xref: /freebsd/sys/fs/nfsserver/nfs_nfsdstate.c (revision 1fb7d2cf999e52e3682174d0c2f20cb3baf414f3)
1 /*-
2  * SPDX-License-Identifier: BSD-2-Clause-FreeBSD
3  *
4  * Copyright (c) 2009 Rick Macklem, University of Guelph
5  * All rights reserved.
6  *
7  * Redistribution and use in source and binary forms, with or without
8  * modification, are permitted provided that the following conditions
9  * are met:
10  * 1. Redistributions of source code must retain the above copyright
11  *    notice, this list of conditions and the following disclaimer.
12  * 2. Redistributions in binary form must reproduce the above copyright
13  *    notice, this list of conditions and the following disclaimer in the
14  *    documentation and/or other materials provided with the distribution.
15  *
16  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
17  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
18  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
19  * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
20  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
21  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
22  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
23  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
24  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
25  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
26  * SUCH DAMAGE.
27  *
28  */
29 
30 #include <sys/cdefs.h>
31 __FBSDID("$FreeBSD$");
32 
33 #include "opt_inet.h"
34 #include "opt_inet6.h"
35 #include <sys/extattr.h>
36 #include <fs/nfs/nfsport.h>
37 
38 int nfsrv_issuedelegs = 0;
39 int nfsrv_dolocallocks = 0;
40 struct nfsv4lock nfsv4rootfs_lock;
41 time_t nfsdev_time = 0;
42 int nfsrv_layouthashsize;
43 volatile int nfsrv_layoutcnt = 0;
44 
45 NFSD_VNET_DEFINE(struct nfsrv_stablefirst, nfsrv_stablefirst);
46 
47 NFSD_VNET_DECLARE(int, nfsrv_numnfsd);
48 NFSD_VNET_DECLARE(struct nfsstatsv1 *, nfsstatsv1_p);
49 
50 extern uint32_t nfs_srvmaxio;
51 extern int nfsrv_lease;
52 extern struct timeval nfsboottime;
53 extern u_int32_t newnfs_true, newnfs_false;
54 extern struct mtx nfsrv_dslock_mtx;
55 extern struct mtx nfsrv_recalllock_mtx;
56 extern struct mtx nfsrv_dontlistlock_mtx;
57 extern int nfsd_debuglevel;
58 extern u_int nfsrv_dsdirsize;
59 extern struct nfsdevicehead nfsrv_devidhead;
60 extern int nfsrv_doflexfile;
61 extern int nfsrv_maxpnfsmirror;
62 NFSV4ROOTLOCKMUTEX;
63 NFSSTATESPINLOCK;
64 extern struct nfsdontlisthead nfsrv_dontlisthead;
65 extern volatile int nfsrv_devidcnt;
66 extern struct nfslayouthead nfsrv_recalllisthead;
67 extern char *nfsrv_zeropnfsdat;
68 
69 SYSCTL_DECL(_vfs_nfsd);
70 int	nfsrv_statehashsize = NFSSTATEHASHSIZE;
71 SYSCTL_INT(_vfs_nfsd, OID_AUTO, statehashsize, CTLFLAG_RDTUN,
72     &nfsrv_statehashsize, 0,
73     "Size of state hash table set via loader.conf");
74 
75 int	nfsrv_clienthashsize = NFSCLIENTHASHSIZE;
76 SYSCTL_INT(_vfs_nfsd, OID_AUTO, clienthashsize, CTLFLAG_RDTUN,
77     &nfsrv_clienthashsize, 0,
78     "Size of client hash table set via loader.conf");
79 
80 int	nfsrv_lockhashsize = NFSLOCKHASHSIZE;
81 SYSCTL_INT(_vfs_nfsd, OID_AUTO, fhhashsize, CTLFLAG_RDTUN,
82     &nfsrv_lockhashsize, 0,
83     "Size of file handle hash table set via loader.conf");
84 
85 int	nfsrv_sessionhashsize = NFSSESSIONHASHSIZE;
86 SYSCTL_INT(_vfs_nfsd, OID_AUTO, sessionhashsize, CTLFLAG_RDTUN,
87     &nfsrv_sessionhashsize, 0,
88     "Size of session hash table set via loader.conf");
89 
90 int	nfsrv_layouthighwater = NFSLAYOUTHIGHWATER;
91 SYSCTL_INT(_vfs_nfsd, OID_AUTO, layouthighwater, CTLFLAG_RDTUN,
92     &nfsrv_layouthighwater, 0,
93     "High water mark for number of layouts set via loader.conf");
94 
95 static int	nfsrv_v4statelimit = NFSRV_V4STATELIMIT;
96 SYSCTL_INT(_vfs_nfsd, OID_AUTO, v4statelimit, CTLFLAG_RWTUN,
97     &nfsrv_v4statelimit, 0,
98     "High water limit for NFSv4 opens+locks+delegations");
99 
100 static int	nfsrv_writedelegifpos = 0;
101 SYSCTL_INT(_vfs_nfsd, OID_AUTO, writedelegifpos, CTLFLAG_RW,
102     &nfsrv_writedelegifpos, 0,
103     "Issue a write delegation for read opens if possible");
104 
105 static int	nfsrv_allowreadforwriteopen = 1;
106 SYSCTL_INT(_vfs_nfsd, OID_AUTO, allowreadforwriteopen, CTLFLAG_RW,
107     &nfsrv_allowreadforwriteopen, 0,
108     "Allow Reads to be done with Write Access StateIDs");
109 
110 int	nfsrv_pnfsatime = 0;
111 SYSCTL_INT(_vfs_nfsd, OID_AUTO, pnfsstrictatime, CTLFLAG_RW,
112     &nfsrv_pnfsatime, 0,
113     "For pNFS service, do Getattr ops to keep atime up-to-date");
114 
115 int	nfsrv_flexlinuxhack = 0;
116 SYSCTL_INT(_vfs_nfsd, OID_AUTO, flexlinuxhack, CTLFLAG_RW,
117     &nfsrv_flexlinuxhack, 0,
118     "For Linux clients, hack around Flex File Layout bug");
119 
120 /*
121  * Hash lists for nfs V4.
122  */
123 NFSD_VNET_DEFINE(struct nfsclienthashhead *, nfsclienthash);
124 NFSD_VNET_DEFINE(struct nfslockhashhead *, nfslockhash);
125 NFSD_VNET_DEFINE(struct nfssessionhash *, nfssessionhash);
126 
127 struct nfslayouthash		*nfslayouthash;
128 volatile int nfsrv_dontlistlen = 0;
129 
130 static u_int32_t nfsrv_openpluslock = 0, nfsrv_delegatecnt = 0;
131 static int nfsrv_returnoldstateid = 0, nfsrv_clients = 0;
132 static int nfsrv_clienthighwater = NFSRV_CLIENTHIGHWATER;
133 static int nfsrv_nogsscallback = 0;
134 static volatile int nfsrv_writedelegcnt = 0;
135 static int nfsrv_faildscnt;
136 
137 NFSD_VNET_DEFINE_STATIC(time_t, nfsrvboottime);
138 
139 /* local functions */
140 static void nfsrv_dumpaclient(struct nfsclient *clp,
141     struct nfsd_dumpclients *dumpp);
142 static void nfsrv_freeopenowner(struct nfsstate *stp, int cansleep,
143     NFSPROC_T *p);
144 static int nfsrv_freeopen(struct nfsstate *stp, vnode_t vp, int cansleep,
145     NFSPROC_T *p);
146 static void nfsrv_freelockowner(struct nfsstate *stp, vnode_t vp, int cansleep,
147     NFSPROC_T *p);
148 static void nfsrv_freeallnfslocks(struct nfsstate *stp, vnode_t vp,
149     int cansleep, NFSPROC_T *p);
150 static void nfsrv_freenfslock(struct nfslock *lop);
151 static void nfsrv_freenfslockfile(struct nfslockfile *lfp);
152 static void nfsrv_freedeleg(struct nfsstate *);
153 static int nfsrv_getstate(struct nfsclient *clp, nfsv4stateid_t *stateidp,
154     u_int32_t flags, struct nfsstate **stpp);
155 static void nfsrv_getowner(struct nfsstatehead *hp, struct nfsstate *new_stp,
156     struct nfsstate **stpp);
157 static int nfsrv_getlockfh(vnode_t vp, u_short flags,
158     struct nfslockfile *new_lfp, fhandle_t *nfhp, NFSPROC_T *p);
159 static int nfsrv_getlockfile(u_short flags, struct nfslockfile **new_lfpp,
160     struct nfslockfile **lfpp, fhandle_t *nfhp, int lockit);
161 static void nfsrv_insertlock(struct nfslock *new_lop,
162     struct nfslock *insert_lop, struct nfsstate *stp, struct nfslockfile *lfp);
163 static void nfsrv_updatelock(struct nfsstate *stp, struct nfslock **new_lopp,
164     struct nfslock **other_lopp, struct nfslockfile *lfp);
165 static int nfsrv_getipnumber(u_char *cp);
166 static int nfsrv_checkrestart(nfsquad_t clientid, u_int32_t flags,
167     nfsv4stateid_t *stateidp, int specialid);
168 static int nfsrv_checkgrace(struct nfsrv_descript *nd, struct nfsclient *clp,
169     u_int32_t flags);
170 static int nfsrv_docallback(struct nfsclient *clp, int procnum,
171     nfsv4stateid_t *stateidp, int trunc, fhandle_t *fhp,
172     struct nfsvattr *nap, nfsattrbit_t *attrbitp, int laytype, NFSPROC_T *p);
173 static int nfsrv_cbcallargs(struct nfsrv_descript *nd, struct nfsclient *clp,
174     uint32_t callback, int op, const char *optag, struct nfsdsession **sepp,
175     int *slotposp);
176 static u_int32_t nfsrv_nextclientindex(void);
177 static u_int32_t nfsrv_nextstateindex(struct nfsclient *clp);
178 static void nfsrv_markstable(struct nfsclient *clp);
179 static void nfsrv_markreclaim(struct nfsclient *clp);
180 static int nfsrv_checkstable(struct nfsclient *clp);
181 static int nfsrv_clientconflict(struct nfsclient *clp, int *haslockp, struct
182     vnode *vp, NFSPROC_T *p);
183 static int nfsrv_delegconflict(struct nfsstate *stp, int *haslockp,
184     NFSPROC_T *p, vnode_t vp);
185 static int nfsrv_cleandeleg(vnode_t vp, struct nfslockfile *lfp,
186     struct nfsclient *clp, int *haslockp, NFSPROC_T *p);
187 static int nfsrv_notsamecredname(struct nfsrv_descript *nd,
188     struct nfsclient *clp);
189 static time_t nfsrv_leaseexpiry(void);
190 static void nfsrv_delaydelegtimeout(struct nfsstate *stp);
191 static int nfsrv_checkseqid(struct nfsrv_descript *nd, u_int32_t seqid,
192     struct nfsstate *stp, struct nfsrvcache *op);
193 static int nfsrv_nootherstate(struct nfsstate *stp);
194 static int nfsrv_locallock(vnode_t vp, struct nfslockfile *lfp, int flags,
195     uint64_t first, uint64_t end, struct nfslockconflict *cfp, NFSPROC_T *p);
196 static void nfsrv_localunlock(vnode_t vp, struct nfslockfile *lfp,
197     uint64_t init_first, uint64_t init_end, NFSPROC_T *p);
198 static int nfsrv_dolocal(vnode_t vp, struct nfslockfile *lfp, int flags,
199     int oldflags, uint64_t first, uint64_t end, struct nfslockconflict *cfp,
200     NFSPROC_T *p);
201 static void nfsrv_locallock_rollback(vnode_t vp, struct nfslockfile *lfp,
202     NFSPROC_T *p);
203 static void nfsrv_locallock_commit(struct nfslockfile *lfp, int flags,
204     uint64_t first, uint64_t end);
205 static void nfsrv_locklf(struct nfslockfile *lfp);
206 static void nfsrv_unlocklf(struct nfslockfile *lfp);
207 static struct nfsdsession *nfsrv_findsession(uint8_t *sessionid);
208 static int nfsrv_freesession(struct nfsdsession *sep, uint8_t *sessionid);
209 static int nfsv4_setcbsequence(struct nfsrv_descript *nd, struct nfsclient *clp,
210     int dont_replycache, struct nfsdsession **sepp, int *slotposp);
211 static int nfsv4_getcbsession(struct nfsclient *clp, struct nfsdsession **sepp);
212 static int nfsrv_addlayout(struct nfsrv_descript *nd, struct nfslayout **lypp,
213     nfsv4stateid_t *stateidp, char *layp, int *layoutlenp, NFSPROC_T *p);
214 static void nfsrv_freelayout(struct nfslayouthead *lhp, struct nfslayout *lyp);
215 static void nfsrv_freelayoutlist(nfsquad_t clientid);
216 static void nfsrv_freelayouts(nfsquad_t *clid, fsid_t *fs, int laytype,
217     int iomode);
218 static void nfsrv_freealllayouts(void);
219 static void nfsrv_freedevid(struct nfsdevice *ds);
220 static int nfsrv_setdsserver(char *dspathp, char *mdspathp, NFSPROC_T *p,
221     struct nfsdevice **dsp);
222 static void nfsrv_deleteds(struct nfsdevice *fndds);
223 static void nfsrv_allocdevid(struct nfsdevice *ds, char *addr, char *dnshost);
224 static void nfsrv_freealldevids(void);
225 static void nfsrv_flexlayouterr(struct nfsrv_descript *nd, uint32_t *layp,
226     int maxcnt, NFSPROC_T *p);
227 static int nfsrv_recalllayout(nfsquad_t clid, nfsv4stateid_t *stateidp,
228     fhandle_t *fhp, struct nfslayout *lyp, int changed, int laytype,
229     NFSPROC_T *p);
230 static int nfsrv_findlayout(nfsquad_t *clientidp, fhandle_t *fhp, int laytype,
231     NFSPROC_T *, struct nfslayout **lypp);
232 static int nfsrv_fndclid(nfsquad_t *clidvec, nfsquad_t clid, int clidcnt);
233 static struct nfslayout *nfsrv_filelayout(struct nfsrv_descript *nd, int iomode,
234     fhandle_t *fhp, fhandle_t *dsfhp, char *devid, fsid_t fs);
235 static struct nfslayout *nfsrv_flexlayout(struct nfsrv_descript *nd, int iomode,
236     int mirrorcnt, fhandle_t *fhp, fhandle_t *dsfhp, char *devid, fsid_t fs);
237 static int nfsrv_dontlayout(fhandle_t *fhp);
238 static int nfsrv_createdsfile(vnode_t vp, fhandle_t *fhp, struct pnfsdsfile *pf,
239     vnode_t dvp, struct nfsdevice *ds, struct ucred *cred, NFSPROC_T *p,
240     vnode_t *tvpp);
241 static struct nfsdevice *nfsrv_findmirroredds(struct nfsmount *nmp);
242 
243 /*
244  * Scan the client list for a match and either return the current one,
245  * create a new entry or return an error.
246  * If returning a non-error, the clp structure must either be linked into
247  * the client list or free'd.
248  */
249 int
250 nfsrv_setclient(struct nfsrv_descript *nd, struct nfsclient **new_clpp,
251     nfsquad_t *clientidp, nfsquad_t *confirmp, NFSPROC_T *p)
252 {
253 	struct nfsclient *clp = NULL, *new_clp = *new_clpp;
254 	int i, error = 0, ret;
255 	struct nfsstate *stp, *tstp;
256 #ifdef INET
257 	struct sockaddr_in *sin, *rin;
258 #endif
259 #ifdef INET6
260 	struct sockaddr_in6 *sin6, *rin6;
261 #endif
262 	struct nfsdsession *sep, *nsep;
263 	int zapit = 0, gotit, hasstate = 0, igotlock;
264 	static u_int64_t confirm_index = 0;
265 
266 	/*
267 	 * Check for state resource limit exceeded.
268 	 */
269 	if (nfsrv_openpluslock > nfsrv_v4statelimit) {
270 		error = NFSERR_RESOURCE;
271 		goto out;
272 	}
273 
274 	if (nfsrv_issuedelegs == 0 ||
275 	    ((nd->nd_flag & ND_GSS) != 0 && nfsrv_nogsscallback != 0))
276 		/*
277 		 * Don't do callbacks when delegations are disabled or
278 		 * for AUTH_GSS unless enabled via nfsrv_nogsscallback.
279 		 * If establishing a callback connection is attempted
280 		 * when a firewall is blocking the callback path, the
281 		 * server may wait too long for the connect attempt to
282 		 * succeed during the Open. Some clients, such as Linux,
283 		 * may timeout and give up on the Open before the server
284 		 * replies. Also, since AUTH_GSS callbacks are not
285 		 * yet interoperability tested, they might cause the
286 		 * server to crap out, if they get past the Init call to
287 		 * the client.
288 		 */
289 		new_clp->lc_program = 0;
290 
291 	/* Lock out other nfsd threads */
292 	NFSLOCKV4ROOTMUTEX();
293 	nfsv4_relref(&nfsv4rootfs_lock);
294 	do {
295 		igotlock = nfsv4_lock(&nfsv4rootfs_lock, 1, NULL,
296 		    NFSV4ROOTLOCKMUTEXPTR, NULL);
297 	} while (!igotlock);
298 	NFSUNLOCKV4ROOTMUTEX();
299 
300 	/*
301 	 * Search for a match in the client list.
302 	 */
303 	gotit = i = 0;
304 	while (i < nfsrv_clienthashsize && !gotit) {
305 	    LIST_FOREACH(clp, &NFSD_VNET(nfsclienthash)[i], lc_hash) {
306 		if (new_clp->lc_idlen == clp->lc_idlen &&
307 		    !NFSBCMP(new_clp->lc_id, clp->lc_id, clp->lc_idlen)) {
308 			gotit = 1;
309 			break;
310 		}
311 	    }
312 	    if (gotit == 0)
313 		i++;
314 	}
315 	if (!gotit ||
316 	    (clp->lc_flags & (LCL_NEEDSCONFIRM | LCL_ADMINREVOKED))) {
317 		if ((nd->nd_flag & ND_NFSV41) != 0 && confirmp->lval[1] != 0) {
318 			/*
319 			 * For NFSv4.1, if confirmp->lval[1] is non-zero, the
320 			 * client is trying to update a confirmed clientid.
321 			 */
322 			NFSLOCKV4ROOTMUTEX();
323 			nfsv4_unlock(&nfsv4rootfs_lock, 1);
324 			NFSUNLOCKV4ROOTMUTEX();
325 			confirmp->lval[1] = 0;
326 			error = NFSERR_NOENT;
327 			goto out;
328 		}
329 		/*
330 		 * Get rid of the old one.
331 		 */
332 		if (i != nfsrv_clienthashsize) {
333 			LIST_REMOVE(clp, lc_hash);
334 			nfsrv_cleanclient(clp, p);
335 			nfsrv_freedeleglist(&clp->lc_deleg);
336 			nfsrv_freedeleglist(&clp->lc_olddeleg);
337 			zapit = 1;
338 		}
339 		/*
340 		 * Add it after assigning a client id to it.
341 		 */
342 		new_clp->lc_flags |= LCL_NEEDSCONFIRM;
343 		if ((nd->nd_flag & ND_NFSV41) != 0) {
344 			confirmp->lval[0] = ++confirm_index;
345 			new_clp->lc_confirm.lval[0] = confirmp->lval[0] - 1;
346 		} else
347 			confirmp->qval = new_clp->lc_confirm.qval =
348 			    ++confirm_index;
349 		clientidp->lval[0] = new_clp->lc_clientid.lval[0] =
350 		    NFSD_VNET(nfsrvboottime);
351 		clientidp->lval[1] = new_clp->lc_clientid.lval[1] =
352 		    nfsrv_nextclientindex();
353 		new_clp->lc_stateindex = 0;
354 		new_clp->lc_statemaxindex = 0;
355 		new_clp->lc_prevsess = 0;
356 		new_clp->lc_cbref = 0;
357 		new_clp->lc_expiry = nfsrv_leaseexpiry();
358 		LIST_INIT(&new_clp->lc_open);
359 		LIST_INIT(&new_clp->lc_deleg);
360 		LIST_INIT(&new_clp->lc_olddeleg);
361 		LIST_INIT(&new_clp->lc_session);
362 		for (i = 0; i < nfsrv_statehashsize; i++)
363 			LIST_INIT(&new_clp->lc_stateid[i]);
364 		LIST_INSERT_HEAD(NFSCLIENTHASH(new_clp->lc_clientid), new_clp,
365 		    lc_hash);
366 		NFSD_VNET(nfsstatsv1_p)->srvclients++;
367 		nfsrv_openpluslock++;
368 		nfsrv_clients++;
369 		NFSLOCKV4ROOTMUTEX();
370 		nfsv4_unlock(&nfsv4rootfs_lock, 1);
371 		NFSUNLOCKV4ROOTMUTEX();
372 		if (zapit)
373 			nfsrv_zapclient(clp, p);
374 		*new_clpp = NULL;
375 		goto out;
376 	}
377 
378 	/*
379 	 * Now, handle the cases where the id is already issued.
380 	 */
381 	if (nfsrv_notsamecredname(nd, clp)) {
382 	    /*
383 	     * Check to see if there is expired state that should go away.
384 	     */
385 	    if (clp->lc_expiry < NFSD_MONOSEC &&
386 	        (!LIST_EMPTY(&clp->lc_open) || !LIST_EMPTY(&clp->lc_deleg))) {
387 		nfsrv_cleanclient(clp, p);
388 		nfsrv_freedeleglist(&clp->lc_deleg);
389 	    }
390 
391 	    /*
392 	     * If there is outstanding state, then reply NFSERR_CLIDINUSE per
393 	     * RFC3530 Sec. 8.1.2 last para.
394 	     */
395 	    if (!LIST_EMPTY(&clp->lc_deleg)) {
396 		hasstate = 1;
397 	    } else if (LIST_EMPTY(&clp->lc_open)) {
398 		hasstate = 0;
399 	    } else {
400 		hasstate = 0;
401 		/* Look for an Open on the OpenOwner */
402 		LIST_FOREACH(stp, &clp->lc_open, ls_list) {
403 		    if (!LIST_EMPTY(&stp->ls_open)) {
404 			hasstate = 1;
405 			break;
406 		    }
407 		}
408 	    }
409 	    if (hasstate) {
410 		/*
411 		 * If the uid doesn't match, return NFSERR_CLIDINUSE after
412 		 * filling out the correct ipaddr and portnum.
413 		 */
414 		switch (clp->lc_req.nr_nam->sa_family) {
415 #ifdef INET
416 		case AF_INET:
417 			sin = (struct sockaddr_in *)new_clp->lc_req.nr_nam;
418 			rin = (struct sockaddr_in *)clp->lc_req.nr_nam;
419 			sin->sin_addr.s_addr = rin->sin_addr.s_addr;
420 			sin->sin_port = rin->sin_port;
421 			break;
422 #endif
423 #ifdef INET6
424 		case AF_INET6:
425 			sin6 = (struct sockaddr_in6 *)new_clp->lc_req.nr_nam;
426 			rin6 = (struct sockaddr_in6 *)clp->lc_req.nr_nam;
427 			sin6->sin6_addr = rin6->sin6_addr;
428 			sin6->sin6_port = rin6->sin6_port;
429 			break;
430 #endif
431 		}
432 		NFSLOCKV4ROOTMUTEX();
433 		nfsv4_unlock(&nfsv4rootfs_lock, 1);
434 		NFSUNLOCKV4ROOTMUTEX();
435 		error = NFSERR_CLIDINUSE;
436 		goto out;
437 	    }
438 	}
439 
440 	if (NFSBCMP(new_clp->lc_verf, clp->lc_verf, NFSX_VERF)) {
441 		/*
442 		 * If the verifier has changed, the client has rebooted
443 		 * and a new client id is issued. The old state info
444 		 * can be thrown away once the SETCLIENTID_CONFIRM occurs.
445 		 */
446 		LIST_REMOVE(clp, lc_hash);
447 
448 		/* Get rid of all sessions on this clientid. */
449 		LIST_FOREACH_SAFE(sep, &clp->lc_session, sess_list, nsep) {
450 			ret = nfsrv_freesession(sep, NULL);
451 			if (ret != 0)
452 				printf("nfsrv_setclient: verifier changed free"
453 				    " session failed=%d\n", ret);
454 		}
455 
456 		new_clp->lc_flags |= LCL_NEEDSCONFIRM;
457 		if ((nd->nd_flag & ND_NFSV41) != 0) {
458 			confirmp->lval[0] = ++confirm_index;
459 			new_clp->lc_confirm.lval[0] = confirmp->lval[0] - 1;
460 		} else
461 			confirmp->qval = new_clp->lc_confirm.qval =
462 			    ++confirm_index;
463 		clientidp->lval[0] = new_clp->lc_clientid.lval[0] =
464 		    NFSD_VNET(nfsrvboottime);
465 		clientidp->lval[1] = new_clp->lc_clientid.lval[1] =
466 		    nfsrv_nextclientindex();
467 		new_clp->lc_stateindex = 0;
468 		new_clp->lc_statemaxindex = 0;
469 		new_clp->lc_prevsess = 0;
470 		new_clp->lc_cbref = 0;
471 		new_clp->lc_expiry = nfsrv_leaseexpiry();
472 
473 		/*
474 		 * Save the state until confirmed.
475 		 */
476 		LIST_NEWHEAD(&new_clp->lc_open, &clp->lc_open, ls_list);
477 		LIST_FOREACH(tstp, &new_clp->lc_open, ls_list)
478 			tstp->ls_clp = new_clp;
479 		LIST_NEWHEAD(&new_clp->lc_deleg, &clp->lc_deleg, ls_list);
480 		LIST_FOREACH(tstp, &new_clp->lc_deleg, ls_list)
481 			tstp->ls_clp = new_clp;
482 		LIST_NEWHEAD(&new_clp->lc_olddeleg, &clp->lc_olddeleg,
483 		    ls_list);
484 		LIST_FOREACH(tstp, &new_clp->lc_olddeleg, ls_list)
485 			tstp->ls_clp = new_clp;
486 		for (i = 0; i < nfsrv_statehashsize; i++) {
487 			LIST_NEWHEAD(&new_clp->lc_stateid[i],
488 			    &clp->lc_stateid[i], ls_hash);
489 			LIST_FOREACH(tstp, &new_clp->lc_stateid[i], ls_hash)
490 				tstp->ls_clp = new_clp;
491 		}
492 		LIST_INIT(&new_clp->lc_session);
493 		LIST_INSERT_HEAD(NFSCLIENTHASH(new_clp->lc_clientid), new_clp,
494 		    lc_hash);
495 		NFSD_VNET(nfsstatsv1_p)->srvclients++;
496 		nfsrv_openpluslock++;
497 		nfsrv_clients++;
498 		NFSLOCKV4ROOTMUTEX();
499 		nfsv4_unlock(&nfsv4rootfs_lock, 1);
500 		NFSUNLOCKV4ROOTMUTEX();
501 
502 		/*
503 		 * Must wait until any outstanding callback on the old clp
504 		 * completes.
505 		 */
506 		NFSLOCKSTATE();
507 		while (clp->lc_cbref) {
508 			clp->lc_flags |= LCL_WAKEUPWANTED;
509 			(void)mtx_sleep(clp, NFSSTATEMUTEXPTR, PZERO - 1,
510 			    "nfsd clp", 10 * hz);
511 		}
512 		NFSUNLOCKSTATE();
513 		nfsrv_zapclient(clp, p);
514 		*new_clpp = NULL;
515 		goto out;
516 	}
517 
518 	/* For NFSv4.1, mark that we found a confirmed clientid. */
519 	if ((nd->nd_flag & ND_NFSV41) != 0) {
520 		clientidp->lval[0] = clp->lc_clientid.lval[0];
521 		clientidp->lval[1] = clp->lc_clientid.lval[1];
522 		confirmp->lval[0] = 0;	/* Ignored by client */
523 		confirmp->lval[1] = 1;
524 	} else {
525 		/*
526 		 * id and verifier match, so update the net address info
527 		 * and get rid of any existing callback authentication
528 		 * handle, so a new one will be acquired.
529 		 */
530 		LIST_REMOVE(clp, lc_hash);
531 		new_clp->lc_flags |= (LCL_NEEDSCONFIRM | LCL_DONTCLEAN);
532 		new_clp->lc_expiry = nfsrv_leaseexpiry();
533 		confirmp->qval = new_clp->lc_confirm.qval = ++confirm_index;
534 		clientidp->lval[0] = new_clp->lc_clientid.lval[0] =
535 		    clp->lc_clientid.lval[0];
536 		clientidp->lval[1] = new_clp->lc_clientid.lval[1] =
537 		    clp->lc_clientid.lval[1];
538 		new_clp->lc_delegtime = clp->lc_delegtime;
539 		new_clp->lc_stateindex = clp->lc_stateindex;
540 		new_clp->lc_statemaxindex = clp->lc_statemaxindex;
541 		new_clp->lc_cbref = 0;
542 		LIST_NEWHEAD(&new_clp->lc_open, &clp->lc_open, ls_list);
543 		LIST_FOREACH(tstp, &new_clp->lc_open, ls_list)
544 			tstp->ls_clp = new_clp;
545 		LIST_NEWHEAD(&new_clp->lc_deleg, &clp->lc_deleg, ls_list);
546 		LIST_FOREACH(tstp, &new_clp->lc_deleg, ls_list)
547 			tstp->ls_clp = new_clp;
548 		LIST_NEWHEAD(&new_clp->lc_olddeleg, &clp->lc_olddeleg, ls_list);
549 		LIST_FOREACH(tstp, &new_clp->lc_olddeleg, ls_list)
550 			tstp->ls_clp = new_clp;
551 		for (i = 0; i < nfsrv_statehashsize; i++) {
552 			LIST_NEWHEAD(&new_clp->lc_stateid[i],
553 			    &clp->lc_stateid[i], ls_hash);
554 			LIST_FOREACH(tstp, &new_clp->lc_stateid[i], ls_hash)
555 				tstp->ls_clp = new_clp;
556 		}
557 		LIST_INIT(&new_clp->lc_session);
558 		LIST_INSERT_HEAD(NFSCLIENTHASH(new_clp->lc_clientid), new_clp,
559 		    lc_hash);
560 		NFSD_VNET(nfsstatsv1_p)->srvclients++;
561 		nfsrv_openpluslock++;
562 		nfsrv_clients++;
563 	}
564 	NFSLOCKV4ROOTMUTEX();
565 	nfsv4_unlock(&nfsv4rootfs_lock, 1);
566 	NFSUNLOCKV4ROOTMUTEX();
567 
568 	if ((nd->nd_flag & ND_NFSV41) == 0) {
569 		/*
570 		 * Must wait until any outstanding callback on the old clp
571 		 * completes.
572 		 */
573 		NFSLOCKSTATE();
574 		while (clp->lc_cbref) {
575 			clp->lc_flags |= LCL_WAKEUPWANTED;
576 			(void)mtx_sleep(clp, NFSSTATEMUTEXPTR, PZERO - 1,
577 			    "nfsdclp", 10 * hz);
578 		}
579 		NFSUNLOCKSTATE();
580 		nfsrv_zapclient(clp, p);
581 		*new_clpp = NULL;
582 	}
583 
584 out:
585 	NFSEXITCODE2(error, nd);
586 	return (error);
587 }
588 
589 /*
590  * Check to see if the client id exists and optionally confirm it.
591  */
592 int
593 nfsrv_getclient(nfsquad_t clientid, int opflags, struct nfsclient **clpp,
594     struct nfsdsession *nsep, nfsquad_t confirm, uint32_t cbprogram,
595     struct nfsrv_descript *nd, NFSPROC_T *p)
596 {
597 	struct nfsclient *clp;
598 	struct nfsstate *stp;
599 	int i;
600 	struct nfsclienthashhead *hp;
601 	int error = 0, igotlock, doneok;
602 	struct nfssessionhash *shp;
603 	struct nfsdsession *sep;
604 	uint64_t sessid[2];
605 	bool sess_replay;
606 	static uint64_t next_sess = 0;
607 
608 	if (clpp)
609 		*clpp = NULL;
610 	if ((nd == NULL || (nd->nd_flag & ND_NFSV41) == 0 ||
611 	    opflags != CLOPS_RENEW) && NFSD_VNET(nfsrvboottime) !=
612 	    clientid.lval[0]) {
613 		error = NFSERR_STALECLIENTID;
614 		goto out;
615 	}
616 
617 	/*
618 	 * If called with opflags == CLOPS_RENEW, the State Lock is
619 	 * already held. Otherwise, we need to get either that or,
620 	 * for the case of Confirm, lock out the nfsd threads.
621 	 */
622 	if (opflags & CLOPS_CONFIRM) {
623 		NFSLOCKV4ROOTMUTEX();
624 		nfsv4_relref(&nfsv4rootfs_lock);
625 		do {
626 			igotlock = nfsv4_lock(&nfsv4rootfs_lock, 1, NULL,
627 			    NFSV4ROOTLOCKMUTEXPTR, NULL);
628 		} while (!igotlock);
629 		/*
630 		 * Create a new sessionid here, since we need to do it where
631 		 * there is a mutex held to serialize update of next_sess.
632 		 */
633 		if ((nd->nd_flag & ND_NFSV41) != 0) {
634 			sessid[0] = ++next_sess;
635 			sessid[1] = clientid.qval;
636 		}
637 		NFSUNLOCKV4ROOTMUTEX();
638 	} else if (opflags != CLOPS_RENEW) {
639 		NFSLOCKSTATE();
640 	}
641 
642 	/* For NFSv4.1, the clp is acquired from the associated session. */
643 	if (nd != NULL && (nd->nd_flag & ND_NFSV41) != 0 &&
644 	    opflags == CLOPS_RENEW) {
645 		clp = NULL;
646 		if ((nd->nd_flag & ND_HASSEQUENCE) != 0) {
647 			shp = NFSSESSIONHASH(nd->nd_sessionid);
648 			NFSLOCKSESSION(shp);
649 			sep = nfsrv_findsession(nd->nd_sessionid);
650 			if (sep != NULL)
651 				clp = sep->sess_clp;
652 			NFSUNLOCKSESSION(shp);
653 		}
654 	} else {
655 		hp = NFSCLIENTHASH(clientid);
656 		LIST_FOREACH(clp, hp, lc_hash) {
657 			if (clp->lc_clientid.lval[1] == clientid.lval[1])
658 				break;
659 		}
660 	}
661 	if (clp == NULL) {
662 		if (opflags & CLOPS_CONFIRM)
663 			error = NFSERR_STALECLIENTID;
664 		else
665 			error = NFSERR_EXPIRED;
666 	} else if (clp->lc_flags & LCL_ADMINREVOKED) {
667 		/*
668 		 * If marked admin revoked, just return the error.
669 		 */
670 		error = NFSERR_ADMINREVOKED;
671 	}
672 	if (error) {
673 		if (opflags & CLOPS_CONFIRM) {
674 			NFSLOCKV4ROOTMUTEX();
675 			nfsv4_unlock(&nfsv4rootfs_lock, 1);
676 			NFSUNLOCKV4ROOTMUTEX();
677 		} else if (opflags != CLOPS_RENEW) {
678 			NFSUNLOCKSTATE();
679 		}
680 		goto out;
681 	}
682 
683 	/*
684 	 * Perform any operations specified by the opflags.
685 	 */
686 	if (opflags & CLOPS_CONFIRM) {
687 		sess_replay = false;
688 		if ((nd->nd_flag & ND_NFSV41) != 0) {
689 		    /*
690 		     * For the case where lc_confirm.lval[0] == confirm.lval[0],
691 		     * use the new session, but with the previous sessionid.
692 		     * This is not exactly what the RFC describes, but should
693 		     * result in the same reply as the previous CreateSession.
694 		     */
695 		    if (clp->lc_confirm.lval[0] + 1 == confirm.lval[0]) {
696 			clp->lc_confirm.lval[0] = confirm.lval[0];
697 			clp->lc_prevsess = sessid[0];
698 		    } else if (clp->lc_confirm.lval[0] == confirm.lval[0]) {
699 			if (clp->lc_prevsess == 0)
700 			    error = NFSERR_SEQMISORDERED;
701 			else
702 			    sessid[0] = clp->lc_prevsess;
703 			sess_replay = true;
704 		    } else
705 			error = NFSERR_SEQMISORDERED;
706 		} else if ((nd->nd_flag & ND_NFSV41) == 0 &&
707 		     clp->lc_confirm.qval != confirm.qval)
708 			error = NFSERR_STALECLIENTID;
709 		if (error == 0 && nfsrv_notsamecredname(nd, clp))
710 			error = NFSERR_CLIDINUSE;
711 
712 		if (!error) {
713 		    if ((clp->lc_flags & (LCL_NEEDSCONFIRM | LCL_DONTCLEAN)) ==
714 			LCL_NEEDSCONFIRM) {
715 			/*
716 			 * Hang onto the delegations (as old delegations)
717 			 * for an Open with CLAIM_DELEGATE_PREV unless in
718 			 * grace, but get rid of the rest of the state.
719 			 */
720 			nfsrv_cleanclient(clp, p);
721 			nfsrv_freedeleglist(&clp->lc_olddeleg);
722 			if (nfsrv_checkgrace(nd, clp, 0)) {
723 			    /* In grace, so just delete delegations */
724 			    nfsrv_freedeleglist(&clp->lc_deleg);
725 			} else {
726 			    LIST_FOREACH(stp, &clp->lc_deleg, ls_list)
727 				stp->ls_flags |= NFSLCK_OLDDELEG;
728 			    clp->lc_delegtime = NFSD_MONOSEC +
729 				nfsrv_lease + NFSRV_LEASEDELTA;
730 			    LIST_NEWHEAD(&clp->lc_olddeleg, &clp->lc_deleg,
731 				ls_list);
732 			}
733 			if ((nd->nd_flag & ND_NFSV41) != 0)
734 			    clp->lc_program = cbprogram;
735 		    }
736 		    clp->lc_flags &= ~(LCL_NEEDSCONFIRM | LCL_DONTCLEAN);
737 		    if (clp->lc_program)
738 			clp->lc_flags |= LCL_NEEDSCBNULL;
739 		    /* For NFSv4.1, link the session onto the client. */
740 		    if (nsep != NULL) {
741 			/* Hold a reference on the xprt for a backchannel. */
742 			if ((nsep->sess_crflags & NFSV4CRSESS_CONNBACKCHAN)
743 			    != 0 && !sess_replay) {
744 			    if (clp->lc_req.nr_client == NULL)
745 				clp->lc_req.nr_client = (struct __rpc_client *)
746 				    clnt_bck_create(nd->nd_xprt->xp_socket,
747 				    cbprogram, NFSV4_CBVERS);
748 			    if (clp->lc_req.nr_client != NULL) {
749 				SVC_ACQUIRE(nd->nd_xprt);
750 				CLNT_ACQUIRE(clp->lc_req.nr_client);
751 				nd->nd_xprt->xp_p2 = clp->lc_req.nr_client;
752 				/* Disable idle timeout. */
753 				nd->nd_xprt->xp_idletimeout = 0;
754 				nsep->sess_cbsess.nfsess_xprt = nd->nd_xprt;
755 			    } else
756 				nsep->sess_crflags &= ~NFSV4CRSESS_CONNBACKCHAN;
757 			}
758 			NFSBCOPY(sessid, nsep->sess_sessionid,
759 			    NFSX_V4SESSIONID);
760 			NFSBCOPY(sessid, nsep->sess_cbsess.nfsess_sessionid,
761 			    NFSX_V4SESSIONID);
762 			if (!sess_replay) {
763 			    shp = NFSSESSIONHASH(nsep->sess_sessionid);
764 			    NFSLOCKSTATE();
765 			    NFSLOCKSESSION(shp);
766 			    LIST_INSERT_HEAD(&shp->list, nsep, sess_hash);
767 			    LIST_INSERT_HEAD(&clp->lc_session, nsep, sess_list);
768 			    nsep->sess_clp = clp;
769 			    NFSUNLOCKSESSION(shp);
770 			    NFSUNLOCKSTATE();
771 			}
772 		    }
773 		}
774 	} else if (clp->lc_flags & LCL_NEEDSCONFIRM) {
775 		error = NFSERR_EXPIRED;
776 	}
777 
778 	/*
779 	 * If called by the Renew Op, we must check the principal.
780 	 */
781 	if (!error && (opflags & CLOPS_RENEWOP)) {
782 	    if (nfsrv_notsamecredname(nd, clp)) {
783 		doneok = 0;
784 		for (i = 0; i < nfsrv_statehashsize && doneok == 0; i++) {
785 		    LIST_FOREACH(stp, &clp->lc_stateid[i], ls_hash) {
786 			if ((stp->ls_flags & NFSLCK_OPEN) &&
787 			    stp->ls_uid == nd->nd_cred->cr_uid) {
788 				doneok = 1;
789 				break;
790 			}
791 		    }
792 		}
793 		if (!doneok)
794 			error = NFSERR_ACCES;
795 	    }
796 	    if (!error && (clp->lc_flags & LCL_CBDOWN))
797 		error = NFSERR_CBPATHDOWN;
798 	}
799 	if ((!error || error == NFSERR_CBPATHDOWN) &&
800 	     (opflags & CLOPS_RENEW)) {
801 		clp->lc_expiry = nfsrv_leaseexpiry();
802 	}
803 	if (opflags & CLOPS_CONFIRM) {
804 		NFSLOCKV4ROOTMUTEX();
805 		nfsv4_unlock(&nfsv4rootfs_lock, 1);
806 		NFSUNLOCKV4ROOTMUTEX();
807 	} else if (opflags != CLOPS_RENEW) {
808 		NFSUNLOCKSTATE();
809 	}
810 	if (clpp)
811 		*clpp = clp;
812 
813 out:
814 	NFSEXITCODE2(error, nd);
815 	return (error);
816 }
817 
818 /*
819  * Perform the NFSv4.1 destroy clientid.
820  */
821 int
822 nfsrv_destroyclient(nfsquad_t clientid, NFSPROC_T *p)
823 {
824 	struct nfsclient *clp;
825 	struct nfsclienthashhead *hp;
826 	int error = 0, i, igotlock;
827 
828 	if (NFSD_VNET(nfsrvboottime) != clientid.lval[0]) {
829 		error = NFSERR_STALECLIENTID;
830 		goto out;
831 	}
832 
833 	/* Lock out other nfsd threads */
834 	NFSLOCKV4ROOTMUTEX();
835 	nfsv4_relref(&nfsv4rootfs_lock);
836 	do {
837 		igotlock = nfsv4_lock(&nfsv4rootfs_lock, 1, NULL,
838 		    NFSV4ROOTLOCKMUTEXPTR, NULL);
839 	} while (igotlock == 0);
840 	NFSUNLOCKV4ROOTMUTEX();
841 
842 	hp = NFSCLIENTHASH(clientid);
843 	LIST_FOREACH(clp, hp, lc_hash) {
844 		if (clp->lc_clientid.lval[1] == clientid.lval[1])
845 			break;
846 	}
847 	if (clp == NULL) {
848 		NFSLOCKV4ROOTMUTEX();
849 		nfsv4_unlock(&nfsv4rootfs_lock, 1);
850 		NFSUNLOCKV4ROOTMUTEX();
851 		/* Just return ok, since it is gone. */
852 		goto out;
853 	}
854 
855 	/*
856 	 * Free up all layouts on the clientid.  Should the client return the
857 	 * layouts?
858 	 */
859 	nfsrv_freelayoutlist(clientid);
860 
861 	/* Scan for state on the clientid. */
862 	for (i = 0; i < nfsrv_statehashsize; i++)
863 		if (!LIST_EMPTY(&clp->lc_stateid[i])) {
864 			NFSLOCKV4ROOTMUTEX();
865 			nfsv4_unlock(&nfsv4rootfs_lock, 1);
866 			NFSUNLOCKV4ROOTMUTEX();
867 			error = NFSERR_CLIENTIDBUSY;
868 			goto out;
869 		}
870 	if (!LIST_EMPTY(&clp->lc_session) || !LIST_EMPTY(&clp->lc_deleg)) {
871 		NFSLOCKV4ROOTMUTEX();
872 		nfsv4_unlock(&nfsv4rootfs_lock, 1);
873 		NFSUNLOCKV4ROOTMUTEX();
874 		error = NFSERR_CLIENTIDBUSY;
875 		goto out;
876 	}
877 
878 	/* Destroy the clientid and return ok. */
879 	nfsrv_cleanclient(clp, p);
880 	nfsrv_freedeleglist(&clp->lc_deleg);
881 	nfsrv_freedeleglist(&clp->lc_olddeleg);
882 	LIST_REMOVE(clp, lc_hash);
883 	NFSLOCKV4ROOTMUTEX();
884 	nfsv4_unlock(&nfsv4rootfs_lock, 1);
885 	NFSUNLOCKV4ROOTMUTEX();
886 	nfsrv_zapclient(clp, p);
887 out:
888 	NFSEXITCODE2(error, nd);
889 	return (error);
890 }
891 
892 /*
893  * Called from the new nfssvc syscall to admin revoke a clientid.
894  * Returns 0 for success, error otherwise.
895  */
896 int
897 nfsrv_adminrevoke(struct nfsd_clid *revokep, NFSPROC_T *p)
898 {
899 	struct nfsclient *clp = NULL;
900 	int i, error = 0;
901 	int gotit, igotlock;
902 
903 	/*
904 	 * First, lock out the nfsd so that state won't change while the
905 	 * revocation record is being written to the stable storage restart
906 	 * file.
907 	 */
908 	NFSLOCKV4ROOTMUTEX();
909 	do {
910 		igotlock = nfsv4_lock(&nfsv4rootfs_lock, 1, NULL,
911 		    NFSV4ROOTLOCKMUTEXPTR, NULL);
912 	} while (!igotlock);
913 	NFSUNLOCKV4ROOTMUTEX();
914 
915 	/*
916 	 * Search for a match in the client list.
917 	 */
918 	gotit = i = 0;
919 	while (i < nfsrv_clienthashsize && !gotit) {
920 	    LIST_FOREACH(clp, &NFSD_VNET(nfsclienthash)[i], lc_hash) {
921 		if (revokep->nclid_idlen == clp->lc_idlen &&
922 		    !NFSBCMP(revokep->nclid_id, clp->lc_id, clp->lc_idlen)) {
923 			gotit = 1;
924 			break;
925 		}
926 	    }
927 	    i++;
928 	}
929 	if (!gotit) {
930 		NFSLOCKV4ROOTMUTEX();
931 		nfsv4_unlock(&nfsv4rootfs_lock, 0);
932 		NFSUNLOCKV4ROOTMUTEX();
933 		error = EPERM;
934 		goto out;
935 	}
936 
937 	/*
938 	 * Now, write out the revocation record
939 	 */
940 	nfsrv_writestable(clp->lc_id, clp->lc_idlen, NFSNST_REVOKE, p);
941 	nfsrv_backupstable();
942 
943 	/*
944 	 * and clear out the state, marking the clientid revoked.
945 	 */
946 	clp->lc_flags &= ~LCL_CALLBACKSON;
947 	clp->lc_flags |= LCL_ADMINREVOKED;
948 	nfsrv_cleanclient(clp, p);
949 	nfsrv_freedeleglist(&clp->lc_deleg);
950 	nfsrv_freedeleglist(&clp->lc_olddeleg);
951 	NFSLOCKV4ROOTMUTEX();
952 	nfsv4_unlock(&nfsv4rootfs_lock, 0);
953 	NFSUNLOCKV4ROOTMUTEX();
954 
955 out:
956 	NFSEXITCODE(error);
957 	return (error);
958 }
959 
960 /*
961  * Dump out stats for all clients. Called from nfssvc(2), that is used
962  * nfsstatsv1.
963  */
964 void
965 nfsrv_dumpclients(struct nfsd_dumpclients *dumpp, int maxcnt)
966 {
967 	struct nfsclient *clp;
968 	int i = 0, cnt = 0;
969 
970 	/*
971 	 * First, get a reference on the nfsv4rootfs_lock so that an
972 	 * exclusive lock cannot be acquired while dumping the clients.
973 	 */
974 	NFSLOCKV4ROOTMUTEX();
975 	nfsv4_getref(&nfsv4rootfs_lock, NULL, NFSV4ROOTLOCKMUTEXPTR, NULL);
976 	NFSUNLOCKV4ROOTMUTEX();
977 	NFSLOCKSTATE();
978 	/*
979 	 * Rattle through the client lists until done.
980 	 */
981 	while (i < nfsrv_clienthashsize && cnt < maxcnt) {
982 	    clp = LIST_FIRST(&NFSD_VNET(nfsclienthash)[i]);
983 	    while (clp != LIST_END(&NFSD_VNET(nfsclienthash)[i]) && cnt <
984 		maxcnt) {
985 		nfsrv_dumpaclient(clp, &dumpp[cnt]);
986 		cnt++;
987 		clp = LIST_NEXT(clp, lc_hash);
988 	    }
989 	    i++;
990 	}
991 	if (cnt < maxcnt)
992 	    dumpp[cnt].ndcl_clid.nclid_idlen = 0;
993 	NFSUNLOCKSTATE();
994 	NFSLOCKV4ROOTMUTEX();
995 	nfsv4_relref(&nfsv4rootfs_lock);
996 	NFSUNLOCKV4ROOTMUTEX();
997 }
998 
999 /*
1000  * Dump stats for a client. Must be called with the NFSSTATELOCK and spl'd.
1001  */
1002 static void
1003 nfsrv_dumpaclient(struct nfsclient *clp, struct nfsd_dumpclients *dumpp)
1004 {
1005 	struct nfsstate *stp, *openstp, *lckownstp;
1006 	struct nfslock *lop;
1007 	sa_family_t af;
1008 #ifdef INET
1009 	struct sockaddr_in *rin;
1010 #endif
1011 #ifdef INET6
1012 	struct sockaddr_in6 *rin6;
1013 #endif
1014 
1015 	dumpp->ndcl_nopenowners = dumpp->ndcl_nlockowners = 0;
1016 	dumpp->ndcl_nopens = dumpp->ndcl_nlocks = 0;
1017 	dumpp->ndcl_ndelegs = dumpp->ndcl_nolddelegs = 0;
1018 	dumpp->ndcl_flags = clp->lc_flags;
1019 	dumpp->ndcl_clid.nclid_idlen = clp->lc_idlen;
1020 	NFSBCOPY(clp->lc_id, dumpp->ndcl_clid.nclid_id, clp->lc_idlen);
1021 	af = clp->lc_req.nr_nam->sa_family;
1022 	dumpp->ndcl_addrfam = af;
1023 	switch (af) {
1024 #ifdef INET
1025 	case AF_INET:
1026 		rin = (struct sockaddr_in *)clp->lc_req.nr_nam;
1027 		dumpp->ndcl_cbaddr.sin_addr = rin->sin_addr;
1028 		break;
1029 #endif
1030 #ifdef INET6
1031 	case AF_INET6:
1032 		rin6 = (struct sockaddr_in6 *)clp->lc_req.nr_nam;
1033 		dumpp->ndcl_cbaddr.sin6_addr = rin6->sin6_addr;
1034 		break;
1035 #endif
1036 	}
1037 
1038 	/*
1039 	 * Now, scan the state lists and total up the opens and locks.
1040 	 */
1041 	LIST_FOREACH(stp, &clp->lc_open, ls_list) {
1042 	    dumpp->ndcl_nopenowners++;
1043 	    LIST_FOREACH(openstp, &stp->ls_open, ls_list) {
1044 		dumpp->ndcl_nopens++;
1045 		LIST_FOREACH(lckownstp, &openstp->ls_open, ls_list) {
1046 		    dumpp->ndcl_nlockowners++;
1047 		    LIST_FOREACH(lop, &lckownstp->ls_lock, lo_lckowner) {
1048 			dumpp->ndcl_nlocks++;
1049 		    }
1050 		}
1051 	    }
1052 	}
1053 
1054 	/*
1055 	 * and the delegation lists.
1056 	 */
1057 	LIST_FOREACH(stp, &clp->lc_deleg, ls_list) {
1058 	    dumpp->ndcl_ndelegs++;
1059 	}
1060 	LIST_FOREACH(stp, &clp->lc_olddeleg, ls_list) {
1061 	    dumpp->ndcl_nolddelegs++;
1062 	}
1063 }
1064 
1065 /*
1066  * Dump out lock stats for a file.
1067  */
1068 void
1069 nfsrv_dumplocks(vnode_t vp, struct nfsd_dumplocks *ldumpp, int maxcnt,
1070     NFSPROC_T *p)
1071 {
1072 	struct nfsstate *stp;
1073 	struct nfslock *lop;
1074 	int cnt = 0;
1075 	struct nfslockfile *lfp;
1076 	sa_family_t af;
1077 #ifdef INET
1078 	struct sockaddr_in *rin;
1079 #endif
1080 #ifdef INET6
1081 	struct sockaddr_in6 *rin6;
1082 #endif
1083 	int ret;
1084 	fhandle_t nfh;
1085 
1086 	ret = nfsrv_getlockfh(vp, 0, NULL, &nfh, p);
1087 	/*
1088 	 * First, get a reference on the nfsv4rootfs_lock so that an
1089 	 * exclusive lock on it cannot be acquired while dumping the locks.
1090 	 */
1091 	NFSLOCKV4ROOTMUTEX();
1092 	nfsv4_getref(&nfsv4rootfs_lock, NULL, NFSV4ROOTLOCKMUTEXPTR, NULL);
1093 	NFSUNLOCKV4ROOTMUTEX();
1094 	NFSLOCKSTATE();
1095 	if (!ret)
1096 		ret = nfsrv_getlockfile(0, NULL, &lfp, &nfh, 0);
1097 	if (ret) {
1098 		ldumpp[0].ndlck_clid.nclid_idlen = 0;
1099 		NFSUNLOCKSTATE();
1100 		NFSLOCKV4ROOTMUTEX();
1101 		nfsv4_relref(&nfsv4rootfs_lock);
1102 		NFSUNLOCKV4ROOTMUTEX();
1103 		return;
1104 	}
1105 
1106 	/*
1107 	 * For each open share on file, dump it out.
1108 	 */
1109 	stp = LIST_FIRST(&lfp->lf_open);
1110 	while (stp != LIST_END(&lfp->lf_open) && cnt < maxcnt) {
1111 		ldumpp[cnt].ndlck_flags = stp->ls_flags;
1112 		ldumpp[cnt].ndlck_stateid.seqid = stp->ls_stateid.seqid;
1113 		ldumpp[cnt].ndlck_stateid.other[0] = stp->ls_stateid.other[0];
1114 		ldumpp[cnt].ndlck_stateid.other[1] = stp->ls_stateid.other[1];
1115 		ldumpp[cnt].ndlck_stateid.other[2] = stp->ls_stateid.other[2];
1116 		ldumpp[cnt].ndlck_owner.nclid_idlen =
1117 		    stp->ls_openowner->ls_ownerlen;
1118 		NFSBCOPY(stp->ls_openowner->ls_owner,
1119 		    ldumpp[cnt].ndlck_owner.nclid_id,
1120 		    stp->ls_openowner->ls_ownerlen);
1121 		ldumpp[cnt].ndlck_clid.nclid_idlen = stp->ls_clp->lc_idlen;
1122 		NFSBCOPY(stp->ls_clp->lc_id, ldumpp[cnt].ndlck_clid.nclid_id,
1123 		    stp->ls_clp->lc_idlen);
1124 		af = stp->ls_clp->lc_req.nr_nam->sa_family;
1125 		ldumpp[cnt].ndlck_addrfam = af;
1126 		switch (af) {
1127 #ifdef INET
1128 		case AF_INET:
1129 			rin = (struct sockaddr_in *)stp->ls_clp->lc_req.nr_nam;
1130 			ldumpp[cnt].ndlck_cbaddr.sin_addr = rin->sin_addr;
1131 			break;
1132 #endif
1133 #ifdef INET6
1134 		case AF_INET6:
1135 			rin6 = (struct sockaddr_in6 *)
1136 			    stp->ls_clp->lc_req.nr_nam;
1137 			ldumpp[cnt].ndlck_cbaddr.sin6_addr = rin6->sin6_addr;
1138 			break;
1139 #endif
1140 		}
1141 		stp = LIST_NEXT(stp, ls_file);
1142 		cnt++;
1143 	}
1144 
1145 	/*
1146 	 * and all locks.
1147 	 */
1148 	lop = LIST_FIRST(&lfp->lf_lock);
1149 	while (lop != LIST_END(&lfp->lf_lock) && cnt < maxcnt) {
1150 		stp = lop->lo_stp;
1151 		ldumpp[cnt].ndlck_flags = lop->lo_flags;
1152 		ldumpp[cnt].ndlck_first = lop->lo_first;
1153 		ldumpp[cnt].ndlck_end = lop->lo_end;
1154 		ldumpp[cnt].ndlck_stateid.seqid = stp->ls_stateid.seqid;
1155 		ldumpp[cnt].ndlck_stateid.other[0] = stp->ls_stateid.other[0];
1156 		ldumpp[cnt].ndlck_stateid.other[1] = stp->ls_stateid.other[1];
1157 		ldumpp[cnt].ndlck_stateid.other[2] = stp->ls_stateid.other[2];
1158 		ldumpp[cnt].ndlck_owner.nclid_idlen = stp->ls_ownerlen;
1159 		NFSBCOPY(stp->ls_owner, ldumpp[cnt].ndlck_owner.nclid_id,
1160 		    stp->ls_ownerlen);
1161 		ldumpp[cnt].ndlck_clid.nclid_idlen = stp->ls_clp->lc_idlen;
1162 		NFSBCOPY(stp->ls_clp->lc_id, ldumpp[cnt].ndlck_clid.nclid_id,
1163 		    stp->ls_clp->lc_idlen);
1164 		af = stp->ls_clp->lc_req.nr_nam->sa_family;
1165 		ldumpp[cnt].ndlck_addrfam = af;
1166 		switch (af) {
1167 #ifdef INET
1168 		case AF_INET:
1169 			rin = (struct sockaddr_in *)stp->ls_clp->lc_req.nr_nam;
1170 			ldumpp[cnt].ndlck_cbaddr.sin_addr = rin->sin_addr;
1171 			break;
1172 #endif
1173 #ifdef INET6
1174 		case AF_INET6:
1175 			rin6 = (struct sockaddr_in6 *)
1176 			    stp->ls_clp->lc_req.nr_nam;
1177 			ldumpp[cnt].ndlck_cbaddr.sin6_addr = rin6->sin6_addr;
1178 			break;
1179 #endif
1180 		}
1181 		lop = LIST_NEXT(lop, lo_lckfile);
1182 		cnt++;
1183 	}
1184 
1185 	/*
1186 	 * and the delegations.
1187 	 */
1188 	stp = LIST_FIRST(&lfp->lf_deleg);
1189 	while (stp != LIST_END(&lfp->lf_deleg) && cnt < maxcnt) {
1190 		ldumpp[cnt].ndlck_flags = stp->ls_flags;
1191 		ldumpp[cnt].ndlck_stateid.seqid = stp->ls_stateid.seqid;
1192 		ldumpp[cnt].ndlck_stateid.other[0] = stp->ls_stateid.other[0];
1193 		ldumpp[cnt].ndlck_stateid.other[1] = stp->ls_stateid.other[1];
1194 		ldumpp[cnt].ndlck_stateid.other[2] = stp->ls_stateid.other[2];
1195 		ldumpp[cnt].ndlck_owner.nclid_idlen = 0;
1196 		ldumpp[cnt].ndlck_clid.nclid_idlen = stp->ls_clp->lc_idlen;
1197 		NFSBCOPY(stp->ls_clp->lc_id, ldumpp[cnt].ndlck_clid.nclid_id,
1198 		    stp->ls_clp->lc_idlen);
1199 		af = stp->ls_clp->lc_req.nr_nam->sa_family;
1200 		ldumpp[cnt].ndlck_addrfam = af;
1201 		switch (af) {
1202 #ifdef INET
1203 		case AF_INET:
1204 			rin = (struct sockaddr_in *)stp->ls_clp->lc_req.nr_nam;
1205 			ldumpp[cnt].ndlck_cbaddr.sin_addr = rin->sin_addr;
1206 			break;
1207 #endif
1208 #ifdef INET6
1209 		case AF_INET6:
1210 			rin6 = (struct sockaddr_in6 *)
1211 			    stp->ls_clp->lc_req.nr_nam;
1212 			ldumpp[cnt].ndlck_cbaddr.sin6_addr = rin6->sin6_addr;
1213 			break;
1214 #endif
1215 		}
1216 		stp = LIST_NEXT(stp, ls_file);
1217 		cnt++;
1218 	}
1219 
1220 	/*
1221 	 * If list isn't full, mark end of list by setting the client name
1222 	 * to zero length.
1223 	 */
1224 	if (cnt < maxcnt)
1225 		ldumpp[cnt].ndlck_clid.nclid_idlen = 0;
1226 	NFSUNLOCKSTATE();
1227 	NFSLOCKV4ROOTMUTEX();
1228 	nfsv4_relref(&nfsv4rootfs_lock);
1229 	NFSUNLOCKV4ROOTMUTEX();
1230 }
1231 
1232 /*
1233  * Server timer routine. It can scan any linked list, so long
1234  * as it holds the spin/mutex lock and there is no exclusive lock on
1235  * nfsv4rootfs_lock.
1236  * (For OpenBSD, a kthread is ok. For FreeBSD, I think it is ok
1237  *  to do this from a callout, since the spin locks work. For
1238  *  Darwin, I'm not sure what will work correctly yet.)
1239  * Should be called once per second.
1240  */
1241 void
1242 nfsrv_servertimer(void *arg __unused)
1243 {
1244 	struct nfsclient *clp, *nclp;
1245 	struct nfsstate *stp, *nstp;
1246 	int got_ref, i;
1247 
1248 	/*
1249 	 * Make sure nfsboottime is set. This is used by V3 as well
1250 	 * as V4. Note that nfsboottime is not nfsrvboottime, which is
1251 	 * only used by the V4 server for leases.
1252 	 */
1253 	if (nfsboottime.tv_sec == 0)
1254 		NFSSETBOOTTIME(nfsboottime);
1255 
1256 	/*
1257 	 * If server hasn't started yet, just return.
1258 	 */
1259 	NFSLOCKSTATE();
1260 	if (NFSD_VNET(nfsrv_stablefirst).nsf_eograce == 0) {
1261 		NFSUNLOCKSTATE();
1262 		return;
1263 	}
1264 	if (!(NFSD_VNET(nfsrv_stablefirst).nsf_flags & NFSNSF_UPDATEDONE)) {
1265 		if (!(NFSD_VNET(nfsrv_stablefirst).nsf_flags &
1266 		      NFSNSF_GRACEOVER) &&
1267 		    NFSD_MONOSEC > NFSD_VNET(nfsrv_stablefirst).nsf_eograce)
1268 			NFSD_VNET(nfsrv_stablefirst).nsf_flags |=
1269 			    (NFSNSF_GRACEOVER | NFSNSF_NEEDLOCK);
1270 		NFSUNLOCKSTATE();
1271 		return;
1272 	}
1273 
1274 	/*
1275 	 * Try and get a reference count on the nfsv4rootfs_lock so that
1276 	 * no nfsd thread can acquire an exclusive lock on it before this
1277 	 * call is done. If it is already exclusively locked, just return.
1278 	 */
1279 	NFSLOCKV4ROOTMUTEX();
1280 	got_ref = nfsv4_getref_nonblock(&nfsv4rootfs_lock);
1281 	NFSUNLOCKV4ROOTMUTEX();
1282 	if (got_ref == 0) {
1283 		NFSUNLOCKSTATE();
1284 		return;
1285 	}
1286 
1287 	/*
1288 	 * For each client...
1289 	 */
1290 	for (i = 0; i < nfsrv_clienthashsize; i++) {
1291 	    clp = LIST_FIRST(&NFSD_VNET(nfsclienthash)[i]);
1292 	    while (clp != LIST_END(&NFSD_VNET(nfsclienthash)[i])) {
1293 		nclp = LIST_NEXT(clp, lc_hash);
1294 		if (!(clp->lc_flags & LCL_EXPIREIT)) {
1295 		    if (((clp->lc_expiry + NFSRV_STALELEASE) < NFSD_MONOSEC
1296 			 && ((LIST_EMPTY(&clp->lc_deleg)
1297 			      && LIST_EMPTY(&clp->lc_open)) ||
1298 			     nfsrv_clients > nfsrv_clienthighwater)) ||
1299 			(clp->lc_expiry + NFSRV_MOULDYLEASE) < NFSD_MONOSEC ||
1300 			(clp->lc_expiry < NFSD_MONOSEC &&
1301 			 (nfsrv_openpluslock * 10 / 9) > nfsrv_v4statelimit)) {
1302 			/*
1303 			 * Lease has expired several nfsrv_lease times ago:
1304 			 * PLUS
1305 			 *    - no state is associated with it
1306 			 *    OR
1307 			 *    - above high water mark for number of clients
1308 			 *      (nfsrv_clienthighwater should be large enough
1309 			 *       that this only occurs when clients fail to
1310 			 *       use the same nfs_client_id4.id. Maybe somewhat
1311 			 *       higher that the maximum number of clients that
1312 			 *       will mount this server?)
1313 			 * OR
1314 			 * Lease has expired a very long time ago
1315 			 * OR
1316 			 * Lease has expired PLUS the number of opens + locks
1317 			 * has exceeded 90% of capacity
1318 			 *
1319 			 * --> Mark for expiry. The actual expiry will be done
1320 			 *     by an nfsd sometime soon.
1321 			 */
1322 			clp->lc_flags |= LCL_EXPIREIT;
1323 			NFSD_VNET(nfsrv_stablefirst).nsf_flags |=
1324 			    (NFSNSF_NEEDLOCK | NFSNSF_EXPIREDCLIENT);
1325 		    } else {
1326 			/*
1327 			 * If there are no opens, increment no open tick cnt
1328 			 * If time exceeds NFSNOOPEN, mark it to be thrown away
1329 			 * otherwise, if there is an open, reset no open time
1330 			 * Hopefully, this will avoid excessive re-creation
1331 			 * of open owners and subsequent open confirms.
1332 			 */
1333 			stp = LIST_FIRST(&clp->lc_open);
1334 			while (stp != LIST_END(&clp->lc_open)) {
1335 				nstp = LIST_NEXT(stp, ls_list);
1336 				if (LIST_EMPTY(&stp->ls_open)) {
1337 					stp->ls_noopens++;
1338 					if (stp->ls_noopens > NFSNOOPEN ||
1339 					    (nfsrv_openpluslock * 2) >
1340 					    nfsrv_v4statelimit)
1341 						NFSD_VNET(nfsrv_stablefirst).nsf_flags |=
1342 							NFSNSF_NOOPENS;
1343 				} else {
1344 					stp->ls_noopens = 0;
1345 				}
1346 				stp = nstp;
1347 			}
1348 		    }
1349 		}
1350 		clp = nclp;
1351 	    }
1352 	}
1353 	NFSUNLOCKSTATE();
1354 	NFSLOCKV4ROOTMUTEX();
1355 	nfsv4_relref(&nfsv4rootfs_lock);
1356 	NFSUNLOCKV4ROOTMUTEX();
1357 }
1358 
1359 /*
1360  * The following set of functions free up the various data structures.
1361  */
1362 /*
1363  * Clear out all open/lock state related to this nfsclient.
1364  * Caller must hold an exclusive lock on nfsv4rootfs_lock, so that
1365  * there are no other active nfsd threads.
1366  */
1367 void
1368 nfsrv_cleanclient(struct nfsclient *clp, NFSPROC_T *p)
1369 {
1370 	struct nfsstate *stp, *nstp;
1371 	struct nfsdsession *sep, *nsep;
1372 
1373 	LIST_FOREACH_SAFE(stp, &clp->lc_open, ls_list, nstp)
1374 		nfsrv_freeopenowner(stp, 1, p);
1375 	if ((clp->lc_flags & LCL_ADMINREVOKED) == 0)
1376 		LIST_FOREACH_SAFE(sep, &clp->lc_session, sess_list, nsep)
1377 			(void)nfsrv_freesession(sep, NULL);
1378 }
1379 
1380 /*
1381  * Free a client that has been cleaned. It should also already have been
1382  * removed from the lists.
1383  * (Just to be safe w.r.t. newnfs_disconnect(), call this function when
1384  *  softclock interrupts are enabled.)
1385  */
1386 void
1387 nfsrv_zapclient(struct nfsclient *clp, NFSPROC_T *p)
1388 {
1389 
1390 #ifdef notyet
1391 	if ((clp->lc_flags & (LCL_GSS | LCL_CALLBACKSON)) ==
1392 	     (LCL_GSS | LCL_CALLBACKSON) &&
1393 	    (clp->lc_hand.nfsh_flag & NFSG_COMPLETE) &&
1394 	    clp->lc_handlelen > 0) {
1395 		clp->lc_hand.nfsh_flag &= ~NFSG_COMPLETE;
1396 		clp->lc_hand.nfsh_flag |= NFSG_DESTROYED;
1397 		(void) nfsrv_docallback(clp, NFSV4PROC_CBNULL,
1398 			NULL, 0, NULL, NULL, NULL, 0, p);
1399 	}
1400 #endif
1401 	newnfs_disconnect(NULL, &clp->lc_req);
1402 	free(clp->lc_req.nr_nam, M_SONAME);
1403 	NFSFREEMUTEX(&clp->lc_req.nr_mtx);
1404 	free(clp->lc_stateid, M_NFSDCLIENT);
1405 	free(clp, M_NFSDCLIENT);
1406 	NFSLOCKSTATE();
1407 	NFSD_VNET(nfsstatsv1_p)->srvclients--;
1408 	nfsrv_openpluslock--;
1409 	nfsrv_clients--;
1410 	NFSUNLOCKSTATE();
1411 }
1412 
1413 /*
1414  * Free a list of delegation state structures.
1415  * (This function will also free all nfslockfile structures that no
1416  *  longer have associated state.)
1417  */
1418 void
1419 nfsrv_freedeleglist(struct nfsstatehead *sthp)
1420 {
1421 	struct nfsstate *stp, *nstp;
1422 
1423 	LIST_FOREACH_SAFE(stp, sthp, ls_list, nstp) {
1424 		nfsrv_freedeleg(stp);
1425 	}
1426 	LIST_INIT(sthp);
1427 }
1428 
1429 /*
1430  * Free up a delegation.
1431  */
1432 static void
1433 nfsrv_freedeleg(struct nfsstate *stp)
1434 {
1435 	struct nfslockfile *lfp;
1436 
1437 	LIST_REMOVE(stp, ls_hash);
1438 	LIST_REMOVE(stp, ls_list);
1439 	LIST_REMOVE(stp, ls_file);
1440 	if ((stp->ls_flags & NFSLCK_DELEGWRITE) != 0)
1441 		nfsrv_writedelegcnt--;
1442 	lfp = stp->ls_lfp;
1443 	if (LIST_EMPTY(&lfp->lf_open) &&
1444 	    LIST_EMPTY(&lfp->lf_lock) && LIST_EMPTY(&lfp->lf_deleg) &&
1445 	    LIST_EMPTY(&lfp->lf_locallock) && LIST_EMPTY(&lfp->lf_rollback) &&
1446 	    lfp->lf_usecount == 0 &&
1447 	    nfsv4_testlock(&lfp->lf_locallock_lck) == 0)
1448 		nfsrv_freenfslockfile(lfp);
1449 	free(stp, M_NFSDSTATE);
1450 	NFSD_VNET(nfsstatsv1_p)->srvdelegates--;
1451 	nfsrv_openpluslock--;
1452 	nfsrv_delegatecnt--;
1453 }
1454 
1455 /*
1456  * This function frees an open owner and all associated opens.
1457  */
1458 static void
1459 nfsrv_freeopenowner(struct nfsstate *stp, int cansleep, NFSPROC_T *p)
1460 {
1461 	struct nfsstate *nstp, *tstp;
1462 
1463 	LIST_REMOVE(stp, ls_list);
1464 	/*
1465 	 * Now, free all associated opens.
1466 	 */
1467 	nstp = LIST_FIRST(&stp->ls_open);
1468 	while (nstp != LIST_END(&stp->ls_open)) {
1469 		tstp = nstp;
1470 		nstp = LIST_NEXT(nstp, ls_list);
1471 		(void) nfsrv_freeopen(tstp, NULL, cansleep, p);
1472 	}
1473 	if (stp->ls_op)
1474 		nfsrvd_derefcache(stp->ls_op);
1475 	free(stp, M_NFSDSTATE);
1476 	NFSD_VNET(nfsstatsv1_p)->srvopenowners--;
1477 	nfsrv_openpluslock--;
1478 }
1479 
1480 /*
1481  * This function frees an open (nfsstate open structure) with all associated
1482  * lock_owners and locks. It also frees the nfslockfile structure iff there
1483  * are no other opens on the file.
1484  * Returns 1 if it free'd the nfslockfile, 0 otherwise.
1485  */
1486 static int
1487 nfsrv_freeopen(struct nfsstate *stp, vnode_t vp, int cansleep, NFSPROC_T *p)
1488 {
1489 	struct nfsstate *nstp, *tstp;
1490 	struct nfslockfile *lfp;
1491 	int ret;
1492 
1493 	LIST_REMOVE(stp, ls_hash);
1494 	LIST_REMOVE(stp, ls_list);
1495 	LIST_REMOVE(stp, ls_file);
1496 
1497 	lfp = stp->ls_lfp;
1498 	/*
1499 	 * Now, free all lockowners associated with this open.
1500 	 */
1501 	LIST_FOREACH_SAFE(tstp, &stp->ls_open, ls_list, nstp)
1502 		nfsrv_freelockowner(tstp, vp, cansleep, p);
1503 
1504 	/*
1505 	 * The nfslockfile is freed here if there are no locks
1506 	 * associated with the open.
1507 	 * If there are locks associated with the open, the
1508 	 * nfslockfile structure can be freed via nfsrv_freelockowner().
1509 	 * Acquire the state mutex to avoid races with calls to
1510 	 * nfsrv_getlockfile().
1511 	 */
1512 	if (cansleep != 0)
1513 		NFSLOCKSTATE();
1514 	if (lfp != NULL && LIST_EMPTY(&lfp->lf_open) &&
1515 	    LIST_EMPTY(&lfp->lf_deleg) && LIST_EMPTY(&lfp->lf_lock) &&
1516 	    LIST_EMPTY(&lfp->lf_locallock) && LIST_EMPTY(&lfp->lf_rollback) &&
1517 	    lfp->lf_usecount == 0 &&
1518 	    (cansleep != 0 || nfsv4_testlock(&lfp->lf_locallock_lck) == 0)) {
1519 		nfsrv_freenfslockfile(lfp);
1520 		ret = 1;
1521 	} else
1522 		ret = 0;
1523 	if (cansleep != 0)
1524 		NFSUNLOCKSTATE();
1525 	free(stp, M_NFSDSTATE);
1526 	NFSD_VNET(nfsstatsv1_p)->srvopens--;
1527 	nfsrv_openpluslock--;
1528 	return (ret);
1529 }
1530 
1531 /*
1532  * Frees a lockowner and all associated locks.
1533  */
1534 static void
1535 nfsrv_freelockowner(struct nfsstate *stp, vnode_t vp, int cansleep,
1536     NFSPROC_T *p)
1537 {
1538 
1539 	LIST_REMOVE(stp, ls_hash);
1540 	LIST_REMOVE(stp, ls_list);
1541 	nfsrv_freeallnfslocks(stp, vp, cansleep, p);
1542 	if (stp->ls_op)
1543 		nfsrvd_derefcache(stp->ls_op);
1544 	free(stp, M_NFSDSTATE);
1545 	NFSD_VNET(nfsstatsv1_p)->srvlockowners--;
1546 	nfsrv_openpluslock--;
1547 }
1548 
1549 /*
1550  * Free all the nfs locks on a lockowner.
1551  */
1552 static void
1553 nfsrv_freeallnfslocks(struct nfsstate *stp, vnode_t vp, int cansleep,
1554     NFSPROC_T *p)
1555 {
1556 	struct nfslock *lop, *nlop;
1557 	struct nfsrollback *rlp, *nrlp;
1558 	struct nfslockfile *lfp = NULL;
1559 	int gottvp = 0;
1560 	vnode_t tvp = NULL;
1561 	uint64_t first, end;
1562 
1563 	if (vp != NULL)
1564 		ASSERT_VOP_UNLOCKED(vp, "nfsrv_freeallnfslocks: vnode locked");
1565 	lop = LIST_FIRST(&stp->ls_lock);
1566 	while (lop != LIST_END(&stp->ls_lock)) {
1567 		nlop = LIST_NEXT(lop, lo_lckowner);
1568 		/*
1569 		 * Since all locks should be for the same file, lfp should
1570 		 * not change.
1571 		 */
1572 		if (lfp == NULL)
1573 			lfp = lop->lo_lfp;
1574 		else if (lfp != lop->lo_lfp)
1575 			panic("allnfslocks");
1576 		/*
1577 		 * If vp is NULL and cansleep != 0, a vnode must be acquired
1578 		 * from the file handle. This only occurs when called from
1579 		 * nfsrv_cleanclient().
1580 		 */
1581 		if (gottvp == 0) {
1582 			if (nfsrv_dolocallocks == 0)
1583 				tvp = NULL;
1584 			else if (vp == NULL && cansleep != 0) {
1585 				tvp = nfsvno_getvp(&lfp->lf_fh);
1586 				if (tvp != NULL)
1587 					NFSVOPUNLOCK(tvp);
1588 			} else
1589 				tvp = vp;
1590 			gottvp = 1;
1591 		}
1592 
1593 		if (tvp != NULL) {
1594 			if (cansleep == 0)
1595 				panic("allnfs2");
1596 			first = lop->lo_first;
1597 			end = lop->lo_end;
1598 			nfsrv_freenfslock(lop);
1599 			nfsrv_localunlock(tvp, lfp, first, end, p);
1600 			LIST_FOREACH_SAFE(rlp, &lfp->lf_rollback, rlck_list,
1601 			    nrlp)
1602 				free(rlp, M_NFSDROLLBACK);
1603 			LIST_INIT(&lfp->lf_rollback);
1604 		} else
1605 			nfsrv_freenfslock(lop);
1606 		lop = nlop;
1607 	}
1608 	if (vp == NULL && tvp != NULL)
1609 		vrele(tvp);
1610 }
1611 
1612 /*
1613  * Free an nfslock structure.
1614  */
1615 static void
1616 nfsrv_freenfslock(struct nfslock *lop)
1617 {
1618 
1619 	if (lop->lo_lckfile.le_prev != NULL) {
1620 		LIST_REMOVE(lop, lo_lckfile);
1621 		NFSD_VNET(nfsstatsv1_p)->srvlocks--;
1622 		nfsrv_openpluslock--;
1623 	}
1624 	LIST_REMOVE(lop, lo_lckowner);
1625 	free(lop, M_NFSDLOCK);
1626 }
1627 
1628 /*
1629  * This function frees an nfslockfile structure.
1630  */
1631 static void
1632 nfsrv_freenfslockfile(struct nfslockfile *lfp)
1633 {
1634 
1635 	LIST_REMOVE(lfp, lf_hash);
1636 	free(lfp, M_NFSDLOCKFILE);
1637 }
1638 
1639 /*
1640  * This function looks up an nfsstate structure via stateid.
1641  */
1642 static int
1643 nfsrv_getstate(struct nfsclient *clp, nfsv4stateid_t *stateidp, __unused u_int32_t flags,
1644     struct nfsstate **stpp)
1645 {
1646 	struct nfsstate *stp;
1647 	struct nfsstatehead *hp;
1648 	int error = 0;
1649 
1650 	*stpp = NULL;
1651 	hp = NFSSTATEHASH(clp, *stateidp);
1652 	LIST_FOREACH(stp, hp, ls_hash) {
1653 		if (!NFSBCMP(stp->ls_stateid.other, stateidp->other,
1654 			NFSX_STATEIDOTHER))
1655 			break;
1656 	}
1657 
1658 	/*
1659 	 * If no state id in list, return NFSERR_BADSTATEID.
1660 	 */
1661 	if (stp == LIST_END(hp)) {
1662 		error = NFSERR_BADSTATEID;
1663 		goto out;
1664 	}
1665 	*stpp = stp;
1666 
1667 out:
1668 	NFSEXITCODE(error);
1669 	return (error);
1670 }
1671 
1672 /*
1673  * This function gets an nfsstate structure via owner string.
1674  */
1675 static void
1676 nfsrv_getowner(struct nfsstatehead *hp, struct nfsstate *new_stp,
1677     struct nfsstate **stpp)
1678 {
1679 	struct nfsstate *stp;
1680 
1681 	*stpp = NULL;
1682 	LIST_FOREACH(stp, hp, ls_list) {
1683 		if (new_stp->ls_ownerlen == stp->ls_ownerlen &&
1684 		  !NFSBCMP(new_stp->ls_owner,stp->ls_owner,stp->ls_ownerlen)) {
1685 			*stpp = stp;
1686 			return;
1687 		}
1688 	}
1689 }
1690 
1691 /*
1692  * Lock control function called to update lock status.
1693  * Returns 0 upon success, -1 if there is no lock and the flags indicate
1694  * that one isn't to be created and an NFSERR_xxx for other errors.
1695  * The structures new_stp and new_lop are passed in as pointers that should
1696  * be set to NULL if the structure is used and shouldn't be free'd.
1697  * For the NFSLCK_TEST and NFSLCK_CHECK cases, the structures are
1698  * never used and can safely be allocated on the stack. For all other
1699  * cases, *new_stpp and *new_lopp should be malloc'd before the call,
1700  * in case they are used.
1701  */
1702 int
1703 nfsrv_lockctrl(vnode_t vp, struct nfsstate **new_stpp,
1704     struct nfslock **new_lopp, struct nfslockconflict *cfp,
1705     nfsquad_t clientid, nfsv4stateid_t *stateidp,
1706     __unused struct nfsexstuff *exp,
1707     struct nfsrv_descript *nd, NFSPROC_T *p)
1708 {
1709 	struct nfslock *lop;
1710 	struct nfsstate *new_stp = *new_stpp;
1711 	struct nfslock *new_lop = *new_lopp;
1712 	struct nfsstate *tstp, *mystp, *nstp;
1713 	int specialid = 0;
1714 	struct nfslockfile *lfp;
1715 	struct nfslock *other_lop = NULL;
1716 	struct nfsstate *stp, *lckstp = NULL;
1717 	struct nfsclient *clp = NULL;
1718 	u_int32_t bits;
1719 	int error = 0, haslock = 0, ret, reterr;
1720 	int getlckret, delegation = 0, filestruct_locked, vnode_unlocked = 0;
1721 	fhandle_t nfh;
1722 	uint64_t first, end;
1723 	uint32_t lock_flags;
1724 
1725 	if (new_stp->ls_flags & (NFSLCK_CHECK | NFSLCK_SETATTR)) {
1726 		/*
1727 		 * Note the special cases of "all 1s" or "all 0s" stateids and
1728 		 * let reads with all 1s go ahead.
1729 		 */
1730 		if (new_stp->ls_stateid.seqid == 0x0 &&
1731 		    new_stp->ls_stateid.other[0] == 0x0 &&
1732 		    new_stp->ls_stateid.other[1] == 0x0 &&
1733 		    new_stp->ls_stateid.other[2] == 0x0)
1734 			specialid = 1;
1735 		else if (new_stp->ls_stateid.seqid == 0xffffffff &&
1736 		    new_stp->ls_stateid.other[0] == 0xffffffff &&
1737 		    new_stp->ls_stateid.other[1] == 0xffffffff &&
1738 		    new_stp->ls_stateid.other[2] == 0xffffffff)
1739 			specialid = 2;
1740 	}
1741 
1742 	/*
1743 	 * Check for restart conditions (client and server).
1744 	 */
1745 	error = nfsrv_checkrestart(clientid, new_stp->ls_flags,
1746 	    &new_stp->ls_stateid, specialid);
1747 	if (error)
1748 		goto out;
1749 
1750 	/*
1751 	 * Check for state resource limit exceeded.
1752 	 */
1753 	if ((new_stp->ls_flags & NFSLCK_LOCK) &&
1754 	    nfsrv_openpluslock > nfsrv_v4statelimit) {
1755 		error = NFSERR_RESOURCE;
1756 		goto out;
1757 	}
1758 
1759 	/*
1760 	 * For the lock case, get another nfslock structure,
1761 	 * just in case we need it.
1762 	 * Malloc now, before we start sifting through the linked lists,
1763 	 * in case we have to wait for memory.
1764 	 */
1765 tryagain:
1766 	if (new_stp->ls_flags & NFSLCK_LOCK)
1767 		other_lop = malloc(sizeof (struct nfslock),
1768 		    M_NFSDLOCK, M_WAITOK);
1769 	filestruct_locked = 0;
1770 	reterr = 0;
1771 	lfp = NULL;
1772 
1773 	/*
1774 	 * Get the lockfile structure for CFH now, so we can do a sanity
1775 	 * check against the stateid, before incrementing the seqid#, since
1776 	 * we want to return NFSERR_BADSTATEID on failure and the seqid#
1777 	 * shouldn't be incremented for this case.
1778 	 * If nfsrv_getlockfile() returns -1, it means "not found", which
1779 	 * will be handled later.
1780 	 * If we are doing Lock/LockU and local locking is enabled, sleep
1781 	 * lock the nfslockfile structure.
1782 	 */
1783 	getlckret = nfsrv_getlockfh(vp, new_stp->ls_flags, NULL, &nfh, p);
1784 	NFSLOCKSTATE();
1785 	if (getlckret == 0) {
1786 		if ((new_stp->ls_flags & (NFSLCK_LOCK | NFSLCK_UNLOCK)) != 0 &&
1787 		    nfsrv_dolocallocks != 0 && nd->nd_repstat == 0) {
1788 			getlckret = nfsrv_getlockfile(new_stp->ls_flags, NULL,
1789 			    &lfp, &nfh, 1);
1790 			if (getlckret == 0)
1791 				filestruct_locked = 1;
1792 		} else
1793 			getlckret = nfsrv_getlockfile(new_stp->ls_flags, NULL,
1794 			    &lfp, &nfh, 0);
1795 	}
1796 	if (getlckret != 0 && getlckret != -1)
1797 		reterr = getlckret;
1798 
1799 	if (filestruct_locked != 0) {
1800 		LIST_INIT(&lfp->lf_rollback);
1801 		if ((new_stp->ls_flags & NFSLCK_LOCK)) {
1802 			/*
1803 			 * For local locking, do the advisory locking now, so
1804 			 * that any conflict can be detected. A failure later
1805 			 * can be rolled back locally. If an error is returned,
1806 			 * struct nfslockfile has been unlocked and any local
1807 			 * locking rolled back.
1808 			 */
1809 			NFSUNLOCKSTATE();
1810 			if (vnode_unlocked == 0) {
1811 				ASSERT_VOP_ELOCKED(vp, "nfsrv_lockctrl1");
1812 				vnode_unlocked = 1;
1813 				NFSVOPUNLOCK(vp);
1814 			}
1815 			reterr = nfsrv_locallock(vp, lfp,
1816 			    (new_lop->lo_flags & (NFSLCK_READ | NFSLCK_WRITE)),
1817 			    new_lop->lo_first, new_lop->lo_end, cfp, p);
1818 			NFSLOCKSTATE();
1819 		}
1820 	}
1821 
1822 	if (specialid == 0) {
1823 	    if (new_stp->ls_flags & NFSLCK_TEST) {
1824 		/*
1825 		 * RFC 3530 does not list LockT as an op that renews a
1826 		 * lease, but the consensus seems to be that it is ok
1827 		 * for a server to do so.
1828 		 */
1829 		error = nfsrv_getclient(clientid, CLOPS_RENEW, &clp, NULL,
1830 		    (nfsquad_t)((u_quad_t)0), 0, nd, p);
1831 
1832 		/*
1833 		 * Since NFSERR_EXPIRED, NFSERR_ADMINREVOKED are not valid
1834 		 * error returns for LockT, just go ahead and test for a lock,
1835 		 * since there are no locks for this client, but other locks
1836 		 * can conflict. (ie. same client will always be false)
1837 		 */
1838 		if (error == NFSERR_EXPIRED || error == NFSERR_ADMINREVOKED)
1839 		    error = 0;
1840 		lckstp = new_stp;
1841 	    } else {
1842 	      error = nfsrv_getclient(clientid, CLOPS_RENEW, &clp, NULL,
1843 		(nfsquad_t)((u_quad_t)0), 0, nd, p);
1844 	      if (error == 0)
1845 		/*
1846 		 * Look up the stateid
1847 		 */
1848 		error = nfsrv_getstate(clp, &new_stp->ls_stateid,
1849 		  new_stp->ls_flags, &stp);
1850 	      /*
1851 	       * do some sanity checks for an unconfirmed open or a
1852 	       * stateid that refers to the wrong file, for an open stateid
1853 	       */
1854 	      if (error == 0 && (stp->ls_flags & NFSLCK_OPEN) &&
1855 		  ((stp->ls_openowner->ls_flags & NFSLCK_NEEDSCONFIRM) ||
1856 		   (getlckret == 0 && stp->ls_lfp != lfp))){
1857 		      /*
1858 		       * NFSLCK_SETATTR should return OK rather than NFSERR_BADSTATEID
1859 		       * The only exception is using SETATTR with SIZE.
1860 		       * */
1861                     if ((new_stp->ls_flags &
1862                          (NFSLCK_SETATTR | NFSLCK_CHECK)) != NFSLCK_SETATTR)
1863 			     error = NFSERR_BADSTATEID;
1864 	      }
1865 
1866 		if (error == 0 &&
1867 		  (stp->ls_flags & (NFSLCK_DELEGREAD | NFSLCK_DELEGWRITE)) &&
1868 		  getlckret == 0 && stp->ls_lfp != lfp)
1869 			error = NFSERR_BADSTATEID;
1870 
1871 	      /*
1872 	       * If the lockowner stateid doesn't refer to the same file,
1873 	       * I believe that is considered ok, since some clients will
1874 	       * only create a single lockowner and use that for all locks
1875 	       * on all files.
1876 	       * For now, log it as a diagnostic, instead of considering it
1877 	       * a BadStateid.
1878 	       */
1879 	      if (error == 0 && (stp->ls_flags &
1880 		  (NFSLCK_OPEN | NFSLCK_DELEGREAD | NFSLCK_DELEGWRITE)) == 0 &&
1881 		  getlckret == 0 && stp->ls_lfp != lfp) {
1882 #ifdef DIAGNOSTIC
1883 		  printf("Got a lock statid for different file open\n");
1884 #endif
1885 		  /*
1886 		  error = NFSERR_BADSTATEID;
1887 		  */
1888 	      }
1889 
1890 	      if (error == 0) {
1891 		    if (new_stp->ls_flags & NFSLCK_OPENTOLOCK) {
1892 			/*
1893 			 * If haslock set, we've already checked the seqid.
1894 			 */
1895 			if (!haslock) {
1896 			    if (stp->ls_flags & NFSLCK_OPEN)
1897 				error = nfsrv_checkseqid(nd, new_stp->ls_seq,
1898 				    stp->ls_openowner, new_stp->ls_op);
1899 			    else
1900 				error = NFSERR_BADSTATEID;
1901 			}
1902 			if (!error)
1903 			    nfsrv_getowner(&stp->ls_open, new_stp, &lckstp);
1904 			if (lckstp) {
1905 			    /*
1906 			     * For NFSv4.1 and NFSv4.2 allow an
1907 			     * open_to_lock_owner when the lock_owner already
1908 			     * exists.  Just clear NFSLCK_OPENTOLOCK so that
1909 			     * a new lock_owner will not be created.
1910 			     * RFC7530 states that the error for NFSv4.0
1911 			     * is NFS4ERR_BAD_SEQID.
1912 			     */
1913 			    if ((nd->nd_flag & ND_NFSV41) != 0)
1914 				new_stp->ls_flags &= ~NFSLCK_OPENTOLOCK;
1915 			    else
1916 				error = NFSERR_BADSEQID;
1917 			} else
1918 			    lckstp = new_stp;
1919 		    } else if (new_stp->ls_flags&(NFSLCK_LOCK|NFSLCK_UNLOCK)) {
1920 			/*
1921 			 * If haslock set, ditto above.
1922 			 */
1923 			if (!haslock) {
1924 			    if (stp->ls_flags & NFSLCK_OPEN)
1925 				error = NFSERR_BADSTATEID;
1926 			    else
1927 				error = nfsrv_checkseqid(nd, new_stp->ls_seq,
1928 				    stp, new_stp->ls_op);
1929 			}
1930 			lckstp = stp;
1931 		    } else {
1932 			lckstp = stp;
1933 		    }
1934 	      }
1935 	      /*
1936 	       * If the seqid part of the stateid isn't the same, return
1937 	       * NFSERR_OLDSTATEID for cases other than I/O Ops.
1938 	       * For I/O Ops, only return NFSERR_OLDSTATEID if
1939 	       * nfsrv_returnoldstateid is set. (The consensus on the email
1940 	       * list was that most clients would prefer to not receive
1941 	       * NFSERR_OLDSTATEID for I/O Ops, but the RFC suggests that that
1942 	       * is what will happen, so I use the nfsrv_returnoldstateid to
1943 	       * allow for either server configuration.)
1944 	       */
1945 	      if (!error && stp->ls_stateid.seqid!=new_stp->ls_stateid.seqid &&
1946 		  (((nd->nd_flag & ND_NFSV41) == 0 &&
1947 		   (!(new_stp->ls_flags & NFSLCK_CHECK) ||
1948 		    nfsrv_returnoldstateid)) ||
1949 		   ((nd->nd_flag & ND_NFSV41) != 0 &&
1950 		    new_stp->ls_stateid.seqid != 0)))
1951 		    error = NFSERR_OLDSTATEID;
1952 	    }
1953 	}
1954 
1955 	/*
1956 	 * Now we can check for grace.
1957 	 */
1958 	if (!error)
1959 		error = nfsrv_checkgrace(nd, clp, new_stp->ls_flags);
1960 	if ((new_stp->ls_flags & NFSLCK_RECLAIM) && !error &&
1961 		nfsrv_checkstable(clp))
1962 		error = NFSERR_NOGRACE;
1963 	/*
1964 	 * If we successfully Reclaimed state, note that.
1965 	 */
1966 	if ((new_stp->ls_flags & NFSLCK_RECLAIM) && !error)
1967 		nfsrv_markstable(clp);
1968 
1969 	/*
1970 	 * At this point, either error == NFSERR_BADSTATEID or the
1971 	 * seqid# has been updated, so we can return any error.
1972 	 * If error == 0, there may be an error in:
1973 	 *    nd_repstat - Set by the calling function.
1974 	 *    reterr - Set above, if getting the nfslockfile structure
1975 	 *       or acquiring the local lock failed.
1976 	 *    (If both of these are set, nd_repstat should probably be
1977 	 *     returned, since that error was detected before this
1978 	 *     function call.)
1979 	 */
1980 	if (error != 0 || nd->nd_repstat != 0 || reterr != 0) {
1981 		if (error == 0) {
1982 			if (nd->nd_repstat != 0)
1983 				error = nd->nd_repstat;
1984 			else
1985 				error = reterr;
1986 		}
1987 		if (filestruct_locked != 0) {
1988 			/* Roll back local locks. */
1989 			NFSUNLOCKSTATE();
1990 			if (vnode_unlocked == 0) {
1991 				ASSERT_VOP_ELOCKED(vp, "nfsrv_lockctrl2");
1992 				vnode_unlocked = 1;
1993 				NFSVOPUNLOCK(vp);
1994 			}
1995 			nfsrv_locallock_rollback(vp, lfp, p);
1996 			NFSLOCKSTATE();
1997 			nfsrv_unlocklf(lfp);
1998 		}
1999 		NFSUNLOCKSTATE();
2000 		goto out;
2001 	}
2002 
2003 	/*
2004 	 * Check the nfsrv_getlockfile return.
2005 	 * Returned -1 if no structure found.
2006 	 */
2007 	if (getlckret == -1) {
2008 		error = NFSERR_EXPIRED;
2009 		/*
2010 		 * Called from lockt, so no lock is OK.
2011 		 */
2012 		if (new_stp->ls_flags & NFSLCK_TEST) {
2013 			error = 0;
2014 		} else if (new_stp->ls_flags &
2015 		    (NFSLCK_CHECK | NFSLCK_SETATTR)) {
2016 			/*
2017 			 * Called to check for a lock, OK if the stateid is all
2018 			 * 1s or all 0s, but there should be an nfsstate
2019 			 * otherwise.
2020 			 * (ie. If there is no open, I'll assume no share
2021 			 *  deny bits.)
2022 			 */
2023 			if (specialid)
2024 				error = 0;
2025 			else
2026 				error = NFSERR_BADSTATEID;
2027 		}
2028 		NFSUNLOCKSTATE();
2029 		goto out;
2030 	}
2031 
2032 	/*
2033 	 * For NFSLCK_CHECK and NFSLCK_LOCK, test for a share conflict.
2034 	 * For NFSLCK_CHECK, allow a read if write access is granted,
2035 	 * but check for a deny. For NFSLCK_LOCK, require correct access,
2036 	 * which implies a conflicting deny can't exist.
2037 	 */
2038 	if (new_stp->ls_flags & (NFSLCK_CHECK | NFSLCK_LOCK)) {
2039 	    /*
2040 	     * Four kinds of state id:
2041 	     * - specialid (all 0s or all 1s), only for NFSLCK_CHECK
2042 	     * - stateid for an open
2043 	     * - stateid for a delegation
2044 	     * - stateid for a lock owner
2045 	     */
2046 	    if (!specialid) {
2047 		if (stp->ls_flags & (NFSLCK_DELEGREAD | NFSLCK_DELEGWRITE)) {
2048 		    delegation = 1;
2049 		    mystp = stp;
2050 		    nfsrv_delaydelegtimeout(stp);
2051 	        } else if (stp->ls_flags & NFSLCK_OPEN) {
2052 		    mystp = stp;
2053 		} else {
2054 		    mystp = stp->ls_openstp;
2055 		}
2056 		/*
2057 		 * If locking or checking, require correct access
2058 		 * bit set.
2059 		 */
2060 		if (((new_stp->ls_flags & NFSLCK_LOCK) &&
2061 		     !((new_lop->lo_flags >> NFSLCK_LOCKSHIFT) &
2062 		       mystp->ls_flags & NFSLCK_ACCESSBITS)) ||
2063 		    ((new_stp->ls_flags & (NFSLCK_CHECK|NFSLCK_READACCESS)) ==
2064 		      (NFSLCK_CHECK | NFSLCK_READACCESS) &&
2065 		     !(mystp->ls_flags & NFSLCK_READACCESS) &&
2066 		     nfsrv_allowreadforwriteopen == 0) ||
2067 		    ((new_stp->ls_flags & (NFSLCK_CHECK|NFSLCK_WRITEACCESS)) ==
2068 		      (NFSLCK_CHECK | NFSLCK_WRITEACCESS) &&
2069 		     !(mystp->ls_flags & NFSLCK_WRITEACCESS))) {
2070 			if (filestruct_locked != 0) {
2071 				/* Roll back local locks. */
2072 				NFSUNLOCKSTATE();
2073 				if (vnode_unlocked == 0) {
2074 					ASSERT_VOP_ELOCKED(vp,
2075 					    "nfsrv_lockctrl3");
2076 					vnode_unlocked = 1;
2077 					NFSVOPUNLOCK(vp);
2078 				}
2079 				nfsrv_locallock_rollback(vp, lfp, p);
2080 				NFSLOCKSTATE();
2081 				nfsrv_unlocklf(lfp);
2082 			}
2083 			NFSUNLOCKSTATE();
2084 			error = NFSERR_OPENMODE;
2085 			goto out;
2086 		}
2087 	    } else
2088 		mystp = NULL;
2089 	    if ((new_stp->ls_flags & NFSLCK_CHECK) && !delegation) {
2090 		/*
2091 		 * Check for a conflicting deny bit.
2092 		 */
2093 		LIST_FOREACH(tstp, &lfp->lf_open, ls_file) {
2094 		    if (tstp != mystp) {
2095 			bits = tstp->ls_flags;
2096 			bits >>= NFSLCK_SHIFT;
2097 			if (new_stp->ls_flags & bits & NFSLCK_ACCESSBITS) {
2098 			    KASSERT(vnode_unlocked == 0,
2099 				("nfsrv_lockctrl: vnode unlocked1"));
2100 			    ret = nfsrv_clientconflict(tstp->ls_clp, &haslock,
2101 				vp, p);
2102 			    if (ret == 1) {
2103 				/*
2104 				* nfsrv_clientconflict unlocks state
2105 				 * when it returns non-zero.
2106 				 */
2107 				lckstp = NULL;
2108 				goto tryagain;
2109 			    }
2110 			    if (ret == 0)
2111 				NFSUNLOCKSTATE();
2112 			    if (ret == 2)
2113 				error = NFSERR_PERM;
2114 			    else
2115 				error = NFSERR_OPENMODE;
2116 			    goto out;
2117 			}
2118 		    }
2119 		}
2120 
2121 		/* We're outta here */
2122 		NFSUNLOCKSTATE();
2123 		goto out;
2124 	    }
2125 	}
2126 
2127 	/*
2128 	 * For setattr, just get rid of all the Delegations for other clients.
2129 	 */
2130 	if (new_stp->ls_flags & NFSLCK_SETATTR) {
2131 		KASSERT(vnode_unlocked == 0,
2132 		    ("nfsrv_lockctrl: vnode unlocked2"));
2133 		ret = nfsrv_cleandeleg(vp, lfp, clp, &haslock, p);
2134 		if (ret) {
2135 			/*
2136 			 * nfsrv_cleandeleg() unlocks state when it
2137 			 * returns non-zero.
2138 			 */
2139 			if (ret == -1) {
2140 				lckstp = NULL;
2141 				goto tryagain;
2142 			}
2143 			error = ret;
2144 			goto out;
2145 		}
2146 		if (!(new_stp->ls_flags & NFSLCK_CHECK) ||
2147 		    (LIST_EMPTY(&lfp->lf_open) && LIST_EMPTY(&lfp->lf_lock) &&
2148 		     LIST_EMPTY(&lfp->lf_deleg))) {
2149 			NFSUNLOCKSTATE();
2150 			goto out;
2151 		}
2152 	}
2153 
2154 	/*
2155 	 * Check for a conflicting delegation. If one is found, call
2156 	 * nfsrv_delegconflict() to handle it. If the v4root lock hasn't
2157 	 * been set yet, it will get the lock. Otherwise, it will recall
2158 	 * the delegation. Then, we try try again...
2159 	 * I currently believe the conflict algorithm to be:
2160 	 * For Lock Ops (Lock/LockT/LockU)
2161 	 * - there is a conflict iff a different client has a write delegation
2162 	 * For Reading (Read Op)
2163 	 * - there is a conflict iff a different client has a write delegation
2164 	 *   (the specialids are always a different client)
2165 	 * For Writing (Write/Setattr of size)
2166 	 * - there is a conflict if a different client has any delegation
2167 	 * - there is a conflict if the same client has a read delegation
2168 	 *   (I don't understand why this isn't allowed, but that seems to be
2169 	 *    the current consensus?)
2170 	 */
2171 	tstp = LIST_FIRST(&lfp->lf_deleg);
2172 	while (tstp != LIST_END(&lfp->lf_deleg)) {
2173 	    nstp = LIST_NEXT(tstp, ls_file);
2174 	    if ((((new_stp->ls_flags&(NFSLCK_LOCK|NFSLCK_UNLOCK|NFSLCK_TEST))||
2175 		 ((new_stp->ls_flags & NFSLCK_CHECK) &&
2176 		  (new_lop->lo_flags & NFSLCK_READ))) &&
2177 		  clp != tstp->ls_clp &&
2178 		 (tstp->ls_flags & NFSLCK_DELEGWRITE)) ||
2179 		 ((new_stp->ls_flags & NFSLCK_CHECK) &&
2180 		   (new_lop->lo_flags & NFSLCK_WRITE) &&
2181 		  (clp != tstp->ls_clp ||
2182 		   (tstp->ls_flags & NFSLCK_DELEGREAD)))) {
2183 		ret = 0;
2184 		if (filestruct_locked != 0) {
2185 			/* Roll back local locks. */
2186 			NFSUNLOCKSTATE();
2187 			if (vnode_unlocked == 0) {
2188 				ASSERT_VOP_ELOCKED(vp, "nfsrv_lockctrl4");
2189 				NFSVOPUNLOCK(vp);
2190 			}
2191 			nfsrv_locallock_rollback(vp, lfp, p);
2192 			NFSLOCKSTATE();
2193 			nfsrv_unlocklf(lfp);
2194 			NFSUNLOCKSTATE();
2195 			NFSVOPLOCK(vp, LK_EXCLUSIVE | LK_RETRY);
2196 			vnode_unlocked = 0;
2197 			if (VN_IS_DOOMED(vp))
2198 				ret = NFSERR_SERVERFAULT;
2199 			NFSLOCKSTATE();
2200 		}
2201 		if (ret == 0)
2202 			ret = nfsrv_delegconflict(tstp, &haslock, p, vp);
2203 		if (ret) {
2204 		    /*
2205 		     * nfsrv_delegconflict unlocks state when it
2206 		     * returns non-zero, which it always does.
2207 		     */
2208 		    if (other_lop) {
2209 			free(other_lop, M_NFSDLOCK);
2210 			other_lop = NULL;
2211 		    }
2212 		    if (ret == -1) {
2213 			lckstp = NULL;
2214 			goto tryagain;
2215 		    }
2216 		    error = ret;
2217 		    goto out;
2218 		}
2219 		/* Never gets here. */
2220 	    }
2221 	    tstp = nstp;
2222 	}
2223 
2224 	/*
2225 	 * Handle the unlock case by calling nfsrv_updatelock().
2226 	 * (Should I have done some access checking above for unlock? For now,
2227 	 *  just let it happen.)
2228 	 */
2229 	if (new_stp->ls_flags & NFSLCK_UNLOCK) {
2230 		first = new_lop->lo_first;
2231 		end = new_lop->lo_end;
2232 		nfsrv_updatelock(stp, new_lopp, &other_lop, lfp);
2233 		stateidp->seqid = ++(stp->ls_stateid.seqid);
2234 		if ((nd->nd_flag & ND_NFSV41) != 0 && stateidp->seqid == 0)
2235 			stateidp->seqid = stp->ls_stateid.seqid = 1;
2236 		stateidp->other[0] = stp->ls_stateid.other[0];
2237 		stateidp->other[1] = stp->ls_stateid.other[1];
2238 		stateidp->other[2] = stp->ls_stateid.other[2];
2239 		if (filestruct_locked != 0) {
2240 			NFSUNLOCKSTATE();
2241 			if (vnode_unlocked == 0) {
2242 				ASSERT_VOP_ELOCKED(vp, "nfsrv_lockctrl5");
2243 				vnode_unlocked = 1;
2244 				NFSVOPUNLOCK(vp);
2245 			}
2246 			/* Update the local locks. */
2247 			nfsrv_localunlock(vp, lfp, first, end, p);
2248 			NFSLOCKSTATE();
2249 			nfsrv_unlocklf(lfp);
2250 		}
2251 		NFSUNLOCKSTATE();
2252 		goto out;
2253 	}
2254 
2255 	/*
2256 	 * Search for a conflicting lock. A lock conflicts if:
2257 	 * - the lock range overlaps and
2258 	 * - at least one lock is a write lock and
2259 	 * - it is not owned by the same lock owner
2260 	 */
2261 	if (!delegation) {
2262 	  LIST_FOREACH(lop, &lfp->lf_lock, lo_lckfile) {
2263 	    if (new_lop->lo_end > lop->lo_first &&
2264 		new_lop->lo_first < lop->lo_end &&
2265 		(new_lop->lo_flags == NFSLCK_WRITE ||
2266 		 lop->lo_flags == NFSLCK_WRITE) &&
2267 		lckstp != lop->lo_stp &&
2268 		(clp != lop->lo_stp->ls_clp ||
2269 		 lckstp->ls_ownerlen != lop->lo_stp->ls_ownerlen ||
2270 		 NFSBCMP(lckstp->ls_owner, lop->lo_stp->ls_owner,
2271 		    lckstp->ls_ownerlen))) {
2272 		if (other_lop) {
2273 		    free(other_lop, M_NFSDLOCK);
2274 		    other_lop = NULL;
2275 		}
2276 		if (vnode_unlocked != 0)
2277 		    ret = nfsrv_clientconflict(lop->lo_stp->ls_clp, &haslock,
2278 			NULL, p);
2279 		else
2280 		    ret = nfsrv_clientconflict(lop->lo_stp->ls_clp, &haslock,
2281 			vp, p);
2282 		if (ret == 1) {
2283 		    if (filestruct_locked != 0) {
2284 			if (vnode_unlocked == 0) {
2285 				ASSERT_VOP_ELOCKED(vp, "nfsrv_lockctrl6");
2286 				NFSVOPUNLOCK(vp);
2287 			}
2288 			/* Roll back local locks. */
2289 			nfsrv_locallock_rollback(vp, lfp, p);
2290 			NFSLOCKSTATE();
2291 			nfsrv_unlocklf(lfp);
2292 			NFSUNLOCKSTATE();
2293 			NFSVOPLOCK(vp, LK_EXCLUSIVE | LK_RETRY);
2294 			vnode_unlocked = 0;
2295 			if (VN_IS_DOOMED(vp)) {
2296 				error = NFSERR_SERVERFAULT;
2297 				goto out;
2298 			}
2299 		    }
2300 		    /*
2301 		     * nfsrv_clientconflict() unlocks state when it
2302 		     * returns non-zero.
2303 		     */
2304 		    lckstp = NULL;
2305 		    goto tryagain;
2306 		}
2307 		/*
2308 		 * Found a conflicting lock, so record the conflict and
2309 		 * return the error.
2310 		 */
2311 		if (cfp != NULL && ret == 0) {
2312 		    cfp->cl_clientid.lval[0]=lop->lo_stp->ls_stateid.other[0];
2313 		    cfp->cl_clientid.lval[1]=lop->lo_stp->ls_stateid.other[1];
2314 		    cfp->cl_first = lop->lo_first;
2315 		    cfp->cl_end = lop->lo_end;
2316 		    cfp->cl_flags = lop->lo_flags;
2317 		    cfp->cl_ownerlen = lop->lo_stp->ls_ownerlen;
2318 		    NFSBCOPY(lop->lo_stp->ls_owner, cfp->cl_owner,
2319 			cfp->cl_ownerlen);
2320 		}
2321 		if (ret == 2)
2322 		    error = NFSERR_PERM;
2323 		else if (new_stp->ls_flags & NFSLCK_RECLAIM)
2324 		    error = NFSERR_RECLAIMCONFLICT;
2325 		else if (new_stp->ls_flags & NFSLCK_CHECK)
2326 		    error = NFSERR_LOCKED;
2327 		else
2328 		    error = NFSERR_DENIED;
2329 		if (filestruct_locked != 0 && ret == 0) {
2330 			/* Roll back local locks. */
2331 			NFSUNLOCKSTATE();
2332 			if (vnode_unlocked == 0) {
2333 				ASSERT_VOP_ELOCKED(vp, "nfsrv_lockctrl7");
2334 				vnode_unlocked = 1;
2335 				NFSVOPUNLOCK(vp);
2336 			}
2337 			nfsrv_locallock_rollback(vp, lfp, p);
2338 			NFSLOCKSTATE();
2339 			nfsrv_unlocklf(lfp);
2340 		}
2341 		if (ret == 0)
2342 			NFSUNLOCKSTATE();
2343 		goto out;
2344 	    }
2345 	  }
2346 	}
2347 
2348 	/*
2349 	 * We only get here if there was no lock that conflicted.
2350 	 */
2351 	if (new_stp->ls_flags & (NFSLCK_TEST | NFSLCK_CHECK)) {
2352 		NFSUNLOCKSTATE();
2353 		goto out;
2354 	}
2355 
2356 	/*
2357 	 * We only get here when we are creating or modifying a lock.
2358 	 * There are two variants:
2359 	 * - exist_lock_owner where lock_owner exists
2360 	 * - open_to_lock_owner with new lock_owner
2361 	 */
2362 	first = new_lop->lo_first;
2363 	end = new_lop->lo_end;
2364 	lock_flags = new_lop->lo_flags;
2365 	if (!(new_stp->ls_flags & NFSLCK_OPENTOLOCK)) {
2366 		nfsrv_updatelock(lckstp, new_lopp, &other_lop, lfp);
2367 		stateidp->seqid = ++(lckstp->ls_stateid.seqid);
2368 		if ((nd->nd_flag & ND_NFSV41) != 0 && stateidp->seqid == 0)
2369 			stateidp->seqid = lckstp->ls_stateid.seqid = 1;
2370 		stateidp->other[0] = lckstp->ls_stateid.other[0];
2371 		stateidp->other[1] = lckstp->ls_stateid.other[1];
2372 		stateidp->other[2] = lckstp->ls_stateid.other[2];
2373 	} else {
2374 		/*
2375 		 * The new open_to_lock_owner case.
2376 		 * Link the new nfsstate into the lists.
2377 		 */
2378 		new_stp->ls_seq = new_stp->ls_opentolockseq;
2379 		nfsrvd_refcache(new_stp->ls_op);
2380 		stateidp->seqid = new_stp->ls_stateid.seqid = 1;
2381 		stateidp->other[0] = new_stp->ls_stateid.other[0] =
2382 		    clp->lc_clientid.lval[0];
2383 		stateidp->other[1] = new_stp->ls_stateid.other[1] =
2384 		    clp->lc_clientid.lval[1];
2385 		stateidp->other[2] = new_stp->ls_stateid.other[2] =
2386 		    nfsrv_nextstateindex(clp);
2387 		new_stp->ls_clp = clp;
2388 		LIST_INIT(&new_stp->ls_lock);
2389 		new_stp->ls_openstp = stp;
2390 		new_stp->ls_lfp = lfp;
2391 		nfsrv_insertlock(new_lop, (struct nfslock *)new_stp, new_stp,
2392 		    lfp);
2393 		LIST_INSERT_HEAD(NFSSTATEHASH(clp, new_stp->ls_stateid),
2394 		    new_stp, ls_hash);
2395 		LIST_INSERT_HEAD(&stp->ls_open, new_stp, ls_list);
2396 		*new_lopp = NULL;
2397 		*new_stpp = NULL;
2398 		NFSD_VNET(nfsstatsv1_p)->srvlockowners++;
2399 		nfsrv_openpluslock++;
2400 	}
2401 	if (filestruct_locked != 0) {
2402 		NFSUNLOCKSTATE();
2403 		nfsrv_locallock_commit(lfp, lock_flags, first, end);
2404 		NFSLOCKSTATE();
2405 		nfsrv_unlocklf(lfp);
2406 	}
2407 	NFSUNLOCKSTATE();
2408 
2409 out:
2410 	if (haslock) {
2411 		NFSLOCKV4ROOTMUTEX();
2412 		nfsv4_unlock(&nfsv4rootfs_lock, 1);
2413 		NFSUNLOCKV4ROOTMUTEX();
2414 	}
2415 	if (vnode_unlocked != 0) {
2416 		NFSVOPLOCK(vp, LK_EXCLUSIVE | LK_RETRY);
2417 		if (error == 0 && VN_IS_DOOMED(vp))
2418 			error = NFSERR_SERVERFAULT;
2419 	}
2420 	if (other_lop)
2421 		free(other_lop, M_NFSDLOCK);
2422 	NFSEXITCODE2(error, nd);
2423 	return (error);
2424 }
2425 
2426 /*
2427  * Check for state errors for Open.
2428  * repstat is passed back out as an error if more critical errors
2429  * are not detected.
2430  */
2431 int
2432 nfsrv_opencheck(nfsquad_t clientid, nfsv4stateid_t *stateidp,
2433     struct nfsstate *new_stp, vnode_t vp, struct nfsrv_descript *nd,
2434     NFSPROC_T *p, int repstat)
2435 {
2436 	struct nfsstate *stp, *nstp;
2437 	struct nfsclient *clp;
2438 	struct nfsstate *ownerstp;
2439 	struct nfslockfile *lfp, *new_lfp;
2440 	int error = 0, haslock = 0, ret, readonly = 0, getfhret = 0;
2441 
2442 	if ((new_stp->ls_flags & NFSLCK_SHAREBITS) == NFSLCK_READACCESS)
2443 		readonly = 1;
2444 	/*
2445 	 * Check for restart conditions (client and server).
2446 	 */
2447 	error = nfsrv_checkrestart(clientid, new_stp->ls_flags,
2448 		&new_stp->ls_stateid, 0);
2449 	if (error)
2450 		goto out;
2451 
2452 	/*
2453 	 * Check for state resource limit exceeded.
2454 	 * Technically this should be SMP protected, but the worst
2455 	 * case error is "out by one or two" on the count when it
2456 	 * returns NFSERR_RESOURCE and the limit is just a rather
2457 	 * arbitrary high water mark, so no harm is done.
2458 	 */
2459 	if (nfsrv_openpluslock > nfsrv_v4statelimit) {
2460 		error = NFSERR_RESOURCE;
2461 		goto out;
2462 	}
2463 
2464 tryagain:
2465 	new_lfp = malloc(sizeof (struct nfslockfile),
2466 	    M_NFSDLOCKFILE, M_WAITOK);
2467 	if (vp)
2468 		getfhret = nfsrv_getlockfh(vp, new_stp->ls_flags, new_lfp,
2469 		    NULL, p);
2470 	NFSLOCKSTATE();
2471 	/*
2472 	 * Get the nfsclient structure.
2473 	 */
2474 	error = nfsrv_getclient(clientid, CLOPS_RENEW, &clp, NULL,
2475 	    (nfsquad_t)((u_quad_t)0), 0, nd, p);
2476 
2477 	/*
2478 	 * Look up the open owner. See if it needs confirmation and
2479 	 * check the seq#, as required.
2480 	 */
2481 	if (!error)
2482 		nfsrv_getowner(&clp->lc_open, new_stp, &ownerstp);
2483 
2484 	if (!error && ownerstp) {
2485 		error = nfsrv_checkseqid(nd, new_stp->ls_seq, ownerstp,
2486 		    new_stp->ls_op);
2487 		/*
2488 		 * If the OpenOwner hasn't been confirmed, assume the
2489 		 * old one was a replay and this one is ok.
2490 		 * See: RFC3530 Sec. 14.2.18.
2491 		 */
2492 		if (error == NFSERR_BADSEQID &&
2493 		    (ownerstp->ls_flags & NFSLCK_NEEDSCONFIRM))
2494 			error = 0;
2495 	}
2496 
2497 	/*
2498 	 * Check for grace.
2499 	 */
2500 	if (!error)
2501 		error = nfsrv_checkgrace(nd, clp, new_stp->ls_flags);
2502 	if ((new_stp->ls_flags & NFSLCK_RECLAIM) && !error &&
2503 		nfsrv_checkstable(clp))
2504 		error = NFSERR_NOGRACE;
2505 
2506 	/*
2507 	 * If none of the above errors occurred, let repstat be
2508 	 * returned.
2509 	 */
2510 	if (repstat && !error)
2511 		error = repstat;
2512 	if (error) {
2513 		NFSUNLOCKSTATE();
2514 		if (haslock) {
2515 			NFSLOCKV4ROOTMUTEX();
2516 			nfsv4_unlock(&nfsv4rootfs_lock, 1);
2517 			NFSUNLOCKV4ROOTMUTEX();
2518 		}
2519 		free(new_lfp, M_NFSDLOCKFILE);
2520 		goto out;
2521 	}
2522 
2523 	/*
2524 	 * If vp == NULL, the file doesn't exist yet, so return ok.
2525 	 * (This always happens on the first pass, so haslock must be 0.)
2526 	 */
2527 	if (vp == NULL) {
2528 		NFSUNLOCKSTATE();
2529 		free(new_lfp, M_NFSDLOCKFILE);
2530 		goto out;
2531 	}
2532 
2533 	/*
2534 	 * Get the structure for the underlying file.
2535 	 */
2536 	if (getfhret)
2537 		error = getfhret;
2538 	else
2539 		error = nfsrv_getlockfile(new_stp->ls_flags, &new_lfp, &lfp,
2540 		    NULL, 0);
2541 	if (new_lfp)
2542 		free(new_lfp, M_NFSDLOCKFILE);
2543 	if (error) {
2544 		NFSUNLOCKSTATE();
2545 		if (haslock) {
2546 			NFSLOCKV4ROOTMUTEX();
2547 			nfsv4_unlock(&nfsv4rootfs_lock, 1);
2548 			NFSUNLOCKV4ROOTMUTEX();
2549 		}
2550 		goto out;
2551 	}
2552 
2553 	/*
2554 	 * Search for a conflicting open/share.
2555 	 */
2556 	if (new_stp->ls_flags & NFSLCK_DELEGCUR) {
2557 	    /*
2558 	     * For Delegate_Cur, search for the matching Delegation,
2559 	     * which indicates no conflict.
2560 	     * An old delegation should have been recovered by the
2561 	     * client doing a Claim_DELEGATE_Prev, so I won't let
2562 	     * it match and return NFSERR_EXPIRED. Should I let it
2563 	     * match?
2564 	     */
2565 	    LIST_FOREACH(stp, &lfp->lf_deleg, ls_file) {
2566 		if (!(stp->ls_flags & NFSLCK_OLDDELEG) &&
2567 		    (((nd->nd_flag & ND_NFSV41) != 0 &&
2568 		    stateidp->seqid == 0) ||
2569 		    stateidp->seqid == stp->ls_stateid.seqid) &&
2570 		    !NFSBCMP(stateidp->other, stp->ls_stateid.other,
2571 			  NFSX_STATEIDOTHER))
2572 			break;
2573 	    }
2574 	    if (stp == LIST_END(&lfp->lf_deleg) ||
2575 		((new_stp->ls_flags & NFSLCK_WRITEACCESS) &&
2576 		 (stp->ls_flags & NFSLCK_DELEGREAD))) {
2577 		NFSUNLOCKSTATE();
2578 		if (haslock) {
2579 			NFSLOCKV4ROOTMUTEX();
2580 			nfsv4_unlock(&nfsv4rootfs_lock, 1);
2581 			NFSUNLOCKV4ROOTMUTEX();
2582 		}
2583 		error = NFSERR_EXPIRED;
2584 		goto out;
2585 	    }
2586 	}
2587 
2588 	/*
2589 	 * Check for access/deny bit conflicts. I check for the same
2590 	 * owner as well, in case the client didn't bother.
2591 	 */
2592 	LIST_FOREACH(stp, &lfp->lf_open, ls_file) {
2593 		if (!(new_stp->ls_flags & NFSLCK_DELEGCUR) &&
2594 		    (((new_stp->ls_flags & NFSLCK_ACCESSBITS) &
2595 		      ((stp->ls_flags>>NFSLCK_SHIFT) & NFSLCK_ACCESSBITS))||
2596 		     ((stp->ls_flags & NFSLCK_ACCESSBITS) &
2597 		      ((new_stp->ls_flags>>NFSLCK_SHIFT)&NFSLCK_ACCESSBITS)))){
2598 			ret = nfsrv_clientconflict(stp->ls_clp,&haslock,vp,p);
2599 			if (ret == 1) {
2600 				/*
2601 				 * nfsrv_clientconflict() unlocks
2602 				 * state when it returns non-zero.
2603 				 */
2604 				goto tryagain;
2605 			}
2606 			if (ret == 2)
2607 				error = NFSERR_PERM;
2608 			else if (new_stp->ls_flags & NFSLCK_RECLAIM)
2609 				error = NFSERR_RECLAIMCONFLICT;
2610 			else
2611 				error = NFSERR_SHAREDENIED;
2612 			if (ret == 0)
2613 				NFSUNLOCKSTATE();
2614 			if (haslock) {
2615 				NFSLOCKV4ROOTMUTEX();
2616 				nfsv4_unlock(&nfsv4rootfs_lock, 1);
2617 				NFSUNLOCKV4ROOTMUTEX();
2618 			}
2619 			goto out;
2620 		}
2621 	}
2622 
2623 	/*
2624 	 * Check for a conflicting delegation. If one is found, call
2625 	 * nfsrv_delegconflict() to handle it. If the v4root lock hasn't
2626 	 * been set yet, it will get the lock. Otherwise, it will recall
2627 	 * the delegation. Then, we try try again...
2628 	 * (If NFSLCK_DELEGCUR is set, it has a delegation, so there
2629 	 *  isn't a conflict.)
2630 	 * I currently believe the conflict algorithm to be:
2631 	 * For Open with Read Access and Deny None
2632 	 * - there is a conflict iff a different client has a write delegation
2633 	 * For Open with other Write Access or any Deny except None
2634 	 * - there is a conflict if a different client has any delegation
2635 	 * - there is a conflict if the same client has a read delegation
2636 	 *   (The current consensus is that this last case should be
2637 	 *    considered a conflict since the client with a read delegation
2638 	 *    could have done an Open with ReadAccess and WriteDeny
2639 	 *    locally and then not have checked for the WriteDeny.)
2640 	 * Don't check for a Reclaim, since that will be dealt with
2641 	 * by nfsrv_openctrl().
2642 	 */
2643 	if (!(new_stp->ls_flags &
2644 		(NFSLCK_DELEGPREV | NFSLCK_DELEGCUR | NFSLCK_RECLAIM))) {
2645 	    stp = LIST_FIRST(&lfp->lf_deleg);
2646 	    while (stp != LIST_END(&lfp->lf_deleg)) {
2647 		nstp = LIST_NEXT(stp, ls_file);
2648 		if ((readonly && stp->ls_clp != clp &&
2649 		       (stp->ls_flags & NFSLCK_DELEGWRITE)) ||
2650 		    (!readonly && (stp->ls_clp != clp ||
2651 		         (stp->ls_flags & NFSLCK_DELEGREAD)))) {
2652 			ret = nfsrv_delegconflict(stp, &haslock, p, vp);
2653 			if (ret) {
2654 			    /*
2655 			     * nfsrv_delegconflict() unlocks state
2656 			     * when it returns non-zero.
2657 			     */
2658 			    if (ret == -1)
2659 				goto tryagain;
2660 			    error = ret;
2661 			    goto out;
2662 			}
2663 		}
2664 		stp = nstp;
2665 	    }
2666 	}
2667 	NFSUNLOCKSTATE();
2668 	if (haslock) {
2669 		NFSLOCKV4ROOTMUTEX();
2670 		nfsv4_unlock(&nfsv4rootfs_lock, 1);
2671 		NFSUNLOCKV4ROOTMUTEX();
2672 	}
2673 
2674 out:
2675 	NFSEXITCODE2(error, nd);
2676 	return (error);
2677 }
2678 
2679 /*
2680  * Open control function to create/update open state for an open.
2681  */
2682 int
2683 nfsrv_openctrl(struct nfsrv_descript *nd, vnode_t vp,
2684     struct nfsstate **new_stpp, nfsquad_t clientid, nfsv4stateid_t *stateidp,
2685     nfsv4stateid_t *delegstateidp, u_int32_t *rflagsp, struct nfsexstuff *exp,
2686     NFSPROC_T *p, u_quad_t filerev)
2687 {
2688 	struct nfsstate *new_stp = *new_stpp;
2689 	struct nfsstate *stp, *nstp;
2690 	struct nfsstate *openstp = NULL, *new_open, *ownerstp, *new_deleg;
2691 	struct nfslockfile *lfp, *new_lfp;
2692 	struct nfsclient *clp;
2693 	int error = 0, haslock = 0, ret, delegate = 1, writedeleg = 1;
2694 	int readonly = 0, cbret = 1, getfhret = 0;
2695 	int gotstate = 0, len = 0;
2696 	u_char *clidp = NULL;
2697 
2698 	if ((new_stp->ls_flags & NFSLCK_SHAREBITS) == NFSLCK_READACCESS)
2699 		readonly = 1;
2700 	/*
2701 	 * Check for restart conditions (client and server).
2702 	 * (Paranoia, should have been detected by nfsrv_opencheck().)
2703 	 * If an error does show up, return NFSERR_EXPIRED, since the
2704 	 * the seqid# has already been incremented.
2705 	 */
2706 	error = nfsrv_checkrestart(clientid, new_stp->ls_flags,
2707 	    &new_stp->ls_stateid, 0);
2708 	if (error) {
2709 		printf("Nfsd: openctrl unexpected restart err=%d\n",
2710 		    error);
2711 		error = NFSERR_EXPIRED;
2712 		goto out;
2713 	}
2714 
2715 	clidp = malloc(NFSV4_OPAQUELIMIT, M_TEMP, M_WAITOK);
2716 tryagain:
2717 	new_lfp = malloc(sizeof (struct nfslockfile),
2718 	    M_NFSDLOCKFILE, M_WAITOK);
2719 	new_open = malloc(sizeof (struct nfsstate),
2720 	    M_NFSDSTATE, M_WAITOK);
2721 	new_deleg = malloc(sizeof (struct nfsstate),
2722 	    M_NFSDSTATE, M_WAITOK);
2723 	getfhret = nfsrv_getlockfh(vp, new_stp->ls_flags, new_lfp,
2724 	    NULL, p);
2725 	NFSLOCKSTATE();
2726 	/*
2727 	 * Get the client structure. Since the linked lists could be changed
2728 	 * by other nfsd processes if this process does a tsleep(), one of
2729 	 * two things must be done.
2730 	 * 1 - don't tsleep()
2731 	 * or
2732 	 * 2 - get the nfsv4_lock() { indicated by haslock == 1 }
2733 	 *     before using the lists, since this lock stops the other
2734 	 *     nfsd. This should only be used for rare cases, since it
2735 	 *     essentially single threads the nfsd.
2736 	 *     At this time, it is only done for cases where the stable
2737 	 *     storage file must be written prior to completion of state
2738 	 *     expiration.
2739 	 */
2740 	error = nfsrv_getclient(clientid, CLOPS_RENEW, &clp, NULL,
2741 	    (nfsquad_t)((u_quad_t)0), 0, nd, p);
2742 	if (!error && (clp->lc_flags & LCL_NEEDSCBNULL) &&
2743 	    clp->lc_program) {
2744 		/*
2745 		 * This happens on the first open for a client
2746 		 * that supports callbacks.
2747 		 */
2748 		NFSUNLOCKSTATE();
2749 		/*
2750 		 * Although nfsrv_docallback() will sleep, clp won't
2751 		 * go away, since they are only removed when the
2752 		 * nfsv4_lock() has blocked the nfsd threads. The
2753 		 * fields in clp can change, but having multiple
2754 		 * threads do this Null callback RPC should be
2755 		 * harmless.
2756 		 */
2757 		cbret = nfsrv_docallback(clp, NFSV4PROC_CBNULL,
2758 		    NULL, 0, NULL, NULL, NULL, 0, p);
2759 		NFSLOCKSTATE();
2760 		clp->lc_flags &= ~LCL_NEEDSCBNULL;
2761 		if (!cbret)
2762 			clp->lc_flags |= LCL_CALLBACKSON;
2763 	}
2764 
2765 	/*
2766 	 * Look up the open owner. See if it needs confirmation and
2767 	 * check the seq#, as required.
2768 	 */
2769 	if (!error)
2770 		nfsrv_getowner(&clp->lc_open, new_stp, &ownerstp);
2771 
2772 	if (error) {
2773 		NFSUNLOCKSTATE();
2774 		printf("Nfsd: openctrl unexpected state err=%d\n",
2775 			error);
2776 		free(new_lfp, M_NFSDLOCKFILE);
2777 		free(new_open, M_NFSDSTATE);
2778 		free(new_deleg, M_NFSDSTATE);
2779 		if (haslock) {
2780 			NFSLOCKV4ROOTMUTEX();
2781 			nfsv4_unlock(&nfsv4rootfs_lock, 1);
2782 			NFSUNLOCKV4ROOTMUTEX();
2783 		}
2784 		error = NFSERR_EXPIRED;
2785 		goto out;
2786 	}
2787 
2788 	if (new_stp->ls_flags & NFSLCK_RECLAIM)
2789 		nfsrv_markstable(clp);
2790 
2791 	/*
2792 	 * Get the structure for the underlying file.
2793 	 */
2794 	if (getfhret)
2795 		error = getfhret;
2796 	else
2797 		error = nfsrv_getlockfile(new_stp->ls_flags, &new_lfp, &lfp,
2798 		    NULL, 0);
2799 	if (new_lfp)
2800 		free(new_lfp, M_NFSDLOCKFILE);
2801 	if (error) {
2802 		NFSUNLOCKSTATE();
2803 		printf("Nfsd openctrl unexpected getlockfile err=%d\n",
2804 		    error);
2805 		free(new_open, M_NFSDSTATE);
2806 		free(new_deleg, M_NFSDSTATE);
2807 		if (haslock) {
2808 			NFSLOCKV4ROOTMUTEX();
2809 			nfsv4_unlock(&nfsv4rootfs_lock, 1);
2810 			NFSUNLOCKV4ROOTMUTEX();
2811 		}
2812 		goto out;
2813 	}
2814 
2815 	/*
2816 	 * Search for a conflicting open/share.
2817 	 */
2818 	if (new_stp->ls_flags & NFSLCK_DELEGCUR) {
2819 	    /*
2820 	     * For Delegate_Cur, search for the matching Delegation,
2821 	     * which indicates no conflict.
2822 	     * An old delegation should have been recovered by the
2823 	     * client doing a Claim_DELEGATE_Prev, so I won't let
2824 	     * it match and return NFSERR_EXPIRED. Should I let it
2825 	     * match?
2826 	     */
2827 	    LIST_FOREACH(stp, &lfp->lf_deleg, ls_file) {
2828 		if (!(stp->ls_flags & NFSLCK_OLDDELEG) &&
2829 		    (((nd->nd_flag & ND_NFSV41) != 0 &&
2830 		    stateidp->seqid == 0) ||
2831 		    stateidp->seqid == stp->ls_stateid.seqid) &&
2832 		    !NFSBCMP(stateidp->other, stp->ls_stateid.other,
2833 			NFSX_STATEIDOTHER))
2834 			break;
2835 	    }
2836 	    if (stp == LIST_END(&lfp->lf_deleg) ||
2837 		((new_stp->ls_flags & NFSLCK_WRITEACCESS) &&
2838 		 (stp->ls_flags & NFSLCK_DELEGREAD))) {
2839 		NFSUNLOCKSTATE();
2840 		printf("Nfsd openctrl unexpected expiry\n");
2841 		free(new_open, M_NFSDSTATE);
2842 		free(new_deleg, M_NFSDSTATE);
2843 		if (haslock) {
2844 			NFSLOCKV4ROOTMUTEX();
2845 			nfsv4_unlock(&nfsv4rootfs_lock, 1);
2846 			NFSUNLOCKV4ROOTMUTEX();
2847 		}
2848 		error = NFSERR_EXPIRED;
2849 		goto out;
2850 	    }
2851 
2852 	    /*
2853 	     * Don't issue a Delegation, since one already exists and
2854 	     * delay delegation timeout, as required.
2855 	     */
2856 	    delegate = 0;
2857 	    nfsrv_delaydelegtimeout(stp);
2858 	}
2859 
2860 	/*
2861 	 * Check for access/deny bit conflicts. I also check for the
2862 	 * same owner, since the client might not have bothered to check.
2863 	 * Also, note an open for the same file and owner, if found,
2864 	 * which is all we do here for Delegate_Cur, since conflict
2865 	 * checking is already done.
2866 	 */
2867 	LIST_FOREACH(stp, &lfp->lf_open, ls_file) {
2868 		if (ownerstp && stp->ls_openowner == ownerstp)
2869 			openstp = stp;
2870 		if (!(new_stp->ls_flags & NFSLCK_DELEGCUR)) {
2871 		    /*
2872 		     * If another client has the file open, the only
2873 		     * delegation that can be issued is a Read delegation
2874 		     * and only if it is a Read open with Deny none.
2875 		     */
2876 		    if (clp != stp->ls_clp) {
2877 			if ((stp->ls_flags & NFSLCK_SHAREBITS) ==
2878 			    NFSLCK_READACCESS)
2879 			    writedeleg = 0;
2880 			else
2881 			    delegate = 0;
2882 		    }
2883 		    if(((new_stp->ls_flags & NFSLCK_ACCESSBITS) &
2884 		        ((stp->ls_flags>>NFSLCK_SHIFT) & NFSLCK_ACCESSBITS))||
2885 		       ((stp->ls_flags & NFSLCK_ACCESSBITS) &
2886 		        ((new_stp->ls_flags>>NFSLCK_SHIFT)&NFSLCK_ACCESSBITS))){
2887 			ret = nfsrv_clientconflict(stp->ls_clp,&haslock,vp,p);
2888 			if (ret == 1) {
2889 				/*
2890 				 * nfsrv_clientconflict() unlocks state
2891 				 * when it returns non-zero.
2892 				 */
2893 				free(new_open, M_NFSDSTATE);
2894 				free(new_deleg, M_NFSDSTATE);
2895 				openstp = NULL;
2896 				goto tryagain;
2897 			}
2898 			if (ret == 2)
2899 				error = NFSERR_PERM;
2900 			else if (new_stp->ls_flags & NFSLCK_RECLAIM)
2901 				error = NFSERR_RECLAIMCONFLICT;
2902 			else
2903 				error = NFSERR_SHAREDENIED;
2904 			if (ret == 0)
2905 				NFSUNLOCKSTATE();
2906 			if (haslock) {
2907 				NFSLOCKV4ROOTMUTEX();
2908 				nfsv4_unlock(&nfsv4rootfs_lock, 1);
2909 				NFSUNLOCKV4ROOTMUTEX();
2910 			}
2911 			free(new_open, M_NFSDSTATE);
2912 			free(new_deleg, M_NFSDSTATE);
2913 			printf("nfsd openctrl unexpected client cnfl\n");
2914 			goto out;
2915 		    }
2916 		}
2917 	}
2918 
2919 	/*
2920 	 * Check for a conflicting delegation. If one is found, call
2921 	 * nfsrv_delegconflict() to handle it. If the v4root lock hasn't
2922 	 * been set yet, it will get the lock. Otherwise, it will recall
2923 	 * the delegation. Then, we try try again...
2924 	 * (If NFSLCK_DELEGCUR is set, it has a delegation, so there
2925 	 *  isn't a conflict.)
2926 	 * I currently believe the conflict algorithm to be:
2927 	 * For Open with Read Access and Deny None
2928 	 * - there is a conflict iff a different client has a write delegation
2929 	 * For Open with other Write Access or any Deny except None
2930 	 * - there is a conflict if a different client has any delegation
2931 	 * - there is a conflict if the same client has a read delegation
2932 	 *   (The current consensus is that this last case should be
2933 	 *    considered a conflict since the client with a read delegation
2934 	 *    could have done an Open with ReadAccess and WriteDeny
2935 	 *    locally and then not have checked for the WriteDeny.)
2936 	 */
2937 	if (!(new_stp->ls_flags & (NFSLCK_DELEGPREV | NFSLCK_DELEGCUR))) {
2938 	    stp = LIST_FIRST(&lfp->lf_deleg);
2939 	    while (stp != LIST_END(&lfp->lf_deleg)) {
2940 		nstp = LIST_NEXT(stp, ls_file);
2941 		if (stp->ls_clp != clp && (stp->ls_flags & NFSLCK_DELEGREAD))
2942 			writedeleg = 0;
2943 		else
2944 			delegate = 0;
2945 		if ((readonly && stp->ls_clp != clp &&
2946 		       (stp->ls_flags & NFSLCK_DELEGWRITE)) ||
2947 		    (!readonly && (stp->ls_clp != clp ||
2948 		         (stp->ls_flags & NFSLCK_DELEGREAD)))) {
2949 		    if (new_stp->ls_flags & NFSLCK_RECLAIM) {
2950 			delegate = 2;
2951 		    } else {
2952 			ret = nfsrv_delegconflict(stp, &haslock, p, vp);
2953 			if (ret) {
2954 			    /*
2955 			     * nfsrv_delegconflict() unlocks state
2956 			     * when it returns non-zero.
2957 			     */
2958 			    printf("Nfsd openctrl unexpected deleg cnfl\n");
2959 			    free(new_open, M_NFSDSTATE);
2960 			    free(new_deleg, M_NFSDSTATE);
2961 			    if (ret == -1) {
2962 				openstp = NULL;
2963 				goto tryagain;
2964 			    }
2965 			    error = ret;
2966 			    goto out;
2967 			}
2968 		    }
2969 		}
2970 		stp = nstp;
2971 	    }
2972 	}
2973 
2974 	/*
2975 	 * We only get here if there was no open that conflicted.
2976 	 * If an open for the owner exists, or in the access/deny bits.
2977 	 * Otherwise it is a new open. If the open_owner hasn't been
2978 	 * confirmed, replace the open with the new one needing confirmation,
2979 	 * otherwise add the open.
2980 	 */
2981 	if (new_stp->ls_flags & NFSLCK_DELEGPREV) {
2982 	    /*
2983 	     * Handle NFSLCK_DELEGPREV by searching the old delegations for
2984 	     * a match. If found, just move the old delegation to the current
2985 	     * delegation list and issue open. If not found, return
2986 	     * NFSERR_EXPIRED.
2987 	     */
2988 	    LIST_FOREACH(stp, &clp->lc_olddeleg, ls_list) {
2989 		if (stp->ls_lfp == lfp) {
2990 		    /* Found it */
2991 		    if (stp->ls_clp != clp)
2992 			panic("olddeleg clp");
2993 		    LIST_REMOVE(stp, ls_list);
2994 		    LIST_REMOVE(stp, ls_hash);
2995 		    stp->ls_flags &= ~NFSLCK_OLDDELEG;
2996 		    stp->ls_stateid.seqid = delegstateidp->seqid = 1;
2997 		    stp->ls_stateid.other[0] = delegstateidp->other[0] =
2998 			clp->lc_clientid.lval[0];
2999 		    stp->ls_stateid.other[1] = delegstateidp->other[1] =
3000 			clp->lc_clientid.lval[1];
3001 		    stp->ls_stateid.other[2] = delegstateidp->other[2] =
3002 			nfsrv_nextstateindex(clp);
3003 		    stp->ls_compref = nd->nd_compref;
3004 		    LIST_INSERT_HEAD(&clp->lc_deleg, stp, ls_list);
3005 		    LIST_INSERT_HEAD(NFSSTATEHASH(clp,
3006 			stp->ls_stateid), stp, ls_hash);
3007 		    if (stp->ls_flags & NFSLCK_DELEGWRITE)
3008 			*rflagsp |= NFSV4OPEN_WRITEDELEGATE;
3009 		    else
3010 			*rflagsp |= NFSV4OPEN_READDELEGATE;
3011 		    clp->lc_delegtime = NFSD_MONOSEC +
3012 			nfsrv_lease + NFSRV_LEASEDELTA;
3013 
3014 		    /*
3015 		     * Now, do the associated open.
3016 		     */
3017 		    new_open->ls_stateid.seqid = 1;
3018 		    new_open->ls_stateid.other[0] = clp->lc_clientid.lval[0];
3019 		    new_open->ls_stateid.other[1] = clp->lc_clientid.lval[1];
3020 		    new_open->ls_stateid.other[2] = nfsrv_nextstateindex(clp);
3021 		    new_open->ls_flags = (new_stp->ls_flags&NFSLCK_DENYBITS)|
3022 			NFSLCK_OPEN;
3023 		    if (stp->ls_flags & NFSLCK_DELEGWRITE)
3024 			new_open->ls_flags |= (NFSLCK_READACCESS |
3025 			    NFSLCK_WRITEACCESS);
3026 		    else
3027 			new_open->ls_flags |= NFSLCK_READACCESS;
3028 		    new_open->ls_uid = new_stp->ls_uid;
3029 		    new_open->ls_lfp = lfp;
3030 		    new_open->ls_clp = clp;
3031 		    LIST_INIT(&new_open->ls_open);
3032 		    LIST_INSERT_HEAD(&lfp->lf_open, new_open, ls_file);
3033 		    LIST_INSERT_HEAD(NFSSTATEHASH(clp, new_open->ls_stateid),
3034 			new_open, ls_hash);
3035 		    /*
3036 		     * and handle the open owner
3037 		     */
3038 		    if (ownerstp) {
3039 			new_open->ls_openowner = ownerstp;
3040 			LIST_INSERT_HEAD(&ownerstp->ls_open,new_open,ls_list);
3041 		    } else {
3042 			new_open->ls_openowner = new_stp;
3043 			new_stp->ls_flags = 0;
3044 			nfsrvd_refcache(new_stp->ls_op);
3045 			new_stp->ls_noopens = 0;
3046 			LIST_INIT(&new_stp->ls_open);
3047 			LIST_INSERT_HEAD(&new_stp->ls_open, new_open, ls_list);
3048 			LIST_INSERT_HEAD(&clp->lc_open, new_stp, ls_list);
3049 			*new_stpp = NULL;
3050 			NFSD_VNET(nfsstatsv1_p)->srvopenowners++;
3051 			nfsrv_openpluslock++;
3052 		    }
3053 		    openstp = new_open;
3054 		    new_open = NULL;
3055 		    NFSD_VNET(nfsstatsv1_p)->srvopens++;
3056 		    nfsrv_openpluslock++;
3057 		    break;
3058 		}
3059 	    }
3060 	    if (stp == LIST_END(&clp->lc_olddeleg))
3061 		error = NFSERR_EXPIRED;
3062 	} else if (new_stp->ls_flags & (NFSLCK_DELEGREAD | NFSLCK_DELEGWRITE)) {
3063 	    /*
3064 	     * Scan to see that no delegation for this client and file
3065 	     * doesn't already exist.
3066 	     * There also shouldn't yet be an Open for this file and
3067 	     * openowner.
3068 	     */
3069 	    LIST_FOREACH(stp, &lfp->lf_deleg, ls_file) {
3070 		if (stp->ls_clp == clp)
3071 		    break;
3072 	    }
3073 	    if (stp == LIST_END(&lfp->lf_deleg) && openstp == NULL) {
3074 		/*
3075 		 * This is the Claim_Previous case with a delegation
3076 		 * type != Delegate_None.
3077 		 */
3078 		/*
3079 		 * First, add the delegation. (Although we must issue the
3080 		 * delegation, we can also ask for an immediate return.)
3081 		 */
3082 		new_deleg->ls_stateid.seqid = delegstateidp->seqid = 1;
3083 		new_deleg->ls_stateid.other[0] = delegstateidp->other[0] =
3084 		    clp->lc_clientid.lval[0];
3085 		new_deleg->ls_stateid.other[1] = delegstateidp->other[1] =
3086 		    clp->lc_clientid.lval[1];
3087 		new_deleg->ls_stateid.other[2] = delegstateidp->other[2] =
3088 		    nfsrv_nextstateindex(clp);
3089 		if (new_stp->ls_flags & NFSLCK_DELEGWRITE) {
3090 		    new_deleg->ls_flags = (NFSLCK_DELEGWRITE |
3091 			NFSLCK_READACCESS | NFSLCK_WRITEACCESS);
3092 		    *rflagsp |= NFSV4OPEN_WRITEDELEGATE;
3093 		    nfsrv_writedelegcnt++;
3094 		} else {
3095 		    new_deleg->ls_flags = (NFSLCK_DELEGREAD |
3096 			NFSLCK_READACCESS);
3097 		    *rflagsp |= NFSV4OPEN_READDELEGATE;
3098 		}
3099 		new_deleg->ls_uid = new_stp->ls_uid;
3100 		new_deleg->ls_lfp = lfp;
3101 		new_deleg->ls_clp = clp;
3102 		new_deleg->ls_filerev = filerev;
3103 		new_deleg->ls_compref = nd->nd_compref;
3104 		new_deleg->ls_lastrecall = 0;
3105 		LIST_INSERT_HEAD(&lfp->lf_deleg, new_deleg, ls_file);
3106 		LIST_INSERT_HEAD(NFSSTATEHASH(clp,
3107 		    new_deleg->ls_stateid), new_deleg, ls_hash);
3108 		LIST_INSERT_HEAD(&clp->lc_deleg, new_deleg, ls_list);
3109 		new_deleg = NULL;
3110 		if (delegate == 2 || nfsrv_issuedelegs == 0 ||
3111 		    (clp->lc_flags & (LCL_CALLBACKSON | LCL_CBDOWN)) !=
3112 		     LCL_CALLBACKSON ||
3113 		    NFSRV_V4DELEGLIMIT(nfsrv_delegatecnt) ||
3114 		    !NFSVNO_DELEGOK(vp))
3115 		    *rflagsp |= NFSV4OPEN_RECALL;
3116 		NFSD_VNET(nfsstatsv1_p)->srvdelegates++;
3117 		nfsrv_openpluslock++;
3118 		nfsrv_delegatecnt++;
3119 
3120 		/*
3121 		 * Now, do the associated open.
3122 		 */
3123 		new_open->ls_stateid.seqid = 1;
3124 		new_open->ls_stateid.other[0] = clp->lc_clientid.lval[0];
3125 		new_open->ls_stateid.other[1] = clp->lc_clientid.lval[1];
3126 		new_open->ls_stateid.other[2] = nfsrv_nextstateindex(clp);
3127 		new_open->ls_flags = (new_stp->ls_flags & NFSLCK_DENYBITS) |
3128 		    NFSLCK_OPEN;
3129 		if (new_stp->ls_flags & NFSLCK_DELEGWRITE)
3130 			new_open->ls_flags |= (NFSLCK_READACCESS |
3131 			    NFSLCK_WRITEACCESS);
3132 		else
3133 			new_open->ls_flags |= NFSLCK_READACCESS;
3134 		new_open->ls_uid = new_stp->ls_uid;
3135 		new_open->ls_lfp = lfp;
3136 		new_open->ls_clp = clp;
3137 		LIST_INIT(&new_open->ls_open);
3138 		LIST_INSERT_HEAD(&lfp->lf_open, new_open, ls_file);
3139 		LIST_INSERT_HEAD(NFSSTATEHASH(clp, new_open->ls_stateid),
3140 		   new_open, ls_hash);
3141 		/*
3142 		 * and handle the open owner
3143 		 */
3144 		if (ownerstp) {
3145 		    new_open->ls_openowner = ownerstp;
3146 		    LIST_INSERT_HEAD(&ownerstp->ls_open, new_open, ls_list);
3147 		} else {
3148 		    new_open->ls_openowner = new_stp;
3149 		    new_stp->ls_flags = 0;
3150 		    nfsrvd_refcache(new_stp->ls_op);
3151 		    new_stp->ls_noopens = 0;
3152 		    LIST_INIT(&new_stp->ls_open);
3153 		    LIST_INSERT_HEAD(&new_stp->ls_open, new_open, ls_list);
3154 		    LIST_INSERT_HEAD(&clp->lc_open, new_stp, ls_list);
3155 		    *new_stpp = NULL;
3156 		    NFSD_VNET(nfsstatsv1_p)->srvopenowners++;
3157 		    nfsrv_openpluslock++;
3158 		}
3159 		openstp = new_open;
3160 		new_open = NULL;
3161 		NFSD_VNET(nfsstatsv1_p)->srvopens++;
3162 		nfsrv_openpluslock++;
3163 	    } else {
3164 		error = NFSERR_RECLAIMCONFLICT;
3165 	    }
3166 	} else if (ownerstp) {
3167 		if (ownerstp->ls_flags & NFSLCK_NEEDSCONFIRM) {
3168 		    /* Replace the open */
3169 		    if (ownerstp->ls_op)
3170 			nfsrvd_derefcache(ownerstp->ls_op);
3171 		    ownerstp->ls_op = new_stp->ls_op;
3172 		    nfsrvd_refcache(ownerstp->ls_op);
3173 		    ownerstp->ls_seq = new_stp->ls_seq;
3174 		    *rflagsp |= NFSV4OPEN_RESULTCONFIRM;
3175 		    stp = LIST_FIRST(&ownerstp->ls_open);
3176 		    stp->ls_flags = (new_stp->ls_flags & NFSLCK_SHAREBITS) |
3177 			NFSLCK_OPEN;
3178 		    stp->ls_stateid.seqid = 1;
3179 		    stp->ls_uid = new_stp->ls_uid;
3180 		    if (lfp != stp->ls_lfp) {
3181 			LIST_REMOVE(stp, ls_file);
3182 			LIST_INSERT_HEAD(&lfp->lf_open, stp, ls_file);
3183 			stp->ls_lfp = lfp;
3184 		    }
3185 		    openstp = stp;
3186 		} else if (openstp) {
3187 		    openstp->ls_flags |= (new_stp->ls_flags & NFSLCK_SHAREBITS);
3188 		    openstp->ls_stateid.seqid++;
3189 		    if ((nd->nd_flag & ND_NFSV41) != 0 &&
3190 			openstp->ls_stateid.seqid == 0)
3191 			openstp->ls_stateid.seqid = 1;
3192 
3193 		    /*
3194 		     * This is where we can choose to issue a delegation.
3195 		     */
3196 		    if ((new_stp->ls_flags & NFSLCK_WANTNODELEG) != 0)
3197 			*rflagsp |= NFSV4OPEN_WDNOTWANTED;
3198 		    else if (nfsrv_issuedelegs == 0)
3199 			*rflagsp |= NFSV4OPEN_WDSUPPFTYPE;
3200 		    else if (NFSRV_V4DELEGLIMIT(nfsrv_delegatecnt))
3201 			*rflagsp |= NFSV4OPEN_WDRESOURCE;
3202 		    else if (delegate == 0 || writedeleg == 0 ||
3203 			NFSVNO_EXRDONLY(exp) || (readonly != 0 &&
3204 			nfsrv_writedelegifpos == 0) ||
3205 			!NFSVNO_DELEGOK(vp) ||
3206 			(new_stp->ls_flags & NFSLCK_WANTRDELEG) != 0 ||
3207 			(clp->lc_flags & (LCL_CALLBACKSON | LCL_CBDOWN)) !=
3208 			 LCL_CALLBACKSON)
3209 			*rflagsp |= NFSV4OPEN_WDCONTENTION;
3210 		    else {
3211 			new_deleg->ls_stateid.seqid = delegstateidp->seqid = 1;
3212 			new_deleg->ls_stateid.other[0] = delegstateidp->other[0]
3213 			    = clp->lc_clientid.lval[0];
3214 			new_deleg->ls_stateid.other[1] = delegstateidp->other[1]
3215 			    = clp->lc_clientid.lval[1];
3216 			new_deleg->ls_stateid.other[2] = delegstateidp->other[2]
3217 			    = nfsrv_nextstateindex(clp);
3218 			new_deleg->ls_flags = (NFSLCK_DELEGWRITE |
3219 			    NFSLCK_READACCESS | NFSLCK_WRITEACCESS);
3220 			*rflagsp |= NFSV4OPEN_WRITEDELEGATE;
3221 			new_deleg->ls_uid = new_stp->ls_uid;
3222 			new_deleg->ls_lfp = lfp;
3223 			new_deleg->ls_clp = clp;
3224 			new_deleg->ls_filerev = filerev;
3225 			new_deleg->ls_compref = nd->nd_compref;
3226 			new_deleg->ls_lastrecall = 0;
3227 			nfsrv_writedelegcnt++;
3228 			LIST_INSERT_HEAD(&lfp->lf_deleg, new_deleg, ls_file);
3229 			LIST_INSERT_HEAD(NFSSTATEHASH(clp,
3230 			    new_deleg->ls_stateid), new_deleg, ls_hash);
3231 			LIST_INSERT_HEAD(&clp->lc_deleg, new_deleg, ls_list);
3232 			new_deleg = NULL;
3233 			NFSD_VNET(nfsstatsv1_p)->srvdelegates++;
3234 			nfsrv_openpluslock++;
3235 			nfsrv_delegatecnt++;
3236 		    }
3237 		} else {
3238 		    new_open->ls_stateid.seqid = 1;
3239 		    new_open->ls_stateid.other[0] = clp->lc_clientid.lval[0];
3240 		    new_open->ls_stateid.other[1] = clp->lc_clientid.lval[1];
3241 		    new_open->ls_stateid.other[2] = nfsrv_nextstateindex(clp);
3242 		    new_open->ls_flags = (new_stp->ls_flags & NFSLCK_SHAREBITS)|
3243 			NFSLCK_OPEN;
3244 		    new_open->ls_uid = new_stp->ls_uid;
3245 		    new_open->ls_openowner = ownerstp;
3246 		    new_open->ls_lfp = lfp;
3247 		    new_open->ls_clp = clp;
3248 		    LIST_INIT(&new_open->ls_open);
3249 		    LIST_INSERT_HEAD(&lfp->lf_open, new_open, ls_file);
3250 		    LIST_INSERT_HEAD(&ownerstp->ls_open, new_open, ls_list);
3251 		    LIST_INSERT_HEAD(NFSSTATEHASH(clp, new_open->ls_stateid),
3252 			new_open, ls_hash);
3253 		    openstp = new_open;
3254 		    new_open = NULL;
3255 		    NFSD_VNET(nfsstatsv1_p)->srvopens++;
3256 		    nfsrv_openpluslock++;
3257 
3258 		    /*
3259 		     * This is where we can choose to issue a delegation.
3260 		     */
3261 		    if ((new_stp->ls_flags & NFSLCK_WANTNODELEG) != 0)
3262 			*rflagsp |= NFSV4OPEN_WDNOTWANTED;
3263 		    else if (nfsrv_issuedelegs == 0)
3264 			*rflagsp |= NFSV4OPEN_WDSUPPFTYPE;
3265 		    else if (NFSRV_V4DELEGLIMIT(nfsrv_delegatecnt))
3266 			*rflagsp |= NFSV4OPEN_WDRESOURCE;
3267 		    else if (delegate == 0 || (writedeleg == 0 &&
3268 			readonly == 0) || !NFSVNO_DELEGOK(vp) ||
3269 			(clp->lc_flags & (LCL_CALLBACKSON | LCL_CBDOWN)) !=
3270 			 LCL_CALLBACKSON)
3271 			*rflagsp |= NFSV4OPEN_WDCONTENTION;
3272 		    else {
3273 			new_deleg->ls_stateid.seqid = delegstateidp->seqid = 1;
3274 			new_deleg->ls_stateid.other[0] = delegstateidp->other[0]
3275 			    = clp->lc_clientid.lval[0];
3276 			new_deleg->ls_stateid.other[1] = delegstateidp->other[1]
3277 			    = clp->lc_clientid.lval[1];
3278 			new_deleg->ls_stateid.other[2] = delegstateidp->other[2]
3279 			    = nfsrv_nextstateindex(clp);
3280 			if (writedeleg && !NFSVNO_EXRDONLY(exp) &&
3281 			    (nfsrv_writedelegifpos || !readonly) &&
3282 			    (new_stp->ls_flags & NFSLCK_WANTRDELEG) == 0) {
3283 			    new_deleg->ls_flags = (NFSLCK_DELEGWRITE |
3284 				NFSLCK_READACCESS | NFSLCK_WRITEACCESS);
3285 			    *rflagsp |= NFSV4OPEN_WRITEDELEGATE;
3286 			    nfsrv_writedelegcnt++;
3287 			} else {
3288 			    new_deleg->ls_flags = (NFSLCK_DELEGREAD |
3289 				NFSLCK_READACCESS);
3290 			    *rflagsp |= NFSV4OPEN_READDELEGATE;
3291 			}
3292 			new_deleg->ls_uid = new_stp->ls_uid;
3293 			new_deleg->ls_lfp = lfp;
3294 			new_deleg->ls_clp = clp;
3295 			new_deleg->ls_filerev = filerev;
3296 			new_deleg->ls_compref = nd->nd_compref;
3297 			new_deleg->ls_lastrecall = 0;
3298 			LIST_INSERT_HEAD(&lfp->lf_deleg, new_deleg, ls_file);
3299 			LIST_INSERT_HEAD(NFSSTATEHASH(clp,
3300 			    new_deleg->ls_stateid), new_deleg, ls_hash);
3301 			LIST_INSERT_HEAD(&clp->lc_deleg, new_deleg, ls_list);
3302 			new_deleg = NULL;
3303 			NFSD_VNET(nfsstatsv1_p)->srvdelegates++;
3304 			nfsrv_openpluslock++;
3305 			nfsrv_delegatecnt++;
3306 		    }
3307 		}
3308 	} else {
3309 		/*
3310 		 * New owner case. Start the open_owner sequence with a
3311 		 * Needs confirmation (unless a reclaim) and hang the
3312 		 * new open off it.
3313 		 */
3314 		new_open->ls_stateid.seqid = 1;
3315 		new_open->ls_stateid.other[0] = clp->lc_clientid.lval[0];
3316 		new_open->ls_stateid.other[1] = clp->lc_clientid.lval[1];
3317 		new_open->ls_stateid.other[2] = nfsrv_nextstateindex(clp);
3318 		new_open->ls_flags = (new_stp->ls_flags & NFSLCK_SHAREBITS) |
3319 		    NFSLCK_OPEN;
3320 		new_open->ls_uid = new_stp->ls_uid;
3321 		LIST_INIT(&new_open->ls_open);
3322 		new_open->ls_openowner = new_stp;
3323 		new_open->ls_lfp = lfp;
3324 		new_open->ls_clp = clp;
3325 		LIST_INSERT_HEAD(&lfp->lf_open, new_open, ls_file);
3326 		if (new_stp->ls_flags & NFSLCK_RECLAIM) {
3327 			new_stp->ls_flags = 0;
3328 		} else if ((nd->nd_flag & ND_NFSV41) != 0) {
3329 			/* NFSv4.1 never needs confirmation. */
3330 			new_stp->ls_flags = 0;
3331 
3332 			/*
3333 			 * This is where we can choose to issue a delegation.
3334 			 */
3335 			if (delegate && nfsrv_issuedelegs &&
3336 			    (writedeleg || readonly) &&
3337 			    (clp->lc_flags & (LCL_CALLBACKSON | LCL_CBDOWN)) ==
3338 			     LCL_CALLBACKSON &&
3339 			    !NFSRV_V4DELEGLIMIT(nfsrv_delegatecnt) &&
3340 			    NFSVNO_DELEGOK(vp) &&
3341 			    ((nd->nd_flag & ND_NFSV41) == 0 ||
3342 			     (new_stp->ls_flags & NFSLCK_WANTNODELEG) == 0)) {
3343 				new_deleg->ls_stateid.seqid =
3344 				    delegstateidp->seqid = 1;
3345 				new_deleg->ls_stateid.other[0] =
3346 				    delegstateidp->other[0]
3347 				    = clp->lc_clientid.lval[0];
3348 				new_deleg->ls_stateid.other[1] =
3349 				    delegstateidp->other[1]
3350 				    = clp->lc_clientid.lval[1];
3351 				new_deleg->ls_stateid.other[2] =
3352 				    delegstateidp->other[2]
3353 				    = nfsrv_nextstateindex(clp);
3354 				if (writedeleg && !NFSVNO_EXRDONLY(exp) &&
3355 				    (nfsrv_writedelegifpos || !readonly) &&
3356 				    ((nd->nd_flag & ND_NFSV41) == 0 ||
3357 				     (new_stp->ls_flags & NFSLCK_WANTRDELEG) ==
3358 				     0)) {
3359 					new_deleg->ls_flags =
3360 					    (NFSLCK_DELEGWRITE |
3361 					     NFSLCK_READACCESS |
3362 					     NFSLCK_WRITEACCESS);
3363 					*rflagsp |= NFSV4OPEN_WRITEDELEGATE;
3364 					nfsrv_writedelegcnt++;
3365 				} else {
3366 					new_deleg->ls_flags =
3367 					    (NFSLCK_DELEGREAD |
3368 					     NFSLCK_READACCESS);
3369 					*rflagsp |= NFSV4OPEN_READDELEGATE;
3370 				}
3371 				new_deleg->ls_uid = new_stp->ls_uid;
3372 				new_deleg->ls_lfp = lfp;
3373 				new_deleg->ls_clp = clp;
3374 				new_deleg->ls_filerev = filerev;
3375 				new_deleg->ls_compref = nd->nd_compref;
3376 				new_deleg->ls_lastrecall = 0;
3377 				LIST_INSERT_HEAD(&lfp->lf_deleg, new_deleg,
3378 				    ls_file);
3379 				LIST_INSERT_HEAD(NFSSTATEHASH(clp,
3380 				    new_deleg->ls_stateid), new_deleg, ls_hash);
3381 				LIST_INSERT_HEAD(&clp->lc_deleg, new_deleg,
3382 				    ls_list);
3383 				new_deleg = NULL;
3384 				NFSD_VNET(nfsstatsv1_p)->srvdelegates++;
3385 				nfsrv_openpluslock++;
3386 				nfsrv_delegatecnt++;
3387 			}
3388 			/*
3389 			 * Since NFSv4.1 never does an OpenConfirm, the first
3390 			 * open state will be acquired here.
3391 			 */
3392 			if (!(clp->lc_flags & LCL_STAMPEDSTABLE)) {
3393 				clp->lc_flags |= LCL_STAMPEDSTABLE;
3394 				len = clp->lc_idlen;
3395 				NFSBCOPY(clp->lc_id, clidp, len);
3396 				gotstate = 1;
3397 			}
3398 		} else {
3399 			*rflagsp |= NFSV4OPEN_RESULTCONFIRM;
3400 			new_stp->ls_flags = NFSLCK_NEEDSCONFIRM;
3401 		}
3402 		nfsrvd_refcache(new_stp->ls_op);
3403 		new_stp->ls_noopens = 0;
3404 		LIST_INIT(&new_stp->ls_open);
3405 		LIST_INSERT_HEAD(&new_stp->ls_open, new_open, ls_list);
3406 		LIST_INSERT_HEAD(&clp->lc_open, new_stp, ls_list);
3407 		LIST_INSERT_HEAD(NFSSTATEHASH(clp, new_open->ls_stateid),
3408 		    new_open, ls_hash);
3409 		openstp = new_open;
3410 		new_open = NULL;
3411 		*new_stpp = NULL;
3412 		NFSD_VNET(nfsstatsv1_p)->srvopens++;
3413 		nfsrv_openpluslock++;
3414 		NFSD_VNET(nfsstatsv1_p)->srvopenowners++;
3415 		nfsrv_openpluslock++;
3416 	}
3417 	if (!error) {
3418 		stateidp->seqid = openstp->ls_stateid.seqid;
3419 		stateidp->other[0] = openstp->ls_stateid.other[0];
3420 		stateidp->other[1] = openstp->ls_stateid.other[1];
3421 		stateidp->other[2] = openstp->ls_stateid.other[2];
3422 	}
3423 	NFSUNLOCKSTATE();
3424 	if (haslock) {
3425 		NFSLOCKV4ROOTMUTEX();
3426 		nfsv4_unlock(&nfsv4rootfs_lock, 1);
3427 		NFSUNLOCKV4ROOTMUTEX();
3428 	}
3429 	if (new_open)
3430 		free(new_open, M_NFSDSTATE);
3431 	if (new_deleg)
3432 		free(new_deleg, M_NFSDSTATE);
3433 
3434 	/*
3435 	 * If the NFSv4.1 client just acquired its first open, write a timestamp
3436 	 * to the stable storage file.
3437 	 */
3438 	if (gotstate != 0) {
3439 		nfsrv_writestable(clidp, len, NFSNST_NEWSTATE, p);
3440 		nfsrv_backupstable();
3441 	}
3442 
3443 out:
3444 	free(clidp, M_TEMP);
3445 	NFSEXITCODE2(error, nd);
3446 	return (error);
3447 }
3448 
3449 /*
3450  * Open update. Does the confirm, downgrade and close.
3451  */
3452 int
3453 nfsrv_openupdate(vnode_t vp, struct nfsstate *new_stp, nfsquad_t clientid,
3454     nfsv4stateid_t *stateidp, struct nfsrv_descript *nd, NFSPROC_T *p,
3455     int *retwriteaccessp)
3456 {
3457 	struct nfsstate *stp;
3458 	struct nfsclient *clp;
3459 	struct nfslockfile *lfp;
3460 	u_int32_t bits;
3461 	int error = 0, gotstate = 0, len = 0;
3462 	u_char *clidp = NULL;
3463 
3464 	/*
3465 	 * Check for restart conditions (client and server).
3466 	 */
3467 	error = nfsrv_checkrestart(clientid, new_stp->ls_flags,
3468 	    &new_stp->ls_stateid, 0);
3469 	if (error)
3470 		goto out;
3471 
3472 	clidp = malloc(NFSV4_OPAQUELIMIT, M_TEMP, M_WAITOK);
3473 	NFSLOCKSTATE();
3474 	/*
3475 	 * Get the open structure via clientid and stateid.
3476 	 */
3477 	error = nfsrv_getclient(clientid, CLOPS_RENEW, &clp, NULL,
3478 	    (nfsquad_t)((u_quad_t)0), 0, nd, p);
3479 	if (!error)
3480 		error = nfsrv_getstate(clp, &new_stp->ls_stateid,
3481 		    new_stp->ls_flags, &stp);
3482 
3483 	/*
3484 	 * Sanity check the open.
3485 	 */
3486 	if (!error && (!(stp->ls_flags & NFSLCK_OPEN) ||
3487 		(!(new_stp->ls_flags & NFSLCK_CONFIRM) &&
3488 		 (stp->ls_openowner->ls_flags & NFSLCK_NEEDSCONFIRM)) ||
3489 		((new_stp->ls_flags & NFSLCK_CONFIRM) &&
3490 		 (!(stp->ls_openowner->ls_flags & NFSLCK_NEEDSCONFIRM)))))
3491 		error = NFSERR_BADSTATEID;
3492 
3493 	if (!error)
3494 		error = nfsrv_checkseqid(nd, new_stp->ls_seq,
3495 		    stp->ls_openowner, new_stp->ls_op);
3496 	if (!error && stp->ls_stateid.seqid != new_stp->ls_stateid.seqid &&
3497 	    (((nd->nd_flag & ND_NFSV41) == 0 &&
3498 	      !(new_stp->ls_flags & NFSLCK_CONFIRM)) ||
3499 	     ((nd->nd_flag & ND_NFSV41) != 0 &&
3500 	      new_stp->ls_stateid.seqid != 0)))
3501 		error = NFSERR_OLDSTATEID;
3502 	if (!error && vp->v_type != VREG) {
3503 		if (vp->v_type == VDIR)
3504 			error = NFSERR_ISDIR;
3505 		else
3506 			error = NFSERR_INVAL;
3507 	}
3508 
3509 	if (error) {
3510 		/*
3511 		 * If a client tries to confirm an Open with a bad
3512 		 * seqid# and there are no byte range locks or other Opens
3513 		 * on the openowner, just throw it away, so the next use of the
3514 		 * openowner will start a fresh seq#.
3515 		 */
3516 		if (error == NFSERR_BADSEQID &&
3517 		    (new_stp->ls_flags & NFSLCK_CONFIRM) &&
3518 		    nfsrv_nootherstate(stp))
3519 			nfsrv_freeopenowner(stp->ls_openowner, 0, p);
3520 		NFSUNLOCKSTATE();
3521 		goto out;
3522 	}
3523 
3524 	/*
3525 	 * Set the return stateid.
3526 	 */
3527 	stateidp->seqid = stp->ls_stateid.seqid + 1;
3528 	if ((nd->nd_flag & ND_NFSV41) != 0 && stateidp->seqid == 0)
3529 		stateidp->seqid = 1;
3530 	stateidp->other[0] = stp->ls_stateid.other[0];
3531 	stateidp->other[1] = stp->ls_stateid.other[1];
3532 	stateidp->other[2] = stp->ls_stateid.other[2];
3533 	/*
3534 	 * Now, handle the three cases.
3535 	 */
3536 	if (new_stp->ls_flags & NFSLCK_CONFIRM) {
3537 		/*
3538 		 * If the open doesn't need confirmation, it seems to me that
3539 		 * there is a client error, but I'll just log it and keep going?
3540 		 */
3541 		if (!(stp->ls_openowner->ls_flags & NFSLCK_NEEDSCONFIRM))
3542 			printf("Nfsv4d: stray open confirm\n");
3543 		stp->ls_openowner->ls_flags = 0;
3544 		stp->ls_stateid.seqid++;
3545 		if ((nd->nd_flag & ND_NFSV41) != 0 &&
3546 		    stp->ls_stateid.seqid == 0)
3547 			stp->ls_stateid.seqid = 1;
3548 		if (!(clp->lc_flags & LCL_STAMPEDSTABLE)) {
3549 			clp->lc_flags |= LCL_STAMPEDSTABLE;
3550 			len = clp->lc_idlen;
3551 			NFSBCOPY(clp->lc_id, clidp, len);
3552 			gotstate = 1;
3553 		}
3554 		NFSUNLOCKSTATE();
3555 	} else if (new_stp->ls_flags & NFSLCK_CLOSE) {
3556 		lfp = stp->ls_lfp;
3557 		if (retwriteaccessp != NULL) {
3558 			if ((stp->ls_flags & NFSLCK_WRITEACCESS) != 0)
3559 				*retwriteaccessp = 1;
3560 			else
3561 				*retwriteaccessp = 0;
3562 		}
3563 		if (nfsrv_dolocallocks != 0 && !LIST_EMPTY(&stp->ls_open)) {
3564 			/* Get the lf lock */
3565 			nfsrv_locklf(lfp);
3566 			NFSUNLOCKSTATE();
3567 			ASSERT_VOP_ELOCKED(vp, "nfsrv_openupdate");
3568 			NFSVOPUNLOCK(vp);
3569 			if (nfsrv_freeopen(stp, vp, 1, p) == 0) {
3570 				NFSLOCKSTATE();
3571 				nfsrv_unlocklf(lfp);
3572 				NFSUNLOCKSTATE();
3573 			}
3574 			NFSVOPLOCK(vp, LK_EXCLUSIVE | LK_RETRY);
3575 		} else {
3576 			(void) nfsrv_freeopen(stp, NULL, 0, p);
3577 			NFSUNLOCKSTATE();
3578 		}
3579 	} else {
3580 		/*
3581 		 * Update the share bits, making sure that the new set are a
3582 		 * subset of the old ones.
3583 		 */
3584 		bits = (new_stp->ls_flags & NFSLCK_SHAREBITS);
3585 		if (~(stp->ls_flags) & bits) {
3586 			NFSUNLOCKSTATE();
3587 			error = NFSERR_INVAL;
3588 			goto out;
3589 		}
3590 		stp->ls_flags = (bits | NFSLCK_OPEN);
3591 		stp->ls_stateid.seqid++;
3592 		if ((nd->nd_flag & ND_NFSV41) != 0 &&
3593 		    stp->ls_stateid.seqid == 0)
3594 			stp->ls_stateid.seqid = 1;
3595 		NFSUNLOCKSTATE();
3596 	}
3597 
3598 	/*
3599 	 * If the client just confirmed its first open, write a timestamp
3600 	 * to the stable storage file.
3601 	 */
3602 	if (gotstate != 0) {
3603 		nfsrv_writestable(clidp, len, NFSNST_NEWSTATE, p);
3604 		nfsrv_backupstable();
3605 	}
3606 
3607 out:
3608 	free(clidp, M_TEMP);
3609 	NFSEXITCODE2(error, nd);
3610 	return (error);
3611 }
3612 
3613 /*
3614  * Delegation update. Does the purge and return.
3615  */
3616 int
3617 nfsrv_delegupdate(struct nfsrv_descript *nd, nfsquad_t clientid,
3618     nfsv4stateid_t *stateidp, vnode_t vp, int op, struct ucred *cred,
3619     NFSPROC_T *p, int *retwriteaccessp)
3620 {
3621 	struct nfsstate *stp;
3622 	struct nfsclient *clp;
3623 	int error = 0;
3624 	fhandle_t fh;
3625 
3626 	/*
3627 	 * Do a sanity check against the file handle for DelegReturn.
3628 	 */
3629 	if (vp) {
3630 		error = nfsvno_getfh(vp, &fh, p);
3631 		if (error)
3632 			goto out;
3633 	}
3634 	/*
3635 	 * Check for restart conditions (client and server).
3636 	 */
3637 	if (op == NFSV4OP_DELEGRETURN)
3638 		error = nfsrv_checkrestart(clientid, NFSLCK_DELEGRETURN,
3639 			stateidp, 0);
3640 	else
3641 		error = nfsrv_checkrestart(clientid, NFSLCK_DELEGPURGE,
3642 			stateidp, 0);
3643 
3644 	NFSLOCKSTATE();
3645 	/*
3646 	 * Get the open structure via clientid and stateid.
3647 	 */
3648 	if (!error)
3649 	    error = nfsrv_getclient(clientid, CLOPS_RENEW, &clp, NULL,
3650 		(nfsquad_t)((u_quad_t)0), 0, nd, p);
3651 	if (error) {
3652 		if (error == NFSERR_CBPATHDOWN)
3653 			error = 0;
3654 		if (error == NFSERR_STALECLIENTID && op == NFSV4OP_DELEGRETURN)
3655 			error = NFSERR_STALESTATEID;
3656 	}
3657 	if (!error && op == NFSV4OP_DELEGRETURN) {
3658 	    error = nfsrv_getstate(clp, stateidp, NFSLCK_DELEGRETURN, &stp);
3659 	    if (!error && stp->ls_stateid.seqid != stateidp->seqid &&
3660 		((nd->nd_flag & ND_NFSV41) == 0 || stateidp->seqid != 0))
3661 		error = NFSERR_OLDSTATEID;
3662 	}
3663 	/*
3664 	 * NFSERR_EXPIRED means that the state has gone away,
3665 	 * so Delegations have been purged. Just return ok.
3666 	 */
3667 	if (error == NFSERR_EXPIRED && op == NFSV4OP_DELEGPURGE) {
3668 		NFSUNLOCKSTATE();
3669 		error = 0;
3670 		goto out;
3671 	}
3672 	if (error) {
3673 		NFSUNLOCKSTATE();
3674 		goto out;
3675 	}
3676 
3677 	if (op == NFSV4OP_DELEGRETURN) {
3678 		if (NFSBCMP((caddr_t)&fh, (caddr_t)&stp->ls_lfp->lf_fh,
3679 		    sizeof (fhandle_t))) {
3680 			NFSUNLOCKSTATE();
3681 			error = NFSERR_BADSTATEID;
3682 			goto out;
3683 		}
3684 		if (retwriteaccessp != NULL) {
3685 			if ((stp->ls_flags & NFSLCK_DELEGWRITE) != 0)
3686 				*retwriteaccessp = 1;
3687 			else
3688 				*retwriteaccessp = 0;
3689 		}
3690 		nfsrv_freedeleg(stp);
3691 	} else {
3692 		nfsrv_freedeleglist(&clp->lc_olddeleg);
3693 	}
3694 	NFSUNLOCKSTATE();
3695 	error = 0;
3696 
3697 out:
3698 	NFSEXITCODE(error);
3699 	return (error);
3700 }
3701 
3702 /*
3703  * Release lock owner.
3704  */
3705 int
3706 nfsrv_releaselckown(struct nfsstate *new_stp, nfsquad_t clientid,
3707     NFSPROC_T *p)
3708 {
3709 	struct nfsstate *stp, *nstp, *openstp, *ownstp;
3710 	struct nfsclient *clp;
3711 	int error = 0;
3712 
3713 	/*
3714 	 * Check for restart conditions (client and server).
3715 	 */
3716 	error = nfsrv_checkrestart(clientid, new_stp->ls_flags,
3717 	    &new_stp->ls_stateid, 0);
3718 	if (error)
3719 		goto out;
3720 
3721 	NFSLOCKSTATE();
3722 	/*
3723 	 * Get the lock owner by name.
3724 	 */
3725 	error = nfsrv_getclient(clientid, CLOPS_RENEW, &clp, NULL,
3726 	    (nfsquad_t)((u_quad_t)0), 0, NULL, p);
3727 	if (error) {
3728 		NFSUNLOCKSTATE();
3729 		goto out;
3730 	}
3731 	LIST_FOREACH(ownstp, &clp->lc_open, ls_list) {
3732 	    LIST_FOREACH(openstp, &ownstp->ls_open, ls_list) {
3733 		stp = LIST_FIRST(&openstp->ls_open);
3734 		while (stp != LIST_END(&openstp->ls_open)) {
3735 		    nstp = LIST_NEXT(stp, ls_list);
3736 		    /*
3737 		     * If the owner matches, check for locks and
3738 		     * then free or return an error.
3739 		     */
3740 		    if (stp->ls_ownerlen == new_stp->ls_ownerlen &&
3741 			!NFSBCMP(stp->ls_owner, new_stp->ls_owner,
3742 			 stp->ls_ownerlen)){
3743 			if (LIST_EMPTY(&stp->ls_lock)) {
3744 			    nfsrv_freelockowner(stp, NULL, 0, p);
3745 			} else {
3746 			    NFSUNLOCKSTATE();
3747 			    error = NFSERR_LOCKSHELD;
3748 			    goto out;
3749 			}
3750 		    }
3751 		    stp = nstp;
3752 		}
3753 	    }
3754 	}
3755 	NFSUNLOCKSTATE();
3756 
3757 out:
3758 	NFSEXITCODE(error);
3759 	return (error);
3760 }
3761 
3762 /*
3763  * Get the file handle for a lock structure.
3764  */
3765 static int
3766 nfsrv_getlockfh(vnode_t vp, u_short flags, struct nfslockfile *new_lfp,
3767     fhandle_t *nfhp, NFSPROC_T *p)
3768 {
3769 	fhandle_t *fhp = NULL;
3770 	int error;
3771 
3772 	/*
3773 	 * For lock, use the new nfslock structure, otherwise just
3774 	 * a fhandle_t on the stack.
3775 	 */
3776 	if (flags & NFSLCK_OPEN) {
3777 		KASSERT(new_lfp != NULL, ("nfsrv_getlockfh: new_lfp NULL"));
3778 		fhp = &new_lfp->lf_fh;
3779 	} else if (nfhp) {
3780 		fhp = nfhp;
3781 	} else {
3782 		panic("nfsrv_getlockfh");
3783 	}
3784 	error = nfsvno_getfh(vp, fhp, p);
3785 	NFSEXITCODE(error);
3786 	return (error);
3787 }
3788 
3789 /*
3790  * Get an nfs lock structure. Allocate one, as required, and return a
3791  * pointer to it.
3792  * Returns an NFSERR_xxx upon failure or -1 to indicate no current lock.
3793  */
3794 static int
3795 nfsrv_getlockfile(u_short flags, struct nfslockfile **new_lfpp,
3796     struct nfslockfile **lfpp, fhandle_t *nfhp, int lockit)
3797 {
3798 	struct nfslockfile *lfp;
3799 	fhandle_t *fhp = NULL, *tfhp;
3800 	struct nfslockhashhead *hp;
3801 	struct nfslockfile *new_lfp = NULL;
3802 
3803 	/*
3804 	 * For lock, use the new nfslock structure, otherwise just
3805 	 * a fhandle_t on the stack.
3806 	 */
3807 	if (flags & NFSLCK_OPEN) {
3808 		new_lfp = *new_lfpp;
3809 		fhp = &new_lfp->lf_fh;
3810 	} else if (nfhp) {
3811 		fhp = nfhp;
3812 	} else {
3813 		panic("nfsrv_getlockfile");
3814 	}
3815 
3816 	hp = NFSLOCKHASH(fhp);
3817 	LIST_FOREACH(lfp, hp, lf_hash) {
3818 		tfhp = &lfp->lf_fh;
3819 		if (NFSVNO_CMPFH(fhp, tfhp)) {
3820 			if (lockit)
3821 				nfsrv_locklf(lfp);
3822 			*lfpp = lfp;
3823 			return (0);
3824 		}
3825 	}
3826 	if (!(flags & NFSLCK_OPEN))
3827 		return (-1);
3828 
3829 	/*
3830 	 * No match, so chain the new one into the list.
3831 	 */
3832 	LIST_INIT(&new_lfp->lf_open);
3833 	LIST_INIT(&new_lfp->lf_lock);
3834 	LIST_INIT(&new_lfp->lf_deleg);
3835 	LIST_INIT(&new_lfp->lf_locallock);
3836 	LIST_INIT(&new_lfp->lf_rollback);
3837 	new_lfp->lf_locallock_lck.nfslock_usecnt = 0;
3838 	new_lfp->lf_locallock_lck.nfslock_lock = 0;
3839 	new_lfp->lf_usecount = 0;
3840 	LIST_INSERT_HEAD(hp, new_lfp, lf_hash);
3841 	*lfpp = new_lfp;
3842 	*new_lfpp = NULL;
3843 	return (0);
3844 }
3845 
3846 /*
3847  * This function adds a nfslock lock structure to the list for the associated
3848  * nfsstate and nfslockfile structures. It will be inserted after the
3849  * entry pointed at by insert_lop.
3850  */
3851 static void
3852 nfsrv_insertlock(struct nfslock *new_lop, struct nfslock *insert_lop,
3853     struct nfsstate *stp, struct nfslockfile *lfp)
3854 {
3855 	struct nfslock *lop, *nlop;
3856 
3857 	new_lop->lo_stp = stp;
3858 	new_lop->lo_lfp = lfp;
3859 
3860 	if (stp != NULL) {
3861 		/* Insert in increasing lo_first order */
3862 		lop = LIST_FIRST(&lfp->lf_lock);
3863 		if (lop == LIST_END(&lfp->lf_lock) ||
3864 		    new_lop->lo_first <= lop->lo_first) {
3865 			LIST_INSERT_HEAD(&lfp->lf_lock, new_lop, lo_lckfile);
3866 		} else {
3867 			nlop = LIST_NEXT(lop, lo_lckfile);
3868 			while (nlop != LIST_END(&lfp->lf_lock) &&
3869 			       nlop->lo_first < new_lop->lo_first) {
3870 				lop = nlop;
3871 				nlop = LIST_NEXT(lop, lo_lckfile);
3872 			}
3873 			LIST_INSERT_AFTER(lop, new_lop, lo_lckfile);
3874 		}
3875 	} else {
3876 		new_lop->lo_lckfile.le_prev = NULL;	/* list not used */
3877 	}
3878 
3879 	/*
3880 	 * Insert after insert_lop, which is overloaded as stp or lfp for
3881 	 * an empty list.
3882 	 */
3883 	if (stp == NULL && (struct nfslockfile *)insert_lop == lfp)
3884 		LIST_INSERT_HEAD(&lfp->lf_locallock, new_lop, lo_lckowner);
3885 	else if ((struct nfsstate *)insert_lop == stp)
3886 		LIST_INSERT_HEAD(&stp->ls_lock, new_lop, lo_lckowner);
3887 	else
3888 		LIST_INSERT_AFTER(insert_lop, new_lop, lo_lckowner);
3889 	if (stp != NULL) {
3890 		NFSD_VNET(nfsstatsv1_p)->srvlocks++;
3891 		nfsrv_openpluslock++;
3892 	}
3893 }
3894 
3895 /*
3896  * This function updates the locking for a lock owner and given file. It
3897  * maintains a list of lock ranges ordered on increasing file offset that
3898  * are NFSLCK_READ or NFSLCK_WRITE and non-overlapping (aka POSIX style).
3899  * It always adds new_lop to the list and sometimes uses the one pointed
3900  * at by other_lopp.
3901  */
3902 static void
3903 nfsrv_updatelock(struct nfsstate *stp, struct nfslock **new_lopp,
3904     struct nfslock **other_lopp, struct nfslockfile *lfp)
3905 {
3906 	struct nfslock *new_lop = *new_lopp;
3907 	struct nfslock *lop, *tlop, *ilop;
3908 	struct nfslock *other_lop = *other_lopp;
3909 	int unlock = 0, myfile = 0;
3910 	u_int64_t tmp;
3911 
3912 	/*
3913 	 * Work down the list until the lock is merged.
3914 	 */
3915 	if (new_lop->lo_flags & NFSLCK_UNLOCK)
3916 		unlock = 1;
3917 	if (stp != NULL) {
3918 		ilop = (struct nfslock *)stp;
3919 		lop = LIST_FIRST(&stp->ls_lock);
3920 	} else {
3921 		ilop = (struct nfslock *)lfp;
3922 		lop = LIST_FIRST(&lfp->lf_locallock);
3923 	}
3924 	while (lop != NULL) {
3925 	    /*
3926 	     * Only check locks for this file that aren't before the start of
3927 	     * new lock's range.
3928 	     */
3929 	    if (lop->lo_lfp == lfp) {
3930 	      myfile = 1;
3931 	      if (lop->lo_end >= new_lop->lo_first) {
3932 		if (new_lop->lo_end < lop->lo_first) {
3933 			/*
3934 			 * If the new lock ends before the start of the
3935 			 * current lock's range, no merge, just insert
3936 			 * the new lock.
3937 			 */
3938 			break;
3939 		}
3940 		if (new_lop->lo_flags == lop->lo_flags ||
3941 		    (new_lop->lo_first <= lop->lo_first &&
3942 		     new_lop->lo_end >= lop->lo_end)) {
3943 			/*
3944 			 * This lock can be absorbed by the new lock/unlock.
3945 			 * This happens when it covers the entire range
3946 			 * of the old lock or is contiguous
3947 			 * with the old lock and is of the same type or an
3948 			 * unlock.
3949 			 */
3950 			if (lop->lo_first < new_lop->lo_first)
3951 				new_lop->lo_first = lop->lo_first;
3952 			if (lop->lo_end > new_lop->lo_end)
3953 				new_lop->lo_end = lop->lo_end;
3954 			tlop = lop;
3955 			lop = LIST_NEXT(lop, lo_lckowner);
3956 			nfsrv_freenfslock(tlop);
3957 			continue;
3958 		}
3959 
3960 		/*
3961 		 * All these cases are for contiguous locks that are not the
3962 		 * same type, so they can't be merged.
3963 		 */
3964 		if (new_lop->lo_first <= lop->lo_first) {
3965 			/*
3966 			 * This case is where the new lock overlaps with the
3967 			 * first part of the old lock. Move the start of the
3968 			 * old lock to just past the end of the new lock. The
3969 			 * new lock will be inserted in front of the old, since
3970 			 * ilop hasn't been updated. (We are done now.)
3971 			 */
3972 			lop->lo_first = new_lop->lo_end;
3973 			break;
3974 		}
3975 		if (new_lop->lo_end >= lop->lo_end) {
3976 			/*
3977 			 * This case is where the new lock overlaps with the
3978 			 * end of the old lock's range. Move the old lock's
3979 			 * end to just before the new lock's first and insert
3980 			 * the new lock after the old lock.
3981 			 * Might not be done yet, since the new lock could
3982 			 * overlap further locks with higher ranges.
3983 			 */
3984 			lop->lo_end = new_lop->lo_first;
3985 			ilop = lop;
3986 			lop = LIST_NEXT(lop, lo_lckowner);
3987 			continue;
3988 		}
3989 		/*
3990 		 * The final case is where the new lock's range is in the
3991 		 * middle of the current lock's and splits the current lock
3992 		 * up. Use *other_lopp to handle the second part of the
3993 		 * split old lock range. (We are done now.)
3994 		 * For unlock, we use new_lop as other_lop and tmp, since
3995 		 * other_lop and new_lop are the same for this case.
3996 		 * We noted the unlock case above, so we don't need
3997 		 * new_lop->lo_flags any longer.
3998 		 */
3999 		tmp = new_lop->lo_first;
4000 		if (other_lop == NULL) {
4001 			if (!unlock)
4002 				panic("nfsd srv update unlock");
4003 			other_lop = new_lop;
4004 			*new_lopp = NULL;
4005 		}
4006 		other_lop->lo_first = new_lop->lo_end;
4007 		other_lop->lo_end = lop->lo_end;
4008 		other_lop->lo_flags = lop->lo_flags;
4009 		other_lop->lo_stp = stp;
4010 		other_lop->lo_lfp = lfp;
4011 		lop->lo_end = tmp;
4012 		nfsrv_insertlock(other_lop, lop, stp, lfp);
4013 		*other_lopp = NULL;
4014 		ilop = lop;
4015 		break;
4016 	      }
4017 	    }
4018 	    ilop = lop;
4019 	    lop = LIST_NEXT(lop, lo_lckowner);
4020 	    if (myfile && (lop == NULL || lop->lo_lfp != lfp))
4021 		break;
4022 	}
4023 
4024 	/*
4025 	 * Insert the new lock in the list at the appropriate place.
4026 	 */
4027 	if (!unlock) {
4028 		nfsrv_insertlock(new_lop, ilop, stp, lfp);
4029 		*new_lopp = NULL;
4030 	}
4031 }
4032 
4033 /*
4034  * This function handles sequencing of locks, etc.
4035  * It returns an error that indicates what the caller should do.
4036  */
4037 static int
4038 nfsrv_checkseqid(struct nfsrv_descript *nd, u_int32_t seqid,
4039     struct nfsstate *stp, struct nfsrvcache *op)
4040 {
4041 	int error = 0;
4042 
4043 	if ((nd->nd_flag & ND_NFSV41) != 0)
4044 		/* NFSv4.1 ignores the open_seqid and lock_seqid. */
4045 		goto out;
4046 	if (op != nd->nd_rp)
4047 		panic("nfsrvstate checkseqid");
4048 	if (!(op->rc_flag & RC_INPROG))
4049 		panic("nfsrvstate not inprog");
4050 	if (stp->ls_op && stp->ls_op->rc_refcnt <= 0) {
4051 		printf("refcnt=%d\n", stp->ls_op->rc_refcnt);
4052 		panic("nfsrvstate op refcnt");
4053 	}
4054 
4055 	/* If ND_ERELOOKUP is set, the seqid has already been handled. */
4056 	if ((nd->nd_flag & ND_ERELOOKUP) != 0)
4057 		goto out;
4058 
4059 	if ((stp->ls_seq + 1) == seqid) {
4060 		if (stp->ls_op)
4061 			nfsrvd_derefcache(stp->ls_op);
4062 		stp->ls_op = op;
4063 		nfsrvd_refcache(op);
4064 		stp->ls_seq = seqid;
4065 		goto out;
4066 	} else if (stp->ls_seq == seqid && stp->ls_op &&
4067 		op->rc_xid == stp->ls_op->rc_xid &&
4068 		op->rc_refcnt == 0 &&
4069 		op->rc_reqlen == stp->ls_op->rc_reqlen &&
4070 		op->rc_cksum == stp->ls_op->rc_cksum) {
4071 		if (stp->ls_op->rc_flag & RC_INPROG) {
4072 			error = NFSERR_DONTREPLY;
4073 			goto out;
4074 		}
4075 		nd->nd_rp = stp->ls_op;
4076 		nd->nd_rp->rc_flag |= RC_INPROG;
4077 		nfsrvd_delcache(op);
4078 		error = NFSERR_REPLYFROMCACHE;
4079 		goto out;
4080 	}
4081 	error = NFSERR_BADSEQID;
4082 
4083 out:
4084 	NFSEXITCODE2(error, nd);
4085 	return (error);
4086 }
4087 
4088 /*
4089  * Get the client ip address for callbacks. If the strings can't be parsed,
4090  * just set lc_program to 0 to indicate no callbacks are possible.
4091  * (For cases where the address can't be parsed or is 0.0.0.0.0.0, set
4092  *  the address to the client's transport address. This won't be used
4093  *  for callbacks, but can be printed out by nfsstats for info.)
4094  * Return error if the xdr can't be parsed, 0 otherwise.
4095  */
4096 int
4097 nfsrv_getclientipaddr(struct nfsrv_descript *nd, struct nfsclient *clp)
4098 {
4099 	u_int32_t *tl;
4100 	u_char *cp, *cp2;
4101 	int i, j, maxalen = 0, minalen = 0;
4102 	sa_family_t af;
4103 #ifdef INET
4104 	struct sockaddr_in *rin = NULL, *sin;
4105 #endif
4106 #ifdef INET6
4107 	struct sockaddr_in6 *rin6 = NULL, *sin6;
4108 #endif
4109 	u_char *addr;
4110 	int error = 0, cantparse = 0;
4111 	union {
4112 		in_addr_t ival;
4113 		u_char cval[4];
4114 	} ip;
4115 	union {
4116 		in_port_t sval;
4117 		u_char cval[2];
4118 	} port;
4119 
4120 	/* 8 is the maximum length of the port# string. */
4121 	addr = malloc(INET6_ADDRSTRLEN + 8, M_TEMP, M_WAITOK);
4122 	clp->lc_req.nr_client = NULL;
4123 	clp->lc_req.nr_lock = 0;
4124 	af = AF_UNSPEC;
4125 	NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
4126 	i = fxdr_unsigned(int, *tl);
4127 	if (i >= 3 && i <= 4) {
4128 		error = nfsrv_mtostr(nd, addr, i);
4129 		if (error)
4130 			goto nfsmout;
4131 #ifdef INET
4132 		if (!strcmp(addr, "tcp")) {
4133 			clp->lc_flags |= LCL_TCPCALLBACK;
4134 			clp->lc_req.nr_sotype = SOCK_STREAM;
4135 			clp->lc_req.nr_soproto = IPPROTO_TCP;
4136 			af = AF_INET;
4137 		} else if (!strcmp(addr, "udp")) {
4138 			clp->lc_req.nr_sotype = SOCK_DGRAM;
4139 			clp->lc_req.nr_soproto = IPPROTO_UDP;
4140 			af = AF_INET;
4141 		}
4142 #endif
4143 #ifdef INET6
4144 		if (af == AF_UNSPEC) {
4145 			if (!strcmp(addr, "tcp6")) {
4146 				clp->lc_flags |= LCL_TCPCALLBACK;
4147 				clp->lc_req.nr_sotype = SOCK_STREAM;
4148 				clp->lc_req.nr_soproto = IPPROTO_TCP;
4149 				af = AF_INET6;
4150 			} else if (!strcmp(addr, "udp6")) {
4151 				clp->lc_req.nr_sotype = SOCK_DGRAM;
4152 				clp->lc_req.nr_soproto = IPPROTO_UDP;
4153 				af = AF_INET6;
4154 			}
4155 		}
4156 #endif
4157 		if (af == AF_UNSPEC) {
4158 			cantparse = 1;
4159 		}
4160 	} else {
4161 		cantparse = 1;
4162 		if (i > 0) {
4163 			error = nfsm_advance(nd, NFSM_RNDUP(i), -1);
4164 			if (error)
4165 				goto nfsmout;
4166 		}
4167 	}
4168 	/*
4169 	 * The caller has allocated clp->lc_req.nr_nam to be large enough
4170 	 * for either AF_INET or AF_INET6 and zeroed out the contents.
4171 	 * maxalen is set to the maximum length of the host IP address string
4172 	 * plus 8 for the maximum length of the port#.
4173 	 * minalen is set to the minimum length of the host IP address string
4174 	 * plus 4 for the minimum length of the port#.
4175 	 * These lengths do not include NULL termination,
4176 	 * so INET[6]_ADDRSTRLEN - 1 is used in the calculations.
4177 	 */
4178 	switch (af) {
4179 #ifdef INET
4180 	case AF_INET:
4181 		rin = (struct sockaddr_in *)clp->lc_req.nr_nam;
4182 		rin->sin_family = AF_INET;
4183 		rin->sin_len = sizeof(struct sockaddr_in);
4184 		maxalen = INET_ADDRSTRLEN - 1 + 8;
4185 		minalen = 7 + 4;
4186 		break;
4187 #endif
4188 #ifdef INET6
4189 	case AF_INET6:
4190 		rin6 = (struct sockaddr_in6 *)clp->lc_req.nr_nam;
4191 		rin6->sin6_family = AF_INET6;
4192 		rin6->sin6_len = sizeof(struct sockaddr_in6);
4193 		maxalen = INET6_ADDRSTRLEN - 1 + 8;
4194 		minalen = 3 + 4;
4195 		break;
4196 #endif
4197 	}
4198 	NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
4199 	i = fxdr_unsigned(int, *tl);
4200 	if (i < 0) {
4201 		error = NFSERR_BADXDR;
4202 		goto nfsmout;
4203 	} else if (i == 0) {
4204 		cantparse = 1;
4205 	} else if (!cantparse && i <= maxalen && i >= minalen) {
4206 		error = nfsrv_mtostr(nd, addr, i);
4207 		if (error)
4208 			goto nfsmout;
4209 
4210 		/*
4211 		 * Parse out the address fields. We expect 6 decimal numbers
4212 		 * separated by '.'s for AF_INET and two decimal numbers
4213 		 * preceeded by '.'s for AF_INET6.
4214 		 */
4215 		cp = NULL;
4216 		switch (af) {
4217 #ifdef INET6
4218 		/*
4219 		 * For AF_INET6, first parse the host address.
4220 		 */
4221 		case AF_INET6:
4222 			cp = strchr(addr, '.');
4223 			if (cp != NULL) {
4224 				*cp++ = '\0';
4225 				if (inet_pton(af, addr, &rin6->sin6_addr) == 1)
4226 					i = 4;
4227 				else {
4228 					cp = NULL;
4229 					cantparse = 1;
4230 				}
4231 			}
4232 			break;
4233 #endif
4234 #ifdef INET
4235 		case AF_INET:
4236 			cp = addr;
4237 			i = 0;
4238 			break;
4239 #endif
4240 		}
4241 		while (cp != NULL && *cp && i < 6) {
4242 			cp2 = cp;
4243 			while (*cp2 && *cp2 != '.')
4244 				cp2++;
4245 			if (*cp2)
4246 				*cp2++ = '\0';
4247 			else if (i != 5) {
4248 				cantparse = 1;
4249 				break;
4250 			}
4251 			j = nfsrv_getipnumber(cp);
4252 			if (j >= 0) {
4253 				if (i < 4)
4254 					ip.cval[3 - i] = j;
4255 				else
4256 					port.cval[5 - i] = j;
4257 			} else {
4258 				cantparse = 1;
4259 				break;
4260 			}
4261 			cp = cp2;
4262 			i++;
4263 		}
4264 		if (!cantparse) {
4265 			/*
4266 			 * The host address INADDR_ANY is (mis)used to indicate
4267 			 * "there is no valid callback address".
4268 			 */
4269 			switch (af) {
4270 #ifdef INET6
4271 			case AF_INET6:
4272 				if (!IN6_ARE_ADDR_EQUAL(&rin6->sin6_addr,
4273 				    &in6addr_any))
4274 					rin6->sin6_port = htons(port.sval);
4275 				else
4276 					cantparse = 1;
4277 				break;
4278 #endif
4279 #ifdef INET
4280 			case AF_INET:
4281 				if (ip.ival != INADDR_ANY) {
4282 					rin->sin_addr.s_addr = htonl(ip.ival);
4283 					rin->sin_port = htons(port.sval);
4284 				} else {
4285 					cantparse = 1;
4286 				}
4287 				break;
4288 #endif
4289 			}
4290 		}
4291 	} else {
4292 		cantparse = 1;
4293 		if (i > 0) {
4294 			error = nfsm_advance(nd, NFSM_RNDUP(i), -1);
4295 			if (error)
4296 				goto nfsmout;
4297 		}
4298 	}
4299 	if (cantparse) {
4300 		switch (nd->nd_nam->sa_family) {
4301 #ifdef INET
4302 		case AF_INET:
4303 			sin = (struct sockaddr_in *)nd->nd_nam;
4304 			rin = (struct sockaddr_in *)clp->lc_req.nr_nam;
4305 			rin->sin_family = AF_INET;
4306 			rin->sin_len = sizeof(struct sockaddr_in);
4307 			rin->sin_addr.s_addr = sin->sin_addr.s_addr;
4308 			rin->sin_port = 0x0;
4309 			break;
4310 #endif
4311 #ifdef INET6
4312 		case AF_INET6:
4313 			sin6 = (struct sockaddr_in6 *)nd->nd_nam;
4314 			rin6 = (struct sockaddr_in6 *)clp->lc_req.nr_nam;
4315 			rin6->sin6_family = AF_INET6;
4316 			rin6->sin6_len = sizeof(struct sockaddr_in6);
4317 			rin6->sin6_addr = sin6->sin6_addr;
4318 			rin6->sin6_port = 0x0;
4319 			break;
4320 #endif
4321 		}
4322 		clp->lc_program = 0;
4323 	}
4324 nfsmout:
4325 	free(addr, M_TEMP);
4326 	NFSEXITCODE2(error, nd);
4327 	return (error);
4328 }
4329 
4330 /*
4331  * Turn a string of up to three decimal digits into a number. Return -1 upon
4332  * error.
4333  */
4334 static int
4335 nfsrv_getipnumber(u_char *cp)
4336 {
4337 	int i = 0, j = 0;
4338 
4339 	while (*cp) {
4340 		if (j > 2 || *cp < '0' || *cp > '9')
4341 			return (-1);
4342 		i *= 10;
4343 		i += (*cp - '0');
4344 		cp++;
4345 		j++;
4346 	}
4347 	if (i < 256)
4348 		return (i);
4349 	return (-1);
4350 }
4351 
4352 /*
4353  * This function checks for restart conditions.
4354  */
4355 static int
4356 nfsrv_checkrestart(nfsquad_t clientid, u_int32_t flags,
4357     nfsv4stateid_t *stateidp, int specialid)
4358 {
4359 	int ret = 0;
4360 
4361 	/*
4362 	 * First check for a server restart. Open, LockT, ReleaseLockOwner
4363 	 * and DelegPurge have a clientid, the rest a stateid.
4364 	 */
4365 	if (flags &
4366 	    (NFSLCK_OPEN | NFSLCK_TEST | NFSLCK_RELEASE | NFSLCK_DELEGPURGE)) {
4367 		if (clientid.lval[0] != NFSD_VNET(nfsrvboottime)) {
4368 			ret = NFSERR_STALECLIENTID;
4369 			goto out;
4370 		}
4371 	} else if (stateidp->other[0] != NFSD_VNET(nfsrvboottime) &&
4372 		specialid == 0) {
4373 		ret = NFSERR_STALESTATEID;
4374 		goto out;
4375 	}
4376 
4377 	/*
4378 	 * Read, Write, Setattr and LockT can return NFSERR_GRACE and do
4379 	 * not use a lock/open owner seqid#, so the check can be done now.
4380 	 * (The others will be checked, as required, later.)
4381 	 */
4382 	if (!(flags & (NFSLCK_CHECK | NFSLCK_TEST)))
4383 		goto out;
4384 
4385 	NFSLOCKSTATE();
4386 	ret = nfsrv_checkgrace(NULL, NULL, flags);
4387 	NFSUNLOCKSTATE();
4388 
4389 out:
4390 	NFSEXITCODE(ret);
4391 	return (ret);
4392 }
4393 
4394 /*
4395  * Check for grace.
4396  */
4397 static int
4398 nfsrv_checkgrace(struct nfsrv_descript *nd, struct nfsclient *clp,
4399     u_int32_t flags)
4400 {
4401 	int error = 0, notreclaimed;
4402 	struct nfsrv_stable *sp;
4403 
4404 	if ((NFSD_VNET(nfsrv_stablefirst).nsf_flags & (NFSNSF_UPDATEDONE |
4405 	     NFSNSF_GRACEOVER)) == 0) {
4406 		/*
4407 		 * First, check to see if all of the clients have done a
4408 		 * ReclaimComplete.  If so, grace can end now.
4409 		 */
4410 		notreclaimed = 0;
4411 		LIST_FOREACH(sp, &NFSD_VNET(nfsrv_stablefirst).nsf_head,
4412 		    nst_list) {
4413 			if ((sp->nst_flag & NFSNST_RECLAIMED) == 0) {
4414 				notreclaimed = 1;
4415 				break;
4416 			}
4417 		}
4418 		if (notreclaimed == 0)
4419 			NFSD_VNET(nfsrv_stablefirst).nsf_flags |=
4420 			    (NFSNSF_GRACEOVER | NFSNSF_NEEDLOCK);
4421 	}
4422 
4423 	if ((NFSD_VNET(nfsrv_stablefirst).nsf_flags & NFSNSF_GRACEOVER) != 0) {
4424 		if (flags & NFSLCK_RECLAIM) {
4425 			error = NFSERR_NOGRACE;
4426 			goto out;
4427 		}
4428 	} else {
4429 		if (!(flags & NFSLCK_RECLAIM)) {
4430 			error = NFSERR_GRACE;
4431 			goto out;
4432 		}
4433 		if (nd != NULL && clp != NULL &&
4434 		    (nd->nd_flag & ND_NFSV41) != 0 &&
4435 		    (clp->lc_flags & LCL_RECLAIMCOMPLETE) != 0) {
4436 			error = NFSERR_NOGRACE;
4437 			goto out;
4438 		}
4439 
4440 		/*
4441 		 * If grace is almost over and we are still getting Reclaims,
4442 		 * extend grace a bit.
4443 		 */
4444 		if ((NFSD_MONOSEC + NFSRV_LEASEDELTA) >
4445 		    NFSD_VNET(nfsrv_stablefirst).nsf_eograce)
4446 			NFSD_VNET(nfsrv_stablefirst).nsf_eograce =
4447 				NFSD_MONOSEC + NFSRV_LEASEDELTA;
4448 	}
4449 
4450 out:
4451 	NFSEXITCODE(error);
4452 	return (error);
4453 }
4454 
4455 /*
4456  * Do a server callback.
4457  * The "trunc" argument is slightly overloaded and refers to different
4458  * boolean arguments for CBRECALL and CBLAYOUTRECALL.
4459  */
4460 static int
4461 nfsrv_docallback(struct nfsclient *clp, int procnum, nfsv4stateid_t *stateidp,
4462     int trunc, fhandle_t *fhp, struct nfsvattr *nap, nfsattrbit_t *attrbitp,
4463     int laytype, NFSPROC_T *p)
4464 {
4465 	struct mbuf *m;
4466 	u_int32_t *tl;
4467 	struct nfsrv_descript *nd;
4468 	struct ucred *cred;
4469 	int error = 0, slotpos;
4470 	u_int32_t callback;
4471 	struct nfsdsession *sep = NULL;
4472 	uint64_t tval;
4473 	bool dotls;
4474 
4475 	nd = malloc(sizeof(*nd), M_TEMP, M_WAITOK | M_ZERO);
4476 	cred = newnfs_getcred();
4477 	NFSLOCKSTATE();	/* mostly for lc_cbref++ */
4478 	if (clp->lc_flags & LCL_NEEDSCONFIRM) {
4479 		NFSUNLOCKSTATE();
4480 		panic("docallb");
4481 	}
4482 	clp->lc_cbref++;
4483 
4484 	/*
4485 	 * Fill the callback program# and version into the request
4486 	 * structure for newnfs_connect() to use.
4487 	 */
4488 	clp->lc_req.nr_prog = clp->lc_program;
4489 #ifdef notnow
4490 	if ((clp->lc_flags & LCL_NFSV41) != 0)
4491 		clp->lc_req.nr_vers = NFSV41_CBVERS;
4492 	else
4493 #endif
4494 		clp->lc_req.nr_vers = NFSV4_CBVERS;
4495 
4496 	/*
4497 	 * First, fill in some of the fields of nd and cr.
4498 	 */
4499 	nd->nd_flag = ND_NFSV4;
4500 	if (clp->lc_flags & LCL_GSS)
4501 		nd->nd_flag |= ND_KERBV;
4502 	if ((clp->lc_flags & LCL_NFSV41) != 0)
4503 		nd->nd_flag |= ND_NFSV41;
4504 	if ((clp->lc_flags & LCL_NFSV42) != 0)
4505 		nd->nd_flag |= ND_NFSV42;
4506 	nd->nd_repstat = 0;
4507 	cred->cr_uid = clp->lc_uid;
4508 	cred->cr_gid = clp->lc_gid;
4509 	callback = clp->lc_callback;
4510 	NFSUNLOCKSTATE();
4511 	cred->cr_ngroups = 1;
4512 
4513 	/*
4514 	 * Get the first mbuf for the request.
4515 	 */
4516 	MGET(m, M_WAITOK, MT_DATA);
4517 	m->m_len = 0;
4518 	nd->nd_mreq = nd->nd_mb = m;
4519 	nd->nd_bpos = mtod(m, caddr_t);
4520 
4521 	/*
4522 	 * and build the callback request.
4523 	 */
4524 	if (procnum == NFSV4OP_CBGETATTR) {
4525 		nd->nd_procnum = NFSV4PROC_CBCOMPOUND;
4526 		error = nfsrv_cbcallargs(nd, clp, callback, NFSV4OP_CBGETATTR,
4527 		    "CB Getattr", &sep, &slotpos);
4528 		if (error != 0) {
4529 			m_freem(nd->nd_mreq);
4530 			goto errout;
4531 		}
4532 		(void)nfsm_fhtom(NULL, nd, (u_int8_t *)fhp, NFSX_MYFH, 0);
4533 		(void)nfsrv_putattrbit(nd, attrbitp);
4534 	} else if (procnum == NFSV4OP_CBRECALL) {
4535 		nd->nd_procnum = NFSV4PROC_CBCOMPOUND;
4536 		error = nfsrv_cbcallargs(nd, clp, callback, NFSV4OP_CBRECALL,
4537 		    "CB Recall", &sep, &slotpos);
4538 		if (error != 0) {
4539 			m_freem(nd->nd_mreq);
4540 			goto errout;
4541 		}
4542 		NFSM_BUILD(tl, u_int32_t *, NFSX_UNSIGNED + NFSX_STATEID);
4543 		*tl++ = txdr_unsigned(stateidp->seqid);
4544 		NFSBCOPY((caddr_t)stateidp->other, (caddr_t)tl,
4545 		    NFSX_STATEIDOTHER);
4546 		tl += (NFSX_STATEIDOTHER / NFSX_UNSIGNED);
4547 		if (trunc)
4548 			*tl = newnfs_true;
4549 		else
4550 			*tl = newnfs_false;
4551 		(void)nfsm_fhtom(NULL, nd, (u_int8_t *)fhp, NFSX_MYFH, 0);
4552 	} else if (procnum == NFSV4OP_CBLAYOUTRECALL) {
4553 		NFSD_DEBUG(4, "docallback layout recall\n");
4554 		nd->nd_procnum = NFSV4PROC_CBCOMPOUND;
4555 		error = nfsrv_cbcallargs(nd, clp, callback,
4556 		    NFSV4OP_CBLAYOUTRECALL, "CB Reclayout", &sep, &slotpos);
4557 		NFSD_DEBUG(4, "aft cbcallargs=%d\n", error);
4558 		if (error != 0) {
4559 			m_freem(nd->nd_mreq);
4560 			goto errout;
4561 		}
4562 		NFSM_BUILD(tl, u_int32_t *, 4 * NFSX_UNSIGNED);
4563 		*tl++ = txdr_unsigned(laytype);
4564 		*tl++ = txdr_unsigned(NFSLAYOUTIOMODE_ANY);
4565 		if (trunc)
4566 			*tl++ = newnfs_true;
4567 		else
4568 			*tl++ = newnfs_false;
4569 		*tl = txdr_unsigned(NFSV4LAYOUTRET_FILE);
4570 		(void)nfsm_fhtom(NULL, nd, (uint8_t *)fhp, NFSX_MYFH, 0);
4571 		NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_HYPER + NFSX_STATEID);
4572 		tval = 0;
4573 		txdr_hyper(tval, tl); tl += 2;
4574 		tval = UINT64_MAX;
4575 		txdr_hyper(tval, tl); tl += 2;
4576 		*tl++ = txdr_unsigned(stateidp->seqid);
4577 		NFSBCOPY(stateidp->other, tl, NFSX_STATEIDOTHER);
4578 		tl += (NFSX_STATEIDOTHER / NFSX_UNSIGNED);
4579 		NFSD_DEBUG(4, "aft args\n");
4580 	} else if (procnum == NFSV4PROC_CBNULL) {
4581 		nd->nd_procnum = NFSV4PROC_CBNULL;
4582 		if ((clp->lc_flags & LCL_NFSV41) != 0) {
4583 			error = nfsv4_getcbsession(clp, &sep);
4584 			if (error != 0) {
4585 				m_freem(nd->nd_mreq);
4586 				goto errout;
4587 			}
4588 		}
4589 	} else {
4590 		error = NFSERR_SERVERFAULT;
4591 		m_freem(nd->nd_mreq);
4592 		goto errout;
4593 	}
4594 
4595 	/*
4596 	 * Call newnfs_connect(), as required, and then newnfs_request().
4597 	 */
4598 	dotls = false;
4599 	if ((clp->lc_flags & LCL_TLSCB) != 0)
4600 		dotls = true;
4601 	(void) newnfs_sndlock(&clp->lc_req.nr_lock);
4602 	if (clp->lc_req.nr_client == NULL) {
4603 		if ((clp->lc_flags & LCL_NFSV41) != 0) {
4604 			error = ECONNREFUSED;
4605 			if (procnum != NFSV4PROC_CBNULL)
4606 				nfsv4_freeslot(&sep->sess_cbsess, slotpos,
4607 				    true);
4608 			nfsrv_freesession(sep, NULL);
4609 		} else if (nd->nd_procnum == NFSV4PROC_CBNULL)
4610 			error = newnfs_connect(NULL, &clp->lc_req, cred,
4611 			    NULL, 1, dotls, &clp->lc_req.nr_client);
4612 		else
4613 			error = newnfs_connect(NULL, &clp->lc_req, cred,
4614 			    NULL, 3, dotls, &clp->lc_req.nr_client);
4615 	}
4616 	newnfs_sndunlock(&clp->lc_req.nr_lock);
4617 	NFSD_DEBUG(4, "aft sndunlock=%d\n", error);
4618 	if (!error) {
4619 		if ((nd->nd_flag & ND_NFSV41) != 0) {
4620 			KASSERT(sep != NULL, ("sep NULL"));
4621 			if (sep->sess_cbsess.nfsess_xprt != NULL)
4622 				error = newnfs_request(nd, NULL, clp,
4623 				    &clp->lc_req, NULL, NULL, cred,
4624 				    clp->lc_program, clp->lc_req.nr_vers, NULL,
4625 				    1, NULL, &sep->sess_cbsess);
4626 			else {
4627 				/*
4628 				 * This should probably never occur, but if a
4629 				 * client somehow does an RPC without a
4630 				 * SequenceID Op that causes a callback just
4631 				 * after the nfsd threads have been terminated
4632 				 * and restared we could conceivably get here
4633 				 * without a backchannel xprt.
4634 				 */
4635 				printf("nfsrv_docallback: no xprt\n");
4636 				error = ECONNREFUSED;
4637 			}
4638 			NFSD_DEBUG(4, "aft newnfs_request=%d\n", error);
4639 			if (error != 0 && procnum != NFSV4PROC_CBNULL) {
4640 				/*
4641 				 * It is likely that the callback was never
4642 				 * processed by the client and, as such,
4643 				 * the sequence# for the session slot needs
4644 				 * to be backed up by one to avoid a
4645 				 * NFSERR_SEQMISORDERED error reply.
4646 				 * For the unlikely case where the callback
4647 				 * was processed by the client, this will
4648 				 * make the next callback on the slot
4649 				 * appear to be a retry.
4650 				 * Since callbacks never specify that the
4651 				 * reply be cached, this "apparent retry"
4652 				 * should not be a problem.
4653 				 */
4654 				nfsv4_freeslot(&sep->sess_cbsess, slotpos,
4655 				    true);
4656 			}
4657 			nfsrv_freesession(sep, NULL);
4658 		} else
4659 			error = newnfs_request(nd, NULL, clp, &clp->lc_req,
4660 			    NULL, NULL, cred, clp->lc_program,
4661 			    clp->lc_req.nr_vers, NULL, 1, NULL, NULL);
4662 	}
4663 errout:
4664 	NFSFREECRED(cred);
4665 
4666 	/*
4667 	 * If error is set here, the Callback path isn't working
4668 	 * properly, so twiddle the appropriate LCL_ flags.
4669 	 * (nd_repstat != 0 indicates the Callback path is working,
4670 	 *  but the callback failed on the client.)
4671 	 */
4672 	if (error) {
4673 		/*
4674 		 * Mark the callback pathway down, which disabled issuing
4675 		 * of delegations and gets Renew to return NFSERR_CBPATHDOWN.
4676 		 */
4677 		NFSLOCKSTATE();
4678 		clp->lc_flags |= LCL_CBDOWN;
4679 		NFSUNLOCKSTATE();
4680 	} else {
4681 		/*
4682 		 * Callback worked. If the callback path was down, disable
4683 		 * callbacks, so no more delegations will be issued. (This
4684 		 * is done on the assumption that the callback pathway is
4685 		 * flakey.)
4686 		 */
4687 		NFSLOCKSTATE();
4688 		if (clp->lc_flags & LCL_CBDOWN)
4689 			clp->lc_flags &= ~(LCL_CBDOWN | LCL_CALLBACKSON);
4690 		NFSUNLOCKSTATE();
4691 		if (nd->nd_repstat) {
4692 			error = nd->nd_repstat;
4693 			NFSD_DEBUG(1, "nfsrv_docallback op=%d err=%d\n",
4694 			    procnum, error);
4695 		} else if (error == 0 && procnum == NFSV4OP_CBGETATTR)
4696 			error = nfsv4_loadattr(nd, NULL, nap, NULL, NULL, 0,
4697 			    NULL, NULL, NULL, NULL, NULL, 0, NULL, NULL, NULL,
4698 			    p, NULL);
4699 		m_freem(nd->nd_mrep);
4700 	}
4701 	NFSLOCKSTATE();
4702 	clp->lc_cbref--;
4703 	if ((clp->lc_flags & LCL_WAKEUPWANTED) && clp->lc_cbref == 0) {
4704 		clp->lc_flags &= ~LCL_WAKEUPWANTED;
4705 		wakeup(clp);
4706 	}
4707 	NFSUNLOCKSTATE();
4708 
4709 	free(nd, M_TEMP);
4710 	NFSEXITCODE(error);
4711 	return (error);
4712 }
4713 
4714 /*
4715  * Set up the compound RPC for the callback.
4716  */
4717 static int
4718 nfsrv_cbcallargs(struct nfsrv_descript *nd, struct nfsclient *clp,
4719     uint32_t callback, int op, const char *optag, struct nfsdsession **sepp,
4720     int *slotposp)
4721 {
4722 	uint32_t *tl;
4723 	int error, len;
4724 
4725 	len = strlen(optag);
4726 	(void)nfsm_strtom(nd, optag, len);
4727 	NFSM_BUILD(tl, uint32_t *, 4 * NFSX_UNSIGNED);
4728 	if ((nd->nd_flag & ND_NFSV41) != 0) {
4729 		if ((nd->nd_flag & ND_NFSV42) != 0)
4730 			*tl++ = txdr_unsigned(NFSV42_MINORVERSION);
4731 		else
4732 			*tl++ = txdr_unsigned(NFSV41_MINORVERSION);
4733 		*tl++ = txdr_unsigned(callback);
4734 		*tl++ = txdr_unsigned(2);
4735 		*tl = txdr_unsigned(NFSV4OP_CBSEQUENCE);
4736 		error = nfsv4_setcbsequence(nd, clp, 1, sepp, slotposp);
4737 		if (error != 0)
4738 			return (error);
4739 		NFSM_BUILD(tl, u_int32_t *, NFSX_UNSIGNED);
4740 		*tl = txdr_unsigned(op);
4741 	} else {
4742 		*tl++ = txdr_unsigned(NFSV4_MINORVERSION);
4743 		*tl++ = txdr_unsigned(callback);
4744 		*tl++ = txdr_unsigned(1);
4745 		*tl = txdr_unsigned(op);
4746 	}
4747 	return (0);
4748 }
4749 
4750 /*
4751  * Return the next index# for a clientid. Mostly just increment and return
4752  * the next one, but... if the 32bit unsigned does actually wrap around,
4753  * it should be rebooted.
4754  * At an average rate of one new client per second, it will wrap around in
4755  * approximately 136 years. (I think the server will have been shut
4756  * down or rebooted before then.)
4757  */
4758 static u_int32_t
4759 nfsrv_nextclientindex(void)
4760 {
4761 	static u_int32_t client_index = 0;
4762 
4763 	client_index++;
4764 	if (client_index != 0)
4765 		return (client_index);
4766 
4767 	printf("%s: out of clientids\n", __func__);
4768 	return (client_index);
4769 }
4770 
4771 /*
4772  * Return the next index# for a stateid. Mostly just increment and return
4773  * the next one, but... if the 32bit unsigned does actually wrap around
4774  * (will a BSD server stay up that long?), find
4775  * new start and end values.
4776  */
4777 static u_int32_t
4778 nfsrv_nextstateindex(struct nfsclient *clp)
4779 {
4780 	struct nfsstate *stp;
4781 	int i;
4782 	u_int32_t canuse, min_index, max_index;
4783 
4784 	if (!(clp->lc_flags & LCL_INDEXNOTOK)) {
4785 		clp->lc_stateindex++;
4786 		if (clp->lc_stateindex != clp->lc_statemaxindex)
4787 			return (clp->lc_stateindex);
4788 	}
4789 
4790 	/*
4791 	 * Yuck, we've hit the end.
4792 	 * Look for a new min and max.
4793 	 */
4794 	min_index = 0;
4795 	max_index = 0xffffffff;
4796 	for (i = 0; i < nfsrv_statehashsize; i++) {
4797 	    LIST_FOREACH(stp, &clp->lc_stateid[i], ls_hash) {
4798 		if (stp->ls_stateid.other[2] > 0x80000000) {
4799 		    if (stp->ls_stateid.other[2] < max_index)
4800 			max_index = stp->ls_stateid.other[2];
4801 		} else {
4802 		    if (stp->ls_stateid.other[2] > min_index)
4803 			min_index = stp->ls_stateid.other[2];
4804 		}
4805 	    }
4806 	}
4807 
4808 	/*
4809 	 * Yikes, highly unlikely, but I'll handle it anyhow.
4810 	 */
4811 	if (min_index == 0x80000000 && max_index == 0x80000001) {
4812 	    canuse = 0;
4813 	    /*
4814 	     * Loop around until we find an unused entry. Return that
4815 	     * and set LCL_INDEXNOTOK, so the search will continue next time.
4816 	     * (This is one of those rare cases where a goto is the
4817 	     *  cleanest way to code the loop.)
4818 	     */
4819 tryagain:
4820 	    for (i = 0; i < nfsrv_statehashsize; i++) {
4821 		LIST_FOREACH(stp, &clp->lc_stateid[i], ls_hash) {
4822 		    if (stp->ls_stateid.other[2] == canuse) {
4823 			canuse++;
4824 			goto tryagain;
4825 		    }
4826 		}
4827 	    }
4828 	    clp->lc_flags |= LCL_INDEXNOTOK;
4829 	    return (canuse);
4830 	}
4831 
4832 	/*
4833 	 * Ok to start again from min + 1.
4834 	 */
4835 	clp->lc_stateindex = min_index + 1;
4836 	clp->lc_statemaxindex = max_index;
4837 	clp->lc_flags &= ~LCL_INDEXNOTOK;
4838 	return (clp->lc_stateindex);
4839 }
4840 
4841 /*
4842  * The following functions handle the stable storage file that deals with
4843  * the edge conditions described in RFC3530 Sec. 8.6.3.
4844  * The file is as follows:
4845  * - a single record at the beginning that has the lease time of the
4846  *   previous server instance (before the last reboot) and the nfsrvboottime
4847  *   values for the previous server boots.
4848  *   These previous boot times are used to ensure that the current
4849  *   nfsrvboottime does not, somehow, get set to a previous one.
4850  *   (This is important so that Stale ClientIDs and StateIDs can
4851  *    be recognized.)
4852  *   The number of previous nfsvrboottime values precedes the list.
4853  * - followed by some number of appended records with:
4854  *   - client id string
4855  *   - flag that indicates it is a record revoking state via lease
4856  *     expiration or similar
4857  *     OR has successfully acquired state.
4858  * These structures vary in length, with the client string at the end, up
4859  * to NFSV4_OPAQUELIMIT in size.
4860  *
4861  * At the end of the grace period, the file is truncated, the first
4862  * record is rewritten with updated information and any acquired state
4863  * records for successful reclaims of state are written.
4864  *
4865  * Subsequent records are appended when the first state is issued to
4866  * a client and when state is revoked for a client.
4867  *
4868  * When reading the file in, state issued records that come later in
4869  * the file override older ones, since the append log is in cronological order.
4870  * If, for some reason, the file can't be read, the grace period is
4871  * immediately terminated and all reclaims get NFSERR_NOGRACE.
4872  */
4873 
4874 /*
4875  * Read in the stable storage file. Called by nfssvc() before the nfsd
4876  * processes start servicing requests.
4877  */
4878 void
4879 nfsrv_setupstable(NFSPROC_T *p)
4880 {
4881 	struct nfsrv_stablefirst *sf = &NFSD_VNET(nfsrv_stablefirst);
4882 	struct nfsrv_stable *sp, *nsp;
4883 	struct nfst_rec *tsp;
4884 	int error, i, tryagain;
4885 	off_t off = 0;
4886 	ssize_t aresid, len;
4887 
4888 	/*
4889 	 * If NFSNSF_UPDATEDONE is set, this is a restart of the nfsds without
4890 	 * a reboot, so state has not been lost.
4891 	 */
4892 	if (sf->nsf_flags & NFSNSF_UPDATEDONE)
4893 		return;
4894 	/*
4895 	 * Set Grace over just until the file reads successfully.
4896 	 */
4897 	NFSD_VNET(nfsrvboottime) = time_second;
4898 	LIST_INIT(&sf->nsf_head);
4899 	sf->nsf_flags = (NFSNSF_GRACEOVER | NFSNSF_NEEDLOCK);
4900 	sf->nsf_eograce = NFSD_MONOSEC + NFSRV_LEASEDELTA;
4901 	if (sf->nsf_fp == NULL)
4902 		return;
4903 	error = NFSD_RDWR(UIO_READ, NFSFPVNODE(sf->nsf_fp),
4904 	    (caddr_t)&sf->nsf_rec, sizeof (struct nfsf_rec), off, UIO_SYSSPACE,
4905 	    0, NFSFPCRED(sf->nsf_fp), &aresid, p);
4906 	if (error || aresid || sf->nsf_numboots == 0 ||
4907 		sf->nsf_numboots > NFSNSF_MAXNUMBOOTS)
4908 		return;
4909 
4910 	/*
4911 	 * Now, read in the boottimes.
4912 	 */
4913 	sf->nsf_bootvals = (time_t *)malloc((sf->nsf_numboots + 1) *
4914 		sizeof(time_t), M_TEMP, M_WAITOK);
4915 	off = sizeof (struct nfsf_rec);
4916 	error = NFSD_RDWR(UIO_READ, NFSFPVNODE(sf->nsf_fp),
4917 	    (caddr_t)sf->nsf_bootvals, sf->nsf_numboots * sizeof (time_t), off,
4918 	    UIO_SYSSPACE, 0, NFSFPCRED(sf->nsf_fp), &aresid, p);
4919 	if (error || aresid) {
4920 		free(sf->nsf_bootvals, M_TEMP);
4921 		sf->nsf_bootvals = NULL;
4922 		return;
4923 	}
4924 
4925 	/*
4926 	 * Make sure this nfsrvboottime is different from all recorded
4927 	 * previous ones.
4928 	 */
4929 	do {
4930 		tryagain = 0;
4931 		for (i = 0; i < sf->nsf_numboots; i++) {
4932 			if (NFSD_VNET(nfsrvboottime) == sf->nsf_bootvals[i]) {
4933 				NFSD_VNET(nfsrvboottime)++;
4934 				tryagain = 1;
4935 				break;
4936 			}
4937 		}
4938 	} while (tryagain);
4939 
4940 	sf->nsf_flags |= NFSNSF_OK;
4941 	off += (sf->nsf_numboots * sizeof (time_t));
4942 
4943 	/*
4944 	 * Read through the file, building a list of records for grace
4945 	 * checking.
4946 	 * Each record is between sizeof (struct nfst_rec) and
4947 	 * sizeof (struct nfst_rec) + NFSV4_OPAQUELIMIT - 1
4948 	 * and is actually sizeof (struct nfst_rec) + nst_len - 1.
4949 	 */
4950 	tsp = (struct nfst_rec *)malloc(sizeof (struct nfst_rec) +
4951 		NFSV4_OPAQUELIMIT - 1, M_TEMP, M_WAITOK);
4952 	do {
4953 	    error = NFSD_RDWR(UIO_READ, NFSFPVNODE(sf->nsf_fp),
4954 	        (caddr_t)tsp, sizeof (struct nfst_rec) + NFSV4_OPAQUELIMIT - 1,
4955 	        off, UIO_SYSSPACE, 0, NFSFPCRED(sf->nsf_fp), &aresid, p);
4956 	    len = (sizeof (struct nfst_rec) + NFSV4_OPAQUELIMIT - 1) - aresid;
4957 	    if (error || (len > 0 && (len < sizeof (struct nfst_rec) ||
4958 		len < (sizeof (struct nfst_rec) + tsp->len - 1)))) {
4959 		/*
4960 		 * Yuck, the file has been corrupted, so just return
4961 		 * after clearing out any restart state, so the grace period
4962 		 * is over.
4963 		 */
4964 		LIST_FOREACH_SAFE(sp, &sf->nsf_head, nst_list, nsp) {
4965 			LIST_REMOVE(sp, nst_list);
4966 			free(sp, M_TEMP);
4967 		}
4968 		free(tsp, M_TEMP);
4969 		sf->nsf_flags &= ~NFSNSF_OK;
4970 		free(sf->nsf_bootvals, M_TEMP);
4971 		sf->nsf_bootvals = NULL;
4972 		return;
4973 	    }
4974 	    if (len > 0) {
4975 		off += sizeof (struct nfst_rec) + tsp->len - 1;
4976 		/*
4977 		 * Search the list for a matching client.
4978 		 */
4979 		LIST_FOREACH(sp, &sf->nsf_head, nst_list) {
4980 			if (tsp->len == sp->nst_len &&
4981 			    !NFSBCMP(tsp->client, sp->nst_client, tsp->len))
4982 				break;
4983 		}
4984 		if (sp == LIST_END(&sf->nsf_head)) {
4985 			sp = (struct nfsrv_stable *)malloc(tsp->len +
4986 				sizeof (struct nfsrv_stable) - 1, M_TEMP,
4987 				M_WAITOK);
4988 			NFSBCOPY((caddr_t)tsp, (caddr_t)&sp->nst_rec,
4989 				sizeof (struct nfst_rec) + tsp->len - 1);
4990 			LIST_INSERT_HEAD(&sf->nsf_head, sp, nst_list);
4991 		} else {
4992 			if (tsp->flag == NFSNST_REVOKE)
4993 				sp->nst_flag |= NFSNST_REVOKE;
4994 			else
4995 				/*
4996 				 * A subsequent timestamp indicates the client
4997 				 * did a setclientid/confirm and any previous
4998 				 * revoke is no longer relevant.
4999 				 */
5000 				sp->nst_flag &= ~NFSNST_REVOKE;
5001 		}
5002 	    }
5003 	} while (len > 0);
5004 	free(tsp, M_TEMP);
5005 	sf->nsf_flags = NFSNSF_OK;
5006 	sf->nsf_eograce = NFSD_MONOSEC + sf->nsf_lease +
5007 		NFSRV_LEASEDELTA;
5008 }
5009 
5010 /*
5011  * Update the stable storage file, now that the grace period is over.
5012  */
5013 void
5014 nfsrv_updatestable(NFSPROC_T *p)
5015 {
5016 	struct nfsrv_stablefirst *sf = &NFSD_VNET(nfsrv_stablefirst);
5017 	struct nfsrv_stable *sp, *nsp;
5018 	int i;
5019 	struct nfsvattr nva;
5020 	vnode_t vp;
5021 #if defined(__FreeBSD_version) && (__FreeBSD_version >= 500000)
5022 	mount_t mp = NULL;
5023 #endif
5024 	int error;
5025 
5026 	if (sf->nsf_fp == NULL || (sf->nsf_flags & NFSNSF_UPDATEDONE))
5027 		return;
5028 	sf->nsf_flags |= NFSNSF_UPDATEDONE;
5029 	/*
5030 	 * Ok, we need to rewrite the stable storage file.
5031 	 * - truncate to 0 length
5032 	 * - write the new first structure
5033 	 * - loop through the data structures, writing out any that
5034 	 *   have timestamps older than the old boot
5035 	 */
5036 	if (sf->nsf_bootvals) {
5037 		sf->nsf_numboots++;
5038 		for (i = sf->nsf_numboots - 2; i >= 0; i--)
5039 			sf->nsf_bootvals[i + 1] = sf->nsf_bootvals[i];
5040 	} else {
5041 		sf->nsf_numboots = 1;
5042 		sf->nsf_bootvals = (time_t *)malloc(sizeof(time_t),
5043 			M_TEMP, M_WAITOK);
5044 	}
5045 	sf->nsf_bootvals[0] = NFSD_VNET(nfsrvboottime);
5046 	sf->nsf_lease = nfsrv_lease;
5047 	NFSVNO_ATTRINIT(&nva);
5048 	NFSVNO_SETATTRVAL(&nva, size, 0);
5049 	vp = NFSFPVNODE(sf->nsf_fp);
5050 	vn_start_write(vp, &mp, V_WAIT);
5051 	if (NFSVOPLOCK(vp, LK_EXCLUSIVE) == 0) {
5052 		error = nfsvno_setattr(vp, &nva, NFSFPCRED(sf->nsf_fp), p,
5053 		    NULL);
5054 		NFSVOPUNLOCK(vp);
5055 	} else
5056 		error = EPERM;
5057 	vn_finished_write(mp);
5058 	if (!error)
5059 	    error = NFSD_RDWR(UIO_WRITE, vp,
5060 		(caddr_t)&sf->nsf_rec, sizeof (struct nfsf_rec), (off_t)0,
5061 		UIO_SYSSPACE, IO_SYNC, NFSFPCRED(sf->nsf_fp), NULL, p);
5062 	if (!error)
5063 	    error = NFSD_RDWR(UIO_WRITE, vp,
5064 		(caddr_t)sf->nsf_bootvals,
5065 		sf->nsf_numboots * sizeof (time_t),
5066 		(off_t)(sizeof (struct nfsf_rec)),
5067 		UIO_SYSSPACE, IO_SYNC, NFSFPCRED(sf->nsf_fp), NULL, p);
5068 	free(sf->nsf_bootvals, M_TEMP);
5069 	sf->nsf_bootvals = NULL;
5070 	if (error) {
5071 		sf->nsf_flags &= ~NFSNSF_OK;
5072 		printf("EEK! Can't write NfsV4 stable storage file\n");
5073 		return;
5074 	}
5075 	sf->nsf_flags |= NFSNSF_OK;
5076 
5077 	/*
5078 	 * Loop through the list and write out timestamp records for
5079 	 * any clients that successfully reclaimed state.
5080 	 */
5081 	LIST_FOREACH_SAFE(sp, &sf->nsf_head, nst_list, nsp) {
5082 		if (sp->nst_flag & NFSNST_GOTSTATE) {
5083 			nfsrv_writestable(sp->nst_client, sp->nst_len,
5084 				NFSNST_NEWSTATE, p);
5085 			sp->nst_clp->lc_flags |= LCL_STAMPEDSTABLE;
5086 		}
5087 		LIST_REMOVE(sp, nst_list);
5088 		free(sp, M_TEMP);
5089 	}
5090 	nfsrv_backupstable();
5091 }
5092 
5093 /*
5094  * Append a record to the stable storage file.
5095  */
5096 void
5097 nfsrv_writestable(u_char *client, int len, int flag, NFSPROC_T *p)
5098 {
5099 	struct nfsrv_stablefirst *sf = &NFSD_VNET(nfsrv_stablefirst);
5100 	struct nfst_rec *sp;
5101 	int error;
5102 
5103 	if (!(sf->nsf_flags & NFSNSF_OK) || sf->nsf_fp == NULL)
5104 		return;
5105 	sp = (struct nfst_rec *)malloc(sizeof (struct nfst_rec) +
5106 		len - 1, M_TEMP, M_WAITOK);
5107 	sp->len = len;
5108 	NFSBCOPY(client, sp->client, len);
5109 	sp->flag = flag;
5110 	error = NFSD_RDWR(UIO_WRITE, NFSFPVNODE(sf->nsf_fp),
5111 	    (caddr_t)sp, sizeof (struct nfst_rec) + len - 1, (off_t)0,
5112 	    UIO_SYSSPACE, (IO_SYNC | IO_APPEND), NFSFPCRED(sf->nsf_fp), NULL, p);
5113 	free(sp, M_TEMP);
5114 	if (error) {
5115 		sf->nsf_flags &= ~NFSNSF_OK;
5116 		printf("EEK! Can't write NfsV4 stable storage file\n");
5117 	}
5118 }
5119 
5120 /*
5121  * This function is called during the grace period to mark a client
5122  * that successfully reclaimed state.
5123  */
5124 static void
5125 nfsrv_markstable(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 mark it and set the nfsclient back pointer.
5142 	 */
5143 	sp->nst_flag |= NFSNST_GOTSTATE;
5144 	sp->nst_clp = clp;
5145 }
5146 
5147 /*
5148  * This function is called when a NFSv4.1 client does a ReclaimComplete.
5149  * Very similar to nfsrv_markstable(), except for the flag being set.
5150  */
5151 static void
5152 nfsrv_markreclaim(struct nfsclient *clp)
5153 {
5154 	struct nfsrv_stable *sp;
5155 
5156 	/*
5157 	 * First find the client structure.
5158 	 */
5159 	LIST_FOREACH(sp, &NFSD_VNET(nfsrv_stablefirst).nsf_head, nst_list) {
5160 		if (sp->nst_len == clp->lc_idlen &&
5161 		    !NFSBCMP(sp->nst_client, clp->lc_id, sp->nst_len))
5162 			break;
5163 	}
5164 	if (sp == LIST_END(&NFSD_VNET(nfsrv_stablefirst).nsf_head))
5165 		return;
5166 
5167 	/*
5168 	 * Now, just set the flag.
5169 	 */
5170 	sp->nst_flag |= NFSNST_RECLAIMED;
5171 }
5172 
5173 /*
5174  * This function is called for a reclaim, to see if it gets grace.
5175  * It returns 0 if a reclaim is allowed, 1 otherwise.
5176  */
5177 static int
5178 nfsrv_checkstable(struct nfsclient *clp)
5179 {
5180 	struct nfsrv_stable *sp;
5181 
5182 	/*
5183 	 * First, find the entry for the client.
5184 	 */
5185 	LIST_FOREACH(sp, &NFSD_VNET(nfsrv_stablefirst).nsf_head, nst_list) {
5186 		if (sp->nst_len == clp->lc_idlen &&
5187 		    !NFSBCMP(sp->nst_client, clp->lc_id, sp->nst_len))
5188 			break;
5189 	}
5190 
5191 	/*
5192 	 * If not in the list, state was revoked or no state was issued
5193 	 * since the previous reboot, a reclaim is denied.
5194 	 */
5195 	if (sp == LIST_END(&NFSD_VNET(nfsrv_stablefirst).nsf_head) ||
5196 	    (sp->nst_flag & NFSNST_REVOKE) ||
5197 	    !(NFSD_VNET(nfsrv_stablefirst).nsf_flags & NFSNSF_OK))
5198 		return (1);
5199 	return (0);
5200 }
5201 
5202 /*
5203  * Test for and try to clear out a conflicting client. This is called by
5204  * nfsrv_lockctrl() and nfsrv_openctrl() when conflicts with other clients
5205  * a found.
5206  * The trick here is that it can't revoke a conflicting client with an
5207  * expired lease unless it holds the v4root lock, so...
5208  * If no v4root lock, get the lock and return 1 to indicate "try again".
5209  * Return 0 to indicate the conflict can't be revoked and 1 to indicate
5210  * the revocation worked and the conflicting client is "bye, bye", so it
5211  * can be tried again.
5212  * Return 2 to indicate that the vnode is VIRF_DOOMED after NFSVOPLOCK().
5213  * Unlocks State before a non-zero value is returned.
5214  */
5215 static int
5216 nfsrv_clientconflict(struct nfsclient *clp, int *haslockp, vnode_t vp,
5217     NFSPROC_T *p)
5218 {
5219 	int gotlock, lktype = 0;
5220 
5221 	/*
5222 	 * If lease hasn't expired, we can't fix it.
5223 	 */
5224 	if (clp->lc_expiry >= NFSD_MONOSEC ||
5225 	    !(NFSD_VNET(nfsrv_stablefirst).nsf_flags & NFSNSF_UPDATEDONE))
5226 		return (0);
5227 	if (*haslockp == 0) {
5228 		NFSUNLOCKSTATE();
5229 		if (vp != NULL) {
5230 			lktype = NFSVOPISLOCKED(vp);
5231 			NFSVOPUNLOCK(vp);
5232 		}
5233 		NFSLOCKV4ROOTMUTEX();
5234 		nfsv4_relref(&nfsv4rootfs_lock);
5235 		do {
5236 			gotlock = nfsv4_lock(&nfsv4rootfs_lock, 1, NULL,
5237 			    NFSV4ROOTLOCKMUTEXPTR, NULL);
5238 		} while (!gotlock);
5239 		NFSUNLOCKV4ROOTMUTEX();
5240 		*haslockp = 1;
5241 		if (vp != NULL) {
5242 			NFSVOPLOCK(vp, lktype | LK_RETRY);
5243 			if (VN_IS_DOOMED(vp))
5244 				return (2);
5245 		}
5246 		return (1);
5247 	}
5248 	NFSUNLOCKSTATE();
5249 
5250 	/*
5251 	 * Ok, we can expire the conflicting client.
5252 	 */
5253 	nfsrv_writestable(clp->lc_id, clp->lc_idlen, NFSNST_REVOKE, p);
5254 	nfsrv_backupstable();
5255 	nfsrv_cleanclient(clp, p);
5256 	nfsrv_freedeleglist(&clp->lc_deleg);
5257 	nfsrv_freedeleglist(&clp->lc_olddeleg);
5258 	LIST_REMOVE(clp, lc_hash);
5259 	nfsrv_zapclient(clp, p);
5260 	return (1);
5261 }
5262 
5263 /*
5264  * Resolve a delegation conflict.
5265  * Returns 0 to indicate the conflict was resolved without sleeping.
5266  * Return -1 to indicate that the caller should check for conflicts again.
5267  * Return > 0 for an error that should be returned, normally NFSERR_DELAY.
5268  *
5269  * Also, manipulate the nfsv4root_lock, as required. It isn't changed
5270  * for a return of 0, since there was no sleep and it could be required
5271  * later. It is released for a return of NFSERR_DELAY, since the caller
5272  * will return that error. It is released when a sleep was done waiting
5273  * for the delegation to be returned or expire (so that other nfsds can
5274  * handle ops). Then, it must be acquired for the write to stable storage.
5275  * (This function is somewhat similar to nfsrv_clientconflict(), but
5276  *  the semantics differ in a couple of subtle ways. The return of 0
5277  *  indicates the conflict was resolved without sleeping here, not
5278  *  that the conflict can't be resolved and the handling of nfsv4root_lock
5279  *  differs, as noted above.)
5280  * Unlocks State before returning a non-zero value.
5281  */
5282 static int
5283 nfsrv_delegconflict(struct nfsstate *stp, int *haslockp, NFSPROC_T *p,
5284     vnode_t vp)
5285 {
5286 	struct nfsclient *clp = stp->ls_clp;
5287 	int gotlock, error, lktype = 0, retrycnt, zapped_clp;
5288 	nfsv4stateid_t tstateid;
5289 	fhandle_t tfh;
5290 
5291 	/*
5292 	 * If the conflict is with an old delegation...
5293 	 */
5294 	if (stp->ls_flags & NFSLCK_OLDDELEG) {
5295 		/*
5296 		 * You can delete it, if it has expired.
5297 		 */
5298 		if (clp->lc_delegtime < NFSD_MONOSEC) {
5299 			nfsrv_freedeleg(stp);
5300 			NFSUNLOCKSTATE();
5301 			error = -1;
5302 			goto out;
5303 		}
5304 		NFSUNLOCKSTATE();
5305 		/*
5306 		 * During this delay, the old delegation could expire or it
5307 		 * could be recovered by the client via an Open with
5308 		 * CLAIM_DELEGATE_PREV.
5309 		 * Release the nfsv4root_lock, if held.
5310 		 */
5311 		if (*haslockp) {
5312 			*haslockp = 0;
5313 			NFSLOCKV4ROOTMUTEX();
5314 			nfsv4_unlock(&nfsv4rootfs_lock, 1);
5315 			NFSUNLOCKV4ROOTMUTEX();
5316 		}
5317 		error = NFSERR_DELAY;
5318 		goto out;
5319 	}
5320 
5321 	/*
5322 	 * It's a current delegation, so:
5323 	 * - check to see if the delegation has expired
5324 	 *   - if so, get the v4root lock and then expire it
5325 	 */
5326 	if ((stp->ls_flags & NFSLCK_DELEGRECALL) == 0 || (stp->ls_lastrecall <
5327 	    NFSD_MONOSEC && clp->lc_expiry >= NFSD_MONOSEC &&
5328 	    stp->ls_delegtime >= NFSD_MONOSEC)) {
5329 		/*
5330 		 * - do a recall callback, since not yet done
5331 		 * For now, never allow truncate to be set. To use
5332 		 * truncate safely, it must be guaranteed that the
5333 		 * Remove, Rename or Setattr with size of 0 will
5334 		 * succeed and that would require major changes to
5335 		 * the VFS/Vnode OPs.
5336 		 * Set the expiry time large enough so that it won't expire
5337 		 * until after the callback, then set it correctly, once
5338 		 * the callback is done. (The delegation will now time
5339 		 * out whether or not the Recall worked ok. The timeout
5340 		 * will be extended when ops are done on the delegation
5341 		 * stateid, up to the timelimit.)
5342 		 */
5343 		if ((stp->ls_flags & NFSLCK_DELEGRECALL) == 0) {
5344 			stp->ls_delegtime = NFSD_MONOSEC + (2 * nfsrv_lease) +
5345 			    NFSRV_LEASEDELTA;
5346 			stp->ls_delegtimelimit = NFSD_MONOSEC + (6 *
5347 			    nfsrv_lease) + NFSRV_LEASEDELTA;
5348 			stp->ls_flags |= NFSLCK_DELEGRECALL;
5349 		}
5350 		stp->ls_lastrecall = time_uptime + 1;
5351 
5352 		/*
5353 		 * Loop NFSRV_CBRETRYCNT times while the CBRecall replies
5354 		 * NFSERR_BADSTATEID or NFSERR_BADHANDLE. This is done
5355 		 * in order to try and avoid a race that could happen
5356 		 * when a CBRecall request passed the Open reply with
5357 		 * the delegation in it when transitting the network.
5358 		 * Since nfsrv_docallback will sleep, don't use stp after
5359 		 * the call.
5360 		 */
5361 		NFSBCOPY((caddr_t)&stp->ls_stateid, (caddr_t)&tstateid,
5362 		    sizeof (tstateid));
5363 		NFSBCOPY((caddr_t)&stp->ls_lfp->lf_fh, (caddr_t)&tfh,
5364 		    sizeof (tfh));
5365 		NFSUNLOCKSTATE();
5366 		if (*haslockp) {
5367 			*haslockp = 0;
5368 			NFSLOCKV4ROOTMUTEX();
5369 			nfsv4_unlock(&nfsv4rootfs_lock, 1);
5370 			NFSUNLOCKV4ROOTMUTEX();
5371 		}
5372 		retrycnt = 0;
5373 		do {
5374 		    error = nfsrv_docallback(clp, NFSV4OP_CBRECALL,
5375 			&tstateid, 0, &tfh, NULL, NULL, 0, p);
5376 		    retrycnt++;
5377 		} while ((error == NFSERR_BADSTATEID ||
5378 		    error == NFSERR_BADHANDLE) && retrycnt < NFSV4_CBRETRYCNT);
5379 		error = NFSERR_DELAY;
5380 		goto out;
5381 	}
5382 
5383 	if (clp->lc_expiry >= NFSD_MONOSEC &&
5384 	    stp->ls_delegtime >= NFSD_MONOSEC) {
5385 		NFSUNLOCKSTATE();
5386 		/*
5387 		 * A recall has been done, but it has not yet expired.
5388 		 * So, RETURN_DELAY.
5389 		 */
5390 		if (*haslockp) {
5391 			*haslockp = 0;
5392 			NFSLOCKV4ROOTMUTEX();
5393 			nfsv4_unlock(&nfsv4rootfs_lock, 1);
5394 			NFSUNLOCKV4ROOTMUTEX();
5395 		}
5396 		error = NFSERR_DELAY;
5397 		goto out;
5398 	}
5399 
5400 	/*
5401 	 * If we don't yet have the lock, just get it and then return,
5402 	 * since we need that before deleting expired state, such as
5403 	 * this delegation.
5404 	 * When getting the lock, unlock the vnode, so other nfsds that
5405 	 * are in progress, won't get stuck waiting for the vnode lock.
5406 	 */
5407 	if (*haslockp == 0) {
5408 		NFSUNLOCKSTATE();
5409 		if (vp != NULL) {
5410 			lktype = NFSVOPISLOCKED(vp);
5411 			NFSVOPUNLOCK(vp);
5412 		}
5413 		NFSLOCKV4ROOTMUTEX();
5414 		nfsv4_relref(&nfsv4rootfs_lock);
5415 		do {
5416 			gotlock = nfsv4_lock(&nfsv4rootfs_lock, 1, NULL,
5417 			    NFSV4ROOTLOCKMUTEXPTR, NULL);
5418 		} while (!gotlock);
5419 		NFSUNLOCKV4ROOTMUTEX();
5420 		*haslockp = 1;
5421 		if (vp != NULL) {
5422 			NFSVOPLOCK(vp, lktype | LK_RETRY);
5423 			if (VN_IS_DOOMED(vp)) {
5424 				*haslockp = 0;
5425 				NFSLOCKV4ROOTMUTEX();
5426 				nfsv4_unlock(&nfsv4rootfs_lock, 1);
5427 				NFSUNLOCKV4ROOTMUTEX();
5428 				error = NFSERR_PERM;
5429 				goto out;
5430 			}
5431 		}
5432 		error = -1;
5433 		goto out;
5434 	}
5435 
5436 	NFSUNLOCKSTATE();
5437 	/*
5438 	 * Ok, we can delete the expired delegation.
5439 	 * First, write the Revoke record to stable storage and then
5440 	 * clear out the conflict.
5441 	 * Since all other nfsd threads are now blocked, we can safely
5442 	 * sleep without the state changing.
5443 	 */
5444 	nfsrv_writestable(clp->lc_id, clp->lc_idlen, NFSNST_REVOKE, p);
5445 	nfsrv_backupstable();
5446 	if (clp->lc_expiry < NFSD_MONOSEC) {
5447 		nfsrv_cleanclient(clp, p);
5448 		nfsrv_freedeleglist(&clp->lc_deleg);
5449 		nfsrv_freedeleglist(&clp->lc_olddeleg);
5450 		LIST_REMOVE(clp, lc_hash);
5451 		zapped_clp = 1;
5452 	} else {
5453 		nfsrv_freedeleg(stp);
5454 		zapped_clp = 0;
5455 	}
5456 	if (zapped_clp)
5457 		nfsrv_zapclient(clp, p);
5458 	error = -1;
5459 
5460 out:
5461 	NFSEXITCODE(error);
5462 	return (error);
5463 }
5464 
5465 /*
5466  * Check for a remove allowed, if remove is set to 1 and get rid of
5467  * delegations.
5468  */
5469 int
5470 nfsrv_checkremove(vnode_t vp, int remove, struct nfsrv_descript *nd,
5471     nfsquad_t clientid, NFSPROC_T *p)
5472 {
5473 	struct nfsclient *clp;
5474 	struct nfsstate *stp;
5475 	struct nfslockfile *lfp;
5476 	int error, haslock = 0;
5477 	fhandle_t nfh;
5478 
5479 	clp = NULL;
5480 	/*
5481 	 * First, get the lock file structure.
5482 	 * (A return of -1 means no associated state, so remove ok.)
5483 	 */
5484 	error = nfsrv_getlockfh(vp, NFSLCK_CHECK, NULL, &nfh, p);
5485 tryagain:
5486 	NFSLOCKSTATE();
5487 	if (error == 0 && clientid.qval != 0)
5488 		error = nfsrv_getclient(clientid, CLOPS_RENEW, &clp, NULL,
5489 		    (nfsquad_t)((u_quad_t)0), 0, nd, p);
5490 	if (!error)
5491 		error = nfsrv_getlockfile(NFSLCK_CHECK, NULL, &lfp, &nfh, 0);
5492 	if (error) {
5493 		NFSUNLOCKSTATE();
5494 		if (haslock) {
5495 			NFSLOCKV4ROOTMUTEX();
5496 			nfsv4_unlock(&nfsv4rootfs_lock, 1);
5497 			NFSUNLOCKV4ROOTMUTEX();
5498 		}
5499 		if (error == -1)
5500 			error = 0;
5501 		goto out;
5502 	}
5503 
5504 	/*
5505 	 * Now, we must Recall any delegations.
5506 	 */
5507 	error = nfsrv_cleandeleg(vp, lfp, clp, &haslock, p);
5508 	if (error) {
5509 		/*
5510 		 * nfsrv_cleandeleg() unlocks state for non-zero
5511 		 * return.
5512 		 */
5513 		if (error == -1)
5514 			goto tryagain;
5515 		if (haslock) {
5516 			NFSLOCKV4ROOTMUTEX();
5517 			nfsv4_unlock(&nfsv4rootfs_lock, 1);
5518 			NFSUNLOCKV4ROOTMUTEX();
5519 		}
5520 		goto out;
5521 	}
5522 
5523 	/*
5524 	 * Now, look for a conflicting open share.
5525 	 */
5526 	if (remove) {
5527 		/*
5528 		 * If the entry in the directory was the last reference to the
5529 		 * corresponding filesystem object, the object can be destroyed
5530 		 * */
5531 		if(lfp->lf_usecount>1)
5532 			LIST_FOREACH(stp, &lfp->lf_open, ls_file) {
5533 				if (stp->ls_flags & NFSLCK_WRITEDENY) {
5534 					error = NFSERR_FILEOPEN;
5535 					break;
5536 				}
5537 			}
5538 	}
5539 
5540 	NFSUNLOCKSTATE();
5541 	if (haslock) {
5542 		NFSLOCKV4ROOTMUTEX();
5543 		nfsv4_unlock(&nfsv4rootfs_lock, 1);
5544 		NFSUNLOCKV4ROOTMUTEX();
5545 	}
5546 
5547 out:
5548 	NFSEXITCODE(error);
5549 	return (error);
5550 }
5551 
5552 /*
5553  * Clear out all delegations for the file referred to by lfp.
5554  * May return NFSERR_DELAY, if there will be a delay waiting for
5555  * delegations to expire.
5556  * Returns -1 to indicate it slept while recalling a delegation.
5557  * This function has the side effect of deleting the nfslockfile structure,
5558  * if it no longer has associated state and didn't have to sleep.
5559  * Unlocks State before a non-zero value is returned.
5560  */
5561 static int
5562 nfsrv_cleandeleg(vnode_t vp, struct nfslockfile *lfp,
5563     struct nfsclient *clp, int *haslockp, NFSPROC_T *p)
5564 {
5565 	struct nfsstate *stp, *nstp;
5566 	int ret = 0;
5567 
5568 	stp = LIST_FIRST(&lfp->lf_deleg);
5569 	while (stp != LIST_END(&lfp->lf_deleg)) {
5570 		nstp = LIST_NEXT(stp, ls_file);
5571 		if (stp->ls_clp != clp) {
5572 			ret = nfsrv_delegconflict(stp, haslockp, p, vp);
5573 			if (ret) {
5574 				/*
5575 				 * nfsrv_delegconflict() unlocks state
5576 				 * when it returns non-zero.
5577 				 */
5578 				goto out;
5579 			}
5580 		}
5581 		stp = nstp;
5582 	}
5583 out:
5584 	NFSEXITCODE(ret);
5585 	return (ret);
5586 }
5587 
5588 /*
5589  * There are certain operations that, when being done outside of NFSv4,
5590  * require that any NFSv4 delegation for the file be recalled.
5591  * This function is to be called for those cases:
5592  * VOP_RENAME() - When a delegation is being recalled for any reason,
5593  *	the client may have to do Opens against the server, using the file's
5594  *	final component name. If the file has been renamed on the server,
5595  *	that component name will be incorrect and the Open will fail.
5596  * VOP_REMOVE() - Theoretically, a client could Open a file after it has
5597  *	been removed on the server, if there is a delegation issued to
5598  *	that client for the file. I say "theoretically" since clients
5599  *	normally do an Access Op before the Open and that Access Op will
5600  *	fail with ESTALE. Note that NFSv2 and 3 don't even do Opens, so
5601  *	they will detect the file's removal in the same manner. (There is
5602  *	one case where RFC3530 allows a client to do an Open without first
5603  *	doing an Access Op, which is passage of a check against the ACE
5604  *	returned with a Write delegation, but current practice is to ignore
5605  *	the ACE and always do an Access Op.)
5606  *	Since the functions can only be called with an unlocked vnode, this
5607  *	can't be done at this time.
5608  * VOP_ADVLOCK() - When a client holds a delegation, it can issue byte range
5609  *	locks locally in the client, which are not visible to the server. To
5610  *	deal with this, issuing of delegations for a vnode must be disabled
5611  *	and all delegations for the vnode recalled. This is done via the
5612  *	second function, using the VV_DISABLEDELEG vflag on the vnode.
5613  */
5614 void
5615 nfsd_recalldelegation(vnode_t vp, NFSPROC_T *p)
5616 {
5617 	time_t starttime;
5618 	int error;
5619 
5620 	/*
5621 	 * First, check to see if the server is currently running and it has
5622 	 * been called for a regular file when issuing delegations.
5623 	 */
5624 	if (NFSD_VNET(nfsrv_numnfsd) == 0 || vp->v_type != VREG ||
5625 	    nfsrv_issuedelegs == 0)
5626 		return;
5627 
5628 	KASSERT((NFSVOPISLOCKED(vp) != LK_EXCLUSIVE), ("vp %p is locked", vp));
5629 	/*
5630 	 * First, get a reference on the nfsv4rootfs_lock so that an
5631 	 * exclusive lock cannot be acquired by another thread.
5632 	 */
5633 	NFSLOCKV4ROOTMUTEX();
5634 	nfsv4_getref(&nfsv4rootfs_lock, NULL, NFSV4ROOTLOCKMUTEXPTR, NULL);
5635 	NFSUNLOCKV4ROOTMUTEX();
5636 
5637 	/*
5638 	 * Now, call nfsrv_checkremove() in a loop while it returns
5639 	 * NFSERR_DELAY. Return upon any other error or when timed out.
5640 	 */
5641 	starttime = NFSD_MONOSEC;
5642 	do {
5643 		if (NFSVOPLOCK(vp, LK_EXCLUSIVE) == 0) {
5644 			error = nfsrv_checkremove(vp, 0, NULL,
5645 			    (nfsquad_t)((u_quad_t)0), p);
5646 			NFSVOPUNLOCK(vp);
5647 		} else
5648 			error = EPERM;
5649 		if (error == NFSERR_DELAY) {
5650 			if (NFSD_MONOSEC - starttime > NFS_REMOVETIMEO)
5651 				break;
5652 			/* Sleep for a short period of time */
5653 			(void) nfs_catnap(PZERO, 0, "nfsremove");
5654 		}
5655 	} while (error == NFSERR_DELAY);
5656 	NFSLOCKV4ROOTMUTEX();
5657 	nfsv4_relref(&nfsv4rootfs_lock);
5658 	NFSUNLOCKV4ROOTMUTEX();
5659 }
5660 
5661 void
5662 nfsd_disabledelegation(vnode_t vp, NFSPROC_T *p)
5663 {
5664 
5665 #ifdef VV_DISABLEDELEG
5666 	/*
5667 	 * First, flag issuance of delegations disabled.
5668 	 */
5669 	atomic_set_long(&vp->v_vflag, VV_DISABLEDELEG);
5670 #endif
5671 
5672 	/*
5673 	 * Then call nfsd_recalldelegation() to get rid of all extant
5674 	 * delegations.
5675 	 */
5676 	nfsd_recalldelegation(vp, p);
5677 }
5678 
5679 /*
5680  * Check for conflicting locks, etc. and then get rid of delegations.
5681  * (At one point I thought that I should get rid of delegations for any
5682  *  Setattr, since it could potentially disallow the I/O op (read or write)
5683  *  allowed by the delegation. However, Setattr Ops that aren't changing
5684  *  the size get a stateid of all 0s, so you can't tell if it is a delegation
5685  *  for the same client or a different one, so I decided to only get rid
5686  *  of delegations for other clients when the size is being changed.)
5687  * In general, a Setattr can disable NFS I/O Ops that are outstanding, such
5688  * as Write backs, even if there is no delegation, so it really isn't any
5689  * different?)
5690  */
5691 int
5692 nfsrv_checksetattr(vnode_t vp, struct nfsrv_descript *nd,
5693     nfsv4stateid_t *stateidp, struct nfsvattr *nvap, nfsattrbit_t *attrbitp,
5694     struct nfsexstuff *exp, NFSPROC_T *p)
5695 {
5696 	struct nfsstate st, *stp = &st;
5697 	struct nfslock lo, *lop = &lo;
5698 	int error = 0;
5699 	nfsquad_t clientid;
5700 
5701 	if (NFSISSET_ATTRBIT(attrbitp, NFSATTRBIT_SIZE)) {
5702 		stp->ls_flags = (NFSLCK_CHECK | NFSLCK_WRITEACCESS);
5703 		lop->lo_first = nvap->na_size;
5704 	} else {
5705 		stp->ls_flags = 0;
5706 		lop->lo_first = 0;
5707 	}
5708 	if (NFSISSET_ATTRBIT(attrbitp, NFSATTRBIT_OWNER) ||
5709 	    NFSISSET_ATTRBIT(attrbitp, NFSATTRBIT_OWNERGROUP) ||
5710 	    NFSISSET_ATTRBIT(attrbitp, NFSATTRBIT_MODE) ||
5711 	    NFSISSET_ATTRBIT(attrbitp, NFSATTRBIT_ACL))
5712 		stp->ls_flags |= NFSLCK_SETATTR;
5713 	if (stp->ls_flags == 0)
5714 		goto out;
5715 	lop->lo_end = NFS64BITSSET;
5716 	lop->lo_flags = NFSLCK_WRITE;
5717 	stp->ls_ownerlen = 0;
5718 	stp->ls_op = NULL;
5719 	stp->ls_uid = nd->nd_cred->cr_uid;
5720 	stp->ls_stateid.seqid = stateidp->seqid;
5721 	clientid.lval[0] = stp->ls_stateid.other[0] = stateidp->other[0];
5722 	clientid.lval[1] = stp->ls_stateid.other[1] = stateidp->other[1];
5723 	stp->ls_stateid.other[2] = stateidp->other[2];
5724 	error = nfsrv_lockctrl(vp, &stp, &lop, NULL, clientid,
5725 	    stateidp, exp, nd, p);
5726 
5727 out:
5728 	NFSEXITCODE2(error, nd);
5729 	return (error);
5730 }
5731 
5732 /*
5733  * Check for a write delegation and do a CBGETATTR if there is one, updating
5734  * the attributes, as required.
5735  * Should I return an error if I can't get the attributes? (For now, I'll
5736  * just return ok.
5737  */
5738 int
5739 nfsrv_checkgetattr(struct nfsrv_descript *nd, vnode_t vp,
5740     struct nfsvattr *nvap, nfsattrbit_t *attrbitp, NFSPROC_T *p)
5741 {
5742 	struct nfsstate *stp;
5743 	struct nfslockfile *lfp;
5744 	struct nfsclient *clp;
5745 	struct nfsvattr nva;
5746 	fhandle_t nfh;
5747 	int error = 0;
5748 	nfsattrbit_t cbbits;
5749 	u_quad_t delegfilerev;
5750 
5751 	NFSCBGETATTR_ATTRBIT(attrbitp, &cbbits);
5752 	if (!NFSNONZERO_ATTRBIT(&cbbits))
5753 		goto out;
5754 	if (nfsrv_writedelegcnt == 0)
5755 		goto out;
5756 
5757 	/*
5758 	 * Get the lock file structure.
5759 	 * (A return of -1 means no associated state, so return ok.)
5760 	 */
5761 	error = nfsrv_getlockfh(vp, NFSLCK_CHECK, NULL, &nfh, p);
5762 	NFSLOCKSTATE();
5763 	if (!error)
5764 		error = nfsrv_getlockfile(NFSLCK_CHECK, NULL, &lfp, &nfh, 0);
5765 	if (error) {
5766 		NFSUNLOCKSTATE();
5767 		if (error == -1)
5768 			error = 0;
5769 		goto out;
5770 	}
5771 
5772 	/*
5773 	 * Now, look for a write delegation.
5774 	 */
5775 	LIST_FOREACH(stp, &lfp->lf_deleg, ls_file) {
5776 		if (stp->ls_flags & NFSLCK_DELEGWRITE)
5777 			break;
5778 	}
5779 	if (stp == LIST_END(&lfp->lf_deleg)) {
5780 		NFSUNLOCKSTATE();
5781 		goto out;
5782 	}
5783 	clp = stp->ls_clp;
5784 
5785 	/* If the clientid is not confirmed, ignore the delegation. */
5786 	if (clp->lc_flags & LCL_NEEDSCONFIRM) {
5787 		NFSUNLOCKSTATE();
5788 		goto out;
5789 	}
5790 
5791 	delegfilerev = stp->ls_filerev;
5792 	/*
5793 	 * If the Write delegation was issued as a part of this Compound RPC
5794 	 * or if we have an Implied Clientid (used in a previous Op in this
5795 	 * compound) and it is the client the delegation was issued to,
5796 	 * just return ok.
5797 	 * I also assume that it is from the same client iff the network
5798 	 * host IP address is the same as the callback address. (Not
5799 	 * exactly correct by the RFC, but avoids a lot of Getattr
5800 	 * callbacks.)
5801 	 */
5802 	if (nd->nd_compref == stp->ls_compref ||
5803 	    ((nd->nd_flag & ND_IMPLIEDCLID) &&
5804 	     clp->lc_clientid.qval == nd->nd_clientid.qval) ||
5805 	     nfsaddr2_match(clp->lc_req.nr_nam, nd->nd_nam)) {
5806 		NFSUNLOCKSTATE();
5807 		goto out;
5808 	}
5809 
5810 	/*
5811 	 * We are now done with the delegation state structure,
5812 	 * so the statelock can be released and we can now tsleep().
5813 	 */
5814 
5815 	/*
5816 	 * Now, we must do the CB Getattr callback, to see if Change or Size
5817 	 * has changed.
5818 	 */
5819 	if (clp->lc_expiry >= NFSD_MONOSEC) {
5820 		NFSUNLOCKSTATE();
5821 		NFSVNO_ATTRINIT(&nva);
5822 		nva.na_filerev = NFS64BITSSET;
5823 		error = nfsrv_docallback(clp, NFSV4OP_CBGETATTR, NULL,
5824 		    0, &nfh, &nva, &cbbits, 0, p);
5825 		if (!error) {
5826 			if ((nva.na_filerev != NFS64BITSSET &&
5827 			    nva.na_filerev > delegfilerev) ||
5828 			    (NFSVNO_ISSETSIZE(&nva) &&
5829 			     nva.na_size != nvap->na_size)) {
5830 				error = nfsvno_updfilerev(vp, nvap, nd, p);
5831 				if (NFSVNO_ISSETSIZE(&nva))
5832 					nvap->na_size = nva.na_size;
5833 			}
5834 		} else
5835 			error = 0;	/* Ignore callback errors for now. */
5836 	} else {
5837 		NFSUNLOCKSTATE();
5838 	}
5839 
5840 out:
5841 	NFSEXITCODE2(error, nd);
5842 	return (error);
5843 }
5844 
5845 /*
5846  * This function looks for openowners that haven't had any opens for
5847  * a while and throws them away. Called by an nfsd when NFSNSF_NOOPENS
5848  * is set.
5849  */
5850 void
5851 nfsrv_throwawayopens(NFSPROC_T *p)
5852 {
5853 	struct nfsclient *clp, *nclp;
5854 	struct nfsstate *stp, *nstp;
5855 	int i;
5856 
5857 	NFSLOCKSTATE();
5858 	NFSD_VNET(nfsrv_stablefirst).nsf_flags &= ~NFSNSF_NOOPENS;
5859 	/*
5860 	 * For each client...
5861 	 */
5862 	for (i = 0; i < nfsrv_clienthashsize; i++) {
5863 	    LIST_FOREACH_SAFE(clp, &NFSD_VNET(nfsclienthash)[i], lc_hash,
5864 		nclp) {
5865 		LIST_FOREACH_SAFE(stp, &clp->lc_open, ls_list, nstp) {
5866 			if (LIST_EMPTY(&stp->ls_open) &&
5867 			    (stp->ls_noopens > NFSNOOPEN ||
5868 			     (nfsrv_openpluslock * 2) >
5869 			     nfsrv_v4statelimit))
5870 				nfsrv_freeopenowner(stp, 0, p);
5871 		}
5872 	    }
5873 	}
5874 	NFSUNLOCKSTATE();
5875 }
5876 
5877 /*
5878  * This function checks to see if the credentials are the same.
5879  * The check for same credentials is needed for state management operations
5880  * for NFSv4.0 where 1 is returned if not same, 0 is returned otherwise.
5881  */
5882 static int
5883 nfsrv_notsamecredname(struct nfsrv_descript *nd, struct nfsclient *clp)
5884 {
5885 
5886 	/* For NFSv4.1/4.2, SP4_NONE always allows this. */
5887 	if ((nd->nd_flag & ND_NFSV41) != 0)
5888 		return (0);
5889 
5890 	if (nd->nd_flag & ND_GSS) {
5891 		if (!(clp->lc_flags & LCL_GSS))
5892 			return (1);
5893 		if (clp->lc_flags & LCL_NAME) {
5894 			if (nd->nd_princlen != clp->lc_namelen ||
5895 			    NFSBCMP(nd->nd_principal, clp->lc_name,
5896 				clp->lc_namelen))
5897 				return (1);
5898 			else
5899 				return (0);
5900 		}
5901 		if (nd->nd_cred->cr_uid == clp->lc_uid)
5902 			return (0);
5903 		else
5904 			return (1);
5905 	} else if (clp->lc_flags & LCL_GSS)
5906 		return (1);
5907 	/*
5908 	 * For AUTH_SYS, allow the same uid or root. (This is underspecified
5909 	 * in RFC3530, which talks about principals, but doesn't say anything
5910 	 * about uids for AUTH_SYS.)
5911 	 */
5912 	if (nd->nd_cred->cr_uid == clp->lc_uid || nd->nd_cred->cr_uid == 0)
5913 		return (0);
5914 	else
5915 		return (1);
5916 }
5917 
5918 /*
5919  * Calculate the lease expiry time.
5920  */
5921 static time_t
5922 nfsrv_leaseexpiry(void)
5923 {
5924 
5925 	if (NFSD_VNET(nfsrv_stablefirst).nsf_eograce > NFSD_MONOSEC)
5926 		return (NFSD_MONOSEC + 2 * (nfsrv_lease + NFSRV_LEASEDELTA));
5927 	return (NFSD_MONOSEC + nfsrv_lease + NFSRV_LEASEDELTA);
5928 }
5929 
5930 /*
5931  * Delay the delegation timeout as far as ls_delegtimelimit, as required.
5932  */
5933 static void
5934 nfsrv_delaydelegtimeout(struct nfsstate *stp)
5935 {
5936 
5937 	if ((stp->ls_flags & NFSLCK_DELEGRECALL) == 0)
5938 		return;
5939 
5940 	if ((stp->ls_delegtime + 15) > NFSD_MONOSEC &&
5941 	    stp->ls_delegtime < stp->ls_delegtimelimit) {
5942 		stp->ls_delegtime += nfsrv_lease;
5943 		if (stp->ls_delegtime > stp->ls_delegtimelimit)
5944 			stp->ls_delegtime = stp->ls_delegtimelimit;
5945 	}
5946 }
5947 
5948 /*
5949  * This function checks to see if there is any other state associated
5950  * with the openowner for this Open.
5951  * It returns 1 if there is no other state, 0 otherwise.
5952  */
5953 static int
5954 nfsrv_nootherstate(struct nfsstate *stp)
5955 {
5956 	struct nfsstate *tstp;
5957 
5958 	LIST_FOREACH(tstp, &stp->ls_openowner->ls_open, ls_list) {
5959 		if (tstp != stp || !LIST_EMPTY(&tstp->ls_lock))
5960 			return (0);
5961 	}
5962 	return (1);
5963 }
5964 
5965 /*
5966  * Create a list of lock deltas (changes to local byte range locking
5967  * that can be rolled back using the list) and apply the changes via
5968  * nfsvno_advlock(). Optionally, lock the list. It is expected that either
5969  * the rollback or update function will be called after this.
5970  * It returns an error (and rolls back, as required), if any nfsvno_advlock()
5971  * call fails. If it returns an error, it will unlock the list.
5972  */
5973 static int
5974 nfsrv_locallock(vnode_t vp, struct nfslockfile *lfp, int flags,
5975     uint64_t first, uint64_t end, struct nfslockconflict *cfp, NFSPROC_T *p)
5976 {
5977 	struct nfslock *lop, *nlop;
5978 	int error = 0;
5979 
5980 	/* Loop through the list of locks. */
5981 	lop = LIST_FIRST(&lfp->lf_locallock);
5982 	while (first < end && lop != NULL) {
5983 		nlop = LIST_NEXT(lop, lo_lckowner);
5984 		if (first >= lop->lo_end) {
5985 			/* not there yet */
5986 			lop = nlop;
5987 		} else if (first < lop->lo_first) {
5988 			/* new one starts before entry in list */
5989 			if (end <= lop->lo_first) {
5990 				/* no overlap between old and new */
5991 				error = nfsrv_dolocal(vp, lfp, flags,
5992 				    NFSLCK_UNLOCK, first, end, cfp, p);
5993 				if (error != 0)
5994 					break;
5995 				first = end;
5996 			} else {
5997 				/* handle fragment overlapped with new one */
5998 				error = nfsrv_dolocal(vp, lfp, flags,
5999 				    NFSLCK_UNLOCK, first, lop->lo_first, cfp,
6000 				    p);
6001 				if (error != 0)
6002 					break;
6003 				first = lop->lo_first;
6004 			}
6005 		} else {
6006 			/* new one overlaps this entry in list */
6007 			if (end <= lop->lo_end) {
6008 				/* overlaps all of new one */
6009 				error = nfsrv_dolocal(vp, lfp, flags,
6010 				    lop->lo_flags, first, end, cfp, p);
6011 				if (error != 0)
6012 					break;
6013 				first = end;
6014 			} else {
6015 				/* handle fragment overlapped with new one */
6016 				error = nfsrv_dolocal(vp, lfp, flags,
6017 				    lop->lo_flags, first, lop->lo_end, cfp, p);
6018 				if (error != 0)
6019 					break;
6020 				first = lop->lo_end;
6021 				lop = nlop;
6022 			}
6023 		}
6024 	}
6025 	if (first < end && error == 0)
6026 		/* handle fragment past end of list */
6027 		error = nfsrv_dolocal(vp, lfp, flags, NFSLCK_UNLOCK, first,
6028 		    end, cfp, p);
6029 
6030 	NFSEXITCODE(error);
6031 	return (error);
6032 }
6033 
6034 /*
6035  * Local lock unlock. Unlock all byte ranges that are no longer locked
6036  * by NFSv4. To do this, unlock any subranges of first-->end that
6037  * do not overlap with the byte ranges of any lock in the lfp->lf_lock
6038  * list. This list has all locks for the file held by other
6039  * <clientid, lockowner> tuples. The list is ordered by increasing
6040  * lo_first value, but may have entries that overlap each other, for
6041  * the case of read locks.
6042  */
6043 static void
6044 nfsrv_localunlock(vnode_t vp, struct nfslockfile *lfp, uint64_t init_first,
6045     uint64_t init_end, NFSPROC_T *p)
6046 {
6047 	struct nfslock *lop;
6048 	uint64_t first, end, prevfirst __unused;
6049 
6050 	first = init_first;
6051 	end = init_end;
6052 	while (first < init_end) {
6053 		/* Loop through all nfs locks, adjusting first and end */
6054 		prevfirst = 0;
6055 		LIST_FOREACH(lop, &lfp->lf_lock, lo_lckfile) {
6056 			KASSERT(prevfirst <= lop->lo_first,
6057 			    ("nfsv4 locks out of order"));
6058 			KASSERT(lop->lo_first < lop->lo_end,
6059 			    ("nfsv4 bogus lock"));
6060 			prevfirst = lop->lo_first;
6061 			if (first >= lop->lo_first &&
6062 			    first < lop->lo_end)
6063 				/*
6064 				 * Overlaps with initial part, so trim
6065 				 * off that initial part by moving first past
6066 				 * it.
6067 				 */
6068 				first = lop->lo_end;
6069 			else if (end > lop->lo_first &&
6070 			    lop->lo_first > first) {
6071 				/*
6072 				 * This lock defines the end of the
6073 				 * segment to unlock, so set end to the
6074 				 * start of it and break out of the loop.
6075 				 */
6076 				end = lop->lo_first;
6077 				break;
6078 			}
6079 			if (first >= end)
6080 				/*
6081 				 * There is no segment left to do, so
6082 				 * break out of this loop and then exit
6083 				 * the outer while() since first will be set
6084 				 * to end, which must equal init_end here.
6085 				 */
6086 				break;
6087 		}
6088 		if (first < end) {
6089 			/* Unlock this segment */
6090 			(void) nfsrv_dolocal(vp, lfp, NFSLCK_UNLOCK,
6091 			    NFSLCK_READ, first, end, NULL, p);
6092 			nfsrv_locallock_commit(lfp, NFSLCK_UNLOCK,
6093 			    first, end);
6094 		}
6095 		/*
6096 		 * Now move past this segment and look for any further
6097 		 * segment in the range, if there is one.
6098 		 */
6099 		first = end;
6100 		end = init_end;
6101 	}
6102 }
6103 
6104 /*
6105  * Do the local lock operation and update the rollback list, as required.
6106  * Perform the rollback and return the error if nfsvno_advlock() fails.
6107  */
6108 static int
6109 nfsrv_dolocal(vnode_t vp, struct nfslockfile *lfp, int flags, int oldflags,
6110     uint64_t first, uint64_t end, struct nfslockconflict *cfp, NFSPROC_T *p)
6111 {
6112 	struct nfsrollback *rlp;
6113 	int error = 0, ltype, oldltype;
6114 
6115 	if (flags & NFSLCK_WRITE)
6116 		ltype = F_WRLCK;
6117 	else if (flags & NFSLCK_READ)
6118 		ltype = F_RDLCK;
6119 	else
6120 		ltype = F_UNLCK;
6121 	if (oldflags & NFSLCK_WRITE)
6122 		oldltype = F_WRLCK;
6123 	else if (oldflags & NFSLCK_READ)
6124 		oldltype = F_RDLCK;
6125 	else
6126 		oldltype = F_UNLCK;
6127 	if (ltype == oldltype || (oldltype == F_WRLCK && ltype == F_RDLCK))
6128 		/* nothing to do */
6129 		goto out;
6130 	error = nfsvno_advlock(vp, ltype, first, end, p);
6131 	if (error != 0) {
6132 		if (cfp != NULL) {
6133 			cfp->cl_clientid.lval[0] = 0;
6134 			cfp->cl_clientid.lval[1] = 0;
6135 			cfp->cl_first = 0;
6136 			cfp->cl_end = NFS64BITSSET;
6137 			cfp->cl_flags = NFSLCK_WRITE;
6138 			cfp->cl_ownerlen = 5;
6139 			NFSBCOPY("LOCAL", cfp->cl_owner, 5);
6140 		}
6141 		nfsrv_locallock_rollback(vp, lfp, p);
6142 	} else if (ltype != F_UNLCK) {
6143 		rlp = malloc(sizeof (struct nfsrollback), M_NFSDROLLBACK,
6144 		    M_WAITOK);
6145 		rlp->rlck_first = first;
6146 		rlp->rlck_end = end;
6147 		rlp->rlck_type = oldltype;
6148 		LIST_INSERT_HEAD(&lfp->lf_rollback, rlp, rlck_list);
6149 	}
6150 
6151 out:
6152 	NFSEXITCODE(error);
6153 	return (error);
6154 }
6155 
6156 /*
6157  * Roll back local lock changes and free up the rollback list.
6158  */
6159 static void
6160 nfsrv_locallock_rollback(vnode_t vp, struct nfslockfile *lfp, NFSPROC_T *p)
6161 {
6162 	struct nfsrollback *rlp, *nrlp;
6163 
6164 	LIST_FOREACH_SAFE(rlp, &lfp->lf_rollback, rlck_list, nrlp) {
6165 		(void) nfsvno_advlock(vp, rlp->rlck_type, rlp->rlck_first,
6166 		    rlp->rlck_end, p);
6167 		free(rlp, M_NFSDROLLBACK);
6168 	}
6169 	LIST_INIT(&lfp->lf_rollback);
6170 }
6171 
6172 /*
6173  * Update local lock list and delete rollback list (ie now committed to the
6174  * local locks). Most of the work is done by the internal function.
6175  */
6176 static void
6177 nfsrv_locallock_commit(struct nfslockfile *lfp, int flags, uint64_t first,
6178     uint64_t end)
6179 {
6180 	struct nfsrollback *rlp, *nrlp;
6181 	struct nfslock *new_lop, *other_lop;
6182 
6183 	new_lop = malloc(sizeof (struct nfslock), M_NFSDLOCK, M_WAITOK);
6184 	if (flags & (NFSLCK_READ | NFSLCK_WRITE))
6185 		other_lop = malloc(sizeof (struct nfslock), M_NFSDLOCK,
6186 		    M_WAITOK);
6187 	else
6188 		other_lop = NULL;
6189 	new_lop->lo_flags = flags;
6190 	new_lop->lo_first = first;
6191 	new_lop->lo_end = end;
6192 	nfsrv_updatelock(NULL, &new_lop, &other_lop, lfp);
6193 	if (new_lop != NULL)
6194 		free(new_lop, M_NFSDLOCK);
6195 	if (other_lop != NULL)
6196 		free(other_lop, M_NFSDLOCK);
6197 
6198 	/* and get rid of the rollback list */
6199 	LIST_FOREACH_SAFE(rlp, &lfp->lf_rollback, rlck_list, nrlp)
6200 		free(rlp, M_NFSDROLLBACK);
6201 	LIST_INIT(&lfp->lf_rollback);
6202 }
6203 
6204 /*
6205  * Lock the struct nfslockfile for local lock updating.
6206  */
6207 static void
6208 nfsrv_locklf(struct nfslockfile *lfp)
6209 {
6210 	int gotlock;
6211 
6212 	/* lf_usecount ensures *lfp won't be free'd */
6213 	lfp->lf_usecount++;
6214 	do {
6215 		gotlock = nfsv4_lock(&lfp->lf_locallock_lck, 1, NULL,
6216 		    NFSSTATEMUTEXPTR, NULL);
6217 	} while (gotlock == 0);
6218 	lfp->lf_usecount--;
6219 }
6220 
6221 /*
6222  * Unlock the struct nfslockfile after local lock updating.
6223  */
6224 static void
6225 nfsrv_unlocklf(struct nfslockfile *lfp)
6226 {
6227 
6228 	nfsv4_unlock(&lfp->lf_locallock_lck, 0);
6229 }
6230 
6231 /*
6232  * Clear out all state for the NFSv4 server.
6233  * Must be called by a thread that can sleep when no nfsds are running.
6234  */
6235 void
6236 nfsrv_throwawayallstate(NFSPROC_T *p)
6237 {
6238 	struct nfsclient *clp, *nclp;
6239 	struct nfslockfile *lfp, *nlfp;
6240 	int i;
6241 
6242 	/*
6243 	 * For each client, clean out the state and then free the structure.
6244 	 */
6245 	for (i = 0; i < nfsrv_clienthashsize; i++) {
6246 		LIST_FOREACH_SAFE(clp, &NFSD_VNET(nfsclienthash)[i], lc_hash,
6247 		    nclp) {
6248 			nfsrv_cleanclient(clp, p);
6249 			nfsrv_freedeleglist(&clp->lc_deleg);
6250 			nfsrv_freedeleglist(&clp->lc_olddeleg);
6251 			free(clp->lc_stateid, M_NFSDCLIENT);
6252 			free(clp, M_NFSDCLIENT);
6253 		}
6254 	}
6255 
6256 	/*
6257 	 * Also, free up any remaining lock file structures.
6258 	 */
6259 	for (i = 0; i < nfsrv_lockhashsize; i++) {
6260 		LIST_FOREACH_SAFE(lfp, &NFSD_VNET(nfslockhash)[i], lf_hash,
6261 		    nlfp) {
6262 			printf("nfsd unload: fnd a lock file struct\n");
6263 			nfsrv_freenfslockfile(lfp);
6264 		}
6265 	}
6266 
6267 	/* And get rid of the deviceid structures and layouts. */
6268 	nfsrv_freealllayoutsanddevids();
6269 }
6270 
6271 /*
6272  * Check the sequence# for the session and slot provided as an argument.
6273  * Also, renew the lease if the session will return NFS_OK.
6274  */
6275 int
6276 nfsrv_checksequence(struct nfsrv_descript *nd, uint32_t sequenceid,
6277     uint32_t *highest_slotidp, uint32_t *target_highest_slotidp, int cache_this,
6278     uint32_t *sflagsp, NFSPROC_T *p)
6279 {
6280 	struct nfsdsession *sep;
6281 	struct nfssessionhash *shp;
6282 	int error;
6283 
6284 	shp = NFSSESSIONHASH(nd->nd_sessionid);
6285 	NFSLOCKSESSION(shp);
6286 	sep = nfsrv_findsession(nd->nd_sessionid);
6287 	if (sep == NULL) {
6288 		NFSUNLOCKSESSION(shp);
6289 		return (NFSERR_BADSESSION);
6290 	}
6291 	error = nfsv4_seqsession(sequenceid, nd->nd_slotid, *highest_slotidp,
6292 	    sep->sess_slots, NULL, NFSV4_SLOTS - 1);
6293 	if (error != 0) {
6294 		NFSUNLOCKSESSION(shp);
6295 		return (error);
6296 	}
6297 	if (cache_this != 0)
6298 		nd->nd_flag |= ND_SAVEREPLY;
6299 	/* Renew the lease. */
6300 	sep->sess_clp->lc_expiry = nfsrv_leaseexpiry();
6301 	nd->nd_clientid.qval = sep->sess_clp->lc_clientid.qval;
6302 	nd->nd_flag |= ND_IMPLIEDCLID;
6303 
6304 	/* Save maximum request and reply sizes. */
6305 	nd->nd_maxreq = sep->sess_maxreq;
6306 	nd->nd_maxresp = sep->sess_maxresp;
6307 
6308 	*sflagsp = 0;
6309 	if (sep->sess_clp->lc_req.nr_client == NULL ||
6310 	    (sep->sess_clp->lc_flags & LCL_CBDOWN) != 0)
6311 		*sflagsp |= NFSV4SEQ_CBPATHDOWN;
6312 	NFSUNLOCKSESSION(shp);
6313 	if (error == NFSERR_EXPIRED) {
6314 		*sflagsp |= NFSV4SEQ_EXPIREDALLSTATEREVOKED;
6315 		error = 0;
6316 	} else if (error == NFSERR_ADMINREVOKED) {
6317 		*sflagsp |= NFSV4SEQ_ADMINSTATEREVOKED;
6318 		error = 0;
6319 	}
6320 	*highest_slotidp = *target_highest_slotidp = NFSV4_SLOTS - 1;
6321 	return (0);
6322 }
6323 
6324 /*
6325  * Check/set reclaim complete for this session/clientid.
6326  */
6327 int
6328 nfsrv_checkreclaimcomplete(struct nfsrv_descript *nd, int onefs)
6329 {
6330 	struct nfsdsession *sep;
6331 	struct nfssessionhash *shp;
6332 	int error = 0;
6333 
6334 	shp = NFSSESSIONHASH(nd->nd_sessionid);
6335 	NFSLOCKSTATE();
6336 	NFSLOCKSESSION(shp);
6337 	sep = nfsrv_findsession(nd->nd_sessionid);
6338 	if (sep == NULL) {
6339 		NFSUNLOCKSESSION(shp);
6340 		NFSUNLOCKSTATE();
6341 		return (NFSERR_BADSESSION);
6342 	}
6343 
6344 	if (onefs != 0)
6345 		sep->sess_clp->lc_flags |= LCL_RECLAIMONEFS;
6346 		/* Check to see if reclaim complete has already happened. */
6347 	else if ((sep->sess_clp->lc_flags & LCL_RECLAIMCOMPLETE) != 0)
6348 		error = NFSERR_COMPLETEALREADY;
6349 	else {
6350 		sep->sess_clp->lc_flags |= LCL_RECLAIMCOMPLETE;
6351 		nfsrv_markreclaim(sep->sess_clp);
6352 	}
6353 	NFSUNLOCKSESSION(shp);
6354 	NFSUNLOCKSTATE();
6355 	return (error);
6356 }
6357 
6358 /*
6359  * Cache the reply in a session slot.
6360  */
6361 void
6362 nfsrv_cache_session(struct nfsrv_descript *nd, struct mbuf **m)
6363 {
6364 	struct nfsdsession *sep;
6365 	struct nfssessionhash *shp;
6366 	char *buf, *cp;
6367 #ifdef INET
6368 	struct sockaddr_in *sin;
6369 #endif
6370 #ifdef INET6
6371 	struct sockaddr_in6 *sin6;
6372 #endif
6373 
6374 	shp = NFSSESSIONHASH(nd->nd_sessionid);
6375 	NFSLOCKSESSION(shp);
6376 	sep = nfsrv_findsession(nd->nd_sessionid);
6377 	if (sep == NULL) {
6378 		NFSUNLOCKSESSION(shp);
6379 		if ((NFSD_VNET(nfsrv_stablefirst).nsf_flags &
6380 		     NFSNSF_GRACEOVER) != 0) {
6381 			buf = malloc(INET6_ADDRSTRLEN, M_TEMP, M_WAITOK);
6382 			switch (nd->nd_nam->sa_family) {
6383 #ifdef INET
6384 			case AF_INET:
6385 				sin = (struct sockaddr_in *)nd->nd_nam;
6386 				cp = inet_ntop(sin->sin_family,
6387 				    &sin->sin_addr.s_addr, buf,
6388 				    INET6_ADDRSTRLEN);
6389 				break;
6390 #endif
6391 #ifdef INET6
6392 			case AF_INET6:
6393 				sin6 = (struct sockaddr_in6 *)nd->nd_nam;
6394 				cp = inet_ntop(sin6->sin6_family,
6395 				    &sin6->sin6_addr, buf, INET6_ADDRSTRLEN);
6396 				break;
6397 #endif
6398 			default:
6399 				cp = NULL;
6400 			}
6401 			if (cp != NULL)
6402 				printf("nfsrv_cache_session: no session "
6403 				    "IPaddr=%s, check NFS clients for unique "
6404 				    "/etc/hostid's\n", cp);
6405 			else
6406 				printf("nfsrv_cache_session: no session, "
6407 				    "check NFS clients for unique "
6408 				    "/etc/hostid's\n");
6409 			free(buf, M_TEMP);
6410 		}
6411 		m_freem(*m);
6412 		return;
6413 	}
6414 	nfsv4_seqsess_cacherep(nd->nd_slotid, sep->sess_slots, nd->nd_repstat,
6415 	    m);
6416 	NFSUNLOCKSESSION(shp);
6417 }
6418 
6419 /*
6420  * Search for a session that matches the sessionid.
6421  */
6422 static struct nfsdsession *
6423 nfsrv_findsession(uint8_t *sessionid)
6424 {
6425 	struct nfsdsession *sep;
6426 	struct nfssessionhash *shp;
6427 
6428 	shp = NFSSESSIONHASH(sessionid);
6429 	LIST_FOREACH(sep, &shp->list, sess_hash) {
6430 		if (!NFSBCMP(sessionid, sep->sess_sessionid, NFSX_V4SESSIONID))
6431 			break;
6432 	}
6433 	return (sep);
6434 }
6435 
6436 /*
6437  * Destroy a session.
6438  */
6439 int
6440 nfsrv_destroysession(struct nfsrv_descript *nd, uint8_t *sessionid)
6441 {
6442 	int error, igotlock, samesess;
6443 
6444 	samesess = 0;
6445 	if (!NFSBCMP(sessionid, nd->nd_sessionid, NFSX_V4SESSIONID) &&
6446 	    (nd->nd_flag & ND_HASSEQUENCE) != 0) {
6447 		samesess = 1;
6448 		if ((nd->nd_flag & ND_LASTOP) == 0)
6449 			return (NFSERR_BADSESSION);
6450 	}
6451 
6452 	/* Lock out other nfsd threads */
6453 	NFSLOCKV4ROOTMUTEX();
6454 	nfsv4_relref(&nfsv4rootfs_lock);
6455 	do {
6456 		igotlock = nfsv4_lock(&nfsv4rootfs_lock, 1, NULL,
6457 		    NFSV4ROOTLOCKMUTEXPTR, NULL);
6458 	} while (igotlock == 0);
6459 	NFSUNLOCKV4ROOTMUTEX();
6460 
6461 	error = nfsrv_freesession(NULL, sessionid);
6462 	if (error == 0 && samesess != 0)
6463 		nd->nd_flag &= ~ND_HASSEQUENCE;
6464 
6465 	NFSLOCKV4ROOTMUTEX();
6466 	nfsv4_unlock(&nfsv4rootfs_lock, 1);
6467 	NFSUNLOCKV4ROOTMUTEX();
6468 	return (error);
6469 }
6470 
6471 /*
6472  * Bind a connection to a session.
6473  * For now, only certain variants are supported, since the current session
6474  * structure can only handle a single backchannel entry, which will be
6475  * applied to all connections if it is set.
6476  */
6477 int
6478 nfsrv_bindconnsess(struct nfsrv_descript *nd, uint8_t *sessionid, int *foreaftp)
6479 {
6480 	struct nfssessionhash *shp;
6481 	struct nfsdsession *sep;
6482 	struct nfsclient *clp;
6483 	SVCXPRT *savxprt;
6484 	int error;
6485 
6486 	error = 0;
6487 	savxprt = NULL;
6488 	shp = NFSSESSIONHASH(sessionid);
6489 	NFSLOCKSTATE();
6490 	NFSLOCKSESSION(shp);
6491 	sep = nfsrv_findsession(sessionid);
6492 	if (sep != NULL) {
6493 		clp = sep->sess_clp;
6494 		if (*foreaftp == NFSCDFC4_BACK ||
6495 		    *foreaftp == NFSCDFC4_BACK_OR_BOTH ||
6496 		    *foreaftp == NFSCDFC4_FORE_OR_BOTH) {
6497 			/* Try to set up a backchannel. */
6498 			if (clp->lc_req.nr_client == NULL) {
6499 				NFSD_DEBUG(2, "nfsrv_bindconnsess: acquire "
6500 				    "backchannel\n");
6501 				clp->lc_req.nr_client = (struct __rpc_client *)
6502 				    clnt_bck_create(nd->nd_xprt->xp_socket,
6503 				    sep->sess_cbprogram, NFSV4_CBVERS);
6504 			}
6505 			if (clp->lc_req.nr_client != NULL) {
6506 				NFSD_DEBUG(2, "nfsrv_bindconnsess: set up "
6507 				    "backchannel\n");
6508 				savxprt = sep->sess_cbsess.nfsess_xprt;
6509 				SVC_ACQUIRE(nd->nd_xprt);
6510 				CLNT_ACQUIRE(clp->lc_req.nr_client);
6511 				nd->nd_xprt->xp_p2 = clp->lc_req.nr_client;
6512 				/* Disable idle timeout. */
6513 				nd->nd_xprt->xp_idletimeout = 0;
6514 				sep->sess_cbsess.nfsess_xprt = nd->nd_xprt;
6515 				sep->sess_crflags |= NFSV4CRSESS_CONNBACKCHAN;
6516 				clp->lc_flags |= LCL_DONEBINDCONN |
6517 				    LCL_NEEDSCBNULL;
6518 				clp->lc_flags &= ~LCL_CBDOWN;
6519 				if (*foreaftp == NFSCDFS4_BACK)
6520 					*foreaftp = NFSCDFS4_BACK;
6521 				else
6522 					*foreaftp = NFSCDFS4_BOTH;
6523 			} else if (*foreaftp != NFSCDFC4_BACK) {
6524 				NFSD_DEBUG(2, "nfsrv_bindconnsess: can't set "
6525 				    "up backchannel\n");
6526 				sep->sess_crflags &= ~NFSV4CRSESS_CONNBACKCHAN;
6527 				clp->lc_flags |= LCL_DONEBINDCONN;
6528 				*foreaftp = NFSCDFS4_FORE;
6529 			} else {
6530 				error = NFSERR_NOTSUPP;
6531 				printf("nfsrv_bindconnsess: Can't add "
6532 				    "backchannel\n");
6533 			}
6534 		} else {
6535 			NFSD_DEBUG(2, "nfsrv_bindconnsess: Set forechannel\n");
6536 			clp->lc_flags |= LCL_DONEBINDCONN;
6537 			*foreaftp = NFSCDFS4_FORE;
6538 		}
6539 	} else
6540 		error = NFSERR_BADSESSION;
6541 	NFSUNLOCKSESSION(shp);
6542 	NFSUNLOCKSTATE();
6543 	if (savxprt != NULL)
6544 		SVC_RELEASE(savxprt);
6545 	return (error);
6546 }
6547 
6548 /*
6549  * Free up a session structure.
6550  */
6551 static int
6552 nfsrv_freesession(struct nfsdsession *sep, uint8_t *sessionid)
6553 {
6554 	struct nfssessionhash *shp;
6555 	int i;
6556 
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 		sep->sess_refcnt--;
6568 		if (sep->sess_refcnt > 0) {
6569 			NFSUNLOCKSESSION(shp);
6570 			NFSUNLOCKSTATE();
6571 			return (NFSERR_BACKCHANBUSY);
6572 		}
6573 		LIST_REMOVE(sep, sess_hash);
6574 		LIST_REMOVE(sep, sess_list);
6575 	}
6576 	NFSUNLOCKSESSION(shp);
6577 	NFSUNLOCKSTATE();
6578 	if (sep == NULL)
6579 		return (NFSERR_BADSESSION);
6580 	for (i = 0; i < NFSV4_SLOTS; i++)
6581 		if (sep->sess_slots[i].nfssl_reply != NULL)
6582 			m_freem(sep->sess_slots[i].nfssl_reply);
6583 	if (sep->sess_cbsess.nfsess_xprt != NULL)
6584 		SVC_RELEASE(sep->sess_cbsess.nfsess_xprt);
6585 	free(sep, M_NFSDSESSION);
6586 	return (0);
6587 }
6588 
6589 /*
6590  * Free a stateid.
6591  * RFC5661 says that it should fail when there are associated opens, locks
6592  * or delegations. Since stateids represent opens, I don't see how you can
6593  * free an open stateid (it will be free'd when closed), so this function
6594  * only works for lock stateids (freeing the lock_owner) or delegations.
6595  */
6596 int
6597 nfsrv_freestateid(struct nfsrv_descript *nd, nfsv4stateid_t *stateidp,
6598     NFSPROC_T *p)
6599 {
6600 	struct nfsclient *clp;
6601 	struct nfsstate *stp;
6602 	int error;
6603 
6604 	NFSLOCKSTATE();
6605 	/*
6606 	 * Look up the stateid
6607 	 */
6608 	error = nfsrv_getclient((nfsquad_t)((u_quad_t)0), CLOPS_RENEW, &clp,
6609 	    NULL, (nfsquad_t)((u_quad_t)0), 0, nd, p);
6610 	if (error == 0) {
6611 		/* First, check for a delegation. */
6612 		LIST_FOREACH(stp, &clp->lc_deleg, ls_list) {
6613 			if (!NFSBCMP(stp->ls_stateid.other, stateidp->other,
6614 			    NFSX_STATEIDOTHER))
6615 				break;
6616 		}
6617 		if (stp != NULL) {
6618 			nfsrv_freedeleg(stp);
6619 			NFSUNLOCKSTATE();
6620 			return (error);
6621 		}
6622 	}
6623 	/* Not a delegation, try for a lock_owner. */
6624 	if (error == 0)
6625 		error = nfsrv_getstate(clp, stateidp, 0, &stp);
6626 	if (error == 0 && ((stp->ls_flags & (NFSLCK_OPEN | NFSLCK_DELEGREAD |
6627 	    NFSLCK_DELEGWRITE)) != 0 || (stp->ls_flags & NFSLCK_LOCK) == 0))
6628 		/* Not a lock_owner stateid. */
6629 		error = NFSERR_LOCKSHELD;
6630 	if (error == 0 && !LIST_EMPTY(&stp->ls_lock))
6631 		error = NFSERR_LOCKSHELD;
6632 	if (error == 0)
6633 		nfsrv_freelockowner(stp, NULL, 0, p);
6634 	NFSUNLOCKSTATE();
6635 	return (error);
6636 }
6637 
6638 /*
6639  * Test a stateid.
6640  */
6641 int
6642 nfsrv_teststateid(struct nfsrv_descript *nd, nfsv4stateid_t *stateidp,
6643     NFSPROC_T *p)
6644 {
6645 	struct nfsclient *clp;
6646 	struct nfsstate *stp;
6647 	int error;
6648 
6649 	NFSLOCKSTATE();
6650 	/*
6651 	 * Look up the stateid
6652 	 */
6653 	error = nfsrv_getclient((nfsquad_t)((u_quad_t)0), CLOPS_RENEW, &clp,
6654 	    NULL, (nfsquad_t)((u_quad_t)0), 0, nd, p);
6655 	if (error == 0)
6656 		error = nfsrv_getstate(clp, stateidp, 0, &stp);
6657 	if (error == 0 && stateidp->seqid != 0 &&
6658 	    SEQ_LT(stateidp->seqid, stp->ls_stateid.seqid))
6659 		error = NFSERR_OLDSTATEID;
6660 	NFSUNLOCKSTATE();
6661 	return (error);
6662 }
6663 
6664 /*
6665  * Generate the xdr for an NFSv4.1 CBSequence Operation.
6666  */
6667 static int
6668 nfsv4_setcbsequence(struct nfsrv_descript *nd, struct nfsclient *clp,
6669     int dont_replycache, struct nfsdsession **sepp, int *slotposp)
6670 {
6671 	struct nfsdsession *sep;
6672 	uint32_t *tl, slotseq = 0;
6673 	int maxslot;
6674 	uint8_t sessionid[NFSX_V4SESSIONID];
6675 	int error;
6676 
6677 	error = nfsv4_getcbsession(clp, sepp);
6678 	if (error != 0)
6679 		return (error);
6680 	sep = *sepp;
6681 	nfsv4_sequencelookup(NULL, &sep->sess_cbsess, slotposp, &maxslot,
6682 	    &slotseq, sessionid, true);
6683 	KASSERT(maxslot >= 0, ("nfsv4_setcbsequence neg maxslot"));
6684 
6685 	/* Build the Sequence arguments. */
6686 	NFSM_BUILD(tl, uint32_t *, NFSX_V4SESSIONID + 5 * NFSX_UNSIGNED);
6687 	bcopy(sessionid, tl, NFSX_V4SESSIONID);
6688 	tl += NFSX_V4SESSIONID / NFSX_UNSIGNED;
6689 	nd->nd_slotseq = tl;
6690 	nd->nd_slotid = *slotposp;
6691 	nd->nd_flag |= ND_HASSLOTID;
6692 	*tl++ = txdr_unsigned(slotseq);
6693 	*tl++ = txdr_unsigned(*slotposp);
6694 	*tl++ = txdr_unsigned(maxslot);
6695 	if (dont_replycache == 0)
6696 		*tl++ = newnfs_true;
6697 	else
6698 		*tl++ = newnfs_false;
6699 	*tl = 0;			/* No referring call list, for now. */
6700 	nd->nd_flag |= ND_HASSEQUENCE;
6701 	return (0);
6702 }
6703 
6704 /*
6705  * Get a session for the callback.
6706  */
6707 static int
6708 nfsv4_getcbsession(struct nfsclient *clp, struct nfsdsession **sepp)
6709 {
6710 	struct nfsdsession *sep;
6711 
6712 	NFSLOCKSTATE();
6713 	LIST_FOREACH(sep, &clp->lc_session, sess_list) {
6714 		if ((sep->sess_crflags & NFSV4CRSESS_CONNBACKCHAN) != 0)
6715 			break;
6716 	}
6717 	if (sep == NULL) {
6718 		NFSUNLOCKSTATE();
6719 		return (NFSERR_BADSESSION);
6720 	}
6721 	sep->sess_refcnt++;
6722 	*sepp = sep;
6723 	NFSUNLOCKSTATE();
6724 	return (0);
6725 }
6726 
6727 /*
6728  * Free up all backchannel xprts.  This needs to be done when the nfsd threads
6729  * exit, since those transports will all be going away.
6730  * This is only called after all the nfsd threads are done performing RPCs,
6731  * so locking shouldn't be an issue.
6732  */
6733 void
6734 nfsrv_freeallbackchannel_xprts(void)
6735 {
6736 	struct nfsdsession *sep;
6737 	struct nfsclient *clp;
6738 	SVCXPRT *xprt;
6739 	int i;
6740 
6741 	for (i = 0; i < nfsrv_clienthashsize; i++) {
6742 		LIST_FOREACH(clp, &NFSD_VNET(nfsclienthash)[i], lc_hash) {
6743 			LIST_FOREACH(sep, &clp->lc_session, sess_list) {
6744 				xprt = sep->sess_cbsess.nfsess_xprt;
6745 				sep->sess_cbsess.nfsess_xprt = NULL;
6746 				if (xprt != NULL)
6747 					SVC_RELEASE(xprt);
6748 			}
6749 		}
6750 	}
6751 }
6752 
6753 /*
6754  * Do a layout commit.  Actually just call nfsrv_updatemdsattr().
6755  * I have no idea if the rest of these arguments will ever be useful?
6756  */
6757 int
6758 nfsrv_layoutcommit(struct nfsrv_descript *nd, vnode_t vp, int layouttype,
6759     int hasnewoff, uint64_t newoff, uint64_t offset, uint64_t len,
6760     int hasnewmtime, struct timespec *newmtimep, int reclaim,
6761     nfsv4stateid_t *stateidp, int maxcnt, char *layp, int *hasnewsizep,
6762     uint64_t *newsizep, struct ucred *cred, NFSPROC_T *p)
6763 {
6764 	struct nfsvattr na;
6765 	int error;
6766 
6767 	error = nfsrv_updatemdsattr(vp, &na, p);
6768 	if (error == 0) {
6769 		*hasnewsizep = 1;
6770 		*newsizep = na.na_size;
6771 	}
6772 	return (error);
6773 }
6774 
6775 /*
6776  * Try and get a layout.
6777  */
6778 int
6779 nfsrv_layoutget(struct nfsrv_descript *nd, vnode_t vp, struct nfsexstuff *exp,
6780     int layouttype, int *iomode, uint64_t *offset, uint64_t *len,
6781     uint64_t minlen, nfsv4stateid_t *stateidp, int maxcnt, int *retonclose,
6782     int *layoutlenp, char *layp, struct ucred *cred, NFSPROC_T *p)
6783 {
6784 	struct nfslayouthash *lhyp;
6785 	struct nfslayout *lyp;
6786 	char *devid;
6787 	fhandle_t fh, *dsfhp;
6788 	int error, mirrorcnt;
6789 
6790 	if (nfsrv_devidcnt == 0)
6791 		return (NFSERR_UNKNLAYOUTTYPE);
6792 
6793 	if (*offset != 0)
6794 		printf("nfsrv_layoutget: off=%ju len=%ju\n", (uintmax_t)*offset,
6795 		    (uintmax_t)*len);
6796 	error = nfsvno_getfh(vp, &fh, p);
6797 	NFSD_DEBUG(4, "layoutget getfh=%d\n", error);
6798 	if (error != 0)
6799 		return (error);
6800 
6801 	/*
6802 	 * For now, all layouts are for entire files.
6803 	 * Only issue Read/Write layouts if requested for a non-readonly fs.
6804 	 */
6805 	if (NFSVNO_EXRDONLY(exp)) {
6806 		if (*iomode == NFSLAYOUTIOMODE_RW)
6807 			return (NFSERR_LAYOUTTRYLATER);
6808 		*iomode = NFSLAYOUTIOMODE_READ;
6809 	}
6810 	if (*iomode != NFSLAYOUTIOMODE_RW)
6811 		*iomode = NFSLAYOUTIOMODE_READ;
6812 
6813 	/*
6814 	 * Check to see if a write layout can be issued for this file.
6815 	 * This is used during mirror recovery to avoid RW layouts being
6816 	 * issued for a file while it is being copied to the recovered
6817 	 * mirror.
6818 	 */
6819 	if (*iomode == NFSLAYOUTIOMODE_RW && nfsrv_dontlayout(&fh) != 0)
6820 		return (NFSERR_LAYOUTTRYLATER);
6821 
6822 	*retonclose = 0;
6823 	*offset = 0;
6824 	*len = UINT64_MAX;
6825 
6826 	/* First, see if a layout already exists and return if found. */
6827 	lhyp = NFSLAYOUTHASH(&fh);
6828 	NFSLOCKLAYOUT(lhyp);
6829 	error = nfsrv_findlayout(&nd->nd_clientid, &fh, layouttype, p, &lyp);
6830 	NFSD_DEBUG(4, "layoutget findlay=%d\n", error);
6831 	/*
6832 	 * Not sure if the seqid must be the same, so I won't check it.
6833 	 */
6834 	if (error == 0 && (stateidp->other[0] != lyp->lay_stateid.other[0] ||
6835 	    stateidp->other[1] != lyp->lay_stateid.other[1] ||
6836 	    stateidp->other[2] != lyp->lay_stateid.other[2])) {
6837 		if ((lyp->lay_flags & NFSLAY_CALLB) == 0) {
6838 			NFSUNLOCKLAYOUT(lhyp);
6839 			NFSD_DEBUG(1, "ret bad stateid\n");
6840 			return (NFSERR_BADSTATEID);
6841 		}
6842 		/*
6843 		 * I believe we get here because there is a race between
6844 		 * the client processing the CBLAYOUTRECALL and the layout
6845 		 * being deleted here on the server.
6846 		 * The client has now done a LayoutGet with a non-layout
6847 		 * stateid, as it would when there is no layout.
6848 		 * As such, free this layout and set error == NFSERR_BADSTATEID
6849 		 * so the code below will create a new layout structure as
6850 		 * would happen if no layout was found.
6851 		 * "lyp" will be set before being used below, but set it NULL
6852 		 * as a safety belt.
6853 		 */
6854 		nfsrv_freelayout(&lhyp->list, lyp);
6855 		lyp = NULL;
6856 		error = NFSERR_BADSTATEID;
6857 	}
6858 	if (error == 0) {
6859 		if (lyp->lay_layoutlen > maxcnt) {
6860 			NFSUNLOCKLAYOUT(lhyp);
6861 			NFSD_DEBUG(1, "ret layout too small\n");
6862 			return (NFSERR_TOOSMALL);
6863 		}
6864 		if (*iomode == NFSLAYOUTIOMODE_RW) {
6865 			if ((lyp->lay_flags & NFSLAY_NOSPC) != 0) {
6866 				NFSUNLOCKLAYOUT(lhyp);
6867 				NFSD_DEBUG(1, "ret layout nospace\n");
6868 				return (NFSERR_NOSPC);
6869 			}
6870 			lyp->lay_flags |= NFSLAY_RW;
6871 		} else
6872 			lyp->lay_flags |= NFSLAY_READ;
6873 		NFSBCOPY(lyp->lay_xdr, layp, lyp->lay_layoutlen);
6874 		*layoutlenp = lyp->lay_layoutlen;
6875 		if (++lyp->lay_stateid.seqid == 0)
6876 			lyp->lay_stateid.seqid = 1;
6877 		stateidp->seqid = lyp->lay_stateid.seqid;
6878 		NFSUNLOCKLAYOUT(lhyp);
6879 		NFSD_DEBUG(4, "ret fnd layout\n");
6880 		return (0);
6881 	}
6882 	NFSUNLOCKLAYOUT(lhyp);
6883 
6884 	/* Find the device id and file handle. */
6885 	dsfhp = malloc(sizeof(fhandle_t) * NFSDEV_MAXMIRRORS, M_TEMP, M_WAITOK);
6886 	devid = malloc(NFSX_V4DEVICEID * NFSDEV_MAXMIRRORS, M_TEMP, M_WAITOK);
6887 	error = nfsrv_dsgetdevandfh(vp, p, &mirrorcnt, dsfhp, devid);
6888 	NFSD_DEBUG(4, "layoutget devandfh=%d\n", error);
6889 	if (error == 0) {
6890 		if (layouttype == NFSLAYOUT_NFSV4_1_FILES) {
6891 			if (NFSX_V4FILELAYOUT > maxcnt)
6892 				error = NFSERR_TOOSMALL;
6893 			else
6894 				lyp = nfsrv_filelayout(nd, *iomode, &fh, dsfhp,
6895 				    devid, vp->v_mount->mnt_stat.f_fsid);
6896 		} else {
6897 			if (NFSX_V4FLEXLAYOUT(mirrorcnt) > maxcnt)
6898 				error = NFSERR_TOOSMALL;
6899 			else
6900 				lyp = nfsrv_flexlayout(nd, *iomode, mirrorcnt,
6901 				    &fh, dsfhp, devid,
6902 				    vp->v_mount->mnt_stat.f_fsid);
6903 		}
6904 	}
6905 	free(dsfhp, M_TEMP);
6906 	free(devid, M_TEMP);
6907 	if (error != 0)
6908 		return (error);
6909 
6910 	/*
6911 	 * Now, add this layout to the list.
6912 	 */
6913 	error = nfsrv_addlayout(nd, &lyp, stateidp, layp, layoutlenp, p);
6914 	NFSD_DEBUG(4, "layoutget addl=%d\n", error);
6915 	/*
6916 	 * The lyp will be set to NULL by nfsrv_addlayout() if it
6917 	 * linked the new structure into the lists.
6918 	 */
6919 	free(lyp, M_NFSDSTATE);
6920 	return (error);
6921 }
6922 
6923 /*
6924  * Generate a File Layout.
6925  */
6926 static struct nfslayout *
6927 nfsrv_filelayout(struct nfsrv_descript *nd, int iomode, fhandle_t *fhp,
6928     fhandle_t *dsfhp, char *devid, fsid_t fs)
6929 {
6930 	uint32_t *tl;
6931 	struct nfslayout *lyp;
6932 	uint64_t pattern_offset;
6933 
6934 	lyp = malloc(sizeof(struct nfslayout) + NFSX_V4FILELAYOUT, M_NFSDSTATE,
6935 	    M_WAITOK | M_ZERO);
6936 	lyp->lay_type = NFSLAYOUT_NFSV4_1_FILES;
6937 	if (iomode == NFSLAYOUTIOMODE_RW)
6938 		lyp->lay_flags = NFSLAY_RW;
6939 	else
6940 		lyp->lay_flags = NFSLAY_READ;
6941 	NFSBCOPY(fhp, &lyp->lay_fh, sizeof(*fhp));
6942 	lyp->lay_clientid.qval = nd->nd_clientid.qval;
6943 	lyp->lay_fsid = fs;
6944 	NFSBCOPY(devid, lyp->lay_deviceid, NFSX_V4DEVICEID);
6945 
6946 	/* Fill in the xdr for the files layout. */
6947 	tl = (uint32_t *)lyp->lay_xdr;
6948 	NFSBCOPY(devid, tl, NFSX_V4DEVICEID);		/* Device ID. */
6949 	tl += (NFSX_V4DEVICEID / NFSX_UNSIGNED);
6950 
6951 	/* Set the stripe size to the maximum I/O size. */
6952 	*tl++ = txdr_unsigned(nfs_srvmaxio & NFSFLAYUTIL_STRIPE_MASK);
6953 	*tl++ = 0;					/* 1st stripe index. */
6954 	pattern_offset = 0;
6955 	txdr_hyper(pattern_offset, tl); tl += 2;	/* Pattern offset. */
6956 	*tl++ = txdr_unsigned(1);			/* 1 file handle. */
6957 	*tl++ = txdr_unsigned(NFSX_V4PNFSFH);
6958 	NFSBCOPY(dsfhp, tl, sizeof(*dsfhp));
6959 	lyp->lay_layoutlen = NFSX_V4FILELAYOUT;
6960 	return (lyp);
6961 }
6962 
6963 #define	FLEX_OWNERID	"999"
6964 #define	FLEX_UID0	"0"
6965 /*
6966  * Generate a Flex File Layout.
6967  * The FLEX_OWNERID can be any string of 3 decimal digits. Although this
6968  * string goes on the wire, it isn't supposed to be used by the client,
6969  * since this server uses tight coupling.
6970  * Although not recommended by the spec., if vfs.nfsd.flexlinuxhack=1 use
6971  * a string of "0". This works around the Linux Flex File Layout driver bug
6972  * which uses the synthetic uid/gid strings for the "tightly coupled" case.
6973  */
6974 static struct nfslayout *
6975 nfsrv_flexlayout(struct nfsrv_descript *nd, int iomode, int mirrorcnt,
6976     fhandle_t *fhp, fhandle_t *dsfhp, char *devid, fsid_t fs)
6977 {
6978 	uint32_t *tl;
6979 	struct nfslayout *lyp;
6980 	uint64_t lenval;
6981 	int i;
6982 
6983 	lyp = malloc(sizeof(struct nfslayout) + NFSX_V4FLEXLAYOUT(mirrorcnt),
6984 	    M_NFSDSTATE, M_WAITOK | M_ZERO);
6985 	lyp->lay_type = NFSLAYOUT_FLEXFILE;
6986 	if (iomode == NFSLAYOUTIOMODE_RW)
6987 		lyp->lay_flags = NFSLAY_RW;
6988 	else
6989 		lyp->lay_flags = NFSLAY_READ;
6990 	NFSBCOPY(fhp, &lyp->lay_fh, sizeof(*fhp));
6991 	lyp->lay_clientid.qval = nd->nd_clientid.qval;
6992 	lyp->lay_fsid = fs;
6993 	lyp->lay_mirrorcnt = mirrorcnt;
6994 	NFSBCOPY(devid, lyp->lay_deviceid, NFSX_V4DEVICEID);
6995 
6996 	/* Fill in the xdr for the files layout. */
6997 	tl = (uint32_t *)lyp->lay_xdr;
6998 	lenval = 0;
6999 	txdr_hyper(lenval, tl); tl += 2;		/* Stripe unit. */
7000 	*tl++ = txdr_unsigned(mirrorcnt);		/* # of mirrors. */
7001 	for (i = 0; i < mirrorcnt; i++) {
7002 		*tl++ = txdr_unsigned(1);		/* One stripe. */
7003 		NFSBCOPY(devid, tl, NFSX_V4DEVICEID);	/* Device ID. */
7004 		tl += (NFSX_V4DEVICEID / NFSX_UNSIGNED);
7005 		devid += NFSX_V4DEVICEID;
7006 		*tl++ = txdr_unsigned(1);		/* Efficiency. */
7007 		*tl++ = 0;				/* Proxy Stateid. */
7008 		*tl++ = 0x55555555;
7009 		*tl++ = 0x55555555;
7010 		*tl++ = 0x55555555;
7011 		*tl++ = txdr_unsigned(1);		/* 1 file handle. */
7012 		*tl++ = txdr_unsigned(NFSX_V4PNFSFH);
7013 		NFSBCOPY(dsfhp, tl, sizeof(*dsfhp));
7014 		tl += (NFSM_RNDUP(NFSX_V4PNFSFH) / NFSX_UNSIGNED);
7015 		dsfhp++;
7016 		if (nfsrv_flexlinuxhack != 0) {
7017 			*tl++ = txdr_unsigned(strlen(FLEX_UID0));
7018 			*tl = 0;		/* 0 pad string. */
7019 			NFSBCOPY(FLEX_UID0, tl++, strlen(FLEX_UID0));
7020 			*tl++ = txdr_unsigned(strlen(FLEX_UID0));
7021 			*tl = 0;		/* 0 pad string. */
7022 			NFSBCOPY(FLEX_UID0, tl++, strlen(FLEX_UID0));
7023 		} else {
7024 			*tl++ = txdr_unsigned(strlen(FLEX_OWNERID));
7025 			NFSBCOPY(FLEX_OWNERID, tl++, NFSX_UNSIGNED);
7026 			*tl++ = txdr_unsigned(strlen(FLEX_OWNERID));
7027 			NFSBCOPY(FLEX_OWNERID, tl++, NFSX_UNSIGNED);
7028 		}
7029 	}
7030 	*tl++ = txdr_unsigned(0);		/* ff_flags. */
7031 	*tl = txdr_unsigned(60);		/* Status interval hint. */
7032 	lyp->lay_layoutlen = NFSX_V4FLEXLAYOUT(mirrorcnt);
7033 	return (lyp);
7034 }
7035 
7036 /*
7037  * Parse and process Flex File errors returned via LayoutReturn.
7038  */
7039 static void
7040 nfsrv_flexlayouterr(struct nfsrv_descript *nd, uint32_t *layp, int maxcnt,
7041     NFSPROC_T *p)
7042 {
7043 	uint32_t *tl;
7044 	int cnt, errcnt, i, j, opnum, stat;
7045 	char devid[NFSX_V4DEVICEID];
7046 
7047 	tl = layp;
7048 	maxcnt -= NFSX_UNSIGNED;
7049 	if (maxcnt > 0)
7050 		cnt = fxdr_unsigned(int, *tl++);
7051 	else
7052 		cnt = 0;
7053 	NFSD_DEBUG(4, "flexlayouterr cnt=%d\n", cnt);
7054 	for (i = 0; i < cnt; i++) {
7055 		maxcnt -= NFSX_STATEID + 2 * NFSX_HYPER +
7056 		    NFSX_UNSIGNED;
7057 		if (maxcnt <= 0)
7058 			break;
7059 		/* Skip offset, length and stateid for now. */
7060 		tl += (4 + NFSX_STATEID / NFSX_UNSIGNED);
7061 		errcnt = fxdr_unsigned(int, *tl++);
7062 		NFSD_DEBUG(4, "flexlayouterr errcnt=%d\n", errcnt);
7063 		for (j = 0; j < errcnt; j++) {
7064 			maxcnt -= NFSX_V4DEVICEID + 2 * NFSX_UNSIGNED;
7065 			if (maxcnt < 0)
7066 				break;
7067 			NFSBCOPY(tl, devid, NFSX_V4DEVICEID);
7068 			tl += (NFSX_V4DEVICEID / NFSX_UNSIGNED);
7069 			stat = fxdr_unsigned(int, *tl++);
7070 			opnum = fxdr_unsigned(int, *tl++);
7071 			NFSD_DEBUG(4, "flexlayouterr op=%d stat=%d\n", opnum,
7072 			    stat);
7073 			/*
7074 			 * Except for NFSERR_ACCES, NFSERR_STALE and
7075 			 * NFSERR_NOSPC errors, disable the mirror.
7076 			 */
7077 			if (stat != NFSERR_ACCES && stat != NFSERR_STALE &&
7078 			    stat != NFSERR_NOSPC)
7079 				nfsrv_delds(devid, p);
7080 
7081 			/* For NFSERR_NOSPC, mark all devids and layouts. */
7082 			if (stat == NFSERR_NOSPC)
7083 				nfsrv_marknospc(devid, true);
7084 		}
7085 	}
7086 }
7087 
7088 /*
7089  * This function removes all flex file layouts which has a mirror with
7090  * a device id that matches the argument.
7091  * Called when the DS represented by the device id has failed.
7092  */
7093 void
7094 nfsrv_flexmirrordel(char *devid, NFSPROC_T *p)
7095 {
7096 	uint32_t *tl;
7097 	struct nfslayout *lyp, *nlyp;
7098 	struct nfslayouthash *lhyp;
7099 	struct nfslayouthead loclyp;
7100 	int i, j;
7101 
7102 	NFSD_DEBUG(4, "flexmirrordel\n");
7103 	/* Move all layouts found onto a local list. */
7104 	TAILQ_INIT(&loclyp);
7105 	for (i = 0; i < nfsrv_layouthashsize; i++) {
7106 		lhyp = &nfslayouthash[i];
7107 		NFSLOCKLAYOUT(lhyp);
7108 		TAILQ_FOREACH_SAFE(lyp, &lhyp->list, lay_list, nlyp) {
7109 			if (lyp->lay_type == NFSLAYOUT_FLEXFILE &&
7110 			    lyp->lay_mirrorcnt > 1) {
7111 				NFSD_DEBUG(4, "possible match\n");
7112 				tl = lyp->lay_xdr;
7113 				tl += 3;
7114 				for (j = 0; j < lyp->lay_mirrorcnt; j++) {
7115 					tl++;
7116 					if (NFSBCMP(devid, tl, NFSX_V4DEVICEID)
7117 					    == 0) {
7118 						/* Found one. */
7119 						NFSD_DEBUG(4, "fnd one\n");
7120 						TAILQ_REMOVE(&lhyp->list, lyp,
7121 						    lay_list);
7122 						TAILQ_INSERT_HEAD(&loclyp, lyp,
7123 						    lay_list);
7124 						break;
7125 					}
7126 					tl += (NFSX_V4DEVICEID / NFSX_UNSIGNED +
7127 					    NFSM_RNDUP(NFSX_V4PNFSFH) /
7128 					    NFSX_UNSIGNED + 11 * NFSX_UNSIGNED);
7129 				}
7130 			}
7131 		}
7132 		NFSUNLOCKLAYOUT(lhyp);
7133 	}
7134 
7135 	/* Now, try to do a Layout recall for each one found. */
7136 	TAILQ_FOREACH_SAFE(lyp, &loclyp, lay_list, nlyp) {
7137 		NFSD_DEBUG(4, "do layout recall\n");
7138 		/*
7139 		 * The layout stateid.seqid needs to be incremented
7140 		 * before doing a LAYOUT_RECALL callback.
7141 		 */
7142 		if (++lyp->lay_stateid.seqid == 0)
7143 			lyp->lay_stateid.seqid = 1;
7144 		nfsrv_recalllayout(lyp->lay_clientid, &lyp->lay_stateid,
7145 		    &lyp->lay_fh, lyp, 1, lyp->lay_type, p);
7146 		nfsrv_freelayout(&loclyp, lyp);
7147 	}
7148 }
7149 
7150 /*
7151  * Do a recall callback to the client for this layout.
7152  */
7153 static int
7154 nfsrv_recalllayout(nfsquad_t clid, nfsv4stateid_t *stateidp, fhandle_t *fhp,
7155     struct nfslayout *lyp, int changed, int laytype, NFSPROC_T *p)
7156 {
7157 	struct nfsclient *clp;
7158 	int error;
7159 
7160 	NFSD_DEBUG(4, "nfsrv_recalllayout\n");
7161 	error = nfsrv_getclient(clid, 0, &clp, NULL, (nfsquad_t)((u_quad_t)0),
7162 	    0, NULL, p);
7163 	NFSD_DEBUG(4, "aft nfsrv_getclient=%d\n", error);
7164 	if (error != 0) {
7165 		printf("nfsrv_recalllayout: getclient err=%d\n", error);
7166 		return (error);
7167 	}
7168 	if ((clp->lc_flags & LCL_NFSV41) != 0) {
7169 		error = nfsrv_docallback(clp, NFSV4OP_CBLAYOUTRECALL,
7170 		    stateidp, changed, fhp, NULL, NULL, laytype, p);
7171 		/* If lyp != NULL, handle an error return here. */
7172 		if (error != 0 && lyp != NULL) {
7173 			NFSDRECALLLOCK();
7174 			/*
7175 			 * Mark it returned, since no layout recall
7176 			 * has been done.
7177 			 * All errors seem to be non-recoverable, although
7178 			 * NFSERR_NOMATCHLAYOUT is a normal event.
7179 			 */
7180 			if ((lyp->lay_flags & NFSLAY_RECALL) != 0) {
7181 				lyp->lay_flags |= NFSLAY_RETURNED;
7182 				wakeup(lyp);
7183 			}
7184 			NFSDRECALLUNLOCK();
7185 			if (error != NFSERR_NOMATCHLAYOUT)
7186 				printf("nfsrv_recalllayout: err=%d\n", error);
7187 		}
7188 	} else
7189 		printf("nfsrv_recalllayout: clp not NFSv4.1\n");
7190 	return (error);
7191 }
7192 
7193 /*
7194  * Find a layout to recall when we exceed our high water mark.
7195  */
7196 void
7197 nfsrv_recalloldlayout(NFSPROC_T *p)
7198 {
7199 	struct nfslayouthash *lhyp;
7200 	struct nfslayout *lyp;
7201 	nfsquad_t clientid;
7202 	nfsv4stateid_t stateid;
7203 	fhandle_t fh;
7204 	int error, laytype = 0, ret;
7205 
7206 	lhyp = &nfslayouthash[arc4random() % nfsrv_layouthashsize];
7207 	NFSLOCKLAYOUT(lhyp);
7208 	TAILQ_FOREACH_REVERSE(lyp, &lhyp->list, nfslayouthead, lay_list) {
7209 		if ((lyp->lay_flags & NFSLAY_CALLB) == 0) {
7210 			lyp->lay_flags |= NFSLAY_CALLB;
7211 			/*
7212 			 * The layout stateid.seqid needs to be incremented
7213 			 * before doing a LAYOUT_RECALL callback.
7214 			 */
7215 			if (++lyp->lay_stateid.seqid == 0)
7216 				lyp->lay_stateid.seqid = 1;
7217 			clientid = lyp->lay_clientid;
7218 			stateid = lyp->lay_stateid;
7219 			NFSBCOPY(&lyp->lay_fh, &fh, sizeof(fh));
7220 			laytype = lyp->lay_type;
7221 			break;
7222 		}
7223 	}
7224 	NFSUNLOCKLAYOUT(lhyp);
7225 	if (lyp != NULL) {
7226 		error = nfsrv_recalllayout(clientid, &stateid, &fh, NULL, 0,
7227 		    laytype, p);
7228 		if (error != 0 && error != NFSERR_NOMATCHLAYOUT)
7229 			NFSD_DEBUG(4, "recallold=%d\n", error);
7230 		if (error != 0) {
7231 			NFSLOCKLAYOUT(lhyp);
7232 			/*
7233 			 * Since the hash list was unlocked, we need to
7234 			 * find it again.
7235 			 */
7236 			ret = nfsrv_findlayout(&clientid, &fh, laytype, p,
7237 			    &lyp);
7238 			if (ret == 0 &&
7239 			    (lyp->lay_flags & NFSLAY_CALLB) != 0 &&
7240 			    lyp->lay_stateid.other[0] == stateid.other[0] &&
7241 			    lyp->lay_stateid.other[1] == stateid.other[1] &&
7242 			    lyp->lay_stateid.other[2] == stateid.other[2]) {
7243 				/*
7244 				 * The client no longer knows this layout, so
7245 				 * it can be free'd now.
7246 				 */
7247 				if (error == NFSERR_NOMATCHLAYOUT)
7248 					nfsrv_freelayout(&lhyp->list, lyp);
7249 				else {
7250 					/*
7251 					 * Leave it to be tried later by
7252 					 * clearing NFSLAY_CALLB and moving
7253 					 * it to the head of the list, so it
7254 					 * won't be tried again for a while.
7255 					 */
7256 					lyp->lay_flags &= ~NFSLAY_CALLB;
7257 					TAILQ_REMOVE(&lhyp->list, lyp,
7258 					    lay_list);
7259 					TAILQ_INSERT_HEAD(&lhyp->list, lyp,
7260 					    lay_list);
7261 				}
7262 			}
7263 			NFSUNLOCKLAYOUT(lhyp);
7264 		}
7265 	}
7266 }
7267 
7268 /*
7269  * Try and return layout(s).
7270  */
7271 int
7272 nfsrv_layoutreturn(struct nfsrv_descript *nd, vnode_t vp,
7273     int layouttype, int iomode, uint64_t offset, uint64_t len, int reclaim,
7274     int kind, nfsv4stateid_t *stateidp, int maxcnt, uint32_t *layp, int *fndp,
7275     struct ucred *cred, NFSPROC_T *p)
7276 {
7277 	struct nfsvattr na;
7278 	struct nfslayouthash *lhyp;
7279 	struct nfslayout *lyp;
7280 	fhandle_t fh;
7281 	int error = 0;
7282 
7283 	*fndp = 0;
7284 	if (kind == NFSV4LAYOUTRET_FILE) {
7285 		error = nfsvno_getfh(vp, &fh, p);
7286 		if (error == 0) {
7287 			error = nfsrv_updatemdsattr(vp, &na, p);
7288 			if (error != 0)
7289 				printf("nfsrv_layoutreturn: updatemdsattr"
7290 				    " failed=%d\n", error);
7291 		}
7292 		if (error == 0) {
7293 			if (reclaim == newnfs_true) {
7294 				error = nfsrv_checkgrace(NULL, NULL,
7295 				    NFSLCK_RECLAIM);
7296 				if (error != NFSERR_NOGRACE)
7297 					error = 0;
7298 				return (error);
7299 			}
7300 			lhyp = NFSLAYOUTHASH(&fh);
7301 			NFSDRECALLLOCK();
7302 			NFSLOCKLAYOUT(lhyp);
7303 			error = nfsrv_findlayout(&nd->nd_clientid, &fh,
7304 			    layouttype, p, &lyp);
7305 			NFSD_DEBUG(4, "layoutret findlay=%d\n", error);
7306 			if (error == 0 &&
7307 			    stateidp->other[0] == lyp->lay_stateid.other[0] &&
7308 			    stateidp->other[1] == lyp->lay_stateid.other[1] &&
7309 			    stateidp->other[2] == lyp->lay_stateid.other[2]) {
7310 				NFSD_DEBUG(4, "nfsrv_layoutreturn: stateid %d"
7311 				    " %x %x %x laystateid %d %x %x %x"
7312 				    " off=%ju len=%ju flgs=0x%x\n",
7313 				    stateidp->seqid, stateidp->other[0],
7314 				    stateidp->other[1], stateidp->other[2],
7315 				    lyp->lay_stateid.seqid,
7316 				    lyp->lay_stateid.other[0],
7317 				    lyp->lay_stateid.other[1],
7318 				    lyp->lay_stateid.other[2],
7319 				    (uintmax_t)offset, (uintmax_t)len,
7320 				    lyp->lay_flags);
7321 				if (++lyp->lay_stateid.seqid == 0)
7322 					lyp->lay_stateid.seqid = 1;
7323 				stateidp->seqid = lyp->lay_stateid.seqid;
7324 				if (offset == 0 && len == UINT64_MAX) {
7325 					if ((iomode & NFSLAYOUTIOMODE_READ) !=
7326 					    0)
7327 						lyp->lay_flags &= ~NFSLAY_READ;
7328 					if ((iomode & NFSLAYOUTIOMODE_RW) != 0)
7329 						lyp->lay_flags &= ~NFSLAY_RW;
7330 					if ((lyp->lay_flags & (NFSLAY_READ |
7331 					    NFSLAY_RW)) == 0)
7332 						nfsrv_freelayout(&lhyp->list,
7333 						    lyp);
7334 					else
7335 						*fndp = 1;
7336 				} else
7337 					*fndp = 1;
7338 			}
7339 			NFSUNLOCKLAYOUT(lhyp);
7340 			/* Search the nfsrv_recalllist for a match. */
7341 			TAILQ_FOREACH(lyp, &nfsrv_recalllisthead, lay_list) {
7342 				if (NFSBCMP(&lyp->lay_fh, &fh,
7343 				    sizeof(fh)) == 0 &&
7344 				    lyp->lay_clientid.qval ==
7345 				    nd->nd_clientid.qval &&
7346 				    stateidp->other[0] ==
7347 				    lyp->lay_stateid.other[0] &&
7348 				    stateidp->other[1] ==
7349 				    lyp->lay_stateid.other[1] &&
7350 				    stateidp->other[2] ==
7351 				    lyp->lay_stateid.other[2]) {
7352 					lyp->lay_flags |= NFSLAY_RETURNED;
7353 					wakeup(lyp);
7354 					error = 0;
7355 				}
7356 			}
7357 			NFSDRECALLUNLOCK();
7358 		}
7359 		if (layouttype == NFSLAYOUT_FLEXFILE && layp != NULL)
7360 			nfsrv_flexlayouterr(nd, layp, maxcnt, p);
7361 	} else if (kind == NFSV4LAYOUTRET_FSID)
7362 		nfsrv_freelayouts(&nd->nd_clientid,
7363 		    &vp->v_mount->mnt_stat.f_fsid, layouttype, iomode);
7364 	else if (kind == NFSV4LAYOUTRET_ALL)
7365 		nfsrv_freelayouts(&nd->nd_clientid, NULL, layouttype, iomode);
7366 	else
7367 		error = NFSERR_INVAL;
7368 	if (error == -1)
7369 		error = 0;
7370 	return (error);
7371 }
7372 
7373 /*
7374  * Look for an existing layout.
7375  */
7376 static int
7377 nfsrv_findlayout(nfsquad_t *clientidp, fhandle_t *fhp, int laytype,
7378     NFSPROC_T *p, struct nfslayout **lypp)
7379 {
7380 	struct nfslayouthash *lhyp;
7381 	struct nfslayout *lyp;
7382 	int ret;
7383 
7384 	*lypp = NULL;
7385 	ret = 0;
7386 	lhyp = NFSLAYOUTHASH(fhp);
7387 	TAILQ_FOREACH(lyp, &lhyp->list, lay_list) {
7388 		if (NFSBCMP(&lyp->lay_fh, fhp, sizeof(*fhp)) == 0 &&
7389 		    lyp->lay_clientid.qval == clientidp->qval &&
7390 		    lyp->lay_type == laytype)
7391 			break;
7392 	}
7393 	if (lyp != NULL)
7394 		*lypp = lyp;
7395 	else
7396 		ret = -1;
7397 	return (ret);
7398 }
7399 
7400 /*
7401  * Add the new layout, as required.
7402  */
7403 static int
7404 nfsrv_addlayout(struct nfsrv_descript *nd, struct nfslayout **lypp,
7405     nfsv4stateid_t *stateidp, char *layp, int *layoutlenp, NFSPROC_T *p)
7406 {
7407 	struct nfsclient *clp;
7408 	struct nfslayouthash *lhyp;
7409 	struct nfslayout *lyp, *nlyp;
7410 	fhandle_t *fhp;
7411 	int error;
7412 
7413 	KASSERT((nd->nd_flag & ND_IMPLIEDCLID) != 0,
7414 	    ("nfsrv_layoutget: no nd_clientid\n"));
7415 	lyp = *lypp;
7416 	fhp = &lyp->lay_fh;
7417 	NFSLOCKSTATE();
7418 	error = nfsrv_getclient((nfsquad_t)((u_quad_t)0), CLOPS_RENEW, &clp,
7419 	    NULL, (nfsquad_t)((u_quad_t)0), 0, nd, p);
7420 	if (error != 0) {
7421 		NFSUNLOCKSTATE();
7422 		return (error);
7423 	}
7424 	lyp->lay_stateid.seqid = stateidp->seqid = 1;
7425 	lyp->lay_stateid.other[0] = stateidp->other[0] =
7426 	    clp->lc_clientid.lval[0];
7427 	lyp->lay_stateid.other[1] = stateidp->other[1] =
7428 	    clp->lc_clientid.lval[1];
7429 	lyp->lay_stateid.other[2] = stateidp->other[2] =
7430 	    nfsrv_nextstateindex(clp);
7431 	NFSUNLOCKSTATE();
7432 
7433 	lhyp = NFSLAYOUTHASH(fhp);
7434 	NFSLOCKLAYOUT(lhyp);
7435 	TAILQ_FOREACH(nlyp, &lhyp->list, lay_list) {
7436 		if (NFSBCMP(&nlyp->lay_fh, fhp, sizeof(*fhp)) == 0 &&
7437 		    nlyp->lay_clientid.qval == nd->nd_clientid.qval)
7438 			break;
7439 	}
7440 	if (nlyp != NULL) {
7441 		/* A layout already exists, so use it. */
7442 		nlyp->lay_flags |= (lyp->lay_flags & (NFSLAY_READ | NFSLAY_RW));
7443 		NFSBCOPY(nlyp->lay_xdr, layp, nlyp->lay_layoutlen);
7444 		*layoutlenp = nlyp->lay_layoutlen;
7445 		if (++nlyp->lay_stateid.seqid == 0)
7446 			nlyp->lay_stateid.seqid = 1;
7447 		stateidp->seqid = nlyp->lay_stateid.seqid;
7448 		stateidp->other[0] = nlyp->lay_stateid.other[0];
7449 		stateidp->other[1] = nlyp->lay_stateid.other[1];
7450 		stateidp->other[2] = nlyp->lay_stateid.other[2];
7451 		NFSUNLOCKLAYOUT(lhyp);
7452 		return (0);
7453 	}
7454 
7455 	/* Insert the new layout in the lists. */
7456 	*lypp = NULL;
7457 	atomic_add_int(&nfsrv_layoutcnt, 1);
7458 	NFSD_VNET(nfsstatsv1_p)->srvlayouts++;
7459 	NFSBCOPY(lyp->lay_xdr, layp, lyp->lay_layoutlen);
7460 	*layoutlenp = lyp->lay_layoutlen;
7461 	TAILQ_INSERT_HEAD(&lhyp->list, lyp, lay_list);
7462 	NFSUNLOCKLAYOUT(lhyp);
7463 	return (0);
7464 }
7465 
7466 /*
7467  * Get the devinfo for a deviceid.
7468  */
7469 int
7470 nfsrv_getdevinfo(char *devid, int layouttype, uint32_t *maxcnt,
7471     uint32_t *notify, int *devaddrlen, char **devaddr)
7472 {
7473 	struct nfsdevice *ds;
7474 
7475 	if ((layouttype != NFSLAYOUT_NFSV4_1_FILES && layouttype !=
7476 	     NFSLAYOUT_FLEXFILE) ||
7477 	    (nfsrv_maxpnfsmirror > 1 && layouttype == NFSLAYOUT_NFSV4_1_FILES))
7478 		return (NFSERR_UNKNLAYOUTTYPE);
7479 
7480 	/*
7481 	 * Now, search for the device id.  Note that the structures won't go
7482 	 * away, but the order changes in the list.  As such, the lock only
7483 	 * needs to be held during the search through the list.
7484 	 */
7485 	NFSDDSLOCK();
7486 	TAILQ_FOREACH(ds, &nfsrv_devidhead, nfsdev_list) {
7487 		if (NFSBCMP(devid, ds->nfsdev_deviceid, NFSX_V4DEVICEID) == 0 &&
7488 		    ds->nfsdev_nmp != NULL)
7489 			break;
7490 	}
7491 	NFSDDSUNLOCK();
7492 	if (ds == NULL)
7493 		return (NFSERR_NOENT);
7494 
7495 	/* If the correct nfsdev_XXXXaddrlen is > 0, we have the device info. */
7496 	*devaddrlen = 0;
7497 	if (layouttype == NFSLAYOUT_NFSV4_1_FILES) {
7498 		*devaddrlen = ds->nfsdev_fileaddrlen;
7499 		*devaddr = ds->nfsdev_fileaddr;
7500 	} else if (layouttype == NFSLAYOUT_FLEXFILE) {
7501 		*devaddrlen = ds->nfsdev_flexaddrlen;
7502 		*devaddr = ds->nfsdev_flexaddr;
7503 	}
7504 	if (*devaddrlen == 0)
7505 		return (NFSERR_UNKNLAYOUTTYPE);
7506 
7507 	/*
7508 	 * The XDR overhead is 3 unsigned values: layout_type,
7509 	 * length_of_address and notify bitmap.
7510 	 * If the notify array is changed to not all zeros, the
7511 	 * count of unsigned values must be increased.
7512 	 */
7513 	if (*maxcnt > 0 && *maxcnt < NFSM_RNDUP(*devaddrlen) +
7514 	    3 * NFSX_UNSIGNED) {
7515 		*maxcnt = NFSM_RNDUP(*devaddrlen) + 3 * NFSX_UNSIGNED;
7516 		return (NFSERR_TOOSMALL);
7517 	}
7518 	return (0);
7519 }
7520 
7521 /*
7522  * Free a list of layout state structures.
7523  */
7524 static void
7525 nfsrv_freelayoutlist(nfsquad_t clientid)
7526 {
7527 	struct nfslayouthash *lhyp;
7528 	struct nfslayout *lyp, *nlyp;
7529 	int i;
7530 
7531 	for (i = 0; i < nfsrv_layouthashsize; i++) {
7532 		lhyp = &nfslayouthash[i];
7533 		NFSLOCKLAYOUT(lhyp);
7534 		TAILQ_FOREACH_SAFE(lyp, &lhyp->list, lay_list, nlyp) {
7535 			if (lyp->lay_clientid.qval == clientid.qval)
7536 				nfsrv_freelayout(&lhyp->list, lyp);
7537 		}
7538 		NFSUNLOCKLAYOUT(lhyp);
7539 	}
7540 }
7541 
7542 /*
7543  * Free up a layout.
7544  */
7545 static void
7546 nfsrv_freelayout(struct nfslayouthead *lhp, struct nfslayout *lyp)
7547 {
7548 
7549 	NFSD_DEBUG(4, "Freelayout=%p\n", lyp);
7550 	atomic_add_int(&nfsrv_layoutcnt, -1);
7551 	NFSD_VNET(nfsstatsv1_p)->srvlayouts--;
7552 	TAILQ_REMOVE(lhp, lyp, lay_list);
7553 	free(lyp, M_NFSDSTATE);
7554 }
7555 
7556 /*
7557  * Free up a device id.
7558  */
7559 void
7560 nfsrv_freeonedevid(struct nfsdevice *ds)
7561 {
7562 	int i;
7563 
7564 	atomic_add_int(&nfsrv_devidcnt, -1);
7565 	vrele(ds->nfsdev_dvp);
7566 	for (i = 0; i < nfsrv_dsdirsize; i++)
7567 		if (ds->nfsdev_dsdir[i] != NULL)
7568 			vrele(ds->nfsdev_dsdir[i]);
7569 	free(ds->nfsdev_fileaddr, M_NFSDSTATE);
7570 	free(ds->nfsdev_flexaddr, M_NFSDSTATE);
7571 	free(ds->nfsdev_host, M_NFSDSTATE);
7572 	free(ds, M_NFSDSTATE);
7573 }
7574 
7575 /*
7576  * Free up a device id and its mirrors.
7577  */
7578 static void
7579 nfsrv_freedevid(struct nfsdevice *ds)
7580 {
7581 
7582 	TAILQ_REMOVE(&nfsrv_devidhead, ds, nfsdev_list);
7583 	nfsrv_freeonedevid(ds);
7584 }
7585 
7586 /*
7587  * Free all layouts and device ids.
7588  * Done when the nfsd threads are shut down since there may be a new
7589  * modified device id list created when the nfsd is restarted.
7590  */
7591 void
7592 nfsrv_freealllayoutsanddevids(void)
7593 {
7594 	struct nfsdontlist *mrp, *nmrp;
7595 	struct nfslayout *lyp, *nlyp;
7596 
7597 	/* Get rid of the deviceid structures. */
7598 	nfsrv_freealldevids();
7599 	TAILQ_INIT(&nfsrv_devidhead);
7600 	nfsrv_devidcnt = 0;
7601 
7602 	/* Get rid of all layouts. */
7603 	nfsrv_freealllayouts();
7604 
7605 	/* Get rid of any nfsdontlist entries. */
7606 	LIST_FOREACH_SAFE(mrp, &nfsrv_dontlisthead, nfsmr_list, nmrp)
7607 		free(mrp, M_NFSDSTATE);
7608 	LIST_INIT(&nfsrv_dontlisthead);
7609 	nfsrv_dontlistlen = 0;
7610 
7611 	/* Free layouts in the recall list. */
7612 	TAILQ_FOREACH_SAFE(lyp, &nfsrv_recalllisthead, lay_list, nlyp)
7613 		nfsrv_freelayout(&nfsrv_recalllisthead, lyp);
7614 	TAILQ_INIT(&nfsrv_recalllisthead);
7615 }
7616 
7617 /*
7618  * Free layouts that match the arguments.
7619  */
7620 static void
7621 nfsrv_freelayouts(nfsquad_t *clid, fsid_t *fs, int laytype, int iomode)
7622 {
7623 	struct nfslayouthash *lhyp;
7624 	struct nfslayout *lyp, *nlyp;
7625 	int i;
7626 
7627 	for (i = 0; i < nfsrv_layouthashsize; i++) {
7628 		lhyp = &nfslayouthash[i];
7629 		NFSLOCKLAYOUT(lhyp);
7630 		TAILQ_FOREACH_SAFE(lyp, &lhyp->list, lay_list, nlyp) {
7631 			if (clid->qval != lyp->lay_clientid.qval)
7632 				continue;
7633 			if (fs != NULL && fsidcmp(fs, &lyp->lay_fsid) != 0)
7634 				continue;
7635 			if (laytype != lyp->lay_type)
7636 				continue;
7637 			if ((iomode & NFSLAYOUTIOMODE_READ) != 0)
7638 				lyp->lay_flags &= ~NFSLAY_READ;
7639 			if ((iomode & NFSLAYOUTIOMODE_RW) != 0)
7640 				lyp->lay_flags &= ~NFSLAY_RW;
7641 			if ((lyp->lay_flags & (NFSLAY_READ | NFSLAY_RW)) == 0)
7642 				nfsrv_freelayout(&lhyp->list, lyp);
7643 		}
7644 		NFSUNLOCKLAYOUT(lhyp);
7645 	}
7646 }
7647 
7648 /*
7649  * Free all layouts for the argument file.
7650  */
7651 void
7652 nfsrv_freefilelayouts(fhandle_t *fhp)
7653 {
7654 	struct nfslayouthash *lhyp;
7655 	struct nfslayout *lyp, *nlyp;
7656 
7657 	lhyp = NFSLAYOUTHASH(fhp);
7658 	NFSLOCKLAYOUT(lhyp);
7659 	TAILQ_FOREACH_SAFE(lyp, &lhyp->list, lay_list, nlyp) {
7660 		if (NFSBCMP(&lyp->lay_fh, fhp, sizeof(*fhp)) == 0)
7661 			nfsrv_freelayout(&lhyp->list, lyp);
7662 	}
7663 	NFSUNLOCKLAYOUT(lhyp);
7664 }
7665 
7666 /*
7667  * Free all layouts.
7668  */
7669 static void
7670 nfsrv_freealllayouts(void)
7671 {
7672 	struct nfslayouthash *lhyp;
7673 	struct nfslayout *lyp, *nlyp;
7674 	int i;
7675 
7676 	for (i = 0; i < nfsrv_layouthashsize; i++) {
7677 		lhyp = &nfslayouthash[i];
7678 		NFSLOCKLAYOUT(lhyp);
7679 		TAILQ_FOREACH_SAFE(lyp, &lhyp->list, lay_list, nlyp)
7680 			nfsrv_freelayout(&lhyp->list, lyp);
7681 		NFSUNLOCKLAYOUT(lhyp);
7682 	}
7683 }
7684 
7685 /*
7686  * Look up the mount path for the DS server.
7687  */
7688 static int
7689 nfsrv_setdsserver(char *dspathp, char *mdspathp, NFSPROC_T *p,
7690     struct nfsdevice **dsp)
7691 {
7692 	struct nameidata nd;
7693 	struct nfsdevice *ds;
7694 	struct mount *mp;
7695 	int error, i;
7696 	char *dsdirpath;
7697 	size_t dsdirsize;
7698 
7699 	NFSD_DEBUG(4, "setdssrv path=%s\n", dspathp);
7700 	*dsp = NULL;
7701 	if (jailed(p->td_ucred)) {
7702 		printf("A pNFS nfsd cannot run in a jail\n");
7703 		return (EPERM);
7704 	}
7705 	NDINIT(&nd, LOOKUP, FOLLOW | LOCKSHARED | LOCKLEAF, UIO_SYSSPACE,
7706 	    dspathp);
7707 	error = namei(&nd);
7708 	NFSD_DEBUG(4, "lookup=%d\n", error);
7709 	if (error != 0)
7710 		return (error);
7711 	if (nd.ni_vp->v_type != VDIR) {
7712 		vput(nd.ni_vp);
7713 		NFSD_DEBUG(4, "dspath not dir\n");
7714 		return (ENOTDIR);
7715 	}
7716 	if (strcmp(nd.ni_vp->v_mount->mnt_vfc->vfc_name, "nfs") != 0) {
7717 		vput(nd.ni_vp);
7718 		NFSD_DEBUG(4, "dspath not an NFS mount\n");
7719 		return (ENXIO);
7720 	}
7721 
7722 	/*
7723 	 * Allocate a DS server structure with the NFS mounted directory
7724 	 * vnode reference counted, so that a non-forced dismount will
7725 	 * fail with EBUSY.
7726 	 * This structure is always linked into the list, even if an error
7727 	 * is being returned.  The caller will free the entire list upon
7728 	 * an error return.
7729 	 */
7730 	*dsp = ds = malloc(sizeof(*ds) + nfsrv_dsdirsize * sizeof(vnode_t),
7731 	    M_NFSDSTATE, M_WAITOK | M_ZERO);
7732 	ds->nfsdev_dvp = nd.ni_vp;
7733 	ds->nfsdev_nmp = VFSTONFS(nd.ni_vp->v_mount);
7734 	NFSVOPUNLOCK(nd.ni_vp);
7735 
7736 	dsdirsize = strlen(dspathp) + 16;
7737 	dsdirpath = malloc(dsdirsize, M_TEMP, M_WAITOK);
7738 	/* Now, create the DS directory structures. */
7739 	for (i = 0; i < nfsrv_dsdirsize; i++) {
7740 		snprintf(dsdirpath, dsdirsize, "%s/ds%d", dspathp, i);
7741 		NDINIT(&nd, LOOKUP, FOLLOW | LOCKSHARED | LOCKLEAF,
7742 		    UIO_SYSSPACE, dsdirpath);
7743 		error = namei(&nd);
7744 		NFSD_DEBUG(4, "dsdirpath=%s lookup=%d\n", dsdirpath, error);
7745 		if (error != 0)
7746 			break;
7747 		if (nd.ni_vp->v_type != VDIR) {
7748 			vput(nd.ni_vp);
7749 			error = ENOTDIR;
7750 			NFSD_DEBUG(4, "dsdirpath not a VDIR\n");
7751 			break;
7752 		}
7753 		if (strcmp(nd.ni_vp->v_mount->mnt_vfc->vfc_name, "nfs") != 0) {
7754 			vput(nd.ni_vp);
7755 			error = ENXIO;
7756 			NFSD_DEBUG(4, "dsdirpath not an NFS mount\n");
7757 			break;
7758 		}
7759 		ds->nfsdev_dsdir[i] = nd.ni_vp;
7760 		NFSVOPUNLOCK(nd.ni_vp);
7761 	}
7762 	free(dsdirpath, M_TEMP);
7763 
7764 	if (strlen(mdspathp) > 0) {
7765 		/*
7766 		 * This DS stores file for a specific MDS exported file
7767 		 * system.
7768 		 */
7769 		NDINIT(&nd, LOOKUP, FOLLOW | LOCKSHARED | LOCKLEAF,
7770 		    UIO_SYSSPACE, mdspathp);
7771 		error = namei(&nd);
7772 		NFSD_DEBUG(4, "mds lookup=%d\n", error);
7773 		if (error != 0)
7774 			goto out;
7775 		if (nd.ni_vp->v_type != VDIR) {
7776 			vput(nd.ni_vp);
7777 			error = ENOTDIR;
7778 			NFSD_DEBUG(4, "mdspath not dir\n");
7779 			goto out;
7780 		}
7781 		mp = nd.ni_vp->v_mount;
7782 		if ((mp->mnt_flag & MNT_EXPORTED) == 0) {
7783 			vput(nd.ni_vp);
7784 			error = ENXIO;
7785 			NFSD_DEBUG(4, "mdspath not an exported fs\n");
7786 			goto out;
7787 		}
7788 		ds->nfsdev_mdsfsid = mp->mnt_stat.f_fsid;
7789 		ds->nfsdev_mdsisset = 1;
7790 		vput(nd.ni_vp);
7791 	}
7792 
7793 out:
7794 	TAILQ_INSERT_TAIL(&nfsrv_devidhead, ds, nfsdev_list);
7795 	atomic_add_int(&nfsrv_devidcnt, 1);
7796 	return (error);
7797 }
7798 
7799 /*
7800  * Look up the mount path for the DS server and delete it.
7801  */
7802 int
7803 nfsrv_deldsserver(int op, char *dspathp, NFSPROC_T *p)
7804 {
7805 	struct mount *mp;
7806 	struct nfsmount *nmp;
7807 	struct nfsdevice *ds;
7808 	int error;
7809 
7810 	NFSD_DEBUG(4, "deldssrv path=%s\n", dspathp);
7811 	/*
7812 	 * Search for the path in the mount list.  Avoid looking the path
7813 	 * up, since this mount point may be hung, with associated locked
7814 	 * vnodes, etc.
7815 	 * Set NFSMNTP_CANCELRPCS so that any forced dismount will be blocked
7816 	 * until this completes.
7817 	 * As noted in the man page, this should be done before any forced
7818 	 * dismount on the mount point, but at least the handshake on
7819 	 * NFSMNTP_CANCELRPCS should make it safe.
7820 	 */
7821 	error = 0;
7822 	ds = NULL;
7823 	nmp = NULL;
7824 	mtx_lock(&mountlist_mtx);
7825 	TAILQ_FOREACH(mp, &mountlist, mnt_list) {
7826 		if (strcmp(mp->mnt_stat.f_mntonname, dspathp) == 0 &&
7827 		    strcmp(mp->mnt_stat.f_fstypename, "nfs") == 0 &&
7828 		    mp->mnt_data != NULL) {
7829 			nmp = VFSTONFS(mp);
7830 			NFSLOCKMNT(nmp);
7831 			if ((nmp->nm_privflag & (NFSMNTP_FORCEDISM |
7832 			     NFSMNTP_CANCELRPCS)) == 0) {
7833 				nmp->nm_privflag |= NFSMNTP_CANCELRPCS;
7834 				NFSUNLOCKMNT(nmp);
7835 			} else {
7836 				NFSUNLOCKMNT(nmp);
7837 				nmp = NULL;
7838 			}
7839 			break;
7840 		}
7841 	}
7842 	mtx_unlock(&mountlist_mtx);
7843 
7844 	if (nmp != NULL) {
7845 		ds = nfsrv_deldsnmp(op, nmp, p);
7846 		NFSD_DEBUG(4, "deldsnmp=%p\n", ds);
7847 		if (ds != NULL) {
7848 			nfsrv_killrpcs(nmp);
7849 			NFSD_DEBUG(4, "aft killrpcs\n");
7850 		} else
7851 			error = ENXIO;
7852 		NFSLOCKMNT(nmp);
7853 		nmp->nm_privflag &= ~NFSMNTP_CANCELRPCS;
7854 		wakeup(nmp);
7855 		NFSUNLOCKMNT(nmp);
7856 	} else
7857 		error = EINVAL;
7858 	return (error);
7859 }
7860 
7861 /*
7862  * Search for and remove a DS entry which matches the "nmp" argument.
7863  * The nfsdevice structure pointer is returned so that the caller can
7864  * free it via nfsrv_freeonedevid().
7865  * For the forced case, do not try to do LayoutRecalls, since the server
7866  * must be shut down now anyhow.
7867  */
7868 struct nfsdevice *
7869 nfsrv_deldsnmp(int op, struct nfsmount *nmp, NFSPROC_T *p)
7870 {
7871 	struct nfsdevice *fndds;
7872 
7873 	NFSD_DEBUG(4, "deldsdvp\n");
7874 	NFSDDSLOCK();
7875 	if (op == PNFSDOP_FORCEDELDS)
7876 		fndds = nfsv4_findmirror(nmp);
7877 	else
7878 		fndds = nfsrv_findmirroredds(nmp);
7879 	if (fndds != NULL)
7880 		nfsrv_deleteds(fndds);
7881 	NFSDDSUNLOCK();
7882 	if (fndds != NULL) {
7883 		if (op != PNFSDOP_FORCEDELDS)
7884 			nfsrv_flexmirrordel(fndds->nfsdev_deviceid, p);
7885 		printf("pNFS server: mirror %s failed\n", fndds->nfsdev_host);
7886 	}
7887 	return (fndds);
7888 }
7889 
7890 /*
7891  * Similar to nfsrv_deldsnmp(), except that the DS is indicated by deviceid.
7892  * This function also calls nfsrv_killrpcs() to unblock RPCs on the mount
7893  * point.
7894  * Also, returns an error instead of the nfsdevice found.
7895  */
7896 int
7897 nfsrv_delds(char *devid, NFSPROC_T *p)
7898 {
7899 	struct nfsdevice *ds, *fndds;
7900 	struct nfsmount *nmp;
7901 	int fndmirror;
7902 
7903 	NFSD_DEBUG(4, "delds\n");
7904 	/*
7905 	 * Search the DS server list for a match with devid.
7906 	 * Remove the DS entry if found and there is a mirror.
7907 	 */
7908 	fndds = NULL;
7909 	nmp = NULL;
7910 	fndmirror = 0;
7911 	NFSDDSLOCK();
7912 	TAILQ_FOREACH(ds, &nfsrv_devidhead, nfsdev_list) {
7913 		if (NFSBCMP(ds->nfsdev_deviceid, devid, NFSX_V4DEVICEID) == 0 &&
7914 		    ds->nfsdev_nmp != NULL) {
7915 			NFSD_DEBUG(4, "fnd main ds\n");
7916 			fndds = ds;
7917 			break;
7918 		}
7919 	}
7920 	if (fndds == NULL) {
7921 		NFSDDSUNLOCK();
7922 		return (ENXIO);
7923 	}
7924 	if (fndds->nfsdev_mdsisset == 0 && nfsrv_faildscnt > 0)
7925 		fndmirror = 1;
7926 	else if (fndds->nfsdev_mdsisset != 0) {
7927 		/* For the fsid is set case, search for a mirror. */
7928 		TAILQ_FOREACH(ds, &nfsrv_devidhead, nfsdev_list) {
7929 			if (ds != fndds && ds->nfsdev_nmp != NULL &&
7930 			    ds->nfsdev_mdsisset != 0 &&
7931 			    fsidcmp(&ds->nfsdev_mdsfsid,
7932 			    &fndds->nfsdev_mdsfsid) == 0) {
7933 				fndmirror = 1;
7934 				break;
7935 			}
7936 		}
7937 	}
7938 	if (fndmirror != 0) {
7939 		nmp = fndds->nfsdev_nmp;
7940 		NFSLOCKMNT(nmp);
7941 		if ((nmp->nm_privflag & (NFSMNTP_FORCEDISM |
7942 		     NFSMNTP_CANCELRPCS)) == 0) {
7943 			nmp->nm_privflag |= NFSMNTP_CANCELRPCS;
7944 			NFSUNLOCKMNT(nmp);
7945 			nfsrv_deleteds(fndds);
7946 		} else {
7947 			NFSUNLOCKMNT(nmp);
7948 			nmp = NULL;
7949 		}
7950 	}
7951 	NFSDDSUNLOCK();
7952 	if (nmp != NULL) {
7953 		nfsrv_flexmirrordel(fndds->nfsdev_deviceid, p);
7954 		printf("pNFS server: mirror %s failed\n", fndds->nfsdev_host);
7955 		nfsrv_killrpcs(nmp);
7956 		NFSLOCKMNT(nmp);
7957 		nmp->nm_privflag &= ~NFSMNTP_CANCELRPCS;
7958 		wakeup(nmp);
7959 		NFSUNLOCKMNT(nmp);
7960 		return (0);
7961 	}
7962 	return (ENXIO);
7963 }
7964 
7965 /*
7966  * Mark a DS as disabled by setting nfsdev_nmp = NULL.
7967  */
7968 static void
7969 nfsrv_deleteds(struct nfsdevice *fndds)
7970 {
7971 
7972 	NFSD_DEBUG(4, "deleteds: deleting a mirror\n");
7973 	fndds->nfsdev_nmp = NULL;
7974 	if (fndds->nfsdev_mdsisset == 0)
7975 		nfsrv_faildscnt--;
7976 }
7977 
7978 /*
7979  * Fill in the addr structures for the File and Flex File layouts.
7980  */
7981 static void
7982 nfsrv_allocdevid(struct nfsdevice *ds, char *addr, char *dnshost)
7983 {
7984 	uint32_t *tl;
7985 	char *netprot;
7986 	int addrlen;
7987 	static uint64_t new_devid = 0;
7988 
7989 	if (strchr(addr, ':') != NULL)
7990 		netprot = "tcp6";
7991 	else
7992 		netprot = "tcp";
7993 
7994 	/* Fill in the device id. */
7995 	NFSBCOPY(&nfsdev_time, ds->nfsdev_deviceid, sizeof(nfsdev_time));
7996 	new_devid++;
7997 	NFSBCOPY(&new_devid, &ds->nfsdev_deviceid[sizeof(nfsdev_time)],
7998 	    sizeof(new_devid));
7999 
8000 	/*
8001 	 * Fill in the file addr (actually the nfsv4_file_layout_ds_addr4
8002 	 * as defined in RFC5661) in XDR.
8003 	 */
8004 	addrlen = NFSM_RNDUP(strlen(addr)) + NFSM_RNDUP(strlen(netprot)) +
8005 	    6 * NFSX_UNSIGNED;
8006 	NFSD_DEBUG(4, "hn=%s addr=%s netprot=%s\n", dnshost, addr, netprot);
8007 	ds->nfsdev_fileaddrlen = addrlen;
8008 	tl = malloc(addrlen, M_NFSDSTATE, M_WAITOK | M_ZERO);
8009 	ds->nfsdev_fileaddr = (char *)tl;
8010 	*tl++ = txdr_unsigned(1);		/* One stripe with index 0. */
8011 	*tl++ = 0;
8012 	*tl++ = txdr_unsigned(1);		/* One multipath list */
8013 	*tl++ = txdr_unsigned(1);		/* with one entry in it. */
8014 	/* The netaddr for this one entry. */
8015 	*tl++ = txdr_unsigned(strlen(netprot));
8016 	NFSBCOPY(netprot, tl, strlen(netprot));
8017 	tl += (NFSM_RNDUP(strlen(netprot)) / NFSX_UNSIGNED);
8018 	*tl++ = txdr_unsigned(strlen(addr));
8019 	NFSBCOPY(addr, tl, strlen(addr));
8020 
8021 	/*
8022 	 * Fill in the flex file addr (actually the ff_device_addr4
8023 	 * as defined for Flexible File Layout) in XDR.
8024 	 */
8025 	addrlen = NFSM_RNDUP(strlen(addr)) + NFSM_RNDUP(strlen(netprot)) +
8026 	    14 * NFSX_UNSIGNED;
8027 	ds->nfsdev_flexaddrlen = addrlen;
8028 	tl = malloc(addrlen, M_NFSDSTATE, M_WAITOK | M_ZERO);
8029 	ds->nfsdev_flexaddr = (char *)tl;
8030 	*tl++ = txdr_unsigned(1);		/* One multipath entry. */
8031 	/* The netaddr for this one entry. */
8032 	*tl++ = txdr_unsigned(strlen(netprot));
8033 	NFSBCOPY(netprot, tl, strlen(netprot));
8034 	tl += (NFSM_RNDUP(strlen(netprot)) / NFSX_UNSIGNED);
8035 	*tl++ = txdr_unsigned(strlen(addr));
8036 	NFSBCOPY(addr, tl, strlen(addr));
8037 	tl += (NFSM_RNDUP(strlen(addr)) / NFSX_UNSIGNED);
8038 	*tl++ = txdr_unsigned(2);		/* Two NFS Versions. */
8039 	*tl++ = txdr_unsigned(NFS_VER4);	/* NFSv4. */
8040 	*tl++ = txdr_unsigned(NFSV42_MINORVERSION); /* Minor version 2. */
8041 	*tl++ = txdr_unsigned(nfs_srvmaxio);	/* DS max rsize. */
8042 	*tl++ = txdr_unsigned(nfs_srvmaxio);	/* DS max wsize. */
8043 	*tl++ = newnfs_true;			/* Tightly coupled. */
8044 	*tl++ = txdr_unsigned(NFS_VER4);	/* NFSv4. */
8045 	*tl++ = txdr_unsigned(NFSV41_MINORVERSION); /* Minor version 1. */
8046 	*tl++ = txdr_unsigned(nfs_srvmaxio);	/* DS max rsize. */
8047 	*tl++ = txdr_unsigned(nfs_srvmaxio);	/* DS max wsize. */
8048 	*tl = newnfs_true;			/* Tightly coupled. */
8049 
8050 	ds->nfsdev_hostnamelen = strlen(dnshost);
8051 	ds->nfsdev_host = malloc(ds->nfsdev_hostnamelen + 1, M_NFSDSTATE,
8052 	    M_WAITOK);
8053 	NFSBCOPY(dnshost, ds->nfsdev_host, ds->nfsdev_hostnamelen + 1);
8054 }
8055 
8056 /*
8057  * Create the device id list.
8058  * Return 0 if the nfsd threads are to run and ENXIO if the "-p" argument
8059  * is misconfigured.
8060  */
8061 int
8062 nfsrv_createdevids(struct nfsd_nfsd_args *args, NFSPROC_T *p)
8063 {
8064 	struct nfsdevice *ds;
8065 	char *addrp, *dnshostp, *dspathp, *mdspathp;
8066 	int error, i;
8067 
8068 	addrp = args->addr;
8069 	dnshostp = args->dnshost;
8070 	dspathp = args->dspath;
8071 	mdspathp = args->mdspath;
8072 	nfsrv_maxpnfsmirror = args->mirrorcnt;
8073 	if (addrp == NULL || dnshostp == NULL || dspathp == NULL ||
8074 	    mdspathp == NULL)
8075 		return (0);
8076 
8077 	/*
8078 	 * Loop around for each nul-terminated string in args->addr,
8079 	 * args->dnshost, args->dnspath and args->mdspath.
8080 	 */
8081 	while (addrp < (args->addr + args->addrlen) &&
8082 	    dnshostp < (args->dnshost + args->dnshostlen) &&
8083 	    dspathp < (args->dspath + args->dspathlen) &&
8084 	    mdspathp < (args->mdspath + args->mdspathlen)) {
8085 		error = nfsrv_setdsserver(dspathp, mdspathp, p, &ds);
8086 		if (error != 0) {
8087 			/* Free all DS servers. */
8088 			nfsrv_freealldevids();
8089 			nfsrv_devidcnt = 0;
8090 			return (ENXIO);
8091 		}
8092 		nfsrv_allocdevid(ds, addrp, dnshostp);
8093 		addrp += (strlen(addrp) + 1);
8094 		dnshostp += (strlen(dnshostp) + 1);
8095 		dspathp += (strlen(dspathp) + 1);
8096 		mdspathp += (strlen(mdspathp) + 1);
8097 	}
8098 	if (nfsrv_devidcnt < nfsrv_maxpnfsmirror) {
8099 		/* Free all DS servers. */
8100 		nfsrv_freealldevids();
8101 		nfsrv_devidcnt = 0;
8102 		nfsrv_maxpnfsmirror = 1;
8103 		return (ENXIO);
8104 	}
8105 	/* We can fail at most one less DS than the mirror level. */
8106 	nfsrv_faildscnt = nfsrv_maxpnfsmirror - 1;
8107 
8108 	/*
8109 	 * Allocate the nfslayout hash table now, since this is a pNFS server.
8110 	 * Make it 1% of the high water mark and at least 100.
8111 	 */
8112 	if (nfslayouthash == NULL) {
8113 		nfsrv_layouthashsize = nfsrv_layouthighwater / 100;
8114 		if (nfsrv_layouthashsize < 100)
8115 			nfsrv_layouthashsize = 100;
8116 		nfslayouthash = mallocarray(nfsrv_layouthashsize,
8117 		    sizeof(struct nfslayouthash), M_NFSDSESSION, M_WAITOK |
8118 		    M_ZERO);
8119 		for (i = 0; i < nfsrv_layouthashsize; i++) {
8120 			mtx_init(&nfslayouthash[i].mtx, "nfslm", NULL, MTX_DEF);
8121 			TAILQ_INIT(&nfslayouthash[i].list);
8122 		}
8123 	}
8124 	return (0);
8125 }
8126 
8127 /*
8128  * Free all device ids.
8129  */
8130 static void
8131 nfsrv_freealldevids(void)
8132 {
8133 	struct nfsdevice *ds, *nds;
8134 
8135 	TAILQ_FOREACH_SAFE(ds, &nfsrv_devidhead, nfsdev_list, nds)
8136 		nfsrv_freedevid(ds);
8137 }
8138 
8139 /*
8140  * Check to see if there is a Read/Write Layout plus either:
8141  * - A Write Delegation
8142  * or
8143  * - An Open with Write_access.
8144  * Return 1 if this is the case and 0 otherwise.
8145  * This function is used by nfsrv_proxyds() to decide if doing a Proxy
8146  * Getattr RPC to the Data Server (DS) is necessary.
8147  */
8148 #define	NFSCLIDVECSIZE	6
8149 int
8150 nfsrv_checkdsattr(vnode_t vp, NFSPROC_T *p)
8151 {
8152 	fhandle_t fh, *tfhp;
8153 	struct nfsstate *stp;
8154 	struct nfslayout *lyp;
8155 	struct nfslayouthash *lhyp;
8156 	struct nfslockhashhead *hp;
8157 	struct nfslockfile *lfp;
8158 	nfsquad_t clid[NFSCLIDVECSIZE];
8159 	int clidcnt, ret;
8160 
8161 	ret = nfsvno_getfh(vp, &fh, p);
8162 	if (ret != 0)
8163 		return (0);
8164 
8165 	/* First check for a Read/Write Layout. */
8166 	clidcnt = 0;
8167 	lhyp = NFSLAYOUTHASH(&fh);
8168 	NFSLOCKLAYOUT(lhyp);
8169 	TAILQ_FOREACH(lyp, &lhyp->list, lay_list) {
8170 		if (NFSBCMP(&lyp->lay_fh, &fh, sizeof(fh)) == 0 &&
8171 		    ((lyp->lay_flags & NFSLAY_RW) != 0 ||
8172 		     ((lyp->lay_flags & NFSLAY_READ) != 0 &&
8173 		      nfsrv_pnfsatime != 0))) {
8174 			if (clidcnt < NFSCLIDVECSIZE)
8175 				clid[clidcnt].qval = lyp->lay_clientid.qval;
8176 			clidcnt++;
8177 		}
8178 	}
8179 	NFSUNLOCKLAYOUT(lhyp);
8180 	if (clidcnt == 0) {
8181 		/* None found, so return 0. */
8182 		return (0);
8183 	}
8184 
8185 	/* Get the nfslockfile for this fh. */
8186 	NFSLOCKSTATE();
8187 	hp = NFSLOCKHASH(&fh);
8188 	LIST_FOREACH(lfp, hp, lf_hash) {
8189 		tfhp = &lfp->lf_fh;
8190 		if (NFSVNO_CMPFH(&fh, tfhp))
8191 			break;
8192 	}
8193 	if (lfp == NULL) {
8194 		/* None found, so return 0. */
8195 		NFSUNLOCKSTATE();
8196 		return (0);
8197 	}
8198 
8199 	/* Now, look for a Write delegation for this clientid. */
8200 	LIST_FOREACH(stp, &lfp->lf_deleg, ls_file) {
8201 		if ((stp->ls_flags & NFSLCK_DELEGWRITE) != 0 &&
8202 		    nfsrv_fndclid(clid, stp->ls_clp->lc_clientid, clidcnt) != 0)
8203 			break;
8204 	}
8205 	if (stp != NULL) {
8206 		/* Found one, so return 1. */
8207 		NFSUNLOCKSTATE();
8208 		return (1);
8209 	}
8210 
8211 	/* No Write delegation, so look for an Open with Write_access. */
8212 	LIST_FOREACH(stp, &lfp->lf_open, ls_file) {
8213 		KASSERT((stp->ls_flags & NFSLCK_OPEN) != 0,
8214 		    ("nfsrv_checkdsattr: Non-open in Open list\n"));
8215 		if ((stp->ls_flags & NFSLCK_WRITEACCESS) != 0 &&
8216 		    nfsrv_fndclid(clid, stp->ls_clp->lc_clientid, clidcnt) != 0)
8217 			break;
8218 	}
8219 	NFSUNLOCKSTATE();
8220 	if (stp != NULL)
8221 		return (1);
8222 	return (0);
8223 }
8224 
8225 /*
8226  * Look for a matching clientid in the vector. Return 1 if one might match.
8227  */
8228 static int
8229 nfsrv_fndclid(nfsquad_t *clidvec, nfsquad_t clid, int clidcnt)
8230 {
8231 	int i;
8232 
8233 	/* If too many for the vector, return 1 since there might be a match. */
8234 	if (clidcnt > NFSCLIDVECSIZE)
8235 		return (1);
8236 
8237 	for (i = 0; i < clidcnt; i++)
8238 		if (clidvec[i].qval == clid.qval)
8239 			return (1);
8240 	return (0);
8241 }
8242 
8243 /*
8244  * Check the don't list for "vp" and see if issuing an rw layout is allowed.
8245  * Return 1 if issuing an rw layout isn't allowed, 0 otherwise.
8246  */
8247 static int
8248 nfsrv_dontlayout(fhandle_t *fhp)
8249 {
8250 	struct nfsdontlist *mrp;
8251 	int ret;
8252 
8253 	if (nfsrv_dontlistlen == 0)
8254 		return (0);
8255 	ret = 0;
8256 	NFSDDONTLISTLOCK();
8257 	LIST_FOREACH(mrp, &nfsrv_dontlisthead, nfsmr_list) {
8258 		if (NFSBCMP(fhp, &mrp->nfsmr_fh, sizeof(*fhp)) == 0 &&
8259 		    (mrp->nfsmr_flags & NFSMR_DONTLAYOUT) != 0) {
8260 			ret = 1;
8261 			break;
8262 		}
8263 	}
8264 	NFSDDONTLISTUNLOCK();
8265 	return (ret);
8266 }
8267 
8268 #define	PNFSDS_COPYSIZ	65536
8269 /*
8270  * Create a new file on a DS and copy the contents of an extant DS file to it.
8271  * This can be used for recovery of a DS file onto a recovered DS.
8272  * The steps are:
8273  * - When called, the MDS file's vnode is locked, blocking LayoutGet operations.
8274  * - Disable issuing of read/write layouts for the file via the nfsdontlist,
8275  *   so that they will be disabled after the MDS file's vnode is unlocked.
8276  * - Set up the nfsrv_recalllist so that recall of read/write layouts can
8277  *   be done.
8278  * - Unlock the MDS file's vnode, so that the client(s) can perform proxied
8279  *   writes, LayoutCommits and LayoutReturns for the file when completing the
8280  *   LayoutReturn requested by the LayoutRecall callback.
8281  * - Issue a LayoutRecall callback for all read/write layouts and wait for
8282  *   them to be returned. (If the LayoutRecall callback replies
8283  *   NFSERR_NOMATCHLAYOUT, they are gone and no LayoutReturn is needed.)
8284  * - Exclusively lock the MDS file's vnode.  This ensures that no proxied
8285  *   writes are in progress or can occur during the DS file copy.
8286  *   It also blocks Setattr operations.
8287  * - Create the file on the recovered mirror.
8288  * - Copy the file from the operational DS.
8289  * - Copy any ACL from the MDS file to the new DS file.
8290  * - Set the modify time of the new DS file to that of the MDS file.
8291  * - Update the extended attribute for the MDS file.
8292  * - Enable issuing of rw layouts by deleting the nfsdontlist entry.
8293  * - The caller will unlock the MDS file's vnode allowing operations
8294  *   to continue normally, since it is now on the mirror again.
8295  */
8296 int
8297 nfsrv_copymr(vnode_t vp, vnode_t fvp, vnode_t dvp, struct nfsdevice *ds,
8298     struct pnfsdsfile *pf, struct pnfsdsfile *wpf, int mirrorcnt,
8299     struct ucred *cred, NFSPROC_T *p)
8300 {
8301 	struct nfsdontlist *mrp, *nmrp;
8302 	struct nfslayouthash *lhyp;
8303 	struct nfslayout *lyp, *nlyp;
8304 	struct nfslayouthead thl;
8305 	struct mount *mp, *tvmp;
8306 	struct acl *aclp;
8307 	struct vattr va;
8308 	struct timespec mtime;
8309 	fhandle_t fh;
8310 	vnode_t tvp;
8311 	off_t rdpos, wrpos;
8312 	ssize_t aresid;
8313 	char *dat;
8314 	int didprintf, ret, retacl, xfer;
8315 
8316 	ASSERT_VOP_LOCKED(fvp, "nfsrv_copymr fvp");
8317 	ASSERT_VOP_LOCKED(vp, "nfsrv_copymr vp");
8318 	/*
8319 	 * Allocate a nfsdontlist entry and set the NFSMR_DONTLAYOUT flag
8320 	 * so that no more RW layouts will get issued.
8321 	 */
8322 	ret = nfsvno_getfh(vp, &fh, p);
8323 	if (ret != 0) {
8324 		NFSD_DEBUG(4, "nfsrv_copymr: getfh=%d\n", ret);
8325 		return (ret);
8326 	}
8327 	nmrp = malloc(sizeof(*nmrp), M_NFSDSTATE, M_WAITOK);
8328 	nmrp->nfsmr_flags = NFSMR_DONTLAYOUT;
8329 	NFSBCOPY(&fh, &nmrp->nfsmr_fh, sizeof(fh));
8330 	NFSDDONTLISTLOCK();
8331 	LIST_FOREACH(mrp, &nfsrv_dontlisthead, nfsmr_list) {
8332 		if (NFSBCMP(&fh, &mrp->nfsmr_fh, sizeof(fh)) == 0)
8333 			break;
8334 	}
8335 	if (mrp == NULL) {
8336 		LIST_INSERT_HEAD(&nfsrv_dontlisthead, nmrp, nfsmr_list);
8337 		mrp = nmrp;
8338 		nmrp = NULL;
8339 		nfsrv_dontlistlen++;
8340 		NFSD_DEBUG(4, "nfsrv_copymr: in dontlist\n");
8341 	} else {
8342 		NFSDDONTLISTUNLOCK();
8343 		free(nmrp, M_NFSDSTATE);
8344 		NFSD_DEBUG(4, "nfsrv_copymr: dup dontlist\n");
8345 		return (ENXIO);
8346 	}
8347 	NFSDDONTLISTUNLOCK();
8348 
8349 	/*
8350 	 * Search for all RW layouts for this file.  Move them to the
8351 	 * recall list, so they can be recalled and their return noted.
8352 	 */
8353 	lhyp = NFSLAYOUTHASH(&fh);
8354 	NFSDRECALLLOCK();
8355 	NFSLOCKLAYOUT(lhyp);
8356 	TAILQ_FOREACH_SAFE(lyp, &lhyp->list, lay_list, nlyp) {
8357 		if (NFSBCMP(&lyp->lay_fh, &fh, sizeof(fh)) == 0 &&
8358 		    (lyp->lay_flags & NFSLAY_RW) != 0) {
8359 			TAILQ_REMOVE(&lhyp->list, lyp, lay_list);
8360 			TAILQ_INSERT_HEAD(&nfsrv_recalllisthead, lyp, lay_list);
8361 			lyp->lay_trycnt = 0;
8362 		}
8363 	}
8364 	NFSUNLOCKLAYOUT(lhyp);
8365 	NFSDRECALLUNLOCK();
8366 
8367 	ret = 0;
8368 	mp = tvmp = NULL;
8369 	didprintf = 0;
8370 	TAILQ_INIT(&thl);
8371 	/* Unlock the MDS vp, so that a LayoutReturn can be done on it. */
8372 	NFSVOPUNLOCK(vp);
8373 	/* Now, do a recall for all layouts not yet recalled. */
8374 tryagain:
8375 	NFSDRECALLLOCK();
8376 	TAILQ_FOREACH(lyp, &nfsrv_recalllisthead, lay_list) {
8377 		if (NFSBCMP(&lyp->lay_fh, &fh, sizeof(fh)) == 0 &&
8378 		    (lyp->lay_flags & NFSLAY_RECALL) == 0) {
8379 			lyp->lay_flags |= NFSLAY_RECALL;
8380 			/*
8381 			 * The layout stateid.seqid needs to be incremented
8382 			 * before doing a LAYOUT_RECALL callback.
8383 			 */
8384 			if (++lyp->lay_stateid.seqid == 0)
8385 				lyp->lay_stateid.seqid = 1;
8386 			NFSDRECALLUNLOCK();
8387 			nfsrv_recalllayout(lyp->lay_clientid, &lyp->lay_stateid,
8388 			    &lyp->lay_fh, lyp, 0, lyp->lay_type, p);
8389 			NFSD_DEBUG(4, "nfsrv_copymr: recalled layout\n");
8390 			goto tryagain;
8391 		}
8392 	}
8393 
8394 	/* Now wait for them to be returned. */
8395 tryagain2:
8396 	TAILQ_FOREACH(lyp, &nfsrv_recalllisthead, lay_list) {
8397 		if (NFSBCMP(&lyp->lay_fh, &fh, sizeof(fh)) == 0) {
8398 			if ((lyp->lay_flags & NFSLAY_RETURNED) != 0) {
8399 				TAILQ_REMOVE(&nfsrv_recalllisthead, lyp,
8400 				    lay_list);
8401 				TAILQ_INSERT_HEAD(&thl, lyp, lay_list);
8402 				NFSD_DEBUG(4,
8403 				    "nfsrv_copymr: layout returned\n");
8404 			} else {
8405 				lyp->lay_trycnt++;
8406 				ret = mtx_sleep(lyp, NFSDRECALLMUTEXPTR,
8407 				    PVFS | PCATCH, "nfsmrl", hz);
8408 				NFSD_DEBUG(4, "nfsrv_copymr: aft sleep=%d\n",
8409 				    ret);
8410 				if (ret == EINTR || ret == ERESTART)
8411 					break;
8412 				if ((lyp->lay_flags & NFSLAY_RETURNED) == 0) {
8413 					/*
8414 					 * Give up after 60sec and return
8415 					 * ENXIO, failing the copymr.
8416 					 * This layout will remain on the
8417 					 * recalllist.  It can only be cleared
8418 					 * by restarting the nfsd.
8419 					 * This seems the safe way to handle
8420 					 * it, since it cannot be safely copied
8421 					 * with an outstanding RW layout.
8422 					 */
8423 					if (lyp->lay_trycnt >= 60) {
8424 						ret = ENXIO;
8425 						break;
8426 					}
8427 					if (didprintf == 0) {
8428 						printf("nfsrv_copymr: layout "
8429 						    "not returned\n");
8430 						didprintf = 1;
8431 					}
8432 				}
8433 			}
8434 			goto tryagain2;
8435 		}
8436 	}
8437 	NFSDRECALLUNLOCK();
8438 	/* We can now get rid of the layouts that have been returned. */
8439 	TAILQ_FOREACH_SAFE(lyp, &thl, lay_list, nlyp)
8440 		nfsrv_freelayout(&thl, lyp);
8441 
8442 	/*
8443 	 * Do the vn_start_write() calls here, before the MDS vnode is
8444 	 * locked and the tvp is created (locked) in the NFS file system
8445 	 * that dvp is in.
8446 	 * For tvmp, this probably isn't necessary, since it will be an
8447 	 * NFS mount and they are not suspendable at this time.
8448 	 */
8449 	if (ret == 0)
8450 		ret = vn_start_write(vp, &mp, V_WAIT | V_PCATCH);
8451 	if (ret == 0) {
8452 		tvmp = dvp->v_mount;
8453 		ret = vn_start_write(NULL, &tvmp, V_WAIT | V_PCATCH);
8454 	}
8455 
8456 	/*
8457 	 * LK_EXCLUSIVE lock the MDS vnode, so that any
8458 	 * proxied writes through the MDS will be blocked until we have
8459 	 * completed the copy and update of the extended attributes.
8460 	 * This will also ensure that any attributes and ACL will not be
8461 	 * changed until the copy is complete.
8462 	 */
8463 	NFSVOPLOCK(vp, LK_EXCLUSIVE | LK_RETRY);
8464 	if (ret == 0 && VN_IS_DOOMED(vp)) {
8465 		NFSD_DEBUG(4, "nfsrv_copymr: lk_exclusive doomed\n");
8466 		ret = ESTALE;
8467 	}
8468 
8469 	/* Create the data file on the recovered DS. */
8470 	if (ret == 0)
8471 		ret = nfsrv_createdsfile(vp, &fh, pf, dvp, ds, cred, p, &tvp);
8472 
8473 	/* Copy the DS file, if created successfully. */
8474 	if (ret == 0) {
8475 		/*
8476 		 * Get any NFSv4 ACL on the MDS file, so that it can be set
8477 		 * on the new DS file.
8478 		 */
8479 		aclp = acl_alloc(M_WAITOK | M_ZERO);
8480 		retacl = VOP_GETACL(vp, ACL_TYPE_NFS4, aclp, cred, p);
8481 		if (retacl != 0 && retacl != ENOATTR)
8482 			NFSD_DEBUG(1, "nfsrv_copymr: vop_getacl=%d\n", retacl);
8483 		dat = malloc(PNFSDS_COPYSIZ, M_TEMP, M_WAITOK);
8484 		/* Malloc a block of 0s used to check for holes. */
8485 		if (nfsrv_zeropnfsdat == NULL)
8486 			nfsrv_zeropnfsdat = malloc(PNFSDS_COPYSIZ, M_TEMP,
8487 			    M_WAITOK | M_ZERO);
8488 		rdpos = wrpos = 0;
8489 		ret = VOP_GETATTR(fvp, &va, cred);
8490 		aresid = 0;
8491 		while (ret == 0 && aresid == 0) {
8492 			ret = vn_rdwr(UIO_READ, fvp, dat, PNFSDS_COPYSIZ,
8493 			    rdpos, UIO_SYSSPACE, IO_NODELOCKED, cred, NULL,
8494 			    &aresid, p);
8495 			xfer = PNFSDS_COPYSIZ - aresid;
8496 			if (ret == 0 && xfer > 0) {
8497 				rdpos += xfer;
8498 				/*
8499 				 * Skip the write for holes, except for the
8500 				 * last block.
8501 				 */
8502 				if (xfer < PNFSDS_COPYSIZ || rdpos ==
8503 				    va.va_size || NFSBCMP(dat,
8504 				    nfsrv_zeropnfsdat, PNFSDS_COPYSIZ) != 0)
8505 					ret = vn_rdwr(UIO_WRITE, tvp, dat, xfer,
8506 					    wrpos, UIO_SYSSPACE, IO_NODELOCKED,
8507 					    cred, NULL, NULL, p);
8508 				if (ret == 0)
8509 					wrpos += xfer;
8510 			}
8511 		}
8512 
8513 		/* If there is an ACL and the copy succeeded, set the ACL. */
8514 		if (ret == 0 && retacl == 0) {
8515 			ret = VOP_SETACL(tvp, ACL_TYPE_NFS4, aclp, cred, p);
8516 			/*
8517 			 * Don't consider these as errors, since VOP_GETACL()
8518 			 * can return an ACL when they are not actually
8519 			 * supported.  For example, for UFS, VOP_GETACL()
8520 			 * will return a trivial ACL based on the uid/gid/mode
8521 			 * when there is no ACL on the file.
8522 			 * This case should be recognized as a trivial ACL
8523 			 * by UFS's VOP_SETACL() and succeed, but...
8524 			 */
8525 			if (ret == ENOATTR || ret == EOPNOTSUPP || ret == EPERM)
8526 				ret = 0;
8527 		}
8528 
8529 		if (ret == 0)
8530 			ret = VOP_FSYNC(tvp, MNT_WAIT, p);
8531 
8532 		/* Set the DS data file's modify time that of the MDS file. */
8533 		if (ret == 0)
8534 			ret = VOP_GETATTR(vp, &va, cred);
8535 		if (ret == 0) {
8536 			mtime = va.va_mtime;
8537 			VATTR_NULL(&va);
8538 			va.va_mtime = mtime;
8539 			ret = VOP_SETATTR(tvp, &va, cred);
8540 		}
8541 
8542 		vput(tvp);
8543 		acl_free(aclp);
8544 		free(dat, M_TEMP);
8545 	}
8546 	if (tvmp != NULL)
8547 		vn_finished_write(tvmp);
8548 
8549 	/* Update the extended attributes for the newly created DS file. */
8550 	if (ret == 0)
8551 		ret = vn_extattr_set(vp, IO_NODELOCKED,
8552 		    EXTATTR_NAMESPACE_SYSTEM, "pnfsd.dsfile",
8553 		    sizeof(*wpf) * mirrorcnt, (char *)wpf, p);
8554 	if (mp != NULL)
8555 		vn_finished_write(mp);
8556 
8557 	/* Get rid of the dontlist entry, so that Layouts can be issued. */
8558 	NFSDDONTLISTLOCK();
8559 	LIST_REMOVE(mrp, nfsmr_list);
8560 	NFSDDONTLISTUNLOCK();
8561 	free(mrp, M_NFSDSTATE);
8562 	return (ret);
8563 }
8564 
8565 /*
8566  * Create a data storage file on the recovered DS.
8567  */
8568 static int
8569 nfsrv_createdsfile(vnode_t vp, fhandle_t *fhp, struct pnfsdsfile *pf,
8570     vnode_t dvp, struct nfsdevice *ds, struct ucred *cred, NFSPROC_T *p,
8571     vnode_t *tvpp)
8572 {
8573 	struct vattr va, nva;
8574 	int error;
8575 
8576 	/* Make data file name based on FH. */
8577 	error = VOP_GETATTR(vp, &va, cred);
8578 	if (error == 0) {
8579 		/* Set the attributes for "vp" to Setattr the DS vp. */
8580 		VATTR_NULL(&nva);
8581 		nva.va_uid = va.va_uid;
8582 		nva.va_gid = va.va_gid;
8583 		nva.va_mode = va.va_mode;
8584 		nva.va_size = 0;
8585 		VATTR_NULL(&va);
8586 		va.va_type = VREG;
8587 		va.va_mode = nva.va_mode;
8588 		NFSD_DEBUG(4, "nfsrv_dscreatefile: dvp=%p pf=%p\n", dvp, pf);
8589 		error = nfsrv_dscreate(dvp, &va, &nva, fhp, pf, NULL,
8590 		    pf->dsf_filename, cred, p, tvpp);
8591 	}
8592 	return (error);
8593 }
8594 
8595 /*
8596  * Look up the MDS file shared locked, and then get the extended attribute
8597  * to find the extant DS file to be copied to the new mirror.
8598  * If successful, *vpp is set to the MDS file's vp and *nvpp is
8599  * set to a DS data file for the MDS file, both exclusively locked.
8600  * The "buf" argument has the pnfsdsfile structure from the MDS file
8601  * in it and buflen is set to its length.
8602  */
8603 int
8604 nfsrv_mdscopymr(char *mdspathp, char *dspathp, char *curdspathp, char *buf,
8605     int *buflenp, char *fname, NFSPROC_T *p, struct vnode **vpp,
8606     struct vnode **nvpp, struct pnfsdsfile **pfp, struct nfsdevice **dsp,
8607     struct nfsdevice **fdsp)
8608 {
8609 	struct nameidata nd;
8610 	struct vnode *vp, *curvp;
8611 	struct pnfsdsfile *pf;
8612 	struct nfsmount *nmp, *curnmp;
8613 	int dsdir, error, mirrorcnt, ippos;
8614 
8615 	vp = NULL;
8616 	curvp = NULL;
8617 	curnmp = NULL;
8618 	*dsp = NULL;
8619 	*fdsp = NULL;
8620 	if (dspathp == NULL && curdspathp != NULL)
8621 		return (EPERM);
8622 
8623 	/*
8624 	 * Look up the MDS file shared locked.  The lock will be upgraded
8625 	 * to an exclusive lock after any rw layouts have been returned.
8626 	 */
8627 	NFSD_DEBUG(4, "mdsopen path=%s\n", mdspathp);
8628 	NDINIT(&nd, LOOKUP, FOLLOW | LOCKSHARED | LOCKLEAF, UIO_SYSSPACE,
8629 	    mdspathp);
8630 	error = namei(&nd);
8631 	NFSD_DEBUG(4, "lookup=%d\n", error);
8632 	if (error != 0)
8633 		return (error);
8634 	if (nd.ni_vp->v_type != VREG) {
8635 		vput(nd.ni_vp);
8636 		NFSD_DEBUG(4, "mdspath not reg\n");
8637 		return (EISDIR);
8638 	}
8639 	vp = nd.ni_vp;
8640 
8641 	if (curdspathp != NULL) {
8642 		/*
8643 		 * Look up the current DS path and find the nfsdev structure for
8644 		 * it.
8645 		 */
8646 		NFSD_DEBUG(4, "curmdsdev path=%s\n", curdspathp);
8647 		NDINIT(&nd, LOOKUP, FOLLOW | LOCKSHARED | LOCKLEAF,
8648 		    UIO_SYSSPACE, curdspathp);
8649 		error = namei(&nd);
8650 		NFSD_DEBUG(4, "ds lookup=%d\n", error);
8651 		if (error != 0) {
8652 			vput(vp);
8653 			return (error);
8654 		}
8655 		if (nd.ni_vp->v_type != VDIR) {
8656 			vput(nd.ni_vp);
8657 			vput(vp);
8658 			NFSD_DEBUG(4, "curdspath not dir\n");
8659 			return (ENOTDIR);
8660 		}
8661 		if (strcmp(nd.ni_vp->v_mount->mnt_vfc->vfc_name, "nfs") != 0) {
8662 			vput(nd.ni_vp);
8663 			vput(vp);
8664 			NFSD_DEBUG(4, "curdspath not an NFS mount\n");
8665 			return (ENXIO);
8666 		}
8667 		curnmp = VFSTONFS(nd.ni_vp->v_mount);
8668 
8669 		/* Search the nfsdev list for a match. */
8670 		NFSDDSLOCK();
8671 		*fdsp = nfsv4_findmirror(curnmp);
8672 		NFSDDSUNLOCK();
8673 		if (*fdsp == NULL)
8674 			curnmp = NULL;
8675 		if (curnmp == NULL) {
8676 			vput(nd.ni_vp);
8677 			vput(vp);
8678 			NFSD_DEBUG(4, "mdscopymr: no current ds\n");
8679 			return (ENXIO);
8680 		}
8681 		curvp = nd.ni_vp;
8682 	}
8683 
8684 	if (dspathp != NULL) {
8685 		/* Look up the nfsdev path and find the nfsdev structure. */
8686 		NFSD_DEBUG(4, "mdsdev path=%s\n", dspathp);
8687 		NDINIT(&nd, LOOKUP, FOLLOW | LOCKSHARED | LOCKLEAF,
8688 		    UIO_SYSSPACE, dspathp);
8689 		error = namei(&nd);
8690 		NFSD_DEBUG(4, "ds lookup=%d\n", error);
8691 		if (error != 0) {
8692 			vput(vp);
8693 			if (curvp != NULL)
8694 				vput(curvp);
8695 			return (error);
8696 		}
8697 		if (nd.ni_vp->v_type != VDIR || nd.ni_vp == curvp) {
8698 			vput(nd.ni_vp);
8699 			vput(vp);
8700 			if (curvp != NULL)
8701 				vput(curvp);
8702 			NFSD_DEBUG(4, "dspath not dir\n");
8703 			if (nd.ni_vp == curvp)
8704 				return (EPERM);
8705 			return (ENOTDIR);
8706 		}
8707 		if (strcmp(nd.ni_vp->v_mount->mnt_vfc->vfc_name, "nfs") != 0) {
8708 			vput(nd.ni_vp);
8709 			vput(vp);
8710 			if (curvp != NULL)
8711 				vput(curvp);
8712 			NFSD_DEBUG(4, "dspath not an NFS mount\n");
8713 			return (ENXIO);
8714 		}
8715 		nmp = VFSTONFS(nd.ni_vp->v_mount);
8716 
8717 		/*
8718 		 * Search the nfsdevice list for a match.  If curnmp == NULL,
8719 		 * this is a recovery and there must be a mirror.
8720 		 */
8721 		NFSDDSLOCK();
8722 		if (curnmp == NULL)
8723 			*dsp = nfsrv_findmirroredds(nmp);
8724 		else
8725 			*dsp = nfsv4_findmirror(nmp);
8726 		NFSDDSUNLOCK();
8727 		if (*dsp == NULL) {
8728 			vput(nd.ni_vp);
8729 			vput(vp);
8730 			if (curvp != NULL)
8731 				vput(curvp);
8732 			NFSD_DEBUG(4, "mdscopymr: no ds\n");
8733 			return (ENXIO);
8734 		}
8735 	} else {
8736 		nd.ni_vp = NULL;
8737 		nmp = NULL;
8738 	}
8739 
8740 	/*
8741 	 * Get a vp for an available DS data file using the extended
8742 	 * attribute on the MDS file.
8743 	 * If there is a valid entry for the new DS in the extended attribute
8744 	 * on the MDS file (as checked via the nmp argument),
8745 	 * nfsrv_dsgetsockmnt() returns EEXIST, so no copying will occur.
8746 	 */
8747 	error = nfsrv_dsgetsockmnt(vp, 0, buf, buflenp, &mirrorcnt, p,
8748 	    NULL, NULL, NULL, fname, nvpp, &nmp, curnmp, &ippos, &dsdir);
8749 	if (curvp != NULL)
8750 		vput(curvp);
8751 	if (nd.ni_vp == NULL) {
8752 		if (error == 0 && nmp != NULL) {
8753 			/* Search the nfsdev list for a match. */
8754 			NFSDDSLOCK();
8755 			*dsp = nfsrv_findmirroredds(nmp);
8756 			NFSDDSUNLOCK();
8757 		}
8758 		if (error == 0 && (nmp == NULL || *dsp == NULL)) {
8759 			if (nvpp != NULL && *nvpp != NULL) {
8760 				vput(*nvpp);
8761 				*nvpp = NULL;
8762 			}
8763 			error = ENXIO;
8764 		}
8765 	} else
8766 		vput(nd.ni_vp);
8767 
8768 	/*
8769 	 * When dspathp != NULL and curdspathp == NULL, this is a recovery
8770 	 * and is only allowed if there is a 0.0.0.0 IP address entry.
8771 	 * When curdspathp != NULL, the ippos will be set to that entry.
8772 	 */
8773 	if (error == 0 && dspathp != NULL && ippos == -1) {
8774 		if (nvpp != NULL && *nvpp != NULL) {
8775 			vput(*nvpp);
8776 			*nvpp = NULL;
8777 		}
8778 		error = ENXIO;
8779 	}
8780 	if (error == 0) {
8781 		*vpp = vp;
8782 
8783 		pf = (struct pnfsdsfile *)buf;
8784 		if (ippos == -1) {
8785 			/* If no zeroip pnfsdsfile, add one. */
8786 			ippos = *buflenp / sizeof(*pf);
8787 			*buflenp += sizeof(*pf);
8788 			pf += ippos;
8789 			pf->dsf_dir = dsdir;
8790 			strlcpy(pf->dsf_filename, fname,
8791 			    sizeof(pf->dsf_filename));
8792 		} else
8793 			pf += ippos;
8794 		*pfp = pf;
8795 	} else
8796 		vput(vp);
8797 	return (error);
8798 }
8799 
8800 /*
8801  * Search for a matching pnfsd mirror device structure, base on the nmp arg.
8802  * Return one if found, NULL otherwise.
8803  */
8804 static struct nfsdevice *
8805 nfsrv_findmirroredds(struct nfsmount *nmp)
8806 {
8807 	struct nfsdevice *ds, *fndds;
8808 	int fndmirror;
8809 
8810 	mtx_assert(NFSDDSMUTEXPTR, MA_OWNED);
8811 	/*
8812 	 * Search the DS server list for a match with nmp.
8813 	 * Remove the DS entry if found and there is a mirror.
8814 	 */
8815 	fndds = NULL;
8816 	fndmirror = 0;
8817 	if (nfsrv_devidcnt == 0)
8818 		return (fndds);
8819 	TAILQ_FOREACH(ds, &nfsrv_devidhead, nfsdev_list) {
8820 		if (ds->nfsdev_nmp == nmp) {
8821 			NFSD_DEBUG(4, "nfsrv_findmirroredds: fnd main ds\n");
8822 			fndds = ds;
8823 			break;
8824 		}
8825 	}
8826 	if (fndds == NULL)
8827 		return (fndds);
8828 	if (fndds->nfsdev_mdsisset == 0 && nfsrv_faildscnt > 0)
8829 		fndmirror = 1;
8830 	else if (fndds->nfsdev_mdsisset != 0) {
8831 		/* For the fsid is set case, search for a mirror. */
8832 		TAILQ_FOREACH(ds, &nfsrv_devidhead, nfsdev_list) {
8833 			if (ds != fndds && ds->nfsdev_nmp != NULL &&
8834 			    ds->nfsdev_mdsisset != 0 &&
8835 			    fsidcmp(&ds->nfsdev_mdsfsid,
8836 			    &fndds->nfsdev_mdsfsid) == 0) {
8837 				fndmirror = 1;
8838 				break;
8839 			}
8840 		}
8841 	}
8842 	if (fndmirror == 0) {
8843 		NFSD_DEBUG(4, "nfsrv_findmirroredds: no mirror for DS\n");
8844 		return (NULL);
8845 	}
8846 	return (fndds);
8847 }
8848 
8849 /*
8850  * Mark the appropriate devid and all associated layout as "out of space".
8851  */
8852 void
8853 nfsrv_marknospc(char *devid, bool setit)
8854 {
8855 	struct nfsdevice *ds;
8856 	struct nfslayout *lyp;
8857 	struct nfslayouthash *lhyp;
8858 	int i;
8859 
8860 	NFSDDSLOCK();
8861 	TAILQ_FOREACH(ds, &nfsrv_devidhead, nfsdev_list) {
8862 		if (NFSBCMP(ds->nfsdev_deviceid, devid, NFSX_V4DEVICEID) == 0) {
8863 			NFSD_DEBUG(1, "nfsrv_marknospc: devid %d\n", setit);
8864 			ds->nfsdev_nospc = setit;
8865 		}
8866 	}
8867 	NFSDDSUNLOCK();
8868 
8869 	for (i = 0; i < nfsrv_layouthashsize; i++) {
8870 		lhyp = &nfslayouthash[i];
8871 		NFSLOCKLAYOUT(lhyp);
8872 		TAILQ_FOREACH(lyp, &lhyp->list, lay_list) {
8873 			if (NFSBCMP(lyp->lay_deviceid, devid,
8874 			    NFSX_V4DEVICEID) == 0) {
8875 				NFSD_DEBUG(1, "nfsrv_marknospc: layout %d\n",
8876 				    setit);
8877 				if (setit)
8878 					lyp->lay_flags |= NFSLAY_NOSPC;
8879 				else
8880 					lyp->lay_flags &= ~NFSLAY_NOSPC;
8881 			}
8882 		}
8883 		NFSUNLOCKLAYOUT(lhyp);
8884 	}
8885 }
8886