xref: /freebsd/sys/fs/nfsserver/nfs_nfsdstate.c (revision 3332f1b444d4a73238e9f59cca27bfc95fe936bd)
1 /*-
2  * SPDX-License-Identifier: BSD-2-Clause-FreeBSD
3  *
4  * Copyright (c) 2009 Rick Macklem, University of Guelph
5  * All rights reserved.
6  *
7  * Redistribution and use in source and binary forms, with or without
8  * modification, are permitted provided that the following conditions
9  * are met:
10  * 1. Redistributions of source code must retain the above copyright
11  *    notice, this list of conditions and the following disclaimer.
12  * 2. Redistributions in binary form must reproduce the above copyright
13  *    notice, this list of conditions and the following disclaimer in the
14  *    documentation and/or other materials provided with the distribution.
15  *
16  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
17  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
18  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
19  * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
20  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
21  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
22  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
23  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
24  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
25  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
26  * SUCH DAMAGE.
27  *
28  */
29 
30 #include <sys/cdefs.h>
31 __FBSDID("$FreeBSD$");
32 
33 #include "opt_inet.h"
34 #include "opt_inet6.h"
35 #include <sys/extattr.h>
36 #include <fs/nfs/nfsport.h>
37 
38 struct nfsrv_stablefirst nfsrv_stablefirst;
39 int nfsrv_issuedelegs = 0;
40 int nfsrv_dolocallocks = 0;
41 struct nfsv4lock nfsv4rootfs_lock;
42 time_t nfsdev_time = 0;
43 int nfsrv_layouthashsize;
44 volatile int nfsrv_layoutcnt = 0;
45 extern uint32_t nfs_srvmaxio;
46 
47 extern int newnfs_numnfsd;
48 extern struct nfsstatsv1 nfsstatsv1;
49 extern int nfsrv_lease;
50 extern struct timeval nfsboottime;
51 extern u_int32_t newnfs_true, newnfs_false;
52 extern struct mtx nfsrv_dslock_mtx;
53 extern struct mtx nfsrv_recalllock_mtx;
54 extern struct mtx nfsrv_dontlistlock_mtx;
55 extern int nfsd_debuglevel;
56 extern u_int nfsrv_dsdirsize;
57 extern struct nfsdevicehead nfsrv_devidhead;
58 extern int nfsrv_doflexfile;
59 extern int nfsrv_maxpnfsmirror;
60 NFSV4ROOTLOCKMUTEX;
61 NFSSTATESPINLOCK;
62 extern struct nfsdontlisthead nfsrv_dontlisthead;
63 extern volatile int nfsrv_devidcnt;
64 extern struct nfslayouthead nfsrv_recalllisthead;
65 extern char *nfsrv_zeropnfsdat;
66 
67 SYSCTL_DECL(_vfs_nfsd);
68 int	nfsrv_statehashsize = NFSSTATEHASHSIZE;
69 SYSCTL_INT(_vfs_nfsd, OID_AUTO, statehashsize, CTLFLAG_RDTUN,
70     &nfsrv_statehashsize, 0,
71     "Size of state hash table set via loader.conf");
72 
73 int	nfsrv_clienthashsize = NFSCLIENTHASHSIZE;
74 SYSCTL_INT(_vfs_nfsd, OID_AUTO, clienthashsize, CTLFLAG_RDTUN,
75     &nfsrv_clienthashsize, 0,
76     "Size of client hash table set via loader.conf");
77 
78 int	nfsrv_lockhashsize = NFSLOCKHASHSIZE;
79 SYSCTL_INT(_vfs_nfsd, OID_AUTO, fhhashsize, CTLFLAG_RDTUN,
80     &nfsrv_lockhashsize, 0,
81     "Size of file handle hash table set via loader.conf");
82 
83 int	nfsrv_sessionhashsize = NFSSESSIONHASHSIZE;
84 SYSCTL_INT(_vfs_nfsd, OID_AUTO, sessionhashsize, CTLFLAG_RDTUN,
85     &nfsrv_sessionhashsize, 0,
86     "Size of session hash table set via loader.conf");
87 
88 int	nfsrv_layouthighwater = NFSLAYOUTHIGHWATER;
89 SYSCTL_INT(_vfs_nfsd, OID_AUTO, layouthighwater, CTLFLAG_RDTUN,
90     &nfsrv_layouthighwater, 0,
91     "High water mark for number of layouts set via loader.conf");
92 
93 static int	nfsrv_v4statelimit = NFSRV_V4STATELIMIT;
94 SYSCTL_INT(_vfs_nfsd, OID_AUTO, v4statelimit, CTLFLAG_RWTUN,
95     &nfsrv_v4statelimit, 0,
96     "High water limit for NFSv4 opens+locks+delegations");
97 
98 static int	nfsrv_writedelegifpos = 0;
99 SYSCTL_INT(_vfs_nfsd, OID_AUTO, writedelegifpos, CTLFLAG_RW,
100     &nfsrv_writedelegifpos, 0,
101     "Issue a write delegation for read opens if possible");
102 
103 static int	nfsrv_allowreadforwriteopen = 1;
104 SYSCTL_INT(_vfs_nfsd, OID_AUTO, allowreadforwriteopen, CTLFLAG_RW,
105     &nfsrv_allowreadforwriteopen, 0,
106     "Allow Reads to be done with Write Access StateIDs");
107 
108 int	nfsrv_pnfsatime = 0;
109 SYSCTL_INT(_vfs_nfsd, OID_AUTO, pnfsstrictatime, CTLFLAG_RW,
110     &nfsrv_pnfsatime, 0,
111     "For pNFS service, do Getattr ops to keep atime up-to-date");
112 
113 int	nfsrv_flexlinuxhack = 0;
114 SYSCTL_INT(_vfs_nfsd, OID_AUTO, flexlinuxhack, CTLFLAG_RW,
115     &nfsrv_flexlinuxhack, 0,
116     "For Linux clients, hack around Flex File Layout bug");
117 
118 /*
119  * Hash lists for nfs V4.
120  */
121 struct nfsclienthashhead	*nfsclienthash;
122 struct nfslockhashhead		*nfslockhash;
123 struct nfssessionhash		*nfssessionhash;
124 struct nfslayouthash		*nfslayouthash;
125 volatile int nfsrv_dontlistlen = 0;
126 
127 static u_int32_t nfsrv_openpluslock = 0, nfsrv_delegatecnt = 0;
128 static time_t nfsrvboottime;
129 static int nfsrv_returnoldstateid = 0, nfsrv_clients = 0;
130 static int nfsrv_clienthighwater = NFSRV_CLIENTHIGHWATER;
131 static int nfsrv_nogsscallback = 0;
132 static volatile int nfsrv_writedelegcnt = 0;
133 static int nfsrv_faildscnt;
134 
135 /* local functions */
136 static void nfsrv_dumpaclient(struct nfsclient *clp,
137     struct nfsd_dumpclients *dumpp);
138 static void nfsrv_freeopenowner(struct nfsstate *stp, int cansleep,
139     NFSPROC_T *p);
140 static int nfsrv_freeopen(struct nfsstate *stp, vnode_t vp, int cansleep,
141     NFSPROC_T *p);
142 static void nfsrv_freelockowner(struct nfsstate *stp, vnode_t vp, int cansleep,
143     NFSPROC_T *p);
144 static void nfsrv_freeallnfslocks(struct nfsstate *stp, vnode_t vp,
145     int cansleep, NFSPROC_T *p);
146 static void nfsrv_freenfslock(struct nfslock *lop);
147 static void nfsrv_freenfslockfile(struct nfslockfile *lfp);
148 static void nfsrv_freedeleg(struct nfsstate *);
149 static int nfsrv_getstate(struct nfsclient *clp, nfsv4stateid_t *stateidp,
150     u_int32_t flags, struct nfsstate **stpp);
151 static void nfsrv_getowner(struct nfsstatehead *hp, struct nfsstate *new_stp,
152     struct nfsstate **stpp);
153 static int nfsrv_getlockfh(vnode_t vp, u_short flags,
154     struct nfslockfile *new_lfp, fhandle_t *nfhp, NFSPROC_T *p);
155 static int nfsrv_getlockfile(u_short flags, struct nfslockfile **new_lfpp,
156     struct nfslockfile **lfpp, fhandle_t *nfhp, int lockit);
157 static void nfsrv_insertlock(struct nfslock *new_lop,
158     struct nfslock *insert_lop, struct nfsstate *stp, struct nfslockfile *lfp);
159 static void nfsrv_updatelock(struct nfsstate *stp, struct nfslock **new_lopp,
160     struct nfslock **other_lopp, struct nfslockfile *lfp);
161 static int nfsrv_getipnumber(u_char *cp);
162 static int nfsrv_checkrestart(nfsquad_t clientid, u_int32_t flags,
163     nfsv4stateid_t *stateidp, int specialid);
164 static int nfsrv_checkgrace(struct nfsrv_descript *nd, struct nfsclient *clp,
165     u_int32_t flags);
166 static int nfsrv_docallback(struct nfsclient *clp, int procnum,
167     nfsv4stateid_t *stateidp, int trunc, fhandle_t *fhp,
168     struct nfsvattr *nap, nfsattrbit_t *attrbitp, int laytype, NFSPROC_T *p);
169 static int nfsrv_cbcallargs(struct nfsrv_descript *nd, struct nfsclient *clp,
170     uint32_t callback, int op, const char *optag, struct nfsdsession **sepp,
171     int *slotposp);
172 static u_int32_t nfsrv_nextclientindex(void);
173 static u_int32_t nfsrv_nextstateindex(struct nfsclient *clp);
174 static void nfsrv_markstable(struct nfsclient *clp);
175 static void nfsrv_markreclaim(struct nfsclient *clp);
176 static int nfsrv_checkstable(struct nfsclient *clp);
177 static int nfsrv_clientconflict(struct nfsclient *clp, int *haslockp, struct
178     vnode *vp, NFSPROC_T *p);
179 static int nfsrv_delegconflict(struct nfsstate *stp, int *haslockp,
180     NFSPROC_T *p, vnode_t vp);
181 static int nfsrv_cleandeleg(vnode_t vp, struct nfslockfile *lfp,
182     struct nfsclient *clp, int *haslockp, NFSPROC_T *p);
183 static int nfsrv_notsamecredname(struct nfsrv_descript *nd,
184     struct nfsclient *clp);
185 static time_t nfsrv_leaseexpiry(void);
186 static void nfsrv_delaydelegtimeout(struct nfsstate *stp);
187 static int nfsrv_checkseqid(struct nfsrv_descript *nd, u_int32_t seqid,
188     struct nfsstate *stp, struct nfsrvcache *op);
189 static int nfsrv_nootherstate(struct nfsstate *stp);
190 static int nfsrv_locallock(vnode_t vp, struct nfslockfile *lfp, int flags,
191     uint64_t first, uint64_t end, struct nfslockconflict *cfp, NFSPROC_T *p);
192 static void nfsrv_localunlock(vnode_t vp, struct nfslockfile *lfp,
193     uint64_t init_first, uint64_t init_end, NFSPROC_T *p);
194 static int nfsrv_dolocal(vnode_t vp, struct nfslockfile *lfp, int flags,
195     int oldflags, uint64_t first, uint64_t end, struct nfslockconflict *cfp,
196     NFSPROC_T *p);
197 static void nfsrv_locallock_rollback(vnode_t vp, struct nfslockfile *lfp,
198     NFSPROC_T *p);
199 static void nfsrv_locallock_commit(struct nfslockfile *lfp, int flags,
200     uint64_t first, uint64_t end);
201 static void nfsrv_locklf(struct nfslockfile *lfp);
202 static void nfsrv_unlocklf(struct nfslockfile *lfp);
203 static struct nfsdsession *nfsrv_findsession(uint8_t *sessionid);
204 static int nfsrv_freesession(struct nfsdsession *sep, uint8_t *sessionid);
205 static int nfsv4_setcbsequence(struct nfsrv_descript *nd, struct nfsclient *clp,
206     int dont_replycache, struct nfsdsession **sepp, int *slotposp);
207 static int nfsv4_getcbsession(struct nfsclient *clp, struct nfsdsession **sepp);
208 static int nfsrv_addlayout(struct nfsrv_descript *nd, struct nfslayout **lypp,
209     nfsv4stateid_t *stateidp, char *layp, int *layoutlenp, NFSPROC_T *p);
210 static void nfsrv_freelayout(struct nfslayouthead *lhp, struct nfslayout *lyp);
211 static void nfsrv_freelayoutlist(nfsquad_t clientid);
212 static void nfsrv_freelayouts(nfsquad_t *clid, fsid_t *fs, int laytype,
213     int iomode);
214 static void nfsrv_freealllayouts(void);
215 static void nfsrv_freedevid(struct nfsdevice *ds);
216 static int nfsrv_setdsserver(char *dspathp, char *mdspathp, NFSPROC_T *p,
217     struct nfsdevice **dsp);
218 static void nfsrv_deleteds(struct nfsdevice *fndds);
219 static void nfsrv_allocdevid(struct nfsdevice *ds, char *addr, char *dnshost);
220 static void nfsrv_freealldevids(void);
221 static void nfsrv_flexlayouterr(struct nfsrv_descript *nd, uint32_t *layp,
222     int maxcnt, NFSPROC_T *p);
223 static int nfsrv_recalllayout(nfsquad_t clid, nfsv4stateid_t *stateidp,
224     fhandle_t *fhp, struct nfslayout *lyp, int changed, int laytype,
225     NFSPROC_T *p);
226 static int nfsrv_findlayout(nfsquad_t *clientidp, fhandle_t *fhp, int laytype,
227     NFSPROC_T *, struct nfslayout **lypp);
228 static int nfsrv_fndclid(nfsquad_t *clidvec, nfsquad_t clid, int clidcnt);
229 static struct nfslayout *nfsrv_filelayout(struct nfsrv_descript *nd, int iomode,
230     fhandle_t *fhp, fhandle_t *dsfhp, char *devid, fsid_t fs);
231 static struct nfslayout *nfsrv_flexlayout(struct nfsrv_descript *nd, int iomode,
232     int mirrorcnt, fhandle_t *fhp, fhandle_t *dsfhp, char *devid, fsid_t fs);
233 static int nfsrv_dontlayout(fhandle_t *fhp);
234 static int nfsrv_createdsfile(vnode_t vp, fhandle_t *fhp, struct pnfsdsfile *pf,
235     vnode_t dvp, struct nfsdevice *ds, struct ucred *cred, NFSPROC_T *p,
236     vnode_t *tvpp);
237 static struct nfsdevice *nfsrv_findmirroredds(struct nfsmount *nmp);
238 
239 /*
240  * Scan the client list for a match and either return the current one,
241  * create a new entry or return an error.
242  * If returning a non-error, the clp structure must either be linked into
243  * the client list or free'd.
244  */
245 int
246 nfsrv_setclient(struct nfsrv_descript *nd, struct nfsclient **new_clpp,
247     nfsquad_t *clientidp, nfsquad_t *confirmp, NFSPROC_T *p)
248 {
249 	struct nfsclient *clp = NULL, *new_clp = *new_clpp;
250 	int i, error = 0, ret;
251 	struct nfsstate *stp, *tstp;
252 #ifdef INET
253 	struct sockaddr_in *sin, *rin;
254 #endif
255 #ifdef INET6
256 	struct sockaddr_in6 *sin6, *rin6;
257 #endif
258 	struct nfsdsession *sep, *nsep;
259 	int zapit = 0, gotit, hasstate = 0, igotlock;
260 	static u_int64_t confirm_index = 0;
261 
262 	/*
263 	 * Check for state resource limit exceeded.
264 	 */
265 	if (nfsrv_openpluslock > nfsrv_v4statelimit) {
266 		error = NFSERR_RESOURCE;
267 		goto out;
268 	}
269 
270 	if (nfsrv_issuedelegs == 0 ||
271 	    ((nd->nd_flag & ND_GSS) != 0 && nfsrv_nogsscallback != 0))
272 		/*
273 		 * Don't do callbacks when delegations are disabled or
274 		 * for AUTH_GSS unless enabled via nfsrv_nogsscallback.
275 		 * If establishing a callback connection is attempted
276 		 * when a firewall is blocking the callback path, the
277 		 * server may wait too long for the connect attempt to
278 		 * succeed during the Open. Some clients, such as Linux,
279 		 * may timeout and give up on the Open before the server
280 		 * replies. Also, since AUTH_GSS callbacks are not
281 		 * yet interoperability tested, they might cause the
282 		 * server to crap out, if they get past the Init call to
283 		 * the client.
284 		 */
285 		new_clp->lc_program = 0;
286 
287 	/* Lock out other nfsd threads */
288 	NFSLOCKV4ROOTMUTEX();
289 	nfsv4_relref(&nfsv4rootfs_lock);
290 	do {
291 		igotlock = nfsv4_lock(&nfsv4rootfs_lock, 1, NULL,
292 		    NFSV4ROOTLOCKMUTEXPTR, NULL);
293 	} while (!igotlock);
294 	NFSUNLOCKV4ROOTMUTEX();
295 
296 	/*
297 	 * Search for a match in the client list.
298 	 */
299 	gotit = i = 0;
300 	while (i < nfsrv_clienthashsize && !gotit) {
301 	    LIST_FOREACH(clp, &nfsclienthash[i], lc_hash) {
302 		if (new_clp->lc_idlen == clp->lc_idlen &&
303 		    !NFSBCMP(new_clp->lc_id, clp->lc_id, clp->lc_idlen)) {
304 			gotit = 1;
305 			break;
306 		}
307 	    }
308 	    if (gotit == 0)
309 		i++;
310 	}
311 	if (!gotit ||
312 	    (clp->lc_flags & (LCL_NEEDSCONFIRM | LCL_ADMINREVOKED))) {
313 		if ((nd->nd_flag & ND_NFSV41) != 0 && confirmp->lval[1] != 0) {
314 			/*
315 			 * For NFSv4.1, if confirmp->lval[1] is non-zero, the
316 			 * client is trying to update a confirmed clientid.
317 			 */
318 			NFSLOCKV4ROOTMUTEX();
319 			nfsv4_unlock(&nfsv4rootfs_lock, 1);
320 			NFSUNLOCKV4ROOTMUTEX();
321 			confirmp->lval[1] = 0;
322 			error = NFSERR_NOENT;
323 			goto out;
324 		}
325 		/*
326 		 * Get rid of the old one.
327 		 */
328 		if (i != nfsrv_clienthashsize) {
329 			LIST_REMOVE(clp, lc_hash);
330 			nfsrv_cleanclient(clp, p);
331 			nfsrv_freedeleglist(&clp->lc_deleg);
332 			nfsrv_freedeleglist(&clp->lc_olddeleg);
333 			zapit = 1;
334 		}
335 		/*
336 		 * Add it after assigning a client id to it.
337 		 */
338 		new_clp->lc_flags |= LCL_NEEDSCONFIRM;
339 		if ((nd->nd_flag & ND_NFSV41) != 0)
340 			new_clp->lc_confirm.lval[0] = confirmp->lval[0] =
341 			    ++confirm_index;
342 		else
343 			confirmp->qval = new_clp->lc_confirm.qval =
344 			    ++confirm_index;
345 		clientidp->lval[0] = new_clp->lc_clientid.lval[0] =
346 		    (u_int32_t)nfsrvboottime;
347 		clientidp->lval[1] = new_clp->lc_clientid.lval[1] =
348 		    nfsrv_nextclientindex();
349 		new_clp->lc_stateindex = 0;
350 		new_clp->lc_statemaxindex = 0;
351 		new_clp->lc_cbref = 0;
352 		new_clp->lc_expiry = nfsrv_leaseexpiry();
353 		LIST_INIT(&new_clp->lc_open);
354 		LIST_INIT(&new_clp->lc_deleg);
355 		LIST_INIT(&new_clp->lc_olddeleg);
356 		LIST_INIT(&new_clp->lc_session);
357 		for (i = 0; i < nfsrv_statehashsize; i++)
358 			LIST_INIT(&new_clp->lc_stateid[i]);
359 		LIST_INSERT_HEAD(NFSCLIENTHASH(new_clp->lc_clientid), new_clp,
360 		    lc_hash);
361 		nfsstatsv1.srvclients++;
362 		nfsrv_openpluslock++;
363 		nfsrv_clients++;
364 		NFSLOCKV4ROOTMUTEX();
365 		nfsv4_unlock(&nfsv4rootfs_lock, 1);
366 		NFSUNLOCKV4ROOTMUTEX();
367 		if (zapit)
368 			nfsrv_zapclient(clp, p);
369 		*new_clpp = NULL;
370 		goto out;
371 	}
372 
373 	/*
374 	 * Now, handle the cases where the id is already issued.
375 	 */
376 	if (nfsrv_notsamecredname(nd, clp)) {
377 	    /*
378 	     * Check to see if there is expired state that should go away.
379 	     */
380 	    if (clp->lc_expiry < NFSD_MONOSEC &&
381 	        (!LIST_EMPTY(&clp->lc_open) || !LIST_EMPTY(&clp->lc_deleg))) {
382 		nfsrv_cleanclient(clp, p);
383 		nfsrv_freedeleglist(&clp->lc_deleg);
384 	    }
385 
386 	    /*
387 	     * If there is outstanding state, then reply NFSERR_CLIDINUSE per
388 	     * RFC3530 Sec. 8.1.2 last para.
389 	     */
390 	    if (!LIST_EMPTY(&clp->lc_deleg)) {
391 		hasstate = 1;
392 	    } else if (LIST_EMPTY(&clp->lc_open)) {
393 		hasstate = 0;
394 	    } else {
395 		hasstate = 0;
396 		/* Look for an Open on the OpenOwner */
397 		LIST_FOREACH(stp, &clp->lc_open, ls_list) {
398 		    if (!LIST_EMPTY(&stp->ls_open)) {
399 			hasstate = 1;
400 			break;
401 		    }
402 		}
403 	    }
404 	    if (hasstate) {
405 		/*
406 		 * If the uid doesn't match, return NFSERR_CLIDINUSE after
407 		 * filling out the correct ipaddr and portnum.
408 		 */
409 		switch (clp->lc_req.nr_nam->sa_family) {
410 #ifdef INET
411 		case AF_INET:
412 			sin = (struct sockaddr_in *)new_clp->lc_req.nr_nam;
413 			rin = (struct sockaddr_in *)clp->lc_req.nr_nam;
414 			sin->sin_addr.s_addr = rin->sin_addr.s_addr;
415 			sin->sin_port = rin->sin_port;
416 			break;
417 #endif
418 #ifdef INET6
419 		case AF_INET6:
420 			sin6 = (struct sockaddr_in6 *)new_clp->lc_req.nr_nam;
421 			rin6 = (struct sockaddr_in6 *)clp->lc_req.nr_nam;
422 			sin6->sin6_addr = rin6->sin6_addr;
423 			sin6->sin6_port = rin6->sin6_port;
424 			break;
425 #endif
426 		}
427 		NFSLOCKV4ROOTMUTEX();
428 		nfsv4_unlock(&nfsv4rootfs_lock, 1);
429 		NFSUNLOCKV4ROOTMUTEX();
430 		error = NFSERR_CLIDINUSE;
431 		goto out;
432 	    }
433 	}
434 
435 	if (NFSBCMP(new_clp->lc_verf, clp->lc_verf, NFSX_VERF)) {
436 		/*
437 		 * If the verifier has changed, the client has rebooted
438 		 * and a new client id is issued. The old state info
439 		 * can be thrown away once the SETCLIENTID_CONFIRM occurs.
440 		 */
441 		LIST_REMOVE(clp, lc_hash);
442 
443 		/* Get rid of all sessions on this clientid. */
444 		LIST_FOREACH_SAFE(sep, &clp->lc_session, sess_list, nsep) {
445 			ret = nfsrv_freesession(sep, NULL);
446 			if (ret != 0)
447 				printf("nfsrv_setclient: verifier changed free"
448 				    " session failed=%d\n", ret);
449 		}
450 
451 		new_clp->lc_flags |= LCL_NEEDSCONFIRM;
452 		if ((nd->nd_flag & ND_NFSV41) != 0)
453 			new_clp->lc_confirm.lval[0] = confirmp->lval[0] =
454 			    ++confirm_index;
455 		else
456 			confirmp->qval = new_clp->lc_confirm.qval =
457 			    ++confirm_index;
458 		clientidp->lval[0] = new_clp->lc_clientid.lval[0] =
459 		    nfsrvboottime;
460 		clientidp->lval[1] = new_clp->lc_clientid.lval[1] =
461 		    nfsrv_nextclientindex();
462 		new_clp->lc_stateindex = 0;
463 		new_clp->lc_statemaxindex = 0;
464 		new_clp->lc_cbref = 0;
465 		new_clp->lc_expiry = nfsrv_leaseexpiry();
466 
467 		/*
468 		 * Save the state until confirmed.
469 		 */
470 		LIST_NEWHEAD(&new_clp->lc_open, &clp->lc_open, ls_list);
471 		LIST_FOREACH(tstp, &new_clp->lc_open, ls_list)
472 			tstp->ls_clp = new_clp;
473 		LIST_NEWHEAD(&new_clp->lc_deleg, &clp->lc_deleg, ls_list);
474 		LIST_FOREACH(tstp, &new_clp->lc_deleg, ls_list)
475 			tstp->ls_clp = new_clp;
476 		LIST_NEWHEAD(&new_clp->lc_olddeleg, &clp->lc_olddeleg,
477 		    ls_list);
478 		LIST_FOREACH(tstp, &new_clp->lc_olddeleg, ls_list)
479 			tstp->ls_clp = new_clp;
480 		for (i = 0; i < nfsrv_statehashsize; i++) {
481 			LIST_NEWHEAD(&new_clp->lc_stateid[i],
482 			    &clp->lc_stateid[i], ls_hash);
483 			LIST_FOREACH(tstp, &new_clp->lc_stateid[i], ls_hash)
484 				tstp->ls_clp = new_clp;
485 		}
486 		LIST_INIT(&new_clp->lc_session);
487 		LIST_INSERT_HEAD(NFSCLIENTHASH(new_clp->lc_clientid), new_clp,
488 		    lc_hash);
489 		nfsstatsv1.srvclients++;
490 		nfsrv_openpluslock++;
491 		nfsrv_clients++;
492 		NFSLOCKV4ROOTMUTEX();
493 		nfsv4_unlock(&nfsv4rootfs_lock, 1);
494 		NFSUNLOCKV4ROOTMUTEX();
495 
496 		/*
497 		 * Must wait until any outstanding callback on the old clp
498 		 * completes.
499 		 */
500 		NFSLOCKSTATE();
501 		while (clp->lc_cbref) {
502 			clp->lc_flags |= LCL_WAKEUPWANTED;
503 			(void)mtx_sleep(clp, NFSSTATEMUTEXPTR, PZERO - 1,
504 			    "nfsd clp", 10 * hz);
505 		}
506 		NFSUNLOCKSTATE();
507 		nfsrv_zapclient(clp, p);
508 		*new_clpp = NULL;
509 		goto out;
510 	}
511 
512 	/* For NFSv4.1, mark that we found a confirmed clientid. */
513 	if ((nd->nd_flag & ND_NFSV41) != 0) {
514 		clientidp->lval[0] = clp->lc_clientid.lval[0];
515 		clientidp->lval[1] = clp->lc_clientid.lval[1];
516 		confirmp->lval[0] = 0;	/* Ignored by client */
517 		confirmp->lval[1] = 1;
518 	} else {
519 		/*
520 		 * id and verifier match, so update the net address info
521 		 * and get rid of any existing callback authentication
522 		 * handle, so a new one will be acquired.
523 		 */
524 		LIST_REMOVE(clp, lc_hash);
525 		new_clp->lc_flags |= (LCL_NEEDSCONFIRM | LCL_DONTCLEAN);
526 		new_clp->lc_expiry = nfsrv_leaseexpiry();
527 		confirmp->qval = new_clp->lc_confirm.qval = ++confirm_index;
528 		clientidp->lval[0] = new_clp->lc_clientid.lval[0] =
529 		    clp->lc_clientid.lval[0];
530 		clientidp->lval[1] = new_clp->lc_clientid.lval[1] =
531 		    clp->lc_clientid.lval[1];
532 		new_clp->lc_delegtime = clp->lc_delegtime;
533 		new_clp->lc_stateindex = clp->lc_stateindex;
534 		new_clp->lc_statemaxindex = clp->lc_statemaxindex;
535 		new_clp->lc_cbref = 0;
536 		LIST_NEWHEAD(&new_clp->lc_open, &clp->lc_open, ls_list);
537 		LIST_FOREACH(tstp, &new_clp->lc_open, ls_list)
538 			tstp->ls_clp = new_clp;
539 		LIST_NEWHEAD(&new_clp->lc_deleg, &clp->lc_deleg, ls_list);
540 		LIST_FOREACH(tstp, &new_clp->lc_deleg, ls_list)
541 			tstp->ls_clp = new_clp;
542 		LIST_NEWHEAD(&new_clp->lc_olddeleg, &clp->lc_olddeleg, ls_list);
543 		LIST_FOREACH(tstp, &new_clp->lc_olddeleg, ls_list)
544 			tstp->ls_clp = new_clp;
545 		for (i = 0; i < nfsrv_statehashsize; i++) {
546 			LIST_NEWHEAD(&new_clp->lc_stateid[i],
547 			    &clp->lc_stateid[i], ls_hash);
548 			LIST_FOREACH(tstp, &new_clp->lc_stateid[i], ls_hash)
549 				tstp->ls_clp = new_clp;
550 		}
551 		LIST_INIT(&new_clp->lc_session);
552 		LIST_INSERT_HEAD(NFSCLIENTHASH(new_clp->lc_clientid), new_clp,
553 		    lc_hash);
554 		nfsstatsv1.srvclients++;
555 		nfsrv_openpluslock++;
556 		nfsrv_clients++;
557 	}
558 	NFSLOCKV4ROOTMUTEX();
559 	nfsv4_unlock(&nfsv4rootfs_lock, 1);
560 	NFSUNLOCKV4ROOTMUTEX();
561 
562 	if ((nd->nd_flag & ND_NFSV41) == 0) {
563 		/*
564 		 * Must wait until any outstanding callback on the old clp
565 		 * completes.
566 		 */
567 		NFSLOCKSTATE();
568 		while (clp->lc_cbref) {
569 			clp->lc_flags |= LCL_WAKEUPWANTED;
570 			(void)mtx_sleep(clp, NFSSTATEMUTEXPTR, PZERO - 1,
571 			    "nfsdclp", 10 * hz);
572 		}
573 		NFSUNLOCKSTATE();
574 		nfsrv_zapclient(clp, p);
575 		*new_clpp = NULL;
576 	}
577 
578 out:
579 	NFSEXITCODE2(error, nd);
580 	return (error);
581 }
582 
583 /*
584  * Check to see if the client id exists and optionally confirm it.
585  */
586 int
587 nfsrv_getclient(nfsquad_t clientid, int opflags, struct nfsclient **clpp,
588     struct nfsdsession *nsep, nfsquad_t confirm, uint32_t cbprogram,
589     struct nfsrv_descript *nd, NFSPROC_T *p)
590 {
591 	struct nfsclient *clp;
592 	struct nfsstate *stp;
593 	int i;
594 	struct nfsclienthashhead *hp;
595 	int error = 0, igotlock, doneok;
596 	struct nfssessionhash *shp;
597 	struct nfsdsession *sep;
598 	uint64_t sessid[2];
599 	static uint64_t next_sess = 0;
600 
601 	if (clpp)
602 		*clpp = NULL;
603 	if ((nd == NULL || (nd->nd_flag & ND_NFSV41) == 0 ||
604 	    opflags != CLOPS_RENEW) && nfsrvboottime != clientid.lval[0]) {
605 		error = NFSERR_STALECLIENTID;
606 		goto out;
607 	}
608 
609 	/*
610 	 * If called with opflags == CLOPS_RENEW, the State Lock is
611 	 * already held. Otherwise, we need to get either that or,
612 	 * for the case of Confirm, lock out the nfsd threads.
613 	 */
614 	if (opflags & CLOPS_CONFIRM) {
615 		NFSLOCKV4ROOTMUTEX();
616 		nfsv4_relref(&nfsv4rootfs_lock);
617 		do {
618 			igotlock = nfsv4_lock(&nfsv4rootfs_lock, 1, NULL,
619 			    NFSV4ROOTLOCKMUTEXPTR, NULL);
620 		} while (!igotlock);
621 		/*
622 		 * Create a new sessionid here, since we need to do it where
623 		 * there is a mutex held to serialize update of next_sess.
624 		 */
625 		if ((nd->nd_flag & ND_NFSV41) != 0) {
626 			sessid[0] = ++next_sess;
627 			sessid[1] = clientid.qval;
628 		}
629 		NFSUNLOCKV4ROOTMUTEX();
630 	} else if (opflags != CLOPS_RENEW) {
631 		NFSLOCKSTATE();
632 	}
633 
634 	/* For NFSv4.1, the clp is acquired from the associated session. */
635 	if (nd != NULL && (nd->nd_flag & ND_NFSV41) != 0 &&
636 	    opflags == CLOPS_RENEW) {
637 		clp = NULL;
638 		if ((nd->nd_flag & ND_HASSEQUENCE) != 0) {
639 			shp = NFSSESSIONHASH(nd->nd_sessionid);
640 			NFSLOCKSESSION(shp);
641 			sep = nfsrv_findsession(nd->nd_sessionid);
642 			if (sep != NULL)
643 				clp = sep->sess_clp;
644 			NFSUNLOCKSESSION(shp);
645 		}
646 	} else {
647 		hp = NFSCLIENTHASH(clientid);
648 		LIST_FOREACH(clp, hp, lc_hash) {
649 			if (clp->lc_clientid.lval[1] == clientid.lval[1])
650 				break;
651 		}
652 	}
653 	if (clp == NULL) {
654 		if (opflags & CLOPS_CONFIRM)
655 			error = NFSERR_STALECLIENTID;
656 		else
657 			error = NFSERR_EXPIRED;
658 	} else if (clp->lc_flags & LCL_ADMINREVOKED) {
659 		/*
660 		 * If marked admin revoked, just return the error.
661 		 */
662 		error = NFSERR_ADMINREVOKED;
663 	}
664 	if (error) {
665 		if (opflags & CLOPS_CONFIRM) {
666 			NFSLOCKV4ROOTMUTEX();
667 			nfsv4_unlock(&nfsv4rootfs_lock, 1);
668 			NFSUNLOCKV4ROOTMUTEX();
669 		} else if (opflags != CLOPS_RENEW) {
670 			NFSUNLOCKSTATE();
671 		}
672 		goto out;
673 	}
674 
675 	/*
676 	 * Perform any operations specified by the opflags.
677 	 */
678 	if (opflags & CLOPS_CONFIRM) {
679 		if (((nd->nd_flag & ND_NFSV41) != 0 &&
680 		     clp->lc_confirm.lval[0] != confirm.lval[0]) ||
681 		    ((nd->nd_flag & ND_NFSV41) == 0 &&
682 		     clp->lc_confirm.qval != confirm.qval))
683 			error = NFSERR_STALECLIENTID;
684 		else if (nfsrv_notsamecredname(nd, clp))
685 			error = NFSERR_CLIDINUSE;
686 
687 		if (!error) {
688 		    if ((clp->lc_flags & (LCL_NEEDSCONFIRM | LCL_DONTCLEAN)) ==
689 			LCL_NEEDSCONFIRM) {
690 			/*
691 			 * Hang onto the delegations (as old delegations)
692 			 * for an Open with CLAIM_DELEGATE_PREV unless in
693 			 * grace, but get rid of the rest of the state.
694 			 */
695 			nfsrv_cleanclient(clp, p);
696 			nfsrv_freedeleglist(&clp->lc_olddeleg);
697 			if (nfsrv_checkgrace(nd, clp, 0)) {
698 			    /* In grace, so just delete delegations */
699 			    nfsrv_freedeleglist(&clp->lc_deleg);
700 			} else {
701 			    LIST_FOREACH(stp, &clp->lc_deleg, ls_list)
702 				stp->ls_flags |= NFSLCK_OLDDELEG;
703 			    clp->lc_delegtime = NFSD_MONOSEC +
704 				nfsrv_lease + NFSRV_LEASEDELTA;
705 			    LIST_NEWHEAD(&clp->lc_olddeleg, &clp->lc_deleg,
706 				ls_list);
707 			}
708 			if ((nd->nd_flag & ND_NFSV41) != 0)
709 			    clp->lc_program = cbprogram;
710 		    }
711 		    clp->lc_flags &= ~(LCL_NEEDSCONFIRM | LCL_DONTCLEAN);
712 		    if (clp->lc_program)
713 			clp->lc_flags |= LCL_NEEDSCBNULL;
714 		    /* For NFSv4.1, link the session onto the client. */
715 		    if (nsep != NULL) {
716 			/* Hold a reference on the xprt for a backchannel. */
717 			if ((nsep->sess_crflags & NFSV4CRSESS_CONNBACKCHAN)
718 			    != 0) {
719 			    if (clp->lc_req.nr_client == NULL)
720 				clp->lc_req.nr_client = (struct __rpc_client *)
721 				    clnt_bck_create(nd->nd_xprt->xp_socket,
722 				    cbprogram, NFSV4_CBVERS);
723 			    if (clp->lc_req.nr_client != NULL) {
724 				SVC_ACQUIRE(nd->nd_xprt);
725 				CLNT_ACQUIRE(clp->lc_req.nr_client);
726 				nd->nd_xprt->xp_p2 = clp->lc_req.nr_client;
727 				/* Disable idle timeout. */
728 				nd->nd_xprt->xp_idletimeout = 0;
729 				nsep->sess_cbsess.nfsess_xprt = nd->nd_xprt;
730 			    } else
731 				nsep->sess_crflags &= ~NFSV4CRSESS_CONNBACKCHAN;
732 			}
733 			NFSBCOPY(sessid, nsep->sess_sessionid,
734 			    NFSX_V4SESSIONID);
735 			NFSBCOPY(sessid, nsep->sess_cbsess.nfsess_sessionid,
736 			    NFSX_V4SESSIONID);
737 			shp = NFSSESSIONHASH(nsep->sess_sessionid);
738 			NFSLOCKSTATE();
739 			NFSLOCKSESSION(shp);
740 			LIST_INSERT_HEAD(&shp->list, nsep, sess_hash);
741 			LIST_INSERT_HEAD(&clp->lc_session, nsep, sess_list);
742 			nsep->sess_clp = clp;
743 			NFSUNLOCKSESSION(shp);
744 			NFSUNLOCKSTATE();
745 		    }
746 		}
747 	} else if (clp->lc_flags & LCL_NEEDSCONFIRM) {
748 		error = NFSERR_EXPIRED;
749 	}
750 
751 	/*
752 	 * If called by the Renew Op, we must check the principal.
753 	 */
754 	if (!error && (opflags & CLOPS_RENEWOP)) {
755 	    if (nfsrv_notsamecredname(nd, clp)) {
756 		doneok = 0;
757 		for (i = 0; i < nfsrv_statehashsize && doneok == 0; i++) {
758 		    LIST_FOREACH(stp, &clp->lc_stateid[i], ls_hash) {
759 			if ((stp->ls_flags & NFSLCK_OPEN) &&
760 			    stp->ls_uid == nd->nd_cred->cr_uid) {
761 				doneok = 1;
762 				break;
763 			}
764 		    }
765 		}
766 		if (!doneok)
767 			error = NFSERR_ACCES;
768 	    }
769 	    if (!error && (clp->lc_flags & LCL_CBDOWN))
770 		error = NFSERR_CBPATHDOWN;
771 	}
772 	if ((!error || error == NFSERR_CBPATHDOWN) &&
773 	     (opflags & CLOPS_RENEW)) {
774 		clp->lc_expiry = nfsrv_leaseexpiry();
775 	}
776 	if (opflags & CLOPS_CONFIRM) {
777 		NFSLOCKV4ROOTMUTEX();
778 		nfsv4_unlock(&nfsv4rootfs_lock, 1);
779 		NFSUNLOCKV4ROOTMUTEX();
780 	} else if (opflags != CLOPS_RENEW) {
781 		NFSUNLOCKSTATE();
782 	}
783 	if (clpp)
784 		*clpp = clp;
785 
786 out:
787 	NFSEXITCODE2(error, nd);
788 	return (error);
789 }
790 
791 /*
792  * Perform the NFSv4.1 destroy clientid.
793  */
794 int
795 nfsrv_destroyclient(nfsquad_t clientid, NFSPROC_T *p)
796 {
797 	struct nfsclient *clp;
798 	struct nfsclienthashhead *hp;
799 	int error = 0, i, igotlock;
800 
801 	if (nfsrvboottime != clientid.lval[0]) {
802 		error = NFSERR_STALECLIENTID;
803 		goto out;
804 	}
805 
806 	/* Lock out other nfsd threads */
807 	NFSLOCKV4ROOTMUTEX();
808 	nfsv4_relref(&nfsv4rootfs_lock);
809 	do {
810 		igotlock = nfsv4_lock(&nfsv4rootfs_lock, 1, NULL,
811 		    NFSV4ROOTLOCKMUTEXPTR, NULL);
812 	} while (igotlock == 0);
813 	NFSUNLOCKV4ROOTMUTEX();
814 
815 	hp = NFSCLIENTHASH(clientid);
816 	LIST_FOREACH(clp, hp, lc_hash) {
817 		if (clp->lc_clientid.lval[1] == clientid.lval[1])
818 			break;
819 	}
820 	if (clp == NULL) {
821 		NFSLOCKV4ROOTMUTEX();
822 		nfsv4_unlock(&nfsv4rootfs_lock, 1);
823 		NFSUNLOCKV4ROOTMUTEX();
824 		/* Just return ok, since it is gone. */
825 		goto out;
826 	}
827 
828 	/*
829 	 * Free up all layouts on the clientid.  Should the client return the
830 	 * layouts?
831 	 */
832 	nfsrv_freelayoutlist(clientid);
833 
834 	/* Scan for state on the clientid. */
835 	for (i = 0; i < nfsrv_statehashsize; i++)
836 		if (!LIST_EMPTY(&clp->lc_stateid[i])) {
837 			NFSLOCKV4ROOTMUTEX();
838 			nfsv4_unlock(&nfsv4rootfs_lock, 1);
839 			NFSUNLOCKV4ROOTMUTEX();
840 			error = NFSERR_CLIENTIDBUSY;
841 			goto out;
842 		}
843 	if (!LIST_EMPTY(&clp->lc_session) || !LIST_EMPTY(&clp->lc_deleg)) {
844 		NFSLOCKV4ROOTMUTEX();
845 		nfsv4_unlock(&nfsv4rootfs_lock, 1);
846 		NFSUNLOCKV4ROOTMUTEX();
847 		error = NFSERR_CLIENTIDBUSY;
848 		goto out;
849 	}
850 
851 	/* Destroy the clientid and return ok. */
852 	nfsrv_cleanclient(clp, p);
853 	nfsrv_freedeleglist(&clp->lc_deleg);
854 	nfsrv_freedeleglist(&clp->lc_olddeleg);
855 	LIST_REMOVE(clp, lc_hash);
856 	NFSLOCKV4ROOTMUTEX();
857 	nfsv4_unlock(&nfsv4rootfs_lock, 1);
858 	NFSUNLOCKV4ROOTMUTEX();
859 	nfsrv_zapclient(clp, p);
860 out:
861 	NFSEXITCODE2(error, nd);
862 	return (error);
863 }
864 
865 /*
866  * Called from the new nfssvc syscall to admin revoke a clientid.
867  * Returns 0 for success, error otherwise.
868  */
869 int
870 nfsrv_adminrevoke(struct nfsd_clid *revokep, NFSPROC_T *p)
871 {
872 	struct nfsclient *clp = NULL;
873 	int i, error = 0;
874 	int gotit, igotlock;
875 
876 	/*
877 	 * First, lock out the nfsd so that state won't change while the
878 	 * revocation record is being written to the stable storage restart
879 	 * file.
880 	 */
881 	NFSLOCKV4ROOTMUTEX();
882 	do {
883 		igotlock = nfsv4_lock(&nfsv4rootfs_lock, 1, NULL,
884 		    NFSV4ROOTLOCKMUTEXPTR, NULL);
885 	} while (!igotlock);
886 	NFSUNLOCKV4ROOTMUTEX();
887 
888 	/*
889 	 * Search for a match in the client list.
890 	 */
891 	gotit = i = 0;
892 	while (i < nfsrv_clienthashsize && !gotit) {
893 	    LIST_FOREACH(clp, &nfsclienthash[i], lc_hash) {
894 		if (revokep->nclid_idlen == clp->lc_idlen &&
895 		    !NFSBCMP(revokep->nclid_id, clp->lc_id, clp->lc_idlen)) {
896 			gotit = 1;
897 			break;
898 		}
899 	    }
900 	    i++;
901 	}
902 	if (!gotit) {
903 		NFSLOCKV4ROOTMUTEX();
904 		nfsv4_unlock(&nfsv4rootfs_lock, 0);
905 		NFSUNLOCKV4ROOTMUTEX();
906 		error = EPERM;
907 		goto out;
908 	}
909 
910 	/*
911 	 * Now, write out the revocation record
912 	 */
913 	nfsrv_writestable(clp->lc_id, clp->lc_idlen, NFSNST_REVOKE, p);
914 	nfsrv_backupstable();
915 
916 	/*
917 	 * and clear out the state, marking the clientid revoked.
918 	 */
919 	clp->lc_flags &= ~LCL_CALLBACKSON;
920 	clp->lc_flags |= LCL_ADMINREVOKED;
921 	nfsrv_cleanclient(clp, p);
922 	nfsrv_freedeleglist(&clp->lc_deleg);
923 	nfsrv_freedeleglist(&clp->lc_olddeleg);
924 	NFSLOCKV4ROOTMUTEX();
925 	nfsv4_unlock(&nfsv4rootfs_lock, 0);
926 	NFSUNLOCKV4ROOTMUTEX();
927 
928 out:
929 	NFSEXITCODE(error);
930 	return (error);
931 }
932 
933 /*
934  * Dump out stats for all clients. Called from nfssvc(2), that is used
935  * nfsstatsv1.
936  */
937 void
938 nfsrv_dumpclients(struct nfsd_dumpclients *dumpp, int maxcnt)
939 {
940 	struct nfsclient *clp;
941 	int i = 0, cnt = 0;
942 
943 	/*
944 	 * First, get a reference on the nfsv4rootfs_lock so that an
945 	 * exclusive lock cannot be acquired while dumping the clients.
946 	 */
947 	NFSLOCKV4ROOTMUTEX();
948 	nfsv4_getref(&nfsv4rootfs_lock, NULL, NFSV4ROOTLOCKMUTEXPTR, NULL);
949 	NFSUNLOCKV4ROOTMUTEX();
950 	NFSLOCKSTATE();
951 	/*
952 	 * Rattle through the client lists until done.
953 	 */
954 	while (i < nfsrv_clienthashsize && cnt < maxcnt) {
955 	    clp = LIST_FIRST(&nfsclienthash[i]);
956 	    while (clp != LIST_END(&nfsclienthash[i]) && cnt < maxcnt) {
957 		nfsrv_dumpaclient(clp, &dumpp[cnt]);
958 		cnt++;
959 		clp = LIST_NEXT(clp, lc_hash);
960 	    }
961 	    i++;
962 	}
963 	if (cnt < maxcnt)
964 	    dumpp[cnt].ndcl_clid.nclid_idlen = 0;
965 	NFSUNLOCKSTATE();
966 	NFSLOCKV4ROOTMUTEX();
967 	nfsv4_relref(&nfsv4rootfs_lock);
968 	NFSUNLOCKV4ROOTMUTEX();
969 }
970 
971 /*
972  * Dump stats for a client. Must be called with the NFSSTATELOCK and spl'd.
973  */
974 static void
975 nfsrv_dumpaclient(struct nfsclient *clp, struct nfsd_dumpclients *dumpp)
976 {
977 	struct nfsstate *stp, *openstp, *lckownstp;
978 	struct nfslock *lop;
979 	sa_family_t af;
980 #ifdef INET
981 	struct sockaddr_in *rin;
982 #endif
983 #ifdef INET6
984 	struct sockaddr_in6 *rin6;
985 #endif
986 
987 	dumpp->ndcl_nopenowners = dumpp->ndcl_nlockowners = 0;
988 	dumpp->ndcl_nopens = dumpp->ndcl_nlocks = 0;
989 	dumpp->ndcl_ndelegs = dumpp->ndcl_nolddelegs = 0;
990 	dumpp->ndcl_flags = clp->lc_flags;
991 	dumpp->ndcl_clid.nclid_idlen = clp->lc_idlen;
992 	NFSBCOPY(clp->lc_id, dumpp->ndcl_clid.nclid_id, clp->lc_idlen);
993 	af = clp->lc_req.nr_nam->sa_family;
994 	dumpp->ndcl_addrfam = af;
995 	switch (af) {
996 #ifdef INET
997 	case AF_INET:
998 		rin = (struct sockaddr_in *)clp->lc_req.nr_nam;
999 		dumpp->ndcl_cbaddr.sin_addr = rin->sin_addr;
1000 		break;
1001 #endif
1002 #ifdef INET6
1003 	case AF_INET6:
1004 		rin6 = (struct sockaddr_in6 *)clp->lc_req.nr_nam;
1005 		dumpp->ndcl_cbaddr.sin6_addr = rin6->sin6_addr;
1006 		break;
1007 #endif
1008 	}
1009 
1010 	/*
1011 	 * Now, scan the state lists and total up the opens and locks.
1012 	 */
1013 	LIST_FOREACH(stp, &clp->lc_open, ls_list) {
1014 	    dumpp->ndcl_nopenowners++;
1015 	    LIST_FOREACH(openstp, &stp->ls_open, ls_list) {
1016 		dumpp->ndcl_nopens++;
1017 		LIST_FOREACH(lckownstp, &openstp->ls_open, ls_list) {
1018 		    dumpp->ndcl_nlockowners++;
1019 		    LIST_FOREACH(lop, &lckownstp->ls_lock, lo_lckowner) {
1020 			dumpp->ndcl_nlocks++;
1021 		    }
1022 		}
1023 	    }
1024 	}
1025 
1026 	/*
1027 	 * and the delegation lists.
1028 	 */
1029 	LIST_FOREACH(stp, &clp->lc_deleg, ls_list) {
1030 	    dumpp->ndcl_ndelegs++;
1031 	}
1032 	LIST_FOREACH(stp, &clp->lc_olddeleg, ls_list) {
1033 	    dumpp->ndcl_nolddelegs++;
1034 	}
1035 }
1036 
1037 /*
1038  * Dump out lock stats for a file.
1039  */
1040 void
1041 nfsrv_dumplocks(vnode_t vp, struct nfsd_dumplocks *ldumpp, int maxcnt,
1042     NFSPROC_T *p)
1043 {
1044 	struct nfsstate *stp;
1045 	struct nfslock *lop;
1046 	int cnt = 0;
1047 	struct nfslockfile *lfp;
1048 	sa_family_t af;
1049 #ifdef INET
1050 	struct sockaddr_in *rin;
1051 #endif
1052 #ifdef INET6
1053 	struct sockaddr_in6 *rin6;
1054 #endif
1055 	int ret;
1056 	fhandle_t nfh;
1057 
1058 	ret = nfsrv_getlockfh(vp, 0, NULL, &nfh, p);
1059 	/*
1060 	 * First, get a reference on the nfsv4rootfs_lock so that an
1061 	 * exclusive lock on it cannot be acquired while dumping the locks.
1062 	 */
1063 	NFSLOCKV4ROOTMUTEX();
1064 	nfsv4_getref(&nfsv4rootfs_lock, NULL, NFSV4ROOTLOCKMUTEXPTR, NULL);
1065 	NFSUNLOCKV4ROOTMUTEX();
1066 	NFSLOCKSTATE();
1067 	if (!ret)
1068 		ret = nfsrv_getlockfile(0, NULL, &lfp, &nfh, 0);
1069 	if (ret) {
1070 		ldumpp[0].ndlck_clid.nclid_idlen = 0;
1071 		NFSUNLOCKSTATE();
1072 		NFSLOCKV4ROOTMUTEX();
1073 		nfsv4_relref(&nfsv4rootfs_lock);
1074 		NFSUNLOCKV4ROOTMUTEX();
1075 		return;
1076 	}
1077 
1078 	/*
1079 	 * For each open share on file, dump it out.
1080 	 */
1081 	stp = LIST_FIRST(&lfp->lf_open);
1082 	while (stp != LIST_END(&lfp->lf_open) && cnt < maxcnt) {
1083 		ldumpp[cnt].ndlck_flags = stp->ls_flags;
1084 		ldumpp[cnt].ndlck_stateid.seqid = stp->ls_stateid.seqid;
1085 		ldumpp[cnt].ndlck_stateid.other[0] = stp->ls_stateid.other[0];
1086 		ldumpp[cnt].ndlck_stateid.other[1] = stp->ls_stateid.other[1];
1087 		ldumpp[cnt].ndlck_stateid.other[2] = stp->ls_stateid.other[2];
1088 		ldumpp[cnt].ndlck_owner.nclid_idlen =
1089 		    stp->ls_openowner->ls_ownerlen;
1090 		NFSBCOPY(stp->ls_openowner->ls_owner,
1091 		    ldumpp[cnt].ndlck_owner.nclid_id,
1092 		    stp->ls_openowner->ls_ownerlen);
1093 		ldumpp[cnt].ndlck_clid.nclid_idlen = stp->ls_clp->lc_idlen;
1094 		NFSBCOPY(stp->ls_clp->lc_id, ldumpp[cnt].ndlck_clid.nclid_id,
1095 		    stp->ls_clp->lc_idlen);
1096 		af = stp->ls_clp->lc_req.nr_nam->sa_family;
1097 		ldumpp[cnt].ndlck_addrfam = af;
1098 		switch (af) {
1099 #ifdef INET
1100 		case AF_INET:
1101 			rin = (struct sockaddr_in *)stp->ls_clp->lc_req.nr_nam;
1102 			ldumpp[cnt].ndlck_cbaddr.sin_addr = rin->sin_addr;
1103 			break;
1104 #endif
1105 #ifdef INET6
1106 		case AF_INET6:
1107 			rin6 = (struct sockaddr_in6 *)
1108 			    stp->ls_clp->lc_req.nr_nam;
1109 			ldumpp[cnt].ndlck_cbaddr.sin6_addr = rin6->sin6_addr;
1110 			break;
1111 #endif
1112 		}
1113 		stp = LIST_NEXT(stp, ls_file);
1114 		cnt++;
1115 	}
1116 
1117 	/*
1118 	 * and all locks.
1119 	 */
1120 	lop = LIST_FIRST(&lfp->lf_lock);
1121 	while (lop != LIST_END(&lfp->lf_lock) && cnt < maxcnt) {
1122 		stp = lop->lo_stp;
1123 		ldumpp[cnt].ndlck_flags = lop->lo_flags;
1124 		ldumpp[cnt].ndlck_first = lop->lo_first;
1125 		ldumpp[cnt].ndlck_end = lop->lo_end;
1126 		ldumpp[cnt].ndlck_stateid.seqid = stp->ls_stateid.seqid;
1127 		ldumpp[cnt].ndlck_stateid.other[0] = stp->ls_stateid.other[0];
1128 		ldumpp[cnt].ndlck_stateid.other[1] = stp->ls_stateid.other[1];
1129 		ldumpp[cnt].ndlck_stateid.other[2] = stp->ls_stateid.other[2];
1130 		ldumpp[cnt].ndlck_owner.nclid_idlen = stp->ls_ownerlen;
1131 		NFSBCOPY(stp->ls_owner, ldumpp[cnt].ndlck_owner.nclid_id,
1132 		    stp->ls_ownerlen);
1133 		ldumpp[cnt].ndlck_clid.nclid_idlen = stp->ls_clp->lc_idlen;
1134 		NFSBCOPY(stp->ls_clp->lc_id, ldumpp[cnt].ndlck_clid.nclid_id,
1135 		    stp->ls_clp->lc_idlen);
1136 		af = stp->ls_clp->lc_req.nr_nam->sa_family;
1137 		ldumpp[cnt].ndlck_addrfam = af;
1138 		switch (af) {
1139 #ifdef INET
1140 		case AF_INET:
1141 			rin = (struct sockaddr_in *)stp->ls_clp->lc_req.nr_nam;
1142 			ldumpp[cnt].ndlck_cbaddr.sin_addr = rin->sin_addr;
1143 			break;
1144 #endif
1145 #ifdef INET6
1146 		case AF_INET6:
1147 			rin6 = (struct sockaddr_in6 *)
1148 			    stp->ls_clp->lc_req.nr_nam;
1149 			ldumpp[cnt].ndlck_cbaddr.sin6_addr = rin6->sin6_addr;
1150 			break;
1151 #endif
1152 		}
1153 		lop = LIST_NEXT(lop, lo_lckfile);
1154 		cnt++;
1155 	}
1156 
1157 	/*
1158 	 * and the delegations.
1159 	 */
1160 	stp = LIST_FIRST(&lfp->lf_deleg);
1161 	while (stp != LIST_END(&lfp->lf_deleg) && cnt < maxcnt) {
1162 		ldumpp[cnt].ndlck_flags = stp->ls_flags;
1163 		ldumpp[cnt].ndlck_stateid.seqid = stp->ls_stateid.seqid;
1164 		ldumpp[cnt].ndlck_stateid.other[0] = stp->ls_stateid.other[0];
1165 		ldumpp[cnt].ndlck_stateid.other[1] = stp->ls_stateid.other[1];
1166 		ldumpp[cnt].ndlck_stateid.other[2] = stp->ls_stateid.other[2];
1167 		ldumpp[cnt].ndlck_owner.nclid_idlen = 0;
1168 		ldumpp[cnt].ndlck_clid.nclid_idlen = stp->ls_clp->lc_idlen;
1169 		NFSBCOPY(stp->ls_clp->lc_id, ldumpp[cnt].ndlck_clid.nclid_id,
1170 		    stp->ls_clp->lc_idlen);
1171 		af = stp->ls_clp->lc_req.nr_nam->sa_family;
1172 		ldumpp[cnt].ndlck_addrfam = af;
1173 		switch (af) {
1174 #ifdef INET
1175 		case AF_INET:
1176 			rin = (struct sockaddr_in *)stp->ls_clp->lc_req.nr_nam;
1177 			ldumpp[cnt].ndlck_cbaddr.sin_addr = rin->sin_addr;
1178 			break;
1179 #endif
1180 #ifdef INET6
1181 		case AF_INET6:
1182 			rin6 = (struct sockaddr_in6 *)
1183 			    stp->ls_clp->lc_req.nr_nam;
1184 			ldumpp[cnt].ndlck_cbaddr.sin6_addr = rin6->sin6_addr;
1185 			break;
1186 #endif
1187 		}
1188 		stp = LIST_NEXT(stp, ls_file);
1189 		cnt++;
1190 	}
1191 
1192 	/*
1193 	 * If list isn't full, mark end of list by setting the client name
1194 	 * to zero length.
1195 	 */
1196 	if (cnt < maxcnt)
1197 		ldumpp[cnt].ndlck_clid.nclid_idlen = 0;
1198 	NFSUNLOCKSTATE();
1199 	NFSLOCKV4ROOTMUTEX();
1200 	nfsv4_relref(&nfsv4rootfs_lock);
1201 	NFSUNLOCKV4ROOTMUTEX();
1202 }
1203 
1204 /*
1205  * Server timer routine. It can scan any linked list, so long
1206  * as it holds the spin/mutex lock and there is no exclusive lock on
1207  * nfsv4rootfs_lock.
1208  * (For OpenBSD, a kthread is ok. For FreeBSD, I think it is ok
1209  *  to do this from a callout, since the spin locks work. For
1210  *  Darwin, I'm not sure what will work correctly yet.)
1211  * Should be called once per second.
1212  */
1213 void
1214 nfsrv_servertimer(void)
1215 {
1216 	struct nfsclient *clp, *nclp;
1217 	struct nfsstate *stp, *nstp;
1218 	int got_ref, i;
1219 
1220 	/*
1221 	 * Make sure nfsboottime is set. This is used by V3 as well
1222 	 * as V4. Note that nfsboottime is not nfsrvboottime, which is
1223 	 * only used by the V4 server for leases.
1224 	 */
1225 	if (nfsboottime.tv_sec == 0)
1226 		NFSSETBOOTTIME(nfsboottime);
1227 
1228 	/*
1229 	 * If server hasn't started yet, just return.
1230 	 */
1231 	NFSLOCKSTATE();
1232 	if (nfsrv_stablefirst.nsf_eograce == 0) {
1233 		NFSUNLOCKSTATE();
1234 		return;
1235 	}
1236 	if (!(nfsrv_stablefirst.nsf_flags & NFSNSF_UPDATEDONE)) {
1237 		if (!(nfsrv_stablefirst.nsf_flags & NFSNSF_GRACEOVER) &&
1238 		    NFSD_MONOSEC > nfsrv_stablefirst.nsf_eograce)
1239 			nfsrv_stablefirst.nsf_flags |=
1240 			    (NFSNSF_GRACEOVER | NFSNSF_NEEDLOCK);
1241 		NFSUNLOCKSTATE();
1242 		return;
1243 	}
1244 
1245 	/*
1246 	 * Try and get a reference count on the nfsv4rootfs_lock so that
1247 	 * no nfsd thread can acquire an exclusive lock on it before this
1248 	 * call is done. If it is already exclusively locked, just return.
1249 	 */
1250 	NFSLOCKV4ROOTMUTEX();
1251 	got_ref = nfsv4_getref_nonblock(&nfsv4rootfs_lock);
1252 	NFSUNLOCKV4ROOTMUTEX();
1253 	if (got_ref == 0) {
1254 		NFSUNLOCKSTATE();
1255 		return;
1256 	}
1257 
1258 	/*
1259 	 * For each client...
1260 	 */
1261 	for (i = 0; i < nfsrv_clienthashsize; i++) {
1262 	    clp = LIST_FIRST(&nfsclienthash[i]);
1263 	    while (clp != LIST_END(&nfsclienthash[i])) {
1264 		nclp = LIST_NEXT(clp, lc_hash);
1265 		if (!(clp->lc_flags & LCL_EXPIREIT)) {
1266 		    if (((clp->lc_expiry + NFSRV_STALELEASE) < NFSD_MONOSEC
1267 			 && ((LIST_EMPTY(&clp->lc_deleg)
1268 			      && LIST_EMPTY(&clp->lc_open)) ||
1269 			     nfsrv_clients > nfsrv_clienthighwater)) ||
1270 			(clp->lc_expiry + NFSRV_MOULDYLEASE) < NFSD_MONOSEC ||
1271 			(clp->lc_expiry < NFSD_MONOSEC &&
1272 			 (nfsrv_openpluslock * 10 / 9) > nfsrv_v4statelimit)) {
1273 			/*
1274 			 * Lease has expired several nfsrv_lease times ago:
1275 			 * PLUS
1276 			 *    - no state is associated with it
1277 			 *    OR
1278 			 *    - above high water mark for number of clients
1279 			 *      (nfsrv_clienthighwater should be large enough
1280 			 *       that this only occurs when clients fail to
1281 			 *       use the same nfs_client_id4.id. Maybe somewhat
1282 			 *       higher that the maximum number of clients that
1283 			 *       will mount this server?)
1284 			 * OR
1285 			 * Lease has expired a very long time ago
1286 			 * OR
1287 			 * Lease has expired PLUS the number of opens + locks
1288 			 * has exceeded 90% of capacity
1289 			 *
1290 			 * --> Mark for expiry. The actual expiry will be done
1291 			 *     by an nfsd sometime soon.
1292 			 */
1293 			clp->lc_flags |= LCL_EXPIREIT;
1294 			nfsrv_stablefirst.nsf_flags |=
1295 			    (NFSNSF_NEEDLOCK | NFSNSF_EXPIREDCLIENT);
1296 		    } else {
1297 			/*
1298 			 * If there are no opens, increment no open tick cnt
1299 			 * If time exceeds NFSNOOPEN, mark it to be thrown away
1300 			 * otherwise, if there is an open, reset no open time
1301 			 * Hopefully, this will avoid excessive re-creation
1302 			 * of open owners and subsequent open confirms.
1303 			 */
1304 			stp = LIST_FIRST(&clp->lc_open);
1305 			while (stp != LIST_END(&clp->lc_open)) {
1306 				nstp = LIST_NEXT(stp, ls_list);
1307 				if (LIST_EMPTY(&stp->ls_open)) {
1308 					stp->ls_noopens++;
1309 					if (stp->ls_noopens > NFSNOOPEN ||
1310 					    (nfsrv_openpluslock * 2) >
1311 					    nfsrv_v4statelimit)
1312 						nfsrv_stablefirst.nsf_flags |=
1313 							NFSNSF_NOOPENS;
1314 				} else {
1315 					stp->ls_noopens = 0;
1316 				}
1317 				stp = nstp;
1318 			}
1319 		    }
1320 		}
1321 		clp = nclp;
1322 	    }
1323 	}
1324 	NFSUNLOCKSTATE();
1325 	NFSLOCKV4ROOTMUTEX();
1326 	nfsv4_relref(&nfsv4rootfs_lock);
1327 	NFSUNLOCKV4ROOTMUTEX();
1328 }
1329 
1330 /*
1331  * The following set of functions free up the various data structures.
1332  */
1333 /*
1334  * Clear out all open/lock state related to this nfsclient.
1335  * Caller must hold an exclusive lock on nfsv4rootfs_lock, so that
1336  * there are no other active nfsd threads.
1337  */
1338 void
1339 nfsrv_cleanclient(struct nfsclient *clp, NFSPROC_T *p)
1340 {
1341 	struct nfsstate *stp, *nstp;
1342 	struct nfsdsession *sep, *nsep;
1343 
1344 	LIST_FOREACH_SAFE(stp, &clp->lc_open, ls_list, nstp)
1345 		nfsrv_freeopenowner(stp, 1, p);
1346 	if ((clp->lc_flags & LCL_ADMINREVOKED) == 0)
1347 		LIST_FOREACH_SAFE(sep, &clp->lc_session, sess_list, nsep)
1348 			(void)nfsrv_freesession(sep, NULL);
1349 }
1350 
1351 /*
1352  * Free a client that has been cleaned. It should also already have been
1353  * removed from the lists.
1354  * (Just to be safe w.r.t. newnfs_disconnect(), call this function when
1355  *  softclock interrupts are enabled.)
1356  */
1357 void
1358 nfsrv_zapclient(struct nfsclient *clp, NFSPROC_T *p)
1359 {
1360 
1361 #ifdef notyet
1362 	if ((clp->lc_flags & (LCL_GSS | LCL_CALLBACKSON)) ==
1363 	     (LCL_GSS | LCL_CALLBACKSON) &&
1364 	    (clp->lc_hand.nfsh_flag & NFSG_COMPLETE) &&
1365 	    clp->lc_handlelen > 0) {
1366 		clp->lc_hand.nfsh_flag &= ~NFSG_COMPLETE;
1367 		clp->lc_hand.nfsh_flag |= NFSG_DESTROYED;
1368 		(void) nfsrv_docallback(clp, NFSV4PROC_CBNULL,
1369 			NULL, 0, NULL, NULL, NULL, 0, p);
1370 	}
1371 #endif
1372 	newnfs_disconnect(NULL, &clp->lc_req);
1373 	free(clp->lc_req.nr_nam, M_SONAME);
1374 	NFSFREEMUTEX(&clp->lc_req.nr_mtx);
1375 	free(clp->lc_stateid, M_NFSDCLIENT);
1376 	free(clp, M_NFSDCLIENT);
1377 	NFSLOCKSTATE();
1378 	nfsstatsv1.srvclients--;
1379 	nfsrv_openpluslock--;
1380 	nfsrv_clients--;
1381 	NFSUNLOCKSTATE();
1382 }
1383 
1384 /*
1385  * Free a list of delegation state structures.
1386  * (This function will also free all nfslockfile structures that no
1387  *  longer have associated state.)
1388  */
1389 void
1390 nfsrv_freedeleglist(struct nfsstatehead *sthp)
1391 {
1392 	struct nfsstate *stp, *nstp;
1393 
1394 	LIST_FOREACH_SAFE(stp, sthp, ls_list, nstp) {
1395 		nfsrv_freedeleg(stp);
1396 	}
1397 	LIST_INIT(sthp);
1398 }
1399 
1400 /*
1401  * Free up a delegation.
1402  */
1403 static void
1404 nfsrv_freedeleg(struct nfsstate *stp)
1405 {
1406 	struct nfslockfile *lfp;
1407 
1408 	LIST_REMOVE(stp, ls_hash);
1409 	LIST_REMOVE(stp, ls_list);
1410 	LIST_REMOVE(stp, ls_file);
1411 	if ((stp->ls_flags & NFSLCK_DELEGWRITE) != 0)
1412 		nfsrv_writedelegcnt--;
1413 	lfp = stp->ls_lfp;
1414 	if (LIST_EMPTY(&lfp->lf_open) &&
1415 	    LIST_EMPTY(&lfp->lf_lock) && LIST_EMPTY(&lfp->lf_deleg) &&
1416 	    LIST_EMPTY(&lfp->lf_locallock) && LIST_EMPTY(&lfp->lf_rollback) &&
1417 	    lfp->lf_usecount == 0 &&
1418 	    nfsv4_testlock(&lfp->lf_locallock_lck) == 0)
1419 		nfsrv_freenfslockfile(lfp);
1420 	free(stp, M_NFSDSTATE);
1421 	nfsstatsv1.srvdelegates--;
1422 	nfsrv_openpluslock--;
1423 	nfsrv_delegatecnt--;
1424 }
1425 
1426 /*
1427  * This function frees an open owner and all associated opens.
1428  */
1429 static void
1430 nfsrv_freeopenowner(struct nfsstate *stp, int cansleep, NFSPROC_T *p)
1431 {
1432 	struct nfsstate *nstp, *tstp;
1433 
1434 	LIST_REMOVE(stp, ls_list);
1435 	/*
1436 	 * Now, free all associated opens.
1437 	 */
1438 	nstp = LIST_FIRST(&stp->ls_open);
1439 	while (nstp != LIST_END(&stp->ls_open)) {
1440 		tstp = nstp;
1441 		nstp = LIST_NEXT(nstp, ls_list);
1442 		(void) nfsrv_freeopen(tstp, NULL, cansleep, p);
1443 	}
1444 	if (stp->ls_op)
1445 		nfsrvd_derefcache(stp->ls_op);
1446 	free(stp, M_NFSDSTATE);
1447 	nfsstatsv1.srvopenowners--;
1448 	nfsrv_openpluslock--;
1449 }
1450 
1451 /*
1452  * This function frees an open (nfsstate open structure) with all associated
1453  * lock_owners and locks. It also frees the nfslockfile structure iff there
1454  * are no other opens on the file.
1455  * Returns 1 if it free'd the nfslockfile, 0 otherwise.
1456  */
1457 static int
1458 nfsrv_freeopen(struct nfsstate *stp, vnode_t vp, int cansleep, NFSPROC_T *p)
1459 {
1460 	struct nfsstate *nstp, *tstp;
1461 	struct nfslockfile *lfp;
1462 	int ret;
1463 
1464 	LIST_REMOVE(stp, ls_hash);
1465 	LIST_REMOVE(stp, ls_list);
1466 	LIST_REMOVE(stp, ls_file);
1467 
1468 	lfp = stp->ls_lfp;
1469 	/*
1470 	 * Now, free all lockowners associated with this open.
1471 	 */
1472 	LIST_FOREACH_SAFE(tstp, &stp->ls_open, ls_list, nstp)
1473 		nfsrv_freelockowner(tstp, vp, cansleep, p);
1474 
1475 	/*
1476 	 * The nfslockfile is freed here if there are no locks
1477 	 * associated with the open.
1478 	 * If there are locks associated with the open, the
1479 	 * nfslockfile structure can be freed via nfsrv_freelockowner().
1480 	 * Acquire the state mutex to avoid races with calls to
1481 	 * nfsrv_getlockfile().
1482 	 */
1483 	if (cansleep != 0)
1484 		NFSLOCKSTATE();
1485 	if (lfp != NULL && LIST_EMPTY(&lfp->lf_open) &&
1486 	    LIST_EMPTY(&lfp->lf_deleg) && LIST_EMPTY(&lfp->lf_lock) &&
1487 	    LIST_EMPTY(&lfp->lf_locallock) && LIST_EMPTY(&lfp->lf_rollback) &&
1488 	    lfp->lf_usecount == 0 &&
1489 	    (cansleep != 0 || nfsv4_testlock(&lfp->lf_locallock_lck) == 0)) {
1490 		nfsrv_freenfslockfile(lfp);
1491 		ret = 1;
1492 	} else
1493 		ret = 0;
1494 	if (cansleep != 0)
1495 		NFSUNLOCKSTATE();
1496 	free(stp, M_NFSDSTATE);
1497 	nfsstatsv1.srvopens--;
1498 	nfsrv_openpluslock--;
1499 	return (ret);
1500 }
1501 
1502 /*
1503  * Frees a lockowner and all associated locks.
1504  */
1505 static void
1506 nfsrv_freelockowner(struct nfsstate *stp, vnode_t vp, int cansleep,
1507     NFSPROC_T *p)
1508 {
1509 
1510 	LIST_REMOVE(stp, ls_hash);
1511 	LIST_REMOVE(stp, ls_list);
1512 	nfsrv_freeallnfslocks(stp, vp, cansleep, p);
1513 	if (stp->ls_op)
1514 		nfsrvd_derefcache(stp->ls_op);
1515 	free(stp, M_NFSDSTATE);
1516 	nfsstatsv1.srvlockowners--;
1517 	nfsrv_openpluslock--;
1518 }
1519 
1520 /*
1521  * Free all the nfs locks on a lockowner.
1522  */
1523 static void
1524 nfsrv_freeallnfslocks(struct nfsstate *stp, vnode_t vp, int cansleep,
1525     NFSPROC_T *p)
1526 {
1527 	struct nfslock *lop, *nlop;
1528 	struct nfsrollback *rlp, *nrlp;
1529 	struct nfslockfile *lfp = NULL;
1530 	int gottvp = 0;
1531 	vnode_t tvp = NULL;
1532 	uint64_t first, end;
1533 
1534 	if (vp != NULL)
1535 		ASSERT_VOP_UNLOCKED(vp, "nfsrv_freeallnfslocks: vnode locked");
1536 	lop = LIST_FIRST(&stp->ls_lock);
1537 	while (lop != LIST_END(&stp->ls_lock)) {
1538 		nlop = LIST_NEXT(lop, lo_lckowner);
1539 		/*
1540 		 * Since all locks should be for the same file, lfp should
1541 		 * not change.
1542 		 */
1543 		if (lfp == NULL)
1544 			lfp = lop->lo_lfp;
1545 		else if (lfp != lop->lo_lfp)
1546 			panic("allnfslocks");
1547 		/*
1548 		 * If vp is NULL and cansleep != 0, a vnode must be acquired
1549 		 * from the file handle. This only occurs when called from
1550 		 * nfsrv_cleanclient().
1551 		 */
1552 		if (gottvp == 0) {
1553 			if (nfsrv_dolocallocks == 0)
1554 				tvp = NULL;
1555 			else if (vp == NULL && cansleep != 0) {
1556 				tvp = nfsvno_getvp(&lfp->lf_fh);
1557 				if (tvp != NULL)
1558 					NFSVOPUNLOCK(tvp);
1559 			} else
1560 				tvp = vp;
1561 			gottvp = 1;
1562 		}
1563 
1564 		if (tvp != NULL) {
1565 			if (cansleep == 0)
1566 				panic("allnfs2");
1567 			first = lop->lo_first;
1568 			end = lop->lo_end;
1569 			nfsrv_freenfslock(lop);
1570 			nfsrv_localunlock(tvp, lfp, first, end, p);
1571 			LIST_FOREACH_SAFE(rlp, &lfp->lf_rollback, rlck_list,
1572 			    nrlp)
1573 				free(rlp, M_NFSDROLLBACK);
1574 			LIST_INIT(&lfp->lf_rollback);
1575 		} else
1576 			nfsrv_freenfslock(lop);
1577 		lop = nlop;
1578 	}
1579 	if (vp == NULL && tvp != NULL)
1580 		vrele(tvp);
1581 }
1582 
1583 /*
1584  * Free an nfslock structure.
1585  */
1586 static void
1587 nfsrv_freenfslock(struct nfslock *lop)
1588 {
1589 
1590 	if (lop->lo_lckfile.le_prev != NULL) {
1591 		LIST_REMOVE(lop, lo_lckfile);
1592 		nfsstatsv1.srvlocks--;
1593 		nfsrv_openpluslock--;
1594 	}
1595 	LIST_REMOVE(lop, lo_lckowner);
1596 	free(lop, M_NFSDLOCK);
1597 }
1598 
1599 /*
1600  * This function frees an nfslockfile structure.
1601  */
1602 static void
1603 nfsrv_freenfslockfile(struct nfslockfile *lfp)
1604 {
1605 
1606 	LIST_REMOVE(lfp, lf_hash);
1607 	free(lfp, M_NFSDLOCKFILE);
1608 }
1609 
1610 /*
1611  * This function looks up an nfsstate structure via stateid.
1612  */
1613 static int
1614 nfsrv_getstate(struct nfsclient *clp, nfsv4stateid_t *stateidp, __unused u_int32_t flags,
1615     struct nfsstate **stpp)
1616 {
1617 	struct nfsstate *stp;
1618 	struct nfsstatehead *hp;
1619 	int error = 0;
1620 
1621 	*stpp = NULL;
1622 	hp = NFSSTATEHASH(clp, *stateidp);
1623 	LIST_FOREACH(stp, hp, ls_hash) {
1624 		if (!NFSBCMP(stp->ls_stateid.other, stateidp->other,
1625 			NFSX_STATEIDOTHER))
1626 			break;
1627 	}
1628 
1629 	/*
1630 	 * If no state id in list, return NFSERR_BADSTATEID.
1631 	 */
1632 	if (stp == LIST_END(hp)) {
1633 		error = NFSERR_BADSTATEID;
1634 		goto out;
1635 	}
1636 	*stpp = stp;
1637 
1638 out:
1639 	NFSEXITCODE(error);
1640 	return (error);
1641 }
1642 
1643 /*
1644  * This function gets an nfsstate structure via owner string.
1645  */
1646 static void
1647 nfsrv_getowner(struct nfsstatehead *hp, struct nfsstate *new_stp,
1648     struct nfsstate **stpp)
1649 {
1650 	struct nfsstate *stp;
1651 
1652 	*stpp = NULL;
1653 	LIST_FOREACH(stp, hp, ls_list) {
1654 		if (new_stp->ls_ownerlen == stp->ls_ownerlen &&
1655 		  !NFSBCMP(new_stp->ls_owner,stp->ls_owner,stp->ls_ownerlen)) {
1656 			*stpp = stp;
1657 			return;
1658 		}
1659 	}
1660 }
1661 
1662 /*
1663  * Lock control function called to update lock status.
1664  * Returns 0 upon success, -1 if there is no lock and the flags indicate
1665  * that one isn't to be created and an NFSERR_xxx for other errors.
1666  * The structures new_stp and new_lop are passed in as pointers that should
1667  * be set to NULL if the structure is used and shouldn't be free'd.
1668  * For the NFSLCK_TEST and NFSLCK_CHECK cases, the structures are
1669  * never used and can safely be allocated on the stack. For all other
1670  * cases, *new_stpp and *new_lopp should be malloc'd before the call,
1671  * in case they are used.
1672  */
1673 int
1674 nfsrv_lockctrl(vnode_t vp, struct nfsstate **new_stpp,
1675     struct nfslock **new_lopp, struct nfslockconflict *cfp,
1676     nfsquad_t clientid, nfsv4stateid_t *stateidp,
1677     __unused struct nfsexstuff *exp,
1678     struct nfsrv_descript *nd, NFSPROC_T *p)
1679 {
1680 	struct nfslock *lop;
1681 	struct nfsstate *new_stp = *new_stpp;
1682 	struct nfslock *new_lop = *new_lopp;
1683 	struct nfsstate *tstp, *mystp, *nstp;
1684 	int specialid = 0;
1685 	struct nfslockfile *lfp;
1686 	struct nfslock *other_lop = NULL;
1687 	struct nfsstate *stp, *lckstp = NULL;
1688 	struct nfsclient *clp = NULL;
1689 	u_int32_t bits;
1690 	int error = 0, haslock = 0, ret, reterr;
1691 	int getlckret, delegation = 0, filestruct_locked, vnode_unlocked = 0;
1692 	fhandle_t nfh;
1693 	uint64_t first, end;
1694 	uint32_t lock_flags;
1695 
1696 	if (new_stp->ls_flags & (NFSLCK_CHECK | NFSLCK_SETATTR)) {
1697 		/*
1698 		 * Note the special cases of "all 1s" or "all 0s" stateids and
1699 		 * let reads with all 1s go ahead.
1700 		 */
1701 		if (new_stp->ls_stateid.seqid == 0x0 &&
1702 		    new_stp->ls_stateid.other[0] == 0x0 &&
1703 		    new_stp->ls_stateid.other[1] == 0x0 &&
1704 		    new_stp->ls_stateid.other[2] == 0x0)
1705 			specialid = 1;
1706 		else if (new_stp->ls_stateid.seqid == 0xffffffff &&
1707 		    new_stp->ls_stateid.other[0] == 0xffffffff &&
1708 		    new_stp->ls_stateid.other[1] == 0xffffffff &&
1709 		    new_stp->ls_stateid.other[2] == 0xffffffff)
1710 			specialid = 2;
1711 	}
1712 
1713 	/*
1714 	 * Check for restart conditions (client and server).
1715 	 */
1716 	error = nfsrv_checkrestart(clientid, new_stp->ls_flags,
1717 	    &new_stp->ls_stateid, specialid);
1718 	if (error)
1719 		goto out;
1720 
1721 	/*
1722 	 * Check for state resource limit exceeded.
1723 	 */
1724 	if ((new_stp->ls_flags & NFSLCK_LOCK) &&
1725 	    nfsrv_openpluslock > nfsrv_v4statelimit) {
1726 		error = NFSERR_RESOURCE;
1727 		goto out;
1728 	}
1729 
1730 	/*
1731 	 * For the lock case, get another nfslock structure,
1732 	 * just in case we need it.
1733 	 * Malloc now, before we start sifting through the linked lists,
1734 	 * in case we have to wait for memory.
1735 	 */
1736 tryagain:
1737 	if (new_stp->ls_flags & NFSLCK_LOCK)
1738 		other_lop = malloc(sizeof (struct nfslock),
1739 		    M_NFSDLOCK, M_WAITOK);
1740 	filestruct_locked = 0;
1741 	reterr = 0;
1742 	lfp = NULL;
1743 
1744 	/*
1745 	 * Get the lockfile structure for CFH now, so we can do a sanity
1746 	 * check against the stateid, before incrementing the seqid#, since
1747 	 * we want to return NFSERR_BADSTATEID on failure and the seqid#
1748 	 * shouldn't be incremented for this case.
1749 	 * If nfsrv_getlockfile() returns -1, it means "not found", which
1750 	 * will be handled later.
1751 	 * If we are doing Lock/LockU and local locking is enabled, sleep
1752 	 * lock the nfslockfile structure.
1753 	 */
1754 	getlckret = nfsrv_getlockfh(vp, new_stp->ls_flags, NULL, &nfh, p);
1755 	NFSLOCKSTATE();
1756 	if (getlckret == 0) {
1757 		if ((new_stp->ls_flags & (NFSLCK_LOCK | NFSLCK_UNLOCK)) != 0 &&
1758 		    nfsrv_dolocallocks != 0 && nd->nd_repstat == 0) {
1759 			getlckret = nfsrv_getlockfile(new_stp->ls_flags, NULL,
1760 			    &lfp, &nfh, 1);
1761 			if (getlckret == 0)
1762 				filestruct_locked = 1;
1763 		} else
1764 			getlckret = nfsrv_getlockfile(new_stp->ls_flags, NULL,
1765 			    &lfp, &nfh, 0);
1766 	}
1767 	if (getlckret != 0 && getlckret != -1)
1768 		reterr = getlckret;
1769 
1770 	if (filestruct_locked != 0) {
1771 		LIST_INIT(&lfp->lf_rollback);
1772 		if ((new_stp->ls_flags & NFSLCK_LOCK)) {
1773 			/*
1774 			 * For local locking, do the advisory locking now, so
1775 			 * that any conflict can be detected. A failure later
1776 			 * can be rolled back locally. If an error is returned,
1777 			 * struct nfslockfile has been unlocked and any local
1778 			 * locking rolled back.
1779 			 */
1780 			NFSUNLOCKSTATE();
1781 			if (vnode_unlocked == 0) {
1782 				ASSERT_VOP_ELOCKED(vp, "nfsrv_lockctrl1");
1783 				vnode_unlocked = 1;
1784 				NFSVOPUNLOCK(vp);
1785 			}
1786 			reterr = nfsrv_locallock(vp, lfp,
1787 			    (new_lop->lo_flags & (NFSLCK_READ | NFSLCK_WRITE)),
1788 			    new_lop->lo_first, new_lop->lo_end, cfp, p);
1789 			NFSLOCKSTATE();
1790 		}
1791 	}
1792 
1793 	if (specialid == 0) {
1794 	    if (new_stp->ls_flags & NFSLCK_TEST) {
1795 		/*
1796 		 * RFC 3530 does not list LockT as an op that renews a
1797 		 * lease, but the consensus seems to be that it is ok
1798 		 * for a server to do so.
1799 		 */
1800 		error = nfsrv_getclient(clientid, CLOPS_RENEW, &clp, NULL,
1801 		    (nfsquad_t)((u_quad_t)0), 0, nd, p);
1802 
1803 		/*
1804 		 * Since NFSERR_EXPIRED, NFSERR_ADMINREVOKED are not valid
1805 		 * error returns for LockT, just go ahead and test for a lock,
1806 		 * since there are no locks for this client, but other locks
1807 		 * can conflict. (ie. same client will always be false)
1808 		 */
1809 		if (error == NFSERR_EXPIRED || error == NFSERR_ADMINREVOKED)
1810 		    error = 0;
1811 		lckstp = new_stp;
1812 	    } else {
1813 	      error = nfsrv_getclient(clientid, CLOPS_RENEW, &clp, NULL,
1814 		(nfsquad_t)((u_quad_t)0), 0, nd, p);
1815 	      if (error == 0)
1816 		/*
1817 		 * Look up the stateid
1818 		 */
1819 		error = nfsrv_getstate(clp, &new_stp->ls_stateid,
1820 		  new_stp->ls_flags, &stp);
1821 	      /*
1822 	       * do some sanity checks for an unconfirmed open or a
1823 	       * stateid that refers to the wrong file, for an open stateid
1824 	       */
1825 	      if (error == 0 && (stp->ls_flags & NFSLCK_OPEN) &&
1826 		  ((stp->ls_openowner->ls_flags & NFSLCK_NEEDSCONFIRM) ||
1827 		   (getlckret == 0 && stp->ls_lfp != lfp))){
1828 		      /*
1829 		       * NFSLCK_SETATTR should return OK rather than NFSERR_BADSTATEID
1830 		       * The only exception is using SETATTR with SIZE.
1831 		       * */
1832                     if ((new_stp->ls_flags &
1833                          (NFSLCK_SETATTR | NFSLCK_CHECK)) != NFSLCK_SETATTR)
1834 			     error = NFSERR_BADSTATEID;
1835 	      }
1836 
1837 		if (error == 0 &&
1838 		  (stp->ls_flags & (NFSLCK_DELEGREAD | NFSLCK_DELEGWRITE)) &&
1839 		  getlckret == 0 && stp->ls_lfp != lfp)
1840 			error = NFSERR_BADSTATEID;
1841 
1842 	      /*
1843 	       * If the lockowner stateid doesn't refer to the same file,
1844 	       * I believe that is considered ok, since some clients will
1845 	       * only create a single lockowner and use that for all locks
1846 	       * on all files.
1847 	       * For now, log it as a diagnostic, instead of considering it
1848 	       * a BadStateid.
1849 	       */
1850 	      if (error == 0 && (stp->ls_flags &
1851 		  (NFSLCK_OPEN | NFSLCK_DELEGREAD | NFSLCK_DELEGWRITE)) == 0 &&
1852 		  getlckret == 0 && stp->ls_lfp != lfp) {
1853 #ifdef DIAGNOSTIC
1854 		  printf("Got a lock statid for different file open\n");
1855 #endif
1856 		  /*
1857 		  error = NFSERR_BADSTATEID;
1858 		  */
1859 	      }
1860 
1861 	      if (error == 0) {
1862 		    if (new_stp->ls_flags & NFSLCK_OPENTOLOCK) {
1863 			/*
1864 			 * If haslock set, we've already checked the seqid.
1865 			 */
1866 			if (!haslock) {
1867 			    if (stp->ls_flags & NFSLCK_OPEN)
1868 				error = nfsrv_checkseqid(nd, new_stp->ls_seq,
1869 				    stp->ls_openowner, new_stp->ls_op);
1870 			    else
1871 				error = NFSERR_BADSTATEID;
1872 			}
1873 			if (!error)
1874 			    nfsrv_getowner(&stp->ls_open, new_stp, &lckstp);
1875 			if (lckstp) {
1876 			    /*
1877 			     * For NFSv4.1 and NFSv4.2 allow an
1878 			     * open_to_lock_owner when the lock_owner already
1879 			     * exists.  Just clear NFSLCK_OPENTOLOCK so that
1880 			     * a new lock_owner will not be created.
1881 			     * RFC7530 states that the error for NFSv4.0
1882 			     * is NFS4ERR_BAD_SEQID.
1883 			     */
1884 			    if ((nd->nd_flag & ND_NFSV41) != 0)
1885 				new_stp->ls_flags &= ~NFSLCK_OPENTOLOCK;
1886 			    else
1887 				error = NFSERR_BADSEQID;
1888 			} else
1889 			    lckstp = new_stp;
1890 		    } else if (new_stp->ls_flags&(NFSLCK_LOCK|NFSLCK_UNLOCK)) {
1891 			/*
1892 			 * If haslock set, ditto above.
1893 			 */
1894 			if (!haslock) {
1895 			    if (stp->ls_flags & NFSLCK_OPEN)
1896 				error = NFSERR_BADSTATEID;
1897 			    else
1898 				error = nfsrv_checkseqid(nd, new_stp->ls_seq,
1899 				    stp, new_stp->ls_op);
1900 			}
1901 			lckstp = stp;
1902 		    } else {
1903 			lckstp = stp;
1904 		    }
1905 	      }
1906 	      /*
1907 	       * If the seqid part of the stateid isn't the same, return
1908 	       * NFSERR_OLDSTATEID for cases other than I/O Ops.
1909 	       * For I/O Ops, only return NFSERR_OLDSTATEID if
1910 	       * nfsrv_returnoldstateid is set. (The consensus on the email
1911 	       * list was that most clients would prefer to not receive
1912 	       * NFSERR_OLDSTATEID for I/O Ops, but the RFC suggests that that
1913 	       * is what will happen, so I use the nfsrv_returnoldstateid to
1914 	       * allow for either server configuration.)
1915 	       */
1916 	      if (!error && stp->ls_stateid.seqid!=new_stp->ls_stateid.seqid &&
1917 		  (((nd->nd_flag & ND_NFSV41) == 0 &&
1918 		   (!(new_stp->ls_flags & NFSLCK_CHECK) ||
1919 		    nfsrv_returnoldstateid)) ||
1920 		   ((nd->nd_flag & ND_NFSV41) != 0 &&
1921 		    new_stp->ls_stateid.seqid != 0)))
1922 		    error = NFSERR_OLDSTATEID;
1923 	    }
1924 	}
1925 
1926 	/*
1927 	 * Now we can check for grace.
1928 	 */
1929 	if (!error)
1930 		error = nfsrv_checkgrace(nd, clp, new_stp->ls_flags);
1931 	if ((new_stp->ls_flags & NFSLCK_RECLAIM) && !error &&
1932 		nfsrv_checkstable(clp))
1933 		error = NFSERR_NOGRACE;
1934 	/*
1935 	 * If we successfully Reclaimed state, note that.
1936 	 */
1937 	if ((new_stp->ls_flags & NFSLCK_RECLAIM) && !error)
1938 		nfsrv_markstable(clp);
1939 
1940 	/*
1941 	 * At this point, either error == NFSERR_BADSTATEID or the
1942 	 * seqid# has been updated, so we can return any error.
1943 	 * If error == 0, there may be an error in:
1944 	 *    nd_repstat - Set by the calling function.
1945 	 *    reterr - Set above, if getting the nfslockfile structure
1946 	 *       or acquiring the local lock failed.
1947 	 *    (If both of these are set, nd_repstat should probably be
1948 	 *     returned, since that error was detected before this
1949 	 *     function call.)
1950 	 */
1951 	if (error != 0 || nd->nd_repstat != 0 || reterr != 0) {
1952 		if (error == 0) {
1953 			if (nd->nd_repstat != 0)
1954 				error = nd->nd_repstat;
1955 			else
1956 				error = reterr;
1957 		}
1958 		if (filestruct_locked != 0) {
1959 			/* Roll back local locks. */
1960 			NFSUNLOCKSTATE();
1961 			if (vnode_unlocked == 0) {
1962 				ASSERT_VOP_ELOCKED(vp, "nfsrv_lockctrl2");
1963 				vnode_unlocked = 1;
1964 				NFSVOPUNLOCK(vp);
1965 			}
1966 			nfsrv_locallock_rollback(vp, lfp, p);
1967 			NFSLOCKSTATE();
1968 			nfsrv_unlocklf(lfp);
1969 		}
1970 		NFSUNLOCKSTATE();
1971 		goto out;
1972 	}
1973 
1974 	/*
1975 	 * Check the nfsrv_getlockfile return.
1976 	 * Returned -1 if no structure found.
1977 	 */
1978 	if (getlckret == -1) {
1979 		error = NFSERR_EXPIRED;
1980 		/*
1981 		 * Called from lockt, so no lock is OK.
1982 		 */
1983 		if (new_stp->ls_flags & NFSLCK_TEST) {
1984 			error = 0;
1985 		} else if (new_stp->ls_flags &
1986 		    (NFSLCK_CHECK | NFSLCK_SETATTR)) {
1987 			/*
1988 			 * Called to check for a lock, OK if the stateid is all
1989 			 * 1s or all 0s, but there should be an nfsstate
1990 			 * otherwise.
1991 			 * (ie. If there is no open, I'll assume no share
1992 			 *  deny bits.)
1993 			 */
1994 			if (specialid)
1995 				error = 0;
1996 			else
1997 				error = NFSERR_BADSTATEID;
1998 		}
1999 		NFSUNLOCKSTATE();
2000 		goto out;
2001 	}
2002 
2003 	/*
2004 	 * For NFSLCK_CHECK and NFSLCK_LOCK, test for a share conflict.
2005 	 * For NFSLCK_CHECK, allow a read if write access is granted,
2006 	 * but check for a deny. For NFSLCK_LOCK, require correct access,
2007 	 * which implies a conflicting deny can't exist.
2008 	 */
2009 	if (new_stp->ls_flags & (NFSLCK_CHECK | NFSLCK_LOCK)) {
2010 	    /*
2011 	     * Four kinds of state id:
2012 	     * - specialid (all 0s or all 1s), only for NFSLCK_CHECK
2013 	     * - stateid for an open
2014 	     * - stateid for a delegation
2015 	     * - stateid for a lock owner
2016 	     */
2017 	    if (!specialid) {
2018 		if (stp->ls_flags & (NFSLCK_DELEGREAD | NFSLCK_DELEGWRITE)) {
2019 		    delegation = 1;
2020 		    mystp = stp;
2021 		    nfsrv_delaydelegtimeout(stp);
2022 	        } else if (stp->ls_flags & NFSLCK_OPEN) {
2023 		    mystp = stp;
2024 		} else {
2025 		    mystp = stp->ls_openstp;
2026 		}
2027 		/*
2028 		 * If locking or checking, require correct access
2029 		 * bit set.
2030 		 */
2031 		if (((new_stp->ls_flags & NFSLCK_LOCK) &&
2032 		     !((new_lop->lo_flags >> NFSLCK_LOCKSHIFT) &
2033 		       mystp->ls_flags & NFSLCK_ACCESSBITS)) ||
2034 		    ((new_stp->ls_flags & (NFSLCK_CHECK|NFSLCK_READACCESS)) ==
2035 		      (NFSLCK_CHECK | NFSLCK_READACCESS) &&
2036 		     !(mystp->ls_flags & NFSLCK_READACCESS) &&
2037 		     nfsrv_allowreadforwriteopen == 0) ||
2038 		    ((new_stp->ls_flags & (NFSLCK_CHECK|NFSLCK_WRITEACCESS)) ==
2039 		      (NFSLCK_CHECK | NFSLCK_WRITEACCESS) &&
2040 		     !(mystp->ls_flags & NFSLCK_WRITEACCESS))) {
2041 			if (filestruct_locked != 0) {
2042 				/* Roll back local locks. */
2043 				NFSUNLOCKSTATE();
2044 				if (vnode_unlocked == 0) {
2045 					ASSERT_VOP_ELOCKED(vp,
2046 					    "nfsrv_lockctrl3");
2047 					vnode_unlocked = 1;
2048 					NFSVOPUNLOCK(vp);
2049 				}
2050 				nfsrv_locallock_rollback(vp, lfp, p);
2051 				NFSLOCKSTATE();
2052 				nfsrv_unlocklf(lfp);
2053 			}
2054 			NFSUNLOCKSTATE();
2055 			error = NFSERR_OPENMODE;
2056 			goto out;
2057 		}
2058 	    } else
2059 		mystp = NULL;
2060 	    if ((new_stp->ls_flags & NFSLCK_CHECK) && !delegation) {
2061 		/*
2062 		 * Check for a conflicting deny bit.
2063 		 */
2064 		LIST_FOREACH(tstp, &lfp->lf_open, ls_file) {
2065 		    if (tstp != mystp) {
2066 			bits = tstp->ls_flags;
2067 			bits >>= NFSLCK_SHIFT;
2068 			if (new_stp->ls_flags & bits & NFSLCK_ACCESSBITS) {
2069 			    KASSERT(vnode_unlocked == 0,
2070 				("nfsrv_lockctrl: vnode unlocked1"));
2071 			    ret = nfsrv_clientconflict(tstp->ls_clp, &haslock,
2072 				vp, p);
2073 			    if (ret == 1) {
2074 				/*
2075 				* nfsrv_clientconflict unlocks state
2076 				 * when it returns non-zero.
2077 				 */
2078 				lckstp = NULL;
2079 				goto tryagain;
2080 			    }
2081 			    if (ret == 0)
2082 				NFSUNLOCKSTATE();
2083 			    if (ret == 2)
2084 				error = NFSERR_PERM;
2085 			    else
2086 				error = NFSERR_OPENMODE;
2087 			    goto out;
2088 			}
2089 		    }
2090 		}
2091 
2092 		/* We're outta here */
2093 		NFSUNLOCKSTATE();
2094 		goto out;
2095 	    }
2096 	}
2097 
2098 	/*
2099 	 * For setattr, just get rid of all the Delegations for other clients.
2100 	 */
2101 	if (new_stp->ls_flags & NFSLCK_SETATTR) {
2102 		KASSERT(vnode_unlocked == 0,
2103 		    ("nfsrv_lockctrl: vnode unlocked2"));
2104 		ret = nfsrv_cleandeleg(vp, lfp, clp, &haslock, p);
2105 		if (ret) {
2106 			/*
2107 			 * nfsrv_cleandeleg() unlocks state when it
2108 			 * returns non-zero.
2109 			 */
2110 			if (ret == -1) {
2111 				lckstp = NULL;
2112 				goto tryagain;
2113 			}
2114 			error = ret;
2115 			goto out;
2116 		}
2117 		if (!(new_stp->ls_flags & NFSLCK_CHECK) ||
2118 		    (LIST_EMPTY(&lfp->lf_open) && LIST_EMPTY(&lfp->lf_lock) &&
2119 		     LIST_EMPTY(&lfp->lf_deleg))) {
2120 			NFSUNLOCKSTATE();
2121 			goto out;
2122 		}
2123 	}
2124 
2125 	/*
2126 	 * Check for a conflicting delegation. If one is found, call
2127 	 * nfsrv_delegconflict() to handle it. If the v4root lock hasn't
2128 	 * been set yet, it will get the lock. Otherwise, it will recall
2129 	 * the delegation. Then, we try try again...
2130 	 * I currently believe the conflict algorithm to be:
2131 	 * For Lock Ops (Lock/LockT/LockU)
2132 	 * - there is a conflict iff a different client has a write delegation
2133 	 * For Reading (Read Op)
2134 	 * - there is a conflict iff a different client has a write delegation
2135 	 *   (the specialids are always a different client)
2136 	 * For Writing (Write/Setattr of size)
2137 	 * - there is a conflict if a different client has any delegation
2138 	 * - there is a conflict if the same client has a read delegation
2139 	 *   (I don't understand why this isn't allowed, but that seems to be
2140 	 *    the current consensus?)
2141 	 */
2142 	tstp = LIST_FIRST(&lfp->lf_deleg);
2143 	while (tstp != LIST_END(&lfp->lf_deleg)) {
2144 	    nstp = LIST_NEXT(tstp, ls_file);
2145 	    if ((((new_stp->ls_flags&(NFSLCK_LOCK|NFSLCK_UNLOCK|NFSLCK_TEST))||
2146 		 ((new_stp->ls_flags & NFSLCK_CHECK) &&
2147 		  (new_lop->lo_flags & NFSLCK_READ))) &&
2148 		  clp != tstp->ls_clp &&
2149 		 (tstp->ls_flags & NFSLCK_DELEGWRITE)) ||
2150 		 ((new_stp->ls_flags & NFSLCK_CHECK) &&
2151 		   (new_lop->lo_flags & NFSLCK_WRITE) &&
2152 		  (clp != tstp->ls_clp ||
2153 		   (tstp->ls_flags & NFSLCK_DELEGREAD)))) {
2154 		ret = 0;
2155 		if (filestruct_locked != 0) {
2156 			/* Roll back local locks. */
2157 			NFSUNLOCKSTATE();
2158 			if (vnode_unlocked == 0) {
2159 				ASSERT_VOP_ELOCKED(vp, "nfsrv_lockctrl4");
2160 				NFSVOPUNLOCK(vp);
2161 			}
2162 			nfsrv_locallock_rollback(vp, lfp, p);
2163 			NFSLOCKSTATE();
2164 			nfsrv_unlocklf(lfp);
2165 			NFSUNLOCKSTATE();
2166 			NFSVOPLOCK(vp, LK_EXCLUSIVE | LK_RETRY);
2167 			vnode_unlocked = 0;
2168 			if (VN_IS_DOOMED(vp))
2169 				ret = NFSERR_SERVERFAULT;
2170 			NFSLOCKSTATE();
2171 		}
2172 		if (ret == 0)
2173 			ret = nfsrv_delegconflict(tstp, &haslock, p, vp);
2174 		if (ret) {
2175 		    /*
2176 		     * nfsrv_delegconflict unlocks state when it
2177 		     * returns non-zero, which it always does.
2178 		     */
2179 		    if (other_lop) {
2180 			free(other_lop, M_NFSDLOCK);
2181 			other_lop = NULL;
2182 		    }
2183 		    if (ret == -1) {
2184 			lckstp = NULL;
2185 			goto tryagain;
2186 		    }
2187 		    error = ret;
2188 		    goto out;
2189 		}
2190 		/* Never gets here. */
2191 	    }
2192 	    tstp = nstp;
2193 	}
2194 
2195 	/*
2196 	 * Handle the unlock case by calling nfsrv_updatelock().
2197 	 * (Should I have done some access checking above for unlock? For now,
2198 	 *  just let it happen.)
2199 	 */
2200 	if (new_stp->ls_flags & NFSLCK_UNLOCK) {
2201 		first = new_lop->lo_first;
2202 		end = new_lop->lo_end;
2203 		nfsrv_updatelock(stp, new_lopp, &other_lop, lfp);
2204 		stateidp->seqid = ++(stp->ls_stateid.seqid);
2205 		if ((nd->nd_flag & ND_NFSV41) != 0 && stateidp->seqid == 0)
2206 			stateidp->seqid = stp->ls_stateid.seqid = 1;
2207 		stateidp->other[0] = stp->ls_stateid.other[0];
2208 		stateidp->other[1] = stp->ls_stateid.other[1];
2209 		stateidp->other[2] = stp->ls_stateid.other[2];
2210 		if (filestruct_locked != 0) {
2211 			NFSUNLOCKSTATE();
2212 			if (vnode_unlocked == 0) {
2213 				ASSERT_VOP_ELOCKED(vp, "nfsrv_lockctrl5");
2214 				vnode_unlocked = 1;
2215 				NFSVOPUNLOCK(vp);
2216 			}
2217 			/* Update the local locks. */
2218 			nfsrv_localunlock(vp, lfp, first, end, p);
2219 			NFSLOCKSTATE();
2220 			nfsrv_unlocklf(lfp);
2221 		}
2222 		NFSUNLOCKSTATE();
2223 		goto out;
2224 	}
2225 
2226 	/*
2227 	 * Search for a conflicting lock. A lock conflicts if:
2228 	 * - the lock range overlaps and
2229 	 * - at least one lock is a write lock and
2230 	 * - it is not owned by the same lock owner
2231 	 */
2232 	if (!delegation) {
2233 	  LIST_FOREACH(lop, &lfp->lf_lock, lo_lckfile) {
2234 	    if (new_lop->lo_end > lop->lo_first &&
2235 		new_lop->lo_first < lop->lo_end &&
2236 		(new_lop->lo_flags == NFSLCK_WRITE ||
2237 		 lop->lo_flags == NFSLCK_WRITE) &&
2238 		lckstp != lop->lo_stp &&
2239 		(clp != lop->lo_stp->ls_clp ||
2240 		 lckstp->ls_ownerlen != lop->lo_stp->ls_ownerlen ||
2241 		 NFSBCMP(lckstp->ls_owner, lop->lo_stp->ls_owner,
2242 		    lckstp->ls_ownerlen))) {
2243 		if (other_lop) {
2244 		    free(other_lop, M_NFSDLOCK);
2245 		    other_lop = NULL;
2246 		}
2247 		if (vnode_unlocked != 0)
2248 		    ret = nfsrv_clientconflict(lop->lo_stp->ls_clp, &haslock,
2249 			NULL, p);
2250 		else
2251 		    ret = nfsrv_clientconflict(lop->lo_stp->ls_clp, &haslock,
2252 			vp, p);
2253 		if (ret == 1) {
2254 		    if (filestruct_locked != 0) {
2255 			if (vnode_unlocked == 0) {
2256 				ASSERT_VOP_ELOCKED(vp, "nfsrv_lockctrl6");
2257 				NFSVOPUNLOCK(vp);
2258 			}
2259 			/* Roll back local locks. */
2260 			nfsrv_locallock_rollback(vp, lfp, p);
2261 			NFSLOCKSTATE();
2262 			nfsrv_unlocklf(lfp);
2263 			NFSUNLOCKSTATE();
2264 			NFSVOPLOCK(vp, LK_EXCLUSIVE | LK_RETRY);
2265 			vnode_unlocked = 0;
2266 			if (VN_IS_DOOMED(vp)) {
2267 				error = NFSERR_SERVERFAULT;
2268 				goto out;
2269 			}
2270 		    }
2271 		    /*
2272 		     * nfsrv_clientconflict() unlocks state when it
2273 		     * returns non-zero.
2274 		     */
2275 		    lckstp = NULL;
2276 		    goto tryagain;
2277 		}
2278 		/*
2279 		 * Found a conflicting lock, so record the conflict and
2280 		 * return the error.
2281 		 */
2282 		if (cfp != NULL && ret == 0) {
2283 		    cfp->cl_clientid.lval[0]=lop->lo_stp->ls_stateid.other[0];
2284 		    cfp->cl_clientid.lval[1]=lop->lo_stp->ls_stateid.other[1];
2285 		    cfp->cl_first = lop->lo_first;
2286 		    cfp->cl_end = lop->lo_end;
2287 		    cfp->cl_flags = lop->lo_flags;
2288 		    cfp->cl_ownerlen = lop->lo_stp->ls_ownerlen;
2289 		    NFSBCOPY(lop->lo_stp->ls_owner, cfp->cl_owner,
2290 			cfp->cl_ownerlen);
2291 		}
2292 		if (ret == 2)
2293 		    error = NFSERR_PERM;
2294 		else if (new_stp->ls_flags & NFSLCK_RECLAIM)
2295 		    error = NFSERR_RECLAIMCONFLICT;
2296 		else if (new_stp->ls_flags & NFSLCK_CHECK)
2297 		    error = NFSERR_LOCKED;
2298 		else
2299 		    error = NFSERR_DENIED;
2300 		if (filestruct_locked != 0 && ret == 0) {
2301 			/* Roll back local locks. */
2302 			NFSUNLOCKSTATE();
2303 			if (vnode_unlocked == 0) {
2304 				ASSERT_VOP_ELOCKED(vp, "nfsrv_lockctrl7");
2305 				vnode_unlocked = 1;
2306 				NFSVOPUNLOCK(vp);
2307 			}
2308 			nfsrv_locallock_rollback(vp, lfp, p);
2309 			NFSLOCKSTATE();
2310 			nfsrv_unlocklf(lfp);
2311 		}
2312 		if (ret == 0)
2313 			NFSUNLOCKSTATE();
2314 		goto out;
2315 	    }
2316 	  }
2317 	}
2318 
2319 	/*
2320 	 * We only get here if there was no lock that conflicted.
2321 	 */
2322 	if (new_stp->ls_flags & (NFSLCK_TEST | NFSLCK_CHECK)) {
2323 		NFSUNLOCKSTATE();
2324 		goto out;
2325 	}
2326 
2327 	/*
2328 	 * We only get here when we are creating or modifying a lock.
2329 	 * There are two variants:
2330 	 * - exist_lock_owner where lock_owner exists
2331 	 * - open_to_lock_owner with new lock_owner
2332 	 */
2333 	first = new_lop->lo_first;
2334 	end = new_lop->lo_end;
2335 	lock_flags = new_lop->lo_flags;
2336 	if (!(new_stp->ls_flags & NFSLCK_OPENTOLOCK)) {
2337 		nfsrv_updatelock(lckstp, new_lopp, &other_lop, lfp);
2338 		stateidp->seqid = ++(lckstp->ls_stateid.seqid);
2339 		if ((nd->nd_flag & ND_NFSV41) != 0 && stateidp->seqid == 0)
2340 			stateidp->seqid = lckstp->ls_stateid.seqid = 1;
2341 		stateidp->other[0] = lckstp->ls_stateid.other[0];
2342 		stateidp->other[1] = lckstp->ls_stateid.other[1];
2343 		stateidp->other[2] = lckstp->ls_stateid.other[2];
2344 	} else {
2345 		/*
2346 		 * The new open_to_lock_owner case.
2347 		 * Link the new nfsstate into the lists.
2348 		 */
2349 		new_stp->ls_seq = new_stp->ls_opentolockseq;
2350 		nfsrvd_refcache(new_stp->ls_op);
2351 		stateidp->seqid = new_stp->ls_stateid.seqid = 1;
2352 		stateidp->other[0] = new_stp->ls_stateid.other[0] =
2353 		    clp->lc_clientid.lval[0];
2354 		stateidp->other[1] = new_stp->ls_stateid.other[1] =
2355 		    clp->lc_clientid.lval[1];
2356 		stateidp->other[2] = new_stp->ls_stateid.other[2] =
2357 		    nfsrv_nextstateindex(clp);
2358 		new_stp->ls_clp = clp;
2359 		LIST_INIT(&new_stp->ls_lock);
2360 		new_stp->ls_openstp = stp;
2361 		new_stp->ls_lfp = lfp;
2362 		nfsrv_insertlock(new_lop, (struct nfslock *)new_stp, new_stp,
2363 		    lfp);
2364 		LIST_INSERT_HEAD(NFSSTATEHASH(clp, new_stp->ls_stateid),
2365 		    new_stp, ls_hash);
2366 		LIST_INSERT_HEAD(&stp->ls_open, new_stp, ls_list);
2367 		*new_lopp = NULL;
2368 		*new_stpp = NULL;
2369 		nfsstatsv1.srvlockowners++;
2370 		nfsrv_openpluslock++;
2371 	}
2372 	if (filestruct_locked != 0) {
2373 		NFSUNLOCKSTATE();
2374 		nfsrv_locallock_commit(lfp, lock_flags, first, end);
2375 		NFSLOCKSTATE();
2376 		nfsrv_unlocklf(lfp);
2377 	}
2378 	NFSUNLOCKSTATE();
2379 
2380 out:
2381 	if (haslock) {
2382 		NFSLOCKV4ROOTMUTEX();
2383 		nfsv4_unlock(&nfsv4rootfs_lock, 1);
2384 		NFSUNLOCKV4ROOTMUTEX();
2385 	}
2386 	if (vnode_unlocked != 0) {
2387 		NFSVOPLOCK(vp, LK_EXCLUSIVE | LK_RETRY);
2388 		if (error == 0 && VN_IS_DOOMED(vp))
2389 			error = NFSERR_SERVERFAULT;
2390 	}
2391 	if (other_lop)
2392 		free(other_lop, M_NFSDLOCK);
2393 	NFSEXITCODE2(error, nd);
2394 	return (error);
2395 }
2396 
2397 /*
2398  * Check for state errors for Open.
2399  * repstat is passed back out as an error if more critical errors
2400  * are not detected.
2401  */
2402 int
2403 nfsrv_opencheck(nfsquad_t clientid, nfsv4stateid_t *stateidp,
2404     struct nfsstate *new_stp, vnode_t vp, struct nfsrv_descript *nd,
2405     NFSPROC_T *p, int repstat)
2406 {
2407 	struct nfsstate *stp, *nstp;
2408 	struct nfsclient *clp;
2409 	struct nfsstate *ownerstp;
2410 	struct nfslockfile *lfp, *new_lfp;
2411 	int error = 0, haslock = 0, ret, readonly = 0, getfhret = 0;
2412 
2413 	if ((new_stp->ls_flags & NFSLCK_SHAREBITS) == NFSLCK_READACCESS)
2414 		readonly = 1;
2415 	/*
2416 	 * Check for restart conditions (client and server).
2417 	 */
2418 	error = nfsrv_checkrestart(clientid, new_stp->ls_flags,
2419 		&new_stp->ls_stateid, 0);
2420 	if (error)
2421 		goto out;
2422 
2423 	/*
2424 	 * Check for state resource limit exceeded.
2425 	 * Technically this should be SMP protected, but the worst
2426 	 * case error is "out by one or two" on the count when it
2427 	 * returns NFSERR_RESOURCE and the limit is just a rather
2428 	 * arbitrary high water mark, so no harm is done.
2429 	 */
2430 	if (nfsrv_openpluslock > nfsrv_v4statelimit) {
2431 		error = NFSERR_RESOURCE;
2432 		goto out;
2433 	}
2434 
2435 tryagain:
2436 	new_lfp = malloc(sizeof (struct nfslockfile),
2437 	    M_NFSDLOCKFILE, M_WAITOK);
2438 	if (vp)
2439 		getfhret = nfsrv_getlockfh(vp, new_stp->ls_flags, new_lfp,
2440 		    NULL, p);
2441 	NFSLOCKSTATE();
2442 	/*
2443 	 * Get the nfsclient structure.
2444 	 */
2445 	error = nfsrv_getclient(clientid, CLOPS_RENEW, &clp, NULL,
2446 	    (nfsquad_t)((u_quad_t)0), 0, nd, p);
2447 
2448 	/*
2449 	 * Look up the open owner. See if it needs confirmation and
2450 	 * check the seq#, as required.
2451 	 */
2452 	if (!error)
2453 		nfsrv_getowner(&clp->lc_open, new_stp, &ownerstp);
2454 
2455 	if (!error && ownerstp) {
2456 		error = nfsrv_checkseqid(nd, new_stp->ls_seq, ownerstp,
2457 		    new_stp->ls_op);
2458 		/*
2459 		 * If the OpenOwner hasn't been confirmed, assume the
2460 		 * old one was a replay and this one is ok.
2461 		 * See: RFC3530 Sec. 14.2.18.
2462 		 */
2463 		if (error == NFSERR_BADSEQID &&
2464 		    (ownerstp->ls_flags & NFSLCK_NEEDSCONFIRM))
2465 			error = 0;
2466 	}
2467 
2468 	/*
2469 	 * Check for grace.
2470 	 */
2471 	if (!error)
2472 		error = nfsrv_checkgrace(nd, clp, new_stp->ls_flags);
2473 	if ((new_stp->ls_flags & NFSLCK_RECLAIM) && !error &&
2474 		nfsrv_checkstable(clp))
2475 		error = NFSERR_NOGRACE;
2476 
2477 	/*
2478 	 * If none of the above errors occurred, let repstat be
2479 	 * returned.
2480 	 */
2481 	if (repstat && !error)
2482 		error = repstat;
2483 	if (error) {
2484 		NFSUNLOCKSTATE();
2485 		if (haslock) {
2486 			NFSLOCKV4ROOTMUTEX();
2487 			nfsv4_unlock(&nfsv4rootfs_lock, 1);
2488 			NFSUNLOCKV4ROOTMUTEX();
2489 		}
2490 		free(new_lfp, M_NFSDLOCKFILE);
2491 		goto out;
2492 	}
2493 
2494 	/*
2495 	 * If vp == NULL, the file doesn't exist yet, so return ok.
2496 	 * (This always happens on the first pass, so haslock must be 0.)
2497 	 */
2498 	if (vp == NULL) {
2499 		NFSUNLOCKSTATE();
2500 		free(new_lfp, M_NFSDLOCKFILE);
2501 		goto out;
2502 	}
2503 
2504 	/*
2505 	 * Get the structure for the underlying file.
2506 	 */
2507 	if (getfhret)
2508 		error = getfhret;
2509 	else
2510 		error = nfsrv_getlockfile(new_stp->ls_flags, &new_lfp, &lfp,
2511 		    NULL, 0);
2512 	if (new_lfp)
2513 		free(new_lfp, M_NFSDLOCKFILE);
2514 	if (error) {
2515 		NFSUNLOCKSTATE();
2516 		if (haslock) {
2517 			NFSLOCKV4ROOTMUTEX();
2518 			nfsv4_unlock(&nfsv4rootfs_lock, 1);
2519 			NFSUNLOCKV4ROOTMUTEX();
2520 		}
2521 		goto out;
2522 	}
2523 
2524 	/*
2525 	 * Search for a conflicting open/share.
2526 	 */
2527 	if (new_stp->ls_flags & NFSLCK_DELEGCUR) {
2528 	    /*
2529 	     * For Delegate_Cur, search for the matching Delegation,
2530 	     * which indicates no conflict.
2531 	     * An old delegation should have been recovered by the
2532 	     * client doing a Claim_DELEGATE_Prev, so I won't let
2533 	     * it match and return NFSERR_EXPIRED. Should I let it
2534 	     * match?
2535 	     */
2536 	    LIST_FOREACH(stp, &lfp->lf_deleg, ls_file) {
2537 		if (!(stp->ls_flags & NFSLCK_OLDDELEG) &&
2538 		    (((nd->nd_flag & ND_NFSV41) != 0 &&
2539 		    stateidp->seqid == 0) ||
2540 		    stateidp->seqid == stp->ls_stateid.seqid) &&
2541 		    !NFSBCMP(stateidp->other, stp->ls_stateid.other,
2542 			  NFSX_STATEIDOTHER))
2543 			break;
2544 	    }
2545 	    if (stp == LIST_END(&lfp->lf_deleg) ||
2546 		((new_stp->ls_flags & NFSLCK_WRITEACCESS) &&
2547 		 (stp->ls_flags & NFSLCK_DELEGREAD))) {
2548 		NFSUNLOCKSTATE();
2549 		if (haslock) {
2550 			NFSLOCKV4ROOTMUTEX();
2551 			nfsv4_unlock(&nfsv4rootfs_lock, 1);
2552 			NFSUNLOCKV4ROOTMUTEX();
2553 		}
2554 		error = NFSERR_EXPIRED;
2555 		goto out;
2556 	    }
2557 	}
2558 
2559 	/*
2560 	 * Check for access/deny bit conflicts. I check for the same
2561 	 * owner as well, in case the client didn't bother.
2562 	 */
2563 	LIST_FOREACH(stp, &lfp->lf_open, ls_file) {
2564 		if (!(new_stp->ls_flags & NFSLCK_DELEGCUR) &&
2565 		    (((new_stp->ls_flags & NFSLCK_ACCESSBITS) &
2566 		      ((stp->ls_flags>>NFSLCK_SHIFT) & NFSLCK_ACCESSBITS))||
2567 		     ((stp->ls_flags & NFSLCK_ACCESSBITS) &
2568 		      ((new_stp->ls_flags>>NFSLCK_SHIFT)&NFSLCK_ACCESSBITS)))){
2569 			ret = nfsrv_clientconflict(stp->ls_clp,&haslock,vp,p);
2570 			if (ret == 1) {
2571 				/*
2572 				 * nfsrv_clientconflict() unlocks
2573 				 * state when it returns non-zero.
2574 				 */
2575 				goto tryagain;
2576 			}
2577 			if (ret == 2)
2578 				error = NFSERR_PERM;
2579 			else if (new_stp->ls_flags & NFSLCK_RECLAIM)
2580 				error = NFSERR_RECLAIMCONFLICT;
2581 			else
2582 				error = NFSERR_SHAREDENIED;
2583 			if (ret == 0)
2584 				NFSUNLOCKSTATE();
2585 			if (haslock) {
2586 				NFSLOCKV4ROOTMUTEX();
2587 				nfsv4_unlock(&nfsv4rootfs_lock, 1);
2588 				NFSUNLOCKV4ROOTMUTEX();
2589 			}
2590 			goto out;
2591 		}
2592 	}
2593 
2594 	/*
2595 	 * Check for a conflicting delegation. If one is found, call
2596 	 * nfsrv_delegconflict() to handle it. If the v4root lock hasn't
2597 	 * been set yet, it will get the lock. Otherwise, it will recall
2598 	 * the delegation. Then, we try try again...
2599 	 * (If NFSLCK_DELEGCUR is set, it has a delegation, so there
2600 	 *  isn't a conflict.)
2601 	 * I currently believe the conflict algorithm to be:
2602 	 * For Open with Read Access and Deny None
2603 	 * - there is a conflict iff a different client has a write delegation
2604 	 * For Open with other Write Access or any Deny except None
2605 	 * - there is a conflict if a different client has any delegation
2606 	 * - there is a conflict if the same client has a read delegation
2607 	 *   (The current consensus is that this last case should be
2608 	 *    considered a conflict since the client with a read delegation
2609 	 *    could have done an Open with ReadAccess and WriteDeny
2610 	 *    locally and then not have checked for the WriteDeny.)
2611 	 * Don't check for a Reclaim, since that will be dealt with
2612 	 * by nfsrv_openctrl().
2613 	 */
2614 	if (!(new_stp->ls_flags &
2615 		(NFSLCK_DELEGPREV | NFSLCK_DELEGCUR | NFSLCK_RECLAIM))) {
2616 	    stp = LIST_FIRST(&lfp->lf_deleg);
2617 	    while (stp != LIST_END(&lfp->lf_deleg)) {
2618 		nstp = LIST_NEXT(stp, ls_file);
2619 		if ((readonly && stp->ls_clp != clp &&
2620 		       (stp->ls_flags & NFSLCK_DELEGWRITE)) ||
2621 		    (!readonly && (stp->ls_clp != clp ||
2622 		         (stp->ls_flags & NFSLCK_DELEGREAD)))) {
2623 			ret = nfsrv_delegconflict(stp, &haslock, p, vp);
2624 			if (ret) {
2625 			    /*
2626 			     * nfsrv_delegconflict() unlocks state
2627 			     * when it returns non-zero.
2628 			     */
2629 			    if (ret == -1)
2630 				goto tryagain;
2631 			    error = ret;
2632 			    goto out;
2633 			}
2634 		}
2635 		stp = nstp;
2636 	    }
2637 	}
2638 	NFSUNLOCKSTATE();
2639 	if (haslock) {
2640 		NFSLOCKV4ROOTMUTEX();
2641 		nfsv4_unlock(&nfsv4rootfs_lock, 1);
2642 		NFSUNLOCKV4ROOTMUTEX();
2643 	}
2644 
2645 out:
2646 	NFSEXITCODE2(error, nd);
2647 	return (error);
2648 }
2649 
2650 /*
2651  * Open control function to create/update open state for an open.
2652  */
2653 int
2654 nfsrv_openctrl(struct nfsrv_descript *nd, vnode_t vp,
2655     struct nfsstate **new_stpp, nfsquad_t clientid, nfsv4stateid_t *stateidp,
2656     nfsv4stateid_t *delegstateidp, u_int32_t *rflagsp, struct nfsexstuff *exp,
2657     NFSPROC_T *p, u_quad_t filerev)
2658 {
2659 	struct nfsstate *new_stp = *new_stpp;
2660 	struct nfsstate *stp, *nstp;
2661 	struct nfsstate *openstp = NULL, *new_open, *ownerstp, *new_deleg;
2662 	struct nfslockfile *lfp, *new_lfp;
2663 	struct nfsclient *clp;
2664 	int error = 0, haslock = 0, ret, delegate = 1, writedeleg = 1;
2665 	int readonly = 0, cbret = 1, getfhret = 0;
2666 	int gotstate = 0, len = 0;
2667 	u_char *clidp = NULL;
2668 
2669 	if ((new_stp->ls_flags & NFSLCK_SHAREBITS) == NFSLCK_READACCESS)
2670 		readonly = 1;
2671 	/*
2672 	 * Check for restart conditions (client and server).
2673 	 * (Paranoia, should have been detected by nfsrv_opencheck().)
2674 	 * If an error does show up, return NFSERR_EXPIRED, since the
2675 	 * the seqid# has already been incremented.
2676 	 */
2677 	error = nfsrv_checkrestart(clientid, new_stp->ls_flags,
2678 	    &new_stp->ls_stateid, 0);
2679 	if (error) {
2680 		printf("Nfsd: openctrl unexpected restart err=%d\n",
2681 		    error);
2682 		error = NFSERR_EXPIRED;
2683 		goto out;
2684 	}
2685 
2686 	clidp = malloc(NFSV4_OPAQUELIMIT, M_TEMP, M_WAITOK);
2687 tryagain:
2688 	new_lfp = malloc(sizeof (struct nfslockfile),
2689 	    M_NFSDLOCKFILE, M_WAITOK);
2690 	new_open = malloc(sizeof (struct nfsstate),
2691 	    M_NFSDSTATE, M_WAITOK);
2692 	new_deleg = malloc(sizeof (struct nfsstate),
2693 	    M_NFSDSTATE, M_WAITOK);
2694 	getfhret = nfsrv_getlockfh(vp, new_stp->ls_flags, new_lfp,
2695 	    NULL, p);
2696 	NFSLOCKSTATE();
2697 	/*
2698 	 * Get the client structure. Since the linked lists could be changed
2699 	 * by other nfsd processes if this process does a tsleep(), one of
2700 	 * two things must be done.
2701 	 * 1 - don't tsleep()
2702 	 * or
2703 	 * 2 - get the nfsv4_lock() { indicated by haslock == 1 }
2704 	 *     before using the lists, since this lock stops the other
2705 	 *     nfsd. This should only be used for rare cases, since it
2706 	 *     essentially single threads the nfsd.
2707 	 *     At this time, it is only done for cases where the stable
2708 	 *     storage file must be written prior to completion of state
2709 	 *     expiration.
2710 	 */
2711 	error = nfsrv_getclient(clientid, CLOPS_RENEW, &clp, NULL,
2712 	    (nfsquad_t)((u_quad_t)0), 0, nd, p);
2713 	if (!error && (clp->lc_flags & LCL_NEEDSCBNULL) &&
2714 	    clp->lc_program) {
2715 		/*
2716 		 * This happens on the first open for a client
2717 		 * that supports callbacks.
2718 		 */
2719 		NFSUNLOCKSTATE();
2720 		/*
2721 		 * Although nfsrv_docallback() will sleep, clp won't
2722 		 * go away, since they are only removed when the
2723 		 * nfsv4_lock() has blocked the nfsd threads. The
2724 		 * fields in clp can change, but having multiple
2725 		 * threads do this Null callback RPC should be
2726 		 * harmless.
2727 		 */
2728 		cbret = nfsrv_docallback(clp, NFSV4PROC_CBNULL,
2729 		    NULL, 0, NULL, NULL, NULL, 0, p);
2730 		NFSLOCKSTATE();
2731 		clp->lc_flags &= ~LCL_NEEDSCBNULL;
2732 		if (!cbret)
2733 			clp->lc_flags |= LCL_CALLBACKSON;
2734 	}
2735 
2736 	/*
2737 	 * Look up the open owner. See if it needs confirmation and
2738 	 * check the seq#, as required.
2739 	 */
2740 	if (!error)
2741 		nfsrv_getowner(&clp->lc_open, new_stp, &ownerstp);
2742 
2743 	if (error) {
2744 		NFSUNLOCKSTATE();
2745 		printf("Nfsd: openctrl unexpected state err=%d\n",
2746 			error);
2747 		free(new_lfp, M_NFSDLOCKFILE);
2748 		free(new_open, M_NFSDSTATE);
2749 		free(new_deleg, M_NFSDSTATE);
2750 		if (haslock) {
2751 			NFSLOCKV4ROOTMUTEX();
2752 			nfsv4_unlock(&nfsv4rootfs_lock, 1);
2753 			NFSUNLOCKV4ROOTMUTEX();
2754 		}
2755 		error = NFSERR_EXPIRED;
2756 		goto out;
2757 	}
2758 
2759 	if (new_stp->ls_flags & NFSLCK_RECLAIM)
2760 		nfsrv_markstable(clp);
2761 
2762 	/*
2763 	 * Get the structure for the underlying file.
2764 	 */
2765 	if (getfhret)
2766 		error = getfhret;
2767 	else
2768 		error = nfsrv_getlockfile(new_stp->ls_flags, &new_lfp, &lfp,
2769 		    NULL, 0);
2770 	if (new_lfp)
2771 		free(new_lfp, M_NFSDLOCKFILE);
2772 	if (error) {
2773 		NFSUNLOCKSTATE();
2774 		printf("Nfsd openctrl unexpected getlockfile err=%d\n",
2775 		    error);
2776 		free(new_open, M_NFSDSTATE);
2777 		free(new_deleg, M_NFSDSTATE);
2778 		if (haslock) {
2779 			NFSLOCKV4ROOTMUTEX();
2780 			nfsv4_unlock(&nfsv4rootfs_lock, 1);
2781 			NFSUNLOCKV4ROOTMUTEX();
2782 		}
2783 		goto out;
2784 	}
2785 
2786 	/*
2787 	 * Search for a conflicting open/share.
2788 	 */
2789 	if (new_stp->ls_flags & NFSLCK_DELEGCUR) {
2790 	    /*
2791 	     * For Delegate_Cur, search for the matching Delegation,
2792 	     * which indicates no conflict.
2793 	     * An old delegation should have been recovered by the
2794 	     * client doing a Claim_DELEGATE_Prev, so I won't let
2795 	     * it match and return NFSERR_EXPIRED. Should I let it
2796 	     * match?
2797 	     */
2798 	    LIST_FOREACH(stp, &lfp->lf_deleg, ls_file) {
2799 		if (!(stp->ls_flags & NFSLCK_OLDDELEG) &&
2800 		    (((nd->nd_flag & ND_NFSV41) != 0 &&
2801 		    stateidp->seqid == 0) ||
2802 		    stateidp->seqid == stp->ls_stateid.seqid) &&
2803 		    !NFSBCMP(stateidp->other, stp->ls_stateid.other,
2804 			NFSX_STATEIDOTHER))
2805 			break;
2806 	    }
2807 	    if (stp == LIST_END(&lfp->lf_deleg) ||
2808 		((new_stp->ls_flags & NFSLCK_WRITEACCESS) &&
2809 		 (stp->ls_flags & NFSLCK_DELEGREAD))) {
2810 		NFSUNLOCKSTATE();
2811 		printf("Nfsd openctrl unexpected expiry\n");
2812 		free(new_open, M_NFSDSTATE);
2813 		free(new_deleg, M_NFSDSTATE);
2814 		if (haslock) {
2815 			NFSLOCKV4ROOTMUTEX();
2816 			nfsv4_unlock(&nfsv4rootfs_lock, 1);
2817 			NFSUNLOCKV4ROOTMUTEX();
2818 		}
2819 		error = NFSERR_EXPIRED;
2820 		goto out;
2821 	    }
2822 
2823 	    /*
2824 	     * Don't issue a Delegation, since one already exists and
2825 	     * delay delegation timeout, as required.
2826 	     */
2827 	    delegate = 0;
2828 	    nfsrv_delaydelegtimeout(stp);
2829 	}
2830 
2831 	/*
2832 	 * Check for access/deny bit conflicts. I also check for the
2833 	 * same owner, since the client might not have bothered to check.
2834 	 * Also, note an open for the same file and owner, if found,
2835 	 * which is all we do here for Delegate_Cur, since conflict
2836 	 * checking is already done.
2837 	 */
2838 	LIST_FOREACH(stp, &lfp->lf_open, ls_file) {
2839 		if (ownerstp && stp->ls_openowner == ownerstp)
2840 			openstp = stp;
2841 		if (!(new_stp->ls_flags & NFSLCK_DELEGCUR)) {
2842 		    /*
2843 		     * If another client has the file open, the only
2844 		     * delegation that can be issued is a Read delegation
2845 		     * and only if it is a Read open with Deny none.
2846 		     */
2847 		    if (clp != stp->ls_clp) {
2848 			if ((stp->ls_flags & NFSLCK_SHAREBITS) ==
2849 			    NFSLCK_READACCESS)
2850 			    writedeleg = 0;
2851 			else
2852 			    delegate = 0;
2853 		    }
2854 		    if(((new_stp->ls_flags & NFSLCK_ACCESSBITS) &
2855 		        ((stp->ls_flags>>NFSLCK_SHIFT) & NFSLCK_ACCESSBITS))||
2856 		       ((stp->ls_flags & NFSLCK_ACCESSBITS) &
2857 		        ((new_stp->ls_flags>>NFSLCK_SHIFT)&NFSLCK_ACCESSBITS))){
2858 			ret = nfsrv_clientconflict(stp->ls_clp,&haslock,vp,p);
2859 			if (ret == 1) {
2860 				/*
2861 				 * nfsrv_clientconflict() unlocks state
2862 				 * when it returns non-zero.
2863 				 */
2864 				free(new_open, M_NFSDSTATE);
2865 				free(new_deleg, M_NFSDSTATE);
2866 				openstp = NULL;
2867 				goto tryagain;
2868 			}
2869 			if (ret == 2)
2870 				error = NFSERR_PERM;
2871 			else if (new_stp->ls_flags & NFSLCK_RECLAIM)
2872 				error = NFSERR_RECLAIMCONFLICT;
2873 			else
2874 				error = NFSERR_SHAREDENIED;
2875 			if (ret == 0)
2876 				NFSUNLOCKSTATE();
2877 			if (haslock) {
2878 				NFSLOCKV4ROOTMUTEX();
2879 				nfsv4_unlock(&nfsv4rootfs_lock, 1);
2880 				NFSUNLOCKV4ROOTMUTEX();
2881 			}
2882 			free(new_open, M_NFSDSTATE);
2883 			free(new_deleg, M_NFSDSTATE);
2884 			printf("nfsd openctrl unexpected client cnfl\n");
2885 			goto out;
2886 		    }
2887 		}
2888 	}
2889 
2890 	/*
2891 	 * Check for a conflicting delegation. If one is found, call
2892 	 * nfsrv_delegconflict() to handle it. If the v4root lock hasn't
2893 	 * been set yet, it will get the lock. Otherwise, it will recall
2894 	 * the delegation. Then, we try try again...
2895 	 * (If NFSLCK_DELEGCUR is set, it has a delegation, so there
2896 	 *  isn't a conflict.)
2897 	 * I currently believe the conflict algorithm to be:
2898 	 * For Open with Read Access and Deny None
2899 	 * - there is a conflict iff a different client has a write delegation
2900 	 * For Open with other Write Access or any Deny except None
2901 	 * - there is a conflict if a different client has any delegation
2902 	 * - there is a conflict if the same client has a read delegation
2903 	 *   (The current consensus is that this last case should be
2904 	 *    considered a conflict since the client with a read delegation
2905 	 *    could have done an Open with ReadAccess and WriteDeny
2906 	 *    locally and then not have checked for the WriteDeny.)
2907 	 */
2908 	if (!(new_stp->ls_flags & (NFSLCK_DELEGPREV | NFSLCK_DELEGCUR))) {
2909 	    stp = LIST_FIRST(&lfp->lf_deleg);
2910 	    while (stp != LIST_END(&lfp->lf_deleg)) {
2911 		nstp = LIST_NEXT(stp, ls_file);
2912 		if (stp->ls_clp != clp && (stp->ls_flags & NFSLCK_DELEGREAD))
2913 			writedeleg = 0;
2914 		else
2915 			delegate = 0;
2916 		if ((readonly && stp->ls_clp != clp &&
2917 		       (stp->ls_flags & NFSLCK_DELEGWRITE)) ||
2918 		    (!readonly && (stp->ls_clp != clp ||
2919 		         (stp->ls_flags & NFSLCK_DELEGREAD)))) {
2920 		    if (new_stp->ls_flags & NFSLCK_RECLAIM) {
2921 			delegate = 2;
2922 		    } else {
2923 			ret = nfsrv_delegconflict(stp, &haslock, p, vp);
2924 			if (ret) {
2925 			    /*
2926 			     * nfsrv_delegconflict() unlocks state
2927 			     * when it returns non-zero.
2928 			     */
2929 			    printf("Nfsd openctrl unexpected deleg cnfl\n");
2930 			    free(new_open, M_NFSDSTATE);
2931 			    free(new_deleg, M_NFSDSTATE);
2932 			    if (ret == -1) {
2933 				openstp = NULL;
2934 				goto tryagain;
2935 			    }
2936 			    error = ret;
2937 			    goto out;
2938 			}
2939 		    }
2940 		}
2941 		stp = nstp;
2942 	    }
2943 	}
2944 
2945 	/*
2946 	 * We only get here if there was no open that conflicted.
2947 	 * If an open for the owner exists, or in the access/deny bits.
2948 	 * Otherwise it is a new open. If the open_owner hasn't been
2949 	 * confirmed, replace the open with the new one needing confirmation,
2950 	 * otherwise add the open.
2951 	 */
2952 	if (new_stp->ls_flags & NFSLCK_DELEGPREV) {
2953 	    /*
2954 	     * Handle NFSLCK_DELEGPREV by searching the old delegations for
2955 	     * a match. If found, just move the old delegation to the current
2956 	     * delegation list and issue open. If not found, return
2957 	     * NFSERR_EXPIRED.
2958 	     */
2959 	    LIST_FOREACH(stp, &clp->lc_olddeleg, ls_list) {
2960 		if (stp->ls_lfp == lfp) {
2961 		    /* Found it */
2962 		    if (stp->ls_clp != clp)
2963 			panic("olddeleg clp");
2964 		    LIST_REMOVE(stp, ls_list);
2965 		    LIST_REMOVE(stp, ls_hash);
2966 		    stp->ls_flags &= ~NFSLCK_OLDDELEG;
2967 		    stp->ls_stateid.seqid = delegstateidp->seqid = 1;
2968 		    stp->ls_stateid.other[0] = delegstateidp->other[0] =
2969 			clp->lc_clientid.lval[0];
2970 		    stp->ls_stateid.other[1] = delegstateidp->other[1] =
2971 			clp->lc_clientid.lval[1];
2972 		    stp->ls_stateid.other[2] = delegstateidp->other[2] =
2973 			nfsrv_nextstateindex(clp);
2974 		    stp->ls_compref = nd->nd_compref;
2975 		    LIST_INSERT_HEAD(&clp->lc_deleg, stp, ls_list);
2976 		    LIST_INSERT_HEAD(NFSSTATEHASH(clp,
2977 			stp->ls_stateid), stp, ls_hash);
2978 		    if (stp->ls_flags & NFSLCK_DELEGWRITE)
2979 			*rflagsp |= NFSV4OPEN_WRITEDELEGATE;
2980 		    else
2981 			*rflagsp |= NFSV4OPEN_READDELEGATE;
2982 		    clp->lc_delegtime = NFSD_MONOSEC +
2983 			nfsrv_lease + NFSRV_LEASEDELTA;
2984 
2985 		    /*
2986 		     * Now, do the associated open.
2987 		     */
2988 		    new_open->ls_stateid.seqid = 1;
2989 		    new_open->ls_stateid.other[0] = clp->lc_clientid.lval[0];
2990 		    new_open->ls_stateid.other[1] = clp->lc_clientid.lval[1];
2991 		    new_open->ls_stateid.other[2] = nfsrv_nextstateindex(clp);
2992 		    new_open->ls_flags = (new_stp->ls_flags&NFSLCK_DENYBITS)|
2993 			NFSLCK_OPEN;
2994 		    if (stp->ls_flags & NFSLCK_DELEGWRITE)
2995 			new_open->ls_flags |= (NFSLCK_READACCESS |
2996 			    NFSLCK_WRITEACCESS);
2997 		    else
2998 			new_open->ls_flags |= NFSLCK_READACCESS;
2999 		    new_open->ls_uid = new_stp->ls_uid;
3000 		    new_open->ls_lfp = lfp;
3001 		    new_open->ls_clp = clp;
3002 		    LIST_INIT(&new_open->ls_open);
3003 		    LIST_INSERT_HEAD(&lfp->lf_open, new_open, ls_file);
3004 		    LIST_INSERT_HEAD(NFSSTATEHASH(clp, new_open->ls_stateid),
3005 			new_open, ls_hash);
3006 		    /*
3007 		     * and handle the open owner
3008 		     */
3009 		    if (ownerstp) {
3010 			new_open->ls_openowner = ownerstp;
3011 			LIST_INSERT_HEAD(&ownerstp->ls_open,new_open,ls_list);
3012 		    } else {
3013 			new_open->ls_openowner = new_stp;
3014 			new_stp->ls_flags = 0;
3015 			nfsrvd_refcache(new_stp->ls_op);
3016 			new_stp->ls_noopens = 0;
3017 			LIST_INIT(&new_stp->ls_open);
3018 			LIST_INSERT_HEAD(&new_stp->ls_open, new_open, ls_list);
3019 			LIST_INSERT_HEAD(&clp->lc_open, new_stp, ls_list);
3020 			*new_stpp = NULL;
3021 			nfsstatsv1.srvopenowners++;
3022 			nfsrv_openpluslock++;
3023 		    }
3024 		    openstp = new_open;
3025 		    new_open = NULL;
3026 		    nfsstatsv1.srvopens++;
3027 		    nfsrv_openpluslock++;
3028 		    break;
3029 		}
3030 	    }
3031 	    if (stp == LIST_END(&clp->lc_olddeleg))
3032 		error = NFSERR_EXPIRED;
3033 	} else if (new_stp->ls_flags & (NFSLCK_DELEGREAD | NFSLCK_DELEGWRITE)) {
3034 	    /*
3035 	     * Scan to see that no delegation for this client and file
3036 	     * doesn't already exist.
3037 	     * There also shouldn't yet be an Open for this file and
3038 	     * openowner.
3039 	     */
3040 	    LIST_FOREACH(stp, &lfp->lf_deleg, ls_file) {
3041 		if (stp->ls_clp == clp)
3042 		    break;
3043 	    }
3044 	    if (stp == LIST_END(&lfp->lf_deleg) && openstp == NULL) {
3045 		/*
3046 		 * This is the Claim_Previous case with a delegation
3047 		 * type != Delegate_None.
3048 		 */
3049 		/*
3050 		 * First, add the delegation. (Although we must issue the
3051 		 * delegation, we can also ask for an immediate return.)
3052 		 */
3053 		new_deleg->ls_stateid.seqid = delegstateidp->seqid = 1;
3054 		new_deleg->ls_stateid.other[0] = delegstateidp->other[0] =
3055 		    clp->lc_clientid.lval[0];
3056 		new_deleg->ls_stateid.other[1] = delegstateidp->other[1] =
3057 		    clp->lc_clientid.lval[1];
3058 		new_deleg->ls_stateid.other[2] = delegstateidp->other[2] =
3059 		    nfsrv_nextstateindex(clp);
3060 		if (new_stp->ls_flags & NFSLCK_DELEGWRITE) {
3061 		    new_deleg->ls_flags = (NFSLCK_DELEGWRITE |
3062 			NFSLCK_READACCESS | NFSLCK_WRITEACCESS);
3063 		    *rflagsp |= NFSV4OPEN_WRITEDELEGATE;
3064 		    nfsrv_writedelegcnt++;
3065 		} else {
3066 		    new_deleg->ls_flags = (NFSLCK_DELEGREAD |
3067 			NFSLCK_READACCESS);
3068 		    *rflagsp |= NFSV4OPEN_READDELEGATE;
3069 		}
3070 		new_deleg->ls_uid = new_stp->ls_uid;
3071 		new_deleg->ls_lfp = lfp;
3072 		new_deleg->ls_clp = clp;
3073 		new_deleg->ls_filerev = filerev;
3074 		new_deleg->ls_compref = nd->nd_compref;
3075 		new_deleg->ls_lastrecall = 0;
3076 		LIST_INSERT_HEAD(&lfp->lf_deleg, new_deleg, ls_file);
3077 		LIST_INSERT_HEAD(NFSSTATEHASH(clp,
3078 		    new_deleg->ls_stateid), new_deleg, ls_hash);
3079 		LIST_INSERT_HEAD(&clp->lc_deleg, new_deleg, ls_list);
3080 		new_deleg = NULL;
3081 		if (delegate == 2 || nfsrv_issuedelegs == 0 ||
3082 		    (clp->lc_flags & (LCL_CALLBACKSON | LCL_CBDOWN)) !=
3083 		     LCL_CALLBACKSON ||
3084 		    NFSRV_V4DELEGLIMIT(nfsrv_delegatecnt) ||
3085 		    !NFSVNO_DELEGOK(vp))
3086 		    *rflagsp |= NFSV4OPEN_RECALL;
3087 		nfsstatsv1.srvdelegates++;
3088 		nfsrv_openpluslock++;
3089 		nfsrv_delegatecnt++;
3090 
3091 		/*
3092 		 * Now, do the associated open.
3093 		 */
3094 		new_open->ls_stateid.seqid = 1;
3095 		new_open->ls_stateid.other[0] = clp->lc_clientid.lval[0];
3096 		new_open->ls_stateid.other[1] = clp->lc_clientid.lval[1];
3097 		new_open->ls_stateid.other[2] = nfsrv_nextstateindex(clp);
3098 		new_open->ls_flags = (new_stp->ls_flags & NFSLCK_DENYBITS) |
3099 		    NFSLCK_OPEN;
3100 		if (new_stp->ls_flags & NFSLCK_DELEGWRITE)
3101 			new_open->ls_flags |= (NFSLCK_READACCESS |
3102 			    NFSLCK_WRITEACCESS);
3103 		else
3104 			new_open->ls_flags |= NFSLCK_READACCESS;
3105 		new_open->ls_uid = new_stp->ls_uid;
3106 		new_open->ls_lfp = lfp;
3107 		new_open->ls_clp = clp;
3108 		LIST_INIT(&new_open->ls_open);
3109 		LIST_INSERT_HEAD(&lfp->lf_open, new_open, ls_file);
3110 		LIST_INSERT_HEAD(NFSSTATEHASH(clp, new_open->ls_stateid),
3111 		   new_open, ls_hash);
3112 		/*
3113 		 * and handle the open owner
3114 		 */
3115 		if (ownerstp) {
3116 		    new_open->ls_openowner = ownerstp;
3117 		    LIST_INSERT_HEAD(&ownerstp->ls_open, new_open, ls_list);
3118 		} else {
3119 		    new_open->ls_openowner = new_stp;
3120 		    new_stp->ls_flags = 0;
3121 		    nfsrvd_refcache(new_stp->ls_op);
3122 		    new_stp->ls_noopens = 0;
3123 		    LIST_INIT(&new_stp->ls_open);
3124 		    LIST_INSERT_HEAD(&new_stp->ls_open, new_open, ls_list);
3125 		    LIST_INSERT_HEAD(&clp->lc_open, new_stp, ls_list);
3126 		    *new_stpp = NULL;
3127 		    nfsstatsv1.srvopenowners++;
3128 		    nfsrv_openpluslock++;
3129 		}
3130 		openstp = new_open;
3131 		new_open = NULL;
3132 		nfsstatsv1.srvopens++;
3133 		nfsrv_openpluslock++;
3134 	    } else {
3135 		error = NFSERR_RECLAIMCONFLICT;
3136 	    }
3137 	} else if (ownerstp) {
3138 		if (ownerstp->ls_flags & NFSLCK_NEEDSCONFIRM) {
3139 		    /* Replace the open */
3140 		    if (ownerstp->ls_op)
3141 			nfsrvd_derefcache(ownerstp->ls_op);
3142 		    ownerstp->ls_op = new_stp->ls_op;
3143 		    nfsrvd_refcache(ownerstp->ls_op);
3144 		    ownerstp->ls_seq = new_stp->ls_seq;
3145 		    *rflagsp |= NFSV4OPEN_RESULTCONFIRM;
3146 		    stp = LIST_FIRST(&ownerstp->ls_open);
3147 		    stp->ls_flags = (new_stp->ls_flags & NFSLCK_SHAREBITS) |
3148 			NFSLCK_OPEN;
3149 		    stp->ls_stateid.seqid = 1;
3150 		    stp->ls_uid = new_stp->ls_uid;
3151 		    if (lfp != stp->ls_lfp) {
3152 			LIST_REMOVE(stp, ls_file);
3153 			LIST_INSERT_HEAD(&lfp->lf_open, stp, ls_file);
3154 			stp->ls_lfp = lfp;
3155 		    }
3156 		    openstp = stp;
3157 		} else if (openstp) {
3158 		    openstp->ls_flags |= (new_stp->ls_flags & NFSLCK_SHAREBITS);
3159 		    openstp->ls_stateid.seqid++;
3160 		    if ((nd->nd_flag & ND_NFSV41) != 0 &&
3161 			openstp->ls_stateid.seqid == 0)
3162 			openstp->ls_stateid.seqid = 1;
3163 
3164 		    /*
3165 		     * This is where we can choose to issue a delegation.
3166 		     */
3167 		    if ((new_stp->ls_flags & NFSLCK_WANTNODELEG) != 0)
3168 			*rflagsp |= NFSV4OPEN_WDNOTWANTED;
3169 		    else if (nfsrv_issuedelegs == 0)
3170 			*rflagsp |= NFSV4OPEN_WDSUPPFTYPE;
3171 		    else if (NFSRV_V4DELEGLIMIT(nfsrv_delegatecnt))
3172 			*rflagsp |= NFSV4OPEN_WDRESOURCE;
3173 		    else if (delegate == 0 || writedeleg == 0 ||
3174 			NFSVNO_EXRDONLY(exp) || (readonly != 0 &&
3175 			nfsrv_writedelegifpos == 0) ||
3176 			!NFSVNO_DELEGOK(vp) ||
3177 			(new_stp->ls_flags & NFSLCK_WANTRDELEG) != 0 ||
3178 			(clp->lc_flags & (LCL_CALLBACKSON | LCL_CBDOWN)) !=
3179 			 LCL_CALLBACKSON)
3180 			*rflagsp |= NFSV4OPEN_WDCONTENTION;
3181 		    else {
3182 			new_deleg->ls_stateid.seqid = delegstateidp->seqid = 1;
3183 			new_deleg->ls_stateid.other[0] = delegstateidp->other[0]
3184 			    = clp->lc_clientid.lval[0];
3185 			new_deleg->ls_stateid.other[1] = delegstateidp->other[1]
3186 			    = clp->lc_clientid.lval[1];
3187 			new_deleg->ls_stateid.other[2] = delegstateidp->other[2]
3188 			    = nfsrv_nextstateindex(clp);
3189 			new_deleg->ls_flags = (NFSLCK_DELEGWRITE |
3190 			    NFSLCK_READACCESS | NFSLCK_WRITEACCESS);
3191 			*rflagsp |= NFSV4OPEN_WRITEDELEGATE;
3192 			new_deleg->ls_uid = new_stp->ls_uid;
3193 			new_deleg->ls_lfp = lfp;
3194 			new_deleg->ls_clp = clp;
3195 			new_deleg->ls_filerev = filerev;
3196 			new_deleg->ls_compref = nd->nd_compref;
3197 			new_deleg->ls_lastrecall = 0;
3198 			nfsrv_writedelegcnt++;
3199 			LIST_INSERT_HEAD(&lfp->lf_deleg, new_deleg, ls_file);
3200 			LIST_INSERT_HEAD(NFSSTATEHASH(clp,
3201 			    new_deleg->ls_stateid), new_deleg, ls_hash);
3202 			LIST_INSERT_HEAD(&clp->lc_deleg, new_deleg, ls_list);
3203 			new_deleg = NULL;
3204 			nfsstatsv1.srvdelegates++;
3205 			nfsrv_openpluslock++;
3206 			nfsrv_delegatecnt++;
3207 		    }
3208 		} else {
3209 		    new_open->ls_stateid.seqid = 1;
3210 		    new_open->ls_stateid.other[0] = clp->lc_clientid.lval[0];
3211 		    new_open->ls_stateid.other[1] = clp->lc_clientid.lval[1];
3212 		    new_open->ls_stateid.other[2] = nfsrv_nextstateindex(clp);
3213 		    new_open->ls_flags = (new_stp->ls_flags & NFSLCK_SHAREBITS)|
3214 			NFSLCK_OPEN;
3215 		    new_open->ls_uid = new_stp->ls_uid;
3216 		    new_open->ls_openowner = ownerstp;
3217 		    new_open->ls_lfp = lfp;
3218 		    new_open->ls_clp = clp;
3219 		    LIST_INIT(&new_open->ls_open);
3220 		    LIST_INSERT_HEAD(&lfp->lf_open, new_open, ls_file);
3221 		    LIST_INSERT_HEAD(&ownerstp->ls_open, new_open, ls_list);
3222 		    LIST_INSERT_HEAD(NFSSTATEHASH(clp, new_open->ls_stateid),
3223 			new_open, ls_hash);
3224 		    openstp = new_open;
3225 		    new_open = NULL;
3226 		    nfsstatsv1.srvopens++;
3227 		    nfsrv_openpluslock++;
3228 
3229 		    /*
3230 		     * This is where we can choose to issue a delegation.
3231 		     */
3232 		    if ((new_stp->ls_flags & NFSLCK_WANTNODELEG) != 0)
3233 			*rflagsp |= NFSV4OPEN_WDNOTWANTED;
3234 		    else if (nfsrv_issuedelegs == 0)
3235 			*rflagsp |= NFSV4OPEN_WDSUPPFTYPE;
3236 		    else if (NFSRV_V4DELEGLIMIT(nfsrv_delegatecnt))
3237 			*rflagsp |= NFSV4OPEN_WDRESOURCE;
3238 		    else if (delegate == 0 || (writedeleg == 0 &&
3239 			readonly == 0) || !NFSVNO_DELEGOK(vp) ||
3240 			(clp->lc_flags & (LCL_CALLBACKSON | LCL_CBDOWN)) !=
3241 			 LCL_CALLBACKSON)
3242 			*rflagsp |= NFSV4OPEN_WDCONTENTION;
3243 		    else {
3244 			new_deleg->ls_stateid.seqid = delegstateidp->seqid = 1;
3245 			new_deleg->ls_stateid.other[0] = delegstateidp->other[0]
3246 			    = clp->lc_clientid.lval[0];
3247 			new_deleg->ls_stateid.other[1] = delegstateidp->other[1]
3248 			    = clp->lc_clientid.lval[1];
3249 			new_deleg->ls_stateid.other[2] = delegstateidp->other[2]
3250 			    = nfsrv_nextstateindex(clp);
3251 			if (writedeleg && !NFSVNO_EXRDONLY(exp) &&
3252 			    (nfsrv_writedelegifpos || !readonly) &&
3253 			    (new_stp->ls_flags & NFSLCK_WANTRDELEG) == 0) {
3254 			    new_deleg->ls_flags = (NFSLCK_DELEGWRITE |
3255 				NFSLCK_READACCESS | NFSLCK_WRITEACCESS);
3256 			    *rflagsp |= NFSV4OPEN_WRITEDELEGATE;
3257 			    nfsrv_writedelegcnt++;
3258 			} else {
3259 			    new_deleg->ls_flags = (NFSLCK_DELEGREAD |
3260 				NFSLCK_READACCESS);
3261 			    *rflagsp |= NFSV4OPEN_READDELEGATE;
3262 			}
3263 			new_deleg->ls_uid = new_stp->ls_uid;
3264 			new_deleg->ls_lfp = lfp;
3265 			new_deleg->ls_clp = clp;
3266 			new_deleg->ls_filerev = filerev;
3267 			new_deleg->ls_compref = nd->nd_compref;
3268 			new_deleg->ls_lastrecall = 0;
3269 			LIST_INSERT_HEAD(&lfp->lf_deleg, new_deleg, ls_file);
3270 			LIST_INSERT_HEAD(NFSSTATEHASH(clp,
3271 			    new_deleg->ls_stateid), new_deleg, ls_hash);
3272 			LIST_INSERT_HEAD(&clp->lc_deleg, new_deleg, ls_list);
3273 			new_deleg = NULL;
3274 			nfsstatsv1.srvdelegates++;
3275 			nfsrv_openpluslock++;
3276 			nfsrv_delegatecnt++;
3277 		    }
3278 		}
3279 	} else {
3280 		/*
3281 		 * New owner case. Start the open_owner sequence with a
3282 		 * Needs confirmation (unless a reclaim) and hang the
3283 		 * new open off it.
3284 		 */
3285 		new_open->ls_stateid.seqid = 1;
3286 		new_open->ls_stateid.other[0] = clp->lc_clientid.lval[0];
3287 		new_open->ls_stateid.other[1] = clp->lc_clientid.lval[1];
3288 		new_open->ls_stateid.other[2] = nfsrv_nextstateindex(clp);
3289 		new_open->ls_flags = (new_stp->ls_flags & NFSLCK_SHAREBITS) |
3290 		    NFSLCK_OPEN;
3291 		new_open->ls_uid = new_stp->ls_uid;
3292 		LIST_INIT(&new_open->ls_open);
3293 		new_open->ls_openowner = new_stp;
3294 		new_open->ls_lfp = lfp;
3295 		new_open->ls_clp = clp;
3296 		LIST_INSERT_HEAD(&lfp->lf_open, new_open, ls_file);
3297 		if (new_stp->ls_flags & NFSLCK_RECLAIM) {
3298 			new_stp->ls_flags = 0;
3299 		} else if ((nd->nd_flag & ND_NFSV41) != 0) {
3300 			/* NFSv4.1 never needs confirmation. */
3301 			new_stp->ls_flags = 0;
3302 
3303 			/*
3304 			 * This is where we can choose to issue a delegation.
3305 			 */
3306 			if (delegate && nfsrv_issuedelegs &&
3307 			    (writedeleg || readonly) &&
3308 			    (clp->lc_flags & (LCL_CALLBACKSON | LCL_CBDOWN)) ==
3309 			     LCL_CALLBACKSON &&
3310 			    !NFSRV_V4DELEGLIMIT(nfsrv_delegatecnt) &&
3311 			    NFSVNO_DELEGOK(vp) &&
3312 			    ((nd->nd_flag & ND_NFSV41) == 0 ||
3313 			     (new_stp->ls_flags & NFSLCK_WANTNODELEG) == 0)) {
3314 				new_deleg->ls_stateid.seqid =
3315 				    delegstateidp->seqid = 1;
3316 				new_deleg->ls_stateid.other[0] =
3317 				    delegstateidp->other[0]
3318 				    = clp->lc_clientid.lval[0];
3319 				new_deleg->ls_stateid.other[1] =
3320 				    delegstateidp->other[1]
3321 				    = clp->lc_clientid.lval[1];
3322 				new_deleg->ls_stateid.other[2] =
3323 				    delegstateidp->other[2]
3324 				    = nfsrv_nextstateindex(clp);
3325 				if (writedeleg && !NFSVNO_EXRDONLY(exp) &&
3326 				    (nfsrv_writedelegifpos || !readonly) &&
3327 				    ((nd->nd_flag & ND_NFSV41) == 0 ||
3328 				     (new_stp->ls_flags & NFSLCK_WANTRDELEG) ==
3329 				     0)) {
3330 					new_deleg->ls_flags =
3331 					    (NFSLCK_DELEGWRITE |
3332 					     NFSLCK_READACCESS |
3333 					     NFSLCK_WRITEACCESS);
3334 					*rflagsp |= NFSV4OPEN_WRITEDELEGATE;
3335 					nfsrv_writedelegcnt++;
3336 				} else {
3337 					new_deleg->ls_flags =
3338 					    (NFSLCK_DELEGREAD |
3339 					     NFSLCK_READACCESS);
3340 					*rflagsp |= NFSV4OPEN_READDELEGATE;
3341 				}
3342 				new_deleg->ls_uid = new_stp->ls_uid;
3343 				new_deleg->ls_lfp = lfp;
3344 				new_deleg->ls_clp = clp;
3345 				new_deleg->ls_filerev = filerev;
3346 				new_deleg->ls_compref = nd->nd_compref;
3347 				new_deleg->ls_lastrecall = 0;
3348 				LIST_INSERT_HEAD(&lfp->lf_deleg, new_deleg,
3349 				    ls_file);
3350 				LIST_INSERT_HEAD(NFSSTATEHASH(clp,
3351 				    new_deleg->ls_stateid), new_deleg, ls_hash);
3352 				LIST_INSERT_HEAD(&clp->lc_deleg, new_deleg,
3353 				    ls_list);
3354 				new_deleg = NULL;
3355 				nfsstatsv1.srvdelegates++;
3356 				nfsrv_openpluslock++;
3357 				nfsrv_delegatecnt++;
3358 			}
3359 			/*
3360 			 * Since NFSv4.1 never does an OpenConfirm, the first
3361 			 * open state will be acquired here.
3362 			 */
3363 			if (!(clp->lc_flags & LCL_STAMPEDSTABLE)) {
3364 				clp->lc_flags |= LCL_STAMPEDSTABLE;
3365 				len = clp->lc_idlen;
3366 				NFSBCOPY(clp->lc_id, clidp, len);
3367 				gotstate = 1;
3368 			}
3369 		} else {
3370 			*rflagsp |= NFSV4OPEN_RESULTCONFIRM;
3371 			new_stp->ls_flags = NFSLCK_NEEDSCONFIRM;
3372 		}
3373 		nfsrvd_refcache(new_stp->ls_op);
3374 		new_stp->ls_noopens = 0;
3375 		LIST_INIT(&new_stp->ls_open);
3376 		LIST_INSERT_HEAD(&new_stp->ls_open, new_open, ls_list);
3377 		LIST_INSERT_HEAD(&clp->lc_open, new_stp, ls_list);
3378 		LIST_INSERT_HEAD(NFSSTATEHASH(clp, new_open->ls_stateid),
3379 		    new_open, ls_hash);
3380 		openstp = new_open;
3381 		new_open = NULL;
3382 		*new_stpp = NULL;
3383 		nfsstatsv1.srvopens++;
3384 		nfsrv_openpluslock++;
3385 		nfsstatsv1.srvopenowners++;
3386 		nfsrv_openpluslock++;
3387 	}
3388 	if (!error) {
3389 		stateidp->seqid = openstp->ls_stateid.seqid;
3390 		stateidp->other[0] = openstp->ls_stateid.other[0];
3391 		stateidp->other[1] = openstp->ls_stateid.other[1];
3392 		stateidp->other[2] = openstp->ls_stateid.other[2];
3393 	}
3394 	NFSUNLOCKSTATE();
3395 	if (haslock) {
3396 		NFSLOCKV4ROOTMUTEX();
3397 		nfsv4_unlock(&nfsv4rootfs_lock, 1);
3398 		NFSUNLOCKV4ROOTMUTEX();
3399 	}
3400 	if (new_open)
3401 		free(new_open, M_NFSDSTATE);
3402 	if (new_deleg)
3403 		free(new_deleg, M_NFSDSTATE);
3404 
3405 	/*
3406 	 * If the NFSv4.1 client just acquired its first open, write a timestamp
3407 	 * to the stable storage file.
3408 	 */
3409 	if (gotstate != 0) {
3410 		nfsrv_writestable(clidp, len, NFSNST_NEWSTATE, p);
3411 		nfsrv_backupstable();
3412 	}
3413 
3414 out:
3415 	free(clidp, M_TEMP);
3416 	NFSEXITCODE2(error, nd);
3417 	return (error);
3418 }
3419 
3420 /*
3421  * Open update. Does the confirm, downgrade and close.
3422  */
3423 int
3424 nfsrv_openupdate(vnode_t vp, struct nfsstate *new_stp, nfsquad_t clientid,
3425     nfsv4stateid_t *stateidp, struct nfsrv_descript *nd, NFSPROC_T *p,
3426     int *retwriteaccessp)
3427 {
3428 	struct nfsstate *stp;
3429 	struct nfsclient *clp;
3430 	struct nfslockfile *lfp;
3431 	u_int32_t bits;
3432 	int error = 0, gotstate = 0, len = 0;
3433 	u_char *clidp = NULL;
3434 
3435 	/*
3436 	 * Check for restart conditions (client and server).
3437 	 */
3438 	error = nfsrv_checkrestart(clientid, new_stp->ls_flags,
3439 	    &new_stp->ls_stateid, 0);
3440 	if (error)
3441 		goto out;
3442 
3443 	clidp = malloc(NFSV4_OPAQUELIMIT, M_TEMP, M_WAITOK);
3444 	NFSLOCKSTATE();
3445 	/*
3446 	 * Get the open structure via clientid and stateid.
3447 	 */
3448 	error = nfsrv_getclient(clientid, CLOPS_RENEW, &clp, NULL,
3449 	    (nfsquad_t)((u_quad_t)0), 0, nd, p);
3450 	if (!error)
3451 		error = nfsrv_getstate(clp, &new_stp->ls_stateid,
3452 		    new_stp->ls_flags, &stp);
3453 
3454 	/*
3455 	 * Sanity check the open.
3456 	 */
3457 	if (!error && (!(stp->ls_flags & NFSLCK_OPEN) ||
3458 		(!(new_stp->ls_flags & NFSLCK_CONFIRM) &&
3459 		 (stp->ls_openowner->ls_flags & NFSLCK_NEEDSCONFIRM)) ||
3460 		((new_stp->ls_flags & NFSLCK_CONFIRM) &&
3461 		 (!(stp->ls_openowner->ls_flags & NFSLCK_NEEDSCONFIRM)))))
3462 		error = NFSERR_BADSTATEID;
3463 
3464 	if (!error)
3465 		error = nfsrv_checkseqid(nd, new_stp->ls_seq,
3466 		    stp->ls_openowner, new_stp->ls_op);
3467 	if (!error && stp->ls_stateid.seqid != new_stp->ls_stateid.seqid &&
3468 	    (((nd->nd_flag & ND_NFSV41) == 0 &&
3469 	      !(new_stp->ls_flags & NFSLCK_CONFIRM)) ||
3470 	     ((nd->nd_flag & ND_NFSV41) != 0 &&
3471 	      new_stp->ls_stateid.seqid != 0)))
3472 		error = NFSERR_OLDSTATEID;
3473 	if (!error && vnode_vtype(vp) != VREG) {
3474 		if (vnode_vtype(vp) == VDIR)
3475 			error = NFSERR_ISDIR;
3476 		else
3477 			error = NFSERR_INVAL;
3478 	}
3479 
3480 	if (error) {
3481 		/*
3482 		 * If a client tries to confirm an Open with a bad
3483 		 * seqid# and there are no byte range locks or other Opens
3484 		 * on the openowner, just throw it away, so the next use of the
3485 		 * openowner will start a fresh seq#.
3486 		 */
3487 		if (error == NFSERR_BADSEQID &&
3488 		    (new_stp->ls_flags & NFSLCK_CONFIRM) &&
3489 		    nfsrv_nootherstate(stp))
3490 			nfsrv_freeopenowner(stp->ls_openowner, 0, p);
3491 		NFSUNLOCKSTATE();
3492 		goto out;
3493 	}
3494 
3495 	/*
3496 	 * Set the return stateid.
3497 	 */
3498 	stateidp->seqid = stp->ls_stateid.seqid + 1;
3499 	if ((nd->nd_flag & ND_NFSV41) != 0 && stateidp->seqid == 0)
3500 		stateidp->seqid = 1;
3501 	stateidp->other[0] = stp->ls_stateid.other[0];
3502 	stateidp->other[1] = stp->ls_stateid.other[1];
3503 	stateidp->other[2] = stp->ls_stateid.other[2];
3504 	/*
3505 	 * Now, handle the three cases.
3506 	 */
3507 	if (new_stp->ls_flags & NFSLCK_CONFIRM) {
3508 		/*
3509 		 * If the open doesn't need confirmation, it seems to me that
3510 		 * there is a client error, but I'll just log it and keep going?
3511 		 */
3512 		if (!(stp->ls_openowner->ls_flags & NFSLCK_NEEDSCONFIRM))
3513 			printf("Nfsv4d: stray open confirm\n");
3514 		stp->ls_openowner->ls_flags = 0;
3515 		stp->ls_stateid.seqid++;
3516 		if ((nd->nd_flag & ND_NFSV41) != 0 &&
3517 		    stp->ls_stateid.seqid == 0)
3518 			stp->ls_stateid.seqid = 1;
3519 		if (!(clp->lc_flags & LCL_STAMPEDSTABLE)) {
3520 			clp->lc_flags |= LCL_STAMPEDSTABLE;
3521 			len = clp->lc_idlen;
3522 			NFSBCOPY(clp->lc_id, clidp, len);
3523 			gotstate = 1;
3524 		}
3525 		NFSUNLOCKSTATE();
3526 	} else if (new_stp->ls_flags & NFSLCK_CLOSE) {
3527 		lfp = stp->ls_lfp;
3528 		if (retwriteaccessp != NULL) {
3529 			if ((stp->ls_flags & NFSLCK_WRITEACCESS) != 0)
3530 				*retwriteaccessp = 1;
3531 			else
3532 				*retwriteaccessp = 0;
3533 		}
3534 		if (nfsrv_dolocallocks != 0 && !LIST_EMPTY(&stp->ls_open)) {
3535 			/* Get the lf lock */
3536 			nfsrv_locklf(lfp);
3537 			NFSUNLOCKSTATE();
3538 			ASSERT_VOP_ELOCKED(vp, "nfsrv_openupdate");
3539 			NFSVOPUNLOCK(vp);
3540 			if (nfsrv_freeopen(stp, vp, 1, p) == 0) {
3541 				NFSLOCKSTATE();
3542 				nfsrv_unlocklf(lfp);
3543 				NFSUNLOCKSTATE();
3544 			}
3545 			NFSVOPLOCK(vp, LK_EXCLUSIVE | LK_RETRY);
3546 		} else {
3547 			(void) nfsrv_freeopen(stp, NULL, 0, p);
3548 			NFSUNLOCKSTATE();
3549 		}
3550 	} else {
3551 		/*
3552 		 * Update the share bits, making sure that the new set are a
3553 		 * subset of the old ones.
3554 		 */
3555 		bits = (new_stp->ls_flags & NFSLCK_SHAREBITS);
3556 		if (~(stp->ls_flags) & bits) {
3557 			NFSUNLOCKSTATE();
3558 			error = NFSERR_INVAL;
3559 			goto out;
3560 		}
3561 		stp->ls_flags = (bits | NFSLCK_OPEN);
3562 		stp->ls_stateid.seqid++;
3563 		if ((nd->nd_flag & ND_NFSV41) != 0 &&
3564 		    stp->ls_stateid.seqid == 0)
3565 			stp->ls_stateid.seqid = 1;
3566 		NFSUNLOCKSTATE();
3567 	}
3568 
3569 	/*
3570 	 * If the client just confirmed its first open, write a timestamp
3571 	 * to the stable storage file.
3572 	 */
3573 	if (gotstate != 0) {
3574 		nfsrv_writestable(clidp, len, NFSNST_NEWSTATE, p);
3575 		nfsrv_backupstable();
3576 	}
3577 
3578 out:
3579 	free(clidp, M_TEMP);
3580 	NFSEXITCODE2(error, nd);
3581 	return (error);
3582 }
3583 
3584 /*
3585  * Delegation update. Does the purge and return.
3586  */
3587 int
3588 nfsrv_delegupdate(struct nfsrv_descript *nd, nfsquad_t clientid,
3589     nfsv4stateid_t *stateidp, vnode_t vp, int op, struct ucred *cred,
3590     NFSPROC_T *p, int *retwriteaccessp)
3591 {
3592 	struct nfsstate *stp;
3593 	struct nfsclient *clp;
3594 	int error = 0;
3595 	fhandle_t fh;
3596 
3597 	/*
3598 	 * Do a sanity check against the file handle for DelegReturn.
3599 	 */
3600 	if (vp) {
3601 		error = nfsvno_getfh(vp, &fh, p);
3602 		if (error)
3603 			goto out;
3604 	}
3605 	/*
3606 	 * Check for restart conditions (client and server).
3607 	 */
3608 	if (op == NFSV4OP_DELEGRETURN)
3609 		error = nfsrv_checkrestart(clientid, NFSLCK_DELEGRETURN,
3610 			stateidp, 0);
3611 	else
3612 		error = nfsrv_checkrestart(clientid, NFSLCK_DELEGPURGE,
3613 			stateidp, 0);
3614 
3615 	NFSLOCKSTATE();
3616 	/*
3617 	 * Get the open structure via clientid and stateid.
3618 	 */
3619 	if (!error)
3620 	    error = nfsrv_getclient(clientid, CLOPS_RENEW, &clp, NULL,
3621 		(nfsquad_t)((u_quad_t)0), 0, nd, p);
3622 	if (error) {
3623 		if (error == NFSERR_CBPATHDOWN)
3624 			error = 0;
3625 		if (error == NFSERR_STALECLIENTID && op == NFSV4OP_DELEGRETURN)
3626 			error = NFSERR_STALESTATEID;
3627 	}
3628 	if (!error && op == NFSV4OP_DELEGRETURN) {
3629 	    error = nfsrv_getstate(clp, stateidp, NFSLCK_DELEGRETURN, &stp);
3630 	    if (!error && stp->ls_stateid.seqid != stateidp->seqid &&
3631 		((nd->nd_flag & ND_NFSV41) == 0 || stateidp->seqid != 0))
3632 		error = NFSERR_OLDSTATEID;
3633 	}
3634 	/*
3635 	 * NFSERR_EXPIRED means that the state has gone away,
3636 	 * so Delegations have been purged. Just return ok.
3637 	 */
3638 	if (error == NFSERR_EXPIRED && op == NFSV4OP_DELEGPURGE) {
3639 		NFSUNLOCKSTATE();
3640 		error = 0;
3641 		goto out;
3642 	}
3643 	if (error) {
3644 		NFSUNLOCKSTATE();
3645 		goto out;
3646 	}
3647 
3648 	if (op == NFSV4OP_DELEGRETURN) {
3649 		if (NFSBCMP((caddr_t)&fh, (caddr_t)&stp->ls_lfp->lf_fh,
3650 		    sizeof (fhandle_t))) {
3651 			NFSUNLOCKSTATE();
3652 			error = NFSERR_BADSTATEID;
3653 			goto out;
3654 		}
3655 		if (retwriteaccessp != NULL) {
3656 			if ((stp->ls_flags & NFSLCK_DELEGWRITE) != 0)
3657 				*retwriteaccessp = 1;
3658 			else
3659 				*retwriteaccessp = 0;
3660 		}
3661 		nfsrv_freedeleg(stp);
3662 	} else {
3663 		nfsrv_freedeleglist(&clp->lc_olddeleg);
3664 	}
3665 	NFSUNLOCKSTATE();
3666 	error = 0;
3667 
3668 out:
3669 	NFSEXITCODE(error);
3670 	return (error);
3671 }
3672 
3673 /*
3674  * Release lock owner.
3675  */
3676 int
3677 nfsrv_releaselckown(struct nfsstate *new_stp, nfsquad_t clientid,
3678     NFSPROC_T *p)
3679 {
3680 	struct nfsstate *stp, *nstp, *openstp, *ownstp;
3681 	struct nfsclient *clp;
3682 	int error = 0;
3683 
3684 	/*
3685 	 * Check for restart conditions (client and server).
3686 	 */
3687 	error = nfsrv_checkrestart(clientid, new_stp->ls_flags,
3688 	    &new_stp->ls_stateid, 0);
3689 	if (error)
3690 		goto out;
3691 
3692 	NFSLOCKSTATE();
3693 	/*
3694 	 * Get the lock owner by name.
3695 	 */
3696 	error = nfsrv_getclient(clientid, CLOPS_RENEW, &clp, NULL,
3697 	    (nfsquad_t)((u_quad_t)0), 0, NULL, p);
3698 	if (error) {
3699 		NFSUNLOCKSTATE();
3700 		goto out;
3701 	}
3702 	LIST_FOREACH(ownstp, &clp->lc_open, ls_list) {
3703 	    LIST_FOREACH(openstp, &ownstp->ls_open, ls_list) {
3704 		stp = LIST_FIRST(&openstp->ls_open);
3705 		while (stp != LIST_END(&openstp->ls_open)) {
3706 		    nstp = LIST_NEXT(stp, ls_list);
3707 		    /*
3708 		     * If the owner matches, check for locks and
3709 		     * then free or return an error.
3710 		     */
3711 		    if (stp->ls_ownerlen == new_stp->ls_ownerlen &&
3712 			!NFSBCMP(stp->ls_owner, new_stp->ls_owner,
3713 			 stp->ls_ownerlen)){
3714 			if (LIST_EMPTY(&stp->ls_lock)) {
3715 			    nfsrv_freelockowner(stp, NULL, 0, p);
3716 			} else {
3717 			    NFSUNLOCKSTATE();
3718 			    error = NFSERR_LOCKSHELD;
3719 			    goto out;
3720 			}
3721 		    }
3722 		    stp = nstp;
3723 		}
3724 	    }
3725 	}
3726 	NFSUNLOCKSTATE();
3727 
3728 out:
3729 	NFSEXITCODE(error);
3730 	return (error);
3731 }
3732 
3733 /*
3734  * Get the file handle for a lock structure.
3735  */
3736 static int
3737 nfsrv_getlockfh(vnode_t vp, u_short flags, struct nfslockfile *new_lfp,
3738     fhandle_t *nfhp, NFSPROC_T *p)
3739 {
3740 	fhandle_t *fhp = NULL;
3741 	int error;
3742 
3743 	/*
3744 	 * For lock, use the new nfslock structure, otherwise just
3745 	 * a fhandle_t on the stack.
3746 	 */
3747 	if (flags & NFSLCK_OPEN) {
3748 		KASSERT(new_lfp != NULL, ("nfsrv_getlockfh: new_lfp NULL"));
3749 		fhp = &new_lfp->lf_fh;
3750 	} else if (nfhp) {
3751 		fhp = nfhp;
3752 	} else {
3753 		panic("nfsrv_getlockfh");
3754 	}
3755 	error = nfsvno_getfh(vp, fhp, p);
3756 	NFSEXITCODE(error);
3757 	return (error);
3758 }
3759 
3760 /*
3761  * Get an nfs lock structure. Allocate one, as required, and return a
3762  * pointer to it.
3763  * Returns an NFSERR_xxx upon failure or -1 to indicate no current lock.
3764  */
3765 static int
3766 nfsrv_getlockfile(u_short flags, struct nfslockfile **new_lfpp,
3767     struct nfslockfile **lfpp, fhandle_t *nfhp, int lockit)
3768 {
3769 	struct nfslockfile *lfp;
3770 	fhandle_t *fhp = NULL, *tfhp;
3771 	struct nfslockhashhead *hp;
3772 	struct nfslockfile *new_lfp = NULL;
3773 
3774 	/*
3775 	 * For lock, use the new nfslock structure, otherwise just
3776 	 * a fhandle_t on the stack.
3777 	 */
3778 	if (flags & NFSLCK_OPEN) {
3779 		new_lfp = *new_lfpp;
3780 		fhp = &new_lfp->lf_fh;
3781 	} else if (nfhp) {
3782 		fhp = nfhp;
3783 	} else {
3784 		panic("nfsrv_getlockfile");
3785 	}
3786 
3787 	hp = NFSLOCKHASH(fhp);
3788 	LIST_FOREACH(lfp, hp, lf_hash) {
3789 		tfhp = &lfp->lf_fh;
3790 		if (NFSVNO_CMPFH(fhp, tfhp)) {
3791 			if (lockit)
3792 				nfsrv_locklf(lfp);
3793 			*lfpp = lfp;
3794 			return (0);
3795 		}
3796 	}
3797 	if (!(flags & NFSLCK_OPEN))
3798 		return (-1);
3799 
3800 	/*
3801 	 * No match, so chain the new one into the list.
3802 	 */
3803 	LIST_INIT(&new_lfp->lf_open);
3804 	LIST_INIT(&new_lfp->lf_lock);
3805 	LIST_INIT(&new_lfp->lf_deleg);
3806 	LIST_INIT(&new_lfp->lf_locallock);
3807 	LIST_INIT(&new_lfp->lf_rollback);
3808 	new_lfp->lf_locallock_lck.nfslock_usecnt = 0;
3809 	new_lfp->lf_locallock_lck.nfslock_lock = 0;
3810 	new_lfp->lf_usecount = 0;
3811 	LIST_INSERT_HEAD(hp, new_lfp, lf_hash);
3812 	*lfpp = new_lfp;
3813 	*new_lfpp = NULL;
3814 	return (0);
3815 }
3816 
3817 /*
3818  * This function adds a nfslock lock structure to the list for the associated
3819  * nfsstate and nfslockfile structures. It will be inserted after the
3820  * entry pointed at by insert_lop.
3821  */
3822 static void
3823 nfsrv_insertlock(struct nfslock *new_lop, struct nfslock *insert_lop,
3824     struct nfsstate *stp, struct nfslockfile *lfp)
3825 {
3826 	struct nfslock *lop, *nlop;
3827 
3828 	new_lop->lo_stp = stp;
3829 	new_lop->lo_lfp = lfp;
3830 
3831 	if (stp != NULL) {
3832 		/* Insert in increasing lo_first order */
3833 		lop = LIST_FIRST(&lfp->lf_lock);
3834 		if (lop == LIST_END(&lfp->lf_lock) ||
3835 		    new_lop->lo_first <= lop->lo_first) {
3836 			LIST_INSERT_HEAD(&lfp->lf_lock, new_lop, lo_lckfile);
3837 		} else {
3838 			nlop = LIST_NEXT(lop, lo_lckfile);
3839 			while (nlop != LIST_END(&lfp->lf_lock) &&
3840 			       nlop->lo_first < new_lop->lo_first) {
3841 				lop = nlop;
3842 				nlop = LIST_NEXT(lop, lo_lckfile);
3843 			}
3844 			LIST_INSERT_AFTER(lop, new_lop, lo_lckfile);
3845 		}
3846 	} else {
3847 		new_lop->lo_lckfile.le_prev = NULL;	/* list not used */
3848 	}
3849 
3850 	/*
3851 	 * Insert after insert_lop, which is overloaded as stp or lfp for
3852 	 * an empty list.
3853 	 */
3854 	if (stp == NULL && (struct nfslockfile *)insert_lop == lfp)
3855 		LIST_INSERT_HEAD(&lfp->lf_locallock, new_lop, lo_lckowner);
3856 	else if ((struct nfsstate *)insert_lop == stp)
3857 		LIST_INSERT_HEAD(&stp->ls_lock, new_lop, lo_lckowner);
3858 	else
3859 		LIST_INSERT_AFTER(insert_lop, new_lop, lo_lckowner);
3860 	if (stp != NULL) {
3861 		nfsstatsv1.srvlocks++;
3862 		nfsrv_openpluslock++;
3863 	}
3864 }
3865 
3866 /*
3867  * This function updates the locking for a lock owner and given file. It
3868  * maintains a list of lock ranges ordered on increasing file offset that
3869  * are NFSLCK_READ or NFSLCK_WRITE and non-overlapping (aka POSIX style).
3870  * It always adds new_lop to the list and sometimes uses the one pointed
3871  * at by other_lopp.
3872  */
3873 static void
3874 nfsrv_updatelock(struct nfsstate *stp, struct nfslock **new_lopp,
3875     struct nfslock **other_lopp, struct nfslockfile *lfp)
3876 {
3877 	struct nfslock *new_lop = *new_lopp;
3878 	struct nfslock *lop, *tlop, *ilop;
3879 	struct nfslock *other_lop = *other_lopp;
3880 	int unlock = 0, myfile = 0;
3881 	u_int64_t tmp;
3882 
3883 	/*
3884 	 * Work down the list until the lock is merged.
3885 	 */
3886 	if (new_lop->lo_flags & NFSLCK_UNLOCK)
3887 		unlock = 1;
3888 	if (stp != NULL) {
3889 		ilop = (struct nfslock *)stp;
3890 		lop = LIST_FIRST(&stp->ls_lock);
3891 	} else {
3892 		ilop = (struct nfslock *)lfp;
3893 		lop = LIST_FIRST(&lfp->lf_locallock);
3894 	}
3895 	while (lop != NULL) {
3896 	    /*
3897 	     * Only check locks for this file that aren't before the start of
3898 	     * new lock's range.
3899 	     */
3900 	    if (lop->lo_lfp == lfp) {
3901 	      myfile = 1;
3902 	      if (lop->lo_end >= new_lop->lo_first) {
3903 		if (new_lop->lo_end < lop->lo_first) {
3904 			/*
3905 			 * If the new lock ends before the start of the
3906 			 * current lock's range, no merge, just insert
3907 			 * the new lock.
3908 			 */
3909 			break;
3910 		}
3911 		if (new_lop->lo_flags == lop->lo_flags ||
3912 		    (new_lop->lo_first <= lop->lo_first &&
3913 		     new_lop->lo_end >= lop->lo_end)) {
3914 			/*
3915 			 * This lock can be absorbed by the new lock/unlock.
3916 			 * This happens when it covers the entire range
3917 			 * of the old lock or is contiguous
3918 			 * with the old lock and is of the same type or an
3919 			 * unlock.
3920 			 */
3921 			if (lop->lo_first < new_lop->lo_first)
3922 				new_lop->lo_first = lop->lo_first;
3923 			if (lop->lo_end > new_lop->lo_end)
3924 				new_lop->lo_end = lop->lo_end;
3925 			tlop = lop;
3926 			lop = LIST_NEXT(lop, lo_lckowner);
3927 			nfsrv_freenfslock(tlop);
3928 			continue;
3929 		}
3930 
3931 		/*
3932 		 * All these cases are for contiguous locks that are not the
3933 		 * same type, so they can't be merged.
3934 		 */
3935 		if (new_lop->lo_first <= lop->lo_first) {
3936 			/*
3937 			 * This case is where the new lock overlaps with the
3938 			 * first part of the old lock. Move the start of the
3939 			 * old lock to just past the end of the new lock. The
3940 			 * new lock will be inserted in front of the old, since
3941 			 * ilop hasn't been updated. (We are done now.)
3942 			 */
3943 			lop->lo_first = new_lop->lo_end;
3944 			break;
3945 		}
3946 		if (new_lop->lo_end >= lop->lo_end) {
3947 			/*
3948 			 * This case is where the new lock overlaps with the
3949 			 * end of the old lock's range. Move the old lock's
3950 			 * end to just before the new lock's first and insert
3951 			 * the new lock after the old lock.
3952 			 * Might not be done yet, since the new lock could
3953 			 * overlap further locks with higher ranges.
3954 			 */
3955 			lop->lo_end = new_lop->lo_first;
3956 			ilop = lop;
3957 			lop = LIST_NEXT(lop, lo_lckowner);
3958 			continue;
3959 		}
3960 		/*
3961 		 * The final case is where the new lock's range is in the
3962 		 * middle of the current lock's and splits the current lock
3963 		 * up. Use *other_lopp to handle the second part of the
3964 		 * split old lock range. (We are done now.)
3965 		 * For unlock, we use new_lop as other_lop and tmp, since
3966 		 * other_lop and new_lop are the same for this case.
3967 		 * We noted the unlock case above, so we don't need
3968 		 * new_lop->lo_flags any longer.
3969 		 */
3970 		tmp = new_lop->lo_first;
3971 		if (other_lop == NULL) {
3972 			if (!unlock)
3973 				panic("nfsd srv update unlock");
3974 			other_lop = new_lop;
3975 			*new_lopp = NULL;
3976 		}
3977 		other_lop->lo_first = new_lop->lo_end;
3978 		other_lop->lo_end = lop->lo_end;
3979 		other_lop->lo_flags = lop->lo_flags;
3980 		other_lop->lo_stp = stp;
3981 		other_lop->lo_lfp = lfp;
3982 		lop->lo_end = tmp;
3983 		nfsrv_insertlock(other_lop, lop, stp, lfp);
3984 		*other_lopp = NULL;
3985 		ilop = lop;
3986 		break;
3987 	      }
3988 	    }
3989 	    ilop = lop;
3990 	    lop = LIST_NEXT(lop, lo_lckowner);
3991 	    if (myfile && (lop == NULL || lop->lo_lfp != lfp))
3992 		break;
3993 	}
3994 
3995 	/*
3996 	 * Insert the new lock in the list at the appropriate place.
3997 	 */
3998 	if (!unlock) {
3999 		nfsrv_insertlock(new_lop, ilop, stp, lfp);
4000 		*new_lopp = NULL;
4001 	}
4002 }
4003 
4004 /*
4005  * This function handles sequencing of locks, etc.
4006  * It returns an error that indicates what the caller should do.
4007  */
4008 static int
4009 nfsrv_checkseqid(struct nfsrv_descript *nd, u_int32_t seqid,
4010     struct nfsstate *stp, struct nfsrvcache *op)
4011 {
4012 	int error = 0;
4013 
4014 	if ((nd->nd_flag & ND_NFSV41) != 0)
4015 		/* NFSv4.1 ignores the open_seqid and lock_seqid. */
4016 		goto out;
4017 	if (op != nd->nd_rp)
4018 		panic("nfsrvstate checkseqid");
4019 	if (!(op->rc_flag & RC_INPROG))
4020 		panic("nfsrvstate not inprog");
4021 	if (stp->ls_op && stp->ls_op->rc_refcnt <= 0) {
4022 		printf("refcnt=%d\n", stp->ls_op->rc_refcnt);
4023 		panic("nfsrvstate op refcnt");
4024 	}
4025 
4026 	/* If ND_ERELOOKUP is set, the seqid has already been handled. */
4027 	if ((nd->nd_flag & ND_ERELOOKUP) != 0)
4028 		goto out;
4029 
4030 	if ((stp->ls_seq + 1) == seqid) {
4031 		if (stp->ls_op)
4032 			nfsrvd_derefcache(stp->ls_op);
4033 		stp->ls_op = op;
4034 		nfsrvd_refcache(op);
4035 		stp->ls_seq = seqid;
4036 		goto out;
4037 	} else if (stp->ls_seq == seqid && stp->ls_op &&
4038 		op->rc_xid == stp->ls_op->rc_xid &&
4039 		op->rc_refcnt == 0 &&
4040 		op->rc_reqlen == stp->ls_op->rc_reqlen &&
4041 		op->rc_cksum == stp->ls_op->rc_cksum) {
4042 		if (stp->ls_op->rc_flag & RC_INPROG) {
4043 			error = NFSERR_DONTREPLY;
4044 			goto out;
4045 		}
4046 		nd->nd_rp = stp->ls_op;
4047 		nd->nd_rp->rc_flag |= RC_INPROG;
4048 		nfsrvd_delcache(op);
4049 		error = NFSERR_REPLYFROMCACHE;
4050 		goto out;
4051 	}
4052 	error = NFSERR_BADSEQID;
4053 
4054 out:
4055 	NFSEXITCODE2(error, nd);
4056 	return (error);
4057 }
4058 
4059 /*
4060  * Get the client ip address for callbacks. If the strings can't be parsed,
4061  * just set lc_program to 0 to indicate no callbacks are possible.
4062  * (For cases where the address can't be parsed or is 0.0.0.0.0.0, set
4063  *  the address to the client's transport address. This won't be used
4064  *  for callbacks, but can be printed out by nfsstats for info.)
4065  * Return error if the xdr can't be parsed, 0 otherwise.
4066  */
4067 int
4068 nfsrv_getclientipaddr(struct nfsrv_descript *nd, struct nfsclient *clp)
4069 {
4070 	u_int32_t *tl;
4071 	u_char *cp, *cp2;
4072 	int i, j, maxalen = 0, minalen = 0;
4073 	sa_family_t af;
4074 #ifdef INET
4075 	struct sockaddr_in *rin = NULL, *sin;
4076 #endif
4077 #ifdef INET6
4078 	struct sockaddr_in6 *rin6 = NULL, *sin6;
4079 #endif
4080 	u_char *addr;
4081 	int error = 0, cantparse = 0;
4082 	union {
4083 		in_addr_t ival;
4084 		u_char cval[4];
4085 	} ip;
4086 	union {
4087 		in_port_t sval;
4088 		u_char cval[2];
4089 	} port;
4090 
4091 	/* 8 is the maximum length of the port# string. */
4092 	addr = malloc(INET6_ADDRSTRLEN + 8, M_TEMP, M_WAITOK);
4093 	clp->lc_req.nr_client = NULL;
4094 	clp->lc_req.nr_lock = 0;
4095 	af = AF_UNSPEC;
4096 	NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
4097 	i = fxdr_unsigned(int, *tl);
4098 	if (i >= 3 && i <= 4) {
4099 		error = nfsrv_mtostr(nd, addr, i);
4100 		if (error)
4101 			goto nfsmout;
4102 #ifdef INET
4103 		if (!strcmp(addr, "tcp")) {
4104 			clp->lc_flags |= LCL_TCPCALLBACK;
4105 			clp->lc_req.nr_sotype = SOCK_STREAM;
4106 			clp->lc_req.nr_soproto = IPPROTO_TCP;
4107 			af = AF_INET;
4108 		} else if (!strcmp(addr, "udp")) {
4109 			clp->lc_req.nr_sotype = SOCK_DGRAM;
4110 			clp->lc_req.nr_soproto = IPPROTO_UDP;
4111 			af = AF_INET;
4112 		}
4113 #endif
4114 #ifdef INET6
4115 		if (af == AF_UNSPEC) {
4116 			if (!strcmp(addr, "tcp6")) {
4117 				clp->lc_flags |= LCL_TCPCALLBACK;
4118 				clp->lc_req.nr_sotype = SOCK_STREAM;
4119 				clp->lc_req.nr_soproto = IPPROTO_TCP;
4120 				af = AF_INET6;
4121 			} else if (!strcmp(addr, "udp6")) {
4122 				clp->lc_req.nr_sotype = SOCK_DGRAM;
4123 				clp->lc_req.nr_soproto = IPPROTO_UDP;
4124 				af = AF_INET6;
4125 			}
4126 		}
4127 #endif
4128 		if (af == AF_UNSPEC) {
4129 			cantparse = 1;
4130 		}
4131 	} else {
4132 		cantparse = 1;
4133 		if (i > 0) {
4134 			error = nfsm_advance(nd, NFSM_RNDUP(i), -1);
4135 			if (error)
4136 				goto nfsmout;
4137 		}
4138 	}
4139 	/*
4140 	 * The caller has allocated clp->lc_req.nr_nam to be large enough
4141 	 * for either AF_INET or AF_INET6 and zeroed out the contents.
4142 	 * maxalen is set to the maximum length of the host IP address string
4143 	 * plus 8 for the maximum length of the port#.
4144 	 * minalen is set to the minimum length of the host IP address string
4145 	 * plus 4 for the minimum length of the port#.
4146 	 * These lengths do not include NULL termination,
4147 	 * so INET[6]_ADDRSTRLEN - 1 is used in the calculations.
4148 	 */
4149 	switch (af) {
4150 #ifdef INET
4151 	case AF_INET:
4152 		rin = (struct sockaddr_in *)clp->lc_req.nr_nam;
4153 		rin->sin_family = AF_INET;
4154 		rin->sin_len = sizeof(struct sockaddr_in);
4155 		maxalen = INET_ADDRSTRLEN - 1 + 8;
4156 		minalen = 7 + 4;
4157 		break;
4158 #endif
4159 #ifdef INET6
4160 	case AF_INET6:
4161 		rin6 = (struct sockaddr_in6 *)clp->lc_req.nr_nam;
4162 		rin6->sin6_family = AF_INET6;
4163 		rin6->sin6_len = sizeof(struct sockaddr_in6);
4164 		maxalen = INET6_ADDRSTRLEN - 1 + 8;
4165 		minalen = 3 + 4;
4166 		break;
4167 #endif
4168 	}
4169 	NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
4170 	i = fxdr_unsigned(int, *tl);
4171 	if (i < 0) {
4172 		error = NFSERR_BADXDR;
4173 		goto nfsmout;
4174 	} else if (i == 0) {
4175 		cantparse = 1;
4176 	} else if (!cantparse && i <= maxalen && i >= minalen) {
4177 		error = nfsrv_mtostr(nd, addr, i);
4178 		if (error)
4179 			goto nfsmout;
4180 
4181 		/*
4182 		 * Parse out the address fields. We expect 6 decimal numbers
4183 		 * separated by '.'s for AF_INET and two decimal numbers
4184 		 * preceeded by '.'s for AF_INET6.
4185 		 */
4186 		cp = NULL;
4187 		switch (af) {
4188 #ifdef INET6
4189 		/*
4190 		 * For AF_INET6, first parse the host address.
4191 		 */
4192 		case AF_INET6:
4193 			cp = strchr(addr, '.');
4194 			if (cp != NULL) {
4195 				*cp++ = '\0';
4196 				if (inet_pton(af, addr, &rin6->sin6_addr) == 1)
4197 					i = 4;
4198 				else {
4199 					cp = NULL;
4200 					cantparse = 1;
4201 				}
4202 			}
4203 			break;
4204 #endif
4205 #ifdef INET
4206 		case AF_INET:
4207 			cp = addr;
4208 			i = 0;
4209 			break;
4210 #endif
4211 		}
4212 		while (cp != NULL && *cp && i < 6) {
4213 			cp2 = cp;
4214 			while (*cp2 && *cp2 != '.')
4215 				cp2++;
4216 			if (*cp2)
4217 				*cp2++ = '\0';
4218 			else if (i != 5) {
4219 				cantparse = 1;
4220 				break;
4221 			}
4222 			j = nfsrv_getipnumber(cp);
4223 			if (j >= 0) {
4224 				if (i < 4)
4225 					ip.cval[3 - i] = j;
4226 				else
4227 					port.cval[5 - i] = j;
4228 			} else {
4229 				cantparse = 1;
4230 				break;
4231 			}
4232 			cp = cp2;
4233 			i++;
4234 		}
4235 		if (!cantparse) {
4236 			/*
4237 			 * The host address INADDR_ANY is (mis)used to indicate
4238 			 * "there is no valid callback address".
4239 			 */
4240 			switch (af) {
4241 #ifdef INET6
4242 			case AF_INET6:
4243 				if (!IN6_ARE_ADDR_EQUAL(&rin6->sin6_addr,
4244 				    &in6addr_any))
4245 					rin6->sin6_port = htons(port.sval);
4246 				else
4247 					cantparse = 1;
4248 				break;
4249 #endif
4250 #ifdef INET
4251 			case AF_INET:
4252 				if (ip.ival != INADDR_ANY) {
4253 					rin->sin_addr.s_addr = htonl(ip.ival);
4254 					rin->sin_port = htons(port.sval);
4255 				} else {
4256 					cantparse = 1;
4257 				}
4258 				break;
4259 #endif
4260 			}
4261 		}
4262 	} else {
4263 		cantparse = 1;
4264 		if (i > 0) {
4265 			error = nfsm_advance(nd, NFSM_RNDUP(i), -1);
4266 			if (error)
4267 				goto nfsmout;
4268 		}
4269 	}
4270 	if (cantparse) {
4271 		switch (nd->nd_nam->sa_family) {
4272 #ifdef INET
4273 		case AF_INET:
4274 			sin = (struct sockaddr_in *)nd->nd_nam;
4275 			rin = (struct sockaddr_in *)clp->lc_req.nr_nam;
4276 			rin->sin_family = AF_INET;
4277 			rin->sin_len = sizeof(struct sockaddr_in);
4278 			rin->sin_addr.s_addr = sin->sin_addr.s_addr;
4279 			rin->sin_port = 0x0;
4280 			break;
4281 #endif
4282 #ifdef INET6
4283 		case AF_INET6:
4284 			sin6 = (struct sockaddr_in6 *)nd->nd_nam;
4285 			rin6 = (struct sockaddr_in6 *)clp->lc_req.nr_nam;
4286 			rin6->sin6_family = AF_INET6;
4287 			rin6->sin6_len = sizeof(struct sockaddr_in6);
4288 			rin6->sin6_addr = sin6->sin6_addr;
4289 			rin6->sin6_port = 0x0;
4290 			break;
4291 #endif
4292 		}
4293 		clp->lc_program = 0;
4294 	}
4295 nfsmout:
4296 	free(addr, M_TEMP);
4297 	NFSEXITCODE2(error, nd);
4298 	return (error);
4299 }
4300 
4301 /*
4302  * Turn a string of up to three decimal digits into a number. Return -1 upon
4303  * error.
4304  */
4305 static int
4306 nfsrv_getipnumber(u_char *cp)
4307 {
4308 	int i = 0, j = 0;
4309 
4310 	while (*cp) {
4311 		if (j > 2 || *cp < '0' || *cp > '9')
4312 			return (-1);
4313 		i *= 10;
4314 		i += (*cp - '0');
4315 		cp++;
4316 		j++;
4317 	}
4318 	if (i < 256)
4319 		return (i);
4320 	return (-1);
4321 }
4322 
4323 /*
4324  * This function checks for restart conditions.
4325  */
4326 static int
4327 nfsrv_checkrestart(nfsquad_t clientid, u_int32_t flags,
4328     nfsv4stateid_t *stateidp, int specialid)
4329 {
4330 	int ret = 0;
4331 
4332 	/*
4333 	 * First check for a server restart. Open, LockT, ReleaseLockOwner
4334 	 * and DelegPurge have a clientid, the rest a stateid.
4335 	 */
4336 	if (flags &
4337 	    (NFSLCK_OPEN | NFSLCK_TEST | NFSLCK_RELEASE | NFSLCK_DELEGPURGE)) {
4338 		if (clientid.lval[0] != nfsrvboottime) {
4339 			ret = NFSERR_STALECLIENTID;
4340 			goto out;
4341 		}
4342 	} else if (stateidp->other[0] != nfsrvboottime &&
4343 		specialid == 0) {
4344 		ret = NFSERR_STALESTATEID;
4345 		goto out;
4346 	}
4347 
4348 	/*
4349 	 * Read, Write, Setattr and LockT can return NFSERR_GRACE and do
4350 	 * not use a lock/open owner seqid#, so the check can be done now.
4351 	 * (The others will be checked, as required, later.)
4352 	 */
4353 	if (!(flags & (NFSLCK_CHECK | NFSLCK_TEST)))
4354 		goto out;
4355 
4356 	NFSLOCKSTATE();
4357 	ret = nfsrv_checkgrace(NULL, NULL, flags);
4358 	NFSUNLOCKSTATE();
4359 
4360 out:
4361 	NFSEXITCODE(ret);
4362 	return (ret);
4363 }
4364 
4365 /*
4366  * Check for grace.
4367  */
4368 static int
4369 nfsrv_checkgrace(struct nfsrv_descript *nd, struct nfsclient *clp,
4370     u_int32_t flags)
4371 {
4372 	int error = 0, notreclaimed;
4373 	struct nfsrv_stable *sp;
4374 
4375 	if ((nfsrv_stablefirst.nsf_flags & (NFSNSF_UPDATEDONE |
4376 	     NFSNSF_GRACEOVER)) == 0) {
4377 		/*
4378 		 * First, check to see if all of the clients have done a
4379 		 * ReclaimComplete.  If so, grace can end now.
4380 		 */
4381 		notreclaimed = 0;
4382 		LIST_FOREACH(sp, &nfsrv_stablefirst.nsf_head, nst_list) {
4383 			if ((sp->nst_flag & NFSNST_RECLAIMED) == 0) {
4384 				notreclaimed = 1;
4385 				break;
4386 			}
4387 		}
4388 		if (notreclaimed == 0)
4389 			nfsrv_stablefirst.nsf_flags |= (NFSNSF_GRACEOVER |
4390 			    NFSNSF_NEEDLOCK);
4391 	}
4392 
4393 	if ((nfsrv_stablefirst.nsf_flags & NFSNSF_GRACEOVER) != 0) {
4394 		if (flags & NFSLCK_RECLAIM) {
4395 			error = NFSERR_NOGRACE;
4396 			goto out;
4397 		}
4398 	} else {
4399 		if (!(flags & NFSLCK_RECLAIM)) {
4400 			error = NFSERR_GRACE;
4401 			goto out;
4402 		}
4403 		if (nd != NULL && clp != NULL &&
4404 		    (nd->nd_flag & ND_NFSV41) != 0 &&
4405 		    (clp->lc_flags & LCL_RECLAIMCOMPLETE) != 0) {
4406 			error = NFSERR_NOGRACE;
4407 			goto out;
4408 		}
4409 
4410 		/*
4411 		 * If grace is almost over and we are still getting Reclaims,
4412 		 * extend grace a bit.
4413 		 */
4414 		if ((NFSD_MONOSEC + NFSRV_LEASEDELTA) >
4415 		    nfsrv_stablefirst.nsf_eograce)
4416 			nfsrv_stablefirst.nsf_eograce = NFSD_MONOSEC +
4417 				NFSRV_LEASEDELTA;
4418 	}
4419 
4420 out:
4421 	NFSEXITCODE(error);
4422 	return (error);
4423 }
4424 
4425 /*
4426  * Do a server callback.
4427  * The "trunc" argument is slightly overloaded and refers to different
4428  * boolean arguments for CBRECALL and CBLAYOUTRECALL.
4429  */
4430 static int
4431 nfsrv_docallback(struct nfsclient *clp, int procnum, nfsv4stateid_t *stateidp,
4432     int trunc, fhandle_t *fhp, struct nfsvattr *nap, nfsattrbit_t *attrbitp,
4433     int laytype, NFSPROC_T *p)
4434 {
4435 	struct mbuf *m;
4436 	u_int32_t *tl;
4437 	struct nfsrv_descript *nd;
4438 	struct ucred *cred;
4439 	int error = 0, slotpos;
4440 	u_int32_t callback;
4441 	struct nfsdsession *sep = NULL;
4442 	uint64_t tval;
4443 	bool dotls;
4444 
4445 	nd = malloc(sizeof(*nd), M_TEMP, M_WAITOK | M_ZERO);
4446 	cred = newnfs_getcred();
4447 	NFSLOCKSTATE();	/* mostly for lc_cbref++ */
4448 	if (clp->lc_flags & LCL_NEEDSCONFIRM) {
4449 		NFSUNLOCKSTATE();
4450 		panic("docallb");
4451 	}
4452 	clp->lc_cbref++;
4453 
4454 	/*
4455 	 * Fill the callback program# and version into the request
4456 	 * structure for newnfs_connect() to use.
4457 	 */
4458 	clp->lc_req.nr_prog = clp->lc_program;
4459 #ifdef notnow
4460 	if ((clp->lc_flags & LCL_NFSV41) != 0)
4461 		clp->lc_req.nr_vers = NFSV41_CBVERS;
4462 	else
4463 #endif
4464 		clp->lc_req.nr_vers = NFSV4_CBVERS;
4465 
4466 	/*
4467 	 * First, fill in some of the fields of nd and cr.
4468 	 */
4469 	nd->nd_flag = ND_NFSV4;
4470 	if (clp->lc_flags & LCL_GSS)
4471 		nd->nd_flag |= ND_KERBV;
4472 	if ((clp->lc_flags & LCL_NFSV41) != 0)
4473 		nd->nd_flag |= ND_NFSV41;
4474 	if ((clp->lc_flags & LCL_NFSV42) != 0)
4475 		nd->nd_flag |= ND_NFSV42;
4476 	nd->nd_repstat = 0;
4477 	cred->cr_uid = clp->lc_uid;
4478 	cred->cr_gid = clp->lc_gid;
4479 	callback = clp->lc_callback;
4480 	NFSUNLOCKSTATE();
4481 	cred->cr_ngroups = 1;
4482 
4483 	/*
4484 	 * Get the first mbuf for the request.
4485 	 */
4486 	MGET(m, M_WAITOK, MT_DATA);
4487 	m->m_len = 0;
4488 	nd->nd_mreq = nd->nd_mb = m;
4489 	nd->nd_bpos = mtod(m, caddr_t);
4490 
4491 	/*
4492 	 * and build the callback request.
4493 	 */
4494 	if (procnum == NFSV4OP_CBGETATTR) {
4495 		nd->nd_procnum = NFSV4PROC_CBCOMPOUND;
4496 		error = nfsrv_cbcallargs(nd, clp, callback, NFSV4OP_CBGETATTR,
4497 		    "CB Getattr", &sep, &slotpos);
4498 		if (error != 0) {
4499 			m_freem(nd->nd_mreq);
4500 			goto errout;
4501 		}
4502 		(void)nfsm_fhtom(nd, (u_int8_t *)fhp, NFSX_MYFH, 0);
4503 		(void)nfsrv_putattrbit(nd, attrbitp);
4504 	} else if (procnum == NFSV4OP_CBRECALL) {
4505 		nd->nd_procnum = NFSV4PROC_CBCOMPOUND;
4506 		error = nfsrv_cbcallargs(nd, clp, callback, NFSV4OP_CBRECALL,
4507 		    "CB Recall", &sep, &slotpos);
4508 		if (error != 0) {
4509 			m_freem(nd->nd_mreq);
4510 			goto errout;
4511 		}
4512 		NFSM_BUILD(tl, u_int32_t *, NFSX_UNSIGNED + NFSX_STATEID);
4513 		*tl++ = txdr_unsigned(stateidp->seqid);
4514 		NFSBCOPY((caddr_t)stateidp->other, (caddr_t)tl,
4515 		    NFSX_STATEIDOTHER);
4516 		tl += (NFSX_STATEIDOTHER / NFSX_UNSIGNED);
4517 		if (trunc)
4518 			*tl = newnfs_true;
4519 		else
4520 			*tl = newnfs_false;
4521 		(void)nfsm_fhtom(nd, (u_int8_t *)fhp, NFSX_MYFH, 0);
4522 	} else if (procnum == NFSV4OP_CBLAYOUTRECALL) {
4523 		NFSD_DEBUG(4, "docallback layout recall\n");
4524 		nd->nd_procnum = NFSV4PROC_CBCOMPOUND;
4525 		error = nfsrv_cbcallargs(nd, clp, callback,
4526 		    NFSV4OP_CBLAYOUTRECALL, "CB Reclayout", &sep, &slotpos);
4527 		NFSD_DEBUG(4, "aft cbcallargs=%d\n", error);
4528 		if (error != 0) {
4529 			m_freem(nd->nd_mreq);
4530 			goto errout;
4531 		}
4532 		NFSM_BUILD(tl, u_int32_t *, 4 * NFSX_UNSIGNED);
4533 		*tl++ = txdr_unsigned(laytype);
4534 		*tl++ = txdr_unsigned(NFSLAYOUTIOMODE_ANY);
4535 		if (trunc)
4536 			*tl++ = newnfs_true;
4537 		else
4538 			*tl++ = newnfs_false;
4539 		*tl = txdr_unsigned(NFSV4LAYOUTRET_FILE);
4540 		nfsm_fhtom(nd, (uint8_t *)fhp, NFSX_MYFH, 0);
4541 		NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_HYPER + NFSX_STATEID);
4542 		tval = 0;
4543 		txdr_hyper(tval, tl); tl += 2;
4544 		tval = UINT64_MAX;
4545 		txdr_hyper(tval, tl); tl += 2;
4546 		*tl++ = txdr_unsigned(stateidp->seqid);
4547 		NFSBCOPY(stateidp->other, tl, NFSX_STATEIDOTHER);
4548 		tl += (NFSX_STATEIDOTHER / NFSX_UNSIGNED);
4549 		NFSD_DEBUG(4, "aft args\n");
4550 	} else if (procnum == NFSV4PROC_CBNULL) {
4551 		nd->nd_procnum = NFSV4PROC_CBNULL;
4552 		if ((clp->lc_flags & LCL_NFSV41) != 0) {
4553 			error = nfsv4_getcbsession(clp, &sep);
4554 			if (error != 0) {
4555 				m_freem(nd->nd_mreq);
4556 				goto errout;
4557 			}
4558 		}
4559 	} else {
4560 		error = NFSERR_SERVERFAULT;
4561 		m_freem(nd->nd_mreq);
4562 		goto errout;
4563 	}
4564 
4565 	/*
4566 	 * Call newnfs_connect(), as required, and then newnfs_request().
4567 	 */
4568 	dotls = false;
4569 	if ((clp->lc_flags & LCL_TLSCB) != 0)
4570 		dotls = true;
4571 	(void) newnfs_sndlock(&clp->lc_req.nr_lock);
4572 	if (clp->lc_req.nr_client == NULL) {
4573 		if ((clp->lc_flags & LCL_NFSV41) != 0) {
4574 			error = ECONNREFUSED;
4575 			if (procnum != NFSV4PROC_CBNULL)
4576 				nfsv4_freeslot(&sep->sess_cbsess, slotpos,
4577 				    true);
4578 			nfsrv_freesession(sep, NULL);
4579 		} else if (nd->nd_procnum == NFSV4PROC_CBNULL)
4580 			error = newnfs_connect(NULL, &clp->lc_req, cred,
4581 			    NULL, 1, dotls, &clp->lc_req.nr_client);
4582 		else
4583 			error = newnfs_connect(NULL, &clp->lc_req, cred,
4584 			    NULL, 3, dotls, &clp->lc_req.nr_client);
4585 	}
4586 	newnfs_sndunlock(&clp->lc_req.nr_lock);
4587 	NFSD_DEBUG(4, "aft sndunlock=%d\n", error);
4588 	if (!error) {
4589 		if ((nd->nd_flag & ND_NFSV41) != 0) {
4590 			KASSERT(sep != NULL, ("sep NULL"));
4591 			if (sep->sess_cbsess.nfsess_xprt != NULL)
4592 				error = newnfs_request(nd, NULL, clp,
4593 				    &clp->lc_req, NULL, NULL, cred,
4594 				    clp->lc_program, clp->lc_req.nr_vers, NULL,
4595 				    1, NULL, &sep->sess_cbsess);
4596 			else {
4597 				/*
4598 				 * This should probably never occur, but if a
4599 				 * client somehow does an RPC without a
4600 				 * SequenceID Op that causes a callback just
4601 				 * after the nfsd threads have been terminated
4602 				 * and restared we could conceivably get here
4603 				 * without a backchannel xprt.
4604 				 */
4605 				printf("nfsrv_docallback: no xprt\n");
4606 				error = ECONNREFUSED;
4607 			}
4608 			NFSD_DEBUG(4, "aft newnfs_request=%d\n", error);
4609 			if (error != 0 && procnum != NFSV4PROC_CBNULL) {
4610 				/*
4611 				 * It is likely that the callback was never
4612 				 * processed by the client and, as such,
4613 				 * the sequence# for the session slot needs
4614 				 * to be backed up by one to avoid a
4615 				 * NFSERR_SEQMISORDERED error reply.
4616 				 * For the unlikely case where the callback
4617 				 * was processed by the client, this will
4618 				 * make the next callback on the slot
4619 				 * appear to be a retry.
4620 				 * Since callbacks never specify that the
4621 				 * reply be cached, this "apparent retry"
4622 				 * should not be a problem.
4623 				 */
4624 				nfsv4_freeslot(&sep->sess_cbsess, slotpos,
4625 				    true);
4626 			}
4627 			nfsrv_freesession(sep, NULL);
4628 		} else
4629 			error = newnfs_request(nd, NULL, clp, &clp->lc_req,
4630 			    NULL, NULL, cred, clp->lc_program,
4631 			    clp->lc_req.nr_vers, NULL, 1, NULL, NULL);
4632 	}
4633 errout:
4634 	NFSFREECRED(cred);
4635 
4636 	/*
4637 	 * If error is set here, the Callback path isn't working
4638 	 * properly, so twiddle the appropriate LCL_ flags.
4639 	 * (nd_repstat != 0 indicates the Callback path is working,
4640 	 *  but the callback failed on the client.)
4641 	 */
4642 	if (error) {
4643 		/*
4644 		 * Mark the callback pathway down, which disabled issuing
4645 		 * of delegations and gets Renew to return NFSERR_CBPATHDOWN.
4646 		 */
4647 		NFSLOCKSTATE();
4648 		clp->lc_flags |= LCL_CBDOWN;
4649 		NFSUNLOCKSTATE();
4650 	} else {
4651 		/*
4652 		 * Callback worked. If the callback path was down, disable
4653 		 * callbacks, so no more delegations will be issued. (This
4654 		 * is done on the assumption that the callback pathway is
4655 		 * flakey.)
4656 		 */
4657 		NFSLOCKSTATE();
4658 		if (clp->lc_flags & LCL_CBDOWN)
4659 			clp->lc_flags &= ~(LCL_CBDOWN | LCL_CALLBACKSON);
4660 		NFSUNLOCKSTATE();
4661 		if (nd->nd_repstat) {
4662 			error = nd->nd_repstat;
4663 			NFSD_DEBUG(1, "nfsrv_docallback op=%d err=%d\n",
4664 			    procnum, error);
4665 		} else if (error == 0 && procnum == NFSV4OP_CBGETATTR)
4666 			error = nfsv4_loadattr(nd, NULL, nap, NULL, NULL, 0,
4667 			    NULL, NULL, NULL, NULL, NULL, 0, NULL, NULL, NULL,
4668 			    p, NULL);
4669 		m_freem(nd->nd_mrep);
4670 	}
4671 	NFSLOCKSTATE();
4672 	clp->lc_cbref--;
4673 	if ((clp->lc_flags & LCL_WAKEUPWANTED) && clp->lc_cbref == 0) {
4674 		clp->lc_flags &= ~LCL_WAKEUPWANTED;
4675 		wakeup(clp);
4676 	}
4677 	NFSUNLOCKSTATE();
4678 
4679 	free(nd, M_TEMP);
4680 	NFSEXITCODE(error);
4681 	return (error);
4682 }
4683 
4684 /*
4685  * Set up the compound RPC for the callback.
4686  */
4687 static int
4688 nfsrv_cbcallargs(struct nfsrv_descript *nd, struct nfsclient *clp,
4689     uint32_t callback, int op, const char *optag, struct nfsdsession **sepp,
4690     int *slotposp)
4691 {
4692 	uint32_t *tl;
4693 	int error, len;
4694 
4695 	len = strlen(optag);
4696 	(void)nfsm_strtom(nd, optag, len);
4697 	NFSM_BUILD(tl, uint32_t *, 4 * NFSX_UNSIGNED);
4698 	if ((nd->nd_flag & ND_NFSV41) != 0) {
4699 		if ((nd->nd_flag & ND_NFSV42) != 0)
4700 			*tl++ = txdr_unsigned(NFSV42_MINORVERSION);
4701 		else
4702 			*tl++ = txdr_unsigned(NFSV41_MINORVERSION);
4703 		*tl++ = txdr_unsigned(callback);
4704 		*tl++ = txdr_unsigned(2);
4705 		*tl = txdr_unsigned(NFSV4OP_CBSEQUENCE);
4706 		error = nfsv4_setcbsequence(nd, clp, 1, sepp, slotposp);
4707 		if (error != 0)
4708 			return (error);
4709 		NFSM_BUILD(tl, u_int32_t *, NFSX_UNSIGNED);
4710 		*tl = txdr_unsigned(op);
4711 	} else {
4712 		*tl++ = txdr_unsigned(NFSV4_MINORVERSION);
4713 		*tl++ = txdr_unsigned(callback);
4714 		*tl++ = txdr_unsigned(1);
4715 		*tl = txdr_unsigned(op);
4716 	}
4717 	return (0);
4718 }
4719 
4720 /*
4721  * Return the next index# for a clientid. Mostly just increment and return
4722  * the next one, but... if the 32bit unsigned does actually wrap around,
4723  * it should be rebooted.
4724  * At an average rate of one new client per second, it will wrap around in
4725  * approximately 136 years. (I think the server will have been shut
4726  * down or rebooted before then.)
4727  */
4728 static u_int32_t
4729 nfsrv_nextclientindex(void)
4730 {
4731 	static u_int32_t client_index = 0;
4732 
4733 	client_index++;
4734 	if (client_index != 0)
4735 		return (client_index);
4736 
4737 	printf("%s: out of clientids\n", __func__);
4738 	return (client_index);
4739 }
4740 
4741 /*
4742  * Return the next index# for a stateid. Mostly just increment and return
4743  * the next one, but... if the 32bit unsigned does actually wrap around
4744  * (will a BSD server stay up that long?), find
4745  * new start and end values.
4746  */
4747 static u_int32_t
4748 nfsrv_nextstateindex(struct nfsclient *clp)
4749 {
4750 	struct nfsstate *stp;
4751 	int i;
4752 	u_int32_t canuse, min_index, max_index;
4753 
4754 	if (!(clp->lc_flags & LCL_INDEXNOTOK)) {
4755 		clp->lc_stateindex++;
4756 		if (clp->lc_stateindex != clp->lc_statemaxindex)
4757 			return (clp->lc_stateindex);
4758 	}
4759 
4760 	/*
4761 	 * Yuck, we've hit the end.
4762 	 * Look for a new min and max.
4763 	 */
4764 	min_index = 0;
4765 	max_index = 0xffffffff;
4766 	for (i = 0; i < nfsrv_statehashsize; i++) {
4767 	    LIST_FOREACH(stp, &clp->lc_stateid[i], ls_hash) {
4768 		if (stp->ls_stateid.other[2] > 0x80000000) {
4769 		    if (stp->ls_stateid.other[2] < max_index)
4770 			max_index = stp->ls_stateid.other[2];
4771 		} else {
4772 		    if (stp->ls_stateid.other[2] > min_index)
4773 			min_index = stp->ls_stateid.other[2];
4774 		}
4775 	    }
4776 	}
4777 
4778 	/*
4779 	 * Yikes, highly unlikely, but I'll handle it anyhow.
4780 	 */
4781 	if (min_index == 0x80000000 && max_index == 0x80000001) {
4782 	    canuse = 0;
4783 	    /*
4784 	     * Loop around until we find an unused entry. Return that
4785 	     * and set LCL_INDEXNOTOK, so the search will continue next time.
4786 	     * (This is one of those rare cases where a goto is the
4787 	     *  cleanest way to code the loop.)
4788 	     */
4789 tryagain:
4790 	    for (i = 0; i < nfsrv_statehashsize; i++) {
4791 		LIST_FOREACH(stp, &clp->lc_stateid[i], ls_hash) {
4792 		    if (stp->ls_stateid.other[2] == canuse) {
4793 			canuse++;
4794 			goto tryagain;
4795 		    }
4796 		}
4797 	    }
4798 	    clp->lc_flags |= LCL_INDEXNOTOK;
4799 	    return (canuse);
4800 	}
4801 
4802 	/*
4803 	 * Ok to start again from min + 1.
4804 	 */
4805 	clp->lc_stateindex = min_index + 1;
4806 	clp->lc_statemaxindex = max_index;
4807 	clp->lc_flags &= ~LCL_INDEXNOTOK;
4808 	return (clp->lc_stateindex);
4809 }
4810 
4811 /*
4812  * The following functions handle the stable storage file that deals with
4813  * the edge conditions described in RFC3530 Sec. 8.6.3.
4814  * The file is as follows:
4815  * - a single record at the beginning that has the lease time of the
4816  *   previous server instance (before the last reboot) and the nfsrvboottime
4817  *   values for the previous server boots.
4818  *   These previous boot times are used to ensure that the current
4819  *   nfsrvboottime does not, somehow, get set to a previous one.
4820  *   (This is important so that Stale ClientIDs and StateIDs can
4821  *    be recognized.)
4822  *   The number of previous nfsvrboottime values precedes the list.
4823  * - followed by some number of appended records with:
4824  *   - client id string
4825  *   - flag that indicates it is a record revoking state via lease
4826  *     expiration or similar
4827  *     OR has successfully acquired state.
4828  * These structures vary in length, with the client string at the end, up
4829  * to NFSV4_OPAQUELIMIT in size.
4830  *
4831  * At the end of the grace period, the file is truncated, the first
4832  * record is rewritten with updated information and any acquired state
4833  * records for successful reclaims of state are written.
4834  *
4835  * Subsequent records are appended when the first state is issued to
4836  * a client and when state is revoked for a client.
4837  *
4838  * When reading the file in, state issued records that come later in
4839  * the file override older ones, since the append log is in cronological order.
4840  * If, for some reason, the file can't be read, the grace period is
4841  * immediately terminated and all reclaims get NFSERR_NOGRACE.
4842  */
4843 
4844 /*
4845  * Read in the stable storage file. Called by nfssvc() before the nfsd
4846  * processes start servicing requests.
4847  */
4848 void
4849 nfsrv_setupstable(NFSPROC_T *p)
4850 {
4851 	struct nfsrv_stablefirst *sf = &nfsrv_stablefirst;
4852 	struct nfsrv_stable *sp, *nsp;
4853 	struct nfst_rec *tsp;
4854 	int error, i, tryagain;
4855 	off_t off = 0;
4856 	ssize_t aresid, len;
4857 
4858 	/*
4859 	 * If NFSNSF_UPDATEDONE is set, this is a restart of the nfsds without
4860 	 * a reboot, so state has not been lost.
4861 	 */
4862 	if (sf->nsf_flags & NFSNSF_UPDATEDONE)
4863 		return;
4864 	/*
4865 	 * Set Grace over just until the file reads successfully.
4866 	 */
4867 	nfsrvboottime = time_second;
4868 	LIST_INIT(&sf->nsf_head);
4869 	sf->nsf_flags = (NFSNSF_GRACEOVER | NFSNSF_NEEDLOCK);
4870 	sf->nsf_eograce = NFSD_MONOSEC + NFSRV_LEASEDELTA;
4871 	if (sf->nsf_fp == NULL)
4872 		return;
4873 	error = NFSD_RDWR(UIO_READ, NFSFPVNODE(sf->nsf_fp),
4874 	    (caddr_t)&sf->nsf_rec, sizeof (struct nfsf_rec), off, UIO_SYSSPACE,
4875 	    0, NFSFPCRED(sf->nsf_fp), &aresid, p);
4876 	if (error || aresid || sf->nsf_numboots == 0 ||
4877 		sf->nsf_numboots > NFSNSF_MAXNUMBOOTS)
4878 		return;
4879 
4880 	/*
4881 	 * Now, read in the boottimes.
4882 	 */
4883 	sf->nsf_bootvals = (time_t *)malloc((sf->nsf_numboots + 1) *
4884 		sizeof (time_t), M_TEMP, M_WAITOK);
4885 	off = sizeof (struct nfsf_rec);
4886 	error = NFSD_RDWR(UIO_READ, NFSFPVNODE(sf->nsf_fp),
4887 	    (caddr_t)sf->nsf_bootvals, sf->nsf_numboots * sizeof (time_t), off,
4888 	    UIO_SYSSPACE, 0, NFSFPCRED(sf->nsf_fp), &aresid, p);
4889 	if (error || aresid) {
4890 		free(sf->nsf_bootvals, M_TEMP);
4891 		sf->nsf_bootvals = NULL;
4892 		return;
4893 	}
4894 
4895 	/*
4896 	 * Make sure this nfsrvboottime is different from all recorded
4897 	 * previous ones.
4898 	 */
4899 	do {
4900 		tryagain = 0;
4901 		for (i = 0; i < sf->nsf_numboots; i++) {
4902 			if (nfsrvboottime == sf->nsf_bootvals[i]) {
4903 				nfsrvboottime++;
4904 				tryagain = 1;
4905 				break;
4906 			}
4907 		}
4908 	} while (tryagain);
4909 
4910 	sf->nsf_flags |= NFSNSF_OK;
4911 	off += (sf->nsf_numboots * sizeof (time_t));
4912 
4913 	/*
4914 	 * Read through the file, building a list of records for grace
4915 	 * checking.
4916 	 * Each record is between sizeof (struct nfst_rec) and
4917 	 * sizeof (struct nfst_rec) + NFSV4_OPAQUELIMIT - 1
4918 	 * and is actually sizeof (struct nfst_rec) + nst_len - 1.
4919 	 */
4920 	tsp = (struct nfst_rec *)malloc(sizeof (struct nfst_rec) +
4921 		NFSV4_OPAQUELIMIT - 1, M_TEMP, M_WAITOK);
4922 	do {
4923 	    error = NFSD_RDWR(UIO_READ, NFSFPVNODE(sf->nsf_fp),
4924 	        (caddr_t)tsp, sizeof (struct nfst_rec) + NFSV4_OPAQUELIMIT - 1,
4925 	        off, UIO_SYSSPACE, 0, NFSFPCRED(sf->nsf_fp), &aresid, p);
4926 	    len = (sizeof (struct nfst_rec) + NFSV4_OPAQUELIMIT - 1) - aresid;
4927 	    if (error || (len > 0 && (len < sizeof (struct nfst_rec) ||
4928 		len < (sizeof (struct nfst_rec) + tsp->len - 1)))) {
4929 		/*
4930 		 * Yuck, the file has been corrupted, so just return
4931 		 * after clearing out any restart state, so the grace period
4932 		 * is over.
4933 		 */
4934 		LIST_FOREACH_SAFE(sp, &sf->nsf_head, nst_list, nsp) {
4935 			LIST_REMOVE(sp, nst_list);
4936 			free(sp, M_TEMP);
4937 		}
4938 		free(tsp, M_TEMP);
4939 		sf->nsf_flags &= ~NFSNSF_OK;
4940 		free(sf->nsf_bootvals, M_TEMP);
4941 		sf->nsf_bootvals = NULL;
4942 		return;
4943 	    }
4944 	    if (len > 0) {
4945 		off += sizeof (struct nfst_rec) + tsp->len - 1;
4946 		/*
4947 		 * Search the list for a matching client.
4948 		 */
4949 		LIST_FOREACH(sp, &sf->nsf_head, nst_list) {
4950 			if (tsp->len == sp->nst_len &&
4951 			    !NFSBCMP(tsp->client, sp->nst_client, tsp->len))
4952 				break;
4953 		}
4954 		if (sp == LIST_END(&sf->nsf_head)) {
4955 			sp = (struct nfsrv_stable *)malloc(tsp->len +
4956 				sizeof (struct nfsrv_stable) - 1, M_TEMP,
4957 				M_WAITOK);
4958 			NFSBCOPY((caddr_t)tsp, (caddr_t)&sp->nst_rec,
4959 				sizeof (struct nfst_rec) + tsp->len - 1);
4960 			LIST_INSERT_HEAD(&sf->nsf_head, sp, nst_list);
4961 		} else {
4962 			if (tsp->flag == NFSNST_REVOKE)
4963 				sp->nst_flag |= NFSNST_REVOKE;
4964 			else
4965 				/*
4966 				 * A subsequent timestamp indicates the client
4967 				 * did a setclientid/confirm and any previous
4968 				 * revoke is no longer relevant.
4969 				 */
4970 				sp->nst_flag &= ~NFSNST_REVOKE;
4971 		}
4972 	    }
4973 	} while (len > 0);
4974 	free(tsp, M_TEMP);
4975 	sf->nsf_flags = NFSNSF_OK;
4976 	sf->nsf_eograce = NFSD_MONOSEC + sf->nsf_lease +
4977 		NFSRV_LEASEDELTA;
4978 }
4979 
4980 /*
4981  * Update the stable storage file, now that the grace period is over.
4982  */
4983 void
4984 nfsrv_updatestable(NFSPROC_T *p)
4985 {
4986 	struct nfsrv_stablefirst *sf = &nfsrv_stablefirst;
4987 	struct nfsrv_stable *sp, *nsp;
4988 	int i;
4989 	struct nfsvattr nva;
4990 	vnode_t vp;
4991 #if defined(__FreeBSD_version) && (__FreeBSD_version >= 500000)
4992 	mount_t mp = NULL;
4993 #endif
4994 	int error;
4995 
4996 	if (sf->nsf_fp == NULL || (sf->nsf_flags & NFSNSF_UPDATEDONE))
4997 		return;
4998 	sf->nsf_flags |= NFSNSF_UPDATEDONE;
4999 	/*
5000 	 * Ok, we need to rewrite the stable storage file.
5001 	 * - truncate to 0 length
5002 	 * - write the new first structure
5003 	 * - loop through the data structures, writing out any that
5004 	 *   have timestamps older than the old boot
5005 	 */
5006 	if (sf->nsf_bootvals) {
5007 		sf->nsf_numboots++;
5008 		for (i = sf->nsf_numboots - 2; i >= 0; i--)
5009 			sf->nsf_bootvals[i + 1] = sf->nsf_bootvals[i];
5010 	} else {
5011 		sf->nsf_numboots = 1;
5012 		sf->nsf_bootvals = (time_t *)malloc(sizeof (time_t),
5013 			M_TEMP, M_WAITOK);
5014 	}
5015 	sf->nsf_bootvals[0] = nfsrvboottime;
5016 	sf->nsf_lease = nfsrv_lease;
5017 	NFSVNO_ATTRINIT(&nva);
5018 	NFSVNO_SETATTRVAL(&nva, size, 0);
5019 	vp = NFSFPVNODE(sf->nsf_fp);
5020 	vn_start_write(vp, &mp, V_WAIT);
5021 	if (NFSVOPLOCK(vp, LK_EXCLUSIVE) == 0) {
5022 		error = nfsvno_setattr(vp, &nva, NFSFPCRED(sf->nsf_fp), p,
5023 		    NULL);
5024 		NFSVOPUNLOCK(vp);
5025 	} else
5026 		error = EPERM;
5027 	vn_finished_write(mp);
5028 	if (!error)
5029 	    error = NFSD_RDWR(UIO_WRITE, vp,
5030 		(caddr_t)&sf->nsf_rec, sizeof (struct nfsf_rec), (off_t)0,
5031 		UIO_SYSSPACE, IO_SYNC, NFSFPCRED(sf->nsf_fp), NULL, p);
5032 	if (!error)
5033 	    error = NFSD_RDWR(UIO_WRITE, vp,
5034 		(caddr_t)sf->nsf_bootvals,
5035 		sf->nsf_numboots * sizeof (time_t),
5036 		(off_t)(sizeof (struct nfsf_rec)),
5037 		UIO_SYSSPACE, IO_SYNC, NFSFPCRED(sf->nsf_fp), NULL, p);
5038 	free(sf->nsf_bootvals, M_TEMP);
5039 	sf->nsf_bootvals = NULL;
5040 	if (error) {
5041 		sf->nsf_flags &= ~NFSNSF_OK;
5042 		printf("EEK! Can't write NfsV4 stable storage file\n");
5043 		return;
5044 	}
5045 	sf->nsf_flags |= NFSNSF_OK;
5046 
5047 	/*
5048 	 * Loop through the list and write out timestamp records for
5049 	 * any clients that successfully reclaimed state.
5050 	 */
5051 	LIST_FOREACH_SAFE(sp, &sf->nsf_head, nst_list, nsp) {
5052 		if (sp->nst_flag & NFSNST_GOTSTATE) {
5053 			nfsrv_writestable(sp->nst_client, sp->nst_len,
5054 				NFSNST_NEWSTATE, p);
5055 			sp->nst_clp->lc_flags |= LCL_STAMPEDSTABLE;
5056 		}
5057 		LIST_REMOVE(sp, nst_list);
5058 		free(sp, M_TEMP);
5059 	}
5060 	nfsrv_backupstable();
5061 }
5062 
5063 /*
5064  * Append a record to the stable storage file.
5065  */
5066 void
5067 nfsrv_writestable(u_char *client, int len, int flag, NFSPROC_T *p)
5068 {
5069 	struct nfsrv_stablefirst *sf = &nfsrv_stablefirst;
5070 	struct nfst_rec *sp;
5071 	int error;
5072 
5073 	if (!(sf->nsf_flags & NFSNSF_OK) || sf->nsf_fp == NULL)
5074 		return;
5075 	sp = (struct nfst_rec *)malloc(sizeof (struct nfst_rec) +
5076 		len - 1, M_TEMP, M_WAITOK);
5077 	sp->len = len;
5078 	NFSBCOPY(client, sp->client, len);
5079 	sp->flag = flag;
5080 	error = NFSD_RDWR(UIO_WRITE, NFSFPVNODE(sf->nsf_fp),
5081 	    (caddr_t)sp, sizeof (struct nfst_rec) + len - 1, (off_t)0,
5082 	    UIO_SYSSPACE, (IO_SYNC | IO_APPEND), NFSFPCRED(sf->nsf_fp), NULL, p);
5083 	free(sp, M_TEMP);
5084 	if (error) {
5085 		sf->nsf_flags &= ~NFSNSF_OK;
5086 		printf("EEK! Can't write NfsV4 stable storage file\n");
5087 	}
5088 }
5089 
5090 /*
5091  * This function is called during the grace period to mark a client
5092  * that successfully reclaimed state.
5093  */
5094 static void
5095 nfsrv_markstable(struct nfsclient *clp)
5096 {
5097 	struct nfsrv_stable *sp;
5098 
5099 	/*
5100 	 * First find the client structure.
5101 	 */
5102 	LIST_FOREACH(sp, &nfsrv_stablefirst.nsf_head, nst_list) {
5103 		if (sp->nst_len == clp->lc_idlen &&
5104 		    !NFSBCMP(sp->nst_client, clp->lc_id, sp->nst_len))
5105 			break;
5106 	}
5107 	if (sp == LIST_END(&nfsrv_stablefirst.nsf_head))
5108 		return;
5109 
5110 	/*
5111 	 * Now, just mark it and set the nfsclient back pointer.
5112 	 */
5113 	sp->nst_flag |= NFSNST_GOTSTATE;
5114 	sp->nst_clp = clp;
5115 }
5116 
5117 /*
5118  * This function is called when a NFSv4.1 client does a ReclaimComplete.
5119  * Very similar to nfsrv_markstable(), except for the flag being set.
5120  */
5121 static void
5122 nfsrv_markreclaim(struct nfsclient *clp)
5123 {
5124 	struct nfsrv_stable *sp;
5125 
5126 	/*
5127 	 * First find the client structure.
5128 	 */
5129 	LIST_FOREACH(sp, &nfsrv_stablefirst.nsf_head, nst_list) {
5130 		if (sp->nst_len == clp->lc_idlen &&
5131 		    !NFSBCMP(sp->nst_client, clp->lc_id, sp->nst_len))
5132 			break;
5133 	}
5134 	if (sp == LIST_END(&nfsrv_stablefirst.nsf_head))
5135 		return;
5136 
5137 	/*
5138 	 * Now, just set the flag.
5139 	 */
5140 	sp->nst_flag |= NFSNST_RECLAIMED;
5141 }
5142 
5143 /*
5144  * This function is called for a reclaim, to see if it gets grace.
5145  * It returns 0 if a reclaim is allowed, 1 otherwise.
5146  */
5147 static int
5148 nfsrv_checkstable(struct nfsclient *clp)
5149 {
5150 	struct nfsrv_stable *sp;
5151 
5152 	/*
5153 	 * First, find the entry for the client.
5154 	 */
5155 	LIST_FOREACH(sp, &nfsrv_stablefirst.nsf_head, nst_list) {
5156 		if (sp->nst_len == clp->lc_idlen &&
5157 		    !NFSBCMP(sp->nst_client, clp->lc_id, sp->nst_len))
5158 			break;
5159 	}
5160 
5161 	/*
5162 	 * If not in the list, state was revoked or no state was issued
5163 	 * since the previous reboot, a reclaim is denied.
5164 	 */
5165 	if (sp == LIST_END(&nfsrv_stablefirst.nsf_head) ||
5166 	    (sp->nst_flag & NFSNST_REVOKE) ||
5167 	    !(nfsrv_stablefirst.nsf_flags & NFSNSF_OK))
5168 		return (1);
5169 	return (0);
5170 }
5171 
5172 /*
5173  * Test for and try to clear out a conflicting client. This is called by
5174  * nfsrv_lockctrl() and nfsrv_openctrl() when conflicts with other clients
5175  * a found.
5176  * The trick here is that it can't revoke a conflicting client with an
5177  * expired lease unless it holds the v4root lock, so...
5178  * If no v4root lock, get the lock and return 1 to indicate "try again".
5179  * Return 0 to indicate the conflict can't be revoked and 1 to indicate
5180  * the revocation worked and the conflicting client is "bye, bye", so it
5181  * can be tried again.
5182  * Return 2 to indicate that the vnode is VIRF_DOOMED after NFSVOPLOCK().
5183  * Unlocks State before a non-zero value is returned.
5184  */
5185 static int
5186 nfsrv_clientconflict(struct nfsclient *clp, int *haslockp, vnode_t vp,
5187     NFSPROC_T *p)
5188 {
5189 	int gotlock, lktype = 0;
5190 
5191 	/*
5192 	 * If lease hasn't expired, we can't fix it.
5193 	 */
5194 	if (clp->lc_expiry >= NFSD_MONOSEC ||
5195 	    !(nfsrv_stablefirst.nsf_flags & NFSNSF_UPDATEDONE))
5196 		return (0);
5197 	if (*haslockp == 0) {
5198 		NFSUNLOCKSTATE();
5199 		if (vp != NULL) {
5200 			lktype = NFSVOPISLOCKED(vp);
5201 			NFSVOPUNLOCK(vp);
5202 		}
5203 		NFSLOCKV4ROOTMUTEX();
5204 		nfsv4_relref(&nfsv4rootfs_lock);
5205 		do {
5206 			gotlock = nfsv4_lock(&nfsv4rootfs_lock, 1, NULL,
5207 			    NFSV4ROOTLOCKMUTEXPTR, NULL);
5208 		} while (!gotlock);
5209 		NFSUNLOCKV4ROOTMUTEX();
5210 		*haslockp = 1;
5211 		if (vp != NULL) {
5212 			NFSVOPLOCK(vp, lktype | LK_RETRY);
5213 			if (VN_IS_DOOMED(vp))
5214 				return (2);
5215 		}
5216 		return (1);
5217 	}
5218 	NFSUNLOCKSTATE();
5219 
5220 	/*
5221 	 * Ok, we can expire the conflicting client.
5222 	 */
5223 	nfsrv_writestable(clp->lc_id, clp->lc_idlen, NFSNST_REVOKE, p);
5224 	nfsrv_backupstable();
5225 	nfsrv_cleanclient(clp, p);
5226 	nfsrv_freedeleglist(&clp->lc_deleg);
5227 	nfsrv_freedeleglist(&clp->lc_olddeleg);
5228 	LIST_REMOVE(clp, lc_hash);
5229 	nfsrv_zapclient(clp, p);
5230 	return (1);
5231 }
5232 
5233 /*
5234  * Resolve a delegation conflict.
5235  * Returns 0 to indicate the conflict was resolved without sleeping.
5236  * Return -1 to indicate that the caller should check for conflicts again.
5237  * Return > 0 for an error that should be returned, normally NFSERR_DELAY.
5238  *
5239  * Also, manipulate the nfsv4root_lock, as required. It isn't changed
5240  * for a return of 0, since there was no sleep and it could be required
5241  * later. It is released for a return of NFSERR_DELAY, since the caller
5242  * will return that error. It is released when a sleep was done waiting
5243  * for the delegation to be returned or expire (so that other nfsds can
5244  * handle ops). Then, it must be acquired for the write to stable storage.
5245  * (This function is somewhat similar to nfsrv_clientconflict(), but
5246  *  the semantics differ in a couple of subtle ways. The return of 0
5247  *  indicates the conflict was resolved without sleeping here, not
5248  *  that the conflict can't be resolved and the handling of nfsv4root_lock
5249  *  differs, as noted above.)
5250  * Unlocks State before returning a non-zero value.
5251  */
5252 static int
5253 nfsrv_delegconflict(struct nfsstate *stp, int *haslockp, NFSPROC_T *p,
5254     vnode_t vp)
5255 {
5256 	struct nfsclient *clp = stp->ls_clp;
5257 	int gotlock, error, lktype = 0, retrycnt, zapped_clp;
5258 	nfsv4stateid_t tstateid;
5259 	fhandle_t tfh;
5260 
5261 	/*
5262 	 * If the conflict is with an old delegation...
5263 	 */
5264 	if (stp->ls_flags & NFSLCK_OLDDELEG) {
5265 		/*
5266 		 * You can delete it, if it has expired.
5267 		 */
5268 		if (clp->lc_delegtime < NFSD_MONOSEC) {
5269 			nfsrv_freedeleg(stp);
5270 			NFSUNLOCKSTATE();
5271 			error = -1;
5272 			goto out;
5273 		}
5274 		NFSUNLOCKSTATE();
5275 		/*
5276 		 * During this delay, the old delegation could expire or it
5277 		 * could be recovered by the client via an Open with
5278 		 * CLAIM_DELEGATE_PREV.
5279 		 * Release the nfsv4root_lock, if held.
5280 		 */
5281 		if (*haslockp) {
5282 			*haslockp = 0;
5283 			NFSLOCKV4ROOTMUTEX();
5284 			nfsv4_unlock(&nfsv4rootfs_lock, 1);
5285 			NFSUNLOCKV4ROOTMUTEX();
5286 		}
5287 		error = NFSERR_DELAY;
5288 		goto out;
5289 	}
5290 
5291 	/*
5292 	 * It's a current delegation, so:
5293 	 * - check to see if the delegation has expired
5294 	 *   - if so, get the v4root lock and then expire it
5295 	 */
5296 	if ((stp->ls_flags & NFSLCK_DELEGRECALL) == 0 || (stp->ls_lastrecall <
5297 	    NFSD_MONOSEC && clp->lc_expiry >= NFSD_MONOSEC &&
5298 	    stp->ls_delegtime >= NFSD_MONOSEC)) {
5299 		/*
5300 		 * - do a recall callback, since not yet done
5301 		 * For now, never allow truncate to be set. To use
5302 		 * truncate safely, it must be guaranteed that the
5303 		 * Remove, Rename or Setattr with size of 0 will
5304 		 * succeed and that would require major changes to
5305 		 * the VFS/Vnode OPs.
5306 		 * Set the expiry time large enough so that it won't expire
5307 		 * until after the callback, then set it correctly, once
5308 		 * the callback is done. (The delegation will now time
5309 		 * out whether or not the Recall worked ok. The timeout
5310 		 * will be extended when ops are done on the delegation
5311 		 * stateid, up to the timelimit.)
5312 		 */
5313 		if ((stp->ls_flags & NFSLCK_DELEGRECALL) == 0) {
5314 			stp->ls_delegtime = NFSD_MONOSEC + (2 * nfsrv_lease) +
5315 			    NFSRV_LEASEDELTA;
5316 			stp->ls_delegtimelimit = NFSD_MONOSEC + (6 *
5317 			    nfsrv_lease) + NFSRV_LEASEDELTA;
5318 			stp->ls_flags |= NFSLCK_DELEGRECALL;
5319 		}
5320 		stp->ls_lastrecall = time_uptime + 1;
5321 
5322 		/*
5323 		 * Loop NFSRV_CBRETRYCNT times while the CBRecall replies
5324 		 * NFSERR_BADSTATEID or NFSERR_BADHANDLE. This is done
5325 		 * in order to try and avoid a race that could happen
5326 		 * when a CBRecall request passed the Open reply with
5327 		 * the delegation in it when transitting the network.
5328 		 * Since nfsrv_docallback will sleep, don't use stp after
5329 		 * the call.
5330 		 */
5331 		NFSBCOPY((caddr_t)&stp->ls_stateid, (caddr_t)&tstateid,
5332 		    sizeof (tstateid));
5333 		NFSBCOPY((caddr_t)&stp->ls_lfp->lf_fh, (caddr_t)&tfh,
5334 		    sizeof (tfh));
5335 		NFSUNLOCKSTATE();
5336 		if (*haslockp) {
5337 			*haslockp = 0;
5338 			NFSLOCKV4ROOTMUTEX();
5339 			nfsv4_unlock(&nfsv4rootfs_lock, 1);
5340 			NFSUNLOCKV4ROOTMUTEX();
5341 		}
5342 		retrycnt = 0;
5343 		do {
5344 		    error = nfsrv_docallback(clp, NFSV4OP_CBRECALL,
5345 			&tstateid, 0, &tfh, NULL, NULL, 0, p);
5346 		    retrycnt++;
5347 		} while ((error == NFSERR_BADSTATEID ||
5348 		    error == NFSERR_BADHANDLE) && retrycnt < NFSV4_CBRETRYCNT);
5349 		error = NFSERR_DELAY;
5350 		goto out;
5351 	}
5352 
5353 	if (clp->lc_expiry >= NFSD_MONOSEC &&
5354 	    stp->ls_delegtime >= NFSD_MONOSEC) {
5355 		NFSUNLOCKSTATE();
5356 		/*
5357 		 * A recall has been done, but it has not yet expired.
5358 		 * So, RETURN_DELAY.
5359 		 */
5360 		if (*haslockp) {
5361 			*haslockp = 0;
5362 			NFSLOCKV4ROOTMUTEX();
5363 			nfsv4_unlock(&nfsv4rootfs_lock, 1);
5364 			NFSUNLOCKV4ROOTMUTEX();
5365 		}
5366 		error = NFSERR_DELAY;
5367 		goto out;
5368 	}
5369 
5370 	/*
5371 	 * If we don't yet have the lock, just get it and then return,
5372 	 * since we need that before deleting expired state, such as
5373 	 * this delegation.
5374 	 * When getting the lock, unlock the vnode, so other nfsds that
5375 	 * are in progress, won't get stuck waiting for the vnode lock.
5376 	 */
5377 	if (*haslockp == 0) {
5378 		NFSUNLOCKSTATE();
5379 		if (vp != NULL) {
5380 			lktype = NFSVOPISLOCKED(vp);
5381 			NFSVOPUNLOCK(vp);
5382 		}
5383 		NFSLOCKV4ROOTMUTEX();
5384 		nfsv4_relref(&nfsv4rootfs_lock);
5385 		do {
5386 			gotlock = nfsv4_lock(&nfsv4rootfs_lock, 1, NULL,
5387 			    NFSV4ROOTLOCKMUTEXPTR, NULL);
5388 		} while (!gotlock);
5389 		NFSUNLOCKV4ROOTMUTEX();
5390 		*haslockp = 1;
5391 		if (vp != NULL) {
5392 			NFSVOPLOCK(vp, lktype | LK_RETRY);
5393 			if (VN_IS_DOOMED(vp)) {
5394 				*haslockp = 0;
5395 				NFSLOCKV4ROOTMUTEX();
5396 				nfsv4_unlock(&nfsv4rootfs_lock, 1);
5397 				NFSUNLOCKV4ROOTMUTEX();
5398 				error = NFSERR_PERM;
5399 				goto out;
5400 			}
5401 		}
5402 		error = -1;
5403 		goto out;
5404 	}
5405 
5406 	NFSUNLOCKSTATE();
5407 	/*
5408 	 * Ok, we can delete the expired delegation.
5409 	 * First, write the Revoke record to stable storage and then
5410 	 * clear out the conflict.
5411 	 * Since all other nfsd threads are now blocked, we can safely
5412 	 * sleep without the state changing.
5413 	 */
5414 	nfsrv_writestable(clp->lc_id, clp->lc_idlen, NFSNST_REVOKE, p);
5415 	nfsrv_backupstable();
5416 	if (clp->lc_expiry < NFSD_MONOSEC) {
5417 		nfsrv_cleanclient(clp, p);
5418 		nfsrv_freedeleglist(&clp->lc_deleg);
5419 		nfsrv_freedeleglist(&clp->lc_olddeleg);
5420 		LIST_REMOVE(clp, lc_hash);
5421 		zapped_clp = 1;
5422 	} else {
5423 		nfsrv_freedeleg(stp);
5424 		zapped_clp = 0;
5425 	}
5426 	if (zapped_clp)
5427 		nfsrv_zapclient(clp, p);
5428 	error = -1;
5429 
5430 out:
5431 	NFSEXITCODE(error);
5432 	return (error);
5433 }
5434 
5435 /*
5436  * Check for a remove allowed, if remove is set to 1 and get rid of
5437  * delegations.
5438  */
5439 int
5440 nfsrv_checkremove(vnode_t vp, int remove, struct nfsrv_descript *nd,
5441     nfsquad_t clientid, NFSPROC_T *p)
5442 {
5443 	struct nfsclient *clp;
5444 	struct nfsstate *stp;
5445 	struct nfslockfile *lfp;
5446 	int error, haslock = 0;
5447 	fhandle_t nfh;
5448 
5449 	clp = NULL;
5450 	/*
5451 	 * First, get the lock file structure.
5452 	 * (A return of -1 means no associated state, so remove ok.)
5453 	 */
5454 	error = nfsrv_getlockfh(vp, NFSLCK_CHECK, NULL, &nfh, p);
5455 tryagain:
5456 	NFSLOCKSTATE();
5457 	if (error == 0 && clientid.qval != 0)
5458 		error = nfsrv_getclient(clientid, CLOPS_RENEW, &clp, NULL,
5459 		    (nfsquad_t)((u_quad_t)0), 0, nd, p);
5460 	if (!error)
5461 		error = nfsrv_getlockfile(NFSLCK_CHECK, NULL, &lfp, &nfh, 0);
5462 	if (error) {
5463 		NFSUNLOCKSTATE();
5464 		if (haslock) {
5465 			NFSLOCKV4ROOTMUTEX();
5466 			nfsv4_unlock(&nfsv4rootfs_lock, 1);
5467 			NFSUNLOCKV4ROOTMUTEX();
5468 		}
5469 		if (error == -1)
5470 			error = 0;
5471 		goto out;
5472 	}
5473 
5474 	/*
5475 	 * Now, we must Recall any delegations.
5476 	 */
5477 	error = nfsrv_cleandeleg(vp, lfp, clp, &haslock, p);
5478 	if (error) {
5479 		/*
5480 		 * nfsrv_cleandeleg() unlocks state for non-zero
5481 		 * return.
5482 		 */
5483 		if (error == -1)
5484 			goto tryagain;
5485 		if (haslock) {
5486 			NFSLOCKV4ROOTMUTEX();
5487 			nfsv4_unlock(&nfsv4rootfs_lock, 1);
5488 			NFSUNLOCKV4ROOTMUTEX();
5489 		}
5490 		goto out;
5491 	}
5492 
5493 	/*
5494 	 * Now, look for a conflicting open share.
5495 	 */
5496 	if (remove) {
5497 		/*
5498 		 * If the entry in the directory was the last reference to the
5499 		 * corresponding filesystem object, the object can be destroyed
5500 		 * */
5501 		if(lfp->lf_usecount>1)
5502 			LIST_FOREACH(stp, &lfp->lf_open, ls_file) {
5503 				if (stp->ls_flags & NFSLCK_WRITEDENY) {
5504 					error = NFSERR_FILEOPEN;
5505 					break;
5506 				}
5507 			}
5508 	}
5509 
5510 	NFSUNLOCKSTATE();
5511 	if (haslock) {
5512 		NFSLOCKV4ROOTMUTEX();
5513 		nfsv4_unlock(&nfsv4rootfs_lock, 1);
5514 		NFSUNLOCKV4ROOTMUTEX();
5515 	}
5516 
5517 out:
5518 	NFSEXITCODE(error);
5519 	return (error);
5520 }
5521 
5522 /*
5523  * Clear out all delegations for the file referred to by lfp.
5524  * May return NFSERR_DELAY, if there will be a delay waiting for
5525  * delegations to expire.
5526  * Returns -1 to indicate it slept while recalling a delegation.
5527  * This function has the side effect of deleting the nfslockfile structure,
5528  * if it no longer has associated state and didn't have to sleep.
5529  * Unlocks State before a non-zero value is returned.
5530  */
5531 static int
5532 nfsrv_cleandeleg(vnode_t vp, struct nfslockfile *lfp,
5533     struct nfsclient *clp, int *haslockp, NFSPROC_T *p)
5534 {
5535 	struct nfsstate *stp, *nstp;
5536 	int ret = 0;
5537 
5538 	stp = LIST_FIRST(&lfp->lf_deleg);
5539 	while (stp != LIST_END(&lfp->lf_deleg)) {
5540 		nstp = LIST_NEXT(stp, ls_file);
5541 		if (stp->ls_clp != clp) {
5542 			ret = nfsrv_delegconflict(stp, haslockp, p, vp);
5543 			if (ret) {
5544 				/*
5545 				 * nfsrv_delegconflict() unlocks state
5546 				 * when it returns non-zero.
5547 				 */
5548 				goto out;
5549 			}
5550 		}
5551 		stp = nstp;
5552 	}
5553 out:
5554 	NFSEXITCODE(ret);
5555 	return (ret);
5556 }
5557 
5558 /*
5559  * There are certain operations that, when being done outside of NFSv4,
5560  * require that any NFSv4 delegation for the file be recalled.
5561  * This function is to be called for those cases:
5562  * VOP_RENAME() - When a delegation is being recalled for any reason,
5563  *	the client may have to do Opens against the server, using the file's
5564  *	final component name. If the file has been renamed on the server,
5565  *	that component name will be incorrect and the Open will fail.
5566  * VOP_REMOVE() - Theoretically, a client could Open a file after it has
5567  *	been removed on the server, if there is a delegation issued to
5568  *	that client for the file. I say "theoretically" since clients
5569  *	normally do an Access Op before the Open and that Access Op will
5570  *	fail with ESTALE. Note that NFSv2 and 3 don't even do Opens, so
5571  *	they will detect the file's removal in the same manner. (There is
5572  *	one case where RFC3530 allows a client to do an Open without first
5573  *	doing an Access Op, which is passage of a check against the ACE
5574  *	returned with a Write delegation, but current practice is to ignore
5575  *	the ACE and always do an Access Op.)
5576  *	Since the functions can only be called with an unlocked vnode, this
5577  *	can't be done at this time.
5578  * VOP_ADVLOCK() - When a client holds a delegation, it can issue byte range
5579  *	locks locally in the client, which are not visible to the server. To
5580  *	deal with this, issuing of delegations for a vnode must be disabled
5581  *	and all delegations for the vnode recalled. This is done via the
5582  *	second function, using the VV_DISABLEDELEG vflag on the vnode.
5583  */
5584 void
5585 nfsd_recalldelegation(vnode_t vp, NFSPROC_T *p)
5586 {
5587 	time_t starttime;
5588 	int error;
5589 
5590 	/*
5591 	 * First, check to see if the server is currently running and it has
5592 	 * been called for a regular file when issuing delegations.
5593 	 */
5594 	if (newnfs_numnfsd == 0 || vp->v_type != VREG ||
5595 	    nfsrv_issuedelegs == 0)
5596 		return;
5597 
5598 	KASSERT((NFSVOPISLOCKED(vp) != LK_EXCLUSIVE), ("vp %p is locked", vp));
5599 	/*
5600 	 * First, get a reference on the nfsv4rootfs_lock so that an
5601 	 * exclusive lock cannot be acquired by another thread.
5602 	 */
5603 	NFSLOCKV4ROOTMUTEX();
5604 	nfsv4_getref(&nfsv4rootfs_lock, NULL, NFSV4ROOTLOCKMUTEXPTR, NULL);
5605 	NFSUNLOCKV4ROOTMUTEX();
5606 
5607 	/*
5608 	 * Now, call nfsrv_checkremove() in a loop while it returns
5609 	 * NFSERR_DELAY. Return upon any other error or when timed out.
5610 	 */
5611 	starttime = NFSD_MONOSEC;
5612 	do {
5613 		if (NFSVOPLOCK(vp, LK_EXCLUSIVE) == 0) {
5614 			error = nfsrv_checkremove(vp, 0, NULL,
5615 			    (nfsquad_t)((u_quad_t)0), p);
5616 			NFSVOPUNLOCK(vp);
5617 		} else
5618 			error = EPERM;
5619 		if (error == NFSERR_DELAY) {
5620 			if (NFSD_MONOSEC - starttime > NFS_REMOVETIMEO)
5621 				break;
5622 			/* Sleep for a short period of time */
5623 			(void) nfs_catnap(PZERO, 0, "nfsremove");
5624 		}
5625 	} while (error == NFSERR_DELAY);
5626 	NFSLOCKV4ROOTMUTEX();
5627 	nfsv4_relref(&nfsv4rootfs_lock);
5628 	NFSUNLOCKV4ROOTMUTEX();
5629 }
5630 
5631 void
5632 nfsd_disabledelegation(vnode_t vp, NFSPROC_T *p)
5633 {
5634 
5635 #ifdef VV_DISABLEDELEG
5636 	/*
5637 	 * First, flag issuance of delegations disabled.
5638 	 */
5639 	atomic_set_long(&vp->v_vflag, VV_DISABLEDELEG);
5640 #endif
5641 
5642 	/*
5643 	 * Then call nfsd_recalldelegation() to get rid of all extant
5644 	 * delegations.
5645 	 */
5646 	nfsd_recalldelegation(vp, p);
5647 }
5648 
5649 /*
5650  * Check for conflicting locks, etc. and then get rid of delegations.
5651  * (At one point I thought that I should get rid of delegations for any
5652  *  Setattr, since it could potentially disallow the I/O op (read or write)
5653  *  allowed by the delegation. However, Setattr Ops that aren't changing
5654  *  the size get a stateid of all 0s, so you can't tell if it is a delegation
5655  *  for the same client or a different one, so I decided to only get rid
5656  *  of delegations for other clients when the size is being changed.)
5657  * In general, a Setattr can disable NFS I/O Ops that are outstanding, such
5658  * as Write backs, even if there is no delegation, so it really isn't any
5659  * different?)
5660  */
5661 int
5662 nfsrv_checksetattr(vnode_t vp, struct nfsrv_descript *nd,
5663     nfsv4stateid_t *stateidp, struct nfsvattr *nvap, nfsattrbit_t *attrbitp,
5664     struct nfsexstuff *exp, NFSPROC_T *p)
5665 {
5666 	struct nfsstate st, *stp = &st;
5667 	struct nfslock lo, *lop = &lo;
5668 	int error = 0;
5669 	nfsquad_t clientid;
5670 
5671 	if (NFSISSET_ATTRBIT(attrbitp, NFSATTRBIT_SIZE)) {
5672 		stp->ls_flags = (NFSLCK_CHECK | NFSLCK_WRITEACCESS);
5673 		lop->lo_first = nvap->na_size;
5674 	} else {
5675 		stp->ls_flags = 0;
5676 		lop->lo_first = 0;
5677 	}
5678 	if (NFSISSET_ATTRBIT(attrbitp, NFSATTRBIT_OWNER) ||
5679 	    NFSISSET_ATTRBIT(attrbitp, NFSATTRBIT_OWNERGROUP) ||
5680 	    NFSISSET_ATTRBIT(attrbitp, NFSATTRBIT_MODE) ||
5681 	    NFSISSET_ATTRBIT(attrbitp, NFSATTRBIT_ACL))
5682 		stp->ls_flags |= NFSLCK_SETATTR;
5683 	if (stp->ls_flags == 0)
5684 		goto out;
5685 	lop->lo_end = NFS64BITSSET;
5686 	lop->lo_flags = NFSLCK_WRITE;
5687 	stp->ls_ownerlen = 0;
5688 	stp->ls_op = NULL;
5689 	stp->ls_uid = nd->nd_cred->cr_uid;
5690 	stp->ls_stateid.seqid = stateidp->seqid;
5691 	clientid.lval[0] = stp->ls_stateid.other[0] = stateidp->other[0];
5692 	clientid.lval[1] = stp->ls_stateid.other[1] = stateidp->other[1];
5693 	stp->ls_stateid.other[2] = stateidp->other[2];
5694 	error = nfsrv_lockctrl(vp, &stp, &lop, NULL, clientid,
5695 	    stateidp, exp, nd, p);
5696 
5697 out:
5698 	NFSEXITCODE2(error, nd);
5699 	return (error);
5700 }
5701 
5702 /*
5703  * Check for a write delegation and do a CBGETATTR if there is one, updating
5704  * the attributes, as required.
5705  * Should I return an error if I can't get the attributes? (For now, I'll
5706  * just return ok.
5707  */
5708 int
5709 nfsrv_checkgetattr(struct nfsrv_descript *nd, vnode_t vp,
5710     struct nfsvattr *nvap, nfsattrbit_t *attrbitp, NFSPROC_T *p)
5711 {
5712 	struct nfsstate *stp;
5713 	struct nfslockfile *lfp;
5714 	struct nfsclient *clp;
5715 	struct nfsvattr nva;
5716 	fhandle_t nfh;
5717 	int error = 0;
5718 	nfsattrbit_t cbbits;
5719 	u_quad_t delegfilerev;
5720 
5721 	NFSCBGETATTR_ATTRBIT(attrbitp, &cbbits);
5722 	if (!NFSNONZERO_ATTRBIT(&cbbits))
5723 		goto out;
5724 	if (nfsrv_writedelegcnt == 0)
5725 		goto out;
5726 
5727 	/*
5728 	 * Get the lock file structure.
5729 	 * (A return of -1 means no associated state, so return ok.)
5730 	 */
5731 	error = nfsrv_getlockfh(vp, NFSLCK_CHECK, NULL, &nfh, p);
5732 	NFSLOCKSTATE();
5733 	if (!error)
5734 		error = nfsrv_getlockfile(NFSLCK_CHECK, NULL, &lfp, &nfh, 0);
5735 	if (error) {
5736 		NFSUNLOCKSTATE();
5737 		if (error == -1)
5738 			error = 0;
5739 		goto out;
5740 	}
5741 
5742 	/*
5743 	 * Now, look for a write delegation.
5744 	 */
5745 	LIST_FOREACH(stp, &lfp->lf_deleg, ls_file) {
5746 		if (stp->ls_flags & NFSLCK_DELEGWRITE)
5747 			break;
5748 	}
5749 	if (stp == LIST_END(&lfp->lf_deleg)) {
5750 		NFSUNLOCKSTATE();
5751 		goto out;
5752 	}
5753 	clp = stp->ls_clp;
5754 
5755 	/* If the clientid is not confirmed, ignore the delegation. */
5756 	if (clp->lc_flags & LCL_NEEDSCONFIRM) {
5757 		NFSUNLOCKSTATE();
5758 		goto out;
5759 	}
5760 
5761 	delegfilerev = stp->ls_filerev;
5762 	/*
5763 	 * If the Write delegation was issued as a part of this Compound RPC
5764 	 * or if we have an Implied Clientid (used in a previous Op in this
5765 	 * compound) and it is the client the delegation was issued to,
5766 	 * just return ok.
5767 	 * I also assume that it is from the same client iff the network
5768 	 * host IP address is the same as the callback address. (Not
5769 	 * exactly correct by the RFC, but avoids a lot of Getattr
5770 	 * callbacks.)
5771 	 */
5772 	if (nd->nd_compref == stp->ls_compref ||
5773 	    ((nd->nd_flag & ND_IMPLIEDCLID) &&
5774 	     clp->lc_clientid.qval == nd->nd_clientid.qval) ||
5775 	     nfsaddr2_match(clp->lc_req.nr_nam, nd->nd_nam)) {
5776 		NFSUNLOCKSTATE();
5777 		goto out;
5778 	}
5779 
5780 	/*
5781 	 * We are now done with the delegation state structure,
5782 	 * so the statelock can be released and we can now tsleep().
5783 	 */
5784 
5785 	/*
5786 	 * Now, we must do the CB Getattr callback, to see if Change or Size
5787 	 * has changed.
5788 	 */
5789 	if (clp->lc_expiry >= NFSD_MONOSEC) {
5790 		NFSUNLOCKSTATE();
5791 		NFSVNO_ATTRINIT(&nva);
5792 		nva.na_filerev = NFS64BITSSET;
5793 		error = nfsrv_docallback(clp, NFSV4OP_CBGETATTR, NULL,
5794 		    0, &nfh, &nva, &cbbits, 0, p);
5795 		if (!error) {
5796 			if ((nva.na_filerev != NFS64BITSSET &&
5797 			    nva.na_filerev > delegfilerev) ||
5798 			    (NFSVNO_ISSETSIZE(&nva) &&
5799 			     nva.na_size != nvap->na_size)) {
5800 				error = nfsvno_updfilerev(vp, nvap, nd, p);
5801 				if (NFSVNO_ISSETSIZE(&nva))
5802 					nvap->na_size = nva.na_size;
5803 			}
5804 		} else
5805 			error = 0;	/* Ignore callback errors for now. */
5806 	} else {
5807 		NFSUNLOCKSTATE();
5808 	}
5809 
5810 out:
5811 	NFSEXITCODE2(error, nd);
5812 	return (error);
5813 }
5814 
5815 /*
5816  * This function looks for openowners that haven't had any opens for
5817  * a while and throws them away. Called by an nfsd when NFSNSF_NOOPENS
5818  * is set.
5819  */
5820 void
5821 nfsrv_throwawayopens(NFSPROC_T *p)
5822 {
5823 	struct nfsclient *clp, *nclp;
5824 	struct nfsstate *stp, *nstp;
5825 	int i;
5826 
5827 	NFSLOCKSTATE();
5828 	nfsrv_stablefirst.nsf_flags &= ~NFSNSF_NOOPENS;
5829 	/*
5830 	 * For each client...
5831 	 */
5832 	for (i = 0; i < nfsrv_clienthashsize; i++) {
5833 	    LIST_FOREACH_SAFE(clp, &nfsclienthash[i], lc_hash, nclp) {
5834 		LIST_FOREACH_SAFE(stp, &clp->lc_open, ls_list, nstp) {
5835 			if (LIST_EMPTY(&stp->ls_open) &&
5836 			    (stp->ls_noopens > NFSNOOPEN ||
5837 			     (nfsrv_openpluslock * 2) >
5838 			     nfsrv_v4statelimit))
5839 				nfsrv_freeopenowner(stp, 0, p);
5840 		}
5841 	    }
5842 	}
5843 	NFSUNLOCKSTATE();
5844 }
5845 
5846 /*
5847  * This function checks to see if the credentials are the same.
5848  * Returns 1 for not same, 0 otherwise.
5849  */
5850 static int
5851 nfsrv_notsamecredname(struct nfsrv_descript *nd, struct nfsclient *clp)
5852 {
5853 
5854 	if (nd->nd_flag & ND_GSS) {
5855 		if (!(clp->lc_flags & LCL_GSS))
5856 			return (1);
5857 		if (clp->lc_flags & LCL_NAME) {
5858 			if (nd->nd_princlen != clp->lc_namelen ||
5859 			    NFSBCMP(nd->nd_principal, clp->lc_name,
5860 				clp->lc_namelen))
5861 				return (1);
5862 			else
5863 				return (0);
5864 		}
5865 		if (nd->nd_cred->cr_uid == clp->lc_uid)
5866 			return (0);
5867 		else
5868 			return (1);
5869 	} else if (clp->lc_flags & LCL_GSS)
5870 		return (1);
5871 	/*
5872 	 * For AUTH_SYS, allow the same uid or root. (This is underspecified
5873 	 * in RFC3530, which talks about principals, but doesn't say anything
5874 	 * about uids for AUTH_SYS.)
5875 	 */
5876 	if (nd->nd_cred->cr_uid == clp->lc_uid || nd->nd_cred->cr_uid == 0)
5877 		return (0);
5878 	else
5879 		return (1);
5880 }
5881 
5882 /*
5883  * Calculate the lease expiry time.
5884  */
5885 static time_t
5886 nfsrv_leaseexpiry(void)
5887 {
5888 
5889 	if (nfsrv_stablefirst.nsf_eograce > NFSD_MONOSEC)
5890 		return (NFSD_MONOSEC + 2 * (nfsrv_lease + NFSRV_LEASEDELTA));
5891 	return (NFSD_MONOSEC + nfsrv_lease + NFSRV_LEASEDELTA);
5892 }
5893 
5894 /*
5895  * Delay the delegation timeout as far as ls_delegtimelimit, as required.
5896  */
5897 static void
5898 nfsrv_delaydelegtimeout(struct nfsstate *stp)
5899 {
5900 
5901 	if ((stp->ls_flags & NFSLCK_DELEGRECALL) == 0)
5902 		return;
5903 
5904 	if ((stp->ls_delegtime + 15) > NFSD_MONOSEC &&
5905 	    stp->ls_delegtime < stp->ls_delegtimelimit) {
5906 		stp->ls_delegtime += nfsrv_lease;
5907 		if (stp->ls_delegtime > stp->ls_delegtimelimit)
5908 			stp->ls_delegtime = stp->ls_delegtimelimit;
5909 	}
5910 }
5911 
5912 /*
5913  * This function checks to see if there is any other state associated
5914  * with the openowner for this Open.
5915  * It returns 1 if there is no other state, 0 otherwise.
5916  */
5917 static int
5918 nfsrv_nootherstate(struct nfsstate *stp)
5919 {
5920 	struct nfsstate *tstp;
5921 
5922 	LIST_FOREACH(tstp, &stp->ls_openowner->ls_open, ls_list) {
5923 		if (tstp != stp || !LIST_EMPTY(&tstp->ls_lock))
5924 			return (0);
5925 	}
5926 	return (1);
5927 }
5928 
5929 /*
5930  * Create a list of lock deltas (changes to local byte range locking
5931  * that can be rolled back using the list) and apply the changes via
5932  * nfsvno_advlock(). Optionally, lock the list. It is expected that either
5933  * the rollback or update function will be called after this.
5934  * It returns an error (and rolls back, as required), if any nfsvno_advlock()
5935  * call fails. If it returns an error, it will unlock the list.
5936  */
5937 static int
5938 nfsrv_locallock(vnode_t vp, struct nfslockfile *lfp, int flags,
5939     uint64_t first, uint64_t end, struct nfslockconflict *cfp, NFSPROC_T *p)
5940 {
5941 	struct nfslock *lop, *nlop;
5942 	int error = 0;
5943 
5944 	/* Loop through the list of locks. */
5945 	lop = LIST_FIRST(&lfp->lf_locallock);
5946 	while (first < end && lop != NULL) {
5947 		nlop = LIST_NEXT(lop, lo_lckowner);
5948 		if (first >= lop->lo_end) {
5949 			/* not there yet */
5950 			lop = nlop;
5951 		} else if (first < lop->lo_first) {
5952 			/* new one starts before entry in list */
5953 			if (end <= lop->lo_first) {
5954 				/* no overlap between old and new */
5955 				error = nfsrv_dolocal(vp, lfp, flags,
5956 				    NFSLCK_UNLOCK, first, end, cfp, p);
5957 				if (error != 0)
5958 					break;
5959 				first = end;
5960 			} else {
5961 				/* handle fragment overlapped with new one */
5962 				error = nfsrv_dolocal(vp, lfp, flags,
5963 				    NFSLCK_UNLOCK, first, lop->lo_first, cfp,
5964 				    p);
5965 				if (error != 0)
5966 					break;
5967 				first = lop->lo_first;
5968 			}
5969 		} else {
5970 			/* new one overlaps this entry in list */
5971 			if (end <= lop->lo_end) {
5972 				/* overlaps all of new one */
5973 				error = nfsrv_dolocal(vp, lfp, flags,
5974 				    lop->lo_flags, first, end, cfp, p);
5975 				if (error != 0)
5976 					break;
5977 				first = end;
5978 			} else {
5979 				/* handle fragment overlapped with new one */
5980 				error = nfsrv_dolocal(vp, lfp, flags,
5981 				    lop->lo_flags, first, lop->lo_end, cfp, p);
5982 				if (error != 0)
5983 					break;
5984 				first = lop->lo_end;
5985 				lop = nlop;
5986 			}
5987 		}
5988 	}
5989 	if (first < end && error == 0)
5990 		/* handle fragment past end of list */
5991 		error = nfsrv_dolocal(vp, lfp, flags, NFSLCK_UNLOCK, first,
5992 		    end, cfp, p);
5993 
5994 	NFSEXITCODE(error);
5995 	return (error);
5996 }
5997 
5998 /*
5999  * Local lock unlock. Unlock all byte ranges that are no longer locked
6000  * by NFSv4. To do this, unlock any subranges of first-->end that
6001  * do not overlap with the byte ranges of any lock in the lfp->lf_lock
6002  * list. This list has all locks for the file held by other
6003  * <clientid, lockowner> tuples. The list is ordered by increasing
6004  * lo_first value, but may have entries that overlap each other, for
6005  * the case of read locks.
6006  */
6007 static void
6008 nfsrv_localunlock(vnode_t vp, struct nfslockfile *lfp, uint64_t init_first,
6009     uint64_t init_end, NFSPROC_T *p)
6010 {
6011 	struct nfslock *lop;
6012 	uint64_t first, end, prevfirst __unused;
6013 
6014 	first = init_first;
6015 	end = init_end;
6016 	while (first < init_end) {
6017 		/* Loop through all nfs locks, adjusting first and end */
6018 		prevfirst = 0;
6019 		LIST_FOREACH(lop, &lfp->lf_lock, lo_lckfile) {
6020 			KASSERT(prevfirst <= lop->lo_first,
6021 			    ("nfsv4 locks out of order"));
6022 			KASSERT(lop->lo_first < lop->lo_end,
6023 			    ("nfsv4 bogus lock"));
6024 			prevfirst = lop->lo_first;
6025 			if (first >= lop->lo_first &&
6026 			    first < lop->lo_end)
6027 				/*
6028 				 * Overlaps with initial part, so trim
6029 				 * off that initial part by moving first past
6030 				 * it.
6031 				 */
6032 				first = lop->lo_end;
6033 			else if (end > lop->lo_first &&
6034 			    lop->lo_first > first) {
6035 				/*
6036 				 * This lock defines the end of the
6037 				 * segment to unlock, so set end to the
6038 				 * start of it and break out of the loop.
6039 				 */
6040 				end = lop->lo_first;
6041 				break;
6042 			}
6043 			if (first >= end)
6044 				/*
6045 				 * There is no segment left to do, so
6046 				 * break out of this loop and then exit
6047 				 * the outer while() since first will be set
6048 				 * to end, which must equal init_end here.
6049 				 */
6050 				break;
6051 		}
6052 		if (first < end) {
6053 			/* Unlock this segment */
6054 			(void) nfsrv_dolocal(vp, lfp, NFSLCK_UNLOCK,
6055 			    NFSLCK_READ, first, end, NULL, p);
6056 			nfsrv_locallock_commit(lfp, NFSLCK_UNLOCK,
6057 			    first, end);
6058 		}
6059 		/*
6060 		 * Now move past this segment and look for any further
6061 		 * segment in the range, if there is one.
6062 		 */
6063 		first = end;
6064 		end = init_end;
6065 	}
6066 }
6067 
6068 /*
6069  * Do the local lock operation and update the rollback list, as required.
6070  * Perform the rollback and return the error if nfsvno_advlock() fails.
6071  */
6072 static int
6073 nfsrv_dolocal(vnode_t vp, struct nfslockfile *lfp, int flags, int oldflags,
6074     uint64_t first, uint64_t end, struct nfslockconflict *cfp, NFSPROC_T *p)
6075 {
6076 	struct nfsrollback *rlp;
6077 	int error = 0, ltype, oldltype;
6078 
6079 	if (flags & NFSLCK_WRITE)
6080 		ltype = F_WRLCK;
6081 	else if (flags & NFSLCK_READ)
6082 		ltype = F_RDLCK;
6083 	else
6084 		ltype = F_UNLCK;
6085 	if (oldflags & NFSLCK_WRITE)
6086 		oldltype = F_WRLCK;
6087 	else if (oldflags & NFSLCK_READ)
6088 		oldltype = F_RDLCK;
6089 	else
6090 		oldltype = F_UNLCK;
6091 	if (ltype == oldltype || (oldltype == F_WRLCK && ltype == F_RDLCK))
6092 		/* nothing to do */
6093 		goto out;
6094 	error = nfsvno_advlock(vp, ltype, first, end, p);
6095 	if (error != 0) {
6096 		if (cfp != NULL) {
6097 			cfp->cl_clientid.lval[0] = 0;
6098 			cfp->cl_clientid.lval[1] = 0;
6099 			cfp->cl_first = 0;
6100 			cfp->cl_end = NFS64BITSSET;
6101 			cfp->cl_flags = NFSLCK_WRITE;
6102 			cfp->cl_ownerlen = 5;
6103 			NFSBCOPY("LOCAL", cfp->cl_owner, 5);
6104 		}
6105 		nfsrv_locallock_rollback(vp, lfp, p);
6106 	} else if (ltype != F_UNLCK) {
6107 		rlp = malloc(sizeof (struct nfsrollback), M_NFSDROLLBACK,
6108 		    M_WAITOK);
6109 		rlp->rlck_first = first;
6110 		rlp->rlck_end = end;
6111 		rlp->rlck_type = oldltype;
6112 		LIST_INSERT_HEAD(&lfp->lf_rollback, rlp, rlck_list);
6113 	}
6114 
6115 out:
6116 	NFSEXITCODE(error);
6117 	return (error);
6118 }
6119 
6120 /*
6121  * Roll back local lock changes and free up the rollback list.
6122  */
6123 static void
6124 nfsrv_locallock_rollback(vnode_t vp, struct nfslockfile *lfp, NFSPROC_T *p)
6125 {
6126 	struct nfsrollback *rlp, *nrlp;
6127 
6128 	LIST_FOREACH_SAFE(rlp, &lfp->lf_rollback, rlck_list, nrlp) {
6129 		(void) nfsvno_advlock(vp, rlp->rlck_type, rlp->rlck_first,
6130 		    rlp->rlck_end, p);
6131 		free(rlp, M_NFSDROLLBACK);
6132 	}
6133 	LIST_INIT(&lfp->lf_rollback);
6134 }
6135 
6136 /*
6137  * Update local lock list and delete rollback list (ie now committed to the
6138  * local locks). Most of the work is done by the internal function.
6139  */
6140 static void
6141 nfsrv_locallock_commit(struct nfslockfile *lfp, int flags, uint64_t first,
6142     uint64_t end)
6143 {
6144 	struct nfsrollback *rlp, *nrlp;
6145 	struct nfslock *new_lop, *other_lop;
6146 
6147 	new_lop = malloc(sizeof (struct nfslock), M_NFSDLOCK, M_WAITOK);
6148 	if (flags & (NFSLCK_READ | NFSLCK_WRITE))
6149 		other_lop = malloc(sizeof (struct nfslock), M_NFSDLOCK,
6150 		    M_WAITOK);
6151 	else
6152 		other_lop = NULL;
6153 	new_lop->lo_flags = flags;
6154 	new_lop->lo_first = first;
6155 	new_lop->lo_end = end;
6156 	nfsrv_updatelock(NULL, &new_lop, &other_lop, lfp);
6157 	if (new_lop != NULL)
6158 		free(new_lop, M_NFSDLOCK);
6159 	if (other_lop != NULL)
6160 		free(other_lop, M_NFSDLOCK);
6161 
6162 	/* and get rid of the rollback list */
6163 	LIST_FOREACH_SAFE(rlp, &lfp->lf_rollback, rlck_list, nrlp)
6164 		free(rlp, M_NFSDROLLBACK);
6165 	LIST_INIT(&lfp->lf_rollback);
6166 }
6167 
6168 /*
6169  * Lock the struct nfslockfile for local lock updating.
6170  */
6171 static void
6172 nfsrv_locklf(struct nfslockfile *lfp)
6173 {
6174 	int gotlock;
6175 
6176 	/* lf_usecount ensures *lfp won't be free'd */
6177 	lfp->lf_usecount++;
6178 	do {
6179 		gotlock = nfsv4_lock(&lfp->lf_locallock_lck, 1, NULL,
6180 		    NFSSTATEMUTEXPTR, NULL);
6181 	} while (gotlock == 0);
6182 	lfp->lf_usecount--;
6183 }
6184 
6185 /*
6186  * Unlock the struct nfslockfile after local lock updating.
6187  */
6188 static void
6189 nfsrv_unlocklf(struct nfslockfile *lfp)
6190 {
6191 
6192 	nfsv4_unlock(&lfp->lf_locallock_lck, 0);
6193 }
6194 
6195 /*
6196  * Clear out all state for the NFSv4 server.
6197  * Must be called by a thread that can sleep when no nfsds are running.
6198  */
6199 void
6200 nfsrv_throwawayallstate(NFSPROC_T *p)
6201 {
6202 	struct nfsclient *clp, *nclp;
6203 	struct nfslockfile *lfp, *nlfp;
6204 	int i;
6205 
6206 	/*
6207 	 * For each client, clean out the state and then free the structure.
6208 	 */
6209 	for (i = 0; i < nfsrv_clienthashsize; i++) {
6210 		LIST_FOREACH_SAFE(clp, &nfsclienthash[i], lc_hash, nclp) {
6211 			nfsrv_cleanclient(clp, p);
6212 			nfsrv_freedeleglist(&clp->lc_deleg);
6213 			nfsrv_freedeleglist(&clp->lc_olddeleg);
6214 			free(clp->lc_stateid, M_NFSDCLIENT);
6215 			free(clp, M_NFSDCLIENT);
6216 		}
6217 	}
6218 
6219 	/*
6220 	 * Also, free up any remaining lock file structures.
6221 	 */
6222 	for (i = 0; i < nfsrv_lockhashsize; i++) {
6223 		LIST_FOREACH_SAFE(lfp, &nfslockhash[i], lf_hash, nlfp) {
6224 			printf("nfsd unload: fnd a lock file struct\n");
6225 			nfsrv_freenfslockfile(lfp);
6226 		}
6227 	}
6228 
6229 	/* And get rid of the deviceid structures and layouts. */
6230 	nfsrv_freealllayoutsanddevids();
6231 }
6232 
6233 /*
6234  * Check the sequence# for the session and slot provided as an argument.
6235  * Also, renew the lease if the session will return NFS_OK.
6236  */
6237 int
6238 nfsrv_checksequence(struct nfsrv_descript *nd, uint32_t sequenceid,
6239     uint32_t *highest_slotidp, uint32_t *target_highest_slotidp, int cache_this,
6240     uint32_t *sflagsp, NFSPROC_T *p)
6241 {
6242 	struct nfsdsession *sep;
6243 	struct nfssessionhash *shp;
6244 	int error;
6245 
6246 	shp = NFSSESSIONHASH(nd->nd_sessionid);
6247 	NFSLOCKSESSION(shp);
6248 	sep = nfsrv_findsession(nd->nd_sessionid);
6249 	if (sep == NULL) {
6250 		NFSUNLOCKSESSION(shp);
6251 		return (NFSERR_BADSESSION);
6252 	}
6253 	error = nfsv4_seqsession(sequenceid, nd->nd_slotid, *highest_slotidp,
6254 	    sep->sess_slots, NULL, NFSV4_SLOTS - 1);
6255 	if (error != 0) {
6256 		NFSUNLOCKSESSION(shp);
6257 		return (error);
6258 	}
6259 	if (cache_this != 0)
6260 		nd->nd_flag |= ND_SAVEREPLY;
6261 	/* Renew the lease. */
6262 	sep->sess_clp->lc_expiry = nfsrv_leaseexpiry();
6263 	nd->nd_clientid.qval = sep->sess_clp->lc_clientid.qval;
6264 	nd->nd_flag |= ND_IMPLIEDCLID;
6265 
6266 	/* Save maximum request and reply sizes. */
6267 	nd->nd_maxreq = sep->sess_maxreq;
6268 	nd->nd_maxresp = sep->sess_maxresp;
6269 
6270 	*sflagsp = 0;
6271 	if (sep->sess_clp->lc_req.nr_client == NULL ||
6272 	    (sep->sess_clp->lc_flags & LCL_CBDOWN) != 0)
6273 		*sflagsp |= NFSV4SEQ_CBPATHDOWN;
6274 	NFSUNLOCKSESSION(shp);
6275 	if (error == NFSERR_EXPIRED) {
6276 		*sflagsp |= NFSV4SEQ_EXPIREDALLSTATEREVOKED;
6277 		error = 0;
6278 	} else if (error == NFSERR_ADMINREVOKED) {
6279 		*sflagsp |= NFSV4SEQ_ADMINSTATEREVOKED;
6280 		error = 0;
6281 	}
6282 	*highest_slotidp = *target_highest_slotidp = NFSV4_SLOTS - 1;
6283 	return (0);
6284 }
6285 
6286 /*
6287  * Check/set reclaim complete for this session/clientid.
6288  */
6289 int
6290 nfsrv_checkreclaimcomplete(struct nfsrv_descript *nd, int onefs)
6291 {
6292 	struct nfsdsession *sep;
6293 	struct nfssessionhash *shp;
6294 	int error = 0;
6295 
6296 	shp = NFSSESSIONHASH(nd->nd_sessionid);
6297 	NFSLOCKSTATE();
6298 	NFSLOCKSESSION(shp);
6299 	sep = nfsrv_findsession(nd->nd_sessionid);
6300 	if (sep == NULL) {
6301 		NFSUNLOCKSESSION(shp);
6302 		NFSUNLOCKSTATE();
6303 		return (NFSERR_BADSESSION);
6304 	}
6305 
6306 	if (onefs != 0)
6307 		sep->sess_clp->lc_flags |= LCL_RECLAIMONEFS;
6308 		/* Check to see if reclaim complete has already happened. */
6309 	else if ((sep->sess_clp->lc_flags & LCL_RECLAIMCOMPLETE) != 0)
6310 		error = NFSERR_COMPLETEALREADY;
6311 	else {
6312 		sep->sess_clp->lc_flags |= LCL_RECLAIMCOMPLETE;
6313 		nfsrv_markreclaim(sep->sess_clp);
6314 	}
6315 	NFSUNLOCKSESSION(shp);
6316 	NFSUNLOCKSTATE();
6317 	return (error);
6318 }
6319 
6320 /*
6321  * Cache the reply in a session slot.
6322  */
6323 void
6324 nfsrv_cache_session(struct nfsrv_descript *nd, struct mbuf **m)
6325 {
6326 	struct nfsdsession *sep;
6327 	struct nfssessionhash *shp;
6328 	char *buf, *cp;
6329 #ifdef INET
6330 	struct sockaddr_in *sin;
6331 #endif
6332 #ifdef INET6
6333 	struct sockaddr_in6 *sin6;
6334 #endif
6335 
6336 	shp = NFSSESSIONHASH(nd->nd_sessionid);
6337 	NFSLOCKSESSION(shp);
6338 	sep = nfsrv_findsession(nd->nd_sessionid);
6339 	if (sep == NULL) {
6340 		NFSUNLOCKSESSION(shp);
6341 		if ((nfsrv_stablefirst.nsf_flags & NFSNSF_GRACEOVER) != 0) {
6342 			buf = malloc(INET6_ADDRSTRLEN, M_TEMP, M_WAITOK);
6343 			switch (nd->nd_nam->sa_family) {
6344 #ifdef INET
6345 			case AF_INET:
6346 				sin = (struct sockaddr_in *)nd->nd_nam;
6347 				cp = inet_ntop(sin->sin_family,
6348 				    &sin->sin_addr.s_addr, buf,
6349 				    INET6_ADDRSTRLEN);
6350 				break;
6351 #endif
6352 #ifdef INET6
6353 			case AF_INET6:
6354 				sin6 = (struct sockaddr_in6 *)nd->nd_nam;
6355 				cp = inet_ntop(sin6->sin6_family,
6356 				    &sin6->sin6_addr, buf, INET6_ADDRSTRLEN);
6357 				break;
6358 #endif
6359 			default:
6360 				cp = NULL;
6361 			}
6362 			if (cp != NULL)
6363 				printf("nfsrv_cache_session: no session "
6364 				    "IPaddr=%s\n", cp);
6365 			else
6366 				printf("nfsrv_cache_session: no session\n");
6367 			free(buf, M_TEMP);
6368 		}
6369 		m_freem(*m);
6370 		return;
6371 	}
6372 	nfsv4_seqsess_cacherep(nd->nd_slotid, sep->sess_slots, nd->nd_repstat,
6373 	    m);
6374 	NFSUNLOCKSESSION(shp);
6375 }
6376 
6377 /*
6378  * Search for a session that matches the sessionid.
6379  */
6380 static struct nfsdsession *
6381 nfsrv_findsession(uint8_t *sessionid)
6382 {
6383 	struct nfsdsession *sep;
6384 	struct nfssessionhash *shp;
6385 
6386 	shp = NFSSESSIONHASH(sessionid);
6387 	LIST_FOREACH(sep, &shp->list, sess_hash) {
6388 		if (!NFSBCMP(sessionid, sep->sess_sessionid, NFSX_V4SESSIONID))
6389 			break;
6390 	}
6391 	return (sep);
6392 }
6393 
6394 /*
6395  * Destroy a session.
6396  */
6397 int
6398 nfsrv_destroysession(struct nfsrv_descript *nd, uint8_t *sessionid)
6399 {
6400 	int error, igotlock, samesess;
6401 
6402 	samesess = 0;
6403 	if (!NFSBCMP(sessionid, nd->nd_sessionid, NFSX_V4SESSIONID) &&
6404 	    (nd->nd_flag & ND_HASSEQUENCE) != 0) {
6405 		samesess = 1;
6406 		if ((nd->nd_flag & ND_LASTOP) == 0)
6407 			return (NFSERR_BADSESSION);
6408 	}
6409 
6410 	/* Lock out other nfsd threads */
6411 	NFSLOCKV4ROOTMUTEX();
6412 	nfsv4_relref(&nfsv4rootfs_lock);
6413 	do {
6414 		igotlock = nfsv4_lock(&nfsv4rootfs_lock, 1, NULL,
6415 		    NFSV4ROOTLOCKMUTEXPTR, NULL);
6416 	} while (igotlock == 0);
6417 	NFSUNLOCKV4ROOTMUTEX();
6418 
6419 	error = nfsrv_freesession(NULL, sessionid);
6420 	if (error == 0 && samesess != 0)
6421 		nd->nd_flag &= ~ND_HASSEQUENCE;
6422 
6423 	NFSLOCKV4ROOTMUTEX();
6424 	nfsv4_unlock(&nfsv4rootfs_lock, 1);
6425 	NFSUNLOCKV4ROOTMUTEX();
6426 	return (error);
6427 }
6428 
6429 /*
6430  * Bind a connection to a session.
6431  * For now, only certain variants are supported, since the current session
6432  * structure can only handle a single backchannel entry, which will be
6433  * applied to all connections if it is set.
6434  */
6435 int
6436 nfsrv_bindconnsess(struct nfsrv_descript *nd, uint8_t *sessionid, int *foreaftp)
6437 {
6438 	struct nfssessionhash *shp;
6439 	struct nfsdsession *sep;
6440 	struct nfsclient *clp;
6441 	SVCXPRT *savxprt;
6442 	int error;
6443 
6444 	error = 0;
6445 	savxprt = NULL;
6446 	shp = NFSSESSIONHASH(sessionid);
6447 	NFSLOCKSTATE();
6448 	NFSLOCKSESSION(shp);
6449 	sep = nfsrv_findsession(sessionid);
6450 	if (sep != NULL) {
6451 		clp = sep->sess_clp;
6452 		if (*foreaftp == NFSCDFC4_BACK ||
6453 		    *foreaftp == NFSCDFC4_BACK_OR_BOTH ||
6454 		    *foreaftp == NFSCDFC4_FORE_OR_BOTH) {
6455 			/* Try to set up a backchannel. */
6456 			if (clp->lc_req.nr_client == NULL) {
6457 				NFSD_DEBUG(2, "nfsrv_bindconnsess: acquire "
6458 				    "backchannel\n");
6459 				clp->lc_req.nr_client = (struct __rpc_client *)
6460 				    clnt_bck_create(nd->nd_xprt->xp_socket,
6461 				    sep->sess_cbprogram, NFSV4_CBVERS);
6462 			}
6463 			if (clp->lc_req.nr_client != NULL) {
6464 				NFSD_DEBUG(2, "nfsrv_bindconnsess: set up "
6465 				    "backchannel\n");
6466 				savxprt = sep->sess_cbsess.nfsess_xprt;
6467 				SVC_ACQUIRE(nd->nd_xprt);
6468 				CLNT_ACQUIRE(clp->lc_req.nr_client);
6469 				nd->nd_xprt->xp_p2 = clp->lc_req.nr_client;
6470 				/* Disable idle timeout. */
6471 				nd->nd_xprt->xp_idletimeout = 0;
6472 				sep->sess_cbsess.nfsess_xprt = nd->nd_xprt;
6473 				sep->sess_crflags |= NFSV4CRSESS_CONNBACKCHAN;
6474 				clp->lc_flags |= LCL_DONEBINDCONN |
6475 				    LCL_NEEDSCBNULL;
6476 				clp->lc_flags &= ~LCL_CBDOWN;
6477 				if (*foreaftp == NFSCDFS4_BACK)
6478 					*foreaftp = NFSCDFS4_BACK;
6479 				else
6480 					*foreaftp = NFSCDFS4_BOTH;
6481 			} else if (*foreaftp != NFSCDFC4_BACK) {
6482 				NFSD_DEBUG(2, "nfsrv_bindconnsess: can't set "
6483 				    "up backchannel\n");
6484 				sep->sess_crflags &= ~NFSV4CRSESS_CONNBACKCHAN;
6485 				clp->lc_flags |= LCL_DONEBINDCONN;
6486 				*foreaftp = NFSCDFS4_FORE;
6487 			} else {
6488 				error = NFSERR_NOTSUPP;
6489 				printf("nfsrv_bindconnsess: Can't add "
6490 				    "backchannel\n");
6491 			}
6492 		} else {
6493 			NFSD_DEBUG(2, "nfsrv_bindconnsess: Set forechannel\n");
6494 			clp->lc_flags |= LCL_DONEBINDCONN;
6495 			*foreaftp = NFSCDFS4_FORE;
6496 		}
6497 	} else
6498 		error = NFSERR_BADSESSION;
6499 	NFSUNLOCKSESSION(shp);
6500 	NFSUNLOCKSTATE();
6501 	if (savxprt != NULL)
6502 		SVC_RELEASE(savxprt);
6503 	return (error);
6504 }
6505 
6506 /*
6507  * Free up a session structure.
6508  */
6509 static int
6510 nfsrv_freesession(struct nfsdsession *sep, uint8_t *sessionid)
6511 {
6512 	struct nfssessionhash *shp;
6513 	int i;
6514 
6515 	NFSLOCKSTATE();
6516 	if (sep == NULL) {
6517 		shp = NFSSESSIONHASH(sessionid);
6518 		NFSLOCKSESSION(shp);
6519 		sep = nfsrv_findsession(sessionid);
6520 	} else {
6521 		shp = NFSSESSIONHASH(sep->sess_sessionid);
6522 		NFSLOCKSESSION(shp);
6523 	}
6524 	if (sep != NULL) {
6525 		sep->sess_refcnt--;
6526 		if (sep->sess_refcnt > 0) {
6527 			NFSUNLOCKSESSION(shp);
6528 			NFSUNLOCKSTATE();
6529 			return (NFSERR_BACKCHANBUSY);
6530 		}
6531 		LIST_REMOVE(sep, sess_hash);
6532 		LIST_REMOVE(sep, sess_list);
6533 	}
6534 	NFSUNLOCKSESSION(shp);
6535 	NFSUNLOCKSTATE();
6536 	if (sep == NULL)
6537 		return (NFSERR_BADSESSION);
6538 	for (i = 0; i < NFSV4_SLOTS; i++)
6539 		if (sep->sess_slots[i].nfssl_reply != NULL)
6540 			m_freem(sep->sess_slots[i].nfssl_reply);
6541 	if (sep->sess_cbsess.nfsess_xprt != NULL)
6542 		SVC_RELEASE(sep->sess_cbsess.nfsess_xprt);
6543 	free(sep, M_NFSDSESSION);
6544 	return (0);
6545 }
6546 
6547 /*
6548  * Free a stateid.
6549  * RFC5661 says that it should fail when there are associated opens, locks
6550  * or delegations. Since stateids represent opens, I don't see how you can
6551  * free an open stateid (it will be free'd when closed), so this function
6552  * only works for lock stateids (freeing the lock_owner) or delegations.
6553  */
6554 int
6555 nfsrv_freestateid(struct nfsrv_descript *nd, nfsv4stateid_t *stateidp,
6556     NFSPROC_T *p)
6557 {
6558 	struct nfsclient *clp;
6559 	struct nfsstate *stp;
6560 	int error;
6561 
6562 	NFSLOCKSTATE();
6563 	/*
6564 	 * Look up the stateid
6565 	 */
6566 	error = nfsrv_getclient((nfsquad_t)((u_quad_t)0), CLOPS_RENEW, &clp,
6567 	    NULL, (nfsquad_t)((u_quad_t)0), 0, nd, p);
6568 	if (error == 0) {
6569 		/* First, check for a delegation. */
6570 		LIST_FOREACH(stp, &clp->lc_deleg, ls_list) {
6571 			if (!NFSBCMP(stp->ls_stateid.other, stateidp->other,
6572 			    NFSX_STATEIDOTHER))
6573 				break;
6574 		}
6575 		if (stp != NULL) {
6576 			nfsrv_freedeleg(stp);
6577 			NFSUNLOCKSTATE();
6578 			return (error);
6579 		}
6580 	}
6581 	/* Not a delegation, try for a lock_owner. */
6582 	if (error == 0)
6583 		error = nfsrv_getstate(clp, stateidp, 0, &stp);
6584 	if (error == 0 && ((stp->ls_flags & (NFSLCK_OPEN | NFSLCK_DELEGREAD |
6585 	    NFSLCK_DELEGWRITE)) != 0 || (stp->ls_flags & NFSLCK_LOCK) == 0))
6586 		/* Not a lock_owner stateid. */
6587 		error = NFSERR_LOCKSHELD;
6588 	if (error == 0 && !LIST_EMPTY(&stp->ls_lock))
6589 		error = NFSERR_LOCKSHELD;
6590 	if (error == 0)
6591 		nfsrv_freelockowner(stp, NULL, 0, p);
6592 	NFSUNLOCKSTATE();
6593 	return (error);
6594 }
6595 
6596 /*
6597  * Test a stateid.
6598  */
6599 int
6600 nfsrv_teststateid(struct nfsrv_descript *nd, nfsv4stateid_t *stateidp,
6601     NFSPROC_T *p)
6602 {
6603 	struct nfsclient *clp;
6604 	struct nfsstate *stp;
6605 	int error;
6606 
6607 	NFSLOCKSTATE();
6608 	/*
6609 	 * Look up the stateid
6610 	 */
6611 	error = nfsrv_getclient((nfsquad_t)((u_quad_t)0), CLOPS_RENEW, &clp,
6612 	    NULL, (nfsquad_t)((u_quad_t)0), 0, nd, p);
6613 	if (error == 0)
6614 		error = nfsrv_getstate(clp, stateidp, 0, &stp);
6615 	if (error == 0 && stateidp->seqid != 0 &&
6616 	    SEQ_LT(stateidp->seqid, stp->ls_stateid.seqid))
6617 		error = NFSERR_OLDSTATEID;
6618 	NFSUNLOCKSTATE();
6619 	return (error);
6620 }
6621 
6622 /*
6623  * Generate the xdr for an NFSv4.1 CBSequence Operation.
6624  */
6625 static int
6626 nfsv4_setcbsequence(struct nfsrv_descript *nd, struct nfsclient *clp,
6627     int dont_replycache, struct nfsdsession **sepp, int *slotposp)
6628 {
6629 	struct nfsdsession *sep;
6630 	uint32_t *tl, slotseq = 0;
6631 	int maxslot;
6632 	uint8_t sessionid[NFSX_V4SESSIONID];
6633 	int error;
6634 
6635 	error = nfsv4_getcbsession(clp, sepp);
6636 	if (error != 0)
6637 		return (error);
6638 	sep = *sepp;
6639 	(void)nfsv4_sequencelookup(NULL, &sep->sess_cbsess, slotposp, &maxslot,
6640 	    &slotseq, sessionid);
6641 	KASSERT(maxslot >= 0, ("nfsv4_setcbsequence neg maxslot"));
6642 
6643 	/* Build the Sequence arguments. */
6644 	NFSM_BUILD(tl, uint32_t *, NFSX_V4SESSIONID + 5 * NFSX_UNSIGNED);
6645 	bcopy(sessionid, tl, NFSX_V4SESSIONID);
6646 	tl += NFSX_V4SESSIONID / NFSX_UNSIGNED;
6647 	nd->nd_slotseq = tl;
6648 	*tl++ = txdr_unsigned(slotseq);
6649 	*tl++ = txdr_unsigned(*slotposp);
6650 	*tl++ = txdr_unsigned(maxslot);
6651 	if (dont_replycache == 0)
6652 		*tl++ = newnfs_true;
6653 	else
6654 		*tl++ = newnfs_false;
6655 	*tl = 0;			/* No referring call list, for now. */
6656 	nd->nd_flag |= ND_HASSEQUENCE;
6657 	return (0);
6658 }
6659 
6660 /*
6661  * Get a session for the callback.
6662  */
6663 static int
6664 nfsv4_getcbsession(struct nfsclient *clp, struct nfsdsession **sepp)
6665 {
6666 	struct nfsdsession *sep;
6667 
6668 	NFSLOCKSTATE();
6669 	LIST_FOREACH(sep, &clp->lc_session, sess_list) {
6670 		if ((sep->sess_crflags & NFSV4CRSESS_CONNBACKCHAN) != 0)
6671 			break;
6672 	}
6673 	if (sep == NULL) {
6674 		NFSUNLOCKSTATE();
6675 		return (NFSERR_BADSESSION);
6676 	}
6677 	sep->sess_refcnt++;
6678 	*sepp = sep;
6679 	NFSUNLOCKSTATE();
6680 	return (0);
6681 }
6682 
6683 /*
6684  * Free up all backchannel xprts.  This needs to be done when the nfsd threads
6685  * exit, since those transports will all be going away.
6686  * This is only called after all the nfsd threads are done performing RPCs,
6687  * so locking shouldn't be an issue.
6688  */
6689 void
6690 nfsrv_freeallbackchannel_xprts(void)
6691 {
6692 	struct nfsdsession *sep;
6693 	struct nfsclient *clp;
6694 	SVCXPRT *xprt;
6695 	int i;
6696 
6697 	for (i = 0; i < nfsrv_clienthashsize; i++) {
6698 		LIST_FOREACH(clp, &nfsclienthash[i], lc_hash) {
6699 			LIST_FOREACH(sep, &clp->lc_session, sess_list) {
6700 				xprt = sep->sess_cbsess.nfsess_xprt;
6701 				sep->sess_cbsess.nfsess_xprt = NULL;
6702 				if (xprt != NULL)
6703 					SVC_RELEASE(xprt);
6704 			}
6705 		}
6706 	}
6707 }
6708 
6709 /*
6710  * Do a layout commit.  Actually just call nfsrv_updatemdsattr().
6711  * I have no idea if the rest of these arguments will ever be useful?
6712  */
6713 int
6714 nfsrv_layoutcommit(struct nfsrv_descript *nd, vnode_t vp, int layouttype,
6715     int hasnewoff, uint64_t newoff, uint64_t offset, uint64_t len,
6716     int hasnewmtime, struct timespec *newmtimep, int reclaim,
6717     nfsv4stateid_t *stateidp, int maxcnt, char *layp, int *hasnewsizep,
6718     uint64_t *newsizep, struct ucred *cred, NFSPROC_T *p)
6719 {
6720 	struct nfsvattr na;
6721 	int error;
6722 
6723 	error = nfsrv_updatemdsattr(vp, &na, p);
6724 	if (error == 0) {
6725 		*hasnewsizep = 1;
6726 		*newsizep = na.na_size;
6727 	}
6728 	return (error);
6729 }
6730 
6731 /*
6732  * Try and get a layout.
6733  */
6734 int
6735 nfsrv_layoutget(struct nfsrv_descript *nd, vnode_t vp, struct nfsexstuff *exp,
6736     int layouttype, int *iomode, uint64_t *offset, uint64_t *len,
6737     uint64_t minlen, nfsv4stateid_t *stateidp, int maxcnt, int *retonclose,
6738     int *layoutlenp, char *layp, struct ucred *cred, NFSPROC_T *p)
6739 {
6740 	struct nfslayouthash *lhyp;
6741 	struct nfslayout *lyp;
6742 	char *devid;
6743 	fhandle_t fh, *dsfhp;
6744 	int error, mirrorcnt;
6745 
6746 	if (nfsrv_devidcnt == 0)
6747 		return (NFSERR_UNKNLAYOUTTYPE);
6748 
6749 	if (*offset != 0)
6750 		printf("nfsrv_layoutget: off=%ju len=%ju\n", (uintmax_t)*offset,
6751 		    (uintmax_t)*len);
6752 	error = nfsvno_getfh(vp, &fh, p);
6753 	NFSD_DEBUG(4, "layoutget getfh=%d\n", error);
6754 	if (error != 0)
6755 		return (error);
6756 
6757 	/*
6758 	 * For now, all layouts are for entire files.
6759 	 * Only issue Read/Write layouts if requested for a non-readonly fs.
6760 	 */
6761 	if (NFSVNO_EXRDONLY(exp)) {
6762 		if (*iomode == NFSLAYOUTIOMODE_RW)
6763 			return (NFSERR_LAYOUTTRYLATER);
6764 		*iomode = NFSLAYOUTIOMODE_READ;
6765 	}
6766 	if (*iomode != NFSLAYOUTIOMODE_RW)
6767 		*iomode = NFSLAYOUTIOMODE_READ;
6768 
6769 	/*
6770 	 * Check to see if a write layout can be issued for this file.
6771 	 * This is used during mirror recovery to avoid RW layouts being
6772 	 * issued for a file while it is being copied to the recovered
6773 	 * mirror.
6774 	 */
6775 	if (*iomode == NFSLAYOUTIOMODE_RW && nfsrv_dontlayout(&fh) != 0)
6776 		return (NFSERR_LAYOUTTRYLATER);
6777 
6778 	*retonclose = 0;
6779 	*offset = 0;
6780 	*len = UINT64_MAX;
6781 
6782 	/* First, see if a layout already exists and return if found. */
6783 	lhyp = NFSLAYOUTHASH(&fh);
6784 	NFSLOCKLAYOUT(lhyp);
6785 	error = nfsrv_findlayout(&nd->nd_clientid, &fh, layouttype, p, &lyp);
6786 	NFSD_DEBUG(4, "layoutget findlay=%d\n", error);
6787 	/*
6788 	 * Not sure if the seqid must be the same, so I won't check it.
6789 	 */
6790 	if (error == 0 && (stateidp->other[0] != lyp->lay_stateid.other[0] ||
6791 	    stateidp->other[1] != lyp->lay_stateid.other[1] ||
6792 	    stateidp->other[2] != lyp->lay_stateid.other[2])) {
6793 		if ((lyp->lay_flags & NFSLAY_CALLB) == 0) {
6794 			NFSUNLOCKLAYOUT(lhyp);
6795 			NFSD_DEBUG(1, "ret bad stateid\n");
6796 			return (NFSERR_BADSTATEID);
6797 		}
6798 		/*
6799 		 * I believe we get here because there is a race between
6800 		 * the client processing the CBLAYOUTRECALL and the layout
6801 		 * being deleted here on the server.
6802 		 * The client has now done a LayoutGet with a non-layout
6803 		 * stateid, as it would when there is no layout.
6804 		 * As such, free this layout and set error == NFSERR_BADSTATEID
6805 		 * so the code below will create a new layout structure as
6806 		 * would happen if no layout was found.
6807 		 * "lyp" will be set before being used below, but set it NULL
6808 		 * as a safety belt.
6809 		 */
6810 		nfsrv_freelayout(&lhyp->list, lyp);
6811 		lyp = NULL;
6812 		error = NFSERR_BADSTATEID;
6813 	}
6814 	if (error == 0) {
6815 		if (lyp->lay_layoutlen > maxcnt) {
6816 			NFSUNLOCKLAYOUT(lhyp);
6817 			NFSD_DEBUG(1, "ret layout too small\n");
6818 			return (NFSERR_TOOSMALL);
6819 		}
6820 		if (*iomode == NFSLAYOUTIOMODE_RW) {
6821 			if ((lyp->lay_flags & NFSLAY_NOSPC) != 0) {
6822 				NFSUNLOCKLAYOUT(lhyp);
6823 				NFSD_DEBUG(1, "ret layout nospace\n");
6824 				return (NFSERR_NOSPC);
6825 			}
6826 			lyp->lay_flags |= NFSLAY_RW;
6827 		} else
6828 			lyp->lay_flags |= NFSLAY_READ;
6829 		NFSBCOPY(lyp->lay_xdr, layp, lyp->lay_layoutlen);
6830 		*layoutlenp = lyp->lay_layoutlen;
6831 		if (++lyp->lay_stateid.seqid == 0)
6832 			lyp->lay_stateid.seqid = 1;
6833 		stateidp->seqid = lyp->lay_stateid.seqid;
6834 		NFSUNLOCKLAYOUT(lhyp);
6835 		NFSD_DEBUG(4, "ret fnd layout\n");
6836 		return (0);
6837 	}
6838 	NFSUNLOCKLAYOUT(lhyp);
6839 
6840 	/* Find the device id and file handle. */
6841 	dsfhp = malloc(sizeof(fhandle_t) * NFSDEV_MAXMIRRORS, M_TEMP, M_WAITOK);
6842 	devid = malloc(NFSX_V4DEVICEID * NFSDEV_MAXMIRRORS, M_TEMP, M_WAITOK);
6843 	error = nfsrv_dsgetdevandfh(vp, p, &mirrorcnt, dsfhp, devid);
6844 	NFSD_DEBUG(4, "layoutget devandfh=%d\n", error);
6845 	if (error == 0) {
6846 		if (layouttype == NFSLAYOUT_NFSV4_1_FILES) {
6847 			if (NFSX_V4FILELAYOUT > maxcnt)
6848 				error = NFSERR_TOOSMALL;
6849 			else
6850 				lyp = nfsrv_filelayout(nd, *iomode, &fh, dsfhp,
6851 				    devid, vp->v_mount->mnt_stat.f_fsid);
6852 		} else {
6853 			if (NFSX_V4FLEXLAYOUT(mirrorcnt) > maxcnt)
6854 				error = NFSERR_TOOSMALL;
6855 			else
6856 				lyp = nfsrv_flexlayout(nd, *iomode, mirrorcnt,
6857 				    &fh, dsfhp, devid,
6858 				    vp->v_mount->mnt_stat.f_fsid);
6859 		}
6860 	}
6861 	free(dsfhp, M_TEMP);
6862 	free(devid, M_TEMP);
6863 	if (error != 0)
6864 		return (error);
6865 
6866 	/*
6867 	 * Now, add this layout to the list.
6868 	 */
6869 	error = nfsrv_addlayout(nd, &lyp, stateidp, layp, layoutlenp, p);
6870 	NFSD_DEBUG(4, "layoutget addl=%d\n", error);
6871 	/*
6872 	 * The lyp will be set to NULL by nfsrv_addlayout() if it
6873 	 * linked the new structure into the lists.
6874 	 */
6875 	free(lyp, M_NFSDSTATE);
6876 	return (error);
6877 }
6878 
6879 /*
6880  * Generate a File Layout.
6881  */
6882 static struct nfslayout *
6883 nfsrv_filelayout(struct nfsrv_descript *nd, int iomode, fhandle_t *fhp,
6884     fhandle_t *dsfhp, char *devid, fsid_t fs)
6885 {
6886 	uint32_t *tl;
6887 	struct nfslayout *lyp;
6888 	uint64_t pattern_offset;
6889 
6890 	lyp = malloc(sizeof(struct nfslayout) + NFSX_V4FILELAYOUT, M_NFSDSTATE,
6891 	    M_WAITOK | M_ZERO);
6892 	lyp->lay_type = NFSLAYOUT_NFSV4_1_FILES;
6893 	if (iomode == NFSLAYOUTIOMODE_RW)
6894 		lyp->lay_flags = NFSLAY_RW;
6895 	else
6896 		lyp->lay_flags = NFSLAY_READ;
6897 	NFSBCOPY(fhp, &lyp->lay_fh, sizeof(*fhp));
6898 	lyp->lay_clientid.qval = nd->nd_clientid.qval;
6899 	lyp->lay_fsid = fs;
6900 	NFSBCOPY(devid, lyp->lay_deviceid, NFSX_V4DEVICEID);
6901 
6902 	/* Fill in the xdr for the files layout. */
6903 	tl = (uint32_t *)lyp->lay_xdr;
6904 	NFSBCOPY(devid, tl, NFSX_V4DEVICEID);		/* Device ID. */
6905 	tl += (NFSX_V4DEVICEID / NFSX_UNSIGNED);
6906 
6907 	/* Set the stripe size to the maximum I/O size. */
6908 	*tl++ = txdr_unsigned(nfs_srvmaxio & NFSFLAYUTIL_STRIPE_MASK);
6909 	*tl++ = 0;					/* 1st stripe index. */
6910 	pattern_offset = 0;
6911 	txdr_hyper(pattern_offset, tl); tl += 2;	/* Pattern offset. */
6912 	*tl++ = txdr_unsigned(1);			/* 1 file handle. */
6913 	*tl++ = txdr_unsigned(NFSX_V4PNFSFH);
6914 	NFSBCOPY(dsfhp, tl, sizeof(*dsfhp));
6915 	lyp->lay_layoutlen = NFSX_V4FILELAYOUT;
6916 	return (lyp);
6917 }
6918 
6919 #define	FLEX_OWNERID	"999"
6920 #define	FLEX_UID0	"0"
6921 /*
6922  * Generate a Flex File Layout.
6923  * The FLEX_OWNERID can be any string of 3 decimal digits. Although this
6924  * string goes on the wire, it isn't supposed to be used by the client,
6925  * since this server uses tight coupling.
6926  * Although not recommended by the spec., if vfs.nfsd.flexlinuxhack=1 use
6927  * a string of "0". This works around the Linux Flex File Layout driver bug
6928  * which uses the synthetic uid/gid strings for the "tightly coupled" case.
6929  */
6930 static struct nfslayout *
6931 nfsrv_flexlayout(struct nfsrv_descript *nd, int iomode, int mirrorcnt,
6932     fhandle_t *fhp, fhandle_t *dsfhp, char *devid, fsid_t fs)
6933 {
6934 	uint32_t *tl;
6935 	struct nfslayout *lyp;
6936 	uint64_t lenval;
6937 	int i;
6938 
6939 	lyp = malloc(sizeof(struct nfslayout) + NFSX_V4FLEXLAYOUT(mirrorcnt),
6940 	    M_NFSDSTATE, M_WAITOK | M_ZERO);
6941 	lyp->lay_type = NFSLAYOUT_FLEXFILE;
6942 	if (iomode == NFSLAYOUTIOMODE_RW)
6943 		lyp->lay_flags = NFSLAY_RW;
6944 	else
6945 		lyp->lay_flags = NFSLAY_READ;
6946 	NFSBCOPY(fhp, &lyp->lay_fh, sizeof(*fhp));
6947 	lyp->lay_clientid.qval = nd->nd_clientid.qval;
6948 	lyp->lay_fsid = fs;
6949 	lyp->lay_mirrorcnt = mirrorcnt;
6950 	NFSBCOPY(devid, lyp->lay_deviceid, NFSX_V4DEVICEID);
6951 
6952 	/* Fill in the xdr for the files layout. */
6953 	tl = (uint32_t *)lyp->lay_xdr;
6954 	lenval = 0;
6955 	txdr_hyper(lenval, tl); tl += 2;		/* Stripe unit. */
6956 	*tl++ = txdr_unsigned(mirrorcnt);		/* # of mirrors. */
6957 	for (i = 0; i < mirrorcnt; i++) {
6958 		*tl++ = txdr_unsigned(1);		/* One stripe. */
6959 		NFSBCOPY(devid, tl, NFSX_V4DEVICEID);	/* Device ID. */
6960 		tl += (NFSX_V4DEVICEID / NFSX_UNSIGNED);
6961 		devid += NFSX_V4DEVICEID;
6962 		*tl++ = txdr_unsigned(1);		/* Efficiency. */
6963 		*tl++ = 0;				/* Proxy Stateid. */
6964 		*tl++ = 0x55555555;
6965 		*tl++ = 0x55555555;
6966 		*tl++ = 0x55555555;
6967 		*tl++ = txdr_unsigned(1);		/* 1 file handle. */
6968 		*tl++ = txdr_unsigned(NFSX_V4PNFSFH);
6969 		NFSBCOPY(dsfhp, tl, sizeof(*dsfhp));
6970 		tl += (NFSM_RNDUP(NFSX_V4PNFSFH) / NFSX_UNSIGNED);
6971 		dsfhp++;
6972 		if (nfsrv_flexlinuxhack != 0) {
6973 			*tl++ = txdr_unsigned(strlen(FLEX_UID0));
6974 			*tl = 0;		/* 0 pad string. */
6975 			NFSBCOPY(FLEX_UID0, tl++, strlen(FLEX_UID0));
6976 			*tl++ = txdr_unsigned(strlen(FLEX_UID0));
6977 			*tl = 0;		/* 0 pad string. */
6978 			NFSBCOPY(FLEX_UID0, tl++, strlen(FLEX_UID0));
6979 		} else {
6980 			*tl++ = txdr_unsigned(strlen(FLEX_OWNERID));
6981 			NFSBCOPY(FLEX_OWNERID, tl++, NFSX_UNSIGNED);
6982 			*tl++ = txdr_unsigned(strlen(FLEX_OWNERID));
6983 			NFSBCOPY(FLEX_OWNERID, tl++, NFSX_UNSIGNED);
6984 		}
6985 	}
6986 	*tl++ = txdr_unsigned(0);		/* ff_flags. */
6987 	*tl = txdr_unsigned(60);		/* Status interval hint. */
6988 	lyp->lay_layoutlen = NFSX_V4FLEXLAYOUT(mirrorcnt);
6989 	return (lyp);
6990 }
6991 
6992 /*
6993  * Parse and process Flex File errors returned via LayoutReturn.
6994  */
6995 static void
6996 nfsrv_flexlayouterr(struct nfsrv_descript *nd, uint32_t *layp, int maxcnt,
6997     NFSPROC_T *p)
6998 {
6999 	uint32_t *tl;
7000 	int cnt, errcnt, i, j, opnum, stat;
7001 	char devid[NFSX_V4DEVICEID];
7002 
7003 	tl = layp;
7004 	cnt = fxdr_unsigned(int, *tl++);
7005 	NFSD_DEBUG(4, "flexlayouterr cnt=%d\n", cnt);
7006 	for (i = 0; i < cnt; i++) {
7007 		/* Skip offset, length and stateid for now. */
7008 		tl += (4 + NFSX_STATEID / NFSX_UNSIGNED);
7009 		errcnt = fxdr_unsigned(int, *tl++);
7010 		NFSD_DEBUG(4, "flexlayouterr errcnt=%d\n", errcnt);
7011 		for (j = 0; j < errcnt; j++) {
7012 			NFSBCOPY(tl, devid, NFSX_V4DEVICEID);
7013 			tl += (NFSX_V4DEVICEID / NFSX_UNSIGNED);
7014 			stat = fxdr_unsigned(int, *tl++);
7015 			opnum = fxdr_unsigned(int, *tl++);
7016 			NFSD_DEBUG(4, "flexlayouterr op=%d stat=%d\n", opnum,
7017 			    stat);
7018 			/*
7019 			 * Except for NFSERR_ACCES, NFSERR_STALE and
7020 			 * NFSERR_NOSPC errors, disable the mirror.
7021 			 */
7022 			if (stat != NFSERR_ACCES && stat != NFSERR_STALE &&
7023 			    stat != NFSERR_NOSPC)
7024 				nfsrv_delds(devid, p);
7025 
7026 			/* For NFSERR_NOSPC, mark all devids and layouts. */
7027 			if (stat == NFSERR_NOSPC)
7028 				nfsrv_marknospc(devid, true);
7029 		}
7030 	}
7031 }
7032 
7033 /*
7034  * This function removes all flex file layouts which has a mirror with
7035  * a device id that matches the argument.
7036  * Called when the DS represented by the device id has failed.
7037  */
7038 void
7039 nfsrv_flexmirrordel(char *devid, NFSPROC_T *p)
7040 {
7041 	uint32_t *tl;
7042 	struct nfslayout *lyp, *nlyp;
7043 	struct nfslayouthash *lhyp;
7044 	struct nfslayouthead loclyp;
7045 	int i, j;
7046 
7047 	NFSD_DEBUG(4, "flexmirrordel\n");
7048 	/* Move all layouts found onto a local list. */
7049 	TAILQ_INIT(&loclyp);
7050 	for (i = 0; i < nfsrv_layouthashsize; i++) {
7051 		lhyp = &nfslayouthash[i];
7052 		NFSLOCKLAYOUT(lhyp);
7053 		TAILQ_FOREACH_SAFE(lyp, &lhyp->list, lay_list, nlyp) {
7054 			if (lyp->lay_type == NFSLAYOUT_FLEXFILE &&
7055 			    lyp->lay_mirrorcnt > 1) {
7056 				NFSD_DEBUG(4, "possible match\n");
7057 				tl = lyp->lay_xdr;
7058 				tl += 3;
7059 				for (j = 0; j < lyp->lay_mirrorcnt; j++) {
7060 					tl++;
7061 					if (NFSBCMP(devid, tl, NFSX_V4DEVICEID)
7062 					    == 0) {
7063 						/* Found one. */
7064 						NFSD_DEBUG(4, "fnd one\n");
7065 						TAILQ_REMOVE(&lhyp->list, lyp,
7066 						    lay_list);
7067 						TAILQ_INSERT_HEAD(&loclyp, lyp,
7068 						    lay_list);
7069 						break;
7070 					}
7071 					tl += (NFSX_V4DEVICEID / NFSX_UNSIGNED +
7072 					    NFSM_RNDUP(NFSX_V4PNFSFH) /
7073 					    NFSX_UNSIGNED + 11 * NFSX_UNSIGNED);
7074 				}
7075 			}
7076 		}
7077 		NFSUNLOCKLAYOUT(lhyp);
7078 	}
7079 
7080 	/* Now, try to do a Layout recall for each one found. */
7081 	TAILQ_FOREACH_SAFE(lyp, &loclyp, lay_list, nlyp) {
7082 		NFSD_DEBUG(4, "do layout recall\n");
7083 		/*
7084 		 * The layout stateid.seqid needs to be incremented
7085 		 * before doing a LAYOUT_RECALL callback.
7086 		 */
7087 		if (++lyp->lay_stateid.seqid == 0)
7088 			lyp->lay_stateid.seqid = 1;
7089 		nfsrv_recalllayout(lyp->lay_clientid, &lyp->lay_stateid,
7090 		    &lyp->lay_fh, lyp, 1, lyp->lay_type, p);
7091 		nfsrv_freelayout(&loclyp, lyp);
7092 	}
7093 }
7094 
7095 /*
7096  * Do a recall callback to the client for this layout.
7097  */
7098 static int
7099 nfsrv_recalllayout(nfsquad_t clid, nfsv4stateid_t *stateidp, fhandle_t *fhp,
7100     struct nfslayout *lyp, int changed, int laytype, NFSPROC_T *p)
7101 {
7102 	struct nfsclient *clp;
7103 	int error;
7104 
7105 	NFSD_DEBUG(4, "nfsrv_recalllayout\n");
7106 	error = nfsrv_getclient(clid, 0, &clp, NULL, (nfsquad_t)((u_quad_t)0),
7107 	    0, NULL, p);
7108 	NFSD_DEBUG(4, "aft nfsrv_getclient=%d\n", error);
7109 	if (error != 0) {
7110 		printf("nfsrv_recalllayout: getclient err=%d\n", error);
7111 		return (error);
7112 	}
7113 	if ((clp->lc_flags & LCL_NFSV41) != 0) {
7114 		error = nfsrv_docallback(clp, NFSV4OP_CBLAYOUTRECALL,
7115 		    stateidp, changed, fhp, NULL, NULL, laytype, p);
7116 		/* If lyp != NULL, handle an error return here. */
7117 		if (error != 0 && lyp != NULL) {
7118 			NFSDRECALLLOCK();
7119 			/*
7120 			 * Mark it returned, since no layout recall
7121 			 * has been done.
7122 			 * All errors seem to be non-recoverable, although
7123 			 * NFSERR_NOMATCHLAYOUT is a normal event.
7124 			 */
7125 			if ((lyp->lay_flags & NFSLAY_RECALL) != 0) {
7126 				lyp->lay_flags |= NFSLAY_RETURNED;
7127 				wakeup(lyp);
7128 			}
7129 			NFSDRECALLUNLOCK();
7130 			if (error != NFSERR_NOMATCHLAYOUT)
7131 				printf("nfsrv_recalllayout: err=%d\n", error);
7132 		}
7133 	} else
7134 		printf("nfsrv_recalllayout: clp not NFSv4.1\n");
7135 	return (error);
7136 }
7137 
7138 /*
7139  * Find a layout to recall when we exceed our high water mark.
7140  */
7141 void
7142 nfsrv_recalloldlayout(NFSPROC_T *p)
7143 {
7144 	struct nfslayouthash *lhyp;
7145 	struct nfslayout *lyp;
7146 	nfsquad_t clientid;
7147 	nfsv4stateid_t stateid;
7148 	fhandle_t fh;
7149 	int error, laytype = 0, ret;
7150 
7151 	lhyp = &nfslayouthash[arc4random() % nfsrv_layouthashsize];
7152 	NFSLOCKLAYOUT(lhyp);
7153 	TAILQ_FOREACH_REVERSE(lyp, &lhyp->list, nfslayouthead, lay_list) {
7154 		if ((lyp->lay_flags & NFSLAY_CALLB) == 0) {
7155 			lyp->lay_flags |= NFSLAY_CALLB;
7156 			/*
7157 			 * The layout stateid.seqid needs to be incremented
7158 			 * before doing a LAYOUT_RECALL callback.
7159 			 */
7160 			if (++lyp->lay_stateid.seqid == 0)
7161 				lyp->lay_stateid.seqid = 1;
7162 			clientid = lyp->lay_clientid;
7163 			stateid = lyp->lay_stateid;
7164 			NFSBCOPY(&lyp->lay_fh, &fh, sizeof(fh));
7165 			laytype = lyp->lay_type;
7166 			break;
7167 		}
7168 	}
7169 	NFSUNLOCKLAYOUT(lhyp);
7170 	if (lyp != NULL) {
7171 		error = nfsrv_recalllayout(clientid, &stateid, &fh, NULL, 0,
7172 		    laytype, p);
7173 		if (error != 0 && error != NFSERR_NOMATCHLAYOUT)
7174 			NFSD_DEBUG(4, "recallold=%d\n", error);
7175 		if (error != 0) {
7176 			NFSLOCKLAYOUT(lhyp);
7177 			/*
7178 			 * Since the hash list was unlocked, we need to
7179 			 * find it again.
7180 			 */
7181 			ret = nfsrv_findlayout(&clientid, &fh, laytype, p,
7182 			    &lyp);
7183 			if (ret == 0 &&
7184 			    (lyp->lay_flags & NFSLAY_CALLB) != 0 &&
7185 			    lyp->lay_stateid.other[0] == stateid.other[0] &&
7186 			    lyp->lay_stateid.other[1] == stateid.other[1] &&
7187 			    lyp->lay_stateid.other[2] == stateid.other[2]) {
7188 				/*
7189 				 * The client no longer knows this layout, so
7190 				 * it can be free'd now.
7191 				 */
7192 				if (error == NFSERR_NOMATCHLAYOUT)
7193 					nfsrv_freelayout(&lhyp->list, lyp);
7194 				else {
7195 					/*
7196 					 * Leave it to be tried later by
7197 					 * clearing NFSLAY_CALLB and moving
7198 					 * it to the head of the list, so it
7199 					 * won't be tried again for a while.
7200 					 */
7201 					lyp->lay_flags &= ~NFSLAY_CALLB;
7202 					TAILQ_REMOVE(&lhyp->list, lyp,
7203 					    lay_list);
7204 					TAILQ_INSERT_HEAD(&lhyp->list, lyp,
7205 					    lay_list);
7206 				}
7207 			}
7208 			NFSUNLOCKLAYOUT(lhyp);
7209 		}
7210 	}
7211 }
7212 
7213 /*
7214  * Try and return layout(s).
7215  */
7216 int
7217 nfsrv_layoutreturn(struct nfsrv_descript *nd, vnode_t vp,
7218     int layouttype, int iomode, uint64_t offset, uint64_t len, int reclaim,
7219     int kind, nfsv4stateid_t *stateidp, int maxcnt, uint32_t *layp, int *fndp,
7220     struct ucred *cred, NFSPROC_T *p)
7221 {
7222 	struct nfsvattr na;
7223 	struct nfslayouthash *lhyp;
7224 	struct nfslayout *lyp;
7225 	fhandle_t fh;
7226 	int error = 0;
7227 
7228 	*fndp = 0;
7229 	if (kind == NFSV4LAYOUTRET_FILE) {
7230 		error = nfsvno_getfh(vp, &fh, p);
7231 		if (error == 0) {
7232 			error = nfsrv_updatemdsattr(vp, &na, p);
7233 			if (error != 0)
7234 				printf("nfsrv_layoutreturn: updatemdsattr"
7235 				    " failed=%d\n", error);
7236 		}
7237 		if (error == 0) {
7238 			if (reclaim == newnfs_true) {
7239 				error = nfsrv_checkgrace(NULL, NULL,
7240 				    NFSLCK_RECLAIM);
7241 				if (error != NFSERR_NOGRACE)
7242 					error = 0;
7243 				return (error);
7244 			}
7245 			lhyp = NFSLAYOUTHASH(&fh);
7246 			NFSDRECALLLOCK();
7247 			NFSLOCKLAYOUT(lhyp);
7248 			error = nfsrv_findlayout(&nd->nd_clientid, &fh,
7249 			    layouttype, p, &lyp);
7250 			NFSD_DEBUG(4, "layoutret findlay=%d\n", error);
7251 			if (error == 0 &&
7252 			    stateidp->other[0] == lyp->lay_stateid.other[0] &&
7253 			    stateidp->other[1] == lyp->lay_stateid.other[1] &&
7254 			    stateidp->other[2] == lyp->lay_stateid.other[2]) {
7255 				NFSD_DEBUG(4, "nfsrv_layoutreturn: stateid %d"
7256 				    " %x %x %x laystateid %d %x %x %x"
7257 				    " off=%ju len=%ju flgs=0x%x\n",
7258 				    stateidp->seqid, stateidp->other[0],
7259 				    stateidp->other[1], stateidp->other[2],
7260 				    lyp->lay_stateid.seqid,
7261 				    lyp->lay_stateid.other[0],
7262 				    lyp->lay_stateid.other[1],
7263 				    lyp->lay_stateid.other[2],
7264 				    (uintmax_t)offset, (uintmax_t)len,
7265 				    lyp->lay_flags);
7266 				if (++lyp->lay_stateid.seqid == 0)
7267 					lyp->lay_stateid.seqid = 1;
7268 				stateidp->seqid = lyp->lay_stateid.seqid;
7269 				if (offset == 0 && len == UINT64_MAX) {
7270 					if ((iomode & NFSLAYOUTIOMODE_READ) !=
7271 					    0)
7272 						lyp->lay_flags &= ~NFSLAY_READ;
7273 					if ((iomode & NFSLAYOUTIOMODE_RW) != 0)
7274 						lyp->lay_flags &= ~NFSLAY_RW;
7275 					if ((lyp->lay_flags & (NFSLAY_READ |
7276 					    NFSLAY_RW)) == 0)
7277 						nfsrv_freelayout(&lhyp->list,
7278 						    lyp);
7279 					else
7280 						*fndp = 1;
7281 				} else
7282 					*fndp = 1;
7283 			}
7284 			NFSUNLOCKLAYOUT(lhyp);
7285 			/* Search the nfsrv_recalllist for a match. */
7286 			TAILQ_FOREACH(lyp, &nfsrv_recalllisthead, lay_list) {
7287 				if (NFSBCMP(&lyp->lay_fh, &fh,
7288 				    sizeof(fh)) == 0 &&
7289 				    lyp->lay_clientid.qval ==
7290 				    nd->nd_clientid.qval &&
7291 				    stateidp->other[0] ==
7292 				    lyp->lay_stateid.other[0] &&
7293 				    stateidp->other[1] ==
7294 				    lyp->lay_stateid.other[1] &&
7295 				    stateidp->other[2] ==
7296 				    lyp->lay_stateid.other[2]) {
7297 					lyp->lay_flags |= NFSLAY_RETURNED;
7298 					wakeup(lyp);
7299 					error = 0;
7300 				}
7301 			}
7302 			NFSDRECALLUNLOCK();
7303 		}
7304 		if (layouttype == NFSLAYOUT_FLEXFILE)
7305 			nfsrv_flexlayouterr(nd, layp, maxcnt, p);
7306 	} else if (kind == NFSV4LAYOUTRET_FSID)
7307 		nfsrv_freelayouts(&nd->nd_clientid,
7308 		    &vp->v_mount->mnt_stat.f_fsid, layouttype, iomode);
7309 	else if (kind == NFSV4LAYOUTRET_ALL)
7310 		nfsrv_freelayouts(&nd->nd_clientid, NULL, layouttype, iomode);
7311 	else
7312 		error = NFSERR_INVAL;
7313 	if (error == -1)
7314 		error = 0;
7315 	return (error);
7316 }
7317 
7318 /*
7319  * Look for an existing layout.
7320  */
7321 static int
7322 nfsrv_findlayout(nfsquad_t *clientidp, fhandle_t *fhp, int laytype,
7323     NFSPROC_T *p, struct nfslayout **lypp)
7324 {
7325 	struct nfslayouthash *lhyp;
7326 	struct nfslayout *lyp;
7327 	int ret;
7328 
7329 	*lypp = NULL;
7330 	ret = 0;
7331 	lhyp = NFSLAYOUTHASH(fhp);
7332 	TAILQ_FOREACH(lyp, &lhyp->list, lay_list) {
7333 		if (NFSBCMP(&lyp->lay_fh, fhp, sizeof(*fhp)) == 0 &&
7334 		    lyp->lay_clientid.qval == clientidp->qval &&
7335 		    lyp->lay_type == laytype)
7336 			break;
7337 	}
7338 	if (lyp != NULL)
7339 		*lypp = lyp;
7340 	else
7341 		ret = -1;
7342 	return (ret);
7343 }
7344 
7345 /*
7346  * Add the new layout, as required.
7347  */
7348 static int
7349 nfsrv_addlayout(struct nfsrv_descript *nd, struct nfslayout **lypp,
7350     nfsv4stateid_t *stateidp, char *layp, int *layoutlenp, NFSPROC_T *p)
7351 {
7352 	struct nfsclient *clp;
7353 	struct nfslayouthash *lhyp;
7354 	struct nfslayout *lyp, *nlyp;
7355 	fhandle_t *fhp;
7356 	int error;
7357 
7358 	KASSERT((nd->nd_flag & ND_IMPLIEDCLID) != 0,
7359 	    ("nfsrv_layoutget: no nd_clientid\n"));
7360 	lyp = *lypp;
7361 	fhp = &lyp->lay_fh;
7362 	NFSLOCKSTATE();
7363 	error = nfsrv_getclient((nfsquad_t)((u_quad_t)0), CLOPS_RENEW, &clp,
7364 	    NULL, (nfsquad_t)((u_quad_t)0), 0, nd, p);
7365 	if (error != 0) {
7366 		NFSUNLOCKSTATE();
7367 		return (error);
7368 	}
7369 	lyp->lay_stateid.seqid = stateidp->seqid = 1;
7370 	lyp->lay_stateid.other[0] = stateidp->other[0] =
7371 	    clp->lc_clientid.lval[0];
7372 	lyp->lay_stateid.other[1] = stateidp->other[1] =
7373 	    clp->lc_clientid.lval[1];
7374 	lyp->lay_stateid.other[2] = stateidp->other[2] =
7375 	    nfsrv_nextstateindex(clp);
7376 	NFSUNLOCKSTATE();
7377 
7378 	lhyp = NFSLAYOUTHASH(fhp);
7379 	NFSLOCKLAYOUT(lhyp);
7380 	TAILQ_FOREACH(nlyp, &lhyp->list, lay_list) {
7381 		if (NFSBCMP(&nlyp->lay_fh, fhp, sizeof(*fhp)) == 0 &&
7382 		    nlyp->lay_clientid.qval == nd->nd_clientid.qval)
7383 			break;
7384 	}
7385 	if (nlyp != NULL) {
7386 		/* A layout already exists, so use it. */
7387 		nlyp->lay_flags |= (lyp->lay_flags & (NFSLAY_READ | NFSLAY_RW));
7388 		NFSBCOPY(nlyp->lay_xdr, layp, nlyp->lay_layoutlen);
7389 		*layoutlenp = nlyp->lay_layoutlen;
7390 		if (++nlyp->lay_stateid.seqid == 0)
7391 			nlyp->lay_stateid.seqid = 1;
7392 		stateidp->seqid = nlyp->lay_stateid.seqid;
7393 		stateidp->other[0] = nlyp->lay_stateid.other[0];
7394 		stateidp->other[1] = nlyp->lay_stateid.other[1];
7395 		stateidp->other[2] = nlyp->lay_stateid.other[2];
7396 		NFSUNLOCKLAYOUT(lhyp);
7397 		return (0);
7398 	}
7399 
7400 	/* Insert the new layout in the lists. */
7401 	*lypp = NULL;
7402 	atomic_add_int(&nfsrv_layoutcnt, 1);
7403 	NFSBCOPY(lyp->lay_xdr, layp, lyp->lay_layoutlen);
7404 	*layoutlenp = lyp->lay_layoutlen;
7405 	TAILQ_INSERT_HEAD(&lhyp->list, lyp, lay_list);
7406 	NFSUNLOCKLAYOUT(lhyp);
7407 	return (0);
7408 }
7409 
7410 /*
7411  * Get the devinfo for a deviceid.
7412  */
7413 int
7414 nfsrv_getdevinfo(char *devid, int layouttype, uint32_t *maxcnt,
7415     uint32_t *notify, int *devaddrlen, char **devaddr)
7416 {
7417 	struct nfsdevice *ds;
7418 
7419 	if ((layouttype != NFSLAYOUT_NFSV4_1_FILES && layouttype !=
7420 	     NFSLAYOUT_FLEXFILE) ||
7421 	    (nfsrv_maxpnfsmirror > 1 && layouttype == NFSLAYOUT_NFSV4_1_FILES))
7422 		return (NFSERR_UNKNLAYOUTTYPE);
7423 
7424 	/*
7425 	 * Now, search for the device id.  Note that the structures won't go
7426 	 * away, but the order changes in the list.  As such, the lock only
7427 	 * needs to be held during the search through the list.
7428 	 */
7429 	NFSDDSLOCK();
7430 	TAILQ_FOREACH(ds, &nfsrv_devidhead, nfsdev_list) {
7431 		if (NFSBCMP(devid, ds->nfsdev_deviceid, NFSX_V4DEVICEID) == 0 &&
7432 		    ds->nfsdev_nmp != NULL)
7433 			break;
7434 	}
7435 	NFSDDSUNLOCK();
7436 	if (ds == NULL)
7437 		return (NFSERR_NOENT);
7438 
7439 	/* If the correct nfsdev_XXXXaddrlen is > 0, we have the device info. */
7440 	*devaddrlen = 0;
7441 	if (layouttype == NFSLAYOUT_NFSV4_1_FILES) {
7442 		*devaddrlen = ds->nfsdev_fileaddrlen;
7443 		*devaddr = ds->nfsdev_fileaddr;
7444 	} else if (layouttype == NFSLAYOUT_FLEXFILE) {
7445 		*devaddrlen = ds->nfsdev_flexaddrlen;
7446 		*devaddr = ds->nfsdev_flexaddr;
7447 	}
7448 	if (*devaddrlen == 0)
7449 		return (NFSERR_UNKNLAYOUTTYPE);
7450 
7451 	/*
7452 	 * The XDR overhead is 3 unsigned values: layout_type,
7453 	 * length_of_address and notify bitmap.
7454 	 * If the notify array is changed to not all zeros, the
7455 	 * count of unsigned values must be increased.
7456 	 */
7457 	if (*maxcnt > 0 && *maxcnt < NFSM_RNDUP(*devaddrlen) +
7458 	    3 * NFSX_UNSIGNED) {
7459 		*maxcnt = NFSM_RNDUP(*devaddrlen) + 3 * NFSX_UNSIGNED;
7460 		return (NFSERR_TOOSMALL);
7461 	}
7462 	return (0);
7463 }
7464 
7465 /*
7466  * Free a list of layout state structures.
7467  */
7468 static void
7469 nfsrv_freelayoutlist(nfsquad_t clientid)
7470 {
7471 	struct nfslayouthash *lhyp;
7472 	struct nfslayout *lyp, *nlyp;
7473 	int i;
7474 
7475 	for (i = 0; i < nfsrv_layouthashsize; i++) {
7476 		lhyp = &nfslayouthash[i];
7477 		NFSLOCKLAYOUT(lhyp);
7478 		TAILQ_FOREACH_SAFE(lyp, &lhyp->list, lay_list, nlyp) {
7479 			if (lyp->lay_clientid.qval == clientid.qval)
7480 				nfsrv_freelayout(&lhyp->list, lyp);
7481 		}
7482 		NFSUNLOCKLAYOUT(lhyp);
7483 	}
7484 }
7485 
7486 /*
7487  * Free up a layout.
7488  */
7489 static void
7490 nfsrv_freelayout(struct nfslayouthead *lhp, struct nfslayout *lyp)
7491 {
7492 
7493 	NFSD_DEBUG(4, "Freelayout=%p\n", lyp);
7494 	atomic_add_int(&nfsrv_layoutcnt, -1);
7495 	TAILQ_REMOVE(lhp, lyp, lay_list);
7496 	free(lyp, M_NFSDSTATE);
7497 }
7498 
7499 /*
7500  * Free up a device id.
7501  */
7502 void
7503 nfsrv_freeonedevid(struct nfsdevice *ds)
7504 {
7505 	int i;
7506 
7507 	atomic_add_int(&nfsrv_devidcnt, -1);
7508 	vrele(ds->nfsdev_dvp);
7509 	for (i = 0; i < nfsrv_dsdirsize; i++)
7510 		if (ds->nfsdev_dsdir[i] != NULL)
7511 			vrele(ds->nfsdev_dsdir[i]);
7512 	free(ds->nfsdev_fileaddr, M_NFSDSTATE);
7513 	free(ds->nfsdev_flexaddr, M_NFSDSTATE);
7514 	free(ds->nfsdev_host, M_NFSDSTATE);
7515 	free(ds, M_NFSDSTATE);
7516 }
7517 
7518 /*
7519  * Free up a device id and its mirrors.
7520  */
7521 static void
7522 nfsrv_freedevid(struct nfsdevice *ds)
7523 {
7524 
7525 	TAILQ_REMOVE(&nfsrv_devidhead, ds, nfsdev_list);
7526 	nfsrv_freeonedevid(ds);
7527 }
7528 
7529 /*
7530  * Free all layouts and device ids.
7531  * Done when the nfsd threads are shut down since there may be a new
7532  * modified device id list created when the nfsd is restarted.
7533  */
7534 void
7535 nfsrv_freealllayoutsanddevids(void)
7536 {
7537 	struct nfsdontlist *mrp, *nmrp;
7538 	struct nfslayout *lyp, *nlyp;
7539 
7540 	/* Get rid of the deviceid structures. */
7541 	nfsrv_freealldevids();
7542 	TAILQ_INIT(&nfsrv_devidhead);
7543 	nfsrv_devidcnt = 0;
7544 
7545 	/* Get rid of all layouts. */
7546 	nfsrv_freealllayouts();
7547 
7548 	/* Get rid of any nfsdontlist entries. */
7549 	LIST_FOREACH_SAFE(mrp, &nfsrv_dontlisthead, nfsmr_list, nmrp)
7550 		free(mrp, M_NFSDSTATE);
7551 	LIST_INIT(&nfsrv_dontlisthead);
7552 	nfsrv_dontlistlen = 0;
7553 
7554 	/* Free layouts in the recall list. */
7555 	TAILQ_FOREACH_SAFE(lyp, &nfsrv_recalllisthead, lay_list, nlyp)
7556 		nfsrv_freelayout(&nfsrv_recalllisthead, lyp);
7557 	TAILQ_INIT(&nfsrv_recalllisthead);
7558 }
7559 
7560 /*
7561  * Free layouts that match the arguments.
7562  */
7563 static void
7564 nfsrv_freelayouts(nfsquad_t *clid, fsid_t *fs, int laytype, int iomode)
7565 {
7566 	struct nfslayouthash *lhyp;
7567 	struct nfslayout *lyp, *nlyp;
7568 	int i;
7569 
7570 	for (i = 0; i < nfsrv_layouthashsize; i++) {
7571 		lhyp = &nfslayouthash[i];
7572 		NFSLOCKLAYOUT(lhyp);
7573 		TAILQ_FOREACH_SAFE(lyp, &lhyp->list, lay_list, nlyp) {
7574 			if (clid->qval != lyp->lay_clientid.qval)
7575 				continue;
7576 			if (fs != NULL && fsidcmp(fs, &lyp->lay_fsid) != 0)
7577 				continue;
7578 			if (laytype != lyp->lay_type)
7579 				continue;
7580 			if ((iomode & NFSLAYOUTIOMODE_READ) != 0)
7581 				lyp->lay_flags &= ~NFSLAY_READ;
7582 			if ((iomode & NFSLAYOUTIOMODE_RW) != 0)
7583 				lyp->lay_flags &= ~NFSLAY_RW;
7584 			if ((lyp->lay_flags & (NFSLAY_READ | NFSLAY_RW)) == 0)
7585 				nfsrv_freelayout(&lhyp->list, lyp);
7586 		}
7587 		NFSUNLOCKLAYOUT(lhyp);
7588 	}
7589 }
7590 
7591 /*
7592  * Free all layouts for the argument file.
7593  */
7594 void
7595 nfsrv_freefilelayouts(fhandle_t *fhp)
7596 {
7597 	struct nfslayouthash *lhyp;
7598 	struct nfslayout *lyp, *nlyp;
7599 
7600 	lhyp = NFSLAYOUTHASH(fhp);
7601 	NFSLOCKLAYOUT(lhyp);
7602 	TAILQ_FOREACH_SAFE(lyp, &lhyp->list, lay_list, nlyp) {
7603 		if (NFSBCMP(&lyp->lay_fh, fhp, sizeof(*fhp)) == 0)
7604 			nfsrv_freelayout(&lhyp->list, lyp);
7605 	}
7606 	NFSUNLOCKLAYOUT(lhyp);
7607 }
7608 
7609 /*
7610  * Free all layouts.
7611  */
7612 static void
7613 nfsrv_freealllayouts(void)
7614 {
7615 	struct nfslayouthash *lhyp;
7616 	struct nfslayout *lyp, *nlyp;
7617 	int i;
7618 
7619 	for (i = 0; i < nfsrv_layouthashsize; i++) {
7620 		lhyp = &nfslayouthash[i];
7621 		NFSLOCKLAYOUT(lhyp);
7622 		TAILQ_FOREACH_SAFE(lyp, &lhyp->list, lay_list, nlyp)
7623 			nfsrv_freelayout(&lhyp->list, lyp);
7624 		NFSUNLOCKLAYOUT(lhyp);
7625 	}
7626 }
7627 
7628 /*
7629  * Look up the mount path for the DS server.
7630  */
7631 static int
7632 nfsrv_setdsserver(char *dspathp, char *mdspathp, NFSPROC_T *p,
7633     struct nfsdevice **dsp)
7634 {
7635 	struct nameidata nd;
7636 	struct nfsdevice *ds;
7637 	struct mount *mp;
7638 	int error, i;
7639 	char *dsdirpath;
7640 	size_t dsdirsize;
7641 
7642 	NFSD_DEBUG(4, "setdssrv path=%s\n", dspathp);
7643 	*dsp = NULL;
7644 	NDINIT(&nd, LOOKUP, FOLLOW | LOCKSHARED | LOCKLEAF, UIO_SYSSPACE,
7645 	    dspathp, p);
7646 	error = namei(&nd);
7647 	NFSD_DEBUG(4, "lookup=%d\n", error);
7648 	if (error != 0)
7649 		return (error);
7650 	if (nd.ni_vp->v_type != VDIR) {
7651 		vput(nd.ni_vp);
7652 		NFSD_DEBUG(4, "dspath not dir\n");
7653 		return (ENOTDIR);
7654 	}
7655 	if (strcmp(nd.ni_vp->v_mount->mnt_vfc->vfc_name, "nfs") != 0) {
7656 		vput(nd.ni_vp);
7657 		NFSD_DEBUG(4, "dspath not an NFS mount\n");
7658 		return (ENXIO);
7659 	}
7660 
7661 	/*
7662 	 * Allocate a DS server structure with the NFS mounted directory
7663 	 * vnode reference counted, so that a non-forced dismount will
7664 	 * fail with EBUSY.
7665 	 * This structure is always linked into the list, even if an error
7666 	 * is being returned.  The caller will free the entire list upon
7667 	 * an error return.
7668 	 */
7669 	*dsp = ds = malloc(sizeof(*ds) + nfsrv_dsdirsize * sizeof(vnode_t),
7670 	    M_NFSDSTATE, M_WAITOK | M_ZERO);
7671 	ds->nfsdev_dvp = nd.ni_vp;
7672 	ds->nfsdev_nmp = VFSTONFS(nd.ni_vp->v_mount);
7673 	NFSVOPUNLOCK(nd.ni_vp);
7674 
7675 	dsdirsize = strlen(dspathp) + 16;
7676 	dsdirpath = malloc(dsdirsize, M_TEMP, M_WAITOK);
7677 	/* Now, create the DS directory structures. */
7678 	for (i = 0; i < nfsrv_dsdirsize; i++) {
7679 		snprintf(dsdirpath, dsdirsize, "%s/ds%d", dspathp, i);
7680 		NDINIT(&nd, LOOKUP, FOLLOW | LOCKSHARED | LOCKLEAF,
7681 		    UIO_SYSSPACE, dsdirpath, p);
7682 		error = namei(&nd);
7683 		NFSD_DEBUG(4, "dsdirpath=%s lookup=%d\n", dsdirpath, error);
7684 		if (error != 0)
7685 			break;
7686 		if (nd.ni_vp->v_type != VDIR) {
7687 			vput(nd.ni_vp);
7688 			error = ENOTDIR;
7689 			NFSD_DEBUG(4, "dsdirpath not a VDIR\n");
7690 			break;
7691 		}
7692 		if (strcmp(nd.ni_vp->v_mount->mnt_vfc->vfc_name, "nfs") != 0) {
7693 			vput(nd.ni_vp);
7694 			error = ENXIO;
7695 			NFSD_DEBUG(4, "dsdirpath not an NFS mount\n");
7696 			break;
7697 		}
7698 		ds->nfsdev_dsdir[i] = nd.ni_vp;
7699 		NFSVOPUNLOCK(nd.ni_vp);
7700 	}
7701 	free(dsdirpath, M_TEMP);
7702 
7703 	if (strlen(mdspathp) > 0) {
7704 		/*
7705 		 * This DS stores file for a specific MDS exported file
7706 		 * system.
7707 		 */
7708 		NDINIT(&nd, LOOKUP, FOLLOW | LOCKSHARED | LOCKLEAF,
7709 		    UIO_SYSSPACE, mdspathp, p);
7710 		error = namei(&nd);
7711 		NFSD_DEBUG(4, "mds lookup=%d\n", error);
7712 		if (error != 0)
7713 			goto out;
7714 		if (nd.ni_vp->v_type != VDIR) {
7715 			vput(nd.ni_vp);
7716 			error = ENOTDIR;
7717 			NFSD_DEBUG(4, "mdspath not dir\n");
7718 			goto out;
7719 		}
7720 		mp = nd.ni_vp->v_mount;
7721 		if ((mp->mnt_flag & MNT_EXPORTED) == 0) {
7722 			vput(nd.ni_vp);
7723 			error = ENXIO;
7724 			NFSD_DEBUG(4, "mdspath not an exported fs\n");
7725 			goto out;
7726 		}
7727 		ds->nfsdev_mdsfsid = mp->mnt_stat.f_fsid;
7728 		ds->nfsdev_mdsisset = 1;
7729 		vput(nd.ni_vp);
7730 	}
7731 
7732 out:
7733 	TAILQ_INSERT_TAIL(&nfsrv_devidhead, ds, nfsdev_list);
7734 	atomic_add_int(&nfsrv_devidcnt, 1);
7735 	return (error);
7736 }
7737 
7738 /*
7739  * Look up the mount path for the DS server and delete it.
7740  */
7741 int
7742 nfsrv_deldsserver(int op, char *dspathp, NFSPROC_T *p)
7743 {
7744 	struct mount *mp;
7745 	struct nfsmount *nmp;
7746 	struct nfsdevice *ds;
7747 	int error;
7748 
7749 	NFSD_DEBUG(4, "deldssrv path=%s\n", dspathp);
7750 	/*
7751 	 * Search for the path in the mount list.  Avoid looking the path
7752 	 * up, since this mount point may be hung, with associated locked
7753 	 * vnodes, etc.
7754 	 * Set NFSMNTP_CANCELRPCS so that any forced dismount will be blocked
7755 	 * until this completes.
7756 	 * As noted in the man page, this should be done before any forced
7757 	 * dismount on the mount point, but at least the handshake on
7758 	 * NFSMNTP_CANCELRPCS should make it safe.
7759 	 */
7760 	error = 0;
7761 	ds = NULL;
7762 	nmp = NULL;
7763 	mtx_lock(&mountlist_mtx);
7764 	TAILQ_FOREACH(mp, &mountlist, mnt_list) {
7765 		if (strcmp(mp->mnt_stat.f_mntonname, dspathp) == 0 &&
7766 		    strcmp(mp->mnt_stat.f_fstypename, "nfs") == 0 &&
7767 		    mp->mnt_data != NULL) {
7768 			nmp = VFSTONFS(mp);
7769 			NFSLOCKMNT(nmp);
7770 			if ((nmp->nm_privflag & (NFSMNTP_FORCEDISM |
7771 			     NFSMNTP_CANCELRPCS)) == 0) {
7772 				nmp->nm_privflag |= NFSMNTP_CANCELRPCS;
7773 				NFSUNLOCKMNT(nmp);
7774 			} else {
7775 				NFSUNLOCKMNT(nmp);
7776 				nmp = NULL;
7777 			}
7778 			break;
7779 		}
7780 	}
7781 	mtx_unlock(&mountlist_mtx);
7782 
7783 	if (nmp != NULL) {
7784 		ds = nfsrv_deldsnmp(op, nmp, p);
7785 		NFSD_DEBUG(4, "deldsnmp=%p\n", ds);
7786 		if (ds != NULL) {
7787 			nfsrv_killrpcs(nmp);
7788 			NFSD_DEBUG(4, "aft killrpcs\n");
7789 		} else
7790 			error = ENXIO;
7791 		NFSLOCKMNT(nmp);
7792 		nmp->nm_privflag &= ~NFSMNTP_CANCELRPCS;
7793 		wakeup(nmp);
7794 		NFSUNLOCKMNT(nmp);
7795 	} else
7796 		error = EINVAL;
7797 	return (error);
7798 }
7799 
7800 /*
7801  * Search for and remove a DS entry which matches the "nmp" argument.
7802  * The nfsdevice structure pointer is returned so that the caller can
7803  * free it via nfsrv_freeonedevid().
7804  * For the forced case, do not try to do LayoutRecalls, since the server
7805  * must be shut down now anyhow.
7806  */
7807 struct nfsdevice *
7808 nfsrv_deldsnmp(int op, struct nfsmount *nmp, NFSPROC_T *p)
7809 {
7810 	struct nfsdevice *fndds;
7811 
7812 	NFSD_DEBUG(4, "deldsdvp\n");
7813 	NFSDDSLOCK();
7814 	if (op == PNFSDOP_FORCEDELDS)
7815 		fndds = nfsv4_findmirror(nmp);
7816 	else
7817 		fndds = nfsrv_findmirroredds(nmp);
7818 	if (fndds != NULL)
7819 		nfsrv_deleteds(fndds);
7820 	NFSDDSUNLOCK();
7821 	if (fndds != NULL) {
7822 		if (op != PNFSDOP_FORCEDELDS)
7823 			nfsrv_flexmirrordel(fndds->nfsdev_deviceid, p);
7824 		printf("pNFS server: mirror %s failed\n", fndds->nfsdev_host);
7825 	}
7826 	return (fndds);
7827 }
7828 
7829 /*
7830  * Similar to nfsrv_deldsnmp(), except that the DS is indicated by deviceid.
7831  * This function also calls nfsrv_killrpcs() to unblock RPCs on the mount
7832  * point.
7833  * Also, returns an error instead of the nfsdevice found.
7834  */
7835 int
7836 nfsrv_delds(char *devid, NFSPROC_T *p)
7837 {
7838 	struct nfsdevice *ds, *fndds;
7839 	struct nfsmount *nmp;
7840 	int fndmirror;
7841 
7842 	NFSD_DEBUG(4, "delds\n");
7843 	/*
7844 	 * Search the DS server list for a match with devid.
7845 	 * Remove the DS entry if found and there is a mirror.
7846 	 */
7847 	fndds = NULL;
7848 	nmp = NULL;
7849 	fndmirror = 0;
7850 	NFSDDSLOCK();
7851 	TAILQ_FOREACH(ds, &nfsrv_devidhead, nfsdev_list) {
7852 		if (NFSBCMP(ds->nfsdev_deviceid, devid, NFSX_V4DEVICEID) == 0 &&
7853 		    ds->nfsdev_nmp != NULL) {
7854 			NFSD_DEBUG(4, "fnd main ds\n");
7855 			fndds = ds;
7856 			break;
7857 		}
7858 	}
7859 	if (fndds == NULL) {
7860 		NFSDDSUNLOCK();
7861 		return (ENXIO);
7862 	}
7863 	if (fndds->nfsdev_mdsisset == 0 && nfsrv_faildscnt > 0)
7864 		fndmirror = 1;
7865 	else if (fndds->nfsdev_mdsisset != 0) {
7866 		/* For the fsid is set case, search for a mirror. */
7867 		TAILQ_FOREACH(ds, &nfsrv_devidhead, nfsdev_list) {
7868 			if (ds != fndds && ds->nfsdev_nmp != NULL &&
7869 			    ds->nfsdev_mdsisset != 0 &&
7870 			    fsidcmp(&ds->nfsdev_mdsfsid,
7871 			    &fndds->nfsdev_mdsfsid) == 0) {
7872 				fndmirror = 1;
7873 				break;
7874 			}
7875 		}
7876 	}
7877 	if (fndmirror != 0) {
7878 		nmp = fndds->nfsdev_nmp;
7879 		NFSLOCKMNT(nmp);
7880 		if ((nmp->nm_privflag & (NFSMNTP_FORCEDISM |
7881 		     NFSMNTP_CANCELRPCS)) == 0) {
7882 			nmp->nm_privflag |= NFSMNTP_CANCELRPCS;
7883 			NFSUNLOCKMNT(nmp);
7884 			nfsrv_deleteds(fndds);
7885 		} else {
7886 			NFSUNLOCKMNT(nmp);
7887 			nmp = NULL;
7888 		}
7889 	}
7890 	NFSDDSUNLOCK();
7891 	if (nmp != NULL) {
7892 		nfsrv_flexmirrordel(fndds->nfsdev_deviceid, p);
7893 		printf("pNFS server: mirror %s failed\n", fndds->nfsdev_host);
7894 		nfsrv_killrpcs(nmp);
7895 		NFSLOCKMNT(nmp);
7896 		nmp->nm_privflag &= ~NFSMNTP_CANCELRPCS;
7897 		wakeup(nmp);
7898 		NFSUNLOCKMNT(nmp);
7899 		return (0);
7900 	}
7901 	return (ENXIO);
7902 }
7903 
7904 /*
7905  * Mark a DS as disabled by setting nfsdev_nmp = NULL.
7906  */
7907 static void
7908 nfsrv_deleteds(struct nfsdevice *fndds)
7909 {
7910 
7911 	NFSD_DEBUG(4, "deleteds: deleting a mirror\n");
7912 	fndds->nfsdev_nmp = NULL;
7913 	if (fndds->nfsdev_mdsisset == 0)
7914 		nfsrv_faildscnt--;
7915 }
7916 
7917 /*
7918  * Fill in the addr structures for the File and Flex File layouts.
7919  */
7920 static void
7921 nfsrv_allocdevid(struct nfsdevice *ds, char *addr, char *dnshost)
7922 {
7923 	uint32_t *tl;
7924 	char *netprot;
7925 	int addrlen;
7926 	static uint64_t new_devid = 0;
7927 
7928 	if (strchr(addr, ':') != NULL)
7929 		netprot = "tcp6";
7930 	else
7931 		netprot = "tcp";
7932 
7933 	/* Fill in the device id. */
7934 	NFSBCOPY(&nfsdev_time, ds->nfsdev_deviceid, sizeof(nfsdev_time));
7935 	new_devid++;
7936 	NFSBCOPY(&new_devid, &ds->nfsdev_deviceid[sizeof(nfsdev_time)],
7937 	    sizeof(new_devid));
7938 
7939 	/*
7940 	 * Fill in the file addr (actually the nfsv4_file_layout_ds_addr4
7941 	 * as defined in RFC5661) in XDR.
7942 	 */
7943 	addrlen = NFSM_RNDUP(strlen(addr)) + NFSM_RNDUP(strlen(netprot)) +
7944 	    6 * NFSX_UNSIGNED;
7945 	NFSD_DEBUG(4, "hn=%s addr=%s netprot=%s\n", dnshost, addr, netprot);
7946 	ds->nfsdev_fileaddrlen = addrlen;
7947 	tl = malloc(addrlen, M_NFSDSTATE, M_WAITOK | M_ZERO);
7948 	ds->nfsdev_fileaddr = (char *)tl;
7949 	*tl++ = txdr_unsigned(1);		/* One stripe with index 0. */
7950 	*tl++ = 0;
7951 	*tl++ = txdr_unsigned(1);		/* One multipath list */
7952 	*tl++ = txdr_unsigned(1);		/* with one entry in it. */
7953 	/* The netaddr for this one entry. */
7954 	*tl++ = txdr_unsigned(strlen(netprot));
7955 	NFSBCOPY(netprot, tl, strlen(netprot));
7956 	tl += (NFSM_RNDUP(strlen(netprot)) / NFSX_UNSIGNED);
7957 	*tl++ = txdr_unsigned(strlen(addr));
7958 	NFSBCOPY(addr, tl, strlen(addr));
7959 
7960 	/*
7961 	 * Fill in the flex file addr (actually the ff_device_addr4
7962 	 * as defined for Flexible File Layout) in XDR.
7963 	 */
7964 	addrlen = NFSM_RNDUP(strlen(addr)) + NFSM_RNDUP(strlen(netprot)) +
7965 	    14 * NFSX_UNSIGNED;
7966 	ds->nfsdev_flexaddrlen = addrlen;
7967 	tl = malloc(addrlen, M_NFSDSTATE, M_WAITOK | M_ZERO);
7968 	ds->nfsdev_flexaddr = (char *)tl;
7969 	*tl++ = txdr_unsigned(1);		/* One multipath entry. */
7970 	/* The netaddr for this one entry. */
7971 	*tl++ = txdr_unsigned(strlen(netprot));
7972 	NFSBCOPY(netprot, tl, strlen(netprot));
7973 	tl += (NFSM_RNDUP(strlen(netprot)) / NFSX_UNSIGNED);
7974 	*tl++ = txdr_unsigned(strlen(addr));
7975 	NFSBCOPY(addr, tl, strlen(addr));
7976 	tl += (NFSM_RNDUP(strlen(addr)) / NFSX_UNSIGNED);
7977 	*tl++ = txdr_unsigned(2);		/* Two NFS Versions. */
7978 	*tl++ = txdr_unsigned(NFS_VER4);	/* NFSv4. */
7979 	*tl++ = txdr_unsigned(NFSV42_MINORVERSION); /* Minor version 2. */
7980 	*tl++ = txdr_unsigned(nfs_srvmaxio);	/* DS max rsize. */
7981 	*tl++ = txdr_unsigned(nfs_srvmaxio);	/* DS max wsize. */
7982 	*tl++ = newnfs_true;			/* Tightly coupled. */
7983 	*tl++ = txdr_unsigned(NFS_VER4);	/* NFSv4. */
7984 	*tl++ = txdr_unsigned(NFSV41_MINORVERSION); /* Minor version 1. */
7985 	*tl++ = txdr_unsigned(nfs_srvmaxio);	/* DS max rsize. */
7986 	*tl++ = txdr_unsigned(nfs_srvmaxio);	/* DS max wsize. */
7987 	*tl = newnfs_true;			/* Tightly coupled. */
7988 
7989 	ds->nfsdev_hostnamelen = strlen(dnshost);
7990 	ds->nfsdev_host = malloc(ds->nfsdev_hostnamelen + 1, M_NFSDSTATE,
7991 	    M_WAITOK);
7992 	NFSBCOPY(dnshost, ds->nfsdev_host, ds->nfsdev_hostnamelen + 1);
7993 }
7994 
7995 /*
7996  * Create the device id list.
7997  * Return 0 if the nfsd threads are to run and ENXIO if the "-p" argument
7998  * is misconfigured.
7999  */
8000 int
8001 nfsrv_createdevids(struct nfsd_nfsd_args *args, NFSPROC_T *p)
8002 {
8003 	struct nfsdevice *ds;
8004 	char *addrp, *dnshostp, *dspathp, *mdspathp;
8005 	int error, i;
8006 
8007 	addrp = args->addr;
8008 	dnshostp = args->dnshost;
8009 	dspathp = args->dspath;
8010 	mdspathp = args->mdspath;
8011 	nfsrv_maxpnfsmirror = args->mirrorcnt;
8012 	if (addrp == NULL || dnshostp == NULL || dspathp == NULL ||
8013 	    mdspathp == NULL)
8014 		return (0);
8015 
8016 	/*
8017 	 * Loop around for each nul-terminated string in args->addr,
8018 	 * args->dnshost, args->dnspath and args->mdspath.
8019 	 */
8020 	while (addrp < (args->addr + args->addrlen) &&
8021 	    dnshostp < (args->dnshost + args->dnshostlen) &&
8022 	    dspathp < (args->dspath + args->dspathlen) &&
8023 	    mdspathp < (args->mdspath + args->mdspathlen)) {
8024 		error = nfsrv_setdsserver(dspathp, mdspathp, p, &ds);
8025 		if (error != 0) {
8026 			/* Free all DS servers. */
8027 			nfsrv_freealldevids();
8028 			nfsrv_devidcnt = 0;
8029 			return (ENXIO);
8030 		}
8031 		nfsrv_allocdevid(ds, addrp, dnshostp);
8032 		addrp += (strlen(addrp) + 1);
8033 		dnshostp += (strlen(dnshostp) + 1);
8034 		dspathp += (strlen(dspathp) + 1);
8035 		mdspathp += (strlen(mdspathp) + 1);
8036 	}
8037 	if (nfsrv_devidcnt < nfsrv_maxpnfsmirror) {
8038 		/* Free all DS servers. */
8039 		nfsrv_freealldevids();
8040 		nfsrv_devidcnt = 0;
8041 		nfsrv_maxpnfsmirror = 1;
8042 		return (ENXIO);
8043 	}
8044 	/* We can fail at most one less DS than the mirror level. */
8045 	nfsrv_faildscnt = nfsrv_maxpnfsmirror - 1;
8046 
8047 	/*
8048 	 * Allocate the nfslayout hash table now, since this is a pNFS server.
8049 	 * Make it 1% of the high water mark and at least 100.
8050 	 */
8051 	if (nfslayouthash == NULL) {
8052 		nfsrv_layouthashsize = nfsrv_layouthighwater / 100;
8053 		if (nfsrv_layouthashsize < 100)
8054 			nfsrv_layouthashsize = 100;
8055 		nfslayouthash = mallocarray(nfsrv_layouthashsize,
8056 		    sizeof(struct nfslayouthash), M_NFSDSESSION, M_WAITOK |
8057 		    M_ZERO);
8058 		for (i = 0; i < nfsrv_layouthashsize; i++) {
8059 			mtx_init(&nfslayouthash[i].mtx, "nfslm", NULL, MTX_DEF);
8060 			TAILQ_INIT(&nfslayouthash[i].list);
8061 		}
8062 	}
8063 	return (0);
8064 }
8065 
8066 /*
8067  * Free all device ids.
8068  */
8069 static void
8070 nfsrv_freealldevids(void)
8071 {
8072 	struct nfsdevice *ds, *nds;
8073 
8074 	TAILQ_FOREACH_SAFE(ds, &nfsrv_devidhead, nfsdev_list, nds)
8075 		nfsrv_freedevid(ds);
8076 }
8077 
8078 /*
8079  * Check to see if there is a Read/Write Layout plus either:
8080  * - A Write Delegation
8081  * or
8082  * - An Open with Write_access.
8083  * Return 1 if this is the case and 0 otherwise.
8084  * This function is used by nfsrv_proxyds() to decide if doing a Proxy
8085  * Getattr RPC to the Data Server (DS) is necessary.
8086  */
8087 #define	NFSCLIDVECSIZE	6
8088 int
8089 nfsrv_checkdsattr(vnode_t vp, NFSPROC_T *p)
8090 {
8091 	fhandle_t fh, *tfhp;
8092 	struct nfsstate *stp;
8093 	struct nfslayout *lyp;
8094 	struct nfslayouthash *lhyp;
8095 	struct nfslockhashhead *hp;
8096 	struct nfslockfile *lfp;
8097 	nfsquad_t clid[NFSCLIDVECSIZE];
8098 	int clidcnt, ret;
8099 
8100 	ret = nfsvno_getfh(vp, &fh, p);
8101 	if (ret != 0)
8102 		return (0);
8103 
8104 	/* First check for a Read/Write Layout. */
8105 	clidcnt = 0;
8106 	lhyp = NFSLAYOUTHASH(&fh);
8107 	NFSLOCKLAYOUT(lhyp);
8108 	TAILQ_FOREACH(lyp, &lhyp->list, lay_list) {
8109 		if (NFSBCMP(&lyp->lay_fh, &fh, sizeof(fh)) == 0 &&
8110 		    ((lyp->lay_flags & NFSLAY_RW) != 0 ||
8111 		     ((lyp->lay_flags & NFSLAY_READ) != 0 &&
8112 		      nfsrv_pnfsatime != 0))) {
8113 			if (clidcnt < NFSCLIDVECSIZE)
8114 				clid[clidcnt].qval = lyp->lay_clientid.qval;
8115 			clidcnt++;
8116 		}
8117 	}
8118 	NFSUNLOCKLAYOUT(lhyp);
8119 	if (clidcnt == 0) {
8120 		/* None found, so return 0. */
8121 		return (0);
8122 	}
8123 
8124 	/* Get the nfslockfile for this fh. */
8125 	NFSLOCKSTATE();
8126 	hp = NFSLOCKHASH(&fh);
8127 	LIST_FOREACH(lfp, hp, lf_hash) {
8128 		tfhp = &lfp->lf_fh;
8129 		if (NFSVNO_CMPFH(&fh, tfhp))
8130 			break;
8131 	}
8132 	if (lfp == NULL) {
8133 		/* None found, so return 0. */
8134 		NFSUNLOCKSTATE();
8135 		return (0);
8136 	}
8137 
8138 	/* Now, look for a Write delegation for this clientid. */
8139 	LIST_FOREACH(stp, &lfp->lf_deleg, ls_file) {
8140 		if ((stp->ls_flags & NFSLCK_DELEGWRITE) != 0 &&
8141 		    nfsrv_fndclid(clid, stp->ls_clp->lc_clientid, clidcnt) != 0)
8142 			break;
8143 	}
8144 	if (stp != NULL) {
8145 		/* Found one, so return 1. */
8146 		NFSUNLOCKSTATE();
8147 		return (1);
8148 	}
8149 
8150 	/* No Write delegation, so look for an Open with Write_access. */
8151 	LIST_FOREACH(stp, &lfp->lf_open, ls_file) {
8152 		KASSERT((stp->ls_flags & NFSLCK_OPEN) != 0,
8153 		    ("nfsrv_checkdsattr: Non-open in Open list\n"));
8154 		if ((stp->ls_flags & NFSLCK_WRITEACCESS) != 0 &&
8155 		    nfsrv_fndclid(clid, stp->ls_clp->lc_clientid, clidcnt) != 0)
8156 			break;
8157 	}
8158 	NFSUNLOCKSTATE();
8159 	if (stp != NULL)
8160 		return (1);
8161 	return (0);
8162 }
8163 
8164 /*
8165  * Look for a matching clientid in the vector. Return 1 if one might match.
8166  */
8167 static int
8168 nfsrv_fndclid(nfsquad_t *clidvec, nfsquad_t clid, int clidcnt)
8169 {
8170 	int i;
8171 
8172 	/* If too many for the vector, return 1 since there might be a match. */
8173 	if (clidcnt > NFSCLIDVECSIZE)
8174 		return (1);
8175 
8176 	for (i = 0; i < clidcnt; i++)
8177 		if (clidvec[i].qval == clid.qval)
8178 			return (1);
8179 	return (0);
8180 }
8181 
8182 /*
8183  * Check the don't list for "vp" and see if issuing an rw layout is allowed.
8184  * Return 1 if issuing an rw layout isn't allowed, 0 otherwise.
8185  */
8186 static int
8187 nfsrv_dontlayout(fhandle_t *fhp)
8188 {
8189 	struct nfsdontlist *mrp;
8190 	int ret;
8191 
8192 	if (nfsrv_dontlistlen == 0)
8193 		return (0);
8194 	ret = 0;
8195 	NFSDDONTLISTLOCK();
8196 	LIST_FOREACH(mrp, &nfsrv_dontlisthead, nfsmr_list) {
8197 		if (NFSBCMP(fhp, &mrp->nfsmr_fh, sizeof(*fhp)) == 0 &&
8198 		    (mrp->nfsmr_flags & NFSMR_DONTLAYOUT) != 0) {
8199 			ret = 1;
8200 			break;
8201 		}
8202 	}
8203 	NFSDDONTLISTUNLOCK();
8204 	return (ret);
8205 }
8206 
8207 #define	PNFSDS_COPYSIZ	65536
8208 /*
8209  * Create a new file on a DS and copy the contents of an extant DS file to it.
8210  * This can be used for recovery of a DS file onto a recovered DS.
8211  * The steps are:
8212  * - When called, the MDS file's vnode is locked, blocking LayoutGet operations.
8213  * - Disable issuing of read/write layouts for the file via the nfsdontlist,
8214  *   so that they will be disabled after the MDS file's vnode is unlocked.
8215  * - Set up the nfsrv_recalllist so that recall of read/write layouts can
8216  *   be done.
8217  * - Unlock the MDS file's vnode, so that the client(s) can perform proxied
8218  *   writes, LayoutCommits and LayoutReturns for the file when completing the
8219  *   LayoutReturn requested by the LayoutRecall callback.
8220  * - Issue a LayoutRecall callback for all read/write layouts and wait for
8221  *   them to be returned. (If the LayoutRecall callback replies
8222  *   NFSERR_NOMATCHLAYOUT, they are gone and no LayoutReturn is needed.)
8223  * - Exclusively lock the MDS file's vnode.  This ensures that no proxied
8224  *   writes are in progress or can occur during the DS file copy.
8225  *   It also blocks Setattr operations.
8226  * - Create the file on the recovered mirror.
8227  * - Copy the file from the operational DS.
8228  * - Copy any ACL from the MDS file to the new DS file.
8229  * - Set the modify time of the new DS file to that of the MDS file.
8230  * - Update the extended attribute for the MDS file.
8231  * - Enable issuing of rw layouts by deleting the nfsdontlist entry.
8232  * - The caller will unlock the MDS file's vnode allowing operations
8233  *   to continue normally, since it is now on the mirror again.
8234  */
8235 int
8236 nfsrv_copymr(vnode_t vp, vnode_t fvp, vnode_t dvp, struct nfsdevice *ds,
8237     struct pnfsdsfile *pf, struct pnfsdsfile *wpf, int mirrorcnt,
8238     struct ucred *cred, NFSPROC_T *p)
8239 {
8240 	struct nfsdontlist *mrp, *nmrp;
8241 	struct nfslayouthash *lhyp;
8242 	struct nfslayout *lyp, *nlyp;
8243 	struct nfslayouthead thl;
8244 	struct mount *mp, *tvmp;
8245 	struct acl *aclp;
8246 	struct vattr va;
8247 	struct timespec mtime;
8248 	fhandle_t fh;
8249 	vnode_t tvp;
8250 	off_t rdpos, wrpos;
8251 	ssize_t aresid;
8252 	char *dat;
8253 	int didprintf, ret, retacl, xfer;
8254 
8255 	ASSERT_VOP_LOCKED(fvp, "nfsrv_copymr fvp");
8256 	ASSERT_VOP_LOCKED(vp, "nfsrv_copymr vp");
8257 	/*
8258 	 * Allocate a nfsdontlist entry and set the NFSMR_DONTLAYOUT flag
8259 	 * so that no more RW layouts will get issued.
8260 	 */
8261 	ret = nfsvno_getfh(vp, &fh, p);
8262 	if (ret != 0) {
8263 		NFSD_DEBUG(4, "nfsrv_copymr: getfh=%d\n", ret);
8264 		return (ret);
8265 	}
8266 	nmrp = malloc(sizeof(*nmrp), M_NFSDSTATE, M_WAITOK);
8267 	nmrp->nfsmr_flags = NFSMR_DONTLAYOUT;
8268 	NFSBCOPY(&fh, &nmrp->nfsmr_fh, sizeof(fh));
8269 	NFSDDONTLISTLOCK();
8270 	LIST_FOREACH(mrp, &nfsrv_dontlisthead, nfsmr_list) {
8271 		if (NFSBCMP(&fh, &mrp->nfsmr_fh, sizeof(fh)) == 0)
8272 			break;
8273 	}
8274 	if (mrp == NULL) {
8275 		LIST_INSERT_HEAD(&nfsrv_dontlisthead, nmrp, nfsmr_list);
8276 		mrp = nmrp;
8277 		nmrp = NULL;
8278 		nfsrv_dontlistlen++;
8279 		NFSD_DEBUG(4, "nfsrv_copymr: in dontlist\n");
8280 	} else {
8281 		NFSDDONTLISTUNLOCK();
8282 		free(nmrp, M_NFSDSTATE);
8283 		NFSD_DEBUG(4, "nfsrv_copymr: dup dontlist\n");
8284 		return (ENXIO);
8285 	}
8286 	NFSDDONTLISTUNLOCK();
8287 
8288 	/*
8289 	 * Search for all RW layouts for this file.  Move them to the
8290 	 * recall list, so they can be recalled and their return noted.
8291 	 */
8292 	lhyp = NFSLAYOUTHASH(&fh);
8293 	NFSDRECALLLOCK();
8294 	NFSLOCKLAYOUT(lhyp);
8295 	TAILQ_FOREACH_SAFE(lyp, &lhyp->list, lay_list, nlyp) {
8296 		if (NFSBCMP(&lyp->lay_fh, &fh, sizeof(fh)) == 0 &&
8297 		    (lyp->lay_flags & NFSLAY_RW) != 0) {
8298 			TAILQ_REMOVE(&lhyp->list, lyp, lay_list);
8299 			TAILQ_INSERT_HEAD(&nfsrv_recalllisthead, lyp, lay_list);
8300 			lyp->lay_trycnt = 0;
8301 		}
8302 	}
8303 	NFSUNLOCKLAYOUT(lhyp);
8304 	NFSDRECALLUNLOCK();
8305 
8306 	ret = 0;
8307 	mp = tvmp = NULL;
8308 	didprintf = 0;
8309 	TAILQ_INIT(&thl);
8310 	/* Unlock the MDS vp, so that a LayoutReturn can be done on it. */
8311 	NFSVOPUNLOCK(vp);
8312 	/* Now, do a recall for all layouts not yet recalled. */
8313 tryagain:
8314 	NFSDRECALLLOCK();
8315 	TAILQ_FOREACH(lyp, &nfsrv_recalllisthead, lay_list) {
8316 		if (NFSBCMP(&lyp->lay_fh, &fh, sizeof(fh)) == 0 &&
8317 		    (lyp->lay_flags & NFSLAY_RECALL) == 0) {
8318 			lyp->lay_flags |= NFSLAY_RECALL;
8319 			/*
8320 			 * The layout stateid.seqid needs to be incremented
8321 			 * before doing a LAYOUT_RECALL callback.
8322 			 */
8323 			if (++lyp->lay_stateid.seqid == 0)
8324 				lyp->lay_stateid.seqid = 1;
8325 			NFSDRECALLUNLOCK();
8326 			nfsrv_recalllayout(lyp->lay_clientid, &lyp->lay_stateid,
8327 			    &lyp->lay_fh, lyp, 0, lyp->lay_type, p);
8328 			NFSD_DEBUG(4, "nfsrv_copymr: recalled layout\n");
8329 			goto tryagain;
8330 		}
8331 	}
8332 
8333 	/* Now wait for them to be returned. */
8334 tryagain2:
8335 	TAILQ_FOREACH(lyp, &nfsrv_recalllisthead, lay_list) {
8336 		if (NFSBCMP(&lyp->lay_fh, &fh, sizeof(fh)) == 0) {
8337 			if ((lyp->lay_flags & NFSLAY_RETURNED) != 0) {
8338 				TAILQ_REMOVE(&nfsrv_recalllisthead, lyp,
8339 				    lay_list);
8340 				TAILQ_INSERT_HEAD(&thl, lyp, lay_list);
8341 				NFSD_DEBUG(4,
8342 				    "nfsrv_copymr: layout returned\n");
8343 			} else {
8344 				lyp->lay_trycnt++;
8345 				ret = mtx_sleep(lyp, NFSDRECALLMUTEXPTR,
8346 				    PVFS | PCATCH, "nfsmrl", hz);
8347 				NFSD_DEBUG(4, "nfsrv_copymr: aft sleep=%d\n",
8348 				    ret);
8349 				if (ret == EINTR || ret == ERESTART)
8350 					break;
8351 				if ((lyp->lay_flags & NFSLAY_RETURNED) == 0) {
8352 					/*
8353 					 * Give up after 60sec and return
8354 					 * ENXIO, failing the copymr.
8355 					 * This layout will remain on the
8356 					 * recalllist.  It can only be cleared
8357 					 * by restarting the nfsd.
8358 					 * This seems the safe way to handle
8359 					 * it, since it cannot be safely copied
8360 					 * with an outstanding RW layout.
8361 					 */
8362 					if (lyp->lay_trycnt >= 60) {
8363 						ret = ENXIO;
8364 						break;
8365 					}
8366 					if (didprintf == 0) {
8367 						printf("nfsrv_copymr: layout "
8368 						    "not returned\n");
8369 						didprintf = 1;
8370 					}
8371 				}
8372 			}
8373 			goto tryagain2;
8374 		}
8375 	}
8376 	NFSDRECALLUNLOCK();
8377 	/* We can now get rid of the layouts that have been returned. */
8378 	TAILQ_FOREACH_SAFE(lyp, &thl, lay_list, nlyp)
8379 		nfsrv_freelayout(&thl, lyp);
8380 
8381 	/*
8382 	 * Do the vn_start_write() calls here, before the MDS vnode is
8383 	 * locked and the tvp is created (locked) in the NFS file system
8384 	 * that dvp is in.
8385 	 * For tvmp, this probably isn't necessary, since it will be an
8386 	 * NFS mount and they are not suspendable at this time.
8387 	 */
8388 	if (ret == 0)
8389 		ret = vn_start_write(vp, &mp, V_WAIT | PCATCH);
8390 	if (ret == 0) {
8391 		tvmp = dvp->v_mount;
8392 		ret = vn_start_write(NULL, &tvmp, V_WAIT | PCATCH);
8393 	}
8394 
8395 	/*
8396 	 * LK_EXCLUSIVE lock the MDS vnode, so that any
8397 	 * proxied writes through the MDS will be blocked until we have
8398 	 * completed the copy and update of the extended attributes.
8399 	 * This will also ensure that any attributes and ACL will not be
8400 	 * changed until the copy is complete.
8401 	 */
8402 	NFSVOPLOCK(vp, LK_EXCLUSIVE | LK_RETRY);
8403 	if (ret == 0 && VN_IS_DOOMED(vp)) {
8404 		NFSD_DEBUG(4, "nfsrv_copymr: lk_exclusive doomed\n");
8405 		ret = ESTALE;
8406 	}
8407 
8408 	/* Create the data file on the recovered DS. */
8409 	if (ret == 0)
8410 		ret = nfsrv_createdsfile(vp, &fh, pf, dvp, ds, cred, p, &tvp);
8411 
8412 	/* Copy the DS file, if created successfully. */
8413 	if (ret == 0) {
8414 		/*
8415 		 * Get any NFSv4 ACL on the MDS file, so that it can be set
8416 		 * on the new DS file.
8417 		 */
8418 		aclp = acl_alloc(M_WAITOK | M_ZERO);
8419 		retacl = VOP_GETACL(vp, ACL_TYPE_NFS4, aclp, cred, p);
8420 		if (retacl != 0 && retacl != ENOATTR)
8421 			NFSD_DEBUG(1, "nfsrv_copymr: vop_getacl=%d\n", retacl);
8422 		dat = malloc(PNFSDS_COPYSIZ, M_TEMP, M_WAITOK);
8423 		/* Malloc a block of 0s used to check for holes. */
8424 		if (nfsrv_zeropnfsdat == NULL)
8425 			nfsrv_zeropnfsdat = malloc(PNFSDS_COPYSIZ, M_TEMP,
8426 			    M_WAITOK | M_ZERO);
8427 		rdpos = wrpos = 0;
8428 		ret = VOP_GETATTR(fvp, &va, cred);
8429 		aresid = 0;
8430 		while (ret == 0 && aresid == 0) {
8431 			ret = vn_rdwr(UIO_READ, fvp, dat, PNFSDS_COPYSIZ,
8432 			    rdpos, UIO_SYSSPACE, IO_NODELOCKED, cred, NULL,
8433 			    &aresid, p);
8434 			xfer = PNFSDS_COPYSIZ - aresid;
8435 			if (ret == 0 && xfer > 0) {
8436 				rdpos += xfer;
8437 				/*
8438 				 * Skip the write for holes, except for the
8439 				 * last block.
8440 				 */
8441 				if (xfer < PNFSDS_COPYSIZ || rdpos ==
8442 				    va.va_size || NFSBCMP(dat,
8443 				    nfsrv_zeropnfsdat, PNFSDS_COPYSIZ) != 0)
8444 					ret = vn_rdwr(UIO_WRITE, tvp, dat, xfer,
8445 					    wrpos, UIO_SYSSPACE, IO_NODELOCKED,
8446 					    cred, NULL, NULL, p);
8447 				if (ret == 0)
8448 					wrpos += xfer;
8449 			}
8450 		}
8451 
8452 		/* If there is an ACL and the copy succeeded, set the ACL. */
8453 		if (ret == 0 && retacl == 0) {
8454 			ret = VOP_SETACL(tvp, ACL_TYPE_NFS4, aclp, cred, p);
8455 			/*
8456 			 * Don't consider these as errors, since VOP_GETACL()
8457 			 * can return an ACL when they are not actually
8458 			 * supported.  For example, for UFS, VOP_GETACL()
8459 			 * will return a trivial ACL based on the uid/gid/mode
8460 			 * when there is no ACL on the file.
8461 			 * This case should be recognized as a trivial ACL
8462 			 * by UFS's VOP_SETACL() and succeed, but...
8463 			 */
8464 			if (ret == ENOATTR || ret == EOPNOTSUPP || ret == EPERM)
8465 				ret = 0;
8466 		}
8467 
8468 		if (ret == 0)
8469 			ret = VOP_FSYNC(tvp, MNT_WAIT, p);
8470 
8471 		/* Set the DS data file's modify time that of the MDS file. */
8472 		if (ret == 0)
8473 			ret = VOP_GETATTR(vp, &va, cred);
8474 		if (ret == 0) {
8475 			mtime = va.va_mtime;
8476 			VATTR_NULL(&va);
8477 			va.va_mtime = mtime;
8478 			ret = VOP_SETATTR(tvp, &va, cred);
8479 		}
8480 
8481 		vput(tvp);
8482 		acl_free(aclp);
8483 		free(dat, M_TEMP);
8484 	}
8485 	if (tvmp != NULL)
8486 		vn_finished_write(tvmp);
8487 
8488 	/* Update the extended attributes for the newly created DS file. */
8489 	if (ret == 0)
8490 		ret = vn_extattr_set(vp, IO_NODELOCKED,
8491 		    EXTATTR_NAMESPACE_SYSTEM, "pnfsd.dsfile",
8492 		    sizeof(*wpf) * mirrorcnt, (char *)wpf, p);
8493 	if (mp != NULL)
8494 		vn_finished_write(mp);
8495 
8496 	/* Get rid of the dontlist entry, so that Layouts can be issued. */
8497 	NFSDDONTLISTLOCK();
8498 	LIST_REMOVE(mrp, nfsmr_list);
8499 	NFSDDONTLISTUNLOCK();
8500 	free(mrp, M_NFSDSTATE);
8501 	return (ret);
8502 }
8503 
8504 /*
8505  * Create a data storage file on the recovered DS.
8506  */
8507 static int
8508 nfsrv_createdsfile(vnode_t vp, fhandle_t *fhp, struct pnfsdsfile *pf,
8509     vnode_t dvp, struct nfsdevice *ds, struct ucred *cred, NFSPROC_T *p,
8510     vnode_t *tvpp)
8511 {
8512 	struct vattr va, nva;
8513 	int error;
8514 
8515 	/* Make data file name based on FH. */
8516 	error = VOP_GETATTR(vp, &va, cred);
8517 	if (error == 0) {
8518 		/* Set the attributes for "vp" to Setattr the DS vp. */
8519 		VATTR_NULL(&nva);
8520 		nva.va_uid = va.va_uid;
8521 		nva.va_gid = va.va_gid;
8522 		nva.va_mode = va.va_mode;
8523 		nva.va_size = 0;
8524 		VATTR_NULL(&va);
8525 		va.va_type = VREG;
8526 		va.va_mode = nva.va_mode;
8527 		NFSD_DEBUG(4, "nfsrv_dscreatefile: dvp=%p pf=%p\n", dvp, pf);
8528 		error = nfsrv_dscreate(dvp, &va, &nva, fhp, pf, NULL,
8529 		    pf->dsf_filename, cred, p, tvpp);
8530 	}
8531 	return (error);
8532 }
8533 
8534 /*
8535  * Look up the MDS file shared locked, and then get the extended attribute
8536  * to find the extant DS file to be copied to the new mirror.
8537  * If successful, *vpp is set to the MDS file's vp and *nvpp is
8538  * set to a DS data file for the MDS file, both exclusively locked.
8539  * The "buf" argument has the pnfsdsfile structure from the MDS file
8540  * in it and buflen is set to its length.
8541  */
8542 int
8543 nfsrv_mdscopymr(char *mdspathp, char *dspathp, char *curdspathp, char *buf,
8544     int *buflenp, char *fname, NFSPROC_T *p, struct vnode **vpp,
8545     struct vnode **nvpp, struct pnfsdsfile **pfp, struct nfsdevice **dsp,
8546     struct nfsdevice **fdsp)
8547 {
8548 	struct nameidata nd;
8549 	struct vnode *vp, *curvp;
8550 	struct pnfsdsfile *pf;
8551 	struct nfsmount *nmp, *curnmp;
8552 	int dsdir, error, mirrorcnt, ippos;
8553 
8554 	vp = NULL;
8555 	curvp = NULL;
8556 	curnmp = NULL;
8557 	*dsp = NULL;
8558 	*fdsp = NULL;
8559 	if (dspathp == NULL && curdspathp != NULL)
8560 		return (EPERM);
8561 
8562 	/*
8563 	 * Look up the MDS file shared locked.  The lock will be upgraded
8564 	 * to an exclusive lock after any rw layouts have been returned.
8565 	 */
8566 	NFSD_DEBUG(4, "mdsopen path=%s\n", mdspathp);
8567 	NDINIT(&nd, LOOKUP, FOLLOW | LOCKSHARED | LOCKLEAF, UIO_SYSSPACE,
8568 	    mdspathp, p);
8569 	error = namei(&nd);
8570 	NFSD_DEBUG(4, "lookup=%d\n", error);
8571 	if (error != 0)
8572 		return (error);
8573 	if (nd.ni_vp->v_type != VREG) {
8574 		vput(nd.ni_vp);
8575 		NFSD_DEBUG(4, "mdspath not reg\n");
8576 		return (EISDIR);
8577 	}
8578 	vp = nd.ni_vp;
8579 
8580 	if (curdspathp != NULL) {
8581 		/*
8582 		 * Look up the current DS path and find the nfsdev structure for
8583 		 * it.
8584 		 */
8585 		NFSD_DEBUG(4, "curmdsdev path=%s\n", curdspathp);
8586 		NDINIT(&nd, LOOKUP, FOLLOW | LOCKSHARED | LOCKLEAF,
8587 		    UIO_SYSSPACE, curdspathp, p);
8588 		error = namei(&nd);
8589 		NFSD_DEBUG(4, "ds lookup=%d\n", error);
8590 		if (error != 0) {
8591 			vput(vp);
8592 			return (error);
8593 		}
8594 		if (nd.ni_vp->v_type != VDIR) {
8595 			vput(nd.ni_vp);
8596 			vput(vp);
8597 			NFSD_DEBUG(4, "curdspath not dir\n");
8598 			return (ENOTDIR);
8599 		}
8600 		if (strcmp(nd.ni_vp->v_mount->mnt_vfc->vfc_name, "nfs") != 0) {
8601 			vput(nd.ni_vp);
8602 			vput(vp);
8603 			NFSD_DEBUG(4, "curdspath not an NFS mount\n");
8604 			return (ENXIO);
8605 		}
8606 		curnmp = VFSTONFS(nd.ni_vp->v_mount);
8607 
8608 		/* Search the nfsdev list for a match. */
8609 		NFSDDSLOCK();
8610 		*fdsp = nfsv4_findmirror(curnmp);
8611 		NFSDDSUNLOCK();
8612 		if (*fdsp == NULL)
8613 			curnmp = NULL;
8614 		if (curnmp == NULL) {
8615 			vput(nd.ni_vp);
8616 			vput(vp);
8617 			NFSD_DEBUG(4, "mdscopymr: no current ds\n");
8618 			return (ENXIO);
8619 		}
8620 		curvp = nd.ni_vp;
8621 	}
8622 
8623 	if (dspathp != NULL) {
8624 		/* Look up the nfsdev path and find the nfsdev structure. */
8625 		NFSD_DEBUG(4, "mdsdev path=%s\n", dspathp);
8626 		NDINIT(&nd, LOOKUP, FOLLOW | LOCKSHARED | LOCKLEAF,
8627 		    UIO_SYSSPACE, dspathp, p);
8628 		error = namei(&nd);
8629 		NFSD_DEBUG(4, "ds lookup=%d\n", error);
8630 		if (error != 0) {
8631 			vput(vp);
8632 			if (curvp != NULL)
8633 				vput(curvp);
8634 			return (error);
8635 		}
8636 		if (nd.ni_vp->v_type != VDIR || nd.ni_vp == curvp) {
8637 			vput(nd.ni_vp);
8638 			vput(vp);
8639 			if (curvp != NULL)
8640 				vput(curvp);
8641 			NFSD_DEBUG(4, "dspath not dir\n");
8642 			if (nd.ni_vp == curvp)
8643 				return (EPERM);
8644 			return (ENOTDIR);
8645 		}
8646 		if (strcmp(nd.ni_vp->v_mount->mnt_vfc->vfc_name, "nfs") != 0) {
8647 			vput(nd.ni_vp);
8648 			vput(vp);
8649 			if (curvp != NULL)
8650 				vput(curvp);
8651 			NFSD_DEBUG(4, "dspath not an NFS mount\n");
8652 			return (ENXIO);
8653 		}
8654 		nmp = VFSTONFS(nd.ni_vp->v_mount);
8655 
8656 		/*
8657 		 * Search the nfsdevice list for a match.  If curnmp == NULL,
8658 		 * this is a recovery and there must be a mirror.
8659 		 */
8660 		NFSDDSLOCK();
8661 		if (curnmp == NULL)
8662 			*dsp = nfsrv_findmirroredds(nmp);
8663 		else
8664 			*dsp = nfsv4_findmirror(nmp);
8665 		NFSDDSUNLOCK();
8666 		if (*dsp == NULL) {
8667 			vput(nd.ni_vp);
8668 			vput(vp);
8669 			if (curvp != NULL)
8670 				vput(curvp);
8671 			NFSD_DEBUG(4, "mdscopymr: no ds\n");
8672 			return (ENXIO);
8673 		}
8674 	} else {
8675 		nd.ni_vp = NULL;
8676 		nmp = NULL;
8677 	}
8678 
8679 	/*
8680 	 * Get a vp for an available DS data file using the extended
8681 	 * attribute on the MDS file.
8682 	 * If there is a valid entry for the new DS in the extended attribute
8683 	 * on the MDS file (as checked via the nmp argument),
8684 	 * nfsrv_dsgetsockmnt() returns EEXIST, so no copying will occur.
8685 	 */
8686 	error = nfsrv_dsgetsockmnt(vp, 0, buf, buflenp, &mirrorcnt, p,
8687 	    NULL, NULL, NULL, fname, nvpp, &nmp, curnmp, &ippos, &dsdir);
8688 	if (curvp != NULL)
8689 		vput(curvp);
8690 	if (nd.ni_vp == NULL) {
8691 		if (error == 0 && nmp != NULL) {
8692 			/* Search the nfsdev list for a match. */
8693 			NFSDDSLOCK();
8694 			*dsp = nfsrv_findmirroredds(nmp);
8695 			NFSDDSUNLOCK();
8696 		}
8697 		if (error == 0 && (nmp == NULL || *dsp == NULL)) {
8698 			if (nvpp != NULL && *nvpp != NULL) {
8699 				vput(*nvpp);
8700 				*nvpp = NULL;
8701 			}
8702 			error = ENXIO;
8703 		}
8704 	} else
8705 		vput(nd.ni_vp);
8706 
8707 	/*
8708 	 * When dspathp != NULL and curdspathp == NULL, this is a recovery
8709 	 * and is only allowed if there is a 0.0.0.0 IP address entry.
8710 	 * When curdspathp != NULL, the ippos will be set to that entry.
8711 	 */
8712 	if (error == 0 && dspathp != NULL && ippos == -1) {
8713 		if (nvpp != NULL && *nvpp != NULL) {
8714 			vput(*nvpp);
8715 			*nvpp = NULL;
8716 		}
8717 		error = ENXIO;
8718 	}
8719 	if (error == 0) {
8720 		*vpp = vp;
8721 
8722 		pf = (struct pnfsdsfile *)buf;
8723 		if (ippos == -1) {
8724 			/* If no zeroip pnfsdsfile, add one. */
8725 			ippos = *buflenp / sizeof(*pf);
8726 			*buflenp += sizeof(*pf);
8727 			pf += ippos;
8728 			pf->dsf_dir = dsdir;
8729 			strlcpy(pf->dsf_filename, fname,
8730 			    sizeof(pf->dsf_filename));
8731 		} else
8732 			pf += ippos;
8733 		*pfp = pf;
8734 	} else
8735 		vput(vp);
8736 	return (error);
8737 }
8738 
8739 /*
8740  * Search for a matching pnfsd mirror device structure, base on the nmp arg.
8741  * Return one if found, NULL otherwise.
8742  */
8743 static struct nfsdevice *
8744 nfsrv_findmirroredds(struct nfsmount *nmp)
8745 {
8746 	struct nfsdevice *ds, *fndds;
8747 	int fndmirror;
8748 
8749 	mtx_assert(NFSDDSMUTEXPTR, MA_OWNED);
8750 	/*
8751 	 * Search the DS server list for a match with nmp.
8752 	 * Remove the DS entry if found and there is a mirror.
8753 	 */
8754 	fndds = NULL;
8755 	fndmirror = 0;
8756 	if (nfsrv_devidcnt == 0)
8757 		return (fndds);
8758 	TAILQ_FOREACH(ds, &nfsrv_devidhead, nfsdev_list) {
8759 		if (ds->nfsdev_nmp == nmp) {
8760 			NFSD_DEBUG(4, "nfsrv_findmirroredds: fnd main ds\n");
8761 			fndds = ds;
8762 			break;
8763 		}
8764 	}
8765 	if (fndds == NULL)
8766 		return (fndds);
8767 	if (fndds->nfsdev_mdsisset == 0 && nfsrv_faildscnt > 0)
8768 		fndmirror = 1;
8769 	else if (fndds->nfsdev_mdsisset != 0) {
8770 		/* For the fsid is set case, search for a mirror. */
8771 		TAILQ_FOREACH(ds, &nfsrv_devidhead, nfsdev_list) {
8772 			if (ds != fndds && ds->nfsdev_nmp != NULL &&
8773 			    ds->nfsdev_mdsisset != 0 &&
8774 			    fsidcmp(&ds->nfsdev_mdsfsid,
8775 			    &fndds->nfsdev_mdsfsid) == 0) {
8776 				fndmirror = 1;
8777 				break;
8778 			}
8779 		}
8780 	}
8781 	if (fndmirror == 0) {
8782 		NFSD_DEBUG(4, "nfsrv_findmirroredds: no mirror for DS\n");
8783 		return (NULL);
8784 	}
8785 	return (fndds);
8786 }
8787 
8788 /*
8789  * Mark the appropriate devid and all associated layout as "out of space".
8790  */
8791 void
8792 nfsrv_marknospc(char *devid, bool setit)
8793 {
8794 	struct nfsdevice *ds;
8795 	struct nfslayout *lyp;
8796 	struct nfslayouthash *lhyp;
8797 	int i;
8798 
8799 	NFSDDSLOCK();
8800 	TAILQ_FOREACH(ds, &nfsrv_devidhead, nfsdev_list) {
8801 		if (NFSBCMP(ds->nfsdev_deviceid, devid, NFSX_V4DEVICEID) == 0) {
8802 			NFSD_DEBUG(1, "nfsrv_marknospc: devid %d\n", setit);
8803 			ds->nfsdev_nospc = setit;
8804 		}
8805 	}
8806 	NFSDDSUNLOCK();
8807 
8808 	for (i = 0; i < nfsrv_layouthashsize; i++) {
8809 		lhyp = &nfslayouthash[i];
8810 		NFSLOCKLAYOUT(lhyp);
8811 		TAILQ_FOREACH(lyp, &lhyp->list, lay_list) {
8812 			if (NFSBCMP(lyp->lay_deviceid, devid,
8813 			    NFSX_V4DEVICEID) == 0) {
8814 				NFSD_DEBUG(1, "nfsrv_marknospc: layout %d\n",
8815 				    setit);
8816 				if (setit)
8817 					lyp->lay_flags |= NFSLAY_NOSPC;
8818 				else
8819 					lyp->lay_flags &= ~NFSLAY_NOSPC;
8820 			}
8821 		}
8822 		NFSUNLOCKLAYOUT(lhyp);
8823 	}
8824 }
8825