfsrv_openctrl() when conflicts with other clients
5183  * a found.
5184  * The trick here is that it can't revoke a conflicting client with an
5185  * expired lease unless it holds the v4root lock, so...
5186  * If no v4root lock, get the lock and return 1 to indicate "try again".
5187  * Return 0 to indicate the conflict can't be revoked and 1 to indicate
5188  * the revocation worked and the conflicting client is "bye, bye", so it
5189  * can be tried again.
5190  * Return 2 to indicate that the vnode is VIRF_DOOMED after NFSVOPLOCK().
5191  * Unlocks State before a non-zero value is returned.
5192  */
5193 static int
5194 nfsrv_clientconflict(struct nfsclient *clp, int *haslockp, vnode_t vp,
5195     NFSPROC_T *p)
5196 {
5197 	int gotlock, lktype = 0;
5198 
5199 	/*
5200 	 * If lease hasn't expired, we can't fix it.
5201 	 */
5202 	if (clp->lc_expiry >= NFSD_MONOSEC ||
5203 	    !(NFSD_VNET(nfsrv_stablefirst).nsf_flags & NFSNSF_UPDATEDONE))
5204 		return (0);
5205 	if (*haslockp == 0) {
5206 		NFSUNLOCKSTATE();
5207 		if (vp != NULL) {
5208 			lktype = NFSVOPISLOCKED(vp);
5209 			NFSVOPUNLOCK(vp);
5210 		}
5211 		NFSLOCKV4ROOTMUTEX();
5212 		nfsv4_relref(&nfsv4rootfs_lock);
5213 		do {
5214 			gotlock = nfsv4_lock(&nfsv4rootfs_lock, 1, NULL,
5215 			    NFSV4ROOTLOCKMUTEXPTR, NULL);
5216 		} while (!gotlock);
5217 		NFSUNLOCKV4ROOTMUTEX();
5218 		*haslockp = 1;
5219 		if (vp != NULL) {
5220 			NFSVOPLOCK(vp, lktype | LK_RETRY);
5221 			if (VN_IS_DOOMED(vp))
5222 				return (2);
5223 		}
5224 		return (1);
5225 	}
5226 	NFSUNLOCKSTATE();
5227 
5228 	/*
5229 	 * Ok, we can expire the conflicting client.
5230 	 */
5231 	nfsrv_writestable(clp->lc_id, clp->lc_idlen, NFSNST_REVOKE, p);
5232 	nfsrv_backupstable();
5233 	nfsrv_cleanclient(clp, p, false, NULL);
5234 	nfsrv_freedeleglist(&clp->lc_deleg);
5235 	nfsrv_freedeleglist(&clp->lc_olddeleg);
5236 	LIST_REMOVE(clp, lc_hash);
5237 	nfsrv_zapclient(clp, p);
5238 	return (1);
5239 }
5240 
5241 /*
5242  * Resolve a delegation conflict.
5243  * Returns 0 to indicate the conflict was resolved without sleeping.
5244  * Return -1 to indicate that the caller should check for conflicts again.
5245  * Return > 0 for an error that should be returned, normally NFSERR_DELAY.
5246  *
5247  * Also, manipulate the nfsv4root_lock, as required. It isn't changed
5248  * for a return of 0, since there was no sleep and it could be required
5249  * later. It is released for a return of NFSERR_DELAY, since the caller
5250  * will return that error. It is released when a sleep was done waiting
5251  * for the delegation to be returned or expire (so that other nfsds can
5252  * handle ops). Then, it must be acquired for the write to stable storage.
5253  * (This function is somewhat similar to nfsrv_clientconflict(), but
5254  *  the semantics differ in a couple of subtle ways. The return of 0
5255  *  indicates the conflict was resolved without sleeping here, not
5256  *  that the conflict can't be resolved and the handling of nfsv4root_lock
5257  *  differs, as noted above.)
5258  * Unlocks State before returning a non-zero value.
5259  */
5260 static int
5261 nfsrv_delegconflict(struct nfsstate *stp, int *haslockp, NFSPROC_T *p,
5262     vnode_t vp)
5263 {
5264 	struct nfsclient *clp = stp->ls_clp;
5265 	int gotlock, error, lktype = 0, retrycnt, zapped_clp;
5266 	nfsv4stateid_t tstateid;
5267 	fhandle_t tfh;
5268 
5269 	/*
5270 	 * If the conflict is with an old delegation...
5271 	 */
5272 	if (stp->ls_flags & NFSLCK_OLDDELEG) {
5273 		/*
5274 		 * You can delete it, if it has expired.
5275 		 */
5276 		if (clp->lc_delegtime < NFSD_MONOSEC) {
5277 			nfsrv_freedeleg(stp);
5278 			NFSUNLOCKSTATE();
5279 			error = -1;
5280 			goto out;
5281 		}
5282 		NFSUNLOCKSTATE();
5283 		/*
5284 		 * During this delay, the old delegation could expire or it
5285 		 * could be recovered by the client via an Open with
5286 		 * CLAIM_DELEGATE_PREV.
5287 		 * Release the nfsv4root_lock, if held.
5288 		 */
5289 		if (*haslockp) {
5290 			*haslockp = 0;
5291 			NFSLOCKV4ROOTMUTEX();
5292 			nfsv4_unlock(&nfsv4rootfs_lock, 1);
5293 			NFSUNLOCKV4ROOTMUTEX();
5294 		}
5295 		error = NFSERR_DELAY;
5296 		goto out;
5297 	}
5298 
5299 	/*
5300 	 * It's a current delegation, so:
5301 	 * - check to see if the delegation has expired
5302 	 *   - if so, get the v4root lock and then expire it
5303 	 */
5304 	if ((stp->ls_flags & NFSLCK_DELEGRECALL) == 0 || (stp->ls_lastrecall <
5305 	    NFSD_MONOSEC && clp->lc_expiry >= NFSD_MONOSEC &&
5306 	    stp->ls_delegtime >= NFSD_MONOSEC)) {
5307 		/*
5308 		 * - do a recall callback, since not yet done
5309 		 * For now, never allow truncate to be set. To use
5310 		 * truncate safely, it must be guaranteed that the
5311 		 * Remove, Rename or Setattr with size of 0 will
5312 		 * succeed and that would require major changes to
5313 		 * the VFS/Vnode OPs.
5314 		 * Set the expiry time large enough so that it won't expire
5315 		 * until after the callback, then set it correctly, once
5316 		 * the callback is done. (The delegation will now time
5317 		 * out whether or not the Recall worked ok. The timeout
5318 		 * will be extended when ops are done on the delegation
5319 		 * stateid, up to the timelimit.)
5320 		 */
5321 		if ((stp->ls_flags & NFSLCK_DELEGRECALL) == 0) {
5322 			stp->ls_delegtime = NFSD_MONOSEC + (2 * nfsrv_lease) +
5323 			    NFSRV_LEASEDELTA;
5324 			stp->ls_delegtimelimit = NFSD_MONOSEC + (6 *
5325 			    nfsrv_lease) + NFSRV_LEASEDELTA;
5326 			stp->ls_flags |= NFSLCK_DELEGRECALL;
5327 		}
5328 		stp->ls_lastrecall = time_uptime + 1;
5329 
5330 		/*
5331 		 * Loop NFSRV_CBRETRYCNT times while the CBRecall replies
5332 		 * NFSERR_BADSTATEID or NFSERR_BADHANDLE. This is done
5333 		 * in order to try and avoid a race that could happen
5334 		 * when a CBRecall request passed the Open reply with
5335 		 * the delegation in it when transitting the network.
5336 		 * Since nfsrv_docallback will sleep, don't use stp after
5337 		 * the call.
5338 		 */
5339 		NFSBCOPY((caddr_t)&stp->ls_stateid, (caddr_t)&tstateid,
5340 		    sizeof (tstateid));
5341 		NFSBCOPY((caddr_t)&stp->ls_lfp->lf_fh, (caddr_t)&tfh,
5342 		    sizeof (tfh));
5343 		NFSUNLOCKSTATE();
5344 		if (*haslockp) {
5345 			*haslockp = 0;
5346 			NFSLOCKV4ROOTMUTEX();
5347 			nfsv4_unlock(&nfsv4rootfs_lock, 1);
5348 			NFSUNLOCKV4ROOTMUTEX();
5349 		}
5350 		retrycnt = 0;
5351 		do {
5352 		    error = nfsrv_docallback(clp, NFSV4OP_CBRECALL,
5353 			&tstateid, 0, &tfh, NULL, NULL, 0, p);
5354 		    retrycnt++;
5355 		} while ((error == NFSERR_BADSTATEID ||
5356 		    error == NFSERR_BADHANDLE) && retrycnt < NFSV4_CBRETRYCNT);
5357 		error = NFSERR_DELAY;
5358 		goto out;
5359 	}
5360 
5361 	if (clp->lc_expiry >= NFSD_MONOSEC &&
5362 	    stp->ls_delegtime >= NFSD_MONOSEC) {
5363 		NFSUNLOCKSTATE();
5364 		/*
5365 		 * A recall has been done, but it has not yet expired.
5366 		 * So, RETURN_DELAY.
5367 		 */
5368 		if (*haslockp) {
5369 			*haslockp = 0;
5370 			NFSLOCKV4ROOTMUTEX();
5371 			nfsv4_unlock(&nfsv4rootfs_lock, 1);
5372 			NFSUNLOCKV4ROOTMUTEX();
5373 		}
5374 		error = NFSERR_DELAY;
5375 		goto out;
5376 	}
5377 
5378 	/*
5379 	 * If we don't yet have the lock, just get it and then return,
5380 	 * since we need that before deleting expired state, such as
5381 	 * this delegation.
5382 	 * When getting the lock, unlock the vnode, so other nfsds that
5383 	 * are in progress, won't get stuck waiting for the vnode lock.
5384 	 */
5385 	if (*haslockp == 0) {
5386 		NFSUNLOCKSTATE();
5387 		if (vp != NULL) {
5388 			lktype = NFSVOPISLOCKED(vp);
5389 			NFSVOPUNLOCK(vp);
5390 		}
5391 		NFSLOCKV4ROOTMUTEX();
5392 		nfsv4_relref(&nfsv4rootfs_lock);
5393 		do {
5394 			gotlock = nfsv4_lock(&nfsv4rootfs_lock, 1, NULL,
5395 			    NFSV4ROOTLOCKMUTEXPTR, NULL);
5396 		} while (!gotlock);
5397 		NFSUNLOCKV4ROOTMUTEX();
5398 		*haslockp = 1;
5399 		if (vp != NULL) {
5400 			NFSVOPLOCK(vp, lktype | LK_RETRY);
5401 			if (VN_IS_DOOMED(vp)) {
5402 				*haslockp = 0;
5403 				NFSLOCKV4ROOTMUTEX();
5404 				nfsv4_unlock(&nfsv4rootfs_lock, 1);
5405 				NFSUNLOCKV4ROOTMUTEX();
5406 				error = NFSERR_PERM;
5407 				goto out;
5408 			}
5409 		}
5410 		error = -1;
5411 		goto out;
5412 	}
5413 
5414 	NFSUNLOCKSTATE();
5415 	/*
5416 	 * Ok, we can delete the expired delegation.
5417 	 * First, write the Revoke record to stable storage and then
5418 	 * clear out the conflict.
5419 	 * Since all other nfsd threads are now blocked, we can safely
5420 	 * sleep without the state changing.
5421 	 */
5422 	nfsrv_writestable(clp->lc_id, clp->lc_idlen, NFSNST_REVOKE, p);
5423 	nfsrv_backupstable();
5424 	if (clp->lc_expiry < NFSD_MONOSEC) {
5425 		nfsrv_cleanclient(clp, p, false, NULL);
5426 		nfsrv_freedeleglist(&clp->lc_deleg);
5427 		nfsrv_freedeleglist(&clp->lc_olddeleg);
5428 		LIST_REMOVE(clp, lc_hash);
5429 		zapped_clp = 1;
5430 	} else {
5431 		nfsrv_freedeleg(stp);
5432 		zapped_clp = 0;
5433 	}
5434 	if (zapped_clp)
5435 		nfsrv_zapclient(clp, p);
5436 	error = -1;
5437 
5438 out:
5439 	NFSEXITCODE(error);
5440 	return (error);
5441 }
5442 
5443 /*
5444  * Check for a remove allowed, if remove is set to 1 and get rid of
5445  * delegations.
5446  */
5447 int
5448 nfsrv_checkremove(vnode_t vp, int remove, struct nfsrv_descript *nd,
5449     nfsquad_t clientid, NFSPROC_T *p)
5450 {
5451 	struct nfsclient *clp;
5452 	struct nfsstate *stp;
5453 	struct nfslockfile *lfp;
5454 	int error, haslock = 0;
5455 	fhandle_t nfh;
5456 
5457 	clp = NULL;
5458 	/*
5459 	 * First, get the lock file structure.
5460 	 * (A return of -1 means no associated state, so remove ok.)
5461 	 */
5462 	error = nfsrv_getlockfh(vp, NFSLCK_CHECK, NULL, &nfh, p);
5463 tryagain:
5464 	NFSLOCKSTATE();
5465 	if (error == 0 && clientid.qval != 0)
5466 		error = nfsrv_getclient(clientid, CLOPS_RENEW, &clp, NULL,
5467 		    (nfsquad_t)((u_quad_t)0), 0, nd, p);
5468 	if (!error)
5469 		error = nfsrv_getlockfile(NFSLCK_CHECK, NULL, &lfp, &nfh, 0);
5470 	if (error) {
5471 		NFSUNLOCKSTATE();
5472 		if (haslock) {
5473 			NFSLOCKV4ROOTMUTEX();
5474 			nfsv4_unlock(&nfsv4rootfs_lock, 1);
5475 			NFSUNLOCKV4ROOTMUTEX();
5476 		}
5477 		if (error == -1)
5478 			error = 0;
5479 		goto out;
5480 	}
5481 
5482 	/*
5483 	 * Now, we must Recall any delegations.
5484 	 */
5485 	error = nfsrv_cleandeleg(vp, lfp, clp, &haslock, p);
5486 	if (error) {
5487 		/*
5488 		 * nfsrv_cleandeleg() unlocks state for non-zero
5489 		 * return.
5490 		 */
5491 		if (error == -1)
5492 			goto tryagain;
5493 		if (haslock) {
5494 			NFSLOCKV4ROOTMUTEX();
5495 			nfsv4_unlock(&nfsv4rootfs_lock, 1);
5496 			NFSUNLOCKV4ROOTMUTEX();
5497 		}
5498 		goto out;
5499 	}
5500 
5501 	/*
5502 	 * Now, look for a conflicting open share.
5503 	 */
5504 	if (remove) {
5505 		/*
5506 		 * If the entry in the directory was the last reference to the
5507 		 * corresponding filesystem object, the object can be destroyed
5508 		 * */
5509 		if(lfp->lf_usecount>1)
5510 			LIST_FOREACH(stp, &lfp->lf_open, ls_file) {
5511 				if (stp->ls_flags & NFSLCK_WRITEDENY) {
5512 					error = NFSERR_FILEOPEN;
5513 					break;
5514 				}
5515 			}
5516 	}
5517 
5518 	NFSUNLOCKSTATE();
5519 	if (haslock) {
5520 		NFSLOCKV4ROOTMUTEX();
5521 		nfsv4_unlock(&nfsv4rootfs_lock, 1);
5522 		NFSUNLOCKV4ROOTMUTEX();
5523 	}
5524 
5525 out:
5526 	NFSEXITCODE(error);
5527 	return (error);
5528 }
5529 
5530 /*
5531  * Clear out all delegations for the file referred to by lfp.
5532  * May return NFSERR_DELAY, if there will be a delay waiting for
5533  * delegations to expire.
5534  * Returns -1 to indicate it slept while recalling a delegation.
5535  * This function has the side effect of deleting the nfslockfile structure,
5536  * if it no longer has associated state and didn't have to sleep.
5537  * Unlocks State before a non-zero value is returned.
5538  */
5539 static int
5540 nfsrv_cleandeleg(vnode_t vp, struct nfslockfile *lfp,
5541     struct nfsclient *clp, int *haslockp, NFSPROC_T *p)
5542 {
5543 	struct nfsstate *stp, *nstp;
5544 	int ret = 0;
5545 
5546 	stp = LIST_FIRST(&lfp->lf_deleg);
5547 	while (stp != LIST_END(&lfp->lf_deleg)) {
5548 		nstp = LIST_NEXT(stp, ls_file);
5549 		if (stp->ls_clp != clp) {
5550 			ret = nfsrv_delegconflict(stp, haslockp, p, vp);
5551 			if (ret) {
5552 				/*
5553 				 * nfsrv_delegconflict() unlocks state
5554 				 * when it returns non-zero.
5555 				 */
5556 				goto out;
5557 			}
5558 		}
5559 		stp = nstp;
5560 	}
5561 out:
5562 	NFSEXITCODE(ret);
5563 	return (ret);
5564 }
5565 
5566 /*
5567  * There are certain operations that, when being done outside of NFSv4,
5568  * require that any NFSv4 delegation for the file be recalled.
5569  * This function is to be called for those cases:
5570  * VOP_RENAME() - When a delegation is being recalled for any reason,
5571  *	the client may have to do Opens against the server, using the file's
5572  *	final component name. If the file has been renamed on the server,
5573  *	that component name will be incorrect and the Open will fail.
5574  * VOP_REMOVE() - Theoretically, a client could Open a file after it has
5575  *	been removed on the server, if there is a delegation issued to
5576  *	that client for the file. I say "theoretically" since clients
5577  *	normally do an Access Op before the Open and that Access Op will
5578  *	fail with ESTALE. Note that NFSv2 and 3 don't even do Opens, so
5579  *	they will detect the file's removal in the same manner. (There is
5580  *	one case where RFC3530 allows a client to do an Open without first
5581  *	doing an Access Op, which is passage of a check against the ACE
5582  *	returned with a Write delegation, but current practice is to ignore
5583  *	the ACE and always do an Access Op.)
5584  *	Since the functions can only be called with an unlocked vnode, this
5585  *	can't be done at this time.
5586  * VOP_ADVLOCK() - When a client holds a delegation, it can issue byte range
5587  *	locks locally in the client, which are not visible to the server. To
5588  *	deal with this, issuing of delegations for a vnode must be disabled
5589  *	and all delegations for the vnode recalled. This is done via the
5590  *	second function, using the VV_DISABLEDELEG vflag on the vnode.
5591  */
5592 void
5593 nfsd_recalldelegation(vnode_t vp, NFSPROC_T *p)
5594 {
5595 	time_t starttime;
5596 	int error;
5597 
5598 	/*
5599 	 * First, check to see if the server is currently running and it has
5600 	 * been called for a regular file when issuing delegations.
5601 	 */
5602 	if (NFSD_VNET(nfsrv_numnfsd) == 0 || vp->v_type != VREG ||
5603 	    nfsrv_issuedelegs == 0)
5604 		return;
5605 
5606 	KASSERT((NFSVOPISLOCKED(vp) != LK_EXCLUSIVE), ("vp %p is locked", vp));
5607 	/*
5608 	 * First, get a reference on the nfsv4rootfs_lock so that an
5609 	 * exclusive lock cannot be acquired by another thread.
5610 	 */
5611 	NFSLOCKV4ROOTMUTEX();
5612 	nfsv4_getref(&nfsv4rootfs_lock, NULL, NFSV4ROOTLOCKMUTEXPTR, NULL);
5613 	NFSUNLOCKV4ROOTMUTEX();
5614 
5615 	/*
5616 	 * Now, call nfsrv_checkremove() in a loop while it returns
5617 	 * NFSERR_DELAY. Return upon any other error or when timed out.
5618 	 */
5619 	starttime = NFSD_MONOSEC;
5620 	do {
5621 		if (NFSVOPLOCK(vp, LK_EXCLUSIVE) == 0) {
5622 			error = nfsrv_checkremove(vp, 0, NULL,
5623 			    (nfsquad_t)((u_quad_t)0), p);
5624 			NFSVOPUNLOCK(vp);
5625 		} else
5626 			error = EPERM;
5627 		if (error == NFSERR_DELAY) {
5628 			if (NFSD_MONOSEC - starttime > NFS_REMOVETIMEO)
5629 				break;
5630 			/* Sleep for a short period of time */
5631 			(void) nfs_catnap(PZERO, 0, "nfsremove");
5632 		}
5633 	} while (error == NFSERR_DELAY);
5634 	NFSLOCKV4ROOTMUTEX();
5635 	nfsv4_relref(&nfsv4rootfs_lock);
5636 	NFSUNLOCKV4ROOTMUTEX();
5637 }
5638 
5639 void
5640 nfsd_disabledelegation(vnode_t vp, NFSPROC_T *p)
5641 {
5642 
5643 #ifdef VV_DISABLEDELEG
5644 	/*
5645 	 * First, flag issuance of delegations disabled.
5646 	 */
5647 	atomic_set_long(&vp->v_vflag, VV_DISABLEDELEG);
5648 #endif
5649 
5650 	/*
5651 	 * Then call nfsd_recalldelegation() to get rid of all extant
5652 	 * delegations.
5653 	 */
5654 	nfsd_recalldelegation(vp, p);
5655 }
5656 
5657 /*
5658  * Check for conflicting locks, etc. and then get rid of delegations.
5659  * (At one point I thought that I should get rid of delegations for any
5660  *  Setattr, since it could potentially disallow the I/O op (read or write)
5661  *  allowed by the delegation. However, Setattr Ops that aren't changing
5662  *  the size get a stateid of all 0s, so you can't tell if it is a delegation
5663  *  for the same client or a different one, so I decided to only get rid
5664  *  of delegations for other clients when the size is being changed.)
5665  * In general, a Setattr can disable NFS I/O Ops that are outstanding, such
5666  * as Write backs, even if there is no delegation, so it really isn't any
5667  * different?)
5668  */
5669 int
5670 nfsrv_checksetattr(vnode_t vp, struct nfsrv_descript *nd,
5671     nfsv4stateid_t *stateidp, struct nfsvattr *nvap, nfsattrbit_t *attrbitp,
5672     struct nfsexstuff *exp, NFSPROC_T *p)
5673 {
5674 	struct nfsstate st, *stp = &st;
5675 	struct nfslock lo, *lop = &lo;
5676 	int error = 0;
5677 	nfsquad_t clientid;
5678 
5679 	if (NFSISSET_ATTRBIT(attrbitp, NFSATTRBIT_SIZE)) {
5680 		stp->ls_flags = (NFSLCK_CHECK | NFSLCK_WRITEACCESS);
5681 		lop->lo_first = nvap->na_size;
5682 	} else {
5683 		stp->ls_flags = 0;
5684 		lop->lo_first = 0;
5685 	}
5686 	if (NFSISSET_ATTRBIT(attrbitp, NFSATTRBIT_OWNER) ||
5687 	    NFSISSET_ATTRBIT(attrbitp, NFSATTRBIT_OWNERGROUP) ||
5688 	    NFSISSET_ATTRBIT(attrbitp, NFSATTRBIT_MODE) ||
5689 	    NFSISSET_ATTRBIT(attrbitp, NFSATTRBIT_ACL))
5690 		stp->ls_flags |= NFSLCK_SETATTR;
5691 	if (stp->ls_flags == 0)
5692 		goto out;
5693 	lop->lo_end = NFS64BITSSET;
5694 	lop->lo_flags = NFSLCK_WRITE;
5695 	stp->ls_ownerlen = 0;
5696 	stp->ls_op = NULL;
5697 	stp->ls_uid = nd->nd_cred->cr_uid;
5698 	stp->ls_stateid.seqid = stateidp->seqid;
5699 	clientid.lval[0] = stp->ls_stateid.other[0] = stateidp->other[0];
5700 	clientid.lval[1] = stp->ls_stateid.other[1] = stateidp->other[1];
5701 	stp->ls_stateid.other[2] = stateidp->other[2];
5702 	error = nfsrv_lockctrl(vp, &stp, &lop, NULL, clientid,
5703 	    stateidp, exp, nd, p);
5704 
5705 out:
5706 	NFSEXITCODE2(error, nd);
5707 	return (error);
5708 }
5709 
5710 /*
5711  * Check for a write delegation and do a CBGETATTR if there is one, updating
5712  * the attributes, as required.
5713  * Should I return an error if I can't get the attributes? (For now, I'll
5714  * just return ok.
5715  */
5716 int
5717 nfsrv_checkgetattr(struct nfsrv_descript *nd, vnode_t vp,
5718     struct nfsvattr *nvap, nfsattrbit_t *attrbitp, NFSPROC_T *p)
5719 {
5720 	struct nfsstate *stp;
5721 	struct nfslockfile *lfp;
5722 	struct nfsclient *clp;
5723 	struct nfsvattr nva;
5724 	fhandle_t nfh;
5725 	int error = 0;
5726 	nfsattrbit_t cbbits;
5727 	u_quad_t delegfilerev;
5728 
5729 	NFSCBGETATTR_ATTRBIT(attrbitp, &cbbits);
5730 	if (!NFSNONZERO_ATTRBIT(&cbbits))
5731 		goto out;
5732 	if (nfsrv_writedelegcnt == 0)
5733 		goto out;
5734 
5735 	/*
5736 	 * Get the lock file structure.
5737 	 * (A return of -1 means no associated state, so return ok.)
5738 	 */
5739 	error = nfsrv_getlockfh(vp, NFSLCK_CHECK, NULL, &nfh, p);
5740 	NFSLOCKSTATE();
5741 	if (!error)
5742 		error = nfsrv_getlockfile(NFSLCK_CHECK, NULL, &lfp, &nfh, 0);
5743 	if (error) {
5744 		NFSUNLOCKSTATE();
5745 		if (error == -1)
5746 			error = 0;
5747 		goto out;
5748 	}
5749 
5750 	/*
5751 	 * Now, look for a write delegation.
5752 	 */
5753 	LIST_FOREACH(stp, &lfp->lf_deleg, ls_file) {
5754 		if (stp->ls_flags & NFSLCK_DELEGWRITE)
5755 			break;
5756 	}
5757 	if (stp == LIST_END(&lfp->lf_deleg)) {
5758 		NFSUNLOCKSTATE();
5759 		goto out;
5760 	}
5761 	clp = stp->ls_clp;
5762 
5763 	/* If the clientid is not confirmed, ignore the delegation. */
5764 	if (clp->lc_flags & LCL_NEEDSCONFIRM) {
5765 		NFSUNLOCKSTATE();
5766 		goto out;
5767 	}
5768 
5769 	delegfilerev = stp->ls_filerev;
5770 	/*
5771 	 * If the Write delegation was issued as a part of this Compound RPC
5772 	 * or if we have an Implied Clientid (used in a previous Op in this
5773 	 * compound) and it is the client the delegation was issued to,
5774 	 * just return ok.
5775 	 * I also assume that it is from the same client iff the network
5776 	 * host IP address is the same as the callback address. (Not
5777 	 * exactly correct by the RFC, but avoids a lot of Getattr
5778 	 * callbacks.)
5779 	 */
5780 	if (nd->nd_compref == stp->ls_compref ||
5781 	    ((nd->nd_flag & ND_IMPLIEDCLID) &&
5782 	     clp->lc_clientid.qval == nd->nd_clientid.qval) ||
5783 	     nfsaddr2_match(clp->lc_req.nr_nam, nd->nd_nam)) {
5784 		NFSUNLOCKSTATE();
5785 		goto out;
5786 	}
5787 
5788 	/*
5789 	 * We are now done with the delegation state structure,
5790 	 * so the statelock can be released and we can now tsleep().
5791 	 */
5792 
5793 	/*
5794 	 * Now, we must do the CB Getattr callback, to see if Change or Size
5795 	 * has changed.
5796 	 */
5797 	if (clp->lc_expiry >= NFSD_MONOSEC) {
5798 		NFSUNLOCKSTATE();
5799 		NFSVNO_ATTRINIT(&nva);
5800 		nva.na_filerev = NFS64BITSSET;
5801 		error = nfsrv_docallback(clp, NFSV4OP_CBGETATTR, NULL,
5802 		    0, &nfh, &nva, &cbbits, 0, p);
5803 		if (!error) {
5804 			if ((nva.na_filerev != NFS64BITSSET &&
5805 			    nva.na_filerev > delegfilerev) ||
5806 			    (NFSVNO_ISSETSIZE(&nva) &&
5807 			     nva.na_size != nvap->na_size)) {
5808 				error = nfsvno_updfilerev(vp, nvap, nd, p);
5809 				if (NFSVNO_ISSETSIZE(&nva))
5810 					nvap->na_size = nva.na_size;
5811 			}
5812 		} else
5813 			error = 0;	/* Ignore callback errors for now. */
5814 	} else {
5815 		NFSUNLOCKSTATE();
5816 	}
5817 
5818 out:
5819 	NFSEXITCODE2(error, nd);
5820 	return (error);
5821 }
5822 
5823 /*
5824  * This function looks for openowners that haven't had any opens for
5825  * a while and throws them away. Called by an nfsd when NFSNSF_NOOPENS
5826  * is set.
5827  */
5828 void
5829 nfsrv_throwawayopens(NFSPROC_T *p)
5830 {
5831 	struct nfsclient *clp, *nclp;
5832 	struct nfsstate *stp, *nstp;
5833 	int i;
5834 
5835 	NFSLOCKSTATE();
5836 	NFSD_VNET(nfsrv_stablefirst).nsf_flags &= ~NFSNSF_NOOPENS;
5837 	/*
5838 	 * For each client...
5839 	 */
5840 	for (i = 0; i < nfsrv_clienthashsize; i++) {
5841 	    LIST_FOREACH_SAFE(clp, &NFSD_VNET(nfsclienthash)[i], lc_hash,
5842 		nclp) {
5843 		LIST_FOREACH_SAFE(stp, &clp->lc_open, ls_list, nstp) {
5844 			if (LIST_EMPTY(&stp->ls_open) &&
5845 			    (stp->ls_noopens > NFSNOOPEN ||
5846 			     (nfsrv_openpluslock * 2) >
5847 			     nfsrv_v4statelimit))
5848 				nfsrv_freeopenowner(stp, 0, p);
5849 		}
5850 	    }
5851 	}
5852 	NFSUNLOCKSTATE();
5853 }
5854 
5855 /*
5856  * This function checks to see if the credentials are the same.
5857  * The check for same credentials is needed for state management operations
5858  * for NFSv4.0 or NFSv4.1/4.2 when SP4_MACH_CRED is configured via
5859  * ExchangeID.
5860  * Returns 1 for not same, 0 otherwise.
5861  */
5862 static int
5863 nfsrv_notsamecredname(int op, struct nfsrv_descript *nd, struct nfsclient *clp)
5864 {
5865 
5866 	/* Check for the SP4_MACH_CRED case. */
5867 	if (op != 0 && nfsrv_checkmachcred(op, nd, clp) != 0)
5868 		return (1);
5869 
5870 	/* For NFSv4.1/4.2, SP4_NONE always allows this. */
5871 	if ((nd->nd_flag & ND_NFSV41) != 0)
5872 		return (0);
5873 
5874 	if (nd->nd_flag & ND_GSS) {
5875 		if (!(clp->lc_flags & LCL_GSS))
5876 			return (1);
5877 		if (clp->lc_flags & LCL_NAME) {
5878 			if (nd->nd_princlen != clp->lc_namelen ||
5879 			    NFSBCMP(nd->nd_principal, clp->lc_name,
5880 				clp->lc_namelen))
5881 				return (1);
5882 			else
5883 				return (0);
5884 		}
5885 		if (nd->nd_cred->cr_uid == clp->lc_uid)
5886 			return (0);
5887 		else
5888 			return (1);
5889 	} else if (clp->lc_flags & LCL_GSS)
5890 		return (1);
5891 	/*
5892 	 * For AUTH_SYS, allow the same uid or root. (This is underspecified
5893 	 * in RFC3530, which talks about principals, but doesn't say anything
5894 	 * about uids for AUTH_SYS.)
5895 	 */
5896 	if (nd->nd_cred->cr_uid == clp->lc_uid || nd->nd_cred->cr_uid == 0)
5897 		return (0);
5898 	else
5899 		return (1);
5900 }
5901 
5902 /*
5903  * Calculate the lease expiry time.
5904  */
5905 static time_t
5906 nfsrv_leaseexpiry(void)
5907 {
5908 
5909 	if (NFSD_VNET(nfsrv_stablefirst).nsf_eograce > NFSD_MONOSEC)
5910 		return (NFSD_MONOSEC + 2 * (nfsrv_lease + NFSRV_LEASEDELTA));
5911 	return (NFSD_MONOSEC + nfsrv_lease + NFSRV_LEASEDELTA);
5912 }
5913 
5914 /*
5915  * Delay the delegation timeout as far as ls_delegtimelimit, as required.
5916  */
5917 static void
5918 nfsrv_delaydelegtimeout(struct nfsstate *stp)
5919 {
5920 
5921 	if ((stp->ls_flags & NFSLCK_DELEGRECALL) == 0)
5922 		return;
5923 
5924 	if ((stp->ls_delegtime + 15) > NFSD_MONOSEC &&
5925 	    stp->ls_delegtime < stp->ls_delegtimelimit) {
5926 		stp->ls_delegtime += nfsrv_lease;
5927 		if (stp->ls_delegtime > stp->ls_delegtimelimit)
5928 			stp->ls_delegtime = stp->ls_delegtimelimit;
5929 	}
5930 }
5931 
5932 /*
5933  * This function checks to see if there is any other state associated
5934  * with the openowner for this Open.
5935  * It returns 1 if there is no other state, 0 otherwise.
5936  */
5937 static int
5938 nfsrv_nootherstate(struct nfsstate *stp)
5939 {
5940 	struct nfsstate *tstp;
5941 
5942 	LIST_FOREACH(tstp, &stp->ls_openowner->ls_open, ls_list) {
5943 		if (tstp != stp || !LIST_EMPTY(&tstp->ls_lock))
5944 			return (0);
5945 	}
5946 	return (1);
5947 }
5948 
5949 /*
5950  * Create a list of lock deltas (changes to local byte range locking
5951  * that can be rolled back using the list) and apply the changes via
5952  * nfsvno_advlock(). Optionally, lock the list. It is expected that either
5953  * the rollback or update function will be called after this.
5954  * It returns an error (and rolls back, as required), if any nfsvno_advlock()
5955  * call fails. If it returns an error, it will unlock the list.
5956  */
5957 static int
5958 nfsrv_locallock(vnode_t vp, struct nfslockfile *lfp, int flags,
5959     uint64_t first, uint64_t end, struct nfslockconflict *cfp, NFSPROC_T *p)
5960 {
5961 	struct nfslock *lop, *nlop;
5962 	int error = 0;
5963 
5964 	/* Loop through the list of locks. */
5965 	lop = LIST_FIRST(&lfp->lf_locallock);
5966 	while (first < end && lop != NULL) {
5967 		nlop = LIST_NEXT(lop, lo_lckowner);
5968 		if (first >= lop->lo_end) {
5969 			/* not there yet */
5970 			lop = nlop;
5971 		} else if (first < lop->lo_first) {
5972 			/* new one starts before entry in list */
5973 			if (end <= lop->lo_first) {
5974 				/* no overlap between old and new */
5975 				error = nfsrv_dolocal(vp, lfp, flags,
5976 				    NFSLCK_UNLOCK, first, end, cfp, p);
5977 				if (error != 0)
5978 					break;
5979 				first = end;
5980 			} else {
5981 				/* handle fragment overlapped with new one */
5982 				error = nfsrv_dolocal(vp, lfp, flags,
5983 				    NFSLCK_UNLOCK, first, lop->lo_first, cfp,
5984 				    p);
5985 				if (error != 0)
5986 					break;
5987 				first = lop->lo_first;
5988 			}
5989 		} else {
5990 			/* new one overlaps this entry in list */
5991 			if (end <= lop->lo_end) {
5992 				/* overlaps all of new one */
5993 				error = nfsrv_dolocal(vp, lfp, flags,
5994 				    lop->lo_flags, first, end, cfp, p);
5995 				if (error != 0)
5996 					break;
5997 				first = end;
5998 			} else {
5999 				/* handle fragment overlapped with new one */
6000 				error = nfsrv_dolocal(vp, lfp, flags,
6001 				    lop->lo_flags, first, lop->lo_end, cfp, p);
6002 				if (error != 0)
6003 					break;
6004 				first = lop->lo_end;
6005 				lop = nlop;
6006 			}
6007 		}
6008 	}
6009 	if (first < end && error == 0)
6010 		/* handle fragment past end of list */
6011 		error = nfsrv_dolocal(vp, lfp, flags, NFSLCK_UNLOCK, first,
6012 		    end, cfp, p);
6013 
6014 	NFSEXITCODE(error);
6015 	return (error);
6016 }
6017 
6018 /*
6019  * Local lock unlock. Unlock all byte ranges that are no longer locked
6020  * by NFSv4. To do this, unlock any subranges of first-->end that
6021  * do not overlap with the byte ranges of any lock in the lfp->lf_lock
6022  * list. This list has all locks for the file held by other
6023  * <clientid, lockowner> tuples. The list is ordered by increasing
6024  * lo_first value, but may have entries that overlap each other, for
6025  * the case of read locks.
6026  */
6027 static void
6028 nfsrv_localunlock(vnode_t vp, struct nfslockfile *lfp, uint64_t init_first,
6029     uint64_t init_end, NFSPROC_T *p)
6030 {
6031 	struct nfslock *lop;
6032 	uint64_t first, end, prevfirst __unused;
6033 
6034 	first = init_first;
6035 	end = init_end;
6036 	while (first < init_end) {
6037 		/* Loop through all nfs locks, adjusting first and end */
6038 		prevfirst = 0;
6039 		LIST_FOREACH(lop, &lfp->lf_lock, lo_lckfile) {
6040 			KASSERT(prevfirst <= lop->lo_first,
6041 			    ("nfsv4 locks out of order"));
6042 			KASSERT(lop->lo_first < lop->lo_end,
6043 			    ("nfsv4 bogus lock"));
6044 			prevfirst = lop->lo_first;
6045 			if (first >= lop->lo_first &&
6046 			    first < lop->lo_end)
6047 				/*
6048 				 * Overlaps with initial part, so trim
6049 				 * off that initial part by moving first past
6050 				 * it.
6051 				 */
6052 				first = lop->lo_end;
6053 			else if (end > lop->lo_first &&
6054 			    lop->lo_first > first) {
6055 				/*
6056 				 * This lock defines the end of the
6057 				 * segment to unlock, so set end to the
6058 				 * start of it and break out of the loop.
6059 				 */
6060 				end = lop->lo_first;
6061 				break;
6062 			}
6063 			if (first >= end)
6064 				/*
6065 				 * There is no segment left to do, so
6066 				 * break out of this loop and then exit
6067 				 * the outer while() since first will be set
6068 				 * to end, which must equal init_end here.
6069 				 */
6070 				break;
6071 		}
6072 		if (first < end) {
6073 			/* Unlock this segment */
6074 			(void) nfsrv_dolocal(vp, lfp, NFSLCK_UNLOCK,
6075 			    NFSLCK_READ, first, end, NULL, p);
6076 			nfsrv_locallock_commit(lfp, NFSLCK_UNLOCK,
6077 			    first, end);
6078 		}
6079 		/*
6080 		 * Now move past this segment and look for any further
6081 		 * segment in the range, if there is one.
6082 		 */
6083 		first = end;
6084 		end = init_end;
6085 	}
6086 }
6087 
6088 /*
6089  * Do the local lock operation and update the rollback list, as required.
6090  * Perform the rollback and return the error if nfsvno_advlock() fails.
6091  */
6092 static int
6093 nfsrv_dolocal(vnode_t vp, struct nfslockfile *lfp, int flags, int oldflags,
6094     uint64_t first, uint64_t end, struct nfslockconflict *cfp, NFSPROC_T *p)
6095 {
6096 	struct nfsrollback *rlp;
6097 	int error = 0, ltype, oldltype;
6098 
6099 	if (flags & NFSLCK_WRITE)
6100 		ltype = F_WRLCK;
6101 	else if (flags & NFSLCK_READ)
6102 		ltype = F_RDLCK;
6103 	else
6104 		ltype = F_UNLCK;
6105 	if (oldflags & NFSLCK_WRITE)
6106 		oldltype = F_WRLCK;
6107 	else if (oldflags & NFSLCK_READ)
6108 		oldltype = F_RDLCK;
6109 	else
6110 		oldltype = F_UNLCK;
6111 	if (ltype == oldltype || (oldltype == F_WRLCK && ltype == F_RDLCK))
6112 		/* nothing to do */
6113 		goto out;
6114 	error = nfsvno_advlock(vp, ltype, first, end, p);
6115 	if (error != 0) {
6116 		if (cfp != NULL) {
6117 			cfp->cl_clientid.lval[0] = 0;
6118 			cfp->cl_clientid.lval[1] = 0;
6119 			cfp->cl_first = 0;
6120 			cfp->cl_end = NFS64BITSSET;
6121 			cfp->cl_flags = NFSLCK_WRITE;
6122 			cfp->cl_ownerlen = 5;
6123 			NFSBCOPY("LOCAL", cfp->cl_owner, 5);
6124 		}
6125 		nfsrv_locallock_rollback(vp, lfp, p);
6126 	} else if (ltype != F_UNLCK) {
6127 		rlp = malloc(sizeof (struct nfsrollback), M_NFSDROLLBACK,
6128 		    M_WAITOK);
6129 		rlp->rlck_first = first;
6130 		rlp->rlck_end = end;
6131 		rlp->rlck_type = oldltype;
6132 		LIST_INSERT_HEAD(&lfp->lf_rollback, rlp, rlck_list);
6133 	}
6134 
6135 out:
6136 	NFSEXITCODE(error);
6137 	return (error);
6138 }
6139 
6140 /*
6141  * Roll back local lock changes and free up the rollback list.
6142  */
6143 static void
6144 nfsrv_locallock_rollback(vnode_t vp, struct nfslockfile *lfp, NFSPROC_T *p)
6145 {
6146 	struct nfsrollback *rlp, *nrlp;
6147 
6148 	LIST_FOREACH_SAFE(rlp, &lfp->lf_rollback, rlck_list, nrlp) {
6149 		(void) nfsvno_advlock(vp, rlp->rlck_type, rlp->rlck_first,
6150 		    rlp->rlck_end, p);
6151 		free(rlp, M_NFSDROLLBACK);
6152 	}
6153 	LIST_INIT(&lfp->lf_rollback);
6154 }
6155 
6156 /*
6157  * Update local lock list and delete rollback list (ie now committed to the
6158  * local locks). Most of the work is done by the internal function.
6159  */
6160 static void
6161 nfsrv_locallock_commit(struct nfslockfile *lfp, int flags, uint64_t first,
6162     uint64_t end)
6163 {
6164 	struct nfsrollback *rlp, *nrlp;
6165 	struct nfslock *new_lop, *other_lop;
6166 
6167 	new_lop = malloc(sizeof (struct nfslock), M_NFSDLOCK, M_WAITOK);
6168 	if (flags & (NFSLCK_READ | NFSLCK_WRITE))
6169 		other_lop = malloc(sizeof (struct nfslock), M_NFSDLOCK,
6170 		    M_WAITOK);
6171 	else
6172 		other_lop = NULL;
6173 	new_lop->lo_flags = flags;
6174 	new_lop->lo_first = first;
6175 	new_lop->lo_end = end;
6176 	nfsrv_updatelock(NULL, &new_lop, &other_lop, lfp);
6177 	if (new_lop != NULL)
6178 		free(new_lop, M_NFSDLOCK);
6179 	if (other_lop != NULL)
6180 		free(other_lop, M_NFSDLOCK);
6181 
6182 	/* and get rid of the rollback list */
6183 	LIST_FOREACH_SAFE(rlp, &lfp->lf_rollback, rlck_list, nrlp)
6184 		free(rlp, M_NFSDROLLBACK);
6185 	LIST_INIT(&lfp->lf_rollback);
6186 }
6187 
6188 /*
6189  * Lock the struct nfslockfile for local lock updating.
6190  */
6191 static void
6192 nfsrv_locklf(struct nfslockfile *lfp)
6193 {
6194 	int gotlock;
6195 
6196 	/* lf_usecount ensures *lfp won't be free'd */
6197 	lfp->lf_usecount++;
6198 	do {
6199 		gotlock = nfsv4_lock(&lfp->lf_locallock_lck, 1, NULL,
6200 		    NFSSTATEMUTEXPTR, NULL);
6201 	} while (gotlock == 0);
6202 	lfp->lf_usecount--;
6203 }
6204 
6205 /*
6206  * Unlock the struct nfslockfile after local lock updating.
6207  */
6208 static void
6209 nfsrv_unlocklf(struct nfslockfile *lfp)
6210 {
6211 
6212 	nfsv4_unlock(&lfp->lf_locallock_lck, 0);
6213 }
6214 
6215 /*
6216  * Clear out all state for the NFSv4 server.
6217  * Must be called by a thread that can sleep when no nfsds are running.
6218  */
6219 void
6220 nfsrv_throwawayallstate(NFSPROC_T *p)
6221 {
6222 	struct nfsclient *clp, *nclp;
6223 	struct nfslockfile *lfp, *nlfp;
6224 	int i;
6225 
6226 	/*
6227 	 * For each client, clean out the state and then free the structure.
6228 	 */
6229 	for (i = 0; i < nfsrv_clienthashsize; i++) {
6230 		LIST_FOREACH_SAFE(clp, &NFSD_VNET(nfsclienthash)[i], lc_hash,
6231 		    nclp) {
6232 			nfsrv_cleanclient(clp, p, false, NULL);
6233 			nfsrv_freedeleglist(&clp->lc_deleg);
6234 			nfsrv_freedeleglist(&clp->lc_olddeleg);
6235 			free(clp->lc_stateid, M_NFSDCLIENT);
6236 			free(clp, M_NFSDCLIENT);
6237 		}
6238 	}
6239 
6240 	/*
6241 	 * Also, free up any remaining lock file structures.
6242 	 */
6243 	for (i = 0; i < nfsrv_lockhashsize; i++) {
6244 		LIST_FOREACH_SAFE(lfp, &NFSD_VNET(nfslockhash)[i], lf_hash,
6245 		    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 	/* Handle the SP4_MECH_CRED case for NFSv4.1/4.2. */
6289 	if ((sep->sess_clp->lc_flags & LCL_MACHCRED) != 0 &&
6290 	    (nd->nd_flag & (ND_GSSINTEGRITY | ND_GSSPRIVACY)) != 0 &&
6291 	    nd->nd_princlen == sep->sess_clp->lc_namelen &&
6292 	    !NFSBCMP(sep->sess_clp->lc_name, nd->nd_principal,
6293 	    nd->nd_princlen)) {
6294 		nd->nd_flag |= ND_MACHCRED;
6295 		NFSSET_OPBIT(&nd->nd_allowops, &sep->sess_clp->lc_allowops);
6296 	}
6297 
6298 	/* Save maximum request and reply sizes. */
6299 	nd->nd_maxreq = sep->sess_maxreq;
6300 	nd->nd_maxresp = sep->sess_maxresp;
6301 
6302 	*sflagsp = 0;
6303 	if (sep->sess_clp->lc_req.nr_client == NULL ||
6304 	    (sep->sess_clp->lc_flags & LCL_CBDOWN) != 0)
6305 		*sflagsp |= NFSV4SEQ_CBPATHDOWN;
6306 	NFSUNLOCKSESSION(shp);
6307 	if (error == NFSERR_EXPIRED) {
6308 		*sflagsp |= NFSV4SEQ_EXPIREDALLSTATEREVOKED;
6309 		error = 0;
6310 	} else if (error == NFSERR_ADMINREVOKED) {
6311 		*sflagsp |= NFSV4SEQ_ADMINSTATEREVOKED;
6312 		error = 0;
6313 	}
6314 	*highest_slotidp = *target_highest_slotidp = NFSV4_SLOTS - 1;
6315 	return (0);
6316 }
6317 
6318 /*
6319  * Check/set reclaim complete for this session/clientid.
6320  */
6321 int
6322 nfsrv_checkreclaimcomplete(struct nfsrv_descript *nd, int onefs)
6323 {
6324 	struct nfsdsession *sep;
6325 	struct nfssessionhash *shp;
6326 	int error = 0;
6327 
6328 	shp = NFSSESSIONHASH(nd->nd_sessionid);
6329 	NFSLOCKSTATE();
6330 	NFSLOCKSESSION(shp);
6331 	sep = nfsrv_findsession(nd->nd_sessionid);
6332 	if (sep == NULL) {
6333 		NFSUNLOCKSESSION(shp);
6334 		NFSUNLOCKSTATE();
6335 		return (NFSERR_BADSESSION);
6336 	}
6337 
6338 	if (onefs != 0)
6339 		sep->sess_clp->lc_flags |= LCL_RECLAIMONEFS;
6340 		/* Check to see if reclaim complete has already happened. */
6341 	else if ((sep->sess_clp->lc_flags & LCL_RECLAIMCOMPLETE) != 0)
6342 		error = NFSERR_COMPLETEALREADY;
6343 	else {
6344 		sep->sess_clp->lc_flags |= LCL_RECLAIMCOMPLETE;
6345 		nfsrv_markreclaim(sep->sess_clp);
6346 	}
6347 	NFSUNLOCKSESSION(shp);
6348 	NFSUNLOCKSTATE();
6349 	return (error);
6350 }
6351 
6352 /*
6353  * Cache the reply in a session slot.
6354  */
6355 void
6356 nfsrv_cache_session(struct nfsrv_descript *nd, struct mbuf **m)
6357 {
6358 	struct nfsdsession *sep;
6359 	struct nfssessionhash *shp;
6360 	char *buf, *cp;
6361 #ifdef INET
6362 	struct sockaddr_in *sin;
6363 #endif
6364 #ifdef INET6
6365 	struct sockaddr_in6 *sin6;
6366 #endif
6367 
6368 	shp = NFSSESSIONHASH(nd->nd_sessionid);
6369 	NFSLOCKSESSION(shp);
6370 	sep = nfsrv_findsession(nd->nd_sessionid);
6371 	if (sep == NULL) {
6372 		NFSUNLOCKSESSION(shp);
6373 		if ((NFSD_VNET(nfsrv_stablefirst).nsf_flags &
6374 		     NFSNSF_GRACEOVER) != 0) {
6375 			buf = malloc(INET6_ADDRSTRLEN, M_TEMP, M_WAITOK);
6376 			switch (nd->nd_nam->sa_family) {
6377 #ifdef INET
6378 			case AF_INET:
6379 				sin = (struct sockaddr_in *)nd->nd_nam;
6380 				cp = inet_ntop(sin->sin_family,
6381 				    &sin->sin_addr.s_addr, buf,
6382 				    INET6_ADDRSTRLEN);
6383 				break;
6384 #endif
6385 #ifdef INET6
6386 			case AF_INET6:
6387 				sin6 = (struct sockaddr_in6 *)nd->nd_nam;
6388 				cp = inet_ntop(sin6->sin6_family,
6389 				    &sin6->sin6_addr, buf, INET6_ADDRSTRLEN);
6390 				break;
6391 #endif
6392 			default:
6393 				cp = NULL;
6394 			}
6395 			if (cp != NULL)
6396 				printf("nfsrv_cache_session: no session "
6397 				    "IPaddr=%s, check NFS clients for unique "
6398 				    "/etc/hostid's\n", cp);
6399 			else
6400 				printf("nfsrv_cache_session: no session, "
6401 				    "check NFS clients for unique "
6402 				    "/etc/hostid's\n");
6403 			free(buf, M_TEMP);
6404 		}
6405 		m_freem(*m);
6406 		return;
6407 	}
6408 	nfsv4_seqsess_cacherep(nd->nd_slotid, sep->sess_slots, nd->nd_repstat,
6409 	    m);
6410 	NFSUNLOCKSESSION(shp);
6411 }
6412 
6413 /*
6414  * Search for a session that matches the sessionid.
6415  */
6416 static struct nfsdsession *
6417 nfsrv_findsession(uint8_t *sessionid)
6418 {
6419 	struct nfsdsession *sep;
6420 	struct nfssessionhash *shp;
6421 
6422 	shp = NFSSESSIONHASH(sessionid);
6423 	LIST_FOREACH(sep, &shp->list, sess_hash) {
6424 		if (!NFSBCMP(sessionid, sep->sess_sessionid, NFSX_V4SESSIONID))
6425 			break;
6426 	}
6427 	return (sep);
6428 }
6429 
6430 /*
6431  * Destroy a session.
6432  */
6433 int
6434 nfsrv_destroysession(struct nfsrv_descript *nd, uint8_t *sessionid)
6435 {
6436 	int error, igotlock, samesess;
6437 
6438 	samesess = 0;
6439 	if (!NFSBCMP(sessionid, nd->nd_sessionid, NFSX_V4SESSIONID) &&
6440 	    (nd->nd_flag & ND_HASSEQUENCE) != 0) {
6441 		samesess = 1;
6442 		if ((nd->nd_flag & ND_LASTOP) == 0)
6443 			return (NFSERR_BADSESSION);
6444 	}
6445 
6446 	/* Lock out other nfsd threads */
6447 	NFSLOCKV4ROOTMUTEX();
6448 	nfsv4_relref(&nfsv4rootfs_lock);
6449 	do {
6450 		igotlock = nfsv4_lock(&nfsv4rootfs_lock, 1, NULL,
6451 		    NFSV4ROOTLOCKMUTEXPTR, NULL);
6452 	} while (igotlock == 0);
6453 	NFSUNLOCKV4ROOTMUTEX();
6454 
6455 	error = nfsrv_freesession(nd, NULL, sessionid, false, NULL);
6456 	if (error == 0 && samesess != 0)
6457 		nd->nd_flag &= ~ND_HASSEQUENCE;
6458 
6459 	NFSLOCKV4ROOTMUTEX();
6460 	nfsv4_unlock(&nfsv4rootfs_lock, 1);
6461 	NFSUNLOCKV4ROOTMUTEX();
6462 	return (error);
6463 }
6464 
6465 /*
6466  * Bind a connection to a session.
6467  * For now, only certain variants are supported, since the current session
6468  * structure can only handle a single backchannel entry, which will be
6469  * applied to all connections if it is set.
6470  */
6471 int
6472 nfsrv_bindconnsess(struct nfsrv_descript *nd, uint8_t *sessionid, int *foreaftp)
6473 {
6474 	struct nfssessionhash *shp;
6475 	struct nfsdsession *sep;
6476 	struct nfsclient *clp;
6477 	SVCXPRT *savxprt;
6478 	int error;
6479 
6480 	error = 0;
6481 	savxprt = NULL;
6482 	shp = NFSSESSIONHASH(sessionid);
6483 	NFSLOCKSTATE();
6484 	NFSLOCKSESSION(shp);
6485 	sep = nfsrv_findsession(sessionid);
6486 	if (sep != NULL) {
6487 		clp = sep->sess_clp;
6488 		error = nfsrv_checkmachcred(NFSV4OP_BINDCONNTOSESS, nd, clp);
6489 		if (error != 0)
6490 			goto out;
6491 		if (*foreaftp == NFSCDFC4_BACK ||
6492 		    *foreaftp == NFSCDFC4_BACK_OR_BOTH ||
6493 		    *foreaftp == NFSCDFC4_FORE_OR_BOTH) {
6494 			/* Try to set up a backchannel. */
6495 			if (clp->lc_req.nr_client == NULL) {
6496 				NFSD_DEBUG(2, "nfsrv_bindconnsess: acquire "
6497 				    "backchannel\n");
6498 				clp->lc_req.nr_client = (struct __rpc_client *)
6499 				    clnt_bck_create(nd->nd_xprt->xp_socket,
6500 				    sep->sess_cbprogram, NFSV4_CBVERS);
6501 			}
6502 			if (clp->lc_req.nr_client != NULL) {
6503 				NFSD_DEBUG(2, "nfsrv_bindconnsess: set up "
6504 				    "backchannel\n");
6505 				savxprt = sep->sess_cbsess.nfsess_xprt;
6506 				SVC_ACQUIRE(nd->nd_xprt);
6507 				CLNT_ACQUIRE(clp->lc_req.nr_client);
6508 				nd->nd_xprt->xp_p2 = clp->lc_req.nr_client;
6509 				/* Disable idle timeout. */
6510 				nd->nd_xprt->xp_idletimeout = 0;
6511 				sep->sess_cbsess.nfsess_xprt = nd->nd_xprt;
6512 				sep->sess_crflags |= NFSV4CRSESS_CONNBACKCHAN;
6513 				clp->lc_flags |= LCL_DONEBINDCONN |
6514 				    LCL_NEEDSCBNULL;
6515 				clp->lc_flags &= ~LCL_CBDOWN;
6516 				if (*foreaftp == NFSCDFS4_BACK)
6517 					*foreaftp = NFSCDFS4_BACK;
6518 				else
6519 					*foreaftp = NFSCDFS4_BOTH;
6520 			} else if (*foreaftp != NFSCDFC4_BACK) {
6521 				NFSD_DEBUG(2, "nfsrv_bindconnsess: can't set "
6522 				    "up backchannel\n");
6523 				sep->sess_crflags &= ~NFSV4CRSESS_CONNBACKCHAN;
6524 				clp->lc_flags |= LCL_DONEBINDCONN;
6525 				*foreaftp = NFSCDFS4_FORE;
6526 			} else {
6527 				error = NFSERR_NOTSUPP;
6528 				printf("nfsrv_bindconnsess: Can't add "
6529 				    "backchannel\n");
6530 			}
6531 		} else {
6532 			NFSD_DEBUG(2, "nfsrv_bindconnsess: Set forechannel\n");
6533 			clp->lc_flags |= LCL_DONEBINDCONN;
6534 			*foreaftp = NFSCDFS4_FORE;
6535 		}
6536 	} else
6537 		error = NFSERR_BADSESSION;
6538 out:
6539 	NFSUNLOCKSESSION(shp);
6540 	NFSUNLOCKSTATE();
6541 	if (savxprt != NULL)
6542 		SVC_RELEASE(savxprt);
6543 	return (error);
6544 }
6545 
6546 /*
6547  * Free up a session structure.
6548  */
6549 static int
6550 nfsrv_freesession(struct nfsrv_descript *nd, struct nfsdsession *sep,
6551     uint8_t *sessionid, bool locked, SVCXPRT **old_xprtp)
6552 {
6553 	struct nfssessionhash *shp;
6554 	int i;
6555 
6556 	if (!locked)
6557 		NFSLOCKSTATE();
6558 	if (sep == NULL) {
6559 		shp = NFSSESSIONHASH(sessionid);
6560 		NFSLOCKSESSION(shp);
6561 		sep = nfsrv_findsession(sessionid);
6562 	} else {
6563 		shp = NFSSESSIONHASH(sep->sess_sessionid);
6564 		NFSLOCKSESSION(shp);
6565 	}
6566 	if (sep != NULL) {
6567 		/* Check for the SP4_MACH_CRED case. */
6568 		if (nd != NULL && nfsrv_checkmachcred(NFSV4OP_DESTROYSESSION,
6569 		    nd, sep->sess_clp) != 0) {
6570 			NFSUNLOCKSESSION(shp);
6571 			if (!locked)
6572 				NFSUNLOCKSTATE();
6573 			return (NFSERR_AUTHERR | AUTH_TOOWEAK);
6574 		}
6575 
6576 		sep->sess_refcnt--;
6577 		if (sep->sess_refcnt > 0) {
6578 			NFSUNLOCKSESSION(shp);
6579 			if (!locked)
6580 				NFSUNLOCKSTATE();
6581 			return (NFSERR_BACKCHANBUSY);
6582 		}
6583 		LIST_REMOVE(sep, sess_hash);
6584 		LIST_REMOVE(sep, sess_list);
6585 	}
6586 	NFSUNLOCKSESSION(shp);
6587 	if (!locked)
6588 		NFSUNLOCKSTATE();
6589 	if (sep == NULL)
6590 		return (NFSERR_BADSESSION);
6591 	for (i = 0; i < NFSV4_SLOTS; i++)
6592 		if (sep->sess_slots[i].nfssl_reply != NULL)
6593 			m_freem(sep->sess_slots[i].nfssl_reply);
6594 	if (!locked) {
6595 		if (sep->sess_cbsess.nfsess_xprt != NULL)
6596 			SVC_RELEASE(sep->sess_cbsess.nfsess_xprt);
6597 		if (old_xprtp != NULL)
6598 			*old_xprtp = NULL;
6599 	} else if (old_xprtp != NULL)
6600 		*old_xprtp = sep->sess_cbsess.nfsess_xprt;
6601 	free(sep, M_NFSDSESSION);
6602 	return (0);
6603 }
6604 
6605 /*
6606  * Free a stateid.
6607  * RFC5661 says that it should fail when there are associated opens, locks
6608  * or delegations. Since stateids represent opens, I don't see how you can
6609  * free an open stateid (it will be free'd when closed), so this function
6610  * only works for lock stateids (freeing the lock_owner) or delegations.
6611  */
6612 int
6613 nfsrv_freestateid(struct nfsrv_descript *nd, nfsv4stateid_t *stateidp,
6614     NFSPROC_T *p)
6615 {
6616 	struct nfsclient *clp;
6617 	struct nfsstate *stp;
6618 	int error;
6619 
6620 	NFSLOCKSTATE();
6621 	/*
6622 	 * Look up the stateid
6623 	 */
6624 	error = nfsrv_getclient((nfsquad_t)((u_quad_t)0), CLOPS_RENEW, &clp,
6625 	    NULL, (nfsquad_t)((u_quad_t)0), 0, nd, p);
6626 	if (error == 0) {
6627 		/* First, check for a delegation. */
6628 		LIST_FOREACH(stp, &clp->lc_deleg, ls_list) {
6629 			if (!NFSBCMP(stp->ls_stateid.other, stateidp->other,
6630 			    NFSX_STATEIDOTHER))
6631 				break;
6632 		}
6633 		if (stp != NULL) {
6634 			nfsrv_freedeleg(stp);
6635 			NFSUNLOCKSTATE();
6636 			return (error);
6637 		}
6638 	}
6639 	/* Not a delegation, try for a lock_owner. */
6640 	if (error == 0)
6641 		error = nfsrv_getstate(clp, stateidp, 0, &stp);
6642 	if (error == 0 && ((stp->ls_flags & (NFSLCK_OPEN | NFSLCK_DELEGREAD |
6643 	    NFSLCK_DELEGWRITE)) != 0 || (stp->ls_flags & NFSLCK_LOCK) == 0))
6644 		/* Not a lock_owner stateid. */
6645 		error = NFSERR_LOCKSHELD;
6646 	if (error == 0 && !LIST_EMPTY(&stp->ls_lock))
6647 		error = NFSERR_LOCKSHELD;
6648 	if (error == 0)
6649 		nfsrv_freelockowner(stp, NULL, 0, p);
6650 	NFSUNLOCKSTATE();
6651 	return (error);
6652 }
6653 
6654 /*
6655  * Test a stateid.
6656  */
6657 int
6658 nfsrv_teststateid(struct nfsrv_descript *nd, nfsv4stateid_t *stateidp,
6659     NFSPROC_T *p)
6660 {
6661 	struct nfsclient *clp;
6662 	struct nfsstate *stp;
6663 	int error;
6664 
6665 	NFSLOCKSTATE();
6666 	/*
6667 	 * Look up the stateid
6668 	 */
6669 	error = nfsrv_getclient((nfsquad_t)((u_quad_t)0), CLOPS_RENEW, &clp,
6670 	    NULL, (nfsquad_t)((u_quad_t)0), 0, nd, p);
6671 	if (error == 0)
6672 		error = nfsrv_getstate(clp, stateidp, 0, &stp);
6673 	if (error == 0 && stateidp->seqid != 0 &&
6674 	    SEQ_LT(stateidp->seqid, stp->ls_stateid.seqid))
6675 		error = NFSERR_OLDSTATEID;
6676 	NFSUNLOCKSTATE();
6677 	return (error);
6678 }
6679 
6680 /*
6681  * Generate the xdr for an NFSv4.1 CBSequence Operation.
6682  */
6683 static int
6684 nfsv4_setcbsequence(struct nfsrv_descript *nd, struct nfsclient *clp,
6685     int dont_replycache, struct nfsdsession **sepp, int *slotposp)
6686 {
6687 	struct nfsdsession *sep;
6688 	uint32_t *tl, slotseq = 0;
6689 	int maxslot;
6690 	uint8_t sessionid[NFSX_V4SESSIONID];
6691 	int error;
6692 
6693 	error = nfsv4_getcbsession(clp, sepp);
6694 	if (error != 0)
6695 		return (error);
6696 	sep = *sepp;
6697 	nfsv4_sequencelookup(NULL, &sep->sess_cbsess, slotposp, &maxslot,
6698 	    &slotseq, sessionid, true);
6699 	KASSERT(maxslot >= 0, ("nfsv4_setcbsequence neg maxslot"));
6700 
6701 	/* Build the Sequence arguments. */
6702 	NFSM_BUILD(tl, uint32_t *, NFSX_V4SESSIONID + 5 * NFSX_UNSIGNED);
6703 	bcopy(sessionid, tl, NFSX_V4SESSIONID);
6704 	tl += NFSX_V4SESSIONID / NFSX_UNSIGNED;
6705 	nd->nd_slotseq = tl;
6706 	nd->nd_slotid = *slotposp;
6707 	nd->nd_flag |= ND_HASSLOTID;
6708 	*tl++ = txdr_unsigned(slotseq);
6709 	*tl++ = txdr_unsigned(*slotposp);
6710 	*tl++ = txdr_unsigned(maxslot);
6711 	if (dont_replycache == 0)
6712 		*tl++ = newnfs_true;
6713 	else
6714 		*tl++ = newnfs_false;
6715 	*tl = 0;			/* No referring call list, for now. */
6716 	nd->nd_flag |= ND_HASSEQUENCE;
6717 	return (0);
6718 }
6719 
6720 /*
6721  * Get a session for the callback.
6722  */
6723 static int
6724 nfsv4_getcbsession(struct nfsclient *clp, struct nfsdsession **sepp)
6725 {
6726 	struct nfsdsession *sep;
6727 
6728 	NFSLOCKSTATE();
6729 	LIST_FOREACH(sep, &clp->lc_session, sess_list) {
6730 		if ((sep->sess_crflags & NFSV4CRSESS_CONNBACKCHAN) != 0)
6731 			break;
6732 	}
6733 	if (sep == NULL) {
6734 		NFSUNLOCKSTATE();
6735 		return (NFSERR_BADSESSION);
6736 	}
6737 	sep->sess_refcnt++;
6738 	*sepp = sep;
6739 	NFSUNLOCKSTATE();
6740 	return (0);
6741 }
6742 
6743 /*
6744  * Free up all backchannel xprts.  This needs to be done when the nfsd threads
6745  * exit, since those transports will all be going away.
6746  * This is only called after all the nfsd threads are done performing RPCs,
6747  * so locking shouldn't be an issue.
6748  */
6749 void
6750 nfsrv_freeallbackchannel_xprts(void)
6751 {
6752 	struct nfsdsession *sep;
6753 	struct nfsclient *clp;
6754 	SVCXPRT *xprt;
6755 	int i;
6756 
6757 	for (i = 0; i < nfsrv_clienthashsize; i++) {
6758 		LIST_FOREACH(clp, &NFSD_VNET(nfsclienthash)[i], lc_hash) {
6759 			LIST_FOREACH(sep, &clp->lc_session, sess_list) {
6760 				xprt = sep->sess_cbsess.nfsess_xprt;
6761 				sep->sess_cbsess.nfsess_xprt = NULL;
6762 				if (xprt != NULL)
6763 					SVC_RELEASE(xprt);
6764 			}
6765 		}
6766 	}
6767 }
6768 
6769 /*
6770  * Do a layout commit.  Actually just call nfsrv_updatemdsattr().
6771  * I have no idea if the rest of these arguments will ever be useful?
6772  */
6773 int
6774 nfsrv_layoutcommit(struct nfsrv_descript *nd, vnode_t vp, int layouttype,
6775     int hasnewoff, uint64_t newoff, uint64_t offset, uint64_t len,
6776     int hasnewmtime, struct timespec *newmtimep, int reclaim,
6777     nfsv4stateid_t *stateidp, int maxcnt, char *layp, int *hasnewsizep,
6778     uint64_t *newsizep, struct ucred *cred, NFSPROC_T *p)
6779 {
6780 	struct nfsvattr na;
6781 	int error;
6782 
6783 	error = nfsrv_updatemdsattr(vp, &na, p);
6784 	if (error == 0) {
6785 		*hasnewsizep = 1;
6786 		*newsizep = na.na_size;
6787 	}
6788 	return (error);
6789 }
6790 
6791 /*
6792  * Try and get a layout.
6793  */
6794 int
6795 nfsrv_layoutget(struct nfsrv_descript *nd, vnode_t vp, struct nfsexstuff *exp,
6796     int layouttype, int *iomode, uint64_t *offset, uint64_t *len,
6797     uint64_t minlen, nfsv4stateid_t *stateidp, int maxcnt, int *retonclose,
6798     int *layoutlenp, char *layp, struct ucred *cred, NFSPROC_T *p)
6799 {
6800 	struct nfslayouthash *lhyp;
6801 	struct nfslayout *lyp;
6802 	char *devid;
6803 	fhandle_t fh, *dsfhp;
6804 	int error, mirrorcnt;
6805 
6806 	if (nfsrv_devidcnt == 0)
6807 		return (NFSERR_UNKNLAYOUTTYPE);
6808 
6809 	if (*offset != 0)
6810 		printf("nfsrv_layoutget: off=%ju len=%ju\n", (uintmax_t)*offset,
6811 		    (uintmax_t)*len);
6812 	error = nfsvno_getfh(vp, &fh, p);
6813 	NFSD_DEBUG(4, "layoutget getfh=%d\n", error);
6814 	if (error != 0)
6815 		return (error);
6816 
6817 	/*
6818 	 * For now, all layouts are for entire files.
6819 	 * Only issue Read/Write layouts if requested for a non-readonly fs.
6820 	 */
6821 	if (NFSVNO_EXRDONLY(exp)) {
6822 		if (*iomode == NFSLAYOUTIOMODE_RW)
6823 			return (NFSERR_LAYOUTTRYLATER);
6824 		*iomode = NFSLAYOUTIOMODE_READ;
6825 	}
6826 	if (*iomode != NFSLAYOUTIOMODE_RW)
6827 		*iomode = NFSLAYOUTIOMODE_READ;
6828 
6829 	/*
6830 	 * Check to see if a write layout can be issued for this file.
6831 	 * This is used during mirror recovery to avoid RW layouts being
6832 	 * issued for a file while it is being copied to the recovered
6833 	 * mirror.
6834 	 */
6835 	if (*iomode == NFSLAYOUTIOMODE_RW && nfsrv_dontlayout(&fh) != 0)
6836 		return (NFSERR_LAYOUTTRYLATER);
6837 
6838 	*retonclose = 0;
6839 	*offset = 0;
6840 	*len = UINT64_MAX;
6841 
6842 	/* First, see if a layout already exists and return if found. */
6843 	lhyp = NFSLAYOUTHASH(&fh);
6844 	NFSLOCKLAYOUT(lhyp);
6845 	error = nfsrv_findlayout(&nd->nd_clientid, &fh, layouttype, p, &lyp);
6846 	NFSD_DEBUG(4, "layoutget findlay=%d\n", error);
6847 	/*
6848 	 * Not sure if the seqid must be the same, so I won't check it.
6849 	 */
6850 	if (error == 0 && (stateidp->other[0] != lyp->lay_stateid.other[0] ||
6851 	    stateidp->other[1] != lyp->lay_stateid.other[1] ||
6852 	    stateidp->other[2] != lyp->lay_stateid.other[2])) {
6853 		if ((lyp->lay_flags & NFSLAY_CALLB) == 0) {
6854 			NFSUNLOCKLAYOUT(lhyp);
6855 			NFSD_DEBUG(1, "ret bad stateid\n");
6856 			return (NFSERR_BADSTATEID);
6857 		}
6858 		/*
6859 		 * I believe we get here because there is a race between
6860 		 * the client processing the CBLAYOUTRECALL and the layout
6861 		 * being deleted here on the server.
6862 		 * The client has now done a LayoutGet with a non-layout
6863 		 * stateid, as it would when there is no layout.
6864 		 * As such, free this layout and set error == NFSERR_BADSTATEID
6865 		 * so the code below will create a new layout structure as
6866 		 * would happen if no layout was found.
6867 		 * "lyp" will be set before being used below, but set it NULL
6868 		 * as a safety belt.
6869 		 */
6870 		nfsrv_freelayout(&lhyp->list, lyp);
6871 		lyp = NULL;
6872 		error = NFSERR_BADSTATEID;
6873 	}
6874 	if (error == 0) {
6875 		if (lyp->lay_layoutlen > maxcnt) {
6876 			NFSUNLOCKLAYOUT(lhyp);
6877 			NFSD_DEBUG(1, "ret layout too small\n");
6878 			return (NFSERR_TOOSMALL);
6879 		}
6880 		if (*iomode == NFSLAYOUTIOMODE_RW) {
6881 			if ((lyp->lay_flags & NFSLAY_NOSPC) != 0) {
6882 				NFSUNLOCKLAYOUT(lhyp);
6883 				NFSD_DEBUG(1, "ret layout nospace\n");
6884 				return (NFSERR_NOSPC);
6885 			}
6886 			lyp->lay_flags |= NFSLAY_RW;
6887 		} else
6888 			lyp->lay_flags |= NFSLAY_READ;
6889 		NFSBCOPY(lyp->lay_xdr, layp, lyp->lay_layoutlen);
6890 		*layoutlenp = lyp->lay_layoutlen;
6891 		if (++lyp->lay_stateid.seqid == 0)
6892 			lyp->lay_stateid.seqid = 1;
6893 		stateidp->seqid = lyp->lay_stateid.seqid;
6894 		NFSUNLOCKLAYOUT(lhyp);
6895 		NFSD_DEBUG(4, "ret fnd layout\n");
6896 		return (0);
6897 	}
6898 	NFSUNLOCKLAYOUT(lhyp);
6899 
6900 	/* Find the device id and file handle. */
6901 	dsfhp = malloc(sizeof(fhandle_t) * NFSDEV_MAXMIRRORS, M_TEMP, M_WAITOK);
6902 	devid = malloc(NFSX_V4DEVICEID * NFSDEV_MAXMIRRORS, M_TEMP, M_WAITOK);
6903 	error = nfsrv_dsgetdevandfh(vp, p, &mirrorcnt, dsfhp, devid);
6904 	NFSD_DEBUG(4, "layoutget devandfh=%d\n", error);
6905 	if (error == 0) {
6906 		if (layouttype == NFSLAYOUT_NFSV4_1_FILES) {
6907 			if (NFSX_V4FILELAYOUT > maxcnt)
6908 				error = NFSERR_TOOSMALL;
6909 			else
6910 				lyp = nfsrv_filelayout(nd, *iomode, &fh, dsfhp,
6911 				    devid, vp->v_mount->mnt_stat.f_fsid);
6912 		} else {
6913 			if (NFSX_V4FLEXLAYOUT(mirrorcnt) > maxcnt)
6914 				error = NFSERR_TOOSMALL;
6915 			else
6916 				lyp = nfsrv_flexlayout(nd, *iomode, mirrorcnt,
6917 				    &fh, dsfhp, devid,
6918 				    vp->v_mount->mnt_stat.f_fsid);
6919 		}
6920 	}
6921 	free(dsfhp, M_TEMP);
6922 	free(devid, M_TEMP);
6923 	if (error != 0)
6924 		return (error);
6925 
6926 	/*
6927 	 * Now, add this layout to the list.
6928 	 */
6929 	error = nfsrv_addlayout(nd, &lyp, stateidp, layp, layoutlenp, p);
6930 	NFSD_DEBUG(4, "layoutget addl=%d\n", error);
6931 	/*
6932 	 * The lyp will be set to NULL by nfsrv_addlayout() if it
6933 	 * linked the new structure into the lists.
6934 	 */
6935 	free(lyp, M_NFSDSTATE);
6936 	return (error);
6937 }
6938 
6939 /*
6940  * Generate a File Layout.
6941  */
6942 static struct nfslayout *
6943 nfsrv_filelayout(struct nfsrv_descript *nd, int iomode, fhandle_t *fhp,
6944     fhandle_t *dsfhp, char *devid, fsid_t fs)
6945 {
6946 	uint32_t *tl;
6947 	struct nfslayout *lyp;
6948 	uint64_t pattern_offset;
6949 
6950 	lyp = malloc(sizeof(struct nfslayout) + NFSX_V4FILELAYOUT, M_NFSDSTATE,
6951 	    M_WAITOK | M_ZERO);
6952 	lyp->lay_type = NFSLAYOUT_NFSV4_1_FILES;
6953 	if (iomode == NFSLAYOUTIOMODE_RW)
6954 		lyp->lay_flags = NFSLAY_RW;
6955 	else
6956 		lyp->lay_flags = NFSLAY_READ;
6957 	NFSBCOPY(fhp, &lyp->lay_fh, sizeof(*fhp));
6958 	lyp->lay_clientid.qval = nd->nd_clientid.qval;
6959 	lyp->lay_fsid = fs;
6960 	NFSBCOPY(devid, lyp->lay_deviceid, NFSX_V4DEVICEID);
6961 
6962 	/* Fill in the xdr for the files layout. */
6963 	tl = (uint32_t *)lyp->lay_xdr;
6964 	NFSBCOPY(devid, tl, NFSX_V4DEVICEID);		/* Device ID. */
6965 	tl += (NFSX_V4DEVICEID / NFSX_UNSIGNED);
6966 
6967 	/* Set the stripe size to the maximum I/O size. */
6968 	*tl++ = txdr_unsigned(nfs_srvmaxio & NFSFLAYUTIL_STRIPE_MASK);
6969 	*tl++ = 0;					/* 1st stripe index. */
6970 	pattern_offset = 0;
6971 	txdr_hyper(pattern_offset, tl); tl += 2;	/* Pattern offset. */
6972 	*tl++ = txdr_unsigned(1);			/* 1 file handle. */
6973 	*tl++ = txdr_unsigned(NFSX_V4PNFSFH);
6974 	NFSBCOPY(dsfhp, tl, sizeof(*dsfhp));
6975 	lyp->lay_layoutlen = NFSX_V4FILELAYOUT;
6976 	return (lyp);
6977 }
6978 
6979 #define	FLEX_OWNERID	"999"
6980 #define	FLEX_UID0	"0"
6981 /*
6982  * Generate a Flex File Layout.
6983  * The FLEX_OWNERID can be any string of 3 decimal digits. Although this
6984  * string goes on the wire, it isn't supposed to be used by the client,
6985  * since this server uses tight coupling.
6986  * Although not recommended by the spec., if vfs.nfsd.flexlinuxhack=1 use
6987  * a string of "0". This works around the Linux Flex File Layout driver bug
6988  * which uses the synthetic uid/gid strings for the "tightly coupled" case.
6989  */
6990 static struct nfslayout *
6991 nfsrv_flexlayout(struct nfsrv_descript *nd, int iomode, int mirrorcnt,
6992     fhandle_t *fhp, fhandle_t *dsfhp, char *devid, fsid_t fs)
6993 {
6994 	uint32_t *tl;
6995 	struct nfslayout *lyp;
6996 	uint64_t lenval;
6997 	int i;
6998 
6999 	lyp = malloc(sizeof(struct nfslayout) + NFSX_V4FLEXLAYOUT(mirrorcnt),
7000 	    M_NFSDSTATE, M_WAITOK | M_ZERO);
7001 	lyp->lay_type = NFSLAYOUT_FLEXFILE;
7002 	if (iomode == NFSLAYOUTIOMODE_RW)
7003 		lyp->lay_flags = NFSLAY_RW;
7004 	else
7005 		lyp->lay_flags = NFSLAY_READ;
7006 	NFSBCOPY(fhp, &lyp->lay_fh, sizeof(*fhp));
7007 	lyp->lay_clientid.qval = nd->nd_clientid.qval;
7008 	lyp->lay_fsid = fs;
7009 	lyp->lay_mirrorcnt = mirrorcnt;
7010 	NFSBCOPY(devid, lyp->lay_deviceid, NFSX_V4DEVICEID);
7011 
7012 	/* Fill in the xdr for the files layout. */
7013 	tl = (uint32_t *)lyp->lay_xdr;
7014 	lenval = 0;
7015 	txdr_hyper(lenval, tl); tl += 2;		/* Stripe unit. */
7016 	*tl++ = txdr_unsigned(mirrorcnt);		/* # of mirrors. */
7017 	for (i = 0; i < mirrorcnt; i++) {
7018 		*tl++ = txdr_unsigned(1);		/* One stripe. */
7019 		NFSBCOPY(devid, tl, NFSX_V4DEVICEID);	/* Device ID. */
7020 		tl += (NFSX_V4DEVICEID / NFSX_UNSIGNED);
7021 		devid += NFSX_V4DEVICEID;
7022 		*tl++ = txdr_unsigned(1);		/* Efficiency. */
7023 		*tl++ = 0;				/* Proxy Stateid. */
7024 		*tl++ = 0x55555555;
7025 		*tl++ = 0x55555555;
7026 		*tl++ = 0x55555555;
7027 		*tl++ = txdr_unsigned(1);		/* 1 file handle. */
7028 		*tl++ = txdr_unsigned(NFSX_V4PNFSFH);
7029 		NFSBCOPY(dsfhp, tl, sizeof(*dsfhp));
7030 		tl += (NFSM_RNDUP(NFSX_V4PNFSFH) / NFSX_UNSIGNED);
7031 		dsfhp++;
7032 		if (nfsrv_flexlinuxhack != 0) {
7033 			*tl++ = txdr_unsigned(strlen(FLEX_UID0));
7034 			*tl = 0;		/* 0 pad string. */
7035 			NFSBCOPY(FLEX_UID0, tl++, strlen(FLEX_UID0));
7036 			*tl++ = txdr_unsigned(strlen(FLEX_UID0));
7037 			*tl = 0;		/* 0 pad string. */
7038 			NFSBCOPY(FLEX_UID0, tl++, strlen(FLEX_UID0));
7039 		} else {
7040 			*tl++ = txdr_unsigned(strlen(FLEX_OWNERID));
7041 			NFSBCOPY(FLEX_OWNERID, tl++, NFSX_UNSIGNED);
7042 			*tl++ = txdr_unsigned(strlen(FLEX_OWNERID));
7043 			NFSBCOPY(FLEX_OWNERID, tl++, NFSX_UNSIGNED);
7044 		}
7045 	}
7046 	*tl++ = txdr_unsigned(0);		/* ff_flags. */
7047 	*tl = txdr_unsigned(60);		/* Status interval hint. */
7048 	lyp->lay_layoutlen = NFSX_V4FLEXLAYOUT(mirrorcnt);
7049 	return (lyp);
7050 }
7051 
7052 /*
7053  * Parse and process Flex File errors returned via LayoutReturn.
7054  */
7055 static void
7056 nfsrv_flexlayouterr(struct nfsrv_descript *nd, uint32_t *layp, int maxcnt,
7057     NFSPROC_T *p)
7058 {
7059 	uint32_t *tl;
7060 	int cnt, errcnt, i, j, opnum, stat;
7061 	char devid[NFSX_V4DEVICEID];
7062 
7063 	tl = layp;
7064 	maxcnt -= NFSX_UNSIGNED;
7065 	if (maxcnt > 0)
7066 		cnt = fxdr_unsigned(int, *tl++);
7067 	else
7068 		cnt = 0;
7069 	NFSD_DEBUG(4, "flexlayouterr cnt=%d\n", cnt);
7070 	for (i = 0; i < cnt; i++) {
7071 		maxcnt -= NFSX_STATEID + 2 * NFSX_HYPER +
7072 		    NFSX_UNSIGNED;
7073 		if (maxcnt <= 0)
7074 			break;
7075 		/* Skip offset, length and stateid for now. */
7076 		tl += (4 + NFSX_STATEID / NFSX_UNSIGNED);
7077 		errcnt = fxdr_unsigned(int, *tl++);
7078 		NFSD_DEBUG(4, "flexlayouterr errcnt=%d\n", errcnt);
7079 		for (j = 0; j < errcnt; j++) {
7080 			maxcnt -= NFSX_V4DEVICEID + 2 * NFSX_UNSIGNED;
7081 			if (maxcnt < 0)
7082 				break;
7083 			NFSBCOPY(tl, devid, NFSX_V4DEVICEID);
7084 			tl += (NFSX_V4DEVICEID / NFSX_UNSIGNED);
7085 			stat = fxdr_unsigned(int, *tl++);
7086 			opnum = fxdr_unsigned(int, *tl++);
7087 			NFSD_DEBUG(4, "flexlayouterr op=%d stat=%d\n", opnum,
7088 			    stat);
7089 			/*
7090 			 * Except for NFSERR_ACCES, NFSERR_STALE and
7091 			 * NFSERR_NOSPC errors, disable the mirror.
7092 			 */
7093 			if (stat != NFSERR_ACCES && stat != NFSERR_STALE &&
7094 			    stat != NFSERR_NOSPC)
7095 				nfsrv_delds(devid, p);
7096 
7097 			/* For NFSERR_NOSPC, mark all devids and layouts. */
7098 			if (stat == NFSERR_NOSPC)
7099 				nfsrv_marknospc(devid, true);
7100 		}
7101 	}
7102 }
7103 
7104 /*
7105  * This function removes all flex file layouts which has a mirror with
7106  * a device id that matches the argument.
7107  * Called when the DS represented by the device id has failed.
7108  */
7109 void
7110 nfsrv_flexmirrordel(char *devid, NFSPROC_T *p)
7111 {
7112 	uint32_t *tl;
7113 	struct nfslayout *lyp, *nlyp;
7114 	struct nfslayouthash *lhyp;
7115 	struct nfslayouthead loclyp;
7116 	int i, j;
7117 
7118 	NFSD_DEBUG(4, "flexmirrordel\n");
7119 	/* Move all layouts found onto a local list. */
7120 	TAILQ_INIT(&loclyp);
7121 	for (i = 0; i < nfsrv_layouthashsize; i++) {
7122 		lhyp = &nfslayouthash[i];
7123 		NFSLOCKLAYOUT(lhyp);
7124 		TAILQ_FOREACH_SAFE(lyp, &lhyp->list, lay_list, nlyp) {
7125 			if (lyp->lay_type == NFSLAYOUT_FLEXFILE &&
7126 			    lyp->lay_mirrorcnt > 1) {
7127 				NFSD_DEBUG(4, "possible match\n");
7128 				tl = lyp->lay_xdr;
7129 				tl += 3;
7130 				for (j = 0; j < lyp->lay_mirrorcnt; j++) {
7131 					tl++;
7132 					if (NFSBCMP(devid, tl, NFSX_V4DEVICEID)
7133 					    == 0) {
7134 						/* Found one. */
7135 						NFSD_DEBUG(4, "fnd one\n");
7136 						TAILQ_REMOVE(&lhyp->list, lyp,
7137 						    lay_list);
7138 						TAILQ_INSERT_HEAD(&loclyp, lyp,
7139 						    lay_list);
7140 						break;
7141 					}
7142 					tl += (NFSX_V4DEVICEID / NFSX_UNSIGNED +
7143 					    NFSM_RNDUP(NFSX_V4PNFSFH) /
7144 					    NFSX_UNSIGNED + 11 * NFSX_UNSIGNED);
7145 				}
7146 			}
7147 		}
7148 		NFSUNLOCKLAYOUT(lhyp);
7149 	}
7150 
7151 	/* Now, try to do a Layout recall for each one found. */
7152 	TAILQ_FOREACH_SAFE(lyp, &loclyp, lay_list, nlyp) {
7153 		NFSD_DEBUG(4, "do layout recall\n");
7154 		/*
7155 		 * The layout stateid.seqid needs to be incremented
7156 		 * before doing a LAYOUT_RECALL callback.
7157 		 */
7158 		if (++lyp->lay_stateid.seqid == 0)
7159 			lyp->lay_stateid.seqid = 1;
7160 		nfsrv_recalllayout(lyp->lay_clientid, &lyp->lay_stateid,
7161 		    &lyp->lay_fh, lyp, 1, lyp->lay_type, p);
7162 		nfsrv_freelayout(&loclyp, lyp);
7163 	}
7164 }
7165 
7166 /*
7167  * Do a recall callback to the client for this layout.
7168  */
7169 static int
7170 nfsrv_recalllayout(nfsquad_t clid, nfsv4stateid_t *stateidp, fhandle_t *fhp,
7171     struct nfslayout *lyp, int changed, int laytype, NFSPROC_T *p)
7172 {
7173 	struct nfsclient *clp;
7174 	int error;
7175 
7176 	NFSD_DEBUG(4, "nfsrv_recalllayout\n");
7177 	error = nfsrv_getclient(clid, 0, &clp, NULL, (nfsquad_t)((u_quad_t)0),
7178 	    0, NULL, p);
7179 	NFSD_DEBUG(4, "aft nfsrv_getclient=%d\n", error);
7180 	if (error != 0) {
7181 		printf("nfsrv_recalllayout: getclient err=%d\n", error);
7182 		return (error);
7183 	}
7184 	if ((clp->lc_flags & LCL_NFSV41) != 0) {
7185 		error = nfsrv_docallback(clp, NFSV4OP_CBLAYOUTRECALL,
7186 		    stateidp, changed, fhp, NULL, NULL, laytype, p);
7187 		/* If lyp != NULL, handle an error return here. */
7188 		if (error != 0 && lyp != NULL) {
7189 			NFSDRECALLLOCK();
7190 			/*
7191 			 * Mark it returned, since no layout recall
7192 			 * has been done.
7193 			 * All errors seem to be non-recoverable, although
7194 			 * NFSERR_NOMATCHLAYOUT is a normal event.
7195 			 */
7196 			if ((lyp->lay_flags & NFSLAY_RECALL) != 0) {
7197 				lyp->lay_flags |= NFSLAY_RETURNED;
7198 				wakeup(lyp);
7199 			}
7200 			NFSDRECALLUNLOCK();
7201 			if (error != NFSERR_NOMATCHLAYOUT)
7202 				printf("nfsrv_recalllayout: err=%d\n", error);
7203 		}
7204 	} else
7205 		printf("nfsrv_recalllayout: clp not NFSv4.1\n");
7206 	return (error);
7207 }
7208 
7209 /*
7210  * Find a layout to recall when we exceed our high water mark.
7211  */
7212 void
7213 nfsrv_recalloldlayout(NFSPROC_T *p)
7214 {
7215 	struct nfslayouthash *lhyp;
7216 	struct nfslayout *lyp;
7217 	nfsquad_t clientid;
7218 	nfsv4stateid_t stateid;
7219 	fhandle_t fh;
7220 	int error, laytype = 0, ret;
7221 
7222 	lhyp = &nfslayouthash[arc4random() % nfsrv_layouthashsize];
7223 	NFSLOCKLAYOUT(lhyp);
7224 	TAILQ_FOREACH_REVERSE(lyp, &lhyp->list, nfslayouthead, lay_list) {
7225 		if ((lyp->lay_flags & NFSLAY_CALLB) == 0) {
7226 			lyp->lay_flags |= NFSLAY_CALLB;
7227 			/*
7228 			 * The layout stateid.seqid needs to be incremented
7229 			 * before doing a LAYOUT_RECALL callback.
7230 			 */
7231 			if (++lyp->lay_stateid.seqid == 0)
7232 				lyp->lay_stateid.seqid = 1;
7233 			clientid = lyp->lay_clientid;
7234 			stateid = lyp->lay_stateid;
7235 			NFSBCOPY(&lyp->lay_fh, &fh, sizeof(fh));
7236 			laytype = lyp->lay_type;
7237 			break;
7238 		}
7239 	}
7240 	NFSUNLOCKLAYOUT(lhyp);
7241 	if (lyp != NULL) {
7242 		error = nfsrv_recalllayout(clientid, &stateid, &fh, NULL, 0,
7243 		    laytype, p);
7244 		if (error != 0 && error != NFSERR_NOMATCHLAYOUT)
7245 			NFSD_DEBUG(4, "recallold=%d\n", error);
7246 		if (error != 0) {
7247 			NFSLOCKLAYOUT(lhyp);
7248 			/*
7249 			 * Since the hash list was unlocked, we need to
7250 			 * find it again.
7251 			 */
7252 			ret = nfsrv_findlayout(&clientid, &fh, laytype, p,
7253 			    &lyp);
7254 			if (ret == 0 &&
7255 			    (lyp->lay_flags & NFSLAY_CALLB) != 0 &&
7256 			    lyp->lay_stateid.other[0] == stateid.other[0] &&
7257 			    lyp->lay_stateid.other[1] == stateid.other[1] &&
7258 			    lyp->lay_stateid.other[2] == stateid.other[2]) {
7259 				/*
7260 				 * The client no longer knows this layout, so
7261 				 * it can be free'd now.
7262 				 */
7263 				if (error == NFSERR_NOMATCHLAYOUT)
7264 					nfsrv_freelayout(&lhyp->list, lyp);
7265 				else {
7266 					/*
7267 					 * Leave it to be tried later by
7268 					 * clearing NFSLAY_CALLB and moving
7269 					 * it to the head of the list, so it
7270 					 * won't be tried again for a while.
7271 					 */
7272 					lyp->lay_flags &= ~NFSLAY_CALLB;
7273 					TAILQ_REMOVE(&lhyp->list, lyp,
7274 					    lay_list);
7275 					TAILQ_INSERT_HEAD(&lhyp->list, lyp,
7276 					    lay_list);
7277 				}
7278 			}
7279 			NFSUNLOCKLAYOUT(lhyp);
7280 		}
7281 	}
7282 }
7283 
7284 /*
7285  * Try and return layout(s).
7286  */
7287 int
7288 nfsrv_layoutreturn(struct nfsrv_descript *nd, vnode_t vp,
7289     int layouttype, int iomode, uint64_t offset, uint64_t len, int reclaim,
7290     int kind, nfsv4stateid_t *stateidp, int maxcnt, uint32_t *layp, int *fndp,
7291     struct ucred *cred, NFSPROC_T *p)
7292 {
7293 	struct nfsvattr na;
7294 	struct nfslayouthash *lhyp;
7295 	struct nfslayout *lyp;
7296 	fhandle_t fh;
7297 	int error = 0;
7298 
7299 	*fndp = 0;
7300 	if (kind == NFSV4LAYOUTRET_FILE) {
7301 		error = nfsvno_getfh(vp, &fh, p);
7302 		if (error == 0) {
7303 			error = nfsrv_updatemdsattr(vp, &na, p);
7304 			if (error != 0)
7305 				printf("nfsrv_layoutreturn: updatemdsattr"
7306 				    " failed=%d\n", error);
7307 		}
7308 		if (error == 0) {
7309 			if (reclaim == newnfs_true) {
7310 				error = nfsrv_checkgrace(NULL, NULL,
7311 				    NFSLCK_RECLAIM);
7312 				if (error != NFSERR_NOGRACE)
7313 					error = 0;
7314 				return (error);
7315 			}
7316 			lhyp = NFSLAYOUTHASH(&fh);
7317 			NFSDRECALLLOCK();
7318 			NFSLOCKLAYOUT(lhyp);
7319 			error = nfsrv_findlayout(&nd->nd_clientid, &fh,
7320 			    layouttype, p, &lyp);
7321 			NFSD_DEBUG(4, "layoutret findlay=%d\n", error);
7322 			if (error == 0 &&
7323 			    stateidp->other[0] == lyp->lay_stateid.other[0] &&
7324 			    stateidp->other[1] == lyp->lay_stateid.other[1] &&
7325 			    stateidp->other[2] == lyp->lay_stateid.other[2]) {
7326 				NFSD_DEBUG(4, "nfsrv_layoutreturn: stateid %d"
7327 				    " %x %x %x laystateid %d %x %x %x"
7328 				    " off=%ju len=%ju flgs=0x%x\n",
7329 				    stateidp->seqid, stateidp->other[0],
7330 				    stateidp->other[1], stateidp->other[2],
7331 				    lyp->lay_stateid.seqid,
7332 				    lyp->lay_stateid.other[0],
7333 				    lyp->lay_stateid.other[1],
7334 				    lyp->lay_stateid.other[2],
7335 				    (uintmax_t)offset, (uintmax_t)len,
7336 				    lyp->lay_flags);
7337 				if (++lyp->lay_stateid.seqid == 0)
7338 					lyp->lay_stateid.seqid = 1;
7339 				stateidp->seqid = lyp->lay_stateid.seqid;
7340 				if (offset == 0 && len == UINT64_MAX) {
7341 					if ((iomode & NFSLAYOUTIOMODE_READ) !=
7342 					    0)
7343 						lyp->lay_flags &= ~NFSLAY_READ;
7344 					if ((iomode & NFSLAYOUTIOMODE_RW) != 0)
7345 						lyp->lay_flags &= ~NFSLAY_RW;
7346 					if ((lyp->lay_flags & (NFSLAY_READ |
7347 					    NFSLAY_RW)) == 0)
7348 						nfsrv_freelayout(&lhyp->list,
7349 						    lyp);
7350 					else
7351 						*fndp = 1;
7352 				} else
7353 					*fndp = 1;
7354 			}
7355 			NFSUNLOCKLAYOUT(lhyp);
7356 			/* Search the nfsrv_recalllist for a match. */
7357 			TAILQ_FOREACH(lyp, &nfsrv_recalllisthead, lay_list) {
7358 				if (NFSBCMP(&lyp->lay_fh, &fh,
7359 				    sizeof(fh)) == 0 &&
7360 				    lyp->lay_clientid.qval ==
7361 				    nd->nd_clientid.qval &&
7362 				    stateidp->other[0] ==
7363 				    lyp->lay_stateid.other[0] &&
7364 				    stateidp->other[1] ==
7365 				    lyp->lay_stateid.other[1] &&
7366 				    stateidp->other[2] ==
7367 				    lyp->lay_stateid.other[2]) {
7368 					lyp->lay_flags |= NFSLAY_RETURNED;
7369 					wakeup(lyp);
7370 					error = 0;
7371 				}
7372 			}
7373 			NFSDRECALLUNLOCK();
7374 		}
7375 		if (layouttype == NFSLAYOUT_FLEXFILE && layp != NULL)
7376 			nfsrv_flexlayouterr(nd, layp, maxcnt, p);
7377 	} else if (kind == NFSV4LAYOUTRET_FSID)
7378 		nfsrv_freelayouts(&nd->nd_clientid,
7379 		    &vp->v_mount->mnt_stat.f_fsid, layouttype, iomode);
7380 	else if (kind == NFSV4LAYOUTRET_ALL)
7381 		nfsrv_freelayouts(&nd->nd_clientid, NULL, layouttype, iomode);
7382 	else
7383 		error = NFSERR_INVAL;
7384 	if (error == -1)
7385 		error = 0;
7386 	return (error);
7387 }
7388 
7389 /*
7390  * Look for an existing layout.
7391  */
7392 static int
7393 nfsrv_findlayout(nfsquad_t *clientidp, fhandle_t *fhp, int laytype,
7394     NFSPROC_T *p, struct nfslayout **lypp)
7395 {
7396 	struct nfslayouthash *lhyp;
7397 	struct nfslayout *lyp;
7398 	int ret;
7399 
7400 	*lypp = NULL;
7401 	ret = 0;
7402 	lhyp = NFSLAYOUTHASH(fhp);
7403 	TAILQ_FOREACH(lyp, &lhyp->list, lay_list) {
7404 		if (NFSBCMP(&lyp->lay_fh, fhp, sizeof(*fhp)) == 0 &&
7405 		    lyp->lay_clientid.qval == clientidp->qval &&
7406 		    lyp->lay_type == laytype)
7407 			break;
7408 	}
7409 	if (lyp != NULL)
7410 		*lypp = lyp;
7411 	else
7412 		ret = -1;
7413 	return (ret);
7414 }
7415 
7416 /*
7417  * Add the new layout, as required.
7418  */
7419 static int
7420 nfsrv_addlayout(struct nfsrv_descript *nd, struct nfslayout **lypp,
7421     nfsv4stateid_t *stateidp, char *layp, int *layoutlenp, NFSPROC_T *p)
7422 {
7423 	struct nfsclient *clp;
7424 	struct nfslayouthash *lhyp;
7425 	struct nfslayout *lyp, *nlyp;
7426 	fhandle_t *fhp;
7427 	int error;
7428 
7429 	KASSERT((nd->nd_flag & ND_IMPLIEDCLID) != 0,
7430 	    ("nfsrv_layoutget: no nd_clientid\n"));
7431 	lyp = *lypp;
7432 	fhp = &lyp->lay_fh;
7433 	NFSLOCKSTATE();
7434 	error = nfsrv_getclient((nfsquad_t)((u_quad_t)0), CLOPS_RENEW, &clp,
7435 	    NULL, (nfsquad_t)((u_quad_t)0), 0, nd, p);
7436 	if (error != 0) {
7437 		NFSUNLOCKSTATE();
7438 		return (error);
7439 	}
7440 	lyp->lay_stateid.seqid = stateidp->seqid = 1;
7441 	lyp->lay_stateid.other[0] = stateidp->other[0] =
7442 	    clp->lc_clientid.lval[0];
7443 	lyp->lay_stateid.other[1] = stateidp->other[1] =
7444 	    clp->lc_clientid.lval[1];
7445 	lyp->lay_stateid.other[2] = stateidp->other[2] =
7446 	    nfsrv_nextstateindex(clp);
7447 	NFSUNLOCKSTATE();
7448 
7449 	lhyp = NFSLAYOUTHASH(fhp);
7450 	NFSLOCKLAYOUT(lhyp);
7451 	TAILQ_FOREACH(nlyp, &lhyp->list, lay_list) {
7452 		if (NFSBCMP(&nlyp->lay_fh, fhp, sizeof(*fhp)) == 0 &&
7453 		    nlyp->lay_clientid.qval == nd->nd_clientid.qval)
7454 			break;
7455 	}
7456 	if (nlyp != NULL) {
7457 		/* A layout already exists, so use it. */
7458 		nlyp->lay_flags |= (lyp->lay_flags & (NFSLAY_READ | NFSLAY_RW));
7459 		NFSBCOPY(nlyp->lay_xdr, layp, nlyp->lay_layoutlen);
7460 		*layoutlenp = nlyp->lay_layoutlen;
7461 		if (++nlyp->lay_stateid.seqid == 0)
7462 			nlyp->lay_stateid.seqid = 1;
7463 		stateidp->seqid = nlyp->lay_stateid.seqid;
7464 		stateidp->other[0] = nlyp->lay_stateid.other[0];
7465 		stateidp->other[1] = nlyp->lay_stateid.other[1];
7466 		stateidp->other[2] = nlyp->lay_stateid.other[2];
7467 		NFSUNLOCKLAYOUT(lhyp);
7468 		return (0);
7469 	}
7470 
7471 	/* Insert the new layout in the lists. */
7472 	*lypp = NULL;
7473 	atomic_add_int(&nfsrv_layoutcnt, 1);
7474 	NFSD_VNET(nfsstatsv1_p)->srvlayouts++;
7475 	NFSBCOPY(lyp->lay_xdr, layp, lyp->lay_layoutlen);
7476 	*layoutlenp = lyp->lay_layoutlen;
7477 	TAILQ_INSERT_HEAD(&lhyp->list, lyp, lay_list);
7478 	NFSUNLOCKLAYOUT(lhyp);
7479 	return (0);
7480 }
7481 
7482 /*
7483  * Get the devinfo for a deviceid.
7484  */
7485 int
7486 nfsrv_getdevinfo(char *devid, int layouttype, uint32_t *maxcnt,
7487     uint32_t *notify, int *devaddrlen, char **devaddr)
7488 {
7489 	struct nfsdevice *ds;
7490 
7491 	if ((layouttype != NFSLAYOUT_NFSV4_1_FILES && layouttype !=
7492 	     NFSLAYOUT_FLEXFILE) ||
7493 	    (nfsrv_maxpnfsmirror > 1 && layouttype == NFSLAYOUT_NFSV4_1_FILES))
7494 		return (NFSERR_UNKNLAYOUTTYPE);
7495 
7496 	/*
7497 	 * Now, search for the device id.  Note that the structures won't go
7498 	 * away, but the order changes in the list.  As such, the lock only
7499 	 * needs to be held during the search through the list.
7500 	 */
7501 	NFSDDSLOCK();
7502 	TAILQ_FOREACH(ds, &nfsrv_devidhead, nfsdev_list) {
7503 		if (NFSBCMP(devid, ds->nfsdev_deviceid, NFSX_V4DEVICEID) == 0 &&
7504 		    ds->nfsdev_nmp != NULL)
7505 			break;
7506 	}
7507 	NFSDDSUNLOCK();
7508 	if (ds == NULL)
7509 		return (NFSERR_NOENT);
7510 
7511 	/* If the correct nfsdev_XXXXaddrlen is > 0, we have the device info. */
7512 	*devaddrlen = 0;
7513 	if (layouttype == NFSLAYOUT_NFSV4_1_FILES) {
7514 		*devaddrlen = ds->nfsdev_fileaddrlen;
7515 		*devaddr = ds->nfsdev_fileaddr;
7516 	} else if (layouttype == NFSLAYOUT_FLEXFILE) {
7517 		*devaddrlen = ds->nfsdev_flexaddrlen;
7518 		*devaddr = ds->nfsdev_flexaddr;
7519 	}
7520 	if (*devaddrlen == 0)
7521 		return (NFSERR_UNKNLAYOUTTYPE);
7522 
7523 	/*
7524 	 * The XDR overhead is 3 unsigned values: layout_type,
7525 	 * length_of_address and notify bitmap.
7526 	 * If the notify array is changed to not all zeros, the
7527 	 * count of unsigned values must be increased.
7528 	 */
7529 	if (*maxcnt > 0 && *maxcnt < NFSM_RNDUP(*devaddrlen) +
7530 	    3 * NFSX_UNSIGNED) {
7531 		*maxcnt = NFSM_RNDUP(*devaddrlen) + 3 * NFSX_UNSIGNED;
7532 		return (NFSERR_TOOSMALL);
7533 	}
7534 	return (0);
7535 }
7536 
7537 /*
7538  * Free a list of layout state structures.
7539  */
7540 static void
7541 nfsrv_freelayoutlist(nfsquad_t clientid)
7542 {
7543 	struct nfslayouthash *lhyp;
7544 	struct nfslayout *lyp, *nlyp;
7545 	int i;
7546 
7547 	for (i = 0; i < nfsrv_layouthashsize; i++) {
7548 		lhyp = &nfslayouthash[i];
7549 		NFSLOCKLAYOUT(lhyp);
7550 		TAILQ_FOREACH_SAFE(lyp, &lhyp->list, lay_list, nlyp) {
7551 			if (lyp->lay_clientid.qval == clientid.qval)
7552 				nfsrv_freelayout(&lhyp->list, lyp);
7553 		}
7554 		NFSUNLOCKLAYOUT(lhyp);
7555 	}
7556 }
7557 
7558 /*
7559  * Free up a layout.
7560  */
7561 static void
7562 nfsrv_freelayout(struct nfslayouthead *lhp, struct nfslayout *lyp)
7563 {
7564 
7565 	NFSD_DEBUG(4, "Freelayout=%p\n", lyp);
7566 	atomic_add_int(&nfsrv_layoutcnt, -1);
7567 	NFSD_VNET(nfsstatsv1_p)->srvlayouts--;
7568 	TAILQ_REMOVE(lhp, lyp, lay_list);
7569 	free(lyp, M_NFSDSTATE);
7570 }
7571 
7572 /*
7573  * Free up a device id.
7574  */
7575 void
7576 nfsrv_freeonedevid(struct nfsdevice *ds)
7577 {
7578 	int i;
7579 
7580 	atomic_add_int(&nfsrv_devidcnt, -1);
7581 	vrele(ds->nfsdev_dvp);
7582 	for (i = 0; i < nfsrv_dsdirsize; i++)
7583 		if (ds->nfsdev_dsdir[i] != NULL)
7584 			vrele(ds->nfsdev_dsdir[i]);
7585 	free(ds->nfsdev_fileaddr, M_NFSDSTATE);
7586 	free(ds->nfsdev_flexaddr, M_NFSDSTATE);
7587 	free(ds->nfsdev_host, M_NFSDSTATE);
7588 	free(ds, M_NFSDSTATE);
7589 }
7590 
7591 /*
7592  * Free up a device id and its mirrors.
7593  */
7594 static void
7595 nfsrv_freedevid(struct nfsdevice *ds)
7596 {
7597 
7598 	TAILQ_REMOVE(&nfsrv_devidhead, ds, nfsdev_list);
7599 	nfsrv_freeonedevid(ds);
7600 }
7601 
7602 /*
7603  * Free all layouts and device ids.
7604  * Done when the nfsd threads are shut down since there may be a new
7605  * modified device id list created when the nfsd is restarted.
7606  */
7607 void
7608 nfsrv_freealllayoutsanddevids(void)
7609 {
7610 	struct nfsdontlist *mrp, *nmrp;
7611 	struct nfslayout *lyp, *nlyp;
7612 
7613 	/* Get rid of the deviceid structures. */
7614 	nfsrv_freealldevids();
7615 	TAILQ_INIT(&nfsrv_devidhead);
7616 	nfsrv_devidcnt = 0;
7617 
7618 	/* Get rid of all layouts. */
7619 	nfsrv_freealllayouts();
7620 
7621 	/* Get rid of any nfsdontlist entries. */
7622 	LIST_FOREACH_SAFE(mrp, &nfsrv_dontlisthead, nfsmr_list, nmrp)
7623 		free(mrp, M_NFSDSTATE);
7624 	LIST_INIT(&nfsrv_dontlisthead);
7625 	nfsrv_dontlistlen = 0;
7626 
7627 	/* Free layouts in the recall list. */
7628 	TAILQ_FOREACH_SAFE(lyp, &nfsrv_recalllisthead, lay_list, nlyp)
7629 		nfsrv_freelayout(&nfsrv_recalllisthead, lyp);
7630 	TAILQ_INIT(&nfsrv_recalllisthead);
7631 }
7632 
7633 /*
7634  * Free layouts that match the arguments.
7635  */
7636 static void
7637 nfsrv_freelayouts(nfsquad_t *clid, fsid_t *fs, int laytype, int iomode)
7638 {
7639 	struct nfslayouthash *lhyp;
7640 	struct nfslayout *lyp, *nlyp;
7641 	int i;
7642 
7643 	for (i = 0; i < nfsrv_layouthashsize; i++) {
7644 		lhyp = &nfslayouthash[i];
7645 		NFSLOCKLAYOUT(lhyp);
7646 		TAILQ_FOREACH_SAFE(lyp, &lhyp->list, lay_list, nlyp) {
7647 			if (clid->qval != lyp->lay_clientid.qval)
7648 				continue;
7649 			if (fs != NULL && fsidcmp(fs, &lyp->lay_fsid) != 0)
7650 				continue;
7651 			if (laytype != lyp->lay_type)
7652 				continue;
7653 			if ((iomode & NFSLAYOUTIOMODE_READ) != 0)
7654 				lyp->lay_flags &= ~NFSLAY_READ;
7655 			if ((iomode & NFSLAYOUTIOMODE_RW) != 0)
7656 				lyp->lay_flags &= ~NFSLAY_RW;
7657 			if ((lyp->lay_flags & (NFSLAY_READ | NFSLAY_RW)) == 0)
7658 				nfsrv_freelayout(&lhyp->list, lyp);
7659 		}
7660 		NFSUNLOCKLAYOUT(lhyp);
7661 	}
7662 }
7663 
7664 /*
7665  * Free all layouts for the argument file.
7666  */
7667 void
7668 nfsrv_freefilelayouts(fhandle_t *fhp)
7669 {
7670 	struct nfslayouthash *lhyp;
7671 	struct nfslayout *lyp, *nlyp;
7672 
7673 	lhyp = NFSLAYOUTHASH(fhp);
7674 	NFSLOCKLAYOUT(lhyp);
7675 	TAILQ_FOREACH_SAFE(lyp, &lhyp->list, lay_list, nlyp) {
7676 		if (NFSBCMP(&lyp->lay_fh, fhp, sizeof(*fhp)) == 0)
7677 			nfsrv_freelayout(&lhyp->list, lyp);
7678 	}
7679 	NFSUNLOCKLAYOUT(lhyp);
7680 }
7681 
7682 /*
7683  * Free all layouts.
7684  */
7685 static void
7686 nfsrv_freealllayouts(void)
7687 {
7688 	struct nfslayouthash *lhyp;
7689 	struct nfslayout *lyp, *nlyp;
7690 	int i;
7691 
7692 	for (i = 0; i < nfsrv_layouthashsize; i++) {
7693 		lhyp = &nfslayouthash[i];
7694 		NFSLOCKLAYOUT(lhyp);
7695 		TAILQ_FOREACH_SAFE(lyp, &lhyp->list, lay_list, nlyp)
7696 			nfsrv_freelayout(&lhyp->list, lyp);
7697 		NFSUNLOCKLAYOUT(lhyp);
7698 	}
7699 }
7700 
7701 /*
7702  * Look up the mount path for the DS server.
7703  */
7704 static int
7705 nfsrv_setdsserver(char *dspathp, char *mdspathp, NFSPROC_T *p,
7706     struct nfsdevice **dsp)
7707 {
7708 	struct nameidata nd;
7709 	struct nfsdevice *ds;
7710 	struct mount *mp;
7711 	int error, i;
7712 	char *dsdirpath;
7713 	size_t dsdirsize;
7714 
7715 	NFSD_DEBUG(4, "setdssrv path=%s\n", dspathp);
7716 	*dsp = NULL;
7717 	if (jailed(p->td_ucred)) {
7718 		printf("A pNFS nfsd cannot run in a jail\n");
7719 		return (EPERM);
7720 	}
7721 	NDINIT(&nd, LOOKUP, FOLLOW | LOCKSHARED | LOCKLEAF, UIO_SYSSPACE,
7722 	    dspathp);
7723 	error = namei(&nd);
7724 	NFSD_DEBUG(4, "lookup=%d\n", error);
7725 	if (error != 0)
7726 		return (error);
7727 	if (nd.ni_vp->v_type != VDIR) {
7728 		vput(nd.ni_vp);
7729 		NFSD_DEBUG(4, "dspath not dir\n");
7730 		return (ENOTDIR);
7731 	}
7732 	if (strcmp(nd.ni_vp->v_mount->mnt_vfc->vfc_name, "nfs") != 0) {
7733 		vput(nd.ni_vp);
7734 		NFSD_DEBUG(4, "dspath not an NFS mount\n");
7735 		return (ENXIO);
7736 	}
7737 
7738 	/*
7739 	 * Allocate a DS server structure with the NFS mounted directory
7740 	 * vnode reference counted, so that a non-forced dismount will
7741 	 * fail with EBUSY.
7742 	 * This structure is always linked into the list, even if an error
7743 	 * is being returned.  The caller will free the entire list upon
7744 	 * an error return.
7745 	 */
7746 	*dsp = ds = malloc(sizeof(*ds) + nfsrv_dsdirsize * sizeof(vnode_t),
7747 	    M_NFSDSTATE, M_WAITOK | M_ZERO);
7748 	ds->nfsdev_dvp = nd.ni_vp;
7749 	ds->nfsdev_nmp = VFSTONFS(nd.ni_vp->v_mount);
7750 	NFSVOPUNLOCK(nd.ni_vp);
7751 
7752 	dsdirsize = strlen(dspathp) + 16;
7753 	dsdirpath = malloc(dsdirsize, M_TEMP, M_WAITOK);
7754 	/* Now, create the DS directory structures. */
7755 	for (i = 0; i < nfsrv_dsdirsize; i++) {
7756 		snprintf(dsdirpath, dsdirsize, "%s/ds%d", dspathp, i);
7757 		NDINIT(&nd, LOOKUP, FOLLOW | LOCKSHARED | LOCKLEAF,
7758 		    UIO_SYSSPACE, dsdirpath);
7759 		error = namei(&nd);
7760 		NFSD_DEBUG(4, "dsdirpath=%s lookup=%d\n", dsdirpath, error);
7761 		if (error != 0)
7762 			break;
7763 		if (nd.ni_vp->v_type != VDIR) {
7764 			vput(nd.ni_vp);
7765 			error = ENOTDIR;
7766 			NFSD_DEBUG(4, "dsdirpath not a VDIR\n");
7767 			break;
7768 		}
7769 		if (strcmp(nd.ni_vp->v_mount->mnt_vfc->vfc_name, "nfs") != 0) {
7770 			vput(nd.ni_vp);
7771 			error = ENXIO;
7772 			NFSD_DEBUG(4, "dsdirpath not an NFS mount\n");
7773 			break;
7774 		}
7775 		ds->nfsdev_dsdir[i] = nd.ni_vp;
7776 		NFSVOPUNLOCK(nd.ni_vp);
7777 	}
7778 	free(dsdirpath, M_TEMP);
7779 
7780 	if (strlen(mdspathp) > 0) {
7781 		/*
7782 		 * This DS stores file for a specific MDS exported file
7783 		 * system.
7784 		 */
7785 		NDINIT(&nd, LOOKUP, FOLLOW | LOCKSHARED | LOCKLEAF,
7786 		    UIO_SYSSPACE, mdspathp);
7787 		error = namei(&nd);
7788 		NFSD_DEBUG(4, "mds lookup=%d\n", error);
7789 		if (error != 0)
7790 			goto out;
7791 		if (nd.ni_vp->v_type != VDIR) {
7792 			vput(nd.ni_vp);
7793 			error = ENOTDIR;
7794 			NFSD_DEBUG(4, "mdspath not dir\n");
7795 			goto out;
7796 		}
7797 		mp = nd.ni_vp->v_mount;
7798 		if ((mp->mnt_flag & MNT_EXPORTED) == 0) {
7799 			vput(nd.ni_vp);
7800 			error = ENXIO;
7801 			NFSD_DEBUG(4, "mdspath not an exported fs\n");
7802 			goto out;
7803 		}
7804 		ds->nfsdev_mdsfsid = mp->mnt_stat.f_fsid;
7805 		ds->nfsdev_mdsisset = 1;
7806 		vput(nd.ni_vp);
7807 	}
7808 
7809 out:
7810 	TAILQ_INSERT_TAIL(&nfsrv_devidhead, ds, nfsdev_list);
7811 	atomic_add_int(&nfsrv_devidcnt, 1);
7812 	return (error);
7813 }
7814 
7815 /*
7816  * Look up the mount path for the DS server and delete it.
7817  */
7818 int
7819 nfsrv_deldsserver(int op, char *dspathp, NFSPROC_T *p)
7820 {
7821 	struct mount *mp;
7822 	struct nfsmount *nmp;
7823 	struct nfsdevice *ds;
7824 	int error;
7825 
7826 	NFSD_DEBUG(4, "deldssrv path=%s\n", dspathp);
7827 	/*
7828 	 * Search for the path in the mount list.  Avoid looking the path
7829 	 * up, since this mount point may be hung, with associated locked
7830 	 * vnodes, etc.
7831 	 * Set NFSMNTP_CANCELRPCS so that any forced dismount will be blocked
7832 	 * until this completes.
7833 	 * As noted in the man page, this should be done before any forced
7834 	 * dismount on the mount point, but at least the handshake on
7835 	 * NFSMNTP_CANCELRPCS should make it safe.
7836 	 */
7837 	error = 0;
7838 	ds = NULL;
7839 	nmp = NULL;
7840 	mtx_lock(&mountlist_mtx);
7841 	TAILQ_FOREACH(mp, &mountlist, mnt_list) {
7842 		if (strcmp(mp->mnt_stat.f_mntonname, dspathp) == 0 &&
7843 		    strcmp(mp->mnt_stat.f_fstypename, "nfs") == 0 &&
7844 		    mp->mnt_data != NULL) {
7845 			nmp = VFSTONFS(mp);
7846 			NFSLOCKMNT(nmp);
7847 			if ((nmp->nm_privflag & (NFSMNTP_FORCEDISM |
7848 			     NFSMNTP_CANCELRPCS)) == 0) {
7849 				nmp->nm_privflag |= NFSMNTP_CANCELRPCS;
7850 				NFSUNLOCKMNT(nmp);
7851 			} else {
7852 				NFSUNLOCKMNT(nmp);
7853 				nmp = NULL;
7854 			}
7855 			break;
7856 		}
7857 	}
7858 	mtx_unlock(&mountlist_mtx);
7859 
7860 	if (nmp != NULL) {
7861 		ds = nfsrv_deldsnmp(op, nmp, p);
7862 		NFSD_DEBUG(4, "deldsnmp=%p\n", ds);
7863 		if (ds != NULL) {
7864 			nfsrv_killrpcs(nmp);
7865 			NFSD_DEBUG(4, "aft killrpcs\n");
7866 		} else
7867 			error = ENXIO;
7868 		NFSLOCKMNT(nmp);
7869 		nmp->nm_privflag &= ~NFSMNTP_CANCELRPCS;
7870 		wakeup(nmp);
7871 		NFSUNLOCKMNT(nmp);
7872 	} else
7873 		error = EINVAL;
7874 	return (error);
7875 }
7876 
7877 /*
7878  * Search for and remove a DS entry which matches the "nmp" argument.
7879  * The nfsdevice structure pointer is returned so that the caller can
7880  * free it via nfsrv_freeonedevid().
7881  * For the forced case, do not try to do LayoutRecalls, since the server
7882  * must be shut down now anyhow.
7883  */
7884 struct nfsdevice *
7885 nfsrv_deldsnmp(int op, struct nfsmount *nmp, NFSPROC_T *p)
7886 {
7887 	struct nfsdevice *fndds;
7888 
7889 	NFSD_DEBUG(4, "deldsdvp\n");
7890 	NFSDDSLOCK();
7891 	if (op == PNFSDOP_FORCEDELDS)
7892 		fndds = nfsv4_findmirror(nmp);
7893 	else
7894 		fndds = nfsrv_findmirroredds(nmp);
7895 	if (fndds != NULL)
7896 		nfsrv_deleteds(fndds);
7897 	NFSDDSUNLOCK();
7898 	if (fndds != NULL) {
7899 		if (op != PNFSDOP_FORCEDELDS)
7900 			nfsrv_flexmirrordel(fndds->nfsdev_deviceid, p);
7901 		printf("pNFS server: mirror %s failed\n", fndds->nfsdev_host);
7902 	}
7903 	return (fndds);
7904 }
7905 
7906 /*
7907  * Similar to nfsrv_deldsnmp(), except that the DS is indicated by deviceid.
7908  * This function also calls nfsrv_killrpcs() to unblock RPCs on the mount
7909  * point.
7910  * Also, returns an error instead of the nfsdevice found.
7911  */
7912 int
7913 nfsrv_delds(char *devid, NFSPROC_T *p)
7914 {
7915 	struct nfsdevice *ds, *fndds;
7916 	struct nfsmount *nmp;
7917 	int fndmirror;
7918 
7919 	NFSD_DEBUG(4, "delds\n");
7920 	/*
7921 	 * Search the DS server list for a match with devid.
7922 	 * Remove the DS entry if found and there is a mirror.
7923 	 */
7924 	fndds = NULL;
7925 	nmp = NULL;
7926 	fndmirror = 0;
7927 	NFSDDSLOCK();
7928 	TAILQ_FOREACH(ds, &nfsrv_devidhead, nfsdev_list) {
7929 		if (NFSBCMP(ds->nfsdev_deviceid, devid, NFSX_V4DEVICEID) == 0 &&
7930 		    ds->nfsdev_nmp != NULL) {
7931 			NFSD_DEBUG(4, "fnd main ds\n");
7932 			fndds = ds;
7933 			break;
7934 		}
7935 	}
7936 	if (fndds == NULL) {
7937 		NFSDDSUNLOCK();
7938 		return (ENXIO);
7939 	}
7940 	if (fndds->nfsdev_mdsisset == 0 && nfsrv_faildscnt > 0)
7941 		fndmirror = 1;
7942 	else if (fndds->nfsdev_mdsisset != 0) {
7943 		/* For the fsid is set case, search for a mirror. */
7944 		TAILQ_FOREACH(ds, &nfsrv_devidhead, nfsdev_list) {
7945 			if (ds != fndds && ds->nfsdev_nmp != NULL &&
7946 			    ds->nfsdev_mdsisset != 0 &&
7947 			    fsidcmp(&ds->nfsdev_mdsfsid,
7948 			    &fndds->nfsdev_mdsfsid) == 0) {
7949 				fndmirror = 1;
7950 				break;
7951 			}
7952 		}
7953 	}
7954 	if (fndmirror != 0) {
7955 		nmp = fndds->nfsdev_nmp;
7956 		NFSLOCKMNT(nmp);
7957 		if ((nmp->nm_privflag & (NFSMNTP_FORCEDISM |
7958 		     NFSMNTP_CANCELRPCS)) == 0) {
7959 			nmp->nm_privflag |= NFSMNTP_CANCELRPCS;
7960 			NFSUNLOCKMNT(nmp);
7961 			nfsrv_deleteds(fndds);
7962 		} else {
7963 			NFSUNLOCKMNT(nmp);
7964 			nmp = NULL;
7965 		}
7966 	}
7967 	NFSDDSUNLOCK();
7968 	if (nmp != NULL) {
7969 		nfsrv_flexmirrordel(fndds->nfsdev_deviceid, p);
7970 		printf("pNFS server: mirror %s failed\n", fndds->nfsdev_host);
7971 		nfsrv_killrpcs(nmp);
7972 		NFSLOCKMNT(nmp);
7973 		nmp->nm_privflag &= ~NFSMNTP_CANCELRPCS;
7974 		wakeup(nmp);
7975 		NFSUNLOCKMNT(nmp);
7976 		return (0);
7977 	}
7978 	return (ENXIO);
7979 }
7980 
7981 /*
7982  * Mark a DS as disabled by setting nfsdev_nmp = NULL.
7983  */
7984 static void
7985 nfsrv_deleteds(struct nfsdevice *fndds)
7986 {
7987 
7988 	NFSD_DEBUG(4, "deleteds: deleting a mirror\n");
7989 	fndds->nfsdev_nmp = NULL;
7990 	if (fndds->nfsdev_mdsisset == 0)
7991 		nfsrv_faildscnt--;
7992 }
7993 
7994 /*
7995  * Fill in the addr structures for the File and Flex File layouts.
7996  */
7997 static void
7998 nfsrv_allocdevid(struct nfsdevice *ds, char *addr, char *dnshost)
7999 {
8000 	uint32_t *tl;
8001 	char *netprot;
8002 	int addrlen;
8003 	static uint64_t new_devid = 0;
8004 
8005 	if (strchr(addr, ':') != NULL)
8006 		netprot = "tcp6";
8007 	else
8008 		netprot = "tcp";
8009 
8010 	/* Fill in the device id. */
8011 	NFSBCOPY(&nfsdev_time, ds->nfsdev_deviceid, sizeof(nfsdev_time));
8012 	new_devid++;
8013 	NFSBCOPY(&new_devid, &ds->nfsdev_deviceid[sizeof(nfsdev_time)],
8014 	    sizeof(new_devid));
8015 
8016 	/*
8017 	 * Fill in the file addr (actually the nfsv4_file_layout_ds_addr4
8018 	 * as defined in RFC5661) in XDR.
8019 	 */
8020 	addrlen = NFSM_RNDUP(strlen(addr)) + NFSM_RNDUP(strlen(netprot)) +
8021 	    6 * NFSX_UNSIGNED;
8022 	NFSD_DEBUG(4, "hn=%s addr=%s netprot=%s\n", dnshost, addr, netprot);
8023 	ds->nfsdev_fileaddrlen = addrlen;
8024 	tl = malloc(addrlen, M_NFSDSTATE, M_WAITOK | M_ZERO);
8025 	ds->nfsdev_fileaddr = (char *)tl;
8026 	*tl++ = txdr_unsigned(1);		/* One stripe with index 0. */
8027 	*tl++ = 0;
8028 	*tl++ = txdr_unsigned(1);		/* One multipath list */
8029 	*tl++ = txdr_unsigned(1);		/* with one entry in it. */
8030 	/* The netaddr for this one entry. */
8031 	*tl++ = txdr_unsigned(strlen(netprot));
8032 	NFSBCOPY(netprot, tl, strlen(netprot));
8033 	tl += (NFSM_RNDUP(strlen(netprot)) / NFSX_UNSIGNED);
8034 	*tl++ = txdr_unsigned(strlen(addr));
8035 	NFSBCOPY(addr, tl, strlen(addr));
8036 
8037 	/*
8038 	 * Fill in the flex file addr (actually the ff_device_addr4
8039 	 * as defined for Flexible File Layout) in XDR.
8040 	 */
8041 	addrlen = NFSM_RNDUP(strlen(addr)) + NFSM_RNDUP(strlen(netprot)) +
8042 	    14 * NFSX_UNSIGNED;
8043 	ds->nfsdev_flexaddrlen = addrlen;
8044 	tl = malloc(addrlen, M_NFSDSTATE, M_WAITOK | M_ZERO);
8045 	ds->nfsdev_flexaddr = (char *)tl;
8046 	*tl++ = txdr_unsigned(1);		/* One multipath entry. */
8047 	/* The netaddr for this one entry. */
8048 	*tl++ = txdr_unsigned(strlen(netprot));
8049 	NFSBCOPY(netprot, tl, strlen(netprot));
8050 	tl += (NFSM_RNDUP(strlen(netprot)) / NFSX_UNSIGNED);
8051 	*tl++ = txdr_unsigned(strlen(addr));
8052 	NFSBCOPY(addr, tl, strlen(addr));
8053 	tl += (NFSM_RNDUP(strlen(addr)) / NFSX_UNSIGNED);
8054 	*tl++ = txdr_unsigned(2);		/* Two NFS Versions. */
8055 	*tl++ = txdr_unsigned(NFS_VER4);	/* NFSv4. */
8056 	*tl++ = txdr_unsigned(NFSV42_MINORVERSION); /* Minor version 2. */
8057 	*tl++ = txdr_unsigned(nfs_srvmaxio);	/* DS max rsize. */
8058 	*tl++ = txdr_unsigned(nfs_srvmaxio);	/* DS max wsize. */
8059 	*tl++ = newnfs_true;			/* Tightly coupled. */
8060 	*tl++ = txdr_unsigned(NFS_VER4);	/* NFSv4. */
8061 	*tl++ = txdr_unsigned(NFSV41_MINORVERSION); /* Minor version 1. */
8062 	*tl++ = txdr_unsigned(nfs_srvmaxio);	/* DS max rsize. */
8063 	*tl++ = txdr_unsigned(nfs_srvmaxio);	/* DS max wsize. */
8064 	*tl = newnfs_true;			/* Tightly coupled. */
8065 
8066 	ds->nfsdev_hostnamelen = strlen(dnshost);
8067 	ds->nfsdev_host = malloc(ds->nfsdev_hostnamelen + 1, M_NFSDSTATE,
8068 	    M_WAITOK);
8069 	NFSBCOPY(dnshost, ds->nfsdev_host, ds->nfsdev_hostnamelen + 1);
8070 }
8071 
8072 /*
8073  * Create the device id list.
8074  * Return 0 if the nfsd threads are to run and ENXIO if the "-p" argument
8075  * is misconfigured.
8076  */
8077 int
8078 nfsrv_createdevids(struct nfsd_nfsd_args *args, NFSPROC_T *p)
8079 {
8080 	struct nfsdevice *ds;
8081 	char *addrp, *dnshostp, *dspathp, *mdspathp;
8082 	int error, i;
8083 
8084 	addrp = args->addr;
8085 	dnshostp = args->dnshost;
8086 	dspathp = args->dspath;
8087 	mdspathp = args->mdspath;
8088 	nfsrv_maxpnfsmirror = args->mirrorcnt;
8089 	if (addrp == NULL || dnshostp == NULL || dspathp == NULL ||
8090 	    mdspathp == NULL)
8091 		return (0);
8092 
8093 	/*
8094 	 * Loop around for each nul-terminated string in args->addr,
8095 	 * args->dnshost, args->dnspath and args->mdspath.
8096 	 */
8097 	while (addrp < (args->addr + args->addrlen) &&
8098 	    dnshostp < (args->dnshost + args->dnshostlen) &&
8099 	    dspathp < (args->dspath + args->dspathlen) &&
8100 	    mdspathp < (args->mdspath + args->mdspathlen)) {
8101 		error = nfsrv_setdsserver(dspathp, mdspathp, p, &ds);
8102 		if (error != 0) {
8103 			/* Free all DS servers. */
8104 			nfsrv_freealldevids();
8105 			nfsrv_devidcnt = 0;
8106 			return (ENXIO);
8107 		}
8108 		nfsrv_allocdevid(ds, addrp, dnshostp);
8109 		addrp += (strlen(addrp) + 1);
8110 		dnshostp += (strlen(dnshostp) + 1);
8111 		dspathp += (strlen(dspathp) + 1);
8112 		mdspathp += (strlen(mdspathp) + 1);
8113 	}
8114 	if (nfsrv_devidcnt < nfsrv_maxpnfsmirror) {
8115 		/* Free all DS servers. */
8116 		nfsrv_freealldevids();
8117 		nfsrv_devidcnt = 0;
8118 		nfsrv_maxpnfsmirror = 1;
8119 		return (ENXIO);
8120 	}
8121 	/* We can fail at most one less DS than the mirror level. */
8122 	nfsrv_faildscnt = nfsrv_maxpnfsmirror - 1;
8123 
8124 	/*
8125 	 * Allocate the nfslayout hash table now, since this is a pNFS server.
8126 	 * Make it 1% of the high water mark and at least 100.
8127 	 */
8128 	if (nfslayouthash == NULL) {
8129 		nfsrv_layouthashsize = nfsrv_layouthighwater / 100;
8130 		if (nfsrv_layouthashsize < 100)
8131 			nfsrv_layouthashsize = 100;
8132 		nfslayouthash = mallocarray(nfsrv_layouthashsize,
8133 		    sizeof(struct nfslayouthash), M_NFSDSESSION, M_WAITOK |
8134 		    M_ZERO);
8135 		for (i = 0; i < nfsrv_layouthashsize; i++) {
8136 			mtx_init(&nfslayouthash[i].mtx, "nfslm", NULL, MTX_DEF);
8137 			TAILQ_INIT(&nfslayouthash[i].list);
8138 		}
8139 	}
8140 	return (0);
8141 }
8142 
8143 /*
8144  * Free all device ids.
8145  */
8146 static void
8147 nfsrv_freealldevids(void)
8148 {
8149 	struct nfsdevice *ds, *nds;
8150 
8151 	TAILQ_FOREACH_SAFE(ds, &nfsrv_devidhead, nfsdev_list, nds)
8152 		nfsrv_freedevid(ds);
8153 }
8154 
8155 /*
8156  * Check to see if there is a Read/Write Layout plus either:
8157  * - A Write Delegation
8158  * or
8159  * - An Open with Write_access.
8160  * Return 1 if this is the case and 0 otherwise.
8161  * This function is used by nfsrv_proxyds() to decide if doing a Proxy
8162  * Getattr RPC to the Data Server (DS) is necessary.
8163  */
8164 #define	NFSCLIDVECSIZE	6
8165 int
8166 nfsrv_checkdsattr(vnode_t vp, NFSPROC_T *p)
8167 {
8168 	fhandle_t fh, *tfhp;
8169 	struct nfsstate *stp;
8170 	struct nfslayout *lyp;
8171 	struct nfslayouthash *lhyp;
8172 	struct nfslockhashhead *hp;
8173 	struct nfslockfile *lfp;
8174 	nfsquad_t clid[NFSCLIDVECSIZE];
8175 	int clidcnt, ret;
8176 
8177 	ret = nfsvno_getfh(vp, &fh, p);
8178 	if (ret != 0)
8179 		return (0);
8180 
8181 	/* First check for a Read/Write Layout. */
8182 	clidcnt = 0;
8183 	lhyp = NFSLAYOUTHASH(&fh);
8184 	NFSLOCKLAYOUT(lhyp);
8185 	TAILQ_FOREACH(lyp, &lhyp->list, lay_list) {
8186 		if (NFSBCMP(&lyp->lay_fh, &fh, sizeof(fh)) == 0 &&
8187 		    ((lyp->lay_flags & NFSLAY_RW) != 0 ||
8188 		     ((lyp->lay_flags & NFSLAY_READ) != 0 &&
8189 		      nfsrv_pnfsatime != 0))) {
8190 			if (clidcnt < NFSCLIDVECSIZE)
8191 				clid[clidcnt].qval = lyp->lay_clientid.qval;
8192 			clidcnt++;
8193 		}
8194 	}
8195 	NFSUNLOCKLAYOUT(lhyp);
8196 	if (clidcnt == 0) {
8197 		/* None found, so return 0. */
8198 		return (0);
8199 	}
8200 
8201 	/* Get the nfslockfile for this fh. */
8202 	NFSLOCKSTATE();
8203 	hp = NFSLOCKHASH(&fh);
8204 	LIST_FOREACH(lfp, hp, lf_hash) {
8205 		tfhp = &lfp->lf_fh;
8206 		if (NFSVNO_CMPFH(&fh, tfhp))
8207 			break;
8208 	}
8209 	if (lfp == NULL) {
8210 		/* None found, so return 0. */
8211 		NFSUNLOCKSTATE();
8212 		return (0);
8213 	}
8214 
8215 	/* Now, look for a Write delegation for this clientid. */
8216 	LIST_FOREACH(stp, &lfp->lf_deleg, ls_file) {
8217 		if ((stp->ls_flags & NFSLCK_DELEGWRITE) != 0 &&
8218 		    nfsrv_fndclid(clid, stp->ls_clp->lc_clientid, clidcnt) != 0)
8219 			break;
8220 	}
8221 	if (stp != NULL) {
8222 		/* Found one, so return 1. */
8223 		NFSUNLOCKSTATE();
8224 		return (1);
8225 	}
8226 
8227 	/* No Write delegation, so look for an Open with Write_access. */
8228 	LIST_FOREACH(stp, &lfp->lf_open, ls_file) {
8229 		KASSERT((stp->ls_flags & NFSLCK_OPEN) != 0,
8230 		    ("nfsrv_checkdsattr: Non-open in Open list\n"));
8231 		if ((stp->ls_flags & NFSLCK_WRITEACCESS) != 0 &&
8232 		    nfsrv_fndclid(clid, stp->ls_clp->lc_clientid, clidcnt) != 0)
8233 			break;
8234 	}
8235 	NFSUNLOCKSTATE();
8236 	if (stp != NULL)
8237 		return (1);
8238 	return (0);
8239 }
8240 
8241 /*
8242  * Look for a matching clientid in the vector. Return 1 if one might match.
8243  */
8244 static int
8245 nfsrv_fndclid(nfsquad_t *clidvec, nfsquad_t clid, int clidcnt)
8246 {
8247 	int i;
8248 
8249 	/* If too many for the vector, return 1 since there might be a match. */
8250 	if (clidcnt > NFSCLIDVECSIZE)
8251 		return (1);
8252 
8253 	for (i = 0; i < clidcnt; i++)
8254 		if (clidvec[i].qval == clid.qval)
8255 			return (1);
8256 	return (0);
8257 }
8258 
8259 /*
8260  * Check the don't list for "vp" and see if issuing an rw layout is allowed.
8261  * Return 1 if issuing an rw layout isn't allowed, 0 otherwise.
8262  */
8263 static int
8264 nfsrv_dontlayout(fhandle_t *fhp)
8265 {
8266 	struct nfsdontlist *mrp;
8267 	int ret;
8268 
8269 	if (nfsrv_dontlistlen == 0)
8270 		return (0);
8271 	ret = 0;
8272 	NFSDDONTLISTLOCK();
8273 	LIST_FOREACH(mrp, &nfsrv_dontlisthead, nfsmr_list) {
8274 		if (NFSBCMP(fhp, &mrp->nfsmr_fh, sizeof(*fhp)) == 0 &&
8275 		    (mrp->nfsmr_flags & NFSMR_DONTLAYOUT) != 0) {
8276 			ret = 1;
8277 			break;
8278 		}
8279 	}
8280 	NFSDDONTLISTUNLOCK();
8281 	return (ret);
8282 }
8283 
8284 #define	PNFSDS_COPYSIZ	65536
8285 /*
8286  * Create a new file on a DS and copy the contents of an extant DS file to it.
8287  * This can be used for recovery of a DS file onto a recovered DS.
8288  * The steps are:
8289  * - When called, the MDS file's vnode is locked, blocking LayoutGet operations.
8290  * - Disable issuing of read/write layouts for the file via the nfsdontlist,
8291  *   so that they will be disabled after the MDS file's vnode is unlocked.
8292  * - Set up the nfsrv_recalllist so that recall of read/write layouts can
8293  *   be done.
8294  * - Unlock the MDS file's vnode, so that the client(s) can perform proxied
8295  *   writes, LayoutCommits and LayoutReturns for the file when completing the
8296  *   LayoutReturn requested by the LayoutRecall callback.
8297  * - Issue a LayoutRecall callback for all read/write layouts and wait for
8298  *   them to be returned. (If the LayoutRecall callback replies
8299  *   NFSERR_NOMATCHLAYOUT, they are gone and no LayoutReturn is needed.)
8300  * - Exclusively lock the MDS file's vnode.  This ensures that no proxied
8301  *   writes are in progress or can occur during the DS file copy.
8302  *   It also blocks Setattr operations.
8303  * - Create the file on the recovered mirror.
8304  * - Copy the file from the operational DS.
8305  * - Copy any ACL from the MDS file to the new DS file.
8306  * - Set the modify time of the new DS file to that of the MDS file.
8307  * - Update the extended attribute for the MDS file.
8308  * - Enable issuing of rw layouts by deleting the nfsdontlist entry.
8309  * - The caller will unlock the MDS file's vnode allowing operations
8310  *   to continue normally, since it is now on the mirror again.
8311  */
8312 int
8313 nfsrv_copymr(vnode_t vp, vnode_t fvp, vnode_t dvp, struct nfsdevice *ds,
8314     struct pnfsdsfile *pf, struct pnfsdsfile *wpf, int mirrorcnt,
8315     struct ucred *cred, NFSPROC_T *p)
8316 {
8317 	struct nfsdontlist *mrp, *nmrp;
8318 	struct nfslayouthash *lhyp;
8319 	struct nfslayout *lyp, *nlyp;
8320 	struct nfslayouthead thl;
8321 	struct mount *mp, *tvmp;
8322 	struct acl *aclp;
8323 	struct vattr va;
8324 	struct timespec mtime;
8325 	fhandle_t fh;
8326 	vnode_t tvp;
8327 	off_t rdpos, wrpos;
8328 	ssize_t aresid;
8329 	char *dat;
8330 	int didprintf, ret, retacl, xfer;
8331 
8332 	ASSERT_VOP_LOCKED(fvp, "nfsrv_copymr fvp");
8333 	ASSERT_VOP_LOCKED(vp, "nfsrv_copymr vp");
8334 	/*
8335 	 * Allocate a nfsdontlist entry and set the NFSMR_DONTLAYOUT flag
8336 	 * so that no more RW layouts will get issued.
8337 	 */
8338 	ret = nfsvno_getfh(vp, &fh, p);
8339 	if (ret != 0) {
8340 		NFSD_DEBUG(4, "nfsrv_copymr: getfh=%d\n", ret);
8341 		return (ret);
8342 	}
8343 	nmrp = malloc(sizeof(*nmrp), M_NFSDSTATE, M_WAITOK);
8344 	nmrp->nfsmr_flags = NFSMR_DONTLAYOUT;
8345 	NFSBCOPY(&fh, &nmrp->nfsmr_fh, sizeof(fh));
8346 	NFSDDONTLISTLOCK();
8347 	LIST_FOREACH(mrp, &nfsrv_dontlisthead, nfsmr_list) {
8348 		if (NFSBCMP(&fh, &mrp->nfsmr_fh, sizeof(fh)) == 0)
8349 			break;
8350 	}
8351 	if (mrp == NULL) {
8352 		LIST_INSERT_HEAD(&nfsrv_dontlisthead, nmrp, nfsmr_list);
8353 		mrp = nmrp;
8354 		nmrp = NULL;
8355 		nfsrv_dontlistlen++;
8356 		NFSD_DEBUG(4, "nfsrv_copymr: in dontlist\n");
8357 	} else {
8358 		NFSDDONTLISTUNLOCK();
8359 		free(nmrp, M_NFSDSTATE);
8360 		NFSD_DEBUG(4, "nfsrv_copymr: dup dontlist\n");
8361 		return (ENXIO);
8362 	}
8363 	NFSDDONTLISTUNLOCK();
8364 
8365 	/*
8366 	 * Search for all RW layouts for this file.  Move them to the
8367 	 * recall list, so they can be recalled and their return noted.
8368 	 */
8369 	lhyp = NFSLAYOUTHASH(&fh);
8370 	NFSDRECALLLOCK();
8371 	NFSLOCKLAYOUT(lhyp);
8372 	TAILQ_FOREACH_SAFE(lyp, &lhyp->list, lay_list, nlyp) {
8373 		if (NFSBCMP(&lyp->lay_fh, &fh, sizeof(fh)) == 0 &&
8374 		    (lyp->lay_flags & NFSLAY_RW) != 0) {
8375 			TAILQ_REMOVE(&lhyp->list, lyp, lay_list);
8376 			TAILQ_INSERT_HEAD(&nfsrv_recalllisthead, lyp, lay_list);
8377 			lyp->lay_trycnt = 0;
8378 		}
8379 	}
8380 	NFSUNLOCKLAYOUT(lhyp);
8381 	NFSDRECALLUNLOCK();
8382 
8383 	ret = 0;
8384 	mp = tvmp = NULL;
8385 	didprintf = 0;
8386 	TAILQ_INIT(&thl);
8387 	/* Unlock the MDS vp, so that a LayoutReturn can be done on it. */
8388 	NFSVOPUNLOCK(vp);
8389 	/* Now, do a recall for all layouts not yet recalled. */
8390 tryagain:
8391 	NFSDRECALLLOCK();
8392 	TAILQ_FOREACH(lyp, &nfsrv_recalllisthead, lay_list) {
8393 		if (NFSBCMP(&lyp->lay_fh, &fh, sizeof(fh)) == 0 &&
8394 		    (lyp->lay_flags & NFSLAY_RECALL) == 0) {
8395 			lyp->lay_flags |= NFSLAY_RECALL;
8396 			/*
8397 			 * The layout stateid.seqid needs to be incremented
8398 			 * before doing a LAYOUT_RECALL callback.
8399 			 */
8400 			if (++lyp->lay_stateid.seqid == 0)
8401 				lyp->lay_stateid.seqid = 1;
8402 			NFSDRECALLUNLOCK();
8403 			nfsrv_recalllayout(lyp->lay_clientid, &lyp->lay_stateid,
8404 			    &lyp->lay_fh, lyp, 0, lyp->lay_type, p);
8405 			NFSD_DEBUG(4, "nfsrv_copymr: recalled layout\n");
8406 			goto tryagain;
8407 		}
8408 	}
8409 
8410 	/* Now wait for them to be returned. */
8411 tryagain2:
8412 	TAILQ_FOREACH(lyp, &nfsrv_recalllisthead, lay_list) {
8413 		if (NFSBCMP(&lyp->lay_fh, &fh, sizeof(fh)) == 0) {
8414 			if ((lyp->lay_flags & NFSLAY_RETURNED) != 0) {
8415 				TAILQ_REMOVE(&nfsrv_recalllisthead, lyp,
8416 				    lay_list);
8417 				TAILQ_INSERT_HEAD(&thl, lyp, lay_list);
8418 				NFSD_DEBUG(4,
8419 				    "nfsrv_copymr: layout returned\n");
8420 			} else {
8421 				lyp->lay_trycnt++;
8422 				ret = mtx_sleep(lyp, NFSDRECALLMUTEXPTR,
8423 				    PVFS | PCATCH, "nfsmrl", hz);
8424 				NFSD_DEBUG(4, "nfsrv_copymr: aft sleep=%d\n",
8425 				    ret);
8426 				if (ret == EINTR || ret == ERESTART)
8427 					break;
8428 				if ((lyp->lay_flags & NFSLAY_RETURNED) == 0) {
8429 					/*
8430 					 * Give up after 60sec and return
8431 					 * ENXIO, failing the copymr.
8432 					 * This layout will remain on the
8433 					 * recalllist.  It can only be cleared
8434 					 * by restarting the nfsd.
8435 					 * This seems the safe way to handle
8436 					 * it, since it cannot be safely copied
8437 					 * with an outstanding RW layout.
8438 					 */
8439 					if (lyp->lay_trycnt >= 60) {
8440 						ret = ENXIO;
8441 						break;
8442 					}
8443 					if (didprintf == 0) {
8444 						printf("nfsrv_copymr: layout "
8445 						    "not returned\n");
8446 						didprintf = 1;
8447 					}
8448 				}
8449 			}
8450 			goto tryagain2;
8451 		}
8452 	}
8453 	NFSDRECALLUNLOCK();
8454 	/* We can now get rid of the layouts that have been returned. */
8455 	TAILQ_FOREACH_SAFE(lyp, &thl, lay_list, nlyp)
8456 		nfsrv_freelayout(&thl, lyp);
8457 
8458 	/*
8459 	 * Do the vn_start_write() calls here, before the MDS vnode is
8460 	 * locked and the tvp is created (locked) in the NFS file system
8461 	 * that dvp is in.
8462 	 * For tvmp, this probably isn't necessary, since it will be an
8463 	 * NFS mount and they are not suspendable at this time.
8464 	 */
8465 	if (ret == 0)
8466 		ret = vn_start_write(vp, &mp, V_WAIT | V_PCATCH);
8467 	if (ret == 0) {
8468 		tvmp = dvp->v_mount;
8469 		ret = vn_start_write(NULL, &tvmp, V_WAIT | V_PCATCH);
8470 	}
8471 
8472 	/*
8473 	 * LK_EXCLUSIVE lock the MDS vnode, so that any
8474 	 * proxied writes through the MDS will be blocked until we have
8475 	 * completed the copy and update of the extended attributes.
8476 	 * This will also ensure that any attributes and ACL will not be
8477 	 * changed until the copy is complete.
8478 	 */
8479 	NFSVOPLOCK(vp, LK_EXCLUSIVE | LK_RETRY);
8480 	if (ret == 0 && VN_IS_DOOMED(vp)) {
8481 		NFSD_DEBUG(4, "nfsrv_copymr: lk_exclusive doomed\n");
8482 		ret = ESTALE;
8483 	}
8484 
8485 	/* Create the data file on the recovered DS. */
8486 	if (ret == 0)
8487 		ret = nfsrv_createdsfile(vp, &fh, pf, dvp, ds, cred, p, &tvp);
8488 
8489 	/* Copy the DS file, if created successfully. */
8490 	if (ret == 0) {
8491 		/*
8492 		 * Get any NFSv4 ACL on the MDS file, so that it can be set
8493 		 * on the new DS file.
8494 		 */
8495 		aclp = acl_alloc(M_WAITOK | M_ZERO);
8496 		retacl = VOP_GETACL(vp, ACL_TYPE_NFS4, aclp, cred, p);
8497 		if (retacl != 0 && retacl != ENOATTR)
8498 			NFSD_DEBUG(1, "nfsrv_copymr: vop_getacl=%d\n", retacl);
8499 		dat = malloc(PNFSDS_COPYSIZ, M_TEMP, M_WAITOK);
8500 		/* Malloc a block of 0s used to check for holes. */
8501 		if (nfsrv_zeropnfsdat == NULL)
8502 			nfsrv_zeropnfsdat = malloc(PNFSDS_COPYSIZ, M_TEMP,
8503 			    M_WAITOK | M_ZERO);
8504 		rdpos = wrpos = 0;
8505 		ret = VOP_GETATTR(fvp, &va, cred);
8506 		aresid = 0;
8507 		while (ret == 0 && aresid == 0) {
8508 			ret = vn_rdwr(UIO_READ, fvp, dat, PNFSDS_COPYSIZ,
8509 			    rdpos, UIO_SYSSPACE, IO_NODELOCKED, cred, NULL,
8510 			    &aresid, p);
8511 			xfer = PNFSDS_COPYSIZ - aresid;
8512 			if (ret == 0 && xfer > 0) {
8513 				rdpos += xfer;
8514 				/*
8515 				 * Skip the write for holes, except for the
8516 				 * last block.
8517 				 */
8518 				if (xfer < PNFSDS_COPYSIZ || rdpos ==
8519 				    va.va_size || NFSBCMP(dat,
8520 				    nfsrv_zeropnfsdat, PNFSDS_COPYSIZ) != 0)
8521 					ret = vn_rdwr(UIO_WRITE, tvp, dat, xfer,
8522 					    wrpos, UIO_SYSSPACE, IO_NODELOCKED,
8523 					    cred, NULL, NULL, p);
8524 				if (ret == 0)
8525 					wrpos += xfer;
8526 			}
8527 		}
8528 
8529 		/* If there is an ACL and the copy succeeded, set the ACL. */
8530 		if (ret == 0 && retacl == 0) {
8531 			ret = VOP_SETACL(tvp, ACL_TYPE_NFS4, aclp, cred, p);
8532 			/*
8533 			 * Don't consider these as errors, since VOP_GETACL()
8534 			 * can return an ACL when they are not actually
8535 			 * supported.  For example, for UFS, VOP_GETACL()
8536 			 * will return a trivial ACL based on the uid/gid/mode
8537 			 * when there is no ACL on the file.
8538 			 * This case should be recognized as a trivial ACL
8539 			 * by UFS's VOP_SETACL() and succeed, but...
8540 			 */
8541 			if (ret == ENOATTR || ret == EOPNOTSUPP || ret == EPERM)
8542 				ret = 0;
8543 		}
8544 
8545 		if (ret == 0)
8546 			ret = VOP_FSYNC(tvp, MNT_WAIT, p);
8547 
8548 		/* Set the DS data file's modify time that of the MDS file. */
8549 		if (ret == 0)
8550 			ret = VOP_GETATTR(vp, &va, cred);
8551 		if (ret == 0) {
8552 			mtime = va.va_mtime;
8553 			VATTR_NULL(&va);
8554 			va.va_mtime = mtime;
8555 			ret = VOP_SETATTR(tvp, &va, cred);
8556 		}
8557 
8558 		vput(tvp);
8559 		acl_free(aclp);
8560 		free(dat, M_TEMP);
8561 	}
8562 	if (tvmp != NULL)
8563 		vn_finished_write(tvmp);
8564 
8565 	/* Update the extended attributes for the newly created DS file. */
8566 	if (ret == 0)
8567 		ret = vn_extattr_set(vp, IO_NODELOCKED,
8568 		    EXTATTR_NAMESPACE_SYSTEM, "pnfsd.dsfile",
8569 		    sizeof(*wpf) * mirrorcnt, (char *)wpf, p);
8570 	if (mp != NULL)
8571 		vn_finished_write(mp);
8572 
8573 	/* Get rid of the dontlist entry, so that Layouts can be issued. */
8574 	NFSDDONTLISTLOCK();
8575 	LIST_REMOVE(mrp, nfsmr_list);
8576 	NFSDDONTLISTUNLOCK();
8577 	free(mrp, M_NFSDSTATE);
8578 	return (ret);
8579 }
8580 
8581 /*
8582  * Create a data storage file on the recovered DS.
8583  */
8584 static int
8585 nfsrv_createdsfile(vnode_t vp, fhandle_t *fhp, struct pnfsdsfile *pf,
8586     vnode_t dvp, struct nfsdevice *ds, struct ucred *cred, NFSPROC_T *p,
8587     vnode_t *tvpp)
8588 {
8589 	struct vattr va, nva;
8590 	int error;
8591 
8592 	/* Make data file name based on FH. */
8593 	error = VOP_GETATTR(vp, &va, cred);
8594 	if (error == 0) {
8595 		/* Set the attributes for "vp" to Setattr the DS vp. */
8596 		VATTR_NULL(&nva);
8597 		nva.va_uid = va.va_uid;
8598 		nva.va_gid = va.va_gid;
8599 		nva.va_mode = va.va_mode;
8600 		nva.va_size = 0;
8601 		VATTR_NULL(&va);
8602 		va.va_type = VREG;
8603 		va.va_mode = nva.va_mode;
8604 		NFSD_DEBUG(4, "nfsrv_dscreatefile: dvp=%p pf=%p\n", dvp, pf);
8605 		error = nfsrv_dscreate(dvp, &va, &nva, fhp, pf, NULL,
8606 		    pf->dsf_filename, cred, p, tvpp);
8607 	}
8608 	return (error);
8609 }
8610 
8611 /*
8612  * Look up the MDS file shared locked, and then get the extended attribute
8613  * to find the extant DS file to be copied to the new mirror.
8614  * If successful, *vpp is set to the MDS file's vp and *nvpp is
8615  * set to a DS data file for the MDS file, both exclusively locked.
8616  * The "buf" argument has the pnfsdsfile structure from the MDS file
8617  * in it and buflen is set to its length.
8618  */
8619 int
8620 nfsrv_mdscopymr(char *mdspathp, char *dspathp, char *curdspathp, char *buf,
8621     int *buflenp, char *fname, NFSPROC_T *p, struct vnode **vpp,
8622     struct vnode **nvpp, struct pnfsdsfile **pfp, struct nfsdevice **dsp,
8623     struct nfsdevice **fdsp)
8624 {
8625 	struct nameidata nd;
8626 	struct vnode *vp, *curvp;
8627 	struct pnfsdsfile *pf;
8628 	struct nfsmount *nmp, *curnmp;
8629 	int dsdir, error, mirrorcnt, ippos;
8630 
8631 	vp = NULL;
8632 	curvp = NULL;
8633 	curnmp = NULL;
8634 	*dsp = NULL;
8635 	*fdsp = NULL;
8636 	if (dspathp == NULL && curdspathp != NULL)
8637 		return (EPERM);
8638 
8639 	/*
8640 	 * Look up the MDS file shared locked.  The lock will be upgraded
8641 	 * to an exclusive lock after any rw layouts have been returned.
8642 	 */
8643 	NFSD_DEBUG(4, "mdsopen path=%s\n", mdspathp);
8644 	NDINIT(&nd, LOOKUP, FOLLOW | LOCKSHARED | LOCKLEAF, UIO_SYSSPACE,
8645 	    mdspathp);
8646 	error = namei(&nd);
8647 	NFSD_DEBUG(4, "lookup=%d\n", error);
8648 	if (error != 0)
8649 		return (error);
8650 	if (nd.ni_vp->v_type != VREG) {
8651 		vput(nd.ni_vp);
8652 		NFSD_DEBUG(4, "mdspath not reg\n");
8653 		return (EISDIR);
8654 	}
8655 	vp = nd.ni_vp;
8656 
8657 	if (curdspathp != NULL) {
8658 		/*
8659 		 * Look up the current DS path and find the nfsdev structure for
8660 		 * it.
8661 		 */
8662 		NFSD_DEBUG(4, "curmdsdev path=%s\n", curdspathp);
8663 		NDINIT(&nd, LOOKUP, FOLLOW | LOCKSHARED | LOCKLEAF,
8664 		    UIO_SYSSPACE, curdspathp);
8665 		error = namei(&nd);
8666 		NFSD_DEBUG(4, "ds lookup=%d\n", error);
8667 		if (error != 0) {
8668 			vput(vp);
8669 			return (error);
8670 		}
8671 		if (nd.ni_vp->v_type != VDIR) {
8672 			vput(nd.ni_vp);
8673 			vput(vp);
8674 			NFSD_DEBUG(4, "curdspath not dir\n");
8675 			return (ENOTDIR);
8676 		}
8677 		if (strcmp(nd.ni_vp->v_mount->mnt_vfc->vfc_name, "nfs") != 0) {
8678 			vput(nd.ni_vp);
8679 			vput(vp);
8680 			NFSD_DEBUG(4, "curdspath not an NFS mount\n");
8681 			return (ENXIO);
8682 		}
8683 		curnmp = VFSTONFS(nd.ni_vp->v_mount);
8684 
8685 		/* Search the nfsdev list for a match. */
8686 		NFSDDSLOCK();
8687 		*fdsp = nfsv4_findmirror(curnmp);
8688 		NFSDDSUNLOCK();
8689 		if (*fdsp == NULL)
8690 			curnmp = NULL;
8691 		if (curnmp == NULL) {
8692 			vput(nd.ni_vp);
8693 			vput(vp);
8694 			NFSD_DEBUG(4, "mdscopymr: no current ds\n");
8695 			return (ENXIO);
8696 		}
8697 		curvp = nd.ni_vp;
8698 	}
8699 
8700 	if (dspathp != NULL) {
8701 		/* Look up the nfsdev path and find the nfsdev structure. */
8702 		NFSD_DEBUG(4, "mdsdev path=%s\n", dspathp);
8703 		NDINIT(&nd, LOOKUP, FOLLOW | LOCKSHARED | LOCKLEAF,
8704 		    UIO_SYSSPACE, dspathp);
8705 		error = namei(&nd);
8706 		NFSD_DEBUG(4, "ds lookup=%d\n", error);
8707 		if (error != 0) {
8708 			vput(vp);
8709 			if (curvp != NULL)
8710 				vput(curvp);
8711 			return (error);
8712 		}
8713 		if (nd.ni_vp->v_type != VDIR || nd.ni_vp == curvp) {
8714 			vput(nd.ni_vp);
8715 			vput(vp);
8716 			if (curvp != NULL)
8717 				vput(curvp);
8718 			NFSD_DEBUG(4, "dspath not dir\n");
8719 			if (nd.ni_vp == curvp)
8720 				return (EPERM);
8721 			return (ENOTDIR);
8722 		}
8723 		if (strcmp(nd.ni_vp->v_mount->mnt_vfc->vfc_name, "nfs") != 0) {
8724 			vput(nd.ni_vp);
8725 			vput(vp);
8726 			if (curvp != NULL)
8727 				vput(curvp);
8728 			NFSD_DEBUG(4, "dspath not an NFS mount\n");
8729 			return (ENXIO);
8730 		}
8731 		nmp = VFSTONFS(nd.ni_vp->v_mount);
8732 
8733 		/*
8734 		 * Search the nfsdevice list for a match.  If curnmp == NULL,
8735 		 * this is a recovery and there must be a mirror.
8736 		 */
8737 		NFSDDSLOCK();
8738 		if (curnmp == NULL)
8739 			*dsp = nfsrv_findmirroredds(nmp);
8740 		else
8741 			*dsp = nfsv4_findmirror(nmp);
8742 		NFSDDSUNLOCK();
8743 		if (*dsp == NULL) {
8744 			vput(nd.ni_vp);
8745 			vput(vp);
8746 			if (curvp != NULL)
8747 				vput(curvp);
8748 			NFSD_DEBUG(4, "mdscopymr: no ds\n");
8749 			return (ENXIO);
8750 		}
8751 	} else {
8752 		nd.ni_vp = NULL;
8753 		nmp = NULL;
8754 	}
8755 
8756 	/*
8757 	 * Get a vp for an available DS data file using the extended
8758 	 * attribute on the MDS file.
8759 	 * If there is a valid entry for the new DS in the extended attribute
8760 	 * on the MDS file (as checked via the nmp argument),
8761 	 * nfsrv_dsgetsockmnt() returns EEXIST, so no copying will occur.
8762 	 */
8763 	error = nfsrv_dsgetsockmnt(vp, 0, buf, buflenp, &mirrorcnt, p,
8764 	    NULL, NULL, NULL, fname, nvpp, &nmp, curnmp, &ippos, &dsdir);
8765 	if (curvp != NULL)
8766 		vput(curvp);
8767 	if (nd.ni_vp == NULL) {
8768 		if (error == 0 && nmp != NULL) {
8769 			/* Search the nfsdev list for a match. */
8770 			NFSDDSLOCK();
8771 			*dsp = nfsrv_findmirroredds(nmp);
8772 			NFSDDSUNLOCK();
8773 		}
8774 		if (error == 0 && (nmp == NULL || *dsp == NULL)) {
8775 			if (nvpp != NULL && *nvpp != NULL) {
8776 				vput(*nvpp);
8777 				*nvpp = NULL;
8778 			}
8779 			error = ENXIO;
8780 		}
8781 	} else
8782 		vput(nd.ni_vp);
8783 
8784 	/*
8785 	 * When dspathp != NULL and curdspathp == NULL, this is a recovery
8786 	 * and is only allowed if there is a 0.0.0.0 IP address entry.
8787 	 * When curdspathp != NULL, the ippos will be set to that entry.
8788 	 */
8789 	if (error == 0 && dspathp != NULL && ippos == -1) {
8790 		if (nvpp != NULL && *nvpp != NULL) {
8791 			vput(*nvpp);
8792 			*nvpp = NULL;
8793 		}
8794 		error = ENXIO;
8795 	}
8796 	if (error == 0) {
8797 		*vpp = vp;
8798 
8799 		pf = (struct pnfsdsfile *)buf;
8800 		if (ippos == -1) {
8801 			/* If no zeroip pnfsdsfile, add one. */
8802 			ippos = *buflenp / sizeof(*pf);
8803 			*buflenp += sizeof(*pf);
8804 			pf += ippos;
8805 			pf->dsf_dir = dsdir;
8806 			strlcpy(pf->dsf_filename, fname,
8807 			    sizeof(pf->dsf_filename));
8808 		} else
8809 			pf += ippos;
8810 		*pfp = pf;
8811 	} else
8812 		vput(vp);
8813 	return (error);
8814 }
8815 
8816 /*
8817  * Search for a matching pnfsd mirror device structure, base on the nmp arg.
8818  * Return one if found, NULL otherwise.
8819  */
8820 static struct nfsdevice *
8821 nfsrv_findmirroredds(struct nfsmount *nmp)
8822 {
8823 	struct nfsdevice *ds, *fndds;
8824 	int fndmirror;
8825 
8826 	mtx_assert(NFSDDSMUTEXPTR, MA_OWNED);
8827 	/*
8828 	 * Search the DS server list for a match with nmp.
8829 	 * Remove the DS entry if found and there is a mirror.
8830 	 */
8831 	fndds = NULL;
8832 	fndmirror = 0;
8833 	if (nfsrv_devidcnt == 0)
8834 		return (fndds);
8835 	TAILQ_FOREACH(ds, &nfsrv_devidhead, nfsdev_list) {
8836 		if (ds->nfsdev_nmp == nmp) {
8837 			NFSD_DEBUG(4, "nfsrv_findmirroredds: fnd main ds\n");
8838 			fndds = ds;
8839 			break;
8840 		}
8841 	}
8842 	if (fndds == NULL)
8843 		return (fndds);
8844 	if (fndds->nfsdev_mdsisset == 0 && nfsrv_faildscnt > 0)
8845 		fndmirror = 1;
8846 	else if (fndds->nfsdev_mdsisset != 0) {
8847 		/* For the fsid is set case, search for a mirror. */
8848 		TAILQ_FOREACH(ds, &nfsrv_devidhead, nfsdev_list) {
8849 			if (ds != fndds && ds->nfsdev_nmp != NULL &&
8850 			    ds->nfsdev_mdsisset != 0 &&
8851 			    fsidcmp(&ds->nfsdev_mdsfsid,
8852 			    &fndds->nfsdev_mdsfsid) == 0) {
8853 				fndmirror = 1;
8854 				break;
8855 			}
8856 		}
8857 	}
8858 	if (fndmirror == 0) {
8859 		NFSD_DEBUG(4, "nfsrv_findmirroredds: no mirror for DS\n");
8860 		return (NULL);
8861 	}
8862 	return (fndds);
8863 }
8864 
8865 /*
8866  * Mark the appropriate devid and all associated layout as "out of space".
8867  */
8868 void
8869 nfsrv_marknospc(char *devid, bool setit)
8870 {
8871 	struct nfsdevice *ds;
8872 	struct nfslayout *lyp;
8873 	struct nfslayouthash *lhyp;
8874 	int i;
8875 
8876 	NFSDDSLOCK();
8877 	TAILQ_FOREACH(ds, &nfsrv_devidhead, nfsdev_list) {
8878 		if (NFSBCMP(ds->nfsdev_deviceid, devid, NFSX_V4DEVICEID) == 0) {
8879 			NFSD_DEBUG(1, "nfsrv_marknospc: devid %d\n", setit);
8880 			ds->nfsdev_nospc = setit;
8881 		}
8882 	}
8883 	NFSDDSUNLOCK();
8884 
8885 	for (i = 0; i < nfsrv_layouthashsize; i++) {
8886 		lhyp = &nfslayouthash[i];
8887 		NFSLOCKLAYOUT(lhyp);
8888 		TAILQ_FOREACH(lyp, &lhyp->list, lay_list) {
8889 			if (NFSBCMP(lyp->lay_deviceid, devid,
8890 			    NFSX_V4DEVICEID) == 0) {
8891 				NFSD_DEBUG(1, "nfsrv_marknospc: layout %d\n",
8892 				    setit);
8893 				if (setit)
8894 					lyp->lay_flags |= NFSLAY_NOSPC;
8895 				else
8896 					lyp->lay_flags &= ~NFSLAY_NOSPC;
8897 			}
8898 		}
8899 		NFSUNLOCKLAYOUT(lhyp);
8900 	}
8901 }
8902 
8903 /*
8904  * Check to see if SP4_MACH_CRED is in use and, if it is, check that the
8905  * correct machine credential is being used.
8906  */
8907 static int
8908 nfsrv_checkmachcred(int op, struct nfsrv_descript *nd, struct nfsclient *clp)
8909 {
8910 
8911 	if ((clp->lc_flags & LCL_MACHCRED) == 0 ||
8912 	    !NFSISSET_OPBIT(&clp->lc_mustops, op))
8913 		return (0);
8914 	KASSERT((nd->nd_flag & ND_NFSV41) != 0,
8915 	    ("nfsrv_checkmachcred: MachCred for NFSv4.0"));
8916 	if ((nd->nd_flag & (ND_GSSINTEGRITY | ND_GSSPRIVACY)) != 0 &&
8917 	    nd->nd_princlen == clp->lc_namelen &&
8918 	    !NFSBCMP(nd->nd_principal, clp->lc_name, nd->nd_princlen))
8919 		return (0);
8920 	return (NFSERR_AUTHERR | AUTH_TOOWEAK);
8921 }
8922 
8923 /*
8924  * Issue a delegation and, optionally set rflagsp for why not.
8925  */
8926 static void
8927 nfsrv_issuedelegation(struct vnode *vp, struct nfsclient *clp,
8928     struct nfsrv_descript *nd, int delegate, int writedeleg, int readonly,
8929     u_quad_t filerev, uint64_t rdonly, struct nfsstate **new_delegp,
8930     struct nfsstate *new_stp, struct nfslockfile *lfp, uint32_t *rflagsp,
8931     nfsv4stateid_t *delegstateidp)
8932 {
8933 	struct nfsstate *up_deleg, *new_deleg;
8934 
8935 	new_deleg = *new_delegp;
8936 	up_deleg = LIST_FIRST(&lfp->lf_deleg);
8937 	if ((new_stp->ls_flags & NFSLCK_WANTNODELEG) != 0)
8938 		*rflagsp |= NFSV4OPEN_WDNOTWANTED;
8939 	else if (nfsrv_issuedelegs == 0)
8940 		*rflagsp |= NFSV4OPEN_WDSUPPFTYPE;
8941 	else if (NFSRV_V4DELEGLIMIT(nfsrv_delegatecnt))
8942 		*rflagsp |= NFSV4OPEN_WDRESOURCE;
8943 	else if (delegate == 0 || !NFSVNO_DELEGOK(vp) ||
8944 	    (writedeleg == 0 && (readonly == 0 ||
8945 	    (new_stp->ls_flags & NFSLCK_WANTWDELEG) != 0)) ||
8946 	    (clp->lc_flags & (LCL_CALLBACKSON | LCL_CBDOWN)) !=
8947 	     LCL_CALLBACKSON) {
8948 		/* Is this a downgrade attempt? */
8949 		if (up_deleg != NULL && up_deleg->ls_clp == clp &&
8950 		    (up_deleg->ls_flags & NFSLCK_DELEGWRITE) != 0 &&
8951 		    (new_stp->ls_flags & NFSLCK_WANTRDELEG) != 0)
8952 			*rflagsp |= NFSV4OPEN_WDNOTSUPPDOWNGRADE;
8953 		else
8954 			*rflagsp |= NFSV4OPEN_WDCONTENTION;
8955 	} else if (up_deleg != NULL &&
8956 	    (up_deleg->ls_flags & NFSLCK_DELEGREAD) != 0 &&
8957 	    (new_stp->ls_flags & NFSLCK_WANTWDELEG) != 0) {
8958 		/* This is an atomic upgrade. */
8959 		up_deleg->ls_stateid.seqid++;
8960 		delegstateidp->seqid = up_deleg->ls_stateid.seqid;
8961 		delegstateidp->other[0] = up_deleg->ls_stateid.other[0];
8962 		delegstateidp->other[1] = up_deleg->ls_stateid.other[1];
8963 		delegstateidp->other[2] = up_deleg->ls_stateid.other[2];
8964 		up_deleg->ls_flags = (NFSLCK_DELEGWRITE |
8965 		    NFSLCK_READACCESS | NFSLCK_WRITEACCESS);
8966 		*rflagsp |= NFSV4OPEN_WRITEDELEGATE;
8967 		nfsrv_writedelegcnt++;
8968 	} else {
8969 		new_deleg->ls_stateid.seqid = delegstateidp->seqid = 1;
8970 		new_deleg->ls_stateid.other[0] = delegstateidp->other[0]
8971 		    = clp->lc_clientid.lval[0];
8972 		new_deleg->ls_stateid.other[1] = delegstateidp->other[1]
8973 		    = clp->lc_clientid.lval[1];
8974 		new_deleg->ls_stateid.other[2] = delegstateidp->other[2]
8975 		    = nfsrv_nextstateindex(clp);
8976 		if (writedeleg && !rdonly &&
8977 		    (nfsrv_writedelegifpos || !readonly) &&
8978 		    (new_stp->ls_flags & (NFSLCK_WANTRDELEG |
8979 		     NFSLCK_WANTWDELEG)) != NFSLCK_WANTRDELEG) {
8980 			new_deleg->ls_flags = (NFSLCK_DELEGWRITE |
8981 			    NFSLCK_READACCESS | NFSLCK_WRITEACCESS);
8982 			*rflagsp |= NFSV4OPEN_WRITEDELEGATE;
8983 			nfsrv_writedelegcnt++;
8984 		} else {
8985 			new_deleg->ls_flags = (NFSLCK_DELEGREAD |
8986 			    NFSLCK_READACCESS);
8987 			*rflagsp |= NFSV4OPEN_READDELEGATE;
8988 		}
8989 		new_deleg->ls_uid = new_stp->ls_uid;
8990 		new_deleg->ls_lfp = lfp;
8991 		new_deleg->ls_clp = clp;
8992 		new_deleg->ls_filerev = filerev;
8993 		new_deleg->ls_compref = nd->nd_compref;
8994 		new_deleg->ls_lastrecall = 0;
8995 		LIST_INSERT_HEAD(&lfp->lf_deleg, new_deleg, ls_file);
8996 		LIST_INSERT_HEAD(NFSSTATEHASH(clp, new_deleg->ls_stateid),
8997 		    new_deleg, ls_hash);
8998 		LIST_INSERT_HEAD(&clp->lc_deleg, new_deleg, ls_list);
8999 		*new_delegp = NULL;
9000 		NFSD_VNET(nfsstatsv1_p)->srvdelegates++;
9001 		nfsrv_openpluslock++;
9002 		nfsrv_delegatecnt++;
9003 	}
9004 }
9005